Cholinergic Reinforcement Signaling Is Impaired by Amyloidosis Prior to Its Synaptic Loss

Abstract

Alzheimer's disease (AD) is associated with amyloidosis and dysfunction of the cholinergic system, which is crucial for learning and memory. However, the nature of acetylcholine signaling within regions of cholinergic-dependent plasticity and how it changes with experience is poorly understood, much less the impact of amyloidosis on this signaling. Therefore, we optically measure the release profile of acetylcholine to unexpected, predicted, and predictive events in visual cortex (VC)—a site of known cholinergic-dependent plasticity—in a preclinical mouse model of AD that develops amyloidosis. We find that acetylcholine exhibits reinforcement signaling qualities, reporting behaviorally relevant outcomes and displaying release profiles to predictive and predicted events that change as a consequence of experience. We identify three stages of amyloidosis occurring before the degeneration of cholinergic synapses within VC and observe that cholinergic responses in amyloid-bearing mice become impaired over these stages, diverging progressively from age- and sex-matched littermate controls. In particular, amyloidosis degrades the signaling of unexpected rewards and punishments, and attenuates the experience-dependent (1) increase of cholinergic responses to outcome predictive visual cues, and (2) decrease of cholinergic responses to predicted outcomes. Hyperactive spontaneous acetylcholine release occurring transiently at the onset of impaired cholinergic signaling is also observed, further implicating disrupted cholinergic activity as an early functional biomarker in AD. Our findings suggest that acetylcholine acts as a reinforcement signal that is impaired by amyloidosis before pathologic degeneration of the cholinergic system, providing a deeper understanding of the effects of amyloidosis on acetylcholine signaling and informing future interventions for AD.

SIGNIFICANCE STATEMENT The cholinergic system is especially vulnerable to the neurotoxic effects of amyloidosis, a hallmark of Alzheimer's disease (AD). Though amyloid-induced cholinergic synaptic loss is thought in part to account for learning and memory impairments in AD, little is known regarding how amyloid impacts signaling of the cholinergic system before its anatomic degeneration. Optical measurement of acetylcholine (ACh) release in a mouse model of AD that develops amyloidosis reveals that ACh signals reinforcement and outcome prediction that is disrupted by amyloidosis before cholinergic degeneration. These observations have important scientific and clinical implications: they implicate ACh signaling as an early functional biomarker, provide a deeper understanding of the action of acetylcholine, and inform on when and how intervention may best ameliorate cognitive decline in AD.

Introduction

The cholinergic system, which has been implicated in a number of cognitive processes (Everitt and Robbins, 1997; Lin et al., 2015; Disney and Higley, 2020) including arousal (Teles-Grilo Ruivo et al., 2017; Lohani et al., 2022), attention (Herrero et al., 2008; Parikh and Sarter, 2008), sensory modulation (Rodriguez et al., 2004; Alitto and Dan, 2013; Pinto et al., 2013; Eggermann et al., 2014), and learning and memory (Bakin and Weinberger, 1996; Hasselmo, 2006; Kang and Vaucher, 2009; Jiang et al., 2016; Crouse et al., 2020), is especially susceptible to the neurotoxic effects of amyloidosis (Bell et al., 2006)—a hallmark of Alzheimer's disease (AD; Selkoe, 2000). The importance of acetylcholine (ACh) to cognition and vulnerability to amyloidosis intersect two once dominant theories of AD—the cholinergic hypothesis (Francis et al., 1999), and the amyloid hypothesis (Selkoe, 2000)—which have waned as treatments either elevating basal levels of acetylcholine or clearing amyloid have met with modest cognitive benefit. We posit that approaches elevating cholinergic tone (Becker et al., 2008) or clearing amyloidosis (van Dyck et al., 2023) have yet to achieve their full promise because of our incomplete understanding of (1) the role of ACh in cognition and (2) the damage caused to the cholinergic system by nascent amyloidosis, occurring at a time well preceding typical clinical intervention. Improving our understanding of the impact of emerging amyloidosis on the signaling profile of acetylcholine, therefore, informs on the presumed limitations of these interventional approaches and may impart renewed interest in the interaction of amyloid and acetylcholine in AD.

As cholinergic neurons are particularly vulnerable to amyloid β (Aβ) peptides (Perry, 1980; Whitehouse et al., 1981; Lehéricy et al., 1993; Bruno et al., 2009; Knowles et al., 2009; Liu et al., 2015b), we reasoned that brain areas of cholinergic-dependent learning and memory are key sites to investigate the consequences of amyloidosis on cholinergic signaling. Though higher-order brain areas associated with decision-making have been intensively investigated as they relate to AD, there has been a call for greater focus on the consequences of AD in sensory areas (Albers et al., 2015). One such site is the primary visual cortex (VC; Bear and Singer, 1986; Gavornik and Bear, 2014), where pairing visual stimuli with subsequent reward has been shown to lead to the emergence of activity that predicts the time of future reward (Shuler and Bear, 2006; Zold and Hussain Shuler, 2015; Shuler, 2016; Monk et al., 2020) and informs the timing of reward-seeking actions (Namboodiri et al., 2015; Levy et al., 2017). Learning relationships between visual cues and what they portend is not a property unique to the visual cortex of rodents, as it is observed in human visual cortex (Serences, 2008; Cravo et al., 2013; Salvioni et al., 2013), where, specifically, visually cued interval timing activity has also been reported (Yu et al., 2022).

How such timing activity in VC can be learned is described in a formal model (Gavornik et al., 2009; Huertas et al., 2015; Namboodiri et al., 2015), which proposes the presence of a reinforcement signal conveying the receipt of reward to visual cortex. Our research subsequently identified basal forebrain cholinergic innervation as the source of this reinforcement signal, demonstrating that it is both necessary (Chubykin et al., 2013) and sufficient (Liu et al., 2015a) to teach VC to produce visually cued interval timing. These observations contribute to a growing body of evidence that the cholinergic system responds to behaviorally relevant outcomes, including aversive as well as appetitive events (Lin and Nicolelis, 2008; Hangya et al., 2015; Monosov et al., 2015), exhibits reinforcement signaling qualities (Kuchibhotla et al., 2017; Guo et al., 2019; Crouse et al., 2020; Hegedüs et al., 2023), and broadcasts that information widely throughout the cortical mantel (Mesulam et al., 1983; Rieck and Carey, 1984; Semba et al., 1988; Dinopoulos et al., 1989; Robertson et al., 1990). While these findings advanced the case that acetylcholine acts to signal reinforcement, the dynamics of acetylcholine release to outcome and outcome-predictive cues within cortical sites of cholinergic-dependent learning is yet to be adequately characterized.

Given the above observations, we hypothesized that cholinergic signaling within VC will report outcomes and exhibit features indicative of reinforcement signaling, and that this signaling will progressively be impaired by nascent amyloidosis occurring before cholinergic degeneration. To test this hypothesis, we expressed a genetically encoded acetylcholine sensor (Jing et al., 2020) in VC of a mouse model of preclinical AD that develops amyloidosis (Ferretti et al., 2011). We then measured cholinergic responses to behaviorally relevant outcomes and to outcome-predictive cues at various stages of amyloidosis preceding cholinergic degeneration. By comparing cholinergic responses to age- and sex-matched WT littermate controls, we dissociate the effects of emerging amyloidosis from that of aging and other presumptive drivers of AD.

Materials and Methods

All procedures reported here are in accordance with the United States National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals and were approved by the Johns Hopkins University School of Medicine Animal Care and Use Committee (approval #MO20M293). In accordance with NIH policy, experiments were conducted in both female and male mice.

Animals

The McGill-Thy1-APP mouse (APP mouse) is a preclinical mouse model of AD designed to display a slowly progressing amyloidosis by overexpressing the human amyloid precursor protein (APP) with Swedish and Indiana mutations responsible for many cases of familial AD (Ferretti et al., 2011). These mice gradually accumulate Aβ in neurons, followed by extracellular accumulation of Aβ, and develop cognitive impairments that worsen as the amyloid pathology progresses (Ferretti et al., 2011). Selection of this model mouse was made on the following three qualifying characteristics met uniquely by the McGill Thy-1 APP mouse: (1) specificity to human APP (hAPP) with familial AD mutations, (2) slow progression, and (3) intact vascularization. McGill-Thy1-APP mice were kept hemizygous by breeding wild-type (WT) C57BL/6J females with homozygous McGill-Thy1-APP males. Offspring were genotyped (Ferretti et al., 2011), and negative littermates were used as age-matched controls for the transgenic mice from each litter. Animals were housed socially by sex in a temperature- and humidity-controlled vivarium on a 12 h light/dark cycle and had access to food and water ad libitum until recruited for experiments. To prevent genetic drift from the background, every five generations WT C57BL/6J females were obtained from The Jackson Laboratory and used for breeding. Otherwise, negative littermate females from our colony were paired with positive males. Seventy-two mice were used for fiber photometry, being allocated to appetitive (36 mice) and aversive (36 mice) cohorts equally by genotype with the following age and sex (male/female) distribution: 3 months of age, 10/6; 4.5 months, 6/4; 6 months, 6/4. All APP mice (n = 36) used in photometry were used for biochemical characterization of pathologic stages of amyloidosis.

Surgical procedures and habituation

Mice of the appropriate age were taken from the main colony and brought to our satellite vivarium where they were individually housed during experimentation. Surgical procedures were separated in the following two steps: (1) headbar implant and skull mapping, and (2) sensor infection and fiber optic implant. For all surgeries, mice were anesthetized with a ketamine/xylazine cocktail, delivering 100 mg/kg ketamine and 10 mg/kg xylazine. Mice were administered a presurgical dose of dexamethasone (5 mg/kg) to prevent inflammation and lidocaine at the wound margin (7 mg/kg, s.c.), and received buprenorphine postsurgically for analgesia (0.1 mg/kg).

Headbar implant and skull mapping

The skin over the parietal and frontal bones was excised, and the periosteum was removed with 3% hydrogen peroxide. The area above VC was marked stereotaxically at 3 mm lateral to λ and covered with a silicone cap for future craniotomy. The rest of the skull was covered with cyanoacrylate glue (Gel Super Glue; catalog #1363589, Loctite), and a custom stainless steel headbar was fixed above the frontal bones. Implanted mice were left to recover for 4 d and then habituated to head fixation. Mice received two habituation sessions for each of 3 d, with head fixation times increasing gradually across sessions (5, 10, 15, 20, 30, 40 min).

Once habituated to head fixation, mice underwent the second surgical procedure for sensor infection and fiber optic implant. Following anesthesia, mice were placed in a stereotaxic frame (Kopf Instruments) that was modified to accommodate headbar fixation. The silicone cap above VC was removed, and a small craniotomy was made with a dental drill. Three hundred nanoliters of undiluted acetylcholine sensor (AAV9 hSyn ACh3.0 GRAB sensor, also known as ACh4.3; genomic titer, ≥1 * 10¹³ genome copies/ml; WZ Biosciences) was slowly injected (Nanoject, Drummond Scientific) unilaterally into VC (anteroposterior, 0.00 mm; mediolateral, ±3 mm; dorsoventral, 0.30 mm; relative to λ) over 5 min, with the injection pipette being left in place for 10 min before removal. A fiber optic cannula (200 μm, NA 0.37; Neurophotometrics) was then lowered to the brain surface using a 20° tilt on the stereotaxic arm. Small adjustments were made to the angle of approach so that the fiber touched the dura perpendicularly. The fiber optic was then secured to the headcap with cyanoacrylate glue. Animals were left to recover for 4 d before pretraining.

Pretraining

Mice in the appetitive conditioning cohort were placed on water restriction for 2 d before commencing pretraining and were weighed daily, with supplemental water given if their weight dropped below 85% of their fully hydrated weight. Once water was restricted, mice commenced pretraining, where they were head fixed within the behavioral rig containing a solenoid-controlled water spout, and exposed to sham and rewarded sessions.

Rewarded sessions.

Mice were presented with rewarded trials during which a 1.5 μl water reward was delivered following a random intertrial interval (ITI) varying between 3 and 10 s.

Sham sessions.

Mice were left in the experimental setup for 20 min, and no behavioral stimuli were delivered. Aversive cohort mice only experienced sham sessions at this stage. For both appetitive and aversive cohorts, fluorescence signals were recorded in every session, with the next stage of training being initiated on the third day of stable δF/F signals.

Unexpected visual cues and reward, appetitive cohort

Mice received two sessions per day: one visual stimulus (VS) session and one rewarded session. During each session, mice received 100 stimulus trials (75%) and 33 sham trials (25%), interleaved randomly. Trials were separated by a random ITI varying between 3 and 10 s. For VS trials, a short 100 ms binocular flash of green light was delivered with custom made LED goggles, followed by a period of 3 s where nothing happened (see Fig. 4, schematic 1). During sham trials, nothing happened for 3 s (see Fig. 4, schematic 1). Reward sessions were similar in structure to VS trials except that a 1.5 μl water reward replaced the visual stimulus. This stage of training lasted 3 d.

Figure 4.

Effect of appetitive trace conditioning on evoked cholinergic transients. Three-month-old WT mice were conditioned to associate a VS and a water reward delayed by 1.5 s. A, Three stages of conditioning. B, Example heat maps and PSTHs showing session examples of cholinergic responses to stimuli across trace conditioning. C–E, Quantification of the trial-averaged cholinergic response (C), the evoked response magnitude (D), and the probability (E) of cholinergic transients to the conditioned stimulus (visual cue, green) and the unconditioned stimulus (water, blue) across days of conditioning. Following ANOVAs (Table 1), stars indicate significance (Tukey's test) from initial responses to CS and US only cues.

Unexpected visual cues and shocks, aversive cohort

The first 3 d contained only one VS session, while the first shock session was administered on day 4 (see Fig. 5, schematic 1). Shock sessions were similar to reward and VS sessions except that a mild tail shock (10 μA for 100 ms) was delivered (Monk et al., 2021) as a stimulus (see Fig. 5, schematic 1).

Figure 5.

Effect of aversive trace conditioning on cholinergic signaling. Three-month-old WT mice were conditioned to associate a VS and a mild tail shock delayed by 1.5 s. A, Four stages of conditioning. B, Example heat maps and PSTHs showing session examples of cholinergic responses to stimuli across trace conditioning. C–E, Quantification of the trial-averaged cholinergic response (C), the evoked response magnitude (D), and the probability (E) of cholinergic transients to the conditioned stimulus (visual cue, green) and the unconditioned stimulus (shock, magenta) across days of conditioning. Following ANOVAs (Table 1), stars indicate significance (Tukey's test) from initial responses to CS- and US-only cues.

Appetitive and aversive trace conditioning

Appetitive and aversive trace conditioning was conducted in two stages. During the first stage, mice were presented with paired trials and sham trials. During paired trials, the VS preceded the reward or shock by 1.5 s. During the second stage, catch trials were introduced where rewards or shocks were omitted following the VS. During the first stage, mice received 75% paired trials and 25% sham trials. During the second stage, mice received 50% paired trials, 25% catch trials, and 25% sham trials. Trial numbers were adjusted so that mice always received at least 100 paired trials per session.

Fiber photometry

Fluorescent signals were acquired using a photometry system (system #FP3002, Neurophotometrics). Sampling frequency was set at 90 Hz and contained three stages per cycle, so that each signal was acquired at 30 Hz. The first stage was 470 nm illumination, followed by 415 nm (isosbestic), then a blank stage where no LED was illuminated. The separation of illumination minimizes bleed-through, while the blank stage reduces total illumination time and photobleaching. The system was turned on and allowed to run for 15 min before the first recording session of the day so that the LEDs stabilize and minimize signal fluctuations. LED powers were kept at 1% for both 470 and 415 nm for an output of ∼15 mW at the fiber tip. A “Serial String read” node was added to the Bonsai workflow, and all behavioral events were relayed by serial communication from the Arduino-controlled behavioral setup so that events and photometry signals were time stamped with the same clock. To normalize the signal, compensating for bleaching and motion artifacts, each 470 nm frame was subtracted by its corresponding 415 nm frame before δf/f calculation. Baseline fluorescence value (f_b) was calculated on isosbestic-subtracted data averaging the 100 lowest value frames in a recording session. For each timeframe, the δf/f value was calculated as (f_x – f_b)/f_b, where f_x is the isosbestic-subtracted signal fluorescence value at any time frame.

Detection and quantification of cholinergic transients

Cholinergic transients were detected and quantified using normalized δf/f values. For each recording session, a detection threshold was first established to separate noise-related fluctuations from acetylcholine-mediated fluctuations. To establish a threshold value, the δf/f signal for the full session was ordered in ascending order of signal intensity and plotted. Ordered signal traces typically show two slopes. A first milder slope covers data points of lower value and represents background. The remainder of data points follow a higher slope and represent transient-related fluctuations. A value >95% of background data points is selected as the threshold for the session. Transient magnitude was calculated using the trapezoidal method (trapz MATLAB function) to approximate the area under the curve (AUC) from the first to the last point of the transient. The first point included for quantification is the last point below threshold, preceding a series of at least three consecutive suprathreshold points. The last point included for quantification is the first point back to subthreshold values. Transients initiating within 200 ms of stimulus onset were considered “evoked.” Transients initiating outside of the stimulus time envelope were considered “spontaneous.” Because of the presence of the unconditioned stimulus (US) in trace conditioning, the AUC of conditioned stimulus (CS) evoked transients during paired trials were never quantified beyond 1.5 s after stimulus onset. For quantification of spontaneous ACh transient magnitude, ACh transients detected within this window were realigned to their start times before averaging. The probability of evoked transients was calculated by dividing the number of transients that initiated within 200 ms of stimulus presentation by the number of presentations. For the probability of spontaneous transients, the total number of transients occurring during the ITI periods and sham trials was divided by the total time spent in ITI and sham trials, then multiplied by 200 to have a value corresponding to the probability in a 200 ms time bin. For trial-averaged cholinergic response quantification, trials were first aligned by time to stimulus presentation (time 0). Following alignment, the signal between time 0 and 1 s was averaged across trials. The AUC of the averaged trace represents the trial-averaged cholinergic response. While evoked-transient magnitude only considers time envelopes following stimuli that contain a transient, the trial-averaged cholinergic response accounts for all presentations, including trials without a detected transient. Sham trials were used to calculate the stimulus energy of spontaneous transients.

Animal inclusion criteria

Following each pretraining session, δf/f values were calculated and visualized. When transients emerged from the background signal, quantification of AUCs was performed for the session. From that moment on, the mean transient AUCs were plotted every day. To be included for group comparisons of transient properties, the average, an animal's average AUC of spontaneous transients had to be within a stable δf/f plateau for twelve days (Fig. 1B). Animals that either did not exhibit transients (n = 55) or showed transients at some point but did not meet stability criteria (n = 58) were excluded.

Figure 1.

Cholinergic responses to unexpected visual cues, rewards, punishments. A, Example fluorescence micrograph showing expression of the ACh3.0 GRAB sensor following immunohistochemical detection. Scale bar, 200 μm. V2MM, Secondary visual cortex, mediomedial area. B, Following viral transduction, fluorescence signals were recorded by fiber photometry and the average AUC of spontaneous transients was calculated. Over sessions, the mean values increased and then remained stable across the course of experimentation. The green line at time 0 represents the first day included for group comparisons, while the red line indicates the last. A three-way ANOVA, with repeated measures for time, showed no effect of genotype (F_(1,206) = 1.1, p = 0.34), pathological stage (F_(2,206) = 0.19, p = 0.95), or experimental day (F_(7,206) = 0.91, p = 0.41) on average spontaneous transients. C–G, Unexpected VSs, rewards (water droplets), or punishments (mild tail shocks) were presented to 3-month-old WT mice during independent sessions, intermixed with sham events (sham). C, Example session heat map and trial-averaged cholinergic responses for each event type: sham, visual cue (VS), reward (water), and punishment (shock). Time zero aligned to the time of stimulus presentation. D, Example trace from a single trial. E–G, The population average across the 3-month-old WT cohort of the trial-averaged cholinergic response (E), the magnitude of detected ACh transients (F), and the probability of evoking ACh transients (G) for sham, VS, reward, or punishment events. Color-coded stars indicate significance from groups to the left. Significance was calculated with a Tukey's test following a significant group effect by one-way ANOVA (Table 1).

Tissue processing

Following behavioral conditioning and fiber photometry signal acquisition, mice were overdosed with pentobarbital and perfused transcardially with cold saline. The brains were dissected, and the hemisphere with the fiber implant was immersed in 4% PFA for 48 h at 4°C, while the contralateral hemisphere was snap frozen in liquid nitrogen and kept for ELISAs.

ELISAs

Soluble and insoluble Aβ-40 and Aβ-42 levels were quantified using ELISA kits (EZBRAIN-SET, Sigma-Aldrich) following manufacturer instructions. Weighed VC samples were homogenized in RIPA buffer with protease inhibitors (provided with the EZBRAIN-SET) with a handheld homogenizer. Lysates were centrifuged at 13,000 rpm at 4°C for 10 min. Supernatants containing the soluble fraction were transferred to a tube, and proper weight-based dilutions were confirmed by bicinchoninic acid assay (BCA Assay, Sigma-Aldrich). Pellets containing the insoluble fraction were resuspended in 2× volume of 70% formic acid, sonicated for 10 min, and resuspended in 15× volume of 1 m Tris, pH 7.4. Resuspended samples were centrifuged at 13,000 rpm at 4°C for 10 min, and the supernatants were transferred to a new tube and used as samples for ELISA. Standards and samples were loaded on the plate in duplicate and incubated overnight at 4°C. On the next day, the plates were washed 5× with washing solution, incubated for 1 h with the detection antibodies provided, and then washed 5×. The substrate solution was added to the wells and allowed to develop for 15 min on a plate shaker. Stop solution was added to the wells, and absorbance was read at 450 and 590 nm on a plate reader (model ELx808, BioTek). The difference in absorbance units was logged. The standard curve was fitted linearly and used to calculate Aβ concentration in averaged sample duplicates.

Immunofluorescence

Perfused brains were immersion fixed in 4% PFA in PBS for 48 h at 4°C. The brains were then transferred to a 30% sucrose in PBS solution and left to equilibrate for 48 h. Brain hemispheres were cut 50 μm thick and collected in PBS. Alternate sections of VC were allocated for ACh3.0 and vesicular acetylcholine transporter (VaChT) immunostaining. The former was used to verify sensor expression, the latter to quantify the density of cortical cholinergic varicosities (Allard et al., 2012b). Both stains were done using the same protocol. Sections were washed 3 × 10 min in PBS and then blocked for 1 h in 10% normal goat serum (NGS) in PBS+T (PBS with Triton X-100 at 0.1%). Following the blocking step, the sections were incubated with the primary antibody diluted in 5% NGS in PBS+T overnight on a plate shaker at 4°C. The primary antibodies were either a chicken anti-GFP (used 1:1000; Aves Labs) or a rabbit anti-VaChT (used 1:2000; catalog #139103, Synaptic Systems) to stain for the sensor and cholinergic axons, respectively. Following primary antibody incubation, the sections were washed 3 × 10 min with PBS, then incubated with the secondary antibodies diluted 1:500 in 5% NGS in PBS for 2 h. The secondary antibodies used were a goat anti-rabbit Alexa Fluor 594 (catalog #111-585-144, Jackson ImmunoResearch) or a goat anti-chicken Alexa Fluor 488 (catalog #111-545-045, Jackson ImmunoResearch) for the VaChT and sensor staining, respectively. Sections were then washed 3× in PBS, mounted on glass slides and coverslipped with Fluoromount-G (Thermo Fisher Scientific).

Slide imaging and quantification

Immunofluorescence for the sensor signal was imaged with a fluorescence microscope (model BZ-X800, Keyence) equipped with the appropriate filter cubes and a 10× dry objective. For imaging cortical cholinergic varicosities, a microscope (model LSM 880, Zeiss) was used. An Objective Plan-Apochromat 40×/1.4 Oil DIC M27 (Zeiss) was used to make 20-μm-thick z-stacks centered around layers II/III (supragranular) and layers V/VI (infragranular). Three stacks were imaged for both infragranular and supragranular regions per section, and four sections containing VC were imaged per mouse. The Alexa Fluor 594 signal was imaged on a track, and a control track for Alexa Fluor 488 was added. The control track was used to assess lipofuscin autofluorescence signal. To control for autofluorescence and unspecific binding of the secondary antibody, control sections were included for each animal. Sections used as controls were not incubated with the primary antibody but were otherwise treated like the experimental tissue, undergoing blocking steps and secondary antibody incubations. Experimental and control sections were imaged using fixed laser and detector gain settings. Quantification of confocal micrographs was performed with ImageJ (NIH). First, the 20-μm-thick stack was projected on a single z-plane using a maximum intensity projection. A signal intensity threshold was automatically determined using the “Triangle” algorithm. Following thresholding, a size criterion excluding all structures <0.3 μm² was used as an additional filter. A binary mask image of the structures passing criteria was generated for quality control, and the number of immunoreactivity structures was calculated. Importantly, negative controls were analyzed with the same method and failed to produce varicosity counts.

Experimental design and statistical analyses

Sample size was estimated before conducting the study based in part on prior reports of cholinergic signaling detected by the indicator used (Jing et al., 2020) and by factoring the number of planned comparisons. ACh signaling was compared by ANOVA across experimental factors (genotype: WT vs APP; pathologic stage: early, mid, late) and by mixed-effects ANOVA with days of conditioning as the repeated measure (as in appetitive and aversive trace conditioning). Following the determination of the significant F-ratio in global ANOVAs, Tukey's HSD was used for post hoc testing where appropriate. Significance annotations were as follows: *p < 0.05, **p < 0.01, ***p < 0.001. Summary statistics are reported as the mean ± SEM unless otherwise noted. Perievent time histograms show the mean and SE. Analyses and graphing were conducted in MATLAB, and the figure layout in Adobe Illustrator.

Data availability

Long-term archiving of biochemistry and fiber photometry data will be managed by Johns Hopkins Data Services using the Johns Hopkins Research Data Repository. The data and analysis code is available at https://doi.org/10.7281/T1/AVOEQR.

Results

Rewards and punishments evoke rapid responses (∼20 ms) from cholinergic neurons in the basal forebrain (Hangya et al., 2015), which provide the major source of acetylcholine to the cerebral cortex (Mesulam et al., 1983; Rye et al., 1984; Woolf, 1991; Parikh and Sarter, 2006). Acetylcholine activation and release in the rodent visual cortex has been observed following visual stimulation using microdialysis (Laplante et al., 2005) as well as under various locomotive (Reimer et al., 2016; Jing et al., 2020) and behavioral (Lohani et al., 2022; Collins et al., 2023) states using optical sensors. However, neither the release of acetylcholine in VC in response to rewarding and punishing events nor the effect of associative learning on the response profile of acetylcholine has been previously reported. To characterize acetylcholine release in VC, we infected mice with a genetically encoded acetylcholine sensor (Jing et al., 2020) via an AAV9 viral delivery system. This approach led to the reliable expression of the sensor in VC (Fig. 1A,B), where background fluorescence levels did not differ significantly between groups or timepoints (two-way ANOVA, effects of genotype and pathologic stage; Appetitive cohort: genotype: F_(1,24) = 0.36, p = 0.55; stage: F_(2,24) = 0.11, p = 0.89; interaction: F_(2,24) = 0.04, p = 0.96; Aversive cohort: genotype: F_(1,24) = 0.50, p = 0.48; stage: F_(2,24) = 0.31, p = 0.73; interaction: F_(2,24) = 1.25, p = 0.30). This sensor expression afforded a means to assess cholinergic responses to behaviorally neutral (VS; binocular 100 ms light flash), appetitive (1.5 μl water rewards), and aversive (mild tail shock, 10 μA) events, occurring unexpectedly. Additionally, by pairing visual cues with future reward or punishment occurring at a delay, as in appetitive and trace conditioning, cholinergic responses to predictive visual cues and to predicted outcomes could also be measured, thus testing the impact of associative learning on evoked cholinergic response profiles. Mice habituated to head fixation were surgically infected to express the acetylcholine sensor and implanted supradurally with an optic fiber above VC. Mice (n = 72; 36 male/female) were first presented with sessions of unexpected VSs while cholinergic responses in VC were recorded by fiber photometry, followed by sessions of unexpected reward in one cohort (n = 36) or unexpected punishment in another cohort (n = 36). These cohorts were then subsequently imaged under consecutive days of appetitive or aversive trace conditioning, respectively.

Cholinergic responses to unexpected, behaviorally neutral, appetitive, and aversive stimuli

Unexpected visual stimulation, rewards, and punishments evoked cholinergic responses within the VC of 3-month-old WT mice well in excess of spontaneous events occurring in sham trials (Fig. 1C,E–G, Table 1, Figure 1 ANOVA). To quantify these responses, we subtracted the isosbestic control channel and analyzed the δf/f signal, taking the following measurements: (1) the trial-averaged cholinergic response; (2) the magnitude of detected cholinergic transients; and (3) the probability of evoking cholinergic transients (see Materials and Methods). Unexpected rewarding and punishing events elicited robust and comparable trial-averaged cholinergic responses (Fig. 1C,E). Though modest compared with these behaviorally relevant events, the unexpected VS also elicited a cholinergic response (Fig. 1C,E). As the trial-averaged cholinergic response is a product of both the magnitude of an evoked cholinergic transient as well as its probability of being elicited, we assessed these response features independently. The magnitude of detected cholinergic transients evoked by visual stimulation was significantly greater than the magnitude of detected spontaneous cholinergic transients but significantly smaller than transients evoked by rewards or by punishments (Fig. 1F). Similarly, the probability of a visually evoked cholinergic transient (36%) was significantly greater than that of spontaneous activity (2.6%), yet significantly smaller than that evoked by either reward (87%) or punishment (88%; Fig. 1G).

Table 1.

The effect of aging and amyloidosis on cholinergic responses to unexpected, predictive, and predicted events

		Trial-averaged response		ACh magnitude		ACh probability
		F	p	F	p	F	p
		Figure 1, one-way ANOVA, effect of stimulus type on transient properties (df = 2, n = 5)
Transient type		132.233649	6.69E-09	162.8270654	3.37E-12	198.6308108	6.35E-10
		Figure 3, two-way ANOVAs, effects of genotype and stage of emerging amyloidosis on cholinergic response (n = 36)
Spontaneous	Genotype (df = 1)	1.702356374	2.04E-01	1.168929541	0.29036604	1.121160445	3.00E-01
	Stage (df = 2)	3.816132142	3.64E-02	0.216570743	0.806832125	3.514928926	4.58E-02
	Interaction	5.697483133	9.45E-03	0.165277743	0.848614813	6.322376366	6.23E-03
VS evoked	Genotype (df = 1)	0.151206714	7.01E-01	1.31122727	0.263463699	0.467567988	5.01E-01
	Stage (df = 2)	2.219579022	1.30E-01	2.114599755	0.14261071	1.132213298	3.39E-01
	Interaction	0.027994814	9.72E-01	1.005331447	0.380818339	0.157096096	8.55E-01
Reward evoked	Genotype (df = 1)	42.36156119	9.89E-07	72.2792485	1.06E-08	9.451553637	5.20E-03
	Stage (df = 2)	8.370933496	1.75E-03	10.23596403	6.10E-04	6.118530434	7.12E-03
	Interaction	9.16799811	1.10E-03	12.02530079	2.41E-04	10.39175654	5.61E-04
Shock evoked	Genotype (df = 1)	85.67892398	2.17E-09	61.00672683	4.81E-08	9.487381258	5.13E-03
	Stage (df = 2)	34.11350015	9.64E-08	25.44910314	1.17E-06	5.947072936	7.98E-03
	Interaction	47.30551758	4.71E-09	26.67584016	7.96E-07	14.5198378	7.37E-05
		Figure 4, repeated-measures ANOVA, effect of time (n = 5)
VS evoked	Effect of time (df = 4)	63.91705197	4.43E-13	18.3589272	1.66E-07	69.2514684	1.82E-13
Reward evoked	Effect of time (df = 4)	17.63836242	2.39E-07	13.1512496	3.17E-06	5.356820958	1.91E-03
		Figure 5, repeated-measures ANOVA, effect of time (n = 5)
VS evoked	Effect of time (df = 4)	180.0095153	2.23E-15	102.1927133	5.14E-13	121.851042	9.60E-14
Shock evoked	Effect of time (df = 4)	19.09280727	1.32E-06	15.8619951	5.38E-06	5.616265123	3.38E-03
		Figure 7, three-way ANOVA, effects of genotype, stage and training day (n = 36)
VS evoked	Genotype (df = 1)	39.25773983	3.55E-09	16.38713947	8.15E-05	23.7809998	2.67E-06
	Stage (df = 2)	59.97060957	5.49E-20	6.207094486	2.56E-03	65.5995228	2.47E-21
	Day (df = 4)	438.356798	3.98E-89	89.76103085	7.81E-44	411.4638627	3.72E-87
	Genotype * stage	55.46258443	7.22E-19	6.088303895	2.85E-03	61.43498386	2.42E-20
	Genotype * day	1.764704314	1.23E-01	2.984436936	1.34E-02	1.375470579	2.36E-01
	Stage * day	2.747764945	3.84E-03	1.982986415	3.86E-02	2.437317842	1.00E-02
Reward evoked	Genotype (df = 1)	176.2067772	2.72E-27	253.2037763	2.47E-34	15.09062604	1.52E-04
	Stage (df = 2)	54.09143019	1.61E-18	40.83088684	5.93E-15	38.90166519	2.11E-14
	Day (df = 4)	60.73844111	1.07E-34	32.7113047	1.23E-22	26.86732928	1.80E-19
	Genotype * stage	66.63631575	1.41E-21	40.52272379	7.25E-15	50.97583688	1.03E-17
	Genotype * day	6.13418844	3.27E-05	2.556612215	2.97E-02	3.966338455	2.06E-03
	Stage * day	5.249978247	1.26E-06	6.053975438	9.88E-08	1.178328945	3.09E-01
		Figure 8, three-way ANOVA, effects of genotype, stage, and training day (n = 36)
VS evoked	Genotype (df = 1)	84.03408423	1.04E-15	102.1874769	5.10E-18	0.826499497	3.65E-01
	Stage (df = 2)	53.02684271	1.68E-17	33.01306799	2.74E-12	51.92895776	3.07E-17
	Day	672.7955997	7.02E-85	203.2195363	2.02E-54	803.8968637	1.28E-89
	Genotype*Stage	55.95974247	3.45E-18	47.31233884	4.14E-16	34.56235142	9.95E-13
	Genotype*Day	6.426979494	9.58E-05	12.31585788	1.67E-08	1.164653415	3.30E-01
	Stage*Day	2.671713041	9.53E-03	3.347600722	1.63E-03	2.646625744	1.02E-02
Shock evoked	Genotype (df = 1)	179.3110289	4.15E-26	223.7627353	7.00E-30	9.034818985	3.19E-03
	Stage (df = 2)	98.14279311	1.45E-26	92.98078186	1.15E-25	24.19207741	1.22E-09
	Day (df = 3)	76.79160203	4.49E-33	47.52639321	1.95E-24	24.31849321	5.56E-15
	Genotype*Stage	117.9325805	9.15E-30	84.67022125	3.74E-24	46.88631216	5.29E-16
	Genotype*Day	6.663530369	6.65E-05	4.350529714	2.48E-03	2.10193956	8.43E-02
	Stage*Day	2.484876816	1.54E-02	2.620926744	1.09E-02	0.293971292	9.67E-01

Table of ANOVAs. Figure 1, One-way ANOVAs for effect of unexpected stimuli (VS, reward, punishment) on the trial-averaged ACh response, its magnitude, and probability. Figure 3, Two-way ANOVAs assessing spontaneous, VS-, reward-, and shock-evoked ACh responses (trial average, magnitude, and probability). Figure 4, Repeated-measures ANOVAs for VS- and reward-evoked ACh responses in 3 male WT mice across days of appetitive conditioning. Figure 5, Repeated-measures ANOVAs for VS- and punishment-evoked ACh responses in 3 m WT mice across days of aversive conditioning. Figure 7, Three-way mixed-effect ANOVA with day of appetitive conditioning as the repeated measure, contrasting genotype (WT vs APP) and pathologic stage (early, mid, late), for VS- and reward-evoked ACh response metrics. Figure 8, Three-way mixed-effect ANOVA with day of aversive conditioning the repeated measure, contrasting genotype (WT vs APP) and pathologic stage (early, mid, late), for VS- and punishment-evoked ACh response metrics. Significant p-values in bold.

Early, mid, and late stages of emerging amyloidosis before cholinergic synapse degeneration

As mice do not develop amyloidosis natively and are amenable to genetic manipulation, mice present an opportunity to test presumptive drivers of AD, such as familial AD-related mutations of amyloid precursor protein that give rise to amyloidosis, in dissociation from other putative causes. To determine the effect of amyloidosis on cholinergic signaling, we therefore compared responses from the APP mouse model of preclinical AD to those of their WT controls. The McGill-Thy1-APP mouse was selected as it overexpresses the hAPP, with Swedish and Indiana mutations responsible for many cases of familial AD (Ferretti et al., 2011), which leads to a slowly progressing amyloidosis affording the study of functional changes in cholinergic signaling that precede synaptic loss (Ferretti et al., 2011). Significant cholinergic synaptic loss is known, however, to occur by 12 months of age (Ferretti et al., 2011), which we determine to be the case in VC of 12 male APP mice, observing 28% fewer cholinergic varicosities in infragranular layers in APP than in WT mice (p = 0.0083, Tukey's test), following significant two-way ANOVA [effects of genotype: F_(1,16) = 25, p = 0.00,013; cortical layer (supragranular and infragranular): F_(1,16) = 31, p = 0.000041; interaction: F_(1,16) = 20, p = 0.00,043]. Therefore, in our effort to assess functional changes in male and female mice during nascent amyloidosis, before pathologic degeneration of the cholinergic system, we selected 3, 4.5, and 6 months as time points of interest. To characterize the progression of amyloidosis and to ascertain whether there was any impact on the number of cholinergic synapses over these timepoints within VC, we quantified the amounts of Aβ-40 and Aβ-42 peptides from soluble and insoluble fractions by sandwich ELISA (Englund et al., 2007), and counted cortical cholinergic varicosities in supragranular and infragranular layers of VC by imaging VaChT immunoreactivity under confocal microscopy (Allard et al., 2011, 2012a,b; Ferretti et al., 2011, 2012).

As expected, Aβ gradually accumulates, as observed in significant increases in soluble and insoluble Aβ-40 and Aβ-42 peptide concentration, as animals age. By plotting soluble against insoluble levels of Aβ, three clusters are readily identified, which correspond to the age of the animal, except for 3-month-old females who exhibited an advanced amyloidosis consistent with 4.5-month-old animals (Fig. 2A). The age and sex distribution of these groups was confirmed by k-means clustering. Using these clusters, we then assessed whether there was any difference in the density of cholinergic varicosities between APP and WT sex- and age-matched controls and found that there was, importantly, no detectable difference between cohorts, and no cholinergic synaptic loss across the early, mid, and late stages (Fig. 2B, Table 2, ANOVAs and post hoc test). Using these clusters, we sorted mice into corresponding early, mid, and late stages of nascent amyloidosis preceding cholinergic degeneration.

Figure 2.

Early, mid, and late stages of amyloidosis preceding cholinergic synapse degeneration. Clustering of Aβ-40 and Aβ-42 affords the identification of an early, mid, and late stage of accumulation that precedes cholinergic synapse degeneration. A, Aβ-40 (left) and Aβ-42 (right) expression levels from 3-, 4.5-, and 6-month-old APP mice. The x-axes and y-axes represent insoluble and soluble levels, respectively. B, Quantification of visual cortex cholinergic varicosities per 1.8 × 10⁶ µm³ by supragranular and infragranular layers in APP and WT mice. Density of cholinergic varicosities is unchanging with age and does not significantly differ between WT and APP mice (Table 2, ANOVAs and post hoc tests).

Table 2.

Cholinergic synaptic density in infragranular layers and supragranular layers does not significantly differ between WT and APP mice across stages

	F	p
Figure 2B, three-way ANOVA effects of genotype, stage, and cortical layers
Genotype (df = 1)	0.30	0.59
Stage (df = 2)	0.00	1.00
Layer (df = 1)	433.33	5.97E-40
Genotype * stage	0.84	0.43
Genotype * layer	1.84	0.18
Stage * layer	4.14	0.02

Comparison	p Values infragranular	p Values supragranular
Figure 2B, Tukey's post hoc test results, comparisons between genotypes
Early, APP * WT	0.577652913	0.872421701
Mid, APP * WT	0.402601188	0.482321323
Late, APP * WT	0.489947204	0.483833254

Figure 2B, Three-way ANOVA. Quantification of cortical cholinergic varicosities per 1.8 × 10⁶ µm³ by supragranular and infragranular layers in APP and WT mice across stages. Figure 2B, Tukey's post hoc tests. Density of cortical cholinergic varicosities does not differ between WT and APP mice within any stage.

Effect of amyloidosis on cholinergic responses to unexpected events

Having classified amyloidosis before cholinergic degeneration into early, mid, and late stages, we next sought to identify the impact of amyloidosis on cholinergic signaling over its progression (Fig. 3, Table 1, Figure 3 for ANOVAs). To do so, cholinergic responses to unexpected visual, reward, and punishment events were imaged in age- and sex-matched WT and APP mice at these stages.

Figure 3.

Effect of amyloidosis on spontaneous and evoked cholinergic transients. The effect of amyloidosis on cholinergic signaling is assessed across early, mid, and late stages preceding cholinergic degeneration by comparing responses to spontaneous, and behaviorally neutral, appetitive, and aversive events between APP and WT mice. A–C, Quantification across the APP (darkened colored bars) and WT (lightened colored bars) cohorts of trial-averaged cholinergic responses (A), cholinergic transient magnitude (B), and cholinergic transient probability (C), for each event type over early, mid, and late stages. D, Example sessions at the late stage show typical responses for WT (bottom row) versus APP (top row) mice to unexpected visual (left column), reward (middle column), and shock (right column) cues, exemplifying the decrease of evoked cholinergic response caused by amyloidosis to behaviorally relevant appetitive and aversive, but not behaviorally neutral, cueing. Stars (A–C) indicate significant differences within a stage of amyloidosis for APP mice and their WT age- and sex-matched controls; post hoc Tukey's test following significant ANOVAs (Table 1).

Recordings from WT mice determined the stability of cholinergic responses to these events across aging in healthy animals, exhibiting markedly consistent responses at the population level at early, mid, and late stages (Fig. 3A–C, lightened colored bars). Reward-evoked responses (Fig. 3D, top middle), for example, remain constant across stages (Fig. 3A, light blue) and are significantly greater than VS-evoked (Fig. 3D, top right, light green) or spontaneous ACh (light purple) responses, and do not exhibit a difference in either their evoked magnitude (Fig. 3B) or probability (Fig. 3C) of response. The same holds for cholinergic responses to punishment (light magenta), VS (light green), and spontaneous (light purple) events (Fig. 3A–D,F).

We next compare the cholinergic responses of WT (Fig. 3A–C, lightened colored bars) to APP (Fig. 3A–C, darkened colored bars) across the three stages. Before reporting measures of the strength of response, however, first we note that the latencies to VS, reward, and punishment, which are short and exhibit low variability (respectively: 135 ± 21, 139 ± 20, and 119 ± 19; latencies in milliseconds ±SD), do not differ by genotype, stage, or their interaction (three-way ANOVA of genotype, stage, and stimulus; effect of genotype: F_(1,8293) = 0.460, p = 0.50; pathologic stage: F_(2,8293) = 2.217, p = 0.11; genotype × stage: F_(2,8293) = 0.236, p = 0.79). Punishment does, however, elicit response at a lower latency than VS or reward (punishment vs VS, p = 0.002; punishment vs reward, p = 0.0003).

At the early stage, the strength of response to VS, rewards, and punishments as well as spontaneous events are indistinguishable between WT and APP animals (Fig. 3A–C). By the mid stage, however, APP mice begin to exhibit deficits in their cholinergic responses to behaviorally relevant rewards (Fig. 3A–C, dark blue) and punishments (Fig. 3A–C, dark magenta), though not, notably, to VS (Fig. 3A–C, dark green). At the mid stage, cholinergic deficits are characterized by significant reductions in the trial-averaged cholinergic response for both rewarding and punishing events (Fig. 3A), being driven by deficits in the magnitude of evoked cholinergic transients (Fig. 3B). We also note that spontaneous events in APP mice occurring within the sham event response window exhibit an approximately twofold increase in their probability of occurrence at the mid stage of nascent amyloidosis (probability of a spontaneous transient in any 200 ms time window: WT= 0.030; APP = 0.054), though their magnitudes remain indistinguishable to that of WT controls (Fig. 3B, purple bars). Deficits in cholinergic signaling to unexpected behaviorally relevant events (rewards and punishments) progressively deepen by the late stage, driven by a drop in their evoked probability both for reward and punishment (Fig. 3C), as well as a further decrease in the magnitude of evoked responses to punishments (Fig. 3B). Amyloidosis is thus observed to lead to a progressive impairment of cholinergic signaling to unexpected behaviorally relevant events over these stages. In contrast, cholinergic signaling of unexpected VS does not exhibit a diminishment by these measures and are indistinguishable to that observed for WT controls.

Cholinergic responses to outcome-predictive visual stimuli and to predictable rewarding and punishing outcomes

The cholinergic system has been postulated to serve as a reinforcement signal (Chubykin et al., 2013; Hangya et al., 2015; Liu et al., 2015a; Sturgill et al., 2020; Hegedüs et al., 2023), for which there is supporting evidence in the visual cortex. Therefore, we wished to determine whether and how cholinergic responses in VC change to visual cues, rewards, and punishments when visual cues are predictive of future rewarding or punishing outcomes. Specifically, we exposed 3-month-old WT mice to repeated sessions in which visual cues were predictive of reward (appetitive cohort) or punishment (aversive cohort) at a fixed delay (1.5 s; Figs. 4A, 5A). In the appetitive cohort (Fig. 4, Table 1, Figure 4 for ANOVAs), VS-evoked (green) trial-averaged cholinergic responses (Fig. 4C) increased as trace conditioning progressed, accompanied by a corresponding decrease in cholinergic responding to subsequent rewards (Fig. 4C, blue). This augmenting response to the reward-predictive visual cue, and decrementing response to the predicted reward, is a product of changes in both the magnitude (Fig. 4D) and probability (Fig. 4E) of evoked cholinergic transients, increasing to the VS and decreasing to the reward.

Qualitatively similar effects are observed over aversive trace conditioning (Fig. 5, Table 1, Figure 5 for ANOVAs). Here too, visually evoked cholinergic responses to the predictive VS (Fig. 5C, green) augment with conditioning, being driven by increases in the magnitude and probability of cholinergic transients, while punishment-evoked responses (Fig. 5C, magenta) diminish with conditioning, being driven by decreases in the magnitude (Fig. 5D) and probability (Fig. 5E) of evoked cholinergic transients. In appetitive and aversive cohorts, the inclusion of catch trials, in which reward or punishment was withheld, allowed for cholinergic activity at the time of expected (but not received) outcome to be assessed. When withheld, no positive or negative deflection of the cholinergic activity was observed at the time of the expected outcome, be it reward or punishment.

Effect of amyloidosis on cholinergic signaling of outcome predictive, and outcome-predicted, events

Having identified conditioning-induced changes in cholinergic signaling to predictive and predicted cues, we next sought to determine the effect that emerging amyloidosis may have on this outcome prediction signaling. To assess the impact of amyloidosis, APP mice at the early, mid, and late stages, underwent appetitive and aversive trace conditioning, with their age- and sex-matched littermates serving as controls. Whereas WT responses are consistent following conditioning across the early, mid, and late stages—exhibiting augmenting responses to outcome-predictive visual cues and suppressing cholinergic responses to predicted outcomes—APP mice exhibit progressively deficient experience-dependent changes in cholinergic signaling. Differential experience-dependent effects are assessed, as below, by quantifying changes in the duration (Fig. 6) and strength (magnitude and probability) of acetylcholine response profiles (Figs. 7, 8).

Figure 6.

Differential effects of conditioning on ACh response duration. Return-to-threshold times of evoked ACh responses prior to conditioning (circles) and following conditioning (x's), for each animal in aversive (left) and appetitive (right) cohorts, grouped by WT versus APP and split across stage (early, mid, late). Conditioning results in a significant increase in ACh response duration following either appetitive or aversive conditioning, with a greater increase following aversive conditioning. APP mice diverge from WT in aversive conditioning by the mid stage and continuing in the late stage, with the degree of response increase being foreshortened as amyloidosis progresses.

Figure 7.

Effect of amyloidosis on cholinergic appetitive conditioning. Cholinergic responses to conditioned stimuli (visual cues, green) and unconditioned stimuli (water reward, blue) were assessed across days of appetitive trace conditioning at early, mid, and late stages preceding cholinergic degeneration in APP mice (dashed lines) and compared with those observed in their WT controls (solid lines). Changes in cholinergic responses, as measured by their trial-averaged response (top row), transient magnitude (middle row), and transient probability (bottom row), are shown relative to preconditioning responses to VS and Reward. Following ANOVAs (Table 1), stars indicate a significant difference between APP and WT responses in their pairwise comparisons (Tukey's test).

Figure 8.

Effect of amyloidosis on aversive conditioning. Cholinergic responses to conditioned stimuli (visual cues, green) and unconditioned stimuli (shock punishment, magenta) were assessed across days of aversive trace conditioning at early, mid, and late stages preceding cholinergic degeneration in APP mice (dashed lines) and compared with those observed in their WT controls (solid lines). Cholinergic responses, as measured by their trial-averaged response (top row), transient magnitude (middle row), and transient probability (bottom row), are shown relative to preconditioning responses to VS and punishment. Following ANOVAs (Table 1), stars indicate a significant difference between APP and WT responses in their pairwise comparisons (Tukey's test).

First, the inclusion of catch trials affords a means of assessing the impact of conditioning on the duration of cholinergic signaling to outcome-predictive visual cues, and whether genotype and/or stage has any differential effect on VS-evoked response duration. Taking the return times to threshold as a measure of ACh response duration, we find that conditioning, be it appetitive or aversive, leads to a significant increase in the duration of the VS-evoked response in WT and APP mice at all stages, compared with VS-evoked responses before conditioning (three-way mixed-effect ANOVA on the return to baseline; effect of appetitive conditioning: F_(1,50) = 64.70, p = 1.42E-10; effect of aversive conditioning: F_(1,50) = 1754.46, p = 1.32E-40), an effect that is consistent across all mice (Fig. 6). The ACh response duration to VS of WT mice increased from 1.32 ± 0.093 s preappetitive conditioning to 1.55 ± 0.090 s postconditioning (mean ± SEM), and from 1.37 ± 0.088 s pre-aversive conditioning to 2.3 ± 0.120 s postconditioning. While conditioning results in an increase in response duration in either case, response duration was greater following aversive than appetitive conditioning (t test; p = 0.04E-19). The ACh response duration to VS of APP mice also increased, from 1.38 ± 0.140 s pre-appetitive conditioning to 1.53 ± 0.120 s postconditioning, and from 1.34 ± 0.089 s pre-aversive conditioning to 2.12 ± 0.110 s postconditioning. However, splitting by stage and comparing to WT reveals that APP mice diverge from WT mice, with the degree to which their response duration is augmented by aversive conditioning being significantly foreshortened as amyloidosis advances in APP mice. Specifically, by the mid stage (2.05 ± 0.080 vs 2.30 ± 0.015 s), and continuing in the late stage (2.08 ± 0.035 vs 2.40 ± 0.10 s) following aversive conditioning, APP response durations were foreshortened compared with WT mice (p = 0.0014 and p = 0.0015 for mid and late stages, respectively).

Second, to quantify the effect of amyloidosis on experience-dependent changes in the strength of cholinergic signaling, we normalized, within subjects, the cholinergic responses during associative pairing to the same events (VS, reward, punishment) on days immediately before pairing. Plotting the change in cholinergic responding to the conditioned and unconditioned stimuli, we then compared WT (solid lines) to APP (dashed lines) mice over successive sessions when pairing was performed at early, mid, or late stages in appetitive cohorts (Fig. 7, Table 1, Figure 7 for ANOVAs) and in aversive cohorts (Fig. 8, Table 1, Figure 8 for ANOVAs).

In the appetitive cohort, at the early stage, the trial-averaged cholinergic responses of APP and WT animals (Fig. 7, top left)—when so normalized to their preconditioning values—were indistinguishable, exhibiting by the end of conditioning a ∼2.3-fold increase to reward-predicting visual cues (Fig. 7, green), and a ∼31% reduction in responses to predicted rewards (Fig. 7, blue). VS-evoked cholinergic transient magnitude increased ∼20% (Fig. 7, middle left) while cholinergic transient magnitude evoked by reward (Fig. 7, bottom left) decreased ∼16%. The probability of VS-evoked cholinergic transients increased approximately twofold, while the probability of reward-evoked cholinergic transients decreased ∼20%. By mid stage, deficiencies in conditioning-induced changes to cholinergic signaling begin to emerge in APP mice as the suppression of cholinergic responding to predictable reward (as observed in WT mice) begins to falter. By the late stage of emerging amyloidosis, the augmented response to reward-predicting VS as well as the suppressed response to predicted reward observed in WT controls is impaired in APP mice. Specifically, VS-evoked cholinergic responses in APP mice (Fig. 7, dashed green) exhibit significantly diminished changes in trial-averaged responses (Fig. 7, top right), driven by decreases in transient magnitude (Fig. 7, middle right) and evoked probabilities (Fig. 7, bottom right). At the same time, conditioning-induced suppression of reward-evoked cholinergic responses (as observed in WT mice) is significantly diminished in APP mice (Fig. 7, right, dashed blue), being indistinguishable from preconditioning values over any metric (one-way repeated-measures ANOVA; effect of training day, trial-averaged response: F_(5,24) = 1.98, p = 0.12; transient magnitude: F_(5,24) = 1.37, p = 0.27; probability of transient: F_(5,24) = 1.22, p = 0.33). Notably, while these deficits in cholinergic signaling emerge across stages in APP mice, licking behavior does not differ significantly between WT and APP mice across these stages [two-way ANOVA on the number of licks in a rewarded lick bout (6.4); effect of genotype: F_(1,24) = 1.03, p = 0.32; pathologic stage: F_(2,24) = 0.79, p = 0.46; interaction: F_(2,24) = 0.018, p = 0.98; two-way ANOVA on the number of licks in a spontaneous lick bout (3.2); effect of genotype: F_(1,24) = 0.71, p = 0.4; pathologic stage: F_(2,24) = 0.073, p = 0.89; interaction: F_(2,24) = 0.12, p = 0.89; two-way ANOVA on the probability of a spontaneous lick bout in 0.1 s (0.0037); effect of genotype: F_(1,24) = 0.025, p = 0.8; pathologic stage: F_(2,24) = 0.06; p = 0.94; interaction: F_(2,24) = 0.25, p = 0.78].

In the aversive cohort, amyloidosis causes deficits in experience-dependent cholinergic signaling that are similar to those observed under appetitive conditioning. Augmenting cholinergic responses to punishment-predicting VS and suppressing cholinergic responses to predicted punishments—as observed in WT mice—become deficient in APP mice with the progression of amyloidosis (Fig. 8, Table 1, Figure 8 for ANOVAs). As observed under appetitive conditions, at the early stage of nascent amyloidosis, cholinergic responses do not differ between WT and APP mice. VS-evoked responses (Fig. 8, green) increase by the end of conditioning to ∼3.5-fold their preconditioning value (Fig. 8, top left), driven by an ∼1.6-fold increase in evoked cholinergic response magnitude (Fig. 8, middle left) and an ∼2.2-fold increase in the probability of evoking a response (Fig. 8, bottom left). Punishment-evoked cholinergic responses (Fig. 8, magenta) fall over the course of conditioning to ∼73% of their preconditioning value, accounted for by a 17% drop in the evoked cholinergic transient magnitude and a ∼12% drop in the probability of evoking a cholinergic transient. By the mid stage (Fig. 8, middle column), however, APP mice express impaired cholinergic signaling to punishment-predicting VS (Fig. 8, dashed darkened green), failing to augment their response to the same degree as that observed in their WT controls (Fig. 8, solid bright green). In contrast to APP mice in the appetitive cohort, this impairment to the outcome-predictive VS occurred earlier in the progression of amyloidosis. Suppression of cholinergic signaling evoked by the predicted punishment in WT mice (Fig. 8, solid bright magenta) was impaired in APP mice (Fig. 8, dashed darkened magenta) by the mid stage (akin to that observed in the appetitive cohort). These deficits in APP mice are also observed in the late stage and are driven by decreases in the change in cholinergic response magnitudes (Fig. 8, right column, middle) and probabilities (Fig. 8, right column, bottom) across conditioning.

Hyperactive cholinergic activity occurs at the onset of amyloidosis-induced deficits in cholinergic signaling

Until now, we have focused principally on cholinergic responses to visual, rewarding, and punishing events in WT and APP mice, withholding discussion of the effects of amyloidosis on spontaneous cholinergic transients. We note, however, that spontaneous events in APP mice exhibit a significant (approximately twofold) increase over WT controls in the probability of their occurrence, which occurs just as cholinergic signaling begins to exhibit deficiencies (Fig. 3C). Though this increase in the probability of spontaneous transients is too small to lead to a detectable change in the trial-average response within the sham event time window (Fig. 3A, spontaneous), this observation prompted us to determine whether the total cumulative release of ACh across the entire session significantly differs between WT and APP mice. Surprisingly, despite decreased cholinergic signaling to unexpected events in APP mice, an increased probability of spontaneous transients (Fig. 9, top row) in these mice led to a significantly higher cumulative cholinergic signal (Fig. 9, bottom row) than in WT mice, occurring at the mid stage of nascent amyloidosis (Fig. 9, Table 3, ANOVAs). Thus, a transient hyperactivity in spontaneous acetylcholine release occurs at the onset of detectable impairments in cholinergic outcome and prediction signaling.

Figure 9.

APP mice exhibit tonic cholinergic hyperactivity at the onset of cholinergic dysfunction. Quantification of cholinergic activity reveals at the mid pathological stage an increased probability of spontaneous cholinergic transients in APP mice, leading to an overall hyperactive release of acetylcholine across the session. A, B, Probability of spontaneous transients during reward and shock sessions, respectively. Probabilities represent the likelihood of transient in any 200 ms time window in the absence of behavioral stimulus. Error bars represent the SE. C, D, Total cholinergic response within a session for reward (C) and punishment (D) sessions, respectively. Values represent the session cumulative cholinergic response, calculated by adding AUC values from all detected transients during a 1 h recording period. This calculation was done for both reward and shock sessions and does not differentiate between spontaneous and evoked transients. Following ANOVAs (Table 3), stars represent significant differences between APP and WT mice of the same stage. Despite this time including event-driven release (which is diminished in APP mice), APP mice exhibit an overall cholinergic hyperactivity at the mid stage compared with WT controls.

Table 3.

The effect of genotype and stage of amyloidosis on spontaneous cholinergic activity

Parameter tested	Genotype (df = 1)		Pathologic stage (df = 2)		Interaction (df = 2)
	F	p	F	p	F	p
Figure 9, two-way ANOVA, effects of genotype and pathologic stage
Probability of spontaneous transients in rewarded trials	1.121160445	3.00E-01	3.51E + 00	4.58E-02	6.322376366	6.23E-03
Probability of spontaneous transients in shock trials	16.57845946	4.40E-04	9.43E + 00	9.50E-04	3.424534352	4.92E-02
Total cumulative energy in rewarded trials	0.563153766	4.60E-01	4.04E + 00	3.08E-02	6.038374819	7.51E-03
Total cumulative energy in shock trials	11.64275947	2.29E-03	10.0043799	6.92E-04	5.132141141	1.39E-02

The probability of spontaneous transients, and the resulting cumulative cholinergic response across a session, significantly differ between WT and APP mice under appetitive and trace conditioning. This difference is accounted for by spontaneous hyperactivity at the mid stage in APP mice, which occurs at the onset of deficient cholinergic signaling.

Discussion

The importance of the cholinergic system to learning and memory and its susceptibility to amyloidosis has been central to efforts addressing Alzheimer's disease. By quantifying its response profile in wild-type and APP mice, we further the case that cortical cholinergic innervation signals reinforcement and is degraded by amyloidosis before its degeneration. In quantifying Aβ peptide accrual in a mouse model of preclinical AD, we identified three stages of amyloidosis preceding the degeneration of cholinergic synapses within VC and demonstrate that cholinergic responses become impaired over these stages, diverging progressively from age- and sex-matched controls. Specifically, while responses to unexpected visual stimuli remain intact, responses to unexpected rewards or punishments diminish with the progression of amyloidosis. Atop this effect, progressing amyloidosis attenuates both (1) augmenting cholinergic responses to outcome predictive visual cues as well as (2) the suppression of cholinergic responses to predicted outcomes, as occurs normally in wild-type littermates. In addition, spontaneous cholinergic events are transiently hyperactive at the onset of amyloid-induced impairments of cholinergic signaling, further implicating dysregulated cholinergic activity as an early functional biomarker. Together these observations advance the case that ACh can act as a reinforcement signal, which is impaired by amyloidosis before pathologic degeneration of cholinergic synapses.

Reinforcement signaling by acetylcholine in the visual cortex

By observing prompt and robust cholinergic responses to rewards as well as punishments, we corroborate independent lines of evidence that the cholinergic system conveys behaviorally relevant outcomes (Hangya et al., 2015), be they appetitive or aversive (Harrison et al., 2016), to cortex (Eggermann et al., 2014; Guo et al., 2019), and specifically to VC (Chubykin et al., 2013; Liu et al., 2015a) as previously predicted (Gavornik et al., 2009; Huertas et al., 2015). Though measured within visual cortex, unexpected rewards and punishments evoked cholinergic transients that occurred with probabilities and magnitudes significantly higher than those elicited by unexpected visual cues. When visual cues are predictive of future outcome, however, cholinergic responses to these cues augment, while responses evoked by predictable rewarding or punishing outcomes attenuate. These observations broadly agree with prior reports of neural responses to outcomes and outcome prediction in the basal forebrain (Richardson and DeLong, 1991; Avila and Lin, 2014; Zhang et al., 2019), and from cholinergic basal forebrain neurons specifically (Kuchibhotla et al., 2017; Hegedüs et al., 2023), using different approaches, such as calcium imaging of basal forebrain cholinergic neurons (BFCNs; Robert et al., 2021), antidromic optotagging (Guo et al., 2019), as well as optical and amperometric measurement of acetylcholine release within the basal forebrain (Hanson et al., 2021) and medial prefrontal cortex (Parikh et al., 2007), respectively. Cholinergic signaling to reward and acquired cue prediction has also been measured in the amygdala (Crouse et al., 2020; Sturgill et al., 2020) where augmenting its release experimentally resulted in improved task acquisition (Crouse et al., 2020). These observations collectively advance the case that ACh displays and acts with reinforcement signaling qualities (though not necessarily to the exclusion of other reported roles).

Appetitive and aversive trace conditioning of the same visual cue under like conditions allows for cholinergic response profiles to be contrasted. When occurring unexpectedly, reward and punishment evoked comparable cholinergic responses, and, when paired with visual cueing, evoked similar experience-dependent changes across conditioning. One notable difference, however, is that while visually cued cholinergic responses begin to return to baseline soon after the CS, responses remain near peak values over the CS–US interval following aversive conditioning (Figs. 4B, 5B, 6). This difference in cholinergic signaling agrees with prior observations regarding the effects of appetitive and aversive conditioning: whereas cholinergic responses were observed to bridge the CS–US interval following aversive conditioning of a predictive auditory stimulus (Guo et al., 2019), reward-predicting tones evoked cholinergic transients that quickly returned to baseline without spanning the interval following appetitive conditioning (Crouse et al., 2020). This difference in resulting acetylcholine response profile may explain why appetitive conditioning leads to the emergence of cued interval timing activity to expected reward by VC neurons (Chubykin et al., 2013), whereas aversive trace conditioning results in suppression of cue-evoked responses (Monk et al., 2021). As transient optogenetic activation of cholinergic axons at a delay is sufficient to condition neural reports of that interval (Liu et al., 2015a), it is possible that sustained elevated cholinergic signaling between the CS and US, as in the aversive case, suppresses cue-evoked responses, preventing VC neurons from reporting the delay to aversive outcomes.

Together, robust transient signaling of outcomes along with experience-dependent changes to predictive and predicted events add to a growing body of evidence that acetylcholine can act as a reinforcement signal. Its properties differ in a number of ways, however, from canonical reward prediction error (RPE) signaling as formalized in temporal difference reinforcement learning (Sutton and Barto, 1998), and as typically ascribed to dopaminergic neurons (Schultz et al., 1997; but see Jeong et al., 2022). Namely, observed cholinergic responses differ from RPE by (1) exhibiting activation to aversive as well as appetitive events, (2) modifying similarly under aversive and appetitive conditioning, and (3) lacking a paucity of release at the time of expected but unreceived outcomes. These features indicate a reinforcement signal of a distinct form whose properties could potentially be understood in terms of an absolute prediction error, as recently put forward, where experimentally observed activity of cholinergic basal forebrain neurons is modeled as an unsigned, weighted prediction error (Hegedüs et al., 2023). Alternatively, an unsigned prediction error could similarly describe a “surprise signal” conveying unexpected salience (Lin and Nicolelis, 2008; Avila and Lin, 2014; Zhang et al., 2019). Signaling unexpected salience, useful for mechanisms of attention, and signaling unsigned prediction errors, useful for learning, need not be mutually exclusive, however.

Amyloidosis degrades cholinergic reinforcement signaling

How might amyloidosis exert its effects on cholinergic signaling before degeneration? A clue may be found in how amyloidosis degrades cholinergic signaling to unexpected appetitive and aversive events, but not, curiously, to visual cues at any stage preceding cholinergic degeneration investigated (Fig. 3). One possibility for this difference may be that VS-related transients are reported by a different and less vulnerable cholinergic neuron subpopulation from the basal forebrain, or, perhaps in part, local to VC (Granger et al., 2020), than those conveying reward and punishment events. Functional (Robert et al., 2021) and molecular specialization of BFCNs lends credence to this notion, as cholinergic neurons within different basal forebrain nuclei express different subunit receptor components with varying Aβ susceptibility (George et al., 2021). Indeed, oligomeric Aβ has been shown to associate with the α7β2 subunits of nicotinic receptors on BFCNs and increase their intrinsic excitability (George et al., 2021), which may also explain the cholinergic hyperactivity observed here at the onset of altered cholinergic signaling (Fig. 9). Another possibility is that Aβ directly interferes with cholinergic release at target sites, being most readily revealed under high-demand events such as rewards and punishments. Aβ has in fact been shown to impair choline transport (Parikh et al., 2014; Cuddy et al., 2015, 2017) and choline acetyltransferase activity (Nunes-Tavares et al., 2012; Kumar et al., 2018), both of which could impair the release of acetylcholine as revealed under heavy demand.

Impaired learning of outcome-prediction signaling in cortex, as results when disrupting cholinergic signaling of outcomes (Chubykin et al., 2013), may in turn lead to impaired experience-dependent changes in responses of cholinergic neurons to predictive and predicted cues. While conditioning produced an increase in CS-evoked cholinergic responding in controls, it also revealed an amyloid-induced deficit in signaling to these outcome predictive cues (Figs. 7, 8). In the mid and late stages of incipient amyloidosis preceding cholinergic degeneration, impaired experience-dependent augmentation in the CS-evoked cholinergic response was observed (Figs. 7, 8) in amyloid mice. Though impaired augmenting responses to outcome predictive cues is consistent with the idea that Aβ may reduce the capacity of cholinergic neurons to release large amounts of neurotransmitter, it may also indicate a failure of outcome-predictive input to BFCNs to drive augmented outcome prediction signaling by BFCNs. At the same time, impaired experience-dependent reductions in the US-evoked cholinergic response is also observed (Figs. 7, 8) in amyloid mice. Together, these impairments in experience-dependent responding to predictive and predicted events suggest rather a failure of plasticity upstream of or onto BFCNs, which can be accounted for in a model where cued intervals and outcomes are learned within cortex and conveyed to reinforcement signaling areas (Huertas et al., 2015; Namboodiri et al., 2015; Cone and Shouval, 2021).

Conclusion

Here, we determined that amyloidosis impairs outcome and outcome prediction signaling within cortex by the cholinergic system before its synaptic loss. These observations have important clinical and scientific implications. They deepen the connection between amyloidosis and impaired cholinergic signaling as a cause of degrading learning and memory in AD. Aberrant cholinergic signaling preceding pathologic degeneration is also identified as a potential early functional biomarker. These findings point to when and how intervention may best ameliorate cognitive decline in AD and advance mouse visual cortex as a test bed to assess to what degree cholinergic signaling may be rescued by early intervention. Interventional approaches applied at the onset of irregular cholinergic activity, but preceding its degeneration, may best forestall cognitive decline. This includes rescuing cholinergic signaling from the consequences of amyloidosis by targeted clearance of toxic forms of Aβ, pharmacologically countering amyloid-induced cholinergic hyperexcitability at its onset, and promoting cholinergic innervation (Burke et al., 2013; Tucker et al., 2015; Rofo et al., 2022; van Dyck et al., 2023; Zott et al., 2023). In advanced stages of AD where irreversible damage to the cholinergic system has been sustained, neuroprosthetic control of remaining cholinergic innervation may provide a means to mimic the dynamic signaling of outcomes and outcome-predictive events of acetylcholine, improving learning and memory.