CD44 Expressed on Cancer-Associated Fibroblasts Is a Functional Molecule Supporting the Stemness and Drug Resistance of Malignant Cancer Cells in the Tumor Microenvironment


  • Yumi Kinugasa,

    1. Department of Signal Transduction, Research Institute for Microbial Diseases, Osaka University, Osaka, Japan
    Search for more papers by this author
  • Takahiro Matsui,

    1. Department of Signal Transduction, Research Institute for Microbial Diseases, Osaka University, Osaka, Japan
    2. Department of Hematology and Oncology, Osaka University Graduate School of Medicine, Osaka, Japan
    Search for more papers by this author
  • Nobuyuki Takakura

    Corresponding author
    1. Department of Signal Transduction, Research Institute for Microbial Diseases, Osaka University, Osaka, Japan
    2. Japan Science and Technology Agency, Tokyo, Japan
    • Correspondence: Nobuyuki Takakura, M.D., Ph.D., Department of Signal Transduction, Research Institute for Microbial Diseases, Osaka University, 3-1 Yamada-oka, Suita, Osaka 565-0871, Japan. Telephone: +81-6-6879-8312; Fax: +81-6-6879-8314; e-mail:

    Search for more papers by this author


Cells constituting the tumor microenvironment are attractive targets for developing new cancer therapies. Here we show that cancer-associated fibroblasts (CAFs) support tumor growth in vivo and maintain the stemness of cancer stem/initiating cells in an in vitro model using an established CAF cell line. We found that CD44 is abundantly expressed on CAFs. This molecule is a cancer stem cell marker in several tumors, but its role in tumorigenesis when expressed by CAFs has not been investigated. It is generally accepted that hypoxic and hyponutritional conditions are triggers of cancer malignancy. We found that CAFs strongly express CD44 in hypoxic and avascular areas in the tumor and that its expression on established CAFs is upregulated under hypoxic and hyponutritional conditions in vitro. In addition, CAF CD44-positivity in tumor tissues was increased after treatment with inhibitors of angiogenesis. Using cocultures and tumor sphere formation assays, CAFs from wild-type mice were found to sustain the stemness of cancer stem/initiating cells, while CD44-deficient CAFs did not. Furthermore, CD44 was involved in malignant cancer cell drug resistance mechanisms. In conclusion, our study suggests that CD44 on CAFs is a functional molecule contributing to the maintenance of cancer stem cell populations in the tumor microenvironment. Stem Cells 2014;32:145–156


Tumor stromal tissue consists of endothelial cells forming blood vessels, activated fibroblasts, infiltrating inflammatory cells, and extracellular matrix. Activated fibroblasts known as cancer-associated fibroblasts (CAFs) are critical for tumor development, progression, and metastasis in the tumor microenvironment [1, 2]. Recent studies suggest that there are several different subpopulations of CAFs expressing partially overlapping markers, including α-smooth muscle actin (αSMA), fibroblast-activated protein, fibroblast-specific protein-1 (S100A4), neuron-glial antigen-2 (NG-2), and platelet-derived growth factor β-receptor (PDGFRβ) [3, 4]. CAFs are known to enhance angiogenesis by secreting factors which activate endothelial cells and pericytes [5, 6]. It has been reported that there are several mechanisms for the development of CAFs in tumors. Local fibroblasts or fibroblast precursors, stimulated by PDGF or transforming growth factor-β (TGF-β) family members, are considered as the major source of CAFs [1, 3]. Tumor-associated PDGFRβ+ progenitor perivascular cells are recruited from bone marrow, differentiate into pericytes, and regulate vessel stability and vascular survival in tumors [7]. TGF-β1 induces the endothelial to mesenchymal transition, which is a source of CAFs [8, 9]. It is generally accepted that CAFs support survival or proliferation of cancer cells. In addition, their possible role as niche cells in supporting stemness of cancer stem cells (CSCs) has been suggested.

CSCs have been reported to possess self-renewal ability and to differentiate into cancer cells. It has been reported that sphere forming capacity reflected the self-renewal ability, while serial transplantation capacity reflected the character of progenitor cells [10]. In this study, we have used the sphere formation assay to determine the character of CSCs. Some markers used for the identification of CSCs have been determined, including cell surface molecules such as CD19, CD20, CD24, CD34, CD38, CD44, CD90, CD133, and others [11], as well as the detoxifying enzyme, aldehyde dehydrogenase one (ALDH1) [12-15]. ALDH1 plays an essential role in maintaining stemness and is the regular biological pathway for normal stem cells as well as CSCs. ALDH1 was also investigated as a specific marker for identifying and isolating normal and malignant human colonic stem cells [13]. Previously, we have reported that transcription of the PSF1 gene correlates with malignancy of cancer cells [16]. PSF1 is a member of the GINS (Go-Ichi-Ni-San) complex associated with the formation of the DNA replication fork. We found that PSF1 is required for acute proliferation of normal stem cells such as epiblasts in the embryo and hematopoietic stem cells in bone marrow [17, 18]. PSF1 expression visualized by enhanced green fluorescent protein (EGFP) under the transcriptional control of the PSF1 promoter in lung cancer cells (LLC-PSF1p-EGFP cells) indicated that a high level of PSF1 transcriptional activity in cancer cells (PSF1-EGFPhigh) derived from tumors in vivo was associated with a higher invasive, tumorigenic, and metastatic activity compared with PSF1-EGFPlow cancer cells. In addition, PSF1-EGFPhigh cancer cells had an embryonic stem cell-like gene expression profile [16]. Moreover, we recently found that PSF1-EGFPhigh cancer cells showed greater drug resistance than PSF1-EGFPlow cells after administration of anti-cancer drugs [19]. Therefore, we concluded that PSF1 promoter activity correlates with cancer stemness [16].

In vitro culture systems for the maintenance of CSCs are useful to understand how stemness of CSCs is maintained in the niche. In this study, we established a stromal cell line which supports the growth of tumors in vivo and the stemness of CSCs in vitro. We found that these stromal cells strongly express CD44 and documented a role for stromal CD44 in maintaining stemness of CSCs. CD44 is a type I transmembrane glycoprotein, an adhesion molecule that is upregulated following tissue injury, and is implicated in many chronic inflammatory diseases [20, 21]. It has been reported that some variant isoforms of CD44 (CD44v) induced a metastatic phenotype in cancer cells [22]. Therefore, CD44v is implicated as one of the cell surface markers associated with CSCs in several types of cancers [23]. Recently, CD44v has been showed to regulate redox status by interacting with a glutamate-cysteine transporter, and thereby promoting tumor growth [24]. In contrast, the shortest, standard, or hematopoietic isoform of CD44 is expressed at high levels on the membrane of a variety of cells [25]. However, to the best of our knowledge, the biological significance of CD44 expressed on CAFs has not yet been determined. Here, we assessed whether stromal CD44 plays a role in maintaining the stemness of CSCs by comparing CAFs from tumors induced in wild-type or CD44-mutant mice using CSC models as described above.

Materials and Methods

Cell Culture and Cell Line Derivation

The mouse melanoma cell line B16, the Lewis lung carcinoma cell line LLC, and the endothelial cell line bEND.3 were maintained in Dulbecco's modified Eagle's medium (DMEM) (Sigma, St. Louis, MO, with 10% fetal bovine serum (FBS; Sigma), and penicillin/streptomycin (P/S, Invitrogen, Carlsbad, CA, The human colorectal cancer cell line HT29 was maintained in RPMI-1640 (Sigma) with 10% FBS and P/S. Cells were cultured at 37°C in a humidified 5% CO2 atmosphere.

Tumor stromal cells defined as CAFs were isolated from B16 tumors generated by subcutaneous injection into CAG-EGFP/C57BL/6 mice (The Jackson Laboratory, BarHarbor, ME,, C57BL/6 mice (SLC, Shizuoka, Japan,, or CD44 knockout (KO) mice (The Jackson Laboratory) and grown in DMEM with 10% FBS and P/S. We designated CAFs derived from CD44-KO mice, KO-CAFs. CAFs were immortalized by transfection with the simian virus 40 large T antigen and cloned as described previously [26]. pCMV-E2-Crimson Vector constructs (Clontech, Mountain view, CA, were transfected into CAFs and HT29 cells by Lipofectamine 2000 (Invitrogen) and stably transfected cells were selected using G418 (Invitrogen). To establish KO-CAFs stably expressing CD44, the V5-tagged mouse CD44 gene was transfected into KO-CAFs as previously reported [26].

Preparation of Cell Extracts and Western Blotting

Cells were washed with phosphate-buffered saline (PBS), lysed in SDS sample buffer (125 mM Tris-HCl, pH 6.8, 2% sodium dodecyl sulfate, 10% glycerol, 0.003% bromophenol blue, 5% 2-mercaptoethanol). Cell extracts were analyzed by immunoblotting using rabbit anti-NG2 antibody (Millipore, Bedford, MA,, mouse anti-αSMA antibody (Sigma), rat anti-mouse/human CD44 antibody (IM7, eBioscience, San Diego, CA,, or mouse anti-β-actin antibody (Sigma) as the primary antibodies. Horseradish peroxidase (HRP)-conjugated anti-rabbit, mouse, or rat IgG antibody (The Jackson Laboratory) was used as the secondary antibody. The immunoreactive proteins were visualized using the ECL Prime Western Blotting Detection system (GE Healthcare, Buckinghamshire, U.K., The blots were scanned using the imaging densitometer LAS-3000 mini (Fujifilm, Tokyo, Japan).


Seven- to eight-week-old C57BL/6 female mice (for the LLC and B16 experiments) and KSN/Slc female mice of the same age (for HT29 experiments) were purchased from Japan SLC. All animal studies were approved by the Osaka University Animal Care and Use Committee. Subcutaneous inoculation was performed by injecting 1 or 2 × 106 cells into the flanks of mice. For coinjection analysis, subcutaneous inoculation was performed by injecting 1 × 103 B16 cells and 1 × 103 WT-CAFs (positive for CD44) or 1 × 103 LLC cells and 1 × 103 WT-CAFs (positive for CD44) into the flanks. Tumor volumes were measured with calipers and calculated as width × width × length × 0.52.

Immunohistochemical and Immunofluorescence Staining

Tissue fixation, preparation of tissue sections, and staining with antibodies was performed as described previously [17]. For immunofluorescence staining, fluorescein isothiocyanate (FITC)- or phycoerythrin (PE)-conjugated rat anti-CD31 (BD Biosciences, San Jose, CA,, and the DyLight 488 or 649 (Thermo, Rockford, IL, anti-mouse CD44 antibody MS44 [26] were used. For immunohistochemical staining, IM7 or MS44 was used as the first antibody. HRP-conjugated secondary antibody was obtained from BioSource (Camarillo, CA, The sections were counterstained with hematoxylin. Tumor hypoxia was measured 90 minutes after injection of 60 mg/kg pimonidazole hydrochloride into tumor-bearing mice. To detect the formation of pimonidazole adducts, tumor tissue sections were immunostained with anti-Hypoxiprobe-1 antibody (Hypoxiprobe, Inc. Burlington, MA, following the manufacturer's instructions. Stained sections were assessed under a microscope (CTR 5500, LEICA, Wetzlar, Germany,

Flow Cytometric Analysis

Single-cell suspensions from tumors were prepared using a standard protocol. Fluorescence-activated cell sorting (FACS) analysis was performed using a FACS Calibur (BD Biosciences), and cell sorting was done using a FACSAria (BD Biosciences) as described previously [17]. The antibodies used for flow cytometry were PE (Dojin Chemical Laboratories, Kumamoto, Japan, MS44, FITC-conjugated rat anti-lineage markers (a mixture of ter119, Gr-1, Mac-1, B220, CD4, and CD8), and CD31 antibodies (all purchased from BD Biosciences). For the PSF1-EGFPhigh population, the 5% most brightly fluorescing cells were sorted, and for the PSF1-EGFPlow population, the 5% least fluorescent, as previously reported [16].

CSC/Cancer Initiating Cell Culture and Sphere Formation

To exclude contamination by cells other than HT29 cells, E2-Crimson was stably transfected into HT29 cells. We used E2-Crimson positive and ALDH1-bright cells obtained from HT29 human colon cancer-bearing mice as CSCs. ALDH1 enzymatic activity in HT29 cells was analyzed using ALDEFLUOR kits (Stem Cell Technologies, Vancouver, BC, Canada, with ALDH substrate BODIPY-aminoacetaldehyde according to the manufacturer's instructions. As a negative control to confirm the specificity of ALDEFLUOR labeling, cells were treated with 1.6 mM diethylaminobenzaldehyde, a specific ALDH inhibitor (Supporting Information Fig. S1A). One million ALDH1-bright cells were cultured with or without 1 × 104 cell lined WT- or KO-CAFs, or freshly isolated WT- or KO-CAFs in six-well plates for 7 days. For the indirect coculture experiments, CAFs were placed in the upper chamber of six-well hanging cell culture inserts (polyethylene terephthalate membranes with 0.4 µm pores) (Millicell; Millipore). In addition, as described in the Introduction section [15], PSF1-EGFPhigh LLC cells were derived from LLC tumor-bearing mice as cancer initiating cells (CICs). Ten million sorted PSF1-EGFPhigh cells were cultured with 1 × 104 cell-lined WT- or KO-CAFs, or freshly isolated WT- or KO-CAFs in six-well plates for 9 days. The number of cells showing high EGFP intensity (above 5 × 102) was calculated among 104 total tumor cells by flow cytometry.

Cultured CSCs/CICs were sorted selectively and plated at a density of 5 × 103 cells in 6-cm low cell binding dishes (NUNC) for tumor sphere-forming assays. Cells were grown in DMEM or RPMI-1640 media containing 20 ng/mL epidermal growth factor (EGF, PeproTech, London, U.K.,, 10 ng/mL basic fibroblast growth factor (PeproTech), 10 mM HEPES, and B-27 supplement (Invitrogen, Gibco). For the LLC cells, we added N-2 supplement (Invitrogen, Gibco). Two weeks later, we counted the number of spheres under a microscope with a ×4 objective lens (IX70, Olympus, Tokyo, Japan, and presented the average number of spheres per field.

Coculture of CSC Cells and CAFs

LLC-PSF1p-EGFP cells were cocultured with or without CAFs in 10-cm culture dishes (1.5 × 105 of each per dish) in media with or without 100 ng/mL 5-fluorouracil (5-FU) for 3 days. Cultured cancer cells (EGFP-positive cells) were sorted selectively and RNA extracted.

Drug Treatment

The anti-vascular endothelial growth factor (VEGF) neutralizing monoclonal antibody bevacizumab was purchased from Roche (Basel, Switzerland, HT29 tumor-bearing mice were treated with i.v. bevacizumab or human control IgG (The Jackson ImmunoResearch Laboratory) at a dose of 5 mg/kg b.wt. [27] twice weekly for 6 weeks. 5-FU was obtained from Kyowa Hakko Kogyo (Tokyo, Japan, LLC tumor-bearing mice were treated with 5-FU or saline as a control intraperitoneally at a dose of 60 mg/kg b.wt. three times weekly for 2 weeks.

In Vitro Colony Formation

Isolated PSF1-EGFP-positive cells obtained from LLC-PSF1p-EGFP-bearing WT or CD44-KO mice were plated on 10-cm culture dishes (500 cells per dish) and cultured in media with or without 100 ng/mL 5-FU.

Quantitative Polymerase Chain Reaction

Total RNA was isolated using RNAeasy Kits (Qiagen, Valencia, CA, according to the manufacturer's instructions. RNA was reverse-transcribed using the ExScript RT Reagent Kit (Takara, Kyoto, Japan, Quantitative reverse-transcription polymerase chain reaction (qRT-PCR) was performed using SYBR Premix Ex Taq II (Takara) on an Mx3000 system (Stratagene, LA Jolla, CA, Levels of the specific amplified cDNAs were normalized to the level of glyceraldehyde-3-phosphate dehydrogenase housekeeping control cDNA. We used the following primer sets: 5′-CATCACCATCTTCCAGGAGCG-3′ and 5′-GAG GGGCCATCCACAGTCTTC-3 for GAPDH, and 5′-GACACTGATGC TTCTGAACTGA-3′ and 5′-GCAAAGTACGCCAAC AAGTAAAT-3′ for multidrug resistance one (Mdr1).

Apoptosis Assay

Cells were stained with allophycocyanin-labeled Annexin V (BD Biosciences) according to the manufacturer's instructions. EGFP-positive and Annexin V-positive cells were quantified using the FACS Calibur.

Statistical Analysis

Results are expressed as the mean ± SEM. Student's t test was used for statistical analysis. Differences were considered statistically significant when p < .05.


Isolation of Tumor Stromal Cell Populations from Tumor-Bearing Mice and Establishment of a CAF Cell Line

PDGFRβ is a functional receptor expressed on vessel-associated pericytes and fibroblasts [28]. To establish CAF cell lines, B16 mouse melanoma cells were subcutaneously inoculated into mice, and PDGFRβ-positive, hematopoietic lineage marker or CD31-negative, cells were sorted from the B16 tumor mass and immortalized by transfection of the SV40 gene as previously reported [26]. These isolated CAFs had a typical spindle-shaped morphology (Fig. 1A) and most expressed the mesenchymal markers PDGFRβ and integrin β1 (CD29), as well as the well-known CAF markers NG2 and αSMA at the protein level (Fig. 1B, 1C). To test whether established CAFs support tumor growth, we injected mice with tumor cells together with CAFs. Established CAFs were found to significantly enhance tumor growth of both B16 melanoma and LLC cells (Fig. 1D). We confirmed by histology that growth of tumor supported by CAFs was not caused by dominant growth of the CAFs themselves and that tumorigenic capacity of CAFs alone was not observed (data not shown).

Figure 1.

Established CAFs enhance tumor growth in vivo. (A): Morphology of CAFs established from B16 melanoma-bearing mice. (B): Fluorescence-activated cell sorting analysis of PDGFRβ and integrin β1 expression on CAFs. (C): Western blotting of NG2 and αSMA expression in CAFs. Mouse endothelial cell line bEnd.3 cells were used as a negative control. β-Actin was used for the internal control. (D): Support of tumor growth by CAFs. Gross appearance of tumors growing in mice injected with B16 cells (upper panels) and LLC cells (lower panels) with or without CAFs. Quantitative analyses of tumor size are shown in the right-hand panels. Data are means ± SEM. ****, p < .05; ***, p < .01 (n > 3). Abbreviations: CAFs, cancer-associated fibroblasts; NG2, neuron-glial antigen-2; PDGFRβ, platelet-derived growth factor receptor β; αSMA, α smooth muscle actin.

CAFs Maintain the Stemness of CSCs/CICs via Direct Interactions

We hypothesized that CAFs would be able to maintain the stemness of CSCs/CICs. To test this, we performed coculture assays of CSCs/CICs and CAFs. We isolated ALDH1-bright cancer cells from the HT29 tumor and confirmed that they formed tumor spheres (Supporting Information Fig. S1A, S1B), suggesting that these ALDH1-bright cells are CSCs, as previously reported [12-15]. First, ALDH1-bright HT29 cells were cultured with or without CAFs for 7 days. We then determined whether ALDH1 enzymatic activity was maintained by the CAFs. As shown in Figure 2A, 2B, the number of ALDH1-bright cells in cocultures with CAFs was higher than in HT29 mono-cultures. In addition to the number of ALDH1-bright cells, we found that the intensity of expression of ALDH1 by individual cells was higher in the coculture group. Next, to assess the self-renewal capacity of CSCs retained after coculture or mono-culture, we plated cells at clonal density and characterized tumor sphere formation. The number of tumor spheres per field was 1.33-fold higher in HT29 cells cocultured with CAFs relative to mono-cultures (Fig. 2B). Inhibition of direct cell to cell contact by means of cell culture inserts separating HT29 cells from CAFs resulted in abrogation of enhanced tumor-sphere formation, although the number of ALDH1-bright cells was not affected (Fig. 2A, 2B). Consistent with these results, conditioned media from CAF cultures supported the maintenance of ALDH1-bright cells but did not enhance tumor sphere formation (Fig. 2C, 2D). These results indicate that molecules secreted from CAFs can support phenotype but not stemness of CSCs, and that direct cell–cell contact between CAFs and CSCs is required for the maintenance of the latter.

Figure 2.

CAFs support the stemness of cancer stem/initiating cells via direct interactions. (A): Fluorescence-activated cell sorting (FACS) plots and percentages of ALDH1-bright cells in the HT29 cells cultured with (CAF+) or without (CAF−) CAFs. ALDH1-bright cells were also cultured with CAFs separated by culture inserts (CAF+/insert). (B): Number of tumor spheres generated by cultured HT29 as in (A). Data are means ± SEM. *, p < .001 (n = 20). (C): FACS plots and percentages of ALDH1-bright cells. ALDH1-bright cells were cultured under the following conditions: CAF−, CAF+, and CM-CAF. (D): Number of tumor spheres per field generated by HT29 cells cultured as in (C). Data show the mean ± SEM. *, p < .001 (n = 20). (E): EGFP intensity by FACS analysis of LLC-PSF1p-EGFP cells with or without CAFs. Cell culture insert (insert) was used for separation of CAFs and cancer cells in each condition. Data are means± SEM. *, p < .001 (n = 3). (F): Number of tumor spheres per field generated by LLC-PSF1p-EGFP cells cultured as in (E). Data are means ± SEM. *, p < .001 (n = 20). Abbreviations: ALDH, aldehyde dehyrogenase; CAFs, cancer-associated fibroblasts; CAF−, ALDH1-bright cells alone; CAF+, ALDH1-bright cells together with CAF; CM-CAF, ALDH1-bright cells together with conditioned media from CAF cultures; EGFP, enhanced green fluorescent protein; SSC, side scatter.

Next, we exploited the higher PSF1 promoter activity in LLC cells to detect CSCs/CICs as described above [16]. We confirmed that PSF1-EGFPhigh LLC cells were able to form more tumor spheres than PSF1-EGFPlow LLC cells in culture (Supporting Information Fig. S1C, S1D). EGFP intensity of LLC-PSF1p-EGFP cells cocultured with CAFs was significantly higher than mono-cultured LLC-PSF1p-EGFP cells (Fig. 2E) and they formed 2.65-fold greater numbers of spheres per field (Fig. 2F). Using cell culture inserts, it was found that coculture with CAFs had no effect on EGFP intensity and tumor sphere formation (Fig. 2E, 2F).

Strong CD44 Expression on CAFs in the Hypovascular Area of the Tumor Microenvironment

Because we had generated a monoclonal antibody specific for CD44 by immunizing with CAFs [26], we investigated the distribution of CD44-positive CAFs in the tumor microenvironment. As we reported previously [26], our anti-mouse CD44 monoclonal antibody MS44 recognizes mouse but not human CD44. Therefore, using MS44 for immunohistostaining in tumor tissues generated by injection of human cancer cells into mice allowed us to distinguish between stromal and tumor CD44 expression (Supporting Information Fig. S2). Using this approach, we analyzed CD44 expression in stromal tissues of HT29 tumors, focusing on the relationship between CD44 and hypoxia (which was a feature of most tumors caused by limited oxygen diffusion into the avascular areas of tumors [29]) because it had been reported that CSCs were occasionally found in hypoxic areas [30, 31], the so-called hypoxic niche, and in vascular areas, the so-called perivascular niche [32]. Determining the localization of CD44-positive CAFs, CD31-positive endothelial cells and hypoxia by pimonidazole staining revealed strongly CD44-positive stromal cells in hypoxic areas and only weak CD44-positive stromal cells in nonhypoxic vascular areas (Fig. 3A). This suggests that CD44-positive CAFs might support cancer cell survival in an avascular area.

Figure 3.

Strong stromal CD44 expression in the avascular area in the tumor microenvironment. Section from HT29 tumor was stained with DyLight 649-conjugated anti-CD44, clone MS44 (blue), and phycoerythrin-conjugated anti-CD31 (red) antibodies. The hypoxic region in the tumor was visualized by pimonidazole staining (green). Right-hand panels are a higher magnification of the area indicated by boxes in the middle bottom panel.

Stromal CD44 Expression Increases Under Hypovascularity in the Tumor Microenvironment

Next, to analyze whether the expression of stromal CD44 is altered in the hypovascular area in tumors, we treated the tumor-bearing mice with anti-VEGF neutralizing antibody. Treatment with anti-VEGF antibody inhibited HT29 tumor growth as previously reported [33] and this decreased the number of CD31-positive blood vessels in the tumor; however, the number of CD44-positive cells was not altered, although the strength of expression seemed to increase (Fig. 4A). To confirm this, the intensity of CD44 expression was quantified by flow cytometry. As shown in Figure 4B, CD44 expression was enhanced by treatment with anti-VEGF neutralizing antibody. To assess whether hypoxia regulates upregulation of CD44 in CAFs, CAFs were cultured under hypoxic condition. Unexpectedly, we found that CD44 expression was not induced but was reduced on our established CAFs under hypoxic conditions in vitro when compared with that under normoxic conditions. However, CD44 expression was induced under both hypoxic and hyponutritional conditions (Supporting Information Fig. S3). This suggests that CD44 expression on CAFs may be regulated with several complex conditions, that is, hypoxia, hyponutrition, in the hypovascular area of a tumor. Taken together, these results indicate that CD44 on CAFs may interact with cancer cells especially in locations without blood vessels.

Figure 4.

Upregulation of CD44 expression on stromal cells in tumors treated with angiogenesis inhibitors. (A): Immunofluorescent staining of tumor sections derived from tumors formed by HT29 cells inoculated into nude mice with or without anti-VEGF neutralizing antibody or control human IgG. Sections were stained with anti-CD31 antibody (upper panels) or anti-CD44 antibody (lower panels). (B): Cells from tumor tissue of HT29 tumor-bearing mice were stained with antihematopoietic lineage markers, CD31 and CD44 antibodies and analyzed by flow cytometry. The percentages of CD44-positive cells are shown in each panel and the negative threshold of CD44 in controls (no-treatment) is shown by a red line. Abbreviation: VEGF, vascular endothelial growth factor.

CD44 on CAFs Is Required for the Maintenance of the Stemness of CSCs/CICs

In initial experiments using cocultures and sphere formation assays we found that established CAFs could maintain the stemness of CSCs/CICs. To determine whether CD44 on CAFs is involved in this, we established CD44-negative CAFs (KO-CAFs) from a CD44-KO mouse. KO-CAFs retained characteristic spindle-shaped morphology and expressed mesenchymal or CAF markers (Supporting Information Fig. S4A–S4C) as in wild-type CD44-positive CAFs (WT-CAFs, Fig. 1A–1C). We performed genotyping of CD44 in WT-CAFs, KO-CAFs, and B16 cells and showed that KO-CAFs were certainly derived from CD44 KO mice but not B16 cells (Supporting Information Fig. S4D). Their growth kinetics were also unaffected (Supporting Information Fig. S4E). Using these KO-CAFs, we assessed the role of CD44 in the same assays as shown above in Figure 2. The number of ALDH1-bright cells in and spheres produced by HT29 cells after coculturing with KO-CAFs was less than in cocultures with WT-CAFs (Fig. 5A, 5B). Similar results were obtained when observing promoter activity of the PSF1 gene and sphere formation of LLC-PSF1p-EGFP cells after coculturing with KO-CAFs or WT-CAFs (Fig. 5C, 5D). To confirm that CD44 on CAFs is indeed responsible for these results, we performed CD44 rescue experiments in these cultures. The results showed that KO-CAFs stably transfected with mouse CD44 (KO+ CD44, Supporting Information Fig. S4F) regained the ability to increase CSCs/CICs and tumor sphere formation of both HT29 and LLC-PSF1p-EGFP cells (Fig. 5A–5D). Furthermore, we showed that CD44 is an effective molecule to maintain the stemness of CSCs/CICs by using small interfering RNA (siRNA)s against CD44 (Supporting Information Fig. S5). Additionally, using freshly sorted CAFs from WT mice or CD44-KO mice, we obtained similar results as that observed in WT-CAFs and KO-CAFs described above (Supporting Information Fig. S6). Taken together, these results suggest that CD44 expressed on CAFs is involved in the maintenance of the stemness of CSCs/CICs.

Figure 5.

CD44 expressed on CAFs is required to maintain the stemness of cancer stem cells/cancer initiating cells. (A): Fluorescence-activated cell sorting (FACS) analysis of ALDH1-bright cells from cultured HT29 in the presence or absence of CAFs. ALDH1-bright cells were cultured under the following conditions: CAF−, WT, KO, and KO+CD44. (B): Quantitative evaluation of the number of tumor spheres per field generated from HT29 cells cultured as indicated in (A). Data are means ± SEM. *, p < .001 (n = 20). (C): FACS analysis of EGFP intensity in LLC-PSF1p-EGFP cells cultured as in (A). Data are means ± SEM. *, p < .001; ****, p < .05 (n = 3). (D): Quantitative evaluation of the number of tumor spheres per field generated from LLC-PSF1p-EGFP cells cultured as indicated in (C). Data are means ± SEM. *, p < .001 (n = 20). Abbreviations: ALDH, aldehyde dehydrogenase; CAFs, cancer-activated fibroblasts; CAF−, ALDH1-bright cells alone; KO, ALDH1-bright cells with KO-CAF; KO+CD44, ALDH1-bright cells with KO-CAF transfected with mouse CD44; SSC, side scatter; WT, ALDH1-bright cells with WT-CAF.

CD44 Expressed on CAFs Is Involved in Cancer Drug Resistance

Next we tested the ability of CD44 expressed by stromal cells to maintain CSCs/CICs in vivo. LLC-PSF1p-EGFP cells were inoculated into WT mice or CD44-KO mice. Tumor growth was not significantly different in these animals (Fig. 6A), showing that host CD44 did not alter tumor growth in vivo. Moreover, PSF1 gene promoter activity in the tumors showed that the lack of CD44 did not affect peak EGFP intensity of LLC-PSF1p-EGFP cells (Fig. 6B). These results suggest that cancer cell proliferation was not affected by the absence of host CD44. Next, we focused on the qualities of mouse-derived LLC cells in terms of the stemness of CSCs/CICs. Interestingly, LLC cells derived from tumor tissue in CD44-KO mice had a lower capacity to form tumor spheres compared to LLC cells derived from WT mice (Fig. 6C). These results showing a disconnection between CSC activity and tumor growth are in accordance with a recent report showing that CSCs do not always generate a larger tumor compared with differentiated cancer progenitors differentiated from CSCs [10].

Figure 6.

Association of stromal CD44 with tumor drug resistance. (A): Tumor volume in LLC-PSF1p-EGFP-bearing WT or CD44-KO mice. There were no significant differences in tumor volume between these two groups (n = 6). (B): Fluorescence-activated cell sorting (FACS) analysis of EGFP intensity in cells derived from tumor as indicated in (A). (C): Quantitative evaluation of the number of tumor spheres generated by the same number of LLC-PSF1p-EGFP cells from tumor tissues in WT or CD44-KO mice. Data are means ± SEM. *, p < .001 (n = 20). (D): FACS analysis of EGFP intensity in LLC-PSF1p-EGFP cells from WT or CD44-KO mice treated with saline (control, purple) or 60 mg/kg of 5-FU (green). (E): Mdr1 mRNA expression level was analyzed by quantitative polymerase chain reaction (PCR). cDNA was prepared from LLC-PSF1p-EGFP cells cultured with or without WT-CAFs or KO-CAFs in the presence or the absence of 5-FU (100 mg/mL). (F): Quantitative evaluation of the number of cancer cell colonies. Cancer cells derived from tumor generated by LLC-PSF1p-EGFP cells in WT or CD44-KO mice were cultured with or without 100 ng/mL of 5-FU. Data are means ± SEM. *, p < .001 (n = 3). (G): Scheme of apoptosis assay. 1. Isolated CICs (PSF1-EGFPhigh cells obtained from LLC-PSF1p-EGFP-bearing WT mice) were plated in six-well plates with or without WT-CAFs or KO-CAFs [CICs ± CAFs (WT/KO)] and cultured in media containing 1% fetal bovine serum for 1 day (upper) or with cell culture inserts for 3 days (lower). 2. We changed media of CICs ± CAFs with or without 100 ng/mL of 5-FU and incubated for 2 days. 3. CICs in the insert were put in the well in which CICs ± CAFs had been cultured, and then cocultured for 3 days. (H): Quantification of apoptotic cells by Annexin V staining in EGFP-positive cells cultured as in (G). Data are means ± SEM. ****, p < .05 (n = 3). Abbreviations: CICs, cancer initiating cells; CAF, cancer-activated fibroblasts; 5-FU, 5-flurouracil; EGFP, enhanced green fluorescent protein; KO, knockout; Mdr1, multidrug resistance one; WT, wild type.

Additionally, we sought associations between host CD44 expression and cancer cell drug resistance. We inoculated LLC-PSF1p-EGFP cells into WT and CD44-KO mice and quantified EGFP intensity after treatment with the anti-cancer drug 5-FU. PSF1-EGFPhigh cells remained in the WT host but not in the CD44-KO mice after treatment with 5-FU (Fig. 6D). Recently, we reported that PSF1-EGFPhigh cells still present after 5-FU treatment generate secondary tumors from very small numbers of cells in serial transplantation analysis [19]. This suggests that CD44 on CAFs protects CSCs/CICs from cell death otherwise caused by anti-cancer drugs. In order to confirm this, expression levels of Mdr1 mRNA in LLC-PSF1p-EGFP cells, cultured with or without WT-CAFs or KO-CAFs in the presence or absence of 5-FU, were determined. As shown in Figure 6E, when cultured in the presence of 5-FU, the expression level of Mdr1 increased in cells under all conditions compared with cultures in the absence of 5-FU. Mdr1 gene expression was significantly increased in LLC-PSF1p-EGFP cells cocultured with WT-CAFs. We then determined whether CD44 on CAFs influenced drug resistance in LLC-PSF1p-EGFP cells. After inoculation of LLC-PSF1p-EGFP cells into WT or CD44-KO mice, PSF1-EGFPhigh cells (10% most bright) were sorted from tumors and cultured for 18 days with or without 100 ng/mL 5-FU. Significantly fewer colonies were formed by cells from CD44-KO mice than WT mice when cultured in the presence of 5-FU (Fig. 6F). This suggests that CAFs produce factors preventing CSC/CIC cell death and that CD44 expressed by the CAFs is involved in this process.

Finally, to test the above possibility, we conducted stepwise culture experiments. First, we sorted PSF1-EGFPhigh cells (10% most bright) from LLC-PSF1p-EGFP tumor tissue, cultured these cells mixed directly with WT-CAFs or KO-CAFs with 1% FCS-containing hyponutrient media for 1 day, and added 5-FU for 2 days. Subsequently, PSF1-EGFPhigh cells isolated from tumors were cultured for 3 days on cell culture inserts placed into the culture plates in which CAFs and CICs had been cocultured in the presence of 5-FU (Fig. 6G). As shown in Figure 6H, there were less apoptotic cells in cultures where WT-CAFs were used than in cultures where KO-CAFs were used. The number of apoptotic cells in the latter was the same as in cultures where CAFs were not initially used (Fig. 6H). These results suggest that CAFs produce and secrete factors having antiapoptotic effects on CICs and that CD44 on CAFs plays some role in the production of these factors.


Studies on CD44 have been focused on the roles of variant forms of this molecule in cancer cells, especially cancer stem cells [23]. In contrast, here we show that the standard form of CD44 expressed on CAFs has a role in supporting CSCs/CICs.

Although the origins of CAFs have been extensively explored and they are considered to be resident local fibroblasts, bone marrow-derived progenitor cells, transdifferential epithelial cells and others [6, 9, 34-38], their precise cellular origins and functional contributions to tumor growth still remain unclear. However, molecules produced from CAFs affecting tumor growth and maintenance are gradually being defined.

It has been reported that CAFs increasingly acquire two autocrine signaling loops, mediated by TGF-β and stromal cell-derived factor 1 (SDF-1) cytokines, which both act in autostimulatory and cross-communicating fashions in invasive human breast carcinomas [39]. In addition, in the earliest stages of inflammation-induced tumor development, the bone marrow undergoes remodeling, mediated in part by TGF-β, resulting in mesenchymal stem cell-derived CAFs which promote tumor progression through SDF-1 signaling [40]. We determined the expression of TGF-β and SDF-1 in our established WT-CAFs and KO-CAFs by qRT-PCR. As expected, the results showed that WT-CAFs expressed SDF-1 at higher levels than KO-CAFs and, moreover, that WT-CAFs cocultured with tumor cells expressed SDF-1 at higher levels than mono-cultured WT-CAFs (data not shown). However, contrary to expectations, WT-CAFs expressed lower levels of TGF-β than KO-CAFs. This suggests that there is a reciprocal relationship between the CD44 expression and TGF-β expression in CAFs.

Furthermore, according to the results shown in Figure 2, direct contact between CAFs and CSCs/CICs is necessary for maintaining stemness, at least in the case of our established CAF cell line. Our results suggest that CD44 expression influences the expression of many other molecules and is concerned with maintenance of CSCs/CICs. Further determination of the precise molecular mechanisms of how CD44 affects molecules associating with stemness is required.

Recently, inhibitors of angiogenesis have attracted much attention in cancer therapeutics, but these agents alone are not sufficient to kill all cancer cells [41]. We have discovered that abundant stromal cells expressing PDGFRβ are found in dormant tumors (unpublished data). In addition, we found that under these conditions PDGFRβ-positive CAFs formed tube-like structures similar to vessels. Therefore, it is possible that such CAFs function as pipes, similar to blood vessels, and supply oxygen and nutrients to tumor tissue after treatment with angiogenesis-disrupting agents. However, we failed to detect intravenously injected lectin or biotin by histochemical staining in such PDGFRβ-positive CAF tube-like structures, suggesting that there was no blood flow through them. Regardless, CD44-positive CAFs are detected in both vascularized and nonvascularized areas of tumors (Fig. 3) and seem to connect with CD31-positive endothelial cells. Therefore, it remains possible that blood vessels provide their intraluminal nutrient to cancer cells in avascular areas through intracellular exchange of molecules between endothelial cells and CD44-positive CAFs.

We found that CD44 expression in the tumor microenvironment increases after treatment with anti-VEGF neutralizing antibody (Fig. 4). Because we used the anti-mouse CD44-specific antibody, MS44, and detected mouse CD44, these CD44 positive cells were considered tumor stromal cells, mainly CAFs, but not human HT29 colon cancer cells. In CSCs/CICs, coculture with CD44-positive CAFs (WT-CAFs) influenced their ability to maintain stem-like characteristics including high expression of stem cell markers and self-renewing sphere formation via direct cell–cell interactions. This suggests that CD44 upregulated on CAFs might be involved in cancer cell survival and stemness even in the regions without blood vessels.

Taken together, our data suggest that the use of agents directed against these CAFs would enable the complete elimination of cancer cells by targeting CD44. However, CD44 is expressed on the membrane of many different cell types. Therefore, drugs would need to be delivered specifically to CD44-positive CAFs in the tumor microenvironment. This could represent a promising new candidate cancer treatment strategy.


We demonstrated the role of CD44 expressed on CAFs in tumor microenvironment. CD44 was abundantly expressed on CAFs in hypoxic and avascular areas in the tumor, and CD44 positivity in tumor tissues was increased after treatment with inhibitors of angiogenesis. Using cocultures and tumor sphere formation assays, CAFs from wild-type mice were found to sustain the stemness of CSC/CICs, while CD44-deficient CAFs did not. These suggest that CD44 on CAFs might interact with cancer cells to support cancer cell survival in hypovascular areas and contribute to the maintenance of CSC populations in the tumor microenvironment. Furthermore, CD44 was involved in malignant cancer cell drug resistance mechanisms. It is hoped that CD44 will became a promising new candidate for cancer treatment.


We thank Keisho Fukuhara and Noriko Fujimoto for assistance; members of the RIMD Core Instrumentation Facility, Kojiro Nakamura, Yuko Kabumoto, and Yukiko Uchikawa, for technical assistance with the FACSAria.

Author Contributions

Y.K.: conception and design, data analysis and interpretation, and manuscript writing; T.M.: data analysis and interpretation; N.T.: conception and design, financial support, administrative support, manuscript writing, and final approval of manuscript.

Disclosure of Potential Conflict of Interest

The authors indicate no potential conflict of interests.