MICon Contamination Detection Workflow for Next-Generation Sequencing Laboratories Using Microhaplotype Loci and Supervised Learning

Abstract

Innovation in sequencing instrumentation is increasing the per-batch data volumes and decreasing the per-base costs. Multiplexed chemistry protocols after the addition of index tags have further contributed to efficient and cost-effective sequencer utilization. With these pooled processing strategies, however, comes an increased risk of sample contamination. Sample contamination poses a risk of missing critical variants in a patient sample or wrongly reporting variants derived from the contaminant, which are particularly relevant issues in oncology specimen testing in which low variant allele frequencies have clinical relevance. Small custom-targeted next-generation sequencing (NGS) panels yield limited variants and pose challenges in delineating true somatic variants versus contamination calls. A number of popular contamination identification tools have the ability to perform well in whole-genome/exome sequencing data; however, in smaller gene panels, there are fewer variant candidates for the tools to perform accurately. To prevent clinical reporting of potentially contaminated samples in small next-generation sequencing panels, we have developed MICon (Microhaplotype Contamination detection), a novel contamination detection model that uses microhaplotype site variant allele frequencies. In a heterogeneous hold-out test cohort of 210 samples, the model displayed state-of-the-art performance with an area under the receiver-operating characteristic curve of 0.995. Innovation in sequencing instrumentation is increasing the per-batch data volumes and decreasing the per-base costs. Multiplexed chemistry protocols after the addition of index tags have further contributed to efficient and cost-effective sequencer utilization. With these pooled processing strategies, however, comes an increased risk of sample contamination. Sample contamination poses a risk of missing critical variants in a patient sample or wrongly reporting variants derived from the contaminant, which are particularly relevant issues in oncology specimen testing in which low variant allele frequencies have clinical relevance. Small custom-targeted next-generation sequencing (NGS) panels yield limited variants and pose challenges in delineating true somatic variants versus contamination calls. A number of popular contamination identification tools have the ability to perform well in whole-genome/exome sequencing data; however, in smaller gene panels, there are fewer variant candidates for the tools to perform accurately. To prevent clinical reporting of potentially contaminated samples in small next-generation sequencing panels, we have developed MICon (Microhaplotype Contamination detection), a novel contamination detection model that uses microhaplotype site variant allele frequencies. In a heterogeneous hold-out test cohort of 210 samples, the model displayed state-of-the-art performance with an area under the receiver-operating characteristic curve of 0.995. Next-generation sequencing (NGS)-targeted panels are powerful tools capable of identifying genetic variants, thereby providing diagnostic and therapeutic solutions.1Lee H. Martinez-Agosto J.A. Rexach J. Fogel B.L. Next generation sequencing in clinical diagnosis.Lancet Neurol. 2019; 18: 426Abstract Full Text Full Text PDF PubMed Scopus (6) Google Scholar High-depth sequencing of genomic regions facilitates detecting variants at low allele fractions (<5%) such as subclonal genomic variants in cancer.2Koboldt D.C. Best practices for variant calling in clinical sequencing.Genome Med. 2020; 12: 91Crossref PubMed Scopus (93) Google Scholar Individual sample barcodes introduced during library preparation (indexing) permit the processing and analysis of many samples per run, in turn leveraging the efficiency of current high-throughput modern NGS platforms; however, with multiplex processing methods, the potential for sample cross-contamination is an ongoing concern. Sample contamination events can occur with manual techniques, as well as in the setting of automated systems (eg, drips or aerosols during liquid handling steps and general protocol splashes from plate seals being removed).3Fiévet A. Bernard V. Tenreiro H. Dehainault C. Girard E. Deshaies V. Hupe P. Delattre O. Stern M.-H. Stoppa-Lyonnet D. Golmard Lisa Houdayer Claude ART-DeCo: easy tool for detection and characterization of cross-contamination of DNA samples in diagnostic next-generation sequencing analysis.Eur J Hum Genet. 2019; 27: 792-800Crossref PubMed Scopus (10) Google Scholar Contamination events can significantly and adversely complicate interpretation of NGS results, especially for oncology panels in which subclonal alterations of clinical significance are commonly encountered; this in turn can result in reporting of variants falsely derived from the contaminating samples or, in some cases, skewing of allele frequencies resulting in a dropout of reportable calls. Hence, detecting contamination is a crucial quality control measure in clinical NGS laboratories. Sample integrity can be ascertained by using different methods, including initial and postsequencing single nucleotide polymorphism (SNP) or short tandem repeat sample profiling (“fingerprinting”),4Anslinger K. Graw M. Bayer B. Deconvolution of blood-blood mixtures using DEPArray[TM] separated single cell STR profiling.Rechtsmedizin. 2019; 29: 30-40Crossref Scopus (16) Google Scholar or via introduction of exogenous sample-specific tags before library preparation; however, these methods add further cost and complexity in routine NGS testing. Postsequencing sample intrinsic approaches such as SNP analysis of unexpected non-diploid allelic fractions have been shown to be useful in identifying sample cross-contamination.3Fiévet A. Bernard V. Tenreiro H. Dehainault C. Girard E. Deshaies V. Hupe P. Delattre O. Stern M.-H. Stoppa-Lyonnet D. Golmard Lisa Houdayer Claude ART-DeCo: easy tool for detection and characterization of cross-contamination of DNA samples in diagnostic next-generation sequencing analysis.Eur J Hum Genet. 2019; 27: 792-800Crossref PubMed Scopus (10) Google Scholar Bioinformatics tools for assessing sample-level SNP distributions can be efficacious for identifying potential contamination events but generally require a large enough survey of informative SNPs to provide confidence, and this condition may not be adequately satisfied in targeted smaller NGS panels. More recently, multi-allelic SNPs in microhaplotype (MH) sites have been proven to be useful in sample mixture deconvolution.5Kidd K.K. Pakstis A.J. Speed W.C. Lagace R. Chang J. Wootton S. Ihuegbu N. Microhaplotype loci are a powerful new type of forensic marker.Forensic Sci Int Genet Suppl Ser. 2013; 4: e123-e124Abstract Full Text Full Text PDF Scopus (69) Google Scholar, 6Kidd K.K. Pakstis A.J. Speed W.C. Lagace R. Wootton S. Chang J. Selecting microhaplotypes optimized for different purposes.Electrophoresis. 2018; 39: 2815-2823Crossref PubMed Scopus (29) Google Scholar, 7Oldoni F. Kidd K.K. Podini D. Microhaplotypes in forensic genetics.Forensic Sci Int Genet. 2019; 38: 54-69Abstract Full Text Full Text PDF PubMed Scopus (90) Google Scholar MHs are naturally occurring short genomic DNA segments (typically <300 bp) that are characterized by the presence of two or more closely linked SNPs that present three or more allelic combinations or haplotypes.7Oldoni F. Kidd K.K. Podini D. Microhaplotypes in forensic genetics.Forensic Sci Int Genet. 2019; 38: 54-69Abstract Full Text Full Text PDF PubMed Scopus (90) Google Scholar The short distance between SNPs in these loci imparts a low genetic recombination rate. The level of heterozygosity of MH-associated SNPs is dependent on several factors, including historical accumulation of allelic variants at various positions within the targeted region, incidence of rare crossover events, occurrence of random genetic drift, and/or selection. Because MH regions possess multi-SNP haplotypes, they provide a richer degree of genotyping information than a singleton SNP marker. Also, MH SNPs are prevalent across various populations in the gnomAD database, as illustrated in Supplemental Figure S1.8Karczewski K.J. Francioli L.C. Tiao G. Cummings B.B. Alföldi J. Wang Q. et al.The mutational constraint spectrum quantified from variation in 141,456 humans.Nature. 2020; 581: 434-443Crossref PubMed Scopus (3859) Google Scholar Given the impressive ability to distinguish individuals in forensic and population ancestry applications,5Kidd K.K. Pakstis A.J. Speed W.C. Lagace R. Chang J. Wootton S. Ihuegbu N. Microhaplotype loci are a powerful new type of forensic marker.Forensic Sci Int Genet Suppl Ser. 2013; 4: e123-e124Abstract Full Text Full Text PDF Scopus (69) Google Scholar, 6Kidd K.K. Pakstis A.J. Speed W.C. Lagace R. Wootton S. Chang J. Selecting microhaplotypes optimized for different purposes.Electrophoresis. 2018; 39: 2815-2823Crossref PubMed Scopus (29) Google Scholar, 7Oldoni F. Kidd K.K. Podini D. Microhaplotypes in forensic genetics.Forensic Sci Int Genet. 2019; 38: 54-69Abstract Full Text Full Text PDF PubMed Scopus (90) Google Scholar we hypothesized that MH would represent an alternative biologically intrinsic approach to sample cross-contamination detection in a high-volume oncology NGS clinical testing environment. Using a targeted NGS myeloid neoplasm panel design that incorporates 27 MH regions, we introduce a novel method of identifying cross-contamination using the heterozygosity profiles of SNPs within and across the MH regions. The allelic frequencies of the multi-SNP haplotypes, the number of genotypes in the MH regions, and the contamination estimation score from verifyBamID9Jun G. Flickinger M. Hetrick K.N. Romm J.M. Doheny K.F. Abecasis G.R. Boehnke M. Kang H.M. Detecting and estimating contamination of human DNA samples in sequencing and array-based genotype data.Am J Hum Genet. 2012; 91: 839-848Abstract Full Text Full Text PDF PubMed Scopus (255) Google Scholar all provide signals capable of distinguishing contaminated samples from noncontaminated samples. We thus report on the training, validation, and testing of a state-of-the-art binary classifier for contamination identification. This model was built by using a heterogeneous cohort of noncontaminated patient samples, in silico contaminated sample mixtures, and gold standard laboratory-generated or therapy-related true sample admixtures. A targeted NGS panel for evaluation of hematologic neoplasms, specifically of myeloid origin (eg, acute myeloid leukemia, myelodysplastic syndrome, myeloproliferative neoplasm, myelodysplastic/myeloproliferative neoplasm), was utilized. This assay targets 47 genes, and the total target size is 95 kb. This test is run at the time of diagnosis or disease relapse to facilitate diagnostic classification and provide prognostic or therapeutic information for clinical management. In parallel, a non-Hodgkin lymphoma panel for evaluation of mature lymphoid neoplasms was also employed; this test is likewise run at the time of diagnosis or disease relapse to provide diagnostic, prognostic, or therapeutic information for clinical management. This assay targets 77 genes, and the total target size is 178 kb. Probe pools for each panel were designed through the xGen Custom Hybridization Capture Panel service (Integrated DNA Technologies, Coralville, IA), and samples were sequenced by using the NovaSeq 6000 and MiSeq instruments (Illumina, San Diego, CA) in 151 bp paired-end sequencing mode. In addition to designing hybridization probes in the relevant target genomic regions for these panels, probes were added to capture 27 MH sites (Supplemental Table S1) to help evaluate for contamination. The list of 27 MH loci was initially selected empirically and subsequently modified to include optimum sites.6Kidd K.K. Pakstis A.J. Speed W.C. Lagace R. Wootton S. Chang J. Selecting microhaplotypes optimized for different purposes.Electrophoresis. 2018; 39: 2815-2823Crossref PubMed Scopus (29) Google Scholar The MH loci were chosen to emphasize high effective allele values, which correlate with the ability to positively distinguish among individual genotypes in sample mixtures (random match probability). Additional selection criteria included maximum locus lengths <250 bp to accommodate efficient probe coverage, distribution across genomic regions, and the presence of >2 tandem SNP alleles in the majority of MH loci. Ultimately, 92 SNPs in total were targeted across the 27 MH sites. On a per-sample basis, the sequenced reads were aligned to GRCh37 (hg19) reference genome using bwa-mem with default settings from Sentieon (San Jose, CA) tools (version 0.7.17-r1188). Genomic variants were called in MH sites by using TNhaplotyper2 (sentieon-genomics-202112) with default settings from Sentieon tools.10Freed D. Aldana R. Weber J.A. Edwards J.S. The Sentieon Genomics Tools—a fast and accurate solution to variant calling from next-generation sequence data.bioRxiv. 2017; ([Preprint] doi: 10.1101/115717)Google Scholar,11Li H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM.arXiv. 2013; ([Preprint] doi: 10.48550/arXiv.1303.3997)Google Scholar Samples sequenced utilizing both targeted NGS panel designs were used for the analyses. The primary specimen source for the lymphoma panel was formalin-fixed, paraffin-embedded tissue; fresh bone marrow or peripheral blood specimens were the sources for the myeloid panel. Because the primary specimen source for each panel is different, samples sequenced across both panels allowed the creation of a model built from DNA that is both challenging to work with due to issues such as fragmentation and cross-linking (formalin-fixed, paraffin-embedded tissue) and higher quality DNA that is ideal to work with because long strands of DNA at ample concentration are easily obtainable. A cohort of 244 normal (noncontaminated) samples, 127 laboratory mixtures (101 contaminated via dilution and 26 chimeric post–stem cell transplant samples), and 476 in silico contaminated samples were included in the study. Figure 1 shows the sample selection for training, validation, and testing of the binary classification model for detecting contamination. Single contaminant and multiple sample contaminant scenarios were used in the creation of in silico contaminated samples as described in Supplemental Table S2. Methods for in silico contamination are also detailed in Supplemental Table S2. In addition, the experiment was designed with laboratory mix samples that were quantitated, normalized, and mixed at various levels/percentages of contamination. Supplemental Table S2 and Supplemental Figure S2 outline the sample mixes for the simulated and laboratory mix contaminated samples, respectively. A cross-contamination event introduces drift in the allelic frequencies of homozygous and heterozygous SNPs. Figure 2 explains the effect of a contamination event on a sample, with outcomes including reduction of allelic frequency of heterozygous and homozygous SNPs and/or introduction of a contaminant allele that was not present in the sample. Distribution of allelic frequencies of MH SNPs in 107 normal samples from the training data set was assessed to understand the pattern for heterozygous and homozygous variant SNPs. Variants with allele frequencies <3% (approximate for the designed NGS were to calls. was that the heterozygous MH SNPs were a distribution to allelic and the homozygous MH SNPs a at Figure shows the SNP allelic that are from noncontaminated samples. were heterozygous and homozygous allele frequency this approach the allele frequency of individual SNP regions and a high score to 1 the allele frequency is with a genome or a low score to is The were as shown in Figure the MH SNPs the normal samples distribution (noncontaminated) have high for this and MH SNPs from contaminated samples have to as in Figure MH SNP allele frequency was by the of the score from the across all SNPs in the MH resulting in per MH used was the number of total combinations of genotypes that can occur at MH This for the number of alleles in each MH and the is to have a higher for contaminated samples. added was the contamination score from the as a distribution of contamination to the two Figure shows a distribution of across the normal and contaminated samples, the of these a model to contamination In a clinical laboratory an model clinical and to understand the resulting to and the to report a or as a quality control M. M. P. and of an for of Med. 4: PubMed Scopus Google Scholar Hence, are popular in the as they are easily to and perform E. P. M. and of a model for of 2018; PubMed Scopus Google Scholar these an model was G. A. V. B. O. M. P. R. V. J. A. D. M. M. E. in 12: to the data; this model the of standard but on the to prevent and H. and selection via the B. Scopus Google Scholar was by using S. M. a next-generation of the on for Scopus Google Scholar (version for the model on The in the that of was by using and a of was to the The to the model include and for To with of the a was on a per-sample basis, which explains the that in the classification of the of Figure an of such a B. M. J. for the of during 2018; PubMed Scopus Google Scholar The help with the in of sample testing quality control and the to a clinical run and patient sample MICon (Microhaplotype Contamination was popular methods such as and and A. Bernard V. Tenreiro H. Dehainault C. Girard E. Deshaies V. Hupe P. Delattre O. Stern M.-H. Stoppa-Lyonnet D. Golmard Lisa Houdayer Claude ART-DeCo: easy tool for detection and characterization of cross-contamination of DNA samples in diagnostic next-generation sequencing analysis.Eur J Hum Genet. 2019; 27: 792-800Crossref PubMed Scopus (10) Google G. Flickinger M. Hetrick K.N. Romm J.M. Doheny K.F. Abecasis G.R. Boehnke M. Kang H.M. Detecting and estimating contamination of human DNA samples in sequencing and array-based genotype data.Am J Hum Genet. 2012; 91: 839-848Abstract Full Text Full Text PDF PubMed Scopus (255) Google K. A. E. M. G. estimating cross-contamination of human samples in next-generation sequencing 27: PubMed Scopus Google Scholar MICon a classification for the samples with and the tools provide a of The from the individual tools with the true were used for the area under the receiver-operating characteristic curve using (version and the tool is with the The used for individual with the tool are detailed in the MICon at under the model was on the and were to a maximum of in the data The model was used to the contamination in the hold-out test cohort of 210 samples normal samples and contaminated samples), an of of to of to and of to The for the test set is also Figure the for samples that were as normal and of the model and in The as in the training data set per is in Figure A approach to model of the on Scholar there are MH sites that are more informative of presence of contamination with MH sites. This can be to the presence of alleles in MH sites across the training data set and the Figure the of MICon and in MICon was as the by among the of the 210 test samples variant candidates for of contamination for the tool and a score of for all these they were thus as normal samples. This the score for analysis on the that the from MICon were all to of the low contamination level in the laboratory and selected Supplemental Figure the and the for with the that MICon is a accurate NGS contamination detection tool that methods to sample contamination at various from to sample per contamination level are illustrated in Supplemental Figure In addition, MICon easy interpretation of the classification through as in Figure and Supplemental Figure The and the data used are in the under the clinical variants in NGS testing and M. S. S. A. and for the interpretation and reporting of sequence variants in a of the for of and of 2017; Full Text Full Text PDF PubMed Scopus Google Scholar The presence of sample cross-contamination result in clinical and patient especially reporting variants in oncology A. Bernard V. Tenreiro H. Dehainault C. Girard E. Deshaies V. Hupe P. Delattre O. Stern M.-H. Stoppa-Lyonnet D. Golmard Lisa Houdayer Claude ART-DeCo: easy tool for detection and characterization of cross-contamination of DNA samples in diagnostic next-generation sequencing analysis.Eur J Hum Genet. 2019; 27: 792-800Crossref PubMed Scopus (10) Google Scholar detection of sample cross-contamination is thus critical to samples and repeat and the general integrity of clinical MH sites have been proven to be powerful in and forensic K.K. Pakstis A.J. Speed W.C. Lagace R. Chang J. Wootton S. Ihuegbu N. Microhaplotype loci are a powerful new type of forensic marker.Forensic Sci Int Genet Suppl Ser. 2013; 4: e123-e124Abstract Full Text Full Text PDF Scopus (69) Google F. Kidd K.K. Podini D. Microhaplotypes in forensic genetics.Forensic Sci Int Genet. 2019; 38: 54-69Abstract Full Text Full Text PDF PubMed Scopus (90) Google A.J. R. Kidd Kidd K.K. as informative SNPs and ancestry J Hum Genet. 2012; PubMed Scopus Google Scholar and for contamination detection in targeted NGS panels as a further novel MHs are easily as single with standard target regions in NGS panel design with added The MICon approach described shows the of genomic target regions and MH sites for the detection of clinical genetic variants and cross-contamination in individual clinical samples. In addition, we developed a state-of-the-art model to contamination that using thereby a clinical MICon high with and popular tools for detecting contamination. The of popular contamination methods in the is that they of detection in smaller targeted NGS panels. methods are target and this is challenging because smaller targeted panels fewer variants, for estimating contamination. The MICon method profiling of enough informative SNPs by the MH sites as for a classification thereby providing the to contamination. to MICon are including of of MH loci to SNP allele and for as more samples are A potential in using method is the for designing the NGS or probe capture chemistry to target the MH sites in Supplemental Table panels at these sites and thus the variant to contamination. Custom design approaches or are in these of these MH regions can quality and can be an to clinical NGS a approach and set of MH for contamination detection standard target in targeted NGS however, the approach described is generally to high-volume NGS testing in which contamination a concern. MH and analysis using MICon are also of potential in assessing in cell transplant being for myeloid disease by In MICon a and efficient method to contamination in a targeted NGS panel of these or MH regions in targeted NGS panels and the of a model approach as detailed in this facilitate contamination identification in NGS and data quality and We A. for useful for Supplemental Figure of samples mixed at various of contamination in laboratory to contamination detection The various of and the number of samples mixed at the of Figure Figure the for a contaminated sample as normal normal and contaminated This sample a for classification as normal at which the is at is also that this sample was contaminated at and three samples were mixed sample at at and the at to the low of contamination from multiple samples, is that the contamination not the microhaplotype sites and not a pattern of contamination. can from the that the is Hence, classification with the is that this was a and the sample not of contamination. This sample a for classification as normal at which the is at is also that this sample was contaminated at to the low of is that the contamination not the microhaplotype sites and not pattern of contamination. can from the that the is classification with is that this was a and the sample not of contamination. This sample a of and the for classification as contaminated sample was which the is at and was thus a can be via the that this as Hence, the classification with the is that this was a Figure with Supplemental Table with Supplemental Table S2

Topics & Concepts