Microglia-to-neuron signaling links APOE4 and inflammation to enhanced neuronal lipid metabolism and network activity

Significance

Microglia impact neuronal function, evident by neuroinflammation’s role in neurodegenerative diseases, including Alzheimer’s disease (AD). APOE4, the strongest genetic risk factor for AD, alters microglial activation and neuronal excitability, but the mechanism by which APOE4 impacts inflammatory microglial regulation of neuronal function is unclear. Using monocultures of iPSC-derived microglia and neurons indicates that intercellular signaling by conditioned media (CM) or exosomes mediates microglial regulation of neuronal network function. Inflammatory APOE4 microglia increase neuronal hyperexcitability more than APOE3 microglia, Neuronal APOE genotype plays a modest role in the response to microglial signaling. Metabolism of neuronal lipid droplets (LD) is required to sustain neuronal network activity, and microglia-driven increases in network activity increase neuronal LD metabolism, linking microglia-driven hyperexcitability to neuronal metabolic stress.

Abstract

Microglia regulate neuronal circuit plasticity. Disrupting their homeostatic function has detrimental effects on neuronal circuit health. Neuroinflammation contributes to the onset and progression of neurodegenerative diseases, including Alzheimer’s disease (AD), with several microglial activation genes linked to increased risk for these conditions. Inflammatory microglia alter neuronal excitability, inducing metabolic strain. Interestingly, expression of APOE4, the strongest genetic risk factor for AD, affects both microglial activation and neuronal excitability, highlighting the interplay between lipid metabolism, inflammation, and neuronal function. It remains unclear how microglial inflammatory state is conveyed to neurons to affect circuit function and whether APOE4 expression alters this intercellular communication. Here, we use a reductionist model of human iPSC-derived microglial and neuronal monocultures to dissect how the APOE genotype in each cell type independently contributes to microglial regulation of neuronal activity during inflammation. Conditioned media (CM) from LPS-stimulated microglia increased neuronal network activity, assessed by calcium imaging, with APOE4 microglial CM driving greater neuronal activity than APOE3 CM. Both APOE3 and APOE4 neurons increase network activity in response to CM treatments, while APOE4 neurons uniquely increase presynaptic puncta in response to APOE4 microglial CM. CM-derived exosomes from LPS-stimulated microglia can mediate increases to network activity. Finally, increased network activity is accompanied by increased lipid droplet (LD) metabolism, and blocking LD metabolism abolishes network activity. These findings illuminate how microglia-to-neuron communication drives inflammation-induced changes in neuronal circuit function, demonstrate a role for neuronal LDs in network activity, and support a potential mechanism through which APOE4 increases neuronal excitability.

Microglia perform essential roles in neuronal circuit function and synaptic plasticity, mediating synaptic pruning and modulating synaptic strength (1–3). As the brain’s resident immune cells, they are also the main drivers of neuroinflammation, a phenomenon implicated in neurodegenerative diseases, including Alzheimer’s disease (AD). Microglia are activated in response to pathogens, injury, and other insults, altering their gene expression and function to recruit relevant cells to resolve the inflammatory process. Although the extent to which this disruption in microglial homeostatic function affects neuronal health is not fully understood, multiple genome-wide association studies have linked microglial activation genes, including TREM2, APOE, BIN1, and SPI1, with an increased risk of AD (4–7). Thus, understanding the mechanisms by which microglial inflammatory responses alter neuronal function is essential for better understanding the early mechanisms underlying neurodegeneration.

Inflammatory microglia alter neuronal network activity and metabolism. In vivo treatment with lipopolysaccharide (LPS), a commonly used model of inflammation, can increase neuronal excitability, induce seizures, and—upon prolonged exposure—lead to oxidative stress, synapse dysfunction, and cell death (8, 9). Additionally, as microglia adopt an inflammatory state, they undergo a metabolic shift that impacts surrounding cells. Neuronal metabolism specifically is altered by inflammatory microglia, as neurons are known to have an altered rate of respiration and they can switch to alternate energy sources to sustain network activity (10, 11). Identifying how microglial inflammatory responses are communicated to neurons and how they affect neuronal metabolism is critical for understanding the vulnerabilities contributing to neuroinflammation-driven neurodegeneration.

A key mechanism that could mediate microglia-to-neuron communication is the release of exosomes. Exosomes are membrane-bound particles that contain a variety of functional cargoes, including proteins, RNAs, and metabolites, which change based on donor cell type and state (12–14). Microglia, like most cells, secrete exosomes under basal conditions, but inflammatory microglia secrete exosomes with distinct cargoes that can propagate the inflammatory phenotype to recipient microglia (15–18). Exosomes from inflammatory microglia have also been shown to inhibit neuroprogenitor proliferation (19) and potentially impair cognitive function, as blocking exosome release in an LPS mouse model improved behavioral outcomes (20). Although their role in regulating circuit function is not fully understood, microglial exosomes are promising candidates for relaying inflammatory cues to alter neuronal excitability.

Among the genes associated with increased risk for AD, APOE stands out for its strong influence on both microglial reactivity and neuronal function. APOE, which encodes the Apolipoprotein E lipid carrier, has three allelic variants of which APOE4 is the strongest genetic risk factor for late-onset AD (21). In microglia, APOE4 expression drives a proinflammatory transcriptional profile, characterized by lower expression of homeostatic genes and higher expression of inflammatory cytokine genes (22–26). Additionally, APOE4 microglia accumulate lipid droplets (LDs)—a hallmark of their inflammatory response—even in the absence of a pathological stimulus and they can induce tau phosphorylation, apoptosis, and disrupt neuronal network activity (22, 27). Interestingly, recent work demonstrated that APOE4-related neuronal pathologies improved in the absence of microglia in an AD mouse model, suggesting that microglial APOE4 expression could drive neurodegeneration (28). In addition to the effects of APOE4 expression in microglia, neuronal APOE4 expression is also associated with increases in neuronal excitability (25, 29, 30), posing the question of whether neuronal APOE genotype could influence their susceptibility to excitotoxicity. However, it remains unclear how APOE4 expression in each cell type independently contributes to microglia-to-neuron signaling during inflammation, a question that is difficult to address in vivo, where multiple cell types contribute to the inflammatory response.

In this study, we investigate microglia-to-neuron signaling in the context of inflammation. Using induced pluripotent stem cell (iPSC)-derived microglial and neuronal monocultures, we find that exosomes from inflammatory microglia increase synchronized neuronal network activity. Assessing the bioactivity of microglial conditioned media (CM) shows that CM from LPS-stimulated inflammatory microglia similarly increases neuronal network activity. We use CRISPR editing to dissect the contributions of APOE3 and isoAPOE4 genotypes in microglia-to-neuron signaling. Signaling from inflammatory isoAPOE4 microglia drives increased network activity in neurons, regardless of neuronal APOE genotype, while APOE4 neurons uniquely increase presynaptic puncta in response to APOE4 microglial CM. Considering the cellular energy source for the increased neuronal activity, we find that CM from inflammatory microglia decrease neuronal LD load. We identify neuronal LD metabolism as a mechanism that is required to sustain network activity and that adapts with microglia-driven hyperexcitability.

Results

Generating and Characterizing iPSC-Derived iNs and Inflammatory iMGLs.

To enable our study of how APOE genotype influences microglial-to-neuron signaling during inflammation, we CRISPR-edited an established APOE3/E3 iPSC line (31) to change amino acid residue 112 from cysteine to arginine and generate isogenic APOE4/E4 iPSCs (isoAPOE4; SI Appendix, Fig. S1A). iPSC lines were karyotyped and characterized via a single nucleotide polymorphism (SNP) array to confirm euploidy and they were evaluated for their expression of pluripotency markers TRA 1-60 and SSEA4 by flow cytometry (SI Appendix, Fig. S1 B and C). Expression of pluripotency markers Nanog, SSEA-4, SOX2, and Oct-4 was also confirmed in both iPSC lines via immunocytochemistry (SI Appendix, Fig. S1 D and E).

Induced neurons (iNs) were generated following a Classic NGN2 induction protocol (SI Appendix, Fig. S2A) (32) with modifications to enhance neuronal maturation (Enhanced NGN2 neural induction; SI Appendix, Fig. S2B) (33). As previously reported, by day 24, iNs generated with a Classic NGN2 induction express APOE, pan neuronal markers MAP2 and TUBB3, and excitatory synapse markers VGLUT1 and VGLUT2 at higher levels than inhibitory markers like GAD1 and GAD2, which we confirmed by RT-qPCR (SI Appendix, Fig. S2 A and F). To enhance neuronal maturation, we used a prolonged 49-day protocol with additional neurotrophic factors (SI Appendix, Fig. S2B). Immunocytochemistry revealed that like iNs generated with a Classic NGN2 induction, iNs generated via an Enhanced NGN2 induction protocol express excitatory synaptic markers vGlut1 and 2 while lacking expression of inhibitory marker Gad67 (SI Appendix, Fig. S2E). We further characterized iNs via immunocytochemistry to evaluate neuronal marker NeuN and cortical marker Cux1 protein expression (SI Appendix, Fig. S2C). With a Classic NGN2 induction of 24 d (SI Appendix, Fig. S2A), 86.7% of APOE3 and 89% of isoAPOE4 iNs express NeuN, and 72.9 and 67.3% of NeuN-expressing cells also express cortical marker Cux1 (SI Appendix, Fig. S2D). Compared to Classic NGN2 D24 iNs, however, Enhanced D49 iNs have greater proportions of both NeuN-expressing cells. NeuN+ cells increase significantly to 94.4 and 94.1% of APOE3 and isoAPOE4 iNs, respectively, with Cux1+ cells averaging 93.8 and 88.8% of NeuN+ cells (SI Appendix, Fig. S2D). Indicative of their greater maturity, Enhanced D49 iNs also display spontaneous network activity, which we assessed via live imaging with calcium indicator Fluo-4AM (Movie S1), a feature rarely observed in iNs generated with Classic protocol even at D45 (Movie S4).

To generate induced microglia-like cells (iMGLs), we followed an established protocol in which iPSCs are first differentiated into CD43-expressing hematopoietic progenitors (HPCs) and further differentiated into mature microglia over the course of 7 wk (34) (SI Appendix, Fig. S3A). The expression of CD43 in HPCs was characterized via immunocytochemistry, and we found 93.7 and 92.5% of HPCs express CD43 in APOE3 and isoAPOE4 cell lines, respectively, with little variation across batches (SEM = 1.4 and 1.6; SI Appendix, Fig. S3B). Mature iMGLs were functionally validated via a phagocytosis assay with fluorescent beads. In APOE3 iMGL cultures, 89% of cells displayed phagocytic activity, while isoAPOE4 iMGL cultures had a significantly larger proportion of phagocytic cells at 95% (SI Appendix, Fig. S3C). We observed via immunocytochemistry that a large proportion of both APOE3 and isoAPOE4 iMGLs express microglial markers PU.1 (93.4 and 97.9%, respectively) and Iba1 (89.3 and 89.7%, respectively) with no significant difference between APOE genotypes (SI Appendix, Fig. S3D). Immunocytochemistry also confirmed that APOE3 and isoAPOE4 iMGLs express microglial markers Trem2 and P2RY12 (SI Appendix, Fig. S3E). Via RT-qPCR, we observed similar expression levels of TREM2, IBA1, P2RY12, and APOE in APOE3 compared to isoAPOE4 iMGLs (SI Appendix, Fig. S3G).

Given previous reports of APOE4 microglia having a disrupted homeostatic state and enhanced inflammatory response compared to APOE3 microglia (22, 27, 35), we wanted to evaluate iMGL signaling under basal and inflammatory conditions. To induce a robust inflammatory response in iMGLs, we treated cells with 50 ng/mL LPS every other day for a week, allowing exosomes and other signaling factors produced as part of the inflammatory response to accumulate in the CM (Fig. 1A). Compared to PBS, LPS treatment caused clear morphological changes, including hypertrophic cell bodies, and increased expression of inflammatory cytokine genes IL6, IL10, and CCL2 evaluated by RT-qPCR in both APOE3 and isoAPOE4 iMGLs (SI Appendix, Fig. S3G). In APOE3 iMGLs, LPS treatment decreased expression of homeostatic P2RY12; however, in isoAPOE4 iMGLs, LPS did not affect P2RY12 expression, suggesting an altered inflammatory response (SI Appendix, Fig. S3G). LPS treatment also caused accumulation of LDs in both APOE3 and isoAPOE4 iMGLs compared to PBS controls, but isoAPOE4 iMGLs had a significantly greater LD load than their APOE3 counterparts both with PBS- and LPS treatments (SI Appendix, Fig. S3F). We assessed the expression of genes involved in LD metabolism and found no differences in PLIN2, ACAT1, or DDHD2 expression between APOE3 and isoAPOE4 iMGL with or without LPS treatment (SI Appendix, Fig. S3G). We did, however, observe differences in the expression of DGAT-1, which encodes an enzyme involved in the synthesis of triacylglycerol (TAG), a major component of LDs. Surprisingly, DGAT-1 expression was significantly lower in isoAPOE4 iMGLs compared to APOE3 iMGLs and, whereas DGAT-1 expression decreased with LPS treatment in APOE3 iMGLs, it did not change in isoAPOE4 iMGLs treated with LPS (SI Appendix, Fig. S3G). This could reflect a compensatory mechanism in iMGLs with high LD load (36). Overall, these data show that a week-long LPS treatment successfully induces a robust inflammatory response in both APOE3 and isoAPOE4 iMGLs, with isoAPOE4 iMGLs having a distinct inflammatory phenotype that could alter downstream intercellular signaling.

Fig. 1.

Exosomes from inflammatory iMGLs increase iN network activity. (A) Differentiation timelines for iMGLs and iNs. (B) Representative brightfield images of D43 iMGLs after a 7-d treatment with PBS or 50 ng/mL LPS. (C) Transmission electron micrographs of iMGL exosomes. (D) Schematic of treatment conditions. Exosomes were isolated from APOE3 iMGL conditioned media (CM) via sequential ultracentrifugation, and APOE3 D46 iNs received either PBS or exosomes from PBS- or LPS-treated iMGLs (PBS-Exos and LPS-Exos, respectively). (E) Representative raster plots of Ca⁺⁺-imaging recordings in APOE3 iNs at D48. Imaging with calcium indicator Fluo4-AM was conducted after 48 h treatment with PBS or iMGL Exos. Blue tick marks represent coordinated events, defined as instances where >50% of cells spiked within a 250 ms window (5 frames). Red tick marks represent synchronized events, defined as instances where >50% of cells spiked within a 50 ms window (1 frame). (F–H) Enlarged examples of activity bursts over 0.5 s. (F) Correlated activity where only 23% of cells spiked within a 250 ms window, not meeting criteria for a coordinated event. (G) Coordinated event in which 65% of cells spiked within a 250 ms window. (H) Synchronous event in which 59% of cells fired within a 50 ms window. (I and J) Violin plots of spike frequency (I) and amplitude (J) per cell normalized as the fold change relative to the mean of PBS-treated iNs. (N = 665 to 1,132 cells from 2 independent experiments). (K and L) Violin plots showing number of coordinated events (K) and synchronized events (L) per 100 s recording. (N = 6 to 12 recordings, 1 to 3 recordings per well from 2 independent experiments). (M and N) Violin plots showing the fraction of cells engaged in each coordinated event (M; N = 43 to 89 coordinated events from 2 independent experiments) or synchronized event (N; N = 18 to 53 synchronized events from 2 independent experiments). Violin plot lines represent median (solid) and quartiles (dashed). Median and quartile values can be found in SI Appendix, Table S4. One-way ANOVA with Tukey test for multiple comparisons. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Exosomes from Inflammatory Microglia Increase Neuronal Network Activity.

To assess the contribution of exosomes in microglia-to-neuron signaling in the context of inflammation, we harvested exosomes from PBS- and LPS-treated APOE3 iMGLs (Fig. 1 A and B). iMGL exosomes were isolated from CM via sequential ultracentrifugation (SI Appendix, Fig. S4A), as we previously described (37), and their characteristic cup-like morphology was identified by transmission electron microscopy (Fig. 1C). Additionally, we demonstrated by Western blot that iMGL exosomes contain exosomal markers Alix and Flotillin-1 (SI Appendix, Fig. S4B).

We treated D46 APOE3 iNs with either PBS (vehicle) or purified exosomes from PBS- or LPS-treated iMGLs (Fig. 1D, PBS, PBS-exos, LPS-exos). Exosome samples were normalized based on a protein assay, and equal amounts were administered across conditions. After 48 h, we imaged neuronal activity as Ca⁺⁺ transients, hereafter referred to as “spikes,” in iN cultures over 100 s epochs using the Ca⁺⁺ indicator Fluo4-AM (Fig. 1E). Assessing individual neuron firing rates, as the frequency of Ca⁺⁺ transients, we found that APOE3 neurons significantly increased spike frequency when treated with either PBS- or LPS-exosomes compared to the control treatment with PBS alone (Fig. 1I). Additionally, spike amplitude increased with LPS-exosomes compared to PBS controls (Fig. 1J).

iN cultures exhibited repeated bouts of network activity throughout the 100 s epochs, with varying proportions of neurons engaged in each bout (Fig. 1E and Movies S1–S3). This led us to wonder whether exosomes, which increased the neuronal firing rate, could also increase network connectivity, increasing the frequency of—and the proportion of cells recruited to—network events. To address this question, we quantified two types of network activity: coordinated and synchronized events (Fig. 1 F–H). We defined coordinated events as instances where >50% of neurons in the field of view (FOV) spiked within a broad 250 ms window, representing the temporally dispersed network activity (Fig. 1G). We then defined synchronized events as the subset of coordinated events in which >50% of neurons in the FOV were coactive within a shorter 50msec window, representing only the temporally precise activity of neuronal networks with robust synaptic connectivity (Fig. 1H).

We quantified the incidence of coordinated (Fig. 1K) and synchronized events (Fig. 1L) as well as the fraction of cells active in both types of network events (Fig. 1 M and N). Compared to treatment with PBS, treatment with PBS-exosomes did not affect the incidence of coordinated events in iN cultures (Fig. 1K) but rather increased the synchronicity of coordinated activity resulting in a greater number of synchronized events (Fig. 1L). PBS-exosome treatment in iN cultures also promoted the recruitment of additional neurons into both coordinated and synchronized events, increasing the fractions of coactive cells compared to control PBS treatment (Fig. 1 M and N). Compared to PBS treatment, treatment with LPS-exosomes increased the incidence of global coordinated events and enhanced synchronicity in coordinated networks, increasing also the numbers of synchronized events in iN cultures (Fig. 1 K and L). LPS-exosomes, however, recruited significantly fewer neurons to coordinated events than PBS-exosomes and did not increase the fraction of coactive cells in synchronized events compared to control PBS treatment (Fig. 1 M and N). This suggests that while PBS-exosomes enhance the synchronicity of existing coordinated activity and expand synchronized neuronal networks, LPS-exosomes primarily increase the incidence of coordinated and synchronized events without promoting the recruitment of additional neurons into these networks.

In summary, APOE3 iMGL exosomes increase neuronal spiking frequency independent of microglial inflammatory status. PBS- and LPS-exosomes, however, have distinct effects on network activity. While PBS-exosomes appear to foster circuit refinement, LPS-exosomes drive a more generalized increase in the activity of both strongly and weakly connected neuronal networks.

Inflammatory APOE4 Microglial Signaling Drives Greater Network Activity than APOE3.

We next sought to evaluate how microglial and neuronal APOE genotype affects microglia-to-neuron signaling via CM in the context of inflammation. Using both APOE3 and isoAPOE4 iN cultures, we performed a half-media exchange on iNs with CM from PBS- and LPS-treated APOE3 and isoAPOE4 iMGLs (Fig. 2A). To control for the dilution of neuronal media, we used PBS- or LPS-unconditioned media (UCM) as control treatments. After 48 h, we imaged spontaneous neuronal activity in APOE3 iN and isoAPOE4 iN cultures (Fig. 2 B and G). Treating APOE3 and isoAPOE4 iNs with either PBC UCM or LPS UCM showed comparable low baseline activity in iNs of both genotypes, which is likely due to the dilution of neurotrophic factors in neuronal media. Treatment with CM from PBS-treated APOE3 iMGLs decreased Ca⁺⁺ spike amplitude in APOE3 iNs while increasing Ca⁺⁺ spike amplitude in isoAPOE4 iNs (Fig. 2 C and H). CM from PBS-treated isoAPOE4 iMGLs, however, did not affect Ca⁺⁺ spike amplitude in iNs of either APOE genotype. CM from LPS-treated APOE3 and isoAPOE4 iMGLs increased Ca⁺⁺ spike amplitude in both APOE3 and isoAPOE4 iNs, with isoAPOE4 LPS CM causing a significantly greater Ca⁺⁺ spike amplitude than APOE3 LPS CM (Fig. 2 C and H).

Fig. 2.

Inflammatory APOE4 microglial signaling drives greater neuronal network activity than APOE3. (A) Schematic of experimental design. APOE3 (B–F, blue) and isogenic APOE4 (G–K, red) iNs received a half-media exchange with either unconditioned media (UCM) or iMGL CM and underwent Ca⁺⁺-imaging with Fluo4-AM 48 h later. (B and G) Representative raster plots of Ca⁺⁺-imaging recordings from APOE3 (B) and isogenic APOE4 (G) neurons treated with UCM or CM from PBS- or LPS-treated iMGLs. Red tick marks represent synchronized events. (C, D, H, and I) Violin plots of spike amplitude (C and H) and spike frequency (D and I) plotted as fold change relative to the average of PBS UCM-treated iNs (N = 1223 to 3211 cells from 4 independent experiments; Statistical comparisons were performed separately within the PBS-treated and LPS-treated groups using one-way ANOVAs followed by Tukey tests). (E, F, J, and K) Violin plots showing number of coordinated events (E and J) and synchronized events (F and K) per 100 s recording (N = 26 to 33 recordings, 1 to 2 recordings per well from 4 independent experiments; Statistical comparisons were performed separately within the PBS-treated and LPS-treated groups using Kruskal–Wallis tests with Dunn’s multiple comparison tests). (L–O) Violin plots showing statistical comparisons in spike frequency (L), spike amplitude (M), coordinated event count (N), and synchronized event count (O) across neuronal APOE genotype for iNs treated with LPS-stimulated iMGL CM. For L–O, statistical comparisons were performed with one-way ANOVA with Sidak multiple comparisons test. Violin plot lines represent median (solid) and quartiles (dashed), and values can be found in SI Appendix, Table S4. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

LPS-treated iMGL CM showed a genotype-dependent effect with isoAPOE4 CM driving a greater increase in neuronal activity than APOE3 CM (Fig. 2 D and I). In addition, LPS-CM increased both coordinated and synchronized events in iNs independent of their APOE genotype (Fig. 2 E, F, J, and K) and this increase in coordinated events occurs by recruitment of neurons into the highly coactive networks (SI Appendix, Fig. S5 B and D). By contrast, CM from PBS-treated iMGLs increased spike frequency in iNs independent of the genotypes of the iMGLs or the iNs but had no impact on coordinated or synchronized events (Fig. 2 D–F and I–K). Overall, these data indicate that inflammatory iMGL CM has a distinct effect on neuronal network activity and synchronicity.

We performed additional post hoc comparisons to evaluate whether neuronal APOE4 genotype further increases neuronal activity in response to CM from inflammatory iMGLs. When treated with either APOE3 or isoAPOE4 LPS iMGL CM, isoAPOE4 iNs showed significantly higher Ca⁺⁺ spike amplitude and Ca⁺⁺ spike frequency compared to APOE3 iNs (Fig. 2 L and M). IsoAPOE4 iNs treated with isoAPOE4 LPS iMGL CM significantly increased both coordinated and synchronized events compared to APOE3 iNs treated with APOE3 LPS iMGL CM. We did not observe a difference in coordinated or synchronized event count between APOE3 and isoAPOE4 iNs when receiving the same LPS iMGL CM treatment (Fig. 2 N and O). These findings suggest that APOE4 genotype in neurons increases neuronal excitability in response to CM but APOE4 genotype in microglia is responsible for driving increased network activity during inflammation.

Increased network activity could result from structural changes to synapses. We immunolabeled iNs for the presynaptic marker synapsin and quantified synapsin+ puncta (SI Appendix, Fig. S5 E and F). isoAPOE4 iNs increased synapsin+ puncta following treatment with both PBS and LPS isoAPOE4 iMGL CM (SI Appendix, Fig. S5 F, Right) but APOE3 iNs were not affected (SI Appendix, Fig. S5 F, Left) These data suggest that isoAPOE4 iNs exhibit greater synaptic plasticity than their APOE3 counterparts and that isoAPOE4 iMGL CM might have a unique ability to remodel presynaptic elements in isoAPOE4 iNs. Additionally, these findings show that structural changes to presynaptic elements are not required to increase network activity, suggesting that other mechanisms, such as increased intrinsic excitability may contribute to greater network activity. Overall, our findings suggest that microglial APOE genotype modulates their signaling capabilities, especially under inflammatory conditions. Interestingly, although we detected similar changes to network activity in both APOE3 and isoAPOE4 iNs in response to these treatment conditions, the underlying structural mechanisms mediating these changes in activity may depend on neuronal APOE genotype.

Increased iN Network Activity Is Accompanied by a Depletion of Neuronal LDs.

Given the robust increase in neuronal network activity that we observed, and its known energetic cost, we questioned how neurons could fuel such sudden changes in energy demand. In many cell types, LDs can serve as energy stores and the TAGs that make up LDs can be metabolized to fuel a variety of cellular processes (38). Although neurons are not thought to rely on lipids as an energy source, a recent study demonstrated that neurons can utilize LD-derived free fatty acids (FFAs) to sustain synaptic activity (39). We investigated whether the conditions that increased neuronal network activity affected LD load in neurons by staining CM-treated D48 iNs with BODIPY-493/503 (Fig. 3A). To capture LDs in the entirety of neuronal cell bodies, we quantified LDs using maximum intensity projections of 10 μm confocal z-stacks acquired in 1 μm steps and normalized LDs to nuclei count. We found that in APOE3 iNs, isoAPOE4 PBS iMGL CM decreased LDs compared to PBS UCM (Fig. 3 B, Left) while in APOE4 iNs neither APOE3 nor isoAPOE4 PBS CM altered neuronal LD load (Fig. 3 B, Right). Compared to LPS UCM, treatment with both APOE3 and isoAPOE4 LPS iMGL CM significantly decreased LDs in iN cultures of both APOE genotypes (Fig. 3B). To rule out potential artifacts arising from the 10 μm projections, we performed alternate analysis and quantified LDs in single confocal planes and normalized LD count to the area of βIII-tubulin-positive neurites. We found similar, although smaller, decreases in LD load with this alternate analysis (SI Appendix, Fig. S6 A and B). Interestingly, the LPS-CM treatments which decreased neuronal LD load are the same conditions that increased network activity, supporting our hypothesis that LD metabolism might help neurons meet the energy demand driven by increased network activity.

Fig. 3.

Inflammatory iMGL CM decreases neuronal lipid droplet (LD) load. D48 or D24 iNs were treated with UCM or CM from PBS- or LPS-treated iMGLs for 48 h and immunolabeled for βIII-tubulin (red). LDs were labeled with BODIPY-493/503 (green), and nuclei were labeled with DAPI (white). (A) Representative images of UCM or CM-treated APOE3 (Top, blue) and isoAPOE4 (Bottom, red) D48 iNs. Images are max intensity projections of 10 μm confocal z-stacks taken at 1 μm steps and cropped for visualization. (B) Quantification of LDs in APOE3 (Left, blue) and isoAPOE4 (Right, red) iNs. LDs were normalized to DAPI-labeled nuclei per field of view, averaged by well, and graphed as the fold change relative to UCM PBS conditions (N = 9 to 16 wells from 3 to 4 independent experiments). (C) Representative images of CM-treated D24 iNs and (D) quantification of LDs/nuclei (N = 9 to 17 wells from 2 to 3 independent experiments). Statistical comparisons were performed separately within the PBS-treated and LPS-treated groups using one-way ANOVAs followed by Tukey’s post hoc tests. Bar graphs represent the mean with SEM. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

We then tested whether changes in LD load were present in less mature neuronal cultures in which circuit activity is less robust. For this, we used the Classic NGN2 induction method (32) with minor modifications (SI Appendix, Fig. S2A). We treated D24 iNs with the same iMGL CM conditions described above and used BOIDPY-493/503 to label and quantify LDs (Fig. 3 C and D). While D24 iNs had fewer LDs than the more mature D48 iNs (SI Appendix, Fig. S6C), they had a similar response to the iMGL CM treatments. In APOE3 iNs, CM-treatment from APOE3 and isoAPOE4 PBS-iMGLs did not alter LD load, while in isoAPOE4 iNs, CM-treatment from APOE3 PBS-iMGLs decreased LDs (Fig. 3D). Like more mature D48 iNs, D24 iNs of both APOE3 and isoAPOE4 genotypes had significantly decreased LDs when treated with CM from LPS-treated iMGLs, regardless of iMGL APOE genotype (Fig. 3D). This suggests that inflammatory microglia can modulate neuronal LD metabolism in iN cultures at different maturation stages, and this modulation is not dependent on preexisting iN network activity. Overall, these results support our hypothesis that neuronal LD stores may be metabolized to support the increased energy demand of changes to spontaneous activity.

LD Metabolism Is Required to Sustain Network Activity.

To further investigate LD metabolism in iN cultures, we employed an additional pair of APOE3 and isoAPOE4 iPSC lines to ensure that our observations were not specific to a single isogenic background (SI Appendix, Fig. S7). The additional iPSC lines appeared genetically normal after characterization with SNP array, although karyotyping showed that 25% of APOE3 iPSCs (5/20 cells tested) had trisomy of chromosome 1 (SI Appendix, Fig. S7 A and B). Both APOE3 and isoAPOE4 iPSC lines showed a high level of expression of pluripotency markers TRA 1-60 and SSEA-4 by FACS and expressed Nanog, SSEA-4, SOX2, and Oct-4 by immunocytochemistry (SI Appendix, Fig. S7 A–D). We also confirmed that D24 iNs generated from both iPSC lines express neuronal markers NeuN, Cux1, vGlut1, and vGlut2 by immunocytochemistry (SI Appendix, Fig. S7 E–G), and confirmed the expression of APOE, MAP2, and TUBB3 as well as synaptic markers like VGLUT1 and VGLUT2 via RT-qPCR (SI Appendix, Fig. S7H).

To investigate whether LDs are actively metabolized in iNs under basal conditions, we treated iN cultures with KLH45, an inhibitor of neuron-specific lipase DDHD2 which breaks down LD TGs into FFAs that can be used as energy (40). We hypothesized that if iNs continuously metabolize LDs under basal conditions to support functions such as synaptic activity, blocking LD metabolism would result in a rapid accumulation of LDs (Fig. 4A). We conducted a time-point experiment where we treated APOE3 and isoAPOE4 D24 iNs with 5 μM KLH45 or vehicle (EtOH) for 2, 8, or 24 h. Strikingly, we found that iNs treated with KLH45 rapidly accumulate LDs. Within 8 h of exposure to KLH45, iNs show more than a 2-fold increase in number of LDs, and this accumulation further increases by 24 h (SI Appendix, Fig. S8 A–D). Interestingly, both APOE3 and isoAPOE4 D24 iNs accumulated LDs at similar rates. When we evaluated the expression of genes involved in LD metabolism, there was also no difference in the expression levels of DDHD2, PLIN-1, DGAT-1, and ATGL between APOE3 and isoAPOE4 iNs (SI Appendix, Figs. S2F and S6H).

Fig. 4.

LD metabolism is required for neuronal network activity. (A) Schematic of KLH45 mechanism of action and our experimental hypothesis. DDHD2 sits on the surface of LDs and breaks down triglycerides (TGs) into free fatty acids (FFAs). KLH45 inhibits DDHD2, which in turn results in cells accumulating LDs. We hypothesized that if LD metabolism supports neuronal network activity, KLH45 would result in a decrease of network activity. (B–E) D48 iNs were treated with vehicle or 5 μM KLH45 for 24 h and analyzed for relative changes in LDs and synapses. Schematics created in BioRender.com. (B) Representative images of D48 iNs immunolabeled for βIII-tubulin (red) with LDs labeled with BODIPY-493/503 (green) and nuclei labeled with DAPI (white). Images are max intensity projections of 10 μm confocal z-stacks taken at 1 μm steps and cropped for visualization. (C) Quantification of LDs per field of view, normalized to neurite area, and averaged by well. Data are shown as fold change relative to the APOE3 Vehicle condition (N = 9 to 10 wells from 2 independent experiments; Two-way ANOVA with Sidak correction for multiple comparisons). (D) Representative images of iNs immunolabeled for βIII-tubulin (white) and synapsin 1/2 (red) with nuclei labeled with DAPI (blue). Confocal images were cropped for visualization. (E) Quantification of synapsin puncta per field of view, normalized to neurite area, and averaged by well. Data are shown as fold change relative to the APOE3 Vehicle condition (N = 9 to 10 wells from 2 independent experiments; Two-way ANOVA with Sidak correction for multiple comparisons). (F) Schematic of KLH45 treatment timeline used for Ca⁺⁺-imaging with Fluo4-AM in D48 iNs. iNs received a 24 h pretreatment with either vehicle (EtOH) or 5 μM KLH45. KLH45 was either removed (KLH45 Pre) or kept (KLH45 Pre+Post) during Ca⁺⁺-imaging. (G) Representative raster plots. Red tick marks represent synchronized events. (H–M) Quantification of Ca⁺⁺-imaging features for APOE3 (H–J, blue) and isoAPOE4 (K-M, red) iNs. (H and K) Quantification of average spike frequency per cell normalized as the fold change to the average of Veh-treated iNs (N = 326 to 3236 cells from 2 independent experiments). (I and L) Quantification of average spike amplitude per cell, normalized as the fold change to the average of Veh-treated iNs (N = 326 to 3236 cells from 2 independent experiments). (J and M) Violin plots showing quantification of coordinated events per 100 s recording session (N = 17 to 20 recordings, 1 to 2 recordings per well from 2 independent experiments). One-way ANOVA with Tukey correction for multiple comparisons. Bar graphs represent the mean, and error bars represent SEM. Violin plot lines represent median (solid) and quartiles (dashed), and values can be found in SI Appendix, Table S4. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

We then looked at D48 Enhanced iNs and found that a 24 h KLH45 treatment also results in an accumulation of LDs (Fig. 4 B and C and SI Appendix, Fig. S8 G and H). Interestingly, isoAPOE4 iNs had a higher abundance of LDs compared to APOE3 iNs both with vehicle and KLH45 treatment, a difference we did not observe in D24 iNs. This suggests that neuronal APOE genotype could affect neuronal LD metabolism in more mature iN cultures.

To test whether LD metabolism is required to sustain spontaneous neuronal network activity, we performed calcium imaging on D48 iNs pretreated with vehicle or KLH45 for 24 h, with the drug treatment either maintained or removed during the calcium imaging (Fig. 4 F and G). In both APOE3 and isoAPOE4 iNs, we saw a marked decrease in spike frequency and amplitude with KLH45 pretreatment, regardless of whether the drug was maintained during the imaging period (Fig. 4 H, I, K, and L). KLH45 also caused a stark decline in coordinated events (Fig. 4 J and M). Interestingly, changes in neuronal activity were not accompanied by changes in the number of presynaptic synapsin puncta (Fig. 4 D–E). These results were validated in our second set of APOE3 and isoAPOE4 cell lines (SI Appendix, Fig. S8 I–Q). To rule out the possibility of the decline in neuronal activity being caused by KLH45-induced toxicity, we used confocal imaging to quantify healthy, noncondensed nuclei based on size and DAPI signal intensity. We found no changes in the number of healthy nuclei per field of view (SI Appendix, Fig. S8 E and F). Overall, these findings support our hypothesis that LD metabolism is required to sustain spontaneous network activity in iN cultures, and the decrease in activity driven by blocking LD metabolism is not preceded by detectable structural changes to presynaptic compartments.

Discussion

In this study, we show that microglia drive neuronal activity through secreted factors, including exosomes, and that they specifically enhance network activity under inflammatory conditions. We demonstrate that microglial APOE4 expression further increases neuronal excitability and show that neuronal LD metabolism is essential to sustain spontaneous network activity. These findings provide a mechanistic link between inflammation, neuronal excitability, and lipid metabolism, all of which are implicated in AD.

While CM from resting microglia drives an increase in Ca⁺⁺ spike frequency, CM from LPS-stimulated microglia not only causes a greater increase in spike frequency but also enhances synchronous network activity in iN cultures (Fig. 2). Notably, exosomes isolated from microglial CM recapitulate these signaling bioactivities; while both PBS- and LPS-exosomes increase spike frequency, only LPS-exosomes enhance global network activity (Fig. 1 E–K), positioning exosomes as potential mediators of this microglia-neuron regulation of activity. While previous work has shown that LPS-induced inflammation can increase neuronal activity, most of these studies were done in vivo and in acute brain slices, which contain astrocytes and other cell types that contribute to the inflammatory response (9, 41, 42). Our reductionist approach allowed us to show that signaling from microglia alone can directly alter neuronal activity during LPS-induced inflammation. Furthermore, microglial signaling increases network activity in more mature D48 iN cultures, but not in relatively immature D24 iN cultures, which lack basal network activity, indicating that microglial signaling bioactivity depends on the response capability of recipient cells. Interestingly, prior work on microglia-derived microvesicles (MVs)—which differ from exosomes in cellular origin and cargo—showed that both resting and LPS-stimulated MVs increased neuronal excitability to similar extents (43). A direct comparison of microglial exosomes and MVs, as well as an assessment of exosome- and MV-depleted CM, is needed to clarify the specific contributions of exosomes in microglia-to-neuron signaling.

Our reductionist experimental design also allowed us to investigate the independent contributions of microglial and neuronal APOE genotype to microglial modulation of neuronal activity. Given the known influence of APOE4 genotype on microglial reactivity (22–24, 26) and our observed role of microglia exosome signaling in modulating network activity, we investigated how microglial APOE4 genotype affects this modulation. Resting APOE4 microglia had more LDs than their APOE3 counterparts, and this difference was exacerbated upon LPS-stimulation (SI Appendix, Fig. S3F). This aligns with other reports of APOE4 genotype increasing microglial reactivity and LD accumulation (22, 25, 27, 36, 44). When evaluating CM bioactivity, we found that resting APOE4 microglial CM increases neuronal activity after the 2-d treatment compared to APOE3. We further observed that APOE4 amplifies the inflammatory response, with LPS-treated APOE4 microglia driving a greater increase in network activity than APOE3 (Fig. 2). Thus, our findings suggest that microglial APOE4 can drive neuronal hyperexcitability via intercellular signaling, particularly during inflammation. Notably, a recent study in iPSC-derived microglia found a 7-d treatment with APOE4 resting microglial CM reduced network activity in neurons (27). Given that we observed a striking depletion of LDs accompanying the increased network activity with a 2-d CM treatment and that blocking LD metabolism abolished neuronal activity, we speculate that prolonged CM exposure may deplete neuronal LDs, impairing their ability to sustain network activity. More work is needed to understand the timescale at which microglial CM regulates neuronal activity and the underlying mechanisms that mediates this modulation. Additionally, other microglial activation pathways, including purinergic signaling, are known to also alter neuronal network activity and synchronicity (45). Whether the APOE4 genotype affects microglia–neuron signaling in the context of additional activation pathways merits further investigation.

Neuronal APOE4 expression is associated with neuronal hyperexcitability preceding neurodegeneration in diseases like AD (46). A variety of mechanisms have been implicated in this effect, including neuronal APOE4 driving increased synaptic protein expression and excitatory activity (25) and decreased inhibitory tone in brain regions vulnerable to degeneration (29, 47). Recent work also suggests that neuron-specific APOE genotype drives this excitation–inhibition imbalance (29). Although we show that microglial APOE4 genotype drives an increase in neuronal activity, we did not observe consistent evidence of neuronal APOE genotype affecting excitability. While PBS-treated APOE4 neurons had increased network activity compared to APOE3 neurons (Fig. 1), we did not see the same trend in the control conditions of media exchange (Fig. 2) and KLH45 treatment experiments (Fig. 4). We also observed similar increases in activity in both APOE3 and APOE4 neurons upon treatment with microglial CM (Fig. 2). Although we detected increased synapsin puncta following treatment with APOE4 microglial CM, this was not correlated with increased network activity. Specifically, unlike APOE3 neurons which had no change to presynaptic elements in response to microglial CM, APOE4 neurons increased presynaptic puncta when treated with APOE4 microglial CM. This suggests that neuronal APOE could be affecting the underlying mechanisms that give rise to hyperexcitability in other models. Given that several lines of work show that certain neuron subpopulations are more susceptible to APOE and AD-driven hyperexcitability (46, 47), more work is needed to characterize the extent to which APOE genotype affects neuronal function in models of susceptible neuronal subtypes.

It is well established that neuronal activity is energetically costly, that increases in synaptic activity drive adaptive changes to neuronal metabolism (48, 49) and that hiPSC-derived neuronal models of AD exhibit metabolic dysfunction (25, 50). While the direct contribution of LD metabolism in fueling neuronal activity is unclear, there is a growing body of evidence supporting a critical role for LD metabolism in maintaining neuronal health. Loss-of-function mutations in the neuron-specific TAG lipase DDHD2, for example, cause spastic paraplegia, a neurodegenerative disorder characterized by massive neuronal LD accumulation, limb weakness, and intellectual disability (40, 51, 52). In our NGN2-induced neurons, we found LDs in both D24 and D48 cultures, with older neurons displaying a greater LD load (Fig. 3). Upon treatment with LPS-stimulated microglial CM, which increased neuronal activity, we observed a reduction in LD abundance—suggesting upregulated LD metabolism (Fig. 3). Given the recent reports demonstrating that neuronal activity drives DDHD2-mediated FFA synthesis (53) and that neuronal LD-derived FFAs can be used as oxidative phosphorylation fuel during periods of heightened activity (39), we hypothesized that our iNs might use LDs to fuel both spontaneous and CM-driven activity. We probed LD dynamics by inhibiting the LD lipase DDHD2 using KLH45. In young (D24) iNs, KLH45 treatment led to a significant accumulation of LDs within just 8 h, indicating high metabolic demand underlies ongoing LD turnover even in the absence of network-wide activity (SI Appendix, Fig. S8 A and B). This LD turnover could support membrane remodeling, synapse formation, and low-level activity in these developing neurons, consistent with an association of LD triglyceride metabolism with axon growth following nerve crush (54). In more mature (D48) iNs, which exhibit robust network activity, KLH45 treatment for 24 h led to an accumulation of LDs and concurrent suppression of network activity (Fig. 4 G–M). These data indicate that LDs serve distinct metabolic demands in more mature D48 cultures with active neuronal networks where DDHD2-mediated LD metabolism is required to sustain spontaneous network activity compared to immature rapidly growing D24 iN cultures with little network activity. We propose that neuronal LDs serve as energy reservoirs supporting neuronal network activity and that blocking their metabolism impairs this function. Since neurons are particularly sensitive to oxidative stress from fatty acid catabolism (55), future studies investigating how neuronal LD metabolism is regulated and whether reliance on this pathway contributes to neurodegeneration under conditions of inflammation, APOE4 expression, and heightened energy demand would be informative.

Materials and Methods

All Materials and Methods are described in detail in SI Appendix. Briefly, iPSC-derived microglia-like cells (iMGLs) were generated following the protocol from McQuade and Blurton-Jones (56) and iPSC-derived neurons (iNs) were generated by direct induction with forced expression of NGN2. To induce an inflammatory state, iMGLs were stimulated with 50 ng/mL lipopolysaccharide (LPS), or with vehicle (PBS) as a control, every other day for a week after which CM and exosomes were evaluated for their bioactivity based on iN calcium activity and LD load. Exosomes were isolated from iMGL CM by sequential ultracentrifugation, normalized by protein assay, and administered to iNs in 5μL of PBS. To evaluate iN LD dynamics, iNs were treated with 5 μM KLH45 or vehicle (EtOH) for 24 h, unless otherwise specified. Calcium imaging was performed with Fluo-4AM at 20fps to evaluate neuronal activity in response to iMGL exosomes, CM, and KLH45. To analyze neuronal calcium activity, we first used the analysis pipeline Suite2p (https://github.com/MouseLand/suite2p) to identify cell regions of interest (ROIs) and extract fluorescent signals. Then, features such as spike frequency, spike amplitude, correlated and synchronized network events, and coactive neuron fractions were extracted as described in SI Appendix. Neuronal LDs were labeled with BODIPY-493/503 (1 μg/mL; Invitrogen D3922) and immunocytochemistry was performed as described in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Movie S1.

Neuronal activity in APOE3 iNs after a 48hr treatment with PBS. We used calcium indicator Fluo4-AM and imaged iNs at 20 frames per second. This video is a 225.75μm x 225.75μm cropped region of the original 660.48μm x 660.48μm field of view used for analysis in Fig. 1E. Scale bar is 50μm. Video plays in real time.

Movie S2.

Neuronal activity in APOE3 iNs after a 48hr treatment with PBSexosomes from APOE3 iMGLs. We used calcium indicator Fluo4-AM and imaged iNs at 20 frames per second. This video is a 225.75μm x 225.75μm cropped region of the original 660.48μm x 660.48μm field of view used for analysis in Fig. 1E. Scale bar is 50μm. Video plays in real time.

Movie S3.

Neuronal activity in APOE3 iNs after a 48hr treatment with LPSexosomes from APOE3 iMGLs. We used calcium indicator Fluo4-AM and imaged iNs at 20 frames per second. This video is a 225.75μm x 225.75μm cropped region of the original 660.48μm x 660.48μm field of view used for analysis in Fig. 1E. Scale bar is 50μm. Video plays in real time.

Movie S4.

Neuronal activity in APOE3 iNs at D45 generated with Classic NGN2 Induction. We used calcium indicator Fluo4-AM and imaged iNs at 20 frames per second. This video is a 225.75μm x 225.75μm cropped region of the original 660.48μm x 660.48μm field of view. Scale bar is 50μm. Video plays in real time.

Data, Materials, and Software Availability

All study data are included in the article, deposited in DANDI Archive (RRID: SCR-017571) (57) and/or supporting information.