|Home | About | Journals | Submit | Contact Us | Français|
Despite evidence that obsessive-compulsive disorder (OCD) is a familial neuropsychiatric condition, progress aimed at identifying genetic determinants of the disorder has been slow. The OCD Collaborative Genetic Study (OCGS) has identified several OCD susceptibility loci through linkage analysis.
In this study we investigate two regions on chromosomes 15q and 1q by first refining the linkage region using additional STRP markers. We then performed association analysis on SNP genotyped (markers placed every 2-4kb) in the linkage regions in the OCGS sample of 376 rigorously phenotyped affected families.
Three SNPs are most strongly associated with OCD: rs11854486 (p=0.00005 [0.046 after adjustment for multiple tests]; Genetic Relative Risk (GRR)=11.1 homozygous and 1.6 heterozygous) and rs4625687 (p=0.00007(after adjustment = 0.06); GRR=2.4) on 15q; and rs4387163 (p=0.0002 (after adjustment= 0.08); GRR=1.97) on 1q. The first SNP is adjacent to NANOGP8, the second SNP is in MEIS2, and the third is150kb between PBX1 and LMX1A.
All the genes implicated by association signals are homeobox genes and are intimately involved in neurodevelopment. PBX1 and MEIS2 exert their effects by the formation of a heterodimeric complex which is involved in development of the striatum, a brain region involved in the pathophysiology of OCD. NANOGP8 is a retrogene of NANOG, a homeobox transcription factor known to be involved in regulation of neuronal development. These findings need replication; but support the hypothesis that genes involved in striatal development are implicated in the pathogenesis of OCD.
Obsessive-compulsive disorder (OCD) is a psychiatric condition first described over a century ago (Janet, 1903). The pathognomonic features of the disorder are persistent, intrusive, senseless thoughts and impulses (obsessions) and repetitive, intentional behaviors (compulsions). Patients with the disorder recognize that their thoughts and behaviors are excessive and unreasonable, and they struggle to resist them. The lifetime prevalence of OCD is estimated to be 1-3%, based on population-based surveys conducted in many communities nationally and internationally, (Karno and Golding 1991; Weissman et al 1994). Conventional behavioral and medication treatments for this chronic, disabling disorder, while helpful, generally bring only partial relief. No cure exists. While neurocircuitry including the striatum and its key functional connections has been consistently implicated in OCD, application of that knowledge to treatment remains very limited (Greenberg et al 2010). The discovery of genes is crucial for elucidation of pathogenic mechanisms and for developing new rational treatments.
Several family studies of OCD indicate a genetic component for its origin (Black et al 1992; do Rosario-Campos et al 2005; Lenane et al 1990; Pauls et al 1995; Nestadt et al 2000). The Johns Hopkins OCD family study (Nestadt et al 2000) reported that the prevalence of OCD in the first-degree relatives of probands was 11.7%, compared to 2.7% in the relatives of controls; also, the prevalence of OCD in the siblings of early onset probands was 17.9% (λ sib =7.8). There are several reports of monozygotic twins concordant for obsessive-compulsive symptoms (Woodruff and Pitts, 1964; Inyoue, 1965; Carey and Gottesman II, 1981).
The OCD Collaborative Genetics Study (OCGS) conducted by our group, was the first large-scale genome-wide linkage scan searching for genetic loci that predispose individuals to OCD (Shugart et al 2006); this study included 219 families. The strongest signal was observed at chromosome 3q27/28, with a LOD score of 2.67 (P-value=0.0002). Regions on chromosomes 1, 6, 7, and 15 also showed evidence of linkage, with P-values <0.01(Shugart et al 2006).
The aim of the current study is to follow up the chromosomal regions identified in our linkage study using genetic association methods. This report describes the most convincing results in regions on chromosomes 15q and 1q, and centers on four genes (Meis homeobox 2 (MEIS2), pre-B-cell leukemia homeobox 1 (PBX1), LIM homeobox transcription factor 1, alpha (LMX1A), and Nanog homeobox pseudogene 8 (NANOGP8) that not only have the most significant findings, but suggest a novel and testable hypothesis about the origin of OCD.
The OCGS, which commenced in 2001, is an ongoing, NIMH-funded collaboration among investigators at six sites in the US (Brown University, Columbia University, Johns Hopkins University (JHU), Massachusetts General Hospital, University of California at Los Angeles, and the National Institute of Mental Health). The methods are described in detail elsewhere (Samuels et al 2006) and summarized below.
The OCGS ascertained families with OCD-affected sibling pairs, and extended these when possible through affected first- and second-degree relatives; in addition, two sites also collected other pedigrees with multiple-affected relatives when these were available. Subjects were recruited into the study from outpatient and inpatient clinics, referrals from clinicians in the community, web sites, media advertisements, self-help groups, and annual conventions of the Obsessive Compulsive Disorder Foundation. To be considered affected, a subject had to meet Diagnostic and Statistical Manual (DSM)-IV OCD diagnostic criteria at any time in his/her life (APA, 2000). Probands were included if, in addition to meeting DSM-IV criteria, their first onset of obsessions and/or compulsions occurred before 18 years of age. Probands with schizophrenia, severe mental retardation, Tourette disorder, or secondary OCD (OCD occurring exclusively in the context of depression) were excluded. Subjects had to be at least 7 years old to participate in the study. Written, informed consent (or assent, for children) to study procedures was obtained before the clinical interview. The protocol was approved by the institutional review board at each site.
Diagnostic assessments were conducted by psychiatrists or PhD-level psychologists experienced with clinical evaluations using the OCGS assessment package, modified and developed for the study, as a semi-structured format for the evaluation of psychopathology. The OCD section was adapted from the Schedule for Affective Disorders and Schizophrenia-Lifetime Anxiety Version (SADS-LA-R; Mannuzza et al 1986) and included detailed screening questions; the Yale Brown Obsessive Compulsive Scale (YBOCS; Goodman et al 1989) and symptom checklist, and additional questions on onset, course, and treatment of symptoms. A similar section was developed for assessing tics, Tourette disorder, and other tic disorders. The Structured Clinical Interview for DSM-IV (SCID; Spitzer et al 1992) was used for assessing other major Axis I diagnoses and a semi-structured assessment protocol was used for additional diagnoses of interest. In adults, items from the Structured Instrument for the Diagnosis of DSM-IV Personality Disorders (SIDP; Pfohl et al 1989) were used to assess the presence of personality disorders. The Family Informant Schedule and Criteria (Mannuzza et al 1985) was used to obtain additional information about each subject from a knowledgeable informant. For subjects who had received psychiatric treatment, consent was obtained to review relevant medical records and to contact treatment providers, if such information was deemed useful for making diagnoses. Examiners completed a narrative formulation for each case.
The JHU OCD diagnostic assignment checklist was used to collate all the clinical information from a variety of sources (the semi-structured direct interview, case formulation, informant interview, and medical records). The checklist presents logical algorithms with specified rules, allowing assignment of definite, probable, absent, or unknown for each disorder. All psychiatric diagnoses were made according to strict DSM-IV criteria (APA, 2000). At each site, each case was reviewed independently by two expert diagnosticians who reviewed all case materials. Final diagnoses were assigned by diagnosticians at Johns Hopkins University. If all criteria required for having the disorder were met, then a ‘definite’ diagnosis was given. If any required criterion was clearly not met, then the diagnosis was considered ‘absent’. If it appeared likely that the subject had the diagnosis, but the diagnosticians could not be certain of a given criterion (required for definite diagnosis), then the diagnosis was made at the ‘probable’ level. If the diagnosticians could not be sure of the presence or absence of a given diagnosis, then that diagnosis was recorded as ‘unknown’.
Blood samples were collected from all affected probands, their parents, and their affected relatives. If available, blood samples from unaffected relatives were also collected to help determine phase used for computing identity by descent (IBD) sharing probabilities.
Different approaches were taken and different genotyping laboratories used for the two chromosomal regions. Specifically, the region on chromosome 15q was studied in a single fine-mapping phase, whereas, the region on chromosomes 1q was studied in three phases, such that the final phase represented the result of sequential testing. Furthermore, the sample size increased between the initial linkage study and the association studies.
Fine mapping of the regions on chromosome 15q and 1q, using 14 short tandem repeat polymorphic markers (STRP) markers across 12.8Mb region of 15q, and 21 STRP markers spaced every 1.2Mb across a 25.8Mb region on 1q, was conducted at deCODE genetics service laboratory in Iceland. Samples were analyzed by capillary electrophoresis utilizing ABI DNA analyzers. Alleles were called by proprietary deCODE Allele Caller (DAC) software. The genotype success rate was 96.1%. A total of 945 subjects in 214 families were genotyped. Compared to the genome-wide linkage scan, five families were excluded from the linkage fine mapping due to insufficient amount of DNA for genotyping.
Two thousand, two hundred-and-fifty-five single nucleotide polymorphisms (SNPs) were genotyped in a 10.7 Mb region with average spacing of 4.8kb in the 15q linkage region at deCODE, in Iceland. Fine mapping of the refined linkage region on 1q, using 860 SNP markers spaced every 6.3kb across a 5.4 Mb region (1q23-24), was conducted at the Illumina service laboratory. Public databases (dbSNP Build 126) were used to select the (SNPs) genotyped by Illumina with their BeadArray system with the GoldenGate assay (Steemers and Gunderson, 2005). We gave preference to validated SNPs with minor allele frequency >0.05 and high anticipated Illumina assay success rate (>80%). The genotype success rate was 99.46%.
We included the families described above and an additional 162 families (631 individuals) recruited and evaluated after the initial study in both the 15q 1nd 1q association studies; i.e. a total of 376 families, including 1,576 individuals. Among these new families, 73 had at least one OCD-affected sibling pair or other relative pair; 89 were trio families. 999 participants in this study were diagnosed with probable or definite OCD. 622 (62.3%) of these participants were female and 208 (20.8%) were younger than eighteen years; the average age of the cases was 35 years and their means score on the Y-BOCS (lifetime worst episode) was 24.6(SD 7.2).
An additional 160 SNP markers were genotyped, at the deCODE service laboratory in Iceland using the same methods, in regions identified based on the significance of the signal in either the prior fine mapping phase, or that were in a gene or a conserved region according to the University of California Santa Cruz (UCSC) database.
Multipoint allele-sharing methods were used to assess evidence for linkage by testing the null hypothesis given = 0, where is the extent deviated from expected IBD sharing (Kong and Cox, 1997). Using Merlin version 0.10.2 (Abecasis et al 2002), we computed the Kong and Cox logarithm of odds (LOD)all (KACall) statistic.
We used the Family-Based Association Test (FBAT; Horvath et al 2001) to test each SNP separately for association with OCD under the additive model. Then, we used the Haplotype-Based Association Test (HBAT; Horvath et al 2001) to test haplotypes of SNPs for association with OCD. We chose to use predefined blocks to specify the haplotypes for testing. Finally, we used the gtrr function of the GenAssoc STATA (Clayton, 2007) to obtain the estimated odds ratio as well as calculate global likelihood ratio test P-values. Empirical P-values for each significant association signal were also calculated with FBAT using 100,000 permutations.
Two-point P-values from the FBAT were corrected for multiple-marker testing in the presence of linkage disequilibrium (LD). The P-values Adjusted for Correlated Tests (PACT) approach was employed to adjust for multiple testing (Conneely and Boehnke, 2007).
Estimates of the interaction between identified markers were accomplished by performing conditional logistic regression. The transmission disequilibrium test (TDT) test is the McNemar’s chi-squared test, and can be embedded in the more general conditional logistic regression framework. With two or more markers, the “case” is the transmission of a parental haplotype instead of an allele, and the “control” is non-transmission of a parental haplotype. Therefore, with L markers considered simultaneously, each case-parent trio contributes two entries in a 2L × 2L matrix. . The test of association tests the asymmetry of such a matrix. The test can be performed in conditional logistic regression with L covariates of exposure and possible interaction terms.
The linkage signal at 15q increased only marginally after genotyping additional STRP markers in the original 214 families; five families were not genotyped. The KACall increased from 1.32 (p = 0.007) in the original genome wide linkage scan to 1.43 (p = 0.005) at marker DG15S14.
SNPs were genotyped every 4.3kb across the 1-LOD linkage region. As shown in Figure 1, SNP rs4625687, in the gene MEIS2, was associated with OCD (p = 0.000079; empirical p = 0.00012; adjusted for multiple tests, p= 0.06). The GRR at this SNP was 2.4 (95% C.I.: 1.5, 3.8). The SNP rs11854468, located approximately 64 kb downstream from a NANOG homeobox transcription factor (OMIM 607937) pseudogene, NANOGP8, was significantly associated with OCD (p=0.00005; adjusted for multiple tests p=0.05; GRR= 11.1, 95% C.I.:100.0,1.4 (GG) and GRR=1.6, 95% C.I.:2.3,1.1 (GA). Several SNPs in or near formin1 (FMN1) were significantly associated with OCD; the strongest association was at rs345804 (p= 0.00049; GRR = 1.6, 95% C.I.: 1.2, 2.0) in an expressed sequence tag (AI040235) about 5.7kb upstream of FMN1. One associated SNP rs2306277 in exon 1 of FMN1 is reported in Polyphen as a missense mutation with “probable protein damage.” No interesting haplotype was identified in the 15q region.
The linkage signal in this region increased after genotyping additional STRP markers in the original sample of families (214 families). The KACall increased from 1.42 (p = 0.005) in the original genome wide linkage scan to 2.06 (p = 0.001) at marker D1S2762 (figure 2).
SNPs were genotyped, on average every 6.3kb, across the 5.4 Mb region of the 1-LOD interval of the linkage peak. As shown in Table 1 and figure 3, five SNPs between PBX1 and LMX1A were associated with OCD. rs4387163 gave the most significant signal (p=0.0002; empirical p=0.0002; adjusted for multiple tests p= 0.08) with a genetic relative risk (GRR) of 1.97 (95% C.I.:1.33, 2.92) for the homozygote (G/G) and 1.58 (95% C.I.:1.19, 2.10) for the heterozygote (G/A). Haploview (confidence interval method) identified three contiguous haplotype blocks in the region of this signal, as described in Figure 3. The first two blocks contained SNPs with signals of p<0.01. The most common haplotype in Block 1 (frequency = 0.55) was associated with OCD (p = 0.0012); the most common haplotype in Block 2 associated with OCD had a frequency = 0.55 (p = 0.0004); and an extended haplotype across both blocks had a frequency = 0.52 and was also associated with OCD (P = 0.0004). A fourth haplotype based on “pair-wise tagging” with a threshold of r2>0.5, computed in Haploview (most common haplotype frequency = 0.526), provided similar results (p= 0.0002).
Evidence for statistical interaction was investigated using conditional logic regression. The best markers on 1q are rs4387163 and rs986362 and those on 15q are rs4625687, rs4244559, and rs1464284. No evidence for interaction was detected in tests including each of the markers on one chromosome with those on the other chromosome. It should be noted that the power to detect evidence of interaction is limited; a much larger sample size than this one would be required to determine the presence of statistical interaction.
We explored the clinical data for association between clinical characteristics and the associated allele of SNP rs4625687 in MEIS2. Three clinical characteristics were significantly associated with this allele; the presence of any tic disorder (p < 0.05); body dysmorphic disorder (p<0.05); and the symmetry/ordering symptom factor, derived by factor analysis (Pinto et al 2008), from the overall range of OCD symptoms (p=0.02).
This study represents a classical approach for the identification of genes underlying common and complex disorders, such as the disabling neuropsychiatric condition, OCD. The approach entailed a genome-wide linkage scan in a large sample of rigorously assessed families, each with multiple members diagnosed with DSM-IV OCD (Shugart et al 2006). Chromosomal regions with the strongest linkage signals were then investigated further; first with additional linkage fine-mapping using STRP markers, and then with staged, association fine-mapping using SNP markers.
The original linkage study suggested that six regions were most promising for further investigation. After additional linkage and association fine-mapping, two regions on chromosomes 15q and 1q, and, more specifically, two genes in each region, were associated with OCD. These genes are MEIS2 and NANOGP8 (15q) and PBX1 and LMX1A (1q).
The strongest signals on chromosome 15q are at rs11854468 (p=0.000054), located approximately 60 kb from NANOGP8, and at rs4625687 (p=0.00007), located in the intron of MEIS2. There is another strong signal at rs11858145 (p=0.0007) located in a coding region of FMN1. The best signal on chromosome 1q is at rs4387163 (p= 0.0002) which is located at 1q23.3. The GRR at this locus is 1.9. This SNP is located 150kb equidistant from both PBX1 and LMX1A. SNP rs11854468 on 15q remained significant after adjustment for multiple testing (p<0.05). Although statistically this is the most robust signal, rs4625687 in the MEIS2 gene (p after adjustment = 0.065) remains an interesting signal, because of its location and the signals at surrounding SNPs.
An outstanding feature of these results is that PBX1, MEIS2, LMX1A, and the NANOG-related NANOGP8 gene are all transcription factors, or closely related to transcription factors, important in neurodevelopment. While we recognize that there are many transcription factors involved in neurodevelopment, and even more that could plausibly be implicated in some hypothetical relationship to the origin of OCD, we assert that the probability is exceedingly small that genes with these functional properties would be identified. Specifically, of the thousands of genes across the genome there are approximately 233 Homeobox genes identified; therefore identifying these four genes by chance is highly improbable.
MEIS2, PBX1, LMX1A, and NANOG code for homeobox proteins: the former two are members of the three amino acid loop extension (TALE) superclass, whereas LMX1A is a member of the LIM superclass. All are cofactors for transcription regulation and have important roles in brain development. MEIS2 and PBX are intimately involved as cofactors with homeobox (HOX) and myogenic differentiation (MyoD) genes (Berkes et al 2004). It has been shown that MEIS2 and PBX1 form a heterodimeric complex which acts together with HOX genes to activate transcription (Sagerström, 2004) and MEIS2 has been found to be the preferred partner of PBX1 (Chang, et al 1997). NANOG also binds to PBX1 which regulate its function (Chan et al., 2009). NANOGP8 is a retrogene (Zhang, et al., 2006) which belongs to a family of 11 NANOG pseudogenes. The NANOGP8 gene in 15q14 is unique among NANOG pseudogenes in that it has an intact open reading frame and is actively transcribed (Jeter et al 2009). It is the only NANOG-related gene that is uniquely expressed in humans compared to chimpanzees (Fairbanks et al 2006). FMN1 is a regulator of another important transcription factor in neurodevelopment, glioma-associated oncogene family zinc finger 3 (GLI3; Zeller et al 1999; Zuniga & Zeller, 1999).
Given the relationship between MEIS2 and PBX1 and the HOX class of transcription factors, and in particular the interaction of PBX1 and HOXB8 (Knoepfler et al 2001; Neuteboom et al 1995; Sanchez et al 1997), it is fascinating that Greer and Capecchi (2002) reported that HOXB8 knockout mice exhibit compulsive hair pulling behavior. This group recently reported that this behavior is attributable to bone marrow-derived microglia and hypothesize an immunologic basis to the behavior (Chen et al 2010). It should be noted that both MEIS2 and PBX1 are also intimately involved in hematopoiesis (Ficara et al 2008; Wilson et al 2009).
MEIS2 and PBX1 have been shown to be expressed in the developing rat striatum (Cecconi et al 1997; Toresson et al 2009), as has NANOG (Molero et al 2009). Takahashi et al (2008) suggest that MEIS2 and PBX1 might play a “pivotal role in establishing the neurons of the islands of Calleja, within the ventral striatum, as well as striosomal neurons in the striatum” (Takahashi et al 2008). These transcription factors have also been reported to be associated with the ephA8 regulatory sequence in mesencephalon development (Sungbo et al 2007). MEIS2 (with ISLET1) has also been shown to activate transcription through I56ii (a conserved regulatory element of DLX genes) during mid-gestation. Additionally, both PBX1 and MEIS2 have been shown to have a functional link to SOX3, which is involved in specifying neural fate (Mojsin and Stevanovic, 2009). Determining the pathways through which these and other transcription factors interact in the development of the relevant brain regions is in progress (Long et al 2009a; Long et al 2009b), and will ultimately enhance our understanding of brain development and possibly the origin of neuropsychiatric conditions.
Either of these two genes (i.e. MEIS2 and PBX1) alone could be considered interesting candidate genes involved in the pathogenesis of OCD (Takahashi et al 2008). However, identifying signals in the region of both genes, on different chromosomes, makes these findings more compelling; both are involved in brain development in a brain region, the striatum, that is the primary focus of interest in OCD; and they exert their developmental effect in tandem, forming a heterodimeric complex for this purpose. The genetic effects of FMN1 and NANOGP8 also plausibly impinge on neurodevelopment, albeit in different ways. While FMN1 directly supports cell structure via actin/microtubules and regulates a neurodevelopment gene, GLI3, NANOGP8 is related to NANOG which maintains stem cell pluripotency, and therefore conceivably affects neurogenesis generally, and early stage serotonin neurons specifically (Bethea et al 2009).
An obvious step to investigate a possible inter-relationship between these two genes in the origin of OCD is to determine whether there is a statistical interaction between them in their association with OCD. We did not find evidence for statistical interaction in this sample using the method described above; however, definitive identification of interaction would require a far larger sample of families than were available in this study.
LMX1A is adjacent to PBX1 on 1q and both are equidistant to the association signal in that region. LMX1A acts in a pathway that involves the development of dopamine neurons in the midbrain. It has also been reported to be involved in postnatal brain development and the maintenance of neuronal function (Andersson et al 2006). This gene is therefore also an interesting candidate in OCD. Dopaminergic function has been hypothesized to be involved in the etiology (and treatment) of OCD (Graybiel and Rauch, 2000).
FMN1 is a member of the formin family of proteins that remodel the actin and microtubule cytoskeletons (Goode and Eck, 2007); however, it also directly regulates GLI3, a key gene in development, including cortical morphology (Friedrichs et al 2008; Theil, 2005). Thus, a plausible explanation for the involvement of FMN1 in OCD may be related to the involvement of microtubule and actin cytoskeleton in neural migration (Kerjan and Gleeson, 2007) or in the gene network that regulates brain morphology (Yu et al., 2009; Quinn et al., 2009).
The likelihood of genetic heterogeneity is a fundamental concern in conducting studies such as this one. We investigated whether any clinical characteristic of the cases were associated with SNP rs4625687 in MEIS2. Only this SNP was investigated, out of several potential SNPs, as an exploratory exercise. The presence of body dysmorphic disorder, tics, and symmetry/ordering OCD symptoms were significantly more common. It should be noted that probands with Tourette Disorder were excluded. Nevertheless, these three clinical features frequently co-occur in OCD cases who exhibit tics; this and suggests that a ‘tic-form’ of OCD may be a relevant sub-phenotype underlying these findings.
Patients with OCD typically score extremely high on ‘Neuroticism’, a well-characterized personality trait in the psychological literature (Samuels et al 2000). It is therefore interesting that a WGA study of ‘neuroticism’ reported a single nucleotide polymorphism (SNP) in PBX1 as one of the ten best signals (7.18 × 10−6), and this SNP was most significant for the “anxiety” facet of neuroticism (van den Oord et al 2008). The hypothesis that ‘neuroticism’ and OCD share a common etiology would be borne out if the relationship between PBX1 and both ‘neuroticism’ and OCD were replicated.
The original linkage findings, and the subsequent fine mapping linkage results, do not meet statistical significance, as defined by Lander and Kruglyak (1995). Nevertheless, after fine-mapping the LOD score was 2.06 (p=0.001) on chromosome 1 and 1.43 (p=0.0035) on chromosome 15. There could be several reasons that these signals did not meet that threshold, including power and genetic heterogeneity. Furthermore, because the linkage signal does not reach the conventional level of significance, it could be argued that the significance of the association signal ought to be adjusted for across the entire genome. Nevertheless, this is the largest OCD sample studied and therefore these are the most likely known genomic regions to harbor OCD susceptibility genes. Therefore, it seems appropriate to both investigate these chromosomal regions and to test for significance within the linkage region. This limitation emphasizes the need for replication of these findings.
The association signals do not specifically identify which of the genes may be involved in the genesis of OCD. For instance the strongest signal on chromosome 1 is between LMX1A and PBX1. We therefore discuss the potential relationship of both genes to the development of OCD.
In conclusion, the genes with the strongest signals in this systematic genome-wide linkage and association study are all Homeobox genes, with transcription factor activity. They are important in development, and possibly to ongoing neurogenesis in brain regions strongly implicated in the pathophysiology of OCD. Thus, these findings provide important and novel evidence that OCD is a neurodevelopmental condition. The hypothesis that emerges is that developmental influences on the neuronal architecture in the striatum results in a vulnerability to OCD. An alternate hypothesis is that individuals with OCD have a deficit in neurogenesis that prevents/ limits appropriate learning mechanisms, resulting in the pathognomonic symptoms of OCD.
The authors thank the many families participating in the study; David Housman, MD, Kathleen Merikangas, PhD, Alec Wilson, PhD, and Dani Fallin PhD, for consultation; and Dalin Li PhD for analytic support; and clinicians and coordinators at each site. Yin Yao Shugart worked on this project as approved outside activities. The views expressed in this manuscript do not necessarily represent the views of the NIMH, NIH, HHS, or the United States Government.
This work was funded by the National Institute of Health grants R01-MH-50214, and NIH/NCRR/OPD-GCRC RR00052.
Financial disclosures & Conflict of Interest:
The authors report no disclosures.