MerTK-triggered TGFβ1 autocrine signal regulates microglial response to neurodegeneration

Abstract

Microglial phagocytosis exerts essential roles in neurodegeneration, but how phagocytic processes may reciprocally regulate microglia remains incompletely understood. Here, we report that microglial response in the mouse model of pathological axonal degeneration depends on the phagocytic receptor MerTK. The MerTK-triggered downstream phospholipase C signal is sufficient to induce the up-regulation of PU.1 and IRF8, the two central transcription factors governing microglial functions. Chromatin immunoprecipitation-sequencing analyses identify that PU.1 and IRF8 directly target the gene locus of TGFβ1, and disruption of this PU.1-IRF8 targeting site abolishes the induction of microglial TGFβ1 during neurodegeneration. Of importance, neurodegeneration-induced TGFβ1 acts in an autocrine manner, and the microglia-specific deletion of TGFβ1 or its receptors TGFβR1 or TGFβR2 blocks microglial response. Moreover, microglial TGFβ1 autocrine signaling similarly occurs in the 5×FAD mouse model of Alzheimer’s disease and in human patients. These results have delineated an important mechanism underlying microglial response to neurodegeneration.

The interplay between microglial phagocytosis and function is not completely understood. Here, the authors show that in response to neurodegeneration, the phagocytic receptor MerTK in microglia triggers PU.1 and IRF8, which elicit TGFβ1 as an autocrine signal. This signaling pathway also occurs in Alzheimer’s disease mouse model and patient cells.

Introduction

As unique residential immune cells of the central nervous system (CNS), microglia are essential for neurodevelopment and neurophysiology^1–4. Unlike other glial cells, i.e., astrocytes and oligodendrocytes, microglia originate from myeloid-lineage progenitor cells of the yolk sac at early embryonic stages^5–7. During the CNS development, microglia participate in the engulfment of apoptotic neuronal cells and the pruning of synaptic connections, which represent vital steps in the precise establishment and maintenance of neurocircuits throughout adulthood^8–11. In particular, microglia can initiate phagocytosis to “eat-me” signals, e.g., phosphatidylserine exposed on the outer leaflet of the plasma membrane of neurons or their axons undergoing developmental degeneration. Phosphatidylserine molecules are recognized by specific bridging proteins such as GAS6 and PROS1, which then link the downstream phagocytic receptors of the TAM (TYRO3, AXL, and MerTK) family^12–14. At the same time, phosphatidylserine may directly engage additional receptors expressed by microglia, e.g., TIM4 and TREM2, to trigger phagocytic events^15,16. Besides the phosphatidylserine-dependent pathways, the complement factors C1q and C3 play critical roles in opsonizing synaptic structures for microglial engulfment via the complement receptors such as CR1 or CR3^17,18.

In addition to their involvement in the CNS homeostasis, microglial phagocytosis occurs broadly under disease conditions, e.g., neural damage, pathogenic infection, and neurodegeneration. Microglia have long been known for their capacity to clear cellular debris left by neuronal or axonal death following injuries or strokes, which is indispensable for eliciting neuroinflammation and promoting tissue repair^8,19–21. Also, microglial phagocytosis of pathogen-infected cells prevents infections in the CNS^22–24. Moreover, the complex roles of microglia in neurodegenerative diseases have become increasingly appreciated^4,25–29. For instance, microglia are responsible for the phagocytic removal of protein aggregates such as Aβ plaques, which helps delay the onset and progression of Alzheimer’s disease^30–33. On the other hand, microglia responding to neurodegenerative cues release a repertoire of proinflammatory factors or aberrantly engulf synaptic structures, leading to further destruction of neurocircuits and exacerbation of disease conditions^34–36.

It has been documented that phagocytosis can influence the immune functions of macrophages, the myeloid-lineage immune cells in peripheral tissues with functional similarities to microglia. One of the first examples is that the phagocytosis of apoptotic cells by human macrophages could diminish the expression of proinflammatory cytokines³⁷. Since then, studies have elucidated the immunomodulatory action of specific phagocytic receptors in various contexts. For instance, MerTK suppressed the lipopolysaccharide-induced production of TNFα in mouse macrophages by down-regulating the NF-κB signal³⁸. Also, MerTK and other TAM family receptors could function as the pleiotropic inhibitors of innate immunity in mouse macrophages or dendritic cells^39,40. Notably, several reports showed that blocking the phagocytic process in microglia would interfere with their response under physiological or neurodegenerative conditions^13,33,41. However, the mechanism linking phagocytosis to microglial response remains incompletely understood. Previous studies by colleagues and us have demonstrated that the transcription factors PU.1 (also known as SPI1) and IRF8 cooperatively target a collection of microglial activation-related genes during neurological insults^42–46. Whether a specific phagocytic signal may engage this central transcriptional module is unclear.

In this study, we report that microglial response in the mouse model of pathological axonal degeneration depends on the phagocytic receptor MerTK. Chromatin immunoprecipitation-sequencing analyses then reveal that PU.1 and IRF8 directly target the Tgfb1 gene, with disruption of the PU.1-IRF8 targeting site abolishing microglial TGFβ1 induction during neurodegeneration. Of importance, the microglia-specific deletion of TGFβ1 or the receptors TGFβR1 or TGFβR2 blocks microglial response. Moreover, this TGFβ1 autocrine signal similarly occurs in the 5×FAD mouse model of Alzheimer’s disease and in human patients. These results have thus delineated a critical mechanism of microglial response to neurodegeneration.

Results

Microglial response to pathological axonal degeneration depends on MerTK

As the entry point, we utilized the mouse procedure of optic nerve injury, a standard model of pathological axonal degeneration in the CNS. A unique advantage of this model is that optic nerves do not contain neuronal cell bodies, thereby essentially excluding any secondary effect of neuronal damage. Studies have shown that axons of retinal ganglion cells projecting in injured optic nerves undergo a highly synchronized, stereotyped neurodegenerative process, known as Wallerian degeneration, eliciting the profound microglia-mediated neuroinflammation^47–50. In addition, phosphatidylserine would become exposed on the outer leaflet of the membrane of injury-induced degenerating axons to trigger phagocytosis processes⁵¹, rendering this model ideal for investigating the function of phagocytic signals in microglia. Importantly, our previous work combining linkage tracing, parabiosis, and microglia-specific genetic deletion demonstrated that this microglial response did not involve circulating peripheral immune cells⁴⁵. To verify this notion, we performed parabiosis between wild-type and H11-ZsGreen mice (Supplementary Fig. 1A). The FACS analysis of peripheral blood from the wild-type mice in parabiotic pairs showed that approximately 50% of CD45⁺ total immune cells were ZsGreen⁺, confirming the success of the procedure (Supplementary Fig. 1B). However, no ZsGreen⁺ cells were detected in the injured optic nerves of those wild-type mice in parabiosis, suggesting a minor involvement of circulating immune cells in this experimental setup (Supplementary Fig. 1C). Meanwhile, we acknowledge the possibility that microglia from other segments of the optic nerve or the retina may migrate into the injured nerve tissue.

We first assessed the transcriptomic profile of mouse microglia by RNA sequencing (RNA-seq), focusing on the central components of phosphatidylserine-dependent phagocytosis pathways (Supplementary Table 1). Consistent with previous reports¹³, the phagocytic receptor Mertk had a higher expression than the other two TAM family members, Tyro3 and Axl. The bridging proteins Gas6 and Pros1 were also highly expressed in mouse microglia, indicating their capacity to initiate phosphatidylserine-dependent phagocytosis. We generated Cx3cr1^CreER/+; Mertk ^fl/fl mice to achieve the inducible, microglia-specific deletion of MerTK in adult mice to examine its function in microglial response to pathological axonal degeneration. As expected, there was a profound loss of MerTK in the optic nerves of Cx3cr1^CreER/+; Mertk ^fl/fl mice without or with optic injury, as determined by qPCR (Fig. 1C) and immunostaining (Supplementary Fig. 2C). We noticed a detectable level of MerTK remaining in the optic nerves of Cx3cr1^CreER/+; Mertk ^fl/fl mice, reflecting the fact that other non-microglial cell types, e.g., astrocytes, have phagocytic capacity^52–54.

Fig. 1. Microglial response to pathological axonal degeneration depends on the phagocytic receptor MerTK.

A–J 8-week-old Cx3cr1^CreER/+; Mertk^fl/fl and control Cx3cr1^CreER/+; Mertk ^+/+ female mice were subjected to the 4-hydroxytamoxifen-induced Cre-recombinase activity and then utilized for the model of optic nerve injury. Microglial expression of PU.1 (A) or IRF8 (B) in the optic nerves at 3 days post-injury was assessed by co-immunostaining with CD11b, and representative images are shown. Arrowheads exemplify the PU.1⁺ microglia. Note that IRF8 is a transcription factor with nuclear staining, and the green-channel fluorescence signal on the boundary of optic nerves was non-specific staining. C mRNA levels of Mertk in the optic nerves at 3 days post-injury were determined by qPCR. n = 3 mice per condition, mean ± SEM, two-way ANOVA test. Microglia densities (D), PU.1⁺ microglia (E), and IRF8⁺ microglia (F) were quantified. n = 5 mice per condition, mean ± SEM, two-way ANOVA test. G–J Microglia were FACS-sorted from the optic nerves and analyzed by RNA-seq. Pooled RNA-seq data from 4 mice per condition are shown. G GO enrichment analysis of microglia from the injured optic nerves of control Cx3cr1^CreER/+; Mertk ^+/+ versus Cx3cr1^CreER/+; Mertk^fl/fl mice. Expression levels of homeostatic markers (H), signature genes for microglial response to neurodegenerative insults (I), and proinflammatory cytokines and chemokines (J) in microglia of the indicated conditions.

Because the gene loci of several common microglial markers, e.g., Iba1 and P2ry12, are under the transcriptional control of PU.1 and IRF8⁴⁵, which may interfere with immunostaining assessment with the genetic manipulation of this signaling axis (see below), we chose CD11b, which is not directly targeted by these two transcription factors, to visualize microglia throughout this study. We utilized Tmem119^CreER/+; Rosa26-LSL-tdTomato mice to validate the microglial identity of CD11b⁺ cells in the injured optic nerves (Supplementary Fig. 1D). Of importance, while there was a dramatic upregulation of microglial PU.1 in the injured optic nerves of control Cx3cr1^CreER/+; Mertk^+/+ mice, this response was blunted in Cx3cr1^CreER/+; Mertk^fl/fl mice (Fig. 1A, E). Also, the neurodegeneration-induced IRF8 expression in microglia was diminished with the MerTK deletion (Fig. 1B, F). To exclude the potential “leakage” of the Cre-recombinase activity with Cx3cr1^CreER/+, we included the vehicle-treated Cx3cr1^CreER/+; Mertk ^fl/fl mice as the additional control, whose microglial PU.1 upregulation appeared indistinguishable from that of tamoxifen-treated Cx3cr1^CreER/+; Mertk^+/+ mice (Supplementary Fig. 2A).

In accordance with the central role of PU.1 and IRF8 in microglial response to neurodegeneration, microglial proliferation markedly decreased in the injured optic nerves of Cx3cr1^CreER/+; Mertk ^fl/fl mice (Fig. 1D). Moreover, Gene Ontology (GO) enrichment analysis of the transcriptomic profiles of microglia in the injured optic nerves of Cx3cr1^CreER/+; Mertk^fl/fl mice showed a significant downregulation of pathways central to microglial activation, including microglial response (cell chemotaxis, glial cell differentiation, and mononuclear cell proliferation), cell cycle (mitotic cell cycle phase transition and positive regulation of cell cycle), and phagocytosis (reactive oxygen species metabolic process and protein localization to endosome) (Fig. 1G). In particular, the MerTK deletion blocked the reduction of classic homeostatic markers in microglia of the injured optic nerves, e.g., P2ry12, P2ry13, Sall1, and Tmem119 (Fig. 1H), while inhibiting the expression of signature genes commonly recognized for microglial response to neurodegeneration⁵⁵, e.g., Cst7, Ctsd, Lpl, and Spp1 (Fig. 1I), and inflammatory cytokines and chemokines, e.g., Il1b, Ccl2, Ccl4, and Cxcl2 (Fig. 1J). On the other hand, the MerTK deletion did not significantly affect the expression of homeostatic markers or activation-related genes in microglia of the uninjured optic nerves (Fig. 1H–J), except for Cd68 exhibiting an increasing trend.

Mouse microglia expressed the phagocytic receptor AXL, albeit at a lower level compared to MerTK (Supplementary Table 1). We examined the microglia-specific deletion of AXL in adult Cx3cr1^CreER/+; Axl ^fl/fl mice. In contrast to those observed with the MerTK deletion, the AXL deletion did not alter the upregulation of microglial PU.1 in the injured optic nerves (Supplementary Fig. 2D and 2F). Also, microglial proliferation following pathological axonal degeneration was not affected (Supplementary Fig. 2E). Together, these results supported the critical role of MerTK in microglial response to neurodegeneration.

We investigated the molecular mechanism linking phagocytosis to microglial response. Microglia were FACS-sorted from Cx3cr1^CreER/+; Mertk ^fl/fl or control Cx3cr1^CreER/+; Mertk ^+/+ mice and in vitro cultured in the presence of phosphatidylserine to trigger the phagocytic signal. Cultured microglia expressed the classic markers, e.g., TMEM119 and IBA1, confirming their cellular identity (Fig. 2A). While the IRF8 upregulation was evident in Cx3cr1^CreER/+; Mertk ^+/+ microglia, it became mitigated in Cx3cr1^CreER/+; Mertk ^fl/fl microglia (Fig. 2B, C). Previous studies have documented that MerTK can engage downstream mTOR and phospholipase C signals^56–58. We thus FACS-sorted microglia from wild-type mice and in vitro treated the specific mTOR inhibitor, rapamycin, or the phospholipase C inhibitor, U-73122. The phospholipase C inhibition effectively reduced microglial IRF8 expression, whereas the mTOR blockage showed no significant effect (Fig. 2D, E). We further examined whether the phospholipase C signal would be sufficient to mimic the MerTK-triggered phagocytic signal. Cx3cr1^CreER/+; Rosa26-LSL-hM3D(Gq) mice were bred for the chemogenetic activation of phospholipase C in microglia without the necessity of optic nerve injury (Fig. 2H). Indeed, the chemogenetic ligand clozapine N-oxide (CNO) elicited the microglial upregulation of PU.1 (Fig. 2I, K) and IRF8 (Fig. 2J, L) in the optic nerves of Cx3cr1^CreER/+; Rosa26-LSL-hM3D(Gq) mice. In contrast, CNO did not alter microglial PU.1 or IRF8 expression in the optic nerves of control Rosa26-LSL-hM3D(Gq) mice, ruling out any non-specific action of this chemogenetic ligand.

Fig. 2. MerTK-triggered phospholipase C signal is sufficient to induce microglial PU.1 and IRF8 expression.

A Microglia were FACS-sorted from the optic nerves of 8-week-old C57BL/6 wild-type female mice (3 mice) and in vitro cultured in the presence of phosphatidylserine. The cells were visualized by the co-immunostaining of TMEM119 and IBA1, and representative images are shown. B, C 8-week-old Cx3cr1^CreER/+; Mertk^fl/fl and control Cx3cr1^CreER/+; Mertk ^+/+ female mice were subjected to the 4-hydroxytamoxifen-induced Cre-recombinase activity. Microglia were FACS-sorted and in vitro cultured in the presence of phosphatidylserine. Microglial expression of IRF8 was assessed by co-immunostaining with CD11b. B Representative images are shown. C IRF8⁺ microglia were quantified. n = 3 mice per condition, mean ± SEM, two-sided Student’s t test. D–G Microglia were FACS-sorted from 8-week-old C57BL/6 wild-type female mice (3 mice per condition) and in vitro cultured in the presence of phosphatidylserine. The cells were then treated with the mTOR inhibitor rapamycin or the phospholipase C inhibitor U-73122. Microglial expression of IRF8 and phospho-SMAD2 (p-SMAD2) was assessed by co-immunostaining with CD11b. Representative images of IRF8 (D) or p-SMAD2 (F) are shown. Mean fluorescence intensities of IRF8 (E) and p-SMAD2 (G) in microglia were quantified. n = 50 cells, mean ± SEM, one-way ANOVA test. H–L 8-week-old Cx3cr1^CreER/+; Rosa26-LSL-hM3D(Gq) and control Rosa26-LSL-hM3D(Gq) female mice were subjected to the 4-hydroxytamoxifen-induced Cre-recombinase activity and then treated with the chemogenetic ligand CNO. H Diagram of the experimental procedure. Note that the mice were not subjected to the model of optic nerve injury. Microglial expression of PU.1 (I) or IRF8 (J) in the optic nerves was assessed by co-immunostaining with CD11b, and representative images are shown. PU.1⁺ microglia (K) and IRF8⁺ microglia (L) were quantified. n = 5 mice for Rosa26-LSL-hM3D(Gq) and n = 6 mice for Cx3cr1^CreER/+; Rosa26-LSL-hM3D(Gq), mean ± SEM, two-sided Student’s t test.

TGFβ1 autocrine signal regulates microglial response to neurodegeneration

We investigated the downstream action of MerTK-triggered PU.1 and IRF8 in microglial response in the model of optic nerve injury. The chromatin immunoprecipitation-sequencing (ChIP-seq) datasets of PU.1- or IRF8-binding sites in mouse microglia, as reported in our previous work⁴⁵, were exploited for analyses. While the MerTK signal could induce PU.1 and IRF8 expression, the two transcription factors did not control Mertk or other genes of phosphatidylserine-dependent phagocytosis pathways (Supplementary Fig. 3A). Studies have demonstrated the essential functions of TGFβ signaling in microglial development and functions^59–61, and we observed that the gene locus of Tgfb1 was directly targeted by both PU.1 and IRF8 (Fig. 3A). In contrast, the loci of other TGFβ family members, e.g., Tgfb2 and Tgfb3, exhibited no prominent PU.1- or IRF8-binding sites. Also, while previous works by colleagues and us showed that LRRC33 participated in the presentation of the latent TGFβ1 protein in microglia^62,63, PU.1 and IRF8 did not target the gene loci of Lrrc33 or its related family member Lrrc32 (Fig. 3A). In addition, the two TGFβ receptor subunits, Tgfbr1 and Tgfbr2, were not subjected to the transcriptional control of PU.1 or IRF8 (Fig. 3A).

Fig. 3. IRF8 and PU.1 target microglial TGFβ1 expression in pathological axonal degeneration.

A 8-week-old C57BL/6 wild-type female mice were subjected to the model of optic nerve injury. IRF8- or PU.1-targeted sites on the indicated gene loci in microglia were assessed by ChIP-seq analyses. B, C 8-week-old Cx3cr1^CreER/+; Irf8 ^fl/fl and control Cx3cr1^CreER/+; Irf8^+/+ female mice were subjected to the 4-hydroxytamoxifen-induced Cre-recombinase activity and then utilized for the model of optic nerve injury. Microglial TGFβ1 expression in the optic nerves at 3 days post-injury was assessed by co-immunostaining with CD11b. B Representative images are shown. C Microglia densities and TGFβ1⁺ microglia were quantified. n = 6 mice per condition, mean ± SEM, two-way ANOVA test. D–FTgfb1^AA/AA mice were generated in which the composite motif (TTTCACTTCC) of the IRF8- and PU.1-targeted site on the Tgfb1 gene was mutated (TATCACTACC). D Diagram of the gene targeting to obtain the Tgfb1^AA/+ allele. E, F 8-week-old Tgfb1^AA/AA and control Tgfb1 ^+/+ female mice were subjected to optic nerve injury. Microglial TGFβ1 expression in the optic nerves at 3 days post-injury was assessed by co-immunostaining with CD11b. E Representative images are shown. F TGFβ1⁺ microglia were quantified. n = 6 mice for Tgfb1 ^+/+ and n = 5 mice for Tgfb1^AA/AA, mean ± SEM, two-way ANOVA test.

We thus examined the PU.1/IRF8-targeted TGFβ1 expression. The microglia-specific deletion of IRF8 was achieved in adult Cx3cr1^CreER/+; Irf8 ^fl/fl mice, which helped circumvent the known developmental deficits of microglia with the IRF8 deletion^64,65. In the control Cx3cr1^CreER/+; Irf8^+/+ mice, there was a profound upregulation of microglial TGFβ1 expression in the injured optic nerves (Fig. 3B, C). Of note, because TGFβ1 is located within the endosomal system before its secretion, its immunostaining appeared in the cytoplasm of microglia^66,67. Such TGFβ1 expression in microglia was mostly abolished in the injured optic nerves of Cx3cr1^CreER/+; Irf8^fl/fl mice (Fig. 3B, C), supporting its transcriptional regulation by IRF8. Also, microglial proliferation in the injured optic nerves markedly decreased with this adult deletion of IRF8 (Fig. 3C), consistent with previous studies^42,43,45. Moreover, GO enrichment analyses showed that, similar to the MerTK deletion, microglia from the injured optic nerves of Cx3cr1^CreER/+; Irf8^fl/fl mice had a downregulation of central pathways in microglial activation, including the TGFβ pathway (Supplementary Fig. 4E). In addition, the IRF8 deletion reduced the expression of signature genes for microglial response to neurodegeneration (Fig. 5F) and inflammatory cytokines and chemokines (Fig. 5G) in the injured optic nerves. In parallel, we sought to examine the PU.1 function in microglial TGFβ1 expression. However, due to the indispensable role of PU.1 in the survival of myeloid-lineage cells^64,68,69, the microglia-specific deletion of PU.1 in adult Cx3cr1^CreER/+; Pu.1^fl/fl mice caused the widespread loss of microglia in the optic nerves, precluding the accurate assessment in this mouse model as we previously reported⁴⁵. In further support of the PU.1/IRF8-targeted TGFβ1 expression, we generated Tgfb1^AA/AA mice in which the composite motif of PU.1-IRF8 targeting site on the Tgfb1 gene, i.e., TTTCACTTCC, was mutated to TATCACTACC (Fig. 3D). Notably, this mutation of the PU.1-IRF8 targeting site did not affect the baseline levels of microglial TGFβ1 expression (Fig. 3F). On the other hand, the neurodegeneration-induced TGFβ1 expression was diminished in the injured optic nerves of Tgfb1^AA/AA compared to control Tgfb1^+/+ mice (Fig. 3E, F).

Fig. 5. An essential role of TGFβ1 autocrine signal in microglial response to pathological axonal degeneration.

A–C 8-week-old Cx3cr1^CreER/+; Tgfbr1 ^fl/fl and control Cx3cr1^CreER/+; Tgfbr1^+/+ female mice were subjected to the 4-hydroxytamoxifen-induced Cre-recombinase activity and then utilized for the model of optic nerve injury. Microglial phospho-SMAD2 (p-SMAD2) in the optic nerves at 3 days post-injury was assessed by co-immunostaining with CD11b. A Representative images are shown. Microglia densities (B) and p-SMAD2⁺ microglia (C) were quantified. n = 3 mice per condition, mean ± SEM, two-way ANOVA test. D 8-week-old Cx3cr1^CreER/+; Tgfb1 ^fl/fl, Cx3cr1^CreER/+; Tgfbr1 ^fl/fl, and control Cx3cr1^CreER/+ female mice were subjected to the 4-hydroxytamoxifen-induced Cre-recombinase activity and then utilized for the model of optic nerve injury. Microglia were FACS-sorted from the injured optic nerves and analyzed by RNA-seq. Pooled RNA-seq data from 4 mice per condition are shown for GO enrichment analysis of microglia from the injured optic nerves of control Cx3cr1^CreER/+ versus Cx3cr1^CreER/+; Tgfb1 ^fl/fl mice and control Cx3cr1^CreER/+ versus Cx3cr1^CreER/+; Tgfbr1 ^fl/fl mice. E–G 8-week-old Cx3cr1^CreER/+; Irf8 ^fl/fl, Cx3cr1^CreER/+; Tgfb1 ^fl/fl, Cx3cr1^CreER/+; Tgfbr1 ^fl/fl, and control Cx3cr1^CreER/+ female mice were subjected to the 4-hydroxytamoxifen-induced Cre-recombinase activity and then utilized for the model of optic nerve injury. Microglia were FACS-sorted from the optic nerves and analyzed by RNA-seq. Pooled RNA-seq data from 4 mice per condition are shown. Expression levels of homeostatic markers (E), signature genes for microglial response to neurodegenerative insults (F), and proinflammatory cytokines and chemokines (G) in microglia of the indicated conditions.

We next investigated the role of microglia-derived TGFβ1 in pathological axonal degeneration. Cx3cr1^CreER/+; Tgfb1 ^fl/fl mice were bred for the microglia-specific deletion of TGFβ1 in adult mice. The neurodegeneration-induced TGFβ1 expression in microglia became abrogated in the optic nerves of Cx3cr1^CreER/+; Tgfb1 ^fl/fl mice, as assessed by immunostaining (Fig. 4A, D). Meanwhile, we noticed remaining TGFβ1 expression in the injured optic nerves of Cx3cr1^CreER/+; Tgfb1 ^fl/fl mice, which partially overlapped with GFAP⁺ processes, likely implicating astrocytes as a non-microglial source of TGFβ1 (Supplementary Fig. 3B). Of importance, microglial proliferation evidently decreased in the injured optic nerves of Cx3cr1^CreER/+; Tgfb1 ^fl/fl mice (Fig. 4C), suggesting that TGFβ1 could be an autocrine signal to promote microglial response. In support of this notion, while phospho-SMAD2 (p-SMAD2), a central downstream component of TGFβ signaling, was mostly absent in the uninjured optic nerves, it emerged exclusively in microglia following injury (Fig. 4B, E). Because p-SMAD2 is a transcriptional factor activated by the TGFβR signaling, its immunostaining is primarily located within microglial nuclei⁶⁷. This appearance of microglial p-SMAD2 was strongly suppressed in Cx3cr1^CreER/+; Tgfb1 ^fl/fl mice (Fig. 4B, E).

Fig. 4. Microglial TGFβ1 acts as an autocrine signal in pathological axonal degeneration.

A–E 8-week-old Cx3cr1^CreER/+; Tgfb1 ^fl/fl and control Cx3cr1^CreER/+; Tgfb1^+/+ female mice were subjected to the 4-hydroxytamoxifen-induced Cre-recombinase activity and then utilized for the model of optic nerve injury. Microglial expression of TGFβ1 (A) or phospho-SMAD2 (p-SMAD2) (B) in the optic nerves at 3 days post-injury was assessed by co-immunostaining with CD11b, and representative images are shown. Microglia densities (C), TGFβ1⁺ microglia (D), and p-SMAD2⁺ microglia (E) were quantified. n = 3 mice for Cx3cr1^CreER/+; Tgfb1^+/+ and n = 4 mice for Cx3cr1^CreER/+; Tgfb1 ^fl/fl, mean ± SEM, two-way ANOVA test.

We further pursued the genetic deletion of the TGFβ receptor subunits in microglial response to pathological axonal degeneration. Notably, the gene loci of Tgfbr2 (68.39 cM) and Cx3cr1 (71.37 cM) are both on chromosome 9 of the mouse genome, with their genetic linkage of 2.98 cM. As a result, the second round of gene targeting was essential to generate the Cx3cr1^CreER/+; Tgfbr2 ^fl/+ allele (Supplementary Fig. 3C). The RNA-seq of FACS-sorted microglia from the uninjured optic nerves of Cx3cr1^CreER/+; Tgfbr2^fl/fl mice confirmed a significant loss of Tgfbr2 (Supplementary Fig. 3D). This blockage of TGFβ receptor signaling abrogated the neurodegeneration-induced appearance of microglial p-SMAD2 (Supplementary Fig. 3E and 3G). Also, microglial proliferation in the injured optic nerves of Cx3cr1^CreER/+; Tgfbr2^fl/fl mice was profoundly reduced (Supplementary Fig. 3F). In parallel, the critical function of TGFβ1 signaling in microglial response was verified by the specific deletion of TGFβR1. The RNA-seq of FACS-sorted microglia from the uninjured optic nerves of Cx3cr1^CreER/+; Tgfbr1^fl/fl mice confirmed the successful depletion of Tgfbr1 (Supplementary Fig. 3D). Similar to the findings with the TGFβR2 deletion, the neurodegeneration-induced p-SMAD2 in microglia (Fig. 5A, C) and microglial proliferation (Fig. 5B) became blunted in the injured optic nerves of Cx3cr1^CreER/+; Tgfbr1^fl/fl compared to control Cx3cr1^CreER/+; Tgfbr1^+/+ mice. Moreover, in support of an upstream action of MerTK, microglial p-SMAD2 was significantly inhibited in the injured optic nerves of Cx3cr1^CreER/+; Mertk^fl/fl mice (Supplementary Fig. 2B). In addition, the inhibition of phospholipase C but not mTOR reduced the p-SMAD2 levels of in vitro cultured wild-type microglia (Fig. 2F, G). GO enrichment analyses showed that similar to the MerTK or IRF8 deletion, microglia from the injured optic nerves of Cx3cr1^CreER/+; Tgfb1 ^fl/fl or Cx3cr1^CreER/+; Tgfbr1^fl/fl mice exhibited a significant downregulation of microglial activation-related pathways, including the TGFβ pathway as expected (Fig. 5D). Also, the expression of signature genes for microglial response to neurodegeneration (Fig. 5F) and inflammatory cytokines and chemokines (Fig. 5G) was mitigated in microglia from the injured optic nerves of Cx3cr1^CreER/+; Tgfb1 ^fl/fl and Cx3cr1^CreER/+; Tgfbr1^fl/fl mice. Intriguingly, in contrast to observation with the MerTK deletion (Fig. 1H), the microglia-specific deletion of IRF8, TGFβ1, or TGFβR1 did not affect the neurodegeneration-induced reduction of homeostatic markers in microglia (Fig. 5E), implicating that additional pathways may be responsible for this cellular event. Together, these results revealed a critical role of MerTK-triggered PU.1/IRF8-targeted TGFβ1 autocrine signal in microglial response to pathological axonal degeneration.

Recent studies have shown that the TGFβ1 deletion in adult mouse microglia disrupts homeostasis and induces inflammatory-like phenotypes in brain regions^66,67. However, we observed that the genetic blockage of IRF8-targeted TGFβ1 autocrine signal did not affect microglial density or morphology in the uninjured optic nerves of adult Cx3cr1^CreER/+; Irf8 ^fl/fl (Fig. 3C), Cx3cr1^CreER/+; Tgfb1 ^fl/fl (Fig. 4A, C), Cx3cr1^CreER/+; Tgfbr1^fl/fl (Fig. 5A, B), and Cx3cr1^CreER/+; Tgfbr2^fl/fl (Supplementary Fig. 3E and 3F), at least in the timeframe of our experimental setup. Also, the RNA-seq analyses confirmed that microglial expression of homeostatic markers (Supplementary Fig. 4A) and activation-associated markers of microglia (Supplementary Fig. 4B and 4C) did not significantly differ in the uninjured optic nerves of Cx3cr1^CreER/+; Irf8 ^fl/fl, Cx3cr1^CreER/+; Tgfb1 ^fl/fl, and Cx3cr1^CreER/+; Tgfbr1^fl/fl. In addition, overall transcriptomic profiles of microglia in the uninjured optic nerves were mostly unaffected by those genetic manipulations (Supplementary Fig. 4D). To reconcile such discrepancies, we assessed microglia in brain regions, i.e., cerebral cortex and hippocampus, of adult Cx3cr1^CreER/+; Mertk^fl/fl, Cx3cr1^CreER/+; Tgfb1^fl/fl, Cx3cr1^CreER/+; Tgfbr1^fl/fl, and Cx3cr1^CreER/+; Tgfbr2^fl/fl mice. Consistent with published reports^66,67, the homeostatic marker P2RY12 was significantly diminished in microglia of those mouse lines (Supplementary Fig. 5A–5C). Also, microglial expression of PU.1 (Supplementary Fig. 6A to 6C) and IRF8 (Supplementary Fig. 6D to 6F) became elevated in the cerebral cortex and the hippocampus of adult Cx3cr1^CreER/+; Mertk^fl/fl, Cx3cr1^CreER/+; Tgfb1^fl/fl, Cx3cr1^CreER/+; Tgfbr1^fl/fl, and Cx3cr1^CreER/+; Tgfbr2^fl/fl mice, supporting an inflammatory-like phenotype. These results appeared in agreement with the notion that white matter microglia in the optic nerves may display distinct expression patterns compared to those in other brain regions^70–72.

Microglial TGFβ1 autocrine signal in Alzheimer’s disease

We explored the disease relevance of TGFβ1 autocrine signal in microglial response to neurodegeneration. We first examined the 5×FAD mice, a common model of Alzheimer’s disease^73,74. Reminiscent of the observations in the model of optic nerve injury, there was an upregulated expression of PU.1 (Fig. 6A–D) and IRF8 (Fig. 6E–H) in microglia surrounding Aβ plaques in the cerebral cortex and the hippocampus of 9-month-old or 15-month-old 5×FAD mice compared to 24-month-old wild-type mice. Also, microglial TGFβ1 expression trended higher in those brain regions of 9-month-old 5×FAD mice and was significantly elevated in 15-month-old 5×FAD mice (Supplementary Fig. 7A to 7D). Accordingly, while there was a minor presence of microglial p-SMAD2 in the cerebral cortex and the hippocampus of 24-month-old wild-type mice, its levels profoundly increased around Aβ plaques in 15-month-old 5×FAD mice (Supplementary Fig. 7E to 7H).

Fig. 6. Microglial TGFβ1 signal in the mouse model of Alzheimer’s disease.

A–H Brain tissues were collected from 9-month-old or 15-month-old 5×FAD and 24-month-old C57BL/6 wild-type male mice. Microglial expression of PU.1 (A–D) and IRF8 (E–H) in the cerebral cortex or the hippocampus were examined by co-immunostaining with CD11b and Aβ. Representative images of microglial PU.1 expression at low (A) and high magnification (B) are shown. Densities of PU.1⁺ microglia in the cerebral cortex (C) and the hippocampus (D) were quantified. n = 4 mice for 24-month-old wild-type, n = 5 mice for 9-month-old 5×FAD, and n = 3 mice for 15-month-old 5×FAD, mean ± SEM, one-way ANOVA test. Representative images of microglial IRF8 expression at low (E) and high magnification (F) are shown. Densities of IRF8⁺ microglia in the cerebral cortex (G) and the hippocampus (H) were quantified. n = 4 mice for 24-month-old wild-type, n = 5 mice for 9-month-old 5×FAD, and n = 3 mice for 15-month-old 5×FAD, mean ± SEM, one-way ANOVA test.

We next looked into the MerTK-triggered TGFβ1 signal in human microglia. The single-cell RNA sequencing (scRNA-seq) datasets of the Human Protein Atlas showed MERTK as the highest expressed one among central components of phagocytosis pathways in human brain microglia. Meanwhile, AXL exhibited a lower level, and TYRO3 was barely detectable (Supplementary Table 1). The bridging molecules such as GAS6 and PROS1 were expressed by human microglia, implicating their capacity for phosphatidylserine-dependent phagocytosis. Clinical evidence has suggested that lower expression levels of PU.1 are associated with a reduced risk of Alzheimer’s disease⁷⁵. In addition, human brain microglia would express IRF8 in the condition of Alzheimer’s disease⁴⁶. We thus analyzed the published single-nucleus RNA sequencing (snRNA-seq) datasets of total cells in the prefrontal cortices of Alzheimer’s disease patients⁴⁶. After quality control, a total of 36,038 cells from the prefrontal cortices of 14 patients (information in Supplementary Table 2) were obtained. Besides neuronal populations such as excitatory and inhibitory neurons, several non-neuronal cell types could be defined, e.g., oligodendrocytes, astrocytes, microglia, and endothelial cells (Fig. 7A, B). Indeed, microglia in the snRNA-seq datasets showed prominent MERTK expression (Fig. 7C). MERTK was also expressed by astrocytes, in line with astrocytes being another type of phagocytic cells in the CNS^52–54. Of importance, we revealed the distinct expression of PU.1 and IRF8 in the microglia of Alzheimer’s disease patients (Fig. 7C). Moreover, though TGFB1 was distributed among neuronal and non-neuronal cells, its expression was significantly enriched in microglia (Fig. 7C). In parallel, we replicated those findings in a separate collection of snRNA-seq datasets of Alzheimer’s disease patients⁷⁶. After quality control, 16,868 cells pooled from 23 patient samples (information in Supplementary Table 2) were utilized for analyses, among which the microglial population was defined (Fig. 7D, E). Again, MERTK was expressed by both microglia and astrocytes (Fig. 7F). In addition, PU.1 and IRF8 were evidently detected in microglia of Alzheimer’s disease patients (Fig. 7F). Moreover, microglia represented the major TGFB1-expressing population among all the neuronal and non-neuronal cells of patients’ cortical tissues (Fig. 7F). In light of those snRNA-seq analyses, we finally examined the cortical tissue sections of control donors and Alzheimer’s disease patients (information in Supplementary Table 3). Immunofluorescence staining revealed that while microglial TGFβ1 expression appeared low or hardly present in control donors (Supplementary Fig. 8A), its levels became enriched in microglia of Alzheimer’s disease patients (Supplementary Fig. 8B). More importantly, we detected the profound p-SMAD2 immunostaining within microglia of Alzheimer’s disease patients but rarely in those of control donors (Fig. 8A–C). Of note, other non-neuronal cells, e.g., astrocytes or oligodendrocytes, might also respond to TGFβ1 and exhibit p-SMAD2 immunostaining. Together, these results have supported the TGFβ1 autocrine signal of human microglia in response to Alzheimer’s disease.

Fig. 7. Microglial TGFβ1 expression in the prefrontal cortices of Alzheimer’s disease patients.

A–C The published snRNA-seq datasets of total cells in the prefrontal cortices of Alzheimer’s disease patients (syn21670836, information in Supplementary Table 2) were analyzed. A UMAP plot of defined neuronal and non-neuronal cell types. B Violin plots of the expression of signature genes in defined cell types. OPCs, oligodendrocyte progenitor cells. C The expression of MERTK, PU.1, IRF8, or TGFB1 was projected onto the UMAP plot of defined cell types. D–F The published snRNA-seq datasets of total cells in the prefrontal cortices of Alzheimer’s disease patients (syn18681734, information in Supplementary Table 2) were analyzed. A UMAP plot of defined neuronal and non-neuronal cell types. B Violin plots of the expression of signature genes in defined cell types. OPCs, oligodendrocyte progenitor cells. C The expression of MERTK, PU.1, IRF8, or TGFB1 was projected onto the UMAP plot of defined cell types.

Fig. 8. Microglial TGFβ1 signal in the prefrontal cortices of Alzheimer’s disease patients.

A–C Microglial expression of phospho-SMAD2 (p-SMAD2) in the cerebral cortex of control donors or Alzheimer’s disease patients (information in Supplementary Table 3) was examined by co-immunostaining with IBA1. Representative images of control donors (A) or Alzheimer’s disease patients (B) at low and high magnification are shown. C Mean fluorescence intensity of p-SMAD2 in microglia was quantified. n = 8 patients per condition, mean ± SEM, two-sided Student’s t test. D TGFβ1 autocrine signal regulates microglial response to neurodegeneration.

Discussion

Extensive research has established the broad involvement of microglial phagocytosis in neurodegenerative diseases^{32,36,77–79}. However, the reciprocal regulation of microglial functions by a specific phagocytic signal has yet to be comprehensively characterized. Previous studies by colleagues and us have identified the cooperative action of PU.1 and IRF8 as a central mechanism underlying microglial response to neurological insults^42–46. Our current work has built on this finding to show that the phagocytic receptor MerTK can engage PU.1 and IRF8, revealing a functional link between phagocytosis and this transcriptional module. Whether a similar connection exists for other microglial phagocytosis pathways, e.g., those mediated by TREM2 or complement receptors, is a tempting possibility to explore. Notably, Ca²⁺ is an essential second messenger in the phospholipase C signal. Previous reports have suggested the critical roles of Ca²⁺ in microglia under disease conditions^8,80,81. For instance, damage to a single neuron would cause rapid Ca²⁺ transients in surrounding microglia via P2Y receptors⁸². Also, the microglia-specific induction of intracellular Ca²⁺ could exaggerate neuroinflammation and neuropathic pain⁸³. In addition, activating the mechanosensitive Piezo1 channel in microglia by Aβ plaques triggered Ca²⁺ signals that stimulated cell migration⁸⁴. Given our finding that the phospholipase C signal is sufficient to upregulate PU.1 and IRF8, whether those Ca²⁺-inducing stimuli may converge on the PU.1-IRF8 transcriptional mechanism warrants in-depth examination.

TGFβ signaling has been documented to be indispensable for microglial development and function, with various cellular sources of TGFβ suggested^59–61. Our current work showed that PU.1 and IRF8 can directly target the TGFβ1 gene expression, which acts as an autocrine signal in the context of neurodegeneration. We note that due to the challenge of obtaining the microglia of optic nerves for ChIP-seq analyses, particularly from a mutant mouse line, the mutation of the IRF8-PU.1 targeting site in the Tgfb1 gene remains to be further characterized. Nevertheless, the microglia-specific deletion of TGFβ1 or its receptors TGFβR1 or TGFβR2 diminished their response to pathological axonal degeneration in adult mice. Intriguingly, such manipulations of TGFβ1 autocrine signaling did not affect the neurodegeneration-induced downregulation of homeostatic markers in microglia, implicating that additional pathways may be responsible for this cellular event.

Recent reports showed that the TGFβ1 deletion in microglia disrupted their homeostasis in the brain of adult mice, leading to inflammatory-like phenotypes^66,67. Consistent with this notion, we observed that the genetic blockage of the MerTK-IRF8-TGFβ1 signaling axis led to microglial upregulation of PU.1 and IRF8 in brain regions. This opposite effect likely reflects the fact that white matter microglia in the optic nerves may display distinct expression patterns compared to those in other CNS regions^70–72. In addition, it is conceivable that TGFβ1 can exert a bidirectional role in microglia under homeostatic versus disease conditions.

In further support of the clinical relevance of microglial TGFβ1 autocrine signal, we demonstrated that it also occurred in the 5×FAD mouse model of Alzheimer’s disease, as well as in Alzheimer’s disease patients. Whether this signaling axis participates in other neurological disorders, e.g., Parkinson’s disease, frontotemporal dementia, or amyotrophic lateral sclerosis, awaits further clinical investigations. Moreover, microglial response has been implicated in several psychiatric diseases, e.g., major depressive disorder^85,86, and we recently observed the involvement of PU.1-IRF8 transcriptional action in a mouse model of depressive-like behaviors⁸⁷. Whether the MerTK-triggered phagocytosis pathway and TGFβ1 autocrine signal may be engaged in those mental illnesses calls for future research attention.

In sum, this study has delineated a molecular pathway of phagocytosis-triggered TGFβ1 autocrine signaling in the microglial response to neurodegeneration (Fig. 8D) with broad diagnostic and therapeutic implications.

Methods

Human brain samples

Paraffin sections of the cortical tissues of control donors (2 males and 6 females; 66- to 93-year-old) or Alzheimer’s disease patients (3 males and 5 females; 78- to 88-year-old) were obtained from the Chinese Academy of Medical Sciences—Peking Union Medical College Human Brain Bank (information in Supplementary Table 3) in compliance with informed consent from each patient and the protocol approved by the Institutional Review Board of Peking University (#IRB00001052-22104).

Mouse information

All the experimental procedures in mice were performed in compliance with the protocol approved by the Institutional Animal Care and Use Committee (IACUC) of Peking University (#LSC-YangJ-3). Mice were maintained on the 12-hr/12-hr light/dark cycle (light period 7:00 am ~ 7:00 pm) at 22 – 26 °C and relative humidity of 40 ~ 60%, with the chow diet and water available ad libitum. Mice utilized in the experiments were males or females of 8 to 12 weeks old, unless otherwise specified.

All the mouse lines were on the C57BL/6 background. C57BL/6 wild-type mice were purchased from Charles River International. Cx3cr1^CreER/+ (#020940), Tmem119^CreER/+(#031820), Rosa26-LSL-tdTomato (#007914), Rosa26-LSL-hM3D(Gq)-DREADD (#026220), and 5×FAD (#034848) mice were from the Jackson Laboratory. H11-ZsGreen (#T006163) mice were from GemPharmatech. Mertk ^fl/+ (loxP sites flanking the exon 2, #S-CKO-03713), Axl ^fl/+ (loxP sites flanking the exon 5 and exon 6, #S-CKO-09532), Irf8 ^fl/+ (loxP sites flanking the exon 5, #S-CKO-03026), and Tgfbr1^fl/+ (loxP sites flanking the exon 5 and exon 6, #S-CKO-06237) were from Cyagen Inc. Tgfb1^fl/+ mice (loxP sites flanking the exon 2, #110164) were from Biocytogen Pharmaceuticals Co. Tgfbr2 ^fl/+ mice (loxP sites flanking the exon 4, #NM-CKO-200228) were from the Shanghai Model Organisms Center Inc.

The gene loci of Tgfbr2 (68.39 cM) and Cx3cr1 (71.37 cM) are on chromosome 9 of the mouse genome, with their genetic linkage of 2.98 cM. As a result, it was necessary to conduct the second round of gene targeting to generate the Cx3cr1^CreER/+; Tgfbr2 ^fl/+ allele. A targeting vector that contained the loxP sites flanking the exon 4 of Tgfbr2 was constructed and microinjected with the CRISPR/Cas9 system into the Cx3cr1^CreER/CreER fertilized oocytes. The resulting offspring were screened for the targeted Cx3cr1^CreER/+; Tgfbr2 ^fl/+ allele and backcrossed with C57BL/6 wild-type mice for two generations. Cx3cr1^CreER/+; Tgfbr2 ^fl/+ mice were then bred with Tgfbr2 ^fl/+ to obtain Cx3cr1^CreER/+; Tgfbr2 ^fl/fl mice for experiments.

For the disruption of the composite motif (TTTCACTTCC) of the IRF8-PU.1 targeting site on the Tgfb1 gene, a targeting vector carrying the mutations (g.25406417 T > A and g.25406423 T > A) was constructed. The linearized vector was microinjected with the CRISPR/Cas9 system into the fertilized oocytes of C57BL/6 wild-type mice. The resulting offspring were screened for the targeted Tgfb1^AA/+ allele (TATCACTACC), which was backcrossed with C57BL/6 wild-type mice for two generations and then bred to produce Tgfb1^AA/AA and control Tgfb1^+/+ littermates for experiments.

Mouse procedures

For the induction of Cre-recombinase activity in the mouse lines carrying Cx3cr1^CreER/+, 4-hydroxytamoxifen (Sigma-Aldrich) was formulated in DMSO / Kolliphor-EL / 5% sucrose (v:v:v = 1:3:6) and daily administered via oral gavage at 20 mg/kg of body weight for 8 days. The control condition received the vehicle formulation without 4-hydroxytamoxifen. The mice were utilized for experiments 4 weeks after the last gavage unless otherwise specified.

For the chemogenetic manipulation, clozapine N-oxide (CNO) dissolved in sterile phosphate-buffered saline (PBS) was intraperitoneally administered to mice at 2.5 mg/kg of body weight every 12 h for three times. 12 h after the last injection, the mice were subjected to tissue harvesting.

For the model of optic nerve injury, the mice were anesthetized with 3% isoflurane, and the topical antibiotic ointment was applied to both eyes. An incision was made on the superior conjunctiva of the left eye, and the optic nerve was exposed by a pair of blunt forceps. The crush injury was performed for 5 s using a pair of fine-tip forceps (Fine Science Tools) at 1 mm distal from the eyeball. The sham surgery included all the steps except that the optic nerve was untouched⁸⁸.

For the parabiosis procedure, each pair of mice was housed together for 1 week before the surgery. The mice were anesthetized with 3% isoflurane, and the skin on one side of each mouse was shaved and prepared with iodine and alcohol. A longitudinal incision was made along the side of each mouse, and the skin was carefully separated from the underlying connective tissues. A longitudinal incision of approximately 10 mm was then made on the exposed peritoneum of each mouse. The incision sites of the two mice were sutured together to establish the connection of the vascular systems. In addition, the scapulae on the incision sides of the mice were sutured together to help hold the parabiotic pair. The skin incisions of the mice were closed together by surgical staples. The mice were utilized for experiments 2 weeks after the procedure.

Fluorescence-activated cell sorting (FACS)

The peripheral blood of each mouse was collected via retro-orbital bleeding into PBS containing 5 mM Na-EDTA (pH 8.0). The cells were centrifuged at 500 × g for 5 min, and ammonium-chloride-potassium (ACK) buffer was added to lyse red blood cells. The cells were centrifuged again at 500 × g for 5 min and re-suspended in Hank’s balanced salt solution (HBSS; Thermo Fisher Scientific) containing 10 mM EDTA-Na (pH 8.0) and 2% heat-inactivated fetal bovine serum (HI-FBS; Sigma-Aldrich) for staining with PE/Cyanine7 anti-mouse CD45 (1:200, BioLegend, #103114, RRID:AB_312979). The stained cells were processed on the BD LSRFortessa, and the FACS data were analyzed by FlowJo (https://www.flowjo.com).

RNA sequencing (RNA-seq)

Mice of the indicated conditions were anesthetized with 3% isoflurane and euthanized by cervical dislocation. Optic nerves were dissected and minced into small pieces on ice. The tissues were digested in RPMI 1640 (Thermo Fisher Scientific) / 0.1 mg/ml Liberase TL (Roche) / 20 μg/ml DNase I (Sigma-Aldrich) / 10 mM HEPES / 3% HI-FBS at 37 ^oC for 15 min. The tissues were mashed through a 70-μm cell strainer and centrifuged at 500 × g for 5 min. The cells were re-suspended in PBS / 2% HI-FBS / 10 mM EDTA-Na (pH 8.0) and stained with PE/Cyanine7 anti-mouse CD45 (1:200, BioLegend, #103114, RRID:AB_312979) and FITC anti-mouse CD11b (1:200, BioLegend, #101206, RRID:AB_312788). FACS-stained cells were processed on the Beckman MoFlo Astrios EQ, and microglia were defined as CD45^low CD11b⁺ by the common standard⁵⁵.

Microglia FACS-sorted from uninjured or injured optic nerves were subjected to RNA-seq library preparation. Total RNAs were extracted and reverse-transcribed using a template-switching oligo (BGI Genomics). cDNAs then underwent pre-amplification and tagmentation to add sequencing adapters, and the index PCR was performed to incorporate sample-specific barcodes. The libraries were purified using VAHTS DNA Clean Beads (Vazyme) and assessed for quality and concentration using the Qubit (Thermo Fisher Scientific) and the Bio-Fragment Analyzer (BiOptic). The libraries were sequenced on the MGI DNBSEQ-T7 platform using 150 bp paired-end reads. Sequencing files were aligned to the reference mouse genome using STAR (v2.7.11a) (https://code.google.com/archive/p/rna-star/). Gene expression levels were quantified with FeatureCounts included in the Subread package (https://subread.sourceforge.net/). Gene expression levels were quantified as transcripts per million (TPM) or fragments per kilobase per million (FPKM) and were visualized with the heatmap package (https://bioconductor.org/packages/release/bioc/html/heatmaps.html) in R. Differentially expressed genes in microglia of the indicated conditions were analyzed for Gene Ontology (GO) enrichment using the R package clusterProfiler (version 4.12.6) (https://bioconductor.org/packages//release/bioc/html/clusterProfiler.html). Significantly enriched biological processes were presented as dot plots with ggplot2. Pearson correlation coefficients were calculated by the R package corrplot (https://github.com/taiyun/corrplot).

Quantitative real-time PCR (qPCR)

For the qPCR analyses of gene expression in optic nerves, mice of the indicated conditions were anesthetized with 3% isoflurane and euthanized by cervical dislocation. Optic nerves were dissected, and total RNAs were extracted by the RNeasy Mini Kit (Qiagen), reverse-transcribed by the PrimeScript RT Reagent Kit with gDNA Eraser (Takara), and analyzed by the SYBR Green Real-Time PCR Kit (Thermo Fisher Scientific). Cyclophilin mRNA levels were utilized as the internal control. qPCR primers in this study were Cyclophilin-Forward: TGGAGAGCACCAAGACAGACA; Cyclophilin-Reverse: TGCCGGAGTCGACAATGAT; Mertk-Forward: CTCGGGGCACATCATTCA; Mertk-Reverse: TACGACCCATTGTCTGAGCG.

Immunofluorescence staining

Mice of the indicated conditions were anesthetized with 3% isoflurane and transcardially perfused with PBS / 50 μg/ml heparin, followed by PBS / 1% PFA / 50 μg/ml heparin. The brain or optic nerves were dissected, and the tissues were post-fixed in PBS / 1% PFA at 4 °C overnight and then cryopreserved in PBS / 30% sucrose at 4 °C overnight for 8-μm cryosectioning. The sections were immunostained with the intended primary antibodies in PBS / 0.1% Tween-20 / 5% normal donkey serum. The primary antibodies used for mouse tissue sections were rat anti-CD11b (1:1000, BD Biosciences, #553308, RRID:AB_394772), rabbit anti-MerTK (1:500, Thermo Fisher Scientific, #14-5751-82, RRID:AB_2688282), rabbit anti-IRF8 (1:500, Cell Signaling Technology, #98344, RRID:AB_3083757), rabbit anti-PU.1 (1:500, Cell Signaling Technology, #2258, RRID:AB_10693421), rabbit anti-TGFβ1 (1:500, Thermo Fisher Scientific, #PA1-29032, RRID:AB_2202039), rabbit anti-phospho-SMAD2 (1:500, Cell Signaling Technology, #18338, RRID:AB_2798798), chicken anti-GFAP (1:500, Abcam, #ab4674, RRID:AB_304558), rat anti-P2RY12 (1:1000, BioLegend, #848002, RRID:AB_2650634), and mouse anti-Aβ (1:1000, BioLegend, #803001, RRID:AB_2564653). The tissue sections were further immunolabeled with the corresponding secondary antibodies Alexa Fluor 488-conjugated donkey anti-rabbit IgG (1:500, Thermo Fisher Scientific, #A32790, RRID:AB_2762833), Alexa Fluor 568-conjugated donkey anti-rat IgG (1:500, Thermo Fisher Scientific, #A78946, RRID:AB_2910653), Alexa Fluor 568-conjugated donkey anti-chicken IgY(H + L) (1:500, Thermo Fisher Scientific, #A78950, RRID:AB_2921072), and Alexa Fluor 647-conjugated donkey anti-mouse IgG (1:500, Thermo Fisher Scientific, #A-31571, RRID:AB_162542).

For the immunostaining of human cortical tissues, 5-μm paraffin sections were incubated with Biodewax and Clear Solution (Servicebio) for 10 min three times and dehydrated in 100% ethanol for 5 min three times. For antigen retrieval, the sections were heated at 98^oC for 8 min, cooled down for 8 min, and heated at 98 ^oC for 7 min in EDTA Antigen Retrieval Buffer (Servicebio). The sections were then washed with PBS for 5 min three times and immunostained with the intended primary antibodies in PBS / 0.1% Tween-20 / 5% normal donkey serum at room temperature overnight. The primary antibodies used for human brain sections were rat anti-IBA1 (1:500, Abcam, #ab283346, RRID:AB_3065282), rabbit anti-TGFβ1 (1:500, Thermo Fisher Scientific, #PA1-29032, RRID:AB_2202039), and rabbit anti-phospho-SMAD2 (1:500, Thermo Fisher Scientific, #44-244 G, RRID:AB_2533614). The tissue sections were further immunolabeled with Alexa Fluor 488-conjugated donkey anti-rabbit IgG (1:500, Thermo Fisher Scientific, #A32790, RRID:AB_2762833) and Alexa Fluor 568-conjugated donkey anti-rat IgG (1:500, Thermo Fisher Scientific, #A78946, RRID:AB_2910653).

For low-magnification images, immunostained mouse or human tissue sections were scanned on a Zeiss Axio Scan.Z1 slide scanner using a 4× objective. High-magnification images were acquired on a Nikon A1R laser-scanning confocal microscope controlled by NIS-Elements software with a 60× oil-immersion objective. DAPI, Alexa Fluor 488, Alexa Fluor 568, and Alexa Fluor 647 were excited with the 405-, 488-, 561-, and 640-nm laser lines, respectively. Confocal images were obtained at 1024×1024 pixels with the pinhole set to 1.2 Airy units. Laser power and detector gain/offset were kept constant for all the images within each experiment. Microglial densities and the immunofluorescence features of PU.1, IRF8, TGFβ1, and phospho-SMAD2 were quantified in ImageJ (https://imagej.net/ij).

In vitro treatments of microglia

Mice of the indicated conditions were anesthetized with 3% isoflurane and euthanized by cervical dislocation. Brain tissues were dissected and minced into small pieces on ice. The tissues were digested in RPMI 1640 / 0.1 mg/ml Liberase TL / 20μg/ml DNase I / 10 mM HEPES / 3% HI-FBS at 37 ^oC for 20 min. The tissues were mashed through a 70-μm cell strainer and centrifuged at 500 × g for 5 min. Cell pellets were re-suspended in 4 ml of 37% Percoll (GE Healthcare) and added on top of 4 ml of 70% Percoll in a 15-ml tube. An additional 4 ml of 30% Percoll and 2 ml HBSS were sequentially overlain. Percoll gradients were centrifuged at 340 × g for 20 min, and the cell layers between 37% and 70% Percoll were carefully collected. The cells were diluted with HBSS / 2% HI-FBS / 10 mM EDTA-Na (pH 8.0) and centrifuged at 500 g for 5 min. The cells were re-suspended in HBSS / 2% HI-FBS / 10 mM EDTA-Na (pH 8.0) and stained with PE/Cyanine7 anti-mouse CD45 (1:200, BioLegend, #103114, RRID:AB_312979) and FITC anti-mouse CD11b (1:200, BioLegend, #101206, RRID:AB_312788). CD45^low CD11b⁺ microglia were sorted on BD FACSAria Fusion.

Microglia were cultured in 24-well plates pre-coated with poly-L-lysine and the phagocytosis-ligand phosphatidylserine in RPMI 1640 / 1× insulin-transferrin-selenium (Thermo Fisher Scientific) / 1× non-essential amino acids (Thermo Fisher Scientific) / 100U/ml penicillin / 100μg/ml streptomycin / 25 ng/ml M-CSF (Peprotech). Microglia were rested for 2 h and then treated with a final concentration of 100 nM rapamycin or 1μΜ U-73122. The cells were fixed with 2% PFA at 12 hr post-culture and immunostained with rat anti-CD11b (1:1000, BD Biosciences, #553308, RRID:AB_394772), rat anti-IBA1 (1:1000, Abcam, #ab283346, RRID:AB_3065282), rabbit anti-TMEM119 (1:1000, Cell Signaling Technology, #90840, RRID:AB_2928137), rabbit anti-IRF8 (1:1000, Cell Signaling Technology, #98344, RRID:AB_3083757), and rabbit anti-phospho-SMAD2 (1:1000, Cell Signaling Technology, #18338, RRID:AB_2798798). The cells were further immunolabeled by Alexa Fluor 488-conjugated donkey anti-rabbit IgG (1:500, Thermo Fisher Scientific, #A32790, RRID:AB_2762833) and Alexa Fluor 568-conjugated donkey anti-rat IgG (1:500, Thermo Fisher Scientific, #A78946, RRID:AB_2910653).

Immunostained microglia were imaged on a Nikon A1R laser-scanning confocal microscope controlled by NIS-Elements software with a 60× oil-immersion objective. DAPI, Alexa Fluor 568, and Alexa Fluor 647 were excited with the 405-, 561-, and 640-nm laser lines, respectively. Confocal images were obtained at 1024×1024 pixels with the pinhole set to 1.2 Airy units. Laser power and detector gain/offset were kept constant for all the images within each experiment. Microglial densities and the immunofluorescence features of IRF8 and phospho-SMAD2 were quantified in ImageJ (https://imagej.net/ij).

Chromatin immunoprecipitation sequencing (ChIP-seq)

ChIP-seq analyses of PU.1- or IRF8-targeted sites in mouse microglia of injured optic nerves were reported by our previous study⁴⁵. The ChIP-seq datasets were available in the Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra) with the accession numbers SRR6963590 (PU.1) and SRR6963589 (IRF8). The datasets were mapped to the mouse genome (mm9) with Bowtie 2 (https://bowtie-bio.sourceforge.net/index.shtml). PU.1- or IRF8-targeted sites on the indicated gene loci were visualized by Integrative Genomics Viewer_2.16.2 (https://igv.org/).

Single-nucleus RNA sequencing (snRNA-seq)

The published snRNA-seq datasets of the total cells of prefrontal cortices of Alzheimer’s disease patients, syn18681734⁷⁶ and syn21670836⁴⁶, were obtained from the Synapse platform (www.synapse.org). We selected the sequencing data of patients with a final consensus cognitive score of 4 or greater for analyses (information in Supplementary Table 2). For the syn18681734 dataset, data preprocessing was performed according to the manufacturer’s instructions (https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/ advanced/references), and CellRanger (v7.0.1) was used for sequence alignment and gene expression quantification. For the syn21670836 dataset, the gene count files were available from the original provider. For quality control, we applied filters for genes expressed in 10 or fewer cells, cells expressing fewer than 200 genes, mitochondrial reads exceeding 10%, or ribosomal genes for more than 20%. We further employed Solo⁸⁹ in scvi-tools to remove doublets, using the top 2000 variant genes and default parameters.

We utilized single-cell Variational Inference (scVI)⁹⁰ in scvi-tools to integrate the snRNA-seq data of different patients, following the software instructions (https://docs.scvi-tools.org/en/stable/tutorials/index.html). SCANPY⁹¹ was used for clustering the integrated data, and the resulting H5ad files were converted into the Seurat objects using Sceasy (https://github.com/cellgeni/sceasy). Finally, we performed the analyses in RStudio with R v4.2.25, utilizing Seurat v.4’s FindAllMarkers⁹² to determine and visualize the markers for each cluster. For the determination of cell types, we referred to PanglaoDB⁹³ and the information provided in the original datasets. The feature plots were generated using scCustomize (https://samuel-marsh.github.io/scCustomize/).

Statistical methods

Student’s t test (two-sided) or ANOVA (one-way or two-way with post hoc tests) was performed by GraphPad Prism (http://www.graphpad.com/scientific-software/prism). Student’s t test was used to compare two groups, and ANOVA was utilized for multiple group comparisons.

All the sample points (n) shown in the figures represent biological replicates, i.e., culture wells, mice, or patients. The results were reported as mean ± SEM (standard error of the mean). The exact value of n and the statistical test description are included in the figure legends, and the exact p-values are presented in the figures where appropriate.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Y.H., J.Y., and Y.C. conceived and designed this study. Y.H., Z.Z., K.F., J.Y., and Y.C. performed the experiments and analyzed the results. F.L. and H.Z. processed the ChIP-seq datasets. Z.D. and J.C. analyzed the RNA-seq, scRNA-seq, and snRNA-seq datasets. Y.H., J.Y., and Y.C. prepared the manuscript.

Peer review

Peer review information

Nature Communications thanks Gabriel McKinsey and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The source data generated in this study are provided in the Supplementary Information/Source Data file. Source data are available for Figs. 1C-1F, 2C, 2E, 2G, 2K, 2L, 3C, 3F, 4C-4E, 5B, 5C, 6C, 6D, 6G, 6H, 8C, S2E, S2F, S3F, S3G, S5B, S5C, S6B, S6C, S6E, S6F, S7C, S7D, S7G, and S7H. The RNA-seq data generated in this study have been deposited in the Sequence Read Archive database under accession code PRJNA1380057. The ChIP-seq datasets of PU.1 and IRF8 generated by our previous study are available in the Sequence Read Archive database under accession numbers SRR6963590 and SRR6963589, respectively. The published snRNA-seq datasets re-analyzed in this study are available in the Synapse platform database under accession numbers syn18681734 (https://www.synapse.org/Synapse:syn18681734) and syn21670836 (https://www.synapse.org/Synapse:syn21670836). Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-69189-3.