Gene-Expression Profile Implicates eIF2-, mTOR-, and Hypoxia-Signaling Pathways in the Phenotypic Shift from A1 to A2 Cell Populations.

Expression patterns for 660 genes underwent a significant change in the transition of Fib-L cells to Myofib-L cells [t tests with P < 0.05 and fold change of median transcripts per million (TPM) >2 or <0.5]. Pathway analysis showed that differentially expressed genes (DEGs) were clearly enriched in 26 pathways (SI Appendix, Fig. S4), with nine pathways showing a significant predicted activity pattern (Fig. 2A). The top enriched pathways, i.e., eIF2 signaling, mTOR signaling, and hypoxia signaling in the cardiovascular system, mediate responses to hypoxic stress. Given a higher proportion of Myofib-L in FD eyes, this change indicated that these hypoxia-related signal pathways may be activated in FD eyes.

Open in a separate window

Fig. 2.

Pathways and transcription factors underlying gene-expression changes during the transition of A1 to A2. (A) Pathways with enrichment P < 0.01 and absolute activation Z-score >2 are shown. Almost all pathways were significantly activated in the A2 population except for PPAR/RXRα, which was inhibited. (B) Hierarchical cluster analysis of the A2 population. The x axis indicates the genes with large CV² values, and the y axis indicates the single scleral cells in the A2 population. The A2 population contained four subpopulations [the color gradient from yellow to red indicates the cosine distance (CD) to the A1 cell type, with those in yellow color being the closest to A1 cells and the red-colored ones being the farthest away]. Light to dark orange subpopulations represent intermediate cells. (C) Expression dynamics of Hif1a among different populations and subpopulations. The x axis is ranked by CD to A1, with the nearest subpopulation on the left (A2_1) and the farthest subpopulation on the right (A2_4). The A2_2 and A2_3 subpopulations are interspersed between them. Data are expressed as medians (interquartile ranges); *P < 0.05, Wilcoxon-rank test. (D) Relationship between human pathologic myopia-risk genes (red ovals) and genes in the HIF-1α–signaling pathway (blue ovals) according to the PPI network. The risk gene for pathologic myopia with the most interactions was GATA4, which is an important transcription activator involved in the developmental process. SNPs in this gene were reported to be associated with myopia in a French population (27).

To validate the signaling pathways identified from scRNA-seq analysis in myopic scleras, we used RT-PCR to detect mRNA expression levels of eight DEGs from the top three pathways (SI Appendix, Table S4) in mice with 2 d of FD. In agreement with scRNA-seq analysis, scleral mRNA levels of these eight genes in the FD eyes were significantly increased compared with those in the control eyes (SI Appendix, Fig. S5A). Differential expression profiles of these genes after FD were specific to the sclera because expression in the retina remained unchanged (SI Appendix, Fig. S5B). Additionally, we compared the scRNA-seq data with that generated using bulk scleral tissue RNA sequencing (RNA-seq) to show similarities in expression features (SI Appendix, Fig. S6). The apparent consistency gave us more confidence in the scRNA-seq data. These data confirm that eIF2-, mTOR-, and hypoxia-signaling pathways in scleral fibroblasts are involved in myopia development.

To find the key modulators of the transition from Fib-L to Myofib-L, we divided the Myofib-L population into four subpopulations based on the results of hierarchical clustering (Fig. 2B). These four subpopulations were further ranked by expression cosine distance with the Fib-L population. This kind of ranking may indicate the process of the transition of Fib-L into Myofib-L in the sclera. Next, we employed Hypergeometric Optimization of Motif Enrichment software on 864 DEGs among the four subpopulations to identify candidate transcription factors mediating this transition. Thirty-six transcription factors were identified (SI Appendix, Table S5), including Smad4 and Hif1a, which are effective transcription factors of the TGF-β–signaling pathway and the hypoxia-inducible factor-1α (HIF-1α) pathway, respectively. Seven transcription factors were expressed with a TPM >2 in at least one subpopulation, while only Hif1a was expressed over a dynamic range in association with the anticipated Fib-L–to–Myofib-L transition, with the highest level of expression occurring at the last stage (Fig. 2C). These results suggest that HIF-1α plays an important role in mediating the transition from fibroblasts to myofibroblasts.

The HIF-1α–Signaling Pathway Is Related to Genes Associated with Human Myopia.

We next explored possible interactions between the top three significant signaling pathways and the development of myopia in humans. Using protein–protein interaction (PPI) networks, we investigated interactions of the risk genes of human myopia (SI Appendix, Table S6), determined in a previous genome-wide association study and linkage analysis (28, 29), with genes involved in these pathways. Of 145 myopia-risk genes, only nine had interactions with the eIF2-signaling pathway, and six had interactions with the mTOR pathway. On the other hand, nearly one-third of the myopia-risk genes interacted with genes involved in the HIF-1α–signaling pathway (45/145) (SI Appendix, Fig. S7). Of 27 pathologic myopia-risk genes, only three interacted with the eIF2-signaling pathway and one with the mTOR-signaling pathway. However, genes associated with pathologic myopia showed a borderline significance, with 10 of the 27 risk genes interacting with the HIF-1α–signaling pathway (P = 0.07, hyper-geometric distribution test) (Fig. 2D). Significant numbers of interactions were identified between genes in the HIF-1α pathway and the risk genes of myopia (mean interactions = 3.925, P = 0.001, bootstrapping) and pathologic myopia (mean interactions = 3.2, P = 0.036, bootstrapping).

Although myopia-risk genes exhibited no significant interactions with DEGs between the Fib-L and Myofib-L populations (P = 0.68, hyper-geometric distribution test), 27 pathologic myopia-risk genes had significant interactions with 40 of the 660 DEGs (P = 0.00467, hyper-geometric distribution test) (SI Appendix, Table S7). In addition, 6 of 141 human myopia risk genes (PTPN5, NFIA, GNL2, KCNMA1, TBC1D23, and SPTAN1) that have mouse homologs were found in DEGs between the Fib-L and Myofib-L (P = 0.0409, bootstrapping). However, none of the 27 human pathologic myopia-risk genes was observed in the DEGs. These findings emphasize that genes involved in the scleral cellular phenotypical changes in mouse model of myopia are strongly implicated in human myopia development.

Localized Scleral Hypoxia Is a Common Feature of Myopia.

Given that hypoxia is a common factor among the risk genes of human pathologic myopia, we further determined the dynamic changes in HIF-1α protein, which is an indicator of tissue hypoxia (30), in mice with 2 d and 2 wk of FD. Scleral HIF-1α levels in the FD eyes were significantly increased compared with those in the control eyes after 2 d and 2 wk of FD (Fig. 3 A and B). No significant differences in retinal HIF-1α levels were detected between FD and control eyes at these two time points (Fig. 3 C and D). As expected, increased axial elongation was observed after 2 wk of FD (SI Appendix, Fig. S8). Such findings indicate the development of localized scleral hypoxia during myopia development.

Open in a separate window

Fig. 3.

Dynamic changes in the HIF-1α protein level during myopia development in mice. Western blot analysis of scleral (A and B) and retinal (C and D) HIF-1α expression in mice in response to FD for 2 d and 2 wk. Only right eyes from age-matched normal animals are shown. The bar graphs represent relative levels of HIF-1α from experiments (n = 3–6). Data are expressed as mean ± SEM. *P < 0.05; **P < 0.01; Student’s t test.

Furthermore, we determined if scleral hypoxia is a common feature of myopia by measuring the changes in HIF-1α protein level in response to two different myopia-induction paradigms: FD and negative lens induction (LI) in guinea pigs. Both myopic stimuli significantly increased the scleral HIF-1α levels compared with those in the control eyes after 2 d and 1 wk of myopia induction and then declined to a level similar to that in the control eyes after 2 d of recovery (Fig. 4 A–D). Similar to mice, we did not detect significant changes in the retinal HIF-1α levels between the myopia-induced (FD or LI) eyes and their respective fellow controls at these time points (Fig. 4 E–H) despite the dynamic changes in eye growth at the above time points (SI Appendix, Fig. S9). These results strengthen our hypothesis that scleral hypoxia is a common feature of myopia.

Open in a separate window

Fig. 4.

Dynamic changes in HIF-1α protein level during myopia development in guinea pigs. (A–D) Western blot analysis of scleral HIF-1α expression in guinea pigs in response to FD (A and B) or LI (C and D) for 2 d or 1 wk and 2 d after recovery (Rec) from 1 wk of myopia induction. (E–H) Corresponding retinal HIF-1α expression levels in response to FD (E and F) or LI (G and H) at these time points. Only right eyes from age-matched normal animals are shown. Bar graphs represent HIF-1α levels from experiments (n = 3–4). Data are expressed as mean ± SEM. *P < 0.05; ***P < 0.001; Student’s t test.

Hypoxia Exposure-Induced Myofibroblast Transdifferentiation in Human Scleral Fibroblasts.

To further explore the role of scleral hypoxic involvement in myopia, we used Western blots to examine myofibroblast transdifferentiation and collagen production in human scleral fibroblasts (HSFs) after reducing the levels of ambient oxygen to 5% O₂. In HSFs exposed to the hypoxic condition, there was increased expression of HIF-1α protein (Fig. 5A), the focal adhesion proteins vinculin (Fig. 5B) and paxillin (Fig. 5C), and the myofibroblast marker α-SMA (Fig. 5D). Simultaneously, there was a decrease in the expression of COL1A1 (Fig. 5E). Taken together, these results indicate that hypoxia may affect myopia development via scleral myofibroblast transdifferentiation.

Open in a separate window

Fig. 5.

Myofibroblast transdifferentiation in response to hypoxia in vitro. (A) HSFs exposed to 5% O₂ were analyzed by Western blot for the protein expression of HIF-1A, vinculin, paxillin, α-SMA, and COL1A1. (B–E) Bar graphs represent vinculin (B), paxillin (C), α-SMA (D), and COL1A1 (E) levels, respectively (n = 4). Data are expressed as mean ± SEM. *P < 0.05; **P < 0.01. Treatment duration: 2, 4, 7, or 10 h vs. control (0 h); Student’s t test.

Antihypoxia Drugs Inhibited the Progression of Experimental Myopia Without Affecting Normal Ocular Growth in Guinea Pigs.

We further asked if amelioration of hypoxia affects myopia progression. The FD eyes of guinea pigs were given periocular injections of the antihypoxic drugs salidroside (31) or formononetin (32). After the injection of salidroside (10 µg per eye) or formononetin (5.0 µg per eye) for 4 wk, the scleral levels of HIF-1α, COL1α1, and α-SMA were determined by Western blot (Fig. 6 A and B). Injection of salidroside suppressed the up-regulation of HIF-1α (Fig. 6A). It also suppressed the down-regulation of COL1α1 in FD eyes (Fig. 6A). There were no changes in α-SMA levels in FD and control eyes in either the vehicle- or salidroside-injected group (Fig. 6A). Similarly, injection of formononetin suppressed the up-regulation of HIF-1α and the down-regulation of COL1α1 in FD eyes (Fig. 6B), while the α-SMA levels in FD and control eyes remained unchanged in both the vehicle- and the formononetin-injected group (Fig. 6B).

Open in a separate window

Fig. 6.

Effect of antihypoxic drugs on myopia in guinea pigs with FD for 4 wk. (A and B) Protein levels of scleral HIF-1α, COL1α1, and α-SMA were detected by Western blot after periocular injection in FD eyes of salidroside(Salid) (10 μg per eye) (A) or formononetin (Formo) (5 μg per eye) (B) for 4 wk. Bar graphs represent HIF-1α, COL1α1, and α-SMA levels from experiments (n = 4–6). Data are expressed as mean ± SEM. *P < 0.05; Student’s t test. (C–H) Interocular differences (injected eye minus fellow untreated eye) in refraction (C and F), axial length (D and G), and vitreous chamber depth (E and H) in FD guinea pigs before and after 4 wk of treatment with normal saline (vehicle control, n = 13), salidroside,1 μg per eye (n = 10) or 10 μg per eye (n = 16) (C–E) or with 0.1% dimethyl sulfoxide (vehicle control, n = 18) or formononetin, 0.5 μg per eye (n = 19) or 5 μg per eye (n = 17) In F–H, data are expressed as mean ± SEM. *P < 0.05; **P < 0.01; ***P < 0.001; two-way repeated-measures ANOVA with Bonferroni multiple comparison. AL, axial length; D, diopter; DMSO, dimethyl sulfoxide; NS, normal saline; VCD, vitreous chamber depth.

Importantly, daily injections of salidroside at the higher dose (10 μg per eye) for 4 wk significantly inhibited the development of myopia in FD guinea pigs compared with the lower dose (1 μg per eye) and vehicle injection (Fig. 6C). Elongation of axial length and vitreous chamber depth was also significantly inhibited with a higher dose of salidroside compared with the lower dosage and vehicle injection (Fig. 6 D and E). Daily injection with formononetin (0.5 μg per eye and 5.0 μg per eye) for 4 wk significantly inhibited myopia at both dosages (Fig. 6F). The elongation of axial length was also significantly inhibited at these dosages of formononetin (Fig. 6G), while the increase in the vitreous chamber depth showed a nonsignificant inhibition (Fig. 6H). In contrast, injecting normal guinea pigs with salidroside or formononetin did not induce any changes in refraction, axial length, or vitreous chamber depth (SI Appendix, Fig. S10). Collectively, these results indicate that scleral hypoxia plays critical roles during myopia development and that amelioration of hypoxia can suppress myopia progression through the down-regulation of HIF-1α and restoration of scleral collagen levels.

scRNA-Seq Reveals a Scleral Phenotypic Switch from Fib-L to Myofib-L via Activation of Hypoxic Signal During Myopia Development.

Here we used single-cell analysis to resolve the mechanism of myofibroblast transdifferentiation involved in myopia development. Traditional population-based genome-wide approaches are incapable of resolving this unsynchronized cellular phenotype shift process. In our current study, to obtain a sufficient number of cells for the scRNA-seq analysis, we separately pooled the FD scleral tissues and the control scleral tissues from six to eight animals to produce each suspension. This procedure could have limited the power of analysis. Despite this, we took advantage of the unique ability of the scRNA-seq technique to circumvent the cellular heterogeneity of sclera tissues. Thus, we identified two cellular populations with distinct gene-expression profiles and found a significant difference between them during myopia progression. Specifically, we found that the proportion of Myofib-L cells increased in sclera during myopia development, as evidenced by an increase in α-SMA expression. This is consistent with a previous report showing up-regulation of scleral α-SMA in myopic tree shrews (35). Scleral ECM remodeling occurs as a consequence of increases in myofibroblast transdifferentiation and is associated with the down-regulation of Col1a1 and Col1a2 gene expression, whereas degradative proteinases undergo up-regulation leading to increases in collagen turnover (20). These findings suggest an important role of the fibroblast-to-myofibroblast transdifferentiation during myopia development. This view is supported by an in vitro finding that imposed stress mediates myofibroblast transdifferentiation in fibroblast-populated collagen lattices that mimic the scleral microenvironment during myopia development (35). Also, an in vivo study showed that increasing the intraocular pressure for an extended period of time inhibited the ocular elongation in tree shrews, whose sclera is characterized by the presence of myofibroblasts (36). The transdifferentiated myofibroblast population in the sclera provides an offsetting contractile force that resists the persistent outward-directed force of intraocular pressure.

Notably, the myofibroblast phenotypic changes during myopia development are different from those occurring during wound healing (37). Unlike the transient type of myofibroblasts observed during wound healing, the myofibroblasts transdifferentiated during myopia development express lower levels of Col1a1 and Col1a2. During wound healing, myofibroblasts exhibit a profibrotic phenotype with up-regulation of ECM components such as fibronectin and collagen. This phenotypic difference may be due to variations in the structural architecture of granulation tissues, i.e., wound healing is characterized by the formation of new capillaries containing vital nutrients (38), as opposed to the avascular scleral stroma. Thus, even though the scleral myofibroblasts acquire a contractile phenotype, the ability to mediate ECM remodeling is dependent on environmental cues (i.e., scleral hypoxia microenvironment).

Multiple lines of experimental evidence establish scleral hypoxia as a critical contributing factor in myopia development. First, scRNA-seq analysis revealed that myopia development is associated with the transdifferentiation of fibroblast to myofibroblast, whose major functional differences arise from activation of hypoxia-related pathways, including eIF2 signaling, mTOR signaling, and hypoxia signaling (Fig. 2A). Many genes that are involved in the HIF-1α–signaling pathway are associated with risk genes involved in human myopia or pathologic myopia (Fig. 2D and SI Appendix, Fig. S7). Second, while Guo et al. (39, 40) showed the possible involvement of scleral HIF-1α mRNA in myopia, our results showed that scleral HIF-1α protein levels increased in response to myopia induction in both mice and guinea pigs at 2 d (Figs. 3 and ​and4).4). Although we did not detect a change in HIF-1α levels in retina, we cannot exclude the possibility that the retina might undergo hypoxia-induced changes if the treatment persists longer. Third, in vitro exposure to hypoxia triggered the transdifferentiation of HSFs with increased expression of α-SMA, paxillin, and vinculin, along with reduced COL1A1 expression (Fig. 5), suggesting that the increase in scleral myofibroblast formation during myopia progression is associated with scleral hypoxia. Our results are consistent with previous studies that showed that hypoxia can enhance myofibroblast transdifferentiation in other tissues, such as skin (41), artery adventitia (42), and nasal polyps (43). Finally, antihypoxic treatments inhibited myopia development along with reduced levels of scleral HIF-1α (Fig. 6). Collectively, these findings suggest that the scleral microenvironment becomes hypoxic during myopia development, which triggers myofibroblast transdifferentiation, leading to reduced collagen biosynthesis.

The hypoxic signals may act as a common mediator for several signaling pathways during myopia. Our results showed that the elevated phosphorylation levels of eIF2α and mTOR were also suppressed by the antihypoxic drugs salidroside and formononetin in FD guinea pigs, suggesting the interaction between the hypoxic signal and the eIF2α and mTOR signal during myopia. Studies in renal tubular cells showed that hypoxia can enhance myofibroblast transdifferentiation through their interaction with TGF-β (44). As TGF-β is considered critical for myopia development (20), it may regulate scleral ECM remodeling through hypoxic signaling. Further experiments are needed to clarify the potential functional interactions of HIF-1α and TGF-β in scleral ECM remodeling.

Hypoxia May Be the Long-Sought Myopic Visual Signal Transduction Mechanism.

Demonstration of scleral hypoxia during myopia development raises an important question: How can myopic visual signal transduction events mediate scleral hypoxia? Wallman and Winawer (12) speculated that visual signals are transduced in the retina and choroid and then are translated to produce mediators that affect myofibroblast transdifferentiation and ECM remodeling in the sclera. These changes then weaken the scleral framework, thereby accommodating increases in ocular elongation. The choroidal layer is uniquely positioned to translate and transfer myopic visual signals to the sclera, where they could affect ECM remodeling and ocular growth. However, the identity of the mediators acting between the choroid and sclera is unknown. Numerous studies have shown an association between myopia and choroidal thinning (45). In humans, myopic eyes with higher dioptric values have thinner choroids than those in fellow eyes with lower diopter values (46), and prolonged near work elicits an accommodation process that is associated with myopia development and choroidal thinning (47, 48). Choroidal thinning in humans with high myopia is also associated with reductions in both its stromal and vascular components (49). Further, changes in choroidal blood flow precede changes in choroidal thickness subsequent to a recovery period from FD in animal models (50). In addition, the increase of α-SMA⁺ cells in response to FD appeared mostly in the scleral layer close to choroid (51), suggesting the role of choroid in the transdifferentiation of scleral fibroblasts. Thus, altered choroidal vasculature and decreased choroidal blood flow may cause scleral hypoxia that triggers scleral myofibroblast transdifferentiation and altered scleral biomechanics.

Taking these findings together, we propose that myopia-related visual signals cause a decrease in choroidal capillary patency and blood flow, resulting in reduced levels of oxygen and nutrient supply to the neighboring avascular sclera (Fig. 7). Scleral hypoxia thus ensues, promoting myofibroblast transdifferentiation along with declines in collagen production through the accumulation of HIF-1α. These changes contribute to scleral thinning and weakening, ultimately resulting in excessive axial elongation. Further experiments are needed to clarify the detailed mechanisms between choroid and sclera.

Open in a separate window

Fig. 7.

Paradigm for myofibroblast transdifferentiation involved in the pathogenesis of myopia. (A) The choroidal layer thins rapidly followed by scleral thinning that results from ECM remodeling during myopic visual stimulus. (B) These myopia-related visual signals decrease choroidal blood flow that leads to an insufficient supply of oxygen and nutrients to the avascular sclera. Scleral fibroblasts first sense and quickly respond to the altered extracellular microenvironment through the accumulation of HIF-1α and enhanced phosphorylation levels of eIF2α and mTOR. This induces a phenotypic shift of fibroblast-like cells toward myofibroblast-like cells as well as a decrease in collagen production. As a result, the sclera becomes thinner and weaker. Consequently, axial length increases, and myopia ensues.

Animal Experiments.

The treatment and care of the animals was conducted according to the Association for Research in Vision and Ophthalmology statement for the use of animals in ophthalmic and vision research, and the protocol for handling animals was approved by the Animal Care and Ethics Committee at Wenzhou Medical University, Wenzhou, China (approval number: WYDW2014-0015).

Experimental myopia in mice (male C57BL/6 mice, about 3 wk old) was induced by monocular FD as previously reported (59). Two days after FD, the whole sclera of each eye was harvested separately for scRNA-seq analysis. To validate the changes in DEGs from scRNA-seq analysis, we used RT-PCR to examine eight genes (SI Appendix, Tables S4 and S8) in the scleras and retinas from mice after 2 d of FD and from age-matched normal mice. To validate the major finding relating to the role of hypoxia in myopia, we used Western blots to determine the dynamic changes in HIF-1α protein in retinas and scleras from mice after 2 d or 2 wk of FD. As it is unlikely that changes in refractive state and axial length will be apparent after a brief period of FD (2 d), and given that the mice may develop cataract or keratitis in response to anesthetic injections (ketamine) given twice in 2 d, we measured the refraction and ocular biometrics only at 2 wk after FD. These parameters were measured in independent sets of mice or guinea pigs so that the ocular tissues from those animals undergoing HIF-1α protein-expression assays were not affected by normal visual stimulus during ocular measurements before harvest.

In another set of experiments aimed at identifying the effects of the specific eIF2- and mTOR-signaling pathways on myopia development, FD mice were injected i.p. daily with GSK2606414, an eIF2α kinase PERK inhibitor, or everolimus, an mTOR inhibitor, for 2 wk. Age-matched normal mice were injected i.p. daily with salubrinal, an eIF2α dephosphorylation inhibitor, or MHY1485, a mTOR phosphorylation activator, for the same duration. Refraction was measured before and after 2 wk of injections. The effects of GSK2606414 and MHY1485 on the phosphorylation levels of eIF2α and mTOR, respectively, in the scleras were evaluated by Western blots in another set of FD and normal mice after two daily injections.

In addition, we used guinea pigs (Cavia porcellus, the English short hair stock, about 3 wk old) as a second in vivo model of induced myopia to validate the role of hypoxia signaling. Experimental myopia was induced via monocular FD or −4.00 diopter negative LI as previously reported (60, 61). The changes in HIF-1α protein expression were also examined in the scleras and retinas from guinea pigs exposed to FD or LI for 2 d or 1 wk and after 2 d of recovery from 1 wk of myopia induction. The refraction and ocular biometrics were repeatedly measured in an independent set of guinea pigs at the above time points. The larger eyes of guinea pigs enabled periocular injection of the antihypoxic drugs salidroside and formononetin in FD animals for 4 wk to determine if they ameliorated the induced myopia and altered the expression of hypoxia-related molecules. The effects of these drugs on normal ocular development were also determined in age-matched normal animals for 4 wk.

Experimental Hypoxia in Vitro in HSFs.

To investigate the potential role of hypoxia on scleral ECM remodeling, HSFs were established and used as previously described (62). The cells were grown in DMEM supplemented with 10% FBS (Gibco) and 2 mM glutamine. All cells were grown at 37 °C in a 5% CO₂, humidified incubator. Cells were seeded at a density of 2.5 × 10⁵ in 35-mm culture dishes for 24 h before hypoxic treatment. Growth medium was replaced when the cells reached 70–80% confluence; then the HSFs were exposed to the hypoxic protocol. The cells were placed in a three-gas incubator (Thermo Fisher Scientific) that maintained a low O₂ level of 5% for 2, 4, 7, or 10 h. Cells maintained under the normoxic condition (21% O₂) served as the control. At the end of the designated periods, the cells were lysed, and protein levels of HIF-1A, COL1A1, α-SMA, paxillin, and vinculin were determined by Western blot analysis.

Drug Preparation and in Vivo Injection, scRNA-Seq Procedure, RT-PCR, and Western Blot Analysis.

Details of drug preparation and in vivo injection, the scRNA-seq procedure, RT-PCR, and Western blot analysis are provided in SI Appendix, SI Materials and Methods.

The study was supported by National Natural Science Foundation of China Grants 81670886, 81422007, 81470659, 81170870, and 3147119; National Key Research and Development Program of China Grant 2016YFC20160905200; Natural Science Foundation of Zhejiang Province Grants LZ14H120001 and LQ16H120006; the Zhejiang Provincial Program for the Cultivation of High-Level Innovative Health Talents; the National Young Excellent Talents Support Program; and Chinese Academy of Sciences Strategic Priority Research Program (XDB13020500).