2.2. Genotyping

We selected four IFNL genetic variants for the study: rs368234815, rs12979860, rs117648444, and rs28416813 (Figure 1a). Genomic DNA was isolated from blood using the QIAamp DNA mini kit (Qiagen). The concentration and quality of the isolated genomic DNA was measured using absorbance spectroscopy (Nanodrop, ThermoFisher). The genotyping of all the samples were carried out using competitive allele‐specific polymerase chain reaction (KASP, LGC Genomics, UK) method for the SNPs rs12979860, rs117648444, and rs28416813. The assay was carried out in Quantstudio 5 (Applied Biosystems) real‐time PCR machine. For genotyping of rs368234815, a subset of seventy‐seven samples were sequenced using Sanger sequencing from an amplicon generated from the primers: (1) IL28B4kbxhoRev: 5’‐GATATCCTC GAGCCCGGATTTCAGGAC‐3′ and (2) IL28B4kbKpnFor: 5’‐GATATCGGTACCGCCCTGGACGGGAAAG‐3′. The Genbank reference sequence numbers for: IFNL3 is NM_001346937.2 and for IFNL4 is NM_001276254.2.

3.2. Distribution of variance between the groups within dominant and recessive models of IFNL genetic variants in the combined cohort or in male and female sub‐cohorts

Our study design is loosely based on the concept of Mendelian Randomization (Adam, 2019), wherein random segregation of alleles occurs during meiosis and that any association that the alleles may show with any phenotype(s) in the offspring should be free from confounding. We tested if the variance that is naturally present in the data that we collected for the 30 phenotypes had a random distribution between the two groups formed under either the dominant or recessive models for the IFNL genetic variants. We saw some highly significant associations in many of the phenotypes with p‐values ranging from 5×10⁻² to 3×10⁻³⁵ (not shown). In order to filter out the strongest associations, we set the following criteria: (a) a p‐value <4.1×10⁻⁴ (after Bonferroni correction for 120 tests: two SNPs under two models for 30 phenotypes); (b) if it is a dominant model, at least one of the two SNPs should have a significant p‐value; and (c) if it is a recessive model, both SNPs should show a significant p‐value for association. We saw that eight phenotypes were able to survive these stringent criteria and are shown in Figure 2. Next, we tested if the phenotype variance we saw between the groups was randomly distributed between the males and females (Figure 2). We chose only the dominant model to compare the distribution among the two genders as the sample numbers within the gender sub‐cohorts were more than 30 under this model in order to justify the use of a parametric test like F‐test (Ghasemi & Zahediasl, 2012). The variance distribution between the genotype groups under the dominant model was more significantly different in males compared to the females in many of the phenotypes except in blood glucose levels, where females contributed to the bulk of the association that we saw in the combined cohort. In some phenotypes like GGT and TSH, both males and females contributed to the significant association seen in the combined cohort, whereas in triglycerides, the association in males and females was in opposite directions (Figure 2). These results suggest to us that the IFNL genetic variants could potentially influence multiple phenotypes in humans in a gender‐specific manner.

Open in a separate window

FIGURE 2

Non‐random distribution of variance according to IFNL SNPs and gender in multiple phenotypes in the study cohort. Heat map shows the p‐value distribution in the combined cohort and the gender sub‐cohorts after applying Bonferroni correction for multiple testing as described in the Results section; the darker color is suggestive of highly significant association of the variance between the major homozygotes group and the group with the remaining individuals (dominant model for the minor allele). The p‐values were derived from the F‐statistic. The results from only the dominant model are shown here. *Triglyceride levels qualified for the criteria by showing significant p‐values for both the SNPs under the recessive model even though the dominant model showed insignificant p‐values in the combined cohort; however, they showed highly significant p‐values for both the SNPs under the dominant model only when the cohort was stratified on gender and interestingly the association was in opposite directions in males vs. females. Abbreviations: VLDL‐cholesterol: Very Low‐Density Lipoprotein‐cholesterol, SGOT: Serum Glutamic Oxaloacetic Transaminase, GGT: Gamma Glutamyl Transferase, TSH: Thyroid Stimulating Hormone.

3.3. Significant associations between multiple phenotypes and IFNL genetic variants

Since the data for all the phenotypes showed a non‐normal distribution (Kolmogorov–Smirnov test; p‐value <1.1×10^‐8), we used a non‐parametric test (Wilcoxon Mann–Witney U test) to look for significant differences in the midpoints of the quantitative traits in each of the groups being compared. We found no association with any of the 30 phenotypes with either of the SNPs rs12979860 or rs28416813 using either the dominant or the recessive models (Table 1; data not shown for all phenotypes). It is known from several previous studies, including ours (Astudillo et al., 2020; Chinnaswamy et al., 2017; Eslam et al., 2015), that gender can be an effect modifier with IFNL genetic variants. Therefore, we stratified the cohort into males and females and tested all the parameters under the two models and also based on IFN‐λ4 copy number and activity status (Table 1; data shown only for significant associations; a total of 270 tests were carried out: 120 for the two SNPs under two models for 30 phenotypes; a total of 150 tests for five comparisons of IFN‐λ4 copy number and activity status with 30 phenotypes). There were several significant associations (p<0.05) that we observed in males (a total of 15 out of a total of 270 tests conducted) and a few in females (a total of 5 out of a total of 270 tests conducted). The associations that were significant (p<0.05) were not randomly distributed across males and females (p=0.02; Chi‐square test). This suggests that there is a gender‐specific effect of the IFNL genetic variants on up to 11 phenotypes (Table 1).

TABLE 1

Association of blood profile phenotypes with IFNL3/4 polymorphisms.

	Combined cohort				Male sub‐cohort				Female sub‐cohort
	rs12979860 (C>T)		rs28416813 (C>G)		rs12979860 (C>T)		rs28416813 (C>G)		rs12979860 (C>T)		rs28416813 (C>G)
	N=544		N=547		N=278		N=281		N=266		N=266
	C/C=317		C/C=323		C/C=161		C/C=163		C/C=156		C/C=160
	C/T=187		C/G=185		C/T=96		C/G=99		C/T=91		C/G=86
	T/T=40		G/G=39		T/T=21		G/G=19		T/T=19		G/G=20
	Dom	Rec	Dom	Rec	Dom	Rec	Dom	Rec	Dom	Rec	Dom	Rec
Platelet count	0.66	0.09	0.43	0.25	0.80	0.85	0.70	0.96	0.37	0.02	0.12	0.14
% of Monocytes	0.14	0.29	0.14	0.94	0.007	0.11	0.01	0.24	0.50	0.87	0.76	0.24
SGOT	0.21	0.49	0.09	0.90	0.02	0.05	0.01	0.09	0.65	0.32	0.94	0.10
Albumin	0.22	0.20	0.34	0.29	0.31	0.84	0.50	0.73	0.35	0.01	0.32	0.04
Globulin	0.98	0.68	0.62	0.83	0.38	0.03	0.50	0.12	0.43	0.13	0.16	0.33
Uric Acid	0.81	0.27	0.86	0.10	0.86	0.01	0.77	0.02	0.59	0.90	0.32	0.67

Open in a separate window

	Combined cohort					Male sub‐cohort					Female sub‐cohort
	P70_1 vs S70*	P70_1 vs No‐IFNL4	P70_2 vs S70*	P70_2 vs No‐IFNL4	P70_1 vs P70_2	P70_1 vs S70*	P70_1 vs No‐IFNL4	P70_2 vs S70*	P70_2 vs No‐IFNL4	P70_1 vs P70_2	P70_1 vs S70*	P70_1 vs No‐IFNL4	P70_2 vs S70*	P70_2 vs No‐IFNL4	P70_1 vs P70_2
% of Eosinophils	0.02	0.45	0.26	0.55	0.38	0.47	0.88	0.51	0.93	0.96	0.01	0.20	0.39	0.43	0.18
% of Monocytes	0.61	0.27	0.63	0.71	0.85	0.78	0.03	1	0.28	0.84	0.53	0.56	0.47	0.45	0.66
Triglycerides	0.75	0.92	0.03	0.01	0.01	0.64	0.80	0.13	0.01	0.01	0.20	0.97	0.15	0.49	0.44
VLDL	0.75	0.96	0.03	0.01	0.01	0.64	0.80	0.13	0.01	0.01	0.20	0.95	0.15	0.50	0.43
ALP	0.01	0.08	0.13	0.43	0.82	0.01	0.05	0.10	0.28	0.75	0.70	0.58	0.77	0.89	0.98
Creatinine	0.22	0.36	0.56	0.94	0.61	0.94	0.73	0.28	0.16	0.26	0.05	0.19	0.04	0.17	0.67
Uric Acid	0.63	0.85	0.44	0.25	0.23	0.35	0.26	0.11	0.03	0.01	0.95	0.65	0.93	0.97	0.85

Open in a separate window

Note: Only significantly associated (p<0.05) phenotypes out of the 30 phenotypes in either the male or female sub‐cohort are shown here. A total of 270 tests were carried out: 120 for the two SNPs under two models for 30 phenotypes; a total of 150 tests for five comparisons of IFN‐λ4 copy and activity status with 30 phenotypes. All the p‐values shown are from Wilcoxon Mann Whitney U‐test. Dominant (Dom) and recessive (Rec) models for the minor alleles of each SNP (e.g., C/C vs C/T + T/T for rs12979860 dominant model, C/C + C/T vs T/T for rs12979860 recessive model) and IFN‐L4 status (No IFNL4 or 1 copy of IFNL4‐P70 or 2 copies of IFNL4‐P70, and 1 copy of IFNL4‐S70*) were used for the comparisons. p‐values <0.05 were considered significant and highlighted in bold. VLDL‐cholesterol: Very Low‐Density Lipoprotein‐cholesterol, SGOT: Serum Glutamic Oxaloacetic Transaminase, ALP: Alkaline Phosphatase

^*The IFNL4‐S70 group has some IFNL4‐P70 alleles as heterozygotes along with IFNL4‐S70 alleles; the combined cohort has 44 copies of IFNL4‐S70 alleles and 10 copies of IFNL4‐P70 alleles, the males sub‐cohort has 24 IFNL4‐S70 alleles and 7 copies of IFNL4‐P70 alleles, while the female sub‐cohort has 20 IFNL4‐S70 alleles and 3 copies of IFNL4‐P70 alleles.

Next, we tested if several demographic and lifestyle parameters that we collected as part of the metadata could be potential confounders in our significant results shown for the 11 phenotypes in Table 1. We had collected a total of 11 demographic and lifestyle factors in our metadata (Table S3). For males and females separately, we tested if any of these factors significantly associated with the two SNPs rs12979860 and rs28416813 based on the groupings under dominant or recessive models and also groupings based on IFN‐λ4 copy number and activity status using Chi‐square tests. None of the demographic/lifestyle factors showed a significant association (p<0.05) with either of the SNP or based on IFN‐λ4 copy number and activity status in males (data not shown); however, in females, we saw three factors (housing type, annual income, and source of drinking water) showing a significant association (p<0.05) with either of the SNPs or with the IFN‐λ4 copy number and activity status (Table 2). We deemed these three variables to be potential confounders in the significant association we saw in females with the IFNL variants highlighted in Table 1. Out of the four phenotypes that showed significant association with IFNL variants in females (shown in Table 1), two, i.e., platelet counts and albumin levels also significantly (p<0.05) associated with at least one of the three demographic/lifestyle variables (Table 2), suggesting that the latter could be confounders in our results on significant association between IFNL genetic variants and the two phenotypes.

Only those significant associations between IFNL genetic variants and peripheral blood phenotypes that were free from confounding by the 11 demographic/lifestyle factors (as shown in Table S3) were chosen and adjusted for age using multiple linear regression. We saw that the only two significant associations we had seen in females, i.e., with eosinophil % and creatinine levels became insignificant after adjusting for age (Table S4; p‐values only for the genetic variants are shown in the table). The remaining significantly associated phenotypes in males are shown as violin plots in Figures 3 and ​and4.4. Monocytes and SGOT levels showed significant association after adjusting for age in males under the dominant model, while uric acid levels were significantly associated under the recessive model (Figure 3a–c).

Open in a separate window

FIGURE 3

Multiple phenotypes significantly associate with IFNL SNPs only in males. (a–d) Violin plots showing the distribution of data around the median (thick dashed line shows median and the two thin lines separate the quartiles in each half) for the phenotypes shown according to the IFNL SNPs under the dominant or recessive models (for a, b and c) or additive model (for d) for the minor allele as shown. All the p‐values shown were for the effect of the individual SNPs derived from multiple linear regression analysis with age as the covariate (fully described in Table S4). No significant effect of the SNPs was seen in the female sub‐cohort.

Open in a separate window

FIGURE 4

Multiple phenotypes significantly associate with groupings based on IFN‐λ4 copy number and status. (a–d) The IFN‐λ4 status was inferred from the genotype information at rs12979860 and rs117648444 for each individual before grouping them in to one of the four groups‐ no‐IFN‐λ4, P70‐1 (having a single copy of the high activity P70 variant of IFN‐λ4), P70‐2 (having two copies of the high activity P70 variant of IFN‐λ4) and S70‐1 (while the large majority of the individuals in this group had one copy of the low‐activity S70 variant of IFN‐λ4, a minority of them had diplotypes, i.e. had one copy of the high activity P70 variant as well, as explained in e). (e) The IFN‐λ4 status based on copy number and activity level explained. A minority (10 in the combined, 07 in the males and 03 in the female sub‐cohort) of individuals had both a high activity and low‐activity IFN‐λ4 variant in them; this grouping strategy allowed us to have more power in testing the S70 variant due to the low MAF of rs117648444 SNP (see Figure 1b). All the p‐values shown were for the effect of the individual SNPs/variants derived from multiple linear regression analysis with age and gender (for the combined cohort) or age only (for the individual gender sub‐cohorts) as the covariate(s) (fully described in Table S4).

While the recessive model did not show any significant association between rs12979860 and globulin levels (p=0.05), an additive model showed a significant association after adjusting for age (Figure 3d). The additive models for the other three phenotypes shown in Figure 3 did not improve on the significance level (data not shown) showed by dominant or recessive models tested, supporting our assumption that the latter models have better power in detecting a significant association with IFNL genetic variants.

When we tested for absence or presence of one or two copies of P70 IFN‐λ4 (P70‐1 or P70‐2) or one copy of S70 IFN‐λ4 (S70‐1) (there was only one individual with two copies of S70 variant allele and was excluded from the analysis and there were ten individuals in the combined cohort, seven individuals in the males and three in the females sub‐cohort that were also homozygous for the ΔG allele at rs368234815 and heterozygous at the rs117648444 position and hence had both S70 and P70 variant copies in them as shown in Figure 4e; this strategy was adopted to have better power in the analysis involving the S70 group as the sample sizes with only one copy of S70 variant was low due to the lower frequency of the minor allele in our population) based on the genotype data at SNPs rs12979860 (inferred for rs368234815) and rs117648444, we saw that P70‐2 carrying males had significantly lower uric acid levels than no‐IFN‐λ4 or P70‐1‐carrying individuals (Figure 4a). However, with monocyte % levels, P70‐1 showed a significant association with the no‐IFN‐λ4 genotype (Figure 4b). With both Triglyceride and VLDL, having two copies of P70 proved beneficial in maintaining significantly lower levels of the lipids when compared to having no‐IFN‐λ4 or one copy of P70 (Figure 4c); even though the combined group had significant p‐values, they seem to be driven by the male sub‐cohort as females did not show such an association. In both these lipid profiles or in monocyte or uric acid levels, the S70 variant does not seem to have any significant association with the other groups (Figure 4a‐c); however, in ALP levels, the low‐activity S70 variant showed significant association with the P70 variant (one copy) in the combined group and the association further strengthened only in the male sub‐cohort (Figure 4d). Previous studies have used a different strategy to examine the effect of the weaker S70 variant compared to the stronger P70 variant (Gadalla et al., 2020; Terczyńska‐Dyla et al., 2014). They have compared three groupings instead of the four groupings we have adopted in Figure 4. The P70 variants: both one and two copies, including some one copy S70 variants are placed in a single group and compared against no‐IFN‐λ4 carriers and those that carry only one copy of S70 variant (Terczyńska‐Dyla et al., 2014). We too tried this strategy and tested all the five phenotypes shown in Figure 4 and found that except for monocyte levels (%), in the comparison with no‐IFN‐λ4 vs. P70, no other comparison either in the combined or males cohort showed a significant association after adjusting for age and gender (Table S5). Our strategy therefore has captured the dosage effect of the P70 variant (in case of triglycerides and VLDL levels, Figure 4c) and also the weak vs strong IFN‐λ4 effect (ALP in Figure 4d) among others, better than this alternate strategy adopted in previous reports (Gadalla et al., 2020; Terczyńska‐Dyla et al., 2014) (Figure 4 and Table S5). The discrepancies in results from the two strategies could be related to issues of sample sizes (especially of the S70 carriers) and therefore, statistical power and/or different nature and size of the effects of the weak vs. strong IFN‐λ4 variants.

3.4. Predictor variables of monocyte, SGOT, uric acid, and globulin levels in males using multiple linear regression analysis

We used the significant associations we saw in males for either of the SNPs rs12979860 and rs28416813 with the four phenotypes: monocyte, SGOT, uric acid, and globulin levels (Table 1 and Figure ​Figure3),3), to build a linear regression model to predict the phenotype levels using the 11 different demographic/lifestyle factors shown in Table S3 as covariates (Table 3). Only data from two phenotypes, i.e., monocyte and uric acid levels, were able to yield a linear regression model that had significance based on F‐statistic p‐value (Table 3). However, even in these two phenotypes, the variation seen in the dependent variable that was explained by all the independent variables was =/<10%, suggesting the inefficiency of the model. Nonetheless, the IFNL genetic variants remained significant in their effect on both monocyte and uric acid levels in this multiple regression model. The SNP rs28416813 also had a significant effect on predicting SGOT levels in the multiple regression model, even though the model itself was insignificant.

DGR thanks DST‐INSPIRE Govt. of India for the junior and senior research fellowship; the authors thank Indranil Mukhopadhyay, Indian Statistical Institute (ISI), Kolkata and Aloke Kar, Samsiddhi Bhattacharjee, Anusuya Chakrabarty and Analabha Basu for helpful discussions. This work was supported by the DBT/Wellcome Trust India Alliance Fellowship (grant No. IA/I/17/1/503122) awarded to SC. The authors thank the Director, NIBMG, Kalyani for the facilities and encouragement. SC and DGR would like to dedicate this article to late Prof. Saurabh Ghosh, ISI, Kolkata.