68 Rapid detection of somatic variants in human leukocyte antigen class 1 genes from solid tumor samples

Ramit Bharanikumar, Karl R. Beutner, Aly A. Khan, Jason PereraRegular 2021 · DOI

Abstract

Human Leukocyte Antigen (HLA) class 1 proteins are essential for the presentation of tumor-specific neoantigens to CD8+ T-cells. Somatic mutations within HLA genes represent a critical immune escape mechanism, potentially diminishing the efficacy of Immune Checkpoint Blockade (ICB) therapy. However, the high degree of polymorphism and the presence of pseudogenes in the HLA locus make somatic variant calling exceptionally challenging for standard sequencing pipelines. We present a rapid, modular algorithm designed to detect these variants from next-generation sequencing (NGS) data. Validated against TCGA datasets and PacBio long-read sequencing, our method demonstrates a 10x increase in computational speed compared to current gold standards while maintaining high sensitivity. Furthermore, we identified a transcriptomic signature of Nonsense Mediated Decay (NMD) in samples with nonsense mutations, providing a biological link between HLA mutations and immune evasion.

Introduction

The Human Leukocyte Antigen (HLA) class 1 genes (HLA-A, -B, and -C) are fundamental to the adaptive immune response against malignancies. These genes encode surface proteins that present intracellular peptides to CD8+ cytotoxic T-lymphocytes. When a cell becomes cancerous, these proteins display neoantigens that signal the immune system to attack. However, tumors can acquire somatic mutations in their HLA genes to "hide" from the immune system, a primary driver of resistance to immunotherapy.

Despite their clinical importance, HLA genes are notoriously difficult to analyze. The HLA locus is characterized by extreme polymorphism and the presence of highly homologous pseudogenes, which frequently lead to alignment errors and false-positive variant calls in standard genomic pipelines. Existing tools often require significant computational resources and time, limiting their utility in high-throughput clinical or research settings.

Methods

To address these challenges, we developed a modular algorithm optimized for speed and accuracy in HLA variant detection. The workflow requires matched tumor and normal sequencing (BAM) files and the patient’s specific HLA type as input. The pipeline consists of four primary stages:

  1. Patient-Specific Alignment: Rather than using a generic reference genome, the algorithm aligns reads directly to the full-length HLA class 1 sequences corresponding to the individual’s specific HLA alleles.
  2. Error Filtering: A dedicated module filters reads to minimize noise from sequencing artifacts.
  3. Cross-Mapping Correction: The algorithm specifically accounts for the high sequence similarity between different HLA genes and pseudogenes, ensuring reads are correctly attributed to the proper locus.
  4. Somatic Calling and Annotation: By comparing tumor and normal samples, the tool identifies somatic variants and provides functional annotations to characterize their potential impact on protein function.

Results and Validation

The efficacy of the algorithm was validated using 41 samples from The Cancer Genome Atlas (TCGA) containing 46 previously characterized HLA mutations. Our method successfully identified 43 of these 46 mutations (93.5% sensitivity) with correct functional annotations. Crucially, the algorithm achieved a computational run-time an order of magnitude faster than current gold-standard approaches.

To further confirm the accuracy of our calls, we compared our results against eight mutations validated by PacBio long-read sequencing. Our algorithm detected 100% (8/8) of these high-confidence variants.

Beyond detection, we leveraged multi-modal TCGA data to explore the functional consequences of these mutations. By analyzing RNA-seq data from samples harboring nonsense mutations, we identified a distinct transcriptomic signature associated with the Nonsense Mediated Decay (NMD) pathway. NMD is a cellular surveillance mechanism that degrades mRNA transcripts containing premature stop codons. This finding suggests a clear biological mechanism: somatic HLA mutations trigger the destruction of HLA mRNA before it can be translated, effectively removing the "billboards" the immune system uses to identify the tumor.

Discussion and Conclusions

Our findings demonstrate that this modular algorithm provides a reliable and highly efficient solution for detecting somatic HLA mutations in solid tumors. By achieving a 10x speedup over existing tools, this approach moves HLA variant calling from a specialized research task to a viable component of high-throughput clinical diagnostics.

While the algorithm missed 3 of the 46 known TCGA mutations—likely due to low sequencing coverage or complex variants—its overall sensitivity and speed represent a significant advancement. The modular architecture of the pipeline allows for future expansion to include HLA class 2 genes and non-classical HLA loci, which also play vital roles in the tumor microenvironment. Ultimately, rapid HLA variant detection will be essential for better predicting patient responses to immunotherapy and understanding the mechanisms of tumor evolution.

Summary updated

Abstract

<h3>Background</h3> Human Leukocyte Antigens (HLA) class 1 proteins are important for recognizing tumor specific mutations (neoantigens) and presenting them to CD8+ T-cells. Somatic mutations in HLA genes can potentially reduce the set of neoantigens available for presentation to T-cells, providing a possible immune escape mechanism for tumors. The presence of mutations in HLA genes may thus affect an individual’s response to Immune Checkpoint Blockade therapy but calling somatic variants in HLA genes is challenging given high degrees of polymorphism and the presence of pseudogenes in HLA loci. Here, we present a rapid, modular algorithm to detect somatic variants in HLA class 1 genes from next-generation sequencing data. <h3>Methods</h3> Our method takes as input matched tumor and normal sequencing (BAM) files and a patient‘s HLA type. Our first step involves mapping and aligning reads to the full-length HLA class 1 genes of an individual patient. Next, we filter reads to minimize sequencing errors and process them to account for potential cross-mapping between genes in the broader HLA locus. Our algorithm uses the final set of processed reads from matched tumor and normal sequencing files to call somatic variants. The final set of variants are then provided with a genetic variant annotation that summarizes their functional effect. <h3>Results</h3> We validated the efficacy of our approach using TCGA samples that were previously known to contain somatic HLA mutations. Out of a total of 46 mutations previously characterized from 41 samples, our algorithm detected 43 mutations with the correct functional annotation and did so with a computational run-time that is an order of magnitude faster than the current gold-standard approach. Eight of these mutations were previously validated using PacBio long-read sequencing, and all of these validated mutations were detected by our method. Additionally, we analyzed transcriptomic data in samples containing nonsense mutations and detected a transcriptomic signature in RNA-seq data that corresponded to the Nonsense Mediated Decay pathway. <h3>Conclusions</h3> Our findings show that our algorithm can reliably and rapidly detect somatic mutations in HLA class 1 classical genes. We leveraged multi-modal TCGA data and modules developed at Tempus to link samples with nonsense mutations to the nonsense mediated decay pathway. This allowed us to posit a possible biological mechanism that could result in increased resistance to immunotherapy because of somatic HLA mutations. Finally, we emphasize that our approach is modular and can be extended to call somatic variants in other non-classical and class 2 HLA genes.
BibTeX
@inproceedings{bharanikumar2021rapid,
  author = {Ramit Bharanikumar and Karl R. Beutner and Aly A. Khan and Jason Perera},
  title = {{68 Rapid detection of somatic variants in human leukocyte antigen class 1 genes from solid tumor samples}},
  booktitle = {Regular},
  year = {2021},
  doi = {10.1136/jitc-2021-sitc2021.068},
}
Figure