Macrophages may promote cancer growth via a GM-CSF/HB-EGF paracrine loop that is enhanced by CXCL12

The blood and histological samples used in our study were in compliance with Institutional Review Board regulations.

Highly purified human mononuclear phagocytes and neutrophils were isolated from the buffy coats [21] of blood samples from healthy volunteers. HeLa (human cervical carcinoma), DLD-1 (human colon adenocarcinoma) and Balb/c 3T3 (Swiss mouse embryo) cell lines (purchased from ATCC, Manassas, VA) and HUVEC (human umbilical vein endothelial cells, purchased from Cambrex, Walkersville, VA) were also used. Non-adherent and adherent cells were grown in RPMI-1640 medium and DMEM or TC199 + 10% FCS (complete medium; Invitrogen, Carlsbad, CA), respectively. Cells were treated with 200 ng/mL CXCL12 (Peprotech, London, UK) or 25 ng/mL GM-CSF (Genetics Institute, Boston, MA) or 25 ng/mL HB-EGF or 100 μg/mL anti-HB-EGF or 100 μg/mL anti-GM-CSF neutralising monoclonal antibody (mAb) (R&D Systems, Minneapolis, MN) or isotypic control immunoglobulins. After growing in cultures for the appropriate times in different conditions, the cells were either lysed for total RNA extraction or used for functional assays. In some experiments, the conditioned medium was replaced with fresh medium after 24 hours of stimulation and the cells were then maintained in culture for up to 48 hours. Cell-free supernatants (SN) were stored at -80°C.

Histological samples were obtained by hepatic lobectomy to excise metastatic nodules derived from colon cancer. After surgical excision, samples were put in buffered formalin, treated in an automated processor and embedded in paraffin. Four micrometre-thick slices were cut from paraffin blocks onto adhesive-coated slides. Cytological samples were obtained by allowing cells to grow on the slides. Antibodies (Ab) used included the following: CD163 (clone 10D6, 1/200; Novocastra, Newcastle-upon-Tyne, UK), CXCL10 and CXCR4 (both rabbit polyclonal, 1/500 and 1/100, respectively; Abcam, Cambridge, UK), CXCL12 (clone 7918, 1/100), GM-CSF (clone 3209, 1/100), HER1 (1/100), HER4 (1/100), and HB-EGF (clone 125923, 1/200) (all purchased from R&D Systems). Antigen retrieval was performed for all antibodies in a hot bath for 30 minutes at pH 6 (except for GM-CSF retrieval, which was performed at pH 8). For GM-CSF and HB-EGF no H₂O₂ blocking was performed. As controls, sequential sections or cytological slides were incubated with the Ab diluent and indifferent isotypic Ab. All procedures were performed on an automated stainer (Bond, Vision Biosystems, Melbourne, AU) using a polymer detection system (NovoLink, Novocastra).

Following stimulation for 20 minutes or 24 hours with CXCL12, 1 × 10⁶ cells were incubated with a goat anti-human HB-EGF IgG polyclonal Ab (Oncogene, San Diego, CA) or an IgG control Ab. After washing, 5 μL of phycoerythrin-conjugated swine anti-goat IgG (Caltag Laboratories, Burlingame, CA) was added. The cells were then analysed on a FACSCalibur (Becton-Dickinson, Mountain View, CA) using the FlowJo 8.8.2 program (Tree Star, Ashland, OR).

Cellular release of HB-EGF was evaluated by measuring its effect on HER1 phosphorylation at tyrosine 1068 (Y1068) using HeLa and DLD-1 as target cells. Approximately 3 × 10⁶ effector cells (mononuclear phagocytes and neutrophils) were seeded in the upper chamber of a 0.4 μm transwell co-culture system (Costar, Cambridge, MA) and were stimulated with CXCL12 for 20 minutes. In a separate experiment, 3 × 10⁶ mononuclear phagocytes and neutrophils were cultured in the presence or absence of CXCL12 for 24 hours and their conditioned medium was added to the target cells for 20 minutes. HeLa cells were also stimulated for 20 minutes with recombinant human HB-EGF. Subconfluent HeLa cells that had been starved for 24 hours were incubated in the presence or absence of the anti-HB-EGF neutralising Ab (R&D Systems). Target cells were harvested for protein extraction.

Cell pellets were lysed for 30 minutes in 1 mL ice-cold cell extraction buffer (Biosource, Camarillo, CA), which was supplemented with a protease inhibitor cocktail (Sigma, St. Louis, MO) and 1 mM PMSF (Sigma). After centrifugation at 13,000 rpm for 10 minutes at 4°C, aliquots of the SN were stored at -80°C.

Approximately 1 × 10⁷ HeLa cells, which were either untreated or treated with HB-EGF for 20 minutes, were harvested for HER1 immunoprecipitation, reduction, alkylation and tryptic digestion. Ten microliters of the peptide mixture was analysed by means of an LCMS/MS system [22]. The data handling for the identification of phosphorylated residues was performed according to Guo et al. [23]. HB-EGF-induced HER1 phosphorylation was evaluated at Y992, Y1045, Y1068, Y1086, S1142, Y1148, Y1173, and expressed as phosphorylation ratio (phosphorylation after stimulus/basal phosphorylation).

1. HER1 pY1068 and ERK1/2 pTpY185/187. Total HER1 versus HER1 pY1068 and total ERK1/2 versus ERK1/2 pTpY-185/187 were measured in cell protein extracts using a commercially available ELISA (Biosource). The results were calculated as the ratio of phosphorylated molecules to total molecules and expressed as the percentage of phosphorylation. 2. HB-EGF and GM-CSF. The cell-free culture SN was assayed for soluble HB-EGF protein using a specific ELISA developed in our laboratory [20] and for GM-CSF using a commercial ELISA (R&D Systems). Only SN previously purified from HB-EGF or GM-CSF by specific crosslinking (Invitrogen) were subsequently used as agonistic SN to induce HB-EGF or GM-CSF, respectively, in cell stimulation experiments followed by ELISA tests. 3. Apoptosis. 72-hour internucleosomal DNA fragmentation as an index of apoptosis was evaluated using a cell death detection ELISA (Roche Diagnostics, Mannheim, DE) and expressed as the mean ± SD of triplicate determinations of the enrichment factor (absorbance of unstimulated cells/absorbance of stimulated cells).

Total RNA was extracted from the indicated cells. Northern blot analysis (with 10 μg RNA/lane) was performed as described [19]. Filters were hybridised with ³²P-labelled cDNA probes complementary to the HB-EGF RNA sequence, which were generated in our laboratory or with a ³²P-labelled plasmid containing a cDNA probe that detected the β-actin RNA sequence.

Total cellular RNA was purified by the QIAamp RNA Blood mini kit (Qiagen, Hilden, DE) + DNase digestion (Qiagen). Templates were generated from 2 μg of RNA with the Superscript III First-Strand Synthesis System kit (Invitrogen). cDNA for GM-CSF was amplified using the following primers (Invitrogen): 5'-CTCAGAAATGTTTGACCTCCAG-3' (sense) and 5'-TGACAAGCAGAAAGTCCTTCAG-3' (antisense) for a 221 bp amplicon. The GAPDH internal control primers (SuperArray Bioscience, Frederick, MD) were used to co-amplify a specific fragment that was 436 bp long. After 30 cycles at 95°C for 15", 55°C for 30" and 72°C for 30" in a Veriti thermal cycler (Applied Biosystems, Foster City, CA) using the AmpliTaq Gold PCR master mix (Applied Biosystems), the reactions were analysed by agarose gel electrophoresis and recorded using a VersaDoc imaging system (Bio-Rad, Hercules, CA).

Hela, DLD-1, HUVEC and Balb/c 3T3 cells were seeded in 96-well plates in complete medium and allowed to adhere for 24 hours. They were then incubated in the absence or presence of the anti-HB-EGF neutralizing Ab (R&D Systems) and were stimulated with 25 ng/mL HB-EGF. In selected experiments, HeLa and DLD-1 were deprived of serum and cultured in the absence or presence of HB-EGF. Cell proliferation was measured by MTT incorporation [19] at 24, 48 and 72 hours and expressed as the proliferation index. Apoptosis was measurerd as the internucleosomal DNA fragmentation enrichment factor at 72 hours (see ELISA).

To silence the expression of GM-CSF, HeLa and DLD-1 cells were seeded on glass slides (10⁵ cells/mL), stimulated with 200 ng/mL CXCL12 and/or 25 ng/mL HB-EGF or SN from GM-CSF-stimulated mononuclear phagocytes and transfected 48 hours using HiPerFect Transfection Reagent and siRNA (2 nM and 5 nM respectively, GM-CSF siRNA no. SI03037272 and AllStars negative control siRNA - Alexa Fluor 488; Qiagen). Transfection efficiency (> 95%) was determined by flow cytometry on trypsinized cells. Effective silencing of GM-CSF was determined by testing the induction and release of 1. GM-CSF in cancer cells (using immunohistochemistry and ELISA, after a 24-hour stimulation with HB-EGF or mononuclear phagocytes conditioned medium) and 2. HB-EGF in mononuclear phagocytes (using flow cytometry and ELISA, after a 24-hour stimulation with silenced cancer cell SN).

Student's t-test, Mann-Whitney U test and Kruskall-Wallis ANOVA by ranks were used. Differences were considered significant for p values < 0.05.

We analysed the histological pattern of surgical samples from 15 patients aged 60 to 79 who underwent hepatic lobectomy in order to excise metastatic colon cancer nodules. In Figure 1, we show the representative histology from a 76-year-old patient who had such a procedure. Serial preparations of a subglissonian metastatic nodule were stained by immunohistochemistry for CD68, CXCL10, CD163, CXCR4, CXCL12, GM-CSF, HER1, HER4 and HB-EGF. Macrophages, which were found to infiltrate the metastatic region often creating a bridge between perivascular zones and metastases, were intensely positive for CD68, CXCR4, GM-CSF and HB-EGF; some were CXCL12-positive. Of interest, macrophages stained positive for both CXCL10 (M1-marker) and CD163 (M2-marker). Though we could not perform a double staining, the distribution of the expression of CXCL10 and CD163 suggests a cellular co-expression rather than distinct populations of cells. In any case, the macrophages were not definitely polarized towards a typical M2 pattern but instead showed a mixed M1/M2 pattern. The cancer cells were positive for CXCR4, CXCL12, HER1, HER4 (a chemotactic receptor that responds to HB-EGF), and GM-CSF. The cellular distribution of ligands and receptors suggested specific interplays that were tested in the following experiments performed on the HeLa and DLD-1 cancer cell lines, which express the same pattern of molecules, and on human mononuclear phagocytes ex vivo.

Under basal conditions, mononuclear phagocytes express HB-EGF (Figure 2A, B). When we stimulated mononuclear phagocytes with 200 ng/mL CXCL12, the membrane density of HB-EGF was initially reduced (at 20 minutes), but was increased after 24 hours (Figure 2A). After a 2-hour stimulation, constitutive HB-EGF mRNA transcripts were up-regulated in mononuclear phagocytes (Figure 2B), and this effect increased up to 24 hours (data not shown). This induction of HB-EGF was cell type specific. For instance, we show in Figure 2B that neutrophils did not express HB-EGF transcripts either basally or after CXCL12 stimulation. A 1-hour stimulation of mononuclear phagocytes with CXCL12 determined an increase in the HB-EGF concentration in the culture supernatants (Figure 2C). Thus, CXCL12-dependent signals induce mononuclear phagocytes to release HB-EGF from the cell membrane, increase the amounts of HB-EGF transcripts at 2 hours, and upregulate HB-EGF synthesis, leading to an increase in membrane-bound HB-EGF at 24 hours.

To test if the stimulation of mononuclear phagocytes with CXCL12 could induce HER1 transactivation in bystander cells via HB-EGF shedding, we performed transwell co-cultures, in which we analysed HER1 phosphorylation at tyrosine 1068. Tyrosine 1068, a major site of autophosphorylation that is associated with the activation of Ras, MEK and ERK1/2 [24] was chosen after performing mass spectrometry analysis of ligand-dependent HER1 phosphorylation in HeLa cells. Mass spectrometry confirmed that 25 ng/mL HB-EGF induced a phosphorylation pattern different from that induced by other HER1 ligands [23, 24] and that Y1068 phosphorylation was induced by HB-EGF in either HeLa or DLD-1 cells (Figure 3A). Moreover, HB-EGF-dependent phosphorylation was coupled to phosphorylation of ERK1/2 at threonine 185 and tyrosine 187 (Figure 3B), as expected [24].

To determine if CXCL12 induces the transactivation of HER1, we performed the transwell experiments depicted in Figure 4. Mononuclear phagocytes (and neutrophils as negative control) in the upper chamber were stimulated with 200 ng/mL CXCL12 and HER1-positive HeLa, DLD-1, Balb/c 3T3 cells and HUVEC were used in the lower chamber as target cells. After 20 minutes, the cells were analysed to assess HER1 phosphorylation at tyrosine 1068. Figure 4A depicts several negative controls. The direct stimulation of HeLa, DLD-1, Balb/c 3T3 cells and HUVEC with CXCL12 did not induce HER1 phosphorylation. Unstimulated mononuclear phagocytes did not induce HER1 phosphorylation in the target cells. Neutrophils, which do not express HB-EGF [20], were treated with CXCL12 and this treatment did not lead to phosphorylation of HER1 at tyrosine 1068 in the target cells. In contrast, as depicted in Figure 4B, treatment of mononuclear phagocytes with CXCL12 led to the strong phosphorylation of Y1068 in all of the target cells. Pre-treatment with 100 μg/mL of neutralising anti-HB-EGF, but not its corresponding controls, inhibited the transactivation of HER1. Finally, supernatants from CXCL12-stimulated neutrophils, which did not produce HB-EGF, were not effective (SN1, Figure 5A, B, C). Mononuclear phagocytes-derived supernatants (SN2, Figure 5A) contained factors that led to HER1 phosphorylation in, and proliferation of, HeLa and DLD-1 cells (Figure 5B, C) as well as HUVEC and Balb/c 3T3 cell proliferation (SN2, Figure 6A). Blockade of HB-EGF with the neutralising Ab abolished the phosphorylation and the proliferation of the cells (Figures 4B; 5B, C; 6A). However, this effect did not occur when using indifferent isotypic immunoglobulins. Thus, CXCL12 induced the release of functional HB-EGF from mononuclear phagocytes, transactivation of HER1 and proliferation of cancer cells (HeLa and DLD-1), fibroblasts (Balb/c 3T3 cells) and endothelial cells (HUVEC). This occurred both in transwell co-cultures and after adding conditioned medium (from cultures of mononuclear phagocytes stimulated with CXCL12) to the target cells. Moreover, the metastatic colon cancer cells stained positive for HER4 (Figure 1), through which HB-EGF exerts powerful chemotactic activity [19]. Thus, HB-EGF can induce cancer cell chemotaxis and proliferation as well as microenvironment-targeted angiogenic signals. Finally, Figure 6B shows that HB-EGF conferred upon HeLa and DLD-1 cells both proliferative and anti-apoptotic signals; these latter signals clearly emerged under starvation conditions, as indicated by the statistically significant reduction in mono/oligonucleosomes released into the cytoplasm.

In addition, when HeLa and DLD-1 cancer cells were stimulated with 200 ng/mL CXCL12 and/or 25 ng/mL HB-EGF, GM-CSF proteins were detected by immunocytochemistry after 24 hours and new GM-CSF transcripts (as assessed by RT-PCR) appeared after 2 hours (Figure 7A, B). Conditioned medium obtained from cancer cells contained GM-CSF (Figure 8A) and induced HB-EGF expression in, and release from, mononuclear phagocytes (Figures 7C; 8B). Inhibitory anti-GM-CSF mAbs significantly reduced the production of HB-EGF (Figure 8B). Thus, CXCL12 and HB-EGF induced GM-CSF expression in HeLa and DLD-1 cancer cells.

As described above, CXCL12 was shown to prompt mononuclear phagocytes and cancer cells to release HB-EGF and GM-CSF, respectively. On the other hand, we have previous evidence showing that GM-CSF is a strong inducer of HB-EGF expression in mononuclear phagocytes [19, 20]. If HB-EGF released by mononuclear phagocytes can trigger the production of GM-CSF in cancer cells, a possible GM-CSF/HB-EGF paracrine loop may exist that is initially activated by CXCL12. Thus, we tested (i) HeLa and DLD-1 cancer cells for the production of GM-CSF upon HB-EGF stimulation and (ii) mononuclear phagocytes for the production of HB-EGF upon GM-CSF stimulation. This choice was based on the known differential receptor expression in mononuclear phagocytes, as opposed to cancer cells, which are usually negative for the GM-CSF receptor. Figure 7 depicts the experiments suggesting that a paracrine loop exists between Mø and HeLa or DLD-1 cancer cells. When these cancer cells were stimulated with 200 ng/mL CXCL12 and/or 25 ng/mL HB-EGF, they produced and released GM-CSF (Figures 7A, B; 8A). When mononuclear phagocytes were stimulated with CXCL12 and/or 25 ng/mL GM-CSF, they produced and released HB-EGF (Figures 2; 7B, C, D; 8B). HB-EGF mRNA transcripts and membrane protein levels were increased after 2 hours (Figures 2B; 7B) and after 24 hours of stimulation (Figures 2A, C; 7C, D; 8B). These results were reproduced by the addition of conditioned medium from mononuclear phagocytes to cancer cells and vice versa (Figures 7A, C; 8). Inhibitory anti-HB-EGF and anti-GM-CSF Abs significantly reduced the production of GM-CSF by cancer cells (Figure 8A) and HB-EGF by mononuclear phagocytes (Figure 8B), respectively. Thus, a paracrine loop may exist between mononuclear phagocytes and HeLa or DLD-1 cancer cells, as exemplified in Figure 7E. CXCL12 seems to act as an additive or synergistic agent in this positive feedback loop. The expression of these molecules as found in patient biopsy samples (Figure 1) is in line with and corroborates our findings in vitro.

To further test the GM-CSF/HB-EGF loop between cancer cells and macrophages, we established a siRNA sequence targeted against GM-CSF. This siRNA efficiently blocked GM-CSF induction and release in HeLa and DLD-1 cells in response to both HB-EGF (Figure 9A) and conditioned medium from mononuclear phagocytes (data not shown). Similarly, inhibitory anti-HB-EGF Abs significantly reduced the production of GM-CSF by cancer cells (Figure 8A). Therefore, HB-EGF genuinely and specifically induced GM-CSF expression in these cells. The supernatant from silenced cancer cells lost the ability to upregulate HB-EGF in mononuclear phagocytes (Figure 9B) and inhibitory anti-GM-CSF Abs blocked the induction of HB-EGF (Figure 8B), showing that the upregulation was specifically due to GM-CSF. Finally, mononuclear phagocytes treated with supernatants from GM-CSF-silenced cancer cells lost the ability to induce GM-CSF in non-silenced cancer cells (Figure 9C). Thus, GM-CSF silencing data complement the ligand-blocking experiments obtained with inhibitory Abs, and strengthen our experimental evidence supporting the existence of an activated GM-CSF/HB-EGF loop between cancer cells and mononuclear phagocytes. When available, HB-EGF specifically stimulates cancer cells to produce GM-CSF, and the reciprocal availability of the two factors activates a positive feedback loop between them (Figure 7E).