contralateral to ipsilateral hemifield (green), there was a decrease compared with the contralateral location on no-swap trials (desaturated green; green stars: p < 0.01). For average MUA (Figure 4A), there was a complete inversion. The spike rates after a saccade that shifted the remembered location into a given hemifield almost exactly matched those for a static memory in the same hemifield. Around the time of the saccade, there was also increased spiking for both swap conditions, relative to the no-swap trials (S symbols: p < 0.01, shift condition main effect). Nevertheless, later in the delay, the predicted inversion effect was dominant (X symbols: p < 0.01, interaction effect). For decoding accuracy (Figure 4B), ipsilateral-shifting trials (green) were near the value of constant ipsilateral trials (desaturated brown). Contralateral-shifting trials (orange) exhibited a bump of increased accuracy after the saccade (brown stars: p < 0.01, paired t test) but subsequently declined and never attained the level of constant contralateral trials (desaturated green). This imperfect transfer of information may explain why behavioral performance was significantly decreased in swap trials.
