DMXAA Causes Tumor Site-Specific Vascular Disruption in Murine Non-Small Cell Lung Cancer, and like the Endogenous Non-Canonical Cyclic Dinucleotide STING Agonist, 2′3′-cGAMP, Induces M2 Macrophage Repolarization

The vascular disrupting agent 5,6-dimethylxanthenone-4-acetic acid (DMXAA), a murine agonist of the stimulator of interferon genes (STING), appears to target the tumor vasculature primarily as a result of stimulating pro-inflammatory cytokine production from tumor-associated macrophages (TAMs). Since there were relatively few reports of DMXAA effects in genetically-engineered mutant mice (GEMM), and models of non-small cell lung cancer (NSCLC) in particular, we examined both the effectiveness and macrophage dependence of DMXAA in various NSCLC models. The DMXAA responses of primary adenocarcinomas in K-rasLA1/+ transgenic mice, as well as syngeneic subcutaneous and metastatic tumors, generated by a p53R172HΔg/+; K-rasLA1/+ NSCLC line (344SQ-ELuc), were assessed both by in vivo bioluminescence imaging as well as by histopathology. Macrophage-dependence of DMXAA effects was explored by clodronate liposome-mediated TAM depletion. Furthermore, a comparison of the vascular structure between subcutaneous tumors and metastases was carried out using micro-computed tomography (micro-CT). Interestingly, in contrast to the characteristic hemorrhagic necrosis produced by DMXAA in 344SQ-ELuc subcutaneous tumors, this agent failed to cause hemorrhagic necrosis of either 344SQ-ELuc-derived metastases or autochthonous K-rasLA1/+ NSCLCs. In addition, we found that clodronate liposome-mediated depletion of TAMs in 344SQ-ELuc subcutaneous tumors led to non-hemorrhagic necrosis due to tumor feeding-vessel occlusion. Since NSCLC were comprised exclusively of TAMs with anti-inflammatory M2-like phenotype, the ability of DMXAA to re-educate M2-polarized macrophages was examined. Using various macrophage phenotypic markers, we found that the STING agonists, DMXAA and the non-canonical endogenous cyclic dinucleotide, 2′3′-cGAMP, were both capable of re-educating M2 cells towards an M1 phenotype. Our findings demonstrate that the choice of preclinical model and the anatomical site of a tumor can determine the vascular disrupting effectiveness of DMXAA, and they also support the idea of STING agonists having therapeutic utility as TAM repolarizing agents.


Introduction
Strategies targeting the tumor vasculature represent an attractive approach in cancer therapy, and as such there has been much interest in a class of drugs known as vascular disrupting agents (VDA) [1,2]. The VDA, 5,6-dimethylxanthenone-4-acetic acid (DMXAA; a.k.a. ASA-404) specifically targets immature and unstable vasculature of solid tumors, leading to thrombosis, hemorrhage, and necrosis [3]. In a host of preclinical studies involving many different tumor types and primarily in subcutaneous tumor models, DMXAA has demonstrated potent antitumor activity [4][5][6]. In addition, synergies were observed when DMXAA was used in conjunction with cytotoxic chemotherapeutic agents that targeted the viable rim of tumor cells that typically survive DMXAA treatment [3,[7][8][9].
There is evidence that the actions of DMXAA on tumor vasculature involve both direct and indirect effects, via targeting of the endothelium, and macrophages, respectively. The latter appear to be the most important, and are the result of DMXAA-triggered release of tumor-associated macrophage (TAM)-derived factors, such as TNF-a and NO [5,7,[10][11][12], together with contributions from various other cytokines and chemokines [2,[6][7][8]. Following success in preclinical studies, the impetus for moving DMXAA into a Phase III trial for NSCLC stemmed largely from the observed increase in overall survival reported in a previous Phase II trial [13,14]. However, the larger trial, and other Phase II trials, failed to produce favorable outcomes [15][16][17]. This raised the question as to why there was such a discrepancy between the positive results obtained using preclinical animal models and the results in the clinic. Recently, it was shown that DMXAA exhibits differential effects on murine and human macrophages [18], and also that the stimulator of interferon genes (STING), was a receptor for DMXAA [19][20][21]. The finding that DMXAA was unable to activate human STING provided a salient explanation for the failure of this agent in the human clinical trials [20,21].
With the goal of gaining further insight into additional variables accounting for the differential effects of DMXAA between preclinical and clinical trials, we examined the effects of this agent in several mouse models, including: (a) syngeneic subcutaneous and metastatic tumors due to a cell line (344SQ-ELuc) derived from the p53 R172HDg/+ K-ras LA1/+ genetically engineered mutant mouse (GEMM) model of NSCLC [22][23][24]; (b) primary lung adenocarcinomas arising in the K-ras LA1/+ model of NSCLC; and (c) subcutaneous and metastatic tumors due to the human MDA-MB-231 breast cancer cell line [25]. Consistent with previous preclinical studies, DMXAA led to massive hemorrhagic necrosis in subcutaneously grown breast cancer and NSCLC cell line tumors. In contrast, neither autochthonic lung adenocarcinomas arising in K-ras LA1/+ transgenic mice [26], nor metastases derived from intracardiac injections of syngeneic 344SQ-ELuc NSCLC cells showed responses to DMXAA administration. In addition, we found that clodronate liposome-mediated macrophage depletion [27] abrogated DMXAA-induced intra-tumoral hemorrhagic necrosis in 344SQ-ELuc subcutaneous tumors.
Although the macrophage-derived factors thought to mediate the anti-tumor effects of DMXAA are characteristic of M1 polarized cells, we found that the murine NSCLC TAMs were primarily M2-like. Thus, the responses of M1 and M2 macrophages to DMXAA were investigated. We found that M2 polarized bone marrow-derived macrophages, as well as M2-like TAMs could be re-polarized to an M1 phenotype by DMXAA. We further found that the endogenous STING ligand, 2939-cGAMP, produced a similar repolarization phenotype. TAM reeducation represented a plausible mechanism whereby M2-like TAMs were able to mediate vascular disruption in response to DMXAA. Our results lend support to the idea of using STING agonists as TAM repolarizing agents, and they also highlight the importance of testing agents on a variety of preclinical models. In addition, our study highlights the growing awareness of the utility of GEMMs for preclinical drug studies [28,29].

NSCLC TAMs and Clodronate Liposome-mediated TAM Depletion
Consistent with the evidence that macrophages play an important role in the action of DMXAA [11,30], we found that 344SQ-ELuc subcutaneous tumors contained large numbers of infiltrating macrophages, as detected by ionized Ca 2+ -binding adaptor (Iba)-1 immunostaining. These cells were concentrated primarily at the tumor periphery ( Figure 1A), and the vast Figure 1. Effectiveness of clodronate-mediated TAM depletion varied depending on tumor site. Representative 344SQ-ELuc subcutaneous tumor sections were stained with antibodies against Iba-1 (A) and Arg-1 (B), demonstrating abundant M2-like macrophages primarily at the tumor periphery. In contrast, much lower and variable macrophage infiltration was present in either the K-ras LA1/+ primary NSCLCs (C) or in 344SQ-ELuc metastases (D). (E) MTT assay conducted on BMDM (Mw) or 344SQ-ELuc cells to assess potential cytotoxicity in response to either Clodrolip (Clod) or empty liposomes (EL). (F) Representative 344SQ-ELuc subcutaneous tumor and kidney metastases sections from Clod-treated mice were stained with Iba-1 showing that TAM depletion only occurred in subcutaneous tumors (T = tumor, K = kidney). Scale bar = 100 mm (A-B) 50 mm (C, D, F). Data represent the mean 6 SEM. doi:10.1371/journal.pone.0099988.g001 majority of the TAMs were positive for the murine M2 marker, arginase (Arg)-1 ( Figure 1B). In contrast, spontaneously arising Kras LA1/+ lung adenocarcinomas contained a relative paucity of TAMs, with only a few scattered cells located at the tumor periphery ( Figure 1C). To obtain metastases, 344SQ-ELuc NSCLC cells were introduced via intracardiac, left-ventricle (LV), injection in syngeneic mice. This route reproducibly resulted in multiple kidney, adrenal gland, lung, and visceral fat pad metastases ( Figure S1). These metastases contained variable levels of macrophage infiltration ( Figure 1D) but at considerably lower levels than were observed in the 344SQ-ELuc subcutaneous tumors.
To investigate the role of macrophages in mediating DMXAA vascular disruption, we depleted macrophages using clodronateencapsulated liposomes (Clodrolip) [27]. A comparison of the effects of in vitro Clodrolip treatment of 344SQ-ELuc cells and bone marrow-derived macrophages (BMDM) demonstrated an ,100-fold greater sensitivity of macrophages ( Figure 1E). In vivo, Clodrolip led to an ,50% decrease in CD11b + F4/80 + marrow cell populations ( Figure S2A-B), and although we obtained ,100% depletion of TAMs within subcutaneous 344SQ-ELuc tumors, no evidence of TAM depletion was seen in 344SQ-ELuc metastases ( Figure 1F). TAMs can play a supportive role during tumor development, and consistent with this, we found that Clodrolip-mediated TAM depletion slowed the growth of 344SQ-ELuc subcutaneous tumors while growth of metastases was unaffected, presumably due to the inability to obtain macrophage depletion ( Figure S2C-F).

Depletion of Macrophages in Subcutaneous 344SQ-ELuc Tumors Alters the Response to DMXAA
Consistent with previous reports [22], 344SQ-ELuc NSCLC subcutaneous tumors respond dramatically to DMXAA, with a marked (,2-logs) decrease in bioluminescence (BLI) signals post- Note that Clod plus DMXAA treated tumors did not exhibit intra-tumoral hemorrhage, but rather showed thrombosis and hemorrhage confined to larger feeding vessels (yellow asterisk). Data represent the mean 6 SEM. doi:10.1371/journal.pone.0099988.g002 drug injection (Figure 2A-B). This was accompanied by vascular thrombosis and hemorrhage in the tumor periphery, and by the development of extensive central necrosis ( Figure 2C). The drop in BLI following DMXAA treatment was not due to direct tumor cell toxicity since DMXAA had no detrimental effect on 344SQ-ELuc cell viability ( Figure S3). Instead, tumor BLI signal loss was attributable to greatly diminished blood, and hence luciferin substrate, perfusion which would diminish ATP-dependent light production. While decreased perfusion could conceivably have resulted from reversible vasoconstriction, given the massive tumor necrosis observed, it was more likely that decreased light emission was the result of tumor vessel thrombosis and rupture.
Since TAMs could be efficiently depleted by Clodrolip in the 344SQ-ELuc subcutaneous tumors, we evaluated the responses of these tumors to DMXAA. Mice treated with empty liposomes (EL) served as the control group. DMXAA still provoked a dramatic drop (,2-logs) in bioluminescence signal intensity ( Figure 2D), however, this occurred in the absence of the characteristic intratumoral thrombosis and hemorrhage. Instead, DMXAA led to massive non-hemorrhagic, 'dry', necrosis ( Figure 2E). The latter appeared to result from DMXAA-induced occlusion of tumorfeeding vessels ( Figure 2F). Although it was conceivable that Clodrolip had sensitized the feeding vessels to DMXAA, this result suggested the VDA effects of DMXAA were not confined to the intra-tumoral vasculature, but rather that tumor-feeding arterioles had also been compromised by this agent. Indeed, DMXAA may have a direct effect on endothelial cells [3,[7][8][9]. While it is plausible that clodronate treatment may have sensitized the tumor vasculature, for example, by increasing vessel fragility or interfering with pericyte coverage, we have found that large tumor-feeding vessel thrombosis is also present after DMXAA administration to non-Clodrolip exposed mice (data not shown). The results suggest that DMXAA mediates its effects in two ways: by causing macrophage-induced intra-tumoral microvessel disruption and by causing thrombosis of tumor-feeding vessels. Regardless, it was clear that subcutaneous 344SQ-ELuc tumor TAM depletion was effective in preventing DMXAA-induced intra-tumoral hemorrhagic thrombosis.
Both DMXAA and the Non-canonical Cyclic Dinucleotide 2939-cGAMP are able to Re-educate M2 Macrophages towards an M1 Pro-inflammatory Phenotype Since the intra-tumoral hemorrhagic response of subcutaneous 344SQ-ELuc tumors to DMXAA was dependent on TAMs, we next examined the effect of DMXAA directly on macrophages. DMXAA is known to stimulate the production of cytokines, such as TNFa, CXCL10 (IP-10), MIP-1a, and MIP-1b, [7,10], in order to mediate vascular disruption, and these factors are typical of M1 macrophage responses. However, we found that the majority of 344SQELuc TAMs were M2-like ( Figure 1A Table 1 (for the raw data see Table S1). Although the majority of the up-regulated cytokines were characteristic of M1-polarized cells (e.g. CXCL10, CXCL9, IL-1a and b), we also found that compared to the M1 cells, the IL-4 polarized M2 macrophages displayed 2-10 fold greater inductions of these factors in response to DMXAA, indicative of their reeducation towards an M1 phenotype [31][32][33]. Furthermore, analysis of RNA transcripts provided additional evidence that M2 macrophages had been shifted towards an M1 phenotype, as demonstrated by decrease in the M2 markers, Arg-1 and Fizz1, and acquisition of the M1 markers, iNOS and IL-12p40 ( Figure 3A). The repolarizing effect of DMXAA was even evident at relatively low concentrations (e.g. 5 mg/ml) of this agent ( Figure 3B). Consistent with STING-TBK1 pathway activation [34], by reverse-phase protein array, we found that DMXAAmediated upregulation of the NF-kB pathway as shown by increased p65 phosphorylation in M2 macrophages ( Figure S4B).
Since one of the targets of DMXAA is murine STING, we tested another STING agonist, the non-canonical endogenous cyclic dinucleotide 2939-cGAMP [35,36] on M2 macrophages. As was the case with DMXAA, we found that 2939-cGAMP administration to M2-polarized macrophages dramatically increased interferon-b, as well as expression of M1 markers iNOS and IL-12p40, and this was accompanied by decreased expression of the M2 markers, Arg-1 and Fizz1 ( Figure 3C-D). Together, our results using these chemically distinct agonists indicated that STING activation was the key factor responsible for the observed M2-to-M1 macrophage re-education. RNA transcripts from spleens of mice treated in vivo with DMXAA also demonstrated induction of iNOS and downregulation of Arg-1 ( Figure 4A), as well as diminished anti-Arg-1 immunohistochemical staining ( Figure 4B). Importantly, subcutaneous tumor lysates also demonstrated evidence of DMXAA-mediated repolarization ( Figure 4C), with diminished Arg-1 staining being evident as early as 6 hours post-DMXAA exposure ( Figure 4D). In summary, these results suggest that STING activation can mediate M2-like TAM re-education.

Discordant Effects of DMXAA on 344SQ-ELuc Subcutaneous Versus Metastatic Tumors
In contrast to the dramatic results obtained using 344SQ-ELuc subcutaneous tumors, DMXAA treatment of 344SQ-ELuc metastases yielded no decrease in photon emission rates ( Figure 5A-B), with the tumors remaining histologically similar to controls after this treatment ( Figure 5C). To confirm this effect was not peculiar to the 344SQ-ELuc cell line, we also compared the effects of DMXAA on subcutaneous versus metastatic tumors generated in MDA-MB-231-Luc2 human breast cancer xenografts. Similar to 344SQ-Eluc tumors, the MDA-MB-231-Luc2 model demonstrated dramatic hemorrhagic necrosis of subcutaneous tumors but not bone metastases ( Figure 6).
Subcutaneous tumors grow rapidly up to ,1 cm 3 , whereas the multiple metastases generated by 344SQ-ELuc cells are not able of reach this size owing to the lethal tumor burden that would ensue. Therefore, we evaluated the effect of DMXAA on subcutaneous 344SQ-ELuc tumors having a size (,2 mm 3 ) comparable to that of the metastases. As with the large subcutaneous tumors, DMXAA administration to mice with small subcutaneous tumors still led to ,2-log decreases in photon emission at both 6 and 24 hours ( Figure 5D-E). This was also accompanied by the development of pathology similar to that of the large subcutaneous were also taken from triplicate samples of M2-polarized macrophages exposed to 20 or 40 mg/ml 2939-cGAMP plus LF2000 for 6 and 24 hours in vitro. LF2000 alone served as the control (designated as 'M2 alone' in the graphs). 2939-cGAMP led to down-regulation of Arg-1 and Fizz1, and dramatic increases in iNOS and IL-12p40 expression in a dose-dependent manner. (D) IFN-b induction provided an indication of STING activation in response to 2939-cGAMP, with strong inductions at 6 hours that returned to baseline by 24 hours (*p,0.05, **p,0.01, ***p,0.001). doi:10.1371/journal.pone.0099988.g003 tumors ( Figure 5F). Thus, differences in tumor volume did not account for the differential effects of DMXAA on subcutaneous versus metastatic tumors.

Primary Pulmonary Adenocarcinomas Arising in K-Ras la1/+ Transgenic Mice Show No Response to DMXAA
We next evaluated the effects of DMXAA on spontaneously arising primary NSCLCs in the K-ras LA1/+ transgenic model [26] of lung cancer. Again, in contrast to the dramatic results seen with the subcutaneous tumors, DMXAA treatment of K-ras LA1/+ mice (,150 days old) produced no discernable histological effect on the lung adenocarcinomas ( Figure 7A). It was possible that the variable level of macrophage infiltration amongst different tumor sites may have accounted for the inconsistent responses to DMXAA. Thus, while primary K-ras LA1/+ lung tumors have even fewer numbers infiltrating macrophages ( Figure 1C) than systemic metastases, subcutaneous tumors show abundant infiltrates of macrophages and neutrophils (data not shown) in the tumor periphery ( Figure 7B). Thus, the level of TAM infiltration could be one potential variable determining whether DMXAA will cause vascular disruption and hemorrhagic necrosis.

Primary NSCLC Tumors Exhibit Increased Evans Blue Dye Permeability
DMXAA is known to target the unstable, leaky vasculature of tumors [3,[7][8][9], thus, we investigated whether diminished permeability might be a potential factor rendering the primary tumor vasculature resistant to DMXAA. Using a modified Miles assay, we injected mice with the Evans blue dye to look for interstitial leakage in K-ras LA1/+ lung adenomas and adenocarcinomas, and in 344SQ-ELuc subcutaneous tumors. Primary lung neoplasms (adenomas and adenocarcinomas) demonstrated similar dye permeability as subcutaneous tumors ( Figure 8A). Frozen sections of K-ras LA1/+ lung ( Figure 8B), and subcutaneous tumors ( Figure 8C) demonstrated similar dye leakage into tumors at both sites (see also Figure S5). Hence, the possibility that vessel impermeability to small molecules such as DMXAA was a factor accounting for the inability of this agent to cause vascular disruption in the primary tumors was discounted.

Subcutaneous and Metastatic 344SQ-Eluc Tumors Exhibit Differences in Vasculature Structure
Since differences in tumor vascular bed structure could be a factor accounting for the differential responses to DMXAA, we compared subcutaneous 344SQ-ELuc tumors and p53 R172HDg/+ Kras LA1/+ lung adenocarcinoma-derived spontaneous metastases using a 3D-microcomputed tomography (micro-CT) imaging quantification method we previously described [22]. Interestingly, 3D renderings of 344SQ-ELuc subcutaneous tumors and p53 R172HDg/+ K-ras LA1/+ metastases displayed significant differences with respect to overall appearance ( Figure 9A), vessel thickness ( Figure 9B), and most strikingly, with respect to internal avascular areas, as demonstrated by a sphere-filling computational technique [22] ( Figure 9C). Quantification of vessel parameters demonstrated a significant drop in vessel density (VV/TV) ( Figure 9D) as well as overall vessel number (V.N) and connectivity (Conn.D) in subcutaneous tumors ( Figure S6). Interestingly, there was a similar pattern in vessel thickness (V.Th) when compared as a percentage of vessels present, although the metastases had larger-diameter vessels ( Figure 9E). Vessel separation (V.Sp), on the other hand, demonstrated a significant difference between the two different tumor sites, with subcutaneous tumors having markedly less small-diameter spheres (an indication of well vascularized areas) ( Figure 9F), thus confirming the presence of larger avascular/ischemic areas in the subcutaneous tumors (as visualized by the red spheres in Figure 9C). Thus, it was plausible that differences in the structure of the tumor vascular network between the different tumor sites might also contribute to the differential responses to DMXAA.

Discussion
Our findings using both DMXAA and 2939-cGAMP suggest that STING activation was the common factor leading to M2 macrophage re-polarization, a process that undoubtedly played a role in mediating the vascular disrupting effects of DMXAA we observed on the subcutaneous 344SQ-ELuc tumors. Interestingly, however, we found that the vascular disrupting effects of DMXAA on subcutaneous tumors did not extend to either the 344SQ-ELuc metastases obtained following intracardiac injection of these cells, or to the spontaneously arising tumors in the K-ras LA1/+ GEMM model of NSCLC. Indeed, the majority of successful pre-clinical studies evaluating the utility of DMXAA were carried out in subcutaneous tumor models, with relatively few studies examining the effects of DMXAA on tumors in other anatomical sites [37,38]. Echoing our results, experiments employing a different vascular disrupting drug, flavone acetic acid (FAA), vessel disruption was seen in subcutaneous tumors, but not systemic tumors [39]. Thus, although the inability of DMXAA to activate human STING provided an obvious reason for failure of DMXAA in human cancer trials [20,21], our results nevertheless suggest that vascular disruption might not occur in either primary or metastatic human NSCLC if human STING agonists were administered. With regard to correcting this defect, considerable efforts are now underway involving the development of stable cyclic dinucleotide analogs that will allow human STING activation [40].
Differences in the density of TAM infiltration amongst different tumor sites may have been one of several factors accounting for the differential effects of DMXAA on subcutaneous versus 344SQ-ELuc metastases. Supporting the key role of infiltrating TAMs, we found that their depletion in subcutaneous 344SQ-ELuc tumors prevented DMXAA-induced intra-tumoral hemorrhagic necrosis. In the case of 344SQ-ELuc subcutaneous tumors, TAMs were present as a dense rim at the tumor periphery and were thus well positioned to support DMXAA-induced vascular disruption via the production of pro-inflammatory mediators. There were extensive regions of ischemic necrosis invariably present within the subcutaneous 344SQ-ELuc tumors, signifying the availability of macrophage activating environmental factors such as hypoxia and the Toll-like receptor (TLR) 4-activating protein, high mobility group, HMGB1 protein [41]. Thus, in addition to quantitative differences in macrophage infiltration density between tumor sites, there may have been substantial qualitative differences between the TAMs at different tumor sites.
Macrophages are inherently plastic, with the M1-and M2polarized phenotypes representing the extremes of a spectrum [33,42,43]. Although TAMs are often M2-like, there is now evidence that such TAMs can be re-polarized towards an M1 phenotype that can inhibit tumor growth [31,32,44,45]. Herein, we show that the STING agonists DMXAA and 2939-cGAMP are both able to repolarize M2-polarized marrow-derived macrophages in vitro, and we also provide evidence that DMXAA is able to shift M2-like macrophages towards an M1-like phenotype in vivo. The latter results were in agreement with a recent study reporting an M2-like to M1 shift in TAM populations in response to DMXAA treatment that was also were accompanied by an antitumor effect [46]. Intra-tumoral TAM repolarization would provide a source of pro-inflammatory cytokines and chemokines, and reduce the levels of vascular endothelial growth factor, effects that could contribute towards either vascular disruption, or stabilization, respectively. TAM re-education with STING agonists may play a role in other important processes, including promotion of anti-tumor adaptive immune responses that are dependent on type I interferons and dendritic cell activation [47,48].
The sensitivity of tumor vasculature to DMXAA is thought to be due to the immature and irregular vascular patterning within tumors [2,7]. To produce subcutaneous tumors, large numbers of cancer cells are implanted and these divide rapidly, rendering them sensitive to chemotherapeutic agents [49]. Growth of cells introduced in this manner may outstrip the angiogenic capacity of the host, leading to the development of large regions of ischemia and necrosis that can promote macrophage infiltration and activation. In both the 344SQ-ELuc metastases and the autochthonic NSCLC tumors there were no areas of necrosis, in contrast, subcutaneous 344SQ-ELuc tumors contained large necrotic regions and were densely populated with macrophages. It is possible that the ischemia and necrosis that typifies subcutaneous tumors renders their vessels more susceptible to DMXAA. In addition, it is plausible that compared to vessels derived from other vascular beds, dermal vasculature-derived tumor vessels are inherently unstable, and hence more vulnerable to DMXAA. Co-option of mature vessels may be a feature of both systemic metastases and primary lung adenocarcinomas, and such vessels, like other normal vascular beds in the animal, would be predicted to be refractory to the vascular disrupting effects of DMXAA. Our finding of structural differences in the vasculature between tumors at different anatomical sites lends some support the latter notion, an idea that was reinforced by the apparent inability of clodronate liposomes to cause macrophage deletion in 344SQ-ELuc metastases. Indeed, the latter observation provided a functional indication of structural differences between subcutaneous tumors and metastases. In summary, and as depicted in Figure S7, the features of the subcutaneous 344SQ-ELuc tumor vasculature may render them susceptible vascular disrupting effects of DMXAA. In contrast, the 344SQ-ELuc metastases, having arisen not only within different vascular beds, but importantly, from a very much smaller initial tumor colonizing cell numbers, contain vessels that are resistant to this agent.
In view of the evidence that DMXAA, acting via its effects on TAMs and/or dendritic cells, has the capacity to augment adaptive cytotoxic T cell anti-tumor activity, reviewed in [48], it would be of considerable interest to determine whether chronic administration of STING agonists might similarly lead to spontaneous immunity against 344SQ-ELuc metastases and primary lung adenocarcinomas. The fact that 344SQ-ELuc metastases do not undergo hemorrhagic necrosis in response to DMXAA would actually be of benefit in this setting, since it would allow anti-tumor immunity to be readily quantified via changes in photon emission rates.
Recent studies suggest GEMMs may be able to more faithfully mimic their human counterparts, not only with respect to genetic alterations, but also in their ability to predict responses to therapy [28,29,50,51]. Thus, while primary orthotopic or subcutaneous models are useful for initial drug screening, new agents also need to be evaluated using a range of preclinical models before their use in humans is contemplated [49]. Interestingly, we found that metastases resulting from intra-cardiac injection of 344SQ-ELuc cells failed to respond to DMXAA, suggesting that GEMMderived cell lines might serve as effective surrogates for the corresponding slowly growing autochthonous cancers. Regardless of whether or not STING agonists are ultimately found to cause vascular disruption in human cancer, the potential for such agents to repolarize TAMs will render them useful additions to the anticancer armamentarium.

Mice
Male 129/Sv mice (6-12 week old) were used for syngeneic tumor studies. Transgenic mice harboring the p53 R172HDg/+ and Kras LA1/+ mutations were kindly provided Dr. G. Lozano (University of Texas) [23,26,52]. Mice were maintained on standard mouse chow (Pico-Vac Lab Mouse Diet #5062), and housed in a specific pathogen-free barrier facility with ethics approval from the University of Calgary Animal Care Committee (protocols M10063 and M08112) and in accordance with Canadian Council on Animal Care guidelines.

Tumor Generation and Drug Treatments
To generate subcutaneous tumors, 5610 5 344SQ-ELuc cells in 100 ml PBS were injected in both posterior flanks of mice. Metastatic tumors were generated via intracardiac injection of 1610 4 344SQ-ELuc cells in 100 ml PBS into the left ventricles (LV) of mice anaesthetized with 100 mg/kg ketamine and 6 mg/ kg xylazine, given intra-peritoneally (i.p.). Tumor growth was monitored every 2-4 days via BLI. Mice were shaved to minimize light signal attenuation. Once tumors were established (day 10 for systemic metastases; day 7 or day 14 for subcutaneous tumors), mice were given 25 mg/kg of DMXAA (D5817, Sigma-Aldrich), or DMSO vehicle by i.p. injection. BLI was carried out at 6 and 24 hours as previously described [22]. K-ras LA1/+ mice, aged ,150 days received 25 mg/kg DMXAA or DMSO and were sacrificed 6 hours later. TAM depletion was carried out via i.p. injection of Clodrolip (2 mg/20 g mouse) starting at day 21 prior to tumor cell inoculation, and was maintained with 1 mg/20 g mouse given every 3 days for the duration of the experiment. Empty liposomes (EL) served as the controls [27].

Reverse Phase Protein Array (RPPA)
M2-polarized (40 ng/ml IL-4 for 48 hours, N = 3) macrophages were treated with 20 mg/ml DMXAA or DMSO vehicle for 30 min. Cells were then lysed and protein denatured in SDS buffer and samples sent for RPPA analysis (RPPA Core Facility, University of Texas, MD Anderson Cancer Center, Houston, TX). Differential abundance of various proteins and/or their phosphorylation status in response to DMXAA was assessed.

Vascular Permeability
Vessel permeability in mice with subcutaneous 344SQ-ELuc tumors and aged K-ras LA1/+ mice (,200 days old) was examined using a modified Miles Assay. Mice were given a 4 ml/g dose of 2% Evans Blue dye (Sigma) made up in 0.9% saline by i.p. injection. After 5 hours, mice were perfused with PBS followed by 4% PFA using a Masterflex C/L perfusion pump (Cole-Parmer). Tissues were harvested and lungs were inflated via a tracheal cannula prior to sinking them in 30% sucrose/PBS and OCT embedding (Tissue-Tek). Whole tissues were imaged on a Stemi SV 11 dissection microscope (Zeiss). Samples were cryosectioned at 10 mm and imaged on an Axio Imager A2 (Zeiss) under a Cy5 filter.

Macrophage Polarization and Supernatant Cytokine Assay
BMDM were seeded in 6-well plates at 2610 6 cells/well and polarized for 48 hours with the addition of 50 ng/ml LPS (List Biological Labs) and 50 ng/ml IFNc (Cedarlane) for M1, or 40 ng/ml IL-4 (R&D Systems) for M2 at 37uC in a 5% CO 2 humidified atmosphere. Cells were re-plated in triplicate in 96-well plates, 8610 5 cells/well, in media containing 20 mg/ml DMXAA or DMSO control for 24 hours. Supernatants were subjected to a mouse 32-plex cytokine/chemokine discovery array (Luminex) (EVE Technologies, Calgary, Alberta).

RT-PCR
Subcutaneous tumors and control spleens were snap frozen in liquid N 2 prior to being homogenized in QIAzol lysis reagent (Qiagen). M1 and M2 polarized macrophages were treated with 20 mg/ml DMXAA (or dose response) or DMSO vehicle for an additional 24 hours. In addition, M2 polarized macrophages were treated with 20 mg/ml or 40 mg/ml 2939-cGAMP (InvivoGen) in the presence of Lipofectamine-2000 (LF2000, Invitrogen). Cells were lysed with QIAzol lysis reagent (Qiagen) and RNA was extracted with chloroform and isopropanol. To make cDNA, 1 mg of RNA was treated with DNAse (Promega) followed by RT-PCR with 10 mM dNTPs, random primers (Roche) and Superscript II reverse transcriptase (Invitrogen). Real-time PCR of cDNAs was carried out on the LightCycler using the LightCycler FastStart DNA MasterPLUS SYBR Green kit (Roche). Data were normalized to b-actin mRNA. Primer sequences were as previously described [44,54,55].

Statistical Methods
Samples were compared using a student's t-test, with Welch's correction on samples with unequal variance. P values of ,0.05 were considered significant.