- Email: [email protected]
Functional connectivity in the rat brain: a complex network approach
Available online at www.sciencedirect.com
Magnetic Resonance Imaging 28 (2010) 1200 – 1209
Review articles
Functional connectivity in the rat brain: a complex network approach Angelo Bifone a,⁎, Alessandro Gozzi a , Adam J. Schwarz b b
a Italian Institute of Technology, Center for Nanotechnology Innovation, IIT@NEST, Piazza San Silvestro 12, Pisa, Italy Translational Imaging, Exploratory and Program Medicine, Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, IN 46285, USA Received 6 November 2009; revised 30 June 2010; accepted 13 July 2010
Abstract Functional connectivity analyses of fMRI data can provide a wealth of information on the brain functional organization and have been widely applied to the study of the human brain. More recently, these methods have been extended to preclinical species, thus providing a powerful translational tool. Here, we review methods and findings of functional connectivity studies in the rat. More specifically, we focus on correlation analysis of pharmacological MRI (phMRI) responses, an approach that has enabled mapping the patterns of connectivity underlying major neurotransmitter systems in vivo. We also review the use of novel statistical approaches based on a network representation of the functional connectivity and their application to the study of the rat brain functional architecture. © 2010 Elsevier Inc. All rights reserved. Keywords: Functional connectivity; Pharmacological MRI (phMRI); fMRI; Complex networks; Community structure
1. Introduction Functional magnetic resonance imaging (fMRI) has been instrumental in the study of brain functional segregation, i.e., the functional specialization of discrete brain regions engaged by specific stimuli or tasks [1]. To this end, fMRI time series are typically analyzed with massively univariate methods, testing the hypothesis of a greater (or lower) activity during execution of a task than at baseline on a pixelby-pixel basis (spatial smoothing notwithstanding) [2]. Functional specialization is then inferred from the spatial segregation of activated pixels in anatomically or functionally defined brain regions, further emphasized by the application of statistical thresholds for the significance of fMRI activation. While the evidence of functional specialization of human brain cortical regions in response to many perceptual, cognitive or emotional stimuli appears compelling, the integrated activity of multiple brain areas is involved even in the simplest sensorimotor tasks [3], in keeping with the dichotomic principles of functional segregation and integration underlying the brain's functional organization. Multi-
⁎ Corresponding author. E-mail addresses: [email protected], [email protected] (A. Bifone) 0730-725X/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2010.07.001
variate analyses of fMRI time series have been developed to assess the distributed nature of brain neurophysiological response and to address interactivity among different structures [4,5]. All of these approaches rely on the evaluation of some kind of correlation or covariance between spatially remote neurofunctional events. In this context, statistical dependencies among signals originating in different brain regions are interpreted in terms of functional connectivity, as opposed to structural connectivity, which denotes the presence of physical neuronal connections between remote brain structures [6]. While the concept of functional connectivity is thought to capture the flow of information between brain regions, and the parallel nature of brain functional processes, it should be noted that it does not necessarily imply the presence of actual physical links between the regions involved nor a causal relationship between spatially remote activations. In order to determine the influence that a neural system exerts over another, a causal model can be fitted to the data to define the effective connectivity between different regions [7]. In the case of fMRI, effective connectivity is inferred from the application of an anatomically constrained model to the analysis of interregional covariance of activity [8]. Functional connectivity has been investigated in both humans and laboratory animals with a number of techniques and using different definitions of the notion of co-varying
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
activity. At the cellular level, functional connectivity has been measured by assessing the temporal coherence in the spiking activity of individual neurons [9]. On a larger spatial scale, electroencephalography (EEG) has been applied to evaluate cross-correlation between signals from pairs of scalp electrodes and their coherence in specific frequency bands [10,11] under evoked response or resting conditions. Interregional correlations in the metabolic responses across subjects have been used to investigate functional connectivity from PET studies with FDG in humans [12,13] or from autoradiographic data in laboratory animals [14]. The inception of fMRI, a neurofunctional imaging method presenting good spatial and temporal resolution, has greatly expanded the repertoire of methods to study functional interactivity in the human brain, at rest or while performing a variety of tasks. Increased flexibility has also enabled the exploration of different measures of interregional covariance, including temporal correlation, and correlations in the fMRI response amplitude across conditions and subjects (for a review of these methods, see Horwitz [15] and references therein). It should be noted that while all these approaches represent legitimate definitions of functional connectivity, they interrogate different aspects of the regional interdependence and are not necessarily equivalent. The recent discovery that spontaneous, low-frequency fluctuations in the fMRI signals (based on the BOLD effect) from the human brain in the absence of an explicit cognitive task or stimulus exhibit coherent patterns within defined networks has opened an interesting avenue of investigation, often referred to as ‘resting-state fMRI’ [16,17]. By way of example, a network of functional connectivity corresponding to brain regions whose activity is higher at rest than during a range of cognitive tasks has been identified and interpreted as evidence in support of the existence of a ‘default mode’ of baseline brain function [18]. Interestingly, alterations in resting state functional connectivity have been observed under a number of pathological conditions [19], including Alzheimer [20], multiple sclerosis [21] and schizophrenia [22]. While there is a large body of work assessing functional connectivity in the human brain following the intense research in this area in the past decade, the extension of these methods to preclinical species, and particularly to rodents, has been more recent (notwithstanding the original demonstration of coupled responses by Soncrant et al. [14] in the late 1980s). The study of functional connectivity in laboratory animals, where imaging can be combined with more invasive methods, may provide a better understanding of the physiological basis of the interregional correlations in fMRI signals observed in humans and of the origin of the alterations in functional connectivity observed in pathological states. However, limitations also exist, mostly related to the use of anesthetic agents that reduce animal motion and stress, and to the substantially more limited repertoire of functional stimuli that can be applied in laboratory animals.
1201
Several studies in laboratory animals have followed the path of human resting state fMRI exploiting the BOLD effect. Synchronous low-frequency fluctuations were detected bilaterally in the primary somatosensory cortex of the rat brain at rest under alpha-chloralose anesthesia [23]. Interestingly, a significant correlation between fMRI responses and delta oscillations measured with epidural EEG was observed, thus suggesting a link between spontaneous fMRI fluctuations and the underlying neuronal electrical activity. Similarly correlated fluctuations were also detected in the rat under isoflurane [24], halothane [25] or medetomidine anesthesia [26–28]. All these studies used a seed-correlation region approach and demonstrated encouraging correlations between bilaterally symmetrical cortical regions. However, coherent networks of connectivity akin to those observed in humans (e.g., the default mode network) have not been observed in the rat using this method to date. A conceptually different approach to the study of functional connectivity in rodents has been explored using pharmacological MRI (phMRI), i.e., fMRI applied to map spatio-temporal patterns of brain activity induced by acute pharmacological challenge [29]. The main merits of this approach lie in its ability to elicit robust and reliable activations even under anesthesia conditions and to enable selective stimulation of different neurotransmitter systems, thus providing a playground to study the neurochemical bases of fMRI responses and their connectivity. Several recent studies using this approach have shown exquisite delineation of focal patterns of correlated responses corresponding to key neurotransmitter pathways [30–34]. Moreover, network analyses of functional connectivity patterns obtained with phMRI have provided the first evidence of organized networks of functional connectivity in the rat brain, thus demonstrating the potential of this method to explore the brain's functional architecture in animal models. Here, we summarize the key principles and methodological aspects of correlation analysis of phMRI data and review recent studies exploring pharmacologically induced or modulated functional connectivity in the rat brain. Moreover, we discuss novel methods based on a network representation of functional connectivity data and their application to resolve the functional organization of the rat brain.
2. Correlation analysis in phMRI In a typical phMRI experiment, a drug challenge is administered systemically (e.g., intravenously or intraperitoneally) during the acquisition of the fMRI time series [35–37]. The time resolution of fMRI is determined by the time scale of the hemodynamic response following neuronal excitation and is of the order of seconds. This temporal resolution is sufficient to capture the time course of the slow spontaneous fluctuations measured by resting state fMRI (typically below 0.15 Hz). The phMRI signal changes
1202
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
following a drug challenge are much slower (typically lasting tens of minutes or more) than the hemodynamic response to a single neural ‘event’ [38]. Moreover, the signal time course reflects drug kinetics and pharmacodynamic effects and, despite rigorous control over animal physiology and experimental parameters, a degree of variability in the relative response amplitude in different brain regions across subjects is typically observed [37,39]. This variability can be leveraged by an alternative approach based on the calculation of interregional correlations in the response amplitude across subjects [30], in the fashion of the procedures applied in metabolic PET or autoradiography studies [13,14]. Schematically, the correlation analysis procedure used in phMRI studies can be summarized in terms of the following steps. (A) Individual subject time series are co-registered to a common stereotaxic space [40]; (B) Time courses from individual image voxels are extracted for each subject, and suitable signal regressors are fitted to the data to determine response amplitudes. Data-driven signal decompositions strategies (e.g., wavelet-cluster analysis [41,42]) have proven particularly effective in providing signal regressors that fit experimental data with excellent sensitivity, particularly in the case of highly dynamic phMRI signal changes. In the present context, we understand response amplitude to represent the magnitude of the post-injection signal change elicited by the acute drug challenge. These values provide voxel-specific vectors of response amplitude across subjects. (C) Inter-subject correlations are then calculated for each voxel in reference to a selected ‘seed’ region, using the vector of response amplitudes from Step B. (D) Finally, statistical correlation maps, i.e., maps of voxels whose response amplitude correlates signifi-
cantly (in a statistical sense) with those in the reference region are generated. This approach leverages variations in the spatial profile of the response observed across subjects following drug challenge. Fig. 1 provides an outline of this process and an exaggerated visual impression of the intersubject variability in the phMRI response profile. 3. Application to mapping functional connectivity in neurotransmitter systems The first demonstration of this approach applied correlation analysis of regional cerebral blood volume changes in the rat induced by acute challenge with D-amphetamine and fluoxetine, two widely used (and abused) drugs that target key monoaminergic systems [30]. One of the main effects of D-amphetamine in the brain is the stimulation of release of dopamine (DA), as well as other monoamines, resulting in increased levels of DA at the terminals of dopaminergic neurons [43]. The psychoactive and reinforcing properties of D-amphetamine are thought to be mediated by stimulation of the mesolimbic DA pathway originating from the ventral tegmental area (VTA) in the midbrain and projecting to the ventral part of the striatum, the nucleus accumbens [43]. Hence, the VTA represents the natural choice of seed correlation region for the analysis of the responses to D -amphetamine. Correlation maps referenced to the VTA clearly delineated the parallel major axes of this pathway forward through the medial forebrain bundle to the ventral striatum (Fig. 2). Other downstream patterns of functional connectivity could be identified by moving the reference region along the functional connectivity pathway [31]. With the use of the shell of the nucleus accumbens as a seed region, a functional connectivity pattern largely coincident with that of the VTA, but extending to the medial pre-frontal cortex, was identified, consistent with the
Fig. 1. Schematic overview of the principle underlying across-subject functional correlation. Different subjects show different spatial profiles of responses to the pharmacological challenge. This variability is leveraged to extract response vectors for each image voxel (left-hand panel). Intersubject correlations are then calculated for each pair of image voxels or for a voxel in reference to a seed region (right-hand panel; a–e indicate representative responses in brain regions A and B across different individual subjects).
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
Fig. 2. Map of the responses to D-amphetamine covarying (across-subjects) with that in the ventral tegmental area (VTA). The pattern of connectivity delineates the mesolimbic DA pathway extending forward to the striatum.
known reciprocal projections of this region to the accumbens. Moving further down the DA pathway, broader cortical involvement, including somatosensory and motor cortices, was observed. In this way, it was possible to discriminate between connectivity in the primary dopaminergic projections and connectivity with downstream structures. Fluoxetine (Prozac) is a canonical selective serotonin reuptake inhibitor [44]; stimulation of the serotonin system is thought to mediate the antidepressant properties of this drug. The main source of serotonergic projections to the forebrain originates in the raphe nuclei. Also in this case, correlation analysis using the raphe as a reference region showed a focal delineation of major ascending serotonergic projections to the forebrain and cortex [30], and a network of subcortical structures including the hippocampus and amygdala, a pattern of connectivity consistent with the drug's mechanism of action. For both drugs, the neuroanatomically specific distributions obtained by the seed region connectivity approach were far more meaningful than the results of massively univariate maps of mean amplitude differences between drug and vehicle groups. The latter identified widespread strong responses in cortical regions, but the compelling subcortical circuits were resolved only when the covariation in response between different brain regions was examined. This study represented the first in vivo demonstration of functional connectivity in discrete neurotransmitter systems [30]. The correlation analysis was guided by a judicious
1203
choice of seed region that was informed by previous knowledge of the systems under investigation. While the confirmatory evidence produced in this study represented convincing validation of the method, approaches that are independent of the specific choice of reference region would be desirable for more exploratory studies where the mechanism of action and behavioral/clinical effects of the drug are less well characterized. A more general approach to study patterns of connectivity from phMRI data was proposed in Ref. [31], where an hypothesis-free cluster analysis was applied to resolve networks of correlated brain activity stimulated by Damphetamine without prior definition of a seed correlation region. A volume-of-interest (VOI) parcellation of the brain volume into Nv=48 bilateral brain structures was delineated automatically using a 3D digital reconstruction of the Paxinos and Watson rat brain atlas co-registered with the MRI rat brain template [40]. Pairwise correlations within the set of selected structures were calculated by computing the Pearson linear correlation coefficient r between the response vectors in each pair of structures, yielding a 48×48 correlation matrix summarizing the similarities in response profile among the selected VOIs (Fig. 3). In order to identify groups of brain structures within the full correlation matrix that were closely coupled in their response to Damphetamine, these authors applied a k-means clustering algorithm [45], an optimization-based method that minimizes a distance measure between the Nv response vectors. In the case of a D-amphetamine phMRI data set, the algorithm indicated that the 48 response vectors should be optimally clustered into three groups in order to maximize intercluster over intracluster dispersion in the normalized cross-subject response profiles. Interestingly, one of the clusters consisted of brain structures involved in the mesolimbic and nigrostriatal DA pathways, including the VTA and substantia nigra (SN), together with ventral structures along the major axes of these pathways (Fig. 3C), a finding consistent with the mesolimbic signature identified in the seed region approach. It is important to emphasize that the clustering results did not identify merely a trivial solution in which strongly responding regions were grouped together while those responding more weakly were clustered apart. Instead, key brain areas involved in primary DA pathways emerged naturally from the data in the form of a group of tightly interconnected structures, thus overcoming the limitations imposed by the need to introduce a priori knowledge in the analysis. While Damphetamine also stimulates other monoamines, e.g., norepinephrine, the DA system appears to show the tightest connectivity under the effects of this drug. In the phMRI approach, a drug challenge serves the purpose of eliciting a brain response whose intersubject variability provides a basis to calculate interregional correlations. From a pharmacological point of view, it is also important to ascertain whether drug treatment can modulate brain functional connectivity, since this might shed
1204
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
Fig. 3. Interregional correlations in response to D-amphetamine. (A) Cross-correlation matrix for the 48 brain structures examined. (B) Connections between the 48 structures in sagittal view. Each region is indicated by a black dot and the connections by red lines. (C) A cluster of structures identified by the k-clustering algorithm, corresponding to the primary projection of the mesolimbic DA system.
light on the drug mechanism of action at a systems level. One interesting approach to addressing this problem has been proposed in Ref. [32], where the effects of a compound of interest were assessed by examining how it modulated the correlated responses to a different probe compound. Specifically, this study investigated the effects of a DA D3 receptor antagonist, a putative treatment of addiction, on the pattern of functional connectivity stimulated by D-amphetamine. The DA D3 receptor has a particularly focal distribution in the brain, with high levels of expression in dopaminergically innervated mesolimbic brain structures, including the nucleus accumbens, involved in reward-related processes [46,47]. D-Amphetamine was previously shown to elicit phMRI responses that were highly correlated in these key dopaminergic circuits and provided the ideal probe to study the effects of the blockade of D3 receptors on the brain functional connectivity. SB277011A, a potent and selective DA D3 receptor antagonist [47], or its vehicle, was administered in two separate groups of rats 30 min prior to an intravenous D-amphetamine challenge. The temporal interval between the drug administration and the amphetamine challenge was sufficient for SB277011A to reach significant unbound concentration in the brain and for its acute effects on cerebral hemodynamics to subside, so that identical CBV baselines could be obtained in the drug and
vehicle groups at the moment of the D-amphetamine challenge. In order to identify brain regions associated with changes in functional connectivity between the two Damphetamine groups, the phMRI responses to amphetamine were acquired and Pearson linear correlation coefficients summarizing the correlation between the responses in each pair of VOIs were calculated for each group. These were then converted to z-statistics using Fisher's r-to-z transformation [40] resulting in a matrix of the z-statistics for each group corresponding to the pairwise correlations between responses in each pair of VOIs. Direct comparison of the zstatistics corresponding to each cell in the matrix then indicated the pairs of brain regions in which the correlation differed between the two groups. It should, however, be emphasized that the simple approach used by the authors is only one of the many statistical options that could be applied to estimate intergroup differences in connectivity patterns, and several alternative approaches have been proposed [48,49]. Nevertheless, main changes in inter-region correlations associated with the VTA, a region rich in DA D3 receptors, were observed. Specifically, these involved connectivity of the VTA with the ventral subiculum and ventral CA3 fields in the hippocampus, dorsolateral and midline dorsal thalamus, piriform cortex and superior colliculi. In the
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
midline dorsal thalamus and hippocampus, the relationship was altered from a significantly positive correlation with the VTA to a significantly negative one, with a dramatic inversion of the sign of the connectivity. The thalamus and hippocampus are important components of a limbic brain circuit and thought to be involved in drug addiction, providing modulatory inputs to the prefrontal cortex, an important brain region involved in goaldirected behavior. The findings of this study, of a reversed correlation between responses in the VTA and the dorsal thalamus, and a reduced correlation with the ventral striatum, suggest that a modified functional connectivity within this circuit may play a key role in the efficacy of selective DA D3 receptor antagonists in attenuating drug-seeking behavior [50]. Methodologically, this study extended the applicability of intersubject functional connectivity analyses of phMRI data to antagonist–agonist experiments, where modulations in the correlation structure underlying the response to a probe signal may be detected. Although the examples provided are based on the exploitation of CBV-based phMRI, it should be emphasized that the conceptual framework underlying the approach is modality independent and can be straightforwardly applied to different neuroimaging modalities (i.e., BOLD-phMRI) and brain mapping techniques. A recent study [51] has significantly expanded the area of application of these methods, by providing a remarkable example of the use of phMRI, intersubject functional connectivity and advanced mouse genetics to provide a direct link between cells, circuits and behavior. Specifically, the authors combined phMRI and focal pharmacogenetic silencing to spatially resolve behavior-specific circuits controlled by a discrete neuronal population expressed in the central nucleus of the amygdala (CeA), a brain region involved in the control of emotional and behavioral response to fear [52]. Reversible suppression of neural activity in a subset (‘type I’) of CeA neurons was achieved by inducing cell-specific re-expression of the serotonin 1A receptor (Htr1aR) in mice devoid of the endogenous receptor (the resulting mice called Htr1aCeA), a strategy recently described by Tsetsenis et al. [53]. The Htr1aR is coupled with inward rectifying potassium channels whose endogenous or exogenous activation produces rapid and reversible membrane hyperpolarization of the cells expressing the receptor. Hence, the use of systemically administered Htr1aR agonists (like the selective compound 8-OHDPAT) can be exploited to obtain cell-specific inhibition of spontaneous neuronal firing in behaving animals [53]. By combining fMRI with this pharmacogenetic inhibition strategy, the authors were able to identify a novel pathway in the mammalian brain apt to regulate passive and active components of fear responses. Specifically, selective inhibition of type I CeA cells led to widespread increased cortical arousal as visualized by phMRI upon acute administration of the selective Htr1AR agonist 8-OH-DPAT. Intersubject functional connectivity analyses of phMRI time courses were critical in resolving the neuronal circuitry underlying
1205
the increased cortical activity. Significant correlation was found between the CeA and several ventral forebrain cholinergic nuclei such as the substantia innominata and the diagonal band, and the same cholinergic nuclei exhibited tight correlations with the cortical areas activated upon inhibition of type I CeA neurons. The involvement of the cholinergic system was confirmed in additional phMRI studies showing that the cortical arousal was blocked by central (but not by peripheral) cholinergic antagonists. Remarkably, an analysis of the behavioral correlates of cortical activation in Htr1aCeA mice highlighted a pivotal role of type I CeA cells as suppressors of cholinergic activity and exploratory behavior and as promoters of freezing (passive) fear responses, thus leading to the identification of a novel neural pathway that biases fear responses toward either passive or active coping strategies. Methodologically, this work expands the applicability of functional connectivity analyses on genetically modified mouse models and provides at the same time a compelling demonstration of the combined use neuroimaging methods and advanced pharmacogenetic systems as a new powerful paradigm to identify and resolve behaviorally relevant neural circuits in the living brain.
4. Complex network analysis and community structure of functional connectivity in the rat brain Functional connectivity analyses of neuroimaging data aim to elucidate the relationships between signals originating in spatially distinct brain regions. This emphasis on interaction between different brain structures is a good conceptual match for representing the data as a graph, or network, of nodes and links [54]. In this representation, image voxels or parcellated brain regions represent the nodes, and a measure of similarity or correlation in their responses defines the edges linking the nodes [55–58]. In recent years, networks from a wide variety of fields, describing connection patterns as diverse as social collaborations, metabolic pathways or air traffic, have been characterized and found to exhibit rich behaviors beyond those of simple random networks; accordingly, they are referred to as ‘complex networks’ [59]. Network analysis of functional connectivity data to date has concentrated on the characterization of global properties of the network, which can reveal much about the properties of the system. For example, networks derived from human brain imaging data have been shown to possess a scale-free degree distribution and exhibit ‘small world’ behavior — a finding that has implications for information transfer in the brain [54]. Complex networks of functional connectivity derived from phMRI data in the rat have also been recently investigated [33,34]. In these articles, the nodes were defined as individual voxels (0.128 mm3) in the 3D image volume, and the strength of the edge between each pair of nodes (i,j) was based on the Pearson correlation coefficient
1206
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
rij between the intersubject response amplitude vectors in the two voxels. This value was converted into an equivalent z-statistic using Fisher's r-to-z transformation. The magnitudes of these normalized correlation values were used to describe the strength of the correlation between each pair of nodes and used to construct an edge weight matrix Wij. Finally, the edge weight matrix was thresholded and binarized to define a binary adjacency matrix. Analysis of the structural properties of the resulting networks outlined interesting topological features similar to those that have been reported for human brain networks [54], including a small-world structure and a long-tailed distribution of node degrees (i.e., number of connections for certain node), indicative of the presence of hubs, namely, a subset of highly connected nodes in cortical regions. Beyond global characterization, the coexistence of functional segregation and integration in brain activity [60] suggests that some degree of modularity, i.e., subnetworks or clusters of more tightly linked nodes, might exist within functional connectivity networks (cf. Ref. [31]). The issue of the identification of such groups of nodes characterized by denser connections between their constituents than to other nodes outside the group has received considerable recent attention in the field of complex networks. This concept originated in the study of social relationships and has been consequently dubbed ‘community structure.’ The first application of this concept to the analysis of brain functional connectivity networks has been recently demonstrated using phMRI data in rodents challenged with fluoxetine [33]. In this article, a networktheoretic community structure algorithm, based on maximization of a mathematical formalism of ‘modularity,’ was applied to resolve functionally and anatomically segregated communities within a functional connectivity network derived from phMRI responses. Specifically, an algorithm based upon maximizing the value of a mathematical definition of modularity Q [61,62] was applied to identify the number and composition of subnetworks within a widely distributed network of functional connectivity obtained under pharmacological challenge with fluoxetine. For a certain partition of the network, Q measures the difference between the fraction of the edges connecting nodes within communities and the same fraction in the case of a randomly connected network with the same partition. The closer the value of Q is to its theoretical maximum of 1, the stronger the community structure, i.e., the more modular the network. Algorithms seeking a network partition that maximizes modularity yield an estimate of the number of communities into which the network should be optimally split, the composition of each community and an associated value of max(Q). In the case of functional connectivity networks, an optimal partition can reveal the system-level functional organization of the brain under the particular experimental conditions studied, as well as provide a measure of the emergent modularity. The application of this method to the functional connectivity
network in the rat challenged with fluoxetine revealed three communities of nodes (Fig. 4). The pixels in the two largest communities were symmetrically distributed between the left and right hemispheres, and their distributions corresponded closely to known anatomical and functional subdivisions of the rat brain. It is important to emphasize that division was obtained from a network representation based on image pixels with no imposition of symmetry nor any prior anatomical constraints. Interestingly, one of the communities comprised nodes corresponding primarily to subcortical structures — in the striatum, thalamus and amygdala — but also to regions of the hippocampus and entorhinal, medial pre-frontal and cingulate cortices (Fig. 4A). Again, this finding was consistent with the results of the seed region analysis [30]. The finding of pixels in the cingulate and prefrontal areas grouped with those in striatal and thalamic structures is consistent with the fact that these cortical regions are the main cortical target of input from the basal ganglia via extensive reciprocal connections with the thalamus [63]. These cortical regions are anatomically similar to other regions of the cortex yet functionally distinct. Similarly, pixels in the entorhinal cortex and in the
Fig. 4. Anatomical representation of the three communities identified by modularity maximization in a functional network derived from phMRI responses to fluoxetine.
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
hippocampus were assigned to the same community, reflecting the dense connections between these structures. Most other cortical nodes, including those located in the motor, somatosensory and visual cortices, were assigned to a second community (Fig. 4B), while a third community (Fig. 4C) mainly comprised pixels near the brain edge, the ventricles and in white matter and the cerebellum. The community structure approach is somewhat akin to other multivariate methods, like principal component analysis and independent component analysis, which have been extensively applied to seek structure within imaging data in a model-independent way [2,64,65]. Unlike these, however, this community structure approach explicitly takes into account the topology of the functional connections, with communities defined on the basis of link density and distribution. This concept may be more readily interpretable in biological terms than measures such as the orthogonality [2,66] or statistical independence [64] of spatial modes that are optimized by other algorithms. A key point in this approach is that the identification of communities within a functional imaging network contains information on both segregation and integration. A partition of the overall network into smaller subunits suggests a degree of functional segregation in the response, whereas the set of brain regions — not necessarily contiguous — identified within each subnetwork reflects their integrated action in response to the experimental stimulus. Importantly, the value of the modularity Q provides a measure of the degree of functional segregation in the network and may provide the basis for an operational definition of this. This first demonstration of the potential of community structure analysis of functional connectivity in preclinical species has been rapidly translated to human functional connectivity, and several studies extending this approach to resting state fMRI have subsequently appeared in the literature [54,67]. However, a few questions remain open and are the object of active investigation. For example, binarization of the adjacency matrix was used to make the problem computationally tractable. However, application of a threshold to retain only the strongest correlations ignores information that may be contained in the distribution of the weaker links, and the extension to fully weighted functional connectivity networks is a natural development for this approach, dependent on the availability of computationally efficient partitioning algorithms. Also, introduction of a direction to the edges linking pairs of nodes, perhaps using Granger causality or related concepts, would provide a means of creating directed graphs of connectivity that may be informative of the roles played by different nodes. Lastly, and most importantly, the origin of structured patterns of correlated responses is still largely unknown. Specifically, it is unclear whether the patterns of functional connectivity reflect interregional correlations induced by the drug challenge itself, or the intrinsic organization of the brain, perhaps determined by the structure of the underlying neuronal substrate.
1207
A follow-up article applied community structure analysis to address this question [34]. To this end, the community structures of functional connectivity networks under three different pharmacological challenges, D-amphetamine, fluoxetine and nicotine, were investigated and the emergent community structures under different pharmacological conditions were compared to discriminate between connectivity patterns that are stimulus specific and those independent of the particular neurotransmitter system(s) engaged by the drug, which might thus correspond to general features of the rat brain functional architecture. Interestingly, common features across all three networks revealed two groups of tightly coupled brain structures that responded as functional units independent of the drug, including a network involving the prefrontal cortex and subcortical regions extending from the striatum to the amygdala. This suggests that this network of functional connectivity may reflect a general feature of the brain organization and is consistent with evidence showing strong intrinsic connectivity between neurons in these brain structures [68]. 5. Conclusion Functional connectivity analysis of functional neuroimaging data is a rapidly developing area of research that has brought about much progress in our understanding of the brain functional architecture. While most of the studies in the literature pertain to human subjects, functional connectivity methods and concepts have been recently extended, mutatis mutandis, to preclinical species. Here, we have reviewed the recent literature in this field, with particular emphasis on approaches based on pharmacological MRI, an imaging approach based on stimulation or modulation of brain activity with pharmacological challenges. While conceptually different from the human resting-state fMRI, which relies on spontaneous fluctuations of the fMRI signal in the brain in the absence of external stimuli, the phMRI approach has proven to be a powerful tool to explore functional connectivity in rodents and to map a variety of different neurotransmitter pathways. Pharmacological MRI has also provided a useful playground to explore novel statistical methods of analysis of functional connectivity represented in terms of complex networks. Particularly, network partitioning methods have made it possible for the first time to resolve biologically and anatomically meaningful patterns of correlated responses in the rat brain, thus providing a novel and powerful method to investigate the brain functional architecture in preclinical species. References [1] Posner MI, Petersen SE, Fox PT, Raichle ME. Localization of cognitive operations in the human brain. Science 1988;240:1627–31. [2] Friston KJ. Analyzing brain images: principles and overview. In: Frackowiak RSJ, Friston KJ, Frith C, Dolan R, Mazziotta JC, editors. Human Brain Function. Academic Press; 1997. p. 25–41.
1208
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
[3] Luria A. The working brain. New York: Basic Books; 1973. [4] Rogers BP, Morgan VL, Newton AT, Gore JC. Assessing functional connectivity in the human brain by fMRI. Magn Reson Imag 2007;25: 1347–57. [5] Friston KJ, Jezzard P, Turner R. Analysis of functional MRI timeseries. Hum Brain Mapp 1994;1:153–71. [6] Ramnani N, Behrens TE, Penny W, Matthews PM. New approaches for exploring anatomical and functional connectivity in the human brain. Biol Psychiatry 2004;56:613–9. [7] Friston KJ. Statistical parametric mapping and other analyses of functional imaging data. In: Toga AW, Mazziotta JC, editors. Brain mapping: the methods. San Diego: Academic Press; 1996. p. 363–86. [8] Buchel C, Friston K. Assessing interactions among neuronal systems using functional neuroimaging. Neural Netw 2000;13:871–82. [9] Gerstein GL, Perkel DH. Simultaneously recorded trains of action potentials: analysis and functional interpretation. Science 1969;164: 828–30. [10] Adey WR, Walter DO, Hendrix CE. Computer techniques in correlation and spectral analyses of cerebral slow waves during discriminative behavior. Exp Neurol 1961;3:501–24. [11] Pfurtscheller G, Andrew C. Event-related changes of band power and coherence: methodology and interpretation. J Clin Neurophysiol 1999;16:512–9. [12] Bartlett EJ, Brown JW, Wolf AP, Brodie JD. Correlations between glucose metabolic rates in brain regions of healthy male adults at rest and during language stimulation. Brain Lang 1987;32:1–18. [13] Horwitz B, Duara R, Rapoport SI. Intercorrelations of glucose metabolic rates between brain regions: application to healthy males in a state of reduced sensory input. J Cereb Blood Flow Metab 1984;4: 484–99. [14] Soncrant TT, Horwitz B, Holloway HW, Rapoport SI. The pattern of functional coupling of brain regions in the awake rat. Brain Res 1986;369:1–11. [15] Horwitz B. The elusive concept of brain connectivity. NeuroImage 2003;19:466–70. [16] Beckmann CF, DeLuca M, Devlin JT, Smith SM. Investigations into resting-state connectivity using independent component analysis. Philos Trans R Soc Lond B Biol Sci 2005;360:1001–13. [17] Biswal B, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn Reson Med 1995;34:537–41. [18] Greicius MD, Krasnow B, Reiss AL, Menon V. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis. Proc Natl Acad Sci U S A 2003;100:253–8. [19] Greicius M. Resting-state functional connectivity in neuropsychiatric disorders. Curr Opin Neurol 2008;21:424–30. [20] Li SJ, Li Z, Wu G, Zhang MJ, Franczak M, Antuono PG. Alzheimer disease: evaluation of a functional MR imaging index as a marker. Radiology 2002;225:253–9. [21] Lowe MJ, Phillips MD, Lurito JT, Mattson D, Dzemidzic M, Mathews VP. Multiple sclerosis: low-frequency temporal blood oxygen leveldependent fluctuations indicate reduced functional connectivity initial results. Radiology 2002;224:184–92. [22] Zhou Y, Shu N, Liu Y, Song M, Hao Y, Liu H, et al. Altered restingstate functional connectivity and anatomical connectivity of hippocampus in schizophrenia. Schizophr Res 2008;100:120–32. [23] Lu H, Zuo Y, Gu H, Waltz JA, Zhan W, Scholl CA, et al. Synchronized delta oscillations correlate with the resting-state functional MRI signal. Proc Natl Acad Sci U S A 2007;104:18265–9. [24] Kannurpatti SS, Biswal BB, Kim YR, Rosen BR. Spatio-temporal characteristics of low-frequency BOLD signal fluctuations in isoflurane-anesthetized rat brain. NeuroImage 2008;40:1738–47. [25] Majeed W, Magnuson M, Keilholz SD. Spatiotemporal dynamics of low frequency fluctuations in BOLD fMRI of the rat. J Magn Reson Imaging 2009;30:384–93. [26] Pawela CP, Biswal BB, Hudetz AG, Schulte ML, Li R, Jones SR, et al. A protocol for use of medetomidine anesthesia in rats for extended
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45] [46]
studies using task-induced BOLD contrast and resting-state functional connectivity. NeuroImage 2009;46:1137–47. Pawela CP, Biswal BB, Cho YR, Kao DS, Li R, Jones SR, et al. Resting-state functional connectivity of the rat brain. Magn Reson Med 2008;59:1021–9. Zhao F, Zhao T, Zhou L, Wu Q, Hu X. BOLD study of stimulationinduced neural activity and resting-state connectivity in medetomidinesedated rat. NeuroImage 2008;39:248–60. Leslie RA, James MF. Pharmacological magnetic resonance imaging: a new application for functional MRI. Trends Pharmacol Sci 2000;21: 314–8. Schwarz AJ, Gozzi A, Reese T, Bifone A. In vivo mapping of functional connectivity in neurotransmitter systems using pharmacological MRI. NeuroImage 2007;34:1627–36. Schwarz AJ, Gozzi A, Reese T, Bifone A. Functional connectivity in the pharmacologically activated brain: resolving networks of correlated responses to D-amphetamine. Magn Reson Med 2007;57:704–13. Schwarz AJ, Gozzi A, Reese T, Heidbreder CA, Bifone A. Pharmacological modulation of functional connectivity: the correlation structure underlying the phMRI response to D-amphetamine modified by selective dopamine D3 receptor antagonist SB277011A. Magn Reson Imaging 2007;25:811–20. Schwarz AJ, Gozzi A, Bifone A. Community structure and modularity in networks of correlated brain activity. Magn Reson Imaging 2008;26: 914–20. Schwarz AJ, Gozzi A, Bifone A. Community structure in networks of functional connectivity: resolving functional organization in the rat brain with pharmacological MRI. NeuroImage 2009;47:302–11. Gozzi A, Schwarz A, Reese T, Bertani S, Crestan V, Bifone A. Regionspecific effects of nicotine on brain activity: a pharmacological MRI study in the drug-naïve rat. Neuropsychopharmacology 2006;31:1690–703. Gozzi A, Large C, Schwarz A, Bertani S, Crestan V, Bifone A. Differential effects of antipsychotic and glutamatergic agents on the phMRI response to phencyclidine. Neuropsychopharmacology 2008;33:1690–703. Schwarz AJ, Zocchi A, Reese T, Gozzi A, Garzotti M, Varnier G, et al. Concurrent pharmacological MRI and in situ microdialysis of cocaine reveal a complex relationship between the central hemodynamic response and local dopamine concentration. NeuroImage 2004;23: 296–304. Schwarz AJ, Reese T, Gozzi A, Bifone A. Functional MRI using intravascular contrast agents: detrending of the relative cerebrovascular (rCBV) time course. Magn Reson Imaging 2003;21:1191–200. Schwarz A, Whitcher B, Reese T, Gozzi A, Bifone A. Group-level data-driven phMRI analysis. Book of Abstracts: Thirteenth Annual Meeting of the International Society of Magnetic Resonance in Medicine, 13; 2005. p. 157. Schwarz AJ, Danckaert A, Reese T, Gozzi A, Paxinos G, Watson C, et al. A stereotaxic MRI template set for the rat brain with tissue class distribution maps and co-registered anatomical atlas: application to pharmacological MRI. NeuroImage 2006;32:538–50. Schwarz AJ, Whitcher B, Gozzi A, Reese T, Bifone A. Study-level wavelet cluster analysis and data-driven signal models in pharmacological MRI. J Neurosci Methods 2006;159:346–60. Whitcher B, Schwarz AJ, Barjat H, Smart SC, Grundy RI, James MF. Wavelet-based cluster analysis: data-driven grouping of voxel time courses with application to perfusion-weighted and pharmacological MRI of the rat brain. NeuroImage 2005;24:281–95. Everitt BJ, Robbins TW. Neural systems of reinforcement for drug addiction: from actions to habits to compulsion. Nat Neurosci 2005;8: 1481–9. Stahl SM. Mechanism of action of serotonin selective reuptake inhibitors: serotonin receptors and pathways mediate therapeutic effects and side effects. J Affect Disord 1998;51:215–35. Everitt BJ, Landau S, Leese M. Cluster analysis. London: Arnold; 2001. Schwarz A, Gozzi A, Reese T, Bertani S, Crestan V, Hagan J, et al. Selective dopamine D(3) receptor antagonist SB-277011-A potentiates
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
[47]
[48]
[49] [50]
[51]
[52] [53]
[54]
[55]
[56]
phMRI response to acute amphetamine challenge in the rat brain. Synapse 2004;54:1–10. Heidbreder CA, Gardner EL, Xi ZX, Thanos PK, Mugnaini M, Hagan JJ, et al. The role of central dopamine D3 receptors in drug addiction: a review of pharmacological evidence. Brain Res Brain Res Rev 2005;49:77–105. Holmes AP, Blair RC, Watson JDG, Ford I. Nonparametric analysis of statistic images from functional mapping experiments. J Cereb Blood Flow Metab 1996;16:7–22. Rogers BP, Gore JC. Empirical comparison of sources of variation for FMRI connectivity analysis. PLoS ONE 2008:3. Cervo L, Cocco A, Petrella C, Heidbreder CA, Bendotti C, Mennini T. SB-277011-A, a selective dopamine D3 receptor antagonist, reduces cocaine-seeking behavior in response to drug-associated stimuli in rats. Proc Society for Neuroscience 2003. Gozzi A, Apar J, Giovanelli A, Bertollini C, Crestan V, Schwarz AJ, et al. A neural switch for active and passive fear. Neuron 2010 doi:10.1016/j.neuron.2010.07.008. LeDoux J. The emotional brain, fear, and the amygdala. Cell Mol Neurobiol 2003;23:727–38. Tsetsenis T, Ma XH, Lo Iacono L, Beck SG, Gross C. Suppression of conditioning to ambiguous cues by pharmacogenetic inhibition of the dentate gyrus. Nat Neurosci 2007;10:896–902. Bullmore E, Sporns O. Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci 2009;10:186–98. Salvador R, Suckling J, Schwarzbauer C, Bullmore E. Undirected graphs of frequency-dependent functional connectivity in whole brain networks. Philos Trans R Soc Lond B Biol Sci 2005;360:937–46. Eguiluz VM, Chialvo DR, Cecchi GA, Baliki M, Apkarian AV. Scalefree brain functional networks. Phys Rev Lett 2005;018102:94.
1209
[57] Achard S, Salvador R, Whitcher B, Suckling J, Bullmore E. A resilient, low-frequency, small-world human brain functional network with highly connected association cortical hubs. J Neurosci 2006;26: 63–72. [58] Achard S, Bullmore E. Efficiency and cost of economical brain functional networks. PLoS Comput Biol 2007;e17:3. [59] Strogatz SH. Exploring complex networks. Nature 2001;410:268–76. [60] Tononi G, Sporns O, Edelman GM. A measure for brain complexity: relating functional segregation and integration in the nervous system. Proc Natl Acad Sci U S A 1994;91:5033–7. [61] Newman ME, Girvan M. Finding and evaluating community structure in networks. Phys Rev E Stat Nonlin Soft Matter Phys 2004;69: 026113. [62] Newman ME. Modularity and community structure in networks. Proc Natl Acad Sci U S A 2006;103:8577–82. [63] Uylings HB, Groenewegen HJ, Kolb B. Do rats have a prefrontal cortex? Behav Brain Res 2003;146:3–17. [64] McKeown MJ, Makeig S, Brown GG, Jung TP, Kindermann SS, Bell AJ, et al. Analysis of fMRI data by blind separation into independent spatial components. Hum Brain Mapp 1998;6:160–88. [65] Beckmann CF, Smith SM. Probabilistic independent component analysis for functional magnetic resonance imaging. IEEE Trans Med Imaging 2004;23:137–52. [66] Friston KJ, Frith CD, Liddle PF, Frackowiak RS. Functional connectivity: the principal-component analysis of large (PET) data sets. JCBFM 1993;13:5–14. [67] Meunier D, Achard S, Morcom A, Bullmore E. Age-related changes in modular organization of human brain functional networks. NeuroImage 2009;44:715–23. [68] Palomero-Gallagher N, Zilles K. The rat nervous system. In: Paxinos G, editor. London: Elsevier; 2010. p. 729–57.
Magnetic Resonance Imaging 28 (2010) 1200 – 1209
Review articles
Functional connectivity in the rat brain: a complex network approach Angelo Bifone a,⁎, Alessandro Gozzi a , Adam J. Schwarz b b
a Italian Institute of Technology, Center for Nanotechnology Innovation, IIT@NEST, Piazza San Silvestro 12, Pisa, Italy Translational Imaging, Exploratory and Program Medicine, Lilly Research Laboratories, Eli Lilly and Company, Indianapolis, IN 46285, USA Received 6 November 2009; revised 30 June 2010; accepted 13 July 2010
Abstract Functional connectivity analyses of fMRI data can provide a wealth of information on the brain functional organization and have been widely applied to the study of the human brain. More recently, these methods have been extended to preclinical species, thus providing a powerful translational tool. Here, we review methods and findings of functional connectivity studies in the rat. More specifically, we focus on correlation analysis of pharmacological MRI (phMRI) responses, an approach that has enabled mapping the patterns of connectivity underlying major neurotransmitter systems in vivo. We also review the use of novel statistical approaches based on a network representation of the functional connectivity and their application to the study of the rat brain functional architecture. © 2010 Elsevier Inc. All rights reserved. Keywords: Functional connectivity; Pharmacological MRI (phMRI); fMRI; Complex networks; Community structure
1. Introduction Functional magnetic resonance imaging (fMRI) has been instrumental in the study of brain functional segregation, i.e., the functional specialization of discrete brain regions engaged by specific stimuli or tasks [1]. To this end, fMRI time series are typically analyzed with massively univariate methods, testing the hypothesis of a greater (or lower) activity during execution of a task than at baseline on a pixelby-pixel basis (spatial smoothing notwithstanding) [2]. Functional specialization is then inferred from the spatial segregation of activated pixels in anatomically or functionally defined brain regions, further emphasized by the application of statistical thresholds for the significance of fMRI activation. While the evidence of functional specialization of human brain cortical regions in response to many perceptual, cognitive or emotional stimuli appears compelling, the integrated activity of multiple brain areas is involved even in the simplest sensorimotor tasks [3], in keeping with the dichotomic principles of functional segregation and integration underlying the brain's functional organization. Multi-
⁎ Corresponding author. E-mail addresses: [email protected], [email protected] (A. Bifone) 0730-725X/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2010.07.001
variate analyses of fMRI time series have been developed to assess the distributed nature of brain neurophysiological response and to address interactivity among different structures [4,5]. All of these approaches rely on the evaluation of some kind of correlation or covariance between spatially remote neurofunctional events. In this context, statistical dependencies among signals originating in different brain regions are interpreted in terms of functional connectivity, as opposed to structural connectivity, which denotes the presence of physical neuronal connections between remote brain structures [6]. While the concept of functional connectivity is thought to capture the flow of information between brain regions, and the parallel nature of brain functional processes, it should be noted that it does not necessarily imply the presence of actual physical links between the regions involved nor a causal relationship between spatially remote activations. In order to determine the influence that a neural system exerts over another, a causal model can be fitted to the data to define the effective connectivity between different regions [7]. In the case of fMRI, effective connectivity is inferred from the application of an anatomically constrained model to the analysis of interregional covariance of activity [8]. Functional connectivity has been investigated in both humans and laboratory animals with a number of techniques and using different definitions of the notion of co-varying
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
activity. At the cellular level, functional connectivity has been measured by assessing the temporal coherence in the spiking activity of individual neurons [9]. On a larger spatial scale, electroencephalography (EEG) has been applied to evaluate cross-correlation between signals from pairs of scalp electrodes and their coherence in specific frequency bands [10,11] under evoked response or resting conditions. Interregional correlations in the metabolic responses across subjects have been used to investigate functional connectivity from PET studies with FDG in humans [12,13] or from autoradiographic data in laboratory animals [14]. The inception of fMRI, a neurofunctional imaging method presenting good spatial and temporal resolution, has greatly expanded the repertoire of methods to study functional interactivity in the human brain, at rest or while performing a variety of tasks. Increased flexibility has also enabled the exploration of different measures of interregional covariance, including temporal correlation, and correlations in the fMRI response amplitude across conditions and subjects (for a review of these methods, see Horwitz [15] and references therein). It should be noted that while all these approaches represent legitimate definitions of functional connectivity, they interrogate different aspects of the regional interdependence and are not necessarily equivalent. The recent discovery that spontaneous, low-frequency fluctuations in the fMRI signals (based on the BOLD effect) from the human brain in the absence of an explicit cognitive task or stimulus exhibit coherent patterns within defined networks has opened an interesting avenue of investigation, often referred to as ‘resting-state fMRI’ [16,17]. By way of example, a network of functional connectivity corresponding to brain regions whose activity is higher at rest than during a range of cognitive tasks has been identified and interpreted as evidence in support of the existence of a ‘default mode’ of baseline brain function [18]. Interestingly, alterations in resting state functional connectivity have been observed under a number of pathological conditions [19], including Alzheimer [20], multiple sclerosis [21] and schizophrenia [22]. While there is a large body of work assessing functional connectivity in the human brain following the intense research in this area in the past decade, the extension of these methods to preclinical species, and particularly to rodents, has been more recent (notwithstanding the original demonstration of coupled responses by Soncrant et al. [14] in the late 1980s). The study of functional connectivity in laboratory animals, where imaging can be combined with more invasive methods, may provide a better understanding of the physiological basis of the interregional correlations in fMRI signals observed in humans and of the origin of the alterations in functional connectivity observed in pathological states. However, limitations also exist, mostly related to the use of anesthetic agents that reduce animal motion and stress, and to the substantially more limited repertoire of functional stimuli that can be applied in laboratory animals.
1201
Several studies in laboratory animals have followed the path of human resting state fMRI exploiting the BOLD effect. Synchronous low-frequency fluctuations were detected bilaterally in the primary somatosensory cortex of the rat brain at rest under alpha-chloralose anesthesia [23]. Interestingly, a significant correlation between fMRI responses and delta oscillations measured with epidural EEG was observed, thus suggesting a link between spontaneous fMRI fluctuations and the underlying neuronal electrical activity. Similarly correlated fluctuations were also detected in the rat under isoflurane [24], halothane [25] or medetomidine anesthesia [26–28]. All these studies used a seed-correlation region approach and demonstrated encouraging correlations between bilaterally symmetrical cortical regions. However, coherent networks of connectivity akin to those observed in humans (e.g., the default mode network) have not been observed in the rat using this method to date. A conceptually different approach to the study of functional connectivity in rodents has been explored using pharmacological MRI (phMRI), i.e., fMRI applied to map spatio-temporal patterns of brain activity induced by acute pharmacological challenge [29]. The main merits of this approach lie in its ability to elicit robust and reliable activations even under anesthesia conditions and to enable selective stimulation of different neurotransmitter systems, thus providing a playground to study the neurochemical bases of fMRI responses and their connectivity. Several recent studies using this approach have shown exquisite delineation of focal patterns of correlated responses corresponding to key neurotransmitter pathways [30–34]. Moreover, network analyses of functional connectivity patterns obtained with phMRI have provided the first evidence of organized networks of functional connectivity in the rat brain, thus demonstrating the potential of this method to explore the brain's functional architecture in animal models. Here, we summarize the key principles and methodological aspects of correlation analysis of phMRI data and review recent studies exploring pharmacologically induced or modulated functional connectivity in the rat brain. Moreover, we discuss novel methods based on a network representation of functional connectivity data and their application to resolve the functional organization of the rat brain.
2. Correlation analysis in phMRI In a typical phMRI experiment, a drug challenge is administered systemically (e.g., intravenously or intraperitoneally) during the acquisition of the fMRI time series [35–37]. The time resolution of fMRI is determined by the time scale of the hemodynamic response following neuronal excitation and is of the order of seconds. This temporal resolution is sufficient to capture the time course of the slow spontaneous fluctuations measured by resting state fMRI (typically below 0.15 Hz). The phMRI signal changes
1202
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
following a drug challenge are much slower (typically lasting tens of minutes or more) than the hemodynamic response to a single neural ‘event’ [38]. Moreover, the signal time course reflects drug kinetics and pharmacodynamic effects and, despite rigorous control over animal physiology and experimental parameters, a degree of variability in the relative response amplitude in different brain regions across subjects is typically observed [37,39]. This variability can be leveraged by an alternative approach based on the calculation of interregional correlations in the response amplitude across subjects [30], in the fashion of the procedures applied in metabolic PET or autoradiography studies [13,14]. Schematically, the correlation analysis procedure used in phMRI studies can be summarized in terms of the following steps. (A) Individual subject time series are co-registered to a common stereotaxic space [40]; (B) Time courses from individual image voxels are extracted for each subject, and suitable signal regressors are fitted to the data to determine response amplitudes. Data-driven signal decompositions strategies (e.g., wavelet-cluster analysis [41,42]) have proven particularly effective in providing signal regressors that fit experimental data with excellent sensitivity, particularly in the case of highly dynamic phMRI signal changes. In the present context, we understand response amplitude to represent the magnitude of the post-injection signal change elicited by the acute drug challenge. These values provide voxel-specific vectors of response amplitude across subjects. (C) Inter-subject correlations are then calculated for each voxel in reference to a selected ‘seed’ region, using the vector of response amplitudes from Step B. (D) Finally, statistical correlation maps, i.e., maps of voxels whose response amplitude correlates signifi-
cantly (in a statistical sense) with those in the reference region are generated. This approach leverages variations in the spatial profile of the response observed across subjects following drug challenge. Fig. 1 provides an outline of this process and an exaggerated visual impression of the intersubject variability in the phMRI response profile. 3. Application to mapping functional connectivity in neurotransmitter systems The first demonstration of this approach applied correlation analysis of regional cerebral blood volume changes in the rat induced by acute challenge with D-amphetamine and fluoxetine, two widely used (and abused) drugs that target key monoaminergic systems [30]. One of the main effects of D-amphetamine in the brain is the stimulation of release of dopamine (DA), as well as other monoamines, resulting in increased levels of DA at the terminals of dopaminergic neurons [43]. The psychoactive and reinforcing properties of D-amphetamine are thought to be mediated by stimulation of the mesolimbic DA pathway originating from the ventral tegmental area (VTA) in the midbrain and projecting to the ventral part of the striatum, the nucleus accumbens [43]. Hence, the VTA represents the natural choice of seed correlation region for the analysis of the responses to D -amphetamine. Correlation maps referenced to the VTA clearly delineated the parallel major axes of this pathway forward through the medial forebrain bundle to the ventral striatum (Fig. 2). Other downstream patterns of functional connectivity could be identified by moving the reference region along the functional connectivity pathway [31]. With the use of the shell of the nucleus accumbens as a seed region, a functional connectivity pattern largely coincident with that of the VTA, but extending to the medial pre-frontal cortex, was identified, consistent with the
Fig. 1. Schematic overview of the principle underlying across-subject functional correlation. Different subjects show different spatial profiles of responses to the pharmacological challenge. This variability is leveraged to extract response vectors for each image voxel (left-hand panel). Intersubject correlations are then calculated for each pair of image voxels or for a voxel in reference to a seed region (right-hand panel; a–e indicate representative responses in brain regions A and B across different individual subjects).
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
Fig. 2. Map of the responses to D-amphetamine covarying (across-subjects) with that in the ventral tegmental area (VTA). The pattern of connectivity delineates the mesolimbic DA pathway extending forward to the striatum.
known reciprocal projections of this region to the accumbens. Moving further down the DA pathway, broader cortical involvement, including somatosensory and motor cortices, was observed. In this way, it was possible to discriminate between connectivity in the primary dopaminergic projections and connectivity with downstream structures. Fluoxetine (Prozac) is a canonical selective serotonin reuptake inhibitor [44]; stimulation of the serotonin system is thought to mediate the antidepressant properties of this drug. The main source of serotonergic projections to the forebrain originates in the raphe nuclei. Also in this case, correlation analysis using the raphe as a reference region showed a focal delineation of major ascending serotonergic projections to the forebrain and cortex [30], and a network of subcortical structures including the hippocampus and amygdala, a pattern of connectivity consistent with the drug's mechanism of action. For both drugs, the neuroanatomically specific distributions obtained by the seed region connectivity approach were far more meaningful than the results of massively univariate maps of mean amplitude differences between drug and vehicle groups. The latter identified widespread strong responses in cortical regions, but the compelling subcortical circuits were resolved only when the covariation in response between different brain regions was examined. This study represented the first in vivo demonstration of functional connectivity in discrete neurotransmitter systems [30]. The correlation analysis was guided by a judicious
1203
choice of seed region that was informed by previous knowledge of the systems under investigation. While the confirmatory evidence produced in this study represented convincing validation of the method, approaches that are independent of the specific choice of reference region would be desirable for more exploratory studies where the mechanism of action and behavioral/clinical effects of the drug are less well characterized. A more general approach to study patterns of connectivity from phMRI data was proposed in Ref. [31], where an hypothesis-free cluster analysis was applied to resolve networks of correlated brain activity stimulated by Damphetamine without prior definition of a seed correlation region. A volume-of-interest (VOI) parcellation of the brain volume into Nv=48 bilateral brain structures was delineated automatically using a 3D digital reconstruction of the Paxinos and Watson rat brain atlas co-registered with the MRI rat brain template [40]. Pairwise correlations within the set of selected structures were calculated by computing the Pearson linear correlation coefficient r between the response vectors in each pair of structures, yielding a 48×48 correlation matrix summarizing the similarities in response profile among the selected VOIs (Fig. 3). In order to identify groups of brain structures within the full correlation matrix that were closely coupled in their response to Damphetamine, these authors applied a k-means clustering algorithm [45], an optimization-based method that minimizes a distance measure between the Nv response vectors. In the case of a D-amphetamine phMRI data set, the algorithm indicated that the 48 response vectors should be optimally clustered into three groups in order to maximize intercluster over intracluster dispersion in the normalized cross-subject response profiles. Interestingly, one of the clusters consisted of brain structures involved in the mesolimbic and nigrostriatal DA pathways, including the VTA and substantia nigra (SN), together with ventral structures along the major axes of these pathways (Fig. 3C), a finding consistent with the mesolimbic signature identified in the seed region approach. It is important to emphasize that the clustering results did not identify merely a trivial solution in which strongly responding regions were grouped together while those responding more weakly were clustered apart. Instead, key brain areas involved in primary DA pathways emerged naturally from the data in the form of a group of tightly interconnected structures, thus overcoming the limitations imposed by the need to introduce a priori knowledge in the analysis. While Damphetamine also stimulates other monoamines, e.g., norepinephrine, the DA system appears to show the tightest connectivity under the effects of this drug. In the phMRI approach, a drug challenge serves the purpose of eliciting a brain response whose intersubject variability provides a basis to calculate interregional correlations. From a pharmacological point of view, it is also important to ascertain whether drug treatment can modulate brain functional connectivity, since this might shed
1204
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
Fig. 3. Interregional correlations in response to D-amphetamine. (A) Cross-correlation matrix for the 48 brain structures examined. (B) Connections between the 48 structures in sagittal view. Each region is indicated by a black dot and the connections by red lines. (C) A cluster of structures identified by the k-clustering algorithm, corresponding to the primary projection of the mesolimbic DA system.
light on the drug mechanism of action at a systems level. One interesting approach to addressing this problem has been proposed in Ref. [32], where the effects of a compound of interest were assessed by examining how it modulated the correlated responses to a different probe compound. Specifically, this study investigated the effects of a DA D3 receptor antagonist, a putative treatment of addiction, on the pattern of functional connectivity stimulated by D-amphetamine. The DA D3 receptor has a particularly focal distribution in the brain, with high levels of expression in dopaminergically innervated mesolimbic brain structures, including the nucleus accumbens, involved in reward-related processes [46,47]. D-Amphetamine was previously shown to elicit phMRI responses that were highly correlated in these key dopaminergic circuits and provided the ideal probe to study the effects of the blockade of D3 receptors on the brain functional connectivity. SB277011A, a potent and selective DA D3 receptor antagonist [47], or its vehicle, was administered in two separate groups of rats 30 min prior to an intravenous D-amphetamine challenge. The temporal interval between the drug administration and the amphetamine challenge was sufficient for SB277011A to reach significant unbound concentration in the brain and for its acute effects on cerebral hemodynamics to subside, so that identical CBV baselines could be obtained in the drug and
vehicle groups at the moment of the D-amphetamine challenge. In order to identify brain regions associated with changes in functional connectivity between the two Damphetamine groups, the phMRI responses to amphetamine were acquired and Pearson linear correlation coefficients summarizing the correlation between the responses in each pair of VOIs were calculated for each group. These were then converted to z-statistics using Fisher's r-to-z transformation [40] resulting in a matrix of the z-statistics for each group corresponding to the pairwise correlations between responses in each pair of VOIs. Direct comparison of the zstatistics corresponding to each cell in the matrix then indicated the pairs of brain regions in which the correlation differed between the two groups. It should, however, be emphasized that the simple approach used by the authors is only one of the many statistical options that could be applied to estimate intergroup differences in connectivity patterns, and several alternative approaches have been proposed [48,49]. Nevertheless, main changes in inter-region correlations associated with the VTA, a region rich in DA D3 receptors, were observed. Specifically, these involved connectivity of the VTA with the ventral subiculum and ventral CA3 fields in the hippocampus, dorsolateral and midline dorsal thalamus, piriform cortex and superior colliculi. In the
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
midline dorsal thalamus and hippocampus, the relationship was altered from a significantly positive correlation with the VTA to a significantly negative one, with a dramatic inversion of the sign of the connectivity. The thalamus and hippocampus are important components of a limbic brain circuit and thought to be involved in drug addiction, providing modulatory inputs to the prefrontal cortex, an important brain region involved in goaldirected behavior. The findings of this study, of a reversed correlation between responses in the VTA and the dorsal thalamus, and a reduced correlation with the ventral striatum, suggest that a modified functional connectivity within this circuit may play a key role in the efficacy of selective DA D3 receptor antagonists in attenuating drug-seeking behavior [50]. Methodologically, this study extended the applicability of intersubject functional connectivity analyses of phMRI data to antagonist–agonist experiments, where modulations in the correlation structure underlying the response to a probe signal may be detected. Although the examples provided are based on the exploitation of CBV-based phMRI, it should be emphasized that the conceptual framework underlying the approach is modality independent and can be straightforwardly applied to different neuroimaging modalities (i.e., BOLD-phMRI) and brain mapping techniques. A recent study [51] has significantly expanded the area of application of these methods, by providing a remarkable example of the use of phMRI, intersubject functional connectivity and advanced mouse genetics to provide a direct link between cells, circuits and behavior. Specifically, the authors combined phMRI and focal pharmacogenetic silencing to spatially resolve behavior-specific circuits controlled by a discrete neuronal population expressed in the central nucleus of the amygdala (CeA), a brain region involved in the control of emotional and behavioral response to fear [52]. Reversible suppression of neural activity in a subset (‘type I’) of CeA neurons was achieved by inducing cell-specific re-expression of the serotonin 1A receptor (Htr1aR) in mice devoid of the endogenous receptor (the resulting mice called Htr1aCeA), a strategy recently described by Tsetsenis et al. [53]. The Htr1aR is coupled with inward rectifying potassium channels whose endogenous or exogenous activation produces rapid and reversible membrane hyperpolarization of the cells expressing the receptor. Hence, the use of systemically administered Htr1aR agonists (like the selective compound 8-OHDPAT) can be exploited to obtain cell-specific inhibition of spontaneous neuronal firing in behaving animals [53]. By combining fMRI with this pharmacogenetic inhibition strategy, the authors were able to identify a novel pathway in the mammalian brain apt to regulate passive and active components of fear responses. Specifically, selective inhibition of type I CeA cells led to widespread increased cortical arousal as visualized by phMRI upon acute administration of the selective Htr1AR agonist 8-OH-DPAT. Intersubject functional connectivity analyses of phMRI time courses were critical in resolving the neuronal circuitry underlying
1205
the increased cortical activity. Significant correlation was found between the CeA and several ventral forebrain cholinergic nuclei such as the substantia innominata and the diagonal band, and the same cholinergic nuclei exhibited tight correlations with the cortical areas activated upon inhibition of type I CeA neurons. The involvement of the cholinergic system was confirmed in additional phMRI studies showing that the cortical arousal was blocked by central (but not by peripheral) cholinergic antagonists. Remarkably, an analysis of the behavioral correlates of cortical activation in Htr1aCeA mice highlighted a pivotal role of type I CeA cells as suppressors of cholinergic activity and exploratory behavior and as promoters of freezing (passive) fear responses, thus leading to the identification of a novel neural pathway that biases fear responses toward either passive or active coping strategies. Methodologically, this work expands the applicability of functional connectivity analyses on genetically modified mouse models and provides at the same time a compelling demonstration of the combined use neuroimaging methods and advanced pharmacogenetic systems as a new powerful paradigm to identify and resolve behaviorally relevant neural circuits in the living brain.
4. Complex network analysis and community structure of functional connectivity in the rat brain Functional connectivity analyses of neuroimaging data aim to elucidate the relationships between signals originating in spatially distinct brain regions. This emphasis on interaction between different brain structures is a good conceptual match for representing the data as a graph, or network, of nodes and links [54]. In this representation, image voxels or parcellated brain regions represent the nodes, and a measure of similarity or correlation in their responses defines the edges linking the nodes [55–58]. In recent years, networks from a wide variety of fields, describing connection patterns as diverse as social collaborations, metabolic pathways or air traffic, have been characterized and found to exhibit rich behaviors beyond those of simple random networks; accordingly, they are referred to as ‘complex networks’ [59]. Network analysis of functional connectivity data to date has concentrated on the characterization of global properties of the network, which can reveal much about the properties of the system. For example, networks derived from human brain imaging data have been shown to possess a scale-free degree distribution and exhibit ‘small world’ behavior — a finding that has implications for information transfer in the brain [54]. Complex networks of functional connectivity derived from phMRI data in the rat have also been recently investigated [33,34]. In these articles, the nodes were defined as individual voxels (0.128 mm3) in the 3D image volume, and the strength of the edge between each pair of nodes (i,j) was based on the Pearson correlation coefficient
1206
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
rij between the intersubject response amplitude vectors in the two voxels. This value was converted into an equivalent z-statistic using Fisher's r-to-z transformation. The magnitudes of these normalized correlation values were used to describe the strength of the correlation between each pair of nodes and used to construct an edge weight matrix Wij. Finally, the edge weight matrix was thresholded and binarized to define a binary adjacency matrix. Analysis of the structural properties of the resulting networks outlined interesting topological features similar to those that have been reported for human brain networks [54], including a small-world structure and a long-tailed distribution of node degrees (i.e., number of connections for certain node), indicative of the presence of hubs, namely, a subset of highly connected nodes in cortical regions. Beyond global characterization, the coexistence of functional segregation and integration in brain activity [60] suggests that some degree of modularity, i.e., subnetworks or clusters of more tightly linked nodes, might exist within functional connectivity networks (cf. Ref. [31]). The issue of the identification of such groups of nodes characterized by denser connections between their constituents than to other nodes outside the group has received considerable recent attention in the field of complex networks. This concept originated in the study of social relationships and has been consequently dubbed ‘community structure.’ The first application of this concept to the analysis of brain functional connectivity networks has been recently demonstrated using phMRI data in rodents challenged with fluoxetine [33]. In this article, a networktheoretic community structure algorithm, based on maximization of a mathematical formalism of ‘modularity,’ was applied to resolve functionally and anatomically segregated communities within a functional connectivity network derived from phMRI responses. Specifically, an algorithm based upon maximizing the value of a mathematical definition of modularity Q [61,62] was applied to identify the number and composition of subnetworks within a widely distributed network of functional connectivity obtained under pharmacological challenge with fluoxetine. For a certain partition of the network, Q measures the difference between the fraction of the edges connecting nodes within communities and the same fraction in the case of a randomly connected network with the same partition. The closer the value of Q is to its theoretical maximum of 1, the stronger the community structure, i.e., the more modular the network. Algorithms seeking a network partition that maximizes modularity yield an estimate of the number of communities into which the network should be optimally split, the composition of each community and an associated value of max(Q). In the case of functional connectivity networks, an optimal partition can reveal the system-level functional organization of the brain under the particular experimental conditions studied, as well as provide a measure of the emergent modularity. The application of this method to the functional connectivity
network in the rat challenged with fluoxetine revealed three communities of nodes (Fig. 4). The pixels in the two largest communities were symmetrically distributed between the left and right hemispheres, and their distributions corresponded closely to known anatomical and functional subdivisions of the rat brain. It is important to emphasize that division was obtained from a network representation based on image pixels with no imposition of symmetry nor any prior anatomical constraints. Interestingly, one of the communities comprised nodes corresponding primarily to subcortical structures — in the striatum, thalamus and amygdala — but also to regions of the hippocampus and entorhinal, medial pre-frontal and cingulate cortices (Fig. 4A). Again, this finding was consistent with the results of the seed region analysis [30]. The finding of pixels in the cingulate and prefrontal areas grouped with those in striatal and thalamic structures is consistent with the fact that these cortical regions are the main cortical target of input from the basal ganglia via extensive reciprocal connections with the thalamus [63]. These cortical regions are anatomically similar to other regions of the cortex yet functionally distinct. Similarly, pixels in the entorhinal cortex and in the
Fig. 4. Anatomical representation of the three communities identified by modularity maximization in a functional network derived from phMRI responses to fluoxetine.
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
hippocampus were assigned to the same community, reflecting the dense connections between these structures. Most other cortical nodes, including those located in the motor, somatosensory and visual cortices, were assigned to a second community (Fig. 4B), while a third community (Fig. 4C) mainly comprised pixels near the brain edge, the ventricles and in white matter and the cerebellum. The community structure approach is somewhat akin to other multivariate methods, like principal component analysis and independent component analysis, which have been extensively applied to seek structure within imaging data in a model-independent way [2,64,65]. Unlike these, however, this community structure approach explicitly takes into account the topology of the functional connections, with communities defined on the basis of link density and distribution. This concept may be more readily interpretable in biological terms than measures such as the orthogonality [2,66] or statistical independence [64] of spatial modes that are optimized by other algorithms. A key point in this approach is that the identification of communities within a functional imaging network contains information on both segregation and integration. A partition of the overall network into smaller subunits suggests a degree of functional segregation in the response, whereas the set of brain regions — not necessarily contiguous — identified within each subnetwork reflects their integrated action in response to the experimental stimulus. Importantly, the value of the modularity Q provides a measure of the degree of functional segregation in the network and may provide the basis for an operational definition of this. This first demonstration of the potential of community structure analysis of functional connectivity in preclinical species has been rapidly translated to human functional connectivity, and several studies extending this approach to resting state fMRI have subsequently appeared in the literature [54,67]. However, a few questions remain open and are the object of active investigation. For example, binarization of the adjacency matrix was used to make the problem computationally tractable. However, application of a threshold to retain only the strongest correlations ignores information that may be contained in the distribution of the weaker links, and the extension to fully weighted functional connectivity networks is a natural development for this approach, dependent on the availability of computationally efficient partitioning algorithms. Also, introduction of a direction to the edges linking pairs of nodes, perhaps using Granger causality or related concepts, would provide a means of creating directed graphs of connectivity that may be informative of the roles played by different nodes. Lastly, and most importantly, the origin of structured patterns of correlated responses is still largely unknown. Specifically, it is unclear whether the patterns of functional connectivity reflect interregional correlations induced by the drug challenge itself, or the intrinsic organization of the brain, perhaps determined by the structure of the underlying neuronal substrate.
1207
A follow-up article applied community structure analysis to address this question [34]. To this end, the community structures of functional connectivity networks under three different pharmacological challenges, D-amphetamine, fluoxetine and nicotine, were investigated and the emergent community structures under different pharmacological conditions were compared to discriminate between connectivity patterns that are stimulus specific and those independent of the particular neurotransmitter system(s) engaged by the drug, which might thus correspond to general features of the rat brain functional architecture. Interestingly, common features across all three networks revealed two groups of tightly coupled brain structures that responded as functional units independent of the drug, including a network involving the prefrontal cortex and subcortical regions extending from the striatum to the amygdala. This suggests that this network of functional connectivity may reflect a general feature of the brain organization and is consistent with evidence showing strong intrinsic connectivity between neurons in these brain structures [68]. 5. Conclusion Functional connectivity analysis of functional neuroimaging data is a rapidly developing area of research that has brought about much progress in our understanding of the brain functional architecture. While most of the studies in the literature pertain to human subjects, functional connectivity methods and concepts have been recently extended, mutatis mutandis, to preclinical species. Here, we have reviewed the recent literature in this field, with particular emphasis on approaches based on pharmacological MRI, an imaging approach based on stimulation or modulation of brain activity with pharmacological challenges. While conceptually different from the human resting-state fMRI, which relies on spontaneous fluctuations of the fMRI signal in the brain in the absence of external stimuli, the phMRI approach has proven to be a powerful tool to explore functional connectivity in rodents and to map a variety of different neurotransmitter pathways. Pharmacological MRI has also provided a useful playground to explore novel statistical methods of analysis of functional connectivity represented in terms of complex networks. Particularly, network partitioning methods have made it possible for the first time to resolve biologically and anatomically meaningful patterns of correlated responses in the rat brain, thus providing a novel and powerful method to investigate the brain functional architecture in preclinical species. References [1] Posner MI, Petersen SE, Fox PT, Raichle ME. Localization of cognitive operations in the human brain. Science 1988;240:1627–31. [2] Friston KJ. Analyzing brain images: principles and overview. In: Frackowiak RSJ, Friston KJ, Frith C, Dolan R, Mazziotta JC, editors. Human Brain Function. Academic Press; 1997. p. 25–41.
1208
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
[3] Luria A. The working brain. New York: Basic Books; 1973. [4] Rogers BP, Morgan VL, Newton AT, Gore JC. Assessing functional connectivity in the human brain by fMRI. Magn Reson Imag 2007;25: 1347–57. [5] Friston KJ, Jezzard P, Turner R. Analysis of functional MRI timeseries. Hum Brain Mapp 1994;1:153–71. [6] Ramnani N, Behrens TE, Penny W, Matthews PM. New approaches for exploring anatomical and functional connectivity in the human brain. Biol Psychiatry 2004;56:613–9. [7] Friston KJ. Statistical parametric mapping and other analyses of functional imaging data. In: Toga AW, Mazziotta JC, editors. Brain mapping: the methods. San Diego: Academic Press; 1996. p. 363–86. [8] Buchel C, Friston K. Assessing interactions among neuronal systems using functional neuroimaging. Neural Netw 2000;13:871–82. [9] Gerstein GL, Perkel DH. Simultaneously recorded trains of action potentials: analysis and functional interpretation. Science 1969;164: 828–30. [10] Adey WR, Walter DO, Hendrix CE. Computer techniques in correlation and spectral analyses of cerebral slow waves during discriminative behavior. Exp Neurol 1961;3:501–24. [11] Pfurtscheller G, Andrew C. Event-related changes of band power and coherence: methodology and interpretation. J Clin Neurophysiol 1999;16:512–9. [12] Bartlett EJ, Brown JW, Wolf AP, Brodie JD. Correlations between glucose metabolic rates in brain regions of healthy male adults at rest and during language stimulation. Brain Lang 1987;32:1–18. [13] Horwitz B, Duara R, Rapoport SI. Intercorrelations of glucose metabolic rates between brain regions: application to healthy males in a state of reduced sensory input. J Cereb Blood Flow Metab 1984;4: 484–99. [14] Soncrant TT, Horwitz B, Holloway HW, Rapoport SI. The pattern of functional coupling of brain regions in the awake rat. Brain Res 1986;369:1–11. [15] Horwitz B. The elusive concept of brain connectivity. NeuroImage 2003;19:466–70. [16] Beckmann CF, DeLuca M, Devlin JT, Smith SM. Investigations into resting-state connectivity using independent component analysis. Philos Trans R Soc Lond B Biol Sci 2005;360:1001–13. [17] Biswal B, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn Reson Med 1995;34:537–41. [18] Greicius MD, Krasnow B, Reiss AL, Menon V. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis. Proc Natl Acad Sci U S A 2003;100:253–8. [19] Greicius M. Resting-state functional connectivity in neuropsychiatric disorders. Curr Opin Neurol 2008;21:424–30. [20] Li SJ, Li Z, Wu G, Zhang MJ, Franczak M, Antuono PG. Alzheimer disease: evaluation of a functional MR imaging index as a marker. Radiology 2002;225:253–9. [21] Lowe MJ, Phillips MD, Lurito JT, Mattson D, Dzemidzic M, Mathews VP. Multiple sclerosis: low-frequency temporal blood oxygen leveldependent fluctuations indicate reduced functional connectivity initial results. Radiology 2002;224:184–92. [22] Zhou Y, Shu N, Liu Y, Song M, Hao Y, Liu H, et al. Altered restingstate functional connectivity and anatomical connectivity of hippocampus in schizophrenia. Schizophr Res 2008;100:120–32. [23] Lu H, Zuo Y, Gu H, Waltz JA, Zhan W, Scholl CA, et al. Synchronized delta oscillations correlate with the resting-state functional MRI signal. Proc Natl Acad Sci U S A 2007;104:18265–9. [24] Kannurpatti SS, Biswal BB, Kim YR, Rosen BR. Spatio-temporal characteristics of low-frequency BOLD signal fluctuations in isoflurane-anesthetized rat brain. NeuroImage 2008;40:1738–47. [25] Majeed W, Magnuson M, Keilholz SD. Spatiotemporal dynamics of low frequency fluctuations in BOLD fMRI of the rat. J Magn Reson Imaging 2009;30:384–93. [26] Pawela CP, Biswal BB, Hudetz AG, Schulte ML, Li R, Jones SR, et al. A protocol for use of medetomidine anesthesia in rats for extended
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45] [46]
studies using task-induced BOLD contrast and resting-state functional connectivity. NeuroImage 2009;46:1137–47. Pawela CP, Biswal BB, Cho YR, Kao DS, Li R, Jones SR, et al. Resting-state functional connectivity of the rat brain. Magn Reson Med 2008;59:1021–9. Zhao F, Zhao T, Zhou L, Wu Q, Hu X. BOLD study of stimulationinduced neural activity and resting-state connectivity in medetomidinesedated rat. NeuroImage 2008;39:248–60. Leslie RA, James MF. Pharmacological magnetic resonance imaging: a new application for functional MRI. Trends Pharmacol Sci 2000;21: 314–8. Schwarz AJ, Gozzi A, Reese T, Bifone A. In vivo mapping of functional connectivity in neurotransmitter systems using pharmacological MRI. NeuroImage 2007;34:1627–36. Schwarz AJ, Gozzi A, Reese T, Bifone A. Functional connectivity in the pharmacologically activated brain: resolving networks of correlated responses to D-amphetamine. Magn Reson Med 2007;57:704–13. Schwarz AJ, Gozzi A, Reese T, Heidbreder CA, Bifone A. Pharmacological modulation of functional connectivity: the correlation structure underlying the phMRI response to D-amphetamine modified by selective dopamine D3 receptor antagonist SB277011A. Magn Reson Imaging 2007;25:811–20. Schwarz AJ, Gozzi A, Bifone A. Community structure and modularity in networks of correlated brain activity. Magn Reson Imaging 2008;26: 914–20. Schwarz AJ, Gozzi A, Bifone A. Community structure in networks of functional connectivity: resolving functional organization in the rat brain with pharmacological MRI. NeuroImage 2009;47:302–11. Gozzi A, Schwarz A, Reese T, Bertani S, Crestan V, Bifone A. Regionspecific effects of nicotine on brain activity: a pharmacological MRI study in the drug-naïve rat. Neuropsychopharmacology 2006;31:1690–703. Gozzi A, Large C, Schwarz A, Bertani S, Crestan V, Bifone A. Differential effects of antipsychotic and glutamatergic agents on the phMRI response to phencyclidine. Neuropsychopharmacology 2008;33:1690–703. Schwarz AJ, Zocchi A, Reese T, Gozzi A, Garzotti M, Varnier G, et al. Concurrent pharmacological MRI and in situ microdialysis of cocaine reveal a complex relationship between the central hemodynamic response and local dopamine concentration. NeuroImage 2004;23: 296–304. Schwarz AJ, Reese T, Gozzi A, Bifone A. Functional MRI using intravascular contrast agents: detrending of the relative cerebrovascular (rCBV) time course. Magn Reson Imaging 2003;21:1191–200. Schwarz A, Whitcher B, Reese T, Gozzi A, Bifone A. Group-level data-driven phMRI analysis. Book of Abstracts: Thirteenth Annual Meeting of the International Society of Magnetic Resonance in Medicine, 13; 2005. p. 157. Schwarz AJ, Danckaert A, Reese T, Gozzi A, Paxinos G, Watson C, et al. A stereotaxic MRI template set for the rat brain with tissue class distribution maps and co-registered anatomical atlas: application to pharmacological MRI. NeuroImage 2006;32:538–50. Schwarz AJ, Whitcher B, Gozzi A, Reese T, Bifone A. Study-level wavelet cluster analysis and data-driven signal models in pharmacological MRI. J Neurosci Methods 2006;159:346–60. Whitcher B, Schwarz AJ, Barjat H, Smart SC, Grundy RI, James MF. Wavelet-based cluster analysis: data-driven grouping of voxel time courses with application to perfusion-weighted and pharmacological MRI of the rat brain. NeuroImage 2005;24:281–95. Everitt BJ, Robbins TW. Neural systems of reinforcement for drug addiction: from actions to habits to compulsion. Nat Neurosci 2005;8: 1481–9. Stahl SM. Mechanism of action of serotonin selective reuptake inhibitors: serotonin receptors and pathways mediate therapeutic effects and side effects. J Affect Disord 1998;51:215–35. Everitt BJ, Landau S, Leese M. Cluster analysis. London: Arnold; 2001. Schwarz A, Gozzi A, Reese T, Bertani S, Crestan V, Hagan J, et al. Selective dopamine D(3) receptor antagonist SB-277011-A potentiates
A. Bifone et al. / Magnetic Resonance Imaging 28 (2010) 1200–1209
[47]
[48]
[49] [50]
[51]
[52] [53]
[54]
[55]
[56]
phMRI response to acute amphetamine challenge in the rat brain. Synapse 2004;54:1–10. Heidbreder CA, Gardner EL, Xi ZX, Thanos PK, Mugnaini M, Hagan JJ, et al. The role of central dopamine D3 receptors in drug addiction: a review of pharmacological evidence. Brain Res Brain Res Rev 2005;49:77–105. Holmes AP, Blair RC, Watson JDG, Ford I. Nonparametric analysis of statistic images from functional mapping experiments. J Cereb Blood Flow Metab 1996;16:7–22. Rogers BP, Gore JC. Empirical comparison of sources of variation for FMRI connectivity analysis. PLoS ONE 2008:3. Cervo L, Cocco A, Petrella C, Heidbreder CA, Bendotti C, Mennini T. SB-277011-A, a selective dopamine D3 receptor antagonist, reduces cocaine-seeking behavior in response to drug-associated stimuli in rats. Proc Society for Neuroscience 2003. Gozzi A, Apar J, Giovanelli A, Bertollini C, Crestan V, Schwarz AJ, et al. A neural switch for active and passive fear. Neuron 2010 doi:10.1016/j.neuron.2010.07.008. LeDoux J. The emotional brain, fear, and the amygdala. Cell Mol Neurobiol 2003;23:727–38. Tsetsenis T, Ma XH, Lo Iacono L, Beck SG, Gross C. Suppression of conditioning to ambiguous cues by pharmacogenetic inhibition of the dentate gyrus. Nat Neurosci 2007;10:896–902. Bullmore E, Sporns O. Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci 2009;10:186–98. Salvador R, Suckling J, Schwarzbauer C, Bullmore E. Undirected graphs of frequency-dependent functional connectivity in whole brain networks. Philos Trans R Soc Lond B Biol Sci 2005;360:937–46. Eguiluz VM, Chialvo DR, Cecchi GA, Baliki M, Apkarian AV. Scalefree brain functional networks. Phys Rev Lett 2005;018102:94.
1209
[57] Achard S, Salvador R, Whitcher B, Suckling J, Bullmore E. A resilient, low-frequency, small-world human brain functional network with highly connected association cortical hubs. J Neurosci 2006;26: 63–72. [58] Achard S, Bullmore E. Efficiency and cost of economical brain functional networks. PLoS Comput Biol 2007;e17:3. [59] Strogatz SH. Exploring complex networks. Nature 2001;410:268–76. [60] Tononi G, Sporns O, Edelman GM. A measure for brain complexity: relating functional segregation and integration in the nervous system. Proc Natl Acad Sci U S A 1994;91:5033–7. [61] Newman ME, Girvan M. Finding and evaluating community structure in networks. Phys Rev E Stat Nonlin Soft Matter Phys 2004;69: 026113. [62] Newman ME. Modularity and community structure in networks. Proc Natl Acad Sci U S A 2006;103:8577–82. [63] Uylings HB, Groenewegen HJ, Kolb B. Do rats have a prefrontal cortex? Behav Brain Res 2003;146:3–17. [64] McKeown MJ, Makeig S, Brown GG, Jung TP, Kindermann SS, Bell AJ, et al. Analysis of fMRI data by blind separation into independent spatial components. Hum Brain Mapp 1998;6:160–88. [65] Beckmann CF, Smith SM. Probabilistic independent component analysis for functional magnetic resonance imaging. IEEE Trans Med Imaging 2004;23:137–52. [66] Friston KJ, Frith CD, Liddle PF, Frackowiak RS. Functional connectivity: the principal-component analysis of large (PET) data sets. JCBFM 1993;13:5–14. [67] Meunier D, Achard S, Morcom A, Bullmore E. Age-related changes in modular organization of human brain functional networks. NeuroImage 2009;44:715–23. [68] Palomero-Gallagher N, Zilles K. The rat nervous system. In: Paxinos G, editor. London: Elsevier; 2010. p. 729–57.