Figure 4. Distinct state-space positions at Go cue predict distinct outcomes.

(A) Alternative concepts for proactive inhibition, illustrated using a simplified rise-to-threshold framework (Brown and Heathcote, 2008; Noorani and Carpenter, 2016; Verbruggen and Logan, 2009). (B). Comparison of GP population state between Maybe-Stop-Contra trials (including both contra- and ipsi-instructing Go cues and Stop trials) and No-Stop trials (± 100ms around Go cue; same state-space as Figure 3). Filled circles indicate epochs of significant Euclidean distance between two trajectories (permutation test on each 4 ms time bin, p<0.05). (C). Permutation tests (same format as Figure 3). Just before the Go cue (-100-0ms) the Maybe-Stop state was significantly shifted away from action initiation, and in the ipsi direction. (D) Breakdown of GP state for trials with contra Go cues, by distinct trial outcomes. (E) Quantification of D, comparing evolution of activity along Initiation and Selection Axes on correct contra trials (blue), incorrect action selections (light green) and RT limit errors (brown; failure to initiate movement within 800ms). Thicker lines indicate epochs of significant difference to the Correct trajectory (permutation test on each 4 ms time bin, p<0.05).

Figure 4—figure supplement 1. Neural population results for individual rats, and corresponding behavior.

(A) Comparing proactive shifts along Initiation and Selection Axes for all rats together (left) and for individual rats. Rats 2,4 and 6 were grouped together as they had fewer recorded neurons. In left plots, thicker lines indicate epochs of significant difference between two conditions (permutation test on each 4 ms time bin, p<0.05). Note that Rat three had the largest Selection Axis bias towards ipsiversive movements before the Go cue (and a bias towards movement on the Initiation Axis). (B) RT results for the same animal groupings. In all cases there was a greater slowing of contra than ipsi movements, consistent with a selective proactive inhibition effect. However, Rat three showed a speeding of ipsi movements compared to the No-Stop condition, consistent with an ipsiversive bias and no overall movement inhibition.

Figure 4—figure supplement 2. Trial-history dependence.

(A) (Left) On Maybe-Stop trials that followed Stop trials (‘After-Stop’), rats were more likely to succeed in stopping (Wilcoxon signed rank test, z = 2.67, p=0.008) and showed increased RT (Wilcoxon signed rank test, z = 4.46, p=8.02 × 10⁻⁶), compared to trials that followed Go trials. (Right) On No-Stop trials that followed error trials (‘After-Error’), rats were more likely to make RT limit errors (Wilcoxon signed rank test, z = 3.03, p=0.002) and showed increased RT (Wilcoxon signed rank test, z = 5.42, p=5.95 × 10⁻⁸). (B) Corresponding apparent shifts along the Initiation Axis did not reach significance (permutation tests, analysis epoch: −100–0 ms before Go cue).