Table S1. Genomic position and AUC ROC for smoking status predicted by methylation level of each CpG locus.

Table S2. Power analysis to test the different PRSs for association with smoking status in SzP and controls, estimated by AVENGEME. Values used in calculations as well as estimated power are shown.

Table S3. Percentage of variance explained of smoking status (and 95% CI) on the liability scale corrected for ascertainment bias24 in SzP and controls by the different PRSs.

Table S4. Effect of different smoking prevalence values on R2 on the liability scale and corrected by ascertainment bias. Prevalence values from 10 percentage points lower to 10 percentage points higher than those presented in main text were tested. Bold letter indicates the values presented in main text, based on the prevalence determined in the CLAMORS study25.

Table S5. Results of association analysis between cross-disorder PRS and smoking status in SzP and controls.

Table S6. SNP heritability and genetic correlations, estimated by using LD Score regression, used for calculation of multitrait PRS by SMTpred. Genetic correlations are shown above the diagonal, significance of the genetic correlations is shown below the diagonal, and SNP heritabilities on the diagonal.

Table S7. Statistical power and results of association analysis between SmkInit PRS and schizophrenic status.

Figure S1. Determination of smoking status by methylation analysis. A) Correlation of methylation levels among the 8 CpG loci. All correlations were significant at the Bonferroni's correction level (P < 1.79 x 10-3, corrected for 28 tests). The CpG loci are named following the nomenclature of Tian et al1 B) Determination of optimal cut-off point for Youden's index value. C) Methylation level distribution of smokers and non-smokers in the training sample. The solid vertical line represents the optimal cut-off point, and the dashed vertical lines represent the 95% CI of the cut-off point, delimiting the gray zone.

Figure S2. Percentage of variance explained of smoking status (and 95% CI) in SzP and controls by SmkInit PRS differentiating between individuals whose smoking status was determined by clinical interview or methylation analysis.

Figure S3. Correlation between the PRSs estimated with thresholding and with EB-PRS for SmkInit. All correlations were significant at the Bonferroni-correction P < 1.01 x 10-4 level (corrected for 496 tests).

Figure S4. Post-hoc power analysis for the interaction test in logistic regression. Data was simulated from SzP and controls, giving our sample size and proportion of SzP and smokers by group, considering a model with two independent variables, a normal variable, PRS, and a dichotomous variable, schizophrenic status. Beta coefficient for PRS was set to the value for ADHD PRS estimated in univariate analysis in our data. The results were not affected by using beta coefficients of other PRS for main effects (data not shown). This process was replicated 5,000 times using different values of the interaction effect, that is, the absolute difference in beta coefficients of PRS in SzP versus controls. Power was estimated as the proportion of significant results at the PRS x schizophrenic status interaction term among all replicates. Vertical lines correspond to the observed values in our data for the interaction effect between schizophrenic status and the corresponding PRS (Table 2).