INTRODUCTION

Autism is a pervasive developmental disorder characterized by the presence of atypically developing or impaired development in social interaction and communication, and markedly restricted, repetitive activities and interests [1]. While autism has been incorporated into the diagnosis of Autism Spectrum Disorder (ASD) in DSM-V [2], patients with autism generally have a moderate to severe level of illness often with a clinically significant intellectual impairment that requires substantial care or support. The prevalence of ASD diagnosis has increased rapidly in recent decades [3], reaching 16.8 per 1000 children (one in 59) according to the latest report from the Autism and Developmental Disabilities Monitoring (ADDM) Network in the United States [4], although this increase might be in part attributed to the change in the diagnostic criteria. Even though twin studies of autism suggest a strong genetic component [5,6], the genetic etiology of autism is yet evident. Genetic research about autism has uncovered many moderately penetrant rare variants that may explain up to 35% of cases with this diagnosis. While recent genome-wide association studies suggest that the genetic risk of autism mostly resides with common variants, which may act additively [7,8], so far only a few common genetic variants have been associated with autism through several sizeable genome-wide association studies [9–13]. A recent large-scale genome-wide combined analysis of multiple major psychiatric disorders indicate that common genetic variants shared between ADHD and ASD are mainly located at genomic regions with CNVs that have been associated with ASD [14].

Previous studies have implicated genetic variants in ULK4 as risk factors for multiple psychiatric disorders [15–17]. Unc-51-like kinase 4 (ULK4), located at 3p22.1 (Chr3: 41, 246, 599–41, 962, 440, hg38) is comprised of 37 exons and plays an important role in neuronal development [15]. Deletions spanning exons 21–34 of ULK4 have been found in 4 out of 3,391 patients with schizophrenia from the International Schizophrenia Consortium but none of 3181 control subjects [17]. The CNVs, deletions removing exons 33 and 34 of a large splice variant of ULK4, have been found in patients with schizophrenia (2/708), bipolar disorder (2/1136) and autism (1/507) compared with 37 of 98,022 controls in the Icelandic population [15]. Another study detected an additional seven cases of CNVs involving ULK4 in 5891 patients with developmental disorders (DD) that often have co-occurrence with ASD [16]. While common genetic variants at ULK4 have not been associated with schizophrenia in the past genome-wide association studies from either the Psychiatric Genomic Consortium (PGC) or a sizeable study of Han Chinese populations [18], the data from the PGC showed potentially suggestive association signals (p = 0.0001) at rs17210774 in intron 33 of ULK4 with bipolar disorder and rs1722850 (p = 0.001) flanking ULK4 with major depressive disorder [15].

Apart from these reports, previous studies have associated ULK4 common variants with hematologic and cardiologic disorders such as multiple myeloma, acute aortic dissections, and hypertension at genome-wide significance. A meta-analysis of 19 studies (n = 29,378 subjects) identified rs1717027 at ULK4 associated with diastolic blood pressure [19]. The missense variants rs1052501 and rs2272007 in complete linkage disequilibrium (LD) in ULK4 (r² = 0.98 and D' = 1) have been associated with multiple myeloma at genome-wide significance [20]. In addition, genome-wide association studies have identified ULK4 variants associated with blood pressure trait and hypertension in both European ancestry and Han Chinese populations [21,22]. Therefore, ULK4 might have a pleiotropic effect on complex human disorders including autism.

We conducted the present study to examine associations of common variants in ULK4 with autism using a family cohort of samples in Han Chinese, followed by an analysis of three family cohorts of European ancestry samples. SNP rs2272007, a non-synonymous variant was associated with autism in the Han Chinese. SNP rs2272007 and several others in LD were strongly associated with ULK4 expression in postmortem human prefrontal cortex, multiple brain areas, and other human tissues including artery and whole blood in an independent dataset. Finally, the minor risk allele of rs2272007 for autism was associated with poorer treatment response in patients with schizophrenia treated with olanzapine in an antipsychotic trial.

METHODS

Chinese autism sample: The subjects of autism case-parent triad sample were initially recruited for a genome-wide association study (GWAS) and all were Han Chinese [9]. A senior child psychiatrist made the diagnosis of Autistic Disorder according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision (DSM-IV-TR). An independent child psychiatrist confirmed the diagnosis with a semi-structured interview, which included a parent interview in which a psychiatrist reviewed children's symptoms with parents and a child interview [23]. Subjects with Asperger’s syndrome and pervasive developmental disorder not otherwise specified (PPD-NOS) were excluded. Therefore, the cases were typically autism, belonging to autism spectrum disorder according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V). Neurological examination and check of features and signs were performed to exclude specific genetic diseases such as Down's syndrome, Angel's syndrome, and Fragile X syndrome. The karyotype analysis was performed for each subject to eliminate chromosomal abnormalities, and CNVs were screened.

European ancestry sample: Three cohorts of European ancestry samples were also analyzed, and they were from previous GWAS of autism [9]. The samples were the Autism Genetic Research Exchange (AGRE) sample, the Autism Genome Project (AGP) and the Simons Foundation for Autism Research Initiative (SAFRI) Base samples. All three cohorts are family-based samples including multiplex pedigree. The recruitments have been described in the previous studies [10,24,25].

Subjects for antipsychotic treatment: Data from the Clinical Antipsychotic Trials in Intervention Effectiveness (CATIE) [26] was used to examine the association of genetic variant at ULK4 with antipsychotic treatment response in patients with schizophrenia, a neurodevelopmental disorder that may share common genetic etiology with autism. CNVs that disrupted ULK4 have been found in schizophrenia. Considering the sample size, we only included patients with schizophrenia of European ancestry and who completed the Phase 1/1A of the CATIE study, in which patients were randomly assigned to one of the five medications (olanzapine, perphenazine, quetiapine, ziprasidone, and risperidone), for the analysis. The primary outcome of the CATIE study was the time to discontinuing use of antipsychotic medication.

All data of SNPs were from previous genome-wide studies and genotyped with microarray chips. The Chinese autism triads were genotyped using the HumanHap CNV370K. SNPs located in the genomic region of ULK4 (upstream and downstream 100 kb) were selected for this study. The region was on purpose defined wider in order to include more SNPs for analysis. The detailed genotype data of three European ancestry cohorts in AGRE, AGP, and SFARI have been described in previous studies [10,24,25]. In brief, the AGRE samples were genotyped with the Illumina HumanHap 550 K chip; and the AGP and SFARI samples were genotyped with the HumanHap 1M chip [9]. Genome-wide quality control was applied as follows: individuals with missing SNP call rate > 2% were excluded. SNPs were zeroed out if Mendelian errors > 5% for a SNP and individuals were removed if Mendelian errors > 3% for an individual from the analysis. Sample duplications and cryptic relatedness were checked by the Identity by State (IBS) analysis of genotypes data in 22 chromosomes. One of each related pair (IBD-sharing coefficients > 0.10) was excluded. No sex error was found based on the heterozygote of the X chromosome. Individual genotyping and quality control of the CATIE sample were conducted for a genome-wide association study of schizophrenia [27], and the samples were genotyped with Affymetrix 500 K chip and 164 K chip.

One dataset of the mRNA expression was based on postmortem human prefrontal cortex of fetal and postnatal subjects. The sample was collected at the Clinical Brain Disorders Branch of the National Institute of Mental Health of NIH and data were developed by the Lieber Institute for Brain Development (http://braincloud.jhmi.edu) [28]. Sample subjects included fetal, neonatal, child and adolescent, and adults.

The other dataset was the Genotype-Tissue Expression (GTEx), which consists of whole-genome sequence and RNA-Seq data from about 650 deceased adult donors, with multiple tissue samples collected per donor (https://commonfund.nih.gov/gtex). The GTEx dataset included mRNA expression from many different regions of the brain with variable sample size in specific tissues, which included anterior cingulate cortex (BA24), Caudate nucleus (basal ganglia), Cerebellar hemisphere, Cerebellum, Cortex, Frontal cortex (BA9), Hippocampus, Hypothalamus, Nucleus accumbens (basal ganglia), Putamen (basal ganglia) and Pituitary [29].

Transmission disequilibrium test (TDT) was performed to identify common genetic variants of ULK4 associated with autism in the Chinese and European ancestry samples using PLINK 1.9. The cis-association analysis was performed to test for the association of SNP genotypes with the expression of a gene where the SNPs are located. In the Braincloud gene expression dataset, general linear regression analysis was performed to test the association of SNP genotypes with the mean level of gene expression while adjusting for age, sex, race, and post-mortem interval (in h) and RNA integrity number (RIN). In the expression data of Genotype-Tissue Expression (GTEx), associations of SNPs at ULK4 with its mRNA expression were performed through the online analysis, in which linear regression was used while adjusting for PEER factors, sex, genotyping platform, and three genotype-based principal components in the gene expression dataset of GTEx (https://gtexportal.org/).

The association analysis of ULK4 variants with antipsychotic treatment response was based on the SNPs genotyped in the original genotyping arrays [27] in a 160-kb genomic region around the top signals. Cox proportional hazard model was used to estimate the hazard ratio of discontinued use of antipsychotics [30].

RESULTS

TDT analysis was performed for 15 SNPs in 275 autism case-parent triads of Han Chinese ancestry. The mean age of assessment was 60.69 (SD = 22.29) months and 85% of children were boys. Three cohorts of European ancestry samples were also analyzed as replication samples, including 285 case-parent triads of AGRE, 325 case-parent triads of AGP, and 689 case-parent triads of SAFRI. SNPs in the analysis were located at the genomic region (upstream and downstream 100kb) of ULK4 and genotyped with the Illumina HumanHap 370K Chip. Multiple association signals (OR > 1.65; p < 0.005) were observed at SNPs rs1052501, rs1716975, rs2272007, rs12635286, rs17066958 for allelic transmission disequilibrium from parents to patients with autism (Table 1). Even though some of these SNPs were in strong LD, they passed through the Bonferroni correction for multiple testing. The SNPs rs2272007, rs1716975, and rs1052501 are located in exon 2, 7 and 17 of ULK4, respectively, whereas rs12635286 and rs17066958 are located at an intergenic region between ULK4||TRAK1. The three SNPs in the exons are non-synonymous and in complete LD with each other (pairwise r² = 1 and D' = 1) according to the CHBJPT sample of 1000 Genomes, of which SNPs rs1052501 and rs2272007 are conserved in vertebrates. However, we did not find a consistently nominal association in three family-based samples of the European ancestry (p > 0.05), likely due to that our patient subjects were limited to a diagnosis of simple form of autism rather than autism spectrum disorders that were included in other three samples, some of which were from multiplex families [9].

All SNPs showing signals for association with autism had significant cis-association with mRNA expression in the tissue sample of postmortem human prefrontal cortex (n = 261). The most significant cis-association was observed at rs1052501 (p = 1.77 × 10⁻⁸), rs1716975 and rs2272007 (p = 4.94 × 10⁻⁷) (Figure 1a), which are also in strong LD (pairwise r² > 0.9 and D' = 1) in the CEU sample. We selected rs2272007 as a proxy for all three SNPs for examining cis-association in the sample by the stage of life. Interestingly, SNP rs2272007 was associated with mRNA expression consistently in the sample of fetal (p = 0.039; n = 37), child and adolescent (p = 0.005; n = 30), and adult (p = 0.001; n = 162) comprising of adult African American (p = 0.001; n = 80) and adults of European ancestry (p = 0.018; n = 82) (Figure 1b).

SNP rs2272007 had consistent cis-associations in multiple brain areas and other human tissues in an independent GTEx dataset. We found the most significant cis-association with ULK4 expression for rs2272007 in the sample of tibial nerve (NES = −0.69; p = 1.4 × 10⁻³⁷), pituitary (NES = −1.2; p = 9.6 × 10⁻³¹); and rs2272007 had cis-association in broad brain regions including putamen (NES = −1.3; p = 9.6 × 10⁻²⁰), frontal cortex-Brodmann area (BA9) (NES = −1.1; p = 1.6 × 10⁻¹⁹), anterior cingulate (BA24) (NES = −1.1; p = 1.96 × 10⁻¹⁶), cerebellum (NES = −0.88; p = 3.4 × 10⁻¹⁴), caudate of basal ganglia (NES = −0.75; p = 3.5 × 10−14), nucleus accumbens of basal ganglia (NES = −0.82; p = 9.6 × 10⁻¹³), cerebellar hemisphere (NES = −0.93; p = 3.7 × 10⁻¹³), hypothalamus (NES = −0.59; p = 3.6 × 10⁻¹¹), hippocampus (NES = −0.60; p = 3.2 × 10⁻¹⁰), amygdala (NES = −0.61; p = 8.3 × 10⁻⁸) and substantia nigra (NES = −0.91; p = 1.4 × 10⁻⁷) (Figure 1c). In addition, we noted that rs2272007 had a strong cis-association in other human tissues such as transformed fibroblast cells, artery and whole blood. In all these, the risk-associated alleles were associated with relatively higher expression.

Antipsychotics have been approved for treating behavior problem of autism [31]. Based on the treatment data available in the European ancestry sample, we performed a fast-track analysis to examine the impact of ULK4 variants on antipsychotic drug response in patients with schizophrenia. Of 16 SNPs in the 100kb around the top association signal at rs1716975, we found that only rs2272007, which showed both associated with risk of autism and cis-regulatory effect in postmortem human brain, was genotyped in the CATIE clinical trial data (n = 419). SNP rs2272007 was associated with time to discontinuing the use of antipsychotics in patients treated with olanzapine (p = 0.0034) (Table 2). Two other SNPs rs1717020 and rs9856633, in complete LD with rs2272007 (r² = 0.996 and D' = 1), were also associated (p = 0.0034) with time to the discontinuation in patients treated with olanzapine (n = 95).

In the overall sample, SNPs rs2030431 (p = 0.0003) and rs7651623 (p = 0.0005) were associated with the antipsychotic response. Specifically, the rs2030431 heterozygous TC (minor allele C, HR = 17.25; p = 3.0 × 10⁻⁴) and rs7651623 heterozygous GA (minor allele A, HR = 16.33; p = 5 × 10⁻⁴) were shown a high risk of discontinuing the use of antipsychotics before completing the Phase 1/1A trial (Table 3). Two SNPs rs2030431 and rs7651623 were non-polymorphic (MAF = 0) in the CHBJPT and CEU samples of the 1000 Genomes, but they are also in strong LD (pairwise r² = 0.799 and D' = 1) in the patient sample. SNP rs7651623 is at a putative transcriptional factor binding site (TFBS) and perfect LD with rs2272007 (pairwise r² = 0.027 and D' = 1). In addition, we also found that some SNPs were associated with treatment response in individual drugs (Table 2). For example, compared with the homozygote CC of the major allele that conferred a risk of autism, the heterozygous (CT) of rs2272007 had more than the 4-fold likelihood for discontinuing use of olanzapine medication (HR = 4.22; p = 0.0034). However, we did not find a significant effect of rs2272007 homozygote TT on the time to discontinuation, likely due to the small sample size (n = 4) for a non-continuous outcome.

DISCUSSIONS

Our study provides several lines of evidence that ULK4 may be a susceptibility gene for autism and this has implications for precision medicine. Three nonsynonymous SNPs in complete LD were associated with autism in a Han Chinese sample, and they were shown cis-association with ULK4 expression in postmortem human prefrontal cortex from samples in the different stage of life and different brain regions in an independent dataset. SNP rs2272007, one of the three non-synonymous in LD, also had an impact on antipsychotic treatment response in the primary outcome, time to discontinuing use of medication, in chronic patients with schizophrenia treated with olanzapine.

The association of three nonsynonymous SNPs with autism is likely real, but they are located at three exons, different from those disrupted by the CNVs reported in the previous studies of schizophrenia. First, SNP rs2272007 and the other two were associated with autism and were shown cis-association with the gene expression in postmortem human brain. Although the cis-association analysis performed in the sample of European ancestry populations, the MAFs of rs2272007 are similar in the East Asian (0.149), European (0.191) and American (0.19) populations. Therefore, the gene expression results provide molecular evidence for the genetic association with autism that is found in the sample of Han Chinese. In addition, SNP rs2272007 is a non-synonymous variant causing a lysine-to-arginine substitution. Multiple rare variants, including rs7651623 in a TFBS, were associated with the time to discontinuing use of antipsychotics in the overall sample and were in perfect LD with rs2272007, which was only associated with time to the discontinuation in patients treated with olanzapine. This suggests that the three functional SNPs associated with the risk of autism may play a role in the antipsychotic response. We noted that individuals with the heterozygous of risk allele have a poorer response to olanzapine than those homozygotes of non-risk allele C. However, the mechanism of this relationship between antipsychotic treatment response and the autism-associated variant at ULK4 is not clear. This warrants a further study of the molecular mechanism.

TABLE 1

Table 1. Associations of ULK4 genetic variants with autism in Han Chinese triad sample.

FIGURE 1

Figure 1. Genetic association of SNPs at ULK4 with the risk of autism and cis-association in postmortem human brain. (a) overall genetic association of all SNPs with autism and ULK4 expression (circle, indicates −log p-value of SNPs for association with autism in Han Chinese sample; Triangle, indicates −log p-value of SNPs for cis-association in postmortem human prefrontal cortex; Blue diamond, indicates the top signals for association with autism, and red colored, indicates LD (r²) of top SNPs with other signals; (b) cis-association of rs2272007 in prefrontal cortex across different stage of life and in adult sample by race; (c) cis-association of rs2272007 in multiple tissues of the central nevus system and different brain regions.

TABLE 2

Table 2. Autism-associated SNPs at ULK4 associated with antipsychotic treatment response in patients with schizophrenia in the primary outcome, the time to discontinued use.

TABLE 3

Table 3. Cox proportional hazard model estimated the hazard ratio of discontinued use of antipsychotics in patients with schizophrenia.

It is interesting to mention that previous studies have shown multiple ULK4 functional variants including rs2272007 associated with multiple myeloma and blood pressure at genome-wide significance. A meta-analysis of 19 studies (n = 29,378) identified rs1717027 associated with diastolic blood pressure [19], which is in strong LD (r² > 0.90 and D' = 1) with two non-synonymous SNPs, rs1716975 and rs2272007 that we found here associated with autism. The missense variants rs1052501 and rs2272007 (r² = 0.98 and D' = 1) in ULK4 have been associated with multiple myeloma at genome-wide significance [20]. In addition, genome-wide association studies also identify that genetic variants at ULK4 are associated with blood pressure trait and hypertension in both European ancestry and Han Chinese populations [21,22]. The SNP rs2272007 had stronger cis-association with mRNA expression of ULK4 in the tissue sample of the artery and whole blood, providing molecular evidence for these genetic associations of ULK4 with both diastolic blood pressure and multiple myeloma. These four SNPs at ULK4 are in strong or complete pairwise LD in the CEU (r² > 0.909 and D' = 1), YRI (r² > 0.8 and D' = 1), and CHBJPT (r² = 1 and D' = 1) sample of the 1000 Genomes, which correspond to European ancestry, African and Asian population, respectively.

We do not know the exact biological mechanisms that ULK4 may implicate in the development of neurodevelopment disorders like autism. ULK4 and related genes are involved in cell cycle progression, which influences cell apoptosis and tumorigenesis [20], especially in the nervous system [32]. Autism risk-associated allele was associated with a higher level of gene expressions in postmortem prefrontal cortex and other brain areas or human tissues including artery and whole blood. Because multiple SNPs at ULK4 were associated with multiple myeloma, this suggests that the associated SNPs may result in gain-of-function. In addition, ULK4 deficiency mice are shown to have a significant decrease in thickness and superficial layers of the cortex, and this phenomenon was a uniform distribution in six sections of frontal cortex [16]. As neurons in six layers are generated from subventricular zone (SVZ) progenitors, ULK4 deficiency mice show reduction state of neural progenitors in the S phase, which is pivotal for DNA synthesis, and then influence normal progress of cell cycle. The principal role of G2 in the cell cycle is RNA and microtubules synthesis and it provides raw materials for spindle fiber assembly. Both in vivo and in vitro the ULK4 gene is highly expressed in G2/M phase of neural proliferating cells [16].

AUTHOR CONTRIBUTIONS

FZ designed this study. The clinical genetic data was from previous genome-wide association study of autism designed by KX, JZ and FZ, in which the Han Chinese cohort was designed KX and JZ and carried out by HG, ZH and JO; FZ obtained three cohorts of European ancestry samples (with Dr. David St Clair), gene expression of human brain samples and antipsychotics treatment response dataset. FZ, OJ, HG performed data analysis; JO, FZ and KL drafted the manuscript and all authors contribute to the manuscript significantly.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

ACKNOWLEDGMENTS

This study was supported by the National Natural Science Foundation of China (No: 81601197, 81730036, and 81525007), China Scholarship Council and direct funding from the Lieber Institute for Brain Development. We thank Kristin Bigos from Department of Pharmacology at the Johns Hopkins University School of Medicine for sharing the data and for the comments on the early draft of this manuscript.

How to Cite This Article

Ou J, Li K, Guo H, Xia K, Hu Z, Zhao J, Zhang F. ULK4 Genetic Variants Have Pleiotropic Effect on Risk of Autism, Associated with Brain mRNA Expression and Antipsychotic Treatment Response. J Psychiatry Brain Sci. 2019;4:e190010. https://doi.org/10.20900/jpbs.20190010