The GWAS summary statistics for psoriasis were derived from data published by FinnGen Consortium R9. This study used the “psoriasis” phenotype. The GWAS of psoriasis included 373,338 Finnish adult subjects, including 9,267 cases and 364,071 controls, which excluded subjects with other cancers. Age, sex, top 10 major components, and genotyping batches were corrected during analysis. As a genetic instrumental variable, the GWAS summary data of CM came from the two-stage genome-wide meta-analysis conducted by Matthew et al. (17) that included 13 GWAS datasets from Europe, Australia, the United Kingdom, Athens, and the United States encompassing a total of 15,990 melanoma cases and 26,409 controls.

We use the following five criteria to select instrumental variables (IV). First, the SNP–phenotype association level must reach a genome-wide significance threshold (P< 5*10⁻⁸). Second, the linkage disequilibrium between all SNPs is based on the European 1000 Genomes Project reference panel. Third, SNPs with a secondary allele frequency (MAF) of ≤ 0.01 were removed. Fourth, among the SNPs where R²< 0.001 (clumping window size = 10,000 kb), only those with the lowest P value were retained. Finally, when palindromic SNPs existed, we used the allele frequency information to infer the forward-strand alleles. We used a curated genotype-phenotype database (PhenoScanner) to search for associations between variants used to detect each instrumental variable and other traits that may represent pleiotropic pathways; specific examples are traits associated with hypertension and telomere length, which are recognized risk factors for cutaneous melanoma (18, 19). Variants associated with these and other traits were excluded from sensitivity analysis (using a software default threshold of P< 5 × 10 – 8).

In the present study, we evaluated the causal relationship between psoriasis and CM using various methods, including inverse variance weighting (IVW), MR–Egger regression, MR Pleiotropy RESidual Sum and Outlier (MR-PRESSO), weighted mode, and weighted median. IVW, also known as the inverse variance weighting method, is characterized by ignoring the existence of an intercept term in regression and using the reciprocal of outcome variance (se²) as the weight for fitting. In the absence of heterogeneity and horizontal pleiotropy, IVW estimates are the most reliable (20). MR–Egger regression is conducted under the assumption of InSIDE (instrument strength is independent of the direct effect), which enables us to assess the presence of poly-efficacity with intercept terms. The intercept term being 0 implies that horizontal pleiotropy is absent, and the result of MR–Egger regression is consistent with IVW (21). In addition, MR–Egger regression is also used to judge whether there is horizontal pleiotropy or not. When more than 50% of the instrumental variables are invalid, the weighted median method can estimate causality more correctly. The weighted model estimation method can better detect the causal effect, has less deviation, and a lower class I error rate than the MR–Egger regression when the InSIDE hypothesis is not met (22). In addition, significant outliers were detected using MR-PRESSO (23) tests and MR–Egger regression, and horizontal pleiotropic effects were corrected by removing outliers. We used the global test to check if horizontal pleiotropy among all instruments existed (23). Moreover, we applied Cochran’s Q-test statistics to further examine the heterogeneity among all SNPs. We identify potentially heterogeneous SNPs by conducting a “leave-one-out” analysis by excluding each instrument SNP in turn. Finally, we also performed reverse MR analysis of psoriasis and CM. We adopted methods and settings that were consistent with the forward MR.

The F statistic was calculated by the formula F = ( R 2 1 − R 2 × ( n − k − 1 ) k ) (where R² is the proportion of the variance of the trait explained by the SNP, k is the number of IVs, and N is the sample size of the GWAS of the SNP with the trait), which is used to quantify the strength of the instrument, and a value > 10 is considered sufficient. We used the “TwoSampleMR” R package to calculate the coefficient of determination (R²) of exposure to genetic variants. The R² value was estimated using the formula R 2 = 2 × E A F × ( 1 − E A F ) × ( β ) 2 (where EAF is the effect allele frequency [EAF] of SNP, SD is the standard deviation, and β is the estimated effect size of SNP on the trait). Based on the online MR-power calculation tool (https://sb452.shinyapps.io/power/) (24), we calculated the power of the MR estimates.

All analyses were performed in R (version 4.2.2; R Foundation for Statistical Computing, Vienna, Austria), and MR analysis was conducted based on the TwosampleMR (25) (version 0.5.6) and MR-PRESSO (version 1.0) R packages.

As shown in Table 1 , we identified a total of 31 independent SNPs as instrumental variables in psoriasis, explaining 29.22% of the total variation. The F-statistic range for SNPs was 31.9–1018.0, indicating that SNPs explain the potency strength of exposure effectively. Figures 1A, B show the effect of each SNP locus on CM. Furthermore, the results of IVW estimation suggest that psoriasis is an independent risk factor for CM (odds ratio [OR], 1.69; 95% confidence interval [CI], 1.15–2.48; P = 0.025; Figures 1A, C ). In addition, the results of the other three estimate methods, including weighted mode (OR, 2.42; 95% CI, 1.16–5.06; P = 0.026), MR–Egger regression (OR, 1.71; 95% CI, 1.06–2.76; P = 0.038), and weighted median (OR, 1.77; 95% CI, 1.09–2.87; P = 0.043), further validated the results of IVW estimation.

Cochran’s IVW and MR–Egger Q test showed no significant heterogeneity among these IVs ( Table S1 ). Moreover, no significant change in the estimated causal effect was observed when we excluded each SNP individually during the “leave-one-out” analysis ( Figure 1D ). Therefore, the estimated effect cannot be explained by any single SNP. In addition, according to the MR–Egger regression intercept analysis ( Supplemental Table 2 ) and MR-PRESSO global test ( Supplemental Table 3 ), there was no significant horizontal pleiotropy. After setting the type I error rate as 0.05 and the outcome variable as binary, we obtained an MR-power value of 100%. The above results prove that the results of our MR analysis are reliable and have high efficiency.

We obtained seven SNPs (P< 5 * 10^-8, R²< 0.001) significantly associated with CM from a meta-analysis of large GWASs, which explain 3% of the total variation. Summaries and details of each SNP are provided in Supplementary Table 4 . The F-statistic range for SNPs was 30.1–49.5. The individual SNP effects and the combined effects of each method are shown in Figures 2A, B . The results of the four MR estimate methods all suggested that no association was observed between genetically determined CM and psoriasis ( Figure 2C , P > 0.05). No notable heterogeneity was detected by Cochran’s Q statistics ( Supplementary Table 5 , P > 0.05). The MR–Egger intercept test ( Supplementary Table 6 ) and MR-PRESSO global test ( Supplementary Table 7 ) suggested no evidence of pleiotropy. Moreover, no significant change in the estimated causal effect was observed when we excluded each SNP individually during the “leave-one-out” analysis ( Figure 2D ).

This study had several advantages. To determine the causal relationship between psoriasis and CM without the interference of confounding factors, we performed bidirectional MR analysis. Bidirectional Mendelian randomization analysis is an extension of standard Mendelian randomization analysis. Compared with the latter, the former can investigate whether there is a two-way causal relationship between exposure variables and outcome variables. In addition, genetic variation in CM was derived from the largest available GWAS meta-analysis, ensuring the strength of instruments in MR analyses. We detected and excluded horizontal poly-tropism using MR-PRESSO and MR–Egger regression intercept tests. Finally, there was no overlap between the GWAS summary data for psoriasis and the meta-analysis GWAS summary data for CM, which led to low heterogeneity, low bias, and high accuracy in our study.

We also acknowledge some limitations. First, most of the data used in the analysis mainly came from individuals with European ancestry, which restricts the applicability of our findings to other ethnic groups. Second, due to the lack of detailed clinical data, we were unable to conduct a subgroup analysis, for example, to distinguish between psoriasis vulgaris, arthritic psoriasis, pustular psoriasis, and erythrodermic psoriasis; explore non-linear relationships; or to sort by the severity of psoriasis. A population-based study in Denmark suggested a modestly increased risk of melanoma in patients with mild psoriasis and no increased risk in patients with severe psoriasis and psoriatic arthritis (1). Due to the lack of GWAS data on psoriasis severity classification or strictly defined clinical classification to date, we were unable to use MR methods to study the effect of different severities or different clinical subtypes of psoriasis on CM. Notably, each method used in our analysis had advantages and disadvantages. However, the use of many methods with different assumptions may lead to inconsistent or conflicting outcomes and obscure the conclusions.