Patients with schizophrenia and subjects with mental retardation were ascertained at Rouen Hospital from consecutive hospitalizations in patients with schizophrenia, or from consecutive referral for phenotypic and genetic investigations in subjects with intellectual disability. The ASD sample included patients ascertained from consecutive consultations in four units specialized in autism diagnosis and evaluation located at Rouen, Dijon, Tours (France) and Messina (Italy) and patients directly referred by the French Autism Foundation. Controls, all ascertained at Rouen, were screened with a standardized data sheet derived from the SADS-LA 13 and were required to be free of any psychotic disorder or MR in themselves or in their first degree relatives. All psychiatric diagnoses were established according to DSM4 criteria following review of case notes and direct examination of subjects. The schedule for affective disorder and schizophrenia (SADS-LA) 13 was used for the clinical assessment of all patients with schizophrenia or schizoaffective disorder, the Autism Diagnostic Interview-Revised (ADI-R) 14, the Autism Diagnostic Observation Schedule-general (ADOS-G) 15 or the Childhood Autism Rating Scale (CARS)16 were used for 83% of ASD patients (100% of ASD patients with CNVs). IQs were evaluated using standardized neuropsychological tests i.e. validated mental age-appropriate Weschler scales (WPPSI, WISC or WAIS).

The schizophrenia group included 189 subjects with a diagnosis of schizophrenia and 47 with a diagnosis of schizoaffective disorder. Post morbid IQs were available for 2/3 of patients with schizophrenia; 18% of these subjects had an IQ< 70. The ASD group included 257 patients with autism and 3 with Asperger syndrome. The MR group included 12 patients with developmental language disorder. All MR patients and 2/3 of ASD patients were examined by an experienced clinical geneticist and were also screened for fragile X mutation and karyotype abnormalities. Subjects with large chromosomal anomalies, fragile X syndrome or other established syndromes were not included. Subjects with common environmental etiologies of MR such as fetal alcohol syndrome or birth complications were also excluded. Additional clinical features including intrauterine or postnatal growth retardation, dysmorphic features or malformations were present in 8.5 % of ASD patients and 62 % of subjects with MR. Demographic characteristics of the sample, including a total of 979 Caucasian subjects from France or Italy, are summarized in Table 1.

Blood samples were drawn, after written informed consent, from all included subjects and whenever possible from parents and affected relatives of patients. Ethics committee approval was obtained from all regions where families were recruited.

A medline search with the words CNV, schizophrenia, autism and mental retardation allowed us to select non exhaustively a set of 28 loci with microrearrangements characterized by prior array-CGH analyses, often in a single patient. This set included major candidate CNV loci identified in ASD and SZ before April 2008 as well as 8 functionally related CNV loci identified in MR (Table 2). Each locus contained generally a single disease-associated CNV, but in some cases overlapping CNVs with different boundaries had been described in patients. The gene content of these loci ranged from 1 to 28. At each locus at least one candidate gene had been previously suggested in the seminal reports and was retained for the present analysis. Functionally, most of these candidate genes can be classified in two main categories related to synapse formation and maintenance or neurotransmission.

Copy number variation at each locus was assessed by quantitative multiplex PCR of short fluorescent fragments (QMPSF), a method based on the simultaneous amplification of several short genomic fragments under quantitative conditions 17. For each locus, amplicons were designed in the coding sequence of selected candidate genes. All assays were grouped in three multiplex PCR experiments including 10 short genomic fragments (range 101–301bp) each. Primer sequences and PCR conditions are summarized in supplementary Table S1. DNA fragments generated by QMPSF were separated on an ABI Prism 3100 sequencer (Applied Biosystem) and the resulting fluorescence profiles were analyzed using the Gene Scan 3.7 Software (PE Applied Biosystems). For each patient, the QMPSF profile was superimposed to that generated from a reference subject by adjusting to the same level the peak obtained for a control amplicon corresponding to a short exonic fragment of the PBGD gene. When a copy number variation was detected, further analyses aiming to confirm and delineate the size of the rearrangements were performed using additional dedicated QMPSF assays (supplementary figure S1), array-CGH and/or fluorescence in situ hybridization (FISH) analyses.

Oligonucleotide array-CGH was performed using the Agilent Human Genome CGH microarray 4x44K (Agilent Technologies, Santa Clara, CA, USA). This array contains 44 290 60-mer oligonucleotide probes covering the whole genome with an average spatial resolution of ≈ 30–35 kb. 84% of the probes reside in intragenic regions, and over 30 000 genes are each represented by at least one probe. All experiments were performed according to version 4.0 (June 2006) of the protocol provided by Agilent (“Agilent Oligonucleotide Array-Based CGH for Genomic DNA Analysis”).

FISH analyses were performed on metaphase spreads obtained from peripheral lymphocytes from the patients. Selected Human genomic BAC clones were obtained from the BACPAC resources center (http://bacpac.chori.org).

Sequence analysis of the coding exons of the PRODH gene was performed using primers and PCR conditions previously described 18 on an Applied Biosystem model 3100 automated sequencer (PE Applied Biosystem). Paternity was checked by microsatellite typing.

In patients, plasma proline levels were determined after overnight fasting. All samples were analyzed using ion exchange chromatography on a BIOTRONIK LC 3000 system.

Categorical variables were compared by Fisher’s exact test. Two hypotheses were tested. First, the distribution of the collective set of recurrent/overlapping CNVs found in each group of patients was compared with that found in controls (3 tests). Second, the distribution of each recurrent/overlapping CNV present in our population was compared between each disease group and controls (33 tests). P-values were reported without Bonferroni’s correction.

Among the 28 loci selected for this study (Table 2), only one CNV identical to a previously described disease-associated one (i.e. a 22q11 350kb deletion spanning the PRODH and DGCR6 genes)19 was detected in the control group. This deletion, present in a single control subject, had a low frequency (1/236)similar to that previously reported in Japanese 20 and Canadian 21 populations. A common CNV reciprocal to the expected one (i.e. a 350 kb duplication) was also found at the same locus in 6/236 controls. This CNV, also present in 7/236 patients with schizophrenia, 4/260 ASD patients and 9/247 MR patients was clearly a benign polymorphism. Another overlapping reciprocal CNV (i.e. a 490 kb duplication) was detected at the CHRNA7 locus located on chromosome 15q13.3. This CNV present in 2/236 controls, 1/236 patients with schizophrenia, 1/260 ASD patients and 1/247 MR was equally unrelated to any pathological condition. (Supplementary Table S2)

The proportion of recurrent/overlapping CNVs identified among the 28 selected loci in the present sample totaling 743 affected subjects was 11/28 (40 %). Their collective frequency was 10/236 (4 %) in patients with schizophrenia, 16/260 (6 %) in ASD patients and 13/247 (5.3 %) in MR patients vs 1/236 (0.4 %) in controls, therefore demonstrating a significant excess of these CNVs in each disease group as compared with controls (p= 0.005, p< 0.001 and p= 0.001 respectively, Fisher exact test) (Table 3a). Individual significance for association with ASD was reached for the PRODH/DGCR6 deletion (9/260 in ASD patients vs 1/236 in controls, p= 0.02). None of the cases had more than one of these 28 CNVs. Among the four most prevalent CNVs, three, located in 22q11, 16p11 and 15q13, were flanked by known regions of segmental duplication and resulted most likely from a non allelic homologous recombination mechanism (NAHR). At the 2p16 locus, the NRXN1 gene was recurrently disrupted by a set of partially overlapping deletions spanning either the promoter and first exons of neurexin1 alpha or the exons coding for the middle section of this protein as well as for the proximal region of neurexin1 beta. These rearrangements occurred in a region devoid of any segmental duplication and resulted from another mechanism distinct from NAHR.

Among 27 families in which transmission was tested (69% of the total sample), it was found that only eight CNVs (located in 8p23, 15q11-q13, 15q13, 16p11 and 22q13) had occurred de novo (Table 3a). Paternal age was not significantly different between families with de novo and inherited CNVs (27.2 ± 4.7 y vs 30.7 ± 4.6 y, NS, Mann Whitney U test). In most families, CNVs were transmitted from an apparently unaffected (although not clinically nor neuropsychologically assessed) parent. This includes a partial duplication of the X linked GRIA3 gene, present in a young male autistic patient, which was inherited from the unaffected mother. The 22q11 350 kb deletion spanning the PRODH/DGCR6 locus was also transmitted in 11/11 tested cases. PRODH encodes for the proline dehydrogenase and PRODH deficiency is responsible for type 1 hyperprolinemia, a condition often associated with both cognitive impairment and psychotic symptoms 18. However, hemizygous deletion of the PRODH gene is not sufficient per se to result in hyperprolinemia since only 35–50 % of Velo-cardio-facial syndrome (VCFS) patients, all bearing a single copy of PRODH, exhibit hyperprolinemia 18, 22. Indeed, a reduction of more than 50% of the enzymatic activity is generally required to produce hyperprolinemia 18. The presence of a mutation affecting enzyme activity 23 on the second allele is therefore necessary. To examine this issue, the remaining PRODH allele was sequenced in all affected subjects bearing the 350 kb deletion, and, when possible, plasma proline level was assessed. As shown in Table 4, 14/15 patients harbored a genotype predicted to result in a reduction of at least 70% of enzymatic activity. Among the 10 subjects in whom plasma proline level was assessed, 7 had mild to severe hyperprolinemia and 3 had plasma proline level at the upper boundary of normal values.

Cosegregation of the CNV with pathological conditions was examined in 4 multiplex families in which DNA from affected siblings was available. In family 144, the two sibs with schizophrenia both harbored the PRODH/DGCR6 deletion. In family 11695, the two MR sibs both harbored the 2p16.3 deletion. In family 14390, the 16p11 deletion was present in the proband with developmental language disorder but not in his mentally retarded sib. In family 33, the 16p11 duplication was present in two sibs with schizophrenia as well as in a non affected sibling, but not in a third sibling with schizophrenia.

Combining the results of the present study and those of previous reports, we found that, with the possible exception of the maternally derived 15q13 duplication associated with ASD, none of the observed rearrangement was disease-specific. The 22q11 19 and 2p16 1 deletions were found in the three conditions, whereas the 22q13 deletion found in two ASD patients had already been described in ASD and MR 4, 24–26. The 16p11 and 8p23 rearrangements previously described in ASD 4, 5, were both found in SZ and also in MR patients for the 16p11 rearrangements. The Xp11.4 duplication spanning the TSPAN7 gene, previously described in MR and ASD 4, 27, was found in a patient with SZ, as well as the 17q21 duplication previously described in a patient with MR 28. Two different sized 15q13 duplications encompassing APBA2 were found in one ASD and one SZ patient. A partial duplication of the GRIA3 gene, encompassing the promoter region and the exons coding for the proximal region of GRIA3 was detected in a single autistic patient. Although slightly different by its size and by the number of duplicated exons, this partial duplication is reminiscent from that recently reported in a patient with MR 29. At the 8p23 locus, both gain and loss of material were found, as well as at the 16p11 and 17q21 loci, as recently described 28, 30, 31. This suggests that dosage-sensitive genes, whose expression is finely tuned, are located within these rearranged segments.

From a phenotypic point of view, it is noteworthy that 3 out of 9 patients with schizophrenia bearing a candidate CNV had mild MR in an IQ assessment obtained after the onset of their psychotic symptoms (Table 3b). Although postmorbid IQ is likely to constitute an underestimate of the premorbid level of cognitive functioning in patients with schizophrenia, cognitive deficits testified by severe learning disorder were already noted in these 3 patients during childhood before the onset of their psychotic symptoms, thus supporting the diagnosis of comorbidity between MR and SZ. With the exception of two subjects who had normal cognitive functioning (high-functioning autism), all tested CNV bearing autistic patients had IQs falling in the range of mental retardation, albeit one of them had only mild cognitive dysfunction (Table 3b).

After a first wave of CNV discovery by array-CGH analyses in neuropsychiatric disorders, this study for the first time examines the involvement of a limited number of candidate loci in large samples of patients with different clinical diagnoses. Two strengths of our study design are the inclusion of controls carefully screened for the studied pathologies and of series of patients mostly ascertained through consecutive admissions or consultations and therefore including subjects belonging either to simplex or multiplex families. Given the expected rarity of each variant, our first goal was not to test the association of every individual CNV with SZ, ASD or MR but to determine whether these variants were collectively more frequent in patients with these diseases than among healthy controls. This aim was successfully achieved and, in addition, we were able to obtain suggestive statistical significance for the association between the 22q11 350 kb deletion and ASD. This deleted segment, located within the chromosomal region deleted in VCFS, a contiguous gene syndrome known to be associated with a high frequency of MR, ASD and psychosis, contained two genes PRODH and DGCR6. Although we cannot exclude an involvement of DGCR6 in the neuropsychiatric phenotype of the subjects bearing this CNV, previous work from our group strongly suggests that PRODH is the prime candidate. We had previously shown that hyperprolinemia, resulting from the partial or total inactivation of this enzyme (i) may lead to MR and autism in patients with type 1 hyperprolinemia 18 (ii) is a risk factor for schizoaffective disorder 32 (iii) is inversely correlated with IQ in the VCFS 18. Here, we show that all but one patients harboring this deletion were in fact compound heterozygotes, also bearing mutations affecting enzymatic activity on the second allele. It results in a loss of at least 70% of the predicted PRODH residual activity in 14/15 subjects and in hyperprolinemia in 7/10 assessed subjects.

Second, we show that both de novo CNVs and CNVs inherited from an apparently healthy parent can be found in patients. For transmitted CNVs, the mode of inheritance of the disease was, in some cases recessive (e.g. hyperprolinemia related to the 22q11 deletion) or implied the transmission of an X linked gene by a woman to her son (GRIA3). Consistent with previous reports 4, 8, 33, the 16p11 rearrangements were inherited from an apparently non affected parent in three families. These CNVs, whose estimated frequency in Icelandic population was 5/18,834 (0.03%) for the duplication and 2/18,834 (0.01%) for the deletion 31, should therefore be considered as risk factors rather than fully causative variations. The presence of affected siblings that do not share the CNV, already noted in a previous report 33, does not necessarily rule out the causative implication of these CNVs, but raises the question of intra-familial genetic heterogeneity. This hypothesis, which is not un- plausible for these frequent disorders often characterized by assortative mating, remains speculative since the parents and their relatives were not psychiatrically nor cognitively assessed in these families.

Third and more importantly, our study confirms and extends the conclusions of several recent reports suggesting that a large number of candidate CNVs are not disease-specific but are involved in the expression of different behavioral phenotypes including MR, ASD and SZ. This implies the existence of shared biological pathways between these three neurodevelopmental conditions. These pathways affect chiefly synapse formation and maintenance as well as neurotransmission with a special emphasis on glutamate and GABA. The dysfunction of specific neuronal networks underlying the peculiar symptomatology of each clinical condition most likely depends on additional genetics, epigenetics and environmental factors which remain to be characterized. From a clinical point of view, it should be stressed that despite the diversity of categorical diagnoses, many subjects harboring these CNVs shared some clinical features: 1/3 of patients with schizophrenia and 80% of autistic patients with CNVs had a level of cognitive functioning which falls in the range of mental retardation. This is in accordance with previous reports showing that that point prevalence of schizophrenia is increased by a factor of 3 in subjects with intellectual disabilities 34 and that 3/4 of autistic patients have mental retardation 35. However, it should be remembered that (i) because attention and communication are markedly impaired in children with autism, assessment of their IQs (even performance IQs in non verbal subjects) are notoriously unreliable and (ii) these results were not obtained in a single community based population, but in three disease groups ascertained according to different schemes, a factor whose exact impact is difficult to appreciate but which is likely to have implications related to the phenotypic severity of these subject.

Finally, we would like to point out the utility of targeted methods for CNV analysis, such as the QMPSF method, as a cost effective alternative to a-CGH for the screening of candidate loci in large case-control cohorts. We plan to conduct extensive resequencing of these candidate genes in order to further validate their role in these conditions.