Tumor-Initiating Cells Are Enriched in CD44hi Population in Murine Salivary Gland Tumor

Tumor-initiating cells (T-ICs) discovered in various tumors have been widely reported. However, T-IC populations in salivary gland tumors have yet to be elucidated. Using the established Pleomorphic Adenoma Gene-1 (Plag1) transgenic mouse model of a salivary gland tumor, we identified CD44high (CD44hi) tumor cells, characterized by high levels of CD44 cell surface expression, as the T-ICs for pleomorphic adenomas. These CD44hi tumor cells incorporated 5-bromo-2-deoxyuridine (BrdU), at a lower rate than their CD44negative (CD44neg) counterparts, and also retained BrdU for a long period of time. Cell surface maker analysis revealed that 25% of the CD44hi tumor cells co-express other cancer stem cell markers such as CD133 and CD117. As few as 500 CD44hi tumor cells were sufficient to initiate pleomorphic adenomas in one third of the wildtype mice, whereas more than 1×104 CD44neg cells were needed for the same purpose. In NIH 3T3 cells, Plag1 was capable of activating the gene transcription of Egr1, a known upregulator for CD44. Furthermore, deletion of sequence 81–96 in the Egr1 promoter region abolished the effect of Plag1 on Egr1 upregulation. Our results establish the existence of T-ICs in murine salivary gland tumors, and suggest a potential molecular mechanism for CD44 upregulation.


Introduction
Tumors comprise of a heterogeneous population of cells with varying morphologies and functions. It has been proposed that tumors may not be mere monoclonal expansions of cells. Instead, they may be sustained by a specialized type of tumor-initiating cells (T-ICs), also known as cancer stem cells, which are capable of self-renewal and aberrant differentiation [1]. Many tumors have a hierarchical organization of T-ICs, rapidly dividing cells and differentiated tumor cells. These T-ICs are not only a renewable source of tumor cells, but are also a source of tumor resistance, leading to tumor progression, metastasis and recurrence [2][3][4]. Hence, identifying these cells and their respective cell surface markers are crucial to the understanding of the mechanisms that govern cancers, thereby leading to the development of better and more strategic oncological therapies.
Early experimental evidence for T-ICs came from human acute myeloid leukemia studies. Human CD34 + CD38 2 acute myeloid leukemia cells were capable of initiating leukemia when transplant-ed in non-obese diabetic mice with severe combined immunodeficiency (NOD/SCID mice) [5]. More recently, T-ICs were identified and isolated from epithelial tumors such as breast tumors [6][7][8], glioblastomas [9][10] and colorectal carcinomas [11], where each of these tissue-specific T-ICs displayed distinct cell surface markers. For instance, those discovered in human breast tumors had a subpopulation of CD44 + CD24 2/low tumor cells, capable of initiating and sustaining tumor growth after they were xenografted into NOD/SCID mice. These CD44 + CD24 2/low cells contained mixed populations of epithelial tumor cells, and often phenocopied the histological heterogeneity of their parent tumors [8]. Another example of T-ICs includes the CD133 + tumor cells discovered in glioblastoma multiforme and medulloblastomas of the human brain, where CD133 + subpopulations accounted for 5-30% of the total number of tumor cells. These cells were found to be tumorigenic and could reproduce the phenotypic diversity and differentiation pattern of the parent tumors when injected intracerebrally [12]. In human colorectal cancer, the ability to engraft in vivo in NOD/SCID mice is restricted to EpCAM high/ CD44 + tumor cells.
Tumors originated from EpCAM high /CD44 + cells also maintained a differentiated phenotype and reproduced the phenotypic heterogeneity of their parental lesions [13].
Pleomorphic adenomas are benign tumors found in salivary glands. They are the most common of all salivary gland tumors, and can account for up to 80% of all parotid neoplasms. While most human pleomorphic adenomas are benign and can be easily treated via various forms of curative resections, some of them may recur decades after removing the primary tumor. An even smaller percentage of them can become malignant (carcinoma expleomorphic adenoma) with metastasis to distant and remote sites [14][15]. Investigations on the tumorigenesis of pleomorphic adenomas revealed that the oncogenic activation of Pleomorphic Adenoma Gene 1 (Plag1) played a pivotal role in their development within salivary glands. Various studies including ours verified that Plag1 transgenic mice spontaneously developed salivary gland tumors with similar histopathological features to human pleomorphic adenomas [16][17][18]. Using this model, we sought to establish the presence of T-ICs and identify the corresponding cell surface markers for T-IC isolation. In this study, we provided evidence that CD44 hi cells function as the T-ICs in pleomorphic adenomas.

CD44 protein expression in normal salivary gland or salivary gland tumor cells
Both normal salivary gland and tumor salivary gland cells displayed CD44 staining via FACS analysis. The mean fluorescent intensities (MFI) for CD44 was significantly higher on tumor cells than on wildtype cells (84 vs. 7, p,0.05) (Figure 1a, b). We further characterized the CD44 cells into CD44 hi , CD44 intermediate (CD44 int ) and CD44 neg subpopulation by MFI. We noted that the CD44 hi population registered MFI that was one log higher than that of the CD44 neg population. The percentage of CD44 hi cells in the total salivary gland tumor cell population was 8.6%. This was significantly higher than the percentage of CD44 hi cells in the wildtype salivary glands (0.6%) (Figure 1c). Thus, CD44 protein expression was significantly upregulated in salivary gland tumor cells.
Phenotype of CD44 hi tumor cells CD133 and CD117 are two markers frequently associated with T-ICs [7,10,[19][20][21]. We examined their expression in different CD44 subpopulations of tumor cells by flow cytometry. CD44 hi subpopulation contained cells which co-stained with CD117 and CD133 antibodies. These cells constituted about 25% of the total CD44 hi cells. In contrast, only 6% of CD44 int cells stained for CD117 and CD133, and less than 0.5% of the CD44 neg cells expressing CD117 and CD133 (Figure 2a).
Many studies have revealed that T-ICs do not divide as fast as their non-T-IC counterparts [22][23][24]. To determine if this was the case in salivary gland tumors, we compared the cell cycling rate and label retaining ability of CD44 hi and CD44 neg tumor cells (Figure 2b). After a 5-day pulse of BrdU, about 3.5% CD44 hi cells incorporated BrdU, compared to 19% of CD44 neg cells, indicating that CD44 hi cells were slow-cycling compared to CD44 neg cells. When the second group of mice was similarly pulsed with BrdU for 5 days, followed by a rest period of 4 weeks, about 3% CD44 hi cells were still BrdU positive, whereas only about 5% of CD44 neg cells retained BrdU marker. Therefore, majority of CD44 hi cells retained their BrdU over the course of 4 weeks whereas only a small fraction of CD44 neg cells were able to do so. The slow baseline cycling rate and robust BrdU retaining ability of CD44 hi cells were consistent with stem cell like properties [22][23][24].
Transcription factors such as Oct-4, Nanog, Sox2 and telomerase are implicated in linage commitment or maintenance, and function for cells of stem cell-like properties [25][26][27][28][29]. We performed realtime PCR to assess gene transcript levels for Oct4, Nanog, and Sox2, as well as mouse telomerase reverse transcriptase (mTRET), a catalytic subunit of telomerase, in CD44 hi and CD44 neg tumor cells. Transcript levels for those stem cell related factor were higher in CD44 hi cells than in CD44 neg cells (Figure 2c).

High tumorigenicity of CD44 hi tumor cells in vivo
We have demonstrated with Plag1 transgenic mouse model that pleomorphic adenomas contained a population of CD44 hi cells that co-express other common T-ICs markers and propagate in slow-cycling state, and have upregulated stem cell related transcription factors. To determine if these CD44 hi cells could induce tumors in mice, in vivo tumorigenic experiments were performed. For all our in vivo transfer experiments, immunocompetent wildtype CD57/BL6 mice instead of immunocomprised NOD/SCID mice were used as hosts [30].
The pleomorphic adenoma tumor cells were sorted into various populations according to their surface markers. Comparable cell populations were injected into bilateral salivary glands of the same mouse. Unsorted 1610 5 tumor cells initiated tumors in about two thirds of the recipients, and 1610 4 unsorted tumor cells generated tumors in less than one third of the recipients. In contrast, as few as 1610 3 sorted CD44 hi tumor cells gave rise to tumors in a majority of recipients (7/9); 5610 2 sorted CD44 hi tumor cells generated tumors in one third of the recipients. For CD44 neg tumor cells, 1610 5 cells gave rise to tumors in more than half of recipients (5/9); 1610 4 tumor cells gave rise to tumor in one eighth of the recipients. CD44 hi cells showed 10-20 fold greater ability to initiate tumors in wildtype mice compared with the total number of tumor cells ( Table 1), suggesting that the CD44 hi population was enriched for T-ICs.
We next asked whether CD117 and CD133 can be used as T-ICs markers in our tumor model. When CD44 neg CD117 + CD133 + tumor cells were transplanted at the dose of 1610 4 cells, two out of eight recipients developed a secondary tumor. However, when CD44 hi CD117 + CD133 + triple marker positive tumor cells were sorted and transplanted into the recipients, these cells initiated secondary tumors in one third of the recipients at the dose of 2610 2 cells. Thus, T-ICs of pleomorphic adenomas were further enriched in CD44 hi CD117 + CD133 + populations, but not CD44 neg CD117 + CD133 + populations.
Tumors initiated by CD44 hi or CD44 hi CD117 + CD133 + cells shared similar histological features with the primary tumors (Figure 3a-c). They often form luminal or cyst-like structures, and also contained myoepithelial cells in a disordered pattern. CD44 expression was similar (6-10%) in both the primary and secondary tumors initiated by CD44 hi or CD44 hi CD117 + CD133 + cells transplants (Figure 3a-d). Mean fluorescent intensity analysis of CD44 revealed that all three groups of tumors expressed similar levels of CD44 (Figure 3e). Secondary tumors initiated by CD44 hi or CD44 hi CD117 + CD133 + cells predominantly contained CD44 neg cells. Therefore, CD44 hi tumor cells were able to produce secondary tumors that resemble primary tumors, consisting of both CD44 hi and CD44 neg cells.

Plag1 upregulates Egr1 transcription
Egr1, Early Growth Response 1, a positive regulator of CD44 [31][32], was found to be significantly upregulated in Plag1 transgenic mice tumors (unpublished microarray analysis). Plag1 and Egr1 gene transcript levels were assessed by realtime quantitative PCR with cDNA extracted from CD44 hi and CD44 neg tumor cells. Both Plag1 transcripts and Egr1 transcripts levels were significantly higher in CD44 hi cells than in CD44 neg cells (Figure 4a).
Egr1 promoter sequence analysis revealed 4 putative Plag1 binding consensus sites (Figure 4b). Co-transfection of Plag1 expression vector (pCI-neo-Plag1) and mouse Egr1 reporter construct containing all four binding sequences induced robust luciferase activity. This data indicated that Plag1 could serve as a positive regulator for Egr1 transcription. To identify the promoter region important for Plag1-mediated upregulation of Egr1, truncated forms of Egr1 promoters were constructed to contain only 3 out of 4 putative binding sequences (Figure 4b). Deleting the binding sequences in Egr1 promoter region 2750-2735 or 2301-2287 did not dampen the effects of Plag1 on Egr1 upregulation. However, when the Egr1 promoter sequence in region 81-96 was deleted, upregulation of Egr1 induced by Plag1 was almost completely abolished ( Figure 4c). These results demonstrated that promoter sequences 81-96 were necessary for Plag1-mediated Egr1 upregulation.

Discussion
A majority of T-IC studies employ human tumor xenotransplants that utilize immunocompromised mice without a functional adaptive immune system (T and B lymphocytes) as the recipients of human T-ICs [33]. However, these mice still harbor a functioning innate immune system which could cause cellular 'graft' rejection. Immune rejection of transplanted T-ICs would result in an inaccurate assessment of the frequency of T-ICs in tumors [30,34]. A recent study by Quintana et al. highlighted the impact of immune rejection on T-IC frequency estimations [29]. Using a NOD/SCID mice model with common gamma chains knocked out (NOD/SCID IL2g 2/2 ), they showed that human melanoma cells initiated tumors 100% of the time upon transplantation into NOD/SCID IL2g 2/2 hosts. However, with conventional NOD/SCID recipients, the frequency of tumor-initiating cells was found to be about 0.1% [34]. Intrinsic differences among various immunodeficient mouse strains could account for the T-IC frequency variability, and such discrepancies could greatly confound the interpretation of experimental data. Therefore, assessing the efficacy of T-IC models in syngeneic transplantation systems is a critical and imperative step that formed the foundation and basis of our testing T-ICs in syngeneic, immunocompetent hosts.
We found that salivary gland tumors induced by Plag1 overexpression contained CD44 hi cells that incorporated BrdU at a low cycle rate and retained BrdU better than CD44 neg cells; and that CD44 hi cells have higher levels of transcripts of stem cell related genes, Oct-4, Nanog, Sox2 and telomerase. These data collectively suggest a stem cell-like phenotype for CD44 hi tumor cells. A few hundred transplanted tumor cells were able to initiate new tumors in syngeneic recipients. Secondary tumors shared the same histopathological features and cellular composition as their parent tumors. These findings supported previous reports that identified CD44 as a stem cell marker in colon, prostate cancer, breast, and head and neck squamous cell carcinomas [22,[35][36][37]. These studies and our results (Figure 1) indicated that CD44 expression was widely distributed across a number of tumors and a range of CD44 expression levels-negative to very high levelswere observed in the tumor cells. This broad pattern of CD44   [22], where CD44 hi populations were defined by cells exhibiting a mean fluorescent intensity (MFI) of at least half a log higher than CD44 neg cells. In this study, we defined CD44 hi cells as exhibiting MFI that was one log higher than that of CD44 neg cells. This is a more stringent criterion. Our results revealed that in murine salivary gland tumor, CD44 hi population was enriched for T-ICs as well.
CD113 is a marker expressed on hematopoietic stem cells isolated from cord blood. The use of CD133 as a T-IC marker is still debatable, largely due to the variability in its tissue-specific expressions. For example, in a colon adenocarcinoma xenotransplant model, CD133 + cells did not display increased tumorigenicity [22] even though they have been reportedly found in T-ICs [38]. The intrinsic difference among these tumor models could contribute to this discrepancy. CD117, a cytokine receptor  expressed on hematopoietic stem cells and other cell types, has been detected on the surfaces of various tumors stem cells, including those found in leukemia and lung cancer [20][21]. Our study showed higher CD133 and CD117 expression levels in tandem with CD44 hi cells, when compared with the CD44 neg population. Triple marker CD44 hi CD133 + CD117 + cells possessed potent tumor initiating capacities, indicating that CD44 hi CD133 + CD117 + could be used as a robust marker for T-IC identification in salivary gland tumors. Although CD44 + markers have been widely used to isolate and identify T-ICs, it was not until recent that the functional roles of CD44 in T-ICs were investigated [11]. Du et al. used RNA interference to knockdown CD44 and CD133 in colorectal T-ICs isolated from patients. They showed that the deletion of CD44, and not CD133, strongly prevented clonal formation and inhibited the tumorigenicity of T-ICs in a xenografted model [11]. Therefore, the authors deduced that CD44 not only functioned as a marker for T-ICs, it could also participate in tumorigenic signaling pathways.
Our data showed that both Plag1and Egr1 gene transcriptions were upregulated in murine salivary gland CD44 hi tumor cells compared to CD44 neg cells (Figure 4a). We also demonstrated that Plag1 activated Egr1 gene transcription, and Egr1 promoter region 81-96 was necessary in mediating this process (Figure 4c). The CD44 gene contains an Egr1 binding site at 2301 in the promoter region, and this Egr1 binding motif is indispensable to the stimulus-induced expression of CD44 [31][32]. These data collectively indicate a plausible Plag1-Egr1-CD44 regulatory pathway in the T-ICs of pleomorphic adenomas.

Conclusion
Our study demonstrated that CD44 hi cells function as the T-ICs in pleomorphic adenomas that are capable of generating secondary tumors with similar histopathological features to the parent tumors. In addition, we provided evidence that Plag1 was capable of activating gene transcription of Egr1, an established upregulator for CD44, suggesting that Plag1 may be potentially involved in upregulation of CD44.

MMTV-PLAG1 transgenic mice
Mouse mammary tumor virus (MMTV)-PLAG1 transgenic mice, maintained in C57BL/6 background were generated at Experimental Animal Center, Department of Medical Genetics, Shanghai Jiaotong University School of Medicine [17][18]. All transgenic mice used in this study were genotyped via PCR, and were all heterozygous for the Plag1 transgene. Almost all the transgenic mice developed submandibular gland pleomorphic adenomas between 4-12 weeks old. Age (6-8 weeks old) and sexmatched C57BL/6 wildtype mice were used as controls. All the procedures described were approved by the Animal Use and Care Committee of Shanghai Jiaotong University School of Medicine (approval number: SYKX-2008-0050).

Preparation of single cell suspensions from salivary gland tumors and normal salivary gland tissues
To generate single cell suspensions from salivary gland tumors, tumor samples from Plag1 transgenic mice were minced with scissors, mixed with 0.25% trypsin-EDTA and incubated at 37uC for 1 hour in a shaking water bath. Following incubation, cells were collected after blocking with PBS containing 10% FBS (Gibco, Carlsbad, CA, USA) and serially filtered through 200 and 100 mm nylon meshes. Contaminated red blood cells were removed by osmotic lysis and the filtered cells were suspended in PBS/1%FBS. Similarly, normal salivary gland tissues from C57BL/6 wildtype mice were also prepared into single cell suspensions.
For cell surface staining, cells were counted and transferred to a 15-ml tube before they were washed and re-suspended. To minimize the unspecific binding of antibodies, cells were first incubated with mouse immunoglobulins at 0.06-0.1 mg/1610 6 cells on ice for 10 minutes, after which the samples were washed twice. Fluorescent-conjugated antibodies were then mixed with the re-suspended cells, incubated for another 30 minutes on ice and washed twice again.
To label the cells and investigate their BrdU retaining abilities, the mice were injected with BrdU at 50 mg/kg twice daily for 5 days. They were then divided into two groups, where one was sacrificed immediately and the other group was left alone for 4 weeks. Cell surface staining was then performed on the single cell suspensions before being permeablized with 1% PFA and 0.5% Tween-20 in PBS for 30 minutes at room temperature. This was followed by DNaseI treatment at 100 U/ml (Sigma, St Louis, MO, USA) in a 37uC water bath for 15 minutes. Anti-BrdU antibodies were then added into cells for 45 minutes at room temperature. FACS Caliber (BD, San Jose, CA, USA) was used for cell acquisition and CellQuest (BD, San Jose, CA, USA) was used for analysis. For all analyses, live cells were gated based on 7-AAD negative staining.

Cell sorting and tumorigenicity experiments
Cell sorting was performed using a BD FACSAria cell sorter. Sorted cells were spun down by low speed centrifugation(1000 rpm for 5 min) and re-suspended in RPMI 1640 supplemented with 10% FBS, 20 mM Hepes and 2 mM Lglutamine. In all the experiments, a small aliquot of cells was set aside to confirm cell counts and viability. Once confirmed, cells were diluted to appropriate injection doses and injected into murine submandibular glands with 28G needles and microliter syringes (Hamilton, Reno, NV, USA) under anesthesia. To minimize experimental variability, comparative groups of cells were injected into the contralateral submandibular glands of the same animal. The injected mice were followed up intensively for the subsequent 3 months.