Cell Communication Network factor 4 promotes tumor‐induced immunosuppression in melanoma

Abstract

Cell Communication Network factor 4 (CCN4/WISP1) is a matricellular protein secreted by cancer cells that promotes metastasis by inducing the epithelial–mesenchymal transition. While metastasis limits survival, limited anti‐tumor immunity also associates with poor patient outcomes with recent work linking these two clinical correlates. Motivated by increased CCN4 correlating with dampened anti‐tumor immunity in primary melanoma, we test for a direct causal link by knocking out CCN4 (CCN4 KO) in the B16F0 and YUMM1.7 mouse melanoma models. Tumor growth is reduced when CCN4 KO melanoma cells are implanted in immunocompetent but not in immunodeficient mice. Correspondingly, CD45⁺ tumor‐infiltrating leukocytes are significantly increased in CCN4 KO tumors, with increased natural killer and CD8⁺ T cells and reduced myeloid‐derived suppressor cells (MDSC). Among mechanisms linked to local immunosuppression, CCN4 suppresses IFN‐gamma release by CD8⁺ T cells and enhances tumor secretion of MDSC‐attracting chemokines like CCL2 and CXCL1. Finally, CCN4 KO potentiates the anti‐tumor effect of immune checkpoint blockade (ICB) therapy. Overall, our results suggest that CCN4 promotes tumor‐induced immunosuppression and is a potential target for therapeutic combinations with ICB.

Cell Communication Network factor 4, a secreted matricellular protein that promotes metastasis in melanoma, also suppresses anti‐tumor immunity via autocrine and paracrine mechanisms. CCN4 is a potential target in combination with immune checkpoint blockade therapy.

Introduction

Immune checkpoint blockade (ICB) has transformed the clinical landscape for treating patients diagnosed with melanoma. While these immunotherapies provide a significant benefit to a portion of the patient population, there is still unmet need for identifying targets that can broaden the clinical benefit of immunotherapies (Darvin et al, ²⁰¹⁸). As immune checkpoint blockade relies on inhibitory signals that can be present both in the local tumor microenvironment and in secondary lymphoid organs to augment anti‐tumor immunity (Binnewies et al, ²⁰¹⁸; Fransen et al, ²⁰¹⁸), an ongoing anti‐tumor immune response is a prognostic indicator of a clinical response (Hamid et al, ²⁰¹¹; Daud et al, ²⁰¹⁶). Collateral targets that transform the tumor microenvironment from immunologically cold to hot are natural next steps. To date, a variety of targets have been identified that heat‐up the anti‐tumor immune response, including targets related to developmental pathways like the Wnt signaling pathway (Spranger et al, ²⁰¹⁵; Luo et al, ²⁰¹⁷; Zhan et al, ²⁰¹⁷).

In addition to an immunologically cold tumor, metastatic dissemination also predicts limited patient survival (Siegel et al, ²⁰²⁰). Metastatic dissemination is attributed to re‐engaging developmental pathways that enable malignant cells to decouple from their tissue niche, migrate via the circulation, extravasate into a peripheral tissue, and establish metastatic colonies (Ye & Weinberg, 2015). In previous work, Cell Communication Network factor 4 (CCN4), a secreted matricellular protein produced by activating the Wnt/β‐catenin pathway, promotes metastatic dissemination in melanoma by engaging the epithelial–mesenchymal transition (Deng et al, ^{2019, 2020}). Interestingly, we observed slower tumor growth rates of CCN4 knock‐out variants in immunocompetent (C57BL/6) compared with immunocompromised (NSG) hosts while the parental cell lines exhibited no difference in growth rates. One explanation for this differential response is that CCN4 inhibits anti‐tumor immunity. A secreted protein that promotes metastatic dissemination and simultaneously inhibits anti‐tumor immunity is intriguing as neutralizing this protein therapeutically would impact two key limiters for patient survival. While a secretome screen piqued our interest in CCN4 (Kulkarni et al, ²⁰¹²), the literature related to CCN4 (a.k.a. WISP1) is not well‐developed with less than 500 publications since 1993 listed in PubMed (Leask, 2020). To clarify the potential role that CCN4 plays at the interface of metastasis and anti‐tumor immunity, the objective of this study was to test for clinical correlates in data obtained from humans diagnosed with melanoma and to assess a causal role for CCN4 in regulating host anti‐tumor immunity more directly using immunocompetent mouse models for melanoma.

Results

CCN4 is associated with a reduced anti‐tumor immune contexture in primary melanoma patients

To assess the clinical context, we first tested for possible connection between CCN4 mRNA expression and overall survival of patients diagnosed with primary melanoma and reported in skin cutaneous melanoma (SKCM) arm of the Cancer Genome Atlas (TCGA). Using data from samples obtained at diagnosis from patients with primary melanoma and with complete survival histories for statistical analysis (n = 95), we stratified patients based on CCN4 expression and summarized their overall survival using Kaplan–Meier survival curves (Fig 1A). A Cox proportional hazards model was used to assess covariance of overall survival with CCN4 expression, tumor stage, age at diagnosis, and gender (Appendix Fig S1A and B). CCN4 expression was the only statistically significant covariate (HR 2.24, P‐value = 0.022).

Figure 1. CCN4 is associated with reduced overall survival of patients diagnosed with primary melanoma and a shift in immune contexture.

1. Kaplan–Meier estimate of overall survival of patients diagnosed with primary melanoma stratified by CCN4 transcript abundance, with patients at risk tabulated below graph, as similarly shown in Deng et al (2019). Original data obtained from SKCM arm of TCGA and stratified based on CCN4 mRNA expression (CCN4 high/positive > 1 FPKM: blue, CCN4 low/negative < 1 FPKM: red). P‐value calculated using the Peto & Peto modified Gehan–Wilcoxon test.
2. The proportion of CCN4‐positive melanoma cells obtained from patients with melanoma obtained prior to treatment (o, n = 15) and from patients that did not respond to immune checkpoint blockade (x, n = 10). Values shown as the proportion of CCN4‐positive melanoma cells of the sample ± SE of the sample proportion, given a binomial distribution. A binomial test assessed significance between the proportion of CCN4 positive cells in the sample relative to a null proportion of 1% or less CCN4‐positive melanoma cells. Samples with significantly enriched CCN4‐positive cells are indicated in red.
3. Immune contexture in corresponding primary SKCM tissue samples estimated from bulk RNAseq data using CIBERSORTx deconvolution. Columns ordered from low (left) to high (right) CCN4 expression. Rows hierarchically clustered based on a Euclidian distance metric in R (Ward.D). A non‐parametric Mann–Whitney U‐test assessed significance of difference in immune subset signature between CCN4 high/positive and low/negative groups. The log10 P‐values were color‐coded.

As the TCGA data are derived from a bulk tissue assay that averages across malignant, stromal, and immune cells present within the tissue sample, we used single‐cell RNA‐seq data obtained from melanoma biopsied prior to treatment (n = 15) and following non‐response to immune checkpoint blockade (ICB, n = 10) to assess the frequency of melanoma cells producing CCN4 (Fig 1B). Assuming a null hypothesis of 1% or less of CCN4‐positive cells could be observed by random chance, a binomial test was used to assess whether the proportion of CCN4‐positive cells in each patient sample is greater than the null hypothesis. The samples with a frequency of CCN4‐positive cells greater than 1% with 95% confidence were considered CCN4‐positive. In the treatment naïve cohort, 4 of 15 samples had CCN4‐positive tumors. In the ICB‐resistant cohort, 5 of 10 samples had CCN4‐positive tumors. In comparison, the TCGA SKCM primary melanoma cohort had 26 of 95 samples with CCN4‐positive tumors (CCN4 reads greater than 1 FPKM). Using a Fisher exact test, the difference in CCN4‐positive samples in the treatment naïve group (4 of 15) compared to the TCGA SKCM cohort (26 of 95) was not statistically different (P‐value = 1, odds ratio = 0.965). While the prevalence of CCN4‐positive samples in the ICB‐resistant cohort (5 of 10) seemed to be higher compared to treatment naïve and TCGA SKCM cohorts (P‐value = 0.1535, odds ratio = 2.64), we are unable to conclude with confidence that these samples come from different populations as n = 10 is not sufficiently powered to detect a difference in prevalence between 27 and 50%. Collectively, the results suggest that (i) the frequency of CCN4‐positive tissue samples is similar among all the cohorts, (ii) patients with higher levels of CCN4 expression have a worse outcome, and (iii) separating the population into high and low CCN4 expression subsets is not masking for other common latent variables that may influence overall survival, such as age, sex, and tumor stage.

Given that the clinical data suggest that melanoma patients with high CCN4 expression have a worse outcome, we used digital cytometry to infer whether changes in immune contexture also corresponded with changes in CCN4 expression (Fig 1C). In the SKCM TCGA dataset, gene signatures associated with CD8 T cells (P‐value < 0.01), activated NK cells (P‐value < 0.01), follicular T helper cells (P‐value < 0.05), and lymphocytes (P‐value < 0.05) were reduced while M0 macrophages (P‐value < 0.05), resting NK cells (P‐value < 0.05), resting memory CD4⁺ T cells (P‐value < 0.05), and macrophages (P‐value < 0.01) were increased in the high CCN4 cohort. Collectively, characterizing the immune contexture in human melanoma using digital cytometry suggests that increased CCN4 expression corresponds with a shift in immune response from active anti‐tumor immunity to a dampened immune response with enhanced resting and undifferentiated immune cell phenotypes. Animal models may help clarify mechanistic underpinnings of these clinical observations.

CCN4 knockout in mouse melanoma reduces subcutaneous tumor growth

To analyze the importance of CCN4 for subcutaneous (s.c.) tumor growth and anti‐tumor immune response, we used two mouse melanoma models: the spontaneous B16F0 model and the more clinically relevant YUMM1.7 model displaying Braf ^{V 600 E / WT} Pten^−/−Cdkn2^−/− genotype. B16F0 and YUMM1.7 parental cells secreted 605 ± 15 pg/ml and 451 ± 25 pg/ml, respectively, of CCN4 in 2D culture media. CCN4 KO variants of the B16F0 and YUMM1.7 cell lines were generated through CRISPR/Cas9 methodology and produced undetectable levels of CCN4 under similar culture conditions (Fig 2A). A control for puromycin selection (B16F0‐Ctr) produced CCN4 similar to the parental cell line.

Figure 2. CCN4 knockout suppressed melanoma tumor growth in immunocompetent mice.

• A
  CCN4 secretion after CRISPR/Cas9 knockout in B16F0 and YUMM1.7 cell lines. Cell culture media conditioned for 48 h by the indicated cell line were tested by ELISA (n.d., not detected). Three biological replicates represented as mean ± SEM.
• B–E
  B16F0 knockout created with Homology‐Directed Repair approach with controls for puromycin selection (B16F0‐Ctr) created with pBabe‐puro retrovirus. YUMM1.7 knockout created with double nickase approach with KO1 and KO2 indicating two different clones. Wildtype (WT: red) B16F0 (B, C) and YUMM1.7 (D, E) cells and CCN4 knockout variants (KO1: blue, KO2: black) were subcutaneously injected into (B, D) C57BL/6 immunocompetent and (C, E) NSG immunocompromised mice. Tumor volumes measured by caliper as a function of time after tumor challenge, with n expressed as the number of mice with tumors over the number of mice injected. The tumor growth profiles for every mouse in the different cohorts is provided in Dataset EV1.
• F–H
  The tumor growth model used to interpret the log‐linear tumor growth curves (F). Posterior distributions in the growth rate constant of tumors in NSG mice (G) and in the log‐ratio of the net growth rate constants of tumors in NSG relative to C57BL/6 mice (H). Statistical differences in these distributions were assessed using a chi‐squared test summarized in text.
• I
  In a time‐matched experiment, average tumor weights of both B16F0 and YUMM1.7 were significantly decreased following CCN4 knockout (KO1). Results summarized as mean ± SEM.
• J
  Significant increases in the number of CD45⁺ cells were observed in both B16F0 and YUMM1.7 cell lines following CCN4 knockout (KO1). Results summarized as mean ± SEM.

Data information: In I and J, statistical significance was assessed using a Student’s t‐test with *: 0.01 < P < 0.05, **: 0.001 < P < 0.01, ***: P < 0.001 and results representative of at least six biological replicates.

Following s.c. implantation, we compared the growth trajectories of CCN4 KO variants with parental B16F0 (Fig 2B and C) and YUMM1.7 (Fig 2D and E) cells in immunocompetent C57BL/6 (Fig 2B and D) and in severely immunocompromised NSG (Fig 2C and E) mice. Tumor growth consistently followed a log‐linear growth trajectory, which implies that tumor size at any time point depends on the initial bolus of tumor‐initiating cells (C_{T 0}) and the net growth rate constant (k = k_G−k_D ) (Fig 2F). Between these two parameters, the log growth rate constants were more consistent among replicates than the initial bolus of tumor‐initiating cells, despite injecting the same number of cells. As genetic editing of the cells can alter the intrinsic growth rate, we used tumor growth in NSG mice to estimate the intrinsic growth rate constants (k_G ) for WT and CCN4 KO variants (Fig 2G). We noted that CCN4 KO increased k_G for both cell lines (chi‐squared P‐value < 1E−5). In immunocompetent mice, we assumed that the net growth rate constant reflects both the intrinsic growth rate constant and a loss in tumor size due to immune‐mediated cell death (k_D ), which lumps all mechanisms for anti‐tumor immunity that are present in C57BL/6 mice and absent in NSG mice into a single constant. Of note, any residual anti‐tumor immune function present in NSG mice, albeit nearly negligible, is included in the intrinsic growth rate constant. We can infer the impact of CCN4 KO on immune‐mediated cell death by comparing the difference in net growth rate constant for a given cell line in NSG versus C57BL/6 mice, which here is expressed as a log‐ratio (Fig 2H). Generally, YUMM1.7 variants exhibited lower log‐ratios compared to B16F0 variants (e.g., median log‐ratio B16F0 WT = 0.1991 versus YUMM1.7 WT = 0.063), which is consistent with genetically engineered mouse models being less immunogenic compared to spontaneous tumor models. More importantly, CCN4 KO variants of both cell lines exhibited greater log‐ratios relative to WT cell lines (chi‐squared P‐values all < 1E‐5). KO1 variants for both cell lines were used for subsequent experiments. Of note, B16F0 variants with a knock‐out of DNA (cytosine‐5)‐methyltransferase 3A (DNMT3A‐KO), which served as a CRISPR/Cas9 editing control, exhibited no difference in tumor growth and overall survival (Appendix Fig S2A–E) compared with parental B16F0 cells. After 14 days following tumor challenge in the B16F0 model, tumors were surgically removed and weighted, with CCN4 KO tumors being 1.7‐fold smaller than WT tumors (Fig 2I, P‐value = 0.0003). After 28 days in the YUMM1.7 model, excised CCN4 KO tumors were 3‐fold lighter than WT tumors (Fig 2I; P‐value < 0.0001). Given the different growth response to CCN4 knockout in C57BL/6 versus NSG mice, the results suggest that CCN4 plays a role in modulating the immune system to favor tumor development. Supporting this idea, a 2‐fold increase of infiltrating CD45⁺ leukocytes was detected in CCN4 KO tumors when compared to B16F0‐WT (P‐value = 0.0297) and YUMM1.7‐WT (P‐value = 0.0017) tumors (Fig 2J). Subsequent experiments focused on clarifying CCN4’s role in modulating anti‐tumor immunity.

CCN4 knockout increases CTL and NK effector cell frequency while reducing MDSC infiltration in melanoma tumors

Next, we used flow cytometry to characterize tumor‐infiltrating leukocytes using three different antibody panels that focused on T cells, NK cells, and myeloid cell subsets (Fig 3). In resolving lymphoid subsets, we found that CD3e⁺ T cells (Live CD45⁺ CD3e⁺ events, Fig 3A and B; P‐value = 0.0046), CD4⁺ T cells (Live CD45⁺ CD3e⁺ CD4⁺ CD8a⁻ events, Fig 3A and C; P‐value = 0.0006), CD8⁺ T cells (Live CD45⁺ CD3e⁺ CD8a⁺ CD4⁻ events, Fig 3A and D; P‐value = 0.0004), and NK cells (Live CD45⁺ CD3e⁻ NK1.1⁺ B220^{−/ lo} events, Fig 3A and E; P‐value = 0.0104) were increased in CCN4 KO tumors compared to YUMM1.7‐WT tumors. In terms of the myeloid compartment, a subset of macrophages (Live CD45⁺ CD11b⁺ CD11c⁺ Gr1⁻ F480⁺ MHCII⁺ events, Fig 3F; P‐value < 0.0001) and neutrophils (Live CD45⁺ CD11b ^lo CD11c⁻ Gr1⁺ events, Fig 3A and G; P‐value = 0.0004) were significantly expanded in the CCN4 KO tumors. Interestingly, these macrophages were CD11c⁺ (Fig 3A), a marker associated with pro‐inflammatory M1 TAM (Jeong et al, ²⁰¹⁹). Compared to CD11c⁺ subset, CD11c⁻ macrophages were 100‐times less abundant and not statistically different. No significant differences were detected in the number of dendritic cells (DC: Live CD45⁺ CD11b⁻ CD11c⁺ Gr1⁻ F480⁻ events, Fig 3H; P‐value = 0.0699). We also noted the number of CD8a⁺ DCs (Live CD45⁺ CD3e⁻ CD8a⁺ events) was negligible. Conversely, putative myeloid‐derived suppressor cells (MDSC) were reduced (Live CD45⁺ CD11b⁺ Gr1⁺ events, Fig 3A and I; P‐value = 0.0078), which resulted in the CCN4 KO tumors having 4.4‐fold (P‐value < 0.0001) and 3.5‐fold (P‐value = 0.0013) higher CD8⁺ T cells/MDSC and NK/MDSC ratios, respectively (Fig EV1F–H). Similar results were obtained with the B16F0 tumor model (Fig EV1A–H).

Figure 3. CCN4 knockout increased CTL and NK effector cells and decreased MDSC in the tumor microenvironment.

• A
  Representative flow cytometry data of the frequency of tumor‐infiltrating CD4⁺ and CD8⁺ T cells, NK effector cells, tumor‐associated neutrophils (TAN) and MDSC from the live CD45⁺ compartment in WT and CCN4 knockout (KO1) YUMM1.7 tumors. Gating strategies are illustrated in Appendix Figs S4 and S5. In the right panel, the contour lines enclose 90% of TANs (red line) and MDSC (blue) that were determined based on gating shown in Appendix Fig S5 but here backgated along the GR1/CD11b axes.
• B–I
  Graphs summarizing the change in frequency of immune cells in the YUMM1.7 tumor microenvironment following CCN4 knockout (n = 6 /group). (B) CD3⁺, (C) CD4⁺, (D) CD8⁺, (E) NK cells, (F) CD11c⁺ TAMs, (G) TANs, (H) DCs, and (I) MDSC. Statistical significance was assessed using a Student’s t‐test using six biological replicates with results annotated with *: 0.01 < P < 0.05, **: 0.001 < P < 0.01, and ***: P < 0.001. Results summarized as mean ± SEM.
• J
  CCN4 secretion from CCN4‐inducible cells and associated control cell lines in conditioned media with or without 0.5 μg/ml doxycycline, as measured by ELISA (n.d., not detected). Results from three biological replicates summarized as mean ± SEM.
• K
  Kaplan–Meier summary of the fraction tumor‐free in a 2x2 factor experimental design (n = 5 / group), where Tet‐on vector control (red) versus inducible mCCN4 vector (black) and in the presence (dotted lines) or absence (solid lines) of doxycycline were the two factors.
• L
  Summary of a Cox multivariate proportional hazards analysis where x‐axis of the dot‐and‐whisker plot corresponds to the mean hazard ratio and 95% confidence interval for the indicated variable.
• M
  Representative flow cytometry data of the frequency of tumor‐infiltrating MDSC (GR1⁺ CD11c^−/lo: dotted box) from the live CD45⁺ CD11b⁺ compartment isolated from CCN4‐induced rescue (ID mCCN4⁺ Dox) and CCN4 knockout (YM1.7‐KO1) YUMM1.7 tumors.
• N
  Frequency of MDSC for similarly sized tumors generated from CCN4‐induced rescue (gray triangle), CCN4 knockout (blue square), and WT YUMM1.7 cells (red circles). Pairwise differences were assessed using a two‐sided homoscedastic (pooled variance) Student’s t‐test.

Figure EV1. Comparison of TILs in mice receiving wt versus CCN4 KO B16F0 cells.

• A
  Representative flow cytometry panels comparing CD4⁺ T cells (Live CD45⁺ CD3⁺ CD8⁻ CD4⁺ events), CD8⁺ T cells (Live CD45⁺ CD3⁺ CD8⁺ CD4⁻ events), NK (Live CD45⁺ CD3⁻ NK1.1⁺ events), and NKT (Live CD45⁺ CD3⁺ NK1.1⁺ events) cells, and MDSC (Live CD45⁺ CD11b⁺ Gr1⁺ events) infiltration in tumors derived from WT and CCN4 KO B16F0 cells (left to right columns).
• B–E
  Summary figures for the number of CD3⁺ (B), CD8⁺ (C), and CD4⁺ (D) T cells and NK cells (E) per mm³ tumor (n ≥ 6 mice/group).
• F–H
  Comparison in the change of CD3⁺ T cell:MDSC (F), CD8⁺ T cell:MDSC (G), NK cell:MDSC (H) ratios upon CCN4 KO between the B16F0 and YUMM1.7 models.

Data information: *0.01 < P < 0.05, **0.001 < P < 0.01, ***P < 0.001. Results representative of at least six biological replicates. Statistical significance was assessed using a Student’s t‐test and error bars represent SEM.

Source data are available online for this figure.

To test whether reintroducing CCN4 into CCN4 KO cells rescues the phenotype, we created a CCN4‐inducible variant of CCN4 KO YUMM1.7 cells under the control of doxycycline and a vector control. CCN4 expression was under stringent control of doxycycline, with induced levels similar to wild‐type YUMM1.7 cells (Fig 3J). C57BL/6 mice were challenged with the Tet‐on inducible CCN4 (ID mCCN4) and vector control variants of the YUMM1.7 CCN4 KO cells (Ym1.7‐KO1) and doxycycline using a 2×2 experimental design (Fig 3K). Tumor‐free survival of the mouse cohort was regressed using a Cox proportional hazards model jointly to the presence of the ID mCCN4 expression vector, treatment with DOX, and the combination (Fig 3L). As expected, genetic editing of the Ym1.7‐KO1 cells to include the ID mCCN4 vector reduced the hazard ratio (P‐value < 1e−10) while the treating with doxycycline had no significant effect (P‐value = 0.288). Re‐expression of CCN4 significantly increased tumor development (P‐value < 1e−10). The presence of putative tumor‐infiltrating MDSC was increased by CCN4 re‐expression compared with similarly sized WT and CCN4 KO tumors (Fig 3M and N, P‐value = 0.0022). Collectively, these data suggest that CCN4 skews the immune contexture within the melanoma microenvironment by reducing the infiltration of cells with potential anti‐tumor cytolytic activity, namely CD8⁺ T cells and NK cells, and by enhancing the prevalence of cells with immunosuppressive capacity, namely MDSC.

Melanoma‐produced CCN4 promotes the splenic expansion of G‐MDSC

Considering the spleen’s role in tumor‐induced immunosuppression and particularly for MDSC extramedullary generation (Ugel et al, ²⁰¹²; Jordan et al, ²⁰¹⁷), we analyzed the frequency of MDSC in the spleen of CCN4 KO and WT YUMM1.7 tumor‐bearing (TB) mice. Splenic MDSC were first compared on day 28 after tumor challenge (time‐matched, Fig 4A), where both the percentage (Fig 4B, P‐value = 0.0074) and number of MDSC per gram of spleen (Fig 4C, P‐value = 0.0019) were significantly reduced in CCN4 KO TB mice. Splenomegaly was also notable in mice bearing WT YUMM1.7 tumors (Appendix Fig S3A–C). Moreover, the percentage of CD45⁺ cells and total number per gram spleen of CD11b⁺Gr1⁺ cells in mice with CCN4 KO tumors were similar to the normal values detected in tumor‐free mice (Fig 4B and C, P‐value = 0.8886 and 0.5042, respectively). However, in matching the time point, the results raised an additional concern as CCN4 KO tumors were 3.7‐fold smaller than YUMM1.7‐WT (Fig 4A, P‐value = 0.0277) and MDSC prevalence commonly correlates with tumor burden.

Figure 4. CCN4 expanded MDSC in the spleens of tumor‐bearing mice.

• A–C
  Tumor volume of YUMM1.7 WT and CCN4 KO (KO1) compared when size‐matched and time‐matched on day 28 after tumor challenge. MDSC in the spleens of tumor‐bearing mice expressed as percentage (B) or weight‐basis (C) observed in size‐matched versus time‐matched experimental designs (n = 5 for size‐matched, 4 for time‐matched experiments). Statistical significance was assessed using a Student’s t‐test with results annotated with *: 0.01 < P < 0.05, **: 0.001 < P < 0.01, and ***: P < 0.001. Results summarized as mean ± SEM.
• D
  Representative flow cytometry data of CD11b^+/−/GR‐1^+/− cells found in WT, KO1, and tumor‐free (TF) groups.
• E–H
  Graphs summarizing G‐MDSCs (E, G) and M‐MDSCs (F, H) in the spleens of time‐matched YUMM1.7 WT and CCN4 KO (KO1) tumor‐bearing mice (n = 4/group) expressed as percentage of CD45⁺ (E, F) and total number per gram spleen (G, H). Statistical significance was assessed using a Student’s t‐test with results annotated with *: 0.01 < P < 0.05, **: 0.001 < P < 0.01, and ***P < 0.001. Results summarized as mean ± SEM.
• I
  Representative flow cytometry data of CD11b⁺/Ly6C^hi/low/Ly6G^+/− cells found in WT, KO1, and TF groups.
• J, K
  Suppression of naïve CD8⁺ T‐cell proliferation by (J) freshly isolated G‐MDSCs and (K) total splenocytes from WT, KO1, and TF groups (n = 3 biological replicates/group) on a per cell basis. ANOVA with post hoc Tukey tests were used to assess statistical significance, where different letter (a, b, or c) denotes statistically different groups.
• L
  Cell trace profiles of live CD8⁺ T cells contained within CD45⁺ cells extracted from WT (red) and KO1 (blue) tumors in a time‐matched experiment (TME) and stimulated for three days in vitro with αCD3/αCD28‐loaded beads (representative of n = 3 biological replicates/group), with unstimulated cells as a staining control (gray).

Data information: In panels A‐C and E‐H, dotted line represents average value found in tumor‐free mice.

To test whether CCN4 KO impairs splenic MDSC expansion independent of tumor burden, we analyzed MDSC in mice bearing CCN4 KO and WT YUMM1.7 tumors of similar size (size‐matched, Fig 4A, P‐value = 0.0843). Once again, significant decreases in the percentage (P‐value = 0.0006) and number per gram of spleen (P‐value = 0.0007) of MDSC were observed in CCN4 KO TB mice compared to WT YUMM1.7 counterparts (Fig 4B–D). Given that CCN4 KO reduced MDSC in both time‐matched and size‐matched TB mice, we asked whether CCN4 had a differential effect on the two main MDSC subpopulations. Using a size‐matched experimental design, CD11b⁺Ly6C ^low Ly6G⁺ granulocytic MDSC (G‐MDSC) were the majority of the splenic MDSC in mice with WT YUMM1.7 tumors (Fig 4E–I). Interestingly, a 2.7‐fold and 2.9‐fold reduction was observed in the percentage (Fig 4E; P‐value = 0.0009) and the number (Fig 4G; P‐value = 0.0003) of G‐MDSC per gram of spleen, respectively. In addition, no differences in the frequency of splenic CD11b⁺Ly6C ^hi Ly6G⁻ monocytic MDSC (M‐MDSC) were found in the absence of tumor‐derived CCN4 (Fig 4F and H). We also noted that G‐MDSC (CXCR2⁺ CCR2⁻ CD11b⁺ GR1⁺) were predominantly captured in our initial MDSC gating based on GR1/CD11b staining (Appendix Fig S3D).

Next, we tested whether splenic G‐MDSC, isolated from KO CCN4 and WT YUMM1.7 TB mice, could suppress na¨ıve CD8⁺ T‐cell proliferation stimulated by αCD3/αCD28‐loaded beads (Fig 4J). In a time‐matched design, fresh G‐MDSC induced by CCN4 KO tumors were significantly less suppressive on a per cell basis than those isolated from WT YUMM1.7 TB mice. In addition, CD11b⁺Ly6C ^low Ly6G⁺ cells isolated from tumor‐free (TF) mice were unable to suppress CD8⁺ T cell proliferation. While G‐MDSC were the predominant MDSC subset in TB mice (7% of splenocytes), we also evaluated CD8⁺ T‐cell proliferation after stimulating total splenocytes from TB mice with αCD3/αCD28‐loaded beads (Fig 4K). The precursor frequency of proliferative CD8⁺ T cells (KO1: 0.357 ± 0.022 vs. WT: 0.220 ± 0.011, P‐value = 0.0007) and the fraction that divided at least once (KO1: 0.700 ± 0.020 vs. WT: 0.502 ± 0.028, P‐value = 0.0006) were significantly increased when splenocytes from CCN4 KO TB mice were compared with splenocytes from mice with WT YUMM1.7 tumors. Once cell division occurred, CD8⁺ T splenocytes from CCN4 KO TB mice also had higher indices of proliferation (KO1: 1.72 ± 0.02 vs. WT: 1.52 ± 0.05, P‐value = 0.0040) and division (KO1: 0.61 0.03 vs. WT: 0.33 0.03, P‐value = 0.0002).

When tumor‐infiltrating CD45⁺ cells were stimulated in similar fashion (Fig 4L), live CD8⁺ T cells extracted from CCN4 KO and WT YUMM1.7 tumors had the same precursor frequency of proliferative cells (KO1: 0.884 ± 0.051 vs. WT: 0.884 ± 0.029, P‐value = 0.998) and fraction that divided at least once to in vitro stimulation (KO1: 0.985 ± 0.008 vs. WT: 0.971 ± 0.009, P‐value = 0.109). While CD8⁺ T cells were less prevalent in WT compared to CCN4 KO tumors, proliferation on a per cell basis was also reduced. Specifically, CD8⁺ T cells extracted from CCN4 KO tumors had higher indices of proliferation (KO1: 2.26 ± 0.08 vs. WT: 1.63 ± 0.08, P‐value = 0.0006) and division (KO1: 2.00 ± 0.16 vs. WT: 1.45 ± 0.11, P‐value = 0.009). While we observed a similar effect on CD8 T‐cell proliferation in WT versus CCN4 KO systems when using isolated G‐MDSC as with splenocytes or TILs, the presence of T regulatory cells is a potential additional suppressive cell type in the mixed cell assays. We do note, though, that MDSC were 10‐times more abundant than CD4⁺ T cells in analyzing TIL populations in WT YUMM1.7 tumors. Moreover, we observed no difference in T regulatory cell fraction (Live CD45⁺ CD4⁺ FOXP3⁺ CD25⁺ events) within the CD4⁺ TIL compartment upon CCN4 KO (P‐value = 0.566, Fig EV2A–D). Collectively, these results suggest that melanoma‐derived CCN4 contributes to the splenic expansion of immunosuppressive G‐MDSC.

Figure EV2. The fraction of T regulatory cells (Live CD45⁺ CD4⁺ FOXP3⁺ CD25⁺) within the CD4⁺ TIL compartment was not changed upon CCN4 KO.

• A–D
  In a time‐matched experiment, TILs (A, n = 3 biological replicates/group) and splenocytes (B, n = 4 biological replicates/group) obtained from mice bearing WT (top panels) and CCN4 KO (KO1—bottom panels) YUMM1.7 tumors were analyzed for the presence of T regulatory cells (Tregs) by flow cytometry. Results are summarized in terms of the fraction of Tregs within total live CD45⁺ events (C) and the fraction of Tregs within the total live CD45⁺ CD4⁺ events (D). Statistical significance was assessed using a Student’s t‐test and error bars represent the standard deviation.

Source data are available online for this figure.

CCN4 stimulates the secretion of lactate and MDSC‐attracting chemokines by melanoma cells and directly inhibits CD8⁺ T cells

To identify mechanisms associated with CCN4‐mediated immunomodulation, we analyzed the cytokines, chemokines, and growth factors produced by CCN4 KO and parental WT YUMM1.7 cells in vitro using R&D Systems Mouse XL Cytokine Array (Fig 5A). Interestingly, CCL2 and CXCL1 chemokines were among the most down‐regulated proteins in the tumor‐conditioned media (TCM) from CCN4 KO tumor cells with high basal expression. These chemokines have been previously associated with MDSC recruitment to the tumor and their splenic accumulation (Chun et al, ²⁰¹⁵; Taki et al, ²⁰¹⁸; Wu et al, ²⁰¹⁸). The reduction in CCL2 and CXCL1 were confirmed by ELISA in the TCM of CCN4 KO cells compared to WT YUMM1.7 counterparts (Fig 5B and C, P‐value = 0.0001), as well as in media conditioned by CD45⁻ cells that were isolated from CCN4 KO and WT YUMM1.7 tumors, after 36 h of culture ex vivo (Fig 5D and E, P‐value = 0.0151 and P‐value < 0.0001, respectively). Additionally, CCL2 and CXCL1 were significantly diminished in the serum from CCN4 KO TB mice compared to mice with WT YUMM1.7 tumors (Fig 5F and G). In fact, CCL2 and CXCL1 serum concentrations in CCN4 KO TB mice were not different from normal levels detected in tumor‐free mice (Fig 5F and G). We also observed that G‐MDSC expressed a receptor for CXCL1: CXCR2 (Fig EV3).

Figure 5. Knockout of CCN4 down‐regulated CCL2 and CXCL1 expression and decreased glycolysis and glycolytic capacity in YUMM1.7 melanoma cells.

• A
  Cytokine, chemokine, and growth factor expression by CCN4 KO and WT YUMM1.7 in vitro assayed using R&D Systems' Mouse XL Cytokine Array. Orange dots represent results for specific cytokine probes while blue dots represent positive and negative controls. Dotted lines enclose a null distribution estimated from positive and negative controls.
• B–G
  ELISA results for CCL2 and CXCL1 secretion after 36 h ex vivo culture (n = 3 biological replicates/group). CCL2 assayed in (B) tumor‐conditioned medium (TCM), (D) CD45⁻ medium, and (F) serum. CXCL1 assayed in (C) TCM, (E) CD45⁻ medium, and (G) serum.
• H, I
  Live‐dead staining of CD45⁺ cells isolated from WT and CCN4 KO tumors. Statistical significance was assessed using a Student’s t‐test with results annotated with ***P < 0.001. Results of biological replicates summarized as mean ± SEM.
• J, K
  Analysis of the extracellular acidification rate (ECAR) associated with (J) glycolysis and (K) glycolytic capacity assayed by Seahorse Analyzer in WT and CCN4 KO tumors (n ≥ 12 biological replicates/group) after 36 h ex vivo. Statistical significance was assessed using a Student’s t‐test with results annotated with ***P < 0.001. Results of biological replicates summarized as mean ± SEM.
• L
  Lactate production by isolated CD45⁻ cells after 36 h ex vivo culture (n = 4 biological replicates/group). Statistical significance was assessed using a Student’s t‐test with results annotated with ***P < 0.001. Results of biological replicates summarized as mean ± SEM.

Figure EV3. Glycolysis and glycolytic capacity in YUMM1.7 and B16F0 cells were reduced upon CCN4 KO.

• A, B
  Time course of ECAR analysis using XFe96 Seahorse Analyzer in WT and CCN4 KO tumors derived from YUMM1.7 (A) and B16F0 cells (B) after 36 h ex vivo (n = 3, results representative of one of three biological replicates). Curves represent average ± standard deviation at each time point of three technical replicates.
• C, D
  Time course profiles were analyzed to estimate (C) glycolysis and (D) glycolytic capacity for WT and CCN4 KO tumors derived from B16F0 cell variants. ***P < 0.001. In (C and D), statistical significance was assessed using a Student’s t‐test and error bars represent SEM of seven biological replicates.

Source data are available online for this figure.

IGFBP‐3 (Insulin‐Like Growth Factor‐Binding Protein 3) was another protein showing a significant change between CCN4 KO and WT YUMM1.7 conditioned media with high basal expression (Fig 5A). Considering that both IGFBP‐3 and CCN4 play a role in glycolysis regulation (Muzumdar et al, ²⁰⁰⁶; Mireuta et al, ²⁰¹¹; Ferrand et al, ²⁰¹⁷; Wang et al, ²⁰¹⁹) and that the viability of tumor‐infiltrating CD45⁺ cells was reduced in WT YUMM1.7 tumors when compared to CCN4 KO counterparts (Fig 5H and I, P‐value < 0.0001), we measured the extracellular acidification rate (ECAR) associated with glycolysis and glycolytic capacity in WT and CCN4 KO YUMM1.7 cells using a Seahorse Analyzer (Figs 5J and K and EV3A). Of note, glycolysis (P‐value = 0.0006) and glycolytic capacity (P‐value = 0.0002) were significantly reduced in the absence of CCN4. Similar reduction in glycolysis parameters in CCN4 KO cells was observed in the B16F0 tumor model (Fig EV3B–D).

One consequence of aerobic glycolysis in tumor cells is the release of lactate that acidifies the tumor microenvironment (Huber et al, ²⁰¹⁷), suppresses or induces apoptosis of tumor‐infiltrating lymphocytes (Calcinotto et al, ²⁰¹²; Huber et al, ²⁰¹⁷), and expands MDSC (Husain et al, ²⁰¹³). Therefore, lactate production from CD45⁻ cells, isolated from CCN4 KO and WT YUMM1.7 tumors, was evaluated after 36 h of ex vivo culture. As shown in Fig 5L, CCN4 KO tumor cells secreted less lactate into the extracellular milieu (P‐value < 0.0001), which is consistent with reduced glycolysis and differences in TIL viability.

Given the observed increase in CD8⁺ T cells upon CCN4 KO, we next tested whether CCN4 had a direct impact on CD8⁺ T‐cell function through quantifying target‐specific ex vivo cytokine release. To generate YUMM1.7‐reactive CD8⁺ T cells, we immunized C57BL/6 mice using subcutaneous injection of irradiated YUMM1.7 cells and boosted with live YUMM1.7 cells three days before isolating CD8⁺ T cells. As target cells, we used the CCN4‐inducible variant of CCN4 KO YUMM1.7 cells under the control of doxycycline and the corresponding vector control. IFNγ ELISpots were used to quantify the CD8⁺ T‐cell functional response to the different tumor targets in the absence or presence of tumor‐produced CCN4 (Fig 6A–C). As expected, the highest IFNγ and lowest TNFα responses were against WT and CCN4 KO YUMM1.7 cells, with a seemingly higher IFNγ response to WT YUMM1.7 targets (Fig 6C, P‐value = 0.098). Interestingly, re‐expression of CCN4 by CCN4 KO YUMM1.7 cells following doxycycline induction significantly reduced both IFNγ and TNFα production (P‐value < 0.001), which suggests that CCN4 directly inhibits CD8⁺ T‐cell function.

Figure 6. CCN4 directly inhibited CD8⁺ T cell function.

1. ELISpot for IFNγ release by CD8⁺ T cells using parental YUMM1.7 and CCN4 KO YUMM1.7 (KO1) cells as targets and different amount of in vivo activated CD8⁺ T cells. Statistical significance was assessed using a Student’s t‐test with results annotated with ***P < 0.001. Results of three biological replicates summarized as mean ± SEM.
2. ELISpot for IFNγ release by in vivo activated CD8⁺ T cells with CCN4‐inducible cells as targets in the presence or absence of 0.5 mg/ml doxycycline. Statistical significance was assessed using a Student’s t‐test with results annotated with ***P < 0.001. Results of biological replicates summarized as mean ± SEM.
3. CD8⁺ T cells isolated from the spleens of C57BL/6 mice that rejected YUMM1.7 tumors were assayed by in vitro ELISpot using variants of the YUMM1.7 cell line as targets (WT YUMM1.7 (Ym1.7)—yellow, CCN4 KO YUMM1.7 (Ym1.7‐KO1)—light green, CCN4 KO YUMM1.7 with a blank inducible expression vector (Ym1.7‐KO1‐IDvector)—dark green and blue, CCN4 KO YUMM1.7 with a CCN4 inducible expression vector (Ym1.7‐KO1‐IDmCCN4)—purple and red). Representative images shown under indicated condition with scale bar indicating 1 mm. Variants containing the inducible expression vector were also cultured in the absence (dark green and purple) or presence of 0.5 μg/ml doxycycline (blue and red). CD8⁺ T cells expressing IFNγ and TNFα were quantified following 24‐h co‐culture (bar graph). Results shown as mean ± S.D. for three biological replicates. Statistical significance was assessed using a Student’s t‐test with results annotated with ***P < 0.001 and n.s.: P > 0.05.

Taken together, these results indicate that melanoma‐derived CCN4 directly inhibits CD8⁺ T‐cell function but also has autocrine effects to produce CXCL1 and CCL2 and to stimulate glycolysis and lactate secretion by tumor cells. While cited studies describe the functional implications of these observations, our data suggest, to us, that the autocrine effects of CCN4 contribute to MDSC expansion and recruitment, as well as to the apoptosis of tumor‐infiltrating CD45⁺ cells. A direct inhibitory effect of CCN4 on DC differentiation and maturation were not observed (Fig EV4A–F). As CCN4 is expressed heterogeneously among human melanoma cells (Fig 1B), additional work will be required to parse how autocrine versus paracrine effects propagate from CCN4‐expressing to non‐CCN4‐expressing cells within the tumor microenvironment.

Figure EV4. CCN4 has no direct inhibitory effect on dendritic cell maturation and differentiation.

• A–F
  Bone marrow cells were harvested from femurs and tibias of C57BL/6 mice. For dendritic cell (DC) preparation, 6 × 10⁵ cells/well were cultured for 6 days with 20 ng/ml of GM‐CSF (eBioscience, Thermo Fisher) in 6 well plates. On day 6, the media was removed and DC maturation was induced using 1 μg/ml of LPS (Sigma). Medium conditioned by WT (TCM WT) and CCN4/WISP1 KO (TCM KO) melanoma B16F0 cells was added at 50% final volume during differentiation and maturation of the DC, whereas rmCCN4 (WISP1, R&D) was added at a final concentration of 10 ng/ml. After 24h of LPS treatment, DC were extracted, washed, and incubated with Mouse BD Fc Block (BD Biosciences). Using the gating strategy illustrated in (A), the following antibodies were used to analyze by flow cytometry the efficiency of the DC generation (B–D) and the maturation of these cells (E, F): anti‐mouse CD11c/PE (eBioscience, Thermo Fisher), anti‐mouse IA/IE (MHCII)/AlexaFluor 700 (BioLegend), anti‐mouse CD11b/PerCP‐Cy5.5 (eBioscience, Thermo Fisher), anti‐mouse Gr1/APC (BioLegend), anti‐mouse CD40/FITC (eBioscience, Thermo Fisher), and anti‐mouse CD86/V450 (BD Biosciences). Experiments performed using biological triplicates and b indicates P < 0.01 assessed by ANOVA with Tukey’s ad hoc post‐test. Error bars represent SEM.

Source data are available online for this figure.

CCN4 knockout complements the anti‐tumor effect of immune checkpoint blockade therapy

As CCN4 KO increased T‐cell infiltration, we next compared the anti‐tumor effect of an αPD1 antibody on CCN4 KO and WT YUMM1.7 tumors. Administering αPD1 and isotype control (IC) antibodies started when the tumors reached approximately 100 mm³ in size for all experimental groups. As shown in Fig 7A, CCN4 KO delayed the beginning of αPD1 therapy for 11 days. Of note, a significant reduction in tumor growth was observed for both CCN4 KO and WT YUMM1.7 tumors after three doses of the αPD1 antibody when compared to mice receiving the IC antibody. Interestingly, mice with CCN4 KO tumors treated with αPD1 antibody had a tumor volume of 413.9 ± 94.99 mm³ on day 32, whereas WT YUMM1.7 TB mice receiving IC antibody reached a tumor volume of 1,540 ± 300.7 mm³ on day 21, 4 days after the last antibody dose for each group. Similar results were obtained in the B16F0 model, where administering an αCTLA4 antibody significantly delayed the growth of CCN4 KO but not WT B16F0 tumors (Fig EV5A and B).

Figure 7. CCN4 knockout further promoted the anti‐tumor effect of immune checkpoint blockade therapy.

• A
  Average tumor volumes of mice bearing WT YUMM1.7 (squares and triangles) or CCN4 KO (KO1, circles and inverted triangles) tumors (n = 4 mice/group). Groups were treated with either αPD1 (triangles and inverted triangles) or isotype control (squares and circles) antibodies when the tumors reached 100 mm³ (dotted line). Statistical significance at each time point was assessed using a Student’s t‐test with results annotated with *: 0.01 < P < 0.05, **: 0.001 < P < 0.01, and ***P < 0.001. Results summarized as mean ± SEM.
• B
  Expression of H‐2K^b (top panel) and PD‐L1 (bottom panel) were assayed by flow cytometry in WT (red curves) and CCN4 KO (KO1—blue curves) YUMM1.7 cells with (dotted curves) and without (solid curves) preconditioning with IFNγ. Unstained cells were used as a negative control (shaded curve).
• C–F
  CD45⁻ fraction isolated from WT and CCN4 KO YUMM1.7 tumors in a time‐matched experiment were assayed for H‐2K^b and PD‐L1 expression by flow cytometry. Contour curves enclose 90% (dotted curve) and 50% (solid curves) of CD45⁻ events obtained from WT (red) and CCN4 KO (blue) YUMM1.7 tumors. CD8⁺ T cells expressing PD1 within the tumor (D) and spleen (E) were assayed by flow cytometry in mice bearing WT and CCN4 KO YUMM1.7 tumors (F). Results representative of three biological replicates and summarized as mean ± SD. Statistical significance was assessed using a Student’s t‐test.

Figure EV5. Tumor growth profiles of tumors derived from wt and CCN4 KO B16F0 cells in response to αCTLA4 mAb treatment.

• A, B
  Average tumor volumes of C57BL/6 mice bearing WT B16F0 (A, squares and triangles) or CCN4 KO (B, circles and inverted triangles) tumors inoculated with s.c. injection of 3 × 10⁵ cells. Groups were treated with either αCTLA4 (triangles and inverted triangles) or isotype control (IC: squares and circles). The anti‐CTLA4 and IC antibodies were administered intraperitoneally (i.p.) at a dose of 200 μg/mouse on days 3, 7, and 10 following inoculation (n = 8 mice/group, combined from two independent experiments). ***P < 0.001. Statistical significance of the difference in tumor size at a given time point was assessed using a Student’s t‐test and error bars represent SEM.

Source data are available online for this figure.

While generally the effect of CCN4 KO was consistent between B16F0 and YUMM1.7 models subtle differences in immunosuppression mechanisms seemed to change the susceptibility of these two systems to treatment with a specific immune checkpoint. For instance, in ELISpot assays, we found that CCN4 inhibited antigen‐specific release of IFNγ by CD8⁺ T cells. Separately, we observed that IFNγ upregulated the MHC class I molecule H‐2K ^b and PD‐L1 (CD274) in both WT and CCN4 KO YUMM1.7 cells (Fig 7B). In assaying these two proteins on the CD45⁻ fraction isolated from WT and CCN4 KO YUMM1.7 tumors (Fig 7C), cells with increased expression of both H‐2K ^b and PD‐L1 increased from 12.9 ± 1.9% in WT to 24.5 ± 0.5% in CCN4 KO tumors (n = 3, P‐value = 0.0005). While tumor‐infiltrating CD8⁺ T cells predominantly expressed PD1 compared to splenocytes (Fig 7D–F), we observed no difference in PD1⁺ CD8⁺ TILs upon CCN4 KO (P‐value = 0.766, Fig 7F). It follows then that, by knocking out CCN4, the PD1‐PDL1 axis plays a more dominant role in the YUMM1.7 model, likely through IFNγ cross‐talk, in restraining the T‐cell response and is more sensitive to therapeutic intervention. We also noted that infiltrating T cells and NK cells were almost 10‐times lower per gram tumor in the WT B16F0 model compared with the WT YUMM1.7 model. Those differences coupled with B16F0 sensitivity to αCTLA4 suggest that pathways associated with initiating an anti‐tumor response, such as the extent of clonal T‐cell expansion within secondary lymphoid organs, are key limiters in the B16F0 model. Overall, these data indicate that absence of tumor‐derived CCN4 complements the anti‐tumor effect of immune checkpoint blockade therapy.

Discussion

Recent literature connects the epithelial–mesenchymal transition (EMT) with tumor immune escape (Dongre et al, ²⁰¹⁷; Terry et al, ²⁰¹⁷; Taki et al, ²⁰¹⁸). Interestingly, CCN4 activates EMT‐associated genes in melanoma cells thus increasing metastatic potential (Deng et al, ^{2019, 2020}). Motivated by encouraging clinical correlates, we report that melanoma‐derived CCN4 also stimulated tumor‐induced immunosuppression in mice, particularly by directly suppressing antigen‐induced IFNγ release by CD8⁺ T cells and by expanding and recruiting G‐MDSC. IFNγ action within the tumor microenvironment is a key promoter of anti‐tumor immunity (Alspach et al, ²⁰¹⁹). In addition, tumor‐induced MDSC suppress the proliferation and effector function of T cells and NK cells and interfere with the migration and viability of lymphocytes (Gabrilovich et al, ²⁰¹²). Correspondingly, increased anti‐tumor T cells (CD4⁺ and CD8⁺) and NK cells were observed in CCN4 KO tumors, concomitant with a reduced frequency of MDSC. Moreover, the ability of G‐MDSC to suppress T‐cell proliferation was diminished in mice bearing CCN4 KO tumors. Similar shifts in immune contexture with CCN4 expression were observed in human data. We note that Tao et al (2020) reported that CCN4 alters the myeloid compartment by maintaining tumor‐supportive macrophages in glioblastoma (Tao et al, ²⁰²⁰), although the results are difficult to interpret due to the use of a xenograft and a lack of controls for tumor growth and intrinsic differences in tumor cells due to CCN4 knockdown. While CCN4 is a downstream effector of the Wnt/β‐catenin pathway, no significant difference in tumor‐infiltrating DC was observed. This is in contrast to previous work from Gajewski et al where deficient recruitment of CD103⁺ DC in melanoma tumors with intrinsic β‐catenin signaling was associated with T‐cell exclusion (Spranger et al, ²⁰¹⁵). Of note, Ccn4 was not significantly increased in tumors with a constitutively active β‐catenin compared with Braf ^{V 600 E} /Pten^−/− control mice based on transcriptional profiling using Illumina microarrays, which suggests that transcriptional co‐activators play a role here in shaping cell‐to‐cell communication.

MDSC generation is a complex process mediated by tumor‐derived soluble factors (Condamine & Gabrilovich, 2011). Glycolysis exacerbation expands MDSCs, where a high glycolytic rate promotes MDSC generation by increasing granulocyte–macrophage colony‐stimulating factor (GM‐CSF) and granulocyte‐colony stimulating factor (G‐CSF) in triple‐negative breast cancer models (Li et al, ²⁰¹⁸). Interestingly, the glycolytic rate in CCN4 KO melanoma cells was reduced. Prior work notes that CCN4 promotes glycolysis in laryngeal squamous cell carcinoma (Wang et al, ²⁰¹⁹). CCN4 also interacts with peroxisome proliferator‐activated receptor (PPARγ) to inhibit PPARγ activity (Ferrand et al, ²⁰¹⁷). In turn, PPARγ stimulates adipocyte differentiation (Ferrand et al, ²⁰¹⁷) and inhibits glycolysis (Guo et al, ²⁰¹⁸). Thus, a model where CCN4 stimulates glycolysis by repressing PPARγ could apply to our results with melanoma cells. However, we found no differences in GM‐CSF and G‐CSF secretion in the TCM from CCN4 KO and WT YUMM1.7 cells. Conversely, and associated with impaired glycolysis, lactate secretion was reduced in the absence of CCN4 in melanoma cells. Of note, Husain et al observed a role for tumor‐derived lactate in generating MDSC (Husain et al, ²⁰¹³), suggesting that the reduced MDSC frequency in CCN4 KO TB mice is related to a diminished lactate secretion in the tumor microenvironment. Lactate released by the tumor cells also acidifies the tumor microenvironment (Huber et al, ²⁰¹⁷), which reinforces tumor‐induced immunosuppression by impairing CD8⁺ T‐cell proliferation, cytokine production, and lytic activity (Calcinotto et al, ²⁰¹²; Huber et al, ²⁰¹⁷). In addition, lymphocytes apoptose at the low pH values of the tumor microenvironment (Lugini et al, ²⁰⁰⁶). Correspondingly, we observed that half of CD45⁺ leukocytes in the tumor microenvironment of WT YUMM1.7 tumors were dead compared with around 90% viability in CCN4 KO counterparts.

CCL2 and CXCL1 secretion was also significantly reduced by CCN4 KO YUMM1.7 cells in vitro, as well as in the serum from TB mice and when analyzing the CD45⁻ cells isolated from the s.c. tumors. Of note, CCL2‐CCR2 signaling plays a key role in recruiting MDSC into the tumor microenvironment in glioma, renal and colon carcinomas, lung cancer, and melanoma (Arenberg et al, ²⁰⁰⁰; Huang et al, ²⁰⁰⁷; Hale et al, ²⁰¹⁵; Chang et al, ²⁰¹⁶; Hartwig et al, ²⁰¹⁷; Liang et al, ²⁰¹⁷). CXCL1 interaction with CXCR2 also helps recruit G‐MDSC into ovarian cancer microenvironment, where a high CXCL1 serum concentration correlates with increased tumor‐infiltrating G‐MDSC and a poor prognosis (Taki et al, ²⁰¹⁸). Interestingly, this process is mediated by Snail (SNAI1), a relevant transcription repressor regulating EMT, which stimulates CXCL1 gene expression through a direct binding to the promoter and via NF‐κB activation (Taki et al, ²⁰¹⁸). Additionally, Snail regulates CCL2 production in epithelial cells (Hsu et al, ²⁰¹⁴). Of note, CCN4 activates Snail in YUMM1.7 melanoma cells and Snail overexpression in CCN4 KO cells rescues the metastatic potential of B16F10 cells in vivo (Deng et al, ²⁰¹⁹). While additional experiments would help clarify the conditional dependence among these changes in immune contexture elicited by CCN4, our results are consistent with a model where CCN4‐induced Snail activation promotes CXCL1 and CCL2 secretion directly or through NF‐κB activation, which in turn stimulates G‐ MDSC expansion and suppressive function.

Antibodies targeting CTLA4 (ipilimumab) and PD1 (nivolumab and pembrolizumab) have proven the potential of ICB immunotherapy in patients with different malignancies, like melanoma. However, even when the response rate in melanoma is high compared to other treatments, many patients still do not receive clinical benefit (Schadendorf et al, ²⁰¹⁵; Ribas et al, ²⁰¹⁶; Jenkins et al, ²⁰¹⁸). Since ICB treatment relieves inhibitory signals controlling T‐cell function, combination therapies that increase T‐cell infiltration in the tumor can improve the anti‐tumor effect of ICB. Our results demonstrated that knocking out CCN4 in the tumor cells complemented the anti‐tumor effect of anti‐PD1 and anti‐CTLA4 antibodies in YUMM1.7 and B16F0 mouse melanoma models, respectively. Thus, targeting CCN4 can enhance ICB therapy not only by increasing T‐cell infiltration in the tumor and enhancing local IFNγ production but also through ameliorating MDSC‐mediated suppression. Others have shown that targeting MDSCs, for example by blocking CCL2‐CCR2 interactions, enhances the anti‐tumor effect of ICB therapy even in resistant tumors like glioblastoma (Flores‐Toro et al, ²⁰²⁰). Considering these elements, targeting CCN4 is immunotherapeutic strategy that can potentially impair tumor‐induced immunosuppression and enhance ICB therapy while inhibiting EMT and metastasis formation.

Materials and Methods

Mice

C57BL/6 mice (6‐ to 8‐week‐old, female) and NOD‐scid IL2Rγ^null immunodeficient mice (NSG, 6‐8 week‐old, male) were purchased from Charles River Laboratories and The Jackson Laboratory, respectively. Upon receipt, animals were labeled and randomly assigned to treatment arms/cages, with a density of five mice per cage. All animal experiments were approved by West Virginia University (WVU) Institutional Animal Care and Use Committee and performed at the WVU Animal Facility (IACUC Protocol #1604002138).

Cell culture

All biochemical reagents were obtained from commercial sources and used according to the suppliers’ recommendations unless otherwise indicated. Mouse melanoma cell lines B16F0 and YUMM1.7 were cultured in supplemented DMEM as previously described (Deng et al, ²⁰¹⁹). B16F0 cells (RRID: CVCL 0604) were obtained from the American Tissue Culture Collection (ATCC, Manassas, VA) in 2008. YUMM1.7 cells (RRID: CVCL JK16) were a gift from Drs. William E. Damsky and Marcus W. Bosenberg (Yale University) (Meeth et al, ²⁰¹⁶) and were received in 2017. CCN4‐knockout (CCN4 KO) B16F0 and YUMM1.7 cells were generated using a double nickase‐based CRISPR/Cas9 approach as previously described (Deng et al, ²⁰¹⁹). Additionally, for the B16F0 model, DNMT3A‐ and CCN4‐knockout cells were obtained through transfection with a mix of CRISPR/Cas9 KO and Homology‐Directed Repair (HDR) plasmids, followed by puromycin selection (Deng et al, ²⁰²⁰). Tet‐on inducible mouse CCN4 expression lentiviral vector (IDmCCN4) was constructed with Gateway cloning using Tet‐on destination lentiviral vector pCW57.1 (Addgene Plasmid #41393, a gift from David Root) and pShuttle Gateway PLUS ORF Clone for mouse CCN4 (GC‐Mm21303, GeneCopoeia). Lentiviruses were packaged as described (Deng et al, ²⁰¹⁹) to transduce YUMM1.7 cells with Ccn4 CRISPR knockout (Ym1.7‐KO1). Two pools of Tet‐on variant cells with inducible mCCN4 (Ym1.7‐KO1‐IDmCCN4) or vector control (Ym1.7‐KO1‐IDvector) were obtained after puromycin selection. All cell lines were revived from frozen stock, used within 10‐15 passages, and routinely tested for mycoplasma contamination by PCR.

In vivo tumor growth and ICB immunotherapy

To evaluate the effect of CCN4 KO in B16F0 and YUMM1.7 melanoma tumors, 3 × 10⁵ tumor cells were subcutaneously (s.c.) injected in C57BL/6 and NSG mice. Once palpable, the largest perpendicular diameters of the s.c. tumors were measured unblinded with a caliper twice a week, and the tumor volume was calculated using the formula: π/6 × length × width², where the width is the smaller dimension of the tumor. Using the resulting measurements, the growth rate of the tumors was estimated using a log‐linear tumor growth model and a Markov chain Monte Carlo approach to generate the posterior distribution in the rate parameters, as described previously (Deng et al, ²⁰²⁰). C57BL/6 mice were also injected subcutaneously with Tet‐on variants constructed using CCN4 knockout YUMM1.7 cells (KO1) using a 2×2 factor experimental design, with doxycycline (Sigma‐Aldrich) treatment and the Tet‐on variant cells as the two factors. Doxycycline was delivered by injecting 0.15 ml (10 mg/ml) intraperitoneally at day 0 and orally via consumption (ad lib) of standard mouse chow containing 200 mg dox per 1 kg food (Bio‐Serv). ICB therapy was studied using the InVivoMAbs anti‐mouse PD1 (CD279, clone J43) monoclonal antibody with YUMM1.7 cell variants, anti‐mouse CTLA‐4 (clone UC10‐4F10‐11) monoclonal antibody with B16F0 cell variants, and a polyclonal Armenian hamster IgG as isotype control (IC) (BioXCell, NH) at doses of 200 µg/mouse. C57BL/6 mice were s.c. inoculated with 5 × 10⁵ CCN4 KO and wild type (WT) YUMM1.7 tumor cells. The anti‐PD1 and IC antibodies were administered intraperitoneally (i.p.) on days 0, 4, and 9, considering as day 0 the day when the tumors reached a volume of 100 mm³. All in vivo studies were repeated at least twice with two independent cohorts and with n ≥ 3 in each experimental group. The cohort size was not pre‐specified as this was an exploratory study and not designed to test for a pre‐specified effect.

In vitro suppression of CD8⁺ T‐cell function

To generate YUMM1.7‐reactive CD8⁺ T cells, healthy C57BL/6 mice were inoculated subcutaneously with irradiated YUMM1.7 cells (10⁵/mouse), followed by live YUMM1.7 cells (3 10⁵/mouse) 3 weeks later. The mice without tumors in the following five weeks were maintained. Three days before the assay, the mice were injected again with live YUMM1.7 cells (10⁵/mouse). On the day of assay, the YUMM1.7‐reactive cells were isolated from mouse splenocytes using mouse CD8a⁺ T Cell Isolation Kit (Miltenyi Biotec, Germany), resuspended at 10⁶ cells/ml. A 50 µl (5 × 10⁴) of the YUMM1.7‐reactive CD8⁺ T cells were aliquoted into 96‐well plates for ELISpot assay using Mouse IFNγ/TNFα Double‐Color ELISpot kit (Cellular Technology Limited) following manufacturer’s instructions. Briefly, target tumor cells were stimulated with IFNγ (200 U/ml, or, 20 ng/ml) for 24 h, harvested and resuspended at 2 × 10⁶ cells/ml. A 50 µl (10⁵) of indicated tumor cells were aliquoted in triplicate, with or without doxycycline (Dox, final 0.5 µg/ml). The reactions were incubated at 37°C for 24 h and colored spots were developed, imaged using an Olympus MVX10 Microscope, and counted.

Flow cytometry

Tumors were surgically removed after euthanasia, weighed, and processed into single‐cell suspensions using the Tumor Dissociation Kit, mouse (Miltenyi Biotec) and the manufacturer’s instructions. Single‐cell suspensions from tumors and spleens were stained with specific antibodies or IC using conventional protocols. Live and dead cells were discriminated with Live/Dead Fix‐able Violet Dead Cell Stain Kit (Thermo Fisher Scientific, MA). Fc receptors were blocked with purified rat anti‐mouse CD16/CD32 (Mouse BD Fc Block, BD Biosciences, CA). Anti‐mouse antibodies were used to characterize the lymphoid populations: CD45/BB515 (clone 30‐F11, BD Biosciences), CD3t:/Alexa Fluor 700 (clone 500A2, BioLegend, CA), CD3t:/PE (clone 17A2, Miltenyi Biotec), CD8a/APC (clone 53‐6.7, Miltenyi Biotec), CD4/APC‐Cy7 (clone GK1.5, BD Bio‐ sciences), CD45R/B220/APC (clone RA3‐6B2, BioLegend), NK‐1.1/APC‐Cy7 (clone PK136, BioLegend), CD49b/PerCP‐Cy5.5 (clone DX5, BioLegend), CD25/PerCP‐Cy5.5 (clone PC61.5, eBio‐ science), CD279 (PD‐1)/PE (clone REA802, BioLegend), and FOXP3/PE (clone FJK‐16s, eBioscience). The following antibodies were used to detect the myeloid populations: CD45/BB515 (clone 30‐F11, BD Biosciences), CD11b/PerCP‐Cy5.5 (clone M1/70, Thermo Fisher Scientific), Ly‐6G/Ly‐6C (Gr‐1)/APC (clone RB6‐8C5, BioLegend), CD11c/PE (clone N418, Thermo Fisher Scientific), F4/80/APC (clone BM8, BioLegend), I‐A/I‐E/Alexa Fluor 700 (clone M5/114.15.2, Bi oLegend), Ly‐6G/APC (clone 1A8, BD Biosciences), and Ly‐6C/PE (clone AL‐21, BD Biosciences). Total number of cells in tumors and spleens were determined using SPHERO AccuCount Fluores cent Particles (Spherotech, IL). Expression of H‐2K ^b and PD‐L1 were assayed in the CD45⁻ fraction generated from WT and CCN4 KO YUMM1.7 tumors using the antibodies: CD274 (PD‐L1)/PE (clone 10F.9G2, BioLegend) and H‐2K ^b /APC (clone AF6‐88.5, BioLegend). For comparison, H‐ 2K ^b and PD‐L1 expression was assayed in WT and CCN4 KO YUMM1.7 tumor cells conditioned in the presence or absence of IFNγ (200 U/ml, or, 20 ng/ml) for 24 h, with unstained cells as controls. Events were acquired using a BD LSRFortessa (BD Biosciences) flow cytometer with FACSDiva software, where the fluorescence intensity for each parameter was reported as a pulse area with 18‐bit resolution. Flow cytometric data were exported as FCS3.0 files and analyzed with FCS Express 6.0 (DeNovo Software, CA) and FlowJo 5.7.2 (Tree Star Inc., OR). The typical gating strategies for lymphoid and myeloid cells are shown in Appendix Figs S4A and B and S5, respectively.

T‐cell proliferation and MDSC‐mediated suppression assay

Splenocytes from mice bearing CCN4 KO and WT YUMM1.7 tumors, as well as from tumor‐free mice, were used as effector cells. CD45⁺ cells isolated from tumor‐bearing (TB) mice using the CD45 MicroBeads, mouse (Miltenyi Biotec) were also used as an effector population. All effector cells (1 × 10⁷/ml) were stained with CellTrace Violet Cell Proliferation Kit (Thermo Fisher Scientific) following the manufacturer protocol. Once stained, 5 × 10⁵ effector cells/well were stimulated for 72 h with the T Cell Activation/Expansion Kit, mouse (Miltenyi Biotec) at a 1:1 ratio with anti‐CD3/anti‐CD28‐loaded beads. To evaluate the suppressive function, granulocytic MDSC (G‐MDSC) were isolated using the Myeloid‐Derived Suppressor Cell Isolation Kit, mouse (Miltenyi Biotec) from the spleens of CCN4 KO and WT YUMM1.7 TB mice and tumor‐free mice. G‐MDSC (20% of effector cells) were then co‐incubated for 72 h with 5 × 10⁵ stained naïve splenocytes in the presence of anti‐CD3/anti‐CD28‐loaded beads at 1:1 ratio with effector cells. Proliferation diluted the CellTrace Violet dye, as assayed by flow cytometry. Live CD8⁺ effector cells were identified with Live/Dead Fixable Green Dead Cell Stain Kit (Thermo Fisher Scientific) and anti‐mouse CD8a/APC (clone 53‐6.7, Miltenyi Biotec). Proliferation metrics were quantified using an approach described by Roederer (2011).

Tumor‐conditioned media collection, cytokine array, and ELISA

To collect tumor‐conditioned media (TCM), cells were grown in complete DMEM until 80% confluency, washed with PBS (Cellgro/Corning, NY) and incubated for 48 h in FBS‐free DMEM. TCM were then centrifuged at 3,000 g and 4°C for 15 min with the supernatant collected and filtered. The cytokines, chemokines, and growth factors in TCM were detected with the Proteome Profiler Mouse XL Cytokine Array (R&D Systems, MN), following the manufacturer’s instructions. CCN4 was assayed in TCM using the mouse WISP‐1/CCN4 DuoSet ELISA Kit (R&D Systems). CCL2 and CXCL1 were also quantified with Mouse CCL2/JE/MCP‐1 and Mouse CXCL1/KC DuoSet ELISA kits (R&D Systems), respectively. These chemokines were measured in TCM, obtained from the cell lines in vitro as described above, and from CD45⁻ cells isolated from WT and CCN4 KO tumors by negative selection using a mouse CD45 MicroBeads kit (Miltenyi Biotec). Conditioned media were obtained by culturing 1 × 10⁴ CD45⁻ cells/well for 36 h in DMEM supplemented with 10% FBS. Blood was collected from the submandibular vein of CCN4 KO and WT YUMM1.7 TB mice and the serum was obtained after letting the blood to clot for 1 h at room temperature and performing a 10 min centrifugation at 2,000 g and 4°C and assayed for CCL2 and CXCL1.

Metabolic function assays

WT and CCN4 KO melanoma cells were cultured overnight (1 × 10⁴ cells/well) in Seahorse XFe96 cell culture microplates (Agilent, CA) with complete DMEM. The extracellular acidification rate (ECAR) was measured using a Seahorse XFe96 Analyzer (Agilent) according to the manufacturer’s instructions, which allowed calculating glycolysis and glycolytic capacity. To compare the lactate secretion, live CD45⁻ cells were isolated by negative selection from digested WT and CCN4 KO tumors using the mouse CD45 (TIL) MicroBeads kit (Miltenyi Biotec). A total of 1 × 10⁴ cells/well were cultured for 36 h in DMEM supplemented with 10% dialyzed FBS (Gibco, Thermo Fisher Scientific). Lactate was measured in the conditioned media using the Lactate‐Glo Assay (Promega, WI).

Statistical analysis

Gene expression and clinical profiles for patients diagnosed with stage I to III melanoma (SKCM) from TCGA were downloaded using the “TCGAbiolinks” (V2.8.2) package in R (V3.5.1). Single‐cell RNA sequencing data obtained from tumor samples of patients diagnosed with melanoma that were naive and resistant to immune checkpoint therapy were used with Gene Expression Omnibus accession numbers GSE72056 (Data ref: Tirosh & Izar, 2016; Tirosh et al, ²⁰¹⁶) and GSE115978 (Data ref: Jerby‐Arnon et al, ^2018a; Jerby‐Arnon et al, ^2018b), where non‐zero counts in CCN4 expression was used to designate a CCN4‐positive malignant cell. Statistical enrichment of CCN4‐positive malignant cells was assessed by a binomial test where the observed frequency was compared against a null hypothesis represented by a binomial distribution with a baseline frequency of 1% CCN4‐positive cells. A P‐value represents the probability of the observed or greater frequency being drawn from null distribution. The immune contexture was estimated from the SKCM data obtained from primary melanoma tissue samples using CIBERSORTx and the LM22 immune cell gene signatures (Newman et al, ²⁰¹⁹). Statistical differences in the posterior distributions in tumor growth rate parameters were assessed using a Pearson’s chi‐squared test. Kaplan–Meier analysis, Cox proportional hazards modeling, and Mann–Whitney U‐tests were performed using the “survival” (V2.42‐6), “survminer” (V0.4.2), and “stats” (V3.5.1) packages in R. Unless otherwise specified, quantitative results were summarized as mean ± standard error of measurement (SEM) and overlaid on individual results. Unpaired Student’s t‐test (two‐tailed) or one‐way ANOVA followed by Tukey’s multiple comparison ad hoc post‐test were performed with GraphPad Prism (version 5). A P‐value of < 0.05 was considered statistically significant and denoted as follows: *0.01 < P < 0.05, **0.001 < P < 0.01, and ***P < 0.001.

Author contributions

Audry Fernandez: Conceptualization; Data curation; Formal analysis; Methodology; Writing—original draft; Writing—review & editing. Wentao Deng: Data curation; Formal analysis; Writing—review & editing. Sarah L Mclaughlin: Data curation; Writing—review & editing. Anika C Pirkey: Formal analysis; Writing—original draft; Writing—review & editing. Stephanie L Rellick: Data curation; Formal analysis; Writing—review & editing. Atefeh Razazan: Data curation. David J Klinke: Conceptualization; Software; Formal analysis; Supervision; Funding acquisition; Methodology; Writing—original draft; Project administration; Writing—review & editing.

In addition to the CRediT author contributions listed above, the contributions in detail are:

Conception and design: AF and DJK; Data Acquisition: AF, WD, SLM, SR, and AR; Analysis and interpretation of data: AF, WD, AP, SR, and DJK; Funding acquisition: DJK; Methodology: AF and DJK; Project administration: DJK; Software: DJK; Supervision: DJK; Writing—original draft: AF, AP, and DJK; Writing—review and editing: all authors.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Dataset EV1

Source Data for Expanded View

EMBO reports (2022) 23: e54127.

Data availability

This study included no data deposited in external repositories.