Intracellular Activation of Complement C3 Leads to PD-L1 Antibody Treatment Resistance by Modulating Tumor-Associated Macrophages

The complement system of the innate immune system acts to remove pathogens and trigger release of inflammatory cytokines (1). Although the role of the complement system as an effector system to kill cancer cells is known, evidence suggests that various complement molecules are enriched in the tumor microenvironment (TME) and they facilitate tumor progression through modulating angiogenesis and altering the function of cancer cells or immune cells. For example, C1q (2) and C5a (3) are angiogenic, C7 enhances the stemness of cancer stem cells (4), and C5a enhances the growth of tumors by promoting chronic inflammation (3). Lambris and colleagues found that C5a inhibits the antitumor immune response by recruiting myeloid-derived suppressor cells (MDSCs; refs. 5, 6). Inspired by their findings, our laboratory and others demonstrated that blocking C5a signaling could enhance the efficacy of PD-1/PD-L1 antibody treatment (7, 8). In addition, He and colleagues demonstrated that C3a and C5a could dampen the actions of tumor-infiltrating CD8⁺ T cells through inhibiting autocrine IL10 production (9). These results illustrate that the complement system alters the TME and tumor immunity.

Complement C3 acts as the component of complement activation. Under physiologic conditions, the main source of systemic C3 is hepatocytes. However, almost all cell types can express C3 locally, including myeloid, lymphocytic, fibroblastic, and epithelial cells (10). Evidence suggests that locally produced C3 is involved in tissue regeneration (11), differentiation of MDSCs (12), graft-versus-host disease (13), survival maintenance of naïve CD4⁺ T cells (14), and intestinal tissue damage during mesenteric ischemia (15). Although some evidence suggests C3 found in the TME was generated systemically (16), other evidence shows that C3 produced by tumor-infiltrating CD8⁺ T cells could repress the antitumor response (9). Comparable C3 deposition was observed when ID8-VEGF cells were implanted into C3^−/− or wild-type (WT) recipient mice (16). Moreover, tumor cell–derived C3 was reported to promote cancer cell proliferation (17) and the leptomeningeal metastasis of lung cancer (18). These results indicate that tumor cells might produce sufficient functional C3 protein for TME construction. Nevertheless, the role of tumor cell–derived C3 in antitumor immune suppression remains to be explored.

In this study, we demonstrated that intracellular activation of complement C3 suppressed the infiltration and function of CD8⁺ T cells by promoting the accumulation and immune-suppressive activity of tumor-associated macrophages (TAMs) in a C3aR-dependent, but C5aR-independent, manner. The binding of C3a with its receptor, C3aR, in TAMs activates PI3Kγ signaling and thus promotes TAM immunosuppressive activity. Our data showed that blocking tumor-derived complement C3 is sufficient to remove obstacles that could prevent antitumor efficacy in checkpoint blockade of treatment-resistant tumor models.

CT26 cells (2012), B16F10 cells (2012), 4T1 cells (2014), RAW264.7 (2009), LLC cells (2012), EL4 cells (2012), A549 (2008), CaCO2 (2008), and 3T3 (2009) cells were obtained from the American Type Culture Collection. MC38 was kindly provided by Liufu Deng (Shanghai Jiao Tong University, 2015). All cell lines were examined and authenticated by short tandem repeat profiling in September 2015. All cell lines were Mycoplasma negative and used within 10 passages. All cells were cultured as indicated previously (8). To collect cell supernatants, cells were seeded in 6-well plates at a density of 1 × 10⁶ cells/mL in complete DMEM medium. The cell supernatant was collected 24 hours later, filtered through a 0.22-μm filter and stored at −80°C.

Six- to 8-week-old female C57BL/6, BALB/c, and NU/NU mice were purchased from the Chinese Academy of Medical Sciences (Beijing, China). C3^−/− mice were purchased from The Jackson Laboratory. C3a receptor–deficient (C3aR^−/−) mice on a C57BL/6 background were kindly provided by Dr. Rick A. Wetsel (The University of Texas). C5aR^−/− mice on a C57BL/6 background were obtained from Dr. Craig Gerard (Harvard Medical School). The animal studies have been conducted in accordance with the guidelines of the Institutional Animal Care and Use Committee of the Third Military Medical University.

The C3 concentration in cell supernatants was measured using a mouse complement C3 ELISA kit (Abcam, ab157711) according to its instructions.

A pHBAd-U6-gRNA-CMV-Cas9-T2A-zsgreen plasmid was obtained from Hanbio Biotechnology (China). A Cas9 guide sequence for mouse C3 (5′ GGATGTCACCCTGAGCATCG 3′) was designed using the online program (http://chopchop.cbu.uib.no/), and oligos were synthesized by Hanbio Biotechnology (China). The pHBAd-U6-gRNA-CMV-Cas9-T2A-zsgreen was digested with BbsI, and the annealed oligos were cloned into this plasmid. To make the adenovirus, the pHBAd-U6-gRNA-CMV-Cas9-T2A-zsgreen (with cloned sgRNA) was cotransfected into HEK293 cells with the packaging plasmids h-SNCA and pHBAd-BHG. For C3 silencing, CRISPR control or CRISPR sgRNA targeting C3 viruses were transduced into CT26 or LLC cells. Single clones were generated by the limited-dilution method, and knockout clones were selected by ELISA analysis for the lack of C3 in the cell supernatant.

Synergistic activation mediator (SAM) technology was applied to overexpress C3. Briefly, we first generated B16F10 cell lines with stable CAS9-VP64 expression by lentivirus transfection and then screened with media containing puromycin. B16F10- CAS9-VP64 cells were transfected with a lentivirus coding a Cas9 guide sequence for mouse C3 (5′ AGCAGGTACTTTCAAGCTCC 3′). The cells were selected by G418. All lentivirus used were purchased from Shanghai GeneChem (China).

Tumor cells (for CT26, LLC, and 4T1: 5 × 10⁵ in 100 μL phosphate-buffered saline; for B16F10: 3 × 10⁴ in 100 μL phosphate-buffered saline) were injected subcutaneously into the flank of mice at day 0. For depletion of CD4⁺ T or CD8⁺ T cells, mice were injected intraperitoneally with anti-CD4 (200 μg; clone GK1.5; Bio X Cell) or anti-CD8a (200 μg; clone 2.43; Bio X Cell) on days 5 and 9 after the tumor challenge. For macrophage depletion, 200 μL PBS-liposomes or clodronate liposomes were injected intraperitoneally on days 7, 10, and 13 after the tumor challenge. For PD-L1 blockade, anti–PD-L1 (200 μg; clone 10F.9G2; Bio X Cell) or control IgG2b (200 μg; clone LTF-2; Bio X Cell) was injected intraperitoneally on days 4, 7, 10, and 13 after the tumor challenge. For the pharmacologic blockade of C3aR or C5aR signaling, the C3aR antagonist (C3aRa; SB290157, Cayman; 1 mg/kg daily) or the C5aR antagonist (C5aRa; PMX53, GL Biochem; 1 mg/kg daily) was injected intraperitoneally into C57BL/6 mice from day 1 to day 14. The tumor volume was measured with an electronic caliper every 2 to 3 days in two dimensions (length and width). Excised tumors were weighed and processed for flow cytometry, histology, or isolation of RNA and proteins as indicated.

LLC cells (5 × 10⁵ in 100 μL phosphate-buffered saline) were injected subcutaneously into the flanks of WT or C3aR^−/− mice at day 0. On day 21 after tumor challenge, mice were sacrificed. TAMs from WT or C3aR^−/− mice were enriched by magnetic isolation using EasySep Mouse CD11b Positive Selection Kit II (STEMCELL) according to its manufacturer. Then, TAMs were isolated by FACS. Then, LLC cells (5 × 10⁵) were coinjected with or without WT or C3aR^−/− TAMs (5 × 10⁵) subcutaneously into the flanks of WT mice. Tumor growth was monitored every 3 to 4 days.

CT26 tumors were removed and digested into single-cell suspension. CD11b⁺ cells were enriched by using a CD11b⁺ cell isolation kit (Stemcells). Cells (2 × 10⁵) were then treated with 1 μg/mL rmC3a. Then, cells were fixed and permeabilized using Transcription Factor Phospho Buffer Set (BD) according to its instructions, followed by staining with CD11b (clone M1/70; BioLegend), F4/80 (clone BM8; BioLegend), and phospho-Akt (Ser473; D9E; CST).

CT26 cells (5 × 10⁵ in 100 μL phosphate-buffered saline) were injected subcutaneously into the flank of mice at day 0. IPI-549 (PI3Kγ inhibitor) was dissolved at 5% 1-methyl-2-pyrrolidinone in polyethylene glycol 400 as described before (19). Mice were treated with IPI-549 (15 mg/kg body weight, 100 μL in volume, daily) or vehicles (100 μL in volume, daily) were treated by oral gavage from day 7 to day 17.

To obtain single-cell suspensions, tumors were harvested at the indicated time points and then cut into pieces and digested with 1 mg/mL collagenase I (Gibco) and 1 mg/mL Dispase II (Roche) for 45 minutes at 37°C. Cells were blocked with anti-FcR (clone 93; BioLegend) and then stained with antibodies from BioLegend to PD-L1 (clone 10F.9G2), CD45 (clone 30-F11), CD11b (clone M1/70), Gr-1 (clone RB6-8C5), F4/80 (clone BM8), CD11c (clone N418), CD206 (clone C068C2), I-A/I-E (clone M5/114.15.2), CD8a (clone 53-6.7), CD4 (clone GK1.5), CD3 (clone 17A2), CD140a (clone APA5), and CD45 (clone 30-F11). For IFNγ, TNFα, and IL10 staining, cells were stimulated in vitro with a cell stimulation cocktail (plus protein transport inhibitors) (eBioscience) for 4 hours. Cells were then processed using a fixation and permeabilization kit (BD Bioscience) and stained with antibodies from eBiosciences to IFNγ (clone XMG1.2), TNFα (clone MP6-XT22), and IL10 (clone JES5-16E3). For intracellular Ki67 or Foxp3 staining, cells were processed using a fixation and permeabilization kit (eBioscience) and stained with antibodies against Ki67 (clone B56; BD) or Foxp3 (clone MF-14; BD). Incorporation of 5-ethynyl-2′-deoxyuridine (EdU) was measured using the Click-iT EdU flow cytometry assay kit according to the manufacturer's instructions (Invitrogen). For the in vivo experiments, tumor-bearing mice (day 10 PI) were injected i.p. with 50 μg/g body weight EdU and sacrificed 30 hours later. For the in vitro experiment model, 10 mmol/L EdU was added into the cell culture media. Two hours later, cells were processed and analyzed. All samples were acquired with a CantoII flow cytometer (BD) and analyzed with FlowJo software (TreeStar). An Aria II flow cytometer (BD) was used for cell sorting.

Total RNA was prepared from murine cancer cell lines or tumors using TRIzol reagent according to the manufacturer's instructions, and then reverse transcribed using random hexamers to generate cDNA (Takara). qPCR was performed with SYBR Premix Ex Taq II (Takara) to quantify the relative expression of mRNA. Primers for real-time PCR are listed in Supplementary Table S1. For RNA sequencing, libraries were prepared using an MGIEasy mRNA kit and sequenced by a BGISEQ-500 instrument. Upon sequencing, raw FASTQ files were aligned using a HISAT aligner with default parameters. Aligned fragments were then counted and annotated using the Bowtie2 transcript database. Normalized RSEM were obtained. All P values were adjusted for multiple testing using the Benjamini and Hochberg FDR algorithm.

A normalized mRNA expression data set for human colon adenocarcinoma was downloaded from the cBioPortal for cancer genomics and used to evaluate the correlation between C3 and immunosuppressive cell marker and immunosuppressive cytokine transcript levels (20, 21). This data set includes mRNA profiles for 270 colon tumor samples and was downloaded in December 2016. Spearman correlation analyses were conducted. Differences were considered significant at P < 0.05.

Tumor cells were seeded in a glass bottom cell culture dish (NEST) and cultured overnight. Cells were fixed in 4% paraformaldehyde for 20 minutes at room temperature, followed by permeabilization with 0.3% Triton X-100 and blocked with 1% BSA for 30 minutes at room temperature. For detection of C3a, a primary antibody against C3a (clone I87-1162; BD) was used. For the detection of mouse C3b/iC3b/C3c (clone 2/11; Hycult Biotech), Brefeldin A (1:1,000; BioLegend) was added into culture media for 6 hours before staining, and then the sections were incubated with primary antibodies at 4°C overnight. FITC goat anti-Rat IgG (H + L; 1:1,000; Abcam) was then applied as a secondary antibody. For the detection of human C3b/iC3b/C3c (clone bh6; Hycult Biotech), this antibody was used. FITC goat anti-Mouse IgG (H + L; 1:1,000; Abcam) was used as a secondary antibody. For the staining of human/mouse α-tubulin, α-Tubulin Rabbit Polyclonal Antibody (Beyotime) was used. FITC goat anti-Rabbit IgG (H + L; 1:1,000; Abcam) was used as a secondary antibody. Nuclei were stained with DAPI for 5 minutes. Stained sections were photographed using an Olympus fluorescence microscope.

This experiment was conducted as described previously (22, 23). Briefly, TAMs were isolated by FACS. CD8⁺ T cells were isolated from a tumor-free spleen of BALB/C mouse using the CD8a⁺ T-Cell Isolation kit (Stemcells). Then, isolated T cells were labeled with 5 μmol/L carboxyfluorescein diacetate succinimidyl ester (CFSE; Invitrogen). CFSE-labeled T cells (10⁵) were cultured in a 96-well plate, precoated with anti-CD3ϵ (3.5 μg/mL, clone 145-2C11; BioLegend), and soluble anti-CD28 (1.5 μg/mL, clone 37.51; BioLegend) was added to the medium to induce T-cell proliferation either with or without TAMs at the indicated ratios. Three days (∼72 hours) later, cells were collected and analyzed by flow cytometry.

All data were expressed as the means ± SEM and were analyzed using either two-tailed unpaired Student t test or other statistical methods indicated in the text with GraphPad Prism 7.0 software. For each parameter of all data, *, P < 0.05; **, P < 0.01; ***, P < 0.005.

RNA-sequencing data have been deposited to NCBI (GEO: GSE120274).

To dissect the contribution of individual cell types to C3 production in the TME, we first fractionated a CT26 tumor mass into four populations by flow cytometry (FACS). These four populations comprise implanted tumor cells (CD45⁻PDGFRα⁻), cancer-associated fibroblasts (CAF, CD45⁻PDGFRα⁺), myeloid cells (CD45⁺CD11b⁺), and lymphocytes (CD45⁺CD11b⁻; Fig. 1A). For each population, C3 mRNA expression was measured by quantitative PCR. The results indicated that CAFs and infiltrated lymphocytes expressed little C3, whereas tumor cells and myeloid cells expressed more C3. Although myeloid cells expressed more C3 (∼1.5-fold) than tumor cells (Fig. 1B), in the tumor mass, the number of myeloid cells was less than 1/5 of the number of tumor cells (Fig. 1A), suggesting that tumor cells are the main source of intratumoral C3. To determine whether CT26 tumor cells are unique, we surveyed a series of widely used murine tumor cell lines for C3 production. We found that MC38 colon cancer cells produced little C3, whereas B16F10 melanoma and EL4 T lymphoma cells, as well as other cell lines, including 4T1 breast cancer cells, CT26 colon cancer, and LLC lung cancer cells, produced more C3 (Fig. 1C). A previous report suggested that the intracellular activation of complement C3 in human CD4⁺ T cells sustains T-cell homeostasis (10). To determine whether the intracellular activation of complement C3 also occurs in cancer cells, we assessed the presence of intracellular C3a and C3b/iC3b/C3c in cultured CT26 and LLC cancer cells. We used two antibodies that recognize a C3a or a C3b/iC3b/C3c neoepitope on the cleaved fragment. We found that both C3a and C3b/iC3b/C3c were present in cultured CT26 and LLC cancer cells (Fig. 1D). We also detected intracellular C3b/iC3b/C3c in cultured human A549 and CaCO2 cancer cells (Fig. 1E). This suggests that tumor cell–derived C3 is functional as it is activated intracellularly.

To determine the contribution of tumor cell–derived C3 to tumor growth, we engineered C3-deficient CT26 (CT26-Sg-C3) and LLC cells (LLC-Sg-C3) using CRISPR-Cas9 technology. After validating the efficiency of C3 deletion by ELISA (Fig. 1F), we transplanted CT26 cells subcutaneously into C3-sufficient immunocompetent syngeneic BALB/c mice. Deletion of C3 in CT26 cells delayed tumor growth to a great extent (Fig. 1G). Similarly, retardation of tumor growth in syngeneic C57BL/6J hosts was observed in LLC-Sg-C3 cells, a tumor model that produces high levels of C3 (Fig. 1H). Previously, silencing C3 expression was reported to impair ovarian cancer cell proliferation by its cleaved products C3a and C5a, two potent effectors of the complement system (17). Therefore, the reduced growth of CT26-Sg-C3 and LLC-Sg-C3 tumors observed could reflect the effect of C3 on tumor cell proliferation and survival. We found that C3-deficient and C3-sufficient cell lines had the same proliferation rate (Supplementary Fig. S1A–S1C). Furthermore, we used EdU incorporation assays to examine the proliferative capacity of C3a and/or C5a on tumor cells. Consistent with loss-of-function experiments, the addition of exogenous C3a and/or C5a did not affect proliferation (Supplementary Fig. S1D). These data indicate that the impact of C3 on tumor growth in immunocompetent mice was not intrinsic to tumor cells. Our result was consistent with reports (18, 24) that exogenous C3a does not promote the proliferation of LLC tumor cells, at least not in an anchorage-dependent manner. One possible extrinsic mechanism to explain this is that complement could regulate T-cell immunity, which consequently affects tumor growth (9, 25). To examine this, we performed similar tumor transplantation experiments in nude mice that lack T cells. The absence of T cells in recipient mice completely abolished the difference in growth between C3-deficient and C3-sufficient tumors (Fig.1I and J). Taken together, these results suggest that C3 complement produced by tumor cells is a component of the immunosuppressive TME, and its immunosuppressive mechanism acts upon T cells.

To determine whether systemic C3 could compensate for tumor cell–derived C3, we injected LLC-Sg-C3 or LLC-Sg-Con cells into C3 knockout (C3-KO) or WT mice. Consistent with previous findings, systemic C3 deficiency reduced tumor growth by about 25% (Fig. 1K). We found that there was no difference in tumor growth when LLC-Sg-C3 cells were implanted in WT and C3-KO mice, suggesting that systemic C3 could not compensate for the deficiency of C3 in tumor cells (Fig. 1K). Taken together, these results suggest that although C3 is present in tumor cells at lower concentrations than found in sera, tumor-derived C3 is more important for function as it is activated intracellularly.

The above data suggest that tumor cell–derived C3 acts upon T cells to promote tumor growth. Therefore, we determined the impact of C3 deletion in CT26 cells on T-cell populations in the TME. We observed increased infiltration of CD45⁺ inflammatory cells (Fig. 2A) and CD3⁺ T cells in CT26-Sg-C3 tumors (Fig. 2B). The percentage of CD8⁺ T cells in CD3⁺ T cells increased by about 10% (Fig. 2C), whereas the absolute number of CD8⁺ T cells and CD4⁺ T cells was greatly elevated in CT26-Sg-C3 (Fig. 2D). Moreover, the IFNγ (Fig. 2E) and TNFα (Fig. 2F) producing CD8⁺ T cells were increased in CT26-Sg-C3 tumors compared with in CT26-Sg-Con tumors. The proliferation potential of CD8⁺ T cells (Fig. 2G) and CD4⁺ cells (Fig. 2H) was also increased, as determined by Ki-67 staining.

C3 produced by CD8⁺ T cells suppresses their antitumor immunity by inhibiting their IL10 production (9). However, IL10 production was not detected in CD8⁺ T cells, which is consistent with previous findings (ref. 9; Supplementary Fig. S2A). Moreover, we noticed that the addition of exogenous C3a or C5a did not affect the proliferation of CD8⁺ T cells in vitro (Supplementary Fig. S2B), indicating that tumor-derived C3 functions on CD8⁺ T cells in an indirect manner. To discriminate whether CD8⁺ or CD4⁺ T-cell lineage mediated tumor inhibition caused by C3 deletion, we used monoclonal antibodies (anti-CD4, clone GK1.5; anti-CD8a, clone 2.43, respectively) to deplete the corresponding population in tumor-bearing mice (Supplementary Fig. S3). Our data suggest that control of Sg-con CT26 tumor growth was somewhat dependent on CD8⁺ T cells, as depleting CD8⁺ T cells in Sg-con CT26 tumors only slightly promoted tumor growth. Depleting CD4⁺ T cells in Sg-con CT26 tumors delayed tumor growth (Fig. 2I). Although the loss of CD4⁺ T cells had no effect on CT26-Sg-C3 tumor growth, depletion of CD8⁺ T cells abrogated the tumor growth delay effect caused by C3 deficiency in tumor cells (Fig. 2I). Collectively, these results suggest that, unlike autocrine complement produced by CD8⁺ T cells, C3 generated by tumor cells might serve as a TME regulator to promote CD8⁺ T-cell dysfunction.

Previous studies have suggested that complement C5a can shape the TME by recruiting MDSCs and enhancing their inhibitory functions in a murine cervical cancer model (5). We explored this potential mechanism in our study but found that the percentage and function of CD11b⁺Gr-1⁺ MDSCs in CT26-Sg-C3 and CT26-Sg-Con tumors were comparable (Supplementary Fig. S4A). In addition, the percentages of the MDSC subpopulations, both polymorphonuclear MDSCs (PMN-MDSCs, CD11b⁺Gr-1^high) and mononuclear MDSCs (MO-MDSCs, CD11b⁺Gr-1^int), were comparable between the two groups (Supplementary Fig. S4B). Effector molecule analysis of MDSCs revealed that the expression of ARG1 was comparable between the two groups, whereas expression of NOS2 was about 2-fold higher in MDSCs from CT26-Sg-C3 (Supplementary Fig. S4C). Conversely, a higher percentage of regulatory T cells (Treg; CD4⁺Foxp3⁺) was present in CT26-Sg-C3 tumors, when compared with control tumors (Supplementary Fig. S4D). The increased infiltration of Tregs in CT26-Sg-C3 tumors may have resulted from excessive inflammation.

TAMs were reduced by around 50% within CT26-Sg-C3 tumors compared with that in CT26-Sg-Con tumors (Fig. 3A). TAMs exist as two functionally distinct subtypes: type 1–polarized macrophages (M1, CD11b⁺F4/80⁺MHCII^high/CD11c⁺), which are classically activated and possess potent tumoricidal capacities, and type 2 macrophages (M2, CD11b⁺F4/80⁺CD206⁺), which are alternatively activated and specialize in suppressing antitumor immunity (19, 26). We analyzed the composition of TAMs by the established macrophage polarization markers, MHCII^high/CD11c⁺ for M1 and CD206 for M2. The results showed that deletion of C3 in tumor cells promoted macrophage M1 polarization and inhibited M2 polarization (Fig. 3B–D). To explore the changes of TAMs following C3 deficiency in tumor cells, we applied RNA sequencing to profile the gene expression of TAMs, both from CT26-Sg-C3 and CT26-Sg-Con. RNA-sequencing data revealed that TAMs isolated from CT26-Sg-C3 tumors display a different gene-expression profile than that of CT26-Sg-Con tumors (Fig. 3E). Consistent with phenotype data from FACS analysis (Fig. 3B–D), TAMs from CT26-Sg-C3 tumors presented as an M1-polarized state compared with CT26-Sg-Con tumors (Fig. 3F). Furthermore, genes associated with antigen presentation and immune stimulation were upregulated in TAMs from C3-deficient tumors, whereas genes associated with immune suppression and chemoattraction were inhibited (Fig. 3G).

Functionally, TAMs isolated from CT26-Sg-Con tumors were more suppressive than their counterparts isolated from CT26-Sg-C3 tumors, as determined by the in vitro CD8⁺ T-cell proliferation inhibition assay (Fig. 3H). To further validate the contribution of TAMs on C3-mediated immunosuppression in vivo, we administered the macrophage depletion agent, clodronate-liposome, to mice bearing CT26-Sg-C3 and CT26-Sg-Con tumors. Although tumor growth from both experimental groups was inhibited by the treatment, macrophage depletion abrogated the difference between the two groups (Fig. 3I).

The activation of C3 results in the generation of C3a and C5a, two effectors of the complement system (27). To determine which active component facilitates tumor growth, we blocked C3a and C5a receptors individually in hosts by genetic depletion and pharmaceutical inhibition. C3AR1^−/− recipient mice were resistant to tumor growth, regardless of C3 sufficiency or deficiency in LLC tumor cells (Fig. 4A). The presence of the C5AR1 deletion in recipient mice did not abrogate the antitumor effects elicited by C3 deletion in LLC tumor cells (Fig. 4B). The same phenotypes were also seen with C3aR and C5R antagonist treatments (Fig. 4C). To validate whether C3 is directly acting on TAMs, we coinjected LLC with or without C3aR^+/+ TAMs or C3aR^−/− TAMs subcutaneously into C57BL/6 mice. The addition of C3aR^+/+ TAMs accelerated tumor growth as compared with tumors injected with LLC alone. C3aR^−/− TAM coinjection significantly delayed LLC growth when compared with other two groups (Fig. 4D and E). These results demonstrate C3 acts on TAMs in a C3aR-dependent manner.

To investigate the mechanism of M2 polarization by tumor cell–derived C3, we conducted signaling pathway analysis based on the above-mentioned RNA-sequencing data. Our data revealed that various pathways changed in TAMs isolated from CT26-Sg-C3 tumors. Among them, we found that PI3K/AKT/mTOR signaling was significantly downregulated in TAMs isolated from CT26-Sg-C3 tumors (Fig. 5A). PI3Kγ is expressed in TAMs, and inhibition of PI3Kγ activity by using IPI-549 promotes M1 polarization (19, 28). PI3Kγ signaling is also activated by C3a/C3aR signaling (29). On the basis of the above knowledge, we hypothesized that tumor cell–derived C3 promotes M2 polarization by activating PI3Kγ signaling. Among TAMs stimulated with murine recombinant C3a, we observed increased percentages of phosphorylated-akt (p-akt) positive TAMs along with an increased MFI of p-AKT in p-AKT⁺ TAMs (Fig. 5B and C). To test whether C3a-mediated Akt phosphorylation is C3aR-dependent, we treated cells with C3aR antibody or C3aR antagonist. C3aR antagonist treatment abolished C3a-mediated Akt phosphorylation and C3aR antibody inhibited C3a-mediated Akt phosphorylation (Fig. 5B and C). To test whether C3a-mediated Akt phosphorylation is PI3Kγ-dependent, we used IPI-549 (a PI3Kγ inhibitor) to treat the cells. Our data showed that IPI-549 treatment abrogated C3a-promoted Akt phosphorylation in TAMs (Fig. 5D and E). These results show that C3a could activate PI3Kγ signaling by binding to its receptor C3aR. For in vivo confirmation, we treated CT26-Sg-C3 or CT26-Sg-Con tumor-bearing mice with the PI3Kγ inhibitor, IPI-549, or a vehicle. Consistent with previous reports (19, 28), tumors in CT26-Sg-Con tumor-bearing mice treated with IPI-549 grew more slowly than tumors in mice treated with the vehicle. However, C3-deficient tumor-bearing mice did not respond to IPI-549 treatment (Fig. 5F). These results validate a C3a/C3aR/PI3Kγ signaling axis that regulates M2 polarization in the TME.

Our data from the mouse colon cancer model suggested that local C3 production by tumor cells expanded TAM cells and exhausted CD8⁺ T cells. To validate its clinical relevance, we explored a TCGA data set composed of tumor transcriptomes from 270 colorectal carcinoma patients (30). We found that C3 mRNA expression was correlated with a set of TAM surface markers such as CD68, CD163, and MRC1 (Fig. 6A–C). In addition, a series of immunosuppressive cytokines that are produced byM2 TAMs, such as TGFB1, TGFB3, and IL10, were also correlated with local C3 expression (Fig. 6D–F). The local C3 mRNA production was also associated with markers indicating T-cell exhaustion: PD-1, CTLA4, and TIM3 (Fig. 6G–I). Taken together, we concluded that the local production of C3 complement might aid establishment of an immunosuppressive TME and suppress infiltrating T cells in human colon cancers.

Clinical trials investigating the PD-1 pathway blockade have shown a durable response in subsets of patients with a variety of cancer types (31–34). Unfortunately, when delivered as a monotherapy, only a minority of patients benefit from PD-1 pathway blockade (35). One hypothesis for these failures is that additional TAMs accumulate and provide a mechanism of resistance to PD-1 signaling blockade (19, 28). Our data from animal models and clinical samples indicated that C3 promotes TAM accumulation and enhances its inhibitory potential. Therefore, we hypothesized that tumor cell–derived C3 might represent a resistant factor. In published work (36) and our data using PD-L1 antibody treatment, cell lines with high C3 expression (LLC and 4T1) did not respond to anti–PD-L1 treatment, whereas cell lines (B16F10, EL4, and MC38) with little to no C3 expression tumors responded well (Fig. 1C). We applied SAM technology to activate C3 expression in B16F10 cells (37) and validated the expression of C3 in the B16-C3 cell supernatant by C3 ELISA (Fig. 7A). To understand the association between tumor C3 expression and resistance to PD-L1 antibody treatment, we compared multiple mouse tumor models treated with the PD-L1 antibody. Our results show that mice bearing 4T1 tumors, which express large amounts of C3, are resistant to PD-L1 antibody treatment (Fig. 7B, left). By contrast, mice bearing B16F10 tumors, which express little C3, are more responsive to PD-L1 antibody treatment and benefit from a longer survival (Fig. 7B, middle). However, mice bearing B16F10-C3 tumors lose sensitivity to PD-L1 antibody treatment compared with B16F10 controls (Fig. 7B, right). To test whether combinatorial targeting of these two pathways could overcome the resistance to PD-L1 antibody treatment, we performed checkpoint therapy against C3-deleted CT26 tumors with anti–PD-L1. As a monotherapy, anti–PD-L1 slightly inhibited the growth of C3-sufficient CT26 tumors (Fig. 7C and D). However, sensitivity to anti–PD-L1 was enhanced in mice bearing C3-deficient tumors (Fig. 7C and D). With anti–PD-L1 treatment, 75% (15/20) of C3-deficient CT26 tumors disappeared; these mice achieved long-term survival (Fig. 7E).

To test whether anti–PD-L1 treatment in C3-deficient tumor-bearing mice mediated antitumor responses result in prolonged protective T-cell immunity, we rechallenged mice that underwent complete tumor regression after treatment with 4T1 or CT26 tumor cells. All the mice from the complete tumor regression group rechallenged with CT26 rejected the tumor, whereas mice rechallenged with 4T1 failed to reject the tumor (Fig. 7F and G). These results suggest that anti–PD-L1 treatment in C3-deficient tumors promoted the generation of tumor-specific long-term protective immunity. Mechanically, we observed increased expression of PD-L1 in both CD45⁻ tumor cells and CD11b⁺ myeloid cells from C3-deficient CT26 tumors (Supplementary Fig. S5A and S5B). Furthermore, as previous report suggested that IFNγ-associated gene signatures have been suggested to predict clinical response to PD-1 blockade (38), we assessed the expression of IFNγ-associated genes in CT26-Sg-Con or CT26-Sg-C3 tumors, including B2M, CD8A, CD27, CD274, CXCL9, CLCL10, CXCR6, H2EB1, IDO1, IFNG, LAG3, NKG7, PDCD1LG2, PDL1, and RSMB10. We found that the expression of IFNγ-associated genes was increased in CT26 upon C3 deletion (Supplementary Fig. S5C).

Evidence showing that complement activation functions as an immune suppressor in the TME has inspired interest in regulating the complement system for cancer therapy (5, 7–9, 18). Although the role of systemic complement activation in tumor development has been studied, whether and how tumor cell–derived complements could act on antitumor immunity remain to be explored. Here, we describe a role of tumor cell–derived C3 in tumor immunity. We also suggest a strategy against complement to enhance checkpoint inhibitor therapies. Intracellular activation of complement C3 in the tumor cells plays a role in antitumor immunity, independent of systemic complement activation and T-cell–producing C3. Mechanically, tumor cell–derived C3a could modulate TAMs via C3a-C3aR-PI3Kγ signaling, thereby repressing antitumor immunity. Deletion of C3 in the tumor cells with high C3 expression is sufficient to enhance the efficacy of anti–PD-L1 treatment, whereas transactivation of C3 in tumor cells with low C3 expression could dampen anti–PD-L1 treatment response.

Previous reports with C3^−/− mice have generated variable results for different tumor models: the growth of TC-1 and EO771 breast tumors was retarded (5, 9), there was no difference in ID8-VEGF levels (17), and data from B16F10 melanoma were contradictory (2, 9). These discrepancies might be explained by our findings that different tumor cells produced different amounts of C3: B16F10 cells produced no C3, whereas CT26 and LLC produced high levels of C3. Therefore, any benefit to tumor immunity by deleting host C3 could be tumor-type specific. For robust C3 producers, such as CT26 and LLC, the major source of C3 is the tumor cell (17, 39), but for tumors without C3 expression, such as B16F10, C3 may instead derive from host cells, including CD8⁺ T cells. Accordingly, our results showed that although C3 deficiency in the host slows tumor growth, C3 deletion in cancer cells delayed tumor growth to a greater degree. On the other hand, when looking at C3 activation, tumor cell–derived, local C3 activation might be more important than systemic C3 activation. Systemic C3 is primarily activated via classic, alternative, and lectin pathways (40). However, due to the potential harmful consequences of its cytolytic properties by membrane attack complex, activation of systemic C3 is controlled by soluble and membrane-bound complement regulatory proteins (CRP; ref. 41). Evidence suggests tumor cells express CRPs (42) to avoid potential self-harm, which makes activation of systemic C3 difficult in the extracellular space in the TAM. In contrast, tumor cell–derived C3 avoids the interaction with these extracellular CRPs by intracellular activation. In the present study, both C3a and C3b/iC3b/C3c were found in cultured CT26 and LLC cancer cells. Consistent with our results, intracellular activation of C3 in intestinal epithelial cells has been illustrated in mesenteric ischemia model (15). We found both CT26 and LLC express high CTSL and CTSG, two molecules reported to cleave C3 in human T cells (10, 43). How intracellular activation of C3 in tumor cells occurs remains to be addressed.

Several mechanisms have been proposed for the protumorigenic function of complement (5, 9, 17, 18, 24, 44). To analyze our model in the context of these mechanisms, we conducted a series of experiments. First, with two independent assays we found that deletion of the C3 gene had no effect on the proliferation of tumor cells or in nude mice. This is supported by other groups' findings that epithelial cells express little C3aR or C5aR (45, 46), and exogenous C3a does not promote proliferation of LLC cells (18, 24). These results from nude mice also excluded the possibility that tumor-derived C3 promoted tumor growth by affecting the vascular system. Second, our results showed that deletion of C3 in tumor cells did not alter MDSC composition in the TME, consistent with a previous report (9). Third, we failed to detect any IL10 production by CD8⁺ T cells in the CT26 model. To explain this discrepancy, we speculate that autocrine C3 production by CD8⁺ cells inhibits its own IL10 production, as suggested by Wang and colleagues (9). Fourth, our results showed that deletion of CD4⁺ T cells did not abrogate the tumor growth delay effect caused by C3 deletion in tumor cells, which was inconsistent with a previous report (24). We noted that in this work, CD4⁺ T-cell depletion antibody was used for a total of 8 times in 4 weeks. In this case, we doubt that CD8⁺ T cells would also be affected. CD4⁺ T-cell deletion by antibody has been reported to delay tumor growth in various studies (47), whereas conflicted results were reported in this study (24). Nevertheless, in our present study with CT26 and LLC tumor models, we found that C3-deficient tumors had reduced TAM accumulation and that TAMs exhibited an M1-polarized state, which was consistent with a previous report (44). In this case, why was intracellular and systemic C3 activation in tumor cells immunosuppressive by different mechanisms (TAM vs. MDSC)? First, MDSC recruitment and enhanced functions are C5aR signaling–dependent. However, intracellular C3-activating protease does not cleave C5 (10). Accordingly, tumor cell intracellular C3 activation could not affect MDSC. Second, CD11b⁺F4/80⁺ macrophages highly express C3aR (48), whereas CD11b⁺Gr1⁺cells (namely, MDSCs in tumor) express no C3aR (48). Therefore, tumor cell intracellular C3 activation may modulate TAMs. The above scenario was further supported by these data: (i) tumor cell–derived C3 promoted immunosuppression through C3aR but not C5aR; (ii) TAMs derived from CT26-Sg-C3 tumor exhibited the characteristics of M1, whereas CT26-Sg-Con tumor were M2 polarized; (iii) macrophage depletion abrogated the tumor growth difference between CT26-Sg-C3 and CT26-Sg-Con; (iv) C3aR antagonist inhibited C3a-induced PI3Kγ activation, needed for M2 polarization, and also abolished the difference in tumor growth between CT26-Sg-C3 and CT26-Sg-Con.

Targeting complement as a tumor immunotherapy has been reported (7–9). We and others showed that systemic blockade of C5aR and C3aR could enhance the efficacy of anti–PD-L1/PD-L1 (7–9). However, there are clinical limitations to this strategy. As both C3aR and C5aR provide protection against acute infections, systemic blockade of C3aR and C5aR may cause serious infection in the patients, as suggested by experiments using a mouse model (49, 50). For tumors with robust C3 production, our data showed that blocking tumor cell–derived C3 is sufficient to overcome the resistance to PD-1 signaling blockade. We reasoned that for these types of tumor, blocking C3 via tumor cell–targeted gene-editing techniques may be a safer choice to avoid serious adverse events caused by systemic blockade of both C3aR and C5aR.

Our study suggested two strategies to advance current cancer immunotherapy. First, our data showed that deleting C3 in tumor cells is sufficient to remodel the TME and enhance antitumor immunity. Therefore, strategies such as linking a C3aR antagonist to antibodies against tumor-associated antigens (51) may deliver safe and robust therapeutic effects. Second, by depleting or antagonizing C3, we may bias the remaining immune-suppressive mechanism of the TME toward the PD-1–PD-L1 axis.

No potential conflicts of interest were disclosed.

Conception and design: H. Zha, B. Guo, B. Zhu

Development of methodology: H. Zha, X. Wang, Y. Zhu, C. Hu, Y.Y. Wan, B. Guo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Zha, Y. Zhu, D. Chen, X. Han, F. Yang, J. Gao, Y. Feng, Y. Li, B. Guo

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Zha, X. Wang, X. Han, B. Guo, B. Zhu

Writing, review, and/or revision of the manuscript: H. Zha, Y. Li, Y.Y. Wan, B. Guo, B. Zhu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Zha, C. Shu, Y. Tan, J. Zhang, Y. Li, B. Guo, B. Zhu

Study supervision: B. Guo, B. Zhu

This work was supported by the National Nature Science Foundation of China (No. 81472648 and No. 81620108023 to B. Zhu; No. 31570866 and No. 31870875 to B. Guo). We thank Rick A. Wetsel (The University of Texas) for kindly providing C3aR knockout mice; Craig Gerard (Harvard Medical School) for kindly providing C5aR knockout mice.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.