Modular Representation of Luminance Polarity in the Superficial Layers of Primary Visual Cortex

Abstract

The spatial arrangement of luminance increments (ON) and decrements (OFF) falling on the retina provides a wealth of information used by central visual pathways to construct coherent representations of visual scenes. But how the polarity of luminance change is represented in the activity of cortical circuits remains unclear. Using wide-field epifluorescence and two-photon imaging we demonstrate a robust modular representation of luminance polarity (ON or OFF) in the superficial layers of ferret primary visual cortex. Polarity-specific domains are found with both uniform changes in luminance and single light/dark edges, and include neurons selective for orientation and direction of motion. The integration of orientation and polarity preference is evident in the selectivity and discrimination capabilities of most layer 2/3 neurons. We conclude that polarity selectivity is an integral feature of layer 2/3 neurons, ensuring that the distinction between light and dark stimuli is available for further processing in downstream extrastriate areas.

Figure 1. Modular mapping of polarity preference in layer 2/3 of visual cortex

(A) Single condition ON and OFF responses to uniform luminance polarity transitions visualized with wide-field epifluorescence imaging (PD35 ferret). (B) Polarity difference map reveals the differential ON–OFF response. Scale bars: 1 mm. (C) Polarity preferences of individual layer 2/3 cells overlaid on the wide-field map in (B). Scale bars: 100μm. (D) Representative responses for three neurons highlighted in (C) showing OFF, ON, and non–selective responses. (E) Polarity maps (n=15) show high trial-to-trial correlations within stimulus and weak correlations across stimuli (error bars are mean±SEM, open circles indicate individual animals; * p<0.05, # p< 0.0001). (F) Cellular polarity preferences (ON–OFF ratio, OOR) display a slight dominance for ON responses. (G) Cellular polarity preferences show spatial organization, both horizontally and vertically within layer 2/3. Scale bar: 100 μm. (H) (left) Horizontal clustering was quantified by comparing the neuronal pairwise difference in OOR in each planar field of view as a function of pairwise horizontal distance for actual and position shuffled data (median of 10,000 shuffles shown). Dashed lines indicate linear fits from 0– 300 μm separation. The slope of the linear fit is used to assess the significance of clustering versus shuffle. (right) Distribution of linear slopes for 10,000 shuffles and actual data (red arrow) demonstrates significant horizontal clustering (p < 1e-4). (I) Vertical clustering was computed by comparing the similarity of polarity preference for neurons within a cylinder of 100 μm diameter. Differences in preference within vertical cylinders for actual data (blue) are significantly smaller than those after shuffling neuronal positions (red). (J) Response amplitude in individual neurons is proportional to Michelson contrast. See also Figures S1 and S2, and Movies S1 and S2.

Figure 2. Responses to uniform luminance steps are spatially restricted across visual cortex

(A–B) Response strength varies across cortex for: (A) uniform luminance stimuli (maximum of ON and OFF) and (B) oriented gratings (maximum across orientations). Scale bar: 1 mm for A–C. (C) The response strength index (RSI) also varies across cortical surface. (D–E) Varied response to uniform luminance steps and gratings in layer 2/3 neurons. (D) Representative examples showing grating and uniform luminance evoked responses in individual neurons drawn from the full RSI distribution (E) at positions indicated by colored squares. (E) Distribution of RSI across individual neurons shows the majority of cells are dominated by grating evoked responses. (F) Spatial clustering of cellular RSI (left) corresponds to wide-field organization (right). White box in (C) indicates region shown in (F). Scale bar: 100 μm. (G–H) Spatial clustering of grating and uniform luminance response strength at the cellular level. (G). Neuronal pairwise difference in response strength index (RSI) as a function of pairwise horizontal distance for actual and position shuffled data (median of 10,000 shuffles shown). Dashed lines indicate linear fits from 0–300 μm separation. The slope of the linear fit is used to assess the significance of clustering versus shuffle. (H). Distribution of linear slopes for 10,000 shuffles and actual data (red arrow) demonstrates significant horizontal clustering (p<1e-4). See also Figures S3-5.

Figure 3. Strong polarity selective responses evoked by visual edges

(A) Schematic of edges of the same orientation, but different polarity: ON edge (a bright edge swept across a dark screen) and OFF edge (a dark edge swept across a bright screen). (B–C) Single condition responses to edges with same orientation but opposite polarities visualized with wide-field epifluorescence imaging: (B) ON edge and (C) OFF edge. Scale bars: 1 mm. (D) Representative responses from two highlighted regions in (B) and (C) demonstrate that drifting edge stimuli elicit polarity selective responses (E) Polarity difference map of the single-condition edge maps in (B–C) reveals a differential ON–OFF response to edges of the same orientation. Scale bar: 1 mm. (F) The polarity difference map, averaged across all presented directions, resembles the polarity difference map for the single-direction shown in (E). Scale bar: 1 mm. (G) Wide-field maps from boxed region in (F) overlaid with polarity preference of individual layer 2/3 neurons. Scale bar: 100 μm. (H) Representative responses for three neurons highlighted in (G) showing ON-selective, non-selective, and OFF-selective edge responses. See also Movie S3.

Figure 4. Edges reveal widespread polarity selective responses

(A) Polarity maps evoked by edge (left) and uniform luminance stimuli (right) are highly similar. (B) However relative polarity selectivity for edge versus uniform luminance stimuli varies across cortex in a spatially-structured manner. (C) The map of response strength index (RSI) is well correlated with the pattern of relative polarity selectivity (r=0.36). Stars indicate corresponding reference landmarks. Scale bars in (A–C): 1mm. (D) Wide-field maps from boxed region in (A) overlaid with polarity preference of individual layer 2/3 neurons. Scale bar: 100 μm. (E) Elevated selectivity for polarity is revealed by edge stimuli for neurons dominated by grating responses (low RSI), whereas uniform luminance drives greater selectivity for neurons with weak grating responses (high RSI). Stars indicate p<0.003, post-hoc MW test with Bonferroni-Holm correction for multiple comparisons. Error bars are mean±SEM across neurons within RSI bins. See also Figure S6.

Figure 5. Polarity and orientation maps show high coverage despite a lack of orthogonality

(A–B) Coverage for stimulus space is high at the scale of cortical hypercolumns. (A) Stimulus-space coverage for polarity, orientation and the combination of both features for two search diameters. (B) Coverage is high for all stimulus spaces at a search area diameter roughly matching the scale of a typical hypercolumn (1000 μm). Shaded areas indicate 95^th confidence intervals. Scale bars: 1 mm. (C–F) Polarity and orientation maps lack orthogonal organization. (C) Polarity preference map overlaid with zero polarity contour line (white line). (D) Orientation preference map, overlaid with contour lines (each black line indicates 30° separation). (E). Expanded region of orientation map shown in (D) with orientation contours (black) and polarity zero-contour (white), showing regions of both parallel and perpendicular contours. (F) Angular difference in gradient direction between polarity and orientation maps. An orthogonal geometric relationship would be indicated by an over-representation of pixels near 90 degrees. Scale bars in (C–D): 1 mm. (G–H) Polarity and orientation hypercolumns display structural differences. (G) Polarity hypercolumns are significantly wider than orientation hypercolumns, as quantified by the mean domain spacing. (H) Polarity hypercolumns are more spatially isotropic than orientation hypercolumns. This difference was quantified by computing the mean bandedness (i.e. stripe-likeness) of polarity and orientation hypercolumns. In (G–H): # indicates p<0.001, error bars are mean±SEM, open circles indicate individual animals.

Figure 6. Response to edges depends on luminance polarity and orientation

(A) Edge response is best predicted by intersection of polarity and orientation maps. Top: response masks derived from full field stimuli. Bottom: horizontal OFF edge response. Yellow line denotes response mask boundary. Asterisks highlight regions in which response is poorly predicted by orientation mask. Scale bar: 1 mm. (B) Edge responses match intersection significantly better than either mask individually (# p≪ 0.001). Error bars indicate mean±SEM, and gray points indicate individual edge stimuli. (C) Edge stimuli vary along both orientation and polarity dimensions. (D) Stimulus selectivity along the intersection of both polarity and orientation dimensions is significantly higher than along either polarity or orientation alone. (E) A large majority of imaged pixels exhibit greater selectivity for the intersection of polarity and orientation (83%) than for one dimension alone (polarity: 1%, orientation: 16%). (F) The degree of enhancement in selectivity depends on the response strength index (RSI, r = 0.66). In (D–F): error bars are mean±SEM, gray circles in D,E indicate individual animals.

Figure 7. Cooperative discriminability of edge orientation and polarity

(A) Representative edge tuning curves pooled across edge polarity (red), or at the preferred polarity (blue). (B) Orientation selectivity is significantly enhanced at the preferred polarity. (C) Representative edge-evoked polarity responses either pooled across edge orientations (red) or at the preferred orientation (blue). (D) Polarity selectivity is significantly enhanced at the preferred edge orientation. (E–F) Discriminability (d’ from ROC analysis) of orthogonal orientations (E) and adjacent orientations (F) is significantly greater at the preferred polarity. Colored points in (E–H) indicate median and interquartile range for neurons binned according to RSI (blue: RSI < −0.3, green: RSI −0.3 to 0.3; red: RSI > 0.3, corresponding to grating dominated, balanced, and uniform luminance dominated responses). (G) Discriminability of ON vs. OFF edges is significantly enhanced at the preferred orientation. (H) Combined discriminability along the intersection of preferred orientation and polarity is significantly greater than for either orientation or polarity alone. See also Figure S7.