Introduction

Deviations of eye movement control are established neurophysiological intermediate phenotypes for psychotic disorders that may be useful for advancing gene discovery in psychiatry.1 Impairments are seen in a reduced ability to accurately track slowly moving objects with the eyes2 and to voluntarily suppress a reflexive saccade to a peripheral target on antisaccade tasks.3, 4 Consistent with multiple lines of evidence indicating shared neurobiological alterations and genetic vulnerability across schizophrenia spectrum and psychotic bipolar disorders,5, 6, 7, 8, 9 comparable eye movement deficits have been demonstrated across these groups in first-episode and chronically ill patients, and in their relatives indicating disturbances in brain systems subserving pursuit initiation and maintenance, and inhibitory control.2, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27 We recently reported both smooth pursuit impairments and antisaccade inhibition errors in a large cohort of clinically stabilized psychotic disorder cases and their relatives as part of the Bipolar and Schizophrenia Network on Intermediate Phenotypes (B-SNIP) Consortium Study.28, 29, 30 We found that the initiation of a pursuit movement, which depends on rapid sensorimotor processing, was disturbed in probands and their relatives, while pursuit maintenance, dependent on cognitive predictions of target motion and the most widely used phenotype in prior genetic studies, was mostly impaired in probands.29 Impaired antisaccade task performance was identified in probands and their relatives, reflecting decreased inhibitory behavioral control.28 How these intermediate phenotypes are related to genetic variation across the genome has to date not been comprehensively studied.

The first genetic studies of eye movement abnormalities in psychotic disorders reported linkage between pursuit maintenance ability and microsatellite markers on the short arm of chromosome 6 (6p21-23).31, 32 Subsequent genetic studies using eye movement phenotypes have predominantly focused on single nucleotide polymorphisms (SNPs) in candidate genes for schizophrenia disease risk, for example, catechol-O-methyltransferase and neuregulin-1.33, 34, 35, 36, 37, 38, 39 Lencer et al.40 reported an association of pursuit-initiation impairments in first-episode psychosis patients with a dopamine D2 receptor gene (DRD2), whereas pursuit maintenance was associated to candidate SNPs in metabotropic glutamate receptor 3 protein (GRM3). This finding supports a model of different genes being potentially significant for different aspects of eye movement control.

Despite these initial reports, confirmation and larger scale genome-wide association studies (GWAS) in patients with psychosis are lacking. We report herein a GWAS evaluating genetic associations with three eye movement phenotypes representing (1) initial sensorimotor processing (pursuit acceleration), (2) sustained pursuit (maintenance gain) and (3) voluntary inhibitory control (antisaccade error rate) in probands with psychosis and controls from the B-SNIP sample, with additional exploratory pathway analyses to identify biological networks implicated by top findings.

Materials and methods

Participants

Smooth pursuit and antisaccade measures were assessed in 849 participants (schizophrenia N=230, schizoaffective disorder N=155, psychotic bipolar disorder N=206 and healthy controls N=258) of the B-SNIP consortium for whom DNA and genotyping information were available. In depth descriptions of the overall B-SNIP study design, inclusion and exclusion criteria, clinical ratings and eye movement assessments have been previously described.28, 29, 30 Diagnoses were made by a consensus process using all available clinical information including the Structured Clinical Interview for DSM IV (SCID)41 with collateral information from family members when available. Probands were clinically stable and receiving consistent psychopharmacological treatment for at least 1 month (Table 1; Supplementary Table 1).42, 43, 44, 45, 46, 47

Table 1 Characteristics of probands with psychosis and healthy controls
Full size table

Inclusion criteria for all subjects were (1) age 15–65; (2) WRAT reading score ⩾65;42 (3) no history of neurologic or systemic disease; (4) minimum of 20/40 visual acuity (with or without correction) and (5) no history of substance abuse within the last month or substance dependence within the last three months according to SCID, and negative urine toxicology (MP On-Site 11: One Step Onsite, ref: 60B02-MPB) on assessment day. Inclusion criteria for control subjects additionally included: (1) no personal or family history (first-degree) of psychotic or bipolar disorder; (2) no history of recurrent depression; and (3) no history of psychosis spectrum personality traits defined as meeting full or within one criteria of a cluster A (psychosis spectrum) Axis-II diagnosis. The study was approved by institutional review boards at each study site and written informed consent was obtained prior to study participation.

Eye movement analyses

The eye movement measures (Table 1) utilized as primary outcome measures in genetic analyses included: (1) initial pursuit acceleration (measure of rapid sensorimotor processing during the first 100ms of pursuit assessed by foveo-petal step-ramp stimuli (18.7°/s);29 (2) pursuit maintenance gain (accuracy of matching eye to target velocity during sustained pursuit) using a triangular wave task (18.7°/s);29 and (3) antisaccade error rate defined as the percentage of trials with failed response inhibition from an overlap task,28 (Supplementary Material Methods). Eye movements were acquired with a video-based eye tracker in a darkened room (Eyelink II, SR Research, Ottawa, ON, Canada, sampling rate 500 Hz) with the same testing conditions and hardware used at all B-SNIP sites. Each eye movement measure was standardized using a normative regression approach, transforming data to z-scores including age, race and sex as covariates. This was done to remove variance in data related to demographic parameters from all groups in a similar way, and to facilitate comparison of the magnitude of effects across the different groups and pursuit measures. Our previous analyses with the B-SNIP study sample did not identify significant effects of antipsychotic dosing, anticholinergic loading or other medication effects on eye movement measures in these stably treated patients.28, 29 Furthermore, eye movement measures were shown to be relatively independent from general cognitive deficits indicated by BACS scores.28, 29

Genotyping and imputation

Genomic DNA from participants was isolated from whole blood using standard protocols and genotyped by the Broad Institute using the Illumina Infinium PsychChip array. Quality control (QC) procedures were conducted with PLINK v1.948 following standardized protocols.49 Genetic markers deviating from Hardy–Weinberg Equilibrium (P<10E−6), genotype-inferred sex differing from reported sex, or having call rates <98% were excluded from analyses. We included SNPs that had minor allele frequencies (MAF) ⩾0.01 in case or control groups. Cryptic relatedness was checked with PREST-plus.50 Samples showing a second degree relationship or greater were excluded resulting in 849 participants available for GWAS.

SNPs passing quality control procedures were used for imputation using HAPI-UR for pre-phasing51 and IMPUTE2 for imputation52 using the 1000 Genomes phase 1 data as a reference panel.53 Poorly imputed SNPs were filtered with the resulting imputed SNPs merged back in with the directly genotyped SNPs from the PsychChip for a total of 4 404 269 SNPs passing filtering criteria used for the analyses described herein.

Genetic ancestry assessments were completed with multi-dimensional scaling (MDS) plots relative to 1000 Genomes Project populations. Race stratified analyses represented a division of the two predominating ancestry components. Analyses of both stratified and whole group analyses utilized the first five principle components of ancestry analyses as covariates.

Genome-wide association analyses approach

We used a mixed-modeling approach as implemented in the Efficient Mixed-Model Association eXpedited (EMMAX)-software package,54 which uses an identity by state (IBS) relationship matrix, and the first five eigenvectors from principle components analysis (PCA) included as covariates to reliably account for mixed ethnicity populations. Standardized eye movement measures (see above) were modeled as quantitative trait phenotypes in relation to genetic data. Probands and controls were grouped together for primary analyses with all ancestry groups combined. For secondary analyses, the sample was stratified by the top two genetically derived ancestry groups with follow-up studies in the proband only sample. The rationale for grouping cases and controls together in primary analyses was to take advantage of the wider range of phenotypic variance for the examination of genetic contributions to eye movement control.

To account for multiple testing using imputed data, the genome-wide significance threshold was set at 1 x E−08, which is more conservative than the commonly used GWAS significance threshold of 5 x E−08.55 False discovery rate (FDR) q-statistics further adjusting for multiple analyses of phenotypes and race groups were calculated. FDR q-values for GWAS significant findings remained <0.05 with the exception of rs2010148567 in relation to antisaccade response inhibition in African ancestry (AA), where q=0.09, all collectively indicating low type I error. We define ‘suggestive associations’ as P-values exceeding 5 x E−7 but not meeting 1 x E−8 GWAS significance. The closest gene was assigned to each SNP using BEDTools closest and RefSeq gene annotations from hg19.56

Exploratory pathway analyses

We used Ingenuity Pathway Analysis software (Ingenuity Systems, Redwood City, CA, USA) to conduct exploratory analyses (using the Core Analysis feature) of genes affiliated (within 15 kb) with the top 200 SNP associations identified through GWAS analyses. Associations in all participants were examined separately for each eye movement measure by merging the top 200 associated SNPs from the two primary ancestry group analyses. The expression quantitative trait loci analyses of top SNPs associated with clinical phenotypes were performed using the Genotype-Tissue Expression GTEx Portal (www.gtexportal.org/home) and the United Kingdom Brain Expression Consortium (UKBEC, www.braineac.org).

Results

Initial sensorimotor processing

GWAS of initial pursuit acceleration in all participants

Across participants, the most robust genome-wide significant association was identified with an isolated SNP in the Importin 8 gene (IPO8) at chromosome 12p11.21 (Table 2). In addition, a number of SNPs in a non-coding RNA gene at chromosome 2p12, and in an intergenic region near the mitogen-activated protein kinase MAP3K1 gene at chromosome 5q11.2 showed patterns of suggestive association.

Table 2 Top SNPs associated with eye movement measures in the combined proband-control sample and in subsets defined by ancestry
Full size table

GWAS of initial pursuit acceleration stratified by ancestry

GWAS in participants of predominantly AA (N=300) revealed the aforementioned genome-wide significant association with IPO8, as well as an additional genome-wide significant association with SNPs in the protocadherin 12 (PCDH12) gene at chromosome 5q31.3 (Figure 1a; Table 2). Other polymorphisms with suggestive associations included SNPs ~300 kb upstream of the Neurensin 1 gene (NRSN1) in an intergenic region at chromosome 6p22.3 and a SNP in the LIM domain only protein 7 gene (LMO7) at chromosome 13q22.2.

Figure 1
figure 1

Manhattan plots from genome-wide association studies (GWAS) stratified for participants of predominantly African ancestry (N=300, left side) and participants of predominantly Caucasian ancestry (N=549, right side). Results for the three eye movement measures used as phenotypes in GWAS are depicted: (a) initial pursuit acceleration, (b) pursuit maintenance gain and (c) antisaccade error rate. For more details see Table 2.

Full size image

In participants of predominantly Caucasian ancestry (CA, N=549), initial pursuit acceleration was significantly associated with a SNP representing a missense mutation in CYB5R3 at chromosome 22q13.2 coding for membrane bound cytochrome B5 reductase 3. In addition, the aforementioned SNPs in a non-coding RNA gene at chromosome 2p12 showed suggestive associations.

GWAS of initial pursuit acceleration in probands only

Follow-up analyses in probands across ancestries identified the genome-wide significant association with the SNP in IPO8 that was seen in the whole-study sample, and additionally suggestive association in LMO7. Similarly, follow-up analyses in AA probands revealed the genome-wide significant associations with SNPs in IPO8, PCDH12 and LMO7 (Table 3). In the sub-sample of CA probands, no genome-wide significant association was observed with initial pursuit acceleration.

Table 3 Top SNPs associated with eye movement measures across all probands and in subsets of probands defined by ancestry
Full size table

Sustained pursuit maintenance

GWAS of maintenance gain in all participants

No associations were identified which exceeded our pre-defined threshold for genome-wide significance of 1 × E−8.55 Suggestive associations with pursuit maintenance gain across all participants were identified with a number of SNPs in or around the src Homology-3 (SH3) domain gene (SH3GL2) at chromosome 9p22.2 (Table 2).

GWAS of pursuit maintenance gain stratified by ancestry

Suggestive associations with SH3GL2 were also observed in CA participants only (Table 2; Figure 1b). In AA participants, we identified further suggestive associations with pursuit maintenance ability. These included SNPs in TMPRSS5 at chromosome 11q23.1 encoding a transmembrane serine protease, and in POP7 at chromosome 7q22.1, which is a protein-coding gene related to gene expression and RNA transport. Very close to this region on chromosome 7, we additionally identified a SNP in GGYF1, which encodes a protein believed to act cooperatively with growth factor receptor-bound protein10 (GRB10) to regulate tyrosine kinase receptor signaling.

Follow-up analyses in the proband subsample (Table 3) showed suggestive associations of pursuit maintenance gain with SNPs in POP7, GGYF1 and TMPRSS5 in the subsample of AA probands but not in CA probands.

GWAS of antisaccade error rates

GWAS with antisaccade error rate in AA participants identified one GWAS significant SNP and 25 suggestive SNPs in an intergenic region at chromosome 7 (Table 2; Figure 1c). However, no further associations with error rate were observed in either the whole sample or proband subsamples considered separately.

More details on top 200 SNPs identified in primary and secondary GWAS are given in Supplementary Table 2.

Exploratory pathway analyses

The top 200 SNP associations identified in race stratified GWAS analyses represented 89 distinct genes for initial pursuit acceleration and 103 distinct genes for pursuit maintenance gain (Supplementary Table 3). A top physiological system category identified for both pursuit phenotypes was nervous system development and function, which was represented by ~19% (N=17) of the top genes associated with initial pursuit acceleration and ~22% (N=23) of the top genes associated with pursuit maintenance gain (enrichment P-value range 4.9−10−2–9.8 × 10−4). Noting the similarities in pathway relationships identified, the genes comprising these lists were merged (N=189 unique genes; N=40 genes relevant to the nervous system) and visualized with neural network mapping that highlights the nervous system development and synaptic functioning (Figure 2). This revealed functions including ‘formation of the eye’, ‘eyelid reflex’ and ‘electrophysiology of the eye’; ‘excitatory postsynaptic action potential’; as well as ‘neurological signs’, ‘movement disorders’ and ‘neurodegeneration’. Nineteen of these genes have previously shown evidence for a relationship to psychotic disorders.

Figure 2
figure 2

Summary of Ingenuity Pathway Analysis (IPA) using genes encoding the top 200 SNPs associated with initial pursuit acceleration and pursuit maintenance within the study population (N=849). The functional category nervous system development and function was identified as one of the top five physiological systems represented by these genes. Genes listed from the top have the greatest number of connections to functional categories within the nervous system development category listed in the lower panel. Genes that have shown evidence for a relationship to psychotic disorders are highlighted in red. SNP, single-nucleotide polymorphism.

Full size image

Discussion

In this GWAS, we focused on eye movement measures indexing different neurophysiological aspects of eye movement control known to be impaired in psychotic disorders. We identified novel genome-wide significant findings that may promote understanding of psychosis risk and pathophysiology. First, the most significant associations were found for initial sensorimotor processing with IPO8 at 12p11.21, PCDH12 at 5q31.3, CYB5R3 at 22q13.2 and LMO7 at 13q22.2. These associations were predominantly observed in variants with lower minor frequencies, mostly in AA participants. Second, suggestive associations with sustained pursuit maintenance were observed with protein coding SNPs in and around SH3GL2 at 9p22.2. Third, significant genome-wide association of behavioral response inhibition was observed with a non-coding region at chromosome 7 in AA participants. All genes for which we found significant associations with referring SNPs are expressed in the brain (www.gtexportal.org/home; www.braineac.org). Those variants exceeding our pre-defined GWAS significance threshold also had low FDR statistics after accounting for multiple comparisons. Finally, exploratory pathway analyses of top associated SNPs identified commonalities between genes related to smooth pursuit measures, which consisted of loci previously associated with brain development, neurophysiology, ocular physiology and schizophrenia risk. These findings provide important new genetic information about what has long been one of the most promising familial phenotypes associated with psychotic disorders.1, 2, 57 This said, our findings extend previous reports from large-scale genetic studies showing considerable overlap between schizophrenia spectrum and bipolar disorders.6, 7, 8, 9

Genetic alterations related to initial sensorimotor processing

The IPO8 gene, in which we found a missense mutation significantly associated with initial pursuit acceleration, encodes importin 8, which has a key role in nuclear–cytoplasmic transport of proteins including many miRNAs.58 Importin 8 has also been identified as a component of miRNA-guided regulatory pathways for gene silencing by argonaute proteins, which are ubiquitous proteins found in plants, animals and fungi, leading to mRNA destabilization by transcription repression and translation inhibition.59 Blocking importin 8 reduces the nuclear concentration of argonaute proteins and may thus attenuate mRNA destabilization.59 We found this mutation, to date, only identified in those of AA, specifically associated with pursuit acceleration in AA probands.

Other suggestive associations with initial sensorimotor processing in the whole sample included a non-coding RNA gene (chr2p12), and SNPs ~15 kb upstream of MAP3K1 (chr5q11.2), which encodes a mitogen-activated protein kinase known to regulate apoptosis.60 There were 52 other SNPs in or around MAP3K1 including others upstream of the transcription starts site and two missense variants within the coding region, all with association P-values ranging from 3.4 × 10−5 to 2.2 × 10−7. In addition, expression quantitative trait loci analysis of the top associated SNP (rs1862618) revealed a strong correlation with the expression of the SET domain containing 9 gene (SETD9) (www.gtexportal.org/home) at chromosome 5q11.2, coding for a SET7 class of methyltransferase, which methylates H3K4. This correlation exists across multiple tissue types including regions of the brain and skeletal muscle.

Stratified GWAS by ancestry revealed significant genome-wide associations of sensorimotor processing with a synonymous mutation in Protocadherin 12 (PCDH12) in AA participants, primarily driven by effects observed in AA probands. PCDH12 belongs to a protocadherin gene cluster at chromosome 5q31 that has been previously implicated in schizophrenia and psychosis in non-AA samples.61, 62 PCDH12 encodes a cellular adhesion molecule that has an important role in cell–cell interactions including axonal guidance and synaptic specificity. The association with initial pursuit acceleration suggests that in psychotic disorders alterations of the cadherin-based adhesive system may alter functional connectivity and coherent information processing in brain systems needed for fast visual information processing.63 Putative association of PCDH12 with gyrification asymmetry has also been reported in schizophrenia suggesting its involvement in neurodevelopment and neural network formation.64 More broadly, our sensorimotor processsing related findings are in line with reports from the B-SNIP sample showing associations between genetic variants of the cadherin family and electroencephalogy abnormalities65, 66 and resting state brain activity seen with imaging studies.67

Suggestive associations of rapid sensorimotor processing around the Neurensin 1 gene (NRSN1, chr6p22.3) were observed in AA participants. NRSN1 has been suggested to have an important role in the transduction of nerve signals and for neural plasticity. This may explain why NRSN1 has been previously related to information-processing speed,68 supporting our finding of a specific association with rapid sensorimotor transformation needed during pursuit initiation.

Another genome-wide association was found for a missense mutation in the LMO7 gene coding for LIM domain only protein 7 (chr13q22.2) in AA participants in general, and in AA probands specifically at a genome-wide significant level. LMO7 is involved in protein–protein interactions and transcription, and mutations by alternative splicing in LMO7 have been related to retinal defects and degeneration,69 which could explain why we found a SNP in this gene to be associated with rapid retinal error information processing.

Stratified GWAS in CA participants revealed genome-wide significant association of sensorimotor processing with CYB5R3 (ch22q13.2). Notably, patients with 22q13 deletion syndrome are characterized by autism and schizophrenia-like symptoms.70 In these patients, loss of CYB5R3 has been related to impaired language skills.70

Genetic alterations related to pursuit maintenance

In contrast to pursuit initiation, sustained pursuit maintenance depends upon an established prediction of target velocity, and thus is more dependent on cognitive function. Here across all participants, we found suggestive associations of sustained pursuit maintenance with a region in the SH3GL2 gene (chr9p22.2) encoding Endophilin A1.71 Previous studies in schizophrenia suggest that SH3GL2 is differentially expressed in gray matter of prefrontal cortex in psychosis patients compared to controls.72, 73 Endophilin A1 is implicated in synaptic vesicle endocytosis involving intracellular signaling, calcium homeostasis and neurotransmitter release.73 Specifically, Endophilin A1 is suggested to regulate glutamate release in neurons expressing vesicular glutamate transporter 1.74 This is of interest as we recently found pursuit maintenance being associated with genetic variants in GRM3.40, 75

Genetic alterations related to antisaccade response inhibition

SNPs associated with antisaccade performance in AA participants were identified in an intergenic region at chromosome 7 with the closest defined gene being the non-coding RNA LOC101928283, which is ~230 kb away. An expression quantitative trait loci analysis search for the top 10 SNPs within the United Kingdom Brain Expression Consortium (UKBEC) (www.braineac.org) showed significant association with expression of the gene GPR37 (G protein-coupled receptor 37) within the hippocampus. The encoded protein contains seven transmembrane domains and is found in cell and endoplasmic reticulum membranes. G protein-coupled receptors are involved in translating outside signals into G protein-mediated intracellular effects. A previous GWAS on antisaccade error rate in twins reported suggestive associations with SNPs at chromosome 7 close to the region identified in the present study.76 The same study also revealed genome-wide significant associations with genes at chromosome 2.76 Others reported genome-wide linkage of antisaccade error rate with SNPs at chromosome 3p12 from a schizophrenia family study (COGS).57 Altogether, these findings support the notion that antisaccade error rate may be regarded as a complex polygenic trait.76

Implications from pathway analyses

The 200 top SNPs associated with pursuit acceleration and maintenance gain were enriched for genes related to nervous system development pathways including relevant functions such as eye formation, neuronal action potential and movement disorders. Some of these genes have also been identified in previous disease risk association studies for psychotic disorders. Altogether, these findings support the model of smooth pursuit disturbances representing alterations in brain systems contributing to psychosis disease pathology. They are in line with other findings from the B-SNIP sample that revealed brain system changes related to gene clusters indicating physiological pathways involved in brain development, synaptic transmission and ion channel activity.67, 77

Limitations

Although our findings are novel and potentially heuristically valuable, there are potential limitations. First, although our sample size was large compared to most previous association studies of eye movements in psychosis probands, it is still small for GWAS. To enhance statistical power, we used a combined proband-control sample from the B-SNIP study, which had the benefit of increasing sample size as well as phenotypic variance for genetic association analyses. However, our analyses are not powered to detect smaller genotype–phenotype associations in the individual proband groups. Further research is needed to examine potential disorder-specific effects. Second, some of our more highly associated SNP findings represented those with lower minor allele frequencies (that is, IPO8, PCDH12, CYB3R5, LMO7). Special effort was undertaken to assure SNP genotyping quality and phenotyping for these variants, however these associations require validation and replication, especially with respect to the findings in the subgroup of AA participants.

Conclusions

GWAS using eye movement phenotypes offers a promising approach for advancing pathophysiological models and understanding discrete components of complex multifactorial genetic risks for psychosis. We identified regions of interest for further study including some novel findings in addition to suggestive associations that are consistent with prior disease risk studies. Collectively, these findings highlight the importance of genes related to disease risk alongside other unique genetic contributions to eye movement phenotypes associated with psychotic disorders.