Disentangling Cephalopod Chromatophores Motor Units with Computer Vision
Cephalopod chromatophores are skin pigment organs that enable unmatched camouflage through rapid, flexible and neurally controlled deformation. Although their morphology is well known, the organization of their motor control is not entirely understood. Here, we combine high-resolution videography with a dedicated computer-vision pipeline (CHROMAS) to investigate chromatophore control and their likely innervation in Euprymna berryi and Sepia officinalis. By segmenting chromatophores into radial slices and analyzing anisotropic deformations, we applied dimensionality reduction (PCA) and source separation (ICA) to estimate the number and spatial influence of motor neurons responsible for the control of individual and groups of chromatophores. On average, four independent components were detected (suggesting innervation by four motor neurons), each forming contiguous petal-shaped domains rather than causing uniform expansion. Clustering thousands of components revealed motor units spanning multiple chromatophores, most involving fewer than 14 but occasionally spanning more widely. These motor units displayed a wide variety of geometries, ranging from compact local groups to elongated or fragmented structures; they often overlapped, with repeated co-innervation of chromatophore pairs occurring more often than expected by chance. Expansion was consistently faster and more stereotyped than relaxation, consistent with active contraction (corresponding to chromatophore expansion) and passive recoil (chromatophore contraction). Together, these results show that individual chromatophores are not singular or uniform pixels, but rather contrast elements that can be fractionated into smaller territories, themselves coordinated with those of other chromatophores. This geometry of neural control enables, among others, the generation of "virtual" chromatophores, i.e., functional groupings of adjacent chromatophore territories that act as single units, as well as that of noise in the distribution of pixel shapes.