Neural coding of temperature with a DNA-based spiking chemical neuron

Complex organisms perceive their surroundings with sensory neurons that encode physical stimuli into spikes of electrical activities. The past decades have seen a throve of computing approaches taking inspiration from neurons, including reports of DNA-based chemical neurons that mimic artificial neural networks with chemical reactions. Yet, they lack the physical sensing and temporal coding of sensory biological neurons. Here we report a thermosensory chemical neuron based on DNA and enzymes that spikes with chemical activity when exposed to cold. Surprisingly, this chemical neuron shares deep mathematical similarities with a toy model of a cold nociceptive neuron: they follow a similar bifurcation route between rest and oscillations and avoid artefacts associated with canonical bifurcations (such as irreversibility, damping or untimely spiking). We experimentally demonstrate this robustness by encoding—digitally and analogically—thermal messages into chemical waveforms. This chemical neuron could pave the way for implementing the third generation of neural network models (spiking networks) in DNA and opens the door for associative learning.