Abstract— By combining functional Magnetic Resonance Imaging and acquisition of cardiac pulsation during the execution of a cognitive task we identified areas where task- induced changes in brain activity correlated with individuals' cardiac RMSSD. We show the existence of a relationship between task - induced deactivations and RMSSD. I. INTRODUCTION The relation between brain function, as measured by noninvasive functional Magnetic Resonance Imaging (fMRI) and physiological fluctuations is one of the most debated topics in the last decade [1]. We used fMRI and an inter- individual differences analysis to identify brain regions where the magnitude of task-induced changes in brain activity correlated with individuals’ RMSSD during task performance (brain/behavior correlations). We then determined whether these regions showed task-induced activation or deactivation. We found that brain regions demonstrating brain/behavior correlations, notably including the insula and anterior cingulate cortex, tended to be task- deactivated. Thus, the magnitude of deactivation in task- deactive regions partially reflects modulation of autonomic nervous system activity. II. METHODS Participants: 11 participants (7 males, mean age = 24.6 yrs, std = 3.5) took part in the study. Brain imaging acquisitions: 1 structural (MPRAGE) image was acquired (spatial resolution = 1 mm isotropic, 176 sagittal slices). One fMRI scan was acquired as well (115 scans / 253 sec, temporal resolution = 2.2 sec, spatial resolution = 3x3x3 mm, 0.45 mm slice spacing, 37 axial, parallel to AC - PC slices). During the fMRI session, participants engaged in a mental arithmetic Continuous Performance Task (CPT), which had a 4-cycle on/off structure (ABABABAB), where A = 10 sec rest period; B = 30 sec task period (B). Cardiac data were recorded during the CPT using a photoplethysmograph placed on participants’ left forefinger (sampling frequency = 50 Hz). Cardiac data were evaluated manually and annotated to reflect correct R-R intervals. To define an autonomic index we first established the inter-beat interval (IBI) series in the task performance block (initial 30sec task block) and from the IBI series we derived a series reflecting the root mean square of subsequent differences (RMSSD, [2]) of the IBIs. This resulted in a single autonomic activity indicant for each participant. RMSSD Research supported by European Research Council (ERC) starting grant, NeuroInt to U. H V.Iacovella is with Fondazione Bruno Kessler, Italy (*corresponding author e-mail: iacovella@fbk.eu). Uri Hasson is with the Center For Mind and Brain Sciences, The University of Trento, Italy indexes beat-to-beat variation and reflects a mainly vagal HRV component. Neuroimaging analysis consisted of constructing two statistical parametric maps (SPMs) reflecting features of brain activity, and then relating the two maps. One SPM indicated task-related changes in brain activity. The other SPM was based on inter-individual differences, and identified brain regions where the magnitude of task-related activation correlated with participants’ RMSSD values. 1. Modeling: For each voxel (the 3mm 3 fMRI spatial sampling unit), activation was defined as the correlation between the timeline of the study (i.e., the 30sec on / 10 seconds off cycle) and the voxel’s time series. This was done via a regression model: Voxel timeseries = β * study_timeline + ε. Here β is the regression slope. 2. Group level activation SPM: After obtaining the voxel’s β in step #1 for each participant, a voxel-wise one- sample T-test evaluated whether the mean β for the voxel, across participants, departed from chance; i.e., 0 (statistical significance set at p < .005, T (10) > 2.71). Family-Wise Error (FWE) control for multiple comparisons was implemented via cluster-wise thresholding [3], which identifies contiguous clusters of statistically significant voxels (FWE p < .05 using cluster extent). This defined, on a group-level, task-activated or task-deactivated clusters (see Fig. 1A), and constitutes a validity check for the study, as its results should replicate prior paradigms. 3. Group level correlation SPM: For each voxel, we calculated the correlation (Pearson’s R) between participants’ β values and their RMSSD during task performance (β:RMSSD correlation henceforth). FWE was implemented as described above (single voxel p < .005 uncorrected; FWE controlled using cluster extent, p < .05). This identified clusters where all voxels showed a significant β:RMSSD correlation (see Fig. 1B). 4. Finally, we treated each cluster in which the β:RMSSD correlation was significant (step #3) as a functional ‘region of interest’. For each region we determined if it was associated with task-related activation or deactivation, by calculating the mean β in the cluster per participant, and conducting a T-test against 0. This indicated which of the clusters that showed β:RMSSD correlations were also significantly task-activated or deactivated. This analysis returns a matrix that partitions clusters with β:RMSSD correlations into four types depending on whether the correlation was positive/negative and whether voxels in these clusters tended to be task-active or task-deactive (Fig. 1C). Magnitude of Task-induced Deactivation of Insula and Anterior Cingulate Cortex is related to Inter-individual Differences in RMSSD Vittorio Iacovella*, Uri Hasson