Sex Differences in the Neural Correlates of Affective Experience

Yoshiya Moriguchi; Alexandra Touroutoglou; Bradford C. Dickerson; Lisa Feldman Barrett


Soc Cogn Affect Neurosci. 2014;9(5):591-600. 

In This Article

Material and Methods


Participants were 34 right-handed healthy volunteers (17 male: age range 19–78 years, M = 42.9, s.d. = 22.9; 17 female: age range 22–83 years, M = 43.2, s.d. = 22.9). These participants were originally recruited to study the affective significance of novelty across the lifespan (Moriguchi et al., 2011). In this article, however, we focused on completely different hypotheses, and the analyses reported here do not overlap with those that have been previously published. We already confirmed that there were no age effects on the present results (data not shown; available on request), and therefore, the age effect was not considered in this study. The original sample contained more women than men (Moriguchi et al., 2011), thus the numbers of men and women were not balanced to test the sex difference hypothesis. Therefore, we randomly sampled to have an equal number of participants of men and women. (Note that both the behavioral and imaging results are essentially identical if the full sample is used.) There was no difference in age between the male and female groups [T(32) = 0.037, P = 0.97].

Affective Picture Stimuli

For the fMRI task, 132 full-color images were selected from the International Affective Picture System (IAPS; Lang et al., 1997) for each valence (positive, negative and neutral). Positive and negative pictures were equated for intensity [based on the norms (Lang et al., 2008)]. The task was run using E-Prime experimental software (Psychology Software Tools, Pittsburgh, PA) on a PC, from which images were projected onto a screen in the magnet bore. Participants viewed images through a mirror mounted on the head coil.

Scanning Procedure for Task-Related fMRI

The imaging paradigm consisted of four event-related fMRI runs. During scanning, participants rated their subjective experiences of affective arousal as they viewed each image using a three-point scale (1 = low, 2 = mid and 3 = high) and answered with a three-button response box. These individual arousal ratings were used as a measure of momentary evaluation of their own affective states induced by the IAPS images with different valences in the scanner. Each run was 340 s in length and each image was presented for 3.5 s, with a stimulus onset asynchrony that varied from 4 to 16 s.

Image data were acquired using a Siemens Magnetom Trio Tim 3T whole body high-speed imaging device equipped for echo planar imaging (EPI) (Siemens Medical Systems, Iselin, NJ) with a 12-channel gradient head coil. Expandable foam cushions restricted head movement. After an automated scout image was acquired and shimming procedures were performed to optimize field homogeneity, a T1-weighted 3D MPRAGE sequence (TR/TE/flip angle = 2.53 s/3.39 ms/7°) with an in-plane resolution of 1.3 × 1.0 mm and 1.3 mm slice thickness was collected. Blood-oxygenation-level-dependent fMRI images were acquired using a gradient echo T2*-weighted sequence (TR/TE/flip angle = 2.0 s/30 ms/90°). At the beginning of each scan run, four volumes were acquired and discarded to allow longitudinal magnetization to reach equilibrium.

Processes and Analyses of Imaging Data for Task-Related fMRI

Image processing was carried out using Statistical Parametric Mapping software (SPM5, Wellcome Department of Imaging Neuroscience, London, UK). The EPI images were realigned to the first image of the time series, and co-registered to the participants' T1-weighted images. Then the T1 images were transformed to a template brain in MNI (Montreal Neurological Institute) stereotactic space using high-dimensional warping implemented in SPM5. The parameters for the transformation were applied to the co-registered EPI images. The normalized images were smoothed by a 6-mm FWHM Gaussian kernel.

To test regionally specific effects, a first-level analysis was computed using the general linear model (GLM) with the canonical hemodynamic response function in SPM package, which models typical event-related hemodynamic response. Neural responses associated with the magnitude of momentary emotional experiences (i.e. subjective arousal ratings in the scanner) were assessed on event-by-event basis by parametric modulation analysis, which allows us to estimate the degree to which the hemodynamic response is modulated by the arousal rating in the design matrix. That is, the subjective rating scores were first mean-centered to zero across a run, and each hypothesized event-related hemodynamic response was modulated (multiplied) by the subjective rating score relevant to that event. The modulated hemodynamic time-course models were then concatenated to the nonmodulated conventional hemodynamic GLM model. The beta values estimated for this modulated term—which represent the correlational effect of the subjective rating with the hemodynamic responses—were calculated with this GLM for each individual subject across whole brain, and were then entered into a second-level random-effect analysis (Friston et al., 1999) to allow for population inferences and group analysis to test hypotheses related to sex differences.

Theoretically, we could have done this parametric modulation analysis in each negative, positive and neutral picture condition separately, but valence and subjective arousal are strongly associated, as negative and positive pictures were scored higher on arousal than neutral pictures. Within a single valence, there was a restricted range in arousal (e.g. participants often scored all the neutral pictures in a run 'low'). Therefore, we did not segregate the data into separate valence conditions but analyzed the data including all the three conditions together to maximize the range of subjective arousal ratings. We also ran conventional analyses of sex difference of neural activity in response to affective pictures. The method and results are provided in Figure S1, S2, S3, S4, and Table S1 in Supplemental Information.

Region of Interest Definition

The primary hypotheses tested focused on two nodes of the salience network (bilateral AI and dACC) and the visual cortex (V1–V5). We used a functional-anatomic approach to identify specific regions within each of these cortical areas in the following manner. First, the anatomic areas were identified as bilateral AI, dACC and V1–V5 using the Automated Anatomical Labeling atlas (Tzourio-Mazoyer et al., 2002) from the WFU PickAtlas software (Maldjian et al., 2003). Next, within each of these anatomic areas, we identified the significantly activated clusters from the aforementioned parametric modulation analysis including all participants in the present study (height and extent threshold at P < 0.05, FDR-corrected). Each of these functional-anatomically defined clusters was set as region of interest (ROI). We then extracted, for each subject, the mean parameter estimates (beta values) from the parametric modulation analysis from each ROI using MarsBar software (; Brett et al., 2002). These values were then compared statistically between male and female groups.

In addition to this hypothesis-driven ROI approach, we also performed a whole brain analysis, and verified whether the clusters with significant sex difference from this whole brain analysis were included in our ROIs (the bilateral AI, dACC and V1–V5). For this analysis, we set alpha at P < 0.001 (uncorrected) for peak voxels, and specified a lower limit of 10 contiguous active voxels to constitute a cluster.

We found that, using the method just described for localizing the ROIs, the AI ROI was too large to determine the specific localization of activation within the bilateral AI (i.e. differentiation between ventral or dorsal AI [dAI]). Thus, we also confirmed the detailed localization (i.e. clusters) of sex difference in the parametric modulation analysis by visualizing the activation map [the significant clusters within the AI (P < 0.001, uncorrected, k = 10)] rendered on the surface of the insula (Figure 2A). These clusters were then used to extract parameter estimates as previously stated using MarsBar software, and the estimates in each sex group were illustrated graphically.

Figure 2.

Sex differences in the association between neural responses and subjective arousal ratings. (A) The two upper left brain maps show the clusters within the insula that have significantly stronger association in women than in men between neural responses and subjective arousal ratings (P < 0.005 for the purpose of illustration). Blue line divides insula into dorsal and ventral sectors. Note that these regions were mostly located in more ventral or mid part of AI, not dorsal AI. The upper right brain map shows the clusters within the left primary visual area (V1) that have significantly stronger association in men than in women between neural responses and subjective arousal ratings (P < 0.005 for the purpose of illustration). The bar graphs below the brain map show parameter estimates of modulation effects of subjective ratings, with error bars indicating one standard error of the mean. Rt, right hemisphere; Lt, left hemisphere; vAI, ventral anterior insula (AI); mAI, mid AI; red bar (F), female; blue bar (M), male. Each coordinate below the graphs indicates the peak MNI coordinate (x, y, z mm) of each cluster. (B) Whole brain analysis of correlation between neural response to all IAPS pictures and subjective arousal rating (parametric modulation analysis) in men compared with women. Stronger correlation in women than men (t range 1.8–4.0 for illustration) is colored in red-yellow, and stronger correlation in men than women is colored in blue-green. Greater anterior insula (AI) extending to middle insula (MI) activation was confirmed in women than in men, whereas men show greater V1 association than women. Additional exploratory results demonstrate that women had greater association with arousal in other affective areas including the amygdala (AMG) extending to parahippocampal cortex than men, whereas men had greater association with arousal in ventrolateral prefrontal cortex (vlPFC) (see Table 1 for complete cluster table).

Functional Connectivity Analysis During Task Engagement

Next, we conducted a functional connectivity analysis between dACC and ROIs in the right and left AI. We used the functional connectivity toolbox version 12 ( To define seed ROIs, we used previous meta-analyses (Wager and Barrett, 2004; Wager et al., 2008; Mutschler et al., 2009; Kurth et al., 2010; Deen et al., 2011). The coordinates reported in all these studies overlap with each other, so we selected one study (Kurth et al., 2010) that classified the insular coordinates into several functionally different categories. We selected eight insular MNI coordinates from the category 'emotion' [as affective regions; (−31, 24, −4), (−30, 12, 10), (42, 15, −3),(39, 7, 0), (28, 17, −15), (−31, 24, −6), (42, 8, −6) and (44, 10, −4)] and three from 'attention' [as cognitive/regulatory regions; (−35, 18, 7), (−33, 18, −5) and (36, 19, 3)] to cover a maximally broad area of AI (including both ventral and dorsal portions). Then we created spherical ROIs (4 mm radius) centered at each coordinate using MarsBar software (see Figure 3A). As our goal was to examine sex differences in connectivity between AI and ACC, we performed the functional connectivity analysis using a mask to restrict the search space to voxels in the right and left ACC ROIs (defined by Automated Anatomical Labeling system). The time course of the blood-oxygenation-level-dependent signal in the ROIs in each participant was extracted using the functional connectivity toolbox software, and the data were band-pass filtered with a frequency band from 0.008 to 0.16 Hz, which was confirmed by a spectral analysis as an effective frequency range of the theoretical task-related hemodynamic design in the present study. The correlations of time courses between the insula seed ROIs and the voxels within ACC ROIs were estimated by regression analyses co-varied with the main event-related task effect (i.e. the modeled hemodynamic response to observation of IAPS images during arousal ratings) and other nuisance variables (including white matter, cerebrospinal fluid signal and head motion parameters). Thus, the resulting maps represent voxels within the ACC in which the hemodynamic response during task-related activity is correlated with that of the AI. Next, for group comparison, we identified the statistically significant clusters in which there was a sex difference (height threshold P < 0.001 uncorrected, extent threshold k = 10).

Figure 3.

Sex differences during task performance in the functional connectivity between insula ROIs and the dACC. (A) Seed ROI locations in the AI for functional connectivity analyses. The circles on the AI surface show 11 spherical ROIs. The coordinates at the center of each sphere are the ones reported in a meta-analysis (Kurth et al., 2010). The green and yellow ROIs are associated with 'emotion', and blue ROIs are associated with 'attention'. In 'emotion' coordinates, green ROIs include multimodal integrated regions that might be also activated by the tasks for other different categories (i.e. not 'specific' to emotional tasks). On the other hand, yellow ROIs were the regions specific to emotional tasks (Kurth et al., 2010). (B) The upper figures show the ROIs rendered on brain surface maps (to illustrate insula, a part of temporal/frontal/parietal regions is cropped) and the clusters in ACC with a statistically significant sex difference (men vs women) of functional connectivity (during task performance) to the respective anterior insula (AI) ROI labeled with a colored dot. A blue line divides the insula into dorsal and ventral parts. Functional connectivity from each seed ROI [see (A)] to ACC was calculated. The three ROIs in the AI shown on the map were those that showed statistically significant stronger functional connectivity to the dACC in men than in women. Note that these seed regions in the AI were mostly located in the mid or dorsal AI (not ventral AI). The graphs below the brain map show mean functional connectivity (correlation coefficient r) in each sex group, with error bars indicating one standard error of the mean. Rt, right side; Lt, left side; dAI, dorsal AI; mAI, mid AI; red bar (F), female; blue bar (M), male. Each coordinate described below the graph represents center MNI coordinate (x, y, z, mm) at each ROI.