Elsevier Editorial System(tm) for Biological Psychiatry Manuscript Draft Biological Psychiatry Association between Genes of Disc1 Interactors and Schizophrenia Supports the Role of the Disc1 Pathway in the Etiology of Major Mental Illnesses

Title: Association between genes of DISC1 interactors and schizophrenia supports the role of the DISC1 pathway in the etiology of major mental illnesses Article Type: Archival Report Abstract: BACKGROUND: Disrupted in Schizophrenia 1 (DISC1) is currently one of the most interesting candidate genes for major mental illness, having been demonstrated to associate with schizophrenia, bipolar disorder, major depression, autism, and Asperger's syndrome. We have previously reported a DISC1 haplotype, HEP3, and an NDE1 spanning tag haplotype to associate to schizophrenia in Finnish schizophrenia families. Since both DISC1 and NDE1 display association in our study sample, we hypothesized that other genes interacting with DISC1 may also have a role in the etiology of schizophrenia. METHODS: We selected 11 additional genes encoding components of the "DISC1 pathway" and studied these in our study sample of 476 families including 1857 genotyped individuals. We performed SNP and haplotype association analyses in two independent sets of families. For markers and haplotypes found to be consistently associated in both sets, the overall significance was tested using the combined set of families. RESULTS: We identified three SNPs to be associated with schizophrenia in PDE4D (rs1120303, p = 0.021), PDE4B (rs7412571, p = 0.018) and NDEL1 (rs17806986, p = 0.0038). Greater significance was observed with allelic haplotypes of PDE4D (p = 0.00084), PDE4B (p = 0.0022 and p = 0.029) and NDEL1 (p = 0.0027) that increased or decreased schizophrenia susceptibility. CONCLUSIONS: Our findings with other converging lines of evidence support the underlying importance of DISC1-related molecular pathways in the etiology of schizophrenia and other major mental illnesses. The reviewer asks for clarification as to whether one or more affected individuals from each family were included in the analyses. As described in the results section, initially all the affected individuals were included. If a marker or a haplotype displayed significant evidence for association when tested in stage 1 and stage 2 samples the overall significance was tested using both all affected individuals and one affected individual per family (page 13, paragraph 1). Since association was observed outside of the family structure and the confounding factor of linkage, segregation of these variants has not been pursued. The reviewer also asks for clarification within the figure legends. There we state that SNPs with rs-numbers indicate SNPs that display association to schizophrenia, either independently or in haplotypes. All the haplotype blocks (according to LD structure) were tested for association (as indicated …


Reviewer 1
The reviewer asks for clarification as to whether one or more affected individuals from each family were included in the analyses. As described in the results section, initially all the affected individuals were included. If a marker or a haplotype displayed significant evidence for association when tested in stage 1 and stage 2 samples the overall significance was tested using both all affected individuals and one affected individual per family (page 13, paragraph 1). Since association was observed outside of the family structure and the confounding factor of linkage, segregation of these variants has not been pursued.
The reviewer also asks for clarification within the figure legends. There we state that SNPs with rs-numbers indicate SNPs that display association to schizophrenia, either independently or in haplotypes. All the haplotype blocks (according to LD structure) were tested for association (as indicated in the text, page 9, paragraph 3). This is now stated more clearly in the figure legend.
The reviewer's minor comments have been all been included into the manuscript.

Reviewer 3
The reviewer asks if our study design needs to take into account multiple test correction by Bonferroni correction. Since we elected for a two-stage study design that allows for replication of observations in an independent study sample multiple test correction is not required.
The reviewer wonders also if we detected any association between bipolar disorder and the genes analyzed. Our sample includes also individuals who are affected with bipolar disorder and are siblings to individuals with schizophrenia. However, these individuals were not included in our family based analyses since it has been shown that in families ascertained for schizophrenia the genetic liability for disorders in both schizoaffective disorder and schizophrenia spectrum disorders is increased, whereas bipolar disorder, despite having some overlap in genetic risk, display the same liability as the general population background. Thus no analyses were performed using these individuals. The number of individuals with bipolar disorder in our sample is too small (N=67) to perform separate association analysis for this disorder. We therefore hope that through publication of this work, independent researchers will study these genes with other mental illnesses, including bipolar disorder.
The additional references suggested by the reviewer have been added to the manuscript.

Reviewer 5
The reviewer comments on our use of the term "pathway" in the hypothesis of the "DISC1 pathway", suggesting that "network" is more representative of the genes we have studied. The "DISC1 pathway" hypothesis proposes that genes related to DISC1 may also be disrupted in major mental illnesses, by producing disruptions to the same pathways as risk variants in DISC1. As the majority of the functions and roles of DISC1 through its "hub" effects are yet to be formalized, we do not reliably know what genes are of interest in relevant pathways. We have for this study therefore concentrated our focus on the direct interactors of DISC1, and fully admit that this is more of network than a pathway. However, since we use the term "DISC1 pathway" to refer to the global hypothesis rather than just to these selected genes we feel that our use of the term is justified.
The reviewer requires further discussion on type I and type II errors. The statistical methodology we have used is widely used in genetic studies and has been stated to effectively increase true positives (Wen et al: A two stage design for multiple testing in large-scale association studies, J Hum Genet, 2006 and van den Oord et al: A framework for controlling false discovery rates and minimizing the amount of genotyping in search for disease mutations, Trends Genet, 2003). Although the use of a smaller sample in Stage 1 may be thought to increase the potential for false negatives, we feel this is balanced by the increase in true positives. Further, since the use of traditional multiple test correction is considered to be overly conservative, it can be argued that a two stage design actually decreases false negatives (Wen et al). This is clarified in the manuscript (page 9, paragraph 1).

Word count abstract: 232
Word count text: 3855 ABSTRACT BACKGROUND: Disrupted in Schizophrenia 1 (DISC1) is currently one of the most interesting candidate genes for major mental illness, having been demonstrated to associate with schizophrenia, bipolar disorder, major depression, autism, and Asperger's syndrome. We have previously reported a DISC1 haplotype, HEP3, and an NDE1 spanning tag haplotype to associate to schizophrenia in Finnish schizophrenia families. Since both DISC1 and NDE1 display association in our study sample, we hypothesized that other genes interacting with DISC1 may also have a role in the etiology of schizophrenia. METHODS: We selected 11 additional genes encoding components of the "DISC1 pathway" and studied these in our study sample of 476 families including 1857 genotyped individuals.
We performed SNP and haplotype association analyses in two independent sets of families.
For markers and haplotypes found to be consistently associated in both sets, the overall significance was tested using the combined set of families.
CONCLUSIONS: Our findings with other converging lines of evidence support the underlying importance of DISC1-related molecular pathways in the etiology of schizophrenia and other major mental illnesses.

INTRODUCTION
Disrupted in Schizophrenia 1 (DISC1) is one of the most promising candidate genes for schizophrenia. It was identified as disrupted by a balanced translocation (t(1;11)(q42.1;q14.3)) in a large Scottish family with high prevalence of schizophrenia and other psychiatric disorders (1,2). Independent evidence for DISC1 was found by our work using Finnish cohorts. Initially we found strong and replicated linkage to 1q42 with markers intragenic of DISC1 (3,4). We have since reported an association between allelic haplotypes of DISC1 and schizophrenia (5,6), bipolar disorder (7), autism and Asperger's syndrome (8).
Further studies of the most robust of these haplotypes identified in the original schizophrenia study, have implicated that it predisposes to schizophrenia especially in males (6) and associates with poorer visual working memory performance and reduced grey matter volume in the prefrontal cortex (9,10). To date a number of groups have reported association findings between DISC1 and several neuropsychiatric disorders (7,9,(11)(12)(13)(14)(15)(16)(17)(18)(19)(20), neurocognitive (21-23) and neuroimaging (9,24) phenotypes. DISC1 has also been strongly linked to neuronal development based on numerous indirect studies (9,(25)(26)(27). DISC1 is considered a multifunctional "hub" for many protein interactions acting along several pathways (28). Proteins have been recognized to bind to DISC1 through yeast twohybrid screens or co-immunoprecipitation experiments (26, 29-32). A number of these have since been identified as promising candidates for roles in the etiology of schizophrenia and other mental illnesses. We have previously reported an association between NDE1 and schizophrenia in Finnish schizophrenia families (33). Further, NDE1 and NDEL1 have displayed significant interplay with DISC1 that associates with the disorder (34). PDE4B was initially recognized as a candidate for schizophrenia at a translocation breakpoint between chromosomes 1 and 16 in individuals with schizophrenia (35). Since, several groups have reported association findings between PDE4B and schizophrenia (36-38).FEZ1 has been reported to associate with schizophrenia in the Japanese population (39). This evidence has led to the concept of the "DISC1 pathway", a hypothesis which proposes that disruption of the pathways DISC1 is involved in provides risk to mental illness, not just disruption of DISC1 itself (29, 39, 40).
In the present study, we aimed to investigate DISC1 binding partners as potential schizophrenia susceptibility genes in 476 Finnish families ascertained for schizophrenia. We included 11 genes coding for proteins convincingly reported to bind to and interact with DISC1 (18, 40) . These included NDEL1, PDE4B and FEZ1 mentioned above. Additionally we included PDE4D, PCNT, MAP1A, PAFAH1B1, TUBA1A, TRAF3IP1, ATF4 and ATF5.

Sample
In this study we used the same Finnish schizophrenia family sample that has previously been used in DISC1 related linkage and association analyses in Finland (33), plus an additional set of 18 families. The sample collection method has remained unchanged over the years. The individuals with schizophrenia (probands) are identified from three nationwide data sources; the hospital discharge, disability pension, and reimbursed medication registers (41). The firstdegree family members of each proband were thereafter identified through the Population Register Centre, enabling the construction of pedigrees. Personal data recorded in the Concordance between these two psychiatrists is high (kappa values range from 95% to 99% depending on liability class). However in the case of a disagreement, a third psychiatrist estimated life-time diagnosis and the consensus best-estimate life-time diagnosis was made.
In addition to the proband with schizophrenia, family members with other psychiatric illnesses were also identified. It is therefore possible to define affection using increasingly inclusive liability classes (LC). LC1 consists of individuals diagnosed with schizophrenia, LC2 adds individuals diagnosed with schizoaffective disorder to the sample and LC3 adds individuals affected with schizophrenia spectrum disorders (33) (schizoid, schizotypal and paranoid personality disorders, schizophreniform, delusional and brief psychotic disorder, and psychosis not otherwise specified) , and LC4 adds individuals with bipolar affective disorder or major depression with and without the presence of psychosis. In this study we have restricted our end-state diagnosis phenotype to LC3. It has been shown that in families ascertained for schizophrenia the genetic liability for disorders in both LC2 and LC3 is increased (43), whereas the disorders in LC4, despite having some overlap in genetic risk, display the same liability as the general population background (43-45) . According to this LC3 criterion our study includes a total of 886 affected individuals out of which 725 are genotyped.
In this present study, the family sample was randomly split into two non-overlapping subsamples to be analyzed as exploratory and replication data sets. The exploratory set contained 171 families, and the replication set 305 families. The combined sample set included both sample sets. In addition, a sample of 57 anonymous Finnish parent-offspring trios, representing a random sample of the Finnish population was used as a control sample.

Neuropsychological test methods
For a sub-sample of 186 families a neuropsychological test battery has been administered.
The tests were administered in a fixed order by experienced psychologists or especially trained psychiatric nurses, and the scoring of the tests was performed by experienced psychologists (46). The following variables were used in the analyses.

Genotyping methods
The SNPs were selected from the international HapMap project database, build #16, phase 1 (50). The LD structure for the 11 candidate genes was defined using all the SNPs with a minor allele frequency > 5% in the population of European descent (CEU, Utah residents with ancestry from Northern and Western Europe). We used the Haploview program (51)  Genotyping was performed using the Sequenom platform according to manufacturer's recommendations (54). SNPs were later rejected if they had a genotyping success rate < 80% (6 SNPs). None of the genotyped SNPs showed deviation for Hardy-Weinberg disequilibrium (HWE) (p > 0.001). After these quality control measures 142 SNPs were included in the analyses (Supplementary Information Table 3).

Statistical methods
The SNP genotypes were checked for Mendelian errors using Pedcheck (55). In case of an error in any genotype, all the genotypes for that marker in the relevant family were discarded (Mendelian error rate: 0.000076 / genotype).
Association tests between affection status and gene variants were performed using a two-stage study design described above. Such a design has been demonstrated to increase the identification of true positives while also decreasing the rate of false negative associations compared to using the more traditional, but overly conservative, Bonferroni correction (56, 57).
Two-point analyses were performed using the program Pseudomarker (58). This program performs linkage as well as linkage disequilibrium analyses on samples of mixed form (families and trios being combined here), correcting for the effect of linkage on the association tests, and is able to deal with cases where parental genotypes are not known (58).
To perform haplotype analyses, we identified haplotype blocks according to the LD structure defined by Haploview, as described above. For NDEL1, TRAF3IP1, ATF4, ATF5, MAP1A, TUBA1A, FEZ1 and PCNT only one haplotype block for each gene was present.
PAFAH1B1, PDE4B and PDE4D included 2, 10 and 17 LD blocks, respectively. In ATF4, PDE4B and PDE4D we were able to tag 0, 8 and 16 blocks respectively, in other genes, all the blocks were tagged. Generally, the LD structure in our sample correlates well with the LD structure defined using the HapMap CEU population. However, for NDEL1 the LD pattern in the CEU population predicts two LD blocks meanwhile according to our sample one block was predicted. Hedricks D' between the blocks defined according to CEU population in our sample is 1.00 (compared to 0.39 in CEU population). Therefore we analyzed these blocks as one.
Haplotype association analyses to end-state diagnosis were performed using the two-stage method described previously. We compared the frequencies of the haplotypes in the affected offspring of the schizophrenia families to that of the founders of the control trios. These frequencies were defined by estimating each individual's most likely haplotypes using Simwalk2, a Markov chain Monte Carlo (MCMC) and simulated annealing program (59).
Only haplotype block tagging SNPs were taken into account when constructing the haplotypes. Haplotype association analyses to end-state diagnosis were performed using the chi squared test, testing each possible haplotype against all other haplotypes combined in a 2 x 2 table. A global test was performed in a 2 x n (n = number of alleles) table for the haplotype blocks including one or more haplotypes displaying significant association to affection status in both sample sets. Haplotypes with frequencies < 5% in both the case and control samples where combined together to avoid the deviation that these rare haplotypes might have on the result The SNPs and haplotypes showing evidence for association to end state diagnosis (p < 0.05) in our two stage design were tested for the association with quantitative neurocognitive traits (622 individuals available with quantitative data). These analyses were performed using the QTDT program and the orthogonal model (60, 61). Here the haplotypes were re-coded to form a "bi-allelic" marker, so as to test the hypothesized variant against all others combined.
We used age, sex and affection status according to LC3 as covariates in our analysis.
The study has been accepted by the Ministry of Social Affairs and Health (Finland) and institutional review boards. All subjects have provided written informed consent.

RESULTS
We analyzed single SNP markers and haplotypes for their association to schizophrenia using a two stage study design with random sampling of our original 476 Finnish families.
Association analyses between affection status and SNPs and haplotypes were first performed in a sample of 171 families. SNPs and haplotypes displaying evidence for association with pvalues < 0.05 proceeded to analyses in a second sample consisting of 305 independent families. Markers and haplotypes passing the replication criteria in the second stage were analyzed using both sample sets combined.
In stage one 14 SNPs and 19 haplotypes in NDEL1, PDE4B, PDE4D, PAFAH1B1, MAP1A and TRAF3IP1 met our criteria for progressing to stage two. No markers or haplotypes in FEZ1, PCNT, TUBA1A, ATF4 and ATF5 met this criterion. In stage two associations with three markers and four haplotypes of NDEL1, PDE4D and PDE4B were replicated, the same allele showing association with the disorder in both sample sets. These markers and haplotypes were then tested for their overall significance in the combined sample set (Table 1,   Supplementary Information Table 1 and Supplementary Information Table 2). Table 1 Those SNPs and haplotypes that passed through the two stages are as follows. In PDE4D, Previous reports of DISC1 (6) and the DISC1 pathway genes (33, 38) have displayed distinct sex differences in their allelic association. We tested for any potential sex differences in allele frequencies of the recognized associating variants in our sample and could detect no differentiating affects between female and male offspring in these Finnish families ( Table 2). Table 2 Since some of the affected individuals in the sample are siblings, we also tested the identified haplotype variants using only one randomly selected affected offspring per family to account for the possible confounding effect of linkage in the obtained results. Three out of the four haplotypes remained significantly associated with the following p-values: PDE4D = 0.0055, NDEL1 = 0.0018 and PDE4B protective and risk = 0.0025 and 0.070 respectively. Since DISC1 has been previously shown to be associated with visual working memory in these families (10), and with other quantitative neurocognitive traits in other sample sets, we wanted to test these newly identified variants for their association with such traits. In addition to visual working memory, we included several learning and memory related variables in the analyses. Due to the reduction of the neuropsychologically tested sample size we tested all the recognized SNPs and haplotypes only in the combined sample. Of the seven variants, none associated significantly (p < 0.05) with any of the nine tested traits.

DISCUSSION
We show here that three genes involved in the same intracellular pathways with DISC1 associate significantly with schizophrenia in a Finnish family sample. We identified SNP and haplotype variants in PDE4B, PDE4D and NDEL1 that were either under-represented or over-represented in families ascertained for schizophrenia. The minor allele of rs7412571 in PDE4B was significantly over-represented in affected individuals. The neighboring haplotype block displayed both an over-represented ("risk") and an under-represented ("protective") allele. However the over-represented haplotype did not remain significantly associated when we tested only one affected offspring per family. This would suggest that the under-represented haplotype is of greater importance. PDE4D displayed significant association to schizophrenia with a SNP whose minor allele was significantly under-represented in individuals with schizophrenia. Consistent with this finding, a haplotype including the major allele of this same SNP was significantly over-represented in the cases. The minor allele at the NDEL1 rs17806986 SNP displayed significant under representation to schizophrenia. This SNP was a part of the NDEL1 gene spanning haplotype that was also significantly underrepresented in affected individuals.
Fatemi and colleagues have recently reported on an association between schizophrenia and the major allele of SNP rs1354064 in PDE4B that is located in the same LD block we detected association with (36). We had not included this SNP in this study but our SNP rs2503222 tags the variation of this SNP (r 2 = 0.82). Even though we did not see evidence for this SNP alone, our haplotype finding further supports the importance of this locus. Further, a recent report by Pickard and colleagues also noted a haplotype within PDE4B in the Scottish population to be protective against schizophrenia (38). Although this haplotype is located approximately 70 kb toward the 3' end compared to our haplotype finding, and in an independent LD block, we were able to detect association in the same region as the Scottish haplotype, with the SNP rs7412571, suggesting that the same region might also be involved in the etiology of schizophrenia in the Finnish population. The minor allele of this SNP was over-represented in patients with schizophrenia. This SNP tags (r 2 = 1) two SNPs included in the Scottish haplotype. Consistent with our finding, the major allele of these two SNPs were included in the protective haplotypes in the Scottish study. Two previous reports have highlighted protective SNPs recognized close to 3' end of the gene, although we did not detect any evidence for association in this region (36, 37).
Burdick et al have reported on an association with NDEL1 to schizophrenia. They recognized an NDEL1 spanning haplotype being under-represented in affected individuals and further reported that the G allele of SNP rs1391768 was over represented in affected individuals (34).
In contrast, in our sample this G allele is part of a NDEL1 spanning putative protective haplotype.
NDEL1 has been a strong schizophrenia susceptibility candidate gene (27, 62) based on its role in neuronal migration and neuronal outgrowth (63) . Further, it is part of the same Dynein signaling pathway as PAFAH1B1, another DISC1 binding partner (64), and RELN (65), a protein that is also implicated in the etiology of schizophrenia (66). Previously NDEL1 and its homologue NDE1 have been strongly linked to prenatal and early age neuronal development (63, 64, 67) but recent findings suggest a wider role for NDEL1 (68). According to recent findings, DISC1 has a central role in adult neurogenesis along with NDEL1 in mice. Our finding proposes the involvement of these genes also in the etiology of adult onset disorders.
Further, involvement of these genes in the etiology of schizophrenia would support the hypothesis that the vulnerability for developing these disorders might originate already during embryonic development.
We are the first to report direct association between PDE4D and schizophrenia. Interestingly PDE4D (5q11.2-5q12.1) was located close to a linkage peak (chromosome 5q12.3, LOD = 2.59) in our previous genome wide linkage scan when the analysis was conditioned by absence of previously recognized DISC1 risk haplotype HEP3 (33).
Our findings thus support the previous reports especially for PDE4B. Even though our findings are encouraging, naturally further studies are needed to establish the relevance of these genes in the etiology of psychiatric disorders.
It is well established that patients with schizophrenia have several specific memory deficits (69). Given the evidence supporting the involvement of PDE4B and PDE4D in the etiology of schizophrenia (35-38) and their supposed involvement in memory functions (70,71), it was of interest to investigate if PDE4D and PDE4B would, in addition to association with schizophrenia, display association with learning and memory related quantitative traits.
However, no significant association was detected with the recognized variants. Yet this may be due to the reduced sample size in the quantitative trait analyses compared to association tests using affection status.
Since all these genes are shown to biologically interact with DISC1, it would have been interesting to investigate whether these genes demonstrate further combined effects that increase schizophrenia susceptibility within the DISC1 pathways. However due to the potentially large number of interactions in these pathways, our sample size remains too small for testing this meaningfully (6,33). For the same reason, testing for interaction between the most significant Finnish DISC1 haplotype HEP3 (frequency = 0.088) and the variants recognized in the current study (frequencies ranging from 0.058 to 0.37) remains unreliable as only 0.51 % to 3.3 % of the studied individuals would carry combinations of these recognized variants. Yet, as the variants detected here are associated with schizophrenia independent of DISC1, it demonstrates that alterations in other components on the DISC1 pathway than just DISC1 itself may also influence schizophrenia etiology.
Emerging evidence supports a wider role for DISC1 in the development of various psychiatric and neurodevelopmental disorders with the latest observation being an association to the early age neurodevelopmental disorders autism and Asperger's syndrome (8) . Interestingly altered expression levels at PDE4B were recently reported within individuals affected with autism (72). PDE4B has as well been shown to associate with major depression (73). Further PDE4D has been reported to be associated with neuroticism, a psychological trait evidently highly related to major depression and anxiety (74). This would suggest that like DISC1, the other genes in the DISC1 pathways might play a general role in the development of neurodevelopmental disorders rather than being specific to schizophrenia, potentially being involved in a wider spectrum of psychiatric illness.
We have taken the novel approach of studying candidate genes for schizophrenia based on the known molecular interactions of a previously identified susceptibility gene. This "guided candidate gene approach" is theoretically attractive since it is plausible that variants in genes involved in the same molecular pathways can cause similar phenotypes. DISC1 is an excellent starting point on which to anchor such an effort. First it has been shown to be associated with schizophrenia both in our study sample and others and there is an abundance of independent, multimodal findings implying this gene in schizophrenia. DISC1 is known to function as a "hub" for many protein interactions acting along several pathways supporting this approach for this particular gene. Using this approach we observed significant association between schizophrenia and two memory related phosphodiesterase genes (PDE4B and PDE4D) as well as one neurodevelopmentally important peptidase gene (NDEL1). Combined with our previous observation of association with NDE1, this provides strong evidence supporting the DISC1 related pathways in the susceptibility of schizophrenia, and should inspire further research into these pathways.

ACKNOWLEDGEMENTS
The authors would like to acknowledge Sarah Whittall for help in producing the figures.

CONFLICT OF INTEREST STATEMENT
The authors declare they have no potential conflicts of interest.

SUPPLEMENTATY MATERIAL
Supplementary information is available at Biological Psychiatry Online.