TL;DR
This paper introduces CTAT-LR-fusion, a new computational tool that accurately identifies gene fusions from long-read RNA sequencing data, either alone or combined with short-read data. This advancement is crucial for cancer diagnostics and therapy, as it leverages the higher resolution of long-read sequencing to detect fusion transcripts with greater sensitivity and accuracy than previous methods.
What Problem Does This Paper Address?
Gene fusions, resulting from chromosomal rearrangements, are key drivers in many cancers, acting as diagnostic markers and therapeutic targets. Accurately detecting these fusions is vital for personalized medicine. Traditional methods rely on short-read RNA sequencing (RNA-seq), which, while accurate, struggles to fully resolve complex fusion isoforms and can miss certain fusion events. The advent of long-read sequencing technologies (like PacBio and Oxford Nanopore) offers the potential to capture full-length transcripts, providing unprecedented resolution for fusion detection. However, existing computational tools for long-read fusion detection are relatively new, and their performance, especially in challenging scenarios like single-cell analysis, needs thorough evaluation and improvement. This paper addresses this gap by developing and validating CTAT-LR-fusion, a tool designed to harness the power of long-read sequencing for comprehensive fusion transcript identification.
Methods
CTAT-LR-fusion operates in two main phases:
- Candidate Fusion Detection: It uses a customized version of the minimap2 aligner (ctat-minimap2) to rapidly identify potential fusion reads. This phase focuses on reads that show preliminary alignments to multiple genomic loci, efficiently filtering out non-chimeric reads.
- Contig Modeling, Realignment, and Quantification: Candidate fusion gene pairs are modeled as collinear gene contigs using FusionInspector. The candidate reads are then realigned to these contigs for precise breakpoint identification and quantification based on the number of supporting long-read fusions.
Crucially, CTAT-LR-fusion can also integrate companion short-read RNA-seq data. If available, FusionInspector is used to capture short-read evidence, which is then combined with long-read results for a more robust final report. The tool also includes an interactive IGV-report for visualization.
Datasets
The authors evaluated CTAT-LR-fusion using several datasets:
- Simulated Long Reads: They used existing simulated data from the JAFFAL publication (leveraging Badread) with varying error rates (25% to 5% divergence) and sequencing technologies (PacBio and ONT). They also generated new simulated data using PBSIM3 to better reflect current PacBio HiFi and ONT R10.4.1 error profiles.
- Real Data:
- SeraCare Fusion RNA Mix v4: A commercial reference sample containing 16 known oncogenic fusion transcripts, used for bulk RNA-seq analysis.
- Nine Cancer Cell Lines: Transcriptomes from various cancer cell lines were sequenced using PacBio MAS-ISO-seq.
- Tumor Single Cell Samples: Data from a T-cell infiltrated melanoma tumor and three metastatic high-grade serous ovarian carcinoma (HGSOC) samples were analyzed.
Key Insights and Ideas
- Long Reads are Key for Isoform Resolution: The paper strongly emphasizes that long-read sequencing is essential for fully resolving the complex splicing isoforms of fusion transcripts, which short reads often fragment.
- Combining Long and Short Reads Maximizes Detection: While long reads offer superior isoform resolution, combining them with short reads significantly enhances the overall detection sensitivity and accuracy, capturing a broader spectrum of fusion events.
- Order and Orientation Matter: The authors highlight that correctly reporting the order and orientation of fused genes is critical, as different orientations can have vastly different oncogenic interpretations and clinical implications. CTAT-LR-fusion, along with JAFFAL and pbfusion, excels in this regard.
- Single-Cell Fusion Detection is Challenging but Possible: The study demonstrates the application of CTAT-LR-fusion to single-cell data, revealing that long reads are often superior for detecting fusion-expressing cells, especially for complex or low-abundance fusions.
Key Results
- Superior Accuracy: CTAT-LR-fusion demonstrated top performance in accuracy benchmarks against alternative long-read fusion detection tools, particularly when considering gene order and orientation.
- High Sensitivity: In both bulk and single-cell analyses, long isoform reads generally yielded higher sensitivity for fusion detection compared to short reads.
- SeraCare Control Sample: CTAT-LR-fusion successfully identified all 16 control fusions in the SeraCare sample, outperforming other methods in some aspects.
- Cancer Cell Lines: The tool identified numerous fusion transcripts across nine cancer cell lines, with a significant portion agreed upon by multiple long-read prediction methods.
- Single-Cell Applications: In melanoma and HGSOC single-cell samples, CTAT-LR-fusion identified fusion transcripts that distinguished tumor from normal cells, with long reads often providing more comprehensive evidence. For instance, in one melanoma sample, ~60% of tumor cells with a specific fusion were identified solely by long-read evidence.
Limitations and Future Work
- Benchmarking Challenges: The authors note that benchmarking accuracy in real cancer cell lines is challenging due to the lack of comprehensive, absolute truth sets. They employed a "proxy truth set" based on consensus predictions.
- Artifacts and Normal Fusions: While CTAT-LR-fusion aims to filter artifacts and common normal fusions, the paper acknowledges the ongoing challenge of distinguishing true oncogenic fusions from benign variants.
- Data Integration: The paper suggests that further optimization of integrating long and short read data could yield even greater detection sensitivity.
- Routine Long-Read Sequencing: As long-read sequencing becomes more routine and cost-effective, the capabilities of CTAT-LR-fusion are expected to become increasingly valuable for a wider range of applications.
Summary updated