Introduction

The fundamental nature of the AD disease(s) is conceived as a mixture of proteinopathies and inflammation, with age the biggest risk factor^1–4. In the prodromal phase, the trajectory of the disease, however, may be shaped by selective declines in neuromodulator molecules that are produced subcortically⁵. For example, acetylcholine (ACh) is a neuromodulator, synthesized by the choline acetyltransferase (ChAT) enzyme, that supports arousal, emotion and cognition, activating both metabotropic (muscarinic) and ionotropic (nicotinic) receptors^6–15. During wakefulness, but also during rapid eye movement sleep (REMS), ACh is released widely into the forebrain from key groups of neurons in the cholinergic basal forebrain (BF), which include, among other nuclei, the medial septum (MS)^6,7,16.

Known for nearly half a century, and confirmed numerous times, the selective degeneration of BF-cholinergic neurons is a notable early feature of Alzheimer’s disease (AD)^17–30, emerging before cortex atrophy^5,31,32. It was postulated that this early decline of ACh neurons in humans causes the mild cognitive impairment at the start of AD¹⁸. Cholinergic esterase inhibitors significantly delay the decline of cognition for (some) patients³³, but do not stop the disease progressing²⁴. For these reasons, according to some commentators, interest in the cholinergic system’s failure in AD has languished^24,25. It has also been suggested that neglecting the importance of neuromodulatory systems in the genesis of AD probably also reflects dominant cortico-centric models that assume that higher-order deficits are caused by cortical and hippocampal damage alone¹³. However, these cortico-centric models fail to account for the reliance of these structures on their neurochemical milieu and tend to minimize the critical role of ascending neuroregulatory systems in synchronizing flexible and precise information processing throughout the brain¹³.

Neuromodulators are also essential for orchestrating sleep-wake states^34,35. Genetic knockout studies demonstrate that ACh is required for the generation of REMS, although the circuitry involved is unclear³⁶. Importantly, reductions and alterations in REMS have emerged as strong predictors of clinical AD onset³⁷, alongside alterations of non-rapid eye movement sleep (NREMS)³⁸. Additionally, lower percentages of REMS correlate with smaller volumes of the cholinergic BF³⁹ and are further linked with higher all-cause, cardiovascular, and non-cancer mortality⁴⁰. Both REMS and its associated theta rhythms are hypothesized to promote cognition^41–43, regulate emotion, and dampen stress responses^44–46, processes that are compromised in AD and in individuals living with mild cognitive impairment, as well as in animal models exhibiting widespread amyloid-β (Aβ) accumulation^47–61.

In humans, BF-cholinergic neurons are especially susceptible to Aβ toxicity, irrespective of the AD subtype, and BF atrophy correlates strongly with neocortical Aβ burden^21,62–64. The degeneration of cholinergic neurons, alongside other neuromodulator systems, is posited to potentiate the pathological processes underlying AD^5,13. Thus, it is worth examining the extent to which cholinergic deficits contribute to the disease symptoms. Of note, the MS contains a mixture of ACh, GABA and glutamate neurons which all project to the hippocampus and other areas^{11,16,29,65–68}. Strikingly, Aβ peptide injection into the MS of rats disrupts neuronal activity, hippocampal long-term potentiation, theta rhythm and cognition^69,70.

To test if and how selective amyloid pathology affects MS cholinergic (MS^ChAT) cells, and its consequences for REMS and cognition, we expressed the synthetic App^NL-G-F allele, which comprises a combination of familial AD mutations⁷¹, selectively in mouse MS^ChAT neurons, and over 14 months, compared the phenotype in mice where the App^NL-G-F allele is expressed brain-wide. We found that inducing pathology specifically in the small population of MS^ChAT neurons generates highly similar phenotypes (inter alia, altered cognition and emotional processing, as well as selective REMS deficits) to those of the App^NL-G-F knock-in mouse^49,71, where the allele is expressed in many cell types. Thus, the integrity of the medial septal cholinergic cells underpins a large part of App^NL-G-F model’s phenotypic features. In essence, we have proposed a circuit mechanism and developed an animal model that may capture key aspects of the prodromal phase of AD, which incorporates the selective vulnerability of acetylcholine cells and which suggests how an early and accumulating deficit in REM sleep and associated cognitive changes in the prodromal phases of AD might be produced. Nevertheless, broadcasted amyloid from MS cholinergic neurons also generates disease features as it produced hippocampal astrocyte activation and epileptiform spikes.

Results

Cholinergic medial septal cells are wake- and REMS-active and primarily send projections to the hippocampus and cortex

We first characterized MS^ChAT cell activity across vigilance states. By calcium photometry, MS^ChAT cells were most active in REMS and WAKE, but quiet in NREMS (AAV-DIO-GCaMP6s was injected into the MS of ChAT-Cre mice; Figures 1A-B). Baseline Ca²⁺ transients increased during transitions from NREMS to REMS and NREMS to wakefulness. We next mapped the projection targets of MS^ChAT cells. Many of their axons were detected in hippocampal structures (CA areas, dentate gyrus and subiculum), but also in the medial prefrontal cortex (mPFC; prelimbic and cingulate cortices), primary cortices (somatosensory, motor and visual) and entorhinal cortex, as well as in the olfactory bulb, thalamus, hypothalamus, amygdala, and medial habenula (for the tracing, AAV-DIO-ChR2-EYFP was injected into the MS of ChAT-Cre mice; Figures 1C-D).

Figure 1.

Selective expression of a pathological App allele in cholinergic medial septal neurons generates localized Aβ accumulation and leads to cholinergic neuronal loss

Cre recombinase-dependent AAV transgenes that express the familial App^NL-G-F mutations were constructed (Figure 2A). The viruses AAV-DIO-App^NL-G-F-P2A-EGFP or AAV-DIO-EGFP (as a control) were injected into the MS of ChAT-Cre male and female mice to generate MS^ChAT-App^NL-G-F and MS^ChAT-GFP mice that were then allowed to age for 13 to 14 months (Figure 2A); their phenotypes were then compared in detail with App^{NL-G-F/NL-G-F} global knock-in mice that were aged at the same time (Figure 2D). As an internal control, to determine whether the phenotypes observed in aged MS^ChAT-App^NL-G-F mice were predominantly attributable to the putative loss of MS cholinergic neurons or to the deposition of amyloid, we selectively lesioned MS^ChAT cells independently of amyloid expression. AAV-DIO-Caspase was injected into the MS of ChAT-Cre male and female mice, thereby generating MS^ΔChAT animals, which were subsequently aged until 7-8 months post-surgery, making them around 9-10 and 10-11 months old at the time of sleep and behavioral analysis, respectively.

Figure 2.

For 13- to 14-month-old MS^ChAT-App^NL-G-F mice, immunohistochemical analyses employing the amyloid-specific antibodies D54D2 and 6E10 (Figures 2C and S1A, respectively) revealed a marked accumulation of Aβ deposits in the MS (respectively p = 0.0452 and p < 0.001, Figures 2E and S1B). This amyloid deposition extended beyond the MS into adjacent structures of the BF cholinergic system, including the vertical (VDB) and horizontal (HDB) limbs of the diagonal band of Broca. While D54D2 staining showed only non-significant trend in these regions after multiple-comparison correction (Figure 2E), 6E10 staining revealed significant increases in both VDB and HDB (respectively p < 0.001 and p = 0.0257, Figure S1B). There was also a significant increase in Aβ deposits, assessed using the D54D2 antibody, in the same BF regions of 13- to 14-month-old App^{NL-G-F/NL-G-F} mice (p < 0.001 for MS, VDB and HDB; Figures 2D-E), albeit with a greater magnitude than in the MS^ChAT-App^NL-G-F model (p < 0.001, MS^ChAT-App^NL-G-F versus App^{NL-G-F/NL-G-F} mice).

Notably, in comparison to MS^ChAT-GFP controls, 14-month-old MS^ChAT-App^NL-G-F mice had significantly greater Aβ accumulation in regions innervated by MS^ChAT cells (Figures 3A-C). This accumulation was particularly prominent in the CA1, CA2 and hilar areas of the hippocampus (p = 0.0065, Figures 3A-B), with additional amyloid deposits detected in the thalamus, amygdala, mPFC, primary cortical areas, and the hypothalamus (Figure 3C). However, no increase in Aβ burden was observed in the medial habenula, olfactory bulb, or entorhinal cortex, despite their cholinergic innervation by the MS (see above), nor in the caudate-putamen, where no cholinergic input from the MS was previously detected (see Supplemental information). There was also some colocalization of amyloid and laminin-labeled blood vessels in the hippocampus of MS^ChAT-App^NL-G-Fand, to a much lesser extent, MS^ChAT-GFP mice that was not observed in the BF (Figures S1C-D). This suggests that MS^ChAT cell terminals preferentially innervate hippocampal blood vessels.

Figure 3.

Both MS^ChAT-App^NL-G-F and homozygous App^{NL-G-F/NL-G-F} mice exhibited a loss of MS^ChAT cells relative to their respective controls, with decreases of 23.1% (p < 0.001, Figures 4A, D) and 21.9% (p < 0.001, Figures 4B, D), respectively. App^{NL-G-F/NL-G-F} mice also had fewer ChAT-expressing neurons in the HDB compared with App^WT controls (Figure 4D). In MS^ChAT-App^NL-G-F mice, despite widespread amyloid deposition in the hippocampus, no changes in neuronal (NeuN) staining in hippocampus were observed compared to MS^ChAT-GFP controls (Figure S2).

Figure 4.

The MS also contains GABAergic and glutamatergic neurons and synapses^65–67. Compared to controls, App^{NL-G-F/NL-G-F} mice had reduced GAD67 and VGLUT2 immunoreactivity (collective cell bodies and synapses) in the MS, VDB, and HDB (Figures S3A-D). No such changes were observed between MS^ChAT-App^NL-G-F and MS^ChAT-GFP mice, however (Figures S3A-D). Overall, in the MS^ChAT-App^NL-G-F mice, the intrinsic overproduction of amyloid by MS^ChAT neurons over about a year caused the degeneration of about a quarter of them, but the resulting amyloid in the MS did not affect the neighboring non-ChAT cells or synapses.

In the MS^ΔChAT mice, complete ablation of MS cholinergic neurons was not achieved by the caspase expression. Instead, there was a partial neuronal loss of approximately one third of cells (31.7%, p < 0.001, Figures 4C-D). The number of cholinergic neurons in adjacent BF regions (i.e., VDB and HDB) of MS^ΔChAT mice remained unaltered compared with MS^ChAT-GFP controls (Figure 4D). Although not anticipated, this partial loss of cholinergic neurons turned out to be fortuitous, as the degree of loss was roughly similar to the loss of cholinergic neurons induced by the familial App^NL-G-F allele, further enhancing the utility of comparing MS^ChAT-App^NL-G-F and MS^ΔChAT mice.

Aβ-induced disruption of the MS-cholinergic system hinders memory and reduces anxiety-like behavior

Home-cage locomotor activity, measured 13- to 14-months following AAV injections, was only elevated in the dark phase in App^{NL-G-F/NL-G-F} mice compared with App^WT controls (Figure 5A). Notably, female mice were consistently more active than males across all groups (see Supplemental information). We also assessed baseline stress levels by ELISA-based quantification of fecal corticosterone metabolites (FCM; Figure 5B). MS^ChAT-App^NL-G-F mice showed no significant difference in corticosterone concentrations throughout the day compared to MS^ChAT-GFP controls. By contrast, App knock-in mice displayed elevated overall baseline FCM levels relative to their App^WT counterparts (Figure 5B). In all types of mice, females consistently exhibited higher baseline levels of FCM than males (see Supplemental information).

Figure 5.

Behavioral testing was performed 13- to 14-months following AAV injections. Short-term working memory was assessed using the Y-maze paradigm (Figure 5C). Both MS^ChAT-App^NL-G-F and App^{NL-G-F/NL-G-F} mice performed significantly worse than their corresponding controls MS^ChAT-GFP and App^WT (respectively p < 0.001 and p = 0.0418), indicating impaired spatial working memory. MS^ΔChAT mice likewise performed significantly worse than MS^ChAT-GFP animals (p = 0.0049). In the novel object recognition task (NORT; Figure 5D), MS^ChAT-App^NL-G-F mice had no preference for the novel object as opposed to the MS^ChAT-GFP group (p < 0.001, versus 10 seconds average exploration), suggesting impaired non-spatial object memory; a similar lack of preference was observed in MS^ΔChAT mice. Conversely, App^{NL-G-F/NL-G-F} mice, like App^WT controls, explored the novel object more, indicating intact recognition memory (p < 0.001, versus 10 seconds average exploration). All groups spent comparable amounts of time exploring the two identical objects during the familiarization session of the NORT, confirming equal baseline exploratory behavior (see Supplemental information).

In both the open field and light-dark box tests, MS^ChAT-App^NL-G-F mice exhibited reduced anxiety-like behavior compared with controls, with increased time spent in the central zone of the open field (p = 0.0444, Figure 5E) and the illuminated chamber of the light-dark box (p = 0.0106, Figures 5F). However, in App^{NL-G-F/NL-G-F} mice, reduced anxiety-like behavior was observed only in the open field test, compared to both App^WT controls and MS^ChAT-App^NL-G-F mice (respectively p = 0.0021 and p = 0.0444, Figures 5E-F). In contrast, MS^ΔChAT mice did not exhibit a significant reduction in anxiety-like behavior compared to controls.

All animals exhibited broadly comparable exploratory behavior across the conducted behavioral assessments, although group differences were observed in specific measures, including centre entries in the open field (consistent with reduced anxiety-like behavior), and increased distance travelled in the light-dark box for MS^ΔChAT mice compared with MS^ChAT-GFP and MS^ChAT-App^NL-G-F mice (see Supplemental information).

Amyloid pathology restricted to the MS-cholinergic system induces REMS decreases comparable to those in App^NL-G-F mice

As the mice aged from 4 to 12 months post-AAV injections, we investigated over three time points how amyloid-induced disruption of MS cholinergic neurons affects sleep-wake architecture. In comparison to MS^ChAT-GFP control mice, MS^ChAT-App^NL-G-F animals had unchanged amounts of NREMS and WAKE (Figures S4A and S5A). However, MS^ChAT-App^NL-G-F mice had fewer (p < 0.0042), but prolonged (p < 0.0052), NREMS bouts during the light phase, i.e. there was enhanced NREMS consolidation (Figures S4B-C). Notably, this same pattern regarding NREMS duration, bout length (p = 0.0098), and frequency (p = 0.040) was present in homozygous App^{NL-G-F/NL-G-F} mice when compared with App^WT controls (Figures S4A-C). Of note, App^{NL-G-F/NL-G-F} animals also displayed an overall increase in WAKE time during the light phase (p = 0.0175, Figure S5A).

Contrary to the relatively preserved structure of NREMS and WAKE, marked differences in REMS architecture emerged between MS^ChAT-GFP control and MS^ChAT-App^NL-G-F mice. Over the year, the amount of REMS in MS^ChAT-App^NL-G-F mice lessened relative to baseline controls during the light period, so that by 12 months post-AAV injection MS^ChAT-App^NL-G-F mice had clearly less REMS than controls (p < 0.001; Figure 6A). During the dark phase, reduced REMS amount was also observed in MS^ChAT-App^NL-G-F animals compared to controls (p = 0.0256). Moreover, REMS episodes in these mice became strikingly more consolidated: REMS bouts were longer during the light phase (p = 0.0278, Figure 6B), but there were fewer of them across both light and dark periods (respectively p < 0.001 and p = 0.0062, Figure 6C). During a year, presumably as the MS cholinergic neurons were slowly becoming compromised, all these REMS changes (less REMS, longer and fewer bouts) were phenocopied in App^{NL-G-F/NL-G-F} mice relative to App^WT controls (Figures 6A-C). Nine- to ten-month-old MS^ΔChAT mice also showed reduced REMS duration and fewer REMS bouts in the light phase compared to MS^ChAT-GFP controls (Figures 6D, F).

Figure 6.

To determine any changes in aperiodic and periodic spectral power of the different vigilance states, we applied a “Fitting Oscillations & One-Over-F” (FOOOF)-based spectral analysis^72,73. FOOOF analysis of MS^ChAT-App^NL-G-F mice at 12 months post-AAV injection revealed significant reductions in delta- and theta-band periodic power (1-4.4 Hz and 6-9 Hz, respectively) during REMS across both phases compared to similarly aged MS^ChAT-GFP controls (p < 0.001; Figure 6G), with similar reductions during NREMS (p < 0.001; Figure S4G) and wakefulness (p ≤ 0.0028; Figure S5G). App^{NL-G-F/NL-G-F} mice showed a comparable reduction in delta- and theta-band periodic power during REMS relative to App^WT controls (p ≤ 0.0127; Figure 6G); during NREMS, only theta-band power was reduced (p ≤ 0.0267; Figure S4G), and during wakefulness, only delta-band power was reduced in the dark phase (p = 0.0401, Figure S5G). MS^ΔChAT mice showed a broadly similar pattern, with reductions in delta- and theta-band periodic power during REMS (Figure 6G), NREMS (Figure S4G), and wakefulness (Figure S5G) compared with MS^ChAT-GFP mice, reaching significance in most conditions (p ≤ 0.0289).

Regarding the aperiodic component, MS^ChAT-App^NL-G-F and App^{NL-G-F/NL-G-F} mice both showed a significant reduction in EEG offset during REMS, NREMS, and wakefulness compared to their respective controls across both phases, except for NREMS in App knock-in mice (Figures S6A-C). MS^ΔChAT mice showed an offset reduction only during REMS (Figure S6A). No significant differences in the aperiodic exponent were observed in any group or vigilance state (Figures S6A-C).

Together, these findings demonstrate that cholinergic dysfunction of MS cells, produced either by lesioning about a third of them independently of pathological amyloid expression in MS^ΔChAT mice, or by compromising their health and killing about a quarter of them by expressing the familial App^NL-G-F transgene, produce largely convergent disruptions in sleep architecture and EEG spectral organisation, including the reduction of delta- and theta-band periodic power across vigilance states, and mimicking the changes seen in global App^{NL-G-F/NL-G-F} mice. Nevertheless, subtle differences in REMS architecture and aperiodic EEG features between MS^ChAT-App^NL-G-F and MS^ΔChAT mice suggest that beyond cholinergic loss, amyloid pathology may engage additional mechanisms (see next section).

Cognitive and REMS alterations of MS^ChAT-App^NL-G-F mice can be primarily attributed to the loss of cholinergic neurons, while MS^ChAT amyloid pathology drives additional features beyond cholinergic loss

MS^ΔChAT mice exhibited cognitive and REMS alterations closely resembling those observed in MS^ChAT-App^NL-G-F mice, including reduced spontaneous alternations in the Y-maze and a lack of novelty preference in the NORT (Figures 5C-D), together with decreased REMS duration (Figure 6D). In MS^ΔChAT mice, the aperiodic offset was selectively reduced during REMS (Figure S6A), further supporting a specific link between cholinergic dysfunction and REMS regulation. In line with this, Kendall partial correlation analyses across models displaying cholinergic deficits (MS^ChAT-App^NL-G-F, App^{NL-G-F/NL-G-F} and MS^ΔChAT) revealed a positive association between the number of MS cholinergic neurons and both REMS duration and theta-band periodic power during the light phase (Figure 7A).

Figure 7.

We also looked at several amyloid-associated features: epileptiform activity^74–77, and astrocyte activation^78,79. Both MS^ChAT-App^NL-G-F and App^{NL-G-F/NL-G-F} mice had significantly increased epileptiform spike activity compared to their respective controls, approximately 66.77 ± 13.96 and 129.42 ± 23.44 spikes per 12-h dark period, respectively (Figure 7B), whereas MS^ΔChAT mice, which lack amyloid deposition, did not exhibit astrocyte activation or spikes, indicating a specific contribution of amyloid to these network abnormalities.

The amyloid deposition in MS^ChAT-App^NL-G-F mice was associated with a localized increase in GFAP immunoreactivity within the MS, which was not observed in adjacent BF regions or in MS^ΔChAT mice, and was also evident in the hippocampus in line with amyloid distribution (Figure 7C).

Together, these findings support a primary role for MS cholinergic neuron loss in producing cognitive and REMS alterations, while highlighting additional, distinct contributions of amyloid pathology originating from MS cholinergic cells to hippocampal network dysfunction and astrocyte activation.

Discussion

In this study, we show that inducing a focal AD-like pathology in a small population of ACh-producing subcortical neurons leads, as mice age, to some of the typical changes that mark the prodromal phase of AD, such as the gradual loss of REMS. Remarkably, many phenotypic traits of ageing App^NL-G-F mice, a model for familial AD in which REMS disorder is a known feature, can be reproduced by inducing amyloid pathology solely within MS^ChAT cells. ACh released from these neurons into cortical and hippocampal circuits is crucial for sustaining neuronal network function during a wide variety of behaviors^6,7,12. Our results are threefold: first, a model depicting the gradual decline and selective vulnerability to amyloid of MS^ChAT cells as they age and the broadcasting of this amyloid produced within them brain-wide, which could be of relevance in human pathology; second, how loss of ACh from MS cells gradually degrades REMS generation, the theta rhythm, and the regulation of cognition; and third, an independent genetic lesioning of MS^ChAT cells, showing that it is primarily the loss of these cells (and thus acetylcholine) that likely drives the aberrant REMS and memory phenotypes, whereas deposited/released amyloid contributes additional and distinct effects, particularly on hippocampal network activity and astrocyte activation.

The App^NL-G-F allele causes neurodegeneration in knock-in rats and organoids^80,81. But little cell death has been reported in App^NL-G-F homozygous mice⁷¹. We found, however, that a subset (around 22%) of MS^ChAT cells degenerate in App^NL-G-F mice, and that this cell death is reproduced when amyloid is selectively present in MS^ChAT neurons as they age. Indeed, human cholinergic BF neurons are unusual in having high amounts of intraneuronal Aβ-accumulation, perhaps as Aβ oligomers, even in young adults^21,62. Similarly, in the 5XFAD mouse line, another model which expresses familial APP mutations, amyloid plaques initially form in the “Papez circuit” (mamillary bodies and medial septal neurons) before appearing in other areas⁸². In the MS^ChAT-App^NL-G-F mice, we found that the amyloid within the mouse MS^ChAT neurons killed about 23% of them during just over a year of expression, yet the local amyloid deposits had no discernable effect on the neighboring glutamate and GABA cells/synapses in the MS. The cell death remained cell-autonomous. Thus, our findings confirm that MS^ChAT cells are especially susceptible to amyloid toxicity.

Intriguingly, the amyloid deposits in MS^ChAT-App^NL-G-F mice are not only confined to the MS but also extend into the hippocampus (especially in the CA pyramidal cell layers and dentate hilus), and other areas innervated by MS^ChAT cells. Whether the amyloid remains inside the axons of MS^ChAT neurons or is extracellular, and, if the latter, whether it is released by living MS^ChAT cells is unclear. For other types of neurons, amyloid and/or APP can be transported along axons to terminals and released from synaptic vesicles during neuronal activity^83–86. The anatomical distribution of amyloid in our model, closely matching known septo-hippocampal projections and occurring in the absence of detectable expression in hippocampal neuronal somata, supports a projection-dependent origin of the amyloid rather than amyloid originating from local viral transduction or off-target expression. But despite increased amyloid signal in the hippocampus, NeuN staining did not reveal any significant neuronal loss, indicating that this deposition does not produce overt neurodegeneration in the hippocampus.

The fact that selective ablation of MS^ChAT cells alone, through targeted caspase expression, produces cognitive, sleep, and EEG phenotypes largely overlapping with those observed in MS^ChAT-App^NLG-F mice raises the possibility that amyloid released both in the BF and from MS^ChAT terminals in the hippocampus and other projection areas may not be sufficient per se to drive these sleep/behavioral changes. Rather, it is the loss of ACh that seems critical, a notion further supported by Kendall partial correlation analyses across models, showing that the extent of MS cholinergic cell loss is positively associated with REMS duration and theta activity.

Our results support a dichotomy in which early cholinergic dysfunction may be a primary trigger of these phenotypes, while amyloid accumulation, as it progresses and spreads, contributes additional and distinct effects. Consistent with this, while cholinergic loss alone was sufficient to impair both spatial working memory and object recognition, the presence of chronic, widespread amyloid pathology in App^{NL-G-F/NL-G-F} mice was associated with a more selective cognitive profile, with preserved recognition memory despite deficits in spatial working memory. Thus, amyloid may differentially shape the expression of cognitive impairments across domains, potentially through long-term network adaptations, while more acute and focal disruption of the MS-hippocampal cholinergic system may be more detrimental to recognition memory.

Features selectively associated with amyloid pathology, epileptiform spiking^74–77 and astrocyte activation^78,79, further support the idea that pathology can be broadcast with functional effects. Both MS^ChAT-App^NL-G-F and App^{NL-G-F/NL-G-F} mice exhibited increased epileptiform spike activity, whereas MS^ΔChAT mice did not. In fact, spiking activity in the MS^ChAT-App^NL-G-F mice was about half the amount in App^{NL-G-F/NL-G-F} mice, suggesting that the broadcasted amyloid has a significant impact on network activity. Amyloid originating from cholinergic septal cells was accompanied by increased GFAP immunoreactivity in the MS and hippocampus, consistent with astrocyte activation linked to amyloid deposition^78,79. In humans, the released amyloid from MS^ChAT terminals could initiate further pathology in other neurons, possibly via prion-like mechanisms⁸⁷, thereby amplifying and diversifying disease manifestations. We also cannot exclude that additional neuromodulatory systems, such as noradrenergic inputs from the locus coeruleus, which also degenerate early in Alzheimer’s disease⁸⁸, may also produce similar phenotypes. However, genetic lesioning of noradrenergic locus coeruleus neurons in mice does not influence gross features of NREMS and REMS or their amounts⁸⁹, but it is expected that loss of noradrenergic LC cells would diminish cognitive abilities⁹⁰.

In the hippocampus, a portion of the amyloid originating from the MS^ChAT axons appears to associate with vascular structures, as identified by laminin staining. This suggests that the MS^ChAT cells innervate some blood vessels. Indeed, ACh facilitates cerebral vasodilatation and enhances blood flow⁹¹; on the other hand, reduced vascular reactivity hampers Aβ clearance, potentially contributing to cerebral amyloid angiopathy (CAA)⁹². Supporting this, when cholinergic neurons throughout the forebrain in mice were lesioned with mu-saporin, the loss of cholinergic innervation exacerbated neurovascular impairments and CAA progression in the cortex and hippocampus^93,94.

Acetylcholine is important for REMS: mice with a double knockout of m1 and m3 muscarinic receptor genes have no REMS³⁶. Our study also highlights MS^ChAT cells as key drivers of REMS, theta rhythmicity, and cognitive processing. Similar to cholinergic cells in other basal forebrain nuclei^8,95, we found that MS^ChAT cells are specifically active in WAKE and REMS, as also reported by others⁹⁶. This pattern of activity aligns with their involvement in theta oscillations during these states, as evidenced by the diminished theta power density in both WAKE and REMS episodes of MS^ChAT-App^NL-G-F and MS^ΔChAT mice, which also experienced significantly less REMS. Although REMS generation is classically thought to be primarily generated in the brainstem⁹⁷, and certainly this contains a core part of the REMS circuitry that generates muscle atonia and degenerates in Parkinson’s disease^97,98, many other REMS-generating components have now been found throughout the brain, including in neocortex, basal ganglia, hypothalamus, as well as GABA cells in the medial spetum^{44,68,99–104}. MS^ChAT neurons can now also be added to this list. Importantly, REMS was not abolished but only partially lost in MS^ChAT-App^NL-G-F mice. This is consistent with the distributed REMS-generating model, whereby forebrain cholinergic systems, including MS^ChAT neurons, modulate the quantitative expression and electrophysiological features of REMS.

In our study, control MS^ChAT-GFP mice on a C57BL/6 background exhibited an age-related increase in REMS up to 15 months of age, consistent with previous reports showing preserved or increased REMS at middle age in C57BL/6 mice, with declines occurring later in life¹⁰⁵. The REMS deficit in MS^ChAT-App^NL-G-F mice increases with age, presumably reflecting the gradual degeneration of MS^ChAT cells. Lesions of the MS abolish hippocampal theta¹⁰⁶. Theta activity, including during REMS, is thought to be governed by MS^GABA cells that project to the hippocampus^66–68. Opto-inhibiting MS^GABA cells reduces hippocampal theta⁶⁸, but opto-stimulation of MS^ChAT cells alone failed to elicit hippocampal theta⁹⁶. Despite this, our results indicate that MS^ChAT cells contribute to a component of theta rhythm generation.

While other cell-circuit candidates responsible for the reduced REMS phenotype in App^NL-G-F mice have been proposed, in particular REMS-generating MCH cells in the lateral hypothalamus and ACh neurons in the pontine brainstem^49,107, these associations were correlative and did not definitively implicate those cell types in mediating the REMS decline. Unexpectedly, our results reveal that MS^ChAT neurons may also influence NREMS, accompanied by alterations in EEG spectral power, although the underlying mechanism remains unclear. NREMS disturbances have been observed in App^NL-G-F (although modest) and in other genetic mouse models^60,108, highlighting the broader impact of AD-related pathology on sleep-wake regulation. The fact that, however, in all three mouse models examined (MS^ChAT-App^NL-G-F, App^NL-G-F, and MS^ΔChAT), the remaining REMS bouts were overall longer but less frequent, although the increase in bout length was not consistently observed in MS^ΔChAT mice, may suggest a reorganization within the sleep circuitry, potentially reflecting adaptative changes aimed at preserving aspects of REMS physiology.

Importantly, the question arises about the consequences of less REMS, and also has reduced theta and delta power. Although in mice REMS is dispensable in a laboratory environment³⁶, in humans lower REMS percentages are linked to higher all-cause, cardiovascular, and non-cancer mortality⁴⁰. Both REMS and theta rhythms are hypothesized to contribute to enhancing cognition^41–43, regulating emotions and reducing stress responses^44–46, processes that are compromised in AD and in individuals living with mild cognitive impairment, as well as in animal models exhibiting widespread Aβ accumulation^47–61.

In humans, EEG slowing during REMS may represent a neurophysiological marker distinguishing older adults with memory impairment from cognitively intact peers^61,109–113. However, our findings do not indicate a selective shift in oscillatory slowing per se, but rather a widespread reduction in aperiodic offset across REMS, NREMS, and wakefulness in MS^ChAT-App^NL-G-F mice, with a similar but more state-dependent pattern in App^{NL-G-F/NL-G-F} mice. As the aperiodic offset reflects broadband neural activity and is considered a proxy of global cortical population firing and background cortical gain, these results suggest a reduction in baseline cortical excitability⁷². In line with this interpretation, loss of basal forebrain cholinergic input is known to reduce cortical and hippocampal excitability and population spiking, providing a mechanistic basis for a broadband downward shift in spectral power^15,24. Notably, MS^ΔChAT mice also exhibited a reduction in aperiodic offset, albeit restricted to REMS, indicating that cholinergic loss alone is sufficient to impact this parameter in a state-dependent manner, whereas the more global reduction observed in MS^ChAT-App^NL-G-F mice likely reflects the combined effects of cholinergic dysfunction and amyloid pathology.

Aperiodic components (power decreasing with increasing frequency) vary across vigilance states, supporting a physiological basis for state-dependent modulation of background activity⁷³, and are reliably captured using spectral parameterization approaches that dissociate aperiodic and oscillatory components⁷³. In the context of amyloid-driven network alterations, both modeling and human MEG/EEG studies indicate the coexistence of regional hyper- and hypoexcitability along the Alzheimer’s disease continuum, with inhibitory or homeostatic mechanisms dominating in some circuits and leading to net reductions in broadband activity that are spatially and state dependent¹¹⁴. Thus, rather than reflecting canonical EEG slowing or a uniform state of hyperexcitability, the present results point to a cholinergic-dependent reduction in state-dependent cortical activation, modulated by amyloid pathology, and consistent with emerging views of Alzheimer’s disease as a dynamic condition characterized by region- and stage-specific imbalances in excitation and inhibition^114,115.

Both MS^ChAT-App^NL-G-F and App^NL-G-F mice exhibited reduced anxiety-like behavior, consistent with previous reports in the 3xTg transgenic model of AD¹¹⁶. Because MS^ΔChAT mice showed no change in anxiety, this effect is likely attributable to amyloid in the ventral hippocampus, a region known to regulate anxiety. In our model, the progressive loss of MS^ChAT neurons is expected to markedly diminish cholinergic input to the hippocampus, particularly along the septo-ventral hippocampal axis where we observed dense MS^ChAT innervation. Given that septo-ventral hippocampal cholinergic projections are critically involved in REMS-related theta activity, reduced MS cholinergic tone in our model likely underlies the observed REMS alterations, whereas emotional phenotypes may require a more pronounced cholinergic deficit or the additional presence of amyloid pathology.

MS^ChAT neurons enhance spatial and non-spatial working memory^117,118, and their inputs to the dorsal hippocampus seem more strongly implicated in these processes¹¹⁹. Inhibition or ablation of MS^ChAT projections to the dorsal CA2 region of the hippocampus (where we observed prominent amyloid deposition) impairs social cognition in mice^120,121, which can be restored through the pharmacological increase of cholinergically-driven theta oscillations¹²⁰. Collectively, these behavioral observations underscore the central role of MS^ChAT-hippocampal circuit dysfunction in mediating alterations in risk assessment, working memory, and social cognition. These impairments mirror core symptoms seen in AD patients, including impulsivity, diminished threat perception, short-term memory deficits, and socially inappropriate behavior¹²².

Our findings imply that slow disruptions in MS^ChAT neurons in the early stages of AD gradually reduce REMS, alter NREMS architecture, decrease EEG theta and delta power (with region- and state-specific effects), and produce cognitive impairments and selective alterations in emotional behavior. As suggested by the specific caspase-induced lesions of MS^ChAT cells, neurodegenerative insults affecting these neurons are sufficient to recapitulate key cognitive and sleep phenotypes associated with excessive amyloid production, although emotional alterations appear to be more dependent on the presence of amyloid pathology. Therefore, preserving MS^ChAT cells integrity in a variety of neurodegenerative diseases could help slow decline of patients. Encouragingly, choline supplementation to pregnant App^NL-G-F mice reduces learning and memory impairments and brain Aβ deposition in progeny¹²³; and a year-long administration of an acetylcholinesterase inhibitor protected against prodromal atrophy of MS^ChAT cells in AD patients¹²⁴. This might suggest that keeping MS^ChAT neurons healthier for longer in individuals with AD, or even replacing them, like the approaches being used for dopamine cell loss in Parkinson’s disease, may provide an improved palliative therapy route that could be combined with amyloid-targeting treatments to enhance clinical outcome in AD.

In conclusion, we have provided a circuit-based mechanism and animal model for the prodromal phase of AD, highlighting the selective vulnerability of acetylcholine neurons and offering insight into how early deficits may arise during the prodromal stages of the disease. Our findings support a primary role for a slow loss of MS cholinergic neurons producing cognitive and REMS alterations, while highlighting additional, distinct contributions of amyloid pathology, broadcast from MS cholinergic cells, to hippocampal network dysfunction and astrocyte activation. Our findings underscore the interest in revitalizing the cholinergic hypothesis of AD^5,13,24,25, and suggest that the cholinergic system remains a salient target for the development of disease-modifying interventions⁵.

Methods

Key resources table

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, William Wisden (w.wisden@imperial.ac.uk).

Materials availability

AAVs and plasmids generated in this study have been deposited at Addgene. The accession numbers are listed in the key resources table.

Experimental model details

We used adult App^NL-G-F homozygous mice with the Swedish (KM670/671NL), Beyreuther/Iberian (I716F) and Arctic (E22G) knock-in mutations in the APP gene (these mice were a gift from Takashi Saito, Institute of Brain Science, Nagoya City University Graduate School of Medical Sciences, and Takaomi Saido, RIKEN Centre for Brain Science, Wako, Saitama, Japan)⁷¹. The ChAT-ires-Cre mice (B6;129S6-Chat^tm2(cre)Lowl/J) were from JAX labs (JAX #031661, a gift of Bradford Lowell, Beth Israel Deaconess Medical Center, Harvard)¹²⁵. Animals were maintained under a 12-h light/12-h dark cycle in a controlled environment (21 ± 1 °C; 55 ± 10% relative humidity) with ad libitum access to food and water. Environmental enrichment was provided in the form of PVC tunnels and wooden chew blocks. For all mouse strains, males and females were used in all experiments. All procedures were approved by the Animal Welfare Ethical Review Body at Imperial College London. Experiments were performed in accordance with ARRIVE guidelines and the UK Home Office Animal Procedures Act (1986).

Method details

AAV constructs and production

In the pAAV-hsyn-DIO-App^NL-G-F-P2A-EGFP and pAAV-hsyn-DIO-App^NL-G-F-T2A-mCherry AAV transgenes, the APP^NL-G-F-P2A-EGFP or APP^NL-G-F-T2A-mCherry reading frames, flanked by heterologous Cre recombination sites, are under the control of the human synapsin promoter. The transgenes were made as follows. An App^NL-G-F reading frame, flanked by AscI and AgeI sites, was synthesized by Eurofins (Eurofins Genomics UK Limited) and supplied in a pEX-A258 plasmid. The App^NL-G-F 2267bp band was gel purified from this plasmid as an AscI-App^NL-G-F-AgeI fragment, and ligated into an AAV-hsyn-DIO-P2A-EGFP transgene vector doubly-digested with AscI and AgeI, to generate AAV-hsyn-DIO-App^NL-G-F-P2A-EGFP. A similar procedure generated the pAAV-hsyn-DIO-App^NL-G-F-P2A-mCherry transgene. The transgenes also contained a woodchuck post-transcriptional regulatory element (WPRE) and polyadenylation sequences. These plasmids and their sequences, pAAV-hsyn-DIO-App^NL-G-F-P2A-EGFP and pAAV-hsyn-DIO-App^NL-G-F-P2A-mCherry, have been deposited at Addgene (Addgene plasmids 242648 and 242649, respectively).

We also used the following pAAV transgene plasmids: pAAV-hSyn-DIO-EGFP was a gift from Bryan Roth (Addgene plasmid 50457), pAAV-EF1a-flex-taCasp3-TEVp was a gift from Nirao Shah & Jim Wells¹²⁶ (Addgene plasmid 45580), pAAV-DIO-GCaMP6s transgene was constructed by us and described previously¹²⁷ (Addgene plasmid 184284), pAAV-DIO-ChR2-EYFP was a gift from Karl Deisseroth (Addgene plasmid 20298).

The AAVs consisted of a mixed capsid serotype combining AAV1 and AAV2. To generate AAVs, the adenovirus helper plasmid pFΔ6, the AAV helper plasmids pH21 (AAV1) and pRVI (AAV2), along with the pAAV transgene plasmids, were co-transfected into HEK293 cells. Following this, the resulting AAV particles were purified in-house using heparin columns¹²⁸.

Stereotaxic surgery

Prior to surgical procedures, mice were group-housed and allowed at least one week to acclimatize to the animal facility. Surgeries were conducted at 11-13 weeks of age, after which mice were individually housed for the duration of the in vivo experiments. Mice were randomly assigned to either experimental or control groups. Anesthesia was induced and maintained with 2% isoflurane in oxygen, delivered via inhalation. Preoperative analgesia and anti-inflammatory treatment were administered subcutaneously using buprenorphine (0.1 mg·kg^-1) and carprofen (5 mg·kg^-1). Animals were then positioned and secured in a stereotaxic apparatus (Angle Two; Leica Microsystems) for the duration of the surgical procedure. Postoperatively, carprofen was provided in the drinking water (5 mg·kg^-1 per day), and mice were allowed to recover in their home cages for a minimum of 3 weeks prior to further experimentation.

AAV Injections

Viral vectors were delivered using a stainless steel 33-gauge internal cannula (13 mm, PST3; Hamilton Company) connected to a 10-μl Hamilton microsyringe. Injections were performed at a rate of 0.1 μl·min^-1, with the following constructs and volumes: pAAV-hsyn-DIO-App^NL-G-F-P2A-EGFP (1 μl), pAAV-hSyn-DIO-EGFP (1 μl), a combination of pAAV-EF1a-flex-taCasp3-TEVp and pAAV-hSyn-DIO-EGFP (0.5 μl each), pAAV-DIO-GCaMP6s (0.5 μl), and pAAV-DIO-ChR2-EYFP (1 μl). The cannula was slowly lowered to the target site in the MS (coordinates: 0.05 mm ML, 0.8 mm AP, -4.4 mm DV relative to bregma). Following injection, the cannula was left in place for 5 min before being withdrawn slowly to minimize backflow.

Fiber photometry implantation

In mice receiving AAV-DIO-GCaMP6 injections, a 200-μm diameter, 0.5 NA (numerical aperture) mono-fiber optic cannula (RWD Life Science) was surgically implanted 100 μm above the injection site in the MS and secured with a cured acrylic resin (Simplex Rapid powder and Rapid Liquid, Kemdent). Mice were allowed to recover for 3 weeks prior to photometry recording.

Headstage implantation for EEG/EMG recordings

All mice were implanted with two miniature screw electrodes for EEG recording. Electrodes were prepared by soldering ultra flexible PVC lead EEG wires (NUF30-4046BLACK, Cooner Wire) to the screws and positioned at standard locations: frontal cortex (-1.7 mm ML, +1.5 mm AP relative to bregma) and parietal cortex (-1.7 mm ML, +1.0 mm AP relative to lambda). EMG electrodes (stranded stainless-steel wire; AS634, Cooner Wire) were inserted bilaterally into the neck musculature. All wires were soldered to a 7-pin header, which was secured to the skull using a cured acrylic resin (Simplex Rapid powder and Rapid Liquid, Kemdent).

Sleep-wake analysis

Continuous EEG and EMG recordings were acquired for a minimum of 48 h using untethered Neurologger 2A devices¹²⁹, which were connected to a 7-pin headstage implant. Data were recorded at a sampling rate of 200 Hz with fourfold oversampling. In MATLAB (MathWorks), EEG signals were processed using a high-pass digital filter (highpass function; cutoff frequency: 0.5 Hz, -3 dB), while EMG signals were filtered using a band-pass digital filter (bandpass function; 5-45 Hz, -3 dB).

Sleep scoring

For sleep state classification, 24-h EEG/EMG recordings were analyzed, comprising 12-h light and 12-h dark phases. These recordings were collected 3 times at 4-month intervals following AAV injection or from birth in both App^{NL-G-F/NL-G-F} and App^WT mice. For MS^ΔChAT mice, recordings were collected 6 months post-injection. Vigilance states, i.e. non-rapid eye movement sleep (NREMS), rapid eye movement sleep (REMS) and wakefulness, were manually scored in 5-s epochs using SleepSign for Animal software (KISSEI COMTEC Co.) by an experienced experimenter blinded to treatment conditions.

EEG power spectral analysis

EEG power spectral densities were computed for successive 5-s epochs using the Welch method in MATLAB (pwelch function; frequency range: 1-15 Hz; resolution: 0.2 Hz; non-overlapping Hanning window), a more robust and less noisy variant of fast Fourier transform. Epochs containing artifacts were excluded from spectral analyses. Power spectra were calculated separately for NREMS, REMS, and wakefulness during both light and dark phases. To further characterize spectral features, power spectra were parameterized using the Fitting Oscillations and One-Over-F (FOOOF) algorithm, implemented in Python⁷³. This approach decomposes the power spectrum into an aperiodic (1/f-like) component and superimposed periodic oscillatory peaks, allowing separation of broadband background activity from true oscillatory signals. The aperiodic component was described by its offset, reflecting the overall broadband power level, and its exponent, reflecting the slope of the spectrum and the relative distribution of power across frequencies. Periodic (oscillatory) activity was quantified after removal of the aperiodic component by integrating power within predefined frequency bands. Delta (1-4.4 Hz) and theta (6-9 Hz) periodic power were extracted for each vigilance state and phase. In parallel, a conventional spectral analysis was performed. For each epoch, spectral power within each 0.2 Hz EEG frequency bin was normalized to the total spectral power of that epoch and expressed as a percentage of total EEG power. Delta power and theta band power were then calculated by summing normalized power within the respective frequency ranges.

Interictal epileptiform spikes were identified using an automated detection algorithm based on a previously validated method⁷². Briefly, the approach combines band-pass filtering with envelope-based feature extraction to detect transient events meeting amplitude and temporal criteria consistent with epileptiform discharges.

Calcium photometry

To assess the spontaneous activity of MS cholinergic neurons across sleep-wake states (NREMS, REMS, and wakefulness), we employed fiber photometry using the genetically encoded calcium indicator GCaMP6s, as previously described^127,130. Simultaneous EEG/EMG recordings were acquired to accurately determine vigilance states during photometry sessions. Recordings were conducted minimum 3 weeks post-surgery during the light (rest) phase between zeitgeber time (ZT) 0 and ZT6, with each session lasting 3-6 h. Each animal underwent 3 to 4 recording sessions.

Photometry recording

A 473-nm diode-pumped solid-state blue laser (ADR-800A, Shanghai Laser & Optics Century Co.) was used to excite GCaMP6s. The excitation light was routed through a single-source fluorescence cube (FMC_GFP_FC, Doric Lenses) via an optical fiber patch cord (Ø 200 μm, 0.22 NA, Doric Lenses). Emitted fluorescence passed through the same cube and was transmitted via a multimodal optical patch cord (Ø 200 μm, 0.37 NA, Doric Lenses) connected to the ceramic-implanted fiber optic cannula (Ø 200 μm, 0.37 NA, Doric Lenses) using a split mating sleeve ferrule (Thorlabs). The GCaMP6s emission signal was filtered between 500-550 nm and detected by an amplified photodiode (APD-FC, Doric Lenses). The photodiode output was relayed to a lock-in amplifier (SR810, Stanford Research Systems), which also modulated the laser at 125 Hz. Laser power was maintained at 80 μW at the fiber tip. The demodulated signal was digitized using a CED 1401 Micro Box (Cambridge Electronic Design) and recorded at a sampling rate of 1 kHz using Spike2 software (Cambridge Electronic Design).

Neuronal activity analysis

All photometry data were analyzed using MATLAB. Signals were temporally aligned with the EEG/EMG recordings. Analyses focused on 100-s windows enclosing transitions between vigilance states (NREMS to REMS, NREMS to wakefulness and wakefulness to NREMS). The raw fluorescence signal (F) was normalized to a baseline using the formula¹²⁷:

Following ΔF/F normalization, photometry traces were smoothed on a per-transition basis using a Savitzky-Golay filter (3^rd-order polynomial, 1 s window) implemented in the signal package in R, to reduce high-frequency noise while preserving the slower dynamics of cholinergic signals, and were then averaged across transitions. In some recordings, an initial transient decay in signal intensity was observed; therefore, only sessions with stable photometry signals following this decay period were included in the final analysis.

Histological analysis

Immunohistochemistry

Mice were transcardially perfused with 4% paraformaldehyde (Thermo Scientific) in phosphate-buffered saline (PBS; Sigma). MS^ChAT-App^NL-G-F and MS^ChAT-GFP mice were perfused 13-14 months after AAV injection, whereas App^{NL-G-F/NL-G-F} and App^WT mice were perfused at 13-14 months of age, corresponding to comparable durations of amyloid pathology. At this late time point, in MS^ChAT-App^NL-G-F mice injected with AAV-DIO-App^NL-G-F-2A-EGFP, only sparse and morphologically compromised GFP signal was detectable by immunostaining at the injection site, consistent with the progressive loss of AAV-transduced ChAT-Cre neurons driven by sustained intracellular amyloid burden. MS^ΔChAT mice were perfused 7-8 months post-injection, and ChAT-Cre mice used for cholinergic projection mapping were perfused 1 month after AAV injection. Brains were extracted, post-fixed in 30% sucrose in PBS for cryoprotection, and later coronally sectioned (40 μm thickness) using a HM 450 sliding microtome (Thermo Scientific). Free-floating sections were washed 3 times in PBS (5 min each), followed by epitope retrieval in 0.05% Tween-20 in 10 mM sodium citrate buffer (pH 6.0) at 80-85 °C for 30 min. Sections were then cooled at room temperature for 15 min and washed 3 times in PBS (10 min each). Non-specific binding was blocked by incubating the sections for 1 h in PBS containing 0.2% Triton X-100 and 20% normal donkey or goat serum. Sections were incubated overnight at 4 °C on a shaker with one or two primary antibodies diluted in PBS supplemented with 2% normal serum. The following day, sections were washed 3 times in PBS (10 min each), incubated for 1.5 h at room temperature with the appropriate secondary antibody diluted in PBS with 2% normal serum, and washed again 3 times (10 min each). When two primary antibodies were used, each corresponding secondary antibody incubation was performed sequentially, with intermediate PBS washes (3 times, 10 min each), to prevent cross-reactivity. For triple labeling, a third primary antibody was applied, followed on the subsequent day by the corresponding secondary antibody as described. Nuclear staining was performed by incubating the sections in Hoechst (1:5,000; Invitrogen) for 3 min at room temperature, followed by three washes in PBS (5 min each). The primary antibodies used included chicken anti-GFP (1:1,000; abcam, ab13970), goat anti-ChAT (1:200; Merck, AB144P), rabbit anti-D54D2 (1:500; Cell Signaling Technology, 8243), mouse anti-6E10 (1:1,000; BioLegend, 803001), rabbit anti-laminin (1:350; Merck, L9393), rabbit anti-GAD67 (1:500; Invitrogen, PA5-21397), guinea pig anti-VGluT2 (1:200; Merck, AB2251-I), chicken anti-GFAP (1:1000; abcam, ab4674) and mouse anti-NeuN (1:1000; abcam, ab104224). Secondary antibodies (1:500; Invitrogen) included Alexa Fluor 488-conjugated donkey anti-chicken (A78948), donkey anti-mouse (A-21202), and goat anti-chicken (A-11039), Alexa Fluor 568-conjugated donkey anti-goat (A-11057), donkey anti-mouse (A10037), and goat anti-guinea pig (A-11075), as well as Alexa Fluor 647-conjugated donkey anti-rabbit (A-31573) and goat anti-rabbit (A-21245). Sections were mounted on glass slides using ProLong Gold antifade mounting medium (Invitrogen) and cover-slipped.

Data analysis and quantification

Images were acquired using a Zeiss AxioObserver inverted widefield microscope, controlled with Zen Pro software (Zeiss) at the Facility for Imaging by Light Microscopy, Imperial College London. Illumination was provided by a Lumencor SpectraX LED light source with the following excitation-emission filters: 390/40-450/40 (Hoescht), 470/40-525/50 (Alexa Fluor 488), 550/25-605/70 (Alexa Fluor 568), and 640/30-690/50 (Alexa Fluor 647). Emission was collected with a sCMOS Hamamatsu ORCA-Flash4.0 camera coupled to objectives selected according to the experiment, including 10x/0.3 Ph1 EC Plan-Neofluar, 20x/0.8 DIC Plan Apochromat, and 40x/0.75 Ph2 EC Plan-Neofluar. For large region acquisitions, tiles were configured to overlap by 10% and manual focus maps were generated. Post-acquisition, images were stitched using Zen Pro software. Histological analyses were conducted using the Fiji distribution of ImageJ. Core preinstalled functions were employed for unbiased quantification of neuronal counts and stained areas, utilizing automatic thresholding algorithms included in the Fiji package. Co-localization analyses were performed using the BIOP JaCoP plugin, allowing for quantitative assessment of signal overlap between markers.

Behavioral and baseline stress analysis

All behavioral assessments were conducted 13-14 months post-surgery or afterbirth, and during the animals’ active (dark) phase under red light, specifically between Zeitgeber Time (ZT) 16 and ZT22. Animal behavior was recorded using an overhead camera and analyzed automatically in a genotype-blind manner with the ANY-maze software (Stoelting Co.), except for the light-dark box test which was conducted using the Zantiks LT automated apparatus (Zantiks), ensuring objective and unbiased assessment. Video tracking enabled precise quantification of locomotion, orientation, activity, and total distance traveled.

Y-maze

The Y-maze task was employed to evaluate spontaneous alternation behavior, an index of spatial working memory¹³¹. Each mouse was placed at the center of a Y-shaped maze (arm length: 30 cm, width: 6 cm, wall height: 15 cm), with the orientation of entry arms counterbalanced across subjects. Arm entries and total distance traveled were recorded over a 10-min session. A spontaneous alternation was defined as consecutive entries into three different arms. The percentage of spontaneous alternation was calculated as:

Novel object recognition task (NORT)

To assess non-spatial working memory, the NORT was performed as described previously¹³². During the habituation phase, mice were allowed to explore an empty square arena (45 cm × 45 cm, height: 40 cm) for 10 min. This session also served as the open field test for evaluating anxiety-like behavior (see below). Twenty-four hours later, in the familiarization phase, two identical objects were placed 5 cm from the arena walls, and mice were allowed to explore until a total of 20 s of object interaction was reached. This criterion ensured consistent exploration across animals, irrespective of their inherent activity levels. Another 24 hours later, one familiar object was replaced with a novel one, and mice were again allowed to explore both objects until 20 s of cumulative exploration time was reached. Object type and positioning were randomized between animals and experimental groups, and the time to reach the 20 s exploration criterion was recorded.

Open field

Anxiety-like behavior was evaluated using the open field paradigm. Mice were placed in the center of a square arena (45 cm × 45 cm, height: 40 cm) and allowed to explore freely for 10 min. The time spent in the central zone, defined as 25% of the total area and considered anxiogenic, were measured, as well as the number of entries into the central zone and the total distance traveled. This session also constituted the initial habituation phase for the NORT (see above).

Light-dark box

Bright-light-induced anxiety was assessed using the light-dark box test. The setup included a white bright compartment (23.5 cm × 27 cm, height: 30 cm; ∼150 lux illumination from a translucent floor) connected to a dark compartment (11.5 cm × 27 cm, height: 30 cm) via a 5 cm semi-circle opening. The dark compartment was covered with a removable black semi-translucent lid. Mice were introduced into the center of the dark compartment, facing away from the connecting opening, and allowed to explore both compartments for 5 min. Time spent in the illuminated compartment, the number of transitions between compartments and the total distance traveled were recorded.

Home-cage locomotor activity

Spontaneous locomotor activity in the home cage was continuously monitored for an average of 6.8 consecutive days using custom-built devices modeled after a previously published protocol¹³³. Each unit consisted of a passive infrared sensor connected to a customized microcontroller (M0 AdaLogger, Adafruit #2796), housed within a 3D-printed casing containing a lithium-ion battery for wireless operation. Devices were placed in the cage on the wire lid, allowing uninterrupted monitoring of activity (total activity duration per min) without disturbing the animals.

Assessment of stress hormone levels

Basal stress levels were assessed via fecal corticosterone metabolite (FCM) concentrations using a commercial ELISA kit (ADI-901-097, Enzo), following a previously described protocol optimized and validated in our laboratory¹³⁴. Feces were collected over a 28-h period, divided into five ZT-defined windows: ZT20-ZT0 (final 4 h of dark phase), ZT0-ZT12 (entire light phase), ZT12-ZT16, ZT16-ZT20, and ZT20-ZT0. One week prior to sample collection, animals were habituated to alternating between two uncleaned cages (containing their own scent, enrichment and bedding material), minimizing environmental stress during sampling. At the beginning of the collection period, mice were transferred to the other previously occupied but feces-free cage, and subsequent swaps were performed at each ZT interval. Feces were collected from the previously used cage after each swap and stored at -20 °C until processing. Each sample was homogenized in 5 mL of 95% ethanol per gram of feces, thoroughly ground, and incubated overnight at room temperature. Samples were centrifuged at 3,000 × g for 20 min; the supernatant was collected and centrifuged again at 10,000 × g for 15 min. A 200 μL aliquot of the final supernatant was evaporated to dryness in an open tube under a fume hood. The residue was then reconstituted in 250 μL of assay buffer (provided with the ELISA kit), vortexed, incubated overnight at 4 °C, and stored at -20 °C until analysis. All samples were diluted 1:10 prior to assay and processed in accordance with the manufacturer’s instructions.

Quantification and statistical analysis

All statistical analyses were conducted using R (R Core Team) in RStudio (Posit). A priori power analyses were performed prior to the start of the study using previously collected in-lab data to estimate the required sample size for achieving adequate statistical power (typically α = 0.05 and power ≥ 80%). All datasets satisfied the assumptions necessary for parametric testing, including normality and homoscedasticity, assessed through residual inspection and/or standard tests (e.g., Shapiro-Wilk for normality and Levene’s test for equality of variances, where appropriate).With the exception of photometry recordings and the NORT, which were analyzed using paired and one-sample Student’s t-tests respectively, all other data were analyzed using either standard ANOVA (via the lm function in base R) when no random effects were present, or linear mixed-effects models (via the lmer function from the lme4 R package) when accounting for random factors, such as repeated measures from the same subject or anatomical variability across brain section locations in histological analyses. In this framework, random intercepts were included for both individual animals and atlas-defined brain section level to account for repeated measurements and residual variability across rostro-caudal sampling positions. Although not all animals contributed data to every section level, this approach allowed us to retain spatially resolved measurements without averaging across sections, while accounting for non-independence of observations.

Fixed effects included experimental group, sex, and time (for repeated measures), modeled as interacting categorical variables, thereby allowing the assessment of group × sex interactions, as well as brain region in the analysis of 6E10/laminin colocalization staining. Given that the primary aim of this study was to define group-related phenotypes across the full cohort, results are presented as pooled male and female comparisons for clarity and interpretability. Full statistical outputs, including sex-stratified analyses and interaction effects, are reported in the Supplementary Information. Post hoc pairwise comparisons were conducted using Holm-Bonferroni correction for multiple testing, implemented through the lsmeans and contrast functions from the emmeans R package.

Bivariate associations were assessed using partial Kendall’s tau correlations, controlling for group and sex, as implemented in the R package ppcor.All statistical tests were two-sided, and exact p-values are reported unless otherwise specified. When models included time as a factor, statistical reporting in the main text was restricted to overall group effects to maintain clarity, with detailed pairwise comparisons across time points available in the figures and Supplemental information. No data points were excluded unless predefined technical quality criteria were not met (e.g., insufficient signal quality or failed recordings), in which case the affected measurements were omitted from the corresponding analyses. Sample sizes (n) refer to individual animals unless otherwise indicated. Data are presented as mean ± standard error of the mean (SEM). Full statistical results are provided in the Supplemental information.

Supplemental Figures

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Figure S5.

Figure S6.

Figure S7.

Data availability

All raw data generated during this study and used to calculate the results reported herein, along with the results of all statistical analyses conducted, as well as bespoke scripts used for sleep analysis, calcium photometry, and histological analyses (developed in MATLAB, R, and Fiji/ImageJ), will be deposited in a public repository upon publication. The accession number will be listed in the key resources table.

Acknowledgements

Preparation for this article was supported by the UK Dementia Research Institute (award number UK DRI-5004) through UK DRI Ltd, principally funded by the Medical Research Council (W.W., N.P.F.); the Wellcome Trust (220759/Z/20/Z, to N.P.F. and W.W.); and the Medical Research Council (S.W., N.P.F and W.W., APP9329). The Facility for Imaging by Light Microscopy (FILM) at Imperial College London is part-supported by funding from the Wellcome Trust (104931/Z/14/Z) and BBSRC (BB/L015129/1). For the purposes of Open Access, the authors have applied a Creative Commons Attribution (CC BY) license to any Author Accepted Manuscript Version arising.

Additional information

Author contributions

M.N. performed behavioral experiments, surgeries and AAV injections, EEG and behavioral analysis, histology, all general data analysis and assembled and designed all figures.

W.B. performed surgeries and calcium photometry analysis.

B.A.S. performed surgeries and helped with histology experiments.

C.Y. helped with histology experiments.

L.L. helped with stress hormone level analysis.

K.J. helped with histology experiments.

S.W. helped with histology experiments.

A.L.V. provided the Neurologger 2A devices.

R.Y. performed all molecular biology and AAV transgene construction, AAV production, mouse genotyping.

M.N., N.P.F and W.W. conceived and designed the experiments. N.P.F. and W.W. co-supervised the project. M.N., N.P.F. and W.W. co-wrote the paper.

Funding

UK Dementia Research Institute (UK DRI) (UK DRI-5004)

• Mathieu Nollet

Wellcome Trust (WT)

https://doi.org/10.35802/220759

• William Wisden

UKRI | Medical Research Council (MRC) (APP9329)

• William Wisden