Human breast cancer cell lines (MDA-MB-231, SK-BR-3, and MCF7) and the human embryonic kidney HEK-293T cell line were obtained from the American Type Culture Collection (ATCC). These cell lines were authenticated at ATCC before purchase by their standard short tandem repeat DNA-typing methodology. MDA-MB-231, SK-BR-3, and HEK-293T cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM, Gibco) containing 10% fetal bovine serum (FBS, Gibco). MCF7 cells were cultured in Minimum Essential Media (MEM, ATCC) containing 10% FBS and 0.01 mg/ml human recombinant insulin (Sigma-Aldrich). All cells were incubated at 37°C in a humidified 5% CO₂ atmosphere. Lipid-depleted serum (LDS) were purchased from MACGENE (Catalog. CS021).

pLVX-flag was a gift from Dr. Tao Guo (The First Affiliated Hospital, Dalian Medical University, Dalian, China). pLVX-mVenus-p27K^- was a gift from Dr. Bing Liu (Institute of Cancer Stem Cell, Cancer Center, Dalian Medical University, Dalian, China). VLDLR-I, VLDLR-II, LDLR, and RAP fragments were cloned from human genome cDNA. The primers were as follows: VLDLR-I/II: Forward, 5′-CTACCGGACTCAGATGCCACCATGGGCACGTCCGCGCTCT; Reverse, 5′-TACCCGGTAGAATTATCTAGATCAAGCTAGATCATCATCT. LDLR: Forward, 5′-CCGCGGCCGCGCCACCATGGGGCCCTGGGGCTGGAA; Reverse, 5′-TCCATATGTCACGCCACGTCATCCTCCA. RAP: Forward, 5′-GATCTGGTTCCGCGTGGATCCATGGCGCCGCGGAGGGTCA; Reverse, 5′-GTCACGATGCGGCCGCTCGAGTCAGAGTTCGTTGTGCCGA. DsRed fragment was replaced by the VLDLR-I/II fragment in the pLVX-DsRed-Monomer-N1 vector (Clontech). The LDLR fragment was ligated into the pLVX-TRE3G vector (Clontech). The RAP fragment was recombinated into the pGEX-4T-1 vector (Clontech). The shRNAs specifically targeting VLDLR were cloned into the pLKO-Tet-On-shNC vector. The primers were as follows: shVLDLR-1: Forward, 5′-CCGGGCACAGATGATGATCTAGCTTCTCGAGAAGCTAGATCATCATCTGTGCTTTTTG; Reverse, 5′-AATTCAAAAAGCACAGATGATGATCTAGCTTCTCGAGAAGCTAGATCATCATCTGTGC. shVLDLR-2: Forward, 5′-CCGGGCTTGATTCTAAGTTGCACATCTCGAGATGTGCAACTTAGAATCAAGCTTTTTG; Reverse, 5′-AATTCAAAAAGCTTGATTCTAAGTTGCACATCTCGAGATGTGCAACTTAGAATCAAGC.

Lentiviruses were packaged in HEK-293T cells using the second-generation packaging system plasmid psPAX2 (Addgene) and pMD2.G (Addgene). Transfection was performed with LipoD293 (SignaGen) following the manufacturer’s instructions. The lentivirus was harvested at both 24 h and 48 h after transfection and used for cell infection in the presence of 8 μg/ml polybrene. After infection for 48 h, the infected cells were selected with 2 μg/ml puromycin (Sigma–Aldrich) or 1 mg/ml G418 (Sigma–Aldrich).

GST and GST-RAP fusion protein were expressed in bacteria and then purified. Briefly, recombined plasmid was expressed in E. coli Transetta (DE3) cells. At OD₆₀₀ = 0.5, protein expression was induced with 1 mM IPTG (Solarbio) for 5 h at 25°C. Bacteria were lysed and sonicated in lysis buffer containing 50 mM Tris-HCl (pH 7.4), 300 mM NaCl, 0.5% Triton X-100, PMSF, DTT, lysozyme, and protease inhibitor cocktail (Sigma-Aldrich) at 4°C. After centrifugation at 12,000 g for 15 min at 4°C, the supernatant was used for the protein purification with GST resin (TransGen Biotech) according to the manufacturer’s instructions. Elution of recombinant protein was performed under mild, non-denaturing condition using reduced glutathione. Then, the fractions were collected and analyzed with SDS-PAGE followed by Coomassie brilliant blue staining.

Breast cancer cells were treated with 200 nM GST-RAP fusion protein for 72 h followed by immunofluorescent staining. Briefly, cells grown on slides were fixed with 4% paraformaldehyde. After blocking, slides were stained with anti-GST antibody (Proteintech, 10000-0-AP). Slides were then incubated with Alexa-488 Goat anti-Rabbit IgG (Invitrogen, A11008). Nuclei were stained with DAPI (Thermo Fisher Scientific).

Total RNA was extracted using Trizol reagent (Invitrogen) following the manufacturer’s instructions. The cDNA was generated using the EasyScript One-Step gDNA Removal and cDNA Synthesis SuperMix Kit (TransGen Biotech) according to the manufacturer’s instructions. The amount of cDNA used as the PCR template was equivalent to 60 ng of total RNA. The quantitative PCR reactions were performed by limiting cycle numbers from 28 to 30 dependent on the transcript examined. ACTB was used as an internal control for normalization. The PCR products were separated in a 2% agarose gel and detected by ethidium bromide staining. The primers were as follows: VLDLR: Forward, 5′-CAACCTGAATGATGCCCAAGA; Reverse, 5′-CTTTTGGGGGAACACTGACCT. ACTB: Forward, 5′-TTGCCGACAGGATGCAGAAGGA; Reverse, 5′-AGGTGGACAGCGAGGCCAGGAT. Each experiment was repeated three times.

Cells were lysed on ice with RIPA buffer [150 mM NaCl, 0.5% sodium deoxycholate, 0.1% SDS, 1% NP40, and 50 mM Tris (pH 8.0)] supplemented with protease inhibitor cocktail (Sigma-Aldrich). Lysates were cleared by centrifuging at 12,000 rpm for 15 min. Next, the protein was quantified by the Coomassie brilliant blue dye method. After boiling with loading buffer for 5 min, equal amounts of cellular proteins were loaded to separate in SDS-PAGE and then transferred to a nitrocellulose membrane (Millipore) via submerged transfer. After blocking, the membranes were incubated overnight with primary antibodies at 4°C. Next, the membranes were incubated with HRP-conjugated secondary antibodies at room temperature for 1 h. The signals were visualized using an enhanced chemiluminescence kit (Advansta) according to the manufacturer’s instructions. The following antibodies were used: VLDLR (Proteintech, #19493-1), OCT4 (Abcam, #ab19857), SOX2 (Santa Cruz Biotechnology, #sc-365823), NANOG (Abcam, #ab109250), β-actin (Proteintech, #60008-1), Goat anti-Mouse IgG (HRP conjugated) (Thermo Fisher Scientific, #31430), and Goat anti-Rabbit IgG (HRP conjugated) (Thermo Fisher Scientific, #31460). Each experiment was repeated three times.

Cell proliferation was determined using CCK-8 kits (MedChemExpress) according to the manufacturer’s instructions. Briefly, cells were plated into 96-well microplates at 1 × 10³ cells/well density and the absorbance at 450 nm was measured every 24 h with a multimode plate reader (Perkin Elmer). Each experiment was repeated three times.

Cells were plated into six-well plates at 500 cells/well density. Fresh growth medium was replaced every 3 days. After 7–10 days, colonies were fixed with 4% paraformaldehyde, stained with 0.5% crystal violet (Sigma-Aldrich) solution, washed, and dried. Images of stained plates were captured. Next, bound crystal violet was dissolved by 50% glacial acetic acid solution, and the absorbance was measured at 570 nm with a multimode plate reader (Perkin Elmer). Alternatively, the colony numbers were counted with ImageJ software. Each experiment was repeated three times.

Cells were plated into ultra-low attachment 96-well plates (Corning) at 500 cells/well density and cultured in DMEM/F12 (Gibco) supplemented with B27 (Invitrogen), 20 ng/ml epithelial growth factor (EGF, Sigma-Aldrich), 20 ng/ml basic fibroblast growth factor (bFGF, PeproTech), and 1% methylcellulose (Sigma-Aldrich). EV and VLDLR overexpressed cells were cultured for 10 days, while shNC and shVLDLR cells were cultured for 12 days. Each experiment was repeated three times. For serial sphere experiment, cells were plated into ultra-low attachment six-well plates (Corning) at 5,000 cells/well density and cultured without methylcellulose. One week later, single cells from digested spheres were used for the next generation of sphere formation. The spheres were photographed and counted under an inverted microscope (Olympus).

Propidium iodide (PI, Sigma-Aldrich) staining was used to detect cell cycle distribution. Briefly, the cells were harvested, washed with cold PBS, and fixed in 70% ethanol overnight at 4°C. Cells were washed with PBS; resuspended in 500 µl of binding buffer containing PI, RNase (BD Pharmingen), and 0.01% Triton X-100; and then incubated for 15 min at room temperature in the dark. The cell cycle distribution was analyzed by CytoFLEX flow cytometry (Beckman Coulter). Single cells were gated and were used for cell cycle analysis. Each experiment was repeated three times.

Cell apoptosis was determined by flow cytometry using an Annexin V-FITC/PI kit (Tianjin Sungene Biotech) according to the manufacturer’s instructions. Briefly, adherent and floating cells were collected, washed with PBS, and stained with Annexin V and PI in 1× binding buffer for 15 min at room temperature. Cells were further diluted with the buffer and analyzed using a CytoFLEX flow cytometry (Beckman Coulter). Cell population fractions in four quadrants were analyzed using FlowJo software. Cells were gated and were used for cell apoptosis analysis. The percentage of apoptotic cells (AV⁺/PI^‐ and AV⁺/PI⁺) was calculated using the following formula: Apoptotic rate (%) = (number of apoptotic cells/total number of cells examined) × 100. Each experiment was repeated three times.

Nile red staining was used to assess the lipid content of cells. Nile red powder (Sigma-Aldrich) was dissolved in DMSO to make a 1 mg/ml stock solution, which was kept in the dark at −20°C. This stock solution was then diluted 1:100 in PBS before use. Staining was carried out on both fixed cells and unfixed cells. Cells were fixed in 4% paraformaldehyde and stained with Nile red solution for 15 min at room temperature. Nuclei were stained with DAPI. After washing with PBS, cells were photographed under fluorescent microscopy (Olympus). For FACS analysis, cells were harvested, washed with PBS, and incubated in 500 μl of Nile red solution for 15 min at room temperature. Stained cells were washed with PBS and analyzed using CytoFLEX flow cytometry (Beckman Coulter). Cells were gated and were used for data analysis. Mean fluorescence values were determined using FlowJo software. Each experiment was repeated three times.

All animal studies were approved by the Institutional Animal Care and Use Committee of Dalian Medical University and were carried out in accordance with established institutional guidelines and approved protocols. MDA-MB-231-shNC and MDA-MB-231-shVLDLR cells (1 × 10⁶) in 100 μl PBS containing 50% Matrigel (BD Biosciences) were subcutaneously injected into the left and right dorsal flank of female BALB/c nude mice (4 weeks old), respectively. Drinking water containing 2 mg/ml doxycycline (Dox) and 3% sucrose was used 3 days before subcutaneous inoculation and replaced every 2 days. The body weight of mice and the two perpendicular diameters (length: a, width: b) of tumors were recorded every other day. Tumor volume (V) was calculated according to the standard formula (V = 0.5 × a × b²). At the end of experiment, xenografts were dissected from euthanized mice and then photographed.

Following informed consent and in accordance with the guidelines of the Institutional Research Medical Ethics Committee, breast cancer specimens were obtained from the First Affiliated Hospital of Dalian Medical University.

The molecular subtypes of breast cancer were defined as follows: Luminal A: ER⁺ and/or PR⁺, HER2^-, Ki-67 < 14%. Luminal B (HER2^-): ER⁺ and/or PR⁺, HER2^-, Ki−67 ≥ 14%. Luminal B (HER2⁺): ER⁺ and/or PR⁺, HER2⁺. HER2: ER⁻, PR⁻, HER2⁺. Triple-negative: ER⁻, PR⁻, HER2⁻.

Immunohistochemical (IHC) staining was performed to examine the expression of VLDLR in breast cancer tissues. Briefly, slides were deparaffinized, rehydrated, and subjected to antigen retrieval by heating the sample with citrate buffer (pH 6.0) in a microwave oven. Endogenous peroxidase activity was quenched by 3% H₂O₂. After blocking, sections were incubated with VLDLR primary antibody (1:50; Santa Cruz Biotechnology, #sc-18824) at 4°C overnight. Then, biotinylated secondary antibody, streptavidin-conjugated HRP, and DAB were applied for specific detection. Images were viewed using light microscopy (Olympus).

The IHC staining results were reviewed and scored independently by two pathologists who were blinded to specimens’ clinical information, based on both the intensity of staining and the percentage of positively stained tumor cells. The intensity of protein expression was shown as follows: 0 (no staining), 1 (weak staining), 2 (moderate staining), and 3 (strong staining). The histochemical score (H-score) was calculated as the product of the staining intensity and the percentage of positive cells.

Control (empty vector, EV) and VLDLR-I/II overexpressed cells were lysed with freshly prepared lysis buffer [100 mM NH₄HCO₃ buffer, pH 10.0, 6 M guanidine hydrochloride, and protease inhibitor cocktail (Sigma-Aldrich)] followed by ultrasonication at 4°C for 30 min. The lysate was incubated on ice for 15 min and then pelleted by centrifugation at 12,000 g for 15 min. A BCA protein assay kit (ThermoFisher, China) was used to determine the protein concentration of each sample. A total of 200 μg of protein from each lysate was reduced with 20 mM dithiothreitol (DTT) for 1 h at 55°C followed by protein alkylation with 40 mM iodoacetamide in the dark at room temperature for 30 min. After protein samples were desalted with ultrafiltration device (Vivacon 500, Sartorius) and digested overnight at 37°C with trypsin (Promega 1:50 w/w), the digested peptides were lyophilized and reconstituted in 50 mM HEPES (pH 8.0).

The lyophilized samples were dissolved in 0.1% trifluoroacetic acid (TFA) and quantified using the NanoDrop 2000 (Thermo Fisher Scientific). Approximately 1 μg of sample was run on the Q Exactive HF mass spectrometer coupled with the Ultimate 3000 RSLCnano system to collect data. The mass spectrometric data were processed by Xcalibur software (version 2.1.0, Thermo Fisher Scientific) and raw files were analyzed with MaxQuant software (version 1.5.3.30). All files were searched against the UniProt human reference database using the Andromeda search engine. Digestion mode was set to trypsin with a maximum of two missed cleavages. Carbamidomethylation of cysteine was set as a fixed modification. Variable modifications included N-terminal protein acetylation and methionine oxidation. The precursor mass tolerance and fragment mass tolerance were set at 10 ppm and 0.02 Da, respectively. The false discovery rate (FDR) was set at 0.01 to eliminate low-probability protein identifications.

VLDLR mRNA expression levels of monolayer and sphere-forming cells of breast cancer cell lines (MDA-MB-231-LM2, MCF7, and SUM159) were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). Two original datasets were downloaded (GEO accession no. GSE98239 and GSE43657). VLDLR mRNA expression levels in normal breast tissues and breast cancer tissues were analyzed with the GENT2 database (http://gent2.appex.kr/gent2/). The prognostic significance of VLDLR was evaluated using the PROGgene V2 database (http://www.progtools.net/gene/), GENT2 database (http://gent2.appex.kr/gent2/), and GEPIA2 database (http://gepia2.cancer-pku.cn/#correlation). Gene correlation analysis was done using the GEPIA2 database (http://gepia2.cancer-pku.cn/#correlation).

Experiments were performed at least three times. Statistical analysis was performed using the GraphPad Prism statistical software (GraphPad Prism 8.0.2). All data were expressed as the mean ± SEM. The unpaired t-test was used to compare the difference between two groups. One-way ANOVA followed by Dunnett’s test was used for multiple group comparison. p-value < 0.05 was considered statistically significant (ns: no significance, *p < 0.05, **p < 0.01, ***p < 0.001).

To enrich the BCSCs, three generations of sphere formation (18, 19) were performed on the triple-negative breast cancer cell line MDA-MB-231 (ER⁻PR⁻HER2⁻). The diameter and number of spheres were increased during serial passage ( Figure 1A ), indicating that BCSCs were successfully enriched. Consistently, the expression levels of SOX2 and NANOG, master stemness factors, were increased in spheres compared to monolayer cells (2D) ( Figure 1B ).

Proteomic analysis is a powerful high-throughput technique for understanding cellular protein expression. In this study, isobaric tags for relative and absolute quantification (iTRAQ) technology (20), a novel tool for the detection of protein expression, was used to assess proteomic changes of these three generations of spheres (unpublished data). From the proteomic data, we observed several cancer stem cell-related proteins, including STC1 (21), IL1B (22, 23), ICAM1 (24, 25), CEBPB (26), and HMGA1 (27), were upregulated during serial passage ( Supplementary Figure 1A and Supplementary Table 1 ). Interestingly, VLDLR, a cell surface protein that has not been investigated in BCSCs, was gradually increased in both mRNA and protein level ( Supplementary Figures 1A, B ). VLDLR has two variant forms because of alternative splicing ( Figure 1C ). We further found that both subtypes were gradually increased during serial sphere formation process ( Figure 1B ). We also enriched BCSCs from MCF7 (ER⁺PR⁺HER2⁻) and SK-BR-3 (ER⁻PR⁺HER2⁺) breast cancer cell lines by sphere formation assay for one generation and observed that VLDLR-I in SK-BR-3 and VLDLR-I and VLDLR-II in MCF7 were increased in sphere-forming cells compared to monolayer cells (2D) ( Figure 1D ). Furthermore, increased VLDLR mRNA expression levels were observed in sphere-forming cells from MCF7, SUM159 (ER⁻PR⁻HER2⁻) (28), and MDA-MB-231-LM2 cells (29, 30) ( Supplementary Figure 1C ), which exhibit enhanced lung metastasis capability, using the public GEO datasets (GSE43657 and GSE98239). Furthermore, VLDLR mRNA was significantly upregulated in moderately and poorly differentiated breast cancer tissues compared to well-differentiated compartments ( Figure 1E ). Taken together, these findings indicated that VLDLR was upregulated in BCSCs.

To investigate the biological effect of different subtypes of VLDLR on breast cancer cell stemness, we stably overexpressed VLDLR-I/II in MDA-MB-231 cells ( Supplementary Figure 2A ) and found that the sphere formation capabilities were significantly enhanced by both VLDLR-I and VLDLR-II expression ( Figure 2A ). To prevent the adhesion of adjacent spheres, 1% methylcellulose was added in the sphere medium, resulting in the different morphology of spheres compared with those of serial sphere formation assay. We also developed two stable cell lines transduced with different VLDLR-specific shRNAs that can be induced by Dox. Western blot analysis showed that VLDLR expression was significantly decreased in monolayer MDA-MB-231 cells following Dox treatment for 5 days ( Supplementary Figure 2B ). These stable cell lines were subsequently applied in the sphere formation assay. Results demonstrated that VLDLR knockdown remarkably inhibited stem cell growth, and the numbers of spheres were reduced by more than 50% in VLDLR silenced cells compared to control cells ( Figure 2B ). These data suggested that VLDLR was essential for BCSC expansion.

To further determine the effects of VLDLR on tumor growth in vivo, we subcutaneously inoculated nude mice with MDA-MB-231 stable cells expressing inducible shNC or shVLDLRs into dorsal flank ( Figure 2C ). Dox dissolved in drinking water was employed to trigger the endogenous VLDLR knockdown of the xenograft throughout the experiment. Tumor volume was monitored from day 0 to day 20 after inoculation, and the results showed that the tumors derived from VLDLR silenced cells grew significantly slower than tumors derived from control cells ( Figures 2D, E ). Collectively, these results demonstrated that VLDLR silencing inhibited the tumor development in vivo.

To elucidate the potential mechanisms of VLDLR responsible for the BCSC characteristics, we detected the expression of three master stemness factors. Surprisingly, neither VLDLR-I nor VLDLR-II overexpression had detectable effects on the expression of OCT4, SOX2, and NANOG ( Supplementary Figure 2A ). Consistent with the results from overexpression experiments, no change was observed in the expression of these stemness-related proteins upon VLDLR silencing ( Supplementary Figure 2C ), indicating that the OCT4/SOX2/NANOG regulatory circuit was not affected. We then investigated whether VLDLR knockdown induced cell apoptosis in breast cancer cells. Cell apoptosis was analyzed by flow cytometry, and results showed that the percentages of Annexin V-positive cells were slightly increased but still lower than 3% ( Supplementary Figures 3A, B ). Given that the numbers of spheres were reduced by more than 50% upon VLDLR knockdown ( Figure 2B ), we proposed that the increased cell apoptosis was not enough to explain the inhibited cancer stem cell function in VLDLR silenced cells.

Cell cycle progression is a precondition for the cancer cell proliferation. Upon VLDLR knockdown, we observed increased G0/G1 cell population and concomitant decreased S and G2/M population in MDA-MB-231 cells ( Figures 3A, B ), suggesting that VLDLR silencing induced G0/G1 phase cell cycle arrest. Cells with 2N DNA content were identified in either the G0 or G1 phase. The G1 phase is the first phase in which cells commit to enter the mitotic cycle and prepare to duplicate their DNA in the S phase. The G0 phase, however, is a reversible quiescent state in which cells remain viable but do not proliferate until they are stimulated into an active cell cycle. To distinguish between these two phases, we established a stable cell line expressing mVenus-p27K^- fusion protein, in which the mVenus fluorescent intensity can be used as an indicator of quiescence (31). Cell cycle phase distribution was analyzed by PI staining. As expected, cells with low fluorescent intensity were distributed in all cell cycle phases, while cells with high fluorescent intensity were only distributed in the G0/G1 phase ( Supplementary Figure 4A ). Furthermore, in this stable cell line, the percentage of cells with high mVenus fluorescent intensity was significantly increased following VLDLR knockdown ( Figures 3C, D ), indicating that VLDLR silencing promoted breast cancer cells to enter a quiescent state. Therefore, we concluded that the entry of cellular quiescence contributed to VLDLR knockdown-mediated breast cancer cell growth inhibition. Consistently, cell proliferation and colony formation capacity were remarkably blunted by VLDLR reduction ( Figures 3E–G ), which can be rescued by VLDLR restoration ( Supplementary Figures 4B, C ). In contrast, VLDLR-I and VLDLR-II overexpression markedly enhanced the proliferation and colony growth of cells ( Figures 3H–J ). Taken together, VLDLR is essential for breast cancer cell proliferation.

VLDLR has been reported to play essential roles in lipid metabolism and signal transduction by binding to numerous ligands. We then investigated whether these functions are involved in VLDLR-mediated promotion of breast cancer cell growth.

The 39-kDa RAP, a molecular chaperone, can bind with high affinity to VLDLR, thereby blocking the binding of other ligands to the receptor (6). In this study, GST and the GST-RAP fusion protein were purified ( Supplementary Figure 5A ) and added to the cell culture medium. No significant change in proliferation of MDA-MB-231 cells was found after GST-RAP treatment ( Figure 4A and Supplementary Figure 5B ), indicating that ligand-independent activities may be involved in VLDLR-mediated promotion of breast cancer cell growth.

VLDLR can promote intracellular lipid accumulation through VLDL uptake. To exclude the effect of lipid in the medium, we cultured cells with medium containing 10% FBS or 10% LDS. Interestingly, VLDLR-I and VLDLR-II overexpression in MDA-MB-231 cells significantly promoted cell growth even under lipid depleted condition ( Figure 4B ). Even though VLDLR promoted cell growth at a lower level under lipid-depleted conditions, our results suggested that the lipid-independent function of VLDLR was involved in cancer cell proliferation. Furthermore, cellular lipid was stained with Nile red, a fluorescent neutral lipid stain, and analyzed via flow cytometry. VLDLR exhibits structural similarity to those of the LDLR, except VLDLR has an extra repeat of the cysteine-rich ligand-binding domain ( Figure 4C ). LDLR, an important cell surface receptor, mediates lipid uptake (32), and its role in the progression of familial hypercholesterolemia has been widely studied (33). Therefore, we used LDLR overexpressed cells as a positive control in this experiment. Results showed that the mean fluorescent intensity (MFI) of Nile red in VLDLR-I or VLDLR-II overexpressed cells was not increased compared to control cells ( Figures 4D, E ). Moreover, fixed cells were stained with Nile red and DAPI, and the results were consistent with flow cytometry ( Figure 4F ). In contrast, LDLR overexpression significantly stimulated lipid accumulation in breast cancer cells ( Figures 4D–F ). As expected, VLDLR knockdown did not inhibit lipid accumulation in breast cancer cells ( Supplementary Figures 6A, B ). These results demonstrated that the promotion of cell growth is independent of the lipid metabolism activity of VLDLR in breast cancer cells.

To explore the possible mechanism by which VLDLR-I/II promotes cancer cell proliferation, we performed proteomic analysis in control and VLDLR-I/II overexpressed MDA-MB-231 cells. Results demonstrated that more than half of upregulated and downregulated proteins were shared in VLDLR-I and VLDLR-II overexpressed cells compared to control cells ( Figure 5A , Supplementary Table 2 ), indicating that VLDLR-I and VLDLR-II have similar regulations of protein expression in cancer cells. Biological process enrichment analysis indicated that upregulated proteins upon VLDLR-I/II overexpression were associated with citrate cycle (TCA cycle), carbon metabolism, and ribosome biogenesis ( Figures 5B, C ). Furthermore, in breast cancer tissues, the expression of VLDLR mRNA exhibits positive correlations with gene signatures constituted by genes involved in these pathways ( Figure 5C , lower panel), suggesting that VLDLR expression correlated with elevated energy production and ribosome biogenesis.

To explore the role of VLDLR in breast cancer, we examined the expression levels of VLDLR in 146 cases of breast cancer tissues and 30 control subjects by immunohistochemistry. Although low/negative expression of VLDLR (H-score ≤ 20) was observed in more than half of the breast cancer cases, positive expression of VLDLR (H-score > 20) occurred more frequently in breast cancer tissues than in tumor adjacent normal breast tissues ( Figures 6A, B ). Moreover, strong staining (H-score > 200) was observed in four cases (13.8%) of the triple-negative subtype ( Figure 6B ), which has the highest degree of malignancy and the worst prognosis among all subtypes (34). Furthermore, by analyzing VLDLR expression in the GENT2 database (35), we observed that the VLDLR mRNA expression levels were significantly increased in triple-negative breast cancer (TNBC) compared to other subtypes ( Figure 6C ). Consistently, VLDLR mRNA expression levels were higher in ER- and PR-negative tissues than positive compartments ( Supplementary Figure 7A , left and middle panel). The correlation between VLDLR and HER2 was not observed ( Supplementary Figure 7A , right panel).

To further validate the important role of VLDLR in the outcome of breast cancer patients, we stratified the breast cancer patients based on the VLDLR expression level using a publicly available database (35–37) and found that elevated VLDLR expression was significantly associated with worse overall survival ( Figure 6D and Supplementary Figure 7B ), as well as relapse-free survival ( Figure 6E ). However, this correlation was not observed in the GEPIA2 database ( Supplementary Figure 7C ). We further analyzed overall survival of breast cancer patients in separated subtypes and found that high VLDLR expression significantly correlated with poor prognosis in TNBC patients in contrast to other subtypes ( Supplementary Figure 7D ). Taken together, these results suggested that VLDLR may act as an important protein to promote cancer progression in human breast tumors, especially in TNBC.