Figure 3. Context-dependent working memory drives subspace rotation and representation reorganization in PFC and PPC.

(A) Predicted and empirical rotation angles in four trial epochs for the three models—ordinal, pre-allocate, and overwrite—as well as the prefrontal cortex and the posterior parietal cortex. (B) Subspace rotation angles between the Remember 1^st and the Remember 2^nd contexts across task epochs. Rotation angle variability was assessed via bootstrap resampling of trials and was then compared to a control condition generated from a synthetic dataset (see Methods). (C) Beta-space representation of the first (dashed lines) and second locations (solid lines) during the second delay in PFC (left) and PPC (right). (D) Visualization of spatial code reorganization of the PFC. Neural responses during early (250 ms post-second stimulus offset) and late (1250 ms post-second stimulus offset) phases of the second delay epoch in the Remember 2^nd context were projected into low-dimensional space. Responses were color-coded based on the spatial location of either the first or second stimulus, using the same color-location correspondence shown in panel C. During the second delay period, separability of the first stimulus representation decreased, while separability of the second stimulus representation increased. (E) Quantification of geometrical separability for the first and second stimuli during the second delay epoch in Remember 1^st and Remember 2^nd contexts for both the prefrontal cortex and the posterior parietal cortex. *, **, *** indicate p < 0.05, p < 0.001, p < 0.0001, evaluated using a two-tailed paired t-test.