Local and global dichotomic dysfunction in resting and evoked functional connectivity precedes tauopathy

a Singapore Bioimaging Consortium, Agency for Science, Technology and Research, Singapore b Faculty of Biology, Medicine and Health, The University of Manchester, Manchester, United Kingdom c Department of Radiology and Bioimaging Sciences, Yale School of Medicine, New Haven, CT, USA d Department of Life Sciences, Yeungnam University, Gyeongsan, Gyeongsangbuk‐do 38541, South Korea e Department of Pharmacology, Yong Loo Lin School of Medicine, National University of Singapore f Department of Radiology and Nuclear Medicine & Donders Institute for Brain, Cognition, and Behaviour, Donders Institute, Radboud University Medical Centre


Introduction
Tauopathies are a hallmark in many neurodegenerative disorders, including Alzheimer's disease (AD) 1 . To date, clinical trials targeting later stages of AD pathology have all either failed or made symptoms worse 2 . As such, the focus is now placed on the earliest signs and predictive biomarkers for disease progression. Synaptic dysfunction is one of the earliest events taking place prior to AD 3 . This has been revealed by abnormal functional Magnetic Resonance Imaging (fMRI) patterns during tasks, or at rest, in groups at risk for AD, such as mid-life APOEε4 risk-allele carriers 4 , or amyloid-positive but cognitively-healthy elderly 5 , thus, confirming the need to focus on the pre-pathological stages.
Despite a strong interest in these biomarkers, the underlying mechanisms supporting abnormal connectivity changes are not known. Transgenic animals bearing mutations from familial AD and tauopathies also develop several of the distributed neuronal network dysfunctions found in AD, such as in the early stages of cerebral amyloidosis [6][7][8] ; as such, AD models may provide important mechanistic insights. To understand the physiological basis underlying resting-state fMRI (rsfMRI) dysfunction during pre-tauopathy, we used the triple-transgenic mouse model for AD (3xTgAD) 9 .
3xTgAD mice were positive for phospho-tau, a precursor for the hallmark tauopathy of neurofibrillary tangles, in the amygdala and hippocampus, at 3 and 6 months of age.
The local connectivity loss reported at the same age points was at the stem of larger neuronal network dysfunction. Furthermore, the deficits in local connectivity showed a remarkable overlap with homologous networks affected by tauopathies in humans. We then employed optogenetics in combination with fMRI (ofMRI) to examine the evoked response from photostimulation of the lateral entorhinal cortex (ENTl), a central dysfunctional node in 3xTgAD 10,11 and AD patients [12][13][14] . This revealed a hyperemic response in 3xTgAD relative to wild-type mice in several distal projection areas.
Our observations underscore several of the physiological underpinnings behind local and distal connectivity dysfunctions commonly observed in groups at risk of developing AD and other tauopathies, thus supporting a reconciliation for the apparent discordant results put forward in early-AD subjects 4,5,15 . This study provides a trans-species neurophysiological model for early network failures, which cascade into some of the most severe neuropathologies affecting the aging population.

The ventral-amygdaloid-striatal network is affected during tauopathy in mice and humans
To examine spontaneous fluctuations in brain activity, we recorded rsfMRI in male 3xTgAD and wildtype control mice on the same background strain (129sv/c57bl6) at 3 (N3xTgAD = 19, Ncontrols = 10) and 6 months of age (N3xTgAD = 13, Ncontrols = 10), longitudinally. The rsfMRI protocol employed here was recently compared to others in a multicenter study, which indicated elevated sensitivity and specificity for resting-state networks detected in this dataset relative to other protocols, including an awake mouse protocol 17 . One 3xTgAD mouse developed hydrocephalus, which, despite the acute condition, only marginally affected functional connectivity 18 . extracellular Aβ and tangle deposition. The spatial distribution of these dysfunctions confers added relevance to the 3xTgAD model for the study of early emotional-memory deficits found in patients affected by mild cognitive impairment. Importantly, deficits within the ventral-amygdaloid-striatal system are consistent with behavioral results in young 3xTgAD, where fear and emotional processes are highly affected 23 . Emotional control, regulated by the hippocampal-prefrontal-amygdaloid system, is also affected in pre-AD patients 23,24 , further highlighting the trans-species relevance of our results. Moreover, not all brain areas were affected in comparable manner: the somatosensory cortex of 3xTgAD presented elevated ReHo, reminiscent of previous findings in APP transgenic model 25 . This highlights that pathophysiology does not affect each brain region equally. . c) Pair-wise ROI interactions relative to the left ENTl: decrease of functional connectivity in contralateral ENTl, ACB, BLA at both ages. Increased functional connectivity in somatosensory regions. d) Paired-pulse electrical stimulation of the ENTl is recorded as fEPSPs in the BLA: increased facilitation in 3xTgAD, at both ages. Response to ENTl stimulus pairs calculated as the paired-pulse index (PPI) was reported for intervals of 20, 50, 100, 200, 500 and 1000 ms. * p < 0.05, ** p < 0.005, ***p < 0.001 (3 months old: Ncontrols = 7, N3xTgAD = 6; 6 months old: Ncontrols = 4, N3xTgAD = 4). Immunohistochemistry (AT8/DAPI) reveals phosphorylated tau in the BLA, at both ages (age-matched insets). Top-left inset: example of raw data for control and 3xTgAD in BLA for pulse 1 (P1) and pulse 2 (P2) at 3 months. SSp-II: somatosensory area, lower limb; ENTl: lateral entorhinal cortex; BLA: basolateral amygdala; ACB: nucleus accumbens; fEPSP: field excitatory postsynaptic potential. Scale bar: 100 μm. Changes in dopamine signaling were reported previously in an animal model of cerebral amyloidosis also overexpressing APPswe. In addition, loss of midbrain dopamine (DA) neurons, as well as deficits in hippocampus-to-ACB signaling mediated by DA, has been observed at 3 months of age in Tg2576 mice 26,27 . In AD, alterations in DA levels or DAergic receptors are found to significantly impact synaptic plasticity and hippocampal-memory encoding 28 . Loss of DA receptors, especially D2, has been shown in areas such as the hippocampus, prefrontal cortices and BLA 29,30 in AD patients. PET studies on AD patients also confirm a loss of striatal D2-like receptors 31 . Our results, therefore, bring supporting evidence for an interaction between the DAergic system associated with early cerebral amyloidosis and tauopathy, which leads to synaptic dysfunction.

Local functional connectivity deficits translate into whole-brain network alterations
The ENTl and associated hippocampal areas are fundamental for declarative memory encoding and retrieval 32,33 . Particularly, the ENTl is among the first hubs affected in both human AD 34 and the 3xTgAD model of cerebral amyloidosis and tauopathy 9 (Fig. 1a). As such, the ENTl was targeted for further analysis. To examine distal functional connectivity alterations at rest, we assessed pair-wise ROI interactions relative to the left-hemisphere ENTl. Functional connectivity to the ENTl was decreased in the retrohippocampal and hippocampal regions (Fig. 1c, and e.g. ENTl right hemisphere: Supplementary Fig. 5a), ventral striatum (ACB, Supplementary Fig. 5b), and amygdala (BLA, Supplementary Fig. 5c).
Similarly, an increase in functional connectivity relative to the ENTl was reported in the somatosensory cortex (lower limb, SSp-ll; Supplementary Fig. 5d). These results, focusing on the ENTl-specific network, show similarities to the whole-brain functional connectivity changes assessed with ReHo, presented in an overlapped design in Supplementary Fig. 6.
To confirm the connectivity results, we performed electrophysiological recordings in urethaneanesthetized 3xTgAD and control mice, in vivo, at 3 or 6 months of age. Field excitatory postsynaptic potentials (fEPSP) were assessed within the BLA and dentate gyrus (DG), following electrical stimulation in the ENTl. Paired-pulse stimulation (PPS) protocol, used to assess short-term synaptic plasticity changes, was analyzed through the paired-pulse index (PPI) and revealed a quadruple effect  Fig. 8b, right panel). Taken together, our electrophysiological evidence, assessed in vivo, reveals a hyperexcitable behavior during the evoked neuronal response in disease-relevant 3xTgAD brain regions, suggesting a dichotomous relationship between increased-evoked and reducedspontaneous activity in AD-like vulnerable areas, where functional connectivity is highly compromised at rest. In an exploratory analysis, we examined whole-brain network deficits in 3xTgAD mice at 3 and 6 months of age ( Supplementary Fig. 9a). Alterations were consistent between both age groups, localized within and between regions highlighted in the ReHo analysis, namely, in the amygdaloid/cerebral nuclei (including BLA and ACB), the ENTl and the hippocampal formation ( Fig. 1a, Supplementary Fig. 9b). Importantly, the nodal degree distribution, i.e., the number of affected connections per ROI, was found to overlap with ReHo ( Supplementary Fig. 9b). This indicates that local functional connectivity deficits translate into distal functional connectivity deficits and, in turn, greater network dysfunction. Moreover, regions highlighted in the pair-wise correlation analysis were also found to be enriched in tau aggregates within the AD-like disease progression (amygdala and hippocampus, Supplementary Fig. 2), consistent with tau dispersions across functionally connected networks 35 .
Prolonged photostimulation (500 ms) applied to patched neurons in vitro confirmed an effective ChR2-induced inward current on both control and 3xTgAD mice post hoc ( Supplementary Fig. 11d).
There was no evidence of aberrant spontaneous behavior to photostimulation protocols in freely behaving mice, unlike seizures previously reported following photostimulation of the hippocampus in rats 37 . An unbiased voxel-wise analysis revealed the BOLD signal associated with our modeled response in controls and 3xTgAD at 3 months of age ( Interestingly, optogenetically-evoked activity was mostly confined to the ipsilateral hemisphere, despite the presence of contralateral projections, as predicted by viral tracers (Fig. 2e bottom row, spatial correlation r = 0.36). This supports the notion of a neuronal mechanism that silences the response contralaterally but not ipsilaterally to artificially generated neuronal activity, perhaps via feed-forward active inhibition 38 . The response elicited through photostimulation of ENTl was comparable between mice at 3 and 6 months of age, indicating a stable expression allowing for longitudinal analysis. A negative control carried out in healthy wild-type mice (NmCherry-controls = 9) transfected with mCherry alone (Fig. 2b) did not reveal the presence of a light response, except for a visual-associated response of the lateral geniculate nucleus and superior colliculus, probably due to the fiber illumination received as direct visual response to retinal illumination ( Supplementary Fig. 10d). Hence, we concluded that the response recorded with ofMRI was not associated with potential heating and/or vascular photoreactivity artifacts. Photostimulation at frequencies ranging from 5 to 20 Hz indicated spatially overlapping results ( Supplementary Fig. 10e), in contrast to previous research 39 . In fact, the nonspecific, visually-associated response amplitude was stronger at lower frequencies (Supplementary

Potentiated hemodynamic response and neuronal activity in 3xTgAD
To assess response differences across the brain, a non-parametric second-level analysis comparing the amplitude of activation between 3xTgAD and controls was carried out (Fig. 3, central segment).
Group size at 6 months of age was reduced due to group attrition, e.g., detachment of the implant.
Therefore, group sizes differed between 3 months (Ncontrols = 10, N3xTgAD = 12) and 6 months  Fig. 13). Central panel: Two-sample t-test showing significantly higher response (p < 0.05, corrected) in 3xTgAD compared to controls in AD-like vulnerable regions, such as mPFC and DG, left and right sections respectively. a,d) Representative action potential firing patterns to 120 pA injection in ILA pyramidal cells (Ncontrols = 4, n = 10 / N 3xTgAD = 4, n = 11) and DG granule cells (Ncontrols = 4, n = 29 / N 3xTgAD = 5, n = 18) respectively. b, e) The evoked spike number during various injected currents in ILA pyramidal cells and DG granule cells, respectively. The data are plotted as mean action potential numbers ± 1 SD. The statistical significance is presented with asterisks (*p < 0.05, **p < 0.01 by Mann-Whitney U test). c, f) COPEs are represented for 3xTgAD and controls at both age points for mPFC and DG, respectively.
To confirm the ofMRI results, we examined neuronal excitability ex vivo in two projection areas highlighted above, namely, the hippocampus, specifically DG, and mPFC, specifically the infralimbic area (ILA). Acute brain slice electrophysiology indicated that excitatory neurons of 3xTgAD, in both ILA (Fig. 3ab) and DG (Fig. 3de), were prone to increase the number of spikes derived by current injection compared to controls, but ILA neurons showed the enhancement of afterhyperpolarization (AHP) latency and half-width of the action potential, a phenotype consistent with previous results 40 (Supplementary Table 1), while spike morphology was comparatively unaltered (Supplementary Table   2). To examine how the alterations of the ENTl functional projectome at 3 months related to a loss of functional connectivity at rest, we projected the two responses onto the same template ( Supplementary   Fig. 12, right panel). There, we found that several of the ROIs, presenting a decreased functional connectivity at rest, were responding with greater amplitude to optogenetically-driven neuronal activity. This was the case for regions encompassing the ventral striatum (ACB), the dorsal DG within the hippocampal formation and prefrontal regions.
We concluded, therefore, that disturbed distal connectivity at rest translated to heightened hemodynamic response to elicited neuronal activity with optogenetics, and that this response is accompanied by hyperactivity of the embedded neurons within the affected circuits.

Discussion and conclusions
Neurological disorders, including tauopathies such as AD, are one of our greatest challenges in modern medicine. The absence of disease-modifying treatment for AD represents a major loss for the millions of patients affected by the pathology worldwide. A detailed understanding of the disease mechanisms across spatial and temporal scales constitutes a translational opportunity to facilitate the drug development process. Here, we have determined that the pre-tangle stage of the 3xTgAD mouse model presents functional connectivity patterns overlapping with areas affected by tauopathy in humans. This supports the trans-species relevance of our results.
We determined that distal connectivity disturbances could be explained by local connectivity deficits.
Moreover, we observed a dichotomy between resting activity, that leads to decreased connectivity, and evoked activity that promotes increased metabolic demand and neuronal hyperexcitability. We confirmed fMRI results with electrophysiological recordings that indicated a pathological effect on signal transmission and electrical properties of neurons in vulnerable neuronal circuits. Hence, we demonstrated that fMRI connectivity deficits are rooted in a deeper physiological context. Finally, an in-silico transcriptome analysis indicated a potential association of these results with deficits in the DA system. Importantly, our results connect to two widely examined models of AD pathological progression, namely the tau seeding hypothesis 41 , and a revised amyloid hypothesis 25 . Finally, we examined potential mechanisms underpinning the electrophysiological signatures. Staging the disease progression has been an important question in AD in order to understand mechanisms and to improve diagnostics 42 .
We observed phospho-tau spreading from 3 to 10 months in 3xTg AD model mice and found that this was associated with long-range connectivity deficits in young model animals. This supports the notion of a tight coupling between functional connectivity and tau progression 35 , supported by molecular work by others 43 . The seeding hypothesis can be further connected to other models of axonal degeneration and inflammatory response 44 and, thus, provides a coherent description of the pathophysiology process taking place in AD.
The contribution of different Aβ species (soluble oligomers, fibrils, plaques) has been the subject of an ardent discussion within the community. Here, we demonstrate functional connectivity dysfunction in the absence of amyloid plaques and detectable elevated Aβ levels, similar to previous results in other models 6,7,25 and in subjects at risk of developing AD 4,15 . These results and others contribute to the view that the traditional amyloid cascade hypothesis for AD etiopathogenesis should be updated 45 .
Our results contribute to a modern re-interpretation of the cascade and reconcile contradicting results within the human literature. Decreased resting connectivity patterns translated here into an increased response to optogenetically-driven and electrically-driven neuronal activity, thus, highlighting a possible increase in neuronal excitability following stimulation and increased metabolic demands (Fig.   3, Supplementary Fig. 12 right panel).
The dichotomy of the direction of these changes in resting and evoked activity, specifically within the ENTl network, mirrors several findings in preclinical and mild AD patients. Decreased functional connectivity is found in mild AD patients at rest, in areas related to the DMN, including the hippocampal formation 46 . However, task-based fMRI studies show increased activity in memoryrelated areas in subjects at risk of AD but cognitively still normal (e.g. APOEε4 carriers) suggesting a possible dichotomy in network organization of the brain at rest and in engaged status 47 25 . Our results support the notion that local and distal network dysfunction at rest impairs information transmission and processing. This leads to increased metabolic demand during evoked activity, which putatively leads to circuit exhaustion and further accumulation of Aβ species through increased neuronal activity 50 . It will be important to confirm this model prediction in older 3xTgAD mice.
In sum, the functional deficits found within and relative to the temporal and ventral brain areas in  Table 2. Specifically, male 3xTgAD and control mice on the same background strain (129sv/c57bl6) aged either 3-4 months (N = 6 and N = 7, respectively) or 6-7 months old (N = 4 and N = 4, respectively) were used for electrophysiological recordings in vivo. Additionally, male controls (total N = 29) and 3xTgAD (total N = 31) have been used for the imaging experiments. Specifically, Ncontrols = 10 and N3xTgAD = 19 underwent rsfMRI experiments. Additionally, Ncontrols = 19 and N3xTgAD = 12 mice underwent ofMRI experiments. An a priori power analysis was performed following results in 21 using R (power.t.test), indicating that N = 10 per group for rsfMRI is sufficient to achieve 80% power with the following parameters: delta = 14, SD = 11, two-tailed test, significance threshold = 0.05.

Imaging: Optogenetic surgery
Male 129sv/c57bl6 and 3xTgAD mice (~30 g, N = 19, N = 12 respectively) were anesthetized with a mixture of ketamine/xylazine (ketamine 75 mg/kg, xylazine 10 mg/kg). The head was shaved and cleaned with three wipes of Betadine® and ethanol (70%). Lidocaine was administered subcutaneously, in situ under the scalp. Each animal was kept on a warm pad to prevent hypothermia, and the head was positioned in a stereotaxic frame; protective ophthalmic gel was applied to avoid dryness. A portion of the scalp was removed to expose the skull. The distance between Bregma and Lambda was measured and compared to the standard 4.2 mm reported in the mouse brain atlas 51 . Any deviation from 4.2 mm allowed a proportional adjustment for craniotomy coordinates. Small craniotomies were performed above the left hemisphere with a drill (burr tip 0.9 mm 2 ) at -2.8 from bregma, +4.2 from the midline. Virus injection into ENTl was carried out through this craniotomy at -2.8 to -2.7 mm from brain surface and cannula positioning reached -2.6 mm from the surface.
Coordinates were taken according to the Paxinos mouse brain atlas 51  Functional MRI was acquired 20 min following maintenance infusion onset to allow for the animal state to stabilize. Care was taken to maintain the temperature of the animals at 37 °C.

Imaging: Data acquisition and stimulation protocols
Data were acquired on an 11.75 T (Bruker BioSpin MRI, Ettlingen, Germany) equipped with a BGA-S gradient system, a 72 mm linear volume resonator coil for transmission. A 2×2 phased-array cryogenic surface receiver coil was adopted for the rsfMRI experiment (N = 29) and a 10 mm single-loop surface coil for ofMRI experiments (N = 31). Images were acquired using Paravision 6.0.1 software.
For the rsfMRI data acquisition, an anatomical reference scan was acquired using a spin-echo Hz light pulses were applied for 10 s followed by a 50 s rest period, in a 10-block design fashion. An additional 60 s of rest were recorded after the last block of stimulation ( Supplementary Fig. 10a). The experimental groups (3xTgAD and wild-type mice with ChR2-mCherry) and the negative control group (wild-type mice with mCherry alone) underwent the same imaging protocol, i.e., one restingstate scan, followed by randomized 5 Hz, 10 Hz and 20 Hz evoked fMRI scans. The negative control group was imaged with the same imaging protocol as the experimental groups to exclude potential heating and/or vascular photoreactivity artifacts 54,55 . Additionally, in order to exclude abnormal behavior induced by the photostimulation protocol 37 , all animals underwent the three stimulation sessions (5 Hz, 10 Hz, and 20 Hz) again while awake and freely walking in a behavior-chamber.

Imaging: fMRI analysis
Images were processed using a protocol optimized for the mouse and corrected for spikes (3dDespike, AFNI; 56 , motion (mcflirt, FSL; 57 , and B1 field inhomogeneity (fast). Automatic brain masking was carried out on the EPI using bet, following smoothing with a 0.3 mm 2 kernel (susan) and a 0.01 Hz high-pass filter (fslmaths). Nuisance regression was performed using FIX 8 . Separate classifiers were generated for rsfMRI and ofMRI. The EPIs were registered to the Allen Institute for Brain Science (AIBS) reference template ccfv3 using SyN diffeomorphic image registration (antsIntroduction.sh, Local connectivity was assessed with ReHo (3dReHo) 19,20 . Pair-wise region-of-interest (ROI) analysis was carried out with respect to ROIs defined in the AIBS atlas. Time series extracted with the atlas were cross-correlated to the time series from the ENTl using Pearson's correlation. The ofMRI response was examined using a general linear model (GLM) framework (fsl_glm). The stimulation paradigm and its first derivative were convolved using the default gamma function and used as regressors in the analysis, with motion parameters as covariates. Nomenclature and abbreviations for the brain regions are in accordance with https://atlas.brain-map.org/.
The query returned 30 spatial maps depicting activation voxels in neuroimaging literature associated with the searched term (Fig. 1a).

Anatomical gene expression atlas comparison
The spatial expression profile for 4117 genes was obtained from the anatomical gene expression atlas database using the application programming interface from the AIBS 59 . The spatial correlation between the ReHo second-level statistical map and each of the genes was estimated using Pearson's correlation (fslcc). ReHo-gene correlations were ranked and tested for enrichment of biological processes using Gene Ontology enRIchment anaLysis and visuaLizAtion tool (GOrilla, http://cblgorilla.cs.technion.ac.il/) 60,61 . Enrichment was tested with Fisher's Exact test with FDR correction.

Electrophysiological recordings in vivo: anesthesia and surgical procedure
Anesthesia was induced via intraperitoneal (i.p) injection of urethane (1.5-1. These were first lowered to target locations (BLA target was 4 mm ventral from the brain surface at a 15° angle from vertical in the coronal plane; the DG target was also 4 mm ventral to brain surface but along a vertical plane) and then used to record brain activity during different stimulation protocols.

Electrophysiological recordings in vivo: data acquisition
Data were recorded on a Recorder64 system (Plexon Inc, USA), and saved for offline analysis.
Electrodes were connected to a head stage (fixed gain of x20) and then to a preamplifier for a total gain of x500. Field excitatory postsynaptic potentials (fEPSPs) were recorded at a sampling rate of 5 or 10 kHz using a 12-bit A/D converter and then stored for offline analysis. A low pass filter (1 kHz) was applied to attenuate spiking activity. Electrical stimuli were delivered by a constant-current stimulator (DS3, Digitimer, UK), triggered by analog 5 V square wave pulses from a National Instruments PCI card (PCI-6071E). Timings and types of stimuli to be delivered were controlled through custom-written programs in LabVIEW (v8, National Instruments). Stimulus duration was fixed at 200 μs throughout each protocol. Two different protocols of stimulation were performed: Input/Output, for the Input/Output curve (IOC) analysis and PPS for the PPI analysis. In each mouse, the channel with the most distinctive response (as revealed through current source density analysis post hoc; see below) was selected for further analysis.

Electrophysiological recordings in vivo: stimulation protocols -Input/output curve
The IOC reflects the functional strength of synaptic connectivity: by applying different current intensities, it is possible to analyze how the response (Output voltage) changes as a function of input strength (Input current). The range of current intensities used here was 180, 300, 450 and 600 μA.

Electrophysiological recordings in vivo: stimulation protocols -Paired-pulse stimulation
To measure short-term synaptic plasticity, stimulus current was set to half-maximum of the response obtained in the BLA IOC paradigm, i.e., ~300 μA, and paired-pulses were delivered at this current with different paired-pulse intervals for 20 repetitions each. The range of intervals was 20, 50, 100, 200, 500 and 1000 ms.

Electrophysiological recordings in vivo: data analysis
For the IOC protocol, stimulation at each current intensity was repeated 20 times (runs), the initial slope of the fEPSP response measured for each repetition and then these 20 values were averaged. The mean response to each current step for each mouse was used to plot the IOC of response to current intensity by genotype, age, and ROI. For the PPS protocol, the slope for the initial fEPSP response on a selected channel was measured for both stimuli in each pair. For each pair, the fEPSP response to the second stimulus (P2) was normalized to that of the first (P1) and expressed as PPI ratio (see equation Data are plotted as mean ± 1 standard deviation (SD).

Brain slice preparation and whole-cell current-clamp recordings in ILA and DG
After undergoing the scan at 6 months, 8 mice (Ncontrols = 4, N3xTgAD = 4) from the ofMRI experimental mice were used for a whole-cell patch recording in brain slices. The animals were deeply anesthetized with ketamine/xylazine (0.1 ml/kg) and cardiac perfusion was performed with ice-cold, oxygenated respectively under current-clamp mode (I = 0) to examine whether the neurons were able to follow high-frequency photostimulation. After different frequencies of photostimulation were completed, neurons were shifted to voltage-clamp mode (at -60 mV), and a prolonged square pulse of 500 ms duration was delivered, to further confirm whether ChR2-induced current could be seen in the recorded neurons. The access resistance, membrane resistance, and membrane capacitance were consistently monitored during the experiment to ensure the stability and the health of the cell.

ELISA diagnostic assay
Brains were homogenized in a Tris-HCl buffer and agitated for 30 minutes before being subjected to centrifugation at 6000 g. The supernatant was used for the detection of soluble Aβ40-42 and the pellet was re-suspended in Tris-HCl and 10 μl was used for further processing for insoluble Aβ ELISA as recommended by the ELISA kit (incubation with 5 M Guanidine to solubilize any aggregates and diluted before adding into ELISA plates). Samples were diluted only when necessary (phospho-tau and total tau were diluted 2x, no dilution for Aβ ELISAs). As readings could be affected by the sample volume in each well, we normalized the ELISA results to protein assay results of the same fraction, hence, the final units were in picogram of Aβ or tau in per milligram of protein (pg/mg protein).
After the primary antibody binding step sections were washed 5 times in 1 x PBS for 3 min and then incubated with anti-mouse Alexa 488 or anti-rabbit Alexa 594 for 2 hours followed by washing 3 times with 1 x PBS for 3 minutes. Sections were then mounted with DAPI plus mounting media on slides. All pictures were taken by using a confocal microscope with a 40× objective.

Ex vivo histology
After the completion of the experiments, animals were injected with an overdose of ketamine/xylazine and transcardially perfused with PBS (0.01 M) followed by 4% PFA in 0.01 M PBS. After extraction, the brain was post-fixed in 4% PFA overnight. Brain sections of 50 μm were made with a vibratome (VBT1200s, Leica) and fluorophore expression, together with Hoechst staining, was checked through a confocal microscope Ti-E; DS-Qi2; Fluorescence, SBIC-Nikon Imaging Center, Singapore) for anatomical confirmation of viral injection and fiber optic cannula positioning (Fig. 2).

Statistics and data availability
Descriptive statistics for neuroimaging data are given as mean difference and [95th confidence interval] unless stated otherwise, and graphically represented as 'Gardner-Altman plots' (https://www.estimationstats.com/; 66 ). If not specified, descriptive statistics are provided for left hemisphere ROIs. The statistical threshold for significance was set at p < 0.05, two-tailed. Voxel-wise was carried out with a non-parametric permutation-based (5000 permutations) test (randomize).
Cluster correction was carried out with threshold-free cluster enhancement (tfce). Thresholded tstatistic for one-sample and two-sample t-tests (p < 0.05, tfce corrected) are shown as a color-coded overlay on the AIBS template. ROI analysis was carried out with a linear mixed model using genotype and age as fixed effects and individual intercepts as random effects, using the lme4 package (1.  for R (https://cran.r-project.org/, 3.5.3, "Great Truth"). Significance was assessed with general linear hypothesis tests implemented in the multcomp (1.4-10) package and corrected with the false discovery rate.