Texture is encoded in precise temporal spiking patterns in primate somatosensory cortex

Humans are exquisitely sensitive to the microstructure and material properties of surfaces. In the peripheral nerves, texture information is conveyed via two mechanisms: coarse textural features are encoded in spatial patterns of activation that reflect their spatial layout, and fine features are encoded in highly repeatable, texture-specific temporal spiking patterns evoked as the skin moves across the surface. Here, we examined whether this temporal code is preserved in the responses of neurons in somatosensory cortex. We scanned a diverse set of everyday textures across the fingertip of awake macaques while recording the responses evoked in individual cortical neurons. We found that temporal spiking patterns are highly repeatable across multiple presentations of the same texture, with millisecond precision. As a result, texture identity can be reliably decoded from the temporal patterns themselves, even after information carried in the spike rates is eliminated. However, the combination of rate and timing is more informative than either code in isolation. The temporal precision of the texture response is heterogenous across cortical neurons and depends on the submodality composition of their input and on their location along the somatosensory neuraxis. Furthermore, temporal spiking patterns in cortex dilate and contract with decreases and increases in scanning speed, respectively, and this systematic relationship between speed and patterning may contribute to the observed perceptual invariance to speed. Finally, we find that the quality of a texture percept can be better predicted when these temporal patterns are taken into consideration. We conclude that high-precision spike timing complements rate-based signals to encode texture in somatosensory cortex.


Supplemental
: Confusion matrices of cross-correlation in measured responses and simulated ones. A| Confusion matrices for an example cell, the same example cell used throughout. Each pixel represents the mean pairwise cross-correlation across responses to a training texture (abscissa) and test texture (ordinate) at a resolution of 2 ms. Correlation values scale with firing rate in Poisson neurons and, to a much lesser extent, in measured responses to texture. For the measured responses, correlation values for responses to the same training and test texture (in the diagonal) are higher than mismatched pairs, whereas in the Poisson model, no such structure exists. B-D| Same convention as in panel A with mean values across all cells (B), a subpopulation of PC-like cells (C), and a subpopulation of RA-like cells (D).
Supplemental Figure 4: Temporal resolutions derived from the repeatability and classification analyses. A| Cumulative distributions of resolutions obtained in the repeatability analysis (red) and those yielding maximum performance in the timing-based classification (blue). B| Comparison of the resolution determined from the repeatability analysis and timing-based classification. Each point represents one neuron; the dashed line denotes unity. C| Absolute difference in resolution between repeatability analysis and timingbased classification as a function of each neuron's classification performance. D| In neurons with a classification performance better than 30% (those to the right of the grey line in C), the resolution estimated from the timing-based classification is always within 5 ms of the resolutions estimated from the repeatability analysis, and often within 0.5 ms.
Supplemental Figure 5: Timing-based classification using simulated jittered responses. A| Less jitter in simulated spike trains results in better timing-based classification performance; Kruskal-Wallis test comparing max classification performance for each simulated cell across the three lowest levels of imposed jitter ( 2 = 142.6, p < 0.0001); post-hoc 1-sided Mann-Whitney U tests yield significant differences between 1-ms jitter (maroon) and 2-ms jitter (salmon, z-statistic = 4.0, p < 0.0001) and between 2-ms jitter and 5-ms jitter (blue, z-statistic = 9.2, p < 0.0001) . B| The best classification resolution coincides with the amount of imposed jitter in these simulated responses.
Supplemental Figure 6: Single-cell and population classification using rate, timing, and their mean. A| Single cell classification performance using rate difference, timing maximum cross-correlation, and an average of both (rather than an optimal combination). Violin plots show all values. Boxplots indicate median (center), interquartile range (boxes), and maximum and minimum (whiskers). B| Population classification using rate dissimilarity (red), timing cross-correlation (blue), and both (rather than an optimal combination, purple), averaged across the full neuronal population. Lines indicate the mean across 1000 iterations. Shaded areas denote standard deviation. C| The weighting of rate and timing that yields the highest classification performance for each cell. Each point denotes a single cell's optimal rate contribution. Error bars denote the standard deviation across 141 cells. The mean optimal weighting is 52.4%, which explains why both A and B (in which all cells' weightings are set to 50%) are nearly indistinguishable from Figure 2B and C (in which weighting is optimized for each cell).
Supplemental Figure 7: Responses of PC-like, SA1-like, and RA-like cortical cells. A| Responses of nine example PC-like neurons to five repeated presentations of nine (of 59) textures. Each row is the response of an individual neuron across five repeated presentations of a given texture. B| Same as above, but for nine example SA1-like neurons. C| Same as above, but for nine example RA-like neurons.
Supplemental Figure 8: Timing-based classification based on peripheral and cortical responses for a matched texture set. A| Timingbased classification performance across temporal resolutions for PC-like (orange), SA1-like (green), RA-like (blue), and all remaining unclassified neurons (grey) for a subset of 24 textures that were used in both the cortical and peripheral experiments. Shaded area denotes the standard error across cells. B| Timing-based classification performance for the same 24 textures based on PC (orange), SA1 (green), and RA (blue) afferent responses. Shaded area denotes the standard error across cells. C| The best classification resolutions are similar for cortical and peripheral neurons, but some cortical neurons are temporally imprecise whereas no afferents are. D| The temporal resolutions of PC fibers (orange dashed) and PC-like cortical neurons (orange solid) overlap, whereas some SA1like (green solid) and RA-like cortical neurons (blue solid) are less precise than their afferent counterparts (green dashed and blue dashed, respectively). E| Single-cell classification based on firing rates evoked in cortical and peripheral neurons. Each point represents one neuron's classification performance. Bars indicate means and error bars denote standard deviation across 141 cortical and 39 peripheral neurons. Darker bar denotes cortical cells. F| Same as in E, but for timing-based classification.
Supplemental Figure 9: Classification of textures based on peripheral and cortical population responses. A| Classification of 24 textures (chance = 1/24, 4.2%) based on the rate (rate) and timing (blue) of the response of populations of SA1 fibers. Shaded area denotes the standard deviation across 100 randomly sampled populations of peripheral or cortical neurons. Solid lines represent SA1-like cortical cells (n = 1-25); dashed-lines represent SA1 afferents in the peripheral nerve (n=1-17). B| Same, with PC-like cortical cells (n=1-12) and PC afferents (n=1-7). Shaded area denotes the standard deviation across 100 randomly sampled populations of peripheral or cortical neurons. C| Same, with RA-like cells (n=1-12) and RA afferents (n=1-15). Shaded area denotes the standard deviation across 100 randomly sampled populations of peripheral or cortical neurons.
Supplemental Figure 10: Differences in temporal resolution across cortical fields. A| Single-cell timing-based classification performance across different cortical fields. Each circle represents one neuron's performance; orange, green, and blue circles are color-coded based on the neuron's dominant submodality input (PC, SA1, and RA, respectively). Bars indicate means and error bars denote the standard error across cells (n = 35 in area 3b, 81 in area 1, and 25 in area 2). B| Ratio of timing performance / rate performance for the same cells. Same conventions as in panel A. Dashed line represents equal performance for rate-and timing-based classifiers. Points above the line denote neurons for which timing outperforms rate.
Supplemental Figure 11: Cross-speed classification based on both rate and timing outperforms classification based on rate. Mean cross-speed population classification based on rate (unwarped, red), timing (warped, blue), and the combination of both (purple). As a control, we averaged rate over a population twice the size (large population rate classification, maroon) to test the extent to which the improved performance with rate and timing was simply driven by more predictors. Performance from the large population never exceeds that of rate and timing. Shaded regions denote standard deviation across 1000 iterations of randomly sampled populations of 49 neurons.