Models of microglia depletion and replenishment elicit protective effects to alleviate vascular and neuronal damage in the diabetic murine retina

Due to the fact that streptozotocin does not induce consistent hyperglycemia in female mice, all experiments used male mice 6–8 weeks old. CX3CR1^CreER (JAX stock number: 020940; RRID:IMSR_JAX:020940), Rosa26:TdTomato, (JAX stock number: 007914; RRID:IMSR_JAX:007914), Rosa26:iDTR (JAX stock number: 007900; RRID:IMSR_JAX:007900) and CX3CR1-WT (JAX stock number: 000664; RRID:IMSR_JAX:000664) mice were purchased from The Jackson Laboratory. Rosa26:TdTomato and Rosa26:iDTR mice harbor a TdTomato (TdT) or diphtheria toxin receptor (DTR) gene, respectively, with a STOP sequence flanked by loxP sites. Cre-specific expression removes the STOP sequence allowing expression of a functional TdT or DTR gene. CX3CR1^CreER mice were crossed to Rosa26:TdTomato mice to generate CX3CR1^CreER:R26^TdT mice. CX3CR1^CreER:R26^iDTR mice were generated by crossing CX3CR1^CreER mice to Rosa26:iDTR mice. Mice were maintained at the Laboratory Animal Resource Center at The University of Texas at San Antonio under conventional housing conditions. All experiments were performed in accordance with National Institutes of Health guidelines and approved by UTSA-Institutional Animal Care and Use Committee.

To induce hyperglycemia, mice were injected intra-peritoneally (i.p) once daily for 5 days with 60 mg/kg of streptozotocin (STZ) (Sigma Aldrich catalog number: S0130) [11, 12]. Age-matched non-diabetic controls were administered citrate buffer as a vehicle control. Blood glucose levels were measured once a week using a blood glucose monitor and animals were deemed hyperglycemic when blood glucose levels were > 250 mg/dL. To mimic systemic inflammation due to recurrent infections common in diabetic patients, all STZ-treated mice were given one daily injection of 0.08 mg/kg lipopolysaccharide (LPS) (Sigma Aldrich catalog number: L2637) for 2 days prior to euthanasia [13‐16].

Cre recombinase induction. Non-diabetic CX3CR1^CreER:R26^TdT and CX3CR1^CreER:R26^iDTR mice at 6 weeks of age, were given an injection of 165 mg/kg tamoxifen (TAM) (Sigma Aldrich catalog number: T5648) once a day for 5 days and age-matched controls were administered corn oil [17]. 3-day diphtheria toxin treatment (Fig. 1E). To deplete microglia, 1 month after the last TAM injection, CX3CR1^CreER:R26^iDTR mice were treated with PBS as a control or 25 ng/g diphtheria toxin (DTx) (Sigma Aldrich catalog number: D0564) once daily for 3 days. Tissues were collected 24 h after the last DTx injection. Two-week diphtheria toxin treatment (Fig. 2A). Two weeks following the last TAM injection, diabetes was induced in CX3CR1^CreER:R26^iDTR mice and at 6 weeks of hyperglycemia, DTx was injected once daily for 3 days followed by daily injections (25 ng/g) once every 48 h for two consecutive weeks until 8 weeks of hyperglycemia [17]. In total, mice received 10 DTx injections during this 2-week treatment regimen, the first 3 injections 24 h apart, followed by 7 injections 48 h apart. PBS and DTx treated mice were pulsed with EdU at 7 weeks of diabetes, 1 week after the depletion started to determine the frequency of newly proliferating cells by administering 1 intraperitoneal injection of EdU at 50 mg/kg. Tissues were collected at 24 h after the last DTx injection at 8 weeks of diabetes. Two-week recovery following 2-week diphtheria toxin treatment (Fig. 2A). A separate group of mice was euthanized 2 weeks after the 2-week diphtheria toxin treatment (6-week diabetes + 2-week diphtheria toxin + 2-week recovery). Non-diabetic (ND) groups received citrate buffer instead of STZ. ND, 8-week and 10-week diabetic controls received TAM to induce the Cre recombinase and PBS instead of DTx.

Two-week PLX-5622 induced microglia depletion. Eight weeks following STZ-induced hyperglycemia, CX3CR1-WT mice were fed 0.12% PLX-5622 chow for 2 weeks, and tissues were collected the last day of treatment when mice had progressed to 10 weeks of hyperglycemia (Fig. 5A). Microglia depletion and repopulation for mRNA-seq analysis (Fig. 6A). Six weeks following STZ-induced hyperglycemia, CX3CR1-WT mice were fed 0.12% PLX-5622 (MedChem Express, catalog number: HY-114153) chow (7012, Blue, TD.200435, Envigo) for 2 weeks, and tissues were collected at 8 weeks of hyperglycemia (6-week diabetes + 2-week PLX-5622 depletion). Following the 2-week PLX-5622 microglia depletion, a separate group of mice was transitioned back to normal chow and euthanized 2 weeks after microglia depletion to assess microglia repopulation (6-week diabetes + 2-week PLX-5622 depletion + 2-week repopulation). Non-diabetic (ND) groups received citrate buffer instead of STZ. ND, 6-, 8- and 10-week diabetic non-depleted controls served as normal chow controls.

Mice were transcardially perfused with cold 1 × Hank’s balanced salt solutions (HBSS) (RRID: SCR_021689). For retinal isolation, eyes were enucleated and fixed for 20 min in 4% paraformaldehyde (PFA) and post-fixed in 1% PFA for one additional hour (Sigma Aldrich catalog number: P6148). Dissected retinas were placed in cryoprotection solution (200 mL glycerol, 200 mL 0.4 M Sorenson’s buffer and 600 mL MilliQ water) overnight at 4 °C, and the following day placed in cryostorage solution (500 mL 0.2 M PO₄, 10 g PVP-40, 300 g sucrose and 300 mL ethylene glycol) at − 20 °C. Brain tissues were dissected and fixed in 4% PFA overnight followed by cryoprotection overnight at 4 °C. Free floating brain sections (30 μm) were generated using a freezing microtome at − 20 °C and placed in cryostorage solution at − 20 °C.

Blood was collected from mice via submandibular cheek bleed into tubes with K₂EDTA (BD Microtainer catalog number: 365974). Serum tubes were centrifuged at 2200 rcf for 20 min at 4 °C. Supernatant was isolated and 1 uL of 100X proteinase inhibitor cocktail (PIC) (Roche catalog number: 04693116001) was added to serum supernatants per 100 uL of serum to bring the final PIC concentration to 1X.

Blood was collected from mice via submandibular cheek bleed into 1.5 mL Eppendorf tubes with 30 μL of 5000 U/mL heparin (30 μL of heparin per 500 μL of blood) (Sigma Aldrich catalog number: H3393) and lysed in water for 20 s followed by the addition of 10 × HBSS and 1 × HBSS (supplemented with 10 mM HEPES). Lysed blood was centrifuged at 4 °C for 7 min at 2200 rpm. Mononuclear cells were isolated from the brains and spinal cords as previously described using Percoll gradients [18‐20]. Cellular suspensions were prepared in cell staining buffer (Biolegend catalog number: 420201), blocked and stained as previously described [21]. Antibody cocktail used to label peripheral leukocytes and CNS mononuclear cells CD11b, CD45, Zombie aqua, Ly6C and CD11c. The following cell subsets were identified using the previously stated cocktail: neutrophils (CD45^HiCD11b⁺SSC^Hi), tissue resident (CD45^HiCD11b⁺Ly6C^Lo) and inflammatory (CD45^HiCD11b⁺Ly6C^Hi) macrophages, myeloid-derived (CD45^HiCD11b⁺CD11c⁺) dendritic cells (DCs), conventional DCs (CD45^HiCD11b^–CD11c⁺), CNS-resident microglia (CD11b⁺CD45^LoP2RY12⁺Ly6C^–) and monocyte-derived microglia (CD11b⁺CD45^LoP2RY12⁺Ly6C⁺) [15]. Catalog numbers for the antibodies used in the antibody cocktails to stain cells are outlined in Table 1.

For immunohistochemical analysis, whole retinal preparations or brain tissues were blocked overnight in 10% goat or donkey serum containing 1% Triton-X 100 at 4 °C. Tissues were incubated with primary antibodies overnight at 4 °C in blocking solution (10% goat or donkey serum containing 1% Triton-X 100) followed by 5 washes each at 5 min in PBS with 0.1% Triton-X 100. Tissues were incubated in species-specific secondary antibodies to visualize proteins of interest, rabbit anti-ionized calcium binding adaptor molecule-1 (Iba1), mouse anti-neuronal nuclei (NeuN), mouse anti-β tubulin III (TUJ1), guinea pig anti-RNA-binding protein with multiple splicing (RBPMS), rat anti-glial fibrillary acidic protein (GFAP), rat anti-pecam-1 (CD31), and rabbit anti-fibrinogen (Table 1). EdU detection. Retinal tissues were rinsed with 1 × PBS for 5 min followed by 2 washes in 3% BSA for 5 min. Tissues were then incubated in 0.5% triton/PBS for 20 min. Next tissues were immunolabeled following manufacturer’s instructions for the In vivo EdU Click Kit 647 (Millipore Sigma, catalog number: BCK647-IV-IM-M). Following EdU immunolabeling, tissues were stained with Iba1 as outlined above.

Confocal microscopy was done using a Zeiss 710 NLO confocal microscope and 3D compositions of confocal images were generated using Imaris software v7.2 (Bitplane). Six images were obtained per ¼ retinal leaflet per mouse, 2 images at the central retina nearest the optic nerve, 2 images in the middle of the leaflet and 2 images in the outer leaflet. The quantifications shown, represent the average of these 6 images spanning the 3 aforementioned regions of the retina. For Iba1⁺cells/mm³, NeuN⁺ cells/mm³, and percent immunoreactive area of CD31, fibrinogen, TUJ1 and GFAP quantifications, the following sample sizes were used: n = 9 non-diabetic PBS-treated, n = 5 8-week diabetic PBS-treated, n = 5 8-week diabetic DTx-treated, n = 10 10-week diabetic PBS-recovery and n = 7–8 10-week diabetic DTx-recovery mice. For Edu⁺ cells/mm³ quantification, the following sample sized were used: n = 3 non-diabetic PBS-treated, n = 5 8-week diabetic PBS-treated and n = 5 8-week diabetic DTx-treated mice. For IHC analysis in CX3CR1-WT mice (Fig. 5), the following sample sizes were used: n = 9 non-diabetic normal chow, n = 10 10-week diabetic normal chow and n = 10 10-week diabetic PLX-5622 chow. To quantify Iba1⁺ microglial and NeuN⁺RBPMS⁺ neuronal cell body densities, cells were manually counted in 40 × images using the counter tool in Adobe Photoshop version 21.0.3. To quantify the percent immunoreactive area of TUJ1⁺ axons, GFAP⁺ astrocytes, CD31⁺ blood vessels and fibrinogen, using ImageJ Fiji analysis software (NIH), raw confocal images were converted to 32-bit grayscale and then a global automatic threshold was applied to each image. The percent area occupied by the automatic threshold was recorded as the percent immunoreactive area. Data were normalized by volume based on X, Y and Z coordinates (i.e., 212 µm × 212 µm × Z-stack thickness) to account for changes in confocal Z-stack thickness and images size (scale settings- distance in pixels: 1024; known distance: 206.25). For microglial morphological analyses (microglia phenomics) and to determine the transformation index (TI) of microglia, individual cells were traced using ImageJ Fiji analysis software (NIH) to determine the perimeter and area of a microglia cell and calculated using the equation: perimeter²/4π × area² [22]. The TI was determined for 5 microglia per 40 × image from each image from the central, medial and peripheral retina from 5 mice per treatment and genotype for a total of 15 microglia quantified per animal. Values are expressed as a range from 1 to 100, with a TI value closer to 1 representing a circular, amoeboid microglial cell with fewer and/or shorter cellular processes. Higher TI values represent ramified microglial cells with extensive branching and smaller cell bodies.

RNA was isolated from enucleated retinas by the UT Health San Antonio Biospecimen and Translational Genomics Core Laboratory using the Qiagen RNeasy Kit (Qiagen catalog number: 74104) following tissue homogenization using Zymo Research BashingBead lysis tubes (Zymo research catalog number: S6003-50). For mRNA sequencing, quality control (QC), stranded mRNAseq library prep, and sequencing were performed by UT Health San Antonio Genome Sequencing Facility. The quality of Total RNAs was assessed by Agilent Fragment Analyzer (Agilent Technologies, Santa Clara, CA), and only RNAs with RQN > 7 were used for subsequent mRNA-seq library preparation and sequencing. Approximately 500 ng of total RNA was used for stranded mRNA-seq library preparation by following the NEB Directional mRNA-seq sample preparation guide (New England Biolabs, Ipswich, MA). The first step in the workflow involved purifying the poly-A containing mRNA molecules using poly-T oligo-attached magnetic beads. Following purification, the mRNA was fragmented into small pieces using divalent cautions under elevated temperature. The cleaved RNA fragments were copied into first strand cDNA using reverse transcriptase and random primers. This was followed by second strand cDNA synthesis using DNA Polymerase I and RNase H. Strand specificity was achieved by replacing dTTP with dUTP in the Second Strand Marking Mix (SMM). These cDNA fragments then went through an end repair process, the addition of a single ‘A’ base, and then ligation of the adapters. The products were then purified and enriched with PCR to create the final RNA-seq library. Finally, RNA-seq libraries were subjected to quantification process, pooled for cBot amplification and subsequent 100 bp paired read sequencing run with Illumina NovaSeq 6000 platform. After the sequencing run, demultiplexing with Bcl2fastq2 was employed to generate the fastq file, with the average of 30 M reads per sample. Raw reads were imported into CLC Genomics Workbench v21.0.5. Initially, adaptors were trimmed and remaining reads for each sample were mapped to the annotated mouse genome (GRCm39), followed by differential gene expression (DEG) analysis of diabetic PLX-treated and diabetic PLX-recovery mice against their diabetic controls, diabetic normal chow-treated and diabetic normal chow-recovery, respectively, using the RNA-seq analysis tools within the CLC genomics software. DEG analysis was also performed on all diabetic groups, diabetic before treatment, diabetic normal chow-treated, diabetic PLX-treated, diabetic normal chow-recovery, and diabetic PLX-recovery, against non-diabetic mice. Genes were considered differentially expressed if the FDR P-value was < 0.05. For data mining, DEGs were defined by a cutoff of FDR P-value < 0.05. Principal component analysis (PCA) plots to show sample clustering by treatment type are outlined in Additional file 1: Fig. S8 and S9.

All analyses were conducted using GraphPad Prism v9.2 and a P value < 0.05 was considered statistically significant. Statistical significance is denoted as *P value < 0.05, **P value < 0.01, ***P value < 0.001 and ****P value < 0.0001. Statistical tests performed included a two-tailed parametric unpaired Student's t test with Welch’s correction when comparing two groups. When comparing multiple groups, a two-way ANOVA with the Tukey’s post hoc test was performed, using the treatment type as the first variable and genotype as the second variable.

We used CX3CR1^CreER:R26^TdT mice to confirm that Cre penetrance targets CX3CR1-expressing cells and to validate the feasibility of depleting resident microglia in the retina and brain without affecting peripheral CX3CR1-expressing immune cells (Fig. 1A). We verified Cre expression prevalence (TdT expression) among blood leukocytes (TdT⁺CD11b⁺CD45^Hi) before TAM injection, and 1, 3 and 6 weeks post-TAM administration by flow cytometry (Additional file 1: Fig. S1A). We found that less than 0.02% of blood leukocytes (CD11b⁺CD45^Hi) were TdT⁺ prior to TAM delivery in control mice (0.02 ± 0.01047) (Additional file 1: Fig. S1C). Upon TAM administration, there was a significant increase in the percentage of CD11b⁺CD45^Hi TdT⁺ cells at 1 week (2.083 ± 0.6445, Student’s t test P = 0.0078) and 3 weeks (2.403 ± 0.7707, Student’s t test P = 0.0007) post-TAM from 0.02% TdT⁺ cells in vehicle-treated mice (Additional file 1: Fig. S1C). However, 6 weeks after TAM administration, a time period during which bone marrow turnover allowed repopulation of blood leukocytes, there was a significant reduction in the percentage of CD11b⁺CD45^Hi TdT⁺ cells in TAM-treated mice (0.21 ± 0.1805) showing less than 0.4% of red fluorescent cells (Additional file 1: Fig. S1C). The data at 6 weeks post-TAM (0.21 ± 0.1805) closely resembles the levels found in corn oil-treated controls (0.02%) (Additional file 1: Fig. S1C). These results from TdT expression analysis in CD11b⁺CD45^Hi peripheral leukocytes confirmed that bone marrow turnover replenishes a TdT^Neg population in the periphery (Additional file 1: Fig. S1C). Flow cytometry on brain microglia (CD11b⁺CD45^Lo) showed that the proportion of TdT⁺ cells in control (33.01 ± 1.841) and vehicle groups (35.97 ± 4.479) was comparable and revealed the baseline levels of leakiness of the promoter (Additional file 1: Fig. S1E). At 6 weeks post-TAM (95.77 ± 0.8770, Student’s t test P < 0.0001) administration, greater than 95% of the microglia population was CD11b⁺CD45^LoTdT⁺, confirming a Cre-specific regulation of TdT expression in CX3CR1⁺ cells (Fig. 1E). The percentage of CD11b⁺CD45^Hi CNS infiltrating leukocytes was comparable among control (0.7929 ± 0.5669), vehicle (0.7629 ± 0.7024) and TAM-treated (0.7486 ± 0.4678) mice (Additional file 1: Fig. S1F). These results were also supported by immunofluorescent (IF) analyses of brain and retinal tissues from corn oil and TAM-treated mice (Fig. 1B–D). In both brain and retinal tissues, TdT⁺ cells accounted for ~ 40% of the Iba1⁺ population in corn oil-treated mice, in contrast to TAM-treated mice whose Iba1⁺TdT⁺ population accounted for 100% of the microglia population (Fig. 1B, D). TAM treatment did not alter the overall Iba1⁺ microglia density in the brain and retina (Fig. 1C). These results indicate that the CX3CR1^CreER strain is controllable by TAM serving as a valid model to conditionally express our gene of interest, diphtheria toxin receptor (DTR), at specified time points of disease in the retina, without targeting peripheral CX3CR1-expressing cells.

To deplete microglia, TAM-treated CX3CR1^CreER:R26^iDTR mice were injected with PBS as a control or DTx (25 ng/g of body weight) once daily for 3 days and euthanized the last day of the depletion regimen (Fig. 1E). The overall population of circulating CD11b⁺CD45^Hi blood leukocytes remained unaltered following DTx administration (5.364 ± 1.471), closely resembling PBS (3.938 ± 1.652) treated controls (Additional file 1: Fig. S2B, C). To assess the impact of this acute depletion regimen in neuron and microglial densities, immunofluorescent analysis was used to compare retinal and brain tissues (Fig. 1F–H and Additional file 1: Fig. S2D-H). We focused on the primary visual cortex (PVC) in brain tissues because this is the region where axons synapse from the retina through the optic nerve and the dorsal lateral geniculate [23‐27]. We found that acute microglia depletion resulted in a ~ 30% reduction in the overall Iba1⁺ microglia density in the PVC and did not alter the overall NeuN⁺ neuronal coverage (Additional file 1: Fig. S2D-F). However, retinal tissues were more susceptible to microglia depletion with greater than 60% of Iba1⁺ microglia cell loss after DTx administration (Fig. 1F, G). The retinas of DTx-treated (26.26 ± 2.357) mice did not display any alterations in TUJ1⁺ axonal percent immunoreactive area compared to PBS-treated (28.2 ± 4.471) controls (Fig. 1G), nor changes in astrocyte morphology or distribution (Additional file 1: Fig. S2G-H). These results reveal that this model allows successful depletion of microglia in the murine retina without depleting CX3CR1-expressing peripheral immune cells or eliciting acute neurotoxic effects in CNS tissues.

Next, CX3CR1^CreER:R26^iDTR mice (Fig. 2A) were analyzed after a 2-week DTx treatment and after a 2-week recovery, time points at which mice were diabetic for 8 and 10 weeks, respectively. Non-diabetic (ND), 8-week diabetic and 10-week diabetic PBS-treated groups were used as controls. We assessed peripheral immune cells and found that DTx treatment in diabetic mice did not alter the frequencies of neutrophils (CD45^HiCD11b⁺SSC^Hi), tissue resident (CD45^HiCD11b⁺Ly6C^Lo) and inflammatory (CD45^HiCD11b⁺Ly6C^Hi) macrophages, or myeloid-derived (CD45^HiCD11b⁺CD11c⁺) dendritic cells (DCs) when compared to PBS-treated diabetic controls (Fig. 2B–E, G) (gating strategy Additional file 1: Fig. S3A). However, changes were detected when comparing non-diabetic and diabetic mice to include a decrease in neutrophils, tissue resident macrophages, myeloid DCs, and an increase in inflammatory macrophages in diabetic mice (Fig. 2B–E, G). A significant increase in the CD45^HiCD11b^–CD11c⁺ conventional DCs population was observed when comparing ND and 8-week diabetic groups, but this population returned to baseline levels at 10 weeks of diabetes (Fig. 2F). BioPlex 23-Cytokine analysis on serum samples did not detect significant differences in the presence of cytokines associated with inflammation (IL-1a, IL-1β, IL-13, IL-17, GM-CSF, IFN- γ, MIP-1α, TNF-α, IL-6 and RANTES), proliferation (IL-3, IL-5 and IL-12p40), activation and chemotaxis (CXCL1, eotaxin and CCL-2) when comparing the DTx-recovery mice to their diabetic controls (Additional file 1: Fig. S3B-D). Differences observed in these cytokines and chemokines were due to diabetes itself as 10-week diabetic mice showed significant increases in these cytokines and chemokines when compared to ND mice (Additional file 1: Fig. S3B-D). This data suggests that under inflammatory conditions, DTx treatment does not seem to alter the peripheral immune cell profile, nor the cytokine profile environment in the periphery.

In contrast to acute DTx treatment (Fig. 1E), 2 weeks DTx treatment led to a significant increase in Iba1⁺ cells in the diabetic retina (41,157 ± 11,571) (Fig. 3A) in comparison to the PBS-treated ND (25,038 ± 2705) and diabetic controls (25,459 ± 2567, Student’s t test P = 0.0369) (Fig. 3A, B). We expected to observe a decrease in Iba1⁺ microglia cells after the 2-week DTx regimen, instead we found that ~ 35% of the retinal Iba1⁺ cells were EdU⁺ in the diabetic-DTx treated mice (34.76 ± 6.078), suggesting that long-term DTx treatment does not sustain microglia depletion (Fig. 3C, D). We next assessed microglial morphological changes by measuring their transformation index (TI) (Fig. 3E, F). In comparison to diabetic-PBS mice (20.16 ± 12.64), microglia from diabetic-DTx mice had significantly lower TIs (12.38 ± 8.34, Student’s t-test P < 0.0001) suggesting a higher activation and a phagocytic morphology. In diabetic retinas, microglia in the 2-week recovery group displayed higher TI values (25.4 ± 11.56, Student’s t-test P < 0.0001) in comparison to the corresponding diabetic-PBS control mice (12.55 ± 8.486) that displayed amoeboid microglia with low TI values (Fig. 3E, F). We also detected a significant increase in the number of Iba1⁺ cells (32,904 ± 3923, Student’s t test P = 0.0005) with significantly higher TI’s (33.78 ± 15.45, Student’s t-test P < 0.0006) in non-diabetic 2-week recovery mice in comparison to ND controls (Additional file 1: Fig. S4B, C, E, F). To validate microglial activation and to inquire about the potential origin of repopulating cells in the diabetic CNS, we did flow cytometry to distinguish CNS-resident microglia CD11b⁺CD45^LoP2RY12⁺Ly6C^– and monocyte-derived microglia CD11b⁺CD45^LoP2RY12⁺Ly6C⁺ (Additional file 1: Fig. S3E-H) [15]. Microglia depletion using CX3CR1^CreER:R26^DTA mice demonstrated that microglial repopulation occurs from the resident pool of microglia that were resistant to depletion, and from bone marrow derived Ly6C^Hi monocytes that infiltrate the brain and acquire a microglia-like signature [17]. Therefore, we deemed microglia that were CD11b⁺CD45^LoP2RY12⁺Ly6C⁺ as monocyte-derived microglia due to their expression of Ly6C in comparison to CNS-resident microglia that do not express Ly6C. There was a significant increase in the CNS-resident CD11b⁺CD45^LoP2RY12⁺ microglia population in diabetic-PBS and diabetic mice after 2-week recovery (Additional file 1: Fig. S3F). There was also significant increase in monocyte-derived CD11b⁺CD45^LoP2RY12⁺Ly6C⁺ microglia in diabetic-PBS and diabetic-DTx mice in comparison to ND controls (Additional file 1: Fig. S3G). The frequency of monocyte-derived CD11b⁺CD45^LoP2RY12⁺Ly6C⁺ microglia was sustained in diabetic PBS-control mice, but a robust decrease in CD11b⁺CD45^LoP2RY12⁺Ly6C⁺ microglia was observed in diabetic DTx-recovery mice (Additional file 1: Fig. S3G).

To assess glial and neuronal responses to diphtheria toxin (same experimental design as shown in Fig. 2A), retinal tissues were stained to visualize neurons (NeuN⁺RBPMS⁺), axonal integrity (TUJ1⁺), astrocytes and Müller glia (GFAP⁺), angiogenesis (CD31⁺) and fibrinogen deposition in retinal flat mounts (Fig. 4). Diphtheria toxin treatment and recovery in non-diabetic mice did not cause changes in the number of NeuN⁺RBPMS⁺ cells (352,340 ± 28,343), TUJ1⁺ (42.29 ± 2.014) and GFAP⁺ (43.57 ± 3.759) percent immunoreactive area in comparison to ND PBS controls (Additional file 1: Fig. S4D, G-I). We observed an increase in GFAP⁺ glial cells in diabetic-DTx groups over their PBS-treated controls (Fig. 4B). However, diabetic control mice, both at 8 weeks and 10 weeks of diabetes, showed NeuN⁺RBPMS⁺ neuronal cell loss (284,504 ± 14,011 and 292,792 ± 28,127, respectively) and decreased TUJ1⁺ immunoreactivity (33.28 ± 1.928 and 36.35 ± 4.121, respectively) when compared to diabetic-DTx (NeuN⁺RBPMS⁺: 312,572 ± 54,360; TUJ1%: 49.41 ± 3.729, Student’s t-test P = 0.0001), diabetic at 2 weeks recovery (NeuN⁺RBPMS⁺: 334,078 ± 32,437, Student’s t-test P = 0.0187; TUJ1%: 47.61 ± 5.686, Student’s t-test P = 0.0005) and ND (NeuN⁺RBPMS⁺: 327,108 ± 41,863, Student’s t test P = 0.0182; TUJ1%: 38.84 ± 5.539, Student’s t test P = 0.02) mice (Fig. 4C, D). These data suggest that diphtheria toxin treatment in diabetic CX3CR1^CreER:R26^iDTR mice supports a neuroprotective environment in the diabetic retina.

We identified vascular abnormalities by staining for CD31⁺ blood vessels and classified damaged blood vessels as those containing ruptures and fibrinogen deposits outside of the vasculature (Fig. 4A, E, F). We did not detect changes to the vasculature, nor fibrinogen deposits in the retinas of non-diabetic mice after 2-week recovery (Additional file 1: Fig. S4J-L). Ten weeks diabetic control mice displayed a significant increase in angiogenesis (24.85 ± 4.981, Student’s t-test P = 0.0006) in comparison to ND (14.27 ± 5.735) mice (Fig. 4A, E, F). Angiogenesis was not evident in 8-week diabetic-PBS (16.29 ± 5.232), diabetic-DTx (14.20 ± 2.048) or DTx-recovery (14.89 ± 3.684) mice, closely resembling ND controls (14.27 ± 5.735) (Fig. 4A, E, F). Furthermore, the vasculature in 10-week diabetic mice (PBS controls) had many vascular abnormalities with discontinuous, ruptured blood vessels, and aggregated endothelium deposits throughout the retina that colocalized with fibrinogen deposits (Fig. 4A). In addition to these vascular abnormalities, robust amounts of fibrinogen deposits were detected (13.17 ± 5.666) (Fig. 4A, F). These fibrinogen deposits in diabetic mice were initially detected in 8-week diabetic-PBS mice (7.27 ± 2.558) and only exacerbated as disease progressed to 10 weeks of diabetes (Fig. 4A, F). Conversely, diabetic DTx-treated (CD31%: 14.20 ± 2.048; fibrinogen%: 1.969 ± 1.725) and diabetic mice after 2-week recovery (CD31%: 14.89 ± 3.684; Fibrinogen%: 1.944 ± 2.198) did not show fibrinogen deposits nor vascular damage, similar to ND mice (CD31%: 14.27 ± 5.735; Fibrinogen%: 0.8082 ± 0.7421) (Fig. 4A, E, F).

We compared the degree of microglia depletion in diabetic CX3CR1^CreER:R26^iDTR mice DTx treated for 2 weeks and in diabetic CX3CR1-WT mice, PLX-5622 treated for 2 weeks (Additional file 1: Fig. S6A). Flow cytometric analysis of brain and spinal cord tissues revealed no difference in the percentage of live CD11b⁺CD45^LoZombie^– microglia in DTx-treated CX3CR1^CreER:R26^iDTR mice (94.22 ± 1.469, Student’s t test P = 0.0006) in comparison to their PBS-treated diabetic CX3CR1^CreER:R26^iDTR controls (95.12 ± 0.9706) (Additional file 1: Fig. S6B). We observed a significant reduction in the percentage of live CD11b⁺CD45^LoZombie^– microglia in PLX-5622-treated CX3CR1-WT mice (30.06 ± 4.95, Student’s t test P < 0.0001) in comparison to their diabetic, normal chow CX3CR1-WT controls (99.56 ± 0.1776) (Additional file 1: Fig. S6B). In retinal tissues, the data showed a significant increase in Iba1⁺ cells in the DTx-treated CX3CR1^CreER:R26^iDTR mice (41,157 ± 11,571, Student’s t test P = 0.0345) compared to their diabetic CX3CR1^CreER:R26^iDTR PBS controls (25,459 ± 2567). In PLX-5622 treated CX3CR1-WT mice (6994 ± 7422, Student’s t test P < 0.0001), we detected a robust reduction in Iba1⁺ cells in the retina of compared to their diabetic CX3CR1-WT normal chow controls (28,721 ± 4462) (Additional file 1: Fig. S6C). These results highlight the differences in the kinetics of microglia depletion with DTx and PLX-5622. Following depletion, diabetic DTx-treated CX3CR1^CreER:R26^iDTR mice had a reduction in their TI (12.38 ± 8.34, Student’s t test P < 0.0001) compared to their diabetic, CX3CR1^CreER:R26^iDTR PBS controls (30.73 ± 10.16), indicative of increased activation in DTx-treated mice (Additional file 1: Fig. S6D). In contrast, PLX-5622-treated diabetic CX3CR1-WT mice exhibited a ~ 20% increase in their TI (20.49 ± 8.502, Student’s t test P < 0.0001) compared to their diabetic, normal chow CX3CR1-WT controls (17.03 ± 6.729) shifting microglia to a more branched and ramified morphology (Fig. 5D, E and Additional file 1: Fig. S6D).

Diabetes induced significant TUJ1⁺ axonal loss in diabetic mice (32.08 ± 3.225, Student’s t test P = 0.0095) in comparison to ND mice (35.95 ± 2.703) (Fig. 5F, G). PLX-5622 treatment correlated with an increase in TUJ1⁺ axonal density in diabetic mice (42.15 ± 2.59, Student’s t test P < 0.0001) in comparison to diabetic control mice (Fig. 5F, G). Consistent with previous results (Fig. 4E, F), diabetes induced angiogenesis, ruptured blood vessels and fibrinogen deposition in diabetic mice (CD31: 20.73; fibrinogen: 14.42, Student’s t test P = 0.0015) in comparison to ND mice (CD31: 16.51 ± 2.565; fibrinogen: 0.6982 ± 0.4107) (Fig. 5H–J). This pathology was ameliorated in diabetic PLX-5622-treated retinas which displayed intact vasculature with little fibrinogen extravasation into the diabetic retina (CD31: 14.1 ± 3.115, Student’s t test P = 0.007); fibrinogen: 5.141 ± 1.563, Student’s t test P = 0.0139) (Fig. 5H–J).

We next performed mRNAseq on whole retinas from non-diabetic and diabetic mice with or without microglia depletion to probe general transcriptional changes that would give rise to the neuro- and vasculo-protective effects of transient microglia depletion and repopulation in the diabetic retina (Fig. 6A). Microglia were pharmacologically depleted in 6-week diabetic CX3CR1-WT mice using PLX-5622 and analyzed at three different timepoints: (1) at 6 weeks of diabetes (6D) to establish the retinal transcriptome prior to microglia depletion; (2) after 2 weeks microglia depletion (PLX-treatment), and (3) after 2 weeks of microglia repopulation (PLX-recovery) (Fig. 6A). At each timepoint, ND, diabetic normal chow (NC) treated, and NC recovery were included (Fig. 6A). Microglial activation and expression of disease-associated genes are highly correlated to retinal inflammation and degeneration [8]. We identified 2520 differentially expressed genes (DEGs) significantly upregulated and 1,664 DEGs significantly downregulated in diabetic mice before treatment relative to non-diabetic mice (Fig. 6B). Analysis of diabetic retinas before treatment revealed a significant increase in DEGs associated with DR pathogenesis and microglial activation (Saa3, Ccl7, Madcam1, Tnf, Ccl3, Acod1, Cxcl2, Ccl4, Ch25h, Serpina3n, Fpr1, Mmp13, Hmox1, Slc15a3, Slc7a11, Wfdc17, Csf1 and Abca1), and complement activation (CFB, Fas, C3, C5ar1, C4b, Fcgr3, C1qb) revealing the baseline levels of inflammation prior to depletion (Fig. 6C, D). We identified 994 DEGs significantly upregulated and 831 DEGs significantly downregulated in PLX-treated relative to NC mice and 30 significantly upregulated DEGs and 45 significantly downregulated DEGs in PLX-recovery mice relative to NC-recovery mice (Fig. 6E, H). Genes associated with DR pathogenesis and microglial activation and phagocytosis (Ccl3, Ccl7, Ccr5, Ly86, Cx3cr1, Siglech, Clec7a, Fcgr3, Csf1r, Fcrg1, Itgam, B2m, Msn, Il6st) were significantly downregulated in diabetic PLX-treated and PLX-recovery mice when compared to diabetic NC mice, with the greatest fold changes seen in PLX-treated mice (Fig. 6F, I). Diabetic PLX-treated and PLX-recovery mice displayed a significant downregulation of complement-associated genes to include C3ar1, Ctss, C5ar1, C1qa, C1qb and C1qc (Fig. 6G, J).To corroborate our findings that microglia replenishment is neuroprotective in the diabetic retina, we assessed genes associated with intermediate filament organization (Krt17, Krt16, Krt14, Krt15, Krt6a, Krt6b, Krt5 and Dsp), visual cycle wellness (Rpe65 and Rdh9) and found that these genes were significantly upregulated in diabetic PLX-treated mice compared to diabetic NC controls (Fig. 6G, J). We also identified GFAP, Claudin-1, Claudin-4 and F11r transcripts significantly upregulated in diabetic PLX-treated and PLX-recovery mice indicative of strengthening of the glial limitans (Additional file 1: Fig. S7B-E), and vascular protection (Adamts13 and Csf3) (Additional file 1: Fig. S7F-G).