Nanoparticle‐microglial interaction in the ischemic brain is modulated by injury duration and treatment

Abstract Cerebral ischemia is a major cause of death in both neonates and adults, and currently has no cure. Nanotechnology represents one promising area of therapeutic development for cerebral ischemia due to the ability of nanoparticles to overcome biological barriers in the brain. ex vivo injury models have emerged as a high‐throughput alternative that can recapitulate disease processes and enable nanoscale probing of the brain microenvironment. In this study, we used oxygen–glucose deprivation (OGD) to model ischemic injury and studied nanoparticle interaction with microglia, resident immune cells in the brain that are of increasing interest for therapeutic delivery. By measuring cell death and glutathione production, we evaluated the effect of OGD exposure time and treatment with azithromycin (AZ) on slice health. We found a robust injury response with 0.5 hr of OGD exposure and effective treatment after immediate application of AZ. We observed an OGD‐induced shift in microglial morphology toward increased heterogeneity and circularity, and a decrease in microglial number, which was reversed after treatment. OGD enhanced diffusion of polystyrene‐poly(ethylene glycol) (PS‐PEG) nanoparticles, improving transport and ability to reach target cells. While microglial uptake of dendrimers or quantum dots (QDs) was not enhanced after injury, internalization of PS‐PEG was significantly increased. For PS‐PEG, AZ treatment restored microglial uptake to normal control levels. Our results suggest that different nanoparticle platforms should be carefully screened before application and upon doing so; disease‐mediated changes in the brain microenvironment can be leveraged by nanoscale drug delivery devices for enhanced cell interaction.

vation (OGD) to model ischemic injury and studied nanoparticle interaction with microglia, resident immune cells in the brain that are of increasing interest for therapeutic delivery. By measuring cell death and glutathione production, we evaluated the effect of OGD exposure time and treatment with azithromycin (AZ) on slice health. We found a robust injury response with 0.5 hr of OGD exposure and effective treatment after immediate application of AZ. We observed an OGD-induced shift in microglial morphology toward increased heterogeneity and circularity, and a decrease in microglial number, which was reversed after treatment. OGD enhanced diffusion of polystyrene-poly(ethylene glycol) (PS-PEG) nanoparticles, improving transport and ability to reach target cells. While microglial uptake of dendrimers or quantum dots (QDs) was not enhanced after injury, internalization of PS-PEG was significantly increased. For PS-PEG, AZ treatment restored microglial uptake to normal control levels. Our results suggest that different nanoparticle platforms should be carefully screened before application and upon doing so; disease-mediated changes in the brain microenvironment can be leveraged by nanoscale drug delivery devices for enhanced cell interaction.

| INTRODUCTION
Cerebral ischemia is a major cause of death in both neonates and adults. 1,2 Unfortunately, treatments for both neonatal hypoxiaischemia (HI) and adult stroke provide only modest benefits in mortality and morbidity. 3,4 Investigation of effective treatments for HI continues to be a critical area of research. Microglia, the resident immune cells of the brain are of more recent and special interest for therapeutic targeting in HI. 5 Microglia become activated after ischemic injury, exhibit increased phagocytic behavior, and contribute to neuroinflammatory and reactive oxygen species (ROS) stress that may exacerbate damage in the brain. 6 Thus, an opportunity exists to provide neuroprotection after ischemic injury by designing therapeutics to target and modulate microglial behavior.
A promising strategy for microglial-targeted therapeutic development is the delivery of drugs via nano-sized carriers. Although nanoparticle platforms vary widely in composition, shape, and other physical characteristics, several distinct nanoparticle types have shown an ability to overcome biological barriers to drug delivery in the brain. For example, polymeric nanoparticles (size < 200 nm) with a dense poly(ethylene glycol) (PEG) coating can rapidly penetrate within small pores in the brain extracellular space (ECS) that restrict the diffusion and broad distribution of most therapeutics. 7,8 Adequate intracellular trafficking of therapeutics also presents a major drug delivery challenge, but nanoparticles can leverage existing endocytosis pathways in microglial cells for internalization. Quantum dots (QDs) and poly(amidoamine) (PAMAM) dendrimer nanoparticles, among others, have shown accumulation within activated microglia. [9][10][11][12][13] While nanoparticles can facilitate and enhance drug transport in the brain, these effects are dependent on both nanoparticle characteristics and disease state, 14 requiring further screening and study.
We investigate the role of injury, treatment, and nanoparticle type in driving nanoparticle-microglial interactions in ischemic conditions. We use ex vivo organotypic whole hemisphere (OWH) brain slices, which have emerged as a high-throughput platform for modeling disease processes and screening therapeutic platforms, including nanoparticles. 13,15,16 The ability to obtain multiple OWH slices from a single brain reduces biological variation and enables detailed investigation of disease environments or therapeutic efficacy. OWH slices also preserve functional relationships between neighboring cells and maintain 3D-cytoarchitecture. 15 Importantly, oxygen-glucose deprivation (OGD) has been widely used to model ischemic injury in organotypic slices. 17 OGD brain slice models retain in vivo pathological processes including extracellular glutamate release, neuronal damage, and production of cytokines and oxidative stress markers. [18][19][20] Thus, OGD exposed OWH slices are a powerful tool for evaluation of nanoparticle-cell interactions in ischemic injury.
To establish the degree of injury and microglial response following OGD exposure, we evaluate cytotoxicity and oxidative stress in healthy, OGD exposed, and azithromycin (AZ)-treated OWH brain slices. AZ is an FDA-approved therapy that can suppress both acute and chronic pathologic microglial activation in response to ischemic stroke injury. 21,22 Because microglial behavior correlates with disease severity, 23,24 AZ modulation of microglia provides one avenue to study microglia-nanoparticle interaction in response to treatment. 25 We use a Python-based image analysis technique to quantify the degree of microglial morphological heterogeneity following injury and treatment. We next investigate how injury alters the ability of nanoparticles to diffuse within the brain, an important factor for maximal distribution to reach target microglial cells. Last, we use flow cytometry and immunofluorescent imaging to quantify nanoparticle uptake in microglia based on injury and treatment. We compare three distinct nanoparticle platforms, polystyrene (PS)-PEG, PAMAM dendrimers, and cadmium selenide/cadmium sulfide (CdSe/CdS) core/ shell QDs, to determine the influence of nanoparticle physical characteristics on microglial uptake. In using OWH slices to probe nanoparticle-microglial interaction after disease and treatment, our study informs the design of nanoparticles to leverage the brain microenvironment and target microglial cells for enhanced therapeutic outcome in ischemic conditions.

| Animal experiments and ethics statement
This study was performed in accordance with the guide for the care and use of laboratory animals of the National Institutes of Health 2.2 | Slice culturing, OGD, and treatment OWH brain slice culturing techniques were prepared adapted from previously published methodology. 13 Slice preparation and culturing medias are provided in Supplemental Information. For glutathione and flow cytometry measurements, three brain slices were plated per membrane insert. For all other experiments, one brain slice was plated per membrane insert. After slices rested overnight in the incubator, culture media was removed, and fresh media was added, followed by two more days of rest. Samples underwent OGD after 3 days in vitro (DIV), except for flow cytometry studies, which underwent 2 DIV. The end of the OGD incubation period was defined as time t = 0 hr. For the normal control (NC) condition, slices proceeded directly to t = 0 hr without OGD media exchange. For a subset of groups, 100 μl 1 x PBS containing 0.1 mg superoxide dismutase (SOD, Cu/Zn SOD1 from bovine erythrocytes, Sigma) or 0.75 μg (150 mg/kg brain tissue) AZ (Zithromax) per slice was added directly to the 3 and 0.5 hr OGD slices, respectively. Six-well plates were returned to the CO 2 incubator until further processing.  Dr. Anjali Sharma at the Johns Hopkins Center for Nanomedicine. 28,29 These conjugates are stable at physiological conditions and have been validated for ex vivo application at the concentration used in our study (1 ng/ul). 9 CdSe/CdS core-shell QDs with PEG-methoxy functionality were provided by Dr Vince Holmberg in the Department of Chemical Engineering at the UW, which were proven stable at physiological conditions for ex vivo application. 13 Nanoparticle characterization, multiple particle tracking (MPT) to measure nanoparticle diffusion, trajectory analysis 30 and application of the Amsden obstruction-scaling model to calculate effective pore sizes, including all assumptions, [31][32][33][34] are detailed in Supplemental Information.

| Nanoparticle colocalization in microglia and neurons
To probe nanoparticle interactions with microglia, at t = 1 hr PS-PEG, D-Cy5, and QDs (10 μl of 1 ng/μl) were topically pipetted on NC, 0.5 h OGD, and 0.5 h OGD + AZ OWH slices. The slices were then live-incubated for 4 hr to allow the nanoparticles to diffuse through the brain tissue and interact with microglia. Slices were fixed and imaged or processed for flow cytometry, as described in Supplemental Information.

| VAMPIRE for microglial morphometric analysis
All confocal microscopy images were converted from the .nd2 to .tiff file format. Using Python, all images were separated by RGB channel and labeled with the appropriate cell stain: DAPI for the blue channel and Iba1 for the green channel. [35][36][37] Every image was then split into four quadrants using Image_slicer. Scikit-learn split all images in an 80:20 test-to-train ratio, 38 assuring at least two images for each slice of the three experimental conditions: NC, 0.5 h OGD, and 0.5 h OGD + AZ. Cells from each image were segmented using Cell Profiler and the Cell Profiler pipeline associated with the VAMPIRE package. 39 A model of shape modes was built from all training images using the VAMPIRE package and associated protocol (https://github.com/kukionfr/VAMPIRE_open), 26 and then applied to all images. The shape mode frequencies of individual slices were averaged for resulting distribution plots. Equation (1) was used to calculate the difference in sample shape mode frequency from NC shape mode frequency: where n is the shape mode (1 through 5), x n is the sample frequency for shape mode n, and x NC,n is the NC frequency for shape mode n.
Circularity was calculated with Equation (2): where A is the area and P is the perimeter for each microglia.

| Statistical analysis
All statistical analyzes were carried out in GraphPad Prism (GraphPad Soft-

| OGD time-dependent severity
HI duration can drastically change disease outcomes. 40 Therefore, we first investigated the impact of OGD time on cell viability and the oxidative stress environment. Compared to NC slice cytotoxicity of 11.2%, 0.5, 1.5, and 3 hr OGD exposure times resulted in significant increases in cytotoxicity of 54.3, 33.8, and 32.9% respectively ( Figure 1a). The 0.5 hr OGD cytotoxicity was also significantly higher than that of 1.5 or 3 hr OGD. 0.5, 1.5, and 3 hr OGD exposure times also yielded significantly decreased GSH concentrations of 1.7-fold, 4.1-fold, and 2.3-fold reductions, respectively ( Figure 1b).

| Therapeutic effect of SOD and AZ on OGDinduced injury
Prior to the investigation of AZ effects on nanoparticle interaction with microglia, we evaluated the effect of AZ and SOD on OGDinduced cytotoxicity to confirm therapeutic effects seen in literature.
We have previously shown SOD can attenuate excitotoxic damage in OWH brain slices. 15 SOD addition to 3 hr OGD exposed slices significantly reduced cytotoxicity of 17.1% (p = .036) ( Figure 2a). Having observed the greatest cytotoxicity induced by 0.5 hr OGD compared to NC, we proceeded to investigate the cytotoxicity effect of AZ on 0.5 hr OGD exposed slices. AZ treatment significantly reduced cytotoxicity to 14.0% (p < .001) ( Figure 2b). AZ treatment also significantly increased GSH concentration 1.5-fold compared to that of 0.5 hr OGD (p = .013) (Figure 2c).

| OGD and AZ effects on microglial shape as determined by VAMPIRE
To better understand microglial response to OGD injury, we characterized microglial morphological heterogeneity across disease states.
Microglial morphology and heterogeneity are dependent on the disease environment and are one indicator of microglial phenotype and function. 41,42 Relevant to this study, nanoparticle uptake has been correlated to microglial activation state, 9,43 with similar findings in liver-derived macrophages, where phenotype determined nanoparticle uptake. 44 Confocal images of microglia from each group were used to train a model via VAMPIRE, 26 which resulting in five distinct microglial shape modes and subsequent classification of each microglial cell into its closest matching shape mode. The frequency of the five shape modes for each group is the percentage of microglia that exhibited the given shape mode (Figure 3a). 0.5 hr OGD with AZ treatment resulted in a distribution of shape modes similar to that of NC microglia: shape modes 3, 4, 5, 1, and 2 in order of increasing frequency. Overall, 0.5 hr OGD showed a reduced spread in shape mode frequencies, indicating an increase in microglial shape heterogeneity since all five shape modes were more equally represented. The absolute difference of shape mode frequency from NC shape mode frequency was 4.5 and 2.2% for 0.5 and 0.5 h OGD + AZ, respectively, although the difference between the two groups was not significant (p < .001). NC and OGD + AZ circularity were also significantly different (p < .001). Furthermore, circularity did not correlate to shape mode, with a greater circularity for the 0.5 hr OGD condition for each of the five shape modes (Supplemental Table 1).  Table 1.

| OGD enhances nanoparticle diffusion through the brain ECS
We also anticipated that OGD exposure would alter the ECS through which nanoparticles can move. To estimate the distribution of effective ECS pores, we fit the Amsden obstruction-scaling model for entangled and cross-linked hydrogels 31 to the D eff data in  Figure 1A) and efficiency (Supplemental Figure 1B).

| Microglial uptake of nanoparticles is influenced by disease state and nanoparticle properties
We have confirmed OGD slice health can be recovered following AZ treatment and that particles can readily move within the OGD brain environment. We next sought to understand how injury and treatment influ-  at 24 hr after injury (Figure 6b). Interestingly, we observed a regional difference in microglial vulnerability. In the cortex, area covered by microglia did not change significantly across experimental condition ( Figure 6c), but microglial area in the thalamus was significantly reduced after OGD (p < .001). Treatment with AZ increased microglial coverage (p = .031 compared to OGD) in the thalamus, although still to a reduced level from NC (p < .001) (Figure 6d). Representative confocal images from each region are shown in Figure 6e.

| DISCUSSION
We investigated the nanoparticle-and disease-dependent nature of nanoparticle-microglia interactions using an OWH slice model of ischemic brain injury. Our first goal was to modulate ischemic brain injury severity by increasing OGD exposure times. There was an increase in cytotoxicity for all OGD exposure times compared to that of NC. Surprisingly, 1.5 and 3 h OGD exposure times resulted in lower cytotoxicity than 0.5 hr OGD exposure. Previous studies have demonstrated an increase in cytotoxicity from 0.5 to 1 hr of OGD ex vivo, and 60 min is regarded as the timeframe of OGD-induced neuronal swelling followed by apoptotic and necrotic death. [45][46][47][48] In our study, exposure to 0.5 hr OGD was sufficient to induce significantly different outcomes than the NC condition, in agreement with other findings. 18,49,50 Considering OGD mediates damage via oxidative stress, we measured the concentration of the redox buffering molecule GSH as another measure of slice health. 51 All OGD exposure times reduced GSH concentrations, representing an oxidatively stressed environment. 52 Microglia are potent cellular targets for drug delivery, due to their role propagating pathological processes after ischemic injury. Previous studies have demonstrated microglia-specific drug delivery with PAMAM dendrimers, which was attributed to the increased phagocytic behavior of microglia in an activated state. 6,12,53 Increased microglial uptake of dendrimers is also present after HI in mice in vivo, retinal HI injury in mice in vivo, and maternal inflammation-induced cerebral palsy rabbits in vivo and ex vivo. 9,[54][55][56] In all three models, microglia exhibited robust activation and proliferation, which was Microglial phenotypes in slice culture are also different than those represented in vivo 61 and several studies have shown that nanoparticle uptake is correlated to microglial phenotype. 43,62 Accumulating evidence shows the importance of shape in characterizing microglial phenotype, 63,64 supporting microglial characterization using shape modes independently or in tandem with other classification systems, such as surface marker presentation, transcriptomics, or cytokine expression. Improving microglial characterization is essential to understanding microglial-nanoparticle interactions, especially as the classical M1 and M2 microglial polarization phenotype schema progressively phases out. 65 To better characterize the range of microglial morphologies in our OGD slice model, we used computer-aided morphological analysis through application of the VAMPIRE software package. 0.5 hr OGD-induced changes in microglial shape modes, which were reversed back to NC shape mode distributions upon AZ treatment. AZ has previously been shown to promote an anti-inflammatory phenotypical change of macrophages and microglia. 21 Although there remained some variation in frequency for the five shape modes for 0.5 hr OGD + AZ compared to that of NC, the ranking of shape mode frequencies from least (3) to greatest (2) followed the same ranking as the NC microglia shape mode frequencies. The VAMPIRE package enables in-depth morphological analysis and detection of nuances in microglial shapes that encompass the heterogeneity of microglia that the human eye cannot detect. However, extracting intuitive shape characteristics of microglia from VAMPIRE is nontrivial given the complexity in interpreting the large dataset "machine-vision" classifications of the VAMPIRE package. 26 Independent of shape mode, one readily interpretable morphological distinction between groups was that of microglial circularity, where high circularity corresponds to a nonbranching amoeboid morphology characteristic of a proinflammatory activated state. 64 Although a value of one describes a perfect circle, some circularity values (1.2% of NC, 14.3% of 0.5 hr OGD, and 1.6% of 0.5 hr OGD + AZ) are slightly above 1 (1.12 max) due to errors in computer estimation of pixel perimeters. 66 We found that 0.5 hr OGD increased microglial circularity and AZ prevented amoeboid morphology, reducing circularity levels to closer to that of NC microglia. Further work may better determine the association between nanoparticle uptake and microglial phenotype by analyzing additional cell features, such as degree of branching, branching polarization, and soma size, in combination with transcriptomic analysis, especially for nanoparticle containing cells.
In contrast to dendrimers, PS-PEG nanoparticles exhibited increased microglial accumulation after OGD, and QDs were internalized at roughly equal proportions regardless of disease state. QDs were able to achieve orders of magnitude higher microglial accumulation after administration at the same dose as PS-PEG and D-Cy5. Our results indicate that microglial phagocytosis is highly dependent on nanoparticle platform, and disease-induced changes in microglial behavior are not leveraged equally among all nanoparticle types.
These platforms differ in size, rigidity, and chemical composition which can influence nanoparticle-cell interactions 67,68 suggesting that nanoparticle physicochemical parameters must be well-tuned to achieve accumulation in target cells at sites of injury. For example, rigid lipid nanoparticles could more easily pass through cell membranes compared to less rigid nanoparticles. 69 Previous work also supports microglial uptake of high-rigidity nanoparticles, namely gold nanoparticles in vitro and silica and QD nanoparticles ex vivo and in vivo. 13,70,71 Nanoparticle diffusive ability also plays a role in reaching target microglial cells and in achieving maximal therapeutic impact. 72,73 As evidenced by the MPT results after OGD, PS-PEG nanoparticles exhibited more than 16-fold higher diffusive ability compared to the NC condition. In this study, we directly confirmed diffusivity of PS-PEG, but PEGylated QDs and PAMAM dendrimers have also been shown to move effectively within the brain parenchyma. 8 While it is likely that some populations of particles are truly immobilized within pores, the Amsden obstruction model may be underestimating pore size by assuming nanoparticles are completely inert. PEGylated nanoparticles may interact with microglia, other cell types, and various components of the ECM, which hinders transport and may skew pore size estimates. Although the average pore size was smaller than previously calculated, the range of pore sizes identified in this study is similar to previous findings using MPT analysis in brain slices. 78  One benefit of the OWH slice platform for furthering this work is the ability to study regional variability in response to injury. After both 1.5 and 3 hr OGD, nanoparticles in the striatum had consistently faster D eff compared to those in the cortex. Microglial area coverage also indicated a greater injury response in the midbrain. Only the thalamus, not the cortex, showed a decrease in microglial area coverage after 0.5 hr OGD. Although we did not probe for a mechanism to explain regional differences, such an investigation can have important implications for therapeutic development. One recent study investigated the injury-resistant nature of the hypothalamus region and identified that slow neuronal depolarization in the region may be one native mechanism of neuroprotection. 79 Enhanced therapeutic penetration within diseased brain regions could reduce requisite dose amount and frequency and avoid inadvertent cytotoxicity on healthy tissue. Due to the intricate balance of pro-and anti-inflammatory activity in the brain microenvironment after disease, region specific control can also reduce over-scavenging of ROS or excessive inhibition of inflammatory processes that could interrupt healthy cellular function or exacerbate damage. [80][81][82] Given that ischemic injury manifests in regional patterns in multiple phases, 83,84 the continued investigation of regional variations in the brain could inform therapeutic strategies that are highly advantageous in combating immediate and ongoing regionally-dependent disease sequelae.
Further cross-platform investigation will be important to elucidate advantageous nanoparticle characteristics, in addition to diffusive ability, size, rigidity, and surface functionalization, for microglial-targeted drug delivery after injury. Microglial uptake of additional nanoparticle platforms, such as nanocrystals or polymeric micelles, has been understudied yet may elucidate new therapeutic avenues. Importantly, immediate AZ treatment of OGD-injured slices returned microglial uptake behavior of all nanoparticle types closer to that of NC conditions, suggesting that modulated microglia of recovered brain tissue behave similarly to healthy microglia. While this work primarily studied AZ modulation of slice health and microglial behavior, we also demonstrated that application of SOD or AZ is neuroprotective after ischemic injury. Brain slices treated with either therapeutic had significantly reduced cell death over 24 hr, and AZ additionally prevented a decrease in GSH concentration, indicative of an inhibition of oxidative stress and cell damage. SOD efficacy is well defined by scavenging of superoxide anion, which plays a damaging role in excitotoxicity, but the exact mechanism of AZ therapeutic efficacy remains to be elucidated in the brain. 85,86 Regardless of therapeutic mechanism, therapeutic reversion of OGD-induced changes offers promising implications for drug delivery strategies. If SOD or AZ were delivered via rigid carriers similar to PS-PEG nanoparticles that exhibited increased diffusion and microglial uptake after OGD exposure, drug distribution would favorably accumulate in diseased regions. After therapeutic release and reduction of disease phenotype, subsequently administered nanoparticle doses would preferentially sequester in ongoing injury sites compared to recovering or healthy tissue environments. Further investigation of strategies to modulate nanoparticle diffusivity and microglial uptake may result in more effective methods of nanoparticle-mediated therapeutic delivery.

| CONCLUSION
In this work, we probed the effect of OGD-induced brain injury and AZ treatment on nanoparticle interactions with microglia. First, we determined the effect of OGD exposure on markers of injury severity, including cell death and oxidative stress. We observed significant injury responses after 0.5 hr OGD exposure: a 54.3% increase in cytotoxicity, a 1.7-fold decrease in GSH concentration, and a larger pore distribution in the ECS. We observed an OGD-induced shift in microglial morphology toward more heterogeneity in shapes with overall increased circularity and a decrease in microglial density. Nanoparticle interactions with microglia were dependent on both the nanoparticle platform as well as treatment condition. After 0.5 hr OGD, microglial internalization of PS-PEG was increased, but uptake of QDs or dendrimers was not enhanced, indicating an important role of nanoparticle material identity in determining extent of phagocytosis after injury. OGD injury did not impede nanoparticle mobility; PS-PEG nanoparticles had a 16.7-fold increase in diffusion after 0.5 hr OGD compared to that of NC. Treatment with AZ not only effectively reduced OGD cytotoxicity and GSH depletion, but also reverted nanoparticle uptake behavior of PS-PEG and microglial morphology toward that of NC. This study shows OWH slices enabled region-dependent nanoscale probing of live tissue to identify cellular and microenvironmental changes in diseased and recovering brain that can be leveraged for cell-specific uptake of nanoparticles. In addition, we demonstrate that in ischemic conditions, nanoparticle fate is platformdependent, providing insights into therapeutic strategy for targeting microglial cells to combat neurological disease.

ACKNOWLEDGMENTS
We would like to thank Dr Rangaramanujam Kannan