Less head motion during MRI under task than resting-state conditions

Author’s Accepted Manuscript Less head motion during MRI under task than resting-state conditions Willem Huijbers, Koene R.A. Van Dijk, Meta M. Boenniger, Rüdiger Stirnberg, Monique M.B. Breteler www.elsevier.com

PII: DOI: Reference:

S1053-8119(16)30693-0 http://dx.doi.org/10.1016/j.neuroimage.2016.12.002 YNIMG13623

To appear in: NeuroImage Received date: 20 June 2016 Revised date: 24 November 2016 Accepted date: 1 December 2016 Cite this article as: Willem Huijbers, Koene R.A. Van Dijk, Meta M. Boenniger, Rüdiger Stirnberg and Monique M.B. Breteler, Less head motion during MRI under task than resting-state conditions, NeuroImage, http://dx.doi.org/10.1016/j.neuroimage.2016.12.002 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Less head motion during MRI under task than restingstate conditions Authors: Willem Huijbers1,2,3, Koene R.A. Van Dijk3, Meta M. Boenniger1, Rüdiger Stirnberg4, Monique M.B. Breteler1

Affiliation: 1German Centre for Neurodegenerative Diseases (DZNE), Department of Population Health Sciences, Bonn, Germany. 2

Department of Neurology, Massachusetts General Hospital, Harvard Medical

School, Boston, MA. 3

Athinoula A. Martinos Center for Biomedical Imaging, Department of

Radiology, Massachusetts General Hospital, Charlestown, MA. 4

German Centre for Neurodegenerative Diseases (DZNE), Department of MR

Physics, Bonn, Germany.

Corresponding author: Willem Huijbers, [email protected]

1

Abstract Head motion reduces data quality of neuroimaging data. In three functional magnetic resonance imaging (MRI) experiments we demonstrate that people make less head movements under task than resting-state conditions. In Experiment 1, we observed less head motion during a memory encoding task than during the resting-state condition. In Experiment 2, using publicly shared data from the UCLA Consortium for Neuropsychiatric Phenomics LA5c Study, we again found less head motion during several active task conditions than during a resting-state condition, although some task conditions also showed comparable motion. In the healthy controls, we found more head motion in men than in women and more motion with increasing age. When comparing clinical groups, we found that patients with a clinical diagnosis of bipolar disorder, or schizophrenia, move more compared to healthy controls or patients with ADHD. Both these experiments had a fixed acquisition order across participants, and we could not rule out that a first or last scan during a session might be particularly prone to more head motion. Therefore, we conducted Experiment 3, in which we collected several task and resting-state fMRI runs with an acquisition order counter-balanced. The results of Experiment 3 show again less head motion during several task conditions than during rest. Together these experiments demonstrate that small head motions occur during MRI even with careful instruction to remain still and fixation with foam pillows, but that head motion is lower when participants are engaged in a cognitive task. These finding may inform the choice of functional runs when studying difficult-to-scan populations, such as children or certain patient populations. Our findings also indicate

2

that differences in head motion complicate direct comparisons of measures of functional neuronal networks between task and resting-state fMRI because of potential differences in data quality. In practice, a task to reduce head motion might be especially useful when acquiring structural MRI data such as T1/T2-weighted and diffusion MRI in research and clinical settings.

Keywords: neuroimaging, movement, data quality, artifact, functional MRI

Introduction Head motion is a principal confound when acquiring brain magnetic resonance imaging (MRI) data. Head motion induces artifacts in brain images, may render data useless, and can bias group results (e.g. (Alexander-Bloch et al., 2016; Fellner et al., 2016; Glover and Lee, 1995; Pardoe et al., 2016; Power et al., 2012; Reuter et al., 2015; Satterthwaite et al., 2012; Van Dijk et al., 2012; Yan et al., 2013; Zaitsev et al., 2015). Although methods for motion correction have been improving over the past years, these corrections remain imperfect and come at a cost. (e.g. (Ferrazzi et al., 2014; Goto et al., 2015; Griffanti et al., 2014; Power et al., 2015; Tisdall et al., 2015; Yan et al., 2013; Zaitsev et al., 2016). Prospective motion correction during MRI, with reacquisition strategies, increases the total scan time. Retrospective corrections, for example in functional MRI (fMRI), typically remove corrupted scans and effectively reduce the number of observation in a time series. Thus, strategies that help reduce head motion

3

during MRI can improve data quality for both scientific research and diagnostic purposes in the clinic. In a clinical setting, it is common to play music or movies to increase compliance, especially in children or patients. Inspired by clinical MRI, Vanderwal et al. (2015) demonstrated that a movie could reduce head motion compared to an eyes-open fixation resting-state condition. The authors argued that cognitive engagement that comes with watching a movie reduces head motion (Vanderwal et al., 2015). Extending this argument, we hypothesize that an engaging cognitive task can also reduce head motion during MRI. In a series of three experiments, we investigated head motion during restingstate conditions and several cognitive task conditions typically used in functional MRI experiments (e.g. (Barch et al., 2013; Krienen et al., 2014; Poldrack and Gorgolewski, 2015; 2014; Wager et al., 2007). Our aim was not to examine the direct influence of the cognitive tasks on the blood-oxygen-level dependent (BOLD) signal nor to attempt to segregate task-evoked (extrinsic) and spontaneous (intrinsic) fMRI activity (e.g (Cole et al., 2014; Fransson, 2006; Geerligs et al., 2015; Huijbers et al., 2013; Krienen et al., 2014; Northoff et al., 2010; Smith et al., 2009). Instead, we used the fMRI time-series to estimate head motion. In three experiments, with a total of 369 participants, we explored different task conditions as a strategy to reduce head motion and potentially improve the quality of neuroimaging data.

4

Materials and Methods Overview In Experiment 1, we examined head motion in a passive memory encoding task and under resting-state conditions. In Experiment 2, using functional MRI data from the University of California Los Angeles (UCLA) Neuropsychiatric Phenomics Study, we examined head motion under several cognitive tasks and resting-state conditions. Both Experiment 1 and 2 had a fixed order of conditions across participants, so we could not rule out whether, for instance, the first or last scan in a session might be particularly prone to head motion. Therefore, we conducted Experiment 3 in which we collected fMRI data during several cognitive tasks, a movie, and during resting-state with an acquisition order counter-balanced across participants.

Experiment 1: head motion in resting-state and a passive task condition We recruited 56 participants (age range 20-46, M = 25, SD = 4.74, 32 female) from the Bonn community in the context of our pre-studies for the Rhineland Study, a novel prospective cohort study. All participants provided written informed consent. The study was approved by the medical ethics committee of the Medical Faculty of the University of Bonn. Functional Magnetic Resonance Imaging (fMRI) data were acquired using a 3 Tesla Siemens MAGNETOM Prisma system (Siemens Medical Systems, Erlangen, Germany). The scanner was equipped with a 64-channel phased-array head/neck coil. Auditory 5

stimuli were presented via S14 Insert Earphones (Sensimetric, Malden, USA). Visual stimuli were presented via a monitor located at the head of the magnet bore and seen via a mirror mounted on the head coil. The visual and auditory stimuli were presented using PsychoPy software v1.82 (Peirce, 2007), running on a Windows PC. Using inflatable air pads head movement was minimized and participants were instructed to lie still while the scanner was running. We acquired four fMRI time-series of 140 volumes using echo-planar imaging (EPI). Each volume consisted of 32 axial slices of 3mm thickness with a 0.75mm skip. The repetition time (TR) was 2000ms, echo time (TE) 30ms, flip angle (FA) 84°, readout bandwidth 2300 Hz/pixel, and field of view (FOV) was 192 x 192mm resulting in an effective voxel size of 3.0 x 3.0 x 3.75mm. After a first resting-state fMRI run (REST1) and two fMRI runs using a memory encoding task (ENC1 and ENC2), we collected T1-weighted, T2-weighted, and diffusion MRI data, (which are not included in this manuscript), followed by a second resting-state fMRI run (REST2). The total acquisition time was approximately 48 minutes and the time between the end of ENC2 and the beginning of REST2 was approximately 18 minutes. During REST1 and REST2, a white fixation cross was presented at the center of a black screen. During the task runs we presented visual images of faces or scenes and auditory sounds of vocal or non-vocal sounds, together or separate, using a mixedevent/block design (Visscher et al., 2003). The task did not require any motor responses using a button-box (or otherwise). Participants were instructed to pay attention and were told that their memory for the visual images and for the sounds would be tested after the scanning session. 6

The fMRI data were pre-processed using MATLAB (Mathworks, Natick, MA, USA), the Statistical Parametric Mapping Toolbox (SPM8, UCL, London, UK) and GLM_Flex (MGH, http://mrtools.mgh.harvard.edu/index.php/GLM_Flex, MA, USA). We dropped the first 4 volumes and realigned the time-series to the first volume. From the realignment, we obtained the motion parameters for translation and rotation. To calculate mean translational motion, we first took the relative difference in translations between two consecutive volumes. Next, we combined the mean translation in x, y and z direction into a single number using the root mean square (RMS) (Van Dijk et al., 2012). To calculate mean rotation, we took the Euler angle of the rotation parameters (x, y and z) and combined the Euler angles by averaging the absolute difference between two volumes. The Euler range is expressed in radians and was multiplied by 1000 to obtain a value with a similar scale to the mean translational motion metric (Van Dijk et al., 2012). To calculate the framewise displacement, we first transformed rotation from degrees to millimeters by calculating displacement on the surface of a sphere with a radius 50 mm, which is approximately the mean distance from the cerebral cortex to the center of the head (Jenkinson et al., 2002; Power et al., 2012). Next, we combined root mean squares of the volume-to-volume translation and volume-to-volume rotation and calculated the mean framewise displacement over the total number of volumes in a series. We also estimated the number of motion spikes by counting number of volumes where the volume-to-volume framewise displacement exceeded 0.20 mm. The absolute number of motion spikes was divided by the total number of volumes in a series, to correct for the series duration, and defined as percentile motion spikes. For all statistical analyses, we

7

log-transformed the percentile motion spikes after substituting the zero values with the half of the minimal observed value. We used Pearson’s product moment correlations for associations between variables. For post-hoc comparisons between two conditions we used paired-t-tests (two-sided) and for comparisons between groups we used twosample t-tests (two-sided). The figures were generated using ggplot2 (Hadley, 2009).

Experiment 2: head motion in the UCLA Consortium for Neuropsychiatric Phenomics LA5c Study We downloaded data from 290 participants (age range 22-50, M = 33, SD = 9.25, 124 female) who participated in the UCLA Consortium for Neuropsychiatric Phenomics LA5c Study. The data was obtained via the public database openfMRI (Poldrack et al., 2013; Poldrack and Gorgolewski, 2015) and approved by the UCLA Institutional Review Board The LA5c dataset includes 138 healthy controls, 58 individuals diagnosed with schizophrenia, 49 with bipolar disorder and 45 with attention deficit hyperactivity disorder (ADHD). These participants were recruited from the LA2k study, see also for more details (Bilder et al., 2009; Jalbrzikowski et al., 2012). The fMRI data were acquired using a 3 Tesla Siemens MAGNETOM TrioTim system (Siemens Medical Systems, Erlangen, Germany) on two different days in a counterbalanced design. Day A included fMRI data of a balloon analog risk task (BART) (Helfinstein et al., 2014; Lejuez et al., 2002; Schonberg, 2012) and the encoding (ENC) plus retrieval (RET) phase of a paired associates memory task. Day B included fMRI data of resting-state run (REST), a stop-

8

signal task (STOP) (Congdon et al., 2014), a spatial working memory capacity task (WM) (Montojo et al., 2013) and a task-switching paradigm (SWITCH). The resting-state and task conditions were acquired using the same EPI sequence. Each volume consisted of 34 axial slices of 4mm thickness. The TR was 2000ms, TE 30ms, FA 90°, and FOV was 192 x 192 resulting in an effective voxel size of 3.0 x 3.0 x 4.00 mm. The BART consisted of 267 volumes, ENC of 242, RET of 268, REST of 152, STOP of 184, WM of 291 and SWITCH of 208. During the resting-state condition, participants were asked to remain relaxed and keep their eyes open for five minutes. Visual stimuli were presented using MRIcompatible goggles (Resonance Technologies, Van Nuys, CA) and responses where made via a button-box. We downloaded the dataset and obtained 262 REST runs, 260 WM runs, 199 ENC runs, 258 BART runs, 255 SWITCH runs, 199 RET runs and 256 STOP runs. The task and resting-state data were pre-processed using SPM8 and GLM_Flex. We realigned the time-series to the first volume. From the realignment, we obtained the motion parameters and calculated mean framewise displacement, percentile motion spikes, mean translational motion and mean rotation using the identical methods as described in Experiment 1. The linear regression models were implemented using the companion to Applied Regression Toolbox (Fox and Weisberg, 2011). The linear models included a random intercept for each subject, a within-subject term for task condition (REST, WM, ENC, BART, SWITCH, RET, STOP) and between-subject terms for diagnosis (healthy controls, ADHD, bipolar, schizophrenia), gender (female, male) and age (years). In the follow-up analysis, we also included a linear model with a random intercept for each subject, a within-subject term for condition (REST, STOP), a between-subject term

9

for diagnosis (healthy controls, schizophrenia) and the interaction terms of condition by diagnosis. More details on the fMRI task paradigms can be found on openfMRI.org (Poldrack and Gorgolewski, 2015) or via the UCLA Consortium for Neuropsychiatric Phenomics (http://www.phenomics.ucla.edu/).

Experiment 3: head motion during counterbalanced rest and task conditions We recruited 23 participants (age 28-78, M = 63, SD = 10.70, 13 female) from the Bonn community in the context of our pre-studies for the Rhineland Study, a novel prospective cohort study. All participants provided written informed consent. The study was approved by the medical ethics committee of the Medical Faculty of the University of Bonn. The fMRI data were acquired using the same equipment and scan sequences as described in Experiment 1. We acquired five fMRI time-series of 214 volumes each. We obtained one resting-state run (REST) and four task runs. The task runs consisted of a passive memory encoding task (ENC), identical to Experiment 1 (ENC), a retrieval task (RET), a working memory task (WM) and a movie without sound (MOVIE). The retrieval task was old/new recognition task with novel and previously seen visual stimuli from the memory encoding task. For the working memory, we used an N-back task (Braver et al., 1997). The session order was counter-balanced using two alternate sequences (REST/ENC/RET/WM/MOVIE) and the inverse (MOVIE/WM/RET/ENC/REST). After each participant, we additionally shifted the starting condition circular to the last position. When the retrieval condition occurred before the encoding condition, participants rated if the auditory sound was vocal or non-vocal, instead of whether they heard the sound 10

before or not. Responses in the retrieval and working memory task were made with the right-hand via a button-box. The movie contained selected clips from the 2006 nature documentary “Planet Earth” (British Broadcasting Corporation) and contained no underwater scenes, sweeping views, and limited violent or animal mating behavior. The task, movie and resting-state data were pre-processed using SPM8 and GLM_Flex. We dropped the first 4 volumes and realigned the time-series to the first volume. From the realignment, we obtained the motion parameters and calculated the mean framewise displacement, percentile motion spikes, mean translational motion and mean rotation using the identical methods as described in Experiment 1.

Results Experiment 1: head motion in resting-state and a passive task condition We assessed framewise displacement, percentile motion spikes, translational and rotational head motion for each of the task and resting-state runs. We performed 2x2 repeated measures ANOVAs with factors for condition (TASK/REST) and acquisition order (Run1/Run2). We found greater framewise displacement during resting-state runs (M = 0.156, SD = 0.066) compared to task runs (M = 0.127, SD = 0.047) as demonstrated by a main effect of condition (F(1, 164) = 19.56, p < 0.001). We also found larger framewise displacement across the second two runs (ENC2 and REST2, M = 0.146, SD = 0.061) compared to the first two runs (REST1 and ENC1, M = 0.137, SD = 0.057) as demonstrated by a main effect of acquisition order (F(1, 164) = 8.44, p = 0.004). We did

11

not observe a significant interaction between condition and acquisition order (F(2,164) = 0.002, p = 0.96) indicating that later acquired scans contain larger framewise displacement than earlier acquired scans, irrespective of task condition. See Figure 1 for a plot of the mean framewise displacement across conditions. We found a higher percentile of motion spikes in the resting-state runs (M = 0.225, SD = 0.208) compared to task runs (M = 0.14, SD = 0.151) as demonstrated by a main effect of condition (F(1, 164) = 22.66, p < 0.001). We also found a higher percentile of motion spikes across the second runs (M = 0.193, SD = 0.188) compared to first runs (M = 0.172, SD = 0.185) as demonstrated by a main effect of acquisition order (F(1, 164) = 5.76, p = 0.018). We did not observe a significant interaction between condition and acquisition order (F(2,164) = 0.233, p = 0.63). We found more translational motion during resting-state runs (M = 0.043, SD = 0.020) compared to task runs (M = 0.034, SD = 0.013) as demonstrated by a main effect of condition (F(1, 164) = 27.41, p < 0.001). We also found more translational motion across the second two runs (ENC2 and REST2, M = 0.040, SD = 0.017) compared to the first two runs (REST1 and ENC1, M = 0.037, SD = 0.017) as demonstrated by a main effect of acquisition order (F(1, 164) = 11.91, p < 0.001). We did not observe a significant interaction between condition and acquisition order (F(2,164) = 0.018, p = 0.89). Finally, we also found more rotational motion in the resting-state runs (M = 0.416, SD = 0.24) when compared to the task runs (M = 0.391, SD = 0.20) as demonstrated by a main effect of condition (F(1, 164) = 12.13, p < 0.001). We did not find a difference in rotation between the second runs (M = 0.444, SD = 0.24) compared to the first runs (M = 0.363, SD = 0.018), as the main effect of acquisition order was not significant (F(1, 164)

12

= 1.80, p = 0.181). Nevertheless, we observed an interaction between condition and acquisition order for rotational motion (F(2, 164) = 5.14, p = 0.024). This shows that a counterbalanced order of data acquisition is needed to fully elucidate if the effects are a consequence of task condition or acquisition order (see Experiment 3). === INSERT FIGURE 1 HERE ==

As first additional analysis, we compared translational head motion within the encoding task during the passive rest blocks (eyes open fixation) and the task blocks (visual and auditory blocks). Within the first encoding run, we found more translational head motion in the rest blocks (M = 0.038, SD = 0.014) compared to the task blocks (M = 0.032, SD = 0.012), as confirmed by a paired t-test (T(55) = 6.40, p < 0.001). Within the second encoding run, we also found more head motion in the rest blocks (M = 0.044, SD = 0.024) compared to the task blocks (M = 0.036, SD = 0.015), as confirmed by a paired t-test (T(55) = 3.80, p < 0.001). We did not find significant differences in head motion between the auditory and visual task blocks. As a second additional analysis, we assessed head motion in relation to age and gender, using the first resting-state session from experiment 1. We found a positive correlation (R = 0.398, p = 0.002) between age and mean translational motion in the resting-state condition. Using a two-sample t-test (T(2,54) = 1.66, p = 0.101), we found that women (M = 0.037, SD = 0.020) showed a trend for less translational motion compared to men (M = 0.046, SD = 0.021). As a third additional analysis, we compared head motion in the first and second half of each run. We found that the later half of each run was associated with more head motion (See

13

Supplemental Figure 1A). Also in each split-half, rest was associated with more head motion as compared to the task conditions.

Experiment 2: head motion in the UCLA Consortium for Neuropsychiatric Phenomics LA5c Study To investigate if the difference in motion between the resting-state and task conditions were specific to our passive memory encoding task –in which no motor responses were required– we assessed head motion in the LA5c study. We first examined the restingstate (REST) and task fMRI data in the healthy controls only. We observed the larger framewise displacement during REST (M = 0.179, SD = 0.122), followed by task conditions: WM (M = 0.177, SD = 0.109), ENC (M = 0.159, SD = 0.137), BART (M = 0.167, SD = 0.121), SWITCH (M = 0.153, SD = 0.095), RET (M = 0.148, SD = 0.101) and finally the least framewise displacement in STOP (M = 0.142, SD = 0.100). Figure 2A, top left, shows mean framewise displacement for each task condition in healthy controls. To assess if motion is statistically different between the conditions, we conducted two-sided paired t-tests between all combinations of REST and task in the healthy controls (see Table 1). We observed that REST was associated with significantly larger framewise displacement as compared to the STOP condition and showed a trend for RET and SWITCH. For percentile motion spikes, we found a mean of 0.259 (SD = 0.244) for REST as compared to the task conditions: WM (M = 0.268, SD = 0.257), ENC (M = 0.207, SD = 0.227), BART (M = 0.224, SD = 0.253), SWITCH (M = 0.207, SD = 0.229), RET (M = 0.194, SD = 0.232) and for STOP (M = 0.176, SD = 0.224). See Table 1 for two-sided paired t-tests. We 14

observed the greatest translational motion during REST (M = 0.048, SD = 0.032), followed by task conditions: WM (M = 0.042, SD = 0.024), ENC (M = 0.038, SD = 0.029), BART (M = 0.039, SD = 0.023), SWITCH (M = 0.038, SD = 0.026), RET (M = 0.036, SD = 0.024) and finally the least translational motion in STOP (M = 0.034, SD = 0.025). See Table 1 for two-sided paired t-tests. We observed that REST was associated with significantly more translational motion as compared to each of the task conditions. Within the task conditions only WM and the STOP task were significantly different. For rotational motion, we also found slightly higher mean values for REST (M = 0.565, SD = 0.495) followed by task conditions: WM (M = 0.561, SD = 0.591), BART (M = 0.478, SD = 0.652), ENC (M = 0.460, SD = 0.629), SWITCH (M = 0.443, SD = 0.400), STOP (M = 0.439, SD = 0.488) and finally the least rotational motion in RET (M = 0.414, SD = 0.474). However, with one exception (REST vs. RET) these values were not statistically different (Table 1). The resting-state run was shorter than the task runs. Therefore, we also conducted these analyses truncating each task run to be the same length as the restingstate run; so all runs were of equal duration. Across all conditions and diagnosis groups, this affected neither the actual findings nor the statistical significance (results not shown). Together, the analyses in the healthy controls of the LA5c data demonstrate that the differences in head motion between task and resting-state condition that we observed in Experiment 1 were not specific to either the memory encoding task nor the experimental setting, but generalize to an independent dataset acquired at another institute. Furthermore, the pattern of relatively increased head motion in the restingstate condition was similar regardless of whether task stimuli were presented on a

15

screen at the end of the magnet bore (Experiment 1 / Experiment 3) or via goggles (Experiment 2) === INSERT TABLE 1 HERE === === INSERT FIGURE 2 HERE ===

As a follow-up, we also explored head motion during the resting-state and task conditions in the clinical groups (Figure 2A). Similarly to the healthy control group, we found a pattern of greater framewise displacement in the resting-state condition than most task conditions. Also, the pattern of head motion across conditions was similar across the clinical groups, despite the overall higher motion in patients with bipolar disorder and schizophrenia (Figure 2A). Using a linear regression model, that included a within-subject term for task condition and between-subject terms for diagnosis, gender, and age, we confirmed that most task conditions were associated with smaller framewise displacement compared to resting-state condition (Table 2). We also found that a diagnosis of schizophrenia was associated with larger framewise displacements. With respect to age, the linear model additionally demonstrated larger framewise displacements with increasing age. We also included linear models with percentile motion spikes, translation and rotation (Table 2). In the linear model using framewise displacement, we did not observe an independent effect of gender, although the effect of gender was apparent in the percentile motion spikes and at trend level in rotation.

=== INSERT TABLE 2 HERE ===

16

As a second exploratory analysis, we assessed head motion in relation to diagnosis, age and gender. Using two-sample t-tests, we compared head motion in the resting-state sessions from the healthy controls (n = 131) and patients with ADHD (n = 44), bipolar disorder (n = 37) and schizophrenia (n = 50). The healthy controls (M = 0.179, SD = 0.122) and patients with ADHD (M = 0.166, SD = 0.088) showed smaller framewise displacement compared to patients with bipolar disorder (M = 0.216, SD = 0.112) and schizophrenia (M = 0.301, SD = 0.187). The difference between controls and ADHD was not statistically significant (T(2,173) = 0.666, p = 0.507), whereas the difference between healthy controls and patients with bipolar disorder was trend level (T(2,166) = 1.66, p = 0.099) and the difference between healthy controls and patients with schizophrenia was significant (T(2,179) = 5.17, p < 0.001). Next, we plotted the distribution of mean framewise displacement separately for women and for men across the LAC5 cohort including all conditions and diagnoses (Figure 2B). Across the LA5c cohort women (M = 0.180, SD = 0.120) showed smaller framewise displacement compared to men (M = 0.224, SD = 0.149) during the resting-state condition (T(2,260) = 2.60, p = 0.010). Next, we plotted the regression lines between age and the mean translation motion for the different task condition across the LA5C cohort (Figure 2C). Across the LA5c cohort, we also found a positive correlation (R(260) = 0.307, p < 0.001) between age and mean framewise displacement in the resting-state condition. Finally, we plotted the regression lines between age and the mean framewise displacement from the resting-state condition for the different diagnoses (Figure 2D). Additionally, we tested if a cognitive

17

task can be used to reduce head motion in a patient cohort. Therefore, we examined the mean framewise displacement during the resting-state and stop task (condition) from the healthy controls and schizophrenia patients (diagnosis), see bookends Figure 2A. The linear regression model indicated main effects of diagnosis (beta estimate = 0.123, SE = 0.021, df = 179, p < 0.001), condition (beta estimate = 0.037, SE = 0.010, df = 174, p < 0.001) and a significant interaction (beta estimate = 0.041, SE = 0.019, df = 174, p = 0.034), thus, indicating that a task can reduce head motion in a clinical cohort. These follow-up analyses in the resting-state condition are consistent with the linear models across all task conditions (Table 2). Note that Experiment 1 had a fixed order of conditions across participants, as the resting-state runs were always acquired as the first and last run. In Experiment 2, Day A and B were counter balanced, but the acquisition order on each day was similar. Therefore, we could not exclude the possibility that acquisition order had an effect on head motion. As a third additional analysis, we compared head motion in the first and second half of each run. We found that the later half of each run was associated with more head motion (see Supplemental Figure 1B). Also in each split-half, rest was associated with more head motion as compare to most task conditions.

Experiment 3: head motion during counterbalanced rest and task conditions To investigate if the difference in motion between the resting-state and task condition is influenced by acquisition order, we assessed framewise displacement, percentile motion spikes, translational and rotational head motion in a resting-state run (REST) and 18

several cognitive task runs (RET, WM, ENC, MOVIE), while counter balancing the acquisition order across participants. We used a repeated measures ANOVA with factors for condition (RET/WM/ENC/MOVIE/REST) and acquisition order (run 1 to 5).

We found the greatest mean framewise displacement during REST (M = 0.301, SD = 0.174), followed by ENC (M = 0.248, SD = 0.118), MOVIE (M = 0.244, SD = 0.120), WM (M = 0.226, SD = 0.097), and finally, the least translational motion during RET (M = 0.222, SD = 0.136; Figure 3A), as demonstrated by a main effect of condition (F(4,82) = 3.67, p = 0.009). With regard to acquisition order, we did not observe significant differences due to acquisition order across run one (M = 0.253, SD = 0.139), run two (M = 0.258, SD = 0.142), run three (M = 0.251, SD = 0.139), run four (M = 0.231, SD = 0.127) and run five (M = 0.249, SD = 0.126; see Figure 3B), as demonstrated by an absence of main effect of acquisition order (F(1,82) = 0.753, p = 0.38). We also did not find a significant interaction between task condition and acquisition order (F(4,82) = 0.406, p = 0.80). === INSERT FIGURE 3 HERE == We found more motion spikes during REST (M = 0.554, SD = 0.304), followed by MOVIE (M = 0.475, SD = 0.286), ENC (M = 0.498, SD = 0.303), WM (M = 0.434, SD = 0.242) and RET (M = 0.397, SD = 0.294), as demonstrated by a main effect of condition (F(4, 82) = 2.85, p = 0.029). Again, we found no effect of acquisition order across run one (M = 0.492, SD = 0.290), run two (M = 0.487, SD = 0.283), run three (M = 0.459, SD = 0.312), run four (M = 0.436, SD = 0.282) and run five (M = 0.486, SD = 0.288), as demonstrated

19

by the absence of a significant main effect (F(1, 82) = 0.788, p = 0.377) The interaction between task condition and acquisition order (F(4, 82) = 0.895, p = 0.471) was not significant. To assess if the percentile motion spikes were statistically different between the individual conditions, we conducted two-sided paired t-test between all combinations of rest and task conditions on the log-transformed values (see Table 3). We found the greatest mean translational motion during REST (M = 0.084, SD = 0.052), followed by ENC (M = 0.072, SD = 0.039), MOVIE (M = 0.070, SD = 0.039), WM (M = 0.061, SD = 0.029), and finally, the least translational motion during RET (M = 0.052, SD = 0.025), as demonstrated by a main effect of condition (F(4,82) = 7.45, p < 0.001). We did not observe significant differences due to acquisition order across run one (M = 0.068, SD = 0.042), run two (M = 0.069, SD = 0.035), run three (M = 0.067, SD = 0.042), run four (M = 0.066, SD = 0.039) and run five (M = 0.069, SD = 0.040) as demonstrated by an absence of main effect of acquisition order (F(1,82) = 0.038, p = 0.85). We also did not find a significant interaction between task condition and acquisition order (F(4,82) = 0.444, p = 0.78). We also conducted two-sided paired t-test between all combinations of rest and task conditions (see Table 3). These tests showed that mean translational motion during RET and WM is significantly lower than during REST but not significantly different between ENC and REST nor between MOVIE and REST. For rotational motion, we found no differences across REST (M = 0.57, SD = 0.44), MOVIE (M = 0.49, SD = 0.35), ENC (M = 0.50, SD = 0.35), WM (M = 0.45, SD = 0.19), and RET (M = 0.48, SD = 0.39), as demonstrated by the absence of a significant main effect (F(4, 82) = 1.17, p = 0.33). Similarly, we observed no differences due to acquisition order across run one (M = 0.47,

20

SD = 0.22), run two (M = 0.59, SD = 0.46), run three (M = 0.50, SD = 0.46), run four (M = 0.47, SD = 0.37) and run five (M = 0.48, SD = 0.35), as demonstrated by the absence of a significant main effect (F(1, 82) = 0.584, p = 0.45). We found no interaction between task condition and acquisition order (F(4, 82) = 1.25, p = 0.30). To further assess if rotation was statistically different between the individual conditions, we also conducted two-sided paired t-test between all combinations of rest and tasks and found no significant differences (see Table 3). These analyses extend the results from Experiment 1 and 2, and show that the resting-state is associated with more head motion than most cognitive tasks regardless of the order of acquisition. === INSERT TABLE 3 HERE ===

As an additional analysis, we assessed head motion in relation to age and gender, using the resting-state session from experiment 3. We found a positive trend (R = 0.340, p = 0.112) between age and mean translational motion in the resting-state condition, although this was not significant. Using a two-sample t-test (T (21,2) = 2.31, p = 0.031), we found that women (M = 0.064, SD = 0.037) showed less translational motion compared to men (M = 0.110, SD = 0.058). As a second additional analysis, we compared head motion in the first and second half of each run. We found that the later half of each run was associated with more head motion (see Supplemental Figure 1C). Also in each split-half, rest was associated with more head motion as compared to the task conditions.

21

Discussion In the present study, we observed less head motion under task than resting-state conditions across three experiments. In Experiment 1, using newly collected MRI data (n = 56), we observed more head motion during the resting-state conditions. To test whether our findings of less motion during task than rest were specific to our experimental setting, we conducted Experiment 2 using MRI data from the UCLA Consortium for Neuropsychiatric Phenomics study (LA5c; n = 290) and replicated the findings from Experiment 1. We found less head motion during several cognitive task conditions than during resting-state. Since both experiments had a semi-fixed order of conditions across participants, we could not rule out that a first or last scan during a session might have been particularly prone to head motion. Therefore, we conducted Experiment 3 (n = 23) using newly collected MRI data with a counter-balanced acquisition order. Again, we found less head motion in several cognitive task conditions than during resting-state. We did not observe an effect of acquisition order, consistent with previous work (Ferrazzi et al., 2014; Goto et al., 2015; Griffanti et al., 2014; Power et al., 2015; Tisdall et al., 2015; Yan et al., 2013). One of the conditions in Experiment 3 was watching selected scenes from the BBC nature documentary Planet Earth, and although head motion, as measured by framewise displacement, was numerically lower than during rest (Figure 3), this difference was not statistically significant. The reduced head motion during a movie, as observed by Vanderwal and colleagues (2015), versus our null finding may be explained by the difference in participants or difference in the movie content. We examined healthy older adults while Vanderwal et al. examined

22

children, who tend to move more. Furthermore, our movie also differed in several manners. It lacked a clear narrative, was mostly nonsocial and was presented without sound, which may have made it less engaging. In the context of previous work (Vanderwal et al., 2015), we speculate that the exact content of a movie, and how much it is able to capture the attention of an individual, might influence head motion similarly to different task conditions. In terms of compliance, anecdotally participants from Experiment 3 reported that they found watching the movie most enjoyable out of all conditions tested. In a larger cohort, the difference between rest and a movie is likely to also be statistically significant, and therefore a movie is useful, and perhaps often most practical, to increase subject compliance. In summary, in three independent experiments we demonstrate that cognitive engaging conditions tend to reduce head motion during MRI, despite the requirement to make motor responses via a button box.

One consideration is that our results might not generalize to every possible cognitive task because some other untested cognitive paradigms may induce more head motion than the specific tasks we tested. In Experiment 3, the pattern of reduced head motion was consistent across two active task conditions, with button responses, and a passive task without responses. Although each of the task conditions was associated with less head motion compared to the resting-state condition, with this sample size only the active conditions were significantly different. In the larger sample of Experiment 1, the difference in head motion between the passive encoding task and the resting-state was

23

significant. Thus, we found that both passive and active conditions are beneficial to reduce head motion. In the LA5c data from Experiment 2, after the resting-state condition, the spatial working memory capacity task (Montojo et al., 2013) showed the most head motion. It is possible that some other more cognitive task, for example an oddball task or a comical movie might induce greater head motion compared to the resting-state condition. In general, the resting-state condition can be considered as a relatively un-engaging cognitive condition and boredom or an increased level of awareness of minor or moderate discomfort, may be the reason for significantly more head motion during rest than during tasks. Future studies should clarify to what extend these findings generalize to cognitive tasks that we did not examine in this study.

A second consideration is that we only tested one resting-state condition in Experiment 1 and 3, namely, one in which the participants were instructed to keep their eyes open and fixated on a cross. In experiment 2, no fixation cross was presented and participants were simply asked to keep their eyes open. One prior study compared several commonly used resting-state conditions, including eyes-closed, eyes-open fixation and eyes-open without fixation (Patriat et al., 2013) and found no difference in the amounts of head motion in each resting-state condition. Therefore, we assume that our findings extend to eyes-closed or resting-state without a fixation cross, but this was not explicitly tested. Future studies in specific clinical cohorts could investigate further what task, or what specific resting-state conditions, are best suited to reduce group differences in

24

head motion.

One implication of our finding is that an engaging cognitive task can reduce head motion not only during fMRI, but also during acquisition of other types of scans, including T1weighted, T2-weighted,

FLAIR,

proton density-weighted,

and

diffusion

MRI.

Furthermore, other imaging modalities, for example Positron Emission Tomography, might benefit from offering study participants or patients a cognitive task during data acquisition. We know that head motion impacts data quality in, for example, T1weighted anatomical and diffusion MRI data (Alexander-Bloch et al., 2016; Pardoe et al., 2016; Reuter et al., 2015; Yendiki et al., 2014), and offering study participants a cognitive task may reduce head motion during those scans also. Yet, motion spikes, translational and rotational motion, as well as motion in different directions, can have distinct influences on the MRI signal and could interact with the physical attributes of the MRI sequences (Zaitsev et al. 2016). Thus, what task condition is best suited to reduce head motion (artifacts) might be different for function MRI (typically gradient echo EPI), diffusion MRI (pulsed gradient spin-echo EPI) or volumetric, anatomical imaging such as T1, T2 or FLAIR, and could further depend on parallel imaging (Polimeni et al., 2015), partial Fourier, k-space view ordering or several other MRI parameters. Furthermore, our results show that the type of cognitive tasks could impact the amount of head motion in groups of patients and controls and this can complicate direct comparisons between clinical groups. Our results suggest that cognitive tasks can help

25

reduce group differences in head motion. Therefore, a carefully selected cognitive task can potentially reduce spurious group differences in MRI data that may have been induced by head motion.

Using cognitive tasks during MRI acquisition also represents an opportunity to collect behavioral measures with a limited increase in participant burden. Especially in large-scale neuroimaging studies, the time requested from volunteers and patients is typically substantial (e.g. (Azmak et al., 2015; Caspers et al., 2014; Dagley et al., 2015; Hofman et al., 2015; Miller et al., 2016; Weiner et al., 2015). Acquiring behavioral measures during structural and diffusion MRI is an opportunity for investigators to reduce participant burden and probe cognitive systems. For example, translational researchers recently developed a “virtual treasure hunt” that essentially mimics the Morris water maze (Possin et al., 2016). These kinds of playful examinations are entertaining while at the same time measure aspects of memory function. Alternatively, more classic neuropsychological tests could be modified for computerized assessment and performed in the MRI scanner. This would help reduce the total study duration and could replace traditional neuropsychological examinations or questionnaires collected outside the MRI (Zygouris and Tsolaki, 2015).

Conclusions

26

Taken together these experiments demonstrate that participants who are cognitively engaged tend to make less head movements while inside the MRI scanner. Specifically, the instruction to lie still and look at a fixation cross, as in so-called restingstate conditions, is associated with more head motion than a range of cognitive tasks. Potentially, functional MRI studies in more-difficult-to-scan population, such as children, that aim to quantify the brain’s functional connectivity could benefit from reduced head motion afforded by using a cognitive task. However, task conditions are known to influence functional connectivity and segregation of task-evoked (extrinsic) and spontaneous (intrinsic) connectivity is not straightforward (e.g. (Cole et al., 2014; Fransson, 2006; Geerligs et al., 2015; Huijbers et al., 2013; Krienen et al., 2014; Northoff et al., 2010; Smith et al., 2009). Our results demonstrate that comparisons between task fMRI and resting-state fMRI may be further complicated by confounding head motion. Future studies should take this into account, for example by selecting participants with similar motion in both conditions (Van Dijk et al., 2012; Zeng et al., 2014). We conclude that data quality in none-functional MRI, including MRI in clinical setting, can benefit from engaging patients and participants with a cognitive task.

Acknowledgements The data in experiment two were provided by the UCLA Consortium for Neuropsychiatric Phenomics LA5c study and supported by NIH Roadmap for Medical Research

grants

UL1-DE019580,

RL1MH083268,

RL1MH083269,

RL1DA024853,

RL1MH083270, RL1LM009833, PL1MH083271, and PL1NS062410. We thank the

27

investigators Robert Bilder, Russell Poldrack, Tyrone Cannon, Edythe London, Nelson Freimer, Eliza Congdon, Katherine Karlsgodt, and Fred Sabb for sharing their data publically. We also thank Kerstin Micus, Sara Jansen and Jarmila Schmitz for data collection at the DZNE, Shahid Mohammad for help with data management and Natascha Merten for comments on the cognitive task paradigms.

References Alexander-Bloch, A., Clasen, L., Stockman, M., Ronan, L., Lalonde, F., Giedd, J., Raznahan, A., 2016. Subtle in-scanner motion biases automated measurement of brain anatomy from in vivo MRI. Hum. Brain Mapp. n/a–n/a. doi:10.1002/hbm.23180 Azmak, O., Bayer, H., Caplin, A., Chun, M., Glimcher, P., Koonin, S., Patrinos, A., 2015. Using Big Data to Understand the Human Condition: The Kavli HUMAN Project. Big Data 3, 173–188. doi:10.1089/big.2015.0012 Barch, D.M., Burgess, G.C., Harms, M.P., Petersen, S.E., Schlaggar, B.L., Corbetta, M., Glasser, M.F., Curtiss, S., Dixit, S., Feldt, C., Nolan, D., Bryant, E., Hartley, T., Footer, O., Bjork, J.M., Poldrack, R., Smith, S., Johansen-Berg, H., Snyder, A.Z., Van Essen, D.C., Consortium, F.T.W.-M.H., 2013. Function in the human connectome: Task-fMRI and individual differences in behavior. NeuroImage 80, 169–189. doi:10.1016/j.neuroimage.2013.05.033 Bilder, R.M., Sabb, F.W., Cannon, T.D., London, E.D., Jentsch, J.D., Parker, D.S., Poldrack, R.A., Evans, C., Freimer, N.B., 2009. PHENOMICS: THE SYSTEMATIC STUDY OF PHENOTYPES ON A GENOME-WIDE SCALE. NSC 164, 30–42. doi:10.1016/j.neuroscience.2009.01.027 Braver, T.S., Cohen, J.D., Nystrom, L.E., Jonides, J., Smith, E.E., Noll, D.C., 1997. A parametric study of prefrontal cortex involvement in human working memory. NeuroImage 5, 49–62. doi:10.1006/nimg.1996.0247 Caspers, S., Moebus, S., Lux, S., Pundt, N., Schütz, H., Mühleisen, T.W., Gras, V., Eickhoff, S.B., Romanzetti, S., Stöcker, T., Stirnberg, R., Kirlangic, M.E., Minnerop, M., Pieperhoff, P., Mödder, U., Das, S., Evans, A.C., Jöckel, K.-H., Erbel, R., Cichon, S., Nöthen, M.M., Sturma, D., Bauer, A., Jon Shah, N., Zilles, K., Amunts, K., 2014. Studying variability in human brain aging in a population-based German cohortrationale and design of 1000BRAINS. Front. Ag. Neurosci. 6, 149. doi:10.3389/fnagi.2014.00149 28

Cole, M.W., Bassett, D.S., Power, J.D., Braver, T.S., Petersen, S.E., 2014. Intrinsic and task-evoked network architectures of the human brain. Neuron 83, 238–251. doi:10.1016/j.neuron.2014.05.014 Congdon, E., Altshuler, L.L., Mumford, J.A., Karlsgodt, K.H., Sabb, F.W., Ventura, J., McGough, J.J., London, E.D., Cannon, T.D., Bilder, R.M., Poldrack, R.A., 2014. Neural activation during response inhibition in adult attention-deficit/hyperactivity disorder: Preliminary findings on the effects of medication and symptom severity. Psychiatry Research: Neuroimaging 222, 17–28. doi:10.1016/j.pscychresns.2014.02.002 Dagley, A., LaPoint, M., Huijbers, W., Hedden, T., McLaren, D.G., Chatwal, J.P., Papp, K.V., Amariglio, R.E., Blacker, D., Rentz, D.M., Johnson, K.A., Sperling, R.A., Schultz, A.P., 2015. Harvard Aging Brain Study: Dataset and accessibility. NeuroImage. doi:10.1016/j.neuroimage.2015.03.069 Fellner, M.-C., Volberg, G., Mullinger, K.J., Goldhacker, M., Wimber, M., Greenlee, M.W., Hanslmayr, S., 2016. UNCORRECTED PROOF. NeuroImage 1–13. doi:10.1016/j.neuroimage.2016.03.031 Ferrazzi, G., Murgasova, M.K., Arichi, T., Malamateniou, C., Fox, M.J., Makropoulos, A., Allsop, J., Rutherford, M., Malik, S., Aljabar, P., Hajnal, J.V., 2014. Resting State fMRI in the moving fetus: A robust framework for motion, bias field and spin history correction. NeuroImage 101, 555–568. doi:10.1016/j.neuroimage.2014.06.074 Fox, J., Weisberg, S., 2011. An R Companion to Applied Regression. Sage Publications. Fransson, P., 2006. How default is the default mode of brain function? Neuropsychologia 44, 2836–2845. doi:10.1016/j.neuropsychologia.2006.06.017 Geerligs, L., Rubinov, M., Cam-CAN, Henson, R.N., 2015. State and Trait Components of Functional Connectivity: Individual Differences Vary with Mental State. Journal of Neuroscience 35, 13949–13961. doi:10.1523/JNEUROSCI.1324-15.2015 Glover, G.H., Lee, A.T., 1995. Motion artifacts in fMRI: comparison of 2DFT with PR and spiral scan methods. Magn. Reson. Med. 33, 624–635. Goto, M., Abe, O., Miyati, T., Yamasue, H., Gomi, T., Takeda, T., 2015. Head Motion and Correction Methods in Resting-state Functional MRI. Magn Reson Med Sci. doi:10.2463/mrms.rev.2015-0060 Griffanti, L., Salimi-Khorshidi, G., Beckmann, C.F., Auerbach, E.J., Douaud, G., Sexton, C.E., Zsoldos, E., Ebmeier, K.P., Filippini, N., Mackay, C.E., Moeller, S., Xu, J., Yacoub, E., Baselli, G., Ugurbil, K., Miller, K.L., Smith, S.M., 2014. ICA-based artefact removal and accelerated fMRI acquisition for improved resting state network imaging. NeuroImage 95, 232–247. doi:10.1016/j.neuroimage.2014.03.034 Hadley, W., 2009. ggplot2: elegant graphics for data analysis. Springer New York. Helfinstein, S.M., Schonberg, T., Congdon, E., Karlsgodt, K.H., Mumford, J.A., Sabb, F.W., Cannon, T.D., London, E.D., Bilder, R.M., Poldrack, R.A., 2014. Predicting risky choices from brain activity patterns. Proc. Natl. Acad. Sci. U.S.A. 111, 2470–2475. doi:10.1073/pnas.1321728111 Hofman, A., Brusselle, G.G.O., Murad, S.D., Duijn, C.M., Franco, O.H., Goedegebure, A., Ikram, M.A., Klaver, C.C.W., Nijsten, T.E.C., Peeters, R.P., Stricker, B.H.C., Tiemeier, H.W., Uitterlinden, A.G., Vernooij, M.W., 2015. The Rotterdam Study: 2016 29

objectives and design update. Eur J Epidemiol 30, 661–708. doi:10.1007/s10654015-0082-x Huijbers, W., Schultz, A.P., Vannini, P., McLaren, D.G., Wigman, S.E., Ward, A.M., Hedden, T., Sperling, R.A., 2013. The encoding/retrieval flip: interactions between memory performance and memory stage and relationship to intrinsic cortical networks. J Cogn Neurosci 25, 1125–1165. doi:10.1162/jocn_a_00366 Jalbrzikowski, M., Carter, C., Senturk, D., Chow, C., Hopkins, J.M., Green, M.F., Galvan, A., Cannon, T.D., Bearden, C.E., 2012. Schizophrenia Research. Schizophrenia Research 142, 99–107. doi:10.1016/j.schres.2012.10.007 Jenkinson, M., Bannister, P., Brady, M., Smith, S., 2002. Improved Optimization for the Robust and Accurate Linear Registration and Motion Correction of Brain Images. NeuroImage 17, 825–841. doi:10.1006/nimg.2002.1132 Krienen, F.M., Yeo, B.T.T., Buckner, R.L., 2014. Reconfigurable task-dependent functional coupling modes cluster around a core functional architecture. Philosophical Transactions of the Royal Society B: Biological Sciences 369, 20130526–20130526. doi:10.1038/nrn2961 Lejuez, C.W., Read, J.P., Kahler, C.W., Richards, J.B., Ramsey, S.E., Stuart, G.L., Strong, D.R., Brown, R.A., 2002. Evaluation of a behavioral measure of risk taking: The Balloon Analogue Risk Task (BART). Journal of Experimental Psychology: Applied 8, 75–84. doi:10.1037//1076-898X.8.2.75 Miller, K.L., Alfaro-Almagro, F., Bangerter, N.K., Thomas, D.L., Yacoub, E., Xu, J., Bartsch, A.J., Jbabdi, S., Sotiropoulos, S.N., Andersson, J.L.R., Griffanti, L., Douaud, G., Okell, T.W., Weale, P., Dragonu, I., Garratt, S., Hudson, S., Collins, R., Jenkinson, M., Matthews, P.M., Smith, S.M., 2016. Multimodal population brain imaging in the UK Biobank prospective epidemiological study. Nature Neuroscience. doi:10.1038/nn.4393 Montojo, C.A., Ibrahim, A., Karlsgodt, K.H., Chow, C., Hilton, A.E., Jonas, R.K., Vesagas, T.K., Bearden, C.E., 2013. NeuroImage: Clinical. Elsevier 4, 392–402. doi:10.1016/j.nicl.2014.01.010 Northoff, G., Qin, P., Nakao, T., 2010. Rest-stimulus interaction in the brain: a review. Trends in Neurosciences 33, 277–284. doi:10.1016/j.tins.2010.02.006 Pardoe, H.R., Hiess, R.K., Kuzniecky, R., 2016. Motion and morphometry in clinical and nonclinical populations. NeuroImage 135, 177–185. doi:10.1016/j.neuroimage.2016.05.005 Patriat, R., Molloy, E.K., Meier, T.B., Kirk, G.R., Nair, V.A., Meyerand, M.E., Prabhakaran, V., Birn, R.M., 2013. The effect of resting condition on resting-state fMRI reliability and consistency: A comparison between resting with eyes open, closed, and fixated. NeuroImage 78, 463–473. doi:10.1016/j.neuroimage.2013.04.013 Peirce, J.W., 2007. PsychoPy—Psychophysics software in Python 162, 8–13. doi:10.1016/j.jneumeth.2006.11.017 Poldrack, R.A., Barch, D.M., Mitchell, J.P., Wager, T.D., Wagner, A.D., Devlin, J.T., Cumba, C., Koyejo, O., Milham, M.P., 2013. Toward open sharing of task-based fMRI data: the OpenfMRI project. Front. Neuroinform. 7, 12. doi:10.3389/fninf.2013.00012 Poldrack, R.A., Gorgolewski, K.J., 2015. OpenfMRI: Open sharing of task fMRI data. 30

NeuroImage. doi:10.1016/j.neuroimage.2015.05.073 Poldrack, R.A., Gorgolewski, K.J., 2014. Making big data open: data sharing in neuroimaging. Nature Neuroscience 17, 1510–1517. doi:10.1038/nn.3818 Polimeni, J.R., Bhat, H., Witzel, T., Benner, T., Feiweier, T., Inati, S.J., Renvall, V., Heberlein, K., Wald, L.L., 2015. Reducing sensitivity losses due to respiration and motion in accelerated echo planar imaging by reordering the autocalibration data acquisition. Magn. Reson. Med. 75, 665–679. doi:10.1002/mrm.25628 Possin, K.L., Sanchez, P.E., Anderson-Bergman, C., Fernandez, R., Kerchner, G.A., Johnson, E.T., Davis, A., Lo, I., Bott, N.T., Kiely, T., Fenesy, M.C., Miller, B.L., Kramer, J.H., Finkbeiner, S., 2016. Cross-species translation of the Morris maze for Alzheimer’s disease. Journal of Clinical Investigation 126, 779–783. doi:10.1172/JCI78464DS2 Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L., Petersen, S.E., 2012. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage 59, 2142–2154. doi:10.1016/j.neuroimage.2011.10.018 Power, J.D., Schlaggar, B.L., Petersen, S.E., 2015. Recent progress and outstanding issues in motion correction in resting state fMRI. NeuroImage 105, 536–551. doi:10.1016/j.neuroimage.2014.10.044 Reuter, M., Tisdall, M.D., Qureshi, A., Buckner, R.L., van der Kouwe, A.J.W., Fischl, B., 2015. Head motion during MRI acquisition reduces gray matter volume and thickness estimates. NeuroImage 107, 1–9. doi:10.1016/j.neuroimage.2014.12.006 Satterthwaite, T.D., Wolf, D.H., Loughead, J., Ruparel, K., Elliott, M.A., Hakonarson, H., Gur, R.C., Gur, R.E., 2012. Impact of in-scanner head motion on multiple measures of functional connectivity: Relevance for studies of neurodevelopment in youth. NeuroImage 60, 623–632. doi:10.1016/j.neuroimage.2011.12.063 Schonberg, T., 2012. Decreasing ventromedial prefrontal cortex activity during sequential risk-taking: an fMRI investigation of the balloon analog risk task 1–11. doi:10.3389/fnins.2012.00080/abstract Smith, S.M., Fox, P.T., Miller, K.L., Glahn, D.C., Fox, P.M., Mackay, C.E., Filippini, N., Watkins, K.E., Toro, R., Laird, A.R., 2009. Correspondence of the brain's functional architecture during activation and rest. Proceedings of the National Academy of Sciences 106, 13040–13045. Tisdall, M.D., Reuter, M., Qureshi, A., Buckner, R.L., Fischl, B., van der Kouwe, A.J.W., 2015. Prospective motion correction with volumetric navigators (vNavs) reduces the bias and variance in brain morphometry induced by subject motion. NeuroImage 127, 11–22. doi:10.1016/j.neuroimage.2015.11.054 Van Dijk, K.R.A., Sabuncu, M.R., Buckner, R.L., 2012. The influence of head motion on intrinsic functional connectivity MRI. NeuroImage 59, 431–438. doi:10.1016/j.neuroimage.2011.07.044 Vanderwal, T., Kelly, C., Eilbott, J., Mayes, L.C., Castellanos, F.X., 2015. ACCEPTED MANUSCRIPT. NeuroImage 1–25. doi:10.1016/j.neuroimage.2015.07.069 Visscher, K.M., Miezin, F.M., Kelly, J.E., Buckner, R.L., Donaldson, D.I., McAvoy, M.P., Bhalodia, V.M., Petersen, S.E., 2003. Mixed blocked/event-related designs separate transient and sustained activity in fMRI. NeuroImage 19, 1694–1708. 31

doi:10.1016/S1053-8119(03)00178-2 Wager, T.D., Lindquist, M., Kaplan, L., 2007. Meta-analysis of functional neuroimaging data: current and future directions. Social Cognitive and Affective Neuroscience 2, 150–158. doi:10.1093/scan/nsm015 Weiner, M.W., Veitch, D.P., Aisen, P.S., Beckett, L.A., Cairns, N.J., Cedarbaum, J., Green, R.C., Harvey, D., Jack, C.R., Jagust, W., Luthman, J., Morris, J.C., Petersen, R.C., Saykin, A.J., Shaw, L., Shen, L., Schwarz, A., Toga, A.W., Trojanowski, J.Q., Initiative, A.D.N., 2015. 2014 Update of the Alzheimer’s Disease Neuroimaging Initiative: A review of papers published since its inception. Alzheimer's & Dementia 11, e1–e120. doi:10.1016/j.jalz.2014.11.001 Yan, C.-G., Cheung, B., Kelly, C., Colcombe, S., Craddock, R.C., Di Martino, A., Li, Q., Zuo, X.-N., Castellanos, F.X., Milham, M.P., 2013. A comprehensive assessment of regional variation in the impact of head micromovements on functional connectomics. NeuroImage 76, 183–201. doi:10.1016/j.neuroimage.2013.03.004 Yendiki, A., Koldewyn, K., Kakunoori, S., Kanwisher, N., Fischl, B., 2014. Spurious group differences due to head motion in a diffusion MRI study. NeuroImage 88, 79–90. doi:10.1016/j.neuroimage.2013.11.027 Zaitsev, M., Akin, B., LeVan, P., Knowles, B.R., 2016. Author’s Accepted Manuscript. NeuroImage 1–28. doi:10.1016/j.neuroimage.2016.11.014 Zaitsev, M., Maclaren, J., Herbst, M., 2015. Motion artifacts in MRI: A complex problem with many partial solutions. J. Magn. Reson. Imaging 42, 887–901. doi:10.1002/jmri.24850 Zeng, L.-L., Wang, D., Fox, M.D., Sabuncu, M., Hu, D., Ge, M., Buckner, R.L., Liu, H., 2014. Neurobiological basis of head motion in brain imaging. Proc. Natl. Acad. Sci. U.S.A. 111, 6058–6062. doi:10.1073/pnas.1317424111 Zygouris, S., Tsolaki, M., 2015. Computerized Cognitive Testing for Older Adults: A Review. American Journal of Alzheimer's Disease and Other Dementias 30, 13–28. doi:10.1177/1533317514522852

Figure Legends Figure 1: Head motion in passive encoding task and resting-state conditions. The plot shows mean framewise displacement for functional runs. The runs are shown in acquisition order from left to right. The resting-state conditions (REST1/REST2) are shown in red and the passive memory encoding task conditions (ENC1/ENC2) in cyan.

32

The error bars show the standard deviation, the black line the median, the dots indicate outliers and the stars indicate p-values from a paired t-tests: ***p < 0.001, **p < 0.005.

Figure 2: Head motion in the LA5c Study. The upper panel (2A) shows boxplots of mean framewise displacement in millimeters (mm) for rest and task conditions grouped by diagnosis. The conditions are resting-state (REST) in red, working memory task (WM) in yellow, gambling task (RISK) in green, switching task (SWITCH) in blue and inhibition task (STOP) in purple. From left to right, head motion is shown in controls and patients with ADHD, bipolar disorder or schizophrenia. The error bars indicate the standard deviation, black line the median and dots indicate outliers. For visual purposes only, we capped the y-axis at 0.60 (this occluded 23 out of 1689 data points). These data are included in the boxplots and the statistical analyses. The lower left panel (2B) shows a histogram of mean framewise displacement across all conditions and diagnosis separately for both genders (male = M, female = F). The lower middle panel (2C) shows the linear regression lines of framewise displacement by age separately for each condition. REST is shown in red, WM in yellow, RISK in green, SWITCH in blue and STOP in purple. The lower right panel (2D) shows the linear regression lines of framewise displacement by age separately for each diagnosis. Healthy controls are shown in red, patient with ADHD in green, bipolar disorder in blue and schizophrenia in purple.

33

Figure 3: Head motion during counterbalanced rest and task conditions. The left plot (3A) shows mean framewise displacement by session type. From left to right, the active retrieval task (RET), the active working memory task (WM), the passive memory encoding task (ENC), the movie (MOVIE) and the resting-state condition (REST). The resting-state conditions are shown in red, the movie in blue and the memory encoding task conditions in cyan. The right plot shows mean framewise displacement by acquisition order (1-5). The error bars show the standard deviation, the black line the median and the dots indicate outliers.

34

Fig. 1

35

Fig. 2

36

Fig. 3

37

Tables with Captions Table 1: Paired t-tests between task and resting-state conditions in the healthy controls of the LA5C study. The functional runs (conditions) are denoted as resting-state (REST), spatial working memory capacity task (WM), encoding of paired associates memory task (ENC), balloon analog risk task (BART), task-switching paradigm (SWITCH), retrieval of paired associates memory task (RET) and the stop-signal task (STOP). The top left panel compares means framewise displacement, the bottom left percentile motion spikes, the top right shows translational motion and bottom right rotational motion. Comparisons with the resting-state conditions are shown in black and between task conditions in grey. ***p < 0.001, **p < 0.005, *p < 0.05 and #p < 0.10 (trending). Framewise Displacement WM

ENC

BAR T

-

-

-

W M

REST 0. 89 1

EN C

0. 21 3

0. 26 1

0. 37 7

0. 45 5

0. 06 8

#

0. 09 1

B A RT S W IT C H

RE T

0. 05 1

#

0. 06 8

ST O P

0. 00 8

* *

0. 01 2

#

#

*

0. 6 4 4

-

0. 7 0 8 0. 5 2 4 0. 2 7 2

0. 3 4 5 0. 2 4 4 0. 0 7 8

-

Translational Motion

SWI TCH

RET

-

-

W M

REST 0. 04 2 *

-

E N C

0. 00 6

* *

B A RT S W IT C H

0. 00 3

* *

-

-

0. 7 4 4 0. 4 1 # 6

-

-

RE T

0. 6 9 2

ST O P

38

0. 00 2 < 0. 00 1 < 0. 00 1

* * * * * * * *

WM

ENC

0. 3 4 6 0. 3 4 1

-

0. 3 1 1 0. 0 9 7 0. 0 3 0

BAR T

*

SWI TCH

RET

-

* -

*

-

0. 9 2 0

-

-

-

-

-

-

0. 9 6 7 0. 5 1 # 5 0. 3 2 * 7

0. 9 4 8 0. 4 1 3 0. 2 2 2

0. 4 4 8 0. 2 5 0

-

0. 7 9 4

*

Percentile Motion Spikes WM

ENC

BAR T

-

-

-

W M

REST 0. 68 6

EN C

0. 33 9

0. 18 9

0. 21 2

0. 09 9

0. 08 3

0. 03 3

B A RT S W IT C H

RE T ST O P

0. 05 7 < 0. 00 1

#

# * * *

0. 02 4 < 0. 00 1

#

*

* * * *

0. 8 8 4

-

0. 5 6 1 0. 3 9 0 0. 0 1 2

0. 6 2 3 0. 4 2 3 0. 0 0 * 7

-

Rotational Motion

SWI TCH

RET

-

-

W M

REST 0. 95 3

-

E N C

0. 16 2

B A RT S W IT C H

0. 18 8

-

-

0. 7 1 5 0. 0 * 2 * 9

-

-

0. 1 1 * 5

0. 06 7

#

RE T

0. 04 5

*

ST O P

0. 05 7

#

WM

ENC

BAR T

SWI TCH

RET

0. 1 7 8 0. 2 0 9

-

-

-

-

0. 8 1 6

-

-

-

-

-

-

0. 0 7 6 0. 0 5 0 0. 0 6 5

0. 8 1 # 9 0. 5 8 # 0 0. 7 7 # 4

0. 6 0 3 0. 3 9 9 0. 5 5 7

0. 7 0 4 0. 9 4 9

-

0. 7 4 6

Table 2. Summary of linear models on motion data from LA5c study. Four separate linear mixed models examined mean framewise displacement, percentile motion spikes translational motion and rotational motion. The main effects of task condition, diagnosis, gender and age were estimated in the same model without interaction terms. Estimates are in unstandardized values reflecting the differences with the reference. The conditions include resting-state (REST) contrasted with spatial working memory capacity task (WM), encoding of paired associates memory task (ENC), balloon analog risk task (BART), task-switching paradigm (SWITCH), retrieval of paired associates memory task (RET) and the stop-signal task (STOP). Diagnoses included healthy controls contrasted with attention deficit hyperactivity disorder (ADHD), bipolar disorder (Bipolar) and schizophrenia. Gender is females contrasted with males and age is modeled continuously in years. SE = standard error, df = degrees of freedom, ***p < 0.001, **p < 0.005, *p < 0.05 and #p < 0.10 (trending). Framewise Displacement Translational Motion Predict ors Condit ion

Refere nce

REST

Contrast

WM

Estim ate 0.006

SE 0.0 06

Df 14 15

39

P 0.35 6

Estim ate 0.006

SE 0.0 01

Df 14 15

P < 0.00 ** 1 *

ENC

0.009

0.0 07

14 15

0.19 5

0.007

0.0 02

14 15

BART

0.015

0.0 06

14 15

0.010

0.0 01

14 15

SWITCH

0.029

0.0 06

14 15

0.01 9 * < 0.00 ** 1 *

0.010

0.0 01

14 15

RET

0.014

0.0 07

14 15

*

0.008

0.0 02

14 15

0.021

14 15 26 2 26 2

0.015

Bipolar

0.0 06 0.0 18 0.0 19

** *

ADHD

0.042 0.003

0.0 01 0.0 05 0.0 05

14 15 26 2 26 2

26 2 26 2

0.005

0.0 05 0.0 03

26 2 26 2

26 2

0.04 2 < 0.00 1 0.86 3 0.26 0 < 0.00 1 0.12 9 < 0.00 1

0.001

0.0 00

26 2

STOP Diagn osis

Contro l

Gende r

Femal e

Age

Years

Schizophr enia

0.071

Male

0.019

0.0 17 0.0 13

0.003

0.0 01

0.002 0.010 ** *

** *

Percentile Motion Spikes Predict ors Condit ion

Refere nce REST

Contrast

0.012 0.009 0.170

SE 0.0 75 0.0 81 0.0 75

Df 14 15 14 15 14 15

RET

0.296 0.180

0.0 75 0.0 81

14 15 14 15

STOP

0.635

ADHD

0.080

Bipolar

0.578

0.0 75 0.2 11 0.2 28

14 15 26 2 26 2

WM ENC BART

SWITCH

Diagn osis

Contro l

Estim ate

40

P 0.87 3 0.91 1 0.02 4 < 0.00 1 0.02 7 < 0.00 1 0.70 5 0.01 2

0.023

* ** *

*

** * ** * ** * ** *

# ** *

** *

0.001 0.059

SE 0.0 32 0.0 35 0.0 33

Df 14 15 14 15 14 15

P 0.94 1 0.97 6 0.07 0 #

0.094 0.027

0.0 33 0.0 35

14 15 14 15

0.113 0.005 0.047

0.0 33 0.0 87 0.0 94

14 15 26 2 26 2

0.00 4 0.43 8 < 0.00 1 0.95 6 0.61 6

0.002

** *

** *

Rotational Motion Estim ate

*

< 0.00 1 < 0.00 1 < 0.00 1 < 0.00 1 < 0.00 1 0.60 7 0.05 2 < 0.00 1 0.12 1 < 0.00 1

** *

** *

Gende r

Age

Femal e

Years

Schizophr enia

0.407

0.2 10 0.1 53

26 2 26 2

0.899

Male

0.031

0.0 08

26 2

41

< 0.00 1 0.00 9 < 0.00 1

** *

0.142

** ** *

0.124

0.0 86 0.0 63

26 2 26 2

0.10 3 0.05 2 #

0.009

0.0 03

26 2

0.00 8 *

Table 3: Paired t-tests between task and resting-state conditions with counter-balanced acquisition order. Functional Runs (conditions) are denoted as resting-state (REST), movie (MOVIE), memory encoding task (ENC), working memory task (WM) and memory retrieval task (RET). The top left panel compares means framewise displacement, the bottom left percentile motion spikes, the top right shows translational motion and bottom right rotational motion. Comparisons with the resting-state condition are shown black and between task conditions in grey. **p < 0.005, *p < 0.05 and #p < 0.10 (trending).

Framewise Displacement

RET WM ENC MO VIE

RES MO T VIE 0.0 0.57 43 * 0 0.0 0.64 55 # 8 0.1 0.90 77 3 0.1 41 -

EN C 0.4 * 91 0.5 63

W M 0.9 12

Translational Motion

* RET

-

WM

-

-

-

-

ENC MO VIE

RES T 0.0 * 05 * 0.0 45 * 0.2 65 0.2 10

Percentile Motion Spikes

RET WM ENC MO VIE

RES MO T VIE 0.0 0.34 66 # 8 0.4 0.86 60 5 0.4 0.88 45 5 0.3 64 -

MO VIE 0.11 3 0.44 6 0.88 9 -

EN C 0.0 * 85 0.3 68

W M 0.4 # 07 -

-

-

-

-

Rotational Motion

EN C 0.2 79 0.9 80

W M 0.2 68

RET

-

WM

-

-

-

-

ENC MO VIE

42

RES T 0.3 72 0.2 81 0.4 98 0.4 52

MO VIE 0.88 9 0.74 4 0.94 0

EN C 0.8 29 0.6 88

W M 0.8 52

-

-

-

-

-

-

*

Highlights Resting-state conditions are associated with more head motion Task conditions are associated with less head motion Older adults and men move more during MRI acquisition Bipolar or schizophrenia patients move more than healthy controls Cognitive tasks can reduce head motion for improved MRI data quality

43