Stable knockdowns of ELOVL5 and IGFBP6 genes were performed using RNA interference (Schwankhaus et al., 2014; Nikulin et al., 2018; 2021; Maltseva et al., 2020). DNA oligonucleotides selected for the target sequences in the ELOVL5 and IGFBP6 genes were ligated into the pLVX shRNA1 lentiviral vector (Clontech Laboratories) according to the manufacturer’s protocol. To obtain the control MDA-MB-231 cells, we used the same lentiviral vector pLVX shRNA1 containing shRNA to the Photinus pyralis firefly luciferase gene. Viral particles were obtained in the form of cell-free supernatants using transient transfection of HEK-293T cell line according to the previously described method (Weber et al., 2010; 2012). Supernatants were collected 24 h after transfection, filtered using .45 µm syringe filters and stored at −80°C. Then, 5∙10⁴ MDA-MB-231 cells were cultured in the wells of a 24-well culture plate in 0.5 mL of cell culture medium. After 24 h, 10 μL of the supernatant containing viral particles was added to the wells, and the plate was placed in a cell culture incubator for 24 h. Then the cell culture medium was changed and the cells were incubated for another 24 h. After that, the selection with 1 μg/mL puromycin (Gibco) was carried out for 2 weeks.

Cells were cultured in a complete culture medium consisting of DMEM 4.5 g/L glucose (Gibco) supplemented with 10% vol. fetal bovine serum (Gibco) and 1% vol. antibiotic-antimycotic solution (Gibco). The cells were incubated in a cell culture incubator at 37°C, 5% CO₂ (Sanyo). Subcultivation was performed every 2–3 days using trypsin-EDTA solution (PanEco). Cells were counted after trypan blue (Gibco) staining using Countess automated cell counter (Invitrogen) according to the manufacturer’s protocol.

We analyzed further the transcriptomic data that we have previously published (Nikulin et al., 2021) and deposited it as GSE165854 dataset. GSE165854 contains Human Transcriptome Array 2.0 microarray (Affymetrix) data for MDA-MB-231 cells with shRNA mediated knockdown of either ELOVL5 or IGFBP6 genes and for control MDA-MB-231(LUC) cells. ANOVA with Benjamin-Hochberg correction for multiple comparisons were used to access statistical significance of the differences observed between these cell lines.

The following datasets (Supplementary Table S1) from GEO (Gene Expression Omnibus) were used for correlation analysis: GSE102484 (Cheng et al., 2017), GSE22220 (Camps et al., 2008), GSE3494 (Miller et al., 2005), GSE58644 (Miller et al., 2005), GSE6532 (Loi et al., 2008). We also used data obtained by the METABRIC consortium (Cerami et al., 2012; Curtis et al., 2012) and The Cancer Genome Atlas (TCGA) program (Weinstein et al., 2013).

TAC 4.0 software (Thermo Fisher Scientific) was applied to preprocess raw data from Affymetrix microarrays. To carry out correlation analysis and statistical data processing, we employed the R 4.1 programming language with the RStudio 1.4 integrated development environment. The values of the Pearson correlation coefficient R and the p-values (the significance of the difference of R from zero) were calculated using the “cor.test” function. Correction for multiple comparisons was performed with the Benjamini-Hochberg method. The correlation coefficients with p-values less than .05 were considered significant.

Proteomic data from PXD023892 dataset was analyzed as well. NanoHPLC-MS/MS system coupled with a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific) was utilized to measure expression of proteins in the cells. Student’s t-test with Benjamin-Hochberg correction for multiple comparisons were used to access statistical significance of the observed changes. Detailed procedure was described earlier (Nikulin et al., 2021).

Real-time PCR was used to confirm the changes in the expression of individual genes. RNA isolation was performed using miRNeasy Micro Kit (Qiagen) according to the manufacturer’s protocol. RNA concentration was measured with NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific). Reverse transcription of RNA was performed using the MMLV RT kit (Evrogen) according to the manufacturer’s protocol. The obtained cDNA samples were stored at −20°C. qPCRmix-HS SYBR (Evrogen) was used for RT-PCR performed with DTprime detecting amplifier (DNA Technology).

The oligonucleotide primers used for RT-PCR were designed based on the mRNA sequences of the studied genes from the UCSC Genome Browser database (Kent et al., 2002). Primer selection was performed using Primer-BLAST software (Ye et al., 2012). The possibility of the formation of secondary structures (hairpins), homo- and heterodimers by the primers was assessed using OligoAnalyzer 3.1 software (Owczarzy et al., 2008). EEF1A1 and ACTB were selected as reference genes (Maltseva et al., 2013). The sequences of the primers are presented in Supplementary Table S2. The evaluation of the differences in the expression of the selected genes was carried out using the software REST 2009 v.2.0.13 (Pfaffl et al., 2002; Vandesompele et al., 2002). For each group 3 independently obtained samples of RNA were used to assess expression levels of the selected genes.

Analysis of the enriched biological processes among the genes with increased and decreased expression was carried out using Gene Ontology (GO) database (Ashburner et al., 2000; Gene and Consortium, 2019) and “topGO” package for R programming language. The results were obtained using “weight01″ algorithm, p-values were calculated using Fisher’s exact test.

To analyze the composition of intracellular fatty acids, the cells were cultured in culture flasks with surface area of 25 cm² in 5 mL of complete nutrient medium for 48 h. Samples containing pellets of 1.5 × 10⁶ million cells were prepared for HPLC-MS analysis following a protocol based on a previously published method (Valianpour et al., 2003). The experiment was carried out in two biological replicates.

To study the kinetics of absorption of LС-PUFAs by the cells from the nutrient medium, 2 × 10⁵ cells were seeded into the wells of 24-well culture plates in 500 µL of complete cell culture medium. The plates were then incubated in a cell culture incubator at 37°C, 5% CO₂ (Sanyo) for 24 h. After that, the nutrient medium was replaced with a medium containing 50 μM of one of the studied LC-PUFAs (linoleic acid: LA, α-linolenic acid: ALA, arachidonic acid: AA, and docosahexaenoic acid: DHA) and the plates were incubated in the cell culture incubator until the end of the experiment. Sampling of the nutrient medium for analysis of the concentration of LC-PUFAs was carried out after 2, 4, 9, 22 and 27 h The experiment was carried out in two biological replicates.

To analyze the content of free fatty acids in the selected samples, each of them was combined 1:1 with methanol, thoroughly mixed by vortexing, and incubated at room temperature for 5 min. Next, the samples were centrifuged at 13,000 g for 5 min and the supernatant was transferred into chromatographic vials for analysis.

The composition of fatty acids in the samples was analyzed by HPLC using an Infinity 1200 chromatograph (Agilent Technologies) with a UV detector. Separation of fatty acids was carried out on a ZORBAX Eclipse XDB-C18 reverse-phase chromatographic column (length 150 mm, inner diameter 4.6 mm, sorbent particle diameter 5 µm) (Agilent Technologies), in the gradient elution mode (Supplementary Table S3). Phase A was deionized water with 0.1% vol. formic acid and phase B was acetonitrile with 0.1% vol. formic acid. The flow rate of the mobile phase was constant and equal to 1.5 mL/min. Fatty acid chromatograms were recorded with the UV detector at 194 nm. Concentrations of the analyzed PUFAs in the samples were determined based on the external calibration curves (Supplememtary Figure S1).

GPx4 activity analysis was performed with the Colorimetric Glutathione Peroxidase Assay Kit (Abcam) according to the standard protocol. Cell suspensions of 2 × 10⁶ were used for each test. GPx4 activity was assessed by measurement of the decrease in NADPH absorbance at 340 nm with X-mark plate reader (BioRad). One unit is defined as the amount of enzyme that will cause the oxidation of 1.0 µmol of NADPH to NADP+ under the assay kit condition per minute at 25°C.

The cells were seeded into 96-well plates (20,000 cells/well) and incubated overnight. After 24 h, PUFAs in different concentrations were added into the wells and the plates were incubated for 4 h. Next, the culture medium was removed from the wells, the cells were washed with DPBS, and solution of the dye-indicator of reactive oxygen species H2DCFDA (cell-permeant 2′,7′-dichlorodihydrofluorescein diacetate, final concentration 10 µM) was added, and the plates were incubated for 30 min. After 30 min, the dye was removed, the wells were washed with DPBS and fresh DPBS was added into each well. Then integral fluorescence signal (excitation at 485 nm, emission at 535 nm) was measured on the SpectraMax i3 Plate Reader (Molecular Devices). Photomicrographs of cells were obtained using an inverted fluorescent microscope Axio Observer Z1 (Carl Zeiss).

MDA-MB-231 cells were seeded in a 96-well plate (15∙10³ cells per well), grown overnight and then cultured with 50 μM of various LC-PUFAs for 24 h.

For lipid droplets staining by Oil Red O, cells were washed twice with PBS, fixed with 10% formalin for 45 min, and then rinsed twice in water for 1 min, followed by 5 min in 60% isopropanol. The cells were stained with Oil Red O (1.8 mg/mL in 60% isopropanol) for 15 min and rinsed 5 times with ddH₂O to remove excess stain. Oil Red O stained cells were directly visualized and imaged using an inverted fluorescent microscope Axio Observer Z1 (Carl Zeiss). Quantification of lipid accumulation was achieved by Oil Red O extraction with 100% isopropanol and gentle agitation for 5 min at room temperature. Then the extracts were transferred to a new 96-well plate and absorbance was measured at 492 nm using X-mark plate reader (BioRad).

For lipid droplet staining with BODIPY, cells were washed twice with DPBS, fixed with 10% formalin for 45 min and then washed twice with DPBS for 1 min. Cells were incubated with 2 μM BOBIPY (Lumiprobe) in the dark at 37°C for 60 min and then washed 3 times with DPBS to remove excess dye. Cells stained with BOBIPY were visualized directly with an inverted fluorescent microscope Axio Observer Z1 (Carl Zeiss).

The viability of tumor cells in all cases was determined using the CellTiter 96® AQueous One Solution (Promega) according to the manufacturer’s recommendations. Optical density was measured at 490 nm with X-mark plate reader (BioRad). All analyzed LC-PUFAs were dissolved in ethanol and all other small molecules were dissolved in DMSO.

For the experiments with 2D cultures, 1 × 10⁴ MDA-MB-231 cells per well were seeded into 96-well plates and incubated (37°C, 5% CO2) for 24 h with a tested compound or a combination of compounds. Then the viability was measured. Working concentrations of Erastin and Ferrostatin-1 were choosen based on previously published data (2–80 μM for Erastin and 0.1–2 μM for Ferrostatin-1) (Dixon et al., 2014; Yu et al., 2019; Anthonymuthu et al., 2021; Li et al., 2021; Chen et al., 2022; Wang et al., 2022).

To generate spheroids, MDA-MB-231 cells were cultured in Matrigel Growth Factor Reduced (GFR) Basement Membrane Matrix (Corning) in 24-well plates for 48 h. Then the spheroids were diluted in Matrigel Growth Factor Reduced (GFR) Basement Membrane Matrix (Corning) and seeded into 96-well plate. After solidification of the gel, complete cell culture medium was added into each well.

After 24 h, the cell culture medium was replaced with the control medium or the medium containing single standard of care (SOC) drugs. Clinically relevant concentrations were used in the assay: 371.7 μM for 5-FU, 5.47 μM for Docetaxel, 6.73 μM for Doxorubicin, 89.3 μM for Gemcitabine, 1.75 μM for Methotrexate, and 811 nM for Vinorelbine (Liston and Davis, 2017). Then the cells were incubated for 3 h in cell culture incubator (37°C, 5% CO2), and the medium was replaced with fresh complete cell culture medium or the medium containing Erastin (1 μM) or DHA (100 μM). Then plates were incubated in cell culture incubator (37°C, 5% CO2) for 72 h, and relative number of viable cells was measured. Growth rate of cancer cells was calculated as described previously (Hafner et al., 2016).

ANOVA with Tukey post hoc test has been used to assess statistical significance of the changes in viability or growth rate. The differences were considered statistically significant if p-values were less than 0.05.

To study the activation of apoptosis Dead Cell Apoptosis Kit with Annexin V Alexa Fluor ™ 488 & Propidium Iodide (PI) (Thermo Fisher Scientific) was used according to the manufacturer’s instructions. Cells were treated with different PUFAs then detached and analyzed as described earlier (Nikulin et al., 2021). Analysis of raw data was carried out using FlowJo 10.6.1 software. Chi-squared test with Bonferroni correction for multiple comparisons were used to access statistical significance of the changes in different populations.

As one gene from our prognostic pair, ELOVL5, is directly involved in the elongation of fatty acids, we reanalyzed our previously published transcriptomic (GSE165854) and proteomic (PXD023892) data for a more detailed view of the changes in the molecular pathways related to lipid metabolism Surprisingly, we found a number of biological processes associated with lipid metabolism that were significantly changed not only after knockdown of ELOVL5, but of IGFBP6 as well (Table 1; Table 2). In addition, a lot of the genes involved in these processes changed their expression by at least 1.5 times (FDR p < 0.05) after the knockdown of IGFBP6 gene (Supplementary Figures S2–S16).

Fatty acid biosynthesis was one of the altered biological processes. Elongation is the key step of this pathway (Figure 1A). In the first stage, the condensation reaction of acyl-CoA with malonyl-CoA catalyzed by ELOVL1-7 enzymes occurs, followed by three consecutive reactions catalyzed by enzymes 3-ketoacyl-CoA reductase (HSD17B12 gene), 3-hydroxyacyl-CoA dehydratase (HACD1-4 genes), and trans-2,3-enoyl-CoA reductase (TECR gene) (Moon and Horton, 2003; Leonard et al., 2004; Ikeda et al., 2008). Malonyl-CoA, necessary for the elongation of long fatty acids, is synthesized by enzyme acetyl-CoA carboxylase (ACACA and ACACB genes).

According to the transcriptomic analysis (Figure 1B), expression of several genes from the fatty acids elongation pathway decreased as a result of the knockdown of IGFBP6 gene in MDA-MB-231 cells; namely: ACACB (2.1 times, FDR p = 6.3·10^–9), HACD2 (2.1 times, FDR p = 1.6·10^–8), and TECR (2.0 times, FDR p = 5.8·10^–6). At the same time, the expression of the HACD1 gene slightly increased (1.6 times, FDR p = 2.8·10^–3). A decrease in the content of the TECR protein in MDA-MB-231 cells with the knockdown of IGFBP6 gene was also confirmed by proteomic analysis (1.9 times, FDR p = 2.0·10^–2).

Interestingly, according to the transcriptomic analysis the expression of one of the fatty acids elongases, ELOVL7, substantially dropped as a result of the knockdown of IGFBP6 gene (4.2 times, FDR p = 2,4·10^–10), and it was confirmed by RT-PCR (10.4 times, p = 2,6·10^–2). Moreover, similar but less prominent effect was observed after knockdown of ELOVL5 gene. In this case expression of ELOVL7 decreased 1.8 times (FDR p = 6,0·10^–4). ELOVL7 catalyzes elongation of very long chain saturated as well as unsaturated fatty acids, including n-3 and n-6 PUFAs (Naganuma et al., 2011).

Expression of several enzymes from the pathway of elongation of n-3 and n-6 polyunsaturated fatty acids, which include ELOVL5, changed after the knockdown of IGFBP6 gene (Leonard et al., 2000; Wang et al., 2008; Moon et al., 2009). A slight but statistically significant increase in the expression of ELOVL5, FADS1, and FADS2 genes after the knockdown of IGFBP6 has been noted (ELOVL5 1.4 times, FDR p = 2.0·10^–2; FADS1 1.3 times, FDR p = 2.0·10^–2 and FADS2 1.5 times, FDR p = 3.0·10^–4). The increase in FADS1 and FADS2 mRNAs was confirmed by RT-PCR (FADS1 2.7 times, p = 3.5·10^–2 and FADS2 2.0 times, p = 3.3·10^–2), and the increase in FADS2 protein level in MDA-MB-231 with knockdown of IGFBP6 gene was also confirmed by proteomic analysis (3.4 times, FDR p = 4.2·10^–3).

All described changes suggest that IGFBP6 can be involved in the regulation of lipid metabolism in breast cancer cells. To our knowledge, there are no previously published data on the role of IGFBP6 in the lipid metabolism; therefore, we performed additional experiments to test this relation.

We found that knockdowns of both ELOVL5 and IGFBP6 genes led to a significant change in the content of individual long and very long fatty acids in MDA-MB-231 cells (Figure 1C; Supplementary Table S4). In particular, when the ELOVL5 gene is knocked down, the content of fatty acids C22:4n-6, C20:2 (peak “a", n-6 or n-9) decreased. Also, when either of the genes was knocked down, there was a significant decrease in the content of very long saturated fatty acid C24:0 in the cells. This result is in good agreement with the results of transcriptomic analysis, which demonstrated a decrease in the expression of the ELOVL7 gene in both stable cell lines. Interestingly, after the knockdown of the IGFBP6 gene, a significant decrease in the content of n-3 eicosapentaenoic (EPA, C20:5n-3) and docosahexaenoic (DHA, C22:6n-3) acids in cells was also noted.

In this work we also measured uptake rates of four different PUFAs (LA, AA, ALA and DHA) by the control MDA-MB-231 cells and by the cells with knockdowns of ELOVL5 and IGFBP6 genes. In most cases, the kinetics of changes of PUFAs’ concentrations in cell culture medium could be accurately approximated by the linear model with constant uptake rate over the whole time range (Figure 2A; Supplementary Table S5). At the same time, an exponential model was a better alternative in several cases. Indicating a more efficient uptake, the cases following exponential kinetics evidently benefit from transport molecules and/or specific enzymes utilizing the fatty acid not becoming rate-limiting as the fatty acid concentration decreases. In particular, the kinetics of AA’s uptake quantitatively and qualitatively changed as a result of the knockdown of ELOVL5 and IGFBP6 genes. In both cases the cells with knockdowns demonstrated more efficient consumption.

To compare uptake rates of different PUFAs by different cells in cases of both linear and exponential kinetics, we approximated the initial parts of the exponential curves by straight lines and so obtained maximum initial uptake rates for those cases. We detected significant differences in the PUFAs’ uptake rates after the knockdowns of ELOVL5 and IGFBP6 genes (Figure 2B). For example, we found that uptake rate of AA increased 3.3 times as a result of the decrease in expression of ELOVL5 gene. Similar, but less prominent change (2.2 times) was observed for IGFBP6 gene. In addition, the consumption of ALA was 1.8 and 2.2 times faster after the knockdown of ELOVL5 and IGFBP6 genes respectively. And finally, uptake rate of DHA was significantly higher (2.2 times) for the cells with the knockdown of IGFBP6 gene in comparison to the control cells. All other changes were statistically insignificant.

Interestingly, the expression of genes involved in the transport of fatty acids into the cell (Liu et al., 2015; Zhang et al., 2018) did not change significantly (FDR p > .05) after the knockdown of ELOVL5 or IGFBP6 gene (CD36, SLC27A1-6, FABP1-9 and FABP12 genes were analyzed). The observed changes in uptake rates are thus probably due to the changes in metabolism of LC-PUFAs.

As both ELOVL5 and IGFBP6 genes affect lipid metabolism, we examined the changes in the response of MDA-MB-231 cells to external LC-PUFAs in the culture medium. Overall, we observed that all LC-PUFAs studied can decrease viability of breast cancer cells in vitro (Figure 3). The viability of the cells with the knockdown of IGFBP6 gene was significantly lower in comparison to the control cells for ALA in the range of 25–300 μM (p < 0.05), for DHA in the range of 50–100 μM (p < .05), and for AA in the range of 25–400 μM (p < .05). For the cells with reduced expression of ELOVL5 gene, similar but less prominent effects were seen. The viability of these cells was lower than that of the control cells only at 200 μM for ALA (p < .05), at 400 μM for AA (p < 0.05), and in the range of 25–100 μM (p < .05) for DHA. The effect of LA was not significantly changed by either knockdown (p > .05). While n-3 ALA and n-6 AA turned out to be the least cytotoxic towards both control and knockdown ELOVL5 cells, n-6 AA and n-3 DHA were highly cytotoxic to the knockdown IGFBP6 cells, in line with the significantly greater sensitivity of the latter to all LC-PUFAs, and in good agreement with previous research (Schley et al., 2005). In all, our data demonstrate that reduction in the expression of ELOVL5 or IGFBP6 genes can lead to an increase in the sensitivity of breast cancer cells to various PUFAs.

Interestingly, we also found that some LC-PUFAs have a stimulating effect on the growth of MDA-MB-231 cells. For example, an increase in the viability of the control cells was observed in the range 25–200 μM for ALA, 25–100 μM for DHA and at 25 μM for LA. At the same time, the stimulating effect of these fatty acids on MDA-MB-231 cells with either ELOVL5 or IGFBP6 knockdown was less pronounced or completely absent.

As all tested LC-PUFAs could decrease viability of the MDA-MB-231 breast cancer cells, we studied the effect of these LC-PUFAs on the activation of apoptosis in the cells. The analysis was carried out at 3 and 20 h after the addition of one of the LC-PUFAs (Figure 4; Supplementary Table S6). For the control MDA-MB-231 cells, 3 h after the addition of LA and ALA, a significant increase in the proportion of dead (AV + PI+) cells (from about 5% to 10%) was observed, accompanied by a decrease in the proportion of viable cells. When these fatty acids were added, the proportion of cells at the early stage of apoptosis also slightly increased from about 1.5% to 2.9%–4.1%. At 20 h after the addition of any LC-PUFA, there was a significant increase in the proportion of the control cells in the AV-PI+ region. Moreover, a pronounced peak with increased cell density was visualized in the two-dimensional density maps.

Objects in this area correspond to nuclear fragments without a cell membrane, which can result from necrosis (Sawai and Domae, 2011). Interestingly, during necrosis, cells can also first move into the AV + PI- region, mimicking the behavior of cells at an early stage of apoptosis, and only then move into the area of dead cells (AV + PI-) and into the area of nuclear fragments without a cell membrane. (AV-PI+) (Sawai and Domae, 2011). However, a significant increase in the proportion of cells in the AV-PI+ region is not characteristic of apoptosis, which makes it possible to distinguish between these two mechanisms of cell death (Wlodkowic et al., 2009; Sawai and Domae, 2011; Brauchle et al., 2015). An increase in the proportion of cells in the AV-PI+ is also characteristic of ferroptosis (Chen et al., 2017; Sun et al., 2018). At the same time the proportions of cells in the regions AV + PI+ and AV + PI- can also rise during ferroptosis (Chen et al., 2017; Sun et al., 2018).

The proportion of the control cells in the region corresponding to early apoptosis (AV + PI-) slightly increased 20 h after the addition of LA, ALA, and AA (from about 2% to 4%). The increase was more significant for the DHA (from about 2% to 9%). The proportion of dead cells also increased from 6% to 10%–13% for LA and AA, while the proportion of dead cells did not change for ALA. In case of DHA, most of the cells were in the AV + PI+ area of dead cells (about 45%), and not in the area corresponding to viable cells, as it was observed for other fatty acids.

For the knockdown ELOVL5 cells, a significant increase in the proportion of dead cells and a decrease in the proportion of viable cells was observed 3 h after the addition of ALA, AA, and DHA. This effect was less pronounced for LA. After 20 h, the pattern for the knockdown ELOVL5 cells was largely similar to the results observed in the control cells. One can notice a slightly increased sensitivity of the cells with the knockdown of ELOVL5 gene to ALA and AA, which is in good agreement with the data of the MTS test.

Three hours after the addition of LA, the knockdown IGFBP6 cells started to concentrate in the AV-PI+ region. Also, the number of dead knockdown IGFBP6 cells became more pronounced when DHA was added, compared to the control cells. Other changes after 3 h were similar to the control cells. After 20 h, for shorter LA and ALA, clusters of the knockdown IGFBP6 cells were observed in the AV-PI+ area. The proportions were similar to the control cells. For longer AA and DHA, much fewer events were observed and a significant increase of the proportion of the cells in the region of dead cells (AV + PI+) was also noticed. On the other hand, the increase of cell counts in the AV-PI+ area was less prominent.

Overall, the patterns of death observed here for all tested LC-PUFAs and the cell lines correspond more closely to necrosis or ferroptosis rather than apoptosis.

To assess how the sensitivity of MDA-MB-231 cells to ferroptosis changes upon the knockdown of ELOVL5 or IGFBP6 genes, we treated the cells with ferroptosis inducer Erastin. A ferroptosis inhibitor Ferrostatin-1 was used to prevent the action of Erastin. Our data showed (Figure 5A) that the knockdown of both genes leads to a significant increase in sensitivity to ferroptosis of MDA-MB-231 cells (p < 0.05). Moreover, the effect of Erastin was significantly reduced when Ferrostatin-1 was added to both knockdown cells (p < 0.05), while the increase of viability was insignificant for the control cells (p > 0.05). Thus, the reduction in the expression of ELOVL5 or IGFBP6 genes leads to increased sensitivity to the induction of ferroptosis in breast cancer cells and at the same time makes them more responsive to its inhibition.

Next, we treated MDA-MB-231 cells with combinations of various LC-PUFAs and Ferrostatin-1. Ferrostatin concentrations were the same as in the previous experiment. The choice of working concentrations of fatty acids (300 μM for LA, AA, and ALA, and 200 μM for DHA) was made on the basis of the data obtained on the cellular response to external LC-PUFAs to ensure the cells showed a measurable decrease in viability.

We confirmed that increasing the content of LC-PUFAs in the nutrient medium decreases the viability of MDA-MB-231 cells (Figure 5B). This effect was significantly reduced by Ferrostatin-1 in the knockdown IGFBP6 cells for all LC-PUFAs (p < 0.05) and in the knockdown ELOVL5 for DHA (p < 0.05). Again, the action of Ferrostatin-1 was insignificant for the control cells (p > 0.05). Overall, these data suggest that the mechanisms of cell death induced by Erastin and LC-PUFAs are similar.

GPx4 peroxidase is one of the key enzymes involved in the glutathione pathway of ferroptosis inhibition. It oxidizes glutathione to form glutathione disulfide (GSSG) and reduces the cytotoxic lipid peroxides L-OOH to alcohols simultaneously. Disruption of GPx4 activity leads to a decrease in antioxidant capacity and, consequently, to ferroptosis. Our transcriptomic analysis revealed a significantly reduced level of GPx4 expression in knockdown IGFBP6 cells compared to the control cells.

Therefore, the next important task was to check the activity of this enzyme in all three cell lines. In this assay, GSSG formed by GPx4 is reduced back to 2GSH by glutathione disulfide reductase (GR) using NADPH as a source of electrons and cumene hydroperoxide as a substrate. The rate of oxidation of NADPH is directly reflects the activity of GPx4. This rate was determined by following the change in absorbance of NADPH at 340 nm with a spectrophotometer in kinetic mode and all reactants kept in excess so as to make the action of GPx4 rate limiting in the oxidation of NADPH.

The results obtained (Figure 5C) demonstrated a significant decrease in GPx4 activity in knockdown IGFBP6 cells compared to the control (p < 0.05). These data are in good agreement with the transcriptomic analysis. It should also be noted that the activity of this enzyme also slightly changed after the knockdown of ELOVL5 gene (p < 0.05). Thus, the decrease in GPx4 activity is likely to be the key to lowering antioxidant capacity and increasing cells’ sensitivity to ferroptosis with the knockdowns.

Next, we tested whether the expression of GPX4 gene can also decrease simultaneously with the expression of IGFBP6 gene in breast cancer tissue. The analysis of the publicly available databases of transcriptomes of breast cancer samples showed that GPX4 gene expression positively correlates with IGFBP6 gene expression (i.e. decreases with a decrease in IGFBP6 gene expression) in tumor samples from patients with ER+ breast cancer in 4 analyzed data sets (in total 10 datasets of ER+ breast cancer were analyzed) and in 5 datasets of ER-breast cancer patients (in total 7 datasets of ER-breast cancer were analyzed). Negative correlations have not been observed in this study.

To check whether the knockdown of either ELOVL5 or IGFBP6 gene leads greater accumulation of reactive oxygen species (ROS) in the cells upon addition of external PUFAs, we used ROS indicator 2′,7′-dichlorodihydrofluorescein diacetate (DCFDA, also known as H2DCFDA, DCFH-DA, and DCFH). DCFDA is a cell-permeable chemically reduced form of fluorescein. After cleavage of the acetate groups by intracellular esterases and oxidation, the non-fluorescent form of DCFDA is converted to highly fluorescent DCF.

The results showed (Figure 6A) that for all tested PUFAs fluorescence intensity of DCF normalized to the control culture medium without external PUFAs was higher for the cells with reduced expression of ELOVL5 (ANOVA p < 0.05) and IGFBP6 gene (ANOVA p < 0.05), indicating greater accumulation of ROS in the knockdowns. In addition, knockdown IGFBP6 cells with almost all fatty acids, and especially DHA, demonstrated the highest increase in fluorescence intensity. These data have been confirmed by microscopic observations (Figure 6B). Overall, the greater accumulation of ROS in the knockdowns is in good agreement with lower activity of GPx4 and higher sensitivity of these cells to PUFAs.

One possible way to prevent oxidation of PUFAs is to store them in lipid droplets (Wang et al., 2017; Dierge et al., 2021). We compared the accumulation of lipid droplets in the cells with knockdowns to the control cells with the help of lipid dye Oil Red O. The results (Figure 6C) showed an increase in the accumulation of lipids in MDA-MB-231 cells after addition of external PUFAs to the nutrient medium. Oil Red O extraction was carried out for quantification of lipid accumulation (Figure 6D). We found that knockdown of IGFBP6 gene significantly reduces the formation of lipid droplets under the control conditions (p <0.05) and after treatment with LA (p < 0.05) and AA (p < 0.05). These data were confirmed by staining the cells with another lipid dye BODIPY (Figure S17).

These results are in good agreement with the transcriptomic analysis, which demonstrated significantly reduced expression levels of the AGPAT3, GPAT2, GPAT3, and DGAT1 enzyme genes in the knockdown IGFBP6 cells compared to the control cells (Supplementary Table S7). These enzymes play important roles in the regulation of triacylglycerol (TAG) biosynthesis: AGPAT3 and GPAT2 translocate TAG molecules from the endoplasmic reticulum (ER) into lipid droplets (LD) during membrane fusion; GPAT3 catalyzes the conversion of glycerol-3-phosphate to lysophosphatidic acid during the synthesis of triacylglycerol; and DGAT1 catalyzes the conversion of diacylglycerol and fatty acid esters of coenzyme A to triacylglycerols (Wilfling et al., 2014; Wang et al., 2017; Quiroga et al., 2021). Thus, disruption of TAG biosynthesis and lipid droplet formation in the knockdown IGFBP6 cells appear to be one of the reasons for their increased sensitivity to external LC-PUFAs.

To find out whether induction of ferroptosis by a PUFA or Erastin can significantly improve cytotoxic effects of standard of care (SOC) chemotherapeutic drugs, we tested different combinations on 3D cell models from control MDA-MB-231 cells, as well as from the cells with the knockdowns. The cells were grown in Matrigel and formed spheroids. This method allows one to create more physiologically relevant environment and provides a more gradual access for the drugs to the cells, reflecting better the conditions in vivo.

Overall, almost all tested drugs at clinically relevant concentrations were able to significantly reduce growth rate or even cause cell death (Figure 7). The only exception was methotrexate which was inactive on all the three cell types (p > 0.05). As expected, breast cancer cells after the knockdown of either ELOVL5 or IGFBP6 gene were more sensitive to DHA (p < 0.05). Interestingly, after the knockdown of IGFBP6 gene cells also became more sensitive to docetaxel, doxorubicin, vinorelbine and pure erastin (p < 0.05). On the other hand, decrease in the expression of IGFBP6 gene led to higher resistance to gemcitabine (p < 0.05).

Analysis of the results of the combined treatment showed that combination of 5-FU with DHA resulted in better growth inhibition than pure 5-FU (p < 0.05) and pure DHA (p < 0.05) for control cells. The effect of this combination was still better than pure 5-FU for the cells with knockdown of ELOVL5 (p < 0.05) or IGFBP6 (p < 0.05) gene, however it was indistinguishable from pure DHA (p > 0.05), as DHA itself was much more potent for these cells. The effect of the combination of 5-FU with erastin on the control cells was almost the same as the effects of pure 5-FU (p > 0.05) and pure erastin (p > 0.05). However, erastin significantly improved the effect of 5-FU after the knockdown of ELOVL5 (p < 0.05) or IGFBP6 (p < 0.05) gene. But again, the result of the combination was indistinguishable from pure erastin (p > 0.05).

The inhibitory effect on growth of docetaxel in combination with either DHA or erasin was the same as the effect of pure docetaxel for all three cell types (p > .05). The same results were obtained for doxorubicin (p > 0.05). Very similar to these two drugs was vinorelbine with the exception for the cells with reduced expression of ELOVL5 gene where the combination of vinorelbine with DHA resulted in slightly better growth inhibition than pure vinorelbine (p < .05) and pure DHA (p < 0.05). The effect of methotrexate in combination with DHA or erastin was in all cases better (p < .05) than the effect of pure methotrexate as it was not active as single drug. However, almost in all cases (with one exception for combination with erastin on the control cells) the effect did not differ from pure DHA (p > 0.05) or pure erastin (p > 0.05).

The results of the test for gemcitabine in combination with either DHA or erasin did not differ from pure gemcitabine on the control cells (p > 0.05). On the other hand, the combined treatment of gemcitabine with DHA outperformed pure gemcitabine after the knockdown of either ELOVL5 (p < 0.05) or IGFBP6 (p < 0.05) gene. However, as DHA was much more effective on these cells, the growth inhibitory effect of the combination was indistinguishable from pure DHA (p > .05). Addition of erastin to gemcitabine increased the effect of the later only on the cells with reduced expression of IGFBP6 gene (p < 0.05).

Overall, these results clearly show that both DHA and erastin have the potential to inhibit growth of breast cancer cells. Moreover, in some cases they can enhance the effect of the SOC drugs, especially if the cells have a decreased expression of IGFBP6 gene.