Introduction
A major challenge in modern medicine is to understand cellular and molecular mechanisms underlying common mental illnesses such as schizophrenia, which involve complicated genetic and environmental determinants1. Prevailing etiological hypotheses maintain that the interaction of multiple genetic factors with each other and with environmental risk factors results in neurodevelopmental abnormalities predisposing to illness later in life1. Given this complexity, together with imprecise and differing definitions of phenotype, it is not surprising that statistical association studies to identify risk genes for schizophrenia have not been easily reproduced1, 2. Gene identification for complex polygenic disorders such as schizophrenia will probably require demonstration that risk variants affect the biology of the gene in a manner that converges on key aspects of the biology of the illness3. Indeed, this approach has proven to be crucial in a number of complex genetic disorders, such as adult onset diabetes, in which multiple genes each account for only a very small share of risk but show stronger effects on related intermediate phenotypes even in normal individuals, for example, body mass index4 or glucose-induced insulin release5. Potential biological intermediate phenotypes related to risk for schizophrenia include abnormalities related to hippocampal formation and prefrontal cortices, which are consistently reported in patients with schizophrenia and in their healthy relatives6, 7, 8, 9, 10, 11. Thus, it follows that genes weakly associated with susceptibility for schizophrenia might show relatively robust effects on prefrontal cortex and hippocampal formation function in risk allele–carrying populations.
Here we identify a previously undescribed potential schizophrenia susceptibility gene, KCNH2 (a human ether-a-go-go (ERG)-family potassium channel), by meta-analyses of five independent clinical association data sets. Moving beyond statistical association with clinical diagnosis, we describe its association with several schizophrenia-linked biological intermediate phenotypes in large samples of normal individuals who carry risk-associated alleles. In searching for a molecular mechanism, we discovered a previously undescribed primate- and brain-specific KCNH2 isoform (3.1), encoded in close proximity to the risk-associated single nucleotide polymorphisms (SNPs). Expression of KCNH2-3.1 is specifically increased within the hippocampal formation of individuals with schizophrenia and in normal individuals who carry risk-associated alleles. Expression of KCNH2-3.1 in rodent cortical neurons causes a marked alteration in KCNH2 channel physiology resulting in high-frequency, nonadapting neuronal firing patterns. Such changes in spike frequency may underlie abnormalities in neuronal firing thought to be a fundamental aspect of cortical dysfunction in psychosis12. These convergent results implicate KCNH2 in primate cortical physiology and in the etiology and pathophysiology of psychosis, making it a potentially new target for antipsychotic therapy.
Results
Identification of KCNH2 association with schizophrenia
It has become increasingly apparent that genetic risk for schizophrenia and many other complicated illnesses are not related to major amino acid changes or highly detrimental protein mutations, although these may still occur in rare forms of the illness13. Instead, the molecular effects of susceptibility alleles are more likely to be subtle, relating to the regulation and splicing of transcripts or proteins. Thus, the gene expression profiles of tissue from individuals with the disease are a potentially informative starting point to search for candidate risk genes14, 15. We selected ten genes previously reported as differentially expressed in schizophrenic brain16 and genotyped haplotype-tagging SNPs in 170 families of European descent having an offspring with schizophrenia. Family-based association test (FBAT) analysis17 of gene-based haplotypes showed significant global association with the genomic region of nitric oxide synthase-3 (NOS3; 7q36.1; P < 0.005 and P < 0.05 after Bonferroni correction for ten association tests) with the diagnosis of schizophrenia (Supplementary Table 1 online). In the European sample (CEU) of the HapMap Project18, the block tagged by these SNPs spans portions of NOS3 and KCNH2 (Fig. 1 and Supplementary Data 1 online). The SNP in this 7q36.1 region that was most strongly associated with schizophrenia is within KCNH2 and was still associated with the disease after correction for the total number of SNP tests performed across all ten genes (rs1036145, corrected P = 0.042, Supplementary Table 1).
Figure 1: Genetic association of 7q36.1 with risk for schizophrenia.
Top, inverse log of the P value for single SNPs from association results for the CBDB/NIMH Sibling Study (blue), for the NIMHGI study (black), for the German case-control study (green) and for the pooled five–data set meta-analysis (Pooled, red). Only markers with P values less than 0.1 are shown. A physical map of the region is given and depicts known genes within the region. Bottom, the linkage disequilibrium structure of the genotyped markers for 370 unrelated healthy control subjects of European descent and depicted as r2. Graphic created with the R package snp.plotter57.
Full size image (75 KB)KCNH2 as a potential risk gene for schizophrenia
We genotyped additional SNPs (43 in total) in an expanded sample of 296 families (Clinical Brain Disorders Branch (CBDB) family sample) from the 3' end of KCNH2 to the 3' end of NOS3, including 11 SNPs detected through resequencing of the region in individuals with schizophrenia. Six SNPs were significantly associated with schizophrenia at 
< 0.05 (five with P < 0.01; Fig. 1 and Supplementary Data 1 and 2 online). Most of the associated markers are in moderate to strong linkage disequilibrium, suggesting the region of interest maps to a 
3-kilobase (kb) of intron 2 of KCNH2 (Supplementary Data 1).
We tested for replication of association in a large German case-control data set, typing the six schizophrenia-associated SNPs found in the CBDB sample and one SNP with marginal significance (P = 0.057). We observed significant association at the genotype level for minor-allele homozygotes at four of these SNPs, M16, M17, M19 and M30 (Fig. 1 and Supplementary Data 2). We tested three additional samples with the same six markers: a family–based sample of European ancestry from the US NIMH Genetics Initiative project (NIMHGI)19, a case-control sample from Armenia and a case-control sample from Italy. Although the sizes and ascertainment strategies differed from those of the larger original samples, in these samples we observed significant association of SNPs in this region with schizophrenia and/or the six implicated SNPs tended toward the same direction of association, although not significantly (Fig. 1 and Supplementary Data 2).
Given these results and power limitations, we conducted a combined family-based and case-control meta-analysis with the R package case-control and transmission disequilibrium test meta-analysis package (Catmap)20. This meta-analysis showed SNPs M17, M30, M31 and M33 to be significantly associated with schizophrenia, with M30 showing the greatest odds ratio (1.17) across all samples (P = 0.0054; Supplementary Data 2). 'Leave-one-out' sensitivity analysis to determine whether the meta-analysis results were driven by the initial family-based result and therefore suggestive of a 'winner's curse' phenomenon21 revealed that removal of any of the included studies still led to significant association of KCNH2 with schizophrenia using the four remaining independent studies for SNP M30 (Supplementary Data 2). The odds ratios in the cumulative analysis are comparable with other genes reported as significantly associated with schizophrenia in larger meta-analyses22, 23 and exceed the Venice interim criteria for 'small summary' findings24. Taken together, these results identify a small intronic region of KCNH2 as a potential susceptibility locus for schizophrenia.
Effect of KCNH2 on cognition and cortical structure and function
Given the high expression of KCNH2 within the prefrontal cortex and hippocampal formation25, two brain regions subserving attention and memory processes and consistently implicated in the neuropathology of schizophrenia26, 27, we hypothesized that if risk-associated SNPs affect the biology of these brain regions, cognitive deficits referable to these regions would also be associated with risk genotypes, regardless of disease status. We selected three of the significantly associated SNPs from the meta-analysis (M30, M31 and M33, all of which are in strong linkage disequilibrium with the fourth positively associated SNP, M17) to test for their effects on seven independent summary measures of cognition28 (Supplementary Data 3 online) in a sample of healthy, unrelated controls independent from the previous control samples. The use of healthy controls for genetic association at the level of brain function avoids potential confounders related to chronic illness and medical treatment29. We observed significant association between risk genotypes for each SNP and performance on the intelligence quotient (IQ) and processing speed factor (P = 0.020; Supplementary Data 3). Thus, even in healthy controls, the same alleles observed more frequently in individuals with schizophrenia predict significantly lower IQ and processing speed. It is noteworthy that similar cognitive phenotypes have been especially strongly related to genetic risk for schizophrenia in discordant twin samples7.
We also tested whether the same three SNPs would relate to changes in brain structure and physiology. By magnetic resonance imaging (MRI)-based morphometry, we observed significant volume decreases in hippocampal formation structures ( = 0.05, false discovery rate (FDR) corrected), proportional to allelic load, in normal individuals carrying risk alleles of SNPs M30, M31 and M33 using whole-brain and region of interest analyses (Fig. 2a and Supplementary Data 4 online). We next tested whether the same SNPs would influence the physiology of cognitive processing related to hippocampal formation, again in healthy control carriers of risk alleles at SNPs M30, M31 and M33. We used functional MRI (fMRI) to assess regional activation during incidental encoding of a temporal lobe memory task30 in which task performance was controlled (see Supplementary Data 4 online for demographic statistics). We observed greater activation (Familywise error (FWE) correction = 0.05) of the hippocampal formation within healthy control risk allele carriers (Fig. 2b and Supplementary Data 4). This pattern of inefficient processing of memory information (that is, excessive activity for a fixed level of performance) suggests that the tuning of cortical microcircuits crucial for information processing within the hippocampal formation is relatively disordered in individuals with risk-associated alleles. Similar patterns of inefficient processing in hippocampal formation have been reported for other genes and diseases affecting cognition, for example31, 32.
Figure 2: Association of risk SNPs with cognitive measures, brain structure volumes and regional brain activity during memory-based tasks.
(a) M30 genotype versus hippocampal gray matter volume in healthy control subjects. Heatmap depicts statistical t values of linear decrease in regional gray matter volume from subjects homozygous for the risk associated allele, T, (n = 16) to heterozygote carriers (n = 61) to noncarriers (n = 64). Only voxels corresponding to a P < 0.05 FDR corrected threshold are shown. (b) Percentage blood oxygen level-dependent (BOLD) signal change in healthy controls during the encoding conditions of a declarative memory task in the left posterior hippocampus (Montreal Neurological Institue (MNI) coordinates of peak cluster: -34 mm, -25 mm and -15 mm) showing a significant (P = 0.0003) linear increase of activation of the risk associated allele at M30 in homozygote carriers (n = 14) relative to heterozygote carriers (n = 37) and to noncarriers (n = 28). Plot depicts voxel mean 
s.e.m. Heatmap colors correspond to statistical t values of the degree of increase in BOLD signal with each copy of M30 risk allele (T). Only voxels surviving P < 0.05 FWE correction are shown. (c) Thresholded (P < 0.05 FWE corrected) statistical t-maps and mean s.e.m. percentage BOLD signal change during the executive working memory task in the right DLPFC (MNI coordinates of peak cluster: 26 mm, 30 mm and 42 mm) showing a significant (P < 0.005) linear increase of activation of the risk associated allele at M30 in homozygote carriers (n = 24) relative to heterozygote carriers (n = 71) and to noncarriers (n = 81).
Although hippocampus may be engaged during executive cognition tasks that affect measures of IQ6, 33, recent studies have shown that general intelligence or IQ measures are more closely associated with executive control processes and prefrontal cortex–related structures34, 35, 36. Therefore, to more specifically investigate a connection between risk-associated alleles and executive-linked physiology, we used fMRI to measure cortical engagement during the N-back working memory task, an executive cognition task well validated as engaging prefrontal cortical circuitry37. Again, normal subjects carrying risk-associated alleles of SNPs M30, M31 and M33 showed significantly increased activity in an allele load pattern within the dorsolateral prefrontal cortex (DLPFC), despite controlled task performance (that is, inefficient cortical processing; Fig. 2c and Supplementary Data 4). In conjunction with significantly increased hippocampal formation activity during memory-encoding tasks, again in normal subjects, these data consistently suggest association between risk alleles and impairments in the efficiency of information processing within two areas of the brain implicated in the pathophysiology of schizophrenia.
Risk genotypes and KCNH2 mRNA levels
The association of KCNH2 with schizophrenia and with related brain phenotypes in multiple independent samples lends statistical and biological support to the involvement of this genomic region in risk of illness. However, these findings do not identify the underlying molecular mechanism. We investigated gene expression in human brain and its relationship to our evidence for genetic association (Fig. 3). Using quantitative RT-PCR, we found expression of full-length KCNH2 (1A) to be significantly lower in both the DLPFC (P = 0.016; Fig. 3a) and hippocampus (P = 0.003; Fig. 3a) of individuals with schizophrenia. Because most patients with schizophrenia receive antipsychotic drugs, which bind KCNH2 (ref. 38), some expression differences may be medication artifacts. We therefore measured KCNH2 expression within the frontal cortex of rats after 28 d of treatment with clozapine or haloperidol across a range of doses and did not find significant medication effects (Supplementary Fig. 1 online).
Figure 3: Regional gene expression and association with risk genotype.
(a) Differences in mRNA expression within the hippocampus of 29 subjects with schizophrenia and 59 healthy control subjects and within the DLPFC of a largely overlapping set of 31 subjects with schizophrenia and 69 healthy control subjects. Expression values are normalized fold differences versus the mean of healthy controls. P values represent main effect of diagnosis in the final regression model. (b) Association of M30 risk genotype (T carriers) with KCNH2-3.1 expression within the hippocampus (data from a) with respect to diagnosis. All error bars represent s.e.m.
Full size image (43 KB)To explore the possibility that the molecular mechanism of the clinical associations involves transcriptional regulation, we tested for genotype effects on expression of KCNH2-1A. No significant effects were observed (for example, for M30 P = 0.516; Supplementary Data 5 online). Expression of KCNH2-1B, a minor isoform expressed within the brain25, did not show significant differences between control subjects and individuals with schizophrenia, nor was it significantly associated with risk genotypes (data not shown). NOS3 expression also showed no association with these risk SNPs (for M30 P = 0.90; Supplementary Data 5 and Supplementary Note online). These data suggest that the molecular effects of genetic risk do not affect the expression of KCNH2-1A or KCNH2-1B and may instead relate to more complicated aspects of gene processing, such as the splicing of transcripts or the expression of other isoforms.
Identification of a brain-specific KCNH2 isoform in humans
Using rapid amplification of 5' complementary DNA ends (5'-RACE) on RNA pooled from the DLPFC of ten subjects with schizophrenia and beginning from exon 4 of the full-length transcript (KCNH2-1A; NM_000238) we identified a previously undescribed 5' extension of exon 3 producing an isoform that originates upstream of exon 3 (Fig. 1 and Supplementary Data 1 and 6 online). Further cloning from a human complementary DNA library revealed that the isoform is expressed endogenously and contains all of the downstream exons of the full-length gene (through exon 15 of KCNH2-1A; Fig. 4a and Supplementary Data 7 online). In silico prediction of the longest open reading frame of KCNH2-3.1 suggests that the majority of the 5' extension of exon 3 would be untranslated and that the first methionine is in frame with the conserved portion of KCNH2-1A (Supplementary Data 7). As such, KCNH2-3.1 is expected to be missing the first 102 amino acids of KCNH2-1A, replacing them with six amino acids unique to this isoform. The translation of KCNH2-3.1 and the predicted difference in protein size was confirmed by western blot analysis of transfected HEK cells (Fig. 4b). Western blot analysis of human hippocampus and frontal cortex revealed a protein band of similar molecular weight to that in transfected HEK cells (Fig. 4b). Furthermore, neuroblastoma cells transfected with constructs encoding either or both KCNH2-1A or KCNH2-3.1 show overlapping expression of the protein isoforms on the plasma membrane (Fig. 4c).
Figure 4: Detection and quantification of KCNH2-3.1 mRNA and protein.
(a) Gel with PCR products of KCNH2-1A and KCNH2-3.1 using isoform-specific primer pairs in human heart, hippocampus and fetal brain tissue extracts. PCR products for KCNH2-1A correspond to 373 base pairs (bp) and 463 bp, lanes 1 and 2, respectively, and for KCNH2-3.1 at 388 bp and 478 bp, lanes 3 and 4, respectively. The band at 1.5 kb in Iso1-Ex4 lanes corresponds to small amounts of unspliced RNA or genomic DNA contaminant present in the samples (lane 4). (b) Protein expression of KCNH2-1A and KCNH2-3.1 in human, primate and mouse frontal cortex. Positive control lanes include HEK cells transfected with full-length KCNH2-1A (WT) and KCNH2-3.1. Proteins extracted from human brain regions show two distinct bands, one equivalent in size to that found in KCNH2-3.1–transfected HEK cells. However, the larger protein band observed in human brain did not correspond to the size of KCNH2-1A–transfected HEK proteins. Instead, the larger band occurs at 160 kDa, which is the reported size of KCNH2-1A from in vivo protein extracts, suggesting post-translational modifications. (c) Primary rat neurons transfected with KCNH2-1A or KCNH2-1A–containing vectors.
To assess the potential organ selectivity of KCNH2-3.1, we measured the ratio of KCNH2-3.1 to KCNH2-1A mRNA levels across several tissues, including heart and three different brain regions, by quantitative RT-PCR. We found that the two forms are comparable in their expression within several brain regions (Supplementary Data 6), but KCNH2-3.1 was over three orders of magnitude less abundant than KCNH2-1A within the heart, suggesting a brain-specific role of KCNH2-3.1. As ERG family genes are moderately conserved in sequence across several species39, we expected KCNH2-3.1 to be present within other mammals. By 5'-RACE, we were unable to detect KCNH2-3.1 homologs in mouse brain (data not shown). Moreover, in silico analysis revealed that the 1.1-Kb region unique to KCNH2-3.1 was highly degenerate outside of primates (Supplementary Data 7). Protein expression of KCNH2-3.1 was undetectable in mouse brain but was abundant within rhesus monkey DLPFC (Fig. 4b).
As neuronal gene expression is tightly regulated and carefully orchestrated during development, we measured the expression profiles of KCNH2-1A and KCNH2-3.1 throughout brain development in the prefrontal cortices of 283 subjects, including 39 prenatal samples (14–20 weeks of gestation) (see Supplementary Fig. 2 and Supplementary Data 8 and 9 online for sample information). Isoform expression patterns were similar throughout postnatal life, but we observed marked differences in their relative expression prenatally (Supplementary Fig. 2). KCNH2-3.1 transcript levels are markedly increased prenatally relative to adult levels but then decrease and stabilize shortly after birth (Supplementary Fig. 2). In contrast, KCNH2-1A expression increases throughout prenatal development until it reaches a maximum level that is sustained throughout postnatal life (Supplementary Fig. 2). Although these data offer only a qualitative snapshot of developmental expression patterns, they suggest differential regulation of these isoforms during early brain development and a specific role for KCNH2-3.1 during these early stages.
Risk genotypes and diagnosis predict KCNH2-3.1 mRNA changes
KCNH2-3.1 mRNA expression was significantly increased within the hippocampal formation of subjects with schizophrenia versus healthy control subjects (P = 0.012, Fig. 3a). Moreover, the ratio of KCNH2-3.1 to KCNH2-1A was 2.5-fold higher in individuals with schizophrenia (P = 0.0010). This ratio may reflect how isoforms and homologs of KCNH2 assemble to create heteromeric potassium channels with unique electrophysiological properties25, 40. Additionally, higher levels of KCNH2-3.1 mRNA were significantly associated with risk alleles at SNPS M30, M31 and M33, even within healthy control subjects (Fig. 3b). Specifically, diagnosis and M30 genotype each had independent significant effects on KCNH2-3.1 expression (P = 0.006 and 0.002, respectively) as well as a significant interaction on expression (P = 0.028; Fig. 3b), suggesting that although risk alleles are associated with increased expression of KCNH2-3.1 regardless of subject group, the effect is more pronounced in subjects with schizophrenia (Fig. 3b). We observed this trend irrespective of ethnicity, reducing the potential confounder of genetic background (Supplementary Data 5). We found similar results with M31 and M33, although interactions with diagnosis were not significant (data not shown). These data support the hypothesis that the mechanism by which the disease-associated SNPs contribute to risk for schizophrenia and related phenotypes involves the regulation of KCNH2-3.1 transcription and further suggest that overexpression of this isoform would result in physiological effects related to the pathogenesis of the disorder.
Functional role of KCNH2-3.1 revealed by electrophysiology
KCNH2, (also known as hERG1), a member of the tetrameric ERG-related family of potassium channels39, is best known for its role in slow repolarization during the myocardial action potential, thereby regulating the QT interval41. Its distinct physiological features, namely slow activation, fast inactivation and slow and voltage-dependent deactivation39, 40, 42, make KCNH2 an excellent candidate for controlling neuronal firing patterns and cortical network oscillation43, 44, 45. Modeling work predicts that KCNH2 is crucial for spike-frequency adaptation, or a gradual termination of a train of evoked action potentials, a common firing pattern seen in excitatory neurons in the brain46. Several studies have shown that inhibition of KCNH2 converts low-frequency, adapting to high-frequency, non-adapting neuronal discharges46, 47, a firing pattern necessary for sustained neuronal activity subserving numerous complex cognitive functions48.
The slow deactivation of the KCNH2 channel is mediated by a PYP-like sensor domain (amino acids 25–136) of the KCNH2-1A amino terminus42, most of which is missing in KCNH2-3.1 (Fig. 5a). We performed whole-cell recording on HEK cells transfected with either KCNH2-1A or KCNH2-3.1. When depolarized from -60 mV to various membrane potentials, KCNH2-3.1 was activated faster than KCNH2-1A (Fig. 5b, and Supplementary Fig. 3 online).
Figure 5: Characterization of KCNH2 currents in HEK293T cells expressing KCNH2-1A and KCNH2-3.1.
(a) Schematic diagram of KCNH2-1A and KCNH2-3.1 domain structures. Blue box, PYP-like sensor (PAS) domain. Green box, conserved amino acid sequences between isoforms. Numbers correspond to amino acid positions. (b) Currents evoked by voltage steps (4 s) from resting voltage (VH) of -60 mV to potentials from -100 mV to +50 mV in 10-mV increments, followed by a voltage pulse to -120 mV (5 s). Top, voltage protocol. Middle and bottom: traces (corrected for leak currents) recorded from cells transfected with KCNH2-1A (middle) and KCNH2-3.1 (bottom) cDNAs. (c) Effects of E-4031 on tail currents evoked by a test pulse to -120 mV from holding potential of +50 mV, using the same protocol as in b. Traces recorded from the same cells before (black) and after (red) treatment with 10 
M E-4031 are superimposed. Top and bottom graphs show KCNH2 currents recorded from cells transfected with KCNH2-1A and KCNH2-3.1 cDNAs, respectively. (d) Steady-state activation curves of KCNH2-1A and KCNH2-3.1 tail currents, which were induced by the protocol shown in b. Numbers in parentheses indicate the number of cells recorded. g/gmax, conductance/maximum conductance ratio. (e) Tail currents evoked by voltage steps from +60 mV to potentials between -120 mV and -70 mV in 10-mV increments. Traces in top, middle and bottom graphs represent tail currents of cells transfected with KCNH2-1A, KCNH2-3.1 and both KCNH2-1A and KCNH2-3.1, respectively. (f) Semilogarithmic plot of deactivation time constants of KCNH2-1A and KCNH2-3.1 currents at various repolarizing voltages. Top plot (between -210 mV and -100 mV) corresponds to 
2 for HEK cells transfected with KCNH2-1A alone or cotransfected with KCNH2-1A and KCNH2-3.1, whereas the bottom plots correspond to 1 for all transfection combinations. Data are means s.e.m.
When the membrane potential was hyperpolarized to -120 mV after depolarization, the KCNH2-1A–expressing cells showed a large, slowly deactivating 'tail current', which is completely blocked by the hERG-specific inhibitor E-4031 (ref. 49) (Fig. 5c), suggesting that this current is mediated solely by hERG channels. In contrast, cells transfected with KCNH2-3.1 showed a markedly lower tail current (Fig. 5c), which was also blocked by E-4031. The peak current density-voltage relationship (I-V curve) showed a substantial reduction in the peak amplitudes, particularly when cells were more substantially depolarized initially (Supplementary Fig. 3). In a conductance-voltage plot, KCNH2-3.1 showed a slight, but significantly reduced, steady-state tail current activation (Fig. 5d; V1/2: KCNH2-1A = -16.37 2.55 mV; KCNH2-3.1 = -6.54 4.04 mV, P < 0.05).
When the cells were first depolarized and held at +60 mV and then hyperpolarized at various potentials (Fig. 5e), the inward currents mediated by KCNH2-3.1 decayed toward baseline much faster than those mediated by KCNH2-1A (Fig. 5e). The deactivation phase of the KCNH2-1A currents was best fitted by a double-exponential curve, generating a slow and a fast deactivation constant (1 and 2), whereas that of KCNH2-3.1 could be fitted only by a single-exponential curve, with a much slower 1 at all repolarizing voltages tested (Fig. 5f). Thus, KCNH2-3.1 mediates an inward rectified K+ current with markedly faster deactivation kinetics. Furthermore, when KCNH2-1A was coexpressed with KCNH2-3.1, the deactivation phase of the tail currents could be fitted by a double-exponential curve, with 1 and 2 that were both significantly lower than those in cells expressing KCNH2-1A alone (Fig. 5f; analysis of variance, P < 0.05), suggesting codominant channel behavior intermediate to each channel.
We next examined the functional role of KCNH2-3.1 in cortical neurons. Given that rodent central nervous system neurons express KCNH2-1A, but not KCNH2-3.1, KCNH2-3.1 introduced into these neurons would presumably form complexes with the endogenous KCNH2-1A, thereby changing the deactivation kinetics of the K+ current. In rat primary cortical neurons transfected with a GFP marker alone, application of a voltage protocol similar to that in Figure 5e elicited a series of tail currents that were blocked by E-4031 (Fig. 6a). Subtraction of the currents after E-4031 treatment from control currents generated pure KCNH2-1A–mediated current that showed typical, slow deactivation kinetics (Fig. 6a). In contrast, application of the same protocol to neurons transfected with KCNH2-3.1 (plus GFP marker) elicited a considerably reduced tail current with significantly faster deactivation (Fig. 6a). At -100 mV, the introduction of KCNH2-3.1 reduced the 2 (initial fast component) and 1 (slow component) of tail currents by 83% and 92%, respectively (Fig. 6b). Transfection of human KCNH2-1A into rat cortical neurons resulted in tail currents very similar to those of untransfected or GFP-transfected neurons (Supplementary Fig. 3), further supporting that KCNH2-1A and KCNH2-3.1 have very different deactivation kinetics.
Figure 6: Effect of KCNH2-3.1 on repolarization-induced tail currents and firing patterns in cortical neurons.
(a) KCNH2-mediated tail currents in GFP-(left) and KCNH2-3.1–transfected (right) neurons before and after application of E4031. Top plot diagrams the voltage protocol. Bottom plot represents E-4031–sensitive current (that is, hERG current) generated by subtracting the before inhibitor and after inhibitor currents. (b) Semilogarithmic plot of deactivation time constants of E-4031–sensitive currents at various repolarizing voltages in transfected primary cortical neurons. The -120 mV to -100 mV decay of E-4031–sensitive currents was fitted by double-exponential functions (bottom curves: 1, top curves: 2), whereas the -90 mV to -70 mV time course followed single-exponential functions (). (c) Effect of KCNH2-3.1 overexpression in rat cortical neurons on action potential discharge evoked by long depolarizing pulse (40 pA, 1 s) before (left) and after (right) application of E-4031. (d) Spike frequencies (number of spikes per second) versus applied depolarizing currents in transfected primary cortical neurons. (e) Effect of KCNH2-3.1 on spike frequency adaptation, depicted as instantaneous frequency (inverse of interpulse interval) versus the corresponding spikes interval evoked by a 4-pA depolarizing pulse. (f) Effect of E-4031 on repetitive action potential discharge of transfected cortical neurons. Data are means s.e.m. *P < 0.01.
To test whether endogenous KCNH2-1A and exogenous KCNH2-3.1 form a functional complex, we examined action potentials induced by a step depolarization under the current-clamp conditions. In neurons transfected with KCNH2-3.1, a 1-s depolarization induced significantly more spikes compared with control (untransfected) and GFP-transfected neurons (Fig. 6c). A systematic analysis showed a general increase in spike frequencies in KCNH2-3.1–expressing neurons injected with more than 30 pA depolarizing current (Fig. 6d,e). E-4031 significantly increased firing frequency, even in neurons expressing KCNH2-3.1 (Fig. 6f). Again, the firing behavior of rat neurons expressing human KCNH2-1A was very similar to that of control neurons (Supplementary Fig. 3c,d).
In addition to increasing firing frequency, expression of KCNH2-3.1 seemed to change the firing pattern from a typical, adapting pattern to a nonadapting pattern (Fig. 6c). To better quantify this change, we measured instantaneous frequency (inverse of the first inter-pulse interval versus spike number). Indeed, both control and GFP- expressing neurons exhibited a gradual decline of instantaneous frequency, whereas instantaneous frequency remained stable during the entire course of depolarization in KCNH2-3.1–expressing neurons (Fig. 6e). Taken together, these results suggest that expression of KCNH2-3.1 in human prefrontal cortex significantly increases neuronal excitability and offers a potential mechanism for persistent or overactive neuronal discharges observed in neurons of the prefrontal cortex during working memory tasks48.
Discussion
We report a previously undescribed genetic association with schizophrenia within a small region of intron 2 of KCNH2, which was observed in a meta-analysis of five independent clinical data sets. Our results also suggest that the molecular mechanism of genetic susceptibility relates to changes in gene expression and regulation of an isoform of KCNH2 with unique physiologic properties. The sequence of hypothesis-driven experiments and tests leading to this convergent evidence, from initial genetic screening to genetic association with aspects of human hippocampal structure and hippocampal and DLPFC function and gene expression to confirmatory experiments in tissue culture is summarized in Supplementary Figure 4 online. Although we report here the significant association of several SNPs across independent samples and by meta-analysis, the exact risk structure may vary between populations, whereas the regional location and overall effect on KCNH2 gene processing may remain consistent with what is described here. The splicing mechanism related to variation in this region of KCNH2 is yet to be determined (see Supplementary Note).
To our knowledge, the region of interest on chromosome 7q36.1 has not, to date, been reported as a major locus in the few genome-wide association studies of psychiatric illnesses50, 51. These negative results may reflect the fact that such psychiatric genome-wide association studies have relied largely on Affymetrix 5.0 SNP arrays, whose coverage is notably lacking over this genomic region. Conversely, the P values reported here would not be significant if we corrected for all SNPs in the genome. Scanning the genome with hundreds of thousands of SNPs and employing the necessary rigid statistical correction because of no prior probability of any SNP being positive is a popular strategy at the moment, as it makes no assumptions about biology or function. This strategy has the appeal of a level of statistical significance being clear and incontrovertible. However, statistical significance does not, in and of itself, imply biological significance, nor does it necessarily identify the genes most likely to be important in unraveling new strategies for prevention and treatment.
To move beyond statistical association with clinical diagnosis and to obtain convergent evidence for association between KCNH2 and schizophrenia-related biology, we have performed a series of convergent experiments testing risk-associated SNPs on several intermediate biological phenotypes (Supplementary Fig. 4). Although we believe our deductive, hypothesis-driven strategy minimizes serendipity, it does involve a number of tests. Thus, the potential for spurious association because of multiple testing is necessary to consider. Three of the four SNPs showing association in a meta-analysis of five independent clinical data sets were tested for association with biological phenotypes related to risk for schizophrenia and to molecular processing of KCNH2 transcripts. A consistent pattern of allelic association, involving cognition and structural and functional imaging, was found in normal subjects and in expression of the 3.1 isoform in brain tissue from individuals with schizophrenia and controls of various ethnic backgrounds. The likelihood that by chance the same risk-associated alleles would predict variation in each of these independent phenotypes across these diverse samples and always in the direction of abnormality associated with illness is remote.
Exploration of the potential electrophysiological effects of elevated KCNH2-3.1 levels on neuronal repolarization and spike frequency has noteworthy implications for both normal cortical function and the pathophysiology of psychosis. Cortical information processing is dependent on well orchestrated, persistent neuronal firing within cortical circuits45. For example, neurons in the prefrontal cortex show sustained firing of high-frequency activity during the delay period that is thought to be crucial for working memory48. The cellular mechanism for this sustained firing is not clear. We found that expression of KCNH2-3.1 in cultured cortical neurons results in a much faster deactivation of the hERG channels, leading to a marked increase in spike frequency and a conversion from adapting to predominantly nonadapting firing patterns. Such sustained firing patterns may be important for higher-order cognition and suggest a role for the primate-specific KCNH2-3.1 in normal human cognitive processing. However, this key evolutionary change in activation kinetics may require specific titration of isoform abundances to achieve optimum signal-to-noise relationships. The 2.5-fold higher relative KCNH2-3.1 expression in the schizophrenic brain might imbalance this titration, resulting in abnormally increased neuronal excitability and disruption of neuronal firing patterns and regulation of signal-to-noise ratios12.
The discovery of KCNH2-3.1 may also have therapeutic implications. Some typical and atypical antipsychotic drugs bind and inhibit KCNH2 with affinities comparable to their affinities for dopamine D2 receptors38. Whereas D2 receptor affinity is thought to account for the therapeutic effects of antipsychotics, KCNH2 binding is responsible at least for side effects such as altered QT interval38 or even sudden cardiac failure52. Given that KCNH2 controls neuronal excitability and firing patterns, could the therapeutic effects of antipsychotic drugs also be related to their affinities for the brain-specific isoforms of KCNH2? Although the binding properties of antipsychotic drugs to KCNH2-3.1 remain to be tested, the unique structure of KCNH2-3.1, its role in nonadaptive firing, its low expression in heart and its elevated expression in brains of individuals with schizophrenia and genetic risk carriers make it reasonable to speculate that selective inhibition of KCNH2-3.1, but not of KCNH2-1A, would improve the disorganized firing in schizophrenia brain without eliciting cardiac side effects.
In conclusion, we have identified and characterized a previously undescribed, primate-specific, brain-enriched potassium channel with a potential genetic association with risk for schizophrenia. Healthy control carriers of risk alleles show a schizophrenia-like shift in cognitive traits (IQ and processing speed), hippocampal volume and physiological engagement during memory processing, and in prefrontal physiologic engagement during executive cognition. The mechanism of these associations seems to be related to genetic regulation of KCNH2-3.1, which has unique electrophysiological properties and expressional dependence on risk-associated genotypes. Together, these results may provide new insight into the etiology of schizophrenia and a fresh direction for therapeutic discovery.
Methods
Genetic association cohorts. We included five independent clinical data sets in this study; two family-based cohorts of European ancestry (367 families) and three case control cohorts also of European ancestry (1,158 cases, 1,704 controls) (see Supplementary Methods online for details). We took postmortem brain from the CBDB Postmortem Brain Collection and Life Span Series consisting of tissue from 31 subjects with schizophrenia, 69 control subjects, 39 fetal samples and an additional 244 control subjects covering a range of ages (Supplementary Methods and Supplementary Data 8 online). All research subjects gave written informed consent under a protocol approved by the human subjects research committees of the appropriate institution (see Supplementary Methods) and by the Combined Neuroscience Institutional Review Board of the US National Insistutes of Health (NIH) intramural research program.
Candidate gene screening. We selected 10 candidate genes from those previously reported as differentially expressed between individuals with schizophrenia and healthy controls16. We genotyped SNPs tagging haplotype blocks determined from HapMap Caucasian European (CEU) population (release 15) and overlapping with part or all of the genes of interest on a screening subsample of 175 families (698 subjects of European descent) from the CBDB family dataset. We genotyped all SNPs with 5-exonuclease TaqMan assays as previously described53. We tested haplotype association with the FBAT17, evaluated at = 0.05 using simulation with 1,000 iterations and subjected to Bonferroni correction.
KCNH2 genotyping and resequencing. We genotyped 43 SNPs within a 65.2-kb region of KCNH2 and NOS3 (chromosome 7 bases 150,280,464–150,345,679). We resequenced a total of 13.5 kb in 48 subjects with schizophrenia, including 10.4 kb flanking rs1036145 (chromosome 7 bases 150,299,575–150,309,948) and 3.1 kb upstream of exon 3 (chromosome 7 bases 150,287,750–150,290,865) with high cross-species conservation in the University of California at Santa Cruz genomes database (http://genome.ucsc.edu/). Resequencing found 11 previously undescribed SNPs and no coding polymorphisms (details in Supplementary Methods).
Cognitive testing. Depending on the cognitive test, data were available for between 230 and 330 normal individuals not part of the clinical association data sets. All subjects gave written informed consent under a protocol approved by the Combined Neuroscience Institutional Review Board of the NIH. The tests included tasks aimed at assessing a wide range of cognitive features significantly affected in subjects with schizophrenia and in their unaffected siblings12, 13, 14, 15. Factor analysis of 24 performance scores identified seven factors that explained 68% of the variance on these measures28. We used these seven factors to assess genotype effects on cognitive task performance (see below) and labeled them as follows: verbal memory, N-back, visual memory, processing speed and IQ, card sorting, attention and digit span. Further information regarding the test measures included in each factor is provided in Supplementary Data 3.
Statistical genetics analysis. We analyzed SNP association within families by FBAT17 and by logistic regression in unrelated cases and controls (Supplementary Methods). We performed meta-analyses with the R package Catmap20 based on fixed-effects estimates where the heterogeneity (Q statistic) of genetic effects was not significant at = 0.05 (markers M17, M19, M30, M31 and M33) and random-effects estimates (markers M16 and M20) where significant differences in genetic effects were found across samples (Supplementary Methods).
We tested genotype effects on cognitive intermediate phenotypes in control subjects using linear regression, adjusting for age and gender. We used permutation testing based on 1,000 replicates to examine the pattern of associations of SNPs M30, M31 and M33 with the cognitive factor scores to assess the likelihood that for each SNP the schizophrenia risk-associated allele would be associated with relatively poorer performance on the same factor.
Quantitative real-time PCR. Details about the collection, screening and dissection processes for human brain tissue have been previously described54, and demographic statistics about the brain specimens used here are provided in the Supplementary Methods. We measured gene expression by quantitative real-time RT-PCR. We designed probes to differentiate between KCNH2 isoforms through hybridizing to isoform-specific exons (Supplementary Methods).
Antipsychotic drug effects in rats. Briefly, we measured KCNH2 mRNA in the frontal cortex of male Sprague-Dawley rats (weight 250 g) treated chronically (for 28 d) with clozapine (0.5, 5 or 10 mg per kg body weight) and haloperidol (0.08, 0.6 or 2 mg per kg body weight) using Taqman ABI assay Rn00588515_m1. The expression data of a target gene were normalized to a geometric mean of three control genes. Further details are provided in the Supplementary Methods.
Custom Illumina microarray analysis of fetal tissues. Microarray chips were generated in the National Human Genome Research Institute Microarray Core Facility from 44,544 70-mer probes obtained from the Illumina Oligoset HEEBO (http://www.microarray.org/sfgf/heebo.do). We amplified RNA samples from the lifespan DLPFC collection (500 ng) and labeled them with fluorescent dye (Cy5)(GE Healthcare). We hybridized each sample to microarrays simultaneously with a reference standard (labeled with Cy3) consisting of a pool of RNA from many brain tissue samples. Analysis was done with limma R package55.
Imaging methods. All imaging subjects were healthy volunteers of European descent (not part of the clinical association data) without significant differences in age, gender, IQ score or education across genotype groups (Supplementary Data 4 contains demographic information). All subjects gave written informed consent under a protocol approved by the Combined Neuroscience Institutional Review Board of the NIH. All analyses were done in SPM (Statistical Parametric Mapping, http://www.fil.ion.ucl.ac.uk/spm) with SNPs effects assessed with random effects analyses. For gray matter volumes, we performed optimized voxel-based morphometry following standardized methods with appropriate covariates as described previously22, 23, 24 (Supplementary Methods) (n = 141).
During 3T blood oxygen level–dependent (BOLD) fMRI scanning described previously30, normal subjects performed a declarative memory task (n = 79) and an executive working memory task (N-back, n = 178) to robustly engage the hippocampal formation30 and dorsolateral prefrontal cortex37, respectively (Supplementary Methods). We performed random-effects analyses on a priori–defined regions of interest with genotype as a predictor (Supplementary Methods). We used FDR corrections ( = 0.05) to control for the expected proportion of false positives among suprathreshold voxels.
Isoform 3.1 cloning and expression. We performed 5'-RACE with a 5'-RACE System (v2.0) as recommended by the manufacturer (Invitrogen). We made template from commercially available total RNA from ten human cell lines and tissues (Stratagene; Supplementary Methods). We cloned isoform3.1 cDNA by long PCR with specific primers and high-fidelity DNA polymerase into vector pZero-Blunt (Invitrogen) and then subcloned it into vector pcDNA3 (Invitrogen). We transfected HEK293 cells (American Type Culture Collection) with the KCNH2-1A vector, KCNH2-3.1 vector, or empty vector using standard methods. We measured protein expression by western blotting and immunocytochemistry using standard techniques as described in the Supplementary Methods.
Electrophysiology. We cultured HEK293T (American Type Culture Collection) cells and primary cortical neurons were and transfected them by nucleofection, as previously described56. We identified transfected cells by GFP expression under fluorescence microscopy, and we performed whole-cell recordings as detailed in the Supplementary Methods. We obtained tail currents during 4-s voltage steps shown in the upper panels of Figures 5b, 5e and 6a, and we fitted them by exponential functions and analyzed them in detail as detailed in the Supplementary Methods. We also assessed transfected cortical neurons for repetitive firing properties using a long depolarizing pulse (40 pA, 1 s) as described in the Supplementary Methods.
Accession codes. The sequence for KCNH2-3.1 has been deposited in GenBank with accession number FJ938021.
Note: Supplementary information is available on the Nature Medicine website.
