Residual post-treatment TNBC tumours are enriched with RASAL2 compared to adjacent normal breast tissues

Undifferentiated tumours are associated with poorer prognosis compared to well-differentiated tumours [14]. Using CytoTRACE, a computational framework for predicting differentiation states from single-cell RNA-sequencing data [15, 16], we found that high RASAL2 expression overlapped with less differentiated epithelial cells from TNBC patients (Fig. 1A). Moreover, high RASAL2 expression in these cells was strongly associated with a tumour transcriptomic signature [17] derived from genes enriched in residual viable tumour population of patients treated with pre-operative chemotherapy (Fig. 1B). Further validation using another cohort of TNBC patients [18] showed that RASAL2 was specifically enriched in the most aggressive subpopulation that harbour multiple signatures related to unfavourable clinical outcomes such as metastasis and chemoresistance (Fig. S1A). Collectively, despite the heterogeneity of TNBC, these data suggest that RASAL2 is a common molecular feature in a subset of aggressive TNBC cells.

Fig. 1: Residual tumours, but not adjacent normal tissues, are enriched with RASAL2.

A UMAP plots of epithelial cells from primary TNBC tumours showing correlation between CytoTRACE score (1—least differentiation, 0 – most differentiation) and RASAL2 expression. B Violin plots of RASAL2 expression and residual tumour signature expression in the four clusters of epithelial cells identified in the TNBC tumours in (A). Cluster 1 expressions are significantly higher compared to all other clusters. Data are represented as mean ± SEM. P value by one-way ANOVA. C Heatmap of RASAL2 expression in pre- and post-treatment breast cancer patients [19]. Fold change (FC) was determined relative to pre-treatment expression level. P value by paired T-test. D Dot plots of RASAL2 expression in pre- versus post-treatment TNBC/BRCA-mutant breast cancer patients [20]. Probes that recognise RASAL2 variant 2 (ILMN_1813701 and ILMN_1673455) show significant increase following treatment. Data are represented as mean ± SEM. P value by two-tailed T-test. E Immunohistochemistry of fixed post-treatment TNBC patient breast specimens. RASAL2 (brown stain) was enriched in the tumour compartment versus adjacent normal epithelia. Data are represented as mean ± SEM. P value by two-tailed T-test. Scale bar, 20 µm. F Immunoblotting of fresh post-treatment TNBC patient specimens. RASAL2 was enriched in the tumour (T) versus adjacent normal (N) tissues in TNBC patients. LE long exposure.

Next, we asked how RASAL2 expression changed post-treatment. First, we analysed the transcriptome of patient-matched breast tumour specimens before and after neoadjuvant (pre-operative) chemotherapy [19]. Consistently, RASAL2 was upregulated following treatment in tumours of TNBC patients, but this upregulation was not observed in some of the non-TNBC patients (Fig. 1C). There are at least two known RASAL2 transcriptional variants, variants 1 and 2, where they differ in the 5’ untranslated and coding regions (Fig. S1B). Using an independent patient cohort of TNBC/BRCA-mutant breast cancer patients [20], we found that transcripts of RASAL2 variant 2, but not variant 1, was significantly upregulated following treatment (Figs. 1D and S1C, D). Furthermore, we observed that RASAL2 protein was enriched in post-treatment (residual) TNBC tumour tissue compared to patient-matched adjacent normal breast tissue (Fig. 1E). Immunoblotting of fresh patient-matched residual tumour versus normal breast tissues corroborated this observation (Figs. 1F and S1E). Collectively, these findings suggest that RASAL2 is a tumour cell-specific factor that is upregulated in residual treatment-resistant TNBC.

RASAL2 promotes pan-resistance to cytotoxic agents beyond platinum

We previously reported that platinum resistance is correlated with RASAL2 overexpression [12, 13], but the extent of this correlation with other DNA-damaging agents was largely unexplored. Using the publicly available computational analysis of resistance model [21], we found that RASAL2 expression is highly associated with resistance towards drugs that target the DNA damage checkpoint-related pathways (Fig. S2A). Pharmacogenomic analysis of human cancer cell lines revealed that RASAL2 was broadly associated with resistance to classic DNA-damaging agents (Fig. S2B). Among others, these agents include chemotherapeutic agents that are used in TNBC such as doxorubicin, gemcitabine and SN-38 (payload of FDA-approved antibody-drug conjugate sacituzumab govitecan) (Fig. 2A). In-vitro cytotoxic assays using 2D TNBC cell lines and 3D tumour spheroids confirmed the association. Ectopic expression of RASAL2 was sufficient to render cells more resistant to these drugs (Figs. 2B–D and S2C, D). Conversely, knockdown of endogenous RASAL2 reversed the resistance phenotype (Figs. 2E and S2D). BRCA1-mutant TNBC cells exhibited similar chemosensitivity profiles following RASAL2 overexpression and knockdown (Fig. 2D, E), suggesting that RASAL2-driven chemoresponse is at least in part independent of wild-type BRCA1 function. In a cohort of established TNBC patient-derived xenograft (PDX) mice [22], again, RASAL2 expression is significantly correlated with chemoresponsiveness (Fig. 2F). The PDX model with the highest expression of RASAL2 was doxorubicin-resistant, whereas the model with the lowest expression was highly sensitive (Fig. 2G). Together, these findings show that the effect of RASAL2 in driving resistance is not limited to platinum-based therapy as previously shown, but also to other common DNA-damaging cytotoxic agents.

Fig. 2: RASAL2 promotes pan-resistance to cytotoxic agents beyond platinum.

A Correlation between RASAL2 expression and sensitivity to indicated chemotherapy. The area under percent-viability curves (AUC) was computed as a metric of drug sensitivity, as derived from the Cancer Therapeutics Response Portal. Pearson r and P value are reported. B Cell viability assay. Vector control and RASAL2-overexpressing MDA-MB-468 cells were treated as indicated. Data are represented as mean ± SEM, n = 3 biological replicates. P value by paired T-test. C Spheroid assay. Viability of vector control and RASAL2-overexpressing TNBC spheroids was measured following treatment with vehicle DMSO, doxorubicin (DOXO) or gemcitabine (GEM). Representative images of HCC1806 spheroids are shown. Data are represented as mean ± SEM, n = 3 biological replicates. P value by paired T-test. Scale bar, 250 µm. Cell viability assay. Vector control and RASAL2-overexpressing (D) or -knockdown (E) HCC1937 cells were treated as indicated. Data are represented as mean ± SEM, n = 3 biological replicates. P value by paired T-test. F Correlation between RASAL2 expression and in vivo tumour response to doxorubicin. Mice were treated with vehicle control or 2 mg/kg doxorubicin [22]. Each dot represents an independent TNBC PDX model. Tumour growth inhibition was defined as [1 − (mean volume of treated tumours)/(mean volume of control tumours)] × 100%. Pearson r and P value are reported. G Change in tumour volume following doxorubicin in two TNBC PDX models. Mice were treated as described in (F), n = 8–9 per group. TM00099 tumours had the lowest RASAL2 expression, whereas TM01278 had the highest RASAL2 expression. P value by two-way ANOVA test.

BCL2 is upregulated by RASAL2

We observed that chemotherapy-induced apoptosis by 24 h far more robustly in control TNBC cells compared to RASAL2-overexpressing TNBC cells, as evidenced by concerted upregulation of both ɣH2AX and cleaved caspase 3 (Fig. 3A). Using a classic apoptosis inducer staurosporine, we found that RASAL2 expression indeed rendered cells less susceptible to apoptosis (Fig. 3B). To determine the underlying mechanisms, we performed genome-wide RNA sequencing of the isogenic RASAL2-overexpressing TNBC MDA-MB-468 cell model. Gene set enrichment analysis (GSEA) revealed downregulation of apoptotic transcriptomic signatures in RASAL2-overexpressing cells (Figs. 3C and S3A), in line with the prediction that RASAL2 is a resistance gene against apoptotic signals (Fig. S2A). In The Cancer Genome Atlas (TCGA) clinical cohort [23], TNBC tumours with high RASAL2 expression exhibited attenuated apoptotic signalling compared to those with low RASAL2 expression (Fig. 3D). This clinical correlation was also observed in another independent cohort of TNBC patients [24] (Fig. S3B). We found that anti-apoptotic genes such as BCL2, BCL-XL and BAXI1 were significantly upregulated upon RASAL2 expression, whereas pro-apoptotic genes such as BIK, BIM and BOK were significantly downregulated (Fig. S3C). By ranking all anti-apoptotic genes based on fold change, BCL2 emerged as the topmost elevated factor in RASAL2-overexpressing cells (Fig. 3E). We asked whether this transcriptional upregulation was recapitulated at the protein level. Immunoblotting of TNBC cell lines revealed that BCL2 protein was increased following RASAL2 expression (Fig. 3F). To further support these findings, we performed quantitative immunofluorescence and found that BCL2 protein level was significantly higher in RASAL2-overexpressing cells compared to control cells (Fig. 3G).

Fig. 3: RASAL2 upregulates BCL2.

A Immunoblotting of cisplatin-treated TNBC cells showing upregulation of apoptotic/DNA damage markers, cleaved caspase 3 and γH2AX, respectively. Vector control and RASAL2-overexpressing MDA-MB-468 cells were treated with 3 µM cisplatin for 24 h before being released into fresh drug-free medium for the indicated durations. B Immunoblotting showing upregulation of apoptotic marker, cleaved caspase 3, in vector control but not RASAL2-overexpressing MDA-MB-468 cells following apoptosis induction. Cells were treated with either DMSO or 10 µM staurosporine for 1 h. C GSEA analysis of transcriptomes showing downregulation of apoptosis signatures in RASAL2-overexpressing MDA-MB-468 cells compared to isogenic vector cells. D GSEA analysis of transcriptomes showing downregulation of apoptosis signatures in RASAL2-high TNBC tumours versus RASAL2-low TNBC tumours. TNBC patients from the TCGA cohort were stratified by quartile of RASAL2 expression. E Heatmaps of apoptosis-related genes for isogenic RASAL2-overexpressing MDA-MB-468 cells. BCL2 was the topmost significantly upregulated anti-apoptotic gene in RASAL2-overexpressing cells. F Immunoblotting of isogenic TNBC cell lines showing upregulation of BCL2 protein upon RASAL2 expression. G Quantitative immunofluorescence of BCL2 expression in isogenic RASAL2-overexpressing MDA-MB-468 cell model. Data are represented as mean ± SEM. P value by two-tailed T-test. Scale bar, 40 µm. H Representative immunohistochemistry showing TNBC tumours with high RASAL2/BCL2 protein expression (top panel) versus low RASAL2/BCL2 protein expression (bottom panel). Quantification was done based on the H-score system (Fig. S3E). Scale bar, 50 µm. I Dot plots showing upregulation of BCL2 mRNA expression in post-treatment primary TNBC/BRCA-mutant tumours versus pre-treatment tumours. Data are represented as mean ± SEM. P value by two-tailed T-test. J Immunoblotting of isogenic MDA-MB-468 cells following siRNA knockdown of RASAL2 or BCL2. Knocking down RASAL2 downregulated BCL2 expression, while knocking down BCL2 did not change RASAL2 expression.

We asked whether the association between RASAL2 and BCL2 expression was clinically relevant. The trend of a positive correlation between BCL2 and RASAL2 was observed at the mRNA and protein level in our own TNBC clinical specimens (Figs. 3H and S3D, E). In independent clinical cohorts, where RASAL2 was elevated in TNBC tumours post-treatment (Fig. 1C, D), BCL2 was also significantly elevated post-treatment (Figs. 3I and S3F). BCL2 and RASAL2 are targets of transcription co-factor YAP [12, 25, 26], which itself is activated by RASAL2 via the inactivation of the Hippo pathway [11]. In concordance, we found that overexpressing RASAL2-induced YAP activation, while silencing RASAL2 decreased YAP activation and increased translocation of YAP from the nucleus to cytoplasm (Fig. S3G–I). Accordingly, overexpressing RASAL2 upregulated BCL2 (Fig. 3F–G), while knocking down RASAL2 downregulated BCL2 (Fig. 3J). On the other hand, knocking down BCL2 did not lead to a change in RASAL2 expression (Fig. 3J), suggesting that BCL2 acts downstream of RASAL2. Finally, knocking down YAP reversed the upregulation of BCL2 by RASAL2 (Fig. S3J). Together, these findings suggest that RASAL2 upregulates BCL2 via YAP, and that this axis contributes to the concerted dampening of apoptotic signalling observed in the RASAL2-high context.

CREB1 is the common transcription factor that drives BCL2 and RASAL2 expression

The collective YAP-mediated upregulation of BCL2 and RASAL2 suggested that there might be common transcription factors driving their expression. To determine the upstream regulatory elements, we first employed computational tools JASPAR (a repository of transcription factor binding profiles represented as position frequency matrices [27]) and LASAGNA-Search 2.0 (a database leveraging the Length-Aware Site Alignment Guided by Nucleotide Association algorithm for identifying transcription factor binding sites [28, 29]) to predict transcription factors that most likely bind to the promoters of RASAL2 and BCL2. Our convergent analysis identified 15 candidate transcription factors for both genes (Fig. 4A). We filtered this list by analysing transcriptomics data from patient-matched breast tumour specimens collected before and after neoadjuvant (pre-operative) chemotherapy. We found that, similar to RASAL2 and BCL2, two of the 15 transcription factors, CREB1 and ZEB1, were consistently upregulated in post-treatment samples specifically in TNBC but not in non-TNBC patients [19] (Fig. 4B, top). In particular, CREB1 is a YAP-interacting transcription factor [30] that was associated with not only post-treatment upregulation [19, 20] (Figs. 4B and S4A, B) but was also predictive of poor outcomes in chemotherapy-treated TNBC patients [31] (Fig. S4C). Furthermore, correlative analysis using TNBC patients in the TCGA-BRCA cohort showed that CREB1 demonstrated a significant positive correlation with both RASAL2 and BCL2 expressions (Fig. 4B, bottom).

Fig. 4: Transcription factor CREB1 drives RASAL2 and BCL2 expression.

A Venn diagram showing the overlapped predicted transcription factors binding on the promoters of RASAL2 and BCL2 using computational tools, JASPAR and LASAGNA. B Analyses of candidate transcription factors in breast cancer patient cohorts. Heatmap shows the fold changes of RASAL2, BCL2 and candidate gene transcription factors in patient-matched breast tumour specimens post- versus pre-treatment (top, [19]). Correlation between RASAL2/BCL2 expression and candidate transcription factors in TNBC patients in the TCGA-BRCA cohort (bottom). P value by two-sided Pearson correlation analysis. C Consensus binding motifs of transcription factor CREB1. D Decrease in the relative mRNA expression of RASAL2 and BCL2 following siRNA-mediated knockdown of CREB1 in MDA-MB-468 cells. Data are represented as mean ± SEM, n = 3 biological replicates. P value by two-tailed T-test. E Immunoblotting of TNBC cells transfected with siCREB1 or control siRNA. CREB1, RASAL2 and BCL2 were decreased in expression in cells treated with siCREB1 compared to control. F ChIP-qPCR confirmation of CREB1 binding to predicted sites on RASAL2 and BCL2 promoters. TSS denotes transcription start site. Data are represented as mean ± SEM, n = 3 biological replicates. P value by two-tailed T-test. G Decrease in the relative luciferase units in siCREB1 MDA-MB-468 cells compared to siControl. pRL-CMV Renilla luciferase plasmid was co-transfected for normalisation. Data are represented as mean ± SEM, n = 3 biological replicates. P value by two-way ANOVA. H Decrease in the relative luciferase units in MDA-MB-468 cells with truncated RASAL2 promoter without CREB1-binding sequence compared to those with wild-type (WT) RASAL2 promoter. pRL-CMV Renilla luciferase plasmid was co-transfected for normalisation. Data are represented as mean ± SEM, n = 3 biological replicates. P value by one-way ANOVA test.

Given that CREB1 is the sole candidate transcription factor that demonstrated notable clinical correlations across multiple patient cohorts (Figs. 4B, C and S4A–C), we proceeded to test the functional effect of knocking down CREB1. First, we found a decrease in both mRNA and protein expression levels of RASAL2 and BCL2 upon CREB1 knockdown across all TNBC cell lines tested (Fig. 4D, E and S4D). Second, we assessed the binding of CREB1 to the respective promoters of RASAL2 and BCL2 by ChIP-qPCR. At predicted sites within RASAL2 and BCL2 promoter regions, we found significant enrichment of CREB1 binding (Fig. 4F). Furthermore, CREB1 ChIP-sequencing data demonstrated a significant binding peak at the predicted CREB1 binding site in the RASAL2 promoter region [32], indicating a strong likelihood that CREB1 interacts with the RASAL2 promoter (Fig. S4E). Third, we generated RASAL2 promoter and performed luciferase reporter assay. We observed a significant reduction in luciferase activity when CREB1 was silenced in TNBC cells (Fig. 4G), suggesting that CREB1 is responsible to enabling RASAL2 transcription. Finally, we truncated RASAL2 promoter region harbouring the binding sequence of CREB1, and again found a significant reduction in luciferase activity compared to wild-type promoter (Fig. 4H). Together, these findings show that CREB1 is a transcription factor that is, at least in part, responsible for the upregulation of RASAL2 and BCL2.

BCL2 interacts with RASAL2 on mitochondria

In TNBC cells and patient tumours, we observed that immunofluorescence staining of BCL2 overlapped with RASAL2, suggesting subcellular co-localisation of the two proteins (Fig. 5A, B). To validate these findings, we employed high-resolution confocal microscopy and found that BCL2 colocalised with a subset of RASAL2 (Figs. 5C and S5A). In-silico AlphaFold modelling predicted high confidence protein-protein interaction between BCL2 and the N-terminus of RASAL2 [33, 34] (Fig. 5D), which was consistent with the result of co-immunoprecipitation assay (Fig. 5E). Moreover, unlike the full-length wild-type RASAL2, truncation of the N-terminus of RASAL2 abolished the ability of RASAL2 to bind to BCL2 (Fig. S5B). Given that BCL2 normally localises to mitochondria, we hypothesised that a subset of RASAL2 may be present on these organelles. Subcellular fractionation of multiple mammary cell lines showed that the detection of BCL2 was largely restricted to the mitochondrial fraction, as expected (Fig. 5F). Conversely, RASAL2 was found in both the cytosolic and mitochondrial fractions. Indeed, the mitochondrial presence of RASAL2 was detected in all cell lines used, including non-malignant human mammary line MCF10A and malignant mouse mammary line 4T1, alluding to an evolutionarily conserved feature (Fig. 5F). Mitochondrial labelling in cells further demonstrated an overlap of RASAL2 with the mitochondrial network, corroborating the notion of co-localisation (Figs. 5G and S5C). Collectively, these findings reveal a hitherto undescribed physical interaction between BCL2 and RASAL2, with mitochondria being a common homing site for both proteins.

Fig. 5: Mitochondria is a common homing site for BCL2 and RASAL2.

A Immunofluorescence of BCL2 and RASAL2 in primary TNBC patient tumour. Scale bar, 20 µm. B Immunofluorescence of BCL2 and RASAL2 in TNBC cells. Bottom graph shows the line scan quantification of BCL2 (red) and RASAL2 (green). Scale bar, 30 µm. C Confocal imaging of BCL2 and RASAL2 in TNBC cells. Panels on the left show exemplary co-localisation of signals within the boxed region of the cell. Scale bar, 10 µm. D AlphaFold prediction of the interaction between BCL2 and the N-terminus of RASAL2. pLDDT score (0–100) is a confidence score, and pTM score (0–1) is a metric for the structural congruency between two folded protein structures, with higher scores corresponding to higher confidence. PAE plot of the top ranked model is shown on the right [33, 34]. E Co-immunoprecipitation of BCL2 and RASAL2 in MDA-MB-468 cells. F Immunoblotting of cytoplasmic versus mitochondrial fractions of mammary cell lines. BCL2 was not detected in 4T1 murine cells as the antibody used was reactive only to human. AKT and TOM20 serve as cytoplasmic and mitochondrial markers, respectively. G Confocal imaging of RASAL2 and MitoTracker in TNBC cells. Scale bar, 5 µm.

High BCL2 levels confer mitochondrial resilience against apoptosis induction

BAX translocates to mitochondria and oligomerises in response to apoptotic stimuli, leading to permeabilisation of mitochondrial outer membrane [35, 36]. We hypothesised that increased BCL2 levels in the RASAL2-high context counteract BAX oligomerisation and render mitochondria more robust to permeabilisation. We conducted live-cell imaging of TNBC cells transfected with GFP-tagged BAX. As expected, upon exposure to staurosporine, BAX fluorescence intensity increased over time, consistent with its accumulation following apoptosis induction (Fig. 6A). The kinetics of BAX accumulation in RASAL2-depleted cells was faster than that in control cells, indicating that RASAL2 attenuates BAX accumulation during apoptosis induction (Fig. 6B).

Fig. 6: BCL2 upregulation attenuate mitochondrial depolarisation by attenuating BAX oligomerisation.

A Live-cell imaging. RASAL2 depletion increases the rate of GFP-BAX accumulation (green) in HCC1937 cells following exposure to 20 µM staurosporine. Number denotes time in seconds. Scale bar, 10 µm. B Quantification of change in BAX intensity. Fluorescence intensity of individual BAX foci was tracked over time and quantified, n = 4 foci per condition. Data are represented as mean ± SEM. P value by two-tailed T-test. C Schematic for live mitochondrial outer membrane permeabilisation (MOMP) assay. D MOMP assays revealing attenuated cytochrome c release in RASAL2-overexpressing TNBC cells. TOM20 serves as mitochondrial marker. E JC-1 mitochondrial membrane potential assay. TNBC cells were treated with vehicle DMSO or 5 µM doxorubicin (DOXO), and subsequently stained with JC-1 reagent. JC-1 aggregates (indicating high mitochondrial membrane potential) were observed as red, while JC-1 monomers (indicating low mitochondrial membrane potential) were green. Representative images of vector control and RASAL2-overexpressing TNBC cells are shown. Scale bar, 100 µm. F Quantification of the ratio of integrated intensity of red to green fluorescence in (E). Data are represented as mean ± SEM, n = 5 random fields of view per condition. P value by two-tailed T-test.

A hallmark of apoptosis is the disruption of the mitochondrial outer membrane. To directly test the resilience of mitochondria, we isolated live mitochondria from TNBC cells and measured the level of death signal cytochrome c following apoptosis induction (Fig. 6C). First, consistent with the higher levels of total BCL2 found in RASAL2-high cells (Fig. 3F, G), mitochondria of these cells harbour more BCL2 compared to the isogenic control cells (first versus third columns, Fig. 6D). Second, in line with the presence of RASAL2 in mitochondrial fraction (Fig. 5F), RASAL2 was also detected on intact mitochondria. In fact, similar to BCL2, mitochondrial RASAL2 level was higher in RASAL2-high cells compared to control cells (first two versus last two columns, Fig. 6D). Third, following apoptosis induction by treatment of BAX and tBID, release of cytochrome c was detected in the supernatant. Notably, there was less cytochrome c detected in the supernatant of RASAL2-high mitochondria compared to control cells, supporting the notion that RASAL2-high cells are more apoptosis-tolerant (second versus fourth columns, Fig. 6D). The same observations were made across different TNBC cell lines (Fig. 6D). Finally, to further confirm that RASAL2 bolsters mitochondrial resilience to depolarisation, we used JC-1 dye as a live-cell mitochondrial membrane potential indicator in TNBC cells exposed to chemotherapy (Fig. 6E). At high membrane potentials, JC-1 dye monomer forms aggregates, where it accumulates within the mitochondria. Conversely, at low internal mitochondrial concentrations or low membrane potentials, the dye presents as monomers. Consistent with the findings of the mitochondrial outer membrane permeabilisation assay (Fig. 6C, D), we found significantly more JC-1 aggregates (red) in RASAL2-overexpressing cells compared to control cells upon doxorubicin treatment (Fig. 6E, F), indicating a higher mitochondrial tolerance towards apoptosis induction in the former. This observation was uniformly recapitulated across multiple cell lines and DNA-damaging agents (Figs. 6E, F and S6A, B). Furthermore, consistent with the notion that CREB1 drives RASAL2 expression (Fig. 4), we found significantly reduced accumulation of JC-1 aggregates (red) in CREB1-knockdown cells compared to control cells following chemotherapy treatment (Fig. S6C, D), suggesting a lower mitochondrial tolerance to apoptosis in the absence of CREB1. Together, these findings suggest that RASAL2 bolsters mitochondrial resilience by mitigating membrane depolarisation, presumably via inhibition of BAX oligomerisation owing to high levels of mitochondrial BCL2.