Aphasia is a devastating outcome of stroke. The most widespread current rehabilitation approach for aphasia is speech and language therapy; however, many individuals with post-stroke aphasia still struggle to achieve satisfactory language improvements. Transcranial direct current stimulation (tDCS) shows promise as a complementary treatment to traditional speech and language therapy [see reviews by (1–3)]. Transcranial direct current stimulation (tDCS) is a safe, non-invasive neuromodulation technique that applies low-intensity currents (1–4 mA) between electrodes mounted on the scalp. This current depolarizes or hyperpolarizes neuronal resting membrane potentials and thereby alters cortical excitability (4, 5). Generally, the cortical excitability is increased by anodal tDCS while it is decreased by cathodal tDCS over the same area. Experimental and human studies suggest tDCS elicits long-lasting, polarity-dependent changes in cortical excitability [e.g., (4)]. The after-effects of tDCS are mediated by multiple mechanisms that modify synaptic efficacy similar to those underlying long-term potentiation (LTP) and long-term depression (LDP) of synaptic activity (6–10). Further, synaptic efficacy is most effectively modulated when stimulation coincides with ongoing behavior (e.g., task practice) (11, 12). These state-and activity-dependent principles help explain why pairing tDCS with behavioral therapy can yield additive or synergistic benefits in neurorehabilitation.

The most frequently used approach for tDCS treatment in post-stroke aphasia is stimulation of the intact left hemisphere cortical regions (13–18). An emerging line of work targets the cerebellum [for review, see (19, 100, 101)], particularly the right cerebellum as a nontraditional node for modulating the language network in aphasia. The rationale for stimulating the cerebellum is derived from neuroanatomical, imaging and clinical work including the cerebellar cognitive affective syndrome framework, which establishes the cerebellum’s contributions to higher-order cognition and language functions (20–24). The right cerebellum is anatomically positioned to influence left-hemisphere language circuits via robust, crossed cerebro-cerebellar pathways (25–27). Studies show that cortical projections to pontine nuclei enter the cerebellum via the middle cerebellar peduncle, while Purkinje cell output modulates deep cerebellar nuclei which project through the superior cerebellar peduncle and thalamus back to frontal, parietal, and temporal language areas (28). Cerebellar tDCS is thought to alter Purkinje cell firing and dentate-thalamo-cortical output, thereby shaping the excitability and plasticity of the contralateral cortical regions [for a detailed review see (29)]. Clinically, several studies report that right cerebellar tDCS combined with language therapy improves naming and functional communication in post-stroke aphasia, motivating a closer look at which individuals benefit most (30–34).

Treatment response to neuromodulation varies widely across individuals, and is influenced by diverse factors including demographics (35), lesion characteristics (36–39), baseline aphasia severity (40), and baseline neuroanatomical differences in tissue integrity and organization across relevant gray matter or white matter regions (41–43). White matter architecture may be especially pertinent for cerebellar stimulation, given its reliance on long-range cerebro-cerebellar loops. In our recent work using diffusion tensor imaging (DTI), we found that people with post-stroke aphasia show reduced white matter integrity in tracts connecting the right cerebellum and the lesioned left hemisphere, and that lower integrity in these pathways was associated with poorer picture naming skills (44). Complementary evidence implicates cerebellar peduncle microstructure and right cerebellar gray matter atrophy in aphasia severity and therapy gains (45, 46), paralleling broader findings that neuroanatomical biomarkers predict aphasia severity and responsiveness to treatment. Despite this convergence, no study has directly tested whether baseline properties of cerebro-cerebellar tracts specifically predict or moderate the efficacy of cerebellar tDCS in post-stroke aphasia.

Addressing this gap, the present exploratory study asks whether baseline DTI-derived fractional anisotropy (FA) and mean diffusivity (MD) in cerebro-cerebellar pathways predict or moderate the effects of right cerebellar tDCS, delivered in conjunction with computerized language therapy, on longitudinal language outcomes (post-treatment, 2 weeks post-treatment, and 2 months post-treatment) relative to sham. Our primary aim was to identify tract-specific predictors of treatment effects for the trained outcome (trained naming) and for generalization (untrained naming and functional communication). Based on our prior behavioral and imaging findings, we hypothesized that tracts connecting the stimulated right cerebellum to left-hemisphere language cortex and tracts connecting the left-hemisphere language cortex to the stimulated right cerebellum will be the most informative biomarkers, with greater integrity conferring stronger treatment-related gains. As a secondary aim, we compared anodal and cathodal cerebellar stimulation, testing whether polarity-dependent differences reported previously translate into distinct predictive profiles. We hypothesized that cathodal cerebellar tDCS may be more likely to show treatment and predictive effects, based on previous findings (31, 33).

We briefly present the behavioral effects of tDCS versus sham on trained naming (Naming 80), untrained naming (PNT), and functional communication skills (ASHA FACS CI and QDC scales) at post-treatment, 2 weeks post-treatment, and 2 months post-treatment. Subsequently, we present results that address the question of which white matter tracts predicted (i.e., moderated) the treatment effects. Results for the tDCS polarity groups combined (anode and cathode) are shown first followed by group specific results (anode or cathode). The behavioral results have been previously reported in detail in Sebastian et al. (33) and Kim et al. (47).

In the present study, we asked the question of whether baseline cerebro-cerebellar white matter integrity may predict effects of cerebellar tDCS treatment coupled with computerized aphasia therapy. Our results indicate that baseline cerebro-cerebellar white matter tract integrity predicts treatment response to cerebellar tDCS in post-stroke aphasia rehabilitation. Specifically, higher FA and lower MD in descending cortico-ponto-cerebellar pathways predicted larger gains from right cerebellar tDCS paired with language therapy, with effects sustained for 2 months post-treatment. These white matter findings occurred within the context of behavioral improvements showing that cerebellar tDCS enhanced therapy effects on untrained naming (PNT), and one measure of functional communication (ASHA-FACS QDC). However, cerebellar tDCS did not enhance behavioral therapy effects for trained naming (Naming 80) and another measure of functional communication (ASHA-FACS CI). These findings replicates our prior work and is in line with others’ studies (1, 3, 30–32, 46, 70) showing that pairing cerebellar tDCS with aphasia therapy can yield durable improvements while also highlighting that the magnitude and specificity of adjuvant benefits depend on the language measure tested. For a discussion of the behavioral tDCS effects of trained and untrained tasks, we refer the reader to Sebastian et al. (33) and Kim et al. (47).

Our primary hypothesis was supported: baseline properties of cerebro-cerebellar white-matter tracts predicted or moderated the treatment effects of cerebellar tDCS relative to sham. The most consistent pattern emerged for descending cortico-ponto-cerebellar pathways connecting the lesioned left cerebral cortex to the stimulated right posterolateral cerebellum. For functional communication, the integrity of left-hemisphere descending pathways was the strongest predictor of cerebellar tDCS treatment gains. Specifically, left FPC integrity reliably predicted tDCS improvements on both ASHA-FACS scales (CI and QDC), while right MCP showed associations with untrained naming gains on the PNT. These findings underscore that the cerebellum’s capacity to enhance language recovery depends critically on preserved input from spared left-hemisphere tissue and is consistent with previous findings on the relevance of cortico-ponto-cerebellar pathways for stroke recovery and language processing (44, 67, 68, 75). Notably, distinct white matter pathways also predicted response to language therapy alone (sham condition). For example, left OPC integrity predicted untrained naming (PNT) gains in the sham group, consistent with studies showing that white matter integrity predicts responsiveness to language rehabilitation independently of neurostimulation (43, 76–78). This dissociation suggests that different cortico-ponto-cerebellar pathways may support distinct mechanisms of language recovery.

The trained naming task (Naming 80) showed a different predictive pattern: bilateral pathways were associated with treatment gains, including right FPC, right PPC, and right OPC, tracts connecting the right cerebral cortex to the left (non-stimulated) cerebellum. This divergence likely reflects differences in the neural substrates supporting specific treatment effects versus generalization (78, 79). Trained naming involves repeated practice on a constrained set of items, which may engage more distributed or bilateral cerebellar networks related to learning. Prior neuroimaging work demonstrates that cognitively demanding tasks engage bilateral cerebellar lobules along with prefrontal and parietal cortices (21), supporting this interpretation. The bilateral pattern for trained items suggests that even unilateral cerebellar stimulation can modulate language functions through distributed network effects.

Regarding the ascending DRTC pathway linking the stimulated right posterolateral cerebellum to the lesioned left cerebral cortex, most effects were null, with two exceptions. For functional communication ASHA-FACS CI, higher MD showed a trend-level association with greater gains after tDCS, and this pattern differed significantly from sham, where higher MD was associated with smaller gains. For the trained naming task (Naming 80), higher MD predicted greater gains in the sham condition only. Given that the DRTC is the primary ascending pathway conveying cerebellar output to cortical association areas, and that our prior work demonstrated positive associations between DRTC integrity and naming performance in chronic post-stroke aphasia (44), it was surprising that we did not find more associations with language gains after cerebellar tDCS. This may suggest that descending cortico-cerebellar input is more critical than ascending cerebellar output for determining cerebellar tDCS response. Alternatively, it is possible that sample size limited our ability to detect more subtle modulation effects in this pathway.

The secondary hypothesis that cathodal cerebellar tDCS may be more likely to show treatment and predictive effects was not supported. Contrary to our hypothesis, but consistent with several prior reports (30–33), the polarity effects (anode versus cathode) were mixed across the outcome measures, and predictive relations with baseline tract integrity were largely similar for anodal and cathodal stimulation. Multiple factors likely contribute to polarity inconsistency in the cerebellum (80–83) including neuroanatomical factors interacting with the tDCS stimulation, such as the complexity of gyral folding, the types of neuronal cells engaged, and other individual differences. Therefore, the potential impact of tDCS polarity on treatment efficacy and predictive effects in the specific context of cerebellar tDCS combined with aphasia therapy remains an open question for future research.

A highly consistent pattern across tasks and the descending tracts was that larger cerebellar tDCS treatment gains were associated with a diffusivity profile indicative of higher white-matter integrity [higher FA, lower MD (77, 78, 102)], particularly within the context of stroke-induced white matter degeneration (75, 84, 85). Participants with lower baseline integrity in the descending cortico-ponto-cerebellar pathways tended to show smaller cerebellar tDCS gains, underscoring that the cerebellum’s capacity to influence language depends on the inputs from spared left-hemisphere tissue. This is consistent with anterograde degeneration following cortical lesions (75) and with reports that have shown that white matter alterations in the cerebellum, the superior and middle cerebellar peduncles, and the pons correlate with recovery of motor function after stroke (80, 86). However, it is also important to acknowledge the known challenges of interpreting microstructural tissue integrity from DTI-derived metrics, especially in relatively low-resolution diffusion data (87–89, 102). Nevertheless, the convergent association between better-organized white matter and stronger cerebellar tDCS response parallels broader neuromodulation findings (42, 43, 76, 90, 91), i.e., individuals with relatively preserved structural connectivity within functionally relevant networks and/or adaptive reorganization of spared regions are more likely to benefit (44, 81), in keeping with state-dependent and metaplastic accounts of tDCS (92).

Because we created the cerebro-cerebellar tracts by combining the cerebellar peduncles with the contralateral corticopontine projections, we anticipated that the predictive effects may be driven by either or both of the separate segments. Across composite tracts, language measures, DTI metrics, and tDCS polarity groups, both corticopontine and cerebellar segments contributed relatively equally to predictive effects. For the cerebellar peduncles specifically, predictive effects involved both the stimulated (right) and non-stimulated (left) hemispheres. Notably, a functional dissociation emerged between cerebellar peduncle types: MCP integrity predicted treatment gains across all language measures, whereas SCP integrity predicted only functional communication outcomes (ASHA-FACS CI and QDC). This pattern suggests selective associations between specific cerebellar white matter pathways and language functions. The measures themselves differ meaningfully: ASHA-FACS provides global ratings of functional communication based on clinician and caregiver observations, whereas PNT and Naming80 are objective, performance-based measures of discrete naming ability. This distinction may be critical for understanding the dissociation. The MCP, as part of the afferent pathway, transmits task-specific information from cortex to cerebellum and supports lower-level language functions. In contrast, the SCP, as part of the efferent pathway, conveys cerebellar output back to cortex and supports higher-level language functions (64, 67, 82). Our differential findings may therefore reflect this functional dissociation between cerebellar peduncles, consistent with other evidence of pathway-specific contributions to distinct language processes (45, 66).

The bilaterality of predictive effects from the cerebro-cerebellar tracts or tract segments (e.g., peduncles) was surprising given that the cerebellar stimulation was lateralized (right hemisphere), the cerebro-cerebellar tracts are contralateral, and the cerebellar peduncles do not bridge within the cerebellum. There are several possible explanations to this finding. First, although our montage concentrates the electric field in the right cerebellum, current field modeling indicates some spread to the left cerebellum (32). Second, indirect modulation of the non-targeted cerebellum is plausible via propagation of tDCS effects through structural connections directly between the cerebellar hemispheres (e.g., cerebellar commissure) or indirectly via cross-hemispheric connections between cortices (e.g., corpus callosum) and/or via feedback loops or interactions between cerebro-cerebellar tracts. In either case, indirect modulation of the non-targeted cerebellum may have contributed to its predictive effects. Functional neuroimaging studies provide convergent evidence for such distant effects of cerebellar tDCS, including modulation of BOLD activity and connectivity in contralateral cortical nodes within motor and cognitive networks and changes in language-related circuits (52, 83). Together, these observations suggest that cerebellar tDCS acts as a network-level intervention whose behavioral impact depends on bilateral cerebro-cerebellar loop integrity.

An important point to note is that we did not detect any significant cerebellar tDCS carryover effects (see section 2.8.1), which seems likely because we took several steps to minimize the potential for carryover effects. We used a 2-month break in between phase 1 and phase 2 with the assumption that the tDCS effect will have washed out by that time. In addition, we also adopted an analysis approach utilized by our colleagues (93, 94) to allow for the case that the tDCS effect does not wash out in 2 months. In our analysis, we used the “change in naming or functional communication” as the main outcome, i.e., the score at each post-treatment/follow-up minus the score at baseline of the respective phase. This means that any effect of tDCS versus sham found in phase 2 will not reflect the improvement itself at the level of performance in naming, i.e., the fact that the participants who got tDCS first are naming more accurately than those who got sham. Therefore, any effect of tDCS versus sham that carries over from 2 months of phase 1 to only the level of performance in phase 2 will cancel in the tDCS versus sham comparisons of “change in naming” at phase 2. In this way, we eliminated one of the possible “carryover” effects of tDCS: the improvement at the mere level of performance (94).

This study has several limitations. First, while our sample size is comparable to other cerebellar tDCS studies, it constrained statistical power and precluded more complex modeling approaches, such as including random effects for treatment or timepoint and testing higher-order interactions (e.g., polarity-by-tract-by-time effects). Second, our diffusion acquisition parameters, though clinically pragmatic, were limited in diffusion weighting, gradient count, and voxel size. These constraints precluded advanced preprocessing techniques and reliable decussation estimation, necessitating manual combination of cerebral and cerebellar tract segments. As with all DTI-based studies, our microstructural inferences remain indirect and sensitive to acquisition and modeling choices. Third, we did not differentiate between aphasia subtypes (fluent versus non-fluent) and stroke types (ischemic vs. hemorrhagic), which may influence outcomes (95). However, we note that our treatment targeted naming, a core deficit across aphasia subtypes, and our sample consisted exclusively of chronic stroke patients (>6 months), a phase when acute pathophysiological differences between stroke types have largely resolved. Fourth, we did not include canonical perisylvian language tracts (e.g., arcuate fasciculus, superior longitudinal fasciculus, inferior fronto-occipital fasciculus) as comparison structures. Our focus on cerebro-cerebellar pathways was driven by our research question examining cerebellar stimulation effects; however, without these control tracts, we cannot definitively determine whether our predictive effects are specific to cerebellar circuitry or reflect general language-related white matter integrity. Prior literature shows that damage to these tracts robustly predicts aphasia severity and therapy response (96–99). A critical next step is therefore to test whether cerebellar tract integrity explains unique variance in tDCS response beyond lesion burden and perisylvian tract status, and whether combining these biomarkers improves patient stratification.

In conclusion, this exploratory study provides novel evidence that baseline cerebro-cerebellar tract integrity moderates the efficacy of right cerebellar tDCS delivered in conjunction with aphasia therapy: higher FA and lower MD in these loops predict larger short- and long-term gains on naming and functional communication. The bilateral nature of predictive effects, despite unilateral stimulation, is consistent with network-level propagation documented in functional neuroimaging studies of cerebellar tDCS. Future work should (a) replicate these findings in larger, more diverse cohorts; (b) incorporate canonical language tracts and lesion metrics to test specificity; (c) use advanced diffusion and functional imaging to refine tractography and quantify network-level modulation; and (d) leverage individualized modeling and state-dependent protocols to optimize polarity, dosing, and montage. Together, these steps will enable biomarker-guided, mechanistically informed personalization of adjuvant neuromodulation for post-stroke aphasia.