A trans-ethnic Mendelian randomization study identifies causality of obesity on risk of psoriasis

TO THE EDITOR

Psoriasis is a chronic disorder characterized by cutaneous and systemic manifestations. Epidemiological studies have reported increased comorbidity of psoriasis with numerous complex diseases such as metabolic clinical measurements (Greb et al., 2016; Naito et al., 2016). However, interpretation of the comorbidity remains controversial to date, since causal inference between correlated phenotypes is difficult when depending solely on epidemiological studies. Identification of causal inference between correlated phenotypes has significant clinical impacts, as modification of the causal phenotypes could benefit treatment of the outcome phenotypes. Drugs indicated by the causal phenotypes could also be promising targets of drug repositioning for the outcome phenotypes (Holmes et al., 2017). Therefore, alternative approaches to strengthen causal inference on psoriasis are warranted.

A rising approach is to utilize genetic data for this purpose (Pingault et al., 2018). Genetically-determined phenotype profiles are robust to confounding factors acquired during a lifetime, which could be interpreted as ideal randomization of subjects. Mendelian randomization (MR) is an approach to infer causal inference between phenotypes using the GWAS results (Holmes et al., 2017; Hemani et al., 2018). Due to (i) achievement of large-scale GWAS of a variety of human phenotypes with public data deposit, and (ii) development of MR analytical methods robustly infer causality such as MR-Egger (Burgess et al., 2017), MR is now one of the best approaches to infer causality. Generally, the largest available GWAS result within a single ancestry are used for a MR analysis to afford robust conclusions. Thus, confirmation of the MR analysis results requires additional validation using GWAS with independent ancestry.

Here, we conducted a trans-ethnic MR analysis to estimate causal inference on psoriasis. We obtained the genome-wide summary statistics of the previously reported psoriasis GWAS of the European populations (13,229 cases and 21,543 controls; Tsoi et al., 2017) and Japanese population (282 cases and 426 controls; Hirata et al., 2018; Supplementary Table 1). Both of the psoriasis GWAS were conducted by applying whole-genome sequencing (WGS)-based genotype imputation, which yielded high coverage of the genome-wide variants suitable for the MR analysis (> 6,000,000 variants). As for the European GWAS, the subjects in the cohort collected by 23andMe were excluded due to their policies on summary data sharing. We thus re-conducted the GWAS meta-analysis after sample exclusion for our MR analyses. We admit that the sample size of the Japanese GWAS was relatively smaller, which warrants further accumulation of the subjects.

We focused on metabolic clinical measurements as exposure phenotypes, where the large-scale GWAS results have been released in both populations. We selected in total nine measurements: obesity (body mass index [BMI]), triglyceride, total cholesterol, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, blood sugar, hemoglobin A1c (HbA1c), systolic blood pressure, and diastolic blood pressure (Supplementary Table 2) (Ehret et al., 2011; Scott et al., 2012; Willer et al., 2013; Locke et al., 2015; Wheeler et al., 2017; Akiyama et al., 2017; Kanai et al., 2018; on average 149,958 subjects per trait). After selection of the lead variants (or the proxy single nucleotide polymorphisms [SNPs] in linkage disequilibrium [LD] of r²≥0.5 in the corresponding 1000 Genomes Project phase3v5 populations) at the loci with genome-wide significance threshold (P<5.0×10⁻⁸) and exclusion of the highly pleiotropic locus of the major histocompatibility complex (MHC) region, on average 41.3 loci per trait were obtained.

We adopted two-sample MR, one of the MR analysis approaches that handles summary statistics obtained from separate studies. In addition to the typical method of inverse variance weighted (IVW), we adopted MR analysis based on Egger regression (i.e., MR-Egger), which is statistically less powerful but more robust to bias caused by directional pleiotropy (Burgess et al., 2017). We used the MR-Base platform implemented as a package of R statistical software (Hemani et al., 2018).

For Europeans, significant causality of genetically increased BMI on risk of psoriasis was estimated (β=0.464 and P=3.1×10⁻⁵ in IVW, and β=0.697 and P=0.0093 in MR-Egger; Table 1 and Figure 1). For Japanese, significant causality of BMI on psoriasis risk was also observed in the IVW analysis with a concordant directional effect (β=1.275 and P=0.0069). While the MR-Egger result was not significant (β=1.499 and P=0.27), the effect size estimate was larger than that of IVW. Because phenotype normalization methods were different among the original GWAS, direct comparison of the estimated causal effect sizes of the BMI-associated SNPs on psoriasis between the studies was difficult. We assessed potential bias in the results of MR analyses mostly caused by SNP pleiotropy, by applying a set of sensitivity analyses including heterogeneity test, leave-one-out analysis, and funnel plots implemented in the MR-Base platform, but did not find existence of apparent bias (Supplementary Figure 1). We applied the reverse MR analysis inferring causality of psoriasis on BMI, but did not observe significant causality (P>0.75; Supplementary Figure 2).

Table 1.

Results of the trans-ethnic Mendelian randomization (MR) analyses inferring causality of the clinical metabolic measurements on psoriasis

Trait	MR method	MR result in Europeans		MR result in Japanese		Meta-analysis
		Beta (SE)	P	Beta (SE)	P	P
Body mass index	Inverse variance weighted	0.464 (0.112)	3.1×10−5	1.275 (0.472)	0.0069	1.2×10−6
	MR-Egger	0.697 (0.262)	0.0093	1.499 (1.342)	0.27	0.0088
Triglyceride	Inverse variance weighted	0.085 (0.088)	0.33	0.139 (0.369)	0.71	0.34
	MR-Egger	−0.064 (0.129)	0.62	0.748 (0.561)	0.19	0.56
Total cholesterol	Inverse variance weighted	0.054 (0.063)	0.39	−0.595 (0.476)	0.21	0.78
	MR-Egger	0.126 (0.115)	0.28	−0.663 (0.900)	0.47	0.80
HDL cholesterol	Inverse variance weighted	−0.026 (0.072)	0.72	0.190 (0.318)	0.55	0.87
	MR-Egger	0.024 (0.109)	0.82	−0.048 (0.509)	0.93	0.92
LDL cholesterol	Inverse variance weighted	0.051 (0.054)	0.35	−0.300 (0.385)	0.44	0.91
	MR-Egger	0.180 (0.088)	0.046	−0.134 (0.535)	0.80	0.22
Blood sugar	Inverse variance weighted	−0.537 (0.269)	0.046	0.235 (0.671)	0.73	0.24
	MR-Egger	−0.685 (0.943)	0.47	1.940 (2.550)	0.46	0.99
HbA1c	Inverse variance weighted	−0.143 (0.273)	0.60	−0.289 (0.417)	0.49	0.39
	MR-Egger	0.064 (0.552)	0.91	1.811 (1.661)	0.29	0.41
Systolic blood pressure	Inverse variance weighted	0.006 (0.010)	0.54	1.080 (0.781)	0.17	0.16
	MR-Egger	0.048 (0.033)	0.16	3.818 (2.823)	0.19	0.055
Diastolic blood pressure	Inverse variance weighted	0.017 (0.017)	0.31	1.010 (0.963)	0.29	0.14
	MR-Egger	0.099 (0.058)	0.10	1.560 (3.043)	0.62	0.13

Figure 1. Regression plots of the body mass index (BMI)-associated variants on psoriasis risk.

Dots represent the BMI-associated single nucleotide polymorphisms (SNPs) plotted along with effect size estimates on BMI (x-axis) and psoriasis risk (y-axis) with 95% confidence intervals in the European populations (a) and the Japanese population (b). Regression lines obtained from the Mendelian randomization (MR) analyses are plotted in red (by inverse variance weighted [IVW]) and blue (by MR-Egger).

We then conducted a trans-ethnic meta-analysis of the MR results, using weighted summation of z-scores considering directional concordance of the causal effect estimates. As expected, significant causality of BMI on risk of psoriasis was observed (P=1.2×10⁻⁶ in IVW and P=0.0088 in MR-Egger). While suggestive relationships of LDL cholesterol and blood sugar were observed in Europeans (P<0.05 in IVW or MR-Egger), these associations were not replicated in Japanese, and trans-ethnic MR analyses did not indicate significant causality of these traits (P>0.22).

By utilizing a large-scale GWAS results of psoriasis in European and Japanese populations, our trans-ethnic MR analyses identified a causal link of obesity on risk of psoriasis. This study has a value as one of the initial successful examples of the trans-ethnic MR analysis. Epidemiological studies have pointed a link between higher BMI and increased incidence or severity of psoriasis (Greb et al., 2016; Naito et al., 2016), with which our MR analysis results were concordant. Moreover, because our study validated causality (i.e., directional link) of obesity on psoriasis, interventional improvement of obesity itself could be a promising treatment strategy towards better management of psoriasis. Further application of the MR analysis of psoriasis to a wider range of phenotypes, as well as validations in additional ethnicities, is warranted.