Least squares mean hippocampal volumes across risk groups. CC indicatesconcordant; MZ, monozygotic; DC, discordant; DZ, dizygotic. Error bars representSEM.
Theo G. M. van Erp, Peter A. Saleh, Matti Huttunen, Jouko Lönnqvist, Jaakko Kaprio, Oili Salonen, Leena Valanne, Veli-Pekka Poutanen, Carl-Gustav Standertskjöld-Nordenstam, Tyrone D. Cannon. Hippocampal Volumes in Schizophrenic Twins. Arch Gen Psychiatry. 2004;61(4):346–353. doi:10.1001/archpsyc.61.4.346
The effects of genes and environment on brain abnormalities in schizophrenia
To examine the contributions of genes and environment to hippocampal
volume reduction in schizophrenia.
Population-based twin cohort study.
Seven monozygotic (MZ) twin pairs concordant for schizophrenia and 16
MZ and 32 dizygotic (DZ) twin pairs discordant for schizophrenia, ascertained
so as to be representative of all such probands in a Finnish birth cohort,
along with 28 MZ and 26 DZ healthy comparison twin pairs without a family
history of psychosis.
Main Outcome Measures
Hippocampal volume measurements taken from high-resolution magnetic
Hippocampal volumes of probands were smaller than those of their nonschizophrenic
MZ and DZ co-twins and healthy twins. Hippocampal volumes of probands' non-ill
co-twins were smaller than those of healthy twins, but those of non-ill MZ
and DZ co-twins of schizophrenic patients were similar. The intraclass correlations
for hippocampal volumes among healthy and discordant MZ pairs were larger
than those among the respective DZ pairs. The intraclass correlation for healthy
MZ pairs was larger than that for discordant MZ pairs, and the variance component
estimate for additive genetic effects was lower in discordant twins than in
Although hippocampal volume in healthy individuals is largely affected
by genetic factors, it is subject to substantially greater modulation by environmental
factors in schizophrenic patients and their relatives. The results are discussed
in view of assumptions underlying classic twin methods.
Hippocampal volume reduction is a robust correlate of schizophrenia,1 but the etiology of the deficit remains unclear.2 Several studies3- 11 ofchildren and siblings of schizophrenic patients have suggested that geneticpredisposition and environmental risk factors may contribute to hippocampalvolume reduction in schizophrenia. Twin studies can be used to separate geneticfrom shared and unique environmental effects. Genetic effects would be suggestedby the observation of smaller hippocampal volumes in the healthy monozygotic(MZ) compared with dizygotic (DZ) co-twins from pairs discordant for schizophreniawho share on average 100% and 50% of their genes with affected individuals,respectively, and by higher intraclass correlations (ICCs) for MZ than forDZ pairs discordant for schizophrenia. Shared environmental effects wouldbe suggested by equivalent hippocampal volumes in the healthy MZ comparedwith DZ co-twins of schizophrenic patients and by equivalent ICCs for hippocampalvolume in MZ and DZ pairs discordant for schizophrenia. Unique environmental(or possibly epigenetic) effects would be implicated if probands were observedto have smaller hippocampal volumes than their healthy MZ co-twins and ifthe ICCs were lower in discordant twin pairs than in healthy twin pairs. Ifsample sizes are sufficient, variance components indicating the proportionof the variance in hippocampal volumes explained by additive genetic factorsand by shared and unique environmental factors can be calculated.
To date, 7 articles and 1 abstract from 4 independent twin samples examininghippocampal volumes in schizophrenia have been published.12- 19 Theirresults are equivocal on the question of etiology, with some studies12,14,15 implicating a majorrole for environmental factors and other studies16,17 suggestingthat genetic, shared environmental, or both factors are involved. Furthermore,Weinberger and colleagues18 reported no differencein the hippocampal volumes of probands from concordant and discordant pairs,suggesting that a similar etiology may underlie hippocampal volume reductionsin both proband types.
Neuropsychological studies20- 22 haveshown reduced performance in probands compared with their healthy MZ co-twinson tests thought to be sensitive to temporal lobe functioning, suggestingthat medial temporal lobe structures are at least in part affected by environmentalfactors. One such study23 has even shown strongassociations between difference (proband–co-twin) left hippocampal volumesand difference verbal memory test scores of MZ probands discordant for schizophrenia.Cannon and colleagues22 showed reduced ICCsfor verbal episodic memory in discordant compared with healthy twin pairs.In that study, the ICCs for discordant MZ pairs were larger than those forDZ pairs, suggesting that genetic (or shared environmental) factors may alsoplay a role.
We have now completed magnetic resonance imaging and quantitation onour full series of 16 MZ and 32 DZ twin pairs discordant for schizophrenia,7 MZ pairs concordant for schizophrenia, and matched groups of healthy twins(28 MZ and 26 DZ pairs). Based on the findings reviewed previously herein,we hypothesize that (1) the mean hippocampal volumes of probands from pairsconcordant and discordant for schizophrenia will be similar; (2) the hippocampalvolumes of healthy MZ co-twins of schizophrenic patients will be smaller thanthose of DZ co-twins, whose volumes will be smaller than those of healthytwins, and the ICCs for hippocampal volume in MZ discordant pairs will belarger than those in DZ discordant pairs (implicating a genetic contribution);and (3) the mean hippocampal volumes of probands will be smaller than thoseof their MZ co-twins, the ICCs for hippocampal volume in twin pairs discordantfor schizophrenia will be lower than those in healthy twin pairs and pairsconcordant for schizophrenia, and the variance component for additive geneswill be higher in healthy twins than in twin pairs discordant for schizophrenia(consistent with a larger unique environmental effect in schizophrenia).
The study protocol was reviewed and approved by the institutional reviewboards of the University of California (Los Angeles), the University of Pennsylvania(Philadelphia), and the National Public Health Institute of Finland (Helsinki),and all participants signed institutional review board–approved informedconsent forms.
Participants were drawn from a twin cohort consisting of all same-sextwins born in Finland between 1940 and 1957 (N = 9562 pairs) identified throughthe Finnish national twin registry. Questionnaire-based classification identified2495 MZ twins, 5378 DZ twins, and 1689 twins of unknown zygosity.24 This cohort was screened, for 1969 to 1991, for ahistory of hospitalization, medicine prescription, and work disability dueto psychiatric indication in 3 national computerized databases: the HospitalDischarge Register, the Free Medicine Register, and the Pension Register.25 These searches identified 348 index twin pairs withat least 1 co-twin with a diagnosis of schizophrenia or schizoaffective disorderand 9214 healthy pairs with no schizophrenia diagnosis in either co-twin accordingto any of the 3 sources. After exclusion because of death or emigration, atotal of 260 twins consisting of 60 (27 MZ and 33 DZ) index pairs were chosenrandomly from the available index pairs (n = 229: 50 MZ, 121 DZ, and 58 unknownzygosity), along with 70 (34 MZ and 36 DZ) demographically balanced healthypairs. Index pairs in which, on direct interview, either the proband had adiagnosis of schizoaffective disorder, affective type, or the co-twin hada psychotic disorder diagnosis were excluded (n = 1 concordant MZ pair). Healthypairs were excluded if there was a history of psychosis-related treatmentor work disability in any of their first-degree relatives or if either co-twinwas found, on direct interview, to meet diagnostic criteria for a psychoticdisorder or schizotypal, paranoid, or schizoid personality disorder (n = 15pairs: 6 MZ and 9 DZ). The selected sample of 114 pairs (n = 228) consistedof 8 pairs concordant for schizophrenia (7 MZ and 1 DZ), 51 pairs discordantfor schizophrenia (19 MZ and 32 DZ), and 55 healthy pairs (28 MZ and 27 DZ).High-resolution magnetic resonance images were acquired on 252 of the 260twins. Two images were excluded because of technical problems with the magneticresonance imaging, and 1 was excluded because of a large frontal lobe lesion,leaving 249 images on which hippocampal volumes were measured. After exclusions,the sample (n = 219) on which hippocampal volumes were gathered comprised16 twins concordant for schizophrenia (7 MZ pairs and 1 DZ pair), 94 twinsdiscordant for schizophrenia (16 MZ and 28 DZ pairs and 2 and 4 additionalMZ co-twins and DZ probands, respectively), and 56 MZ (28 pairs) and 53 DZ(26 pairs) healthy comparison individuals.
Each co-twin was interviewed using the StructuredClinical Interview for DSM-III-R Disorders, patientor nonpatient edition,26 by an examiner whowas blind to the zygosity and diagnostic status of their co-twin, and thetwins were assigned diagnoses according to DSM-IV.27 Co-twins and healthy individuals were also interviewedand rated on the cluster A items from the Personality DisorderExamination.28 Diagnostic reliabilitywas excellent (ie, mean ± SD κ = 0.94 ± 0.02)29; final diagnoses were made by consensus among 3 independentraters (T.D.C., M.H., and J.L.) after review of written case reports. Individualswith a psychotic condition were also rated using the Scalefor the Assessment of Positive Symptoms30 andthe Scale for the Assessment of Negative Symptoms.31 Of the 64 probands, 58 (26 MZ and 32 DZ) were diagnosedas having schizophrenia and 8 as having schizoaffective disorder (3 MZ and5 DZ). Five MZ and 2 DZ co-twins of schizophrenic patients had a cluster Apersonality disorder (Table 1).Substance disorder was rated as present when participants were actively abusingalcohol, sedatives, cannabis, stimulants, opioids, cocaine, hallucinogens,or a multitude of other substances as scored by the StructuredClinical Interview for DSM-III-R Disorders.
Zygosity was determined by DNA analysis using the following markers:DIS80 (20 alleles), DI7S30 (13 alleles), apoB (20 alleles), COL2A1 (10 alleles),vWA (9 alleles), and HUMTH01 (6 alleles). Assuming an average heterozygosityrate of 70% per marker, we estimate that this procedure will falsely classifya DZ pair as MZ in approximately 1 of 482 cases.
Magnetic resonance images were acquired on a 1.0-T scanner (SiemensMedical Systems, Iselin, NJ) in the Department of Radiology, University ofHelsinki, using a standard magnetization-prepared rapid gradient echo sequence,with a repetition time of 10 milliseconds, an echo time of 4 milliseconds,a number of excitations equal to 1, and a flip angle of 12°. The imagescomprised 128 sagittal slices, with 1.2-mm slice thickness and no interslicegap. The matrix size was 256 × 256 pixels, corresponding to a fieldof view of 25 cm2 and an in-plane resolution of 0.9766 ×0.9766 mm.
After deleting nonbrain voxels using a conservative automated procedurefollowed by manual removal of nonbrain tissue, the images were segmented intogray matter, white matter, and cerebrospinal fluid using an adaptive, 3-dimensional,Bayesian algorithm33 previously validated forthis purpose.34 To control for differencesin head tilt during acquisition, images were resliced parallel to the anteriorcommissure–posterior commissure plane and saved in sagittal view.
A method for outlining the hippocampal region of interest was developedby 2 of us (P.A.S. and T.G.M.vE.) and is described and depicted in detailin another publication.10 A previous publicationusing the same method reported lower hippocampal volumes in patients comparedwith non-ill siblings and in non-ill siblings compared with healthy individuals,and the range of the volumes in that study10 isvery consistent with that in this study. Briefly, tracings were started onthe most lateral slice on which the hippocampus was first visible, in themost lateral extent of the temporal horn of the lateral ventricle. The inferiorand superior borders were determined by drawing a line through the white matterseparating the hippocampus from the parahippocampal and fusiform gyri andthe alveus separating the hippocampus from the lateral ventricles, respectively.More medially, the anterior hippocampus was separated from the amygdala bya thin line of white matter between the 2 structures. The last slice on whichthe hippocampus was clearly distinct from the amygdala formed the medial border.This roughly corresponds to the second slice medial to the slice on whichthe parahippocampal gyrus separates or 2 slices before the midbrain formsin the temporal horn of the lateral ventricle. The hippocampal volume measuresinclude the cornu Ammonis, the gyrus dentatus, the prosubiculum, and the subiculumproper (see Table 2 for raw hippocampalvolumes). Only voxels in the region of interest that were classified as graymatter were counted. Tracings were performed blindly to diagnosis, birth history,and hemisphere. Interrater and intrarater reliabilities based on 10 caseswere excellent (ICCs >0.95).
Before analysis, data were checked for normality35 andhomogeneity of variance.36 Data were analyzedusing the general linear mixed model with repeated measures (SAS version 6.12;SAS Institute Inc, Cary, NC), correcting for dependency (ie, correlation)among co-twins by treating twin pair as a random variable and adjusting themodel error terms accordingly (Satterthwaite option). Hypotheses pertainingto the mean comparisons were tested by modeling risk group (probands fromMZ concordant pairs, probands from MZ discordant pairs, probands from DZ discordantpairs, healthy MZ co-twins from discordant pairs, healthy DZ co-twins fromdiscordant pairs, and healthy twin pairs) as a fixed-effect predictor whilecovarying for age at imaging, history of substance disorder, sex, total corticalgray matter volume,37 and the interactionsof group with a history of substance disorder and sex (model 1). To test forpossible differences in laterality, hemisphere entered the model as a within-subjectrepeated-measures factor, and a group × hemisphere interaction enteredthe model to test for possible differences in laterality among the groups.Whenever one of these terms contributed statistically significantly to theprediction of hippocampal volume, contrast analyses were performed comparingconditions within the term collapsing over nonsignificant terms in the model.This approach maintains the hypothesis-wise type I error rate at 0.05 becausea predictor's contribution to particular dependent measures is evaluated onlyif its effect is found to vary at the multivariate level. The significanceof each predictor was tested while accounting for all other model terms simultaneously,and where mean differences were hypothesized, 1-tailed tests were used. Theseanalyses were also performed for cortical gray matter while covarying forintracranial volume as a point of comparison to the hippocampus.
Intraclass correlations and their confidence intervals for index (concordantMZ and discordant MZ and DZ) and healthy (MZ and DZ) pairs were calculatedusing the analysis of variance method in SAS (version 6.12) for hippocampal,intracranial, cortical gray matter, and hippocampal corrected for corticalgray matter volumes. Predicted differences in ICCs between pairs were comparedusing 1-tailed t tests.
Finally, variance component analyses were performed using version 1.50dof MX38 to determine the proportion of thevariance in these volumes explained by additive genetic factors and sharedand unique environmental factors in healthy and index twin pairs while covaryingfor age and sex.
There were significant effects for risk group (F5,372 = 9.23; P<.001), hemisphere (F1,372 = 3.95; P = .048), substance abuse (F1,372 = 7.27; P = .007), and intracranial gray matter volume (F1,372 = 45.43; P<.001) in predicting hippocampalvolumes.
Given that the risk group effect did not vary by hemisphere, contrastanalyses were performed on hippocampal volumes collapsed across hemisphere.Probands from MZ concordant pairs, MZ discordant pairs, and DZ discordantpairs had smaller mean hippocampal volumes than healthy individuals (t372 = 4.0, t372 = 4.9, and t372 = 4.4, respectively; P<.001 for all). As predicted (hypothesis 1), none ofthe mean hippocampal volumes of the 3 proband groups differed statisticallysignificantly from each other (Figure 1).Although the hippocampal volumes of the non-ill MZ and DZ co-twins from pairsdiscordant for schizophrenia were smaller than those of healthy control twins(t372 = 2.6; P =.005 and t372 = 3.2; P<.001, respectively), contrary to hypothesis 2, the non-ill MZco-twins did not differ from the non-ill DZ co-twins (t372 = 0.14; P = .45). Consistentwith hypothesis 3, the probands from MZ discordant twin pairs had smallerhippocampal volumes than their non-ill MZ co-twins (t372 = 2.0; P = .02). Probands from DZ discordanttwin pairs did not have significantly smaller hippocampal volumes than theirnon-ill DZ co-twins (t372 = 1.0; P = .15). However, hippocampal volumes from MZ and DZ probandscombined were smaller than those of MZ (t372 = 1.77; P = .04) and DZ (t372 = 1.84; P = .03) co-twins,and these effects were even stronger when hippocampal volumes from MZ concordantprobands were added (t372 = 1.98 and t372 = 2.1, respectively; P = .02 for both).
We replicated our previous findings of larger right than left hippocampalvolumes (t372 = 2.0; P = .02). The hippocampal volumes of individuals diagnosed as havinga substance disorder were smaller than those without such a diagnosis (t372 = −2.7; P<.001).
Percentage-wise, 11 (69%) of the 16 MZ and 19 (68%) of the 28 DZ probandshad smaller hippocampal volumes than their healthy co-twins, and 12 (75%)of the 16 MZ and 25 (89%) of the 28 DZ co-twins had smaller hippocampal volumesthan the average of the healthy twins.
There were significant effects for risk group (F5,181 = 3.0; P = .01), risk group × sex (F5,181 = 3.14; P = .01), substance abuse (F1,181 = 4.6; P = .03), age (F1,181 = 33.8; P<.001), and intracranial volume (F1,402 = 51.32; P<.001) in predicting cortical gray matter volumes.
Contrast analyses showed that probands from pairs concordant for schizophreniahad less cortical gray matter than probands from discordant MZ (t181 = 2.2; P = .03) and DZ (t181 = 2.5; P = .01)pairs, non-ill MZ (t181 = 2.7; P = .009) and DZ (t181 = 3.2; P = .002) co-twins, and healthy twins (t181 = 3.7; P<.001). None ofthe other groups differed from each other.
Cortical gray matter volumes of individuals diagnosed as having a substancedisorder were smaller than those of individuals without such a diagnosis (t181 = −2.2; P =.03). Cortical gray matter volumes in the overall sample seemed to declinewith age (slope = −2.4 [SE = 0.41358462], t181 = −5.8; P<.001, 2-tailed). Femaleconcordant patients had larger cortical gray matter volumes than male concordantpatients (t181 = 2.4; P = .02, 2-tailed), whereas female MZ co-twins and healthy twins hadsmaller cortical gray matter volumes than male MZ co-twins (t181 = −2.3; P = .02, 2-tailed)and male healthy twins (t181 = −2.7; P = .009, 2-tailed), respectively; none of the other groupsshowed sex differences in cortical gray matter volumes.
The ICCs for healthy MZ pairs were larger than those for healthy DZpairs and the ICCs for concordant index MZ pairs were larger than those fordiscordant index MZ and DZ pairs on all measures (Table 3). Although the ICCs for hippocampal volumes in discordantMZ and DZ pairs were similar, those for hippocampal volumes corrected forcortical gray matter, and those for intracranial and cortical gray mattervolume, were larger in discordant MZ compared with DZ pairs (Table 3). Finally, the ICC for hippocampal volumes in healthy MZpairs was larger than that in discordant MZ pairs (t43 = 8.1; P<.001).
In index and healthy pairs, unique environment/error–only modelshad significantly worse fit than additive genes, common environment, uniqueenvironment/error (ACE) models for hippocampal and total gray matter and intracranialvolume (Table 4). Although notsignificantly different from the ACE model, based on parsimony and fit statistics,the AE model is the best-fitting model for all the regions in the healthytwins, whereas the CE model provides the best fit for hippocampal and totalgray matter volume in the index twins and the AE model provides the best fitfor intracranial volume and total hippocampal volume corrected for total corticalgray matter volume (Table 4).Based on this analysis, the variance component for the effect of additivegenes on hippocampal volume corrected for cortical gray matter volume is 71%in the healthy twins and 42% in the discordant twins.
The principal finding of this study is that although hippocampal volumesin healthy twins are highly heritable, those in twins discordant for schizophreniaare subject to substantially great modulation by environmental factors.
The higher ICC for healthy MZ vs DZ pairs and the best fit for the AEmodel in the variance component analysis corroborate findings by other researchers16,39,40 that hippocampalvolume in healthy individuals is highly heritable. A combination of genetic8,16,41,42 andunique environmental12- 15 effectson hippocampal volume in schizophrenia is indicated by (1) smaller hippocampalvolumes in probands compared with their non-ill MZ co-twins, (2) larger ICCsfor hippocampal volume in healthy compared with discordant MZ pairs, (3) highervariance components for additive genes in healthy compared with discordanttwins, and (4) statistically significantly higher ICCs for discordant MZ comparedwith DZ twin hippocampal volume.
Interpretation of the data is based on 3 assumptions underlying theclassic twin design: (1) MZ twins share 100% and DZ twins share, on average,50% of their polymorphic genetic material, and current methods can adequatelyidentify zygosity; (2) the environment shared among MZ and DZ twins is similar;and (3) twins are similar to singletons such that findings in twins can begeneralized to nontwin populations.
Although it has been hypothesized that MZ twins discordant for schizophreniashare less of their polymorphic genetic material than concordant and healthytwins,43- 47 theempirical evidence for this claim remains controversial.44,46,48 Incontrast, obstetric complications seem to increase the risk for schizophrenia,49- 51 are associated withillness discordance,52 and have been relatedto hippocampal volume reduction in schizophrenia.9,10,14 Itcould still be argued that the smaller hippocampal volumes in probands comparedwith non-ill MZ co-twins are due to factors associated with disease status.However, in contrast to this hypothesis, and consistent with the results ofSuddath and colleagues,12 hippocampal volumesof probands were not associated with illness duration after covarying forage (SE = .04, t50 = 1.6; P = .11) or with years receiving neuroleptic medication.
The findings are also interpreted assuming that MZ and DZ twins shareenvironmental exposures to similar degrees. A competing interpretation ofthe similarity in mean hippocampal volume between the MZ and DZ co-twins andthe higher ICC for MZ compared with DZ discordant twin pairs could be thatDZ co-twins experienced more pregnancy complications than MZ co-twins fromdiscordant pairs. In this sample, the frequency of prenatal and perinatalcomplications coded blindly from the original obstetric records on approximatelyhalf the studied twin pairs (T.D.C., unpublished observations, 2000) did notdiffer between discordant MZ and DZ pairs.22 Afull interpretation of the obstetric data is not possible because obstetricdata were recorded by pair and not by individual twin.
The twin method is often criticized for nongeneralizability owing todifferences in the intrauterine and family environment of twins compared withsingletons. However, recent studies show that any differences in cognitiveabilities between twins and their siblings no longer exist at age 5 years53 and that although second-born twins have lower intracranialvolumes than first-born twins, all other volumes are comparable when controllingfor intracranial volume, suggesting that twin studies can provide reliableestimates of heritabilities in brain volume measures and that these can begeneralized to singleton populations.54
The differences in overall cortical gray matter volume were differentfrom those observed for the hippocampus, suggesting that the observed patternof hippocampal volume reduction is not due to global effects on gray matter.Although intracranial gray matter volume was only reduced in concordant MZprobands relative to the other groups, a previous study55 onregional cortical gray matter deficits showed that particular heteromodalcortical regions are affected by genetic liability and disease-related environmentalfactors.
The finding that the hippocampal volumes of probands from concordantpairs do not differ from those of discordant pairs is consistent with thatof Weinberger and colleagues18 in suggestingthat the etiologies underlying the hippocampal volume reductions in these2 types of probands are similar. However, the higher ICCs for concordant MZtwin pairs compared with discordant MZ and DZ twin pairs suggest that theenvironmental factor contributing to discordance may affect dissimilarityof hippocampal volume within twin pairs also.
Previously reported data on verbal episodic memory in the same sample,22 thought to rely on the hippocampus56 andpreviously shown to correlate with hippocampal volume,8,23 matchthe pattern of hippocampal volume reduction and similarities observed in thisstudy.
Strengths of this study are as follows: a random representative populationsample was used such that the results can be generalized to the total populationof twins, probands from concordant and discordant pairs were available suchthat their volumes could be compared directly, concordant and discordant pairswere available such that ICCs could be compared directly, high-resolutionimages were used to make the measurements, high reliabilities were achievedon the measurements, and the rater was blind to the presentation of the images(neurologic/radiologic) during data collection such that potential rater orother orientation biases were eliminated.
Several weaknesses of this study must also be noted. The measurementsonly reflect hippocampal volumes, and it is possible that there are also regionalshape changes, in particular in the Sommer sector. Although the sample sizeis relatively large, the effects under examination are relatively small. Datafor prenatal and perinatal complications were not available for the entiresample, making it impossible to directly examine the effects of specific environmentalfactors on hippocampal volume.
Finally, although hippocampal volume reduction in schizophrenia seemsto be affected by genetic factors and unique environmental factors, it isyet to be determined in which parts of the hippocampal microstructure, andwhen during development, these factors act or interact.
Corresponding author: Tyrone D. Cannon, PhD, Department of Psychology,University of California, Los Angeles, 1285 Franz Hall, Box 951563, Los Angeles,CA 90095 (e-mail: email@example.com).
Submitted for publication October 10, 2002; final revision receivedNovember 5, 2003; accepted November 19, 2003.
This study was supported by grant MH52857 from the National Instituteof Mental Health, Bethesda, Md, and by grant RR00827 to the FIRST BiomedicalInformatics Research Network (http://www.nbirn.net), funded bythe National Center for Research Resources at the National Institutes of Health.