I read the paper by Lencz et al ^{Reference Lencz, Lipsky, DeRosse, Burdick, Kane and Malhotra1} with concern for the future of psychosis genetics. The authors claim that their candidate gene study of BDNF is ‘the first to demonstrate association with schizoaffective disorder but not schizophrenia’ and therefore that ‘BDNF variation is associated with psychiatric disorders with a primary affective component’. To reach this conclusion they argue on the basis of a sample size of 596 individuals against two meta-analyses and two cohort studies with sample sizes between 6 and 26 times larger (Table 1). Each of these studies examined the Val66Met polymorphism (the subject of Lencz et al's report) and reached the conclusion that BDNF genotype does not exert an influence on the development of affective illness whether or not associated with psychosis.

Table 1 Main findings of two recent studies of the Val66Met variation in BDNF in relation to psychiatric diagnosis compared with Lencz et al ^{Reference Lencz, Lipsky, DeRosse, Burdick, Kane and Malhotra1}

	Controls, n	Schizophrenia, n	Schizoaffective disorder, n	Bipolar disorder, n	Depression, n	P
Kanazawa et al Reference Kanazawa, Glatt, Kia-Keating, Yoneda and Tsuang2
Meta-analysis	2035	2955
Meta-analysis	6347			3143		0.161
Chen et al Reference Chen, Lawlor, Lewis, Yuan, Abdollahi and Timpson3
BWHHS	2367				553	0.360
ALSPAC	6242				596	0.834
Meta-analysis	11040				3879	0.537
Lencz et al Reference Lencz, Lipsky, DeRosse, Burdick, Kane and Malhotra1
HC v. Sz	222	211				NS
HC v. (SzAf+Bip+MDD)	222		61	77	29	0.015
Sz v. (SzAf+Bip+MDD)		211	61	77	29	0.08

ALSPAC, Avon Longitudinal Study of Parents and Children; BWHHS, British Women's Heart and Health Study; HC, healthy controls; MDD, major depressive disorder; NS, not significant; Sz, schizophrenia; SzAf, schizaffective disorder

A literature survey indicates that between 2004 and 2009 these authors between them published 25 papers relating to associations of 19 genes with aspects of psychiatric disease. Concerning one gene (FEZ1) they drew negative conclusions, but concerning each of the other 18 they claim a relationship was established. Such a rate of gene discovery would be a remarkable achievement. My review of the linkage literature, ^{Reference Crow4} as represented by the four largest (each >300 sibpairs) studies, suggests that none of Lencz et al's candidate genes were replicated in these systematic searches, and the association study of Sanders et al ^{Reference Sanders, Duan, Levinson, Shi, He and Hou5} that investigated six of them (DISC1, DAOA, HTTLPR, DTNBP1, COMT, DRD2) in 1870 individuals with schizophrenia or schizoaffective disorder and 2002 controls concluded these genes were unrelated to psychosis.

When large numbers of variables are examined, simultaneously alluring relationships can often be discerned that evaporate in the wider context of large and systematic studies. It appears that by ignoring this context Lencz et al are operating an algorithm for generating positive associations in selected data-sets.