STING Sensing of Murine Cytomegalovirus Alters the Tumor Microenvironment to Promote Antitumor Immunity Free

As research continues to focus on discovering novel cancer therapies, the approach of using pathogens to alter the immune microenvironment has emerged as an exciting new strategy. With the U.S. Food and Drug Administration approval of the oncolytic HSV-1 T-VEC in 2015, pathogens, such as vaccinia, Semliki Forrest virus, Toxoplasma gondii, Salmonella, and CMV, have all been tested in preclinical models and have the potential to directly lyse tumor cells and/or alter the tumor environment (1–12). In addition to using pathogens to alter the tumor microenvironment experimentally, research has also focused on how the naturally occurring microbiome in human tumors may positively and negatively alter patient outcomes (13–15). Indeed, both viruses and bacteria have been shown to naturally alter the tumor microenvironment in addition to being used as a potential immunotherapy.

Human CMV (HCMV) is a ubiquitous β–herpes virus that has been repeatedly identified in human tumors including glioblastoma, nonmelanoma skin cancer, colorectal adenocarcinoma, rhabdomyosarcoma, and ovarian cancer (16–20). It is not thought that HCMV initiates tumor development (21). However, the impact of HCMV on tumor growth has been widely debated, and studies conflict as to whether HCMV may have a positive or negative impact on tumor progression (16–20, 22, 23). Thus, it is important to evaluate the effect of CMV in experimental tumor models and define the mechanisms by which CMV alters tumor behavior.

In addition to its potential to alter tumor progression, CMV has also been proposed as a vaccine vector for infectious diseases and cancer (8, 24–41). CMV naturally generates a massive CD8⁺ T cell response in the host that can inflate over time because of blips of reactivation (42). In multiple models, including melanoma, prostate cancer, and head and neck squamous cell carcinoma, murine CMV (MCMV) has been successfully used as a vaccine to generate immunity toward tumor Ags encoded in the viral backbone (8, 34–41). However, the potential for such a therapy depends heavily on how CMV naturally alters the tumor environment. Indeed, there has been experimental evidence that latent MCMV may promote tumor growth and invasiveness in models of breast cancer and glioblastoma, with both of these studies citing increased angiogenesis as partially responsible for the effect (43, 44). In contrast, MCMV infection in models of melanoma and lymphoma can cause tumor regression (8–12). We recently showed that MCMV, when injected intratumorally (i.t.) into a growing B16-F0 melanoma, slowed progression of the lesion, altered the immune compartment of the tumor, and synergized with blockade of the immune checkpoint PD-L1 to clear established tumors and promote long-term immune memory (8). The efficacy of MCMV approximately correlated with the period of time during which the virus is active in the tumor and could be extended by increasing the number of viral injections (9). The virus was found to recruit macrophages via a viral chemokine and infect the macrophages in the tumor. In vitro work also demonstrated that MCMV increased inflammatory transcripts in anti-inflammatory or M2-like macrophages. However, this work did not address the mechanism by which MCMV infection alters tumor-associated macrophages (TAMs), nor did we show that infection of tumor-infiltrating macrophages, per se, was important for the therapeutic effect.

Although the mechanisms by which MCMV activates macrophages have not been determined in the mouse model, it has been shown that HCMV can affect the inflammatory state of the human monocytes through activating the PI3K pathway upon entry into cells (45, 46). Additionally, HCMV is known to activate STING in monocytes to promote type I IFN production (47). In the current study, we show that MCMV activates macrophages through the STING pathway. Expression of STING was critical for MCMV to delay the growth of B16 melanomas, an effect that was most likely the result of type I IFN production in the tumor leading to the production of chemokines that recruited T cells to the tumor environment. Most importantly, we found that infected macrophages were sufficient to recruit CD8⁺ T cells and delay tumor growth in a STING-dependent manner. These data demonstrate that an active MCMV infection delays tumor growth in this model through activation of STING.

C57BL/6J, Tmem173^gt/gt (C57BL/6J-Tmem173gt/J, referred to as STING ^gt//gt), and Tmem173^−/− (B6(Cg)-Tmem173^tm1.2Camb/J, referred to as STING ^−/−) were purchased from The Jackson Laboratory. Ifnar1^−/− mice back-crossed to B6 (48) were a gift from Dr. Thomas Moran (Mount Sinai School of Medicine, New York, NY). Tlr9^−/− (B6.129-Tlr9tm1Aki/Obs) mice were produced by Dr. S. Akira (Osaka University, Osaka, Japan) (49) and generously provided by Dr. Robert Finberg (University of Massachusetts, Worcester, MA). A mix of male and female mice were used for all studies. All mice, on a B6 background, were bred at Thomas Jefferson University from original breeders and used at an age of 6–16 wk. Because of differences in breeding schedules, data from STING^−/− mice (Tmem173^−/−) and STING^gt/gt (mice lacking functional STING, Tmem173^gt/gt) were combined to form the STING-deficient groups for some experiments. This is noted in the figure legends. No notable differences in tumor growth, survival, or doubling time were observed between the STING^−/− and STING^gt/gt animals (Supplemental Fig. 1). The Institutional Animal Care and Use Committee at Thomas Jefferson University reviewed and approved all protocols.

Bone marrow–derived macrophages (BMDMs) were harvested from the femur and tibia of C57BL/6J, Tlr9^−/−, Ifnar1^−/−, STING^gt/gt, and/or STING^−/− animals by flushing the bones with bone marrow macrophage media consisting of DMEM with 10% L929-conditioned media, 10% FBS, and 1% penicillin–streptomycin. The L929-conditioned media were produced by plating 7.2 × 10⁵ L929 cells in T-150 flasks. Supernatant was harvested after 7 d, replaced with fresh media, and harvested again after 14 d. All L929-conditioned media were combined and frozen for later use. Bone marrow was strained through a 70-μm filter after harvest and plated onto nontissue culture–treated petri dishes to facilitate macrophage recovery from the plastic. After 7 d, macrophages were replated at 6 × 10⁵ cells per well in a six-well plate. M2-like macrophages were polarized using 40 ng/ml IL-4 on day 8. On day 9, half of the IL-4–treated BMDMs were infected with K181 MCMV (Fig. 1) or the GFP⁺ SL8-015 MCMV (Fig. 3) at a multiplicity of infection of five for 24 h. M1-like macrophages were also polarized on day 9 by adding 20 ng/ml of IFN-γ and 1 μg/ml LPS. M0 wells were left untreated. Twenty-four hours postinfection, all macrophages were harvested for analysis. All wild-type (WT) MCMV (K181 and SL8-015) were grown as previously described (50).

BMDMs from C567BL/6J, STING^gt/gt, and Ifnar1^−/− animals were harvested and grown as described above. After 7 d, cells were replated and treated with 40 ng/ml IL-4 to produce M2 macrophages or left untreated to produce M0 macrophages. After 24 h, M0 and M2 macrophages were treated with a 1:1 mix of IFN-α and IFN-β in increasing concentrations from 0 to 500 U without removing initial polarizing conditions (Supplemental Fig. 2) to determine the optimal dose of 500 U. The cells were treated with IFN for 24 h before processing the samples for RNA using the RNeasy Plus Mini Kit (QIAGEN). The experiment was repeated with the optimal 500 U dose of type I IFN (Fig. 2).

When B16-F0s were treated with IFN (Fig. 5), 3 × 10⁵ B16-F0s were plated in six-well plates 24 h prior to treatment with IFN. Cells were either treated with 500 U of IFN or left untreated. After 24 h with IFN, the live B16-F0s were counted manually using trypan blue and a hemocytometer and harvested for RNA using the RNeasy Plus Mini Kit (QIAGEN).

B16-F0 cells were purchased from the American Type Culture Collection, grown in DMEM with 1% penicillin–streptomycin and 10% FBS, and frozen in a large batch of aliquots between passage 5 and 13. In all experiments, cells derived from this frozen batch were thawed, passaged once more, and implanted within 7 d. Injected cells were always pigmented, and a batch was certified negative for mycoplasma and other pathogens by IMPACT III testing on 10/30/2017 by IDEXX BioResearch. For tumor implantation, B16-F0s were resuspended in HBSS and injected s.c. in the shaved right flank of the animal, as described previously (51). Tumor growth was monitored by measuring length and width with a six-inch digital caliper (Neiko). When tumors reached 20 mm² in area, they were directly injected with spread-defective ΔgL-MCMV or PBS as a control using an insulin syringe with 5 × 10⁵ PFU of virus in 50–70 μl of PBS every other day for a total of three injections. Spread-defective MCMV (ΔgL, based on the K181 backbone) has been previously characterized and was grown as previously described (52). After i.t. injections, tumors were monitored until they reached 100 mm².

For BMDM transfers (Fig. 8), macrophages from C57BL/6 animals were harvested and cultured as described above. For timing purposes, bone marrow was generally harvested on the same day or on the day before tumors were first implanted into STING^gt/gt animals. When tumors had reached ∼20 mm², BMDMs were polarized to M2-like macrophages for 24 h and then infected with spread-defective MCMV (ΔgL) or left untreated. After another 24 h, 4 × 10⁴ M2 or infected M2 macrophages were transferred i.t. into the B16-F0 melanomas in STING^gt/gt animals. The number of macrophages for transfer were based on macrophage counts in immunofluorescent images extrapolated to the whole tumor volume. After the single injection of macrophages, tumors were monitored to an end point of 100 mm².

Tumors were harvested on day 5 and homogenized in RPMI 1640 containing 10% FBS and 1% penicillin–streptomycin, using a gentleMACS Octo Dissociator (Miltenyi Biotec) on m_lung_01_01 and m_lung_02_01settings. Cells were separated with Lymphoprep (STEMCELL Technologies) according to the manufacturer’s instructions and stained for sorting with Zombie UV Fixable Viability Dye (BioLegend) as well as Abs specific for CD45.2 (clone 104; BioLegend), CD11b (clone M1/70; BioLegend) and Gr-1 (clone RB6-8C5; BioLegend), CD4 (clone RM4-5; BioLegend), CD19 (clone 1D3; eBiosciences), and NK1.1 (clone PK136; BioLegend). Representative FACS plots from sorting are shown in Supplemental Fig. 4.

RNA was extracted from treated BMDMs (Figs. 1, 2, Supplemental Fig. 2), B16-F0s (Fig. 5A), whole tumor homogenate (Fig. 6A), or cells sorted from the tumor (Fig. 6D). Tumor homogenate was processed by pushing tumors through a 70-μm filter to obtain a single-cell suspension before lysing cells for RNA extraction. RNA from all samples was isolated using the RNeasy Mini Kit (QIAGEN), and cDNA was produced using a high-capacity cDNA reverse transcription kit (Applied Biosystems). Transcripts were detected using iTaq Universal SYBR Green with the following primers: Gapdh, forward (for) 5′-tgtccgtcgtggatctgac-3′ and reverse (rev) 5′-cctgcttcaccaccttcttg-3′; Cxcl9, for 5′-cttttcctcttgggcatcat-3′ and rev 5′-gcatcgtgcattccttatca-3′; Cxcl10, for 5′-gctgccgtcattttctgc-3′ and rev 5′-tctcactggcccgtcatc-3′; Tnf, for 5′-tcttctcattcctgcttgtgg-3′ and rev 5′-ggtctgggccatagaactga-3′; Il6, for 5′-tctaattcatatcttcaaccaagagg-3′ and rev 5′-tggtccttagccactccttc-3′; Ifn-nona4, for 5′-aagctgtgtgatgcaacaggt-3′ and rev 5′-ggaacacagtgatcctgtgg-3′; and IFNG, for 5′-gcaaaaggatggtgacatga-3′ and rev 5′-ttcaagacttcaaagagtctgaggta-3′. The transcript reported as Ifna was collected using the IFN-nonα4 primers, which amplify all subtypes of IFN-α except for IFN-α4 because of deviations in sequence homology. All samples were run on the Bio-Rad Laboratories CFX96 Touch Real-Time PCR Detection System. In macrophage studies, transcripts were normalized to the housekeeping gene GAPDH and compared with uninfected M2 macrophages to obtain a ΔΔCT value. Data are expressed as fold change over M2 macrophages using 2^−(ΔΔCT). For whole tumor homogenates, transcript concentrations were normalized to GAPDH in each sample and compared with the averages of the PBS-treated samples in each experiment to obtain a ΔΔCT value. For sorted cells, transcript concentrations were normalized to GAPDH and displayed as relative expression.

BMDMs were harvested from C57BL/6, STING^gt/gt, and Ifnar1^−/− animals and cultured as described above. After resting in culture for 7 d, 100,000 BMDMs were plated in triplicate for each treatment condition in a 96-well black microplate. BMDMs were polarized to M2-like macrophages using 40 ng/ml IL-4 for 24 h. After 24 h, M2-like BMDMs were infected with GFP-expressing SL8-015 MCMV at a multiplicity of infection of five or left untreated. M1-like macrophages were also polarized at this time using 20 ng/ml IFN-γ and 1 μg/ml LPS. After 24 h, the phagocytic potential of the cells was determined using pHrodo Red Escherichia coli BioParticles (Invitrogen) using the protocol provided. After 1 h, the BioParticles were removed, and the cells were gently washed once with PBS. Next, 560/585 readings were taken to determine the phagocytic potential of the cells using a Molecular Devices SpectroMax M2 microplate reader. Immediately after, wells were imaged using a Nikon Eclipse Ti confocal microscope to view both phagocytosis and infection percentage in the cells. Images were analyzed with ImageJ (https://fiji.sc/) (53).

Tumors were rapidly frozen in OCT compound, and tumors were cut into 15–20-μM sections using a Leica CM3050 S Cryostat. Sections were placed in cold acetone for 10 min, rehydrated with TBS for 20 min, and blocked with blocking buffer (TBS plus 3% BSA and 0.1% Tween 20) for 20 min. Sections were stained in blocking buffer for 1 h with Abs from BioLegend specific for F4/80 (clone BM8), CD11b (clone M1/70) and DAPI (Fig. 6B), CD8α (clone 53-6.7), CD11b (clone M1/70), CD45.2 (clone 104) and DAPI (Fig. 7A), or F4/80 (clone BM8), CD11b (clone M1/70), and CD8α (clone 53-6.7) and DAPI (Fig. 8C). Samples were imaged using the Nikon A1R fluorescent confocal microscope. The macrophage number per square millimeter image was calculated from the count of F4/80⁺, CD11b⁺ macrophages per image from six to eight images per tumor and two to three tumors per treatment (Fig. 6C). The CD8⁺ T cell number per square millimeter image was calculated from the count of CD8⁺, CD45.2⁺, CD11b⁻ cells per image from five to eight images per tumor and two to three tumors per treatment group in Fig. 7B or from the count of CD8⁺CD11b⁻ cells per image from 9 to 14 images per tumor and two tumors per treatment group in Fig. 8D. Images were analyzed with ImageJ (https://fiji.sc/) (53).

Statistics were performed using GraphPad Prism v6 or R (v3.3.1 R-project.org). If normally distributed (as assessed by the Kolmogorov–Smirnov test), data were analyzed using a two-tailed t test. If data were nonnormally distributed, a Mann–Whitney U test was performed instead. For tumor doubling times, the tumor measurements were log transformed, and a linear regression was used to model the rate of tumor growth as a function of time. Finally, a log-rank (Mantel–Cox) test was used to compare Kaplan–Meier survival curves.

Previous data from our laboratory shows that i.t. injections of WT MCMV can cause tumor growth delay in a B16-F0 melanoma model (9). This antitumor effect was dependent on the recruitment of macrophages, many of which were infected in the tumor environment. Additionally, viral infection of the macrophages in vitro increased proinflammatory transcripts in M2-polarized macrophages. However, it remains unclear how MCMV altered macrophages and whether this pathway was required for virus-induced tumor growth delay.

Sensing of CMV postinfection can occur through multiple pattern recognition receptors but is most often attributed to TLR9, which senses viral DNA in the endosome, and the cGAS/STING pathway, which senses viral DNA in the cytosol (47, 54–58). Both of these pathways, when activated by CMV, produce a strong type I IFN response (47, 55). However, in human monocytes infected with HCMV, STING is required for induction of type I IFN (47). To test the importance of these pathways in MCMV activation of macrophages, we harvested BMDMs from Tlr9^−/−, STING-deficient (both knockout and golden ticket), and Ifnar1^−/− mice along with WT C57BL/6J controls. Macrophages were polarized to the M2-like state for 24 h to mimic the polarization state of newly arrived TAMs and then infected with MCMV. In line with our previous data, MCMV induced proinflammatory cytokine production in the M2-polarized B6 macrophages, including significant induction of mRNA for Il6, Tnf, Ifna, Cxcl9, and Cxcl10 (Fig. 1A). Loss of TLR9 did not alter the production of these cytokines. However, STING-deficient animals were markedly impaired in the production of Il6, Tnf, and Ifna upon infection (Fig. 1B). Moreover, loss of the type I IFNR IFNAR phenocopied the loss of STING, suggesting that STING function was dependent on release of type I IFN. Thus, STING and IFNAR signaling play a critical role in MCMV-mediated induction of inflammatory cytokines in macrophages.

The loss of MCMV-induced inflammatory cytokine production from both Ifnar1^−/− macrophages and STING-deficient macrophages indicated that IFNAR signaling may be playing a key role secondary to CMV infection. Therefore, we wished to determine whether type I IFN alone, via paracrine signaling, could activate production of these cytokines in macrophages. To test this, we added type I IFN alone (a 1:1 mixture of IFN-α and IFN-β) to unpolarized macrophages (M0) or macrophages that had been polarized toward an M2-like state for 24 h. The concentration of type I IFN was determined by titration experiments (Supplemental Fig. 2). One day after the addition of type I IFN, M0 and M2 macrophages expressed increased levels of Il6, Tnf, Cxcl9, and Cxcl10. In the M0 group, Il6, Cxcl9, and Cxcl10 production was significantly increased, whereas Tnf, Cxcl9, and Cxcl10 production was significantly increased in the M2-polarized group (Fig. 2A). STING-deficient BMDMs also followed a similar pattern (data not shown). In contrast, however, Ifnar1^−/− animals showed no response to the addition of type I IFN, as expected (Fig. 2B). Thus, type I IFN alone is sufficient to induce the production of the inflammatory cytokines observed after MCMV infection.

Because activation of macrophages, as well as active effects from CMV proteins, can lead to changes in phagocytosis and acidification of phagolysosomes (59), we next determined whether STING-deficient and Ifnar1^−/− mice displayed altered phagocytosis and acidification postinfection. BMDMs were again polarized to an M2-like state and then infected 24 h later, this time with a GFP-expressing MCMV. Phagocytic capacity was assessed 24 h later by engulfment of beads coated in a pH-sensitive dye that fluoresces upon entering the phagosome. Infection of WT B6 macrophages by MCMV enhanced phagocytosis of the beads. However, loss of STING did not alter this effect (Fig. 3A, 3B), indicating that the increased phagocytosis occurred in a STING-independent manner. Interestingly, the Ifnar1^−/− macrophages displayed higher phagocytosis activity in the absence of infection, and the addition of MCMV did not increase this capacity further. Importantly, the proportion of infected macrophages was unaffected by the loss of STING or IFNAR (Fig. 3A and data not shown). Thus, differences in infection and phagocytosis also cannot account for the changes in inflammatory cytokine production by B6, STING-deficient, and Ifnar1^−/− BMDMs following MCMV infection. Collectively, these data show that MCMV infection enhances inflammatory cytokine production in a STING- and type I IFN–dependent manner but enhances macrophage phagocytosis independently of the STING pathway.

Having shown that MCMV activated macrophage inflammatory cytokine production via STING and type I IFN, we wished to determine whether loss of these molecules would impact tumor growth delay in the B16-F0 model. To this end, we implanted B16-F0s s.c. into Tlr9^−/−, Ifnar1^−/−, and STING-deficient animals. When tumors had reached 20 mm², i.t. injections of PBS or spread-defective ΔgL-MCMV were administered every other day for a total of three injections. The spread-defective ΔgL-MCMV was used to avoid any enhanced viral replication in the absence of critical viral sensing pathways. Importantly, we have previously shown that ΔgL-MCMV delayed tumor growth equally to WT MCMV (9). In agreement with our previous data, injection of ΔgL-MCMV delayed tumor growth (Fig. 4A), significantly increased survival (Fig. 4B), and significantly increased tumor doubling times (Supplemental Fig. 3) when tumors were implanted in WT B6 mice. Consistent with our results from above, loss of TLR9 did not alter the ability of the virus to delay tumor growth, increase survival, or increase tumor doubling times. However, lack of STING completely prevented MCMV from delaying tumor growth (Fig. 4A), increasing survival (Fig. 4B), or increasing tumor doubling time (Supplemental Fig. 3). Interestingly, Ifnar1^−/− animals displayed some growth delay after injection of ΔgL-MCMV, although it was diminished compared with WT B6 animals (median survival after i.t. MCMV treatment: 18 d in B6 mice versus 11 d in Ifnar1^−/− mice). These data show that STING signaling in the host is necessary for MCMV to delay tumor growth in the B16-F0 model.

Because tumor growth delay was completely lost in STING-deficient mice but only partially impaired in Ifnar1^−/− mice, we speculated that type I IFN, which would still be produced in Ifnar1^−/− mice, might directly affect the B16 tumor cells. To test this, we added 500 U of type I IFN to B16-F0 cells in vitro. Indeed, Cxcl9 and Cxcl10 were significantly increased in B16-F0 cells after exposure to type I IFN with an upward trend in Il6 and Tnf production as well (Fig. 5A). Importantly, we observed no obvious impact on B16 growth or survival after 24 h of type I IFN exposure (Fig. 5B). Thus, we speculate that type I IFN can still alter the inflammatory environment of the tumor by directly affecting the B16-F0 tumor cells.

Our data suggest that STING-induced type I IFN was important for inducing inflammatory cytokine production. To test whether loss of STING and IFNAR altered the cytokine environment in vivo, tumors were harvested 1 d after the final i.t. control PBS or MCMV injection (day 5 overall) from B6, STING-deficient, and Ifnar1^−/− animals. Transcripts for Il6, Tnf, Ifng, and Cxcl9 were increased in all tumors after MCMV infection, indicating their independence from STING or IFNAR in vivo (Fig. 6A). In contrast, STING-deficient animals failed to upregulate transcripts for Cxcl10, indicating that production of this chemokine required STING in the host. Transcripts for type I IFN itself were difficult to detect consistently in these samples, perhaps reflecting the fact that samples were assayed 5 d after the first MCMV injection (Fig. 6A and data not shown). Importantly, macrophages were recruited normally to all tumors regardless of the presence or absence of STING (Fig. 6B, 6C). This result is consistent with our recent study showing that MCMV recruits macrophages to the tumor via the viral chemokine MCK2, which should be independent of host-derived inflammatory cytokines and chemokines (9, 60–62). Next, we sorted CD11b⁺Gr-1⁺ monocytic phagocytes from pooled tumors implanted in WT or STING-deficient mice and assessed their expression of transcripts for Tnf, Ifna, and Cxcl10. In each case, transcript levels were further compared with those expressed by lymphocytes sorted from the same tumors. Notably, we did not observe significant differences in the expression of Ly-6C or MHC class II between WT and STING-deficient monocytic phagocytes at this timepoint (data not shown), both of which were similar to data shown in our recent report (9). As in the total tumor (Fig. 6A), sorted monocytic phagocytes expressed comparable levels of Tnf, irrespective of STING, and Ifn transcripts were again inconsistently detectable (Fig. 6D). However, Cxcl10 transcripts were significantly reduced in the absence of STING, implying a reduction in the amount of type I IFN available in the tumor (Fig. 6D). Interestingly, Cxcl10 transcription in STING-deficient monocytic phagocytes was still significantly increased over the tumor-infiltrating lymphocytes (CD4⁺, CD19⁺, NK1.1⁺) from the same animals, indicating some amount of production of this chemokine. PBS-treated tumors could not be used for comparison because of a general lack of leukocyte infiltration into the lesions.

We have previously shown that i.t. MCMV infection results in the accumulation of CD8⁺ T cells in the tumor and that these T cells are involved in the growth delay (8, 63). Thus, the lack of Cxcl10 expression in the tumor and by TAMs led us to investigate whether CD8⁺ T cells were recruited to the tumor normally in STING-deficient mice. Indeed, there was a significant reduction in the number of CD8⁺ T cells recruited to the tumors in STING-deficient mice (Fig. 7A, 7B). These data imply that STING deficiency impairs inflammatory cytokine and chemokine production, resulting in a failure to engage T cells after MCMV infection.

To this point, our data show that STING is necessary to induce inflammatory cytokine production by infected macrophages and that STING deficiency prevents MCMV from delaying tumor growth. To definitively test whether STING signaling in MCMV-infected macrophages was sufficient to induce tumor growth delay, we infected M2-polarized WT or STING-deficient macrophages with ΔgL-MCMV and injected these macrophages into tumors growing in STING-deficient mice. Because ΔgL-MCMV can produce only noninfectious particles in these infected macrophages (52), this protocol restricts infection to the injected cells. Moreover, because the recipients are STING-deficient, any particles engulfed by host macrophages will be unable to activate STING. As a comparison, mice received uninfected M2 macrophages. In all cases, the number of macrophages injected was based on the counts of macrophages obtained from histological images, which were then extrapolated to the entire tumor volume (see 2Materials and Methods). Critically, transfer of uninfected macrophages did not significantly alter tumor growth (median survival = 7 d) compared with tumors injected with PBS alone (median survival = 8 d; compare Fig. 8B to Fig. 4A). Strikingly, however, a single injection of infected WT macrophages delayed the growth of tumors in STING-deficient animals (Fig. 8A) and significantly increased survival (Fig. 8B) compared with injection of uninfected B6 macrophages. Importantly, this growth delay was lost when tumors received STING-deficient macrophages that were infected by MCMV (Fig. 8A, 8B). Moreover, whereas transfer of infected WT macrophages was sufficient to recruit CD8⁺ T cells into the tumor, this effect was lost when the infected macrophages lacked STING (Fig. 8C, 8D). These data show that MCMV infection of macrophages is sufficient to induce CD8⁺ T cell recruitment and antitumor effects in the B16-F0 model via STING signaling and that infection of TAMs alone can tilt the tumor immune environment toward tumor control.

CMV infects most cells in the body (64, 65), including tumor cells, and it is has been identified in multiple human tumors (16–20), although the precise cells infected in the tumor and the net effect of the virus are still unknown. It has been hypothesized that the virus may lead to increased tumor aggression (16, 43, 44) which can either enhance or hinder targeted therapeutic approaches (66, 67). Additionally, CMV itself has shown promise as a potential vaccine vector (8, 34–41). Thus, it becomes increasingly important that we understand how the virus interacts with its surroundings. Our data show that an active WT MCMV infection can engage antitumor immune responses by recruiting macrophages to the tumor (9) and acting as a STING agonist. As HCMV is also known to both recruit myeloid cells to the site of infection and activate STING in myeloid cells, this mechanism may also be translationally relevant (47, 68–70).

Despite some successful preclinical studies, STING agonists have not yet been successful in the clinic. The first STING agonist to go through clinical trials, DMXAA, failed because of low specificity for human STING (71). Testing of newer compounds is still under way, alone or in combination with checkpoint inhibitors (72–75). However, the conditions under which such agonists might successfully promote antitumor immunity are unclear. Indeed, multiple tumors mutate the STING pathway and the immune environment in the tumor varies greatly, ranging from immune deserts that lack significant immune cell infiltration to immune-replete tumors. Given our previous work showing that i.t. MCMV therapy depended on viral recruitment of myeloid cells to the tumor (9), our data suggest that the composition of the immune cells in the tumor may greatly influence the success of STING agonists.

It is worth noting that pathogen-based STING agonists have been explored in some settings as well. In fact, inactivated modified vaccinia virus Ankara has shown promise as a STING agonist in both murine and human tumor models (7). Although many groups are focused on using vaccinia as an oncolytic agent, inactivating vaccinia removes any effect of viral immune modulatory mechanisms, thereby enhancing the immune stimulation and the antitumor effect (7). The ability of both CMV and vaccinia to target the immune response and not the tumor directly may be particularly valuable because it will not be impacted by tumor-associated mutations within immune stimulatory pathways. Moreover, these approaches may provide synergy with other immune therapies. We previously showed that i.t. MCMV synergized with a PD-L1 blockade to clear B16-F0 tumors (8), which are normally resistant to blockade of the PD-1 pathway (76). Likewise, the Deng laboratory demonstrated that inactivated modified vaccinia virus Ankara was synergistic with the blockade of PD-1, PD-L1 or CTLA-4 (7). These data fit with other work showing that STING agonists can enhance the effect of checkpoint inhibition (77–85), with one group reporting that STING can sensitize tumors otherwise resistant to PD-1 blockade (86).

The data shown in this study indicate that STING-sufficient macrophages infected with ΔgL-MCMV produced an antitumor response in STING-deficient animals. Moreover, our previous work revealed that the recruitment of myeloid cells via the MCMV-encoded chemokine MCK2 was critical for MCMV to induce antitumor immunity (9). Thus, STING in the tumor environment was not sufficient without monocytic phagocytes in the tumor, and we infer that the recruitment of myeloid cells by the virus primed the tumor to respond to a STING agonist. Although we explored the impact of infected macrophages on tumor growth in the current study (Fig. 8), it is important to note that other phagocytic cells may also contribute to the antitumor effects after direct injection of the virus, and STING may be necessary in these cells as well. Future work will explore these possibilities.

Other groups have shown that STING promotes antitumor responses through production of type I IFN in the immune compartment (87). Our data show that inflammatory cytokine expression was impaired in Ifnar1^−/− BMDMs and that type I IFN alone was sufficient to activate chemokine and cytokine production by macrophages and the tumor cells themselves in vitro. Thus, we hypothesize that the effect of type I IFN on the tumor cells is to mediate the partial response to therapy in the IFNAR knockout mice (Fig. 4). However, the complete lack of antitumor effects in the absence of STING may argue that STING expression by tumor-associated myeloid cells (or indeed the presence or absence of myeloid cells in the tumor) may be critical for determining whether type I IFN is produced in sufficient quantities after STING activation. Further studies are therefore warranted with STING agonists in the presence or absence of tumor-associated myeloid cells and, when the tumor cells themselves express or lack the type I IFNR, to precisely define the sources and effects of IFN in the tumor.

Interestingly, our data also showed a lack of Cxcl10 transcript in tumors and sorted TAMs in the absence of STING and an impaired ability of these tumors to recruit CD8⁺ T cells after MCMV infection. Our previous work has shown that CD8⁺ T cells are necessary for full therapeutic effect of the virus (8, 9). Therefore, a major long-term goal is to understand the interactions between macrophages and CD8⁺ T cells in this system. The MCMV infection in this model resulted in rapid tumor growth delay that was typically evident within a few days of viral injection. Such kinetics are not consistent with new priming of tumor-specific CD8⁺ T cells as a result of activation of professional APCs, but rather, they are consistent with reactivation, recruitment, or retention of antitumor T cells that were primed previously when the tumor was injected (88, 89). Therefore, it is possible that STING contributes in three places: 1) STING signaling and type I IFN may serve to induce chemokines that attract pre-existing CD8⁺ T cells into the tumor, where they begin to kill their targets; the CXCL10/CXCR3 axis is an obvious target and we would hypothesize that CXCR3-deficient mice will not receive the benefit from i.t. MCMV infection; 2) type I IFN may support T cell function upon arrival in the tumor by altering the concentrations of inflammatory and suppressive cytokines and/or improving Ag presentation and APC activation within the tumor environment; and 3) STING may contribute to the initial priming of tumor-specific T cells at the tumor after the tumor was implanted but prior to MCMV infection. These mechanisms are not mutually exclusive, and future work will be required to tease apart the mechanisms that govern the interplay between macrophages, CD8⁺ T cells, and the tumor in this system.

In the context of how CMV alters the tumor environment, it is important to note that we only determined the effects of an active MCMV infection in a tumor model in which macrophages were the dominant cell type infected. With multiple groups showing that latent HCMV in the tumor may promote tumor aggression, future work will need to investigate the effects of active or latent CMV infection in multiple models. We hypothesize that CMV may promote different effects, depending on the viral activity, tropism, and tumor location in the body (90). We further hypothesize that myeloid cell recruitment and STING activation will only be evident during an active infection and possibly only when hematopoietic cells are present or recruited and can be infected. In contrast, when HCMV is hypothesized to promote tumor aggressiveness, it is generally thought to be caused by a latent infection or a relatively inactive virus, and it is not clear which cells in the tumor might be infected. This raises the possibility that approaches to reactivate latent virus already in a tumor may lead to STING activation and the promotion of antitumor immunity. In addition, because MCMV can increase myeloid cells and CD8⁺ T cell numbers in the tumor, any infection or reactivation of virus in the tumor could fundamentally alter immunologically “cold” tumors, possibly enabling efficacy of additional immune therapies. In fact, our laboratory has shown previously that PD-L1 blockade combined with i.t. MCMV cleared up to 70% of established B16-F0 melanomas, a response rate much greater than either therapy alone (8). However, it must be noted that inflammation in the local tissue may be problematic in some tumors such as glioblastoma.

In summary, our data suggest that an active CMV infection may profoundly alter the tumor microenvironment by recruiting myeloid cells and activating the STING pathway, thereby engaging the adaptive immune system. These basic mechanistic studies raise the question of whether hematopoietic cell infiltration will be critical to the function of STING therapies and shed light on one way that CMV can alter the tumor microenvironment.

Immunofluorescent images were captured at the Sidney Kimmel Cancer Center Bioimaging Facility (NCI 5 P30 CA-56036) with assistance from Dr. Maria Yolanda Covarrubias. Additional microscopy assistance was provided by Tiago Monteiro-Brás.

C.M.S. has a financial interest in UbiVac CMV for the development of spread-defective CMV-based therapeutics. Neither the funding bodies nor UbiVac CMV had any role in the design of experiments or interpretation of the data. The other authors have no financial conflicts of interest.