ALDH1A1 Activity in Tumor-Initiating Cells Remodels Myeloid-Derived Suppressor Cells to Promote Breast Cancer Progression

Triple-negative breast cancer (TNBC) has been characterized by several aggressive clinical features including high rates of metastasis, recurrence, poor survival, and high breast tumor-initiating cell (BTIC) population compared with other breast cancer subtypes (1). Due to lack of specific therapeutic targets, it appears to be extremely urgent to find a reliable and effective strategy for TNBC.

TICs, also known cancer stem cells, refer to a subgroup of tumor cells that exist within the tumor mass and are characterized by tumorigenesis, self-renewal, and resistance to chemotherapy. They are associated with tumor initiation, growth, and metastasis (2). Different from other TIC markers, aldehyde dehydrogenase (ALDH) is quite unique as it detects endogenous enzyme activity to characterize the “stemness” of TICs rather than cell-surface molecules, such as CD133⁺ and CD24⁻/CD44⁺. The ALDH family, comprising of 19 isozymes that possess important physiologic and toxicological functions (3), is a superfamily of NADP-dependent enzymes that metabolize endogenous and exogenous aldehydes to carboxylic acids (4, 5). ALDH1A1 was confirmed as a top-ranked hit among most of ALDH isozymes that are responsible for the enzyme activity accessed by ALDH substrate metabolism assay (ALDEFLUOR assay; ref. 6). As the biomarker of TICs in breast cancer, prostate cancer, colon cancer, and lung cancer (7, 8), ALDH1A1 plays an important role in the promotion of tumor angiogenesis and metastases, and the acquisition of anticancer drug resistance (9). Although ALDH1A1 has been reported to regulate retinoic acid (RA) biosynthesis and RA signaling in acute promyelocytic leukemia (APL; ref. 10), all-trans RA is not effective in the treatment of solid tumors as it does in APL, indicating that ALDH1A1, as a metabolic enzyme, may have different molecular mechanisms in regulating solid tumor progression.

Emerging literatures show an enrichment of PD-L1 in tissues from ALDH1A1-expressing TNBC patients and lower expression of antigen processing and presentation proteins (TAP-1 and TAP-2) and costimulatory molecules (CD80) in ALDH⁺ breast cancer cells (11, 12). Immune cells, which interact with cancer cells each other in tumor microenvironment (TME), could directly induce the BTIC phenotype and local immunosuppression to promote tumor development (13). Furthermore, BTICs activate and recruit immunosuppressive cells including myeloid-derived suppressor cells (MDSC) and tumor-associated macrophages (TAM) to repress the immune response (14). MDSCs, the major immunosuppressive cells found in breast cancer (15), are defined as a myeloid-originated heterogeneous population of cells that consist of myeloid progenitors and immature macrophages, immature granulocytes, and immature dendritic cells. Two major subsets of MDSCs have been identified: monocytic (M-MDSC) and polymorphonuclear (PMN-MDSC). However, whether and how the functional BTIC marker ALDH1A1 regulates the immunosuppressive TME to promote breast cancer development remains to be elucidated.

Here, we revealed that ALDH1A1, relying on its enzyme activity, decreased the intracellular pH of breast cancer cells to increase TAK1 phosphorylation and activate NFκB signaling and then stimulate GM-CSF secretion, which induced MDSC expansion, and in turn reduced antitumor immunity to facilitate breast cancer progression. Therapeutically, the ALDH1A1 inhibitor disulfiram (DSF) combined with MDSC-depleting agent gemcitabine (GEM) significantly inhibited breast tumor growth. These findings elucidate the function and molecular mechanisms of ALDH1A1 in modulating the interaction of BTICs and MDSCs to promote breast cancer development, which provided a novel therapeutic strategy for malignant breast cancer.

The human breast cancer cell line MDA-MB-231 and murine 4T1 were purchased from ATCC. MDA-MB-231 and 4T1 breast cancer cells were cultured in RPMI-1640 medium supplemented with 5% fetal bovine serum (FBS, Gibco) and 1% pen–strep antibiotic (Beyotime), and HEK293T cells were cultured in DMEM (Thermo Fisher) medium supplemented with 10% FBS and 1% pen–strep antibiotic. All of the cell lines were tested and authenticated. All cells were cultured at 37°C in a humidified atmosphere containing 5% CO₂. The cell lines were Mycoplasma free and authenticated by PCR analysis monthly.

Four- to five-week-old female Balb/c mice or nude mice were purchased from Charles River and housed in standard animal cages under specific pathogen-free conditions in the Department of Laboratory Animal Science of Fudan University. Animal experiments were approved according to the experimental animal guidelines of the Care and Use of Laboratory Animals of Fudan University and approved by the Fudan University Shanghai Cancer Center Institutional Review Board (JS-082). 5 × 10⁴ 4T1 cells or 1 × 10⁶ MDA-MB-231 cells were inoculated into the fourth mammary gland fat pads. Tumor growth was monitored every three days or once a week, and the tumor volume was calculated as 0.5 × length × width × width.

DSF (Selleck), a small-molecule inhibitor of ALDH1 enzymatic activity, was suspended in special solvent that consisted of 2% dimethyl sulfoxide, 40% polyethylene glycol, 5% Tween-80, and 53% water. When the diameter of tumors reached about 2.0 mm, tumor-bearing mice were treated with DSF (50 mg/kg, 5 mg/mL) or the same volume vehicle orally every 3 days for 27 consecutive days. Gemcitabine (GEM, Selleck), a small-molecule inhibitor of DNA synthesis, was suspended in water. For combinational therapy, DSF treatment was administered every three days and an intraperitoneal injection of gemcitabine HCl (50 mg/kg once a week, a total of 2 times) was administered for 2 weeks.

For the combination DSF with PD-L1 antibody, PD-L1 antibody (200 μg/mouse every 2–3 days, clone 10F.9G2, Bio X Cell) or isotype control antibody (200 μg/mouse every 2–3 days, IgG, Sigma-Aldrich) alone or combined with DSF (50 mg/kg, every 2–3 days) were administered for 2 weeks started from day 9.

Single-cell suspensions were prepared from spleen and tumor tissues. Tumor cell suspensions from xenografted tumors were analyzed as previously described. When analyzing tumor cells from xenograft of human MDA-MB-231 model or patient-derived xenograft (PDX), mouse cells were excluded by gating in flow cytometry using H-2Kd-PE. But in 4T1 graft, PE-conjugated anti-mouse lineage antibodies were used for CD45, CD31, CD140b, and CD235a. Single cells from tumors were assessed for their ALDH activity using the ALDEFLUOR Kit (STEMCELL Technologies) following the manufacturer's procedures. When immune cells were analyzed, single-cell suspensions were firstly incubated with anti-mouse CD16/32 (BioLegend). To identify the percentage of MDSCs, the cells were stained with CD45, CD11b, and Gr-1. To analyze the subpopulations of T cells, the cells were labeled with CD45, CD3, CD4, and CD8. To assess the active CD8⁺ T cells, the cells were stained with IFNγ and TNFα. The detailed information of antibodies was shown in Supplementary Table S1. The stained cells were analyzed using a MoFlo Astrios instrument (Beckman Coulter), and data acquisition and analysis were performed using Summit software.

Human ALDH1A1 and mouse Aldh1a1 were amplified with the reverse-transcribed cDNA from MDA-MB-468 and 4T1 cell lines and cloned into the lentiviral vector pSIN-Flag (puromycin-resistant; Addgene), and the authenticity was verified by sequencing, respectively. Site mutation was introduced by primer design to construct ALDH1A1 or Aldh1a1-overexpressing plasmids with different ALDH enzymatic activity. Short hairpin RNA (shRNA) sequences of Aldh1a1, TAK1, and p65 were purchased from Sigma-Aldrich and cloned into the lentiviral vector pLKO.1 (Addgene). All the primers used for plasmid construction were listed in Supplementary Table S2. A highly efficient lentiviral system was used to generate the viruses. The cell lines were infected with the lentiviruses, and the stable cell lines were established. The lentiviral transfection efficiency was more than 90% in all cell lines.

Total RNA was extracted with TRIzol, and the RNA concentration was measured with Nanodrop2000 (Thermo Scientific). Real-time PCR was described previously. TBP was used for mRNA normalization, and the primers used are listed in Supplementary Table S3. For RNA sequencing (RNA-seq), strand-specific RNA-seq libraries were prepared using the NEB Next Ultra Directional RNA Library Prep kit for Illumina (New England Biolabs), and subjected to quality control using a Bioanalyzer 2100 (Agilent) and were sequenced using a HiSeq 3000 (Illumina). Unsupervised clustering was performed using cluster and tree view, and the data were transformed into a heat map. Enrichment pathway analysis of genes was compiled from the gene set enrichment analysis (GSEA) databases.

CD11b⁺Gr-1⁺ MDSCs were sorted from the 4T1 allografts with Aldh1a1-knockdown, Aldh1a1 or K193Q/R-mutant overexpression by using a MoFlo Astrios (Beckman Coulter). CD8⁺ T cells were sorted from the spleens of wild-type Balb/c. CD8⁺ T cells were stimulated by cell stimulation cocktail plus protein transport inhibitor (eBioscience) in a 1.5mL tube at 37°C in 5% CO₂. Activated CD8⁺ T cells (20 × 10⁴) were labeled with CSFE (2 mmol/L, Beyotime) and cocultured with sorted MDSC (20 × 10⁴) in vitro in a 96-well round plate. After three days, cells were harvested and stained with anti-CD8, anti-IFNγ and anti-TNFα (Supplementary Table S1) for assessment the status of CD8⁺ T-cell activation. CD8⁺ T-cell proliferation was determined by measuring the dilution of CSFE with flow cytometry.

BCECF (2′,7′‐bis‐(2‐carboxyethyl) ‐5‐(and‐6) ‐carboxyfluorescein) AM is the most widely used fluorescent indicator for intracellular pH. Cancer cells were seeded in six-well plates and incubated under normal conditions till approximately 70% to 80% confluency. Prepare viable cells in suspension (∼10⁶ cells/mL). Dilute an aliquot of 1 mmol/L BCECF-AM ester stock solution 100‐ to 500‐fold into a physiologic saline buffer such as PBS. Add one volume of aqueous AM ester dispersion to one volume of cell suspension. Incubate for 15 to 60 minutes at 4°C to 37°C. Wash the cells twice with fresh culture medium and then detect the fluorescence at 488 nm by flow cytometry. All experiments were repeated more than 3 times.

Bar graphs were generated with GraphPad Prism 8.0, and all values are reported as the mean ± SEM. A one-way or two-way ANOVA was used for multiple comparisons. Unless otherwise indicated, comparisons between two groups were performed using an unpaired, two-tailed t test. P values of less than 0.05 were considered statistically significant (*, P < 0.05; **, P < 0.01; ***, P < 0.001).

All the data acquired and/or used in the study are available to readers. The reader could get data from corresponding authors.

Our previous research indicated that ALDH isoform expression exhibited a cancer type–specific pattern in different cancers (16, 17). ALDH1A1, as a functional biomarker for BTICs (16, 17), is generally believed to be predominantly responsible for the ALDH enzyme activity of BTICs (18). We also confirmed that ALDH1A1 expression was higher than other ALDH1 isoforms (Supplementary Fig. S1A), and Aldh1a1 knockdown could significantly decrease the ALDH enzyme activity in mouse breast cancer cells 4T1 (Supplementary Fig. S1B). However, it is not clear whether the functions of ALDH1A1 on BTICs and breast cancer development are dependent on its enzyme activity.

Database analysis showed that the K193 residue of ALDH1A1 was highly conserved among ALDH family members except for ALDH16A1 and ALDH18A1. K193 was a vital activity site that could form the chemical bond (hydrogen bond) with NADP, which is a key step for ALDH enzyme activity. To further investigate ALDH1A1 effect on breast cancer development, relying on its enzyme activity but not its expression, we stably overexpressed ALDH1A1 or mutants (K193Q/R) in 4T1 (endogenous Aldh1a1 had been stably knocked down) and human breast cancer cell line MDA-MB-231, which have very low endogenous level of ALDH1A1 and few ALDH⁺ cells (17), via constructing lentiviral vectors with mutated K to Q/R at 193 residue of ALDH1A1 to block its enzyme activity but not the expression (Fig. 1A). Flow cytometry analysis showed that ALDH enzyme activity was extremely increased in the ALDH1A1 group, but not in the K193Q/R-mutant groups (Fig. 1B–D). Based on ALDH enzyme activity, we sorted ALDH⁻ and ALDH⁺ cells from ALDH1A1-overexpressing breast cancer cells, and confirmed that ALDH⁺ cells were more tumorigenic than ALDH⁻ cells (Supplementary Fig. S1C and S1D), but the protein and mRNA level of ALDH1A1 showed no significant difference between two populations (Supplementary Fig. S1E–S1H), which indicated that the difference shown in tumorigenesis ability between ALDH^+/− cells did not rely on the ALDH1A1 expression. We also examined the effects of ALDH1A1 and K193Q/R mutants on BTIC self-renewal and tumorigenesis, and found only ALDH1A1 promoted BTIC self-renewal accessed by mammosphere formation assay (Fig. 1E–G). Moreover, the limited dilution assay (LDA) showed that only ALDH1A1 promoted tumor growth and the BTIC frequency (Fig. 1H; Supplementary Fig. S1I), and the ALDH⁺ BTIC population in tumors was increased in ALDH1A1 tumors compared with K193Q/R mutants (Fig. 1I). These results strongly implied that ALDH1A1 might promote breast cancer development, relying on the enzyme activity, and the activity of ALDH1A1 is partially influenced by some key modifications at the posttranslational level.

Next, we verified effects of ALDH1A1 enzyme activity on breast cancer with its inhibitors. Numerous ALDH inhibitors have been developed in the fight against cancer, including N, N-diethylamino benzaldehyde (DEAB), DIMATE, DSF, CM026, CM037, NCT‐501, CVT‐10216, ALDH423, and CB29 (19). Although there is no specific and effective inhibitor of ALDH1A1, DEAB is a reversible, competitive inhibitor of cytosolic ALDH (ALDH1), more potent than mitochondrial ALDH (ALDH2), especially an excellent substrate for ALDH1A1 (20), and DSF can be metabolized to diethyldithiocarbamate by CYP450 and which acts as an irreversible inhibitor to inhibit ALDH1A1 more potently than ALDH2 in vivo (21), and has been used to treat alcoholism in the clinic. We found DEAB inhibited in vitro ALDH enzyme activity in a dose-dependent manner (Supplementary Fig. S1J and S1K) and mainly suppressed BTIC self-renewal in the ALDH1A1 group (Fig. 1J). We also administered DSF to inhibit ALDH activity in mice bearing ALDH1A1 4T1 tumors and found that DSF repressed tumor growth in the control and ALDH1A1 4T1 groups by decreasing ALDH⁺ cells, which was more remarkable in the ALDH1A1 group (Fig. 1K and L). All these indicated that ALDH1A1 increased ALDH⁺ BTICs and promoted tumor growth relying on its enzyme activity.

To explore the effect of ALDH1A1 on immune microenvironment, we constructed orthotopic allografts with knockdown endogenous ALDH1A1 4T1 in Balb/c mice, and analyzed the alterations of immunosuppressive cells by FACS (Fig. 2A; Supplementary S2A). As expected, knockdown of Aldh1a1 reduced tumor growth (Fig. 2B). And we observed that the myeloid cells (CD11b⁺) were markedly reduced (Supplementary Fig. S2B), which mainly attributed to the decrease of CD11b⁺Gr1⁺ MDSCs (Fig. 2C), but not the change of CD11b⁺F4/80⁺ TAMs (Supplementary Fig. S2C). We also analyzed PMN-MDSCs (CD11b⁺LyG⁺LyC^low) and M-MDSCs (CD11b⁺LyG⁻LyC^high), and found that most MDSCs were PMN-MDSCs and the two subtypes were decreased in both tumors and host spleens (Supplementary Fig. S2D and S2E). The definitive characteristic of MDSCs is potent to suppress various T-cell functions (22). We proved that Aldh1a1 knockdown in 4T1 cells reduced the mRNA expression levels of MDSC immunosuppressive genes such as CYBB1, NCF, PD-L1, S100A8, and S100A9 in MDSCs sorted from 4T1-shAldh1a1 allografts (Supplementary S2F). Moreover, the results of CSFE cell proliferation assays showed that the inhibition capability of sorted MDSCs from 4T1-shAldh1a1 allografts on T-cell proliferation was markedly attenuated (Supplementary S2G). It was also reported that MDSC inhibited the secretion of inflammatory cytokines of CD8⁺ T cells, such as IFNγ and TNFα, to elicit an adaptive immune response (23). So, we analyzed the enrichment of active CD8⁺ T cells in mice bearing 4T1 tumors, and found that IFNγ⁺CD8⁺ and TNFα⁺CD8⁺ T cells were augmented after Aldh1a1 knockdown in tumors and host spleens (Fig. 2D and E). Next, the capability of the sorted MDSCs from Aldh1a1-knockdown allografts on suppressing CD8⁺ T-cell activation was also evaluated in vitro (Supplementary S2H). These results demonstrated that Aldh1a1 promoted MDSC enrichment and suppressed T-cell immunity in breast tumors.

In order to investigate whether ALDH1A1 enzyme activity was necessary for its effects on tumor immunity, we constructed orthotopic allografts with stably overexpressed Aldh1a1 or mutants (K193Q/R) 4T1 (endogenous Aldh1a1 had been stably knocked down) in Balb/c mice. Aldh1a1 promoted tumor growth compared with K193Q/R mutants (Fig. 2F). Moreover, compared with control, MDSCs were increased only in the Aldh1a1 group not in K193Q/R mutants (Fig. 2G). IFNγ⁺CD8⁺ and TNFα⁺CD8⁺ T cells had been decreased only in the Aldh1a1 group (Fig. 2H and I). We also confirmed that MDSCs derived from 4T1-Aldh1a1 tumors have higher expression level of immunosuppressive genes (Supplementary Fig. S2I) and more robust capability on inhibiting the proliferation and activation of active CD8⁺ T cells than those from K193Q/R-mutant tumors (Supplementary Fig. S2J and S2K). Further, we inoculated ALDH⁻ or ALDH⁺ 4T1 cells into Balb/c mice and treated the mice with DSF (Supplementary Fig. S3A). We found that ALDH⁺ cells significantly accelerated tumor growth compared with the ALDH⁻ cells, which was blocked by DSF (Supplementary Fig S3B). FACS analysis showed that DSF reduced the ALDH⁺ BTIC enrichment (Supplementary Fig. S3C). MDSCs were remarkably increased in ALDH⁺ cell–derived tumors and host spleens compared with ALDH⁻ ones, which was also blocked by DSF (Supplementary Fig. S3D and S3E). The infiltration of IFNγ⁺CD8⁺ and TNFα⁺CD8⁺ T cells had an opposite change in tumors and host spleens (Supplementary Fig. S3F–S3H). Similarly, DSF blocked the MDSC enrichment and reversed the decrease of active CD8⁺ T cells in control and Aldh1a1 4T1 tumors and host spleens (Fig. 2J–L). These results indicated that Aldh1a1 enriched MDSC and suppressed T-cell immunity depending on its enzyme activity.

To determine whether ALDH1A1 induced MDSC enrichment to promote tumor growth, we administered anti-Ly6G antibody to deplete MDSC in mice bearing tumors as scheduled in Fig. 2M. The results showed that anti-Ly6G antibody decreased tumor growth in the control group but more remarkably in the Alah1a1 group (Fig. 2N and O). As expected, anti-Ly6G antibody depleted MDSCs and recovered the infiltration of IFNγ⁺ CD8⁺ and TNFα⁺ CD8⁺ T cells in tumors and spleens (Fig. 2P–R). These data suggested that Aldh1a1-stimulated MDSC enrichment relying on enzyme activity, and in turn, MDSCs increased immunosuppression to promote tumor growth.

We next dissected the molecular mechanisms for ALDH1A1 mediating MDSC increase. Firstly, we intraperitoneally injected the cell culture supernatant (conditioned medium, CM) derived from 4T1 into female Balb/c mice (Supplementary Fig. S4A), and results of FACS showed that the total percentage of MDSCs was dramatically increased in spleens (Supplementary Fig. S4B and Fig. 3A), peripheral blood (Fig. 3B), and bone marrow (Fig. 3C) of the Aldh1a1 group compared with the K193Q/R-mutant. These results indicated that some cytokines secreted by Aldh1a1 4T1 might enhance MDSC expansion. To confirm this hypothesis, we analyzed the transcriptional profiles of ALDH1A1 and K193Q/R-mutant MDA-MB-231. Compared with control, many cytokines were elevated in the ALDH1A1 group but not in the K193Q/R-mutant groups; GM-CSF and G-CSF were among the most upregulated ones (Fig. 3D). It has been reported that GM-CSF and G-CSF regulate MDSC development in cancers (24, 25). We speculated that ALDH1A1 might expand MDSC by upregulating GM-CSF and G-CSF. To validate this hypothesis, we initially analyzed the correlation between the mRNA expression of ALDH1A1, GM-CSF, G-CSF, and human MDSC marker CD33 in the TCGA data set (26), and found the mRNA expression of ALDH1A1, CD33, and GM-CSF was positively correlated with each other (Supplementary Fig. S4C–S4E), but G-CSF showed no significant correlations with either CD33 or ALDH1A1 (Supplementary Fig. S4F and S4G). Then, we confirmed that ALDH1A1 increased GM-CSF at both the mRNA (Fig. 3E; Supplementary Fig. S4H) and protein (Fig. 3F; Supplementary Fig. S4I) levels in 4T1 and MDA-MB-231, and Aldh1a1 knockdown had the opposite effect (Supplementary Fig. S4J and S4K). Similarly, GM-CSF mRNA was significantly upregulated in ALDH⁺ 4T1 cells compared with ALDH⁻ cells (Fig. 3G). In addition, we utilized DEAB to inhibit ALDH enzyme activity in vitro and observed that DEAB abolished the upregulation of GM-CSF at mRNA and protein levels in Aldh1a1 4T1 (Supplementary Fig. S4L and Fig. 3H). These studies suggested that ALDH1A1 enzyme activity was responsible for ALDH1A1-mediated GM-CSF upregulation in breast cancer cells.

Next, we investigated whether ALDH1A1 upregulated GM-CSF to promote MDSC expansion. We conducted a noncontact coculture system using 4T1 cells and bone marrow cells (BMC) derived from female Balb/c mice (Fig. 3I) and, after 6 days, analyzed the percentage of CD11b⁺Gr1⁺ MDSCs (Fig. 3J). We observed that MDSCs were expanded in BMCs when being cocultured with Aldh1a1 4T1 (Fig. 3K). A similar effect was observed in BMCs cocultured with ALDH⁺ 4T1 (Fig. 3L). Moreover, we observed DEAB reduced the MDSC development in the Aldh1a1 group (Fig. 3M). We knocked down GM-CSF in Aldh1a1 4T1 (Supplementary Fig. S4M and S4N) and found that the MDSC expansion was blocked by GM-CSF knockdown in cocultured 4T1 cells (Fig. 3N). These results suggested that ALDH1A1 increased GM-CSF secretion and then promoted MDSC expansion relying on its enzyme activity.

To explore how active ALDH1A1 upregulated GM-CSF expression in breast cancer cells, we analyzed the transcriptional profiles of ALDH1A1 and K193Q/R-mutant MDA-MB-231. GSEA indicated that many signaling pathways were deregulated, especially the NFκB pathway, which was positively correlated with ALDH1A1 enzyme activity (Fig. 4A), and downstream genes of the NFκB pathway were enriched in the ALDH1A1 group versus others (Fig. 4B). Previous reports have demonstrated the role of NFκB signaling in regulating GM-CSF expression (27, 28). We then investigated whether ALDH1A1 activated the NFκB pathway and found that overexpression of ALDH1A1, but not K193Q/R mutants, effectively increased the phosphorylation of p65 and IKBα (p-p65 and p-IKBα) in both MDA-MB-231 and 4T1 as shown by Western blots (Fig. 4C and D). Correspondingly, p-p65 and p-IKBα were decreased after the Aldh1a1 knockdown in 4T1 cells (Fig. 4E). Meanwhile, p-p65 and p-IKBα were also upregulated in ALDH⁺ 4T1 cells compared with ALDH⁻ cells (Fig. 4F). These studies indicated that ALDH1A1 enzyme activity might be responsible for NFκB pathway activation. In support of this notion, we utilized DEAB to inhibit ALDH1A1 enzyme activity in vitro and found that DEAB suppressed NFκB signaling in a time-dependent manner (Fig. 4G) and reversed NFκB pathway activation in the ALDH1A1 group (Fig. 4H). Furthermore, p65 knockdown with specific shRNAs (Supplementary Fig. S5A) blocked the upregulation of GM-CSF at both mRNA and protein levels in ALDH1A1 MDA-MB-231 (Supplementary Fig. S5B; Fig. 4I). Thus, active ALDH1A1 upregulated GM-CSF via activating the NFκB signaling.

Furthermore, to explore how ALDH1A1 activate NFκB signaling, we performed the Human Phospho-Kinase Array (Fig. 5A) and found that several phosphokinases, including p-p38, p-ERK1/2, and p-JNK, were markedly increased in the ALDH1A1 group (Fig. 5B). Literature showed that the most extensively studied groups of vertebrate MAPKs to date are ERK1/2, JNKs, and p38 kinases (29). Several pieces of evidence also indicate that MAPKs can participate in regulating NFκB pathway activity (30). More recently, TGFβ-activated kinase 1 (TAK1, also known as MAP3K7) is found to be a key serine/threonine protein kinase of the MAP3K family and has been described as a regulator of NFκB and MAPKs in proinflammatory signaling (31). We conjectured that ALDH1A1 might activate NFκB signaling through TAK1-activated MAPK signaling, and sequentially upregulate GM-CSF expression. Western blots revealed that p-TAK1 was distinctly increased in ALDH1A1 group, but not K193Q/R-mutant groups (Fig. 5C and D). p-TAK1 was also increased significantly in ALDH⁺ 4T1 cells compared with ALDH⁻ cells (Fig. 5E). We also confirmed that p-JNK, p-ERK, and p-p38 were consistently increased in the MDA-MB-231 and 4T1 with high ALDH enzyme activity (Supplementary Fig. S5C–S5E). Meanwhile, TAK1–MAPK signaling was suppressed by the knockdown of Aldh1a1 in 4T1 cells (Supplementary Fig. S5F). Furthermore, we found that inhibition of ALDH1A1 enzyme activity by DEAB suppressed p-TAK1 in ALDH1A1 MDA-MB-231 cells (Fig. 5F). Also, the levels of p-JNK, p-ERK, and p-p38 were decreased by DEAB treatment (Supplementary Fig. S5G). These studies suggested that ALDH1A1 activated TAK1 pathway relying on its enzyme activity.

Next, we investigated whether ALDH1A1 activated NFκB signaling via TAK1. We genetically knocked down TAK1 with specific shRNAs (Supplementary Fig. S5H) and observed that TAK1 knockdown reversed the increase of p-p65 and p-IKBα (Fig. 5G), and both mRNA (Supplementary Fig. S5I) and protein levels of GM-CSF (Fig. 5H) were also significantly decreased by TAK1 knockdown in the ALDH1A1 group. Moreover, p-JNK, p-ERK, and p-p38 levels were also reduced by TAK1 knockdown in the ALDH1A1 group (Supplementary Fig. S5J). In conclusion, these studies suggested that ALDH1A1 upregulated GM-CSF via activating the TAK1–NFκB signaling pathway.

Functionally, p65 or TAK1 knockdown partially reversed the promoting effects of ALDH1A1 on breast cancer cell proliferation in vitro (Supplementary Fig. S5K and S5L), breast tumor growth and GM-CSF secretion in vivo (Fig. 5I and J). In addition, p65 or TAK1 knockdown also reversed the increase of ALDH⁺ BTICs (Fig. 5K) and MDSC enrichment in ALDH1A1 tumors and host spleens (Fig. 5L and M).

All of these studies showed that ALDH1A1 activated the TAK1-NFκB pathway to increase GM-CSF secretion, and subsequently led to the increase of BTICs, enhanced MDSC infiltration and breast cancer progression.

ALDH enzymes irreversibly catalyze the oxidation of both endogenously and exogenously produced aldehydes to their respective carboxylic acids, and then carboxylic acids dissociate to form carboxylate ion and hydrogen ion (Fig. 6A). Utilizing BCECF-AM fluorescent probe to detect the pHi, we found that Aldh1a1-knockdown increased the pHi of 4T1 cells (Supplementary Fig. S6A and S6B), and ALDH1A1 overexpression decreased the pHi in MDA-MB-231 and 4T1 (Fig. 6B and C). Moreover, the pHi was lower in the 4T1 ALDH⁺ cells compared with ALDH⁻ cells (Fig. 6D). Notably, the majority of upregulated differential intracellular metabolites detected by the HPLC-MS assay were organic acids in the ALDH1A1 group but not in K193Q/R mutants versus control (Fig. 6E), which might contribute to the lower pHi in active ALDH1A1 cells. These results demonstrated that ALDH1A1 indeed, relying on enzyme activity, reduced the pHi in breast cancer cells. Enhancing sensitivity of pH-weighted (pH_enh) can be a useful tool for noninvasive pH imaging detected by MRI (32). We evaluated tumor pH by pH_enh imaging in ALDH1A1 or K193Q/R-mutant 4T1 allografted tumors in vivo. The mean value of pH_enh in the Aldha1a group was less than that in control or mutant groups, but it was only significant between Aldh1a1 and K193R mutants (Supplementary Fig. S6C and S6D), which might be due to the tumor heterogeneity and the limited sensitivity of current in vivo pH_enh imaging.

Next, we explored whether low pHi could activate the TAK1–NFκB pathway in vitro. Firstly, we treated MDA-MB-231 with hydrochloric acid (HCl) or sodium hydroxide (NaOH) to adjust extracellular pH (pHe) in a proper pH range (pHe = 6.50–8.30), which also correspondingly adjusted the pHi (Fig. 6F and G). p-TAK1, p-p65, and p-IKBα were all increased in MDA-MB-231 and 4T1 under acidic conditions (Fig. 6H; Supplementary Fig. S6E). Because the research showed that phosphorylation-dependent activation of TAK1 is activated by its specific activator, TAK1-binding protein 1 (TAB1; ref. 33), we performed the in vitro phosphorylation assay with purified TAK1 and TAB1 protein. The results showed that the phosphorylation level of the purified TAK1 protein was dependent on the proper range of pH values in the reaction system (Fig. 6I), indicating a direct effect of pHi on TAK1 activation. Furthermore, we found the GM-CSF mRNA and protein levels were upregulated under acidic conditions in MDA-MB-231 (Fig. 6J and K) and 4T1 (Supplementary Fig. S6F and S6G). Our data indicated that pHi could directly regulate TAK1 activation. These studies indicated that ALDH1A1 decreased the pHi to increase the phosphorylation of TAK1 and activate the TAK1–NFκB pathway, and then enhanced GM-CSF secretion.

Collectively, we observed that ALDH1A1 enzyme activity played critical roles in regulating tumor immune microenvironment to promote breast cancer progression. GEM, a clinically used chemotherapeutic agent to treat metastatic breast cancer, is also reported to enhance antitumor immune responses through its capacity to eliminate MDSCs or trigger “immunogenic cell death” of the tumor cell (34, 35). So, we designed a novel strategy combining DSF with GEM to target the BTIC population and eliminate MDSCs (Fig. 7A). The 4T1 tumor growth and ALDH⁺ BTICs were more inhibited by the combination of DSF and GEM than by either treatment alone (Fig. 7B–D). As expected, GEM efficiently eliminated MDSCs (Fig. 7E and F) and increased infiltrated TNFα⁺CD8⁺ and IFNγ⁺CD8⁺ T cells (Fig. 7G and H), which were reinforced by the combination with DSF. In order to confirm the inhibition on BTICs, we performed a secondary tumorigenesis with an LDA, and found that the combination of DSF and GEM decreased the BTICs more efficiently than any one alone (Fig. 7I). In addition, ALDH signaling seems to be increasingly important for Treg induction and function through the production of RA by multiple cell types (e.g., DCs, macrophages, eosinophils, epithelial cells; ref. 36). In order to evaluate the DSF inhibition on ALDH1A1 of cancer cells more clearly, we treated a TNBC PDX model (USTC11) with DSF and GEM in immunodeficient NOD-SCID mice (Supplementary Fig. S7A), and found that either DSF or GEM inhibited the PDX tumor growth and ALDH⁺ BTICs, but it was not enhanced by the combination, which might be due to the reduced immune cells especially T cells in immunodeficient mice (Supplementary Fig. S7B–S7D).

According to the results of the combination strategy, DSF may play vital roles not only in the inhibition on BTICs but also in the increases of IFNγ and TNFα production of CD8⁺ T cells, indicating the dual effects of DSF on tumors in vivo. Moreover, due to high level of PD-L1 in both tumor cells of TNBC (37) and myeloid cells including myeloid suppressors cells, macrophages, and myeloid dendritic (38), we designed a novel therapy strategy combining DSF and PD-L1 antibody in 4T1 tumor-bearing mice. The tumor volume and tumor weight were obviously reduced in the combination group than either DSF-alone or PD-L1–alone group (Supplementary S7E and S7F). Interestingly, we found that DSF alone or combined with PD-L1 antibody efficiently inhibited the enrichment of MDSCs (Supplementary S7G and S7H) and the significantly promoted the infiltration of TNFα⁺CD8⁺ T cells and IFNγ⁺CD8⁺ T cells (Supplementary S7I and S7J). These results indicated that DSF could enhance the T-cell immunity response and then enhance the treatment effects of PD-L1 antibody. All these results provided a promising idea for developing the new strategy to target both BTICs and the tumor immune microenvironment in malignant cancers including TNBC.

ALDH activity is a hallmark of BTICs and can be identified by the aldefluor assay. ALDH1A1, as a biomarker for predicting poor survival of breast cancer patients, is generally believed to be predominantly responsible for the ALDH enzyme activity of BTICs (18). Apart from ALDH1A1, ALDH1A2 (39), ALDH1A3 (40), ALDH2 (39, 41) and other isoforms also showed the enzyme activity. Of note, it was also shown that ALDH1A3 was the predominant ALDH isozyme responsible for ALDH activity and tumorigenicity in most non–small cell lung cancer (42). Due to different cellular context and cancer-context-specificity, ALDH1A3 expression pattern has a clear discrepancy in the same or different cancers. It has been proved that several ALDH isoforms other than ALDH1A1 were associated with certain cancers, e.g., ALDH3A1 in lung cancer (43) and prostate cancer (44) and ALDH5A1 in breast cancer (45). On the other hand, ALDH1A2 (46) and ALDH2 (47) have been reported to act as tumor suppressors in prostate cancer and liver cancer, respectively. As discussed above, the expression of ALDH isoforms revealed cancer-type–specific expression patterns.

In the current study, we sorted ALDH⁻ and ALDH⁺ cells from ALDH1A1 or K193Q/R-mutant-overexpressing breast cancer cells and confirmed that ALDH⁺ breast cancer cells were more tumorigenic than ALDH⁻ ones. Although the endogenous ALDH1A1 expression is higher in the sorted ALDH⁺ versus ALDH⁻ parental cells, we did not see any difference of ALDH1A1 expression in the sorted ALDH⁺ versus ALDH⁻ breast cancer cells from the stable overexpressing breast cancer cells. These results strongly indicated that ALDH1A1 might promote breast cancer development only relying on the enzyme activity, and the activity of ALDH1A1 is partially regulated at posttranslational level, where some key modifications may influence the activity of ALDH1A1. Recently, it has been reported that the activity of ALDH1A1 is regulated by phosphorylation and phosphorylation acetylation (48). Furthermore, we could not rule out the possibility that overexpression of the K193 Q/R-mutant might have other effects independent of its enzyme activity.

Solid tumors are characterized by a highly acidic microenvironment, and tumor acidity enhances the function of immunosuppressive cells, such as MDSCs and Tregs, and polarizes macrophages to the immunosuppressive phenotype (49–51). Specifically, CD8⁺ T cells tend to become anergic and apoptotic when exposed to a low pH TME (52). Apparently, the intracellular acidification is mainly regulated by Warburg effect, we think ALDH1A1 and other ALDH isoforms also contribute to this progression, which may work by changing intracellular pH locally but not affecting the acid–base homeostasis of cancer cells. Quentin Liu and colleagues demonstrated that lactate dehydrogenase A (LDHA) generated lactate and enhanced the USP28/MYC signaling to promote breast cancer stem-like cells via maintaining a local acidic microenvironment (53). It was almost a foregone that ALDH1A1 did significantly reduce the pHi to activate the phosphorylation of TAK1 in breast cancer cells. However, so far how pHi regulates phosphorylation of TAK1 remains unclear. Nonetheless, we provided important insights into ALDH1A1 in the regulation of tumor MDSCs and tumor immunity.

MDSC-mediated immune suppression is a major mechanism responsible for immune evasion in breast cancer. However, immune microenvironment is very complex, and a slight change of any one part may affect the situation as a whole. In order to better understand the effect of ALDH1A1 on immune response, we also analyzed changes of other immune cells except for MDSCs. Apart from the significant increase of CD11b⁺ myeloid cells (the main proportion of CD45⁺ cells), MDSCs and its subtypes, the proportions of other immune cells such as CD11b⁺F4/80⁺ TAMs, total CD11c⁺ cells, CD11c⁺MHCII⁺ DCs and total CD19⁺ B cells were changed more or less in Aldh1a1 tumors compared with K193Q/R-mutant tumors (Supplementary Fig. S8A–S8H). The regulation of ALDH1A1 enzyme activity in tumor cells on the other immune cell components needs further exploration in future studies.

Py8119 is an additional mouse basal-like breast cancer cell line. As shown in the article (54), aerobic glycolysis controlled myeloid-derived suppressor cells and tumor immunity via a specific CEBPB isoform in TNBC. The authors found that LDHA knockdown in 4T1 cells regulated the infiltration of MDSCs and CD8⁺ T cells using 4T1 allograft model in Balb/c mice, and the most of experimental results were consistent in the Py8119 allografts in C57/B6 mice, which indicated 4T1 and Py8119 had similar properties on inducing tumor immunosuppression. Similar to the article (55), we performed the combination strategy only in 4T1 allografts. In order to evaluate the efficiency of this combination strategy in a more clinic-relevant model, we treated a PDX USTC11 using DSF or GEM. The combination DSF with GEM had a significant additive inhibitory effect on 4T1 allografts in immunocompetent Balb/c mice; however, this inhibitory effect on PDX tumor growth and ALDH⁺ BTICs was not enhanced by the combination in immunodeficient NOD-SCID mice. Notably, DSF or GEM alone decreased the ALDH⁺ BTICs, indicating MDSCs in tumors might help to maintain BTICs as previously reported. About the unsetting dissociation between the effects on tumor progression in different tumor models, it may not only attribute to the competence or incompetence of the immune system between normal and NOD-SCID mice, but also ascribe to the significant discrepancy of tumor growth and different ALDH enzyme activity underline different cancer cells.

Overall, our current studies showed that ALDH1A1, solely relying on its enzyme activity, decreased the pHi of breast cancer cells to activate the TAK1–NFκB signaling pathway leading to the increased GM-CSF secretion, which in turn induced MDSC expansion and reduced antitumor immunity to facilitate the breast cancer progression. Our findings help to elucidate a novel mechanism of ALDH⁺ BTICs cross-interacting with TME and provide a novel therapeutic strategy to treat malignant breast cancer by targeting both BTICs and MDSCs.

No disclosures were reported.

C. Liu: Conceptualization, resources, data curation, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. J. Qiang: Validation, investigation, methodology. Q. Deng: Validation, investigation, visualization, methodology. J. Xia: Validation, investigation, methodology. L. Deng: Methodology. L. Zhou: Methodology. D. Wang: Conceptualization, methodology. X. He: Investigation, visualization, methodology. Y. Liu: Visualization, methodology. B. Zhao: Visualization, methodology. J. Lv: Resources. Z. Yu: Supervision. Q. Lei: Supervision. Z. Shao: Resources, supervision. X. Zhang: Resources, supervision. L. Zhang: Conceptualization, supervision, funding acquisition, writing–original draft, writing–review and editing. S. Liu: Conceptualization, supervision, funding acquisition, writing–review and editing.

Thank you very much for the support of the innovative research team of high-level local university in Shanghai. This research was supported by the National Key Research and Development Program of China (2018YFA0507501, 2020YFA0112300, and 2016YFA0101202), NSFC grants (81773155, 81930075, 81772799, 82073267, 81530075, and 81873893), the Program of Shanghai Academic/Technology Research Leader (20XD1400700), “Ten Thousand Plan”- National High-Level Talents Special Support Plan (WR-YK5202101), Program for Outstanding Medical Academic Leader in Shanghai (2019LJ04), Shenzhen Science and Technology Innovation Commission Project, Shenzhen Municipal Government of China (KQTD20170810160226082), the Fudan University Research Foundation (IDH1340042), the Research Foundation of the Fudan University Shanghai Cancer Center (YJRC1603), and the Shanghai Municipal Science and Technology Major Project (2018SHZX01).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.