Dynamic microglia alterations associate with hippocampal network impairments: A turning point in amyloid pathology progression

Alzheimer ’ s disease is a progressive neurological disorder causing memory loss and cognitive decline. The underlying causes of cognitive deterioration and neurodegeneration remain unclear, leading to a lack of effective strategies to prevent dementia. Recent evidence highlights the role of neuroinflammation, particularly involving microglia, in Alzheimer ’ s disease onset and progression. Characterizing the initial phase of Alzheimer ’ s disease can lead to the discovery of new biomarkers and therapeutic targets, facilitating timely interventions for effective treatments. We used the App NL-G-F knock-in mouse model, which resembles the amyloid pathology and neuroinflammatory characteristics of Alzheimer ’ s disease, to investigate the transition from a pre-plaque to an early plaque stage with a combined functional and molecular approach. Our experiments show a progressive decrease in the power of cognition-relevant hippocampal gamma oscillations during the early stage of amyloid pathology, together with a modification of fast-spiking interneuron intrinsic properties and postsynaptic input. Consistently, transcriptomic analyses revealed that these effects are accompanied by changes in synaptic function-associated pathways. Concurrently, homeostasis-and inflammatory-related microglia signature genes were downregulated. Moreover, we found a decrease in Iba1-positive microglia in the hippocampus that correlates with plaque aggregation and neuronal dysfunction. Collectively, these findings support the hypothesis that microglia play a protective role during the early stages of amyloid pathology by preventing plaque aggregation, supporting neuronal homeostasis, and overall preserving the oscillatory network ’ s functionality. These results suggest that the early alteration of microglia dynamics could be a pivotal event in the progression of Alzheimer ’ s disease, potentially triggering plaque deposition, impairment of fast-spiking interneurons, and the breakdown of the oscillatory circuitry in the hippocampus.


Introduction
Alzheimer's disease (AD) is a chronic multifactorial disorder characterized by progressive memory loss and cognitive impairment, preceded by a decades-long asymptomatic stage (Aisen et al., 2017).The events that drive the onset of the pathology leading to cognitive decline and neurodegeneration in the advanced stages remain unclear.This poor understanding of AD aetiopathogenesis is reflected in the absence of early treatment strategies to prevent the onset of dementia rather than slowing down its progression (Golde, 2022;Mehta et al., 2017).Amyloid beta (Aβ) accumulation is traditionally considered a major AD driver (Braak and Braak, 1991;Hardy et al., 1979), however, growing evidence suggests that Aβ amyloidosis alone is not enough to explain the entire AD pathogenesis (Herrup, 2015).Another significant pathological feature of AD is tauopathy, involving abnormal aggregation and phosphorylation of tau protein (Kolarova et al., 2012), leading to neurofibrillary tangle formation and contributing to neurodegeneration (Spillantini and Goedert, 2013).Furthermore, neuroinflammation has been proposed as an important contributor to AD onset and progression (Leng and Edison, 2021;Passamonti et al., 2019).Indeed, epidemiological studies have reported a positive association between cognitive decline and previous inflammatory events (Holmes, 2021;Bohn et al., 2023).
For this reason, microglia, as the resident immune cell type in the brain, are receiving increasing attention in AD studies.However, whether microglia play a protective or detrimental role during AD progression remains controversial (Leng and Edison, 2021;Hamelin et al., 2018).The beneficial role of microglia in containing plaque deposition and delaying cognitive impairment has been reported by several studies (Feng et al., 2020;Hamelin et al., 2016;Fan et al., 2017;Yang et al., 2023;Femminella et al., 2019).Nonetheless, chronic activation of microglia is a well-known cause of neuroinflammation and neurodegeneration (Fan et al., 2015;Heneka et al., 2005;Mass et al., 2017).
Besides their immune-related functions, microglia are involved in homeostatic processes in the brain, including neuronal and network activity regulation (Karson et al., 2009;Guan et al., 2022;Cserép et al., 1979;Cserép and Pósfai, 2021).One of the consequences of AD is the dysfunction of multiple brain networks, resulting in the inability of the circuits to synchronize in specific oscillation rhythms (De Haan et al., 2012;Smailovic et al., 2018).In particular, brain oscillations in the gamma-frequency band (30-80 Hz) play a pivotal role in higher brain activities like information processing and memory formation (Colgin and Moser, 2010;Jensen et al., 2007;Düzel et al., 2010).We previously reported that disruption of cognition-relevant gamma oscillations and fast-spiking interneurons (FSI) rhythmicity occurs at an early stage of amyloid pathology in the hippocampus of the App NL-G-F mouse model (Arroyo-García et al., 2021).This functional impairment takes place before widespread plaque deposition in the hippocampus.Importantly, restoration of FSI rhythmicity and hippocampal oscillatory activity leads to improved cognitive performance (Arroyo-García et al., 2023;Emre et al., 2022), underlining the importance of hippocampal network dysfunction in AD pathology progression.
Here, we used a combination of functional and molecular approaches to investigate the mechanisms behind the failure of the hippocampal network and the role of microglia at the earliest stages of amyloid pathology.To our knowledge, this is the first time that such an early preplaque stage of amyloidosis progression has been investigated by RNA sequencing in this model.Elucidating the events characterizing the initial phase of the amyloid pathology can generate new perspectives on AD's etiopathogenesis, hence providing new biomarkers and targets for the development of a therapeutic approach to AD based on timely intervention.

Ex vivo gamma oscillations
For Local field potential (LFP) measurements, brains from all age points were used: P30 11).Ex vivo gamma oscillations were induced by adding kainic acid (KA; Tocris Bioscience, Bristol, UK) at a concentration of 100 nM to the extracellular bath and were recorded using borosilicate glass microelectrodes (1,5-2,5 MΩ) filled with ACSF placed in the stratum oriens of the CA3 area of the hippocampus.The oscillations were allowed to stabilize for at least 20 min before recordings were performed.Signals were conditioned using a HumBug 50 Hz noise eliminator (Quest Scientific).For oscillation power spectra Fast Fourier Transformations were obtained from 60 s of LFP recording (segment length 8192 points) using Axograph software (Kagi, Berkeley, CA, USA).Frequency variance data was obtained from the power spectra described above using Axograph X. Gamma power was calculated by integrating the power spectral density from 20 to 80 Hz using Clampfit 11.2.

Intrinsic properties of fast-spiking interneurons
Intrinsic properties and spontaneous excitatory post-synaptic currents (sEPSC) were recorded in whole-cell patch clamp mode from fastspiking interneurons (FSI) in the CA3 stratum radiatum of hippocampal slices from P30 (WT n = 17, App NL-G-F n = 10) and P60 (WT n = 11, App NL-G-F n = 17) brains.Patch clamp recordings were performed with borosilicate glass micropipettes (4-6 MΩ) filled with a potassium-based internal recording solution (in mM: 122.5 K + -gluconate, 8 KCl, 2 Mg2 + ATP, 0,3Na + GTP, 10 HEPES, 0,2 EGTA, 2 MgCl; pH 7.2-7.3;osmolarity 270-280 mosmol/l) containing 2 % of neurobiotin (Vector Laboratories) as a neurotracer.FSI were visualized under an upright microscope using IR-DIC microscopy (Axioskop, Carl Zeis AG, Göttingen, Germany) and identified based on their location, morphology, and response to custom-made current delivery protocols as described previously (Arroyo-García et al., 2021;Arroyo-García et al., 2023).EPSCs were recorded in voltage-clamp configuration with the voltage holding at − 70 mV.Firing and membrane potential properties were recorded in gap-free current-clamp mode.Each cell classified as FSI underwent consequent hyper-and de-polarizing current steps (Input resistance protocol (Balleza-Tapia et al., 2022): 25 steps of 10pA from − 100 pA to +140 pA, 700 ms each; Fig. 2.A) starting from the membrane resting potential and from a − 70 mV membrane potential.After recording, the slice was stored in PFA overnight and then in a 30 % sucrose solution until further experiments.All recordings obtained from FSI were analyzed using a costume-made Python script.sEPSC were analyzed using the Pyabf and Scipy packages.Firing and membrane intrinsic properties were obtained from the abovementioned input resistance protocol using the Pyabf, Ipfx, and Efel packages.
Rheobase (pA), firing threshold (mV), and firing latency (ms) were extracted from the first sweep of the input resistance protocol with at least one action potential.
Mean action potential amplitude (mV), half-width (ms), peak upstroke (V/s), peak downstroke (V/s), rise rate, fall rate, rise time (ms), and fall time (ms) were obtained from the first sweep of the input resistance protocol with four to six action potentials.
Firing rate (Hz) was calculated as the number of action potentials fired during the input resistance protocol, divided by the duration of the protocol.
Ohmic input resistance (GΩ) was calculated by averaging the ohmic input resistance values from subthreshold hyper-and depolarizing current steps in the input resistance protocol (from − 30 to +30 mV) in the absence of action potentials.
Membrane potential at the corresponding current step was extracted from each trace of input resistance protocol.Resting membrane potential (Em0) corresponds to the membrane potential when no current is injected.
Sag (mV) was calculated from the most hyperpolarized sweep (− 100 pA) of the input resistance protocol recorded with the voltage holding at − 70 mV.

RT-qPCR
WT (P30 n = 4) and App NL-G-F (P30 n = 4, P60 n = 4) hippocampal slices were placed in a submerged recording chamber where a solution of ACSF and KA (100 nM) was constantly supplied for 2 h.After 2 h, gamma oscillations were recorded to confirm the network activation.For the control group, hippocampal slices from each group were incubated for 2 h in ACSF without KA.
RNA was purified using the PureLink RNA mini kit with in-column DNAse-I digest (#79254, Qiagen) following the manufacturer's instructions (#12183018A, ThermoFisher).RNA was then quantified in a nanodrop and reverse-transcribed using the SuperScript III First-Strand Synthesis System (#18080051, ThermoFisher), followed by RNAse H digestion. cDNA was then used to run qPCRs using TaqMan Universal PCR master mix.

RNA extraction and bulk RNA sequencing
Hippocampi from sixteen mice (WT P30 n = 4, WT P60 n = 4, App NL- G-F P30 n = 4, App NL-G-F P60 n = 4) were dissected and used for bulk RNA-seq.Total RNA was isolated using the RNeasy Plus mini kit (QIA-GEN #74134) according to the manufacturer's instructions and processed for sequencing at the Bioinformatics and Expression Analysis core facility at Karolinska Institutet.Following the quality control by Agilent Bioanalyzer 2100, samples were processed for library preparation and sequenced using the Illumina Nextseq 2000 P3 (100 cycles) system.All the following analyses were performed in R (version 4.3.0).The DESeq2 (version 1.40,2) package was used to identify differentially expressed genes across different groups.Log fold changes obtained with a Wald test were shrunk using the apeglm method.Differentially expressed genes (DEGs) were identified by log2 fold change (Log2FC) > 0,2 and adjusted p-value < 0,05, as calculated with the Benjamini & Hochberg multiple adjustment method.For the functional classification of DEGs, we used the clusterprofiler (Wu et al., 2021;Among et al., 2012) tool (version 4.8.2).Specifically, Gene Ontology (GO) enrichment analysis was performed to identify the activated and suppressed GO terms (biological processes, molecular function, and cellular component).The number of permutations for the enrichment analysis was set to 10000.GO terms that displayed a p-value < 0,05, as derived by the hypergeometric test were sorted out as statistically significant.Principal component analysis (PCA) on the sequencing data was performed using the VSN package following instructions from the Bioconductor repository.Centroid coordinates for the PCA plot were calculated as the average of the x and y coordinates of individual samples belonging to the same group.

Immunostaining and imaging
) hippocampal slices were used for immunostaining.Non-specific binding was blocked by incubating the sections in a solution of 10 % normal donkey serum (Jackson ImmunoResearch Laboratories, West Grove, PA), 1.5 % Triton X-100 (made Tris buffer saline, TBS) for 2 h at room temperature.Sections were then incubated with primary antibodies rabbit anti-Iba1 (1:1000; WAKO) and mouse anti-human amyloid beta 82E1 (1:1000; IBL) at 4 • C for 48 h and then for 2 h at room temperature with the fluorescent secondary antibodies Alexa-555 donkey anti-rabbit (1:1000; Abcam), Alexa-633 donkey anti-mouse (1:250; Biotium) and Streptavidin 488 (1:1000; Vector laboratories).Hoechst 33342 (Molecular Probes/Life Technologies) was used as a nuclear counterstain.Sections were mounted onto glass slides and coverslipped using ProLong diamond anti-fade reagent (Molecular probes/Life technologies).The sections were imaged using a confocal laser scanning microscope (Zeiss LSM900-Airyscan confocal) at the Biomedicum Imaging Core facility at Karolinska Institutet.

Microglia and amyloid plaque density
Images of the entire hippocampus were acquired, using a 20× objective, by a combination of tiles and a Z-stack of a thickness of 10 µm.These images were processed using the Zen system and their maximum intensity projection was obtained, which allowed for the quantification of the microglia using the Zeiss Zen 3.7 software.Furthermore, the area of the hippocampus was obtained using the lasso tool in the Zeiss software which was also used to acquire the area of the amyloid beta plaques.Microglia and amyloid plaque density as well as plaque area were calculated as a function of the hippocampal area.

Microglia and FSI interaction
In thirty-three of these sections (WT P30 n = 6, WT P60 n = 6, App NL- G-F P30 n = 11, App NL-G-F P60 n = 10, containing one neurobiotin prefilled neuron each) images of the neurobiotin-marked neurons were taken at a magnification of 0,7, using a 63x oil objective along with a Zstack technique of a thickness of 10 µm.Subsequently, these images underwent processing utilizing the Zen system: their maximum intensity projections, as well as a subset of images measuring 1355 x 1355 pixels, were generated.Cell surface creation was carried out using the IMARIS 10.0.1 software, by default parameters (surface detail 0.141 µm) in green (AF488) and red (AF555) channels.Surfaces were masked setting a constant in the voxel intensity.Outside surfaces intensity was set to value = 0 and inside surfaces to value = 1, generating two new binary channels.
Using the IMARIS image processing tool "Channel Arithmetic" by Matlab XTension, the two newly generated binary channels were combined following the formula (chX * chY) resulting in a new channel (pink) that shows only the voxel overlaps from the two binary channels.Voxels with value 1 after the channel arithmetic correspond to the places where voxels of both channels had an overlap in positive signal, indicating the contact surface between cells.Image J software was used to obtain the area of the contacts between microglia and neurons and categorize the contact points into somatic and dendritic.

Statistical analysis
All statistical analyses, excluding the analysis of the bulk RNA sequencing data, were performed using GraphPad Prism 9.4.0.We performed two-sided statistical analyses comparing the App NL-G-F group and the WT at each age.We used a normality and lognormality test (Shapiro-Wilk test) and we excluded outliers after ROUT (Q = 1 %) analysis.We performed one and two-way ANOVA followed by a Holm-Sidak's multiple comparisons test, unpaired t-test, and Pearson

Progressive impairment of hippocampal gamma oscillations in App NL- G-F mice
Previous evidence from our lab shows an early impairment in the hippocampal oscillatory activity during amyloid pathology progression in the App NL-G-F knock-in mouse model (Arroyo-García et al., 2021).We reported that gamma oscillations were strongly decreased in the App NL- G-F mice at P75 when compared to age-matched WT mice, while no differences were observed at P30 (Arroyo-García et al., 2021).In the present study, we first aimed to narrow down the time-point when this shift in hippocampal gamma oscillations occurs.For this purpose, we investigated ex vivo gamma oscillations in hippocampal slices of App NL- G-F and WT mice at P30, P40, P50, and P60 (Fig. 1.A).
App NL-G-F mice exhibited a progressive decrease in gamma power that is significant at P50 compared to P30 mice (App NL-G-F P30 vs P50 p = 0,049; Fig. 1.B, C).Additionally, we confirmed that gamma oscillations were comparable in power comparing the App NL-G-F and WT groups at P30 (WT vs App NL-G-F P30 p = 0,124; Fig. 1.C), while they were significantly decreased at P50 in the App NL-G-F group as compared to agematched WT mice (WT vs App NL-G-F P50 p = 0,017; Fig. 1.C).This gamma oscillation impairment in App NL-G-F mice continued to decrease at P60 (WT vs App NL-G-F P60 p = 0,010, App NL-G-F P30 vs P60 p = 0,026; These results provide a specific time window to investigate the functional and molecular changes in this early stage of amyloid pathology progression and potentially identify new targets to prevent and/ or recover them.Therefore, we selected a time point in the App NL-G-F mice where the oscillatory activity is comparable to the WT mice (P30) and a time point with oscillatory impairment (P60) for further analysis.
After selecting the time points of interest, we next asked if the induction of hippocampal gamma oscillations by KA could promote the expression of immediate early genes (IEG).IEG are rapidly induced in response to neuronal depolarization and play an essential role in neuronal synaptic plasticity and learning-associated processes (Minatohara et al., 2016;Yap and Greenberg, 2018;Kim et al., 2018;Gallo et al., 2018).We therefore quantified Fos and Npas4 gene expression analysis by RT-qPCR in KA-exposed hippocampal slices before and after gamma oscillation impairment in App NL-G-F mice and compared them to a WT group (Fig. 1.A).
Our results show that the network activation with KA caused an induction of Fos (Fig. 1.D, p = 0,003) and Npas4 (Fig. 1.E, p < 0,0001) expression in WT mice.This effect was not observed in App NL-G-F mice at P30 nor at P60 (Fig. 1.D, E), in line with previous studies that report suppression of IEG in the presence of Aβ (Dickey et al., 2004).

Fast-spiking interneuron intrinsic properties and postsynaptic input change during amyloid pathology progression
Fast-spiking interneurons (FSI) are a subtype of GABAergic interneurons that hold a pivotal role in initiating and sustaining gamma oscillations (Cardin et al., 2009;Li et al., 2021;Hu et al., 1979).We have previously reported that FSI spike-gamma coupling is affected during amyloid pathology progression at P75 (Arroyo-García et al., 2021).Here we aimed to assess the functionality of FSI in the App NL-G-F mice.For this reason, we used the whole-cell patch-clamp technique to assess the intrinsic properties and post-synaptic currents of FSI in a resting state (without KA) in WT and App NL-G-F mice at P30 and P60 (Fig. 2.A; Table 1).
To evaluate differences and similarities within FSI across all groups, we performed a three-dimensional principal component analysis (PCA) on the intrinsic properties of FSI (Table 1).This analysis revealed that FSI in the App NL-G-F group at P60 had distinctive intrinsic properties since they clustered in a separate area of the plot compared to the App NL- G-F group at P30 and the WT groups at P30 and P60 (Fig. 2.C).This result indicates that FSI in the App NL-G-F mice at P30 were functionally comparable to those from the WT groups, while at P60 they diverged from the WT groups and the App NL-G-F P30 group, implying an alteration of FSI intrinsic properties during amyloidosis progression (App NL-G-F P30 vs P60; Fig. 2.B).Indeed, FSI in the App NL-G-F group at P60 were depolarized (p = 0,044; Fig. 2.D), and they exhibited a higher firing threshold (p = 0,011; Fig. 2.E) and faster repolarization phase of the action potential (p = 0,049; Fig. 2.F) in comparison with FSI in the App NL-G-F group at P30.Additionally, FSI from P60 App NL-G-F mice received a decreased postsynaptic input, as shown by charge transfer (p = 0,027, Fig. 2.G) and spontaneous excitatory postsynaptic current (sEPSC) frequency (p = 0,022, Fig. 2.H).These findings suggest a change in the FSI membrane properties during the amyloid pathology progression.Importantly, the properties that differed between App NL-G-F P60 and P30 were preserved in the App NL-G-F P30 group as compared to age-matched WT mice (Supp Fig. 2), confirming that FSI functionality was not affected at this time point.
Collectively, such changes in FSI functionality might contribute to the impairment of the hippocampal network (Fig. 1.B, C) and the FSI spike-gamma uncoupling (Arroyo-García et al., 2021) that we observed in the early stage of the amyloid pathology progression.Intrinsic properties of FSI in WT P30 (n = 17), WT P60 (n = 11), App NL-G-F P30 (n = 11) and App NL-G-F P60 (n = 17).Data are presented as mean ± SEM, and "n" indicates the number of cells.

Bulk RNA sequencing reveals synaptic-related transcriptional changes during the early stage of amyloid pathology progression
Next, we performed bulk RNA sequencing to investigate the molecular events before and after the impairment in the hippocampal oscillatory activity and neuronal properties in the App NL-G-F mouse model (Fig. 3.A).
First, we analyzed the dataset from WT mice at P30 vs P60, showing 716 differentially expressed genes (DEG) plausibly due to brain maturation (Supp Fig. 3.A) (Bundy et al., 2017).Then we compared agematched WT vs App NL-G-F groups at P30 and P60 age to evaluate the amyloidogenic effect in the App NL-G-F knock-in model in comparison to WT mice.We found 401 DEG at P30 WT vs App NL-G-F (Supp Fig. 4.A) and 553 DEG at P60 WT vs App NL-G-F (Supp Fig. 4.B).Finally, to assess the effects of amyloid pathology progression we compared the App NL-G-F model at P30 vs P60 and we found 2975 DEG (Supp Fig. 3.B).
To differentiate the transcriptional changes triggered by amyloid pathology (App NL-G-F P30 vs P60) from the ones due to normal brain maturation (WT P30 vs P60), we compared the DEGs from these two groups (Fig. 3.B).We found that 502 DEGs (66 upregulated and 434 downregulated; supp Fig. 5.A) were shared between the two comparisons, suggesting that it is genes that are indeed associated with brain maturation in both WT and App NL-G-F mice.Interestingly, 2473 DEG (1401 upregulated and 1074 downregulated, supp Fig. 5.A) changed exclusively during amyloid pathology progression (App NL-G-F P30 vs P60).
After, we aimed to investigate whether the transcriptional changes during amyloidosis pathology progression detected by RNA sequencing translated to protein level changes (App NL-G-F P30 vs P60).For this purpose, we quantified specific proteins corresponding to the altered genes in our transcriptomic assessment (ErbB4, KCNS3, Nav 1.1, Kv 3.1, Syt 2, GABA A b3, and GluR4) and some additional receptor subunits involved in the FSI and gamma oscillations activity: Glutamate receptor A1, GABA receptor A beta3 (Emre et al., 2022) and glutamate receptor 4 (Verret et al., 2012).Our results showed that GABA receptor 1 alpha (Fig. 3.E, F) and glutamate receptor A1 (Fig. 3.E, G) were significantly downregulated already at P30 in the App NL-G-F mice when compared to WT mice (p = 0,032 and p = 0,009 respectively), and the downregulation is also detected at P60 (p = 0,007 and p = 0,005 respectively).No other changes in protein levels were detected at these time points (Supp Fig 6).

Microglia-related transcriptional changes indicate loss of homeostatic and inflammatory microglia during the early stage of amyloid pathology progression
Given the crucial role of microglia in neuronal network regulation (Guan et al., 2022) and AD (Leng and Edison, 2021;Sarlus et al., 2017), our next step was to investigate transcriptional changes of microgliarelated genes and pathways.
Pathway enrichment analysis (Supp.Table 1) showed that microgliaand inflammation-related processes, including microglia differentiation, acute inflammatory response, phagocytosis, synaptic pruning, and complement systems were suppressed in the App NL-G-F group at P60 compared to P30 (App NL-G-F P30 vs P60; Supp Fig. 7.A).Interestingly, similar pathways are activated in App NL-G-F mice at P30 compared to agematched WT mice (WT vs App NL-G-F P30; Supp Fig. 7.B).Fewer microglia and inflammation-related changes occur with normal brain maturation in the WT group (WT P30 vs P60; Supp Fig. 7.C) and in the App NL-G-F group at P60 compared to age-matched WT mice (WT vs App NL-G-F P60; Supp Fig. 7.D).These transcriptional changes suggest a general activation of microglia and inflammation-related pathways in the App NL-G-F group at P30, followed by their suppression at P60.Furthermore, we evaluated the transcriptional differences of microglia populations across the four groups using a two-dimensional PCA.We selected 2054 microglia genes, representative of homeostatic and inflammatory microglia (Frigerio et al., 2019;Hammond et al., 2019;Sun et al., 2023) from the RNA sequencing dataset (Supp.Table 3).The PCA underlines the similarity of the microglia expression profiles in the WT group at P30 and P60.Conversely, microglia gene expression differs between P30 and P60 in the App NL-G-F mice, with samples from P30 and P60 time points clustering separately at the opposite edges of the PCA plot (Fig. 4.A).A similar sample distribution can be observed when genes representative of single microglia populations are plotted separately (Homeostatic and pruning microglia, 751 genes, Supp Fig. 8.A; activated microglia, 885 genes, Supp Fig. 8.B; interferon responding microglia, 418 genes, Supp Fig. 8.C, Supp Table 3).This result suggests that the microglia transcriptional profile changes to a higher extent during amyloid pathology progression than it would during normal maturation.
To further characterize this event, we analyzed the expression of distinctive genes for different subpopulations of microglia previously reported by single-nucleus RNA sequencing characterization of microglia in the App NL-G-F mouse model (Frigerio et al., 2019): Tmem119, P2ry12, and Cx3cr1 for homeostatic microglia; C1qa, C1qb, C1qc, Ctsb, Ctsd, Fth1 and Lyz2 homeostatic microglia engaged in synaptic pruning; Irf7, Ifitm3, Ifit3, Oasl2, Ifitm1 and Ifit2 for interferon responding microglia; Clec7a, Cst7 and Itgax for activated response microglia.A significant downregulation of genes related to microglia homeostatic and inflammatory states was detected in the App NL-G-F group at P60 compared to P30 (App NL-G-F P30 vs P60; Fig. 4.B; Table 3).Additionally, widely accepted pan-microglia markers such as Aif1, Fcrls, and Hexb (Jurga et al., 2020) followed the same expression pattern, suggesting a suppression of microglia-related genes in App NL-G-F mice at P60 that could indicate a decrease of microglia at this time point.To test this hypothesis, we measured the protein levels of Iba1 (Fig. 4.C, F), TMEM119 (Fig. 4.D, G) and P2RY12 (Fig. 4.E, G) by western blot.Iba1 is the protein encoded by Aif1 and a commonly used pan-microglia marker, and TMEM119 and P2RY12 are expressed by homeostatic microglia.These three markers together can give a comprehensive overview of microglia homeostatic state (Kenkhuis et al., 2022;Butovsky et al., 2014).Interestingly, the level of Iba1 and TMEM119 increased in P30 App NL-G-F mice compared to age-matched WT mice (WT vs App NL- G-F P30, p = 0,009 and p = 0,003 respectively; Fig. 4.C, D) and decreased between P30 and P60 in the App NL-G-F group (App NL-G-F P30 vs P60, p = 0,011 and p = 0,006 respectively; Fig. 4.C, D).The level of P2RY12 tended to decrease between P30 and P60 in the App NL-G-F group (App NL-G- F P30 vs P60, p = 0,057: Fig. 4.E).

Iba1-positive microglia density decreases during the early stage of amyloid pathology and correlates with plaque aggregation and FSI dysfunction
To verify that the microglia-related transcriptional and protein level variations reflected a change in the density of microglia in the hippocampus of App NL-G-F mice, we performed immunofluorescence staining (Fig. 5.B-E) of hippocampal slices containing neurobiotin-marked FSI from our intracellular recordings (Fig. 5.A, 2.A).
Here we showed that microglia density, expressed as the number of Iba1-positive cells by area, increased at P30 in App NL-G-F mice compared to age-matched WT mice (WT vs App NL-G-F P30, p = 0.036; Fig. 5.J).Importantly, we found a decrease of microglia density in the hippocampus of App NL-G-F mice at P60 when compared to P30 (App NL-G-F P30 vs P60, p < 0,0001; Fig. 5.J).These findings validate the dynamic pattern of microglia abundance and gene expression observed by the transcriptomic and protein analysis.Moreover, we found an increase in the microglia density at P90 in the App NL-G-F mice compared to P60 (Supp.Fig. 9).Interestingly when compared the WT P90 and the App NL- G-F P90 the microglia density was similar.This result suggests that the decreased microglia density detected at P60 in App NL-G-F mice is transient (Supp.Fig. 9), underlining the peculiarity of the events taking place at this specific stage of the amyloid pathology progression.
Besides microglia, we stained Aβ to evaluate the stage of plaque development during amyloid pathology progression in the hippocampus.As expected, the hippocampi of WT mice were free of amyloid plaques (Fig. 5.B, C).In contrast, we found small, non-widespread amyloid plaques in the hippocampi of App NL-G-F mice at P60, but not at P30 (Fig. 5.D-E, L).Interestingly, the area occupied by amyloid plaques inversely correlated with the density of Iba1-positive microglia in the hippocampus of P60 App NL-G-F mice (r = -0,671; p = 0,048, Fig. 5.N), suggesting that the reduced number of microglia could play a role in plaque aggregation at this stage of the pathology.Additionally, to investigate whether the decrease of Iba1-positive microglia in P60 App NL-G-F mice affected the interaction between microglia and FSI, we used Imaris to obtain a 3D reconstruction of neurobiotin-marked FSI, labeled during patch-clamp recordings, and microglia (Fig. 5.F-I).We found that the somatic contact area between microglia and FSI is decreased at P60 when compared to P30 in App NL-G-F mice (App NL-G-F P30 vs P60, p = 0,047; Fig. 5.K), while the number of contacts is similar between App NL-G-F P30 and WT P30 and P60 (Fig. 5.K).It should be underlined that we found more FSI without any contact with microglia in App NL-G-F P60 than in the other groups.In percentage, 55 % of FSI has no somatic contacts with microglia in the App NL-G-F group at P60.While the percentage of FSI without somatic microglia contacts in the other groups is 15 % in WTP30, 0 % in WT P60, and 8 % in App NL-G-F P30.This implies that the general functionality of microglia-FSI contact is compromised in App NL-G-F mice at P60 since more the half of the FSI that we patched did not have any somatic contacts with microglia.
Since somatic microglia-neuron interaction has been reported to regulate neuronal activity and excitability (Cserép et al., 1979;Szalay et al., 2016), we explored a potential connection between the area of microglia-FSI somatic interphase and FSI excitability in all the groups.To this end, we tested the correlation between microglia-FSI contact area and the FSI intrinsic properties that were affected in P60 App NL-G-F mice (Fig. 2.D-F), adding the firing rate as an additional parameter to evaluate excitability.Interestingly, the microglia-FSI somatic contact area positively correlated with FSI firing rate and negatively with their firing threshold, indicating that FSI with a smaller microglia-FSI somatic interphase area had a higher firing threshold (r = − 0,548; p = 0,015) and fired at a lower rate (r = 0,519; p = 0,023, Fig. 5.M).
Taken together, our results show a higher density of Iba1-positive microglia in the hippocampus of App NL-G-F mice compared to WT before plaque deposition and functional impairment, which decreases one month later, in correspondence with the beginning of amyloid plaque aggregation and network dysfunction.In addition to this decrease of microglia in the App NL-G-F group at P60, we detected a reduction in the somatic contact area between microglia and FSI, which associates with changes in the firing activity of FSI, as evaluated by the Pearson correlation test.

Discussion
In this study, we show that amyloid pathology in App NL-G-F mice progressively and detrimentally affects the hippocampal circuit responsible for gamma oscillations to a point of network failure, before widespread Aβ plaque load.Concurrently, the intrinsic properties of FSI are compromised during this early stage of the pathology.Interestingly, we found that this early functional impairment is associated with a reduced number of Iba1-positive microglia in the hippocampus and a decrease in their somatic contacts with FSI.We propose that this alteration of microglia dynamics could be connected to the onset of Aβ plaque aggregation and network dysfunction in the hippocampus.
Increasing evidence has shown a crucial role of microglia in AD pathogenesis and progression (Leng and Edison, 2021;Sarlus et al., 2017).However, the involvement of this cell type in AD pathology development and its role in Aβ plaque formation and clearance remains controversial (Leng and Edison, 2021;Hamelin et al., 2018).One of the main difficulties to overcome when studying microglia is their extreme dynamism, the reason why different studies can lead to apparently contradictory results when trying to shed light on their role in AD pathophysiology.Studies have shown the protective function of microglia activation at pre-symptomatic stages of amyloidosis both in animal models (Feng et al., 2020;Yang et al., 2023) and clinical studies (Hamelin et al., 2016;Fan et al., 2015).Indeed, in patients in a preclinical stage of AD, microglia activation correlated with a slower  indicates that the gene did not pass the automatic independent filtering for the indicated comparison, due to a low mean normalised counts.
On the other hand, the contribution of microglia to plaque formation and compaction has been hypothesized (Huang et al., 2021;Casali et al., 2020;Spangenberg et al., 2019).In particular, Spangenberg et al. reported that microglia depletion impairs Aβ plaque formation in 5xFAD mice (Spangenberg et al., 2019).The mechanism behind this process remains unclear.One hypothesis is that excessive accumulation of Aβ in microglia lysosomes leads to cell death, which might promote plaque aggregation by the release of Aβ agglomerates at the site of microglial death (Baik et al., 2016).Additionally, strong evidence suggests that microglia reactivity promotes neurodegeneration by sustaining chronic inflammation at advanced AD stages (Fan et al., 2015;Mass et al., 2017).
In agreement with all these findings, a biphasic model has been proposed to explain microglia changes during AD progression.First, a protective peak of microglia activation arises in an early presymptomatic stage of the pathology, before Aβ plaque formation.Then a second detrimental activation phase promoting neuroinflammation takes place (Fan et al., 2017;Sarlus et al., 2017;Onuska, 2020).
In this study, we found an increased density of Iba1-positive microglia in the hippocampus of App NL-G-F mice compared to WT (Fig. 5.J) at a timepoint (P30) that precedes Aβ plaque aggregation (Fig. 5.D, L) and functional impairment (Fig. 1.B-C).An increased number of microglia cells is considered a sign of microglia activation during inflammatory response (Mander et al., 2006;Sun et al., 2008), in line with studies reporting an inflammatory response of microglia before plaque formation in animal models of amyloid pathology (Heneka et al., 2005;Ferretti et al., 2012;Wright et al., 2013;Spangenberg and Green, 2017).This suggests microglia activation in response to Aβ clearing, which keeps the brain free of Aβ plaques and maintains the hippocampal synaptic function.
However, as the pathology progresses in the App NL-G-F mice (P60), we observe a reduction of Iba1-positive microglia in comparison to the initial increase (App NL-G-F P30 vs P60; Fig. 5.J).This takes place concurrently with the onset of functional impairment (Fig. 1.B) and Aβ plaque aggregation (Fig. 5.E, L) in the hippocampus.The same pattern of changes is reported in the protein level of Iba1 and TMEM119 (Fig. 4.C, D).Moreover, homeostatic and inflammatory microglia genes (Frigerio et al., 2019) are downregulated at P60 compared to P30 (Fig. 4.B), suggesting that the reduction in number affects in general the diverse microglia populations.Furthermore, in our samples the density of Iba1positive microglia in the hippocampus negatively correlated with the hippocampal area occupied by Aβ plaques (Fig. 5.N), suggesting that microglia dynamic alterations might facilitate Aβ plaque aggregation at this time point.
While late microglia states in the App NL-G-F mouse model have been observed and characterized (Frigerio et al., 2019), this is the first time that an increased density of microglia is detected in a pre-plaque stage in this model.We hypothesise that microglia initially engulf Aβ (Lee and Landreth, 2010), keeping the hippocampus clean from Aβ.With the progression of the amyloid pathology, more Aβ is produced until the maximum phagocytic capacity of microglia is reached, and these cells start dying due to the excessive uptake of Aβ (Baik et al., 2016).Importantly, the acid pH of lysosomes has been shown to favour Aβ aggregation (Hu et al., 2009).Compact Aβ accumulated in the lysosome of microglia is then released at the site of microglia death, where it promotes plaque aggregation by working as a seed for further Aβ aggregation (Baik et al., 2016).We speculate that reaming microglia overloads with the constantly increasing production of Aβ, hence a switch toward the pathological phenotypes characterising the later stage of amyloid pathology progression is promoted (Gao et al., 2023).Although the causes and mechanisms behind the switch from protective to detrimental microglia remain to be elucidated (Leng and Edison, 2021), our findings of microglia dynamic alterations between the two activation peaks could represent a turning point in this dynamic process.
Furthermore, we show that this decrease in Iba-1 positive microglia affects not only the total microglia number in the hippocampus but, importantly, its somatic contacts with FSI, which are also decreased (Fig. 5.K).Microglia dynamically interact with neurons, constantly monitoring and regulating their activity (Badimon et al., 2020).Somatic junctions between microglia and neurons play a key role in this homeostatic communication (Cserép et al., 1979;Pósfai et al., 2019).Interestingly, in addition to decreased somatic contacts between microglia and FSI, we observed a downregulation of P2y12r (Fig. 4.B, E), a purinergic receptor that is fundamental for microglia-neuron crosstalk (Cserép et al., 1979), suggesting a loss of homeostatic surveillance of microglia on FSI.When somatic microglia-neuron interaction is compromised, changes in neuronal activity and excitability have been reported (Cserép et al., 1979;Szalay et al., 2016).Consistently, our data show a change in FSI's membrane and firing properties and a decrease in their postsynaptic input (Fig. 2).
We found that microglia-FSI somatic contact area correlated positively with FSI firing rate and negatively with FSI firing threshold, which implies that FSI with smaller somatic contacts with microglia had a higher firing threshold and a lower firing rate (Fig. 5.M).Our findings support the role of microglia surveillance in regulating FSI homeostasis, suggesting that a change in microglia-FSI somatic contacts can affect FSI excitability.Given the high consumption of ATP that FSI require to maintain their firing properties (Li et al., 2021;Hu et al., 1979;Verret et al., 2012), it can be hypothesized that mitochondrial alterations might play a role in the FSI dysfunctions during the early phase of amyloid pathology.This is in line with the finding that microglia surveillance on neurons relies on the mitochondrial release of ATP (Cserép et al., 1979).Interestingly, studies have shown mitochondrial alterations in the App NL-G-F model already at 2 months of age (Naia et al., 2023).

2016
). Gamma oscillations are associated with higher cognitive functions (Jensen et al., 2007), specifically hippocampal gamma oscillations are involved in memory formation and learning processes (Colgin and Moser, 2010;Düzel et al., 2010).Gamma oscillations are implicated in information processing, facilitating the integration of sensory inputs and the coordination of neural networks involved in attentional mechanisms and sensory perception (Fries, 2015).Furthermore, gamma oscillations contribute to learning processes, with studies demonstrating their involvement in associative learning, conditioning, and the encoding of new information into memory (Lisman and Jensen, 2013).Indeed, gamma oscillations play a crucial role in memory encoding and consolidation, promoting the transfer of information from short-term to long-term storage through synchronized neural activity patterns (Colgin, 2016).Additionally, gamma oscillations are implicated in memory retrieval and recall processes, facilitating the reconstruction of stored information from memory traces (Buzśaki and Wang, 2012).
Alterations in gamma oscillation onset are related to cognitive deficits in multiple neurodegenerative diseases, including AD (Mably and Colgin, 2018).High concentrations of Aβ were reported to degrade hippocampal gamma oscillations in an acute amyloidogenic model (Kurudenkandy et al., 2014).Furthermore, we have proven that amyloid pathology leads to FSI desynchronization and gamma oscillation disruption, before widespread Aβ plaque deposition in the hippocampus of App NL-G-F mice (Arroyo-García et al., 2021).We have now shown that resting membrane potential, firing threshold, action potential repolarization phase as well as postsynaptic current frequency and amplitude (Fig. 2) of FSI change during amyloid pathology progression.These neuronal properties are crucial to regulate the intrinsic excitability of FSI (Klemz et al., 2022), hence their modification can explain the FSI loss of rhythmicity and the consequent network failure that we observed in the hippocampus of App NL-G-F mice at P75 in our previous study (Arroyo-García et al., 2021).This suggests that preserving or restoring microglia surveillance on FSI would rescue the FSI firing properties affected during amyloid pathology progression, in line with previous findings reported in other pathology models (Cserép et al., 1979;Szalay et al., 2016).Interestingly, restoration of FSI firing properties and rhythmicity leads to an improvement of gamma oscillation synchrony and cognitive performance (Arroyo-García et al., 2023;Emre et al., 2022;Verret et al., 2012).Additionally, studies aiming at switching microglia phenotype toward a protective phenotype have shown a decrease in Aβ burden and an improvement of cognitive deficits in AD (Lu et al., 2019;Ofengeim et al., 2017).Although further investigation is needed, these findings suggest that controlling microglia dynamic phenotypes during AD pathology progression is a worth exploring strategy to reduce Aβ aggregation and preserve the hippocampal oscillatory activity.
Our transcriptomic analysis revealed that multiple pathways and genes related to synaptic transmission and neuronal excitability regulation are activated in correspondence with the hippocampal network dysfunction in the App NL-G-F model (Fig. 3.C, D).Additionally, GABA receptor 1 alpha (Fig. 3. F) and glutamate receptor A1 protein levels were decreased in the App NL-G-F group at P30 and 60 (Fig. 3.G).These receptor subunits belong to the main inhibitory and excitatory receptors expressed in the hippocampus, namely GABA A (Olsen, 2009) and AMPA (Traynelis et al., 2010).Studies indicate that increased levels of Aβ lead to the internalization of GABA A receptors (Krantic et al., 2012;Ulrich, 2015).Consequently, we propose that amyloid pathology results in reduced protein levels of GABA A receptors at P30 and P60 in App NL-G- F mice.In line with this explanation, we hypothesise that this situation could cause the transcriptional overexpression of Gabra1 in the App NL-G-F group at P60 (Fig. 3.C), plausibly in an attempt to compensate for the prolonged decrease of GABA receptor 1 alpha protein level.In addition, it has been reported the internalization of AMPA receptors due to increased levels of Aβ (Marttinen et al., 2018).Given the importance of AMPA transmission for synaptic plasticity, this finding could explain the lack of increased IEG expression in App NL-G-F mice upon KA-induced neuronal activation (Fig. 1.C, D) (Rao et al., 2006).This is in line with previous work showing a suppression of IEG in the presence of Aβ (Dickey et al., 2004).
Here we propose that microglia play a protective role at earlier stages of amyloidosis preventing plaque aggregation and supporting FSI physiological functionality, which results in a preserved network synchronization in the gamma rhythm.Conversely, microglia dynamic alterations in the early stage of amyloid pathology progression might facilitate plaque aggregation, cause functional impairment in FSI, and, consequently, prevent the synchronization of the hippocampal network in the gamma rhythm.

Conclusions
In conclusion, here we identified an alteration of microglia dynamics during the early stage of amyloid pathology, before the manifestation of widespread Aβ plaque load in the hippocampus.Our findings suggest that microglia dynamic alterations might represent a turning point in AD pathology progression, contributing to Aβ plaque formation, FSI dysfunction, and circuit failure in the hippocampus.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Fig. 1 .
Fig. 1.Progressive impairment of Hippocampal gamma oscillations in App NL-G-F mice.A) Graphic methods for LFP recordings and PCR.B) Representative power spectra from WT and App NL-G-F at different ages.C) Scatter plot of gamma oscillation power in WT and App NL-G-F mice at different ages showing a progressive decrease of the hippocampal network synchrony in App NL-G-F mice.The graph shows a two-way ANOVA with Holm-Sidak's multiple comparisons test at P30 (WT n = 6 vs App NL-G-F n = 6), P40 (WT n = 6 vs App NL-G-F n = 6), P50 (WT n = 6 vs App NL-G-F n = 9), and P60 (WT n = 7 vs App NL-G-F n = 11).Intra-group differences were tested too to evaluate the changes in gamma power at different timepoints in the same group (WT and App NL-G-F P30 vs P40, P30 vs P50, P30 vs P60.D) Two-way ANOVA with Holm-Sidak's multiple comparisons test for Fos normalized expression relative to control showing overexpression in WT KA-exposed slices (WT control n = 4 vs KA n = 4) but not in App NL-G-F KAexposed slices (App NL-G-F P30 control n = 3 vs KA n = 3; App NL-G-F P60 control n = 4 vs KA n = 4).E) Two-way ANOVA with Holm-Sidak's multiple comparisons test for Npas4 normalized expression relative to control showing overexpression in WT KA-exposed slices (WT control n = 4 vs KA n = 4) but not in App NL-G-F KA − exposed slices (App NL-G-F P30 control n = 3 vs KA n = 3; App NL-G-F P60 n = 3 control vs KA n = 3).Data are presented as mean ± SEM, and "n" indicates the number of mice.*p < 0,05, **p < 0,01, ***p < 0,001, **** p < 0,0001.

Fig. 2 .
Fig. 2. Fast-spiking interneurons' intrinsic properties and postsynaptic input change during amyloid pathology progression.A) Graphic methods for whole cell patch clamp recordings from FSI. B.i) Representative phase-plot of the FSI's action potentials from App NL-G-F mice at P30 and P60 showing different action potential shapes and resting membrane potential (Em0) in the two groups.B.ii) Current-clamp recording traces of the representative FSI in B.i. B.iii) Cropped recording of the action potential of the FSI in Bi-ii.C) PCA of the intrinsic properties of WT P30 (n = 17), WT P60 (n = 11), App NL-G-F P30 (n = 10), and App NL-G-F P60 (n = 17) FSI.PC1 accounts for 37 % of variation, PC2 for 34 % and PC3 for 19 %.App NL-G-F P60 FSI exhibit distinct properties that make them cluster separately from the other groups.D-H) Scatter plots of unpaired t-test of FSI's intrinsic properties and post-synaptic input in the App NL-G-F group (P30 n = 11 vs P60 n = 17), showing D) resting membrane potential, E) firing threshold, F) peak downstroke, G) charge transfer, H) sEPSC frequency.Data are presented as mean ± SEM, and "n" indicates the number of neurons.*p < 0,05, **p < 0,01, ***p < 0,001.

Table 1
Intrinsic properties of FSI.

Table 2
Synaptic-specific and FSI-specific gene expression.

Table 3
Representative genes for different microglia populations.
Table showing base mean, log2 fold-change and adjusted p-value of selected genes expressed by distinct microglia populations.Non-available adjusted p-value (NA)