Substituting imputation of HLA antigens for high-resolution HLA typing: Evaluation of a multiethnic population and implications for clinical decision making in transplantation

Abstract

Molecular mismatch analysis for assessment of histocompatibility in transplantation requires high-resolution HLA typing. Algorithms to “guesstimate” high-resolution from low-resolution typing exist, but their accuracy remains unknown. We converted high-resolution, sequence-based, HLA typing of 310 subjects from an ethnically heterogeneous population to low-resolution equivalents and tested the ability of the NMDP HaploStats and HLA Matchmaker programs to impute/reproduce the measured high-resolution HLA type, using the more common “winner-takes-all” approach. Only 35.6% of the HaploStats imputed HLA-A, -B, -C, -DRB1, and -DQB1 haplotypes had no mistakes, and the accuracy was significantly lower for non-Caucasians (29.1%) compared to Caucasians (45.2%) (odds ratio [OR], 0.5; 95% confidence interval [CI], 0.3-0.8; P = .004). HLA Matchmaker was not able to provide high-resolution haplotypes for 45.2% of Caucasian subjects and 63.5% of non-Caucasian subjects (P = .002). Of those with an imputed result, only 10.3% of Caucasians and 4.8% of non-Caucasians had accurate 10-allele high-resolution output. Eplet analysis revealed additional, inaccurate eplets in 37% of individuals, with 22.5% showing at least 2 additional, inaccurate eplets; incorrect eplets were more common among non-Caucasians (OR, 1.8; 95% CI, 1.1-2.9; P = .018). Given this high error rate, caution should be taken before using imputation tools for clinical or research purposes, especially for non-Caucasian individuals. Molecular mismatch analysis for assessment of histocompatibility in transplantation requires high-resolution HLA typing. Algorithms to “guesstimate” high-resolution from low-resolution typing exist, but their accuracy remains unknown. We converted high-resolution, sequence-based, HLA typing of 310 subjects from an ethnically heterogeneous population to low-resolution equivalents and tested the ability of the NMDP HaploStats and HLA Matchmaker programs to impute/reproduce the measured high-resolution HLA type, using the more common “winner-takes-all” approach. Only 35.6% of the HaploStats imputed HLA-A, -B, -C, -DRB1, and -DQB1 haplotypes had no mistakes, and the accuracy was significantly lower for non-Caucasians (29.1%) compared to Caucasians (45.2%) (odds ratio [OR], 0.5; 95% confidence interval [CI], 0.3-0.8; P = .004). HLA Matchmaker was not able to provide high-resolution haplotypes for 45.2% of Caucasian subjects and 63.5% of non-Caucasian subjects (P = .002). Of those with an imputed result, only 10.3% of Caucasians and 4.8% of non-Caucasians had accurate 10-allele high-resolution output. Eplet analysis revealed additional, inaccurate eplets in 37% of individuals, with 22.5% showing at least 2 additional, inaccurate eplets; incorrect eplets were more common among non-Caucasians (OR, 1.8; 95% CI, 1.1-2.9; P = .018). Given this high error rate, caution should be taken before using imputation tools for clinical or research purposes, especially for non-Caucasian individuals. High-resolution human leukocyte antigen (HLA) typing is a time-consuming and expensive process that is typically reserved for hematopoietic stem cell transplantion. Solid organ transplantation currently relies on low- or intermediate-resolution antigen reporting, but recent publications on the value of molecular mismatch analysis highlight the potential role for high-resolution HLA typing in support of clinical decision-making and research in solid organ transplantation. In the absence of available high-resolution typing, investigators have resorted to imputation, focusing on 3 lines of study: (1) assessment of molecular mismatch load as a guide to stratify patients’ risk for developing de novo HLA donor-specific antibodies (DSA) and poor graft outcome; (2) investigation of “epitope matching” for organ allocation; and (3) imputation of donor alleles to determine potential presence of DSA, to guide treatment or perform virtual crossmatch (vXM). The most commonly used imputation programs are the National Marrow Donor Program’s (NMDP) HaploStats and Rene Duquesnoy’s HLA Matchmaker converter programs, both of which suggest a patient’s most likely high-resolution HLA type based on their low-resolution type and ethnicity.1Silva E Alba A Castro A et al.Evaluation of HLA Matchmaker compatibility as predictor of graft survival and presence of Anti-HLA antibodies.Transplant Proc. 2010; 42 (https://doi.org/10.1016/j.transproceed.2009.12.047): 266-269Crossref PubMed Scopus (17) Google Scholar, 2Kosmoliaptsis V Bradley JA Sharples LD et al.Predicting the immunogenicity of human leukocyte antigen class I alloantigens using structural epitope analysis determined by HLAMatchmaker.Transplantation. 2008; 85 (https://doi.org/10.1097/TP.0b013e31817441d6): 1817-1825Crossref PubMed Scopus (60) Google Scholar, 3Philogene MC Amin A Zhou S et al.Eplet mismatch analysis and allograft outcome across racially diverse groups in a pediatric transplant cohort: a single-center analysis.Pediatr Nephrol. 2020; 35 (https://doi.org/10.1007/s00467-019-04344-1): 83-94Crossref PubMed Scopus (16) Google Scholar, 4Tafulo S Malheiro J Santos S et al.Degree of HLA class II eplet mismatch load improves prediction of antibody-mediated rejection in living donor kidney transplantation.Hum Immunol. 2019; 80 (https://doi.org/10.1016/j.humimm.2019.09.010): 966-975Crossref PubMed Scopus (20) Google Scholar, 5Meneghini M Melilli E Martorell J et al.Combining sensitive crossmatch assays with donor/recipient human leukocyte antigen eplet matching predicts living-donor kidney transplant outcome.Kidney Int Reports. 2018; 3 (https://doi.org/10.1016/j.ekir.2018.03.015): 926-938Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar HaploStats uses HLA haplotype frequency reference panels for 21 different ethnicities derived from volunteer potential bone marrow donors in the Be The Match/NMDP database. This software can be used to narrow the search for a potential stem cell donor; importantly, however, all potential prospective donors must undergo subsequent high-resolution confirmatory typing before any clinical intervention. This database can be used for imputation in different ways. The most common approach used in solid organ transplantation research has been the “winner-takes-all” method: the imputed high-resolution HLA type is defined as the most frequent pair of HLA haplotypes in a racial/ethnic group that are consistent with an individual’s self-reported race and low-resolution HLA type. Other approaches include computational schemes based on Rubin’s multiple imputation approach6Rubin DB. Multiple imputation for nonresponse in surveys. John Wiley & Sons, Inc, New York1987Crossref Google Scholar whereas others have focused on estimating the upper and lower probabilities for a particular high-resolution HLA time.7Manski CF Tambur AR Gmeiner M. Predicting kidney transplant outcomes with partial knowledge of HLA mismatch.Proc Natl Acad Sci USA. 2019; 116 (https://doi.org/10.1073/pnas.1911281116): 20339-20345Crossref PubMed Scopus (10) Google Scholar In this study, we aimed to determine accuracy of the commonly used, “winner-takes-all” imputation approach in an ethnically diverse population using existing sequenced high-resolution HLA typing from hematopoietic stem cell transplant donors and recipients. We performed a cross-sectional study of all patients who had high-resolution HLA typing prior to receiving a hematopoietic stem cell transplant at Ann & Robert H Lurie Children’s Hospital of Chicago between January 1, 2006 and February 1, 2018. HLA typing on patients and their self-reported race (Caucasian, Black/African-American, Asian, American Indian/Alaskan Native) and ethnicity (Hispanic or not Hispanic) were collected. If patients reported multiple racial or ethnic categories, the first one listed was used for research purposes. All patients who reported Hispanic ethnicity were designated as Hispanic for research purposes, regardless of the reported race, because of the design of the imputation programs. High-resolution HLA typing at the HLA-A, -B, -C, -DRB1, and -DQB1 loci was obtained for all patients by sequencing using ABI PRISM® 3100 or 3130 xl Genetic Analyzers (Applied Biosystems, Foster City, CA) and analyzed by Assign SBT™ HLA Sequence Analysis Software (Conexio Genomics, Fremantle, Western Australia, Australia). Ambiguities were resolved with allele specific in-house primers and analyzed by AutoAssembler™ DNA Sequence Assembly Software (Applied Biosystems). The high-resolution HLA typing was converted to the corresponding expert assigned serologic, low-resolution HLA type according to “The HLA Dictionary 2008.”8Holdsworth R Hurley CK Marsh SGE et al.The HLA dictionary 2008: a summary of HLA-A, -B, -C, -DRB1/3/4/5, and -DQB1 alleles and their association with serologically defined HLA-A, -B, -C, -DR, and -DQ antigens.Tissue Antigens. 2009; 73 (https://doi.org/10.1111/j.1399-0039.2008.01183.x): 95-170Crossref PubMed Scopus (141) Google Scholar This low-resolution HLA typing were then entered into 2 converter programs, and the outputs were compared to the true high-resolution type reported by the clinical HLA laboratory (Figure 1). Two commonly used converter programs were evaluated in this study. The first was the NMDP’s HaploStats web application (http://www.haplostats.org). User input includes the patient’s HLA typing and 1 of 5 patient racial/ethnic groups. The application’s output includes a list of potential high-resolution genotypes ranked by probability. For the purposes of this study, 2 imputations were performed. The first entered the patient’s HLA-A, -B, and -DR low-resolution typing and requested a high-resolution HLA-A, -B, -C, -DRB1, and -DQB1 (Method 1). The second imputation used HLA-A, -B, -C, -DR, and -DQ at low resolution and requested imputation of all 5 loci (Method 2). The most probable phased genotype for the patient’s self-reported race/ethnicity was recorded as the “imputed” high-resolution HLA typing. The second converter program was the HLA Matchmaker Four-digit Allele Converter Program (v01) (www.epitopes.net/downloads.html), which generates high-resolution allele information for use with the HLA Matchmaker epitope programs. Input includes HLA-A, -B, and -DR at low resolution and 1 of 4 racial/ethnic groups. Output is the most common high-resolution HLA-A, -B, -C, -DR, and -DQ haplotypes for the patient’s race or ethnicity (Method 3). HLA Matchmaker does not recognize some commonly used expert assigned HLA serologic types, including B60-65, B75, DR17, and DR18. In order to provide the best possible results, these low-resolution types were converted to their alternate World Health Organization assigned antigens (B14, B15, and DR3) according to “The HLA Dictionary 2008.”8Holdsworth R Hurley CK Marsh SGE et al.The HLA dictionary 2008: a summary of HLA-A, -B, -C, -DRB1/3/4/5, and -DQB1 alleles and their association with serologically defined HLA-A, -B, -C, -DR, and -DQ antigens.Tissue Antigens. 2009; 73 (https://doi.org/10.1111/j.1399-0039.2008.01183.x): 95-170Crossref PubMed Scopus (141) Google Scholar Output from both programs, a total of 3 imputations, was compared to the sequenced high-resolution HLA typing. The number of incorrectly imputed alleles was reported for each HLA locus as well as overall. Results were subgrouped by self-reported race/ethnicity. Accuracy of the imputed HLA typing was compared between groups using logistic regression. To evaluate the effect of HLA imputation on eplet matching we used the HLA Matchmaker “4ABC Eplet Matching Vs02 prototype” and “5DRDQ Matching Vs2.2 August 2018” programs. Sequenced HLA information was entered as the recipient and imputed high-resolution HLA typing was entered as the donor. This provided information on the number of eplets present in the imputed typing that were not present in the sequenced typing. The “4ABC Eplet Matching Vs02 prototype” compares eplets on class I HLA molecules collectively rather than at each individual locus. This has the potential to undercount immunologically important mismatches; therefore, the program was run separately for each class I locus. Total eplet mismatch (antibody verified and unverified) numbers were used for all analyses. Patient race/ethnicity, success of imputation, and accuracy of imputation and eplet identification were reported using descriptive statistics. Accuracy of imputation was evaluated at the individual, rather than HLA loci, level to better evaluate the clinical implications of imputation. Distribution of racial ethnic groups was compared to population using a as population not groups. The of an accurate for different racial/ethnic groups were compared using logistic regression. The association between and -DQB1 imputation was compared using was used for all analyses. This study was by the Ann & Robert H Lurie Children’s Hospital of Chicago study High-resolution HLA typing and race/ethnicity were available on individuals. The patient population was Asian, and American Indian/Alaskan Two or more were self-reported in of the study population 1). The study Caucasians (P and a of (P but was to the population of the 2018. Scholar compared to the for the study a lower of (P a of (P and by race or ethnicity of study population compared to the population of the of and the for = for kidney and by Indian/Alaskan 2 or more in a We the ability of each imputation to a for the individual’s race/ethnicity. In imputed high-resolution HLA typing from low-resolution HLA-A, -B, and -DR for of individuals. In imputed high-resolution HLA typing from low-resolution HLA-A, -B, -C, -DR, and -DQ for of individuals. The program was to provide imputed typing for 1 Hispanic and 1 American Indian/Alaskan In the HLA Matchmaker program provided a high-resolution for only of were from study to of imputed typing. HLA imputation was significantly among non-Caucasian compared to Caucasians (odds ratio 95% confidence interval [CI], P = .002). HLA Matchmaker not a for the 1 American Indian/Alaskan individual, as the program does not include this race/ethnicity of imputation of a high-resolution HLA type using 3 different with by ethnicity as -B, -B, -C, -DR, was to provide high-resolution HLA typing for 1 Hispanic individual and 1 American Indian/Alaskan individual using HLA-A, -B, -C, = Indian/Alaskan HaploStats was to provide high-resolution HLA typing for 1 Hispanic individual and 1 American Indian/Alaskan individual using HLA-A, -B, -C, in a In imputation using low-resolution HLA-A, -B, and -DR typing, HaploStats imputed all high-resolution alleles in of was at 3 or more alleles in 3). an accurate allele typing was significantly more likely for Caucasians compared to non-Caucasians 95% CI, P Accuracy was poor for and of accurate imputation of high-resolution HLA-A, -B, -C, -DRB1, and alleles low-resolution HLA-A, -B, and typing in and subgrouped by self-reported HLA-A, -B, -DR to -B, -C, -DRB1, -DQB1 imputation (Method = confidence = = = = = = = in a 2 were as the program was not able to provide high-resolution low-resolution HaploStats imputed all high-resolution alleles in 35.6% of was at 3 or more alleles in of 10-allele accuracy was for Caucasians compared to non-Caucasians (29.1%) 0.5; 95% CI, 0.3-0.8; P = .004). accuracy was poor for and of accurate imputation of high-resolution HLA-A, -B, -C, -DRB1, and alleles low-resolution HLA-A, -B, and -DQ in and subgrouped by self-reported HLA-A, -B, -DQ to -B, -C, -DRB1, -DQB1 imputation (Method = confidence = = = = = = in a In the HLA Matchmaker Converter Program provided for individuals. the program imputed all alleles in and had at 3 or more alleles in of 10-allele accuracy was 10.3% for Caucasians and for a that was not 95% CI, P = In a analysis focusing only on the HLA-A, -B, and HLA Matchmaker a high-resolution HLA typing in of of accurate imputation of high-resolution HLA-A, -B, and alleles low-resolution HLA-A, -B, in HLA and subgrouped by self-reported Matchmaker HLA-A, -B, -C, -DRB1, -DQB1 imputation by race/ethnicity (Method = = = = = = = in a HaploStats imputation imputed of class I alleles and of class II alleles imputed of class I alleles and of class II HLA Matchmaker imputation was in of class I alleles and of class II alleles of accurate imputation of high-resolution HLA-A, -B, -C, -DRB1, -DQB1 low-resolution typing into HaploStats and HLA subgrouped by of HLA-A, -B, -C, -DRB1, -DQB1 imputation by incorrect = HLA-A, -B, -DR HLA-A, -B, -C, -DR, -DQ = Matchmaker = in a is to be in high with To the accuracy of based on low-resolution typing, we focused on imputation HLA-A, -B, -DR to high-resolution HLA-A, -B, -C, -DRB1, both alleles were was a at one or both alleles in of individuals. 1 allele was in of with 2 imputed and was of both alleles in of individuals. accurate imputation of both alleles was in of both alleles The association between imputation and had a of (P between accurate imputation of and accurate imputation of HLA-A, are entered into between imputation and of incorrect of incorrect in a Results from imputation which provided the most accurate results, were entered into the HLA Matchmaker eplet matching programs with the sequenced HLA type as the recipient and the HLA type as the donor of this population had at least 1 incorrectly imputed The imputed HLA typing at least 1 eplet that was not present in the sequenced typing in 37% of individuals, with and 2 or 3 eplets in the imputed was more likely to incorrect eplets among non-Caucasians compared to Caucasians 1.8; 95% CI, 1.1-2.9; P = of eplets in the imputed HLA type that are not in the sequenced HLA type, using the HaploStats HLA-A, -B, -C, -DRB1, -DQB1 of eplets in imputed HLA but not in sequenced HLA-A, -B, -C, -DRB1, -DQB1 New = = = confidence New in a In this study, we that a “winner-takes-all” imputation approach to HLA typing provided a low level of accuracy across a of imputation and this accuracy was significantly among non-Caucasian individuals. imputation output was into were more common among non-Caucasian individuals. accurate imputation at the allele not into accurate imputation at between the as the of provide and outcomes for research purposes. HLA typing in these has been recorded only for low-resolution HLA-A, -B, and -DR, which is not for molecular mismatch Given the of the outcome by the multiple have used HaploStats the HLA Matchmaker converter to low-resolution HLA typing into high E Alba A Castro A et al.Evaluation of HLA Matchmaker compatibility as predictor of graft survival and presence of Anti-HLA antibodies.Transplant Proc. 2010; 42 (https://doi.org/10.1016/j.transproceed.2009.12.047): 266-269Crossref PubMed Scopus (17) Google Scholar, 2Kosmoliaptsis V Bradley JA Sharples LD et al.Predicting the immunogenicity of human leukocyte antigen class I alloantigens using structural epitope analysis determined by HLAMatchmaker.Transplantation. 2008; 85 (https://doi.org/10.1097/TP.0b013e31817441d6): 1817-1825Crossref PubMed Scopus (60) Google Scholar, 3Philogene MC Amin A Zhou S et al.Eplet mismatch analysis and allograft outcome across racially diverse groups in a pediatric transplant cohort: a single-center analysis.Pediatr Nephrol. 2020; 35 (https://doi.org/10.1007/s00467-019-04344-1): 83-94Crossref PubMed Scopus (16) Google Scholar, 4Tafulo S Malheiro J Santos S et al.Degree of HLA class II eplet mismatch load improves prediction of antibody-mediated rejection in living donor kidney transplantation.Hum Immunol. 2019; 80 (https://doi.org/10.1016/j.humimm.2019.09.010): 966-975Crossref PubMed Scopus (20) Google Scholar, 5Meneghini M Melilli E Martorell J et al.Combining sensitive crossmatch assays with donor/recipient human leukocyte antigen eplet matching predicts living-donor kidney transplant outcome.Kidney Int Reports. 2018; 3 (https://doi.org/10.1016/j.ekir.2018.03.015): 926-938Abstract Full Text Full Text PDF PubMed Scopus (13) Google Scholar as we have this approach a level of results, showing an accuracy of 45.2% for imputation of a 10-allele HLA type among are to the accuracy reported by in an Caucasian A of an HLA genotype imputation in hematopoietic stem cell donors from 5 PubMed Scopus Google Scholar accuracy was significantly for at the HLA allele eplet identification was inaccurate in of Caucasians and of the imputed HLA type 37% of at least 1 additional, inaccurate and 22.5% had 2 or more additional, inaccurate are to by et imputation with in I eplet in of individuals, and in class II eplet in among a population that was A et in epitope using genotype imputation 2018; PubMed Scopus Google Scholar et incorrect I eplet mismatch identification in and incorrect II identification in of patients using imputation of molecular typing in a population that was = the error to for class I and for II eplet mismatch S and between eplet using and high resolution molecular human leukocyte antigen typing 2018; PubMed Scopus (10) Google Scholar A of 2 eplets in an individual donor recipient be to in a in the eplet mismatch load between the risk V et molecular a for J 2019; Full Text Full Text PDF PubMed Scopus Google Scholar eplet mismatch load is investigation by the and as an approach to in prospective with patients at risk to de novo 2018. Scholar This should not be with using eplet mismatch load in matching or for of In the in of and the that eplets should not be with and eplet mismatch load should not be with of immunogenicity AR P et in of J 2018; Full Text Full Text PDF PubMed Scopus (141) Google et in of J of Scholar We have this in a of patients who were across 2 but antibodies to only 1 of the In this of patients only the antigen with the lower eplet mismatch load developing the antigen with the mismatch AR H R The to HLA from J 2019; PubMed Scopus (20) Google Scholar are in which the recipient de novo the that the donor only a matching or eplet risk the is in the Immunol. 2018; PubMed Scopus Google Scholar study from including a number of non-Caucasian in information on HLA imputation is We that imputation is significantly more likely to in for non-Caucasian individuals. This was especially true for those as Hispanic or an that is consistent with the in and Hispanic M High-resolution HLA alleles and haplotypes in the Immunol. PubMed Scopus Google Scholar were by et showing better accuracy among Caucasians compared with using different imputation The organ in the to the best use of each organ with of Scholar In Caucasian patients were more than as likely as patients to a kidney I in PubMed Scopus Google by in organ in of transplantation among racial/ethnic S E of racial with to kidney transplant the of the kidney 2019; PubMed Scopus Google Scholar Multiple recent A et in epitope using genotype imputation 2018; PubMed Scopus Google of an in pediatric kidney PubMed Scopus Google CF V Donor in pediatric kidney transplantation using and eplet A histocompatibility PubMed Scopus Google Scholar have the use of imputation to epitope matching for of donor In study imputation in incorrect of in 37% of individuals, in the use of imputation in transplant only HLA-A, -B, and -DR have been the of mismatch between donor and has that mismatch is with graft M and kidney transplant J Nephrol. 2018; PubMed Scopus Google Scholar and antibodies are the most common type of donor specific to a kidney et antibodies to poor graft outcome Full Text Full Text PDF PubMed Scopus Google Scholar this is common to on between and that matching at -DR a at M. a for J Nephrol. 2018; PubMed Scopus Google Scholar is a between high-resolution -DR alleles and this is information is at the antigen In study, we a association between imputed and but the was We the for high-resolution typing of rather than on than This information is not only for potential but for mismatch load more to determine a particular is a have multiple imputation and computational approaches for M M et and matching for HLA 2019; 35 PubMed Scopus Google Scholar, A J et of imputation of phased genotypes from HLA Antigens. PubMed Scopus Google Scholar, A J M. high resolution HLA haplotype derived from DNA typing for the donor Immunol. PubMed Scopus Google Scholar of these on haplotype from and of across HLA haplotypes to high-resolution HLA for clinical purposes. For et imputation can be used for the study of HLA association with multiple M S et and risk of multiple PubMed Scopus (13) Google Scholar et an approach to support decision-making at the of donor high-resolution HLA typing is not available but in order to perform This prediction determine is to the organ crossmatch et for of molecular HLA typing in organ and virtual Immunol. 2020; PubMed Scopus (3) Google Scholar for organ the laboratory typically has to resolution typing that can the of imputation. This information is not available in imputation is by et multiple imputation approaches no error or error in the haplotype their output only the lower of has been that by imputation, remains a M M et and kidney allograft PubMed Scopus Google Scholar to this was by et in which both the lower and upper are the of the CF Tambur AR Gmeiner M. Predicting kidney transplant outcomes with partial knowledge of HLA mismatch.Proc Natl Acad Sci USA. 2019; 116 (https://doi.org/10.1073/pnas.1911281116): 20339-20345Crossref PubMed Scopus (10) Google Scholar is important to that this approach is to be tested A potential of this study is the of Hispanic whereas the transplant a of analysis does not include on or typing to specific for these To potential error we have used low-resolution typing as as in of organ the have to level typing The high level of error in imputation that use in clinical is likely to existing between racial/ethnic groups in to and outcomes of transplantation. use in research the of the in molecular mismatch epitope or the presence of at the allele should be using imputation as the for Other imputation should be evaluated for their especially among non-Caucasian transplant to of or high-resolution HLA typing the for imputation and the available for The to and for their with this was by the The of this have of to as by the American of Tambur is on the of and The have no of to The that support the of this study are available on from the corresponding The are not available to or

Topics & Concepts