All cancer cell lines were cultured in DMEM medium supplemented with 10% fetal bovine serum, 1% penicillin/streptomycin, and 1% glutamine (Gibco). The human TNBC cell line MDA-MB-231 was purchased from ATCC, and the mouse TNBC cell line C0321 was generated from Lfng-/- mouse on FVB background as described (93). TNBC PDX cell lines, 2K1 and 4IC, were generated as described (94, 95). The cDNA encoding Notch1-intracellular domain (Notch1-IC) was subcloned into pBABE-puro vector (Cell BIOLABS, INC). Notch1-IC was transfected into MDA-MB-231 cells using Lipofectamine 2000. Stable Notch1-IC expressing and vector control cells were selected under puromycin and Notch1 expression was confirmed by Western blot analysis. For ex-vivo tumor-spheroid/organoids experiments, mouse C0321 cells were transformed with a pmCherry-N1 cloning vector (Life Science Market) using Lipofectamine 2000. The stable mCherry expressing C0321 cells were selected under kanamycin and then enriched with flow cytometry sorting (BD FACSAria II cell sorter, BD Biosciences).

Plasmid encoding APP CTF (APP C99) and Notch1 juxtamembrane region (NOTCH1) were constructed as reported (39). HEK 293T PS1 (PS1+/+, PS2-/-) and PS2 (PS1-/-, PS2+/+) cell lines were established as described in (96). APP C99 and NOTCH1 were transiently transfected into HEK 293T wild type, PS1, and PS2 cells using polyethyleneimine. After 16 h of incubation, fresh media with different concentrations of SS or SF were added. Conditioned media were collected after 24 h and assayed by Aβ ELISA as described (39). H4 cells stably overexpress APP C99 and NOTCH1 were treated with different concentrations of SS. Conditioned media were collected after 24 hours and assayed by Aβ ELISA (39).

Primary mammospheres were obtained as previously explained (14). Briefly, MDA-MB-231, C0321, or PDX cells were cultured in MammoCult Human Medium (STEMCELL Technologies) in ultra-low attachment 6-well plates (Corning). After seven days, the primary mammospheres were dissociated into single cells using trypsin and replated in the presence of different concentrations of SS (5, 10, 25, 50, or 100 μM) alone or in combination with 5 μM MK-2206 (AKT-inhibitor). After one week of treatment, mammospheres with a diameter of >100 μm were counted using a Nikon Eclipse microscope. Results were represented as the percentage of mammospheres where control mammospheres were 100%.

Western blot analysis was performed as previously described (14). Briefly, cells were lysed in RIPA buffer (Santa Cruz Biotechnology) and 1 mM Protease and Phosphatase Inhibitor Cocktail (ThermoScientific). The protein samples were resolved in 7.5% Criterion TGX Precast Gels (Bio-Rad), transferred to PVDF membranes (Immobilon-FL Transfer Membrane, Millipore), and blocked in Odyssey blocking buffer (LI-COR). Membranes were incubated with primary antibodies against Notch1 (D1E11), Notch1-IC (Val1744:D3B8) (Cell Signaling), GAPDH, and β-tubulin (Santa Cruz Biotechnology). Following incubation with secondary antibodies (goat anti-mouse 680RD or goat anti-rabbit 800CW; LI-COR), the results were analyzed using the LI-COR Odyssey imaging system.

T-cell proliferation was measured using CFSE dye dilution flow cytometric measurement as described (97). Briefly, T-cells were enriched from naïve FVB mouse spleen using a T-cell (CD3) isolation kit (Stemcells Technologies). Isolated T-cells were then labeled with 1 μM CFSE and plated in a 24-well culture plate with plate-bound anti-CD3 and anti-CD28 (1 μg/ml each). T-cells were treated with SS (5, 25, or 50 μM) at the beginning of incubation, and T-cell proliferation was measured after 72 h by CFSE dilution using flow cytometry. Bone marrow-derived MDSCs (BM-MDSCs) were generated from FVB mice as described (98). Briefly, bone marrow cells were harvested from FVB mouse femur and tibia bones and cultured with G-CSF, GM-CSF, and IL-6 (20 ng/ml each) for four days to generate BM-MDSCs in the presence or absence of SS (5, 25, or 50 μM). CFSE labeled T-cells were co-cultured with BM-MDSCs at a 4:1 (T-cells: MDSC) ratio with SS (5, 25 or 50 μM) in a 24-well culture plate with plate-bound anti-CD3 and anti-CD28 (1 μg/ml each). T-cells proliferation was measured after 72 h by CFSE dilution using flow cytometry.

Organoids were derived from syngeneic TNBC C0321 tumors as described (94) with a slight modification. Tumors were harvested, minced, and digested at 37^°CC in DMEM/F12 Glutamax complete medium (10% FBS, 1% penicillin/streptomycin; Gibco) containing 1mg/ml type IV collagenase (Gibco). The digested tumor was passed over a 70 μm and a 40 μm strainer to isolate organoids of 40-70 μm. The organoids were resuspended in type I rat tail collagen and plated in 8-well chambers (Nunc™ Lab-Tek™ II Chamber Slide™, ThermoScientific). The organoids-collagen cultures were incubated at 37°CC for 30 min to allow the collagen to solidify. The cultures were then hydrated with DMEM/F12 Glutamax complete medium. The organoids were treated with 1 or 5 μM SS with or without 1 μg/ml α-PD1/α-PDL1. Live cells were stained using CellTracker™ Red CMTPX (Invitrogen), and dead cells were labeled using a cell membrane-impermeable dye, NucGreen™ Dead 488 ReadyProbes™ (Invitrogen). Organoids were imaged on days 4-6 using a BZ-X800 (Keyence) microscope.

All animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) at the Louisiana State University Health Sciences Center (LSUHSC). Tumors were induced by injecting 1 million TNBC C0321 cells into syngeneic mice with 1:1 ratio of Matrigel to PBS into the mammary fat pad of 6-10-weeks old female FVB mice (Jackson Laboratory). Upon detection of a palpable mass, mice were treated with SS alone (60 mg/kg by PO) daily for another 14 days. For combination immunotherapy experiments, palpable tumors were treated with SS (20 mg/kg, daily, PO) alone or in combination with α-PD1 (100 μg/mouse twice per week) for another two weeks. Tumor volume and body mass were monitored for 21 days from the injection of C0321 cells. Tumors were harvested and were either processed for flow cytometry analysis or formalin fixed and paraffin embedded (FFPE). FFPE tissues were sectioned at 4 microns in thickness and stained with H&E to examine tumor morphology. Immunohistochemistry for Notch1 and Jagged1 was performed as previously described (99); antibodies used included a rabbit polyclonal anti-Notch-1 (Abcam, ab27526, 1:500 dilution), and a rabbit polyclonal raised against aminoacids 1110-1223 of human Jagged1 (Santa Cruz Biotechnology, (H-114, 1: 500 dilution). For flow cytometry analysis, tissues were digested with Liberase and tumor single cells suspensions were analyzed for tumor-infiltrating T-cells (CD4 and CD8), dendritic cells (CD11c), MDSCs (CD11b+Gr1), and TAMs (CD11b+F4/80). All cells were gated on the leukocyte markers (CD45+).

RNA sequencing was done at the Translational Genomics Core (TGC) at the Stanley S. Scott Cancer Center, LSUHSC, New Orleans, LA. RNA was isolated from control, or SS-treated tumor tissues using an RNA isolation kit (Qiagen) per the manufacturer’s protocol. RNA integrity was analyzed on Agilent BioAnalyzer 2100 (Agilent). Paired-end libraries (2 x 75) were prepared using the TruSeq Stranded mRNA Library Prep kit, validated, and normalized following the manufacturer (Illumina, San Diego, CA) protocol. Libraries were sequenced in the NextSeq500 using a High Output Kit v2.5, 150 cycles from Illumina.

Bioinformatics analysis of RNA-Seq data was performed at LSUHSC’s Bioinformatics and Data Science Service Center. We processed the data, raw sequence reads to remove probe IDs with very low and or no expression values across all samples, from the gene expression data matrix. We mapped the probes onto the Ensemble database using BioMart, an Ensemble tool to identify the corresponding gene symbols or names (100). The resulting gene expression data set with gene names was normalized using quantile normalization (101). Using normalized data, we performed supervised analysis comparing gene expression levels between treatment and control samples using a t-test implemented in Pomelo2 (102). This unbiased approach was conducted to identify all genes significantly (p < 0.05) responsive to treatment. In addition to the p-values, we computed the log fold change (Log2(FC)). We used the false discovery rate (FDR) to correct for multiple hypothesis testing (103). The resulting set of genes were ranked based on P-value, FDR and log2(FC). Significantly differentially expressed genes were subjected to unsupervised analysis using hierarchical clustering implemented in the Morpheus software package (104) to determine their patterns of expression profiles. For hierarchical clustering we used the Pearson correlation as the measure of distance between pairs of genes and complete linkage as the clustering method. We performed functional analysis using the Ingenuity Pathway Analysis (IPA) software platform (105) the signaling pathways dysregulated in response to treatment. Under this approach, differentially expressed genes responsive to treatment were mapped onto networks and canonical pathways, using IPA. We used Fisher’s exact t-test implemented in IPA to determine the probability of correctly predicting the pathway onto which the gene maps. The pathways were ranked on log-p-values and the significant ones were selected. We performed Gene Ontology (GO) analysis implemented in IPA to categorize genes according to the cellular components, molecular functions and biological processes in which they are involved (106).

Previously, the Golde laboratory and others demonstrated that some NSAIDs, including SS, have γ-secretase modulator (GSM) activity (107–110). SS has both GSM and GSI activity at higher concentration (50). Sulindac is a prodrug, which is metabolized into SS and SF by liver enzymes and colonic bacteria (111). SS is a non-selective COX-1 and COX-2 inhibitor and mediates the anti-inflammatory effects of sulindac (64). SF lacks COX-1 and COX-2 inhibitory activity or anti-inflammatory properties, but retains a number of off-target activities (64). We sought to confirm the GSM activity of sulindac derivatives SS and SF in parallel. Therefore, we tested the effects of SS and SF on γ-secretase cleavage of Notch1 and APP-C99 (β-amyloid precursor peptide, a positive control) using HEK wild-type or presenilin-1 (PS1) and presenilin-2 (PS2) KO cells. PS1 and PS2 are catalytic subunits in the γ-secretase complex which are necessary to cleave amyloid precursor protein, generating β-amyloid (112). Therefore, the PS1 and PS2 KO HEK cells were used as controls for γ-secretase activity. SS significantly inhibited Notch1 and APP C99 cleavage, but SF had a very modest effect (Figures 1A, B). We therefore focused on SS. Next, we tested SS on H4 cells stably overexpressing APP-C99 and Notch1 to confirm GSM activities. SS inhibited the cleavage of Notch1 and APP-C99 in a dose-dependent manner (Figure 1C). Similarly, SS inhibited the γ-secretase mediated release of the intracellular cytoplasmic domains of Notch1 (Notch1-IC in human TNBC MDA-MB-231 cells (Figure 1D). Our results suggest that SS has GSM activities and can be tested as a candidate Notch cleavage inhibitor in TNBC.

The mammosphere formation assay provides an informative and convenient in vitro tool to study sphere-forming CSC (113). However, this assay does not address the complexity of CSC formation and maintenance in an in vivo niche. Previously, we showed that clinical investigational GSIs are not pharmacologically equivalent, and GSI PF-3084014 (nirogacestat) had the most potent mammosphere inhibitory activity in TNBC cell lines (39). Additionally, we observed an additive effect of GSI PF-3084014 on mammosphere growth in combination with AKT inhibitor MK-2206 (14). Here, we sought to determine whether SS alone or in combination with AKT inhibitor MK-2206 has mammosphere inhibitory activity analogous to GSIs. We tested the activity of SS in three models, 1) human TNBC MDA-MB-231 cells, 2) mouse TNBC C0321 cells, and 3) TNBC PDX cells. SS inhibited mammosphere growth in both MDA-MB-231 and mouse C0321 cells in a dose-dependent manner (Figures 2A, B). As previously observed with GSI PF-3084014, when we combined SS with MK-2206, we observed an additive effect on mammosphere growth in both cell lines (Figures 2E, F).

PDXs are important models to test human cancer experimental therapeutics (114). Recently, in collaboration with the Burow lab, we characterized several different PDX models from TNBC patients (95, 115, 116). Cell lines were generated from PDX tissues and plated as monolayers as described (14). We developed mammospheres from those cell lines using MammoCult media as described in the Methods section and then tested the efficacy of SS to inhibit mammosphere growth. We found that SS dose-dependently decreased mammosphere growth in 2K1 PDX (Figure 2C) but not in 4IC PDX (Figure 2D). This PDX model derived from a highly aggressive TNBC that was multi-drug resistant and ultimately fatal (117). Interestingly, when we combined SS with AKT inhibitor MK-2206 we observed an additive effect on mammosphere growth from both PDX models (Figures 2G, H). Overall, these results reveal that SS has single agent anti-mammosphere activity in different TNBC models and enhances the activity of an AKT inhibitor in a multi-drug resistant PDX model. This is consistent with the findings of Bhola et al. (30) who reported that inhibition of the PI3K-mTOR-AKT pathway in TNBC increases Notch1 expression in TNBC, and Notch inhibition restores sensitivity to inhibitors of this pathway.

Next, we explored whether the anti-mammosphere activity of SS depends on Notch signaling. To answer this question, we generated stable MDA-MB-231 cells expressing cleaved, intracellular Notch1 (Notch1-IC) as described in the Methods section. Overexpression of Notch1-IC was confirmed by Western blotting (Supplementary Figure 1). We developed mammospheres from control (vector-transfected) and Notch1-IC overexpressing cells and treated them with SS (5 or 50 µM) or vehicle. Notch1-IC overexpressing cells developed larger mammospheres compared to controls (Figure 3) and were essentially insensitive to SS treatment (Figure 3). These results indicate that overexpression of active Notch1 rescues the anti-mammosphere activity of SS. As an additional control, we tested whether a selective COX-2 inhibitor, rofecoxib, had anti-mammosphere activity in our hands. Our results showed no effect of rofecoxib over a wide range of concentrations (Supplementary Figure 2). These results support the conclusion that the anti-mammosphere activity of SS is unrelated to COX inhibition.

We characterized the tumor microenvironment of a Notch-driven, immune-competent murine transplantable TNBC model developed by the Xu lab from targeted, conditional knockout of Lunatic Fringe (LFng-/-) in mice of FVB background (93). LFng-deficient tumors and cell lines expressed high levels of Notch ligand Jagged1 protein and mRNA levels and showed constitutive Notch activation (93). These tumors recapitulate the molecular profiles of human mesenchymal (claudin-low) and basal-like TNBCs (118). Two transplantable clones were isolated from these tumors (93), a mesenchymal/claudin-low clone (C0321) and a basal-like clone (B5725). Here, we used C0321 cells to develop an immunocompetent, Notch-driven TNBC mouse model by injecting 1 million cells into the mammary fat pad of syngeneic FVB female mice. We characterized tumor growth kinetics and tumor-infiltrating immune cell populations in the model. Three weeks after tumor injection, we harvested tumors to generate single cells suspension for flow cytometric analysis of tumor-infiltrating immune cell populations. We found tumor-infiltrating T-cells (CD4, CD8), TAMs, MDSCs, and immune checkpoints including PD1, Lag3, and CTLA4 in C0321 tumors (Supplementary Figure 3). These findings support the use of this TNBC syngeneic mouse model for in vivo experiments including combination immunotherapy with checkpoint inhibitors.

Several NSAIDs, including sulindac, have received considerable attention as potential chemopreventive agents, as reviewed in (59). In combination with epirubicin, sulindac showed preliminary anti-tumor activity in phase I clinical trials in patients with advanced malignancies, including breast cancer, thus encouraging further investigation (119). Sulindac and docetaxel were tested in a phase II clinical trial in recurrent or metastatic breast cancer (NCT00039520). Yin et al. reported that as single agent sulindac was effective against 4T1 murine breast cancer models and increased the survival of tumor-bearing mice (71). These authors reported that, in their model, sulindac was ineffective in nude mice, suggesting the importance of immune cell-mediated anti-tumor effects of sulindac. Therefore, we wanted to test the efficacy of SS as a single agent using our immunocompetent mouse model (which differs in genetic background from 4T1). We injected C0321 cells into the mammary fat pads of syngeneic FVB female mice. Upon detection of palpable tumors, mice were treated with SS or vehicle for 14 days. SS significantly delayed tumor growth and reduced tumor mass without altering body mass (Figure 4A). H&E histopathology revealed increased leukocyte infiltration within the tumor microenvironment upon treatment with SS (Figure 4B). SS treatment abrogated Notch1 protein expression in the tumors, but increased Jagged1 expression (Figure 4B). Of note, Jagged1 is also a γ-secretase substrate, and recent observations indicate that its cleaved C-terminal fragment has oncogenic activity mediated by a transcriptional complex containing DDX17, SMAD3, and TGIF (120). Whether inhibition of Jagged1 cleavage contributes to the anti-neoplastic activity of SS in this model deserves further investigation.

Next, we analyzed tumor-infiltrating immune cells by flow cytometry (Figure 4C). SS did not affect the percentage of tumor-infiltrating MDSCs or TAMs. SS significantly reduced the number of CD4 T-cells and, although not statistically significant, increased the number of CD8 T-cells upon SS treatment. In the BALB(c)-derived 4T1 model, sulindac caused a significant increase in infiltrating CD8 T-cells, which were required for anti-tumor activity (71). In our model, the increase in CD8 T-cells was not statistically significant, while the most remarkable effect was a significant increase in the number of antigen-presenting CD11c+ DC. These results suggest that SS may be investigated in combination with immunotherapy.

Next, we performed whole transcriptome RNA-sequencing of control and SS-treated C0321 tumors. We identified 131 differentially expressed genes in SS treated tumors compared to control tumors that passed our FDR and significance thresholds (Figure 5A). IPA analysis revealed the most significant pathways following SS treatment (Figure 5B). Notably, the most significantly affected pathway was Antigen Presentation, supporting our flow cytometric results. Additionally, several other immunologically relevant pathways were modulated by SS (PKCθ, OX40, CTLA4, PDL1/PD1, Nur77, DC maturation) as well as pathways relevant to CSC maintenance (Molecular Mechanisms of Cancer, Hypoxia, PI3K/AKT, Wnt, ERK5), all of which cross-talk with Notch. Overall our results are consistent with the hypothesis that SS has a multi-targeted anti-tumor effect in this model, including Notch inhibition, inhibition of CSCs and immune-stimulatory effects.

Notch inhibitors, including GSIs, are effective in preclinical models of TNBC, where they eliminate CSC resistance to chemotherapy (32, 33, 35, 39–42). However, this strategy has limitations when considering immunotherapy, due to the requirement for Notch signaling in T-cell activation, including CD8 effector T-cells that participate in anti-tumor responses (43–46). Active Notch1 expression renders CD8 T-cells highly resistant to MDSCs and increases their anti-tumor activity (73). Thus, systemic suppression of Notch signaling in TNBC is potentially a double-edged sword; it may successfully target CSCs but may also impair anti-tumor immunity. Therefore, we wanted to determine whether SS affects Notch signaling in T-cells or T-cell proliferation. We isolated T-cells from FVB mouse spleens and performed a T-cell proliferation assay in the presence of increasing concentrations of SS. We found that SS did not significantly affect T-cell proliferation (Figure 6A). Consistent with this result, SS did not affect Notch1 cleavage in T-cells at concentrations up to 50 μM, unlike TNBC cells (Figure 6B). We also found that SS treatment did not alter IL-2 secretion level in T-cells (Figure 6C). Interleukin 2 (IL-2), mainly produced by activated T-cells, plays a central role in controlling the immune response (121). Notch1, activated by APCs carrying Notch ligand DLL4, promotes IL-2 secretion in native CD4 T-cells (122). Altogether, these results indicate that SS at the concentrations tested has no T-cell suppressive activity.

MDSCs promote tumor growth by suppressing cytotoxic T-cell functions (123). The number of MDSCs in the tumor microenvironment was not significantly affected by SS. However, there is evidence that COX inhibition (124) and Notch inhibition by an anti-Jagged monoclonal antibody (73) inhibit the T-cell suppressive activity of MDSCs. Thus, we asked whether SS could inhibit the suppressive functions of MDSCs. We generated bone marrow-derived MDSCs (BM-MDSCs) from FVB mice in the presence of GM-CSF, G-CSF, and IL-6 as described in the Methods section. BM-MDSCs were generated in the presence or absence of SS for four days. We performed T-cell proliferation assays following co-culture with MDSCs treated or untreated with SS. We found that SS significantly blocked the suppressive functions of BM-MDSCs. Importantly, when we added SS during T-cell co-culture with BM-MDSCs, in addition to MDSC maturation, we found significant MDSC activity inhibition even at 5 µM SS concentration (Figure 6D). These results suggest a novel anti-immunosuppressive function of SS against MDSCs, which will be further explored in future studies.

Tumor organoids recapitulate tumor heterogeneity and microenvironment in vitro to enable the of study tumor biology and drug testing (125). The co-existence of tumor cells and immune cells in an intact architecture in tumor organoids makes them suitable three-dimensional tumor culture models (125). We generated organoids from C0321 tumors grown in syngeneic FVB mice as described (126). SS, but not SF, used as a single agent significantly increased tumor cell death in C0321 organoids (Supplementary Figure 4). To trace tumor cells within organoids, we developed mCherry-expressing C0321 tumor cells (C0321-mCherry). We generated C0321-mCherry organoids from tumors formed from C0321-mCherry cells following the same protocol we used for unlabeled C0321 cells. C0321-mCherry organoids were treated with SS in the presence or absence of α-PD1. In pilot experiments, we confirmed that SS and α-PD1 caused tumor cell death as single agents and in combination (Supplementary Figure 5). We then treated C0321 organoids with SS at 2 concentrations (1 and 5 μM) alone and in combination with α-PD1. SS alone induced cell death in a concentration-dependent fashion. Similarly, α-PD1 caused a significant increase in cell death compared to control IgG. However, combinations of SS and α-PD1 had remarkably higher activity than either agent alone (Figures 7A, B). These findings provided a rationale for in vivo testing of SS in combination with α-PD1 immunotherapy.

Based on our organoid results, we tested the effects of SS at a sub-optimal dose in combination with α-PD1 in the TNBC syngeneic C0321 FVB mouse model. After detection of palpable C0321 tumors, mice were randomized to one of four treatment arms: vehicle control, SS alone (20 mg/kg), α-PD1 or SS plus α-PD1 for another two weeks. We measured tumor growth and mouse body mass during that time. Each single agent showed anti-tumor activity, but the combination of SS and α-PD1 significantly reduced tumor growth compared to either SS or α-PD1 alone (Figure 8A). We did not detect significant weight loss or diarrhea during treatment in any of the arms (Figure 8B). At the end of the experiment, we harvested tumors. We found that the combination of SS and α-PD1 significantly reduced tumor mass compared to SS or α-PD1 alone (Figure 8C). H&E staining (Figure 8D) revealed that α-PD1 induced necrosis within tumors. Tumors treated with SS had much larger and extensive areas of necrosis. Combination-treated tumors showed increased inflammatory infiltrates. These results show that SS enhances the response of C0321 TNBC tumors to α-PD1 immunotherapy.