Prenatal maternal infection, neurodevelopment and adult schizophrenia: a systematic review of population-based studies

Abstract

Background

Disruption of foetal development by prenatal maternal infection is consistent with a neurodevelopmental model of schizophrenia. Whether specific prenatal infections are involved, their timing and the mechanisms of any effect are all unknown. We addressed these questions through a systematic review of population-based studies.

Method

Electronic and manual searches and rigorous quality assessment yielded 21 studies that included an objective assessment of individual-level prenatal maternal infection and standardized psychotic diagnoses in adult offspring. Methodological differences between studies necessitated a descriptive review.

Results

Results for prenatal maternal non-specific bacterial, respiratory or genital and reproductive infection differed between studies, which reported up to a two- to fivefold increased risk of schizophrenia. Evidence for herpes simplex virus type 2 (HSV-2) and Toxoplasma gondii was mixed; some studies reported up to a doubling of schizophrenia risk. Prenatal HSV-1 or cytomegalovirus (CMV) infections were not associated with increased risk. Exposure to influenza or other infections during early pregnancy may be more harmful than later exposure. Increased proinflammatory cytokines during pregnancy were also associated with risk. Prenatal infection was associated with structural and functional brain abnormalities relevant to schizophrenia.

Conclusions

Prenatal exposure to a range of infections and inflammatory responses may be associated with risk of adult schizophrenia. Larger samples, mediation and animal models should be used to investigate whether there is a ‘sensitive period’ during development, and the effects of prenatal infections on neurodevelopment. Inclusion of genetic and immunological information should help to elucidate to what extent genetic vulnerability to schizophrenia may be explained by vulnerability to infection.

Introduction

Maternal infection during pregnancy with Toxoplasma gondii, rubella, cytomegalovirus (CMV), herpes simplex virus (HSV) and other microbes (i.e. TORCH infections) have long been known to be associated with mental retardation, cerebral hypoplasia, ventriculomegaly and other brain and behavioural abnormalities in the offspring (Remington et al. 2006). The neurodevelopmental theory of schizophrenia postulates abnormalities in early neurodevelopment as a possible cause of the disorder (Murray & Lewis, 1987; Weinberger, 1987). Empirical evidence for this theory comes from birth cohort studies demonstrating delays in motor milestone, and deficits in pre-morbid cognitive function in future cases of schizophrenia (Jones et al. 1994; Khandaker et al. 2011). Consequently, several studies have investigated links between maternal infection during pregnancy and risk of schizophrenia in adult offspring.

Many of the early investigations on this topic were based on data from influenza epidemics (Mednick et al. 1988), which yielded conflicting results (Selten et al. 2010). These studies defined maternal exposure to influenza as being pregnant at the time of an epidemic rather than direct measurement of exposure at the individual level. Thus, the ecological fallacy could be a reason for their discrepant findings (Selten et al. 2010). More recent studies based on general population cohorts have overcome this issue. These have used objective measures to determine exposure to infection during the prenatal period at the individual level (Susser, 1999; Susser et al. 2000; Buka et al. 2001a; Mortensen et al. 2007; Susser et al. 2011). Findings from these studies merit further attention, as they may provide a better understanding of the true nature of this association.

A previous review implicated several prenatal maternal infections with increased risk of schizophrenia in adult offspring (Brown & Derkits, 2010). However, fundamental questions have yet to be settled. These include the timing of prenatal infection or any links between prenatal infection and structural and functional brain abnormalities associated with schizophrenia. These issues are important because they would provide important clues to the relevance of prenatal infections in the causation of adult psychotic disorders. Therefore, associations between prenatal infection and neurodevelopmental outcomes in offspring are of much interest in current epidemiological and preclinical research (Meyer & Feldon, 2010; Brown et al. 2011). We have carried out a systematic review of robust population-based studies to address the following issues: (i) association between adult psychotic illness and specific types of prenatal maternal infection, (ii) association between risk and timing of prenatal infection, and (iii) effects of prenatal infection on neurodevelopment. We have included three new studies published on this topic since the most recent review, contributing evidence on a further 4000 cases and more than 1.1 million controls (Xiao et al. 2009; Mortensen et al. 2010; Nielsen et al. 2011). Findings from these studies provide a new insight into the nature of the association between prenatal maternal infection and psychosis in offspring. We also discuss the biological plausibility of such associations in view of recent human and preclinical research.

Method

Search strategy

We searched Medline, PubMed and EMBASE databases from their respective inceptions to November 2011 for studies of prenatal infection that were based on human samples and published in the English language. Search terms included indexing terms (e.g. MeSH) and free texts: [(prenatal OR in utero OR foetal OR maternal) AND (infection OR inflammation OR illness) AND (schizophrenia OR psychotic disorder)]. We also identified potential studies in the reference lists of included studies and wrote to prominent authors in this field for any further relevant studies.

Study selection

Included studies (i) were based on general population datasets, (ii) cohort or nested case–control in design, (iii) used serological assays or clinical examination to determine exposure to infection during the prenatal period at the individual level and (iv) used contemporary ICD or DSM directions to define the outcome of schizophrenia, non-affective psychosis, affective psychosis and other psychotic disorders. Studies that used self-report of exposure to infections instead of clinical examination or serology were excluded. Genetic high-risk studies and studies that used aggregate data to define exposure, for example pregnancy at the time of an epidemic or incidence or prevalence of certain infection during pregnancy rather than direct measurement of infection at the individual level, were also excluded.

Data extraction

Electronic search and study selection were carried out by two researchers working independently (G.M.K. and J.Z.). They examined all titles and abstracts, obtained full texts of potentially relevant papers and applied inclusion criteria. Any differences in opinion were resolved through discussions with the wider study team. All studies that met inclusion criteria were critically appraised using the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) checklists for observational studies (von Elm et al. 2007), with focus on sampling, measurement of exposure and outcome, follow-up, attrition, analytical strategy, bias, confounding and other methodological issues. At this stage, studies with serious methodological concerns that might have biased their findings were identified and excluded from the review.

Data synthesis and rationale for not proceeding to meta-analysis

There was considerable variation among selected studies with regard to design, measurement of prenatal infection, definition of outcome and methods of case ascertainment. For example, three studies investigated links between schizophrenia and maternal exposure to HSV type 2 (HSV-2) during the prenatal period. To define exposure, one study used the ratio of anti-HSV-2 immunoglobulin (Ig)G concentration between cases and controls (Buka et al. 2008). The rest of the studies used absolute concentration of IgG in maternal serum (Brown et al. 2006; Mortensen et al. 2010). Again, the cut-off values used to define exposure were not the same between these two studies (issues around exposure measurement are considered in detail in the discussion section). Definitions of outcomes were also different. For example, two studies used non-affective psychosis by DSM-IV (Brown et al. 2006; Buka et al. 2008) whereas one used schizophrenia by ICD-10 (Mortensen et al. 2010). Such differences in exposure and outcome measurements were also observed for other infections. Because of this heterogeneity the results of meta-analyses would be difficult to interpret and we therefore present a descriptive review of the studies.

Results

Twenty-one studies were included in the review (Fig. 1). There were 16 studies concerning risk of adult psychotic illness. One of these studies also examined childhood IQ in relation to exposure to influenza in the prenatal period. Five studies looked exclusively into structural and functional brain abnormalities related to schizophrenia. Excluded studies with reasons for exclusion are presented as online supplementary material in Table S1.

Fig. 1.

Study selection process for the systematic review.

Prenatal maternal infection and schizophrenia and other psychoses in adult offspring

We reviewed 16 studies on this topic (Brown et al. 2000, 2004a, b, 2005, 2006; Buka et al. 2001a, b, 2008; Babulas et al. 2006; Mortensen et al. 2007; Clarke et al. 2009; Ellman et al. 2009; Sorensen et al. 2009; Xiao et al. 2009; Mortensen et al. 2010; Nielsen et al. 2011). Eleven of these studies were based on two US birth cohorts, the National Collaborative Perinatal Project (NCPP) cohort (Susser, 1999; Susser et al. 2011) and the Prenatal Determinants of Schizophrenia (PDS) cohort (Susser et al. 2000). The PDS cohort (Susser et al. 2000) was derived from the Child Health and Development Study (CHDS; van dan Berg et al. 1988), a prospective birth cohort study of factors affecting outcomes of pregnancy and child development. The NCPP is a multicentre prospective cohort study involving 13 sites. Studies included in the present review are all based on the samples from Providence, Boston and Philadelphia; together, the first two sites are known as the New England Family Study (Susser, 1999; Susser et al. 2011). The remaining five studies were based on general population cohorts from Denmark and Finland. Characteristics of these cohorts and assessment of exposure and outcome in studies based on these cohorts are presented in Table 1.

Table 1.

Cohort characteristics and assessment of exposure and outcome in studies of prenatal infection

Name of cohort	Cohort characteristics	Duration of follow-up	Source, type and timing of exposure sample	Outcomes and diagnostic criteria	Case ascertainment
NCPP, USA	Born between 1959 and  1966 ; n=25 025 in  Providence, Boston and  Philadelphia  sites combined	Up to age 35 years	Maternal serum collected  during late pregnancy or at  delivery used for serological  tests for specific infections  and cytokines	Affective and non-affective  psychosisa by DSM-IV	Boston and Providence: interview and/  or record linkage with psychiatric  treatment facilities to identify potential  cases, followed by interview using  SCID or medical chart review to  establish diagnosis Philadelphia :  screening of psychiatric database  followed by chart review
PDS, USA	Born between 1959 and  1966 ; n=12 094	Up to mid-thirties	Serology in maternal serum  collected during pregnancy  and delivery, and clinically  diagnosed infection during  pregnancy	SSDb by DSM-IV	Psychiatric in- and out-patient registers  and pharmacy registers were screened  for potential cases, followed by  diagnosis through interview using  DIGS or case-note review
Danish register-  based cohorts	Individuals born since  1981	Up to age 18 and  25 years	Blood spots on filter paper  obtained from 5–7-day-old  neonates used for  immunoassay	Schizophrenia by ICD-10	Danish National Psychiatric Register
Copenhagen  perinatal cohort	Born between 1959 and  1961 ; n=9125	Up to 47 years	Clinically diagnosed and/or  treated maternal infection  (bacterial or viral) during  pregnancy	Schizophrenia by ICD-8  and ICD-10	Danish National Psychiatric Register
Finnish cohort	Individuals born in  Helsinki between 1947  and 1990	–	Maternal hospitalization for  pyelonephritis during  pregnancy	Schizophrenia by ICD-8  to ICD-10	Finnish Hospital Discharge Register

NCPP, National Collaborative Perinatal Project ; PDS, Prenatal Determinants of Schizophrenia ; DIGS, Diagnostic Interview for Genetic Studies ; SSD, schizophrenia spectrum disorder.

^aIncludes schizophrenia, schizo-affective, other non-affective psychosis, schizo-affective bipolar type, bipolar disorder with psychotic features, major depressive disorder with psychotic features and psychosis not otherwise specified (NOS).

^bIncludes schizophrenia, delusional disorder, psychotic disorder NOS, schizo-affective disorder and schizotypal personality disorder.

To define exposure, five studies used prospectively collected records of clinically diagnosed infection. Nine studies used serological assays for various antimicrobial antibodies in maternal blood samples obtained during pregnancy, and a further two used inflammatory cytokines as a proxy of exposure to infections. Serological studies were all nested case–control in design whereas the studies of clinically diagnosed infection involved analyses of entire cohorts. We have summarized the results of ‘serological’ and ‘clinically diagnosed’ studies separately, including effect of adjustment for confounding factors in Table 2 (for an extended version of this table see Table S2 in the online supplementary material). Strengths and weaknesses of individual studies are presented in Table 3.

Table 2.

Population-based studies of prenatal infection and schizophrenia and other psychosis

Study	Setting and design	Case/control	Case definition	Exposures measured	Adjustment for confounding	Main findings
Serological studies  of specific infection
Buka et al. 2001a	NCPP, nested  case–control	27/54	Affective and non-affective  psychosis by DSM-IV	CMV, rubella, Toxoplasma  gondii, human parvovirus  B19, HSV-1, HSV-2 and  Chlamydia	Gender, ethnicity, age,  social class, maternal  mental illness, weight  gain and smoking in  pregnancy	Elevated IgG to HSV-2 in  case mothers, no increase  in IgG to HSV-1, CMV,  T. gondii or rubella
Buka et al. 2008	NCPP, nested  case–control	200/544	Affective and non-affective  psychosis by DSM-IV	HSV-2	Birth city and season,  gender, ethnicity,  maternal education and  parental mental illness	Adjusted OR (95% CI) for  schizophrenic psychosis for  maternal HSV-2  seropositivity 1.8 (1.1–3.0)
Xiao et al. 2009	NCPP, nested  case–control	219/618	Affective and non-affective  psychosis by DSM-IV	T. gondii (serotype I, II  and III)	Birth city, age, gender  and ethnicity	Adjusted OR for affective  psychosis 5.24 (95% CI  1.69–16.49) for exposure to  T. gondii type I
Ellman et al. 2009	NCPP, nested  case–control	96/274	Affective and non-affective  psychosis by DSM-IV	Influenza	Maternal ethnicity, social  class and offspring  gender	Non-significant 1.7-fold  increased risk of  schizophrenia for influenza B
Brown et al. 2004a	PDS, nested  case–control	64/125	SSD by DSM-IV	Influenza	Maternal age, gender and  gestational age of foetus	Unadjusted OR (95% CI) for  SSD for influenza in first  trimester 7.0 (0.7–75.3) and in  first half of pregnancy 3.0  (0.98–10.1)
Brown et al. 2005	PDS, nested  case–control	63/123	SSD by DSM-IV	T. gondii	Date of birth, gender and  maternal age	Adjusted OR for SSD 2.61  (95% CI 1.00–6.82) for  maternal ‘high’ versus ‘low’  IgG antibody titre
Brown et al. 2006	PDS, nested  case–control	60/110	SSD by DSM-IV	HSV-1, HSV-2 and CMV	Age, gender, maternal  ethnicity and education	No association between SSD  and HSV-1, HSV-2 or CMV
Mortensen et al. 2007	Danish, nested  case–control	71/648	Schizophrenia by ICD-10	T. gondii and HSV	Year and place of birth,  gender, family history  of mental illness	OR for schizophrenia 1.79  (95% CI 1.01–3.15) for T.  gondii seropositivity (IgG) ;  HSV-1 and -2 : no association  with schizophrenia/  schizo-affective disorder
Mortensen et al. 2010	Danish, nested  case–control	602/602	Schizophrenia by ICD-10	HSV-2	Gender, family history of  mental illness,  urbanization at place of  birth, immigrant status  and parental age	Incidence rate ratio for  schizophrenia 1.51 (95% CI  1.08–2.10) using cut-off value  of ≥0.2 for serum IgG
Serological studies  of inflammatory  cytokines
Buka et al. 2001b	NCPP, nested  case–control	27/54	Affective and non-  affective psychosis  by DSM-IV	IL-1β, IL-2, IL-4, IL-8,  TNF-α	Gender, ethnicity, age,  social class, maternal  mental illness, weight  gain and smoking in  pregnancy	Increased TNF-α in case  mothers ; OR for psychosis  8.5 (p<0.03) in offspring of  women with both third  trimester infection and  TNF-α levels >75th
Brown et al. 2004b	PDS, nested  case–control	59/118	SSD by DSM-IV	IL-1β, IL-6, IL-8, TNF-α	Foetal age, gender,  maternal age, social class,  ethnicity	Nearly twofold increase  in mean and median values  of IL-8 at second/third  trimester in case mothers
Studies of clinically  diagnosed infection
Brown et al. 2000	PDS,  prospective  cohort	58/7781	SSD by DSM-IV	Respiratory infectiona	Maternal ethnicity,  education and smoking  during pregnancy	Second-trimester infection :  adjusted OR for SSD 2.13  (95% CI 1.05–4.35),  schizophrenia 2.07 (95% CI  0.8–5.36)
Babulas et al. 2006	PDS,  prospective  cohort	71/7723	SSD by DSM-IV	Genital/reproductive  infectionb	Maternal age, ethnicity,  education and mental  illness	Periconceptional infection :  adjusted rate ratio for SSD  5.03 (95% CI 2.00–12.64) ;  no increased risk for  infection during pregnancy
Clarke et al. 2009	Finnish  retrospective  national cohort	9596 exposed  v. 13 808  unexposed	Schizophrenia by ICD-8  to ICD-10	Pyelonephritis	–	First-trimester infection, OR  3.35 (95% CI 0.8–14.0) ; risk  difference 0.51% in those  with a positive family history  versus 0.10% in those without  such history
Sorensen et al. 2009	Copenhagen,  prospective  cohort	153/7788	Schizophrenia by ICD-8  and ICD-10	Bacterial and viral  infectionc	Social class and use of  analgesics in pregnancy	First-trimester bacterial  infection : schizophrenia risk  by age 34, adjusted OR 2.53  (95% CI 1.07–5.96), by age 47,  adjusted OR 2.14 (95% CI  1.06–4.31)
Nielsen et al. 2011	Danish register-  based cohort	3722/1 115 752	Schizophrenia by ICD-8 to  ICD-10	Maternal and paternal  hospitalization for  infectiond	Parental history of  schizophrenia, age and  gender	Adjusted risk ratio for  maternal second-trimester  infection 1.72 (95% CI  1.15–2.46), infection any time  in the past before, during or  after pregnancy 1.24 (95% CI  1.16–1.33) ; paternal infection  any time in the past 1.14  (95% CI 1.05–1.23)

NCPP, National Collaborative Perinatal Project ; PDS, Prenatal Determinants of Schizophrenia ; SSD, schizophrenia spectrum disorder ; CMV, cytomegalovirus ; Ig, immunoglobulin; HSV-1, herpes simplex virus type 1 ; HSV-2, herpes simplex virus type 2 ; IL, interleukin ; TNF, tumour necrosis factor ; OR, odds ratio ; CI, confidence interval.

^aMaternal respiratory infections included tuberculosis, influenza, influenza with pneumonia, bronchopneumonia, atypical pneumonia, pleu-risy, emphysema/viral respiratory infections, acute bronchitis, and upper respiratory infections.

^bGenital/reproductive infections included endometritis, cervicitis, pelvic inflammatory disease, vaginitis, syphilis, condylomata, ‘ venereal disease ’ and gonorrhoea. Periconceptional period=30 days before and after last menstrual period.

^cBacterial infections included sinusitis, tonsillitis, pneumonia, cystitis, pyelonephritis, bacterial venereal infection, and any other bacterial infection. A viral infection was judged to be present if (1) a medical diagnosis had been made (commonly by the general practitioner) or (2) symptoms consistent with minor respiratory illnesses or influenza had been present and the mother had been confined to bed and had a rise in temperature to at least 38 °C.

^dInfections include general infections, skin infections, respiratory infections, infections related to puerperium (mothers), genital and other infections.

Table 3.

Strengths and weaknesses of individual studies

Study	Strengths	Weaknesses
NCPP birth cohort
Buka et al. 2001a	Studied several exposures, included several immunoglobulins, two-stage attempt  to case identification	Small sample and heterogeneous case group; for some cases used  maternal IgG levels without having documented seropositivity
Buka et al. 2001b	First to investigate inflammatory cytokines in these studies, included several  cytokines	Small sample and heterogeneous case group
Buka et al. 2008	Large sample allowing study of schizophrenic and affective psychosis separately	Broad definition of case ; in some cases diagnosis were made by  chart review ; overall high seropositivity in sample due to high  seropositivity among African-Americans
Xiao et al. 2009	Large sample, studied three serotypes of Toxoplasma gondii	Broad definition of case ; in some cases diagnoses were made by  chart review
Ellman et al. 2009	Examined childhood IQ and also adult schizophrenia with regard to foetal  exposure to influenza	Small sample, broad definition of cases
PDS birth cohort
Brown et al. 2000	Analysed entire cohort, large sample	Broad exposure of respiratory infections, under-reporting of  exposure and misclassification of timing of exposure is possible
Brown et al. 2004a	First serological study of prenatal influenza, good estimate of timing of exposure	Exposure based on proxy measure of seroconversion but validated  in a comparison sample with seroconversion data
Brown et al. 2004b	Larger sample size compared to a similar previous study, included several  cytokines	Median values of one cytokine (TNF-α) was lower than in most  studies of this cytokine
Brown et al. 2005	Two-stage exposure assessment, use of more sensitive tests than previous studies	Small sample size
Brown et al. 2006	Outcome more focused than previous studies, analysis restricted to seropositive  mothers	Small sample size
Babulas et al. 2006	Analysed entire cohort, large sample	Broad exposure definition, small number of exposed cases (n=5)
Danish cohorts
Mortensen et al. 2007	Large sample size	Blood samples were missing for 28% of eligible cases, short follow-  up
Sorensen et al. 2009	Analysed entire cohort, large sample, examined both narrow and broad  schizophrenia	Broad categories of exposure, viral and bacterial infections,  misclassification possible
Mortensen et al. 2010	Large sample size, detail family psychiatric history taken into account	Short follow-up
Nielsen et al. 2011	Large sample, included paternal infection and maternal infection before and after  pregnancy, family history of schizophrenia taken into account	Only able to use hospital admission for infection as exposure ;  limited power for analysis of infection during pregnancy
Finnish cohort
Clarke et al. 2009	Large sample size, used sibling comparison group, precise timing of exposure  using hospital admission data	Small number of cases meant low statistical power to examine each  trimester of pregnancy

NCPP, National Collaborative Perinatal Project ; PDS, Prenatal Determinants of Schizophrenia ; IgG, immunoglobulin G; TNF-α, tumour necrosis factor-α.

Serologically determined prenatal maternal infection

HSV

Three out of five studies reported a statistically significant increased risk of schizophrenic psychosis in the adult offspring for exposure to HSV-2 infection during pregnancy (Buka et al. 2001a, 2008; Mortensen et al. 2010). These studies used elevated levels of IgG antibodies to HSV-2 to define exposure, that is seropositivity. A recent Danish study involving 602 cases reported a statistically significant 50% increased risk of adult schizophrenia for prenatal exposure to HSV-2 (Table 2). Risk was slightly attenuated (but still statistically significant) only after adjustment for family history of mental illness (Mortensen et al. 2010). This risk estimate is in line with a previous report from the NCPP birth cohort (Buka et al. 2008).

One study from the PDS cohort reported a 30% increase in risk of schizophrenia spectrum disorder (SSD) for maternal HSV-2 seropositivity during late pregnancy (Brown et al. 2006), but this was not statistically significant. The study also reported similar results for exposures to HSV-1 and CMV. A Danish study also reported no increase in risk of schizophrenia for prenatal exposure to HSV-1 and -2 (Mortensen et al. 2007).

T. gondii

Elevated levels of maternal IgG antibodies to T. gondii were reported to be associated with increased risk of schizophrenic psychosis from two cohorts (Brown et al. 2005; Mortensen et al. 2007). One study from the PDS cohort (Brown et al. 2005) found a more than twofold increase in risk of SSD for maternal ‘high’ IgG antibody titre (≥1:128) during late pregnancy, compared to ‘low’ titre (<1:16), with an odds ratio (OR) of 2.61 and 95% confidence interval (CI) 1.00–6.82. Similar findings (OR 1.79) were observed in the Danish study (Mortensen et al. 2007), which also suggested that elevated maternal IgG to T. gondii was associated specifically with risk of schizophrenia, out of all four outcomes included in the analysis.

One study involving the largest number of cases (n=219) and controls to date investigated prenatal exposure to three specific serotypes of T. gondii in relation to affective and non-psychosis in adult offspring (Xiao et al. 2009). Infection with T. gondii Type I (out of all three serotypes) was reported to be associated with increased risk of affective psychosis (more than fivefold) but not non-affective psychosis (Table 2). However, analysis of non-affective psychosis was based on an insufficient sample size of 11 cases and 29 controls.

Influenza

Two studies were available. One was based on the PDS birth cohort and reported a sevenfold increased risk of SSD for exposure during the first trimester and a threefold increased risk for exposure during the first half of pregnancy (Brown et al. 2004a). However, the sample size was relatively small and none of the risk estimates were statistically significant (Table 2). There was no increase in risk for exposure during late pregnancy. The second study was from the NCPP birth cohort and it assayed maternal blood samples collected at delivery for IgG antibodies to influenza B virus and reported a non-significant 70% increase in risk of schizophrenia in the adult offspring (Ellman et al. 2009).

Inflammatory cytokines

One study from the NCPP cohort suggested that inflammatory cytokines may mediate risk of psychosis related to prenatal infection (Buka et al. 2001b). Among 27 cases of affective and non-affective psychosis and 54 matched controls, Buka et al. (2001b) found that mean tumour necrosis factor (TNF)-α level was significantly higher in case mothers. Furthermore, adult offspring of women with both clinical report of third trimester infection and elevated TNF-α level were eight times more likely to develop psychosis. However, this analysis was based on a small sample. A similar study based on the PDS cohort reported a nearly twofold increase in mean and median values of interleukin (IL)-8 during mid- to late pregnancy in SSD case mothers (Brown et al. 2004b). This association was unchanged in analysis restricted to cases of schizophrenia only.

Clinically diagnosed prenatal maternal infection

General infections

In a large Danish cohort maternal hospitalization for infection during pregnancy was associated with increased risk of schizophrenia in offspring [risk ratio (RR) 1.39, 95% CI 1.18–1.62; Nielsen et al. 2011]. Only second-trimester infection was statistically significant (RR 1.72, 95% CI 1.15–2.46). Of note, maternal infection any time in the past (before, during or after pregnancy combined) was also associated with increased risk of schizophrenia (RR 1.24, 95% CI 1.16–1.33). Maternal infection during pregnancy was not significantly different from that before or after pregnancy in terms of offspring’s risk of schizophrenia (Nielsen et al. 2011).

In the Copenhagen Perinatal Cohort bacterial infection during the first trimester was reported to be associated with a nearly twofold increased risk of schizophrenia at follow-up at both age 34 and 47 years (Sorensen et al. 2009) (Table 1). Risk was also increased, but not statistically significant, for second-trimester exposure to bacterial infections. There was no significant increase in risk for viral infections. However, broad definitions of bacterial and viral infection were used in this cohort.

Respiratory infection

In an analysis of the entire PDS cohort, maternal respiratory infection during the second trimester was reported to be associated with a twofold increased risk of SSD in adult offspring (Brown et al. 2000). A similar increase in risk was observed for schizophrenia. Most exposed cases suffered from an upper respiratory tract infection. Similarly, in the Copenhagen Perinatal Cohort maternal upper respiratory infection during any time in pregnancy was associated with a threefold increased risk of schizophrenia in the offspring by age 47 years.

Genital, reproductive and urinary infection

Maternal genital or reproductive infection during the periconceptional period (i.e. 30 days before and after the last menstrual period) was associated with a five-fold increased risk of SSD in adult offspring in the PDS cohort (Babulas et al. 2006). However, there was no increase in risk for such infections during pregnancy (first, second or third trimester). In another study, maternal hospital admission for genital infection during pregnancy was associated with a twofold increased risk of schizophrenia in adult offspring (Nielsen et al. 2011). Similarly, in the Copenhagen Perinatal Cohort maternal gonococcal infection during any time in pregnancy was associated with a more than threefold increased risk of schizophrenia by age 47 years (Sorensen et al. 2009). One Finnish study did not find any significant association between maternal hospitalization with pyelonephritis during pregnancy and risk of schizophrenia in the offspring (Clarke et al. 2009).

Timing of prenatal maternal infection and risk of adult schizophrenia

Most studies lacked power to address the issue of a ‘sensitive period’, that is maternal infection during which stage of pregnancy may be most harmful. However, in general we observed a trend for greater risk of adult psychotic illness for exposure to infections during early stages of gestation (Table 4). This needs to be interpreted in the context of significant variation among studies in design, exposure and outcome measurement.

Table 4.

Timing of prenatal infection and risk of schizophrenia and other psychotic disorders in adult offspring

Time of exposure	Infectious agent	Effect sizea	Study	Cohort
Around conception	Genital/reproductive infection	5	Babulas et al. 2006	PDS
First trimester	Influenza	7	Brown et al. 2004a	PDS
	Bacterial infection	2	Sorensen et al. 2009	Copenhagen  perinatal cohort
Second trimester	Influenzab	3	Brown et al. 2004a	PDS
	Respiratory infection	2	Brown et al. 2000
Third trimester or at delivery	HSV-2	1.5	Buka et al. 2001, 2008  and Xiao et al. 2009	NCPP
	Toxoplasma gondii	2	Brown et al. 2005	PDS
			Mortensen et al. 2007, 2010	Danish cohorts

NCPP, National Collaborative Perinatal Project ; PDS, Prenatal Determinants of Schizophrenia ; HSV-2, herpes simplex virus type 2.

^aOnly studies that reported an increase in risk in relation to prenatal infection were included in this table ; approximate point estimates shown for guide purpose only ; heterogeneity in exposure measurement, case definition, analysis and measure of risk exists between individual studies, refer to Table 1 for individual study details and results.

^bEarly second-trimester exposure.

Interaction between prenatal maternal infection and other risk factors

Family history of psychosis

Using an additive model, a Finnish study reported evidence of an interaction between family history of psychosis and maternal pyelonephritis during pregnancy (Clarke et al. 2009). In a stratified analysis, the risk difference between those exposed and those unexposed to prenatal infection in the group with a family history of psychosis was five times larger than the risk difference in those without such history.

Preterm birth

In the largest study of HSV-2, involving 602 cases, there was an indication that the incidence rate ratio (IRR) for schizophrenia is slightly higher among exposed individuals born before 36 weeks of gestation (IRR 3.68, 95% CI 0.76–17.79) compared to those born at term (IRR 1.52, 95% CI 1.13–2.03). However, this estimate was based on small numbers: only seven exposed cases and two exposed controls were born preterm (Mortensen et al. 2010).

Maternal sexual and reproductive behaviour

One study from the NCPP birth cohort reported that the mother’s sexual behaviour prior to and during pregnancy was relevant to HSV-2 seropositivity and subsequent risk of psychotic illness in the offspring (Buka et al. 2008). Rates of HSV-2 seropositivity at delivery were similar among case and control mothers who reported use of contraception or lower frequency of intercourse during pregnancy. By contrast, case mothers who reported frequent intercourse or no use of contraception were twice as likely to be seropositive than control mothers.

Prenatal maternal infection and structural and functional brain abnormalities in schizophrenia

We identified six population-based studies, mostly from the PDS and NCPP birth cohorts (Table 5) (Brown et al. 2001, 2009a, b, 2011; Ellman et al. 2009, 2010). Maternal infection or increased inflammatory cytokines during pregnancy were reported to be associated with both structural and functional brain phenotypes relevant to schizophrenia in the offspring.

Table 5.

Prenatal infection and structural and functional brain abnormalities relevant to schizophrenia

Study	Study design	Prenatal  exposure	Main findings
Brown et al. 2009	PDS cohort : structural brain  anomaly in 20 cases of DSM-IV  SSD and seven controls, 12 cases  exposed to prenatal infections	Influenza, respiratory infection, toxoplasma or  maternal/genital reproductive infection measured  by immunoassay or clinical diagnosis	CSP length, mean (S.D.) in mm by group: exposed case (n=12) 4.67  (3.65), unexposed case (n=8) 1.75 (2.87), exposed controls 4.2 (1.4),  unexposed controls 0.35 (0.7). For exposed cases, 11 of 12 (92%)  had CSP length >1.4 mm. For unexposed cases, six of eight (75%)  had CSP length of ≤1.4 mm
Ellman et al. 2010	PDS cohort : structural brain  anomaly in 17 cases of DSM-IV  SSD and eight controls	Concentration of inflammatory cytokine IL-8 in  maternal blood samples collected at second/third  trimester	Among cases, increased IL-8 was associated with significant  increase in ventricular cerebrospinal fluid, significant decrease in  cortical volume in left entorhinal cortex and right posterior  cingulate. No significant association in controls
Brown et al. 2009	PDS cohort : executive function in  24 cases of DSM-IV SSD, eight  cases exposed to prenatal  infections	Influenza during first half of pregnancy or Toxoplasma  gondii at third trimester or at delivery	Exposed cases had increased number of total error, preservative  error and non-preservative error in the Wisconsin Card Sorting  Test, and increased mean time in Part B of the Trail Making Test
Brown et al. 2011	Subset of PDS cohort : adult motor  and cognitive function in 25  cases of SSD (11 exposed and 14  unexposed)	Genital and reproductive infection during any time in  pregnancy	Exposed cases performed significantly worse on verbal memory and  fine motor coordination. In stratified analysis, these differences  were significant only in the African-American group
Ellman et al. 2009	NCPP cohort : IQ at age 7 years by  WISC in 96 cases (60  schizophrenia and 36 affective  psychoses) and 274 controls	Maternal influenza in samples collected at delivery,  exposure=IgG antibody titre >75th percentile	Exposed cases performed worse on verbal IQ (SMD 0.408, p=0.024)  and in the information subset (SMD 0.37, p=0.04) than unexposed  cases. No significant difference in any measure of IQ between  exposed and unexposed controls
Brown et al. 2001	Rubella Birth Defects Evaluation  Project, New York: follow-up of  53 cases until age 33 years for  DSM-IV SSD, cognitive function  measured in childhood and in  adolescence	Congenital rubella	20.8% developed SSD (5.7% schizophrenia), 26.4% affective  disorder, 41.5% no psychiatric illness ; cognitive decline between  childhood and adolescence in 87.5% cases of SSD v. 37.0% in non-  SSD group; degree of IQ decline greater in SSD group than non-  SSD group (average IQ decline 11 points v. 3 points respectively)

PDS, Prenatal Determinants of Schizophrenia ; SSD, schizophrenia and spectrum disorder ; NCPP, National Collaborative Perinatal Project ; WISC, Wechsler Intelligence Scale for Children ; IL-8, interleukin 8 ; IgG, immunoglobulin G; CSP, cavum septum pallucidum; S.D., standard deviation ; SMD, standardized mean difference.

Exposed cases were reported to show significant deficit in childhood and adult verbal IQ, and greater IQ decline during the pre-morbid period. Exposed cases were also reported to show increased ventricular volume, increased length of the cavum septum pallucidum, reduced cortical volume and deficits in executive function in adulthood. However, prenatal infections did not seem to affect some of these parameters, such as childhood verbal IQ or adult ventricular and cortical volume in exposed healthy controls (Ellman et al. 2009, 2010). Some of these studies did not include a healthy comparison group (Brown et al. 2009a, 2011).

Discussion

A possible role of infection and immunity in the aetiology of psychotic illness has been in the focus of neuropsychiatric research for more than a century (Menninger, 1994). Potential disruption of foetal development from prenatal maternal infections is in keeping with the neurodevelopmental theory of schizophrenia. Our review of robust epidemiological evidence suggests that exposure to infectious agents or inflammatory response during pregnancy may be associated with schizophrenic psychosis in the adult offspring.

Risk of schizophrenia in adult offspring was examined in relation to maternal non-specific bacterial, respiratory or genital and reproductive infection during pregnancy. The results differed between studies, within and across infections, with an upper range of a two- to fivefold increased risk in all of the reports reviewed. With regard to particular infectious agents, evidence for HSV-2 and T. gondii was mixed, some studies reporting up to a twofold increased risk of schizophrenic psychosis. There was some indication that exposure to influenza, and other infections in general, during early stages of gestation may be more harmful. No evidence of an association was found between prenatal HSV-1 or CMV infection and adult psychotic illness. There was some suggestion that inflammatory cytokines may mediate risk of psychotic illness associated with prenatal infection. With regard to neurodevelopment, structural and functional brain abnormalities relevant to schizophrenia were reported in exposed cases.

Methodological issues that might have influenced the findings are discussed in the following sections.

Study populations

Much of the evidence on this topic comes from two US birth cohorts from which 15 reports were included (10 from the PDS cohort and five from the NCPP cohort). We observed overlap of participants between the reports of individual infectious exposures within each cohort. Five studies that reported associations with T. gondii, influenza, respiratory infection, genital/reproductive infection and IL-8 are based on the same 71 cases of SSD from the PDS cohort (Brown et al. 2000, 2004a, b, 2005; Babulas et al. 2006). Cases in four studies of structural and functional brain abnormalities related to prenatal infection also came from the same pool (Brown et al. 2009a, b, 2011; Ellman et al. 2010). A similar overlap of cases was noted in some studies from the NCPP cohort (Buka et al. 2001a,b, 2008; Ellman et al. 2009; Xiao et al. 2009).

It is possible that the individual infections were allocated randomly among the mothers. Alternatively, exposure to several infections in the same group of case mothers would suggest that related factors, such as lifestyle, may also be important. Indeed, in the NCPP cohort, the mother’s sexual behaviour during pregnancy was found to be associated with HSV-2 infection and subsequent risk of schizophrenia in the adult offspring (Buka et al. 2008). Alternatively, or in addition, clusters of infections in a few mothers might suggest an innate susceptibility to infection that may itself be related to schizophrenia risk (discussed in biological plausibility). Regardless of the precise explanation for the associations, the validity of some the findings from the NCPP and PDS cohorts has been assisted by independent replications from large Danish datasets.

Because of the small sample size most studies could not address the issue of a sensitive period, with infection being harmful only during a particular stage of gestation. Sample sizes in studies of structural and functional brain abnormalities were also often small. None of them included any mediation model to properly explore the effects of prenatal infection on other indicators of schizophrenia.

Measurement of exposure

Selecting the criteria for exposure in serological studies can be difficult. For example, before deciding on cut-off values for antibodies to infectious agents, such as T. gondii or HSV, the prevalence of that infection in the population at the time and geographical region needs to be considered. Such considerations were given in some of the US studies and the Danish register-based studies.

IgG and IgM antibodies are both generated in response to infection (Janeway et al. 2001). IgM immunoglobulins are usually generated within a few days following systemic infection and are detectable for several months whereas IgG immunoglobulins are generated 1–3 weeks after initial infection and are detectable for several years (Buka et al. 2001a; Janeway et al. 2001). None of the studies of HSV-2 or T. gondii reported an increase in specific IgM antibodies, suggesting acute infections with these agents during pregnancy were unlikely.

Misclassification of exposure may be an issue in some studies. For example, in the Copenhagen Perinatal Cohort relatively broad categories of exposure were used (clinically diagnosed viral and bacterial infection) (Sorensen et al. 2009). Random misclassification would minimize the effect size by introducing bias results towards null.

The source of the prenatal sample may also be an issue. Studies from the NCPP and PDS cohorts used maternal serum samples that were stored at −20 °C for >20 years. Degradation of antibodies and cytokines over time is possible; however, the authors reported that there was no evidence of freeze thawing (Buka et al. 2001a; Brown et al. 2004a). The Danish register-based studies used blood samples collected from neonates 5–7 days post-partum (Mortensen et al. 2007, 2010). However, it is highly likely that antibodies against HSV-2 or T. gondii in these samples represent foetal exposure to maternal infection rather than newly acquired infection after birth. Therefore, these can be regarded as valid measures of prenatal exposure to these infections.

Measurement of outcome

Broad outcomes were used in some cohorts (Table 2). Although studies from the NCPP cohort included both affective and non-affective psychosis, the outcome in the PDS cohort was non-affective psychosis only. This made examination of any links between an infection and specific diagnostic categories difficult. Differences in outcome definition may also account for discrepant findings between some of the studies from these two cohorts. The criteria for defining outcome also varied between studies. Typically, DSM was used in the US and ICD in the European studies.

Differences in methods of case ascertainment may have introduced bias in some studies. For example, in a study of maternal genital or reproductive infection, Babulas et al. (2006) reported that risk of SSD was higher in those where diagnosis was based on review of clinical notes as opposed to face-to-face interview.

Confounding and other methodological issues

Observed associations between schizophrenia in the adult offspring and maternal infections before and after pregnancy (Nielsen et al. 2011) and exposure to several infections in the same group of case mothers (PDS and NCPP cohort) may indicate possible confounding by other factors. Individuals who get one infection may be more likely to get others, so the infections might confound each other or all be markers for a third factor.

The characteristics of women who develop infections may be different from those who do not. Paternal infection was also reported to be associated with increased risk of schizophrenia in offspring (Nielsen et al. 2011). This indicates that poor living conditions and social adversity may be important confounding factors. However, most studies adjusted for social class in their analysis. In two large Danish studies the risk of schizophrenia in offspring associated with prenatal maternal HSV-2 or general infections was attenuated (although it remained statistically significant) after taking into account family history of psychosis (Mortensen et al. 2010; Nielsen et al. 2011). Nearly half of the studies were unable to adjust for this factor. It is possible that observed associations between various prenatal infections and schizophrenia in these studies may be partly confounded by family history of psychosis. However, as only 10–15% of patients with schizophrenia have a positive family history (Gottesman & Shields, 1982), confounding by this factor is unlikely to be the sole explanation for the observed associations. Attenuation of the effect size after adjusting for family history of psychosis in the Danish studies might reflect interaction. Indeed, one study reported evidence of interaction between infection and family history of psychosis using an additive model (Clarke et al. 2009). However, unlike the multiplicative model this approach is often more likely to yield a significant interaction in a large enough sample, and may therefore be less informative (Zammit et al. 2010). Therefore, these findings require replication.

Other methodological issues include the duration of follow-up and over-matching. The duration of follow-up was short in the Danish register-based studies and therefore their findings can be generalized to only early-onset cases of schizophrenia (Mortensen et al. 2007, 2010). As most studies matched their cases and controls on gender, they were not able to explore any gender difference in the effects of prenatal infection.

Biological plausibly of association

Mechanisms by which prenatal infections interfere with foetal development and contribute to risk of adult schizophrenia may include (i) direct interference with neurodevelopment by infectious agents or antibodies, (ii) induction of autoimmunity, (iii) the involvement of mediators of acute infection, (iv) reprogramming of the hypothalamic–pituitary–adrenal (HPA) axis, (v) activation of the innate immune system and effects of inflammatory cytokines, and (vi) interaction between host and microbe genome and/or environmental factors.

Maternal infection during pregnancy with neurotrophic agents, such as T. gondii, HSV and rubella, can interfere directly with foetal neurodevelopment. An extensive body of literature suggests that congenital infections with these agents can lead to mental retardation, cerebral hypoplasia, ventriculomegaly and other brain and behavioural abnormalities in the offspring (Remington et al. 2006). This is in line with findings from birth cohort studies reviewed here that reported structural and functional brain abnormalities relevant to schizophrenia in exposed cases. This is supported by evidence from animal model studies. Studies of prenatal maternal influenza (or immune activation) using experimental mouse models have reported increased ventricular and reduced brain volume, deficits in working memory, and reduced social and exploratory behaviour in offspring (Shi et al. 2003; Meyer & Feldon, 2010). In rhesus monkeys, prenatal maternal influenza has been reported to affect neural development in the offspring, reducing grey matter throughout most of the cortex and decreasing white matter in the parietal cortex (Short et al. 2010). It has been suggested that these brain alterations are likely to be permanent, given that they were still present at the monkey equivalent of older childhood, and thus might increase the likelihood of later behavioural pathology (Short et al. 2010).

None of the studies of T. gondii or HSV reviewed here used microbial culture, and the absence of IgM antibodies argues against acute infections with these agents during pregnancy. However, there is evidence that maternal IgG antibodies can cross the placenta and directly affect the foetal brain through molecular mimicry (Wright et al. 1999).

Infections can also induce autoimmunity, which may lead to further central nervous system (CNS) damage (Albert & Inman, 1999). It has been proposed that certain prenatal infections such as influenza may increase the risk of adult psychotic illness through induction of autoimmunity (Wright & Murray, 1993). Maternal neuronal autoantibodies against cerebellar Purkinje cells have been linked with classic neurodevelopmental disorders, such as autism (Dalton et al. 2003) and dyslexia (Vincent et al. 2002), in offspring. Increased autoimmunity including autoantibodies against ion channels, such as the voltage-gated potassium channel (VGKC), or the N-methyl-d-aspartate (NMDA) receptor is well documented in schizophrenia (Parthasarathi et al. 2006; Dalmau et al. 2011; Zandi et al. 2011). Search for other autoantibodies, in addition to microbe-specific antibodies, in prenatal maternal blood samples should help to elucidate to what extent prenatal infections are associated with autoimmunity.

Developmental effects of acute infections such as influenza may be mediated by hyperthermia and foetal hypoxia (Edwards, 1968; Milunsky et al. 1992; Cannon et al. 2002). The finding of increased risk of schizophrenia for maternal analgesic use during pregnancy in the Copenhagen cohort is in line with such mechanisms (Sorensen et al. 2009). Maternal analgesic use during pregnancy has also been linked with increased risk of childhood non-clinical psychotic symptoms (Gunawardana et al. 2011). However, it is yet to be established whether these are direct effect of the analgesics or of the infection that led to the use of analgesics.

Infection can increase foetal exposure to maternal glucocorticoids by inhibiting the placental enzyme 11-beta-hydroxysteroid dehydrogenase type 2 (Edwards et al. 1993; Johnstone et al. 2005; Seckl & Holmes, 2007; Meyer & Feldon, 2010). Excess glucocorticoids can reprogramme the HPA axis in the offspring, leading to either increased basal secretion or enhanced stress-related secretion of glucocorticoids later in life (Owen et al. 2005; Weinstock, 2008). Thus, increased basal cortisol levels observed in some cases of schizophrenia (Bradley & Dinan, 2010) may be related to maternal infections during pregnancy.

Prenatal maternal infection may explain common links between chronic physical and neuropsychiatric diseases of adult life. It has been suggested that people exposed to excess levels of glucocorticoids during pregnancy may be more likely to develop hypertension, hyperglycaemia, hyperinsulinaemia and hyperactivity of the HPA axis as adults (Barker et al. 1993; Seckl & Holmes, 2007). Similarly, activation of the HPA axis has been reported to be associated with increased blood pressure, insulin resistance, glucose intolerance and hyperlipidaemia (Reynolds et al. 2001). This might explain why schizophrenia patients have a higher risk of cardiovascular disease, diabetes and impaired glucose tolerance, which persists even after taking into account effects of antipsychotic drugs and lifestyle factors (Bushe & Holt, 2004; Curkendall et al. 2004).

As increased risk of psychosis has been observed for a diverse range of prenatal infections, it has been proposed that they may share a common pathway to exert pathogenic effects on the foetal brain. One such mechanism involves activation of the innate immune system and release of proinflammatory cytokines (Meyer et al. 2009). Reports from the NCPP and PDS cohorts suggest inflammatory cytokines may mediate risk of psychotic illness associated with prenatal infection (Buka et al. 2001b; Brown et al. 2004b).

A role of prenatal immune activation in schizophrenia is also supported by evidence from animal model studies (reviewed by Meyer & Feldon, 2010). Mouse models using pregnant females have included simulated viral or bacterial infection or direct injection with a cytokine (IL-6). These have been reported to produce intermediate phenotypes related to schizophrenia in the adult offspring. Some of these phenotypes, such as deficits in sensory gating and abnormal latent inhibition, have been shown to be reversible by treatment with clozapine, an antipsychotic (Smith et al. 2007). Cytokines can be neurotrophic and also neurotoxic depending on the developmental stage and condition (Garver et al. 2003). Elevated levels of cytokines during pregnancy have been reported to be associated with reduced cortical neuronal survival and reduced grey- and white-matter volume in animal studies (Meyer & Feldon, 2010; Short et al. 2010). Prenatal immune activation in animals has also been linked with structural and functional alterations in the mesocorticolimbic dopaminergic system long before the onset of the full spectrum of psychosis-associated behavioural and cognitive abnormalities in adulthood (Meyer & Feldon, 2009).

Characteristics of both the microbial and human genome are likely to determine the pathological response to an infection to a large extent (Yolken & Torrey, 2008). For example, one study reported that infection with T. gondii type I, specifically, out of all three genotypes, was associated with increased risk of psychosis (Xiao et al. 2009). However, genetic determinants of human (host) response to infection include receptors, transcription factors, cytokines and other components of innate immunity (Yolken & Torrey, 2008).

It has been suggested that some genetic vulnerability to schizophrenia may be explained by genetic vulnerability to infection (Shi et al. 2009; Stefansson et al. 2009; Nielsen et al. 2011). Variations in the gene encoding IL-10 are reported to be associated with susceptibility to both CMV infection and schizophrenia (Bocchio Chiavetto et al. 2002; Hurme et al. 2003). More recently, a genome-wide association study (GWAS) of schizophrenia reported an association with a cytokine receptor gene, colony stimulating factor, receptor 2 alpha (CSF2Rα; Lencz et al. 2007). This gene is responsible for regulating important components of the innate immune system such as granulocytes and macrophages, which are the primary source of proinflammatory cytokines.

GWAS have also reported significant associations between schizophrenia and markers close to the major histocompatibility complex (MHC) region on chromosome 6 (Shi et al. 2009; Stefansson et al. 2009). This region includes several immunity-related genes and a histone gene cluster relevant to gene expression. This region contains genes involved in brain development, memory and cognition (Stefansson et al. 2009). Thus, increased risk of adult schizophrenia in the offspring for both paternal and maternal infection, as observed in a Danish sample (Nielsen et al. 2011), may reflect high genetic susceptibility to severe infection in future cases of schizophrenia. It is also possible that prenatal infection, by affecting gene expression or in the presence of pre-existing genetic vulnerabilities, can lead to a distinct or pathological immune response. This, in turn, may lead to CNS alterations, making these individuals susceptible to developing psychotic illness later in life.

Directions for future research

Studies using animal models could address whether or not there is a ‘sensitive period’ for vulnerability, an issue that must be resolved to understand the under-lying biological mechanisms and to consider therapeutic interventions.

Duration of follow-up should cover at least the age of peak incidence of schizophrenia to make findings more generalizable. Effects of prenatal infections need to be examined in relation to specific psychopathology rather than broad diagnostic categories of schizophrenia or non-affective psychosis. Relationships between prenatal infection and well-established intermediate phenotypes of adult schizophrenia, such as childhood motor and cognitive development, should be explored using larger samples. The use of mediation models should be used to elucidate potential effects of prenatal infection on other antecedents of schizophrenia. Future studies should use triangulation of genetic information in addition to inflammatory cytokines and other immunological markers to elucidate effects of prenatal and early life infections on neurodevelopment and adult psychotic illness. Future investigations should also address to what extent genetic vulnerability to schizophrenia may be explained by genetic vulnerability to infection. Indeed, a new generation of general population birth cohorts, including the Norwegian Mother and Child Cohort Study (MoBa; Magnus et al. 2006), the Danish National Birth cohort (Olsen et al. 2001) and the Avon Longitudinal Study of Parents and Children (ALSPAC) cohort (Golding et al. 2001) are currently conducting similar studies of neurodevelopmental outcomes. These cohorts have in-depth biological and other data and (for Norway and Denmark) large enough sample sizes to address issues such as mediation and mechanisms.

Strengths and limitations of this review

Our review is based on rigorous examination of robust population-based studies. Included studies reliably measured exposure at the individual level and defined outcome by contemporary ICD or DSM directives. Cases of psychosis were collected from general population registers or by a two-stage process (screening followed by a diagnostic interview). Individual studies also accounted for several confounding factors in design and analysis. These measures should increase both internal and external validity of the evidence. We included a variety of studies, for example those examining serological and clinically diagnosed infections, that used various epidemiological designs. We were not able to combine results of individual studies in a meta-analysis or carry out formal tests for publication bias. Meta-analyses were deemed unsuitable because of significant variation among studies on several important parameters. Nevertheless, to our knowledge this is the most rigorously conducted independent appraisal of the population-based studies published on this topic to date.