Home
Registration
Speakers
Schedule
Workshops
Brochure
Directions
CME
Hotel Information
Transportation
Contact
IIGC Archive
Poster Presentations
2009 Posters
2010 Posters
2011 Posters
2012 Posters
2013 Posters
2014 Posters
Support

Poster Session at the 8th International Imaging Genetics Conference

The Scientific Program Committee of the International Imaging Genetics Conference invited the scientific community to submit abstracts of their current research to be peer reviewed and considered for inclusion in the scientific program as poster presentations. All accepted abstracts are published on the conference website below and were displayed in the Beckman Center atrium both days of the conference. The poster sessions are an important aspect of the scientific program as they promote greater interaction between the researcher and other scientists.

Poster Presentations:

A1. Title: Heritability of White Matter Development in Adolescence

Authors: B.D. Peters, P.R. Szeszko, J. Radua, T. Ikuta, P. Gruner, P. DeRosse, K. Cameron, J. Cholewa, T. Lencz, A.K. Malhotra

Keywords: adolescence, development, diffusion tensor imaging, superior longitudinal fasciculus, heritability

Aim of Investigation: To assess the heritability of developing white matter (WM) tracts that show the most pronounced change during adolescence.

Methods:Diffusion tensor imaging (DTI) was performed in a sample of 78 healthy subjects aged 8-21 years (mean age 15.3±3.7 years). Voxel-wise analysis of local correlations between age and fractional anisotropy (FA), a putative marker of white matter integrity, was conducted using tract-based spatial statistics (TBSS). In addition, we performed the first meta-analysis of TBSS studies on FA changes during adolescence (n=249). We used a novel meta-analysis method for imaging data (Radua et al., 2011), combining the use of raw t-statistic images of three studies (and our own sample) with the peak coordinates of two studies. Tracts that demonstrated most pronounced FA changes in the combined data- and meta-analysis were segmented with probabilistic tractography () to extract FA values. To assess the heritability of FA within these tracts, sibling pairs (n=8) were each age-matched with non-sibling pairs (mean age difference between the sibling pairs 3.4 years, and between the non-sibling pairs 3.5 years; group difference not significant, t=-0.25, p=0.804). Next, the slopes (unstandardized beta's) between each sibling pair and between each non-sibling pair were measured, and the mean slopes were compared between groups.

Results :In our sample, we observed widespread, bilateral increases of FA with age, and no FA decreases. Findings were most pronounced in the left superior longitudinal fasciculus (SLF), left inferior longitudinal fasciculus, left inferior fronto-occipital fasciculus, and left anterior thalamic radiation. These findings were confirmed by the meta-analysis, and sensitivity analysis indicated that FA increase in the bilateral SLF was the most consistent finding across studies. The heritability analysis of FA of the bilateral SLF indicated the following. In the sibling pairs, mean absolute slope between paired FA values was 0.005, and in the non-sibling pairs 0.012, which was a significant difference (t=-2.03, p=0.031). Mean slope across the entire sample was 0.004.

Conclusions:These data are consistent with earlier DTI studies that identified FA increases from childhood into early adulthood, and particularly highlight increasing connectivity in the SLF during this period. In addition, comparisons between sibling and non-sibling pairs suggest that development of the SLF across adolescence is at least partially heritable. When confirmed in a larger sample, these findings provide a specific anatomical target for genetic studies on normal and aberrant MW development during adolescence.

Acknowledgments: This work was supported by the National Institutes of Health (5P30MH090590-02, R01MH076995). None of the authors have potential conflicts of interest to disclose.

A2. Title: Hierarchical genetic organization of human cortical surface area

Author: Chi-Hua Chen, E.D. Gutierrez, Wes Thompson, Matthew S. Panizzon, Terry L. Jernigan, Christine Fennema-Notestine, Amy J. Jak, Michael C. Neale, Carol E.Franz, Michael J. Lyons, Michael D. Grant, Bruce Fischl, Larry J. Seidman, Ming T.Tsuang,Anders M. Dale and William S. Kremen

Keywords: MRI, cortical surface area, twin, genetics, fuzzy clustering

Aim of Investigation: Area and thickness of the cerebral cortex are highly heritable traits, yet little is known about genetic influences on regional cortical differentiation in humans.

Methods: We parceled cortical surface area into genetic subdivisions using a data-driven, fuzzy clustering technique with MRI data from 406 twins.

Results: Boundaries of the genetic divisions corresponded to meaningful structural and functional regions; however, the divisions represented novel phenotypes different from conventional (non-genetically-based) parcellation systems.

Conclusions: The genetic organization of cortical area was hierarchical, modular and symmetric across hemispheres, and could reflect known evolutionary and developmental trajectories. We also found that subtle differences in genetic influences relate to divisions in human specialization. These findings support the notion that the human cortex is built upon the foundation of its vertebrate ancestors’ brains.

A3. Title: A Relationship between 5-HT1B receptor gene expression and 5-HT1A binding in healthy subjects measured by PET

Authors: Christoph Kraus, Thomas Vanicek, Pia Baldinger, Anette Hartmann, Wolfgang Wadsak, Rupert Lanzenberger

Keywords: Positron Emission Tomography, Serotonin 1A, Serotonin 1B

Aim of Investigation: The serotonin (5-HT) 1A and 1B receptors are the most important inhibitory receptor subtypes of the 5-HT-system. The 5-HT1A receptor is highly expressed in limbic and temporal areas, whereas the 5-HT1B receptor peaks in the occipital cortex and in the globus pallidum. However, both receptor subtypes exhibit overlapping distributions in midbrain areas. Additionally, both receptors were suggested to mediate the delayed response to psychopharmacological treatment with selective serotonin reuptake inhibitors (SSRIs). Recent genetic data demonstrated two single nucleotide polymorphisms (SNPs) of the HTR1B gene (rs6298 and rs6296) affecting the time of response to SSRI treatment. Based on anatomical and functional similarities we hypothesized a reciprocal relationship between these two receptor subtypes. Thus, we aimed here to reveal a 5-HT1B-gene influence on the in vivo 5-HT1A binding as quantified by positron emission tomography.

Methods: PET-Data was acquired from 46 healthy subjects (34 female, 12 male, mean age ± SD= 39.72 ± 15.03 years) using the radioligand [carbonyl-11C]WAY-100635. Furthermore, blood samples from all subjects were drawn, DNA was isolated with the QIAGEN® QiaAMP DNA Mini Kit and genotyping was performed for the chosen SNPs using SEQUENOM® arrays. Based on previously demonstrated 5-HT1A distributions by our group, one region of interest (ROIs) in the midbrain including the dorsal raphe nucleus and one in the amygdala were chosen. The 5HT1A binding (BPND) was quantified using a Logan method as implemented in PMOD 3.3. Multivariate ANOVA in SPSS 18.0 was used to investigate the influence of rs6298 and rs6296 gene status (independent variables) on BPND (dependent variables) with a significance level of p<0.05.

Results: 27 of the 46 subjects where homozygote GG carriers and 19 subjects where heterozygote AG for the HTR1B_rs6298 genotype. None of the 46 healthy subjects where homozygote AA for the genotype HTR1B_rs6298. MANOVA revealed significant effects of HTR1B_rs6298 on 5-HT1A BPND in midbrain areas (F=4,740, P=0.035) and the amygdala (F=4,484, P=0.040). Homozygote GG for this gene had higher 5HT1A BPND in both regions compared to heterozygote AG carriers. The HTR1B_rs6296 gene polymorphism did not show any significant influence on the 5HT1A BPND in both regions (F=2,366 and F=2,505, resp., P>0.05).

Conclusion: These findings suggest that carriers with GG for HTR1B_rs6298 genotype exhibit significantly increased 5-HT1A receptor binding in the midbrain and the amygdala compared to heterozygotes. Since the entire serotonergic system is balanced precisely, alteration of one receptor may lead to disruption of neuronal communication cascades, potentially exiting in clinical symptomatology. This result underpins the interdependence of these two inhibitory serotonergic receptors and might further add to the understanding of delayed SSRI response.

A4 Title: Pediatric Imaging, Neurocognition, and Genomics: The PING Genomics Resource

Authors: Cinnamon S. Bloss, Ondrej Libiger, Burcu F. Darst, Ashley A. Scott-Van Zeeland, Nikki Villarasa, Rebecca Tisch, Natacha Akshoomoff, David G. Amaral, B.J. Casey, Linda Chang, Thomas Ernst, Jeffrey R. Gruen, Walter E. Kaufmann, Tal Kenet, David N. Kennedy, Elizabeth R. Sowell, Anders M. Dale, Terry L. Jernigan, Sarah S. Murray, Nicholas J. Schork

Keywords: imaging-genetics, multi-site collaboration, genomics, data resource

Aim of Investigation: The central goal of the Pediatric Imaging, Neurocognition, and Genomics (PING) Study is to facilitate studies that will shed light on the genomic landscape of the typically-developing human brain. Funded through the American Recovery and Reinvestment Act, PING is focused on the development of a large cross-sectional pediatric imaging-genomics dataset that will be offered as a resource to the scientific community.

Methods: PING is made possible through the efforts of a team of investigators representing 9 data collection sites throughout the U.S. where active developmental research programs are ongoing. Study participants between the ages of 3 and 20 years undergo brief cognitive testing, standardized structural, functional, and diffusion tensor imaging, and provide saliva samples for genomic studies. Genotyping is performed using the Illumina Human660W-Quad BeadChip, and preliminary genetic analyses have been performed using both standard and novel methods.

Results: To date, the PING Genomics Core has received N=1,034 (490 females) samples from which DNA has been extracted and genome-wide genotyping performed. A total of 531,732 single nucleotide polymorphisms (SNPs) have passed initial quality control filters. Preliminary analyses of a subset of 13,951 ancestry-informative SNPs indicate that the sample is highly diverse in terms of genetic background, representing more than 6 world populations. Analyses also suggest some degree of genetic relatedness within the sample, which is not unexpected given that in some cases known sibling pairs and other relative pairs were enrolled. Application of a novel method for annotating the SNP variants measured on the Illumina chip shows the presence of a limited number of functional variants. Proof-of-concept association analyses focused on replication of connectivity-associated SNPs in contactin-associated protein-like 2 (CNTNAP2), a gene implicated in a number of neurodevelopmental disorders, have highlighted the utility of this dataset for evaluating relationships between neuroimaging phenotypes and genetic variants of interest.

Conclusions: Via the PING Study, the scientific community will soon have access to a large cross-sectional pediatric imaging-genomics dataset. These data will enable a new wave of studies that can investigate the typical range of early human brain development with unprecedented sophistication and resolution.

Acknowledgments: NIDA RC2 DA029475 Grant (PI T. Jernigan), Sanofi Discovery-Innovation Grant (PI C. Bloss)

A5 Title: SORL1 and White Matter Tract Integrity: Predicting Risk for Late-Onset Alzheimer's Disease

Authors: Daniel Felsky, Jason Lerch, Mallar M. Chakravarty, Jon Pipitone, Tarek K. Rajji, Nancy J. Lobaugh, Benoit H. Mulsant, Bruce G. Pollock, James L. Kennedy, Aristotle N. Voineskos

Keywords: SORL1, diffusion tensor imaging (DTI), Alzheimer’s disease (AD), uncinate fasciculus

Aim of Investigation: Recently, a three-variant haplotype in the sortilin-related receptor, L-class, A repeat-containing (SORL1) gene has been implicated in late-onset Alzheimer's disease (LOAD). Evidence for the involvement of SORL1 in LOAD pathology is compelling: it is highly expressed in the CNS, binds isoforms of APOE with varying affinity, directs vesicular sorting of neurotrophins, and plays a key role in amyloid precursor protein (APP) processing. A recent study found association with the risk TAT haplotype and white matter (WM) atrophy and increased hyperintensities. However, there has been no investigation examining the relationship of risk variants in this gene with potentially vulnerable white matter circuitry using diffusion tensor imaging. Therefore, we investigated this haplotype's effect on WM circuitry in tracts connecting to the temporal lobe, as a potential predictor of LOAD.

Methods: 94 healthy, right-handed, Caucasian subjects aged 18-86 underwent a DTI scanning protocol on a 1.5T GE scanner with 23 directions and 3 repetitions that were then averaged and registered to the baseline image. Following whole brain tractography, we segmentated white matter tracts connecting to temporal lobe using a clustering method. The three variants in the previously identified SORL1 risk haplotype (rs668387, rs689021, and rs641120) were in perfect linkage disequilibrium in our sample; thus we examined effects of a single variant (rs668387) only. General linear models were used to assess the effect of genotypic group on fractional anisotropy (FA) of WM tracts with age as a covariate.

Results: Genotypic analysis revealed a potentially additive effect of rs668387 with FA of the right uncinate fasciculus (F2,90=3.12, P=0.049). When grouped by T (minor allele) homozygotes vs. C-carriers, there was a significant recessive effect where T homozygotes showed reduced FA of the right uncinate fasciculus (F1,91=4.15, P=0.044) and left fronto-occipital fasciculus (F1,91=4.67, P=0.033).

Conclusions: Our results suggest that variation in the SORL1 gene may confer risk for LOAD via effects on white matter tracts connecting to temporal lobe, that are often disrupted in early AD. Our findings align with a recent study implicating these variants with other white matter phenotypes. Our next steps are to analyze SORL1 risk variants and other neural vulnerability phenotypes for AD.

Aknowledgements: Canadian Institutes of Health Research (CIHR), National Alliance for Research on Schizophrenia and Depression (NARSAD), University of Toronto, Centre for Addiction and Mental Health (CAMH), and CAMH Foundation

A6 Title: Extending Parallel ICA to Three or More Modalities

Authors: David Boutte, Vince D. Calhoun, Jingyu Liu

Keywords: Data Mining, Statistics, ICA

Aim of Investigation: In emerging, empirically intensive fields such as population neuroscience; collection of high dimensional data from several sources is becoming standard practice. Magnetic resonance imagine (MRI) is powerful tool for the acquisition of multiple, reliable brain based phenotypes. However, recent twin studies suggest that MRI based phenotypes alone are not sufficient to explain all the inter-individual variance contained in brain structure. Clearly a large portion of this is due to both the genome and other environmental factors. This necessitates the careful collection and assessment of MRI, genetic, environmental and other relevant measurements. With thousands of genes, MRI measurements, neuropsychological assessments and their combinations, the complexity of analyzing this mountain of data seems overwhelming. How do we identify which of the many possible genetic and environmental factors shape a particular structural and functional property of the brain? A partial solution was suggested using an extension of independent component analysis (ICA) dubbed parallel ICA. ICA is a blind matrix factorization technique used to extract latent, statistically independent components from a set of measurements similar to principle component analysis(PCA). Parallel ICA extends this to concept to two modalities by simultaneously factorizing them while maximizing the correlation between the extracted components. This approach has been primarily applied to MRI and genetic data, however it is unable to incorporate information from more than two modalities, such as that contained in neuropsychological assessments and other environmental factors as well as MRI and genetics. We propose further generalizing this concept to include three, or more, modalities.

Methods: Our generalization of parallel ICA simultaneously factorizes three data sets; such as MRI, genetics and neuropsychological assessments into individual statistically independent components subject maximizing the average pairwise correlation between selected triplets. Our algorithm's performance is demonstrated via simulation and comparison with individual ICAs on simulated data sets.

Results: Results from simulations are presented detailing our generalization's improvement in performance over individual, unlinked ICAs. Simulated MRI, genetic and neuropsychological assessments are presented which have underlying correlated components. In these simulations, our three way ICA generalization produces those component triplets which have the underlying correlation and are closer (in terms of correlation) to the ground truth than components extracted using individual ICA on each modality separately. Pairwise component correlations are tested over a wide range, from 0.9 to 0.2 to asses overall performance and potential over fitting problems. Simulations on more than three modalities are also presented demonstrating our method's applicability to larger problems.

Conclusions: Great strides have been made in gathering new knowledge about genetic and and environmental factors that shape our brains. However, with the increase in available data a method is needed to analyze more complicated interactions between them. Previously, ICA has been applied to many of the concerned modalities separately and parallel ICA to MRI and genetics specifically. We present an extension of parallel ICA which is able to extract linked components from three data modalities. This allows us to incorporate information from not only MRI and genetics but also other measurements as well. Based upon simulation results, this method has the potential to be a powerful analysis tool in situations where multiple, interconnected measurements are collected upon a population.

Acknowledgment: This work was supported by grant 1R21DA027626 from the National Institutes of Health.

A7 Title: Integrating genetic, neuroimaging and behavioral data with correspondence analysis: An illustration in addictive populations.

Authors: Derek Beaton, Francesca Filbey, Hervé Abdi

Keywords: correspondence analysis, multivariate, addiction, SNPs, DTI

A challenge for imaging genetics researchers is to integrate multiple data types such as genetics, neuroimaging and behavioral measures. The structure of these data sets precludes the use of standard techniques because we need to integrate multiple tables of intrinsically different, and multivariate, data types. In our study we show the connections between genetic markers, white matter connectivity (via DTI) and impulsive behaviors in addiction populations. We present one possible solution to address individual and between group differences in complex, multi-table data. Some data in imaging genetics—such as groups, rating scales and SNPs—are nominal variables taking values 0 or 1. Other data, such as the brain imaging data DTI fractional anisotropy can be seen as frequency data that take values between 0 and 1. Nominal and frequency variables can be analyzed with Correspondence Analysis (CA), which is a variant of principal component analysis (PCA) tailored for multivariate qualitative data. CA, however, only analyzes one data set. In order to analyze the links between SNPs, DTI and impulsive behaviors in addiction types, we have developed a multi-table version of CA that integrates features of two techniques: STATIS and Multiple Factor Analysis (MFA). STATIS and MFA analyze the similarity between multiple tables (SNPs, DTI & impulsive behavior rating scales) in order to extract the information common to these data tables. These techniques compute a compromise that is a linear combination of all tables. Next, a generalized PCA of the compromise provides an optimal map of observations and measures. Additionally, each table (SNPs, DTI, behavior & groups) provides a map with respect to the (optimal) compromise. Finally, non-parametric statistics (permutation, bootstrap) are used to assess significance and stability of measures. We tested our approach on existing data from studies of addiction populations (recruited from the Albuquerque area), which included: 2.5×2.5×4mm DTI images, genomic DNA (Illumina chip) and impulsive behavior measured by the Barratt Impulsivity Questionnaire and the Impulsiveness and Sensation Seeking Scale. Our combined approach reveals which SNPs are related to differences in white matter structure and patterns of impulsive behavior between addiction types and individuals. We conclude that this approach is an extremely effective method to examine and test the complex relationships between genetics, neuroimaging and behavioral measures.

This research supported by NIDA 5K01DA021632 to FF

A8 Title:A multivariate investigation in high-dimensional Imaging Genetics studies using sparse Partial Least Squares and dimension reduction

Authors: Edith Le Floch, Philippe Pinel, Bertrand Thirion, Jean-Baptiste Poline, Vincent Frouin, Edouard Duchesnay

Keywords: Imaging Genetics, Multivariate Analysis, Partial Least Squares, Regularization, Dimension Reduction

Aims of investigation : Imaging genetics studies that include a large amount of both neuroimaging and genetic data have become a challenging issue for the last few years. Classical approaches rely on massive univariate pairwise linear analysis (Stein et al. 2010), ignoring the potential interactions between genes or between brain regions. This is why in this work we investigate a multivariate method in order to identify a set of Single Nucleotide Polymorphisms (SNPs) covarying with a set of neuroimaging phenotypes derived from functional Magnetic Resonance Imaging (fMRI).

Methods: PLS regression seems to be a good candidate, as it extracts pairs of covarying latent variables that are linear combinations of the variables from each block. However, the high dimensional setting raises critical overfitting issues with such approaches. To face these issues, regularization or variable selection can be used. Sparse versions of PLS regression, based on L1 regularization, have recently been proposed (LeCao et al. 2008, Vounou et al. 2010). However, whether these multivariate techniques can resist very high dimensions, remains an open question. Thus we propose a two-step strategy combining a first step of univariate feature selection with sparse PLS (sPLS). We compared (s)PLS and the two-step method on a dataset composed of 94 subjects, 622 534 SNPs and 34 regions of interest where lateralization indexes were computed from reading and speech comprehension contrast maps (Pinel et al. 2007). We tested the performance of each method by assessing whether the link obtained between the imaging and SNP datasets could generalize to new unseen test subjects, using a ten-fold cross-validation scheme. This link, the correlation between latent variables obtained on test samples, was then averaged over folds and the significance of this average test correlation was assessed with permutation procedures. Different parameter pairs of univariate filtering and sparsity were tested to investigate their influence on the performance of the two-step method, but we corrected empirical p-values for the number of pairs tested.

Results: We first observed that “sPLS alone” gives very low average test correlations whatever the sparsity. Then the two-step method outperforms sPLS and reaches a significant correlation when 1000 SNPs are retained by the univariate filter and 50 of these remaining SNPs finally selected by sPLS. This leads to the identification of 18 genes, such as PPP2R2B and RBFOX1, which have been reported to be linked with ataxia and a poor coordination of speech and body movements.

Conclusions: The originality of this work is to combine a univariate filter with sPLS in order to face exploratory imaging genetics studies. We show that univariate filtering is necessary to overcome the overfitting issue and still allows sparse PLS to extract a significant association between a multivariate genetic pattern and an imaging multivariate pattern.

A9 Title: Bayesian Mining of Image and Nonspatial Data

Authors: edward h Herskovits, rong chen

Keywords: data mining, bayesian analysis, neuroimaging, genetics

Aim of Investigation: To evaluate a Bayesian method for mining a combination of spatial (e.g., brain magnetic-resonance (MR)) and nonspatial data.

Methods: We have implemented a Bayesian data-mining framework for analyzing MR data--consisting of registered and segmented brain-MR images--in the context of categorical nonspatial data. We call this framework graphical model based multivariate analysis (GAMMA). GAMMA is based on discrete Bayesian networks (BNs), and the K2 data-mining metric, which computes the probability of a given BN model yielding the data at hand. We have also implemented a similar algorithm, without Markov random fields or spatial search, for analyzing microarray data. Using simulated data, we compared our Bayesian approach to conventional statistical approaches (i.e., mass-univariate analysis, and analyses based on general linear models (GLMs)) for analyzing categorical data; we computed sensitivity and specificity for detecting known associations in simulated spatial and non-spatial categorical data sets.

Results: Our BN-based analysis of non-spatial data has greater sensitivity and specificity than mass-univariate analysis based on the Fisher exact statistic. In addition, GAMMA has greater sensitivity and specificity than statistical parametric mapping for voxel-based morphometry or functional-MR analysis. Furthermore, our approach detects nonlinear multivariate associations that GLM-based approaches cannot, for both spatial and non-spatial data. For both spatial and non-spatial data, the Bayesian approach required no user input, whereas p-value thresholds were required for conventional analyses.

Conclusions: Our Bayesian approach to data mining makes more efficient use of data than mass-univariate analyses, and therefore may be more suitable to the analysis of undersampled data, which is commonly the case in image-based research. In addition can detect complex associations that GLM-based and mass-univariate analyses cannot. Another advantage of our approach is that there is no multiple-comparison problem, and no user-set thresholds required, because our algorithms return the most likely BN model that explain the associations manifest in the data at hand.

A10 Title: Schizophrenia-associated risk variant in GAD1 influences cortical thickness and task-induced deactivation patterns during working memory

Authors: Esther Walton, Stefan Brauns, Randy L. Gollub, Tonya White, Thomas H. Wassink, Vince D. Calhoun, Stefan Ehrlich

Keywords: cortical thickness, GAD1, COMT Val/Met, default mode network, schizophrenia

Aim of Investigation: Schizophrenia patients are characterized by reduced cortical thickness in multiple brain regions, a trait which is also considered as an intermediate phenotype for the disorder. Schizophrenia is also associated with reduced expression of the γ-Aminobutyric acid (GABA) synthesizing enzyme glutamic acid decarboxylase (GAD1) and the single nucleotide polymorphism rs3749034 in the corresponding gene. The aim of this study was to test whether this risk SNP is associated with reduced cortical thickness. Since it is known that the GABAergic and dopaminergic systems interact, we also examined whether associations between GAD1 rs3749034 and cortical thickness are modulated by the catechol-O-methyltransferase (COMT) Val158Met genotype. Finally, we investigated the connection between GAD1 and neuronal activity.

Methods: Imaging and genetic data from 94 healthy participants of the Mind Clinical Imaging Consortium (MCIC) study of schizophrenia were used to examine the associations between GAD1 genotype and cortical thickness as well as task-induced deactivation patterns during working memory.

Results: We found a robust reduction of cortical thickness by 7.54% in the left parahippocampal gyrus (PHG), a core region of the default mode network, in G allele homozygotes of GAD1 rs3749034. In stratified analyses according to the COMT Val158Met genotype, cortical thickness reductions of G allele homozygotes were only present in Val allele carriers. In addition, we found aberrant task-induced deactivation in G allele homozygotes during working memory processing.

Conclusions: Our findings support an interaction between GABAergic and dopaminergic schizophrenia risk variants, which jointly influence brain structure and function. The reduced cortical thickness in the PHG and reduced task-induced deactivation in the medial prefrontal cortex underline the significance of the GABAergic system and the default mode network in the etiology of schizophrenia.

A11 Title: Longitudinal genetic models of pediatric brain development

Authors: J. Eric Schmitt, Jay N. Giedd, Aaron F. Alexander-Bloch, Greg Wallace , Liv Clasen, Micheal C. Neale

Keywords: neuroimaging, longitudinal, genetics, twin, pediatric

Aim of Investigation: Several twin and family studies have established that global morphometric measures of brain structure are highly heritable during childhood and adolescence. Although longitudinal neuroimaging studies have demonstrated that the brain is dynamically remodeled in childhood, the importance of genes on driving the observed phenotypic changes remains poorly understood. The purpose of the present study was to quantify time-dependent changes in genetic variance in several morphometric brain measures by combining longitudinal models of growth with quantitative genetic analyses on extended pedigrees.

Methods: Serial high-resolution T1-weighted anatomic MRI brain scans were obtained as part of an ongoing longitudinal brain imaging project at the Child Psychiatry Branch of the National Institute of Mental Health. The sample included pediatric, adolescent, and young adult monozygotic twins (N=249), dizygotic twins (N=131), siblings of twins (N=110), and singleton family members (N=302). Up to 8 scans were performed per individual, with sibships containing up to 5 members. A total of 1748 scans were analyzed using automated segmentation and parcellation algorithms to generate 60 measures of brain volume, surface area, cortical thickness, and gyrification. Genetically informative latent growth curve models were then constructed and analyzed in Open Mx.

Results: Individual differences in brain volumes were significantly (p<0.0001) influenced by genetic factors, with heritabilities generally ranging from 0.8-0.9. With the exception of mesial temporal structures and the right occipital lobe, variance in surface area measures also was significantly influenced by genes. Cortical thickness heritability estimates were lower but also statistically significant (p < 0.001). Individual-specific environmental factors had highly significant (p <0.0001) influence on nearly all measures. Both familial environmental and twin specific environmental factors had weak effects and were removed from the model. Complex, statistically significant patterns of dynamic genetic influence on brain structure were observed for volumes, cortical thickness, and several measures of surface area. For example, lobar brain volumes demonstrated marked decrease in genetic variance with time (p-values ranging from 0.004 to <0.0001). Genetic variance sharply increased in white matter over the age range studied (p <0.0001). Measures of surface area and cortical thickness demonstrated similarly complex patterns.

Conclusions: The present study provides strong evidence that genetic variation plays an important role in driving the observed changes in several brain endophenotypes over childhood and adolescence.

A12 Title: Get Ready for Action: DRD4-7R functional brain activity to unpleasant stimuli

Authors: Jean Gehricke, James Fallon, Cyrus Caburian, Tugan Muftuler, Robert Moyzis

Keywords: Dopamine; Emotion; Risk taking; fMRI

Aim of Investigation: Dopamine is a key neurotransmitter, which regulates attention, emotional processing, motor activation, short-term memory, behavioral inhibition and risk taking. Previous research has shown that the dopamine receptor gene DRD4-7R has been associated with risk taking. It is not surprising that this gene is found more frequently in populations, such as early immigrants to the United States, who had to take great risks to travel long distances. Part of risk taking is an appropriate risk assessment, which includes paying close attention to unpleasant or threatening stimuli. The aim of the study was to examine functional brain activity in response to unpleasant images in individuals with DRD4-7R. Due to the greater propensity for risk taking, individuals with the DRD4-7R gene were expected to show greater functional brain activity in response to unpleasant stimuli in the frontal, temporal, occipital, parietal and limbic lobes, which are involved in paying attention, emotional involvement, and preparing for action.

Methods: Functional Magnetic Brain Imaging (fMRI) activity was studied in 22 participants using a Phillips Achieva 3T MR system. Half of the participants had the DRD4-7R gene while the other half had the more typical DRD4-4R gene. Participants were asked to look at unpleasant and neutral images. All participants were young adults (18 to 25 years of age) and abstained from drugs and alcohol use before the fMRI scan, which was verified with urinary and breath drug screens.

Results: Participants with the DRD4-7R gene showed increased brain activity in response to unpleasant images in all major brain areas, including the frontal, temporal, occipital, parietal and limbic lobes, which are involved in paying attention, showing empathy, and preparing for action. In contrast, participants with the DRD4-4R showed less brain activity, which was limited to parts of the frontal, occipital and parietal lobes, indicating passive processing of information. Intriguingly, no differences were found between the two genotypes in the subjective ratings of the images.

Conclusion: The findings show that people with a genotype that was naturally selected for risk taking pay more attention and are more emotionally involved in viewing unpleasant images.

Acknowledgements: This investigation was supported by Public Health Service research grants K01-DA25131 and M01-RR00827 to Jean Gehricke.

A13 Title: Molecular neuroimaging of reading disability: Allelic variation in DCDC2, KIAA0319, and TTRAP impacts human brain function

Authors: Natalie Cope, John D. Eicher, Haiying Meng, Christopher J. Gibson, Karl Hager , Cheryl Lacadie, Robert K. Fulbright, R. Todd Constable, Grier P. Page, Jeffrey R. Gruen

Keywords: Reading disability, dyslexia, functional MRI, DCDC2, KIAA0319

AIM OF INVESTIGATION: Reading disability (RD) is a complex genetic disorder with unknown etiology. Genes on chromosome 6p22, including DCDC2, KIAA0319, and TTRAP have been identified as candidate genes for RD. Imaging studies have meanwhile shown both functional and structural differences between brains of individuals with and without RD. There are limited association studies performed between RD genes, specifically genes on 6p22, and regional brain activation during reading tasks.

METHODS: Using eighteen variants in DCDC2, KIAA0319 and TTRAP and exhaustive reading measures, we first tested for association with reading performance in 82 parent-offspring families (326 individuals). Next, we evaluated the association of these variants with activation of sixteen brain regions of interest during four functional magnetic resonance imaging-reading tasks.

RESULTS: We nominally replicated associations between reading performance and polymorphisms of DCDC2 and KIAA0319. We then observed a number of associations with brain activation patterns during imaging-reading tasks with all three genes. The strongest association occurred between the left anterior inferior parietal lobe and complex tandem repeat BV677278 in DCDC2 (uncorrected p=0.00003, q=0.06726).

CONCLUSIONS: Activation patterns across regions of interest in the brain are influenced by DCDC2, KIAA0319 and TTRAP. When analyzed together from the same subjects, the combination of genetic and functional imaging data show a possible link between genes and brain functioning during reading tasks in subjects with RD. This study underscores the necessity for future investigations of RD to collect and to integrate behavioral, imaging, and genetic data together to determine etiologies and successful interventions for RD.

ACKNOWLEDGEMENTS: Support for JRG is from the National Institute of Health (NIH) R01 NS43530, and support for JDE is from NIH F31 DC012270 and T32 GM007223. Further support for RKF and CL is from R01 HD046171 and R01 EB006494-01. Support for CJG is from a Yale-Rosenberg Genetics Fellowship. Support for RTC is from R01 NS038467 and NS051622. Yale University has applied for a patent covering the complex tandem repeat and deletion in BV677278 (inventors: JRG and HM), and sublicensed it to JS Genetics, Inc. JRG is a founder and equity holder of JS Genetics, Inc. KH is currently employed by JS Genetics, Inc.

A14 Title: Genetic Determinants of Dopamine Effects on Skill Learning and Cortical Plasticity

Authors: Kristin M. Pearson-Fuhrhop, Brian Minton, Daniel Acevedo, Steven C. Cramer

Keywords: dopamine, gene score, plasticity, TMS, L-Dopa

AIM: Dopamine is important to many brain processes including motor function, learning, and plasticity. However, inconsistent results have been found in humans regarding effects on learning and plasticity by drugs that increase level of dopamine neurotransmission. Such dopaminergic drugs also have a mixed record in studies that aimed to improve outcome after stroke. The reasons for these disparities are unclear. Recent studies suggest that genetic polymorphisms underlie some differences between humans in learning and plasticity. The current study addressed the hypothesis that genetic variation in proteins that affect level of dopamine neurotransmission influence effects of dopaminergic drugs on learning and plasticity.

METHODS: In 50 healthy adults, the presence of genetic polymorphisms was assessed for five proteins that influence level of dopamine neurotransmission: dopamine transporter and catechol-O-methyltransferase, regulators of synaptic dopamine levels; and DRD1, DRD2, and DRD3, dopamine receptor subtypes. Each polymorphism increases dopaminergic tone. To evaluate polymorphism effects collectively, a dopamine gene score was calculated by assigning each gene a score of 0 if the polymorphism was absent and 1 if present, then summing all 5 values. Gene scores thus ranged from 0 (no polymorphisms increasing dopamine neurotransmission present) to 5 (all present). Subjects with the val66met BDNF polymorphism, known to affect cortical plasticity, were excluded. At baseline, performance was measured on a skilled motor task that required subjects to use the right index finger, and thus the first dorsal interosseous (FDI), to move a marble into wells on a test board; and the left motor cortex map area for right FDI was measured at rest. The next day subjects began 3 days of practice of this skilled motor task after ingesting either the dopamine precursor L-Dopa or placebo, in a randomized, double-blind manner. On day 5, task performance and TMS mapping were repeated. Using a crossover design, subjects repeated these procedures 14 days later, ingesting the pill (placebo vs. L-Dopa) not given previously, and using a novel test board.

RESULTS: In the setting of placebo, higher gene score (higher intrinsic dopamine neurotransmission) was associated with poorer motor skill learning (p=0.044). Increasing brain dopamine via L-Dopa administration: [1] affected learning in a manner that varied significantly (p=0.039) with gene score, with reduced learning among those with a lower gene score and improved learning among those with a higher gene score, as compared to placebo; and [2] affected cortical plasticity in a manner that varied significantly (p=0.006) with gene score, with map area shrinking across the week of training among those with a lower gene score and expanding among those with a higher gene score, as compared to placebo. Degree to which L-Dopa boosted learning correlated with degree to which L-Dopa boosted map area expansion in a manner that varied with gene score (p=0.038). Multivariate modeling disclosed that these findings were related to gene score, but not to any individual polymorphism.

CONCLUSIONS: Together, the results suggest that increases in levels of dopamine neurotransmission affect learning and cortical plasticity in a manner that is significantly influenced by genetic factors. This genetic influence was best described by a score based on multiple polymorphisms; such a gene score has proven useful in other human biological systems such as coronary disease, breast cancer, and prostate cancer. These findings provide insight into the variability of dopamine drug effects across subjects, and might allow for a polygenic stratification of patients undergoing a dopamine-based therapy.

ACKNOWLEDGEMENTS: Supported by the American Heart Association, R01 NS058755, and NIH P50 AG16573

A15 Title: Neural Activations During Auditory Oddball Processing As Endophenotypes for Schizophrenia and Psychotic Bipolar Disorder

Authors: Lauren E. Ethridge, Jordan P. Hamm, John R. Shapiro, Ann T. Summerfelt, Sarah K.Keedy, Michael C. Stevens, Godfrey Pearlson, Carol A. Tamminga, Nash N. Boutros, John A. Sweeney, Matcheri S. Keshavan, Gunvant Thaker, Brett A. Clementz

Keywords: psychosis, time-frequency, PCA, endophenotype, relatives

Aim of Investigation: Reduced amplitude of the P300 event-related potential in auditory oddball tasks may characterize schizophrenia (SZ), but is also reported in bipolar disorder. Similarity of auditory processing abnormalities among these diagnoses is uncertain given the frequent (and sometimes unclear) combination of both psychotic and nonpsychotic patients in bipolar samples. The extent to which SZ and bipolar disorder with psychosis (BPP) are similar on neurophysiological function requires considerable additional investigation. Also, typically only latency and amplitude of brain responses at selected sensors and singular time points are used to characterize neural responses. Comprehensive quantification of brain activations involving both spatio-temporal and time-frequency analyses could better identify unique auditory oddball responses among patients with different psychotic disorders.

Methods: Sixty SZ, 60 BPP, and 60 healthy subjects (H) from the multi-site data collection project Bipolar and Schizophrenia Network on Intermediate Phenotypes (B-SNIP) were compared on neural responses during an auditory oddball task using multi-sensor EEG. PCA was used to reduce multi-sensor data prior to evaluating group differences on voltage and frequency of neural responses over time. Variables that discriminated proband groups were compared to data from 53 clinically normal first-degree relatives of SZ (SZrel) and 48 normal first-degree relatives of BPP (BPrel) to evaluate for the possibility that specific measures of brain functioning have endophenotype characteristics.

Results: Linear discriminant analysis revealed five variables that best differentiated psychosis groups: (i) late beta activity to standards and (ii) late beta/gamma activity to targets discriminated BPP from other groups; (iii) mid-latency theta/alpha activity to standards and (iv) target-related voltage at the late N2 response discriminated both psychosis groups from H; and (v) target-related voltage during early N2 discriminated BPP from H. Two of these variables showed potential endophenotype characteristics: late beta activity to standards was also abnormal in BPrel and the late N2 response to targets was abnormal for relatives of all psychosis probands.

Conclusions: Although the P300 significantly differentiated psychotic groups from H, it did not uniquely discriminate groups beyond the above variables. No variable uniquely discriminated SZ, perhaps indicating utility of this task for studying psychosis-associated neurophysiology generally and BPP specifically. Late beta accentuation is unique to BPP and their first degree relatives, suggesting a genetic component to this variable and potential as a useful endophenotype for BPP. N2 amplitude is significantly decreased in all psychosis probands and their relatives, suggesting a more general genetic marker for psychosis risk.

Acknowledgements: Funding for this study was provided by NIH Grants MH077945, MH077862, MH077851, MH078113, and MH085485. Potential conflicts of interest: JAS reports Takeda, Pfizer, and Lilly consulting, and a Janssen research grant. CAT reports Intracellular Therapies (ITI, Inc.), PureTech Ventures, Eli Lilly Pharmaceuticles, Sunovion, Astellas, Merck ad hoc consulting, International Congress on Schizophrenia Research - unpaid volunteer, NAMI unpaid volunteer, American Psychiatric Association - Deputy Editor, and Finnegan Henderson Farabow Garrett & Dunner, LLP – Expert Witness as potential conflicts of interest.

A16 Title: Genetic interactions in focal adhesion and extracellular matrix pathways are associated with increase in ventricle size over time in the Alzheimer’s Disease Neuroimaging Initiative cohort

Authors: Mary Ellen Koran, Shashwath Meda, Tricia Thornton-Wells

Keywords: atrophy, pathway, epistasis, Alzheimer’s, MRI

Aim: Late onset Alzheimer disease (LOAD) has a complex genetic etiology, involving heterogeneity and gene-gene interactions. Recent GWAS in AD have led to the discovery of novel genetic risk factors; however, the investigation of gene-gene interactions in GWAS has been limited. Genetic studies often use binary disease status (case/control) as phenotype, or outcome variable, but studies of complex disease require rich phenotypic information that can be mapped to distinct genetic etiologies, which may involve gene-gene or gene-environment interactions. For brain-based diseases like LOAD, neuroimaging data can provide such quantitative traits (QTs). Many QTs in LOAD have been shown to correlate with disease status and to have greater sensitivity in detecting early pathological changes. In this study, we tested for association of gene-gene interactions with AD-related longitudinal MRI measures from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.

Method: We used genetic and imaging data from ADNI to test for gene-gene associations with AD-related longitudinal MRI changes using an additive model. We applied a pathway-driven approach in which only pairs of SNPs from genes belonging to a common biological pathway were tested. We used linear regression in INTERSNP software to test the ability of 151 million SNP-SNP models to predict 12-, 24- and 36-month annual atrophy rates of 13 different brain regions implicated in AD structural pathology, including medial-temporal lobe structures and the ventricles.

Result: The most significant results were seen in the right and left inferior lateral ventricals (RILV & LILV). 575 significant SNP-SNP interactions were identified in the RILV and 316 were seen in the LILV; 7 SNP-SNP interactions replicated over two of the three time points in the RILV. Four gene-gene interactions replicated across time and in both the RILV and LILV: SLIT3-EPHB1, and SV2C, FN1, and ITGA1 each interacting with LAMC1. Three of these 4 gene-gene replicates belonged to the extracellular matrix (ECM) receptor interaction KEGG pathway, and 2 of the 4 belonged to the focal adhesion (FA) KEGG pathway.

Conclusion: Using a pathway based approach for gene-gene interactions associated with AD related QTs, we were able to identify interactions that not only replicated over time but also over location. We then were able to use current knowledge of pathway interactions to deduce a plausible biological context for these significant interactions. There is a theory in AD that neuron loss is mediated through FA signaling. Proteins involved in FA are activated by β-amyloid and communicate with the ECM to induce cell cycle activation, which activates the loss of cell adhesions and leads to subsequent cell death. The ECM and FA pathways were implicated in AD-related neuron loss seen as MRI measurement changes in this study, supporting the relationship between FA and neuron loss in AD etiology.

A17 Title: Genetic variation within NTRK3 influences white matter integrity in healthy young adults

Authors: Meredith N Braskie, Omid Kohannim, Neda Jahanshad, Ming-Chang Chiang, Marina Barysheva, Kori Johnson, Katie L McMahon, Greig I de Zubicaray, Nicholas G Martin, Margaret J Wright, John M Ringman, Arthur W Toga, Paul M Thompson

Keywords: NTRK3, neurotrophin, white matter, fractional anisotropy, diffusion tensor imaging

Aim: The NTRK3 gene (also known as TRKC) encodes a high affinity receptor for the neurotrophin 3’-nucleotidase (NT3), which has been implicated in oligodendrocyte and myelin development. We previously found that common variants in genes related to neurotrophins and their receptors were associated with a diffusion tensor imaging measure of white matter integrity. Single nucleotide polymorphisms (SNPs) in NTRK3 have been identified as possible risk factors in schizophrenia, bipolar disorder, and obsessive-compulsive hoarding. We assessed how these SNPs related to white matter integrity.

Methods: We scanned 392 healthy adult twins and their siblings (mean age 23.6 ± 2.2 years) using 105-gradient diffusion tensor imaging at 4 Tesla. We assessed 18 NTRK3 SNPs associated with neuropsychiatric disorders, and tested their associations with voxelwise fractional anisotropy (FA), a widely-used measure of white matter integrity. We used multi-SNP regression to evaluate 1) the additive effect of all SNPs in a model acting jointly and 2) effects of individual SNPs in the model after adjusting for the effects of all others. We controlled for family relatedness, age, and sex and used the false discovery rate (FDR) method to correct for voxelwise multiple comparisons. We re-ran the model in a step-wise fashion, each time removing the SNP with the weakest association to FA.

Results: The model that optimally related to white matter FA (based on the highest FDR critical p), included five SNPs (rs1017412, rs2114252, rs16941261, rs3784406, and rs7176429; overall FDR critical p = 0.028; higher values denote stronger effects). White matter effects were broad, including the inferior longitudinal fasciculus and corpus callosum genu - regions frequently impaired in neuropsychiatric disorders. Lower FA in these regions was also associated with variants in genes that encode another neurotrophin receptor and a neurotrophin in an overlapping sample of subjects.

Conclusions: Variants in NTRK3 specifically, and in neurotrophin systems generally, may influence white matter integrity in regions vulnerable to neuropsychiatric disorders.

Acknowledgements: Funded by NIH grants HD050735, EB008432, EB008281, and EB007813, and Australian NHMRC grants 486682 and 389875). Additional funding was from the NIH (P50 AG-16570), the UCLA Easton Consortium for Biomarker and Drug Discovery in Alzheimer’s Disease, the NIH/National Library of Medicine (T15 LM07356) and an ARC Future Fellowship (FT0991634).

A18 Title: The Influence of Ancestry on Pediatric Neuroimaging Phenotypes

Authors: Ondrej Libiger, Cinnamon Bloss, B. F. Darst, A. J. Schork, C. Roddy, N. Akshoomoff, D.G. Amaral, B. J. Casey, L. Chang, T. Ernst, J.R. Gruen, W.E. Kaufmann, T. Kenet, D. N. Kennedy, S. S. Murray, E. R. Sowell, T. L. Jernigan, A.M. Dale, and N.J. Schork.

Keywords: ancestry, genetic background, neuroimaging phenotypes, neuroanatomy, admixture

Aim of Investigation: To explore the relationship between genetic background and neuroimaging-derived neuroanatomical phenotypes in a very large and diverse pediatric sample with both detailed imaging data as well as dense genome-wide genotype information.

Methods: We leveraged data from the Pediatric Imaging, Neurocognition, and Genomics (PING) Study. Funded through the American Recovery and Reinvestment Act, PING is focused on the development of a large cross-sectional pediatric imaging-genomics dataset that will be offered as a resource to the scientific community. Study participants between the ages of 3 and 20 years undergo brief cognitive testing, standardized structural, functional, and diffusion tensor imaging, and provide saliva samples for genomic studies. We leveraged genotype information obtained from the Illumina Human660W-Quad BeadChip and very large collection of individuals with genotype data whose ancestry is known to assign ancestry to each PING study participant. We then used generalized molecular analysis of variance (GAMOVA) and regression analysis to assess correlations between ancestry and imaging-derived neuroanatomical phenotypes while controlling for important covariates such as sex and age.

Results: We show that ancestry and genetic background explain an appreciable fraction of the global neuroanatomical variation for key phenotypes and brain regions. We also describe more detailed studies indicating which populations have genetic backgrounds that are more or less responsible for global neuroanatomical variation. We also find that the degree to which an admixed individual has genetic background contributions from one population or another correlates with neuroanatomical phenotypes as well.

Conclusions: Genetically-determined ancestry is correlated with many imaging-derived neuroanatomical phenotypes in pediatric populations though to varying degrees of strength. Some populations make a stronger contribution to global neuroanatomical variation than others, and the degree of admixture exhibited by an individual correlates with particular neuroanatomical phenotypes. The findings have important implications for both population genetic and human evolutionary studies as well as the clinical use of neuroimaging protocols.

A19 Title: The Implications of Ancestry for Neuroimaging-Guided Pediatric Diagnoses

Authors: Ondrej Libiger, Cinnamon Bloss, B. F. Darst, A. J. Schork, C. Roddy, N. Akshoomoff, D.G. Amaral, B. J. Casey, L. Chang, T. Ernst, J.R. Gruen, W.E. Kaufmann, T. Kenet, D. N. Kennedy, S. S. Murray, E. R. Sowell, T. L. Jernigan, A.M. Dale, and N.J. Schork.

Keywords: ancestry, genetic background, neuroanatomy, neuroimaging, normative reference

Aim of Investigation: To explore the impact of ancestry on the distribution of neuroanatomical phenotypes derived from imaging technologies that may impact clinical interpretations and decisions about an individual’s developing brain morphology.

Methods: We leveraged data from the Pediatric Imaging, Neurocognition, and Genomics (PING) Study. Funded through the American Recovery and Reinvestment Act, PING is focused on the development of a large cross-sectional pediatric imaging-genomics dataset that will be offered as a resource to the scientific community. Study participants between the ages of 3 and 20 years undergo brief cognitive testing, standardized structural, functional, and diffusion tensor imaging, and provide saliva samples for genomic studies. We leveraged genotype information obtained from the Illumina Human660W-Quad BeadChip and very large collection of individuals with genotype data whose ancestry is known to assign ancestry to each PING study participant. We explored the age and sex-specific distributional rank of each individual relative to other individuals both in an ancestry-informed matched subsample and in ancestry uninformed matched subsamples and compared the results.

Results: We show that ancestry and genetic background significantly impact the relative rank of an individual’s neuroanatomical profile relative to other individuals. We described the use of a matching strategy to correct for ancestry. We also show that not having appropriately matched individuals in a reference set for an assessment of a particular individual may impact clinical inference.

Conclusions: Variation and overt differences in genetically-determined ancestry among individuals used as a normative reference for assessing the development of an individual’s neuroanatomical profile can have important negative consequences. Care must be taken to ensure the appropriate construction of reference panels for clinical use.

A20 Title: Integrin signaling pathway predict decline in cerebral integrity: Transcriptome-wide analysis of cerebral aging traits

Authors: Kochunov P, Charlesworth JC, Hong EL, Curran JC, Glahn D, Blangero J

Keywords: transcriptom, DTI, GM thickness, aging, integrin

Aim: We used transcriptome-wide analysis to identify networks of genes that explain intersubject differences in cerebral integrity. The expression data were collected fifteen years prior to collection of cerebral integrity phenotypes, adding a prospective element to the brain measures

Methods: Transcriptome and cerebral integrity data were available for 369 (219F; aged 28-85, average=47.1±12.7 years) participants in the Genetics of Brain Structure and Function study. Genome-wide transcriptional profiles (Illumina Human 6v1) were obtained from fresh lymphocytes. Cerebral integrity phenotypes included whole-brain average thickness of cortical grey matter (GM) and fractional anisotropy of cerebral white matter (WM). Phenotypic correlations each of the 22,053 detectable transcripts and cerebral integrity phenotypes were calculated using a regression model that maximises the additional information provided by familial relationships while adjusting for the effects of age and sex method available in SOLAR. Ingenuity Pathways Analysis software was used for assessing network enrichment.

Results: 1,155 transcripts were significantly correlated (using 20% false-discovery rate) with both cerebral integrity traits. There was a striking enrichment of transcripts for genes involved in the integrin signalling pathway (27 genes; p=1.7x10-4) including significantly correlated transcripts for two alpha integrins (ITGA2B and ITGAL) and four beta integrins (ITGB1, ITGB3, ITGB5 and ITGB7). An additional 21 genes in the ERK (extracellular-regulated kinase) signalling pathway, activated by integrin signalling, were also correlated with GM thickness (p=8x10-3). The negative direction of the correlation suggested that a down-regulation of these processes co-occurred with decline in cerebral integrity.

Conclusions: The integrin signalling pathway is involved in cell-cell and cell-extracellular matrix (ECM) interactions; allowing rapid response to changes in environment. While integrins are generally associated with cell movement and attachment, integrin signalling and the ERK pathway play an important role in modulating neuronal survival by protecting against oxidative stress and apoptosis. Interactions between integrins and the ECM play a major role in neurogenesis such as axonal pathfinding and the regeneration of the sensory nervous system. Overall, we observed that decline of cerebral integrity was associated with down-regulation of signalling pathways that respond to environmental stress and facilitate remodelling and apoptosis. This support the hypothesis that cerebral aging is the consequence of reduced CNS neurogenisis as previously demonstrated in animal models.

A21 Title: Abnormal Temporal Lobe White Matter as a Biomarker for Genetic Risk of Bipolar Disorder

Authors: Philip R. Szeszko, Katie Mahon, Katherine E. Burdick, Toshikazu Ikuta, Patricia Gruner, Anil K. Malhotra

Keywords: bipolar disorder, white matter, unaffected sibling, diffusion tensor imaging

Aim of Investigation: Brain white matter abnormalities have been hypothesized to play a role in the neurobiology of bipolar disorder. The nature of these white matter abnormalities is not well characterized, however, and it is unknown whether they occur following disease onset or represent potential markers of genetic risk. In this study we thus examined the brain white matter using diffusion tensor imaging in patients with bipolar disorder, unaffected siblings of patients and healthy volunteers to identify potential white matter biomarkers of genetic risk. We hypothesized that white matter abnormalities identified in patients with bipolar disorder would also be evident in unaffected siblings compared to healthy volunteers.

Methods: We included 26 (11M/15F) patients with a diagnosis of bipolar disorder (mean age = 40.6, SD = 12.4) assessed using the Structured Clinical Interview for DSM-IV Disorders (SCID). In addition, we included 15 (6M/9F) unaffected siblings (mean age = 42.0, SD = 11.7) of patients with a diagnosis of bipolar disorder and 27 healthy volunteers (15M/12F; mean age = 40.8, SD = 12.5) who were all free from current Axis I disorders and lifetime major mood and psychotic disorders as assessed using the non-patient version of the SCID. All subjects received a diffusion tensor imaging exam on a 3T system that included volumes with diffusion gradients applied along 31 non-parallel directions and five volumes without diffusion weighting. Image processing was conducted using the tract-based spatial statistics program within FSL. A nonparametric voxelwise ANCOVA with subject-type (patients, unaffected siblings, and healthy volunteers) as the between-subjects factor and age as a covariate was carried out using permutation statistics via the Randomise tool in FSL with strict family-wise error correction and threshold-free cluster enhancement.

Results: Examination of fractional anisotropy across the entire white matter skeleton revealed three regions in the right temporal lobe that differed significantly (p< .05; family-wise error corrected) among the 3 groups. Post-hoc analyses indicated that unaffected siblings had fractional anisotropy that was intermediate to and significantly (p < .05) different from healthy volunteers and patients with bipolar disorder across the average of these 3 regions (healthy controls > unaffected siblings > bipolar disorder). Using the three clusters of white matter that differed significantly among groups as seed regions probabilistic tractography indicated that they lie along the inferior frontal occipital fasciculus, a large intrahemispheric association pathway that connects the orbital frontal, temporal, and occipital lobes.

Conclusions: Our results suggest that lower white matter integrity in the right temporal lobe may be a biomarker for genetic risk of bipolar disorder. The abnormal white matter regions identified in patients and siblings lie along a major white matter tract that interconnect brain structures strongly implicated in emotion regulation. It is conceivable that the attenuated nature of these white matter abnormalities present in siblings allow for some preservation of adaptive emotional regulation, whereas the more pronounced alterations observed in patients is related to the marked emotional dysregulation characteristic of BD. Further work is required to elucidate both the genetic mechanisms underlying right temporal white matter integrity in bipolar disorder as well as the functional implications of this abnormality.

A22 Title: Effect of serotonin transporter promoter polymorphism on 5-HTT binding potential in the human brain using [11C]DASB

Authors: Pia Baldinger, Georg Kranz, Andreas Hahn, Dan Rujescu, Siegfried Kasper, Rupert Lanzenberger

Keywords: Depression, PET, 5-HTT, serotonin, polymorphim

Aim of the investigation: Numerous studies investigated the role of the serotonin transporter (5-HTT) in the pathophysiology of psychiatric diseases, given its key function in the mechanism of action of the most commonly used antidepressants. The availability of 5-HTT was shown to be altered in depressive patient using positron-emission-tomography (PET), the potentially underlying agent being a length polymorphism of the 5-HTT gene’s promoter region, 5-HTTLPR. The long L allele was associated with a higher 5-HTT binding potential (BPND) in several brain regions investigated by PET in healthy volunteers. Here, we hypothesized whether 5-HTTLPR may have an effect on 5-HTT BPND in healthy and depressive subjects.

Methods: 14 healthy subjects (4 female, mean age ± SD= 32.07 ± 10.30 years) and 13 drug-free subjects suffering from major depression according to the Structured Clinical Interview for DSM IV (9 female, mean age ± SD= 44.00 ± 6.12 years) underwent a PET scan using the highly selective radioligand [11C]DASB. Additionally, 9ml EDTA blood samples were drawn from each subject for DNA isolation using the QIAGEN® QiaAMP DNA Mini Kit. Genotyping was performed using SEQUENOM®. PET scans were normalized to MNI-space and 5-HTT BPND quantification was carried out in PMOD 3.3 with thalamus, caudate, median (MRN) and dorsal raphe nuclei (DRN) as regions of interest, areas exhibiting high 5-HTT availability. To compare between genotypes, multivariate ANOVA was performed using 5-HTT BPND as dependent factor, health status and genotype as fixed factors using SPSS version 18.0 for Windows.

Results: MANOVA revealed a significant main effect of genotype on 5-HTT BPND (Pillai’s trace: F8, 38= 2.59; p= 0.23), but no main effect of health status and no interaction effect. Significant overall differences between genotype (LL, LS and SS) were found in DRN (F(8,38)=7.87; p=0.003), MRN (F(8, 38)= 4.64; p= 0.021) and caudate (F(8, 38)= 4.71; p= 0.02) and a trend for thalamus (F(8,38)= 3.01; p= 0.071). Post hoc comparisons revealed higher 5-HTT BPND in homozygote L carriers compared to S carriers in the DRN (p=0.04), MRN (p=0.013) and the caudate (p=0.018), corrected for multiple comparisons using Tukey Kramer.

Conclusions: These findings are in line with previous PET studies showing an effect of 5-HTTLPR on 5-HTT BPND in several brain regions in healthy volunteers. Moreover, health status had no influence on 5-HTT BPND, which is in agreement with the great inconsistencies in the literature showing either diminished or increased 5-HTTBP in depression. These results highlight the importance of 5-HTTLPR in neuropsychiatric research, given its significant effects on frequently reported endophenotypes of depression.

Acknowledgments: This research was partly supported by an intramural research grant of the Department of Psychiatry, Medical University of Vienna.

A23 Title: Using Alzheimer’s disease probability scores as a quantitative trait in a genome wide association study

Authors: Ramon Casanova, Shyh- Huei Chen, Mark A. Espeland, Fang-Chi Hsu

Keywords: Alzheimer's disease, large scale regularization, GWAS, machine learning, MRI

Aim of Investigation: The main goals of this work are: 1) to introduce a new index for early prediction of Alzheimer’s disease (AD) based on neuroimaging data that we call AD probability (ADP) index and 2) to present a new application for this type of index by using it as a quantitative trait in a genome-wide association study (GWAS). We propose the use of class-conditional probabilities modeled by regularized logistic regression as a new score to predict AD. Unlike previous methods, our approach does not rely on severe dimension reduction measures, but on large scale regularization where the input space for classification is based on voxels.

Methods: We used baseline MRI and GWAS data from 727 subjects downloaded from the Alzheimer’s disease Neuroimaging Initiative (ADNI) website. Of those, 205 were cognitively normal (CN), 171 were AD patients, and 351 had mild cognitive impaired (MCI). The MRI images were segmented, normalized and smoothed using the Statistical Parametric Mapping software. Quality control of the corresponding genotype data was performed using PLINK software remaining for the analyses a total of 556,860 single nucleotide polymorphisms (SNPs). The gray matter (GM) and white matter (WM) images of the CN and AD participants were vectorized and used as samples for two different classification analyses based on regularized logistic regression as implemented in the GLMNET library. Once the GM and WM models were fitted, they were used to estimate the ADP scores for the 727 subjects. These scores were used as quantitative traits in GWAS analyses. We explored the associations between each SNP and the quantitative traits using linear regression models incorporated in PLINK with adjustment for age, sex, handedness, education, and ten population stratification components.

Results: Using these quantitative traits, we obtained results that are consistent with previous reports but with much higher levels of significance. The APOE rs429358 SNP was mostly significant associated with the ADP GM score ( ), followed by the TOMM40 rs2075650 ( ). The analysis based on the ADP WM score again identified the APOE rs429358 SNP as the most significant SNP though the strength of the association was much weaker ( ).

Conclusions: We have introduced a new index for early prediction of AD that was used as a quantitative trait in a GWAS analysis with promising results. Very often, imaging genetics analyses are based on quantitative measures taken from specific brain areas while here we use a metric based on a set of more discriminative brain regions that describes the closeness of the brain spatial patterns found in a given individual with those found in AD patients. Future work will address the extension of this index to include multimodal information.

Acknowledgements This work was funded by the NHLBI Grants WHIMS MRI 2B and NHLBI-WH-11-10 and the Wake Forest University School of Medicine’s Department of Biostatistical Sciences. We thank the ADNI project for providing access to their database. Data used in this work were obtained from www.loni.ucla.edu/ADNI.

A24 Title: Computational Measurement of the Distance Between Sister Chromatids in Confocal Microscopy Images

Authors: Justina McEvoy, Michael Dyer, Stan Pounds

Keywords: interchromatid distances, subcellular imaging

AIM OF INVESTIGATION: The distance between sister chromatids provides insights regarding the stability of the chromatid cohesion complex during mitosis. Confocal microscopy is used to image the chromasomes and the distance between sister chromatids is determined by interactive visual inspection. This measurement method is labor intensive and has variability due to subjective interpretation of the images. Therefore, we developed a computational method to obtain these inter-chromatid distances.

METHODS: Image thresholding is used to separate the chromatids from the background. Putative chromatid-pairs are then identified as sets of neighboring pixels with intensities that exceed the threshold. Each putative chromatid-pair is then rotated by performing an intensity-weighted principal components analysis and subsequently divided into 5 slices according to the cumulative intensity ordered by the first principal component. Within each slice, intensity modes are identified and the distance between them along the second principal component is determined. The mode coordinates and inter-mode distances are then converted back to the original pixel-scale by inverting the principal components transformation. The cosines of the angles between modes are used to identify and exclude putative chromatids that represent folded or overlapping chromatid pairs. For each remaining putative chromatid pair, the final interchromatid distance is computed as the median of distances between paired intensity modes within each slice. The software produces a detailed report with all measurement results and annotated images for visual evaluation of performance.

RESULTS: The annotated images indicate that the method reliably identifies chromatid pairs, eliminates folded chromatids, and accurately measures the distance for the remaining chromatid pairs.

CONCLUSIONS: It is feasible to computationally measure the distance between sister chromatids in confocal microscopy images. Future research should develop techniques to measure the distance between folded sister chromatids.

Acknowledgements: We gratefully acknowledge funding support the American Lebanese-Syrian Associated Charities (ALSAC) and research grants awarded by the NIH.

A25 Title: Working memory development links MAOA polymorphism to aggressive behavior

Authors: Tim B. Ziermans, Iroise Dumontheil, Chantal Roggeman, Myriam Peyrard-Janvid, Hans Matsson, Juha Kere, Torkel Klingberg

Keywords: working memory; fMRI; aggression; single nucleotide polymorphism; MAOA

A developmental increase in working memory (WM) capacity is an important part of cognitive development and low WM capacity is a risk factor for developing psychopathology. Brain activity is a promising intermediate phenotype for linking genes to behavior and for improving our understanding of the neurobiology of WM development. We investigated gene-brain-behavior relationships by focusing on 18 single nucleotide polymorphisms (SNPs) located in six dopaminergic candidate genes (COMT, SLC6A3/DAT1, DBH, DRD4, DRD5, MAOA). Visuospatial WM (VSWM) brain activity, measured with functional magnetic resonance imaging (fMRI), and VSWM capacity were assessed in a longitudinal study of typically developing children and adolescents. Behavioral problems were evaluated using the Child Behavior Checklist (CBCL). One SNP (rs6609257), located ~ 6.6 kb downstream of the monoamine oxidase A gene (MAOA) on human chromosome X, significantly affected brain activity in a network of frontal, parietal and occipital regions. Increased activity in this network, but not in caudate nucleus or anterior prefrontal regions, was correlated with VSWM capacity, which in turn predicted externalizing (aggressive / oppositional) symptoms, with higher WM capacity associated with fewer externalizing symptoms. There were no direct significant correlations between rs6609257 and behavioral symptoms. These results suggest a mediating role of brain activity and WM capacity in linking the MAOA gene to aggressive behavior during development.

A26 Title: The serotonin-2A receptor gene (5-HTR2A) is associated with structural alterations in Williams Syndrome: a preliminary DARTEL-VBM based study

Authors: Shashwath A. Meda, Tracy P. Wilson, Tricia A. Thornton-Wells

Keywords: sociability, Williams syndrome, 5HTR2A, VBM

Aim: Williams syndrome (WS) is a neurodevelopmental disorder caused by a hemizygous microdeletion on chromosome 7q11 involving approximately 25 genes. An intriguing component of the WS behavioral phenotype is hypersociability contrasted with significant nonsocial fears. Multiple studies have found the serotonin-2A receptor gene (5-HTR2A) to be associated with various neuropsychiatric phenotypes, such as OCD, drug addiction and response to treatment for major depressive disorder. Unschuld et al. (2007) found an association between 5-HTR2A and sociability (and conversely, personality disorders) that might be of relevance to inter-subject variability in the WS hypersociability phenotype. Specifically, the T allele of the rs6311 promoter polymorphism was associated with higher sociability ratings. In this context, we sought to examine brain volumetric changes associated with rs6311 in a cohort of WS subjects.

Methods: We evaluated 24 WS individuals (19-38 yo; 14 males; 80% right handed). All participants provided written informed consent. T1-weighted 3D scans were obtained on all individuals, on a 3T Philips Allegra Scanner housed at the Vanderbilt University Institute of Imaging Science. The rs6311 variant was genotyped for all subjects using a standard Taqman assay. Scans were preprocessed using SPM8 software. Image segmentation was performed using DARTEL-VBM in the VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm). Corresponding gray matter (GM) segmentations from each subject were modulated to adjust for non-linear warping effects. Modulated images were normalized to MNI space and smoothed using an 8mm FWHM isotropic Gaussian kernel. Corrections were directly applied to adjust for variation in global GM volume. A one-way ANOVA assessed voxel-wise group differences in GM volume across the whole brain using the corresponding subject genotypes as fixed factors (CC (N=4), CT (N=16), TT (N=4)).

Results: T allele carriers had reduced regional GM volume, primarily in the hippocampus, parahippocampal gyrus and amygdala, compared to subjects with theCC genotype. In contrast, C allele carriers had reduced GM volume, compared to subjects with the TT genotype, in a more widespread circuitry involving the fronto-temporal regions (including the inferior and middle frontal gyri and inferior and superior temporal gyri), middle occipital gyrus, precentral gyrus and cuneus. Results reported are at a 0.05 uncorrected alpha level.

Conclusion: In summary, our preliminary results indicate that the T allele, which was previously associated with greater sociability, is associated with reduced GM volume in brain areas involved in anxiety and emotion regulation. In addition, our results suggest the C allele is associated with reduced GM volume in areas governing executive functioning, response inhibition, and auditory and language processing, consistent with previous studies suggesting the C allele is associated with OCD, drug addiction and personality disorders, OCD in which the above functionality is often compromised.

A27 Title: The schizophrenia risk gene GAD1 (GAD67) promoter variants and fronto-limbic system disconnectivity

Authors: Lett, T.A.P., Chakravarty, M. M., Lerch, J. P., Felsky, D., Pipitone, J., Daskalakis, Z. J., Mulsant, B.H., Kennedy, J. L., Voineskos, A. N.

Keywords: Schizophrenia, Genetics, GAD1, MRI, fronto-limbic circuitry

Aim of Investigation: Postmortem schizophrenia studies implicate hypofunction of a subset of GABAergic interneurons in the prefrontal cortex and the hippocampus. Single nucleotide polymorphisms (SNPs) in the 5’ flanking region of the glutamic acid decarboxylase (GAD1) gene have been associated with schizophrenia, mRNA levels in prefrontal cortex (PFC), and dorsal PFC activation during working memory tasks. The purpose of this study was to examine GAD1 schizophrenia risk SNPs and their effects on fronto-limbic circuitry.

Methods: We genotyped two functional SNPs in the 5’ flanking region of GAD1 in 69 Caucasian healthy individuals (age range 18-60) that had undergone MRI and DTI scans. The effect of GAD1 genotype on hippocampal volume, temporal and frontal cortical thickness, and fronto-temporal white matter tract integrity was assessed. For DTI, 23 directions and 2 b=0 images were obtained and 3 repetitions of the entire sequence was performed and then averaged. Segmentation and measurement (fractional anisotropy) of the left and right uncinate, arcuate, and cingulate fasciculi was performed. For MRI, automated measures of bilateral frontal and temporal cortical thickness were extracted in addition to bilateral hippocampus volumes.

Results: Both GAD1 SNPs were associated with left hippocampal volume (rs1978340: F65,1=7.658 p(uncorrected) = 0.007; rs3749034: F65,1= 5.513, p(uncorrected) = 0.022). The rs3749034 SNP was also associated with fractional anisotropy in the left uncinate fasciculus (F65,1=5.249, p(uncorrected) = 0.025).

Conclusions: Our results highlight the potential impact of GAD1 promoter variants on fronto-limbic circuitry. They suggest that schizophrenia risk variants in the GAD1 region may confer risk for disease via their effects on this core vulnerability brain network. Our next steps are to precisely localize thickness vulnerability regions in frontal and temporal lobes and relate these findings to core cognitive deficits associated with GABA-ergic dysfunction in schizophrenia.

A28 Title: TwinMARM: Two-stage Multiscale Adaptive Regression Methods for Twin Neuroimaging Data

Authors: Yimei Li, John H. Gilmore, Jiaping Wang, Martin styner, Weili lin, Hongtu Zhu

Keywords: Heritability, Multiscale adptive regression model, smooth,structural equation model, Twin study

Twin imaging studies have been valuable for understanding the relative contribution of the environment and genes on brain structures and their functions. Conventional analyses of twin imaging data include three sequential steps: spatially smoothing imaging data, independently fitting a structural equation model at each voxel, and finally correcting for multiple comparisons. However, conventional analyses are limited due to the same amount of smoothing throughout the whole image, the arbitrary choice of smoothing extent, and the decreased power in detecting environmental and genetic effects introduced by smoothing raw images. The goal of this article is to develop a two-stage multiscale adaptive regression method (TwinMARM) for spatial and adaptive analysis of twin neuroimaging and behavioral data. The first stage is to establish the relationship between twin imaging data and a set of covariates of interest, such as age and gender. The second stage is to disentangle the environmental and genetic influences on brain structures and their functions. In each stage, TwinMARM employs hierarchically nested spheres with increasing radii at each location and then captures spatial dependence among imaging observations via consecutively connected spheres across all voxels. Simulation studies show that our TwinMARM significantly outperforms conventional analyses of twin imaging data. Finally, we use our method to detect statistically significant effects of genetic and environmental variations on white matter structures in a neonatal twin study.

A29 Title: Using Alzheimer’s disease probability scores as a quantitative trait in a genome wide association study

Authors: Ramon Casanova, Shyh- Huei Chen, Mark A. Espeland and Fang-Chi Hsu

Aim of Investigation: The main goals of this work are: 1) to introduce a new index for early prediction of Alzheimer’s disease (AD) based on neuroimaging data that we call AD probability (ADP) index and 2) to present a new application for this type of index by using it as a quantitative trait in a genome-wide association study (GWAS). We propose the use of class-conditional probabilities modeled by regularized logistic regression as a new score to predict AD. Unlike previous methods, our approach does not rely on severe dimension reduction measures, but on large scale regularization where the input space for classification is based on voxels.

Methods: We used baseline MRI and GWAS data from 727 subjects downloaded from the Alzheimer’s disease Neuroimaging Initiative (ADNI) website. Of those, 205 were cognitively normal (CN), 171 were AD patients, and 351 had mild cognitive impaired (MCI). The MRI images were segmented, normalized and smoothed using the Statistical Parametric Mapping software. Quality control of the corresponding genotype data was performed using PLINK software remaining for the analyses a total of 556,860 single nucleotide polymorphisms (SNPs). The gray matter (GM) and white matter (WM) images of the CN and AD participants were vectorized and used as samples for two different classification analyses based on regularized logistic regression as implemented in the GLMNET library. Once the GM and WM models were fitted, they were used to estimate the ADP scores for the 727 subjects. These scores were used as quantitative traits in GWAS analyses. We explored the associations between each SNP and the quantitative traits using linear regression models incorporated in PLINK with adjustment for age, sex, handedness, education, and ten population stratification components.

Results: Using these quantitative traits, we obtained results that are consistent with previous reports but with much higher levels of significance. The APOE rs429358 SNP was mostly significant associated with the ADP GM score (), followed by the TOMM40 rs2075650 (). The analysis based on the ADP WM score again identified the APOE rs429358 SNP as the most significant SNP though the strength of the association was much weaker (). 21104.1−x< p10101.2−x < p12106.6−x < p

Conclusions: We have introduced a new index for early prediction of AD that was used as a quantitative trait in a GWAS analysis with promising results. Very often, imaging genetics analyses are based on quantitative measures taken from specific brain areas while here we use a metric based on a set of more discriminative brain regions that describes the closeness of the brain spatial patterns found in a given individual with those found in AD patients. Future work will address the extension of this index to include multimodal information.

Acknowledgements: This work was funded by the NHLBI Grants WHIMS MRI 2B and NHLBI-WH-11-10 and the Wake Forest University School of Medicine’s Department of Biostatistical Sciences. We thank the ADNI project for providing access to their database. Data used in this work were obtained from www.loni.ucla.edu/ADNI.

A30 Title: Projection Regression Models (PRM) for High Dimensional Imaging Responses

Authors: Ja-an Lin, Hongtu Zhu, Rebecca Knickmeyer, Martin Styner, John Gilmore, and Joseph Ibrahim

Introduction: The statistical power of traditional statistical approaches to detect the association between high dimensional imaging measures and covariates of interests are relatively low. We develop a new statistical tool PRM to address this issue. We use simulation studies to compare PRM with existing methods in terms of both type I errors and power.

Method: Our PRM method consists of dimension reduction and hypothesis testing on the high dimension imaging responses by properly weighting all imaging responses, while accounting for the space of explanatory variables and the case that the sample size N is relatively smaller than the number of components in imaging responses q. The PRM consists of 3 key steps: Pre-screening We first fit a univariate linear regression and test a specific hypothesis of interest to each component of imaging responses. Then, we select a subset of multivariate imaging responses with smallest p-values [1]. Estimation and Projection The weights for the selected imaging responses are estimated by maximizing the ratio of projected variance of explanatory variables under null hypothesis over the projected variance of random errors [2]. Then, we apply sparse principal component analysis on weight estimation by further filtering out irrelevant imaging responses [3]. Then, we project the original multivariate responses vector into a small weighted response vector and then fit a multivariate linear model. Hypothesis Testing We propose a test procedure based on a wild-bootstrap method to test for an association between the projected weighted multivariate imaging measures and the covariates of interest.

Result: Simulation Study The simulated data were generated as follows: 1) 5 explanatory variables: 1 for the additive effect from single nucleotide polymorphysm (SNP) with different minor allele frequencies varying (MAF) from 0.05 to 0.5; 1 for binomial disease status with prevalence rate 0.5; 3 multivarate normal variables with standard deviation 1 and correlation 0.3 2) 4 scenarios of imaging responses: all from multivariate normal distribution; 2 responses have variance 0.9 and correlation 0.6; all other responses have variance 0.1 and the same correlation 0.1. Only the 1st response is associated with SNP effect, equal to 0.5, and the 2nd response is associated with disease status effect, equal to 0.5. The hypothesis of interest is the SNP effect. We compared PRM with the principle component (PC) method and the false discovery rate (FDR) method. The results shown below indicate that the type I errors are controlled better and more stable in PRM, and the power are generally higher than the other two methods. Real Data Analysis: The data set from a neonatal study is used to assess the impact of common SNPs in putative psychiatric genes on early age brain development. The study recruited 237 pregnant women in their 2nd trimester. Each subject had 1 visit with T1-weighted MRI scan, demographic and genetic information collection. There were 47 regions of interest defined from the T1-weighted images by non-linear warping of a parcellation atlas template [4][5]. The demographic information includes gender, gestational age at birth, age after birth in days and intracranial volume (ICV) of the infants. There are 128 male and 109 female infants with average gestational age 264.0 (SD ±18.91), age after born in days 30.2 (SD ±17.80) and ICV 481799.9 (SD ±61528.96). Moreover, 9 genetic variants presented in SNPs from 6 genes were collected. Applying PRM, PCA and FDR on the data, only PRM detected the significance of the SNP rs6675281 on the gene DISC1 (with p-value 0.016) and the SNP rs35753505 on the gene NRG1 (with p-value 0.0136) which have important effect on human early age brain development.

Conclusion: We expect that this new method will lead to new findings with providing more powerful and reliable statistical result in clinical applications, especially in the analysis of high dimensional imaging response.