Mesenchymal-Derived Extracellular Vesicles Enhance Microglia-mediated Synapse Remodeling after Cortical Injury in Rhesus Monkeys

Understanding the microglial neuro-immune interactions in the primate brain is vital to developing therapeutics for cortical injury, such as stroke. Our previous work showed that mesenchymal-derived extracellular vesicles (MSC-EVs) enhanced motor recovery in aged rhesus monkeys post-injury of primary motor cortex (M1), by promoting homeostatic ramified microglia, reducing injury-related neuronal hyperexcitability, and enhancing synaptic plasticity in perilesional cortices. The current study addresses how these injury- and recovery-associated changes relate to structural and molecular interactions between microglia and neuronal synapses. Using multi-labeling immunohistochemistry, high resolution microscopy, and gene expression analysis, we quantified co-expression of synaptic markers (VGLUTs, GLURs, VGAT, GABARs), microglia markers (Iba-1, P2RY12), and C1q, a complement pathway protein for microglia-mediated synapse phagocytosis, in perilesional M1 and premotor cortices (PMC) of monkeys with intravenous infusions of either vehicle (veh) or EVs post-injury. We compared this lesion cohort to aged-matched non-lesion controls. Our findings revealed a lesion-related loss of excitatory synapses in perilesional areas, which was ameliorated by EV treatment. Further, we found region-dependent effects of EV on microglia and C1q expression. In perilesional M1, EV treatment and enhanced functional recovery were associated with increased expression of C1q + hypertrophic microglia, which are thought to have a role in debris-clearance and anti-inflammatory functions. In PMC, EV treatment was associated with decreased C1q + synaptic tagging and microglial-spine contacts. Our results provided evidence that EV treatment facilitated synaptic plasticity by enhancing clearance of acute damage in perilesional M1, and thereby preventing chronic inflammation and excessive synaptic loss in PMC. These mechanisms may act to preserve synaptic cortical motor networks and a balanced normative M1/PMC synaptic connectivity to support functional recovery after injury.


INTRODUCTION
Cortical injury, such as stroke in humans, causes neuronal damage and loss of synaptic connections, leading to signi cant cognitive and behavioral impairments [1,2]. Brain plasticity enables functional recovery after cortical injury, which is modulated by neuro-in ammatory responses [3]. As the immune cells of the brain, microglia can promote neuronal plasticity and recovery via phagocytosis of damaged pre-and post-synaptic elements, and release of neurotrophic factors to facilitate synapse turn-over [4]. On the other hand, chronic in ammatory activity of microglia exacerbates neuronal damage and prevents recovery [5]. However, once in ammation subsides, microglia can secrete anti-in ammatory cytokines such as IL-10 and TGF-β that can promote neuronal plasticity and repair [5,6]. Thus, post-injury recovery is dependent on the interplay and balance of in ammatory responses and facilitation of neuronal synaptic plasticity, which is the key to developing effective therapeutics, but not well understood especially in the primate brain. cortex (M1). Two weeks after the surgery, monkeys then began testing on the motor tasks for 12 weeks to assess the degree and nature of recovery. A separate cohort of aged-matched non-lesion monkeys (n = 3) which were part of a larger study on aging, was used as non-lesion control comparisons.

Surgical lesion of M1 hand representation and postinjury treatment
Surgical procedures to induce cortical injury were employed in the lesion cohort, as described [12,14].
Brie y, each monkey was sedated with ketamine (10 mg/kg) and anesthetized with intravenous sodium pentobarbital (15-25 mg/kg). To limit the surgical lesion to the hand representation of M1 on the hemisphere controlling the dominant hand, the precentral gyrus was electrophysiologically mapped using a small silver ball surface stimulating electrode (with minimum current stimulus amplitude: 1-3 mA) [12,14]. While the stimulating electrode was moved across the precentral gyrus, a trained observer recorded and graded the evoked muscle movements of the hand to create a cortical surface map of the hand area, as described [12]. To induce the lesion, a small glass suction pipette was inserted through an incision in the pia at the dorsal limit of the hand area, and then bluntly separated the penetrating arterioles from the underlying cortex.
At 24 hours and 14 days after the injury, monkeys were treated with either vehicle control or EV intravenously. The EVs were extracted from MSCs harvested from the bone marrow of a young adult monkey, as described [12,14]. For bone marrow extraction, the monkey was sedated with ketamine (10 mg/kg) and anesthetized with intravenous sodium pentobarbital. Bone marrow extracted from the iliac crest was shipped to Henry Ford Health Systems, where the MSCs were isolated and cultured in vitro for EV collection, as described [9,10,12,13]. EVs were then shipped back to Boston University for intravenous administration in monkeys (EVs were administered at 4 × 10 11 particles/kg). At 14 to 16 weeks after the injury, monkeys were sedated with ketamine (10 mg/kg), anesthetized with intravenous sodium pentobarbital (15-25 mg/kg) for perfusion and brain harvesting. Monkeys were perfused by using a two-stage Krebs-PFA perfusion method as described [14]. Ice-cold Krebs-Heinsleit buffer (6.4mM Na 2 HPO 4 , 1.4mM Na 2 PO 4 , 137mM NaCl, 2.7mM KCl, 5mM glucose, 0.3mM CaCl 2 , 1mM MgCl 2 , pH7.4) was used for perfusion and collecting fresh tissue biopsies. A tissue block just ventral to the lesion, containing mainly ventral premotor cortex (vPMC), was harvested from each monkey and then transferred to oxygenated (95% O 2 , 5% CO 2 ) ice-cold Ringer's solution (26mM NaHCO 3 , 124mM NaCl, 2mM KCl, 3mM KH 2 PO 4 , 10mM glucose, 1.3mM MgCl 2 , pH 7.4), as described [14]. This vPMC tissue block was sectioned into 300 µm coronal acute slices with a vibratome for in vitro whole-cell patch clamp recording and intracellular lling [14,17].

Perfusion and brain tissue preparation
The rest of the brain was xed with 4L of 4% paraformaldehyde (30°C, pH7.4) and blocked in situ in the coronal plane [13]. The whole brain was then removed from the skull and cryoprotected in 0.1M phosphate buffer (PB) with 10% glycerol, and 2% DMSO, and then in buffer with 2% DMSO and 20% glycerol [13]. Brains were ash-frozen in -75°C isopentane and stored at -80°C before being cut on a microtome into interrupted series containing eight series of 30µm sections and one series of 60µm sections [13]. Sections were kept in the phosphate buffer with 15% glycerol and stored in -80°C for later processing.
All the sections were incubated in 50mM Glycine for 1 hour and were washed with 0.01M PBS. Antigen retrieval was performed via incubation in 10mM citrate buffer (pH = 8.5) at 60-65°C for 20 minutes.

Intracellular lling during in-vitro whole cell patch clamp recording and immuno-staining of microglia
To study microglia-spine interactions, we used pyramidal neurons intracellularly lled from whole-cell patch clamp recordings conducted for our previous study [14]. During Krebs perfusion, a fresh tissue block was harvested from vPMC and cut into 300-µm-thick coronal brain slices using a vibratome, which were then placed into room temperature oxygenated (95% O2, 5% CO2) Ringer's solution. After 1 hour of equilibration period, individual brain slice was placed in submersion-type recording chambers (Harvard Apparatus), mounted on the stages of Nikon E600 infrared-differential interference contrast microscopes (Micro Video Instruments). Then, in-vitro whole cell patch clamp experiments and intracellular lling were performed on the layer 3 (L3) pyramidal neurons in the perilesional region at room temperature, as described [14,17], to obtain electrophysiological data for our previous study [14]. Electrodes were fabricated on a horizontal Flaming and Brown micropipette puller (model P-87, Sutter Instruments) [17]. Potassium methanesulfonate-based solution (concentrations in mM: 122 KCH3SO3, 2 MgCl2, 5 EGTA, 10 Na-HEPES, pH 7.4; Sigma-Aldrich) with 1% biocytin was used as internal solution in electrodes (resistances of 3-6 MΩ) to ll the pyramidal neurons. After recording, slices were xed in 4% PFA for 2 days. In order to visualize the cells lled with biocytin, slices were incubated in 1% Triton-X in 0.1M PB for 2 hours at room temperature, and in streptavidin-Alexa 488 for 2 days (1:500 in 0.1 M PBS, Invitrogen). After recording, slices were xed in 4% PFA for immunolabeling experiments.
2.6 Confocal imaging and quanti cation of synaptic markers All 30 or 60 µm sections were imaged by a Leica TCS SPE laser scanning confocal microscope with 3 laser lines: 488 nm, 546 nm, and 647 nm (Leica Microsystems). Sections were imaged with a 40x 1.3 N.A. oil objective lens at a resolution of 0.134*0.134*0.5µm. We imaged four elds, spaced 400µm apart, directly underlying the damaged pial surface (from ~ 200 µm to ~ 1200 µm distal to the damaged pial surface) in M1 gray matter, and two elds in PMC gray matter layer 2/3 with intact pial surface ~ 2mm distal to the lesion in each brain section (Fig. 1b). Each confocal image was deconvolved using AutoQuant (Media Cybernetics) and converted to 8-bit images for further analysis.
To quantify the synaptic markers, we analyzed the optical density (percent area labeled) and size of immunolabeled pre-and postsynaptic markers using particle analysis function in FIJI/ImageJ (https://imagej.net/Fiji; 1997-2016; RRID:SCR_002285) [19]. The signal threshold for analysis was set with the Renyi method of FIJI in the rst eld (layer 1/2) and applied to all the other elds of the same section. The physical contacts, or colocalization, between synaptic markers (VGLUT1 & 2, VGAT) and microglial markers (P2RY12 & Iba-1) or complementary marker C1q were analyzed by using the colocalization plugin of FIJI/ImageJ. The percent area of co-locolization was rst obtained and then a colocalization coe cient was calculated based on Mander's method (the percent area colocalized/percent area of marker 1 or maker 2). The average measures of synaptic puncta optical density and microglia-synapse colocalization were calculated for each animal and compared between groups.

Quanti cation of microglia-neuron interactions
We assessed the interaction between lled L3 pyramidal neurons and immuno-labeled microglia in the perilesional ventral premotor cortex. Dual channel imaging was conducted using a Leica TCS SPE laser scanning confocal microscope with 488 nm and 546 nm laser under 63x/1.4 N.A. oil objective lens at a resolution of 0.04*0.04*0.3µm. One apical dendrite and one basal dendrite of each lled cell were followed and scanned from base to tip. Confocal scanned images were montaged, and the dendritic segments were traced and reconstructed in Neurolucida 360 (RRID: SCR_016788; MBF Bioscience). The appositions of microglia (P2RY12/Iba-1+) on dendrites and dendritic spines were counted and categorized as contacts and neighboring. Contacts required overlap of saturated signal from the two channels, while neighboring was identi ed when the signal of two channels were adjacent and the distance was ~ 0.3µm to ~ 1µm. The traced dendrites with markers of microglial apposition were analyzed and exported using Neuroexplorer (v11.01, Microbright eld). The density of appositions (# of appositions/total length of dendrite imaged) was calculated for each dendrite.

Microglia classi cation and reconstruction
We used the NeuroLucida 360 software (RRID: SCR_016788; MBF Bioscience) to classify and reconstruct the microglia. We examined each eld through the entire Z-stack and counted the microglia by marking them based on their morphological phenotypes [13,22]. Speci cally, microglia were classi ed into three categories based on morphology: 1) rami ed microglia were classi ed based on the appearance of a round soma, and thin multipolar primary process. 2) Hypertrophic I microglia were classi ed characterized by slightly enlarged (about 1.5x larger than rami ed) and ovoid somata with slightly thickened processes (about 2x thicker than rami ed). 3) The amoeboid/hypertrophic II had either large somata (> 2x more than rami ed) with very thick and short process, or with highly elongated somata with thick processes almost as thick as the somata diameter. The microglia identi ed by Iba-1 staining and classi ed by morphology, were then classi ed based on C1q expression, thus allowing us to quantify six categories of microglial phenotypes as follows: rami ed (Rami) C1q negative cells (R-), rami ed cells with C1q colocalized in the soma or processes (R+), hypertrophic (Hyper) C1q negative cells (H-), hypertrophic cells with C1q colocalized in the soma or processes (H+), amoeboid (Ame) C1q negative cells (A-) and amoeboid cells with C1q colocalized in the soma (As+).
We then reconstructed the soma and primary process of a subset of the classi ed microglia. Somata were reconstructed in NeuroLucida 360 (Microbright eld, Inc.), using the 3D environment and soma autodetection features. The soma detector sensitivity was maintained between 70-90, the interactive search region ranged between 20-30 µm, and the size constraint was between 2-5 µm. For the somas that were not able to be auto-detected using the above measurements, we manually contoured them by using the Cell Body trace feature in the software and focusing through the Z-stack. In addition, all the primary processes of the microglia were manually contoured using the Dendrite feature since we wanted their thickness to be accurately determined. Once all the microglia were classi ed and reconstructed, we exported the data for microglia using the NeuroLucida Explorer (v11.01, Microbright eld).

RNA isolation and qPCR
Ventral perilesional brain tissue containing the caudal PMC/M1 area of each animal was dissected at euthanasia (14 weeks post-injury), ash-frozen using dry ice, then stored at -80•C until RNA isolation as described in our previous work [13]. Brie y, tissue samples were thawed on dry ice and dissected into 100 mg pieces for each animal before mechanical homogenization using an RNAse free scalpel. Then, tissue samples were chemically triturated using the TRIzol method as follows (ThermoFisher, Waltham, MA); brie y, tissue was placed in TRIzol and passed through an 18-gauge needle to further homogenize tissue. An organic extraction was then performed using chloroform and ethanol, according to the manufacturer's protocol (ThermoFisher, Waltham, MA). The extracted RNA was then air-dried and resuspended in 40 µL of PCR-grade water. RNA purity was checked using UV absorbance ratio at A260/280 with a NanoDrop spectrophotometer (ThermoFisher, Waltham, MA).

Statistical analysis
All data were expressed as box-and-whisker plots and linear plots with means. Statistical analyses were processed in MATLAB (R2020a, MathWorks, Natick, MA) to calculate the average of each eld and each animal. The outcome measures of each synaptic marker and colocalization were compared between groups using one-way ANOVA, with post hoc Fisher's LSD in MATLAB. For pairwise comparison of groups with lesion, Student's t tests were performed for measures of synaptic markers, colocalization, microgliaspine contacts, and qPCR results. Linear correlations between variables were determined using linear regression analyses in MATLAB. Nonmetric multidimensional scaling (NMDS) was performed to analyze similarities among three groups (non-lesion control, veh, EV) based on 9 gene expression outcome variables from qPCR analyses (GRIA1, GRIA2, GRIN1, GRIN2B, GABRA1, GABRA2, GABRA5, GABRD, GABBR2), and 21 synaptic and microglia outcome measures (%area VGLUT1, VGLUT2, VGAT, GLUR2/3, GABA a 1, GABA b R2; % of VGLUT1, VGLUT2 or VGAT with Iba-1; % of Iba-1 with VGLUT1, VGLUT2 or VGAT; % area C1q; % of VGLUT2 with C1q; % C1q with VGLUT2; cell densities of Rami ed, Hypertrophic, Amoeboid C1q + and C1q-microglia), as described [14]. Z-scores were obtained for each variable, and pair-wise comparisons were used to calculate a distance matrix based on squared Euclidean distances. The multidimensional distance matrix was then reduced to two dimensions via NMDS, and the resulting values of each case were plotted, with the distances between data points representing the relative similarities based on the set ofvariables. For the results of microglia reconstruction, 3-way ANOVA with post hoc Fisher's LSD was performed in MATLAB for soma volume and aspect ratio, using three factors: group, phenotype/morphology, and C1q+/-.

Lesion-related reduction of VGLUT2 but not VGLUT1 density in perilesional M1 and PMC
To assess the extent of synapse loss and remodeling with lesion and EV treatment, we determined the optical density (% area labeled) and size of immunolabeled presynaptic and postsynaptic structures in perilesional M1 gray matter and perilesional dorsal PMC (Fig. 1b). Using one-way ANOVA, we assessed the effect of experimental group (non-lesion, lesion + veh and lesion + EV) on the expression of presynaptic axon terminals labeled with VGLUT1 and VGLUT2, which represent two distinct sets of inputs to the cortex. VGLUT1 is known to label mostly terminals from cortico-cortical pathways, while VGLUT2 labels axon terminals from subcortical afferents, mainly the thalamus [23]. We found a signi cant effect of lesion on the density of VGLUT2 + axon terminals in both perilesional M1 and dorsal PMC ( Fig. 2c-d): both EV and veh lesion groups showed signi cantly lower density (% area labeled) of VGLUT2 + puncta compared to the non-lesioned controls, indicating a lesion-related loss of VGLUT2 in M1 and PMC ( Fig. 2c: Fisher's LSD post hoc, M1: p < 0.01; PMC: p < 0.05). In contrast to differences in VGLUT2, VGLUT1 density was not signi cantly impacted by the lesion (Fig. 2a-b). No signi cant treatment effect was found for the density of either VGLUT1 + or VGLUT2 + puncta.
3.2 EV treatment mitigated the lesion-related reduction of postsynaptic GLUR2/3 in PMC In order to assess the effects of lesion and EV treatment on excitatory postsynaptic structures, we analyzed the density of GLUR2/3 AMPA glutamatergic receptor subunits in perilesional M1 gray matter and PMC. The AMPA receptor is the ionotropic glutamate receptor responsible for fast excitatory synaptic transmission [24]. In PMC, GLUR2/3 subunit density was signi cantly reduced in the vehicle treated group compared with non-lesion controls ( Fig. 2e: Fisher's LSD post hoc, p < 0.05), indicative of a lesion-related decrease in e cacy of AMPA synaptic transmission ( Fig. 2e-f). In contrast, the EV treated group did not differ from the non-lesion group in GLUR2/3 subunit density ( Fig. 2e-f). The lesion resulted in a reduction of postsynaptic GLUR2/3 in perilesional cortices, which was ameliorated by EV treatment.

Lesion-related dysregulation of GABAergic postsynaptic receptor subunit expression in perilesional M1 and PMC
We assessed the effects of lesion and EV treatment on inhibitory synapses by analyzing the density of immunolabeled presynaptic VGAT and postsynaptic GABA a 1 and GABA b receptor 2 subunits. In contrast to lesion effects on excitatory presynaptic terminals, no signi cant between-group difference was found for presynaptic VGAT in perilesional M1 or PMC gray matter ( Fig. 3a-b). However, signi cant lesion effects were found on the expression of distinct postsynaptic GABA receptor subunits. Postsynaptically located GABA a 1 subunits are associated with ionotropic receptors responsible for fast inhibitory synaptic transmission, whereas GABA b R2 subunits are associated with metabotropic receptors that mediate slow or tonic inhibition [25][26][27]. There was a signi cant lesion-related reduction of density of GABA a 1 + puncta in both perilesional M1 and PMC ( Fig. 3c-d. M1: Fisher's LSD post hoc, con. vs. veh, p < 0.05; con. vs. EV, p < 0.01; PMC: Fisher's LSD post hoc, con. vs. veh, p = 0.05; con. vs. EV, p < 0.05). In perilesional M1, both groups with lesion had smaller average size of GABA a 1 + puncta ( Fig. 3c: twoway ANOVA, main effect, p < 0.001). However, in PMC, the EV treated group showed a trend of larger GABA a 1 puncta than vehicle group (Fig. 3c, Fisher's LSD post hoc, veh vs. EV, p = 0.1).

EV treatment normalized gene expression of glutamate and GABA receptor subunits in PMC
We also assessed whether the lesion or EV treatment impacted the transcription of glutamate and GABA receptors subunits by using qPCR. T-tests were used to compare between groups. Our results showed a lesion-related increase in GRIA2 (AMPA GLUR2) mRNA expression, which was reduced by EV treatment (Fig. 4a, con vs. veh: p < 0.01; veh vs. EV: p < 0.05). A similar pattern of expression was found for GRIN1 (NMDA NR1), although the differences among groups did not reach statistical signi cance (Fig. 6a, con vs. veh: p = 0.06, con vs. EV: p = 0.14).
Interestingly, our results showed a lesion-related increase in gene expression of GABBR2 (GABA b R2) and GABRD (GABA a ∂), which transcribe GABA receptor subunits that mediate tonic inhibitory currents ( Fig. 4b, p < 0.01). The NMDS plot (Fig. 4c) showed that veh and EV treated monkeys formed distinct clusters. Further the EV-treated cluster overlapped with non-lesion control, indicating that EV treatment shifted gene expression of GLURs and GABARs towards the non-lesion control expression pattern. In contrast, vehicle-treated lesion monkeys were more dissimilar (clustered farther in the NMDS plot) to nonlesion control based on GLURs and GABARs gene expression. These data suggest that EV treatment results in a "normalization" of lesion-related changes in gene expression of AMPA GLURs and GABARs.

Lesion-related increase in microglia interaction with synaptic elements
Based on our nding of lesion-related decreases in excitatory VGLUT2 + presynaptic axon terminals ( Fig. 2c-d) and postsynaptic GLUR2/3 expression (Fig. 2e-f), we then investigated the role of microglia in the processes of synaptic remodeling with lesion and EV treatment. Microglia has been suggested to closely appose synaptic elements in order to either phagocytose damage or release neurotrophic factors [4]. First, we assessed microglia interactions on the post-synaptic structures (dendritic shaft and spines), by quantifying close appositions (including contacts and neighboring) between Iba-1/P2RY12 + microglial processes and dendrites of intracellularly lled L3 pyramidal neurons in vPMC. A microgliadendrite apposition was de ned as a "contact" when microglial process directly overlaps with the dendrite spine or shaft, or as a close "proximity/neighboring" interaction when the microglial process was within 1 micron from the labeled neuronal dendritic spine or shaft ( Fig. 5a-b). For the total microgliadendrite appositions, which includes both direct contacts and neighboring interaction between microglia and dendrites (shafts and spines), the proportion of microglial appositions with spines and shafts were about equal (microglia-spine: ~40-60%). However, both microglia-spine and microglia-shaft interactions exhibited signi cant between-group differences, speci cally on apical but not basal dendrites (Fig. 5c). Apical dendrites from neurons in vehicle group exhibited signi cantly higher density [#of appositions/100 µm dendrite length] of microglia-shaft interaction than those from EV group (Fig. 5c, Fisher's LSD post hoc, veh vs. EV, p < 0.01), as well as a higher density of microglia-spine interaction than those from EV group, with a similar increasing trend as compared to the controls (Fig. 5c, Fisher's LSD post hoc, con. vs. veh, p = 0.06; veh vs. EV, p < 0.05). Further, when looking at speci c compartments, we found that the midapical dendritic segments had signi cantly higher density of microglia-dendrite appositions in pyramidal neurons of vehicle but not EV group, compared with non-lesion control group (Fig. 5d, Fisher's LSD post hoc, con. vs veh, p < 0.05; con. vs EV, p = 0.84).
We then assessed the overlap of markers of microglia, Iba-1 and presynaptic VGLUT2 + axon terminals. The fraction of VGLUT2 colocalized with Iba-1 was greater in both groups with lesions in perilesional M1 ( Fig. 5e-f, Fisher's LSD post hoc, p < 0.01) compared to non-lesion controls.
These results suggest that the lesion was associated with increased microglial appositions on presynaptic VGLUT2 + terminals and postsynaptic structures (dendrites and spines) in perilesional motor cortices, and EV treatment reduced the lesion-related increase in microglial contacts on apical dendrites.

Lesion-related increase in C1q complement receptor expression on VGLUT2 + axon terminals and microglia
The nding of lesion-related reduction of pre-(VGLUT2) and postsynaptic (GLUR2/3) markers coupled with a lesion-related increase in microglia appositions with these synaptic elements suggested a role of microglia in synapse phagocytosis and/or pruning. C1q, as a complement protein, contributes to synapse elimination by initiating the classical complement cascade [28]. In particular, C1q tags apoptotic cells or cellular debris, including damaged synapses, and triggers downstream signaling for phagocytic elimination by macrophages or microglia [28,29]. We immuno-labeled C1q to further assess whether the lesion-related increase in these presumed "microglia-synapse interaction" is associated with synapse phagocytosis. In perilesional M1, the density of C1q + puncta was signi cantly higher in the EV group than in the non-lesion control group (Fig. 6a, d, M1: Fisher's LSD post hoc, con. vs. EV: p < 0.05) while no signi cant between-group difference was found in PMC. Similarly, C1QA transcript levels in the ventral perilesional M1 was signi cantly greater in EV compared to vehicle group (Fig. 6b, qPCR fold change, ttest, veh. vs. EV, p < 0.05). Interestingly, no signi cant between-group differences were found with regards to mRNA transcript levels of C3, a downstream target of C1q. Overall, these data suggest an upregulation of C1q activity near the lesion, which was enhanced by EV treatment. This C1q upregulation was not associated with a downstream upregulation of C3 receptor pathway, suggesting involvement of a different complement pathway cascade. Further, it remained unclear if this EV-mediated C1q upregulation re ects increased C1q tagging of damaged synapses or increased C1q expression within microglia.
To estimate the synapses tagged with C1q, we assessed the colocalization between VGLUT2 + and C1q + puncta (Fig. 6c, e). We found a signi cant lesion-related increase in the fraction of VGLUT2 + puncta colocalized with C1q + in M1 in both treatment groups (Fig. 6c, e, M1: Fisher's LSD post hoc, con. vs. veh, p < 0.05; con. vs. EV, p < 0.05). In M1, about 20-40% of all VGLUT2 + puncta were expressing C1q in the lesion groups, compared to the non-lesion group where only virtually no VGLUT2 + puncta expressed C1q. This lesion-related increase in the fraction of VGLUT2 + tagged with C1q was not signi cant in PMC ( Fig. 6c, e, PMC: Fisher's LSD post hoc, con. vs. veh, p = 0.17; con. vs. EV, p = 0.15), suggesting that C1q tagging of presynaptic VGLUT2 + terminals were prevalent within the M1 area most proximal to the lesion, and was diminished in areas more distal to the lesion. We then assessed whether these C1q + VGLUT2 + puncta were associated with and were within the vicinity of microglia-VGLUT2 contacts (Fig. 6f, i). Thus, we estimated the distance between C1q + VGLUT2 + puncta and microglia Iba-1 + VGLUT2 (C1q + V-Iba-1 + V distance) contacts (Fig. 6f). We found in M1 but not PMC, a trend of shorter C1q + V-Iba-1 + V distance in EV group as compared to the control and vehicle groups (Fig. 6f, Fisher's LSD post hoc, M1: con. vs. EV, p = 0.06; veh. vs. EV, p = 0.06). These results suggested that within perilesional M1, microglial contacts on VGLUT2 + synapses were associated with the C1q tagging on these terminals, which implied an EV-related upregulation of microglial phagocytosis of VGLUT2 + synapses within the area nearest to the lesion.

EV treatment increased C1q expression in hypertrophic microglia
While we found a lesion effect on C1q tagging of VGLUT2 + excitatory boutons, we did not nd EV treatment effects on C1q expression and VGLUT2 tagging. Since microglia produce and store C1q [15], as well as phagocytose C1q tagged debris, we therefore assessed the Iba-1-C1q colocalization. We found a lesion-related increase in the total density (percent area) of Iba-1-C1q colocalization in M1 in both treatment groups (Fig. 7a, Fisher's LSD post hoc, M1: con. vs. veh, p < 0.05; con. vs. EV, p < 0.01). In PMC, only the EV group exhibited a signi cant increase in Iba-1-C1q colocalization density compared to nonlesion control (Fig. 7a, Fisher's LSD post hoc, PMC: con. vs. EV, p < 0.01). Further a treatment effect was found in M1, with EV treated monkeys having signi cantly greater fraction of C1q puncta colocalized with Iba-1 compared to vehicle group (Fig. 7b, M1: Fisher's LSD post hoc, veh. vs. EV, p < 0.05).
Given the ndings from our previous study that EV promoted a morphological shift from in ammatory to homeostatic rami ed microglia [13], we therefore assessed whether this increased C1q + expression in microglia was associated with speci c microglial morphologies that are thought to re ect distinct immune activation states [22]. Rami ed microglia are characterized by small round somata, thin and highly branched processes, thought to be in a surveilling homeostatic state. Upon immune activation, microglia transition to an amoeboid state, which have enlarged somata, with few short and thick processes and are thought to be in a phagocytic state. A transitional state between rami ed and amoeboid states are hypertrophic microglia that have intermediate thick process, more polarized or ovoid intermediate sized somata, and branching can be extensive or not depending on state of transition. Molecular identi cation of microglia speci cally in the 'polarized' hypertrophic states suggest distinct sub-populations that have either downstream anti-in ammatory or pro-in ammatory effects [19].
Assessments of total Iba-1 + microglia by morphological subtype revealed a lesion-related increase in the density of hypertrophic microglia of both veh and EV groups only in M1 (Fig. 7e, t-test, M1 Hypertrophic: con. vs. veh, p < 0.01; con. vs. EV, p < 0.01). This lesion-related increase in hypertrophic microglia in M1 is due to an increase in the density of C1q + hypertrophic microglia in the EV monkeys, while an increase in both C1q + and C1q-hypertrophic microglia in the veh monkeys (Fig. 7f, Density of Hyper C1q + t-test: con. vs. veh, p < 0.05; con. vs. EV, p < 0.01; C1q-: con. vs. veh, p < 0.01). Further in M1, the EV group but not veh group had a higher density of C1q-rami ed microglia than control group (Fig. 7f, Density of Rami C1q-ttest: con. vs. EV, p < 0.05). No signi cant between-group differences in the densities of microglia subtypes were found in PMC (Fig. 7g).
We then assessed whether the density and relative proportion of each microglia phenotype based on morphology and C1q expression were altered by lesion and treatment. Importantly, two-way ANOVA of experimental group*area showed that lesion-related shifts in the distribution of microglia subtypes by C1q expression and morphology were region-dependent. Between-group comparison showed that in M1 but not PMC there was a lesion-related increase in density of total and C1q+ (Hyper+) hypertrophic microglia ( Fig. 7e: Fisher's LSD post hoc, M1 Hyper con. vs. veh.: p < 0.001; con. vs. EV: p < 0.001; Fig. 7f, Fisher's LSD post hoc, M1 Hyper + con. vs. veh.: p < 0.05; con. vs. EV: p < 0.001). This group*area interaction effect was also found in the proportions of microglia subtypes, indicating that lesion differentially shifted the relative proportions of microglia subtypes in a region-dependent matter (Fig. 8ac, two-way ANOVA, area*group interaction, %Rami+: p < 0.01; %Hyper-: p < 0.05; %C1q+: p < 0.05).
Given that we found between-group differences in hypertrophic microglia, we assessed the ratio of C1q + hypertrophic microglia to total hypertrophic microglia in M1 for all three groups (Fig. 8a, box-whisker plots). Our results showed that the in M1 of non-lesion control animals almost all (94%) hypertrophic microglia were C1q+, and this proportion was greater compared to the groups with lesion (Fig. 8a, C1q + Hyper/Total Hyper t-test: con. vs. veh, p < 0.01; con vs. EV, p < 0.05). EV treatment attenuated this lesionrelated proportional decrease in C1q + hypertrophic microglia. In the vehicle group, about 40% of hypertrophic microglia were C1q+; in the EV group, this proportion was signi cantly greater, with about 70% of hypertrophic microglia expressing C1q+ (Fig. 9a, C1q + Hyper/Total Hyper t-test: veh vs. EV, p < 0.01).
3.10 Lesion and EV treatment affects region-speci c expression of microglia phenotypes.
The data above indicated that EV treatment affected speci cally hypertrophic C1q + expression in M1.
However, EV treatment seemed to also affect the lesion-related shifts in between-area differences in distribution of microglial types in M1 and PMC. Notably, in non-lesion control, a between-area difference was found in the density and proportion of C1q + rami ed and hypertrophic microglia. A two-way ANOVA analysis revealed a main effect of area on the density of different microglia subtypes ( Fig. 7d-g, two-way ANOVA, area main effect, Rami+: p < 0.05; Hyper+: p < 0.01; Hyper-: p < 0.05; TotalRami: p < 0.05; TotalHyper: p < 0.01; TotalC1q+: p < 0.01; TotalC1q-: p < 0.01), indicating that microglia phenotypes were expressed differently in M1 and PMC. In M1, majority of both rami ed and hypertrophic microglia were C1q+. In PMC, majority of hypertrophic microglia were C1q-and a more equal distribution of C1q + and C1q-rami ed microglia was found ( Fig. 8a-b). Compared to PMC, M1 had greater density of Rami + but not Rami-microglia, and greater density of both Hyper + and Hyper-microglia ( Fig. 7f-g). Consistently, the proportion of Rami + microglia is greater in M1 than in PMC (Fig. 8c, t-test, con. %Rami + M1 vs. PMC: p < 0.01). Lesion in the vehicle group shifts this distribution to the opposite pattern, with %Rami + microglia lower in M1 than in PMC (Fig. 8c, Fisher's LSD post hoc, %Rami + veh., M1 vs. PMC: p < 0.05). Lesion in the vehicle group also caused relative increase in %Hyper-microglia in M1 compared to PMC, that did not differ signi cantly in the non-lesion control brains (Fig. 8c, Fisher's LSD post hoc, %Hyper-con., M1 vs. PMC: p = 0.07; veh., M1 vs. PMC: p < 0.05). Interestingly, EV treatment seemed to dampen this lesionrelated decrease in %Rami + and increase in % Hyper-in M1 relative to PMC (Fig. 8c).
In summary, between-group within area comparisons showed that in M1, lesion proportionally decreased rami ed microglia, but increased hypertrophic microglia compared to non-lesion control (Fig. 7&8). The opposite pattern was seen in PMC, where lesion shifted to a proportional increase of C1q + rami ed microglia but decrease in C1q-hypertrophic microglia (Fig. 8a-c). Further, there are between-area differences in the normative distribution of these microglia. Our results also demonstrated that EV treatment regulated the post-injury region-dependent expression of microglia subtypes by reversing or attenuating the lesion caused changes.
3.11 Microglia morphological features are dependent on experimental group and C1q expression Overall, our results indicated that lesion and EV treatment facilitated a speci c phenotypic shift in C1q + vs C1q-hypertrophic, and to a lesser extent rami ed, microglia in a region dependent manner. We thus further assessed whether these C1q + vs C1q-hypertrophic and rami ed microglia represented distinct subclasses with unique morphologic features. Using 3D reconstruction methods, we quanti ed morphological features of individual microglia in M1 and assessed the independent and interactive effects of morphology*C1q expression*experimental group (Fig. 8d-g). We were not able to report reconstruction data for C1q-hypertrophic microglia in control animals since this subpopulation was very rare in this group. There were signi cant main effects of group and morphology on the number of primary processes (Fig. 8e: three-way ANOVA, group/morphology main effect, p < 0.05). Speci cally, microglia in both groups with lesion had a greater number of primary processes as compared to the controls.
Rami ed C1q + microglia in both groups with lesion showed a greater number of primary processes as compared to the controls (Fig. 8e, Fisher's LSD post hoc, Rami C1q+: con. vs. veh: p < 0.05; con. vs. EV: p < 0.05). In addition, there was a treatment effect on hypertrophic C1q+/-microglia, which exhibited a greater number of processes in the EV-treated group compared to those in vehicle and control groups (Fig. 8e, Fisher's LSD post hoc, Hyper C1q+: con. vs. EV: p < 0.05; veh. vs. EV: p < 0.05; Hyper C1q-: veh. vs. EV: p < 0.05). For all C1q+/-microglia in the EV group, the hypertrophic microglia had greater numbers of primary processes as compared to the rami ed cells (Fig. 8e, Fisher's LSD post hoc, C1q + Hyper. vs. Rami: p < 0.05; C1q-Hyper. vs. Rami: p < 0.05). These results demonstrated that regardless of C1q expression, microglia from the EV-treated group was characterized by a greater number of primary processes, consistent with our previous work [13].
Signi cant main effects of C1q expression (+/-) and group were found for features of rami ed and hypertrophic microglia somata (Fig. 8f: three-way ANOVA, group/C1q expression main effect, p < 0.01).
Both hypertrophic and rami ed C1q + microglia in EV group but not the veh group had greater surface area of their soma as compared to the non-lesion control group (Fig. 8f, Fisher's LSD post hoc, C1q + Hyper. con. vs. EV: p < 0.05; C1q + Rami. con. vs. EV: p < 0.05). In addition, a signi cant main effect of C1q expression was found for cell body aspect ratio, which was calculated as the maximum diameter of the microglial somata divided by the minimum diameter (Fig. 8g, three-way ANOVA, main effect, p < 0.05).
Aspect ratio can be a measure of microglia polarization and immune activation [30]. Cell bodies of C1q + microglia (rami ed or hypertrophic) exhibited a smaller aspect ratio (rounder) compared to the C1qmicroglia which had a more elongated, oval shape. These results showed that C1q expression on microglia was associated with rounder somata with larger surface areas.
We then assessed whether these unique set of morphological features will reveal distinct clusters of microglia subtypes. For each microglia reconstructed, we plotted a 3D scatter plot of primary process, soma volume, and soma aspect ratio, and annotated the plots based on morphology and C1q expression, cortical region and experimental group (Fig. 8h). 3D scatter plot revealed some clustering of microglia based mainly on morphology and not C1q expression (Fig. 8h). Speci cally, rami ed microglia were associated with smaller soma volume and rounder somata shape (smaller aspect ratio) as compared to the hypertrophic microglia (Fig. 8h, left panel). Our clustering analysis also showed that microglia in the three groups exhibited different morphologies (Fig. 8h, right panel). The non-lesion controls tend to have smaller soma with a broader range of aspect ratio and less primary processes. Microglia in the EV group was shown to have larger soma and more processes, whereas the vehicle group had intermediate soma volume but a higher aspect ratio indicating an oval somata shape (Fig. 8h right panel). In contrast to morphological categories and experimental group, no clustering of microglia based on these 3 morphological features was found between regions (Fig. 8h, middle panel).
We then determined the combined effects of synaptic and microglia outcome measures on the relative (dis)similarities of areas and experimental groups in this study. Thus, we performed non-metric multidimensional reduction and clustering of individual PMC and M1 tissue from each case, based on a distance proximity matrix derived from pair-wise correlation of 21 synaptic and microglia outcome measures (per case: % area VGLUT1, VGLUT2, VGAT, GLUR2/3, GABA a 1, GABA b R2; % of VGLUT1, VGLUT2 or VGAT with Iba-1; % of Iba-1 with VGLUT1, VGLUT2 or VGAT; % area C1q; % of VGLUT2 with C1q; % C1q with VGLUT2; cell densities of Rami ed, Hypertrophic, Amoeboid C1q + and C1q-microglia). Our analyses note that there is a strong separation between the control from the lesion group. Further, within each experimental group, M1 vs PMC are separated. This regional separation is most prominent within the non-lesion control group, highlighting the normative diversity between cortical areas with regards to synaptic and microglial features [17,31].
Our previous study [12] has reported that compared with vehicle-treated monkeys, EV-treated monkeys exhibited enhanced recovery of ne motor function of the hand, evidenced by the fewer number of days to return to pre-operative hand grasp pattern and latency to retrieve food reward. Using the functional recovery data from Moore et al. (2019) and Pessina et al. (2019) [12,32], we used linear regression analyses to determine whether the cellular data reported above were associated with behavioral measures of motor recovery (number of days to return to pre-operative latency and grasp pattern). Our results showed that increased C1q-Iba-1 colocalization in M1-speci cally increased density of C1q + hypertrophic microglia-was associated with a more rapid recovery rate (fewer days to return to preoperative latency; Fig. 9b, R 2 = 0.752, p < 0.01; Fig. 9d, R 2 = 0.533, p < 0.05). In PMC, we found increased portion of VGLUT2 tagged by C1q in PMC was associated with a slower recovery rate (more days to return to preoperative grasp pattern; Fig. 9c, R 2 = 0.589, p < 0.05). Increased density of C1q + rami ed microglia in PMC was also associated with a slower recovery rate (more days to return to preoperative grasp pattern, Fig. 9e, R 2 = 0.49, p < 0.05). Overall, our results indicated that the C1q + hypertrophic microglial expression in perilesional M1 was bene cial for the recovery of ne motor function. In contrast, C1q tagging on VGLUT2 + synapse in PMC was detrimental for recovery, indicated by the slower recovery rate.

DISCUSSION
Previous studies have demonstrated the effects of EVs on shifting microglial morphological phenotypes [13] and ameliorating synaptic imbalance in cortical and spinal motor circuits [14,33], to support recovery of motor function after cortical injury in primary motor cortex (M1). The current study provides a mechanistic link between these previous data, showing complementary effects of EV treatment on synaptic marker and microglial expression, and their structural and molecular relationships. As summarized in Fig. 9f, both lesion and EV treatment modify microglial-synapse relationships and microglial phenotypic expression of the complement pathway initiator protein, C1q, in a region-dependent manner. Compared to vehicle, EV treatment was associated with increased expression of C1q + hypertrophic microglia and decreased expression of C1q-hypertrophic microglia in M1, but decreased expression of C1q + rami ed microglia in PMC (Fig. 9f). These data point to the role of EV-mediated regulation of region and circuit-speci c synaptic plasticity to support recovery of motor function after cortical injury.

Differential effects of injury on excitatory VGLUT1 and VGLUT2 terminals suggest pathway-speci c mechanisms for plasticity
Cortical injury leads to neuronal hyperexcitability, excitotoxicity, and disruption of synaptic transmission in perilesional cortex [3,14]. Glutamate, the major excitatory neurotransmitter in the central nervous system, is stored and transported into synaptic vesicles by presynaptic vesicular glutamate transporters (VGLUTs) that have isoforms differentially expressed across distinct neuronal pathways [34]. In the adult brain, VGLUT1 is expressed mainly by cortico-cortical axons, while VGLUT2 is mainly expressed in subcortical glutamatergic neurons, predominantly in thalamocortical axons [35]. The current results showed signi cant reduction of VGLUT2 in both groups with lesion compared to non-lesion controls, while no signi cant difference was found in VGLUT1. These results suggest that either axon terminals expressing VGLUT2 + may be selectively vulnerable to injury compared to VGLUT1+, or that VGLUT1 + connections have a greater degree of plasticity after injury. This is consistent with literature suggesting that axon terminals expressing VGLUT1 such as those in the hippocampus exhibit higher potential for plasticity than those expressing VGLUT2, such as the climbing bers in the cerebellum [36]. Future experiments will be important to clarify the molecular pathways underlying the differential regulation of VGLUT1 and VGLUT2 expression after injury.

EV treatment ameliorated injury-related dysregulation of glutamate AMPA receptor subunit gene expression in PMC
Glutamatergic AMPA receptor composition is an important determinant of synaptic strength and plasticity [3]. The GLUR2 AMPAR subunit, in particular, controls calcium permeability, thereby affecting AMPARs tra cking, and spine growth [37,38]. In the present study, lesion-related reduction of GLUR2/3 receptors, which is dampened in the EV treated group. However, mRNA expression GRIA2, the gene for GLUR2 AMPA receptor subunit, showed the opposite trend; the lesion-related increase in GRIA2 mRNA was downregulated and normalized by EV. The opposite effects of lesion on GLUR2/3 protein and GRIA2 mRNA may be due to numerous factors. First, the GRIA2 mRNA is related to GLUR2 protein translation, but not GLUR3 receptors. Second, the mRNA-protein discrepancy suggests differences in transcriptional and post-translational regulation of plasticity in response to lesion and lesion-induced hyperexcitability.
Previous work in rodents has shown that increased protein expression of GLUR2 promotes dendritic spines formation and enlargement in rat hippocampal neurons [39]. Upregulation of GRIA2 mRNA and concomitant downregulation of GRIA1 (GLUR1) mRNA was found in response to pharmacologicallyinduced hyperexcitability in neuronal cultures [40]. The current data is consistent with our previous ndings showing lesion-induced hyperexcitability, accompanied by excitatory synapse loss at the electrophysiological and structural level in vPMC [14]. Thus, it is possible that lesion-related GRIA2 mRNA upregulation resulted from increased hyperexcitability, but the downstream mechanisms for GLUR2/3 subunit protein synthesis, tra cking and insertion were impaired, thereby not allowing for spine and synapse growth after injury. Interestingly, the current data suggest that this potential impairment in posttranslational regulation is apparent only in vehicle monkeys, where upregulated GLUR2 mRNA was found together with downregulated GLUR2/3 protein expression. In EV-treated monkeys, GLUR2 mRNA and GLUR2/3 protein levels, which showed the opposite trend from vehicle monkeys, were normalized closer to baseline non-lesion control. This is consistent with our previous work demonstrating that the lesionrelated decrease in excitatory postsynaptic current frequencies and spine loss were ameliorated by EVs [14]. Together, these data suggest that EV-mediated dampening of chronic hyperexcitability in perilesional neurons, can be associated with preventing aberrant plasticity and maintaining glutamatergic synapse growth.

Differential effects of injury on distinct inhibitory neurotransmitter receptor expression
In addition to changes in excitatory neurotransmission, phasic and tonic inhibitory GABAergic transmission, conferred by distinct ionotropic GABA a and metabotropic GABA b receptor subunits [41], also play a complex role in recovery after injury [42,43]. Here, we found in both M1 and PMC, a lesionrelated reduction in the density of GABA a 1, a subunit localized on synaptic membranes mediating fast phasic inhibitory currents [41]. Further, in our qPCR results, we found a lesion-related increase in gene expression of GABA a ∂ (GABRD) and GABA b R2 (GABBR2) subunits, known to mediate tonic inhibitory currents that control overall cell excitability [3,25,44]. The current results are consistent with previous studies in rodent and in vitro models showing that injury results in increased tonic inhibition to prevent excitotoxicity during the acute recovery period [42,43]. However re-establishing phasic inhibitory transmission is needed to support reorganization [42,43]. While the current data did not show a treatment effect with regards to GABA receptors, our previous study showed that EV treatment was associated with a speci c increase in distal apical inhibitory synapses and task-related immediate early gene activation of dendritic-targeting inhibitory interneurons that support recovery of motor function [14]. Thus, cell-type and compartment speci c changes in GABAergic receptor expression across recovery would be important to assess in future work.

EV treatment upregulated the anti-in ammatory C1q + hypertrophic microglia in M1
While activity-dependent, neuronal mechanisms of synaptic plasticity have been well studied, it is only recently that the role of microglia and neuro-immune signaling have been investigated [45,46]. In the healthy brain, microglia can regulate synaptic turn-over through phagocytosis of synapses for pruning, or releasing trophic factors to promote synapse growth [45,47]. After injury, microglia, as the resident macrophages of the brain, are stimulated to mediate clearance of damaged synapses [48]. A crucial part of this microglial mediated neuro-immune signaling is the complement system [15,20]. During periods of active synaptic pruning or after injury, the initiating protein of the classical complement cascade, C1q, is produced by microglia to tag excess or damaged synapses. C1q tagging then triggers downstream deposition and activation of complement effector molecules, such as C3 receptors, which would in turn lead to microglial synapse phagocytosis [28,29,49]. In the present study, the vehicle group showed a lesion-related elevation in C1q tagging of VGLUT2 + axon terminals (C1q+/VGLUT2 + colocalization), coupled with decreased VGLUT2 + density, suggesting greater synapse damage and loss [14]. However, in the EV group, this C1q-synapse tagging was dampened, and coupled with a greater expression of C1 + hypertrophic microglia in perilesional M1 (Fig. 9f). After cortical injury, microglia can be 'immuneactivated' to exhibit distinct macrophage-like phenotypes polarized towards either pro-and antiin ammatory functions [50][51][52]. Synthesis of C1q within anti-in ammatory (M2 macrophage-like) microglia has been shown to be acutely upregulated after cortical injury, playing a protective role, to promote the clearance of apoptotic cells and secretion of anti-in ammatory cytokines, and suppress production of pro-in ammatory cytokines [6, 8] [29,53]. Thus, EV treatment mitigated a sustained chronic pro-in ammatory state, evidenced by increased expression of the protective and debris-clearing C1 + hypertrophic microglia in perilesional cortex 12 weeks post-injury. Indeed, here we show that greater expression of C1q + hypertrophic microglia in perilesional M1 was associated with more rapid recovery of function. Interestingly, we found that the vehicle group exhibited a decrease in the proportion of C1q + but an increase in C1q-hypertrophic microglia in perilesional M1, compared to EV and control groups (Fig. 9f). Thus, these C1q-hypertrophic microglia represents a distinct population, likely belonging to the pro-in ammatory subclass [50,51], which persisted in the vehicle monkeys. These pro-in ammatory microglia can exacerbate neurotoxicity by releasing pro-in ammatory cytokines including TNF-, IL-6, and IL-1β [4,54].
In a mouse model of Alzheimer's disease [55], it was found that early in the disease, C1q 'primed' microglia are anti-in ammatory and protective. However, after C1q release and tagging, the downstream molecular targets of C1q can either promote anti-in ammatory (via C3b receptor activation) or exacerbate chronic pro-in ammatory (via C3a or C5 receptor activation) signaling [29]. Future studies to assess the temporal progression of distinct C1q effectors will be important to further understand the role of EVs in modulating the complement system to support recovery after cortical injury. Nevertheless, the current ndings together with our previous data [13] support the role of EV treatment in inducing an early shift from pro-in ammatory to anti-in ammatory microglia in perilesional cortex, preventing chronic in ammation and damage after cortical injury.

Region-dependent effects of injury and treatment in modulating microglial phagocytosis.
The current ndings revealed novel region-speci c expression of microglial phenotypes that are differentially modulated by injury and EV treatment. In the non-injured brain, rami ed homeostatic microglia predominated. However, there were baseline regional differences in the proportion of C1q + vs C1q-rami ed microglia, with M1 showing a signi cantly greater proportion of C1q + microglia than PMC. These data are consistent with previous work in the rodent and human brain, highlighting diversity across cortical areas with regards to microglial subpopulations [56][57][58][59]. Further, the present ndings suggest innate differences in microglial/C1q dependent synapse turnover between two cortical motor areas, with synapses in M1 likely subject to greater synaptic pruning via the upregulated C1q + rami ed microglia compared to PMC [46,60]. Interestingly, the lesion alone resulted in a reversal of the M1 vs PMC gradient in C1q + rami ed microglia; However, EV treatment mitigated this lesion-related regional shift. Indeed, while increasing expression of C1q + hypertrophic microglia in M1 was associated with a more rapid recovery rate, increased C1q-synapse (VGLUT2) tagging and C1q + rami ed microglia in PMC were associated with a prolonged recovery. These results suggest that microglia phagocytosis can be bene cial in perilesional M1, which likely re ect the neuroprotective effects of clearance of damaged tissue necessary to promote tissue repair and re-establish homeostasis [4]. However, chronic and excessive phagocytosis of synapses in PMC can exacerbate neuronal cell death and loss of connections [5,45,46, 61] that can be detrimental to recovery.

Conclusion
We have elucidated the potential microglia-synapse and complement C1q-related mechanisms of how MSC-EVs treatment can enhance recovery in a monkey model of cortical injury. Cortical lesion results in reduced expression of excitatory and inhibitory synaptic markers in perilesional M1 and PMC, consistent with the functional de cits in excitatory and inhibitory synaptic transmission shown in monkeys [14] and rodents [42]. The current data suggest that EV treatment facilitated synaptic plasticity by regulating microglial activity in a region-dependent manner-enhancing anti-in ammatory C1q + hypertrophic microglia expression in perilesional M1 for clearance of acute damage, and thereby preventing excessive synaptic loss in PMC and chronic synaptic dysfunction. These mechanisms may act to preserve synaptic cortical motor networks and balanced normative M1/PMC synaptic connectivity to support functional recovery after injury [14,[62][63][64]. However, it still remains unclear whether EVs act on microglia directly or indirectly via upstream signaling pathways [4] or via suppressing pro-in ammatory complementary system markers downstream to C1q. Our ndings form the basis of future studies to dissect the molecular effects of MSC-EVs treatment on the complement cascade crucial for recovery after cortical injury, and extend their therapeutic application for acute and chronic neurodegenerative diseases. Experimental design and representative images of immunolabeled markers, lesion, and sampling location. a Experimental work ow: monkeys were trained and tested on a ne motor task -the Hand Dexterity Task, before and after the surgical lesion of M1 as described in Moore et al., 2019. Then, the monkeys were randomly assigned with either vehicle or EV treatment, which were infused IV at 24 hours and 14 days post-injury. The brains were harvested 14 to 16 weeks after the surgery. (a1) During Krebs buffer perfusion, 1-2cm fresh tissue block was harvested from the ventral perilesional motor/premotor cortex: the caudal 1/4 was processed for qPCR and the rostral part was cut into 300 µm acute slices for whole cell patch clamp recording and intracellular lling of layer 3 pyramidal cells. (a2) The remainder of the brain containing the lesion and rostral and dorsal perilesional cortex was xed with paraformaldehyde then cut into serial coronal sections and processed with immunohistochemical labeling. The synaptic markers that were immuno-labeled included: VGLUT1 &2, GLUR2/3, VGAT, GABA a 1, and GABA b R2. Microglia markers were combining labeled Iba-1 and P2RY12. C1q was immuno-  Glutamate and GABA receptors subunit mRNA expression in perilesional cortex. a Fold changes of glutamate receptor subunit gene expression. The gene expression of GRIA2 (GluR2) was signi cantly higher in the veh group, as compared with non-lesion controls (t-test, p < 0.001) and the EV group (t-test, p=0.01). b Fold changes of GABA receptor subunit gene expression. The gene expression of GABRD (GABA a ∂) was signi cantly higher in the EV group as compared with non-lesion controls (t-test, p=0.004). apical spine (p=0.038) appositions compared to EV, and a trend for greater total appositions in the vehicle than in the control was found (con. vs. veh., p=0.058). d Total microglial appositions (contacts & neighboring on spines and shafts) on different segments of apical/basal dendrites in vPMC. Mid-apical dendrites had higher density of microglial contacts only in veh compared to control (Fisher's LSD post hoc, con. vs veh, p=0.03; con. vs EV, p=0.84). Non-lesion control: n=10 from 3 monkeys. Veh group: n=10 from 3 monkeys. EV group: n=7 from 2 monkeys. e The fraction of VGLUT2 colocalized with microglia was higher in both groups with lesions in M1 (Fisher's LSD post hoc, con. vs. veh, p=0.009; con. vs. EV, p<0.001). Non-lesion control: n=3. Veh group: n=4. EV group: n=5. f Representative images of dual channel labeling (left panel) of microglia (red) and VGLUT2 (green), with right panel showing higher resolution of colocalized VGLUT2-microglia label masked in white C1q co-expression on VGLUT2+ axon terminals and Iba1 microglia. a The density (% area label) of C1q+ puncta in perilesional M1 and PMC (Fisher's LSD post hoc, M1: con. vs EV p=0.028; Non-lesion control: n=5. Veh group: n=5. EV group: n=4). b Fold changes of C1QA and C3 gene expression in perilesional M1 (t-test, veh. vs. EV, p=0.027). c The density (% area label) of VGLUT2+ puncta colocalized with C1q+ puncta (Fisher's LSD post hoc, con. vs. veh, p=0.02; con. vs. EV, p=0.02; Non-lesion control: n=3. Veh group: n=4. EV group: n=5). d Representative maximum-projection confocal images of C1q+ puncta immuno-labeling in M1 and PMC. Scale bar: 20µm. e Representative maximum-projection confocal images of 4 optical slices showing dual channel labeling (left panel) of C1q (yellow) and VGLUT2 (green), with colocalized VGLUT2-C1q points masked in white (right panel). Scale bar: 20µm. f The distance between the C1q-VGLUT2 colocalized points and microglia (Iba1)-VGLUT2 colocalized points. Inset shows schematic diagram of how C1q-VGLUT2-Iba1 distance was determined. Non-lesion control: n=3.  C1q. dCell densities of total C1q+ vs C1q-microglia. In M1, C1q-microglia: greater density in lesion than control (t-test, con. vs. veh: p<0.001; con. vs. EV: p=0.004); EV but not veh had greater C1q+ microglia density than control (t-test, con. vs. veh: p=0.07; con. vs. EV: p=0.03). In PMC, greater density of C1q+ microglia in veh than control group (t-test, con. vs. veh: p=0.02). e In M1, hypertrophic microglia: greater Page 36/39 density in lesion compared to control group (t-test, con. vs. veh: p=0.01; con. vs. EV: p=0.008). f In M1, the EV group showed higher density of C1q-rami ed microglia as compared to the control group (t-test, con. vs. EV: p=0.01). Both groups with lesion had higher density of hypertrophic C1q+ microglia as compared to the controls (t-test, con. vs. veh: p=0.047; con. vs. EV: p<0.001). However, only veh had higher density of hypertrophic C1q-microglia as compared to the controls (t-test, con. vs. veh: p=0.005). g In PMC, no between-group difference was found in the density of rami ed or hypertrophic microglia. (d-g non-lesion control: n=3. Veh group: n=4. EV group: n=5) Figure 8 The expression of different microglia phenotypes in M1 and PMC. a Pie charts of %microglia by morphology and C1q expression in M1. Right inset shows the ratio of Hyper+ to the total Hyper microglia in each group (t-test, con. vs. veh: p=0.001; con. vs. EV: p=0.03; veh. vs. EV: p=0.006). b In PMC (Ramiveh < con.: p=0.05; EV vs con.: p=0.49). c Relative proportion of Rami+, Hyper+, Rami-, and Total C1q+ normalized to the total number of microglia counted for each case. Signi cant group*area interaction for %Rami+ (p=0.008), %Hyper-(p=0.02), and %Total C1q+ (p=0.04). A main effect of group for %Hyper+ (p=0.01). Signi cant between-group differences per area: M1 (%Rami+ veh < con, p=0.006; EV vs con, Relationship of synaptic and microglia properties to behavior outcome features. a NMDS plot showing clustering of cases, annotated by experimental group (left) and cortical area (right), based on 21 synaptic and microglia outcome measures (%area VGLUT1, VGLUT2, VGAT, GLUR2/3, GABAA alpha1, GABAB R2; % of VGLUT1, VGLUT2 or VGAT with Iba1; % of IBA1 with VGLUT1, VGLUT2 or VGAT; % area C1q; % of VGLUT2 with C1q; % C1q with VGLUT2; cell densities of Rami ed, Hypertrophic, Amoeboid C1q+ and C1q-microglia). The proximity of points indicates the relative similarity-based pair-wise correlation of these multiple variables. b-e Signi cant linear correlations between synaptic-microglial measures and behavioral outcome measures: b Increased density of C1q and Iba1 colocalization in M1 correlated with faster recovery time (less days to return to pre-operative latency to retrieve food reward; R 2 = 0.752, p=0.002). c Increased fraction of VGLUT2 colocalized with C1q in PMC correlated with slower recovery time (more days return to preoperative grasp pattern; R 2 = 0.589, p=0.016). d Greater expression of C1q+ hypertrophic microglia in M1 was correlated with faster recovery time (R 2 = 0.533, p=0.026). d Greater expression of C1q+ rami ed microglia in PMC was associated with slower recovery time (R 2 = 0.490, p=0.036). f A schematic showing summary of ndings and proposed model of the lesion and EV treatment effects on microglial-synapse modulation and C1q signaling pathways. Cortical lesion in M1 induces acute damaged in neuronal structures that triggers an acute increase in C1q+ signaling cascade to initiate phagocytotic clearance. The veh group had accumulation of further damage and downstream C1q pathway related proteins that sustains a chronic pro-in ammatory response (C1q-hypertrophic microglia). The EV treatment upregulated C1q+ mediated clearance of debris and facilitated an early shift to the anti-in ammatory C1q+ hypertrophic microglia phenotype that persisted in the chronic stages, thereby supporting functional recovery