D2.1 dormancy is immune independent. The D2 series of cells, consisting of D2A1, D2.OR, and D2.1 cell lines that arose from D2 hyperplastic alveolar nodules in female BALB/c mice, were used to investigate immunity in tumor dormancy (13, 14). Each cell line is reported to equally extravasate into the lung parenchyma, but only D2A1 rapidly proliferates, while D2.OR and D2.1 remain dormant in the metastatic setting (13, 15, 16). Tumor behavior in the presence or absence of adaptive immunity was first determined by implanting these cells in the mammary fat pad (MFP) of BALB/c (immunocompetent) or SCID-beige (deficient in T and B cells and defective NK cells) mice. Both D2A1 and D2.OR tumor growth was significantly delayed in the presence of adaptive immunity (Figure 1, A and B). In contrast, D2.1 tumors were unaffected by T, B, and NK cells (Figure 1C). However, when cultured in vitro, D2A1 and D2.1 cells proliferated at approximately the same rate until confluence, at which point D2.1 cells became contact inhibited, resulting in slowed growth (Supplemental Figure 1, A and B; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.174458DS1). To validate the dormant phenotype, D2.1 cells were stably transduced with eGFP by lentiviral vectors and proliferation was assessed in resultant tumors by Ki67 immunofluorescence (IF) at 4 or 12 weeks after MFP implantation in BALB/c animals. Staining revealed that eGFP+ D2.1 tumor cells were minimally positive for Ki67 at 4 weeks, but had significantly elevated Ki67 expression by 12 weeks (Figure 1D). Co-injection of a 1:1 mix of D2A1 and D2.1 cells also resulted in approximately equivalent tumor growth as D2A1 cells alone, suggesting dormant cells do not induce quiescence in otherwise proliferative cells (Supplemental Figure 1C). Collectively, these results demonstrate that D2.1 tumor cells exist as a separate, intrinsically dormant population, unaffected by immune status, that have the eventual capacity to develop into proliferative tumors.

The adaptive immune system does not impact tumor dormancy or long-term persistence in mammary D2.1 tumors. (A) Tumor growth in the mammary fat pad (MFP) of 1 × 106 parental D2A1 cells in BALB/c or SCID-beige mice (n = 5/group). (B) D2.OR tumor growth of 1 × 106 parental cells in BALB/c or SCID-beige mice (n = 5/group). Comparisons shown are at time of first euthanasia and were performed by Šídák’s 2-way ANOVA (A and B). (C) Growth of parental D2.1 cells (1 × 106) implanted into the MFP of BALB/c (n = 5) or SCID-beige (n = 8) mice. (D) Representative images (left) or quantification (right) of Ki67+eGFP+ tumor cells after MFP implantation of 1 × 106 D2.1-eGFP cells in BALB/c mice and collection after 4 or 12 weeks. Tumor sections were stained for nuclei (DAPI, pseudocolor gray), eGFP (green), or Ki67 (purple). Scale bars: 100 μm. Statistical comparison was by 2-tailed t test. (E) Flow cytometric analysis of MHC-I surface expression on cultured D2A1, D2.OR, and D2.1 cells. (F) D2A1-eGFP, D2.OR-eGFP, or D2.1-eGFP single-cell clones (n = 5/group) were implanted (1 × 106 cells/mouse) into the MFP of female BALB/c mice. Tumors were resected at the indicated time point, digested, and cultured ex vivo for a pure tumor cell population. (G) Left: Representative flow plots for eGFP expression of ex vivo D2A1, D2.OR, or D2.1 tumors compared to parental (no eGFP) cells or respective single-cell clones on the day of injection. Right: Quantification of the percentage of cultured tumor cells that maintained eGFP expression after in vivo selection. Statistical comparisons were performed by 1-way ANOVA with Tukey’s post hoc correction. Data are presented as mean ± SEM.

Dormant tumors are immunologically protected from infiltrating T cells. Because D2.1 tumor persistence through the dormant phase was unaffected by adaptive immunity, variation in cell surface expression of major histocompatibility complex class I (MHC-I), programmed death-ligand 1 (PD-L1), and CD47 were investigated as potential mechanisms for dormancy-mediated immune evasion (17, 18). Notably, MHC-I (Figure 1E and Supplemental Figure 1D), PD-L1, and CD47 (Supplemental Figure 1, E and F) are expressed similarly on D2A1, D2.OR, and D2.1 cells, suggesting that these are unlikely to mediate differences in immune resistance. As all D2 lines express comparable cell surface MHC-I, loss of eGFP expression after in vivo implantation was used as a readout for antigen-specific immune selection (Figure 1F) (19). While many D2A1 tumor cells resisted eGFP-specific selection, D2.OR cells displayed the highest sensitivity, with almost complete loss of eGFP (Figure 1G). In contrast, the majority of D2.1 cells maintained eGFP after implantation in immunocompetent, non–eGFP tolerant BALB/c mice. These results are consistent with the recent finding that GFP-specific T cells (just eGFP death inducing, Jedi) are unable to eliminate quiescent subpopulations of mammary tumors despite MHC-I expression (20), and suggest that dormant tumors can be effectively shielded from antigen-specific immunity and selection by other means.

D2.1 dormant tumor cells have a hybrid E/M phenotype and are enriched in mammary progenitor genes associated with late recurrence in humans. Due to the striking phenotypic differences in vivo between dormancy-competent D2.1 cells and proliferative counterparts, we initially performed RNA-seq on D2A1 and D2.1 cell lines. The top significantly enriched pathways in D2.1 cells were largely associated with cellular movement and extracellular matrix (ECM) organization, with a significant downregulation of metabolic pathways involved in DNA and RNA processing (Figure 2A and Supplemental Figure 2A). Notably, D2.1 cells expressed higher transcript levels of multiple mammary epithelial maker genes (e.g., Krt8, Krt14, Krt18, Itgb4, Epcam, and Cdh1) and genes formerly associated with mammary progenitor cells (e.g., Sox9, CD74, Id4, Nrg1, Ptn, and Cd200) (Figure 2B). Genes that regulate epithelial-mesenchymal transition (EMT) and metastasis were also upregulated, including genes associated with ECM interaction (Itga5, Sparc, Mmp9, Cd44, and Fn1) as well as many transcriptional regulators of EMT (Twist2, Snai1, Snai2, and Notch1) (Figure 2B). Subsequent quantitative PCR confirmed the hybrid E/M phenotype of D2.1 cells that expressed high levels of epithelial and mesenchymal genes compared with both D2A1 and D2.OR cells (Supplemental Figure 2B). The ERα gene Esr1 and the ERα target gene Greb1 were also expressed in D2.1 cells (Supplemental Figure 2C) — an interesting corollary, as human ER+ tumors generally display the highest degree of dormancy (5, 21).

D2.1 tumor cells are enriched in mammary progenitor genes and maintain a hybrid E/M profile in vivo associated with immune dysfunction. (A) RNA-seq was performed on D2A1 and D2.1 cells cultured in vitro. Shown are the top 10 upregulated (black fill) and downregulated (white fill) Gene Ontology (GO) Biological Process (BP) pathways in D2.1 compared to D2A1. (B) Heatmap of selected significantly differentially expressed genes related to markers of mammary epithelium, antigen presentation via MHC-I, a mesenchymal phenotype, and mammary epithelial cell differentiation is shown. (C) D2A1 and D2.1 tumors were implanted into the mammary fad pad (1 × 106/mouse) and collected at approximately 500 mm3 for RNA-seq (n = 4/group). Shown are selected differentially expressed genes as in B. (D) GSEA of upregulated gene signatures in D2.1 tumors associated with mammary stem cells, luminal progenitor cells, and luminal mature cells compared to D2A1. (E and F) GSEA of upregulated gene sets in D2.1 tumors associated with late recurrence (E) or late distant metastasis (F) in human BC. (G) GO BP pathway analysis of D2.1 vs. D2A1 tumors in BALB/c mice. Eight of the top 10 differentially expressed pathways were directly related to immune function. (H) CIBERSORT analysis of D2.1 and D2A1 BALB/c tumors. Statistical comparisons were performed by 2-tailed t test and data are presented as mean ± SEM. (I) GSEA of T cell anergy in D2.1 vs. D2A1 BALB/c tumors.

In accordance with an overall mammary progenitor–like expression profile, D2.1 cells were found to be CD44hiCD24lo/– compared with D2.OR and D2A1 cells by flow cytometry, with D2.OR cells having the most CD24 expression (Supplemental Figure 2D) (22, 23). The presence of increased epithelial and mesenchymal genes could also reflect the described “hybrid epithelial/mesenchymal (E/M)” state associated with surface CD44 and CD104 expression (24, 25). Interestingly, surface expression of CD44 and CD104 revealed largely nonoverlapping populations between D2A1, D2.OR, and D2.1 cells (Supplemental Figure 2E). Because D2.OR tumors are immune sensitive (Figure 1G), overt tumors were often rejected in the MFP of BALB/c animals. However, small, residual nodules that reemerged after approximately 100 days phenotypically resembled D2.1 cells based on CD44/CD104 expression (Supplemental Figure 2F). Altogether, these data indicate that D2.1 cells have a hybrid E/M phenotype at baseline that has been associated with metastasis, dormancy, and immunosuppression (24–27), and may represent a particular subpopulation that can reemerge after immune rejection of the tumor at large (as from D2.OR tumors).

D2.1 tumor phenotype is maintained throughout progression. Bulk RNA-seq was also performed on MFP D2A1 and D2.1 tumors from both BALB/c and SCID-beige mice to evaluate phenotypic differences and changes from the selective pressure of adaptive immunity (Supplemental Figure 3A). Overall, D2A1 and D2.1 tumors were largely similar whether in BALB/c or SCID-beige mice (Supplemental Figure 3, B and C). The hybrid E/M and mammary progenitor expression pattern was maintained long-term in D2.1 tumors (Figure 2, C and D) (28). Furthermore, D2.1 tumors were enriched for a genetic profile of late recurrence that we previously developed from human BC samples (29), as well as an independent profile of late distant metastasis also generated from human data (30) (Figure 2, E and F, and Supplemental Table 1). Pathway analysis also revealed that even later-stage D2.1 tumors displayed a less proliferative phenotype when compared with D2A1 tumors in both BALB/c and SCID-beige animals (Supplemental Figure 3, D and E) with upregulated Cdkn1b (p27, a common indicator of tumor dormancy; refs. 20, 31) and reduced expression of genes associated with proliferation (Supplemental Figure 3F). Somewhat surprisingly, Gene Ontology (GO) analysis revealed that 8 of the top 10 upregulated pathways in D2.1 tumors specifically related to immune activation (Figure 2G), and while general markers of T cells (Cd3d, Cd3e, and Cd3g) and cytotoxic T cells (Cd8a and Cd8b1) were associated with tumors from all BALB/c animals, Treg genes (Ctla4 and Foxp3) were only associated with D2.1 tumors in immunocompetent animals (Supplemental Figure 3G). Subsequent CIBERSORT analysis indicated increased CD4+ T follicular helper cells, CD8+ cells, Tregs, and naive B cells in D2.1 compared with D2A1 tumors (Figure 2H), despite the possibility that these immune cells were less functional (Figure 2I).

As the Myc oncogene was significantly upregulated in D2A1 tumors (Figure 2, B and C, and Supplemental Figure 4A), we speculated that the dormant phenotype of D2.1 cells could be reversed by Myc-stimulated proliferation. To test this, we expressed a stabilized form of Myc (T58A) (32) in D2.1 cells (Supplemental Figure 4B). While Myc-T58A expression enhanced proliferation in vitro and in SCID-beige mice, it did not translate to enhanced tumor burden in immunocompetent BALB/c mice (Supplemental Figure 4, C and D). This suggests proproliferative signaling in tumors is insufficient to counteract strong immune editing of proliferative cells, and that a period of dormancy followed by growth is potentially a necessary stage for immune escape in these populations.

Dormant mammary tumors have high levels of infiltrating FoxP3+ Treg cells. In light of the immune-independent nature of D2.1 latency, the presence of immune-related genes in D2.1 tumors was striking. Therefore, D2.1 cells were orthotopically implanted and collected after 35 days for immune analysis, and D2A1 tumors were either collected early at the same final size as D2.1 tumors (14 days after implantation) or after the same total duration (35 days after implantation) to ensure that potential differences were not merely due to tumor burden (Figure 3, A and B). Interestingly, flow cytometry revealed that D2.1 tumors indeed contained significantly more total T cells than D2A1 tumors of the same volume or time in vivo (Figure 3C). While large D2A1 tumors had more infiltrating CD8+ T cells in comparison with both small D2A1 tumors and D2.1 tumors (Figure 3D), D2.1 tumors harbored more total CD4+ cells than D2A1 cells at either stage (Figure 3E). Critically, the Foxp3+CD4+ Treg population was significantly elevated in D2.1 tumors (Figure 3F), ultimately resulting in a decreased CD8+ cell/Treg ratio compared with all D2A1 tumors (Figure 3G).

Dormant tumors are highly infiltrated by T cells but induce a Treg-rich microenvironment. (A) D2A1 or D2.1 cells were implanted into the mammary fat pad of BALB/c mice (1 × 106/mouse). Small, volume-matched D2A1 tumors (“Early D2A1”; n = 6), large, duration-matched D2A1 (“D2A1”; n = 4) tumors, or small D2.1 (n = 4) tumors were collected as indicated. (B) Final volume of tumors in A. (C) Flow cytometric analysis of total T cells (CD45+CD11b–CD4+ or CD45+CD11b–CD8β+) among total immune cells (CD45+) in tumors at time of euthanasia. (D) Percentage of CD8+ cells among total CD45+ cells. (E) Percentage of CD4+ cells among total CD45+ cells. (F) Left: Flow plots of FoxP3+CD4+ cells in D2A1 and D2.1 tumors obtained on day 35 (shown gated on CD45+ cells). Right: Quantification of Tregs in tumors from A. (G) The CD8+ cell to Treg (FoxP3+CD4+) ratio in total T cells by flow cytometry. One-way ANOVA with Tukey’s post hoc analysis was used to compare 3 groups for B–G. (H) Representative IHC image of tumor border from endpoint D2A1 tumors. Dashed lines indicate stromal/tumor interface and arrows indicate positively stained CD8+ (top) or CD4+ (bottom) cells. (I) Representative images of sized-matched D2A1 or D2.1 tumor interiors stained for CD8 (top) or CD4 (bottom). (J and K) Quantification of CD8+ (J) or CD4+ (K) cells/mm2 in D2A1 and D2.1 tumor interiors (n = 5 each). (L) Representative images (left) and quantification (right) of FoxP3+ cells in tumors from H–K. Scale bars: 50 μm (H, I, and L). Density analysis was performed on 5 random fields/tumor using ImageJ software and comparisons were performed via 2-tailed t test (J–L). (M and N) Correlation of endpoint volume and CD8+ density (M) or FoxP3+ density (N) in D2.1 tumors of different sizes (n = 17). Pearson’s correlation is displayed in each plot. Data are presented as mean ± SEM.

To ascertain the spatial relationship of T cells with tumor cells and determine whether differences in T cell profiles were maintained through tumor progression, D2.1 tumors were also collected at a later stage (100 days after implantation) (Supplemental Figure 5B). Staining for CD4+ and CD8+ T cells revealed that both localized at the tumor/stromal interface in size-matched D2A1 tumors (Figure 3H; a feature observed in approximately 26% of TNBCs, ref. 33). However, endpoint D2.1 tumors were highly infiltrated with both CD4+ and CD8+ T cells (Figure 3I), resulting in a significantly increased interior T cell density in comparison with similarly sized D2A1 tumors (Figure 3, J and K). Furthermore, D2.1 tumors harbored a significant population of infiltrating FoxP3+ cells, which likely represents the same FoxP3+CD4+ Treg population observed via flow cytometry based on multicolor IHC (Figure 3L and Supplemental Figure 5C). No correlation in CD8+ cell density was observed collectively across D2.1 tumors of multiple time points and sizes (Figure 3M); however, Foxp3+ cell infiltration was significantly correlated with early, smaller tumors (Figure 3N). Thus, dormant tumors do not necessarily restrict T cell infiltration even after beginning to proliferate, indicating that changes in T cell function may be more relevant than proximity. Additionally, these data suggest that Treg induction is an early and necessary event to establish the dormant tumor niche for long-term survival.

Tumor antigen–specific CD8+ cells do not prevent dormant tumor progression in vivo. Primary mammary tumors were recently reported to induce CD8+ T cells (CD39+PD-1+) that systemically control metastatic dormancy in the lungs in the 4T07 model (10). Contrastingly, analysis of MFP D2A1-eGFP and D2.1-eGFP tumors (Supplemental Figure 5E) revealed that D2A1 tumors contained significantly more CD39+PD-1+CD8+ T cells than D2.1 tumors, with a similar trend seen systemically in the spleens of tumor-bearing animals (Figure 4A). Absent a strong, de novo tumor-specific T cell response, we next investigated whether D2.1 persistence could be overcome with potent and systemic antigen-specific T cells. To exclude the role of the tumor microenvironment, killing assays were initially performed in vitro to assess the cytotoxic effect of antigen-specific CD8+ T cells against D2.1 and D2A1 cells. Jedi T cells, which express a H-2Kd–restricted T cell receptor (TCR) for GFP200–208 (Figure 4B) (34), were activated/expanded ex vivo and cultured with D2A1 or D2.1 eGFP cells. Both D2A1 and D2.1 cells were equivalently lysed by Jedi cells, thereby verifying that both are susceptible to direct killing by activated Jedi CD8+ T cells (Figure 4, C and D). Adoptive transfers were then performed to determine whether externally activated antigen-specific CD8+ T cells could prevent/limit eventual D2.1 tumor outgrowth. Activated Jedi cells were injected intravenously into naive C57BL/6 × BALB/c F1 mice, and 2 days later D2A1 or D2.1 cells expressing eGFP were implanted into the MFP (Figure 4E). While D2A1 tumor outgrowth was completely prevented by Jedi T cells compared with control T cells (Figure 4F), D2.1 tumors were unaffected by Jedi T cells in comparison to controls (Figure 4G). These results indicate that even highly activated antigen-specific CD8+ T cells are incapable of preventing establishment and eventual growth of dormant D2.1 tumors in vivo.

Dormant D2.1 tumors evade antigen-specific killing by activated, tumor-specific CD8+ cells. (A) Representative flow plots (left) of PD-1 and CD39 expression on CD8+ cells from endpoint D2A1 or D2.1 tumors and quantification (right) of PD-1+CD39+ cells among total CD8+ T cells in tumors or spleens of tumor-bearing BALB/c animals. Statistical comparisons were performed by 2-tailed t test. (B) Jedi T cells express a TCR specific for eGFP200–208 that enables antigen-specific targeting of eGFP+ tumor cells. (C) In vitro killing of D2A1 cells plated with activated Jedi T cells at the indicated ratios. (D) In vitro killing of D2.1 cells plated with activated Jedi T cells. Values were normalized to a 0:1 E:T ratio for each condition and were compared by Šídák’s 2-way ANOVA. (E) Ex vivo–expanded Jedi T cells or naive splenocytes were transferred to naive mice and 1 × 106 D2A1 or D2.1 eGFP-expressing cells were implanted into the mammary fat pad after 2 days (n = 5/group). (F and G) Growth plots of D2A1 (F) or D2.1 tumors (G). P values calculated by Šídák’s 2-way ANOVA at end of experiment. Data are presented as mean ± SEM.

D2.1 cells secrete factors that induce CD4+ and FoxP3+ Tregs. Increasing evidence suggests that E/M plasticity, as observed in D2.1 cells, is associated with an overall immunosuppressive microenvironment (27). However, because CD8+ T cells were distinctly present in but unable to eliminate latent tumors, we investigated whether tumor cells themselves altered T cell function directly. The effect of the tumor secretome on CD8+ T cell activation/proliferation was initially to be tested by culturing splenocytes from transgenic Jedi mice with tumor-derived conditioned medium (CM) (35). Whole Jedi splenocytes were stained with CellTrace dye and stimulated with GFP200–208 peptides and anti-CD28 antibodies in the presence of D2A1 or D2.1 CM for 3 days (Figure 5A). Overall, D2.1 CM resulted in significantly more T cells than D2A1 CM (Figure 5B), including both CD8+ (which was stimulated via peptide) and CD4+ cells (Figure 5C). CellTrace staining indicated that CD8+ T cells indeed underwent more divisions in D2.1 CM compared with D2A1 CM (Figure 5D); however, CD4+ T cells displayed more significantly enhanced proliferation in D2.1 CM in comparison with D2A1 CM (Figure 5E). Further experiments revealed a significant increase in Tregs within the CD4+ population (Figure 5F), ultimately resulting in a decreased CD8+ cell/Treg ratio in D2.1 CM compared with D2A1 (Figure 5G). Thus, these data suggest that the disparity between D2A1- and D2.1-induced T cell responses in vivo are likely more attributable to dormant tumor cell promotion of CD4+ Treg proliferation and/or differentiation via soluble factors.

D2.1 cells directly induce CD4+ Tregs that can target in vivo. (A) Whole Jedi splenocytes were stained with CellTrace and stimulated with anti-CD28 antibodies and GFP200–208 peptides in conditioned medium (CM) from D2A1 cells or D2.1 cells for 3 days followed by flow cytometric analysis. (B) Total T cells (CD45+CD11b–CD4+ or CD45+CD11b–CD8β+) among immune cells (CD45+) at the end of culture. (C) Representative flow plots of CD4+ and CD8+ T cells among CD45+ cells and proportion of total CD4+ and CD8+ cells among CD45+ cells. (D and E) Representative flow plots (left) and quantification of divisions (right) of CellTrace-stained CD8+ cells (D) or CD4+ cells (E) in D2A1 or D2.1 CM. (F) Representative plots (left) and quantification (right) of FoxP3+ Tregs in D2A1 and D2.1 CM after 3 days in culture. (G) CD8+/Treg ratio in T cells in D2A1 or D2.1 CM. All conditions were performed in triplicate. Statistical comparisons were performed by 2-tailed t test (B, C, F, and G) or Šídák’s 2-way ANOVA (D and E). (H) Growth of D2.1 tumors (1 × 106 cells/mammary fat pad) treated with anti-CTLA4 antibodies (n = 10) or isotype control (n = 9) upon reaching 150 mm3. (I) D2.1 (1 × 106 cells/mammary fat pad) were implanted into female FoxP3-DTR mice and received 0 ng/g DT (Untreated) or 5 doses of 25 ng/g every 4 days (arrows) beginning on day 35 (DT Early) or day 70 (DT late). (J) D2.1 tumor volume comparing all Untreated (n = 14) mice to DT Early (n = 9) at time of DT Early euthanasia (left) or comparing remaining Untreated mice to DT Late (n = 8 each) at end of experiment (right). Statistical comparisons for H and J were performed by Šidák’s 2-way ANOVA and P values shown are at the final time point. Data are presented as mean ± SEM.

Tregs protect D2.1 tumors. Because CD8+ T cell functionality was ostensibly restricted in D2.1 tumors with an associated increase in Tregs, we estimated that Tregs were a more central determinant of immunosuppression in the dormant tumors. Given the ability of anti-CTLA4 IgG2A/B antibodies to deplete intratumoral Tregs (36, 37), we first treated mice bearing D2.1 tumors biweekly with anti-CTLA4 (clone 9D9, 200 μg/mouse) or isotype controls upon the tumors reaching approximately 150 mm3. This treatment significantly reduced tumor growth (Figure 5H and Supplemental Figure 6A), with a modest reduction in intratumoral Tregs (although total T cell levels were unchanged) (Supplemental Figure 6B). While these results suggested the importance of Tregs in dormant tumors, anti-CTLA4 antibodies can alter costimulation; thus, we wanted to validate these findings in a more specific Treg model. We therefore implanted D2.1 cells into FoxP3-DTR mice, in which Tregs can be ablated through diphtheria toxin (DT) administration (Supplemental Figure 6C). Tumor-bearing animals were treated with DT every 4 days beginning at an early time point (day 35) or late time point (day 70), with a total of 5 doses each (Figure 5I). Notably, both early and late DT treatment resulted in significant reduction in tumor growth (Figure 5J and Supplemental Figure 6, D and E). Late-DT-treated and control tumors were analyzed upon euthanasia, which showed an increase in infiltrated T cells (Supplemental Figure 6F), including an increase in activated CD39+PD-1+CD8+ T cells. Altogether, these data indicate that despite high CD8+ T cell infiltration, the concomitant presence of Tregs in D2.1 tumors restricts antitumor immunity and enables tumor persistence.

DKK3 is crucial for D2.1 tumor persistence. Knowing that tumor-derived secreted factors were sufficient to regulate the T cell landscape we further investigated potential soluble mediators of this effect. The hybrid E/M and mammary progenitor-like signature of D2.1 tumors was typified by high expression of DKK3, a Wnt signaling modulator that is preferentially expressed by CD44+CD24– epithelial stem/progenitor cells within the human mammary gland and is generally enriched across stem cell or “immune privileged” niches, making it an attractive candidate (Figure 6A) (38, 39). Interestingly, DKK3 expression was even greater in D2.1 tumors in immunocompetent animals but not D2A1 (Figure 6, B and C). Elevated DKK3 expression in D2.1 cells compared with D2A1 or D2.OR cells was confirmed via quantitative PCR (Supplemental Figure 7A) and its function was assessed after stable shRNA knockdown in D2.1 cells (Supplemental Figure 7, B and C). Upon MFP implantation, shDKK3 D2.1 tumors displayed significantly reduced growth in vivo compared with shScramble controls, with the majority being completely rejected (5/8 mice; Figure 6D and Supplemental Figure 7D). Subsequent analysis of tumor cells after expansion ex vivo confirmed that nonrejected shDKK3 tumors regained DKK3 expression — likely explaining the persistence of these tumors (Supplemental Figure 7E). To validate these results, 2 independent CRISPR DKK3-knockout (DKK3-KO) D2.1 clones were generated (Supplemental Figure 7, F and G). When implanted into immunocompetent animals DKK3-KO cells were completely rejected (Figure 6E), whereas in SCID-beige animals DKK3-KO cells yielded tumors equivalent to control cells (Figure 6F). Thus, DKK3 expression indeed appears to be a necessary component for immune escape and long-term survival during the dormant phase.

D2.1-derived DKK3 is essential for tumor persistence and mediates T cell fate/function. (A) Differential expression analysis of D2.1 vs. D2A1 tumors in BALB/c animals from Figure 2. Dkk3 is highlighted with an empty circle (log2[fold change] = 4.97; Padj = 3.14 × 10–108). (B) Normalized counts of Dkk3 from bulk RNA-seq of D2.1 or D2A1 tumors implanted into BALB/c or SCID-beige animals. Counts were compared by 1-way ANOVA with Holm-Šídák multiple-comparison test. (C) Quantitative PCR of Dkk3 transcripts in samples from B. P values by 1-way ANOVA with Holm-Šídák multiple-comparison test. (D) Growth of 1 × 106 D2.1 cells expressing shScramble (n = 6) or shDKK3 (n = 8) vectors in the mammary fat pad (MFP) of BALB/c mice. (E) Growth of 1 × 106 D2.1 control (n = 10) or 2 independent DKK3-KO lines (n = 5 each) in the MFP of BALB/c mice. Statistical comparisons were by 2-way ANOVA at end of experiment (D and E). (F) Growth of D2.1 control or DKK3-KO cells (n = 5 each) in the MFP of SCID-beige mice. (G) Bead-isolated Jedi CD8+ cells were cultured in D2.1 shScramble or shDKK3 CM with antigen-presenting cells (APCs) and eGFP200–208 peptide. Representative flow plot (left) and quantification (right) of CD8+ cell divisions in D2.1 CM are shown. (H) Jedi CD8+ cells cultured as in G with the addition of equal numbers of bead-isolated CD4+ cells. (I) Representative plots of CD4+ and CD8+ cells (gated on CD45+; left) and quantification (right) of CD4+ cells among CD45+ cells in D2.1 CM. (J) Representative plots (left) and quantification of Tregs in D2.1 CM. (K) CD8+ cell/Treg ratio after culture in D2.1 CM. (L) IFN-γ ELISA after T cell culture in D2.1 CM. Statistical comparisons were performed by 2-tailed t test (I–K), 1-way ANOVA with Tukey’s correction (L), or Šídák’s 2-way ANOVA (G and H). Data are presented as mean ± SEM.

Tumor-derived DKK3 regulates Tregs to inhibit CD8+ T cell function. The mechanistic underpinnings of DKK3-mediated immune evasion were further validated ex vivo by stimulating isolated CD8+ Jedi T cells mixed with antigen-presenting cells (APCs) in D2.1 shScramble or shDKK3 CM. Unexpectedly, no difference in CD8+ T cell proliferation was detected between D2.1 shDKK3 CM and control (Figure 6G). However, additional experiments using a mix of CD8+ Jedi cells, CD4+ cells, and APCs revealed significantly increased total T cells in shDKK3 CM compared with shScramble (Figure 6H), with increased CD8+ T cell proliferation, suggesting an indirect effect of DKK3 on CD8+ T cells. Consistent with this observation, the CD4+ T cell fraction was significantly reduced in DKK3-silenced D2.1 CM (Figure 6I), with a decrease in FoxP3+CD4+ Tregs (Figure 6J), yielding an increased CD8+ cell/Treg ratio (Figure 6K). Similar results were also observed when using CM from DKK3-KO cells (Supplemental Figure 8, A–C). As a proxy for immune activation, we also assessed the level of IFN-γ in media by ELISA and found that DKK3 knockdown resulted in increased IFN-γ compared with control D2.1 CM and was comparable to culturing in D2A1 CM (Figure 6L). Altogether, these experiments support the previous data and implicate CD4+ T cells, particularly Tregs, as a target of D2.1 regulation via secretion of DKK3, with secondary effects on CD8+ cells.

DKK3 is associated with poor survival and immune evasion in human BC. Given the experimental evidence of DKK3-modulated immune function, DKK3 was assessed in BC data sets obtained from clinical samples to determine the relevance of this pathway in humans. In BC, we found that high Dkk3 expression was significantly correlated with poor survival in Basal and Luminal B tumors in both the Kaplan-Meier Plotter and TCGA data sets (Figure 7, A and B). Interestingly, Dkk3 was inversely correlated with genes indicative of an antitumor immune response (e.g., Cd4, Cd8a, Gzma, Gzmb, and Ifng) in Basal BC across data sets, a pattern not observed with family member Dkk1 or Dkk2 (Figure 7C and Supplemental Table 2). Furthermore, Dkk3 was positively correlated with the ectonucleotidases Nt5e (CD73) and Entpd1 (CD39), which are highly expressed by Tregs and critically mediate peripheral tolerance (40), and negatively correlated with the proliferation marker Mki67 (Figure 7C). Expression of Dkk3 was also higher in primary tumors of BC patients who presented with metastasis, even in ER+ tumors that have higher Dkk3 at baseline (Figure 7D and Supplemental Figure 9A). Single-cell RNA-seq analysis of human BCs revealed that Dkk3 is expressed by multiple cells within the breast microenvironment, including both normal and cancer epithelial cells such as TNBC cells (Supplemental Figure 9, B and C). Thus, in human BC both the stroma and tumor cells themselves likely contribute DKK3 to the microenvironment (41).

DKK3 is associated with poor survival, decreased effector function, and metastasis in human BC and promotes tumor growth in an immunocompetent setting. (A) Survival plots of human BC stratified by high and low Dkk3 expression in the Kaplan-Meier Plotter cohort (Basal = 477, Her2 = 348, Luminal A = 903, Luminal B = 676). (B) Survival plots of TCGA human BCs based on Dkk3 expression. (C) Correlation analysis of Dkk1, Dkk2, and Dkk3 with immune-related genes in TCGA, Kaplan-Meier Plotter, and METABRIC data sets. Individual P values are provided in Supplemental Table 2. (D) Dkk3 expression in MBCProject samples with or without metastases present at time of sample collection. Statistical comparisons were performed by 2-tailed t test. (E) Growth of parental, eGFP-expressing only, or DKK3 eGFP D2.OR mammary fat pad tumors (1 × 106 cells; n = 5 each) in female BALB/c mice. Comparisons shown are by Tukey’s 2-way ANOVA at end of experiment. (F) Representative flow plots (left) and quantification (right) of Tregs in total CD4+ cells in tumors from E. (G) Effector phenotype (CD44+CD62L–) of CD8+ T cells in spleens of mice from E. Statistical analysis of F and G was performed by 2-tailed t test. An outlier was identified for G (right; empty triangle) using the outlier analysis function in GraphPad Prism (ROUT method, q = 1%) and the P value displayed excludes the outlier (P = 0.1309 included). Data are presented as mean ± SEM.

DKK3 promotes immune evasion in rapidly progressing models of BC. To directly assess the role of DKK3 in suppressing antitumor immunity, we utilized lentiviral vectors to stably overexpress DKK3 in D2.OR cells, which we found to be highly sensitive to adaptive immune responses (Supplemental Figure 10A and Figure 1G). While DKK3 expression yielded no significant difference in cellular proliferation in vitro, it did elicit a CD44/CD104 profile similar to D2.1 cells or residual D2.OR tumors (Supplemental Figure 10, B and C). Upon MFP implantation in immunocompetent hosts, DKK3 expression provided a robust engraftment and survival advantage compared with control empty vector D2.OR tumors, which were rejected after stable integration of the puromycin xenoantigen (Supplemental Figure 10D). Prior reports suggest that DKK3 enables MHC-I–mismatched transplantation of embryoid bodies (42); therefore, to test the possibility that DKK3 protected against foreign antigen–specific immunity, a vector containing eGFP in addition to DKK3 was lentivirally transduced into D2.OR cells, which were subsequently transplanted into the MFP of BALB/c mice. As with puromycin-expressing D2.OR cells, we found that D2.OR cells expressing eGFP alone were completely rejected, while parental D2.OR cells formed tumors (Figure 7E). However, D2.OR cells expressing both eGFP and DKK3 successfully engrafted, although at a diminished rate compared with parental cells (Figure 7E). Moreover, examination of infiltrating T cells revealed an increase in Tregs compared with parental tumors (Figure 7F) and a systemic decrease in effector CD8+ cells in the spleen (Figure 7G). Together, these data suggest that DKK3 enables antigen-specific protection in otherwise immune-sensitive tumors and supports a role for DKK3 in Treg generation.

Congruent results were observed when expressing DKK3 in highly proliferative D2A1 cells (Supplemental Figure 10F). We found that DKK3-expressing D2A1 cells grew equivalently whether implanted into the MFP of BALB/c or SCID-beige mice, whereas control tumors were substantially delayed in BALB/c animals (Supplemental Figure 10G), and ELISA of serum at time of euthanasia confirmed that DKK3 expression was maintained (Supplemental Figure 10H). By the end of the experiment, effector CD8+ T cells were significantly reduced in the tumor (Supplemental Figure 10I) and large changes in splenic T cells were detected (Supplemental Figure 10J), with T cells that were shifted toward a naive (CD44–CD62L+) phenotype. Fewer anti-eGFP antibodies were also present in the serum of mice bearing DKK3-expressing D2A1 tumors when accounting for tumor volume (Supplemental Figure 10K). In sum, these in vivo data reveal that DKK3 can restrict adaptive antitumor immunity during multiple stages of tumor progression.