Gain-of-function glutamate receptor interacting protein 1 variants alter GluA2 recycling and surface distribution in patients with autism

We identified an association (P = 0.048) of a common SNP (rs7397862) in intron 15 of GRIP1 within the region encoding PDZ4–6 (exons 12–16) in a case-control study of autism (n = 480) and ethnically matched normal controls (n = 480) (Tables S1 and S2). We built on this initial finding to perform parallel sequencing of all 25 exons and ~50-bp flanking introns of GRIP1 in both cohorts. This analysis identified five missense variants involving highly conserved amino acid residues within or near PDZ4–6 only in autistic patients (Fig. 1A) (21). All five are transmitting variants and each was found in a single family (Table S3). The cumulative allele frequency of the five variants was higher (P = 0.032) in autism compared with ethnically matched controls (Table S2). Additionally, one missense variant, V54I in PDZ1, was found in two patients and two missense variants, Q821E and R869Q near PDZ7, were found in both autism and controls (Fig. 1A). Noncoding variants of interest within the immediate flanking intronic regions of exons 12 to 18 encoding PDZ4–6 include a single base-substitution (T > C) at position −3 of intron 13 in one patient, a three base (TTG) deletion at positions −10 to −12 of intron 11 in three patients and one control, and two common SNPs at positions +13 (rs7397862) and +34 (rs7397861) of intron 15 in both autism patients and controls (Fig. S1).

In particular, two variants, A625T and M794R, were found in families with two affected brothers who had different genotypes for respected GRIP1 variants. This finding permitted a correlation study of the presence of these variants with the severity of autism phenotype, as defined primarily by Autism Diagnostic Interview-Revised (ADI-R) scores (Table 1). In the first family, the affected boy, who was heterozygous for A625T, had relatively (with respect to other components of these instruments) more impairment in ADI-R's Reciprocal Social Interaction, the Autism Diagnostic Observation Schedule's Reciprocal Social Interaction domain, and Vineland Adaptive Behavior Scales’ Socialization scale, compared with his affected brother, who did not carry this variant. In the second family, the boy who was homozygous for M794R also had relatively worse scores on ADI-R's Reciprocal Social Interaction domain, compared with his affected brother, who was heterozygous for this variant. Interestingly, their mother, who was also heterozygous for M794R but not diagnosed with autism, had a history of restricted interests, repetitive behavior, poor eye contact, and eccentricity. The two aforementioned socially more-impaired subjects also had overall higher ADI-R scores and lower cognitive—but not motor—skills than their brothers.

The five GRIP1 variants (I549V, T550M, I586L, A625T, M794R), clustered around PDZ4–6 and found only in autism subjects, were further studied because these domains are known to directly interact with the C terminus of GluA2/3. The amino acid residues I549 and T550 within PDZ4, and I586 and A625 within PDZ5, are identical in a nine-species alignment, and the aliphatic/nonpolar nature of the amino acid at M794 is also conserved (Fig. 1A). Mapping the residues to a published NMR structure of PDZ4/5 (21) revealed that I549 and T550 are located within a six-amino acid motif (DSSITS) of the αB-βF loop in PDZ4, which is essential for stabilizing the PDZ4/5 supermodule. Residue I586 is part of the four-amino acid motif (ELGI) that directly binds the C terminus of GluA2/3, and A625 is a highly conserved residue within a βB loop in PDZ5 (Fig. 1B). We hypothesized that these GRIP1 variants would display altered interactions with GluA2/3 and would affect GluA2/3 function and distribution in neurons.

To address the first possibility, we determined the in vitro interaction of the five variants with C-terminal domains of GluA2/3 (GluA2C and GluA3C) using a yeast two-hybrid assay. Deletion of the C-terminal four amino acids of GluA2/3 (R3-CΔ4) or artificial mutations, K577A_K578A, of GRIP1 are known to completely abolish GRIP1-GluA2/3 interaction. As shown in Fig. 2, we found a consistent increase in the interaction of three variants, I549V, I550M, and I586L with GluA2C, as reflected by β-galactosidase activity (WT, 100.0%; I549V, 120.5 ± 4.2%, P = 0.023; I550M, 119.3 ± 7.1, P = 0.019; I586L, 152.0 ± 5.3%, P < 0.001; A625T, 80.2 ± 5.9%, P = 0.006; M794R, 110.3 ± 2.0%; K577A_K578A, 0.0 ± 0.0%; n = 6 yeast clones) and GluA3C (WT 100.0%; I549V, 114.6 ± 5.7%; I550M, 114.0 ± 4.0; I586L, 133.7 ± 11.0% P = 0,001; A625T, 71.5 ± 2.6% P = 0,01; M794R, 91.7 ± 3.2%; K577A_K578A, 0.0 ± 0.0%; R3-CΔ4, 0.0 ± 0.0%; n = 6). The other two variants, A625T and M794R, showed moderately reduced interaction. Interestingly, I586L showed the most significant increase in the interaction, which is consistent with the NMR structure of PDZ4/5, in which I586 directly binds GluA2C and GluA3C (21).

GRIP1 was recently reported to modulate the activity-dependent trafficking of a pH-sensitive pHluorin-GluA2 fusion protein (pH-GluA2) (22). We therefore examined the effects of altered GRIP1-GluA2/3 interaction of three GRIP1 variants (T550M, I586L, A625T) on pH-GluA2 internalization and recycling in live, transfected hippocampal neurons (23). Expression of both WT and GRIP1 variants in neurons mimics the heterozygous state of these variants in individuals with autism. Compared with WT GRIP1, all three variants showed comparable amplitude of fluorescence changes, reflecting the rate of pHluorin-GluA2 internalization following NMDA stimulation (Fig. 3 A, B, and D) (amplitude, WT 0.70 ± 0.03; T550M 0.67 ± 0.03; I586L, 0.72 ± 0.02; A625T, 0.63 ± 0.03) but a faster recycling of pHluorin-GluA2 as reflected by shorter T1/2, the time for recovery of 50% of fluorescence. The differences in two variants, I586 and T550M, reached statistical significance (Fig. 3 A–C) [t_1/2(min), WT, 6.66 ± 0.64; T550M, 5.37 ± 0.64, P = 0.174; I586L, 3.97 ± 0.38, P = 0.003; A625T, 3.97 ± 0.83, P = 0.026; n = 6–13 neurons]. Consistent with the yeast two-hybrid data, I586L showed the most significant effect on GluA2 recycling. Because Grip1/2-deficient hippocampal neurons display slower GluA2 recycling in the same assay (22), these results suggest a gain of GRIP1 function for these variants on GluA2 recycling.

Finally, we determined the relative abundance of surface versus total GluA2 in hippocampal neurons transfected with individual GRIP1 variants. Among the five variants, three (I586L, A625T, M794R) increased the relative abundance of surface GluA2 (Fig. 4). The differences in two variants, I586L and A625T, reached statistical significance compared with WT GRIP1 (surface/total GluA2 (%), WT, 100.0; I549V, 100.6 ± 9.9; T550M, 106.0 ± 8.7; I586L, 163.9 ± 22.7, P = 0.006; A625T, 141.9 ± 10.8, P < 0.001; M794R, 123.3 ± 14.5; n = 11–22 neurons). These data are consistent with that from the pHluorin-GluA2 recycling studies (Fig. 3).

GRIP1 and GRIP2 are highly homologous proteins that are known to complement each other for certain neuronal functions (16). To understand the role of GRIP1/2 in modulating social behaviors, we studied the loss-of-function phenotype of Grip1/2 double-knockout (DKO) mice. DKO mice were generated by crossing Grip2 conventional KO mice with conditional Grip1 (neuron-specific deletion via Nestin-Cre expression) KO mice on C57BL6/129 strain background (17) to circumvent the embryonic lethality of conventional Grip1 KO mice (24), and Grip1 deletion was then achieved by crossing with Nestin-Cre mice. Cre-dependent Grip1 deletion was verified in brain lysates by Western blot analysis. Control mice (WT) were matched for age, sex, and strain background with DKO mice. A standard battery of behavioral tests, including an open-field, elevated plus maze, Y-maze, intruder-resident test, novel object recognition, sociability toward stranger mice, and prepulse inhibition (PPI) were used. Compared with WT, Grip1/2 DKO mice showed a selective, significant increase in sociability toward stranger mice (empty cage in seconds: WT, 98 ± 9; DKO, 112 ± 11; with stranger mice in seconds: WT, 116 ± 10; DKO, 192 ± 23, P = 0.012; n = 10 mice), impaired PPI (percent of inhibition; WT, 54 ± 25 and DKO, 84 ± 18 at p74; WT, 27 ± 8 and DKO, 79 ± 18 at p78; P = 0.012; WT, 21 ± 5 and DKO, 59 ± 17 at p82; P = 0.035; WT, 16 ± 3 and DKO, 50 ± 15 at p86; P = 0.034; WT, 17 ± 4 and DKO, 41 ± 11 at p90, P = 0.041; n = 10 mice), and impairment in recognizing a novel object (time spent interacting with object in seconds: WT with familiar object, 7.1 ± 1.2; DKO with familiar object, 12.0 ± 1.9, P = 0.043; WT with novel object, 12.0 ± 2.4; DKO with novel object 12.8 ± 2.0) (Fig. 5). Lack of difference for other behavioral tests is depicted in Fig. S2. These results support a role for GRIP1/2 proteins in the modulation of social behaviors in mammals.

We extended a previous linkage to chromosome 12q14 by identifying an association of a common SNP within the genomic region encoding PDZ4–6 of GRIP1 with autism. Sequencing of GRIP1 in autism and matched control cohorts identified five rare missense variants located within or near PDZ4–6 only in the autism cohort. Importantly, these variants showed a higher cumulative mutation load in autism and were found to alter GRIP1 interaction with GluA2/3 in in vitro and neuron-based assays. A positive correlation of two variants with greater autistic behavioral severity, particularly of the core social-interaction domain, as measured by ADI-R, was found in families with affected sibling pairs. More severe genotypes in these families were also linked to greater cognitive impairment. Although three of the four parents in these two families are heterozygous for the respective variants, none of them was diagnosed with autism. Limited behavioral data were available only for the mother who was heterozygous for M794R and was noted having a history of restricted interests, repetitive behavior, poor eye contact, and eccentricity. Together, these results suggest that GRIP1 variants may modify the severity of the deficits in social interactions and cognitive function in autistic patients, but are not sufficient by themselves to cause autism.

Because most patients carrying GRIP1 variants in this study are heterozygotes, we speculate that these variants affect neuronal and behavioral phenotype through a dominant mechanism. Consistent with this hypothesis, we found that three variants (I549V, T550M, and I586L) increased interaction with GluA2, two variants (I586L and A625T) are associated with a faster recycling of GluA2, and three variants (I586L, A625T, and M794R) increased surface GluA2 in hippocampal neurons. Because loss of GRIP1/2 function was associated with reduced interaction and slower GluA2 recycling (22), our data support a gain of GRIP1 function in these variants. Although disruption of GluA2-GRIP interactions in neurons also alters synaptic function and plasticity (19, 25), the effect of GRIP1 variants on these readouts remains to be elucidated.

Among the five variants, I586L most strongly affected GRIP1 interaction with GluA2/3 in a yeast two-hybrid assay, GluA2 recycling, and surface distribution. This finding is consistent with the NMR structure of PDZ4/5, in which I586 is one of the four amino acid residues that bind directly to GluA2/3 (Fig. 1B). We recognize that not all variants exhibited the same profile of changes in different assay systems. This finding suggests that additional factors may be involved, which may not be satisfactorily explained based on a simple model of one-on-one interaction between GRIP1 and GluA2/3. These contributing factors may include the following: (i) PDZ4 and -5 form a protein supermodule that interacts with GluA2/3, but individual PDZ domains play different roles: I549 and T550 are located in PDZ4, which stabilizes the supermodule; I586L and A625T are located in PDZ5, which mediates the direct interaction of GRIP1 with GluA2/3; (ii) multimerizations of GRIP1 with GRIP1, and GRIP1 with GRIP2 likely affect GRIP function in neurons; (iii) PDZ4--6 are known to interact with several other proteins in the glutamatergic and GABAergic signaling pathways. It is possible that some effects of GRIP1 variants are exerted via interactions with one or more of these proteins. Studies of neurons from knock-in mice carrying specific Grip1 variants will be needed to further understand the underlying neuronal mechanisms.

The behavioral studies of Grip1/2 DKO mice also suggest a role for GRIP proteins in the modulation of social behavior and possibly cognitive or memory function. Grip1/2 DKO mice exhibited increased sociability, impaired PPI, and object recognition. Deficits in sociability are one of the core features of autism. Impaired PPI was observed in a small group of adult patients with autism (26) and in mice deficient in neurexin-1α, a gene associated with autism (27). Although we recognize that human autistic phenotypes consist of complex behaviors that may not be fully replicated or interpreted in mouse models, studies of knock-in mice carrying GRIP1 variants may help to elucidate how GRIP1 regulates social behavior and cognitive and memory function.

Identification of rare functional variants is important in identifying molecular pathways that are disturbed in autism. GRIP1 binds key proteins that regulate glutamatergic signaling, including liprin-α (28), PICK1 (29), KIF5 (30), and NEEP21 (31), and also uses the same PDZ4–6 to bind the GABA receptor-associated protein, gephyrin (32) (Fig. 1A). We speculate that functional GRIP1 variants in PDZ4–6 may alter not only glutamatergic but also GABAergic synaptic function, leading to an imbalance in selected neuronal circuits in autism (33). A role for synaptic imbalance in autism is supported by findings in mice carrying an autism-associated mutation (i.e., R451C) (5) in Neuroligin-3. These mice were found to have increased inhibitory synaptic markers, gephyrin, and vGABA transporter, and in the size of inhibitory synaptic responses and spontaneous inhibitory event frequency, suggesting an enhanced inhibitory synaptic transmission affecting a subset of inhibitory interneurons in forebrain through a gain of function mechanism (34).

We thank Dr. Paul Worley for the Nestin-CRE transgenic mice, Jennifer Yocum for help with animal behavioral testing, Yilin Yu for help with animal genotyping, and Eva Andres-Mateos for help with the figures; and the resources provided by the South Carolina Autism Project and Autism Genetic Resource Exchange (AGRE) Consortium and the participating families. This study was supported in part by research Grant #2487 from the Autism Speaks Foundation and Grant R01HD052680 from the National Institute of Child Health and Human Development (to T.W.); a postdoctoral fellowship from the Ministry of Education and Science of Spain (to R.M.); and fellowships from the International Human Frontiers Science Program (LT00399/2008-L) and Australian National Health and Medical Research Council (ID477108) (to V.A.). R.L.H. is an investigator with the Howard Hughes Medical Institute. AGRE is a program of Autism Speaks and is supported, in part, by Grant 1U24MH081810 from the National Institute of Mental Health to Clara M. Lajonchere (PI).