Mitochondrial network adaptations of microglia reveal sex-specific stress response after injury and UCP2 knockout

Summary Mitochondrial networks remodel their connectivity, content, and subcellular localization to support optimized energy production in conditions of increased environmental or cellular stress. Microglia rely on mitochondria to respond to these stressors, however our knowledge about mitochondrial networks and their adaptations in microglia in vivo is limited. Here, we generate a mouse model that selectively labels mitochondria in microglia. We identify that mitochondrial networks are more fragmented with increased content and perinuclear localization in vitro vs. in vivo. Mitochondrial networks adapt similarly in microglia closest to the injury site after optic nerve crush. Preventing microglial UCP2 increase after injury by selective knockout induces cellular stress. This results in mitochondrial hyperfusion in male microglia, a phenotype absent in females due to circulating estrogens. Our results establish the foundation for mitochondrial network analysis of microglia in vivo, emphasizing the importance of mitochondrial-based sex effects of microglia in other pathologies.


INTRODUCTION
Mitochondria have been classically defined as autonomous organelles that support eukaryotic cell function by providing energy through cellular respiration. 1][4][5] In line with these observations, changes in cellular metabolism and mitochondrial-associated gene signatures have been frequently associated to pathological conditions, highlighting mitochondria as key players in cell signaling and disease. 2 In physiological conditions, mitochondrial networks constantly remodel through the process of organelle fission and fusion, termed mitochondrial dynamics. 1Mitochondrial fusion supports matrix content exchange and/or enhanced ATP production, while fission assists redistribution and elimination of dysfunctional organelles. 6Cellular stress, such as increased reactive oxygen species (ROS), leads to imbalanced mitochondrial dynamics often resulting in a more fragmented network.This is reflected in an increased number of small-volume, spherical mitochondria 7 and expedites removal of damaged mitochondria in a process termed mitophagy.To compensate for mitophagy, mitochondrial biogenesis occurs which induces protein synthesis allowing the organelles to increase in size, leading to transient alterations in cellular mitochondrial content. 83][14] Therefore, the adaptations of mitochondrial network connectivity, content, and localization provide valuable insight into cellular stress and energy demands.
Degenerative disease environments increase cellular stress, and failure to mitigate this stress results in its accumulation and cellular damage. 15This is particularly relevant in the context of immune cells, whose primary role is to detect and rapidly react to perturbations in their environment. 2,16,179][20] Several studies have shown that mitochondria are essential to microglial surveillance and response.Mitochondrial network fragmentation is required for the metabolic shift to glycolysis in responsive microglia in vitro. 21This was corroborated in cell isolation studies where aberrant metabolic reprogramming was identified as the cause for microglial dysfunction in Alzheimer's mouse models. 20Microglia are able to adapt to available energy sources to maintain their surveillance functions, 22 however in germ-free mice where fuel sources are scarce, their mitochondrial function diminishes which is reflected in increased mitochondrial mass, ROS and reduced numbers. 23Therefore, mitochondrial network adaptations are necessary for effective microglial responses, and provide important insights into cellular stress and metabolic changes that could predict disease-related phenotypes in the neuronal environment.
7][28] Furthermore, spatial information and environmental factors such as sex are lost in cell isolation studies, which are important contributing factors known to influence microglia. 29A major limitation in addressing mitochondria in microglia in vivo is the lack of tools to effectively visualize mitochondrial networks specifically in microglia.
Here, we generated a mouse model that selectively labels mitochondria in microglia.First, we established the differences between in vitro and in vivo mitochondrial networks.Then, we analyzed spatially isolated microglial populations in the brain using the retina as a model.1][32] We show that after ONC, microglia have more fragmented mitochondrial networks with increased content per cell and altered localization in IPL microglia, while OPL microglia further from the dying neurons only minimally respond.Next, we manipulated cellular stress in microglia by selectively depleting uncoupling protein 2 (UCP2), a mitochondria-localized gene associated with stress mitigation and microglial function. 33,34UCP2-depleted IPL microglia responded similarly to wildtype after ONC with the exception of mitochondrial number.We uncovered a sex-specific difference in stress mitigation, where male UCP2 KO microglia exhibited hyperfused mitochondrial networks after ONC.Ovariectomized UCP2 KO females represented a similar mitochondrial phenotype suggesting that circulating estrogens influence this effect, and highlighting that there are sex-based differences in microglial stress mitigation.

Mitochondrial networks of microglia differ between in vitro and in vivo environments
Visualizing mitochondrial networks of retinal microglia by immunostaining for markers such as TOMM20 (translocase of outer mitochondria membrane 20) results in a dense and ill-defined labeling in the OPL and IPL, where microglia predominantly reside (Figure 1A).Thus, we took advantage of the PhAM fl/fl mouse, which carries a floxed transgene encoding a mitochondrial-localized Dendra2 fluorophore. 35We crossed this mouse with the Cx3cr1 CreERT2 mouse expressing tamoxifen-inducible Cre recombinase in cells of myeloid origin. 36After tamoxifen administration, Cx3cr1 CreERT2/+ /PhAM fl/fl mice, hereafter referred to as wildtype (WT) mice (Figure 1B), microglia labeled with IBA1 (ionized calciumbinding adaptor molecule 1) showed selective expression of the mito-Dendra2 fluorophore (Figure 1C).We verified mito-Dendra2 localization at the mitochondria of microglia by co-labeling with TOMM20 in retinal sections (Figure 1D), as well as in 4-hydroxytamoxifen treated primary mixed glial cultures prepared from WT pups (Figure 1E).
Primary microglia in mixed glial culture differ in their morphological shape compared to microglia in vivo (Figures 1F and 1G).To evaluate the mitochondrial network in both conditions, we generated 3-dimensional (3D) surfaces of both mitochondria and microglia and performed volumetric analyses (Figures 1F, 1G, and S1A, Videos S1, and S2).First, we quantified mitochondrial network connectivity for single cells in which we determined the median mitochondrial volume amongst all organelles of a cell (Figure S1B), the number of mitochondria, and the mean sphericity of mitochondria (Figure S1C).Microglia in vitro exhibited a reduced median mitochondrial volume and increased total number of mitochondria compared to in vivo (Figure 1H), indicating a more fragmented mitochondrial network, even though the sphericity did not differ.When we analyzed the mitochondrial content per cell (Figure S1D), we found that microglia in vitro had a significantly greater content compared to in vivo (Figure 1I), suggesting an imbalance of biogenesis and turnover between conditions, which can be associated with greater oxidative damage. 37To assess subcellular localization of mitochondria, we measured the distance from the centroid of each mitochondrion to the cell soma (Figure S1E) and represented the data in a scatterplot (Figure S1F).We found a greater percentage of a cell's total mitochondrial volume was localized in the perinuclear region closer to the soma in vitro (Figures 1J and S1G).Collectively, these mitochondrial parameters Wilcoxon rank-sum test: p < 0.0001.**p < 0.01, ***p < 0.001, n.s.p > 0.05: not significant.See Table S1 for experimental, retina and cell numbers, statistical tests and corresponding data.

IPL OPL
demonstrate a difference between in vitro and in vivo conditions that align with a greater stress response in microglia in vitro.To substantiate this, we analyzed the expression of the endosomal-lysosomal marker CD68 (cluster of differentiation 68) in microglia (Figures S2A and  S2B),which is upregulated in reactive microglia. 38,39The CD68 volume was significantly greater in vitro and coincided with increased vesicle volumes distributed along the length of processes when compared to microglia in vivo (Figures S2C and S2D).Together, this data emphasizes that microglia in vitro are distinct from microglia in vivo, not only morphologically, but also at the mitochondrial level.

Mitochondrial network alterations occur prominently in microglia proximal to dying neurons
To identify how mitochondrial networks of retinal microglia in vivo adapt to stress conditions such as neuronal death, we took advantage of the optic nerve crush (ONC) model (Figure 2A). 30,40Here, a mechanical injury at the retinal ganglion cell axons posterior to the optic disc induces their apoptosis, 30,41 which we confirmed by an increased number of cleaved-caspase 3 + /RBPMS + -retinal ganglion cells (Figure 2B).Microglia showed morphological adaptation to these environmental changes throughout the IPL after ONC (Figure 2C).At the single cell level, microglia increased expression of CD68 (Figures S3A-S3C), where vesicles were distributed throughout the length of the processes (Figure S3D).This increase was corroborated by increased Cd68 transcript expression in FACS-sorted retinal microglia after ONC (Figure S3E).Accompanying this phenotype, microglia reduced their branching complexity, as verified by the increased number of Sholl intersections closer to the soma (Figure S3F).Next, we aligned mitochondrial network adaptations to this responsive microglial phenotype (Figures 2D  and 2E).Relative to the naive environment, microglia showed significantly reduced median mitochondrial volume and an increased number of mitochondria with greater sphericity after ONC (Figure 2F).Together with increased mitochondrial content (Figure 2G), and the subcellular localization of the mitochondrial population closer to the soma (Figure 2H), this data indicates a more fragmented mitochondrial network with altered content and localization in responsive IPL microglia.
To determine whether microglia residing more distant from the apoptotic retinal ganglion cells show a similar effect, we repeated the above analysis in microglia of the OPL.OPL microglia showed mild morphological adaptations (Figure S3G).Upon single cell analysis (Figures 2I and 2J), OPL microglia had significantly reduced median mitochondrial volume, whereas the number and the sphericity of mitochondria, and total content remained unaltered (Figures 2K and L).Similar to IPL microglia, the subcellular mitochondrial localization shifted toward the soma (Figure 2M), CD68 volume increased (Figures S3H-S3K), and the microglial branching complexity was moderately altered (Figure S3L).This data indicates that proximity to apoptotic neurons is a key determinant in the robustness of the mitochondrial-microglial response, where more distant microglia also exhibit some mitochondrial network alterations.
Selective microglial UCP2 KO increases cellular and mitochondrial stress with no effect on the mitochondrial network Electron leakage during oxidative phosphorylation leads to the formation of mitochondrial superoxides and reactive oxygen species (ROS).Strategies to mitigate ROS include the induction of mild uncoupling to increase the respiration rate and reduce electron leak, or detoxification of superoxides with cellular enzymes such as superoxide dismutase 1 (SOD1). 42However, these mitigation strategies are less effective in conditions of cellular damage or mitochondrial dysfunction, and can lead to ROS accumulation. 435][46][47][48] Upregulation of Ucp2 transcript has been reported in microglia within disease environments, 49,50 and we confirmed increased Ucp2 transcript expression in microglia five days after ONC (Figures S4C and S4D), making Ucp2 a candidate gene to disrupt mitochondrial function and alter ROS accumulation.Welch's t-test: p = 0.0045.**p < 0.01, ***p < 0.001, n.s.p > 0.05: not significant.See Table S2 for retina and cell numbers, statistical tests and corresponding data.

ll OPEN ACCESS
Loss of UCP2 has been reported to elicit cellular ROS accumulation. 51,52Thus, to evaluate this consequence on the mitochondrial network in retinal microglia, we generated microglia-specific UCP2 KO mice by crossing the Ucp2 fl/fl mouse with our mitochondrial-labeled WT mice (Cx3cr1 CreERT2/+ /PhAM fl/fl /Ucp2 fl/fl , Figure 3A).This model, hereafter referred to as UCP2 KO mice, showed microglia-selective mitochondrial labeling (Figure S4E).We confirmed successful Ucp2 knockout at both the transcript (Figure S4F) and protein level (Figures S4G-S4I) in which we performed qRT-PCR of FACS-isolated Dendra2 + -retinal microglia three weeks after tamoxifen induction and Western blot analysis of microglia after 4-OHT (4-hydroxytamoxifen) treatment in primary mixed glial cultures, respectively.To determine the consequences of UCP2knockout on microglial ROS levels, we first compared mitochondrial superoxide in retinal microglia from naive WT and UCP2 KO mice using FACS-sorted MitoSOX stained cells.UCP2 KO mice exhibited a significantly increased percentage of MitoSOX + -microglia (Figure 3B).This coincided with significantly enhanced gene transcript expression of the detoxifying enzyme Sod1 (Figure 3C), together indicating elevated mitochondrial and cellular ROS in UCP2 KO microglia.When we compared the mitochondrial networks of UCP2 KO to WT in the naive condition, we did not detect differences in mitochondrial network connectivity, content, or localization (Figures 3D-3G).Interestingly, the overall expression of CD68 was slightly elevated in UCP2 KO microglia (Figures 3H-J) and aligned with increased CD68 transcript (Figure 3K), whereas the microglial morphology remained unaffected (Figure 3L).Together, this data indicates increased cellular stress in UCP2 KO microglia with no direct effect on the mitochondrial network in the naive environment.

UCP2 KO allows microglial response and mitochondrial adaptations after optic nerve crush
Based on previous studies reporting that UCP2 KO prevents mitochondrial fission and microglial reactivity after high-fat diet, 34,53 we anticipated no mitochondrial fragmentation or microglial response in the ONC injury model.Unexpectedly, ONC UCP2 KO microglia showed increased CD68 volume, where vesicles were localized throughout the length of processes (Figures 4A-4D), and reduced branching complexity (Figure 4E) compared to naive UCP2 KO .This aligned with the mitochondrial network becoming more fragmented as reflected in the reduced median mitochondrial volume and increased sphericity (Figures 4I-4K).At the same time, the number of mitochondrial organelles surprisingly remained similar (Figure 4H, center) even though the mitochondrial content was increased (Figure 4I).This deviation in mitochondria number did not align with the expected changes for mitochondrial connectivity seen in WT ONC microglia (Figures 2K and 2L), indicating differences in UCP2 KO mitochondrial network adaptations.

Loss of UCP2 KO elicits a sexually dimorphic microglial response in ONC
Next, we looked more closely at the relationship between conditions (naive, ONC) and genotypes (WT, UCP2 KO ) using principal component (PC) analysis.PC analysis allows unbiased visualization of six parameters for each analyzed microglia (Figure S1 and STAR Methods) and identification of patterns within the dataset.In naive conditions, WT and UCP2 KO microglial profiles intermingled in the same PC space (Figure 5A), corroborating the results from Figure 3. Naive and ONC microglial profiles separated along the first PC for both WT and UCP2 KO , aligning the with microglial-mitochondrial response in Figures 2 and 4.However, the ONC microglial profiles showed less prominent intermingling  S3 for retina and cell numbers, statistical tests and corresponding data.between genotypes, where UCP2 KO showed greater spread along the second PC (Figure 5A).A recent study indicated that UCP2 loss in microglia results in anxiety phenotypes in male mice, 52 therefore we annotated sex as an additional factor in the PC space.Whereas sexes were intermingled for WT profiles in both naive and ONC conditions (Figure 5B), UCP2 KO microglial profiles diverged along the second PC only after ONC (Figure 5C), suggesting a sexually dimorphic response to ONC in UCP2 KO microglia.

Male UCP2 KO microglia induce mitochondrial hyperfusion to mitigate stress
To identify which of the six parameters differed between sexes, we first referenced the PC loadings where the highest contributing factor along the second PC was mitochondrial content (Figure S5A).When separating the volumetric analyses by sex, UCP2 KO males showed significantly higher mitochondrial content per cell compared to female ONC UCP2 KO conditions (Figure S5B).Additionally, male ONC UCP2 KO exhibited a greater median mitochondrial volume and reduced mitochondrial number compared to female ONC UCP2 KO microglia, aligning with a less fragmented network (Figure S5C).To resolve these sex-based differences, we revisited the UCP2 KO microglial and mitochondrial images (Figures 5D and 5E, and Video S3) and separated the localization scatterplots.Here, we found that male ONC UCP2 KO microglia had a higher frequency of mitochondria with large volumes close to the cell soma relative to female ONC UCP2 KO (Figure 5F).When we compared the largest-volume mitochondrion per cell as a metric for highly connected or hyperfused mitochondria, we found that male UCP2 KO had significantly greater volumes than female UCP2 KO or WT after ONC (Figure 5G), indicating a sexually dimorphic hyperfused mitochondrial phenotype in UCP2 KO microglia.Previously, mitochondrial hyperfusion has been described as a transient phenotype in stress-induced environments that protects cellular and mitochondrial health. 12,13To investigate whether the observed hyperfused mitochondrial networks occurred in response to differences in ROS levels between sexes, we quantified MitoSOX + -microglia after ONC in UCP2 KO for each sex.Males and females showed a comparable level of mitochondrial superoxide (Figure 5H), suggesting a similar production of intrinsic mitochondrial ROS.Interestingly, we detected opposing trends in Sod1 transcript expression in UCP2 KO microglia after ONC, where males had significantly greater Sod1 expression compared to females after ONC (Figure 5I), indicating differences between sexes in the strategy for superoxide detoxification.This difference was more pronounced in WT microglia after ONC, where only males significantly increase Sod1, yet the production of mitochondrial superoxide remained similar in females (Figures S5D and S5E), suggesting sex-based differences in microglial ROS mitigation.
5][56] Furthermore, both microglia and mitochondria express estrogen receptors. 57,58Since females did not implement stress-mitigation strategies via mitochondrial hyperfusion (Figures 5E-5G) or increased Sod1 transcription (Figures 5I and S5D), we evaluated whether circulating estrogens contribute to the sexually divergent mitochondrial phenotype in UCP2 KO after ONC.Thus, we repeated the experiment with ovariectomized UCP2 KO female mice (Figure 6A) and quantified the mitochondrial networks in naive and ONC (Figure 6B).The mitochondrial connectivity in ovariectomized UCP2 KO female microglia exhibited decreased median mitochondrial volume, no change in mitochondria number, and significantly increased mitochondrial sphericity (Figure 6C).Furthermore, the mitochondrial content and their percentage in the perinuclear region increased after ONC (Figures 6D and 6E), together indicating mitochondrial adaptations to ONC in ovariectomized UCP2 KO females.Remarkably, the mitochondrial distribution in ovariectomized ONC UCP2 KO females reflected a similar increase in frequency of large-volume organelles close to the microglia soma (Figure 6E) as observed for UCP2 KO males (Figure 5F).When we analyzed the largest-volume mitochondrion per cell, UCP2 KO ovariectomized females had similarly large volumes as UCP2 KO males (Figure 6F).
To substantiate that ovariectomized ONC UCP2 KO females exhibit a mitochondrial phenotype similar to males, we represented all UCP2 KO microglial profiles in the PC space.As anticipated, the ovariectomized UCP2 KO females separated along the first PC into the microglial profiles for the respective naive or ONC condition (Figure 6G).In the ONC condition, ovariectomized UCP2 KO females trended toward males  KO (left) and ONC UCP2 KO (right, gold) conditions.Point density, pseudo-colored.(E) 3D-filament tracings of microglia from naive UCP2 KO (left) or ONC UCP2 KO (right).Scale bar: 10 mm.Line plot for mean number of Sholl intersections per radial distant from the soma (mm) with 95% confidence interval band.Linear mixed effects model: p < 0.0001.(F-J) Mito-Dendra2 (green) expression in the IBA1-immunostained microglia from (A-B) for naive UCP2 KO microglia (F) or ONC UCP2 KO microglia (G).Next, corresponding 3D-surfaces.Below: Zoom-in of region of interest (dashed line) from image and 3D-surface.Scale bar: 10 mm, zoom-in: 3 mm.(H-J) Mitochondrial parameters.Boxplot minimum and maximum: InterQuartile Range (IQR) around median (center line).Whiskers: 1.5 IQRs.Black diamond: outliers outside of 1.5 IQRs.(H) Mitochondrial network connectivity determined by median mitochondrial volume (left, Wilcoxon rank-sum test: p < 0.0001), number of organelles within a single cell (center, Wilcoxon rank-sum test: p = 0.930881) and mean mitochondrial sphericity (right, Student's t test: p < 0.0001).(I) Percentage of mitochondrial volume per microglial volume.Wilcoxon rank-sum test: p = 0.0003.(J) Mitochondrial localization.Scatterplot depicting mitochondria volume vs. distance from the cell soma (0, origin) of the population of mitochondria in naive UCP2 KO (left) or ONC UCP2 KO (right) microglia.Point density, pseudo-colored.Percentage of total mitochondrial volume localized within the perinuclear region.Student's t test: p < 0.0001.***p < 0.001, n.s .p > 0.05: not significant.See Table S4 for retina and cell numbers, statistical tests and corresponding data.along the second PC (Figure 6H).Aligning with the ONC-responsive mitochondrial phenotype, ovariectomized UCP2 KO females showed increased CD68 volume with greater distribution and reduced branching complexity (Figures 6I-6L).Together, this data illustrates that circulating estrogens play a role in maintaining mitochondrial networks in female microglia, and are a contributing factor to the sexually divergent stress-mitigation strategies in UCP2 KO microglia.

DISCUSSION
Our results provide a new perspective of mitochondrial networks in microglia in vivo, and demonstrate their adaptation to an injury environment and increased cellular stress.Importantly, we established sex-based differences in microglial stress mitigation strategies, which we uncovered at the mitochondrial network level upon UCP2 knockout.
Extensive in vitro studies have shown that cellular energy demands of immune cells critically depend on balanced mitochondrial networks equilibrated by mitochondrial fission and fusion. 16Insight into these dynamics in microglia in vivo are mostly unknown due to technical limitations.Tissue-wide mitochondrial immunostaining requires advanced image segmentation techniques to decipher which mitochondria are within or outside the microglia. 23This necessitates high resolution images and experience in how to determine cell edges to avoid userbiases. 59In the retina, we performed high resolution imaging of TOMM20 immunostaining and were challenged to perform image segmentation of microglial mitochondria for both the retina (Figure 1D), and in 2D-mixed primary glial culture (Figure 1E).In contrast, our mouse model allowed clear image segmentation of microglial mitochondria (Figures 1C, 1F, and 1G).Alternative techniques such as electron microscopy can be used to assess cross-sectional mitochondrial areas, 34,53 however resolving the entire mitochondrial network of an intact microglia would require time-and labor-intensive serial sectioning followed by image segmentation. 60Our mouse model visualizes the mitochondrial network of an entire microglia in vivo and allowed us to resolve elongated, tubular organelles in naive condition, which was absent in vitro (Figures 1F and 1G). 21,24Our results in ONC-responsive microglia in vivo corroborated a study reporting increased mitochondrial mass in responsive microglia isolated from brain tissue, 23 indicating that results in vivo are achievable without microglial isolation and importantly, maintain the spatial and morphological information of the cell.To this end, we identified regional differences in mitochondrial network adaptations between microglia localized in the IPL and OPL after ONC, which are not attainable in cell-dissociation or in vitro studies (Figure 2).Together, this emphasizes that our mouse model provides a valuable strategy for assessing the mitochondrial network of microglia in a region-specific manner without cell-isolation.
Mitochondrial networks are most commonly assessed using connectivity parameters, including organelle volume, number, and sphericity. 61We found that these three parameters reliably identified fragmented networks in ONC-responsive IPL microglia (Figure 2), which was also the case when two of the three parameters were altered, highlighting the importance of using more than one metric to evaluate mitochondrial connectivity in the 3D environment.As an additional parameter, we also assessed the subcellular localization of each mitochondrion within individual microglia.Previous studies have shown that responsive microglia enhance their transcriptional activity, 62 a cellular feature facilitated by perinuclear localization of mitochondria. 10We consistently detected an increased fraction of the mitochondria in the perinuclear region across genotypes and sexes after ONC (Figures 2, 4, 6, and S3).Interestingly, this localization was one of the only two mitochondrial parameters that significantly changed in OPL microglia (Figure 2M).The second parameter was decreased median mitochondrial volume (Figure 2K), which is indicative of mitochondrial fragmentation.Thus, both parameters suggest initial mitochondrial adaptations in microglia after ONC in the OPL niche more distant from the injury site.The combination of subcellular localization and the mitochondrial content per cell were critical to identify the hyperfused mitochondrial network (Figure 5).4][65][66] We found that this phenotype also occurs in microglia in vivo, which we identified with significant increases in large-volume mitochondria (Figures 5F-5H).This metric was most informative for identifying Figure 5. Continued number of mitochondria, % mitochondrial volume per microglia, % mitochondria perinuclear obtained from aforementioned genotype and condition (Figure S1).PC1 and PC2 describe 49% and 20% of the explained variance across the population, respectively.Each subpanel highlights the following comparisons: (A) Naive microglial profiles for WT (gray) and UCP2 KO (tan) and ONC microglial profiles for WT (blue) and UCP2 KO (gold).(B-C) Naive and ONC microglial profiles in male (turquoise) and female (purple) for WT (B) and for UCP2 KO (C).Dashed line: Reference of distributions shown in (A).(D and E) Representative 3D-surface renderings of IBA1 (magenta) and mito-Dendra2 (statistics-based volume spectral coloring) microglia from ONC-induced male UCP2 KO (D) and female UCP2 KO (E).Right: corresponding confocal images of IBA1-immonostained (magenta) microglia and mito-Dendra2 (green) expression.Below: Zoom-in of region of interest (dashed outline) of mito-Dendra2 (green) and corresponding 3D-surface with volume spectral coloring.Scale bar: 10 mm, zoom-in: 3 mm.Volume spectrum: 0 mm 3 (blue) -10 mm 3 (red).(F) Mitochondrial localization.Scatterplot depicting mitochondria volume vs. distance from the cell soma (0, origin) of the population of mitochondria in ONC UCP2 KO male (left) or female (right) microglia.Point density, pseudo-colored.(G) Boxplot of largest mitochondrion per cell separated by sex.Boxplot minimum and maximum: InterQuartile Range (IQR) around median (center line).Whiskers: 1.5 IQRs.Black diamond: outliers outside of 1.5 IQRs.Naive WT: gray.ONC WT: blue.ONC UCP2 KO : gold.Kruskal-Wallis test: p < 0.0001.Selected Conover's post-hoc comparisons with Holm p-adjustment: ONC _WT vs. _UCP2 KO , p < 0.0001; ONC _WT vs. \WT, p = 1.0;ONC _UCP2 KO vs. \UCP2 KO , p = 0.0387; ONC \WT vs. \UCP2 KO , p = 0.1364.(H) Frequency plot of MitoSOX fluorescence from FACSed mito-Dendra2 + retinal microglia for ONC UCP2 KO male (turquoise) and female (purple).Corresponding bar plot of percentage of microglia that are mitoSOX + .Student's t test: 0.6707.(I) Bar plot depicting relative Sod1 transcript expression from FACSed mito-Dendra2 + microglia in UCP2 KO naive and ONC microglia.Student's t test: p = 0.0377.*p < 0.05, n.s.p > 0.05: not significant.See Table S5 for retina and cell numbers, statistical tests, other post-hoc comparisons and corresponding data.hyperfusion in microglia since the unique, 3D morphology of microglia makes it difficult to use a subjective classification method without quantification to identify hyperfused networks, which has been the common methodology for defining mitochondrial hyperfusion in in vitro studies. 12,13,63ur approach to evaluate the effects of increased cellular stress on mitochondrial networks in vivo was to endogenously manipulate stress levels through selective-knockout of UCP2, a negative regulator of ROS.UCP2 KO microglia showed increased intrinsic stress and CD68 expression in the naive environment (Figures 3B, 3C, and 3I-3K), aligning with previous literature. 34,52Although these stress levels were increased, we did not detect any significant mitochondrial network alterations in the naive environment (Figure 3).It is possible that mitochondrial adaptations occurred earlier after knockout.However, our model limited us to a three week timepoint after UCP2-depletion when circulating Cx3Cr1 + -monocytes have repopulated. 67This ensures that UCP2-knockout and mitochondrial-labeling is restricted to the resident microglia and thus confirms that our mitochondrial analysis is exclusive to this population in the event of monocyte infiltration after ONC injury.Nevertheless, we did detect mitochondrial fragmentation and a microglial response in the UCP2 KO ONC environment (Figures 4I and 4J-4L), which was surprising since previous studies reported that UCP2 loss prevented these changes in the hypothalamus by assessing crosssectional area and number of mitochondria in electron microscopy preparations. 34,53On the other hand, Kim et al. 34 reported the prevention of mitochondrial fragmentation only in male mice, while females were largely unaffected in diet-induced obesity paradigms and their mitochondrial phenotypes were not reported.Thus, our results align with these previous reports and refine that UCP2 loss affects mitochondrial networks in a sex-specific manner, such that UCP2 KO induces hyperfusion in male microglia after ONC, while female UCP2 KO microglia still exhibit mitochondrial fragmentation.

A Experimental timeline for ovariectomy
Since male and female microglia responded differently to increased stress via UCP2 KO (Figures 5 and S5), we suspect that estrogens may explain the difference in the female response.Circulating estrogens provide a protective role for neurons under oxidative stress or in injury conditions 55,68 and enhance antioxidant gene expression in female mitochondria. 69,70Indeed, in absence of circulating estrogens, both male and ovariectomized female microglia rely on the same mechanism of mitochondrial hyperfusion to mitigate excess stress from UCP2 KO 12,13 .On the other hand, estrogen may not be the only mechanism.In the absence of circulating estrogens in vitro, female but not male neurons were shown to utilize lipids as a pro-survival fuel source, 71 and in a separate study, it was shown that female microglia support protection from diet-induced obesity by an estrogen-independent increase in CX3CR1 signaling. 72This suggests that sex differences in microglial stress responses may be driven by a combination of circulating estrogens and dimorphisms in transcriptional programs. 57,58,73,74n conclusion, our study highlights the substantial importance of sex-mediated effects in microglia which are reflected in the mitochondrial network, and provides a foundation for mitochondrial network analysis of microglia in vivo.(F) Largest mitochondrion per cell separated by sex.One-way ANOVA: p = 0.0088.Selected Tukey's post-hoc comparisons: ONC _UCP2 KO vs. ONC \UCP2 KO , p = 0.0065; ONC _UCP2 KO vs. ONC \ ov UCP2 KO , p = 0.3297, ONC \UCP2 KO vs. ONC \ ov UCP2 KO , p = 0.1938.(G) Principal component analysis (PCA) of microglia from male, female and ovariectomized UCP2 KO microglia in naive and ONC conditions.The first two principle components (PC) are visualized.Naive and ONC UCP2 KO microglial profiles in male (turquoise), female (purple) and ovariectomized females (orange x-marker).Dashed outline: Reference of distributions shown in (Figure 5A).(H) Comparison of ovariectomized ONC UCP2 KO female (orange x-marker) microglial profiles with male (turquoise, left) or female (purple, right) ONC UCP2 KO microglial profiles.(I) Ovariectomized female ONC UCP2 KO IBA1-immunostained microglia (magenta) from Figure 6B co-labeled with CD68 (cyan) and corresponding 3D-surfaces.Below: Zoom-in of region of interest (dashed line) from image and 3D-surface.Scale bar: 10 mm, zoom-in: 3 mm.(J) Percentage of total CD68 volume per microglia volume in WT and UCP2 KO ovariectomized females in naive and ONC conditions.Boxplot minimum and maximum: InterQuartile Range (IQR) around median (center line).Whiskers: 1.5 IQRs.Black diamond points: outliers outside 1.5 IQRs.Wilcoxon rank-sum test: p < 0.0001.(K) CD68 vesicle localization.Scatterplot depicting CD68 vesicle volume vs. distance from the cell soma (0, origin) of the population of vesicles from ovariectomized females in UCP2 KO naive (left) and UCP2 KO ONC (right) conditions.Point density, pseudo-colored.(L) 3D-filament tracings of microglia from UCP2 KO naive (left) or ONC (right).Scale bar: 10 mm.Line plot for mean number of Sholl intersections per radial distant from the soma (mm) with 95% confidence interval band.Linear mixed effects model: p = 0.0018.*p < 0.05, **p < 0.01, ***p < 0.001, n.s.p > 0.05: not significant.See Table S6 for retina and cell numbers, statistical tests and corresponding data.

Optic nerve crush (ONC) procedure
Mice were anesthetized in an induction chamber with 5% (v/v) isoflurane (Zoetis) supplied with oxygen at a flow rate of 0.6 L/min.After lack of a foot pinch reflex, mice were maintained at 2.5% (v/v) isoflurane applied through a nose cone while on a heating pad to maintain body temperature at 37 C. Proparacaine hydrochloride 0.5% ophthalmic eye drops (Ursapharm Arzneimittel GmbH) were applied to numb the eyes, and subcutaneous injection of 5mg/kg Metacam alleviated pain (Meloxacam, Boehringer Ingelheim).The lateral canthus was de-vascularized by clamping with a hemostat (Fine Science Tools) for 10 seconds.Using a Leica dissection microscope, a lateral canthotomy allowed visualization of the posterior pole.While firmly holding the conjunctiva with a jeweler forceps, the conjunctiva was cut perpendicular to the posterior pole.The surrounding muscle was carefully dissected as to not puncture the vascular plexus.The optic nerve was pinched 1mm from the posterior pole for 4 seconds using a curved N7 self-closing forceps (Dumont).Triple antibiotic ointment was applied to the eye directly after the surgery to prevent infection.

Ovariectomy
Cx3cr1 CreERT2/+ /PhAM fl/fl or Cx3cr1 CreERT2/+ /PhAM fl/fl /Ucp2 fl/fl prepubescent postnatal day 20 female mice were anesthetized using 5% (v/v) isofluorane supplemented with oxygen at a flow rate of 0.6L/min in an induction chamber. 81Mice were maintained in 2% (v/v) isoflurane with the same flow rate after absence of a foot pinch reflex.Above the lumbar spine, the skin was exposed by shaving the fur using an electric razor, then sterilized with 70% (v/v) ethanol.A 1cm midline incision on the lower back allowed gentle dissection of the subcutaneous tissue to expose the muscular fascia and ovarian fat pad.A small incision into the peritoneal cavity for entry exposed the fallopian tube, which was used as a guide to identify the ovary.The ovary was removed via cauterization of the oviduct and blood vessels to prevent bleeding.The muscular fascia was sutured after the fallopian tube was replaced into the peritoneal cavity.After repeating the procedure to remove the contralateral ovary, the midline incision of the skin was sutured.Mice received subcutaneous injection of Metamizol (Sanofi Avenis, 200 mg/kg) during the surgical procedure and meloxicam (Boehringer-Ingelheim, 5mg/kg) to alleviate pain.

Primary microglia isolation
Two T75 flasks containing mixed primary glia cultures prepared in tandem were treated with 1mM 4-hydroxy tamoxifen (Sigma-Aldrich, SML1666-1ML) or vehicle at day 10.Three days later, microglia were isolated using mild trypsinization (0.8mM CaCl 2 in 1xTrypsin-EDTA, ThermoFisher #25300-054) is added to the flask after a washing step with pre-warmed 1X PBS to remove inhibiting serum, then incubate at 37 C for 10-20 minutes, or until the non-microglial cells have just lifted from the flask forming a floating layer.The floating cells were collected with the supernatant, spun down at 10,000xg, washed once with 1X PBS, pelleted and snap-frozen.The microglia remaining on the flask were overlaid with 5ml 1X PBS, removed using a cell scraper, pelleted, and snap-frozen.

Western blotting
Total cellular protein isolation, production of recombinant UCP2 in E. coli, and western blot (WB) was performed as described previously. 46,82For WB analyses 10 mg of total protein were loaded on the SDS-gel, 1 ng of recombinant UCP2 was used as positive control for anti-UCP2 antibody detection.To ensure blotting homogeneity to nitrocellulose membranes, Ponceau S staining was performed.To relate UCP2 expression to loading control, the membranes were stripped, re-blocked and incubated with respective antibodies.The affinity purified antibody against the N-terminal sequence of mouse UCP2 (VGFKATDVPPTATVKF) was designed and evaluated in the Pohl laboratory. 46Peptide synthesis, rabbit immunization and affinity purification were performed by PINEDA Antibody-service GmbH (Berlin, Germany).All other antibody source and concentrations are provided in the key resources table.Recombinant murine UCP2 used as positive control was generated in the Pohl laboratory. 83riefly, E. coli Rosetta DE3 strain (Merck) were transformed with murine UCP2 cDNA expression plasmids and protein was isolated after induction, high-pressure homogenization, and centrifugation.1 mg of isolated inclusion bodies were solubilized in a TE/G buffer (2% N-lauroylsarcosine, 1.3% Triton X-114, 0.3% n-octylpolyoxyethylene, 1 mM DTT, and GTP at pH 7.5), with gradual addition of 50 mg lipid mixture (DOPC:DOPE:CL; 45:45:10 mol%).The mixture was dialyzed in buffer (50 mM Na 2 SO 4 , 10 mM MES, 10 mM Tris, and 0.6 mM EGTA at pH 7.34) after a concentration step.Bio-Beads SM-2 (Bio-Rad) application removed non-ionic detergents and hydroxyapatite columns removed unfolded and aggregated proteins from the dialysate.Micro BCA Protein Assay Kit (Thermo Fisher Scientific) was used for determining protein concentration.

Retina sample preparation for fluorescence-activated cell sorting (FACS)
Adult animals expressing the mito-Dendra2 tag were anesthetized with isoflurane and decapitated.Retinas were immediately explanted and dissected in 1X PBS on ice.Each retina was transferred to a 1.5ml Eppendorf low-adhesion tube filled with the 800ml digestion buffer (1:8:1 Cysteine/EDTA solution (2.5mM Cysteine, 0.5mM EDTA (ethylenediaminetetraacetic acid) in HBSS, 10mM HEPES (4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid) in HBSS, and 10mg/ml Papain, Roche #10108014001), incubated at 37 C for 10 minutes, then centrifuged for 2.5 min at 240xg.Samples were transferred on ice, the supernatant was discarded and 800ml 1mM EDTA in HBSS + 2% (v/v) FBS added.After two washes with 1mM EDTA in HBSS + 2% (v/v) FBS, the digested tissue was triturated 10-15 times with a pulled glass pipette, then filtered through a 70 mm strainer.For the measurement of mitochondrial reactive oxygen species, 5 mM MitoSox (ThermoFisher, #M36008) in HBSS was added to the digested tissue and pipetted to break apart the tissue prior to a 10 minute incubation at 37 C, then the sample was centrifuged for 3 minutes at 304xg at 4 C. Samples were carefully washed two times with 1mM EDTA in HBSS + 2% (v/v) FBS, triturated 10-15 times with a pulled glass pipette, then filtered through a 70 mm strainer.

Fluorescent activated cell sorting (FACS)
FACS was performed using a SONY SH800SFP equipped with a 100 mm nozzle and sorting speed of approximately 8,000 events per second in normal purity mode.Microglia were gated from the population by setting the forward and side scatter and by the fluorescence of the mito-Dendra2 tag.From each retina, 100 microglia were sorted and immediately collected in one well of a 96-well plate (Eppendorf) filled with 5 ml cold lysis buffer from the NEBNextâ Single Cell/Low Input cDNA Synthesis & Amplification Module (New England Bio Labs, #E6421L).After the sorting, the plate was shortly spun down to ensure all cells were collected at the bottom of the well and was immediately processed for cDNA synthesis.For MitoSOX based flow cytometry approximately 500,000 events were acquired for each experimental condition.All collected data were gated based first on forward and side scatter, and then on the fluorescence of the mito-Dendra2 tag and the MitoSOX.Data were analyzed using FlowJo ä v10.8 Software (BD Life Sciences).

cDNA synthesis
Sorted microglia were processed with the NEBNextâ Single Cell/Low Input cDNA Synthesis & Amplification Module (New England Bio Labs, #E6421L) according to the manufacturer's protocol.

Reverse transcription quantitative real-time PCR and gene expression analysis
Primers (Table S11) were designed with the free PrimerQuest Tool from Integrated DNA Technologies (https://eu.idtdna.com/PrimerQuest/Home/Index).To ensure the target sequence of each primer, primers were blasted (https://www.ncbi.nlm.nih.gov/tools/primer-blast/). Their self-complementarity and folding probability were investigated using the UNAFold Web server (http://www.unafold.org/).Primer efficiencies were validated from the slope of four to five serial 1:4 dilutions of cDNA template according to Equation 1. Primers with efficiencies between 90-110% were used.

Efficiency = 2 ^À 1
Slope À 1 Ã 100 (Equation 1) For gene expression analysis, RT-qPCR (Lunaâ Universal qPCR Master Mix; New England BioLabs; M3003L) was performed in 384-well plates (Bio-Rad; HSR4805) on a Roche Lightcycler 480 according to the manufacturer's manual.Total reaction volume was 10ml containing 1ml of 1:10 diluted cDNA as template and a final concentration of 0.25mM for each primer.Cycle conditions were 60 seconds at 95 C for initial denaturation, followed by 40 cycles of denaturation (15 seconds; 95 C) and annealing/extension (30 seconds; 60 C).Each run was completed with a melting curve analysis to confirm amplification of only one amplicon.Each PCR reaction was run in triplicate from which a mean Cq value was calculated and used for further analysis.

Analysis of RT-qPCR results
Fold change differences between each condition and WT naive were calculated according to delta-delta Ct method. 84dCq values were obtained by normalizing mean Cq values to the reference housekeeping gene (GAPDH) measured within the same experiment (Equation 2).ddCq values were then calculated by normalizing dCq values to the respective control condition (WT naive ) within each experimental repetition (Equation 3).Fold changes were obtained by transforming ddCq values from log2-scale to linear scale (Equation 4).These fold changes were used for data visualization.Exclusion criteria were based on IBA1 and GFAP fold change expression to determine purity of cell isolation, which would indicate substantial contamination from astrocytes and Mu ¨ller glia.dCq = Cq reference gene À Cq gene of interest (Equation 2) ddCq = dCq À dCq control condition (Equation 3) Fold change = 2 ðddCqÞ (Equation 4)

Retina dissection and fixation
Animals were briefly anesthetized with isoflurane (Zoetis) until breathing slowed, then cervical dislocation was performed.Retinas were immediately dissected from the enucleated eyes in 1X phosphate buffered saline (PBS) and fixed in 4% (w/v) paraformaldehyde for 30 minutes.After 3x PBS washes, retinas were put in 30% (w/v) sucrose in 1X PBS overnight at 4 C for either -80 C freezer storage or immunostaining.

Figure 1 .
Figure 1.Mitochondria in microglia in vivo establish networks distinct from in vitro (A) Retinal section immunostained for TOMM20 (white).Scale bar: 20 mm.ONL, outer nuclear layer.OPL, outer plexiform layer.INL, inner nuclear layer.IPL, inner plexiform layer.GCL, ganglion cell layer.(B) Schematic of mouse model for mitochondrial visualization with floxed mito-Dendra2 at the Rosa26 locus (PhAM fl/fl ).Pham fl/fl are crossed to Cx3cr1 CreERT2 for microglia-selective mito-Dendra2 expression (green) after receiving tamoxifen injection for three consecutive days.Referred to as WT throughout the study.(C) Overview image of immunostained microglia (IBA1, magenta) in the IPL expressing the mitochondrial label (Dendra2, green) in retinal wholemounts for WT.Scale bar: 50 mm, zoom-in: 20 mm.

Figure 5 .
Figure 5. Male UCP2 KO microglia resort to mitochondrial hyperfusion for stress mitigation (A-C) Principal component analysis (PCA) of microglia from WT and UCP2 KO in naive and ONC conditions.The first two principle components (PC) are visualized.Each dot represents a single microglial profile defined by the following six parameters: microglial Sholl index, CD68 volume, median mitochondrial volume,

FBH
Volume of largest mitochondrionNaive UCP2 KO ♀ ov ONC UCP2 KO ♀ ov Mitochondria in ovariectomized UCP2KO  female after ONC Sex comparison of ovariectomized ONC UCP2KO  females