Microglia-synapse engulfment via PtdSer-TREM2 ameliorates neuronal hyperactivity in Alzheimer's disease models

Abstract

Neuronal hyperactivity is a key feature of early stages of Alzheimer's disease (AD). Genetic studies in AD support that microglia act as potential cellular drivers of disease risk, but the molecular determinants of microglia-synapse engulfment associated with neuronal hyperactivity in AD are unclear. Here, using super-resolution microscopy, 3D-live imaging of co-cultures, and in vivo imaging of lipids in genetic models, we found that spines become hyperactive upon Aβ oligomer stimulation and externalize phosphatidylserine (ePtdSer), a canonical "eat-me" signal. These apoptotic-like spines are targeted by microglia for engulfment via TREM2 leading to amelioration of Aβ oligomer-induced synaptic hyperactivity. We also show the in vivo relevance of ePtdSer-TREM2 signaling in microglia-synapse engulfment in the hAPP NL-F knock-in mouse model of AD. Higher levels of apoptotic-like synapses in mice as well as humans that carry TREM2 loss-of-function variants were also observed. Our work supports that microglia remove hyperactive ePtdSer⁺ synapses in Aβ-relevant context and suggest a potential beneficial role for microglia in the earliest stages of AD.

Keywords: Abeta oligomers; Alzheimer's disease; microglia; pruning; synapses.

Figure 1. Microglia target AD synapses for engulfment via ePtdSer

1. A

   Area under curve (AUC) for pHrodo fluorescence signals in primary microglia at t = 3 h, normalized to respective control, post‐application of synaptosomes (SN) from either AD patients or NDC brains, with and without Annexin‐V (AnnxV) pretreatment. Data are normalized to NDC SN. ∼40 microglia per ROI, two ROIs per well, 2–3 wells per experiment, n = 6 human cases for NDC and AD, n = 3 independent experiments.

2. B

   Representative image of primary microglia simultaneously treated with Aβ oligomer‐bound SN (oAβ‐SN; in pHrodo red [magenta]) and control SN (Ctrl‐SN; in pHrodo deep red [cyan]). Scale bar, 50 μm.

3. C

   pHrodo fluorescence (a.u.) over 10 h (3–5‐min intervals, SNs added to microglia at t = 0) showing a faster rate of increase of oAβ‐SN compared with Ctrl‐SN. Data are normalized to Ctrl‐SN. ∼40 microglia per ROI, two ROIs per well, 2–3 wells per experiment, n = 4 independent experiments.

4. D

   Percentage of pHrodo fluorescence of either oAβ‐SN or Ctrl‐SN in microglia at t = 3 h. ∼40 microglia per ROI, two ROIs per well, 2–3 wells per experiment, n = 3 independent experiments.

5. E

   pHrodo fluorescence with time shown as AUC at 3 h normalized to respective control. AUC of engulfed mouse oAβ‐SN versus Ctrl‐SN with and without AnnxV pretreatment.

6. F

   Time‐lapse images of primary Homer1‐eGFP neurons (green) treated with 50 nM oAβ versus vehicle control. Yellow arrows indicate increasing PSVue550 (magenta) signal on dendritic spines with Aβ oligomer treatment over 45 min. Scale bar, 2 μm.

7. G

   Relative fold change of colocalized PSVue and Homer1‐eGFP signal over time. Two–three ROIs per experiment, n = 3 independent experiments.

8. H

   Orthogonal view of AiryScan image showing colocalization of Homer1‐eGFP (green) and PSVue (magenta) at 1‐h post‐treatment with Aβ oligomer. Scale bar, 2 μm.

9. I

   Quantification of colocalized Homer1‐eGFP with PSVue among total Homer1. Three ROIs per neuron, 4–5 neurons, n = 3 independent experiments.

10. J

    Quantification of colocalized PSVue with Homer1‐eGFP among total PSVue. Three ROIs per neuron, 4–5 neurons, n = 3 independent experiments.

Data information: Data shown as mean ± SEM (A–E) or as mean ± SD (F–J). Each shaded point represents one ROI, and each open point represents the mean of each independent experiment (A, D, E, I, J). Filled point (G) represents the average of all ROIs. One‐way (A and E) or two‐way (G) ANOVA followed by Bonferroni post hoc test, paired (D) or unpaired (I and J) t‐test. P‐values shown as *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Source data are available online for this figure.

Figure EV1. Annexin‐V pretreatment of AD synaptosomes decreases microglial engulfment

1. A

   Crude synaptosomes were prepared using a protocol that yields electrically functional synaptosomes for several hours post‐isolation, which can be depolarized and stimulated with KCl and NMDA, respectively. Electron microscopy of crude synaptosome preparation showing intact pre‐ (**) and postsynaptic sites (*). Scale bar 500 nm.

2. B

   Western blot showing enrichment of synaptic markers PSD‐95 (green, 95 kDa) in synaptosomes (SN) compared with total homogenate (THF) fractions with respect to GAPDH (red, 37 kDa) loading control. One lane represents one mouse.

3. C

   Primary microglia grown with TGFβ express high mRNA levels of microglial genes including Cx3cr1, Itgam, Trem2 but also homeostatic Tmem119 with little contamination from neurons (Map2), astrocytes (Gfap), and oligodendrocytes (Mag). Gene expression normalized to the geomean of three housekeeping genes (Actb, Gapdh, and Rpl32).

4. D

   Microglia treated with NDC (NDC SN), AD (AD SN), or AD synaptosomes pretreated with 10 μg/ml Annexin‐V (AD SN + AnnxV) conjugated to pHrodo red (magenta). Scale bar 50 μm.

5. E

   All isolated human synaptosomes (SN) used in this study have been tested for the presence of Aβ oligomers by western blotting. Exemplar western blot showing higher levels of Aβ trimer (14 kDa), dimer (6.5 kDa), and monomer (3 kDa) in synaptosomes prepared from the frontal cortex of AD (AD SN) patients and NDC (NDC SN) with respect to GAPDH loading control (38 kDa) loading control. One lane represents one patient.

Data information: Data shown as mean ± SEM. Each point represents the average per experimental replicate.

Figure EV2. Aβ oligomers induce focal PtdSer externalization in dendritic spines

1. A

   Representative image of Homer1‐eGFP neuron used in our studies here (green, left panel). In contrast to an apoptotic neuron (magenta, middle panel), where PSVue signals are observed in whole parts of the dying cell, the neurons used in our studies are nonapoptotic with negligent PSVue signal in the soma (right panel). Scale bar 10 μm.

2. B

   Super‐resolution Airyscan images of neuronal PSvue (magenta) staining after 1‐h treatment of 50 nM Aβ oligomer. Note PSVue signals in insets 1 and 2 are not random but are colocalized with or in close vicinity to Homer1‐eGFP signal as indicated by yellow arrowheads. Scale bar 10, 2, and 5 μm.

3. C

   Super‐resolution Airyscan images of PSD95 (cyan), PSVue (magenta) and 6E10 anti‐Aβ antibody (yellow) after 1‐h treatment with Aβ oligomer. Yellow arrowheads indicate the spines showing colocalized signal of PSD95, PSVue, and 6E10. Magnified orthogonal view of a single spine with PSD95, PSVue, and 6E10 colocalization in the inset. Scale bar 2 and 0.5 μm.

Figure 2. Microglia selectively engulf hyperactive ePtdSer⁺ spines and ameliorate Aβ oligomer‐induced neuronal hyperactivity

1. A

   Time‐lapse images of microglia (cyan, labeled with IB4‐647, 3D rendered) internalizing (yellow arrowheads) PSVue⁺ Homer1‐eGFP dendritic spines. Note PSVue⁻ Homer1‐eGFP dendritic spines are left behind (empty yellow arrowheads). Scale bar, 5 μm.

2. B

   Percentage of PSVue⁺ or PSVue⁻ Homer1 mobilized by microglial processes within 10 min of recording. Fifteen to thirty ROIs per experiment, three independent experiments.

3. C

   SRM images of Homer1‐eGFP and CD68 (magenta) lysosomes in microglia. Inset shows orthogonal view of colocalized Homer1‐eGFP and CD68. Scale bar, 5 μm.

4. D

   Representative image of transfected GCaMP7 signal on dendritic spines at 48‐h post‐Aβ oligomer challenge labeled with PSVue, closed and open arrows point to PSVue⁺ and PSVue⁻ spines, respectively. Scale bar, 2 μm.

5. E

   Representative trace of transfected postsynaptic GCaMP7 signals from PSVue⁺ and PSVue⁻ spines at 48‐h post‐treatment with Aβ.

6. F

   Number of spontaneous calcium transients per min in PSVue⁺ and PSVue⁻ spines. ∼10 spines per neuron, ∼5 neurons per experiment, from three independent experiments.

7. G

   (Top panel) Representative traces of spontaneous GCaMP7 signals at 48‐h post‐treatment with Aβ oligomer in dendritic spines from neuron‐only culture (cyan), neuron–microglia co‐culture (magenta), and neuron–microglia co‐culture treated with AnnxV (yellow). (Bottom panel) number of calcium transients per minute before (0 h), at 24‐h, and 48‐h post‐treatment of Aβ oligomer. ∼10 spines per neuron, ∼10 neurons per experiment, from three independent experiments.

8. H

   Percentage of spines showing high frequency (> mean + 2 SD; cyan, magenta, and yellow) compared with low frequency (< mean + 2 SD; gray) of calcium transients in each experimental group.

9. I

   SRM images of neuron‐only, neuron–microglia co‐culture, and neuron–microglia co‐culture treated with Annexin‐V (AnnxV) labeled with PSVue (magenta) 48 h after 50 nM Aβ oligomer treatment; neurons are transfected with GCaMP7‐eGFP (green). Scale bar, 10 μm.

10. J, K

    Number of PSVue puncta per 100 μm² (J) and mean intensities of PSVue signal (K) in neuron‐only culture, and neuron–microglia co‐culture with or without AnnxV. Sixteen ROIs per experiment, n = 3 independent experiments.

Data information: Data shown as mean (B), median (F and G) and frequency distribution (H) or box plots (J and K). The top and the bottom of the boxes represent the 75^th and 25^th percentiles, respectively, and the lines in the middle represent the median. The whiskers represent the highest and lowest values that are not outliers. Central bands of the violin plot represent median and quartiles. Each shaded point represents one ROI, and each open point represents the mean (or median for GCaMP studies) of each independent experiment. Paired t‐test (B) Mann–Whitney test (F), Kruskal–Wallis test followed by Dunn's multiple comparisons test (G) or one‐way ANOVA followed by Bonferroni post hoc test (J and K). P‐values shown ns P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Source data are available online for this figure.

Figure EV3. Primary microglia contact and remove ePtdSer⁺ dendritic spines to resolve neuronal hyperactivity upon acute Aβ oligomer challenge

1. A

   Time‐lapse sequence images of 10 consecutive optical sections (Δz = 0.4 μm) showing primary Homer1‐eGFP neurons (green), PSVue (magenta) and microglia (yellow, labeled with Isolectin B4‐647, IB4‐647). Yellow arrowheads indicate PSVue⁺ Homer1‐eGFP dendritic spines contacted by microglia over 1 h. See Movie EV2. Scale bar 2 μm.

2. B

   Representative image of transfected GCaMP7 signal on dendritic spines at 48‐h post‐oAβ challenge. Scale bar 2 μm.

3. C

   (Top panel) Representative traces of GCaMP7 signal in dendritic spines at 0‐h and 24‐h post‐treatment with Aβ oligomer from neuron‐only culture (cyan), neuron–microglia co‐culture (magenta), and neuron–microglia co‐culture treated with Annexin‐V (AnnxV, yellow). (Bottom panel) Violin plots showing normalized peak intensities (ΔF/F ₀) of spontaneous GCaMP7 signal in dendritic spines before (0 h), at 24‐h and 48‐h post‐treatment with Aβ oligomer. (Bottom right) Relative fold change. ∼10 spines per neuron, ∼10 neurons per experiment, from three independent experiments.

Data information: Data shown as violin plot, central bands represent median and quartiles. Each shaded point represents one ROI, and each open point represents the median of each independent experiment. Kruskal–Wallis test followed by Dunn's multiple comparisons test. P‐values shown as ns P > 0.05; *P < 0.05; **P < 0.01; ****P < 0.0001.

Figure 3. Microglia require TREM2 to engulf synapses in in vitro and in vivo Aβ models

1. A

   Trem2 CV and Trem2 R47H KI primary microglia treated simultaneously with oAβ‐bound synaptosomes (SN) conjugated with pHrodo red (magenta) and control SN, conjugated with pHrodo deep red (cyan). Scale bar, 50 μm.

2. B

   pHrodo fluorescence with time shown as AUC at 3 h. AUC of oAβ‐SN is higher compared with Ctrl‐SN in Trem2 CV but not in Trem2 R47H KI microglia. ∼40 microglia per ROI, two ROIs per well, 2–3 wells per experiment, n = 3 independent experiments.

3. C

   Quantification of the percent of C1qa area covered normalized to WT showing an increase in the NL‐F but not NL‐F KI; Trem2 R47H KI. N = 3–4 animals per genotype.

4. D

   Representative images from hippocampal CA1 from 6‐month‐old WT, Trem2 R47H KI, NL‐F KI, and NL‐F KI; Trem2 R47H KI mice probing for C1qa (cyan) by in situ hybridization (RNAScope) followed by Iba1 (red) immunostaining. Scale bar, 20 μm.

5. E–G

   Hippocampal CA1 SR of 6 mo WT, Trem2 R47H KI, NL‐F KI, and NL‐F KI; Trem2 R47H KI mice immunostained for P2Y12 (red), CD68 (magenta), and Homer1 (green). (E) Orthogonal image showing engulfed Homer1 (green) inside CD68 (magenta)‐immunoreactive lysosomes in P2Y12⁺ (red) microglia. Scale bar, 2 μm. (F) Quantification of engulfment index (volume of Homer1 in CD68/microglial cell volume)*100). Six to nine microglia per animal, n = 6 animals per genotype. (G) Representative 3D surface rendering reconstructions showing increased Homer1 in microglia in NL‐F KI compared with WT, Trem2 R47H KI, and NL‐F KI; Trem2 R47H KI mice. Scale bar, 10 μm.

6. H

   Colocalized puncta density normalized to WT or Trem2 R47H KI accordingly showing decreased synapse density in NL‐F KI but not NL‐F KI; Trem2 R47H KI. Three ROIs per animal, n = 6 animals per genotype.

7. I

   SRM images from the hippocampal CA1 SR of 6 mo WT, Trem2 R47H KI, NL‐F KI and NL‐F KI; Trem2 R47H KI mice immunostained for Synaptotagmin (Syt1/2, magenta) and Homer1 (green), pre‐ and postsynaptic puncta, respectively. Scale bar, 1 μm.

Data information: Data shown as mean ± SEM. Each shaded point represents one ROI, and each open point represents the mean of each independent experiment. Two‐way ANOVA followed by Bonferroni's post hoc test. P‐values shown as ns P > 0.05; *P < 0.05; ***P < 0.001; ****P < 0.0001. Source data are available online for this figure.

Figure EV4. Microglia require functional TREM2 to engulf synapses in vivo

1. A

   Primary microglia (P0‐P4) prepared from either Trem2 CV or Trem2 R47H KI mice treated with dextran conjugated to pHrodo red to compare engulfment of inert particles, showing no difference between the two genotypes.

2. B

   Representative images from hippocampal CA1 stratum lacunosum‐moleculare from 6‐month‐old WT, Trem2 R47H KI, NL‐F KI, and NL‐F KI; Trem2 R47H KI mice probing for Tmem119 (magenta) and Clec7a (green) by in situ hybridization (RNAScope). DAPI shown in blue. Scale bar, 20 μm.

3. C

   Quantification of the percent of area covered either by Tmem119 or Clec7a spots showing higher levels of Tmem119 compared with Clec7a across all genotypes with no difference between genotypes. n = 2–4 animals.

4. D–G

   Hippocampal CA1 stratum radiatum sections from 6‐month‐old WT, Trem2 R47H KI, NL‐F KI, and NL‐F KI; Trem2 R47H KI mice immunostained for P2Y12 (red), CD68 (magenta), and Homer1 (green). 3D surface rendering reconstructions of microglia showing increased Homer1 signal inside microglia in NL‐F KI but not NL‐F KI; Trem2 R47H KI compared with WT and Trem2 R47H KI, respectively. Quantification of microglial P2Y12⁺ cell volume (D), number of CD68⁺ lysosomal vesicles per microglia (E), percentage of P2Y12⁺ volume occupied by CD68⁺ immunoreactive vesicles (F), and total volume of Homer1⁺ material within P2Y12⁺ microglia (G). All volumes are represented in μm³. Six to nine microglia per animal, n = 6 animals per genotype.

Data information: Data shown as mean ± SEM. Each shaded point represents one ROI, and each open point represents the average per experimental replicate. Unpaired t‐test (A) or two‐way ANOVA followed by Bonferroni's post hoc test. P‐values shown as ns P > 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Figure 4. TREM2 loss‐of‐function leads to increased levels of apoptotic‐like synapses in mouse and human brains

1. A

   Hippocampal dentate gyrus hilus of 6 mo NL‐F KI and NL‐F KI; Trem2 R47H KI mice ICV injected with PSVue 643 (yellow) and immunostained for Synaptotagmin 1/2 (magenta) and Homer1 (green). Scale bar, 1 μm.

2. B

   Percentage of synaptic Homer1‐immunoreactive puncta found colocalized with PSVue in NL‐F KI; Trem2 R47H compared with NL‐F KI. Three ROIs per animal, n = 3–4 animals per genotype.

3. C

   Orthogonal SRM image showing colocalized PSVue (yellow) with synaptic markers, Syt1/2 (magenta) and Homer1 (green). Scale bar 200 nm.

4. D

   Number of spontaneous calcium transients per minute at 48‐h post‐treatment of Aβ oligomer in neuron‐only culture (cyan), neuron‐Trem2 CV microglia co‐culture (magenta) or neuron‐ Trem2 R47H KI microglia co‐culture (purple). Approximately ten spines per neuron, ∼10 neurons per experiment, from 3 to 4 independent experiments.

5. E

   Quantification of Bassoon‐immunoreactive puncta density in four different genotypes. Six ROIs per animal, n = 4 animals per genotype.

6. F

   SRM 3D images from the hippocampal CA1 SR of 6 mo WT, Trem2 R47H KI, NL‐F KI, and NL‐F KI; Trem2 R47H KI mice immunostained for Bassoon (cyan), a presynaptic marker enriched in functional active zones. Scale bar, 1 μm.

7. G

   Representative western blots comparing levels of cleaved caspase‐3 (17/19 kDa) and procaspase‐3 (35 kDa) with respect to GAPDH (38 kDa) loading control in synaptosomes isolated from human NDC, AD, NDC TREM2 and AD TREM2 brains. One lane represents one patient.

8. H

   Western blot densitometry analysis showing cleaved caspase‐3 levels (left graph) and the ratio of cleaved caspase‐3/procaspase‐3 (right graph) in synaptosomes (SN) isolated from human NDC, AD, TREM2 NDC, and TREM2 variants. N = 9 NDC, 11 AD, 5 TREM2 NDC, 10 TREM2 AD cases.

Data information: Data shown as mean ± SEM. Each shaded point represents one ROI, and each open point represents the mean (or median for GCaMP studies) of each independent experiment. Central bands of the violin plot (D) represent median and quartiles. The top and the bottom of the box plot (H) represent the 75^th and 25^th percentiles, respectively, and the line represents the median. The whiskers represent the highest and lowest values that are not outliers. Unpaired t‐test (B), Kruskal–Wallis test followed by Dunn's test (D), two‐way (E), or one‐way (H) ANOVA followed by Bonferroni's post hoc test. P‐values shown as ns P > 0.05; *P < 0.05; **P < 0.01, ****P < 0.0001. Source data are available online for this figure.

Figure EV5. Trem2 loss‐of‐function exacerbates synaptic ePtdSer in the NL‐F model of amyloidosis

1. A

   Super‐resolution images from the hippocampal CA1 stratum radiatum of 6‐month‐old NL‐F KI and NL‐F KI; Trem2 R47H KI mice immunostained for presynaptic Synaptotagmin 1/2 (Syt1/2, red) and postsynaptic Homer1 (green). PSVue 643 (magenta) is ICV injected. Upper panels show Syt1/2, Homer1, and PSVue. Lower panels show Homer1 and PSVue only. Scale bar 1 μm.

2. B

   PSVue volume represented in μm³. Three ROIs per animal, n = 3–4 per genotype.

3. C

   Percentage of synaptic Synaptotagmin 1/2 puncta within 0.25 μm of PSVue showing increased percentage of PSVue⁺ synapses in NL‐F KI; Trem2 R47H KI mice compared with NL‐F KI. Three ROIs per animal, n = 3–4 animals per genotype.

4. D

   Percentage of PSVue volume within 0.25 μm of synaptic Synaptotagmin 1/2 puncta. Three ROIs per animal, n = 3–4 animals per genotype.

5. E

   Super‐resolution images from the hippocampal CA1 dentate gyrus hilus of 4‐month‐old WT and J20 Tg mice immunostained for pre‐Synaptotagmin 1/2 (Syt1/2, red) and postsynaptic Homer1 (green). PSVue 643 (magenta) is ICV injected. Top panels show Syt1/2, Homer1, and PSVue. Bottom panels show Homer1 and PSVue only. Insets show either triple (Syt1/2, Homer1 and PSVue; top panel) or double colocalization (Homer1 and PSVue; bottom panel). Scale bar 1 μm.

6. F

   PSVue total volume per ROI (7,500 μm³) showing increased PSVue in J20 Tg mice compared with WT. Three ROIs per animal, n = 3–4 animals per genotype.

7. G

   Percentage of synaptic Homer1 puncta within 0.25 μm of PSVue showing an increase in PSVue⁺ synapses in J20 Tg compared with WT. Three ROIs per animal, n = 3–4 animals per genotype.

8. H

   Percentage of synaptic Synaptotagmin 1/2 puncta within 0.25 μm of PSVue showing an increase in PSVue⁺ synapses in J20 Tg compared with WT. Three ROIs per animal, n = 3–4 animals per genotype.

9. I

   Percentage of PSVue volume within 0.25 μm of synaptic Homer1 puncta. Three ROIs per animal, n = 3–4 animals per genotype.

Data information: Data shown as mean ± SEM. Each shaded point represents one ROI, and each open point represents the average per experimental replicate. Two‐way ANOVA followed by Bonferroni's post hoc test. P‐values shown as ns P > 0.05; *P < 0.05; **P < 0.01.