Role of mitochondria and cardiolipins in growth inhibition of breast cancer cells by retinoic acid

The source and the characteristics of the 16 breast-cancer cell-lines used are available in Additional file 1. The generation of the RARα over-expressing (RARA-C5) and relative control (Vect-C1) clones from MDA-MB-453 cells as well as the RARα silenced (RARsA-sh18) and relative control (Vect-C6) clones from SK-BR-3 cells have been described [11]. The growth of cells was determined with the sulforhodamine assay [11].

Single-cell motility assays were performed on BSA-coated culture wells by time-lapse microscopy, using the Imaging Station CellRˆ (Olympus, Segrate, Italy) and the software Image J (Rasband W, National Institutes of Health, Bethesda, MD).

Untargeted lipidomics studies were performed with Lipostar, a high-throughput software supporting targeted and untargeted liquid-chromatography/mass-spectrometry (LC-MS) lipidomics [23, 28]. Further details on the methodological approach can be found in Additional file 1.

Mitochondria were stained with MitoTarcker Deep Red FM (Invitrogen) according to the manufacturer instructions. Following staining, cells were fixed with 2% formalin and subjected to quantitative FACS analysis, using a fluorescence activated cell sorter (FACS, Becton and Dickinson). In the case of the experiments performed on the SK-BR-3, HCC-1419, MDA-MB-361 and HCC-202 lines, mitotracker stained cells were also subjected to quantitative microscopic analysis using the ImageJ software. Cells were counterstained with Hoechst 3342 (Thermofisher) for the determination of cell nuclei. For each experimental point, a minimum of 200 cells/field in at least 4 fields/experimental triplicate were quantitated. Mitochondria were isolated using a described protocol [29]. The enzymatic activity of mitochondrial complexes was determined on isolated mitochondria [30, 31]. The microviscosity of mitochondrial membranes was measured as described by M. Salmona et al. following staining with 1,6-diphenyl-1,3,5-hexatriene as a fluorescent probe [32].

Three paired biological replicates of each breast-cancer cell-line were grown in DMEMF12 medium containing 5% charcolated FBS (Fetal Bovine Serum, Gibco) for 24 h. Cells were treated with vehicle (DMSO) or ATRA (10^− 6 M) for another 24 h. RNA was extracted with the mRNeasy Mini Kit (QIAGEN). RNA sequencing was performed using the Illumina TruSeq RNA library-preparation kit and sequenced on the Illumina NextSeq500 with paired-end, 150 base pair long reads. The overall quality of sequencing reads was evaluated using FastQC [33]. Sequence alignments to the reference human genome (GRCh38) were performed using STAR (v.2.5.2a). Gene-expression was quantified using the comprehensive annotations available in Gencode [34]. Specifically, we used the v27 release of the Gene Transfer File (GTF). Raw-counts were further processed in the R Statistical environment and downstream differential expression analysis was performed using the DESeq2 pipeline. All the RNA-sequencing data relevant for this study were deposited in the EMBL-EBI Arrayexpress database (Accession No: E-MTAB-8408). Genes characterized by low mean normalized counts were filtered out by the Independent Filtering feature embedded in DESeq2 (alpha = 0.05). DESeq2-computed statistics were used as input for gene-set enrichment testing performed with the pre-ranked version of Camera. Statistical enrichments were determined for gene-sets obtained from the Hallmark (H), which are curated by the Molecular Signature DataBase (MSigDB).

ATRA-sensitive SK-BR-3 and ATRA-resistant HCC-1419 cell lines were grown on plastic Petri dishes and fixed in phosphate buffer 0.12 M pH 7.4 containing 4% paraformaldehyde and 2% glutaraldehyde (Electron Microscopy Sciences, code #16220) for 10 min. Cells were scraped from the dishes, centrifuged at 13,000 rpm and kept in 0.12 M phosphate buffer containing 4% paraformaldehyde and 2% glutaraldehyde for 2 h. After post-fixation in 0.12 M cacodylate buffer containing 1% OsO₄ for 2 h and subsequent dehydration in graded series of ethanol, samples were cleared in propylene oxide, embedded in Epoxy medium (Epon 812 Fluka) and polymerized at 60 °C for 72 h. Ultrathin sections (70 nm thickness) were obtained with a Leica EM UC6 ultramicrotome, counterstained with uranyl acetate and lead citrate and examined with an Energy Filter Transmission Electron Microscope (EFTEM, ZEISS LIBRA® 120) equipped with a yttrium aluminium garnet (YAG) scintillator slow-scan charge-coupled device (CCD) camera (Sharp eye, TRS, Moorenweis, Germany). The numerical density of mitochondria (N_V, n/μm³) was estimated by morphometrical analysis using 30 digitized electron microscope fields of cells for each group acquired by the iTem software (Olympus Soft Imaging Solutions, Germany) and digitally superimposing an orthogonal grid with ImageJ (1.52a version). Briefly, the mitochondrial profile area density (N_A) was estimated by the ratio between the number of mitochondria and the cytoplasmic area. Mitochondrial volume density (V_V) was determined by the ratio of grid points falling over mitochondria divided by the total number of points of the grid contained in the cytoplasm. The numerical density of mitochondria (N_V) was then estimated for each cell using the formula: N_V = (1/β) (N_A ^3/2 / V_V ^1/2), where β is the shape coefficient for ellipsoidal mitochondria, calculated from the ratio of the armonic mean of major and minor axis of mitochondria sections measured on digital images. The mean mitochondrial volume was calculated for each cell as the ratio of mitochondrial volume density V_V and numerical density N_V.

The microviscosity of mitochondrial membranes was measured as described by M. Salmona et al. following staining with 1,6-diphenyl-1,3,5-hexatriene as a fluorescent probe [3]. Isolated mitochondria containing an equal amount of proteins were incubated with 1,6-diphenyl-1,3,5-hexatriene (2 × 10^− 6 M) for 30 min at 37° and the fluorescence polarization values were determined using a fluorescence detector (Infinite F500, TECAN, Switzerland). The fluorescence polarization (FP) value is a function of the emission value (emission = 420 nm), which was detected through analyzers oriented in parallel (FP1) and perpendicular (FP2) to the direction of polarization of the excitation beam (excitation = 365 nm), according to the eq. FP = (FP2 - FP1 / FP2 + FP1).

To conduct the study, we used 16 cell-lines whose degree of sensitivity to the anti-proliferative action of ATRA has been determined [11, 12]. Eight cell-lines come from triple-negative (TN) breast-cancers, while the other 8 cell-lines derive from luminal tumors characterized for HER2 and ER/PR expression. Consistent with the tumor origin, the constitutive gene-expression profiles determined by RNA-sequencing classify the cell-lines in two distinct basal and luminal groups (Fig. 1a). The 16 cell-lines are ranked according to their quantitative response to the anti-proliferative effect of ATRA using the continuous ATRA-score index (Fig. 1b) [11]. One basal (HCC-1599) and 4 luminal (SK-BR-3, HCC-1500, CAMA1 and MDA-MB-361) cell-lines are classified as highly sensitive to ATRA. Three luminal (HCC-202; MDA-MB-175VII; ZR75.1) and 3 basal (MB-157; MDA-MB-157; HS578T) cell-lines are endowed with intermediate sensitivity. One luminal (HCC-1419) and 4 basal (MDA-MB-231; CAL-851; HCC-1187; MDA-MB-436) cell-lines show low sensitivity/resistance to ATRA.

We used a LC/MS-based approach [23] to define the lipidomic profiles of each cell-line grown under basal conditions. We generated a fingerprint composed of 530 chemical features identified as lipid species, some of them in multiple adduct forms. Lipid species were grouped in 23 classes according to their chemical structures. Each class contains a different number of chemical species (Fig. 2a and Additional file 2: Table S1). For instance, Diacylglycerophosphocholines (PCs) consist of 203 features, while 40 cardiolipins (CLs) are identified. PCA of the constitutive lipid profiles separates luminal from basal cell-lines (Fig. 2b). Thus, the two cell-line groups are characterized not only by different gene-expression patterns, but also by different complements of lipids. Compared to the luminal counterparts, basal cell-lines show greater amounts of many TGs, diacylglycerophosphoserines or phosphatidylserines (PSs) and PCs, some diacylglycerophosphoinositols or phosphatidylinositols (PIs), neutral glycosphingolipids (N-glyco-SPs), CLs, lysophosphatidylcholines (LPCs), sphingomyelins (SMs) and 1-alkyl-2-acylglycerophosphocholines/1-alkenyl-2-acylglycerophosphocholines (o-PCs/ p-PCs), as well as all the identified steryl-esters (SEs), monoacylglycerophosphoethanolamines (lysophosphatidylethanolamines, LPEs) and 1-alkenyl-phosphatidylethanolamine (p-PE) (Fig. 2c). In contrast, the levels of 3 triacylglycerols (TGs), 2 PCs, 1 p-PC, 1 ceramide (CER) and 1 diacylglycerophosphoethanolamine (phosphatidylethanolamine, PE) are higher in luminal than basal cell-lines. Although luminal and basal cell-lines show different complements of individual features, the average constitutive levels of the 23 classes of lipids do not vary in the two groups (Additional file 1: Figure S1 and Additional file 2: Table S1).

We compared the mass-spectrometry data obtained in all the cell-lines exposed to DMSO or ATRA (10^− 6 M) for 48 h. The selection of the retinoid concentration and exposure time is based on pilot studies conducted in SK-BR-3 cells. These studies demonstrate that maximal alterations in the lipidomic profiles are observed with 10^− 6 M ATRA, while the 48-h exposure time precedes any significant effect on the number of cells [11].

ATRA causes significant alterations of the lipidomic profiles in all the cell-lines considered (PCA, Additional file 1: Figure S2 and Additional file 2: Table S1). The retinoid determines the largest perturbations in cell-lines characterized by high/intermediate ATRA-sensitivity, regardless of the luminal or basal phenotype. Indeed, the effects triggered by ATRA in low-sensitivity/resistant cells are less evident, which is consistent with an association between the changes in the lipidomic profiles and the retinoid-dependent anti-proliferative effects (Additional file 1: Figure S2). We evaluated the effects exerted by ATRA on each of the 23 classes of lipids identified (Additional file 1: Figures S3-S5). For the majority of the classes, ATRA either triggers variable and cell-specific alterations in the total levels of the corresponding lipid components or causes no perturbation at all (Additional file 1: Figures S6-S8). The most interesting effects of ATRA are observed with N-glycoSPs, LPCs, PSs (Fig. 3) and CLs (Fig. 5).

N-glyco-SPs are a group of cell-membrane sphingolipids and participate in biological processes such as cell adhesion and cell-cell interactions [35]. With the exception of retinoid-resistant HCC-1419 cells, ATRA increases N-glyco-SPs levels in all luminal cell-lines (Fig. 3, left). HCC-1187 is the only basal cell-line where a similar effect is observed. Hence, the ATRA-dependent increase of N-glyco-SPs is specific to luminal cells and it is not associated with the growth-inhibitory action exerted by ATRA in this breast cancer cell type. The point is illustrated by the low R² correlation index calculated for the ATRA/DMSO ratio of the N-glyco-SP mean values determined in each cell-line and the corresponding ATRA-Score (Fig. 3, right).

LPCs result from phospholipase A2-dependent partial hydrolysis of phosphatidylcholines, which removes one of the fatty acid groups [36]. Similar to what is observed with N-glycoSPs, there is a trend towards a retinoid-dependent increase in the levels of LPCs in all luminal cell-lines, regardless of their sensitivity to ATRA (Fig. 3, left). The sole exception is represented by the ZR75.1 cell-line, which shows no alteration in the amounts of LPCs following exposure to ATRA. The effects observed in basal cell-lines are variable and devoid of any association with the ATRA-dependent anti-proliferative action. Taken together the data suggest that the perturbations afforded by ATRA on LPC levels are not involved in the anti-proliferative action of the retinoid in either luminal or basal cell-lines. The conclusion is supported by the relatively low R² correlation index between the ATRA/DMSO ratio of the LPC mean values and the ATRA-score (Fig. 3, right). The LPCs up-regulation observed in luminal cells may be related to ATRA metabolism and bio-disposition. In fact, LPCs are by-products of the transacylation reaction catalyzed by lecithin retinol acyl transferase (LRAT) which results in the conversion of all-trans retinol into all-trans retinyl ester, a storage form of Vitamin A [37].

PSs are cell membrane glycerophospholipids differing in their fatty acid composition and they are involved in cell-cycle signaling and apoptosis [38, 39]. In the four luminal and highly retinoid-sensitive SK-BR-3, HCC-1500, CAMA1 and MDA-MB-361 cell-lines, ATRA causes a reproducible reduction in the total amounts of the PSs identified (Fig. 3, left). With the exception of ZR75.1, a similar effect is not observed in the other luminal cell-lines characterized by intermediate/low sensitivity to the growth-inhibitory action of ATRA. HCC-1599 is the only basal cell-line showing an ATRA-dependent decrease in PSs. The data obtained result in a relatively elevated R² coefficient of correlation between the ATRA/DMSO ratio of the PS mean values determined in each cell-line and the corresponding ATRA-Score (Fig. 3, right). Taken together, our results suggest that the retinoid-dependent alterations in the levels of PSs play a role in the anti-proliferative effect exerted by ATRA predominantly in the context of luminal breast cancer cells. However, this hypothesis needs to be supported by further functional studies.

Given the potential significance of N-glyco-SPs in cell adhesion and cell-cell interactions, the retinoid-dependent up-regulation of this particular subset of lipids in luminal breast cancer cells may have implications for the anti-metastatic and anti-motility effect exerted by ATRA in this type of mammary tumor [27]. To support this idea, we performed studies on the effects exerted by ATRA on the random-motility of four cell lines. To this purpose, we selected the luminal MDA-MB-361 and MDA-MB-175VII cell lines, which are characterized by N-glyco-SP up-regulation upon ATRA exposure, as well as the luminal HCC-1419 and the basal MDA-MB-157 cells, which do not show a retinoid-dependent induction of the same type of lipids. Cells were pretreated with ATRA (10^− 6 M) or vehicle (DMSO) for 16 h and subjected to a random-motility assay for another 24 h (Fig. 4). ATRA causes a significant reduction in the motility of MDA-MB-361 and MDA-MB-175VII cells, while the retinoid does not alter the motility of HCC-1419 and MDA-MB-157 cells. In addition, we previously demonstrated [27] that ATRA reduces the directional motility of SK-BR-3, a luminal cell line whose N-glyco-SP levels are up-regulated by ATRA, although the up-regultaion does not reach statistical significance in our experimental conditions (Fig. 3).

The most interesting pattern of lipid perturbations afforded by ATRA involves CLs, a set of glycerophospholipids consisting of two phosphatidic acid moieties connected to a glycerol backbone (Additional file 1: Figure S9). Given the presence of 4 distinct alkyl chains, the potential complexity of CL individual species is enormous. However, in most animal tissues, CLs contain 18-carbon fatty alkyl chains each characterized by 2 unsaturated bonds [40]. Our lipidomic analysis identifies 40 CL species whose constitutive levels vary in each cell-line and are differentially modulated by ATRA (Additional file 2: Table S1). Across our panel of cell-lines, CLs show the highest levels of correlation with ATRA anti-proliferative effects. In fact, ATRA reduces the overall amounts of CLs in the luminal and highly sensitive, SK-BR-3, HCC-1500, CAMA1 and MDA-MB-361 cells. A similar ATRA-dependent reduction is observed in basal HCC-1599 and MDA-MB-157 cells (Fig. 5a, left). In addition, the ATRA/DMSO ratio of the CL mean-values in each cell-line is inversely correlated with the corresponding ATRA-scores and shows a high R² index (Fig. 5a, right). The observed association supports the idea that the reduction in the overall amounts of CLs contributes to the anti-proliferative action of ATRA in both luminal and basal cell-lines. We evaluated the time and concentration dependence of the CLs decrease induced by ATRA in luminal and retinoid-sensitive SK-BR-3 cells (Additional file 3: Table S2). In this cellular context, the effect of ATRA is relatively late and long-lasting, as a significant reduction in CL levels is evident only at 24 h and it is maintained until 72 h (Fig. 5b). In addition, CLs down-regulation is concentration dependent, as indicated by the results obtained following exposure of SK-BR-3 cells to increasing amounts of ATRA (Fig. 5c).

We focused our attention on the specific down-regulation of CLs afforded by ATRA in sensitive cell-lines. CLs are part of the composite glycerophospholipid metabolic pathway (https://​www.​genome.​jp/​kegg/​pathway.​html). To obtain insights into the mechanism underlying the ATRA-dependent down-regulation of CLs, we evaluated the RNA-sequencing data obtained in all the 16 cell-lines exposed to ATRA (10^− 6 M) or vehicle for 24 h (M. Bolis et al., Array Express ref. No: E-MTAB-8408). Using these data, we determined the effects of ATRA on the expression of the 77 genes involved in the glycerophospholipid metabolic pathway (Additional file 1: Figure S10).

Although specific genes are significantly up- or down-regulated by ATRA in single cell-lines (for instance see the expression pattern of MBOAT1, Additional file 1: Figure S10), there is no significant correlation (low R² values) between the ATRA/DMSO expression ratio of any of these genes and the ATRA-score values. The results support the idea that the ATRA-dependent CLs decrease observed in retinoid sensitive cell-lines cannot be explained by up-regulation of specific genes controlling the catabolism of CLs or down-regulation of genes involved in their biosynthesis. CLs are predominantly located in the inner membrane of mitochondria, where they are involved in oxidative phosphorylation [41]. This suggests that the anti-proliferative effect exerted by ATRA in sensitive cell-lines may be accompanied by perturbations in mitochondrial homeostasis. Pathway enrichment analysis of the RNA-sequencing data supports this idea. In fact, “Oxidative Phosphorylation” lays in third position among the 4 top Hallmark gene-sets enriched for genes down-regulated by ATRA in sensitive luminal and basal cell-lines (Fig. 6a). The other 3 top gene-sets collectively down-regulated by ATRA are “Myc Targets”, “E2F Targets” and “G2M Checkpoint”. The down-regulation of these gene-sets, which control cell-cycle and proliferation, is likely to be associated with the growth-inhibitory action exerted by ATRA in sensitive cell-lines.

The “Oxidative Phosphorylation” gene-set consists of 135 genes coding for mitochondrial proteins [42]. The vast majority of mitochondrial proteins are encoded by nuclear genes. Oxidative Phosphorylation is responsible for the production of ATP from NADH via the activity of 5 large protein complexes (mitochondrial complexes I-V) situated in the inner mitochondrial membrane. ATRA down-regulates approximately half of the “Oxidative Phosphorylation” genes in the luminal and basal cell-lines characterized by high/intermediate sensitivity to the retinoid (see black square brackets in Fig. 6b). In these cell-lines, the level of down-regulation correlates with retinoid-sensitivity. Indeed, the highest downregulation is observed in basal HCC-1599 and luminal SK-BR-3 cells, which are characterized by the 2 top ATRA-scores. With the exception of ND6, all the genes down-regulated by ATRA in sensitive cells are of nuclear origin. Down-regulation of the “Oxidative Phosphorylation” gene-network by ATRA is a relatively early event, as indicated by the RNA-sequencing data obtained in HCC-1599 cells exposed to ATRA (10^− 6 M) for 8 h (Fig. 6b). In fact, a significant decrease in the levels of most of the down-regulated genes is already evident at this time point. This is consistent with the idea that ATRA causes a rapid transcriptional repression of various nuclear genes coding for mitochondrial proteins.

Given the observed down-regulation of CLs and the expression of multiple genes involved in oxidative phosphorylation, we evaluated the action of ATRA on the amounts and function of mitochondria in selected luminal cell-lines characterized by sensitivity and resistance to the anti-proliferative effects of the retinoid. We focused our attention on two couples of homogeneous luminal and HER2⁺ cell-lines (SK-BR-3/HCC-1419 and MDA-MB-361/HCC-202). Indeed, PCA demonstrates that the constitutive lipidomic profiles of the ATRA-sensitive SK-BR-3 and the ATRA-insensitive HCC-1419 cell-lines are very similar (Fig. 2b). Similar constitutive lipidomic profiles are also observed in ATRA-sensitive MDA-MB-361 cells and the much less responsive HCC-202 counterparts.

To determine the amounts of mitochondria, SK-BR-3 and HCC-1419 cells were exposed to ATRA (10^− 6 M) for 24, 48 and 72 h, stained with Mitotracker and subjected to quantitative fluorescence microscopy (Fig. 7a and Fig. 7c). In SK-BR-3 cells, ATRA causes a significant reduction in Mitotracker-associated fluorescence, which is already evident at 24 h and it is maintained at 48 and 72 h (Fig. 7a). The results obtained at 48 h are confirmed by FACS analysis of the Mitotracker-stained cells (Fig. 7b, left). These data are consistent with an ATRA-triggered decrease in the amounts of mitochondria. The phenomenon is validated by measurement of the total amounts of mitochondrial proteins (Fig. 7b, right). The decrease in the amounts of mitochondria caused by ATRA in SK-BR-3 cells is dose-dependent and the phenomenon is already evident at 10^− 8 M ATRA (Fig. 7e). The reduction in the number of mitochondria is likely to be one of the mechanisms at the basis of the anti-proliferative action exerted by ATRA, as indicated by the results obtained in HCC-1419 cells. In fact, exposure of this ATRA-insensitive cell-line to the retinoid causes an early and paradoxical increase in Mitotracker-associated fluorescence, which is observed at 24 h and reverts to baseline by 48 h (Fig. 7c). In addition, FACS analysis shows no difference between HCC-1419 cells exposed to vehicle or ATRA for 48 h (Fig. 7d, left) and the amounts of total mitochondrial proteins are not altered by the retinoid (Fig. 7d, right). The results of the SK-BR-3/HCC-1419 cell pair are consistent with what is observed in the MDA-MB-361/HCC-202 counterpart exposed to 10^− 6 M ATRA for 48 h. In fact, the data obtained with the use of quantitative immuno-histochemical (Fig. 7f/g, lower leftmost diagrams) and FACS (Fig. 7f/g, lower rightmost diagrams) analyses demonstrate a reduction in the amounts of mitochondria only in the sensitive MDA-MB-361.

To validate and extend the data obtained on mitochondria, we performed quantitative morphological studies on these organelles following isolation from the retinoid-sensitive SK-BR-3 and the retinoid-resistant HCC-1419 cell lines exposed to vehicle or ATRA (10^− 6 M) for 48 h (Fig. 8). The transmission electron microscopy results obtained confirm that ATRA causes a reduction in the number of mitochondria in SK-BR-3 cells, as indicated by the significant decrease in the numerical density of these organelles. Conversely, the retinoid does not alter the number of mitochondria in HCC-1419 cells (Fig. 8, left column graphs). In addition, ATRA exerts no significant effect on the size of mitochondria in either cell line (Fig. 8, right column graphs). Finally, the ATRA-dependent reduction in the amounts of mitochondria observed in SK-BR-3 cells is unlikely to be due to an increase in the degradation of these organelles by mitophagy following retinoid-dependent cell damage or stress [43‐45]. In fact, the electron microscopy studies performed in SK-BR-3 and HCC-1419 cells demonstrate the presence of intact and morphologically fit mitochondria in both cell types regardless of ATRA treatment (Fig. 8, electron micrographs). By converse, down-regulation of numerous nuclear genes involved in oxidative phosphorylation (Fig. 6) suggests that ATRA reduces the assembly of mitochondria.

To evaluate whether ATRA exerts significant effects not only on the amounts, but also on the functional activity of mitochondria, we isolated these organelles from SK-BR-3 and HCC-1419 cells treated with vehicle or ATRA (10^− 6 M) for 48 h. We measured the enzymatic activity of respiratory complexes I-V following normalization for the levels of citrate synthase activity (Fig. 9). In sensitive SK-BR-3 cells, ATRA decreases the enzymatic activity of mitochondrial complexes I, III, IV and V. Similar effects are not observed in ATRA-insensitive HCC-1419 cells. Interestingly, treatment of SK-BR-3 cells with ATRA reduces the microviscosity of mitochondrial membranes (Additional file 1: Figure S11), a parameter controlled by CLs, which are known membrane rigidifying agents [46]. The ATRA-dependent decrease in microviscosity starts to be observed at 24 h and it is maintained at 48 h. The time course of the reduction in microviscosity and the decrease in CL levels is similar, suggesting that the two phenomena are associated. The decrease in microviscosity indicates that ATRA causes a relative deficit of mitochondrial CL levels which is likely to be at the basis of the retinoid-dependent reduction in the activity of complexes I, III, IV and V.

The involvement of CLs down-regulation and decreased assembly/function of mitochondria in the sensitivity of breast-cancer cells to ATRA is supported by the data obtained in RARα over-expressing and RARα knock-down cells [11].

In a first set of studies, we took MDA-MB-453 cells stably transfected with a plasmid construct allowing the expression of RARα (RARA-C5) or the corresponding void vector (Vect-C1) [11]. As expected, RARA-C5 cells are characterized by higher RARα expression levels than the Vect-C1 counterparts and the nuclear retinoid receptor is functionally active as indicated by its ATRA-dependent degradation (Fig. 10a, right), a phenomenon associated with transcriptional activation [47, 48]. In line with what is observed in parental MDA-MB-453 cells [11], which are characterized by low RARα levels and ATRA-resistance, Vect-C1 cells do not respond to the retinoid (Fig. 10a, left). In contrast, RARα over-expression renders RARA-C5 cells partially responsive to ATRA growth-inhibitory action [11]. The lipidomic data demonstrate that exposure of the two cell-lines to ATRA (10^− 6 M) for 48 h does not alter the overall levels of CLs in Vect-C1 cells, while it causes a reproducible decrease in RARA-C5 cells (Fig. 10c and Additional file 4: Table S3). In the same experimental conditions, ATRA reduces the number of mitochondria in Mitotracker stained RARA-C5, but not in Vect-C1 cells (Fig. 10d). In addition, ATRA reduces the microviscosity of mitochondrial membranes only in retinoid-responsive RARA-C5 cells (Fig. 10d, black and white column graph).

We conducted similar types of studies in SK-BR-3 cells stably infected with a retroviral construct containing a RARα-targeting shRNA (RARA-sh18) and the corresponding void vector (Vect-C6) [11]. As expected, RARA-sh18 are characterized by lower levels of functionally active RARα than the corresponding Vect-C6 control cell-line (Fig. 10b, right). Consistent with the high ATRA-score of parental SK-BR-3 cells (Fig. 1b) [11], the growth of Vect-C6 cells is decreased by ATRA in a dose-dependent fashion (Fig. 10b, left). RARα knock-down reduces the sensitivity of RARA-sh18 cells to ATRA anti-proliferative effects. Similar to parental SK-BR-3 cells (Fig. 5a), exposure of Vect-C6 to ATRA (10^− 6 M) for 48 h decreases CL levels (Fig. 10e, Additional file 4: Table S3). A decrease in the total amounts of CLs is not observed in RARA-sh18 cells, where the retinoid causes an opposite effect (Fig. 10e). In the same experimental conditions, ATRA reduces the number of mitochondria in Vect-C6 control cells, but not in RARA-sh18 cells (Fig. 10f). The reduction of mitochondria in Vect-C6 is accompanied by a decrease in mitochondrial membrane microviscosity, while this parameter is slightly increased in RARA-sh18 cells (Fig. 10f, black and white column graph).

All these results indicate that the ATRA-induced decrease in CLs and the associated reduction in mitochondrial membrane microviscosity are dependent on RARα, which is a crucial determinant of ATRA anti-tumor activity in breast cancer [11].