One stimulus, two behavioral strategies
The study trained the same mice on two whisker-dependent discrimination tasks. In the 'what' task, mice had to distinguish two sandpaper textures (P100 versus P1200 grit) and lick for reward on the go texture. In the 'where' task, they had to distinguish positions of the same texture. Crucially, the go stimulus was identical in identity and position across both tasks, so any difference in brain activity could not be attributed to the physical stimulus alone.
While imaging, the authors split Hit trials by what the mouse did as the texture approached. In active trials the mouse moved and whisked vigorously during the sensation period, defined as −1 to 1 s relative to texture stop; in passive trials it sat quietly and let the texture touch the whiskers. Activeness, the percentage of active Hit trials, varied substantially across mice but was similar within a mouse across the two tasks and did not differ significantly between tasks across mice (p = 0.12, signed rank test).
This design is what makes the paper's question answerable. Because the stimulus and the task rule were held constant in the comparison of interest, the authors could ask whether brain-wide engagement follows the external task or the animal's own strategy.
Active search and passive touch split the brain
Across three recording preparations — wide-field calcium imaging of the whole dorsal cortex, combined cortical imaging with multi-fiber photometry across the thalamo-cortico-amygdala network, and photometry at 24 cortical and subcortical sites — active and passive Hit trials produced opposite spatial patterns. Active sensation enhanced Barrel cortex and many frontal areas; passive sensation enhanced Barrel cortex together with posterior areas LM and P.
The difference map (active minus passive) made the divergence clearest: a gradient from negative values in posterior areas to positive values in frontal areas during the sensation period, reproduced across mice. A two-way ANOVA on activeness and task type found a significant activeness effect and an insignificant task effect in both the posterior group (activeness F(1) = 104.9, p = 1.54×10⁻²⁴; task p = 0.13) and the frontal group (activeness F(1) = 2186.4, p = 5.86×10⁻¹⁴³; task p = 0.7), with insignificant interactions. The trial counts behind this comparison were 6324 active-what, 5733 active-where, 1895 passive-what and 1751 passive-where trials from six mice.
The thalamus followed a similar axis: posterior thalamic areas such as MGV and MGD were more active in passive trials, while frontal thalamic areas such as RT and POM were more active in active trials. The authors describe the active strategy as engaging a frontal subnetwork including M2, prefrontal cortex, frontal thalamus and hippocampus, and the passive strategy as recruiting a posterior subnetwork including area P, posterior thalamus and amygdala.
Choice signals follow the strategy, not the task
A classifier trained to decode the mouse's choice (Hit versus correct rejection) showed where choice information lived. In cortex, thalamo-cortico-amygdala and subcortical recordings, the posterior group carried significantly higher weights in the passive strategy (cortex p = 6.95×10⁻⁴), while the frontal group carried significantly higher weights in the active strategy (cortex p = 3.67×10⁻⁵). The same pattern held in the thalamo-cortico-amygdala network (posterior p = 1.36×10⁻⁸; frontal p = 1.07×10⁻⁶) and in subcortical areas (posterior p = 1.96×10⁻⁴; frontal p = 6.75×10⁻⁵).
The asymmetry matters. Posterior areas had significantly positive weights in passive trials and significantly negative weights in active trials, consistent with suppression during active search. Frontal groups, by contrast, did not carry choice information significantly different from zero in the passive strategy. The decoding divergence looked similar when the 'what' and 'where' tasks were analyzed separately.
So the answer to where a choice is represented depends on which strategy the animal adopted, not on which question the task posed. That is a different claim from saying the two tasks are processed identically; it says the dominant organizing variable is internal state.
Posterior areas also track history and training order
A multivariate model quantified how much each variable contributed to neuronal responses while correcting for movement, using trial shuffling of movement variables and cross-validated R² (0.1162 ± 0.0178, mean ± SEM, n = 19 mice, during the sensation period averaged across brain areas). Movement predictors had coefficients an order of magnitude larger than non-movement predictors, confirming that behavioral movement dominates brain-wide dynamics.
Within that model, activeness and performance were positively associated with frontal cortex and negatively with posterior cortex. Internal predictors — trial history, which task came first in the day, and tasks per day — showed the opposite pattern, positively associated with posterior areas. The task parameter itself ('what' versus 'where') did not show a clear frontal/posterior divergence.
Training history added another layer. Among the six cortex-imaged mice (three started with 'what', three with 'where'), the predictors for task and first task differed significantly in posterior areas but not frontal areas (p < 0.001, Mann–Whitney U-test), while frontal areas did not differ by training history (p > 0.05). An unsupervised embedding of the coefficient matrix recovered two subgroups that mixed cortical, thalamic and subcortical areas and aligned with the frontal/posterior division.
Task identity appears when frontal dominance drops
The paper's most specific result concerns when 'what' and 'where' information becomes readable. Using unique R² values — the variance explained by one variable after shuffling all others — the authors found opposing gradients: movement was positive in frontal areas and near zero posteriorly, while task was near zero frontally and positive posteriorly. A task modulation index above 0.5 marked predominantly task-related encoding in posterior regions during the sensation period.
Trial-level analysis sharpened this. In trials where M2 responded early (M2-first), frontal areas were strongly engaged and posterior cortex was suppressed, in both tasks; frontal areas such as M2 and ALM still differed between tasks (p < 0.001), which the authors suggest could reflect slightly different movement strategies. In trials where Barrel cortex responded first (BC-first), task encoding appeared in posterior areas: during 'what' trials, areas P, LM and POR showed positive responses, whereas during 'where' trials these areas showed a significant decrease (p < 0.001, Mann–Whitney U-test).
This explains why simple averaging across trials did not reveal task differences: they were likely occluded by M2-mediated suppression. The authors' interpretation is that the posterior subnetwork can encode object identity and location, but only when the frontal subnetwork's dominance is discarded.
A reciprocal account replaces parallel streams
The paper argues against reading its results as support for two parallel cortical streams for identity and location. In a large subset of trials the frontal/posterior divergence was similar in both tasks, which the authors say weakens the two-stream hypothesis. They also find the perception-action framing incomplete: in the passive case the mouse still eventually licks, so some action planning should remain in frontal areas, and prior work from the group showed both M2 and P can maintain working memory.
Instead they propose two alternating streams selected by current internal state. In their account, active sensation lets the frontal subnetwork take control and inhibit the posterior network, possibly through top-down control of M2 onto the thalamic reticular nucleus, which inhibits thalamic nuclei projecting to posterior cortex. Without frontal control, the posterior network integrates sensations with internal representations, as during passive viewing or focused touch.
The authors also note that decreased performance and prolonged learning in passive mice may indicate the posterior subnetwork is recruited in a more stressful context, and they explicitly flag that other movement-related parameters not measured here could explain some neuronal task differences.
What the mouse whisker model leaves open
Several boundaries qualify how far these results travel. The recordings came from male mice only, so sex differences are untested. The imaging preparations sample different fractions of the brain: cortical imaging covered layer 2/3 excitatory neurons across the dorsal cortex, while photometry targeted selected sites, and roughly 80% of tracked fibers were correctly localized to the desired area. The classifier and encoding analyses rest on modest mouse numbers per preparation (n = 6 for cortex, n = 4 for thalamo-cortico-amygdala, n = 9 for subcortex).
The causal language in the paper is explicitly a proposal. The inhibitory M2-to-thalamic-reticular-nucleus route is offered as a possible mechanism, and the authors state that future studies will need to dissect each subnetwork and its gating interactions, in both head-fixed and natural conditions.
The practical priority that follows is a comparison rule rather than a brain-region label: when asking where identity or location is encoded in a behaving animal, first separate active from passive trials. Pooling them can hide posterior task signals behind frontal suppression, and treating 'what' and 'where' as the primary split can miss the strategy effect that the data show is larger.
Sources used in this answer
Brain-wide pathways for object identity and location are dominated by internal strategies
Active search and passive touch recruit distinct frontal and posterior brain-wide subnetworks in mice, and posterior areas encode object identity and location mainly when frontal dominance is set aside.