Thirty-two patients with chronic left subcortical stroke (ischemic or hemorrhagic) and history of aphasia were recruited with the following inclusion criteria: (i) Native Chinese speakers; (ii) over 6 months after stroke; (iii) unilateral subcortical lesion; (iv) ability to perform tests; and (v) absence of prior diagnosis of neurological disease. Based on medical records, all patients had aphasia at the time of stroke and received varied types of speech–language rehabilitation. Thirty age-matched healthy subjects with no history of neurological disease were recruited in the study.

The study was approved by the Institutional Review Board of the First Affiliated Hospital of Sun Yat-sen University, and written informed consent was obtained from all participants before enrollment in the study.

Language performance of all participants was assessed using the Western Aphasia Battery-Revised (WAB-R) (Shewan and Kertesz, 1980), which was translated into Chinese version. The language assessment was administrated and scored by two neurologists (S.H.F and H.J.T). The WAB-R includes separate subtests that provide composite scores for spontaneous speech, repetition, naming/word finding, and auditory-verbal comprehension. The total of these composite scores is the aphasia quotient, a measure of overall aphasia severity ranging from 0 to 100.

Image data were acquired on a Siemens 3T Trio scanner with a 12-channel coil, and three-dimensional T1-weighted structural images were acquired using the magnetization-prepared rapid gradient-echo imaging sequence: TR (relaxation time) = 1,900 ms; TE (echo time) = 2.56 ms; flip angle = 9°; field of view (FOV) = 250 mm × 250 mm; 192 contiguous 1-mm sagittal slices; voxel size = 1 mm × 1 mm × 1 mm). Diffusion-weighted images were acquired using the single-shot echo-planar imaging sequence: TR = 8,700 ms; TE = 90 ms; flip angle = 90°; FOV = 240 mm × 240 mm; 64 3-mm sagittal slices; voxel size = 2 mm × 2 mm × 3 mm; and 64 diffusion volumes weighted with b_max = 1,100 s/mm² and one volume with no diffusion gradient (b₀ = 0 s/mm²).

To determine cortical thickness association with language measures, cortical thickness maps were entered into separate multiple regression analyses using spontaneous speech, naming/word finding, repetition, and auditory-verbal comprehension as explanatory variables controlling for key demographic factors and lesion size as confounding variables. Multiple comparisons were corrected at a voxel-wise threshold of p < 0.005 and a cluster-wise level of p < 0.05, as empirically determined by Monte Carlo Simulation.

To clarify how the nuisance covariates included in the cortical thickness analysis contributed to language outcomes with and without cortical thickness measure in the model, the variables were further introduced into separate hierarchical linear regression with language outcomes as the dependent variables. Separate univariate analyses were performed to address group differences in mean cortical thickness in each identified cluster between patients and healthy subjects by controlling for the above demographic variables. Paired t-tests were conducted to compare white matter measures between the left and right hemispheres. Further, partial correlation analyses were performed to explore the relationships between cortical thickness in each cluster and mean FA values of the tracts connecting stroke lesions and cortical clusters by excluding the indicated demographic variables and lesion size. These analyses were performed using SPSS (version 22).

The present study included 32 patients with subcortical stroke (24 infarcts and 8 hemorrhages) and 30 healthy control subjects. The demographic details and language scores of participants are shown in Table 1 and Supplementary Table 1. There were no significant differences in age, gender, educational level, or handedness between patients and healthy subjects (all p > 0.05). The mean time from stroke onset was 14.94 months (interquartile range = 7.8–35.7 months), and the mean lesion volume was 6.74 cm³ (interquartile range = 3.5–13.3 cm³). Among all patients, 26 suffered from aphasia and 6 patients were fully recovered. In aphasic patients, 25 (78%) were anomic aphasia. Significant differences were observed in the outcomes of naming/word finding [F(1, 56) = 18.31, p < 0.001], spontaneous speech [F(1, 56) = 8.87, p = 0.004] and auditory-verbal comprehension [F(1, 56) = 10.34, p = 0.002], but not repetition [F(1, 56) = 2.17, p = 0.146] in patients compared with healthy subjects when controlling for age, gender, educational level, and handedness.

We explored the potential associations of subcortical lesion location with different language scores by SVR-LSM. The results showed that no subcortical region in the left hemisphere in which damage relates to language scores was identified across patients (Data are not shown).

Next, separate regression analyses were performed to identify cortical thickness changes related to language scores by factoring out confounding variables. The correlation analysis showed education and lesion size significantly related to language outcomes (see Supplementary Table 2). Considering the possible impacts of age and time from stroke onset on cerebral volume (Nitkunan et al., 2011; Walhovd et al., 2011; Brodtmann et al., 2012; Takao et al., 2012), we included age, education level, time from stroke onset and lesion size as the key nuisances in the separate regression model. The results showed that cortical thickness in the left inferior parietal cortex and the posterior superior temporal gyrus was positively related to the auditory-verbal comprehension score (Figure 2Aa and Table 2). In contrast, cortical thickness in the left medial/lateral orbitofrontal gyrus was positively related to spontaneous speech outcome (Figure 2Ba and Table 2). As for naming ability, cortical thickness in the left inferior precentral gyrus, pars orbitalis, and left posterior cingulum was positively related to the naming/word finding score (Figures 2Ca,Cd and Table 2). In addition, cortical thickness in the right lateral orbitofrontal cortex, pars orbitalis, and pars triangularis was positively related to naming/word finding performance (Figure 2Da and Table 2). No negative relationships were found between cortical thickness and language performance.

To clarify the relationships between aphasia outcomes, cortical thickness measures, and confounding factors included in the above cortical thickness analyses, we next performed separate hierarchical regressions using the language scores as dependent variables. Age, education level, time from stroke onset and lesion size were entered first, followed by cortical thickness measures. When cortical thickness measure was excluded, the hierarchical regression analyses showed that only lesion size significantly predicted the abilities of auditory-verbal comprehension [t(27) = −4.11, p < 0.001], spontaneous speech [t(27) = −2.79, p = 0.010], and naming/word finding [t(26) = −3.56, p = 0.001]. When cortical thickness measure was added to the models, cortical thickness in the respective left hemisphere clusters was the only significant predictor for the corresponding language scores (auditory-verbal comprehension cluster: R² change = 0.214, p < 0.001; spontaneous speech cluster: R² change = 0.258, p < 0.001; naming/word finding cluster 1: R² change = 0.171, p = 0.001 and cluster 2: R² change = 0.065, p = 0.024). For naming, when cortical thickness measures in the clusters were all entered to the model, only the right hemisphere cluster was a significant predictor [t(24) = 2.20, p = 0.001]. Residual plots of the relationships between cortical thickness in the identified clusters and language performance scores are shown in Figures 2Ab–Cb,Ce,Db.

Mean cortical thickness in the identified clusters was compared between aphasic patients and healthy subjects by factoring out age, gender, educational level, and handedness. Mean cortical thickness was significantly lower in patients than in healthy subjects in the left hemisphere clusters related to the outcomes of auditory-verbal comprehension [F(1, 56) = 5.13, p = 0.027; Figure 2Ac] and spontaneous speech [F(1, 56) = 5.47, p = 0.023; Figure 2Bc]. For naming/word finding, mean cortical thickness was significantly lower in patients than in healthy subjects in the left hemisphere clusters [cluster 1: F(1, 56) = 18.49, p < 0.001; cluster 2: F(1, 56) = 7.31, p = 0.009; Figures 2Cc,f]. In contrast, no significant difference was observed between groups in the mean cortical thickness in the right hemisphere cluster [F(1, 56) = 0.06, p = 0.804; Figure 2Dc].

White matter tracts connecting stroke lesion to the identified cortical regions and the mirrored homologous tracts were successfully reconstructed for all patients. The representative fiber tracts in a common space across patients are showed in Figures 3Aa–Da. To determine the potential correlates between cortical thickness in the identified regions with integrity of the reconstructed tracts, separate hierarchical regression analyses were performed with cortical thickness in each cluster as a dependent variable. The demographic variables, lesion size and FA values of connecting tracts were added to the models as independent factors. The results showed that mean FA values of the left hemisphere tracts were positively related to cortical thickness in the left hemisphere clusters related to the scores of auditory-verbal comprehension [t(24) = 2.59, p = 0.016], spontaneous speech [t(24) = 2.58, p = 0.017], and naming/word finding [cluster 1: t(24) = 2.80, p = 0.010; cluster 2: t(24) = 2.74, p = 0.011]. Residual plots of the relationships between cortical thickness in the identified clusters and language performance scores are shown in Figures 3Ab–Db. No significant relationships were found between cortical thickness in the right hemisphere cluster for naming/word finding and lesion size or the interhemispheric tracts connecting to lesions (both p > 0.20). In addition, mean FA value of each tract was significantly lower in the left hemisphere than in the right hemisphere homolog [t(31) = −5.49 ∼ –7.17, all p < 0.001] (Figures 3Ac–Dc).

To explore the impacts of perfusion status on cortical thickness, we further compared the differences in cortical thickness between patients without hypoperfusion (NHP) and with HP according to perfusion computer tomography or perfusion weighted images at stroke onset. Female patients were more [χ²(1) = 4.50, p = 0.03] and the lesion size was larger in the HP group than in the NHP group [t(30) = −2.12, p = 0.04]. No significant difference was detected for language performance (all p > 0.80) (see Supplementary Table 3). When controlling for demographic variables and lesion size, a reduction of cortical thickness was identified in the right cuneus regions (cluster size = 1177.58 mm²; coordinates: x = 5.6, y = −67.7, z = 14.0; p = 0.0002) in the HP group relative to the NHP group. There was no significant difference in the left hemisphere between the two subgroups (Figure 4). A post hoc analysis was performed to compare cortical thickness in each cluster identified as above to relate to language performance by controlling for key variables. The results showed that the mean cortical thickness of the individual clusters was not different between the two subgroups (all p > 0.11).

We found relationships between declined cortical thickness and subcortical aphasia outcomes by covarying key confounding variables. Specifically, auditory-verbal comprehension was predicted by cortical thickness in the left temporoparietal regions including the posterior temporal gyrus and inferior parietal cortex. It has been clarified that verbal comprehension relies on cortex considerably beyond the traditional location of Wernicke’s area in the posterior temporal gyrus (Turken and Dronkers, 2011; DeWitt and Rauschecker, 2012). Previous studies have suggested that the left inferior parietal cortex, especially the angular gyrus, is an essential area for speech comprehension as well as production (Mesulam, 1998; Vigneau et al., 2006). In contrast, Hickok and Poeppel (2007) proposed a model of the anatomy of language that excludes the angular gyrus from speech comprehension and production. However, it has been included by more recent studies (Lau et al., 2008; Hartwigsen et al., 2015). With respect to speech production, we found that declined thickness in the left orbitofrontal lobe negatively predicted speech outcome following chronic subcortical stroke. To support this, the PET study in healthy subjects has showed that activity in response to speech production to be localized in the left lateral and medial orbitofrontal cortex (Awad et al., 2007). Additionally, our results showed that reductions of thickness in the left inferior precentral gyrus/pars opercularis gyrus and posterior cingulate were strongly predictive of naming outcome. To support this, the left precentral cortex and inferior frontal gyrus are demonstrated to associate with action naming and the phonology component during object naming (Schwartz et al., 2012). Our results also accord with the observations in the previous study that increased activation could be observed within the left posterior cingulate area in healthy subjects during a picture identification task (Szaflarski et al., 2011).

Homotopic regions are traditionally thought to hinder language recovery by transcallosal disinhibition of the lesioned hemisphere (Belin et al., 1996; Blank et al., 2003; Naeser et al., 2005; Winhuisen et al., 2005). However, recent studies reveal beneficial effects of language homologs in the right hemisphere to post-stroke language recovery (Xing et al., 2016; Hope et al., 2017). Similarly, we found cortical thickness of the right lateral orbitofrontal cortex to be also related to naming outcome. This finding is further supported by a previous study in which damage to the right prefrontal cortex impaired the language ability of choosing appropriate words (Taubner et al., 1999). However, cortical thickness in the right hemisphere regions showed no considerable difference relative to controls and was neither related to lesion size nor integrity of connecting fiber tracts in the present study. This likely reflects premorbid inter-individual differences in the right hemisphere regions that protect from language deficits after left subcortical stroke. Patients with greater premorbid cortical thickness in the right hemisphere might have better recovery of language deficits after left subcortical stroke.

The current study demonstrated that there was a significant reduction of cortical thickness in the left hemisphere regions that related to language outcomes among patients as compared to the control subjects. These results might be indicative of secondary cortical atrophy in regions remote from stroke lesions as indicated in previous studies (Zhang et al., 2014). However, the exact mechanisms for structural impairment of the remote cortex remain unclear. It is assumed that fiber disconnection may account for this finding. By combining structural measurements with tractography, we found positive relationships between cortical thickness in the identified cortex and microstructural integrity of the connecting fiber tracts to subcortical lesions, but no significant correlation was seen the contralateral homologous tracts. Moreover, there was marked reductions of FA values in the tracts of lesional hemisphere relative to their contralateral homologs. These results might be reflective of secondary cortical effects of connecting fibers disruption after chronic subcortical stroke, as proposed by the diaschisis theory (Duering et al., 2012). Supporting this, previous studies have shown that focal cerebral infarction can induce progressive neuronal damage in the ipsilateral thalamus and substantia nigra remote from the lesion site (Nakane et al., 1992; Herve et al., 2005). Our results are also consistent with the findings of several longitudinal studies showing reduced thickness in the connecting cortical regions as a resultant of acute subcortical stroke (Cheng et al., 2015; Duering et al., 2015) or cortical remodeling in the bilateral hemisphere after basal ganglia infarction over several months (Liu et al., 2020). Additionally, retrograde neural degeneration of the retinal nerve fiber has been demonstrated among patients with occipital lobe stroke (Jindahra et al., 2012). Given the reconstructed tracts represent the anatomical connections between cortex and stroke lesions, these data implicate a possible link between subcortical aphasic stroke and ability to engage remote cortex in language processing secondary to fiber degeneration. Nevertheless, the casual effects of white matter disconnection on remote cortical atrophy cannot be determined from our results. The pathological mechanisms underlying the damage of connecting tracts are unclear. It might include retrograde or anterograde degeneration, dendrite shrinkage, and neuronal apoptosis because of loss of synaptic input (Siffrin et al., 2010).

Alternatively, acute subcortical aphasia might be resultant of concurrent cortical HP (Olsen et al., 1986; Wallesch et al., 1997). SPECT and perfusion weighted imaging measures revealed that focal lesion could cause remote cortical HP and hypometabolism (Carrera and Tononi, 2014). No significant differences in language scores were found between the two subgroups when controlling for key variables in the present study. In addition, the cortical thickness analysis showed no deference in cortical thickness of the left hemisphere in patients with HP relative to those with normal perfusion. This suggests that atrophy in the connected cortex was not driven by HP. However, cortical hypometabolism is also suggested in cases with subcortical aphasia (Halkar et al., 1997; de Boissezon et al., 2005; Sebastian et al., 2014). Additionally, specific cortical-subcortical circuits have been suggested to associate with language processing (Teichmann et al., 2015; Akinina et al., 2019; Jacquemot and Bachoud-Lévi, 2021). Thus, multiple neural mechanisms might contribute to aphasia outcomes after chronic subcortical stroke.

There are several limitations in the present study. First, this is a cross-sectional investigation that might preclude the empirical evidence of remote cortical atrophy mediated by neurodegeneration of the connecting fiber tracts. Further longitudinal study is expected to confirm the present findings by providing the dynamics of secondary neurodegeneration in remote cortical regions. Secondly, the DTI findings should be interpreted with caution. The extracellular volume effects of free water have been proposed to contaminate and invalidate DTI metrics (Pierpaoli et al., 1996; Pasternak et al., 2009). Additionally, it has been suggested that free-water correction would improve DTI-based tract reconstruction and tissue specificity (Metzler-Baddeley et al., 2012; Hoy et al., 2014; Guder et al., 2021). Considering the fact that atrophy and neuroinflammation can still be observed even several months after stroke, free-water correction will strengthen the reliability of DTI analysis especially in stroke patients. Thirdly, the number of patients with different perfusion status was relatively small in the subgroup analyses, which also restricted the generalization of the conclusions. Further study is need to confirm the present findings with a larger sample of populations. Fourthly, speech therapy experience across patients might drive recruitment of cerebral cortex capable of supporting language function, thereby affects the language outcome. However, the type and dose of speech therapy are not easily quantifiable as a confounding factor in the current regression analyses. Nevertheless, our results are not likely driven by a specific language therapy because the nature and quantity of speech therapy varies between individuals. Lastly, the lack of targeted therapeutic intervention, such as the restoration of cerebral perfusion, was a limitation of the study. Direct intervention of cerebral flow should permit inferences on the HP effects on cortical thickness changes or language outcomes as previously reported (Hillis et al., 2002). Notably, our results were obtained by using FreeSurfer software that could detect changes in cortical thickness at the submillimeter scale. Additionally, the present study demonstrates the feasibility of investigating the neural base for subcortical aphasia with a combination of surface-based thickness analysis with connecting fiber tracking.