Can cerebellar tDCS improve working memory in early course psychosis?

A crossover test of cerebellar stimulation

The study asked whether excitatory transcranial direct current stimulation (tDCS) over the anterior cerebellum can reduce the verbal working memory deficit seen in early course psychosis. tDCS delivers a weak constant current through scalp electrodes; here the active electrode sat on the midline just below the inion, with the return electrode on the right arm deltoid, at 2 mA for 25 minutes.

The design was a double-blind, randomized, crossover-controlled trial. Twenty-seven participants with an early course psychosis diagnosis received either active stimulation or sham before a verbal working memory fMRI task, then returned about a week later for the counterbalanced session. A separate group of 36 healthy controls received sham only, providing a normative comparison rather than a second treatment arm.

Because each participant served as their own comparison, the primary contrast is stimulation versus sham within the psychosis group. That design controls for stable individual differences in ability, but it also means the headline result is an average across a clinically heterogeneous sample.

The working memory deficit was present at baseline

On the sham visit, the psychosis group showed significantly lower accuracy on the Sternberg verbal working memory task than healthy controls (beta = −2.03, 95% CI −3.47 to −0.59, p = 0.006). The task presents sets of one, three, or five consonants and then probes a single letter, so accuracy indexes how well participants hold and match verbal material.

Two checks make this deficit interpretable. Accuracy fell as set size rose from one to five letters (beta = −1.66, p = 0.0024), confirming the difficulty manipulation worked. And the size of that difficulty effect did not differ significantly between groups (p = 0.19), suggesting the task was not simply harder for patients in a way that would confound the comparison.

This baseline gap matters for reading everything that follows: any stimulation effect is being measured against a group that genuinely underperformed, not against a deficit assumed from theory alone.

Stimulation did not shift accuracy on average

Comparing the stimulation and sham sessions within the psychosis group, verbal working memory accuracy did not differ significantly (beta = 0.36, 95% CI −1.25 to 1.96, p = 0.66, d = 0.016). The effect size is essentially zero, so the average participant gained nothing measurable in accuracy from active stimulation.

The neural manipulation check told a more mixed story. Task-related cerebellar activation showed no differences that survived cluster correction. Connectivity, however, did change: stimulation was associated with heightened intrinsic connectivity in an anterior cerebellum–parietal network and reduced extrinsic connectivity in the 0.08–0.15 Hz range (FDR-corrected p < 0.05).

The authors read this as evidence that the stimulation engaged the cerebellum through frequency and network coupling rather than through a simple activation shift. That interpretation is theirs; the packet supports the connectivity finding, not a claim that this connectivity change caused any behavioral gain.

A responder subgroup drove the heterogeneity

Exploratory analyses examined absolute change in accuracy between conditions rather than the group mean. This split the psychosis sample into 14 responders (d = 0.64), 11 non-responders (d = −0.63), and 2 people with stable ceiling-level performance who were excluded from the contrast.

The split was validated against controls. After stimulation, responders no longer differed significantly from healthy controls (t = 0.887, p = 0.38), whereas non-responders remained significantly less accurate (t = −3.710, p = 0.0003). Responders also started lower: their sham-visit accuracy averaged 93.9% (SD 6.64) versus 96.5% (SD 4.46) for non-responders (d = −0.45, p = 0.001), and they improved more once that baseline difference was accounted for (beta = 7.007, p < 0.00001).

So the people who benefited were those with the larger baseline deficit, not those already performing near the task ceiling. The authors note that counterbalanced ordering makes a simple practice effect an unlikely explanation, though the subgroup analysis itself was exploratory and not a pre-specified primary endpoint.

Responder status tracked pre-existing connectivity

The subgroups differed in cerebellar network organization on the sham day, before any active stimulation. Responders showed greater intrinsic anterior cerebellar network activity, and the groups displayed distinct sham-day patterns of intrinsic and extrinsic connectivity within an anterior cerebello-thalamo-cortical network.

Two alternative explanations were probed. Spatial differences in where the stimulation landed did not distinguish responders from non-responders, even at a lowered exploratory threshold, arguing against inconsistent electrode targeting as the cause. Non-responders instead showed reduced intrinsic connectivity in a dorsolateral prefrontal–anterior cingulate network and altered low-frequency signal in the anterior cerebello-thalamo-cortical network regardless of stimulation condition.

Responders also showed greater internetwork connectivity between the anterior cerebello-thalamo-cortical network and a medial cingulate cortex network, again regardless of condition. The authors interpret these as candidate biotypes: stable network features that mark who is likely to respond, rather than effects produced by the stimulation itself.

Preliminary evidence for a targeted subgroup

The study is small and the responder analysis is explicitly exploratory, described by the authors as a descriptive guide for future treatment studies. The primary behavioral endpoint was null, so the responder contrast should not be read as a confirmed treatment effect. The psychosis sample was also diagnostically mixed, spanning mood disorders with psychotic features and schizophrenia-spectrum diagnoses, and 72% were taking antipsychotics.

Tolerability was reasonable: 90% retention, with dropouts attributed to scanner discomfort or contraindications rather than the tDCS protocol. Tingling or itching under the arm electrode was reported by seven individuals, split nearly evenly between sham and stimulation sessions.

The practical implication the paper supports is a screening one. If cerebellar tDCS is pursued for verbal working memory in early course psychosis, the evidence points to selecting people with a measurable baseline deficit and an anterior cerebellar connectivity profile resembling the responder biotype, rather than treating the diagnosis as a single target.

Sources used in this answer