Parvalbumin-expressing basal forebrain neurons mediate learning from negative experience

Basal forebrain (BF) nuclei are classically defined by cholinergic projection neurons, but GABAergic neurons—including a prominent parvalbumin-expressing (PV) population—are far more numerous and project widely to cortical and subcortical targets. Prior work implicated BFPVNs in regulating cortical gamma oscillations and arousal, and non-selective BF lesions cause learning and attention deficits greater than cholinergic-selective lesions. Non-cholinergic BF neurons can respond phasically to salient stimuli independent of valence. These observations raised the question whether BFPVNs contribute to awake behaviors, particularly associative learning. The authors hypothesized that BFPVNs encode outcome-related signals—especially aversive feedback—and support learning of cue–outcome associations.

The study builds on evidence that: (1) BFPVNs promote arousal and modulate cortical gamma rhythms; (2) degeneration of BF GABAergic (including PV) neurons correlates with cognitive decline in Alzheimer’s disease and aging; (3) broader non-cholinergic BF populations respond to salient stimuli irrespective of valence and predict decision speed; (4) BF GABAergic lesions impair associative learning (e.g., eyeblink conditioning), whereas cholinergic lesions alone do not fully recapitulate deficits. The authors contrast BFPVNs with cholinergic BF neurons (BFCNs), previously shown to encode prediction-related signals for both CS and US, and with somatostatin-expressing BF neurons (BFSOMNs), whose activity profile during behavior was less characterized. This literature motivates investigating whether BFPVNs form a specialized aversive-outcome signaling pathway important for associative learning.

• Subjects: Adult male PV-IRES-Cre mice for in vivo and anatomical experiments; SOM-IRES-Cre mice for BFSOMN photometry controls; PV-IRES-Cre males (P30–60) for acute slice electrophysiology.
• Behavioral task: Head-fixed auditory probabilistic Pavlovian conditioning with two tones predicting different outcome probabilities. Cue 1: 80% reward (water), 10% punishment (air puff), 10% omission. Cue 2: 65% punishment, 25% reward, 10% omission. Variable foreperiod (1–4 s) and short variable delay (200–400 ms) before outcomes.
• Electrophysiology with optotagging: AAV2/5 EF1a DIO-ChR2-eYFP was injected into HDB of PV-Cre mice; movable tetrode microdrives with an optical fiber were implanted. Extracellular single-unit recordings in HDB during task; optogenetic tagging with 1 ms, 473 nm pulses at 20 Hz; SALT test (p<0.01) for identification. n=36 optotagged BFPVNs across n=4 mice; histological verification of recording sites.
• Bulk calcium imaging (fiber photometry): AAV2/9 CAG Flex GCaMP6s injected bilaterally into HDB of PV-Cre mice to record somatic signals and axonal signals in projections (MS, hippocampal CA1, retrosplenial cortex). Dual-wavelength demodulation (465/405 nm), dF/F computation and Z-scoring; recordings during task and during exposure to aversive stimuli (air puff, foot shock, fox odor). SOM-Cre cohorts were also used for BFSOMN photometry comparison.
• Optogenetic manipulation: Conditioned place aversion (CPA) with ChR2 activation of BFPVNs (20 Hz, 473 nm) in one side of a two-chamber arena (ChR2 n=8; eYFP control n=10). Optogenetic inhibition during punishment: ArchT-GFP expressed bilaterally in HDB (n=6 ArchT; n=8 GFP controls); 593 nm continuous 1 s laser starting at air puff onset during Pavlovian training; analysis of anticipatory licking and reaction times.
• Anatomical circuit mapping: Mono-transsynaptic rabies tracing in PV-Cre mice (n=3) to map inputs to HDB BFPVNs (Cre-dependent helper virus + EnvA-pseudotyped, G-deleted rabies). Cell-type identification in MRR using immunostaining for vGluT3 and 5-HT. Anterograde tracing with AAV2/8 or 2/9 CAG Flex eGFP (n=3) to visualize projection targets and estimate axon density. Correlated light and electron microscopy to confirm synaptic contacts onto identified target cell types (PV+, CR+, ChAT+).
• In vitro slice electrophysiology: Channelrhodopsin-assisted circuit mapping (PV-Cre, ChR2-mCherry in HDB projections; n=7 mice). Whole-cell voltage-clamp recordings in MS, CA1, and retrosplenial cortex. Optogenetically evoked IPSCs characterized for amplitude, kinetics, and short-term plasticity (5–40 Hz trains); pharmacology with gabazine to confirm GABAergic transmission.
• Aversive stimuli assays: Presentation of mild foot shocks (0.5–1 mA, 200 ms) and predator odor (2‑methyl‑2‑thiazoline) in naïve GCaMP mice; analysis of first-exposure responses.
• Statistics: Non-parametric tests (Mann–Whitney U, Wilcoxon signed-rank), Pearson correlations, linear regression, ROC analysis with permutation tests for licking differences; PETHs and Z-scoring procedures detailed.
• Sample sizes/examples: Identified BFPVNs n=36; photometry HDB somata sessions n=19; projections MS n=8 sessions, CA1 n=9, RSC n=5; rabies tracing n=3; anterograde tracing n=3; opto inhibition sessions: controls n=34, ArchT n=29; CPA ChR2 n=8, control n=10; BFSOMN photometry in SOM-Cre cohort; aversive stimulus photometry foot shock n=4 mice, fox odor n=4.

• BFPVNs encode aversive outcomes phasically: • In HDB, optotagged BFPVNs showed gradual ramp-like cue responses (33%, n=12/36) and small but significant reward responses (39%, n=14/36), but strong, fast phasic responses to punishment (75%, n=27/36), largely homogeneous across cells. Responses to reward and punishment were not modulated by surprise, unlike BFCNs. • Punishment responses adapted across the session but remained present; peak punishment response decreased from first to second half (Wilcoxon signed-rank p=0.00458; n=27), with baseline firing also changing (p=0.00071). • Unbiased clustering of n=685 HDB neurons revealed two clusters with strong punishment and smaller reward responses encompassing most identified BFPVNs (29/36, 80%). A distinct suppressed-after-reinforcement group (14%) and a putative cholinergic cluster were also identified. • About half of punishment-responsive BFPVNs were burst-firing; bursters were more likely punishment-activated (94%, 17/18) than non-bursters (56%, 10/18).
• Multimodal aversive responsiveness: Bulk calcium signals in HDB BFPVNs were robustly activated by air puff, foot shock, and predator (fox) odor, including at first exposure, indicating generalization across sensory modalities.
• Causal role in associative learning, not direct aversion: • Optogenetic activation of HDB BFPVNs did not induce conditioned place aversion or other stress-like behaviors compared to controls (no side preference or behavioral differences). • Optogenetic inhibition (ArchT) of BFPVNs during air puff impaired learning of cue contingencies: ArchT mice failed to develop differential anticipatory licking to Cue 1 vs Cue 2 in a 0.6–1.1 s window before reinforcement (controls: significant difference, Wilcoxon p=0.001517; ArchT: n.s., p=0.40513). ROC analysis showed a significant group difference over ~0.6–1.1 s post-cue. • Reaction time advantage for Cue 1 over Cue 2 present in controls (Wilcoxon p=0.000472) was abolished in ArchT mice (n.s., p=0.15668); cumulative RT distributions differed between groups.
• Input–output architecture: • Retrograde rabies tracing (n=3) revealed dense inputs from lateral hypothalamus (LH, 33.1% of inputs), preoptic area (7.6%), lateral septum (13.8%), medial septum (10.8%), VDB (6.4%), nucleus accumbens (3.5%), and median raphe region (MRR; highest from brainstem, 3.9%). Unilateral tracing showed predominantly ipsilateral inputs (93.7%). • In MRR, most input cells to BFPVNs were vGluT3+ glutamatergic (59%); a small fraction were also serotonergic (vGluT3+/5‑HT+, 2% of MRR inputs). No 5‑HT‑only inputs detected; remaining inputs included vGluT2 and possibly GABAergic neurons. • Anterograde tracing (n=3) showed major projections to MS/VDB, hippocampus (CA1), retrosplenial cortex (RSC), with additional innervation of limbic structures and paratenial thalamus; sparse projections to medial orbital/infralimbic cortices, lateral septum, and MRR. • Synaptic targets confirmed by correlated light and EM and by in vitro recordings: BFPVNs formed symmetric synapses onto MS ChAT+ and PV+ neurons; onto CA1 PV+ and CR+ interneurons; and onto PV+ neurons in RSC.
• Broadcast of aversive signals: • Fiber photometry of BFPVN axons showed phasic punishment responses in MS, CA1, and RSC, paralleling somatic responses, indicating widespread broadcast of negative outcome signals. • Functional synaptic properties differed by target: RSC IPSCs were smaller and slower with short-term facilitation; MS and CA1 IPSCs were larger/faster with strong frequency-dependent short-term depression. Between-area differences reached statistical significance for some comparisons (e.g., smaller amplitudes in RSC vs MS, p=0.017; latency differences including RSC vs CA1, p=0.030).

The results demonstrate that HDB parvalbumin-expressing GABAergic neurons respond preferentially and phasically to aversive outcomes, generalize across modalities, and causally support learning of cue–outcome contingencies. Unlike BFCNs, BFPVNs did not encode outcome prediction error or surprise effects, and they differed from BFSOMNs, which showed relatively larger reward-related responses and expectation modulation. The lack of conditioned place aversion during BFPVN activation, together with impaired associative learning when BFPVNs were inhibited selectively during punishments, supports a cognitive/arousal role rather than direct aversive motor output control. Anatomical mapping indicates that BFPVNs integrate inputs from aversion-coding regions (LH, MRR) and affective/motivational centers (lateral septum, nucleus accumbens), and send disinhibitory outputs across the limbic system (MS, hippocampus, RSC). Projection photometry showed similar punishment responses across targets, consistent with a broadcast mechanism, while slice physiology revealed target-specific synaptic dynamics that may shape context-dependent impact. Collectively, BFPVNs likely enhance attention/arousal for learning from negative outcomes by rapidly disinhibiting key limbic circuits, complementing cholinergic mechanisms of plasticity.

This work identifies HDB BFPVNs as a long-range inhibitory population that broadcasts phasic aversive-outcome signals across the limbic system and is necessary for forming stimulus–outcome associations in a probabilistic Pavlovian task. The study delineates their multimodal sensitivity to aversive stimuli, specialized input patterns from aversion-related nuclei, broad projections with defined postsynaptic targets, and target-specific synaptic dynamics. These findings advance understanding of how basal forebrain inhibitory circuits contribute to learning from negative experience and attentional/arousal processes. Future work could dissect dominance and dynamics among convergent inputs (e.g., LH vs MRR), resolve potential direct contacts onto pyramidal neurons in cortex/hippocampus, quantify projection-specific axon densities vs response strength in vivo, and examine state-dependent modulation and topographic organization across the HDB to frontal cortical targets.

• Punishment responses adapted within sessions; decreased magnitude could reflect reduced aversiveness, novelty, motivation, or fatigue, which were not fully disentangled.
• Clustering indicated many non-tagged neurons with similar profiles; some may be other HDB types or undetected BFPVNs, limiting precise cell-type attribution across the full population.
• Projection photometry revealed qualitative similarities across targets, but differences in amplitude could not be separated from differences in axon density vs local processing.
• EM and slice data confirmed disinhibitory targets (e.g., PV+, CR+, ChAT+), but potential direct synapses onto pyramidal neurons were not reliably tested.
• Input dominance among LH, MRR, septal, and striatal sources remains unresolved; local HDB circuitry contributions were not exhaustively mapped.
• Experiments were conducted in male mice; sex differences were not assessed.
• Conditioned place aversion assay used specific stimulation parameters; other patterns might yield different behavioral outcomes.