Monkey V1 epidural field potentials provide detailed information about stimulus location, size, shape, and color

The study addresses whether epidural field potentials (EFPs), a low-invasive alternative to subdural ECoG or intracortical recordings, preserve detailed information about visual stimulus features in macaque V1. Prior concerns suggest EFPs may have lower specificity due to greater distance from neural generators and larger cortical spread, yet they offer advantages in safety and artifact robustness. The authors investigate the extent to which V1 EFPs encode stimulus location (retinotopy scale), size (sub-receptive field precision), shape (orientation statistics across hypercolumns), and color (blob/interblob contributions), thereby probing functional specificity at multiple spatial scales. The purpose is to quantify EFP sensitivity and separability for these features and assess whether such modulations are sufficient for reliable single-trial decoding, which is important for basic neuroscience and brain–computer interface applications.

Previous work provides mixed conclusions about epidural signals. Some studies reported higher noise levels and more difficult decoding for EFPs compared to subdural or intracortical signals, while other studies found comparable decoding performance between EFP and LFP and no detrimental dura effects on feature detection. A recent study showed V1 EFPs can yield >90% detection of spatial information from brief fragments, though confounded by concomitant differences in size, orientation, and color. V1’s retinotopy and hypercolumnar organization (orientation/color) suggest EFPs might integrate over multiple columns yet still express feature-specific biases. Prior literature also reports gamma-band signatures for location in high gamma and size/orientation effects in low-to-mid gamma for LFP/ECoG. Color tuning in V1 is often weak and variable at the single-unit level and not strongly clustered, implying uncertain expectations for chromatic modulation in EFPs. These mixed findings motivate a systematic assessment of V1 EFP information content across location, size, shape, and color.

Subjects: Two male macaque monkeys (Macaca mulatta), M1 (13 y, 12.8 kg) and M2 (11 y, 11.5 kg). Both were implanted with an epidural microelectrode array over the left V1 hemisphere, with connectors and a titanium head post fixed to the skull. All procedures were approved by the relevant authorities and followed EU Directive 2010/63 and German regulations. Electrophysiology: Epidural arrays comprised many electrodes (reported totals across analyses ~313 channels), with EFPs recorded at high sampling rates (initially up to 25 kHz, downsampled to 1 kHz after preprocessing). Signals were referenced to a skull reference electrode, low-pass filtered (<300 Hz), notch-filtered at 50 Hz, and wavelet-transformed (Morlet). Power was baseline-normalized using the 500 ms pre-trial period. Receptive field mapping: Epidural receptive fields (ERFs) were obtained using an automated reverse-correlation bar-mapping procedure with oriented moving bars. ERF size was estimated by closed areas surpassing a Z-score criterion, converted to diameter assuming circular shape. Mean EFP receptive field diameters were approximately 3.0 ± 0.46 deg (M1) and 2.8 ± 0.66 deg (M2). Visual stimuli and task: Visual objects appeared at five fixed locations in the lower right visual quadrant at eccentricities of approximately 3.5, 5.8, and 8 deg. At each location, three sizes were used (small, medium, large; diameters around 1.0, 1.2, 1.4 deg). Shapes included two triangles, two squares (differing in global orientation), and one circle. Colors comprised five luminance-matched conditions (four chromatic, one achromatic). Stimuli were presented in sequences while monkeys fixated and performed a dimming-detection task; each object was shown for 300 ms with inter-stimulus blanks. Time–frequency feature selection: For each feature category (location, size, shape, color) and monkey, the most informative time–frequency windows were identified using an ROC-based procedure over wavelet-transformed power, computing AUC variance maps across electrodes. Location effects were most consistent in high-gamma; size and shape in low-to-mid gamma; color showed later modulation. Analyses then averaged baseline-corrected wavelet power within the selected windows per electrode and condition. Feature sensitivity metrics and statistics: Spatial sensitivity was quantified as differences between responses to the closest (assigned) versus second-closest stimulus locations, and analyzed as a function of ERF–stimulus center distance. Size sensitivity was computed as mean absolute power differences across adjacent sizes. Shape sensitivity was computed from pairwise differences among circular, quadrangular, and triangular shapes; electrodes also had an orientation index (OI) from bar mapping. Color selectivity index (CSI) summed absolute pairwise differences across colors normalized by mean response to minimize dependence on absolute activation; early (≈40–90 ms) and late (≈120–175 ms) epochs were compared. Nonparametric tests included Kruskal–Wallis for main effects, post-hoc Tukey–Kramer-corrected pairwise tests, and Wilcoxon signed-rank tests. Effect sizes (e.g., ω2) were reported where applicable. Single-trial decoding: Nonlinear SVMs with RBF kernels (libsvm) were trained to classify single trials based solely on non-spatial features for objects presented at the same location. Feature combinations that elicited high versus low gamma power on average (e.g., large blue circle vs small green triangle) were contrasted. Leave-one-out classification was used per condition, with cross-validated selection of C and gamma. Performance was compared to label-shuffled controls; significance at p < 0.05 (Mann–Whitney U).

- Robust spatial selectivity: - 311/313 electrodes showed significant dependence on stimulus location (Kruskal–Wallis, p < 0.028). - 58% (M1) and 75% (M2) of electrodes responded at their assigned location differently than at any other location (post-hoc tests, all p < 0.05). - Spatial sensitivity decreased rapidly with ERF–stimulus center distance; a half-Gaussian fit to assigned-location responses yielded R2 = 0.497, with strong decline within 1.5 deg. - Differences between closest vs second-closest location were significant at all distances (Wilcoxon Z > 2.501, p < 0.01), with effect sizes large up to 0.75 deg (ω2 ≈ 0.175), medium up to 1.5–2 deg (ω2 ≈ 0.091–0.096), and small up to 3 deg (ω2 ≈ 0.058). - Size sensitivity: - Despite small size steps (≈0.2 deg diameter), gamma power differed significantly across small/medium/large stimuli across a range of ERF–stimulus distances, indicating fine-grained spatial sensitivity beyond ERF boundaries. - Shape/orientation sensitivity: - Many electrodes exhibited a weak but significant orientation bias; median OI ≈ 0.115 among channels with significant bias. - Strong main effect of shape (Kruskal–Wallis χ2 = 148.89, p < 1e−30; df = 4). No difference between the two triangular (p = 0.99) or the two quadrangular shapes (p = 0.91), but all other pairwise comparisons were highly significant (p < 0.001). - 138 (78.4%) and 118 (90.1%) channels (M1 and M2, respectively) were significantly modulated by angularity; at 53.4% of these channels two of the three angularity conditions differed, and at 4.7% all three differed. - Shape sensitivity decreased with ERF–stimulus distance but remained significant up to 3 deg; effect sizes were large across distances (ω > 0.181). - Color sensitivity: - Early epoch (≈40–90 ms): minimal color modulation; 3/313 electrodes significant. - Late epoch (≈120–175 ms): robust color modulation; 210/313 electrodes significant (Kruskal–Wallis p < 0.05). - Blue and red elicited higher gamma power than other colors, on average ~10% more in early and ~30% more in late responses. CSI was significantly larger in the late epoch (Wilcoxon Z = 10.69, p < 1e−25, N = 313), indicating increased hue-specific differences over time. - Time–frequency dissociations: - Location: strongest in high gamma (≥ ~85 Hz) during onset. - Size/shape: low-to-mid gamma (≈30–80 Hz) during onset. - Color: later modulation with maxima in distinct gamma ranges; delayed and lower amplitude relative to other features. - Single-trial decoding: - SVM classification of objects shown at the same location, based solely on non-spatial features, achieved high performance: 75.6 ± 4.6% (M1) and 85.5 ± 1.5% (M2) mean accuracy across electrodes for challenging high-vs-low gamma eliciting contrasts, exceeding shuffled-label controls.

The findings demonstrate that V1 EFPs, despite being mass signals integrating over large neuronal populations, retain detailed and reliable information about multiple stimulus features across spatial scales. Spatial selectivity of gamma-band EFPs is finer than suggested by average ERF size; sensitivity declines steeply within ~1.5 deg from ERF centers. Non-spatial features—small changes in size, differences in geometric shape (orientation statistics), and hue—systematically modulate EFP gamma power in largely consistent ways across electrodes. Shape effects likely reflect stimulus-dependent differences in the spatial–temporal activation of orientation-selective populations across multiple hypercolumns, while color effects—emerging prominently in a later time window—indicate general differences in processing short- and long-wavelength hues (blue/red) relative to mid-wavelength hues, independent of simple luminance differences. Distinct time–frequency ranges for location (high gamma), size/shape (low-to-mid gamma), and color (later gamma bands) further support functional specificity. Crucially, these modulations are robust enough to enable accurate single-trial classification using only non-spatial features for stimuli at the same location. Collectively, the results position EFPs as a selective and practical signal for decoding distributed cortical activity with reduced invasiveness, relevant for long-term clinical applications and basic research on mesoscale network interactions.

Epidural field potentials in macaque V1 contain highly selective information about stimulus location, size, shape, and color. Spatial information modulates at sub-ERF scales, and non-spatial features produce consistent gamma-band signatures—sufficient for robust single-trial decoding based solely on size, shape, and color for stimuli at the same location. These results support the use of EFPs as a minimally invasive, functionally informative signal source for chronic, long-term applications in clinical and BCI settings and for basic research on mesoscale cortical dynamics. Future work should optimize arrays and recording stability to reduce session-wise variability, further map feature-specific time–frequency signatures, and extend assessments to other cortical areas and task contexts.

- The study focuses on V1; generalization to other regions or tasks remains to be established. - Color effects depended on stimulus parameters and appeared primarily in a later time window; differences from other studies (e.g., full-screen vs small stimulus dots) suggest stimulus-dependent variability in chromatic gamma modulation. - Orientation selectivity at the EFP level was small; shape effects may partly reflect differences in the number or spatial distribution of activated neurons rather than pure orientation tuning. - Time–frequency analyses identified broad informative ranges; potential shifts with other stimulus properties (e.g., contrast) were not explored. - Session-wise variability across weeks could affect sensitivity; authors note decoding/sensitivity might improve with reduced variability. - Some reporting shows inconsistencies in channel counts across sections; however, core analyses frequently reference N ≈ 313 electrodes.