Identification of ALDH1A3 as a Viable Therapeutic Target in Breast Cancer Metastasis–Initiating Cells

Globally, breast cancer is the most frequently diagnosed cancer and the leading cause of cancer mortality in women; ultimately approximately 12% of women within the United States will be diagnosed over the course of their lifetimes (1). Breast cancer brain metastases (BCBM) represent a late stage in breast cancer progression; current therapeutic options demonstrate limited efficacy resulting in overall survival (OS) on the order of months (1, 2). The incidence of BCBM continues to increase, due in part to improved management of primary breast cancer and subsequent prolongation of survival (1, 2). Unfortunately, recently developed targeted therapies that control primary disease in breast cancer have diminished efficacy against metastatic brain lesions (1, 2). It is also prudent to note that disparities exist in the prognostication of breast cancer with regard to tumor receptor positivity [i.e., HER2⁺, estrogen (ER⁺), and progesterone (PR⁺); ref. 1]. Inherently, triple-negative breast cancer (TNBC) more frequently progresses to BM as compared with other breast cancer subtypes and is associated with an extremely poor prognosis following BCBM establishment (3–5).

Although the precise governing biology of BCBM remains to be fully elucidated, previous studies have demonstrated the existence of molecular signatures distinct to tumors which have metastasized to and ultimately colonized the brain (6–9). However, these advances have failed to translate to actionable biomarkers and/or molecular targeted therapies for BCBM (10). The poor prognosis also relates to BCBM cell's aggressive proliferation compared with breast cancer cells that colonize other organs (e.g., lung, bone, etc.; refs. 11, 12), suggesting that breast cancer cells acquire brain-primed properties potentially via bidirectional interplay with the blood–brain barrier (BBB) and/or brain microenvironment. Previous genomic characterization of BM has demonstrated that metastatic and primary tumors share a common ancestor, yet continue to evolve independently (i.e., branched evolution; ref. 13). Interestingly, those findings also suggested that clinically actionable alterations present in BCBM are often absent in primary breast biopsies (13).

The establishment of BM has been described as a cascade involving primary breast cancer cell invasion into feeding vessels, dissemination/survival through the circulatory system, adhesion to brain vascular endothelial cells, and extravasation for colonization and ultimately propagation within the brain (14, 15). Arrest of circulating tumor cells (CTC) at branches of capillaries denotes the first step in the transmigration process (16). Active adhesion between CTCs and the vascular wall is required for efficient extravasation, though narrow capillary size alone is capable of ensnaring CTCs (17). Following adhesion, extravasation of CTCs involves massive endothelial remodeling with encapsulation of single cells or clusters of arrested CTCs by the formation of endothelial domes (17). Finally, astrocytes produce inflammatory factors and support colonization/propagation via the activation of STAT1 and NF-κB pathways in BM cells (18).

Cancer stem cells (CSC) have been demonstrated to play a critical role in the initiation of metastasis, whereas other studies have suggested that ALDH-positive cells exhibit stem cell–like properties, thus indicating a potential role in metastatic cascade (19–21). Also, it has been reported that both CD44⁺/CD24⁻ and ALDH1⁺ breast CSCs are enriched in TNBC and may contribute to chemotherapy resistance and tumor metastasis of TNBC (22, 23). Interestingly, our previous work in glioblastoma (GBM) has suggested that ALDH1A3 and CD44 expressions are correlated with the aggressive and radioresistant mesenchymal (MES) subtype as compared with the proneural (PN) subtype (24).

Herein, we sought to identify a novel molecular signature of BCBM in an effort to develop and ultimately engage therapeutic targets such as ALDH1A3. In so doing, we have demonstrated that ALD1A3 is involved in core aspects of BCBM and can be effectively engaged via a novel small-molecule inhibitor (i.e., MF-7).

All work related to human tissue was performed at the University of Alabama at Birmingham (UAB) under an Institutional Review Board–approved protocol (N150219008) compliant with guidelines set forth by NIH. Patients gave informed, written consent for use of their tissues as appropriate in accordance with this permission, and the study protocol conformed to the Declaration of Helsinki. Patient-derived specimens were provided to the corresponding scientists after deidentification of the original tumors.

The chemical structures of test compounds MF-7 and GA11are found in Fig. 6A. They were synthesized as outlined in Supplementary Fig. S1. The commercially available 5-bromopyridin-2-amine (1.00 mmol) was reacted with 2-bromo-1-phenylethan-1-one (1.20 mmol) and potassium carbonate (1.20 mmol) in refluxing ethanol to give the key intermediate 6-bromo-2-phenylimidazo[1,2-a]pyridine. Treatment of the intermediate (1.00 mmol) with the suitable phenylboronic acid (1.50 mmol), in the presence of Pd(OAc)₂ (0.10 mmol), PPh₃ (0.20 mmol), and 2 mL of Na₂CO₃ 2M, in toluene solution, afforded the target inhibitors, MF7 and GA11. Crude compounds were purified by column chromatography (silica gel, ethyl acetate/petroleum ether), recrystallized from the suitable solvent, and characterized by physio-chemical and spectroscopic data (Supplementary Table S1).

MDA-MB-231 (MDA-231) cells were purchased from the ATCC, and MDA-MB-468 (MDA-468) cells were kindly provided by Dr. Amdra R. Frost (UAB). We established a metastatic murine model via the cardiac injection of MDA-MB-231 and MDA-MB-468 using immunocompromised mice (SCID Beige) and then produced a series of derivative cell lines from the resultant tumors within the brain (231-BM and 468-BM), lung (231-LM), or bone (231-BoM). These cells were grown in advanced RPMI 1640 medium (Gibco) supplemented with 1% penicillin/streptomycin, 10% FBS (Atlanta Biologics), and 100x l-glutamine (Gibco). HBEC cells were purchased from the ATCC and cultured in DMEM/F12 medium (Gibco) supplemented with endothelial cell growth factor (Millipore). Cell lines were screened for Mycoplasma every 4 months.

Lentiviral vectors expressing shRNA for ALDH1A3 were purchased from Sigma. Lentiviruses were produced as has been previously described (25). Briefly, 293T cells (Invitrogen) were cotransfected with pLKO.1 vector encoding the shRNA and the helper plasmids for virus production (psPAX2 and pMGD2) using calcium phosphate (Clontech). Lentivirus was harvested at 72 hours after transfection and concentrated 100-fold using Lenti-X concentrator (Clontech). Viral transduction was performed with polybrene according to the manufacturer's protocol. shRNAs targeting ALDH1A3 were produced based on the following clones (Sigma) (shALDH1A3 clone 1: CCGGGCAACCA-ATACTGAAGTTCAACTCGAGTTGAACTTCAGTATTGGTTGCTTTTT; shALDH1A3 clone 2: CCGGGCCGAATACACAGAAGTGAAACTCGAGTTT CACTTCTGTGTATTCGGCTTTTT).

One day before transfection, dissociated cells (4 × 10⁵ cells) were plated in medium without antibiotics. After diluting lipofectamine RNAiMAX reagent in Opti-MEM medium (Gibco) and the desired concentration of siRNA (100 nmol/L) in Opti-MEM medium, both mixtures were combined (1:1 ratio) and incubated at room temperature (RT) for 5 minutes. The siRNA–lipid complex was added to each well and incubated for 6 hours. Experiments were performed 72 hours after transfection.

Total RNA was extracted using RNeasy mini kit (Qiagen) as per the manufacturer's instructions. RNA concentration was determined using Nanodrop One (Thermo Scientific). cDNAs were synthesized using iScript reverse transcription supermix (Bio-Rad) according to the manufacturer's protocol. qRT-PCR was performed on StepOnePlus thermal cycler (Thermo scientific) with SYBR Select Master Mix (Thermo scientific). GAPDH was used as an internal control. Primer sequences are shown in Supplementary Table S2.

Western blotting (WB) was performed as has been previously described (26). Antibodies against ALDH1A3 (rabbit, Sigma), CD44 (mouse, Cell Signaling Technology), Vimentin (rabbit, Cell Signaling Technology), and β-actin (mouse, Cell Signaling Technology) were used. Staining was visualized with Amersham ECL Western Blot System (GE Healthcare).

IHC and immunofluorescence (IF) were performed as previously described (27). For IHC, signals were detected using the DAB substrate kit (Vector). For double staining, donkey IgG H&L (alkaline phosphatase) preadsorbed antibody (Abcam) was used for the targeted primary antibody as a secondary antibody and detected by the liquid fast-red substrate kit (Abcam). The primary antibodies used in this study were as follows: anti-CD44 (mouse, Cell Signaling Technology), anti-Vimentin (mouse, Cell Signaling Technology), and anti-ALDH1A3 (rabbit, Sigma). For IF, primary antibodies were visualized with Alexa Fluor 488 (green)– or 555 (red)–conjugated secondary antibodies (Cell Signaling Technology). Nuclei were counter-stained with Hoechst 33258 (Thermo Fisher).

Viability of GBM cells was determined using an AlamarBlue assay (Thermo Fisher Scientific). Briefly, cells were seeded at 3,000 cells per well in a 96-well plates (excitation at 515–565 nm and emission at 570–610 nm were the parameters employed); a Synergy HTX multi-mode reader (BioTek) was utilized for all experiments.

Coverslips placed in a 24-well plate were coated with laminin (Sigma) diluted in PBS (1:100) for 3 hours. HBEC cells were then seeded on these coverslips and allowed to grow to approximately 90% confluency. Tumor cells were labeled by incubating with MPIO-GFP (Bangs lab) for 24 hours and allowed to adhere to the monolayer for 30 minutes. After washing off the nonadherent cells, coverslips were fixed with 4% paraformaldehyde (PFA).

Transwell inserts were coated with 0.2% gelatin (Sigma) for a minimum of 30 minutes and placed in 24-well plate. HBEC cells (4 × 10⁴) were applied to upper chamber of the inserts and allowed to grow for 3 days. Tumor cells were labeled by incubating with MPIO-GFP for 24 hours and applied (4 × 10⁴), seeded on the upper chamber, and incubated for 24 hours. Inserts were washed with PBS, fixed with 4% PFA for 20 minutes, and washed with PBS twice. The inserts were removed from the plate, and nonmigrated cells were scraped off with cotton swabs. Membranes were cut from the inserts, placed the lower side up, and mounted with prolong gold antifade reagent on a microscope slide.

After labeling with MPIO-GFP (Bangs lab) for 24 hours, 1 × 10⁵ tumor cells were injected into the left ventricle of mouse hearts. Mice were sacrificed at 1 hour following intracardiac injection, and brains were placed in a dish containing ice-cold Hank's Balanced Salt Solution on ice. Brains were then embedded with 6% agarose gel and placed on ice-cold stainless-steel brain matrices (0.5 mm, Coronal, 40–75 g) before sectioning. Vessels in brains were stained with anti-collagen IV antibody (rabbit, Millipore) followed by Alexa Fluor 555 (red)–conjugated secondary antibodies (Cell Signaling Technology).

The ALDEFLUOR kit (StemCell Technologies) was used to isolate populations with high ALDH enzymatic activity according to the manufacturer's protocol via a FACS Aria III (BD Bioscience). Briefly, cells were incubated in ALDEFLUOR assay buffer containing an ALDH substrate (1 μmol/L per 1 × 10⁶ cells) for 30 minutes at RT. Each sample of cells was also stained with 150 μmol/L of diethylaminobenzaldehyde, a specific inhibitor of ALDH, as a negative control.

All animal experiments were performed at the UAB under an Institutional Animal Care and Use Committee–approved protocol (#20290) in accordance with NIH guidelines. Immunocompromised mice (SCID Beige) were purchased from Charles River. For intracardiac injection, mice were anesthetized with ketamine/xylazine and fixed in place. The needle was inserted into the left ventricle of the heart properly from the under-costal space. When blood was observed pumping into the needle tube, we began to slowly inject the cell solution. Mice were placed on the stage warmed at 37°C until they were fully awake. For intracranial injection, dissociated breast cancer cells were stereotactically injected into the striatum of mice as described (28).

Bioluminescence imaging (BLI) was performed as previously described (26). Mice were injected i.p. with 2.5 mg/100 μL solution of XenoLight D-luciferin (PerkinElmer). The tumor luciferase images were captured using an IVIS 100 imaging system (PerkinElmer) 10 minutes after injection of substrate. We also imaged mice with a 9.4T MRI system, Biospec 94/20 (Bruker). Mice were anesthetized with isoflurane, and temperature was maintained via a heated circulating water system during the studies; respiration was monitored with a system from SA Instruments.

Briefly, cDNA libraries for paired end sequencing were prepared using TruSeq Stranded mRNA-Seq Library Preparation Kit (Illumina) according to the manufacturer's protocol. Samples were sequenced with an Illumina HiSeq 2000 system (Illumina), and 100 bp paired-end reads were generated. The raw sequencing and processed data have been submitted to Gene Expression Omnibus with accession number (GSE138361). Sequence reads in fastq format were imported into a local instance of galaxy (galaxy.uabgrid.uab.edu). STAR (version 2.5.3a) was used to align the raw RNA-Seq fastq reads to the human reference genome from Gencode (GRCh38 Release 25; ref. 29).

Gene set enrichment analysis (GSEA) was performed using available online software (http://software.broadinstitute.org/gsea/index.jsp). Gene Ontology (GO) enrichment analysis also was performed using available website (http://geneontology.org/).

Statistical analysis was performed using XLSTAT 2018.5, SPSS statistical package version 25, and Graphpad Prism 7.0 software. All data were presented as the mean ± SD. P values < 0.05 were considered statistically significant. Statistically significant differences in Kaplan–Meier survival curves were determined by the log-rank analysis.

To investigate the clinical impact of BCBM in comparison with primary breast cancer and breast cancer metastases to other organs, we performed analyses of patient OS before and after metastasis. Based on the Centers for Disease Control and Prevention (CDC) database (https://gis.cdc.gov/Cancer/USCS/DataViz.html), there has been a rising trend in breast cancer annual incidence; however, annual deaths have markedly decreased (Fig. 1A). Five-year survival in breast cancer has also demonstrated a positive trend over the past four decades (Supplementary Fig. S2). Following meta-analysis of clinical data from recent studies, we determined that despite reports of elevated incidence of BCBM in HER-2⁺ breast cancer (41%), patient OS in HER-2⁺ BCBM is progressively improving, whereas TNBC BM shows significantly worse prognosis with no observable survival benefit over the past decade (Fig. 1B; Supplementary Table S3; refs. 3–5, 30–41). Collectively, these data demonstrate the significant challenges encountered in the clinical application of existing treatment paradigms and in novel therapeutic development for BCBM.

Recent studies, including our own (24), have identified that epithelial–mesenchymal transition (EMT) of brain tumor cells plays a critical role in their malignant transformation and ultimate resistance to therapy. To clarify the relationship between the three EMT-representative proteins (ALDH1A3, CD44, and Vimentin) and BM, we compared their expression in tumor tissue derived from organs which were seeded by breast cancer [i.e., BM, lung metastasis (LM), and bone metastasis (BoM)] of matched patients. None of the patients displayed metastasis to all three organs and as such we compared the expression between either BM and LM (patient 8) or BM and BoM (patient 6); a preoperative MRI displayed findings typical of BM in patient 8, followed by tumor resection (Fig. 1C and D). IHC staining determined that ALDH1A3 was highly expressed in BM, whereas expression levels of ALDH1A3 were significantly lower in both LM and BoM (Fig. 1E and F; Supplementary Fig. S3). This trend was, to some extent, observed by IHC for CD44 and Vimentin (Fig. 1E and F; Supplementary Fig. S3). These findings indicate that breast cancer cells convert to an MES phenotype when metastasizing to the brain, but not lungs or bones, accompanied by strong induction of ALDH1A3 expression.

To further evaluate the signature of metastasized breast tumors, we established a metastatic murine model via the cardiac injection of human TNBC cells; MDA-MB-231 (MDA-231) and MDA-MB-468 (MDA-468) using immunocompromised mice (SCID Beige). IHC examining resultant metastatic tumors demonstrated that BMs express ALDH1A3, CD44, and Vimentin at markedly higher levels in comparison with their LM and BoM counterparts (Fig. 2A; Supplementary Fig. S4A). These findings indicate that BCBM murine mouse models faithfully recapitulate the ALDH1A3 and MES marker expression previously noted in human BCBM.

From the resultant tumors within the brain, lung, or bone, we produced a series of derivative cell lines and performed RNA sequencing (RNA-seq) of these tumor cells. Subsequent principal component analysis (PCA) of MES gene expression profiles from these derivative lines demonstrated that those cells derived from BM are in fact distinct from primary, LM, and BoM; heatmap analysis of 373 EMT-representative genes does confirm this (Fig. 2B and C; Supplementary Table S4). We selected 13 genes that overlapped between 373 EMT-representative genes and our 31 MES gene set in GBM and performed qPCR. Note that 231 BM showed different expression pattern and highly expressed ALDH1A3, CD44, and Vimentin as compared with cells obtained from primary and other metastatic sites (Fig. 2D and E; Supplementary Fig. S4B). Fluorescence-activated cell sorting (FACS) further confirmed elevated ALDH1A3 expression in 231-BM on the cellular level, compared with MDA-231 (Fig. 2F; Supplementary Fig. S4C). WB and IF also showed the difference in expression of ALDH1A3 at the protein level (Fig. 2G and H). CD44 and Vimentin protein levels were consistently higher in BM, supporting the qRT-PCR results (Supplementary Fig. S4D). Collectively, our data indicate that these mouse models of breast cancer–derived metastases faithfully recapitulate the MES phenotype of human disease, further supporting the notion that ALDH1A3 is selectively upregulated in BMs compared with primary lesions and/or other metastases of breast cancers.

Given the elevated expression of the MES markers (ALDH1A3, CD44, and Vimentin) in BM tissues, we sought to determine the intratumoral spatial distribution and temporal change of these markers throughout the course of initiation and propagation of BM tumors. In sharp contrast to CD44 expression, ALDH1A3 expression was noted to a greater extent at the tumor-stromal cell boundary (Fig. 3A and B; Supplementary Fig. S5).

Of note, the intensity of ALDH1A3 expression was substantially higher at the infiltrating edges compared with the core of the tumor, indicating that ALDH1A3 may play some role in expansion and/or invasion of BM cells into the brain parenchyma (Fig. 3C). To evaluate the temporal changes of ALDH1A3 and CD44 expression from the time of colonization to tumor cell expansion, we investigated ALDH1A3 and CD44 expression in the brain at several time points (i.e., 7, 14, and 28 days) following intracardiac injection of MDA-231. IHC revealed that ALDH1A3 expression was detectable as early as 7 days, whereas CD44 expression gradually increased between days 14 and 28 (Fig. 3D and E). These data suggest that ALDH1A3 may play a role in the early/initial phases of BM formation (i.e., in the transition from circulating breast cancer cells to tumor).

Because ALDH1A3 is preferentially expressed by BM lesions (Fig. 1), we investigated whether ALDH1A3 drives establishment of BM by circulating breast cancer cells. We employed a gene silencing strategy using lentiviral infection with shRNA targeting ALDH1A3 in MDA-231 and MDA-468 cells. qRT-PCR analysis for relative expression of ALDH1A3 in infected MDA-231 and MDA-468 cells confirmed downregulation of ALDH1A3 in these two cell lines (Fig. 4A). shRNA-mediated gene silencing of ALDH1A3 resulted in attenuated cell proliferation of MDA-231 and MDA-468 cells (Fig. 4B). Comparative analysis for adhesion capacity in shALDH1A3- or shNT-infected MDA-231 and MDA-468 cells demonstrated significantly reduced adhesion ability following ALDH1A3 knockdown (Fig. 4C; Supplementary Fig. S6A). Similarly, a migration assay comparing shNT- and shALDH1A3-infected MDA-231 and MDA-468 cells demonstrated significantly reduced migration ability following ALDH1A3 knockdown (Fig. 4D; Supplementary Fig. S6B). In line with such findings, GO enrichment analysis comparing sh/siNT or sh/siALDH1A3 MDA-231 cells demonstrated significant downregulation of gene sets associated with cell–cell adhesion, cell junction, and extracellular matrix (Fig. 4E). GSEA comparing ECM receptor interaction in sh/siNT- and sh/siALDH1A3-infected MDA-231 cells demonstrated inhibition of this pathway following ALDH1A3 knockdown (Fig. 4F). These data strongly suggest that ALDH1A3 plays roles in the cell adhesion and migration of circulating breast cancer cells.

Next, we sought to establish a means of visualizing and monitoring the initial phase of BM formation. To this end, we developed brain slices utilizing mice that received intracardiac injection of GFP-labeled MDA-231 cells. Interestingly, GFP signals were already detected in the capillaries of these mouse brains at 2 hours after intracardiac injection (Fig. 4G). For validation of these findings, we performed intracardiac injection of shNT or shALDH1A3 MDA-231 cells labeled with ultrasmall superparamagnetic iron oxide nanoparticles, followed by T1 weighted MRI at 2 hours and 8 days. MRI demonstrated substantially decreased metastatic foci as early as 2 hours and to a lesser extent at 8 days, in the shALDH1A3 group (Fig. 4H). Collectively, these data indicate that circulating breast cancer cells rapidly colonize in the brain to form metastases in an ALDH1A3-dependent manner as early as 2 hours following intracardiac injection.

Next, we sought to explore the function/phenotype of ALDH1A3 in BCBM. To this end, we employed a gene silencing strategy using lentiviral transduction of shRNA targeting ALDH1A3 (shALDH1A3) in MDA-231 cells. BLI revealed that shALDH1A3 knockdown in MDA-231 cells prior to intracardiac injection significantly reduced the frequency and size of BM compared with nontargeting shRNA (shNT) control (Fig. 5A). Hematoxylin and eosin (H&E) staining of brain slices from these two groups of mice showed that BM formation was substantially diminished upon ALDH1A3 knockdown (Fig. 5B). Suppression of ALDH1A3 expression by shRNA infection was confirmed in all three organs with metastatic tumors, although basal levels of ALDH1A3 expression in LM and BoM were markedly lower than that in BM (Fig. 5C). Interestingly, shALDH1A3 also diminished CD44 and Vimentin expression in BM, reinforcing that ALDH1A3 may promote the expression of MES markers during the initiation phase of BCBM (Fig. 5D). Collectively, these findings support a role for ALDH1A3 during BM initiation and MES phenotype development, without an appreciable effect on LM and BoM formation.

To investigate the BM-specific effect of shALDH1A3, we compared the outcome of intracardiac and intracranial injection of MDA-231 and MDA-468 cells. BLI showed that intracardiac injection leads to the formation of all metastases: BM, LM, and BoM, whereas intracranial injection led to no metastases (Fig. 5E). In brains, small and multiple BM foci were observed in the intracardiac injection model, whereas intracranial injection formed larger, solitary tumors (Fig. 5E and F). As expected, ALDH1A3 knockdown significantly prolonged survival following intracardiac injection of both MDA-231 and MDA-468 (Fig. 5G; Supplementary Fig. S7A). In contrast, the impact of ALDH1A3 knockdown on the survival of these tumor-bearing mice with intracranial injection was less pronounced (Fig. 5H; Supplementary Fig. S7B). These results indicate the essential yet not exclusive role of ALDH1A3 with regard to initial phases of circulating breast cancer cells establishing BM formation.

To evaluate ALDH1A3 targeting as a potential therapeutic strategy in BCBM, we utilized a medicinal chemistry approach to develop a novel small molecule that binds to and inhibits the activity of ALDH1A3. We modified the imidazo [1,2-a] pyridine hit GA11, which we previously described as an effective ALDH1A3 inhibitor, by inserting a fluoro substituent in the para position of the 6-phenyl ring (Fig. 6A; ref. 42). The resultant compound, MF-7, fully preserved the inhibitory activity of the imidazo [1,2-a] pyridine class compound against the target protein when tested in a cell-free assay performed on recombinant human ALDH1A3 (IC₅₀ 22.8±1.6 μmol/L for MF-7 vs. IC₅₀ 4.7±1.7 μmol/L for GA11; Supplementary Fig. S1; Supplementary Table S1), but displayed a significantly increased functional efficacy, compared with GA11, when tested for antiproliferative activity in the MDA-468 cell line; log-dose analysis indicates enhanced response in comparison with NH32, MF-6, and GA11 (Fig. 6B; Supplementary Fig. S8A). Collectively, these findings indicate target engagement of the novel ALDH1A3 inhibitor (MF-7) and cytotoxicity against breast cancer cells in vitro.

To evaluate in vivo, we intraperitoneally administered MF-7 in mice for 8 consecutive days before and after intracardiac injection of MDA-231 and MDA-468 cells (Fig. 6C). This treatment protocol did not cause adverse reactions in the treated mice, yet significantly prolonged their survival compared with the DMSO control group (Fig. 6D). H&E staining of brain slices demonstrated formation of smaller and fewer BM tumors in the MF-7–treated mice compared with the control (Fig. 6E). As expected, the levels of CD44 expression in the MF-7–treated tumors were reduced compared with the control group (Fig. 6F). Surprisingly, and in sharp contrast to the effect on BM formation, the numbers of LM or BoM were paradoxically higher in the MF-7 group as compared with the DMSO group (Fig. 6G and H). We posited that the susceptibility of increased frequency of LM and BoM may be related to alteration of the target organ following inhibition of ALDH1A3 activity. This hypothesis may be explained due to single-cell RNA-seq (scRNA-seq) which determined that ALDH1A3 expression was elevated in normal brain endothelial cells, whereas in the normal lung, ALDH1A3 expression was elevated in fibroblasts (Supplementary Fig. S8B and S8C), indicating variance of ALDH1A3 expression in normal vascular endothelial cells of the target organs (i.e., brain, lung). Collectively, these results suggest that ALDH1A3 inhibition is selectively effective in suppressing BM, but not metastases to other organs (i.e., lung, bone).

Herein, we have highlighted current clinical challenges surrounding the management of BCBM, which is underscored by a lack of recent prognostic and therapeutic advancements in the setting of TNBC. We also sought to clarify the molecular signature associated with BCBM (6–8), and in so doing identified key roles for ALDH1A3 and genes previously associated with EMT. We determined that ALDH1A3 expression is selectively elevated in human BCBM, a finding that is recapitulated by BM cell cultures in vitro as well as the tumors established after BM xenografts in vivo. Further, our data suggest that elevated expression of ALDH1A3 precedes other MES markers (e.g., CD44) during the early phase of BM formation. We demonstrated that ALDH1A3 plays roles in the initiation and cell adhesion phases in the BM developmentary cascade, but not LM and BoM. Genetic inhibition of ALDH1A3 via lentiviral shRNA led to significant survival benefit in BCBM mouse models (56% intracardiac, 32% intracranial). Beside our data, other recent findings reveal organ-specific enhancement of metastasis and unique gene signature expression using multiple sublines derived from BM, LM, and BoM of the parental cell line MDA-MB-231 (43, 44). In BM, ST6GALNAC5 was reported as key molecule to enhance CTC's adhesion to brain endothelial cells and their passage through BBB (6). In addition, SLITRK2, TMEM47, and LYPD1 were identified as upregulated genes in BM compared with LM and BoM (45).

The ALDH superfamily is a group of nicotinamide adenine dinucleotide NAD or NADP⁺-dependent enzymes acting as the primary aldehyde metabolic system in human cells (46). Of note, the level of ALDH enzymatic activity has been used as a CSC marker and seemingly correlates with tumor aggressiveness (47, 48). ALDH1A3 is abnormally expressed in various cancers and CSCs, including breast cancer (49–51). Marcato and colleagues reported that ALDH1A3 is strongly expressed in poorly differentiated and metastatic late-stage breast cancer patients (48), whereas other studies have highlighted the role of ALDH1A3 expression in patients with TNBC which correlates with clinical stage (52, 53). Our prior study determined that ALDH1A3 expression was higher in GBM's aggressive subtype of glioma stem cells (24). Together, ALDH1A3 as CSC-associated enzyme is a mediator of tumorigenicity, metastasis, and therapy resistance as is the case with CSC-associated signaling pathways such as Notch, Wnt, and Hedgehog (54). In melanoma, ALDH1A3 knockout by siRNA significantly decreased the expression of melanoma stem cell–related genes such as CDC42 and USH1C (55). Despite this, ALDH1A3 has yet to be evaluated in BM.

Previously, we had determined that ALDH1A3 plays a role in GBM's PN–MES transition (24). Here, our findings indicate that ALDH1A3 plays roles in CTC adherence to brain vascular endothelial cells and extravasation into brain parenchyma. In addition, the inhibition of ALDH1A3 prolonged xenograft mice survival following both intracardiac and intracranial injection of TNBC cells, albeit to a lesser degree by the latter. Given the antitumor formation effects of ALDH1A3 inhibition in intracranial injection models, it is likely that ALDH1A3′s function involves a multistep process in BCBM which may include effects related to our prior findings in GBM. These data raise the possibility that elevated cancer cell ALDH1A3 activity provides an advantage with regard to both survival and progression of cancer cells within the brain. Future investigations will seek to fully elucidate the mechanisms underlying ALDH1A3′s effects, and in doing determine whether ALDH1A3 is responsible for a shared role in both primary and metastatic CNS cancers.

Our findings suggest significant therapeutic benefit via application of MF-7 for both primary prevention of BCBM in patients with breast cancer and post-BCBM establishment. However, one side-effect of MF-7 administration in intracardiac injection models was elevated susceptibility with regard to LM and BoM formation, which may be related to alteration of the target organs by circulating breast cancer cells following inhibition of ALDH1A3 activity. Although the survival of intracardiac injection models following MF-7 administration exhibited an overall increase (∼20%), suggesting a positive trade-off in cost-benefit, it remains to be seen whether similar findings will be observed in humans. In addition, comparisons in patient quality of life would ultimately be required to assess potential improvement due to reduced BM formation versus reduced LM and BoM. Given that ALDH1A3 expression is elevated in normal endothelial cells of the brain (Supplementary Fig. S7A), likely similar targeting mechanisms may be employed to reduce metastases formation in other organs. Recently studies reported that citral is an effective ALDH1A3 inhibitor to block ALDH1A3-mediated BC growth (54) and that NRAD1 targeting therapy has the potential for the treatment of TNBC and the reduction of CSCs (56). These will be points of interest to be discerned by future BCBM preclinical studies and clinical trials.

Several recently developed therapeutics are under investigation for BCBM; however, the HER2⁺ phenotype of BCBM has dominated recent clinical trials (57–59), potentially owing to increased frequency of HER2⁺ BCBM. Systemic modalities including manipulation of the BBB allowing for delivery of traditional chemotherapeutics, novel chemotherapeutics (e.g., etirinotecan pegol, peptide–paclitaxel conjugates), novel targeted therapies (e.g., pan-HER receptor and CDK4/6 inhibitors), PARP inhibitors, and immunotherapies are all under investigation by preclinical and clinical studies (60). Given the lack of therapeutic modalities with appreciable efficacy in TNBC BM, targeting ALDH1A3 may prove an effective option in this intractable disease. Beyond BCBM, the inhibition of ALDH1A3-mediated pathways may ultimately provide a promising therapeutic approach for cancers displaying MES and/or stem-like signatures (e.g., aggressive subsets of GBM; refs. 24, 61). Interestingly, ALDH1A3 expression in human GBM has been correlated with a worse prognosis and is associated with resistance against Temozolomide treatment with sensitivity to the drug having been re-established in ALDH1A3 knockout cells (62). Accordingly, a number of strategies have begun to emerge that seek to target ALDH1A3 isoforms ranging from immunotherapy (63) to traditional small-molecule–based approaches (64).

In this study, we highlighted a lack of appreciable improvement in OS of TNBC BM over the past two decades. We demonstrated novel roles for ALDH1A3 during the cell migration and adhesion phases of the metastatic cascade in BCBM, but not LM or BoM. Targeting ALDH1A3 proved to be effective via both genetic and pharmacologic inhibition. Our findings present ALDH1A3 as an attractive target for primary prevention and therapeutic management of BCBM, suggest applicability of similar metastasis-guiding molecules to other cancers beyond TNBC, and provide a rationale for the therapeutic targeting of ALDH1A3 in a broader context. The identification of ALDH1A3-specific inhibitors may create a new paradigm for development of targeted therapeutics for BCBM and other metastatic carcinomas.

J.D. Bernstock has an ownership interest (including patents) in CITC Ltd. and Avidea Technologies, and has an unpaid consultant/advisory board relationship with POCKiT Diagnostics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: I. Nakano

Development of methodology: D. Yamashita, M. Minata, V. Coviello, B.A. Tannous, I. Nakano

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Yamashita, M. Minata, S. Yamaguchi, S. Harada, C. La Motta

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Yamashita, M. Minata, S. Yamaguchi, V. Coviello, J.D. Bernstock, S. Harada, B.A. Tannous, C. La Motta

Writing, review, and/or revision of the manuscript: D. Yamashita, A.N. Ibrahim, J.D. Bernstock, S. Harada, B.A. Tannous, C. La Motta, I. Nakano

Study supervision: R.A. Cerione, I. Nakano

The authors wish to thank Dr. Amdra R. Frost (UAB) for MDA-MB-468.This study was supported by the NIH grants R01NS083767, R01NS087913, R01CA183991, and R01CA201402 (I. Nakano).

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.