Development of a high-resolution NGS-based HLA-typing and analysis pipeline

Abstract

The human leukocyte antigen (HLA) complex contains the most polymorphic genes in the human genome. The classical HLA class I and II genes define the specificity of adaptive immune responses. Genetic variation at the HLA genes is associated with susceptibility to autoimmune and infectious diseases and plays a major role in transplantation medicine and immunology. Currently, the HLA genes are characterized using Sanger- or next-generation sequencing (NGS) of a limited amplicon repertoire or labeled oligonucleotides for allele-specific sequences. High-quality NGS-based methods are in proprietary use and not publicly available. Here, we introduce the first highly automated open-kit/open-source HLA-typing method for NGS. The method employs in-solution targeted capturing of the classical class I (HLA-A, HLA-B, HLA-C) and class II HLA genes (HLA-DRB1, HLA-DQA1, HLA-DQB1, HLA-DPA1, HLA-DPB1). The calling algorithm allows for highly confident allele-calling to three-field resolution (cDNA nucleotide variants). The method was validated on 357 commercially available DNA samples with known HLA alleles obtained by classical typing. Our results showed on average an accurate allele call rate of 0.99 in a fully automated manner, identifying also errors in the reference data. Finally, our method provides the flexibility to add further enrichment target regions.

Figure 1.

Analysis software. The figure shows a screenshot of the GUI (graphical user interface) of the analysis software. For this figure, we combined two screenshots of two different data views, separated by the white diagonal line. The left part of the graph shows the achieved single start point coverage (see also Figure 2 for further details). The right part shows the corresponding single start point mappings of the same alignment. The different parts of the user interface are as follows: The list view on the left shows samples that are added for analysis. This sample list consists of the sample name followed by a colored rectangle. The colors are coding for the sample states, which are white (added for analysis), yellow (analysis running), green (analysis finished) and red (analysis failed). The middle part is divided into three sections. At the top, the user can select the locus for the selected sample. The ‘change view’ button at the right switches between coverage and read view. Below, the user finds the NGS data visualization of the determined HLA type. The table at the bottom shows a sorted list of the different possible HLA types. The top most HLA type is the most likely determined by the algorithm. The user can change that order to manually correct possible errors. At the right side of the GUI the user finds two additional tables. The upper table contains a sorted list of alleles that failed the initial QC (see Materials and Methods). The column ‘error’ shows the number of nucleotide positions, for which a QC failed value was calculated. The lower table shows the top 50 alleles that were not covered 100%. The table is sorted by the number of uncovered bases in ascending order. Alleles from these two tables can manually be included in the genotype calling. On the other hand, the user can move alleles from the allele calling to one of these tables. This allows for manual evaluation of the calling algorithm. Low covered alleles that failed the initial pre-filtering step can be added. Also degraded DNA, that is not 100% covered, can be analyzed. At the top of the application window, the user finds a small gray bar. When the mouse pointer hovers this area, a toolbox with buttons for sample adding, starting analysis and analysis report moves down (shown at the bottom, below the status bar). A tutorial that contains more details and an example workflow can be found at our online resource http://www.ikmb.uni-kiel.de/resources/download-tools/software/hlassign.

Figure 2.

Example of an erroneous reference HLA allele. The picture shows the unique start point coverage (top) and the corresponding short sequencing reads (bottom) of exon 2 for three different HLA-B alleles of IHW0994. The red background at the top shows the expected ideal coverage for a perfect single start point mapping with 100 bp reads and a minimum truncated mapping length of 70 bp at the exon boundaries. The black curve shows the sample's unique start point coverage and the corresponding reads are shown in the lower panel of the figure. The reference genotype for HLA-B of this sample is 38:01/51:06 as listed by the commercial provider. As shown above, the 3’ part of that exon is not equally covered, thus shows an abnormal read distribution (see purple lines), when considering 38:01 as a candidate allele. As the only difference between 38:01:01 and 38:02:01 are the nucleotides positioned 32 and 34 bases upstream of the 3’ end at exon 2, it is highly likely that the reference data set is wrong. Here we can show that the most probable HLA-B genotype of IHW0994 is 38:02/51:06 or 38:02:01/51:06:01 at the 6-digit level. This was later confirmed by Luminex® HLA typing in another laboratory.