TL;DR
This paper introduces TRUST, a novel open-source algorithm that enables ultrasensitive detection of T cell receptor (TCR) hypervariable CDR3 regions directly from standard RNA sequencing data of solid tissues. This breakthrough allows researchers to study the T cell immune landscape without the need for specialized TCR sequencing, opening new avenues for understanding tumor immunity and autoimmune diseases.
What Problem Does This Paper Address?
Characterizing the T cell receptor (TCR) repertoire within tissues is crucial for understanding how the immune system interacts with tumors and in autoimmune diseases. Traditionally, this requires dedicated TCR sequencing (TCR-seq), which involves targeted PCR capture and deep sequencing. However, TCR-seq is often unavailable for clinical studies due to limited tissue material and the higher priority placed on comprehensive genomic and transcriptomic profiling. This creates a significant gap: researchers need to profile the tumor and its infiltrating immune cells but lack a cost-effective and readily available method to do so using existing RNA-seq data. This paper addresses this by developing a computational tool to repurpose standard RNA-seq data for TCR repertoire analysis.
Methods
The core of this paper is the development and validation of the TRUST (TCR Repertoire Utilities for Solid Tissue) algorithm. TRUST operates on RNA-seq reads that have already been aligned to a reference genome. Its workflow involves several key steps:
- Library Type Detection: TRUST automatically determines if the input RNA-seq library is paired-end (PE) or single-end (SE).
- Read Sorting for CDR3 Detection: It identifies informative reads or read pairs. For PE libraries, this means keeping pairs where one mate maps to a TCR gene locus and the other is unmapped. For SE libraries, it selects unmapped reads.
- Sorting Reads by Mapping or Consensus Motifs: TRUST further refines these reads by either using the mapping of the paired mate (for PE) to identify TCR gene regions or by searching for conserved CDR3 motifs (for SE).
- Assigning Reads to TCR Genes: Reads are then assigned to specific TCR genes (variable, joining, or constant) based on their mapping location or identified motifs.
- De Novo Assembly: For reads assigned to a TCR gene, TRUST performs pairwise sequence comparisons to assemble reads into contigs based on overlapping sequences. This step is crucial for reconstructing the full CDR3 sequence.
- Realigning to IMGT References and Annotation: Finally, assembled contigs are realigned to IMGT (the international ImMunoGeneTics information system) reference sequences. Contigs that contain a consensus CDR3 motif and have at least six consecutive matched amino acids in their flanking regions are reported as valid CDR3 assemblies, along with their associated variable (V) and joining (J) genes.
A key innovation is TRUST's ability to handle unmapped reads and its robust de novo assembly approach, which allows it to reconstruct CDR3 sequences even when they don't perfectly map to known reference sequences due to variations or errors.
Datasets
The study utilized several datasets for evaluation:
- Formalin-Fixed Paraffin-Embedded (FFPE) Samples: Three FFPE samples from TCGA kidney clear cell carcinoma with available RNA-seq data were used to validate TRUST's ability to call real CDR3 sequences by comparing its output to TCR-seq data from the same samples.
- In Silico Simulations: Simulated RNA-seq data was generated to systematically evaluate TRUST's performance across various read depths, read lengths, and library types (SE and PE). This allowed for precise control over ground truth and comparison with other methods.
- Publicly Available Datasets: Nine samples from public databases were used to test TRUST's applicability to non-cancerous solid tissues. These included six negative control cell line samples from ENCODE and three colon tissue samples from NCBI SRA (some with autoimmune disease conditions).
Key Insights and Ideas
- Repurposing RNA-seq for TCR Analysis: The most significant insight is that standard RNA-seq data, commonly generated for gene expression studies, can be effectively repurposed to profile the TCR repertoire. This bypasses the need for specialized TCR-seq, making immune profiling more accessible and cost-effective.
- High Sensitivity in Low-Coverage Scenarios: TRUST demonstrates remarkable sensitivity in detecting CDR3 sequences, even in samples with low sequencing depth. This is critical for analyzing precious or degraded tissue samples.
- Superior Performance Compared to Existing Tools: In silico simulations showed that TRUST achieved significantly higher recall rates (up to 2.1% at 100M reads) compared to competing methods like MiXCR (0.12%) and iSSAKE (0.0%) under similar low-coverage conditions.
- Enrichment for High-Abundance Clonotypes: TRUST-identified CDR3 sequences were enriched for high-frequency clonotypes, suggesting that the algorithm is adept at capturing the most abundant T cell populations, which are often the most functionally relevant.
- Broad Applicability: The study demonstrated TRUST's utility not only in cancer samples but also in non-cancerous inflammatory tissues, indicating its potential for a wide range of immunological research.
Key Results
- Validation with TCR-seq: Approximately two-thirds of TRUST-assembled CDR3s from FFPE samples were found in the corresponding TCR-seq data, confirming its ability to identify genuine TCR sequences.
- Recall and Precision: In silico simulations showed that TRUST achieved an average recall of 2-4% for RNA-seq data with 100M reads. With advancements in sequencing technology (e.g., PE 150nt libraries at 300M fragments), TRUST can achieve a 9.4% average recall with 96.6% precision.
- Comparison with Competitors: At a read depth of 100M, TRUST achieved a recall of 2.1%, significantly outperforming MiXCR (0.12%) and iSSAKE (0.0%).
- High Abundance Clonotype Recovery: TRUST identified an average of 53.5% of the top 1% most abundant CDR3s.
- Non-Cancerous Tissue Analysis: TRUST successfully detected significantly higher T cell infiltration in inflamed colon tissues compared to normal tissues, aligning with expectations for autoimmune conditions.
Limitations and Future Work
- Coverage Limitations: While TRUST performs well at low sequencing depths, its recall is still dependent on the overall depth of the RNA-seq data. The paper notes that approximately one-tenth of the TCR repertoire can be accurately retrieved by repurposing RNA-seq data.
- FFPE Sample Challenges: DNA degradation in FFPE samples can affect the completeness of the tTIL repertoire captured. Additionally, variations in tissue location between RNA-seq and TCR-seq samples can lead to discrepancies.
- Future Directions: The authors anticipate that TRUST will attract broad interest due to accumulating RNA-seq data and decreasing sequencing costs. They suggest its application to large cancer cohorts and its potential for analyzing non-cancerous inflammatory solid tissues. The development of TRUST represents a significant step towards more integrated and accessible immune profiling in various research fields.
Summary updated