Jumper enables discontinuous transcript assembly in coronaviruses

Palash Sashittal, Chuanyi Zhang, Jian Peng, Mohammed El-Kebir

TL;DR

This paper introduces JUMPER, a novel computational method for assembling viral transcripts from coronaviruses, which utilize a unique "discontinuous transcription" process. JUMPER accurately identifies both known (canonical) and previously unknown (non-canonical) viral transcripts, outperforming existing methods and enabling detailed analysis of viral gene expression, including responses to drug treatments.

What Problem Does This Paper Address?

Eukaryotic gene expression typically involves alternative splicing, where different parts of a gene can be included or excluded to create various mRNA molecules. Coronaviruses, however, employ a distinct mechanism called "discontinuous transcription." Here, the viral RNA polymerase skips over segments of the viral RNA template, creating a set of "discontinuous transcripts" that are still ordered according to the original viral genome. While researchers have identified evidence of these split reads, a comprehensive method to assemble these discontinuous transcripts and quantify their abundances has been lacking. This gap hinders a complete understanding of viral gene expression, especially under different conditions like drug treatment, and limits the ability to study the full viral transcriptome.

Methods

The core of this work is the development of JUMPER, a method designed to tackle the "Discontinuous Transcript Assembly" (DTA) problem. JUMPER takes short-read sequencing data (RNA-seq) as input and models the discontinuous transcription process.

Key methodological aspects include:

  • Segment Graph Construction: JUMPER first constructs a "segment graph" from the aligned sequencing reads. This graph represents contiguous genomic regions as nodes and splice junctions as edges, specifically accounting for the unique constraints of discontinuous transcription.
  • Maximum Likelihood Model: The method employs a maximum likelihood framework to infer the set of transcripts and their abundances that best explain the observed sequencing data. This model accounts for varying transcript lengths and the specific nature of discontinuous transcription.
  • Mixed Integer Linear Programming (MILP): JUMPER uses a sophisticated MILP solver to find the optimal set of transcripts and abundances.
  • Progressive Heuristic: To handle the computational complexity of large datasets, JUMPER implements a progressive heuristic. This approach iteratively builds the set of transcripts, improving efficiency and scalability.
  • Paired-End Read Utilization: The method leverages the additional information provided by paired-end sequencing reads, which can offer stronger evidence for transcript structure.

Datasets

The paper utilizes several real-world datasets for evaluation and validation:

  • Simulated Data: The authors generated simulated RNA-seq data based on a segment graph derived from a SARS-CoV-2 sample. This allowed for controlled evaluation of JUMPER's performance against known ground truth.
  • SARS-CoV-2 Vero Cells: Short-read sequencing data from SARS-CoV-2-infected Vero cells was used. Crucially, this was complemented by long-read sequencing data from the same sample for validation of JUMPER's predictions.
  • SARS-CoV-1 and MERS-CoV: Publicly available short-read sequencing data from cells infected with SARS-CoV-1 and MERS-CoV were analyzed to demonstrate JUMPER's generalizability across different coronaviruses.
  • SARS-CoV-2 A549 Cells (Drug Response): Data from A549 lung alveolar cells infected with SARS-CoV-2, with and without prior treatment with the drug ruxolitinib, were analyzed to investigate viral drug response at the transcript level.

Key Insights and Ideas

  • Discontinuous Transcription is Complex: The study highlights that discontinuous transcription in coronaviruses is not just a simple skipping mechanism but leads to a surprisingly complex transcriptome with numerous non-canonical transcripts.
  • Non-Canonical Transcripts Can Be Functional: Contrary to initial assumptions, many of these non-canonical transcripts appear to encode complete or truncated viral proteins, suggesting they play roles in the viral life cycle.
  • JUMPER's Superiority: JUMPER's specialized approach for discontinuous transcription allows it to significantly outperform general-purpose transcript assembly tools (SCALLOP, STRINGTIE) in both accuracy and the number of transcripts identified.
  • Transcript-Level Drug Response: The ability to analyze viral gene expression at the transcript level opens new avenues for understanding how drugs affect viral infection, revealing subtle changes in transcript abundance that might be missed at the gene level.
  • Generalizability: JUMPER's effectiveness extends beyond SARS-CoV-2, demonstrating its utility for studying other coronaviruses like SARS-CoV-1 and MERS-CoV.

Key Results

  • Simulation Performance: JUMPER achieved a significantly higher F1 score (median 0.255) compared to SCALLOP (median 0.062) and STRINGTIE (median 0.019) on simulated data.
  • SARS-CoV-2 Vero Cells: JUMPER identified transcripts for all known viral proteins except ORF10 and predicted nine non-canonical transcripts, which were subsequently validated by long-read data.
  • SARS-CoV-1 and MERS-CoV: JUMPER reconstructed transcripts for most ORFs, including canonical and non-canonical variants, with consistent abundances across samples.
  • Drug Response Analysis: In A549 cells, JUMPER identified a variant of the spike protein with a deletion near the furin cleavage site that showed a statistically significant increase in abundance after ruxolitinib treatment.

Limitations and Future Work

  • Oligo(dT) Amplification Bias: Currently, JUMPER is best suited for data generated with oligo(dT) amplification, which targets the poly(A) tail. The authors plan to extend the model to accommodate data from SARS-CoV-2-specific primers, which are more common.
  • Fixed Read Length Assumption: The current model assumes a fixed read length, which is a limitation for some long-read sequencing technologies. Future work will address variable read lengths.
  • Mutation and Diversity: The paper suggests studying the impact of mutations and within-host viral diversity on transcriptomes.
  • Optimal Solutions: Exploring the space of potentially multiple optimal solutions to the DTA problem is an area for future research.
  • General Transcript Assembly: Extending JUMPER's capabilities to assemble transcriptomes across multiple genes in eukaryotes is a long-term goal.

Summary updated

Abstract

Genes in SARS-CoV-2 and other viruses in the order of Nidovirales are expressed by a process of discontinuous transcription which is distinct from alternative splicing in eukaryotes and is mediated by the viral RNA-dependent RNA polymerase. Here, we introduce the DISCONTINUOUS TRANSCRIPT ASSEMBLYproblem of finding transcripts and their abundances given an alignment of paired-end short reads under a maximum likelihood model that accounts for varying transcript lengths. We show, using simulations, that our method, JUMPER, outperforms existing methods for classical transcript assembly. On short-read data of SARS-CoV-1, SARS-CoV-2 and MERS-CoV samples, we find that JUMPER not only identifies canonical transcripts that are part of the reference transcriptome, but also predicts expression of non-canonical transcripts that are supported by subsequent orthogonal analyses. Moreover, application of JUMPER on samples with and without treatment reveals viral drug response at the transcript level. As such, JUMPER enables detailed analyses of Nidovirales transcriptomes under varying conditions.
BibTeX
@inproceedings{sashittal2021jumper,
  author = {Palash Sashittal and Chuanyi Zhang and Jian Peng and Mohammed El-Kebir},
  title = {{Jumper enables discontinuous transcript assembly in coronaviruses}},
  booktitle = {Nature Communications},
  year = {2021},
  doi = {10.1038/s41467-021-26944-y},
}
Figure