The PhageExpressionAtlas reveals shared and unique transcriptional patterns across phage-host interactions

The PhageExpressionAtlas is a pioneering bioinformatics resource that standardizes and democratizes access to time-resolved dual RNA-sequencing data from phage infections, enabling the discovery of conserved and unique transcriptional patterns in phage-host interactions and anti-phage defense mechanisms.

The Gist

Imagine a bustling city (the bacteria) that gets invaded by a specialized virus (the phage). When the virus attacks, it doesn't just smash things; it hijacks the city's power grid, shuts down the local government, and forces the city's factories to start building more viruses instead of their usual products.

For a long time, scientists have been able to take "snapshots" of this invasion. They can see which genes (the city's instruction manuals) are turned on or off at different moments. But here's the problem: these snapshots were scattered across thousands of different research papers, stored in different formats, and often locked away in complex data files that only experts could read. It was like having a library where every book was written in a different language and hidden in a different building.

Enter the PhageExpressionAtlas.

Think of the PhageExpressionAtlas as a massive, beautifully organized interactive museum dedicated to these viral invasions. The researchers (Maik Wolfram-Schauerte and colleagues) didn't just build a library; they built a time machine and a translator.

Here is how they did it, using simple analogies:

1. The Great Cleanup Crew (Standardization)

Before, every scientist processed their data differently. One might have measured the "loudness" of a gene in decibels, another in volume, and another in pitch. You couldn't compare them.

• The Solution: The team built a giant, automated factory assembly line (a computer pipeline). They took 42 different "snapshots" (datasets) from 23 different studies and ran them all through this factory.
• The Result: Every dataset was cleaned, measured, and labeled using the exact same ruler. Now, a gene from a virus attacking E. coli can be directly compared to a gene from a virus attacking Staphylococcus. It's like translating every book in the library into the same language so everyone can read them together.

2. The Interactive Map (The Website)

The team built a website where anyone can explore these invasions.

• The Heatmap: Imagine a giant wall of colored lights. Each light represents a gene. When the virus first arrives, the "early" lights flash red. As the virus takes over, the "middle" lights turn blue, and finally, the "late" lights (which build the new virus bodies) glow green. You can watch the whole movie of the infection unfold in real-time.
• The Genome Viewer: This is like a 3D globe of the virus's DNA. As you spin it, you can see which parts of the virus's instruction manual are being read at which time. It helps scientists see if the virus has organized its genes in a smart, logical order (like chapters in a book) or a chaotic mess.

3. The Detective Work (What They Found)

By looking at all these data points together, the scientists discovered some fascinating patterns:

• The "Unfinished Business" Mystery: A huge chunk of the virus's instruction manual consists of "unknown genes"—genes that scientists don't know what they do yet. The Atlas revealed that these mystery genes are active throughout the entire infection, from the very first second to the very last. They aren't just background noise; they are working hard the whole time.
• The Defense Game: Bacteria have their own security systems (like CRISPR, which is like a biological immune system). The Atlas showed that when a virus attacks, the bacteria often try to scream for help (turning up defense genes), but the virus usually has a counter-move.
  • The Twist: Sometimes the bacteria's defense genes actually get quieter during the attack. This suggests that the bacteria might be relying on the security guards they already have on duty, rather than calling in new reinforcements. The virus, meanwhile, has its own "anti-defense" tools that it deploys at specific times to shut down the bacteria's alarms.
• The Timing is Everything: The study showed that the exact timing of when a virus turns on its genes depends heavily on who it is attacking. A virus might act like a sprinter against one type of bacteria and a marathon runner against another. This proves that there is no "one size fits all" strategy in the viral world.

Why Does This Matter?

Think of the PhageExpressionAtlas as the Google Maps for viral infections.

• Before: If you wanted to study a specific virus, you had to drive to 20 different libraries, ask for permission to look at 50 different books, and try to figure out how they fit together.
• Now: You can sit at your computer, type in the name of a virus, and instantly see a map of its entire life cycle, compare it to other viruses, and spot patterns that no single scientist could ever see alone.

This tool democratizes science. It means a student in a small lab or a researcher in a different country can now ask big questions about how viruses work and how we might use them to fight antibiotic-resistant bacteria (phage therapy) without needing a supercomputer or a PhD in coding.

In short, the PhageExpressionAtlas turns a chaotic pile of scattered clues into a clear, connected story of the eternal battle between bacteria and viruses.

Technical Summary

1. Problem Statement

Time-resolved dual RNA-sequencing (dual RNA-seq) has become a standard method for studying bacteriophage-host interactions, providing simultaneous quantification of phage and host transcriptomes. However, the field faces significant barriers to data reuse and integrative analysis:

• Lack of Standardization: Existing datasets are rarely processed uniformly, making cross-study comparisons difficult due to varying pipelines and normalization methods.
• Accessibility Issues: Raw data is often available, but processed, gene-level count tables are not. Furthermore, there is a lack of dedicated databases that support rapid re-analysis and interactive visualization for non-bioinformaticians.
• Fragmented Knowledge: While individual studies yield insights into specific phage-host systems, the potential to uncover conserved transcriptional patterns, defense mechanisms, and counter-strategies across diverse systems remains untapped due to the inability to systematically aggregate and query these data.

2. Methodology

The authors developed the PhageExpressionAtlas, a comprehensive bioinformatics resource designed to store, process, and visualize time-resolved dual RNA-seq data.

Data Collection and Curation

• Scope: The atlas aggregates 42 time-resolved dual RNA-seq datasets from 23 peer-reviewed studies.
• Diversity: The collection covers 25 bacteriophages (84% virulent, 16% temperate) infecting diverse hosts, including clinically relevant pathogens (Staphylococcus aureus, Pseudomonas aeruginosa, Escherichia coli).
• Inclusion Criteria: Datasets must cover at least two time points during infection, have public raw data (or author-provided access), and include sufficient strain/genome metadata.

Unified Processing Pipeline

To ensure comparability, the authors implemented a custom Nextflow pipeline:

1. Preprocessing: Quality control (FastQC), adapter trimming (Cutadapt), and mapping against combined host-phage reference genomes using HISAT2.
2. Quantification: Gene-level counting via featureCounts, retaining primary alignments but allowing multi-mapping reads.
3. Normalization:
   • Removal of rRNA features.
   • Calculation of Transcripts Per Million (TPM) per sample.
   • Aggregation of biological replicates into mean TPM values ("TPM means").
   • Calculation of fractional expression (gene TPM at time $t$ divided by its max TPM across all time points) for profile visualization.
4. Classification: Phage genes are assigned to Early, Middle, or Late infection phases based on temporal expression patterns using two strategies:
   • Class Max: Assigns the phase where expression is maximal.
   • Class Threshold: Assigns the earliest phase where expression exceeds a fraction ($\theta$) of the gene's maximum.

Database and Architecture

• Storage: An SQLite3 database stores four expression matrices (raw counts, per-sample TPM, mean TPM, fractional expression) alongside metadata and genome annotations (GFF).
• Annotation Enhancement: The pipeline integrates Pharokka to improve functional annotation of phage genes and standardize functional categories.
• Frontend/Backend:
  • Backend: Python/Flask with SQLAlchemy for API requests.
  • Frontend: HTML/CSS/JS using Shoelace components, Plotly.js for interactive plots (heatmaps, profile plots), and GoslingJS for genome visualization.
  • Features: Users can search datasets, visualize expression dynamics, classify genes, and inspect genomic context via circular and linear genome views.

3. Key Contributions

1. First Dedicated Resource: The PhageExpressionAtlas is the first centralized, interactive resource specifically for time-resolved dual RNA-seq data in phage research.
2. Standardized Re-analysis: It provides uniformly processed data for 42 datasets, enabling direct comparison across different phage-host systems that was previously impossible.
3. Interactive Visualization: The web interface allows non-experts to explore expression dynamics, classify genes, and visualize genomic organization without coding.
4. Benchmarking Framework: The authors used the atlas to systematically benchmark phage gene classification methods against proxy ground truths.

4. Key Results

Validation and Reproducibility

• The pipeline was validated against the T4-E. coli dataset, showing high concordance ($R^2 > 0.9$) with previously published TPM matrices.
• The atlas successfully replicated key findings from original studies, such as the temporal separation of T4 gene classes (early, middle, late) and the identification of specific transcriptional cascades (e.g., gp33/45/55 complex activating late genes).

Phage Gene Classification Benchmarking

• Using a proxy ground truth (genes involved in lysis, packaging, and assembly), the authors evaluated classification strategies.
• "Class Max" (assigning based on peak expression) demonstrated superior recall and precision compared to "Class Threshold" across diverse datasets.
• Uncharacterized Genes: A significant portion of genes in all temporal classes (early, middle, late) remain uncharacterized, highlighting the utility of temporal profiling as a functional annotation tool.
• Context Dependence: Classification results were sensitive to host strain and growth conditions, emphasizing the need for standardized cross-study analysis.

Transcriptional Dynamics of Defense Systems

• Host Defense: Most host defense systems (CRISPR-Cas, Restriction-Modification, Toxin-Antitoxin) showed a decrease in relative abundance during infection, consistent with global host transcriptome degradation. However, some systems (e.g., AbiD, Ceres) showed transient early induction.
• Phage Anti-Defense:
  • Anti-RM and Anti-TA: Phage-encoded anti-defense modules were often expressed early, aligning with the rapid action of nucleic-acid-targeting defenses.
  • Anti-CBASS: In T4, anti-CBASS genes (e.g., 57B/acb1) showed middle-to-late expression, consistent with the timing of CBASS activation (triggered by cyclic nucleotides).
• Orthologous Analysis:
  • Host housekeeping genes and ribosomal proteins consistently declined in relative abundance.
  • Phage orthologs with known functions (e.g., nrdA for DNA replication, frd for early metabolism) showed conserved temporal peaks (middle and early, respectively) across diverse hosts.

5. Significance and Future Outlook

• Democratization: The atlas lowers the barrier to entry for transcriptomics-driven phage research, allowing biologists to perform integrative cross-study assessments without extensive bioinformatics expertise.
• Mechanistic Insights: By placing defense and counter-defense modules in their native infection contexts, the resource reveals conserved transcriptional strategies and unique adaptations across the phage-host arms race.
• Foundation for AI/ML: The standardized, labeled dataset serves as a crucial training and benchmarking resource for future machine learning models aimed at predicting gene function or class from sequence data.
• Scalability: The authors plan to continuously integrate new datasets and expand modalities to include proteomics and differential RNA-seq, moving toward a holistic multi-omics view of phage infections.

Availability:

• Web Interface: https://phageexpressionatlas.cs.uni-tuebingen.de
• Code & Database: Available on GitHub and Zenodo.