Genome-Wide Association Study of NMDA Receptor Coagonists in Human Cerebrospinal Fluid and Plasma

ORIGINAL ARTICLE Genome-wide association study of NMDA receptor coagonists in human cerebrospinal ﬂuid and plasma

JJ Luykx1,2,3, SC Bakker3, WF Visser4, N Verhoeven-Duif4, JE Buizer-Voskamp3, JM den Heijer3, MPM Boks3, JH Sul5, E Eskin5,6, AP Ori7, RM Cantor6, J Vorstman3, E Strengman6, J DeYoung7, TH Kappen8, E Pariama9, EPA van Dongen10, P Borgdorff9, P Bruins10, TJ de Koning4,11, RS Kahn3 and RA Ophoff3,6,7

The N-methyl-D-aspartate receptor (NMDAR) coagonists glycine, D-serine and L-proline play crucial roles in NMDAR-dependent neurotransmission and are associated with a range of neuropsychiatric disorders. We conducted the ﬁrst genome-wide association study of concentrations of these coagonists and their enantiomers in plasma and cerebrospinal ﬂuid (CSF) of human subjects from the general population (N = 414). Genetic variants at chromosome 22q11.2, located in and near PRODH (proline dehydrogenase), − 10 were associated with L-proline in plasma (β = 0.29; P = 6.38 × 10 ). The missense variant rs17279437 in the proline transporter − 9 SLC6A20 was associated with L-proline in CSF (β = 0.28; P = 9.68 × 10 ). Suggestive evidence of association was found for the − 8 D-serine plasma-CSF ratio at the D-amino-acid oxidase (DAO) gene (β = − 0.28; P = 9.08 × 10 ), whereas a variant in SRR (that encodes serine racemase and is associated with schizophrenia) constituted the most strongly associated locus for the L-serine to D-serine ratio in CSF. All these genes are highly expressed in rodent meninges and choroid plexus, anatomical regions relevant to CSF physiology. The enzymes and transporters they encode may be targeted to further construe the nature of NMDAR coagonist involvement in NMDAR gating. Furthermore, the highlighted genetic variants may be followed up in clinical populations, for example, schizophrenia and 22q11 deletion syndrome. Overall, this targeted metabolomics approach furthers the understanding of NMDAR coagonist concentration variability and sets the stage for non-targeted CSF metabolomics projects.

Molecular Psychiatry (2015) 20, 1557–1564; doi:10.1038/mp.2014.190; published online 10 February 2015

INTRODUCTION and NMDAR coagonists in SCZ comes from the most recent and The N-methyl-D-aspartate receptor (NMDAR) coagonists glycine, extensive genetic study of SCZ, in which robust evidence of D L D association was found for serine racemase (SRR) that encodes an -serine, -proline and possibly -alanine play crucial roles in 28 NMDAR-mediated neurotransmission. Although D-serine primarily enzyme that converts L-toD-serine. 1,2 Heritability estimates for amino acids in serum range from 0.23 mediates synaptic NMDAR gating, glycine acts on extraynaptic 29 2,3 to 0.55. Genome-wide association studies (GWASs) of amino- NMDARs. D-alanine and L-proline, although understudied in preclinical settings, also act on NMDAR-binding sites.4–8 acid levels in urine and serum have yielded quantitative trait loci NMDAR coagonists are involved in neuropsychiatric disorders, (QTLs) in genes encoding enzymes (for example, NAT2 and AGXT2 as demonstrated by pharmacological and genetic findings. For (ref. 30)) and several members of the solute carrier family (for example, SLC2A4, SLC25A1 (ref. 29) and SLC6A20 (ref. 30)). example, hyperprolinemia is associated with schizophrenia (SCZ)9 10 Such associations were detected for (ratios of) several amino acids, and schizoaffective disorder. Decreased levels of D-serine have fl 11,12 whereas L- and D-enantiomers have not yet been subjected to been reported in cerebrospinal uid (CSF) of SCZ patients and GWASs. More importantly, it is unknown how urine and serum low glycine levels were observed in CSF of patients suffering from fi 13 ndings should be interpreted for the central nervous system affective disorders. In addition, individuals who smoke have (CNS) given the presence of the blood–brain barrier. When aberrant levels of D-proline in CSF and plasma, and of D-serine in studying the genetic control of NDMAR coagonists in the CNS, CSF 14 CSF. Furthermore, meta-analytic evidence indicates SCZ patients may be the most informative body fluid as it is most proximate to have increased blood serine levels compared with healthy the CNS. In addition, the critical role of CSF in CNS signaling was 15 controls. Moreover, D-cycloserine, D-alanine and glycine consti- recently demonstrated.31 Furthermore, although knowledge tute (experimental) adjuvant pharmacological options in the regarding the differential effects of amino-acid L- and D-enan- – – treatment of SCZ16 22 and anxiety disorders.23 27 Finally, strong tiomers on the CNS has been accumulating over the past decade, genetic evidence implicating glutamatergic neurotransmission the genetic mechanisms underlying stereoisomer concentration

1Human Neurogenetics Unit, Brain Center Rudolf Magnus, University Medical Center Utrecht (UMCU), Utrecht, The Netherlands; 2Department of Psychiatry, ZNA Hospitals, Antwerp, Belgium; 3Department of Psychiatry, Rudolf Magnus Institute of Neuroscience, University Medical Center Utrecht (UMCU), Utrecht, The Netherlands; 4Department of Metabolic Diseases, University Medical Center Utrecht/Wilhelmina Children's Hospital, Utrecht, The Netherlands; 5Computer Science Department, University of California, Los Angeles, CA, USA; 6Department of Human Genetics, University of California, Los Angeles, CA, USA; 7Center for Neurobehavioral Genetics, Semel Institute for Neuroscience and Human Behavior, University of California Los Angeles, Los Angeles, CA, USA; 8Department of Anesthesiology, Intensive Care and Pain Management, University Medical Center Utrecht (UMCU), Utrecht, The Netherlands; 9Department of Anesthesiology, Intensive Care and Pain Management, Diakonessenhuis Hospital, Utrecht, The Netherlands; 10Department of Anesthesiology, Intensive Care & Pain Management, St Antonius Hospital, Nieuwegein, The Netherlands and 11University of Groningen, University Medical Center Groningen, Department of Genetics, Groningen, The Netherlands. Correspondence: Professor RA Ophoff, Center for Neurobehavioral Genetics, Semel Institute for Neuroscience and Human Behavior, University of California, Los Angeles, 695 Charles E. Young Drive South, Gonda Building, Los Angeles, CA 90095, USA. E-mail: [email protected] Received 18 September 2014; revised 5 November 2014; accepted 8 December 2014; published online 10 February 2015 GWAS of NMDA receptor coagonists JJ Luykx et al 1558 variations remain elusive. For instance, it has not been established frequency ⩾ 0.05), a total of 5 749 079 autosomal SNPs (567 521 geno- 2 whether conversion from L-serine constitutes the prime source of typed) remained for the association analyses. Imputation quality scores (r ) 32,33 fi endogenous D-serine. The discovery of the D-amino acid of imputed SNPs that showed signi cant or suggestive associations were synthesizing enzyme SRR in the mammalian brain and its as follows: rs2904552 and rs3970551 had respective scores of 0.64 and implication in neurotransmission have illustrated the relevance 0.60, whereas the other quality scores ranged between 0.84 and 1.0. Other 34–36 22q11 SNPs mentioned in the discussion were either genotyped— of D-amino-acid signaling molecules for the CNS. As genes rs4819756 (R185W)—or had an imputation r2 of 0.93—rs2008720 (Q19P). that moderate D-amino-acid concentration variation have not To assess the validity of the imputation process for the most significant been established, quantitatively analyzing the genetic mechan- locus, we carried out direct genotyping of this locus (rs2540641). To that isms underlying such traits may elucidate D-amino-acid metabo- end, a custom TaqMan SNP Genotyping Assay (Applied Biosystems—cat lism and transporter pathways. Moreover, clarifying the genetic nr: 4331349; Foster City, CA, USA) was designed. Samples were run underpinnings of NMDAR coagonist concentration variations may according to the manufacturer’s protocol. Briefly, 10 ng of purified DNA deepen the understanding of NMDAR physiology. Such insight, in was combined with SNP Genotyping Assay and TaqMan Universal Master turn, may open avenues for preclinical studies aimed at dissecting Mix II—no UNG (Applied Biosystems—cat nr: 4440040). Samples were run the localization of NMDAR coagonist-dependent gating2 and on Applied Biosystems 7900HT Fast Real-Time PCR System on a 384-well μ clinical applications, for example, improvements in add-on reaction plate using 5 l of total reaction volume per well. The SDS software pharmacotherapy in SCZ and disease prediction, as was shown (v2.4) was used to analyze the data and endpoint allelic discrimination was 37 performed to determine the genotypes. Even better than predicted by the for diabetes. imputation quality score of 0.97, only two genotypes differed between the We thus set out to study the genetic underpinnings of variation two techniques. QQ-plots were generated and genomic inflation factors in NMDAR coagonists in both plasma and CSF by analyzing were computed using the R package (www.r-project.org). genome-wide data of 414 healthy human subjects. We hypothe- sized that our design of measuring both L- and D-amino acids in Neurometabolic quantitative trait locus analyses two body fluid compartments using a tandem mass spectrometer in addition to liquid chromatography would augment power to The statistical analyses described here are based on a previous metabolic trait GWAS.30 Outliers (defined as data points 43 s.d. away from the mean) detect single-nucleotide polymorphisms (SNPs) relevant to were excluded. Distributions deviating from normality (defined by a biological processes underlying amino acid turnover in the CNS. Kolmogorov–Smirnov two-sided P-valueo0.05) were log scaled (see Several genome-wide significant neurometabolic QTLs in genes Supplementary Methods for an overview of the traits that were thus log encoding transporter proteins and enzymes were detected. scaled and therefore for an overview of whether the log-transformed trait was used for further analyses). We applied an age- and sex-adjusted addi- tive linear model in Plink v1.07 (ref. 41). Other covariates tested using MATERIALS AND METHODS Spearman’s rho correlations with neurometabolic traits were: time of day Subjects and collection of samples of sampling, CSF amount (as in 8% of the samplings the amount of CSF suctioned differed 41 ml from the planned 6 ml), use of psychotropic The ethics committee at the University Medical Center Utrecht (UMCU) medication (binary), use of other medication (binary), storage duration at and all local ethics committees approved this study. This research was − 80 °C, current or past self-reported psychiatric diagnosis (binary) and type conducted according to the principles expressed in the Declaration of of elective surgical procedure (arthroscopy vs other). These variables were Helsinki. Written informed consent was obtained from the participants. assessed by telephone interviews, online questionnaires and perioperative Subject description and collection of samples have been described in recordings. Storage time was added as a third covariate to our model, as detail previously (also see Supplementary Methods).38,39 In brief, we this was the only impacting variable on ⩾ 2 NMDAR coagonists at Po0.05 included 414 subjects (i) undergoing spinal anesthesia for minor elective (see Supplementary Methods for the 15 metabolites or ratios influenced by surgical procedures, (ii) aged between 18 and 60 years, and (iii) with four storage duration). We used a genome-wide significance threshold of grandparents born in The Netherlands or other North-Western European − − α =(5×10 8)/4 = 1.25 × 10 8 reflecting a correction for the number of countries (Belgium, Germany, UK, France and Denmark). Each candidate NMDAR coagonists (N = 4). Suggestive evidence for association was set at participant received a personal telephone interview to exclude subjects − Po10 7. Because of the strong correlations between the chiral forms and with self-reported psychotic or major neurological disorders (stroke, brain measured values in CSF and plasma (Supplementary Table S1), we used a tumors and neurodegenerative diseases) and to record any use of less stringent correction factor than correction for the 27 traits (including (psychotropic) medication. Whole blood was collected in EDTA tubes for ratios). Moreover, considering this is the first GWAS of CSF neurometabo- DNA and plasma extraction; 6 ml of CSF were suctioned and stored at lites, we thus increased power at the potential cost of type-I error. As a test − 80 °C. Body fluids (plasma and CSF from the same individuals) were − of robustness we carried out 1/1.25 × 10 8 =8×107 permutations in Plink simultaneously obtained between 800 and 1600 hours. v1.07 to establish whether the null hypothesis—lack of association—could be rejected. This label swapping procedure annuls all genotype–pheno- NMDAR coagonist measurements type relationships, thereby providing a virtual data set sampled under the 41 fi We quantified glycine (that is not chiral) and L- and D-enantiomers of null hypothesis. Permutation testing indeed con rmed the results for serine, proline, and alanine in CSF and plasma using ultra-performance L-proline in plasma (rs2540641 and all variants in complete LD), the liquid chromatography–mass spectrometry, as described previously, that L-proline plasma-CSF ratio (rs17279437) and the CSF L-proline (rs73058498 is, after overnight fasting.40 and rs73060324) associations (that is, all empirical P-values were o1.25 × 10 − 8). An empirical P-value of 2.5 × 10 − 8 was computed for the QTL at rs17279437 with L-proline in CSF. We next computed ratios between Whole-genome genotyping, imputation, quality control plasma and CSF concentrations of the same amino acid and ratios procedures and TaqMan assay between L- and D-stereoisomers within the same compartment (resulting in Genotype data of 414 subjects were collected using the Illumina 1 ratio for glycine and 4 for the other amino acids). In GWASs of serum and HumanOmniExpress Beadchip (San Diego, CA, USA; 730 525 SNPs), at the urine metabolites, using ratios may yield QTLs additional to the ones UCLA Neurosciences Genomics Core (UNGC). All genetic analyses were detected for measured values and increase power.30,37,45,46 Moreover, performed using Plink v1.07.41 Quality control was performed as described targeting ratios between amino acids measured in different body fluids in the Supplementary Methods, leaving 398 individuals. We used the may uncover genetic determinants of transport mechanisms between BEAGLE42 software to phase the genotype data and the minimac43 body compartments. Computation of explained variance, power calcula- software (a computationally efficient implementation of MACH44) for tion, in silico replication and candidate polymorphisms selection are genotype imputation on these 398 subjects. The reference panel used for detailed in the Supplementary Methods. In brief, the Metabolomics GWAS imputing markers in autosomes is the EUR population of 1000 Genomes server30,47 was interrogated for in silico replication of the associations we Phase I version 3 (www.1000genomes.org) that includes 379 individuals detected in plasma. Candidate polymorphisms were retrieved by selecting and ~ 17 million markers. After applying the quality control standards the SNPs reported to be associated at Po5×10− 8 with any amino acid in outlined in the Supplementary Methods (o2% genotyping miss- GWASs of metabolic traits in human body fluids published to date (serum, ingness, Hardy–Weinberg Equilibrium P-value41x10 − 6 and minor allele plasma and urine).29,30,46,48,49

Table 1. Subject characteristics of the 414 genotyped subjects and of measured L- and D-enantiomers after removal of values exceeding the mean ± 3 s.d. (in μmol l − 1)

N Mean s.d. Minimum Maximum

Age (years) 414 40 11 18 60 Sex (male) 299 Storage time (months) 414 7.8 5.0 1 33 Body ﬂuid obtainment (time of day) 414 1100 hours 800 hours 1600 hours Psychiatric diagnosis 17 Use of psychotropic medication 10 L-alanine in CSF 392 31.6 7.92 16.1 72.0 D-alanine in CSF 383 0.16 0.09 0.04 0.56 L-alanine in plasma 367 334 74.0 185 561 D-alanine in plasma 365 0.85 0.47 0.09 2.71 Glycine in CSF 390 6.61 3.03 3.03 35.5 Glycine in plasma 365 165 49.4 5.66 330 L-serine in CSF 388 23.7 3.91 14.7 47.0 D-serine in CSF 393 1.26 0.23 0.78 2.21 L-serine in plasma 367 84.6 16.9 22.0 141 D-serine in plasma 369 1.09 0.26 0.42 1.78 L-proline in CSF 392 0.79 0.60 0.07 6.39 D-proline in CSF 389 0.01 0.01 0.00 0.08 L-proline in plasma 363 176 43.9 86.8 324 D-proline in plasma 362 0.32 0.18 0.01 1.18 Abbreviation: CSF, cerebrospinal ﬂuid.

RESULTS at rs2518810 weakened (P = 0.007 and 0.001, respectively). The Descriptive statistics and quality control associations at these loci in DGCR5 and PRODH, therefore, unlikely Study population and NMDAR coagonist characteristics are independently contribute to the L-proline in plasma associations. fi tabulated in Table 1. Seventeen subjects (4.1%) had a psychiatric We observed no evidence of genome-wide signi cant associations history (depression being the most common disorder) and 10 with the other metabolic traits in plasma. (2.4%) were on psychotropic medication (SSRIs in particular). On The strongest signal in CSF was also found for L-proline. SNP fi average, 4.2 outliers (range: 0–11) were removed per measured rs17279437 in the proline transporter SLC6A20 showed signi cant association with L-proline in CSF (chr 3p21.31; β = 0.28; metabolite, leaving measured values for an average of 378 − 9 subjects (range: 362–393). None of the QQ-plots was suggestive of P = 9.68 × 10 ). Two other SNPs in strong LD with rs17279437 2 fi population stratification as the genomic inflation correction (r = 0.9) showed genome-wide signi cance for L-proline in CSF λ (rs73058498 and rs73060324; chr 3p21.31; β = 0.28; P = 6.57 × factors ( GC) ranged between 0.98 and 1.05. Supplementary − 9 Figure S1 shows the QQ-plots of the associations mentioned in 10 ; Figures 1b and 2b). The same variant as mentioned above for L-proline in CSF— the text, all of which had λGC of 1.00 or 1.01. rs17279437 in the proline transporter SLC6A20—was associated − 9 with the L-proline plasma-CSF ratio (β = − 0.30; P = 2.88 × 10 ; Genome-wide significant loci detected for L-proline in plasma and Figures 1c and 2c). We added L-proline in CSF as a covariate to this CSF association test given the possibility that the association signal Genome-wide association signals in plasma were strongest for was largely explained by L-proline in CSF. Correcting for L-proline L-proline (see Figure 1a for a Manhattan plot). Five variants on in CSF indeed substantially decreased the significance of 22q11 in complete linkage disequilibrium (LD)—intronic in the association for the L-proline plasma-CSF ratio (P = 0.0006), DiGeorge syndrome critical region gene 5 (DGCR5) and 35 kb suggesting this association is mostly driven by L-proline in CSF. from proline dehydrogenase (PRODH)—showed association at the fi fi Strong evidence of in silico replication was found for the most prede ned genome-wide signi cant level (rs2518810, rs2540640, strongly associated locus in plasma (rs2540641 on 22q11): the rs2540641, rs2518814 and rs2540644; chr 22q11.2; standardized fi β − 10 strongest replication at this locus was found for the valine/proline regression coef cient ( ) = 0.29; P = 6.38 × 10 ; see Figure 2a for ratio (P = 5.98 × 10 − 69), whereas other SNPs in strong LD with a regional association plot). Loci within PRODH showed similarly rs2540641 (r240.8) were associated with other proline ratios. strong associations with L-proline in plasma. Two variants in 2 When testing candidate polymorphisms previously associated pairwise LD (r = 0.87) were strongly associated with L-proline in − with NMDAR coagonists in GWASs of other body fluids, we plasma: rs3970551 (β = 0.28; P = 2.62 × 10 9) and rs2904552 − 8 observed no evidence of additional loci involved in these traits (β = 0.27; P = 1.65 × 10 ). Both these SNPs are in LD with the − (see Supplementary Table S2 for a list of SNPs passing a Po10 3 abovementioned SNPs in DGCR5 (LD r2 of 0.60 and 0.54, association threshold). respectively). The latter is a missense variant resulting in an Arg to His change in the PRODH-coding region that corresponds to R431H, a variant with a minor allele frequency = 0.13 (ref. 50) and a CSF D-serine findings suggest involvement of D-amino-acid predicted residual enzymatic activity of around 60%.51 Given the oxidase (DAO) − high LD between the associated SNPs in DGCR5 and PRODH, we Suggestive evidence of association (Po10 7) was found conditioned on SNPs. When conditioning on rs2518810, the between two common intronic variants in complete LD located significance of the associations at rs3970551 and rs2904552 in D-amino-acid oxidase (DAO) and the D-serine plasma-CSF ratio − decreased substantially (P = 0.03 and 0.06, respectively). When (rs2070587 and rs4964766; chr 12q24.11; β = − 0.28; P =9.08×10 8; conditioning on rs3970551 and rs2904552, the association signal Figure 2d).

© 2015 Macmillan Publishers Limited Molecular Psychiatry (2015), 1557 – 1564 GWAS of NMDA receptor coagonists JJ Luykx et al 1560 Computation of explained variances We observed strong genetic evidence for loci influencing L-proline The variances explained by the genome-wide and suggestively levels. A missense variant in proline transporter SLC6A20 and significant SNPs ranged between 3.8 and 8.2% (Table 2). This variants in DGCR5 and PRODH were strongly associated with range is consistent with a previous metabolic trait GWAS30 and L-proline levels in CSF and plasma, respectively. In DAO, suggestive our power calculation (Supplementary Methods). Supplementary evidence of association was found with D-serine levels. Figure S2 provides boxplots of the concentrations per genotype. All highlighted transporter and metabolic genes (SLC6A20, PRODH and DAO) are primarily expressed in the brain and expression patterns are stable across the lifespan. Proline DISCUSSION transporter SLC6A20 is syntenic with two highly homologous To our knowledge, this work represents the first genome-wide genes in mouse and rat, Slc6a20a and Slc6a20b, the latter of which study of NMDAR coagonists and enantiomers in CSF and plasma. is predominantly expressed in the choroid plexus and meninges.52

Figure 1. (a–c) Manhattan plots of genome-wide signiﬁcant neurometabolic quantitative trait loci. (a) L-proline in Plasma. (b) L-proline in CSF. (c) The L-proline plasma-CSF ratio.

Figure 1. Continued.

These are the areas where most CSF transport from and to brain abnormality in renal glycine and proline transport.56 Furthermore, parenchyma occurs, in addition to the choroid plexus being the rs17279437 was recently ranked as the fifth most strongly region where CSF is produced. PRODH encodes PRODH that associated SNP with amino-acid metabolism in human urine.30 catalyzes the transformation of L-proline to pyrroline-5-carbo- Our findings thus portend an extensive influence of this variant on 53 xylate; the mouse orthologue is Prodh. DAO (or DAAO, D-amino- amino-acid metabolism. acid oxidase) is syntenic to Dao that encodes DAO, an enzyme that To our knowledge, the present GWAS constitutes the first on degrades D-serine. In humans, the function of DAO had remained D-amino acids in any body fluid, precluding direct comparisons largely elusive. In addition to Slc6a20b, both Dao and Prodh are with previous findings. Since their discovery in human biological highly expressed in the choroid plexus and meninges.54,55 fluids,57 there has been considerable debate regarding the origin 32 The association between missense variant rs17279437 in and biological functions of D-amino acids. Although not genome- SLC6A20 (corresponding to T199M, a variant that almost wide significant, the SNP most strongly associated with serine (the 56 − 8 completely abrogates proline transport in vitro ) and L-proline D-serine plasma-CSF ratio, intronic rs2070587, P = 9.08 × 10 )is in CSF suggests that SLC6A20 transports L-proline to the human located in DAO. The findings of the current study are highly CSF. L-proline concentrations in plasma on the other hand, were suggestive of a major role for this variant or one that is tagged by strongly associated with several SNPs within DGCR5 and PRODH at it in determining D-serine concentration gradients between CSF chromosome 22q11.2. Little is known about the function of and plasma in humans. How rs2070587 contributes to DAO DGCR5. However, this locus is in close proximity (~35 kb upstream) functionality remains elusive because this SNP does not result in of PRODH. The finding that these variants in DGCR5 tag a missense frameshifts, splice-site variation or altered DAO messenger RNA in variant in PRODH (LD r2 = 0.54) that itself was strongly associated postmortem cerebellar tissue.58 We found similarly strong − 8 with L-proline in plasma (rs2904552, P = 1.65 × 10 ) and the associations—although not suggestively significant by our pre- highlighted interdependency of these 22q11.2 associations may defined criteria—for SRR at rs3744270 on 17p13.3 with the L- − 7 imply that allelic variation at this missense variant mediates serine/D-serine ratio in CSF (β = 0.27; P = 1.10 × 10 ; Figure 2e). L-proline concentrations in plasma. Our genetic data thus point to This being the strongest associated locus for this ratio, our data transport mechanisms primarily steering L-proline concentrations lend support to the notion that in humans SRR variants control L- 59 in CSF, whereas catabolic processes seem to drive L-proline plasma to D-serine conversion. Notably, one of the top loci in the largest concentrations. A GWAS of proline in human serum detected a GWAS for SCZ to date was found for rs4523957 in SRR,28 a SNP in locus in close vicinity to the one we detected,49 which may be due weak LD with rs3744270 (r2 = 0.26); in our study, the SCZ SNP to different SNP array platforms and imputation techniques (based (rs4523957) was associated at P = 0.028 with the L-serine/D-serine on the 1000 Genomes Phase 1 Version 3 release in the current ratio in CSF and at P = 0.018 with D-serine in plasma. Furthermore, study; www.1000genomes.org). Strikingly, SNPs that are known to Srr is also highly expressed in mouse choroid plexus and impact PRODH/POX enzyme activity51 did not show evidence for meninges,55 anatomical regions relevant to CSF physiology. We association with any of the proline traits studied here (all P- found no evidence for involvement of the D-amino-acid oxidase values40.1 for both rs4819756 (R185W) and rs2008720 (Q19P)). activator (DAOA) gene (also known as G72)inD-serine levels (no By demonstrating that the missense variant rs17279437 in association results at Po10 − 4 for this gene ± 20 kb with any of SLC6A20 mediates L-proline concentration variability in the CSF of the serine traits). healthy subjects we extend previous work demonstrating a role Several limitations must be borne in mind, however. The size of for this variant in iminoglycinuria, an autosomal recessive our study population was modest compared with metabolic trait

abL−Proline in Plasma L−Proline in CSF Plotted SNPs Plotted SNPs

10 100 10 100

8 80 Recombination rate (cM/Mb) 8 80 Recombination rate

60 6 6 60 (p−value) (p−value) 10 4 40 10 log (

− 4 40 log cM/Mb −

2 20 ) 2 20

0 0 0 0

18.6 18.8 19 19.2 45.4 45.6 45.8 46 Position on chr22 (Mb) Position on chr3 (Mb)

c L−Proline Plasma−CSF ratio d D−Serine Plasma−CSF ratio Plotted SNPs Plotted SNPs

10 100 10 100

8 80 8 80 R R ecombination rate (cM ecombination rate (cM

6 60 6 60 (p−value) (p−value) 10 10 4 40 log − log 4 40 − / Mb) / Mb) 2 20 2 20

0 0 0 0

45.6 45.8 46 46.2 109 109.2 109.4 109.6 Position on chr3 (Mb) Position on chr12 (Mb)

e L−D Serine ratio in CSF Plotted SNPs

10 100 R

8 80 ecombination rate (cM/Mb)

6 60 (p−value) 10 4 40 log −

2 20

2 2.2 2.4 2.6 Position on chr17 (Mb)

Molecular Psychiatry (2015), 1557 – 1564 © 2015 Macmillan Publishers Limited GWAS of NMDA receptor coagonists JJ Luykx et al 1563 Figure 2. (a–e) Regional association plots of neurometabolic quantitative trait loci. (a) Genome-wide significant L-proline (plasma) association in DGCR5/ 35 kb from PRODH.(b) Genome-wide significant L-proline (CSF) association in SCL6A20.(c) Genome-wide significant L-proline (plasma-CSF ratio) association in SCL6A20. (d) Suggestive D-serine (plasma-CSF ratio) association in DAO.(e) Non-significant L-serine/D-serine ratio (CSF) association in SRR.

Table 2. The top genome-wide signiﬁcant (Po1.25 × 10 −8, in bold) and suggestively signiﬁcant (Po10 − 7) associations are shown per quantitative trait, ordered by increasing P-values

Phenotype SNP Position Locus MAF β P-value R2 (%)

L-proline (P) rs2540641 22q11.2 DGCR5/35 kb from PRODH 0.11 0.30 6.38E − 10a 8.2 L-proline (P/CSF) rs17279437 3p21.31 SLC6A20 0.09 − 0.30 2.88E − 09 3.8 L-proline (CSF) rs73058498 3p21.31 SACM1L / 22 kb from SLC6A20 0.09 0.28 6.57E − 09b 7.8 L-alanine (P/CSF) rs4377332 2q36.1 198 kb from EPHA4 0.36 − 0.28 2.91E − 08 7.5 D-alanine (P/CSF) rs352052 4q13.3 Inter-genic between CXCL3 and CXCL5 0.48 − 0.30 3.38E − 08 6.5 Glycine (CSF) rs115153038 14q32.12 4.2 kb from TC2N 0.05 0.27 3.56E − 08 8.1 D-serine (P/CSF) rs2070587 12q24.11 DAO 0.19 − 0.28 9.08E − 08 7.7 Abbreviations: CSF, cerebrospinal ﬂuid; β, standardized beta; LD, linkage disequilibrium; MAF, minor allele frequency; P, Plasma; PRODH, proline dehydrogenase R2, explained variance; SNP, single-nucleotide polymorphism. Only one SNP per locus is shown. aThis SNP is in LD (r2 = 0.54) with missense variant rs2904552 in − 8 2 PRODH that was also associated with L-proline in plasma (P = 1.65 × 10 ); it is also in LD with rs2023634 (r = 0.71) that is associated with the proline/valine ratio at P = 1.21 × 10 −25 in an independent data set of human serum (N = 1768). bThis SNP is in strong LD (r2 = 0.9) with rs17279437 in SLC6A20.

GWASs in other bodily fluids.29,30,47,49 In addition, owing to the CONFLICT OF INTEREST uniqueness of our sample and the enterprising nature of CSF The authors declare no conflict of interest. collection from the general population, a similarly sized replication cohort with CSF measurements was unavailable. Furthermore, due ACKNOWLEDGMENTS to high correlations between ratios and enantiomers (Supple- mentary Table S1) we corrected for the number of NMDAR We thank drs. JTA Knape, P Vaessen and P Keijzers for their valuable assistance in coagonists instead of all measured values and ratios in the two developing our CSF sampling protocol in the operating rooms and conducting the fl fi CSF samplings. We thank K van Eijk and J van Setten for their assistance in body uids. Although the plasma L-proline association nding developing genetic quality control standards. would survive Bonferroni correction for all these traits (α =5× 10 − 8/27 = 1.85 × 10 − 9), future non-targeted CSF metabolomics projects that include a wider range of neurometabolic traits will REFERENCES require larger sample sizes and replication cohorts. 1 Henneberger C, Papouin T, Oliet SH, Rusakov DA. Long-term potentiation The presented data constitute a proof of principle that CSF is a depends on release of D-serine from astrocytes. Nature 2010; 463:232–236. meaningful and informative body fluid for QTL analyses. Our 2 Papouin T, Ladepeche L, Ruel J, Sacchi S, Labasque M, Hanini M et al. Synaptic and findings may have particular implications for 22q11 deletion extrasynaptic NMDA receptors are gated by different endogenous coagonists. Cell 2012; 150: 633–646. syndrome and hyperprolinemia, both of which may entail 3 Nong Y, Huang YQ, Ju W, Kalia LV, Ahmadian G, Wang YT et al. Glycine binding 10,60,61 cognitive deficits and psychosis. On the basis of the current primes NMDA receptor internalization. Nature 2003; 422:302–307. data, one could speculate that pharmacologically blocking 4 Sakata K, Fukushima T, Minje L, Ogurusu T, Taira H, Mishina M et al. Modulation by SLC6A20 in patients suffering from these disorders may reduce L- and D-isoforms of amino acids of the L-glutamate response of N-methyl-D- aspartate receptors. Biochemistry 1999; 38: 10099–10106. the detrimental effects of excess CNS L-proline. In addition, 5 McBain CJ, Kleckner NW, Wyrick S, Dingledine R. Structural requirements for chromosome 22q11 deletion syndrome is the strongest genetic 62 activation of the glycine coagonist site of N-methyl-D-aspartate receptors risk factor for psychosis and L-proline was recently found to be expressed in Xenopus oocytes. Mol Pharmacol 1989; 36:556–565. 63 the prime metabolite altered in SCZ. Thus, genetic studies of the 6 Martin D, Ault B, Nadler JV. NMDA receptor-mediated depolarizing action of highlighted 22q11.2 variants in SCZ patients may elucidate proline on CA1 pyramidal cells. Eur J Pharmacol 1992; 219:59–66. trajectories pertinent to L-proline metabolism in psychosis. Finally, 7 Henzi V, Reichling DB, Helm SW, MacDermott AB. L-proline activates glutamate recent research has indicated how the localization of NMDAR and glycine receptors in cultured rat dorsal horn neurons. Mol Pharmacol 1992; 41: 793–801. coagonist gating may be deduced from modulation of enzymes 8 Kleckner NW, Dingledine R.. Requirement for glycine in activation of NMDA- 2 and transporters relevant to NMDAR coagonists. The genetic receptors expressed in Xenopus oocytes. Science 1988; 241:835–837. variants, enzymes and transporters highlighted here may accord- 9 Clelland CL, Read LL, Baraldi AN, Bart CP, Pappas CA, Panek LJ et al. Evidence for ingly be targeted to further construe the nature of NMDAR association of hyperprolinemia with schizophrenia and a measure of clinical coagonist involvement in synaptic and extrasynaptic glutamate outcome. Schizophr Res 2011; 131:139–145. signaling. 10 Jacquet H, Demily C, Houy E, Hecketsweiler B, Bou J, Raux G et al. Hyperprolinemia is a risk factor for schizoaffective disorder. Mol Psychiatry 2005; 10: 479–485. In summary, loci in genes encoding both a transporter protein 11 Hashimoto K, Fukushima T, Shimizu E, Komatsu N, Watanabe H, Shinoda N et al. (SLC6A20) and an enzyme (PRODH) were identified. Furthermore, Decreased serum levels of D-serine in patients with schizophrenia: evidence in we provide preliminary evidence that in humans a variant in DAO support of the N-methyl-D-aspartate receptor hypofunction hypothesis of schi- 60 – mainly determines D-serine levels in the CNS, whereas variants in zophrenia. Arch Gen Psychiatry 2003; : 572 576. 12 Bendikov I, Nadri C, Amar S, Panizzutti R, De Miranda J, Wolosker H et al. ACSF SRR affect L-toD-serine conversion in CSF. As biologically tenable and postmortem brain study of D-serine metabolic parameters in schizophrenia. associations in both plasma and CSF were detected, we conclude 90 – fl Schizophr Res 2007; :41 51. that quantitative traits in both body uids constitute informative 13 Frye MA, Tsai GE, Huggins T, Coyle JT, Post RM. Low cerebrospinal fluid glutamate targets for preclinical and human follow-up studies. and glycine in refractory affective disorder. Biol Psychiatry 2007; 61:162–166.

© 2015 Macmillan Publishers Limited Molecular Psychiatry (2015), 1557 – 1564 GWAS of NMDA receptor coagonists JJ Luykx et al 1564 14 Luykx JJ, Bakker SC, van Boxmeer L, Vinkers CH, Smeenk HE, Visser WF et al. 40 Visser WF, Verhoeven-Duif NM, Ophoff R, Bakker S, Klomp LW, Berger R et al. D-amino acid aberrations in cerebrospinal fluid and plasma of smokers. Neuro- A sensitive and simple ultra-high-performance-liquid chromatography-tandem psychopharmacology 2013; 38:2019–2026. mass spectrometry based method for the quantification of d-amino acids in 15 Brouwer A, Luykx JJ, van Boxmeer L, Bakker SC, Kahn RS. NMDA-receptor coa- body fluids. J Chromatogr A 1218: 7130–7136. gonists in serum, plasma, and cerebrospinal fluid of schizophrenia patients: a 41 Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D et al. PLINK: a meta-analysis of case-control studies. Neurosci Biobehav Rev 2013; 37: 1587–1596. tool set for whole-genome association and population-based linkage analyses. 16 Goff DC, Tsai G, Levitt J, Amico E, Manoach D, Schoenfeld DA et al. A placebo- Am J Hum Genet 2007; 81:559–575. controlled trial of D-cycloserine added to conventional neuroleptics in patients 42 Browning SR, Browning BL. Rapid and accurate haplotype phasing and missing- with schizophrenia. Arch Gen psychiatry 1999; 56:21–27. data inference for whole-genome association studies by use of localized haplo- 17 Heresco-Levy U, Ermilov M, Lichtenberg P, Bar G, Javitt DC. High-dose glycine type clustering. Am J Hum Genet 2007; 81: 1084–1097. added to olanzapine and risperidone for the treatment of schizophrenia. Biol 43 Howie B, Fuchsberger C, Stephens M, Marchini J, Abecasis GR. Fast and accurate Psychiatry 2004; 55:165–171. genotype imputation in genome-wide association studies through pre-phasing. 18 Heresco-Levy U, Javitt DC, Ermilov M, Mordel C, Horowitz A, Kelly D. Double-blind, Nature Genet 2012; 44:955–959. placebo-controlled, crossover trial of glycine adjuvant therapy for treatment- 44 Li Y, Willer CJ, Ding J, Scheet P, Abecasis GR. MaCH: using sequence and genotype resistant schizophrenia. Br J Psychiatry 1996; 169:610–617. data to estimate haplotypes and unobserved genotypes. Genet Epidemiol 2010; fi 19 Heresco-Levy U, Javitt DC, Ermilov M, Mordel C, Silipo G, Lichtenstein M. Ef cacy 34:816–834. of high-dose glycine in the treatment of enduring negative symptoms of schi- 45 Petersen AK, Krumsiek J, Wagele B, Theis FJ, Wichmann HE, Gieger C et al. On the 56 – zophrenia. Arch Gen Psychiatry 1999; :29 36. hypothesis-free testing of metabolite ratios in genome-wide and metabolome- 20 Tsai G, Yang P, Chung LC, Lange N, Coyle JT. D-serine added to antipsychotics for wide association studies. BMC Bioinformatics 2012; 13: 120. 44 – the treatment of schizophrenia. Biol Psychiatry 1998; : 1081 1089. 46 Gieger C, Geistlinger L, Altmaier E, Hrabe de Angelis M, Kronenberg F, Meitinger T 21 Tsai GE, Yang P, Chang YC, Chong MY. D-alanine added to antipsychotics for the et al. Genetics meets metabolomics: a genome-wide association study of meta- 59 – treatment of schizophrenia. Biol Psychiatry 2006; :230 234. bolite profiles in human serum. PLoS Genet 2008; 4: e1000282. 22 Labrie V, Wong AH, Roder JC. Contributions of the D-serine pathway to schizo- 47 Shin SY, Fauman EB, Petersen AK, Krumsiek J, Santos R, Huang J et al. An atlas 62 – phrenia. Neuropharmacology 2012; : 1484 1503. of genetic influences on human blood metabolites. Nature Genet 2014; 46: 23 Hofmann SG, Meuret AE, Smits JA, Simon NM, Pollack MH, Eisenmenger K et al. 543–550. Augmentation of exposure therapy with D-cycloserine for social anxiety disorder. 48 Nicholson G, Rantalainen M, Li JV, Maher AD, Malmodin D, Ahmadi KR et al. 63 – Arch Gen Psychiatry 2006; :298 304. A genome-wide metabolic QTL analysis in Europeans implicates two loci shaped 24 Guastella AJ, Richardson R, Lovibond PF, Rapee RM, Gaston JE, Mitchell P et al. by recent positive selection. PLoS Genet 2011; 7: e1002270. A randomized controlled trial of D-cycloserine enhancement of exposure therapy 49 Suhre K, Shin SY, Petersen AK, Mohney RP, Meredith D, Wagele B et al. Human for social anxiety disorder. Biol Psychiatry 2008; 63:544–549. metabolic individuality in biomedical and pharmaceutical research. Nature 2011; 25 Otto MW, Tolin DF, Simon NM, Pearlson GD, Basden S, Meunier SA et al. Efficacy of 477:54–60. d-cycloserine for enhancing response to cognitive-behavior therapy for panic 50 Guilmatre A, Legallic S, Steel G, Willis A, Di Rosa G, Goldenberg A et al. Type I disorder. Biol Psychiatry 2010; 67: 365–370. hyperprolinemia: genotype/phenotype correlations. Hum Mutat 2010; 31: 26 Wilhelm S, Buhlmann U, Tolin DF, Meunier SA, Pearlson GD, Reese HE et al. 961–965. Augmentation of behavior therapy with D-cycloserine for obsessive-compulsive 51 Bender HU, Almashanu S, Steel G, Hu CA, Lin WW, Willis A et al. Func- disorder. Am J Psychiatry 2008; 165:335–341. tional consequences of PRODH missense mutations. Am J Hum Genet 2005; 76: 27 Kushner MG, Kim SW, Donahue C, Thuras P, Adson D, Kotlyar M et al. 409–420. D-cycloserine augmented exposure therapy for obsessive-compulsive disorder. 52 Diez-Roux G, Banfi S, Sultan M, Geffers L, Anand S, Rozado D et al. A high- Biol Psychiatry 2007; 62: 835–838. resolution anatomical atlas of the transcriptome in the mouse embryo. PLoS Biol 28 Schizophrenia Working Group of the Psychiatric Genomics Consortium. Biological 2011; 9: e1000582. insights from 108 schizophrenia-associated genetic loci. Nature 2014; 511: 53 Tanner JJ. Structural biology of proline catabolism. Amino Acids 2008; 35: 421–427. 719–730. 29 Kettunen J, Tukiainen T, Sarin AP, Ortega-Alonso A, Tikkanen E, Lyytikainen LP 54 Magdaleno S, Jensen P, Brumwell CL, Seal A, Lehman K, Asbury A et al. BGEM: an et al. Genome-wide association study identifies multiple loci influencing human in situ hybridization database of gene expression in the embryonic and adult serum metabolite levels. Nature Genet 2012; 44:269–276. 4 30 Suhre K, Wallaschofski H, Raffler J, Friedrich N, Haring R, Michael K et al. mouse nervous system. PLoS Biol 2006; :e86. A genome-wide association study of metabolic traits in human urine. Nature 55 Visel A, Thaller C, Eichele G. GenePaint.org: an atlas of gene expression patterns in 32 – Genet 2011; 43: 565–569. the mouse embryo. Nucleic Acids Res 2004; : : D552 D556. 31 Lehtinen MK, Zappaterra MW, Chen X, Yang YJ, Hill AD, Lun M et al. The cere- 56 Broer S, Bailey CG, Kowalczuk S, Ng C, Vanslambrouck JM, Rodgers H et al. brospinal fluid provides a proliferative niche for neural progenitor cells. Neuron Iminoglycinuria and hyperglycinuria are discrete human phenotypes resulting 2011; 69: 893–905. from complex mutations in proline and glycine transporters. J Clin Invest 2008; 118 – 32 Fuchs SA, Berger R, de Koning TJ. D-serine: the right or wrong isoform? Brain Res :3881 3892. 2011; 1401: 104–117. 57 Dunlop DS, Neidle A, McHale D, Dunlop DM, Lajtha A. The presence of free 141 33 Ma TM, Abazyan S, Abazyan B, Nomura J, Yang C, Seshadri S et al. Pathogenic D-aspartic acid in rodents and man. Biochem Biophys Res Commun 1986; : – disruption of DISC1-serine racemase binding elicits schizophrenia-like behavior 27 32. via D-serine depletion. Mol Psychiatry 2012; 18:557–567. 58 Burnet PW, Eastwood SL, Bristow GC, Godlewska BR, Sikka P, Walker M et al. D- 34 Wolosker H, Blackshaw S, Snyder SH. Serine racemase: a glial enzyme synthesizing amino acid oxidase activity and expression are increased in schizophrenia. Mol D-serine to regulate glutamate-N-methyl-D-aspartate neurotransmission. Proc Psychiatry 2008; 13: 658–660. Natl Acad Sci USA 1999; 96: 13409–13414. 59 De Miranda J, Santoro A, Engelender S, Wolosker H. Human serine racemase: 35 Wolosker H, Sheth KN, Takahashi M, Mothet JP, Brady RO Jr, Ferris CD et al. moleular cloning, genomic organization and functional analysis. Gene 2000; 256: Purification of serine racemase: biosynthesis of the neuromodulator D-serine. Proc 183–188. Natl Acad Sci USA 1999; 96: 721–725. 60 Raux G, Bumsel E, Hecketsweiler B, van Amelsvoort T, Zinkstok J, Manouvrier- 36 Boks MP, Rietkerk T, van de Beek MH, Sommer IE, de Koning TJ, Kahn RS. Hanu S et al. Involvement of hyperprolinemia in cognitive and psychiatric features Reviewing the role of the genes G72 and DAAO in glutamate neurotransmission of the 22q11 deletion syndrome. Hum Mol Genet 2007; 16:83–91. in schizophrenia. Eur Neuropsychopharmacol 2007; 17:567–572. 61 Vorstman JA, Morcus ME, Duijff SN, Klaassen PW, Heineman-de Boer JA, Beemer 37 Wang TJ, Larson MG, Vasan RS, Cheng S, Rhee EP, McCabe E et al. Metabolite FA et al. The 22q11.2 deletion in children: high rate of autistic disorders and early profiles and the risk of developing diabetes. Nat Med 2011; 17:448–453. onset of psychotic symptoms. J Am Acad Child Adolesc Psychiatry 2006; 45: 38 Luykx JJ, Vinkers CH, Bakker SC, Visser WF, van Boxmeer L, Strengman E et al. 1104–1113. A common variant in ERBB4 regulates GABA concentrations in human 62 Sullivan PF, Daly MJ, O'Donovan M. Genetic architectures of psychiatric disorders: cerebrospinal fluid. Neuropsychopharmacology 2012; 37: 2088–2092. the emerging picture and its implications. Nat Rev Genet 2012; 13:537–551. 39 Luykx JJ, Bakker SC, Lentjes E, Neeleman M, Strengman E, Mentink L et al. 63 Oresic M, Tang J, Seppanen-Laakso T, Mattila I, Saarni SE, Saarni SI et al. Meta- Genome-wide association study of monoamine metabolite levels in human bolome in schizophrenia and other psychotic disorders: a general population- cerebrospinal fluid. Mol Psychiatry 2014; 19:228–234. based study. Genome Med 2011; 3:19.

Supplementary Information accompanies the paper on the Molecular Psychiatry website (http://www.nature.com/mp)