Single‐cell RNA sequencing in oral science: Current awareness and perspectives

Abstract The emergence of single‐cell RNA sequencing enables simultaneous sequencing of thousands of cells, making the analysis of cell population heterogeneity more efficient. In recent years, single‐cell RNA sequencing has been used in the investigation of heterogeneous cell populations, cellular developmental trajectories, stochastic gene transcriptional kinetics, and gene regulatory networks, providing strong support in life science research. However, the application of single‐cell RNA sequencing in the field of oral science has not been reviewed comprehensively yet. Therefore, this paper reviews the development and application of single‐cell RNA sequencing in oral science, including fields of tissue development, teeth and jaws diseases, maxillofacial tumors, infections, etc., providing reference and prospects for using single‐cell RNA sequencing in studying the oral diseases, tissue development, and regeneration.

"weighted average". 1 Unlike traditional RNA analysis, single-cell RNA sequencing (scRNA-seq) is characterized by high-throughput and high-resolution transcriptomic analyses of individual cells. The scRNAseq can be utilized in the assessment of heterogeneous cell populations, reconstruction of cellular developmental trajectories, simulation of the stochastic gene transcriptional kinetics, and inference of gene regulatory networks. 2 Since the scRNA-seq was first reported, scRNA-seq has resulted in multiple pioneering studies in life science and medical research. Oral science research involves numerous areas, including oral histological structure and histogenesis, teeth and bone disease, oral maxillofacial deformity, mucosal disease, tumor disease, infectious disease, etc.
Although scRNA-seq has not been widely used in all aspects above, it is currently believed to be an indispensable tool that brings about advances in oral science. Cognizing the progress and prospects of this technique in oral science will contribute to the progress in oral science. Herein, current progress and underlying prospect for scRNA-seq in researches of oral science will be reviewed and discussed in this review.

| THE BRIEF HISTORY OF SCRNA-SEQ DEVELOPMENT
Traditional RNA sequencing requires microgram amounts of total RNA for analysis. Therefore, scRNA-seq was first reported by Tang et al. 3 in 2009 to tackle the problem that the samples obtainable were too scant to implement transcriptome analysis, which preferably can analyze the samples at single-cell level. In Tang et al.'s protocol, scRNA-seq dataset relied on manual manipulation of individual cells, thus it was unable to achieve multiplexing. 4 Single-cell tagged reverse transcription sequencing (STRT-seq), the first multiplexed scRNA-seq, was put forward by Islam et al. 5 in 2011. STRT-seq permits the introduction of a barcode for multiplexing, which allows simultaneous amplification of cDNAs from tens of thousands of cells, thereby reducing pre-processing of cells and reducing cost. 6 Nonetheless, only the 5 0 end of each cDNA is quantified by STRT-seq.
Subsequently, switching mechanism at the 5 0 end of RNA template sequencing (Smart-seq), developed by Ramsköld et al. 7 8 It is demonstrated that CEL-seq yielded more repeatable, sensitive and linear-amplified results compared with the PCR-based protocols.
Nevertheless, in addition to 3 0 bias, CEL-seq has the disadvantage of low sensitivity to lowly expressed transcripts. 8 In addition to the aforementioned strategies, conventional scRNA-seq technologies include Smart-seq2, 9 Single Cell RNA Barcoding, and Sequencing (SCRB-seq), 10 Massively Parallel Single-Cell RNA-Sequencing (MARS-Seq), 11 CEL-seq2 12 et al. They have made progress in automation and reduction of cost and reaction volume. However, they remain labor-intensive and time-consuming. 13 Some droplet-based scRNA-seq technologies, such as inDrop (indexing droplets) RNA sequencing, 14 Drop-seq, 15 and 10Â Genomics, 16 have been invented to make the sequencing of a large number of cells more highthroughput. Hydrogel microspheres are utilized to introduce barcoded DNA oligonucleotides to label each cell in inDrop, and reactions, including cell lysis, marking cells by barcodes, and cDNA synthesis, are performed in droplets. Successively, the cDNA is amplified by IVT. Drop-seq, unlike inDrop, uses barcoded beads to introduce oligonucleotides and PCR to amplify cDNA. 15 Gel bead in Emulsion (GEM), which significantly elevates cell capture efficiency, is the pivotal technology of 10Â Genomics and 10Â barcodes are exploited to further improve the throughput. 16 Summary of these three most prevalent droplet-based scRNA-seq technologies is presented in Table 1. Apart from droplet-based scRNA-seq technologies, nanowell-based scRNA-seq technologies, such as Seq-Well, 18 Microwell-seq, 19 single cell optical phenotyping and expression sequencing (SCOPE-seq) 20 et al., have been developed recently and endowed with the advantages of more simplified preparation of cells, improved optimization of optical imaging, fewer experiment reagents, and lower reaction volume. 13 The history of scRNA-seq development was summarized in Figure 1.

| RE-ANALYSIS OF ORAL HISTOLOGICAL STRUCTURE AND HISTOGENESIS USING SCRNA-SEQ
The profiles of transcriptome, epigenetics, and niche endow each single cell with unique characteristics. Through scRNA-seq, researchers can determine expression profiles of oral histology at single-cell resolution, which can clarify cell subpopulations, record signature genes, track variable cellular development trajectory and recognize intercellular crosstalk.

| Tooth histogenesis
The current restoration of tooth defects and deletions relies on synthetic materials without biological activity. A comprehensive understanding of the cellular and molecular mechanisms that regulate tooth structure and developmental processes will benefit the development of tooth regeneration engineering. 21 The scRNA-seq can help reexamine the oral structure from the perspective of single-cell resolution, facilitating the construction of oral structure profile. The use of single-cell sequencing has been applied for studies on teeth, whereas the focus differs. The scRNA-seq adopted in research from Sharir et al. was aimed to uncover the characteristics, locations, and function of dental stem cell. 22 Previously, the classical model of dental epithelial stem cells (DESCs) from mouse incisors suggests that the progenitor cells in the labial cervical loop (laCL) are in the vicinity to the outer enamel epithelium (OEE) or stellate reticulum (SR), which generate the transit-proliferating inner enamel epithelium (IEE) cells and terminally give rise to all the epithelial progeny cells (Figure 2A). 23 In contrast, the model from Sharir et al. pointed out that during homeostasis the stem cells residing in the IEE differentiated into ameloblasts and a small population of non-ameloblast epithelial cells in OEE and SR ( Figure 2B). Upon injury of proliferative IEE cells, the progenitors from OEE and cells from stratum intermedium (SI) entered cycling phase, differentiating into ameloblasts during the recovery state ( Figure 2C).
These results were further validated by RNAscope or immunofluorescence staining. 22 With scRNA-seq, another research group discovered novel cell types and marker genes of dental epithelial cells in mice.
Dental epithelial cells from incisors of postnatal day 7 mice were reclassified, identifying two novel subpopulations of ameloblasts in secretory stage and some previously unknown dental epithelial cell marker genes. Pseudotime analysis suggested one novel subpopulation of ameloblasts (Dentin Sialophosphoprotein [Dspp] +) differentiates into the other one (Ameloblastin [Ambn] +). 24 Krivanek et al.
applied scRNA-seq to explore the novel histology hierarchy of selfrecycling mouse incisors as well. 25 For example, Ryanodine Receptor 2 (RYR2) + ameloblast subpopulation and Thrombomodulin (THBD) + SI subpopulation, unrecognized cell subpopulations previously, were detected by scRNA-seq. In addition, the research analyzed the similarities and differences between growing mouse incisor and non-growing mouse molar and evaluated how similar mouse tooth model and human tooth biology are. 25 Fresia et al. 26 discussed essential similarities and differences between the three studies above. For example, a more complex hierarchy of mouse incisor epithelial cell The brief history of scRNA-seq development. The three categories, conventional scRNA-seq, droplet-based scRNA-seq and nanowell-based scRNA-seq, are classified according to Choi et al. 13 populations was reported in the studies from Sharir et al. 22 and Krivanek et al. 25 However, possibly owing to the more immature mice model sampled, fewer cell clusters were reported by Chiba et al. 24 It was likely due to variations in the analysis tools and parameters adopted as well. 26 Meanwhile, all three studies discovered a great amount of previously unrecognized markers in the nonameloblast counterpart. 26 The scRNA-seq can help map the differentially expressed genes (DEGs) to their spatial locations from single cell perspective. 27  Combined scRNA-seq with CAGE-seq, Cldn10 was found to be a novel SI marker, 29 as well as FXYD domain-containing ion transport regulator 4 (Fxyd4) and Dspp to be unrecognized ameloblast markers. 28 The mouse incisor serves as a major model to study tooth development. The in-depth exploration of the model by scRNA-seq updates our knowledge of the development and maintenance of tooth.
The comparison in scRNA-seq expression profile of mouse and human teeth revealed the differences between them. Some limitations may exist when using mouse teeth as model for human teeth. 25 Therefore, it is necessary to picture the comprehensive expression profiles of human teeth further by scRNA-seq. The scRNA-seq was used to discover the cellular heterogeneity and molecular signatures in human pulp. [30][31][32][33] From dynamics and differentiation trajectories analysis, endothelial cells exhibit the most dynamic behavior, while only minor differentiation trajectories are found in most dental pulp cell populations. 31 According to the ligand-receptor pair analysis among cell populations, the pulp cells communicated the most with other cell types, while T cells communicated the least. 31 Owing to its differentiation potential, human dental pulp stem cell (hDPSC) is promising for tooth regeneration engineering. 34 scRNA-seq can decipher some unrecognized traits of these stem cells. Based on the scRNA-seq data, Notch3 was found as a marker for hDPSC, which was also identified by lineage tracing in mouse tooth injury model. 35 Furthermore, from scRNA-seq analysis, the fate of hDPSC was possibly determined by the microenvironment it resided. 30,32 It is noteworthy that the cell population analysis on scRNA-seq data showed that monolayer culture of hDPSCs was significantly different from freshly isolated hDPSCs in cellular composition. 36 Hence, it may affect the application of hDPSC in tooth regeneration engineering. 36 Overall, based on scRNA-seq technique, these studies have established a more concrete profile of the cellular hierarchy, identity, position, and function during tooth histogenesis. The scRNA-seq data on teeth improve the interpretation of the cell biology and molecular mechanisms about the histogenesis of teeth, 26 which helps the development of teeth regeneration engineering.

| Oral mucosa
Oral mucosa is an essential barrier for human, and is constantly exposed to commensal microbiota and airborne antigens. Oral mucosal also withstands dietary antigens and frequent damage from mastication. 37 The scRNA-seq can contribute to unraveling the mystery of the unique barrier function of oral mucosa. The scRNA-seq has been used to analyze oral mucosa samples taken from different tissue F I G U R E 2 Reanalysis of location and lineage differentiation of dental epithelial stem cells from mouse incisors by scRNA-seq 22,23 (A) Progenitor cells located in the proximal portion of the outer enamel OEE (red) or SR (orange) of the laCL yielded the IEE cells (blue), and finally contributed to AMBs(green) according to classical model. 23 (B) During homeostasis, the stem cells residing in the IEE differentiated into AMBs, and another small population of the stem cells differentiated into non-AMB epithelial cells in OEE and SR according to novel model via scRNAseq. 22 38 Recently, the first human gingival cell atlas has been set up using scRNA-seq, uncovering the heterogeneity within major gingiva cell populations. 39 Another study by Williams et al. obtained scRNA-seq data from biopsies of the buccal and gingival mucosa, which profiled the cell hierarchy and molecular attributes of the oral mucosa. 40 A specific keratinocyte subpopulation was merely found in the gingiva, but not in the buccal mucosa. The DEGs analysis showed its top expressed genes included antimicrobial and inflammatory factors, and leukocyte chemotaxis pathway was identified as a top pathway, which may act through recruitment of neutrophils. 40 On the other hand, periodontal tissue was collected for scRNA-seq analysis as well. The molecular signatures of mesenchymal stem cells (MSCs) in periodontal tissue were discovered to be similar to that in dental pulp. 31 The spatial distribu-

| Maxillofacial alveolar bones
The 10Â Genomics technology was used to profile single-cell transcriptome of mouse mandibular alveolar bone 42  in alveolar bone but they exhibited higher immunosuppressive activity. 43 The immune cell atlas of alveolar bone revealed by scRNA-seq laid a solid foundation for in-depth research on how alveolar bone plays an essential role in orthodontic tooth movement, and reacts to inflammatory diseases, occlusal stress stimulation etc. For example, it is believed that the overreaction of the innate and acquired immune systems induced by dental plaque deteriorates periodontal tissues in periodontal lesions. 44 The orthodontic tooth movement is characterized by alveolar bone resorption, triggered by periodontal immunoreaction upon mechanical stimulus. 45 Therefore, the scRNA-seq atlas of the immune microenvironment in the alveolar bone may be beneficial for understanding the immune responses during physiologic and pathological bone remodeling from a new perspective.
Modification of the mandibular arch, the most rostral element of the pharyngeal arches, facilitates skeletal structure of vertebrate jaws. 46 The formation of jaws depends on neural crest-derived mesenchyme developed along the proximal-distal as well as the oralaboral axis. 47 The detailed gene expression at single cell resolution provided by scRNA-seq has made it possible to explore the molecular atlases that mediate histogenesis and organogenesis. 48  cancers. 63 Nevertheless, the detailed mechanism of CXCL14 in HPVnegative OSCC has not yet been discovered. 64 65 It was reported that CXCL14 could be used to screen cetuximab-responsive patients. 66 These results suggest that scRNAseq can be used to detect molecular targets of oral cancer that are difficult to be found or ignored by conventional sequencing technology. scRNA-seq data with immunohistochemistry, it was also demonstrated that the location of cells with p-EMT program was at the front edge of the tumor and adjacent to cancer-associated fibroblasts (CAFs), inferring that the p-EMT program played potential roles in tumor invasion. In addition, higher p-EMT scores was associated with nodal metastasis and adverse pathologic properties. 61,70 It was proposed that p-EMT program provides cancer cells with multifunctionality and multi-identity so that these cells have the capacity to adapt various microenvironments. 71 The strong plasticity of cancers was attributed to their adaptability, leading to the difficulty and complexity for cancer treatments. 72 Also, p-EMT program has clinical significance in guidance for the treatment because of its promising prediction of nodal metastases, lymphovascular invasion as well as extranodal extension in OSCC patients. 61 The discovery of p-EMT by scRNA-seq, which is likely to be more aggressive intrinsically and cannot be detected by current histopathologic technology, may point out the requirement for adjuvant therapy without any pathological manifestation. 73 In the future, by applying scRNA-seq, it is of importance to determine whether the cancer cell in the leading edge of tumor is undergoing p-EMT program.  74 In addition, the expression profiles of prognosis-related S100 protein family members in human HPVnegative OSCC were evaluated using the same scRNA-seq data. Combined with the bulk RNA-seq data, S100 Calcium Binding Protein A13

| Salivary gland
(S100A13) was detected as a significant differentially expressed gene.
Upregulation of S100A13 was related to decreased OSCC sensitivity to cisplatin. 75 Moreover, DEG analysis on scRNA-seq data demon-  61 indicating that a common pathogenesis was possessed by these TME cells. Therefore, a similar drug therapeutic protocol may be suitable to these patients.
The fibroblasts in TME of HNSCC were separated into two main subtypes-myofibroblasts (MFBs) and CAFs, and a third minor subtyperesting fibroblasts based on scRNA-seq studies. 61 Some OSCC cases showed a network pattern of arrangement of myofibroblasts which was proved to be responsible for more invasive behavior of the tumor. 78 CAFs have been shown to play a critical part in TME alterations that promote tumorigenesis and progression. 79  of OSCC by scRNA-seq, upregulation of which was reported to increase the OSCC stem-cell like activity. 94 Seventy-four genes with significantly higher expression in breast cancer stem cells than in non-CSCs were identified by scRNA-seq, many of which were not markers of breast cancer, consistent with the idea that there is a shared expression profile among the CSCs from different cancers. 90 The identification of tumor stem cells by scRNA-seq may reveal the underlying mechanisms of tumor therapy resistance and tumor recurrence.
To the best of our knowledge, current research on oral cancer from scRNA-seq perspective focus on squamous cell carcinoma. The illustration for key findings of oral squamous cell carcinoma through scRNA-seq is shown in Figure 3.

| Oral potential malignant disorders (OPMDs)
OPMDs have a statistically increased risk of progression into cancer.
The most common OPMDs are oral submucous fibrosis (OSMF), leukoplakia, lichen planus, and erythroplakia. 95 However, investigations into OPMDs were limited by the lack of means to analyze the small amount of cellular material. 96 With the ability of assessing transcriptome heterogeneity and the requirement of low quantities of starting material, scRNA-seq can solve these problems. The process of oral carcinogenesis is from the normal mucosa to precancerous lesions, and ultimately to cancer. The oral carcinogenesis can be induced by costimulus of arecoline and 4-nitroquinoline 1-oxide (4-NQO) in mouse model. It was shown by scRNA-seq that genes of two cell subtypes which were likely to be essential during carcinogenesis, were associated with the MYC_targets_v1 pathway. 96 These cell subtype markers can be applied to examine patients suffering from OPMD, to identify highrisk populations, and used as a treatment target. 96 As for OSMF, the presence of myofibroblasts (MFBs) and the constant α-smooth muscle actin (α-SMA) expression are regarded as landmark of deteriorating fibrosis and may change OSMF microenvironment, resulting in carcinogenesis. 97 Subsets of myofibroblasts have been identified by scRNA-seq in liver fibrosis, 98 a premalignant lesion of the liver. 99 It is reported that activated MFBs differentiated into various subpopulations, with markers such as of α-SMA, collagens et al. S100 calcium binding protein A6 (S100A6) was considered as a general marker of activated MFB. 98 It is promising that the MFB features identified in liver fibrosis  101 Smoking is also a key risk factor of OMPD. 103 The events emerging in bronchial epithelial cells caused by smoking inferred by scRNA-seq can provide a reference for the research on OMPD resulting from tobacco, because a similar process may arise in the oral mucosal cells of OMPD caused by smoking which can be captured by scRNA-seq. To the best of our knowledge, the studies of OPMD using scRNA-seq are limited, but scRNA-seq will undoubtedly bring new angles to study OPMD in future.

| Oral and maxillofacial deformity
The orofacial clefting (OFC) can be divided into three sub-phenotypes: iPSCs. 107 This research deciphered the potential mechanism of p63 mutation-driven CL/P and provided an underlying treatment method for CL/P.

| Infectious diseases
Infectious diseases often occur in the oral and maxillofacial region.
The complex, and diverse interactions between the host cells and bacteria limit the depth of understanding the infectious diseases by bulk RNA sequencing. 108 The scRNA-seq, a technique powerful enough to identify intercellular heterogeneity and profile distinctive transcriptome reflecting each cell's biological characteristics, 109  contributing to pre-osteoblasts during inflammation after periodontal therapy. 115 Second, the gene expression in the scRNA-seq dataset that was differentially regulated in periodontitis assisted in predicting cellular interactions with the aid of algorithms such as CellPhoneDB.
It was found that the stromal cells were responsible for recruiting immune cells to the damage site of periodontitis, especially the neutrophil. 40 The study by Qian et al. showed that increased intercellular communication arose between macrophages and T/B cells in periodontitis-affected tissues. 113 Signals that interacted with macrophages and other cell types were enriched in immune checkpoint pathways associated with exhaustion of T cells. 114 Activation of Ephrin-Eph signaling mediating crosstalk between endothelial cells and pre-osteoblasts maybe responsible for pathological bone loss in periodontitis. 115 Third, cell-specific expression patterns of periodontitis susceptibility genes were identified. Most of the Mendelian disease genes related to periodontitis were expressed predominantly within the immune cells, except C1S and C1R which were expressed only in fibroblasts. 40 In conclusion, the pathogenesis of periodontitis can be further understood by scRNA-seq.
The advancement of scRNA-seq promotes studying the heterogeneity of cellular response to viral infection. 116   was enriched in many human organs. 122 Research with the combination of bulk RNA-seq and scRNA-seq revealed that the frequency of cells expressing ACE2 in alimentary canal organs was evidently higher than that in lung. 123 The scRNA-seq data from human minor salivary glands and gingival mucosa confirmed that the expression level of SARS-CoV-2 viral entry factor was higher in epithelium. More specifi- and fibrosis in renal tubular cells were relevant to therapeutic failure. 128 Almost identical scRNA-seq results have also been analyzed by Der et al. 129 Keratinocyte of oral mucosal presented similar IFN-response signatures, and thus biopsy of oral mucosal may help the diagnosis on lupus nephritis. In addition, the immune cell landscape in kidneys of SLE patients has been established by scRNA-seq, 130 which may provide a reference for the immune cell profile of oral site of SLE patients.
Rheumatoid arthritis (RA), another type of chronic autoimmune disease mainly affecting synovial joints, results in pain and restricted movement. 131 It is reported that the temporomandibular joint (TMJ) is involved in RA. 132  The volume of sample used by scRNA-seq may be too small to reflect the whole damaged tissues. In addition, the tissue dissociation to obtain single cell cannot maintain the spatial data of the isolated samples. 1 The samples collected for scRNA-seq also have some limitations. For example, the current study collected normal and diseased tissues from periodontitis patients for scRNA-seq. However, periodontitis is a systemic disease. Thus, sequencing samples from patients with periodontitis that are considered normal may not be representative of healthy periodontal tissue. Therefore, it is necessary to collect periodontal tissue from patients with completely healthy periodontium for scRNA-seq. 111 For most scRNA-seq studies, samples were taken at a single time point. Samples should be taken at multiple time points to understand dynamic gene expression heterogeneity. In addition, non-coding RNA acts as an important role in cancer, but scRNA-seq including non-coding RNA is still rare. 133 Moreover, although the necessity of scRNA-seq for microbes has been considered, it is halted by the difficulties including low mRNA transcripts in microbes, non-polyadenylated mRNA of microbes, and bacterial cellwall-based barrier, or cell-membrane-based barrier for cell lysis required for subsequent sequencing. Until recently, a scRNA-seq method named microbial split-pool ligation transcriptomics (micro-SPLiT) has been introduced to solve the difficulties in scRNA-seq for microbes. 134 More method development studies of scRNA-seq should aim for enhanced sensitivity, reduced cost, higher throughput, and decreased technical noise, 13 which will help extensive application of scRNA-seq in basic research and clinical practice.

AUTHOR CONTRIBUTIONS
J. Wu, Y. Ding and J. Wang contributed equally to the manuscript conception, drafting of this review, and wrote this paper; F. Lyu, Q. Tang, J. Song, Z. Luo, Q. Wan and X. Lan drafted and critically revised the manuscript; Z. Xu and L. Chen designed and revised the paper. All authors gave final approval and agreed to be accountable for all aspects of the work.