Behavioral states affect eye position tuning in the parietal cortex of macaques Authors: Kostas Hadjidimitrakis, Francesco Edoardo Vaccari, Marina De Vitis, Matteo Filippini, Stefano Diomedi, Patrizia Fattori Published: 2026 ABSTRACT SummaryPosterior parietal neural populations compute flexible yet stable representations of spatial locations using eye position information, and plan effector movements across various goal-directed contexts. How space is represented during internally driven non-goal-directed behaviors and whether these spatial representations are affected by the different behavioral contexts and the underlying brain states remain understudied. We compared eye position signals in single neurons in macaque parietal cortex (area V6A) during constrained fixations toward visual targets with gaze activity during self-paced fixations in darkness. While the firing rates of V6A neurons were comparable in the two conditions, the spatial tuning of gaze activity was stronger during the spontaneous fixations than during task-constrained fixations. Moreover, neural responses were more variable in constrained fixation. Our results confirm the flexibility of spatial representations in parietal cortex and indicate that eye position signals provide spatial information in a wider range of behavioral contexts beyond visual and visuomotor processing.Keywords: parietal cortex, gaze, eye position, visuomotor, spontaneous activity Introduction Although everyday behavior is characterized by goal-directed movements in response to sensory cues or stimuli, several actions are self-generated and not associated with external events, or immediate goals or rewards. Whether the same neural circuits and populations are activated during both actions, and how, remains largely unanswered. The posterior parietal cortex (PPC) is a key region for goal-directed behavior and is involved in the visual processing of target location and sensory guidance of effector movement.1,2,3,4,5 Eye position is crucial for both these functions6,7 and, not surprisingly, gaze signals influence the activity of almost every PPC area.8,9,10,11,12,13,14,15 Eye position tuning of neural activity occurs in PPC even in the absence of visual stimulation, while the gaze is anchored to a small fixation spot in darkness.10,16 In addition, the tuning of neural activity by eye position depends strongly on the sensory context and task demands. For example, PPC neurons change their spatial tuning when fixation targets are presented in different sensory modalities or backgrounds, or serve as a cue for different responses.17,18,19 Eye position is associated with spatial attention, and areas with gaze-related activity are often also modulated by the focus of attention.20,21 The processing of spatial attention involves mainly the dorsal attentional network (DAN22), a widespread network of parietal and frontal cortex areas.23,24,25 Spatial attention increases the visual responses in visual, parietal and prefrontal cortices26,27,28 and is also reported to decrease response variability in various areas both in the frontal and visual cortices.29,30,31 However, the neural correlates and the relationship between eye position and attention beyond the context of goal-directed tasks are largely unknown. Related to these issues is the concept of behavioral states and how cortical processing is affected by varying levels of attention, arousal, expectancy, and other factors while changes in brain state occur.32,33,34 Arousal influences cortical processing via diffuse noradrenergic pathways35 and is considered to exert an additive effect on neural responses without affecting spatial selectivity.22,24 Furthermore, recent advances suggest various effects of arousal and, more generally, of brain states on thalamo-cortical functional interactions, stimulus-induced variability, interneural correlations and spatial tuning both in visual and in associative cingulate cortex.36,37,38,39 Changes in brain states have been also associated with spontaneous body movements often leading to increased sensory responses in visual cortex,40,41,42 but corresponding evidence from parietal cortex is lacking. Here, we addressed the issue of brain states in PPC by testing how the unconstrained fixation behavioral context affected gaze-related activity with respect to constrained fixations toward visual targets occurring in the context of an instructed-delay, goal-directed task. We recorded neural activity from area V6A in the medial PPC43 that processes signals related to gaze direction and distance12,16 and exhibits modulations during a covert attention task.44 We investigated changes in the eye position-related mean firing rate, neural variability, tuning selectivity, and eye position field structure between the two behavioral states. We expected that the different behavioral engagement in the two contexts would affect V6A neural activity, with higher mean firing rates and lower firing variability during constrained fixation due to increased attention and arousal, as several lines of evidence suggest.26,37,45 Contrary to our expectations, we found that most V6A cells showed small changes in mean firing activity in the two contexts, and, in addition, that spontaneous fixations were associated with lower variability and stronger tuning. These findings could be interpreted in a framework that involves the different effects of multiple state-dependent factors in the two task contexts. Results We recorded the activity of 183 single V6A cells (119 from monkey 1 [M1] and 64 from monkey 2 [M2]), during a visually guided fixation task that included two task periods with a different context (Figure 1). During the first period (spontaneous fixation, Figures 1C and 1E: “FREE”), the animals pressed a button next to their chest (Figure 1A) and were free to move their eyes in darkness while waiting for the switching on of the fixation light-emitting diode (LED). In the second period (visually guided fixation, Figures 1C and 1E: “FIX”), they had to hold their gaze on the illuminated LED for a variable period until a “go” signal (change of LED color) cued them to release the button in order to receive reward. Accordingly, the gaze could shift randomly across different spatial positions during the first period while it was constant on the illuminated LED during the second period (FIX) in each trial. Experimental setup, task sequence, eye position space, and analysis (A) Experimental configuration used for the fixation task. During the task, eye movements were performed toward one of the ten light-emitting diodes (LEDs) on a horizontal panel. (B) Top view of the setup showing the contralateral (left) and ipsilateral (right) configurations of the target panel. Dotted rectangles indicate the fixation locations around the LED targets that were used for the analyses. (C) Task sequence and examples of eye movements during a single trial. Top: Task events and horizontal and vertical eye position traces are shown. Long vertical continuous lines indicate the occurrence of events during the task. From left to right, markers show: trial start (home button [HB] press), fixation target appearance (LEDon), and green to red change of the LED (LEDred). White rectangular boxes marked “FREE” or “FIX” indicate the time intervals used for the subsequent analyses of neural activity. Bottom: velocity profiles for the spontaneous and visually guided saccades shown in the top demonstration. (D) Top view of the setup showing the grid of fixation workspace. The nearest target positions were excluded from analysis due to the lack of eye position samples during the FREE interval. (E) Heat maps illustrating the distribution of fixation locations in FREE (left) and FIX (right) from a number of sessions from a single animal. The fixation locations around the nearest LEDs were rarely fixated during FREE and they were not considered in the analyses. See also Figure S2. One hundred thirty seven of the 183 recorded neurons (99 from M1 and 38 from M2) showed an average firing rate higher than 5 sp/s for at least four different targets and relative gaze position modulation during visually guided fixation (ANOVA, p < 0.05) and were included in the subsequent analyses. Saccades were defined as events in which the gaze velocity exceeds 25°/s. Analyzing the gaze behavior simultaneously with the selected neurons, we found that saccades during the FREE epoch were frequent (2.24 and 1.03 /s, M1 and M2, respectively) and quite large (13.6° [3.5 and 24.7°] for M1; 11.5° [5.4 and 18.6°] for M2; median [25th and 50th percentiles]), as expected while the animal is scanning the dark environment with the head fixed and waiting the next fixation point to turn on (Figure S2A). During FIX, the animal was required to maintain steady fixation on the target. The fixation was monitored online with an electronic window (a few degrees of visual angle) set in the eye tracker. Accordingly, only a few saccades occurred (Figure S2B). They were less frequent (1.01 and 0.39/s for M1 and M2, respectively) and of smaller amplitude (1.3° [0 and 5.5°] for M1; 2.2° [0.9 and 6.2°] for M2; median [25th and 50th percentiles]) than those recorded in the former condition. In line with this behavior, the fixations were shorter during FREE (0.25 s [0.13–0.43 s] for M1; 0.45 s [0.2–0.78 s] for M2) than during FIX (0.45 s [0.13 and 0.83 s] for M1; 0.83 s [0.48 and 1.14 s] for M2; median [25th and 50th percentiles]; Figure S1B). To compare the neural activity relative to the same (or the most similar) eye position in different contexts, we excluded from FREE and FIX eye position and neural signals during the saccades and also peri-saccade intervals to avoid confounding elements (from 50 ms before to 100 ms after each saccade). The remaining data were discretized within 6 × 5 (or 6 × 3) grid squares (10 × 10 cm) and the corresponding firing activity was computed (see STAR Methods). The total amount of retained data in FREE and FIX and the overall distribution are shown in Figure 1E, whereas a detailed comparison of data samples and fixation duration in each grid position is provided in Figure S2C. Figure 2 shows examples of the resulting mean firing activity 2D matrices for three V6A neurons during FREE and FIX. Note that in order to create the firing rate matrices without missing values, the parts of space that the monkey did not fixate were filled with interpolated data (see STAR Methods and Figure S3 for illustration of data in the example neurons before interpolation). In addition, Figure 2 shows the vector fields (Figure 2, arrows that point from each cell grid to the direction of maximum firing rate) that allow a more direct visualization of spatial tuning. The first neuron (Figure 2A) was tested only during fixations in the contralateral (left) and central spaces. In FIX, its activity is overall higher than in FREE, both considering the raw data (mean ± std firing rate FREE: 15.3 ± 14.0 sp/sec; FIX: 21.2 ± 16.9 sp/sec; Wilcoxon rank-sum test, p = 2.2 × 10−12 ; total recorded samples: FREE, 474; FIX, 2,654) and pooling/averaging to compute the tuning maps (mean firing rate FREE vs. FIX: 14.8 vs. 22.8 sp/sec; delta statistically significant: one-sample, two-tailed t test, t-value = 33.7, df = 99, given 100 bootstrap samples, delta std = 2.1, p = 4.7 × 10−56, see STAR Methods for details), and its modulation by gaze position is similar to that in FREE, showing stronger responses for extrapersonal (beyond reach) and contralateral positions. The second neuron (Figure 2B) was tested in the contralateral, central, and ipsilateral spaces. The raw distributions of the collected firing rates were not significantly different (mean ± std firing rate FREE: 15.4 ± 15.7 sp/sec; FIX: 14.5 ± 15.9 sp/sec; Wilcoxon rank-sum test, p = 0.17; total recorded samples: FREE, 485; FIX, 3,765). However, when pooling and averaging data relative to the same fixations location to obtain the tuning maps (and thus intrinsically correcting for slight biases in the gazed positions), the neural activity resulted higher in FIX than in FREE (mean firing rate FREE/FIX: 12.5/13.7 sp/sec; delta statistically significant: one-sample, two-tailed t test, t-value = 11.57, df = 99, delta std = 2.19, p = 4.28 × 10−20 ). The unit displayed a higher activity for intermediate distance locations in both FREE and FIX. The third neuron (Figure 2C) showed a more complex pattern of spatial relationship across the two periods (see also vector correlation section of results). The average activity calculated on the raw data was slightly higher in FIX period (mean ± std firing rate FREE: 10.6 ± 15.6 sp/sec; FIX: 13.5 ± 19.4 sp/sec; Wilcoxon rank-sum test, p = 3.5 × 10−4; total recorded samples: FREE, 625; FIX, 3,797). However, when pooling and averaging for gaze locations, the two activity distributions were not statistically different anymore (data: mean firing rate FREE/FIX: 14.0/14.7 sp/sec; delta not statistically significant: one-sample, two-tailed t test, t-value = 1.97, df = 99, delta std = 2.21, p = 0.0520). Example V6A neurons with tuning of fixation activity in the two contexts (A) Neuron tested only in the central and contralateral space. The two fields were very similar, with higher firing rates (FRs) in FIX for the same gaze position. The field vectors converged on the peak of activity toward the extrapersonal space (white and yellow squares). (B) Neuron tested while the monkey looked in contralateral, central, and ipsilateral space that showed high activity for intermediate and far positions. (C) Neuron tested for positions covering contralateral, central, and ipsilateral space and showing a complex relationship of responses between FREE and FIX. Three example neurons with response fields (colormaps) and superimposed vector fields (arrows) based on mean FRs during the FREE and FIX epochs. Response fields are plotted as a function of gaze direction and distance in an X, Z Cartesian system of coordinates centered on the animal forehead, between the eyes. The arrows (i.e., the vector field) indicate for each spatial location the neural response gradient (that is the direction along which the FR increases). Data are represented as mean. See also Figures S3 and S4. Comparison of mean activity and firing variability in the two behavioral contexts The example neurons of Figure 2 had complex patterns of firing rate and no general trend can be clearly discerned as to whether the overall neural activity changed from FREE to FIX. Overall, 90 (31) for M1 (M2) units displayed firing rates significantly different in FIX than in FREE (one sample, two-tailed t test, p < 0.05 over 100 bootstraps, see STAR Methods). Specifically, 57 (18) for M1 (M2) units were less active in FIX than in FREE, whereas for 33 (13) units it was the opposite, with the firing rate higher in FIX than in FREE. For the remaining 9 (7) units, activity was not different (p > 0.05). Despite the statistical significance, the differences in firing rates were quite small: most neurons in both monkeys had a difference of less than 5 sp/sec (67% and 79% for M1 and M2, respectively). Indeed, at the population level we found a unimodal distribution with a mean centered on 0 sp/sec (not significantly different from 0 for M2, one sample t test, p = 0.522; significantly different from 0 for M1, t test with p = 0.012, but with a median difference of −1.18 sp/s; Figure 3, left). When the mean firing difference between FIX and FREE was expressed as a percentage of the mean activity during FIX (Figure 3, right), although the overall distributions were still centered around 0, we found that for M1 the median difference was −10.2% (significantly different from 0, t test with p = 5 × 10−4), whereas the median difference was only −2.7% for M2 (not significant, t test p = 0.270). There was a significant number of neurons with gain modulations larger than ±30% of FIX activity, especially in M1 (Figures 3B and 3D; M1: 46/99, M2: 10/38). In most cases (33 and 7 for M1 and M2, respectively; Figures 3B and 3D), the activity during FIX decreased by at least 30%, whereas in the remaining (13 and 3 for M1 and M2, respectively), it increased compared to that in FREE. In sum, while the mean firing rate distributions were not significantly different between FREE and FIX or just slightly different in absolute terms, proportional changes in activity were present in approximately one-third of the cells in the population. Mean firing activity differences in the FREE and FIX task phases (A) Histograms showing the distribution of the difference (delta) in mean firing rate between the FREE and FIX epochs in the two monkeys. (B–D) Histograms showing the distribution of the difference (delta) in mean firing rate expressed as percentage of the mean FIX activity in the two monkeys. All distributions were not significantly different (one sample t test, p > 0.05) from zero. Dashed lines indicate median value. Data are represented as mean. We subsequently looked for possible changes in neural response variability. In fact, previous work showed that the firing variability between trials decreased during the presentation of the visual stimulus, an effect that suggested the influence of attention.29,30,31 We measured variability in the firing rate during FREE and FIX using the Fano factor (FF), which was computed as the variance in spike counts divided by the mean spike count for each grid position. Under the assumption that spike counts are Poisson distributed, the FF is expected to be equal to 1 (variance = mean). Even if the firing rate distributions for FREE and FIX were very similar (see previous section), in order to avoid biases related to slightly different distributions, we adopted a partially random subsampling procedure to match the distributions by recalculating the FF (see STAR Methods), and then computed the mean-matched FF. The FF is represented by the slope of the fitted line in Figure 4 (see also STAR Methods). We found that, for M1 (Figure 4A), the distribution-matched FF in FREE/FIX was 1.37/1.48, with the difference being statistically significant (Z-test, p = 0.035). Similarly, for M2, the difference between the matched-FF during FREE was 1.00 and 1.23 during FIX; the result is significant (Z-test, p = 0.031; Figure 4B). In sum, visually guided fixation did not result in a reduction in neural variability compared to spontaneous fixation but appeared to enhance variability in both neural populations. Firing activity variability in the two task contexts (A and B) Firing variability in the two task contexts measured as mean-matched Fano factor (FF). In each image, the FF was computed based on regressing the spike-count variance against the mean for FREE (left scatterplot) or FIX (right scatterplot). Each point represents neural data for one neuron per cell grid (i.e., position in space) combination. The red line represents the slope of the regression (FF). (A) and (B) show results from M1 and M2, respectively. Analysis of spatial selectivity and modulation strength Our findings so far show that the different fixation contexts caused some changes in the statistics of the neural activity. Subsequently, we investigated the selectivity and strength of eye position tuning in FREE and FIX and compared them between the two task contexts. To do so, we computed a pair of tuning indexes, the preference index (PI), and the depth of PI (dPI), as detailed in STAR Methods. Briefly, the PI is a measure of the selectivity of the response that takes into account the magnitude of the neuronal activity for each eye position. A value near 0 indicates the same response for all positions, whereas a value near 1 indicates activity only for one position. The dPI, which also ranges from 0 to 1, measures the tuning strength by comparing the activity between the position that evokes the highest response (preferred) and the one that elicits the lowest (non-preferred position). Values around 0 indicate no difference between the preferred vs. non-preferred positions; values around 0.3 indicate that the preferred position elicits double the activity of the non-preferred position; values around 0.6 indicate the quadruple, and so on. To compare the values of the tuning indexes between the two contexts, we first plotted them as scatterplots, calculating also the confidence intervals with bootstrapping, useful also for statistical testing (Figure S5). To evaluate how these measures of spatial tuning varied from FREE to FIX in individual cells, we calculated the difference between FIX and FREE for each index and for every neuron. A resulting value of zero would indicate that the spatial selectivity (PI) and/or tuning strength (dPI) of a neuron were not affected by task context, whereas positive (negative) values would indicate stronger effects in FIX (FREE). The results of this analysis are shown in Figure 5. In both monkeys, the distributions of the differences between FIX and FREE for both indices were shifted toward negative values, suggesting stronger modulations in FREE than in FIX. In particular, the median PI difference was significantly (Wilcoxon rank-sum test, p < 0.05) lower than zero for both monkeys (Figure 5, left, M1/M2: median ΔPI(FIX−FREE) = −0.115/−0.081). Regarding the dPI, we observed an even stronger bias toward negative values (Figure 5, right, M1/M2: median ΔdPI(FIX−FREE) = −0.109/−0.109). In summary, the analysis of the tuning selectivity (PI) and modulation strength (dPI) revealed that eye position effects were significantly stronger during non-goal directed, spontaneous fixation than in the visually guided fixation condition. This result paralleled the FF analysis shown in Figure 4, which revealed significantly reduced firing rate variability during spontaneous fixation in both monkeys. Tuning selectivity (PI) and tuning strength (dPI) indexes in the two behavioral contexts: FIX and FREE (A and B) Histograms showing the distribution of the difference of preference index (ΔPI(FIX-FREE), left), and depth of PI (ΔPI(FIX-FREE), right) between the FIX and FREE periods in neurons from monkeys 1 (M1, upper histograms) and 2 (M2, lower histograms). Vertical dashed lines correspond to median values; vertical continuous lines correspond to 0. All distributions were unimodal with a bias toward negative values, suggesting higher selectivity and tuning strength during spontaneous than during visually-guided fixation. See also Figure S5. Spatial relationships of the eye position fields in the two behavioral contexts All the analyses presented so far do not consider the overall two-dimensional spatial arrangement of the eye positions tested, that is, the fact that single neurons in our study were recorded while monkeys looked at various points with precise spatial relationships (Figures 1A, 1B, and 1D). To examine this aspect, we conducted two correlation analyses. First, for each neuron we correlated mean neural activity during FREE and FIX for the same grid locations, discarding those positions for which we had no fixation data (Figure 6A). We also performed this simple correlation analysis after interpolating the missing values in the spatial grids (Figure 6B). Second, we performed a vector correlation analysis, which has been previously used to study reference frame representations in the parietal cortex.46,47,48 The vector correlation method was used to characterize the spatial relationship of the 2D eye position fields between the two contexts. Being two-dimensional and more flexible, it allowed both for rotation (positive correlation) and for reflection (negative correlation) of the 2D response fields along a certain axis in order to test for possible spatial relationships. Differently, direct scalar correlation methods are usually applied in the case of one-dimensional data. Note that to obtain matrices for calculating the vector fields (first necessary step for the vector correlation analysis) we had to fill in the missing values through interpolation (see STAR Methods). In order to avoid the generation of artifacts during interpolation, we selected only the neurons for which we had sufficient data during the FREE epoch. To this end, we developed a metric (the damage index, or ‘DAM index’) that ranges from 0 to 1. This metric accounts for the amount of missing data and their spatial distribution, penalising larger concentrations of missing data in specific locations (see STAR Methods). Accordingly, we retained 87 neurons from M1 and 33 from M2 with a DAM index <0.40. Filling the empty matrix elements with the average obtained from the neighboring cells as an alternative method to interpolation provided almost identical results. Both the simple Pearson’s correlation and the vector correlation analyses were applied to the selected units to check the consistency of the results (see aforementioned section). Pearson’s correlation and vector correlation analysis (A and B) Histograms showing the distribution of t Pearson’s correlation coefficient ρ before (A) calculated on the raw neural activity ignoring missing entries (A) or after filling the spatial tuning maps by linear interpolation (B) for the population of neurons modulated in M1 (left) and M2 (right). (C–E) Vector correlation coefficient (C), scale factor β (D), and phase angle θ (E) between the 2D matrices of the responses in FIX versus FREE periods The majority of V6A modulated neurons showed medium range coefficients and scale factors less than 1, indicating significant activity correlations between the two contexts and stronger response modulations in the FREE context. In A–C, the blue and red part of the stacked histograms shows significant (p-value < 0.05) and non-significant (p-value >= 0.05 correlations, respectively. Regarding the simple correlation between the firing rates in FREE and FIX without interpolation (Figure 6A), we obtained median r coefficients equal to 0.15/0.10 for M1/M2 ([−0.13 0.45]/[−0.13 0.46], [25 75]th percentiles). The coefficient distribution was significantly shifted rightward (p = 2.3 × 10−4/0.045 M1/M2, one-sample Wilcoxon signed rank test). We found that most neurons in both animals (M1: 52/87 and M2: 22/32) had an intermediate to high correlation coefficient (<−0.2 or >0.2), indicating medium levels of relationship in spatial tuning when passing from FREE to FIX. However, only for a few neurons this relationship was significant (M1: 13/87, M2: 2/32; correlation p < 0.05). In accordance with the shift of the distribution toward positive values, all the significant correlations were positive. After filling in the missing data via interpolation, the correlation results were similar (Figure 6B). The new distributions had again slightly positive medians of 0.15/0.07 for M1/M2 ([−0.09 0.34]/[−0.14 0.25], [25 75]th percentiles), but for M2 the rightwards shift was not significant (p = 5 × 10−5/0.197 M1/M2, one-sample Wilcoxon signed rank test). Similarly to the results obtained without interpolation, most units showed medium to strong correlations (r < −0.2 or >0.2; M1: 46/87, M2: 17/32), but only a minority of these correlations were significant (M1: 22/87, M2: 4/32; correlation p < 0.05). In accordance with the slight shift of the distribution toward positive values, most of the significant correlations were positive (M1: 20/87, M2: 3/32). The consistency of the results between the two simple correlation approaches supported our choice to investigate the 2D structure of spatial tuning with the more complex method of vector correlation. The vector correlation analysis quantifies the correlation of the activity patterns in FREE and FIX by first computing the gradient of the 2D activity matrices to obtain vector fields and then correlating these vector fields, providing three measures, i.e., the ρ coefficient, the β scale factor, and the angle θ. In particular, a ρ coefficient around +1 (−1) indicates high correlation between the vector fields from FREE and FIX contexts, suggesting very similar activity pattern under rotation (reflection). Accordingly, a negative ρ value means that the vector fields (Figure 2, arrows of the 2D eye position matrices) were reflected along a certain axis, similar to the example neuron of Figure 2C. In turn, β values lower than 1 indicate that the vector fields (i.e., the arrows in Figure 2) that quantify the modulation of activity from one position to the next were larger in FREE than in FIX, whereas β > 1 indicates a weaker tuning in FREE than in FIX. For the example neuron in Figure 2A, the vector correlation analysis produced a coefficient ρ of 0.74, a scale factor β of 1.88, and an angle θ of 22°. The corresponding ρ, β, and θ values for the neurons in Figures 2B and 2C were 0.25, 0.58, and −64.48 and −0.4, 0.43, and 66.98, respectively. These values indicate that single neurons showed a diverse pattern of spatial relationships between the two behavioral contexts. The population results are shown in Figures 6C–6E. In both monkeys, we found a significant number of neurons (M1: 77/87, M2: 28/32) showing intermediate values (<−0.2 or >0.2, Figure 6C), suggesting a moderate degree of spatially related eye position fields between FREE and FIX. When tested against a null distribution obtained by randomly shuffling one of the vector fields, the ρ coefficient resulted significantly different than chance level for 36/87 (10/32) neurons in M1 (M2; empirical p value from the permutation null distribution with 1,000 shuffles <0.05). Overall, all three correlation coefficient measures suggested a moderate spatial relationship in the two contexts. The range of ρ coefficients was paralleled by a wide distribution of phase angles, which suggests a wide degree of rotation or reflection (Figure 6E). Of particular importance to our study is the scale factor β that quantified gain modulation by task context and thus can be seen as an alternative measure to PI and dPI for assessing spatial tuning. As shown in Figure 6D, the distribution of β factor was biased toward values below 1 (median β value in M1/M2: 0.75/0.79), indicating a weakening of spatial tuning obtained from response fields from FREE to FIX. When only the strongly significant correlations (ρ > 0.4 or ρ < −0.4) were considered, this trend was much weaker (median β value in M1/M2: 1.46/1.13). This suggests that in cases with a strong spatial relationship between the eye position fields in the two contexts there were small gain modulations, whereas less spatially related fields showed higher scaling of responses, with a bias toward the FREE gaze context. Taken together, these results suggest weaker gain modulation and lower tuning selectivity and strength of neural activity in V6A by gaze position during goal-directed fixation than during spontaneous fixation. In addition, responses were more variable in visually guided fixation. Discussion We compared eye position signals related to two behavioral contexts of the same task: (1) during internally driven, spontaneous eye fixations that unfolded during the 1-s waiting period in which the monkeys could freely move their gaze in darkness waiting for the visual stimulus to appear and (2) during the context of constrained fixation of visual targets. We found that V6A neurons showed context-dependent eye position modulations that were manifested as moderate changes in the mean firing rate and variability and, more importantly, as significant differences in tuning selectivity and strength and in the 2D structure of the eye position fields. Rather unexpectedly, the spatial tuning of gaze signals was stronger during unconstrained fixations than during goal-directed fixations. These findings could reflect a shift in brain state between the two contexts to ensure target response and adaptation to behavioral demands.34,38,39,49,50 Our hypothesis was that two task contexts would cause shifts in brain states associated with changes in several factors, including spatial attention and arousal. During unconstrained fixation, the monkeys were freely shifting their gaze and attention between several locations in darkness while expecting the appearance of a visual stimulus (LED target) that would become the target of the subsequent goal-directed fixation period. While scanning in darkness to detect a stimulus that is about to appear, the monkeys were allocating their attention between the locations of a previously viewed visual scene, a process very similar to endogenous, top-down orienting of attention.22,24,51,52 Both in humans and in monkeys, top-down signals modulate baseline activity even in the absence of visual stimulation, while the organism covertly attends to a peripheral location for stimulus onset.44,51,53 Despite task differences between the present study and the abovementioned ones regarding the free gaze period (covert vs. overt attention, dark environment vs. blank screen, and presence or absence of a fixation point), our view is that also in our case the voluntary allocation of attention modulated the spontaneous eye position activity via top-down mechanisms. Going from the FREE to the FIX task context, the appearance of the LED stimulus deployed spatial attention exogenously, in a stimulus-driven manner, and caused an overt attention shift (i.e., saccadic eye movement to fixate the LED). This suggests bottom-up processing due to visual saliency. The available evidence indicates that bottom-up and top-down spatial attention engage overlapping neuronal populations in the parietal cortex.54,55,56,57 These findings may explain the unimodal distributions we reported here, pointing to the absence of a segregated neural subpopulation with different properties and are reminiscent of some aspects of the mixed selectivity known to characterize the parietal cortex.58 During the FIX phase, the animals were attentively fixating the LED that was relevant to the behavioral goal. Differently to the FREE phase, where there was no fixation target, and gaze and attention were free to shift independently, in FIX, the focus of attention was aligned with gaze location and was sustained for 1–1.5 s. A previous study in V6A that employed a covert attention task reported that for most of the task-related neurons, activity was inhibited when the focus of attention moved to the fixation point.44 In the same study, the shifting of the attentional focus away from the gaze location resulted in comparable excitatory and inhibitory effects in the population. The balanced excitatory and inhibitory effects were also evident in our study, with very similar mean firing rate distributions across the neural population in the two contexts (Figure 3). Differently, the shift of attention that occurred much more frequently during the FREE context could have contributed to the sharpening of the gaze responses and the subsequent higher selectivity and tuning strength compared to the FIX context (Figure 4). In line with this assumption, several human and monkey neuroimaging and lesion studies confirm that the medial PPC is more active during shifts of spatial attention compared to sustained attention.2,23,59,60,61 The changes in eye position tuning that we observed could be related to other factors associated with the shift of behavioral state, such as arousal, surprise and expectancy. Arousal could affect PPC either directly via the noradrenergic system projections originating from locus coeruleus,35 or indirectly through the connections with the prefrontal cortex,62 which was recently proposed to act as a “choreographer,” modulating the activity of various cortical and subcortical networks during shifts from high to low arousal states.63 In the literature, a state of increased arousal modulates activity in the superior parietal lobule in humans.64 Similarly, high arousal causes large increases in PPC activity in mice, but without enhancing stimulus encoding.65 Given this evidence, we wondered which of the FREE or FIX behavioral states involved higher arousal levels. Our hypothesis was that the FIX state, with constrained fixation and a clear behavioral goal, together with visual attention, would also increase alertness to ensure a fast motor response and reward.22,24 However, our hypothesis might not be true, and it could be the case that the arousal state is higher during FREE epoch, when the animal scans the dark environment with saccades searching for the onset of the visual target. At this regard, saccadic eye movements during visual exploration seem to elicit modulations in the thalamo-cortical interaction, similar to those caused by the transition from low to high arousal states.37 Given the small changes in the firing rates that we found in our study, we assume that there were no significant differences in arousal levels between FREE and FIX. While our findings suggest a minimal contribution of the noradrenergic arousal system to the differences between the two behavioral contexts, an effect of the cholinergic system that also sends diffuse ascending projections to the cortex and is implicated in selective visual attention24 and goal-directed actions66 cannot be excluded. Interestingly, stimulation of the cholinergic system is reported to lead to broader neuronal tuning in the prefrontal cortex during working memory tasks.67 It might be the case also in our study that during FIX there was higher activation of the cholinergic system compared to FREE, that also contributed to the decrease in selectivity and strength of spatial tuning we observed (Figure 4). Beyond the noradrenergic and cholinergic influences, research in the visual system reported that active behavioral states are accompanied by motor activity and pupil dilation that not only evoke stronger sensory responses32,40,41,42 but also affect tuning.38 In particular, state-dependent pupil dilation caused a differential recruitment of rods and cones to adapt to changing behavioral demands and allow the perception of ethological stimuli. In our experimental setup unfortunately pupillometry data were not recorded, so we cannot provide evidence for these aspects. However, during the FREE phase, in addition to increased motor activity (spontaneous saccades), it is reasonable to assume that there was also an increased pupil dilation with respect to FIX due to the dark environment, but also given the expectancy of a visual stimulus.68 In analogous fashion, it might be the case that context-dependent pupil dilation resulted not in the differential recruitment of two receptor systems found in Franke et al.,38 but in changes in tuning in the same population of neurons. The present results extend the role of eye position signals in a wider range of contexts. Eye position signals are keys to how the nervous system builds an internal model of the outside world in order to perceive and act. During visually guided behavior, eye position information is integrated with the retinal image of the objects to enable the coding of the object’s location in body and hand-centered coordinates. Additionally, while perceiving auditory stimuli from targets beyond the field of view, the target location is encoded relative to the head, and eye position information is useful to transform head-centered to eye-centered coordinates.18 In our study, there was a complete absence of visual input during the FREE task context that coincided with the spontaneous saccades. It is likely that some of these saccades were memory-guided saccades of previous fixated target locations and there were attentional shifts toward internal (i.e., visual working memory) representations of target locations.69 It might be the case that eye position signals provided a mapping between the unseen egocentric space with the visual working memory representations and the stronger eye position tuning served to enhance this mapping. The activation of the parieto-occipital cortex, the broader region where V6A is located, without overt visual stimuli during self-generated eye movements has been reported in an early positron emission tomography (PET) study in humans.70 This activation was interpreted as a consequence of the efferent copies of motor commands issued for the spontaneous saccades and it was proposed to serve for the remapping of visual stimuli in space. Our findings add further support to this hypothesis. The fact that tuning functions of a single neuron can change between contexts and task periods is not new. Churchland and Shenoy71 reported that during a reaching task, many neurons in motor and premotor cortices changed their directional tuning between the planning and execution periods. Here, we found that single V6A neurons change their eye position tuning depending on whether the animal is freely moving its gaze in darkness while waiting for the visual target to appear or it is fixating the target while waiting for a cue to produce a response. This property of single neurons to change their tuning to one variable depending on several factors has been demonstrated in numerous studies in various species and described as “mixed” selectivity.58,72,73 This property has been viewed as a strategy to increase computational power by adding flexibility to the role of single cells that are involved in multiple neural processes.72 Similar to the results of our work, a recent study in the inferior parietal cortex reported significant changes in the eye position fields of single neurons during a fixate-to-reach task that involved distinct patterns of visual stimulation across various task stages.19 The authors found that despite the fact that the eye position tuning fields of single neurons changed across task stages, population activity encoded accurately the eye position in a subspace orthogonal (i.e., independent) to the one that coded context information. Whether this mechanism, which enabled a separate readout of eye position and task context by downstream structures, is also implemented in our task remains to be resolved by future studies that could leverage parallel recordings with multielectrode arrays (see limitations of the study). Limitations of the study Unfortunately, our recording procedures and analysis methods that involved very unequal trial durations of the FREE context (see STAR Methods) did not allow us to perform population analyses to probe for such a mechanism. Future studies could leverage parallel recordings with multielectrode arrays to further test this hypothesis. Another potential limitation of our study is the fact that the two task periods differed significantly, especially in terms of visual stimulation. Nevertheless, we expect a minimal confound of gaze signals by the presence of visual stimuli in the FIX period, considering that we recorded from the dorsal part of V6A where, as reported in several works from our group, only half of the cells show visual responses and the vast majority of them respond to complex stimuli such as borders and shadows, which are substantially different from the target LEDs.16,74 It should be noted that there are several studies in PPC where activity was studied with stimuli presented in different sensory modalities (e.g., visual/auditory saccades,18: reaches to visual/somatosensory targets47,75). In sum, our study results add to previous evidence addressing the flexibility of spatial representations in a wide variety of sensory and behavioral contexts. Resource availability Lead contact Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Kostas Hadjidimitrakis (kon.chatzidimitrakis@unibo.it). Materials availability This study did not generate new unique reagents. Data and code availability •The dataset supporting the current study is available from the lead contact upon reasonable request. The dataset supporting the current study is available from the lead contact upon reasonable request. •Custom code used to perform the analyses reported in this paper is publicly available on GitHub: https://github.com/francescovaccari/NeuralDataAnalysis_MATLAB. Custom code used to perform the analyses reported in this paper is publicly available on GitHub: https://github.com/francescovaccari/NeuralDataAnalysis_MATLAB. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments The authors wish to thank Prof. C. Galletti for helpful discussions and comments on an earlier draft, Dr. L. Passarelli and Prof. M. Gamberini for assistance with the surgical procedures, Drs. F. Bertozzi and G. Dal Bo’ for help during recordings, and M. Verdosci and F. Campisi for help in setting up the experimental apparatus. This work was supported by NextGenerationEU (NGEU) and funded by the Ministry of University and Research (MUR), National Recovery and Resilience Plan (NRRP), project MNESYS (PE0000006); this work was supported by Ministry of University and Research (MUR), PRIN2022-2022BK2NPS, NRRP ‘Fit4MedRob-Fit for Medical Robotics’ Grant (# PNC0000007) - BAC: SMART-tremor, cascade call for industrial research and experimental development projects under the PNC0000007 ‘Fit4MedRob – Fit for Medical Robotics’ initiative, Mission 3. Author contributions Conceptualization, K.H. and F.E.V; data curation, K.H., F.E.V., and S.D.; formal analysis and application, K.H and F.E.V.; funding acquisition, P.F.; investigation conduct, K.H. and F.E.V.; methodology and development or design of methodology, K.H., F.E.V., and S.D.; project administration and management, P.F.; resources – provision of study materials, P.F.; software programming and development, F.E.V.; supervision and oversight, K.H., M.D.V., M.F., and P.F.; validation and verification, K.H. and F.E.V.; visualization and preparation, K.H., F.E.V., and S.D.; writing – original draft, K.H. and F.E.V.; writing – review and editing, K.H., F.E.V., M.D.V, M.F., S.D., and P.F. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERExperimental models: Organisms/strainsLong-tailed Macaques (Macaca fascicularis)University of BolognaN/ASoftware and algorithmsMATLABMathWorkshttps://ch.mathworks.com/products/matlab.html?s_tid=hp_products_matlabCustom MATLAB codeUniversity of Bolognahttps://github.com/francescovaccari/NeuralDataAnalysis_MATLAB.Other5-channel MiniMatrixThomas recordinghttps://www.thomasrecording.com/5-electrode-mini-matrixETL-200 Eye tracking systemISCAN, Inchttp://iscaninc.com/standard-systems Experimental model and study participant details This study reports the analysis of a dataset, which has been partly used in a previous study from our lab76 and addressed the tuning of fixation activity for the peripersonal and extrapersonal space. Two young adult male macaque monkeys (Macaca fascicularis), M1 and M2, weighing 4.4 and 6 kg, respectively, were used. Initially, the animals were habituated to sit in a primate chair and interact with the experimenters. Then, a head-restraint system and a recording chamber were surgically implanted under general anesthesia (sodium thiopenthal, 8 mg/kg·h, i.v.), following the procedures reported by Galletti et al.16 A full program of postoperative analgesia (ketorolac tromethamine, 1 mg/kg i.m. immediately after surgery, and 1.6 mg/kg i.m. in the following days) and antibiotic care (Ritardomicina, benzatinic benzylpenicillin + dihydrostreptomycin + streptomycin, 1–1.4 mL/10 kg every 5–6 days) followed surgery. The experiments were performed in accordance with the guidelines of EU Directives (86/609/EEC; 2010/63/EU) and Italian national laws (D.L. 116-92, D.L. 26–2014) on the protection of animals used for scientific purposes. The protocols were approved by the Animal-Welfare Body of the University of Bologna and were authorized by the Italian Ministry of Health (Permits No 47/2008-B; 170/2015-PR; 389/2019-PR; n.645/2024-PR). During training and recording, particular care was taken to avoid any behavioral and clinical signs of pain or distress. Method details Extracellular recordings and the procedures to reconstruct microelectrode penetrations were similar to those described previously (e.g.,77,78,79). Single-cell activity was recorded extracellularly from the anterior bank of the parieto-occipital sulcus. Area V6A was initially recognized on functional grounds, following the criteria described in Galletti et al.,80 and later confirmed based on the cytoarchitectonic criteria by Luppino et al.81 We performed multiple electrode penetrations using a five-channel multielectrode recording system (Thomas Recording). The electrode signals were amplified (at a gain of 10,000) and filtered (bandpass between 0.5 and 5 kHz). Action potentials in each channel were isolated with a waveform discriminator (Multi Spike Detector; Alpha Omega Engineering) and sampled at 100 kHz. The histological reconstructions were performed as detailed previously.74 Briefly, electrode tracks and the approximate location of each recording site were reconstructed on histological sections of the brain on the basis of electrolytic lesions and other cues, such as the coordinates of penetrations within the recording chamber, the types of cortical areas passed through before reaching the region of interest, and the depths of passage points between the gray and white matter. All the neurons included in the present study were assigned to V6A, in particular its dorsal sector (see Figure S1 for visualization of the recording sites). Behavioral task Each monkey sat in a primate chair with the head restrained and faced a horizontal panel that hosted 10 circular LEDs (5 mm in diameter), mounted on the panel at different distances and directions with respect to the eyes (Figure 1A). The horizontal panel was located 10 cm below the level of the eyes, roughly at shoulder level. The target LEDs were arranged in two rows (five LEDs in each row): one central, along the sagittal midline, and one lateral, 17 cm to one side (Figure 1A). Along each row, the LEDs were spaced 15 cm apart. The distance from the eyes to the nearest central LED was 15 cm (Figure 1A). During the experiment, the LED panel was first positioned so that the lateral LEDs occupied the contralateral space with respect to the recording site (Figure 1B, “contralateral”). In cases where a neuron remained well isolated after recording at least 7 trials per position, the panel was translated, in order to present the LED targets in the central and ipsilateral spaces (Figure 1B, “ipsilateral”). The interocular distance of monkeys M1/M2 were 35 and 32 mm, respectively, resulting in vergence angles for central LEDs of 13/12, 7/6, 4/4, 3/3 and 2.5/2.5° for M1/M2. The corresponding vergence values for the lateral LEDs were 11/10, 8/6, 6/4, 5/3 and 3.5/2.5° for M1/M2. The horizontal version of the lateral LEDs from the nearest to the farthest one was 36, 29, 21, 17 and 14°. Since the LEDs were located at the same horizontal level below the eyes, the elevation angle was correlated to the vergence. In the central/lateral row, the elevation angle was 33/24, 18/16, 13/12, 9/9 and 8/7°, from the nearest to the farthest LED, respectively. The monkeys were trained to fixate, in a dark environment, one of the LEDs that was turned on for a variable period of time. The time sequence of the task and examples of spontaneous and instructed eye movements occurring in an individual trial are shown in Figure 1C. A trial began when the monkey pressed a home button (HB, Figure 1A) next to its chest (HB press, Figure 1C), which had to keep pressed while performing the task. After a 1000 ms interval, during which the animal was free to look anywhere, one of the LEDs turned green (LEDon, Figure 1C). The animal had up to 600 ms to make an eye movement to fixate the LED (FIXon, Figure 1C), otherwise the trial was aborted. After either 1 or 1.5 s from the fixation start, the color of the LED changed from green to red, signaling the go signal (LEDRed, Figure 1C). The monkey had to release the button to receive a liquid reward. The monkey’s arm ipsilateral to the recording hemisphere was restrained, so the button could be pressed only with the contralateral hand. Fixation LEDs were chosen in random order. Eye movements consisted of a combination of conjugate (version and elevation) and disconjugate (vergence) movements. Because the monkey was free to look anywhere at the beginning of the task, the initial position of the eyes varied from trial to trial. During LED fixation, eye position was constrained via an electronic window (3 × 3°) centered on each LED. Eye position from both eyes was recorded simultaneously with an infrared (IR) oculometer (ISCAN Inc., Woburn, MA, USA) at a sampling rate of 100 Hz. Stimulus presentation and animal behavior, including eye position signals, were monitored in real time using custom software written in Labview (National Instruments, Austin, TX, USA). Before each recording session, the monkey was required to perform a calibration task separately on each eye. Apart from the period of LED illumination, the experiment was performed in complete darkness. The IR illuminators and cameras were placed approximately 15 cm above the monkeys’ heads, and a hot mirror reflecting IR was placed in front of the monkey faces and tilted 45°. In this configuration, the animal could see through the mirror without perceiving the eye-tracker IR light. To prevent dark adaptation, the lights in the experimental room were switched on at regular intervals across the experimental sessions. Quantification and statistical analysis For each neuron and trial, neural activity (spike count and firing rate) were computed in 100-ms partially overlapping bins (25-ms step), whereas raw eye position signals, originally at 100 Hz, were downsampled interpolating every 25 ms (4 Hz) to match the neural activity sampling. The eye signals of the left and right eyes were combined in gaze coordinates using the following formulas: where xr,l and yr,l correspond to the horizontal and vertical coordinates for the right and left eye, respectively. Eye movement velocity was calculated as follows: During the initial phase of the task when the animals could move the eyes freely, saccades were detected when eye movement velocity was faster than 25°/s.82 We discarded from further analysis the data from 50 ms before to 100 ms after each saccade (Figure 1C, time windows marked “FREE” show retained data) to eliminate pre-, peri- and post-saccadic activity that is present in V6A neurons.76,83 During the visually guided fixation period, even though the eye position was controlled online and constrained, several small amplitude (<5°) saccades still occurred, given the presence of the fixation electronic window (3 × 3°) centered on the target. Accordingly, we applied in FIX the same preprocessing strategy used to detect during FREE i.e., saccade detection threshold as eye traces with velocity >25°/s and removal of activity from 50 ms before to 100 ms after each saccade. Note that this procedure was also useful for removing any larger microsaccades (normally defined as eye movements <1°,84;85 ) and for comparing results from the FREE epoch in a more strictly manner. We could not reliably detect any smaller microsaccades (10_40 arcmin of amplitude) with our eye-tracker configuration, mainly because of constraints in temporal and spatial resolution (100 Hz, lack of calibration with a fake eye to assess spatial resolution as reported by86). A detailed quantitative description of the saccades and fixations in the two different task intervals is provided in Figure S2. To compare the eye position activity between FREE and FIX, we binned the gaze data in a 2D grid located around the panel with the fixation LEDs. The grid spanned from 25 cm left to 25 cm right with respect to mid-eye level and from 20 to 80 cm in frontoparallel distance from the eyes (Figures 1D and 1E). When only the contralateral and central target rows were tested, the grid spanned from 25 cm in the contralateral side to 5 cm in the ipsilateral one. Fixation points exceeding 10 cm above or below the target plane were discarded. Thus, each grid square was contained in a 10 × 10 cm area, and we subsequently calculated the average activity within each square whether at least 5 datapoints had been collected. If not specified otherwise, missing values (when the animal did not fixate at a specific location of the grid) were allowed in the resulting 2D matrices that contained the average activity for all analyses except from the vector correlation and one of the two simple correlation analyses (Figures 6B–6E). For analyses such as the vector correlation analysis that required the calculation of a vector field and was not possible with a matrix with empty entries, we filled in the missing values using the neighboring grid cell values via interpolation or averaging (see also below). To fill any missing values via interpolation, the data were first interpolated along the two matrix dimensions independently and then the averaged values were used to substitute the missing ones. A similar procedure was used for filling with averaging only (data not shown). Alternatively, we adopted a second, more conservative approach of data preprocessing by assigning each bin to the target closest to the fixation point if the distance between the target and fixation point was within 3°. Otherwise, the bin was discarded from the analysis. The main results did not change significantly with this second approach (data not shown). Regarding the neural data, we included for further analyses the neurons that had an average firing rate during visually guided fixation higher than 5 sp/s for at least four LED targets and were modulated by target position (one-way ANOVA, p < 0.05). The PI and dPI were calculated as measures of spatial tuning induced by gaze signals during the spontaneous and visual fixation phases.87 The PI is a measure of the response selectivity that takes into account the magnitude of the neuronal activity for each eye position. It was computed as follows: where n is the number of positions, ri the activity for position i, and rpref the activity for the preferred position (highest activity). The PI value ranges between 0 and 1. A value near 0 indicates the same magnitude of response for all positions, whereas a value near 1 indicates strong preference for a single position. The dPI was computed using the formula: where rnonpref is the activity in the nonpreferred position (lowest activity). The dPI, also ranging from 0 to 1, is a measure of the strength of modulation of the neural activity induced by the change of fixation location. Both indexes were calculated on the real data without filling in any missing values via interpolation or averaging. To obtain a distribution of PI/dPI for each unit for statistical testing, we bootstrapped 100 times the activity from the 2D matrices. Bootstrapping was also used for the comparison of mean activity in each unit (Figure 3). We used eye position activity maps, one for FREE and one for FIX, to compute the average difference between activity in FREE and FIX, obtaining a mean delta value for each unit. For statistically testing on single units, a distribution of delta values for each neuron was obtained by randomly resampling 100 times the eye position maps and computing the difference on each resample. Finally, for each cell, we tested whether the delta distribution obtained via the bootstrapping was significantly different from 0 with a one-sample, two-tailed t test (p < 0.05). When data interpolation was required to analyze the spatial structure of the 2D eye position fields during spontaneous and instructed fixation, we first used the DAM index to discard any units where the missing data could bias the results. Specifically, the DAM index is a metric that considers the number of missing values and how they are spatially distributed in the matrix. We reasoned that, given a fixed amount of information, it would be easier (and more accurate) to reconstruct data when the missing values were distributed across the entire matrix rather than when the values were concentrated in a single, large portion with missing entries, which would render more likely the creation of artifacts (Figures S3 and S4). To account for these aspects, the DAM index increases when the number of missing values increases, and when missing values are clustered one next to the other. The DAM index can be calculated starting from a “damage measure” I that is defined as follows: where Ne represents the total number of elements in a matrix and n_valuesk is the number of non-missing values adjacent to the kth matrix elements (maximum: 8). Since I also depends on matrix size, an upper bound (Imax) and lower bound (Imin) can be found by computing I on a matrix of the same size of the real matrix with all missing values (Imax) or with no-missing values (Imin). The DAM index is then the result of normalization: The DAM index ranges from 0 (no missing data) to 1 (complete absence of data). We ran simulations to compute a threshold DAM value to discard neurons for which we did not have enough data and for which any attempt to reconstruct the missing value via interpolation would have produced too many artifacts. At the beginning of this procedure, for each cell, random elements were removed from the matrix containing the mean firing rate during fixation and the DAM index was computed. The same elements were then reconstructed by interpolating, and R2 was calculated between this new matrix (with data removed and reconstructed to simulate a procedure similar to that for the real data) and the original one. For this procedure, we used the fixation matrices because they contained only a few missing or no missing values. We repeated this procedure 1000 times for each cell. A third-degree polynomial was fitted considering the R2 of the reconstruction (independent variable) and the DAM index (dependent variable) which were related by a non-linear monotonically decreasing function (i.e., lower DAM index for higher R2). Threshold DAM values (DAM index ≤0.4) were extrapolated from the fitted function to ensure an R2 ≥ 0.35 that normally reflects good correlation. Please refer to Figures S3 and S4 for examples of the application of the DAM index in single neurons and for visualization of the concepts behind this approach, respectively. Fano factor analysis The FF is a common measure of variability that considers not only the variability per sè but also the magnitude of the response and has been used in neuroscience for quantifying how much neural variability generally shrinks when a stimulus is presented.29 Briefly, data for each neuron were first binned and divided into grid cells, as previously described. The following procedure was repeated twice, for the two different task contexts.29 The mean spike count and its variance were calculated for each grid cell. All pairs of mean and variance were concatenated in two column vectors for all spatial positions and all neurons of the population (vector length = neurons × grid cells). A single “raw” FF was calculated from the regression slope of the variance against the mean. The standard error of the slope (i.e., the standard error of the FF) was also obtained (fitlm MATLAB function). A “distribution-matched FF” was also computed to ensure that the underlying distribution of spike counts would not affect the results.29 First, for each spike count distribution, bin edges were calculated as 5% quantiles of the total distribution obtained by pooling the mean spike count distributions of the two task contexts. These edges were used to slice each mean spike count distribution in bins, producing a simple discrete probability density function (pdf). Two separate distributions were produced, one for the spike counts observed in the interval of spontaneous and one for the visually guided fixation. Then, each bin of the discrete pdf was rebalanced randomly removing exceeding samples from one of these two distributions. At the end of the procedure, the discrete spike count pdfs for the two task contexts were completely overlapping. The distribution matching was controlled with two-sample Kolmogorov-Smirnov tests (p ≈ 1). The corresponding variance values were also removed. The “distribution-matched FF” was calculated on these new distributions. Considering the random component in subsampling, we repeated the procedure 1000 times to ensure robustness. Vector correlation We applied the vector correlation method88 to quantify the relationship between the two 2D firing rate matrices obtained from the activity during the spontaneous and visual fixation phases. Each 2D matrix was transformed into a 2D vector field that described the gradient of the response (gradient MATLAB function). To calculate the estimate of vector correlation, we used the method first developed by Hanson et al.88 to analyze geographic data. By applying this method, a correlation coefficient ρ that is analogous to the Pearson correlation coefficient was calculated. This coefficient quantified how much the 2D vector fields are related to each other. The method also defines the amount of rotation or reflection and the scaling between the two vector fields. If x and y are the two dimensions of one vector field, and u and v the dimensions of the other, using the following equation from Hanson et al.88 a correlation coefficient ρ is calculated: with ξ = σxuσyv−σxvσyu and s=sng(ξ)=ξ|ξ|σx2, σy2, σu2 and σv2 are the variances of x ,y, u, v and σxu, σyv, σxv and σyu are the covariances of the four dimensions. A scale factor (β) and a phase angle (θ) can also be calculated: The coefficient ρ has a range from −1 to 1, with 1 characterizing a perfect rotational relationship between the two vector fields and −1 denoting that one vector field can be produced by the reflection of the other along a given axis. The phase angle θ has a range from −180 to 180° and quantifies the angle of rotation or reflection that is necessary to align the two vector fields, whereas the factor β represents the gain relationship between the fields. The β value resulting from vector correlation is a factor of scale that indicates how the first set of vectors must be rescaled to fit the second set. In other words, it can be seen as a measure of the relative magnitude between the vectors of the first and second sets and, in our case, of stronger (β > 1) or weaker (β < 1) neural modulations for space in the free phase with respect to the fixation phase. In Hanson et al.,88 an exact method for computing the p-value of the vector correlation is provided. As an alternative, we also implemented a p-value based on a nonparametric permutation test. This procedure involved generating nshuffle surrogate datasets by randomly shuffling the location of the vectors of the second vector field, breaking the eventual correspondence with the first vector field. For each k-th surrogate dataset, the ρk coefficient was computed to form a null distribution. The p-value was obtained as: where ρobs is computed on the original data and count indicates the number of events in which |ρk|>|ρobs| is verified. All analyses were implemented in MATLAB. The main functions can be found at https://github.com/francescovaccari/NeuralDataAnalysis_MATLAB. Footnotes Supplementary data related to this article can be found online at https://doi.org/10.1016/j.isci.2026.117459. Contributor Information Kostas Hadjidimitrakis, Email: kon.chatzidimitrakis@unibo.it. Patrizia Fattori, Email: patrizia.fattori@unibo.it. Supplemental information Document S1. Figures S1–S5 References Associated Data Supplementary Materials Document S1. Figures S1–S5 Data Availability Statement •The dataset supporting the current study is available from the lead contact upon reasonable request. The dataset supporting the current study is available from the lead contact upon reasonable request. •Custom code used to perform the analyses reported in this paper is publicly available on GitHub: https://github.com/francescovaccari/NeuralDataAnalysis_MATLAB. Custom code used to perform the analyses reported in this paper is publicly available on GitHub: https://github.com/francescovaccari/NeuralDataAnalysis_MATLAB. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.