Beyond HLA LOH: Alternative modes of HLA loss are common and vary by cancer type

The human leukocyte antigen (HLA) genes play a pivotal role in immune surveillance of tumors by presenting tumor-derived antigens to cytotoxic T cells, enabling immune recognition and elimination of malignant cells. Loss of heterozygosity (LOH), somatic mutations, epigenetic silencing and structural variations at the HLA locus are well-characterized mechanisms by which tumors evade immune detection, contributing to disease progression and resistance to immunotherapies. However, those mechanisms are rarely considered in conjunction. A comprehensive understanding of these diverse modes of HLA loss is critical for elucidating tumor-immune escape and optimizing immunotherapeutic strategies. We developed an integrated workflow leveraging the Tempus xT assay to detect somatic alterations in HLA genes. Germline HLA alleles were first genotyped using a validated proprietary assay. Patient-specific HLA alleles and decoy sequences were used to realign sequencing reads. The realigned reads were fed into standard variant calling and RNA expression quantification workflows to establish HLA variants and HLA expression. HLA variants were filtered to remove artifacts. HLA expression was normalized and bias corrected using machine learning models trained on over 2,000 samples. Total HLA expression loss was defined as the trimmed mean of M values (TMM) being more than 2 standard deviations below the gene- and tumor-purity-stratified mean. HLA LOH was determined using the validated Tempus HLA LOH device. We applied our workflow to 11,000 cancer samples, assessing HLA genotypes, LOH, somatic variants, and expression. Our HLA loss detection methods showed a high level of concordance, with DNA-based alterations evident at the RNA level. 61% of HLA alleles with LOH and/or loss of function (LOF) variants also displayed loss of HLA RNA expression, which included both allele-specific expression (ASE) and total expression loss from both HLA alleles. 12% of tumors without any DNA loss events displayed RNA-level HLA loss, potentially due to epigenetic regulation, or other causes. The prevalence and molecular mechanisms of HLA loss varied by cancer type. LOH was the dominant mechanism for HLA loss in most cancers, with exceptionally high rates of LOH in head and neck squamous cell carcinoma (46%, 92% of loss events) and lung squamous cell carcinoma (32%, 87% of loss events). Conversely, somatic LOF variants were the dominant mechanism in MSI-H colorectal cancer (22%, 51% of loss events), and expression loss dominated in prostate cancers (21% of cases, 84% of loss events). As HLA-restricted immunotherapies expand, accurate and comprehensive characterization of HLA gene alterations in tumors is increasingly important. Our findings highlight that, beyond LOH, LOF variants and transcriptional silencing are significant contributors to HLA loss in certain cancers and should be routinely assessed to inform immunotherapeutic decision-making.