HLA mismatch combinations associated with decreased risk of relapse: implications for the molecular mechanism

The use of allogeneic hematopoietic stem cell transplantation (HSCT), an established treatment for hematologic malignancies, is associated with several immunologic events with contrary effects in the recipient. In graft-versus-host disease (GVHD), for example, graft immune cells attack host organs, whereas in the graft-versus-leukemia (GVL) effect, they eradicate residual leukemia cells.^1-3  GVL is likely to function not only in hematologic malignancies but also in solid tumors, particularly breast cancer and renal cell carcinoma,^4-6  in which it is referred to as the graft-versus-tumor (GVT) effect. Because both GVL and GVHD are caused by either or both major and minor histocompatibility antigen mismatches between donor and recipient, the beneficial effect of allogeneic HSCT due to GVL is thought to be frequently offset by GVHD. Thus, improving HSCT outcome by separating GVL from GVHD is a key clinical issue. Importantly, however, while most such efforts have been in the area of minor histocompatibility antigen,⁷  few researchers have approached this problem in terms of the major histocompatibility antigen.

We recently identified 16 human leukocyte antigen (HLA) mismatch combinations associated with a high risk of severe acute GVHD. Results showed that the overall number of these high-risk mismatches was strongly associated with the occurrence of severe acute GVHD and poor overall survival (OS).⁸  We speculated that the intensity of GVL and acute GVHD in any particular mismatch might not necessarily be parallel, and that among HLA mismatch combinations not inducing severe acute GVHD, those that induce strong GVL might occur. In other words, the hypotheses of this study were that particular mismatch combinations allow the separation of GVL from acute GVHD and that specific amino acid substitutions in HLA molecules contribute to this mechanism.

As part of efforts to improve donor selection and allogeneic HSCT outcomes, we identified HLA mismatch combinations that resulted in a decreased risk of relapse in all 6 major HLA loci and compared them with mismatch combinations carrying a high risk of severe acute GVHD. Further, we investigated specific amino acid substitution positions in the HLA molecule responsible for a decreased risk of relapse.

This study was conducted using clinical data that were collected prospectively at transplant centers participating in the Japan Marrow Donor Program. Patients who received a first transplant of T cell–replete marrow for a hematologic malignancy from a serologically HLA-A, -B, and -DR antigen-matched unrelated donor between January 1993 and December 2005 through the Japan Marrow Donor Program (n = 4643) were registered. Eligible diagnoses included acute lymphoblastic leukemia (ALL); acute myeloid leukemia (AML), which included only de novo AML; chronic myeloid leukemia (CML); malignant lymphoma (ML); and multiple myeloma (MM).

Patient characteristics are shown in Table 1. A final clinical survey of the patients was completed by December 2006. Informed consent was obtained from patients and donors in accordance with the Declaration of Helsinki, and approval for the study was obtained from the Institutional Review Board of Aichi Cancer Center and the Japan Marrow Donor Program.

Alleles at the HLA-A, -B, -C, -DRB1, -DQB1, and -DPB1 loci were identified by previously described methods in all 4643 pairs at the Japanese Red Cross Tokyo Metropolitan Blood Center.^8,9 

HLA allele mismatch among the donor-recipient pair was scored when the recipient's alleles were not shared by the donor (graft-versus-host vector) for all analyses.

Relapse was defined as the recurrence of malignancy as detected by the parameter by which the malignancy was first detected, namely marrow morphology; flow cytometry; cytogenetic studies, including fluorescence in situ hybridization; electrophoresis; immunofixation assays; polymerase chain reaction-based assays for disease markers; or imaging results. The day of relapse was defined as the day on which the respective clinical, hematologic, cytogenetic, or molecular relapse was recognized.

Amino acid sequences of HLA-Cw and -DPB1 molecules were obtained from the IMGT/HLA sequence database.¹⁰  For example, Tyr99C-Phe99C indicated an amino acid substitution at position 99 in the HLA-C molecule in which the donor had tyrosine and the patient had phenylalanine. Substituted amino acids in HLA-Cw and -DPB1 are summarized in Tables S1 and S2 (available on the Blood website; see the Supplemental Materials link at the top of the online article).

OS rate was assessed using the Kaplan-Meier product limit method. To eliminate the effect of competing risk, the cumulative incidence of relapse was assessed using a previously described method.^11,12  The competing event for relapse was defined as death without relapse. Impact by the factor of interest was assessed using the log rank test. The impact of HLA allele mismatch combinations and the position and type of amino acid substitution (for example, alanine, arginine, and asparagine) in HLA molecules were evaluated using multivariable Cox regression analysis¹³  for OS and the occurrence of acute GVHD, while the risk of relapse was evaluated using the multivariable proportional hazard modeling of subdistribution functions in competing risks.¹⁴ 

HLA mismatch combinations were evaluated for each locus separately. When the locus of interest was evaluated, we allowed the other loci to be mismatched, with the status of such mismatches adjusted for in the same way as other confounders. The HLA match and HLA one-allele mismatched in every locus were analyzed. For example, the A*0206-A*0201 mismatch combination meant that the donor had HLA-A*0206, the recipient had HLA-A*0201, while another HLA-A allele of the donor and recipient was identical. This mismatch was compared with the HLA-A allele match. Mismatch combinations that had 9 or fewer pairs were combined together as “other mismatch.” The model was constructed with mismatch combinations, mismatch status in other loci (match, 1 allele mismatched, and 2 alleles mismatched, as an ordinal variable), and potential confounders. Confounders considered were sex (donor-recipient pair), patient age (linear), donor age (linear), transplant year, type of disease, risk of leukemia relapse (standard, high, and diseases other than leukemia), GVHD prophylaxis (cyclosporine [CSP] vs tacrolimus [FK]), ATG (vs no ATG), and preconditioning (TBI vs non-TBI). These confounders were used in all analyses to maintain the comparability of results.

The impact of position and type of amino acid substitutions in HLA molecules was evaluated in pairs with oneallele mismatched in HLA-Cw and -DPB1 separately. The amino acid positions we analyzed were all positions at which an amino acid was substituted in the respective locus. We analyzed the impact of each amino acid substitution on each position separately. Multivariable models were constructed to include the position and type of amino acid substitution, mismatch status in other loci (match, 1 allele mismatched, and 2 alleles mismatched as an ordinal variable) and the confounders described above. A P value less than .05 was considered statistically significant. All statistical tests were 2-sided. All analyses were performed using STATA version 10.0 (StataCorp, College Station, TX) and R version 2.5.1 (The R Foundation for Statistical Computing, www.r-project.org).

Statistical analyses were validated using the bootstrap resampling method.¹⁵  Briefly, we estimated the measure of association with resampled data drawn repeatedly from the original data. Although approximately 100 to 200 bootstrapped samples are generally sufficient,¹⁶  we used 1 000 bootstrap samples for all analysis validations. Further, we judged the results of analysis as statistically significant only when the results of both base analysis and analysis validation using bootstrap resampling were significant; cases in which the result of base analysis was significant but that of analysis validation using bootstrap resampling was not are indicated by an asterisk next to the P value of the base analysis.

The number of mismatched alleles of HLA-Cw (1 allele mismatched: hazard ratio [HR], 0.68; 95% confidence interval [CI], 0.58-0.80; 2 alleles mismatched: HR, 0.43;95% CI, 0.24-0.75) and HLA-DPB1 (1 allele mismatched: HR, 0.80; 95% CI, 0.70-0.92; 2 alleles mismatched: HR, 0.62; 95% CI, 0.51-0.75) was strongly associated with a decreased risk of relapse. In contrast, no associations were seen for HLA-A (1 allele mismatched: HR, 1.00; 95% CI: 0.82-1.22; 2 alleles mismatched: HR, 0.79; 95% CI, 0.28-2.28), HLA-B (1 allele mismatched: HR, 1.06; 95% CI, 0.79-1.41; 2 alleles mismatched: not applicable), HLA-DRB1 (1 allele mismatched: HR, 0.93; 95% CI, 0.74-1.18; 2 alleles mismatched: HR, 1.18, 95% CI: 0.53-2.63) or HLA-DQB1 (1 allele mismatched: HR, 1.12; 95% CI, 0.90-1.40; 2 alleles mismatched: HR, 0.73; 95% CI, 0.35-1.52; Figure 1; Table 2).

Four mismatch combinations in HLA-Cw and 6 in HLA-DPB1 were significantly associated with a decreased risk of relapse (Tables 3 and S3). In contrast, mismatch combinations in HLA-A, -B, -DRB1, and -DQB1 were not significantly associated with differences in risk of relapse (data not shown). The 10 HLA mismatch combinations associated with lower risks of relapse were Cw*0102-Cw*1402 (HR not estimated due to no event), Cw*0801-Cw*0102 (HR not estimated), Cw*1402-Cw*0304 (HR not estimated), Cw*1502-Cw*1402 (HR, 0.28; 95% CI, 0.09-0.88), DPB1*0402-DPB1*0201 (HR, 0.32, 95% CI, 0.12-0.87), DPB1*0501-DPB1*0201 (HR, 0.67; 95% CI:0.50-0.91), DPB1*0501-DPB1*0401 (HR, 0.36; 95% CI, 0.13-0.98), DPB1*0501-DPB1*0402 (HR, 0.55; 95% CI, 0.33-0.93), DPB1*0901-DPB1*0201 (HR, 0.37; 95% CI, 0.14-0.96), and DPB1*1301-DPB1*0201 (HR not estimated; Tables 3 and S3). All 10 HLA mismatch combinations were also significant on validation analysis using the bootstrap resampling method. We speculated that these mismatch combinations would mainly decrease the risk of relapse due to GVL, so we tentatively call them GVL mismatch combinations.

We evaluated the clinical importance of GVL mismatch combinations in HLA-Cw and -DPB1. All analyses in this section were conducted in matched pairs other than the evaluated locus. In HLA-C mismatch, the small number of patients with GVL mismatch combinations (n = 13) in matched pairs at the allele level for HLA-A, -B, -DRB1, -DQB1, and -DPB1 prevented comprehensive analysis. We evaluated the GVL mismatch combinations of HLA-DPB1 in matched pairs for HLA-A, -B, -Cw, -DRB1, and -DQB1. Pairs with HLA-DPB1 mismatch were divided into 2 groups, those with a GVL mismatch combination and those with mismatch combinations other than GVL mismatch combinations. These were then compared with 12/12 matched pairs for association with severe acute GVHD, relapse, and OS (Table 4). The curve of the cumulative incidence of OS is shown in Figure 2. Multivariable analysis revealed that although OS was similar between the 12/12 matched pairs and the pairs with mismatch combinations other than GVL mismatch combinations, it was significantly improved in pairs with a GVL mismatch combination (Table 4). In terms of mortality due to relapse according to HLA-DPB1 matching status and whether the mismatch combinations were GVL mismatch combinations, the HLA-DPB1 matched group, HLA-DPB1 1 allele mismatched group, and GVL mismatch combination group showed an expected decreased mortality due to relapse (20.0%, 15.3%, and 10.5%, respectively). Further, mortality due to relapse in the GVL mismatch combination group was significantly lower than that in the HLA-DPB1 1 allele mismatched group (P = .049). We conducted the same analyses with stratification by leukemia type (ALL, AML, or CML) and found that the myeloid malignancies (AML and CML) had the same tendency (Table 4). In particular, in CML, GVL mismatch combinations in HLA-DPB1 were associated with a significantly reduced risk of relapse (HR, 0.14; 95% CI, 0.03-0.55) and significantly improved OS relapse (HR, 0.50; 95% CI, 0.25-0.98).

We surveyed all substituted positions in HLA-Cw and -DPB1 and found 159 specific amino acid substitutions at 55 positions in HLA-Cw and 55 specific amino acid substitutions at 19 positions in HLA-DPB1 (Tables S1,S2). Analysis revealed 3 specific amino acid substitutions responsible for a decreased risk of relapse in HLA-C, namely Ser9C-Tyr9C (HR, 0.53; 95% CI, 0.30-0.92), Phe99C-Tyr99C (HR, 0.52, 95% CI, 0.30-0.91), and Arg156C-Leu156C (HR, 0.59; 95% CI, 0.37-0.92). In contrast, no decrease in the risk of relapse was seen for substitutions in HLA-DPB1 (Table 5). However, Tyr9C-Ser9C and Tyr99C-Phe99C were strongly linked (see “Discussion”). These specific amino acid substitutions were all significant on validation analysis using the bootstrap resampling method.

Improving outcomes in allogeneic HSCT for hematologic malignancies by separating GVL from GVHD is considered a key clinical challenge. Here, our analysis demonstrated that several donor-recipient HLA mismatch combinations and specific amino acid substitutions in HLA molecules were associated with a decreased risk of relapse, and, in some cases, no significant increase in the risk of severe acute GVHD. These findings suggest that GVL might be separated from severe acute GVHD by selection of suitable HLA mismatch combinations.

We recently reported 16 significant high-risk HLA allele mismatch combinations for severe acute GVHD in 6 HLA loci, a number of which were highly associated with the occurrence of severe acute GVHD and worse OS.⁸  Of note, a group of pairs with mismatches other than severe acute GVHD high-risk mismatches showed an incidence of severe acute GVHD and OS rates almost equal to those of 12/12 matched pairs. In the present study, we elucidated a total of 10 mismatch combinations that were significantly associated with a decreased risk of relapse, which we termed GVL mismatch combinations. Of course, it is possible that some mismatch combinations not classified as GVL mismatch combinations might actually induce strong GVL. Misclassification might have occurred as a result of insufficient statistical power due to the relatively small number of patients in the subcategories. Among these mismatch combinations, 2 of 4 in HLA-Cw were identical to the severe acute GVHD high-risk combinations; a third had a marginal effect on the occurrence of severe acute GVHD, while the fourth combination was different from acute GVHD high-risk mismatch combinations. In contrast, all 6 mismatch combinations in HLA-DPB1 were different from acute GVHD high-risk mismatch combinations (Table 3). As expected, HLA-A, -B, -Cw, -DRB1, and -DQB1 matched pairs with GVL mismatch combinations of HLA-DPB1 were associated with significantly better OS than 12/12 matched pairs (Table 4; Figure 2), indicating that the beneficial antitumor effect of GVL mismatch combinations in HLA-DPB1 would not be offset by the effect of severe acute GVHD. We speculate that conformational changes of HLA molecules in each mismatch combination control the intensity of the acute GVHD and GVL effect, as described later in “Discussion” and in our previous report⁸ ; namely, conformational changes of HLA molecules in GVL mismatch combinations in HLA-DPB1 induce strong GVL with mild or no acute GVHD. These findings suggest that HLA mismatch selection according to these results might improve HSCT outcomes over those obtained with a complete match. The same tendency was seen for AML and CML, whereas the effect of GVL mismatch combination in the HLA-DPB1 allele in ALL patients would be weaker than in the other leukemia types (Table 4). Comprehensive analyses for ML and MM could not be done because of the small number in each group. Thus, the effects of GVL mismatch combination vary according to disease type and may also change according to other factors, including particular cytogenetic abnormalities.

Recent research has shown that HLA-Cw and -DPB1 mismatch at the allele level is strongly associated with a decreased risk of relapse.^17,18  These findings were confirmed in the present large cohort. In addition, the present study also clarified that the mismatching of 2 alleles in either the HLA-Cw or -DPB1 locus had a stronger association with decreased risk than respective mismatching of one allele. Moreover, no association whatsoever was seen for HLA-A, -B, -DRB1, or -DQB1 (Figure 1; Table 2). Furthermore, all 10 GVL mismatch combinations were elucidated from mismatch combinations of HLA-Cw and HLA-DPB1 (Tables 3 and S3), although we also analyzed HLA-A, -B, -DRB1, and -DQB1. These findings indicate that GVL after allogeneic HSCT is mainly induced by HLA-Cw and -DPB1, not HLA-A, -B, -DRB1 or -DQB1, although the role of each HLA locus might vary with the type of disease.¹⁸  There are 3 possible explanations for this. First, the relative expression of HLA-Cw and -DPB1 on malignant cells may be higher than that on normal hematopoietic cells; second, HLA-Cw and -DPB1 may be preferentially expressed on malignant stem cells; and third, surface expression of a few key molecules—such as major histocompatibility complex (MHC), adhesion, and costimulatory molecules—on malignant cells may determine the effect of each HLA locus on GVL.^19-21  In other words, some molecules might stimulate GVL of HLA-Cw or -DPB1, and other molecules might block GVL of other than HLA-Cw and -DPB1. Further investigation of this question is warranted.

In this study, 3 specific amino acid substitutions responsible for GVL at positions 9, 99, and 156 were identified in HLA-Cw, of which only 2, Ser9C-Tyr9C and Phe99C-Tyr99C, were strongly linked in our sample. We were therefore unable to determine which substitutions are the main contributors to the effect of interest (Table 5). These amino acid positions, 9, 99, and 156, were identical to those we elucidated in our previous study as responsible for severe acute GVHD.⁸  These findings suggest that these 3 amino acid positions are important determinants of alloreactivity. Although position 156 of the HLA molecule has been shown to modify T-cell alloreactivity in vitro in HLA-A2,^22-24  -B35,²⁵  and -B44,²⁶  to our knowledge, the present study is the first to identify positions 9 and 99. On the other hand, substituted amino acids were not necessarily identical. In Ser9C-Tyr9C and Phe99C-Tyr99C substitutions, for example, the substituted amino acid position was identical with that responsible for severe acute GVHD, whereas the substituted amino acids were inverse between donor and recipient, even though both substituted position and amino acids were identical in the Arg156C-Leu156C substitution. These findings suggest that Ser9C-Tyr9C and Phe99C-Tyr99C might play an important role in separating GVL from acute GVHD in HLA-Cw mismatch, although the mechanism requires further molecular clarification.

With regard to specific amino acid substitutions of HLA-DPB1, we found no significant association among these with a decreased risk of relapse. Shaw et al²⁷  reported that mismatches at position 57 and 65 in the HLA-DPB1 molecule were associated with transplant complications, but not with GVHD or relapse, which is consistent with our present data. We speculate that, compared with MHC class I, the conformational diversity of MHC class II and peptide complex hampers the identification of strict rules of association between specific amino acid substitutions in MHC class II molecules and the occurrence of alloreaction such as GVHD and GVL. In HLA class I, binding peptides are held by their ends, whereas peptides bind to HLA class II by attachment in the middle, allowing them to vary greatly in length.²⁸ 

Given that this analysis was conducted using a Japanese cohort of patients who received transplants through the Japan Marrow Donor Program, the applicability of our data to other ethnic groups warrants discussion. We speculate that the effect of alloreaction is a reflection and summation of HLA allele mismatch combinations. Discrepancies in the effect of HLA locus on alloreactions between ethnically diverse transplantation might be explained by the proportions of each HLA mismatch combination in each HLA locus. In HLA-DPB1, on the other hand, the allele variations between white and Japanese populations are relatively close, hence our findings in HLA-DPB1 might also be useful for white populations. Regarding HLA-Cw and killer immunoglobulin-like receptor (KIR) incompatibility, we previously reported adverse effects in unrelated T cell–replete HSCT through the Japan Marrow Donor Program,¹⁸  although Ruggieri et al²⁹  demonstrated that beneficial effects were shown in T-cell depleted haploidentical transplantation. We speculated that in vivo and/or in vitro T-cell depletion could account for this discrepancy.³⁰  Therefore, results for mismatch combinations in HLA-Cw obtained in other populations treated in other settings may differ from our results. Nevertheless, clarification of these questions would require the same study in other ethnic populations.

Given the general acceptance that GVL is more closely correlated with chronic GVHD than acute GVHD,³  separating GVL from chronic GVHD may be more difficult than separating it from acute GVHD. On this basis, our results suggest that GVL could be separated from acute GVHD in HSCT from a specific HLA partially mismatched donor. Clarification of whether GVL can also be separated from chronic GVHD requires further study.

In conclusion, we identified 4 HLA-C and 6 HLA-DPB1 mismatch combinations that decrease the risk of relapse in patients after HSCT. Eight of 10 GVL combinations were different from those responsible for severe acute GVHD. In particular, all 6 GVL combinations in HLA-DPB1 were different. Further, pairs with these GVL combinations of HLA-DPB1 were associated with significantly better OS than completely matched pairs. These findings suggest that donor selection according to these results could separate the occurrence of GVL from acute GVHD, especially in HLA-DPB1. Further, amino acid substitutions on specific positions responsible for this decreased risk of relapse were also elucidated in HLA-C, but not in HLA-DPB1. Our finding that specific amino acid substitutions decrease the risk of relapse might be key to revealing the mechanism of the decreased risk of relapse due to GVL with regard to the HLA molecule.

We thank the staff members of the transplant centers, donor centers, and the Japan Marrow Donor Program office for their generous cooperation; Ms Ryoko Yamauchi for the data management; and Drs Toshitada Takahashi and Setsuko Kawase for their expert technical assistance.

This study was supported in part by Health and Labor Science Research Grant (Research on Allergic disease and Immunology) from the Ministry of Health, Labor, and Welfare (Tokyo, Japan) and the Grant-in-Aid for Cancer Research (19-1) from the Ministry of Health, Labor, and Welfare (Tokyo, Japan).

Contribution: T.K., Y.M., T.S., S.O., and Y.K. participated in the conception of this study; K.K., H.I., and H.S. participated in the assessment of histocompatibility; Y.M. and S.K. participated in the execution of transplantation; T.K. and K.M. participated in the statistical data analysis; T.K. and Y.M. wrote the paper; and all authors checked the final version of the manuscript.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Yasuo Morishima, Department of Hematology and Cell Therapy, Aichi Cancer Center Hospital, 1-1 Kanokoden, Chikusa-ku, Nagoya, 464-8681, Japan; e-mail: ymorisim@aichi-cc.jp.