In vivo mapping of functional connectivity in neurotransmitter systems using pharmacological MRI

www.elsevier.com/locate/ynimg

In vivo mapping of functional connectivity in neurotransmittersystems using pharmacological MRI

Adam J. Schwarz,⁎ Alessandro Gozzi, Torsten Reese, and Angelo Bifone

Department of Neuroimaging, Psychiatry Centre of Excellence in Drug Discovery, GlaxoSmithKline, Via Fleming 4, 37135 Verona, Italy

Received 10 August 2006; revised 7 November 2006; accepted 9 November 2006Available online 26 December 2006

Pharmacological MRI (phMRI) methods map the hemodynamicresponse to drug challenge as a surrogate for changes in neuronalactivity. However, the central effects of drugs can be complex andinclude activity at the primary site of action, downstream effects inother brain regions and direct effects on vasculature and neurovascularcoupling. Univariate analysis, normally applied to phMRI data, doesnot discriminate between these effects, and can result in anatomicallynon-specific activation patterns. We analysed inter-subject correla-tions in the amplitude of the slow phMRI response to map functionallyconnected brain regions recruited in response to pharmacologicalchallenge. Application of D-amphetamine and fluoxetine revealed well-defined functional structure underlying the widespread signal changesdetected via standard methods. Correlated responses were found todelineate key neurotransmitter pathways selectively targeted by thesedrugs, corroborating a tight correspondence between the phMRIresponse and changes in neurotransmitter systems specific to thepharmacological action. In vivo mapping of correlated responses in thisway greatly extends the range of information available from phMRIstudies and provides a new window into the function of neurotrans-mitter systems in the active state. This approach may provide newimportant insights regarding the central systems underlying pharma-cological action.© 2006 Elsevier Inc. All rights reserved.

Abbreviations: Acb, nucleus accumbens; AcbSh, shell of the nucleusaccumbens; Amyg, amygdala; CPu, caudate putamen; DRD, dorsal raphenucleus, dorsal par; Ent, entorhinal cortex; hc, hippocampus; Ins, insularcortex; LHb, lateral habenular nucleus; Mctx, motor cortex; MDthal,mediodorsal thalamus; MeA, medial amygdaloid nucleus; MnR, medianraphe nucleus; PrL, pre-limbic cortex; RLi, rostral linear nucleus of theraphe; RSctx, retrosplenial cortex; SN, substantia nigra; SSctx, somatosen-sory cortex; SupCo, superior colliculus; thal, thalamus; thalMD, midlinedorsal thalamus; Vctx, visual cortex; vhc, ventral hippocampus; VTA,ventral tegmental area.⁎ Corresponding author. Fax: +39 045 821 8375.E-mail address: [email protected] (A.J. Schwarz).Available online on ScienceDirect (www.sciencedirect.com).

1053-8119/$ - see front matter © 2006 Elsevier Inc. All rights reserved.doi:10.1016/j.neuroimage.2006.11.010

Introduction

Understanding the neural substrates underlying pharmacologicalaction on different neurotransmitter systems is a key facet ofneuroscience research and critical in the development of newmedicines in psychiatry and neurology. Imaging technologies arenow well established in clinical research and are playing anincreasingly important role pre-clinically, providing new insightsinto brain function and opportunities for translation to and from theclinic. Recently, pharmacological MRI (phMRI) – the application offunctional MRI (fMRI) methods to examine the central effects ofsystemically administered drugs – has offered an in vivowhole-brainview on resulting changes in brain activity (Leslie and James, 2000;Jenkins et al., 2003). By measuring changes in hemodynamicparameters, supposedly driven by underlying alterations in neuralactivity, this approach has been used to investigate the direct centraleffect of drugs per se and to determine how drugs modulate theresponse to other ‘probe’ compounds (Reese et al., 2000; Marotaet al., 2000; Shah et al., 2004; Kalisch et al., 2004; Schwarz et al.,2004a; Dixon et al., 2005; Chen et al., 2005; Gozzi et al., 2006).

However, the central action of drugs can be complex; observedsignal changes may involve the primary site of action of the drugalong with downstream effects in other parts of the brain.Moreover, certain compounds can also directly or indirectly affectthe vasculature or neurovascular coupling. Standard univariateanalyses of such data – independent group comparisons at eachimage voxel – do not distinguish between these effects; even drugsthat are selective for a particular target can thus give rise to awidespread hemodynamic response, complicating the interpreta-tion of such data. How closely the phMRI response reflectsunderlying changes in neuronal activity, therefore, remains to befully determined. A complementary approach to the analysis offunctional imaging data is to explicitly examine co-varying, orcorrelated, signals between different brain regions. The existenceof correlated changes – across time, conditions or subjects – is theninterpreted as a sign of underlying functional connectivity (Friston,1994; Biswal et al., 1995; Hampson et al., 2002; Worsley et al.,2005). This has been applied to studies of signal changes duringcognitive tasks or resting state brain activity; here, we extend thisapproach to resolve functional connectivity patterns resulting from

1628 A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

the activation of specific neurotransmitter systems by pharmaco-logical challenge.

Because the phMRI signal changes following acute drugchallenge are much slower (typically lasting tens of minutes ormore) than the hemodynamic response to a single neural ‘event’and can affect many brain regions, intra-subject temporalcorrelations in the response may be anatomically non-specificin individuals; variability across subjects then complicates thegeneration of a suitably representative and meaningful groupimage. In this paper, we present an inter-subject functionalconnectivity analysis of the acute phMRI response and apply it toresolve different networks of brain regions mediating relativecerebral blood volume (rCBV) changes in the rat followingchallenge with D-amphetamine or fluoxetine. Both compoundsare widely used (and abused) and represent prototypicalpsychoactive drugs commonly applied to probe the dopamineand serotonin neurotransmitter systems, respectively. D-Ampheta-mine provides a dopaminergic stimulus whose central responsecan be modulated by more selective dopamine ligands andfluoxetine is representative of the selective serotonin re-uptakeinhibitor (SSRI) class of antidepressants. Our results clearlydelineate focal circuits corresponding to known pathways as wellas other distinct patterns of co-varying responses, revealing a richfunctional structure underlying the widespread, non-specificsignal changes detected via standard phMRI group activationmaps.

Methods

Overview of the functional connectivity analysis

The correlation analysis procedure we present in the currentpaper can be summarized in terms of the following steps. Afterspatial co-registration of the individual subject time series:

1. Calculate maps of response amplitude for each subject. In thepresent context, we understand response amplitude to representthe magnitude of the post-injection signal change elicited by theacute drug challenge.

2. Extract time courses from specified volume(s) of interest (VOIs),for each subject, and calculate response amplitudes for each.These values provide VOI-specific vectors of response ampli-tude across subjects.

3. Calculate inter-subject correlation maps referenced to selectedreference (or ‘seed’) region(s), using the vector of responseamplitudes from step 2.

In this way, maps of voxels whose response amplitudecorrelates with those in the reference region can be generated.This approach leverages variations not only in overall magnitudebut in the spatial profile of the response observed across subjectsfollowing drug challenge. Below, we provide details of animplantation of this within the framework of the general linearmodel (GLM).

Acquisition details

Animal preparationAll experiments were carried out in accordance with Italian

regulations governing animal welfare and protection. Protocols

were also reviewed and consented to by a local animal carecommittee, in accordance with the guidelines of the Principles ofLaboratory Animal Care (NIH publication 86-23, revised 1985).

Male Sprague-Dawley rats (N=35, weight (mean±SEM) 287±5 g) were surgically prepared and monitored as detailed previously(Schwarz et al., 2004a; Gozzi et al., 2006) and imaged under 0.8%halothane maintenance anesthesia, neuromuscular blockade andartificial ventilation. The ventilation parameters were adjusted foreach animal such that its blood gas values remained withinphysiological range (pCO2 39±1; pO2 199±3). Peripheral bloodpressure remained within the autoregulatory range associated withhalothane anaesthesia (Zaharchuk et al., 1999; Gozzi et al., 2006).Relative changes in blood pressure following injection werequantified to investigate any central responses correlating acrosssubjects with peripheral blood pressure variations.

MRI protocolMR data were acquired using a Bruker Biospec 4.7T scanner

with a cylindrical volume coil for RF transmit and a Brukerquadrature “rat brain” surface receive coil. We analyzed two studies,in which the animals were challenged with either D-amphetamine orfluoxetine respectively. In each study, a T2-weighted anatomicalimage volume followed by a T2-weighted time series were acquiredfor each subject, both using the RARE sequence (Hennig et al.,1986). After five reference time frames in the time series acquisition,2.67 ml/kg of the blood pool contrast agent Endorem (Guerbet,France) was injected so that subsequent signal changes would reflectalterations in rCBV (Mandeville et al., 1998; Reese et al., 2000).

The specific acquisition details for the D-amphetamine studywere as follows. The T2-weighted anatomical image volume wasacquired using the RARE sequence with RARE factor 32, matrix256×256, FOV 40 mm, 16 contiguous 1 mm slices, TReff=5500 ms, TEeff=76 ms. The images were acquired in the coronalplane, centred 8 mm caudal from the posterior edge of the olfactorybulb. This was followed by a time series acquisition using the samesequence with a reduced image matrix of 128×128 and TReff=2700 ms, TEeff=100 ms, but with the image volume spatiallycoincident with that of the anatomical images. The in-plane pixeldimension was thus 312.5 μm. The acquisition time per time seriesimage volume was 20 s; four successive excitations were averagedto yield a final effective time resolution of 80 s and a total of 64image volumes per subject. Either D-amphetamine (Sigma, Milan,Italy; 1 ml, 1 mg/kg, N=17) or vehicle (saline, 1 ml, N=7) wasinjected via the femoral vein over a period of 1 min, following30 min of equilibration after contrast agent administration.Subsequent signal changes were tracked for ~20 min, capturingthe robust initial rCBV changes following intravenous injection(Schwarz et al., 2004a).

For the fluoxetine study the anatomical image acquisitionparameters were the same as those for D-amphetamine. The timeseries parameters were as per the D-amphetamine study exceptthat it comprised 8 contiguous 2-mm-thick slices. The totalacquisition time per time series image volume was 10 s. Twosuccessive excitations were averaged to yield a final effectivetime resolution of 20 s. A total of 128 image volumes per subjectwere acquired. Either fluoxetine (Sigma, Milan, Italy; 10 mg/kg,1 ml, N=7) or vehicle (water, 1 ml, N=4) was injectedintraperitoneally, over a period of 1 min, 20 min after contrastagent administration. Subsequent signal changes were tracked for~20 min, again capturing the robust initial rCBV response to thedrug.

1629A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

Analysis details

MR image pre-processingAnatomical and time series data were converted to Analyze

(AVW 7.5) format and signal intensity changes in each time serieswere transformed into fractional rCBVon a pixel-wise basis, usinga constrained exponential model of the gradual elimination ofcontrast agent from the blood pool to provide a robust prediction ofpost-injection background signal and remove the worst effects ofthis systematic trend in the resulting rCBV data (Schwarz et al.,2003). Data for each subject were then spatially normalized to astereotaxic rat brain template (Schwarz et al., 2006) by computinga nine degree-of-freedom affine transform for the anatomical imageand applying the resulting transformation matrix to the accom-panying rCBV time series (FSL/FLIRT v.5.2). Finally, the rCBVdata were multiplied by a brain parenchyma mask to remove extra-cranial and CSF contributions.

Time series analysisImage-based time series analysis of the response in individual

subjects was carried out in the framework of the GLM using FSL/FEAT v5.43 in order to calculate maps of the post-injectionresponse amplitude in each subject. The images were spatiallysmoothed with a Gaussian kernel of FWHM=0.8 mm, corre-sponding to 2.56× the in-plane pixel dimension. (In fact, all imageprocessing was performed with the physical pixel dimensions inthe image header scaled up by a factor of 10, in order to ensurecompatibility with any explicit length scales that may be encodedin algorithms designed for use with human data. In the scaled data,then, the acquisition pixel volumes corresponded to 3.125×3.125×10 mm3 for the D-amphetamine study and 3.125×

Fig. 1. Design matrix and example time series from the D-amphetamine study. (a) Thonset and slow offset phases of the typical post-injection signal changes (left columThese latter two columns were orthogonalised to the signal model (regressor of inD-amphetamine challenge, with the full and partial (estimated signal model only) rthe response amplitude. (c) rCBV response amplitudes from the CPu across all

3.125×20 mm3 for the fluoxetine study, and the smoothing kernelto 8 mm.)

Both D-amphetamine and fluoxetine elicited rapid rCBVincreases following injection, reaching maximal values afterapproximately 5 and 3 min, respectively, followed by much slowersignal decreases. The design matrix for each study comprised asignal model function identified by study-level Wavelet ClusterAnalysis (WCA), the temporal derivative of this regressor and alinear ramp (both orthogonalised to the model function) (Schwarzet al., in press). This allows a good model fit to signals whosetemporal response profile can vary slightly across subjects andbrain regions. The design matrix for the D-amphetamine time seriesdata is shown in Fig. 1(a). The coefficients of the signal modelfunction (partial model) within the full regressed model thusprovided rCBV response amplitude values for subsequent group-level calculations. An example time course from the caudateputamen (CPu) of one animal, along with the full and partialregressed models, is shown in Fig. 1(b).

Volumes of interestVolumes of interest (VOIs) corresponding to selected structures

were delineated using a 3D digital reconstruction of the Paxinosand Watson rat brain atlas (Paxinos and Watson, 1998) co-registered with the rat brain template (Schwarz et al., 2006). TheseVOIs were used as reference structures for image-based correlationmaps. The structures were selected a priori based on theirinvolvement in the dopaminergic and serotonergic systems,observed effects of the drugs or associated pathologies (Drevets,2000; Slattery et al., 2005; Salome et al., 2006) and on featuresobserved in the univariate activation maps. They are summarizedin Table 1.

e design matrix comprised three columns: a signal model capturing the rapidn, thick black line), its temporal derivative and a post-injection linear ramp.terest). (b) An example time course, from the CPu of one animal receivingegressed models. The coefficient of the signal model provides a measure ofsubjects in the D-amphetamine study.

Table 1VOIs used as seed regions in the analysis

D-Amphetamine Fluoxetine

VTA Source of mesolimbic dopamine projections. Raphe nuclei Source of serotonergic projections.SN Source of nigrostriatal dopamine projections. CeA Amygdaloid nucleus, implicated in anxiolysis

and central action of SSRIs.AcbSh Forebrain target of mesolimbic dopamine projections,

implicated in reward system.MeA Amygdaloid nucleus, implicated in anxiolysis.

PrL Involved in goal-directed behavior, reciprocal connectionswith sub-cortical regions and reward circuit.

thalMD Abnormal metabolism in major depressivedisorder (MDD).

CPu Receives dopaminergic afferents and projects to cortical regions. CPu Implicated in acute response to antidepressants.Mctx Region responding strongly to D-amphetamine challenge,

implicated in locomotor effects of D-amphetamine.AcbC Abnormal metabolism in MDD, implicated in

acute response to antidepressants.SSctx Region responding strongly to D-amphetamine challenge. vhc Altered cell proliferation in depression.

Region involved in univariate response.

1630 A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

All VOIs were defined as bilateral, with voxels from both brainhemispheres combined together; the univariate maps for bothcompounds are highly symmetrical and we did not hypothesize anyunilateral effects of the systemic drug administration. For eachsubject, a single rCBV time course from each VOI was calculatedby averaging the signal value from all voxels lying within the VOIat each time point. A GLM analysis of the VOI time courses, usingthe same design matrix as described above, yielded a vector ofresponse amplitudes across subjects for each selected brainstructure. The response amplitudes for the CPu are shown forD-amphetamine in Fig. 1(c), illustrating the spread across subjects.

Group mean responseMaps of group mean response were calculated within the GLM

framework at the group level using FSL/FEAT v.5.43 with multi-level Bayesian (FLAME) inference (Beckmann et al., 2003;Woolrich et al., 2004). The design matrix for each study comprisedregressors capturing the mean signal across animals for each group(drug, vehicle). Positive and negative group mean differences fromvehicle were calculated via the contrasts [+/−1, −/+1]. Since weexpected relatively widespread clusters of activation followingsystemic pharmacological challenge, the resulting maps werethresholded using spatially extended clusters of voxels (Worsleyet al., 1992; Friston et al., 1994) determined by z>2.3 and acorrected cluster significance threshold of p=0.05.

Correlation analysisMaps of response amplitudes correlating across subjects with

the responses in a reference brain region were calculated within theGLM framework at the group level using FSL/FEAT v.5.43 withmulti-level Bayesian (FLAME) inference (Beckmann et al., 2003;Woolrich et al., 2004). For each VOI, the design matrix compriseda regressor capturing the group mean signal and another containingthe zero-meaned response vector across the N subjects in the groupfrom the selected reference structure. The reference VOI wasselected a priori for each map, analogous to the ‘seed voxel’approach (Friston, 1994; Horwitz et al., 1998; Fox et al., 2005). Zstatistic images were calculated via contrasts capturing positive andnegative correlations with the reference response. Since weexpected clusters of correlated activity to be substantially largerthan the smoothness of the data, the maps were thresholded usingspatially extended clusters of voxels (Worsley et al., 1992; Fristonet al., 1994) determined by z>2.3 and a corrected clustersignificance threshold of p=0.05/7=0.007 (i.e., corrected for the7 correlation maps calculated for each group).

To assess the potential confound of central haemodynamicchanges being driven by peripheral blood pressure variations, wealso calculated maps of brain regions correlating with the change inmean arterial blood pressure following drug challenge. We used thesame framework as outlined above except with the blood pressurechanges as the regressor rather than response amplitudes from areference VOI.

Results

Functional connectivity maps resolve dopamine pathways in vivo

Challenge with 1 mg/kg D-amphetamine induced widespreadrCBV increases in the rat brain. This is reflected in the ‘activation’map calculated by performing a massively univariate groupcomparison, testing the null hypothesis that the mean amplitudeof the response to amphetamine is the same as that to vehicle ateach pixel independently (Fig. 2(a)). Small negative rCBVchanges were observed in some animals, but no brain regionsexhibited significant negative effects at the group level. Strongesteffects occurred in the insular, piriform and auditory cortices,along with the ventrolateral striatum, but overall the groupactivation map calculated in this way is anatomically rather non-specific.

However, more localized networks of co-varying changes canbe identified within these same data by considering correlations inresponse amplitude across subjects. In this way, we can map brainregions whose response amplitude closely co-varies with that in aselected reference structure. The ventral tegmental area (VTA) inthe midbrain is the source of the mesolimbic dopamine pathwayprojecting to striatal regions in the forebrain and as such representsa natural choice of initial reference structure. Mapping theamphetamine responses correlating with those in the VTA clearlydelineates the parallel major axes of this pathway forward throughthe ventromedial hypothalamus to the ventral striatum (Figs. 2(b–d)). Correlated response foci in the mediodorsal thalamus alsocorrespond to projections from the VTA. Other structures identifiedas functionally connected to the VTA include regions in the ventralhippocampus and the retrosplenial cortex. The other main source ofdopaminergic projections to the forebrain, the substantia nigra,revealed a similar but more limited correlation pattern, comprisingparallel pathways forward through the lateral hypothamalus to theventral striatum.

Continuing along the mesolimbic dopamine pathway, the shellsub-region of the nucleus accumbens exhibits a functional

1631A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

connectivity pattern largely coincident with that of the VTA, butextending to increased regions in the caudate putamen and medialpre-frontal cortex (mPFC). In turn, functional connectivity mapsreferenced to the pre-limbic region of the mPFC (Fig. 3(a)), or thecaudate putamen, have a much broader coverage and include broadregions of the somatosensory and motor cortices. These latter mapsnow include many of the cortical areas identified in the univariategroup comparison. In this way, we can thus discriminate betweenactivity in primary dopaminergic projections and that in brainregions related less directly to the VTA.

To highlight regions correlating with both or one of twoselected structures, the overlap in correlation maps can bedisplayed (Fig. 3(b)). In this example, it can be seen that regions

Fig. 2. (a) Acute challenge with 1 mg/kg D-amphetamine induced widespread r(b) Mapping the responses covarying across subjects with that in the VTA delineRostro-caudal range of slices shown is approximately bregma −8 mm to bregmalinear regression lines (solid red line) and 95% confidence bands (dashed red linp=0.0011. (d) Schematic illustration of the mesolimbic dopamine projectionsthalamus. Both these connections are resolved in the VTA correlation map (b).

in the striatum (dorsal and ventral), mPFC and ventral hippocam-pus correlate with both the shell of the accumbens (AcbSh) andpre-limbic part of the mPFC (PrL). Responses extending ventrallyback from the AcbSh toward the VTA, retrosplenial cortex and thePons correlate preferentially with the AcbSh, whereas responses inmore extensive regions in the sensorimotor cortices and thalamusco-vary more closely with those in the PrL, but are not significantlycorrelated with activity in the mesolimbic dopamine pathway.Indeed, correlation maps referenced to prefrontal, motor andsensorimotor cortices and to the CPu showed very similardistributions, covering the broad forebrain areas shown in red inFig. 3(b). These maps were qualitatively different to the more focalventral pathways correlating with the VTA and AcbSh.

CBV increases in the rat brain, including extensive cortical involvement.ates the mesolimbic dopamine pathway extending forward to the striatum.+4 mm. (c) Scatterplot of amphetamine responses in AcbSh vs. VTA withe) indicated. Regression equation AcbSh=0.0119+0.5956xVTA, r=0.72,from the VTA to the accumbens and lateral habenular nucleus in the

Fig. 3. (a) Maps of regions correlating with the accumbens shell (AcbSh), pre-limbic cortex (PrL) and somatosensory cortex (SSctx) indicate overlapping regionsof functional connectivity, linking the sub-cortical ventral pathway from the VTA to the cortex. (b) Overlap of correlation maps referenced to the AcbSh and PrLregions involved in the mesolimbic dopamine pathway. The maps are binarised versions of those thresholded as described in Methods. AcbSh correlated backalong the mesolimbic DA pathway toward the VTA, as well as many parts of the CPu, but also includes PrL which, in turn, correlates extensively with corticalregions that themselves are not correlated with the VTA or AcbSh. The rostro-caudal range of slices shown is approximately bregma −8 mm to bregma +4 mm.

1632 A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

Functional connectivity in the active serotonin system

In order to test the applicability of this approach to anotherneurotransmitter system, we also investigated functional connec-tivity patterns underlying the response to a different pharmacolo-gical stimulus. Fluoxetine is a selective serotonin reuptakeinhibitor (SSRI) that thereby stimulates the serotonergic systemin the brain. A standard group mean response map of fluoxetineversus vehicle showed a predominantly cortical effect, withadditional foci in the hippocampus and VTA and scattered regionsin the thalamus (Fig. 4(a)). Although serotonin projections includeextensive innervations of the cerebral cortex, many sub-corticalregions also involved in these pathways and known to be involvedin the central effect of fluoxetine are not identified in the univariateactivation map.

By examining correlated responses in the fluoxetine group, wewere again able to identify patterns of functionally connectedregions that revealed structure in the data beyond that identified inthe group comparison map. The main sources of serotonergicprojections to the forebrain are the raphe nuclei. Using the rapheVOI as a reference structure identified a bilateral pathway to theforebrain (Fig. 4(b)). Correlation maps for the thalamic, striatal,amygdaloid and hippocampal structures investigated showed a highdegree of mutual overlap, implying a tight functional connectivitybetween these sub-cortical regions under fluoxetine challenge.Strong correlations with ventral and posterior hippocampus werealso often present. This network is illustrated by the correlation mapreferenced to the medial amygdaloid nucleus (Fig. 5). Thedistribution here is almost entirely sub-cortical and involves regionsnot detected using a standard group mean comparison with thevehicle group.

Group correlation maps were also calculated using the bloodpressure change in each subject following injection as the correlate;no significant central correlations with blood pressure were foundin either of the D-amphetamine or fluoxetine data sets.

Discussion

Identified functional connectivity patterns correspond to specificneurotransmitter pathways

The main actions of D-amphetamine are pre-synaptic, enhan-cing dopamine release and blocking its reuptake by the dopaminetransporter. This results in increased dopamine levels at theterminals of dopaminergic neurons. The psychoactive and reinfor-cing properties of D-amphetamine are mediated by stimulation ofthe mesolimbic dopamine pathway (Everitt and Robbins, 2005),identified by ex vivo techniques as projecting from the VTA in themidbrain to the ventral striatum (Beckstead, 1976; Oades andHalliday, 1987). The correlation maps obtained with D-ampheta-mine in the present study included a striking delineation offunctionally co-varying regions coinciding with this mesolimbicdopamine pathway. Fluoxetine, in contrast, is a canonical selectiveserotonin reuptake inhibitor (SSRI); stimulation of the serotonergicsystem is thought to mediate its antidepressant activity. Auto-radiography experiments have shown that the main source ofserotonergic projections to the forebrain originate in the raphenuclei (Moore et al., 1978). Our results showed responsescorrelating with those in the raphe delineating major ascendingprojections to the forebrain and cortex, and a network of sub-cortical structures including the hippocampus and amygdala. Thesepatterns of correlated activity provide specific connectivity

Fig. 4. Group mean effect and functional connectivity with the raphe nuclei following fluoxetine challenge. (a) Acute challenge with 10 mg/kg fluoxetine inducedvariable rCBV increases in the rat brain, with the strongest changes widespread in cortical regions. Rostro-caudal range of slices shown is approximately bregma−10 mm to bregma +2 mm. (b) In contrast, responses correlating with those in the raphe nuclei were predominantly sub-cortical and included regions inmesencephalic areas, the thalamus, amygdala and caudate putamen. Ascending projections to forebrain regions are visible in the horizontal slice shown at right.

1633A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

fingerprints of the effects of these drugs, consistent with theirmechanisms of action.

To the best of our knowledge this is the first time these functionalconnections, in the acute response to a pharmacological stimulus,have been visualized in vivo. The delineation of these primarypathways corroborates a tight link between the phMRI response andunderlying activity in the neurotransmitter systems targeted by D-amphetamine and fluoxetine. By following these pathways,

Fig. 5. Statistical parametric map of responses to fluoxetine correlating with thoscortical, pattern is observed, including thalamic and striatal regions, as well as ventto other brain regions identified in this map overlapped tightly with this distribufunctionally connected in their response to fluoxetine challenge.

correlation maps referenced to target structures in the forebraininvoked more widespread sets of functionally connected brainregions, including the broad cortical areas involved in the responseto both drugs. These findings suggest that different networks of brainstructures may be involved in the acute response to these drugs; thecortical responses observed in each case may be due to activitydownstream from the primary dopamine and serotonin neurotrans-mitter circuits. Taken together, these results demonstrate that an

e in the medial amygdaloid nucleus (MeA). A robust, predominantly sub-ral/posterior hippocampus and the raphe nuclei. Correlation maps referencedtion, suggesting that these structures comprise a network of brain regions

1634 A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

inter-subject correlation analysis of the central rCBV responseamplitude can resolve different circuits involving brain regionsfunctionally related in their response to pharmacological challenge.

Comparison with other approaches to functional connectivity

Examination of correlated responses in brain imaging data haspreviously been used in both animal and man as an index ofconnectivity in sensory/cognitive (Peterson et al., 1999; Rissmanet al., 2004; Menon and Levitin, 2005; Summerfield et al., 2006) orlesion (Tarrasch et al., 2005) paradigms and also for investigationsof connectivity in the resting state (Biswal et al., 1995; Grady et al.,2001; Hampson et al., 2002; Fox et al., 2005; Liang et al., 2006).We have extended this approach to probe functionally connectedresponses in the pharmacologically activated brain using an inter-subject method, akin to techniques employed in 2-deoxyglucose(Soncrant et al., 1986) and PET (Horwitz et al., 1998). In MRI,most previous connectivity analyses have sought to use therelatively high temporal resolution possible with fMRI time seriesacquisition methods and have considered intra-subject temporalcorrelations. However, signal changes following pharmacologicalchallenge typically comprise a slow monophasic profile, lastingtens of minutes or more. In the present data many brain regionsresponded to the drug and so the intra-subject correlations wereoften non-specific; differences in the correlation maps acrosssubjects then make the generation of a meaningful representativegroup pattern difficult. Temporal correlations within subjects mayalso be sensitive to differences in the shape and amplitude of theresponse profile between brain regions (Marota et al., 2000;Schwarz et al., 2004b) and may be more clearly interpreted in thoseterms. For these reasons, our method employed inter-subjectcorrelations in response amplitude, harnessing the variability inspatial response profile.

In the neuroimaging literature, analyses based on correlatedsignal changes are considered indicative of “functional connectiv-ity” or “effective connectivity” depending upon the degree of apriori anatomical and physiological information invoked (Friston,1994). The former category refers to approaches such as themethod presented in the current paper, where correlated responsesin different brain regions are interpreted as reflecting someunderlying functional connection between these regions. Effectiveconnectivity methods include path analysis (or structural equationmodeling) in which ‘known’ anatomical connections are used tospecify an explicit circuit model (Honey et al., 2003; Stephan,2004; Penny et al., 2004; Schlosser et al., 2006). In bothapproaches, modulation of the connectivity relationships may beelicited by pharmacological treatment. Similarly, modified func-tional relationships may be investigated using the methodpresented here, where the challenge serves as a probe to examinethe treatment effect.

The approach presented here explicitly uses differences in thespatial profile of response across subjects. Group sizes must besuch that the cohort samples a sufficient range of profiles to enablea robust definition of correlation relationships. Clearly, the requiredgroup size for adequate statistical power is a function of thevariability in the spatial profile and the strength of the underlyingcorrelations. In the present data, the fluoxetine cohort in particularwas relatively small. Nevertheless, spatial patterns correspondingto known neurotransmitter pathways were identified in both D-amphetamine and fluoxetine groups, providing confidence that thismethod is able to identify meaningful functional connections

related to effects on the neurotransmitter systems targeted by thedrugs. Although phMRI signal changes can represent a surrogatefor changes in neuronal activity, drugs may also affect the centralhaemodynamic response via peripheral changes in blood pressure,direct action on the vasculature (Choi et al., 2006) or modifyneurovascular coupling (Gsell et al., 2006). Such effects might beexpected to be uncorrelated with action more directly related toneuronal activity (for example, the relatively focal patternsidentified in the present analysis as corresponding to projectionsfrom the VTA and the raphe). However, despite the presence ofwidespread correlation patterns in the data, the absence of anysignificant correlation with peripheral blood pressure changesargues in favor of a central origin for the different patterns offunctional connectivity observed. Other neurotransmitter systemsmay be involved, either via direct effects of the drug at theirassociated receptors or cross-talk following binding at the targetreceptor (see, e.g., Reith et al., 1997). The precise mechanismsunderlying these ‘downstream’ effects remain to be elucidated.

In this paper we have focused on following correlation patternsstarting from known anatomical projections. However, this can beperformed for any brain structure of interest and extended todescribe a correlation structure connecting all pixels. To make theresulting data more tractable, the image data may be parcellatedinto a smaller set of brain structures (Achard et al., 2006). Analysisof brain connectivity networks so derived is an ongoing field ofresearch and applications to phMRI promise to expand further therich body of information available from such data sets. In additionto providing a conceptually different insight compared withstandard univariate group comparisons, this method can alsoprovide information relating to brain structures in which thevariability in response (relative to its mean) precludes detectionwith statistical tests designed to detect group mean differences.

Inter-subject correlations and variability

In this method, variability in the spatial profile of the phMRIresponse is harnessed to resolve functionally related brain regions.Indeed, this represents an important difference compared withcalculating intra-subject temporal correlations on phMRI timeseries, where this inter-subject variability reduces the sensitivity.This inter-subject variability was present in the current data despitea consistent challenge dose and strict control of animal physiology.The variability observed in the spatial rCBV response profile maybe related to the pre-synaptic action of these drugs in neuro-transmitter release and re-uptake blockade, mechanisms sensitiveto the basal tone. The inter-subject variability in our rCBVresponse amplitudes in the VOIs investigated is in fact similar tothat reported for dopamine levels in microdialysis measurements(Melega et al., 1995), although such point-sampled modalitiesrarely sample more than one brain region at a time and henceprovide no information on the spatial variability in neurotransmit-ter concentrations.

Conclusion

We have demonstrated that an analysis of correlated signalchange amplitudes across subjects provides a simple and effectivemethod for in vivo mapping of functional connectivity betweendifferent brain structures recruited in response to pharmacologicalchallenge. Analysis of the phMRI responses to D-amphetamine andfluoxetine revealed a rich structure of functionally connected brain

1635A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

regions that closely reflect known pathways in the neurotransmittersystems targeted by these drugs, providing a functional connectiv-ity fingerprint of their central activity. This correspondenceprovides an important corroboration of the hypothesis that thehemodynamic responses observed are closely related to drug-specific changes in neuronal activity. By resolving functionalrelationships in the pharmacologically activated brain, thisapproach greatly extends the information available from phMRIstudies and promises to be useful in elucidating central mechan-isms underlying pharmacological action.

References

Achard, S., Salvador, R., Whitcher, B., Suckling, J., Bullmore, E., 2006. Aresilient, low-frequency, small-world human brain functional networkwith highly connected association cortical hubs. J. Neurosci. 26, 63–72.

Beckmann, C.F., Jenkinson, M., Smith, S.M., 2003. General multilevellinear modeling for group analysis in fMRI. NeuroImage 20,1052–1063.

Beckstead, R.M., 1976. Convergent thalamic and mesencephalic projectionsto the anterior medial cortex in the rat. J. Comp. Neurol. 166, 403–416.

Biswal, B., Yetkin, F.Z., Haughton, V.M., Hyde, J.S., 1995. Functionalconnectivity in the motor cortex of resting human brain using echo-planar MRI. Magn. Reson. Med. 34, 537–541.

Chen, Y.C., Choi, J.K., Andersen, S.L., Rosen, B.R., Jenkins, B.G., 2005.Mapping dopamine D2/D3 receptor function using pharmacologicalmagnetic resonance imaging. Psychopharmacology (Berlin) 180,705–715.

Choi, J.K., Chen, Y.I., Hamel, E., Jenkins, B.G., 2006. Brain hemodynamicchanges mediated by dopamine receptors: role of the cerebralmicrovasculature in dopamine-mediated neurovascular coupling. Neuro-Image 30, 700–712.

Dixon, A.L., Prior, M., Morris, P.M., Shah, Y.B., Joseph, M.H., Young,A.M.J., 2005. Dopamine antagonist modulation of amphetamineresponse as detected using pharmacological MRI. Neuropharmacology48, 236–245.

Drevets, W.C., 2000. Neuroimaging studies of mood disorders. Biol.Psychiatry 48, 813–829.

Everitt, B.J., Robbins, T.W., 2005. Neural systems of reinforcement for drugaddiction: from actions to habits to compulsion. Nat. Neurosci. 8,1481–1489.

Fox, M.D., Snyder, A.Z., Vincent, J.L., Corbetta, M., Van Essen, D.C.,Raichle, M.E., 2005. The human brain is intrinsically organized intodynamic, anticorrelated functional networks. Proc. Natl. Acad. Sci.U. S. A. 102, 9673–9678.

Friston, K.J., 1994. Functional and effective connectivity in neuroimaging: asynthesis. Hum. Brain Mapp. 2, 56–78.

Friston, K., Worsley, K.J., Frackowiak, R.S., Mazziotta, J., Evans, A.C.,1994. Assessing the significance of focal activations using their spatialextent. Hum. Brain Mapp. 1, 210–220.

Gozzi, A., Schwarz, A.J., Reese, T., Bertani, S., Crestan, V., Bifone, A.,2006. Region-specific effects of nicotine on brain activity: a pharma-cological MRI study in the drug-naïve rat. Neuropsychopharmacology31, 1690–1703.

Grady, C.L., Furey, M.L., Pietrini, P., Horwitz, B., Rapoport, S.I., 2001.Altered brain functional connectivity and impaired short-term memoryin Alzheimer's disease. Brain 124, 739–756.

Gsell, W., Burke, M., Wiedermann, D., Bonvento, G., Silva, A.C., Dauphin,F., Buhrle, C., Hoehn, M., Schwindt, W., 2006. Differential effects ofNMDA and AMPA glutamate receptors on functional magneticresonance imaging signals and evoked neuronal activity during forepawstimulation of the rat. J. Neurosci. 26, 8409–8416.

Hampson, M., Peterson, B.S., Skudlarski, P., Gatenby, J.C., Gore, J.C.,2002. Detection of functional connectivity using temporal correlations inMR images. Hum. Brain Mapp. 15, 247–262.

Hennig, J., Nauerth, A., Friedburg, H., 1986. RARE imaging: a fast imagingmethod for clinical MR. Magn. Reson. Med. 3, 823–833.

Honey, G.D., Suckling, J., Zelaya, F., Long, C., Routledge, C., Jackson, S.,Ng, V., Fletcher, P.C., Williams, S.C., Brown, J., Bullmore, E.T., 2003.Dopaminergic drug effects on physiological connectivity in a humancortico-striato-thalamic system. Brain 126, 1767–1781.

Horwitz, B., Rumsey, J.M., Donohue, B.C., 1998. Functional connectivityof the angular gyrus in normal reading and dyslexia. Proc. Natl. Acad.Sci. U. S. A. 95, 8939–8944.

Jenkins, B.G., Chen, Y.-C.I., Mandeville, J.B., 2003. Pharmacologicalmagnetic resonance imaging (phMRI). In: van Bruggen, N., Roberts, T.(Eds.), Biomedical Imaging in Experimental Neuroscience. CRC Press,New York, pp. 155–209.

Kalisch, R., Salome, N., Platzer, S., Wigger, A., Czisch, M., Sommer, W.,Singewald, N., Heilig, M., Berthele, A., Holsboer, F., Landgraf, R.,Auer, D.P., 2004. High trait anxiety and hyporeactivity to stress of thedorsomedial prefrontal cortex: a combined phMRI and Fos study in rats.NeuroImage 23, 382–391.

Leslie, R.A., James, M.F., 2000. Pharmacological magnetic resonanceimaging: a new application for functional MRI. Trends Pharmacol. Sci.21, 314–318.

Liang, M., Zhou, Y., Jiang, T., Liu, Z., Tian, L., Liu, H., Hao, Y., 2006.Widespread functional disconnectivity in schizophrenia with resting-state functional magnetic resonance imaging. NeuroReport 17, 209–213.

Mandeville, J.B., Marota, J.J.A., Kosofsky, B.E., Keltner, J.R., Weissleder,R., Rosen, B., Weisskoff, R., 1998. Dynamic functional imaging ofrelative cerebral blood volume during rat forepaw stimulation. Magn.Reson. Med. 39, 615–624.

Marota, J.J.A., Mandeville, J.B., Weisskoff, R., Moskowitz, M.A., Rosen,B., Kosofsky, B.E., 2000. Cocaine activation discriminates dopaminer-gic projections by temporal response: an fMRI study in rat. NeuroImage11, 13–23.

Melega, W.P., Williams, A.E., Schmitz, D.A., DiStefano, E.W., Cho, A.K.,1995. Pharmacokinetic and pharmacodynamic analysis of the actions ofD-amphetamine and D-methamphetamine on the dopamine terminal.J. Pharmacol. Exp. Ther. 274, 90–96.

Menon, V., Levitin, D.J., 2005. The rewards of music listening: response andphysiological connectivity of the mesolimbic system. NeuroImage 28,175–184.

Moore, R.Y., Halaris, A.E., Jones, B.E., 1978. Serotonin neurons of themidbrain raphe: ascending projections. J. Comp. Neurol. 180, 417–438.

Oades, R.D., Halliday, G.M., 1987. Ventral tegmental (A10) system:neurobiology: 1. Anatomy and connectivity. Brain Res. 434, 117–165.

Paxinos, G., Watson, C., 1998. The Rat Brain in Stereotaxic Coordinates,4th ed. Academic Press, San Diego.

Penny, W.D., Stephan, K.E., Mechelli, A., Friston, K.J., 2004. Modellingfunctional integration: a comparison of structural equation and dynamiccausal models. NeuroImage 23 (Suppl. 1), S264–S274.

Peterson, B.S., Skudlarski, P., Gatenby, J.C., Zhang, H., Anderson, A.W.,Gore, J.C., 1999. An fMRI study of Stroop word-color interference:evidence for cingulate subregions subserving multiple distributedattentional systems. Biol. Psychiatry 45, 1237–1258.

Reese, T., Bjelke, B., Porszasz, R., Baumann, D., Bochelen, D., Sauter, A.,Rudin, M., 2000. Regional brain activation by bicuculline visualized byfunctional magnetic resonance imaging. Time-resolved assessment ofbicuculline-induced changes in local cerebral blood volume using anintravascular contrast agent. NMR Biomed. 13, 43–49.

Reith, M.E., Li, M.Y., Yan, Q.S., 1997. Extracellular dopamine, norepi-nephrine, and serotonin in the ventral tegmental area and nucleusaccumbens of freely moving rats during intracerebral dialysis followingsystemic administration of cocaine and other uptake blockers.Psychopharmacology (Berlin) 134, 309–317.

Rissman, J., Gazzaley, A., D'Esposito, M., 2004. Measuring functionalconnectivity during distinct stages of a cognitive task. NeuroImage 23,752–763.

Salome, N., Stemmelin, J., Cohen, C., Griebel, G., 2006. Differential roles ofamygdaloid nuclei in the anxiolytic- and antidepressant-like effects of

1636 A.J. Schwarz et al. / NeuroImage 34 (2007) 1627–1636

the V1b receptor antagonist, SSR149415, in rats. Psychopharmacology(Berlin) 187, 237–244.

Schlosser, R.G., Wagner, G., Sauer, H., 2006. Assessing the workingmemory network: studies with functional magnetic resonance imagingand structural equation modeling. Neuroscience 139, 91–103.

Schwarz, A.J., Reese, T., Gozzi, A., Bifone, A., 2003. Functional MRI usingintravascular contrast agents: detrending of the relative cerebrovascular(rCBV) time course. Magn. Reson. Imaging 21, 1191–1200.

Schwarz, A., Gozzi, A., Reese, T., Bertani, S., Crestan, V., Hagan, J.,Heidbreder, C., Bifone, A., 2004a. Selective dopamine D(3) receptorantagonist SB-277011-A potentiates phMRI response to acute amphe-tamine challenge in the rat brain. Synapse 54, 1–10.

Schwarz, A.J., Zocchi, A., Reese, T., Gozzi, A., Garzotti, M., Varnier, G.,Curcuruto, O., Sartori, I., Girlanda, E., Biscaro, B., Crestan, V., Bertani,S., Heidbreder, C., Bifone, A., 2004b. Concurrent pharmacological MRIand in situ microdialysis of cocaine reveal a complex relationshipbetween the central hemodynamic response and local dopamineconcentration. NeuroImage 23, 296–304.

Schwarz, A.J., Danckaert, A., Reese, T., Gozzi, A., Paxinos, G., Watson, C.,Merlo-Pich, E.V., Bifone, A., 2006. A stereotaxic MRI template set forthe rat brain with tissue class distribution maps and co-registeredanatomical atlas: application to pharmacological MRI. NeuroImage 32,538–550.

Schwarz, A.J., Whitcher, B., Gozzi, A., Reese, T., Bifone, A., in press.Study-level Wavelet Cluster Analysis and data-driven signal models inpharmacological MRI. J. Neurosci. Methods (doi:10.1016/j.jneumeth.2006.07.017).

Shah, Y.B., Prior, M.J., Dixon, A.L., Morris, P.G., Marsden, C.A., 2004.Detection of cannabinoid agonist evoked increase in BOLD contrast inrats using functional magnetic resonance imaging. Neuropharmacology46, 379–387.

Slattery, D.A., Morrow, J.A., Hudson, A.L., Hill, D.R., Nutt, D.J., Henry, B.,2005. Comparison of alterations in c-fos and Egr-1 (zif268) expressionthroughout the rat brain following acute administration of differentclasses of antidepressant compounds. Neuropsychopharmacology 30,1278–1287.

Soncrant, T.T., Horwitz, B., Holloway, H.W., Rapoport, S.I., 1986. Thepattern of functional coupling of brain regions in the awake rat. BrainRes. 369, 1–11.

Stephan, K.E., 2004. On the role of general system theory for functionalneuroimaging. J. Anat. 205, 443–470.

Summerfield, C., Greene, M., Wager, T., Egner, T., Hirsch, J., Mangels, J.,2006. Neocortical connectivity during episodic memory formation.PLoS. Biol. 4, e128.

Tarrasch, R., Goelman, G., Daphna, J., Weiner, I., 2005. Long-term functionalconsequences of quinolinic acid striatal lesions and their alterationfollowing an addition of a globus pallidus lesion assessed usingpharmacologicalmagnetic resonance imaging. Exp.Neurol. 196, 244–253.

Woolrich, M.W., Behrens, T.E., Beckmann, C.F., Jenkinson, M., Smith,S.M., 2004. Multilevel linear modelling for fMRI group analysisusing Bayesian inference. NeuroImage 21, 1732–1747.

Worsley, K.J., Evans, A.C., Marrett, S., Neelin, P., 1992. A three-dimensional statistical analysis for CBF activation studies in humanbrain. J. Cereb. Blood Flow Metab. 12, 900–918.

Worsley, K.J., Chen, J.I., Lerch, J., Evans, A.C., 2005. Comparing functionalconnectivity via thresholding correlations and singular value decom-position. Philos. Trans. R. Soc. Lond., B Biol. Sci. 360, 913–920.

Zaharchuk, G., Mandeville, J.B., Bogdanov Jr., A.A., Weissleder, R., Rosen,B.R., Marota, J.J., 1999. Cerebrovascular dynamics of autoregulationand hypoperfusion. An MRI study of CBF and changes in total andmicrovascular cerebral blood volume during hemorrhagic hypotension.Stroke 30, 2197–2204.