Rare Variants in Tissue Inhibitor of Metalloproteinase 2 as a Risk Factor for Schizophrenia: Evidence From Familial and Cohort Analysis

Abstract

Candidate gene and genome-wide association study based common risk variant identification is being complemented by whole exome sequencing (WES)/whole genome sequencing based rare variant discovery in elucidation of genetic landscape of schizophrenia (SZ), a common neuropsychiatric disorder. WES findings of de novo mutations in case-parent trios have further implied genetic etiology, but do not explain the high genetic risk in general populations. Conversely, WES in multiplex families may be an insightful strategy for the identification of highly penetrant rare variants in SZ and possibly enhance our understanding of disease biology. In this study, we analyzed a 5-generation Indian family with multiple members affected with SZ by WES. We identified a rare heterozygous missense variant (NM_003255: c.506C>T; p.Pro169Leu; MAF = 0.0001) in Tissue Inhibitor of Metalloproteinase 2 (TIMP2, 17q25.3) segregating with all 6 affected individuals but not with unaffected members. Linkage analysis indicated a maximum logarithm of the odds score of 1.8, θ = 0 at this locus. The variant was predicted to be damaging by various in silico tools and also disrupt the structural integrity by molecular dynamics simulations. WES based screening of an independent SZ cohort (n = 370) identified 4 additional rare missense variants (p.Leu20Met, p.Ala26Ser, p.Lys48Arg and p. Ile217Leu) and a splice variant rs540397728 (NM_003255:c.232-5T>C), also predicted to be damaging, increasing the likelihood of contribution of this gene to SZ risk. Extensive biochemical and knockout mouse studies suggesting involvement of TIMP2 in neurodevelopmental and behavioral deficits, together with genetic evidence for TIMP2 conferring SZ risk from this study may have possible implications for new therapeutics.

Introduction

Schizophrenia (SZ) is a severe and chronic neuropsychiatric disorder characterized by impairment in thought, behavior, and neurocognitive function. Important role of genetic factors in conferring risk to SZ is well documented over several years now, on the basis of twin, adoption and family studies,¹ together with high heritability estimates (~60%–80%).^2,3 Commensurate efforts are being made to decipher the genetics of psychiatric disorders, particularly SZ. A recent Genome-Wide Association Study (GWAS) meta-analysis of 39,989 individuals with SZ and 113075 controls identified 108 loci associated with SZ with small to modest effects,⁴ supporting the commonly accepted polygenic nature of the disease. Meanwhile GWAS based common risk variant identification has been complemented by whole exome sequencing (WES) and whole genome sequencing based rare variant discovery. WES findings of de novo mutations in case-parent trios have further implied a genetic etiology.⁵,^7–17 Though these mutations were not enriched in specific genes, they support the overall role of neuronal migration, synaptic transmission, signaling, transcriptional regulation, transport, glutamatergic postsynaptic proteins comprising activity-regulated cytoskeleton-associated protein (ARC) and N-methyl-d-aspartate receptor (NMDAR) complexes in SZ etiology.^7,12,18 However, such de novo mutations do not sufficiently explain the high inherited genetic risk observed in general populations. On the other hand, WES in multiplex families may be a fruitful and reliable strategy for the identification of highly penetrant rare variants in common diseases and this may also explain the heritable risk and provide a better understanding of disease biology. Recently, we and others have identified rare variants in a number of functionally relevant genes by WES of multimember affected families. These include UNC13B, SHANK2, SMARCA1, TAAR1 GRM5, LRP1B, and RELN.^19–23 Continuing our efforts in discovery genomics, in this study, we analyzed an Indian family with multiple members affected with SZ across 5 generations. We employed the WES approach to identify highly penetrant rare variant(s) that may co-segregate with disease phenotype in the family and identified a rare missense variant in Tissue Inhibitor of Metalloproteinase 2/TIMP metallopeptidase inhibitor 2 (TIMP2). Further, we identified 5 additional rare variants in this functionally relevant gene from among 370 SZ exomes available in the laboratory, thus providing the first ever genetic evidence for the role of this gene in SZ etiology.

Materials and Methods

Recruitment of Study Subjects

The detailed protocol for recruitment of study cohort has been reported previously.^24,25 Briefly, the samples were recruited from Dr RML hospital New Delhi, and were mostly of north Indian origin. Consensus diagnosis for SZ was made based on DSM IV criteria. Hindi versions of the Diagnostic Interview for Genetic Studies (DIGS) and the Family Interview for Genetic Studies (FIGS) were also administered to all the affected and unaffected individuals used in the study to get additional information. All study protocols were approved by the institutional ethical committees of both participating institutions and all the participants provided written informed consent. Venous blood was drawn from all those individuals and DNA isolation was done by the routine phenol chloroform method, which was further used for genetic analysis.

Family Used for WES

A 5-generation family with 11 members affected with SZ was used in the study (figure 1). Genomic DNA samples from 10 individuals from 4 generations were available, of which DNA from 3 affected members was used for WES. Diagnosis of the single individual from the fifth generation (V.1) could not be confirmed and therefore, that individual was excluded from further analysis.

Fig. 1.

Pedigree of the multiplex SZ family analyzed in the study.

WES and Data Processing

Agilent SureSelect Human All Exon V5 + UTR kit was used for library preparations and sequencing was performed in Illumina HiSeq2000 in paired end sequencing mode, using a commercial facility (Medgenome Labs Pvt. Ltd). Raw data generated were processed using standard tools and software. The protocols used are detailed previously.²¹

Variant Prioritization

DNA from 3 informative individuals (III.3, III.4 and IV.1) from the family (figure 1) was used for WES. Prioritization of the variants was performed as described previously.^26–28 Variants in the combined VCF generated from exome sequence data were annotated using wANNOVAR²⁹ Kggseq³⁰ and nonprotein coding variants removed. As the aim of the study was to identify highly penetrant rare variants, all the variants with minor allele frequency (MAF), MAF > 0.001 in public databases including 1000G, Exome Aggregation Consortium (ExAC r0.3.1) and Genome Aggregation Database (GAD), 6500 exomes, G40,dbSNP along with synonymous variants and those not shared among the 3 affected samples were removed. Variants shared among the 3 affected individuals were screened among the remaining family members by Sanger sequencing/custom target capture sequencing. Subsequently, all variants that were not shared among all the affected individuals in the family, variants from regions with segmental duplications and variants with MAF > 0.001 in non-SZ exome data (n = 150) available in the laboratory were also discounted. Finally, the variants that are present in unaffected individuals in the family were also removed. For the variant annotation and prioritization we used Kggseq software. Based on functional/biological relevance or their prior implication for SZ based on association, linkage, exome sequencing and animal studies, the variants shared among affected individuals were further prioritized. These variants were then confirmed by Sanger sequencing (Primer details in supplementary table 1).

Linkage Analysis.

To evaluate the likely linkage of this gene to SZ, additional analysis using 2 microsatellite markers (D17S949, D17S785) upstream of TIMP2 and 2 downstream markers (D17S784, D17S928) from ABI Linkage Mapping Sets V2.5 was carried out. PCR was performed as per the manufacturer’s protocol and products were run on ABI 3730xl DNA Analyzer at central instrumentation facility at University of Delhi South Campus, and analyzed using peak scanner software. Logarithm of the odds (LOD) scores were calculated.

In silico Analysis of the Prioritized Variant(s)

In silico characterization of the prioritized and Sanger sequencing confirmed variant(s) were performed by 14 different tools SIFT, Polyphen2_HDIV, Polyphen2_HVAR, LRT, MutationAssessor, MutationTaster, FATHOM, PROVEAN, MetaSVM, MetaLR, RadialSVM, LR, Variant Effect Scoring Tool3 (VEST3) and Combined Annotation Dependent Depletion (CADD) score which were all available in dbNSFP2.9.³¹ In addition we used Predictor of human Deleterious Single Nucleotide Polymorphisms (PhD-SNP)³² and CONsensus DELeteriousness (CONDEL).³³ We also examined the evolutionary conservation of the confirmed variants using GERP++_RS phyloP46way_placental, phyloP100way_vertebrate, SiPhy_29way_logOdds, GERP++_NR and for gene level pathogenicity estimation we used haplo-insufficiency estimation and Residual Variation Intolerance Score (RVIS).

Molecular Dynamics Simulation.

The X-ray crystallographic 3D structure of the standalone TIMP2 protein available in the Protein Data Bank (PDB code 1BR9) was used to study the effect of the Pro169Leu variant on the protein structure using Molecular Dynamics simulations (MDs). The Pro169 residue in the wild-type protein structure (present at position 143 in 1BR9 which is missing the first 26 amino acid residues corresponding to a signal peptide) was mutated using the mutagenesis option of the Swiss-PdbViewer (SPDBV) software.³⁴ Both the wild-type and variant structures were subject to a 100-nanosecond simulation run each, using GROMACS 5.1.2 software.^35,36 The force-field used for the simulation was CHARMM27 all-atom force field.^37,38 Both the structures were subject to the same simulation parameters. The protein was solvated using the TIP3P water model in a dodecahedron solvation box. A net positive charge (+2) on the protein structures at physiological pH was balanced by adding 2 chloride ions in the solvent, bringing the entire system to a neutral charge. Energy minimization was performed for a maximum of 500000 steps using the steepest descent algorithm. After the structures converged to a maximum force < 1000 kJ/mol/nm, a 2-step equilibration using the NVT ensemble (isothermal-isochoric) followed by the NPT ensemble (isothermal-isobaric) was performed at 300K, 1.0 atm and 10,000 ps, 300K, respectively. The covalent bonds were constrained using the LINCS (Linear Constraint Solver) algorithm,³⁹ while Particle Mesh Ewald (PME) was used to treat the electrostatic interactions.⁴⁰ A coulomb and Vander Waals interactions cut off radii of 10.0 and 14.0 Ǻ were set and MDs were carried out at 300K for 100 ns with the MD trajectories being saved after every 2.0 ps. The MD trajectories were analyzed to obtain the root-mean-square fluctuation (RMSF) as a function of the Cα atoms of protein residues using the Gromacs gmx rmsf utility.³⁶

Results

WES in the Index Family

The mean target depth of the 3 samples used from the family for WES was 54X and on an average, >99% of the target region at 1X and >87% of the target region at 20X were covered. The mean mapping quality of the samples was 47 and detailed statistics are provided in supplementary table 2. Of the 60 variants following prioritization (table 1), only 2 variants one each in TIMP2 and PIWIL3 segregated in the family. Though PIWIL3 is expressed in the brain and could be functionally relevant, due to the predicted tolerant nature of the variant p.Gly31Glu by all the software used, its poor conservation observed across species and an extremely low CADD score of 0.59 (supplementary table 3) it was removed from further analysis.

Table 1.

Prioritization of Variants in the Whole Exome Data from the Index Family

Filtering Criteria	Number of Remaining Variants
Total coding variants	31573
On removal of variants based on global minor allele frequency (MAF < 0.001) in 1000G, ExAC, Exome Variant Server, CG40, dbSNP, gnomAD browser	1320
On removal of variants based on population specific frequency (MAF < 0.001) in 1000G, ExAC, Exome Variant Server, gnomAD browser	850
Total number of protein disrupting variants (Synonymous variants removed)	605
Total number of homozygous variants shared among 3 affected (Exome sequenced)	0
Total number of heterozygous variants shared among 3 affected (Exome sequenced)	128
Total heterozygous variants present in all affected (Screened by target capture sequencing)	60
On removal of variants from regions with segmental duplication	45
On removal of >4 variants per gene	22
On removal of variants present in in-house control sample MAF > 0.001	15
Total number of variants segregating with the phenotype	2
Number of heterozygous segregating variants functionally relevant based on literature	2

On the other hand, the rare heterozygous missense variant in TIMP2 (17q25.3) (NC_000017.10: g.76851906G>A; NM_003255.4: c.506C>T; NP_003246.1:p.Pro169Leu; Exon 5; MAF 0.0001 in ExAC 0.3.1) in TIMP2 present in all 6 affected individuals but absent in the 3 unaffected members (supplementary figure 1) seemed promising. Expression levels of this gene in the different brain regions were checked using 2 different databases namely BrainSpan (http://www.brainspan.org) and Genotype-Tissue Expression (GTEx) portal (https://www.gtexportal.org/). The protein was expressed across the brain regions in all the developmental stages.

Linkage Analysis.

Linkage between D17S785 (2.42 Mb upstream) and D17S784 (0.95 Mb downstream) was estimated with maximum LOD score 1.8 (θ = 0) confirming its linkage.

In silico Analysis

The index variant p.Pro169Leu was highly conserved across species (supplementary figure 2; supplementary table 3), predicted to be damaging by 13 different in silico tools and also predicted to be among the top 1% most deleterious substitutions in the human genome by CADD (supplementary table 3). Gene level analysis revealed TIMP2 to be among the top 31% of intolerant genes and with haploinsufficiency.

Effects of the Index Variant on TIMP2 Structure

The RMSF analysis of MD trajectories showed that although the fluctuations for the WT and the variant structures were nearly concordant for most of the protein’s length, there were notable deviations in the vicinity of the reported variant at the 143rd position (supplementary figure 3). The stretch of residues between positions 130–155, corresponding to the GH loop and its neighboring regions in the C-terminal domain (CTD), showed a marked deviation in the fluctuations of Cα atoms (to the order of ~0.5 nm to 1.0 nm). This indicates that the variant leads to an increased flexibility and disorder in the backbone atoms of CTD of TIMP2, possibly due to disruption of disulphide bridges which stabilize this domain. This structural instability as a consequence of the proline to leucine amino acid change is suggestive of aberrant functionality of the variant protein.

Additional Variants in TIMP2 Identified in an Independent SZ Cohort.

The index variant and additional variants, if any, in TIMP2 exons were screened in an independent SZ cohort comprising of 370 SZ samples, mostly from north India, recruited from Dr. RML hospital, New Delhi. A total of 5 additional coding variants were identified. Three of the novel variants namely p.Leu20Met and p.Ala26Ser in exon 1 and p.Lys48Arg in exon 2 were seen in one individual each. In addition a rare missense variant rs578083142 (NM_003255.4:c.649A>C; p.Ile217Leu) in exon 5 with MAF = 0.0006, in ExAC r0.3.1 and GAD (South Asian population) and a rare splice variant rs540397728 (NM_003255.4:c.232-5T>C) in exon 3 with MAF = 0.0001 in ExAC r0.3.1 and GAD (South Asian population) were observed in 2 other SZ exomes. All these 5 variants were also conserved across species and predicted to be damaging by 1 or more of the in silico tools (supplementary table 4) and were notably absent in 150 non-SZ exome data available in the laboratory.

Discussion

In complex diseases especially in SZ, a likely role of de novo variants in disease development has been evident from previous case-parent studies.⁴¹ Though high heritability estimated for this disorder reiterates the role of inherited variants, family based studies are still limited. The multiplex family (figure 1) analyzed by WES in this study is indeed informative and based on the justification provided below, the rare heterozygous missense variant (p.Pro169Leu, exon 5) in TIMP2 seems to be the most promising risk variant in this family. Considering that this variant has segregated with the phenotype in 6 independent meiotic events (of the 10 that could be tested) across 2 generations (figure 1), it may be argued that it is unlikely to be a chance event. This is supported by our suggestive linkage findings [LOD score of 1.8 (θ = 0) with 2 flanking markers at 2.42Mb (D17S785) and 0.95Mb (D17S784)]. Considering the generally accepted polygenic nature of SZ, the role of common variants with MAF > 0.001 in this family cannot be ignored. However, this was not addressed in the current study whose focus was to identify only highly penetrant rare variants in a familial setting. Identification of 5 additional rare variants among 370 exomes derived from an independent SZ cohort (supplementary table 4) increased the likelihood of the possible contribution of this gene to SZ risk. This is further strengthened by the observation of 6 variants with MAF ≥ 0.01 nominally associated (P ≤ 0.05) with SZ in the Psychiatric Genomics Consortium (PGC) data (http://www.med.unc.edu/pgc; supplementary table 5).

TIMP2 is expressed throughout brain tissues and at all stages of brain development and adulthood reiterating its functional relevance. Further, likely functional significance of the rare variant derives support from in silico predictions wherein several tools showed this variant to be damaging (supplementary table 3). These predictions are strengthened by the findings of MD based structural analysis (supplementary figure 3). Structural details of the interactions of TIMP2 protein with some of its metalloproteinase binding partners (pro-MMP2,⁴² MMP10,⁴³ MMP13,⁴⁴ and CDMT1-MMP⁴⁵) are well documented. In general, the N-terminal domain (NTD) of TIMP2 has been implicated in the inhibitory interactions with various MMPs,^43–45 with the CTD also possibly involved in some stabilizing interactions.⁴⁶ This non-synonymous variant can potentially disrupt some disulfide bonds which stabilize the CTD and this disruption in the protein structure can contribute to a loss of function, but this warrants functional validation.

TIMP2 is located at 17q25.3, a region which has been previously implicated in both SZ and bipolar disorder (BD) based on family based linkage studies. In a study of 250 BD pedigrees of mainly Caucasian ancestry, a maximum LOD score of 3.63 was observed at this locus.⁴⁷ Another study on 22 Canadian pedigrees of Celtic or German descent with SZ identified linkage at the same locus.⁴⁸ Importance of 17q25.3 locus is also available from another study, which reported a balanced translocation involving chromosome 17 t(9;17)(q33.2;q25.3) in 2 SZ affected members of a small family with diverse psychiatric disorders.⁴⁹ Interestingly, 4 protein coding genes namely ENDOV, NPTX1, RNF213, and RPTOR are reported to be present in the break point region at 17q25.3. Association studies of SNPs at this locus revealed significant association of NPTX1 with BD (P = 0.004) and ENDOV with SZ (P = 0.007) which withstood multiple testing.⁵⁰

Substantial literature on TIMP2 is available based on biochemical and animal studies. TIMP2 is a member of TIMP gene family which are natural inhibitors of the matrix metalloproteinases (MMPs). Of note, decreased expression of TIMP2, MMP2, and MMP9 at both mRNA and protein levels in sera samples of subjects with depression and conversely, a direct correlation between elevated expression of these 3 genes and improved cognitive functions have been reported.⁵¹

TIMP2 has been demonstrated to be involved in extra cellular matrix remodeling together with MMPs.⁵² This in turn is notably involved in various neurodevelopmental processes such as neurogenesis, differentiation, axonal growth, formation of axonal tracts, stabilization/maturation of synapses, regulation of synaptic plasticity, remodeling. In addition its involvement in neuro transmission, long-term potentiation (LTP), long-term depression and myelination have also been reported.^53–57

TIMP2 inhibits angiogenic factor mediated angiogenesis in vivo⁵⁸ and possible role of angiogenesis in the pathophysiology of SZ has been suggested previously.⁵⁹ On the other hand, through the MMP independent activity, TIMP2 is known to induce cell cycle arrest and subsequent neuronal differentiation and neurite outgrowth in PC12 cells.⁶⁰ Varying roles of TIMP2 have been reported in several pathways namely VEGFR, EGFR, FGFR1, AKT, and Wnt/β-catenin signaling,^60–63 all previously implicated in SZ.^6,64–70

Besides these studies, TIMP2 has been studied extensively in mouse models including knockout mice and its role in neurodevelopment and behavioral deficits has been reiterated. Role of TIMP2 in neuronal differentiation and neurite outgrowth,⁶⁰ motor function,⁷¹ synaptic plasticity, hippocampal development, cognition and LTP⁷² have been well documented in these studies on knockout mice. Knockout mice also did not exhibit prepulse inhibition of the startle reflex but showed increased anxiety behavior. These functions have been speculated to be modulated by TIMP2 similar to other ECM proteins like reelin or neural cell adhesion molecule (N-CAM),⁷³ Further, repeated treatment of methamphetamine in rats leads to increase in TIMP2 expression in frontal cortex which has been shown to play an important role in drug induced behavioral sensitization and rewards, confirmed by antisense TIMP2 oligonucleotide treatment. The authors suggested that TIMP2 may be involved in the neural network rearrangement in the mesocorticolimbic dopamine system, particularly through the dopamine D2 receptor.⁷⁴

Taken together, identification of the rare heterozygous missense variant p.Pro169Leu in TIMP2 in the study family, 5 additional rare protein disturbing variants in an independent SZ cohort and a large number of functional studies reported in literature, support the likely contribution of protein coding variants in TIMP2 to SZ etiology. This first genetic evidence could pave the way for new therapeutic opportunities with implications for personalized medicine.

Funding

Grant #BT/MB/Project-Schizophrenia/2012–2013 and #BT/PR2425/Med13/089/2001 to B.K.T. and S.N.D. from the Department of Biotechnology, Government of India, New Delhi, India; Grant #MH093246, #MH063480, and #TW009114 to V.L.N. from National Institute of Mental Health, the Fogarty International Center, USA; Junior and Senior Research Fellowship (09/045(1166)/2012-EMR-I) to J.J. from Council for Scientific and Industrial Research, New Delhi; Junior and Senior Research Fellowship from the Department of Genetics under University Grants Commission (UGC)-Special Assistance Program Meritorious Award scheme to A.S.; and DSK-PDF (BL/13-14/0404) to P.K. from UGC, New Delhi are gratefully acknowledged.