A New HLA Genotyping Algorithm Shows Extremely High Accuracy and Concordance Between Tumor and Normal Samples

Background
HLA genotyping, specifically HLA-A*02:01 is an actionable marker in uveal melanoma for Tebentafusp, a bispecific gp100 peptide-HLA-directed CD3 T cell engager, with ongoing clinical trials investigating its use in HLA-A*02:01 positive cutaneous and mucosal melanoma. However, prior work has shown a discrepancy in this marker between blood and somatic tumor testing (Seedor R et al ASCO 2023). Given therapeutic implications of these discordances, a better understanding of tumor versus normal HLA testing concordance is critical.

Methods
The Tempus HLA genotyping algorithm was run on a cohort of 191 patients, each with a tumor and a normal sample. The reference HLA genotype was determined by an independent lab using the GenDx HLA genotyping kit and long read sequencing of the normal sample. Samples that failed sequencing in either lab were excluded from the analysis. Concordance between tumor and normal samples was additionally evaluated on a cohort of 7,821 solid-tumor samples from a variety of sites, selected from the Tempus multimodal database.

Results
The Tempus HLA genotyping algorithm is highly accurate ( > 98.9% on all HLA genes except DQA1). Specifically, it has an accuracy of 100% on the HLA-A gene, using either the tumor or normal sample. On the larger tumor / normal comparison cohort (7,821 samples), the Tempus HLA genotyping algorithm is highly concordant ( > 99% for all genes except HLA-DQA1 [94.3%]). For the 183 melanoma samples in this cohort, concordance was similar with 100% for HLA-A and HLA-B and 99.5% (182/183) for HLA-C, further confirming that HLA genotyping can be performed accurately on melanoma tumor samples. Investigation of all discordances on HLA genes for which HLA LOH results were available (HLA-A, -B and -C) shows that 84% (88/105) of those discordant samples have somatically lost the allele responsible for the discordance. Of LOH positive discordant samples, 73% (64/88) had a tumor purity ≥70%. Thus, over half (64/105) of the discordances between tumor- and normal-HLA genotyping are caused by high tumor purity HLA LOH positive samples. On samples meeting these criteria, concordance drops to 86%-89%.

Conclusions
The Tempus HLA genotyping algorithm is an extremely accurate laboratory developed test that provides accurate results, regardless of whether the normal or tumor sample is used. Concordance between normal- and tumor-based results is above 99.5% in a large-scale dataset, provided the sample does not have HLA LOH and a tumor purity over 70%. These findings suggest that this test can rapidly screen patients for HLA genotype and match them to appropriate clinical trials but emphasize the importance of a normal sample for high tumor purity specimens.