Serologic Evidence of Prenatal Influenza in the Etiology of Schizophrenia FREE

Many studies suggest that prenatal exposure to influenza is associatedwith adult schizophrenia. In the first study of its kind, Mednick et al¹ demonstrated an increased risk of schizophrenia amongFinnish individuals who had been in the second trimester of gestation duringthe 1957 influenza A2 epidemic. Thus far, among 26 investigations that soughtto replicate the finding, about half reported positive associations, yet someof the more rigorously designed studies have shown no evidence of a relationship.²

This divergence of findings is likely explained, at least partly, byimprecise measurement of the exposure in the populations studied. These studiesdefined "exposure" on the basis of the dates of influenza epidemics in thepopulation, or on maternal recall of influenza infection after pregnancy.It has been argued that this question will remain unresolved until more sophisticatedmethods of documenting the exposure are applied.^3- 4

To more definitively test this hypothesis, we conducted assays for influenzaantibody in serum samples drawn from pregnant women whose offspring developedschizophrenia and in a matched comparison group of pregnant women whose offspringdid not develop schizophrenia. These assays were conducted in the cohort ofthe Prenatal Determinants of Schizophrenia (PDS) study, which featured a largebirth cohort of well-characterized pregnancies with archived maternal serumspecimens prospectively collected throughout pregnancy, and rigorous diagnosticassessments of the schizophrenia outcome.⁵ Theserum samples were tested for antibodies that were specific to the influenzastrains previously documented as having been prevalent in the population during1959 through 1966, the years of the pregnancies. To our knowledge, no previousstudy of influenza and schizophrenia has used serologic methods to documentinfluenza exposure.

The PDS study was fully described in a previous publication⁵ and will therefore only be briefly summarized here.The mothers of the cohort members in the PDS study were enrolled in the ChildHealth and Development Study (CHDS),^6- 7 whichwas conducted from 1959 through 1966. During that period, the CHDS recruitednearly every pregnant woman who received obstetric care from the Kaiser FoundationHealth Plan (KFHP) in Alameda County, California. Hence, all of the offspringof these women were automatically enrolled in KFHP. All subjects assessedin the PDS study provided written informed consent for human investigation.The study protocol was approved by the institutional review boards of theNew York State Psychiatric Institute, New York, and the Kaiser FoundationResearch Institute, Oakland, Calif.

The PDS study cohort consisted of the subsample of 12 094 productsof live births who were members of KFHP from January 1, 1981, through December31, 1997. These dates correspond to the period of case ascertainment, whichbegan in 1981 because computerized records did not exist in KFHP before thatdate. We have previously demonstrated that subjects who remained in KFHP andsubjects lost to follow-up did not differ on a wide range of maternal andpaternal characteristics^5,8 andthat the vast majority of subjects who left KFHP did so before the age of10 years,⁵ suggesting that early manifestationsof schizophrenia did not influence the likelihood of loss to follow-up.

A special feature of the CHDS was the collection of maternal serum samplesduring pregnancy. The serum samples from these pregnancies were frozen immediatelyat −20°C and have been archived at that temperature or below ina single serum repository. All specimens were uniformly handled in accordancewith a strict protocol. Serum samples were available from mid- to late gestationin virtually all pregnancies, and from early gestation in about half of thepregnancies.

The main outcome was schizophrenia and other schizophrenia spectrumdisorders (SSD), defined as schizophrenia, schizoaffective disorder, delusionaldisorder, psychotic disorder not otherwise specified, and schizotypal personalitydisorder, based on previous studies.⁹ Caseascertainment and screening were accomplished by means of a computerized recordlinkage between the CHDS and KFHP identifiers from inpatient, outpatient,and pharmacy registries. Subjects from the hospital registry were screenedfor potential SSD on the basis of their having been given registry diagnosesof codes 295 to 299 in the International Classificationof Diseases, Ninth Revision, and were subsequently judged to be potentialSSD cases after psychiatrist review of all psychiatric and medical records.For the outpatient registry, cases were considered positive on screening ifthey had diagnosis codes of 295, 297, 298, or 299. For the pharmacy registry,subjects screened positive if they received treatment with antipsychotic medications.Among subjects who screened positive for potential SSD (n = 183), 13 had died.Among the 170 remaining potential cases, 146 (86%) were successfully contactedto schedule a diagnostic interview.

Potential cases were administered the Diagnostic Interview for GeneticStudies¹⁰ by clinicians with at least a master'sdegree in a mental health field who were trained for reliability (A.S.B. andE.S.S.). Psychiatric diagnoses, by DSM-IV criteria,were made by consensus of 3 experienced research psychiatrists on the basisof the Diagnostic Interview for Genetic Studies narrative, medical records,and discussions between the interviewer and diagnosticians. Of the 146 contactedpotential cases, 107 (73%) completed the Diagnostic Interview for GeneticStudies. For the 76 potential cases who were not interviewed, medical chartreviews by experienced clinicians were conducted. A research psychiatristreviewed and confirmed all chart review diagnoses. These procedures yielded71 total SSD cases, 44 of whom had received the Diagnostic Interview for GeneticStudies, and 27 of whom were diagnosed by chart review. Among these 71 SSDcases, 64 had available prenatal serum. The diagnoses of these cases wereas follows: schizophrenia (n = 38), schizoaffective disorder (n = 15), delusionaldisorder (n = 1), schizotypal personality disorder (n = 5), and other schizophreniaspectrum psychosis (n = 5) (the last category included subjects diagnosedby chart review who met criteria for either schizophrenia or schizoaffectivedisorder, but in whom there was insufficient information to definitively specifyeither condition). Therefore, 58 (91%) of the 64 cases with prenatal serumhad either schizophrenia or schizoaffective disorder.

To select eligible controls, we first excluded the 71 SSD cases alreadydiagnosed and 318 subjects with major psychiatric disorders other than SSD.Up to 8 matched controls were selected for each case. Controls were matchedto cases on 5 characteristics: membership in KFHP at the time of case ascertainment(ie, the time of first treatment for SSD), date of birth (±28 days),sex, number of maternal blood samples drawn during the index pregnancy, andnumber of weeks after the last menstrual period (referred to as the time post-LMP)of the first maternal blood draw during the index pregnancy (±4 weeks).Matching for KFHP membership ensured that the controls for each case wererepresentative of the population at risk at the time of case ascertainment.For this purpose, we used the KFHP membership registry, which enabled us todefine the cohort members remaining in KFHP at the time that each case wasfirst treated. Birth date was included as a matching factor to ensure thatany degradation in the serum samples over time would be comparable betweencases and controls. Controls were matched on sex. Controls were also matchedby the number and timing of maternal blood draws to ensure sufficient andcomparable serologic data for cases and matched controls. Further detailson control selection in the PDS study were reported by Susser et al.⁵ Because of the need to conserve serum, we randomlyselected 2 from the maximum 8 matched controls per case for the serologicanalysis.

The antigens acquired for testing of serum subsets from the study populationwere those of the prevalent influenza strains from 1959 through 1966 in northernCalifornia. These antigens include A/H2N2/Japan/57, A/H2N2/Japan/62, A/H2N2/Taiwan/64,and B/Massachusetts/66.¹¹ The hemagglutinationinhibition (HAI) method, following Good Laboratory Practice standards as previouslydescribed,^12- 13 was used to assaythe serum for influenza antibody. All available serum samples from each subjectwere tested in duplicate on the same V-bottom plate in serial 2-fold dilutions(1:5, 1:10, 1:20, etc) after receptor-destroying enzyme treatment. The HAIduplicates that differed by more than a factor of 2 were repeated for thatsample. Serum samples were tested with chicken red blood cells from 1 of 2standardized chickens used by the Eastern Virginia Medical School laboratory,Norfolk. The HAI-determined titer was the greatest dilution of serum thatcompletely inhibited agglutination of chicken red blood cells by the testvirus. The test virus or viruses used for each assay were those of the specificcirculating strains to each subtype that were prevalent during the periodduring which each serum sample was collected. If more than one influenza strainwas prevalent in a given period, then separate HAI assays were conducted foreach strain.

Antibodies are known to be stable in stored frozen serum for long intervals.To verify the viability of measurements for influenza antibody in the serumspecimens, we first conducted serologic assays, using the procedure specifiedin the previous section, in 51 CHDS pregnancies giving rise to neither casesnor matched controls in the present study, and who had serum available fromeach trimester. This permitted us to document the occurrence of seroconversion(ie, a 4-fold rise in influenza antibody titer in serial samples), which isdiagnostic of influenza infection.¹⁴ Pregnanciesthat overlapped with known influenza epidemics evidenced a 2-fold increasein prevalence of seroconversion for influenza, compared with pregnancies thatdid not overlap with influenza epidemics. This provided further evidence ofthe feasibility of detecting influenza antibody in these archived serum samplesby means of the HAI technique.

Although serum samples were obtained during gestation in virtually allpregnancies of the CHDS cohort, seroconversion could not be documented inmost pregnancies because of an insufficient number with available serial samplesand inconsistencies with regard to the gestational periods during which thesesamples were drawn. Thus, we developed a method for determining influenzainfection status during pregnancy with the use of only a single antibody titer.For this purpose, we assessed the validity of an antibody titer thresholdvalue (an antibody titer equal to or greater than a specified cutoff value)as a proxy for influenza infection, as documented by seroconversion (4-foldrise in influenza antibody titer). Maternal serum from the noncase, noncontrolsample described in the preceding paragraph was used for this purpose. Thevalidity of a series of antibody titer threshold values (ranging from ≥1:10to ≥1:80) as proxies for influenza infection were tested (a titer of 1:5indicated no infection, and titers of ≥1:160 were very rare). We demonstratedthe following validity parameters for a strain-specific influenza antibodytiter of 1:20 or greater in any serum sample during the pregnancy: sensitivity,100% (7/7); specificity, 95% (42/44); positive predictive value, 78% (7/9);and negative predictive value, 100% (42/42). This supported the use of the1:20 or greater antibody threshold titer in a single serum sample as an adequateproxy for influenza exposure during pregnancy.

The primary outcome measure was case or control status (ie, whethera subject was classified as having SSD). The primary exposure measure wasinfluenza infection, which, in accordance with the results of our validitystudy, was defined as the first occurrence during pregnancy of an influenzaantibody titer of 1:20 or greater. Antibody titers of 1:20 or greater occurringat any point in gestation after the first such titer for a given subject weredesignated as negative (unexposed) in the analysis. This criterion was justifiedfor 3 reasons. First, each subject can be infected only once by a given influenzastrain within any single year. Second, influenza is an acute infection, rarelylasting longer than 1 week,¹⁴ and the serumsamples were always drawn at least 2 weeks apart from one another. Third,positive antibody titers to 2 different influenza strains at different pointsin time during a single pregnancy occurred rarely in our sample.

The analyses were conducted separately by trimester (see "Methods ofStatistical Analysis" for a description of the analyses). We defined trimester1 as 0 to 97 days post-LMP, but because of the delay in recognition of pregnancy,the actual range of first-trimester gestational days in which serum sampleswere available was 46 to 97 days post-LMP (or the latter half of the firsttrimester). We defined trimester 2 as 98 to 187 days post-LMP, and trimester3 as 188 days post-LMP until 3 days after birth. Variables selected a priorias potential confounders included maternal age, paternal age, maternal education,and maternal ethnicity; all of these demographic data were acquired at ornear the time of birth of the child.

Matching via the nested case-control design ensured that cases and theircorresponding controls were followed up for equal periods from birth untilthe date of first treatment for SSD (or until time of last follow-up for controls).Thus, having corrected by design for duration of follow-up, we were able toapply conditional methods for binary data to compare cases and controls oninfluenza exposure. As the first step in evaluating the potential associationbetween influenza exposure in each period of pregnancy and SSD risk, we appliedthe Mantel-Haenszel methods for stratified data.¹⁵ Thisanalysis designated each matched set as its own stratum, within which case-controlstatus (SSD vs not) was cross-classified by the presence or absence of influenzainfection. Separate analyses were conducted for each period of gestation.The Mantel-Haenszel methods yield an estimated odds ratio relating exposureand SSD and a test of the significance of the observed association. Subsequentanalyses were conducted by means of conditional logistic regression¹⁶ to allow for control of other potential confoundingvariables (eg, maternal education). This model was conditioned on an identificationvariable that uniquely identified matched stratum membership. Like the Mantel-Haenszelanalysis, the conditional logistic model provides an estimate of the oddsratio to measure the strength of the effect, as well as a significance testto formally assess the association between prenatal influenza exposure andSSD risk.

The population attributable risk was calculated on the basis of thefollowing formula:

(RR − 1)P₁/RR,

where RR indicates relative rate and P₁ is the proportionof exposed cases.¹⁷ In accordance with theusual practice for case-control studies, we substituted the adjusted oddsratio for the RR. This measure provides a crude indication of the proportionof cases occurring in the studied population that might be prevented by removalof the exposure, if the exposure is indeed a cause.

The distribution of potential confounding characteristics was comparedbetween the matched samples of SSD cases and controls (Table 1). The only characteristics that differed appreciably werematernal education (larger proportion of cases with less than a high schooleducation) and paternal age (increased in cases, as reported in a previouspublication).¹⁸

Among subjects tested in the first trimester of pregnancy, 5 (25%) ofSSD cases were influenza-exposed, compared with 4 (11%) of controls. For thesecond and third trimesters, the proportions of influenza-exposed cases andcontrols were similar to one another (Table2).

The Mantel-Haenszel analysis, controlling for matched set membership,demonstrated that influenza exposure in the first trimester of pregnancy wasassociated with a 7-fold increased risk of SSD, although the result did notachieve statistical significance (P = .08). For influenzaexposure in the second trimester of pregnancy, the risk of SSD was not appreciablyelevated in cases as compared with controls. There was also no increase inrisk of SSD after third-trimester influenza exposure (Table 3).

In the conditional logistic regression analysis, adjustment for maternalage, paternal age, maternal education, or maternal ethnicity had no appreciableimpact on the estimated effect of first-trimester influenza exposure on SSDrisk (results available on request).

In a further analysis, we sought to assess whether the association betweenfirst-trimester influenza and SSD risk extended into the first part of thesecond trimester. Our rationale for this analysis was as follows. First, asnoted previously (see "Data Analysis: Key Analytic Variables"), the serumsamples from the first trimester were obtained during the latter half of thatgestational period (thus bordering on the beginning of the second trimester).Second, the divisions between trimesters are defined arbitrarily, rather thanbeing based on biological mechanisms. This suggests that alternative definitionsof pregnancy intervals may yield valuable additional information. Third, althoughwe did not observe an increased risk of SSD in the second trimester analyzedas a whole (see Table 3), previousecologic studies demonstrated second-trimester specificity (see the introduction).

Hence, in the additional analysis, we divided pregnancy into 2 periodsof exposure: the first half and the second half. The exposed period for thefirst half began on day 0 (as discussed earlier, however, the first availablesample was drawn on day 46) and ended on day 142 post-LMP (the midpoint ofpregnancy). In effect, this period consists of the latter half of the firsttrimester and the first half of the second trimester. The second half of pregnancywas defined as the period from day 143 until 3 days after the terminationof pregnancy.

We then examined the association between SSD risk and influenza forsubjects exposed during each of these 2 periods. Using this definition ofexposure, we found that 9 (21%) of SSD cases, compared with 7 (9%) of controls,had been exposed to influenza in the first half of pregnancy (see Table 2). In the Mantel-Haenszel analysis,influenza exposure during the first half of pregnancy was associated witha 3-fold increased risk of SSD, which was at the margin of statistical significance(P = .052) (see Table 3). The findings were not appreciably altered after adjustmentfor maternal age, paternal age, maternal education, and maternal ethnicity(results available on request). There was no association between SSD riskand influenza exposure in the second half of pregnancy.

The population attributable risk associated with influenza exposurewas 21% for the first trimester and 14% for the first half of pregnancy.

This study was the first, to our knowledge, to use maternal influenzaantibody levels in individual pregnancies to examine the relationship betweenprenatal influenza exposure and schizophrenia. The data indicate that therisk of schizophrenia was increased by a factor of 7 after serologically documentedinfluenza exposure during the first trimester of pregnancy. Prospective acquisitionof the serum samples in a well-characterized, continuously monitored birthcohort and the use of a face-to-face psychiatric diagnostic assessment lendcredence to this result. It should be noted, however, that this finding didnot achieve statistical significance (P = .08) andis based on a small sample.

As noted in the introduction, nearly all previous positive studies ofinfluenza and schizophrenia were specific to exposure during the second trimester.² An additional analysis permitted us to further examinethe compatibility between our findings and those of the previous studies.This analysis showed that exposure to influenza from approximately the midpointof the first trimester to the midpoint of the second trimester increased therisk of schizophrenia by a factor of 3. Although this finding fell slightlyshort of statistical significance (P = .052), itsuggests significant overlap between the present and previous studies withregard to the gestational periods of influenza exposure that were associatedwith an increased risk of schizophrenia.

The fact that these gestational periods were not identical might beexplained by differences in measuring the timing of influenza infection duringpregnancy. In previous studies, the timing of exposure was defined by thedates of the peak periods of influenza epidemics in the population, not serologicassessment of infection in individual pregnancies. In addition, previous investigationsused the date of birth of the offspring to define the trimester or month ofpregnancy; in contrast, we had complete information on the date of the lastmenstrual period, permitting a more precise estimate of gestational age. Moreover,the influenza strains examined varied between the present and previous studies.Many of the previous findings were based on the 1957 influenza A2 epidemic.² In contrast, our study included not only the 1957influenza A2 strain, but 3 additional influenza strains that differed significantlyin antigenicity from the 1957 influenza A2 and other influenza strains examinedin previous studies.

In a recent study, the offspring of pregnant mice infected with influenzavirus on gestational day 9.5 had altered exploratory behavior, decreased contactwith novel objects, and deficits in prepulse inhibition to acoustic startle,the last of which was corrected by antipsychotic medications.¹⁹ Eachof these findings is analogous to abnormalities demonstrated in schizophrenia.In another study, the offspring of pregnant mice infected with influenza atgestational day 9 evidenced significant reductions at birth in reelin-positiveCajal-Retzius cells in the cortex and hippocampus, and diminished areas ofthese brain regions,²⁰ consistent with postmortemstudies of schizophrenia.^21- 23

Since influenza is believed to only rarely cross the placenta, an indirecteffect on fetal brain development is the most plausible pathogenic mechanismlinking it to an increased risk of schizophrenia.²⁴ Onesuch mechanism, previously considered, is that maternal IgG antibodies elicitedby influenza traverse the placenta and cross-react with fetal brain antigensby molecular mimicry, thereby disturbing fetal brain development and increasingvulnerability to schizophrenia.²⁵ Another potentialimmunologic mediator is an influenza-induced excess of maternal cytokines,which may damage the developing fetal brain.²⁶ Thishypothesis is based in part on evidence that elevated levels of cytokinescause neurodevelopmental damage, such as periventricular leukomalacia.^27- 28 In addition, increased cytokinesin cord blood have been reported in neonates who developed cerebral palsyand mental retardation in childhood.²⁹ Finally,in the preclinical study discussed at the beginning of this section,¹⁹ prenatal administration of synthetic double-strandedpoly(I:C), which evokes a strong immune response including cytokine elevations,resulted in prepulse inhibition deficits similar to those found in mice prenatallyinfected with influenza.

Recent evidence, however, may argue instead for a direct effect of influenzainfection on the fetal brain. Aronsson et al³⁰ reportedthat mice that were prenatally infected with human influenza (A/WSN/33 strain)had influenza viral RNA in the brain. The viral RNA persisted for at least90 postnatal days.

Other possible mediating factors²⁴ includehyperthermia, which is teratogenic to animals³¹ andpossibly also to humans³²; fetal hypoxia, whichhas been previously associated with schizophrenia^3,33;and prescribed over-the-counter influenza remedies, including aspirin, whichmay cause central nervous system anomalies.³⁴

Since influenza is common in the population, we used our data to calculatethe population risk for schizophrenia attributable to influenza in early tomidpregnancy. Our data suggest the possibility that up to 14% of schizophreniacases would not have occurred if influenza exposure during early to midpregnancyhad been prevented.

These findings add to a body of work suggesting a relationship betweenin utero exposure to infectious agents and risk of adult schizophrenia.³⁵ In a previous study on the National CollaborativePerinatal Project that used prenatal serum specimens, elevated maternal IgGantibody levels to herpes simplex type 2 virus were found in pregnancies givingrise to adults with psychosis, compared with matched controls.³⁶ Brownet al³⁷ reported that more than 20% of subjectswho were clinically and serologically documented with in utero exposure torubella were diagnosed as having schizophrenia and other SSDs. Maternal upperrespiratory infections have been associated with an increased risk of schizophreniain the offspring.³⁸ Finally, ecologic studieshave demonstrated associations between schizophrenia and prenatal exposureto polio, varicella-zoster, and measles.³⁵ Thesefindings suggest that common downstream effects from several infectious agentson neuronal function may be relevant to the etiopathogenesis of schizophrenia.

We used a proxy measure of influenza exposure during pregnancy. As describedearlier (see "Methods: Validity Study"), influenza during a defined periodis classically assessed by the demonstration of seroconversion (ie, a 4-foldrise in antibody titer). In our primary analysis, we approximated influenzainfection by using antibody titers measured at single points in time. Thismethod, however, was well validated with the use of seroconversion as the"gold standard" in serum samples from our cohort.

It is also worth noting that if elevated influenza-specific IgG antibodyis responsible for mediating the effects of influenza on the risk of schizophrenia,as previously postulated,²⁵ then antibody titersmay provide a more direct and accurate exposure measure than seroconversion.To further address this question, we conducted an additional analysis in whichexposure in each period of pregnancy was defined as an elevated antibody titer(≥1:20) during that period, irrespective of whether within-subject titersfrom samples drawn earlier in pregnancy were elevated. (In the primary analysis,infection was determined on the basis of the first occurrence during pregnancyof an elevated antibody titer, regardless of whether titers drawn later inpregnancy were elevated; see "Data Analysis: Key Analytic Variables" in the"Methods" section.) The respective magnitudes of associations between elevatedinfluenza antibody titers and SSD for each period of pregnancy were similarto those of the primary analysis, suggesting that one cannot differentiatebetween the 2 effects (results available on request).

The serum samples used in the present study had been frozen for morethan 30 years, raising the issue of the stability of the antibodies duringthat time. Several factors argue against compromised protein stability asa factor in this study. First, careful visual inspection of our samples showedlittle evidence of previous freeze-thawing, a major cause of protein breakdown,or of evaporation, which might cause a spurious elevation of antibody levels.Second, in consultation with the lead CHDS investigators, we verified thatthroughout the entire storage period, these samples had been carefully anduniformly handled, and special efforts had been expended to ensure that theyconstantly remained at a temperature of −20°C or lower, which considerablyprotects against protein breakdown. Third, we demonstrated that seroconversion,as measured in our serum specimens, occurred twice as often in specimens drawnduring influenza epidemics as in those from nonepidemic periods (see "Methods:Validity Study"). Fourth, controls were matched to cases on date of birthand gestational timing, and the maternal specimens of cases and controls wereuniformly handled and stored, suggesting that these factors should not havebiased the observed associations.

Our study also did not include data on family history of schizophrenia,which would have permitted the adjustment for possible confounding by thisfactor, and the examination of interaction between prenatal influenza exposureand genetic susceptibility to schizophrenia. We are presently conducting familyhistory and molecular genetic assessments of our cases and matched controlsto further address these questions.

Finally, because our sample sizes were small to modest, and the findingsdid not achieve statistical significance, independent replications are needed.

Using serologic methods, we demonstrated that exposure to influenzaduring early to midpregnancy may play a role in the etiology of schizophrenia.If confirmed by other studies, our findings may ultimately have implicationsfor ameliorating, or possibly preventing, a significant portion of schizophreniacases, for example, by administration of influenza vaccine to women of reproductiveage. Although the precise mechanisms need to be delineated, it may be worthconsidering the question of routine vaccination of nonpregnant women, giventhe possibility that the antibody response to influenza, rather than directinfection, may be responsible for the observed increase in risk of schizophrenia.A further implication of the present findings is that they may stimulate furtherwork on the pathogenesis of schizophrenia.

Correspondence: Alan S. Brown, MD, College of Physicians and Surgeonsof Columbia University, New York State Psychiatric Institute, 1051 RiversideDr, Unit 2, New York, NY 10032 (asb11@columbia.edu).

Submitted for publication January 22, 2003; final revision receivedJune 25, 2003; accepted July 2, 2003.

This study was supported by an Independent Investigator Award and YoungInvestigator Awards from the National Alliance for Research on Schizophreniaand Depression, Great Neck, NY (Dr Brown); grants R01 MH63264 (Dr Brown),K08 MH 01206 (Dr Brown) and MH59342 and R01 MH53147 (Dr Susser) from the NationalInstitute of Mental Health, Bethesda, Md; the Lieber Center for SchizophreniaResearch, Great Neck; the Theodore and Vada Stanley Foundation, Washington,DC (Dr Susser); and grant AG00834 from the National Institutes of Health,Bethesda (Dr Gravenstein).

This study was presented in part at the Annual Meeting of the AmericanCollege of Neuropsychopharmacology, December 10, 2001, Waikoloa, Hawaii; theInternational Congress on Schizophrenia Research, April 2, 2003, ColoradoSprings, Colo; and the Annual Meeting of the Society of Biological Psychiatry,May 20, 2003, San Francisco, Calif.

We acknowledge Barbara van den Berg, MD, Barbara Cohn, PhD, RobertaChristianson, Jerianne Myers, Barbara Stebler, Dolores Malaspina, MD, JillHarkavy-Friedman, PhD, Megan Perrin, Vijoy Varma, MD, Ramin Mojtabai, PhD,and Nancy Sohler, PhD, for their contributions to this work.