Dll1+ quiescent tumor stem cells drive chemoresistance in breast cancer through NF-κB survival pathw

Development of chemoresistance in breast cancer patients greatly increases mortality. Thus, understanding mechanisms underlying breast cancer resistance to chemotherapy is of paramount importance to overcome this clinical challenge. Although activated Notch receptors have been associated with chemoresistance in cancer, the specific Notch ligands and their molecular mechanisms leading to chemoresistance in breast cancer remain elusive. Using conditional knockout and reporter mouse models, we demonstrate that tumor cells expressing the Notch ligand Dll1 is important for tumor growth and metastasis and bear similarities to tumor-initiating cancer cells (TICs) in breast cancer. RNA-seq and ATAC-seq using reporter models and patient data demonstrated that NF-κB activation is downstream of Dll1 and is associated with a chemoresistant phenotype. Finally, pharmacological blocking of Dll1 or NF-κB pathway completely sensitizes Dll1+ tumors to chemotherapy, highlighting therapeutic avenues for chemotherapy resistant breast cancer patients in the near future.

Breast cancer is one of the leading causes of cancer-related death among women1. Despite recent progress in the field, clinical challenges including treatment resistance, recurrence, and metastasis still remain2. Breast cancer is classified into three main subtypes, luminal (expressing estrogen receptor (ER) with or without co-expression of the progesterone receptor (PR) and HER2 gene), HER2-enriched without ER and PR, and Triple-negative breast cancer (TNBC), which does not express any of these receptors3,4. Such intertumoral heterogeneity has so far precluded the identification of “universal” breast cancer therapeutics. Moreover, the known intratumoral heterogeneity of breast cancer is thought to directly contribute to chemotherapeutic resistance, as a small population of tumor cells known as tumor-initiating cells (TICs)/cancer stem cells (CSCs) often are endowed with quiescence, metastatic, and chemotherapy resistance capabilities5,6,7,8,9,10,11 that promote tumor recurrence and metastasis, even after chemotherapeutic treatment12,13.

TICs are present in several solid tumors including breast tumors14, and Notch signaling has been specifically implicated in promoting TICs/CSC function in breast cancer15,16,17,18,19. The Notch pathway is composed of four Notch receptors (Notch1-4) and five ligands (Jagged1, Jagged2, Delta-like1 (Dll1), Dll3, and Dll4). While Notch signaling is dysregulated in breast cancer, and extensive studies support the function of Notch receptors in breast cancer20,21,22,23, the role of specific Notch ligands in promoting TIC function and chemoresistance remains unclear. While global Notch signaling inhibitors reduce tumor growth by targeting TICs, their side effects limit its therapeutic efficacy. Notably, our recent studies highlighted the tumor-promoting function of the Notch ligand Dll1 in luminal breast cancer24. As specific ligand-based therapies are likely to provide a better safety profile than global Notch signaling inhibitors, determining the role of Dll1 in TIC generation/function and Notch-mediated chemotherapeutic resistance, may aid in the identification of effective and better-tolerated therapeutics for breast cancer patients.

In this study, we show that Dll1 is essential for tumor development and metastasis in an Polyoma Middle T (MMTV-PyMT) mammary tumor mouse model. Using reporter models, we further demonstrate a temporal increase of Dll1 expression in tumors as they progress from early to late stages, including metastasis. Mechanistically, we show that within luminal tumors, Dll1+ cells represent quiescent TICs, display a chemotherapeutic resistant gene signature enriched in Hypoxia and NF-κB signaling genes, and are resilient to chemotherapy. Most notably, pharmacological blocking of Dll1 or NF-κB pathway sensitizes tumor cells to chemotherapy and significantly abolishes tumor growth and metastasis in tumor initiation and progression study. Collectively, these data suggest that quiescent Dll1+ TICs are an important regulator of breast tumor progression and metastasis and are responsible for chemoresistance. As such, targeting Dll1 in combination with chemotherapy and NF-κB may be a promising therapeutic strategy for breast cancer patients who are resistant to chemotherapy.

Dll1 conditional knockout reveals an oncogenic function for Dll1 in the luminal MMTV-PyMT but not in MMTV-Wnt1 model

To dissect the role of Dll1 in autochthonous mammary tumors, we crossed Dll1 conditional-knockout (Dll1cKO) mice in which K14-Cre recombinase efficiently knocks out Dll1 in mammary epithelial cells25,26 to MMTV-PyMT to generate PyMT-Dll1cKO animals (Fig. 1a). The MMTV-PyMT model represents a luminal-like breast cancer model with high metastatic potency27,28, thereby permitting analysis of progression from early to late stages of the disease. Dll1 expression was significantly downregulated in tumor cells of PyMT-Dll1cKO mice (Fig. 1b). Moreover, we found that loss of Dll1 led to a dramatic reduction in hyperplasia in PyMT-Dll1cKO compared to PyMT-Dll1WT mammary glands at 8 weeks (Fig. 1c), suggesting that loss of Dll1 affects early events in tumorigenesis, potentially by impacting tumor-initiating populations. In control PyMT-Dll1WT mice, mammary tumors occurred as early as 11 weeks of age, with almost 100% tumor penetrance. In contrast, tumors in PyMT-Dll1cKOmice occurred at 12 weeks of age with a 75% delayed penetrance (Fig. 1d). Moreover, PyMT-Dll1WT mice had a significantly increased tumor burden (Fig. 1e) and lung metastasis (Fig. 1f, g) compared to PyMT-Dll1cKO mice.

To determine if the function of Dll1 was specific to the luminal adenocarcinoma MMTV-PyMT model, we crossed the Dll1cKO mice to MMTV-Wnt1 mice, which represents a basal/mixed tumor model29, to generate Wnt1-Dll1cKO mice (Supplementary Fig. 1a). In contrast to the PyMT model, mammary glands from both Wnt1-Dll1cKO and Wnt1-Dll1WT mice exhibited similar hyperplastic growth (Supplementary Fig. 1b). Furthermore, there was no significant difference in tumor onset in Wnt1-Dll1cKO compared with Wnt1-Dll1WT mice (p = 0.4) (Supplementary Fig. 1c), suggesting a specific tumorigenic function for Dll1 in luminal adenocarcinoma.

A dramatic increase of Dll1+ tumor cells during PyMT tumor progression

The dramatic effect of Dll1 deletion on PyMT tumor formation prompted us to investigate the spatial and temporal distribution of Dll1 expression during tumorigenesis. For this, we generated PyMT-Dll1mCherry reporter mice25,30 and evaluated the expression pattern of Dll1+cells during PyMT-Dll1mCherry tumor progression from normal mammary gland to late-stage tumor. We observed a substantial increase in the number of Dll1+ cells as the mammary gland advanced through the hyperplastic stage to the tumor (Fig. 2a, b), supporting an oncogenic function Dll1 in PyMT tumors. The mRNA levels of Dll1 in sorted mCherry positive and negative populations from hyperplasia and tumors indicated faithful expression of the Dll1 in reporter mice (Fig. 2c). To analyze the status of Dll1+ cells in basal and luminal cell compartments, we performed confocal microscopy using basal (K14 and K5) and luminal (K8) markers along with Dll1 mCherry staining. As previously reported, we found that Dll1 is predominantly expressed in the basal layer of the normal mammary gland25 (Fig. 2d). However, there was a relative increase in the number of Dll1+ luminal (K8+) cells when compared to Dll1+ basal tumor cells (K14+ and K5+) as tumors progressed from hyperplasia stage to late tumor (Fig. 2d, e and Supplementary Fig. 2a, b).

To further investigate the distribution of Dll1+ cells during tumor development, we used PyMT-Dll1-GFP mouse model. We crossed Dll1GFP-IRES-Cre-ERT2 knockin mice (Dll1GFP)31 with MMTV-PyMT mice to generate PyMT-Dll1GFP mice. The GFP expression faithfully recapitulated Dll1 expression (Supplementary Fig. 2c), and similar to PyMT-Dll1mCherrymodel, the number of Dll1+ cells (GFP+) in PyMT-Dll1GFP mice was significantly enhanced, and Dll1+ cells acquired expression of the luminal marker (K8) as tumors progressed in this luminal model (Supplementary Fig. 2d–f and Supplementary Fig. 3a, b). These studies identified the cellular and temporal distribution of Dll1 at different stages of PyMT tumor development using multiple reporter models.

Metastatic lung nodules show increased expression of Dll1 compared to primary tumors

The MMTV-PyMT mouse model develops lung metastasis at a high frequency28. To determine if there is an association between the expression of Dll1 and metastasis, we assessed Dll1 expression in PyMT-Dll1mCh and PyMT-Dll1GFP lung metastatic nodules and compared them to primary tumors. We found higher levels of Dll1 expression in metastatic nodules in both reporter models as compared with the primary tumors (Supplementary Fig. 4a–f). Collectively, our studies from Dll1 conditional knockout (Fig. 1) and reporter mice suggest that Dll1 may be a critical determinant of metastasis in breast cancer.

Dll1+ tumor cells display TIC characteristics

Increased metastasis is often linked to the presence of Tumor-initiating cells (TICs)/Cancer stem cells (CSCs) in cancer32,33,34. To determine if Dll1+ cells display a TIC phenotype in breast tumors, we performed an in vitro tumorsphere assay to assess the tumorigenic potential of Dll1+ and Dll1− tumor cells isolated from PyMT-Dll1mCh tumors. Sorted single PyMT-Dll1+ tumor cells formed a significantly higher number of tumorspheres compared with PyMT-Dll1− tumor cells, suggesting PyMT-Dll1+ cells are enriched in TICs (Fig. 3a). To directly test the tumor-initiating function of Dll1, we transplanted sorted PyMT-Dll1+ or PyMT-Dll1− tumor cells into the mammary fat pad (MFP) of syngeneic C57BL/6 mice. We found that transplanted Dll1+ tumor cells initiated tumor growth before Dll1− tumor cells, suggesting that Dll1+ cells display characteristics of TICs (Fig. 3b). In limiting dilution assay, Dll1+ tumor cells showed a higher capability (1/2791) to initiate xenograft growth compared to Dll1− tumor cells (1/27,341) (Fig. 3c). Intriguingly, once Dll1− tumors started growing, they grew faster than Dll1+ tumors, suggesting a higher proliferative potential of Dll1− tumor cells (Fig. 3d, e). As TICs have a slow proliferating phenotype35, we used in vivo EdU labeling36 to assess tumor cell proliferation. Notably, we found that Dll1+ cells were EdU negative, indicating reduced proliferation compared to Dll1− cells (Fig. 3f), further supporting the TIC-like characteristics of Dll1+ cells. Together, our studies identify Dll1+ cells as potential TICs with reduced proliferative potential in breast cancer.

Dll1-mediated Notch signaling activates the NF-κB pathway in breast tumors

To better understand the molecular signatures between Dll1+ and Dll1− tumor cells in breast cancer, we performed RNA-seq analysis using Dll1+ and Dll1− tumor cells from both PyMT-Dll1mCh and PyMT-Dll1GFP tumors (Supplementary Fig. 5a). As expected, signatures linked to stem cells, invasion and quiescent cells were enriched in Dll1+ tumor cells (Fig. 4a, b and Supplementary Fig. 5b, c), supporting our experimental data that Dll1+ cells have stem-like properties that increase their invasive properties. NF-κB and Hypoxia signaling are intimately linked to chemoresistance through regulation of cell proliferation and survival in several cancers, including breast cancer37,38,39,40. Next, we found activation of NF-κB, hypoxia and Notch pathways enriched in Dll1+ tumor cells (Fig. 4c and Supplementary Fig. 5d, e), which may contribute to the chemoresistance. Finally, we also found increase in metastasis signature to be enriched in Dll1+ tumor cells (Fig. 4d), supporting our in vivo metastasis data (Supplementary Fig. 4a–f). As expected, RNA-seq data demonstrated a ~7.5 higher level of DLL1 expression in this experimental setup. Considering a stringent cut off of >5.5 fold change gene expression in Dll1+ vs. Dll1− tumor cells, we identified a set of 10 genes that are strongly expressed in Dll1+ cells as compared to Dll1− cells (Supplementary Fig. 5f). Notably, 8 out of these 10 genes are related to lung and bone metastasis41,42,43,44,45,46,47,48, suggesting that they could represent potential biomarkers.

To further investigate the mechanistic basis of Dll1-mediated effect on metastasis and cell survival of tumor cells, we performed ATAC-seq to examine the chromatin accessible and restricted regions in Dll1+ and Dll1− tumor cells (Supplementary Fig. 5a). On a very global level, ATAC-seq data do not indicate any dramatic, epigenome-wide changes in chromatin accessibility patterns between Dll1+ and Dll1− cells. This conclusion is based on finding similar numbers of UTR, exon, intergenic, intron, and promoter peaks in both Dll1+ and Dll1−tumor cells (Supplementary Fig. 5g). This finding is rather expected, given that these cells are of the same lineage, and as Dll1 is not itself a chromatin-modifying protein. We next performed an integrated analysis with ATAC-seq and RNA-seq data. For the integrated analysis, we took the “ATAC-seq Up” signature (all genes with a proximal ATAC-seq peak increasing significantly at p-adjusted < 0.05 in Dll1+ samples) and tested it for enrichment in the GFP/mCherry Up (Dll1+) combined RNA-seq ranked list. We found that this ATAC-seq signature is strongly enriched in the RNA-seq data (NES = 2.24, p < 0.001) (Fig. 4e). We then focused analyses on the 23 GSEA “Core Enrichment” genes (Fig. 4e) that are both in the ATAC-seq Up signature, and transcriptionally upregulated in Dll1+ cells using EnrichR online software. Our data shows that of these 23 common upregulated/increased accessibility genes from Dll1+ cells, there are significant enrichment of both Notch signaling and NF-kB signaling (e.g., RelA) (Supplementary Fig. 5h), suggesting that Dll1-mediated Notch signaling may regulate NF- kB1 activity and signaling.

ATAC-seq data also identified the presence of enriched peaks of several of the prometastatic genes identified in the RNA-seq data from Dll1+ cells (volcano plot, Fig. 4f). Dll1+ tumor cells also have NF-kB1 specific open chromatin region/enriched peak compared to Dll1− cells (Fig. 4g). Together, these data indicate that Dll1 activates NF-kB1 signaling in Dll1+ tumor cells, potentially contributing to protumor survival and associated chemoresistance. This observation was further confirmed by an IHC analysis, where Dll1+ tumors showed greater nuclear translocation of NF-kB compared with Dll1− tumors suggesting activation of NF-kB signaling (Fig. 4h, i).

To explore the clinical correlation of our mouse data, we examined our Dll1+ mouse signature with METABRIC, a patient database to analyze correlation of DLL1 to NF-kB in luminal A and B breast cancer patients49,50. Notably, we found that genes from DLL1highpatient tumors show strong enrichment to Dll1+ signature from PyMTGFP tumors or PyMTmCh (top 100 upregulated genes) (Fig. 4j and Supplementary Fig. 6a), highlighting the close resemblance of the mouse model to patients. Also, when patients were stratified based on DLL1 expression, the DLL1high tumors from patients had upregulated NF-kB signatures similar to mouse of RNA-seq data (Fig. 4k, l and Supplementary Fig. 6b, c). Moreover, tumors from luminal A and B patients show a strong positive correlation between DLL1 and NF-kB (R = 0.185, p < 0.001, Fig. 4m). Finally, we found NF-kB motif genesets to be significantly enriched in DLL1+ patient tumors (Luminal A and B) compared to DLL1− patients (Supplementary Fig. 6d) using GSEA C3 compendium. Together, these observations suggest that Dll1-mediated Notch signaling induces activation of NF-kB signaling by regulating NF-kB1, which may contribute to tumor cell survival.

Dll1+ TICs are chemotherapy resistant

TICs are thought to be responsible for resistance to chemotherapies in various cancers51,52. Therefore, we next assessed sensitivity to doxorubicin, an anthracycline chemotherapeutic drug effective in the treatment of a wide range of malignancies, including breast cancer53,54. To functionally test whether Dll1+ cells are resistant to doxorubicin in the PyMT-Dll1mChreporter model, we sorted out Dll1+ and Dll1- tumor cells and treated them in vitro with doxorubicin. We found that Dll1+ cells were resistant to doxorubicin-induced cell death compared to Dll1− cells (Supplementary Fig. 7a, b). Similar to our in vitro studies, treatment of transplanted Dll1+ and Dll1− tumors with doxorubicin revealed that Dll1+ tumors were resistant, whereas Dll1− tumor growth was significantly impaired upon doxorubicin treatment (Fig. 5a–d). Particularly, doxorubicin treatment significantly increased the number of quiescent/slow growing Dll1+ TICs in tumors (Fig. 5e and Supplementary Fig. 7c), potentially due to their evasion of cell death and/or DNA damage. Indeed, further analysis of transplanted tumors revealed that Dll1+ tumors were less prone to cell death compared to Dll1− tumors as seen by cleaved-caspase-3 and TUNEL staining (apoptotic cells) in the tumor cells (Fig. 5f–h). Cell cycle analysis further demonstrated a strong increase in a sub-G1 population (apoptotic cells) of Dll1− cells compared to Dll1+ tumor cells (Supplementary Fig. 8a, b). Finally, γH2AX staining (green staining indicates DNA damage) shows reduced doxorubicin-induced DNA damage in Dll1+ tumor cells compared to Dll1− tumors (Supplementary Fig. 8c–e), potentially explaining the chemotherapeutic resistance of Dll1+tumors (Fig. 5b–d).

Pharmacological inhibition of Dll1 sensitizes Dll1+ tumor cells to chemotherapy

Next, we investigated whether targeting Dll1 in vivo could impact the chemoresistant phenotype of Dll1+ TIC/CSCs. Dll1+ tumor-bearing mice were given either the control vehicle or anti-Dll1-blocking antibody (αDll1)55 or doxorubicin (Dox), or a combination of both (Supplementary Fig. 9a). In this tumor initiation and progression study, Dll1 antibody alone had a modest inhibitory effect on tumor growth, whereas, the combination treatment dramatically reduced tumor growth (Fig. 6a, b), suggesting that Dll1-blocking sensitizes Dll1+ TICs to chemotherapy. Moreover, metastatic events were significantly decreased in mice treated with anti-Dll1 antibody and chemotherapy compared with untreated controls (Fig. 6c–e and Supplementary Fig. 9b). This decrease in tumor growth and progression was associated with reduced nuclear translocation of NF-kB (Fig. 6f, g), suggesting that Dll1-mediated chemoresistance is potentially driven by NF-kB signaling.

There were no observable side effects of Dll1 blocking in mono or in combination with doxorubicin, as body weight of these mice were unaltered (Supplementary Fig. 10a). The activity of the Dll1 antibody was further confirmed by examination of depletion of marginal B-cells in the spleen that depend on Dll1 activity (Supplementary Fig. 10b, c)56,57,58. Taken together, these results delineate that Dll1 targeting may be effective for patients failing chemotherapy and may have a better safety profile during treatment.

Pharmacological inhibition of NF-kB and Dll1 resensitizes chemoresistant Dll1+ luminal tumor cells, phenocopying the effects of Dll1-blocking antibody

Since NF-kB was found to be downstream of Dll1-mediated Notch signaling, we next sought to determine if the NF-kB inhibitor IMD-0354 (IMD) would resensitize tumor cells to chemotherapy, phenocopying the effect of the Dll1-blocking antibody. IMD is a selective IKK2 inhibitor, which disrupts NF-κB signaling, which showes activities in several preclinical models including cancer59,60,61,62. An in vitro tumorsphere assay with Dll1+ tumor cells demonstrated a significant decrease in the number and size of tumorspheres with a combination treatment of IMD and doxorubicin, similar to effects seen with Dll1-blocking antibody and doxorubicin (Fig. 7a–c). Especially, combination treatment of Dll1 antibody, IMD and chemotherapy further decreased the number and size of tumorspheres (Fig. 7a–c), while either IMD or chemotherapy alone had minimum effect on tumor cells. To complement the in vitro data, we treated Dll1+ tumor cells with doxorubicin, Dll1-blocking antibody, IMD-0354 treatment alone or in combinations (Supplementary Fig. 10d). Similar to in vitro results, NF-kB inhibition along with αDll1 ab or doxorubicin was very effective to inhibit tumor growth with no significant toxicity (Fig. 7d, e and Supplementary Fig. 10e). Furthermore, combination treatment targeting Dll1, NF-kB and doxorubicin significantly abolished tumor growth of Dll1+ tumor cells. Thus, our results suggest that activation of NF-kB downstream of Dll1 signaling in tumor cells promotes chemoresistance and blocking all these components together can lead to significant response in thes tumors.

Notch signaling, specifically the Notch receptors, has been associated with chemoresistance in several cancers, including breast cancer, most markedly in TNBC63,64,65,66,67,68, suggesting that targeting this pathway may improve clinical outcome. Our study reveals a central role for the Notch ligand Dll1 in promoting tumor progression, metastasis and chemoresistance in aggressive luminal breast tumors. Identification of a ligand-based therapy to block tumor-specific Notch signaling may provide a safer and effective therapeutic option for breast cancer patients.

Earlier studies from our group have demonstrated that the Notch ligand Dll1 is important for normal mammary stem cell development25. We further showed that Dll1 is preferentially upregulated in luminal breast cancer compared to TNBC/basal cancer and regulates CSC function in these breast cancer cells24. However, at that time, we could not evaluate the spatial and temporal distribution of Dll1 expression during tumor progression and metastasis. Furthermore, we could not determine if Dll1+ cells themselves had TIC/CSC properties due to the lack of appropriate mouse models. In this study, we used Dll1 conditional knockout and reporter models crossed with spontaneous murine models (MMTV-PyMT and MMTV-Wnt1) to determine the spatial and temporal distribution of Dll1 and also determine the function of Dll1+ tumor cells in tumor progression. We found that Dll1+ cells increase dramatically from normal mammary gland to hyperplasia and tumor, suggesting a tumor supporting function of Dll1. We found that Dll1+ tumor cells increase in K8+ luminal compartment compared to basal cells as tumor progresses from hyperplasia to tumor stage in MMTV-PyMT model. Future studies on lineage tracing starting at earlier age in mice, will help to decipher the mechanism behind this increase in K8+ Dll1+ luminal cells in tumors. Using conditional-knockout mice of Dll1, we further demonstrated that Dll1 has a significant tumor-promoting function in MMTV-PyMT tumors but not in MMTV-Wnt1 tumors. It is possible that Dll1 may have tumor-promoting function in other basal tumors, which needs further explorations. We further show that Dll1+ tumor cells are quiescent TICs and Dll1-mediated Notch signaling is one of the major drivers of tumor progression and metastasis in the the MMTV-PyMT model using both reporter and conditional-knockout mouse models. These data are in accordance with the Metastatic Breast Cancer project database (Broad Institute), which highlights the amplification of Dll1 in metastatic breast cancer patients.

Tumor heterogeneity is currently a clinical challenge for breast cancer therapy11,69,70,71. To determine if the Dll1+ tumor cells resemble TICs that could increase tumor burden, metastasis, and chemotherapeutic resistance, the transcriptional signatures of Dll1+ and Dll1− tumor cells were further evaluated by RNA-seq analysis. These data show that Dll1+tumor cells are associated with increased invasiveness and metastatic signatures in GSEA, corroborating experimental data. The RNA-seq data indicated that Notch target genes are activated in Dll1+ tumor cells compared with Dll1− tumor cells. However, we need further evaluations to understand Notch signalling between Dll1+ cells and Notch receptor-expressing cells. Moreover, Dll1+ tumor cells are enriched for stem cell genes and as expected, the Dll1+ cells showed reduced expression of cell cycling genes, while Dll1− cells were highly enriched, highlighting the relative quiescent properties of Dll1+ tumor cells, which is confirmed by EdU assay. Previous studies suggest that CD44hi/CD24lo epithelial cells retain classic TIC features in breast cancer, including tumor-initiating capacity in vivo, tumorsphere formation, and resistance to standard chemotherapy13. We found that Dll1+tumor cells are resistant to the chemotherapeutic drug, doxorubicin. Mechanistically, we showed that Dll1+ tumor cells resist chemotherapy by avoiding drug-induced cell death and DNA damage as seen by cleaved-caspase-3 assay, γH2AX staining, TUNEL and cell cycle assays. Interestingly, Dll1+ cells were also enriched for signatures of NF-κB signaling and hypoxia, common features of cancer progression and chemoresistance72,73,74,75. This observation was further confirmed in human patients data using METABRIC dataset, where we found strong positive correlation between DLL1 and NF-κB1. Moreover, Dll1high gene signatures (top 100 genes) from our mouse models closely resemble genes of DLL1highLuminal A and B patients supporting the clinical significance of our model to human patients. Indeed, the NF-κB pathway can crosstalk with Notch signaling in lung cancer to promote chemoresistance76. Moreover, Hypoxia has long been associated with chemoresistance to several drugs, including doxorubicin, in multiple cancers such as lung, prostate, and breast cancer39,77,78,79. ATAC-seq corroborates RNA-seq data and identified several open chromatin regions for the NF-κB1 gene and other protumorigenic and prometastatic genes, indicating that Dll1-mediated Notch signaling may regulate these pathways to resist chemotherapy. Furthermore, increased nuclear translocation of NF-κB was observed in tumor cells expressing higher levels of Dll1. Blocking of NF-kB activity using the IMD inhibitor in combination with doxorubicin significantly decreased tumor cells viability and stem cells activity in tumorsphere assays, thus phenocopying the effects of the Dll1-blocking antibody. Combination treatment blocking NF-kB activation, Dll1 and chemotherapy shows almost complete respone in tumor growth in vivo.

Together, our data suggest that Dll1-mediated Notch signaling drives metastasis and chemoresistance and may therefore be responsible for higher mortality in breast cancer (Fig. 7f). Mechanistically, our data suggest that Dll1 increases resistance to DNA damage and cell death through the NF-κB pathway. Remarkably, combination therapy targeting Dll1 ligand and NF-κB pathway shows complete response to the doxorubicin in Dll1+ tumor cells. Altogether, our findings demonstrate a promising therapeutic target in breast cancer patients with advanced disease and metastasis, particularly for those who are chemoresistant. Further preclinical studies will assess the potential efficacy of this strategy in the clinics.