Patient samples

The INT-MI series includes HER2 + BC patients who received neoadjuvant treatment with anthracycline-taxane plus trastuzumab between 2009 and 2018 at the Fondazione IRCCS Istituto Nazionale dei Tumori of Milan, Italy (INT). Samples from these patients were collected at baseline (diagnostic biopsies) and after neoadjuvant chemotherapy (surgical samples). All patients received six cycles of chemotherapy and trastuzumab, administered at 3-week intervals. Matched diagnostic biopsies and surgical samples obtained from 32 patients who did not achieve pathological complete response (pCR) upon treatment were subjected to molecular profiling (Table 1). The samples selected for gene expression profiling (GEP) contained at least 70% tumor cells.

Table 1 Clinicopathological characteristics of HER2 + BC patients subjected to molecular profiling analysis

On the other hand, selected nonresponsive (pCR No; n = 18) and responsive (pCR Yes; n = 18) HER2 + BC patients with similar clinical parameters (Table 2) were subjected to multiplex immunostaining and immunolocalization analyses (see below) to evaluate the intratumor percentage (%) of neoplastic stem-like cells with atypical morphology and the HER2 + /CD44v6 + /CD36 + immunophenotype.

Table 2 Clinicopathological characteristics of HER2 + BC patients subjected to analysis of intratumor neoplastic stem-like cell enrichment

All procedures were performed with the ethical approval of the Independent Ethics Committee of the INT, as stated in the Declarations.

Gene expression profiling

Total RNA was extracted from HER2 + BC samples using the miRNeasy Mini Kit (QIAGEN, Hilden, Germany). RNA quality and quantity were assessed using the 2200 TapeStation system (Agilent, Santa Clara, CA, USA) and the Qubit 2.0 Fluorometric Assay (Thermo Fisher Scientific, Waltham, MA, USA), respectively. Gene expression analyses were performed according to the GeneChip WT Pico standard protocols (Affymetrix, Thermo Fisher Scientific). After reverse transcription, cDNA synthesis, amplification, and labeling, the samples were hybridized to human ClariomS arrays for 16 h at 45 °C. After washing and staining using the GeneChip Fluidics Station 450, the arrays were scanned using an Affymetrix Gene Chip Scanner 3000 7G. Primary data were acquired using Affymetrix GeneChip Command Scan Control version 4.0 (developed by Thermo Fisher Scientific). Raw and preprocessed gene expression data are available in the NCBI Gene Expression Omnibus repository under accession number GSE245132.

Bioinformatic analysis

The raw gene expression data were preprocessed using the gene-level signal space transformation–robust multiarray average (SST-RMA) algorithm implemented in Transcriptome Analysis Console software version 4.0.2 (Thermo Fisher Scientific). Preprocessed data were filtered by removing probe sets without associated Entrez gene identifiers. The genes targeted by multiple probe sets were collapsed by selecting the probe set with the highest variance using the collapseRows R function [24]. Differential expression analysis was performed using the limma package (28,367,255). Preranked gene set enrichment analysis (GSEA) was performed using the fGSEA package [25] on hallmarks pathways (v 7.4), a curated set of lipid metabolism-related Gene Ontology Biological Processes (GOBP, v 7.4) that contains the keywords ‘fatty acid’, ‘lipid’ or ‘triglyceride’ (n = 180), cancer stemness-related signaling pathways [26], and CSC signatures [27,28,29,30] (Supplementary Table S1). The gene set variation analysis (GSVA) function from the GSVA R package (version 1.50.5) was also applied to the INT samples using the same gene sets to calculate enrichment scores. These scores were then correlated using Pearson correlation. We also analyzed GSE114082 from baseline and Tru-cut biopsies of locally advanced HER2 + BCs from patients in the TRUP window-of-opportunity trial who received one cycle of trastuzumab monotherapy [31]. All bioinformatic analyses were conducted using R software version 4.2.1.

Tumor cell lines

The human HER2 + BC cell lines were purchased from ATCC (Rockville, MD, USA). BT474 and MDAMB361 cells were grown as monolayer cultures in DMEM (EuroClone, Pero, MI, Italy) supplemented with 10% fetal bovine serum (FBS) (Sigma‒Aldrich, St. Louis, MO, USA), whereas HCC1954, EFM192A, HCC1569, SKBR3, and ZR75.30 cells were grown as monolayer cultures in RPMI1640 medium (EuroClone) supplemented with 10% FBS. The BT474, MDAMB361, HCC1954, SKBR3 and ZR75.30 human tumor cell lines were obtained between 2000 and 2010 and authenticated by short tandem repeat DNA fingerprinting (Eurofins Genomics, Louisville, KY, USA; last verification, January 2022). The HCC1569 and EFM192A cell lines were purchased from ATCC in 2021. All tumor cell lines were cultured in a humidified atmosphere at 37 °C with 5% CO2 and were routinely tested and confirmed to remain free of mycoplasma contamination. Mycoplasma contamination was tested using the MycoAlert Plus Kit (Lonza, Basel, Switzerland).

Flow cytometry and cell sorting

HER2 + BC cell lines were analyzed and sorted based on CD36 expression after incubation with an allophycocyanin (APC)-conjugated anti-CD36 antibody or a matched/corresponding isotype control antibody in the dark at 4 °C for 40 min [clone: 5–271; isotype control: APC Mouse IgG2a κ (Biolegend, San Diego, CA, USA)]. To perform triple immunofluorescence flow cytometry to identify cells with the CD44High/CD24Low phenotype, we incubated cells with an APC-conjugated anti-CD36 antibody (BioLegend; clone: 5–271; isotype control: APC Mouse IgG2a κ), a fluorescein-5-isothiocyanate (FITC)-conjugated anti-CD44 antibody (BioLegend) and a PE/Cy7-conjugated antibody [BioLegend; clone: ML5; isotype control: phycoerythrin–cyanine 7 (PE/Cy7) Mouse IgG2a, κ] for 40 min at 4 °C. To perform double immunofluorescence FACS analysis of CD44 and CD24 expression, we incubated cells with an APC-conjugated anti-CD44 antibody (BioLegend) and a PE/Cy7-conjugated antibody (BioLegend; clone: ML5; isotype control: PE/Cy7 Mouse IgG2a, κ) for 40 min at 4 °C. Immunofluorescence analyses were performed using a BD FACSCanto™ II flow cytometer (BD Biosciences, Franklin Lakes, NJ, USA) or after sorting with a high-speed FACSAria II sorter (BD Biosciences) to ensure high purity (> 95%). In all analyses, the data were processed using the FlowJo™ software package (BD Biosciences). The relative median fluorescence intensity (rMFI) and the percentage of positively stained cells were determined by gating on live cells and subtracting the background (isotype) MFI. 7-Aminoactinomycin D (7-AAD) Staining Solution (BD Biosciences) was added before analysis to gate out nonviable cells.

The FA uptake assay was performed as previously described [32]. Cells were seeded at a density of 400,000 cells/well in a 6-well plate for 72 h before the experiments and cultured in serum-free medium supplemented with 2 μM 4,4-Difluoro-5,7-Dimethyl-4-Bora-3a,4a-Diaza-s-Indacene-3-Hexadecanoic Acid (BODIPY FL C16; Thermo Fisher Scientific) for 30 min at 37 °C. After washing, the cells were harvested and analyzed via flow cytometry. The MFI was compared between the samples and the internal control condition.

Cell treatments

Single-cell suspensions of HCC1954, MDAMB361, EFM192A and HCC1569 cells were plated and grown in a 6-well plate at a density of 150,000–400,000 cells/well (according to the cell line) for 72 h at 37 °C with 5% CO2. To examine the contribution of FA uptake via CD36 in the formation of HER2 + EMT-like CSCs, HCC1954, MDAMB361, EFM192A and HCC1569 cells were plated and grown in a 6-well plate at a density of 150,000–400,000 cells/well (according to the cell line) and incubated in culture media supplemented with FBS depleted (FBS-FA-) or not (FBS-FA +) of FA for 24 h, 48 h and 72 h. To evaluate the effects of selective WNT inhibition, HCC1954, MDAMB361, EFM192A and HCC1569 cells were treated with the WNT inhibitor LGK974 (Selleckchem, Karl-Schmid-Str.14, Munich, DE) as a single agent at final concentrations of 10 μM and 20 μM (HCC1954, MDAMB361 and EFM192A cells) or 15 μM and 30 μM (HCC1569 cells) or with 0.1% DMSO (Sigma‒Aldrich) as an internal control for 24 h and 48 h at 37 °C with 5% CO2. In parallel, to evaluate the effects of selective WNT activation, HCC1954, MDAMB361 and EFM192A cells were treated with the specific WNT agonist SKL2001 (Selleckchem) as a single agent (10, 20 and 40 µM) or with 0.1% DMSO (Sigma‒Aldrich) as an internal control for 24 h. Both the mRNA and protein expression levels of CD36 were measured. To evaluate the effect of CD36 inhibition on CSC subset maintenance, HCC1954, MDAMB361, EFM192A and HCC1569 cells were treated with the CD36 inhibitor sulfo N-succinimidoyl oleate (SSO) (Cayman Chemical, Ann Arbor, MI, USA) as a single agent (100 µM) for 48 h at 37 °C with 5% CO2.

Sulforhodamine B (SRB) cell viability/proliferation assay

To evaluate the antitumor effects of the different drugs, HCC1569, MDAMB361, HCC1954, EFM192A, SKBR3, ZR75.30 and BT474 cells were seeded in 96-well plates at a density of 3,000–5,000 cells per well (according to the cell line) for 72 h before each treatment to allow attachment to the plastic surface. Next, the cells were treated with an increasing concentration gradient of the anti-HER2 lapatinib (Selleckchem) (from 0.05625 μM to 28.8 μM) diluted in DMSO for 24 h, 48 h and 72 h, the WNT antagonist LGK974 (from 5 μM to 80 μM), the WNT agonist SKL2001 (from 5 μM to 80 μM) or the CD36 inhibitor SSO (from 12.5 μM to 400 μM) for different periods (24 h, 48 h and 72 h). Cell viability was assessed using the SRB assay (Sigma–Aldrich) as previously reported [33]. In brief, after each incubation period, the cell monolayers were fixed with 10% trichloroacetic acid and stained with SRB for 30 min. The excess dye was subsequently removed by repeated washing with 1% acetic acid, after which the protein-bound dye was dissolved in 10 mM Tris solution for optical density (OD) measurement at 564 nm. The number of living cells was proportional to the amount of solubilized dye. Cell proliferation (%) was calculated as [(untreated—treated)/(untreated)] × 100 and plotted using GraphPad Prism 5.02.

Multiplex immunostaining and immunolocalization analyses

Formalin-fixed paraffin-embedded (FFPE) tissue samples from human HER2 + BC patients who did or did not achieve pCR [pCR Yes (n = 18) and pCR No (n = 18), respectively] after neoadjuvant therapy were obtained from the archive of the INT. After deparaffinization and rehydration, an Opal 3-Plex Detection Kit was used (Cat. # NEL810001KT, Akoya Biosciences, Marlborough, MA, USA). Antigen retrieval was carried out in pH 9 buffer (cod. RE7119, Leica Biosystems, Buccinasco, Milan, Italy) by initial boiling at 100% power, followed by 20% power for 15 min, using microwave technology. The sections were treated with blocking buffer for 10 min at room temperature before primary antibody incubation. The slides were then incubated with a polymeric horseradish peroxidase (HRP)‐conjugated secondary antibody for 10 min, and signals were visualized using Opal 520 fluorophore‐conjugated tyramide signal amplification (TSA) reagent at a 1:100 dilution. Polymeric HRP catalyzes the covalent deposition of fluorophores around the marker of interest. The slides were again subjected to microwave treatment to strip the primary/secondary antibody complexes to allow staining of the next antigen with the corresponding antibody. The second round of staining was performed by incubation with the second primary antibody followed by incubation with the polymeric HRP-conjugated secondary antibody, incubation with Opal 620 fluorophore‐conjugated TSA reagent at a 1:100 dilution for signal visualization, and microwaving in antigen retrieval buffer. The following primary antibodies were used: mouse anti-human CD44v6 (clone VFF18, 1:500, cod. AB78960, Abcam, Cambridge, UK), rabbit anti-human ALDH1A1 (1:500, cod. GTX123973, GeneTex, Irvine, CA, USA), and rabbit anti-human CD36 (clone D8L9T, 1:200, cod. #14,347, Cell Signaling Technology, Danvers, MA, USA). For combined immunofluorescence and HER2 immunohistochemistry (IHC) analysis, after neutralization of endogenous peroxidase with 3% H2O2 and Fc blocking with 0.4% casein in phosphate buffer solution (PBS; Novocastra, Newcastle upon Tyne, UK), the same sections were incubated with a rabbit anti-HER2/ErbB2 antibody (clone D8F12, 1:250, cod. #4290, Cell Signaling Technology), followed by the components of a Novolink Polymer Detection System (Novocastra) and 3,3’-diaminobenzidine (DAB; Novocastra) as the substrate chromogen. Nuclei were counterstained with 4',6-diamidino-2-phenylindole (DAPI) and Harris hematoxylin (Novocastra). Staining was analyzed under a Zeiss Axioscope A1 microscope equipped with four fluorescence channels for widefield IF visualization. Micrographs were acquired using a Zeiss Axiocam 503 color digital camera connected to Zen 2.0 software (Zeiss, Oberkochen, Germany). Quantitative analysis of staining for multiple markers was performed via HALO image analysis software (v3.2.1851.229; Indica Labs, Albuquerque, NM, USA) and by evaluating the percentage of costaining in five nonoverlapping fields at medium magnification (200 ×).

Mammosphere forming efficiency (MFE) assay

Dissociated cells were seeded in 6-well ultralow attachment plates (Corning, NY, USA) at a density of 2,500–3,000 cells per well in serum-free MammoCult medium (StemCell Technologies, Vancouver, Canada) and incubated for 7 days as previously reported [34]. In brief, engineered cell lines (MDAMB361 shSCR, shCD36 (1), shCD36 (2), HCC1954-Empty Vector and HCC1954-CD36 +) were treated on day 0 with 0.5 µM lapatinib or 0.1% DMSO. Similarly, HCC1954, MDAMB361, EFM192A and HCC1569 cells were treated on day 0 with 0.5 µM lapatinib and the anti-CD36 inhibitor SSO (50 µM) alone or in combination. The effect of CD36-silencing on the MFE (%) was also evaluated in EFM192A and HCC1569 transiently transfected with siCD36 and siSCR, as internal control, for 72 h and followed by in vitro culture under mammoshere formation-promoting conditions. The mammospheres (MS) were counted microscopically on day 7, and representative images were acquired using an EVOS XL Core Cell Imaging System (Thermo Fisher Scientific) (10 × magnification). The mammosphere forming efficiency (MFE) was calculated as the number of MSs divided by the number of single cells that were initially seeded, and MFE values are expressed as % means (± SDs). The MFE inhibition percentage, i.e., the decrease in the capability to form MSs under 3D culture conditions, was calculated as [(untreated MFE-treated MFE)/(untreated MFE)] × 100, as reported previously [34].

Quantitative real-time PCR (qRT‒PCR)

Total RNA from the tested cells was extracted using QIAzol (QIAGEN) according to the manufacturer’s instructions and as reported previously [35], and RNA purity was verified by measuring the 260/280 absorbance ratio. cDNA was reverse transcribed from 1 µg of total RNA in a 20 µl reaction volume with SuperScript III (Invitrogen, Waltham, MA, USA) using oligo(dT) primers, and expression was measured by qRT‒PCR using an Applied Biosystems SYBR® Green dye-based PCR assay in an ABI Prism 7900HT sequence detection system (Applied Biosystems, Foster City, CA, USA). Human CD36 and catenin beta 1 (CTNNB1) transcripts were amplified using the following primer pairs (200 nM each primer): CD36 Forward: TCATGTCTTGCTGTTGATTTGTGA, CD36 Reverse: TGGTTTCTACAAGCTCTGGTTCTTA; CTNNB1 Forward: GTGCTATCTGTCTGCTCTAGTA, CTNNB1 Reverse: CTTCCTGTTTAGTTGCAGCATC. All the experiments were performed in triplicate. Target gene expression data were normalized to actin beta expression data (ACTB Forward: AGGCATCCTCACCCTGAAG; ACTB Reverse: TCCATGTCGTCCCAGTTGGT). Gene expression analysis was performed using the comparative 2Δ (Ct) method with the housekeeping gene ACTB for normalization.

ALDH activity assay

ALDH activity was measured in the tested cells using an ALDEFLUOR assay kit (StemCell Technologies) following the manufacturer’s instructions. In brief, 7.5 × 105 cells were suspended in ALDEFLUOR assay buffer containing the ALDH substrate BODIPY-aminoacetaldehyde (BAAA) and incubated for 45 min at 37 °C. A specific inhibitor of ALDH, diethylaminobenzaldehyde (DEAB), was used to distinguish the ALDH + and ALDH– cell subsets.

Vector construction, lentiviral particle production and tumor cell infection

The lentiviral vector encoding human CD36 was constructed using a third-generation self-inactivating lentiviral system that, based on four different plasmids, offers maximal biosafety. The backbone used was pRRL-sin-cPPt.CMV-GFP.WPRE (kindly provided by Dr. G. Ferrari, San Raffaele Scientific Institute, Milan), in which the GFP sequence was replaced with the cDNA sequence of human CD36. The cDNA sequence of CD36 was excised from pUC19-CD36 (GenScript, Piscataway, NJ, USA) using the restriction enzymes AgeI and SalI and was then ligated into the viral backbone and verified by sequencing. For CD36 silencing, pLKO.1-puro vectors carrying different short hairpin RNA (shRNA) sequences (TRCN0000057000; TRCN00000437740; TRCN00000419016) against CD36 were purchased from Sigma–Aldrich, along with a control nontargeting shRNA vector (SHC005). Lentiviral particles were produced using standard methods. In brief, lentiviral particles were produced by transient cotransfection of the transfer vector constructs (for either expression or silencing of CD36) with the VSV-G-expressing plasmid pMD.G and the third-generation packaging plasmids pMDLg/pRRE and pRSVRev [36] into 293 T cells. The medium was replaced 20 h later with fresh medium. The supernatant was collected 24 h later and used for target cell infection at a 1:2 concentration ratio in the presence of 8 µg/ml polybrene to facilitate viral entry into the cells. HCC1954 cells were infected with CD36-overexpressing or empty vector lentiviral particles to obtain the HCC1954-CD36 + and HCC1954-Empty Vector cell lines, respectively. MDAMB361 cells were infected with CD36-silencing and scrambled lentiviral particles to obtain the MDAMB361-shCD36 and MDAMB361-shSCR cell lines, respectively. The medium was changed the next day, and after 4 days, the transduction efficiency was evaluated by FACS analysis using an anti-CD36 antibody. For CD36 silencing, transduced cells were selected based on puromycin resistance, and CD36 downregulation was evaluated by flow cytometry.

Transient silencing of CD36 and CTNNB1

For transient silencing of CD36, EFM192A and HCC1569 cells at 75–80% confluence were transfected with specified amounts of CD36 small interference RNA (siRNA) (100 nM; Assay ID: 105938) with Lipofectamine® RNAiMAX Reagent (Thermo Fisher Scientific) according to the manufacturer’s protocol. Cells were also transfected with Silencer™ Negative Control No. 1 siRNA (100 nM) (Thermo Fisher Scientific) following the same protocol for use as negative control cells. After 48 h, the transfected cells were tested for CD36 expression at both the transcriptional and translational levels. For transient silencing of CTNNB1, HCC1954, MDAMB361, EFM192A and HCC1569 cells at 75–80% confluence were transfected with specified amounts of CTNNB1 siRNA (100 nM; Assay ID: 146154) with Lipofectamine® RNAiMAX Reagent (Thermo Fisher Scientific) according to the manufacturer’s protocol. Cells were also transfected with Silencer™ Negative Control No. 1 siRNA (100 nM) following the same protocol for use as negative control cells. After 24 h and 48 h, the transfected cells were tested for CD36 expression at both the transcriptional and translational levels.

Western blot analysis

Protein extracts from HCC1954, MDAMB361, EFM192A, HCC1569, CD36 transiently silenced EFM192A and HCC1569, MDAMB361-shSCR and MDAMB361-shCD36 tumor cells were obtained by incubation for 40 min at 0 °C with RIPA Lysis and Extraction Buffer (Thermo Fisher Scientific) in the presence of Halt™ Protease Inhibitor Cocktail (Thermo Fisher Scientific), mixed with gel sample buffer under reducing conditions, and heated for 5 min at 95 °C. Extracted proteins were separated by electrophoresis on precast 4–12% Bis–Tris gels (Invitrogen, Thermo Fisher Scientific). The separated proteins were electrophoretically transferred onto nitrocellulose membranes, which were stained with Ponceau Red to assess protein loading, washed extensively with TBS + 0.5% Tween 20, and saturated for 1 h at room temperature in blocking solution (5% low-fat milk in TBS + 0.1% Tween 20). Primary antibodies in 3% low-fat milk in TBS + 0.1% Tween 20 were then added, and the membranes were incubated for 1 h at room temperature or overnight at 4 °C with gentle shaking. The following mouse monoclonal primary antibodies were used: anti-β-actin-peroxidase (AC-15 clone; 1:50.000; Sigma‒Aldrich) and anti-c-ErbB2/c-Neu (3b-5 clone; 1:300; Calbiochem, Darmstadt, Germany). The following rabbit monoclonal primary antibodies were used: anti-phospho-HER2/ErbB2 (Tyr1221/1222) (6B12 clone; 1:1000; Cell Signaling Technology) and anti-catenin beta 1 (1:1000; Cell Signaling Technology). The membranes were then washed extensively with TBS + 0.5% Tween 20 and incubated with sheep anti-mouse Ig (1:5.000) (Amersham GE Healthcare, Little Chalfont, UK) or with donkey anti-rabbit Ig (1:10.000; Amersham GE Healthcare) for 1 h at room temperature. Signals were detected using enhanced chemiluminescence (ECL; Amersham GE Healthcare).

In vivo tumorigenicity assay

Mice were maintained under pathogen-free conditions at the animal facility of INT. All procedures were performed with ethical approval as stated in the Declarations (Ethical approval). Established and randomized 8-week-old female severe combined immunodeficient mice (SCID) mice were purchased from Charles River Laboratories (Calco, Italy). Mice (n = 5–6) were injected in a mammary fat pad (m.f.p.) with 9 × 105 engineered lentivirally transduced MDAMB361 cells (MDAMB361-shCD36 or MDAMB361shSCR) pretreated in vitro for 72 h with the anti-HER2 inhibitor lapatinib or diluted DMSO as an internal control. Tumor size was evaluated once weekly by caliper measurement, and the approximate volume of each mass was calculated using the formula 0.5 x d12 x d2, where d1 and d2 are the smaller and larger diameters, respectively. Mice were sacrificed when the tumor volume reached ~ 1000 mm3.

Statistical analysis

Statistical analyses were conducted with GraphPad Prism 5.02 software using unpaired or paired two-tailed Student’s t test as appropriate. When p < 0.05, the difference between the compared groups was considered significant. The data are presented as the mean ± standard error of the mean (SEM) values (n ≥ 3 technical replicates). The sample size used in each individual experiment is reported in the corresponding figure legend. Mouse survival was assessed using the Kaplan–Meier method. Linear regression analysis was performed and the Pearson correlation coefficient (r) was calculated to estimate the correlations between 1) the percentages of HER2 + /CD44v6 + cells and HER2 + /CD36 + cells and 2) the percentages of ALDH1A + cells and CD36 + cells in primary HER2 + diagnostic core biopsies obtained from human patients.