Identification of Meibomian gland stem cell populations and mechanisms of aging

Summary

Meibomian glands secrete lipid-rich meibum, which prevents tear evaporation. Aging-related Meibomian gland shrinkage may result in part from stem cell exhaustion and is associated with evaporative dry eye disease, a common condition lacking effective treatment. The identities and niche of Meibomian gland stem cells and the signals controlling their activity are poorly defined. Using snRNA-seq, in vivo lineage tracing, ex vivo live imaging, and genetic studies in mice, we identified markers for stem cell populations that maintain distinct regions of the gland and uncovered Hh signaling as a key regulator of stem cell proliferation. Consistent with this, human Meibomian gland carcinoma exhibited increased Hh signaling. Aged glands displayed decreased Hh and EGF signaling, deficient innervation, and loss of collagen I in niche fibroblasts, indicating that alterations in both glandular epithelial cells and their surrounding microenvironment contribute to age-related degeneration. These findings suggest new approaches to treat aging-associated Meibomian gland loss.

Graphical Abstract

Introduction

Meibomian glands (MGs) are specialized holocrine sebaceous glands (SGs) embedded in the tarsal plate of the upper and lower eyelids. These glands secrete lipid-rich meibum, which protects the eye surface by preventing tear film evaporation. In aging, MGs decrease in size and/or display reduced numbers of acini in humans and in mice, which provide a genetically manipulable model system for delineating molecular mechanisms controlling MG function. MG atrophy is associated with evaporative dry eye disease, a common condition in the aged population that can result in severe vision loss and currently has no effective therapy^1–3.

Each MG consists of multiple acini, which are connected to a central duct through short ductules. The MG central duct is a stratified squamous epithelial structure that includes basal and suprabasal layers^4,5. In MG acini, basal cells proliferate toward the center of acinus, forming meibocytes which undergo sequential differentiation and accumulate lipid content. Upon reaching the center of acinus, fully differentiated meibocytes disintegrate and release meibum which reaches the ocular surface through an orifice in the eyelid margin via the ductule and central duct^4,5. MGs develop independently of hair follicles (HFs) but share similarities with HF-associated SGs. Both types of glands secrete lipids in a holocrine manner and require constant replenishment of lipid-releasing cells from proliferating basal cells⁴.

In addition to other proposed mechanisms, including duct hyper-keratinization, immune cell infiltration, and reduced acinar basal cell proliferation⁶, MG dropout is thought to involve decreased activity of MG epithelial stem cells that normally maintain MGs by both self-renewing and yielding differentiating progeny^7,8,9–11. However, the identities of MG stem cells, the nature of their supporting niche environment, and the mechanisms that control their functions in homeostasis and aging are poorly understood¹². By contrast, single-cell analyses and lineage tracing studies in mice have shown that Lrig1+ and Lgr6+ stem cells from the HF isthmus, which connects to the opening of the SG, contribute to homeostasis of the attached SGs^13–16. Recent evidence suggests that the MG duct and acinus are separately maintained by distinct unipotent KRT14+ stem cells; however, KRT14 is ubiquitously expressed in the MG. MG ductules have also been proposed to harbor stem cells, based on the presence of label-retaining cells (LRCs) in ductules¹², but evidence that these LRCs directly contribute to MG homeostasis is lacking. Lineage tracing with a constitutive Krox20/Egr2-Cre mouse line shows that EGR2+ cells give rise to the MG during development. However, lineage tracing experiments with an inducible Egr2-Cre line to determine whether EGR2+ cells self-renew and give rise to differentiating progeny in adult MGs have not been performed. Thus, specific markers for adult MG stem cells have yet to be identified.

Stem cell activity is tightly regulated by intercellular signaling pathways^17,18. These include the Hedgehog (Hh) pathway, which controls HF stem cell activity¹⁹, regulates SG development²⁰, and promotes proliferation of undifferentiated human sebocytes in vitro²¹. Components of the Hh pathway are expressed in MGs²², and Hh signaling controls the proliferation and differentiation of rat MG epithelial cells in vitro²³; however, it is unknown whether Hh signaling regulates adult MG homeostasis and MG stem cell activity in vivo.

To begin to address these questions, we dissected tarsal plates from young and aged mice and performed single-nucleus RNA sequencing (snRNA-seq) to reveal the cell heterogeneity of adult mouse MGs and the changes in MG epithelial cells and the surrounding tissue niche that take place during aging. We used lineage tracing and ex vivo live imaging assays to test the contributions of cells expressing putative stem cell markers to MG homeostasis. These data identified multiple stem cell populations that can self-renew and produce differentiated progeny to maintain the MG duct and acinus. We found that loss of the Hh receptor Smoothened (SMO) in MG epithelial cells caused decreased proliferation and MG dropout, while forced expression of an activated form of the Hh pathway effector GLI2 (GLI2∆N) promoted expansion of MG stem cells at the expense of differentiated acini. GLI2 and other stem cell markers were expressed in human MG and showed elevated levels in MG carcinoma (MGC), suggesting that MGC involves expansion of MG stem cells. Aged versus young mouse MGs displayed fewer acinar basal cells and surrounding dermal cells expressing downstream Hh pathway genes. In parallel, aged MG exhibited lower levels of EGF signaling, decreased MG innervation, and dermal fibroblast dysfunction, identifying additional mechanisms that could contribute to MG degeneration in aging. In summary, our data uncover stem cell populations and molecular mechanisms that sustain the adult MG and are altered in aging and in human disease.

Results

snRNA-seq identifies multiple distinct populations of MG epithelial cells

The eyelid is a complex environment, containing epidermal, HF, and conjunctival cells closely adjacent to the MGs, making it challenging to isolate pure MG cells or to identify MG contributions in bulk RNA-seq analyses. To analyze transcription in MG subpopulations, and to overcome difficulties in isolating cytoplasmic mRNAs from lipid-rich acinar cells, we utilized snRNA-seq which allows for identification of cell sub-populations within a complex mix. Nuclei were analyzed from the pooled tarsal plates of four 8-week-old and four 21-month-old male mice using two replicates for each age (16 mice analyzed in total). UMAP analysis identified 34 clusters of tarsal plate cell populations (Figure 1A; Figure S1A) and confirmed similar cluster distributions between the four datasets and two conditions (Figure S1B, C, D). Cell populations were annotated based on their expression of known signature genes (Figure S1A). MG ductal basal cells (cluster 3) were marked by expression of Col17a1 and Ki67 (Figure S1A, E, F); ductal suprabasal cells (cluster 8) by high levels of Krt17 and Egr2 (Figure S1A, F); orifice cells (cluster 11) by Krt10, Flg and Lor (Figure S1A, G, H); acinar basal cells (cluster 16) by expression of Pparg, Ki67 and Slc1a3 (Figure 1A, I, J); ductular cells (cluster 17) by moderate expression of Krt17 and Lrig1, and low levels of Slc1a3 and Pparg (Figure S1A, J); differentiating meibocytes (cluster 24) by expression of Fasn and high levels of Pparg (Figure 1A, I, K); and differentiated meibocytes (cluster 18) by expression of Scd4 and lack of Fasn expression (Figure 1A, K, L). Non-MG populations included conjunctival cells (clusters 0–2, 4–7, 9, 10, 12, 13, 25, 28, 30, 31), HF epithelial cells (clusters 14, 19, 26), HF dermal papilla cells (cluster 32), eyelid dermal cells (cluster 15 and 23), immune cells (cluster 20), endothelial cells (clusters 21 and 29), melanocytes (cluster 33), and muscle cells (clusters 22 and 27).

Figure 1. Identification of distinct MG stem cell populations.

(A) UMAP plot of snRNA-seq data from pooled tarsal plate samples from four 8-week-old and four 21-month-old male mice, respectively, with two replicates for each age. (B, C) RNA Velocity (B) and pseudotime analyses (C) predictions of the differentiation trajectories of MG cellular subpopulations. RNA Velocity (B) predicts that ductular cells can differentiate towards either duct or acinus. (D) Scheme for lineage tracing. (E) Lrig1+ cells in MG ductal basal layer (yellow arrows) and acinar basal layer (white arrow) self-renew and generate progeny that contribute to MG duct (light blue arrows) and acinus (pink arrow). (F) Lgr6+ cells in MG duct (yellow arrows) and acinar basal layer (white arrow) self-renew and contribute to MG duct (light blue arrow) and acinus (pink arrow). (G) Axin2+ cells in the MG duct (yellow arrows) and acinar basal layer (white arrow) self-renew and replenish MG duct (light blue arrow) and acinus (pink arrow). (H) At 2 days, Gli2 marks acinar basal cells (white arrowheads), ductules (orange arrowhead), and meibocytes (white arrowheads); at 90 days, lineage-traced cells are present in MG duct (yellow and green blue arrowheads), ductule (orange arrowhead) and acinus (white and pink arrowheads). (I) At 2 days, Slc1a3 marks meibocytes (yellow arrowhead) and acinar basal cells (white arrowhead); at 90 days, lineage tracing labels the acinar basal layer (white arrowhead) and meibocytes (yellow arrowheads). Scale bars: (B-D), 50μm; (E, G-K), 25μm. White dashed lines in (E-I) outline MG acini and ducts. 3 mice of each genotype were analyzed in lineage tracing experiments for each line and stage. See also Figures S1-S4; Video 1; Video 2.

To verify the accuracy of these annotations, we conducted spatial transcriptomic analysis and integrated the outcomes of the two assays. These analyses showed that annotation of MG cell populations based on snRNA-seq data was congruent with their anatomical positions within the tarsal plate (Figure S1M).

Comparison of these data with published single-cell analysis of the HF-associated SG¹⁶ revealed many commonalities, including expression of low levels of Pparg in acinar basal cells. A key difference was that MG acinar basal cells and ductular cells expressed genes associated with neural guidance, including Slit3, Robo2, Sema3a and Epha7, that were not notably expressed in HF-associated SG basal cells¹⁶. In line with this, MGs, but not HF-associated SGs, are directly innervated^24–27.

We used Velocity and pseudotime analyses to analyze the snRNA-seq data for the differentiation trajectories of MG cell populations. These analyses predicted that ductal basal cells give rise to ductal suprabasal cells, while acinar basal cells produce progeny that form differentiating acinar cells, which further differentiate into differentiated acinar cells (Figure 1B, C). These predictions are in line with those for HF-associated SGs¹⁶. Interestingly, ductular cells appeared to contribute to both ductal and acinar basal cell populations (Figure 1B), suggesting that the ductules contain mixtures of stem cells fated to become either duct or acinus, and/or harbor bi-potential stem cells²⁸.

MG acinar basal cells and ductular cells express HF and SG stem cell markers

Specific markers for MG stem cells are poorly defined. As MGs share many characteristics with HF-associated SGs we interrogated the snRNA-seq data for known SG stem cell markers and validated the results using RNAscope assays on independent biological samples of adult MG. Our analysis (Figure S2A, B) showed that expression of Lrig1 in ductular cells and ductal suprabasal cells was similar to that of the putative MG stem cell marker Egr2 (Figure S2C, D). We found that acinar basal cells (cluster 16) expressed the SG stem cell markers Lgr6 (Figure S2A, B; Figure S2F, white arrow) and Slc1a329 (Figure S2A, B; Figure S1J, white arrows), and also exhibited low-level expression of Lrig1 (Figure S2A, B; Figure S2E, white arrow) and Axin2³⁰ (Figure S2A, B; Figure S2G, white arrows), along with expression of Gli2 (Figure S2A, B; Figure S2H, white arrow). Ductular cells (cluster 17) were also enriched for Gli2 expression (Figure S2A, B; Figure S2H, yellow arrows) and expressed high levels of Lrig1 (Figure S2A, B; Figure S2E, yellow arrows), together with low levels of Lgr6, Axin2, and Slc1a3 (Figure S2A, B; yellow arrows in Figure S2F, G and Figure S1J). Slc1a3 expression only localized to the acinus and ductule and was absent from the duct (Figure S2A, B; Figure S1J), like its expression in SGs²⁹. The expression of SG stem cell markers in the acinar basal layer and ductule suggested that these markers may also label stem cells within the MG.

Lrig1-, Lgr6-, Axin2-, Slc1a3-, and Gli2-expressing cells contribute to MG homeostasis

To determine whether Lrig1-, Lgr6- and Axin2-expressing cells self-renew and contribute to MG homeostasis, we performed lineage tracing assays using Lrig1-Cre^ERT2, Lgr6-Cre^ERT2, Axin2-Cre^ERT2, Slc1a3-Cre^ERT2, and Gli2-Cre^ERT2 mice carrying a Rosa26^mTmG or Rosa26^nTnG reporter allele (Figure 1D). After 2 days of tamoxifen induction, we observed Lrig1-Cre^ERT2-labeled cells in MG ducts (Figure 1E, yellow arrows) and in acinar basal cells (Figure 1E, white arrow), in line with expression of endogenous Lrig1. As the complete maturation process of meibocytes from acinar basal cells in mice takes approximately 9 days³¹, we analyzed the descendants of Lrig1-expressing cells after 120 days of tracing, which is sufficient time for labeled cells to undergo multiple rounds of cell renewal. We found that progeny of Lrig1-expressing cells were present in both MG ducts (Figure 1E, light blue arrows) and acini (Figure 1E, pink arrow), indicating that these cells can self-renew and give rise to differentiating progeny.

Similarly, we found that cells labelled by Lgr6-Cre^ERT2 were present in both MG ducts and acini after 2 days of tracing (Figure 1F); after 90 days of tracing, GFP+ descendants of Lgr6-expressing cells were present in both MG ducts and acini (Figure 1F).

Axin2-Cre^ERT2-labeled cells were detected in MG ducts and acini at 2 days after Cre induction; after 120 days, GFP+ descendants of Axin2-expressing cells were observed in MG ducts and acini (Figure 1G).

Interestingly, although at 2 days after Cre induction Gli2-Cre^ERT2-labeled cells were exclusively present in ductules (Figure 1H, orange arrowhead) and acini (Figure 1H, white and pink arrows), at 90 days, GFP+ descendants of Gli2-expressing cells not only replenished the ductule (Figure 1H, orange arrowhead) and acinus (Figure 1H, white and pink arrowheads), but also contributed to the central duct (Figure 1H, yellow and white arrowheads).

In line with snRNA-seq data, at 2 days after Cre induction Slc1a3-labeled cells were only identified in the acinar basal layer (Figure 1I, white arrowhead), meibocytes (Figure 1I, yellow arrowhead) and ductules, and not in the central duct; at 90-days after Cre induction, GFP+ descendants of Slc1a3-expressing cells were present in the acinar basal layer (Figure 1I, white arrowhead) and in meibocytes (Figure1I, yellow arrowheads), but not in the central duct. Thus, Slc1a3 marks acinar, but not duct, stem cells.

Taken together, these data demonstrate that Lrig1-, Lgr6-, Axin2-, Slc1a3, and Gli2-expressing populations contain self-renewing stem cells that persist through multiple rounds of MG cell renewal and give rise to differentiated progeny in the duct and/or acinus to maintain MG homeostasis. Furthermore, in line with the prediction from Velocity analysis that the progeny of ductular stem cells can contribute to ducts and acini (Figure 1B), lineage tracing with Gli2-Cre^ERT2 revealed the presence of descendants of Gli2-expressing ductular cells in the central duct.

Lrig1+ stem cells located in the MG ductule migrate toward the acinus

A previous study identified label-retaining cells in ductules, suggesting that ductules may contain stem cells that can replenish the acinus¹². This hypothesis was supported by Velocity analysis of our snRNA-seq data (Figure 1B). To test this further, we investigated the behavior of Lrig1+ cells in ductules by combining lineage tracing in Lrig1-Cre^ERT2 Rosa26^nTnG mice, and ex vivo live imaging. After 40 days of lineage tracing, which is sufficient for the MG to undergo several turnover cycles, we identified GFP+ cell clones in the MG acinus (Figure S3A, B, yellow arrows), ductule (Figure S3A, B, white arrows), and duct (Figure S3A, B, blue arrows). Tarsal plate explants from these mice were cultured ex vivo for 60 minutes, during which period explants remained proliferative and secreted meibum droplets (Figures S4A-C, Video S1). Time-lapse imaging of tarsal plate explants from lineage-traced Lrig1-Cre^ERT2 Rosa26^nTnG mice over 15 hours of culture identified GFP+ cells or their daughter cells from division within the ductules that moved toward the acinus and settled in the acinar basal layer (Video S2, arrowheads). Thus, Lrig1+ cells in the ductule can contribute to the acinar basal layer.

Hh signaling maintains MG homeostasis

Given our finding that MG stem cells express similar markers to SG stem cells, we hypothesized that MG and SG stem cell activity are regulated by similar mechanisms. Among several signaling pathways important for SG development and homeostasis^32,33, Hedgehog (Hh) signaling plays a critical role²⁰. Interrogation of the snRNA-seq dataset (Figure S5A) and validation using RNAscope (Figure S5B-E) showed that multiple Hh pathway components³⁴ are expressed in the MG. Shh is expressed in a sub-population of acinar basal cells (Figure S5B, green arrows); Ptch1 localizes to acinar basal cells (Figure S5C, green arrows), differentiating meibocytes (Figure S5C, white arrow), and ductular cells (Figure S5C, yellow arrow); Smo is expressed in acinar basal cells (Figure S5D, green arrow), ductular cells (Figure S5D, yellow arrow), and ductal cells (Figure S5D, pink arrow); and Gli1 is expressed in acinar basal cells (Figure S5E, green arrows), differentiating meibocytes (Figure S5E, white arrow), and ductular cells (Figure S5E, yellow arrow). Additionally, we noted expression of Ptch1, Smo and Gli1 in the surrounding stroma (Figure S5C-E, light blue arrows).

To determine whether epithelial Hh signaling is functionally required for adult MG homeostasis, we inducibly deleted the Hh co-receptor Smo in adult MG epithelium by treating Krt14-Cre^ERT2 Smo^fl/fl Rosa26^mTmG mice and Krt14-Cre^ERT2 Rosa26^mTmG littermate controls with tamoxifen for 5 days starting at 8 weeks of age. This resulted in reduced size of acini in Smo-deleted MGs analyzed at 18 or 29 weeks post-tamoxifen treatment (Figure 2A-C). Differentiation of MG acini was not obviously affected (Figure 2D-G), but Smo-deleted MGs exhibited significantly decreased proliferation of acinar basal cells compared with controls (Figure 2H, I, J). Thus, Hh signaling maintains MG acini by promoting the proliferation of acinar basal cells.

Figure 2. Hh signaling promotes and is required for normal levels of acinar basal cell proliferation.

(A-C) Whole-mount fluorescence images of MG #6 and/or MG #7 in (A) KRT14-Cre^ERT2 Rosa26^mTmG control mice 29 weeks after tamoxifen treatment (T29Weeks); (B) KRT14-Cre^ERT2 Smo^fl/fl Rosa26^mTmG mice 18 weeks after tamoxifen treatment (T18 Weeks); (C) KRT14-Cre^ERT2 Smo^fl/fl Rosa26^mTmG mice at T29 Weeks. Yellow asterisks indicate hair follicles; CD, MG central duct. (D-G) IHC for PPARγ (D, E) and FASN (F, G) showing similar expression levels in control (D, F) and Smo-deficient (E, G) MGs at 34 weeks after tamoxifen treatment. (H, I) IHC for Ki-67 showing reduced acinar basal cell proliferation in Smo-deficient MGs at 34 weeks after tamoxifen treatment (I) compared with control (H, green arrowheads). (J) Quantitation of the % of Ki-67+ cells in the acinar basal layer of control and Smo-deficient MGs. Littermate pairs were compared. Statistical significance was calculated using a paired two-tailed Student’s t-test. N=4 control and N=4 Smo-deficient mice were analyzed at each stage. At least 70 acinar basal cells were analyzed per mouse. (K-P) Progressive expansion of acinar basal cell clusters (black arrows) in GLI2∆N-expessing MG after 2 (K, L), 4 (M, N) and 7 (O, P) days of doxycycline treatment. Meibocytes were reduced by 7 days (P, green arrows). (Q-V) Progressive expansion of GLI2∆N+/KRT5+ cells (yellow arrows) and reduction of PLIN2+ meibocytes (white arrows) in Gli2∆N^K5rtTA MG acini after 2 (Q, R), 4 (S, T) and 7 (U, V) days of doxycycline treatment. (W, X) Hyperproliferation of GLI2∆N+ cells in Gli2∆N^K5-rtTA MG acini (yellow arrow). (Y) Quantification of acinar basal cell proliferation. Samples from n=5 control mice lacking K5-rtTA or tetO-GLI2∆N and samples from n=5 Gli2∆N^K5-rtTA mice were analyzed. At least 100 acinar cells were analyzed from each animal. Statistical significance was calculated with unpaired two-tailed Student’s t-test. Data are presented as mean +/− SEM. Scale bars: 50 μm. See also Figure S5.

Forced expression of GLI2∆N in the MG epithelium causes expansion of acinar basal cells

To ask whether Hh signaling is sufficient to promote MG basal cell proliferation, we generated K5-rtTA tetO-GLI2∆N (GLI2∆N^K5rtTA) transgenic mice that expressed an activated form of GLI2 (GLI2∆N) in basal epithelial cells, including those in the MG, in a doxycycline-dependent manner^35,36. GLI2∆N+ acinar basal cells emerged within 2 days of initiating doxycycline treatment (Figure 2K, L, Q, R), and began to expand, starting to replace differentiating PLIN2+ meibocytes by 4 days after induction (Figure 2M, N, S, T). After 7 days, most meibocytes had been replaced by GLI2∆N+ cells (Figure 2O, P, U, V) that showed statistically significantly increased proliferation compared to control acinar basal cells (Figure 2W-Y). Thus, hyper-activation of Hh signaling caused expansion of acinar basal cells by promoting their proliferation.

GLI2∆N promoted proliferation of Lrig1- and Lgr6-expressing cells at the expense of differentiation

To delineate the mechanisms responsible for GLI2∆N-mediated acinar basal cell expansion, we performed bulk RNA-seq on MGs laser-captured from control and GLI2∆N^K5rtTA mice after 4 days of doxycycline treatment (Figure 3A). Gene ontology (GO) enrichment analysis showed that the most significantly upregulated pathways were those involved in regulation of cell proliferation, including proliferation regulators such as Ki67 and Ccnd1 (Figure 3B, C), in line with IF results (Figure 2W-Y). Expression of Hh target genes Gli1, Ptch1, and Hhip was also significantly elevated in GLI2∆N-expressing MGs compared to controls (Figure 3C). The most significantly downregulated genes were those associated with control of lipid metabolism (Figure 3B) such as Pparg, and meibocyte differentiation including Scd3, Scd4, and Plin2 (Figure 3C). RNAscope and IF analyses confirmed increased expression of Gli1 and Ccnd1 and reduced levels of PPARγ and PLIN2 expression in GLI2∆N-expressing MGs compared to controls (Figure 3D-I). Notably, cell populations expressing MG stem cell markers Lrig1 and Lgr6 were expanded upon GLI2∆N expression in MGs (Figure 3C, J-M). Collectively, these data suggest that forced GLI2∆N expression in MG epithelium promotes proliferation, expands the number of cells expressing Lrig1 and Lgr6, and, either directly or indirectly, suppresses meibocyte differentiation.

Figure 3. Forced GLI2∆N expression expands MG stem cell populations and impedes meibocyte differentiation.

(A) Scheme for bulk RNA-seq of laser-captured MG samples. (B) GO enrichment analysis of bulk RNA-seq data from GLI2∆N^K5rtTA and littermate control MGs 4 days after induction showing the top 10 enriched pathways. (C) Volcano plot showing genes upregulated or downregulated at 4 days. N=3 GLI2∆N^K5rtTA samples and n=3 littermate controls lacking K5-rtTA or tetO-GLI2∆N; differentially expressed genes were defined as padj<0.001 and Log₂FC<−0.5 or Log₂FC>0.5. (D-G) RNAscope showing upregulation of Gli1 and Ccnd1 in GLI2∆N^K5rtTA MGs. (H-I) IF data showing decreased PPARγ and PLIN2 expression in GLI2∆N^K5rtTA acini. (J-M) RNAscope showing expanded Lrig1 and Lgr6 expression in GLI2∆N^K5rtTA MGs. Independent samples from 3 Gli2∆N^K5-rtTA mice and 3 littermate controls of genotypes tetO-GLI2∆N or K5-rtTA were used for RNAscope and IF; all mice were doxycycline-treated for 4 days. (N-S) GLI2∆N^K5-rtTA Rosa26^mTmG mice carrying inducible Cre alleles driven by Lrig1 (N, O), Lgr6 (P, Q) or Axin2 (R, S) promoters were tamoxifen induced at P42 to induce Cre activity and placed on oral doxycycline at P72 to induce GLI2∆N expression. mGFP expression (red signal) and GLI2 expression (green signal) were analyzed by IF at P74 (N, P, R) or P82 (O, Q, S). GLI2∆N-expressing cells in the acinar basal layer positive for Lrig1 (N), Lgr6 (P) or Axin2 (R) (yellow arrows) give rise to clones that contribute to MG overgrowth (O, Q, S, white arrows). N=3 samples were analyzed per line per time point. Scale bars: (D-M), 50 μm; (N-S), 25 μm.

Lrig1-, Lrg6- and Axin2-expressing MG stem cells contribute to MG overgrowth driven by GLI2∆N

To test whether Lrig1-, Lgr6- and Axin2-expressing MG stem cells contribute to basal cell expansion caused by forced GLI2∆N expression, we performed lineage tracing in Lrig1-Cre^ERT2 Rosa26^mTmG, Lgr6-Cre^ERT2 Rosa26^mTmG, and Axin2-Cre^ERT2 Rosa26^mTmG mice carrying Gli2∆N^K5rtTA. Adult mice were treated with tamoxifen to induce Cre activity; after 30 days, mice were placed on doxycycline water to induce GLI2∆N expression in MG stem cells. Two days after initiating doxycycline treatment, GFP+/GLI2∆N+ cells were present in the acinar basal layer in each of these lines (Figure 3N, P, R, yellow arrows). After 10 days of doxycycline treatment, we observed expansion of GFP+/GLI2∆N+ clones in all three lines (Figure 3O, Q, S, white arrows). These observations indicate that GLI2∆N-expressing MG stem cells contribute to MG basal expansion.

Human Meibomian gland carcinoma samples express GLI2 and exhibit expansion of LRIG1 and LGR6-expressing undifferentiated cells

Abnormal stem cell activity is frequently observed in cancers^37,38, but the role of stem cells in human Meibomian gland carcinoma (MGC) is unknown. To determine whether expanded GLI2 and stem cell marker expression are features of human MGC, we analyzed 7 control MG samples and 10 MGC samples. We found that in all control MGs examined, GLI2 was predominantly expressed in acinar basal cells (Figure 4A, yellow arrows) and differentiating meibocytes (Figure 4A, white arrows) but was absent in fully differentiated meibocytes (Figure 4A, light blue arrows). By contrast, GLI2 was broadly expressed in all 10 MGC samples (Figure 4B). GLI1 was weakly expressed in acinar basal cells of control MGs (Figure 4C, yellow arrows), but was readily detected in 7 of 10 MGC samples (Figure 4D). LRIG1 was predominantly expressed in the acinar basal layer of control MGs (Figure 4E, yellow arrows), but was broadly and strongly expressed in all 10 MGC samples (Figure 4F). Similarly, LGR6 expression was detected in acinar basal cells of control MGs (Figure 4G, yellow arrow), but was broadly expressed in MGC samples (9/10) (Figure 4H). In control MGs, proliferation was restricted to the basal layer of the acinus (Figure 4I, white arrows), which lacked expression of differentiation marker PLIN2 (Figure 4K, white arrows). However, in MGC samples, proliferative cells were widely present (Figure 4J, yellow arrows) and PLIN2 expression was undetectable in most tumor cells (Figure 4L, yellow arrows). These data indicate that human MGC is characterized by expansion of poorly differentiated and proliferative tumor cells that express MG stem cell markers.

Figure 4. GLI2-, LRIG1- and LGR6-expressing undifferentiated cells are expanded in human MGC.

(A, B) In normal human MG GLI2 protein localizes to acinar basal cells (A, yellow arrows) and differentiating meibocytes (A, white arrows) but is absent from fully differentiated meibocytes (A, light blue arrows); in MGC samples GLI2 is broadly expressed (B). (C, D) In normal human MG, GLI1 mRNA is weakly expressed in some acinar basal cells (C, yellow arrows); GLI1+ cells are widely present in MGCs (D). (E-H) LRIG1+ and LGR6+ cells are primarily present in the acinar basal layer of normal human MGs (E, G, yellow arrows); MGC displays expansion of LRIG1+ and LGR6+ cells (F, H). (I, J) Ki-67+ cells localize to the normal human acinar basal layer (I, white arrows) and are expanded in MGC (J, yellow arrows). (K, L) PLIN2-/KRT14+ cells are restricted to the acinar basal layer in normal human MG (K, white arrows) but are present throughout MGC tissue (L, yellow arrows). 7 normal human MG and 10 human MGC samples were analyzed; representative data are shown. Scale bars: 50 μm.

Aged MG display fewer cells expressing Hh pathway genes

The mechanisms underlying MG dropout in aging remain poorly understood. As previously reported^9,39, we found that in aged (21-month) versus young (8-week) MGs, intracellular distribution of PPARγ was altered, being relatively reduced in the cytoplasm and enriched in the nucleus in aged samples (Figure S6A, B), while the percentage of KRT5+ acinar basal cells that were positive for the proliferation marker Ki-67 was significantly decreased in aged compared with young MGs (Figure S6C, D, E).

As our data revealed a key role for Hh signaling in acinar basal cell proliferation, we asked whether expression of Hh pathway genes was altered in aging. Analysis of snRNA-seq data from young and aged MGs using CellChat, which assesses intercellular communication based on patterns of ligand and cognate receptor expression⁴⁰, predicted that in young mice, IHH ligands secreted from MG ductal cells and differentiating Meibocytes are received by ductular cells, acinar basal cells, and dermal cells (Figure S6F, G). However, in aged mice, the levels of Hh pathway gene expression were decreased to the point that significant Hh-mediated interactions could not be identified (Figure S6H, I). Violin plots derived from snRNA-seq data showed that the percentages of ductular cells expressing Gli2 and acinar basal cells expressing Ptch1 were reduced in aged versus young MGs (Figure 5A). Validation by RNAscope revealed that there were fewer cells expressing Ptch1 in the acinar basal layer, differentiating meibocyte population, and stroma in aged MG (Figure 5C) compared with MG at 8 weeks (Figure 5B). Similarly, there were fewer cells expressing Gli2 in aged compared with young MG, especially in the ductules (Figure 5D, E, white arrows).

Figure 5. Aged MGs exhibit fewer cells with Hh activity and elevated association of GLI2 with acetyl-lysine.

(A) Violin plots derived from snRNA-seq data show relatively reduced percentages of cells expressing Gli2 and Ptch1 in the ductule and acinar basal layer, respectively, in aged compared to young MGs. (B, C) Ptch1 is expressed in acinar basal cells (B, white arrow), some meibocytes (B, yellow arrow), and surrounding stromal cells (B, blue arrow) at 8 weeks; fewer Ptch1-expressing cells are present in aged MG (C). (D, E) Fewer Gli2-expressing cells are present in aged MG particularly in the ductule (D, E, white arrows). (F, G) RNAscope using a pan-Hh probe reveals similar expression of mRNA for Hh ligands in the acini of young (F) and aged (G) MGs. (H, I) PLA for GLI2 and acetyl-lysine shows that close association (<40nm) of GLI2 with acetyl-lysine is elevated in subsets of cells within aged (I) compared with young (H) MG acini (I, white arrows) and stroma (I, blue arrows). (J, K) Association of GLI2 and acetyl-lysine is elevated in the acinar basal layer (K, white arrows) and meibocytes (K, yellow arrows) but not stromal cells (J, K, blue arrows) in K5-rtTA tetO-Cre Hdac1^fl/fl Hdac2^fl/fl (Hdac1/2^DcKO) mice doxycycline induced from P48 and analyzed at P58 compared to control mice of genotype Hdac1^fl/fl Hdac2^fl/fl and lacking K5-rtTA or tetO-Cre. (J). (L, M) Ptch1 expression in control acinar basal cells (L, white arrows) and differentiating acinar cells (L, yellow arrow) is reduced in HDAC1/2-deficient MG acini (M); stromal Ptch1 expression is similar in mutants and controls (L, M, blue arrows). (N) Violin plots of snRNA-seq data indicate that there is a relatively lower percentage of Hdac1-expressing acinar basal cells (cluster #16) but a higher percentage of Hdac2-expressing acinar basal cells and ductular cells (cluster #17) in aged compared to young MG. (O, P) IF for HDAC1/2 shows similar expression in young (O) and aged (P) MGs. White dashed lines in (B-M; O, P) outline MG acini and/or ducts. 3 mice of each age and genotype were used for RNAscope and PLA. Scale bars: 25 μm. See also Figure S6.

Association of GLI2 with acetylated lysine is increased in aged MGs

Despite reduced Hh pathway activity, expression of Hh ligands was maintained in aged MG (Figure 5F, G), suggesting that mechanisms other than decreased ligand expression downregulate Hh signaling in aged MGs. Proximity ligation assays (PLA) revealed that subsets of cells in aged MG acini (Figure 5H, I, white arrows) and surrounding stroma (Figure 5H, I, blue arrows) displayed elevated association of GLI2 protein with acetylated lysine. Acetylation mitigates the transcriptional activity of GLI2⁴¹; thus, this mechanism could potentially account for the observed decrease in the percentages of cells expressing Hh target genes in MGs and their stromal environment.

MGs with epithelial deletion of the histone deacetylases HDAC1 and HDAC2 exhibit reduced proliferation of acinar basal cells⁴². As GLI2 can be deacetylated by HDAC1⁴², we hypothesized that association of GLI2 with acetylated lysine would be increased in the absence of HDAC1/2. In line with this, Hdac1/2-deleted acinar basal and differentiating cells in MGs from 8-week-old K5-rtTA tetO-Cre Hdac1^fl/fl Hdac2^fl/fl mice that had been doxycycline-treated for 10 days, showed increased PLA signals for GLI2 and acetylated lysine (Figure 5J, K, white and yellow arrows). PLA signals in the surrounding stromal cells, where Hdac1/2 were not deleted, were similar in mutant and control mice, providing an internal control for the PLA assay (Figure 5J, K, blue arrows). Furthermore, Ptch1 expression was attenuated within HDAC1/2-deficient acini (Figure 5L, M, white and yellow arrows) but not in the surrounding stroma (Figure 5L, M, blue arrows). Violin plots derived from snRNA-seq data revealed similar numbers of ductular cells expressing Hdac1 and increased numbers expressing Hdac2 in aged versus young MGs; and decreased numbers of acinar basal cells expressing Hdac1 and increased numbers expressing Hdac2 in aged versus young MGs (Figure 5N). IF assays failed to detect a discernable difference in the expression of HDAC1/2 proteins between young and aged MG acini (Figure 5O, P). Thus, mechanisms that affect HDAC1/2 activity, rather than altered expression levels of these proteins, may account for increased association of GLI2 with acetylated lysine in aged MGs.

Taken together these observations suggest that the decreased percentages of cells exhibiting Hh activity in aged MG and the surrounding stroma may result in part from increased levels of GLI2 acetylation that mitigate its transcriptional activity in subsets of acinar and stromal cells.

Decreased HBEGF signaling is associated with MG aging

Abrogation of epithelial Hh signaling in adult life did not result in complete loss of MGs, suggesting important contributions of additional signaling pathways to MG maintenance and the loss of these functions in aging. CellChat analysis of snRNA-seq data from young and aged MGs predicted that the EGF pathway contributes to intercellular communication among MG subpopulations and between MG epithelial cells and the surrounding dermal niche and is attenuated in aged mice (Figure 6A, B). CellChat analysis specifically predicted that HBEGF-EGFR-mediated crosstalk between dermal cells and ductular and acinar basal cells is abolished in aged MG (Figure 6C, green pathways). Analysis of snRNA-seq data showed that the relative percentages of cells expressing Hbegf were reduced in aged versus young differentiating meibocyte, acinar basal, and dermal fibroblast populations (Figure 6D). These data were validated via RNAscope analysis of independent samples which revealed fewer Hbegf-expressing cells and decreased Hbegf expression levels per cell in the aged versus young acinar basal layer (Figure 6E, yellow arrows) and stroma (Figure 6E, white arrows). In line with this, phosphorylation of ERK1/2 (p-ERK1/2), which can be stimulated by EGFR activation⁴³, was decreased in aged MG (Figure 6F). Taken together with known pro-proliferative functions for EGF signaling in cultured human or rabbit MG epithelial cells in vitro^44,45, and in mouse eyelid and MG morphogenesis in vivo⁴⁶, these results suggest that the HBEGF-EGFR axis is important for MG homeostasis and that its decreased activity contributes to reduced proliferation of acinar basal cells in aged MG. Other pathways predicted to be dysregulated in aging included FGF signaling (Figure 6A, B), consistent with the role of FGF signaling in sebaceous gland regeneration¹⁶, and results from bulk RNA-seq of young and old tarsal plates⁴⁷.

Figure 6. HBEGF signaling is disrupted in aged MGs.

(A, B) CellChat analysis predicts decreased EGF signaling in aged MGs. (C) CellChat analysis predicts HBEGF signaling between dermal cells and MG ductular cells (green) at 8-weeks and its absence at 21-months. (D) Violin plots of snRNA-seq data show decreased percentages of Hbegf-expressing cells in the indicated cell populations in aged MG. (E) RNAscope shows reduced Hbegf expression levels per cell and fewer acinar basal cells (yellow arrows) and surrounding dermal cells (white arrows) expressing Hbegf in aged MG. (F) IHC shows that p-ERK1/2 levels are decreased but total ERK1/2 levels are similar in the acini of aged compared with young MG. White dashed lines in (E) and black dashed lines in (F) outline MG acini. Three pairs of 8-week and 21-month samples were used for RNAscope and IHC. Scale bars represent 25 μm.

Aging disrupts the MG microenvironment

To delineate additional mechanisms that may contribute to the aging of MG, we performed GO analysis of differentially expressed genes in snRNA-seq data from young and aged MGs. Unexpectedly, the results showed statistically significant aging-related decreases in the expression levels of neural guidance genes such as Slit3, Robo2, Sema3a and Epha7 in the MG acinar basal layer and surrounding dermal cells (Figure 7A, B, E, F). These data were validated by RNAscope (Figure 7C, D). These findings, together with the known impact of sensory nerves on MG epithelial cells⁴⁸, and decreased peripheral innervation of the lacrimal gland in aged mice^49,50, suggested that altered peripheral nerve function could contribute to MG defects in aging. Consistent with this, IF for the pan-neural marker PGP9.5²⁵ revealed that the density of nerves near MG acinar basal cells was compromised in aged compared to young mice (Figure 7G-I).

Figure 7. Aged MGs exhibit reduced peripheral innervation and collagen I expression.

(A) GO analysis of genes with statistically significantly reduced expression in aged compared with young acinar basal cells (cluster #16). (B) Violin plots for neuronal guidance genes in acinar basal cells. (C, D) Slit3 mRNA expression (yellow arrows) is reduced in aged MG acinar basal cells. (E, F) GO analysis of genes with reduced expression in aged cluster #23 dermal cells (E) and violin plots for neuronal guidance genes (F). (G, H) Reduced numbers of PGP9.5 nerve fibers (red) adjacent to the KRT5+ acinar basal layer (green) in aged mice. (I) Quantification of PGP9.5+ nerve fibers within 5 μM of each KRT5+ acinar basal cell in young and aged MGs. Four pairs of 8-week-old and 21-month-old male mice were used for quantification. At least 185 acinar basal cells were analyzed from each mouse. Significance in (I) was calculated by an unpaired two-tailed Student’s t-test. Data are represented as mean +/− SEM. (J, K) GO analysis of genes with reduced expression in aged cluster #15 dermal fibroblasts (J) and violin plots for type I collagen genes in dermal fibroblasts (K). (L, M) Reduced expression of Col1a1 (red) in aged (M) versus young (L) dermis surrounding MG acini. White dashed lines in (C, D, L, M) outline MG acini and/or ducts. 3 pairs of 8-week and 21-month samples were used for RNAscope, IF and IHC. Scale bars represent 25 μm. See also Figure S7.

In addition to the neuron-regulatory genes, downregulated DEGs in aged eyelid dermal fibroblasts included those responsible for collagen synthesis and extracellular matrix assembly, such as Col1a1, Col1a2 and Col3a1 (Figure 7J-M), in line with previous data indicating disrupted collagen synthesis in aged dermal fibroblasts⁵¹.

By contrast, upregulated DEGs in aged MG acinar basal cells and ductular cells were associated with oxidative and other forms of stress (Figure S7A-D), which contribute to aging in other contexts^52,53, while upregulated DEGs in aged MG dermal cells were associated with inflammation, consistent with immune cell infiltration in aged MGs⁶ (Figure S7E-H).

Taken together, our analyses of aged versus young MGs revealed that MG aging is associated with multiple potential mechanisms, including decreased Hh and EGF signaling, altered function of peripheral neurons, a disrupted dermal microenvironment, and inflammation.

Discussion

Little is known about the molecular identities, locations, and behavior of MG stem cells, the signaling pathways that control their homeostasis, and the nature of their tissue niche. Improved understanding of these properties will be important in identifying effective treatments for evaporative dry eye disease, which are currently lacking. To begin to address these questions we used snRNA-seq and lineage tracing approaches to identify MG stem cells, showed that Hh signaling plays a key role in MG stem cell proliferation, and uncovered decreased Hh and EGF signaling in aged compared with young MGs.

Prior data showed that distinct basal cell populations replenish the central MG ducts and the acini, but specific markers for these stem cell sub-populations had not been identified. We found that basal stem cells marked by Slc1a3 exclusively contribute to the acinus; by contrast, lineage tracing analyses revealed that both duct and acinus were replenished by stem cells that expressed Lrig1, Lgr6, Axin2 or Gli2.

MG ductules, which connect the central duct and the acini, contain label-retaining cells and have been suggested to contribute to MG homeostasis¹². However, experimental evidence for this concept has been lacking. Our snRNA-seq analysis revealed that ductules express genes that are characteristic of both the MG duct (Krt17 and Egr2) and the acinus (Pparg and Slc1a3). Velocity analysis of the snRNA-seq data predicted that ductular cells contribute to both the central duct and the acinus. In line with this, we found that ductular cells express Slc1a3 as well as Lrig1, Lgr6, Axin2, and Gli2. Lineage tracing combined with live imaging of MG explants showed that LRIG1+ cells in the ductule can migrate towards the acinus. Furthermore, Gli2 expression is absent in the central duct, but ductular cells lineage-traced for Gli2 populate the central duct. Taken together, these results indicate that the ductule harbors stem cells for both duct and acinus.

Our snRNA-seq data and analyses of mice lacking the Hh receptor Smo revealed that Hh signaling plays a key role in promoting proliferation of MG stem cells. Consistent with this, forced expression of activated GLI2 in MG epithelial cells resulted in expansion of stem cell populations. We found that, like in mouse GLI2∆N-expressing MGs, human GLI2+ MGC cells are proliferative and poorly differentiated, exhibiting broad and increased expression of the stem cell markers LRIG1 and LGR6, suggesting that they arise from uncontrolled stem cell expansion. To date, genetic mutations of Hh components have not been identified in MGC. This may partially be attributed to the low occurrence of this disease. Alternatively, genes controlling posttranslational modification of GLI2, such as those regulating its acetylation^41,54 and ubiquitination⁵⁵, may be mutated or disrupted in MGC. Further investigation is required to test these potential mechanisms.

The observed decline in Hh pathway activity in aged MG could contribute to aging-associated reduced proliferation and MG dropout. Expression levels of Hh ligands were similar in young and aged MGs; however, association of GLI2 with acetylated lysine, which dampens its transcriptional activity, was increased in aged compared with young MG epithelial and surrounding stromal cells. GLI2 is deacetylated by the histone deacetylases HDAC1/2, and we found that association of GLI2 with acetylated lysine was increased in HDAC1/2-deficient MGs. Consistent with this, we previously showed that depletion of epithelial HDAC1/2 causes decreased MG basal cell proliferation⁴². Taken together, these data support a model in which decreased Hh signaling in aging results in part from altered HDAC1/2 activity that impacts GLI2.

Deletion of epithelial Smo did not result in complete loss of MGs, indicating the existence of additional critical pathways. Analysis of snRNA-seq data using CellChat predicted the existence of HBEGF signaling between dermal fibroblasts and both acinar and ductular basal cells that was strongly decreased in aging. This prediction was validated by analysis of Hbegf and p-ERK1/2 expression in independent samples of young and aged MGs and is consistent with in vitro experiments showing that EGFR activation is sufficient to stimulate the proliferation of MG epithelial cells^44,45. Further investigations will be needed to determine the precise role of HBEGF-EGFR signaling in MG homeostasis and the mechanisms controlling its activity during aging.

The MG is highly innervated^25–27, neurotransmitters can stimulate proliferation of MG epithelial cells in vitro⁴⁸, and impaired innervation and/or neurotransmitter secretion have been observed during aging in other glands such as the exorbital lacrimal gland⁵⁶ and sweat gland⁵⁷. However, the in vivo functions of MG innervation and how these might be altered in aging have not been examined. Our studies revealed that impaired innervation is a previously uncharacterized feature of aged MG and may result from decreased expression of axon guidance factors in aged MG epithelial cells and surrounding dermis^58,59.

In summary, our data identify markers for the distinct stem cell populations that maintain MG ducts and acini and reveal that MG ductules harbor stem cells for both the duct and the acinus. We uncovered Hh signaling as a key pathway regulating MG stem cell proliferation and showed that its activity is enhanced in human MGC and decreased in aging. We further found that decreased HBEGF signaling, reduced peripheral innervation, and an altered dermal microenvironment accompany MG aging and may contribute to reduced epithelial proliferation and MG dropout. These observations provide a new understanding of the mechanisms of MG homeostasis, aging, and tumorigenesis, and suggest Hh signaling as a therapeutic target in MGC and in MG aging. Transient, localized and tightly controlled Hh and EGFR pathway activation could provide therapeutic benefit for patients with evaporative dry eye disease by stimulating stem cell activity, whereas inhibition of the Hh pathway, for instance with SMO inhibitors⁶⁰, may be useful in treatment of MGC.

Limitations of Study

snRNA-seq, spatial transcriptomics, and aging-related experiments were carried out in male mice. While most validation experiments were performed on both male and female samples, lack of inclusion of female samples in the snRNA-seq dataset and in aging-related experiments could mean that sex-specific differences in MG biology and aging were missed in the current study.

STAR Methods

Mice

C57BL/6J mice (Jackson Laboratory #000664) were used for snRNA-seq, spatial transcriptomics, and aging-related experiments. All other mice were maintained on a mixed-strain background. Up to five mice were maintained per cage in a specific pathogen-free barrier facility using standard rodent laboratory chow. Mice of both sexes were analyzed in all experiments except for snRNA-seq, spatial transcriptomics, and aging-related experiments, and no significant sex-based differences were noted. The following mouse lines were utilized: Lrig1-Cre^ERT2 (Jackson Laboratory, #018418), Lgr6-Cre^ERT2 (Jackson Laboratory, #016934), Axin2-Cre^{ERT2 30}, Slc1a3-Cre^ERT2 (European Mouse Mutant Archive, EM:12216), Gli2-Cre^{ERT2 61}, K5-rtTA (Jackson Laboratory, #017519), tetO-GLI2∆N³⁵, Rosa26^mTmG (Jackson Laboratory, #007676), Rosa26^nTnG (Jackson Laboratory, #023537), KRT14:H2BGFP⁶², KRT14-Cre^ERT2 (Jackson Laboratory, #005107), Smo^fl/fl (Jackson Laboratory, #004526), Hdac1^fl/fl Hdac2^{fl/fl 63}, tetO-Cre (Jackson Laboratory #006234). All animal experiments were carried out under approved animal protocols according to institutional guidelines established by the Icahn School of Medicine at Mount Sinai IACUC or the Johns Hopkins University IACUC.

Human MG samples

De-identified human eyelid tissues discarded from unrelated surgical procedures were paraffin-embedded and sectioned for histological analysis, RNAscope and IF.

snRNA-seq

Eight 8-week-old and eight 21-month-old male mice were euthanized, and their eyelids were cut with scissors. The dorsal portion of the eyelids and the surrounding tissues near the tarsal plates were removed using a scalpel. The remaining tarsal plates containing MGs were digested with 0.2% collagenase A in PBS for 1 hour at 37 °C. The tarsal plates were then rinsed twice with PBS and tissues surrounding the MGs were removed with fine-tip forceps. Tissues from four littermate mice were pooled together in PBS for each replicate. The MG tissue-containing solutions were then transferred into 1.5mL centrifuge tubes using Pasteur pipettes. The tissue solution was centrifuged at 300g/min for 5 minutes, the supernatant was carefully removed, and the tissue pellet was rinsed once with PBS. Isolation of nuclei was performed using the protocol (CG000375 Rev B) from 10X Genomics. RNA-seq libraries were prepared based on the 10X Genomics protocol (CG000315 Rev E) using 10000 nuclei per replicate. Single-cell separation was performed using the Chromium Next GEM Chip G Single Cell Kit. The RNA-seq libraries were sequenced in paired-end mode on an Illumina Novaseq platform with the Novaseq S4 flowcell.

Alignment of 10X Genomics data

Cell Ranger Single-Cell Software Suite (v7.0.1, 10x Genomics) was used to align and quantify sequencing data against the mm10 mouse reference genome (refdata-gex-mm10–2020-A).

Analysis of snRNA-seq data

Cell clustering and differential gene expression were analyzed using Seurat⁶⁴ (v4.3.0). For each replicate, high-quality nuclei (defined as nFeature_RNA >200 & nFeature_RNA <4500 & nCount_RNA >500 & < nCount_RNA <25000 & precent.mt < 5) were analyzed. 6676 nuclei were analyzed from 8-Week replicate 1, 10734 from 8-Week replicate 2, 8336 from 21-Month replicate 1, and 7305 from 21-Month replicate 2. The CellCycleScoring function of Seurat was applied to mitigate the effects of cell cycle heterogeneity in the datasets. The SCTransform function of Seurat was used to integrate the four datasets. In total, 34 clusters were identified with a resolution of 1. For Violin plots, the y-axis was plotted against the log normalized data from the data slot of the Seurat object. Seurat employs a global-scaling normalization method “LogNormalize” that normalizes the feature expression (each gene) measurements for each cell by the total expression, multiplies this by a scale factor (10,000 by default), and log-transforms the result.

For RNA velocity and pseudotime analyses, bam files generated from Cell Ranger were processed using Velocyto (v0.6)’s run10x function with default parameters, resulting in loom files that contained counts of unspliced and spliced mRNAs. The loom files were merged with the h5ad files, which covered the meibomian gland clusters. Subsequently, the merged files were preprocessed with scVelo (v0.3)⁶⁵, and RNA velocity and pseudotime analyses were performed using cellDancer (v1.1)⁶⁶.

To identify DEGs, genes encoding ribosomal proteins were initially defined and removed by grepping for “^Rp(sl)” from the datasets before pseudobulk analysis using EdgeR (v3.38.4)⁶⁷. Genes with an FDR (padj)<0.05 and |Log₂FC|>0.5 were considered DEGs and were subjected to gene ontology analysis with clusterProfiler (v4.4.4)⁶⁸.

The interactions among distinct MG cell types (MG ductal basal cells, ductal suprabasal cells, ductular cells, acinar basal cells, differentiating meibocytes, and differentiated meibocytes), along with surrounding dermal cells (two populations were combined in this context), were analyzed using CellChat (v1.6.1)⁴⁰. Briefly, the normalized counts were loaded into CellChat and preprocessed with default parameters. To establish the ligand-receptor interaction database, we selected secreted signaling pathways and utilized the “projectData” function to project gene expression data onto the mouse protein-protein interaction (PPI). To visualize specific signaling pathways, we employed the setting “type=truncatedMean, trim = 0.1” to compute the communication probability (computeCommunProb) except for HH signaling (trim =0.001 was used). Core CellChat functions such as “computeCommunProbPathway”, “aggregateNet”, “netAnalysis_signalingRole”, and “identifyCommunicationPatterns” were applied with standard parameters for downstream analysis.

Spatial transcriptomics

The 10x Visium Spatial Gene Expression kit was employed to construct the spatial transcriptomic library, according to the manufacturer’s instructions. A pair of upper and lower eyelids from an 8-week-old male mouse was dissected, sectioned at 10 μM, and mounted onto the assay slide containing four capture areas (6.5mm x 6.5mm) with spatially barcoded poly T capture probes. After permeabilization, the captured mRNAs underwent reverse transcription, followed by cleaving from the slide. cDNA amplification, fragmentation, end-repair, poly A-tailing, adapter ligation, and sample indexing were performed in accordance with the manufacturer’s protocol. The resulting cDNA libraries were quantified using TapeStation (Agilent) and Qubit (Invitrogen) and then subjected to sequencing in paired-end mode on a NovaSeq instrument (Illumina) at a depth of 50,000–100,000 reads per capture spot. The sequencing data were aligned and quantified using Space Ranger Software Suite (v2.1.0, 10x Genomics) against the reference genome.

Analysis of spatial transcriptomic data

Integration of snRNA-seq data and spatial transcriptomic data was conducted through CytoSPACE (v1.0.5)⁶⁹. Briefly, a non-normalized count-based gene expression file and a cell-type label file were generated from the snRNA-seq datasets. Additionally, a compressed spatial transcriptomes input file for CytoSPACE was created using files from Space Ranger outputs. Subsequently, all the input files were loaded into CytoSPACE with the recommended parameter settings, as detailed on the project’s GitHub repository (https://github.com/digitalcytometry/cytospace).

Immunofluorescence (IF) and immunohistochemistry (IHC)

Paraffin sections were deparaffinized and rehydrated to PBS. Antigen retrieval was performed with antigen unmasking solution. Sections were washed with PBST (PBS+0.1% Tween 20) and blocked with 1% BSA for 30 minutes. Primary antibodies were applied, and the sections were incubated overnight at 4°C. For IF, sections were subjected to three washes of PBST, followed by 1-hour incubation with secondary antibodies. The sections were washed with PBST and mounted with DAPI-containing medium. For IHC, sections were washed with PBST and then incubated with 3% H₂O₂/PBST for 15 minutes, followed by two washes of PBST and 30-minute incubation with biotinylated secondary antibodies. Subsequent steps and DAB staining followed the manufacturer’s instructions. The sections were dehydrated, cleared, and mounted. IF and IHC data were documented using a Leica Microsystems DM5500B microscope equipped with two cameras for fluorescent and bright field images (Leica Microsystems).

RNAscope in situ hybridization

RNAscope was performed on paraffin sections following the user’s guide provided by Advanced Cell Diagnostic (ACD) using probes for Slc1a3, Scd4, Slit3, Hbegf, Col1a1, pan-Hh, Lrig1, Lgr6, Axin2, Ccnd1, Gli1, Gli2, Ptch1, Smo, Shh, LRIG1, LGR6, and GLI1.

Tamoxifen induction

200μl Tamoxifen dissolved in corn oil (10mg/ml) was injected intraperitoneally once for lineage tracing. For gene deletion, daily intraperitoneal injections were performed on five consecutive days.

Ex vivo live imaging of tarsal plate explants

Mice were euthanized, and the eyelids were dissected with scissors. Tissues surrounding the tarsal plates were removed using a surgical scalpel. The tarsal plates were placed in a 35-mm Lumox-bottom dish, immobilized with a custom-built holder, and incubated in DMEM/F12(1:1) medium supplemented with 5% FBS for 2 hours in an environmental chamber supplied with 5% CO₂ at 37°C before imaging. A Leica SP8 laser scanning confocal microscope equipped with Power HyD detectors and an environmental chamber was used to image the cultured explants. Images were captured as z stacks at 3 µm intervals and tissues were imaged every 15 min with low laser power (<3%) with a HC PL APO CS2 10x/0.40 DRY objective lens. Stacked images were maximum projected with Leica LAS X software to generate the time-lapse video and video annotations were performed with Fiji software.

Doxycycline administration

200 µg/ml doxycycline was dissolved in 5% sucrose water and was fed to the mice and replaced every 5 days.

Bulk RNA-seq and data analysis

3 pairs of 6-week-old Gli2∆N^K5rtTA mice and littermate controls were fed with 200µg/ml doxycycline water for 4 days. Mice were euthanized and the upper and lower eyelids were dissected, frozen sectioned at 12µm, and mounted onto polyethylene naphthalate-membrane slides (Leica Microsystems). The MGs from 25 sections from each sample were laser dissected using the gravity-assisted LMD 7000 system (Leica Microsystems) and collected by gravity assistance onto AdhensiveCap. RNA was extracted using the Qiagen All Prep DNA/RNA FFPE Kit according to the manufacturer’s instructions. 50ng RNA from each laser-captured sample was used for RNA-seq library construction using the Trio RNA-Seq^™ library preparation kit from Nugene. The libraries were sequenced on an Illumina HiSeq 4000 platform with the 100bp single read (SR100) method. For data analysis, Salmon (v0.9.1)⁷⁰ was used to count data against the transcriptome defined in Gencode vM18. We then annotated and summarized the transcriptome count data to the gene level with tximeta⁷¹ and performed further annotation with biomaRt (v0.7.0)⁷². Normalizations and statistical analyses were carried out with DESeq2 (v1.20.0)⁷³. Genes with padj<0.001 and |Log₂FC|>0.5 were selected for GO enrichment analysis, which was carried out by clusterProfiler. The results were plotted by ggplot2 (v3.4.0) and volcano plots were generated using EnhancedVolcano (v1.14.0).

Proximity ligation assay (PLA)

PLA was carried out as previously described³⁰ using a DuoLink In Situ PLA kit (Sigma Aldrich). Briefly, 5 μm paraffin sections were dewaxed and rehydrated, followed by blocking with 5% donkey serum/PBS/0.8% Triton X-100 for 30 minutes and incubation with solutions containing primary antibodies overnight at 4 °C. The sections were washed with PBST and incubated in PLA probe anti-rabbit plus or anti-goat plus, and PLA probe anti-rabbit minus for 2 hours. The subsequent ligation and amplification steps followed the manufacturer’s instructions. The sections were washed twice in washing buffer and once in PBS for 20 minutes each before mounting and imaging using a Leica SP8 laser scanning confocal microscope.

Statistical analyses

Statistical analysis and graphical representation were performed using Microsoft Excel 2023. Unpaired or paired two-tailed Student’s t-test was used to calculate statistical significance between two groups of data. P < 0.05 was considered significant. Data were represented as mean +/− SEM.

Supplementary Material

Supplement 8

Video S1. MGs in tarsal plate explants undergo proliferation and secrete lipid droplets during short-term culture. Related to Figure 1.

Supplement 9

Video S2. Lrig1+ MG ductule stem cell progeny can migrate toward acinus and locate in the acinus basal layer. Related to Figure 1.

Supplement 10

Document S1. Figures S1-S7 and legends for Videos S1 and S2.

Data availability statement

The authors declare that the main data supporting the findings of this study are available within the article and its Supplemental Information files. This study did not generate unique new reagents. snRNA-seq, spatial transcriptomics, and bulk RNA-seq datasets will be accessible through the GEO website on publication. Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Sarah E. Millar (sarah.millar@mssm.edu).