The invasion process of cancer cells is associated with the generation of specific stroma, called “cancer-induced stroma”. The main constituents of cancer-induced stroma are inflammatory cells, including lymphocytes, granulocytes, macrophages, the endothelial cells of blood and lymph vessels, pericytes and fibroblasts. Inflammatory cells and endothelial cells are recruited into cancer stroma and involved in tumor immunity1, 2, 3, 4 and neoangiogenesis,5, 6 respectively. Fibroblasts, which are the major component of stroma, are also recruited and can convert into smooth muscle actin-positive fibroblasts, i.e., myofibroblasts or activated fibroblasts, and produce collagens and extracellular matrix proteins in response to several extracellular stimuli. The paracrine signaling interactions between cancer cells and associated fibroblasts play important roles in tumor formation and progression.7, 8, 9 Experimental evidence of fibroblast recruitment into cancer stroma comes from the demonstration that the tumor microenvironment preferentially promotes engraftment of i.v.-injected bone marrow–derived mesenchymal stem cells.10 In addition, when β-galactosidase-transduced human fibroblasts were injected i.p. into SCID mice with an ovarian cancer cell line, they preferentially localized within the cancer stroma but not within the normal tissue stroma.11 Bone marrow–derived fibroblasts contribute to the tumor stromal reaction. We have reported that stroma generated by invasive cancer cells consist of both bone marrow–derived and non-bone marrow–derived activated fibroblasts and that the bone marrow–derived activated fibroblasts were recruited into cancer-induced stroma at a later stage.12 Furthermore, Direkze et al.13 also described the bone marrow contribution to tumor-associated myofibroblasts and fibroblasts. However, the functional implications of bone marrow–derived fibroblasts for cancer growth remain unclear.
Stromal fibroblasts create a context that promotes tumor progression.14, 15 Furthermore, investigators have found evidence that the proliferative activity of stromal fibroblasts in cancer-induced stroma is closely linked to lymph node and distant organ metastasis16, 17 and that soluble factor secretion by stromal fibroblasts influences tumor progression.18, 19 Elucidation of the mechanism by which fibroblasts are recruited into cancer stroma could lead to new insights into not only the mechanisms of cancer progression but also strategies for cancer treatment.
In the present study, we addressed the efficiency and selectivity of fibroblast recruitment into experimental tumors. First, we analyzed whether fibroblast cell lines derived from different or the same organs could be equally recruited into cancer-induced stroma. We found that recruitment of fibroblasts into cancer stroma is not selective with respect to organ origin. Furthermore, we developed a novel method to obtain a fibroblast cell line with higher recruitment efficiency by repeated in vivo passages within cancer stroma. We also found that the biologic phenotype of recruited fibroblasts is distinct from that of nonrecruited fibroblasts.
CA IX, carbonic anhydrase IX; GFP, green fluorescent protein; hFB, human fibroblast; HRP, horseradish peroxidase; MFI, mean fluorescence intensity; MMP, matrix metalloproteinase; RT, reverse transcriptase; SCID, severe combined immunodeficiency.
Material and methods
Six-week-old female SCID mice (C.B-17 background) were purchased from Clea (Tokyo, Japan) and maintained at the National Cancer Center Research Institute East. All animals were maintained under specific pathogen-free, temperature-controlled environmental conditions throughout the study, in accordance with institutional guidelines. Written approval for all animal experiments (K03-011) was obtained from the local Animal Experiments Committee of the National Cancer Center Research Institute.
Cell lines and cell cultures
SV-40 transformed human fibroblast cell lines from bone marrow (KM101, 102, 103, 104 and 105) were originally established from an identical male patient, as previously described.20 These cell lines were cultured in RPMI-1640 medium with 10% heat-inactivated FBS and antibiotics (penicillin and streptomycin). SV-40 transformed fibroblast cell lines from lung (VA-13, IMR-90-SV and MRC-5 SV1 TG1) and skin (W-V) were purchased from Riken (Tsukuba, Japan). These cell lines were cultured in DMEM with 10% heat-inactivated FBS and antibiotics. The human pancreatic cancer cell line Capan-1 was purchased from the ATCC (Rockville, MD). Capan-1 was maintained in RPMI-1640 medium with 20% heat-inactivated FBS and antibiotics. All cells were maintained in a 5% CO2 incubator at 37°C.
Retroviral transfer of GFP and cell sorting
Recombinant retroviruses were produced by cotransfecting the pMX-GFP vector21 (kindly provided by Dr. T. Kitamura, University of Tokyo, Tokyo, Japan) and amphotropic helper virus DNA into 293T cells. Human fibroblast cell lines were infected with recombinant retroviruses by culturing in conditioned medium in the presence of polybrene. GFP-positive fibroblast cell lines were further sorted using a FACScaliber sorting system (Becton Dickinson, San Jose, CA).
Xenotransplantation of GFP-labeled human fibroblast cell lines and Capan-1
Capan-1 cells (5 × 106/animal) were injected into the peritoneal cavities of SCID mice. One hour later, 5 × 106 cells of a GFP-labeled fibroblast cell line were injected into each peritoneal cavity at a different injection site. As we did not know when the cellular interactions between human fibroblasts and Capan-1 cells, which are responsible for efficient recruitment into the cancer stroma, occur, we reinjected GFP-labeled fibroblasts on day 14, a week before tumor removal. Animals were killed on day 21 and the tumor masses and other tissues removed. Control mice received only a Capan-1 injection. Another experiment was performed as follows. Capan-1 cells (5 × 106/animal) were injected into the peritoneal cavities of SCID mice on day 0. Next, 5 × 106 cells of GFP-labeled KM104 cells were injected i.p. on day 0, 7 or 14 after the Capan-1 transplantation; animals were killed on day 21. Tumor volume was calculated by the formula volume = (width2 × length)/2. We measured both tumor volume and tumor wet weight in each sample and confirmed that tumor wet weight was proportional to tumor volume (1.6 ± 0.2 mg/mm3).
Immunohistochemical and immunofluorescence analysis
Xenografted cancer tissues were fixed in 10% formaldehyde for 1 day. Double immunolabeling was performed on sections using an indirect immunoperoxidase technique followed by an indirect alkaline phosphatase technique. Primary antibodies were a rabbit polyclonal anti-topoisomerase IIα (Novocastra, New Castle-upon-Tyne, UK) at 1:100 dilution and a rabbit polyclonal anti-GFP (Molecular Probes, Eugene, OR) at a 1:1,000 dilution. Tissue sections (5 μm) were treated using a microwave-based antigen retrieval technique with 10 mmol/l citrate buffer (pH 6.0) for 20 min at 90°C. Endogenous peroxidases were inactivated with 3% H2O2 in methanol. Sections were incubated for 1 hr with anti-topoisomerase IIα antibody, then with EnVisionTM+System-HRP (Dako, Glostrup, Denmark). Reacted products were stained with diaminobenzidine. Sections were incubated with anti-GFP antibody for 1 hr and then with EnVisionTM+System-AP (Dako). After washing in PBS, the product of the alkaline phosphatase reaction was revealed using 0.5% fuchsin solution.
Double immunofluorescence analysis was performed by sectioning specimens into 3-μm-thick sections and incubating them with the primary antibodies, including rabbit polyclonal anti-GFP-Alexa Fluor 488 (Molecular Probes) at a 1:500 dilution and a mouse monoclonal antihuman cytokeratin (AE1/3, Dako) at a 1:50 dilution. Alexa Fluor 546 antimouse IgG (Molecular Probes) was used as the secondary antibody. Sections were examined using an inverted microscope with an excitation wavelength of 488 nm for Alexa Fluor 488 and 568 nm for Alexa Fluor 546. After mounting, sections were examined using an LSM5 PASCAL confocal imaging system (Carl Zeiss, Oberkochen, Germany). Confocal images were stored as digital files and viewed using Photoshop (Adobe, Mountain View, CA). Surgically resected pancreatic cancer tissues were snap-frozen in liquid nitrogen and stored at −80°C until used. Tissue sections (5 μm) were fixed in 10% formaldehyde for 5 min, and endogenous peroxidases were inactivated with 3% H2O2 in methanol. Primary antibodies included a goat polyclonal anti-CA IX (Santa Cruz Biotechnology, Santa Cruz, CA) at a 1:100 dilution and a mouse monoclonal anti-cytokeratin 8 (Dako, Denmark) at a 1:50 dilution. Sections were incubated for 1 hr, followed by the standard ABC technique (Vector, Burlingame, CA).
Establishment of KM104 subpopulations with high recruitment efficiency into cancer stroma
GFP-labeled KM104 (KM104GFP) and Capan-1 cells were inoculated into the peritoneal cavities of SCID mice. After 21 days, tumors at parapancreatic sites were removed, minced and cultured in 10% FBS RPMI-1640. Under subconfluent conditions, adherent cells were harvested and GFP-positive cells at the same MFI were sorted using the FACScaliber system. After confirming that >90% of each cell line was GFP-positive, the collected GFP-positive KM104 cells were then once again inoculated with Capan-1 into the peritoneal cavities of SCID mice. This procedure was repeated 5 times, and the resulting subpopulation was called KM104GFP-5G.
Real-time quantitative PCR analysis
Quantitative PCR was performed using the LightCycler System (Roche Diagnostics, Indianapolis, IN). An aliquot of 4 μg of the extracted total RNA from the tumor mass was reverse-transcribed using ThermoScript and oligo(dT)20 as the primer (Invitrogen, La Jolla, CA). The obtained cDNA was diluted 1/10 with water, and 1 μl was used for amplification. PCR was performed with the LightCycler FastStart DNA SYBR Green Kit (Roche Diagnostics), according to the manufacturer's protocol. To control for the specificity of the amplification products, a melting curve analysis was performed. No amplification of nonspecific products was observed. In addition, PCR products were gel-separated to confirm a band of the expected size. The relative initial amount of a particular template in the cDNA mixture was extrapolated from a standard curve. The standards, composed of 5 serial dilutions of one of the cDNAs (ranging from 1 × 108 to 1 × 102 copies/μl), were run in parallel with the samples under identical PCR conditions and amplified using the same set of primers. Using the LightCycler software, the amplification curves of samples were plotted against these standard curves to generate gene-specific mRNA copy numbers. The relative amounts of the resulting products were normalized using the β-actin housekeeping gene. Primer sequences were as follows: β-actin 5′-TTGAAGGTAGTTTCGTGGAT-3′, 5′-GAAAATCTGGCACCACACCTT-3′; GFP 5′-AAGTTCATCTGCACCACCG-3′, 5′-TCCTTGAAGAAGATGGTGCG-3′.
Chemotaxis and chemoinvasion assays
The chemotaxis assay was performed using 24-well culture chambers (Becton Dickinson Labware, Bedford, MA) and a polycarbonate filter with an 8 μm pore size, while the chemoinvasion assay was performed using 24-well culture chambers and a growth factor–reduced, Matrigel-coated filter with a pore size of 8 μm (Becton Dickinson Labware). The lower chamber contained 0.6 ml of RPMI-1640 + 0.1% BSA with Capan-1-released chemoattractant(s) or RPMI-1640 + 0.1% BSA as a control. In the upper compartment, 2 × 104 cells/well were placed in triplicate wells and incubated for 6 hr (chemotaxis assay) or for 24 hr (chemoinvasion assay) at 37°C in a humidified incubator with 5% CO2. After incubation, the cells that had passed through the filter into the lower wells were stained with hematoxylin and counted under a microscope in 9 predetermined fields.
We used human H133A oligonucleotide probe arrays (Affymetrix, Santa Clara, CA) for analysis of mRNA expression levels corresponding to 22,284 transcripts. The procedures were conducted according to the supplier's protocols and are thus described briefly. Total RNA was extracted from cultured KM104GFP and KM104GFP-5G by Trizol (Life Technologies, Bethesda, MD) reagent. Every 1 mg of total RNA was used to generate a cRNA probe by T7 transcription. Fragmented cRNA (10 mg) was hybridized to the microarrays in 200 ml of a hybridization cocktail at 45°C for 16 hr in a rotisserie oven set at 60 rpm. The arrays were then washed with nonstringent wash buffer (6 × SSPE) at 25°C followed by stringent wash buffer [100 mM MES (pH6.7), 0.1 M NaCl and 0.01% Tween-20] at 50°C, stained with streptavidin phycoerythrin (Molecular Probes), washed again with 6 × SSPE, stained with biotinylated antistreptavidin IgG followed by a second staining with streptavidin phycoerythrin and washed a third time with 6 × SSPE. The arrays were scanned using the Gene Array scanner (Affymetrix) at 3 μm resolution, and the scanned image was quantitatively analyzed with Microarray Suite 5.0 software (Affymetrix). For normalizing the data to compare mRNA expression levels among samples, we unified 1,000 as an average of average difference (AD) scores corresponding to signal intensities of all probe sets in each sample. For statistical analysis to select genes, Microsoft Excel (Redmond, WA) was used. Each experiment was done in duplicate. Duplicate experiments were conducted completely independently of each other.
In vivo model for recruitment of human fibroblast cell lines into xenotransplanted cancer stroma
Nine human fibroblast cell lines from bone marrow, lung and skin were genetically modified by GFP retroviral transfection; and the GFP-positive cells were sorted. After confirming that >90% of each cell line was GFP-positive (hFBGFP), they were used in the following experiment (Fig. 1a). The human pancreatic cancer cell line Capan-1 was inoculated i.p. into SCID mice, and hFBGFP cells were administered according to the schedule outlined in Figure 1b. By day 21, all Capan-1 cells had formed tumor masses in the peritoneum at the injection site, parapancreas and hilum of the spleen and liver. Although tumor sizes varied among cases, the tumor in the parapancreas region was the largest in each animal. The peripheral area of the Capan-1 xenografts at the parapancreas site was characterized by papillary growth (Fig. 1c), while the central area exhibited a glandular formation surrounded by desmoplastic stroma (Fig. 1e). Immunofluorescence examination, enhanced using a GFP antibody, revealed the presence of hFBGFP within both the papillary stroma (Fig. 1d) and the desmoplastic stroma (Fig. 1f). We examined the presence of hFBGFP in the intestine, liver and spleen by immunohistochemical analysis but did not find any GFP-positive cells. Furthermore, none of the 9 fibroblast cell lines induced metastatic spread of Capan-1 into other organs (data not shown). Table I shows the recruitment efficiency of hFBGFP into the cancer stroma. Although hFBGFP originating from bone marrow did not affect tumor growth, 8 of 9 KM104GFP showed moderate (++) or high (+++) recruitment efficiency. On the contrary, when KM102GFP, 103GFP and 105GFP were injected with Capan-1, the number of cases with GFP-positive clusters was lower than when KM104GFP was injected. hFBGFP originating from lung exhibited variable recruitment efficiency. VA-13GFP promoted tumor growth by about 2-fold and showed the highest recruitment efficiency. However, IMR-90-SVGFP reduced the tumor volume by half, and its recruitment efficiency was low. The recruitment efficiencies of MRC-5 SV1 TG1GFP from lung and W-VGFP from skin were intermediate and low, respectively.
Table I. Recruitment of hFBGFP into Cancer Stroma
Capan-1 cells and GFP-labeled fibroblast cell lines were injected i.p. at a different site. The same GFP-labeled fibroblast cell lines were also injected i.p. 14 days after the initial transplantation. Control mice received only a Capan-1 injection.
Another experiment was performed as follows. Capan-1 cells were injected i.p. on day 0. GFP-labeled KM104 were injected i.p. on day 0, 7 or 14 after the Capan-1 transplantation.
The number of GFP+ clusters was determined on the tumor maximum cut surface. −, 0; +, 1–5; ++, 6–10; +++, ≥11.
Early-phase interactions between KM104GFP and Capan-1 are required for efficient recruitment into cancer stroma
To investigate when the cellular interactions between hFBGFP and Capan-1 that are responsible for efficient recruitment into the cancer stroma occur, we inoculated KM104GFP on day 0, 7 or 14 (Fig. 2). KM104GFP was used for this experiment since this cell line did not affect tumor volume, unlike VA-13GFP (Table I). As expected, the tumor volume was not influenced by this injection protocol. When KM104GFP was administered on day 0, 6 of 7 (85.7%) showed moderate (++) or high (+++) recruitment efficiency, which was almost the same as on both day 0 and day 14 injection (5/6, 83.3%; Table I). However, when KM104GFP was administered on day 14, the recruitment efficiency decreased and no case showed moderate or high recruitment efficiency. These results suggest that early cellular interactions between Capan-1 and KM104GFP are required for efficient recruitment into cancer stroma.
In vivo selection of a KM104GFP subpopulation with a higher recruitment efficiency in Capan-1-induced stroma
To establish a KM104GFP subpopulation with higher recruitment efficiency, we performed 5 cyclic in vivo selections of KM104GFP in Capan-1-induced stroma, as shown in Figure 3a (KM104GFP-5G). The population of GFP-positive cells in KM104GFP and KM104GFP-5G was 95.0% and 96.7%, respectively; and MFI of GFP in KM104GFP and KM104GFP-5G was 339 and 380, respectively (Fig. 3b). Morphologically, KM104GFP-5G exhibited a spindle shape that was similar to KM104GFP (Fig. 3c). Growth of KM104GFP and KM104GFP-5G cells in monolayer culture did not significantly differ, indicating that there is no critical difference in proliferative activity or other growth-determining cellular features of these fibroblast subclones (Fig. 3d). When Capan-1 was injected with KM104GFP-5G, numerous GFP-positive clusters were found within cancer stroma (Fig. 4a). In KM104GFP, 4/7 showed high (+++) and 2/7 showed moderate (++) recruitment efficiency, whereas all cases (7/7) showed high (+++) recruitment efficiency in KM104GFP-5G (Table II), indicating higher recruitment efficiency in KM104GFP-5G. Quantitative RT-PCR analysis revealed that the relative copy numbers of GFP/β-actin (10–2) in tumor masses injected with KM104GFP-5G and in those injected with KM104GFP were 115.8 and 13.8, respectively (Fig. 4b). We double immunostained the cells with GFP and topoisomerase IIα, whose expression is limited mostly to the S-to-G2/M phases of the cell cycle22 (Fig. 5a). The topoisomerase IIα-positive cell ratios for GFP-positive KM104GFP and KM104GFP-5G cells on day 10 xenografts were 37.0 ± 7.3% and 37.1 ± 5.4%, respectively, and those on day 21 xenografts were 27.4 ± 4.5% and 32.9 ± 3.7%, respectively; these differences were not significant (Fig. 5b). These results indicated that higher recruitment efficiency of KM104GFP-5G was not caused by increased proliferative activity.
Table II. Recruitment of KM104GFP and KM104GFP-5G into Cancer Stroma
Capan-1 cells were injected i.p. on day 0. After 1 hr, KM104GFP and KM104GFP-5G were injected i.p. at a different site. The number of GFP+ clusters was determined on the tumor maximum cut surface. −, 0; +, 1–5; ++, 5–10; +++, ≥11.
TV, tumor volume.
61 ± 25
65 ± 31
KM104GFP-5G exhibited higher chemotaxis and chemoinvasion activity
Fibroblast recruitment is thought to involve several sequential events, such as migration, proliferation and survival within the cancer microenvironment. We therefore examined whether KM104GFP and KM104GFP-5G exhibited any differences in chemotaxis or chemoinvasion activity. When KM104GFP and KM104GFP-5G were placed on transwell chambers, the number of migrating cells increased in response to Capan-1-released chemoattractant(s); however, KM104GFP-5G exhibited a 2.4-fold higher migratory capacity than KM104GFP (KM104GFP 28.1 ± 6.0/field vs. KM104GFP-5G 68.0 ± 1.6/field). When KM104GFP and KM104GFP-5G were placed on a Matrigel transwell chamber, the invasion capacity of KM104GFP-5G was 3.1-fold higher than that of KM104GFP (KM104GFP 4.1 ± 1.1/field vs. KM104GFP-5G 12.6 ± 4.6/field) without any chemoattractants and 4.3-fold higher (KM104GFP 22.3 ± 2.9/field vs. KM104GFP-5G 95.5 ± 10.2/field) in response to Capan-1-released chemoattractant(s) (Fig. 6).
Comparison of expression profiles between KM104GFP and KM104GFP-5G
To identify the genes differently expressed in KM104GFP-5G, genomewide screening for genes with different expression patterns in KM104GFP and KM104GFP-5G was performed using a microarray containing 22,284 probes. We performed duplicate microarray analysis and selected the genes with >3-fold up- or downregulation in both experiments. Eight genes with >3-fold upregulation and 6 genes with >3-fold downregulation in KM104GFP-5G were identified (Table III). The upregulated genes were keratin-8 and desmolakin (microfilaments and cytoskeleton), zinc finger protein 198 (transcription and translation), G protein–coupled receptor 56 and CA IX (membrane proteins), mothers against decapentaplegic homolog 2 (signal transduction), lumican (cell–matrix interactions) and endosulfine-α (others). The downregulated genes were protocadherin-17 (cell–matrix interactions), ribosomal protein large P2, CCAAT/enhancer binding protein δ and zinc finger protein-36 (transcription and translation) as well as transducin (beta)-like 20 and hypothetical protein FLJ20375 (others). Significant differences in the expression of genes directly involved in cell migration and/or invasion (MMP family) were not found in KM104GFP-5G.
Table III. cDNA Microarray Search for Genes that are >3-Fold Differentially Expressed in KM104GFP-5G
Genes upregulated in KM104GFP-5G
Zing finger protein 198
G protein–coupled receptor 56
Mothers against decapentaplegic homolog 2
Genes downregulated in KM104GFP-5G
Transducin β-like 20
Ribosomal protein, large P2
CCAAT/enhancer binding protein δ
Zinc finger protein 36
Hypothetical protein FLJ20375
Immunohistochemical staining of CA IX and keratin-8 in fibroblasts in pancreatic ductal cancers
To verify whether expression of the genes identified in microarray experiments could be found in the recruited fibroblasts within cancer stroma, we selected 2 genes (CA IX and keratin-8). The results obtained by microarray were confirmed by real-time quantitative PCR. Relative amounts of transcription for CA IX and keratin-8 in 9 fibroblast cell lines and KM104GFP-5G are summarized in Table IV. Transcripts for CA IX and keratin-8 in KM104GFP-5G were 5.6-fold and 1.4-fold higher, respectively, than in KM104GFP. Then, their protein expression was immunohistochemically analyzed in xenografts and surgically resected pancreatic cancers. In the xenografts, KM104GFP recruited into Capan-1-induced stroma expressed CA IX (Fig. 7a,b) but not keratin-8 (data not shown). Overexpression of CA IX was observed in fibroblasts within pancreatic cancer stroma, with desmoplastic reaction in 6 of 8 cases (Fig. 7c). On the contrary, its expression was weak or under the detection level in fibroblasts within noncancerous pancreatic tissue (Fig. 7d). Overexpression of keratin-8 in fibroblasts within cancer stroma was observed in 2 of 8 cases (Fig. 7e), whereas its expression in fibroblasts was under the detection level in noncancerous pancreatic tissue (Fig. 7f).
Table IV. Expression Profiles for CA IX and Keratin-8 in hFBGFP
Total RNA was extracted from 9 GFP-labeled fibroblast cell lines and KM104GFP-5G, and quantitative RT-PCR was performed. The relative amounts of the resulting products were normalized using β-actin.
MRC-5 SV1 TG1
During cancer invasion, fibroblasts as well as inflammatory cells are recruited into cancer-induced stroma in response to chemotactic signals produced by the cancer cells. One of the first demonstrations of recruitment of genetically modified endothelial cells showed that rat brain endothelial cells can stably engraft into glioma-associated vessels.23 Recently reported evidence has shown that exogenously administered human endothelial cells and fibroblasts are selectively recruited into the microvasculature and stromal compartment, respectively, of human tumors growing i.p. in SCID mice.11 We also demonstrated in the present study that GFP-labeled human fibroblast cell lines were incorporated into cancer stroma. Furthermore, KM104GFP (from bone marrow) and VA-13GFP (from lung) were recruited more efficiently than the other cell lines. These results indicate that the recruitment of fibroblasts into cancer stroma is regulated by the properties of the fibroblasts themselves, which appears to agree with the results of previous in vitro studies showing that fibroblasts within cancer stroma and normal tissue are phenotypically and functionally distinct.24, 25, 26, 27
Interestingly, although KM104GFP was efficiently recruited, it did not influence tumor growth. However, VA-13GFP was efficiently recruited and promoted tumor growth. The integration of fibroblasts in the tumor mass would not contribute to the increase of tumor volume since the number of recruited fibroblasts was <5% in the resected tumor by microscopic analysis. In addition, IMR-90-SVGFP reduced tumor growth significantly, while W-VGFP, which showed similar recruitment efficiency as IMR-90-SVGFP, promoted tumor growth. Therefore, the biologic characteristics of fibroblasts could influence not only the recruitment efficiency of fibroblasts themselves but the growth of cancer cells. We are now investigating the mechanism by which recruited VA-13 promotes xenograft volume. We used KM104GFP, which exhibited high recruitment efficiency but did not affect tumor growth, in the following studies.
When KM104GFP cells were administered on day 14, the recruitment efficiency was obviously lower than when they were administered on day 0. We previously reported that most myofibroblasts within cancer stroma at the early phase of cancer progression were of residual tissue origin, whereas at the late phase, 40% of myofibroblasts were of bone marrow origin.12 These findings imply that the cancer microenvironment at the late phase provides an optimal condition for the recruitment of fibroblasts from the peripheral blood, rather than from the local connective tissue around the cancer nests. As tumor microenvironments of early and late phases might recruit different kinds of fibroblast, we speculate that early phase, but not late phase, Capan-1-induced microenvironment was suitable for the recruitment of KM104GFP.
Fibroblasts within cancer stroma are biologically heterogeneous, and this heterogeneity has many ramifications for studies on tumor biology.28, 29, 30, 31, 32, 33, 34 To select and maintain fibroblast variants with a higher recruitment potential from a heterogeneous population, we performed in vivo selection35 using the KM104GFP cell line. After 5 cyclic in vivo selections within cancer stroma, we obtained a subpopulation with an 8.4-fold higher recruitment potential (KM104GFP-5G). We investigated whether the higher recruitment of KM104GFP-5G in vivo is functionally linked with its migration and invasion capacity in vitro. Although both KM104GFP and KM104GFP-5G exhibited migratory potential in response to Capan-1-released chemoattractant(s), KM104GFP-5G had higher migratory activity. Most interestingly, when KM104GFP and KM104GFP-5G were subjected to a Matrigel invasion assay, KM104GFP-5G also exhibited a higher degree of invasiveness. Several reports have described fibroblasts from pathogenic fibrotic lesions, including cancer tissue, exhibiting “transformed” behavior.15, 36, 37, 38, 39 Fibroblast-like synoviocytes derived from rheumatoid arthritis joints were highly invasive and have functional implications for the matrix degradation of cartilage, whereas cells from normal tissues were not invasive.38 Based on these findings, one may postulate that fibroblasts within cancer stroma have a migratory and invasive capacity similar to that of cancer cells and associate with cancer cells by transmitting reciprocal signals. This hypothesis is also supported by the fact that exogenous KM104GFP administered on day 0 was found in both central and peripheral areas of the cancer tissue on day 21 (Fig. 1c–f).
Considering the results obtained from the oligonucleotide microarray method, we have focused on CA IX and keratin-8, which were significantly upregulated in KM104GFP-5G. Their protein expression was actually upregulated in recruited fibroblasts within cancer stroma of invasive pancreatic cancer cases. CA IX is a novel member of the carbonic anhydrase family that codes for a transmembrane glycoprotein with an extracellular catalytic domain; its expression is strongly induced by hypoxia.40, 41 Colpaert et al.42 reported that CA IX expression was found in not only breast cancer cells but also fibroblasts within the cancer stroma and that expression in fibroblasts was associated with poor prognosis. Keratin-8, an intermediate filament, is mainly expressed in epithelial cells but is also found in human decidual stromal cells, embryonal mesenchymal cells and SV-40 transformed fibroblasts in vitro.43, 44 Moreover, forced expression of keratins 8 and 18 in mouse L cells augmented cell migration and invasion.45 Nakagawa et al.46 analyzed molecular expression profiles of cancer-associated fibroblasts in colon cancer metastasis and indicated that these fibroblasts form a favorable microenvironment for cancer cells. According to their results, genes upregulated in cancer-associated fibroblasts compared to skin fibroblasts included keratin-18, which is not found in normal stromal cells. Taking these observation and our results into consideration, keratin upregulation in fibroblasts may be a key phenomenon for the recruitment into cancer-induced stroma. Further research should address the precise role of these genes with regard to fibroblast recruitment in our model using gene overexpression or silencing methods, although the possible effects of posttranscriptional modifications must be kept in mind.
By targeting the tumor microenvironment, treatment effectiveness could be increased.47, 48 Much work on the biologic mechanisms of fibroblast recruitment is needed, but the present results suggest that fibroblasts could be used as a biologic tracer of cancer cells and could act as an efficient drug delivery system to prevent or slow the local growth of cancer cells.
We are grateful to Dr. T. Kitamura (Institute of Medical Science, University of Tokyo, Tokyo, Japan) for providing the retroviral plasmid pMX-GFP. We also thank Ms. C. Okumura and Ms. Y. Okuhara for technical support as well as Ms. M. Suzaki for preparing the manuscript.