Next‐generation sequencing can reveal in vitro‐generated PCR crossover products: some artifactual sequences correspond to HLA alleles in the IMGT/HLA database

Abstract

The high-resolution human leukocyte antigen (HLA) genotyping assay that we developed using 454 sequencing and Conexio software uses generic polymerase chain reaction (PCR) primers for DRB exon 2. Occasionally, we observed low abundance DRB amplicon sequences that resulted from in vitro PCR 'crossing over' between DRB1 and DRB3/4/5. These hybrid sequences, revealed by the clonal sequencing property of the 454 system, were generally observed at a read depth of 5%-10% of the true alleles. They usually contained at least one mismatch with the IMGT/HLA database, and consequently, were easily recognizable and did not cause a problem for HLA genotyping. Sometimes, however, these artifactual sequences matched a rare allele and the automatic genotype assignment was incorrect. These observations raised two issues: (1) could PCR conditions be modified to reduce such artifacts? and (2) could some of the rare alleles listed in the IMGT/HLA database be artifacts rather than true alleles? Because PCR crossing over occurs during late cycles of PCR, we compared DRB genotypes resulting from 28 and (our standard) 35 cycles of PCR. For all 21 cell line DNAs amplified for 35 cycles, crossover products were detected. In 33% of the cases, these hybrid sequences corresponded to named alleles. With amplification for only 28 cycles, these artifactual sequences were not detectable. To investigate whether some rare alleles in the IMGT/HLA database might be due to PCR artifacts, we analyzed four samples obtained from the investigators who submitted the sequences. In three cases, the sequences were generated from true alleles. In one case, our 454 sequencing revealed an error in the previously submitted sequence.

HAPCAD: An open-source tool to detect PCR crossovers in next-generation sequencing generated HLA data

“…HAPCAD results indicate that of the 1,822 DRB alleles in the IMGT/HLA Database version 3.20.0, at least 1,175 of those alleles could be generated as a hybrid amplicons in 7,318,367 discrete DRB allele combinations, substantially increasing the observed repertoire of possible hybrid amplicons reported in Table 1 and by Holcomb et al [6]. A caveat is that HAPCAD inflates the potential number of likely hybrid amplicons due to the inability to rule out the use of hybrid amplicons as parent alleles as well.…”

“…One consequence of this co-amplification of these closely-related loci is that amplicons may be generated that do not reflect the genome. Specifically, spurious PCR products are generated that represent chimeras of different loci, through a phenomenon termed “jumping” PCR (or PCR crossover), in which a partial PCR product dissociates from its template and anneals to a template generated from a different allele (Figure 1) [5,6]. This produces an amplicon that combines sequences from two different HLA alleles into a hybrid sequence or PCR crossover product.…”

“…This phenomenon can happen with amplification of two alleles of a single locus; however, amplification of multiple DRB loci with a single set of primers appears to increase the likelihood. PCR crossover products defined by Holcomb et al [6], as a majority do not correspond to IMGT/HLA Database alleles and are easily detected due to mismatches between the crossover sequence and allele sequence [6]. The hybrid amplicon subgroup of PCR crossovers can however pass undetected because the sequences are exact matches to IMGT/HLA Database alleles.…”

“…Importantly, whether hybrid amplicons are also reflective of the genome in some cases was tested by Holcomb et al using experimental methods shown to reduce PCR crossover product generation [6], the most significant change reflecting a reduction in PCR cycle number from 35 to 28 cycles. Using samples originally used to characterize an allele within the IMGT/HLA Database [6,7], also observed as PCR crossover in a 35 cycle PCR reaction, Holcomb et al report that three of four allele sequences, here defined as hybrid amplicons, appeared to be true genomic sequence when the PCR protocol was modified to include only 28 cycles. The fourth PCR crossover allele sequence observed to differ from the IMGT/HLA Database allele by a single base validated a sequencing error in the originally reported sequence and was used to correct the IMGT/HLA Database.…”

“…This underscores the need to reduce the occurrence of PCR crossover events and to detect hybrid amplicons should they occur. Although reducing the number of PCR cycles from 35 to 28 cycles severely reduced the presence of hybrid amplicons [6] the possibility still exists that PCR crossover events can occur in future experiments. Consequently, all NGS-based HLA datasets should be scrutinized for potential PCR crossover-generated sequences.…”