The presence of antibodies directed at donor-specific Human Leukocyte Antigen (HLA) (donor-specific antibodies, DSA) is a major cause of antibody-mediated rejection (ABMR) and allograft loss in solid organ transplant recipients. Owing to the advances of HLA antibody identification assays and immunosuppression regimens, current histocompatibility criteria based on the avoidance of unacceptable DSAs prior to transplantation has achieved very favorable short-term kidney transplant outcomes. However, the development of de novo DSA (dnDSA) - anti-donor HLA antibodies that emerge after transplantation, still occurs in 13%-30% of kidney transplant recipients, and is strongly associated with chronic allograft dysfunction and inferior graft survival (1).

It is well established that humoral responses in transplantation are elicited against non-self-epitopes, rather than the entire donor HLA proteins. Therefore, evaluating molecular mismatches, i.e., identifying differential HLA epitopes, between transplant donors and recipients provides a more refined measure for immunological risks over the traditional assessment at the antigen level. Several B cell epitope mismatch algorithms have been developed to evaluate the disparity of amino acid compositions, structures, or physiochemical properties of donor HLA proteins. HLAMatchmaker considers eplets, patches of non-self amino acid residues within a 3.0-3.5 Å radius on the structural surface of donor HLA, as potential targets of recipient B cell humoral response (2). Eplets are considered the functional epitopes that determine the specificity and strength of antigen-antibody interaction (3). Previous studies have demonstrated the correlation of eplet mismatches, particularly at HLA Class II DR and DQ loci, with dnDSA development (4) and T cell-mediated rejection (TCMR) in kidney transplant recipients (5). A subset of eplets is designated as “verified eplets” by the HLAMatchmaker program because their predicted antibody reactivities have been confirmed by studies using human and mouse monoclonal antibodies or polyclonal antibodies from parous women and transplant recipients (6).

A recently developed B cell epitope analysis program, Snow, applies amino acid matching based on a combination of two algorithms – Snowflake, a computational pipeline that calculates solvent accessible surfaces area of individual HLA proteins’ amino acids (7), and Snowball, which correspondingly predicts local surface protrusion of amino acids (8), to provide a refined prediction of mismatched B cell epitopes in transplantation. Studies showed that HLA Class I Snowflake scores significantly correlate with the incidence of DSA in kidney transplant recipients and child-specific HLA antibodies during pregnancy (9). Additional B cell epitope mismatch assessment programs include HLA-EMMA, which determines solvent accessible amino acid mismatches on donor HLA molecules (10), and EMS-3D, which compares the differences of surface electrostatic potential between donor and recipient HLA (11).

Alongside the B cell epitope prediction programs, PIRCHE-II (Predicted Indirectly Recognizable HLA Epitopes) is an algorithm that calculates the numbers of donor HLA-derived peptides that can be presented by recipient’s HLA Class II molecules (12). PIRCHE-II scores represent the numbers of potential T cell epitopes involved in indirect CD4 T cell alloreactivity. Studies have shown that elevated PIRCHE-II scores correlate with the occurrence of dnDSA (12, 13), ABMR (14), TCMR (15, 16), and inferior allograft survival (16). As indirect CD4 T cell allorecognition provides pivotal help to B cell proliferation and differentiation in humoral response against the allograft, we theorize that assessing both T and B cell epitope mismatches would provide a more precise immunological risk stratification for transplant recipients than considering T cell and B cell epitopes separately. The goals of the current study are to investigate the independent and synergistic association of verified eplet, Snow, and PIRCHE-II scores with the incidence of de novo DSA development in a cohort of kidney transplant recipients, and to compare the correlation of the three molecular mismatch scores.

In this retrospective study, we evaluated the correlations between HLA epitope mismatches assessed by three algorithms - Snow, HLAMatchmaker, and an updated version of PIRCHE-II, with de novo DSA development in a cohort of 843 kidney transplant recipients. Our results demonstrated that B cell and T cell epitope mismatches are remarkably increased in dnDSA+ recipients, and importantly, elevated numbers of epitope mismatches are significantly predictive of dnDSA occurrence in kidney transplantation.

To our knowledge, this report is the first to evaluate the utility of the newly developed Snow algorithm for the prediction of dnDSA using local kidney transplant data. Snow takes both the solvent-accessible surface area (Snowflake (7)), and amino acid protrusion (Snowball (8)) of individual HLA alleles into consideration and provides a refined measure to identify potential antibody epitopes. In agreement with previous publication (9), we observed strong correlation between Snow scores and verified eplet mismatches in our patient cohort ( Table 2 ).

One of the strengths of our study lies in the painstaking approach to data analysis. We utilized three HLA molecular mismatch assessment algorithms and compared B cell and T cell epitope mismatches at individual HLA-A, B, C, DRB1, DQB1 loci, or combined Class I and DRB1/DQB1 loci after excluding any transplant pairs who are allele-matched at the locus/loci of analysis to avoid misrepresentation of the molecular mismatch scores. With this refined analysis scheme, we observed strong correlations of individual and combined DRB1/DQB1 epitope mismatch loads with the occurrence of DRB1 and/or DQB1 dnDSA ( Figures 2 and 3 , Supplementary Tables 1 and Supplementary Table 2 ), which is consistent with the findings demonstrated by several other groups (4, 13, 17–19). On the other hand, although the lack of clear correlation between Class I epitope mismatches and the risks for inferior kidney transplant outcomes has been presented in a number of studies (16, 18, 20, 21), our results showed that the combined HLA-A/B/C Snow, vEplet, and PIRCHE-II scores are all distinctly greater in the Class I dnDSA+ patients, compared to the dnDSA- group ( Figure 2 , Supplementary Table 1 ). Furthermore, our study assessed epitope mismatches after adjusting for allele mismatch numbers in each patient. This analysis approach avoids inaccurate interpretation of the molecular mismatch scores when there are differential numbers of allele mismatches among the study subjects. Our data demonstrated that Snow, vEplet, and PIRCHE-II scores per mismatched allele are all significantly elevated in the dnDSA+ group ( Figure 4 , Supplementary Table 3 ), verifying that B cell and T cell epitope mismatches are indeed strong contributors to the development of Class I and DRB1/DQB1 de novo DSAs independent of allele mismatch numbers.

During a typical immune response to alloantigens in transplantation, donor-specific naïve B cells are activated upon the recognition of alloantigens via their surface B cell receptor (BCR). Also through BCR, B cells endocytose donor antigens, degrade them in the lysosomes, and present donor-derived peptides in the form of MHC class II-peptide complexes on their surfaces. The primed B cells then migrate to the border of the T cell zone in the secondary lymphoid organs where they interact with CD4⁺ helper T cells that carry cognate T cell receptor (TCR) specific for the presented donor-derived peptides. This interaction leads to the activation of the helper T cells, which in turn deliver stimulation signals through CD40/CD40L binding and the release of cytokines that trigger the expansion and differentiation of B cells into antibody-secreting plasma cells. As the establishment of humoral alloimmune responses requires the interaction between B and T cells, it is imperative to take both B cell and T cell epitopes into account when evaluating the level of compatibility between transplant donors and recipients. Supporting this notion, our results distinctly showed that the subgroup of patients who score high by both T cell (PIRCHE-II) and B cell (Snow or vEplet) molecular mismatch algorithms exhibited the lowest Class I and DRB1/DQB1 dnDSA-free probabilities, compared to patients with only one of the mismatch scores above the ROC-defined threshold ( Figure 7 ). Furthermore, univariate and multivariate Cox regression also showed an increase of hazard ratios with the combined scores. Although verifying the improvement of dnDSA prediction with the combined B cell and T cell epitope mismatch scores in a larger cohort is warranted, our data highlight the potential of a more refined analysis approach utilizing a combination of B cell and T cell molecular mismatch assessment in donor selection and recipient immunological risk stratification, and calls for the development of computer algorithms that incorporate both T and B cell epitopes in the evaluation of recipient-donor histocompatibility.

Our study has several limitations. First, the study cohort is comprised of kidney recipients from four transplant centers, including one pediatric and three adult programs. As each center has specific protocols for immunosuppression regimens and patient management, the heterogeneity of the study subjects restricts the potential to further examine the immunogenicity of individual epitopes in this cohort. This limitation can be mitigated by using the “2MM1DSA” model conceptualized by Tambur et al. (22), where the immunogenicity of epitopes is evaluated in a patient-specific setting so that other factors that may influence immune responses in the study individuals can be controlled. Secondly, the calculation of epitope mismatches was based on high resolution typing imputed from the antigen level. The challenges of accurate HLA typing imputation, especially for non-Caucasian individuals, have been discussed in several reports (23, 24). We acknowledge that in addition to the intrinsic limitation of high resolution typing imputation, the lack of accurate ethnicity information for some of the study subjects may have further restricted our ability to infer their HLA genotype without error (25). Compared to the single imputation method employed in this study, a multiple imputation algorithm that calculates a “summed” epitope mismatch scores using weighted genotype combinations based on haplotype frequencies may improve the accuracy of epitope mismatch prediction (26, 27). Although we and others (28–31) have shown that there is reasonable concordance between molecular mismatch numbers calculated from imputed and high resolution typing ( Supplementary Figure 1 ), we recognize the criticality of correct HLA allele information for the application of epitope matching assessment in the clinical settings. With the advent of novel NGS platforms, such as Nanopore technology, rapid high resolution typing of deceased donors will soon be achievable (32–34). Another limitation of this study is the lack of data on the clinical outcomes, such as ABMR, graft survival, and patient survival. It is well recognized that the presence of dnDSA is a strong risk factor for ABMR and inferior graft survival (35–37), nonetheless, the association of molecular mismatch loads with clinical outcomes in this study cohort remains to be identified. Moreover, while a comprehensive transplant compatibility evaluation requires patient and donor’s allele-level typing at all of the HLA loci, the calculation of T cell and B cell epitope mismatches in this study were only based on HLA-A, B, C, DRB1, and DQB1 loci due to insufficient information available on DRB345, DQA1, and DPA1/DPB1 typing. Along the same line, the T cell epitope mismatches presented by the PIRCHE-II scores in our study only represent the numbers of donor-derived peptides presented by recipient’s DRB1 antigens. Even though it is generally recognized that DQ and DP molecules have lower surface expression, further analyses that include peptides presented by recipient DQ and DP proteins, as well as incorporate alleles from all HLA loci, are required to verify the observations in this study. Lastly, evaluation of combined Class I and DR+DQ molecular mismatch scores does not allow for risk assessment for dnDSA at individual HLA locus, which may have differential clinical significance depending on their expression levels.

In conclusion, in the current study we utilized three HLA molecular mismatch assessment algorithms and evaluated the correlation of B cell and T cell epitope mismatch loads with de novo DSA development in a large kidney transplant cohort. We demonstrated a significant association of Snow, verified eplet, and PIRCHE-II scores with dnDSA at HLA Class I and DRB1/DQB1 loci, independently and in synergy. Our data highlight the value of applying HLA epitope mismatch evaluation in living donor selection and immunological risk stratification to improve transplant outcomes.