Response: KIR/HLA incompatibility in HIV-1 transmission and importance of the ‘missing self’ model

We appreciate the interest of Drs Behrendt and Zaia in our recent study of allogeneic killer cell immunoglobulin-like receptor (KIR)/HLA combinations in HIV-1 transmission,¹  and we are grateful for the opportunity to respond to their comments. Behrendt and Zaia reanalyzed our data by testing regrouped “missing ligand” combinations consisting of recipient partner KIRs lacking cognate index partner HLAs, reporting that they are not associated with HIV-1 transmission. In our analysis, however, we specifically studied “missing self” combinations, which also take recipient HLA into account, adding the requirement that natural killer (NK) cells have to be educated by self HLA before they can sense the lack of it on incoming target cells.²  Missing self, but not missing ligand, combinations were shown to be beneficial during haploidentical hematopoietic stem cell transplantation as a treatment of acute myeloid leukemia, inducing alloreactive NK cells that help prevent leukemia relapse and graft-versus-host disease.^3,4  Conversely, the “matched” combinations we tested consisted of recipient KIRs recognizing index HLAs regardless of recipient HLA, because recipient NK cells will be inhibited by index HLA regardless of whether they have been educated or not. Our analyses (see Table 2 in Jennes et al¹ ) showed that HIV-1 transmission and lack thereof correlated with 1 such matched combination (recipient KIR2DL3/KIR2DL3 with index HLA-C1/C2) and 1 such missing self combination (recipient KIR2DL1 with HLA-C1/C2 and index HLA-C1/C1), respectively, both of which are diluted out in the regrouped analyses by Behrendt and Zaia.

For reasons of simplicity, we chose to analyze all relevant KIR2DL/HLA-C matched and missing self subset combinations in a single step. Although recipient KIR2DL3/KIR2DL3 with index HLA-C1 and recipient KIR2DL1/HLA-C2 with index HLA-C1/C1 also yielded statistically or near-statistically significant differences (P values of .019 and .075, respectively, not shown in our paper), our subset analyses immediately showed that these differences were only valid for index HLA-C1/C2 and recipient HLA-C1/C2, respectively (P values of .001 and .023, respectively, as shown in our paper). However, we agree that it is not clear why the alternative matched or missing self combinations did not correlate with HIV-1 transmission. We can speculate that HLA-C1/C2 index cells are more infectious because they are more widely protected against recipient NK cell killing (ie, through either KIR2DL1, KIR2DL2, or KIR2DL3), whereas HLA-C1/C2 recipient NK cells are more protective because they are more widely educated for index target cell killing (ie, through either KIR2DL1, KIR2DL2, or KIR2DL3). At least for HLA-C1/C2 index cells, this is corroborated by the functional assays that we performed showing low to no NK cell killing of HLA-C1/C2 targets (see Table 4 in our paper). Similarly, we discovered no missing self combinations for recipient KIR2DL2 or KIR2DL3 in our couple cohort, but this likely stems from the low number of couples in these categories. Thus, although we agree with Behrendt and Zaia that these questions need to be addressed in future studies—preferably by using larger couple cohorts and larger sets of functional experiments as we have acknowledged in our paper—they do not invalidate the conclusions that we have drawn to date.

We did not apply a correction for multiple testing because this study was a hypothesis generating exploratory analysis of observational data and because testing was limited to a number of preplanned and scientifically plausible comparisons.^5,6  In the analysis concerned, we performed 16 such comparisons and not 36 as Behrendt and Zaia assumed; this selective testing explains why the number of couples don’t add up to 100%. With 70 couples included, our analysis had 93% and 71% power to detect the identified matched and missing self combinations at P < .05, respectively. It is unclear to us on what basis they state we had sufficient power to perform only 1 comparison. Even if a standard Bonferroni correction would have been applied, which is far too conservative given interdependence resulting from KIR linkage and allelic distribution of KIR2DL2/KIR2DL3 and HLA-C1/C2, the matched combination would still remain statistically significant. Our study is the first to suggest a role for allogeneic KIR/HLA interactions and alloreactive NK cells in protection against HIV-1 transmission. Because of the novelty of our findings, the relatively small size of our cohort, and that we did not correct for multiple testing, independent confirmation of our findings in other cohorts is certainly warranted.