Deficiency of Aph1B/C-γ-secretase disturbs Nrg1 cleavage and sensorimotor gating that can be reversed with antipsychotic treatment

Abstract

Regulated intramembrane proteolysis by γ-secretase cleaves proteins in their transmembrane domain and is involved in important signaling pathways. At least four different γ-secretase complexes have been identified, but little is known about their biological role and specificity. Previous work has demonstrated the involvement of the Aph1A-γ-secretase complex in Notch signaling, but no specific function could be assigned to Aph1B/C-γ-secretase. We demonstrate here that the Aph1B/C-γ-secretase complex is expressed in brain areas relevant to schizophrenia pathogenesis and that Aph1B/C deficiency causes pharmacological and behavioral abnormalities that can be reversed by antipsychotic drugs. At the molecular level we find accumulation of Nrg1 fragments in the brain of Aph1BC^−/− mice. Our observations gain clinical relevance by the demonstration that a Val-to-Leu mutation in the Nrg1 transmembrane domain, associated with increased risk for schizophrenia, affects γ-secretase cleavage of Nrg1. This finding suggests that dysregulation of intramembrane proteolysis of Nrg1 could increase risk for schizophrenia and related disorders.

Gamma-secretase activity, the cleavage of type I integral membrane proteins in their transmembrane domain, is performed by a protein complex that consists of four core components (1): presenilin, nicastrin, Pen2, and Aph1, likely present in a 1:1:1:1 ratio (2). As two presenilin genes and two Aph1 genes exist, at least four different complexes are possible (3, 4). We and others have previously shown that the Aph1A complexes are crucial for Notch signaling during embryogenesis (5, 6). Analysis of Aph1B gene function is complicated by the rodent-specific duplication of the gene (Aph1C). However, even combined inactivation of Aph1B and Aph1C did not result in a major phenotype (6), raising the question of the physiological function of the Aph1B complex. We demonstrate here in the Aph1BC^−/− mouse disturbed Nrg1 cleavage in the brain, a hypersensitivity to psychotropic drugs, sensorimotor gating abnormalities, and working memory deficits, which mimic aspects of schizophrenia. Schizophrenia is a debilitating psychiatric disorder affecting up to 1% of the general population, in which a dysfunction of the prefrontal cortex (PFC) is implicated. A negative feedback loop maintains normal synchronized activity in the pyramidal neurons of the PFC by activating a network of GABAergic interneurons, which in turn inhibit the pyramidal neurons (7). The NMDA receptors on the dendritic spines of these GABAergic interneurons are crucial transmitters of the signal from the pyramidal neurons (8). The cognitive deficits in schizophrenia, which include impaired working memory and difficulties in planning and executive control, are largely attributed to PFC dysfunction (9). A prominent candidate gene associated with increased risk for schizophrenia is the γ-secretase substrate neuregulin-1 (Nrg1) (10). Neuregulins are cell–cell signaling molecules that play essential roles in the development and functioning of the nervous system (11). Nrg1 signaling via its receptor ErbB4 regulates NMDA receptor function of GABAergic interneurons in the PFC (10, 12, 13), and this regulation is perturbed in schizophrenic patients (14). Nrg1 signaling also regulates GABA release by these interneurons (15) and modifies their levels of α7 nicotinic acetyl choline receptors (16). How Nrg1 is mechanistically involved in the pathogenesis of schizophrenia remains unclear. Membrane-bound full-length Nrg1 is cleaved by ADAM metalloproteases (17) or the β-secretase Bace (18), releasing the extracellular domain that acts as a ligand for ErbB4. However, Nrg1 might also act in a cell-autonomous manner as the remaining membrane-bound Nrg1 fragment is cleaved by γ-secretase, releasing an intracellular domain that might regulate gene transcription (19, 20). Nrg1 cleavage could therefore be an example of regulated intramembrane proteolysis (RIP) (21). Whether disturbances along this signaling cascade have relevance for schizophrenia pathogenesis is an attractive, but unexplored, hypothesis. An indication comes from the observation that the only schizophrenia-associated SNP identified in the Nrg1 ORF causes a Val-to-Leu mutation (22) in its transmembrane domain, raising the question of whether this mutation impairs Nrg1 cleavage by γ-secretase. In the current work we show that this Val-to-Leu substitution indeed affects γ-secretase cleavage of Nrg1 and that γ-secretase processing of Nrg1 is disturbed in Aph1BC^−/− mouse brain. We demonstrate that Aph1B/C expression is prominent in PFC and hippocampus, which have been implied prominently in schizophrenia pathogenesis. An extensive pharmaco-behavioural analysis of the Aph1BC^−/− mouse revealed several subtle phenotypes that overall can be considered as highly relevant for schizophrenia.

Results

Aph1B/C Expression Is Enriched in Brain Areas with Relevance for Schizophrenia.

Northern blot data indicated relatively (i.e., compared with Aph1A) strong expression of Aph1B/C in brain (3), but information about the distribution of Aph1B/C in specific areas is lacking. In situ hybridization histochemistry showed that Aph1B/C is widely and specifically expressed in neurons throughout adult mouse brain. Expression is particularly high in cerebral cortex (especially the prefrontal areas), hippocampus, olfactory bulb, and cerebellum. Aph1B/C expression levels are clearly lower in striatum (Fig. 1 a and b). Within the prefrontal cerebral cortex, expression is stronger in layers II, III, and V compared with layers 4 and 6 (Fig. 1 c–e), and within the hippocampus, Aph1B/C is enriched in CA1 compared with CA3 and dentate gyrus (Fig. 1 f–h). Layer 5 of cerebral cortex and hippocampal CA1 consist of glutamatergic “outflow” neurons that connect to other brain areas such as ventral striatum. Disturbance of Nrg1 signaling in the context of schizophrenia pathogenesis is thought to occur primarily in these corticofugal neurons, which signal to ErbB4 receptors on inhibitory interneurons in the PFC and the hippocampus (15).

Fig. 1.

Expression of Aph1B/C in neurons of the adult mouse brain. (a and b) Color-coded sagittal images of Aph1B/C mRNA in situ hybridizations on adult mouse brain using radioactively labeled antisense (a) and sense (b) probes. High expression is seen in hippocampus (HC), olfactory bulb (OB), PFC (PFC-CTX), and cerebellar cortex (CER). In contrast, expression levels are relatively low in striatum (STR). (c and e) Dark-field microscopy images of Aph1B/C mRNA in situ hybridizations of PFC using radioactively labeled antisense (c) and sense (e) probes. (d) Nonradioactive VGLUT1 labeling shows the density of glutamatergic neurons. Aph1B/C expression is enriched in cortical layers 5 and layers 2–3, is low in layers 4 and 6, and is absent in layer 1. (f and h) Dark-field microscopy images of Aph1B/C mRNA in situ hybridizations of hippocampus using radioactively labeled antisense (f) and sense (h) probes. (g) Nonradioactive VGLUT1 labeling shows the density of glutamatergic neurons. Aph1B/C expression is high in hippocampal area CA1 and almost absent in CA3 and dentate gyrus (DG).

Aph1BC^−/− Mice Have a Deficit in Sensorimotor Gating That Can Be Alleviated by Antipsychotic Drugs.

Aph1BC^−/− mice performed normally on tests for sensory and motor functions, basal activity levels and circadian rhythm, basal fear levels, spatial learning, and long-term memory. As we saw no differences in startle responses in pulse-alone trials [supporting information (SI) Fig. S1] and measured normal brainstem auditory evoked potentials (data not shown), we concluded that the auditory system of Aph1BC^−/− mice was generally intact. However, when we measured prepulse inhibition (PPI), a psychometric measure of sensory gating (23), in 3-month-old Aph1BC^−/− mice of the F₂ generation and compared with WT littermates, we found highly significant impairments with the 110-dB pulse stimulus (Fig. 2a; genotype effect: P < 0.001). For both 74-dB prepulse/110-dB pulse and 78-dB prepulse/110-dB pulse trial types, PPI in the Aph1BC^−/− mice was 70–75% of WT levels. For 100-dB pulse trial types, the effect was less outspoken, yet still significant (genotype effect: P = 0.029). To rule out crucial genetic background effects, we performed analogous PPI measurements after six backcrosses to C57BL/6J and found a PPI deficit of similar magnitude, although absolute levels of PPI differed because of the different genetic background (data not shown). We conclude that the PPI deficit is a robust and specific observation in Aph1BC^−/− mice.

Fig. 2.

Aph1BC^−/− mice have a sensorimotor gating deficit that is alleviated by antipsychotics. (a) PPI was measured in Aph1BC^−/− mice (n = 25, empty bars) and WT littermates (n = 24, filled bars), using different combinations of prepulse and startle pulse intensities (indicated on the x axis). PPI in Aph1BC^−/− mice was significantly lower than in WTs (genotype effect, P < 0.001) when a pulse of 110 db was used as startle stimulus [post hoc Tukey test, P = 0.005 for pp74, P = 0.008 for pp78 (**)]. (b and c) Treatment with clozapine (clz) or haloperidol (hal) (both 1 mg/kg, i.p) alleviated the PPI deficit in the knockout [compared with saline injected (ctr)], as shown for prepulse/pulse combinations pp74-p110 (b) and pp78-p110 (c) (treatment effect, P = 0.003 for pp74, P = 0.005 for pp78; genotype × treatment effect, P < 0.001 for both b and c). Post hoc Tukey tests, P = 0.009 and P = 0.007 (**) for the control group in b and c, respectively. P > 0.05 [not significant (n.s.)] for the clozapine and the haloperidol group in both b and c. n = 20 for both genotypes in b and c.

In human patients and animal models for schizophrenia, the PPI deficit can be attenuated by antipsychotics (24). We evaluated the effect of a typical and an atypical antipsychotic, haloperidol (1 mg/kg, i.p.) and clozapine (1 mg/kg, i.p.), respectively. Both drugs normalized PPI to WT levels (Fig. 2 b and c). Of note, PPI levels were significantly elevated in haloperidol-treated, but not clozapine-treated, WT animals. Conflicting results concerning the effects of different antipsychotics on PPI in WT rodents have been reported (24).

Hypersensitivity to Amphetamine Implies Disturbed Dopaminergic Signaling in the Ventral Striatum of Aph1BC^−/− Mice.

Haloperidol and clozapine share D2 receptor antagonism, which led us to further pharmacobehavioural evaluation of the dopaminergic state of Aph1BC^−/− mice. Systemic administration of amphetamine leads to a hyperlocomotor phenotype in humans and rodents. This effect is mediated by enhanced dopaminergic signaling in the ventral striatum (25). Baseline locomotor activity of Aph1BC^−/− mice was indistinguishable from WT littermates, as measured in standard open field and circadian rhythm assays (data not shown). Amphetamine treatment (3 mg/kg i.p.) elicited a locomotor response in WT and Aph1BC^−/− mice, but the degree of hyperactivity was significantly larger in the knockouts (Fig. 3a; genotype effect, P = 0,008). The effect was dose-dependent in WT and Aph1BC^−/− mice, with a hyperresponse in the knockouts (Fig. 3b; genotype effect, P = 0.004).

Fig. 3.

Abnormalities in the dopaminergic mesolimbic sytem of Aph1BC^−/− mice. (a) Mice were habituated to a novel cage and after 1 h they were injected i.p. with amphetamine or saline (arrow) and monitored for 2 more hours. Aph1BC WT, saline (diamonds, n = 20); Aph1BC^−/−, saline (squares, n = 19); Aph1BC WT, 3 mg/kg amphetamine i.p. (triangles, n = 20); Aph1BC^−/−, 3 mg/kg i.p. (rectangles, n = 19). Amphetamine elicits hyperactivity in WTs and knockouts (drug effect, P < 0.001), but the response to the drug is significantly stronger in Aph1BC^−/− mice (genotype effect, P = 0.008). (b) Activity of Aph1BC WT and knockout mice was summated (area under the curve) over 2 h after injection and normalized to total activity after saline injection. Amphetamine in three different doses (0.3, 1, and 3 mg/kg) causes a dose-dependent hyperactivity phenotype in WT and Aph1BC^−/− mice, but the phenotype is more severe in knockouts (genotype effect, P = 0.004). Post hoc Tukey tests show a significant difference both at 1 mg/kg (*, P = 0.038) and 3 mg/kg (**, P = 0.005). n = 20 for WT, n = 19 for Aph1BC^−/− mice. (c–e) Dopamine metabolites were measured in Aph1BC^−/− mice (empty bars, n = 9) and WT littermates (filled bars, n = 8). In the ventral striatum the dopamine catabolites, expressed as ng/mg wet tissue weight, DOPAC (c) (Student's t test; *, P = 0.018) and HVA (d) [Student's t test; P = 0.1 (borderline)] are increased. In the dorsal striatum, no significant differences are measured. (e) Dopamine turnover, expressed as the sum of the levels of HVA and DOPAC relative to total dopamine, is slightly enhanced in Aph1BC^−/− mice (Student's t test; *, P = 0.026).

Enhanced Dopamine Turnover in the Ventral Striatum of Aph1BC^−/− Mice.

This hypersensitivity to amphetamine at the behavioral level may indicate a greater presynaptic availability of dopamine in relevant brain areas, notably the ventral striatum (26). HPLC analysis of homogenized striatal tissue showed similar total dopamine levels in the ventral striatum of Aph1BC^−/− and WT littermates (data not shown). However, there was a subtle increase in 3,4-dihydroxy-phenylacetic acid (DOPAC) (32%, P = 0.022) and homovanillic acid (HVA) (13%, borderline significant, P = 0.1) (Fig. 3 c and d), indicating a modest enhancement of dopamine turnover in the knockout mice. This enhancement is more clearly demonstrated by the significantly higher (HVA+DOPAC)/dopamine ratio in Aph1BC^−/− mice (Fig. 3e) in the ventral striatum (P = 0.026) compared with WT mice. In the dorsal striatum neither the level of dopamine or its metabolites DOPAC and HVA differed between WTs and knockouts (Fig. 3 c–e). These changes occurred in the absence of any gross morphological change in the dopaminergic pathways from midbrain to striatum, as stereological analysis showed equal numbers of tyrosine hydroxylase-positive cells in both ventral tegmental area and substantia nigra, pars compacta of Aph1BC^−/− and WT mice (data not shown).

Impaired Working Memory in Aph1BC^−/− Mice Points to a Compromised Prefrontal Cortical Function.

The behavioral deficits pointed to aberrations in the functioning of the mesolimbic dopaminergic system. However, Aph1B/C expression in the striatum and midbrain is low, whereas it is very high in the PFC and the hippocampus (Fig. 1). A hyperdopaminergic state of the ventral striatum can be a consequence of a primary PFC dysfunction of glutamatergic signaling as has been suggested by several authors (27, 28). Working memory assays are often used as a behavioral read-out of PFC integrity, in humans, nonhuman primates, and rodents (29). We first examined spatial learning and long-term memory abilities by using the standard version of the Morris water maze task. During the acquisition trial blocks (Fig. 4a), Aph1BC^−/− mice and WT littermates learned the position of the hidden platform (trial block effect, P < 0.001) at a similar speed (no genotype effect). During the probe trial (Fig. 4b) at the end of the 2 training weeks, both groups displayed a clear preference of similar strength for the target quadrant (no genotype effect). Aph1BC^−/− mice and WT littermates were now acquainted with the demands of the water maze task and could be compared on the working memory version. To probe for working memory, the platform location was changed every time at the start of the daily trial block. A trial block consisted of five trials separated from each other by only 5 min, during which the platform location was the same. The first trial served as cue trial, whereas the shortening of the latency to find the platform over the consecutive trials served to measure the efficacy of working memory. Aph1BC^−/− mice needed marginally more time to find the platform when test trials 1–4 were averaged per day (genotype effect; P = 0.041) (Fig. 4c). We also measured the savings between the cue trial and test trial 1, respectively, test trial 3 (for which the distance to the platform was the same as in the cue trial) by subtracting individual latencies and averaging the group mean savings over the 6-day trial period. This way we isolated the working memory effect by eliminating variability caused by factors such as swimming speed or motivation. Aph1BC^−/− mice exhibited significantly less savings between the cue trial and test trial 3 (genotype effect, P = 0.003; Student's t test; P = 0.036), indicating inferior working memory capacity (Fig. 4d).

Fig. 4.

Impaired working memory and hypersensitivity to a noncompetitive NMDA receptor antagonist in Aph1BC^−/− mice. (a) Aph1BC^−/− mice (squares, n = 20) and WT littermates (diamonds, n = 20) were trained in the standard version of the Morris water maze task. Latency to find the hidden platform decreased gradually with number of training days, indicating acquisition of reference memory for the platform location. Learning curves were similar for both genotypes (no genotype effect). (b) After 2 weeks of training, Aph1BC^−/− mice (empty bars, n = 20) and WT littermates (filled bars, n = 20) were subjected to a probe trial. Both mice preferred the target quadrant (TQ), where the platform was located during the acquisition phase, above the opposite (OQ) and adjacent (AQ1 and AQ2) quadrants (no genotype effect). Preference was quantified as the total amount of time spent in a specific quadrant during a 100-s probe trial. (c) Subsequently, Aph1BC^−/− mice (squares, n = 20) and WT littermates (diamonds, n = 20) were assessed for working memory capacity. In daily trial blocks of five swims (cue trial and test trials 1–4), mice had to find a hidden platform that changed location daily. Aph1BC^−/− mice needed significantly more time to find the hidden platform expressed as mean latency of test trials 1–4 (genotype effect; P = 0.041). (d) Savings in escape latency between the cue trial and test trial 1 are similar (t test, P > 0.05) for Aph1BC^−/− mice (squares) and WT littermates (diamonds), but savings between test trial 1 and test trial 3 are smaller for Aph1BC^−/− mice (t test, P = 0.027). Data are given as means of latencies on the specified trials, averaged over 6 trial days, and are expressed for both genotypes as the percentage of the escape latency compared with the cue trial. (e and f) PPI was measured after i.p. injection with saline or different concentrations of MK-801 in Aph1BC^−/− mice (squares, n = 15) and WT littermates (diamonds, n = 16). Results are shown for two different prepulse/pulse combinations: pp74-p110 (e) and pp78-p110 (f). PPI under each drug condition is plotted as the percentage of PPI under placebo. For both trial types, Aph1BC^−/− mice are hypersensitive to MK-801 (genotype effect; P = 0.002 for pp74, P = 0.004). Post hoc Tukey tests reveal significant differences at the 0.5 mg/kg dose (**, P = 0.003 for pp74/p110 trials and *, P = 0.036 for pp78/p110 trials).

Hypersensitivity to MK-801 Confirms Dysregulated NMDA Receptor-Mediated Glutamatergic Signaling.

We investigated the sensitivity of the Aph1BC^−/− mice to a drug that influences glutamatergic signaling. Specifically, MK-801 is a noncompetitive NMDA receptor blocker, and administration of this drug to rodents and primates is a well established pharmacological model of schizophrenia (30). Because in rodents, NMDAR blockers impair PPI (31), we used this assay as a read-out for sensitivity to NMDA receptor blockade. Administration of MK-801 causes a dose-dependent decrease in PPI in WT and knockout mice (Fig. 4 e and f). However, Aph1BC^−/− mice were hypersensitive to the PPI-lowering effects of MK-801 for p110 trials (genotype effect; P = 0.002 and 0.004 for pp74 and pp78 trials, respectively). At a dose of 0,5 mg/kg, PPI is almost completely absent in the knockouts (post hoc Tukey test, P = 0.003 for pp74/p110 trials and P = 0.036 for pp78/p110 trials). Other prepulse/pulse combinations yielded similar results.

γ-Secretase-Dependent Cleavage of Nrg1 Is Impaired in Aph1BC^−/− Mouse Brain and by V321L Substitution.

Two known γ-secretase substrates, ErbB4 and Nrg1, have been implicated in schizophrenia pathogenesis. The levels of the ≈80-kDa ErbB4 C-terminal fragment (CTF), the direct γ-secretase substrate (32), were not altered in membrane-enriched brain extracts from Aph1BC^−/− mice (Fig. 5b). For Nrg1, however, a ≈50- to 55-kDa band that was detected with an antibody against the Nrg1 intracellular tail, accumulated in cortex, hippocampus, pons/medulla, and cerebellum but not in striatum (Fig. 5a). Its size was compatible with that of the expected Nrg1 cleavage product after release of the extracellular domain (11). Because of extensive alternative splicing of Nrg1, other bands are visible on the blot, complicating the interpretation of this experiment. Therefore we overexpressed a specific isoform (type I Nrg1βa) in COS1 cells. A band of similar size was seen that accumulated upon treatment with two different γ-secretase inhibitors (Fig. 5c). Classically, the bio-active Nrg1 fragment (i.e., the ErbB4 ligand) is thought to be produced by cleaving full-length Nrg1 in its ectodomain. For type I Nrg1, this cleavage step leads to the secretion of a shed Nrg1 fragment (33). We immunoprecipated shed Nrg1 from the medium of cultured cells by using an antibody against the HA tag that was inserted immediately C-terminally of the EGF-like domain (Fig. 5c). The amount of shed ectodomain was not affected by the γ-secretase inhibitors, confirming that deficiency of γ-secretase specifically causes accumulation of the remaining membrane-bound fragment.

Fig. 5.

Effect of V321L mutation and Aph1BC deficiency on Nrg1 cleavage by γ-secretase. (a and b) Western blot analysis of membrane fractions of mouse brain homogenates. (a) Antibody recognizing a C-terminal epitope in Nrg1 with “a”-type tail conformation, identifies a ≈50- to 55-kDa Nrg1-CTF that accumulates in specific brain areas of the knockout mice. (b) No differences in the levels of the 80-kDa ErbB4-CTF are seen. (c–e) Western blot analysis of COS1 cells transiently overexpressing Nrg1βa. (c) The anti-Nrg1 a-type tail antibody reveals ≈50- to 55-kDa Nrg1-CTF accumulation upon treatment with γ-secretase inhibitors DAPT and L-685,458. Full-length (FL) Nrg1 and secreted Nrg1 (SUP) are unaffected (CE, cell extract; SUP, supernatant). (d and e) The schizophrenia-associated V321L mutation in type III Nrg1βa (empty bar) leads to significantly elevated levels of the ≈50- to 55-kDa Nrg1-CTF compared with WT Nrg1βa (filled bar). **, P = 0.009; n = 4 for both genotypes.

DNA polymorphisms in the Nrg1 gene have been linked to increased schizophrenia susceptibility. Interestingly, the only SNP located in the Nrg1 coding region causes a Val-321 to Leu (V321L) substitution near the end of the transmembrane domain (numbering is based on type III Nrg1βa) (22). This mutation, when introduced in type III Nrg1βa, causes accumulation of the Nrg1 stub, indicating impaired intramembrane cleavage (Fig. 5 d and e).

Discussion

Using a combination of behavioral and pharmacological experimental paradigms we demonstrate here a specific function for the Aph1B/C γ-secretase complex in vivo. Although the phenotype is subtle, our findings suggest the possibility that dysfunction of RIP of Nrg1 and possibly other substrates contributes to the symptomatology of schizophrenia and related diseases.

PPI measures sensory gating, the process of filtering irrelevant sensory information through cortico-striato-thalamo-cortical pathways (34). PPI deficits are part of schizophrenic symptomatology and are thought to contribute to the cognitive fragmentation typical of psychotic episodes (23). However, PPI deficits are not specific for schizophrenia (23). Schizophrenic patients also suffer from cognitive deficits such as planning difficulties and impaired cognitive flexibility (9). Deficient working memory is used as a read-out for these cognitive symptoms in animal models (29). The Aph1BC^−/− mice indeed display disturbances in both PPI and a working memory task. The pharmacological analysis demonstrates both glutamatergic and dopaminergic alterations. One dominant theory, the “dopamine hypothesis” of schizophrenia, claims a hyperdopaminergic state of the mesolimbic system (35). The Aph1BC^−/− mice display amphetamine hypersensitivity in a locomotor test that depends on mesolimbic dopaminergic signaling (25), and their hyperdopaminergic state is confirmed by the increased ratio of dopamine catabolites to total dopamine in the ventral striatum, indicative of higher synaptic turnover of dopamine. According to the more recently developed “glutamatergic hypothesis” of schizophrenia (27), the hyperdopaminergic state of the ventral striatum is caused by inappropriate activation of corticofugal glutamatergic tracts, originating in the PFC and the hippocampus (28). These outflow tracts connect to the mesolimbic pathway and stimulate dopamine release. The overactivation of glutamatergic neurons in these areas is caused by insufficient activity in a network of GABAergic neurons that inhibit the glutamatergic neurons (7). The working memory impairment and the hypersensitivity of the mice to the noncompetitive NMDA receptor blocker MK-801 point to malfunctioning of this brain circuit. Indeed, NMDA receptors are crucial regulators of the inhibiting GABAergic network in the PFC (8). Several phenotypic characteristics of these mice, e.g., the PPI disturbances, are also found in the APO-SUS rat (36). This rat line has a genomic recombination in the Aph1B/C gene locus (37) and provides an independent line of evidence for the crucial role that Aph1B-γ-secretase plays in normal brain function.

Recent work (J.V.B., L.S., and B.D.S., unpublished work) shows dramatic improvement caused by inactivation of Aph1B/C in an Alzheimer's disease mouse model, while severe, Notch-related side effects like gastrointestinal bleedings, neoplasia, and immunodeficiencies that are characteristic for general γ-secretase inhibition are not observed. The results presented here add a note of caution to specific Aph1B-directed therapy to block Aβ production, which will have to be monitored carefully. It should, however, be stressed that the phenotype we see here is associated with a full deficiency of Aph1B/C. Because a wealth of data supports the contention that the development of schizophrenia depends on a neuronal insult early in life (38), it is possible that suppression of Aph1B/C activity in the brain in adulthood only will have no severe side effects at all. The notion of selective inhibition of specific γ-secretase complexes is, however, only theoretical at this point and needs to be investigated further.

We focused on the processing of Nrg1 and its receptor ErbB4 because the link between Nrg1, cognition, and the pathogenesis of schizophrenia has been made (12–15). We cannot exclude the contribution of deficient processing of other substrates such as N-cadherin (data not shown) or amyloid precursor protein (6) to the phenotype. However, there is certainly specificity in Aph1B/C function as ErbB4 or Syndecan (data not shown) cleavage, and Notch signaling during development (6) and adulthood (J.V.B., L.S., and B.D.S., unpublished work), are not affected in our mice. In this context, we consider it highly significant that the only schizophrenia-associated SNP in the Nrg1 ORF results in decreased γ-secretase cleavage of Nrg1.

The dramatic effects on Nrg1 processing and the absence of effects on ErbB4, combined with the behavioral and pharmacological changes, are most easily interpreted in the context of a Nrg1 loss-of-function hypothesis for schizophrenia (10, 12). Indeed, the phenotypes of Nrg1^+/− mice have been explained as a consequence of reduced Nrg1-to-ErbB4 signaling (39). Recently, schizophrenia-like symptoms were observed in Bace1^−/− mice (40). Bace1 cleavage releases the extracellular domain of Nrg1, thus promoting Nrg1-to-ErbB4 signaling (18). Our results suggest alternative interpretations of these data. Indeed, deficiency both in Nrg1 expression by genetic inactivation or Nrg1 ectodomain shedding by Bace1 deficiency will result in decreased production of the Nrg1 membrane-bound fragment. This fragment is the direct substrate for Aph1B/C γ-secretase. The Nrg1 intracellular domain that subsequently becomes released by γ-secretase has been implicated in gene transcription regulation (19, 20). Thus, we suggest that cell-autonomous effects contribute to the Aph1BC, Bace1, and Nrg1 knockout phenotypes, and it is tempting to speculate that disruption of this so-called “back signaling” could lead to schizophrenia-like behaviors in vivo.

In conclusion, our work sets the stage for further exploration of the finely tuned functionality of the γ-secretase family of proteases. It is clear that the structural heterogeneity of this complex is reflected in subtle functional differences and that the current work has led to an additional perspective on the potential role of dysregulation of RIP in psychiatric disease.

Materials and Methods

For more detailed information, see SI Text.

Cell Culture Experiments.

See SI Text for details. COS1 cells were cultured, transiently transfected, and lysed by using standard techniques. Rat type I Nrg1βa (GenBank accession no. AF194993) was cloned in pSecTagA (Invitrogen) with an HA epitope inserted at the C terminus of the EGF-like domain. Rat type III Nrg1βa (GenBank accession no. AF194438) was cloned in pcDNA4A (Invitrogen) with an HA epitope inserted in front of the Ig-like domain. The Val-to-Leu mutation was made by using the QuikChange II site-directed mutagenesis kit (Stratagene).

Brain Homogenates.

Brain regions of 3-month-old Aph1BC^−/− mice and WT littermates were homogenized in 1% Triton sucrose buffer. Membrane fractions were prepared by ultracentrifugation and resolubilization in 0.1 M phosphate buffer (pH 5.7). Equal amounts of protein were loaded and Western blotting was performed as described (6).

Antibodies.

The antibodies used were anti-Nrg1α/β1/2 (C-20), sc-348, anti-Nrg1α/β1/2 (F-20), sc-537, anti-ErbB4 (C-18), sc-283 (Santa Cruz Biotechnology) and anti-HA HA.11 mAb (Covance).

Aph1BC^−/− Mice.

Aph1BC^−/− mice have been described (6). All behavioral experiments were done by using littermates of the F₂ generation of the C57BL/6J–129/Ola hybrids. Biochemical and behavioral experiments were approved by the Ethical Committee on Animal Experimenting of Katholieke Universiteit Leuven.

In Situ Hybridization.

Radioactive probes were generated by run-off transcription by using full-length mouse Aph1B cDNA cloned into pGEM-T. For details, see SI Text.

PPI.

Acoustic startle responses (ASRs) were recorded by using computerized startle response apparatus and software (Med Associates). The ASR was defined as the amplitude of the first peak in the ballistogram (arbitrary units). PPI was defined as follows (in %): PPI = 100 − ((ASR prepulse + startle pulse)/ASR startle pulse alone) × 100. Whenever drugs were injected, a wash-out period of 3 weeks was included between each injection to avoid carryover effects.

Drugs.

Drugs were purchased from Sigma–Aldrich and injected i.p. at a volume of 10 ml/kg. Haloperidol was dissolved in 0.9% saline/0.1% lactic acid. Clozapine, d-amphetamine, and MK-801 were dissolved in 0.9% saline. γ-Secretase inhibitors DAPT and L-685,458 (Merck) were dissolved in DMSO.

Measurement of Locomotor Activity.

Mice were placed individually in 26.7 cm × 20.7 cm transparent cages (floor area 370 cm²) between three IR photo beams. Beam crossings reflect locomotor “cage” activity and were counted in 5-min bins during a 3-h recording period, using an interfaced computer. After 1 h, mice were injected i.p. with d-amphetamine and immediately placed back to analyze the effect of the compound.

Working Memory Version of the Morris Water Maze.

For details, see SI Text. Mice were first trained on the standard hidden-platform Morris water maze and subsequently tested in the working memory version.

Measurement of Striatal Dopamine and Its Metabolites.

Concentrations of dopamine, DOPAC, and HVA were determined by reversed-phase ion-pair HPLC using homgenized ventral or dorsal striatal tissue. For details, see SI Text.

Statistics.

Effects of genotype and drug treatment in behavioral experiments were statistically evaluated by using two-way repeated measures ANOVA and post hoc Tukey tests. Differences between genotypes in the other experiments were tested for statistical significance with the Student's t test.