واکنش منحصر به فرد در طول تماشای فیلم با تفاوت سنی در کنترل توجه

Abstract Much is known about how age affects the brain during tightly controlled, though largely contrived, experiments, but do these effects extrapolate to everyday life? Naturalistic stimuli, such as movies, closely mimic the real world and provide a window onto the brain's ability to respond in a timely and measured fashion to complex, everyday events. Young adults respond to these stimuli in a highly synchronized fashion, but it remains to be seen how age affects neural responsiveness during naturalistic viewing. To this end, we scanned a large (N = 218), population-based sample from the Cambridge Centre for Ageing and Neuroscience (Cam-CAN) during movie-watching. Intersubject synchronization declined with age, such that older adults' response to the movie was more idiosyncratic. This decreased synchrony related to cognitive measures sensitive to attentional control. Our findings suggest that neural responsivity changes with age, which likely has important implications for real-world event comprehension and memory.

. Introduction Movies have the power to transport your mind from the narrow, impersonal bore of an magnetic resonance imaging (MRI) magnet to a world more synonymous with everyday life, replete with sights, sounds, and language. Despite their complexity, these naturalistic stimuli tend to drive neural activation in the same way across individuals (Hasson et al., 2004 and Hasson et al., 2010), suggesting that our experience of real-world events is largely shared. Although responding in the same way as others is not a perquisite for perception, it does seem to reflect the optimal response to a given stimulus, in that asynchronous responding tends to relate to poor comprehension (Hasson et al., 2009) and memory (Hasson et al., 2008a). This may be because synchronized activity reflects shared attention to the most relevant stimulus in the environment, as nominated by the majority. Empirical work supports this view, as (1) participants' eye movements tend to track the same focal item within each shot (Dorr et al., 2010 and Hasson et al., 2008b), (2) materials that are rated as more engaging tend to yield the highest degree of neural synchronization (Dmochowski et al., 2014), and (3) disruptions to story narrative, and ergo meaning, tend to reduce overlap across participants (Dmochowski et al., 2012 and Hasson et al., 2008b). Although previous work has mainly focused on aspects of the stimulus itself that make it more or less captivating, these findings suggest that individual differences in attentional control should also predict intersubject synchronization. Individuals with greater attentional control should be better able to maintain focus on the movie and should therefore show higher synchronization with others. Individuals of all ages differ in their ability to control the focus of attention, but on average, this ability tends to decline with age (Hasher and Zacks, 1988). For instance, relative to younger adults, older adults are less able to ignore distracting information (May, 1999), prevent reflexive eye movements toward irrelevant onsets (Campbell and Ryan, 2009), and to sustain attention to a task to produce consistent response times (RTs; Hultsch et al., 2002). They also experience more interference from internally generated distraction, such as competing responses during memory retrieval (Healey et al., 2013), and these intrusive thoughts affect their ability to stay on task, especially as task demands increase (Persson et al., 2007 and Sambataro et al., 2010). This suggests that even during task-free, naturalistic viewing, older adults should be less able to sustain attention to a movie and prevent interference from both external (e.g., scanner noise; Stevens et al., 2008) and internal distraction (Mishra et al., 2013). As a result, they should show altered patterns of neural responsiveness and reduced synchronization with others during naturalistic viewing. To test this hypothesis, we obtained functional magnetic resonance imaging (fMRI) data while participants from a large population-based cohort (aged 18–88 years) watched Alfred Hitchcock's “Bang! You're Dead”, a movie previously shown to yield widespread correlations throughout the cortex (Hasson et al., 2010). Functional networks were derived using independent components analysis (ICA; Beckmann and Smith, 2005 and Naci et al., 2014), a data-reduction technique that decomposes the continuous fMRI time series into a set of components (or neural networks), each with an associated spatial map, group-average timecourse, and set of individual timecourses reflecting the level of activation within a given network by a given participant at each time point. A measure of synchronization was then derived for each participant, based on the correlation between their individual timecourse and that of the group. Given age-related declines in attentional control, we expected older adults' network timecourses to show less synchronization with the group-average timecourse. To test the reproducibility of our main finding of interest (i.e., decreased temporal synchrony with age), we also ran a supplementary region of interest (ROI) analysis looking at intersubject correlations in the raw fMRI timecourses of a large number of small ROIs (Craddock et al., 2012). Furthermore, we expected intersubject synchronization to be positively related to measures which are sensitive to attentional control. Specifically, we expected higher synchronization to be associated with better performance on a test of fluid intelligence (widely thought to depend on attentional control; Duncan, 2013, Engle et al., 1999 and Kane and Engle, 2002), but not on measures of crystallized intelligence (or semantic knowledge). Crystallized intelligence is less dependent on attentional control (Cole et al., 2012) and shows a different life span trajectory (Horn and Cattell, 1967). We also gave participants a speeded reaction time (RT) task, in which they had to respond as quickly as possible to visual cues. Here, we expected higher synchronization to be associated with less variable RTs, rather than faster responding per se, as previous work suggests that RT variability is a stronger predictor of attentional control than mean RT itself (MacDonald et al., 2009 and Stuss et al., 2003)

3. Results 3.1. ICA results The ICA analysis estimated 56 components, each with an associated spatial map, timecourse of activation, and set of participant loading values, which indicate the degree to which each participant expresses a given spatiotemporal pattern. The spatial maps for the 10 most strongly expressed of these components are shown in Fig. 1 and are labeled according to the most highly correlated template(s) from Shirer et al., 2012; (Supplementary Table 2). These components closely resemble previously established resting state networks (Allen et al., 2011 and Shirer et al., 2012), correlating most highly with the (1) auditory, (2) visuospatial, (3) language and/or dorsal default mode network (DMN), (4) visual, (5) posterior salience, (6) visual, (7) visual and/or ventral DMN, (8) language, (9) ventral DMN, and (10) language and/or ventral DMN networks. Examining the correlations between pairs of network timecourses, we see that these networks clustered into 2 groups (see Fig. 2A). Auditory and language components were highly correlated during the movie and anticorrelated with visual and attentional networks, whereas the latter networks were highly intercorrelated. Moreover, the auditory and language networks covaried with talking during the movie (Fig. 2B), suggesting that the ICA approach successfully identified neurocognitive networks which correspond to meaningful events within the movie. Participant loading values, or network expression, significantly declined with age across all 10 components of interest (Fig. 3). Moreover, and in line with our predictions, intersubject synchronization, or correlation to the group-average timecourse, declined with age across all components, controlling for education (see Fig. 4, Supplementary Table 3). Figure 5 illustrates this for 1 representative component (the auditory network) by plotting the individual timecourses of the 30 youngest and 30 oldest participants, along with the group-average timecourse and the timecourse of talking during the movie. Although younger adults clearly responded in a synchronized way to the movie, older adults were much more variable in their response. This age difference remains even if we partial out the effects of motion (Supplementary Fig. 1), suggesting that it cannot be explained alone by age differences in head motion during the scan (Power et al., 2012, Satterthwaite et al., 2012 and Van Dijk et al., 2012). Partial correlation between age and individual loading values (controlling for ... Fig. 3. Partial correlation between age and individual loading values (controlling for education) for the 10 components shown in Fig. 1. Error bars represent 95% bootstrap confidence intervals. Abbreviations: dDMN, dorsal default mode network; Lang, language; Post, posterior, vDMN, ventral default mode network. Figure options Scatterplots showing the correlation between age and the correlation of ... Fig. 4. Scatterplots showing the correlation between age and the correlation of individual network timecourses to the group-average timecourse for each of the 10 components of interest shown in Fig. 1. Corresponding correlation values shown in Supplementary Table 3. Figure options Individual timecourses from the auditory network for the 30 youngest and 30 ... Fig. 5. Individual timecourses from the auditory network for the 30 youngest and 30 oldest participants (each line represents a single participant), with the group-average timecourse and the timecourse of talking during the movie plotted in between. Figure options We also calculated an equivalent metric of deviation from the mean in the spatial dimension, by first using dual regression (Damoiseaux et al., 2012 and Filippini et al., 2009) to obtain participant-specific spatial maps and then calculating the correlation between these maps and the original component map, for each component of interest separately. As with the temporal dimension, correlation to the group-average spatial map declined with age across all components (see Supplementary Table 3), although the absolute decline in spatial correlation across the life span was quite small (averaged across the 10 networks, correlations declined from 0.78 in decile 1 to 0.74 in decile 7; see Supplementary Fig. 2). 3.2. ROI analysis results To ensure that the observed age-related decrease in intersubject synchronization was not due to a difference in the network structure of older adults, or specific to the method we applied, we ran a supplementary ROI analysis. To this end, we extracted the mean timecourse from 840 ROIs (Craddock et al., 2012) and then for each ROI separately, calculated the correlation between an individual's timecourse and the mean timecourse of all other participants. Similar to previous work using a voxelwise approach (Hasson et al., 2004), this method yielded robust intersubject correlations throughout the cortex (Fig. 6A), with the strongest synchronization in primary visual and auditory regions and weaker synchronization in sensorimotor and rostral frontal cortex. Importantly, we replicate the main finding of interest from our ICA analysis, showing that intersubject correlations declined with age (after controlling for education) in several regions (Fig. 6B), including middle occipital cortex, intraparietal sulcus, the temporal poles, anterior cingulate, and left superior frontal lobe. There were no regions in which synchrony increased with age. Furthermore, if we split the sample into 2 groups (younger and older), correlate individuals to their age-matched peers, and then compare the strength of these correlations between groups (including education as a covariate), a very similar pattern of results emerge (Fig. 6C), suggesting that older adults are just as dissimilar to each other as they are to younger adults. Taken together, these findings suggest that intersubject synchronization during naturalistic viewing declines with age. Intersubject synchrony results from the region of interest analysis. Significant ... Fig. 6. Intersubject synchrony results from the region of interest analysis. Significant intersubject correlations were seen throughout the cortex [(A); Bonferroni corrected for multiple comparisons]. Synchronization was negatively correlated with age (controlling for education) in several regions [(B); blue thresholded at p < 0.001; violet regions survive Bonferroni correction, p < 0.05/840]. A similar pattern of results is seen panel C if younger (<50 years) and older adults (>65 years) are instead correlated to their age-matched peers and a group contrast is performed with education as a covariate (young > old, blue p < 0.001, violet p < 0.05/840), suggesting that older adults are just as dissimilar from each other as they are to younger adults. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) Figure options 3.3. Intersubject synchronization and measures of attentional control Means and SDs for the cognitive measures are shown per decile in Supplementary Table 1. As expected from previous findings, aging was associated with lower performance on tests sensitive to attentional control, with older adults scoring lower on the Cattell test of fluid intelligence (r = −0.65, p < 0.001) and more variably (r = 0.58, p < 0.001), as well as slower (r = 0.65, p < 0.001), on the Choice RT task, all controlling for education. In contrast, crystallized intelligence increased with age, controlling for education (r = 0.22, p < 0.01). To test whether individuals with greater attentional control show higher synchronization with the group during the movie, we first calculated a single synchronization score for each participant as their mean correlation to the group-average timecourse across the 10 components from our ICA analysis. We then performed separate regression analyses for each of the cognitive measures of interest (i.e., fluid intelligence and choice RTISD as proxies for attentional control; crystallized intelligence, and choice RTmean as control measures, not expected to relate to synchronization). Our predictor variables were age, synchronization score, and the age × synchronization interaction, with education as a covariate. Results from these regressions are shown in Table 1 and scatter plots depicting the relationship between synchronization and each cognitive variable (subdivided into decile subgroups for visualization purposes) are shown in Fig. 7. Higher intersubject synchronization related to lower choice RTISD across the entire sample (indicated by the significant main effect of synchronization in Table 1) and to higher fluid intelligence among middle-to older-aged adults (indicated by the significant interaction between age and synchronization in Table 1). In contrast, synchronization was not related to crystallized intelligence and, if anything, related positively to choice RTmean due to slightly slower responding by young adults who showed greater synchrony with the group (Fig. 7). These findings suggest that individual differences in attentional control relate to intersubject synchronization during natural viewing, particularly among older adults. Table 1. Regressions predicting attentional control as a function of age, education, synchronization score, and the age by synchronization score interaction Outcome variable Predictor variables Model R2 β [95% CI] SE β t p Fluid intelligence 0.56 Agea −0.23 [−0.27, −0.19] 0.02 −10.63 <0.0001 Synchronization 1.86 [−6.20, 9.91] 4.09 0.45 0.65 Age × synchronizationa 0.55 [0.03, 1.08] 0.27 2.09 <0.05 Educationa 1.79 [1.14, 2.44] 0.33 5.43 <0.0001 Crystallized intelligence 0.19 Agea 0.08 [0.04, 0.13] 0.02 3.64 <0.001 Synchronization 5.59 [−2.44, 13.63] 4.08 1.37 0.17 Age × synchronization 0.002 [−0.51, 0.52] 0.26 0.006 0.99 Educationa 2.30 [1.52, 3.09] 0.40 5.76 <0.0001 Choice RTISD 0.66 Age 0.0002 [−0.0002, 0.0006] 0.0002 0.85 0.40 Synchronizationa −0.07 [−0.12, −0.02] 0.03 −2.64 <0.01 Age × synchronization −0.001 [−0.005, 0.002] 0.002 −0.75 0.46 Choice RTmeana 0.28 [0.19, 0.37] 0.05 6.05 <0.0001 Education 0.0002 [−0.005, 0.005] 0.003 0.08 0.93 Choice RTmean 0.70 Agea 0.003 [0.002, 0.004] 0.0005 4.84 <0.0001 Synchronizationa 0.16 [0.04, 0.28] 0.06 2.70 <0.01 Age × synchronization −0.005 [−0.02, 0.006] 0.006 −0.88 0.38 Choice RTISDa 1.60 [1.25, 1.92] 0.17 9.34 <0.0001 Education −0.004 [−0.02, 0.009] 0.007 −0.67 0.51 Statistical models were computed separately for each cognitive measure. Results represent regression parameters for a given cognitive task predicted by age, education, synchronization score, and the age × synchronization score interaction. To control for the high degree of correlation between RTmean and RTISD, the model predicting choice RTISD included choice RTmean as a covariate and the model predicting choice RTmean included choice RTISD as a covariate. Beta values reflect unstandardized regression coefficients. Key: CI, confidence interval; ISD, intraindividual standard deviation; RT, reaction time; SE, standard error. a Significant predictors. Table options Scatterplots showing the relationship between intersubject synchronization and ... Fig. 7. Scatterplots showing the relationship between intersubject synchronization and measures sensitive to attentional control (i.e., fluid intelligence and RT variability), as well as control measures less dependent on attentional control (i.e., crystallized intelligence and mean RT). To show how this relationship changed with age in some cases, data are split into 3 roughly equal subgroups: deciles 1–3 (red, N = 72), deciles 4–5 (blue, N = 81), and deciles 6–7 (green, N = 65). Abbreviations: RT, reaction time; SD, standard deviation. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)