Whole-genome sequencing in multiplex families with psychoses reveals mutations in the SHANK2 and SMARCA1 genes segregating with illness.

A current focus in psychiatric genetics is detection of multiple common risk alleles through very large genome-wide association study analyses. Yet families do exist, albeit rare, that have multiple affected members who are presumed to have a similar inherited cause to their illnesses. We hypothesized that within some of these families there may be rare highly penetrant mutations that segregate with illness. In this exploratory study, the genomes of 90 individuals across nine families were sequenced. Each family included a minimum of three available relatives affected with a psychotic illness and three available unaffected relatives. Twenty-six variants were identified that are private to a family, alter protein sequence, and are transmitted to all sequenced affected individuals within the family. In one family, seven siblings with schizophrenia spectrum disorders each carry a novel private missense variant within the SHANK2 gene. This variant lies within the consensus SH3 protein-binding motif by which SHANK2 may interact with post-synaptic glutamate receptors. In another family, four affected siblings and their unaffected mother each carry a novel private missense variant in the SMARCA1 gene on the X chromosome. Both variants represent candidates that may be causal for psychotic disorders when considered in the context of their transmission pattern and known gene and disease biology.

Introduction

Recently, considerable advances have been made in the understanding of the genetics of psychiatric disorders, particularly schizophrenia. A Genome-Wide Association Study (GWAS) meta-analysis of 39,989 individuals with schizophrenia identified 108 loci with risk alleles for schizophrenia[1]. Notably, the risk alleles were all of modest effect, supporting a model of a highly polygenic disease characterized by small cumulative effects of a large number of common risk alleles. This effort was complemented by two recent exome sequencing studies: one searching for rare variants among 2,536 individuals with schizophrenia[2], and the other searching for de novo causative mutations among 623 trios[3]. The rare variant and de novo studies did not implicate any specific gene, but supported the role of synaptic pathways and the immune system in the etiology of the disease.

Collectively, these studies suggest that a large number of genes can contribute to disease risk and that no particular gene is specifically required or strongly over-represented among the risk loci. While many instances of schizophrenia may be polygenic (i.e., resulting from cumulative effect of modest-effect alleles), a subset may arise from single rare high-penetrance variants, possibly drawn from the large pool of genes influencing neurodevelopmental pathways. For example, several rare Copy Number Variants (CNVs) with a strong contribution to disease risk have been identified[4, 5], a risk possibly amplified by somatic copy number variation in the brain[6]. The high heritability ascertained from twin studies, however, suggests a degree of schizophrenia heritability that is not yet accounted for[7]. Here, we extend the search for rare highly-penetrant causative variants by focusing on families with a high prevalence of schizophrenia, as they may be more likely to harbor such variants. This approach has limitations, but can serve to suggest candidate variants that may be of particular value in the development of experimental models for schizophrenia and drug target discovery. The results reported here were obtained from whole genome sequencing of 83 individuals in 9 families. Variants identified include family private candidate variants in the SHANK2 post-synaptic density scaffolding protein and the SMARCA1 transcriptional regulator that might reasonably be expected to disrupt neuronal development or signaling.

Methods

Sample acquisition

Families with at least three members affected with schizophrenia were identified through advertisements placed primarily in local support group newsletters throughout the USA. The National Alliance for Mental Illness Chapters, initially in the Boston area and then throughout the USA, were approached for referrals and to advertise this study in their local newsletters. Several years ago, LE DeLisi also identified and evaluated families in a similar manner, and cell lines from these families were stored at the Coriell Institute in Camden, New Jersey. For the current analyses, five families’ cell lines (pedigrees p1250, p1271, p1274, p1333, and pSB285) were obtained from the Coriell Institute collection. Whole blood was obtained from an additional four families (pedigrees pVA02, pVA03, pVA04, and pVA07) from the later collection initiated in 2013. IRB approval was obtained at each institution where the data were gathered, and the overall study is currently approved by the VA Boston Healthcare System local IRB. All individuals signed written informed consent for their blood sample to be used for finding genes for risk for schizophrenia. All samples and corresponding clinical information were coded using both family and unique individual codes to mask identities. All individuals were interviewed using the Diagnostic Interview for Genetic Studies (http://www.nimhgenetics.org/interviews/digs_2.0/digs2.0.pdf), and diagnoses were made based on DSM-IV criteria by two independent investigators and consensus with a third if necessary.

Sample ancestry

Ancestry of individuals was assessed using iAdmix software (v0.2)[8] with the bundled HapMap 39 population allele frequencies. All individuals most closely resembled the CEU (Utah residents with Northern and Western European ancestry from the CEPH collection) and TSI (Toscani in Italia) populations, with the sum of the remaining population frequencies never exceeding 6% in any individual. Individuals in the pVA07 pedigree were most similar to TSI (68-78%), while the other individuals were most similar to CEU, with the exception of pVA04 individual 4, who was calculated to be 46% CEU, 51% TSI, 2% JPT (Japanese in Tokyo, Japan). When interviewed, these families all described their ancestry as solely European.

Sample preparation and sequencing

Cryopreserved B-lymphoblastoid cell lines were obtained from the Coriell Institute for Medical Research (Camden, NJ). Cells were thawed and grown in RPMI 1640 base medium supplemented with 15% fetal bovine serum in a T-75 flask, and placed in a humidified growth chamber at 37°C and 5% CO2. Cells were harvested at a density of 5×106 to 1×107 cells. Genomic DNA was isolated from cells using the Gentra Puregene cell kit (Qiagen, Inc.). Cell pellets were sent to Expression Analysis (Durham, NC) for genomic DNA isolation. Genomic DNA libraries were prepared and sequenced by deCODE Genetics in Iceland (Illumina 2×150 bp; 340-450 million reads/sample; 20-45x median CDS coverage). A total of nine families and 90 DNA samples were sequenced. After sequencing, QC analyses were performed and 83 samples remained (Figure 1; Supplementary Methods). For the purposes of reproducing our results, raw sequencing data are available on request.

Figure 1

Pedigrees selected for whole genome sequencing

Numbered individuals were sequenced. The numbers are arbitrary identifiers and are unique within each family. *Individuals with SHANK2 Variant. §Individuals with SMARCA1 variant. Within the diagnostic key, "Miscellaneous Diagnoses" represents diagnoses not otherwise mentioned, such as "anxiety disorder", anorexia, and non-schizophrenia-related personality disorders. However, no family members were diagnosed with autism or mental retardation. The diagnostic key also delineates the grouping of diagnoses into classes designated “healthy”, “affected”, and “other diagnosis”.

Sequence analysis

Detailed documentation of sequence analysis and quality control measures are provided in the Supplementary Methods. In brief, SNP and INDEL discovery was performed in adherence to the May 2015 version of the Genome Analysis Toolkit (GATK)[10] Best Practices recommendations[11, 12]. Reads were aligned to the hg19 reference genome using BWA[13], and then GATK was used to perform base quality score recalibration, indel realignment, duplicate removal, variant discovery across the full sample set, and filtering using variant quality score recalibration. Copy number variants were discovered using Genome STRiP (v2.00.1533)[14], and other structural variants were discovered using Lumpy (v0.2.11)[15]. Variants were annotated using the Ensembl Variant Effect Predictor[16], and SNPs and INDELs were then categorized according to rarity using variant calls and frequency annotations obtained from dbSNP (build v142)[17], 1000 genomes (release v5a)[18], and the Exome Aggregation Consortium (ExAC; Cambridge, MA; release r0.3; http://exac.broadinstitute.org). Annotation and manipulation of the VCF format variant calls utilized custom Java code and the Picard-tools Java library (v1.128; http://broadinstitute.github.io/picard/). We have focused this study on private and very rare variants within individual families that statistically represent very small numbers, and thus were not able to describe an odds ratio or a statistical probability of the null hypothesis. Seven sequenced samples were omitted from the analysis due to QC issues (see Supplementary Methods). Variant calls for both SHANK2 and SMARCA1 were verified with Sanger sequencing for each member of pedigrees p1274 and p1333. The two families served as negative controls for one another, as the SHANK2 variant was specific to pedigree p1274 and the SMARCA1 variant was specific to pedigree p1333.

Code Availability

The code and precise parameters used to run the DNA-Seq alignment and variant calling pipeline are described in detail in the Supplementary Methods. Custom code – used primarily for binning variants by rarity and quality control – is not provided, but is described in sufficient detail in the Supplementary Methods to reproduce the analysis.

Results

The total number of sequenced individuals remaining after application of quality control measures is shown for each pedigree in Table 1. Individuals diagnosed as schizophrenia or schizoaffective were classified as “affected” and served as the basis for candidate variant identification. Although we recognize that some mutations may cross diagnostic boundaries and also contribute to affective disorders, autism, and mental retardation, we chose not to include any affective disorders as "affected" in this initial exploratory search. No individuals in our families had diagnoses of autism or mental retardation. The analysis focused on variants that were (1) predicted to alter protein sequence, (2) present in all sequenced affected individuals, and (3) unique to that pedigree. These variants were classified as family private, and all 26 such variants are listed in Table 2. None of these variants are homozygous in the sequenced individuals, and none overlap with the 108 association loci identified by the recent large meta-GWAS study[1] or the set of de novo mutations in the recent trio exome-seq study[3]. To address the possibility that the uncommonly high disease incidence in our ascertained families simply resulted from unusual concentrations of common risk alleles, all sequenced individuals were analyzed for presence of common protective or risk alleles at index SNPs identified as markers of schizophrenia-associated loci defined with GWAS (see Supplementary Methods). Every individual carried a close balance of “risk” and “protective” common variants, and within each family the proportion of risk and common variants was the same in diseased and healthy individuals, making common variation an unlikely source of disease. For extended analysis, we also considered more relaxed selection criteria, such as including variants with low population-level allele frequencies and instances in which a variant is found in all but one affected individual (see Supplementary Table for extended list of variants and annotations). It is entirely possible that variants that segregate still less strongly with phenotype could contribute or predispose to disease. The filtering we describe below is of necessity somewhat arbitrary, and a full catalog of rare coding variants is provided in the Supplementary Table.

Table 1

Counts of sequenced individuals and variants by family, disease status, and variant classification.

	# Sequenced			# Variants in all Affected
Pedigree	Affected†	Other Diagnosis‡	Healthy	Family Private	Rare
p1250	4	1	2	2	12
p1271	3	5	1	9	29
p1274	6	2	1	4	9
p1333	4	0	2	4	16
pSB285	4	7	5	0	2
pVA02	4	3	1	1	1
pVA03	4	6	4	0	0
pVA04	3	3	1	3	6
pVA07	2	3	2	3	61

Diagnosis of ‘schizophrenia’ or ‘schizoaffective’

Diagnosis other than ‘healthy’, ‘schizophrenia’, or ‘schizoaffective’

Table 2

“Family private” variants present in all affected individuals.

Pedigree	Gene	Affected †	Healthy	Other Diagnosis ‡	Genomic Position	Variant	Impact§
p1250	KIT	4	1 of 2	1 of 1	4q12:55,603,415 T > C	I924T	Neutral (0.86)
p1250	HEXA	4	1 of 2	1 of 1	15q23:72,638,995 C > T	M401I	Neutral (0.76)
p1271	ZMYM6	3	1 of 1	2 of 5	1p34.3:35,453,774 A > C	L970R	Deleterious (1)
p1271	SPAG17	3	1 of 1	1 of 5	1p12:118,509,369 G > C	A2132G	Deleterious (1)
p1271	TCP1	3	0 of 1	2 of 5	6q25.3:160,201,497 C > G	D359H	Deleterious (0.99)
p1271	STAG3	3	1 of 1	1 of 5	7q22.1:99,794,880 T > G	I348S	Deleterious (1)
p1271	STK33	3	0 of 1	3 of 5	11p15.4:8,588,781 TCTGGTTCAGAGCTCA > T	<splice>	n/a
p1271	DDX55	3	1 of 1	1 of 5	12q24.31:124,086,789 A > G	M32V	Neutral (0.93)
p1271	NTRK3	3	0 of 1	2 of 5	15q25.3:88,472,466 C > G	D697H	Deleterious (1)
p1271	GREB1L	3	1 of 1	1 of 5	18q11.2:19,093,839 A > G	N1598S	Deleterious (1)
p1271	CEACAM16	3	0 of 1	3 of 5	19q13.32:45,211,144 G > A	A318T	Neutral (0.79)
p1274	STIP1	6	1 of 1	1 of 2	11q13.1:63,963,130 C > G	P173A	Neutral (0.97)
p1274	SHANK2	6	1 of 1	1 of 2	11q13.4:70,644,595 G > A	A578V	Deleterious (1)
p1274	RP11- 796G6.2	6	0 of 1	2 of 2	14q32.31:102,198,574 G > A	W112*	n/a
p1274	DEF8	6	0 of 1	1 of 2	16q24.3:90,030,588 G > A	R399Q	n/a Deleterious (1)
p1333	RNASEL	4	1 of 2	0 of 0	1q25.3:182,545,525 C > G	<splice>	n/a
p1333	KIAA1524	4	1 of 2	0 of 0	3q13.13:108,301,916 T > C	T89A	Neutral (0.85)
p1333	AVEN	4	1 of 2	0 of 0	15q14:34,331,202 C > T	G16S	Deleterious (1)
p1333	SMARCA1	4	1 of 2	0 of 0	Xq25:128,638,728 C > T	V384M	Deleterious (1)
pVA02	TGM6	4	0 of 1	2 of 3	20p13:2,384,113 G > A	V354I	Deleterious (0.98)
pVA04	HSD17B4	4	0 of 1	2 of 3	5q23.1:118,835,147 G > C	D395H	Neutral (0.86)
pVA04	MAN1A1	4	1 of 1	3 of 3	6q22.31:119,670,217 C > T	G5D	Neutral (0.87)
pVA04	ALPK3	4	0 of 1	2 of 3	15q25.3:85,400,551 G >	S1063N	Deleterious (0.99)
pVA07	BEND7	2	1 of 2	2 of 3	10p13:13,534,853 G > T	P147T	Neutral (0.92)
pVA07	WDFY4	2	1 of 2	2 of 3	10q11.23:49,982,576 T > A	V876D	Deleterious (1)
pVA07	WDR25	2	1 of 2	2 of 3	14q32.2:100,992,319 G > A	R405H	Deleterious (1)

Diagnosis of ‘schizophrenia’ or ‘schizoaffective’

Diagnosis other than ‘healthy’, ‘schizophrenia’, or ‘schizoaffective’

Protein impact is reported as a Carol[55] score aggregated from individual Sift[56] and PolyPhen[57] scores

In order to find rare variants that may be causal for schizophrenia, we further evaluated the 26 “family private” SNPs and INDELs in the context of supporting data, including predicted deleterious impact, transmission genetics, presence in individuals with other diagnoses (i.e. non-healthy and non-affected), and functional characterization. First, we considered the three potential disruptive variants. Family p1271 harbors the only non-SNP variant in our candidate variant set, a 15bp deletion that disrupts the splice donor site of the serine/threonine kinase STK33. However, the impacted splice site joins two untranslated 5’ exons in a non-canonical splice variant of dubious validity. Family p1333 harbors a splice acceptor mutation in RNASEL (ribonuclease L), present in all four affected children and transmitted through the healthy father. RNASEL is involved in interferon antiviral response[19], but no obvious connection to schizophrenia or neuronal biology has been reported. The final disruptive variant is a stop gain variant in the RP11-796G6.2 gene of Family p1274, but this gene appears to have been reclassified as a lincRNA in Ensembl GRCh38 (release 80)[20].

The missense SNPs predicted to have deleterious impact on protein function (Table 2; column labeled ‘Impact’) were also examined. Family p1271 harbors six such variants. Four were dismissed (ZMYM6, SPAG17, STAG3, and GREB1L) as inconsistent with our transmission hypothesis, as they were all transmitted from the married-in mother (individual 245; also the case for the neutral variant in DDX55). The remaining two variants impact the genes TCP1 and NTRK3. Connections between TCP1 and schizophrenia are tenuous. TCP1 is a molecular chaperone thought to be associated with schizophrenia in a Han Chinese population[21], but a follow-up attempt to replicate the result failed[22]. TCP1 was also reported to be significantly downregulated in a small comparison of schizophrenic and healthy postmortem brain samples[23]. NTRK3 is a neurotrophic tyrosine kinase receptor with reported associations with bipolar disorder[24, 25] and schizophrenia[26].

Family p1274 harbors two missense SNPs that are predicted to be deleterious. The first, in the gene DEF8, was transmitted through the bipolar father, and the second, in SHANK2, was transmitted through the healthy mother. DEF8 is very poorly characterized, but does exhibit elevated expression in the brain relative to most other normal tissues (GTEx consortium RNA-Seq dataset[27]). SHANK2 is a post-synaptic scaffolding protein at glutamatergic synapses with several connections to schizophrenia, and the A578V variant localizes to the SH3 consensus protein-binding domain[28-31].

Family p1333 also harbors two potentially deleterious missense SNPs, located in the genes AVEN and SMARCA1. AVEN is a caspase activation inhibitor [32] with a possible schizophrenia association reported in a small fine-mapping study of chromosomal region 15q13-q14[33]. SMARCA1 resides on the X chromosome, and thus exists in a hemizygous state in the three affected males and a heterozygous state subject to X inactivation in the unaffected mother and schizoaffective daughter. SMARCA1 is a member of the SWI/SNF family of chromatin remodeling genes and has multiple connections to neurogenesis and schizophrenia (see Discussion).

The two candidate genes in family pVA07, WDFY4 and WDR25, are both poorly characterized. Analysis of this family was compromised by QC issues that reduced the number of sequenced affected individuals to two. In family pVA04, a single family private potentially deleterious variant in the gene ALPK3 was identified. This kinase has no known connection to neurobiology. Finally, in family pVA02, the gene TGM6 was identified as a potential candidate. TGM6, a member of the transglutaminase family, may play a role in neurogenesis[34] and has been implicated in dominant spinocerebellar ataxias[35].

Existing algorithms for discovery and analysis of structural variants in whole genome sequence are far less robust than those for SNPs and small INDELs. Preliminary attempts to identify structural variants yielded numerous probable false positives (see Supplementary Methods), and were likely further compromised by false negatives and inexact breakpoint localization. However, one interesting family private chr6:chr11 translocation with strong support from reads spanning the predicted breakpoint was observed. This structural variant was transmitted from the healthy mother to all affected children in family p1333 and disrupts the TEAD1 gene, which encodes a transcriptional activator.

Discussion

Through whole genome sequencing of multiplex families, 25 SNPs and one small deletion were identified that potentially alter or disrupt protein sequence, are private to a single family, and are observed in all sequenced affected individuals. This study was not intended as an exhaustive attempt to assign a genetic basis for disease to each pedigree, but rather as an exploratory search designed to identify rare instances in which the schizophrenia spectrum phenotype might result from a single high penetrance variant. This approach has yielded several variants with strong contextual support in the form of pre-existing mechanistic and genetic data, among which the SHANK2 variant in family p1274 is perhaps the most compelling. Rare predicted-damaging variants segregating with disease but without such contextual support were also found (Table 2), and from this study alone we do not exclude these variants as playing a role in disease. If relevant to disease, such variants may play an undiscovered role in neuronal development or signaling.

Along with gene family members SHANK1 and SHANK3, SHANK2 is an integral part of the post-synaptic architecture of glutamatergic synapses. Recently, Peykov and colleagues[36] reported a significant enrichment of rare and common SHANK2 variants in schizophrenic individuals and demonstrated that some variants impaired synaptic clustering in hippocampal neurons. Unlike the variants in the Peykov study, the A578V SHANK2 variant identified in this study lies within the SH3 domain and thus might result in an even more severe phenotype. The Peykov study variants do share one feature with the A578V variant: the observed transmission was from heterozygous unaffected mother to heterozygous affected son (for all five variants that the study was able to trace). While this trend might arise by chance, it is also possible that the transmission from healthy mother to seven sons observed in family p1274 reflects an as yet uncharacterized sex-dependent influence on developmental pathways.

SHANK2 has also previously been reported as being associated with autism, intellectual disability, and other neuropsychiatric disorders[37, 38], as has SHANK1[39]. Interestingly, one familial autism-associated variant included deletion of the SH3 domain, and a population study found overrepresentation of missense variant S557N within SH3 in autism probands[40]. It is unknown how different variants in SHANK2, even within the SH3 domain, might predispose differently to autism, intellectual disability, or schizophrenia - clinically distinct syndromes that in some instances are enriched for the same or similar genetic associations. However, these variants offer tools for further study. Within the SHANK family, SHANK3 also has been implicated in schizophrenia via rare missense and stop-gain variants[41, 42], and a noncoding variant near SHANK1 has been linked to mild cognitive deficits in schizophrenia[43].

While SHANK2 represents a compelling candidate for a high penetrance variant, it alone cannot fully explain disease transmission within the p1274 family. The bipolar father (individual 244; Figure 1) does not harbor the SHANK2 variant, yet he also has a bipolar brother and schizoaffective nephew. It is possible that the family private DEF8 variant, which was transmitted through the father, contributes to the phenotype. However this gene is very poorly characterized and lacks the preponderance of connections to synaptic function and neurodevelopmental disorders of the SHANK2 variant. More broadly, this ambiguity reflects the limitations of the multiplex variant analysis, which involves too few samples to power statistical genetic approaches. In many instances the families might have been enriched for high background levels of variants of low effect size, and in others high-penetrance variants might have escaped notice because they resided in regulatory regions, involved structural variants that evaded detection, or were obscured by our stringent selection criteria.

Several of the other families also harbored variants that represent compelling candidates, though perhaps none as compelling as SHANK2. Multiple variants occur in genes that have previously been reported as having some connection to neurodevelopment or schizophrenia, including NTRK3, SMARCA1, TCP1, and TGM6. However, while interesting, these associations should be viewed with caution, as the number of genes that have some such link within the vast body of literature is no doubt quite large. SMARCA1 is worthy of special note, as similar to the SHANK family, the SMARCA protein family has multiple plausible connections to schizophrenia and brain function. SMARCA1 has been implicated in neurogenesis in several mouse studies[44-46], and in humans a loss of function mutation was recently identified in an individual suffering from microcephaly and intellectual disability[47]. Furthermore, gene family member SMARCA2 has previously been implicated in schizophrenia[48, 49]. The SMARCA1 variant in family p1333 is X-linked, and was the only gene in our list of top candidates (Table 2) that had no wild-type allele expressed in the affected individuals (with the possible exception of the affected daughter, subject to X-inactivation state).

While functional effects of these variants are untested, and this study was not designed to conclusively identify high penetrance disease causing variants, we have identified several plausible candidates, including two particularly compelling candidates in SHANK2 and SMARCA1. Thus, the multiplex pedigree sequencing strategy can be considered as a useful complement to the larger sample size GWAS and rare variant studies that don't examine individuals from the same family. The observations from the pedigrees may make valuable contributions towards understanding the biology of specific genes and biological pathways implicated in schizophrenia.

The identification of multiplex families for genetic studies, however, is difficult, not because they are rare, but rather because the affected individuals are frequently unavailable for study due to suicide, family disruption, and the nature of the illness itself. While it is known that family studies as a whole have calculated morbid risks for schizophrenia as only 8-10 percent to specific first degree relatives[50], other studies surveying family history in general in people with schizophrenia find a much higher percentage of positive history[51]. Multiplex families are potentially enriched for extreme and high penetrance phenotypes, yielding variants that are experimentally useful for future studies. For example, mice with disrupted SHANK2 are viable and show behavioral abnormalities possibly correctable pharmacologically[52-54], which is encouraging for potential use of the SHANK2 pathway to generate tools that reflect molecular and cellular mechanisms of schizophrenia.