SIR–It has recently been reported that genetic variants of proline dehydrogenase (PRODH) increase susceptibility to schizophrenia. 1 This is of interest since PRODH maps within the velo-cardio-facial syndrome (VCFS) region of 22q11, deletion of which is one of the strongest known risk factors for schizophrenia. 2 We have examined this claim using two samples each with high power (> 99%) to detect the effects reported. Our data suggest that genetic variation at PRODH is unlikely to be associated with even modestly increased risk of schizophrenia, and that the earlier report is the result of chance. Liu et al1 reported association between various permutations of SNPs and haplotypes constructed from markers at positions 1766, 1945, and 2026 of PRODH in a small sample (n= 107) of European-American schizophrenic proband–parent trios, in a second sample of 26 trios where the proband had a childhood onset illness, and in a third small sample of 109 adult Afrikaners and 75 controls. Their analysis suggested an association between schizophrenia and a ‘221’haplotype at three loci 1766/1945/2026, defined by alleles G, T, and C, respectively, and this effect was strongest for disease of early onset. We have tested this in a large, well-matched Caucasian DSM IV schizophrenia case (n= 677) and control (679) sample ascertained in the UK and the Republic of Ireland (for details of ascertainment and diagnostic methodology, see Anney et al3). In total, 49 of the subjects had juvenile onset schizophrenia (JOS) defined by hallucinations, delusions or grossly disorganised behaviour at age 17 years or less (mean age at onset 15.8 y, sd= 1.9 years). We also studied a sample of 55 unrelated parent–JOS proband–parent trios from Bulgaria (25 male; mean age at onset 16 years (range 9–17). Assessment and diagnostic methods used were the same as for the UK/Irish sample. Tests for haplotype association between schizophrenia and detected SNPs were performed using EHPLUS4 and PMPLUS. 5 The Bulgarian sample was analysed using eTDT software. 6 It was possible to use this method for haplotypes as the LD structure of these three markers allows all parental and proband haplotypes to be unambiguously assigned by hand. The frequency of the 221 putative risk haplotype in our case–control sample was actually lower (0.11) in cases than in controls (0.13), although not significantly so (OR= 0.86, 95% CI= 0.68–1.09, w2= 1.6, 1 df, P= 0.21). When the UK JOS sample was considered separately, again the risk haplotype was reduced in frequency (0.07), but not significantly (OR= 0.54, 95% CI= 0.25–1.17, w2= 2.4, 1 df, P= 0.12). We also found no evidence for excess transmission of the risk haplotype (17 transmitted, 19 nontransmitted, P= 0.74) in the Bulgarian JOS sample. Although there is no a priori hypothesis, in order to fully explore our data sets, we undertook global tests of haplotype association, and tested the SNPs individually for genotypic and allelic association in the UK case–control sample. Genotypic and allelic data are given in Table 1. We found not even a trend (Po0. 15) for association with any allele, genotype or haplotype in the sample as a whole, or in the UK or Bulgarian JOS samples alone (data not shown). All genotypes were in Hardy–Weinberg equilibrium, and the LD relationships between markers were identical in the UK and Bulgarian samples. Our case–control sample has a power of 99% to detect the effect described1 in European-American adult samples at a= 0.01, yet we found not even a trend for association. As the upper ends of the 95% confidence interval for the OR for the risk haplotype is 1.09, we can exclude even small …