Abstract
Our sense of touch confers to us the ability to perceive textural features over a broad range of spatial scales and material properties, giving rise to a complex sensory experience. To understand the neural basis of texture perception requires that the responses of somatosensory neurons be probed with stimuli that tile the space of spatial scales and material properties experienced during everyday interactions with objects. We have previously shown that neurons in early stages of somatosensory processing – the nerves and somatosensory cortex (S1) – are highly sensitive to texture and carry a representation of texture that is highly informative about the surface but also predicts the evoked sensory experience. In contrast, the texture signals in higher order areas – secondary somatosensory cortex (S2) and the parietal ventral area (PV) – have never been investigated with a rich and naturalistic textural set. To fill this gap, we recorded single-unit activity in S2/PV of macaques while they performed a texture discrimination task. We then characterized the neural responses to texture and compared these to their counterparts in somatosensory cortex (S1). We found that the representation of texture in S2/PV differs markedly from its counterpart in S1. In particular, S2/PV neurons carry a much sparser representation of texture identity and also information about task variables, including the animal’s eventual perceptual decision. S2/PV thus seems to carry a labile representation of texture that reflects task demands rather than faithfully encoding the stimulus.
Introduction
Neuroscientists have established that the primary somatosensory cortex (S1) acts as a high-fidelity recorder of tactile information, providing a detailed map of the world by encoding textures through both spatial patterns and precise temporal vibrations. However, the subsequent hierarchy of processing—specifically the secondary somatosensory cortex (S2) and the parietal ventral area (PV)—has remained relatively unexplored regarding complex, naturalistic textures. This study addresses whether S2/PV simply refines the sensory map provided by S1 or if it transforms sensory data into a fundamentally different representation that incorporates cognitive and task-related variables.
Methods
Single-unit activity was recorded from 158 neurons in the S2/PV regions of two Rhesus macaques using multi-contact linear arrays. The monkeys performed a "same-different" discrimination task where a rotating drum scanned two textures sequentially across their fingertip. Following the stimulus presentation, the monkeys signaled via a saccade whether the two textures were identical. The stimulus set comprised 44 diverse natural and artificial textures, including fabrics, furs, and 3D-printed patterns. This data was compared to a reference dataset of 141 neurons in S1 and human psychophysical ratings of perceived roughness.
Results
From Promiscuous to Sparse Coding
A defining difference between the two regions is the selectivity of individual neurons. S1 neurons are "texture promiscuous," typically responding to almost every texture in the set. In contrast, S2/PV neurons exhibit sparse coding, with most neurons firing only for a small subset (<20%) of textures. This suggests a shift toward high selectivity for specific, complex features rather than general physical properties.
Multiplexing and Loss of Physical Fidelity
While S1 is a strong predictor of perceived "roughness," S2/PV is a poor predictor of this basic physical attribute. Instead, S2/PV neurons appear to multiplex sensory data with cognitive information. Approximately 54% of S2/PV neurons encoded the trial type (same vs. different), and 26% encoded the animal’s upcoming decision. This information emerged shortly after the second stimulus, indicating a "labile" representation that shifts from stimulus-driven to decision-driven. Furthermore, S2/PV largely abandons the precise temporal spike timing used by S1 to encode fine textures, relying instead on simpler firing rate codes.
Discussion
The representation of texture in S2/PV differs markedly from S1, moving away from a faithful sensory recording toward a sparse, task-dependent representation. While S2/PV is significantly less informative about raw texture identity—requiring 100 neurons to reach only 40% decoding accuracy compared to near-perfect accuracy from 25 S1 neurons—it provides a higher-dimensional space that integrates sensory input with behavioral context. A limitation of the study was the monkeys' difficulty with the task (55-69% accuracy), which may have influenced the strength of task-variable representation. Future work aims to define the specific "high-level" features that S2/PV neurons prioritize over basic physical properties like roughness.
Summary updated