- Email: [email protected]
A serial functional connectivity MRI study in healthy individuals assessing the variability of connectivity measures: reduced interhemispheric connectivity in the motor network during continuous performance
Available online at www.sciencedirect.com
Magnetic Resonance Imaging 27 (2009) 1347 – 1359
A serial functional connectivity MRI study in healthy individuals assessing the variability of connectivity measures: reduced interhemispheric connectivity in the motor network during continuous performance Michael Amann⁎, Jochen G. Hirsch, Achim Gass Department of Neurology, University Hospital of Basel, Basel, Switzerland Department of Neuroradiology, University Hospital of Basel, Basel, Switzerland Received 10 December 2008; revised 3 April 2009; accepted 10 May 2009
Abstract To date, little data is available on the reproducibility of functional connectivity MRI (fcMRI) studies. Here, we tested the variability and reproducibility of both the functional connectivity itself and different statistical methods to analyze this phenomenon. In the main part of our study, we repeatedly examined two healthy subjects in 10 sessions over 6 months with fcMRI. Cortical areas involved in motor function were examined under two different cognitive states: during continuous performance (CP) of a flexion/extension task of the fingers of the right hand and while subjects were at rest. Connectivity to left primary motor cortex (lSM1) was calculated by correlation analysis. The resulting correlation coefficients were transformed to z-scores of the standard normal distribution. For each subject, multisession statistical analyses were carried out with the z-score maps of the resting state (RS) and the CP experiments. First, voxel based t tests between the two groups of fcMRI experiments were performed. Second, ROI analyses were carried out for contralateral right SM1 and for supplementary motor area (SMA). For both ROI, mean and maximum z-score were calculated for each experiment. Also, the fraction of significantly (Pb.05) correlated voxels (FCV) in each ROI was calculated. To evaluate the differences between the RS and the CP condition, paired t tests were performed for the mean and maximum z-scores, and Wilcoxon signed ranks tests for matched pairs were carried out for the FCV. All statistical methods and connectivity measures under investigation yielded a distinct loss in left–right SM1 connectivity under the CP condition. For SMA, interindividual differences were apparent. We therefore repeated the fcMRI experiments and the ROI analyses in a group of seven healthy subjects (including the two subjects of the main study). In this substudy, we were able to verify the reduction of left–right SM1 connectivity during unilateral performance. Still, the direction of SMA to lSM1 connectivity change during the CP condition remained undefined as four subjects showed a connectivity increase and three showed a decrease. In summary, we were able to demonstrate a distinct reduction in left– right SM1 synchrony in the CP condition compared to the RS both in the longitudinal and in the multisubject study. This effect was reproducible with all statistical methods and all measures of connectivity under investigation. We conclude that despite intra- and interindividual variability, serial and cross-sectional assessment of functional connectivity reveals stable and reliable results. © 2009 Elsevier Inc. All rights reserved. Keywords: Functional magnetic resonance imaging (fMRI); Functional connectivity; Resting state; Continuous performance; Motor system
1. Introduction Functional connectivity MRI (fcMRI) has become an important tool for exploring the spatio-temporal correlations of spontaneous signal fluctuations in the brain. In resting
⁎ Corresponding author. E-mail address: [email protected] (M. Amann). 0730-725X/$ – see front matter © 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2009.05.016
state (RS) brain, these fluctuations were first measured with fcMRI by Biswal et al. [1]. The authors observed that the low-frequency part (νb0.1 Hz) of the blood oxygen leveldependent (BOLD) signal had a high degree of temporal correlation in regions associated with motor function. Such highly correlated fluctuations were not only reported in the motor network [1–5], but also in the visual system [2,4], in bilateral auditory areas [4], in the language system [6], in the reading circuit [7] and in the “default mode” network [8,9].
1348
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
De Luca et al. [10] found at least five distinct RS networks which were reproducible across different subjects. Other groups have shown that these correlations are modified compared to RS during continuous task performance [6,11–14] and may be altered in neurological and psychiatric disorders like multiple sclerosis [12], Alzheimer's disease [15], depression [16] or autism [17]. Although functional connectivity does not imply a direct causal relationship between cortical regions as it is solely defined as temporal correlations between spatially remote neurophysiological events [18], it has been demonstrated that fcMRI is a valuable tool to describe the status and modulation of cortical interactions. Two main approaches have been employed to identify functional connectivity networks: the first one uses a modelfree technique which decomposes the BOLD signal into a set of independent components [19], while the other approach is based on a correlation analysis towards an element of the putative functionally connected network of interest. The definition of this “seed” region of interest (ROI) can be guided by anatomical atlas knowledge, but in the majority of studies an additional conventional fMRI experiment is used to define the respective seed ROI. In the associated fcMRI experiment, the connectivity hypotheses are then evaluated by correlating the low-frequency signal fluctuations between the seed ROI and the rest of the brain. While in fMRI a model of the hemodynamic response to the task or stimulus can be set up, no such modeling is available in fcMRI. Therefore, statistical inferences are not drawn from an effect size of a certain predictor variable, but only by the strength of temporal correlations itself. In fcMRI, the correlation coefficients are frequently transformed to z-scores of an approximately standard normal distribution using Fisher's transformation to allow for statistical second-level analyses. As in fMRI, the resulting z-score maps are often combined [9,20] or contrasted [21] by group analysis. Based on ROI analysis, other researchers compare the mean z-score [6,7] or the fraction of voxels above a specific significance level [12] or both quantities [22] to draw inferences. The motor network has been well investigated in a vast number of fMRI studies (e.g., Refs. [23–27]). Among others, cortical motor areas include primary (sensori-) motor cortex (SM1) along the central sulcus, supplementary motor areas (SMA) located on the medial surface on the frontal lobes and premotor areas located laterally on the frontal lobes anterior to SM1. Selecting an adequate slice orientation, one can scan the abovementioned areas with sufficient spatial coverage and high temporal resolution. The latter is desirable in fcMRI to avoid signal aliasing of high-frequency BOLD components into the low components of interest, which could mask or affect real correlations. To date, very little data is available on the reproducibility of fcMRI studies. For this reason, we repeatedly examined two subjects over 6 months with fcMRI to test the variability and reproducibility of both the functional
connectivity itself and different statistical methods to investigate this phenomenon. For this purpose, functional connectivity to left primary sensorimotor cortex was analyzed under two different cognitive states: during continuous performance (CP) of a simple unilateral finger motor task and while subjects were at a resting state. We further repeated the fcMRI experiments and the ROI analyses in a group of seven healthy subjects to verify the results of the longitudinal study.
2. Methods 2.1. Subjects Two healthy male subjects without any previous psychiatric, neurological or chronic illness, both 39 years old, participated in the longitudinal study. The Edinburgh Handedness Inventory [28] revealed right handedness for both subjects (handedness score: 87.5 for Subject A; 100 for Subject B). For the multisubject substudy, seven healthy subjects were recruited (including both subjects of the longitudinal study). Subjects (two women) were aged between 28 and 42 years (mean age 35 years) and were all right handed (handedness score 50–100; median 90). Prior to the studies, the participants gave informed consent in accordance to the procedures specified by the local institutional review board. 2.2. Imaging protocol In the main study, both participants underwent 10 MR imaging sessions at intervals of a minimum of 1 week over a time period of 6 months. The measurements were performed on a 3.0-T head scanner (Magnetom Allegra, Siemens Medical, Erlangen, Germany) using the manufacturer's circular polarized transmit–receive head coil. In each imaging session, one fMRI experiment and two fcMRI scans were performed. For each functional scan, a series of T2⁎ sensitive, gradient-recalled echo-planar images were acquired (TE=30 ms, BW=2 kHz/px) with an identical spatial resolution of 3.5×3.5×4 mm 3 . For the fMRI experiment, 25 slices parallel to the inferior border of the corpus callosum were acquired in interleaved order (TR=2 s, flip angle 90°). In the fcMRI experiments, a subset of three slices based on the fMRI volumes were scanned (TR=0.25 s, flip angle 30°), covering putative bilateral primary sensorimotor cortex (SM1). Additionally, a set of 25 proton density-weighted images were acquired in each session (2D turbo spin echo, TR/TE=2.5 s/37 ms; echo train length=9, spatial resolution=0.583×0.583×4 mm3, 25 slices at the same position as in the fMRI scan). Due to a technical upgrade, the multisubject substudy was performed on a 3.0-T whole-body scanner (Magnetom Verio, Siemens Medical, Erlangen, Germany) using the manufacturer's 12-channel receive-only head coil. All imaging parameters were kept identical to the main study.
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
In the fMRI experiment, the subjects performed a blocked design motor task. During the active periods, the subjects were advised to perform a self-paced flexion/extension of the fingers of the right hand with a frequency of roughly 1 Hz; the range of motion was guided in a wooden frame. Performance was controlled visually. Five baseline blocks alternated with four active blocks. In the first and the last baseline block, 15 volumes were acquired, whereas 10 volumes were scanned in the other blocks. Prior to data acquisition, two dummy volumes were acquired to minimize nonequilibrium T1 effects. In the first MR imaging session, the manufacture's built-in online fMRI evaluation was carried out to predefine left SM1. For the fcMRI scans, a subset of three slices were positioned on the putative SM1 of each subject, guided by the results of this preliminary evaluation. In all subsequent sessions, positioning was reproduced as accurately as possible. In the two fcMRI scans, the subjects were advised to close their eyes. During the RS experiment, subjects laid still, while in the CP experiment the subjects were instructed to perform the flexion/extension task continuously with the same frequency and motion range as in the fMRI experiment. In each of the fcMRI runs, 1100 volumes were scanned in 4 min 35 s. As in the fMRI experiment, dummy scans were performed in the first 4 s prior to data acquisition. The order of the two fcMRI experiments was altered randomly over the sessions to prevent systematic effects of attention loss or fatigue. Between all three functional scans, breaks of at least 2 min were kept to avoid possible interactions between the experiments, as it was reported that performance-induced alteration of functional connectivity could persist into subsequent resting periods [29]. 2.3. Data analysis Data evaluation was performed with AFNI [30]. All functional data sets were first adjusted with respect to slice acquisition time. For each session, the fMRI time series was motion corrected to the last volume of the scan. The subsequent fcMRI data sets were also realigned to this reference volume. In none of the series was an internal maximal displacement larger than 1 mm detected. A reference space for each subject was defined by the center volume of the 10 anatomical data sets gained in the 10 different sessions. The functional data sets of each session were registered to this individual space using the registration parameters of the respective anatomical volume. In order to maintain resolution and to reduce spatial autocorrelation, no additional spatial smoothing was performed to the functional data except for the intrinsic effect of realignment. For the functional data sets, brain masks were created using the routine 3dAutomask in AFNI to exclude voxels with low signal levels. The signal time series of the fcMRI experiments were then linearly detrended and low-pass filtered with a cut-off frequency of 0.08 Hz using a rectangular infinite impulse response filter. Due to the short TR of
1349
250 ms in the fcMRI runs, no signal aliasing due to undersampled cardiac or respiratory oxygenation fluctuations was expected. Finally, all functional data sets were intensity normalized with respect to the mean of each individual signal time course. After preprocessing, the fMRI data sets underwent a multiple linear regression (MLR) analysis. A stimulus response model was obtained by convolving the hemodynamic response function with a rectangular function describing the paradigm. In the MLR, the six motion parameters (translation and rotation) were treated as regressors of no interest. The resulting correlation coefficients were transformed to z-values of a normal distribution using Fisher's z-transformation: z = 0:5 log
1+r 1r
ð1Þ
In fMRI data sets, serial autocorrelation broadens and shifts the normal distribution. In order to adjust the z-scores, we therefore performed a least squares fit to the resulting z-score distribution with a mixed-model of the sum of a normal distribution and a Gamma distribution [31]. The z-score of each voxel was then transformed to the z-score of the standard normal distribution by adjusting the mean and the standard deviation of the fitted normal distribution [2]. For each session, the respective fMRI z-score maps were used to define the reference seed regions for the functional connectivity analysis. In the three slices that coincided with those of the fcMRI experiments, a ROI in the left primary sensorimotor (lSM1) cortex was defined by the 27 most significant voxels (nine in each slice) posterior and anterior to the left central sulcus. Reference signal time courses for the fcMRI analyses were then calculated by averaging the signal time courses across all voxels in the activation defined seed ROI for both the RS and the CP run. Correlation was calculated to the mean lSM1 signal time course, while the global brain signal and the six motion parameters were treated as regressors of no interest. Again, the correlation coefficients were transformed to z-scores using Fisher's z-transformation (Eq. (1)) and were corrected by the mixedmodel fit. Without correction, the effects of autocorrelation would be even worse in the fcMRI data due to the temporal filtering and the high data sampling rate [2]. Thus, for each session and each subject, three z-transformed correlation maps were obtained: one adjusted z-score map for the fMRI experiment and two adjusted z-score maps for the RS and the CP run, the last two reflecting functional connectivity to left SM1 for the respective states. 2.4. Effects of fatigue Although a relatively less exhausting task was exercised in the CP runs (4.5 min of restricted finger movement), fatigue likely to reduce interhemispheric connectivity could not be excluded per se [21]. Therefore, an additional statistical test was carried out to the CP data sets of the
1350
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
main study with respect to effects of fatigue. The 1100 time frames of each CP run were separated into three subgroups of early, medium and late phase, and were compared with respect to left SM1 correlations. As this analysis was even more vulnerable to nonequilibrium T1 effects, an additional 20 volumes (spanning a time period of 5 s) at the beginning of each run were discarded. The remaining time frames were divided into three subgroups of 360 volumes. Each subseries was processed in the same way as the data of the whole CP run, including temporal low-pass filtering, intensity normalization, correlation analysis to lSM1 and z-score transformation. An analysis of variance was performed for each subject for the z-transformed correlation maps of the different CP phases, in which the phases were treated as fixed and the session as random variable. 2.5. Multisession composite maps In the main study, a voxel-based multisession statistical analysis was carried out separately for each of the two subjects and for each of the three functional experiments. A two-tailed t test was performed for the z-score maps of the 10 sessions against the null hypothesis to yield a significance map for each experiment. Additionally, a paired t test was
done between the two groups of fcMRI runs. The resulting statistical t-score composite maps were corrected for multiple comparisons with a single voxel probability of Pb.005 and a group adjusted probability of Pb.005; the cluster significance was estimated by Monte Carlo simulation. 2.6. ROI analyses In a multitude of studies concerning functional connectivity, inferences are drawn by comparing statistical quantities (e.g., mean z-score [6,7] or percent of significant activated voxels [12,22]) in specific cortical areas which belong to the regarding functional network. Usually, these target ROIs are anatomically predefined and then refined by the results of a related fMRI experiment. In this work, two areas of the motor network were investigated with regard to lSM1 connectivity: the contralateral (right) SM1 and the SMA. It is well known [26] that unilateral motor tasks induce lateralized activation in primary motor areas, so the approach of refining the ROI in right SM1 (rSM1) by the results of the fMRI experiment would fail in our study. We therefore created a gray matter (GM) mask based on the anatomical data set. The rSM1 was anatomically defined as those GM voxels surrounding the contralateral central sulcus. For the
Fig. 1. Multisession composite maps for Subject A. First row: Significant task-related activation in the blocked design finger motor task. Second and third row: Correlations with lSM1 in RS and during CP of the finger motor task. The t-score maps are corrected for multiple comparisons (Pcorb.005).
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
SMA, anatomical ROI definition was based on Ref. [25], including GM voxels adjacent to the interhemispheric fissure. The posterior border was defined by the tips of precentral sulcus, while the frontal border was determined by the level of anterior commissure, following the functional differentiation into SMA proper and pre-SMA [32]. As the finger motor task does elicit SMA activation, an additional confined SMA ROI could be created for each session separately by excluding all voxels which were not significantly activated in the corresponding fMRI experiment. Because of the lower statistical power of a single fMRI run, a less rigorous threshold was used for the target ROI definition with a single voxel probability of Pb.01. In summary, the ROI analyses of the fcMRI data encompassed GM-defined rSM1, anatomically defined SMA and activation defined SMA; the last one changed in size and shape during the different sessions. For each ROI, mean and maximum z-score were calculated for both the RS and the CP run. Additionally, the fraction of significantly correlated voxels (FCV) in each ROI was calculated. Again, a less stringent threshold (Pb.05) was used because of the reduced contrast-to-noise ratio of the single fcMRI runs. To evaluate differences between the RS and the CP runs, paired t tests were performed for the mean and maximum z-scores, and Wilcoxon signed ranks tests for matched pairs were
1351
carried out for the fraction of significant voxels. In both tests, differences were considered to be significant for Pb.05. As interindividual differences were apparent in the ROI analyses of SMA, we performed a two-way analysis of variance over mean z-scores with subject and condition (RS and CP) as independent factors. To address the question of whether mean z-score and FCV were independent measures, we additionally performed a linear regression of both parameters for each ROI. The data of the multisubject study were evaluated in the same way as those of the longitudinal study. In addition, we wanted to test whether individual lateralization of primary motor activation in the fMRI task correlates with individual interhemispheric SM1 connectivity. For the rSM1-ROI, we therefore performed a linear regression between individual mean z-scores in the fMRI experiments and mean z-scores in the fcMRI experiments. 3. Results 3.1. Effects of fatigue The analysis of variance of the different CP phases (early vs. late, early vs. middle, middle vs. late) revealed no difference for either subject even at a very low significance
Fig. 2. Multisession composite maps for Subject B. First row: Significant task-related activation in the blocked design finger motor task. Second and third row: Correlations with lSM1 in RS and during CP of the finger motor task. The t-score maps are corrected for multiple comparisons (Pcorb.005).
1352
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
Fig. 3. Multisession difference maps between RS and CP correlations with lSM1 (Pcorb.005).
level (corrected Pb.5), thus indicating that no effects of fatigue influenced the CP runs. 3.2. Multisession composite maps The fMRI composite maps of both subjects in the longitudinal study demonstrated distinct motor-related activation in left primary sensorimotor cortex (Brodmann area [BA] 1–3, BA 4), in bilateral premotor areas (BA 6) and in medial supplementary and presupplementary motor areas (BA 6). Also, bilateral superior parietal activation (BA 5 and BA 7) was found in both subjects (Figs. 1 and 2, first row). In right SM1, either no significant activation (Subject A, Fig. 1) or even deactivation (Subject B, Fig. 2) had been observed in
the fMRI composite maps. Additionally, Subject B showed a small cluster of deactivated voxels in posterior SMA. For the RS runs, testing against the null hypothesis revealed correlations to left SM1 in contralateral SM1 encompassing most parts of bilateral central sulcus and the surrounding pre- and postcentral gyrus for both subjects. In addition, bilateral correlation could be observed in superior parietal areas (Figs. 1 and 2, second row). Significant correlations were also found in medial parts of rSM1 and in SMA, but not in pre-SMA. In Subject B, additional bilateral correlations in premotor areas were observed. Subject A showed a cluster of anticorrelated voxels in right middle frontal gyrus; in Subject B, right mid-frontal and right superior frontal anticorrelation was observed and a cluster of
Table 1 Mean and maximum z-scores in rSM1 for both subjects in the 10 fcMRI experiments, representing correlations with lSM1 Session
Subject A
Subject B
Mean z
1 2 3 4 5 6 7 8 9 10 Mean S.D. P Value
Maximum z
Mean z
Maximum z
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
1.003 0.921 0.838 0.974 0.787 0.918 0.259 0.657 0.594 0.803 0.775 0.224 6.2E−04
0.206 0.174 0.007 0.480 0.564 0.232 0.295 0.173 0.364 0.519 0.301 0.178
4.699 4.021 4.097 3.885 3.564 4.447 2.739 2.992 3.021 3.850 3.731 0.647 .001
3.053 2.340 2.510 2.899 2.642 2.646 2.844 2.513 2.957 2.567 2.697 0.230
0.777 1.001 1.031 0.776 0.733 1.367 0.841 1.068 0.932 0.321 0.885 0.273 6.2E−05
0.605 0.559 0.406 0.461 0.216 0.327 0.085 0.035 0.306 -0.462 0.254 0.313
3.470 4.784 3.495 3.657 3.266 4.163 3.214 3.637 3.298 2.508 3.549 0.603 .003
2.792 2.456 2.171 2.356 3.229 2.092 2.939 2.488 2.893 2.071 2.549 0.397
Here, the definition of rSM1 was based on a GM mask. The significance of the differences between both states was calculated by a paired t test.
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359 Table 2 Longitudinal study: FCVs in rSM1 Session
Subject A
Table 4 Longitudinal study: FCV for SMA (GM-based ROI definition) Subject B
Session
Subject A
Resting state Continuous task Resting state Continuous task 1 2 3 4 5 6 7 8 9 10 Median P Value
0.165 0.109 0.143 0.120 0.083 0.109 0.026 0.064 0.068 0.094 0.102 1.95E−03
0.023 0.019 0.015 0.019 0.041 0.019 0.023 0.011 0.049 0.034 0.021
1353
0.067 0.130 0.141 0.100 0.081 0.200 0.081 0.144 0.070 0.041 0.091 1.95E−03
0.059 0.015 0.004 0.007 0.022 0.011 0.004 0.015 0.015 0.011 0.013
1 2 3 4 5 6 7 8 9 10 Median P Value
A voxel was considered to be significantly correlated with rSM1 at a threshold of Pb.05. The definition of rSM1 was GM based; statistical differences between the RS and CP were calculated with Wilcoxon signed ranks test for matched pairs.
anticorrelated voxels were found in pre-SMA. In contrast to the RS, the composite maps of continuous motor task performance (Figs. 1 and 2, third row) demonstrated a significant reduction of interhemispheric correlation in both subjects. Little (Subject A) and no significant correlations (Subject B) to lSM1 were found in the contralateral hemisphere. Compared to the RS, correlations to lSM1 were limited to smaller areas in superior parietal cortex and also in lSM1 itself. In both subjects, small clusters of anticorrelated voxel were found in the left mid-frontal lobe. The difference maps between both states are shown in Fig. 3. As already seen in the composite maps, significant higher RS correlations were found in contralateral SM1 for both subjects. Also, higher RS correlations were observed in contralateral superior parietal cortex and in the lateral aspects of ipsilateral SM1. In Subject B, RS correlations were
Subject B
Resting state Continuous task Resting state Continuous task 0.058 0.078 0.104 0.110 0.097 0.104 0.039 0.117 0.084 0.078 0.091 N.5
0.045 0.013 0.032 0.169 0.026 0.214 0.104 0.149 0.078 0.201 0.091
0.020 0.160 0.093 0.040 0.200 0.100 0.120 0.100 0.047 0.120 0.100 .023
0.020 0.027 0.060 0.027 0.113 0.040 0.067 0.080 0.080 0.080 0.063
Statistical differences between the states were calculated with Wilcoxon signed ranks test for matched pairs.
significantly lower in pre-SMA and in a small area located in the right middle frontal gyrus. 3.3. ROI analyses In the longitudinal study, anatomically defined ROI in rSM1 encompassed 266 voxels for Subject A and 270 voxels for Subject B. The mean z-scores, maximum z-scores and FCV were significantly higher in RS than in CP (Tables 1 and 2) for both subjects. The size of the anatomically defined SMA-ROI was 154 voxels for Subject A and 150 voxels for Subject B. Comparison of mean and maximum z-scores yielded no statistical differences between RS and CP in both subjects (Table 3). By contrast, Subject B had a significantly higher FCV in the RS (Table 4). The activation-defined SMA regions were considerably smaller than the anatomically based ROI. In Subject A, the ROI size varied between 21 and
Table 3 Longitudinal study: mean and maximum z-scores in the SMA, representing correlations with lSM1 Session
Subject A
Subject B
Mean z
1 2 3 4 5 6 7 8 9 10 Mean S.D. P Value
Maximum z
Mean z
Maximum z
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
0.383 0.304 0.461 0.389 0.694 0.266 0.235 0.754 0.325 0.347 0.416 0.175 .184
0.461 0.396 0.346 0.729 0.225 0.807 0.365 0.719 0.741 0.736 0.553 0.213
3.345 3.154 3.989 3.854 3.733 4.014 2.667 3.857 3.610 3.193 3.542 0.439 .223
2.681 2.564 2.917 3.633 2.917 3.558 3.393 4.151 3.336 3.773 3.292 0.511
0.271 0.581 0.713 0.144 0.922 0.431 0.599 0.734 0.335 0.393 0.512 0.239 .421
0.267 0.420 0.355 0.330 0.598 0.459 0.494 0.532 0.589 0.521 0.457 0.112
2.253 3.644 4.521 2.533 4.534 3.276 3.605 4.072 3.340 2.759 3.454 0.786 .346
2.481 2.535 3.422 2.854 3.651 2.304 3.615 3.054 4.615 3.320 3.185 0.695
The definition of the SMA is based on a GM mask. Differences between RS and PC were calculated by a paired t test.
1354
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
Table 5 Longitudinal study: mean and maximum z-scores in the SMA, representing correlations with lSM1 Session
Subject A
Subject B
Mean z
1 2 3 4 5 6 7 8 9 10 Mean S.D. P Value
Maximum z
Mean z
Maximum z
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
0.196 0.894 0.825 0.819 1.420 0.604 0.637 1.018 0.421 0.687 0.752 0.334 .096
0.865 0.798 0.548 1.771 0.390 1.862 1.042 1.509 1.428 1.377 1.159 0.506
2.286 2.578 3.797 2.787 3.733 4.014 2.651 3.221 2.621 2.607 3.029 0.614 .984
2.681 1.716 2.425 3.633 1.702 3.558 3.393 4.151 3.336 3.773 3.037 0.860
0.127 1.294 1.423 0.644 0.937 1.577 1.756 1.713 0.898 1.556 1.192 0.529 .067
0.443 0.624 0.763 0.679 1.168 1.090 1.304 1.159 1.092 0.884 0.921 0.284
2.253 3.644 4.521 2.352 4.440 3.276 3.605 4.072 3.340 2.669 3.417 0.807 .014
2.481 2.535 2.030 1.677 2.669 2.304 3.615 2.204 2.865 2.846 2.523 0.532
The definition of the session-specific SMA-ROI is defined by the voxels that were significantly (Pb.01) activated in the respective fMRI experiment.
57 voxels over the different sessions; in Subject B, the range was between 13 and 41 voxels. Subject B demonstrated significantly higher RS maximum z-scores in the activationdefined ROI. In this subject, a trend towards higher RS mean z-scores was also observed. An even weaker trend was seen in the mean z-scores of Subject A; here the mean z-scores in the continuous task were slightly higher than those of the RS (Table 5). Comparing FCV, the difference between RS and CP was again significant for Subject B but not for Subject A (Table 6). The analysis of variance of mean z-scores in SMA revealed no significant difference between subjects and conditions, but significant interaction between both factors for the activation-defined SMA (P=.017), or respectively a trend towards significance for the anatomically defined SMA (P=.12). FCV and mean z-scores correlated significantly in anatomically defined rSM1 and in activation-defined SMA for both subjects (Figs. 4 and 5). In the anatomically defined
Table 6 Longitudinal study: FCV for the activation-defined SMA ROI Session
Subject A
Subject B
SMA, correlation between FCV and mean z-scores was found for Subject A, but not for Subject B (Fig. 5, first row). In the multisubject study, the extent of the anatomically defined rSM1 varied between 173 and 274 voxels. For rSM1, group mean z-scores and FCV were significantly higher in RS; also, maximum z-scores were higher in RS but the difference missed significance (P=.13) due to one outlier (Tables 7 and 9, second and third column). The individual GM-defined SMA varied between 81 and 199 voxels; the individual activation-defined SMA, between 12 and 99 voxels. For both the anatomically and the activation-defined SMA-ROI, no measure revealed significant RS-CP difference (Tables 8 and 9). In all ROI under evaluation, FCV and mean z-scores correlated significantly in the fcMRI experiments (Fig. 6). In rSM1, we found a significant correlation between individual mean z-score in the fMRI experiment and mean z-score in CP (P=.036) (Fig. 7), but not between the z-score in fMRI and in RS (P=.30). In the respective clustercorrected (Pb.01) z-score maps of the fMRI experiments, five subjects showed deactivation in rSM1, one subject showed activation and in one subject neither significant activation nor deactivation was observed.
Resting state Continuous task Resting state Continuous task 1 2 3 4 5 6 7 8 9 10 Median P Value
0.071 0.095 0.114 0.149 0.242 0.173 0.070 0.194 0.116 0.050 0.115 N.2
0.095 0.000 0.057 0.277 0.030 0.462 0.193 0.250 0.209 0.300 0.201
0.073 0.273 0.238 0.200 0.267 0.167 0.333 0.278 0.118 0.308 0.252 3.91E−03
0.024 0.061 0.095 0.000 0.200 0.056 0.143 0.111 0.118 0.154 0.103
The statistical differences between the states were calculated with Wilcoxon signed ranks test for matched pairs.
4. Discussion The main focus of this study was the evaluation of the temporal variability of functional connectivity and the reproducibility of the results of different statistical analysis methods. For this purpose, we investigated the cortical motor system, with focus on the functional connectivity, of two subjects who were repeatedly scanned over a period of half a year. One drawback of such a two-subject analysis (or rather a twofold single-subject analysis) is the limited possibility to draw inferences onto a larger population. We therefore
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
1355
Fig. 4. Linear regression between mean z-score and FCVs in a GM-defined rSM1-ROI over the 10 sessions of the longitudinal study. Magenta squares: Continuous performance; blue triangles: resting state. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
extended the experiments and the ROI analyses in a group of seven healthy subjects (including the two subjects of the main study). But even in the longitudinal study, some of the effects found parallel the findings of other, group-based studies. The multisession t-score maps obtained from the fMRI runs showed the expected activation pattern for a right-finger motor task with prominent activation in left SM1, in medial SMA and in bilateral premotor areas. We found deactivation in right SM1 in one subject and no activation in the other. This different characteristic was not a mere statistical effect.
Even with a less conservative threshold of Pb.05, Subject A showed no deactivation in rSM1 at the multisession level, while this was the case in Subject B. In the ROI analysis of the multisubject study, one subject (who was not Subject A) had a positive mean z-score in the rSM1-ROI, while the other subjects had negative mean z-scores (Fig. 7). Such interindividual differences in SM1 activation ipsilateral to the performing limp have been reported in various other studies investigating simple unilateral motor tasks. In Ref. [33], for example, ipsilateral deactivation was observed during a simple right-finger opposition task in 13 of 23
Fig. 5. Linear regression between mean z-score and FCV in SMA for the longitudinal study. In the first row, the SMA-ROI is defined by a GM mask; in the second row, SMA is defined by voxels which were significantly activated in the respective fMRI experiment (Pb.01). Magenta squares: Continuous performance; blue triangles: resting state. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
1356
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
Table 7 Cross-sectional study: mean and maximum z-scores in the GM-defined rSM1, representing correlations with lSM1 Subject
SM1r-GM Mean z
Maximum z
Resting state Continuous task Resting state Continuous task A B C D E F G P Value
1.334 0.716 0.841 0.838 0.645 0.764 0.665 .002
0.783 0.538 0.282 0.599 0.061 0.513 −0.069
4.619 3.044 3.054 3.205 2.857 2.908 2.857 .133
3.192 3.645 2.024 3.116 2.363 2.674 2.513
subjects, while the other subjects showed neither significant activation nor deactivation. Here, we demonstrated that such interindividual differences also exist at a multisession level, which points out that these differences are physiologically based. In the analysis of the multisession fcMRI experiments, the composite maps of both subjects under investigation showed considerable differences between RS and CP. In RS, symmetric correlation was found in bilateral SM1 and in superior parietal cortex. This interhemispheric connectivity was almost vanished in unilateral CP. Furthermore, correlation was diluted in parts of SM1 ipsilateral to the seed ROI, especially in the lateral aspects of lSM1 (see also the difference t-score maps in Fig. 3). Peltier et al. [21] demonstrated a significant decrease in interhemispheric SM1 connectivity after the performance of a repetitive unilateral handgrip task due to fatigue. The task used here for CP, finger flexion/extension during 4.5 min, was considerably less exhausting than the task used in Ref. [21] (20 min of gripping a device). In our study, we can exclude the possibility of a fatigue-induced reduction of connectivity as the analysis of variance between the different phases of the CP runs revealed no evidence of fatigue even at very low levels of significance. The interhemispheric connectivity descent in CP was also apparent in the ROI analyses both in the longitudinal and in
the multisubject study. In rSM1, mean z-scores, maximum z-scores and FCV were significantly reduced in CP compared to RS. Furthermore, reduction in mean z-score was accompanied by a decrease of significantly correlated voxels. In the ROI analyses of the SMA, interindividual differences were apparent. Based on the evaluation of the activation-defined ROI, Subject A had a weak trend of higher SMA to lSM1 connectivity in CP which was only observable in the comparison of mean z-scores, but not in FCV or maximum z-scores. In contrast, Subject B had higher connectivity in RS, which was significant in FCV and maximum z-scores and reached almost significance in mean z-scores. In both subjects, there was a strong correlation between mean z-score and FCV (Fig. 5, second row). Based on the analysis of the activation-defined ROI, Subject B showed a reduction of connectivity between SMA and lSM1 in CP, while Subject A did not. Analysis of variance underlines the significance of the interaction between subject and fcMRI condition. In the cross-sectional study, we found additionally in four subjects connectivity increase and in three subjects connectivity reduction in CP. So the direction of SMA to lSM1 connectivity change during our CP condition remained undefined. In the longitudinal study, analyses of the anatomically defined SMA-ROI revealed only one significant difference, namely, Subject B had a higher FCV in RS than in CP. All other differences remained nonsignificant. Interestingly, a correlation between mean z-score and FCV could still be observed in Subject A. Depending on the imaging session and subject, the anatomically defined SMA was three to 11 times larger than the respective activation-defined ROI. Hence, GM voxels that are not directly related to the functional network under evaluation were likely to be included into analysis. Due to this nonspecificity of the anatomically determined ROI, changes in connectivity may be diluted which are otherwise observable in more confined ROI. In SM1, the effect of connectivity change between RS and CP is much more obvious. Even if the anatomically defined ROI had encompassed non–motor-related voxels, the reduced connectivity in CP still remained significant due to the large effect in motor-specific voxels. Even so, the results of our
Table 8 Cross-sectional study: mean and maximum z-scores in the GM and in the activation-defined SMA, representing correlations with lSM1 Subject
SMA-Act
SMA-GM
Mean z
A B C D E F G P Value
Maximum z
Mean z
Maximum z
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
1.559 0.964 0.809 0.946 0.647 1.611 0.535 .779
0.991 1.253 1.399 0.692 1.228 1.032 0.876
2.822 2.697 3.673 3.178 2.655 3.212 1.878 .476
2.110 3.670 4.373 3.475 2.387 2.515 3.238
0.254 0.557 0.707 0.826 0.517 0.950 0.288 .850
0.325 0.506 1.192 0.484 0.234 0.778 0.430
2.822 2.697 3.673 3.178 2.655 3.212 1.878 .268
2.796 3.670 4.373 3.475 2.387 2.515 3.238
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
1357
Table 9 Cross-sectional study: FCV for the three different ROIs Subject
A B C D E F G P Value
rSM1r-GM
SM-Act
SMA-GM
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
0.271 0.160 0.158 0.163 0.099 0.070 0.103 .016
0.097 0.090 0.015 0.063 0.043 0.042 0.017
0.333 0.217 0.145 0.179 0.100 0.294 0.000 N.4
0.111 0.261 0.329 0.074 0.200 0.147 0.133
0.071 0.066 0.139 0.163 0.048 0.228 0.000 N.5
0.032 0.074 0.277 0.065 0.065 0.098 0.074
The statistical differences between the states were calculated with Wilcoxon signed ranks test for matched pairs.
ROI analyses of SMA strongly suggest that it is preferable to work with activation-defined ROI like in Refs. [6,12,22], rather than to use purely anatomically based regions. The regression analysis of mean z-score and FCV in this work confirmed the results of the study of Newton et al. [22], which had investigated the modulation of functional connectivity of SM1 to an audibly paced finger-tapping task with increasing demand. The authors had also reported that the changes in FCV within different ROI (SMA, cerebellum and right auditory cortex) mirrored the changes in mean correlation of the regions (which is equivalent to the mean z-score used in our study). With the exception of the anatomically defined SMA for Subject B in the longitudinal study, it was also demonstrated that both measures correlate
with each other, even if there was no significant dependence on the cognitive states in a specific region (Subject A, activation-defined SMA). Our results therefore support the statement of Newton et al. [22] that it is the number of correlated voxels that drives the observed changes in overall functional connectivity. In both the longitudinal and the cross-sectional study, we found a reduction in bilateral SM1 connectivity during CP, which is in contrast to other fcMRI experiments evaluating connectivity under different mental states, where increased connectivity during performance was observed. These studies investigated either nonmotor systems (e.g., the language-related network [6]) or used bimanual motor tasks [12,14], where the corresponding fMRI task itself
Fig. 6. Multisubject analysis of fcMRI: Linear regression between mean z-score and FCV in rSM1 and in SMA. SMA-act: SMA-ROI is defined by significantly activated voxels in fMRI. SMA-GM: SMA is defined be a GM mask. Magenta squares: Continuous performance; blue triangles: resting state. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
1358
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
Fig. 7. Linear regression between individual mean z-scores in the fMRI task and mean z-scores of the respective CP fcMRI experiment. Mean z-scores are calculated over the right SM1-ROI. Correlation reached significance (P=.036).
elicited bilateral SM1 activation. Here, we were able to demonstrate that unilateral SM1 activation during a conventional fMRI experiment is paralleled by desynchronization of bilateral SM1 during steady-state performance of the same task. In addition, we found indications that individual PC connectivity decrease correlates with stronger suppression of collateral SM1 activation during the unilateral motor task (Fig. 7). In the longitudinal study, the measures of connectivity for each subject showed some variations over the sessions. In the first session, for example, Subject B has almost equal rSM1/ lSM1 correlations in both RS and CP, despite the fact that reduction of bilateral SM1 connectivity during unilateral performance is highly significant at multisession level. This indicates that functional connectivity is not a static measure, even for the same individual in comparable cognitive states. Despite this variability, possibly reflecting individual behavioral changes, all measures of functional connectivity investigated here were stable at group level. In our study, we were able to demonstrate a distinct reduction in left–right SM1 synchrony in the CP state compared to the RS in both individuals in the longitudinal study and in all subjects in the multisubject study. This effect was reproducible over half a year with all statistical methods and all measures of connectivity under investigation. Despite intra- and interindividual variability, serial and crosssectional assessments of functional connectivity reveal stable and reliable results. Acknowledgments The authors thank the volunteers for participating in our studies, as well as the two anonymous reviewers of this article for their valuable suggestions and comments. References [1] Biswal BB, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in the motor cortex of the resting human brain using echoplanar MRI. Magn Reson Med 1995;34:537–41.
[2] Lowe MJ, Mock BJ, Sorenson JA. Functional connectivity in single and multislice echoplanar imaging using resting-state fluctuations. NeuroImage 1998;7:119–32. [3] Xiong J, Parsons LM, Gao JH, Fox PT. Interregional connectivity to primary motor cortex revealed using MRI resting state images. Hum Brain Mapp 1999;8:151–6. [4] Cordes D, Haughton VM, Arfanakis K, Wendt GJ, Turski PA, Moritz CH, et al. Mapping functionally related regions of brain with functional connectivity MR imaging. Am J Neuroradiol 2000;21:1636–44. [5] De Luca M, Smith S, De Stefano N, Federico A, Matthews PM. Blood oxygenation level dependent contrast resting state networks are relevant to functional activity in the neocortical sensorimotor system. Exp Brain Res 2005;167:587–94. [6] Hampson M, Peterson BS, Skudlarski P, Gatenby JC, Gore JC. Detection of functional connectivity using temporal correlations in MR images. Hum Brain Mapp 2002;15:247–62. [7] Hampson M, Tokoglu F, Sun Z, Schafer RJ, Skudlarski P, Gore JC, et al. Connectivity-behavior analysis reveals that functional connectivity between left BA39 and Broca's area varies with reading ability. NeuroImage 2006;31:513–9. [8] Greicius MD, Krasnow B, Reiss AL, Menon V. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis. Proc Natl Acad Sci U S A 2003;100:253–8. [9] Fransson P. Spontaneous low-frequency BOLD signal fluctuations: an fMRI investigation of the resting-state default mode of brain function hypothesis. Hum Brain Mapp 2005;26:15–29. [10] De Luca M, Beckman CF, De Stefano N, Matthews PM, Smith SM. FMRI resting state networks define distinct modes of long-distance interactions in the human brain. NeuroImage 2006;29:1359–67. [11] Lowe MJ, Dzemidzic M, Lurito JT, Mathews VP, Phillips MD. Correlations in low-frequency BOLD fluctuations reflect corticocortical connections. NeuroImage 2000;12:582–7. [12] Lowe MJ, Phillips MD, Lurito JT, Mattson D, Dzemidzic M, Mathews VP. Multiple sclerosis: low-frequency temporal blood oxygen leveldependent fluctuations indicate reduced functional connectivity initial results. Radiology 2002;224:184–92. [13] Hampson M, Olson IR, Leung HC, Skudlarski P, Gore JC. Changes in functional connectivity of human MT/V5 with visual motion input. Neuroreport 2004;15:1315–9. [14] Morgan VL, Price RR. The effect of sensorimotor activation on functional connectivity mapping with MRI. Magn Reson Imaging 2004;22:1069–75. [15] Grecius MD, Srivastava G, Reiss AL, Menon V. Default-mode network activity distinguishes Alzheimer's disease from healthy aging: evidence from functional MRI. Proc Natl Acad Sci U S A 2004;101: 4637–42. [16] Anand A, Li Y, Wang Y, Wu J, Gao S, Bukhari L, et al. Activity and connectivity of brain mood regulating circuit in depression: a functional magnetic resonance study. Biol Psychiatry 2005;57:1079–88. [17] Kennedy DP, Courchesne E. The intrinsic functional organization of the brain is altered in autism. NeuroImage 2008;39:1877–85. [18] Friston KJ, Frith CD, Liddle PF, Frackowiak RS. Functional connectivity: the principal-component analysis of large (PET) data sets. J Cereb Blood Flow Metab 1993;13:5–14. [19] Beckmann CF, DeLuca M, Devlin JT, Smith SM. Investigations into resting-state connectivity using independent component analysis. Philos Trans R Soc B 2005;360:1001–13. [20] Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, Raichle ME. The human brain is intrinsically organized into dynamic, anticorrelated functional networks. Proc Natl Acad Sci U S A 2005; 102:9673–8. [21] Peltier SJ, LaConte SM, Niyazov DM, Liu JZ, Sahgal V, Yue GH, et al. Reductions in interhemispheric motor cortex functional connectivity after muscle fatigue. Brain Res 2005;1057:10–6. [22] Newton AT, Morgan VL, Gore CG. Task demand modulation of steady-state functional connectivity to primary motor cortex. Hum Brain Mapp 2007;28:663–72.
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359 [23] Dettmers C, Fink GR, Lemon RN, Stephan KM, Passingham RE, Silbersweig D, et al. Relation between cerebral activity and force in the motor areas of the human brain. J Neurophysiol 1995;74: 802–15. [24] Yousry TA, Schmid UD, Jassoy AG, Schmidt D, Eisner WE, Reulen HJ, et al. Topography of the cortical motor hand area: prospective study with functional MR imaging and direct motor mapping at surgery. Radiology 1995;195:23–9. [25] Picard N, Strick PL. Motor areas of the medial wall: a review of their location and functional activation. Cereb Cortex 1996;6:342–53. [26] Wexler BE, Fulbright RK, Lacadie CM, Skudlarski P, Kelz MB, Constable RT, et al. An fMRI study of the human cortical motor system response to increasing functional demands. Magn Reson Imaging 1997;15:385–96. [27] Smith JF, Chen K, Johnson S, Morrone-Strupinsky J, Reiman EM, Nelson A, et al. Network analysis of single-subject fMRI during a finger opposition task. NeuroImage 2006;32:325–32.
1359
[28] Oldfield RC. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 1971;9:97–113. [29] Waites AB, Stanislavsky A, Abbott DF, Jackson GD. Effect of prior cognitive state on resting state networks measured with functional connectivity. Hum Brain Mapp 2005;24:59–68. [30] Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996;29:162–73. [31] Hartvig NV, Jensen JL. Spatial mixture modeling of fMRI data. Hum Brain Mapp 2000;11:233–48. [32] Johansen-Berg H, Behrens TE, Robson MD, Drobnjak I, Rushworth MF, Brady JM, et al. Changes in connectivity profiles define functionally distinct regions in human medial frontal cortex. Proc Natl Acad Sci U S A 2004;101:13335–40. [33] Nirkko AC, Ozdoba C, Redmond SM, Bürki M, Schroth G, Hess CW, et al. Different ipsilateral representations for distal and proximal movements in the sensorimotor cortex: activation and deactivation patterns. NeuroImage 2001;13:825–35.
Magnetic Resonance Imaging 27 (2009) 1347 – 1359
A serial functional connectivity MRI study in healthy individuals assessing the variability of connectivity measures: reduced interhemispheric connectivity in the motor network during continuous performance Michael Amann⁎, Jochen G. Hirsch, Achim Gass Department of Neurology, University Hospital of Basel, Basel, Switzerland Department of Neuroradiology, University Hospital of Basel, Basel, Switzerland Received 10 December 2008; revised 3 April 2009; accepted 10 May 2009
Abstract To date, little data is available on the reproducibility of functional connectivity MRI (fcMRI) studies. Here, we tested the variability and reproducibility of both the functional connectivity itself and different statistical methods to analyze this phenomenon. In the main part of our study, we repeatedly examined two healthy subjects in 10 sessions over 6 months with fcMRI. Cortical areas involved in motor function were examined under two different cognitive states: during continuous performance (CP) of a flexion/extension task of the fingers of the right hand and while subjects were at rest. Connectivity to left primary motor cortex (lSM1) was calculated by correlation analysis. The resulting correlation coefficients were transformed to z-scores of the standard normal distribution. For each subject, multisession statistical analyses were carried out with the z-score maps of the resting state (RS) and the CP experiments. First, voxel based t tests between the two groups of fcMRI experiments were performed. Second, ROI analyses were carried out for contralateral right SM1 and for supplementary motor area (SMA). For both ROI, mean and maximum z-score were calculated for each experiment. Also, the fraction of significantly (Pb.05) correlated voxels (FCV) in each ROI was calculated. To evaluate the differences between the RS and the CP condition, paired t tests were performed for the mean and maximum z-scores, and Wilcoxon signed ranks tests for matched pairs were carried out for the FCV. All statistical methods and connectivity measures under investigation yielded a distinct loss in left–right SM1 connectivity under the CP condition. For SMA, interindividual differences were apparent. We therefore repeated the fcMRI experiments and the ROI analyses in a group of seven healthy subjects (including the two subjects of the main study). In this substudy, we were able to verify the reduction of left–right SM1 connectivity during unilateral performance. Still, the direction of SMA to lSM1 connectivity change during the CP condition remained undefined as four subjects showed a connectivity increase and three showed a decrease. In summary, we were able to demonstrate a distinct reduction in left– right SM1 synchrony in the CP condition compared to the RS both in the longitudinal and in the multisubject study. This effect was reproducible with all statistical methods and all measures of connectivity under investigation. We conclude that despite intra- and interindividual variability, serial and cross-sectional assessment of functional connectivity reveals stable and reliable results. © 2009 Elsevier Inc. All rights reserved. Keywords: Functional magnetic resonance imaging (fMRI); Functional connectivity; Resting state; Continuous performance; Motor system
1. Introduction Functional connectivity MRI (fcMRI) has become an important tool for exploring the spatio-temporal correlations of spontaneous signal fluctuations in the brain. In resting
⁎ Corresponding author. E-mail address: [email protected] (M. Amann). 0730-725X/$ – see front matter © 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.mri.2009.05.016
state (RS) brain, these fluctuations were first measured with fcMRI by Biswal et al. [1]. The authors observed that the low-frequency part (νb0.1 Hz) of the blood oxygen leveldependent (BOLD) signal had a high degree of temporal correlation in regions associated with motor function. Such highly correlated fluctuations were not only reported in the motor network [1–5], but also in the visual system [2,4], in bilateral auditory areas [4], in the language system [6], in the reading circuit [7] and in the “default mode” network [8,9].
1348
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
De Luca et al. [10] found at least five distinct RS networks which were reproducible across different subjects. Other groups have shown that these correlations are modified compared to RS during continuous task performance [6,11–14] and may be altered in neurological and psychiatric disorders like multiple sclerosis [12], Alzheimer's disease [15], depression [16] or autism [17]. Although functional connectivity does not imply a direct causal relationship between cortical regions as it is solely defined as temporal correlations between spatially remote neurophysiological events [18], it has been demonstrated that fcMRI is a valuable tool to describe the status and modulation of cortical interactions. Two main approaches have been employed to identify functional connectivity networks: the first one uses a modelfree technique which decomposes the BOLD signal into a set of independent components [19], while the other approach is based on a correlation analysis towards an element of the putative functionally connected network of interest. The definition of this “seed” region of interest (ROI) can be guided by anatomical atlas knowledge, but in the majority of studies an additional conventional fMRI experiment is used to define the respective seed ROI. In the associated fcMRI experiment, the connectivity hypotheses are then evaluated by correlating the low-frequency signal fluctuations between the seed ROI and the rest of the brain. While in fMRI a model of the hemodynamic response to the task or stimulus can be set up, no such modeling is available in fcMRI. Therefore, statistical inferences are not drawn from an effect size of a certain predictor variable, but only by the strength of temporal correlations itself. In fcMRI, the correlation coefficients are frequently transformed to z-scores of an approximately standard normal distribution using Fisher's transformation to allow for statistical second-level analyses. As in fMRI, the resulting z-score maps are often combined [9,20] or contrasted [21] by group analysis. Based on ROI analysis, other researchers compare the mean z-score [6,7] or the fraction of voxels above a specific significance level [12] or both quantities [22] to draw inferences. The motor network has been well investigated in a vast number of fMRI studies (e.g., Refs. [23–27]). Among others, cortical motor areas include primary (sensori-) motor cortex (SM1) along the central sulcus, supplementary motor areas (SMA) located on the medial surface on the frontal lobes and premotor areas located laterally on the frontal lobes anterior to SM1. Selecting an adequate slice orientation, one can scan the abovementioned areas with sufficient spatial coverage and high temporal resolution. The latter is desirable in fcMRI to avoid signal aliasing of high-frequency BOLD components into the low components of interest, which could mask or affect real correlations. To date, very little data is available on the reproducibility of fcMRI studies. For this reason, we repeatedly examined two subjects over 6 months with fcMRI to test the variability and reproducibility of both the functional
connectivity itself and different statistical methods to investigate this phenomenon. For this purpose, functional connectivity to left primary sensorimotor cortex was analyzed under two different cognitive states: during continuous performance (CP) of a simple unilateral finger motor task and while subjects were at a resting state. We further repeated the fcMRI experiments and the ROI analyses in a group of seven healthy subjects to verify the results of the longitudinal study.
2. Methods 2.1. Subjects Two healthy male subjects without any previous psychiatric, neurological or chronic illness, both 39 years old, participated in the longitudinal study. The Edinburgh Handedness Inventory [28] revealed right handedness for both subjects (handedness score: 87.5 for Subject A; 100 for Subject B). For the multisubject substudy, seven healthy subjects were recruited (including both subjects of the longitudinal study). Subjects (two women) were aged between 28 and 42 years (mean age 35 years) and were all right handed (handedness score 50–100; median 90). Prior to the studies, the participants gave informed consent in accordance to the procedures specified by the local institutional review board. 2.2. Imaging protocol In the main study, both participants underwent 10 MR imaging sessions at intervals of a minimum of 1 week over a time period of 6 months. The measurements were performed on a 3.0-T head scanner (Magnetom Allegra, Siemens Medical, Erlangen, Germany) using the manufacturer's circular polarized transmit–receive head coil. In each imaging session, one fMRI experiment and two fcMRI scans were performed. For each functional scan, a series of T2⁎ sensitive, gradient-recalled echo-planar images were acquired (TE=30 ms, BW=2 kHz/px) with an identical spatial resolution of 3.5×3.5×4 mm 3 . For the fMRI experiment, 25 slices parallel to the inferior border of the corpus callosum were acquired in interleaved order (TR=2 s, flip angle 90°). In the fcMRI experiments, a subset of three slices based on the fMRI volumes were scanned (TR=0.25 s, flip angle 30°), covering putative bilateral primary sensorimotor cortex (SM1). Additionally, a set of 25 proton density-weighted images were acquired in each session (2D turbo spin echo, TR/TE=2.5 s/37 ms; echo train length=9, spatial resolution=0.583×0.583×4 mm3, 25 slices at the same position as in the fMRI scan). Due to a technical upgrade, the multisubject substudy was performed on a 3.0-T whole-body scanner (Magnetom Verio, Siemens Medical, Erlangen, Germany) using the manufacturer's 12-channel receive-only head coil. All imaging parameters were kept identical to the main study.
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
In the fMRI experiment, the subjects performed a blocked design motor task. During the active periods, the subjects were advised to perform a self-paced flexion/extension of the fingers of the right hand with a frequency of roughly 1 Hz; the range of motion was guided in a wooden frame. Performance was controlled visually. Five baseline blocks alternated with four active blocks. In the first and the last baseline block, 15 volumes were acquired, whereas 10 volumes were scanned in the other blocks. Prior to data acquisition, two dummy volumes were acquired to minimize nonequilibrium T1 effects. In the first MR imaging session, the manufacture's built-in online fMRI evaluation was carried out to predefine left SM1. For the fcMRI scans, a subset of three slices were positioned on the putative SM1 of each subject, guided by the results of this preliminary evaluation. In all subsequent sessions, positioning was reproduced as accurately as possible. In the two fcMRI scans, the subjects were advised to close their eyes. During the RS experiment, subjects laid still, while in the CP experiment the subjects were instructed to perform the flexion/extension task continuously with the same frequency and motion range as in the fMRI experiment. In each of the fcMRI runs, 1100 volumes were scanned in 4 min 35 s. As in the fMRI experiment, dummy scans were performed in the first 4 s prior to data acquisition. The order of the two fcMRI experiments was altered randomly over the sessions to prevent systematic effects of attention loss or fatigue. Between all three functional scans, breaks of at least 2 min were kept to avoid possible interactions between the experiments, as it was reported that performance-induced alteration of functional connectivity could persist into subsequent resting periods [29]. 2.3. Data analysis Data evaluation was performed with AFNI [30]. All functional data sets were first adjusted with respect to slice acquisition time. For each session, the fMRI time series was motion corrected to the last volume of the scan. The subsequent fcMRI data sets were also realigned to this reference volume. In none of the series was an internal maximal displacement larger than 1 mm detected. A reference space for each subject was defined by the center volume of the 10 anatomical data sets gained in the 10 different sessions. The functional data sets of each session were registered to this individual space using the registration parameters of the respective anatomical volume. In order to maintain resolution and to reduce spatial autocorrelation, no additional spatial smoothing was performed to the functional data except for the intrinsic effect of realignment. For the functional data sets, brain masks were created using the routine 3dAutomask in AFNI to exclude voxels with low signal levels. The signal time series of the fcMRI experiments were then linearly detrended and low-pass filtered with a cut-off frequency of 0.08 Hz using a rectangular infinite impulse response filter. Due to the short TR of
1349
250 ms in the fcMRI runs, no signal aliasing due to undersampled cardiac or respiratory oxygenation fluctuations was expected. Finally, all functional data sets were intensity normalized with respect to the mean of each individual signal time course. After preprocessing, the fMRI data sets underwent a multiple linear regression (MLR) analysis. A stimulus response model was obtained by convolving the hemodynamic response function with a rectangular function describing the paradigm. In the MLR, the six motion parameters (translation and rotation) were treated as regressors of no interest. The resulting correlation coefficients were transformed to z-values of a normal distribution using Fisher's z-transformation: z = 0:5 log
1+r 1r
ð1Þ
In fMRI data sets, serial autocorrelation broadens and shifts the normal distribution. In order to adjust the z-scores, we therefore performed a least squares fit to the resulting z-score distribution with a mixed-model of the sum of a normal distribution and a Gamma distribution [31]. The z-score of each voxel was then transformed to the z-score of the standard normal distribution by adjusting the mean and the standard deviation of the fitted normal distribution [2]. For each session, the respective fMRI z-score maps were used to define the reference seed regions for the functional connectivity analysis. In the three slices that coincided with those of the fcMRI experiments, a ROI in the left primary sensorimotor (lSM1) cortex was defined by the 27 most significant voxels (nine in each slice) posterior and anterior to the left central sulcus. Reference signal time courses for the fcMRI analyses were then calculated by averaging the signal time courses across all voxels in the activation defined seed ROI for both the RS and the CP run. Correlation was calculated to the mean lSM1 signal time course, while the global brain signal and the six motion parameters were treated as regressors of no interest. Again, the correlation coefficients were transformed to z-scores using Fisher's z-transformation (Eq. (1)) and were corrected by the mixedmodel fit. Without correction, the effects of autocorrelation would be even worse in the fcMRI data due to the temporal filtering and the high data sampling rate [2]. Thus, for each session and each subject, three z-transformed correlation maps were obtained: one adjusted z-score map for the fMRI experiment and two adjusted z-score maps for the RS and the CP run, the last two reflecting functional connectivity to left SM1 for the respective states. 2.4. Effects of fatigue Although a relatively less exhausting task was exercised in the CP runs (4.5 min of restricted finger movement), fatigue likely to reduce interhemispheric connectivity could not be excluded per se [21]. Therefore, an additional statistical test was carried out to the CP data sets of the
1350
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
main study with respect to effects of fatigue. The 1100 time frames of each CP run were separated into three subgroups of early, medium and late phase, and were compared with respect to left SM1 correlations. As this analysis was even more vulnerable to nonequilibrium T1 effects, an additional 20 volumes (spanning a time period of 5 s) at the beginning of each run were discarded. The remaining time frames were divided into three subgroups of 360 volumes. Each subseries was processed in the same way as the data of the whole CP run, including temporal low-pass filtering, intensity normalization, correlation analysis to lSM1 and z-score transformation. An analysis of variance was performed for each subject for the z-transformed correlation maps of the different CP phases, in which the phases were treated as fixed and the session as random variable. 2.5. Multisession composite maps In the main study, a voxel-based multisession statistical analysis was carried out separately for each of the two subjects and for each of the three functional experiments. A two-tailed t test was performed for the z-score maps of the 10 sessions against the null hypothesis to yield a significance map for each experiment. Additionally, a paired t test was
done between the two groups of fcMRI runs. The resulting statistical t-score composite maps were corrected for multiple comparisons with a single voxel probability of Pb.005 and a group adjusted probability of Pb.005; the cluster significance was estimated by Monte Carlo simulation. 2.6. ROI analyses In a multitude of studies concerning functional connectivity, inferences are drawn by comparing statistical quantities (e.g., mean z-score [6,7] or percent of significant activated voxels [12,22]) in specific cortical areas which belong to the regarding functional network. Usually, these target ROIs are anatomically predefined and then refined by the results of a related fMRI experiment. In this work, two areas of the motor network were investigated with regard to lSM1 connectivity: the contralateral (right) SM1 and the SMA. It is well known [26] that unilateral motor tasks induce lateralized activation in primary motor areas, so the approach of refining the ROI in right SM1 (rSM1) by the results of the fMRI experiment would fail in our study. We therefore created a gray matter (GM) mask based on the anatomical data set. The rSM1 was anatomically defined as those GM voxels surrounding the contralateral central sulcus. For the
Fig. 1. Multisession composite maps for Subject A. First row: Significant task-related activation in the blocked design finger motor task. Second and third row: Correlations with lSM1 in RS and during CP of the finger motor task. The t-score maps are corrected for multiple comparisons (Pcorb.005).
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
SMA, anatomical ROI definition was based on Ref. [25], including GM voxels adjacent to the interhemispheric fissure. The posterior border was defined by the tips of precentral sulcus, while the frontal border was determined by the level of anterior commissure, following the functional differentiation into SMA proper and pre-SMA [32]. As the finger motor task does elicit SMA activation, an additional confined SMA ROI could be created for each session separately by excluding all voxels which were not significantly activated in the corresponding fMRI experiment. Because of the lower statistical power of a single fMRI run, a less rigorous threshold was used for the target ROI definition with a single voxel probability of Pb.01. In summary, the ROI analyses of the fcMRI data encompassed GM-defined rSM1, anatomically defined SMA and activation defined SMA; the last one changed in size and shape during the different sessions. For each ROI, mean and maximum z-score were calculated for both the RS and the CP run. Additionally, the fraction of significantly correlated voxels (FCV) in each ROI was calculated. Again, a less stringent threshold (Pb.05) was used because of the reduced contrast-to-noise ratio of the single fcMRI runs. To evaluate differences between the RS and the CP runs, paired t tests were performed for the mean and maximum z-scores, and Wilcoxon signed ranks tests for matched pairs were
1351
carried out for the fraction of significant voxels. In both tests, differences were considered to be significant for Pb.05. As interindividual differences were apparent in the ROI analyses of SMA, we performed a two-way analysis of variance over mean z-scores with subject and condition (RS and CP) as independent factors. To address the question of whether mean z-score and FCV were independent measures, we additionally performed a linear regression of both parameters for each ROI. The data of the multisubject study were evaluated in the same way as those of the longitudinal study. In addition, we wanted to test whether individual lateralization of primary motor activation in the fMRI task correlates with individual interhemispheric SM1 connectivity. For the rSM1-ROI, we therefore performed a linear regression between individual mean z-scores in the fMRI experiments and mean z-scores in the fcMRI experiments. 3. Results 3.1. Effects of fatigue The analysis of variance of the different CP phases (early vs. late, early vs. middle, middle vs. late) revealed no difference for either subject even at a very low significance
Fig. 2. Multisession composite maps for Subject B. First row: Significant task-related activation in the blocked design finger motor task. Second and third row: Correlations with lSM1 in RS and during CP of the finger motor task. The t-score maps are corrected for multiple comparisons (Pcorb.005).
1352
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
Fig. 3. Multisession difference maps between RS and CP correlations with lSM1 (Pcorb.005).
level (corrected Pb.5), thus indicating that no effects of fatigue influenced the CP runs. 3.2. Multisession composite maps The fMRI composite maps of both subjects in the longitudinal study demonstrated distinct motor-related activation in left primary sensorimotor cortex (Brodmann area [BA] 1–3, BA 4), in bilateral premotor areas (BA 6) and in medial supplementary and presupplementary motor areas (BA 6). Also, bilateral superior parietal activation (BA 5 and BA 7) was found in both subjects (Figs. 1 and 2, first row). In right SM1, either no significant activation (Subject A, Fig. 1) or even deactivation (Subject B, Fig. 2) had been observed in
the fMRI composite maps. Additionally, Subject B showed a small cluster of deactivated voxels in posterior SMA. For the RS runs, testing against the null hypothesis revealed correlations to left SM1 in contralateral SM1 encompassing most parts of bilateral central sulcus and the surrounding pre- and postcentral gyrus for both subjects. In addition, bilateral correlation could be observed in superior parietal areas (Figs. 1 and 2, second row). Significant correlations were also found in medial parts of rSM1 and in SMA, but not in pre-SMA. In Subject B, additional bilateral correlations in premotor areas were observed. Subject A showed a cluster of anticorrelated voxels in right middle frontal gyrus; in Subject B, right mid-frontal and right superior frontal anticorrelation was observed and a cluster of
Table 1 Mean and maximum z-scores in rSM1 for both subjects in the 10 fcMRI experiments, representing correlations with lSM1 Session
Subject A
Subject B
Mean z
1 2 3 4 5 6 7 8 9 10 Mean S.D. P Value
Maximum z
Mean z
Maximum z
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
1.003 0.921 0.838 0.974 0.787 0.918 0.259 0.657 0.594 0.803 0.775 0.224 6.2E−04
0.206 0.174 0.007 0.480 0.564 0.232 0.295 0.173 0.364 0.519 0.301 0.178
4.699 4.021 4.097 3.885 3.564 4.447 2.739 2.992 3.021 3.850 3.731 0.647 .001
3.053 2.340 2.510 2.899 2.642 2.646 2.844 2.513 2.957 2.567 2.697 0.230
0.777 1.001 1.031 0.776 0.733 1.367 0.841 1.068 0.932 0.321 0.885 0.273 6.2E−05
0.605 0.559 0.406 0.461 0.216 0.327 0.085 0.035 0.306 -0.462 0.254 0.313
3.470 4.784 3.495 3.657 3.266 4.163 3.214 3.637 3.298 2.508 3.549 0.603 .003
2.792 2.456 2.171 2.356 3.229 2.092 2.939 2.488 2.893 2.071 2.549 0.397
Here, the definition of rSM1 was based on a GM mask. The significance of the differences between both states was calculated by a paired t test.
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359 Table 2 Longitudinal study: FCVs in rSM1 Session
Subject A
Table 4 Longitudinal study: FCV for SMA (GM-based ROI definition) Subject B
Session
Subject A
Resting state Continuous task Resting state Continuous task 1 2 3 4 5 6 7 8 9 10 Median P Value
0.165 0.109 0.143 0.120 0.083 0.109 0.026 0.064 0.068 0.094 0.102 1.95E−03
0.023 0.019 0.015 0.019 0.041 0.019 0.023 0.011 0.049 0.034 0.021
1353
0.067 0.130 0.141 0.100 0.081 0.200 0.081 0.144 0.070 0.041 0.091 1.95E−03
0.059 0.015 0.004 0.007 0.022 0.011 0.004 0.015 0.015 0.011 0.013
1 2 3 4 5 6 7 8 9 10 Median P Value
A voxel was considered to be significantly correlated with rSM1 at a threshold of Pb.05. The definition of rSM1 was GM based; statistical differences between the RS and CP were calculated with Wilcoxon signed ranks test for matched pairs.
anticorrelated voxels were found in pre-SMA. In contrast to the RS, the composite maps of continuous motor task performance (Figs. 1 and 2, third row) demonstrated a significant reduction of interhemispheric correlation in both subjects. Little (Subject A) and no significant correlations (Subject B) to lSM1 were found in the contralateral hemisphere. Compared to the RS, correlations to lSM1 were limited to smaller areas in superior parietal cortex and also in lSM1 itself. In both subjects, small clusters of anticorrelated voxel were found in the left mid-frontal lobe. The difference maps between both states are shown in Fig. 3. As already seen in the composite maps, significant higher RS correlations were found in contralateral SM1 for both subjects. Also, higher RS correlations were observed in contralateral superior parietal cortex and in the lateral aspects of ipsilateral SM1. In Subject B, RS correlations were
Subject B
Resting state Continuous task Resting state Continuous task 0.058 0.078 0.104 0.110 0.097 0.104 0.039 0.117 0.084 0.078 0.091 N.5
0.045 0.013 0.032 0.169 0.026 0.214 0.104 0.149 0.078 0.201 0.091
0.020 0.160 0.093 0.040 0.200 0.100 0.120 0.100 0.047 0.120 0.100 .023
0.020 0.027 0.060 0.027 0.113 0.040 0.067 0.080 0.080 0.080 0.063
Statistical differences between the states were calculated with Wilcoxon signed ranks test for matched pairs.
significantly lower in pre-SMA and in a small area located in the right middle frontal gyrus. 3.3. ROI analyses In the longitudinal study, anatomically defined ROI in rSM1 encompassed 266 voxels for Subject A and 270 voxels for Subject B. The mean z-scores, maximum z-scores and FCV were significantly higher in RS than in CP (Tables 1 and 2) for both subjects. The size of the anatomically defined SMA-ROI was 154 voxels for Subject A and 150 voxels for Subject B. Comparison of mean and maximum z-scores yielded no statistical differences between RS and CP in both subjects (Table 3). By contrast, Subject B had a significantly higher FCV in the RS (Table 4). The activation-defined SMA regions were considerably smaller than the anatomically based ROI. In Subject A, the ROI size varied between 21 and
Table 3 Longitudinal study: mean and maximum z-scores in the SMA, representing correlations with lSM1 Session
Subject A
Subject B
Mean z
1 2 3 4 5 6 7 8 9 10 Mean S.D. P Value
Maximum z
Mean z
Maximum z
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
0.383 0.304 0.461 0.389 0.694 0.266 0.235 0.754 0.325 0.347 0.416 0.175 .184
0.461 0.396 0.346 0.729 0.225 0.807 0.365 0.719 0.741 0.736 0.553 0.213
3.345 3.154 3.989 3.854 3.733 4.014 2.667 3.857 3.610 3.193 3.542 0.439 .223
2.681 2.564 2.917 3.633 2.917 3.558 3.393 4.151 3.336 3.773 3.292 0.511
0.271 0.581 0.713 0.144 0.922 0.431 0.599 0.734 0.335 0.393 0.512 0.239 .421
0.267 0.420 0.355 0.330 0.598 0.459 0.494 0.532 0.589 0.521 0.457 0.112
2.253 3.644 4.521 2.533 4.534 3.276 3.605 4.072 3.340 2.759 3.454 0.786 .346
2.481 2.535 3.422 2.854 3.651 2.304 3.615 3.054 4.615 3.320 3.185 0.695
The definition of the SMA is based on a GM mask. Differences between RS and PC were calculated by a paired t test.
1354
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
Table 5 Longitudinal study: mean and maximum z-scores in the SMA, representing correlations with lSM1 Session
Subject A
Subject B
Mean z
1 2 3 4 5 6 7 8 9 10 Mean S.D. P Value
Maximum z
Mean z
Maximum z
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
0.196 0.894 0.825 0.819 1.420 0.604 0.637 1.018 0.421 0.687 0.752 0.334 .096
0.865 0.798 0.548 1.771 0.390 1.862 1.042 1.509 1.428 1.377 1.159 0.506
2.286 2.578 3.797 2.787 3.733 4.014 2.651 3.221 2.621 2.607 3.029 0.614 .984
2.681 1.716 2.425 3.633 1.702 3.558 3.393 4.151 3.336 3.773 3.037 0.860
0.127 1.294 1.423 0.644 0.937 1.577 1.756 1.713 0.898 1.556 1.192 0.529 .067
0.443 0.624 0.763 0.679 1.168 1.090 1.304 1.159 1.092 0.884 0.921 0.284
2.253 3.644 4.521 2.352 4.440 3.276 3.605 4.072 3.340 2.669 3.417 0.807 .014
2.481 2.535 2.030 1.677 2.669 2.304 3.615 2.204 2.865 2.846 2.523 0.532
The definition of the session-specific SMA-ROI is defined by the voxels that were significantly (Pb.01) activated in the respective fMRI experiment.
57 voxels over the different sessions; in Subject B, the range was between 13 and 41 voxels. Subject B demonstrated significantly higher RS maximum z-scores in the activationdefined ROI. In this subject, a trend towards higher RS mean z-scores was also observed. An even weaker trend was seen in the mean z-scores of Subject A; here the mean z-scores in the continuous task were slightly higher than those of the RS (Table 5). Comparing FCV, the difference between RS and CP was again significant for Subject B but not for Subject A (Table 6). The analysis of variance of mean z-scores in SMA revealed no significant difference between subjects and conditions, but significant interaction between both factors for the activation-defined SMA (P=.017), or respectively a trend towards significance for the anatomically defined SMA (P=.12). FCV and mean z-scores correlated significantly in anatomically defined rSM1 and in activation-defined SMA for both subjects (Figs. 4 and 5). In the anatomically defined
Table 6 Longitudinal study: FCV for the activation-defined SMA ROI Session
Subject A
Subject B
SMA, correlation between FCV and mean z-scores was found for Subject A, but not for Subject B (Fig. 5, first row). In the multisubject study, the extent of the anatomically defined rSM1 varied between 173 and 274 voxels. For rSM1, group mean z-scores and FCV were significantly higher in RS; also, maximum z-scores were higher in RS but the difference missed significance (P=.13) due to one outlier (Tables 7 and 9, second and third column). The individual GM-defined SMA varied between 81 and 199 voxels; the individual activation-defined SMA, between 12 and 99 voxels. For both the anatomically and the activation-defined SMA-ROI, no measure revealed significant RS-CP difference (Tables 8 and 9). In all ROI under evaluation, FCV and mean z-scores correlated significantly in the fcMRI experiments (Fig. 6). In rSM1, we found a significant correlation between individual mean z-score in the fMRI experiment and mean z-score in CP (P=.036) (Fig. 7), but not between the z-score in fMRI and in RS (P=.30). In the respective clustercorrected (Pb.01) z-score maps of the fMRI experiments, five subjects showed deactivation in rSM1, one subject showed activation and in one subject neither significant activation nor deactivation was observed.
Resting state Continuous task Resting state Continuous task 1 2 3 4 5 6 7 8 9 10 Median P Value
0.071 0.095 0.114 0.149 0.242 0.173 0.070 0.194 0.116 0.050 0.115 N.2
0.095 0.000 0.057 0.277 0.030 0.462 0.193 0.250 0.209 0.300 0.201
0.073 0.273 0.238 0.200 0.267 0.167 0.333 0.278 0.118 0.308 0.252 3.91E−03
0.024 0.061 0.095 0.000 0.200 0.056 0.143 0.111 0.118 0.154 0.103
The statistical differences between the states were calculated with Wilcoxon signed ranks test for matched pairs.
4. Discussion The main focus of this study was the evaluation of the temporal variability of functional connectivity and the reproducibility of the results of different statistical analysis methods. For this purpose, we investigated the cortical motor system, with focus on the functional connectivity, of two subjects who were repeatedly scanned over a period of half a year. One drawback of such a two-subject analysis (or rather a twofold single-subject analysis) is the limited possibility to draw inferences onto a larger population. We therefore
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
1355
Fig. 4. Linear regression between mean z-score and FCVs in a GM-defined rSM1-ROI over the 10 sessions of the longitudinal study. Magenta squares: Continuous performance; blue triangles: resting state. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
extended the experiments and the ROI analyses in a group of seven healthy subjects (including the two subjects of the main study). But even in the longitudinal study, some of the effects found parallel the findings of other, group-based studies. The multisession t-score maps obtained from the fMRI runs showed the expected activation pattern for a right-finger motor task with prominent activation in left SM1, in medial SMA and in bilateral premotor areas. We found deactivation in right SM1 in one subject and no activation in the other. This different characteristic was not a mere statistical effect.
Even with a less conservative threshold of Pb.05, Subject A showed no deactivation in rSM1 at the multisession level, while this was the case in Subject B. In the ROI analysis of the multisubject study, one subject (who was not Subject A) had a positive mean z-score in the rSM1-ROI, while the other subjects had negative mean z-scores (Fig. 7). Such interindividual differences in SM1 activation ipsilateral to the performing limp have been reported in various other studies investigating simple unilateral motor tasks. In Ref. [33], for example, ipsilateral deactivation was observed during a simple right-finger opposition task in 13 of 23
Fig. 5. Linear regression between mean z-score and FCV in SMA for the longitudinal study. In the first row, the SMA-ROI is defined by a GM mask; in the second row, SMA is defined by voxels which were significantly activated in the respective fMRI experiment (Pb.01). Magenta squares: Continuous performance; blue triangles: resting state. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
1356
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
Table 7 Cross-sectional study: mean and maximum z-scores in the GM-defined rSM1, representing correlations with lSM1 Subject
SM1r-GM Mean z
Maximum z
Resting state Continuous task Resting state Continuous task A B C D E F G P Value
1.334 0.716 0.841 0.838 0.645 0.764 0.665 .002
0.783 0.538 0.282 0.599 0.061 0.513 −0.069
4.619 3.044 3.054 3.205 2.857 2.908 2.857 .133
3.192 3.645 2.024 3.116 2.363 2.674 2.513
subjects, while the other subjects showed neither significant activation nor deactivation. Here, we demonstrated that such interindividual differences also exist at a multisession level, which points out that these differences are physiologically based. In the analysis of the multisession fcMRI experiments, the composite maps of both subjects under investigation showed considerable differences between RS and CP. In RS, symmetric correlation was found in bilateral SM1 and in superior parietal cortex. This interhemispheric connectivity was almost vanished in unilateral CP. Furthermore, correlation was diluted in parts of SM1 ipsilateral to the seed ROI, especially in the lateral aspects of lSM1 (see also the difference t-score maps in Fig. 3). Peltier et al. [21] demonstrated a significant decrease in interhemispheric SM1 connectivity after the performance of a repetitive unilateral handgrip task due to fatigue. The task used here for CP, finger flexion/extension during 4.5 min, was considerably less exhausting than the task used in Ref. [21] (20 min of gripping a device). In our study, we can exclude the possibility of a fatigue-induced reduction of connectivity as the analysis of variance between the different phases of the CP runs revealed no evidence of fatigue even at very low levels of significance. The interhemispheric connectivity descent in CP was also apparent in the ROI analyses both in the longitudinal and in
the multisubject study. In rSM1, mean z-scores, maximum z-scores and FCV were significantly reduced in CP compared to RS. Furthermore, reduction in mean z-score was accompanied by a decrease of significantly correlated voxels. In the ROI analyses of the SMA, interindividual differences were apparent. Based on the evaluation of the activation-defined ROI, Subject A had a weak trend of higher SMA to lSM1 connectivity in CP which was only observable in the comparison of mean z-scores, but not in FCV or maximum z-scores. In contrast, Subject B had higher connectivity in RS, which was significant in FCV and maximum z-scores and reached almost significance in mean z-scores. In both subjects, there was a strong correlation between mean z-score and FCV (Fig. 5, second row). Based on the analysis of the activation-defined ROI, Subject B showed a reduction of connectivity between SMA and lSM1 in CP, while Subject A did not. Analysis of variance underlines the significance of the interaction between subject and fcMRI condition. In the cross-sectional study, we found additionally in four subjects connectivity increase and in three subjects connectivity reduction in CP. So the direction of SMA to lSM1 connectivity change during our CP condition remained undefined. In the longitudinal study, analyses of the anatomically defined SMA-ROI revealed only one significant difference, namely, Subject B had a higher FCV in RS than in CP. All other differences remained nonsignificant. Interestingly, a correlation between mean z-score and FCV could still be observed in Subject A. Depending on the imaging session and subject, the anatomically defined SMA was three to 11 times larger than the respective activation-defined ROI. Hence, GM voxels that are not directly related to the functional network under evaluation were likely to be included into analysis. Due to this nonspecificity of the anatomically determined ROI, changes in connectivity may be diluted which are otherwise observable in more confined ROI. In SM1, the effect of connectivity change between RS and CP is much more obvious. Even if the anatomically defined ROI had encompassed non–motor-related voxels, the reduced connectivity in CP still remained significant due to the large effect in motor-specific voxels. Even so, the results of our
Table 8 Cross-sectional study: mean and maximum z-scores in the GM and in the activation-defined SMA, representing correlations with lSM1 Subject
SMA-Act
SMA-GM
Mean z
A B C D E F G P Value
Maximum z
Mean z
Maximum z
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
1.559 0.964 0.809 0.946 0.647 1.611 0.535 .779
0.991 1.253 1.399 0.692 1.228 1.032 0.876
2.822 2.697 3.673 3.178 2.655 3.212 1.878 .476
2.110 3.670 4.373 3.475 2.387 2.515 3.238
0.254 0.557 0.707 0.826 0.517 0.950 0.288 .850
0.325 0.506 1.192 0.484 0.234 0.778 0.430
2.822 2.697 3.673 3.178 2.655 3.212 1.878 .268
2.796 3.670 4.373 3.475 2.387 2.515 3.238
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
1357
Table 9 Cross-sectional study: FCV for the three different ROIs Subject
A B C D E F G P Value
rSM1r-GM
SM-Act
SMA-GM
Resting state
Continuous task
Resting state
Continuous task
Resting state
Continuous task
0.271 0.160 0.158 0.163 0.099 0.070 0.103 .016
0.097 0.090 0.015 0.063 0.043 0.042 0.017
0.333 0.217 0.145 0.179 0.100 0.294 0.000 N.4
0.111 0.261 0.329 0.074 0.200 0.147 0.133
0.071 0.066 0.139 0.163 0.048 0.228 0.000 N.5
0.032 0.074 0.277 0.065 0.065 0.098 0.074
The statistical differences between the states were calculated with Wilcoxon signed ranks test for matched pairs.
ROI analyses of SMA strongly suggest that it is preferable to work with activation-defined ROI like in Refs. [6,12,22], rather than to use purely anatomically based regions. The regression analysis of mean z-score and FCV in this work confirmed the results of the study of Newton et al. [22], which had investigated the modulation of functional connectivity of SM1 to an audibly paced finger-tapping task with increasing demand. The authors had also reported that the changes in FCV within different ROI (SMA, cerebellum and right auditory cortex) mirrored the changes in mean correlation of the regions (which is equivalent to the mean z-score used in our study). With the exception of the anatomically defined SMA for Subject B in the longitudinal study, it was also demonstrated that both measures correlate
with each other, even if there was no significant dependence on the cognitive states in a specific region (Subject A, activation-defined SMA). Our results therefore support the statement of Newton et al. [22] that it is the number of correlated voxels that drives the observed changes in overall functional connectivity. In both the longitudinal and the cross-sectional study, we found a reduction in bilateral SM1 connectivity during CP, which is in contrast to other fcMRI experiments evaluating connectivity under different mental states, where increased connectivity during performance was observed. These studies investigated either nonmotor systems (e.g., the language-related network [6]) or used bimanual motor tasks [12,14], where the corresponding fMRI task itself
Fig. 6. Multisubject analysis of fcMRI: Linear regression between mean z-score and FCV in rSM1 and in SMA. SMA-act: SMA-ROI is defined by significantly activated voxels in fMRI. SMA-GM: SMA is defined be a GM mask. Magenta squares: Continuous performance; blue triangles: resting state. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
1358
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359
Fig. 7. Linear regression between individual mean z-scores in the fMRI task and mean z-scores of the respective CP fcMRI experiment. Mean z-scores are calculated over the right SM1-ROI. Correlation reached significance (P=.036).
elicited bilateral SM1 activation. Here, we were able to demonstrate that unilateral SM1 activation during a conventional fMRI experiment is paralleled by desynchronization of bilateral SM1 during steady-state performance of the same task. In addition, we found indications that individual PC connectivity decrease correlates with stronger suppression of collateral SM1 activation during the unilateral motor task (Fig. 7). In the longitudinal study, the measures of connectivity for each subject showed some variations over the sessions. In the first session, for example, Subject B has almost equal rSM1/ lSM1 correlations in both RS and CP, despite the fact that reduction of bilateral SM1 connectivity during unilateral performance is highly significant at multisession level. This indicates that functional connectivity is not a static measure, even for the same individual in comparable cognitive states. Despite this variability, possibly reflecting individual behavioral changes, all measures of functional connectivity investigated here were stable at group level. In our study, we were able to demonstrate a distinct reduction in left–right SM1 synchrony in the CP state compared to the RS in both individuals in the longitudinal study and in all subjects in the multisubject study. This effect was reproducible over half a year with all statistical methods and all measures of connectivity under investigation. Despite intra- and interindividual variability, serial and crosssectional assessments of functional connectivity reveal stable and reliable results. Acknowledgments The authors thank the volunteers for participating in our studies, as well as the two anonymous reviewers of this article for their valuable suggestions and comments. References [1] Biswal BB, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in the motor cortex of the resting human brain using echoplanar MRI. Magn Reson Med 1995;34:537–41.
[2] Lowe MJ, Mock BJ, Sorenson JA. Functional connectivity in single and multislice echoplanar imaging using resting-state fluctuations. NeuroImage 1998;7:119–32. [3] Xiong J, Parsons LM, Gao JH, Fox PT. Interregional connectivity to primary motor cortex revealed using MRI resting state images. Hum Brain Mapp 1999;8:151–6. [4] Cordes D, Haughton VM, Arfanakis K, Wendt GJ, Turski PA, Moritz CH, et al. Mapping functionally related regions of brain with functional connectivity MR imaging. Am J Neuroradiol 2000;21:1636–44. [5] De Luca M, Smith S, De Stefano N, Federico A, Matthews PM. Blood oxygenation level dependent contrast resting state networks are relevant to functional activity in the neocortical sensorimotor system. Exp Brain Res 2005;167:587–94. [6] Hampson M, Peterson BS, Skudlarski P, Gatenby JC, Gore JC. Detection of functional connectivity using temporal correlations in MR images. Hum Brain Mapp 2002;15:247–62. [7] Hampson M, Tokoglu F, Sun Z, Schafer RJ, Skudlarski P, Gore JC, et al. Connectivity-behavior analysis reveals that functional connectivity between left BA39 and Broca's area varies with reading ability. NeuroImage 2006;31:513–9. [8] Greicius MD, Krasnow B, Reiss AL, Menon V. Functional connectivity in the resting brain: a network analysis of the default mode hypothesis. Proc Natl Acad Sci U S A 2003;100:253–8. [9] Fransson P. Spontaneous low-frequency BOLD signal fluctuations: an fMRI investigation of the resting-state default mode of brain function hypothesis. Hum Brain Mapp 2005;26:15–29. [10] De Luca M, Beckman CF, De Stefano N, Matthews PM, Smith SM. FMRI resting state networks define distinct modes of long-distance interactions in the human brain. NeuroImage 2006;29:1359–67. [11] Lowe MJ, Dzemidzic M, Lurito JT, Mathews VP, Phillips MD. Correlations in low-frequency BOLD fluctuations reflect corticocortical connections. NeuroImage 2000;12:582–7. [12] Lowe MJ, Phillips MD, Lurito JT, Mattson D, Dzemidzic M, Mathews VP. Multiple sclerosis: low-frequency temporal blood oxygen leveldependent fluctuations indicate reduced functional connectivity initial results. Radiology 2002;224:184–92. [13] Hampson M, Olson IR, Leung HC, Skudlarski P, Gore JC. Changes in functional connectivity of human MT/V5 with visual motion input. Neuroreport 2004;15:1315–9. [14] Morgan VL, Price RR. The effect of sensorimotor activation on functional connectivity mapping with MRI. Magn Reson Imaging 2004;22:1069–75. [15] Grecius MD, Srivastava G, Reiss AL, Menon V. Default-mode network activity distinguishes Alzheimer's disease from healthy aging: evidence from functional MRI. Proc Natl Acad Sci U S A 2004;101: 4637–42. [16] Anand A, Li Y, Wang Y, Wu J, Gao S, Bukhari L, et al. Activity and connectivity of brain mood regulating circuit in depression: a functional magnetic resonance study. Biol Psychiatry 2005;57:1079–88. [17] Kennedy DP, Courchesne E. The intrinsic functional organization of the brain is altered in autism. NeuroImage 2008;39:1877–85. [18] Friston KJ, Frith CD, Liddle PF, Frackowiak RS. Functional connectivity: the principal-component analysis of large (PET) data sets. J Cereb Blood Flow Metab 1993;13:5–14. [19] Beckmann CF, DeLuca M, Devlin JT, Smith SM. Investigations into resting-state connectivity using independent component analysis. Philos Trans R Soc B 2005;360:1001–13. [20] Fox MD, Snyder AZ, Vincent JL, Corbetta M, Van Essen DC, Raichle ME. The human brain is intrinsically organized into dynamic, anticorrelated functional networks. Proc Natl Acad Sci U S A 2005; 102:9673–8. [21] Peltier SJ, LaConte SM, Niyazov DM, Liu JZ, Sahgal V, Yue GH, et al. Reductions in interhemispheric motor cortex functional connectivity after muscle fatigue. Brain Res 2005;1057:10–6. [22] Newton AT, Morgan VL, Gore CG. Task demand modulation of steady-state functional connectivity to primary motor cortex. Hum Brain Mapp 2007;28:663–72.
M. Amann et al. / Magnetic Resonance Imaging 27 (2009) 1347–1359 [23] Dettmers C, Fink GR, Lemon RN, Stephan KM, Passingham RE, Silbersweig D, et al. Relation between cerebral activity and force in the motor areas of the human brain. J Neurophysiol 1995;74: 802–15. [24] Yousry TA, Schmid UD, Jassoy AG, Schmidt D, Eisner WE, Reulen HJ, et al. Topography of the cortical motor hand area: prospective study with functional MR imaging and direct motor mapping at surgery. Radiology 1995;195:23–9. [25] Picard N, Strick PL. Motor areas of the medial wall: a review of their location and functional activation. Cereb Cortex 1996;6:342–53. [26] Wexler BE, Fulbright RK, Lacadie CM, Skudlarski P, Kelz MB, Constable RT, et al. An fMRI study of the human cortical motor system response to increasing functional demands. Magn Reson Imaging 1997;15:385–96. [27] Smith JF, Chen K, Johnson S, Morrone-Strupinsky J, Reiman EM, Nelson A, et al. Network analysis of single-subject fMRI during a finger opposition task. NeuroImage 2006;32:325–32.
1359
[28] Oldfield RC. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia 1971;9:97–113. [29] Waites AB, Stanislavsky A, Abbott DF, Jackson GD. Effect of prior cognitive state on resting state networks measured with functional connectivity. Hum Brain Mapp 2005;24:59–68. [30] Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996;29:162–73. [31] Hartvig NV, Jensen JL. Spatial mixture modeling of fMRI data. Hum Brain Mapp 2000;11:233–48. [32] Johansen-Berg H, Behrens TE, Robson MD, Drobnjak I, Rushworth MF, Brady JM, et al. Changes in connectivity profiles define functionally distinct regions in human medial frontal cortex. Proc Natl Acad Sci U S A 2004;101:13335–40. [33] Nirkko AC, Ozdoba C, Redmond SM, Bürki M, Schroth G, Hess CW, et al. Different ipsilateral representations for distal and proximal movements in the sensorimotor cortex: activation and deactivation patterns. NeuroImage 2001;13:825–35.