Tumor-Derived G-CSF Facilitates Neoplastic Growth through a Granulocytic Myeloid-Derived Suppressor Cell-Dependent Mechanism

1Myeloid-derived suppressor cells (MDSC) constitute heterogeneous populations of monocytic and granulocytic-like cells reflecting various stages of differentiation that are well-regarded to play integral roles in immune suppression during diverse pathologic conditions, notably chronic inflammation, infection, trauma, graft-versus-host disease and neoplasia,,,,,,,. Such regulatory myeloid populations accumulate in the bone marrow, blood, peripheral lymphoid tissues and sites of disease activity, such as cancer. In fact, MDSC have been identified in various human hematologic and non-hematologic malignancies, as well as in both implantable and autochthonous animal tumor models indicating that their existence represents a key component of the neoplastic process. The notion that MDSC constitute a significant barrier to effective anti-pathogen immunity has led to a comprehensive understanding of their mechanisms of immune suppression,,,,,,,,,,,.

2Despite the fact that much attention has been dedicated to unraveling mechanisms by which MDSC mediate immune suppression, a larger gap remains in our understanding of the mechanisms that initiate their development. Understanding how they appear is also critical to the design of new therapeutic strategies that impede MDSC involvement in order to potentiate anti-pathogen immunity. In the case of neoplasia, it is generally thought that tumor-derived factors (TDF) govern diverse facets of MDSC biology, including their mobilization, recruitment and activation,,,,,,,,. In mouse models MDSC are broadly defined as CD11b+Gr-1+ cells,,,. More recently, MDSC have been divided into monocytic and granulocytic subsets, reflecting differential expression of the Ly6C and Ly6G epitopes. Monocytic MDSC are characterized as CD11b+Ly6Chigh Ly6G− or CD11b+Gr-1low cells whereas granulocytic MDSC are defined as CD11b+Ly6Clow Ly6G+ cells or CD11b+Gr-1high,,. Interestingly, recent studies reported that in the vast majority of tumor models, as well as in cancer patients, granulocytic MDSC are a predominant MDSC subset,,,,,. In fact, 70–80% of the tumor-induced MDSC response may consist of granulocytic-like cells compared to 20–30% of the cells reflecting the monocytic lineage,,. It has also been reported that both subsets are equally suppressive on a per cell basis.

3Although much interest has been dedicated to monocytic MDSC biology,,,,,,, less is understood about the origin of granulocytic MDSC. Therefore, in this study we focused on the mechanistic basis of granulocytic MDSC accumulation and its relevance to tumor growth. Given that the MDSC response is a consequence of altered myelopoiesis,,,,,, we reasoned that, when aberrantly expressed, tumor-derived granulocyte-colony stimulating factor (G-CSF) represents a key inflammatory component that facilitates granulocytic MDSC accumulation. Ordinarily, endogenous G-CSF regulates granulopoiesis and has an integral role in neutrophil mobilization in response to various insults. Exogenous G-CSF is also important to overcome neutropenia caused by various anti-neoplastic treatments. However, G-CSF exposure paradoxically can also elicit adverse effects, and inhibit innate and adaptive immunity,; yet, the precise mechanisms by which G-CSF does so remain incompletely understood. Moreover, the idea that G-CSF may not always be beneficial to the host is supported by the findings that G-CSF is aberrantly expressed by diverse human tumors, including head and neck, cervical, ovarian, pancreatic, bladder and leukemia,,,,,.

4Indeed, G-CSF production has also been shown to correlate with aberrant granulocytic or MDSC-like responses in several mouse tumor models,,; however, the mechanistic link between G-CSF and granulocytic MDSC generation has not been determined. Moreover, the correlation between G-CSF and MDSC accumulation has largely focused on the global (CD11b+Gr-1+) MDSC response,, making it difficult to distinguish and conclude the impact of G-CSF on the monocytic vs. granulocytic MDSC subsets. Thus, this study focused on three fundamental questions: 1) Does tumor-derived G-CSF drive granulocytic MDSC generation in vivo? 2) Is the granulocytic MDSC response pro-tumorigenic in vivo? and 3) Does recombinant G-CSF administration elicit a myeloid response phenotypically, functionally and molecularly similar to that of G-CSF-producing tumors? The latter question not only addresses the mechanistic link between G-CSF and granulocytic MDSC development in vivo, but also the potential implications of aphysiologic G-CSF levels in non-neoplastic disease or clinical settings.

5To test our hypothesis, we made use of two implantable orthotopic mouse models of mammary cancer that are proficient at MDSC generation, termed 4T1 and AT-3. 4T1 is a well-characterized mammary tumor model that induces a rapid accumulation of systemic and intratumoral MDSC, broadly defined by their CD11b+Gr-1+ expression,,. The AT-3 mammary tumor cell line, recently generated in our laboratory, was established from a primary mammary carcinoma of the MMTV-PyMT transgenic mouse model, henceforth termed MTAG mice. CD11b+Gr-1+ MDSC responses are observed under autochthonous (i.e., MTAG) and implantable (i.e., AT-3) tumor settings.

6We first screened tissue culture supernatants of 4T1 cells for cytokines and chemokines commonly associated with myelopoiesis and MDSC biology,,,,,,,. We found extremely high levels of G-CSF production, whereas only marginal amounts of all other factors tested were detectable (Fig. 1A). Similarly, AT-3 cells produced high levels of G-CSF (Fig. 1B). Analysis of two additional mammary tumor cell lines, DA-3 and EMT6, yielded substantially high levels of G-CSF (Fig. S1). In contrast to these four cell lines, G-CSF production was not detectable in the CMS4 sarcoma cell line (Fig. 1B). It is worth noting that mice implanted with sarcomas generally display minimal MDSC burden.

7We then examined G-CSF levels in the sera of 4T1 and AT-3 tumor-bearing mice (Fig. 1C). High levels of G-CSF were observed in both groups of tumor-bearing mice, whereas nominal amounts were detectable in the non-tumor-bearing control mice. As an additional control, mice were injected with recombinant G-CSF protein (Fig. 1C). Analysis of G-CSF protein revealed copious amounts in the blood reflecting a concentration range seen under tumor-bearing conditions. Lastly, sera G-CSF levels were measured in separate groups of MTAG mice at discrete stages of autochthonous tumor growth: no palpable tumors, minimal tumor burden (i.e., <200 mm3) and extensive tumor burden (i.e., >1500 mm3). Not surprisingly, G-CSF levels increased with increasing tumor burden. In contrast, little-to-no G-CSF was observed in the non-tumor-bearing controls (Fig. 1D). G-CSF levels in mice with large tumor burdens paralleled the high levels found in the implantable models. These data indicate that G-CSF is produced at high levels in multiple tumor models, and represents a potentially critical player for altered myelopoiesis and granulocytic MDSC generation. These studies also extend observations made elsewhere,, regarding correlations between tumor growth and G-CSF production.

8To gain insights into the role of G-CSF as a potential mediator of granulocytic MDSC development, we compared the MDSC response generated in tumor-bearing mice to mice treated with recombinant mouse G-CSF protein (Fig. 2A). In control mice, the percentage of CD11b+Gr-1+ myeloid cells was <7%. In contrast, treatment of mice with G-CSF protein caused significant splenomegaly, resulting from a massive expansion of CD11b+Gr-1+ myeloid cells (Fig. 2A) with a corresponding decrease in the percentage of cells within the lymphoid compartment (Fig. S2 panel A). The accumulation of such myeloid cells was comparable to the CD11b+Gr-1+ myeloid response during AT-3 or 4T1 tumor growth (Fig. 2A).

9While the co-expression of CD11b and Gr-1 markers encompass the ‘global’ MDSC population, the Gr-1 epitopes, Ly6C and Ly6G are used to delineate monocytic and granulocytic MDSC subsets,. Using these additional phenotypic markers, splenocytes from G-CSF-treated mice displayed a prominent expansion of granulocytic-like cells (81.4% granulocytic vs. 8.2% monocytic) compared to control mice (23.5% granulocytic vs. 16.9% monocytic) (Fig. 2B). Indeed, the phenotypic pattern observed in G-CSF-treated mice paralleled that seen in AT-3 and 4T1 tumor-bearing mice (Fig. 2B). Morphologic analysis confirmed their monocytic and granulocytic lineages (Fig. S2 panel B).

10Next, we examined and compared the suppressive capacity of splenic CD11b+Gr-1+ cells from control, G-CSF-treated and AT-3 tumor-bearing mice, based on inhibition of polyclonal or allo-specific T cell proliferation as readouts of suppressive behavior. Similar to what we observed previously using CD11b+Gr-1+ cells from 4T1 tumor-bearing mice, cells from G-CSF-treated mice also exerted significant suppression of polyclonal T cell proliferation (Fig. 2C). Moreover, we observed that CD11b+Gr-1+ cells from G-CSF-treated and AT-3 tumor-bearing mice, but not control mice significantly suppressed allo-specific T cell proliferation (Fig. 2D). Overall, the CD11b+Gr-1+ responses generated by G-CSF treatment were similar to those of 4T1 tumor-bearing mice in terms of phenotype, morphology and functional suppression.

11Global gene expression studies were performed to determine whether the granulocytic-like MDSC populations from G-CSF treated mice resembled those of tumor-bearing (TB) mice more so than those of the non-tumor-bearing control (i.e., WT) at a molecular level. To that end, splenic CD11b+Gr-1high cell populations from WT, 4T1-TB or G-CSF-treated BALB/c mice (as in Fig. 2) were purified in two independent experiments by flow cytometry (>98% purity) and subjected to whole genome expression profiling using Illumina microarrays (see Materials and Methods). Differential expression analysis of microarray data revealed that the gene expression patterns of cells from G-CSF-treated mice resembled those of 4T1-TB hosts more than of the WT group (Fig. 3). Specifically, we identified 932 and 734 genes showing significant expression changes (i.e., >2-fold change, P<0.01) in 4T1-TB vs. WT comparison and G-CSF vs. WT comparison respectively, and only 22 genes were differentially expressed between 4T1-TB and G-CSF using the same significance criteria (Fig. 3A). Hierarchical clustering based on identified differentially expressed genes showed that G-CSF and 4T1-TB samples were clustered into a group separated from that of the WT samples (Fig. 3B). Overall, based on microarray gene expression profiling analysis, these data indicate that G-CSF administration can recapitulate a granulocytic-like MDSC phenotype highly comparable to that of tumor-bearing mice.

12To test the hypothesis that G-CSF production is causally linked to MDSC generation and/or tumor growth, we sought to inhibit G-CSF in tumor-bearing mice using two different approaches. First, treatment with anti-G-CSF mAb significantly reduced primary tumor growth compared to mice treated with the isotype control Ab (Fig. 4A), suggesting an important pro-tumorigenic role of G-CSF during AT-3 tumor growth. It is unlikely that G-CSF acts in an autocrine fashion, as AT-3 (as well as 4T1) tumor cells do not express detectable cell surface G-CSF receptor.

13Second, we made use of a loss-of-function approach, whereby G-CSF expression was down-regulated in AT-3 cells via RNA interference. Individual stable cell lines were generated using different short-hairpin (shRNA) sequences delivered via retroviral-based methods. G-CSF knockdown, relative to the non-silencing (scrambled sequence) control, was strongly observed with all constructs, as measured by RT-PCR (Fig. 4B). In contrast, genes associated with other aspects of tumor biology, such as PyMT (MTAG tumor-specific), COX-2, VEGF-A and VEGF-C remained unaffected compared to the control (Fig. 4B). G-CSF protein levels in tumor cell supernatants were also significantly reduced (Fig. 4C).

14Since all three constructs appeared to be effective in silencing G-CSF expression, we arbitrarily selected cells transduced with construct 1 for further studies. Using these cells, termed AT-3 shRNA 1, we explored the impact of G-CSF loss on tumor growth and MDSC accumulation in vivo. To that end, syngeneic female B6 mice were implanted orthotopically with AT-3 shRNA 1 cells or the control tumor cells. Consistent with our observations using Ab-based G-CSF blockade (Fig. 4A), AT-3 shRNA 1 tumor cells grew significantly slower than control tumor cells at multiple time points (Fig. 4D). In contrast, both cell lines proliferated similarly in vitro (Fig. S3 panel A), suggesting that in vivo interactions influenced tumor growth patterns. To verify that differences in tumor growth patterns correlated with differences in G-CSF levels, sera from both groups of mice with equal tumor volumes were analyzed for G-CSF protein. We found that sera G-CSF levels were significantly lower in mice bearing AT-3 shRNA 1 tumor cells compared to the control (Fig. S3 panel B). Thus, altering tumor-derived G-CSF levels exerted a significant impact on tumor growth.

15To determine whether differences in tumor-derived G-CSF levels also affected granulocytic MDSC accumulation, we analyzed splenic MDSC frequencies from both groups of mice with equal tumor volumes of 800 – 1000 mm3 (i.e., collected from the control group on day 52 vs shRNA 1 group on day 73). First, we observed significant reductions in splenocyte numbers in mice bearing AT-3 shRNA 1 tumor cells compared to the control (Fig. 5A). Second, to determine whether these splenic differences reflected differences in MDSC populations, we analyzed unfractionated splenocytes for co-expression of the CD11b and Gr-1 markers (Fig. 5B). AT-3 shRNA 1 tumor-bearing mice displayed a greater than two-fold reduction in the percentage of CD11b+Gr-1+ cells compared to the control (19.7% to 48.01% respectively) (Fig. 5B). Third, to delineate which MDSC subset(s) were affected most, we analyzed these preparations for granulocytic and monocytic fractions based on differential expression of the Ly6C and Ly6G epitopes in the gated CD11b+ fraction (Fig. 5C). Importantly, we found significant reductions in the percentage (Fig. S3 panel C) and number (Fig. 5C) of granulocytic MDSC from mice bearing AT-3 shRNA 1 tumor cells compared to the control. Conversely, the monocytic fraction remained largely unaffected, indicating a preferential effect on the granulocytic subset.

16Fourth, to determine whether MDSC were affected qualitatively, splenic CD11b+Gr-1+ cells were isolated and tested for their ability to inhibit allo-specific T cell proliferation (Fig. 5D). MDSC derived from both AT-3 shRNA 1 and control tumor-bearing mice, however, displayed comparable capacity to suppress allo-specific T cell proliferation. These data indicated that G-CSF knockdown in AT-3 tumors can slow tumor growth and reduce granulocytic MDSC burden; however, the suppressive capacity of CD11b+Gr-1+ MDSC was not altered. Thus, tumor-derived G-CSF likely plays a more dominant role in affecting the quantity of granulocytic MDSC, which still translates to a significant effect on primary tumor growth.

17To further demonstrate causality between tumor-derived G-CSF with tumor growth and MDSC accumulation, we made use of a gain-of-function approach. As shown earlier, CMS4 sarcoma cells secrete undetectable levels of G-CSF protein (Fig. 1B). Therefore, CMS4 cells were stably transfected with a murine G-CSF expression plasmid or an empty vector control. Both G-CSF-producing CMS4 cells and parental AT-3 tumor cells displayed high levels of G-CSF mRNA, whereas G-CSF transcript was undetectable in the CMS4 vector control cells (Fig. 6A). Differences in G-CSF expression were confirmed at the protein level (Fig. 6B).

18In a reciprocal fashion to our G-CSF loss-of-function approach (Figs. 4 and 5), we explored the impact of G-CSF gain-of-function on CMS4 tumor growth and MDSC generation in vivo. Indeed, CMS4 tumors expressing G-CSF grew significantly faster than the vector control tumor cells (Fig. 6C). For reference, we observed that 10/10 mice bearing G-CSF-producing CMS4 cells displayed tumor volumes >500 mm3, whereas none of the control cells achieved that level at these time points. Again, this G-CSF effect was not due to intrinsic differences in tumor cell growth in vitro, as proliferation was comparable between both cell lines (Fig. S4 panel A). Moreover, this effect did not reflect an autocrine loop, as both groups of CMS4 tumor cells did not express the G-CSF receptor (Fig. S4 panel B). To verify that differences in tumor growth correlated with differences in G-CSF levels, sera from both groups of mice with equal tumor volumes were analyzed for G-CSF protein. We observed that sera G-CSF levels were significantly higher in mice bearing G-CSF-producing CMS4 cells compared to the control cells (Fig. S4 panel C). Thus, modulation of tumor-derived G-CSF levels exerts a significant impact on tumor growth.

19Mice implanted with G-CSF-producing CMS4 cells, compared to the control, exhibited a massive increase in spleen size and splenocyte number (Fig. 7A and Fig. S4 panel D). Such analyses were performed on mice with equivalent tumor burdens of ∼1000 mm3. It is important to note that the control tumors required a longer time to achieve this tumor volume (i.e., 2 additional weeks). Mice bearing G-CSF-producing CMS4 cells also had an approximately 7-fold higher percentage of CD11b+Gr-1+ cells compared to those implanted with the control cells (Fig. 7B; 48.7% to 7.2%). Moreover, myeloid subset analysis revealed significant increases in the percentage (Fig. S4 panel E) and number (Fig. 7C) of granulocytic cells in mice bearing the G-CSF-producing tumor compared to the control tumor. It is also interesting to note that the number of monocytic cells also increased, although not to the same extent (Fig. 7C).

20To determine whether tumor-derived G-CSF also affected function, we examined the ability of CD11b+Gr-1+ cells from control and G-CSF-producing tumor-bearing mice to inhibit allo-specific T cell proliferation. As with the AT-3 model (Fig. 5D), CD11b+Gr-1+ cells derived from both groups of tumor-bearing mice suppressed T cell proliferation at significant and comparable levels (Fig. 7D). These data indicate that MDSC are generated in this CMS4 model, and that tumor-derived G-CSF likely plays a greater role in the development and accumulation of these cells compared to affecting their suppressive behavior.

21Thus far, our data demonstrate that tumor-derived G-CSF impacts tumor growth, as well as granulocytic MDSC accumulation. However, it remained to be determined whether the effect of G-CSF on tumor growth acted through a MDSC-dependent mechanism. To do so, we adopted two approaches; one was based on a direct assessment of the functional impact of the MDSC subset in the tumor microenvironment, and the second was based on cell depletion (Fig. 8).

22Phenotypic characterization of the primary tumor microenvironment, in this case the 4T1 model, revealed a substantial infiltration of CD11b+Gr-1+ cells with the vast majority bearing a granulocytic phenotype (Fig. 8A and B). Thus, it is clear that granulocytic MDSC can infiltrate or traffic to the tumor microenvironment. To determine whether such cells are also functional at the tumor site, we took advantage an ‘admix’ approach to directly assess the impact of the myeloid population on tumor growth originally described by Moses and colleagues,. The granulocytic fraction was purified from G-CSF-treated mice by flow cytometric sorting of splenic CD11b+ cells co-expressing high levels of Gr-1, as in Fig. 3. Such cells were mixed with 4T1 tumor cells and then implanted into new groups of syngeneic mice. We observed that CD11b+Gr-1high cells from G-CSF-treated mice significantly enhanced tumor growth compared to 4T1 cells injected alone (Fig. 8C). Earlier studies in this 4T1 model showed that CD11b+Gr-1+ cells from non-tumor-bearing mice are not pro-tumorigenic and do not alter tumor growth, compared to tumor cells injected alone. In contrast, CD11b+Gr-1+ cells from 4T1 tumor-bearing mice enhanced tumor growth, compared to tumor cells injected alone or tumor cells admixed with myeloid counterparts from the non-tumor-bearing controls.

23Our second approach was based on Gr-1 cell depletion, widely used in the tumor field,,. Although targeting the Gr-1 populations is not necessarily MDSC-specific, it has been used as a ‘proof-of-concept’ approach to assess the potential role of Gr-1-expressing cells,, in affecting a disease outcome. Besides Gr-1 cell depletion as a loss-of-function approach, to our knowledge no other alternative has been reported that can specifically deplete or inactivate MDSC without potentially impacting the tumor or host compartments. We hypothesized that if G-CSF enhances tumor growth through mobilization of granulocytic MDSC, then depletion or inactivation of such Gr-1-expressing myeloid cells would result in a reduction of tumor growth. To avert a leukopenia-induced moribund state due to repeated injections of the antibody, however, the experiment was carried out until differences in tumor growth were observed between groups. To that end, we returned to our CMS4 model in which G-CSF levels are regulated experimentally.

24Treatment of mice bearing the G-CSF-expressing tumor with anti-Gr-1 mAb reduced tumor growth, which approached levels seen with the vector control tumor (Fig. 8D) (P<0.04 for the various comparisons at day 20). The reduction in tumor growth became more apparent over time, consistent with a time-dependent component for anti-Gr-1 mAb treatment to build up to a biologic effect. In contrast, anti-Gr-l mAb treatment had no effect on the growth of the vector control tumor compared to treatment with the isotype control (Fig. 8D). These data are consistent with the undetectable levels of tumor-derived G-CSF and low MDSC burden in the parental CMS4 model, and reinforces the likelihood that anti-Gr-1 mAb treatment did not exert non-specific effects against tumor growth. Overall, these data reveal that tumor-derived G-CSF potentiates tumor growth via granulocytic MDSC, thus unveiling a previously undescribed ‘G-CSF-granulocytic MDSC-tumor axis’ in regulating tumor progression.

25MDSC are a major regulatory myeloid population that accumulates during a range of pathologic processes and function to inhibit innate and adaptive immunity,,,,,,,. MDSC can be found systemically in the blood and peripheral lymphoid tissues, as well as locally at sites of disease activity. It turns out that granulocytic cells comprise a major component of the MDSC response,,,,,; albeit, the underlying reasons for this remain unclear. Therefore, efforts to understand how they develop will have valuable implications for the design of new therapeutic paradigms. Although a number of other TDF have been linked to diverse elements of MDSC biology, namely VEGF, GM-CSF, IL-1β, IL-6, PGE2, IFN-γ, SCF or IL-17,,,,, none have been rigorously tested to explain the connection between granulocytic MDSC response and tumor growth. Given that the granulocytic MDSC response is a likely manifestation of deregulated granulopoiesis, we hypothesized that inappropriate production of tumor-derived G-CSF contributes to granulocytic MDSC accumulation. Similarly, other groups have found positive correlations between G-CSF production and aberrant granulocytic or MDSC-like responses in several mouse models,,. However, despite the correlation, the causal basis for granulocytic MDSC accumulation remained untested.

26Three systematic in vivo approaches were taken to validate the role of G-CSF in granulocytic MDSC generation: 1) loss-of-function using RNA interference; 2) gain-of-function using a tumor model over-expressing G-CSF protein; and 3) direct injection of purified G-CSF protein. Using multiple approaches, our data support the hypothesis that tumor-derived G-CSF facilitates granulocytic MDSC generation which displays both immunosuppressive and pro-tumorigenic activities. Furthermore, G-CSF protein alone was sufficient to generate granulocytic-like MDSC, which strongly recapitulated phenotypic, functional and molecular characteristics observed with tumor-induced granulocytic MDSC. These data imply that high concentrations of G-CSF acting in a systemic and aphysiologic manner can impair normal myeloid cell development or differentiation, leading to the accumulation of granulocytic-like MDSC. Thus, exogenous or endogenous stores of G-CSF may have paradoxical roles during certain pathologic circumstances, including neoplasia.

27Our data do not necessarily preclude the potential role of host-derived G-CSF. However, the loss- and gain-of-function studies directly implicate a prominent role of tumor-derived G-CSF in the accumulation of granulocytic MDSC. The possibility that a ‘G-CSF-granulocytic MDSC-tumor axis’ may be important in a broader sense is supported by our data in multiple tumor models or approaches, and the fact that G-CSF production has been identified in several human cancer types,,,,,. It is important to note that these data do not exclude the possibility that other TDF play relevant roles in granulocytic MDSC biology, especially in the absence of G-CSF production or in a G-CSF-independent manner. For example, in cases where tumors produce high amounts of GM-CSF, the monocytic fraction tends to accumulate more so than the granulocytic fraction, although the latter subset is still represented,.

28In addition to its affects on granulocytic MDSC burden, we observed that tumor-derived G-CSF affects tumor growth rate. Through G-CSF blockade, as well as G-CSF loss- and gain-of-function approaches, we found that altering G-CSF levels had a direct impact on tumor growth. To directly demonstrate that granulocytic MDSC are pro-tumorigenic activity, highly purified CD11b+Gr-1high cells from G-CSF-treated mice were collected, admixed with 4T1 cells and implanted orthotopically into naïve recipients. We found that such myeloid-tumor admixtures significantly enhanced tumor growth, demonstrating that granulocytic-like populations induced by G-CSF treatment are pro-tumorigenic within the tumor microenvironment. Moreover, depleting Gr-1+ cells in mice bearing G-CSF-producing, but not non-G-CSF-producing tumors facilitated a significant reduction in tumor growth, supporting a causal link between G-CSF-induced granulocytic MDSC accumulation and tumor progression.

29Our data also implicate an important effect of G-CSF on the monocytic MDSC subset. Indeed, modulation of tumor G-CSF levels was accompanied by quantitative changes in the monocytic fraction, albeit, not to the same magnitude. Moreover, our data indicate that the monocytic fraction from tumor-bearing mice expressed appreciable levels of the G-CSF receptor (Fig. S5). Although it remains undetermined whether G-CSF affects the monocytic fraction directly or indirectly, this latter observation suggests that G-CSF may do so in part through a direct mechanism. Thus, the impact of tumor-derived G-CSF may have broader consequences on the overall MDSC response. Finally, it remains to be determined what regulates constitutive tumor-derived G-CSF production in the first place. As this study necessarily focused on the mechanistic basis of a ‘G-CSF-granulocytic MDSC-tumor axis’, we believe these new data now provide the framework for pursuing in detail how G-CSF is regulated by neoplastic cells.

30Taken collectively, we demonstrated: 1) a novel role of G-CSF in myeloid biology, acting at the level of granulocytic MDSC accumulation; 2) that granulocytic MDSC can facilitate tumor growth, indicating a previously unrecognized causal link between this MDSC subset and neoplastic progression; 3) that recombinant G-CSF administration alone to healthy mice can promote granulocytic MDSC development, which may help to explain in part the basis of certain immune-associated adverse events in healthy individuals that receive G-CSF injections to mobilize progenitor cells for use in hematopoietic transplant settings; and 4) that experimental approaches that target the initiating TDF (i.e., G-CSF) leads to tumor growth inhibition under conditions of host immunosurveillance. Current MDSC-based pharmacologic approaches include vitamin D3, retinoic acid (ATRA), KIT-specific antibody, VEGF blockade and gemcitabine (reviewed in ref. 1). Thus, therapeutic targeting of TDF that initiate granulocytic MDSC development may offer additional ways to abrogate MDSC-mediated mechanisms of tumor growth, thereby enhancing immunotherapy efficacy.

31All experiments were conducted and approved under our Institutional Animal Care and Use Committee (IACUC) at Roswell Park Cancer Institute under protocol ID numbers 1108 M and 1117 M and in accordance with institutional regulations, NIH and Public Health Service policies. Female C57BL/6 (B6) or BALB/c (6 – 8 wks of age) mice were obtained from the NCI-Frederick Cancer Research Animal Facility (Frederick, MD). The MMTV-PyMT/B6 transgenic mouse, termed MTAG, expresses the polyomavirus middle T antigen controlled through the MMTV-LTR promoter, originally derived in FVB mice, was backcrossed >10 times onto a C57BL/6 (H-2b) background and kindly provided by S. Gendler (Mayo Clinic, Scottsdale, AZ). MTAG mice develop multifocal mammary carcinomas.

32The 4T1 mammary tumor cell line (ATCC, Manassas, VA) was derived from BALB/c mice and maintained in a RPMI-based medium. The AT-3 tumor cell line was established from a primary mammary gland carcinoma of a MTAG mouse and propagated in a DMEM-based medium. The CMS4 sarcoma line, a kind gift of A. DeLeo (Univ. of Pittsburgh, Pittsburgh, PA), was derived from BALB/c mice and maintained in a RPMI-based medium. In vitro assays were performed in a RPMI-based medium.

33Female BALB/c mice were injected orthotopically in an abdominal mammary gland with 5×104 4T1 cells. Female B6 mice were injected similarly with 5×105 AT-3 cells, including those cell lines used in the knockdown studies. For the G-CSF over-expression tumor studies, syngeneic BALB/c mice were injected subcutaneously into the flank with 4×105 CMS4 cells. In the admix experiments, 5×104 4T1 tumor cells were mixed with MACS-purified, MHC-matched CD11b+Gr-1+ cells at a 2∶1 ratio, respectively, prior to orthotopic implantation. Spleens, serum and tumor tissue were recovered, as appropriate. For the G-CSF blockade studies, B6 mice were injected orthotopically, and on day 7 when tumors became palpable, mice were injected intraperitoneally with 10 µg of either neutralizing anti-G-CSF mAb or rat IgG isotype (R&D Systems) for 8-consecutive days, as described. For the in vivo depletion experiments, BALB/c mice were injected ip with anti-Gr-1 Ab (clone RB6-8C5; BioXcell, West Lebanon, NH) or control Ab (Rat IgG2b clone LTF-2; BioXcell, West Lebanon, NH). Antibodies were administered at 200 µg/mouse once tumors became palpable, and continued twice weekly for the duration of the experiment, similarly as described,,. For all tumor experiments, mice were euthanized when tumor volumes approached 2 cm3 or for other health considerations. Tumor growth was measured 2 – 3 times/week in two dimensions and tumor volume was calculated using the formula (width2×length)/2. For recombinant G-CSF in vivo studies, mice were injected subcutaneously with recombinant G-CSF (10 µg/day, Peprotech, Rocky Hill, NJ) for 5 consecutive days and serum/tissues collected, similarly as described.

34Splenic CD11b+Gr-1+ cells were purified from the indicated groups using similarly established protocols,,,. Selections were performed using CD11b+ magnetic beads from Miltenyi Biotec (Auburn, CA) on an AutoMACs system. The percentage of CD11b+ cells was routinely >95%. Of these cells, >90% co-expressed CD11b and Gr-1 markers in both control and tumor-bearing mice, as previously reported.

35Myeloid preparations were pre-incubated with anti-CD16/32 mAb (BD Biosciences, San Diego, CA) to block Fc receptor binding, followed by incubation with directly conjugated primary mAb, as described. Labeled cells were collected on a FACSCalibur flow cytometer (BD, San Jose, CA) and analyzed by FCS-Express software (De Novo Software, Los Angeles, CA). Antibodies reactive against the following cell surface markers were used (including appropriate isotype controls): CD11b (BioLegend, San Diego, CA), Ly6G, Ly6C, Gr-1 (BD Biosciences) and G-CSF receptor mAb (Abcam, Cambridge, MA).

36Flat-bottomed, 96-well plates were coated with anti-CD3 mAb (1 µg/well) (BD Biosciences). Total splenic T cells were purified from naïve mice using the AutoMACS system, and plated at 5×104 cells/well. Isolated CD11b+Gr-1+ cells were added to the plate at multiple CD11b+Gr-1+ cell:T cell ratios. Plates were incubated for 48 hr at 37°C, after which 3H-thymidine (1 µCi/well) was added for an additional 24 hr. Proliferation determined by measuring 3H-thymidine uptake. For the mixed lymphocyte reaction (MLR), one-way allogeneic H-2b anti-H-2d or H-2d anti-H-2b cultures was established. Responders (1×105 unfractionated splenocytes/well from naïve mice) were cultured with irradiated (20 Gy) stimulators (2×105 unfractionated splenocytes, also from naïve mice) in 96-well, U-bottomed plates. MHC-matched CD11b+Gr-1+ splenocytes were then added to the allo-assay at one or more cell densities. Plates were incubated for 96 hr at 37°C, after which 3H-thymidine was added for an additional 24 hr, and cells harvested as above.

37To efficiently isolate CD11b+ Gr-1high myeloid cells from WT control, 4T1 tumor-bearing and recombinant G-CSF treated mice, we first incubated total splenocytes with CD11b (BioLegend, San Diego, CA) and Gr-1 (BD Biosciences) antibodies as indicated previously. 4T1 tumor-bearing mice had tumor burdens of >1000 mm3, whereas mice treated with recombinant G-CSF were given a dose regimen as previously indicated. The cells were sorted on a FACSAria cell sorter (BD Biosciences) and analyzed using FACSDiva software (BD Biosciences). The FACS resulted in >98% recovery of the CD11b+ Gr-1high population. Total RNA was extracted from 5×106 sorted cells using Qiagen's miRNeasy kit, which was quantified and subsequently used for microarray analysis.

38RNA interference methods were used to silence G-CSF expression in AT-3 tumor cells. Stably, transduced cell lines were generated by the shRNA core facility at Roswell Park. Briefly, the retroviral vector, pSM2, was engineered to contain several independent sequences targeting murine G-CSF expression (henceforth termed, shRNA 1 (71191), 2 (68539), 3 (64790) or non-silencing/scrambled control) (Open Biosystems/Thermo Fisher Scientific, Huntsville, AL). AT-3 tumor cells were transduced with the individual G-CSF shRNA-containing retroviral supernatants. Cells were stably propagated under puromycin selection (2 µg/ml). For the G-CSF over-expression studies, murine G-CSF cDNA was PCR-cloned from AT-3 cells using the following primers: forward: 5′-AAAAAACACCATGGCTCAACTTTCTGCCC-3′; reverse: (5′-AAAAAGCGGCCGCCTAGGCCAAGTGGTGCAGA-3′. Afterwards, the gene was gel-purified, inserted into the pcDNA3.1(+) expression plasmid (Invitrogen, Carlsbad, CA), amplified, and purified under endotoxin-free conditons for use in transfection. CMS4 cells were then transfected with the expression vector or empty vector as a control using Lipofectamine (Invitrogen) per manufacturer's instructions. Cells were stably propagated under G418 (800 µg/ml) selection.

39Cell-free supernatants from tumor cell lines (1×106 cells/ml) were collected after incubation for 24 hr at 37°C. Serum was also collected, and all samples were stored at −80°C until assayed. Comprehensive analysis of cytokines was initially performed using Quansys multipex ELISA services (Quansys Biosciences, Logan, UT). Subsequently, G-CSF levels were quantified using G-CSF-specific ELISA (R&D Systems, Minneapolis, MN).

40Total RNA was isolated using RNeasy Mini kits (Qiagen; Valencia, CA) according to manufacturer's instructions. cDNA was synthesized using the ThermoScript RT-PCR system (Invitrogen). The cDNA was then used as the template for PCR amplification of the indicated murine genes in a PTC-200 thermal cycler (MJ Research, Waltham, MA) under the following standard conditions: 94°C for 2 min, 30 cycles (94°C for 30 sec, 60°C for 30 sec and 72°C for 1 min) and 72°C for 10 min. The following mouse primer sets were used: G-CSF forward: 5′-CTCAACTTTCTGCCCAGAGG-3′ and reverse: 5′-AGCTGGCTTAGGCACTGTGT-3′; PyMT (MTAG) forward: 5′-AGTCACTGCTACTGCACCCAG-3′ and reverse: 5′-CTCTCCTCAGTTCCTCGCTCC-3′; VEGF-A forward: 5′-CTGTGCAGGCTGCTGTAACG-3′ and reverse: 5′-GTTCCCGAAACCCTGAGGAG-3′; VEGF-C forward: 5′-TGTGGGGAAGGAGTTTGGAGC-3′ and reverse: 5′-CGGCAGGAAGTGTGATTGGC-3′; COX-2 forward: 5′-ACAACATCCCCCTCCTGCG-3′ and reverse: 5′-GCTCCTTATTTCCCTTCACACCC-3′, GAPDH forward: 5′-CATCACCATCTTCCAGGAGCG-3′ and reverse: 5′-ACGGACACATTGGGGGTAGG -3′. PCR products were separated on a 1% agarose gel and the images captured with the Chemidoc Imaging System (BioRad, Hercules, CA).

41Total RNA from highly purified splenic CD11b+Gr-1high cells from the different experimental groups was quantified by a ND-1000 spectrophotometer (NanoDrop, Thermo Scientific) and evaluated for degradation using a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). Samples requirements were as follows: RIN >7, an OD 260∶280 of 1.9-2.0, and an OD 260/230 >1.8. The MouseWG-6 whole-genome gene expression array and direct hybridization assay was used for expression profiling (Illumina Inc., San Diego, CA). Initially, we converted 250–500 ng total RNA to cDNA, followed by in vitro transcription to generate biotin-labeled cRNA (Illumina TotalPrep RNA Amplification Kit, Ambion Inc., Austin, TX). 750 ng of the labeled probes were then mixed with hybridization reagents and hybridized overnight at 58°C to the MouseWG-6 v2 BeadChips. Following washing and staining with Cy3-streptavidin conjugate, the BeadChips were imaged using the Illumina iScan Reader to measure fluorescence intensity of each probe, where intensity corresponds to the quantity of the respective mRNA in the original sample. The expression profiles have been deposited in NCBI's Gene Expression Omnibus (GEO) with GSE accession number GSE32209. All data is MIAME compliant.

42BeadChip data files were then analyzed with Illumina's GenomeStudio gene expression module and R-based Bioconductor package to determine gene expression signal levels. Briefly, the raw intensity of Illumina MouseWG-6 gene expression array was scanned and extracted using BeadScan, with the data corrected by background subtraction in GenomeStudio module. The lumi module in the R-based Bioconductor Package was used to transform the expression intensity into log2 scale. The log2 transformed intensity data were normalized using Quantile normalization function. We then performed three separate comparisons for 4T1-TB versus WT, G-CSF versus WT and 4T1-TB versus G-CSF. We used the Limma program in the R-based Bioconductor package to calculate the level of gene differential expression for each comparison. Briefly, a linear model was fit to the data (with cell means corresponding to the different condition and a random effect for array), and selected contrast for each comparison was performed. For each comparison, we obtained the list of differentially expressed genes (≥2-fold change) constrained by P<0.01. Results are reported from data collected on two biologic replicates of each group.

43For comparisons between control and experimental groups, data were recorded as mean ± SEM of the indicated number of mice or experiments. Statistical analysis was determined using a two-sided unpaired t-test or exact Wilcoxon Rank Sum tests, where appropriate. P-values less than 0.05 were considered significant.