Feature-specific threat coding in lateral septum guides defensive action

a, Representative expression of Flp-tdTomato in hippocampal (vCA1, vCA3, vSub) and hypothalamic (AHA, LHAsf, SuM) afferents following CVS-N2cΔG-Flp-tdTomato rabies virus injection into LS of Crhr2-Cre mice. Scale bar, 500um. b, Schematic illustrating the fibre photometry approach for recording calcium dynamics. c, Representative image of GCaMP expression and optical fibre placement in the SuM. Scale bar, 500um.d, Example calcium transient from hypothalamic SuM afferents recorded via fibre photometry. e–g, Representative image (e, Scale bar, 500um) and average calcium activity responses (f–g) to CS and shock events in anterior paraventricular thalamus (aPVT), n = 4 mice. h–i, Average calcium activity aligned to CS-Fail trials and shock across (h) hypothalamic (AHN, LHAsf, SuM) and (i) hippocampal (vCA1, vCA3, vSub) afferents. j–o, GLM analysis quantifying variance explained (R², j) and β coefficients for responses to (k) CS-Success, (l) CS-Failure, (m) anticipatory Av-run, (n) Av-run onset, and (o) shock across afferent regions. One sample t-tests. p, Heatmap depicting relative contributions to explained variance (R²) from behavioral events across afferent inputs, depicting stronger contributions of specific variables depending on subtype. For example, aPVT to Av-run onset, vCA1 to CS-S, vCA3 to CS-F and shock, AHA to CS-S and shock, LHAsf to anticipatory Av-run and shock, and SuM to both CS and shock. vCA1, n = 7 mice; vCA3, n = 5 mice; vSub, n = 5 mice; AHN, n = 3 mice; LHAsf, n = 10 mice; and SuM, n = 15 mice. q, Strategy for rabies-assisted bilateral optogenetic inhibition of LSCrhr2 afferents. Crhr2-Cre mice received helper AAVs in LS to permit retrograde N2c-medated expression of Flp for expression of stGtACR2 (FLEx(FRT)-stGtACR2-eGFP) or eGFP control virus selectively in LSCrhr2-projecting neurons from ventral hippocampus (vHPC), lateral hypothalamus spiny/feeding zone (LHAsf), or supramammillary nucleus (SuM). r, Schematic of photoinhibition timing during active-avoidance (−1 s to shuttle crossing) on Test day. s–v, Behavioral effects of pathway-specific photoinhibition. s, Probability of avoidance (P(avoidance)) across sessions for control (Ctrl) and each input (RM two-way ANOVA, Interaction: P = 0.0021; Bonferroni multiple comparisons post-hoc tests; GFP, Pre vs. Test: P = 0.5974; vHPC, Pre vs. Test: P = 0.0697; LHAsf, Pre vs. Test: P = 0.0147; GFP, Pre vs. Test: P = 0.5561). Two circles = mean, lines = paired sessions per mouse; bars = mean ± SEM. t, Session-wise change in avoidance probability (ΔP(avoid); Test–Pre). Inhibition of vHPC and LHAsf significantly reduces avoidance (One-way ANOVA, P = 0.0067; Dunnett’s multiple comparisons test: GFP vs. vHPC: P = 0.305; GFP vs. LHAsf: P = 0.0039; GFP vs. SuM: P = 0.5268). u, Avoidance latency across sessions. SuM inhibition increases latency to avoid (RM two-way ANOVA, Interaction: P = 0.0821; Bonferroni multiple comparisons post-hoc tests; GFP, Pre vs. Test: P = 0.2868; vHPC, Pre vs. Test: P = 0.9532; LHAsf, Pre vs. Test: P = 0.5305; GFP, Pre vs. Test: P = 0.0020). v, Change in latency (Δ latency) (One-way ANOVA, P = 0.1982; Dunnett’s multiple comparisons test: GFP vs. vHPC: P = 0.7605; GFP vs. LHAsf: P = 0.2317; GFP vs. SuM: P = 0.1349). w–x, Speed aligned to the avoidance shuttle. w, Pre (day 1) and x, Test (day 2) mean speed traces (shading, SEM) for each inhibited pathway in their respective colors. Thick lines indicate significant differences between input and GFP (P < 0.05, permutation test. y, Working model: vHPC provides information about the predictive context of the cue-outcome association, SuM conveys arousal/salience signals critical for avoidance responses, and LHAsf contributes motivational drive for avoidance. LSCrhr2 neurons integrate and transform these inputs to evaluate threat and invigorate defensive actions. *P < 0.05, **P < 0.01, ***P < 0.001,****P < 0.0001.