Notably, the neuron was not active in trials with right-side (0°) or left-side (180°) cues, but only for those two directions (up and down) that were

equally probable to instruct a downward motor goal. The bimodal response profile of the example neuron in Figure 3B in the PMG task matched the prediction of the goal-selection hypothesis, and contradicts the rule-selection hypothesis. The bimodal profile mimicked the response pattern one would expect when averaging (not summing) the two response profiles in the DMG task. This means, the response pattern during planning of two equipotent alternative potential motor goals was an equally weighted linear combination of the response patterns during unambiguous planning of the two respective unique motor goals. In a model-based analysis we quantitatively confirmed this view (see Figures S1 and S4 available online). Bimodal selectivity this website profiles dominated the balanced data set in PRR. The average population activity in the balanced data set shows two stable ridges of activity during the memory period (Figure 3C). Since the cue-position axis marks the location of the spatial cue relative to the preferred direction (PD) of each neuron (as measured in the DMG task), the two ridges indicate that on the population level the direct and inferred goals are represented simultaneously during ambiguous reach planning. For quantitative analysis we characterized the bimodal

Ku0059436 versus unimodal selectivity of each neuron with a direction modality contrast (DMC). Positive DMC indices indicate selectivity for the direct motor goal; negative values indicate selectivity for the inferred motor

goal. Indices close to zero indicate symmetric bimodal tuning (not lack of tuning) since only directionally selective neurons were considered (see Experimental Procedures). The mean DMC of the balanced data set did not significantly deviate from zero (m = 0.001; standard error of the mean [SEM] = 0.021, p > 0.05), indicating that in the balanced data set most neurons had bimodal selectivity profiles ( Figure 3C, TCL inset). The existence of a bimodal neural selectivity pattern in the balanced data set is not sufficient to demonstrate potential motor goal encoding. The monkeys could have preliminarily selected one of the two potential motor goals during the memory period in every trial, and randomly switched their selection from trial to trial. Such switching would be obscured in PMG-CI trials due to the explicit context instruction at the time of the GO cue. The bimodal selectivity pattern revealed by the above analyses would denote an artifact of averaging across inhomogeneous sets of trials in this case (Figure 4A, bottom). With a choice-selective analysis of the free-choice (PMG-NC) trials we can rule out this possibility. We can instead show that both potential motor goals were encoded independently of the monkey’s later choices (Figure 4A, top).