Nature Neuroscience: doi: /nn Supplementary Figure 1. Missense damaging predictions as a function of allele frequency

Transcription

1 Supplementary Figure 1 Missense damaging predictions as a function of allele frequency Percentage of missense variants classified as damaging by eight different classifiers and a classifier consisting of the intersection of classifiers SIFT, PolyPhen-2 HDIV, PolyPhen-2 HVAR, LRT, Mutation Taster, Mutation Assessor, and PROVEAN as a function of minor allele count across 12,332 unrelated individuals from Sweden. Gray and black colors indicate, respectively, variants never observed and variants already observed in the ExAC cohort of 45,376 individuals.

2

3 Supplementary Figure 2 Correlations across ultra-rare variant types counts Relationships between four separate types of coding-sequence and splice-site URV counts across 12,332 unrelated individuals from Sweden: (a) disruptive vs. damaging; (b) disruprive vs. missense non-damaging; (c) damaging vs. missense non-damaging; (d) disruptive vs. synonymous; (e) damaging vs. synonymous; (f) missense non-damaging vs synonymous. Black dots indicate individuals with less than or equal to 100 URVs detected, and red crosses indicate individuals with more than 100 URVs detected. Pearson correlation coefficients are indicated on the top left of each panel.

4 Supplementary Figure 3 Enrichment in schizophrenia cases across ultra-rare variant types Observed enrichment in 4,877 schizophrenia cases compared to 6,203 controls for (a) URVs across all annotations, with non-coding including both intronic and untranslated region variants, (b) missense URVs classified as damaging by classifiers PolyPhen-2 HDIV, PolyPhen-2 HVAR, SIFT, LRT, PROVEAN, FATHMM, Mutation Taster, and Mutation Assessor, as well as missense damaging, inframe indel, protein-protein-contact, splice-acceptor, splice-donor, stop-gained, and frameshift URVs. Enrichment and P values were computed using a linear regression model (left panels) and a logistic regression model (right panels) and correcting for covariates (a) with and (b) without the exclusion of URV individual count. Horizontal bars indicate 95% confidence intervals.

5

6 Supplementary Figure 4 Single gene burden tests for disruptive and damaging variants at different allele frequency thresholds Quantile-quantile plots for burden tests for association with schizophrenia of all Ensembl genes for disruptive and damaging variants. Burden test for association was performed using SKAT software.

7 Supplementary Figure 5 Enrichment of durvs in schizophrenia cases across selected gene sets stratified per variant types and cohorts Burden of durvs across selected gene sets analyzed in this study stratified across disruptive and damaging URVs (left panel) and across previously analyzed exomes and newly available exomes (right panel). Enrichment and P values were computed using a logistic regression model using exome-wide durv count as a covariate to correct for average exome-wide burden (dot-dashed line). Horizontal bars indicate 95% confidence intervals.

8 Supplementary Figure 6 Enrichment of durvs in schizophrenia cases across constrained and unconstrained genes Burden of durvs across loss-of-function intolerant (LoF-intolerant) genes, missense constrained genes and their complementary sets, stratified across disruptive and damaging URVs. Enrichment and P values were computed using a linear regression model (left panels) and a logistic regression model (right panels) and, contrary to most other analyses in this study, without using exome-wide durv count to correct for average exome-wide burden. Horizontal bars indicate 95% confidence intervals. As LoF-intolerant and missense constrained genes were defined outside this study, different enrichments observed within and outside these gene sets cannot be attributed to differential false positive rates between cases and controls.

9 Supplementary Figure 7 Enrichment of durvs in schizophrenia cases across gene sets from different tissues Burden of durvs across 27 tissue expression specific gene sets. Each gene set was generated selecting genes for which expression in a given tissue was at least 5 times the median expression across 27 different human organs and tissues ascertained from 95 individual. Enrichment and P values were computed using a logistic regression model using exome-wide durv count as a covariate to correct for average exome-wide burden (dot-dashed line). Horizontal bars indicate 95% confidence intervals. Brain specific genes were significantly more enriched for durvs in schizophrenia cases than the average gene.

10 Supplementary Figure 8 Enrichment of durvs in schizophrenia cases across gene sets from different brain and neuronal cell types Burden analysis for durvs across genes defined from (a) 11 brain cell types and (b) 3 neuron cell types: excitatory pyramidal neurons (Exc), parvalbumin (PV)-expressing fast-spiking interneurons, vasoactive intestinal peptide (VIP)-expressing interneurons, both for expressed genes and cell-type specific genes. Brain cell type gene sets were generated selecting genes for which log-expression in a given cell type was 0.5 greater than the median log-expression across 11 central nervous system cell types ascertained from developing and mature mouse forebrain. Neuron cell type expressed gene sets were defined as those with more than 50 observed transcripts per million (TPM). Neuron cell type specific gene sets were defined as those observed more than 5 times the minimum expression across the 3 different cell types. Expression profiles for neurons were ascertained from nuclei isolated from adult (8 11 weeks) mouse neocortex. Enrichment and P values were computed using a logistic regression model using exome-wide durv count as a covariate to correct for average exome-wide burden (dot-dashed line). Horizontal bars indicate 95% confidence intervals.

11 Supplementary Figure 9 Enrichment of durvs in schizophrenia cases across different X linked intellectual disability gene sets Burden analysis for durvs across X linked intellectual disability (XLID) genes and developmental disorder genes. X linked ID genes correspond to the union of X linked ID sets of genes as defined from OMIM, GCC, and Chicago. Enrichment was computed separately for males, females, and both groups together. Enrichment and P values were computed using a logistic regression model using exomewide durv count as a covariate to correct for average exome-wide burden (dot-dashed line). Horizontal bars indicate 95% confidence intervals. Larger confidence intervals for males reflect the smaller number of variants observed in males due to carrying half as much X chromosome DNA.

12 Supplementary Figure 10 Average ultra-rare variant types count across sequencing waves Average number of URVs detected in controls and in schizophrenia cases used in this analysis across several classes of variants and sequencing waves. Numbers of controls and cases within each wave is indicated in parentheses. Wave 1 is denoted in red as for this batch, accounting for approximately 1% of the whole cohort, an earlier version of the hybrid-capture procedure was used which captured approximately 10% less of the exome. Vertical bars indicate 95% confidence intervals.

13 Supplementary Figure 11 Duplicate and first degree relationship estimates across cohort Estimates of percentage of genome shared IBD (PI_HAT) and percentage of genome shared IBD1 (Z1) among all pairs from 12,384 samples for which PI_HAT>.35. Estimates were generated with plink command --genome full.

14

15 Supplementary Figure 12 Ultra-rare variants counts relationships with population stratification Covariates within the Swedish exome cohort: (a) ultra-rare SNP and indel counts across cohort 12,334 individuals; (b) distribution of birth year across cohort individuals; (c) relationship of cohort individuals (in black) with 1000 Genomes project phase 1 individuals with respect to the first two principal components with removed outliers (in red); (d) relationship of cohort individuals with respect to the two main Swedish principal components; (e) relationship between principal components tracking Finnish ancestry and URV count; (f) relationship between principal components tracking Northern-Southern Swedish ancestry and URV count. Black crosses indicate individuals with less than or equal to 100 URVs detected, and red crosses indicate individuals with more than 100 URVs detected.

16 Supplementary Figure 13 Gender estimates from genotypes X chromosome inbreeding coefficient (F) and Y-chromosome non-missing genotype calls (YCOUNT) for the entire cohort of 12,384 samples. Male individuals inferred with 47,XXY karyotype (Klinefelter syndrome) are indicated in green and individuals with mismatching reported and genotyped sex are indicated as yellow crosses. Estimates were generated with plink command --check-sex ycount.

17 Supplementary Figure 14 Manhattan plot for association with schizophrenia of common exome variants Manhattan plot for association of exonic variants with schizophrenia phenotype using a logistic regression model with sex and first five principal components as covariates. Variants with P value less than 10-6 include variant rs on chromosome 2 in the UTR5 of genes TYW5 and C2orf47, and seven variants in the MHC region around the HLA genes.

18 Supplementary Figure 15 Missense predictors comparisons for association with schizophrenia P values for enrichment of missense URVs classified as damaging by all 256 possible predictors defined as the combination of damaging definitions from any subset of Polyphen2_HDIV, Polyphen2_HVAR, SIFT, LRT, PROVEAN, FATHMM, Mutation Taster, and Mutation Assessor algorithms. P values were computed using a linear regression model and correcting for covariates including total URV count. Predictors including FATHMM performed significantly worse than predictors excluding FATHMM while 25 predictors performed better than the predictor chosen for the analyses in this manuscript (bright red). Both predictors including all algorithms but FATHMM (bright red) and including LRT, MutationTaster, PolyPhen2 HDIV, PolyPhen2 HVAR, and SIFT (green) performed better than each predictor based on a single algorithm.

19 Covariate URV variance explained (%) durv variance explained (%) URV (Not used) Sex 0.02 NA Birth year Exome capturing kit 0.22 NA 1 st principal component 0.01 NA 2 nd principal component 0.15 NA 3 rd principal component 7.22 NA 4 th principal component 0.39 NA 5 th principal component NA 6 th principal component 1.10 NA 7 th principal component 0.28 NA 8 th principal component NA NA 9 th principal component NA NA 10 th principal component NA NA 11 th principal component 0.12 NA 12 th principal component NA NA 13 th principal component NA NA 14 th principal component NA NA 15 th principal component NA NA 16 th principal component 0.03 NA 17 th principal component 0.01 NA 18 th principal component NA NA 19 th principal component 0.02 NA 20 th principal component NA NA Supplementary Table 1 Variance in number of ultra-rare variants uniquely explained by each covariate Variance of the total number of URVs and durvs across individuals uniquely explained by each covariate used for linear regression and logistic regression analyses. All covariates combined explained 34.68% and 28.87% of the overall variability in, respectively, URV and durv count. Principal components 3 and 5 correlate with, respectively, the amount of Finnish ancestry and the amount of Northern Swedish ancestry and accounted alone for 31.21% of the variability in URV count and 8.49% of the variability in durv count. Including year of birth accounted for an additional 0.25% of the variability in URV count, possibly because younger individuals tended to have more exotic ancestry or because younger individuals recent variants were less likely to be shared with individuals from the rest of the cohort due to slightly longer coalescent paths.

20 Gene symbol P value durvs in cases durvs in controls bp length KL ,039 STXBP ,782 PCYT1A ,104 CPVL ,431 KDM5B ,635 HEPHL ,480 AP5Z ,424 TTN ,053 CSNK1G ,368 KCNH ,591 ZDBF ,065 ITPR ,232 PRSS ,545 CCDC ,352 LAMB ,397 PCSK ,910 PRDM ,483 KIAA ,522 NDST ,652 TMPO ,365 KIAA ,272 CDO OGDHL ,033 SORCS ,540 KDM5C ,683 PLEKHG ,140 JARID ,741 ARFGEF ,550 ASAH ,188 PMS ,589 TRIM ,235 TRAF3IP ,076 GOLGA ,693 MPP ,914 KLHL ,770 WBSCR QRICH ,331 ISYNA ,677 VSTM2A DBNL ,320 THG1L LY ,169 LY75-CD ,622 DNM1L ,211 CHRNE ,482 MTAP TBC1D2B ,892 Supplementary Table 2 Genes with strongest results for association of durvs with schizophrenia List of genes with durvs burden test statistically significant at a = Number of durvs observed in schizophrenia cases (4,877 individuals) and controls (6,203 individuals), as well as coding base pairs length of the gene, are reported.

21 Class Gene set URVs Disruptive m.a.c.=1 (no URVs) m.a.c. 5 (no URVs) m.a.c. 10 (no URVs) m.a.f.<0.1% (no URVs) m.a.f.<0.5% (no URVs) LoF-intolerant Potentially Synaptic All LoF-intolerant Damaging Potentially Synaptic All LoF-intolerant Disruptive and damaging Potentially Synaptic All Supplementary Table 8 Results for non ultra-rare variants association with schizophrenia across key gene sets P value for burden tests using SKAT software across variation classes disruptive, damaging, and combined, across genes that are loss-of-function intolerant (LoF-intolerant), potentially synaptic, and all genes, and across URVs and non-ultra-rare variants up to minor allele count (m.a.c.) 1, 5, and 10 and up to minor allele frequency (m.a.f.) 0.1% and 0.5%. Differently to other gene sets analysis in the rest of the manuscript, these analyses were not corrected for exome-wide enrichment of the same class of variants.