Digital Cell Atlas of Mouse Uterus: From Regenerative Phase to Maturational Phase

Background: Endometria undergo repeated repair and regeneration during the menstrual cycle. Using gene expression to dene the menstrual cycle has been seen in a couple of researches, which shares little consistency. The possible reason lies in the fact that the composition of each specimen is different. The objective of this study was to reconstruct an integrated cell atlas of mouse uterus from the regenerative to maturational endometrium at the single-cell level. Methods: We used published single-cell RNA sequencing data of mice uteri to build an integrated cell atlas, including epithelial cells, stromal cells, and immune cells. Principal components analysis, hierarchical clustering, and other data mining procedures were performed with R software. Functional enrichment analyses were performed to identify the potential biological roles of molecular changes in different mice estrus states. Results: Based on the expression level of proliferating cell nuclear antigen and gene ontology terms, we reconstructed the cell atlas and delineated in detail the transitions that happened during the estrus cycle. We also depicted the transition and transcription factors that shaped the differentiation of ESCs and the mononuclear phagocyte system. Besides, the amounts and functions of immune cells varied sharply in two periods. We also found clues about uterus tissue-resident macrophages and Pdgfrb + Aldh1a2 + Cd34 + endometrial mesenchymal stem cells in vivo. Conclusions: The cell atlas of mouse uterus presented here would improve our understanding of the functional changes that occurred in the endometrium and helped us identify abnormalities that have not been apparent histologically.

The cognition of the immune conditions in the menstrual cycle has been updated rapidly, yet the amounts or the functional traits of multiple immune cell types are still under debate. The human immune cells in decidua have been analyzed thoroughly [4]. In sharp contrast, the features of immune cells in the regenerative phase and the maturational phase have not gained enough attention. Up to today, the complicated immune regulation under the hormone-induced menstrual cycle has left a huge riddle for us to solve.
The estrous cycle in mice averages 4-5 days and is a repetitive but dynamic process, re ecting changes in the levels of estradiol and progesterone secreted by the ovarian follicles [5]. There are different criteria for de ning the estrus cycle of mice [6], but no matter how the cycle is de ned, mice uteri undergo the same hormone change patterns and the recurrences of regeneration and maturation like humans. Therefore, we decided to seek insights and inspirations from mice single-cell RNA sequencing (scRNAseq) data to shed light on human endometrium research.
We used published scRNA-seq data of mice uteri to build an integrated cell atlas of different estrus states, including epithelial cells, stromal cells, and immune cells. We gured out in detail the changes that happened in the transition from the regenerative endometrium to the maturational endometrium. In addition, we reconstructed the pseudotime trajectories of endometrium stromal cells and the mononuclear phagocyte system and found the transcriptional factors shaped cell decision. We also found clues about Pdgfrb + Aldh1a2 + Cd34 + endometrial mesenchymal stem cells in vivo. An in-depth understanding of the functional changes that occurred in the endometrium during different menstrual states will help us tackle endometrial-related abnormalities.
The lter criteria of cells were determined as default. It is not necessary to obtain ethical approval because we used published online datasets.
Results: single-cell RNA sequencing analysis of two mice uteri In order to nd more clues about cells in mice uteri, we analyzed the scRNA-seq data of two 6-to-10-weekold female mice from the single-cell mouse cell atlas (scMCA) [7]. Using the 15 most signi cant principal components of the PCA, we divided the cells into 16 clusters ( Figure 1A-B, Table S1). Then we clari ed the identities of each cluster ( Figure 1C, Table S2). Cluster 0, cluster 1, cluster 2, cluster 3, cluster 10 all highly expressed Col3a1 and Fn1, so we clari ed them into stromal cells. Cluster 4, in the meantime, speci cally expressed Acta2, which made us clarify them into myo broblasts. Cluster 11, however, expressed Acta2 and Myh11 without Col3a1, so cells in cluster 11 belonged to muscle cells. Cluster 5, cluster 9, and cluster 12 all highly expressed Cd68 and Adgre1, so they were macrophages/monocytes. Cd7 and Nkg7 could be found highly expressed in cluster 8, which made cluster 8 natural killer (NK) cells. Ly6d and Cd79a were highly expressed in cluster 15, so we identi ed cluster 15 as B cells. Cluster 13 and cluster 7 both had high expression levels of Krt8, Krt18, and Epcam, making them epithelial cells. Cluster 6 and cluster 14 had high expression levels of Cd34 and Pecam, making them endothelial cells. In this way, we clari ed the identities of 16 cell clusters, building a solid foundation for further analysis.
We identi ed 5 sub-clusters of stromal cells (cluster 0, cluster 1, cluster 2, cluster 3, cluster 10) and 2 subclusters of epithelial cells (cluster 13 and cluster 7). According to their distances in the t-SNE, we divided the 5 sub-clusters of stromal cells into 3 groups: stromal cells 1, stromal cells 2, and stromal cells 3. We compared the proportions of each group in two mice ( Figure 1D). Surprisingly, we found that stromal cells 1 were exclusively in mouse 2, and stromal cells 3 were exclusively in mouse1. Stromal cells 2 could be found both in mouse 1 and mouse 2. This phenomenon was also seen in epithelial cells. Epithelial cells 1 were exclusively in mouse 2, while epithelial cells 2 were exclusively in mouse 1.

Comparison of epithelial cells revealed two uteri of distinct estrus cycles
We believed the different cell contents of two mice had further biological signi cance. We found that proliferating cell nuclear antigen (Pcna) was highly expressed in epithelial cells 1 (mouse 2) (Figure 2A). The protein encoded by this gene is a cofactor of DNA polymerase delta, which helps increase the processivity of leading strand synthesis during DNA replication and is regarded as the marker of proliferation [18]. Therefore, we speculated that the uterus of mouse 2 was in the regenerative phase, and the uterus of mouse1 was in the maturational phase. To further investigate this speculation, we performed GO analysis of differentially expressed genes in each epithelial cell cluster ( Figure 2B-C, Table  S3). The GO terms of epithelial cells 2 (mouse 1) include "positive regulation of cell motility", "response to steroid hormone", "skin development", "cellular response to epidermal growth factor stimulus", "placenta development", and "response to wounding". These terms showed the active rebuilt of the uterus epithelium to prepare for subsequent fertilization in the maturational status. In order to fully show the differences between two epithelia, we extracted all the epithelial cells from two uteri and clustered the cells exclusively. We gained 3 clusters ( Figure 2D, Table S4). According to their Pcna expression levels, cluster 1 and cluster 2 showed proliferating characteristics ( Figure 2E). We concluded that mouse 2 was in the regenerative phase, for it was entirely composed of proliferating epithelial cells. As for mouse 1, only part of the epithelial cells showed proliferating characteristics ( Figure 2F).

The molecular trajectory of stromal cells in the estrus cycle
Endometrial stromal cells (ESC) perform a multitude of functions including hormonal regulation, decidualization, maternal-fetal communications, and embryo receptivity [19]. We further explored the characteristics of stromal cells in each estrus period given their signi cance in the uterus.
Because stromal cells 1 belonged to mouse 2, so it would be de ned as regenerative_stromal, and stromal cells 3 would be de ned as maturational_stromal likewise. Then we performed GO analysis of the differentially expressed genes of these two cell groups to see their functional traits.
The GO terms of regenerative_stromal included aorta morphogenesis ( Figure 3A, Table S3). As we know, the key feature of the regenerative phase is angiogenesis. Besides, protein maturation, collagen catabolic process, and retinoid metabolic process were actively involved in this period.
The maturational_stromal presented more dynamic cell communication ( Figure 3B, Table S3). In this period, the GO term "response to progesterone and other steroid hormones" further validated the estrus states of two mice. We also saw the term "response to transforming growth factor beta". Transforming growth factor-beta receptor 1 mediated signaling has been reported to be required for female reproductive tract integrity and function [20]. In the maturational phase, cell movement was accelerated, which was in accord with our results. GO analysis suggested "positive regulation of cell migration" in this period. In the meantime, the elevated levels of cell communication and material transport were found in maturational_stromal, which constituted a prosperous metabolism network in stromal cells.
As for the GO analysis of myo broblasts (Table S3), terms like "muscle structure development" and "regulation of smooth muscle cell proliferation" con rmed our previous clari cation of cell identity. These two stromal subsets in different physiological states had a huge difference in their functional traits, then we further explored how these two cell groups gained their unique traits by differentially gene expression. We performed pseudotime analysis to reconstruct the trajectory of stromal cells in two physiological states ( Figure 3C). This time we also included myo broblasts to construct a more comprehensive atlas of stromal cells in the uterus. The results showed that there were two differentiation branches: the regenerative stromal cells would transform into myo broblast or maturational stromal cells at the rst decisional point, and a small part of the maturational cells still had the ability to differentiate into myo broblasts at the second branch. In this way, an estrus cycle was completed. The heatmap visualized changes of all the signi cantly branch 1-dependent genes and clustered them into 3 categories by unsupervised clustering ( Figure 3D). Among these genes, the myo broblasts-branch had high expression levels of Acta2 and Adamts1 ( Figure 3E). In the meantime, the maturational_stromal-branch had high expression levels of Col6a4, Fbln2, Tcf4, and Wnt5a ( Figure 3F). Col6a2 and Fbln2 encode extracellular matrix protein, and Tcf4 and Wnt5a are genes involved in Wnt signaling. Wnt signaling is important in the maturational phase, for Wnt signaling is involved in the progesterone-induced regulation of uterine stromal cells [21]. These divergent expression patterns of representative genes further con rmed our previous classi cation. We listed other signi cant branch-dependent genes, which may also play key roles in the stromal cell differentiation and need further validation ( Table 1).
The immune landscape of uterus during the estrus cycle Embryo implantation and tumor progression are similar to some extent [22]. The maternal immune system needs to nd the balance and provide an appropriate environment for the fetus to grow. The current ndings of NK, monocytes, and dendritic cells (DCs) during the menstrual cycle are limited. For example, ndings concerning NK cell number and cytotoxic activity have been con icting [23]. In order to gain reliable results with regard to uterine immune cells at the single-cell level, we used the most 10 signi cant principal components to divide all immune cells into 4 clusters ( Figure 4A, Table S5). The heatmap of the top 50 markers for each cluster showed that we gained a convincing clustering result ( Figure 4B).
Using known markers, we clari ed the identities of each cluster ( Figure 4C). Cluster 0 expressed Adgre1 and Mrc1 so we clari ed them into macrophages. Cluster 1 expressed Nkg7 and Cd7, which made us clarify them into NK cells. Cluster 2 expressed Cd83 and Cd209a, so cells in cluster 2 belonged to DC cells. For their high expression of Ly6c2, we clari ed cluster 3 into monocytes.
Previous researchers have proved the plausibility of using expression levels of transcripts to predict the volume of cell populations in tissues. Levels of transcripts that encode immune-related factors changed during the time-course paralleled the changes observed in the volume fractions of the immune cells [24]. We used the expression levels of marker genes to predict the numbers of the immune cells, and we found the clear difference of immune cell composition in these two estrus states. The proportions of macrophages and NK cells in two states vary considerably ( Figure 4D). Macrophages dominated in the regenerative phase, while NK cells dominated in the maturational phase, which could be proved in previous studies. Macrophages in the regenerative phase have been reported to have a potential role in regeneration and proliferation of the functional layer of the endometrium [25]. The possible mechanism may lie in the fact that estrogen can recruit macrophages and neutrophils into the mouse uterus, while progesterone via its receptor antagonizes the pro-in ammatory activity of estrogen in the mouse uterus [26]. NK cells were reported to be dominant in the maturational phase [27], which played a pivotal role in the tissue homeostasis and endometrial vasculature remodeling that were necessary for embryo implantation and successful pregnancy [28].
How monocytes differentiate into DCs or macrophages are poorly understood. We performed pseudotime analysis of monocytes, macrophages, and DCs, trying to reconstruct the developmental trajectory of the mononuclear phagocyte system ( Figure 4E). The branch point led to two differentiation paths: macrophages or DCs. Heatmap clustered branch-dependent genes into three categories according to their expression patterns ( Figure 4F). The expression levels of marker genes of each branch over pseudotime veri ed the identities of two differentiation paths ( Figure 4G). We listed other signi cant branchdependent genes ( Table 2). Among them, we detected three transcription factors: Mafb, Irf7, and Nr4a1. Mafb was detected in the macrophage branch. Mafb was reported to be essential for monocytemacrophage differentiation in the previous study [29]. Irf7 expression was seen in the monocyte branch and Nr4a1 was expressed highly in the DC branch ( Figure 4G), yet the relations between these two transcription factors and their branches have not been reported yet.
In order to study the different functional traits of immune cells in two states, we drew scatter plots to nd the genes that were visual outliers of the average expression levels of two estrus states, which could highlight genes that exhibited dramatic responses to estrus phase change.
As for NK cells ( Figure 4H), NK cells in the regenerative phase showed intense in ammatory responses with high expression levels of Tnf, Il1b, and S100a8 [30]. It also expressed Cxcl2 and Ccrl2, which respectively promoted the recruitment of neutrophils and themselves [31]. In contrast, genes that were upregulated in the maturational state mostly were immune-suppressive genes like Serpinb9, Stat3, Cd96, and Cd55. NK cells also secreted substances that were advantageous for follow-up pregnancy like Ccl2.
The number of macrophages was higher in the regenerative phase. In this phase, macrophages were proin ammatory by secreting resistin, which is a systemic pro-in ammatory cytokine targeting both leukocytes and adipocytes ( Figure 4I) [32]. However, it also showed high expression levels of selenoprotein Msrb1 and Hes1 to inhibit in ammatory responses [33,34]. As for macrophages in the maturational phase, it mainly showed the M2 phenotype with high expression levels of Il10, Cd206, and Ccl7, which consisted of the characteristics of decidual macrophages [35].
In the regenerative phase, the functions of DC were mainly embodied in in uencing other cells. It secreted Ccl22 to recruit regulatory T (Treg) cells and Ccl5 to recruit CCR5-positive cytokine-induced killer cells [36,37]. Its pro-in ammatory role was also embodied in the high expression of Ccrl2, which played a signi cant role in in ammation [31]. DCs in the maturational phase were immune-suppressive ( Figure  4J). DCs expressed PD-L1, Jchain, and Anxa1, which were correlated with immune tolerance and refrained the secretions of pro-in ammatory cytokines [38,39].
Taken together, we found that the immune responses were suppressed in the maturational phase compared to the regenerative phase by reducing in ammatory cytokines secretion and facilitating differentiation into immune-suppressive cells.

Identify Pdgfrb + Aldh1a2 + Cd34 + endometrial mesenchymal stem cells in vivo
The mesenchymal-to-epithelial transition has been proposed as a possible reason to explain the periodically endometrial epithelial tissue regeneration and postpartum endometrial recovery [45]. A previous study has reported a group of cells expressed both the epithelial cell marker, pan-cytokeratin, and the stromal cell marker, vimentin as well [46]. We tried to identify multipotent ESCs in the scRNA-seq data, so we performed StemID2. We failed to nd the cell cluster that expressed all the proposed stem cell surface molecules like Cd73, Cd90, and Cd105 [47]. We hypothesized that expressions of Cd73, Cd90, and Cd105 may be induced during in-vitro culture. To verify this hypothesis, we used another scRNA-seq data of endometrial tissues [8]. The authors generated expression data of cultured ESCs and uncultured ones. The diversity between these two sets of samples was greater than we expected. The cells from two different culture environments clustered separately ( Figure 5A). The stem cell surface markers mentioned above were expressed differently in the two cell groups. We found that stem cell makers like Cd90 (Thy1), Cd44 had signi cant higher expressions in the cultured group ( Figure 5B), so we decided to use Pdgfrb as the stem cell marker, which was known to be enriched in endometrial mesenchymal stem/stromal cells [48]. StemID showed that cluster 5 may be the possible multipotent ESCs, which had a higher expression of Pdgfrb, a higher entropy score, and more connections with other cell types ( Figure 5C-D, F). It had the potential to differentiate into stromal cells, epithelial cells, and immune cells ( Figure 5E). Besides, this cluster also highly expressed Cd34 and Aldh1a2 ( Figure 5F-I) [49]. Cd34 + Klf4 + stromal-resident stem cells have been reported to directly contribute to endometrial regeneration [50]. The cluster5 highly expressed Cd34, but without detectable expression of Klf4. To conclude, we found the transcriptional signature of endometrial mesenchymal stem cells in vivo. And the cluster 5 that we explored here showed a great possibility to be the cells responsible for re-epithelialization.

Discussion:
There is still a huge void in the uterus-related research at the single-cell level. In humans, the sequence data of Cd13 + stromal and Cd9 + epithelial cells from endometrial tissues have been generated [8].
Researchers also used scRNA-seq to map the temporal transcriptomic changes in cultured primary ESCs along a decidual time-course and in response to the withdrawal of differentiation signals [51]. In mice, the transcriptional pro les of uterine epithelial cells at ve developmental stages, ranging from neonatal to mature stages were analyzed [52] These studies contained a limited number of cell types. A scMCA covering major cell types was completed [7]. It included the uterus and mice' other major organs as well, yet without in-depth analysis of the changes happened in the endometrium at different estrus states and left opportunities for later researchers to delineate the dynamic endometrial cell transformation of all cell types in the estrus cycle at the single-cell level. Our research is a systematical single-cell level study with the special focus on reconstructing the mice cell atlas at different estrus states and showed that different estrus states had sheer different cell composition.
We identify three transcription factors in the differentiation path of mononuclear phagocyte system. Irf7 has been reported to be involved in the regulatory pathway initiated in DCs during their response to microbial stimuli but dispensable in DC development [53]. Nr4a1 has been reported to be the target for modulating the in ammatory phenotype of monocytes and macrophages [54]. In our analyses, Irf7 expression was seen in the monocyte branch, and Nr4a1 was expressed highly in the DC branch, so the relations between these two transcription factors and their branches needed further experiments.
Likewise, understanding the decisional mechanism of broblast-to-myo broblast differentiation may be the key to tackle endometriosis and adenomyosis. Among the highly expressed genes that determined the myo broblast differentiation (Table 1), there were a lot of them haven't been reported, which shed light on future research.
Generally, the immune cells in the uterus endometrium showed a pro-in ammatory tendency in the regenerative phase and an anti-in ammatory tendency in the maturational phase. Yet the macrophages in the regenerative phase also expressed certain immune-suppressive genes, indicating the possibility that tissue-resident macrophages existed and contributed to endometrium repair. The ndings concerning uterus tissue-resident macrophages are rare. Putative tissue-resident macrophages have been reported to be spatially restricted and in association with areas of repaired, re-epithelialized endometrium [55]. In our pseudotime analysis ( Figure 4E), we found a branch in the macrophage differentiation path, which indicated the existence of a subtype of macrophage and may be the uterus tissue-resident macrophages. The molecular pro les of mouse uterus-resident macrophages remain unknown, which needed further research. Our results may provide insights into tissue remodeling and tackle endometrium abnormalities like endometritis.
We reported a novel group of markers for in vivo endometrial mesenchymal stem cells: Pdgfrb, Aldh1a2, and Cd34, which has not been reported before. Many proposed stem cell surface molecules like Cd73, Cd90, and Cd105 failed to up-regulated in vivo endometrial mesenchymal stem cells.
We thoroughly compared the functional traits and molecular pro les for each cell type in these two states ( Figure 6). With a template for the interaction of different cell types at the normal condition, future researchers can better identify abnormalities when compared. We also depicted the transition and transcription factors that shaped the differentiation of ESCs and the mononuclear phagocyte system, and many of them have not been reported before. The genes that have not been reported before can provide inspiration for subsequent basic research. The cell atlas of mouse uterus presented here would improve our understanding of the functional changes that occurred in the endometrium and helped us identify abnormalities that have not been apparent histologically.
However, several limitations of our study should be acknowledged. Firstly, this is an RNA-seq based bioinformatics study and caution must be taken when further extrapolating these results in vivo. Future study is needed to evaluate the presence of a transcript corresponds to its expression at the protein level. Experiments for veri cation of our ndings are crucial in the future. Secondly, the limited number of mice that participated in this study may affect the statistical power and the strength of our conclusions.

Conclusions:
This study is a s systematical single-cell level study to reconstruct the mice cell atlas at different estrus states, including epithelial cells, stromal cells, and immune cells. The functional traits and molecular pro les for each cell type in these two states are thoroughly compared. The mice cell atlas also delineates the transitions that shaped the differentiation of endometrial stromal cells and the mononuclear phagocyte system. This study offers clues about uterus tissue-resident macrophages and Pdgfrb + Aldh1a2 + Cd34 + endometrial mesenchymal stem cells in vivo. Tables: Table 1 Signi cant branch-dependent genes of mice stromal cells in different physiological states