Scrhla-Typing: A Method for Quantifying HLA Transcripts in Single Cells

BACKGROUND

The human leukocyte antigen (HLA) genes are the most polymorphic of the human genome and are central to the adaptive immune response against pathogens, alloantigens, and neoantigens in cancer. In allogeneic hematopoietic cell transplantation (allo-HCT), HLA molecules contribute to graft rejection and survival, to the graft-vs.-leukemia effect, and in many cases to patient responses to checkpoint inhibitor-based immunotherapies. Allele-specific HLA downregulation or complete loss is consistently linked to malignancy recurrence. Therefore, knowledge of HLA genotypes and expression levels are increasingly becoming helpful in risk-stratifying patients ahead of therapy. Despite advances in sequencing technologies, accurate HLA typing remains a challenge due to the high level of homology between the HLA genes and alleles, and there is not a method for HLA typing and allele-specific quantification in situations where more than two genotypes per gene coexist, like in allo-HCT. We present scrHLA-typing, a method to enrich HLA full transcripts, barcoded by single cells, by hybridization capture and accurately call and quantify alleles by long-read sequencing. We test our method on bone marrow aspirates (BMA) of relapsed leukemia post-allo-HCT.

METHODS

After generating a single-cell barcoded whole transcriptome amplified pool (cDNA pool) from BMA mononuclear cells via the 10x Genomics platform, we used custom biotinylated DNA oligonucleotides (hybridization probes) to specifically hybridize and enrich HLA cDNA molecules from the resulting libraries. We then 1) removed template switch oligo (TSO) priming artifacts, 2) amplified enriched products, and 3) ligated them with bell-shaped oligonucleotide adaptors for PacBio long-read sequencing. We developed scrHLAtag, a command line tool used to align the reads to all HLA alleles in the IMGT/HLA reference. In parallel, we developed scrHLAmatrix, a package written in R to predict cell classification based on HLA expression patterns, to predict genotypes per single cell, to eliminate sequencing PCR duplicates, and to summarize read counts into Seurat-compatible matrices, ready for further analyses.

RESULTS

scrHLA-typing libraries were generated from marrow aspirates of 5 pediatric acute myeloid leukemia with suspected post-HCT relapse. Each sample yielded ~600K–900K reads with good complexity (90% of unique barcodes had 3 or fewer PCR-duplicated reads). During the first ‘unsupervised’ scrHLAtag iteration, uniquely mapped HLA alleles against the entire IMGT/HLA reference library were between 3963–6553 per sample. Finding allele candidates accounting for a majority of reads was achieved with scrHLAmatrix, which were then used to restrict the scrHLAtag reference library in later iterations. After 4 to 5 iterations, it converged to the list of putative HLA genotypes of each allogeneic entity in a sample (cross validated by clinically available genotyping). To validate single cell HLA labeling, we used souporcell, a genetic demultiplexing algorithm that labels single cells into distinct allogeneic entities by variant calling aligned reads. In general, our results and souporcell were highly consistent with HLA designation, with mappings to the ‘wrong’ cell occurring less than 0.5% of the time. Moreover, we found the original cDNA capture chemistry (5’ or 3’) influenced scrHLA-typing quality, with better yield and fewer mismappings and false positives in 5’ vs. 3’ captured libraries. Finally, our assay was able to measure statistically significant differential allele expressing per locus per cell, which was prevalent in some patient across the haplotype but not in other patients.

CONCLUSIONS

We developed a method which enables full length sequencing of HLA-transcripts that can be genotyped, traced to, and quantified in single cells. With the reliance on personalized immunotherapy, we anticipate the ability to distinguish patients and/or chimeric entities bearing from not bearing HLA allele-specific differential expression will become useful in future clinical applications.