Drug enrichment and discovery from schizophrenia genome-wide association results: an analysis and visualisation approach

Using successful genome-wide association results in psychiatry for drug repurposing is an ongoing challenge. Databases collecting drug targets and gene annotations are growing and can be harnessed to shed a new light on psychiatric disorders. We used genome-wide association study (GWAS) summary statistics from the Psychiatric Genetics Consortium (PGC) Schizophrenia working group to build a drug repositioning model for schizophrenia. As sample size increases, schizophrenia GWAS results show increasing enrichment for known antipsychotic drugs, selective calcium channel blockers, and antiepileptics. Each of these therapeutical classes targets different gene subnetworks. We identify 123 Bonferroni-significant druggable genes outside the MHC, and 128 FDR-significant biological pathways related to neurons, synapses, genic intolerance, membrane transport, epilepsy, and mental disorders. These results suggest that, in schizophrenia, current well-powered GWAS results can reliably detect known schizophrenia drugs and thus may hold considerable potential for the identification of new therapeutic leads. Moreover, antiepileptics and calcium channel blockers may provide repurposing opportunities. This study also reveals significant pathways in schizophrenia that were not identified previously, and provides a workflow for pathway analysis and drug repurposing using GWAS results.

. Association between druggable genes and schizophrenia in SCZ-PGC2. ( a ) Gene Manhattan plot for 4298 druggable genes, showing the schizophrenia SCZ-PGC2 association (-log 10 (p-value)) as a function of chromosomal position. The gene with the lowest p-value for each chromosome was annotated. The red line indicates the Bonferroni threshold at α = 5%. ( b ) Significant druggable genes (Bonferroni threshold) belonging to the same gene families, with at least 3 significant genes. Figure S2. Association (-log 10 (p-value)) of voltage dependent calcium channels and neurotransmitter receptors with schizophrenia (SCZ-PGC2), with Bonferroni significance threshold: ( a ) voltage-dependent calcium channels, ( b ) nicotinic receptors, ( c ) dopamine receptors, ( d ) serotonin receptors, ( e ) GABA receptors, ( f ) glutamate receptors, ( g ) epinephrine receptors, and ( e ) opioid and somatostatin receptors. Figure S3. Region LocusZoom plots 1 of ( a ) OPRD1 , ( b ) GABBR2 , and ( c ) NOS1 with +/-500Kb windows (NBI build 37). The top SNP is indicated in purple and the colour of all other SNPs is representative of the pairwise r 2 using LD data from 1000 Genomes EUR. Figure S4. STRINGdb 1 protein-protein interaction (PPI) network of 123 druggable genes significant in SCZ-PGC2.  Text S1: Drug gene-sets from K i DB and DGIdb .
Drug gene-sets were extracted from K i DB and DGIdb drug/gene interaction databases. We applied several filters listed in Tables 1-2, and merged the two databases (cf. Table 3).

Number of interactions 32,108
Number of gene-sets 10,922

Number of unique gene-sets 3,622
Number of unique gene-sets of size ≥ 2 2,423 Degenerescence (molecules/gene-set) 3.02 Table 3. Merging K i DB and DGIdb.

Number of interactions 35,098
Number of gene-sets 14,917

Number of unique gene-sets 3,939
Number of unique gene-sets of size ≥ 2 2,737 Degenerescence (molecules/gene-set) 3.79

Text S2: Pathway analysis in MAGMA.
The gene association vector Z with elements can be used in a regression , Z …Z Z gene 1 gene 2 gene n model for each pathway p 1 x x x Z = α p → + β 1p 1p + β 2 2 + β 3 3 + … + ε They are two types of pathway analysis: self-contained and competitive. The self-contained analysis tests whether a pathway is associated or not with the trait; the competitive analysis tests whether genes in the pathway are more associated than other genes. If all parameters are equal β to 0, and if only Z values within pathway p are taken into account ( ), the model is Z p intercept-only and a p-value can be obtained by testing against the null hypothesis α p > 0 . This is the self-contained p-value, testing whether the mean gene association value α p = 0 α p within the pathway is significantly above 0. If, instead, all parameters are not equal to 0 β (competitive analysis), reflects the difference between the gene associations within and β 1p outside the pathway, and is a binary vector with i th element = 1 if the i th gene is within x 1p pathway p and = 0 otherwise. The competitive p-value is obtained by testing (better β 1p > 0 association within the pathway) against the null hypothesis (no difference in association β 1p = 0 within or outside the pathway).
In MAGMA, other variables ( are used to account for gene size, gene density, minor , x …) x 2 3 allele count, and the log of those values. The gene density is the ratio of gene size to the number of SNPs in the gene. Because of the LD between genes the errors may be correlated, therefore a generalized least squares approach is adopted where residuals have the variance , where Σ σ 2 Σ is the gene correlation matrix.
Text S3: Hub genes in PPI network.
The 123 significant druggable genes were used to build two STRINGdb protein-protein interaction subnetworks. One subnetwork, d , was only constructed with the 123 genes (cf. Supplementary Figure S4 ); the other, g , was constructed with the 123 genes as well as all protein-coding genes significant in schizophrenia, for a total of 498 genes. We used normalized node betweenness and node degree measure to find hub genes, averaged between values drawn from d and g subnetworks. The betweenness, normalized betweenness, and degree of each node in the network was computed using the igraph R package.
Betweenness b of a node (gene) is the sum of the fraction of shortest paths that pass through the node. The normalized betweenness is computed as b' = 2* b /( n * n -3* n +2), where n is the number of nodes.
A normalized degree measure was computed for each node in the network, by dividing the degree n of a node in the subnetwork by the degree in the complete network. The normalized degree measure in the significant druggable subnetwork d is obtained by dividing by the degree in the whole druggable network D , and the value for the significant protein-coding subnetwork g is obtained by dividing by the degree in the whole protein-coding network G .