Fig. 4: lPBN modulation by competing needs is consistent with a gating of nociceptive input.

a, Schematic of the model control system for regulating behaviour during pain. Nociceptive input is integrated into an internal ‘pain’ state, and past actions (licks) are integrated over recent history by an internal ‘effort’ state. The model’s learned policy maps pain and effort states to a generated behaviour (lick or don’t lick), which in turn modifies the value of the two states. We study the effect of competing needs by manipulating three different parts of the model (input integration, effort and policy (dotted arrows)). b, Time spent licking paw in 5-min bins of 8 trained models compared with observed mouse behaviour. c,d, Time spent licking paw in phase 1 (c) and phase 2 (d) (n = 8 model runs, n = 9 observed mice). e, Simulated paw-licking behaviour of eight trained models, comparing control with a competing need that reduces nociceptive integration. f,g, Simulated time licking paw in phase 1 (f) and phase 2 (g) in the models (n = 8 model runs per condition; unpaired two-sided t-test, not significant (f), P < 0.001 (g)). h,i, Average dynamics of the pain (h) and effort (i) states of eight trained models, comparing control with a competing need that reduces nociceptive integration. j,k, Simulated paw-licking behaviour (j) and average pain state dynamics (k) of eight trained models, comparing control with a competing need that increases the effort cost of licking. l,m, Simulated paw licking behaviour (l) and average pain state dynamics (m) of eight trained models, comparing control with a competing need that extends the dimension of the policy. Data are mean ± s.e.m. Grey dots represent individual mice or individual model runs.

Source data