In vivo evidence insufficient to conclude that “KIR/HLA incompatibility between sexual partners confers protection against HIV-1 transmission”

To the editor:

Recently, Jennes et al investigated killer immunoglobulin-like receptor (KIR) and corresponding HLA ligand in sexual transmission of HIV-1, hypothesizing that “NK cells from the recipient partner … regulated by KIR … sensing the lack of self HLA on incoming index cells” could eliminate those cells, preventing HIV transmission.¹  Studying couples with HIV-positive index case, the authors reported that 2 of many possible combinations of recipient inhibitory KIR, recipient HLA, and index HLA were associated (positively or inversely) with nontransmission of HIV to the partner. The authors also reported that in vitro killing of CD4+ T cells from HIV-infected patients by natural killer (NK) cells of healthy individuals was enhanced by inhibitory KIR whose ligand was missing from target cells (usually accompanied by activating KIR 3DS1 with ligand Bw4 present in the target).

Although plausible, the authors’ hypothesis that “KIR/HLA incompatibility” enables NK cells to prevent acquisition of HIV is not actually supported by the in vivo data they present. To illustrate, from Table 2 in the article by Jennes et al, we obtain data on 3 scenarios that fit the missing-ligand hypothesis: (1) at-risk partner’s KIR 2DL1 when the index case lacks C2; (2) at-risk 2DL3/2DL3 when the index lacks C1; and (3) at-risk 2DL2 when the index lacks C1. When these KIR/HLA combinations are subjected to univariate analysis (see Table 1), none is associated with HIV-1 transmission. It is unlikely that taking gender into account would substantially alter these results.

No hypothesis-driven risk factor that we test here was tested by Jennes et al. Instead, they narrowed risk factor 1 (as described previously) to the subset of 2DL1+ recipients whose own HLA was C1/C2 and narrowed risk factor 2 (as described previously) to 2DL3/2DL3+index C1/C2, definitions that excluded eligible subjects whose data eliminate the associations that were reported.

When evaluating multiple risk factors, it is essential to adjust criteria for statistical significance.²  Because the authors did not do so, the study far exceeded the accepted 5% chance of claiming an association when none exists. With 70 subject-couples, the study had sufficient statistical power to test 1 prespecified risk factor. Yet the study actually considered 36 risk factors (4 recipient KIR × 3 recipient HLA × 3 index HLA types), ultimately selecting 2 associations whose apparent statistical significance disappears after appropriate adjustment for multiple hypotheses.^3,4 

We recently reported, from in vivo observational study, that individual KIRs of the hematopoietic stem cell donor are associated with reactivation of cytomegalovirus posttransplantation when the recipient lacks corresponding HLA ligand.⁵  Thus, we do not doubt that missing ligand for KIR can play an important role in human infection. However, when results of in vivo observation do not confirm in vitro experiment, as occurred in this study, evidence for the hypothesis remains inconclusive. We agree with Jennes et al that definitive results await a large cohort study whose multivariate analysis takes into account potential confounding factors. Moreover, we caution that such an analysis must simultaneously evaluate individual inhibitory and activating KIRs (in the at-risk partner) together with corresponding HLA ligands (missing vs present in the index case), with appropriate adjustment of statistical significance for multiple hypothesis testing.