Review Papers Claim by Claim

Cognitive processes are disentangled at cortex-wide scales — concise review

The authors used a novel mouse VR delayed match to evidence task combined with cortex-wide widefield GCaMP6s imaging to show that evidence accumulation, short term memory, and choice map onto near orthogonal cortex-wide activity subspaces that become more entangled on error trials, and that only the memory subspace aligns with spontaneous activity timescale hierarchies, suggesting partial dependence on intrinsic circuit timescales Preprint

Detailed review and critique

Summary of goals methods and main findings

The preprint introduces a new delayed match to evidence virtual navigation task in mice, manipulates sample coherence and virtual gain to dissociate evidence accumulation sample and delay durations, and records mesoscale cortex wide GCaMP6s activity while mice perform the task. They apply PCA dPCA and SVM decoders to show that major cognitive components map to near orthogonal cortex wide subspaces that are more entangled during error trials and that the memory subspace is uniquely aligned with a spontaneous activity timescale hierarchy Main

Strengths

• Task design: careful independent manipulations of sensory coherence and virtual gain create a 3D task space allowing dissociation of accumulation duration memory and decision variables, which strengthens causal interpretation of decoding results Task
• Large scale recording and rigorous preprocessing: mesoscale widefield hemodynamic corrected imaging at 20 Hz and standardized preprocessing with motion correction and dual wavelength correction increases signal reliability across cortex Imaging
• Multimethod analysis: combining encoding models dPCA PCA and SVM decoders with permutation testing and simulations to probe timescale estimation shows methodological breadth and internal consistency checks Analysis

Weaknesses limitations and potential biases

1. Reliance on widefield Ca2+ signal and GCaMP6s dynamics: mesoscale fluorescence reports predominantly L2 3 excitatory population activity and is lowpass filtered by Ca kinetics and hemodynamic correction; inference about fast spiking or inhibitory contributions is limited and timescale estimates depend on preprocessing and sensor dynamics Ca
2. Species and circuit specificity: results are from transgenic mice expressing GCaMP6s in L2 3 excitatory neurons and may not generalize to other species layers cell types or circuit motifs; conclusions about cortex wide organization should be tempered by this sampling bias Model
3. Interpretation of orthogonality: dPCA and decoding-axis angle analyses reveal near orthogonality but interpretation that this equals functional disentanglement requires caution: near orthogonality can arise from high dimensional recordings and statistical demixing; permutation controls are used but residual correlations across subspaces remain and causal tests (e.g., perturbations) are not provided Geometry
4. Error trials interpretation: the observation that subspace orthogonality collapses on error trials is compelling but may reflect multiple mechanisms including attentional lapses sensory noise or action variability; further control analyses separating e.g., arousal or running kinematics contributions would strengthen claims Error

Reproducibility and data availability

The methods section is detailed covering task apparatus imaging preprocessing and analysis pipelines and the authors state the full dataset will be publicly released on publication which supports reproducibility; however code and data links are currently absent in the preprint and critical analysis choices such as regularization parameters pixel subsampling and SNR thresholds require exact code to fully reproduce results Data

How convincing are the main claims

The core empirical claim that multiple cognitive processes correspond to largely separable cortex wide subspaces is supported by converging analyses (dPCA PCA SVM) and permutation controls; confidence is moderate to high for the existence of separable statistical subspaces in these mice performing this task but lower for mechanistic claims that these subspaces represent independent dedicated circuit modules without targeted perturbations Main

Specific suggestions to strengthen the manuscript

1. Provide code and data release prior to or concurrent with public preprint posting including preprocessing pipelines and random seeds to cement reproducibility.
2. Perform targeted causal perturbations such as optogenetic silencing of area subsets or temporally precise disruption of putative memory subspace dynamics to test whether subspace axes are causally necessary for behavior.
3. Report control analyses explicitly separating effects of arousal locomotion and hemodynamics on subspace geometry and error-related entanglement (e.g., regress pupil running and residualize traces before dPCA to test robustness).
4. Where possible include single cell or two photon validation of key subspace signatures to assess cell type specificity of the observed mesoscopic subspaces.

Paper scoring quick metrics

Key insight

Statistical near orthogonality of cortex wide subspaces suggests the brain leverages high dimensional population activity to multiplex cognitive processes across shared anatomical tissue while preserving separability at the level of population geometry rather than strict anatomical localization Conceptual

Falsification test

This framework would be falsified if temporally precise perturbations that selectively disrupt a decoding axis (for example optogenetic pattern perturbation projected onto the decoded memory axis) impaired the corresponding behavior without affecting other task components; lack of selective behavioral impairment would argue against the functional separability interpretation How

Actionable next steps and reproducible experiments

1. Publish code and preprocessed datasets and provide notebooks to reproduce dPCA and angle permutation analyses.
2. Design optogenetic perturbations targeted to selectively change activity along decoding axes and measure behavioral specificity.
3. Record single cell electrophysiology or 2 photon imaging in target areas to map cell type contributions and validate that mesoscopic subspaces are not an artifact of pooling.