Figure 4. Correlation between representations of the decision variable by leader and supporter neurons.

The analysis examines the within-trial correlation between the representation of the decision variable by pairs of leader and supporter neurons in epochs bracketing the IEM. The correlation is between the conditional expectations of the spike counts (see STAR Methods).

(A) Correlation between firing rates in the 1^st variant of the two-pulse task. Black bars show the correlations when the pair is informative: during the transfer from leader to supporter neurons, before and after the saccade to T⁰, and from supporter back to the same leader neurons, before and after the completion of smooth-pursuit to the original FP. The informative transfer is shown by the black connector in the cartoon above. Gray bars show the correlation between the same neurons in adjacent uninformative epochs (light gray connectors in the cartoon). The tick and cartoon labels use l and s for leader and supporter, respectively; superscripts indicate the epoch of the sample. Yellow lines and shading show the mean and the two standard deviations of the same correlation statistic under shuffled control (1,000 permutations).

(B) Same analysis applied to the 2^nd variant of the two-pulse task. There is only one informative transfer between the leader and supporter neurons around the one IEM. Same conventions as in (A).

(C) Transfer of the decision variable is mediated between pools of weakly correlated leader (L) and supporter (S) neurons. Each circle represents a pool of neurons (triangles) that share the same response field. For each pool, the filled triangle represents the neuron that is observed (recorded), while the unfilled triangles represent the other neurons that are not recorded in the experiment. The diagram applies to the 1^st variant of the two-pulse task. If the pools contain only one neuron, then perfect transfer of the decision variable from the L-pool to the S-pool and back to the L-pool (blue arrows) might predict no autocorrelation between L^pre and L^post, conditional on S^IEM. However, a single neuron would retain conditional autocorrelation if it were a member of a pool of weakly correlated neurons.