SEARCH

SEARCH BY CITATION

Keywords:

  • Cognitive deficit;
  • endophenotypes;
  • glutamatergic signalling;
  • GRIN2B;
  • GRM3;
  • PRKCA;
  • RAVLT

Abstract

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Acknowledgments
  8. Supporting Information

Neurocognitive dysfunction is a core feature of schizophrenia with particularly prominent deficits in verbal episodic memory. The molecular basis of this memory impairment is poorly understood and its relatedness to normal variation in memory performance is unclear. In this study, we explore, in a sample of cognitively impaired schizophrenia patients, the role of polymorphisms in seven genes recently reported to modulate episodic memory in normal subjects. Three polymorphisms (GRIN2B rs220599, GRM3 rs2189814 and PRKCA rs8074995) were associated with episodic verbal memory in both control and patients with cognitive deficit, but not in cognitively spared patients or the pooled schizophrenia sample. GRM3 and PRKCA acted in opposite directions in patients compared to controls, possibly reflecting an abnormal brain milieu and/or adverse environmental effects in schizophrenia. The encoded proteins balance glutamate signalling vs. excitotoxicity in complex interactions involving the excitatory amino acid transporter 2 (EAAT2), implicated in the dysfunctional glutamatergic signalling in schizophrenia. Double carrier status of the GRM3 and PRKCA minor alleles was associated with lower memory test scores and with increased risk of schizophrenia. Single nucleotide polymorphism (SNP) rs8074995 lies within the PRKCA region spanned by a rare haplotype associated with schizophrenia in a recent UK study and provides further evidence of PRKCA contribution to memory impairment and susceptibility to schizophrenia. Our study supports the utility of parsing the broad phenotype of schizophrenia into component cognitive endophenotypes that reduce heterogeneity and enable the capture of potentially important genetic associations.

‘Memory proper … is the knowledge of a former state of mind after it has already dropped from consciousness'. The pithy definition of episodic memory by William James (1890), a founding father of modern psychology, has been followed by decades of research highlighting the cognitive architecture of memory (Jonides et al. 2008; Tulving 2002) and exploring the underlying neurobiology (Miyashita 2004), which includes strikingly conserved features across species, all the way from Aplysia to complex distributed mnemonic networks in humans and other primates. Episodic memory formation occurs via a cascade of signalling events, involving encoding, storage and retrieval processes. These processes have been extensively studied in animals, but little is known about their genetic underpinnings and the molecular mechanisms of human memory in health and disease.

Based on animal studies, de Quervain and Papassotiropoulos (2006) selected a set of 47 genes, encoding receptors and signalling molecules, to investigate their effect on normal variation in human episodic memory and activation of memory-related brain regions. The analysis of 160 single nucleotide polymorphisms (SNPs) in 336 young healthy adults found a cluster of 9 SNPs in seven genes that showed highly significant association with enhanced episodic memory and with functional magnetic resonance imaging (fMRI)-detectable activation of medial temporal lobe structures during memory tasks. These genes are well known to be related to the signalling cascade involved in hippocampus-dependent memory: GRIN2A, GRIN2B and GRM3 encode glutamate receptors, and ADCY8, PRKACG, CAMK2G and PRKCA encode major enzymes from the signalling pathways (de Quervain & Papassotiropoulos 2006; Kandel 2001; Shobe 2002). Neurocognitive dysfunction is a core feature of schizophrenia (Cirillo & Seidman 2003; Gur et al. 2007; Heinrichs & Zakzanis 1998; Mesholam-Gately et al. 2009), with particularly prominent deficits in verbal episodic memory, primarily due to impaired encoding, rather than forgetting, implicating prefrontal and medial temporal lobe structures (Skelley et al. 2008). The impairment is present in first-episode cases, remains relatively stable over the course of the illness and is not significantly influenced by current antipsychotic therapeutic agents (Gur et al. 2007). Similar, although milder, impairments have been observed among unaffected first-degree relatives of patients (Whyte et al. 2005). Notably, both the extent and severity of memory impairment vary considerably among patients with schizophrenia, from mild or moderate to severe dysfunction comparable to memory disorders seen in neurological diseases (Heinrichs & Zakzanis 1998).

The molecular basis of memory impairment in schizophrenia is poorly understood (Aukes et al. 2008) and its relatedness to normal interindividual variation in memory performance (Lee 2003) is unclear. Furthermore, the heterogeneity of memory performance in schizophrenia may significantly reduce the power to detect genetic influences on impaired encoding and recall. In a previous study (Hallmayer et al. 2005), we identified two distinct, only partially overlapping, clusters of schizophrenia cases, based on their neurocognitive profiles: one characterized by pervasive cognitive deficit (CD), with a consistent, robust memory dysfunction contributing the largest effect size, and another, comprising patients with relatively spared cognition (CS), with mild focal deficits, including memory performance within <1 SD of healthy control subjects. Here, we explore the effects of the polymorphisms reported to modulate normal memory performance (de Quervain & Papassotiropoulos 2006) on episodic memory in cognitively impaired schizophrenia patients.

Methods

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Acknowledgments
  8. Supporting Information

Subjects

The overall sample included a total of 508 individuals (336 schizophrenia patients and 172 normal controls) all of European descent with over 75% of Anglo-Irish ancestry. Patients with schizophrenia were recruited from consecutive admissions to a psychiatric hospital or community mental health centres within the same area. The patient sample is a fairly representative cross-section of service users with severe mental disorder (80% male, mean age 33.9 years, range 17–60 years; mean length of illness 9.8 years). The control subjects (59% male, mean age 40.7 years, range 17–76 years) were recruited by random sampling from local telephone directories, or among Red Cross blood donors, and screened for psychopathology to exclude those with a personal or family history of psychotic illness. Written informed consent was obtained from all participating subjects. The study was approved by the Human Research Ethics Committee of the University of Western Australia and the North Metropolitan Health Area Ethics Committee, Perth, Western Australia.

Diagnostic assessment was based on standardized interviews employing the Schedules for Clinical Assessment in Neuropsychiatry (SCAN; Wing et al. 1990) and scored using the Operational Criteria for Research (OPCRIT) algorithm (Castle et al. 2006). Video-recorded interviews and clinical charts were independently reviewed by two senior clinicians who assigned consensus research diagnoses in terms of International Classification of Diseases, 10th Revision (ICD-10) and Diagnostic and Statistical Manual, 4th Revision (DSM-IV). Patients and controls were administered a battery of tests assessing neurocognitive performance, as described previously (Hallmayer et al. 2005): general cognitive ability (prior and current IQ), verbal memory, sustained attention, working memory and executive function, speed of information processing. All patients were examined during periods of relative clinical stabilization, and confounding of the results by acute psychotic turmoil is unlikely. Verbal memory was tested using the Rey Auditory Verbal Learning Test (RAVLT), a robust reliable task (Lezak et al. 2004; Schoenberg et al. 2006; Spreen & Strauss 1998). Our study involved immediate free recall (RAVLTi) of three consecutive trial presentations of a 15-word list, taxing encoding and short-term recall. In addition, a 20-min delayed recall trial (RAVLTd) was presented, following distraction to prevent active rehearsal, to assess long-term retention and retrieval. Performance data from the multiple domains targeted by the neurocognitive battery were integrated into a composite continuous trait, using a latent structure (grade of membership, GoM) model (Manton et al. 1994; Woodbury et al. 1978). Grade of membership computes simultaneously a number of latent ‘pure types' (PT) and quantifies each individual's degree of similarity to any one of the PTs. Two PTs accounted for >90% of the combined test scores in schizophrenia patients and were accordingly labelled pervasive CD, with memory impairment as a prominent feature; and CS, with mild or patchy deficits in individual domains (Hallmayer et al. 2005). Based on a statistically defined cut-off in the cognitive trait, patients were assigned to a CD cluster (n = 155) or a CS cluster (n = 121). Sixty patients could not be definitively assigned to either of the two clusters.

Genotyping

We genotyped the nine SNPs, reported by de Quervain and Papassotiropoulos (2006) (Table 1) in the 155 cognitively deficient cases and, as a comparison with patients and to the original data, also in the group of controls. To check whether the variants rs220599 (GRIN2B), rs2189814 (GRM3) and rs8074995 (PRKCA), associated with memory performance in the CD cluster, also had an effect in the total schizophrenia and CS samples, these three SNPs were also analysed in the remaining cases. Genotyping was performed by Amplifluor technology (Serologicals Corporation, Atlanta, GA, USA) for rs263249, rs3730386, rs1868291 and rs2189814, and Taqman assays (Applied Biosystems, Foster City, CA, USA) for rs11000787, rs12828473, rs220599, rs6465084 and rs8074995. Centre d’Etude du Polymorphisme Humain (CEPH) trio 1334 samples (Coriell Cell Repository) served as internal controls. Consistency with Hardy–Weinberg equilibrium at P < 0.001 (Wigginton et al. 2005) was examined in the control dataset.

Table 1.  Single nucleotide polymorphisms included in the analysis
GeneChromosomeSNP rs#Sequence variation*MAF (controls)P_HWE (controls)
  1. ADCY8, adenylate cyclase 8; CAMK2G, calcium/calmodulin-dependent protein kinase II gamma; GRIN2A, glutamate receptor, ionotropic, N-methyl-d-aspartate, subunit 2A; GRIN2B, glutamate receptor, ionotropic, N-methyl-d-aspartate, subunit 2B; GRM3, glutamate receptor, metabotropic, 3; HWE, Hardy–Weinberg equilibrium; PRKACG, protein kinase, cAMP-dependent, catalytic, gamma; PRKCA, protein kinase C, alpha.

  2. *Major/minor allele based on coding sequence of gene.

ADCY8 8rs263249C/T0.3261.000
PRKACG 9rs3730386C/G0.3120.722
CAMK2G 10rs11000787G/A0.2680.700
GRIN2B 12rs12828473C/T0.3180.384
  rs220599C/T0.4560.877
GRIN2A 16rs1868291A/G0.0740.604
GRM3 7rs2189814T/C0.3290.729
  rs6465084A/G0.2410.400
PRKCA 17rs8074995G/A0.1320.016

To examine the correspondence between our findings and PRKCA polymorphisms recently reported in association with schizophrenia (Carroll et al. 2010), SNP rs62621676 was genotyped using a polymerase chain reaction (PCR)-based restriction fragment length polymorphism (RFLP) assay (restriction enzyme CviKI-1) in 54 control samples. Details are provided online under Supporting information, Methods.

Statistical analyses

Each SNP was analysed for genotypic association with raw RAVLTi (average of the three trials) and RAVLTd scores. Those with a nominally significant result (P≤ 0.05) were explored further under different genetic models. The effects of age and sex were tested in an initial round of analyses, where they were used as covariates, and subsequently removed if no effect was identified. For direct comparison with the previously published data (de Quervain & Papassotiropoulos 2006), two-marker haplotype analyses were conducted for the GRIN2B and GRM3 SNPs. All analyses were performed with SimHap version 1.0.2 (Carter et al. 2008). Power to detect association (Dupont & Plummer 1990) was estimated at >80% for a difference of 0.5 trait standard deviations in genotypic means (between those carrying the allele and those without) and minor allele frequency (MAF) 0.2–0.5. Linkage disequilibrium (LD) information was analysed using HapMap Phase II and III data and our own control genotypes, with statistics and graphics obtained using HAPLOVIEW 4.1 (Barrett et al. 2005).

Results

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Acknowledgments
  8. Supporting Information

The observed MAFs were comparable to those reported for Caucasians in dbSNP (http://www.ncbi.nlm.nih.gov/projects/SNP/) (Table 1). No polymorphism showed deviation from Hardy–Weinberg equilibrium in the controls.

Age and sex were found in the original (de Quervain & Papassotiropoulos 2006) and in this study to have an effect on memory performance in controls (RAVLTi: age P = 0.0001, sex P = 0.04; RAVLTd: age P = 0.0001, sex P = 0.02), and were used as covariates in the analyses of this group. The lack of effect in the schizophrenia patients (RAVLTi: age P = 0.61, sex P = 0.89; RAVLTd: age P = 0.90, sex P = 0.35) reflects the fact that these potential confounds are taken into account in the GoM analysis (Hallmayer et al. 2005).

The full set of results is presented in Tables S1 and S2. Significant results are shown in Table 2 with regression (β)-coefficients, standard errors and P-values.

Table 2.  Variants showing significant evidence of association (P < 0.05) with memory performance in schizophrenia patients with cognitive deficit and in controls
GeneSNP/haplotypeRAVLTi: β-coefficient (SE), P-valueRAVLTd: β-coefficient (SE), P-valueEnhancing effect in discovery sample
Schizophrenia CD clusterControlsSchizophrenia CD clusterControls
  1. N.S., non-significant.

  2. The best fitting model is additive, unless stated otherwise:

  3. *dominant;

  4. **recessive.

  5. Double carriers of the minor alleles of the two SNPs.

PRKACG rs3730386 C/GN.S.N.S.−0.83 (0.35), P = 0.02*N.S.G
GRIN2B rs12828473 C/TN.S.N.S.N.S.N.S.Lack of TT
 rs220599 C/T−1.13 (0.48), P = 0.02N.S.−0.71 (0.26), P = 0.008N.S. 
 rs12828473-rs220599T-T: −2.30, P = 0.071T-C: −3.59, P = 0.069**T-T: −1.14, P = 0.06**T-C: −3.13, P = 0.005**Lack of TT
GRM3 rs2189814 T/C−1.29 (0.48), P = 0.0074.89 (1.2), P = 6e-05**−0.54 (0.27), P = 0.042.82 (0.69), P = 5e-05**Lack of TG
 rs6465084 A/GN.S.N.S.N.S.N.S. 
 rs2189814C-rs6465084A−1.22, P = 0.0124.89, P = 7.12e-05**−0.51, P = 0.0592.82, P = 6.41e-05** 
PRKCA rs8074995 G/A−1.43 (0.61), P = 0.021.25 (0.69), P = 0.08N.S.N.S.A
GRM3 + PRKCA rs2189814C-rs8074995A−2.33 (0.92), P = 0.013.52 (1.49), P = 0.02N.S.1.76 (0.82), P = 0.03T

Four genes showed significant evidence of association with memory performance in patients (Tables 2 and S1). Three of these (encoding glutamate receptors and a protein kinase) were also associated with memory in our control sample (Table 2), thus replicating the original study (de Quervain & Papassotiropoulos 2006). The Results and Discussion sections below focus on these three genes.

GRIN2B

GRIN2B encodes the NR2B subunit of the NMDA receptor (NMDAR). Significant results were obtained for rs220599 in intron 2, associated with lower RAVLTi and RAVLTd scores in CD patients (Table 2). In the original study, GRIN2B rs12828473 and rs220599 contributed to memory performance as a haplotype rather than individually, with better performance observed in the absence of the T-T haplotype (de Quervain & Papassotiropoulos 2006). Our haplotype analysis supports this finding: we observed a trend or significant association with lower RAVLT scores (immediate as well as delayed recall) in both affected subjects and controls for haplotypes comprising at least one of the minor (T) alleles: T-T in patients and T-C in controls (Table 2).

GRM3

In the GRM3 gene, encoding metabotropic glutamate receptor 3 (mGluR3), evidence of association with memory performance was obtained for rs2189814 (intron 1). It had a highly significant effect on immediate and delayed recall in controls, with greatly enhanced performance in homozygotes for the C allele. The same SNP also showed significant association in CD patients, with a reversal of the direction of effect (Table 2). In the original study (de Quervain & Papassotiropoulos 2006), association with memory was identified for rs2189814–rs6465084 haplotypes, with enhanced performance observed in the absence of the T-G haplotype. Our analyses showed an association of the C-A haplotype with higher RAVLTi and RAVLTd scores in homozygous control subjects, supporting the previous findings (de Quervain & Papassotiropoulos 2006). In patients, the same haplotype had the opposite effect, significantly associated with lower RAVLTi scores and showing a trend for RAVLTd. The observed β-coefficients and P-values suggest that the haplotype association was entirely driven by the rs2189814 C allele.

PRKCA

Single nucleotide polymorphism rs8074995 in intron 16 of PRKCA, the gene encoding the α isoform of protein kinase C (PKC), showed a trend for association with RAVLTi scores in controls (Table 2), in agreement with de Quervain and Papassotiropoulos (2006). In affected subjects, the association with RAVLTi was significant, again with the direction of effect opposite to healthy controls.

The associated polymorphism, rs8074995, is located within the region covered by a rare haplotype, referred to by the authors as C-HAP (Fig. 1) recently reported to segregate with mental disorders in a multiplex family and to show association, as a haplotype and as individual constituent SNPs, in a large case–control sample from the United Kingdom (Carroll et al. 2010). As there are no HapMap data on the C-HAP SNPs, we attempted to analyse their correspondence to rs8074995 by genotyping rs62621676 in 54 control subjects. In addition to being part of the C-HAP haplotype, rs62621676 [in the 3′-untranslated region (3′-UTR) of the gene and ∼8-kb downstream of rs8074995] had an individual main effect on increased risk of schizophrenia and psychosis in the UK study (Carroll et al. 2010). We identified a single heterozygote among the 54 controls, MAF 0.009, close to that found by Carroll et al. (2010).

image

Figure 1. Diagram showing the approximate location of SNPs in PRKCA associated with memory in this study and with schizophrenia in a previous report (Carroll et al. 2010). Black boxes indicate coding exons and the open box represents the 3′-untranslated region. The SNP analysed in the current study is underlined and shown in larger font. Previously published SNPs are shown as dots below the gene; connecting lines indicate an associated haplotype. rs62621679 and rs62621677 are only 2 bp apart. The physical distances between all other SNPs are shown below the gene diagram.

Download figure to PowerPoint

The joint effect of GRM3- and PRKCA-associated alleles

NMDA receptor mGluR3 and PKCα are physically and functionally interacting proteins with closely related transmission/occlusion balancing functions at the synapse (Buchner et al. 1999; De Blasi et al. 2001; Tyszkiewicz et al. 2004). In this study, the associated alleles in both genes had opposite effects in patients relative to controls. The published functional data and our own genetic findings suggested a possibly common molecular mechanism of the observed association. We investigated the effects of double carrier status, hetero- and homozygous (n = 88 total, including 71 schizophrenia cases and 17 controls), for the associated minor alleles of rs2189814/GRM3 and rs8074995/PRKCA on memory performance in schizophrenia patients and on susceptibility to developing the clinically diagnosable disease. The combination was associated with lower RAVLTi scores in cases, with a β-coefficient compounding the individual effects of the two SNPs (Table 2). Association analysis in the overall case–control sample showed that double carriers also had an increased risk of developing schizophrenia [odds ratio (OR) 2.3 [95% confidence interval (CI) = 1.23–4.28], P = 0.008].

Lack of effect of the associated variants in the CS cluster

The three polymorphisms discussed earlier (rs220599, rs2189814 and rs8074995) were analysed post hoc to check whether the observed effects could also be detected in CS schizophrenia cases and in the joint, clinically defined schizophrenia dataset (including the non-CD/non-CS cases). This analysis (Table S3) showed no effect on RAVLTi or RAVLTd test scores either in the total group of affected subjects or in the subset of CS cases. However, rs220599 showed an association (P < 0.05) with current IQ in the total schizophrenia group.

Discussion

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Acknowledgments
  8. Supporting Information

The main question addressed in this study was whether polymorphisms associated with normal variation in episodic memory (de Quervain & Papassotiropoulos 2006) play a role in memory impairment in schizophrenia. Our analyses replicated the original findings for three of the genes investigated (GRIN2B, GRM3 and PRKCA) in our control group. Several methodological and technical differences between the original and this study put certain caveats on the interpretation of our findings. Although targeting the same construct of episodic memory, the verbal memory test employed by de Quervain and Papassotiropoulos (2006) consisted of six consecutive (visual) presentations of five different words each, with immediate recall after each presentation and delayed free recall after 5 min. The RAVLT version used in our study involved three consecutive (auditory) presentations of the same list of 15 words, with immediate recall after each presentation and delayed recall after 20 min. The 15-word list is likely to put a greater processing demand on immediate encoding and recall, and is more likely to uncover pre-existing deficits (especially in patients), compared to the less demanding 5-word lists which may have produced, in a proportion of healthy participants, a ceiling effect resulting in less performance variance overall. This might explain, in part, the absence of a significant genetic association of the tested polymorphisms with immediate recall in the original study.

Our analyses also showed a contribution of GRIN2B, GRM3 and PRKCA to dysfunctional memory performance in schizophrenia with pervasive CD. No significant effects on memory were observed in CS patients and in the total schizophrenia sample. This lack of effect can be interpreted in the context of heterogeneity. The pathogenetic heterogeneity of clinically defined schizophrenia is our fundamental assumption, and the utility of endophenotyping is supported by the findings in this study. At the same time, our previous investigations suggest that the CS cluster harbours further heterogeneity (Hallmayer et al. 2005). This diversity and the small sample size are likely to account for the failure of the CS cluster to mimic either controls or CD patients.

The genes encoding NMDA receptors are obvious candidates targeted in numerous association studies of schizophrenia (Allen et al. 2008; Harrison & Weinberger 2005; Li & He 2007). The GRIN2B and GRM3 SNPs associated with memory performance in schizophrenia patients in this study are located in haplotype blocks comprising SNPs previously reported to be associated with schizophrenia (Fig. S1). The inconsistent findings in previous association studies could arise in part from the composition of the schizophrenia samples, with negative results more likely in samples where patients with significant CD are under-represented.

Unlike the neuronal receptors, PRKCA was implicated in schizophrenia susceptibility for the first time in a recent study (Carroll et al. 2010), where a rare four-marker haplotype (referred to as C-HAP) was found to occur in the homozygous form in the affected members of a multiplex family previously (Williams et al. 2003) linked to chromosome 17. This haplotype was also associated with broadly defined psychotic illness in a large UK case–control sample (Carroll et al. 2010). Association mapping of the entire linked region identified another variant, rs873417, downstream of PRKCA, which reached experiment-wide significance (Carroll et al. 2010).

PRKCA SNP rs8074995, associated with memory in both the original (de Quervain & Papassotiropoulos 2006) and in the present study, falls within the interval covered by that haplotype (Fig. 1) and is located very close to rs62621676 (∼8 kb) and rs873417 (30 kb). HapMap data indicate that rs8074995 and rs873417 are not in strong LD, and the widely divergent MAFs of rs8074995 (0.132 in this study; 0.100 dbSNP) and rs62621676 (0.009 in this study; 0.009 in cases and 0.005 in controls in the study by Carroll et al. 2010) suggest no correlation. While our findings do not constitute a direct replication of the UK study, they support a role of PKC and specifically of variation in the 3′ end of the gene, in susceptibility to schizophrenia and memory impairment in patients.

Two of the associated genes in our investigation showed differences in the direction of effects in affected compared to control subjects. While the GRIN2B rs220599 effect was as expected based on the de Quervain and Papassotiropoulos (2006) results and did not differ between patients and controls, GRM3 and PRKCA showed an intriguing reversal – enhancing memory performance in control subjects, in agreement with the original observations (de Quervain & Papassotiropoulos 2006), and reducing it in schizophrenia patients with cognitive deficit. The associated alleles in these genes complemented each other in increasing the risk of disease and memory impairment.

The encoded proteins mGluR3 and PKCα are closely related by direct and indirect interactions and feedback loops at the synapse, involved in the balancing of glutamatergic signalling and the control of glutamate-induced excitotoxicity. The ‘neuroprotective receptor’ mGluR3 is highly enriched in cortical neurons at both pre- and postsynaptic sites and in glial cells and provides a homeostatic feedback for optimizing the signal-to-noise ratio of glutamatergic neurotransmission (Cartmell & Schoepp 2000; Corti et al. 2007; Tyszkiewicz et al. 2004). In turn, PKC is believed to play a role in glutamate excitotoxicity (Buchner et al. 1999). The mechanisms of these balancing effects involve the transporter EAAT2, which is responsible for the bulk of central nervous system (CNS) glutamate uptake (Beart & O’Shea 2007). EAAT2 levels increase in response to mGluR3 activation in vitro (Aronica et al. 2003) and are reduced in GRM3 knockout mice (Lyon et al. 2008). In turn, PKC is involved in the rapid modulation of EAAT2 activity by inducing its removal from the cell surface (Kalandadze et al. 2002).

Dysfunction of the finely tuned system of glutamatergic signalling is emerging as a major mechanism in schizophrenia pathogenesis (Gaspar et al. 2009; Lewis & Gonzalez-Burgos 2006; Lisman et al. 2008) and changes in EAAT2 expression in schizophrenia brains may be an important contributor to that dysfunction (Lauriat et al. 2006; Matute et al. 2005; Smith et al. 2001). The effects of antipsychotic medication are partly due to normalized glutamate transport, e.g. clozapine reduces cortical EAAT2 expression and function in experimental animals and in vitro (Melone et al. 2001; Vallejo-Illarramendi et al. 2005), and augments NMDA-induced responses through PKC activation in pre-frontal cortex pyramidal neurons (Jardemark et al. 2003). Polymorphic variants that normally contribute to an optimal signal/noise/toxicity balance could, in the schizophrenia brain, add to compromised neuroprotection and further blunting of the NMDAR response, thereby resulting in opposite effects on memory performance in control and affected subjects. Reversal of the direction of effect is not uncommon in association studies (Lin et al. 2007). Statistical modelling of the phenomenon (Clarke & Cardon 2010; Lin et al. 2007) suggests that genuine allele flips may occur due to differential environmental exposures between study populations. In our study, schizophrenia patients and healthy normal controls, although ethnically similar, are likely to have been subject to different environmental exposures, ranging from effects of medications or other substance use to chronic stress. The impact of such effects is likely to be augmented in schizophrenia patients by an abnormal internal brain environment and an inherent genetic vulnerability of brain structures involved in memory processing.

Memory dysfunction in schizophrenia has been extensively investigated, including systematic reviews and meta-analyses (Cirillo & Seidman 2003; Danion et al. 2007; Heinrichs & Zakzanis 1998). Most studies converge on an early episodic memory dysfunction (involving encoding, retrieval but not forgetting) as one of the most robust deficits with large effect sizes (Cohen's d) ranging from 0.91 to 1.41 (Heinrichs & Zakzanis 1998). Recall deficits are thought to result from defective strategic processes at encoding and semantic context binding, resulting, for instance, in an increase in non-list intrusions which impede associative learning (Danion et al. 2007; Grillon et al. 2010). Several subtypes of memory impairment in schizophrenia have been proposed based on relatively small samples ( Bruder et al. 2004; Ilonen et al. 2004; McDermid & Heinrichs 2002; Turetsky et al. 2002). However, few studies have addressed the question whether memory impairment in schizophrenia is a genuine focal deficit or a spill-over effect of a generalized cognitive dysfunction (analogous to Spearman's g) affecting most cognitive domains. Considering the results reported earlier, we examined the effects of the memory-associated SNPs on performance in the tests addressing other cognitive functions (Table S3).

Two findings require a comment. First, current IQ, assessed by the Shipley Institute of Living Scale (SILS) test, showed an association (P < 0.05) with rs220599 in the GRIN2B gene in the total schizophrenia sample. Second, verbal fluency (the FAS version of the Controlled Oral Word Association Test) was associated (P < 0.01) with rs8074995 in the PRKCA gene in the CD cluster of schizophrenia cases. Notably, SILS comprises a vocabulary subtest assessing verbal comprehension and memory, and an abstraction subtest assessing cognitive flexibility. Post hoc analysis showed that the effect of GRIN2B was stronger on the verbal component involving memory (P = 0.002) than on the abstraction component (P = 0.05). Similarly, the phonemic FAS verbal fluency task requires effortful lexical retrieval from the memory store, and engages executive function. As nearly every neuropsychological test targeting a particular cognitive domain, such as verbal memory, practically measures several interrelated but different cognitive operations (Jaeger et al. 2006), these findings are consistent with our interpretation of the results as tending to converge on aspects of verbal memory, rather than pointing to a generalized effect on multiple cognitive domains. We also constructed a correlation matrix to examine other relevant cognitive functions in the same subjects (Table S4A). The partial correlations (adjusted for age and sex) show that (1) immediate and delayed recall measures are significantly correlated within all subgroups (controls and schizophrenia patients, including the CD and CS clusters); (2) immediate and delayed recall are significantly correlated with previous and current IQ, sustained attention (Continuous Performance Test, Identical Pairs version) and executive function (verbal fluency) in the schizophrenia sample as a whole, but not in the normal controls; (3) immediate and delayed recall are uncorrelated with previous and current IQ, sustained attention or executive function within the clusters of CD and CS cases. This finding is consistent with Meehl's (Meehl 1990; Waller 2006) observation that ‘seemingly unrelated’ symptoms of schizophrenia, which are highly correlated in admixture (commingled) samples, are uncorrelated in relatively ‘pure’ samples, supporting a taxonic model of relatively discrete subpopulations. Furthermore, it is consistent with the statistical properties of the GoM model from which the CD/CS clusters were derived, where each defining variable of a ‘PT’ is a priori unconfounded from the effects of any other concurrent variables. To rule out the possibility that the loss of significant correlations within the CD and CS clusters could result from reduced sample sizes and concomitant loss of power, we drew from the total schizophrenia sample three random subsamples, each approximately equal in size to the respective CD and CS clusters. Within each of the random subsamples, the correlations observed in the total schizophrenia sample remained statistically significant (Table S4B).

This analysis supports the rationale for splitting the total schizophrenia sample into case clusters based on cognitive endophenotype profiling prior to genetic association analysis and suggests that the significant memory deficit, which is a salient feature of the CD cluster of schizophrenia patients, is not secondary to a hypothetical generalized dysfunction factor, but a relatively independent contributor to the multicomponent deficit profile characterizing this subset of cases.

References

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Acknowledgments
  8. Supporting Information

Acknowledgments

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Acknowledgments
  8. Supporting Information

The study was supported by grants #37580400 and #37580900 of the National Health and Medical Research Council of Australia to Professor A.J., with funding contribution from the North Metropolitan Health Area, Perth, Western Australia. The authors thank the patients, family members and other volunteers who participated in the study and the staff of the North Metropolitan Health Area mental health services in Western Australia for assistance in patient recruitment.

Supporting Information

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Acknowledgments
  8. Supporting Information

Supporting Information

Additional Supporting Information may be found in the online version of this article:

Figure S1: Diagrams of genes showing the approximate location of SNPs associated with memory in this study and with schizophrenia in previous reports. Open boxes indicate 5’- and 3’-untranslated regions and black boxes indicate coding exons. Single nucleotide polymorphisms analysed in the current study are underlined and shown in bold font. Previously published SNPs are shown as dots below each gene; connecting lines indicate associated haplotypes. Linkage disequilibrium patterns between SNPs analysed in the current study (boxed) and previously reported schizophrenia-associated SNPs, present in HapMap Phase II and III, are shown below each gene diagram. For each gene, the image on the left represents D′ values, and the one on the right r2 values. No genotype information on CEPH samples was available for rs1019385 in GRIN2B and rs6465084 and rs1989796 in GRM3 for this analysis. (a) GRIN2B (NM_000834). The marker nomenclature used in older publications is given in brackets following the rs&num;. Asterisks indicate SNPs showing association in meta-analyses, (b) GRM3 (NM_000840).

Table S1: Genetic association with RAVLT scores by genotype group in schizophrenia patients with cognitive deficit.

Table S2: Genetic association with RAVLT scores by genotype group in the controls (adjusted for age and sex).

Table S3: Raw data on neurocognitive tests by sample subgroup genotypes.

Table S4: (A) Correlation coefficients between neurocognitive variables (age and sex adjusted). (B) Correlation coefficients between neurocognitive variables in the three randomly selected (∼50%) subsets of SZ cases (age and sex adjusted).

FilenameFormatSizeDescription
GBB_679_sm_figS1.doc480KSupporting info item
GBB_679_sm_tableS1.doc45KSupporting info item
GBB_679_sm_tableS2.doc64KSupporting info item
GBB_679_sm_tableS3.doc92KSupporting info item
GBB_679_sm_tableS4.doc93KSupporting info item
GBB_679_sm_methods.doc33KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.