Neural Encoding of Edge and Surface Properties in Macaque Visual Area V4

Abstract

Area V4, a mid-level stage in the primate ventral visual pathway, plays a critical rolein shape perception and object recognition. While V4 neurons are known to be selec- tive for boundary curvature, it remains unclear how they integrate information from edge contours and surface fill properties. Previous work from our laboratory (Popovkina et al., 2019) demonstrated that V4 neurons exhibit diverse response patterns to filled versus out- lined shapes, suggesting heterogeneous encoding strategies. Building on this foundation, we designed a five-condition stimulus set comprising 73 shape × rotation combinations across five fill types (A: solid fill, B: blurred inner fill, C: heavily blurred inner fill, D: con- centric double outline, E: single outline) to dissociate four competing hypotheses about how V4 encodes boundary and surface information. Using Neuropixels NHP 1.0 probes, we recorded 154 well-isolated V4 neurons from one male rhesus macaque during passive fixation. A custom data processing pipeline achieved sub-millisecond temporal alignment between neural recordings and stimulus presentation via sync signal-based sample rate correction. Of 154 visually responsive neurons, 69 (44.8%) met criteria for fill-type se- lectivity (two-way ANOVA p < 0.05 and Fill Selectivity Index ≥ 0.159). Fill-selective neurons formed two dominant subpopulations: filled-preferring neurons (33/69, 48%), showing strongest responses to Types A and B with opponent suppression for outline conditions, and outline-preferring neurons (27/69, 39%), showing the reverse. The Type D (double-outline) condition was critical for distinguishing boundary encoding from cen- ter sampling: the majority of fill-selective neurons showed greater response similarity between Types D and E than between D and any filled condition, consistent with pri- mary reliance on boundary geometry, while a substantial minority showed the opposite pattern. Shape tuning was broadly preserved across fill conditions. Of 69 fill-selective neurons, 73.9% showed significant shape consistency in at least one pairwise fill-type comparison. Critically, 71.0% of neurons showed significant consistency in at least one cross-category comparison (filled vs. outlined conditions), and cross-category Spearman correlations were statistically higher than within-category correlations (mean r = 0.346 vs. 0.324; Wilcoxon W = 800, p = 0.015), indicating that shape representations general- ize across the filled-to-outlined boundary at least as robustly as within surface categories. The distribution of significant pairs was bimodal: 26.1% of neurons showed no significant consistency across any pair, while 34.8% showed significant consistency across all 10 pairs, suggesting two functionally distinct subpopulations differing in the degree to which shape representations are invariant to fill type. These results reveal that V4 encodes fill-type information through parallel subpopulations with distinct preferences, integrates bound- ary and surface signals at the single-neuron level, and does so with a temporal precision that precedes conscious object recognition.