Michael T. Wolfinger

Rational design of mechanically active RNAs

2026-05-12T00:00:00+02:00

xrRNA secondary and ring-like tertiary structure

This paper extends the xrRNA story from comparative RNA virology into explicit RNA engineering. Exoribonuclease-resistant RNAs are structured viral elements that stall 5' to 3' decay through a threaded, mechanically resistant fold. The central question addressed here is whether that function can be specified de novo, rather than inherited from a natural sequence that already encodes it.

The work approaches topology as part of the design objective. Structural features relevant for XRN1 resistance are reduced to an explicit symbolic model, which is then used to generate synthetic candidates that are filtered computationally and validated experimentally. The resulting workflow combines structural constraints, ensemble-based screening, three-dimensional modeling, and molecular-dynamics selection for ring closure and directional force resistance.

One of the clearest mechanistic observations concerns the unequal contribution of the two pseudoknots to xrRNA stability. Pseudoknot 2 acts as the decisive gatekeeper of mechanical resistance, whereas pseudoknot 1 contributes additional stabilization once the fold is established. This distinction turns what initially appears as a descriptive structural feature into a tractable design rule.

Three synthetic constructs were used to progressively narrow the design space. syn-xrRNA1 approaches the intended topology but remains comparatively weak in vitro. syn-xrRNA2 improves the geometry of the critical pseudoknot region and reaches wild-type-like XRN1 resistance. syn-xrRNA3 goes a step further: much of the recognizable sequence signal is removed, yet the RNA still folds into a functional threaded architecture and efficiently resists XRN1 degradation.

To my knowledge, this is the first example of a fully de novo designed RNA in which a mechanically active fold of this complexity was specified from topological constraints and then validated experimentally. In that sense, the paper marks a genuine shift in how RNA design can be approached. Mechanically active RNA elements become accessible through topology and geometry rather than through direct sequence imitation of natural viral RNAs.

For synthetic biology and therapeutic RNA engineering, this opens the possibility of tuning decay resistance and transcript stability without importing long native viral sequence segments wholesale. At the same time, the work remains fundamentally a study in RNA architecture. The main result is that mechanical function can be preserved even after evolutionary sequence ancestry has effectively been removed.

The broader implications extend well beyond flavivirus RNA biology. If mechanically active RNA elements can be specified de novo, then decay control becomes an engineering variable rather than a borrowed viral feature. That changes how one can think about stabilizing synthetic transcripts, shaping RNA lifetime in therapeutic settings, or introducing programmable decay barriers into larger regulatory designs. In RNA therapeutics, where persistence, dosage, and degradation behavior are often as important as coding capacity, this kind of design principle could become a useful addition to the current toolbox. More generally, the work shows that higher-order RNA architecture itself can be treated as a design substrate, not only as something discovered retrospectively in natural molecules.

The present study also closes a conceptual loop with earlier work on natural xrRNAs and structured viral RNAs. Comparative analysis originally established the relevant folds in nature. The current paper asks which parts of that function survive once sequence history is removed and only the underlying topological logic is retained.

The January preprint-stage note remains online as a shorter record of how the project was framed before the peer-review process.

Citation

Rational design of mechanically active RNAs: de novo engineering of functional exoribonuclease-resistant RNAs
Jule Walter, Leonhard Sidl, Katrin Gutenbrunner, Denis Skibinski, Tim Kolberg, Ivo L. Hofacker, Hua-Ting Yao, Mario Mörl, Michael T. Wolfinger
Nucleic Acids Res. 54(9):gkag473 (2026) | doi:10.1093/nar/gkag473 | Article | PDF

How to Run Claude Code with a Local LLM on Apple Silicon

2026-04-17T00:00:00+02:00

This guide shows you how to run Claude Code with a locally hosted large language model on Apple Silicon. The supported integration path is to let LM Studio expose an Anthropic-compatible API endpoint and then point Claude Code at that local endpoint via environment variables.

The examples below use LM Studio on localhost:1234 and a locally loaded model with a 32K context window.

Prerequisites

A Mac with Apple Silicon.
LM Studio installed.
claude installed.
A local model downloaded or imported into LM Studio.
Enough available RAM for the model you want to run.

Install Claude Code

Install Claude Code using one of the current supported methods.

Native install:

curl -fsSL https://claude.ai/install.sh | bash

Or via Homebrew:

brew install --cask claude-code

After installation, verify that the CLI is available:

claude --version
claude doctor

Part 1: Prepare LM Studio

Start LM Studio

Open LM Studio at least once. Then verify that the lms CLI is available in your shell.

Check that lms is installed:

lms --help

If this command is not found, follow the LM Studio CLI setup/bootstrap step from the LM Studio documentation and then reopen your shell.

List locally available models:

lms ls

If your model files were downloaded outside LM Studio, you can import them:

lms import /path/to/model.gguf

Load a Model with a 32K Context Window

Load your chosen model into memory and set the context length explicitly.

If you want to estimate memory usage before loading:

lms load --estimate-only <model_key> --context-length 32768

lms load <model_key> --context-length 32768

You can also assign a stable identifier for API use:

lms load <model_key> --context-length 32768 --identifier qwen-local

To see which models are currently loaded:

lms ps

Start the Local Server

Start LM Studio’s local server:

lms server start --port 1234

Check server status:

lms server status

At this point, LM Studio serves an Anthropic-compatible Messages API on http://localhost:1234/v1/messages.

Part 2: Point Claude Code at LM Studio

Configure Environment Variables

Set Claude Code to use your local LM Studio server:

export ANTHROPIC_BASE_URL=http://localhost:1234
export ANTHROPIC_AUTH_TOKEN=lmstudio

If LM Studio’s Require Authentication option is enabled, replace lmstudio with your LM Studio API token.

When using ANTHROPIC_BASE_URL plus ANTHROPIC_AUTH_TOKEN against LM Studio, Claude Code authenticates against the local endpoint and does not need the usual browser login flow for that session.

Choose the Model

You can select the model when launching Claude Code:

claude --model qwen-local

If you did not assign a custom identifier, use the loaded model name that LM Studio exposes.

Alternatively, set the model via an environment variable:

export ANTHROPIC_MODEL=qwen-local
claude

Part 3: Test the Local Endpoint

Before starting Claude Code, test LM Studio directly against the Anthropic-compatible endpoint.

curl http://localhost:1234/v1/messages \
  -H "content-type: application/json" \
  -H "x-api-key: lmstudio" \
  -d '{
    "model": "qwen-local",
    "max_tokens": 128,
    "messages": [
      {
        "role": "user",
        "content": "Write a one-line hello message."
      }
    ]
  }'

If this request succeeds, Claude Code should be able to use the same local server.

Part 4: Run Claude Code in a Project

Start Claude Code inside your project directory:

cd /path/to/your/project
claude --model qwen-local

Typical workflow:

Ask Claude Code to inspect files in the repository.
Let it propose or apply edits.
Run your local build or test commands.
Review diffs before committing changes.

For example, in a Pelican project:

pelican content

And then commit as usual:

git add .
git commit -m "Update article"

Troubleshooting

Claude Code cannot connect

Check that the LM Studio server is running:

lms server status

Check that the model is loaded:

lms ps

Check that the environment variables are set in the same shell where you launch claude:

echo "$ANTHROPIC_BASE_URL"
echo "$ANTHROPIC_AUTH_TOKEN"

Responses are slow

Agentic coding workloads are context-heavy. A local model may perform noticeably better with a larger context window. A practical starting point is around 25K tokens or more, and 32K is a reasonable target if your hardware can support it.

If performance is poor:

Try a smaller or more coding-oriented model.
Reduce other RAM-heavy workloads.
Lower the context length if the model does not fit comfortably.
Test a shorter prompt first to verify baseline responsiveness.

Need server or prompt diagnostics

Stream LM Studio logs:

lms log stream --source server

Stream model input and output:

lms log stream --source model --filter input,output

Include prediction stats when available:

lms log stream --source model --filter output --stats

Notes

This setup uses LM Studio’s Anthropic-compatible /v1/messages endpoint. It does not require a custom JSON provider file, and it does not use the old /v1/completions integration pattern.

If you want a persistent setup, you can place the same environment variables in your shell profile, for example ~/.zshrc:

export ANTHROPIC_BASE_URL=http://localhost:1234
export ANTHROPIC_AUTH_TOKEN=lmstudio
export ANTHROPIC_MODEL=qwen-local

Summary

With LM Studio serving a local model through its Anthropic-compatible Messages API and Claude Code pointed at that local endpoint, you can run a local-inference coding workflow on Apple Silicon.

In short, the essential steps are:

Load a model in LM Studio with your desired context length.
Start the LM Studio local server.
Set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN.
Launch Claude Code with --model or ANTHROPIC_MODEL.

A first look at rational design of mechanically active RNAs

2026-01-08T00:00:00+01:00

Ring-like tertiary structure of a designed synthetic xrRNA

Exoribonuclease-resistant RNAs (xrRNAs) are among the clearest examples of mechanically active RNAs. Their function depends on a ring-like topology that blocks 5' to 3' decay by enzymes such as XRN1. That makes them attractive design targets, but also difficult ones, because the relevant features are not captured well by conventional secondary structure design alone.

The main interest of this work was whether that kind of function could be approached rationally. The study starts from topology rather than sequence conservation and asks which structural elements are indispensable for XRN1 resistance. In the Aroa virus xrRNA used as a benchmark, the two pseudoknots in the xrRNA do not contribute equally: pseudoknot 2 behaves as the decisive gatekeeper of mechanical resistance, whereas pseudoknot 1 is important but less determinant on its own.

Those observations were then turned into a design workflow. Natural mosquito-borne flavivirus xrRNAs were reduced to a symbolic representation that preserves the three-way junction, the two pseudoknots, and characteristic length constraints between structural elements. Sequence generation was carried out with explicit structural and topological constraints, followed by ensemble-based refinement, SimRNA modeling, and molecular dynamics screening for ring closure and directional force resistance. The point was not simply to inverse-fold a target secondary structure, but to make topology part of the design objective.

The synthetic constructs provide a useful progression. syn-xrRNA1 captured the general architecture in silico but remained too weak experimentally. syn-xrRNA2 strengthened the crucial topological region and reached wild-type-like XRN1 resistance. syn-xrRNA3 then removed most of the familiar evolutionary sequence signal while preserving the geometric and energetic requirements for function. Even in that reduced form, the construct still folded into a bona fidae xrRNA architecture and stalled XRN1 efficiently.

The main conceptual result of the preprint is the following: mechanical RNA function could be approached through topology and geometry without relying on obvious sequence ancestry. The broader implications for synthetic biology and transcript engineering are better discussed in the forthcoming publication post, once the peer-reviewed paper is available in final form.

Citation

Rational design of mechanically active RNAs: de novo engineering of functional exoribonuclease-resistant RNAs
Jule Walter, Leonhard Sidl, Katrin Gutenbrunner, Denis Skibinski, Tim Kolberg, Ivo L. Hofacker, Hua-Ting Yao, Mario Mörl, Michael T. Wolfinger
bioRxiv 2026.01.08.698366 (2026) | doi:10.1101/2026.01.08.698366 | Preprint PDF

Functional RNAs in Virology Special Issue

2025-07-29T00:00:00+02:00

I’m honored to serve as Guest Editor of the Functional RNAs in Virology Special Issue. My goal is to curate a collection of high-quality articles that explore how viral RNAs, both structured and unstructured, control translation, replication, genome packaging, immune evasion, and host adaptation.

Understanding the Role of RNA in Viruses

Far beyond acting as genetic messengers, viral RNAs have emerged as active regulators that shape virtually every step of the infection cycle. From modulating protein synthesis to regulating host interactions and genome replication, these RNAs form a molecular toolkit that viruses rely on for survival and success.

Structured elements such as IRESs, frame-shifting pseudoknots, and exoribonuclease-resistant RNAs (xrRNAs), together with unstructured sequence motifs, form an intricate regulatory network that allows viruses to thrive in diverse cellular environments. Recent discoveries, including computational predictions of highly conserved RNA elements across viral families, emphasize the evolutionary significance of RNA-based regulation.

Understanding how viruses exploit RNA structure and sequence is essential not only to decode their infection strategies but also to identify novel targets for antiviral intervention.

Scope of the Special Issue

The Functional RNAs in Virology Special Issue aims to provide a comprehensive overview of how RNA acts as a regulatory molecule in viral infection. We welcome both original research articles and reviews that explore well-characterized RNA elements as well as emerging discoveries. Topics of interest include, but are not limited to:

Functional RNA structures involved in translation, replication, and genome packaging
Long-range RNA–RNA interactions and RNA conformational switches
Viral non-coding RNAs
Exoribonuclease-resistant RNAs (xrRNAs) and RNA-driven immune evasion
RNA structural conservation and covariation across virus species or lineages
Computational discovery and modeling of functional RNA elements
RNA–protein interactions involving viral RNAs and host factors
RNA structure probing and high-throughput functional RNA screening in viruses

A Timely Focus

In recent years, experimental advances and large-scale sequencing efforts have dramatically expanded our ability to detect and characterize functional RNA elements in viral genomes. This growing body of research is reshaping our understanding of viral strategies and uncovering new opportunities for therapeutic intervention. By focusing on RNA-mediated regulation across viral families, this Special Issue bridges molecular virology, RNA biology, and computational genomics.

Join the Special Issue

This Special Issue will serve as a timely and interdisciplinary resource for researchers working on RNA-based mechanisms in viruses. If you are investigating RNA structures, RNA–protein interactions, or computational RNA analysis in virology, we encourage you to submit your work.

Details & Submission

Join the Conversation

Questions about a potential submission or a related research idea are always welcome. You can get in touch via the contact form or connect with me on LinkedIn.

Exploring RNA Biology with Deep Learning Algorithms

2025-07-18T00:00:00+02:00

I will be serving as Guest Editor for the RNA Biology article collection “Exploring RNA Biology with Deep Learning Algorithms”. My aim is to curate articles that showcase how machine learning models can reveal hidden patterns in sequencing data, predict complex three dimensional RNA shapes with high accuracy and guide the design of novel RNA molecules for both research and therapeutic use.

Why This Collection Matters

RNA plays a central role in virtually every biological process, from gene regulation to the assembly of protein complexes. Yet its complexity poses formidable analytical challenges. Over the past few years, deep learning has transformed the way we decode protein structures and genomic data. Now, we stand on the edge of a similar revolution in RNA biology:

Uncover hidden patterns in high‑throughput sequencing data
Predict RNA modifications and their functional impacts
Model secondary & tertiary structures with unprecedented accuracy
Design synthetic RNAs for therapeutics and synthetic biology
Map RNA–protein interactions and regulatory switches at scale

What We’re Looking For

By pooling insights from biochemists, computational biologists and AI specialists, this special issue aims to chart the next frontier in RNA research. We welcome original research, methods papers and in‑depth reviews on topics including (but not limited to):

Deep learning for transcriptome analysis
AI models of RNA modifications
Secondary & tertiary structure prediction
AI‑driven RNA design & editing
RNA–protein interaction mapping
Automated annotation of RNA architectures

Advisory Panel

I’m pleased to be joined by two leading experts in RNA science as Guest Advisors:

Prof. Yiliang Ding (John Innes Centre): Pioneer of in vivo RNA structure mapping and chemical probing methods
Prof. Qiangfeng Cliff Zhang (Tsinghua University): Specialist in AI‑informed big‑data analysis

Join the Conversation

Questions about a potential submission or a related research idea are always welcome. You can reach me through the contact form or connect with me on LinkedIn.

Conserved RNA regulatory switches in living cells

2025-06-17T00:00:00+02:00

Most transcriptome-wide RNA structure studies still summarize each RNA with a single consensus structure. This paper tackles the more difficult and more realistic problem: many RNAs populate ensembles of alternative conformations, and some of those alternative states act as regulatory switches in living cells. The challenge is to recover such ensembles at transcriptome scale and distinguish functional structural heterogeneity from background noise.

The study combines two ingredients. First, transcriptome-wide MaP-based structure probing data are deconvolved with the DRACO algorithm to infer RNA secondary structure ensembles rather than single structures. Second, the resulting conformations are filtered with an automated conservation framework, DeConStruct, which uses covariation and comparative analysis to prioritize candidate regulatory structures. This combination makes it possible to move from transcriptome-wide structural profiling to the systematic discovery of conserved RNA switches.

In bacteria, the approach identified a substantial set of regions that populate two or more conformations in vivo and recovered known regulatory elements, confirming that the method can detect genuine structural switching behavior. More importantly, it uncovered several previously uncharacterized RNA thermometers in the 5' UTRs of cspG, cspI, cpxP, and lpxP, and then followed these candidates mechanistically during cold adaptation. In this context, the work also resolved a role for the CspE chaperone in regulating lpxP, making the paper a strong example of how ensemble mapping can lead to concrete functional hypotheses.

The eukaryotic part is equally notable. By introducing a dedicated 5'UTR-MaP strategy, the paper extends ensemble-scale RNA structure mapping into human 5' UTRs and identifies structural switches connected to differential open reading frame usage in transcripts such as CKS2 and TXNL4A. This is a useful reminder that RNA structure prediction becomes more powerful when it is treated as an ensemble problem rather than a single-structure problem, especially in regulatory regions where alternative conformations can alter translation behavior.

For computational RNA biology, this is an important paper because it brings together transcriptome-scale probing, ensemble deconvolution, and evolutionary support, which are often treated as separate layers. It therefore sits squarely in the RNA structure prediction space, but in a way that moves beyond static secondary structure models and toward experimentally anchored maps of regulatory structure dynamics in living cells.

Abstract

RNA molecules can populate ensembles of alternative structural conformations; however, comprehensively mapping RNA conformational landscapes within living cells presents notable challenges and has, as such, so far remained elusive. Here, we generate transcriptome-scale maps of RNA secondary structure ensembles in both Escherichia coli and human cells, uncovering features of structurally heterogeneous regions. By combining ensemble deconvolution and covariation analyses, we report the discovery of several bacterial RNA thermometers in the 5′ untranslated regions (UTRs) of the cspG, cspI, cpxP and lpxP mRNAs of Escherichia coli. We mechanistically characterize how these thermometers switch structure in response to cold shock and reveal the CspE chaperone-mediated regulation of lpxP. Furthermore, we introduce a method for the transcriptome-scale mapping of 5′ UTR structures in eukaryotes and leverage it to uncover RNA structural switches regulating the differential usage of open reading frames in the 5′ UTRs of the CKS2 and TXNL4A mRNAs in HEK293 cells. Collectively, this work reveals the complexity of RNA structural dynamics in living cells and provides a resource to accelerate the discovery of regulatory RNA switches.

Citation

Identification of conserved RNA regulatory switches in living cells using RNA secondary structure ensemble mapping and covariation analysis
Ivana Borovská, Chundan Zhang, Sarah-Luisa J. Dülk, Edoardo Morandi, Marta F. S. Cardoso, Billal M. Bourkia, Daphne A. L. van den Homberg, Michael T. Wolfinger, Willem A. Velema, Danny Incarnato
Nat. Biotechnol. (2025) | doi:10.1038/s41587-025-02739-0 | PDF

From structure to function in Musashi-RNA complexes

2025-01-01T00:00:00+01:00

Structural overview of Musashi-RNA complexes in regulatory and viral contexts

Musashi proteins sit at an interesting intersection of RNA biology, structural modeling, and disease relevance. They are sequence-selective RNA-binding proteins that regulate translation and cell-state decisions, but they have also emerged in viral contexts, most notably through proposed interactions with structured flaviviral RNAs. That combination makes Musashi a good example of a system where the mechanistic question is not just whether binding occurs, but how structural recognition translates into biological function.

This review brings together the computational work on that question. Rather than presenting a single new simulation pipeline, it synthesizes what molecular modeling, docking, molecular dynamics, and recent AI-assisted structure prediction have taught us about Musashi-RNA recognition. The focus is on how the two RNA-binding domains of Musashi engage short sequence motifs, how local RNA context modulates accessibility, and how these interactions can be interpreted in both endogenous regulatory RNAs and viral genomes.

One useful contribution of the article is that it connects several strands of prior work that are often read separately. On one side are studies of canonical Musashi recognition, where the problem is binding specificity and domain-level interaction geometry. On the other are virus-oriented analyses asking whether Musashi can plausibly recognize motifs embedded in flaviviral untranslated regions, and what that could imply for replication or neuropathology. The review argues that these are not isolated topics: the same structural principles have to explain both classes of observations.

For that reason, the paper is best read as a bridge between atomistic modeling and functional interpretation. It emphasizes that Musashi-RNA recognition depends on more than a short consensus motif in sequence space. RNA presentation, local fold, and dynamic rearrangement all influence whether a candidate site is likely to be bound in a biologically meaningful way. That is exactly where computational approaches remain useful: not as substitutes for experiment, but as tools for ranking plausible binding modes, testing structural hypotheses, and identifying which motifs deserve deeper validation.

The review also reflects a broader change in RNA structural biology. AlphaFold-class models, improved docking strategies, and longer-timescale simulation workflows have made it easier to generate mechanistic hypotheses for RNA-protein complexes, but the hard part remains interpretation. In the Musashi field, the important advance is not simply higher-confidence coordinates. It is the ability to move from static structural models toward more explicit explanations of specificity, competition, and context dependence in RNA recognition.

From my perspective, that is what makes this article worthwhile. It consolidates a line of work spanning Musashi binding to cellular RNAs, structural refinement of Musashi-RNA complexes, and possible links to viral pathogenesis. For readers interested in RNA-Protein interaction, 3D modeling, or the realistic use of AI in structural biology, it provides a compact map of the field and a clear rationale for where computation can genuinely add insight.

Citation

From Structure to Function: Computational Insights into Musashi-RNA Complexes in the Context of Viral Pathogenesis and Beyond
Nitchakan Darai, Leonhard Sidl, Thanyada Rungrotmongkol, Peter Wolschann, Michael T. Wolfinger
Sci. Asia 51S(1) 2025s013:1-10 (2025) | doi:10.2306/scienceasia1513-1874.2025.s013 | PDF

Bayesian approximation of RNA folding times

2025-01-01T00:00:00+01:00

Bayesian approximation of RNA folding times with KinPFN

This workshop paper presents the core idea behind KinPFN in a compact form: model RNA folding-time distributions directly, instead of recomputing them through thousands of simulator runs for every new case. The target quantity is the cumulative distribution of first passage times, a useful kinetic summary because it reflects how quickly different fractions of molecules reach a structure of interest.

The proposed solution uses prior-data fitted networks, trained not on a large corpus of experimentally measured RNAs, but on synthetic multi-modal distributions chosen to resemble the behavior of real folding-time data. Given just a few example folding times from a kinetics simulator, the model predicts the posterior distribution of folding times and from that the full cumulative distribution function. That makes the method especially attractive where only a small kinetic sample is available but a full distribution estimate would be useful.

What makes this interesting for RNA folding kinetics is the combination of speed and modularity. The method does not require rewriting the underlying simulator or abandoning physically motivated kinetic models. Instead, it acts as an approximation layer that reduces the number of expensive simulations needed for downstream interpretation. The workshop paper therefore functions as a proof of principle for integrating modern probabilistic machine learning into RNA kinetics workflows.

Relative to the full conference paper, this version is shorter and more focused on the central idea, but it already makes the key argument clearly: approximating folding-time distributions can be enough for many practical tasks, and those approximations can be learned efficiently from a synthetic prior. For anyone interested in the intersection of RNA folding kinetics and AI, this paper is a useful entry point into the broader KinPFN project.

It also connects directly to Why kinetic folding matters in RNA design, because many design decisions do not require a perfect kinetic simulation. They require a fast and credible way to compare whether one candidate is likely to behave more cleanly than another.

Abstract

RNA is a dynamic biomolecule with its function largely determined by its folding into complex structures. During the folding process, an RNA traverses through a series of intermediate structural states, with each transition occurring at variable rates that collectively influence the time required to reach the functional form. Understanding these folding kinetics is vital for predicting RNA behavior and optimizing applications in synthetic biology and drug discovery. While in silico kinetic RNA folding simulators are often computationally intensive and time-consuming, accurate approximations of the folding times can already be very informative to assess the efficiency of the folding process. Here, we present KinPFN, a novel approach that leverages prior-data fitted networks to directly model the posterior predictive distribution of RNA folding times. Trained on synthetic data representing arbitrary prior folding times, KinPFN efficiently approximates the cumulative distribution function of RNA folding times in a single forward pass, given only a few initial folding time examples. Our method offers a modular extension to RNA kinetics algorithms, promising significant computational speed-ups orders of magnitude faster, while achieving comparable results.

Citation

Bayesian Approximation of RNA Folding Times
Dominik Scheuer, Frederic Runge, Jörg K.H. Franke, Michael T. Wolfinger, Christoph Flamm, Frank Hutter
ICLR 2025 Workshop on AI for Nucleic Acids (2025) | doi:10.5281/zenodo.15228717 | PDF

KinPFN for RNA folding kinetics

2025-01-01T00:00:00+01:00

KinPFN training on a synthetic prior of RNA folding time distributions

RNA folding kinetics is often summarized in terms of first passage times: how long it takes a molecule to reach a target structure for the first time. Those distributions are informative because they capture more than a single mean folding time, but obtaining them requires many stochastic simulations and quickly becomes expensive. This paper asks whether the entire folding-time distribution can be approximated directly from a small number of example simulations.

The proposed answer is KinPFN, a prior-data fitted network trained on synthetic multi-modal distributions that mimic the shape of RNA first-passage-time distributions. Instead of learning from large collections of labeled RNA molecules, the model is trained to do approximate Bayesian inference on generic folding-time distributions. At application time, it receives only a few observed folding times as context and predicts the full cumulative distribution function in a single forward pass.

That framing is important. The method is not replacing physical simulators with a black box trained on a fixed benchmark set. It is designed as a fast probabilistic approximation layer that can sit on top of existing kinetics tools such as Kinfold and return a useful estimate of the folding-time distribution long before exhaustive simulation would finish. In practice, the paper reports speed-ups of at least 95% while preserving the overall shape of the distribution sufficiently well for downstream analysis.

The real value of the approach is that it makes distribution-level reasoning about folding kinetics much more accessible. Instead of reducing kinetics to a single summary statistic, one can compare broad versus narrow first-passage distributions, detect inefficient folding processes, and screen many more candidates than direct simulation alone would allow. The paper illustrates this in analyses of eukaryotic RNAs and folding-efficiency case studies, where the quality of the approximation is already good enough to be practically informative.

For RNA folding kinetics this is a useful conceptual shift. It shows that machine learning can accelerate kinetic analysis without having to learn RNA folding from scratch. By focusing on posterior approximation rather than direct structure prediction, KinPFN becomes a practical tool for kinetic RNA design workflows, where many candidate sequences need to be compared quickly but full simulation remains computationally restrictive.

That is also why it fits naturally with Why kinetic folding matters in RNA design. In both cases, the practical value comes from deciding which candidates deserve closer attention before a project becomes expensive.

Abstract

RNA is a dynamic biomolecule crucial for cellular regulation, with its function largely determined by its folding into complex structures, while misfolding can lead to multifaceted biological sequelae. During the folding process, RNA traverses through a series of intermediate structural states, with each transition occurring at variable rates that collectively influence the time required to reach the functional form. Understanding these folding kinetics is vital for predicting RNA behavior and optimizing applications in synthetic biology and drug discovery. While in silico kinetic RNA folding simulators are often computationally intensive and time-consuming, accurate approximations of the folding times can already be very informative to assess the efficiency of the folding process. In this work, we present KinPFN, a novel approach that leverages prior-data fitted networks to directly model the posterior predictive distribution of RNA folding times. By training on synthetic data representing arbitrary prior folding times, KinPFN efficiently approximates the cumulative distribution function of RNA folding times in a single forward pass, given only a few initial folding time examples. Our method offers a modular extension to existing RNA kinetics algorithms, promising significant computational speed-ups orders of magnitude faster, while achieving comparable results.

Citation

KinPFN: Bayesian Approximation of RNA Folding Kinetics using Prior-Data Fitted Networks
Dominik Scheuer, Frederic Runge, Jörg K.H. Franke, Michael T. Wolfinger, Christoph Flamm, Frank Hutter
The Thirteenth International Conference on Learning Representations (ICLR'25) (2025) | doi:10.5281/zenodo.15233965 | PDF

3' UTR-biased siRNA production in an insect-specific flavivirus

2024-10-15T00:00:00+02:00

Secondary structure plots of selected insect-specific flaviviruses

Mosquito antiviral immunity is dominated by RNA interference, so the distribution of viral siRNAs can be read as a footprint of how the host sees and processes viral RNA. In many flaviviruses, that footprint is fairly diffuse across the genome. This paper starts from the question of whether that is a general rule or whether different flavivirus groups leave different small-RNA signatures in mosquito cells. The answer is that most flaviviruses do behave broadly as expected, but classical insect-specific flaviviruses stand out sharply, and Kamiti River virus stands out even within that subset.

The main comparative result is the unusually strong 3' UTR bias of viral small interfering RNA, or vsiRNA, production in KRV. More than 95% of KRV-derived vsiRNAs map to the 3' UTR, which is remarkable both because of the scale of the enrichment and because KRV carries an unusually long, highly structured 3' UTR of roughly 1.2 kb. That immediately suggests that the structured non-coding end of the genome is not just a passive repository of regulatory motifs. It appears to dominate how the mosquito RNAi machinery encounters the virus.

That would already be interesting on its own, but the paper becomes much more compelling when it asks whether the obvious candidate explanation is actually correct. Because flavivirus 3' UTRs produce sfRNAs through XRN1 stalling at structured xrRNA elements, it would be easy to assume that the strong vsiRNA signal simply reflects abundant sfRNA production. In other words, one might expect the siRNAs to be coming from the same structured decay intermediates that are already familiar from flavivirus RNA biology.

The experiments show that this explanation is incomplete. For KRV, two major sfRNAs were mapped to predicted XRN1-resistant elements in the 3' UTR, and both species were abundant enough to be clear candidates for shaping the small-RNA profile. But when sfRNA production was reduced in Pacman-deficient mosquito cells, the 3' UTR-biased siRNA pattern did not collapse accordingly. That is the key result of the paper. The striking siRNA enrichment and the presence of sfRNAs coincide in the same region, yet the former does not depend on the latter in any simple causal way.

Methodologically, that is an important distinction. It means the 3' UTR signal cannot be explained away as a trivial by-product of one already known pathway. Biologically, it suggests that KRV or related classical insect-specific flaviviruses generate another RNA species, or another structural context, that is especially accessible to the mosquito RNAi machinery. The 3' UTR may therefore act less like a conventional protected fragment and more like a structured decoy or processing hotspot that redirects the antiviral response.

This fits well with a broader line of work on conserved RNA elements in flavivirus untranslated regions. The paper naturally connects to the comparative analysis of flavivirus 3' UTR architectures, to the broader mosquito-borne flavivirus 3' UTR synthesis, and to studies such as the Zambia insect-specific flavivirus xrRNA paper. Those earlier papers established that structured 3' UTRs are central to flavivirus evolution and host interaction. This study adds a new layer by showing that the same region can also dominate mosquito siRNA production in a way that is not simply reducible to sfRNA biogenesis.

For insect-specific flaviviruses, that makes the result especially valuable. These viruses offer a setting in which mosquito-virus interactions can be studied without the vertebrate half of the arbovirus cycle complicating the picture. KRV therefore becomes more than an odd small-RNA outlier. It becomes a model for asking how structured viral RNAs shape, divert, or absorb antiviral RNAi in arthropod hosts.

Abstract

RNA interference (RNAi) plays an essential role in mosquito antiviral immunity, but it is not known whether viral small interfering RNA (siRNA) profiles differ between mosquito-borne and mosquito-specific viruses. A pan-Orthoflavivirus analysis in Aedes albopictus cells revealed that viral siRNAs were evenly distributed across the viral genome of most representatives of the Flavivirus genus. In contrast, siRNA production was biased toward the 3' untranslated region (UTR) of the genomes of classical insect-specific flaviviruses (cISF), which was most pronounced for Kamiti River virus (KRV), a virus with a unique, 1.2 kb long 3' UTR. KRV-derived siRNAs were produced in high quantities and almost exclusively mapped to the 3' UTR. We mapped the 5' end of KRV subgenomic flavivirus RNAs (sfRNAs), products of the 5'−3' exoribonuclease XRN1/Pacman stalling on secondary RNA structures in the 3' UTR of the viral genome. We found that KRV produces high copy numbers of a long, 1,017 nt sfRNA1 and a short, 421 nt sfRNA2, corresponding to two predicted XRN1-resistant elements. Expression of both sfRNA1 and sfRNA2 was reduced in Pacman-deficient Aedes albopictus cells; however, this did not correlate with a shift in viral siRNA profiles. We suggest that cISFs, particularly KRV, developed a unique mechanism to produce high amounts of siRNAs as a decoy for the antiviral RNAi response in an sfRNA-independent manner.

Citation

Pan-flavivirus analysis reveals sfRNA-independent, 3’UTR-biased siRNA production from an Insect-Specific Flavivirus
Benoit Besson, Gijs J. Overheul, Michael T. Wolfinger, Sandra Junglen, Ronald P. van Rij
J. Virol. e01215-24 (2024) | doi:10.1128/jvi.01215-24 | Preprint PDF

Xinyang flavivirus and a likely tick-only orthoflavivirus clade

2024-05-29T00:00:00+02:00

Structural proteins of Xinyang flavivirus

This paper is interesting because it changes the comparative picture of tick-borne orthoflaviviruses in two ways at once. First, it adds a new geographic data point. Xinyang flavivirus was detected in Haemaphysalis flava ticks in China and groups with Mpulungu flavivirus from Zambia and Ngoye virus from Senegal. That means the basal clade defined earlier from African ticks is not a local curiosity. It extends at least into Asia and looks increasingly like a real, ecologically coherent lineage.

Second, the paper sharpens the argument that this lineage may be vertebrate-independent. Most tick-borne flaviviruses are discussed in terms of transmission cycles involving ticks and vertebrate hosts. XiFV, MPFV, and NGOV do not fit that expectation well. Their dinucleotide composition is closer to classical insect-specific flaviviruses than to vertebrate-infecting tick-borne flaviviruses, and XiFV shares with MPFV the absence of a furin cleavage site in prM that is otherwise common in vertebrate-associated relatives. The picture that emerges is not merely “another unusual flavivirus,” but a basal branch that may have settled on a tick-only life cycle.

That is where the structural work becomes important. The paper does not stop at phylogeny and host-range inference. It also examines the 3' UTR and shows that XiFV retains a recognizable orthoflaviviral RNA architecture, including multiple pseudoknot-containing stem-loops and a prototypical dumbbell element. In other words, even a lineage that may have abandoned vertebrate transmission still preserves a sophisticated structured non-coding region. That makes the system valuable for thinking about which RNA elements belong to the core flaviviral toolkit and which may be associated with particular host or transmission regimes.

This connects naturally to the earlier Mpulungu virus study, which first established the unusual African branch and identified distinctive xrRNA-like features in its 3' UTR. XiFV strengthens that story by showing that the lineage is broader than MPFV alone and by adding another genome against which the structural and ecological hypotheses can be tested. It also sits well beside the Zambia insect-specific flavivirus xrRNA work, where the emphasis was likewise on how non-coding RNA structure helps reveal deeper commonalities across understudied flavivirus groups.

What makes XiFV especially useful is that it sits at the intersection of discovery, comparative genomics, and evolutionary interpretation. The virus broadens the known distribution of this basal clade, supports the idea of a vertebrate-independent tick-only branch, and adds another structured 3' UTR to the small but growing set of non-coding regions that can be compared across unusual orthoflaviviruses. That combination makes the paper more than a geographic range extension. It is a clearer statement that the ecological diversity of tick-borne flaviviruses is wider than the classical vertebrate-centered model suggests.

It also fits naturally with When sequence conservation is not enough to find functional RNA structure, because XiFV reinforces how much of the informative signal sits at the level of conserved architecture rather than straightforward sequence identity.

Abstract

Tick-borne orthoflaviviruses (TBFs) are classified into three conventional groups based on genetics and ecology: mammalian, seabird and probable-TBF group. Recently, a fourth basal group has been identified in Rhipicephalus ticks from Africa: Mpulungu flavivirus (MPFV) in Zambia and Ngoye virus (NGOV) in Senegal. Despite attempts, isolating these viruses in vertebrate and invertebrate cell lines or intracerebral injection of newborn mice with virus-containing homogenates has remained unsuccessful. In this study, we report the discovery of Xinyang flavivirus (XiFV) in Haemaphysalis flava ticks from Xìnyáng, Henan Province, China. Phylogenetic analysis shows that XiFV was most closely related to MPFV and NGOV, marking the first identification of this tick orthoflavivirus group in Asia. We developed a reverse transcriptase quantitative PCR assay to screen wild-collected ticks and egg clutches, with absolute infection rates of 20.75% in adult females and 15.19% in egg clutches, suggesting that XiFV could be potentially spread through transovarial transmission. To examine potential host range, dinucleotide composition analyses revealed that XiFV, MPFV and NGOV share a closer composition to classical insect-specific orthoflaviviruses than to vertebrate-infecting TBFs, suggesting that XiFV could be a tick-only orthoflavivirus. Additionally, both XiFV and MPFV lack a furin cleavage site in the prM protein, unlike other TBFs, suggesting these viruses might exist towards a biased immature particle state. To examine this, chimeric Binjari virus with XIFV- prME (bXiFV) was generated, purified and analysed by SDS-PAGE and negative-stain transmission electron microscopy, suggesting prototypical orthoflavivirus size (~50 nm) and bias towards uncleaved prM. In silico structural analyses of the 3'-untranslated regions show that XiFV forms up to five pseudo-knot-containing stem-loops and a prototypical orthoflavivirus dumbbell element, suggesting the potential for multiple exoribonuclease-resistant RNA structures.

Citation

Xinyang flavivirus, from Haemaphysalis flava ticks in Henan province, China, defines a basal, likely tick-only flavivirus clade
Lan-Lan Wang, Qia Cheng, Natalee D. Newton, Michael T. Wolfinger, Mahali S. Morgan, Andrii Slonchak, Alexander A. Khromykh, Tian-Yin Cheng, Rhys H. Parry
J. Gen. Virol. 105(5) (2024) | doi:10.1099/jgv.0.001991 | PDF

Automated lineage designation from viral genomic data

2024-02-12T00:00:00+01:00

Automated lineage designation of Venezuelan Equine Encephalitis complex viruses

Lineage designation is partly a biological question and partly an information-management problem. Once viral sequencing reaches very large scale, as it did for SARS-CoV-2, manual or community-curated naming systems become increasingly difficult to maintain. Delays accumulate, criteria become harder to apply consistently, and the workload grows faster than the nomenclature process can keep up.

This study presents Autolin, a heuristic framework for automated lineage designation from phylogenetic and genomic data. The main contribution is not a claim that expert curation should disappear, but that a simple and explicit rule-based system can produce lineage assignments at a scale that is difficult to sustain manually.

That matters because lineage systems are only useful when they are both interpretable and sustainable. An automated framework can evaluate very large trees, apply the same criteria repeatedly, and generate designations without depending on ad hoc proposal cycles. In practice, that means faster turnaround and more consistent behavior across datasets with millions of sequences.

Another useful feature of the framework is that it allows prioritization of particular mutations or genes. That makes the method flexible enough to reflect biological or epidemiological priorities rather than treating all sequence variation as equally informative. In other words, the system is not just scalable. It can also be tuned to highlight the parts of a genome that matter most for a given pathogen or surveillance context.

The paper is careful about scope. Automated lineage designation is not the same thing as biological interpretation, and no heuristic can remove the need for expert judgment altogether. What it can do is provide a consistent baseline classification system that remains usable as genomic datasets continue to grow. The fact that the method produces lineage partitions similar to existing curated systems across multiple viruses makes that claim much more credible.

For genomic epidemiology, this is the real value of the framework. It gives researchers a practical way to maintain phylogeny-based nomenclature under conditions where purely manual designation becomes increasingly fragile. That is useful not only for SARS-CoV-2, but also for other rapidly sampled viral systems where scale has already become the defining constraint.

This is also one of the clearest cases in which automation addresses a real bottleneck without pretending to resolve molecular mechanism by itself.

Abstract

Pathogen lineage nomenclature systems are a key component of effective communication and collaboration for researchers and public health workers. Since February 2021, the Pango dynamic lineage nomenclature for SARS-CoV-2 has been sustained by crowdsourced lineage proposals as new isolates were sequenced. This approach is vulnerable to time-critical delays as well as regional and personal bias. Here we developed a simple heuristic approach for dividing phylogenetic trees into lineages, including the prioritization of key mutations or genes. Our implementation is efficient on extremely large phylogenetic trees consisting of millions of sequences and produces similar results to existing manually curated lineage designations when applied to SARS-CoV-2 and other viruses including chikungunya virus, Venezuelan equine encephalitis virus complex and Zika virus. This method offers a simple, automated and consistent approach to pathogen nomenclature that can assist researchers in developing and maintaining phylogeny-based classifications in the face of ever-increasing genomic datasets.

Citation

A framework for automated scalable designation of viral pathogen lineages from genomic data
Jakob McBroome, Adriano de Bernardi Schneider, Cornelius Roemer, Michael T. Wolfinger, Angie S. Hinrichs, Aine N. O’Toole, Chris Ruis, Yatish Turakhia, Andrew Rambaut, and Russell Corbett-Detig
Nature Microbiol. 9:550–560 (2024) | doi:10.1038/s41564-023-01587-5 | PDF

What virus bioinformatics can and cannot tell us about RNA viruses

2023-11-03T00:00:00+01:00

Virus bioinformatics is most useful when it answers a concrete question. Which viral lineages are circulating? Which regions of a genome are conserved? Where do we see evidence for structured RNA elements, and where are we only looking at weak computational hints? Those are the questions that make computational analysis valuable for virology.

For RNA viruses in particular, computational work often sits at the interface of comparative genomics, RNA structure analysis, and molecular epidemiology. It can reveal patterns that are difficult to see from individual experiments alone, but it also has clear limits. Sequence conservation is not the same as functional validation, and a predicted structure is not yet a mechanism.

Comparative Genomics First

The starting point for many useful analyses is still comparative genomics. With enough sequence diversity, alignments and phylogenetic reconstructions can show which parts of a viral genome are stable across isolates, which lineages are diverging, and where unusual patterns deserve closer inspection.

In practical terms, this means we can use viral genome collections to identify conserved untranslated regions, recurring sequence motifs, lineage-specific variation, and genomic segments worth testing experimentally. For fast-moving viruses, this kind of analysis is often the difference between anecdotal interpretation and a reproducible view of the sequence landscape.

Where RNA Structure Prediction Helps

RNA viruses do not just encode proteins. They also encode structure. Terminal regions, internal regulatory elements, long-range interactions, and exoribonuclease-resistant RNAs can all shape replication, translation, packaging, or immune evasion. That makes RNA structure prediction a useful layer on top of sequence analysis.

What computation does well is narrowing the search space. It can help prioritize candidate structures, compare homologous elements across related viruses, estimate how mutations may perturb an ensemble, and point to regions where compensatory change or unusual conservation suggests functional constraint.

What it does less well is prove function on its own. A low free energy structure, even a highly plausible one, does not establish biological relevance unless it is supported by comparative evidence, biochemical data, or perturbation experiments. For that reason, structure prediction is best treated as a hypothesis-generating tool that becomes much more powerful when combined with probing data, mutational analysis, or virological assays.

Genomic Epidemiology Is About Context

Genomic epidemiology adds another dimension: time, geography, and lineage history. Instead of asking only what a genome contains, it asks how genomes are changing across outbreaks, hosts, and regions. For many viral systems, that context is essential for interpreting sequence variation sensibly.

This is particularly important when sequence changes overlap with structured RNA regions. A mutation that looks minor at the amino-acid level can have a larger effect on local RNA folding, and the opposite is also true. Bringing phylogenetic and structural perspectives together is one of the more interesting strengths of virus bioinformatics.

What Good Virus Bioinformatics Looks Like

In my view, good computational work on RNA viruses has three properties:

It starts with a clearly defined biological question.
It distinguishes observation from interpretation.
It makes it easy to see which claims are supported by comparative data, which by structural modeling, and which still require experiment.

That standard matters because viral RNA analyses are easy to oversell. It is tempting to move too quickly from sequence conservation to function, or from predicted structure to therapeutic relevance. The more useful approach is slower and more explicit: identify a signal, test whether it is robust, and then decide what biological interpretation is warranted.

Why This Matters

For RNA virology, computation is no longer just a supporting activity. It is often the layer that connects large public sequence collections, mechanistic RNA hypotheses, and experimental prioritization. Done carefully, it helps us decide which genomes to compare, which structures to test, and which lineage-specific patterns are worth following up.

That is also where a broader One Health perspective becomes genuinely relevant. When viruses move across hosts, vectors, and ecological settings, sequence data alone are not enough and isolated mechanistic results are not enough either. The useful view comes from connecting evolutionary context with molecular interpretation.

A conserved G-quadruplex in the Zika virus 3' terminal region

2023-10-23T00:00:00+02:00

G-Quadruplex in the terminal region of the Zika virus genome

Zika virus (ZIKV) is a positive-strand flavivirus whose untranslated terminal regions contain structured RNA elements with likely regulatory roles. In this study, we focused on a G-rich segment in the 3' terminal region and asked two fairly specific questions: is there convincing evidence that this segment forms a conserved G-quadruplex, and how does that element interact with host helicases?

G-quadruplexes are structurally distinct RNA elements formed by stacked guanine quartets. In viral RNAs they are interesting because they may compete with alternative conformations, alter accessibility, or modulate how host factors engage a structured region. That does not make every predicted G4 biologically relevant, but it does make them worth testing when the underlying sequence signal is conserved.

The central result is that the candidate ZIKV G4 is not just a short-oligo artifact. Across available ZIKV isolates, the G-rich segment is conserved, and biochemical assays support G4 formation in the context of a larger 3' terminal region transcript. Pyridostatin and BG4 binding provided experimental support for that interpretation. This moves the discussion from sequence speculation to a more defensible structural observation.

The second important result is that human DEAD-box helicases interact with this region. Both DDX3X132-607 and DDX17135-555 bind the 3' terminal region, and the DDX17 construct used in the study unfolds the G4. Mechanistically, that is interesting because it suggests the local RNA ensemble is not static. Host proteins may reshape this part of the viral RNA, potentially shifting the balance between alternative conformations or affecting downstream processes that depend on terminal-region architecture.

What the study does not show, however, is just as important. It does not establish that the G4 is required for viral replication in cells, nor does it show that DDX17 binding or unwinding is directly druggable in a useful therapeutic sense. Those are hypotheses that may follow from the data, but they are not conclusions supported by the present experiments. For a system like this, the right next steps are functional perturbation, context-specific virology, and direct tests of how the structured region behaves during infection.

This distinction matters because structured viral RNAs are easy to oversell. A conserved element plus a host-factor interaction is already a meaningful result. It tells us there is a plausible structured RNA feature in the ZIKV 3' terminal region and that at least one host helicase can remodel it. That is strong mechanistic groundwork, even before any translational claims enter the picture.

For RNA virology more broadly, this is exactly where computational and biochemical analysis work well together. Comparative sequence analysis identifies conserved candidates, structural assays test whether those candidates persist in larger transcript contexts, and protein-binding experiments begin to reveal how host factors may shape the RNA landscape. That combined view is much more informative than any one layer on its own.

Abstract

Zika virus (ZIKV) infection remains a worldwide concern, and currently no effective treatments or vaccines are available. Novel therapeutics are an avenue of interest that could probe viral RNA-human protein communication to stop viral replication. One specific RNA structure, G-quadruplexes (G4s), possess various roles in viruses and all domains of life, including transcription and translation regulation and genome stability, and serves as nucleation points for RNA liquid-liquid phase separation. Previous G4 studies on ZIKV using a quadruplex forming G-rich sequences Mapper located a potential G-quadruplex sequence in the 3′ terminal region (TR) and was validated structurally using a 25-mer oligo. It is currently unknown if this structure is conserved and maintained in a large ZIKV RNA transcript and its specific roles in viral replication. Using bioinformatic analysis and biochemical assays, we demonstrate that the ZIKV 3′ TR G4 is conserved across all ZIKV isolates and maintains its structure in a 3′ TR full-length transcript. We further established the G4 formation using pyridostatin and the BG4 G4-recognizing antibody binding assays. Our study also demonstrates that the human DEAD-box helicases, DDX3X132-607 and DDX17135-555, bind to the 3′ TR and that DDX17135-555 unfolds the G4 present in the 3′ TR. These findings provide a path forward in potential therapeutic targeting of DDX3X or DDX17’s binding to the 3′ TR G4 region for novel treatments against ZIKV.

Citation

The 3’ terminal region of Zika virus RNA contains a conserved G-quadruplex and is unfolded by human DDX17
Danielle L. Gemmill, Corey R. Nelson, Maulik D. Badmalia, Higor S. Pereira, Michael T. Wolfinger, and Trushar Patel
Biochem. Cell Biol. 102(1):96–105 (2024) | doi:10.1139/bcb-2023-0036 | PDF

RNA-protein complex refinement using AI modeling and docking

2023-06-09T00:00:00+02:00

Association complex of Musashi RBD1 and RBD with a target RNA

Modeling protein-RNA complexes remains difficult even when reasonably good structures for the individual components are available. The hardest part is often not generating a starting model, but refining the interface in a way that captures flexible RNA segments and produces a biologically plausible binding geometry.

This study addresses that problem with a two-step workflow. First, AlphaFold2 was used to generate a structural model for the RNA-binding domains of the human Musashi-1 (MSI1) protein. Second, the resulting model was refined in the presence of RNA using flexible docking based on parallel cascade selection molecular dynamics (PaCS-MD). The point of the method is not simply to place RNA near a protein surface, but to sample interface rearrangements that matter for complex formation.

Musashi-1 is a useful test case because its RNA recognition has been studied experimentally, which means the resulting models can be checked against known interaction patterns. In the refined complexes, the analysis recovered a core set of residues and nucleotides that are consistent with previous work on MSI1-RNA recognition. That matters more than raw structural novelty. A refinement method is only useful if it recovers contacts that make biochemical sense.

Compared with a more standard template-based workflow built around Phyre2, the PaCS-MD approach produced better-supported association complexes in this system. The main reason is that enhanced sampling gives flexible RNA regions more room to explore realistic conformations during docking, instead of forcing the final model to depend too heavily on manual assembly or rigid starting assumptions.

The method still has clear limits. It depends on having useful initial structural information, and the quality of the final complex remains tied to the quality of both the starting model and the sampling protocol. This is not a general solution to protein-RNA structure prediction from sequence alone. It is a refinement strategy that becomes valuable when there is already enough structural context to make flexible docking meaningful.

That makes the study best understood as a methods contribution. It shows that AI-derived protein models can be combined with enhanced-sampling docking to improve protein-RNA complex refinement, at least for systems like MSI1 where independent evidence exists for the binding interface. For researchers working on RNA-binding proteins, this is a more realistic and useful claim than broad promises about drug discovery.

Abstract

An efficient structural refinement technique for protein-RNA complexes is proposed based on a combination of AI-based modeling and flexible docking. Specifically, an enhanced sampling method called parallel cascade selection molecular dynamics (PaCS-MD) was extended to include flexible docking to construct protein-RNA complexes from those obtained by AI-based modeling (AlphaFold2). With the present technique, the conformational sampling of flexible RNA regions is accelerated by PaCS-MD, enabling one to construct plausible models for protein-RNA complexes. For demonstration, PaCS-MD constructed several protein-RNA complexes of the RNA-binding Musashi-1 (MSI1) family of proteins, which were validated by comparing a group of crucial residues for RNA-binding with experimental complexes. Our analyses suggest that PaCS-MD improves the quality of complex modeling compared to the standard protocol based on template-based modeling (Phyre2). Furthermore, PaCS-MD could also be a beneficial technique for constructing complexes of non-native RNA-binding to proteins.

Citation

A Structural Refinement Technique for Protein-RNA Complexes Using a Combination of AI-based Modeling and Flexible Docking: A Study of Musashi-1 Protein
Nitchakan Darai, Kowit Hengphasatporn, Peter Wolschann, Michael T. Wolfinger, Yasuteru Shigeta, Thanyada Rungrotmongkol, Ryuhei Harada
B. Chem. Soc. Jpn. (2023)

RNA–RNA interaction analysis of Japanese encephalitis virus

2023-03-31T00:00:00+02:00

Graphical abstract

Long-range RNA-RNA interactions are central to the replication cycle of many flaviviruses, but they are not always easy to characterize directly. In Japanese encephalitis virus (JEV), the 5' and 3' terminal regions are expected to interact through conserved cyclization elements, yet the strength, stoichiometry, and structural behavior of that interaction still needed to be tested experimentally.

This study approaches the problem from both sides. Computational analysis was used to identify the most plausible interaction site and to evaluate whether the conserved cyclization sequence is kinetically favored over competing alternatives such as homodimer formation. The predicted interaction was then tested biophysically using SEC-MALS, analytical ultracentrifugation, microscale thermophoresis, and SAXS.

The result is a stronger case for the long-range interaction than computation alone could provide. The 5' and 3' terminal regions interact directly with nanomolar affinity and a 1:1 stoichiometry, and that interaction weakens substantially when the conserved cyclization sequence is disrupted. That is a useful mechanistic result because it ties the predicted long-range pairing to an experimentally measurable binding event.

Just as importantly, the kinetic analysis supports the idea that the cyclization sequence is not merely one possible pairing among many. It appears to be the dominant interaction pathway under the conditions examined, with an advantage over competing self-association states. For flavivirus biology, that matters because genome cyclization is closely tied to replication, and the computational ranking of alternative structures becomes much more informative once it is backed by binding measurements.

The SAXS-based structural model adds another piece of the picture. It does not deliver atomic resolution, but it does suggest that the complex is flexible while still maintaining a stable overall association in solution. That kind of low-resolution structural information is valuable when interpreting long viral RNAs, where conformational heterogeneity is often part of the biology rather than a nuisance to be averaged away.

More broadly, this paper is a good example of how RNA-RNA interaction studies should be done. Sequence analysis alone is too weak, and biophysical characterization without a clear computational hypothesis is inefficient. The combination is what makes the result persuasive: comparative and thermodynamic reasoning narrow the search space, and orthogonal biophysical methods test whether the predicted interaction actually exists and behaves as expected.

Abstract

Numerous viruses utilize essential long-range RNA–RNA genome interactions, specifically flaviviruses. Using Japanese encephalitis virus (JEV) as a model system, we computationally predicted and then biophysically validated and characterized its long-range RNA–RNA genomic interaction. Using multiple RNA computation assessment programs, we determine the primary RNA–RNA interacting site among JEV isolates and numerous related viruses. Following in vitro transcription of RNA, we provide, for the first time, characterization of an RNA–RNA interaction using size-exclusion chromatography coupled with multi-angle light scattering and analytical ultracentrifugation. Next, we demonstrate that the 5′ and 3′ terminal regions of JEV interact with nM affinity using microscale thermophoresis, and this affinity is significantly reduced when the conserved cyclization sequence is not present. Furthermore, we perform computational kinetic analyses validating the cyclization sequence as the primary driver of this RNA–RNA interaction. Finally, we examined the 3D structure of the interaction using small-angle X-ray scattering, revealing a flexible yet stable interaction. This pathway can be adapted and utilized to study various viral and human long-non-coding RNA–RNA interactions and determine their binding affinities, a critical pharmacological property of designing potential therapeutics.

Figures and Data

Citation

Investigating RNA-RNA interactions through computational and biophysical analysis
Tyler Mrozowich, Sean Park, Maria Waldl, Amy Henrickson, Scott Tersteeg, Corey R. Nelson, Anneke De Klerk, Borries Demeler, Ivo L. Hofacker, Michael T. Wolfinger, Trushar R. Patel
Nucleic Acids Res. gkad223 (2023) | doi:10.1093/nar/gkad223 | PDF | Supplement | Figures

Strukturierte RNAs in Viren (in German)

2023-03-23T00:00:00+01:00

Schematic representation of xrRNA exoribonuclease stalling

This article is a short review, but it covers a central idea that has become increasingly important across virology: many viral genomes do not just encode proteins, they also encode structured RNA elements that actively control the infection cycle. In flaviviruses, the 3'UTR is a particularly rich example because it contains conserved folds that regulate replication, translation, host adaptation, and immune evasion.

The focus here is on exoribonuclease-resistant RNAs, or xrRNAs. These elements are interesting because they do not merely bind a host factor or present a passive recognition site. They act mechanically. Their three-dimensional fold stalls the cellular XRN1 exoribonuclease and thereby generates subgenomic flaviviral RNAs (sfRNAs). Those sfRNAs are now known to contribute to pathogenicity, host-range effects, and immune modulation in multiple flavivirus groups.

What makes xrRNAs especially attractive from a structural-biology perspective is that they illustrate RNA function in a very direct way. Their activity depends on conserved topology rather than simple sequence identity. That means they are an ideal example of why RNA structure conservation matters in virology: two viral RNAs can diverge substantially at sequence level while preserving a fold that continues to perform the same mechanistic task.

The review therefore works on two levels. At the biological level, it summarizes why structured 3'UTR elements matter for the viral life cycle and how xrRNAs help viruses manipulate host RNA decay pathways. At the conceptual level, it argues that RNA structures should be treated as evolvable functional modules, not just decorative features of untranslated regions. That framing connects this short piece to a much broader body of comparative work on conserved viral RNAs.

Another useful aspect of the article is that it points beyond natural virus biology toward design. If an RNA fold can reliably block an exoribonuclease in nature, then the same principle can potentially be repurposed in synthetic systems. That is where the connection to RNA design and future therapeutic applications comes in. xrRNAs are interesting not only because they explain flavivirus biology, but because they may serve as modular regulatory parts in engineered RNAs.

This is also why the topic has become relevant in the context of RNA therapeutics and mRNA technology. Structured RNA elements that control stability or degradation are obvious candidates for engineering transcript lifetime and behavior. The point is not that viral xrRNAs can simply be copied into therapeutic constructs without modification, but that they provide a proven natural design principle for mechanically active RNA control.

The piece functions as a compact bridge between comparative viral RNA structure analysis, xrRNA discovery, and later efforts in synthetic and mechanically active RNA design. Even though the article itself is shorter and more magazine-style than a primary research paper, the underlying concept is foundational and deserves more context than a brief teaser. Readers who want to follow that thread into primary research can move from here to Discoveries of Exoribonuclease-Resistant Structures of Insect-Specific Flaviviruses Isolated in Zambia or to Rational design of mechanically active RNAs.

For a more general discussion of why sequence similarity alone often fails to capture these conserved viral RNA elements, see When sequence conservation is not enough to find functional RNA structure.

Figures and Data

Citation

Strukturierte RNAs in Viren (in German)
Roman Ochsenreiter, Michael T. Wolfinger
Biospektrum 29(2):156-158 (2023) | | doi:10.1007/s12268-023-1907-x | PDF | Figures

The link between mRNA vaccine design and barbeque optimization

2022-11-18T00:00:00+01:00

I have been discussing questions around the interface of RNA bioinformatics and Data with Jonas Rashedi in his My Data is better than yours podcast.

RNA recognition by Musashi-1

2022-07-26T00:00:00+02:00

Association complexes of Musashi-1 RBD1 and RBD2 with the canonical target RNA GUAGU

Musashi-1 (MSI1) is an RNA-binding protein involved in stem-cell maintenance, neural development, and post-transcriptional regulation. It also became particularly interesting in the context of Zika virus biology, because Musashi proteins were proposed to interact with viral RNAs in ways that could affect replication and neuropathology. That makes the underlying recognition problem more than a narrow structural question. Understanding how Musashi binds RNA is relevant both for endogenous regulation and for virus-host interaction studies.

This paper focuses on that recognition step at the level of individual binding domains. MSI1 contains two RNA-binding domains, RBD1 and RBD2, both of which recognize short single-stranded RNA motifs centered on a UAG core. The main question is how specific that recognition really is. Sequence logos and motif descriptions are useful, but they do not explain why some candidate pentamers bind more strongly than others or which contacts stabilize the preferred complexes.

The methodological approach is straightforward but well chosen for this kind of problem. The study starts from structural models for the Musashi RNA-binding domains and combines those with several candidate RNA pentamers. These protein-RNA complexes are then evaluated using atomistic molecular dynamics simulations in explicit solvent, followed by binding free-energy estimates using the solvated interaction energy framework. In practice, that means the paper does not rely on a single static docking pose. It asks whether the complexes remain stable over time and whether the interaction energies consistently favor some RNA motifs over others.

An additional point of interest is the use of AlphaFold2 to evaluate the protein side of the system. At the time, this was a useful check on whether predicted Musashi domain structures were close enough to experimentally determined conformations to support downstream simulation. The answer was essentially yes. The predicted structures aligned well with known domain architectures, which made it reasonable to use them as a basis for analyzing RNA recognition.

The main finding is that Musashi binding is selective even among short, closely related RNA pentamers. Out of the four tested motifs, two showed substantially stronger calculated binding to MSI1 RBD1 and RBD2 than the others. The simulations trace that difference back to a more favorable network of stacking, hydrogen-bonding, and electrostatic interactions around the central recognition motif. The work moves the discussion from "Musashi prefers UAG-containing RNAs" to a more specific structural claim about why some local sequence contexts are better accommodated than others.

That is what makes the paper useful. It does not claim to solve RNA-protein recognition in general, and it does not replace experiment. It does provide a mechanistic picture of how Musashi domains engage short RNA targets and how binding specificity can be compared systematically across candidate motifs. For systems where direct structural data are still incomplete, that kind of analysis is often the right intermediate step between motif discovery and functional testing.

This study also laid some groundwork for later Musashi modeling efforts. Once domain-level recognition preferences are characterized at this level, it becomes easier to ask larger questions about Musashi binding to structured cellular RNAs or to viral untranslated regions, and to develop more detailed protein-RNA complex refinement workflows. In that sense, this paper is best viewed as the starting point of a broader Musashi-RNA modeling program rather than a standalone binding-energy exercise.

Abstract

The Musashi (MSI) family of RNA-binding proteins, comprising the two homologs Musashi-1 (MSI1) and Musashi-2 (MSI2), typically regulates translation and is involved in cell proliferation and tumorigenesis. MSI proteins contain two ribonucleoprotein-like RNA-binding domains, RBD1 and RBD2, that bind single-stranded RNA motifs with a central UAG trinucleotide with high affinity and specificity. The finding that MSI also promotes the replication of Zika virus, a neurotropic Flavivirus, has triggered further investigations of the biochemical principles behind MSI–RNA interactions. However, a detailed molecular understanding of the specificity of MSI RBD1/2 interaction with RNA is still missing. Here, we performed computational studies of MSI1–RNA association complexes, investigating different RNA pentamer motifs using molecular dynamics simulations with binding free energy calculations based on the solvated interaction energy method. Simulations with Alphafold2 suggest that predicted MSI protein structures are highly similar to experimentally determined structures. The binding free energies show that two out of four RNA pentamers exhibit a considerably higher binding affinity to MSI1 RBD1 and RBD2, respectively. The obtained structural information on MSI1 RBD1 and RBD2 will be useful for a detailed functional and mechanistic understanding of this type of RNA–protein interactions.

Figures and Data

Citation

Theoretical studies on RNA recognition by Musashi 1 RNA–binding protein
Nitchakan Darai, Panupong Mahalapbutr, Peter Wolschann, Vannajan Sanghiran Lee, Michael T. Wolfinger, Thanyada Rungrotmongkol
Sci. Rep. 12:12137 (2022) | doi:10.1038/s41598-022-16252-w | PDF

Hfq, Crc, and antibiotic resistance in P. aeruginosa

2022-06-23T00:00:00+02:00

Schematic of the mexGHI-opmD operon downregulation by Hfq during carbon catabolite repression

This paper extends the Hfq/Crc/CrcZ story in Pseudomonas aeruginosa from single regulons to a dynamic physiological transition. The biological setting is diauxic growth: the bacterium first consumes a preferred carbon source and then rewires its metabolism once that source is depleted. In Pseudomonas, that transition is controlled in large part by carbon catabolite repression (CCR), with Hfq and Crc repressing many transcripts during growth on preferred substrates and the regulatory RNA CrcZ relieving that repression once the cell shifts to alternative nutrients.

The paper asks what that transition looks like globally and whether it has consequences beyond nutrient utilization. To answer that, the study measures the transcriptome, translatome, and proteome in parallel during and after relief of CCR. That multi-omics design is the real strength of the work. It allows the authors to distinguish changes that occur at RNA abundance, translational efficiency, and protein output, rather than treating "gene expression" as a single layer.

The main mechanistic result is that the mexGHI-opmD operon is upregulated after CCR is relieved, which in turn lowers susceptibility to norfloxacin. This is important because the operon encodes an efflux system with direct consequences for antibiotic response. The paper therefore shows that the shift from preferred to non-preferred carbon sources does not just alter metabolism. It also changes the antimicrobial phenotype of the cell.

That observation fits naturally into a broader line of work on how Hfq, Crc, and CrcZ link metabolism to RNA control in Pseudomonas. The mechanistic basis is laid out in Interplay Between the Catabolite Repression Control Protein Crc, Hfq and RNA in Hfq-Dependent Translational Regulation in Pseudomonas aeruginosa, and the link to carbapenem uptake is developed further in Distinctive Regulation of Carbapenem Susceptibility in Pseudomonas aeruginosa by Hfq. This paper pushes that logic one step further by showing how the same network indirectly reshapes resistance-relevant output during diauxic growth. Nutrient-state sensing and antibiotic susceptibility are tightly coupled through the same RNA-centered regulatory architecture.

The word "indirect" in the title matters. The study does not claim that Hfq binds the mexGHI-opmD operon in a simple one-step regulatory interaction. Instead, the data support a more distributed model in which relief of CCR changes the allocation and activity of Hfq-dependent control, and the effect on the efflux pump emerges from that broader rewiring. That makes the paper more interesting than a straightforward target-identification study, because it emphasizes network-level consequences of post-transcriptional regulation.

From a practical perspective, the result is also useful because it reminds us that antibiotic susceptibility can depend strongly on physiological state. The same bacterium can present a different drug-response profile depending on which nutrients it has consumed and which regulatory program it has entered. That is exactly the kind of context dependence that often complicates antimicrobial treatment and laboratory interpretation.

Methodologically, the paper is a good example of how multi-omics becomes genuinely informative when tied to a clear transition state. Sampling during and after relief of CCR gives the authors a biologically meaningful perturbation, and the combined transcriptome-translatome-proteome view makes it possible to see which responses are broad and which are more specifically post-transcriptional. For readers interested in bacterial RNA regulation, that is a major part of the value.

Taken together with the 2018 and 2020 studies, this paper makes the connection between metabolism, RNA control, and resistance phenotypes especially clear. It shows how the same regulatory machinery is deployed during a physiological growth transition and how that deployment feeds into efflux-mediated drug response.

Abstract

In Pseudomonas aeruginosa, the RNA chaperone Hfq and the catabolite repression protein Crc act in concert to regulate numerous genes during carbon catabolite repression (CCR). After alleviation of CCR, the RNA CrcZ sequesters Hfq/Crc, which leads to a rewiring of gene expression to ensure the consumption of less preferred carbon and nitrogen sources. Here, we performed a multiomics approach by assessing the transcriptome, translatome, and proteome in parallel in P. aeruginosa strain O1 during and after relief of CCR. As Hfq function is impeded by the RNA CrcZ upon relief of CCR, and Hfq is known to impact antibiotic susceptibility in P. aeruginosa, emphasis was laid on links between CCR and antibiotic susceptibility. To this end, we show that the mexGHI-opmD operon encoding an efflux pump for the antibiotic norfloxacin and the virulence factor 5-Methyl-phenazine is upregulated after alleviation of CCR, resulting in a decreased susceptibility to the antibiotic norfloxacin. A model for indirect regulation of the mexGHI-opmD operon by Hfq is presented.

Citation

Rewiring of Gene Expression in Pseudomonas aeruginosa During Diauxic Growth Reveals an Indirect Regulation of the MexGHI-OpmD Efflux Pump by Hfq
Marlena Rozner, Ella Nukarinen, Michael T. Wolfinger, Fabian Amman, Wolfram Weckwerth, Udo Blaesi, Elisabeth Sonnleitner
Front. Microbiol. (2022) 13:919539 | doi:10.3389/fmicb.2022.919539 | PDF

Biophysical characterization of human lincRNA-p21 Alu inverted repeats

2022-01-20T00:00:00+01:00

Structural characterization of sense and antisense lincRNA-p21 Alu inverted repeats

Human lincRNA-p21 is a regulatory long non-coding RNA in the p53 pathway and plays an important role in apoptosis. Its activity depends in part on its interaction with the RNA-binding protein hnRNP-K, and that interaction is likely shaped by RNA tertiary structure rather than sequence alone. Earlier work had already defined the secondary structure of the sense and antisense AluSx1 inverted repeats, but the three-dimensional organization of these RNA domains remained unclear.

This study closes that gap with an integrated biophysical and computational workflow. The sense and antisense AluSx1 regions were transcribed in vitro and first checked for sample quality and homogeneity using size-exclusion chromatography, analytical ultracentrifugation, and SEC-MALS. That part matters because long RNAs are structurally heterogeneous by default, and any useful SAXS model depends on showing that the sample is behaving as a monomeric full-length transcript rather than a mixture of fragments or aggregates.

With that in place, the authors used SEC-SAXS to obtain low-resolution solution structures of both RNAs under near-physiological conditions. Both sense and antisense AluSx1 RNAs adopt elongated, asymmetric conformations rather than compact globular ones. The resulting SAXS envelopes suggest RNAs with extended double-stranded arms connected through junction regions, consistent with a structured but conformationally flexible architecture.

The computational part of the workflow then builds on previously determined secondary structures. Using SimRNA and hydrodynamic filtering, high-resolution atomistic models were generated and compared against the SAXS-derived envelopes. The important point is not that a single exact structure was identified, but that a constrained family of models can be found that agrees with both the biophysical data and the known secondary structure. In practice, this is a useful way to move from SHAPE-informed 2D models toward plausible 3D representations for long non-coding RNAs.

Biologically, that matters because the Alu inverted repeats are thought to contribute to how lincRNA-p21 interacts with protein partners and localizes within the cell. The study supports a picture in which both the sense and antisense regions form structured domains with conserved architectural features, while still retaining enough flexibility to make ensemble-based modeling more appropriate than a single rigid conformation.

More broadly, this paper is useful as a methods example. Long non-coding RNAs are difficult structural targets: they are large, flexible, and hard to crystallize, while sequence-only tertiary prediction remains unreliable. The pipeline used here, combining prior secondary-structure information with AUC, SEC-MALS, SAXS, and restrained computational modeling, is a practical template for deriving experimentally grounded structural models of other lncRNAs as well.

Abstract

Human Long Intergenic Noncoding RNA-p21 (LincRNA-p21) is a regulatory noncoding RNA that plays an important role in promoting apoptosis. LincRNA-p21 is also critical in down-regulating many p53 target genes through its interaction with a p53 repressive complex. The interaction between LincRNA-p21 and the repressive complex is likely dependent on the RNA tertiary structure. Previous studies have determined the two-dimensional secondary structures of the sense and antisense human LincRNA-p21 AluSx1 IRs using SHAPE. However, there were no insights into its three-dimensional structure. Therefore, we in vitro transcribed the sense and antisense regions of LincRNA-p21 AluSx1 Inverted Repeats (IRs) and performed analytical ultracentrifugation, size exclusion chromatography, light scattering, and small angle X-ray scattering (SAXS) studies. Based on these studies, we determined low-resolution, three-dimensional structures of sense and antisense LincRNA-p21. By adapting previously known two-dimensional information, we calculated their sense and antisense high-resolution models and determined that they agree with the low-resolution structures determined using SAXS. Thus, our integrated approach provides insights into the structure of LincRNA-p21 Alu IRs. Our study also offers a viable pipeline for combining the secondary structure information with biophysical and computational studies to obtain high-resolution atomistic models for long noncoding RNAs.

Citation

Biophysical Characterisation of Human LincRNA-p21 Sense and Antisense Alu Inverted Repeats

Michael H. D’Souza, Tyler Mrozowich, Maulik D. Badmalia, Mitchell Geeraert, Angela Frederickson, Amy Henrickson, Borries Demeler, Michael T. Wolfinger, Trushar R. Patel
Nucleic Acids Res. gkac414 (2022) | doi:10.1093/nar/gkac414 | PDF

RNA structure conservation and molecular epidemiology of TBEV

2021-12-17T00:00:00+01:00

Annotated 3'UTR of representative tick-borne encephalitis virus (TBEV) strains

Tick-borne encephalitis virus is a good test case for comparative RNA virology because its 3' UTRs are variable enough to be evolutionarily interesting, but not so unconstrained that every lineage looks unrelated to the others. The paper asks whether that variation is best understood as random sequence drift or as remodeling within a limited structural vocabulary. The answer is clearly the latter. TBEV 3' UTRs contain a highly conserved core domain near the 3' end and a more labile upstream region, yet even the variable part can be described through recurring structured elements rather than unrestricted sequence turnover.

That point matters because it shifts the interpretation of TBEV diversity. If the architecture is built from a restricted set of conserved RNA modules, then lineage differences are not just a matter of "more" or "less" sequence divergence. They reflect different combinations, duplications, and losses of structured elements that are likely under functional constraint. In other words, the paper argues that 3' UTR evolution in TBEV is better understood at the level of RNA architecture than at the level of raw sequence identity alone.

The work fits naturally after the broader comparative flavivirus 3' UTR analysis and alongside the later mosquito-borne flavivirus 3' UTR synthesis. Those papers establish the larger structural vocabulary of flavivirus untranslated regions. This TBEV study then zooms in on one medically important lineage and asks how that vocabulary is reused within a single viral species complex. The narrower scope makes it easier to connect comparative RNA structure directly to questions of subtype diversification and geographic spread.

The molecular epidemiology part is therefore not an add-on. By introducing TBEVnext, the paper places the structural observations into an explicitly spatiotemporal framework. The resulting phylogeny shows that subtype labels and geographic occurrence do not map onto each other in a simple one-to-one way. That makes the dataset useful beyond visualization. It links particular 3' UTR architectures to the broader history of lineage expansion, ecological movement, and sampling across Europe and Asia.

What I find most compelling here is the combination of scales. The paper examines the fine-grained organization of non-coding RNA elements in the viral genome, then asks how those architectures sit inside the long-range history of the virus population. That is exactly what makes comparative RNA virology interesting. Structured RNA elements are not isolated curiosities. They evolve inside real lineages moving through hosts, vectors, and landscapes.

For TBEV specifically, the paper provides a more coherent picture of how structural conservation and epidemiological diversification coexist. The genome is not frozen, and the 3' UTR is certainly not uniform. Even so, the range of viable architectures appears constrained enough to reveal recurring principles. That makes the study a useful bridge between comparative structure prediction and phylodynamic surveillance, and gives the TBEV system a clearer place within the larger flavivirus RNA-structure landscape.

That broader point is exactly what I make more explicitly in When sequence conservation is not enough to find functional RNA structure.

Abstract

Tick-borne encephalitis virus (TBEV) is the etiological agent of tick-borne encephalitis, an infectious disease of the central nervous system that is often associated with severe sequelae in humans. While TBEV is typically classified into three subtypes, recent evidence suggests a more varied range of TBEV subtypes and lineages that differ substantially in their 3’UTR architecture. Building on comparative genomics approaches and thermodynamic modelling, we characterize the TBEV 3’UTR structureome diversity and propose a unified picture of pervasive non-coding RNA (ncRNA) structure conservation. Moreover, we provide an updated phylogeny of TBEV, building on more than 220 publicly available complete genomes, and investigate the molecular epidemiology and phylodynamics with Nextstrain, a web-based visualization framework for real-time pathogen evolution.

Figures and Data

Citation

Evolutionary traits of Tick-borne encephalitis virus: Pervasive non-coding RNA structure conservation and molecular epidemiology
Lena S. Kutschera, Michael T. Wolfinger
Virus Evol. (8):1 veac051 (2022) | doi: 10.1093/ve/veac051 | PDF | Supplementary data

Caveats in deep learning for RNA secondary structure prediction

2021-12-16T00:00:00+01:00

Input/output encoding for predicting RNA paired/unpaired status using a BLSTM

Deep learning for RNA secondary structure prediction has an obvious appeal. If neural networks can infer structure directly from sequence, perhaps they can move past some of the limitations of classical thermodynamic folding. That promise has made the area popular, but it has also brought in a familiar problem from other parts of machine learning. Strong benchmark numbers can hide the fact that a model has learned properties of the dataset rather than properties of the underlying biology.

That is the central concern of this paper. Instead of asking only whether a model performs well on a standard test split, we ask what it has actually learned. Has it captured transferable principles of RNA folding, or has it mostly memorized the structural biases of the RNAs it saw during training? For RNA structure prediction, that distinction matters a great deal. The real task is not to recognize another tRNA-like example from a familiar family. It is to say something useful about RNAs with new sequence-structure relationships.

The methodological approach is deliberately controlled. We use inverse RNA folding to generate synthetic datasets with known structural properties, which lets us vary the amount of bias in the training data instead of merely inheriting whatever bias happened to be present in a benchmark collection. That setup makes it possible to compare model behavior on "more of the same" against genuinely novel structural patterns. The paper is not simply another machine-learning benchmark. It is a stress test for generalization.

The result is sobering but informative. When neural networks are trained on biased datasets, they can perform surprisingly well on held-out sequences that fold into familiar classes of structures. Once the test set contains structures outside those familiar patterns, performance drops sharply. The models generalize across sequence variation far more readily than they generalize across structural novelty. That is a warning sign for anyone hoping to use deep learning as a drop-in replacement for biophysical RNA folding models.

An equally important observation is that removing dataset bias does not magically solve the problem. Even on unbiased synthetic data, several architectures struggle to recover basic structural constraints reliably. Some models predict pairing patterns whose scaling with sequence length is inconsistent with valid secondary structures. Others produce artifacts that resemble pseudoknots or base triples even when the ViennaRNA-style ground truth does not contain such features at all. These are not minor numerical errors. They point to a mismatch between model output and the combinatorial rules that define the object being predicted.

I still find the article important years later because it is not an anti-AI paper. It is a paper about technical honesty. If machine learning is going to contribute meaningfully to RNA biology, evaluation has to distinguish memorization from mechanism. The models also have to respect the structural constraints of RNA rather than merely fitting correlations in a benchmark. The work argues, implicitly, for approaches that combine learning with stronger priors, explicit structure constraints, or experimental information instead of assuming that larger networks alone will fix the problem.

The related practical question is when a structure predictor explains a mechanism well enough to influence an experimental decision. The answer depends less on whether a model is branded as AI and more on whether the evidence is robust, interpretable, and proportionate to the cost of the next step.

Readers who arrive here from an AI angle may also want to look at some of my other work from the opposite direction. In Predicting RNA structures from sequence and probing data, I discuss how experimental structure probing can be integrated with computational prediction. In Conserved RNA regulatory switches in living cells, the focus shifts to transcriptome-scale structural ensembles and experimentally anchored regulatory switches. If you are more interested in dynamic folding than static structure, co-transcriptional RNA-ligand interaction dynamics shows the kind of mechanistic modeling that remains hard to replace with black-box prediction alone.

Abstract

Machine learning (ML) and in particular deep learning techniques have gained popularity for predicting structures from biopolymer sequences. An interesting case is the prediction of RNA secondary structures, where well established biophysics based methods exist. These methods even yield exact solutions under certain simplifying assumptions. Nevertheless, the accuracy of these classical methods is limited and has seen little improvement over the last decade. This makes it an attractive target for machine learning and consequently several deep learning models have been proposed in recent years. In this contribution we discuss limitations of current approaches, in particular due to biases in the training data. Furthermore, we propose to study capabilities and limitations of ML models by first applying them on synthetic data that can not only be generated in arbitrary amounts, but are also guaranteed to be free of biases. We apply this idea by testing several ML models of varying complexity. Finally, we show that the best models are capable of capturing many, but not all, properties of RNA secondary structures. Most severely, the number of predicted base pairs scales quadratically with sequence length, even though a secondary structure can only accommodate a linear number of pairs.

Figures and Data

Citation

Caveats to deep learning approaches to RNA secondary structure prediction
Christoph Flamm, Julia Wielach, Michael T. Wolfinger, Stefan Badelt, Ronny Lorenz, Ivo L. Hofacker
Front. Bioinform. 2:835422 (2022) | doi:10.3389/fbinf.2022.835422 | PDF | Figures

Functional RNA structures in the 3'UTR of Mosquito-Borne Flaviviruses

2021-09-02T00:00:00+02:00

Consensus RNA secondary structures of evolutionarily conserved elements in flavivirus 3' UTRs

This chapter asks a deceptively simple question: what is actually conserved in the 3' UTRs of mosquito-borne flaviviruses once one looks beyond primary sequence? For viruses such as dengue, West Nile, yellow fever, Japanese encephalitis, and Zika virus, the answer matters because the structured 3' UTR is not passive genomic baggage. It is part of the replication program and helps shape genome cyclization, sfRNA production, host adaptation, and the broader regulatory behavior of the viral RNA.

One useful starting point is that mosquito-borne flavivirus 3' UTRs do not consist of one canonical motif repeated from virus to virus. They are built from recurring structural modules, including xrRNAs, dumbbell elements, and terminal stem-loops, but the number, arrangement, and sequence realization of those modules differ across lineages. That means one has to think in terms of architectural conservation rather than exact sequence identity. The chapter pulls those patterns together in one place and shows where the common logic ends and lineage-specific variation begins.

The best known modules in this space are the exoribonuclease-resistant RNAs. xrRNAs stall host 5' to 3' nucleases such as Xrn1 and thereby protect downstream fragments that accumulate as subgenomic flaviviral RNAs. That mechanism has become one of the clearest examples of how RNA structure alone can create a biologically active molecular barrier. The chapter also makes the broader point that xrRNAs are only one part of the story. Mosquito-borne flaviviruses preserve other structured elements whose roles are tied to long-range RNA interactions, replication efficiency, and host-specific regulation.

What makes the chapter valuable is the comparative scope. Instead of focusing on a single virus, it surveys the known mosquito-borne flavivirus diversity and asks which structural motifs recur across the group and which appear to have been reshaped during evolution. That perspective complements the earlier comparative 3' UTR analysis of tick-borne, insect-specific, and no-known-vector flaviviruses, where the emphasis was on broader flavivirus diversity and on the evolutionary distribution of xrRNA-like elements. Here the focus narrows to the mosquito-borne branch, where several of the best studied medically relevant flaviviruses sit.

That narrower focus also helps frame later, more specific papers. The logic of conserved 3' UTR architecture feeds directly into work on TBEV 3' UTR structure conservation and molecular epidemiology, as well as into experimental studies of structure-dependent functions such as the BVDV IRES analysis or later flavivirus papers on cyclization, G-quadruplexes, and small-RNA production. Once these untranslated regions are seen as structured regulatory platforms, many apparently separate virological questions start to line up more naturally.

For me, that is the real contribution of this chapter. It provides a map of the conserved structural vocabulary in mosquito-borne flavivirus 3' UTRs and makes it easier to ask which of those motifs are ancient, which are lineage-specific, and which are most likely to carry experimentally testable function. As a synthesis piece, it is less about a single new mechanistic claim than about organizing a rapidly expanding field into a coherent comparative picture.

That broader comparative point is also the theme of When sequence conservation is not enough to find functional RNA structure, which makes explicit why flaviviral 3'UTRs so often have to be read at the level of structural architecture rather than sequence similarity alone.

Figures and Data

Citation

Functional RNA Structures in the 3’UTR of Mosquito-Borne Flaviviruses
Michael T. Wolfinger, Roman Ochsenreiter, Ivo L. Hofacker
In Virus Bioinformatics, edited by Dmitrij Frishman and Manja Marz, pp65–100. Chapman and Hall/CRC Press (2021) | doi: 10.1201/9781003097679-5 | PDF | Figures

Secondary and tertiary structure of the BVDV IRES

2021-05-15T00:00:00+02:00

3D structure prediction of the BVDV IRES region

Bovine viral diarrhea virus (BVDV) relies on an internal ribosome entry site (IRES) in its 5' untranslated region to drive cap-independent translation. For systems like this, function depends on RNA architecture rather than sequence alone, which makes the structural organization of the IRES worth analyzing in detail.

This study combined SHAPE-MaP probing with secondary structure analysis and tertiary modeling to ask a specific question: how is the BVDV IRES organized in solution, and what evidence supports the proposed pseudoknot in domain III? That is the central structural feature of the post, because pseudoknot architecture is likely to be directly relevant to how the IRES maintains a translation-competent state.

The resulting model supports a modular IRES architecture with three major domains. Much of the secondary structure agrees with earlier work, but the analysis also points to flexibility in domain II and a comparatively stable arrangement in domain III. That contrast is useful: it suggests that not all parts of the IRES contribute in the same way, and that local structural stability may be concentrated in the regions most critical for tertiary organization.

The most important result concerns domain III, where the data support an H-type pseudoknot and a specific local arrangement of helices. The tertiary modeling suggests quasi-coaxial stacking between motifs associated with the pseudoknot, which provides a structural explanation for why this region appears unusually stable. Comparative analysis across Pestivirus genomes further supports the idea that the pseudoknot is not an isolated feature of a single strain, but part of a conserved functional architecture.

That does not mean the computational model alone proves mechanism. What it does provide is a more concrete structural hypothesis for how the BVDV IRES is arranged and why certain elements are likely to matter for translation. In other words, the work narrows the space of plausible architectures and gives experimentalists a clearer basis for targeted perturbation.

This is also where the comparison to related IRES systems becomes useful. BVDV and hepatitis C virus share broad organizational similarities, but the real value of that comparison is not to imply immediate transfer of therapeutic strategy. It is to show that conserved structural logic can emerge across related viral translation elements, making careful cross-system comparison worthwhile.

For RNA virology, this is a good example of where structure probing and computational modeling complement each other well. Probing constrains the plausible secondary structures, comparative analysis highlights conserved base pairs, and tertiary modeling translates those constraints into a geometric hypothesis that can be tested further.

Abstract

The Internal Ribosome Entry Site (IRES) RNA of Bovine viral diarrhea virus (BVDV), an economically significant Pestivirus, is required for the cap-independent translation of viral genomic RNA. Thus, it is essential for viral replication and pathogenesis. We applied a combination of high-throughput biochemical RNA structure probing (SHAPE-MaP) and in silico modeling approaches to gain insight into the secondary and tertiary structures of BVDV IRES RNA. Our study demonstrated that BVDV IRES RNA forms in solution a modular architecture composed of three distinct structural domains (I-III). Two regions within domain III are engaged in tertiary interactions to form an H-type pseudoknot. Computational modeling of the pseudoknot motif provided a fine-grained picture of the tertiary structure and local arrangement of helices in the BVDV IRES. Furthermore, comparative genomics and consensus structure predictions revealed that the pseudoknot is evolutionarily conserved among many Pestivirus species. These studies provide detailed insight into the structural arrangement of BVDV IRES RNA H-type pseudoknot and encompassing motifs that likely contribute to the optimal functionality of viral cap-independent translation element.

Figures and Data

Citation

Insights into the secondary and tertiary structure of the Bovine Viral Diarrhea Virus Internal Ribosome Entry Site
Devadatta Gosavi, Iwona Wower, Irene K Beckmann, Ivo L Hofacker, Jacek Wower, Michael T Wolfinger, Joanna Sztuba-Solinska
RNA Biol. 19(1) 496-506 (2022) | doi:10.1080/15476286.2022.2058818 | PDF | Figures

How Pseudomonas aeruginosa responds to colistin and tobramycin

2021-04-30T00:00:00+02:00

Pathways and functions dysregulated in Pseudomonas aeruginosa upon colistin treatment

This paper looks at a clinically important problem from a systems-biology angle. In cystic fibrosis infections, Pseudomonas aeruginosa is often treated with colistin or tobramycin once many other antibiotics have become ineffective. Both are last-resort drugs, but they stress the cell in very different ways. Colistin primarily targets the envelope, whereas tobramycin disrupts translation after energy-dependent uptake into the cytoplasm. The question behind the paper is therefore not just which resistance genes are already known, but how the bacterium reorganizes its physiology when it is actually exposed to these two drugs under a host-relevant growth condition.

The experimental design is what makes the study particularly useful. PA14 was grown in synthetic cystic fibrosis sputum medium and then challenged with inhibitory concentrations of colistin or tobramycin. Instead of measuring transcript levels alone, the paper combined RNA-seq with ribosome profiling in parallel. That matters because antibiotic stress often perturbs translation directly, and a transcriptome alone can miss the distinction between genes that are transcribed more strongly and genes that are actually being translated more efficiently. In that sense, the study moves one level closer to the physiological response of the cell.

The two antibiotics trigger markedly different expression programs. Colistin elicits a response centered on envelope and oxidative-stress adaptation, including well-known pathways connected to lipid A modification and polymyxin resistance, but also broader changes involving the MexT and AlgU regulons. Tobramycin, by contrast, produces a response that is much more tied to translational stress and metabolic rewiring. The cells alter amino-acid catabolism, lower-TCA-cycle genes, secretion systems, and functions linked to motility and attachment, while at the same time increasing expression of systems involved in stalled-ribosome rescue, tRNA methylation, and toxin-antitoxin modules.

One of the strengths of the paper is that it shows these are not merely generic stress signatures. The colistin and tobramycin responses diverge in ways that reflect the underlying drug mechanisms. Colistin mainly drives a membrane-protective and anti-oxidative program, while tobramycin pushes the bacterium into a state that appears designed to reduce uptake, manage translational damage, and compensate for ribosome disruption. That distinction makes the dataset more than a catalog of differentially expressed genes. It provides a mechanistic map of how Pseudomonas senses and reacts to two antibiotic classes that remain highly relevant in the clinic.

The ribosome-profiling component is especially valuable in the tobramycin case. Since aminoglycosides act on the ribosome, direct translatome information helps reveal the countermeasures used by the cell to keep protein synthesis viable enough for survival. The upregulation of rescue factors, methylation-associated functions, and toxin-antitoxin systems reads as a translational damage-control program. That is precisely the kind of biology that would be harder to infer confidently from mRNA levels alone.

This study fits naturally beside the earlier anoxic CF-sputum transcriptomics work, which asked how the infection-like environment reshapes physiology over time. The 2021 paper starts from that already adapted state and asks what happens when the cells are hit with last-resort antibiotics. The later Hfq/Crc/CrcZ papers approach related questions from a regulatory angle and ask how metabolic and post-transcriptional control feed into drug susceptibility. Read together, these studies make it clear that antibiotic response in Pseudomonas is tightly interwoven with metabolism, envelope state, and translational control.

From a practical perspective, the paper is a reminder that resistance and susceptibility are not static traits. Even when a strain carries known resistance determinants, the acute regulatory response to treatment can still expose weak points or compensatory pathways. That is why datasets like this are useful: they help identify which circuits are activated under drug pressure and which of them might be worth targeting in combination therapies.

Abstract

Pseudomonas aeruginosa has become resistant to most antibiotics, leaving polymyxins and aminoglycosides among the last therapeutic options. This study profiled gene expression and ribosome occupancy in parallel in strain PA14 grown in synthetic cystic fibrosis sputum medium after exposure to colistin or tobramycin. Colistin primarily induced anti-oxidative and envelope-associated responses together with deregulation of the MexT and AlgU regulons, whereas tobramycin caused strong changes in amino-acid catabolism, lower TCA-cycle functions, secretion systems, motility- and attachment-related genes, and pathways involved in stalled-ribosome rescue, tRNA methylation, and toxin-antitoxin systems. The combined RNA-seq/Ribo-seq approach therefore captures both the transcriptional and translational layers of the antibiotic stress response.

Citation

Gene Expression Profiling of Pseudomonas aeruginosa Upon Exposure to Colistin and Tobramycin
Anastasia Cianciulli Sesso, Branislav Lilic, Fabian Amman, Michael T. Wolfinger, Elisabeth Sonnleitner, Udo Blasi
Front. Microbiol. 12:626715 (2021) | doi:10.3389/fmicb.2021.626715 | PDF

Mpulungu virus and unique xrRNAs in a novel African tick flavivirus

2021-03-01T00:00:00+01:00

Exoribonuclease-resistant RNAs (xrRNAs) in the 3'UTR of Mpulungu virus

Mpulungu virus became important because it did not fit comfortably into the usual picture of tick-borne flaviviruses. The genome recovered from a Rhipicephalus tick in Zambia grouped with Ngoye virus from Senegal, defining an unusual African lineage that sat apart from the better known vertebrate-associated tick-borne flaviviruses. At the time, that alone made the paper notable. It expanded the geographic and phylogenetic range of the group and suggested that the ecological diversity of tick-associated flaviviruses had been underestimated.

What gives the paper lasting value is the structural analysis of the untranslated regions. The 5' end of MPFV still looks recognizably flaviviral, but the 3' UTR is where the more interesting divergence appears. Instead of treating the non-coding region as an unusual sequence tail, the paper asks which structured RNA elements are preserved there and whether they remain functionally competent despite the lineage's broader divergence.

The central result is the identification of two exoribonuclease-resistant RNA elements in the 3' UTR of MPFV. These xrRNA structures are not just predicted computationally. The study shows biochemically that both can stall Xrn1, providing direct evidence that this unusual African lineage retains a key flaviviral strategy for generating protected decay intermediates and, likely, subgenomic flaviviral RNAs. That tied the new lineage back to a conserved mechanistic theme in flavivirus RNA biology even while the surrounding genomic context looked quite distinctive.

Seen this way, MPFV was an early sign that unusual tick-associated flaviviruses could preserve the broader flaviviral RNA toolkit while following different ecological trajectories. The later Xinyang flavivirus paper sharpened that interpretation by showing that the lineage was not confined to Africa and by strengthening the case for a basal, likely tick-only clade. Read together, the two papers make more sense than either one does alone. MPFV established the branch and its unusual xrRNA features. XiFV then broadened the clade and shifted the ecological argument away from a vertebrate-centered default.

This also places MPFV naturally beside the Zambian insect-specific flavivirus xrRNA study. In both cases, virus discovery becomes much more informative once it is connected to comparative RNA structure and functional validation. The broader lesson is that understudied flavivirus lineages often reveal their deepest commonalities not through obvious sequence identity, but through preserved structured RNA elements in the untranslated regions.

I would now read this paper less as a warning about an immediate public-health threat and more as a foundational comparative RNA virology study. It identified a novel lineage, mapped its structured 3' UTR, and showed that even in a phylogenetically unusual branch, xrRNA function remains central enough to be retained. That is the kind of result that makes later ecological and evolutionary reinterpretation possible.

It is also a good example of the argument in When sequence conservation is not enough to find functional RNA structure, because the deeper commonality here emerges from preserved structured-RNA logic rather than from easy sequence similarity alone.

Abstract

Tick-borne flaviviruses (TBFVs) infect mammalian hosts through tick bites and can cause various serious illnesses, such as encephalitis and hemorrhagic fevers, both in humans and animals. Despite their importance to public health, there is limited epidemiological information on TBFV infection in Africa. Herein, we report that a novel flavivirus, Mpulungu flavivirus (MPFV), was discovered in a Rhipicephalus muhsamae tick in Zambia. MPFV was found to be genetically related to Ngoye virus detected in ticks in Senegal, and these viruses formed a unique lineage in the genus Flavivirus. Analyses of dinucleotide contents of flaviviruses indicated that MPFV was similar to those of other TBFVs with a typical vertebrate genome signature, suggesting that MPFV may infect vertebrate hosts. Bioinformatic analyses of the secondary structures in the 3′-untranslated regions (UTRs) revealed that MPFV exhibited unique exoribonuclease-resistant RNA (xrRNA) structures. Utilizing biochemical approaches, we clarified that two xrRNA structures of MPFV in the 3′-UTR could prevent exoribonuclease activity. In summary, our findings provide new information regarding the geographical distribution of TBFV and xrRNA structures in the 3′-UTR of flaviviruses.

Citation

An African Tick Flavivirus Forming an Independent Clade Exhibits Unique Exoribonuclease-Resistant RNA Structures in the Genomic 3’-Untranslated Region
Hayato Harima, Yasuko Orba, Shiho Torii, Yongjin Qiu, Masahiro Kajihara, Yoshiki Eto, Naoya Matsuta, Bernard M. Hang’ombe, Yuki Eshita, Kentaro Uemura, Keita Matsuno, Michihito Sasaki, Kentaro Yoshii, Ryo Nakao, William W. Hall, Ayato Takada, Takashi Abe, Michael T. Wolfinger, Martin Simuunza, Hirofumi Sawa
Sci. Rep. 11:4883 (2021) | doi: 10.1038/s41598-021-84365-9 | PDF

Molecular epidemiology and RNA structure in Chikungunya virus

2021-02-08T00:00:00+01:00

Ensemble properties of a lineage-specific structured RNA in Chikungunya virus

Chikungunya virus, an emerging Alphavirus, causes millions of human infections annually and has witnessed outbreaks in Africa and Asia since the 1950s. The research presented here delves into the molecular epidemiology of CHIKV, revealing how single nucleotide variants can impact the virus's RNA structures. Utilizing more than 1000 publicly available CHIKV genomes, the study provides an interactive phylodynamics dataset using Nextstrain, a tool for real-time tracking of pathogen evolution. The data set, publicly available through CHIKVnext, enables the tracking of spatiotemporal and epidemiological aspects of the CHIKV global spread.

The study addresses genotype/phenotype associations in CHIKV by employing comparative approaches, molecular epidemiology concepts, phylogeny reconstruction, and computational RNA biology. The nucleotide divergence within CHIKV lineages is found to be relatively low, often due to geographical constraints and a limited collection period for newer lineages. However, the divergence between different lineages can be considerable, with the West African lineage showing the highest divergence compared to others.

This publication questions the current nomenclature of CHIKV lineages, and suggests that a new system, independent of geography, might be more coherent and beneficial for drug and vaccine development. Despite the low divergence in some established lineages, mutations like A226V in E1 can significantly impact the virus's capacity to replicate. The study further highlights the importance of increased surveillance to identify the virus at the time of introduction rather than during a pandemic.

This research exemplifies the importance of a holistic, interdisciplinary approach in understanding and combating infectious diseases such as CHIKV. The findings, which combine molecular epidemiology and RNA structure prediction, not only pave the way for better diagnostic and treatment methods but also underscore the necessity of a timely response to prevent outbreaks. The One Health approach is particularly pertinent here. By adopting a unified perspective, scientists, healthcare professionals, and policymakers can work together to address the multifaceted challenges posed by CHIKV and similar infectious diseases, ultimately contributing to the well-being of our global ecosystem.

Abstract

Chikungunya virus (CHIKV) is an emerging Alphavirus which causes millions of human infections every year. Outbreaks have been reported in Africa and Asia since the early 1950s, from three CHIKV lineages: West African, East Central South African, and Asian Urban. As new outbreaks occurred in the Americas, individual strains from the known lineages have evolved, creating new monophyletic groups that generated novel geographic-based lineages. Building on a recently updated phylogeny of CHIKV, we report here the availability of an interactive CHIKV phylodynamics dataset, which is based on more than 900 publicly available CHIKV genomes. We provide an interactive view of CHIKV molecular epidemiology built on Nextstrain, a web-based visualization framework for real-time tracking of pathogen evolution. CHIKV molecular epidemiology reveals single nucleotide variants that change the stability and fold of locally stable RNA structures. We propose alternative RNA structure formation in different CHIKV lineages by predicting more than a dozen RNA elements that are subject to perturbation of the structure ensemble upon variation of a single nucleotide.

Figures and Data

Citation

Dynamic Molecular Epidemiology Reveals Lineage-Associated Single-Nucleotide Variants That Alter RNA Structure in Chikungunya Virus
Thomas Spicher, Markus Delitz, Adriano de Bernardi Schneider, Michael T. Wolfinger
Genes 12 (2):239 (2021) | doi:10.3390/genes12020239 | PDF | Figures

Genomic epidemiology of SARS-CoV-2 superspreading events in Austria

2020-12-10T00:00:00+01:00

This paper captures one of the moments when genomic epidemiology became central to public-health interpretation in real time. During the first wave of SARS-CoV-2 in Europe, Austria occupied a particularly informative position because several major transmission chains, including tourism-associated outbreaks, could be linked to well-documented epidemiological contexts. That made the country a strong setting for asking not just where the virus spread, but how superspreading events shaped early pandemic dynamics.

The study combines classical epidemiological tracing with deep whole-genome sequencing of more than 500 viral samples. That pairing is what gives the paper its strength. Sequencing alone can reveal related clusters, but without epidemiological context many details remain ambiguous. Conversely, case tracing alone cannot resolve the mutational microdynamics of transmission chains. Here, the two layers reinforce one another and allow the authors to reconstruct superspreading events with unusual confidence.

One major result is geographic and historical. The paper shows how Austrian clusters, especially those linked to winter tourism, contributed to wider dissemination across Europe during the first pandemic wave. In that sense, the work is a study of mobility and amplification as much as one of virology. Superspreading is not treated as a vague narrative label, but as a process that can be reconstructed from linked genomes and transmission histories.

The second major result is methodological and biological. Because the infection clusters were unusually well defined, the authors could examine how viral variation behaves within and across short transmission chains. This includes low-frequency variants that later became fixed, as well as time-resolved within-host changes. That kind of analysis is difficult in noisier surveillance settings, but here it becomes possible to ask what viral diversity is actually transmitted during person-to-person spread.

The transmission bottleneck estimate is especially notable. The paper reports an average bottleneck of roughly 10^3 viral particles, a result that attracted attention because it put a quantitative scale on SARS-CoV-2 transmission under real outbreak conditions. Whether one is interested in mutation fixation, founder effects, or lineage emergence, that number matters because it constrains how much within-host diversity is likely to pass from one infection to the next.

What makes the paper more than a historical case study is that it demonstrates the analytical power of dense genomic surveillance when combined with validated epidemiological data. The first wave of COVID-19 produced many phylogenetic studies, but fewer had this combination of national-scale sampling, cluster resolution, deep sequencing, and explicit bottleneck analysis. That is why the article remains a useful reference for how pathogen genomics can move beyond tree drawing and toward mechanistic interpretation of transmission.

This article sits somewhat outside structure-centered virology, but it is still very much part of virus bioinformatics. The computational challenge here is different. It is not RNA folding or structured elements, but the integration of sequencing, phylogenetics, and outbreak reconstruction at scale. That broader epidemiological dimension deserves a fuller treatment than an abstract alone.

Abstract

Superspreading events shaped the coronavirus disease 2019 (COVID-19) pandemic, and their rapid identification and containment are essential for disease control. Here, we provide a national-scale analysis of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) superspreading during the first wave of infections in Austria, a country that played a major role in initial virus transmissions in Europe. Capitalizing on Austria’s well-developed epidemiological surveillance system, we identified major SARS-CoV-2 clusters during the first wave of infections and performed deep whole-genome sequencing of more than 500 virus samples. Phylogenetic-epidemiological analysis enabled the reconstruction of superspreading events and charts a map of tourism-related viral spread originating from Austria in spring 2020. Moreover, we exploited epidemiologically well-defined clusters to quantify SARS-CoV-2 mutational dynamics, including the observation of low-frequency mutations that progressed to fixation within the infection chain. Time-resolved virus sequencing unveiled viral mutation dynamics within individuals with COVID-19, and epidemiologically validated infector-infectee pairs enabled us to determine an average transmission bottleneck size of 103 SARS-CoV-2 particles. In conclusion, this study illustrates the power of combining epidemiological analysis with deep viral genome sequencing to unravel the spread of SARS-CoV-2 and to gain fundamental insights into mutational dynamics and transmission properties.

Citation

Genomic Epidemiology of Superspreading Events in Austria Reveals Mutational Dynamics and Transmission Properties of SARS-CoV-2
Alexandra Popa, Jakob-Wendelin Genger, Michael D. Nicholson, Thomas Penz, Daniela Schmid, Stephan W Aberle, Benedikt Agerer, Alexander Lercher, Lukas Endler, Henrique Colaco, Mark Smyth, Michael Schuster, Miguel L. Grau, Francisco Martínez-Jiménez, Oriol Pich, Wegene Borena, Erich Pawelka, Zsofia Keszei, Martin Senekowitsch, Jan Laine, Judith H Aberle, Monika Redlberger-Fritz, Mario Karolyi, Alexander Zoufaly, Sabine Maritschnik, Martin Borkovec, Peter Hufnagl, Manfred Nairz, Günter Weiss, Michael T. Wolfinger, Dorothee von Laer, Giulio Superti-Furga, Nuria Lopez-Bigas, Elisabeth Puchhammer-Stöckl, Franz Allerberger, Franziska Michor, Christoph Bock, Andreas Bergthaler
Sci. Transl. Med. 12 (573):eabe2555 (2020) | doi:10.1126/scitranslmed.abe2555 | PDF

Bi-alignments for incongruent RNA sequence and structure evolution

2020-11-01T00:00:00+01:00

This paper addresses a subtle but important problem in comparative RNA analysis. Many RNA alignment methods assume that sequence evolution and secondary structure evolution are congruent, meaning that homologous nucleotides also remain in corresponding structural roles. That assumption is often useful, and for many classical RNA families it works well enough to support consensus-structure prediction, covariance modeling, and structure-aware multiple alignment. But biology is not always that tidy.

The motivating observation here is that structured RNAs can preserve a recognizable fold even when the exact pairing register has shifted relative to the underlying sequence. In that situation, analogous base pairs are no longer formed by homologous nucleotides, and a single alignment cannot simultaneously do justice to both sequence similarity and structural correspondence. If one insists on aligning the homologous sequence blocks, the structural match degrades. If one instead aligns the analogous stem or loop positions, the sequence alignment begins to look wrong. That is the kind of incongruence this paper sets out to model explicitly.

The methodological contribution is the idea of a bi-alignment. Instead of forcing sequence homology and structural homology into one common alignment, the paper represents them as two coupled alignments: one for sequence, one for structure. A third alignment then relates these two views to each other and accounts for the relative shifts between them. Formally, this turns the problem into a constrained four-way alignment, but the conceptual point is simpler than the formalism may suggest. The method acknowledges that sequence and structure can both be conserved, yet conserved in slightly different coordinate systems.

That framing is useful because it captures evolutionary scenarios that standard consensus models tend to flatten away. One example discussed in the paper is stem sliding, where a stem can effectively move by losing a pair at one end while gaining a new one at the other. The overall hairpin remains recognizable and functionally similar, but the base pairs are no longer anchored to the same homologous nucleotides. A conventional structure-aware alignment tends to overcommit to one notion of correspondence or the other. Bi-alignments provide a way to represent the mismatch between those two notions directly instead of treating it as noise.

The paper is also interesting as a reminder that comparative RNA bioinformatics depends on modeling assumptions that are easy to forget once they become standard. Tools based on the Sankoff idea, consensus folding, or covariance models have been extremely successful, but they are built around the expectation that conserved structure and conserved sequence can be reconciled in one coherent alignment. Bi-alignments are valuable not because that assumption is usually wrong, but because they define what to do in the cases where it is not quite right.

Methodologically, this was a deliberately formal paper. It does not present a flashy biological case study so much as it tries to sharpen the mathematical language for a real comparative-genomics phenomenon. The discussion of miRNA precursors and the exploratory scan through Rfam are there to show that incongruent evolution is not just a pathological corner case. Structured RNAs can drift in ways that preserve shape better than position, and once that happens, the distinction between homology and structural analogy becomes computationally relevant.

Seen from the broader arc of RNA bioinformatics, this work fits into a longer effort to make our models better reflect how RNAs actually evolve. Some papers focus on thermodynamics, some on kinetics, some on experimental constraints, and others on machine learning. This one is about representation: what exactly are we aligning when we compare two structured RNAs? That is a foundational question, and although the paper is mathematically flavored, it speaks to a very biological issue.

For readers interested in RNA methods, the value of the paper is therefore twofold. First, it introduces a concrete framework for handling sequence-structure incongruence. Second, it sharpens the intuition that "RNA structure conservation" is not always equivalent to "the same nucleotides occupy the same structural roles". That distinction matters whenever we try to infer evolutionary relationships from structured non-coding RNAs, especially in families where local motifs can shift while the overall fold remains recognizable.

Citation

Bi-Alignments as Models of Incongruent Evolution of RNA Sequence and Secondary Structure
Maria Waldl, Sebastian Will, Michael T. Wolfinger, Ivo L. Hofacker, Peter F. Stadler
In Computational Intelligence Methods for Bioinformatics and Biostatistics, pp159-170. Springer International Publishing (2020) | doi:10.1007/978-3-030-63061-4_15 | Preprint PDF

Insect-specific flaviviruses from Zambia and their exoribonuclease-resistant RNAs

2020-09-14T00:00:00+02:00

Comparative 3'UTR RNA architecture of insect-specific flaviviruses from Zambia

This paper sits at a point where field virology, genome sequencing, and comparative RNA analysis come together very directly. The biological starting point is straightforward: mosquito populations in Zambia were screened for flaviviruses, and the resulting isolates were characterized at the genomic level. What makes the study especially interesting from my perspective is that it does not stop at virus discovery. It follows through into the untranslated regions and asks whether the newly isolated genomes also carry the conserved structured RNAs that are increasingly recognized as functional elements in flavivirus biology.

The study reports two closely related insect-specific flaviviruses from Culex mosquitoes, a Barkedji virus isolate from Zambia and a more distinct Barkedji-like virus. Both replicate in mosquito cells but not in mammalian or avian cell lines, which places them firmly in the insect-specific part of flavivirus diversity. That host restriction is already biologically interesting, but it also makes these viruses useful comparative models. They let us ask which RNA elements belong to the deeper flaviviral toolkit and which ones track more closely with host range or vector association.

Methodologically, the paper combines several layers of analysis. Viral genomes were recovered from field isolates by next-generation sequencing and end-completion methods, then compared with known flavivirus relatives. The untranslated regions were screened with comparative genomics and RNA secondary structure prediction in order to identify canonical flaviviral building blocks in both the 5' and 3' ends. This is the part where the paper moves beyond cataloguing a new isolate: it uses structural homology, not just sequence similarity, to identify conserved functional candidates.

The main result is that the 3'UTRs of both Zambian isolates contain a recognizable set of structured RNA elements, including xrRNA-like folds, an SL-III element, dumbbell-like structures, and the terminal 3' stem-loop. In other words, even these insect-specific viruses retain the broader architectural logic seen across flaviviral non-coding regions. That matters because flaviviral UTRs evolve quickly at the sequence level, and purely sequence-based annotation can miss homologous elements whose function is preserved through structure rather than exact nucleotide identity.

The paper then takes the crucial extra step and tests the predicted xrRNA candidates experimentally. In vitro Xrn1 resistance assays show that the proposed elements are not just plausible folds on paper. They are able to stall the host exoribonuclease. That functional validation is important. It turns a comparative prediction into evidence that these insect-specific viruses likely generate protected decay intermediates in the same general way as other flaviviruses with established xrRNA biology.

This study helped sharpen a recurring theme in flavivirus RNA biology: structured RNAs in viral UTRs are not decorative sequence features. They are conserved control elements, and they remain informative even in relatively under-sampled corners of flavivirus diversity. The Zambia isolates expanded that picture by showing that xrRNA-associated architecture is not restricted to the most intensively studied human pathogens. It is also present in insect-specific lineages, where it can be studied in an evolutionary context that is less confounded by vertebrate pathogenicity. For a broader comparative view, this paper pairs naturally with Functional RNA Structures in the 3’UTR of Tick-Borne, Insect-Specific and No Known Vector Flaviviruses and Functional RNA Structures in the 3’UTR of Mosquito-Borne Flaviviruses.

This also makes the paper a useful bridge between virus discovery and comparative RNA biology. It starts with surveillance and isolation, but it ends with a mechanistic claim about structured non-coding RNA. That combination is one reason why insect-specific flaviviruses remain so informative: they provide natural experiments in how conserved RNA elements are maintained, duplicated, or remodeled across different ecological niches.

Citation

Discoveries of Exoribonuclease-Resistant Structures of Insect-Specific Flaviviruses Isolated in Zambia
Christida E. Wastika, Hayato Harima, Michihito Sasakai, Bernard M. Hang'ombe, Yuki Eshita, Qiu Yongjin, William W. Hall, Michael T. Wolfinger, Hirofumi Sawa, Yasuko Orba
Viruses 12:1017 (2020) | doi:10.3390/v12091017 | PDF

Hfq regulates carbapenem susceptibility in Pseudomonas aeruginosa

2020-05-26T00:00:00+02:00

Model for differential regulation of oprD and opdP translation by Hfq, sRNAs, Crc, and CrcZ

This paper addresses a clinically relevant question through a very RNA-centered mechanism. In Pseudomonas aeruginosa, carbapenem susceptibility depends in part on whether the antibiotics can enter the cell through outer-membrane porins. Two of the important entry points are OprD and OpdP. What this study shows is that these two porins are not regulated in the same way, even though both are under the influence of the RNA chaperone Hfq. That distinction matters because it links antibiotic susceptibility not just to gene expression in general, but to specific post-transcriptional circuits.

The central result is that Hfq governs oprD and opdP through different regulatory mechanisms. Translation of oprD is repressed through Hfq-dependent riboregulation involving the small RNAs ErsA and Sr0161. By contrast, opdP is not best explained by the same sRNA route. Instead, the data support direct translational repression by an Hfq/Crc complex, the same regulatory logic that is already known from carbon catabolite repression in Pseudomonas. In other words, the two carbapenem entry ports sit under related but clearly distinct post-transcriptional control systems.

Methodologically, the paper disentangles these routes with a clean genetic design. Translational lacZ reporter fusions for oprD and opdP were combined with mutant backgrounds lacking hfq, crc, ersA, or sr0161, together with complementation and ectopic expression experiments. The study also used Hfq co-immunoprecipitation and microscale thermophoresis to show that Hfq can bind relevant regions in both transcripts. That is important because it separates the question of Hfq binding from the question of which regulatory partners actually mediate repression in each case.

The mechanistic picture that emerges is appealingly specific. oprD behaves like a classic Hfq-assisted sRNA target: the porin is repressed through the action of ErsA and Sr0161, especially under conditions such as envelope stress, low oxygen, or stationary phase. opdP, on the other hand, fits better into the carbon-catabolite-repression framework, where Hfq and Crc assemble a repressive complex on target mRNAs. The paper therefore does more than identify Hfq as a regulator. It shows that one global RNA-binding protein can shape antibiotic susceptibility by deploying different regulatory strategies on different mRNA targets.

An especially useful part of the paper is the integration of CrcZ into this story. CrcZ is an Hfq-sequestering RNA whose abundance depends on carbon source and growth state. When CrcZ levels rise, Hfq and Hfq/Crc repression are relieved, which increases translation of both oprD and opdP. That gives the study a broader physiological meaning: susceptibility to carbapenems is not fixed, but can shift with metabolic state because the regulatory RNAs and RNA-protein complexes controlling porin synthesis also respond to nutrient conditions.

This makes the paper relevant beyond one antibiotic class. It is a good example of how bacterial metabolism, RNA regulation, and antimicrobial susceptibility intersect. The most interesting point is not simply that Hfq affects drug response. It is that the effect can be rationalized mechanistically, through differential control of uptake channels and through the antagonistic action of a regulatory RNA that senses the carbon regime of the cell.

This study helps define a recurring theme in Pseudomonas RNA biology: carbon catabolite repression and Hfq-dependent regulation are not abstract layers sitting above physiology. They have direct consequences for traits that matter clinically, including biofilm behavior, nutrient adaptation, and susceptibility to antibiotics. This 2020 paper is one of the clearest demonstrations of that connection at the level of individual uptake systems. It also connects naturally to Interplay Between the Catabolite Repression Control Protein Crc, Hfq and RNA in Hfq-Dependent Translational Regulation in Pseudomonas aeruginosa and to Rewiring of Gene Expression in Pseudomonas aeruginosa During Diauxic Growth Reveals an Indirect Regulation of the MexGHI-OpmD Efflux Pump by Hfq.

Citation

Distinctive Regulation of Carbapenem Susceptibility in Pseudomonas aeruginosa by Hfq
Elisabeth Sonnleitner, Petra Pusic, Michael T. Wolfinger, Udo Blasi
Front. Microbiol. 11:1001 (2020) | doi:10.3389/fmicb.2020.01001 | PDF

Chikungunya virus phylogeny and lineage-specific RNA structures

2019-08-29T00:00:00+02:00

Lineage-specific 3' UTR architecture in Chikungunya virus

Chikungunya virus (CHIKV) is usually discussed in terms of a small set of named geographic lineages, but that picture has become increasingly inadequate as more genome sequences have accumulated. In this study we revisited CHIKV evolution using 598 complete genomes and asked two connected questions: how well the traditional lineage labels still reflect the phylogeny, and whether the distinct clades also differ in the organization of their untranslated regions.

The phylogenetic analysis was based on the two viral open reading frames and used a maximum-likelihood framework with explicit support assessment. One of the central results is that the historically defined ECSA group is not best understood as a single lineage. Instead, the data support several African sublineages with different evolutionary histories, including Eastern, Middle, and Southern African components. That distinction matters because the major epidemic clades outside Africa can be traced back to different parts of this broader African diversity. In the revised view used in the paper, the Indian Ocean lineage derives from Eastern African ancestors, while the South American outbreak strain is linked to Middle African taxa.

The second part of the paper focuses on the CHIKV 3' UTR, a region known to carry repeated sequence elements but whose structural organization had remained unclear. Using comparative RNA analysis, covariance-model based homology searches, and consensus secondary structure prediction, we found that the repeat-rich 3' UTR is not just a collection of duplicated sequence blocks. Instead, it follows an alternating architecture of conserved structured and unstructured elements. The structured part consists of stem-loop elements termed SL-a and SL-b, a Y-shaped element SL-Y, and a conserved sequence element near the terminus, while the intervening repeat regions remain predominantly unpaired.

This structured-versus-unstructured organization is one of the more interesting aspects of the study. The unstructured repeat regions are unlikely to be mere spacers. Because they remain accessible, they may act as structural insulators between neighboring RNA elements or provide binding surfaces for host proteins. The paper points in particular to repeated UAG motifs in these accessible regions, which makes interactions with proteins such as Musashi plausible, although that remains a functional hypothesis rather than a demonstrated mechanism.

Another useful outcome is that the 3' UTR architectures differ systematically between lineages. Some elements are shared across essentially all CHIKV groups, while others vary in copy number or arrangement, and recent American isolates of the Asian Urban lineage show evidence of partial duplication events. This suggests that CHIKV evolution in the 3' UTR is not just sequence drift. Rather, the virus appears to rearrange modular RNA building blocks while preserving an overall architectural pattern. That is a more informative model than treating the repeat region as a simple low-complexity tail.

From a computational RNA perspective, this paper is also a good example of what comparative genomics can do even when strong covariation support is limited. CHIKV 3' UTR sequences are often too similar, and the available sampling is too outbreak-biased, to expect the kind of covariation signal seen in deeply diverged structured RNAs. Even so, combining phylogeny, repeated architecture, local thermodynamic stability, and homology modeling is enough to identify plausible conserved RNA elements and to formulate experimentally testable hypotheses about their role in replication, host adaptation, and lineage-specific biology.

Abstract

Chikungunya virus (CHIKV), a mosquito-borne alphavirus of the family Togaviridae, has recently emerged in the Americas from lineages from two continents: Asia and Africa. Historically, CHIKV circulated as at least four lineages worldwide with both enzootic and epidemic transmission cycles. To understand the recent patterns of emergence and the current status of the CHIKV spread, updated analyses of the viral genetic data and metadata are needed. Here, we performed phylogenetic and comparative genomics screens of CHIKV genomes, taking advantage of the public availability of many recently sequenced isolates. Based on these new data and analyses, we derive a revised phylogeny from nucleotide sequences in coding regions. Using this phylogeny, we uncover the presence of several distinct lineages in Africa that were previously considered a single one. In parallel, we performed thermodynamic modeling of CHIKV untranslated regions (UTRs), which revealed evolutionarily conserved structured and unstructured RNA elements in the 3’UTR. We provide evidence for duplication events in recently emerged American isolates of the Asian CHIKV lineage and propose the existence of a flexible 3’UTR architecture among different CHIKV lineages.

Figures and Data

Citation

Updated Phylogeny of Chikungunya Virus Suggests Lineage-Specific RNA Architecture
Adriano de Bernardi Schneider, Roman Ochsenreiter, Reilly Hostager, Ivo L. Hofacker, Daniel Janies, Michael T. Wolfinger
Viruses 11:798 (2019) | doi:10.3390/v11090798 | PDF | Figures

Musashi binding elements in Zika virus 3'UTR

2019-05-06T00:00:00+02:00

Accessibility of Musashi binding elements in Zika virus and related flaviviruses

Musashi-1 is an RNA-binding protein that is highly expressed in neural stem and progenitor cells, and earlier experimental work had suggested that it can bind the Zika virus 3' UTR and enhance viral replication. That makes the structural context of Musashi binding elements in viral RNA immediately relevant: binding depends not just on the presence of a UAG core motif, but on whether that motif is exposed in a single-stranded region that is accessible to the protein.

This paper addresses exactly that question with a thermodynamic RNA-structure model. We screened curated flavivirus 3' UTRs for UAG motifs and evaluated how accessible each site is in its native sequence context. Rather than just folding a single minimum free energy structure, the analysis uses opening energies derived from ensemble calculations in sliding windows and compares them to dinucleotide-shuffled sequence controls. The resulting z scores make it possible to ask whether a given Musashi binding element is unusually exposed relative to what would be expected from local sequence composition alone.

For Zika virus, the result is clear. In both the African and the Asian/American lineages, UAG belongs to the most accessible trinucleotides in the 3' UTR, and the surrounding pentanucleotide context remains accessible as well. The Asian/American lineage is especially striking: in the Brazilian isolate used here, all UAG motifs in the 3' UTR show negative opening-energy z scores, meaning that they are consistently predicted to occur in unpaired structural contexts. That fits well with the idea that Musashi binding is structurally favored in epidemic ZIKV strains.

The comparative part of the study is just as important. The same calculation was carried out across mosquito-borne, tick-borne, insect-specific, and no-known-vector flaviviruses. ZIKV from the Asian/American lineage ranked as the most Musashi-accessible among the mosquito-borne flaviviruses in the dataset, but it was not the only virus with accessible Musashi motifs. Some West Nile, yellow fever, Powassan, Karshi, and insect-specific flaviviruses also contain strongly accessible UAG sites. That does not mean that all of these viruses share the same neuropathology as ZIKV, but it does show that ZIKV is not uniquely equipped with Musashi-compatible RNA motifs.

Another useful result is that Musashi binding elements are not randomly scattered across flavivirus 3' UTRs. By mapping the motifs onto conserved structural elements, the study found that dumbbell elements in particular often carry two conserved UAG motifs in a shared structural context. That is interesting because dumbbell RNAs are already known as functionally important 3' UTR elements in many flaviviruses. The combination of structural conservation and repeated Musashi-compatible motifs suggests that these sites are worth testing experimentally rather than treating them as incidental sequence matches.

Conceptually, this work is less about proving a single mechanism and more about building a tractable computational screen for host-factor compatibility in viral RNAs. Accessibility alone is not enough to predict congenital infection or neurotropism, and the paper is careful not to claim that it is. But opening-energy calculations do provide a useful way to prioritize candidate binding sites and candidate viruses for follow-up experiments. In that sense, the study connects RNA secondary structure thermodynamics with a concrete virological question: which flaviviruses place Musashi binding motifs in structural contexts that are likely to matter biologically?

Abstract

Zika virus (ZIKV) belongs to a class of neurotropic viruses that have the ability to cause congenital infection, which can result in microcephaly or fetal demise. Recently, the RNA-binding protein Musashi-1 (Msi1), which mediates the maintenance and self-renewal of stem cells and acts as a translational regulator, has been associated with promoting ZIKV replication, neurotropism, and pathology. Msi1 predominantly binds to single-stranded motifs in the 3′ untranslated region (UTR) of RNA that contain a UAG trinucleotide in their core. We systematically analyzed the properties of Musashi binding elements (MBEs) in the 3′UTR of flaviviruses with a thermodynamic model for RNA folding. Our results indicate that MBEs in ZIKV 3′UTRs occur predominantly in unpaired, single-stranded structural context, thus corroborating experimental observations by a biophysical model of RNA structure formation. Statistical analysis and comparison with related viruses show that ZIKV MBEs are maximally accessible among mosquito-borne flaviviruses. Our study addresses the broader question of whether other emerging arboviruses can cause similar neurotropic effects through the same mechanism in the developing fetus by establishing a link between the biophysical properties of viral RNA and teratogenicity. Moreover, our thermodynamic model can explain recent experimental findings and predict the Msi1-related neurotropic potential of other viruses.

Figures and Data

Citation

Musashi Binding Elements in Zika and Related Flavivirus 3’UTRs: A Comparative Study in Silico
Adriano de Bernardi Schneider, Michael T. Wolfinger
Sci. Rep. 9(1):6911 (2019) | doi:10.1038/s41598-019-43390-5 | PDF | Figures

Comparative genomics of flavivirus 3' UTR RNA structures

2019-03-25T00:00:00+01:00

Conserved RNA structures in flavivirus 3' UTRs

The 3' untranslated regions of flaviviruses contain far more than generic sequence context. They carry structured RNA elements involved in cyclization, replication, encapsidation, and resistance to host exonucleases. That makes them a natural target for comparative analysis, especially in viral groups where functional annotation is still sparse.

This paper takes a comparative genomics approach to the 3' UTRs of tick-borne, insect-specific, and no-known-vector flaviviruses. The question is not simply whether these viruses contain structured RNA, but how conserved architectures are distributed across lineages and where different flavivirus groups seem to have arrived at different structural solutions.

One of the main results is that exoribonuclease-resistant RNAs (xrRNAs), which had already been characterized experimentally in other flavivirus contexts, are more widely distributed than a narrow lineage-specific view would suggest. The analysis supports their presence in tick-borne and no-known-vector flaviviruses, reinforcing the idea that resistance to 5' to 3' degradation is a broadly reused structural strategy in flavivirus evolution.

The study also points to a repeated or cascaded organization of duplicated RNA structures in insect-specific flaviviruses. That observation matters because it suggests that flavivirus 3' UTRs are not static collections of conserved motifs. They are better viewed as flexible scaffolds on which evolution can duplicate, repurpose, and elaborate structured RNA elements while maintaining core functions.

Comparative genomics is especially valuable for this kind of result. In many cases, experimental structure determination across an entire viral clade is unrealistic. Sequence comparison, consensus folding, and covariation-aware reasoning make it possible to identify plausible conserved elements first and then ask which of them are most worth testing in the lab.

For work on xrRNAs, this broader evolutionary view is particularly useful. It helps separate features that are likely to be deeply conserved from those that may have arisen independently or been remodeled in particular viral groups. In turn, that sharpens how we think about structure-function relationships in flavivirus non-coding regions.

I return to that general issue in When sequence conservation is not enough to find functional RNA structure, which argues that many of the most informative signals in viral untranslated regions live at the level of conserved architecture rather than primary sequence identity.

Abstract

Untranslated regions (UTRs) of flaviviruses contain a large number of RNA structural elements involved in mediating the viral life cycle, including cyclisation, replication, and encapsidation. Here we report on a comparative genomics approach to characterize evolutionarily conserved RNAs in the 3'UTR of tick-borne, insect-specific and no-known-vector flaviviruses in silico. Our data support the wide distribution of previously experimentally characterized exoribonuclease resistant RNAs (xrRNAs) within tick-borne and no-known-vector flaviviruses and provide evidence for the existence of a cascade of duplicated RNA structures within insect-specific flaviviruses. On a broader scale, our findings indicate that viral 3'UTRs represent a flexible scaffold for evolution to come up with novel xrRNAs.

Figures and Data

Citation

Functional RNA Structures in the 3’UTR of Tick-Borne, Insect-Specific and No Known Vector Flaviviruses
Roman Ochsenreiter, Ivo L. Hofacker, Michael T. Wolfinger
Viruses 11:298 (2019) | doi:10.3390/v11030298 | PDF | Figures

A moonlighting role for archaeal aIF5A

2019-03-05T00:00:00+01:00

This paper asks a clean mechanistic question about a deeply conserved factor. In eukaryotes, eIF5A is best known as a translation factor that helps ribosomes move through problematic peptide motifs, especially polyproline stretches. Archaea carry the homologous protein aIF5A, but for a long time its precise role remained much less clear. The central contribution of this study is that it does not treat archaeal aIF5A as a simple copy of the eukaryotic factor. Instead, it tests whether the protein might do more than one job.

The first important result is physiological. Using CRISPR-based knockdown in Sulfolobus solfataricus, the study shows that lowering aIF5A levels produces a strong growth defect. That finding matters because it confirms that the protein is not peripheral. Even before getting to the biochemical details, the phenotype already argues that aIF5A is central to archaeal cell biology.

The more interesting part is the functional interpretation. The paper presents evidence that aIF5A is involved in translation, which is the expected direction given what is known for the eukaryotic homolog. But it then adds a less expected observation: purified Sso aIF5A also shows endoribonucleolytic activity in vitro. That is the basis for the "moonlighting" claim in the title. The argument is not that the canonical translation role disappears, but that the archaeal factor may combine that role with a second one in RNA processing or turnover.

What makes the paper interesting is precisely this shift from homology-based expectation to experimentally grounded nuance. Conserved proteins are often described by analogy to their better-characterized counterparts in other domains of life. That is a useful starting point, but it can flatten genuine biological differences. Here, the archaeal ortholog appears to preserve the core importance of the factor while also pointing to a broader role in RNA metabolism. For archaeal molecular biology, that is a more informative outcome than simply confirming conservation.

Methodologically, the paper combines genetics and biochemistry in a straightforward but effective way. The CRISPR knockdown establishes in vivo relevance, while the in vitro assays test what the purified factor can actually do to RNA. That division of labor between cell-based phenotype and biochemical activity is important, because the moonlighting interpretation requires both pieces. Growth retardation alone would not distinguish translation from other RNA-related functions, and biochemical RNase activity alone would not prove cellular significance. Taken together, though, the results make the hypothesis credible.

The broader interest of the paper is that it fits a recurring theme in RNA biology: proteins that are first classified in one pathway often turn out to participate in others once experimental tools become good enough. RNA-binding proteins, helicases, metabolic enzymes, and translation factors repeatedly cross those boundaries. This study puts archaeal aIF5A into that category. It suggests that the organization of RNA metabolism and translation in archaea may be more interconnected than a textbook factor-by-factor view would imply.

This article sits in archaeal molecular biology and asks a classic functional question about a conserved factor with a potentially dual role. It is a useful reminder that RNA biology is not only about RNA molecules themselves, but also about the proteins whose activities connect translation, processing, and decay. In that respect it pairs well with The SmAP1/2 Proteins of the Crenarchaeon Sulfolobus Solfataricus Interact with the Exosome and Stimulate A-Rich Tailing of Transcripts, which explores a different archaeal route into RNA metabolism.

Citation

Indications for a Moonlighting Function of Translation Factor aIF5A in the Crenarchaeum Sulfolobus Solfataricus
Flavia Bassani, Isabelle Anna Zink, Thomas Pribasnig, Michael T. Wolfinger, Alice Romagnoli, Armin Resch, Christa Schleper, Udo Blasi, Anna La Teana
RNA Biol. 16(5):675-685 (2019) | doi:10.1080/15476286.2019.1582953 | PDF

Metabolic control of Hfq-dependent antibiotic susceptibility in Pseudomonas aeruginosa

2018-11-09T00:00:00+01:00

Increased CrcZ levels correlate with increased gentamicin susceptibility in Pseudomonas aeruginosa

This paper takes the Hfq story in Pseudomonas aeruginosa one step further. Instead of asking only which genes are controlled by Hfq, it asks whether that regulatory layer can be exploited to make the bacterium more susceptible to antibiotics. That is a more intervention-oriented question, and it gives the work a slightly different flavor from the mechanistic porin paper that followed later. The central idea is that Hfq is not just another global regulator. Because it sits at the intersection of carbon catabolite repression, RNA-mediated control, membrane physiology, and stress adaptation, changing Hfq availability could shift several antibiotic-relevant traits at once.

The starting point is the observation that deletion of hfq increases susceptibility of P. aeruginosa to multiple classes of clinically relevant antibiotics. The paper then asks why that happens and whether the same effect can be reproduced in a less drastic and more physiologically meaningful way. That second question is where CrcZ becomes important. CrcZ is an RNA decoy that sequesters Hfq and thereby counteracts Hfq/Crc-mediated repression. If higher CrcZ levels reduce effective Hfq activity, then metabolic conditions that induce CrcZ, or artificial overexpression of CrcZ, might sensitize cells to antibiotics without deleting a global regulator outright.

Methodologically, the study combines several layers of evidence. It compares antibiotic susceptibility of wild-type and hfq deletion strains in both PAO1 and PA14 backgrounds, uses RNA-seq to identify antibiotic-relevant processes influenced by Hfq, and then ties those transcriptomic changes to physiological phenotypes such as membrane potential, c-di-GMP levels, and biofilm formation. In parallel, the paper tests whether higher CrcZ levels, induced either by plasmid overexpression or by growth on non-preferred carbon sources, can reproduce the sensitizing effect. That combination matters because it turns the paper from a descriptive regulatory study into a proof-of-principle for metabolic re-sensitization.

The result is not a single neat pathway but a broader systems picture. Loss of Hfq affects multiple determinants of antibiotic susceptibility at once, from uptake and efflux functions to membrane-associated traits and energy metabolism. The paper therefore argues that the increased susceptibility of the hfq mutant is a composite phenotype rather than the consequence of one master resistance gene. The biology is messier that way, but it is also more realistic. Antibiotic response in Pseudomonas is usually shaped by overlapping physiological layers rather than by a single switch.

What makes the paper especially interesting is that the CrcZ experiments point toward a controllable lever. Overproducing CrcZ increases sensitivity to gentamicin, and growth on oxaloacetate, a non-preferred carbon source that induces CrcZ, shifts susceptibility in the same direction. In complex synthetic cystic fibrosis sputum medium, adding oxaloacetate also lowers the MIC for gentamicin and cefepime. That gives the paper its practical angle: metabolic context can be used to bias the regulatory state of the bacterium toward a more antibiotic-sensitive phenotype.

The paper is an important bridge to later work on oprD, opdP, Crc, and carbon catabolite repression. It establishes the broader principle that Hfq-dependent post-transcriptional regulation is tied to antibiotic susceptibility in a metabolically responsive way. Later studies dissect individual branches of that logic in more detail. Here, the emphasis stays on the network level, where Hfq availability, modulated by CrcZ, changes how the cell handles drugs across several mechanistic layers.

The paper is also a good example of why bacterial RNA regulation matters beyond classical sRNA target maps. Regulatory RNAs such as CrcZ can have indirect but clinically meaningful effects because they redistribute the activity of central RNA-binding proteins. In this case, a carbon-source-sensitive decoy RNA becomes part of a strategy for shifting susceptibility to aminoglycosides and beta-lactams. That is a strong illustration of how metabolism and antimicrobial response are coupled in opportunistic pathogens.

Citation

Harnessing Metabolic Regulation to Increase Hfq-Dependent Antibiotic Susceptibility in Pseudomonas aeruginosa
Petra Pusic, Elisabeth Sonnleitner, Beatrice Krennmayr, Dorothea Agnes Heitzinger, Michael T. Wolfinger, Armin Resch, Udo Blasi
Front. Microbiol. 9:2709 (2018) | doi:10.3389/fmicb.2018.02709 | PDF

Searching for Telomerase RNAs in Saccharomycetes is TERribly difficult

2018-07-26T00:00:00+02:00

Telomerase RNA is a frustrating target for computational annotation. Functionally, it is essential: without it, telomerase cannot maintain chromosome ends. But unlike many better-behaved non-coding RNAs, telomerase RNAs often evolve rapidly in primary sequence, vary strongly in length, and tolerate substantial structural reorganization. That makes them precisely the kind of molecule that defeats simple homology search.

This paper focuses on that problem in Saccharomycete yeasts. The central question is not whether telomerase RNAs exist in these genomes, but how one can find them when neither sequence conservation nor a single fixed structural model is strong enough to carry the search on its own. The title is not rhetorical. The paper is genuinely about why this annotation problem is difficult, and what a realistic computational strategy looks like when the target family is both fast-evolving and structurally plastic.

The study combines multiple approaches rather than relying on one decisive signal. Sequence similarity, comparative context, and ViennaRNA-supported structural reasoning are used together to build and refine search models across subgroups. That point is important because it captures a recurring lesson in non-coding RNA bioinformatics: when the biology is heterogeneous, the right answer is often not a more aggressive version of a single method, but a carefully staged combination of several weak but complementary signals.

Even with that broader strategy, the outcome is only partially complete, and that is one of the strengths of the paper. Instead of overstating success, it documents the limits of the current search space. The authors identify 27 new telomerase RNAs, but only within the subgroup Saccharomycetaceae, and even there different phylogenetic subgroups require different search models. More distant branches of Saccharomycotina remain unresolved. In other words, the paper is as much about the boundaries of current annotation methodology as it is about the annotations themselves.

That honesty makes the paper more useful than a narrower success story would have been. Telomerase RNAs are a classic example of a family where absence of annotation is not evidence of absence. They are simply hard to find. By spelling out which features help, which ones fail, and where the search breaks down, the paper becomes a methodological reference for anyone working on difficult structured RNAs that have retained function while drifting in sequence and architecture.

The broader significance is easy to miss if one focuses only on yeast telomeres. This is really a paper about ncRNA discoverability. Many computational pipelines work best on families with strong covariation support, stable consensus motifs, or relatively conserved lengths. Telomerase RNA violates those expectations. As a result, the paper becomes a useful case study in how to adapt comparative RNA annotation strategies when the target family sits at the edge of what standard homology models can capture.

The same underlying issue appears repeatedly across RNA biology: biologically important RNAs are not always easy to recognize from sequence alone, and structure-aware comparative methods become essential precisely where simple pipelines fail. Telomerase RNA in yeasts is one of the clearest examples of that problem.

Abstract

The telomerase RNA in yeasts is large, usually >1000 nt, and contains functional elements that have been extensively studied experimentally in several disparate species. Nevertheless, they are very difficult to detect by homology-based methods and so far have escaped annotation in the majority of the genomes of Saccharomycotina. This is a consequence of sequences that evolve rapidly at nucleotide level, are subject to large variations in size, and are highly plastic with respect to their secondary structures. Here, we report on a survey that was aimed at closing this gap in RNA annotation. Despite considerable efforts and the combination of a variety of different methods, it was only partially successful. While 27 new telomerase RNAs were identified, we had to restrict our efforts to the subgroup Saccharomycetacea because even this narrow subgroup was diverse enough to require different search models for different phylogenetic subgroups. More distant branches of the Saccharomycotina remain without annotated telomerase RNA.

Citation

TERribly Difficult: Searching for Telomerase RNAs in Saccharomycetes
Maria Waldl, Bernhard C. Thiel, Roman Ochsenreiter, Alexander Holzenleiter, João Victor de Araujo Oliveira, Maria Emília M.T. Walter, Michael T. Wolfinger, Peter F. Stadler
Genes 9(8),372 (2018) | doi: 10.3390/genes9080372 | PDF

Co-transcriptional riboswitch modeling with ViennaRNA

2018-07-01T00:00:00+02:00

Population dynamics of competing riboswitch conformations during cotranscriptional folding

Riboswitches are a good reminder that RNA function is often kinetic rather than purely thermodynamic. For many ligand-sensing RNAs, the biologically relevant question is not simply which structure has the lowest free energy at equilibrium. What matters is which structures become available during transcription, how long they persist, and whether ligand binding happens early enough to redirect the folding pathway before an alternative conformation takes over.

That is the problem addressed in this paper. The goal is to model cotranscriptional RNA-ligand interaction dynamics efficiently enough to say something mechanistic about a kinetically controlled riboswitch, without having to rely entirely on huge numbers of direct folding trajectories. The approach builds on the energy-landscape framework developed in earlier work and extends it to a setting where both the RNA chain and the ligand-binding state change over time.

The methodological core is an extension of the BARRIERS/BarMap style of coarse graining. For each transcript length, the folding landscape is partitioned into macrostates, representing basins of attraction around locally stable structures. Those macrostates are then linked between successive transcript lengths so that population densities can be transferred forward as transcription proceeds. What makes this paper new is that ligand competence is incorporated directly into that framework. Structures that can bind ligand are treated differently from those that cannot, and ligand binding or unbinding is modeled in a concentration-dependent way rather than as a static annotation.

The simplifying assumption is deliberately clear: ligand binding is treated in an all-or-none fashion for a defined binding-competent structure. That sounds restrictive, but it is exactly the kind of abstraction that makes this model useful. It keeps the state space manageable while still capturing the central competition between RNA folding timescales and metabolite binding timescales. In practice, that is the core of many riboswitch mechanisms.

The case study is the I-A 2'-deoxyguanosine (2'dG) riboswitch from Mesoplasma florum, which had already been characterized experimentally by NMR. That makes it a strong benchmark system, because the computational model can be compared against an independently studied folding mechanism rather than against another computational prediction. The figure above captures the main idea well: as the transcript elongates, the dominant population shifts through a series of metastable states, and ligand binding changes which path remains accessible.

The main result is that this landscape-based treatment reproduces the logic of a kinetically controlled riboswitch surprisingly well. The model shows how an early binding-competent aptamer state can capture ligand and thereby steer the RNA away from competing conformations that would otherwise dominate later. The riboswitch cannot be understood from a single end-state structure alone. Its behavior depends on the timing of transcription, the accessibility of intermediate states, and the concentration-dependent opportunity for ligand capture during folding.

I still think this is one of the more useful methodological papers in the riboswitch area. It turns a vague statement like "cotranscriptional effects matter" into a concrete computational workflow. Instead of treating kinetics as an afterthought, the paper makes the dynamic landscape itself the object of study. This is relevant not only for natural riboswitches, but also for synthetic biology, where one often wants to know whether a designed switch is merely structurally plausible or actually kinetically workable.

The paper also connects two strands of RNA research that are often discussed separately: landscape-based folding kinetics and ligand-regulated RNA control. Bringing them together makes it possible to use coarse-grained kinetics not just for descriptive folding studies, but for mechanism-aware analysis of regulatory RNAs. That is a meaningful step beyond equilibrium folding and a useful basis for later in silico screening of switch designs before experimental validation.

That screening logic sits very close to Why kinetic folding matters in RNA design. A construct can look plausible at equilibrium and still fail once timing and pathway dependence are taken seriously.

Citation

Efficient Computation of Cotranscriptional RNA-Ligand Interaction Dynamics
Michael T. Wolfinger, Christoph Flamm, Ivo L. Hofacker
Methods 143:70–76 (2018) | doi: 10.1016/j.ymeth.2018.04.036 | PDF

In silico design of ligand-triggered RNA switches

2018-07-01T00:00:00+02:00

Designing a useful ligand-triggered RNA switch is not just a matter of finding a sequence that can fold into two states. The real challenge is to engineer a sequence whose relevant conformations, ligand competence, and switching kinetics all fit the intended mechanism.

This work lays out a concrete in silico workflow for that problem. It starts from a well-characterized aptamer, exemplified here with the theophylline aptamer, and treats ligand binding as one structurally defined state of the switch. The design task then becomes balancing that binding-competent conformation against an alternative fold that disrupts it in the absence of ligand.

One useful contribution of the paper is that it makes the objective function explicit. Instead of treating switch design as a vague search for "good" sequences, the workflow defines quantitative criteria for what the sequence should do. That matters because RNA design tends to fail when the desired mechanism is underspecified. If the design objective does not encode the structural logic of the switch, ranking candidate sequences quickly becomes guesswork.

The other key point is kinetics. For many RNA design problems, equilibrium structure alone is not enough. A candidate may satisfy static structural constraints and still perform poorly if the switching pathway is too slow, too indirect, or dominated by off-target intermediates. That is why the workflow includes an analysis of folding kinetics instead of stopping at secondary structure prediction.

The result is a design pipeline that helps filter and rank candidate sequences before any experimental work begins. It does not guarantee that a proposed switch will function in a cellular context, and it does not replace experimental validation. It does offer a principled way to reduce the search space and focus attention on sequences whose structures and dynamic behavior are at least consistent with the intended design.

For computational RNA biology, that is the real value of this kind of work. It turns riboswitch design from an intuition-led exercise into an optimization problem with explicit structural and kinetic criteria. That framing also makes the workflow adaptable. Once the design assumptions are clear, the same logic can be extended to other aptamers, other switching scenarios, and more elaborate regulatory mechanisms.

The same pre-synthesis question appears whenever design objectives, competing folds, and likely failure modes have to be examined before a candidate sequence becomes expensive.

Abstract

This contribution sketches a work flow to design an RNA switch that is able to adapt two structural conformations in a ligand-dependent way. A well characterized RNA aptamer, i.e., knowing its Kd and adaptive structural features, is an essential ingredient of the described design process. We exemplify the principles using the well-known theophylline aptamer throughout this work. The aptamer in its ligand-binding competent structure represents one structural conformation of the switch while an alternative fold that disrupts the binding-competent structure forms the other conformation. To keep it simple we do not incorporate any regulatory mechanism to control transcription or translation. We elucidate a commonly used design process by explicitly dissecting and explaining the necessary steps in detail. We developed a novel objective function which specifies the mechanistics of this simple, ligand-triggered riboswitch and describe an extensive in silico analysis pipeline to evaluate important kinetic properties of the designed sequences. This protocol and the developed software can be easily extended or adapted to fit novel design scenarios and thus can serve as a template for future needs.

Citation

In silico design of ligand triggered RNA switches
Sven Findeiß, Stefan Hammer, Michael T. Wolfinger, Felix Kühnl, Christoph Flamm, Ivo L.Hofacker
Methods 143:90-101 (2018) | doi: 10.1016/j.ymeth.2018.04.003 | Preprint PDF

How Crc modulates Hfq-dependent RNA regulation in Pseudomonas aeruginosa

2018-01-29T00:00:00+01:00

Crc stabilizes Hfq/RNA complexes in vitro

This paper is the mechanistic centerpiece of the Pseudomonas aeruginosa Hfq/Crc/CrcZ story. Earlier work had already made it clear that carbon catabolite repression in Pseudomonas does not work like the textbook cAMP-CRP systems familiar from enteric bacteria. Instead, it operates largely at the post-transcriptional level and depends on the RNA chaperone Hfq, the catabolite repression control protein Crc, and the regulatory RNA CrcZ. What had remained unclear was how Crc actually contributes to Hfq-dependent repression. This paper addresses exactly that point.

The key result is that Crc is not simply another independent regulator acting in parallel to Hfq. Rather, it enhances Hfq-mediated translational repression by forming higher-order assemblies with Hfq and A-rich target RNAs. The study shows that Crc does not bind productively to Hfq alone, nor to RNA alone, under the relevant conditions. Instead, RNA bound to the distal side of Hfq provides the context in which Crc can engage. In that configuration, Crc stabilizes the Hfq/RNA complex and thereby strengthens translational silencing of catabolic mRNAs.

That finding matters because it changes the conceptual picture of carbon catabolite repression in Pseudomonas. Hfq is the primary RNA-binding repressor, but Crc acts as a proteinaceous modulator that increases the lifetime and effectiveness of the repressive complex. This is more interesting than a simple cofactor model, because it shows how an RNA-binding protein can be tuned by another protein through assembly on a shared RNA substrate. In that sense, the paper is not just about Pseudomonas physiology. It is also about a general principle of post-transcriptional control.

Methodologically, the paper is unusually rich. It combines RNA-seq with bacterial two-hybrid assays, co-immunoprecipitation, microscale thermophoresis, electrophoretic mobility shift assays, UV and chemical cross-linking, and structural interpretation of the interaction surfaces. That breadth is important because the claim is mechanistic and multi-part: Crc and Hfq associate in vivo, the association depends on the right RNA context, Crc contacts both Hfq and RNA in the complex, and the resulting assembly is more stable than the Hfq/RNA complex alone. No single technique would have been enough to make that case convincingly.

Another important aspect of the paper is that Crc does not just strengthen one branch of Hfq activity. It also interferes with access of at least one regulatory sRNA to the proximal side of Hfq. In the experiments shown here, the presence of Crc reduces effective binding of the sRNA PrrF2 to Hfq, and corresponding in vivo data suggest that this can affect sRNA-mediated riboregulation. That point gives the paper broader significance: Crc is not merely a helper for catabolic repression, it can also bias how Hfq is allocated between carbon-source prioritization and other RNA-regulatory tasks.

This is where the physiological interpretation becomes especially useful. The authors propose a working model in which Crc helps prioritize Hfq function toward the utilization of favored carbon sources during carbon catabolite repression. In other words, when the cell grows on a preferred substrate, Crc helps keep Hfq focused on shutting down unnecessary catabolic programs. When CrcZ accumulates after relief of CCR, Hfq is sequestered away from these target mRNAs and repression is lifted. That logic is exactly what later papers on antibiotic susceptibility and porin regulation build on.

This paper provides the molecular explanation for several phenotypes that can otherwise look disconnected. Harnessing Metabolic Regulation to Increase Hfq-Dependent Antibiotic Susceptibility in Pseudomonas aeruginosa uses CrcZ to shift Hfq-dependent antibiotic susceptibility. Distinctive Regulation of Carbapenem Susceptibility in Pseudomonas aeruginosa by Hfq shows that opdP is controlled through an Hfq/Crc complex while oprD follows a different Hfq/sRNA route. Rewiring of Gene Expression in Pseudomonas aeruginosa During Diauxic Growth Reveals an Indirect Regulation of the MexGHI-OpmD Efflux Pump by Hfq expands that logic to broader transcriptome rewiring. This 2018 NAR paper lays out the physical and biochemical basis of that regulatory architecture.

For readers interested in bacterial RNA biology more generally, the study is also notable because it gives one of the clearest examples of how an RNA chaperone can be functionally redirected by another protein. Hfq is often discussed mainly in the context of sRNA-mediated regulation. Here, the emphasis shifts to a different side of its biology: target-specific translational repression shaped by an interacting partner and by the availability of competing RNAs. That makes the paper both mechanistically satisfying and central to the broader Pseudomonas cluster.

Citation

Interplay Between the Catabolite Repression Control Protein Crc, Hfq and RNA in Hfq-Dependent Translational Regulation in Pseudomonas aeruginosa
Elisabeth Sonnleitner, Alexander Wulf, Sebastien Campagne, Xue-Yuan Pei, Michael T. Wolfinger, Giada Forlani, Konstantin Prindl, Laetitia Abdou, Armin Resch, Frederic Allain, Ben Luisi, Henning Urlaub, Udo Blasi
Nucleic Acids Res. 46:1470-1485 (2018) | doi:10.1093/nar/gkx1245 | PDF

PaiI links anaerobic small-RNA regulation to denitrification in Pseudomonas aeruginosa

2017-11-23T00:00:00+01:00

PaiI deletion causes transient nitrite accumulation and reduced nitrite reductase activity during anaerobic growth

This paper continues the anaerobic Pseudomonas aeruginosa story, but from a more focused regulatory angle. Instead of asking how the whole transcriptome changes during oxygen limitation, it asks whether a specific small RNA helps the bacterium adapt to nitrate-respiring growth. That is an important question because chronic Pseudomonas infections in cystic fibrosis lungs often involve oxygen-poor, biofilm-associated conditions where denitrification becomes physiologically relevant. If there are dedicated anaerobiosis-induced sRNAs in this setting, they are likely to be part of the fine-tuning layer that sits on top of the classical transcriptional denitrification cascade.

The main result is the identification and first characterization of PaiI, a small RNA that is strongly induced under anaerobic conditions in the presence of nitrate. Its expression depends on the NarXL two-component system, which immediately places it inside the nitrate-responsive regulatory network rather than as a generic stress transcript. The paper then shows that PaiI is not just a marker of anaerobiosis. A paiI deletion mutant displays a clear physiological phenotype under denitrifying conditions, particularly when glucose is used as the carbon source.

Methodologically, the paper grows out of earlier RNA-seq work on PA14 under anoxic conditions. Candidate sRNAs were identified from the anaerobic transcriptome data and then followed up experimentally by Northern blotting, promoter analysis, mutant construction, and physiological assays. That progression matters: the study starts with a transcriptomics observation, but it does not stop there. It moves quickly into targeted genetics and phenotype measurements, which is what makes PaiI credible as a functional regulator rather than just another induced RNA band on a blot.

The key phenotype is a defect in efficient denitrification. In the absence of PaiI, the cultures accumulate more nitrite and show reduced nitrite reductase activity, indicating a problem in the conversion of nitrite to nitric oxide. The deletion strain is also impaired in anaerobic growth on glucose, and that phenotype can be reconciled with reduced glucose uptake under these conditions. In other words, the paper ties a small RNA to a very concrete physiological bottleneck within nitrate respiration rather than to an abstract stress response.

An interesting aspect of the study is that the effect appears to be indirect. The transcriptome data did not reveal major changes in the abundance of the canonical nir transcripts, and the nitrite reductase protein itself was not simply lost in the mutant. Overexpression of dnr could complement the deletion phenotype, which places PaiI functionally close to the denitrification control circuitry without reducing it to a trivial one-step mechanism. That makes the biology more subtle, but also more realistic: many small RNAs in bacteria shape pathway output through network effects rather than by acting as on/off switches for one obvious target.

The in vivo angle strengthens the paper further. The paiI deletion strain was impaired in colonizing murine tumors, a model that contains hypoxic or anaerobic regions and therefore stresses the same nitrate-respiring physiology that the in vitro assays probe. That does not turn PaiI into a classical virulence factor paper, but it does show that the anaerobic-growth phenotype is not just a laboratory curiosity. The small RNA matters under host-relevant low-oxygen conditions.

Seen together with the 2016 PA14 anoxic-transcriptome study, this work is a nice example of how broad RNA-seq surveys can lead to more focused mechanistic follow-up. The earlier paper mapped the large-scale physiological shift into long-term anaerobic growth. This one takes one candidate from that landscape and asks what it actually does. The answer is that PaiI helps the cell execute denitrification efficiently, especially at the nitrite-reduction step, and thereby supports anaerobic adaptation.

For readers interested in bacterial RNA biology, the paper is also a reminder that Pseudomonas sRNA research is not limited to Hfq-dependent envelope or carbon-catabolite regulation. Under anaerobic conditions, small RNAs can also intersect directly with respiration-linked physiology. That makes PaiI a useful addition to the site’s broader non-coding RNA theme, even though the biological setting is very different from the viral and structural RNA work elsewhere on the page.

Citation

The Anaerobically Induced sRNA PaiI Affects Denitrification in Pseudomonas aeruginosa PA14
Muralidhar Tata, Fabian Amman, Vinay Pawar, Michael T. Wolfinger, Siegfried Weiss, Susanne Haussler, Udo Blasi
Front. Microbiol. 8:2312 (2017) | doi:10.3389/fmicb.2017.02312 | PDF

How archaeal Sm-like proteins shape RNA tailing

2017-05-18T00:00:00+02:00

This paper sits in a part of RNA biology that is less visible than translation or transcription, but just as fundamental: what happens to RNA after it has been made. In bacteria, eukaryotes, and archaea, RNA stability depends on a network of RNA-binding proteins, exonucleases, and tailing activities that mark transcripts for processing or decay. In archaea, one of the central players in that machinery is the exosome. What this paper adds is a mechanistic link between the archaeal exosome and the small Sm-like proteins SmAP1 and SmAP2.

The main finding is that these archaeal Sm-like proteins are not just generic RNA binders floating elsewhere in the cell. In Sulfolobus solfataricus, SmAP1 and SmAP2 physically interact with the exosome and stimulate the addition of A-rich tails to transcripts. That matters because tailing is not a decorative end modification. It changes how RNAs are recognized and processed, and it can promote RNA turnover by giving decay machinery a more accessible handle.

Conceptually, the paper is interesting because it places archaeal SmAP proteins into a functional role that goes beyond the simple statement "they bind RNA". Sm and Lsm family proteins are broadly associated with RNA metabolism across the tree of life, but the exact wiring differs by system. Here, the authors show that the archaeal versions are tied directly to an RNA-degradation complex and can influence its output. That gives the proteins a concrete place in the architecture of archaeal RNA homeostasis.

Methodologically, the study combines protein interaction assays with functional readouts of RNA tailing. The interaction side establishes that SmAP1/2 and the exosome are associated, while the biochemical assays show that this association is not merely structural. The presence of the Sm-like proteins changes the behavior of the exosome by stimulating A-rich tail addition. That combination of physical interaction and functional consequence is what makes the conclusion convincing.

The biological picture that emerges is that RNA fate in archaea is coordinated through multiprotein assemblies rather than isolated enzymes acting one by one. An exosome subunit may perform the catalytic step, but the efficiency and transcript context of that step can be shaped by companion RNA-binding proteins. That is exactly the kind of regulatory layering that makes RNA metabolism interesting: the key question is often not whether an enzyme exists, but how access to RNA substrates is organized.

This also gives the paper broader significance beyond Sulfolobus itself. Archaeal RNA biology is often discussed as a mix of bacterial-style and eukaryotic-style features, but studies like this show that it has its own logic. Sm-like proteins, tailing reactions, and exosome function are all familiar individually, yet their combination in archaea produces a distinct regulatory system. That makes the paper useful both as a mechanistic study and as a reminder that archaeal post-transcriptional control deserves to be understood on its own terms.

This article forms a natural pair with A moonlighting role for archaeal aIF5A. Both papers are about archaeal proteins that sit at the interface between canonical textbook roles and broader RNA-metabolic functions. Together they show how proteins shape the life cycle of RNA in archaea.

Citation

The SmAP1/2 Proteins of the Crenarchaeon Sulfolobus Solfataricus Interact with the Exosome and Stimulate A-Rich Tailing of Transcripts
Birgit Märtens, Linlin Hou, Fabian Amman, Michael T. Wolfinger, Elena Evguenieva-Hackenberg, Udo Bläsi
Nucleic Acids Res. 45:7938-7949 (2017) | doi:10.1093/nar/gkx437 | PDF

Co-transcriptional folding and metastable states in riboswitch function

2017-01-31T00:00:00+01:00

Riboswitches regulate gene expression by changing conformation in response to ligand binding, but the decisive structural events often happen while the RNA is still being transcribed. This study focuses on the type I-A 2'dG-sensing riboswitch from Mesoplasma florum and asks how transcript length, ligand binding, and metastable intermediates interact during that process.

What makes this paper stand out is the experimental resolution of that question. Rather than inferring intermediates only indirectly, the study used NMR spectroscopy to characterize all relevant transcription intermediates of the riboswitch at single-nucleotide resolution. That provides a rare view of how the accessible conformational space changes as the RNA grows, and it shows very clearly that riboswitch function cannot be understood from the final full-length structure alone.

The methodological approach is worth emphasizing because it differs from many purely computational studies. The RNA was analyzed as a series of length-defined transcription intermediates, allowing the authors to ask, step by step, which helices and alternative folds become available at each stage of transcription. In effect, the work turns cotranscriptional folding into an experimentally accessible sequence of structural snapshots. That is exactly the kind of information that is usually missing when one tries to explain riboswitch behavior from equilibrium models alone.

The main finding is that ligand responsiveness is tightly coupled to transcript length and to the presence of metastable conformations. Certain intermediates transiently expose a binding-competent arrangement, while slightly longer transcripts can already open competing structural options that redirect the switch. In other words, the riboswitch operates within a narrow kinetic window. The regulatory outcome depends on whether ligand binding happens at the right moment, before alternative folds become dominant.

That point matters well beyond this one riboswitch. The paper makes a broader argument that metastable RNA states are not just folding noise on the way to the “real” structure. They can be the mechanistically decisive states. For transcriptional riboswitches in particular, regulation emerges from the coupling of synthesis, folding, and binding, not from equilibrium thermodynamics alone. This is one of the clearest experimental demonstrations of that idea.

For readers interested in RNA design or synthetic biology, this is also the real lesson of the paper. If a regulatory RNA works by passing through a specific sequence of transient states, then designing only for the final minimum free energy structure is not enough. One has to think in terms of folding pathways and timing. That perspective became the basis for later computational work on the same 2'dG riboswitch system, including the follow-up landscape-based analysis linked below.

Abstract

Gene repression induced by the formation of transcriptional terminators represents a prime example for the coupling of RNA synthesis, folding, and regulation. In this context, mapping the changes in available conformational space of transcription intermediates during RNA synthesis is important to understand riboswitch function. A majority of riboswitches, an important class of small metabolite-sensing regulatory RNAs, act as transcriptional regulators, but the dependence of ligand binding and the subsequent allosteric conformational switch on mRNA transcript length has not yet been investigated. We show a strict fine-tuning of binding and sequence-dependent alterations of conformational space by structural analysis of all relevant transcription intermediates at single-nucleotide resolution for the I-A type 2′dG-sensing riboswitch from Mesoplasma f lorum by NMR spectroscopy. Our results provide a general framework to dissect the coupling of synthesis and folding essential for riboswitch function, revealing the importance of metastable states for RNA-based gene regulation.

Citation

NMR Structural Profiling of Transcriptional Intermediates Reveals Riboswitch Regulation by Metastable RNA Conformations
Christina Helmling, Anna Wacker, Michael T. Wolfinger, Ivo L. Hofacker, Martin Hengsbach, Boris Fürtig, Harald Schwalbe
J. Am. Chem. Soc. 139 (7):2647–56 (2017) | doi:10.1021/jacs.6b10429

CrcZ cross-regulates anoxic biofilm formation in Pseudomonas aeruginosa

2016-12-21T00:00:00+01:00

CrcZ levels modulate viable biomass in anoxic Pseudomonas aeruginosa biofilms

This paper is the point where the CrcZ story starts to expand beyond carbon metabolism. CrcZ was already known as the decoy RNA that sequesters Hfq when carbon catabolite repression is relieved, thereby allowing expression of genes needed for the use of less preferred carbon sources. What this study asks is whether that same competition for Hfq can spill over into a completely different physiological process: formation of anaerobic biofilms by Pseudomonas aeruginosa under cystic-fibrosis-like conditions.

That question turns out to be well chosen. The paper shows that CrcZ is by far the most abundant Hfq-bound regulatory RNA in PA14 anoxic biofilms grown in synthetic cystic fibrosis sputum medium. Since Hfq itself proves to be important for anaerobic biofilm formation, this immediately suggests a potential cross-regulatory mechanism. If CrcZ soaks up a substantial fraction of Hfq under these conditions, then it may indirectly reshape a broad set of Hfq-dependent processes that have nothing to do with carbon uptake in the narrow sense.

The main biological result is that this is exactly what happens. Deleting crcZ increases anoxic biofilm formation, while overproducing CrcZ reduces it to a level comparable with the hfq deletion mutant. Confocal microscopy and biomass quantification further show that CrcZ levels influence the balance between viable and dead cells in these biofilms. In other words, CrcZ is not just a marker of altered metabolic state. It actively constrains the development of anaerobic biofilms by competing for Hfq.

Methodologically, the study combines RNA-centric and physiology-centric approaches in a useful way. Hfq-bound RNAs were identified by co-immunoprecipitation and RNA-seq, which established the strong enrichment of CrcZ in the bound fraction. This was then followed by targeted analysis of crcZ, hfq, and combined mutant or overexpression strains under anoxic biofilm conditions. The authors also used transcriptome analysis of the hfq mutant and physiological readouts such as metabolic activity, redox balance, crystal-violet assays, and confocal imaging. That range is important because the claim is inherently indirect: CrcZ does not form biofilms itself, it changes the availability of a global RNA chaperone that then alters multiple downstream pathways.

The broader implication is that regulatory decoy RNAs can cross-regulate functions outside the pathways for which they were first discovered. Here, CrcZ couples the nutritional state of the cell to anaerobic biofilm formation by redistributing Hfq. That is a conceptually strong result, because it shows how one abundant RNA can bias the use of a central post-transcriptional regulator toward one physiological program and away from another.

This paper also provides the bridge to several later Pseudomonas studies. The 2018 NAR paper explains at the molecular level how Crc, Hfq, and RNA assemble into repressive complexes during carbon catabolite repression. The 2018 metabolic-sensitization paper uses CrcZ-mediated Hfq sequestration to alter antibiotic susceptibility. The 2020 porin paper then dissects how Hfq and Crc regulate specific antibiotic entry pathways. This 2016 study is where the physiological reach of CrcZ first becomes obvious.

It is also worth noting that the setting here is highly relevant to chronic infection biology. The experiments were done in a medium designed to mimic cystic fibrosis sputum and under oxygen-limited conditions that approximate the microenvironments in established infections. That makes the work more than a basic-regulation paper. It shows that carbon-responsive RNA control is wired into a host-relevant persistence phenotype.

Citation

Cross-Regulation by CrcZ RNA Controls Anoxic Biofilm Formation in Pseudomonas aeruginosa
Petra Pusic, Muralidhar Tata, Michael T. Wolfinger, Elisabeth Sonnleitner, Susanne Haussler, Udo Blasi
Sci. Rep. 6:39621 (2016) | doi:10.1038/srep39621 | PDF

How inverted SINEs repress gene expression

2016-10-25T00:00:00+02:00

Reporter constructs showing that inverted SINE pairs reduce luciferase output and RNA levels

This paper takes a familiar genome-annotation object, the short interspersed element, and asks a sharper regulatory question: what happens when nearby SINEs occur in inverted orientation within transcripts? In primates, the most prominent examples are Alu elements. They are abundant, often located in introns or untranslated regions, and are well known to form double-stranded RNA structures when inverted copies appear close enough to base pair. The interesting point is that such structures had already been linked to several possible outcomes, including RNA editing, nuclear retention, and translational control, but it was still unclear what their dominant transcriptome-wide effect actually is.

The first part of the paper therefore looks at the genome and transcriptome at scale. Using human annotation data and ENCODE RNA-seq profiles, the study shows two related patterns. First, closely spaced inverted Alu pairs are less common than tandemly arranged pairs, suggesting that they are disfavored over evolutionary time. Second, transcripts carrying inverted SINE arrangements are expressed at lower levels than comparable transcripts with tandem or single SINEs. That matters because it turns a collection of anecdotal locus-specific observations into a broader statistical signal.

The paper then does the more important thing and tests mechanism directly. Reporter constructs carrying natural or engineered SINE arrangements in their 3'UTRs were compared in cell culture. These experiments show that inverted SINEs reduce both reporter protein output and RNA abundance, whereas tandem arrangements have a weaker effect. The repression also scales with the quality of the double-stranded structure: constructs that can form more perfect intramolecular duplexes show stronger loss of expression. So the phenomenon is not simply “having an Alu” but having an arrangement that favors structured pairing within the transcript.

What makes the paper especially interesting is where it ends up mechanistically. The authors explicitly tested several obvious double-stranded-RNA explanations and found that they do not account for the effect. The repression was not explained by ADAR-mediated editing, not by STAUFEN binding, and not by the classical cytoplasmic dsRNA sensors one might first suspect. Instead, the evidence pointed toward a different mechanism: transcriptional elongation by RNA polymerase II becomes less efficient across constructs carrying inverted SINEs, leading to reduced transcript output. That is a stronger and more surprising conclusion than simply saying that dsRNA structures destabilize mature RNA.

In other words, the paper moves the problem upstream. The crucial effect of inverted SINEs is not only on the fate of an already completed transcript, but on the process of making that transcript in the first place. If the polymerase is slowed or impaired by the sequence and structural context associated with inverted repeats, then the cell pays a transcriptional cost for placing too much intramolecular pairing potential into transcribed regions. That provides a plausible mechanistic reason why such arrangements are comparatively underrepresented in the genome.

This is also why the paper remains interesting beyond transposon biology. It connects RNA structure, repetitive-sequence organization, and transcriptional control in one framework. Genome evolution is usually discussed in terms of sequence composition and selection on coding potential, but here the selective pressure may also act through the physical behavior of the resulting RNA and its impact on Pol II. That is a genuinely integrative idea.

From the perspective of the rest of the site, this study belongs to the broader non-coding RNA and RNA-structure theme, even though the biological setting is very different from viral UTRs or bacterial sRNAs. The common thread is that RNA structure is not passive. Whether in regulatory RNAs, untranslated regions, or repetitive elements, it can feed back onto gene expression in concrete mechanistic ways. In this case, inverted SINEs are a large-scale example of that principle in mammalian transcriptomes.

Citation

Transcriptome-Wide Effects of Inverted SINEs on Gene Expression and Their Impact on RNA Polymerase II Activity
Mansoureh Tajaddod, Andrea Tanzer, Konstantin Licht, Michael T. Wolfinger, Stefan Badelt, Florian Huber, Oliver Pusch, Sandy Schopoff, Michael Janisiw, Ivo Hofacker, Michael F. Jantsch
Genome Biol. 17:220 (2016) | doi:10.1186/s13059-016-1083-0 | PDF

Why bats and humans respond differently to filovirus infection

2016-10-07T00:00:00+02:00

This paper addresses one of the most important biological contrasts in filovirus research. Ebola and Marburg viruses cause severe, often fatal disease in humans, yet bats are thought to serve as natural hosts without showing the same destructive pathology. That difference is not just an ecological curiosity. It points to a host-response problem: what do bat cells do differently, and what does the human cell response look like when infection becomes damaging instead of tolerated?

The study approaches that question through comparative transcriptomics. Human HuH7 cells and bat R06E-J cells were infected with either Ebola virus or Marburg virus, and RNA-seq profiles were collected across several time points representing early, intermediate, and later phases of the infection cycle. That design matters because host response is dynamic. A single endpoint would blur together primary responses, downstream stress programs, and viral amplification effects. By sampling the time course, the paper can distinguish faster from slower trajectories and compare how two related filoviruses reshape host transcription in two very different cellular backgrounds.

One of the clearest results is kinetic. Filovirus replication proceeds more rapidly in the human cells than in the bat cells. That observation already hints at a mechanistic difference in permissiveness or control. The transcriptomic analysis then shows that the infected cells do not simply differ in the magnitude of a common response. They differ in which genes, motifs, and pathways become most strongly engaged. Among the most prominent regulated classes are chemokine ligands and transcription factors, indicating that immune signaling and regulatory rewiring are central features of the infection response.

The pathway-level results are especially useful. The paper reports strong activation of the JAK/STAT axis, induction of several dual-specificity phosphatases (DUSP genes) linked to MAP kinase regulation, and upregulation of PPP1R15A, a marker connected to endoplasmic-reticulum stress and stress-induced cell-death programs. That combination is informative because it suggests that the host response is not only antiviral in a narrow sense. It also involves broader stress adaptation and signaling-control modules that may shape whether infection remains contained or progresses toward pathology.

Another important contribution is infrastructural rather than purely biological. At the time, the transcriptional response of bat cells to filovirus infection had not been characterized in comparable depth, and even the human-cell picture was incomplete. The study therefore had to do more than differential-expression testing alone. It also established a transcriptomic resource, including de novo assembly work for the bat system, to make cross-species comparison possible. That resource-building aspect is easy to miss, but it is a major reason why the paper has remained useful.

What makes the study compelling is that it treats the bat-human contrast as a systems-level problem. The question is not reduced to one receptor, one interferon gene, or one viral antagonist. Instead, the authors look at coordinated host programs: transcription factors, activity motifs, pathways, and infection-stage-dependent responses. That is the right scale for a problem where tolerance likely emerges from network behavior rather than a single switch.

Seen from today’s perspective, the paper also illustrates an approach that has only become more important: using comparative host transcriptomics to identify candidate cellular states associated with tolerance, resilience, or severe disease. In that sense, this is not just a filovirus paper. It is part of a broader shift toward understanding infection through host regulatory landscapes rather than viral replication alone.

The article stands apart from structure-centered virology because it shows a different register of virus bioinformatics: large-scale RNA-seq analysis, de novo transcriptome reconstruction, pathway interpretation, and cross-species comparison. That broader systems perspective matters because host response, viral evolution, and structured RNA biology often end up informing one another even when the immediate question is different.

Citation

Differential Transcriptional Responses to Ebola and Marburg Virus Infection in Bat and Human Cells
Martin Hölzer, Verena Krähling, Fabian Amman, Emanuel Barth, Stephan H. Bernhart, Victor Carmelo, Maximilian Collatz, Gero Doose, Florian Eggenhofer, Jan Ewald, Jörg Fallmann, Lasse M. Feldhahn, Markus Fricke, Juliane Gebauer, Andreas J. Gruber, Franziska Hufsky, Henrike Indrischek, Sabina Kanton, Jörg Linde, Nelly Mostajo, Roman Ochsenreiter, Konstantin Riege, Lorena Rivarola-Duarte, Abdullah H. Sahyoun, Sita J. Saunders, Stefan E. Seemann, Andrea Tanzer, Bertram Vogel, Stefanie Wehner, Michael T. Wolfinger, Rolf Backofen, Jan Gorodkin, Ivo Grosse, Ivo L. Hofacker, Steve Hoffmann, Christoph Kaleta, Peter F. Stadler, Stephan Becker, Manja Marz
Sci. Rep. 6:34589 (2016) | doi:10.1038/srep34589 | PDF

Predicting RNA structures from sequence and probing data

2016-07-01T00:00:00+02:00

RNA secondary structure prediction is sometimes presented as if there were a clean historical break between "old biophysics" and "new AI". The field did not develop that way. Long before the current machine-learning wave, RNA bioinformatics had already built a sophisticated toolkit around thermodynamic folding, ensemble analysis, comparative evidence, and experimental structure probing. This review comes from that earlier period and explains the core logic of the field without mistaking benchmark performance for mechanistic understanding.

The review begins from the classical thermodynamic view of RNA folding. Dynamic programming algorithms can efficiently compute minimum free energy structures, base-pairing probabilities, and partition-function ensembles under a well-defined energy model. These methods remain powerful because they do more than output a single structure. They give an explicit physical interpretation of structural alternatives, uncertainty, and energetic tradeoffs. RNA structure prediction therefore cannot be reduced to "find the one correct fold". For many RNAs, the ensemble itself is the relevant biological object.

At the same time, purely sequence-based thermodynamic prediction has obvious limits. Energy parameters are imperfect, tertiary interactions are usually treated only indirectly, and the energetically optimal structure is not always the biologically realized one. This becomes especially clear for regulatory RNAs, long transcripts, and systems shaped by kinetics, ligand binding, proteins, or cellular context. The review lays out these limitations clearly.

The article provides an overview of how chemical and enzymatic structure probing can be integrated with folding algorithms. Methods such as SHAPE, PARS, and related probing strategies provide nucleotide-resolution information about local flexibility. The computational question is how to convert those experimental readouts into something a folding algorithm can use. The review discusses approaches based on pseudo-energies and soft constraints, where probing reactivities perturb the thermodynamic model rather than replacing it outright. That choice keeps the prediction grounded in base-pairing energetics while still allowing experimental evidence to influence the result.

This combination of experiment and computation was, and remains, one of the most productive ideas in RNA structure prediction. Probing data can help discriminate among near-optimal folds, recover structures that sequence-only models miss, and improve the interpretation of structural ensembles. The review does not oversell the approach, though. Experimental data are noisy, condition-dependent, and often indirect. A reactivity profile is not itself a structure. It still has to be interpreted through a model, and the quality of the result depends on both the experiment and the computational framework used to incorporate it.

The article is optimistic about combining data with theory, but it does not pretend that more data automatically solve the inference problem. Prediction improves when models encode the right constraints and when external evidence is incorporated thoughtfully, not simply when another layer of complexity is added.

The same issue comes up whenever a computational result is used to justify an experimental move. At that point the question is what level of structural evidence is actually enough for a design choice, a mutational plan, or a mechanistic claim.

The review connects several levels of the field at once, from classical RNA folding algorithms and ensemble thinking to experimental probing and the practical business of combining them. It also remains a reminder that the most reliable structural insight often comes from combining complementary sources of information rather than choosing between "physics" and "data".

SHAPE directed RNA folding with the ViennaRNA Package is the more implementation-focused companion piece describing specific SHAPE integration strategies in ViennaRNA, and Caveats in deep learning for RNA secondary structure prediction picks up the same problem from the later AI period.

Citation

Predicting RNA Structures from Sequence and Probing Data
Ronny Lorenz, Michael T. Wolfinger, Andrea Tanzer, Ivo L. Hofacker
Methods 103:86–98 (2016) | doi:10.1016/j.ymeth.2016.04.004 | PDF

The MazF regulon and post-transcriptional stress adaptation in Escherichia coli

2016-07-01T00:00:00+02:00

Poly-seq workflow and MazF-dependent processing of ribosomes and mRNAs in Escherichia coli

This paper tackles a question that sits at the intersection of RNA biology, bacterial stress physiology, and translation control: what exactly does the toxin MazF do to the cell’s gene-expression program once it is activated under stress? MazF is often introduced as part of a toxin-antitoxin system, but that label alone does not explain the biology. The more interesting issue is whether MazF simply shuts translation down indiscriminately, or whether it creates a more selective post-transcriptional program that helps cells adapt to harsh conditions.

The central claim of this study is that MazF does not merely destroy RNA in a nonspecific way. Instead, it reprograms translation. MazF cleaves a defined subset of mRNAs and also processes the 16S rRNA within mature ribosomes, removing the anti-Shine-Dalgarno region and thereby generating what had previously been described as stress-ribosomes. Those remodeled ribosomes are no longer equivalent to canonical ribosomes: they become better suited to translate MazF-processed transcripts. That immediately turns the problem into a systems-level one. If both the messages and the decoding machinery are being changed, then the relevant output is not just the transcriptome, but the translatome.

That is where the methodological contribution of the paper comes in. To define this MazF-dependent regulatory layer, we used Poly-seq, combining polysome fractionation with RNA-seq. The point of the method is that it captures intact polysome-associated transcripts rather than only short protected fragments. This makes it possible to ask, in parallel, which RNAs are present, which have been processed, and which are actually associated with translating ribosomes after MazF induction. For this question, that is exactly the right scale of observation.

The results are interesting because they resist a simplistic interpretation. One might expect a dedicated stress regulon in the narrow sense, enriched mainly for classical stress-defense factors. Instead, the MazF-regulon spans a much broader range of cellular functions. The corresponding protein products are not limited to a textbook “general stress response” module. That is one of the reasons the paper is useful: it argues that MazF-dependent translational reprogramming is not just an emergency on-off switch, but a more distributed mechanism for rapidly altering what the cell is able to synthesize under severe stress.

This broader view also changes how one thinks about selective translation in bacteria. In many discussions of bacterial regulation, transcription still gets the starring role, with translational control treated as a local fine-tuning layer. MazF is a strong counterexample. Here, post-transcriptional regulation acts globally enough to reshape the functional output of the cell, and it does so on a timescale that makes sense for acute stress adaptation. The paper therefore highlights selective translation as a major regulatory mechanism in its own right rather than a secondary detail downstream of transcription.

Another important aspect is the connection to persistence. Toxin-antitoxin systems have long been discussed in relation to persister-cell formation, but the mechanistic links often remain vague. This work does not reduce that problem to a single answer, yet it makes a concrete contribution: if MazF introduces heterogeneity in which transcripts remain translation-competent and which ribosomes are available to decode them, then it provides a plausible route by which a stressed bacterial population can diversify its phenotypic state. That makes MazF a candidate effector of harsh-stress adaptation rather than a passive marker of it.

From my perspective, one of the strengths of this paper is that it combines a fairly elegant conceptual model with a method that can actually test it. The figure from the paper captures the logic well: MazF induction, separation of total and polysome-associated RNA, sequencing-based identification of processed transcripts, and direct evidence for MazF-dependent remodeling of the ribosome itself. That is what turns the “MazF regulon” from a loose idea into something experimentally definable.

So although the title uses the word “toolbox”, the main message is not simply that MazF controls a list of genes. It is that bacteria can implement a fast post-transcriptional stress response by simultaneously editing the message pool and the translation machinery. For anyone interested in bacterial RNA biology, translational control, or stress-induced phenotypic diversification, that remains a useful framework.

Abstract

Flexible adaptation to environmental stress is vital for bacteria. An energy-efficient post-transcriptional stress response mechanism in Escherichia coli is governed by the toxin MazF. After stress-induced activation the endoribonuclease MazF processes a distinct subset of transcripts as well as the 16S ribosomal RNA in the context of mature ribosomes. As these ‘stress-ribosomes’ are specific for the MazF-processed mRNAs, the translational program is changed. To identify this ‘MazF-regulon’ we employed Poly-seq (polysome fractionation coupled with RNA-seq analysis) and analyzed alterations introduced into the transcriptome and translatome after mazF overexpression. Unexpectedly, our results reveal that the corresponding protein products are involved in all cellular processes and do not particularly contribute to the general stress response. Moreover, our findings suggest that translational reprogramming serves as a fast-track reaction to harsh stress and highlight the so far underestimated significance of selective translation as a global regulatory mechanism in gene expression.

Citation

The MazF-Regulon: A Toolbox for the Post-Transcriptional Stress Response in Escherichia Coli
Martina Sauert, Michael T. Wolfinger, Oliver Vesper, Christian Müller, Konstantin Byrgazov, Isabella Moll
Nucleic Acids Res. 44 (14): 6660–6675 (2016) | doi:10.1093/nar/gkw115 | PDF

RNA-seq profiling of Pseudomonas aeruginosa under anoxic cystic fibrosis-like growth

2016-01-28T00:00:00+01:00

Heatmap summarizing pathway-level transcriptional changes after short-term and long-term anoxic growth of Pseudomonas aeruginosa PA14

This paper moves away from RNA structure and into a very different biological setting: transcriptome adaptation of Pseudomonas aeruginosa in a cystic fibrosis-like environment. The problem is clinically important and mechanistically rich. In the lungs of cystic fibrosis patients, P. aeruginosa often encounters oxygen-limited mucus and establishes persistent biofilm-associated infections. That means the interesting question is not just how the bacterium behaves in standard aerobic culture, but how its gene-expression program changes when it transitions into sustained anoxic growth under conditions that better resemble the host environment.

That is the motivation for this study. We grew the clinical isolate PA14 in synthetic cystic fibrosis sputum medium and compared three states by RNA-seq: planktonic aerobic growth, an early response 30 minutes after the shift to anaerobiosis, and mature anoxic biofilm growth after 96 hours. This design matters because it separates acute oxygen limitation from the later, more stable transcriptional program associated with long-term adaptation. In other words, the paper is not only about “anoxia versus oxygen”, but about the trajectory from short-term response to chronic biofilm physiology.

Methodologically, this was a straightforward but strong transcriptomics experiment for its time. Directional RNA-seq libraries were prepared from biological replicates after rRNA depletion, sequenced on an Illumina platform, and mapped against the PA14 reference genome. The resulting differential-expression profiles were then interpreted at both the individual-gene and pathway level. What makes the paper useful is that it does not stop at a catalog of changed transcripts. It follows the transcriptome with mutant screening and targeted validation for selected genes, which helps connect the sequencing result back to biofilm phenotypes and antibiotic-related traits.

The main biological picture is clear. Short-term anoxia and long-term anoxic biofilm growth are not interchangeable states. Prolonged biofilm growth in the cystic-fibrosis-like medium is accompanied by broad pathway-level rewiring, including strong changes in denitrification and sulfur metabolism, shifts in central carbon metabolism, altered expression of virulence-associated functions, and changes in membrane or envelope-related pathways. The pathway heatmap from the paper captures this nicely: by the time the culture reaches the 96-hour biofilm state, the transcriptome has moved well beyond the immediate oxygen-starvation response.

One of the more practically relevant findings concerns antibiotic tolerance. The study reports decreased abundance of oprD transcripts and increased abundance of the mexCD-oprJ efflux operon in long-term anoxic biofilms. That combination is mechanistically interesting because it offers a plausible transcriptional explanation for altered susceptibility patterns under these chronic-growth conditions. The anoxic biofilm state does not just affect metabolism. It also reshapes how the bacterium may respond to antimicrobial pressure.

The study also identified candidate functions required for sustained anoxic biofilm growth through follow-up mutant analysis. That part is important because it turns the paper from a descriptive RNA-seq survey into a resource for bacterial physiology. Transcriptome profiling is most useful when it helps prioritize which genes or pathways deserve closer functional study, and this paper does that well in the context of chronic Pseudomonas adaptation.

From today’s perspective, the paper is an early example of a style of infection-related transcriptomics that has become standard: move closer to the host-relevant environment, distinguish transient from persistent states, and interpret gene-expression change in terms of physiology rather than as an undifferentiated list of up- and downregulated genes. For P. aeruginosa, that is especially important because oxygen limitation, nitrate respiration, biofilm formation, and drug tolerance are deeply entangled in chronic infection.

Rewiring of Gene Expression in Pseudomonas aeruginosa During Diauxic Growth Reveals an Indirect Regulation of the MexGHI-OpmD Efflux Pump by Hfq continues the same general theme from a different angle. Transcriptome-level changes in Pseudomonas are often easiest to understand when metabolism, physiology, and antibiotic response are considered together.

Abstract

The opportunistic human pathogen Pseudomonas aeruginosa can thrive under microaerophilic to anaerobic conditions in the lungs of cystic fibrosis patients. RNASeq based comparative RNA profiling of the clinical isolate PA14 cultured in synthetic cystic fibrosis medium was performed after planktonic growth (OD600 = 2.0; P), 30 min after shift to anaerobiosis (A-30) and after anaerobic biofilm growth for 96h (B-96) with the aim to reveal differentially regulated functions impacting on sustained anoxic biofilm formation as well as on tolerance towards different antibiotics. Most notably, functions involved in sulfur metabolism were found to be up-regulated in B-96 cells when compared to A-30 cells. Based on the transcriptome studies a set of transposon mutants were screened, which revealed novel functions involved in anoxic biofilm growth. In addition, these studies revealed a decreased and an increased abundance of the oprD and the mexCD-oprJ operon transcripts, respectively, in B-96 cells, which may explain their increased tolerance towards meropenem and to antibiotics that are expelled by the MexCD-OprD efflux pump.

Citation

RNA-Seq Based Transcriptional Profiling of Pseudomonas Aeruginosa Pa14 After Short- and Long-Term Anoxic Cultivation in Synthetic Cystic Fibrosis Sputum Medium
Muralidhar Tata, Michael T. Wolfinger, Fabian Amman, Nicole Roschanski, Andreas Dötsch, Elisabeth Sonnleitner, Susanne Häussler, Udo Bläsi
PLoS ONE 11 (1): e0147811 (2016) | doi:10.1371/journal.pone.0147811 | PDF

SHAPE directed RNA folding with the ViennaRNA Package

2015-09-02T00:00:00+02:00

Deigan method adjusts the energetics of stacked base pairs

This paper sits at a practical intersection. One has a sequence, probing data, and the need to combine both in a single folding model. Rather than introducing a wholly new formalism, the paper shows how SHAPE reactivities can be incorporated into the ViennaRNA Package through soft constraints, so experimental evidence can steer prediction without displacing the thermodynamic model underneath.

Sequence-based folding is often not enough, especially for larger RNAs, regulatory elements, or transcripts with several plausible alternatives. At the same time, probing experiments do not hand over a finished secondary structure. They report on local flexibility and structural context nucleotide by nucleotide. The central task is to turn those measurements into something a folding algorithm can use without pretending the data are exact. The paper stays in that narrow space between experiment and model.

Methodologically, the paper compares three influential SHAPE-integration strategies and implements them in a common ViennaRNA framework. The Deigan approach converts reactivities into pseudo-energies that act mainly on stacked pairs, which makes it direct and relatively light. The Zarringhalam method reads reactivities in terms of paired and unpaired propensities and spreads penalties more broadly across the structure. The Washietl approach takes a more global view and infers a perturbation of the energy model that reconciles thermodynamic folding with the probing signal while keeping the intervention small. Seeing these methods side by side in one software environment makes their assumptions much easier to compare.

That comparative aspect is one of the strongest features of the paper. It does not argue for one universal recipe. Instead, it shows the choices that matter. How strongly should experimental data push against the default energy model? Is it better to treat reactivities locally, or to interpret them at the ensemble level? What should happen when SHAPE data favor a structure the default parameterization would otherwise underweight? These are the real inference questions behind experiment-guided RNA folding.

The benchmark section matters for the same reason. The paper evaluates the methods on RNAs with known reference structures and looks not only at minimum free energy predictions but also at ensemble properties derived from partition function calculations. That broader view is important. SHAPE data do not simply improve one final fold. Very often they sharpen the whole distribution of plausible structures, which makes base-pair probabilities and structural uncertainty more informative. In practice, careful SHAPE-guided folding does better than sequence-only thermodynamic prediction, particularly when several folds are close in free energy.

The paper also makes clear that ViennaRNA is more than a set of folding executables. It is a framework in which the energy model can be extended in a controlled way. Published SHAPE datasets can be reanalyzed, methods can be compared under the same machinery, and users are not forced into one opaque implementation.

The paper is also a reminder that some of the best progress in RNA structure prediction comes from combining different kinds of evidence rather than replacing one paradigm with another. Even now, that lesson still holds. Experimental probing, ensemble thinking, and physically interpretable constraints remain central when the goal is biological understanding rather than a better benchmark number.

This SHAPE paper belongs squarely in that territory, because its value lies in changing which structural hypotheses remain plausible once experimental data are brought in.

For a more direct discussion of what SHAPE and related chemical probing results can and cannot justify in practice, see How to interpret SHAPE and chemical probing data for RNA structure decisions.

Predicting RNA structures from sequence and probing data places SHAPE integration into the larger classical RNA-structure field, and Caveats to deep learning approaches to RNA secondary structure prediction carries the discussion into the later AI period, where the same inference problem reappears in a different modeling language.

For a deeper methodological breakdown, the Supplementary Data remain worth reading. They contain the detailed parameter choices, benchmark setup, and implementation notes behind the SHAPE-directed folding routines.

Abstract

Chemical mapping experiments allow for nucleotide resolution assessment of RNA structure. We demonstrate that different strategies of integrating probing data with thermodynamics-based RNA secondary structure prediction algorithms can be implemented by means of soft constraints. This amounts to incorporating suitable pseudo-energies into the standard energy model for RNA secondary structures. As a showcase application for this new feature of the ViennaRNA Package we compare three distinct, previously published strategies to utilize SHAPE reactivities for structure prediction. The new tool is benchmarked on a set of RNAs with known reference structure.

The capability for SHAPE directed RNA folding is part of the upcoming release of the ViennaRNA Package 2.2, for which a preliminary release is already freely available at https://www.tbi.univie.ac.at/RNA.

Citation

SHAPE directed RNA folding
Ronny Lorenz, Dominik Luntzer, Ivo L. Hofacker, Peter F. Stadler, Michael T. Wolfinger
Bioinformatics 32: 145–47 (2016) | doi:10.1093/bioinformatics/btv523 | PDF

mRNA degradation occurs on ribosome complexes in Drosophila cells

2015-08-12T00:00:00+02:00

Workflow for sequencing decapped mRNA intermediates and summary of their abundance on ribosome complexes versus whole-cell lysate

This paper sits in a different part of my publication list than the RNA structure and folding work, but in hindsight it connects to it more than it may seem at first glance. Before I started to investigate viral xrRNAs and the ways in which structured RNAs can resist exonucleolytic decay, I was interested in a more general question: where does 5' to 3' mRNA degradation actually take place in the cell, and how tightly is it coupled to translation?

At the time, the coupling between translation and mRNA turnover was already widely appreciated, but the physical location of that coupling was still not entirely clear. One could imagine a handoff model in which translating ribosomes and decay machineries act in sequence, or a more direct model in which degrading mRNAs remain associated with ribosome-containing complexes during turnover. This study addressed that distinction experimentally in Drosophila cells and asked whether the same picture also applies to miRNA-mediated decay.

The approach combined several complementary readouts. Polysome profiling and ribosome affinity purification were used to isolate ribosome-associated complexes. These preparations were then probed for canonical translation factors, general deadenylation and decapping components, and key miRNA pathway proteins such as AGO1 and GW182. The crucial control was that these associations depended on intact RNA, arguing that the decay factors were not simply sticking nonspecifically to ribosomal proteins but were traveling with messenger ribonucleoprotein complexes engaged with mRNA.

The most informative part of the paper, in my view, is the sequencing-based analysis of decapped degradation intermediates. The workflow captured 5' monophosphorylated mRNA fragments from ribosome complexes and compared them with corresponding fragments from whole-cell lysate. That moved the study beyond a few marker transcripts or individual immunoblots. Instead, it asked the question transcriptome-wide: are decapped intermediates generally depleted from ribosome-associated fractions, or do they remain there at similar relative abundance?

The answer was strikingly broad. Roughly 93% of the detected decapped fragments were present at essentially the same relative abundance in ribosome complexes as in the total lysate. That is a strong argument against the idea that decapping is followed by rapid physical separation from translational assemblies for most transcripts. Rather, the data support a model in which bulk 5' to 3' decay commonly proceeds on ribosome-associated mRNPs, with miRNA-mediated decay fitting into the same overall framework rather than constituting a fully separate compartmentalized process.

That result matters biologically because it shifts the emphasis from static cellular compartments to dynamic functional coupling. It suggests that translation, deadenylation, decapping, and exonucleolytic degradation should often be understood as parts of the same continuum of messenger RNA handling. For miRNA biology this was especially relevant: AGO1 and GW182 were not just abstract regulatory labels but factors connected to ribosome-associated decay states of target RNAs.

Seen from later work on xrRNAs, this paper also sharpened the next question. If much of 5' to 3' decay is happening on ribosome-associated messenger complexes, then structured RNA elements that can impede exonucleases become even more interesting, because they do not act in an isolated degradation chamber. They act in the middle of an already crowded and highly coordinated post-transcriptional environment. In that sense, this study predates the xrRNA projects but helped define the decay-centered perspective from which those later questions became compelling.

So although this is not an RNA structure paper, it is an important mechanistic one. It combines classical biochemical fractionation with a transcriptome-wide sequencing readout to make a fairly clean point: in Drosophila cells, both general and miRNA-mediated mRNA degradation are closely tied to ribosome-associated complexes. For anyone interested in how gene expression is controlled after transcription, that is a useful result to keep in mind.

Abstract

The translation and degradation of mRNAs are two key steps in gene expression that are highly regulated and targeted by many factors, including microRNAs (miRNAs). While it is well established that translation and mRNA degradation are tightly coupled, it is still not entirely clear where in the cell mRNA degradation takes place. In this study, we investigated the possibility of mRNA degradation on the ribosome in Drosophila cells. Using polysome profiles and ribosome affinity purification, we could demonstrate the copurification of various deadenylation and decapping factors with ribosome complexes. Also, AGO1 and GW182, two key factors in the miRNA-mediated mRNA degradation pathway, were associated with ribosome complexes. Their copurification was dependent on intact mRNAs, suggesting the association of these factors with the mRNA rather than the ribosome itself. Furthermore, we isolated decapped mRNA degradation intermediates from ribosome complexes and performed high-throughput sequencing analysis. Interestingly, 93% of the decapped mRNA fragments (approximately 12,000) could be detected at the same relative abundance on ribosome complexes and in cell lysates. In summary, our findings strongly indicate the association of the majority of bulk mRNAs as well as mRNAs targeted by miRNAs with the ribosome during their degradation.

Citation

General and miRNA-mediated mRNA degradation occurs on ribosome complexes in Drosophila cells
Sanja Antic, Michael T. Wolfinger, Anna Skucha, Stefanie Hosiner and Silke Dorner
Mol. Cell. Biol. 35(13), 2309-2320 (2015) | doi:10.1128/MCB.01346-14 | PDF

Building efficient NGS analysis pipelines with ViennaNGS

2015-03-02T00:00:00+01:00

Representative ViennaNGS utilities and tutorial resource requirements from the paper

In the early years of NGS bioinformatics, many analyses were still held together by ad hoc lab scripts. That worked for one project at a time, but it did not scale well: code was hard to reuse, pipelines were difficult to document, and routine file-format handling often had to be reimplemented over and over again. ViennaNGS grew out of that environment. The goal was not to offer yet another one-click analysis suite, but to provide reusable building blocks for people who needed to assemble their own workflows.

That distinction matters. ViennaNGS is not a fixed pipeline in the narrow sense. It is a toolbox: a collection of Perl modules and utility scripts intended to support the development of custom next-generation sequencing workflows. In practical terms, that meant exposing common operations as reusable library code while also shipping command-line utilities that served as both reference implementations and ready-to-use helpers for everyday tasks.

The scope of the package reflects the pain points of early NGS data analysis. ViennaNGS included support for feature extraction and format conversion across common genomics file types, mapping statistics, expression quantification and normalization, splice-junction analysis, motif-data handling, and the automated construction of Assembly and Track Hubs for the UCSC genome browser. It also wrapped widely used command-line tools, which helped make larger pipelines less repetitive and easier to script consistently.

What I still like about this paper is that it treats software architecture as a scientific problem rather than an afterthought. The key issue was not only whether one could write a script for a given analysis step, but whether that step could be turned into a reusable, composable module that fit into a larger workflow. That is a very different mindset from publishing a single-purpose utility. It is closer to infrastructure design, and that is why the project remained useful after specific assays or preferred file formats evolved.

The paper also pays attention to usability in a way that was easy to overlook at the time. ViennaNGS was distributed through familiar community channels, including GitHub and CPAN, and it came with tutorials that demonstrated real pipeline construction rather than isolated API calls. That may sound mundane, but it was important. For many academic bioinformatics tools, the bottleneck is not raw capability. It is whether anyone besides the original author can actually deploy the software in a real analysis environment.

ViennaNGS belongs to an earlier generation of workflow engineering, before systems like Snakemake or Nextflow became default answers for many users. That does not make the underlying ideas obsolete. The need for modularity, explicit file-format handling, reusable wrappers, and browser-ready outputs has not gone away. If anything, the paper reads as a snapshot of a period when the field was still learning how much of NGS bioinformatics depended on software organization rather than on any one algorithm.

The project was especially useful in research settings where workflows had to remain flexible. Not every NGS problem fits a canned pipeline, and many biologically interesting questions still require custom combinations of standard processing steps. ViennaNGS tried to make that custom work less brittle by packaging recurring operations into a coherent toolbox. That is the reason I think it still deserves attention: it addressed the engineering layer of genomics analysis at a time when that layer was often neglected.

The ViennaNGS suite is available through GitHub and CPAN.

Abstract

Recent achievements in next-generation sequencing (NGS) technologies lead to a high demand for reuseable software components to easily compile customized analysis workflows for big genomics data. We present ViennaNGS, an integrated collection of Perl modules focused on building efficient pipelines for NGS data processing. It comes with functionality for extracting and converting features from common NGS file formats, computation and evaluation of read mapping statistics, as well as normalization of RNA abundance. Moreover, ViennaNGS provides software components for identification and characterization of splice junctions from RNA-seq data, parsing and condensing sequence motif data, automated construction of Assembly and Track Hubs for the UCSC genome browser, as well as wrapper routines for a set of commonly used NGS command line tools.

Citation

ViennaNGS: A toolbox for building efficient next-generation sequencing analysis pipelines
Michael T. Wolfinger, Jörg Fallmann, Florian Eggenhofer, Fabian Amman
F1000Research 4:50 (2015) | doi: 10.12688/f1000research.6157.2 | PDF

Memory-efficient exploration of RNA energy landscapes

2014-06-12T00:00:00+02:00

This paper is easiest to understand as the RNA-focused continuation of earlier landscape work. Barrier Trees of Degenerate Landscapes introduced a rigorous way to represent basins and saddle points with barrier trees. Exploring the Lower Part of Discrete Polymer Model Energy Landscapes addressed how to explore the low-energy portion of a large landscape efficiently enough to make such representations practical. By 2014, the natural next step was clear: how do we make that style of exact landscape analysis workable for RNA secondary structures without running out of memory?

That is the problem addressed here. RNA folding dynamics can be modeled as motion on a large discrete landscape of secondary structures, with transitions between neighboring states and a macrostate decomposition into basins around local minima. In principle, that landscape contains exactly the information one would want for kinetics. In practice, however, exact exploration becomes difficult very quickly as sequence length increases. The bottleneck is no longer the conceptual framework but the size of the state space and the memory needed to represent it.

The methodological contribution of this paper is a local flooding variant of the earlier global flooding strategy. Instead of trying to hold a much larger explored region in memory all at once, the algorithm constructs exact macrostate transition models in a more localized and memory-efficient way. That sounds like an implementation detail, but it is the sort of implementation detail that determines whether a mathematically elegant method can actually be used on realistic RNA examples.

What makes this paper important is that it preserves exactness at the macrostate level while reducing the computational cost enough to widen the range of tractable systems. The work also compares exact transition models with two barrier-based approximations, showing where coarse approximations can become misleading. In other words, this is not just an optimization paper. It is also a paper about when approximation is acceptable and when a more faithful landscape representation changes the conclusions.

In the broader energy-landscape story, this article closes a loop. The early papers established the language of barrier trees and flooding-style exploration in abstract discrete systems and lattice polymers. This 2014 paper brings those ideas back to RNA in a way that is directly useful for folding kinetics. It shows that the older landscape concepts were not just mathematically interesting. They could be re-engineered into practical infrastructure for RNA analysis a decade later.

That is why I still view this paper as an important bridge between theory and application. It does not propose a radically new conceptual framework. Instead, it makes an existing framework usable at a scale where it becomes relevant for real RNA questions. In computational biology, that kind of engineering step is often what determines whether a good idea remains a paper concept or becomes a durable method.

Abstract

Motivation: Energy landscapes provide a valuable means for studying the folding dynamics of short RNA molecules in detail by modeling all possible structures and their transitions. Higher abstraction levels based on a macro-state decomposition of the landscape enable the study of larger systems; however, they are still restricted by huge memory requirements of exact approaches.

Results: We present a highly parallelizable local enumeration scheme that enables the computation of exact macro-state transition models with highly reduced memory requirements. The approach is evaluated on RNA secondary structure landscapes using a gradient basin definition for macro-states. Furthermore, we demonstrate the need for exact transition models by comparing two barrier-based approaches, and perform a detailed investigation of gradient basins in RNA energy landscapes.

Availability and implementation: Source code is part of the C++ Energy Landscape Library.

Citation

Memory-efficient RNA energy landscape exploration
Martin Mann, Marcel Kucharík, Christoph Flamm, Michael T. Wolfinger
Bioinformatics 30(18):2584-2591 (2014) | doi: 10.1093/bioinformatics/btu337 | PDF

Landscape Context

This paper builds directly on the original landscape-analysis sequence:

Barrier Trees of Degenerate Landscapes
Christoph Flamm, Ivo L. Hofacker, Peter F. Stadler, Michael T. Wolfinger
Z. Phys. Chem. 216:155-173 (2002) | doi:10.1524/zpch.2002.216.2.155 | Preprint PDF

Efficient Computation of RNA Folding Dynamics
Michael T. Wolfinger, W. Andreas Svrcek-Seiler, Christoph Flamm, Ivo L. Hofacker, Peter F. Stadler
J. Phys. A: Math. Gen. 37(17):4731-4741 (2004) | doi:10.1088/0305-4470/37/17/005 | PDF

Exploring the Lower Part of Discrete Polymer Model Energy Landscapes
Michael T. Wolfinger, Sebastian Will, Ivo L. Hofacker, Rolf Backofen, Peter F. Stadler
Europhys. Lett. 74(4):726-732 (2006) | doi:10.1209/epl/i2005-10577-0 | Preprint PDF

How to compute normalized RNA-seq expression from multicov files

2014-04-15T00:00:00+02:00

Whenever it comes to analyzing RNA-seq experiments, there is a need for comparing expression data at a quantitative level. Consider a scenario where samples were taken from different conditions and subjected to Illumina sequencing. Whether those samples were multiplexed or sequenced on a single lane each, one generally gets a different number of raw reads from each sample, refelcting experimental and technical biases inherent in the RNA-seq protocols. Various measures for normalization of RNA-seq samples have been proposed, the most widely used being RPKM (reads per kilobase per million). While RPKM tries to account for different sequencing depth by normalizing by the number of reads sequenced in a specific sample, divided by 10^6. This very step causes a systematic bias, as has been shown here:

Measurement of mRNA abundance using RNA-seq data: RPKM measure is inconsistent among samples
Günter P. Wagner, Koryu Kin, Vincent J. Lynch
Theory Biosci. 131, 281-285 (2012) | doi:10.1007/s12064-012-0162-3

RNA-Seq gene expression estimation with read mapping uncertainty
Bo Li, Victor Ruotti, Ron M. Stewart, James A. Thomson, Colin N. Dewey
Bioinformatics 6(4), 493-500. (2009) | doi:10.1093/bioinformatics/btp692

The central point of these papers is to work out an alternative measure for RNA-seq expression abundance that resembles as closely as possible the relative molar concentration (rmc) of each RNA species present in a sample. It is easy to see that the average rmc across genes has to be a constant that only depends on the number of genes mapped in an RNA-seq experiment.

One example of measures that fulfills the invariant average criterion is Transcript per million (TPM), being defined as

where t_g is a proxy for the number of transcripts that can be explained by a certain number of mapped reads and T is the sum of all t_g over all genes. If one is interested in mRNA abundance, the average TPM - and thus the average rmc is inversely proportional to the number of features present in a reference annotation.

Practically, TPM values for individual genes can be computed from read count tables, ie. tables that give the number of reads overlapping a specific gene. Typical programs for obtaining read count tables are htseq-count or multiBamCov (see bedtools multicov).

I have recently implemented normalize_multicov.pl, a tool for computing normalized RNA-seq expression in terms of TPM from multicov files. It is part of the ViennaNGS Perl Modules for NGS analysis and very easy to use: Just provide it the output of a bedtols multicov run on your data as well as the read length used for sequencing your samples and get back a normalized multicov file of your samples in terms of TPM. That's all ...

Bacterial transcription start site annotation from dRNA-seq data

2014-04-13T00:00:00+02:00

Statistical classification and evaluation scheme used by TSSAR for dRNA-seq-based TSS annotation

Identifying bacterial transcription start sites from differential RNA-seq data used to be a painfully manual task. Researchers would inspect mapped reads in genome browsers, compare TEX-treated and untreated libraries by eye, and then decide which positions looked like genuine primary transcript starts. That process was slow, difficult to reproduce, and inevitably shaped by user-specific thresholds and biases.

TSSAR was designed to solve exactly that problem. The aim was not just to call more TSS automatically, but to provide a statistically principled way to do so. Differential RNA-seq enriches primary transcripts by treating one library with terminator exonuclease, so the key signal is an excess of read starts at a genomic position in the TEX-treated sample relative to the untreated control. TSSAR turns that intuition into a formal model instead of leaving it at the level of browser-based pattern recognition.

The methodological core is a local count model for read starts within transcriptionally active regions. Counts in individual libraries are modeled in a way that leads naturally to a Skellam-distributed difference between treated and untreated start counts. That provides a direct statistical basis for deciding whether an observed enrichment is likely to reflect a genuine primary transcript start rather than noise. The output is not just a list of coordinates, but an annotated classification into primary, internal, antisense, and orphan signals that can be used downstream.

That combination of statistics and annotation logic is what made the tool useful in practice. TSSAR does not simply replace manual curation with a fixed arbitrary cutoff. It automates the decision process while still respecting the structure of dRNA-seq experiments. In the paper, the method is benchmarked against manual annotations and simple cutoff-based alternatives, and it performs substantially better at recovering both curated Helicobacter pylori TSS annotations and experimentally validated sites.

One reason the paper still feels current is its software architecture. TSSAR was not designed as two separate products, one local and one web-based. Instead, the workflow was intentionally split across a local client and a RESTful web service that depend on each other. The local component handles the preprocessing of mapped NGS data, while the processed data are then submitted to the service for the actual TSS-oriented statistical analysis and annotation steps. In other words, the client and service form a coupled pipeline rather than two interchangeable access modes.

That distinction matters, because it reflects how the method was meant to be used in practice. The computationally and format-specific preprocessing stayed close to the user's data and local environment, while the web service centralized the analysis logic and reporting layer. This design reduced the burden of reproducing a fairly specialized workflow in many separate installations, but it also means the local and remote components should not be described as independently useful alternatives. The architecture is integrated by design.

That architectural decision may be just as important as the underlying statistics. A method only has impact if people can actually use it, and TSSAR addressed that by combining client-side preprocessing with a service-backed analysis workflow. That is one reason it remained relevant well beyond the immediate paper.

The broader biological significance is also straightforward. Better TSS annotation improves our view of bacterial transcriptome architecture: operon boundaries, antisense transcription, condition-specific initiation, and the regulatory logic of promoter usage. In that sense, TSSAR is not just a niche utility for one sequencing protocol. It is infrastructure for studying how bacterial gene regulation is organized at the transcript level.

This paper is clearly a different topic from RNA folding or landscape analysis, but the underlying engineering mindset is similar. The interesting part is not only the biology. It is also how to turn a noisy, high-dimensional data source into a usable and reproducible computational workflow. That is exactly the kind of problem where good statistical assumptions, careful software design, and sensible interfaces make the difference between a promising idea and a method that people keep using.

Abstract

Background: Differential RNA sequencing dRNA-seq is a high-throughput screening technique designed to examine the architecture of bacterial operons in general and the precise position of transcription start sites (TSS) in particular. Hitherto, dRNA-seq data were analyzed by visualizing the sequencing reads mapped to the reference genome and manually annotating reliable positions. This is very labor intensive and, due to the subjectivity, biased.

Results: Here, we present TSSAR, a tool for automated de-novo TSS annotation from dRNA-seq data that respects the statistics of dRNA-seq libraries. TSSAR uses the premise that the number of sequencing reads starting at a certain genomic position within a transcriptional active region follows a Poisson distribution with a parameter that depends on the local strength of expression. The differences of two dRNA-seq library counts thus follow a Skellam distribution. This provides a statistical basis to identify significantly enriched primary transcripts.

We assessed the performance by analyzing a publicly available dRNA-seq data set using TSSAR and two simple approaches that utilize user-defined score cutoffs. We evaluated the power of reproducing the manual TSS annotation. Furthermore, the same data set was used to reproduce 74 experimentally validated TSS in H. pylori from reliable techniques such as RACE or primer extension. Both analyses showed that TSSAR outperforms the static cutoff-dependent approaches.

Conclusions: Having an automated and efficient tool for analyzing dRNA-seq data facilitates the use of the dRNA-seq technique and promotes its application to more sophisticated analysis. For instance, monitoring the plasticity and dynamics of the transcriptomal architecture triggered by different stimuli and growth conditions becomes possible.

The main asset of a novel tool for dRNA-seq analysis that reaches out to a broad user community is usability. As such, we provide TSSAR both as intuitive RESTful Web service http://rna.tbi.univie.ac.at/TSSAR together with a set of post-processing and analysis tools, as well as a stand-alone version for use in high-throughput dRNA-seq data analysis pipelines.

Citation

TSSAR: TSS annotation regime for dRNA-seq data
Fabian Amman, Michael T. Wolfinger, Ronny Lorenz, Ivo L. Hofacker, Peter F. Stadler, Sven Findeiß
BMC Bioinformatics 15:89 (2014) | doi: 10.1186/1471-2105-15-89 | PDF

BarMap and RNA folding on dynamic energy landscapes

2010-07-01T00:00:00+02:00

RNA folding rarely happens on a fixed energy landscape. During transcription, degradation, ligand binding, or temperature shifts, the available structure space changes over time, and with it the kinetic pathways. This paper addresses exactly that nonstationary setting by extending landscape-based RNA kinetics from a single static landscape to a sequence of related landscapes connected through time.

The central idea of BarMap is to treat each time point or perturbation step as a snapshot landscape with its own macrostates, and then define a mapping between neighboring snapshots. Population densities can then be transferred from one coarse-grained landscape to the next instead of starting the kinetics calculation from scratch each time. In effect, the expensive landscape analysis becomes a preprocessing step, while the temporal evolution is handled by moving populations across linked barrier-tree representations.

That is a particularly natural fit for co-transcriptional folding, where the RNA chain grows one nucleotide at a time and each elongation step slightly reshapes the landscape. But the same formalism also applies to temperature changes, refolding after cleavage, and mechanically constrained scenarios. The paper therefore broadens the energy-landscape view of RNA folding kinetics into a framework for dynamic landscapes, which is much closer to the situations encountered by regulatory RNAs in cells.

From a methodological perspective, this is a useful complement to trajectory-based simulators. It does not aim to reconstruct every folding path explicitly. Instead, it keeps the coarse-grained view at the level of basins, barriers, and saddle points, while allowing those basins themselves to change over time. That makes it possible to study larger and more complex nonstationary folding scenarios than a direct simulation strategy would usually permit.

BarMap is therefore one of the key papers in the co-transcriptional RNA folding, RNA folding kinetics, and energy landscapes cluster. It formalizes the idea that kinetics should often be thought of as motion across a sequence of changing landscapes, not just as diffusion on a single fixed one.

Abstract

Dynamical changes of RNA secondary structures play an important role in the function of many regulatory RNAs. Such kinetic effects, especially in time-variable and externally triggered systems, are usually investigated by means of extensive and expensive simulations of large sets of individual folding trajectories. Here we describe the theoretical foundations of a generic approach that not only allows the direct computation of approximate population densities but also reduces the efforts required to analyze the folding energy landscapes to a one-time preprocessing step. The basic idea is to consider the kinetics on individual landscapes and to model external triggers and environmental changes as small but discrete changes in the landscapes. A ‘‘barmap’’ links macrostates of temporally adjacent landscapes and defines the transfer of population densities from one ‘‘snapshot’’ to the next. Implemented in the BarMap software, this approach makes it feasible to study folding processes at the level of basins, saddle points, and barriers for many nonstationary scenarios, including temperature changes, cotranscriptional folding, refolding in consequence to degradation, and mechanically constrained kinetics, as in the case of the translocation of a polymer through a pore.

Citation

BarMap: RNA Folding on Dynamic Energy Landscapes
Ivo L. Hofacker, Christoph Flamm, Christian Heine, Michael T. Wolfinger, Gerik Scheuermann, Peter F. Stadler
RNA 16:1308–1316 (2010) | doi:10.1261/rna.2093310 | PDF

Folding kinetics of large RNAs

2008-06-01T00:00:00+02:00

Classical RNA folding kinetics methods become difficult to scale once sequences move beyond the size range of small regulatory RNAs. This paper addresses that limitation with Kinwalker, a heuristic that predicts folding trajectories by combining locally optimal substructures rather than trying to explore the full kinetic landscape explicitly.

The algorithm starts from a simple but powerful observation: many important metastable intermediates appear to be composed of locally favorable structural building blocks. Kinwalker therefore precomputes thermodynamically optimal local structures by dynamic programming, then assembles them stepwise into a global folding trajectory while resolving conflicts between overlapping alternatives. Folding events are interleaved with transcription events, which makes the method particularly suitable for co-transcriptional folding problems.

That design gives the method a very different operating regime from exhaustive simulation. Instead of aiming for exact kinetics on all possible paths, it attempts to recover the dominant trajectory structure for much larger RNAs than earlier methods could handle. The paper reports applicability to sequences of roughly 1500 nucleotides, which was a substantial extension at the time and made many biologically relevant systems computationally accessible.

The validation examples are important here. Kinwalker reproduces several experimentally studied folding scenarios, including delayed cloverleaf formation in bacteriophage RNAs, the ASR riboswitch, and the Hok system. These are exactly the kinds of RNAs where kinetic and co-transcriptional effects matter more than the equilibrium minimum free energy structure alone. In that sense, the paper is not just a scaling story. It is also a strong argument that useful kinetic prediction can come from carefully chosen heuristics when full landscape exploration is infeasible.

This paper is a useful reference point for anyone interested in how to bridge thermodynamic RNA structure prediction with folding-pathway analysis for longer molecules.

Abstract

We introduce here a heuristic approach to kinetic RNA folding that constructs secondary structures by stepwise combination of building blocks. These blocks correspond to sub-sequences and their thermodynamically optimal structures. These are determined by the standard dynamic programming approach to RNA folding. Folding trajectories are modeled at base pair resolution using the Morgan-Higgs heuristic and a barrier tree based heuristic to connect combinations of the local building blocks. Implemented in the program Kinwalker, the algorithm allows co-transcriptional folding and can be used to fold sequences of up to about 1500 nucleotides in length. A detailed comparison with several well-studied examples from the literature, including the delayed folding of bacteriophage cloverleaf structures, the ASR riboswitch, and the Hok RNA, shows an excellent agreement of predicted trajectories and experimental evidence. The software is available as part of the Vienna RNA Package.

Citation

Folding Kinetics of Large RNAs
Michael Geis, Christoph Flamm, Michael T. Wolfinger, Andrea Tanzer, Ivo L. Hofacker, Martin Middendorf, Christian Mandl, Peter F. Stadler, Caroline Thurner
J. Mol. Biol. 379(1):160–173 (2008) | doi:10.1016/j.jmb.2008.02.064 | Preprint PDF

Visualization of lattice-based protein folding simulations

2006-06-01T00:00:00+02:00

This paper sits slightly aside from the more algorithmic energy-landscape work, but it addresses a very practical problem that appears as soon as folding simulations become nontrivial: once you can generate trajectories, transition graphs, or large collections of low-energy conformations, how do you actually inspect them in a way that helps scientific reasoning? For lattice-based protein folding models, the underlying state space is discrete and mathematically convenient, yet still hard to understand from tables of coordinates or lists of moves alone.

The contribution of this conference paper is therefore less about proposing a new folding algorithm and more about making simulation output intelligible. The work focuses on visualization strategies for lattice-based protein folding simulations, with the goal of showing both structural change and dynamical progression. In this setting, the central challenge is that one wants to see several things at once: the conformation of the polymer, the evolution of a trajectory through state space, and the energetic relationships between alternative folds.

That is why this paper fits well with the surrounding early landscape papers. Once folding is understood as motion on an energy landscape, visualization becomes more than presentation. It becomes an analytical tool. Good visual representations can help identify metastable states, transitions between basins, and differences between trajectories that would otherwise remain hidden in raw simulation output. In that sense, this work complements the barrier-tree and kinetics papers by asking how researchers can explore and communicate the structures and transitions those methods produce.

The methodological approach combines lattice-based folding simulations with interactive and comparative visual encodings of conformational change. Rather than reducing the problem to a single static snapshot, the paper considers how to represent discrete polymer conformations over time and how to connect structural states to their dynamic relationships. For a field that often depends on abstract state spaces, this matters. Visualization is one of the few ways to make model behavior legible enough to support debugging, hypothesis generation, and explanation.

The paper also reminds us that bioinformatics is not only about prediction accuracy or algorithmic complexity. There is a downstream interpretability problem as well. Whether one studies lattice proteins, RNA folding landscapes, or modern molecular simulations, the question remains similar: how can complex structural dynamics be turned into a form that scientists can inspect, compare, and reason about effectively? This early work tackles that question directly.

The most natural context is the sequence of papers on energy landscapes. Those articles develop the computational machinery for coarse-graining, exploring, and simulating folding landscapes. This one addresses the complementary issue of how to visualize such processes once the computations are done. It is best read as part of that early methodological cluster rather than as an isolated visualization exercise.

Citation

Visualization of Lattice-Based Protein Folding Simulations
Sebastian Pötzsch, Gerik Scheuermann, Peter F. Stadler, Michael T. Wolfinger, Christoph Flamm
In IV '06 Proceedings of the Conference on Information Visualization, pp89-94. Washington, DC, USA: IEEE Computer Society (2006) | doi:10.1109/IV.2006.127

Exploring the Lower Part of Discrete Polymer Model Energy Landscapes

2006-04-14T00:00:00+02:00

This paper continues the early energy-landscape line of work, but shifts the emphasis from formal representation to practical exploration. Once one accepts that barrier trees are a useful way to summarize a landscape, the next problem is obvious: how do we obtain the relevant low-energy portion of the landscape in the first place, especially when exhaustive enumeration is too expensive?

That is the question addressed here. The paper proposes a generic algorithm for "flooding" the lower part of a discrete energy landscape starting from optimal and near-optimal conformations. Instead of trying to enumerate the full state space, the method selectively expands the part of the landscape that is actually relevant for basin structure, barrier heights, and subsequent dynamics analysis. The examples in the paper use lattice proteins in two and three dimensions, but the underlying idea is more general.

The methodological point is important. Many landscape-analysis methods become impractical not because the mathematics is unclear, but because the raw search space grows too quickly. This paper tackles that bottleneck directly. Starting from structures produced by constraint-based search, it incrementally explores states below a chosen energy threshold and uses that information to reconstruct the low-energy topology of the landscape. In effect, it provides a computational route from a discrete optimization problem to a barrier-tree style description without requiring full enumeration of all states.

What makes this useful is that the low-energy region is often exactly where the interesting folding behavior lives. Local minima, kinetically relevant basins, and the barriers separating them are typically concentrated there. If one can map that part of the landscape reliably, one already has much of what is needed for coarse-grained dynamics, metastability analysis, and structure-space visualization. The paper therefore acts as a bridge between abstract landscape theory and practically usable exploration algorithms.

This is also why the work became relevant beyond lattice proteins. The algorithm is presented in a problem-independent way and helped shape later approaches for RNA landscape exploration, where the same tension appears again: the full structure space is enormous, but the subset relevant for folding kinetics is much smaller and more structured. In that sense, the paper is part of the conceptual path from discrete polymer models to later RNA-focused landscape methods.

For readers looking back from today's perspective, the significance of this work is not that it solves all exploration problems once and for all. It does something more realistic and more durable: it shows how to target the informative part of a large landscape efficiently enough that subsequent topology and dynamics analysis become possible. That remains one of the central engineering challenges in landscape analysis, regardless of whether the underlying system is a lattice polymer, an RNA secondary structure ensemble, or some other discrete configuration space.

Abstract

We present a generic, problem-independent algorithm for exploration of the low-energy portion of the energy landscape of discrete systems and apply it to the energy landscape of lattice proteins. Starting from a set of optimal and near-optimal conformations derived from a constraint-based search technique, we are able to selectively investigate the lower part of lattice protein energy landscapes in two and three dimensions. This novel approach allows, in contrast to exhaustive enumeration, for an efficient calculation of optimal and near-optimal structures below a given energy threshold and is only limited by the available amount of memory. A straightforward application of the algorithm is the calculation of barrier trees (representing the energy landscape), which then allows dynamics studies based on landscape theory.

Citation

Exploring the Lower Part of Discrete Polymer Model Energy Landscapes
Michael T. Wolfinger, Sebastian Will, Ivo L. Hofacker, Rolf Backofen, Peter F. Stadler
Europhys. Lett. 74(4): 726–32 (2006) | doi:10.1209/epl/i2005-10577-0 | Preprint PDF

Efficient computation of RNA folding dynamics

2004-04-14T00:00:00+02:00

RNA folding dynamics on a coarse-grained energy landscape

This paper addresses a classical problem in RNA folding kinetics: exact or near-exact stochastic simulation quickly becomes expensive, but many biologically interesting questions depend on the dynamic behavior of secondary structure landscapes rather than on a single minimum-energy fold. The solution proposed here is to coarse-grain the secondary structure landscape using barrier trees built from local minima and their connecting saddle points.

Within that framework, RNA folding dynamics can be approximated by a master-equation approach on macrostates rather than on the full structure space. Instead of simulating huge numbers of trajectories, one numerically integrates the population flow between barrier-separated basins of attraction. This makes it possible to recover the folding behavior of bistable RNAs and metastable switches with much higher computational efficiency than direct trajectory-based simulations.

The important point is not just speed. Barrier trees provide a natural way to connect energy landscapes with kinetics, because they preserve the local minima, the barriers between them, and therefore the transitions most relevant for folding pathways. In the paper, the resulting dynamics agree reasonably well with stochastic folding simulations while extending the accessible timescales and molecule sizes. At the time, this was a significant step toward making RNA folding kinetics computationally tractable beyond toy examples.

This work is also foundational for later developments in RNA energy-landscape analysis. Many subsequent approaches to metastability, coarse-grained kinetics, and dynamic landscape mapping build on the same idea that one can separate the enormous RNA structure space into kinetically meaningful basins and work at that level instead of trying to resolve every elementary step directly.

For today’s readers, the paper remains relevant because it frames RNA folding kinetics as an energy-landscape problem rather than just a simulation problem. That perspective is still central when thinking about RNA switches, delayed folding, metastable intermediates, and the design of molecules whose function depends on how they move through structure space rather than only on where they end up.

Citation

Efficient Computation of RNA Folding Dynamics
Michael T. Wolfinger, W. Andreas Svrcek-Seiler, Christoph Flamm, Ivo L. Hofacker, Peter F. Stadler
J. Phys. A: Math. Gen. 37(17):4731–4741 (2004) | doi:10.1088/0305-4470/37/17/005 | PDF

Barrier Trees of Degenerate Landscapes

2002-07-01T00:00:00+02:00

This is one of the earliest papers in the landscape-analysis thread that later became central to my work on RNA folding kinetics. It is also one of the most mathematical. But the reason it became widely cited is actually quite practical: it gave people a compact and rigorous way to describe the large-scale topology of complex discrete landscapes.

The core problem is easy to state. If a system has many local minima, one needs some way to summarize which minima belong to which basins, where the relevant saddle points lie, and how difficult it is to move from one basin to another. In non-degenerate landscapes, that decomposition is conceptually straightforward. In degenerate landscapes, however, many states can share the same energy, and the clean separation into minima, saddles, and basins becomes much less obvious. That is exactly the setting this paper addresses.

The contribution of the paper is to extend the concept of a barrier tree to such degenerate cases and to implement that formalism in the barriers program. The input is deliberately general: one needs only a finite set of states, a neighborhood relation, and an energy or fitness value assigned to each state. From that information, the method constructs a tree in which leaves correspond to minima or minimal basins and internal nodes represent the saddle-like connections that merge them at increasing energy thresholds.

Why is that useful? Because the tree compresses a huge and otherwise unmanageable state space into something interpretable. Instead of staring at an enormous graph of configurations, one obtains a hierarchical representation of the landscape: which minima are close, which barriers are high, and which large-scale basins dominate the topology. That immediately makes the landscape more accessible for kinetics, visualization, and coarse graining.

This is also the point where the paper moves beyond pure mathematical formalism. Once barriers are represented explicitly, they can be used to estimate transition frequencies and to define macrostates for dynamical models. That connection turned out to be extremely important. Much of the later work on RNA energy landscapes, metastability, folding kinetics, and dynamic landscape mapping depends on exactly this move from raw microstates to a barrier-structured macrostate view.

Another reason the paper has aged well is that it is not specific to RNA. The formalism is deliberately problem-independent and applies to discrete landscapes more broadly, including spin systems, lattice models, and other optimization settings. That generality made barrier trees useful outside their original context and helped establish them as a standard conceptual tool in landscape analysis.

For readers coming from RNA biology, the significance of this paper is easiest to appreciate in hindsight. Many later methods in this blog, including coarse-grained RNA folding kinetics and the exploration of dynamic landscapes, rely on the assumption that one can meaningfully partition structure space into basins connected by barriers. This 2002 paper is one of the places where that assumption was made rigorous for degenerate discrete systems.

So while the article is mathematically inclined, the underlying message is simple and still relevant: if you want to understand a complex landscape, you need more than a list of low-energy states. You need a representation of how those states are connected. Barrier trees provided exactly that representation, and the barriers tool made it computationally usable.

Citation

Barrier Trees of Degenerate Landscapes
Christoph Flamm, Ivo L. Hofacker, Peter F. Stadler, Michael T. Wolfinger
Z. Phys. Chem. 216: 155–73 (2002) | doi:10.1524/zpch.2002.216.2.155 | Preprint PDF

Michael T. Wolfinger

Rational design of mechanically active RNAs

Citation

How to Run Claude Code with a Local LLM on Apple Silicon

Prerequisites

Install Claude Code

Part 1: Prepare LM Studio

Start LM Studio

Load a Model with a 32K Context Window

Start the Local Server

Part 2: Point Claude Code at LM Studio

Configure Environment Variables

Choose the Model

Part 3: Test the Local Endpoint

Part 4: Run Claude Code in a Project

Troubleshooting

Claude Code cannot connect

Responses are slow

Need server or prompt diagnostics

Notes

Summary

A first look at rational design of mechanically active RNAs

Citation

Functional RNAs in Virology Special Issue

Understanding the Role of RNA in Viruses

Scope of the Special Issue

A Timely Focus

Join the Special Issue

Details & Submission

Join the Conversation

Exploring RNA Biology with Deep Learning Algorithms

Why This Collection Matters

What We’re Looking For

Advisory Panel

Join the Conversation

Conserved RNA regulatory switches in living cells

Citation

From structure to function in Musashi-RNA complexes

Citation

See Also

Bayesian approximation of RNA folding times

Citation

KinPFN for RNA folding kinetics

Citation

3' UTR-biased siRNA production in an insect-specific flavivirus

Citation

Xinyang flavivirus and a likely tick-only orthoflavivirus clade

Citation

See Also

Automated lineage designation from viral genomic data

Citation

What virus bioinformatics can and cannot tell us about RNA viruses

Comparative Genomics First

Where RNA Structure Prediction Helps

Genomic Epidemiology Is About Context

What Good Virus Bioinformatics Looks Like

Why This Matters

A conserved G-quadruplex in the Zika virus 3' terminal region

Citation

RNA-protein complex refinement using AI modeling and docking

Citation

See Also

RNA–RNA interaction analysis of Japanese encephalitis virus

Figures and Data

Citation

Strukturierte RNAs in Viren (in German)

Figures and Data

Citation

The link between mRNA vaccine design and barbeque optimization

RNA recognition by Musashi-1

Figures and Data

Citation

See Also

Hfq, Crc, and antibiotic resistance in P. aeruginosa

Citation

Biophysical characterization of human lincRNA-p21 Alu inverted repeats

Citation

RNA structure conservation and molecular epidemiology of TBEV

Figures and Data

Citation