Human leukocyte antigen DRB1*04:05 and clozapine-induced agranulocytosis/granulocytopenia

To the Editor

Hirakawa et al. (2019) presented a case report on the association between clozapine-induced agranulocytosis (CIA) and human leukocyte antigen (HLA). They contended that HLA-DRB1*04:05 may be associated with CIA in the Japanese population; in this population, HLA-B*59:01 was the strongest risk factor for CIA (n = 22) and clozapine-induced granulocytopenia (CIG; n = 28), with odds ratio (OR) ~10 (Saito et al., 2016).

Their rationale was based on the HLA types of the patients not harboring HLA-B*59:01 but harboring HLA-DRB1*04:05. We previously reported a modest association between HLA-DRB1*04:05 and the presence of both CIA and CIG (CIAG; OR = 3.4, p = 0.000052; 50 patients with CIAG vs 1891 ‘healthy comparison subjects’; Saito et al., 2016). In fact, we claimed that HLA-DRB1*04:05 did not pose a genuine risk but was only a marker, because HLA-DRB1*04:05 (allele frequency [AF] of 11.8%; http://hla.or.jp/med/frequency_search/en/allele/) and HLA-B*59:01 (AF = 2.0%) are in linkage disequilibrium in the population (D′ = 0.95 and r² = 0.1); the conditional analysis revealed a significantly higher p value (0.034; Saito et al., 2016).

To clarify the association between HLA-DRB1*04:05 and CIAG, CIA and CIG (in other words, for extracting the ‘pure’ effect of HLA-DRB1*04:05 on these phenotypes), we recruited an additional 899 healthy comparison subjects (total: 2790) and examined the association of HLA-DRB1*04:05 by excluding the subjects with HLA-B*59:01 (7 subjects with CIA, 5 subjects with CIG and 84 healthy comparison subjects had HLA-B*59:01). In this analysis (of 2706 healthy compassion subjects, 563 had HLA-DRB1*04:05), we found no significant association between HLA-DRB1*04:05 and CIAG (OR = 2.0, p = 0.068; of 38 subjects with CIAG, 13 subjects had HLA-DRB1*04:05), CIA (OR = 2.5, p = 0.10; of 15 subjects with CIA, 6 had HLA-DRB1*04:05), or CIG (OR = 1.7, p = 0.30; of 23 subjects with CIG, 7 had HLA-DRB1*04:05; Supplementary Table1).

From these results, we could not unequivocally conclude that HLA-DRB1*04:05 is an independent risk factor for CIAG if the effect of the HLA-DRB1*04:05 follows the pharmacogenomics concept (i.e. large effect size): The power in this analysis was sufficient (at minimum, OR = 4.0 to keep 80% power in the dominant model; α = 0.05, control AF = 0.21 in CIA comparison). In contrast, our small sample size precluded conclusive results if (1) the effect size was small or (2) specific (possibly rare) haplotypes had striking effects. Nevertheless, even in either case, we have to stress that the clinical utility of HLA-DRB1*04:05 is limited (Arranz and Kapur, 2008). Furthermore, in any cases, the modest association for HLA-DRB1*04:05 that we reported was never attributable to such situations, since the trend for the association (best p = 0.068) was null after correction for multiple testing according to the number of HLA-DRB1 alleles (p ~ 1.00, a 31-fold correction based on our previous report (Saito et al., 2016); Supplementary Table 1).