YNIMG-11797; No. of pages: 13; 4C: 3, 4, 6, 7, 8 NeuroImage xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

NeuroImage journal homepage: www.elsevier.com/locate/ynimg

F

Chris Tailby a,b,⁎, Richard A.J. Masterton a,c, Jenny Y. Huang c, Graeme D. Jackson a,c,1, David F. Abbott a,c,1,⁎

4 5 6

a

7

a r t i c l e

8 17 9 10

Article history: Accepted 18 November 2014 Available online xxxx

11 12 Q3 13 14 15 16

Keywords: Functional MRI Brain networks Functional connectivity BOLD Resting state

The Florey Institute of Neuroscience and Mental Health, The University of Melbourne, Austin Hospital, Victoria, Australia Department of Psychology, The University of Melbourne, Victoria, Australia Department of Medicine, The University of Melbourne, Victoria, Australia

i n f o

a b s t r a c t

Resting state functional connectivity (rFC) is used to identify functionally related brain areas without requiring subjects to perform specific tasks. Previous work suggests that prior brain state, as determined by the activity engaged in immediately prior to collection of resting state data, can influence the networks recovered by rFC analyses. We determined the prevalence and network specificity of rFC changes induced by manipulations of prior state (including an unstructured (unconstrained) state, and language and motor tasks). Three blocks of rest data (one after each of the specified prior states) were acquired on each of 25 subjects. We hypothesised that prior state induced changes in rFC would be greatest within the networks most actively recruited by that prior state. Changes in rFC were greatest following the motor task and, contrary to our hypothesis, were not network specific. This was demonstrated by comparing (1) the timecourses within a set of ROIs selected on the basis of taskrelated de/activation, and (2) seed-based whole brain voxel-wise connectivity maps, seeded from local maxima in the task-related de/activation maps. Changes in connectivity strength tended to manifest as increases in rFC relative to that in the unstructured rest state, with change maps resembling partially complete maps of the primary sensory cortices and the cognitive control network. The majority of rFC changes occurred in areas moderately (but not weakly) connected to the seeds. Constrained prior states were associated with lower acrossparticipant variance in rFC. This systematic investigation of the effect of prior brain state on rFC indicates that the rFC changes induced by prior brain state occur both in brain networks related to that brain activity and in networks nominally unrelated to that brain activity. © 2014 Published by Elsevier Inc.

D

R

R

E

C

T

c

E

b

O

Q23Q1

R O

2

Resting state functional connectivity changes induced by prior brain state are not network specific

P

1

40 38 37 39

Introduction

42

Resting state functional connectivity (rFC) has assumed a prominent position in the investigation of large scale neural networks in the human brain (Bandettini, 2009; Fox and Raichle, 2007; Greicius, 2008; van den Heuvel and Pol, 2010; Zuo et al., 2010). The networks revealed by rFC resemble those identified via task-related activation studies (Biswal et al., 1995). Further, rFC analyses are appealing as the data are comparatively easily acquired, and they can be performed even in clinical populations in whom task execution is compromised. It is now apparent that rFC maps are not fixed, stable entities but rather exhibit variation across a variety of timescales, from seconds to

45 46 47 48 49 50 51

U

43 44

N C O

41

⁎ Corresponding authors at: The Florey Institute of Neuroscience and Mental Health, Melbourne Brain Centre, 245 Burgundy Street Heidelberg, Victoria 3084, Australia. Fax: +61 3 9035 7307. E-mail addresses: [email protected] (C. Tailby), [email protected] (R.A.J. Masterton), [email protected] (J.Y. Huang), [email protected] (G.D. Jackson), david.abbott@florey.edu.au (D.F. Abbott). 1 Joint senior authors.

minutes to days (Chang and Glover, 2010; Guo et al., 2012; Hasson et al., 2009; Kang et al., 2011; Mannfolk et al., 2011; Shehzad et al., 2009; Soares et al., 2013; Stevens et al., 2010; Wang et al., 2012; Zuo et al., 2010). In the present study we focus on rFC changes occurring over periods of minutes. Prior studies examining changes in connectivity at this time scale have compared rFC before and after some manipulation, such as execution of a motor task (Duff et al., 2008; Peltier et al., 2005), or a cognitive task such as language (Waites et al., 2005) or working memory task (Gordon et al., 2014). Such studies converge on evidence that prior brain state can influence subsequent rFC, with the changes hypothesised to reflect factors such as fatigue (Esposito et al., 2014; Peltier et al., 2005), changes in cognitive set (Waites et al., 2005) and/or learning/consolidation (Gordon et al., 2014). The potential influence of prior brain state on rs-fcMRI has important ramifications for group studies. Many research centres uniformly collect rest data across otherwise different experimental protocols, and there is a strong attraction to pool such data in order to increase power. The potential bias introduced by such pooling could also influence analyses based upon large, multicentre data sharing initiatives,

http://dx.doi.org/10.1016/j.neuroimage.2014.11.037 1053-8119/© 2014 Published by Elsevier Inc.

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36

52 53 54 55 56 57 58 59 Q4 60 61 62 63 64 65 66 67 68 69 70

106 107 108

Twenty-five healthy volunteers participated in the study (17 male; age, mean ± SD: 24.6 ± 5.5 years, range: 17–40). All protocols were approved by the relevant institutional Human Research Ethics Committee.

109

In-scanner procedures and cognitive activation paradigms

110

Subjects were scanned continuously for 450 volumes, alternating between periods of “extended rest” (90 volumes) and block design “task” periods (90 volumes) according to the following sequence: rest1, task1, rest2, task2, and rest3. During the extended rest periods, subjects viewed a black screen and were instructed to stay awake with eyes open, and refrain from any overt or covert cognitive or motor activities. During the task periods, subjects performed one of two block design tasks: a language task — Orthographic Lexical Retrieval (OLR), and a motor task — finger tapping (MOTOR); task order was counter-balanced across subjects. Both block design tasks alternated between 10 TRs of active phase and 10 TRs of baseline phase, completing four active phases embedded within five baseline phases. During the baseline phases of both tasks, subjects viewed a black screen with a white cross (“+”) at the centre, and were instructed to relax. During the active phase of the OLR task (Wood et al., 2001), a covert adaptation of the Controlled Oral Word Association Test (Strauss et al., 2006), a letter was displayed at the centre of the screen, and then after five TRs another letter was presented. Participants were instructed to think of as many words as possible beginning with the current letter, but to avoid using proper nouns or numbers, repeating words or adding a suffix to a previously retrieved word. During the active phase of the MOTOR task, the word “Move” was presented at the centre of the screen, and

96 97 98 99 100 101

111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131

C

94 95

E

92 93

R

90 91

R

88 89

O

86 87

C

84 85

N

82 83

U

80 81

Image acquisition

142

The fMRI studies were carried out with a 3 T GE Signa LX whole body scanner (General Electric, Milwaukee, WI), using a standard birdcage quadrature head coil. Functional images were acquired as a series of gradient-recalled echo planar imaging (GR-EPI) volumes (TE = 40 ms). Images for the first 13 participants were acquired using a TR of 3.6 s at a voxel resolution of 1.95 mm × 1.95 mm × (4 mm thick +1 mm gap) (25 oblique slices); images for the final 12 participants were acquired using a TR of 3.2 s at a voxel resolution of 3.44 mm × 3.44 mm (3.2 mm thick + 0.2 mm gap) (40 oblique slices). The data from the two different scanners therefore contained the same number of image volumes, corresponding to slightly different total experiment durations. Due to a technical error the initial rest period for one participant contained 50 rather than 90 volumes.

143 144

F

Participants

78 79

O

105

77

R O

Material and methods

75 76

132 133

P

104

73 74

subjects were required to tap their left index finger in time with a 1.0 Hz metronome played to them over headphones. The metronome was also played throughout the baseline period of the MOTOR task, during which the words “Don't Move” were presented. We refer to the resting state data collected during the initial period as unstructured, and that collected after the OLR and MOTOR tasks as post-OLR and post-MOTOR, respectively. Unstructured refers to the fact that, relative to the post-OLR and post-MOTOR rest periods, the prior brain state in the unstructured rest is not as tightly constrained across participants.

134 135 136 137 138 139 140 141

145 146 147 148 149 150 151 152 153 154 155

Image processing

156

The collected images were pre-processed using Statistical Parametric Mapping software (SPM8 release 4667; Wellcome Department of Imaging Neuroscience, London, UK) with the aid of the iBrain™ analysis toolbox for SPM (Abbott et al., 2011) and iBrain™ (Abbott and Jackson, 2001). Images were first slice-time corrected, realigned, then spatially normalised to an in-house EPI template (constructed from 30 healthy control brains not including the present participants, as described in detail in Waites et al. (2005)) that approximates the SPM standard space (Montreal Neurological Institute). Normalised images were written out at 2 × 2 × 2 mm resolution, then smoothed with an isotropic Gaussian kernel (full-width-at-half-maximum = 8 mm).

157

Analysis of activation paradigms

168

Statistical analysis of the functional imaging data was conducted in SPM8 with the aid of the iBrain™ analysis toolbox for SPM using a general linear model. Standard single subject analyses were conducted on each participant's OLR and MOTOR tasks. The BOLD response of the task compared to baseline state was modelled assuming the SPM canonical hemodynamic response function (HRF), and comprised the effect of interest. In addition, the six rigid body transformation parameters estimated during image realignment were included in the model as effects of no interest. Prior to estimation, the fMRI data and design matrix were high-pass filtered (cut-off = 128 s) and pre-whitened using a firstorder autoregressive process (Friston et al., 2002). Session specific grand mean scaling was used. From these analyses we used contrasts of parameter estimates of task against baseline (OLR-baseline and MOTOR-baseline) as inputs to group level one-sample t-tests of the OLR and MOTOR tasks.

169

Seed selection

184

For analyses of rFC we selected seeds on the basis of task-related activation on the OLR and MOTOR paradigms. We adopted this approach for consistency with our previous published work, examining prior brain state effects on functional connectivity in the language system (Waites et al., 2006). Specifically, we defined five motor and five

185 186

T

102 103

such as the 1000 Functional Connectomes Project (www.nitrc.org/ projects/fcon_1000/). These hypothesised mechanisms by which prior brain state influences rFC, outlined above, suggest that any rFC changes should be network specific. For instance, if fatigue lies behind the rFC changes observed following a finger movement task (Peltier et al., 2005), then one might hypothesise that such changes are restricted to the motor system. The question of the specificity of these effects remains unclear. Gordon et al. (2014) examined connectivity within and between the task positive network (TPN) and the default mode network (DMN) immediately before and after execution of a working memory task. They observed alterations of rFC both within the TPN, and between the TPN and DMN. This result is similar to the earlier work of Grigg and Grady (2010), who showed variable connectivity from precuneus (in the DMN) to a set of brain regions resembling the TPN and primary sensory cortices when comparing rest data before and after a period of task execution. These data suggest that both intra- and inter-network changes in rFC can be induced by prior brain state. However comparisons relative to the DMN may constitute a special case of internetwork change as the DMN has frequently been conceptualised as diametrically opposed to other brain networks, particularly the TPN. Here we address the question of the network specificity of rFC changes induced by immediately prior brain state using a novel experimental design. We collected three sets of rest data: an initial rest period acquired upon entering the scanner, a second rest period following execution of an in-scanner language task, and a third rest period following execution of an in-scanner motor task (with order of the language and motor tasks counter-balanced). For each rest block we calculated seeded rFC analyses, with seeds located in the language network, the motor network, and the default mode network. This design enabled us to evaluate two hypotheses: (1) that systematic variation of prior brain state results in systematic group level alterations in rFC; and (2) that alterations in rFC induced by prior brain state exhibit network specificity.

D

71 72

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

E

2

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

158 159 160 161 162 163 164 165 166 167

170 171 172 173 174 175 176 177 178 179 180 181 182 183

187 188 189

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

202 203 204 Q5 205 206 207 208 209 210 211 212 213 214

Functional connectivity analyses

216

230 231

We performed two types of connectivity analyses, each based upon seed timecourses defined as the average timecourse within 5 mm spheres (corresponding to 81 voxel ROIs) centred on each of the coordinates listed in Table 1. In the first analysis, which we refer to as ROI-toROI connectivity, for each participant we calculated the Pearson correlation coefficient between all possible ROI timecourse pairs, yielding a 10 × 10 correlation matrix for each participant (see Fig. 3). In the second analysis, which we refer to as connectivity maps, for each participant we created separate voxel-wise connectivity maps (covering all within brain voxels) for each seed listed in Table 1 (resulting in 10 connectivity maps per participant). These were created by calculating the Pearson correlation coefficient between the seed timecourse and the ith voxel's timecourse and storing this value in the ith voxel. The r values in these two different analyses (ROI-to-ROI, connectivity maps) were converted to z values using Fisher's r-to-z transform before performing statistical analyses.

t1:1 t1:2

Table 1 Coordinates of the language and motor network seeds used in the rFC analyses.

226 227 228 229

t1:3

C

E

Network/brain region

t1:4 t1:5 t1:6 t1:7 t1:8 t1:9 t1:10 t1:11 t1:12 t1:13 t1:14 t1:15 t1:16

Fig. 1. Methods. A total of 450 volumes were acquired in each participant, alternating between 90 volumes of rest and 90 volumes of task. Both tasks – OLR and MOTOR – were of block design, alternating between baseline (10 volumes) and active (10 volumes) conditions within the 90 volumes of the task period. Task order (OLR, MOTOR) was counter-balanced across participants.

Statistical analysis of changes in ROI-to-ROI connectivity We examined how connectivity within and between networks (language, motor, default mode) was affected by prior brain state by looking at changes in ROI-to-ROI connectivity. Specifically, we divided the correlation matrix described above into regions representing connectivity among the language seeds (language–language), regions representing connectivity among the motor seeds (motor–motor), regions representing connectivity between the language and motor

OLR

MOTOR

+64

+64

+44

+56

+20

+44

+0

+32

R

224 225

R

223

N C O

221 222

U

219 220

90

90

90

T

215

217 218

10 10 10 10 10 10 10 10 10

90

Rest 3

F

Analyses of rFC were conducted separately on each of the 90 volume (~5 min) blocks of rest data (rest1, rest2, rest3). Correlation-based estimates of connection strength have been shown to stabilise after approximately 5 min of BOLD signal acquisition (Van Dijk et al., 2010; Whitlow et al., 2011). Following preprocessing as described above (Section Image processing), additional estimates of physiological noise were removed from the timecourse in each voxel prior to calculating functional connectivity. Data were corrected via linear regression against the estimated rigid body realignment parameters, white matter signals, and ventricular system signals (the average whole brain signal was not used in the estimation of physiological noise). In-scanner head movements in excess of 0.5 mm were corrected via linear regression against stick functions placed at the time of (and the two subsequent time points following) the movement (Lemieux et al., 2007). Data were then bandpass filtered via linear regression against discrete cosine functions to exclude frequencies outside the range 0.01–0.8 Hz.

200 201

10 10 10 10 10 10 10 10 10

Task 2

O

199

196

Rest 2

R O

Additional preprocessing for functional connectivity analyses

194 195

Vols: 90

Task 1

P

198

192 193

Rest 1

E

197

language network seeds on the basis of de/activation peaks in the clusters returned by the group level SPM analyses of the task activation data. This resulted in the selection of three seeds in regions of activation and two seeds in regions of deactivation in the MOTOR task. From the OLR task this procedure resulted in the selection of four activation based seeds, and one deactivation based seed (close to the language network seeds we used in our previously published work (Waites et al., 2006)). Seed coordinates are reported in Table 1.

D

190 191

3

Language Left middle frontal gyrus Left inferior frontal gyrus Paracingulate gyrus Left precentral gyrus Precuneus — language Motor Right postcentral gyrus — inferior Right postcentral gyrus — superior Supplementary motor cortex Left postcentral gyrus Precuneus — motor

MNI coordinates x

y

z

−52 −34 0 −52 12

12 24 0 −6 −68

20 0 68 48 18

54 40 −4 −40 4

−20 −28 −8 −28 −54

46 56 62 48 30

−8.00

8.00

−0.00

0.00

t-value Fig. 2. Task-related activation on the language (OLR) task (left) and the motor (MOTOR) task (right). Peaks in the de/activation images, used to define seeds/ROIs, are shown as black dots. Left hemisphere is shown on the right (radiological convention).

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

232 233 234 235 236 237 238 239

4

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

post-OLR

post-MOTOR

F

O L O R: O LR L M O LR : L F M LR : G M OT P IF : O TO OR OL L P aC G rC G R : R R: M M : R P Pc G o O O TO TO Po CG un R R: CG inf :L S s Po MA up C G

G IF G : L aC G LR : P rC n O P cu LR O R: L : P G inf L LR oC up O Gs P O A C M :R o R R P : S CG n R TO : Po cu R O TO M TO O R: L : P R M O M TO TO O M MO

unstructured

R O

O

-1.0 0.0 1.0 z transformed r-value

0.8 0.7

unstructured

P

post−MOTOR

D

0.5

post−OLR

E

0.4

T

0.3 0.2

C

Average r−to−z value

0.6

E

0.1

R

0 −0.1

Motor−Motor

Lang−Motor

Lang−DMN

Motor−DMN

R

Lang−Lang

242 243 244 245 246 247 248 Q6 249 250 251 252 253 254 255

seeds (language–motor), regions representing connectivity between the language and default mode seeds (language–DMN), and regions representing connectivity between the motor and default mode seeds (motor–DMN). For each participant we calculated the average r-to-z values within these regions, and compared these average values across the different rest states using two-way repeated measures ANCOVA (factor 1: network pairing, five levels; factor 2: rest state, three levels). The model also included two covariates: (i) image acquisition protocol (TR and voxel size, see Section Image acquisition above) was coded as a binary variable, and (ii) temporal order of the second and third rest periods was dummy coded (1 and −1, with the unstructured rest coded as 0). This approach enabled us to ask whether there was a main effect of prior brain state on resting state connectivity, and whether there was a significant interaction (indicating that the effect of prior state differs between networks). We did not examine for a main effect of seed location as such an effect was expected a priori and considered to be of no

U

240 241

N

C

O

Fig. 3. ROI based comparison of intra- and inter-network connectivity as a function of prior brain state. The panel at top left shows the segmentation of the 10 × 10 correlation matrix into inter- and intra-network connectivity regions (see main text). Only those entries below the diagonal are displayed. The three rightmost images in the top row show the average ROI-to-ROI connectivity matrices (across participants) in the unstructured, post-OLR and post-MOTOR rest states; again, only those entries below the main diagonal are displayed. The bar plot in the lower panel shows how inter- and intra-network connectivity changes between the different rest states (error bars report two standard errors of the mean). Average network–network connectivity changes, when they occur, tend to occur uniformly across all combinations of network pairings (i.e. post-motor greater than post-OLR greater than unstructured). There is no interaction between network pairing and prior state.

interest (i.e. a main effect of seed location would simply tell us that there was a significant difference between two or more of the connectivity maps derived from the different seeds, which we know a priori to be the case). Significant F-tests (p b 0.05) were followed up with pairwise comparisons (Tukey's least significant difference, p b 0.05).

256 257

Statistical analysis of changes in voxel-wise functional connectivity maps We performed group level voxel-wise analyses using as inputs the r-to-z transformed FC maps calculated for each seed location in each resting state in each participant. We first performed a two-way repeated measures ANCOVA, implemented using the Flexible factorial option within the Factorial design module of SPM8. We specified the following factors: subject, prior brain state (three levels: no-prior task, post-OLR, post-MOTOR) and seed location (10 levels: see Table 1). Image acquisition protocol and temporal order of the prior states were included as covariates, as described above. From this model we

261

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

258 259 260

262 263 264 265 266 267 268 269 270

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

294 295

300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323

FC max

325 326 327 328

Results

T

292 293

C

290 291

E

288 289

R

286 287

R

284 285

N C O

282 283

U

280 281

F

We also performed a set of exploratory analyses, aimed at characterising how frequently prior brain state dependent changes in FC would arise in analyses based upon comparisons of voxel-wise connectivity maps derived from a single seed, a common approach in the literature. In this section, for each seed location we used paired t-tests to contrast all three possible pairings of resting state (unstructured vs. post-OLR; unstructured vs. post-MOTOR; post-OLR vs. post-MOTOR). SPMs resulting from these paired t-tests were feature thresholded at p b 0.01 (uncorrected), followed by cluster based family-wise error correction (FWEc) at p b 0.05. We were interested to examine whether significant changes in connectivity, when present, tend to occur randomly throughout the network associated with a given seed, or whether they tend to occur in voxels with a particular relation to the seed (such as strong or weak connectivity to the seed). In order to examine this question we calculated histograms of rFC across the whole brain (i.e. all voxels within the within brain mask), and rFC within the subset of voxels exhibiting significant connectivity change. Specifically, we calculated, separately for each seed and rest state, the average of the z-transformed rFC maps across all 25 participants. From the three resulting average rFC maps associated with each seed (one for each rest state) we derived a maximum observed connectivity map, FCmax. FCmax contains, at each voxel, the maximum absolute strength of the z-transformed r value across the three average connectivity maps for that seed (preserving the sign of the correlation):

278 279

O

299

277

0n 1 n n FC n;unstructed Σ FC n:post‐OLR Σ FC n:post‐Motor C BΣ 1 C ;1 ;1 ¼ maxsignðabsÞ B @ A n n n

where FCn,unstructured, FCn,post-OLR, and FCn,post-MOTOR are the z-transformed rFC maps of participant n in the unstructured, post-OLR, and postMOTOR rest states, respectively, for a given seed, and the maxsign(abs) function selects the maximum by absolute value whilst preserving the sign of the result. We then transformed these maximum connectivity

329 330 331 332 333 334 335 336 337 338 339 340 341 342

Analysis of activation paradigms

343

SPM-t maps for the OLR and MOTOR tasks are shown in Figs. 2a and b, respectively. These show typical language and sensorimotor activation patterns: there is left dominant activation on the OLR task, and right sensorimotor activation on the MOTOR task; default mode deactivation is present on both tasks. De/activation peaks were selected as seed coordinates (black dots) for subsequent FC analyses (Table 1).

344

Prior brain state dependent functional connectivity changes: ROI-to-ROI

350

R O

Prevalence of significant changes in connectivity maps

275 276

P

298

273 274

maps back into units of r and calculated the histogram of these maximum r values across the whole brain, and within the subset of voxels in which significant changes in rFC were observed (Fig. 7). In order to examine how across-participant variance changes as a function of prior brain state we employed a similar histogram approach. For each prior state (×3) and seed (×10) we calculated the across-participant variance in connectivity strength in each within brain voxel, then derived a histogram of these values across the whole brain (yielding 10 histograms – one for each seed – per rest state; 30 histograms total). For each rest state we then calculated an average histogram of variance values, by taking the mean of the 10 histograms derived from that state. The average histograms of across-participant variance in each prior state are shown in Fig. 8.

D

296 297

examined for a main effect of prior cognitive state (evaluating the hypothesis that prior brain state alters connectivity), and an interaction between prior state and seed location (evaluating the hypothesis that the effect of prior brain state is network specific; F-tests: feature inducing p b 0.001 unc., cluster extent p b 0.05 estimated via permutation testing). As in the ROI-to-ROI analyses, we did not consider a main effect of seed location. Custom scripts written in Matlab, harnessing the functionality provided by SPM, were used to implement permutation testing (n = 1000) of the alpha b 0.05 cluster extent threshold for F-tests. Specifically, we used the same two-way repeated measures ANCOVA model described above, however on each permutation for each participant we shuffled the rest state label (no-prior state, post-OLR, post-MOTOR) separately for each seed location. In other words, we carried out within-subject, within-seed permutation on the factor of prior state. The covariates entering the model were adjusted accordingly on each permutation (for instance, if the actual temporal order of the prior states for the jth subject was unstructured, post-OLR, post-MOTOR – corresponding to temporal order covariate weights of [0, 1, −1] – and the permuted order for the mth seed of the jth subject on the nth permutation was post-MOTOR, unstructured, post-OLR, the temporal order covariates corresponding to those inputs into the model were adjusted to [− 1, 0, 1]). We used t-tests and random field theory to follow up the main effect of prior cognitive state, as well as to examine the covariates of scanner protocol and cognitive task order (feature inducing p b 0.001 unc., FWEc b 0.05). The t-contrast on the scanner protocol regressor revealed significant clusters exclusively in white matter regions abutting the ventricles, and is not considered further.

In each rest state we calculated connectivity between the timecourses within ten ROIs (5 mm radius spheres) centred on the coordinates presented in Table 1, yielding a 10 × 10 correlation matrix that defines the connectivity between each pair of ROIs (Fig. 3). We divided the correlation matrix into five regions (see Methods) on the basis of inter- and intra-network connectivity, averaging the (r-to-z transformed) correlation values within each region. This yielded measures of inter- and intra-network connectivity in the three rest states, enabling us to examine both for prior-state dependent connectivity changes occurring within a given network (e.g. language-to-language connectivity) as well as changes occurring between networks (e.g. language-to-motor connectivity). Mean (across participants) connectivity matrices for each rest state are shown in the top row of Fig. 3; mean (across participants) values of the inter- and intra-network connectivity measures are plotted at the lower right of Fig. 3. Inspection of the plot indicates that average network–network connectivity changes, when they occur, tend to occur uniformly across all combinations of network pairings: post-Motor greater than post-OLR greater than unstructured. Were changes in connectivity network–network specific – for instance if connectivity within the language network ROIs increased in the post-OLR data, but connectivity within the motor network remained unchanged – the plots in Fig. 3 would depart from such consistency. We investigated this statistically by conducting a two-way (factor 1: network–network connectivity, five levels; factor 2: rest state, 3 levels) repeated measures ANCOVA on the data shown in Fig. 3. This returned a non-significant interaction (F(8,192) = 0.855, p N .05) between network–network rFC and rest state. The data therefore do not support the hypothesis that prior state induced changes in connectivity occur in a network specific manner. There was, however, a main effect of prior brain state (F(2,48) = 7.02, p b 0.01). Follow-up contrasts (least significant difference) indicated that average network–network connectivity in the post-motor state was significantly greater than that in the unstructured state (p = 0.002); there was a trend towards stronger connectivity in the postlanguage state relative to the unstructured (p = 0.055), and connectivity in the post-language and post-motor states were not different from one another (p = 0.087). Thus, our ROI based approach indicates that (1) post-explicit-task rest, especially post-motor rest, is associated

E

271 272

5

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

345 346 347 348 349

351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389

6

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

Unstructured

post-OLR

A

B

C

+64

+56

+44

+64

+56

+44

+64

+56

+44

+32

+20

+0

+32

+20

+0

+32

+20

+0

F

+64

+56

+44

+64

+56

+44

+32

+20

+0

+32

+20

+0

Language seed

P

H

+64

+56

+44

+64

+56

+32

+20

+0

+32

+20

+32

+56

+44

+20

+0

I

+44

+64

+56

+44

+0

+32

+20

+0

−0.70

−0.00 0.00 z transformed r

0.70

E

C

T

E

Default Mode seed

D

G

+64

F

E

O

D

R O

Motor seed

post-MOTOR

390 391

R

R

Fig. 4. Group connectivity maps. Mean, across participants, of the r-to-z maps derived from the separate rest blocks (columns), seeded from right postcentral gyrus (motor seed: top row, A–C), left middle frontal gyrus (language seed: middle row, D–F), and precuneus (default mode network seed: bottom row, G–I). The axial slices shown include the slices present in Fig. 2. There appears to be a tendency for overall connectivity to increase in the post-OLR and post-MOTOR rest blocks relative to the unstructured rest block (“hot” colours are more prominent in the centre and right columns); this is evaluated quantitatively in Figs. 5 and 6. Left hemisphere is shown on the right (radiological convention).

394 395

Prior brain state dependent functional connectivity changes: connectivity maps

396

The preceding ROI-to-ROI based connectivity analyses provide a description of changes in connectivity within select a priori regions of interest. In order to investigate the presence of potential effects occurring in regions outside of these a priori regions of interest we also conducted analyses based on seeded connectivity maps, in which connectivity is estimated between a given seed and every voxel in the brain. Fig. 4 shows the mean (across participants) of the r-to-z maps for three example seeds (rows) in the three different rest blocks (columns): a motor network seed — top row; a language network seed — middle row; a default mode network seed — bottom row (black spot shows seed location). Inspection of the figure reveals a tendency for overall connectivity to increase in the post-OLR and post-MOTOR rest blocks relative to the unstructured rest block (“hot” colours are more prominent in the centre and right columns). We used repeated measures ANCOVA to assess whether changes in the connectivity maps between rest states and seed locations were

399 400 401 402 403 404 405 406 407 408 409 410 411

C

N

U

397 398

O

392 393

with stronger network connectivity (relative to our unstructured condition), and (2) that when prior state induced changes in connectivity occur, they occur in a relatively global, as opposed to network specific, manner.

statistically significant. Specifically, we directly compared connectivity between the different rest states at each seed location by conducting a voxel-wise two-way repeated measures ANCOVA (factor 1: prior brain state, 3 levels; factor 2: seed location, 10 levels — see Table 1; covariate: image acquisition protocol) using the r-to-z transformed seeded connectivity maps from each participant as inputs. We first examined for a significant interaction between prior brain state and seed location. Permutation testing (see Material and methods) using a featureinducing threshold of p b 0.001 (uncorrected) yielded a cluster extent threshold of 155 voxels (p b 0.05). No clusters exceed this criterion, with the largest cluster extent being 43 voxels. Our data therefore do not provide evidence for a significant interaction between prior brain state and seed location. Thus, as in our ROI based analysis of rFC, analysis of seeded connectivity maps does not support the hypothesis that prior brain state alters rFC in a network dependent manner. We next examined for a main effect of prior brain state. Thresholding at p b 0.001 and a cluster extent ≥ 147 voxels (see Methods) yielded 8 significant clusters (Fig. 5). There were a number of clusters that were substantially larger than the estimated extent threshold. The largest cluster (1545 voxels) was located in left sensorimotor cortex, the next largest (1063 voxels) in bilateral occipital cortex, and the next two largest located in left (710 voxels) and right (592 voxels) primary auditory cortices. The remaining significant clusters were in superior mesial frontal cortex (295 voxels), right (181 voxels) and left (150 voxels) anterior insula, and left

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

412 413 414 415 416 417 418 419 Q7 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 Q8 435

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

7

post-MOTOR unstructured

Main effect of prior brain state

+62

+68

+56

+50

+56

+44

+28

+14

+8

+2

−8

+62

O

F

+68

+44

+28

+14

+8

+2

−8

R

R

E

C

T

E

D

P

R O

+50

N C O

0.0 F-value

12.0

Fig. 5. Main effect of prior brain state (thresholded at p b 0.001 (unc.), cluster extent ≥ 147 voxels). Language seeds: cyan; Motor seeds: green. Colour bar shows F-values. The network of regions in which significant rFC change occurs appears largely distinct from those recruited by the OLR and motor tasks (compare with Fig. 2). Left hemisphere is shown on the right (radiological convention).

437 438 439 440 441 442 443 444 445 446 447 448 449 450

cingulate gyrus (162 voxels). The pattern of regions of significant connectivity change resembles the superimposition of maps of the primary cortices (but for the absence of right sensorimotor cortex, the region activated by the motor task) and elements of the so-called cognitive control/salience network (anterior cingulate/SMA, bilateral anterior inferior insula; Dosenbach et al., 2007). Follow-up t-contrasts, comparing rFC in each rest state averaged across all 10 seeds [i.e. collapsing across all levels of seed location], revealed that the main effect of prior state was driven primarily by increases in rFC between the unstructured condition and the post-MOTOR condition (Fig. 6). No clusters survived thresholding (ui b 0.001, FWEc b 0.05) in the comparisons of unstructured vs. post-OLR, and post-OLR vs. post-MOTOR. The findings of a main effect of prior state, driven largely by increases in connectivity in the post-MOTOR data, are consistent with the ROI-to-ROI analysis

U

436

0.0 t-value

8.0

Fig. 6. The main effect of prior brain state is driven principally by differences between the post-MOTOR and unstructured conditions (thresholded at p b 0.001 (unc.), FWEc b 0.05), all of which manifest as increases in connectivity in the post-MOTOR condition. Slices and conventions as those in Fig. 5. There were no significant clusters in the comparison of postOLR vs. unstructured, and post-OLR vs. post-MOTOR. Colour bar shows t-values. Left hemisphere is shown on the right (radiological convention).

presented above (see Section Prior brain state dependent functional 451 connectivity changes: ROI-to-ROI and Fig. 2). 452 Q9 rFC changes are common, irrespective of seed location

453

In the preceding sections we examined prior brain state induced rFC changes using an ROI driven approach and a voxel-wise seeded connectivity approach, each based upon considering connectivity from 10 seed locations and utilising repeated measures ANCOVA. In this section, which is exploratory in nature, we consider the prevalence of rFC changes that would result when the analyses were based upon consideration of only a single seed location, as is frequently employed in the literature. For each seed location we constructed paired t-tests, using the r-to-z transformed seeded connectivity maps as inputs, contrasting each of

454 455

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

456 457 458 459 Q10 460 461 462

8

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

x 104

A

3000

2

Number of voxels

1.0

1 0

2500

B

histogram of r-values across the whole brain

−1

0

histogram of r-values within those voxels exhibiting significant changes in connectivity

1

2000 1500 1000

0.8 Proportion of voxels

3500

0.6

0.4

F

0.2

0.0

−1

−0.5

0

0.5

1

−1

Correlation coefficient (r)

−0.5

R O

0

O

500

0

0.5

1

Correlation coefficient (r)

473 474

Regions in which group level, state dependent rFC changes are observed

487

Inspection of Fig. 5 suggests that prior brain state related rFC changes tend to occur in regions spatially remote from the seeds. The network of regions in which significant rFC change occurs appears largely distinct from those recruited by the OLR and motor tasks (compare Figs. 2 and 5). Given the frequently observed similarity between task activation based and connectivity based maps, this suggests that connectivity changes tend to occur among those voxels that do not show the strongest connectivity with the seed. We sought to quantify this observation by exploring, via histograms of r values, the strength of correlation in all within brain voxels, and within only those voxels in which significant change occurred in the paired t-tests described in Section rFC changes are common, irrespective of seed location (see Material and methods). Fig. 7 shows the mean of these histograms, calculated across all ten seeds. Fig. 7 indicates that significant connectivity change is most common among those voxels whose correlation with the seed is as great or greater than the whole brain average, though only rarely in voxels of very strong connectivity or negative connectivity (the absence of significant changes among the most strongly correlated voxels – r N ~.65 – is likely due, at least in part, to an artefact of seed-based between-state

488

E

seeds exhibited clusters of significantly altered connectivity (all of which reflected greater connectivity in the post-MOTOR data); and in that of post-OLR state with post-MOTOR data 6 of 10 seeds exhibited clusters of significantly altered connectivity (all of which reflected greater connectivity in the post-MOTOR data). Of the 10 seeds, five showed significant changes in connectivity in all three of the rest state comparisons, and only one (the precuneus seed derived from deactivation in the motor task) did not reveal any significant changes in connectivity in any of the rest state comparisons. Across the 30 comparisons that we performed 21 of the SPMs contained one or more significant clusters (with a total of 53 significant clusters across these 21 SPMs). This is substantially in excess of the 1–2 SPMs with significant clusters (30 × 0.05) that would be expected to arise by chance were there no effect of prior state on functional connectivity.

T

C

E

R

R

Influence of prior brain state on variance in rFC 0.12

O

471 472

unstructured

0.10

C

469 470

0.08 0.06 0.04

post−OLR post−Motor

N

467 468

U

465 466

the three possible pairings of rest data (post-OLR vs. unstructured, postMOTOR vs. unstructured, post-MOTOR vs. post-OLR). We feature thresholded these t-tests at p b 0.01, FWEc b 0.05, and then counted the number of suprathreshold clusters. Results are summarised in Table 2. Prior brain state induced changes in functional connectivity were common in these paired t-tests. In the comparison of unstructured and post-OLR data 6 of 10 seeds exhibited clusters of significantly altered connectivity (all of which reflected greater connectivity in the post-OLR data); in that of unstructured with post-MOTOR data 9 of 10

proportion of voxels

463 464

D

P

Fig. 7. A, Histogram of r values, collapsed across the ten seeds, within the whole brain (grey), and within the subset of voxels in which significant changes in connectivity were observed (red). Grey and red triangles below the y =0 line show mean r values across the whole brain and among voxels of significant connectivity change, respectively. The y-axis in the main figure is truncated to facilitate inspection of the regions of the histogram showing significant change in r values, which comprise only a small fraction of voxels across the whole brain (see inset). This plot reveals that significant connectivity change is most common among those voxels whose correlation with the seed is as great or greater than the whole brain average, though only rarely in voxels of very strong connectivity or negative connectivity. B, Cumulative histogram of the data shown in A reveals a clear dearth of weak to negative r values within those voxels in which significant changes in connectivity occur.

0.02 0.00

0

0.05

0.1

0.15

variance (z transformed r value) Fig. 8. Whole brain histograms of the variance (calculated across participants, within each voxel) of the z-transformed r values in the different rest blocks. Each trace shows the mean of the ten histograms derived separately for each seed in the corresponding rest block; shaded polygons enclose ±2 standard errors of the mean (shaded region for the postMOTOR data encloses ±1 SEM so as to reveal the post-OLR data, which would otherwise be obscured). The variance distribution of connectivity following a structured task is lower compared to that following the unstructured task.

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

475 476 477 478 479 480 481 482 483 484 485 486

489 490 491 492 493 494 495 496 497 498 499 500 Q11 501 502 503 504 505 506 507

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

t2:3

Table 2 Number of suprathreshold clusters (p b 0.01, FWEc b 0.05) identifying significant changes in connectivity (increases or decreases) in the paired t-tests for each seed location. Resting state connectivity comparison

t2:4 t2:5 t2:6 t2:7 t2:8 t2:9 t2:10 t2:11 t2:12 t2:13 t2:14 t2:15

Language seeds

Motor seeds

Seed region

Unstructured vs. post-OLR

Unstructured vs. post-MOTOR

Post-OLR vs. post-MOTOR

Marginal means

L MFG L IFG PaCG L PrCG Pcunlang R PoCGinf R PoCGsup SMA L PoCG Pcunmotor Marginal means

1 2 2 0 4 1 0 2 0 0 1.2

3 1 1 5 5 2 3 5 3 0 2.8

1 1 5 1 0 1 0 4 0 0 1.3

1.66 1.33 2.66 2 3 1.33 1.0 3.66 1.0 0 1.77

F

t2:1 t2:2

9

2.13

1.4

Seed region (see Table 1 for corresponding coordinates): L = left; R = right; MFG = middle frontal gyrus; IFG = inferior frontal gyrus; PaCG = paracingulate gyrus; Pcun = precuneus; PoCG = postcentral gyrus; PrCG = precentral gyrus; SMA = supplementary motor cortex; sup = superior; inf = inferior; lang = language activation defined; motor = motor activation defined.

508

513

comparisons, an issue to which we return in Discussion). The whole brain mean r value, averaged across the ten seeds, was 0.20, lower than the mean r value of 0.28 observed among voxels in the clusters of significant connectivity change. Overall, this indicates that changes in connectivity, when present, occur in a subset of those voxels that show moderate positive correlation with the seed.

514

Priming reduces across-participant variance

515

522

The above results indicate that constraining prior brain state can produce changes in the mean (across participants) strength of correlation between different brain regions. We examined whether it also affects variance by calculating, for each condition, whole brain histograms of the variance (across participants) of the z-transformed r values at each voxel (Fig. 8). Considered relative to the unstructured data, the post-OLR and post-MOTOR data show that constraining prior brain state reduces across-participant variance in connectivity strength.

523

Considerations of time and global signal regression in the analysis

520 521

524

U

N C O

R

R

Whilst we explicitly manipulated prior brain state in this experi525 ment, the different rest data sets also differ in terms of time of acquisi526 tion (relative to entering the scanner, see Fig. 1). We examined for an 527 effect of time by performing a t-contrast on the covariate in the repeat528 ed measures ANCOVA (described in Section Statistical analysis of 529 Q12 changes in voxel-wise functional connectivity maps and reported in 530 Section Prior brain state dependent functional connectivity changes: 531 connectivity maps) that coded for rest block (Rest 2 or Rest 3). This con532 trast revealed stronger connectivity in third rest period relative to the 533 second rest period, principally in bilateral planum temporale, bilateral 534 convexity sensorimotor cortex, and bilateral temporo-occipital strips 535 running along a fusiform–parahippocampal axis. 536 We also repeated our exploratory analyses, based on paired t-tests 537 contrasting the different rest blocks (T1 vs. T2, T1 vs. T3, and T2 vs. 538 T3). These data also revealed significant clusters of connectivity change 539 (not shown). Thus, the ANCOVA and t-test analyses suggest that time in 540 scanner also influences rFC. In general, connectivity increased with time 541 (see also Duff et al., 2008), with the majority of significant clusters 542 reflecting greater connectivity in the later time point relative to the ear543 lier time point. 544 We sought to examine the unique contribution of prior brain state 545 by including time as a covariate in paired t-tests on the data obtained 546 from rest blocks 2 and 3, the post-OLR and post-MOTOR rest data sets 547 in which time is appropriately counter-balanced. Including time as a co548 variate did not change the results reported in the third column of 549 Table 2 (the post-OLR vs. post-MOTOR comparison), indicating that 550 the effect of prior task is unchanged after first (linearly) removing the 551 effect of time. A counterpart analysis, examining for effects of time

552

Discussion

578

We have investigated the effect of prior brain state on resting state functional connectivity (rFC), collecting rest data following three different prior brain states from a sample of 25 healthy controls. We hypothesised that rFC changes would show a degree of network specificity such that, for instance, connectivity within a motor network would be largely unaffected by prior performance of a language task. We found, however, that rFC changes, when present, do not respect such network “boundaries”. Analyses based upon (1) the comparison of timecourses within a set of ROIs, and (2) voxel-wise seeded functional connectivity maps, both indicated that when prior brain state influences rFC it does so in a largely network independent manner. Connectivity changes were not restricted to those areas activated by the prior in-scanner task, but also occurred in other brain regions. Significant functional connectivity changes tended to occur in voxels of moderate connectivity with the seed. Covariance analyses revealed, in addition to the effect of prior brain state, an independent effect of time-in-scanner (which itself may be conceptualised as a form of prior brain state). Acrossparticipant variance was reduced in the rest data acquired following explicitly specified prior brain states (post-OLR and post-MOTOR) relative

579 580

D

P

R O

after including prior task as a covariate also revealed significant clusters of connectivity change (on average 3.2 clusters per seed in the comparison of T2 vs. T3). Thus, both prior brain state and time exert independent effects on rFC. A second factor that we considered in additional analyses is the issue of global signal regression, a topic of considerable debate. There is an emerging consensus that denoising is best accomplished by regressing out motion, white matter signal, and ventricular signal (as well as removing other physiological sources of noise – e.g. cardioballistic, respiratory – where possible), but not regressing out the global signal (Birn, 2012; Gotts et al., 2013; Saad et al., 2012). This approach was adopted above, however we also ran a second set of analyses in which we included the global signal as a nuisance regressor during denoising. With the global signal removed we observed one or more significant clusters in 15 of the 30 rest state comparisons (with a total of 31 significant clusters overall, see Supplementary Table 1). Whilst this is lower than the 21 SPMs containing significant clusters that we observed with global signal regression omitted (Table 2), it is still well in excess of the 1–2 clusters expected by chance (30 × 0.05). The contrasts in which significant clusters were most frequently observed were also altered by inclusion of the global signal: rFC changes were most common when comparing the unstructured condition with either of the two post-task conditions (postOLR and post-MOTOR). Thus, in the context of the present data, global signal regression alters the pattern of correlations between regions (Gotts et al., 2013; Saad et al., 2012), and the manner in which these patterns are affected by prior brain state.

E

518 519

T

516 517

C

511 512

E

509 510

O

t2:16 t2:17

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577

581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597

605 606

609

The main conclusion from this study is that, relative to a relatively unconstrained prior brain state (our unstructured rest block, corresponding to volumes acquired upon participants entering the scanner), 612 imposing a uniform prior brain state (having all participants execute a 613 given in-scanner paradigm) increases connectivity within the brain in 614 a non-specific manner. This effect was strongest for the post-MOTOR 615 data, and the same trend was apparent in the post-OLR data. This find616 ing is contrary to our hypothesis that such changes would exhibit net617 work specificity. It is, however, similar to the findings reported by Duff 618 et al. (2008); these authors did not discuss the tendency, observed 619 across all ROIs they examined, for resting state functional connectivity 620 to increase following a motor task. Before considering the basis of this 621 non-specific increase in connectivity we consider how the ROI-to-ROI 622 based approach and the connectivity maps approach relate to one an623 other. In so doing, we also draw attention to a feature of seed-based con624 nectivity analyses that precludes rFC changes in regions proximal to the 625 seed. 626 In the ROI-to-ROI based approach, connectivity changes were appar627 ent in each network–network pairing, manifesting as increases in con628 nectivity (Fig. 3). In the voxel-wise approach (Fig. 5), where the ROIs 629 in the ROI-to-ROI approach now serve as seeds, significant changes in 630 rFC were only observed in regions spatially removed from the seed loca631 tions; no changes in connectivity were observed in voxels overlapping 632 the seed regions. This apparent conflict between the ROI-to-ROI and 633 voxel-wise approaches, however, results from an “artefact” at the seed 634 locations inherent to the connectivity approach itself. Specifically, it 635 stems from the disproportionately large variance (across seed-based 636 connectivity maps, within a given rest state) that occurs at the seed lo637 cations. Consider a collection of voxels, Vx, within and adjacent to seed x. 638 They will have r ≈ 1 in the connectivity map generated from seed x, but 639 generally much lower r values in the connectivity map generated from 640 the other seeds. Now consider a single voxel within Vx. When the main 641 effect of rest state is calculated for this given voxel the (z transformed) r 642 values are effectively averaged across the connectivity maps generated 643 from each of the spatially separate seeds in each of n subjects. Thus, for 644 the given voxel the distribution of r values will contain n r values ≈ 1 645 (from the n maps, one per participant, generated from seed x) and a 646 larger collection of weaker r values (from the connectivity maps gener647 ated for each participant from all of the other seeds). This translates into 648 large variance for the given voxel, and indeed for all voxels within Vx. 649 Conversely, for any voxel that is spatially remote from all of the seed 650 locations, we expect a priori that the distribution of r values will be 651 more tightly distributed about the group mean (because there is no lon652 ger a collection of n r values ≈ 1 biasing the mean and variance of that 653 voxel). This results in smaller variance, and therefore greater potential 654 to reveal significant changes in mean r. Thus, there is a bias against a 655 main effect of prior state being expressed at the seed locations them656 selves using the seeded connectivity map approach. 657 This bias influences not only the main effect of prior state, but holds 658 also for the analyses of simple effects (t-tests) contrasting a pair of con659 nectivity maps derived from a given seed following two different prior 660 Q13 states (Section rFC changes are common, irrespective of seed location). 661 The utility of a between-state comparison is predicated on the

C

E

R

R

O

C

N

U

662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684

Why does connectivity tend to increase and variance decrease relative to 685 the initial rest period? 686 Our data show that, in general, connectivity across participants tended to increase relative to the initial unstructured rest period. This tendency for increased connectivity was apparent whether the data were analysed with respect to prior cognitive task, or whether the data were organised with respect to time in scanner, with these two factors making independent contributions. A similar effect has been reported by Duff et al. (2008). We speculate that at least two factors may contribute to this tendency for connectivity to increase relative to the unstructured rest. The first is that during the initial rest each subject's brain state was unconstrained, common only in that they had just entered the MR environment. Thus, it could be argued that the thought processes engaged in whilst at rest during this initial period were maximally free to vary across individuals, and therefore maximally heterogeneous in terms of the network patterns engaged. For instance participant w may be particularly anxious about being in the scanner environment for the first time, participant x may be engaged principally in visual imagery based day dreaming, participant y in silently “playing” their favourite song in their head, and participant z in covertly reciting their to-do list. This heterogeneity could be expected to give rise to inflated variance in the connectivity strengths for a given network across participants, as indeed we observed (Fig. 8). Thus, using a common form of prior brain state appears to homogenise the nature of the rest state across individuals. By reducing variance, utilisation of a common prior brain state could therefore serve to increase power in group-analyses of resting state functional connectivity. Homogenisation of the rest state may also occur within individuals across time. That is, the focussed performance of a given task in the prior state may lead to a more stable brain state in the minutes following, which may explain an increase in connectivity post-task. The regions showing the main effect of significant increase in connectivity include sensory and task positive areas (Fig. 5), located in networks that to some degree might be expected to be recruited by engagement of both prior tasks. The common recruitment of these regions by both tasks may contribute to their generally greater connectivity post-task. Recent developments in methods for examining dynamic connectivity states may assist investigation of these issues; however this is beyond the scope of the present study. A second factor that could account for the tendency towards increases in connectivity relative to the initial rest is physiological changes related

T

610 611

F

604

O

Constraining prior brain state increases connectivity across brain networks

602 603

R O

608

600 601

assumption that the connectivity values in a given voxel are free to vary between those states. This is (effectively) not the case for any voxels within or proximal to a (multivoxel) seed as their r values will be close to 1.0 in each rest state (a seed is always correlated with itself). Note that this is necessarily the case, even if the nature of the signal in seed voxel y changes considerably across different rest states. Thus, our observation that connectivity changes tend to occur in regions of moderate but not high connectivity with the seed likely, at least in part, reflects an artefact stemming from the nature of the seed-based approach. An alternative viewpoint could be provided by an analysis based on whole brain connectivity methods, comparing – for instance – degree centrality between the different rest states (Masterton et al., 2012), or using data driven approaches such as probabilistic independent component analysis of resting state fMRI (Beckmann et al., 2005; Bhaganagarapu et al., 2013). This is the subject of ongoing work in our laboratory. The ROI-to-ROI analyses do not suffer from such biases at the seed locations themselves (because, unlike in the voxel-wise connectivity map approach, the correlation of seed x with itself is excluded from the analysis), and hence are an important compliment to the connectivity map analyses. The ROI-to-ROI analyses enabled us to examine changes between the seed regions themselves, whereas the voxelwise analyses enabled us to examine changes outside of the seed regions.

P

607

to the initial rest period, in which prior state was comparatively unconstrained (unstructured). Overall, these data clearly demonstrate that prior brain state is capable of exerting a significant influence on rFC. Our results extend previous work (Duff et al., 2008; Gordon et al., 2014; Grigg and Grady, 2010; Harrison et al., 2008; Hasson et al., 2009; Klingner et al., 2012; Peltier et al., 2005; Stevens et al., 2010; Waites et al., 2005) by showing that (1) prior brain state induced rFC changes are not necessarily specific to the network engaged by that prior state, and (2) that constraining prior brain state across individuals can reduce the variance in rFC that is observed across participants.

D

598 599

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

E

10

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

726 727

Properties of the regions in which rFC induced changes occur

757

U

N C O

R

R

E

C

Fig. 7 shows that prior brain state induced FC changes occur princi758 pally in regions of moderate positive connectivity. This therefore im759 plies that, from a spatial perspective, the rFC changes occur outside of 760 the major “hubs” (regions of highest connectivity) of rFC networks. 761 This is consistent with the observations of others ((Kang et al., 2011; 762 Shehzad et al., 2009); discussed below), that test–retest changes in 763 rFC are more common in regions of weak to moderate rFC, and are 764 not observed in the major hubs associated with a given network (see 765 Section Constraining prior brain state increases connectivity across 766 Q15 brain networks for consideration of why changes in rFC are not “re767 vealed” at the seed locations themselves). 768 We hypothesised that rFC changes would be network specific: that, 769 for instance, following execution of the MOTOR task connectivity chang770 es would occur primarily within sensorimotor regions recruited by the 771 task, with little change in connectivity to other networks. Inspection of 772 Fig. 5 (which reports the main effect of prior brain state) reveals that 773 whilst some of the clusters of significant change do lie within sensorimo774 tor regions, others however are proximal to primary visual and auditory 775 cortices, and proximal to the superior mesial and anterior insular compo776 nents of the task positive network. The follow-up contrasts on the main 777 effect of prior brain state revealed that these changes were increases in 778 connectivity, indicating that – at least for the seeds used here – there 779 was a tendency for connectivity with primary sensory and motor cortices 780 and the task positive network to increase relative to the initial rest state. 781 Overall, the results presented in Figs. 5 and 7 indicate that significant 782 changes in rFC are not restricted to the brain areas most strongly asso783 ciated with the prior brain state, and that they tend to occur in regions 784 characterised by moderate, but not weak or negative, connectivity. 785

Prior brain state induced changes in functional connectivity are common

786

In our exploratory analyses we cluster corrected each SPM at FWEc b 0.05. As the seeds were defined by maxima on activation

787

790 791 792 793 794 795 796 797 798 799 800 801 802

O

F

788 789

R O

Comparison with previous work investigating changes across different 803 timescales 804

P

A number of recent studies have documented within-subject temporal variability of rFC maps (Anderson et al., 2011; Berns et al., 2013; Chang and Glover, 2010; Esposito et al., 2014; Harmelech et al., 2013; Kang et al., 2011; Shehzad et al., 2009; Van Dijk et al., 2010; Zuo et al., 2010). Kang et al. (2011) quantified temporal variability of seed-based rFC within long rest periods. They observed that the voxels most stably associated with the seed were those associated with the greatest r values. Similarly, Shehzad et al. (2009) assessed the test–retest reliability of rFC measures and found that reliability increased with increasing strength of functional connectivity [see also (Anderson et al., 2011; Damoiseaux et al., 2006; Van Dijk et al., 2010; Wang et al., 2011)]. These findings align with ours, suggesting that the strongest connections remain relatively stable, but that the strength of association between moderately connected regions varies over time (though see Section Constraining prior brain state increases connectivity across brain networks for discussion of the absence of significant rFC in the most strongly connected voxels). The results of a number of other studies focussing on different aspects of rFC bear upon the central question of the current work: whether prior brain state can systematically bias connectivity networks across individuals. That rFC networks are subject to brain-state-dependent modulation has been shown by comparing rFC at rest with rFC during task execution (Esposito et al., 2006; Fransson, 2006). A second body of work has shown that new learning (a change in brain state) can be accompanied by changes in rFC (Albert et al., 2009; Esposito et al., 2014; Jolles et al., 2013; Sun et al., 2007; Tambini et al., 2010), a further demonstration of experience dependent modulation of connectivity. A third group of studies, most akin to the present study, examine sets of rest data acquired in a single scanning session, separated by performance of a particular task (Barnes et al., 2009; Duff et al., 2008; Grigg and Grady, 2010; Harrison et al., 2008; Hasson et al., 2009; Klingner et al., 2012; Peltier et al., 2005; Stevens et al., 2010; Waites et al., 2005). A subset of these studies focussed primarily on changes in the timeseries spectra of the different sets of rest data (Barnes et al., 2009; Duff et al., 2008). Whilst such changes do not necessarily indicate changes in rFC per se, they nonetheless demonstrate the influence of brain state on intrinsic activity. A second subset of these studies looked explicitly at the influence of an immediately preceding cognitive task on rFC (Duff et al., 2008; Gordon et al., 2014; Grigg and Grady, 2010; Harrison et al., 2008; Hasson et al., 2009; Klingner et al., 2012; Peltier et al., 2005; Stevens et al., 2010; Waites et al., 2005). For instance Gordon et al. (2014) measured rFC within and between the task positive network and the default mode network before, during, and immediately following a working memory task. They found that relative to the initial rest period, there were changes in inter- and intra-network connectivity during task execution, and that these changes persisted into the posttask rest period. Interestingly, they observed that individuals with a

D

T

756

tasks we expect there to be some relationship between the seeds themselves (i.e. they will not be orthogonal). Our aim in these exploratory analyses, however, was to evaluate the prevalence of significant clusters of rFC change across typical analyses, rather than select a single example as proof of a prior brain state effect. Thus, beyond applying FWEc b 0.05 for each SPM in these exploratory analyses, we elected not to further correct for the multiple contrasts we performed. Under the null hypothesis one would expect to only have observed about one or two significant clusters across the 30 comparisons we conducted (10 seeds × 3 rest state comparisons = 30 total comparisons × 0.05 = 1.5 expected SPMs with significant clusters across all comparisons); we instead observed a total of 21 SPMs with significant clusters (with a total of 53 significant clusters across these 21 SPMs), well in excess of that expected simply by chance. Thus prior brain state induced rFC changes are common, and as such identify a factor that merits consideration in connectivity studies.

E

to time in scanner. It may be that variables such as respiration rate, heart rate, blood pressure, CO2, wakefulness, anxiousness, etc., vary over time as 728 participants settle into the scanning session. Different tasks might pro729 duce, for instance, differential levels of mental effort or different auto730 nomic activity patterns, with the potential for slow recovery processes 731 to bleed over into subsequent rest periods, thereby affecting resting 732 state connectivity (Birn et al., 2006; Chang et al., 2009). At the time our 733 data were acquired we did not have the capacity to record such physio734 logical variables, so unfortunately we are unable to consider them in our 735 analyses. 736 Similarly, recent work has shown that subject motion can influence 737 rFC by producing widespread burst-like coherent fluctuations in the 738 BOLD timecourse, sufficient to result in spurious group rFC differences 739 when there are differing amounts of head motion in otherwise matched 740 groups (Power et al., 2012; Satterthwaite et al., 2012; Van Dijk et al., 741 2012). We therefore considered whether participant motion might ac742 count for effects in our data. In order to control for motion effects our 743 data preprocessing pipeline included regressing out volumes associated 744 with scan-to-scan head movements in excess of 0.5 mm (Methods 745 Q14 Section Additional preprocessing for functional connectivity analyses). 746 The total number of motion corrections applied was small (unstructured: 747 2 corrections total, across 2 participants; post-OLR: 4 corrections total, 748 across 3 participants; post-MOTOR: 5 corrections total, across 3 partici749 pants). Further, the root mean square motion (estimated from the rigid 750 body realignment parameters) did not differ significantly between rest 751 periods [repeated measures ANOVA, F = 0.05, p N 0.6; rest1 geometric 752 μ (SD) = 0.042 mm (0.035 mm), rest2 = 0.048 mm (0.034 mm), 753 rest3 = 0.049 mm (0.03 mm)]. We therefore consider it unlikely that dif754 ferent degrees of motion accounts for the prior brain state induced rFC dif755 ferences we observed.

11

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

The problem of interpreting alterations in rFC

884 885

900

A substantial body of work now confirms that prior brain state is capable of influencing resting state dynamics. Importantly, the rFC changes occurred within the context of a single scanning procedure lasting less than 30 min, a period orders of magnitude too brief for the changes to have been mediated by structural cerebral changes. This has important implications for studies of rFC changes associated with disease states [see (Fornito and Bullmore, 2010; Fox and Raichle, 2007; Greicius, 2008; van den Heuvel and Pol, 2010) for reviews], be they psychiatric (e.g. depression, schizophrenia, PTSD) or neuropathological (e.g. Alzheimer disease, multiple sclerosis, ALS, stroke, epilepsy): to the extent that prior state can influence rFC, between group differences revealed by comparing rFC of normal controls with neurological or psychiatric groups may reflect psychological rather than structural/neuropathological processes. Such psychological effects could alter both the strength of rFC between brain regions and the variance in rFC values within and between groups. Using a common prior state could serve to reduce variance across subjects.

901

Conclusions

902

Our data indicate that at both the single participant and the group level, changes in rFC contingent upon prior brain state are common. Such rFC changes manifest as changes in both the strength and variance of connectivity. The most important practical issue arising from this work is the demonstration that not all rest states are equivalent (Grigg and Grady, 2010). The potential import of such context dependent FC changes in pooled intra- and inter-institutional data sets is uncertain, but likely depends on the nature of the research question asked. If rest state data obtained within different temporal contexts are pooled in an effort to increase power, the increase in numbers may be offset to some extent by an increase in variance. Conversely, utilising a common prior state across individuals could serve to reduce variance, and this

896 897 898 899

903 904 905 906 907 908 909 910 911 912 913

E

R

R

O

894 895

C

892 893

N

890 891

U

888 889

C

883

886 887

could be exploited to increase power. Further, it is important to ensure that whenever rFC is being compared between two groups, the two groups do not systematically differ in terms of the cognitive context within which the rest data were collected. Any such differences could produce a significant between group effect that simply reflects differences in prior state, rather than differences based upon the variable of interest used to segregate the groups. Without awareness and control of prior brain state, changes in rFC may thus be incorrectly attributed to changes in the underlying neurobiology associated with subject categorisation. Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.neuroimage.2014.11.037.

914 915 916 917 918 919 920 921 922 923 924 925 926

F

Acknowledgments

P

Disclosure

R O

O

We thank Anthony B. Waites and Alexandra Stanislavsky for assistance with data collection. This study was supported by the National Health and Medical Research Council of Australia (Programme Grant 628952 and practitioner fellowship 527800 to GDJ), the Austin Hospital Medical Research Foundation and the Operational Infrastructure Support Program of the State Government of Victoria, Australia.

927 928 929 930 931 932 933 934 935 936

The authors report no conflicts of interest.

937

References

938

Abbott, D.F., Jackson, G.D., 2001. iBrain-software for analysis and visualisation of functional MR images. NeuroImage 13, s59. Abbott, D.F., Masterton, R.A.J., Waites, A.B., Bhaganagarapu, K., Pell, G., Harvey, M.R., Sharma, G.J., Jackson, G.D., 2011. The iBrainAnalysis Toolbox for SPM. Proc. 17th Annual Meeting of the Organisation for Human Brain Mapping, Quebec City, Canada. Albert, N.B., Robertson, E.M., Miall, R.C., 2009. The resting human brain and motor learning. Curr. Biol. 19, 1023–1027. Anderson, J.S., Ferguson, M.A., Lopez-Larson, M., Yurgelun-Todd, D., 2011. Reproducibility of single-subject functional connectivity measurements. AJNR Am. J. Neuroradiol. 32, 548–555. Bandettini, P.A., 2009. What's new in neuroimaging methods? Ann. N. Y. Acad. Sci. 1156, 260–293. Barnes, A., Bullmore, E.T., Suckling, J., 2009. Endogenous human brain dynamics recover slowly following cognitive effort. PLoS ONE 4, e6626. Beckmann, C.F., DeLuca, M., Devlin, J.T., Smith, S.M., 2005. Investigations into resting-state connectivity using independent component analysis. Philos. Trans. R. Soc. B Biol. Sci. 360, 1001–1013. Berns, G.S., Blaine, K., Prietula, M.J., Pye, B.E., 2013. Short- and long-term effects of a novel on connectivity in the brain. Brain Connect. (epub ahead of print). Bhaganagarapu, K., Jackson, G.D., Abbott, D.F., 2013. An automated method for identifying artifact in independent component analysis of resting-state fMRI. Front. Hum. Neurosci. 7. Birn, R.M., 2012. The role of physiological noise in resting-state functional connectivity. NeuroImage 62, 864–870. Birn, R.M., Diamond, J.B., Smith, M.A., Bandettini, P.A., 2006. Separating respiratoryvariation-related neuronal-activity-related fluctuations in fluctuations from fMRI. NeuroImage 31, 1536–1548. Biswal, B., Yetkin, F.Z., Haughton, V.M., Hyde, J.S., 1995. Functional connectivity in the motor cortex of resting human brain using echo-planar MRI. Magn. Reson. Med. 34, 537–541. Chang, C., Glover, G.H., 2010. Time–frequency dynamics of resting-state brain connectivity measured with fMRI. NeuroImage 50, 81–98. Chang, C., Cunningham, J.P., Glover, G.H., 2009. Influence of heart rate on the BOLD signal: the cardiac response function. NeuroImage 44, 857–869. Damoiseaux, J.S., Rombouts, S.A., Barkhof, F., Scheltens, P., Stam, C.J., Smith, S.M., Beckmann, C.F., 2006. Consistent resting-state networks across healthy subjects. Proc. Natl. Acad. Sci. U. S. A. 103, 13848–13853. Dosenbach, N.U., Fair, D.A., Miezin, F.M., Cohen, A.L., Wenger, K.K., Dosenbach, R.A., Fox, M.D., Snyder, A.Z., Vincent, J.L., Raichle, M.E., Schlaggar, B.L., Petersen, S.E., 2007. Distinct brain networks for adaptive and stable task control in humans. Proc. Natl. Acad. Sci. U. S. A. 104, 11073–11078. Duff, E.P., Johnston, L.A., Xiong, J.H., Fox, P.T., Mareels, I., Egan, G.F., 2008. The power of spectral density analysis for mapping endogenous BOLD signal fluctuations. Hum. Brain Mapp. 29, 778–790. Esposito, F., Bertolino, A., Scarabino, T., Latoffe, V., Blasi, G., Popolizio, T., Tedeschi, G., Cirillo, S., Goebel, R., Di Salle, F., 2006. Independent component model of the default-mode brain function: assessing the impact of active thinking. Brain Res. Bull. 70, 263–269. Esposito, F., Otto, T., Zijlstra, F.R.H., Goebel, R., 2014. Spatially distributed effects of mental exhaustion on resting-state FMRI networks. PloS ONE 9.

939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 Q18 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987

T

more coordinated default mode network during the pre-task resting state and during the working memory task exhibited higher scores on 854 a self-reported attention scale, suggesting a relationship between rFC 855 and behaviour. 856 Hasson et al. (2009) suggested three (not mutually exclusive) possi857 ble mechanisms that might give rise to prior state induced changes in 858 rFC: that regions of rFC change, 1) might be involved in post-task func859 tions such as rehearsal, elaboration, or encoding of the prior state into 860 memory; 2) may facilitate active disengagement from the previous 861 task (with regional variation in a task specific manner); 3) may reflect 862 “cognitive inertia” in which internal processes that mediate the prior 863 task permeate into the subsequent rest period. The first two mecha864 nisms are much less likely to manifest in our data given the consider865 ably longer rest periods we employed (12 second rests in (Hasson 866 et al., 2009), versus ~5 minute rest periods used here). The third, “cog867 Q16 nitive inertia”, overlaps with the psychological account of increased rFC 868 relative to the initial rest period that we offered above (Section Why 869 does connectivity tend to increase and variance decrease relative to 870 Q17 the initial rest period?). To these mechanisms we also add the possibil871 ity of physiological changes precipitated by prior state, as discussed in 872 Section Why does connectivity tend to increase and variance decrease 873 relative to the initial rest period? above. 874 Collectively, our study along with the aforementioned studies 875 show that prior brain state can induce changes in connectivity with876 in a single scanning session, with changes tending to occur in regions 877 of moderate connectivity. The present work adds to these findings, 878 demonstrating that changes in connectivity are not restricted to 879 the brain network(s) most strongly recruited during the prior brain 880 state, but can also occur in networks unrelated to the prior state, 881 and that prior state can serve to reduce across-participant voxel882 wise variance in connectivity strength.

D

852 853

E

12

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

C. Tailby et al. / NeuroImage xxx (2014) xxx–xxx

E

C

F

O

R O

P

D

U

N C O

R

1092

Saad, Z.S., Gotts, S.J., Murphy, K., Chen, G., Jo, H.J., Martin, A., Cox, R.W., 2012. Trouble at rest: how correlation patterns and group differences become distorted after global signal regression. Brain Connect. 2, 25–32. Satterthwaite, T.D., Wolf, D.H., Loughead, J., Ruparel, K., Elliott, M.A., Hakonarson, H., Gur, R.C., Gur, R.E., 2012. Impact of in-scanner head motion on multiple measures of functional connectivity: Relevance for studies of neurodevelopment in youth. NeuroImage 60, 623–632. Shehzad, Z., Kelly, A.M., Reiss, P.T., Gee, D.G., Gotimer, K., Uddin, L.Q., Lee, S.H., Margulies, D.S., Roy, A.K., Biswal, B.B., Petkova, E., Castellanos, F.X., Milham, M.P., 2009. The resting brain: unconstrained yet reliable. Cereb. Cortex 19, 2209–2229. Soares, J.M., Sampaio, A., Ferreira, L.M., Santos, N.C., Marques, P., Marques, F., Palha, J.A., Cerqueira, J.J., Sousa, N., 2013. Stress impact on resting state brain networks. PloS ONE 8. Stevens, W.D., Buckner, R.L., Schacter, D.L., 2010. Correlated low-frequency BOLD fluctuations in the resting human brain are modulated by recent experience in categorypreferential visual regions. Cereb. Cortex 20, 1997–2006. Strauss, E., Sherman, E.M.S., Spreen, O., 2006. A Compendium of Neuropsychological Tests: Administration, Norms and Commentary. 3rd ed. Oxford University Press, New York. Sun, F.T., Miller, L.M., Rao, A.A., D'Esposito, M., 2007. Functional connectivity of cortical networks involved in bimanual motor sequence learning. Cereb. Cortex 17, 1227–1234. Tambini, A., Ketz, N., Davachi, L., 2010. Enhanced brain correlations during rest are related to memory for recent experiences. Neuron 65, 280–290. van den Heuvel, M.P., Pol, H.E.H., 2010. Exploring the brain network: a review on restingstate fMRI functional connectivity. Eur. Neuropsychopharmacol. 20, 519–534. Van Dijk, K.R., Hedden, T., Venkataraman, A., Evans, K.C., Lazar, S.W., Buckner, R.L., 2010. Intrinsic functional connectivity as a tool for human connectomics: theory, properties, and optimization. J. Neurophysiol. 103, 297–321. Van Dijk, K.R.A., Sabuncu, M.R., Buckner, R.L., 2012. The influence of head motion on intrinsic functional connectivity MRI. NeuroImage 59, 431–438. Waites, A.B., Stanislavsky, A., Abbott, D.F., Jackson, G.D., 2005. Effect of prior cognitive state on resting state networks measured with functional connectivity. Hum. Brain Mapp. 24, 59–68. Waites, A.B., Briellmann, R.S., Saling, M.M., Abbott, D.F., Jackson, G.D., 2006. Functional connectivity networks are disrupted in left temporal lobe epilepsy. Ann. Neurol. 59, 335–343. Wang, J.H., Zuo, X.N., Gohel, S., Milham, M.P., Biswal, B.B., He, Y., 2011. Graph theoretical analysis of functional brain networks: test–retest evaluation on short- and long-term resting-state functional MRI data. PloS ONE 6, e21976. Wang, Z.J., Liu, J.M., Zhong, N., Qin, Y.L., Zhou, H.Y., Li, K.C., 2012. Changes in the brain intrinsic organization in both on-task state and post-task resting state. NeuroImage 62, 394–407. Whitlow, C.T., Casanova, R., Maldjian, J.A., 2011. Effect of resting-state functional MR imaging duration on stability of graph theory metrics of brain network connectivity. Radiology 259, 516–524. Wood, A.G., Saling, M.M., Abbott, D.F., Jackson, G.D., 2001. A neurocognitive account of frontal lobe involvement in orthographic lexical retrieval: an fMRI study. NeuroImage 14, 162–169. Zuo, X.N., Kelly, C., Adelstein, J.S., Klein, D.F., Castellanos, F.X., Milham, M.P., 2010. Reliable intrinsic connectivity networks: test–retest evaluation using ICA and dual regression approach. NeuroImage 49, 2163–2177.

E

T

Fornito, A., Bullmore, E.T., 2010. What can spontaneous fluctuations of the blood oxygenation-level-dependent signal tell us about psychiatric disorders? Curr. Opin. Psychiatry 23, 239–249. Fox, M.D., Raichle, M.E., 2007. Spontaneous fluctuations in brain activity observed with functional magnetic resonance imaging. Nat. Rev. Neurosci. 8, 700–711. Fransson, P., 2006. How default is the default mode of brain function? Further evidence from intrinsic BOLD signal fluctuations. Neuropsychologia 44, 2836–2845. Friston, K.J., Glaser, D.E., Henson, R.N., Kiebel, S., Phillips, C., Ashburner, J., 2002. Classical and Bayesian inference in neuroimaging: applications. NeuroImage 16, 484–512. Gordon, E.M., Breeden, A.L., Bean, S.E., Vaidya, C.J., 2014. Working memory-related changes in functional connectivity persist beyond task disengagement. Hum. Brain Mapp. 35, 1004–1017. Gotts, S.J., Saad, Z.S., Jo, H.J., Wallace, G.L., Cox, R.W., Martin, A., 2013. The perils of global signal regression for group comparisons: a case study of Autism Spectrum Disorders. Front. Hum. Neurosci. 7. Greicius, M., 2008. Resting-state functional connectivity in neuropsychiatric disorders. Curr. Opin. Neurol. 21, 424–430. Grigg, O., Grady, C.L., 2010. Task-related effects on the temporal and spatial dynamics of resting-state functional connectivity in the default network. PloS ONE 5, e13311. Guo, C.C., Kurth, F., Zhou, J., Mayer, E.A., Eickhoff, S.B., Kramer, J.H., Seeley, W.W., 2012. One-year test–retest reliability of intrinsic connectivity network fMRI in older adults. NeuroImage 61, 1471–1483. Harmelech, T., Preminger, S., Wertman, E., Malach, R., 2013. The day-after effect: long term, Hebbian-like restructuring of resting-state fMRI patterns induced by a single epoch of cortical activation. J. Neurosci. 33, 9488–9497. Harrison, B.J., Pujol, J., Ortiz, H., Fornito, A., Pantelis, C., Yucel, M., 2008. Modulation of brain resting-state networks by sad mood induction. PloS ONE 3, e1794. Hasson, U., Nusbaum, H.C., Small, S.L., 2009. Task-dependent organization of brain regions active during rest. Proc. Natl. Acad. Sci. U. S. A. 106, 10841–10846. Jolles, D.D., van Buchem, M.A., Crone, E.A., Rombouts, S.A., 2013. Functional brain connectivity at rest changes after working memory training. Hum. Brain Mapp. 34, 396–406. Kang, J., Wang, L., Yan, C., Wang, J., Liang, X., He, Y., 2011. Characterizing dynamic functional connectivity in the resting brain using variable parameter regression and Kalman filtering approaches. NeuroImage 56, 1222–1234. Klingner, C.M., Hasler, C., Brodoehl, S., Axer, H., Witte, O.W., 2012. Perceptual plasticity is mediated by connectivity changes of the medial thalamic nucleus. Hum. Brain Mapp. Lemieux, L., Salek-Haddadi, A., Lund, T.E., Laufs, H., Carmichael, D., 2007. Modelling large motion events in fMRI studies of patients with epilepsy. Magn. Reson. Imaging 25, 894–901. Mannfolk, P., Nilsson, M., Hansson, H., Stahlberg, F., Fransson, P., Weibull, A., Svensson, J., Wirestam, R., Olsrud, J., 2011. Can resting-state functional MRI serve as a complement to task-based mapping of sensorimotor function? A test–retest reliability study in healthy volunteers. J. Magn. Reson. Imaging 34, 511–517. Masterton, R.A., Carney, P.W., Jackson, G.D., 2012. Cortical and thalamic resting-state functional connectivity is altered in childhood absence epilepsy. Epilepsy Res. 99, 327–334. Peltier, S.J., LaConte, S.M., Niyazov, D.M., Liu, J.Z., Sahgal, V., Yue, G.H., Hu, X.P., 2005. Reductions in interhemispheric motor cortex functional connectivity after muscle fatigue. Brain Res. 1057, 10–16. Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L., Petersen, S.E., 2012. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage 59, 2142–2154.

R

988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 Q19 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039

13

Please cite this article as: Tailby, C., et al., Resting state functional connectivity changes induced by prior brain state are not network specific, NeuroImage (2014), http://dx.doi.org/10.1016/j.neuroimage.2014.11.037

1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091