Fig. 8: A generalized linear model confirms enhanced conjunctive coding in NAc-projecting neurons.

a Schematic of the generalized linear model (GLM) (left). Example modeling approach for a triple conjunctive neuron (right). Upper traces show true calcium activity (downsampled and normalized) of test dataset (thin black line) as well as predictions of original model (thick gray) and one example for each shuffled feature. Bottom histograms show R² distributions for 100 shuffled models for each feature. Dotted line represents value of 95^th percentile, thick black line represents original model’s R². b Increased proportions of dHPC^→NAc neurons modulated by position (χ²(1, 5372) = 93.634, P < 0.001), velocity (χ²(1, 5372) = 141.86, P < 0.001), and licking (χ²(1, 5372) = 10.050, P = 0.0015). c Increased proportions of conjunctive coding in dHPC^→NAc neurons for position & velocity (χ²(1, 5372) = 163.97, P < 0.001), position & licking (χ²(1, 5372) = 26.029, P < 0.001), velocity & licking (χ²(1, 5372) = 27.145, P < 0.001), and position & velocity & licking (χ²(1, 5372) = 34.993, P < 0.001). d Venn diagrams showing overlap of neurons GLM-classified as modulated by position, velocity and licking in dHPC^- (top) and dHPC^→NAc (bottom) neurons. e Proportions of n-feature coding neurons in dHPC^- (left) and dHPC^→NAc (right) populations. Non-coding neurons are overrepresented in dHPC^- neurons (χ²(1, 5372) = 35.382, P < 0.001); single-coding neurons are comparably distributed (χ²(1, 5372) = 0.0057, P = 0.940); dual-coding (χ²(1, 5372) = 61.336, P < 0.001) and triple-coding (χ²(1, 5372) = 30.447, P < 0.001) neurons are overrepresented in dHPC^→NAc neurons. f, g A linear classifier to decode the presence of the reward zone. f Example true presence of reward zone (thin line) and decoder predictions based on non-conjunctive neurons (pink; top) and conjunctive-coding neurons (dark purple; bottom). g Conjunctive-coding neurons allow a linear decoder to classify the presence of reward zone more accurately than non-conjunctive coding neurons (Wilcoxon’s W(18) = 14.0, P(two-tailed) <0.001, n = 19 imaging sessions). The box represents quartiles and whiskers represent outlier-corrected minima and maxima of the distribution. Statistical tests used in b, c, e are chi-squared tests. **P < 0.01, ***P < 0.001. Source data are provided as a Source Data file.