Anterior cingulate input tunes striatal dynamics for context-dependent social preference in a mouse model of autism

A selective, context-dependent social deficit

The paper asks why social preference in autism varies by context. Using a two-sided choice task with a corridor between options, the authors compared Shank3 knockout (KO) mice with wild-type (WT) mice in two situations: a social versus nonsocial side (context 1) and a female versus male side (context 2). In context 1, WT mice spent more time on the social side while KO mice spent more time on the nonsocial side, with a significant genotype × side interaction. In context 2, both genotypes spent more time on the female side and no genotype × side interaction appeared.

The deficit therefore is not a blanket loss of social interest. KO mice also oriented their heads toward the social object less even while inside the social zone, and the same core pattern replicated in the classic three-chamber assay (14 mice per genotype there, 10 per genotype in the two-choice task). Control analyses found no genotype differences in repeated crossings or crossing latency, arguing against a general locomotion or chamber-transition explanation. Female KO mice showed the same context-1 alteration and preserved context-2 responses, and the phenotype persisted when the social stimulus was an unfamiliar KO mouse rather than a WT mouse.

DMS activity is required for normal preference

The dorsomedial striatum (DMS) is the striatal region the study focuses on, a goal-directed reward structure made largely of GABAergic medium spiny neurons (MSNs). To test whether DMS activity is necessary, the authors expressed the inhibitory receptor hM4Di in DMS neurons and gave Compound 21 (C21). Suppressing DMS population activity diminished social preference in both contexts and reduced time oriented toward the social or female stimulus. That establishes necessity but not yet a genotype-specific mechanism.

The more informative step was recording DMS activity directly. Using miniature two-photon calcium imaging with GCaMP7f in MSNs, the authors recorded 1,195 MSNs in 7 WT mice and 1,154 MSNs in 7 KO mice. Calcium-event rates showed a significant genotype × condition interaction in the social–nonsocial comparison, with WT mice relatively higher in the nonsocial condition and KO mice differing in the opposite direction. The key readout, however, was not firing rate but how population activity was organized.

Population embeddings track which choice is preferred

The paper summarizes DMS population activity as embeddings: low-dimensional representations of many neurons' activity at once, so that similar population states sit close together. In trial-endpoint analyses, these endpoints correspond to neural representations of the two stimulus-associated sides, approximated behaviourally by the animal entering predefined stimulus zones. This is a different measure from calcium-event rate, and the two should not be conflated.

The striking result is a cross-context geometry. Endpoint clusters grouped as Social–Female and Nonsocial–Male in WT mice, but as Nonsocial–Female and Social–Male in KO mice, and the same organization appeared in condition-averaged population-vector similarity. In other words, the embedding grouped the two contexts by which option the animal preferred, and the grouping flipped with genotype in line with the behavioural flip. Within-genotype and between-genotype embedding consistency, k-nearest-neighbour location decoding and trajectory analyses supported this reading, with 7 mice per genotype.

A weakened ACC–DMS pathway in KO mice

Functional ultrasound imaging showed reduced functional connectivity between the ACC and DMS in KO mice relative to WT (13 WT and 12 KO mice), including within prefrontal comparisons. Anatomical tracing separated two corticostriatal routes: ACC-derived axons and prelimbic (PL)-derived axons showed minimal overlap, and ACC axons did not colocalize with the striosome marker mu-opioid receptor (MOR), whereas PL axons clearly did. The ACC–DMS and PL–DMS pathways are therefore distinct, with the ACC projection avoiding the striosome compartment.

Molecular work pointed the same way. Using pTRAP, single-nucleus RNA sequencing and a multiplex in situ strategy that labelled striatal cells receiving ACC or PL input, the authors found molecular alterations more pronounced in the striatal matrix than in striosomes. The striosome/matrix fluorescence ratio measure came from 12 sections from 3 mice, so the compartment comparison rests on a small anatomical sample even though the sequencing adds breadth. Together these results describe a specific, matrix-associated ACC–DMS impairment rather than a general cortical or striatal defect.

Inhibiting ACC input erases the genotype difference

If ACC input is what organizes the genotype-specific DMS dynamics, removing it should collapse the difference. That is what happened: when the authors perturbed ACC functional inputs in both WT and KO mice, the embedding features of the two genotypes became indistinguishable by K-Means clustering. This is a manipulation result, not merely a correlation, and it identifies ACC input as necessary for the divergent striatal representations.

Behavioural pathway manipulations sharpened the specificity. Bilateral photoinhibition of the ACC–DMS pathway, using eNpHR3.0 in ACC and fibres in DMS, produced similar context-dependent social preference deficits in WT and KO mice, effectively converting WT preference into an autism-like phenotype with reduced social preference. Head orientation toward the social side was not significantly altered. Inhibiting the PL–DMS pathway with stGtACR2, by contrast, failed to abolish the behavioural differences between genotypes. The pathway comparison, not just the pathway effect, is the point: ACC–DMS dysfunction, but not PL–DMS dysfunction, underlies the context-dependent abnormality.

Activation shifts dynamics and improves preference

The corrective experiment used ChrimsonR to activate ACC axons in DMS with 635 nm laser pulses at 20 Hz. In KO mice, this activation shifted DMS population-state organization toward WT-associated patterns and improved social preference. The effect was context-specific: in context 1 (social versus nonsocial) activation altered how time was distributed between sides, whereas in context 2 (male versus female) no comparable interaction appeared and time on either side was not significantly altered. Head orientation was not significantly affected in either context.

Timing mattered. In a closed-loop version, light was delivered only when the animal entered predefined stimulus-associated zones, corresponding to the embedding endpoints identified offline. Zone-entry-triggered stimulation produced broadly similar behavioural effects to whole-session activation, with a significant stimulation × side interaction in context 1 and none in context 2. Delivering light instead while mice ran in the corridor between sides had limited impact on social preference in either context. The authors interpret this as cortical input preferentially influencing preference during stimulus-zone entry epochs associated with endpoint representations, rather than during corridor traversal; that interpretation is theirs, and the closed-loop groups were 7 KO mice.

Why the pathway, not the preference score, is the takeaway

The paper's most decision-relevant claim is a dissociation: the same Shank3 deletion leaves female-over-male preference intact while reducing social-over-nonsocial preference, and the ACC–DMS pathway tracks the impaired context rather than social behaviour in general. For readers comparing autism models or evaluating social preference as a marker, this argues against treating a single preference index as a fixed readout. The context, the stimulus pairing and the population-level organization of DMS activity all change what the number means.

Two boundaries qualify how far this travels. The work is in adult Shank3 KO mice, and the authors note that social preference can vary with age, so the developmental picture is not established here. The restricted corridor apparatus standardized behaviour for neural-dynamics analysis but, as the authors state, may limit direct translational relevance to more naturalistic settings; the three-chamber replication supports robustness of the behaviour, not of the naturalistic setting. Within those limits, the priority is the pathway comparison: ACC–DMS manipulation moved KO dynamics toward WT patterns and improved preference in the social–nonsocial context, while PL–DMS inhibition did not abolish the genotype difference, making ACC input to the DMS the more specific target this study identifies.

Sources used in this answer