Comprehensive integrative analyses identify GLT8D1 and CSNK2B as schizophrenia risk genes

Recent genome-wide association studies (GWAS) have identified multiple risk loci that show strong associations with schizophrenia. However, pinpointing the potential causal genes at the reported loci remains a major challenge. Here we identify candidate causal genes for schizophrenia using an integrative genomic approach. Sherlock integrative analysis shows that ALMS1, GLT8D1, and CSNK2B are schizophrenia risk genes, which are validated using independent brain expression quantitative trait loci (eQTL) data and integrative analysis method (SMR). Consistently, gene expression analysis in schizophrenia cases and controls further supports the potential role of these three genes in the pathogenesis of schizophrenia. Finally, we show that GLT8D1 and CSNK2B knockdown promote the proliferation and inhibit the differentiation abilities of neural stem cells, and alter morphology and synaptic transmission of neurons. These convergent lines of evidence suggest that the ALMS1, CSNK2B, and GLT8D1 genes may be involved in pathophysiology of schizophrenia.

1. The brain eQTL data mainly used in this study included 193 normal human subjects, which is relatively smaller than a recent publication (Mads et al. Am J Hum Genet. 2017 Jun 1;100(6):885-894.), which integrated GWASs with 24 studies of expression quantitative trait loci. In addition, PGC2 schizophrenia dataset is also one of the studied 57 GWASs in the Mads et al's report. It will be better for the authors to refer to these studies and discuss the advantages and shortcomes of the current study.
2. I am wondering whether the authors have western blotting results after transfections the RNAi vectors in the cell-based assay, which might provide clearer evidences for the related downstream pathways after knocking-down GLT8D1.
Reviewer #4 (Remarks to the Author): Peng et al performed GWAS studies and identified 10 candidate genes that contribute to schizophrenia risk. Of these 10 candidate genes, the authors identified ALMS1, CSNK2B and GLT8D1 as the most promising candidates and validated their expression in the human schizophrenia patient cases. Furthermore, they demonstrated that GLT8D1 knockdown using shRNA promotes the selfrenew and proliferation abilities of cultured neural stem cells (NSCs) and inhibits NSCs differentiation into astrocytes and neurons, providing the evidence of the potential role of identified risk genes in neurodevelopment. Overall, this study identified novel schizophrenia risk genes from GWAS analysis, validated their expression and hippocampal structures in human schizophrenia patient cases. However, I have major concerns of the functional studies they performed in this study. The points are listed as below:

Line 152: PPI analysis, What is PPI?
Response: We apologize for the confusion, PPI is abbreviation of protein-protein interaction, and we have clarified this in the revised manuscript. Page 11, line 18.

Line 170: Network-based prioritization. How is this analysis corrected for?
Response: We thank the reviewer for pointing this out. Following the reviewer's suggestion, we provided more detailed information about the correction of network-based prioritization in the revised manuscript. Page 13, lines 19-22, Page 14, lines 1-3, Page 14, lines 8-11. 7. I see that there are a lot of non-independant analysis performed and then used as validation for the other ones. This does not seem to be completely valid. Could the authors find the same results if they only applied non-related (but complementary) analysis approaches ?
Response: We appreciate the reviewer for this valuable suggestion. According to the reviewer's suggestion, we only included the non-related (but complementary) analysis approaches and performed the integrative analysis again. The detailed procedures are as follows: First, we performed Sherlock analysis twice by using independent brain eQTL datasets, and only retained the result from the initial stage for subsequent prioritization. Second, we used an independent integrative approach (i.e., SMR) and performed integrative analysis on the same eQTL dataset, following the reviewer's suggestion, we also excluded the result form SMR and performed prioritization. Therefore, results from step 5 and step 6 in Supplementary Figure S1 were excluded from prioritization. Consistent with our original findings, we obtained the same results (i.e., ALMS1, CSNK2B and GLT8D1 are the risk genes with highest prioritization scores). More detailed information about prioritization can be found in Supplementary Material. Supplementary Figure S1. Response: We thank the reviewer for pointing this out. According to the reviewer's suggestion, we have corrected this in the revised manuscript. Page 24, lines 3-6.
11. Are the brain tissue expression comparisons adjusted for multiple testing (multiple genes in multiple tissues ? Also, sample size of this group is quite small (19 per group or 34 per group?) what a priori power is there to find these effects ?
Response: We understand the reviewer's concern. Brain tissue expression comparisons were not adjusted for multiple testing. However, convergent lines of evidence suggest dysregulation of ALMS1 gene in schizophrenia cases. First, ALMS1 was significantly up-regulated in hippocampus of schizophrenia cases (P=0.006, GSE53987) compared with controls, while CSNK2B (P=0.005, GSE53987) and GLT8D1 (P=0.006, GSE53987) were significantly down-regulated in hippocampus of schizophrenia cases. When corrected for multiple testing (Benjamini-Hochberg procedure), these three genes still showed significant dysregulation in hippocampus of schizophrenia cases. Second, ALMS1 is consistently up-regulated in schizophrenia cases in four independent expression datasets ( Figure  12. where the same brain tissues available for both brain tissue sets? (GSE53987 and GSE12649) If yes, was there any replication of findings?
Response: Both GSE53987 and GSE12649 contain the prefrontal cortex tissues from schizophrenia cases and controls. ALMS1 showed a trend of up-regulation (P=0.056) in GSE53987 ( Figure 3A, middle panel). In GSE12649, ALMS1 was also significantly up-regulated (P=0.019) (Supplementary Figure S5). GSE53987 also includes hippocampal tissues, CSNK2B (P=0.005) and GLT8D1 (P=0.006) were significantly down-regulated in hippocampus of schizophrenia cases ( Figure 3B,C). The expression of CSNK2B and GLT8D1 in the prefrontal cortex tissues did not show significant difference in GSE53987 and GSE12649.

The results presented on line 387 should be presented earlier (as replication of initial results) please revise
Response: We thank the reviewer for this valuable suggestion. Following the reviewer's suggestion, we have revised the manuscript accordingly. Page 21, lines 1-22, Page 22, lines 1-11. 4

It is not clear why the CSNK2B gene is followed further if the expression results are negative for this gene.
Response: We understand the reviewer's concern. To identify genes whose expression change may contribute to schizophrenia risk, we first used Sherlock to integrate genetic associations from GWAS and brain eQTL. Based on the 10 top genes identified by Sherlock integrative analysis, we then used the convergent functional genomics method to prioritize the most promising risk genes for schizophrenia among these 10 genes (Supplementary Figure S1). As stated above, expression of CSNK2B gene was significantly down-regulated in hippocampus of schizophrenia cases compared with controls. In addition, we used evidence from different layers (e.g., PPI, co-expression, association with brain structure and cognitive functions, etc) to prioritize the top genes. That is, evidence from each layer only contributes one point to the final score of each gene. Based on the above evidence, CSNK2B may represent a promising candidate risk gene for schizophrenia.

The CFG analysis appears a bit misleading since it is not clear what information on which genes is being used. If the authors only used the info on the three best candidates described throughout the paper, then is obvious that these should get the best results (circular analysis). If other genes were in included, where are the results of the other analysis ?
Response: We are sorry for the confusion. We first identified genes whose expression change may confer schizophrenia risk using Sherlock integrative analysis. We identified 10 top genes (corrected P <0.05) (Table 1a). To prioritize the most promising risk genes among the 10 top genes, we used CFG analysis to integrate the evidence from different layers. If fact, we included the top 10 genes through the analysis and the score of each gene was calculated at each prioritization step. For clarity, we included a flowchart to illustrate the detailed prioritization procedure (Supplementary Figure S1).

The discussion is a bit confusing. Is not clear to me the rationale to perform all these analysis.
It is clear that many of these steps are not independent from each other and they should be combined.
Response: We are sorry for the confusion. According to the reviewer's comment, we only included the non-related (but complementary) analysis approaches and performed the integrative analysis again. The detailed procedures are as follows: First, as we performed Sherlock analysis twice by using independent brain eQTL datasets, we only retained the result from the initial stage for subsequent prioritization. Second, we used an independent integrative approach (i.e., SMR) and performed integrative analysis on the same eQTL dataset, following the reviewer's suggestion, we also excluded the result form SMR and performed prioritization. That is, results from step 2 and step 3 in Supplementary Figure S1 were excluded from prioritization. Consistent with our original findings, we obtained same results (i.e., ALMS1, CSNK2B and GLT8D1 are the risk genes with highest prioritization scores). More detailed information about prioritization can be found in Supplementary Material. Supplementary Figure S1. 5

What is the variance explained of SCZ, or any of the brain traits tested by the 3 genes discovered by the authors?
Response: We are sorry that we could not calculate the variance explained by the three genes as we could not access the genotype data of the significant SNPs near these three genes. However, in the original paper by PGC2 (1), the authors performed polygenic risk score profiling and they found that about 3.4% SCZ variance can be explained by genome-wide significant loci (assuming independent SNP effects). Considering that 108 independent risk loci have been identified by PGC2, each locus only explains very small variance (3.4%/108=0.031%). Assuming independent SNP effects, the variance explained by these three risk loci is less than 0.094%. Nevertheless, it should be noted this is an estimate, as we could not access the original genotype data, we could not do an accurate calculation.
The paper will benefit tremendously from a figure (of sorts) walking the reader through all the steps done.
Response: We thank the reviewer for this valuable suggestion. Following the reviewer's suggestion, we included a flowchart to illustrate the workflow of this study. Supplementary Figure S1.

Reviewer #2 (Remarks to the Author):
The manuscript by Luo and colleagues performed comprehensive analysis of genetic and genomic data related to SCZ, and discovered three new risk genes. The authors started with integrating brain eQTL and SCZ GWAS data using a statistical tool, Sherlock, and identified 10 candidate genes. With additional analysis, they focused on three of these genes, and supported their relevance in SCZ with a number of lines of evidence, such as temporal expression patterns in brain and association with related psychiatric traits. Overall, the results supporting these genes are quite overwhelming. Most interestingly, GLT8D1 seems a very novel gene whose function was largely unknown before this study. The authors provided clear experiment evidence of this gene using Neural Stem Cell model. Response: We thank the reviewer for the positive comment on our work.

Comments:
The novelty of the findings is not entirely clear. Two of these genes, CSNK2B and GLT8D1, seem to have high-confidence GWAS loci nearby (Table 1). In the published GWAS paper of SCZ, what were the genes assigned to these two loci?
Response: We understand the reviewer's concern. CSNK2B is located in MHC region (Chr 6). Genetic variants in MHC region showed the most significant association with schizophrenia in a recent GWAS of schizophrenia (PGC2) (1). However, due to the high level of linkage disequilibrium in MHC region, it is challenging to pinpoint the causal genes at this locus. A recent study showed that C4A (located in MHC region) may represent an authentic causal gene for schizophrenia at MHC region (2). However, it is unknown whether other genes in MHC region contributes to schizophrenia risk. GLT8D1 is located in Chromosome 3. Genetic variants near GLT8D1 showed genome-wide significant association with schizophrenia (1). Nevertheless, this region also contains several other genes, including  PBRM1, GNL3, GLT8D1, SNORD19, NEK4, ITIH1, ITIH3 and ITIH4.
Since the MHC region contains numerous highly linked genes, no genes were assigned to the MHC region in the published GWAS of schizophrenia (1). And four genes (GNL3, GLT8D1, ITIH1 and ITIH3) were assigned to the loci containing GLT8D1 gene.
Though the PGC2 has reported 108 independent loci associated with schizophrenia (1). However, the reported loci usually span a large chromosomal region and contain many genes. Thus, it is difficult to pinpoint the causal gene (or genes) at the reported loci. In addition, it is not known whether the genetic risk variants contribute to schizophrenia susceptibility through affecting protein function (such as missense and nonsense variants) or regulating gene expression. Our study suggests that ALMS1, GLT8D1 and CSNK2B may represent promising candidate genes for the reported loci and it is likely that genetic variants near these genes may confer schizophrenia risk by regulating the expression level of these genes. Page 32, lines 18-22, Page 33, lines 1-5.
Or is there any published study that has already linked these two genes with the two GWAS loci?
Response: We performed literature search (using following keywords: (1) GLT8D1 and Schizophrenia; (2) CSNK2B and Schizophrenia) and found that no study has reported the association between GLT8D1 and schizophrenia. However, a study published in 2014 showed that ITIH3 polymorphism may confer susceptibility to psychiatric disorders by altering the expression levels of GLT8D1 (3). No published study showed the association between CSNK2B and schizophrenia.

A related question is that the authors used a list of GWAS significant genes. Is this list based on the nearest genes of GWAS loci?
Response: We thank the reviewer for pointing this out. GWAS significant genes were mainly extracted from the largest GWAS of schizophrenia so far (Supplementary Table 3 Table S2.
The authors made statements in quite a few places without statistical justifications: 303: "most of the 10 top candidate genes identified by Sherlock were widely expressed in human brain regions": Should quantify the statistical evidence here.
Response: Following the reviewer's suggestion, we have added the exact number of genes in the revised manuscript. In addition, we also quantified the statistical evidence in the revised manuscript. Page 11, lines 4-16; Page 22, line 22; Page 23, line 1.

323: "frequently interacted with proteins encoded by GWAS significant schizophrenia risk genes". No statistical evidence/p-value.
Response: We appreciate the reviewer for this valuable suggestion. According to the reviewer's suggestion, we provided statistical evidence (i.e., P-value) of the PPI analysis in the revised manuscript. Page 12, lines 2-11, Page 23, lines 18-20.

310: "Permutation test showed that the observed expression pattern is unlikely due to chance effect". What is being tested here? The details of this test should be provided in the text (at least in Methods), not in supplement.
Response: Sherlock integrative analysis (using brain eQTL from Myers et al) identified 10 schizophrenia risk genes. We analyzed the expression pattern of these genes in developing and adult human brain and found that schizophrenia risk genes identified by Sherlock have higher expression level at early developmental stage than later stage (Figure 1). To evaluate if the observed expression pattern is expected by chance or not (i.e., if the genes identified by Sherlock analysis showed higher expression at early developmental stage than later stage), we performed permutation test. Briefly, genes that match the number of Sherlock genes in human genome were randomly selected (8 genes each time, though Sherlock analysis identified 10 genes, 2 of them (LOC376138 and LOC375768) were not found in the BrainSpan dataset) and their expression pattern in developing and adult human brain were assessed. 1000 permutations were performed to obtain the P value (the proportion of tests that have same or extreme expression pattern than the observed expression pattern).

161: Co-expression analysis. It is not clear what the authors tested: number of genes co-expressed with known SCZ genes? Or the average level of co-expression?
The key details should be provided here.
Response: We are sorry for the confusion. We tested the number of known SCZ genes co-expressed with genes identified by Sherlock integrative analysis. According to the reviewer's suggestion, we have provided the details of the co-expression analysis in the revised manuscript. Page 12, lines 18-22, Page 13, lines 1-12.

Reviewer #3 (Remarks to the Author):
In this study, Yang et al. integrated association signals of a largest GWAS report (PGC2) and brain eQTL data, and carried out a series of statistical analysis. In total, 10 candidate causal genes were predicted, while 3 of them, ALMS1, CSNK2B and GLT8D1 were highlighted by the authors, with validation analyses with different methods or in independent data. The current manuscript was well written, and the analysis is sound.
Response: We thank the reviewer for the positive comment on our work.

I only have the following 2 points: 1. The brain eQTL data mainly used in this study included 193 normal human subjects, which is relatively smaller than a recent publication (Mads et al. Am J Hum Genet. 2017 Jun 1;100(6):885-894.), which integrated GWASs with 24 studies of expression quantitative trait loci. In addition, PGC2 schizophrenia dataset is also one of the studied 57 GWASs in the Mads et al's report. It will be better for the authors to refer to these studies and discuss the advantages and shortcomes of the current study.
Response: We appreciate the reviewer for this valuable suggestion. Following the reviewer's suggestion, we have referred to the study of Mads et al and discussed the advantages and shortcomes of the current study in the revised manuscript. Page 36, lines 10-22.

I am wondering whether the authors have western blotting results after transfections the RNAi vectors in the cell-based assay, which might provide clearer evidences for the related downstream pathways after knocking-down GLT8D1.
Response: We thank the reviewer for this valuable suggestion. In the revised manuscript (Page 29, lines 1-11, Figure 4 and Supplementary Figure S10), we first validated the RNAi knockdown efficiency targeting to GLT8D1 and CSNK2B by both real-time PCR and western blot assays, and further explored the effects on NSCs self-renew and found that the expression of the well-characterized stemness marker genes of NSCs (such as Klf4, Sox2, Nanog and Nestin) was significantly up-regulated by depleting GLT8D1 and CSNK2B in neural stem cells compared with controls, which could be Wnt signaling pathway dependent (Supplementary Figure S10). However, limited by the lack of transcriptional and proteome data affected by GLT8D1 and CSNK2B knockdown, the detail mechanisms by which GLT8D1 and CSNK2B regulate schizophrenia need to be further verified in future.

Reviewer #4 (Remarks to the Author):
Yang et al performed GWAS studies and identified 10 candidate genes that contribute to schizophrenia risk. Of these 10 candidate genes, the authors identified ALMS1, CSNK2B and GLT8D1 as the most promising candidates and validated their expression in the human schizophrenia patient cases. Furthermore, they demonstrated that GLT8D1 knockdown using shRNA promotes the self-renew and proliferation abilities of cultured neural stem cells (NSCs) and inhibits NSCs differentiation into astrocytes and neurons, providing the evidence of the potential role of identified risk genes in neurodevelopment. Overall, this study identified novel schizophrenia risk genes from GWAS analysis, validated their expression and hippocampal structures in human schizophrenia patient cases.
Response: We thank the reviewer for the positive comment on our work.
However, I have major concerns of the functional studies they performed in this study. The points are listed as below:

The authors performed all of their analysis in human cases, including GWAS, candidate gene expression analysis, and hippocampal structure analysis but the functional studies were performed in the mouse early postnatal (P7) NSC cultures. It is unclear whether human NSCs express these genes and whether the expression pattern of these genes in the mouse hippocampus is similar to that in human.
Response: We understand the reviewer's concern. We explored the expression of AMLS1, GLT8D1 and CSNK2B in human NSCs using expression data from Lafaille et al (5). The Log2 transformed expression level of AMLS1, GLT8D1 and CSNK2B were 7.85, 6.69 and 12.22. We compared the expression level of these three genes with SOX2, a gene that is highly expressed in neural stem cells (SOX2 is a marker for NSCs). The Log2 transformed expression level of SOX2 is 12.66. These expression data shows that CSNK2B is highly expressed, whereas ALMS1 and GLT8D1 are moderately expressed in human NSCs.
Moreover, it is unclear why postnatal mouse NSC culture was used, since schizophrenia was considered as a neurodevelopmental disorder. Mouse embryonic culture appears to be more relevant.
Response: We thank the reviewer for this valuable suggestion. We agree with the reviewer that schizophrenia was considered as a neurodevelopmental disorder. Nevertheless, multiple lines of evidence suggest that dysfunction of hippocampus plays an important role in schizophrenia. We thus used postnatal mouse NSCs in the original manuscript.
Following the reviewer's suggestion, we repeated the experiments using NSCs from the embryonic mouse (E14.5) and observed similar results. Again, we found that knockdown of GLT8D1 and CSNK2B promoted proliferation and inhibited differentiation of NSCs, respectively. We have included these new results in the revised manuscript.

To best validate the neurodevelopmental deficits associated with human schizophrenia, patient-derived induced pluripotent stem cell (iPSC) culture should be considered in this context.
Response: We agree with the reviewer that patient-derived induced pluripotent stem cell (iPSC) culture may be useful for validation the neurodevelopmental deficits. However, currently, it is difficult for us to conduct this experiment as it is difficult for us to obtain live cells from schizophrenia patients and we are not familiar with iPSC technology. In addition, although the expression of AMLS1, GLT8D1 and CSNK2B was significantly dysregulated in brain tissues of schizophrenia patients in population compared with controls, we are not sure if the expressions of these genes were also dysregulated in NSCs derived from individual schizophrenia cases especially after using iPSC system. Thus, we were not able to perform this assay at current stage.
2. The authors state that ALMS1, CSNK2B and GLT8D1 play active roles in brain development, and showed that reduced GLT8D1 expression led to deficits in NSC proliferation and differentiation. However, they have not shown whether ALMS1 and CSNK2B exhibit similar NSC phenotypes. Despite that other studies showed the role of ALMS1 and CSNK2B in the central nervous systems, these phenotypes are not relevant to hippocampal structure deficits observed in schizophrenia patient cases.
Response: We thank the reviewer for this valuable suggestion. According to the reviewer's suggestion, we performed additional experiments in the revised manuscript. Our expression analysis showed that CSNK2B was also significantly down-regulated in schizophrenia cases compared with controls ( Figure 3C). We thus used shRNA to knockdown CSNK2B in embryonic NSCs. As stated above accordingly, knockdown of GLT8D1 and CSNK2B promoted proliferation and inhibited differentiation of NSCs. We have included these new results in the revised manuscript. Figure  Compared with GLT8D1 and CSNK2B (which showed significant down-regulation in schizophrenia cases), ALMS1 was significantly up-regulated in brains of schizophrenia cases ( Figure 3A). To mimic the effects of ALMS1 up-regulation on NSCs, we tried to clone AMLS1 and construct an over-expression vector. Nevertheless, we found that the coding sequence of ALMS1 is quite large (over 12 kb), though we tried multiple strategies, even by purchasing the commercial AMLS1 expression vector, to solve the clone problem, so far we failed. Due to the technical difficulty and limiting time, we thus were not able not investigate the function of ALMS1 in this study. However, Heydet et al. found that Alms1 is localized at the base of cilia in hypothalamic neurons and a truncating mutation of Alms1 reduces the number of hypothalamic neuronal cilia in mice (6). Intriguingly, ALMS1 is physically interacted with PCM1 (encoded by a schizophrenia risk gene) (7) and Ceroni et al. (European Society of Human Genetics 2016: Title: P09.045A-Case report: exome sequencing of a family with childhood disintegrative disorder) (http://www.well.ox.ac.uk/presentations-and-posters) found that both of ALMS1 and PCM1 genes are mutated in cases of a family with childhood disintegrative disorder (a rare and severe neurodevelopmental disorder). These studies indicated that ALMS1 plays a vital role in brain development and dysfunction of ALSM1 causes neurodevelopmental disorder.
3. The authors showed GLT8D1 knockdown in mouse NSCs led to increased self-renewal, and decreased neuronal and glial differentiation. The authors used GFAP marker for astrocytes. However, in the NSC culture, it has been widely accepted that NSCs give rise to oligodendrocytes besides neurons and astrocytes. This oligo marker should be quantified as well.
Response: We thank the reviewer for this valuable suggestion. Following the reviewer's suggestion, we carried out additional experiments through quantifying the oligo marker (O4). We found knockdown of GLT8D1 and CSNK2B significantly impaired the differentiation of NSCs into oligodendrocytes. We included these new results in the revised manuscript. Figure 5C

Schizophrenia is considered as a synaptic disorder characterized by aberrant synapses (both excitatory and inhibitory). Dendritic and synaptic phenotypes should be included in the analysis associated with GLT8D1 knockdown, including dendritic length, branches, crossings, and functional synaptic characterizations (mIPSCs and mEPSCs).
Response: We appreciate the reviewer for this valuable suggestion. According to the reviewer's suggestion, we performed additional experiments to address these questions. We found that knockdown of GLT8D1 and CSNK2B significantly affected the dendritic length and branches of neurons. In addition to the dendritic phenotypes, synaptic transmission was also affected in GLT8D1 and CSNK2B knockdown neurons. These results suggest that the risk genes identified in this study (i.e., GLT8D1 and CSNK2B) may confer risk of schizophrenia through affecting neurodevelopment and synaptic transmission. We included these new results in the revised manuscript. Figure  Again, we really appreciate the reviewers for their insightful and valuable comments and suggestions, which help us to improve the quality of the manuscript significantly. We hope that our revised work is satisfactory, and are happy to further improve it if needed.