In this study, we included 1,351 individuals who received the COVID-19 vaccine and were recruited in three Italian hospitals: Fondazione IRCCS Istituto Neurologico Carlo Besta in Milan (n = 306), Azienda Ospedaliero-Universitaria Senese in Siena (n = 689), and Fondazione IRCCS Casa Sollievo della Sofferenza, San Giovanni Rotondo (FG) (n = 356; Table 1). The recruitment period spanned from December 27th 2020 and May 15th 2021. Participants with European origin, as determined by principal component analysis (PCA, Supplementary Fig. 1) were included in the genetic analyses. The cohort consisted of Italian vaccinees who were primarily hospital workers, with a predominant female representation (66.5%), and a median age of 48 years (range: 19–84). The measurement of IgG levels was performed at a median time of 40 days after the administration of the second vaccine dose (interquartile range, IQR = 68). IgG levels ranged from 12.64 to 6056 BAU/ml, with a median of 801.5 BAU/ml.
Table 1
Personal and clinical characteristics of vaccinated subjects included in the genetic analyses.
Characteristic | Total (= 1351) | Milan (= 306) | Siena (= 689) | SGR (= 356) |
Age (years), median (range) | 48 (19–84) | 47 (25–84) | 43 (19–78) | 54 (21–67) |
Sex, n (%) | | | | |
male | 452 (33.5) | 86 (28.1) | 218 (31.6) | 148 (41.6) |
female | 899 (66.5) | 220 (71.9) | 471 (68.4) | 208 (58.4) |
Days between vaccination and serological test, median (IQR) | 40 (68) | 30 (5) | 97 (37) | 30 (0) |
Serum Ab anti-SARS-CoV2 (BAU/ml), median (range) | 801.5 (12.64–6056) | 1343 (41.21–5680) | 440.5 (13.63–5680) | 1519 (12.64–6056) |
Normalized IgG values inversely correlated with age at vaccination (beta=-0.012, SE = 0.0019, P-value = 6.5x10− 11) and time (in days) passed between vaccination and serum collection for antibody measurement (beta=-0.016, SE = 0.00099, P-value < 2.0x10− 16). Additionally, we observed lower levels of IgG in individuals from Siena than those recruited in the other two cohorts (beta=-0.18, SE = 0.078, P-value = 0.024); the mean IgG levels of Siena subjects differed of 1,020 and 1,101 BAU/ml from the mean of Milan and San Giovanni Rotondo individuals, respectively. Antibody quantity was not significantly different between females and males (P-value = 0.19).
We carried out a genome-wide association analysis between normalized IgG levels and the genotypes of 521,859 variants, including in the linear regression model sex, age at vaccination, recruiting center, the first 5 principal components (PCs), and the time interval between the second vaccine dose and the serological as potential confounders. The results are reported in the Manhattan plot, shown in Fig. 1 (and listed in Supplementary Table 1). A statistically significant locus was identified on chromosome 6, in the HLA locus, with 40 variants associated with a nominal P-value < 5.0x10− 8. These variants spanned a region from 29.8 Mbp to 33.5 Mbp and the lead variant, rs2499, mapped in the 3’UTR of HLA-A gene (beta = 0.31, SE = 0.045, P-value = 1.19x10− 11).
Zooming in the locus, we observed that there was not a single association signal (Fig. 2A). Indeed, there were other regions, in addition to the one led by rs2499 at position 29,945,765, although this locus has the highest number of significantly associated variants, spanning from 29.8 Mbp to 30.1 Mbp. An additional peak of association (beta = 0.27, SE = 0.048, P-value = 4.5x10− 8) was led by rs28366135 (at position 31,396,328), that maps less than 20 kbp upstream the HLA-B gene. Then, we observed other two variants (rs454875, beta=-0.23, SE = 0.042, P-value = 4.10x10− 8; rs28688207, beta=-0.28, SE = 0.052, P-value = 5.25x10− 8) in positions 32,245,231 and 32,660,883, respectively. The latter was a splice acceptor variant of HLA-DQB1 gene. We calculated the linkage disequilibrium (LD) between rs2499 and these variants, and we observed that they were not in LD (D’=0.0030 with rs28366135, D’=0.28 with rs454875, and D’=0.0043 with rs28688207), suggesting that the three loci are independent. Indeed, in a linear regression analysis with the genotype of rs2499 as an additional covariate, we observed that the other two loci (led by rs28366135 and rs28688207, respectively) remained statistically significant (Fig. 2B).
Looking at four-digits HLA haplotypes (n = 204), we observed that in our analysis the HLA-A*03:01 was the most significantly associated with IgG levels (beta = 0.29, SE = 0.047, P-value = 5.79x10− 10), followed by HLA-C*12:02 (beta=-0.48, SE = 0.10, P-value = 2.81x10− 6), HLA-DQB1*06:01 (beta=-0.47, SE = 0.10, P-value = 2.85x10− 6), HLA-DRB1*15:02 (beta=-0.46, SE = 0.10, P-value = 8.50x10− 6), and HLA-B*52:01 (beta=-0.44, SE = 0.10, P-value = 1.62x10− 5). Considering the 107 two-digits HLA haplotypes, the top significant one was HLA-A*03 (beta = 0.28, SE = 0.044, P-value = 2.43x10− 10), and then we found HLA-B*52 (beta=-0.44, SE = 0.10, P-value = 1.62x10− 5), and HLA-DQB1*05 (beta=-0.15, SE = 0.035, P-value = 3.11x10− 5). The HLA-DQB1*06, identified by Mentzer et al. 3, was not significantly associated with IgG levels in our sample (P-value = 0.43). All results are reported in Supplementary Table 2.