《认知心理学》课程教学资源（书籍文献）聋人听觉皮层在感知觉功能重组中的作用 Reorganization of large-scale brain networks in deaf signing adults - The role of auditory cortex in functional reorganization following deafness.pdf

Neuropsychologia 166 (2022) 108139 Available online 4 January 2022 0028-3932/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Reorganization of large-scale brain networks in deaf signing adults: The role of auditory cortex in functional reorganization following deafness Josefine Andin a,* , Emil Holmer a,b a Linnaeus Centre HEAD, Department of Behavioural Sciences and Learning, Linkoping ¨ University, SE, 581 83, Link¨ oping, Sweden b Center for Medical Image Science and Visualization, Link¨ oping University, Sweden ARTICLE INFO Keywords: Deaf signers Deafness Large-scale brain networks ICA Functional connectivity Superior temporal cortex ABSTRACT If the brain is deprived of input from one or more senses during development, functional and structural reorganization of the deprived regions takes place. However, little is known about how sensory deprivation affects large-scale brain networks. In the present study, we use data-driven independent component analysis (ICA) to characterize large-scale brain networks in 15 deaf early signers and 24 hearing non-signers based on resting-state functional MRI data. We found differences between the groups in independent components representing the left lateralized control network, the default network, the ventral somatomotor network, and the attention network. In addition, we showed stronger functional connectivity for deaf compared to hearing individuals from the middle and superior temporal cortices to the cingulate cortex, insular cortex, cuneus and precuneus, supramarginal gyrus, supplementary motor area, and cerebellum crus 1, and stronger connectivity for hearing nonsigners to hippocampus, middle and superior frontal gyri, pre- and postcentral gyri, and cerebellum crus 8. These results show that deafness induces large-scale network reorganization, with the middle/superior temporal cortex as a central node of plasticity. Cross-modal reorganization may be associated with behavioral adaptations to the environment, including superior ability in some visual functions such as visual working memory and visual attention, in deaf signers. 1. Introduction Cortical areas that are deprived of sensory input during development reorganize to respond to the preserved senses (Bavelier and Neville, 2002; Merabet and Pascual-Leone, 2010). Thus, individuals who are deprived of sensory input in one modality, due to e.g., deafness or blindness, hold an important clue to understanding brain reorganization. Early deafness has repeatedly been associated with reorganization of the auditory cortex (Cardin et al., 2018; Emmorey et al., 2011; Karns et al., 2012; Malaia et al., 2014), but there is also some evidence of reorganization beyond this region (Bonna et al., 2020; Li et al., 2016). In the present study, we apply a data-driven approach, i.e., independent component analysis (ICA) on resting-state fMRI data, to further our understanding of how network organization differs between deaf and hearing individuals. This will in turn have implications for how findings from functional connectivity studies that investigate network nodes can be interpreted. The focus of this study is on an adult deaf population for whom sign language learning has been optimized by early sign language interventions and bilingual schooling with sign language as the main mode of communication (Bagga-Gupta, 2004; Meristo et al., 2007). The combination of fluent language skills and lack of auditory input makes deaf signers a highly valuable study population for investigating cortical reorganization due to sensory deprivation. However, with the introduction of cochlear implants in almost every deaf infant in the Western world, including Sweden where this study is situated, the current cohort is likely to represent one of the last cohorts of its kind. The lack of sensory input induces changes in brain regions associated with the processing of the remaining senses, as well as in the region typically used for processing the sense that is lost. These changes are associated with behavioral adaptations, and sometimes even superior skills compared to individuals with no sensory impairment (for a review see Merabet and Pascual-Leone, 2010). For example, the lack of visual input in blind individuals is associated with recruitment of the visual cortex, i.e., the sensory-deprived area, for tactile Braille reading (Reich et al., 2011), sound localization (Gougoux et al., 2005), and verbal processing (Amedi et al., 2004). Superior processing of auditory stimuli has further been associated with altered processing in areas of the remaining senses, i.e., in the superior temporal cortex (Stevens and * Corresponding author. Linkopings ¨ universitet, IBL, I-huset, SE, 581 83, Linkoping, ¨ Sweden. E-mail addresses: josefine.andin@liu.se (J. Andin), emil.holmer@liu.se (E. Holmer). Contents lists available at ScienceDirect Neuropsychologia journal homepage: www.elsevier.com/locate/neuropsychologia https://doi.org/10.1016/j.neuropsychologia.2021.108139 Received 30 June 2021; Received in revised form 17 December 2021; Accepted 31 December 2021

Neuropsychologia 166 (2022) 108139 2 Weaver, 2009). Similarly, deaf individuals recruit the sensory-deprived area, auditory cortex, for visual (Andin et al., 2021; Bottari et al., 2014; Cardin et al., 2013; Emmorey et al., 2011; Karns et al., 2012) and vibrotactile perception (Auer et al., 2007; Karns et al., 2012), as well as cognitive tasks (Andin et al., 2021; Cardin et al., 2013, 2018; Ding et al., 2015; Twomey et al., 2017). These neural changes may also be related to superior behavioral performances in for example visual attention (for an overview, see Bavelier et al., 2006; MacSweeney and Cardin, 2015). However, few studies have investigated the recruitment of areas of the remaining senses for deaf individuals, e.g., visual cortex. Brain imaging studies have typically investigated task-based differences in distinct, pre-defined regions, primarily in the superior temporal cortex. However, the brain is organized in large-scale brain networks where several brain regions communicate and work in sync with each other, both during specific tasks and during rest (e.g., Petersen and Sporns, 2015; Thomas Yeo et al., 2011). To understand the role of a specific region, it is necessary to investigate inter-regional associations at a network level (Uddin et al., 2019). There is no consensus on network taxonomy, but Uddin et al. (2019) recently proposed a six-network solution as a universal taxonomy of large scale brain networks. The six networks were given anatomical names and include the occipital network, the pericentral network, the dorsal frontoparietal network, the lateral frontoparietal network, the midcingulo-insular network, and the medial frontoparietal network. Each network was further connected to a cognitive domain and to core regions, outlined in Table 1 together with main behavioral functions connected to each network. In the present paper, we primarily refer to the networks based on their cognitive domain. To understand how large-scale networks are organized, functional connectivity analysis is used. This is typically done by investigation of functional connectivity between regions of interest (ROI-to-ROI analysis) or from a seed to all voxels in the brain (seed-to-voxel analysis). Several studies have explored functional connectivity between superior temporal regions and other brain regions in deaf individuals. For example, we recently showed enhanced functional connectivity from the auditory cortex to several brain regions, primarily in the visual cortex, for deaf signers compared to hearing non-signers during a visual working memory task (Andin et al., 2021). Bola et al. (2017) also found stronger functional connectivity between auditory and visual cortices for deaf individuals when administering a rhythm discrimination task in the visual modality. During rest, studies have shown stronger connectivity for deaf compared to hearing individuals between superior temporal cortex and posterior cingulate, precuneus and the intraparietal lobule (Malaia et al., 2014), anterior cingulate cortex (Ding et al., 2016), insula (Ding et al., 2016; Striem-Amit et al., 2016), calcarine sulcus (Shiell et al., 2014), visual (Benetti et al., 2021) and frontal regions (Cardin et al., 2018; Ding et al., 2016). However, weaker functional connectivity between temporal regions and the visual word form area (Wang et al., 2015) and somatomotor areas (Bonna et al., 2020) has also been demonstrated. Several auditory and visual brain regions have further been shown to be structurally connected in both deaf and hearing individuals (Li et al., 2015). However, Shiell et al. (2014) suggested that the connection between auditory and visual regions is enhanced during visual processing due to the lack of auditory processing in deaf populations. Although most studies have focused on the superior temporal cortices, some studies have found reorganization beyond these regions. Connectivity from visual cortices has been found to be stronger for deaf compared to hearing individuals to frontoparietal areas (Bonna et al., 2020; Dell Ducas et al., 2021), the default network, and the salience network (Dell Ducas et al., 2021), and weaker to the language regions, both functionally (Li et al., 2016) and structurally (Dell Ducas et al., 2021; Li et al., 2015). Li et al. (2016) investigated functional connectivity from the limbic system and found stronger connectivity to both visual and language processing regions for deaf, compared to hearing, individuals. Reorganization has further been confirmed by morphological changes in deaf individuals for the occipital cortex (Allen et al., 2013; P´enicaud et al., 2012) and in temporal regions (Hribar et al., 2014; Kumar and Mishra, 2018; Shibata, 2007). ICA has been used to characterize large-scale brain networks in, and between, different populations. While seed- and ROI-based functional connectivity analyses are appropriate when the purpose is to describe group differences in connectivity pattern for the pre-selected seeds, these analyses do not reveal whether group differences exist in the underlying network structures. To investigate network separation, ICA can be used instead. Wang et al. (2014) applied ICA to compare congenital blind and sighted individuals and found differences between groups in the visual network, which is mainly involved in the processing of visual stimuli, and in the salience network, which is engaged for attention switching to salient stimuli. Dell Ducas et al. (2021) used ICA to identify difference between deaf and hearing individuals in the spatial extent of regions within the default network. Further, for deaf compared to hearing cats, Stolzberg et al. (2018) showed patterns of altered functional connectivity in networks including auditory, visual, cingulate, and somatosensory regions. However, to the best of our knowledge, no previous study used ICA to fully characterize large-scale networks in deaf signing individuals, which can further our understanding of plasticity due to deafness and sign language use. In the present study, we use data-driven ICA aiming to characterize group differences in resting-state functional connectivity between deaf signers and hearing non-signers in the organization of large-scale brain networks. 2. Materials and method 2.1. Participants Fifteen deaf early signers (eight female) and twenty-four hearing non-signers (twelve female) were included in the study. There were no significant group differences for gender distribution, non-verbal cognitive ability (tested using the Visual puzzles subtest from Weschler Adult Intelligence Scale), or level of education. However, there was a group difference in age with the deaf signers (M = 35.0, SD = 7.8) being significantly older than hearing non-signers (M = 26.5, SD = 7.5), t(37) Table 1 Overview over large-scale brain networks, based on Uddin et al. (2019). Anatomical name Cognitive domain Core regions Main behavioral functions Occipital network Visual Occipital lobe Visual processing Pericentral network Somatomotor Motor cortex, somatosensory cortex Motor processing, somatosensory processing Dorsal frontoparietal network Attention Superior parietal lobule, intraparietal sulcus, middle temporal complex, frontal eye field Visuospatial attention; top-down processing of stimuli and responses Lateral frontoparietal network Control Lateral prefrontal cortex, middle frontal gyrus, anterior inferior parietal lobule, intraparietal sulcus Executive functions; goal-oriented cognition, working memory, inhibition, switching Midcingulo-insular network Salience Anterior insula, anterior midcinulate cortex Detection of salient information Medial frontoparietal network Default Medial prefrontal cortex, posterior cingulate cortex, posterior inferior parietal lobule Goal directed cognitvion, monitoring the environment, processing of associative representations, elaboration of events etc J. Andin and E. Holmer

Neuropsychologia 166 (2022) 108139 3 = 3.4, p = 0.002. Inclusion criteria were normal or corrected-to-normal vision, right-handedness, and normal or above normal non-verbal cognitive ability. Exclusion criteria included claustrophobia, pregnancy, and having non-MR compatible implants. Nine of the participants had their deafness discovered at birth, while the remaining six were between 6 month and 3 years when their deafness was discovered. All deaf signers were considered early signers, using Swedish Sign Language (Svenskt Teckenspråk; STS) as their primary language, performing on par with an independent sample of deaf native signers on the STS sentence repetition test (Schonstr ¨ om ¨ and Hauser, 2021). Five participants were signed with from birth and nine reported starting acquisition of STS before the age of three. For one participant, age of acquisition was missing. None of the participants relied on hearing aids for verbal communication, although two participants used hearing aids for sound awareness. The hearing participants were native Swedish speakers without any knowledge of STS. The study was approved by the regional ethical review board in Linkoping ¨ (Dnr, 2016/344–31) and was conducted in accordance with the Declaration of Helsinki. Participants gave their written informed consent and were compensated for their participation. 2.2. Image acquisition Structural and functional MRI data were collected with a Siemens Magnetom Prisma 3T scanner (Siemens Healthcare, GmbH) at the Center for Medical Image Science and Visualization (Linko¨ping University, Sweden) using a 64-channel head coil. The scanning started with acquisition of structural images using a T1-weighted three-dimensional magnetizationprepared rapid gradient echo (MPRAGE) sequence with the following parameters: FOV = 288 × 288, acquisition matrix = 208 × 288 × 288, voxel size = 0.90 × 0.86 × 0.86 mm, TR = 2300 ms, TE = 2.36 ms, TI = 900 ms, FA = 8◦. Resting-state data was acquired at the end of the scanning after the participants had performed four task-EPI runs (Andin et al., 2021), using a BOLD multi-plex EPI sequence during a 10-min scan with the following parameters: FOV = 192 × 192 mm, voxel size = 3 × 3 × 3 mm, TR = 1340 ms, TE = 30 ms, FA = 69◦, number of slices = 48, 440 vol, interleaved/simultaneous acquisition. 2.3. Data processing Preprocessing was performed using the default pipeline in CONN functional connectivity toolbox (Version 20.b; www.nitric.org/proj ects/conn, RRID: SCR_009550) running under Matlab R2018a (The MathWorks Inc., Natick, MA). The preprocessing steps included functional realignment, unwarping and co-registration to the first scan, slicetiming correction to adjust for temporal misalignment between slices, outlier detection by computation of framewise displacement (outliers defined as displacement >0.9 mm or BOLD signal change >5 SD.), normalization into standard MNI space, structural segmentation into grey matter, white matter and CSF tissue classes, and smoothing using a Gaussian kernel of 8 mm full width half maximum to increase signal-tonoise ratio. The realignment parameters and the noise components from the outlier detection were used as first-level covariates. Linear regression using the anatomical component-based noise correction (aCompCor) algorithm was implemented to remove effect from subject specific physiological noise such as white matter and cerebrospinal areas, motion parameters, outlier scans (scrubbing) and session-related slow trends. Quality assurance checks showed that there were no group differences in number of scrubbed slices, max motion, or global signal change. However, the deaf group had significantly higher mean motion, i.e., the absolute displacement of each brain volume compared to the previous estimated from the x, y and z translation parameters (DS: M = 0.15, SD = 0.05; HN: M = 0.12, SD = 0.04; t(37) = 2.4, p = 0.023), why this parameter was included as a covariate in all group analyses. Secondlevel covariates thus included group (deaf early signers/hearing nonsigners), age (mean-centered), and the mean motion parameter from the realignment step. Denoising included linear regression of potential confounding effects and temporal processing using bandpass filtering (0.008, 0.09 Hz). To identify networks that are functionally connected, independent component analysis (ICA) was performed by estimating spatially independent patterns in the fMRI data. Independent components across both groups were determined using a G1 FastICA algorithm for component definition at the group-level and GICA 3 subject-level back projection. Dimensionality reduction was set to 64. ICA was performed with the number of components set to eight, sixteen, twenty-four, and thirty-two. Each analysis was visually inspected and matched to the six networks described by Uddin et al. (2019), and automatically to the network templates included in CONN. The twenty-four-component setting rendered the best overall solution. Nine of the twenty-four components matched the large scale brain networks by Uddin et al. (2019) and were included in further analyses (Fig. 1). Generally, a lower number of components (around 20) are used when the aim is to identify functional large-scale networks, as in the present study, while larger number of components (above 100) are used for brain parcellation (Ray et al., 2013). 2.4. Data analysis To identify group-differences in the nine independent components, the spatial maps of each component were analyzed using betweensubjects contrasts with age and mean motion as covariate (1, − 1, 0, 0). Results were voxel thresholded at p < 0.001 uncorrected, and cluster thresholded at p < 0.05 using False Discovery Rate to control for type 1 errors. For independent components with significant differences between groups, the significant clusters were exported and used as seeds in seed-to-voxel analyses. Group differences in functional connectivity were investigated using the same contrasts and the same thresholding as for the ICA. Further, functional connectivity measures and effect size (beta values) of each significant cluster were extracted. 3. Results 3.1. Independent component analysis The nine independent components that best matched the six networks proposed by Uddin et al. (2019) are presented in Fig. 1. For three networks, two separate components were chosen since they represented typical sub-networks. Thus, the visual network (occipital) was represented by a medial and a lateral component (Fig. 1a and b), the somatomotor network (pericentral) was represented by a ventral and a dorsal component (Fig. 1c and d), and the control network (lateral frontoparietal) by a left and a right component (Fig. 1e and f). The attention network (dorsal frontoparietal), salience network (the midcingulo-insular), and the default network (medial frontoparietal) were best captured by one single component each (Fig. 1g–i). Group differences, with stronger connectivity for deaf signers compared to hearing non-signers, were found in the default network component. Stronger connectivity for hearing non-signers compared to deaf signers were found in the ventral somatomotor network, the left control network, and the attention network (Fig. 2). The four independent components in which group differences were identified match different large-scale networks. However, all significant clusters in the between-group analysis except one were found in the superior and middle temporal regions. For the default network component, peak voxels were found in the left temporal pole, right superior and left middle temporal gyri. For the component representing the left control network, peak voxels were found in bilateral superior temporal gyrus and right pallidum. The ventral somatomotor network component showed peak voxels in left superior temporal gyrus, while the component representing the attention network had peak voxels in right superior temporal gyrus (Table 2). J. Andin and E. Holmer

Neuropsychologia 166 (2022) 108139 4 3.2. Seed-to-voxel analysis We localized group differences in temporal regions and the right palladium (Table 2). To further investigate how these regions might differ in connectivity across groups, we used the clusters (n = 8) as masks in subsequent exploratory seed-to-voxel analysis. Group, age, and the mean motion parameter were used as second-level covariates. The seed regions overlapped with each other and the resulting connectivity maps showed considerable similarities (Fig. 3A-E). The seed in right pallidum resulted in no significant connections and is therefore not included in the figure. Deaf signers showed stronger connectivity between the temporal seeds and targets in the cingulate cortex, insular cortex, superior temporal cortex, cuneus and precuneus, supramarginal gyrus, supplementary motor area, and cerebellum crus 1. Hearing nonsigners showed stronger connectivity to clusters including hippocampus, middle/superior frontal gyrus, pre- and postcentral gyrus, and cerebellum crus 8. Details of connectivity measures are provided in Table S1. 4. Discussion We sought to investigate large-scale brain networks in deaf signers, and how these differ from hearing non-signers. Our findings confirm that temporal regions are subject to cross-modal reorganization in deaf individuals. For the nine components that overlapped with the largescale networks defined by Uddin et al. (2019), group differences in functional connectivity were found in four. All differences except one were located in middle and superior temporal regions. These temporal regions further showed connectivity differences between groups, dispersed across brain regions in frontal, parietal, and temporal regions as well as the cerebellum. The results suggest that deafness induces large-scale brain network reorganization which may be associated with behavioral adaptation due to the lack of access to auditory input in deaf individuals. 4.1. Large-scale brain networks We show here that several large-scale brain networks are similar Fig. 1. Large-scale brain networks identified in the independent component analysis. The visual network (occipital network) divided into A) medial part and B) lateral part. The somatomotor network (pericentral network) divided into C) ventral part and D) dorsal part. The control network (lateral frontoparietal network) divided into E) right part and F) left part. G) Attention network (dorsal frontoparietal network), H) salience network (midcingulo-insular network), and I) default network (medial frontoparietal network). Yellow represents regions positively correlated within the network, and purple represents regions negatively correlated. Brighter color represents stronger correlation. J. Andin and E. Holmer

Neuropsychologia 166 (2022) 108139 5 across deaf and hearing individuals. However, network organization of the left control, the default, the ventral somatomotor, and the attention networks seems to differ between groups. Specifically, middle and superior temporal regions are differently involved in these networks. Thus, using pre-defined network nodes based on hearing populations to compare network connectivity differences across deaf and hearing individuals might bias results, and confound interpretations. Previous studies have found differences in engagement of the control network and the somatomotor network between deaf and hearing individuals and suggested this to be related to sign language and visual processing (Bonna et al., 2020; Cardin et al., 2018; Okada et al., 2016). We also found differences in connectivity within the attention network. Since this network is important for orientation towards external stimuli, including visual orientation and target detection, it is reasonable to assume that it might be reorganized in sign language users. Further, others have reported stronger functional connectivity for deaf compared to hearing individuals related to the default network (Bonna et al., 2020; Dell Ducas et al., 2021; Malaia et al., 2014) and between the default and the visual network (Bonna et al., 2020). While Bonna et al. (2020) suggested that the differences were associated with increased network integration following sensory deficits, Malaia et al. (2014) suggested a link to visual language processing since nodes within the default mode network are responsive to the processing of visual language. Another explanation for observed differences between deaf and hearing individuals might be the influence of scanner noise, which in hearing individuals has been shown to contribute to suppression of the default network (e.g., Gaab et al., 2008). It should be noted that during scanning, hearing, but not deaf, individuals have auditory input, and this might influence how large-scale brain networks are captured in the signal. Except for effects on the default mode network (Gaab et al., 2008), scanner noise has also been shown to decrease activation in somatomotor networks (Andoh et al., 2017). Thus, the group differences we found in the somatomotor network might be even larger than what we would have found if scanner noise had been absent for the hearing Fig. 2. Group difference in A) left control network (left lateral frontoparietal network), B) default network (medial frontoparietal network), C) ventral somatomotor network (ventral pericentral network), and D) attention network (dorsal frontoparietal network). Red represents clusters more strongly associated to the component for deaf early signers compared to hearing non-signers and Blue represent clusters more strongly associated to the component for hearing non-signers compared to deaf early signers. Table 2 Peak coordinates for cluster with group differences. Network Region of the peaka Voxels Peak MNI coordinates DS > HN HN > DS x y z t pFDR t pFDR Left control network Left Superior Temporal Gyrus 189 − 56 − 14 0 4.55 <.001 Right Pallidum 164 22 − 2 6 5.52 <.001 Right Superior Temporal Gyrus 124 62 − 20 8 4.87 <.001 Default network Left Middle Temporal Gyrus 189 − 56 − 16 0 5.63 <.001 Left Superior Temporal Pole 146 − 46 12 − 14 5.38 <.001 Right Superior Temporal Gyrus 144 52 − 18 − 6 6.05 <.001 Ventral somatomotor network Left Superior Temporal Gyrus 150 − 52 − 20 8 4.63 <.001 Attention network Right Superior Temporal Gyrus 544 56 − 20 − 2 5.61 <.001 a Labels using AAL. J. Andin and E. Holmer