NOTCH Signaling via WNT Regulates the Proliferation of Alternative, CCR2-Independent Tumor-Associated Macrophages in Hepatocellular Carcinoma

Macrophages infiltrating tumors, known as tumor-associated macrophages (TAM), are pivotally involved in tumor initiation, progression, and metastasis (1, 2). TAMs are broadly classified into the M2 macrophage activation subtype, attributing to their upregulated expression of M2 markers including IL10, mannose receptor, and arginase, as well as their enhanced activities in suppressing antitumor immunity and promoting tumor neovascularization (1–3). Compelling evidence has demonstrated that the majority of TAMs are derived from CCR2⁺ inflammatory monocytes in bone marrow (BM), therefore targeting CCR2⁺ monocytes to block TAM input has been recognized as a promising therapy (2–4). However, it is increasingly clear that TAMs constitute heterogeneous populations (5, 6). In mouse mammary cancer models and human breast cancers, as well as in transplanted murine Lewis lung carcinoma (LLC), spontaneous lung cancer, and glioma, TAMs originate from monocyte input but can maintain their population by local proliferation (5–10). On the other hand, embryonic hematopoiesis–derived tissue-resident macrophages, which are maintained by local mitogenic signals such as colony-stimulating factor (CSF)-1 and IL4 (11–14), also participate in tumor progression and metastasis as shown in pancreatic ductal adenocarcinoma (15) and certain lung metastasis events (16). It is important to dissect the mechanisms for the regulation of these distinct monocyte and macrophage populations in tumors for more efficient intervention.

Hepatocellular carcinoma (HCC) is a leading cause of cancer-related death worldwide (17). In liver, resident macrophages or Kupffer cells (KC) are primarily generated during embryogenesis and sustained by self-renewal, but recent evidence has demonstrated their input from BM also (18–20). KCs produce IL6 to promote diethylnitrosamine-induced HCC and are responsible for the gender disparity in HCC (21–23). KCs also play a key role in cholangiocellular carcinogenesis (24). Monocyte-derived macrophages, on the other hand, are responsible for the clearance of senescent premalignant hepatocytes (25). However, precise contributions of different macrophage populations and their regulation in hepatic cancers have not been fully understood.

The recombination signal binding protein-Jk (RBPj)-mediated NOTCH signaling regulates cell differentiation and plasticity in cooperation with other pathways (26, 27). NOTCH signaling plays a critical role in differentiation and functional plasticity of TAMs (7, 28, 29). NOTCH signaling is required for monocyte-derived TAM differentiation as shown in myeloid-specific Rbpj knockout mediated by CD11c-Cre in mice (7). On the other hand, forced Notch activation by myeloid-specific overexpression of Notch intracellular domain (NICD) mediated by LyzM-Cre results in attenuated TAM phenotypes in mice (29). In this study, we show that myeloid-specific Rbpj knockout blocked TAM differentiation from monocytes in HCC as reported previously in breast cancer models (7), but a TAM population with KC-like phenotype expanded via proliferation and constituted an alternative source of TAMs to facilitate tumor growth and hepatic metastasis. Our evidence also supported that NOTCH-WNT signaling governed the expansion of these KC-like TAMs in HCCs.

Human HCC biopsies were obtained from patients hospitalized in the Department of Hepatobiliary Surgery, Xijing Hospital, Fourth Military Medical University (Xi'an, China), and staged according to the AJCC Cancer Staging Manual (8th ed.; Supplementary Table S1). The use of human samples was approved by the Ethics Committee of Xijing Hospital and conformed to Declaration of Helsinki. Written informed consent was obtained from all patients involved.

LyzM-Cre (stock #019096, Jackson Laboratory), Rbpj-floxed (Rbpj^f; ref. 30), and Ccr2 knockout (stock #004999, Jackson Laboratory) mice were brought about on the C57BL/6J background in a specific pathogen-free facility. Mice were mated and genotyped by PCR. Male littermates of 8- to 10-weeks old were used in experiments. All animal experiments were reviewed and approved by the Animal Experiment Administration Committee of the Fourth Military Medical University to ensure ethnical and humane treatment of animals.

LLC, Hepa1-6 (HCC), and CMT93 (colorectal cancer) cells were obtained from the ATCC repository in 2015. These cells were authenticated by both morphologic profiling and short tandem repeat profiling and tested by PCR to exclude Mycoplasma contamination. Cells were maintained in DMEM supplemented with 10% FCS and 2 mmol/L l-glutamine (Invitrogen). Both LLC and CMT93 cells were derived from the C57BL strain (H-2^b). Hepa1-6 cells were derived from C57L mice (H-2^b) but bear the same MHC genotype as that of C57BL strain. So we tentatively used Hepa1-6 cells to establish orthotopic HCC models in mice on C57BL background. For lung cancer models, LLC cells (5 × 10⁶ cells/200 μL) were inoculated subcutaneously on the right side of the back of recipient mice. For orthotopic HCC models, mice were anesthetized by intraperitoneal injection of 0.6% pentobarbital sodium (10 μL/g, Sigma-Aldrich). Hepa1-6 cells (5 × 10⁶ cells in 30 μL Matrigel, Sigma) were injected into liver parenchyma of left lobe. Mice were sacrificed 3 weeks after inoculation. Tumors were weighed and tumor volume was estimated as (L × S²) × 0.52 (L, long diameter; S, short diameter). In some experiments, Hepa1-6 cells, CMT93, or LLC cells were transduced with a lentivirus expressing firefly luciferase (Genechem) following the manufacturer's protocol, and hepatic tumor growth was monitored using an in vivo imaging system (IVIS; Xenogen, Perkin-Elmer). For liver metastasis, CMT93 cells (1 × 10⁶) were injected into spleen parenchyma of mice. For lung metastasis, LLC cells (1 × 10⁶) were injected via tail vein. Both liver and lung metastasis was monitored using IVIS. ICG-001 (MedChem Express) was administered by intraperitoneal injection (5 mg/kg in saline; Supplementary Fig. S1A–S1D for tumor models).

Mice were perfused with PBS before tumors were collected and fixed in cold 4% paraformaldehyde and treated with 30% sucrose at 4 °C. Hematoxylin and eosin staining was performed routinely, and pictures were taken under a microscope (BX51, Olympus) equipped with a CCD camera (DP70, Olympus). For immunofluorescence, samples were embedded in optimum cutting temperature compound and cryosectioned at 5 or 60-μm thickness, followed by immunostaining according to standard protocols with antibodies listed in Supplementary Table S2. Nuclei were counterstained with Hoechst 33258 (Sigma). Images were collected under a laser scanning confocal microscope (FV1000, Olympus) or a fluorescence microscope (BX51, Olympus). Three-dimensional reconstruction of images was performed with the NIS-Elements Viewer 4.20 software (Nikon) after CD31 staining.

Murine primary KCs were isolated by density gradient centrifugation and cultured as described previously (30). For transfection with siRNA, KCs (5 × 10⁵) were seeded in 24-well plates, and transfected with synthetic siRNA or negative control (NC; 50 nmol/L, RiboBio) using Lipofectamine 2000 (Invitrogen) following the manufacturer's protocol. Medium was replaced with normal medium 4 hours after adding the liposome–nucleotide mixture, and cells were assayed 24 hours later. In some experiments, KCs were treated with Hepa1-6 conditional medium (CM) or Wnt3a (100 ng/mL) for 24 hours. Monocytes were isolated by FACS according to the markers described previously (31) and stimulated with Hepa1-6 CM for 24 hours before FACS analysis of intracellular Ki67.

For reporter assay, the murine β-catenin (β-Ctnn) promoter fragment (−1,237–+131, Genebank accession # NC_000075.6) was cloned by PCR to construct pCtnnb1-pro-luc using pGL3-basic (Invitrogen; ref. 32). KCs (1 × 10⁶) were transfected with 1 μg pCtnnb1-pro-luc and 50 ng pRL-TK by Nucleofector (Lonza) according to the recommended protocol, and cultured further with Hepa1-6 CM and a γ-secretase inhibitor (GSI; 2.5, 5.0, 10 μmol/L; DAPT, Sigma) for 24 hours. Luciferase activity was detected using a Dual-Luciferase Reporter Assay Kit (Promega) and a chemiluminometer (Luminoskan Ascent, Labsystems).

Short hairpin RNA (shRNA) targeting β-catenin (Shβ-Ctnn) and the NC were inserted into GV493 lentivirus vector that expresses GFP (Genechem). HEK293T cells were transfected to prepare viral particles. KCs from RBPj cKO and control mice were stably transduced with Shβ-Ctnn or NC shRNA using the lentiviral particles. The infected GFP-labeled KCs (2.5 × 10⁵) were mixed with Hepa1-6 cells (5 × 10⁶) in 30 μL Matrigel, and were orthotopically inoculated in liver of wild-type mice. Mice were maintained for 3 weeks after the inoculation. In addition, adenovirus-expressing NICD (Genechem) was applied to infect wild-type KCs according to the manufacturer's protocol.

Mice were perfused with PBS and tumors were dissected and finely minced, followed by incubation with 1 mg/mL collagenase V and 4 μg/mL DNase I (Roche) at 37 °C for 1 hour. Single-cell suspensions were obtained by passing through a 70-μm cell strainer. After erythrolysis, cells were resuspended in FACS buffer, and incubated with antibodies listed in Supplementary Table S2 for 20 minutes in dark on ice. Intracellular staining was performed with a permeabilization buffer (eBioscience) and stained with antibodies. Cells were analyzed with a FACSCalibur or a FACSAria II flow cytometer (BD Immunocytometry Systems), and data were analyzed using the FlowJo 7.6.1 Software (TreeStar). Cell viability was evaluated with 7-amino-actinomycin D (BD Biosciences). Cell sorting was performed using the FACSAria II Flow Cytometer (BD Biosciences).

Serum IL10 in mice was determined with an ELISA Kit (eBioscience) according to the manufacturer's instructions. Wnt3a level was also determined with a kit (MyBioSource).

Total RNA was extracted from tissue or cell samples using the TRizol Reagent (Invitrogen), and reverse transcribed into cDNA with the PrimeScript RT Reagent Kit (Takara Biotechnology). qPCR was performed using the SYBR Premix ExTaq II (Takara Biotechnology) and Applied Biosystems 7500 Real-time PCR System, with β-actin as a reference control. Primers are listed in Supplementary Table S3.

Cells were lysed with the RIPA Lysis Buffer (Beyotime Biotechnology) containing 10 mmol/L phenylmethanesulfonyl fluoride. Samples were separated by SDS-PAGE and electro-transferred onto polyvinylidene fluoride membranes. Membranes were blocked with 5% skim milk in PBS-0.1% Tween 20, and then incubated at 4 °C overnight with primary antibodies, followed by washing and incubation with secondary antibodies at room temperature for 1 hour. After washing, blots were developed with enhanced chemiluminescence (Pierce, Thermo Fisher Scientific) and detected using ChemiScope Imaging System (Clinx Science Instruments). β-ACTIN was used as an internal reference. Antibodies are listed in Supplementary Table S2.

Quantification of histologic image was conducted with Image Pro Plus 6.0 Software (Media Cybernetics). Statistical analyses were performed with GraphPad Prism 5 software. The comparisons between groups were undertaken using paired and unpaired Student t tests or one-way ANOVA with Tukey multiple comparison test. Results were expressed as means ± SD. P < 0.05 was considered as significant.

NOTCH signaling appeared activated in TAMs, likely by ligands expressed by tumor or microenvironmental cells (Supplementary Fig. S1E). Myeloid-specific Rbpj disruption with LyzM-Cre (RBPj cKO) resulted in retarded growth of subcutaneous LLC tumors as compared with the control (LyzM-Cre-RBPj^f/+; Supplementary Figs. S1A and S2A–S2C). FACS analyses showed that RBPj cKO reduced TAMs (F4/80^hiCD11b^hi) and increased F4/80^intCD11b^hiLy6C^hi monocytes (Supplementary Fig. S2D), likely due to blocked differentiation of monocyte-derived TAMs (moTAM) as demonstrated by CD11c-Cre–mediated Rbpj knockout in murine breast cancer (7). Consistently, immunofluorescence showed that accompanied with reduced F4/80⁺ TAMs, CD31⁺ vascular staining decreased significantly in subcutaneous LLC tumors from RBPj cKO mice (Supplementary Fig. S2E). Therefore, consistent with previous reports (7), myeloid NOTCH blockade by LyzM-Cre–mediated Rbpj knockout reduced TAMs and suppressed the growth of subcutaneous LLC tumors.

In liver, macrophage development appeared normal in static RBPj cKO mice (Supplementary Fig. S3A and S3B). We found that, unexpectedly, orthotopic HCC tumors established by inoculation of Hepa1-6 cells in the liver (Supplementary Fig. S1B) grew significantly larger in the Rbpj cKO mice than in the control (Fig. 1A and B; Supplementary Fig. S3C–S3E). Tumor-infiltrating macrophages and vasculature increased as shown by immunofluorescence (Fig. 1C). FACS showed that CD8⁺ T cells decreased (Fig. 1D and E). LyzM-Cre specifically activates LoxP-mediated DNA recombination in macrophages and neutrophils (33). In RBPj cKO mice, while total CD11b⁺ myeloid cells (Fig. 1D and F) and MHCII^hiCD11c^hi dendritic cells (Supplementary Fig. S4A) in hepatic Hep1-6 tumors appeared unchanged, and F4/80⁻CD11b⁺Ly6G^hi tumor-associated neutrophils (TAN) showed a tendency of decrease, but not statistically significant, the F4/80^hiCD11b⁺Ly6G⁻ TAM population increased significantly (Fig. 1D and G). Tumor IL10 mRNA and serum IL10 increased in RBPj cKO mice bearing orthotopic Hepa1-6 tumors (Fig. 1H). These data indicated that despite repressing subcutaneous LLC tumors, blocking NOTCH signaling by RBPj cKO promoted orthotopic Hepa1-6 tumor growth with increased TAMs.

To determine the identity of increased TAMs in hepatic Hepa1-6 tumors of RBPj cKO mice, single-cell suspensions from orthotopic Hepa1-6 tumors of perfused mice were analyzed by FACS after staining with CD11b and F4/80 (Fig. 2A; refs. 34, 35). The F4/80^hiCD11b^lo macrophage population (G₁), which normally represents KCs in quiescence but may also include monocyte-derived macrophages (20), increased significantly in both cell percentage and number (Fig. 2B). These TAMs were Ly6G-negative and Ly6C-low/negative, and expressed mature TAM markers (VCAM1 and MHCII) and KC markers (MARCO and TIM4) and negative for the eosinophil marker SiglecF (Fig. 2C). Moreover, they expressed higher IL10 in RBPj cKO mice (Fig. 2D). In FACS-sorted F4/80^hiCD11b^lo TAMs, qRT-PCR confirmed reduced expression of NOTCH downstream genes Hes1 and Hey1 (Supplementary Fig. S4B), and increased IL10, and PD-L1 and PD-L2 mRNA accompanied by decreased IL12 and IL1β expression (Fig. 2E). FACS with Ly6C staining further showed that the F4/80⁺ population contained low percentages of Ly6C^hi monocytes and Ly6C^int moTAMs, but the majority (around 90%) of this population was Ly6C^lo/− macrophages that resembled KCs (36). In this F4/80⁺ population from hepatic Hepa1-6 tumors in RBPj cKO mice, monocytes increased, moTAMs decreased, and Ly6C^lo/− TAMs increased mildly but significantly (Fig. 2F). These F4/80⁺CD11b^loLy6C^lo/− TAMs were then tentatively named as KC-like TAMs (kclTAM), although their ontogeny might be complex (35, 36). Further analysis revealed that the F4/80^hiCD11b^lo population (Fig. 2A) and the F4/80⁺CD11b^loLy6C^lo/− population (Fig. 2F) were the same population of kclTAMs cells (Supplementary Fig. S4C and S4D). Although moTAMs decreased, the total TAM number still accumulated significantly with the increase of kclTAMs in RBPj cKO mice (Supplementary Fig. S4E). Notably, kclTAMs dominated the total TAMs both in RBPj cKO and Ctrl mice (>90%), with higher proportion in the RBPj cKO ones (Supplementary Fig. S4E). Coinoculation of Hepa1-6 with KCs indeed promoted HCC tumor growth, suggesting a protumor role of kclTAMs (Supplementary Fig. S4F). In subcutaneous LLC tumors, consistent with previous findings in breast cancer models (7), staining with anti-CD11b, F4/80, and Ly6C showed that monocytes (F4/80⁺CD11b^hiLy6C^hi) increased and moTAMs (F4/80⁺CD11b^hiLy6C^int) decreased upon NOTCH blockade, respectively (Supplementary Fig. S4G and S4H). These results suggested that although NOTCH blockade disrupted moTAMs differentiation from inflammatory monocytes, kclTAMs increased alternatively, participating in accelerated growth of hepatic Hepa1-6 tumors.

Next, we analyzed mechanisms underlying increased kclTAMs in RBPj cKO mice. Apoptosis did not change in F4/80^hiCD11b^lo TAMs from RBPj cKO mice (Supplementary Fig. S5A and S5B). In contrast, cytoplasmic staining of Ki67 indicated that Ki67⁺ cells increased remarkably in the F4/80^hiCD11b^lo TAM population in hepatic Hepa1-6 tumors from RBPj cKO mice, suggesting increased proliferation (Fig. 3A and B). However, no significant increase in Ki67⁺ cells was detected in the monocyte/moTAM (F4/80^intCD11b^int) or neutrophil (F4/80⁻CD11b^hi) compartment (Supplementary Fig. S5C and S5D). Immunofluorescence staining confirmed that F4/80⁺Ki67⁺ TAMs increased in hepatic Hepa1-6 tumors from RBPj cKO mice (Fig. 3C). Next, KCs were isolated from RBPj cKO and control mice (Supplementary Fig. S5E) and cultured with Hepa1-6 CM, which may provide mitogenic stimulation for KCs. EdU incorporation assay showed more EdU⁺ cells in Rbpj-deficient KCs than that of the control (Fig. 3D and E). Cell-cycle analysis showed that the G₂–M proportion increased in Rbpj-deficient KCs as compared with the control (Fig. 3F). In normal KCs treated with Hepa1-6 CM and GSI, a NOTCH signal inhibitor, more proliferative EdU⁺ cells were detected compared with that of the control (Supplementary Fig. S5F and S5G). However, cytoplasmic Ki67 staining of FACS-sorted BM monocytes stimulated with Hepa1-6 CM showed that the percentage of Ki67⁺ RBPj cKO monocytes displayed a tendency of decrease but not statistical significance, whose phenotype was still different from that of RBPj cKO kclTAMs (Supplementary Fig. S6A–S6C). Together, these data suggested that RBPj cKO increased kclTAMs by promoting their proliferation in hepatic tumors.

To further investigate the macrophage subpopulations in HCC in RBPj cKO mice, we bred mice to obtain RBPj cKO on Ccr2^−/− background. Ccr2 knockout reduced the growth of hepatic Hepa1-6 tumors accompanied by a 34.1% (7.81 ± 0.86 in control vs. 5.14 ± 0.46 in CCR2 knockout) reduction in the F4/80^hiCD11b^lo population (Fig. 4A and B; Supplementary Fig. S7A). F4/80^intCD11b^int monocytes/moTAMs decreased more significantly (Fig. 4A and B). However, when NOTCH signaling was interrupted by RBPj cKO on the Ccr2^−/− background, the growth of hepatic Hepa1-6 tumors increased as compared with the Ccr2^−/− control (Fig. 4C and D; Supplementary Fig. S7B–S7D). Flow cytometry showed that, compared with the Ccr2^−/− control, the F4/80^hiCD11b^lo population increased by 28.3% (5.41 ± 0.30 in Ccr2^−/− control to 6.94 ± 0.13 in Ccr2^−/− RBPj cKO) in hepatic Hepa1-6 tumors from RBPj cKO mice (Fig. 4E and F). Anti-Ki67 and anti-IL10 staining showed that cells in the F4/80^hiCD11b^lo population were more proliferative and expressed more IL10 (Fig. 4E and F). Staining of Ly6C indicated that while monocytes and moTAMs were almost absent likely due to Ccr2 deficiency, F4/80⁺CD11b^loLy6C^lo/− kclTAMs increased (Fig. 4G and H). These data further suggested that NOTCH blockade in myeloid cells promoted kclTAM proliferation and IL10 expression independent of CCR2 in HCC.

Next, we investigated the molecular mechanisms of enhanced proliferation in RBPj cKO kclTAMs. The expression of CSF-1R, CSF-2Rb2, and IL4R in F4/80^hiCD11b^lo TAMs did not change in RBPj cKO as detected by qRT-PCR (Supplementary Fig. S8A). We then examined the activity of WNT signaling, which is repressed by NOTCH activation (27), by immunofluorescence staining with anti-F4/80 and anti-β-catenin. In hepatic tumors from RBPj cKO mice, the mean fluorescence intensity of both total and nuclear β-catenin in F4/80⁺ TAMs increased significantly (Fig. 5A and B). The expression of WNT downstream molecules Axin2, β-catenin, c-Myc, and cyclin D1 increased significantly in FACS-sorted F4/80^hiCD11b^lo TAMs from RBPj cKO mice, as determined by qRT-PCR (Fig. 5C). The protein level of β-CATENIN, activated β-CATENIN (a-β-ctnn), and c-MYC also increased in these cells (Fig. 5D). Rbpj-knockout KCs cultured with Hepa1-6–derived CM displayed upregulated mRNA expression of c-Myc, β-catenin, cyclin D1, and Axin2 (Fig. 5E), as well as protein expression of β-CATENIN, a-β-CTNN, and c-MYC (Fig. 5F). We also carried out a reporter assay by transfecting KCs with pCtnnb1-pro-luc and cultured in the presence of Hepa1-6 CM and GSI. The result showed that blocking NOTCH signaling upregulated β-catenin mRNA, and increased luciferase expression driven by the β-catenin promoter (Fig. 5G and H). ELISA showed that Hepa1-6 secreted WNT3a, which could stimulate the proliferation of KCs (Supplementary Fig. S8B and S8C). Surprisingly, compared with control KCs, RBPj cKO KCs cultured in vitro appeared to exhibit upregulation of WNT signaling independent of exogenous Notch ligands. Thus, we supposed that KCs could express Notch ligands themselves and signal to each other. Indeed, Hepa1-6 CM stimulated KCs to express JAG1 and elevated the expression of NICD with more translocation into the nucleus (Supplementary Fig. S8D and S8E), suggesting activation of NOTCH signaling. To further confirm whether the NOTCH activation of KCs was dependent on the JAG1 ligand expressed, we treated KCs with Jag1 siRNA, which efficiently knocked down the expression of JAG1, as well as that of NICD (Supplementary Fig. S8F and S8G). Next, we used the adenovirus to overexpress NICD in Hepa1-6 CM-treated KCs and found that the protein level of β-CATENIN and a-β-CTNN, and the mRNA level of β-catenin and Axin2 were downregulated with reduced proliferative capacity of KCs as measured by EdU staining (Supplementary Fig. S8H and S8J). These results suggested that NOTCH signaling could negatively regulate the activation of WNT/β-CATENIN signaling and the proliferation of kclTAMs.

We next accessed whether blocking WNT signaling in vivo could inhibit kclTAM proliferation in hepatic Hepa1-6 tumors in RBPj cKO mice with ICG-001, an antagonist of WNT/β-CATENIN pathway (37). Injection of ICG-001 significantly suppressed growth of orthotopic Hepa1-6 tumors in both control and RBPj cKO mice (Supplementary Fig. S9A). FACS analyses showed that the F4/80^hiCD11b^lo population decreased significantly upon ICG-001 injection (Fig. 6A and B). In this population, Ki67⁺ proliferative cells and IL10-expressing cells, both of which increased in hepatic tumors of RBPj cKO mice, decreased significantly (Fig. 6A and B). CD8⁺ T cells increased, whereas regulatory T (Treg) cells decreased (Fig. 6C; Supplementary Fig. S9B and S9C). These data suggested that WNT signaling was required for enhanced proliferation and IL10 expression in Rbpj-deficient kclTAMs. However, reduced tumor growth might not be solely attributed to decreased kclTAMs in the presence of ICG-001, because Wnt signaling plays multiple roles in different cell types in tumor.

In cultured RBPj cKO KCs, β-catenin siRNA repressed β-CATENIN and also AXIN2 expression, and c-MYC siRNA repressed c-MYC expression efficiently (Supplementary Fig. S10A–S10D). EdU staining indicated that knockdown of β-catenin abrogated the enhanced proliferation of Rbpj-deficient KCs (Fig. 6D; Supplementary Fig. S10E). Moreover, knockdown of β-catenin or c-Myc in Rbpj-deficient KCs with siRNAs reversed the downregulated IL12 and IL1β and upregulated IL10 as shown by qRT-PCR (Fig. 6E and F). These data suggested that WNT/β-CATENIN signaling was required for not only accelerated proliferation, but also likely for their protumor cytokine secretion of Rbpj-deficient kclTAMs, likely through upregulated c-MYC (38).

To further show that WNT signaling mediated kclTAM proliferation under RBPj cKO, we transfected KCs from RBPj cKO and control mice with β-catenin shRNA or NC shRNA (both labeled with GFP), and coinoculated these KCs with Hepa1-6 cells in liver of wild-type mice. KCs were transfected with high efficiency, and β-catenin was effectively knocked down with downregulated Axin2 expression (Supplementary Fig. S10F–S10H). The result showed that Rbpj-deficient KCs significantly promoted tumor growth, but this effect was abrogated by β-catenin shRNA transfection (Supplementary Fig. S11A). FACS analysis showed that kclTAMs, which contained the GFP⁺ KCs, increased in the Rbpj-deficient group, but this increase was canceled by β-catenin shRNA transfection (Supplementary Fig. S11B and S11C). Moreover, GFP⁺ RBPj cKO KCs exhibited stronger proliferation as assessed by Ki67 staining, which was also canceled by β-catenin shRNA transfection (Supplementary Fig. S11B, S11D, and S11E). These data support the conclusion that increased kclTAM proliferation in orthotopic Hepa1-6 tumors of RBPj cKO mice was mediated by WNT signaling pathway.

To test the effect of myeloid NOTCH blockade on hepatic metastasis, we inoculated luciferase⁺ CMT93 colorectal cancer cells into spleen of RBPj cKO and control mice (Supplementary Fig. S1C). Live imaging indicated that hepatic metastatic tumors were significantly larger in RBPj cKO mice than that in the control (Fig. 7A; Supplementary Fig. S12A). Histologic staining showed that the hepatic replacement areas in RBPj cKO mice were significantly larger (Fig. 7B). Flow cytometry indicated that F4/80^hiCD11b^lo TAMs increased significantly in metastatic tumors from RBPj cKO mice (Fig. 7C and D). These cells were more proliferative and expressed higher IL10 (Fig. 7C and D). Taken together, these results indicated that NOTCH blockade in myeloid cells promoted liver metastasis of colorectal cancer.

To confirm the liver specificity of this observation, we established lung metastasis model in control and RBPj cKO mice by injection of luciferase-expressing LLC cells via tail vein (Supplementary Fig. S1D). Different from liver metastasis, live imaging showed that lung metastatic tumors were significantly smaller in RBPj cKO mice than that in the control (Supplementary Fig. S12B and S12C). Macroscopic tumor counting indicated that the lungs of RBPj cKO mice possessed less metastatic foci (>2 mm; Supplementary Fig. S12D and S12E). Flow cytometry indicated that F4/80^hiCD11b^hi TAMs decreased significantly in metastatic tumors from RBPj cKO mice with more F4/80^intCD11b^intLy6C^hi monocytes (Supplementary Fig. S12F–S12H), which suggested blocked differentiation of moTAMs. However, the proliferative capacity of these TAMs (F4/80^hiCD11b^hi) appeared unaffected as determined by Ki67 intracellular staining (Supplementary Fig. S12F and S12I). Taken together, these results indicated that NOTCH blockade in myeloid cells suppressed lung metastasis, which suggested that the phenotype that RBPj cKO promoted tumor growth was liver specific.

To further confirm negative regulation of WNT signaling by Notch activation in TAMs in patient samples, we collected biopsies from patients with HCC with different disease grades (Supplementary Table S1). Immunofluorescence staining of tumor sections showed that in CD68⁺ macrophages, the level of nuclear NICD was negatively correlated with that of β-CATENIN (P < 0.0001; Fig. 7E and F), which positively correlated with higher, more malignant HCC grades (Fig. 7G; Supplementary Fig. S12J). These data suggested that inhibition of NOTCH signaling likely led to enhanced WNT activation in TAMs, which marks higher disease grades in patients with HCC.

Alteration in NOTCH signaling is frequently found when macrophages are challenged by stress-derived damage-associated molecular pattern and infection-derived pathogen-associated molecular pattern signals, which are enriched in liver sinusoids. As for our experimental system, NOTCH signaling was activated in TAMs of orthotopic Hepa1-6 HCC tumors (Supplementary Fig. S1E), possibly by ligands expressed by tumor or microenvironmental cells. Previous reports have shown that NOTCH signaling is of prominent importance to both differentiation and functional plasticity of TAMs in various tumor models (7, 28, 29). In this study, we attempted to access the role of NOTCH signaling in HCC TAMs subsets.

In this study, we disrupted canonical NOTCH signaling in myeloid cells by knockout of Rbpj with LyzM-Cre (33). NOTCH blockade mediated by this Cre transgene resulted in reduced tumor growth in the subcutaneous lung cancer model, likely attributed to reduced monocyte-to-TAM differentiation and hence TAM input, as demonstrated previously (7). However, the scenario appears different in liver, which is the largest reservoir of immunosuppressive-resident macrophages (35). When NOTCH was blocked by myeloid-specific Rbpj ablation, growth of hepatic Hepa1-6 tumors was accelerated accompanied by increased TAMs. The dominating TAMs exhibited certain surface markers of KCs, including F4/80^hiCD11b^loLy6C^lo/−TIM4⁺MARCO⁺, and were tentatively named as kclTAMs in this study. Similar to many other cancers, monocytes are recruited and differentiate into TAMs in CCR2- and NOTCH-dependent ways to promote HCC growth. However, our data have shown that despite impeded monocyte-to-TAM differentiation, disruption of NOTCH signaling increased kclTAMs in HCC instead through enhanced in situ proliferation (Supplementary Fig. S13). It has been documented that KCs are with heterogeneous origins, namely embryonic hematopoiesis–generated KCs and BM monocyte–derived KCs (20, 35). The results reported here could not clarify whether these kclTAMs originate from bona fide KCs, or differentiated from BM-derived or extramedullary-derived monocytes, or even transformed from moTAMs. The possibility existed that in the total TAMs pool of HCC, KCs might account for only a small fraction. More experiments are required to answer this question by using fate-tracing strategies and/or gene expression profiling. In a general way, our data suggests that irrespective of origins, kclTAMs as a whole expanded under the control of NOTCH signaling. NOTCH signaling appears to repress their proliferation, so that when NOTCH was blocked by RBPj cKO, this inhibitory role was canceled. Of note, NOTCH blockade resulted in a 2- to 3-fold decrease of tumor size in breast cancers (7) and subcutaneous LLC (Supplementary Fig. S2), but an approximately 2.5-fold increase in orthotopic HCC, suggesting that kclTAMs compensated functionally the loss of moTAMs under NOTCH deficiency.

Ablation or pharmaceutical inhibition of CCR2 suppresses HCC growth, highlighting the role of CCR2-mediated monocyte recruitment in HCC (4). In our experimental system of orthotopic HCC tumors, however, the phenotype of increased proliferating kclTAMs and accelerated tumor growth in RBPj cKO mice still existed under Ccr2 knockout background. Therefore, although targeting CCR2⁺ monocyte–derived TAMs via CCL2–CCR2 axis has been suggested as a promising therapy for HCC (4), this treatment might not be as effective in the absence of potential NOTCH signaling–mediated suppression of kclTAMs.

In addition, KCs are physiologically immune-suppressive through secretion of IL10 and expression of PD-L1/2 to maintain liver tolerance (35, 39). Notch deficiency increased the expression of IL10 and PD-L1/2 and reduced the expression of IL12 and IL1β in kclTAMs of orthotopic HCC, likely leading to exacerbated protumor activity. Indeed, our data indicated that CD8⁺ cytotoxic T cells decreased in cKO HCC tumors (Fig 1D and E). However, more functional experiments are still needed to investigate the downstream cellular and molecular regulatory mechanisms of RBPj cKO kclTAMs. Therefore, myeloid-specific NOTCH deficiency could not only compensate the decreased TAM input from monocytes with increased local kclTAM proliferation, but could probably also elicit stronger protumor activity of kclTAMs, participating in accelerated HCC growth. Importantly, however, the phenomenon discussed above was only based on the orthotopic Hepa1-6 HCC model, and many other HCC cell lines are still needed to confirm the conclusion.

The WNT/β-CATENIN signaling pathway plays a critical role in liver development, physiology, and pathology (40). WNT ligands are produced by different hepatic cell populations such as liver sinusoidal endothelial cells, and support the replenishment of hepatocytes and regulate their metabolism (41). WNT/β-CATENIN pathway is frequently upregulated in HCC and participates in maintenance of tumor-initiating cells, drug resistance, angiogenesis, and metastasis (42). Upregulated WNT ligands are detected in human HCC and HCC cell lines (43). Our data showed that Hepa1-6 cells secreted WNT3a, a typical ligand activating canonical WNT/β-CATENIN pathway, which could serve as a mitogen for kclTAMs. However, we could not exclude that Hepa1-6 produced other WNT ligands or other types of mitogens regulating the proliferation of kclTAMs. Our data indicate that NOTCH deficiency can result in expansion of kclTAMs by upregulating β-CATENIN, the key mediator of WNT pathway. In addition, WNT signaling also regulates macrophage polarization under different disease contexts (44, 45), and c-MYC has been reported to promote M2-like phenotype of TAMs (38). Consistently, our findings indicated that the strengthened protumor secretory phenotype of Rbpj-deficient kclTAMs appeared to be dependent on upregulated c-Myc downstream to β-catenin (Fig 6E and F). In human samples, increased CD68⁺ TAMs with activated β-CATENIN is associated with advanced tumor stages of HCC patients. These findings add further mechanistic evidence for targeting WNT signaling in patients with HCC.

The mutual regulation of NOTCH and WNT signaling pathways are critically involved in numerous developmental processes, including hematopoiesis, by influencing cell proliferation and differentiation (27). A NOTCH-on/WNT-off or WNT-on/NOTCH-off pattern has been noticed in various signaling settings, in which NOTCH pathway and WNT pathway regulate each other at different signaling levels (27). In this study, NOTCH activation in KCs appeared to suppress WNT signaling by downregulating β-catenin expression at the mRNA level. But detailed effect of NICD/RBPj complex or its downstream HES family transcription regulators on β-catenin promoter and the underlying mechanisms require further clarification. Moreover, we could not exclude that NOTCH may inhibit the transcription mediated by β-catenin through protein–protein interactions, which has been revealed in several models (46). It would be interesting to ask whether the NOTCH–WNT axis functions as a hepatic niche regulating KC homeostasis under physiologic and/or pathologic conditions, as suggested in recent studies (20).

Myeloid-specific Rbpj deficiency promoted the hepatic metastatic growth of CMT93 colorectal cancer with more proliferative kclTAMs, which expressed higher IL10. This phenotype was consistent with that in orthotopic Hepa1-6 HCC. Myeloid-specific Rbpj deficiency in liver seemed to provide a prometastatic niche for hepatic metastasis of colorectal cancer. KCs play bimodal roles in colorectal cancer liver metastasis (47, 48). At early stage, KCs prevent liver metastasis by efficiently killing metastatic cancer cells through phagocytosis. However, the activities of KCs are modulated by other cells including cancer cells and cytokines in liver sinusoids during liver metastasis, often leading to increased colorectal cancer cell arrest and colonization in liver (47). The variation in tumoricidal role of KCs could be somewhat excluded because we did not observe a significant difference in tumor growth between control and RBPj cKO group at early stage of this metastasis model (Supplementary Fig. S12A). Our results suggested that the variation of prometastatic role of KCs might be more important with Rbpj-deficient kclTAMs expressing higher IL10. However, the precise regulatory mechanism of this immunosuppressive cytokine still needs further investigation. Besides, it remains to be elucidated whether RBPj cKO facilitated colonization of colorectal cancer cells in liver at advanced stage and whether activated WNT signaling participate in this metastatic model after myeloid NOTCH blockade. It would also be interesting to determine the corresponding activation status of NOTCH and WNT signals in human biopsies of liver metastasis. Moreover, APC deficiency is a highly frequent driving mutation of colorectal cancer (49). Whether APC mutation, which leads to enhanced WNT signaling, in KCs could facilitate liver metastasis would be worthy of further investigation.

Myeloid NOTCH blockade exerted distinct effects on different metastatic models. Compared with the control mice, the growth of subcutaneous LLC and lung metastasis were both inhibited but that of orthotopic Hepa1-6 HCC and CMT93 liver metastasis both promoted in the RBPj cKO mice. This probably suggested that the phenotype that RBPj cKO promoted tumor growth was a liver-specific phenomenon. Of note, from the metastatic model alone, RBPj cKO promoted liver metastasis but repressed lung metastasis. We considered the contribution of different macrophage subsets as a possible mechanism. As for the liver metastasis, the injected CMT93 cells via spleen arrived at the liver sinusoids and directly interacted with KCs, and NOTCH blockade promoted not only the proliferation but also the TAM phenotype (such as the IL10 secretion) of KCs to facilitate liver metastasis (see Fig. 7C and D). As for the lung metastasis model via tail vein injection, Qian and colleagues reported previously that in the metastasis process of breast cancer cells to the lung, monocytes were recruited and differentiated into TAMs or metastasis-associated macrophages, which facilitated lung metastasis by promoting cancer cell extravasation and survival (50, 51). We supposed that RBPj cKO also impeded the differentiation of moTAMs, and thus lung metastasis was repressed.

No potential conflicts of interest were disclosed.

Conception and design: H.-Y. Qin, H. Han

Development of methodology: J.-L. Zhao, K.-F. Dou, H.-Y. Qin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y.-C. Ye, J.-L. Zhao, C.-C. Gao, Y. Yang, S.-Q. Liang, Y.-Y. Lu, L. Wang, S.-Q. Yue, K.-F. Dou, H.-Y. Qin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y.-C. Ye, J.-L. Zhao, C.-C. Gao, Y.-Y. Lu, L. Wang, S.-Q. Yue, K.-F. Dou, H.-Y. Qin, H. Han

Writing, review, and/or revision of the manuscript: K.-F. Dou, H.-Y. Qin, H. Han

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.-T. Lu, K.-F. Dou, H.-Y. Qin

Study supervision: K.-F. Dou, H.-Y. Qin

This study was supported by National Natural Science Foundation of China (81530018, 31371474, 31570878, 31730041, and 31130019) and the Ministry of Science and Technology of China (2015CB553702).

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.