We trained head-fixed mice to perform a cross-modal sensory selection task³⁹ in which they switched between “respond-to-touch” and “respond-to-light” rules in different blocks of trials (Fig. 1a,b). During the respond-to-touch rule, mice responded to a tactile stimulus by licking to the right reward port, and to withhold licking following a visual stimulus. During the respond-to-light rule, mice responded to a visual stimulus by licking to the left reward port, and withheld licking following a tactile stimulus. Blocks of trials under the two rules alternated multiple times in a session (Fig. 1b; 4-6 blocks per session, ∼60 trials per block). No cue was immediately provided after rule switching, so mice detected the rule change through reward availability over the first few trials. On the 9th trial, a drop of water from the correct reward port was released following the stimulus to ensure switching (Fig. 1b, black dots).

The two stimulus-response rules determined the behavioral relevance of sensory stimuli. For example, tactile stimuli were behaviorally relevant in respond-to-touch blocks but irrelevant in respond-to-light blocks. Four trial outcomes were defined based on the behavioral relevance of a sensory stimulus and on the response of the mouse. Correct licking responses following a relevant stimulus were “hits”, and correct withholding of responses following an irrelevant stimulus were “correct rejections”. Failed responses to a relevant stimulus were “misses”, and incorrect licking responses (following an irrelevant stimulus and/or at the incorrect port) were “false alarms”. Two sensory modalities and four trial outcomes composed eight trial types in the cross-modal sensory selection task (Fig. 1c). Only hit trials were rewarded with a drop of water, and the other trial types were neither rewarded nor punished. We used stimulus strengths (tactile stimulus: single whisker, 20 Hz, 150 ms, ∼800° s⁻¹; visual stimulus: 470 nm LED, 150 ms, ∼3 μW) that yielded performance of approximately 75% correct for both respond-to-touch and respond-to-light blocks (Fig. 1d; touch: 74 ± 1%, light: 75 ± 1% (mean ± sem), p = 0.56, paired sample t-test, n = 12 mice). Thus, mice flexibly responded to tactile and visual stimuli in a rule-dependent manner.

To examine how the task rules influenced the sensorimotor transformation occurring in the tactile processing stream, we performed single-unit recordings from sensory and motor cortical areas including S1, S2, MM and ALM using 64-channel silicon probes (Fig. 1e-g and Fig. S1a-h). In tactile stimulus trials, some neurons in sensory and motor cortical areas showed prominent touch-evoked responses in both block types (e.g. Fig. S1c). Most of these touch-evoked responses appeared enhanced and sustained at timepoints prior to licking in the respond-to-touch (tactile hit, “tHit”) blocks when compared with similar timepoints in the respond-to-light (tactile correct rejection, “tCR”) blocks (Fig.1f and Fig. S1a,e). In contrast, some neurons only showed increased activity when mice made lick responses, regardless of the block type (Fig. S1d,f,h). Neurons could also show a mixture of these two response types (Fig. S1b,g).

We first determined to what extent the rules modulated touch-evoked activity by comparing tactile correct trials between respond-to-touch and respond-to-light blocks (tHit vs tCR). We restricted analysis to the 150 ms period of stimulus delivery, to focus on rule-dependent processing that was not influenced by overt movements (97% of lick onsets occurred >150 ms after stimulus onset). We used ideal observer analysis to determine how well trial-by-trial activity of a single neuron could discriminate between the task rules. We found that 8-16% of neurons in these cortical areas had area under the receiver-operating curve (AUC) values significantly different from 0.5, and therefore discriminated between tHit and tCR trials (Fig. 1h; S1: 11.9%, n = 177; S2: 8%, n = 162; MM: 16.4%, n = 140; ALM: 12.1%, n = 256; criterion to be considered significant: Bonferroni corrected 95% CI on AUC did not include 0.5). Moreover, the distribution of AUC values for discriminative neurons showed that the majority had enhanced responses (AUC > 0.5) to tactile stimuli that were behaviorally relevant. Therefore, touch-evoked responses were overall enhanced by the relevance of tactile stimuli according to the current stimulus-response rule.

In our cross-modal selection task, rules were block-based and there were no cues to indicate the current rule (except for the rule transition cue on the 9^th trial of each block), so the mice were required to maintain rule information during the inter-trial interval (ITI). To test if ITI activity of neurons in the somatosensory and motor cortical areas reflected task rules, we first analyzed neural activity during the one second window preceding stimulus delivery. Trials with licking in this time window were removed to minimize possible movement effects on pre-stimulus activity (Methods). We found that some cortical neurons showed obvious changes in their pre-stimulus activity across blocks (Fig. 2a). A preference for the respond-to-touch rule (that is, with activity higher in respond-to-touch blocks compared with respond-to-light blocks) and a preference for the respond-to-light rule were both observed (Fig. 2a,b). We next calculated discriminability between block types for each neuron to see how well an ideal observer could categorize the current trial’s task rule on the basis of pre-stimulus activity (−100 to 0 ms from stimulus onset). To ensure mice were in the correct state to treat the following stimuli according to the rule, only correct tactile and visual trials were included. Less than 5% of neurons in S1 and S2 showed significant rule discriminability, while MM and ALM had around 10-20% significant neurons (Fig. 2c; S1: 4.5%, n= 177; S2: 2.5%, n= 162; MM: 21.4%, n= 140; ALM: 10.2%, n= 256; Bonferroni corrected 95% CI on AUC did not include 0.5).

We also calculated the ability of each neuron to discriminate between tactile vs visual stimuli in windows before (−100 to 0 ms) or after (0 to 100 ms) stimulus onset, to serve as negative and positive controls, respectively, for our use of ideal observer analysis. As expected, single-unit activity before stimulus onset did not discriminate between tactile and visual trials (Fig. 2d; S1: 0%; S2: 0%; MM: 0%; ALM: 0.8%). After stimulus onset, more than 35% of neurons in the sensory cortical areas and approximately 15% of neurons in the motor cortical areas showed significant stimulus discriminability (Fig. 2e; S1: 37.3%; S2: 35.2%; MM: 15%; ALM: 14.1%).

The rule dependence of pre-stimulus activity indicates that neurons were in different states immediately prior to stimulus onset, suggesting that their responses to the subsequent sensory input might also differ. We therefore next investigated early touch-evoked responses (0 to 50 ms from stimulus onset) in those neurons that showed significant rule discriminability prior to stimulus delivery. We found that neurons with stronger preference for the respond-to-touch rule before stimulus onset showed larger touch-evoked activity in the respond-to-touch blocks, whereas neurons with a preference for the respond-to-light rule showed larger touch-evoked activity in the respond-to-light blocks (Fig. 2f; Pearson correlation; S1: r = 0.69, p = 0.056, 8 neurons; S2: r = 0.91, p = 0.093, 4 neurons; MM: r = 0.93, p < 0.001, 30 neurons; ALM: r = 0.83, p < 0.001, 26 neurons). This result suggests that configuration of pre-stimulus states could play a role in achieving rule dependence of sensory processing.

Together, these results demonstrate that the pre-stimulus activity of single units in the sensory and motor cortical areas reflected task rules, with better discrimination of rules by units in the motor areas.

We next investigated the function of rule-dependent pre-stimulus activity from the perspective of neural population dynamics. In a dynamical system, state variables change based on their current values. This implies that the evolution of neural population activity during a trial will depend in part on the activity state of the population at the beginning of the trial^36,40. We hypothesized that pre-stimulus activity—i.e. the state of the population at the start of the trial—would be set in such a way as to enable processing of the upcoming sensory signals according to the appropriate rule. Predictions from this hypothesis are that: (1) the pre-stimulus state of a neural population could be used to decode the current task rule; (2) the rule-dependent separation of neural subspaces before and after the tactile stimulus onset should be correlated; (3) pre-stimulus states would shift when the mice switched between rules; and (4) perturbations of the pre-stimulus state should disrupt task performance. We tested each prediction via the following analyses and experiments.

We first determined whether the pre-stimulus population state could be used to decode the current task rule. For each session, we used linear discriminant analysis (LDA) to obtain a classification accuracy for block type (respond-to-touch vs respond-to-light) based on the pre-stimulus activity (−100 to 0 ms relative to stimulus onset) of simultaneously recorded neurons on correct trials (Fig. 3a). Pre-stimulus states in S1, S2, MM and ALM could each be used to decode the block type (Fig. 3b; medians of classification accuracy [true vs shuffled]: S1 [0.61 vs 0.5], 10 sessions; S2 [0.62 vs 0.53], 8 sessions; MM [0.7 vs 0.52], 9 sessions; ALM [0.68 vs 0.55], 13 sessions). Support vector machine (SVM) and Random Forest classifiers showed similar decoding abilities (Fig. S2a,b; medians of classification accuracy [true vs shuffled]; SVM: S1 [0.6 vs 0.53], S2 [0.61 vs 0.51], MM [0.71 vs 0.51], ALM [0.65 vs 0.52]; Random Forests: S1 [0.59 vs 0.52], S2 [0.6 vs 0.52], MM [0.65 vs 0.49], ALM [0.7 vs 0.5]). Interestingly, activity in the motor cortical areas allowed robust block-type classification (no overlap between bootstrap 95% CIs for the true and shuffled data, 95% CIs for true vs shuffled data: MM [0.60,0.73] vs [0.48, 0.53]; ALM [0.56,0.66] vs [0.49, 0.54]). In contrast, the sensory cortical areas showed limited block-type discriminability (Fig. 3c; bootstrap 95% CIs for the true data were above 0.5 but overlapped with the bootstrap 95% CIs for the shuffled data, 95% CIs for true vs shuffled data: S1 [0.52,0.61] vs [0.49,0.55]; S2 [0.53,0.63] vs [0.49, 0.54]). In positive and negative control analyses, we found that neural population activity could be used to discriminate between stimulus types (tactile vs visual) after stimulus onset (Fig. 3e and Fig. S2d; 0 to 100 ms relative to stimulus onset) but not before stimulus onset (Fig. 3d and Fig. S2c; from −100 to 0 ms).

Together these results show that each of the two task rules was associated with a different pattern of pre-stimulus population activity in sensory and motor cortical areas, with a larger difference in the motor areas.

We found pre-stimulus population activity was rule-dependent across sensory and motor cortical areas. We next asked if these rule-dependent pre-stimulus states affect post-stimulus neural activity. We investigated the relationship between the difference in pre-stimulus states between tHits and tCRs and the divergence of their subsequent neural trajectories. To assess this for the four cortical areas, we quantified how the tHit and tCR trajectories diverged from each other by calculating the Euclidean distance between matching time points for all possible pairs of tHit and tCR trajectories for a given session and then averaging these for the session (Fig. 4a,b; window of analysis: −100 to 150 ms relative to stimulus onset; 10 ms bins; using the top 3 PCs; Methods). The resulting time series of distance values from all sessions (n = 40) were then ranked according to their means over the 100 ms period preceding stimulus onset and split into two groups, those above and below the median. The top 50% group showed a larger mean distance between tHit and tCR trajectories after stimulus onset compared with the bottom 50% group^41,42 (Fig. 4c; p < 0.001, permutation test). This result shows that the difference in population responses to the tactile stimuli under the two rules is commensurate with the difference in pre-stimulus states.

Although the full dimensionality of a neural state space is equal to the number of neurons, correlations among neurons typically cause dynamics to occur within a subspace of lower dimensionality^40,43. Population activity associated with each task rule might not only follow distinct trajectories, but could also occur within different neural subspaces. To test this, we calculated the overlap between the subspaces associated with tHit and tCR trials^35,44,45 (Fig. 4d; see Methods). We expect that, if tHit and tCR trial activity occupies largely overlapping subspaces, then the neural dimensions capturing most of the tHit activity will also explain much of the tCR activity (Fig. 4d, middle). Conversely, if they occupy largely distinct subspaces, then the dimensions capturing most of the tHit activity will explain little of the tCR activity (Fig. 4d, right). For each session, tHit trials were divided randomly into equally-sized “reference” and “control” groups. The reference tHit trials were then used to perform a principal component analysis (PCA; using data 0 to 150 ms from stimulus onset; 10 ms bins). We projected activity from tCR trials and from the control tHit trials into the space of the top-three principal components obtained from the reference group PCA, then calculated and normalized their variance explained (Methods; using data 0 to 150 ms from stimulus onset). For S1, S2, MM and ALM, subspace overlaps for tCR trials were significantly lower than the corresponding subspace overlaps for the control tHit trials (Fig. 4e, dark purple vs gray symbols; tCR – control tHit: S1 [−0.23], 8 sessions, p = 0.0016; S2 [−0.23], 7 sessions, p = 0.0086; MM [−0.36], 5 sessions, p = <0.001; ALM [−0.35], 11 sessions, p < 0.001, paired-t test). This finding suggests that different neural subspaces were used for processing tactile stimuli under each of the two rules, in both sensory and motor cortical areas.

We next asked if the rule-dependent separation of subspaces we observed began prior to the tactile stimulus. This would provide evidence that differences in tactile stimulus processing follow from differences in the state of the neural population at the time the stimulus is received. We found that subspace overlaps for tCR trials before stimulus onset (from −100 to 0 ms) were significantly lower than the subspace overlaps for the control tHit trials (Fig. 4e, light purple vs gray symbols; pre-stimulus-onset tCR – control tHit: S1 [−0.28], p = 0.001; S2 [−0.31], p < 0.001; MM [−0.38], p < 0.001; ALM [−0.33], p < 0.001, paired-t test). Moreover, across the cortical areas, the subspace overlaps for tCR trials calculated from periods before and after the tactile stimulus onset were correlated (Fig. 4f; Pearson correlation, r = 0.68, p < 0.001; 31 sessions). This indicates that the shift between neural subspaces associated with each rule occurred prior to stimulus delivery and should thus impact processing of the stimulus.

Together, our results suggest that cortical populations are “pre-configured” according to the current rule, such that an incoming sensory signal leads to distinct processing and ultimately distinct actions.

Gating of sensory information involves changing how sensory information is represented and read out³². This can be achieved by shifting sensory and/or choice coding dimensions in the population activity space^34,46. In the previous section, we showed that neural subspaces containing trial dynamics changed between the two task rules. We next asked whether stimulus and choice coding dimensions within these subspaces also shifted with task rules. For each task rule, we estimated coding dimensions (CDs) that maximally discriminated the neural trajectories for different conditions^47,48 (Fig. S3a-d; Methods). Since mice rarely licked the wrong port for a given block (Fig. 1b), right-lick and no-lick trials were used to obtain choice coding dimensions for the respond-to-touch blocks, and left-lick and no-lick trials for the respond-to-light blocks. To assess whether stimulus and choice CDs changed with task rule, we calculated for each session the dot product between the CDs obtained from respond-to-touch and respond-to-light blocks. We then used the magnitude of this dot product as an unsigned measure of the relative orientations of the CDs. In ALM, the dot product magnitudes calculated between stimulus CDs for the two block types were not significantly different from those calculated after shuffling trial-type labels (Fig. S3e; significance defined as non-overlap of 95% CIs). This suggests that the stimulus CD in a respond-to-touch block had an orientation unrelated to that in a respond-to-light block. In contrast, we found that S1, S2 and MM had stimulus CDs that were significantly aligned between the two block types (Fig. S3e; magnitude of dot product between the respond-to-touch stimulus CDs and the respond-to-light stimulus CDs, mean ± 95% CI for true vs shuffled data: S1: 0.5 ± [0.34, 0.66] vs 0.21 ± [0.12, 0.34]; S2: 0.62 ± [0.43, 0.78] vs 0.22 ± [0.13, 0.31]; MM: 0.48 ± [0.38, 0.59] vs 0.24 ± [0.16, 0.33]; ALM: 0.33 ± [0.2, 0.47] vs 0.21 ± [0.13, 0.31]). In contrast, the choice CDs for the two block types were not aligned well in S1, S2, MM, or ALM (Fig. S3f; magnitude of dot product between the respond-to-touch choice CD and the respond-to-light choice CD, mean ± 95% CI for true vs shuffled data: S1: 0.28 ± [0.15, 0.43] vs 0.21 ± [0.12, 0.33]; S2: 0.22 ± [0.11, 0.33] vs 0.21 ± [0.13, 0.32]; MM: 0.22 ± [0.13, 0.33] vs 0.22 ± [0.14, 0.3]; ALM: 0.27 ± [0.16, 0.39] vs 0.21 ± [0.13, 0.31]). These results suggest that the different subspaces for tactile processing under the two rules result at least in part from changes in choice coding dimensions.

Task rules switched multiple (3-5) times in each behavioral session. Mice detected a rule switch either through trial and error during the first few trials after the switch, or when a drop of water from the correct reward port was given on the 9th trial (which served as a cue to ensure that mice switched by this point; Fig. 1b). We used the first hit trial as the mark of a successful behavioral switch and found that mice switched before or immediately after the cue (Fig. 5a). To analyze how pre-stimulus states changed over the course of a rule transition, we defined a “transition period” that spanned from the first trial after a block switch until the first hit trial of the new block. In addition, we divided each transition period into “early” and “late” parts based on the occurrence of the first false alarm trial of the new block (Fig. 5b). We considered the first false alarm trial to be the point at which the mouse first received feedback to indicate that a block change had occurred.

We hypothesized that the behavioral change that marked a successful switch in rule application would be accompanied by a neural change. Specifically, that the pre-stimulus population state would progress from that typical of the prior type of block to that typical of the new type of block in parallel with the behavioral shift. To test this, we trained an LDA classifier to discriminate respond-to-touch block and respond-to-light block trials using pre-stimulus neural activity. We used 90% of the correct trials as training data and tested classifier performance on the held-out 10% of correct trials (Fig. 5c,d). Using our “transition period” definition as described above, we tested classifier performance on “early” transition and “late” transition trials (Fig. 5b). For respond-to-touch to respond-to-light block transitions, the fractions of trials classified as respond-to-touch for MM and ALM decreased progressively over the course of the transition (Fig. 5c; rank correlation of the fractions calculated for each of the separate periods spanning the transition, Kendall’s tau, mean ± 95% CI: MM: −0.39 ± [−0.67, −0.11], ALM: −0.29 ± [−0.54, −0.04]). Similarly, the fractions of trials classified as respond-to-touch increased progressively over the course of transitions from respond-to-light to respond-to-touch blocks (Fig. 5d; Kendall’s tau, mean ± 95% CI: MM: 0.37 ± [0.07, 0.63], ALM: 0.27 ± [0.03, 0.49]). Accuracies for classification of trials by block type based on S1 and S2 activity were uniformly poor and showed no clear trends across block transitions (Fig. S4; mean ± 95% CI on Kendall’s tau for respond-to-touch → respond-to-light transitions: S1: −0.16 ± [−0.43, 0.1], S2: −0.15 ± [−0.54, 0.21]; for respond-to-light → respond-to-touch transitions: S1: 0.21 ± [−0.07, 0.5], S2: 0.25 ± [−0.17, 0.63]). Together, these results indicate that the pre-stimulus states of neural populations in MM and ALM shifted over the course of block transitions in a manner commensurate with the behavioral shift in rule application.

So far, we have shown that pre-stimulus neural population states in the sensory and motor cortical areas differed between the two task rules. This suggests that pre-stimulus activity may play a critical role in our task. To test this, we bilaterally inhibited the different cortical areas shortly before stimulus onset via optogenetic activation of parvalbumin-positive (PV) GABAergic neurons^18,49 (Fig. 6a, middle panel; −0.8 to 0 seconds from stimulus onset). Additionally, we included two other areas of the dorsal cortex, the anteromedial part of the motor cortex (AMM) and the posterior parietal cortex (PPC). We also performed negative control (sham) sessions that were identical except that the optogenetic light path was obstructed.

We defined detection sensitivity for tactile stimuli as the difference between tactile hit rate and visual false alarm rate during the response-to-touch blocks. Tactile detection sensitivity was significantly decreased when MM and ALM but not S1 and S2 were inhibited during the pre-stimulus period (Fig. 6c; criterion to be considered significant: 95% CI on Δ tactile sensitivity did not include 0; S1/S2: [−0.47, 0.08], 4 mice, 10 sessions; MM: [−0.65, −0.12], 4 mice, 11 sessions; ALM: [−0.68, −0.02], 4 mice, 10 sessions). This was primarily due to a reduction in the tactile hit rate (Fig. S5b; 95% CI on Δ tactile hit rate: S1/S2 [−0.31, 0.01]; MM [−0.49, −0.15]; ALM [−0.46, −0.07]). Inhibition of S1, S2, MM, and ALM during a 2-second window starting at the time of the stimulus onset served as a positive control (Fig. 6a, right panel). Consistent with previous studies^18,50, inhibition of these cortical areas after stimulus onset reduced detection sensitivity for tactile stimuli (Fig. 6d; 95% CI on Δ tactile sensitivity: S1/S2 [−0.67, −0.08]; MM [−0.6, −0.34]; ALM [−0.55, −0.27]). Neither the negative control (sham) condition, nor inhibition of AMM before stimulus onset, showed an effect on the detection sensitivity for tactile stimuli (Fig. 6c; sham: [−0.12, 0.13], 7 mice, 28 sessions; AMM [−0.19, 0.06], 7 mice, 21 sessions). Together, our results suggest that the pre-stimulus network activity states in MM and ALM play an important role in rule-dependent tactile processing.

We defined detection sensitivity for visual stimuli as the difference between visual hit rate and tactile false alarm rate during the respond-to-light blocks. Visual detection sensitivity was not affected when S1 and S2 were inhibited before (Fig. S6g; 95% CI on Δ visual sensitivity: [−0.27, 0.05]) or after (Fig. S6h; [−0.44, 0.04]) the stimulus onset. For pre-stimulus-onset inhibition of the motor cortical areas, visual detection sensitivity was decreased when MM and AMM but not ALM were suppressed (Fig. S6g; MM [−0.6, −0.18]; ALM [−0.5, 0.04]; AMM [−0.35, −0.08]). Inhibition of any of the three motor cortical areas after stimulus onset caused a reduction in visual sensitivity (Fig. S6h; MM [−0.83., −0.52]; ALM [−0.84, −0.39]; AMM [−0.72, −0.36]). Together, these results indicate that pre-stimulus states of neural populations in the medial parts of motor cortex, such as MM and AMM, are critical for processing visual stimuli in a rule-dependent manner.

PPC is involved in multisensory processing^51–53 and decision-making^54–58. Here we tested for a critical role of PPC in our cross-modal selection task (7 mice, 21 sessions total). Detection sensitivities for both tactile and visual stimuli were decreased when PPC was inhibited before (Fig. 6c; Δ tactile sensitivity: [−0.4, −0.06]; Fig. S6g; Δ visual sensitivity [−0.31, −0.05]) or after the stimulus onset (Fig. 6d; Δ tactile sensitivity: [−0.43, −0.04]; Fig. S6h; Δ visual sensitivity: [−0.54, −0.07]). These reductions in tactile and visual detection sensitivities were caused by a decrease in hit rate and/or an increase in false alarm rate. In general, inhibition of PPC before stimulus onset decreased the hit rate (Fig. S5b; Δ tactile hit [−0.36, −0.05]; Fig. S6b; Δ visual hit [−0.26, 0]), whereas inhibition of PPC after stimulus onset increased the false alarm rate (Fig. S5f; Δ visual false alarm [0.13, 0.38]; Fig. S6f; Δ tactile false alarm [0, 0.25]).

It is possible that disruption of pre-stimulus states may affect aspects of tactile sensory processing and/or lick production that are unrelated to rules. To exclude this possibility, in a new cohort of mice (n = 5), we inhibited each cortical area in either the pre- or the post-stimulus-onset period during performance of a simple tactile detection task (Fig. 6e-g). In this task, mice had only to report with Go/NoGo licking whether a whisker was deflected, without rule-switching components to the task or the need to suppress responses to distracting stimuli. We found that tactile sensitivity was decreased when the sensory and motor cortical areas were inhibited after but not before stimulus onset (Fig. 6f,g; 95% CI on Δ tactile sensitivity for pre-stimulus-onset inhibition: S1/S2 [−0.34, 0.18], 3 mice, 3 sessions; MM [−0.23, 0.06], 3 mice, 5 sessions; ALM [−0.28, 0.07], 3 mice, 4 sessions; Δ tactile sensitivity for post-stimulus-onset inhibition: S1/S2 [−0.59, −0.31]; MM [−0.59, −0.25]; ALM [−0.45, −0.1]). This was mainly caused by the decrease in hit rate in the post-stimulus-onset inhibition (Fig. S5i; 95% CI on Δ tactile hit rate: S1/S2 [−0.61, −0.28]; MM [−0.67, −0.2]; ALM [−0.73, −0.05]). Inhibition of AMM and PPC did not influence tactile sensitivity in either inhibition condition (Fig. 6f,g; Δ tactile sensitivity for pre-stimulus-onset inhibition: AMM [−0.25, 0.16], 3 mice, 4 sessions; PPC [−0.12, 0.1], 3 mice, 4 sessions; Δ tactile sensitivity for post-stimulus-onset inhibition: AMM [−0.17, 0.18]; PPC [−0.14, 0.3]). Together, these results show that inhibition immediately prior to stimulus delivery did not impact the performance of a simple tactile detection task in which the stimulus-response rule remained fixed.

We conducted an additional analysis to rule out the possibility that the behavioral effects of cortical inhibition we observed could be due simply to a deficit in lick production per se. Specifically, we compared the probability of licking in laser-only trials (catch trials where there was no sensory stimulus) with the probability of licking during intertrial intervals, for both the cross-modal selection task and the simple tactile detection task (Fig. 6h). Lick probability was unaffected during S1, S2, MM and ALM experiments for both tasks, indicating that the behavioral effects were not due to an inability to lick (Fig. 6i, j; 95% CI on Δ lick probability for cross-modal selection task: S1/S2 [−0.18, 0.24]; MM [−0.31, 0.03]; ALM [−0.24, 0.16]; Δ lick probability for simple tactile detection task: S1/S2 [−0.13, 0.31]; MM [−0.06, 0.45]; ALM [−0.18, 0.34]).

Taken together, our results suggest that the pre-stimulus states of motor cortical networks play an important role in rule-dependent sensorimotor transformations.