Quantification and statistical analysis

Membrane potential recordings were made in current clamp mode sampled at 10 kHz. Data were analyzed offline using Axograph and custom scripts in Igor Pro and MATLAB. Action potentials (spikes) were detected using custom scripts in Igor Pro using Neuromatic v2.6i plug-in. A first derivative transformation was performed on the membrane potential trace defining dV/dt. Then a constant dV/dt threshold was defined (for an individual neuron)—peaks above the threshold defined a single spike Peristimulus time histograms (PTH) of firing rate were made by binning detected spikes in 1 s bins, defining spikes/s (Hz). Traces depicting mean and line and shading indicate the mean firing rate (Hz) ± SEM. For the dose-response curves of TLHON firing rate vs. temperature derivative, spikes were binned into 25 ms bins, and the spike count per bin was used to calculate the firing rate. For the corresponding stimulus sweep, the peak value of the temperature derivative, dT/dt, was detected, and a time interval where dT/dt was at least 90% of its maximum or minimum was defined. The firing rate within this 90% sampling interval was detected and averaged across the sampling interval to give the cell’s firing rate at the time when the rate of temperature change was fastest. The pre-stimulus baseline firing rate was defined as the mean firing rate in the 5 s immediately preceding the stimulus onset. Temperature recordings were digitized at 10 kHz and smoothed with a 1000 point span. Smoothed traces were then differentiated and smoothed with a 10,000 point span to generate temperature rate traces where the line and shading indicate the mean rate (°C/s) ± SEM. To test for a significant relationship between temperature and steady state firing rate for TLHONs (Fig. 4), a 1-way ANOVA and test for multiple comparisons were used to determine significance; an asterisk means p < 0.05. To test for a relationship between the rate of temperature change and firing rate (Fig. 4) for TPN-III and TLHONs, data were fit with linear functions, and shading indicates the 95% CI. For TLHONs, the fit only includes the portion of the graph containing responses (firing rates that that are above the background firing rate). To determine whether slow thermal change (<0.25 °C/s) elicits responses in TLHONs and TPN-IIIs (i.e., increases in spiking rate from baseline, Fig. 4i,j), background subtracted firing rates at peak stimulus were extracted and binned. A 1-sample t-test was used to test if increases in firing rates are different from zero (n.s.= not different from zero, asterisk = different from zero; p < 0.05).

To test for differences in behavior between controls and silenced flies, a 2-way ANOVA was used (p < 0.05; Figs. 1–3, and Supplementary Fig. ³). Attraction/Avoidance indexes for each genotype were compared by two-way ANOVA, and asterisks denote a statistically significant interaction between the Gal4 or LexA driver and the UAS or LexAop effector (see figure for the specific N of biological replicates). Kolmogorov–Smirnov tests were used to confirm a normally distributed sample. Homogeneity of variance for each data set was confirmed using Levene’s test (threshold p < 0.05). To test for the role of TLHON in turning behavior for single flies (Fig. 5), the ratios of turns vs crosses was computed and analyzed using a GLMM (Generalized Linear Mixed Models) with fly ID as a random effect and Wald testing to determine significance (threshold P = 0.05). To compare the occurrence of early vs late turns, we used a Chi-square test to determine whether bins contained equal counts (p < 0.05). See Supplementary Data ¹ for precise statistical information on Figs. 1–5, ​,77 and Supplementary Fig. ³.