A Baseline for the Multivariate C...
SYSTEMS NEUROSCIENCE critical functions such as vision, audition, motor planning, and directing attention (Calhoun et al., 2002a Beckmann et al., 2005 Damoiseaux et al., 2006 Smith et al., 2009). These networks show surprisingly consistent, though not identical, patterns of activa- tion in the presence or absence of a particular task (Calhoun et al., 2008a Harrison et al., 2008 Laird et al., 2009 Smith et al., 2009), and are often acquired while subjects are at rest. Despite evalua- tion during a relatively unconstrained state, resting-state networks (RSNs) exhibit high reproducibility (Damoiseaux et al., 2006) and 1 IntroductIon Measurement of the blood oxygen level-dependent (BOLD) signal with functional magnetic resonance imaging (fMRI) has become a powerful tool for studying large-scale in vivo brain function. Following the seminal discovery by Biswal et al. (1995) that distinct brain regions exhibit synchronous fluctuations in intrinsic activity, our understanding of so-called functional connectivity has grown substantially. Several different methods have successfully deline- ated a large number of temporally coherent networks that subserve A baseline for the multivariate comparison of resting-state networks Elena A. Allen1*, Erik B. Erhardt1, Eswar Damaraju1, William Gruner1,2, Judith M. Segall1,3, Rogers F. Silva1,2, Martin Havlicek1,2, Srinivas Rachakonda1, Jill Fries1, Ravi Kalyanam1,2, Andrew M. Michael1, Arvind Caprihan1, Jessica A. Turner1,4, Tom Eichele5, Steven Adelsheim4, Angela D. Bryan1,6,7, Juan Bustillo4,8, Vincent P. Clark1,6,8, Sarah W. Feldstein Ewing1, Francesca Filbey1,9, Corey C. Ford10, Kent Hutchison1,6,8, Rex E. Jung1,11, Kent A. Kiehl1,6,8, Piyadasa Kodituwakku12, Yuko M. Komesu13, Andrew R. Mayer1,10, Godfrey D. Pearlson14,15,16, John P. Phillips1,10, Joseph R. Sadek4,17, Michael Stevens14,15, Ursina Teuscher1,6, Robert J. Thoma1,4,6 and Vince D. Calhoun1,2,4,8,15 1 The Mind Research Network, Albuquerque, NM, USA 2 Department of Electrical and Computer Engineering, University of New Mexico, Albuquerque, NM, USA 3 Department of Family and Community Medicine, University of New Mexico, Albuquerque, NM, USA 4 Department of Psychiatry, University of New Mexico, Albuquerque, NM, USA 5 Department of Biological and Medical Psychology, Faculty of Psychology, University of Bergen, Bergen, Norway 6 Department of Psychology, University of New Mexico, Albuquerque, NM, USA 7 Center on Alcoholism Substance Abuse and Addiction, University of New Mexico, Albuquerque, NM, USA 8 Department of Neuroscience, University of New Mexico, Albuquerque, NM, USA 9 Center for BrainHealth, School of Behavioral and Brain Sciences, University of Texas at Dallas, Dallas, TX, USA 10 Department of Neurology, University of New Mexico, Albuquerque, NM, USA 11 Department of Neurosurgery, University of New Mexico, Albuquerque, NM, USA 12 Center for Development and Disability, University of New Mexico, Albuquerque, NM, USA 13 Department Obstetrics and Gynecology, University of New Mexico, Albuquerque, NM, USA 14 Olin Neuropsychiatry Research Center, The Institute of Living, Hartford, CT, USA 15 Department of Psychiatry, Yale University, New Haven, CT, USA 16 Department of Neurobiology, Yale University, New Haven, CT, USA 17 Behavioral Health Care Line, New Mexico VA Health Care System, Albuquerque, NM, USA As the size of functional and structural MRI datasets expands, it becomes increasingly important to establish a baseline from which diagnostic relevance may be determined, a processing strategy that efficiently prepares data for analysis, and a statistical approach that identifies important effects in a manner that is both robust and reproducible. In this paper, we introduce a multivariate analytic approach that optimizes sensitivity and reduces unnecessary testing. We demonstrate the utility of this mega-analytic approach by identifying the effects of age and gender on the resting-state networks (RSNs) of 603 healthy adolescents and adults (mean age: 23.4 years, range: 12���71 years). Data were collected on the same scanner, preprocessed using an automated analysis pipeline based in SPM, and studied using group independent component analysis. RSNs were identified and evaluated in terms of three primary outcome measures: time course spectral power, spatial map intensity, and functional network connectivity. Results revealed robust effects of age on all three outcome measures, largely indicating decreases in network coherence and connectivity with increasing age. Gender effects were of smaller magnitude but suggested stronger intra-network connectivity in females and more inter-network connectivity in males, particularly with regard to sensorimotor networks. These findings, along with the analysis approach and statistical framework described here, provide a useful baseline for future investigations of brain networks in health and disease. Keywords: fMRI, functional connectivity, resting-state, independent component analysis, connectome Edited by: Silvina G. Horovitz, National Institutes of Health, USA Reviewed by: Scott K. Holland, Cincinnati Children���s Research Foundation, USA *Correspondence: Elena A. Allen, The Mind Research Network, 1101 Yale Boulevard NE, Albuquerque, NM 87106, USA. e-mail: firstname.lastname@example.org Frontiers in Systems Neuroscience www.frontiersin.org February 2011 | Volume 5 | Article 2 | 1 Original research article published: 04 February 2011 doi: 10.3389/fnsys.2011.00002
moderate to high test-retest reliability (Franco et al., 2009 Shehzad et al., 2009 Zuo et al., 2010), suggesting a robust examination of the intrinsic functional architecture, or ���connectome,��� of the human brain (Biswal et al., 2010). Because functional connectivity between regions is believed to characterize large-scale system integrity (Van Dijk et al., 2010), there is great interest in understanding the variability of these net- works in normal development and clinical contexts. Studies of the default-mode network (DMN), a set of brain regions preferentially active when subjects are not focused on the external environment (Raichle et al., 2001 Buckner et al., 2008), have established that this network not only shows a high degree of heritability (Glahn et al., 2010), but also shows alterations in a number of different neuro- logical disorders (see Greicius, 2008 and Broyd et al., 2009 for recent reviews). For example, in autism, functional connectivity between DMN regions is substantially reduced, though coactivation within the dorsal attention network, a set of brain regions implicated in directing attention during cognitively demanding tasks, appears relatively unaffected (Kennedy and Courchesne, 2008). The parallel between altered connectivity (specific to regions associated with internal, self-referential processes) and symptoms that character- ize autism suggests that straightforward investigations into func- tional connectivity can elucidate the etiology of complex disorders. Similar success has been found with regard to schizophrenia, where impaired modulation of the DMN has been observed in schizo- phrenia patients as well as their first-degree relatives, identifying an endophenotype based on large-scale connectivity (Whitfield- Gabrieli et al., 2009 Abbott et al., 2010). Furthermore, increased connectivity between particular DMN regions is associated with the severity of positive symptoms, suggesting a correspondence between specific ���hyperconnectivity��� and psychosis (Garrity et al., 2007 Whitfield-Gabrieli et al., 2009). The spectral properties of net- work activation in schizophrenia have also been explored, revealing a signature of reduced low frequency power and increased high frequency power in the DMN as well as many other RSNs (Garrity et al., 2007 Calhoun et al., 2008b, 2009). While multiple aspects of intrinsic functional connectivity show potential for clinical applications, the utility of network evalua- tion as a reliable diagnostic tool depends on the ability to inter- pret aberrant findings in the presence of an appropriate baseline. Fundamental factors, such as age and gender, are expected to exert large influences on functional connectivity based on their strong associations with underlying anatomy. For instance, most cortical regions show rapid gray matter loss as the brain matures through adolescence, followed by more gradual reductions in adulthood and advanced aging (Good et al., 2001 Sowell et al., 2003 Tamnes et al., 2010), though this trend is heterogeneous across structures and par- ticularly variable in subcortical regions (��stby et al., 2009). White matter shows a different developmental trajectory, with volume and tract integrity peaking in adulthood (approximately 25���35 years of age) then declining slowly with age (Sowell et al., 2003 Sullivan and Pfefferbaum, 2006 Tamnes et al., 2010). Structural differences are also observed between genders effects are smaller and some findings lack consistency, however studies concur that females show modest increases in gray matter volume localized to frontal, tempo- ral, and parietal cortices and basal ganglia (BG) structures (Good et al., 2001 Luders et al., 2005, 2009 Sowell et al., 2007). As anticipated, recent investigations have identified effects of age and gender on functional connectivity. With regard to age, reports suggest network maturation in childhood (Szaflarski et al., 2006 Karunanayaka et al., 2007 Fair et al., 2008), pro- gressive decreases in network mutual influences throughout adolescences into adulthood (Stevens et al., 2009), followed by decreases in functional connectivity and coherence in middle and late adulthood (Andrews-Hanna et al., 2007 Damoiseaux et al., 2008 Esposito et al., 2008 Koch et al., 2009 Biswal et al., 2010). Gender-related differences have received less attention but there appears to be some consensus of slightly greater con- nectivity in females localized to the precuneus and posterior cingulate cortex (Bluhm et al., 2008 Biswal et al., 2010). While these studies establish the influence of age and gender on func- tional connectivity, most unfortunately limit their investigations to the DMN, creating a dearth of reported effects with regard to other regions. In part, this bias reflects the unique function of the DMN related to internal mental processes and the desire to explicitly explore the ���cognitive baseline��� (Raichle et al., 2001). However, the relatively narrow scope of prior studies may also reflect the difficulty and somewhat overwhelming nature of investigations of full brain connectivity (Bullmore and Sporns, 2009). As the dimensions of data increase, so do the challenges associated with each analysis step, extending from data collection and processing to interpretation and visualization (Biswal et al., 2010 Costafreda, 2010). Given the need for a more comprehensive understanding of functional connectivity and the methodological challenges associ- ated with such a pursuit, the current study has two primary goals. First, we aim to present a statistical framework optimized for the analysis of large datasets that can be easily applied to investiga- tions in other areas. We advocate a hierarchical approach where multivariate models are used first to identify important covariates, reducing the number of subsequent univariate tests and decreasing the risk of spurious findings. For multivariate analyses, we exploit the autoregressive structure common to many types of data and recommend appropriate dimension reduction of response variables to enhance the sensitivity and specificity of model estimation. Our second goal is to apply this statistical framework in a detailed and careful investigation of the effects of age and gender on large- scale resting-state functional connectivity throughout the brain. To this end, we focus our analysis on data from a large number of healthy subjects (M = 603) collected on a single instrument, and employ group independent components analysis (GICA) to identify a set of robust and reliable RSNs (Calhoun et al., 2001). We examine the effects of age and gender on three ICA-derived outcome vari- ables describing distinct but complementary facets of functional connectivity. These include (1) the power spectra of RSN time course (TCs), related to level of coherent activity within a network (2) the intensities of RSN spatial map (SMs), related to the con- nectivity and degree of coactivation within a network and (3) the functional network connectivity (FNC Jafri et al., 2008), related to the connectivity between networks. Furthermore, we consider these outcome measures as a function of local gray matter concentration (GMC) to determine the extent to which functional changes reflect those observed in the structural domain (Damoiseaux et al., 2008 Glahn et al., 2010). Frontiers in Systems Neuroscience www.frontiersin.org February 2011 | Volume 5 | Article 2 | 2 Allen et al. A baseline for network comparisons
13 and 30 years old). We therefore use a normalizing transforma- tion, log(age), to reduce the leverage of older subjects in regression analyses (Figure 2A). 2.2 data acquIsItIon All images were collected on a 3-Tesla Siemens Trio scanner with a 12-channel radio frequency coil. High resolution T1-weighted structural images were acquired with a five-echo MPRAGE sequence with TE = 1.64, 3.5, 5.36, 7.22, 9.08 ms, TR = 2.53 s, TI = 1.2 s, flip angle = 7��, number of excitations = 1, slice thickness = 1 mm, field of view = 256 mm, resolution = 256 �� 256. T2*-weighted functional images were acquired using a gradient-echo EPI sequence with TE = 29 ms, TR = 2 s, flip angle = 75��, slice thickness = 3.5 mm, slice gap = 1.05 mm, field of view 240 mm, matrix size = 64 �� 64, voxel size = 3.75 mm �� 3.75 mm �� 4.55 mm. Resting-state scans were a minimum of 5 min, 4 s in duration (152 volumes). Any additional volumes were discarded to match data quantity across participants. Subjects were instructed to keep their eyes open during the scan and stare passively at a foveally presented fixation cross, as this is suggested to facilitate network delineation compared to eyes-closed conditions (Van Dijk et al., 2010). 2.3 data PreProcessIng Functional and structural MRI data were preprocessed using an automated preprocessing pipeline and neuroinformatics system (Figure 1, step 1) developed at MRN (Bockholt et al., 2009) and based around SPM51. Following the completion of a scan, data are automatically archived and copied to an analysis directory where preprocessing is performed. In the functional data pipeline, the first four volumes are discarded to remove T1 equilibration effects, images are realigned using INRIalign, and slice-timing correction is applied using the middle slice as the reference frame. Data are then spatially normalized into the standard Montreal Neurological Institute (MNI) space (Friston et al., 1995), resliced to 3 mm �� 3 mm �� 3 mm voxels, and smoothed using a Gaussian kernel with a full-width at half-maximum (FWHM) of 10 mm. To ensure quality and consistency of spatial normalization across subjects we calculate the spatial correlation between each subjects normalized data and the EPI template, as well as the degree of intersection between the EPI mask (determined by retaining voxels greater than the mean of the distribution) and the subject mask (determined by the same criteria). For the analysis presented here, these two metrics flagged datasets from 35 subjects, three of which were uncorrectable due to incomplete brain coverage and one that was unusable due to large signal dropout. The remaining 31 scans were corrected by manually reorienting the original images (shift in the yaw direction), then were re-preprocessed through the pipeline. Subsequent to automated preprocessing, the data were intensity- normalized to improve the accuracy and test-retest reliability of independent components analysis (ICA) output (Allen et al., 2010). Intensity normalization divides the time series of each voxel by its average intensity, converting data to percent signal change units. For the structural data pipeline, tissue classification, bias cor- rection, image registration, and spatial normalization were auto- matically performed using voxel-based morphometry (VBM) in Using the described statistical approach, we identify numerous effects of age and gender on different aspects of functional con- nectivity throughout cortical and subcortical structures. Our results corroborate previous observations and provide novel findings that motivate future in-depth investigations. 2 MaterIals and Methods 2.1 PartIcIPants This analysis combines existing data from 603 subjects scanned on the same scanner and spread across 34 studies and 18 princi- pal investigators at the Mind Research Network (MRN). Informed consent was obtained from all subjects according to institutional guidelines at the University of New Mexico (UNM) and all data were anonymized prior to group analysis. None of the participants were taking psychoactive medications at the time of the scan or had a history of neurological or psychiatric disorders. Subjects were excluded from analysis if their functional scans showed extreme motion (maximum translation 6 mm, roughly two voxels) or showed poor spatial normalization to the EPI template (see below). Subjects were also excluded if they maintained high levels of sub- stance use (smoking an average of 11 or more cigarettes per day drinking 2.5 or more drinks per day). Table 1 provides characteristics of the participants under inves- tigation. The sample is nearly balanced on gender (305 females), and the age distributions for genders are very similar. Because the sample is overwhelmingly right-handed (46 ambidextrous or left- handed individuals) and preliminary tests showed no handedness effects, we do not consider handedness from this point forward. Similarly, because participants in the white racial category are over- represented, and some studies did not collect racial information, we do not consider race from this point forward. Age is right skewed with only seven people older than 50 and the majority of individu- als in adolescence or young adulthood (80% of subjects between 1http://www.fil.ion.ucl.ac.uk/spm/software/spm5 Table 1 | Demographic information. Distributions for primary variables gender and age, as well as secondary variables handedness and race. N % N % Gender 603 100 Handedness* Male 298 49.4 Right 508 91.7 Female 305 50.6 Left and ambi 46 8.3 Mean SD Min. 25% 50% 75% Max. Age (years) 23.4 9.2 12 17 21 27 71 Male 23.8 9.1 12 17 21 26 71 Female 23.1 9.3 12 16 21 27 55 Race* LTN AI ASN NH AA WH MOR N 14 26 12 1 27 276 20 % 4 7 3 0.3 7 73 5 LTN, Latino AI, American Indian/Alaska native ASN, Asian NH, native Hawaiian or other Pacific islander AA, Black or African American WH, White MOR, more than one race. *Race and handedness frequencies and percentages do not account for missing values. Frontiers in Systems Neuroscience www.frontiersin.org February 2011 | Volume 5 | Article 2 | 3 Allen et al. A baseline for network comparisons