Gray matter asymmetry in asymptomatic carotid stenosis

Abstract Even clinically “asymptomatic” carotid stenosis is associated with multidomain cognitive impairment, gray matter (GM) atrophy, and silent lesion. However, the links between them remain unclear. Using structural MRI data, we examined GM asymmetry index (AI) and white matter hyperintensity (WMH) in 24 patients with severe asymptomatic carotid stenosis (SACS), 24 comorbidity‐matched controls, and independent samples of 84 elderly controls and 22 young adults. As compared to controls, SACS patients showed worse verbal memories, higher WMH burden, and right‐lateralized GM in posterior middle temporal and mouth‐somatomotor regions. These clusters extended to pars triangularis, lateral temporal, and cerebellar regions, when compared with young adults. Further, a full‐path of WMH burden (X), GM volume (atrophy, M1), AI (asymmetry, M2), and neuropsychological variables (Y) through a serial mediation model was analyzed. This analysis identified that left‐dominated GM atrophy and right‐lateralized asymmetry in the posterior middle temporal cortex mediated the relationship between WMH burden and recall memory in SACS patients. These results suggest that the unbalanced hemispheric atrophy in the posterior middle temporal cortex is crucial to mediating relationship between WMH burden and verbal recall memories, which may underlie accelerated aging and cognitive deterioration in patients with SACS and other vascular cognitive impairment.

precursor pathology to cerebrovascular disease is silent white matter damage, which promotes GM loss and consequent cognitive impairment. Clues could be found in neurodegenerative diseases, such as mild cognitive impairment (MCI) and Alzheimer's disease (AD). Recent evidence suggests that WMH burden predicts verbal memory deficits and is mediated by lateral temporal atrophy in AD (Swardfager et al., 2018). GM asymmetry, affected by hemispheric atrophy, tends to increase with age and alters in neurodegenerative diseases including MCI and AD (Minkova et al., 2017;Toga & Thompson, 2003). In addition, WMH burden is associated with brain atrophy and predicts dementia progression (Habes et al., 2016;Wen, Sachdev, Chen, & Anstey, 2006).
Advances in neuroimaging, meanwhile, have greatly enhanced our understanding regarding brain aging and macroscale organization.
Multimodal imaging data sharing, for example, not only provides a basis for discovery, validation, and metanalytical synthesis, but also provides perspectives on cortical hierarchy centered on the default mode network (Margulies et al., 2016) and semantic representations of abstract cognitive operations (Huth, de Heer, Griffiths, Theunissen, & Gallant, 2016) in the human brain. These advances enable us to better understand multifactorial interplay underlying carotid stenosis from a neuroimaging framework.
In the present study, we sought to investigate the following key research questions (as illustrated in Figure 1) that related to cognitive decline in unilateral severe (>70%) asymptomatic carotid stenosis (SACS). We tested whether there are unbalanced hemispheric atrophies and whether these atrophies influence the relationship between WMH and the cognitive function. Further, we attempted to interpret these interrelationships under a macroscale cortical organization.
These questions were investigated using two additional in-house samples of young adults and elderly populations and a public life span dataset and metanalytical decoding. Specifically, we examined the voxel-wise GM asymmetry and explored the relationships among hemispheric GM asymmetry, WMH, and neuropsychological performance in patients with SACS.

| MATERIALS AND METHODS
This study was approved by the Medical Ethics Committee of Zhongnan Hospital of Wuhan University, and all participants gave written informed consent.

| Cognitive assessments
All participants (i.e., 24 SACS patients and 24 HC) completed a comprehensive cognitive battery within 7 days of the MRI scan, including (a) general cognitive assessment and screening using the MMSE (Tombaugh & McIntyre, 1992) and Montreal Cognitive Assessment Beijing Version (Nasreddine et al., 2005); (b) perceptual-motor speed, memory, and visual scanning using the Digit Symbol (Strauss, Sherman, & Spreen, 2006); (c) verbal learning, memory, encoding, and retrieval using the Rey Auditory Verbal Learning Tests (RAVLT; Schmidt, 1996). A full description of the assessment procedures can be found in our recent publications (Gao et al., 2019;Gao et al., 2021;Wang et al., 2017;Xiao et al., 2018).
Young adults were scanned using a dedicated 3.0 T Siemens PRISMA scanner (Siemens, Germany), with a 64-channel head coil and similar parameters. Detailed descriptions of the MRI scan protocols have been reported in previous studies (Gao et al., 2019;Gao et al., 2021;Wang et al., 2017;Xiao et al., 2018).

| GM asymmetry index (AI)
The AI was calculated following previous research (Kurth, Gaser, & Luders, 2015;Kurth, MacKenzie-Graham, Toga, & Luders, 2015). The segmented GM and WM images from the VBM analysis were normalized and used to compute GM AI. Specifically, the normalized symmetrical GM and WM images were left-right flipped. The flipped and nonflipped GM and WM images were warped and normalized into the MNI space to create an analysis-specific symmetric DARTEL template.
A right-hemispheric mask was then applied. The GM AI was calculated with the following formula: where i1 was the warped nonflipped GM images, i2 was the warped flipped GM images, and i3 was the right-hemispheric mask. Negative AIs indicate leftward GM, while positive AIs indicate rightward GM. The AI images were smoothed using an 8-mm FWMH smooth kernel.
Given the side of stenosis likely affects the direction of AI, we obtained individual signed Z-score maps by subtracting the mean and divided by the standard deviation of the pooled HC (n = 24). The Zmap of each patient was then thresholded (jZj ≥ 3.1 and cluster size ≥10 voxels) within the right hemispheric mask, which resulted in an AI probability map (superimposed Z map).

| White matter hyperintensity (WMH)
WMH size was calculated based on T1 and 3D T2-FLAIR images using the Lesion Segmentation Tool (LST, www.applied-statistics.de/lst; Schmidt et al., 2012) as described in its tutorial and our recent publications (Gao et al., 2019;Gao et al., 2021). Lesions were segmented by the lesion growth algorithm with a default initial threshold (κ = 0.3; Schmidt et al., 2012).

| Statistical analysis
Five SACS patients were excluded due to artifacts or low weight scores. Demographic and clinical data were contrasted between groups using two-sample independent t-tests for continuous variables and using Chi-square tests for the categorical variable (i.e., gender) in SPSS 16.0 (SPSS Inc., Chicago, IL), with a significance level of p < .05.
Imaging data were contrasted between groups using two-sample ttests with total intracranial volume (TIV), gender, and age as covariates of no interest. The results were corrected for spatial extent across the GM (VBM) or right side hemisphere mask (AI) using a voxel-wise threshold of p < .001 with a cluster-wise family wise error (FWE) rate correction at p < .05.
We also examined Pearson's linear correlations between mean AI within each significant cluster, behavioral measurements, and WMH burden in SACS patients. All correlation results were corrected for multiple comparisons using the Bonferroni approach at a significance of p < .05.  Figure 1). We applied a bootstrapping method with 10,000 samples and a 95% confidence interval (CI) for indirect effect inference. A 95% CI that does not contain zero indicates a significant indirect effect (Hayes, 2018).

| Metanalytical mapping and lifespan GM asymmetry trajectory
To verify the functional role and attached network system of the pMTG, we performed functional connectivity and metanalytical mapping with the bilateral pMTG seeds (MNI coordinates: ±58, À28, À2; spheres with 5 mm radius) from the NeuroSynth database (www. neurosynth.org; Yarkoni, Poldrack, Nichols, Van Essen, & Wager, 2011). Further, we performed a metanalysis with the term "recall" to obtain metanalytical activation maps.
Additionally, to delineate the developmental trajectory of AI over a broad age range of 18-80 years, we applied the same methodology in the SALD dataset and examined both linear and quadratic age effects.
Additionally, we found significantly greater rightward AI in lateral temporal, orofacial somatomotor, pars triangularis (Broca's area), and cerebellar regions in SACS patients vs. young adults (Figure 3b, middle panel). The latter two clusters also consistently presented in the comparison between the HC and the young adults ( Figure 3b, right panel).
In the SALD validation dataset, GM volumes in the inferior frontal/pars triangularis and orofacial somatomotor regions showed significantly age-related rightward shifts (with both linear and quadratic age effects), but the pMTG did not show a trend of age-related effect

| Associations between AI, behavioral, and lesion parameters in SACS patients
After controlling for the GM atrophy and AI, the direct effect of WMH on delayed recall was no longer significant (d' = 56.49, p = .058). Finally, the bootstrap procedure revealed a significant indirect effect that was significantly different from zero (a Â b Â c = À36.62, 95% CI [À64.62, À6.94; Figure 6d).

| WMH to immediate recall
Similar mediation effect was found in the immediate recall. The total effect of WMH on immediate recall was statistically significant

| DISCUSSION
Using a high-dimensional voxel-wise GM asymmetry algorithm (Kurth, Gaser, & Luders, 2015;Kurth, MacKenzie-Graham, et al., 2015), we identified rightward GM shifts in a language/speech pathway in SACS patients. Further, the mediation analysis indicated that atrophy in the left pMTG and asymmetry in the right pMTG serially mediated the relationship between WMH and verbal memory. Using metanalytical decoding and functional connectivity gradient map, we showed that the observed differences between SACS and controls in AI spanned across orofacial somatosensory/motor primary and temporal association cortices (two extreme ends of the principal gradient) and were in a potential "recall memory" network. These findings may facilitate a better understanding of the neuroanatomical basis of cognitive impairment in carotid stenosis.

| Between-group differences in AI
We replicated GM atrophies as shown in previous research in asymptomatic carotid stenosis (Avelar et al., 2016) and extended the analysis to the asymmetric atrophy (i.e., AI). This analysis identified pronounced rightward GM in SACS patients in a semantic/speechprocessing pathway that supports abstract cognitive functions F I G U R E 2 Group differences in VBM. (a) Primary and (b) validation results of group differences between SACS patients and controls in VBM, cluster-level p < .05, FWE corrected. FWE, family wise error; L, left; R, right; SACS, severe asymptomatic carotid stenosis; VBM, voxel-based morphometry. These maps were rendered by using the Surf Ice tool (https://www.nitrc.org/projects/surfice/) and MRIcroGL (https://www.nitrc. org/projects/mricrogl) including verbal learning and semantic memory, and primary functions including orofacial motor and speech (Skeide & Friederici, 2016).
These findings, corresponding with the carotid territory, suggest a left-dominated atrophy and consequently right-lateralized GM shifts.
It is possibly manifested as hemispheric differences in vulnerability, hemodynamics, functional reserve, or silent lesions. Early dementia (Toga & Thompson, 2003)

| A path potentially mediating the relationship between WMH and recall memory
While asymptomatic carotid stenosis is known to be associated with atrophy and WMH, there is a gap among atrophy, WMH, and the cognitive function in individuals with SACS. Both clinical experience and recent evidence (de Weerd et al., 2014;Grau-Olivares et al., 2010;Lal et al., 2017;Lattanzi et al., 2019;Lattanzi et al., 2020) suggest that these factors are likely to promote cognitive deterioration in SACS.
Our data demonstrated that atrophy and asymmetry in bilateral pMTG serially mediate the relationship between WMH and verbal memory, which provides insight into the link between neuroanatomical path and both carotid stenosis and verbal memory impairment. Similar to findings in AD patients (Swardfager et al., 2018), we found that WMH loads affect recall memory through AI alone or a combination with atrophy and asymmetry sequentially in SACS patients. These paths account for 19% (path abc) or 35% (path a 2 c) variances of total effects in delayed recall, and 18% (path abc) or 37% (path a 2 c) in immediate recall memory, respectively. The covariates of no interest, i.e., TIV, age, and education, differently interact with the mediators.
One example is the verbal learning task (Miotto et al., 2006;Miotto et al., 2014), in which subjects are presented with words auditorily and asked to recall the words as much as they can remember immediately or shortly. Completion of this task recruits dorsolateral prefrontal, orofacial sensorimotor, and inferior frontal regions, but suppresses activity in posterior temporal cortex (Miotto et al., 2006;Miotto et al., 2014). Many of these regions, however, were involved in the rightward AI in the SACS patients and in the "recall" topography, suggesting cortical reorganization at the macroscale.

| Situating the AI differences on a macroscale hierarchical organization and aging trajectory
A recent milestone in the neuroimaging community is the discovery of hierarchical gradients embedded in the cerebral cortex (Margulies et al., 2016). These gradients define a geometric space of humans brain function and verify the Dual Origin Theory (Goulas, Margulies, T A B L E 1 Regions showing significant group differences in gray matter asymmetry for both primary and validation analyses Bezgin, & Hilgetag, 2019), in which the cerebral cortex is phylogenetically evolved from two distinct systems: amygdala-centric and hippocampocentric systems. The hippocampocentric system is crucial for advanced cognitive functions and corresponds to the principal gradient. The lower end of the principal gradient anchors to the primary sensorimotor cortex and the top end to the default mode network (Raichle, 2015), and it engages in brain functions from simple perceptual and motor to progressively abstract function spectrums, including semantic retrieval and verbal memory.
Our data identified a link between the AI differences and the prin- . The threshold level in B was set to Z value ≥3.1, corresponding to p < .001. The red boxes outline patients with left carotid stenosis. Superimposed probability maps based on the thresholded Z maps with both increased and decreased AI summed across the SACS patients after thresholded (jZj ≥ 3.1) across SACS patients were displayed, in cortical (c) and slice (d) views. The color bar indicates number of patients with significant rightward (hot tone, up to 11) or leftward (cold tone, up to 4) AI in the same brain region. AI, asymmetry index; HC, healthy controls; SACS, severe asymptomatic carotid stenosis. These maps were visualized by using the BrainNet viewer (https:// www.nitrc.org/projects/bnv/) and MRIcroGL (https://www.nitrc.org/projects/mricrogl) The associations between neuropsychological measures, WMH, and GM asymmetries in SACS patients. (a) Immediate recall was significantly negatively correlated with AI in the pMTG. (b) Delayed recall was significantly negatively correlated with AI in the pMTG. (c) Adjusted WMH size (i.e., the ratio of WMH size and total intracranial volume) was significantly positively correlated with AI in OPER, PrG, pMTG, and the thalamus. AI, asymmetry index; OPER, operculum insulae; pMTG, posterior middle temporal gyrus; PrG, precentral gyrus; SACS, severe asymptomatic carotid stenosis; THALA, thalamus; WMH, white matter hyperintensity These results suggest that SACS, a potential chronic hypoperfusion condition, may associate with reorganization in the principal gradient (i.e., the first two gradients).
An exception is that we did not find a significant mediating role of hippocampus in recall memory. It is possible due to the fact that hippocampus is relatively small and cannot survive after multiple corrections. On the other hand, from a network perspective (Fox, 2018), the pMTG is part of the extending network that involves in semantic processing and verbal memory.
Consistent with a recent report (Kong et al., 2018), we showed significant age-related rightward shifts in inferior frontal and somatomotor regions in the SALD lifespan dataset. However, when using the bilateral pMTG as targets, we did not find significant linear nor quartic effects. These results suggest that SACS-related AI asymmetric changes commix with both cognitive impairment and accelerated aging.

| Clinical relevance
Considering the focal alterations in the semantic and speech circuit, our findings have some potential clinical applications. For example, intensive training (e.g., speech, aerobic exercises, and attentional training; Kurth, Gaser, & Luders, 2015;Kurth, MacKenzie-Graham, et al., 2015) alters GM asymmetries and has shown therapeutic effects in primary progressive aphasia (Bonakdarpour et al., 2019).
Recently, the orofacial area serves as a target of brain computer interface for neural decoding and speech synthesis (Anumanchipalli, Chartier, & Chang, 2019). These applications suggest that speech training may renormalize AI in patients with SACS.

| Limitations
The present study has a few potential limitations that are worth noting. First, we analyzed changes in neuroanatomical pathways, a determinant in cognitive impairment of asymptomatic carotid stenosis and also changes in hemodynamics (Lattanzi et al., 2018;Lattanzi et al., 2019;Lattanzi et al., 2020); however, the effects on the neuroanatomical basis are still poorly understood, and future research is needed to further elaborate on this contribution. Second, the average age of the SACS subjects was >60 years, a stage where normal aging and pathologies (such as worsening hemodynamics and WMH burden) intricately intertwined, and thus longitudinal dataset might be needed For each connection, the coefficient (a, b, c, a 2 , and b 2 ), its SE, and p-value (p) are shown above the line. The mediation effect of the hypothesized serial pathway (a Â b Â c) and its 95% CI are shown in red, the mediation effect of left pMTG atrophy (a 2 Â c) in green, and the mediation effect of atrophy alone (a Â b 2 ) in blue. Direct effects are shown in orange; d is the direct effect adjusted only for covariates and d' is the direct effect adjusted for covariates and indirect effects. Covariate effects on mediators or outcomes with p < .05 are shown in gray. DR, delayed recall; IR, immediate recall; pMTG, posterior middle temporal gyrus; TIV, total intracranial volume. The brain maps were rendered by using the Surf Ice tool (https://www.nitrc.org/projects/ surfice/) to address how pathologies contribute to the morphological alterations. Future research should focus on neuroanatomical changes after carotid revascularization, and this is indeed one of the main goals of randomized clinical trials in this community (Howard et al., 2017;Marshall, Asllani, Pavol, Cheung, & Lazar, 2017). Increasing evidence suggests that revascularization can simultaneously improve cerebral hemodynamics, reverse cognitive deterioration (Lattanzi et al., 2019;Lattanzi et al., 2020), and even renormalize functional connectivity in the advanced networks (Porcu et al., 2019;Porcu et al., 2020). Third, we did not take into account the role of clinically silent lacunar infarcts in cognitive impairment in the sample of patients analyzed.
This is important because more than half of the patients with a firstever lacunar stroke and without cognitive impairment show minor neuropsychological alterations (Blanco-Rojas et al., 2013;Grau-Olivares et al., 2007). These minor disturbances are mainly related to the presence of clinically silent lacunar infarcts at this early stage of the disease (Grau-Olivares et al., 2010). Fourth, this study included independent samples across a wide age span. The structural data processing used a standard group-wise symmetric template, which may bring age-(and disease-) related minor mis-registrations. Fifth, we included MRI data from two different scanners (i.e., Siemens Trio and Prisma). There might be differences caused by different scanners in image intensity/contrast that may lead to differences in tissue segmentations. Nevertheless, the calculation of AI is unlikely to be affected, as AI is a unitless metric. Finally, we did not include factors such as smoking, drinking, and comorbidities into the path-analysis model, and thus it is unknown how these factors may affect the model. Future studies using more sophisticated models may establish these relationships.

| CONCLUSIONS
In conclusion, SACS is associated with left-dominated atrophy and consequently right-lateralized GM shifts in the lateral temporal regions. Furthermore, unbalanced hemispheric atrophy in the lateral temporal region is crucial in mediating relationship between WMH burden and recall memory, which may underlie accelerated aging and cognitive deterioration in SACS and other VCI.

ACKNOWLEDGMENTS
This study was supported by the National Natural Science Foundation of China (grants numbers: 82001799 and 81771819). All authors disclosed no other relevant relationships.

AUTHOR CONTRIBUTIONS
Lei Gao: Collected data, designed the methods, performed analysis, and wrote the manuscript. Yaqiong Xiao: Revised the manuscript, performed the analysis, interpreted, and discussed the results. Haibo Xu: Conceived the study, interpreted and discussed the results, and revised the manuscript. All the authors contributed to the revision of the manuscript.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author.