Mapping molecular HLA typing data to UNOS antigen equivalents for improved virtual crossmatch

Background Virtual crossmatch utilizes HLA typing and antibody screen assay data as a part of organ offers in deceased donor allocation systems. Histocompatibility labs must convert molecular HLA typings to antigen equivalencies for entry into the United Network for Organ Sharing (UNOS) UNet system. While an Organ Procurement and Transplantation Network (OPTN) policy document provides general guidelines for conversion, the process is complex because no antigen mapping table is available. We present a UNOS antigen equivalency table for all IMGT/HLA alleles at the A, B, C, DRB1, DRB3/4/5, DQA1, and DQB1 loci. Methods An automated script was developed to generate a UNOS antigen equivalency table. Data sources used in the conversion algorithm included the World Marrow Donor Association(WMDA) antigen table, the HLA Dictionary, and UNOS-provided tables. To validate antigen mappings, we converted National Marrow Donor Program (NMDP) high resolution allele frequencies to antigen equivalents and compared with the UNOS Calculated Panel Reactive Antibodies (CPRA) reference panel. Results Normalized frequency similarity scores between independent NMDP and UNOS panels for 4 US population categories (Caucasian, Hispanic, African American and Asian/Pacific Islander) ranged from 0.85 to 0.97, indicating correct antigen mapping. An open source web application (ALLele to ANtigen (“ALLAN”)) and web services were also developed to map unambiguous and ambiguous HLA typing data to UNOS antigen equivalents based on NMDP population-specific allele frequencies (http://www.transplanttoolbox.org). Conclusions This tool sets a foundation for using molecular HLA typing to compute the virtual crossmatch and may aid in reducing typing discrepancies in UNet.


Introduction
Human leukocyte antigen (HLA) specificities were originally distinguished by allosera from multiparous women or individuals with multiple blood transfusions using cell-based testing strategies. HLA typing is currently performed by more accurate and specific molecular methods that assay nucleotide sequences. The IMGT/HLA database was developed to catalog the growing number of known HLA gene sequences 1 . While the naming of HLA sequences has its roots in the antigen nomenclature from serologic typing 2 , new HLA alleles are named based on sequence similarity to known alleles.
The United Network for Organ Sharing (UNOS) organ allocation system UNet utilizes donor and recipient HLA typing information and a list of unacceptable HLA antigens determined from antibody assays to perform virtual crossmatch as part of organ offers 3,4 . While the more specific IMGT/HLA allele nomenclature is used to interpret DNA-based typing data, only serologic antigen equivalents are accepted by the UNet 5 .
Antigen groups are used to represent unacceptable HLA specificities because a single allo-HLA antibody may react similarly against several distinct HLA alleles within the same antigen group.
Histocompatibility labs face a data management challenge in mapping molecular-level HLA data to UNOS antigen equivalents for entry into UNet. While the Organ Procurement and Transplantation Network (OPTN) published a policy document in 2003 with general guidelines for mapping molecular data to antigen equivalents 6 , there is a need to automate the mapping process with a standardized mapping table that includes 8 every IMGT/HLA allele. UNOS deemed it infeasible to maintain such a table at the time.
Because of the difficulty in synthesizing antigen equivalency information from multiple reference data sources, there is likely to be some variability in reporting of HLA typing results into UNOS systems, even among commercial typing platforms.
A rapid turnaround for HLA typing is necessary for limiting cold ischemic time, therefore deceased donors are typed by sequence-specific primer (SSP), sequence-specific oligonucleotide probe (SSOP), or real-time polymerase chain reaction (PCR) based methods rather than next-generation sequencing (NGS). These methods rarely achieve allele-level specificity. Hence, there is also a need to interpret ambiguous HLA typings in the context of population HLA frequencies, as OPTN guidelines call for reporting UNOS antigens based on the most common HLA allele.
To meet these challenges in HLA data reporting for virtual crossmatch, we were inspired by matching systems for hematopoietic stem cell transplantation (HSCT) that have the capability to represent HLA typing data in IMGT/HLA allele nomenclature and match to donors whose HLA is represented in antigen nomenclature 7

UNOS/OPTN Policy Documents for "Interpretation of HLA Typing Results for
Entry into UNet": We have summarized the rules for histocompatibility labs to assign UNOS antigens for IMGT/HLA alleles in Table 1. These rules were developed based on interpretation of guidelines published in 2003 in an OPTN policy document 6 . The OPTN documents 5,6 contain tables of UNOS antigen equivalencies for some IMGT/HLA alleles that differ from WHO-assigned antigens. These tables include some 4-digit antigens representing alleles that may be readily identified by low resolution molecular testing as well as antigens that were not recognized by WHO serologic nomenclature at the time 1 0 of policy implementation. The UNOS Histocompatibility Committee issued updates in 2017 that specified antigen assignments for some additional IMGT/HLA alleles 10 .

HLA Dictionary 2008:
To standardize HLA antigen assignments for stem cell donor registries, the WMDA maintains an HLA Dictionary 11 . The 2008 version of the HLA Dictionary describes serologic equivalents of alleles for 6 loci (A, B, C, DRB1, DRB3/4/5, and DQB1). The data sources used to compile the dictionary include the following: WHO serological assignments as described in the Nomenclature report of 2004 12 , data submitted by individual laboratories to WHO committee for factors of HLA system, the international cell exchange program that has 200 laboratories participating in evaluating the standardized sera, NMDP-assigned serological equivalents, and data submitted for 13 th International HLA and Immunogenetics Workshop (IHIWS). Besides these sources that assigned serotypes based on the testing of cells with sera, the dictionary also lists antigens that were computationally assigned by a neural network 13 .
Final consensus antigen assignments are listed as an expert-assigned type. IMGT/HLA alleles have WHO antigen assignments with unambiguous serology. WHO assignments were either made by the nomenclature committee when the allele sequence was submitted or were based on serologic testing data from the HLA Dictionary 11 . Possible serological antigens are given when there is no serologic testing data and are separated by a slash ("/") when there is more than one possibility.
Assumed antigens are typically based on first field of the IMGT/HLA allele name. These frequencies have been derived from 6.59 million volunteer donors from the Be The Match registry for HSCT. The NMDP US frequencies also include the Native American broad race and 21 detailed race/ethnic subcategories that could potentially be selected to make more precise antigen assignments.   15 or as NMDP multiple allele codes (MACs) 16 ). For ambiguous HLA typings, a US population must be selected, as the most probable antigens are calculated given the population-specific allele frequencies and the antigen mappings for the alleles. An upto-date list of MACs is accessed via the NMDP MAC web service (https://hml.nmdp.org/mac/). A function for reverse mapping of an unacceptable UNOS antigen to a list of IMGT/HLA alleles is also provided. DQA1 antigen mapping for ambiguous typing is not included in this release, as interpretation requires populationspecific allele frequencies that are under development by NMDP.

Command Line Tool, Web Application, and Web Services:
We developed a web application named ALLele to ANtigen ("ALLAN") for users to enter molecular HLA typings and perform UNOS antigen conversion. The tool is available at http://www.transplanttoolbox.org. This web tool was implemented using the Django web framework 17 . As a programming interface, Representational State Transfer (RESTful) 18 enabled web services available at http://www.transplanttoolbox.org/tool_services were 1 3 developed using Django REST Framework 19 . An example Python script that illustrates how to access the web services is also provided in our GitHub source code repository (https://github.com/lgragert/hla-who-to-unos). For advanced users who wish to run the conversion tool locally, a Python command line tool and pip installable package "transplanttoolbox-allan" are also available (https://pypi.org/project/transplanttoolboxallan/).

Antigen Frequencies and Similarity Index Calculation
Reference antigen frequencies for 4 broad US race/ethnic categories (CAU, AFA, HIS, and API) were generated by mapping NMDP haplotype frequencies to UNOS antigen equivalencies. To assess accuracy in antigen assignments we used a normalized similarity index, "I F " 20 , to compare the converted NMDP antigen frequencies to an independent antigen-level frequency dataset of previous UNOS donors 21 that is currently used operationally as the reference panel for measuring CPRA values.

Guidelines
An algorithm was designed to map IMGT/HLA alleles to antigens following the UNOS guidelines. Our interpretation of the rule precedence for antigen mapping guidelines is shown as a schematic diagram in Figure 1. The antigen assignments found in tables from UNOS histocompatibility committee updates, UNOS antigen equivalency tables for specific alleles (Table 1: Suggested UNOS serology equivalents for molecular types 1 4 (2003)) 6 , and 4-digit antigens defined by UNOS were given the highest precedence 5 .
For the next level of precedence, WHO-assigned antigens from the WMDA file were assigned (see Rule 3 in Table 1) followed by antigens from HLA Dictionary 2008 11 . No antigens are assigned for DQA1 locus alleles in the WMDA file, however, UNOS has added 2 nd -field allele digits as DQA1 antigen equivalencies in their latest update (e.g. DQA01:01 for DQA1*01:01:01:01).
If antigens could not be assigned from the data sources listed above, then possible and assumed antigens from the WMDA file were used. Cases where we fell back to assigning 2-digit allele equivalencies when information from all other data sources was lacking were mostly restricted to B and C locus alleles. When in doubt, we erred on the side of assigning a broader antigen category rather than assuming a split antigen.
A complete conversion

Precedence Order when Multiple Rules for Assigning Antigen Specificity may Apply
Multiple guidelines for antigen assignments may apply to a single IMGT/HLA allele, therefore an order of precedence for these guidelines must be set. When the UNOS 1 5 antigen assignment listed in the UNOS-provided  Table 2.

Most Probable Antigen for Ambiguous HLA Typing
As per OPTN guidelines, for a group of possible alleles in an ambiguous HLA typing, the serologic equivalent of the most common antigen should be entered. We apply reference population-specific allele frequencies from the NMDP population categories to make the assignment. We show that for a particular ambiguous HLA typing, the most probable antigen for the B locus may vary depending on the broad race/ethnic category selected (Table 3).

"ALLAN": IMGT/HLA ALLele to UNOS ANtigen conversion tool
The conversion functions described above are available via the ALLAN web tool at http://www.transplanttoolbox.org. When the HLA typing is unambiguous, users may select to enter either a single allele or a list of alleles. When the HLA typing contains typing ambiguity, users may enter the typing either in GL string or MAC format ( Table   1  6 3). The output shows the UNOS antigen assignments and Bw4/6 epitopes. Entry of ambiguous HLA typings requires the selection of a US race/ethnic group and returns the probability distribution of all the possible antigens considering the allele frequencies in the selected race/ethnic group. A complete user guide for the tool is available on the website. For advanced users, we also developed a Python script as a command line tool with similar functionality, which can be accessed via our GitHub repository (https://github.com/lgragert//hla-who-to-unos). We have also developed web services as a programming interface for researchers and laboratory information management systems. Client applications such as HLA typing software platforms or other lab information systems can send HLA typing and race/ethnicity information to the service, and UNOS antigens (including probabilities) are returned. The endpoint URLs and commands for these services are shown in Table 4.

Antigen Mappings in Commercially Available Software Programs
Most histocompatibility labs rely on commercial software to analyze HLA typing results and perform antigen conversion. Currently, the software platforms from leading vendors in this domain, i.e. One Lambda (HLA Fusion) and Immucor (LIFECODES ® MATCH IT! ® ), perform antigen mapping for a limited number of IMGT/HLA alleles but would benefit from a complete standardized table. "HLA Fusion" does mapping for 4,927 alleles and "MATCH IT!" does mapping for 4,439 of the 16,886 alleles. Compliance with IMGT database updates for new releases of HLA alleles and UNOS updates represents a major maintenance burden for these software vendors. Vendors must rely on available sources such as the HLA Dictionary, which has not been updated since 2008. 1 7 The WMDA does provide quarterly updates, but again vendors are left with choosing between "Unambiguous associations," "Possible associations," "Assumed associations," and "Expert assigned exceptions," and have to reconcile these options with customer specific requests. As a result, some antigen mappings made by these software packages differ from those recommended by UNOS. A list of discrepancies from apparent UNOS guidelines, apart from the missing mappings, is shown in Table 5. The commercial software vendors are in agreement that the issues arising from independent maintenance and interpretation of antigen conversion rules will be alleviated by having a consensus set of mappings.

Reference Antigen Frequencies for Six HLA Loci in 4 Broad US Races
To validate the antigen assignments, we converted published NMDP high resolution haplotype frequencies 14 into antigen equivalents and compared antigen frequencies at each HLA locus to the UNOS CPRA HLA reference panel. Antigen-level NMDP frequency tables for 4 US broad race/ethnic categories are available as Supplementary   Table 3. The normalized "I F " similarity index for these races ranged from 0.85 to 0.97, indicating accuracy in antigen assignments (Table 6). One cause of disparity between panels was that the UNOS DQ typing data was more often reported as broad DQ1 antigens rather than split DQ5 and DQ6 antigens. Meanwhile, the only high resolution IMGT/HLA alleles in the NMDP panel that mapped to broad DQ1 were the relatively uncommon alleles DQB1*06:11 and DQB1*06:12.

Discussion
The OPTN guidelines for HLA allele to antigen mapping require that histocompatibility labs refer to several different data sources which as we have shown here can lead to differences in antigen assignment. To streamline this process, we present a complete mapping table between all current IMGT/HLA alleles and UNOS antigens that was developed based on published OPTN guidelines. This conversion table report gives a standardized source for antigen mapping that can be updated in tandem with every new IMGT/HLA database release.
Our web tool, ALLAN, offers a convenient mapping of HLA typing data to UNOS antigens for entry into UNet, including resolving HLA typing ambiguities in antigen assignments as shown in Table 3. This will aid in streamlining HLA typing that is performed for deceased donors, which has limited time for experimentally resolving ambiguities. Molecular methods such as SSP may not always resolve null expression variants in time, however, our tool helps in predicting the probability of null alleles and therefore a blank antigen for a specified population.
One potential cause of unexpected positive tissue crossmatch 22   In recent updates, UNOS increased the number of 4-digit unacceptable antigens that can be reported. Because many of these 4-digit antigens are not included in the UNOS antigen frequencies for CPRA 26  This highlights the need for an improved curation process that would lead to the availability of public Bw4 and Bw6 epitope assignment resources.
Few low (L) or questionable (Q) expression alleles were assigned antigens in the HLA dictionary and we have assigned antigens to all these alleles as if they were expressed.
While L alleles have been confirmed to have weak surface expression that can elicit alloimmunity, little information is available for Q alleles 27 . The current version of IMGT/HLA database includes 75 Q alleles, yet only 8 of these have been observed unambiguously in the NMDP registry donor file. Further examination of mRNA transcripts, cell surface expression, and tests of alloimmunity are needed to make definitive assignments for these alleles, but we prefer to err on the side of caution. The 2 1 assigning of antigens to L/Q alleles intends to avoid cases of false negative virtual crossmatch at a potential cost of increasing positive virtual crossmatch predictions.
OPTN guidelines do not call for histocompatibility laboratories to analyze HLA sequences. Therefore, decisions on antigen assignments are best maintained in the reference data sources such as the IMGT/HLA database, and the HLA dictionary. However, both WHO and the HLA dictionary call B*14:03 a B14. As another example, in a recent report 30 these authors describe the serological specificity of B*15:33 as a B62.
OPTN guidelines call for the WHO assignment, which is B15 31 . If the allele lacked a WHO assignment, the 2008 HLA dictionary would be used to assign a B62. For the UNOS assignment to become B62, the WHO antigen would need to change or UNOS would need to specify an exception to the rule. In summary, these complex cases cannot all be sorted out here. We believe such decisions are best left to UNOS, WHO, or the next HLA Dictionary.
Standardized antigen assignments for molecular typing are necessary for the UNOS organ allocation system to evolve to incorporate DNA-based HLA assignments and epitope-based matching strategies 32 . We anticipate future updates to our tool to support changes in the UNOS CPRA to include antigens for the DQA1, DPA1, and DPB1 loci. 2 2 These loci together account for >60% of the unacceptable antigens in candidates with CPRA > 99% 33,34 . We intend to develop additional tools that would allow the NMDP high resolution frequency panel to be used for allele-level CPRA calculations incorporating these additional loci. Imputation tools based on NMDP frequency data are also useful for estimating amino acid assignments from antigen-level data to perform epitope analysis with tools such as PIRCHE: predicted indirectly recognizable HLA epitopes 35 .
The antigen mapping tools we developed here will take UNOS one step closer to their long-term goal of maximally utilizing HLA molecular typing data in solid organ allocation systems.   *Rule notation is used in Figure 1 as well as supplementary Table 1 for easy review 3 2       Highest precedence for antigen assignment to IMGT/HLA alleles was given to UNOS resources, followed by WHO antigens, the HLA dictionary, and the WMDA antigen file. If an antigen could not be assigned to an allele from these antigen equivalency reference sources, then the first-field of the IMGT/HLA allele name was assigned as the antigen.