Figures and data
Stimulation-evoked activity of ON-, OFF-, and NEUTRAL-cells in the RVM.
a, Schematic of the RVM and its descending projections to the spinal dorsal horn, illustrating the modulatory roles of ON- and OFF-cells in nociceptive processing. b, Representative spike rasters (top) and rate plots (bottom) from simultaneously recorded ON-, OFF-, and NEUTRAL-cells during repeated noxious heat trials. ON- and OFF-cells exhibit the canonical reciprocal dynamics around paw withdrawal (solid green line), whereas NEUTRAL-cells show no clear modulation. Notably, alternating dominance of ON- and OFF-cells is also observed during inter-trial intervals, suggesting ongoing, stimulus-independent dynamics. c, Population peri-stimulus time histograms (PSTHs) for ON- and OFF-cells, aligned either to the time of paw withdrawal (top) or to the heat stimulus onset (bottom). Population firing is sharply time-locked to the withdrawal reflex but not to the onset of heat application. Withdrawal-aligned PSTHs show steeper slopes and no delay compared to heat-aligned PSTHs, consistent with the nocifensive response, rather than the stimulus as such, being the primary driver of RVM activity.
Bayesian modeling of ON- and OFF-cell population dynamics during noxious stimulation.
a, The probabilistic model used to describe withdrawal-aligned population firing rates. b, Candidate response–recovery functions per cell type. For ON-cells, single vs. double exponential (exp.) recovery models were compared; for OFF-cells, double exp. vs. linear + exp. recovery models were evaluated. c-f, Models were fitted to the population firing rates per cell type and model variant. The single exp. recovery model missed the sharp ON-cell peak (e) and the double exp. model underestimated the OFF-cell recovery rate (f). g-h, Expected Log Predictive Density (ELPD) comparison of models, confirming that the double exp. recovery best fit ON-cells, while the linear + exp. model best fit OFF-cells. Whilst the linear + exp. model performed best for OFF-cells, the improvement over the double exp. model was minimal. i, Combined ON- and OFF-cell model fits around the withdrawal and population data in gray (left). Posterior switching timepoint estimates (ts) per cell type (right), showing that ON-cells peak slightly before withdrawal, while OFF-cells exhibit post-withdrawal suppression.
Ongoing activity of RVM neurons reveals distinct low-frequency signatures in ON- and OFF-cells.
a, Example spike rasters (top) and rate plots (bottom) from simultaneously recorded ON-, OFF-, and NEUTRAL-cells during unstimulated periods. ON- and OFF-cells exhibit irregular, burst-like activity with slow co-fluctuations, whereas NEUTRAL-cells fire regularly at a high rate. b-d, Population-averaged power spectra of ongoing activity for ON-, OFF-, and NEUTRAL-cells. ON- and OFF-cells show clear low-frequency peaks just below T = 300 s (5 min), with associated harmonics, while NEUTRAL-cells lack such structure and instead show strong high-frequency components (10–100 Hz). e, Coefficient of variation (CV) versus mean interspike interval (ISI) for all recorded cells. NEUTRAL-cells form a distinct cluster characterized by low CV and short ISIs (regular high-rate firing), whereas ON- and OFF-cells show broad CV distributions and longer ISIs, consistent with irregular burst dynamics.
Period-length scans of the negative log marginal likelihood (NLML) from periodic Gaussian process fits to ongoing firing in ON-, OFF-, and NEUTRAL-cells.
ON- and OFF-cells show distinct minima at 300 s and its harmonics, whereas NEUTRAL-cells exhibit flat profiles.
Periodic Gaussian process (GP) modeling of ongoing ON-, OFF-, and NEUTRAL-cell activity.
a, Example fits for single ON-, OFF-, and NEUTRAL-cells using a periodic GP kernel. GP fits were robust to noise or occasional periods of silence in the training data (e.g. ON-cell trace, 410-550 s). Models were trained on all data except the final 300 s (training region: solid line) and tasked with predicting the held-out test segment (dotted line), without using test data during fitting. b, Pseudo-R2 values for training and test sets. ON-cells achieve significant predictive performance on test data, OFF-cells show weak predictability, while NEUTRAL-cells fail to generalize (test Pseudo-R2<0). c, Posterior hyperparameters across cell classes. ON- and OFF-cells exhibit positive dispersion parameters α and tightly clustered period estimates near 300 s, consistent with quasi-periodic firing. NEUTRAL-cells show negative dispersion and less coherent period estimates, reflecting their regular, high-rate activity. d, Phase aligned GP rates adjusted for found period length showed unique signatures for each cell.
Posterior hyperparameters from the fitted latent GP model.
Analysis of heart rate coherence against cell firing rate.
a, Example normed firing rates and heart rates for ON-, OFF-, and NEUTRAL-cells, showing both visibly coherent and incoherent activity. b, Coherence against frequency for the same cells in (a), showing the max-statistic significance level (black line). c, Total number of significant coherences per frequency for all cells. d, Combined averaged heart rate power spectrum over all animals, showing clear heartbeat-related peaks > 1 Hz as well as low frequency peaks. Lowest frequency peak indicated in red.
