Chemoresistance to the combination of doxorubicin and docetaxel is efficiently achieved by loss of genetic material.
There is a clinical need to predict response following chemotherapy treatment and avoid unwarranted treatment and a cancer relapse. We analysed the clinical survival data for standard of care neoadjuvant taxane-anthracycline (docetaxel and doxorubicin) frontline combination therapy comparing TNBC with the remaining subtypes. Results showed that TNBC patients displayed significantly worse distant recurrence free survival (GSE25066, ****p = 1.8e − 6) (Fig. 1A). To evaluate the essential pathways observed in TNBC patients in response to neoadjuvant docetaxel and doxorubicin frontline combination, we analyzed the gene expression changes in treated patients, comparing TNBC to the Luminal/HER2 + subtypes (GSE25066)(16). Molecular Signatures Database curated gene set analysis resulted in differential induction of 88 genes, while 130 genes were repressed (Supplementary Table 1). The gene signatures were able to distinctly separate TNBC (Basal) from the remaining subtypes in GSE25066 (Fig S1A/B) and an independent dataset (Fig. 1B). Gene ontology molecular functions analysis of the TNBC-distinguishing gene signatures revealed four of the top 10 ranked, contained pathways with known DNA binding functions. Of these four, two included pathways with double strand DNA binding, which is functionally dominated by DNA repair proteins and associated functions (Fig. 1C). Although chemoresistance has been attributed to cancer stem-cells, no gene ontology annotated terms or molecular functions were assigned to stem-like functions or phenotypes. This may suggest a dominant role for DNA repair pathways in gene expression response to docetaxel and doxorubicin. In line with this, we observe a positive correlation between our above gene expression signatures (induced and repressed genes) and a gene expression signature, based on the expression of 230 genes as measure for homologous DNA repair deficiency (HRD) in basal breast cancer patients (Fig. 1D).
In order to evaluate the chemotherapy resistance mechanisms driving TNBC recurrence, we grew four TNBC cell lines MDA-MB-231, Bt549, Hs578t and MDA-MB-468 and exposed them to the frontline therapy combination of doxorubicin and docetaxel (ratio 100:1) starting at 0.1 nM docetaxel and 10 nM doxorubicin and concluding with 2 nM docetaxel and 200 nM doxorubicin. The concentration was chosen to mimic the clinical administration of bolus doxorubicin and infusion of docetaxel (17). Resistance was achieved over 6–8 months with escalating doses of the chemotherapy combination. Chemotherapy resistance has been shown to correlate with acquisition of new genomic aberrations and adaptive copy-number evolution (18). To investigate this, we performed metaphase spread counts comparing chromosome numbers between sensitive and resistant paired lines. On a macromolecular level all resistant cell lines had an overall loss of genetic material compared to sensitive parental cells. There was significant loss in chromosome count in the MDA-MB-231 (Fig. 1E) and BT549 (Fig. 1F) and trended lower in Hs578t (Fig. 1G) and MDA-MB-468 cells (Fig. 1H) but did not reach statistical significance. The two statistically significant lines correlated with higher IC50 values achieved for docetaxel (Fig. 1I) and there was a strong correlation between chromosome loss and resistance to docetaxel (Fig. 1J). This observation was repeated with doxorubicin (Fig. 1K/L). This suggests that adaptation to DNA damaging agents requires or results in loss of genetic material. A more detailed analysis of copy number variation in resistant MDA-MB-231 and MDA-MB-468 lines confirmed the observations made with our metaphase spread data (Fig S2). The more highly resistant MDA-MB-231 genetics displayed a greater increase in loss of heterozygosity and a reduction in copy number gains compared to the MDA-MB-468 resistant line (Fig S2). In addition, there was a positive correlation between the TNBC-distinguishing gene signatures and several measures of genomic instability, including the CIN70 gene expression signature (functional aneuploidy), overall aneuploidy and chromosome arm aneuploidy burden (Fig S3). Overall, we observed a correlation between chemotherapy-induced gene expression specifically in TNBC and DDR pathway activation/genome instability. We also observed a correlation between genome instability and levels of resistance achieved.
Chemoresistance drives a shift towards homologous recombination to sustain genome stability.
To evaluate the role of the individual DNA repair pathways in the observed chemoresistance profiles, we analysed the specific cellular double strand DNA break repair response. The two main double strand break DNA repair pathways are the error prone non-homologous end-joining pathway and high-fidelity homologous recombination pathway. We utilized the classical enzyme directed double strand DNA break of GFP and the MDA-MB-231 matched pair sensitive and resistant lines. The first assay was a measure of homologous recombination activity and required a sister template of GFP (iGFP) and active pathway to recapitulate the fluorescence of an enzyme cleaved GFP construct (Fig. 2A). In contrast to our expectations, related to genetic losses via high levels of error prone non-homologous end joining repair, the MDA-MB-231RES cells displayed significantly enhanced homologous recombination activity with 7-fold higher activity compared to MDA-MB-231SENS (**p < 0.001, Fig. 2B). The second assay was a measure of non-homologous end joining via recapitulation of the upstream gene cassette with active promoter for GFP expression (Fig. 2C). Although to a much lesser extent, MDA-MB-231RES cells also displayed enhanced non-homologous end-joining 2.8-fold (***p < 0.001, Fig. 2D). Thus, acquired chemotherapy resistant cells rely upon enhanced DNA repair in both of the main double strand break repair pathways.
To determine whether the adapted cells are reliant upon any specific pathway, we inhibited repair with specific small molecule inhibitors in the context of cell death via DNA double strand breaks induced by gamma irradiation (Fig. 2E). A cellular response to ATM inhibition would represent a general activation of double strand break response, while a response to ATR inhibition would represent a reliance upon single strand break response repair. Specific targeting of key repair proteins provides further insights. Cell death in response to DNA-PK inhibition represents a reliance upon non-homologous end joining, while cell death induced by RAD51 inhibition shows cellular reliance upon homologous recombination. Finally, PARP inhibition is a measure of reliance upon alternative non-homologous end joining. Comparing cell death induced by each of the inhibitors, we observed a switch from an equal distribution between the available repair mechanisms of alternative-non homologous end joining, non-homologous end joining and homologous recombination pathways in MDA-MB-231SENS to a dominant homologous recombination dependent phenotype in MDA-MB-231RES (Fig. 2E, bottom panel). The same shift towards homologous recombination was observed when DNA double strand breaks were induced by the combination of docetaxel and doxorubicin in MDA-MB-231 lines (Fig. 2F). The BT549 cell line, like MDA-MB-231, also displayed significant difference in chromosome count and, consistent with this, displayed the same shift towards reliance on the homologous recombination pathway when resistant to the combination of docetaxel and doxorubicin (Fig. 2G).
To evaluate whether the increased reliance on homologous recombination was related to higher levels of DNA damage, RAD51 function or even longer time spent in S-G2 phase where homologous recombination is active, we analyzed the resolution of DNA double strands breaks over a 24-hour time course in BT549 and MDA-MB-231 cells, respectively (Fig. 2H and S4). Double strand breaks are visualized within the nucleus by gamma-H2AX foci marks. Active double strand break repair is visualized and attributed to homologous recombination as RAD51 foci, while the switch to activate non-homologous end joining is visualized with 53BP1 foci. Compared to BT549SENS cells, (Fig. 2H) BT549RES cells (Fig. 2I) had a higher RAD51 foci count per nucleus, although these formed with slower kinetics than in the sensitive cells. Although the cells display slower RAD51 kinetics, BT549RES cells resulted in enhanced resolution of gamma-H2Ax foci (DNA double strand breaks) at the 24-hour time point, as compared to the BT549SENS cells treated with the same dose of IR (Fig. 2H vs 2I). BT549RES and BT549SENS displayed similarly low levels of 53BP1 induction following IR treatment, consistent with similar signaling towards non-homologous end joining (Fig. 2H vs 2I). This data supports our inhibitor results and showed that resistant cells utilize homologous recombination, sustain the same amplitude of double strand breaks as sensitive cells but resolve these breaks more efficiently.
Sensitization of chemoresistant cells by targeting RAD51 and homologous recombination.
Having established an enhanced homologous recombination repair phenotype in chemoresistant cell lines, we wondered if targeting of the key repair protein RAD51 could sensitize these cells. Utilizing a validated small molecule inhibitor of RAD51 (19), MDA-MB-231RES were actually significantly more sensitive to the inhibitor than MDA-MB-231SENS cells (Fig. 3A). The combination of chemotherapy and RAD51 inhibition achieved resensitization and equivalent levels of cell death independent of chemoresistance status (Fig. 3A). Conversely, stabilization of RAD51 with a small molecule (RS-1)(20) congruently enhanced chemotherapy resistance in MDA-MB-231RES (Fig. 3B). Therefore, RAD51 function demonstrated a direct role of homologous recombination in chemoresistance.
The targeting of RAD51 with chemotherapy induced cell death irrespective of chemoresistance. This provides the capability to successfully target the heterogeneity often observed in TNBC. However, we find that the functionally available repair pathways in sensitive and resistant lines are very different. We evaluated the gene expression profile as a readout for DNA damage response to the combination of chemotherapy and RAD51i (Fig. 3A). Utilizing a targeted array of 21 DNA repair related genes and cluster analysis per individual gene in MDA-MB-231SENS cells we revealed 15/21 of the genes were most highly induced by chemotherapy and RAD51i (Fig. 3C). The Euclidean distance arranged these genes into three main clusters, each of which were derived of both key non-homologous end joining and homologous recombination pathways, cluster I-BRCA1 and DNA-PK, cluster II-KU70 and BRCA2 and cluster III PARP1 and KU80. This was representative of an even distribution of DDR pathways utilized and similar to what was observed in our small molecule inhibition studies in MDA-MB-231SENS cells (Fig. 2F). The induction of gene expression is a representation of individual gene response however we were curious about the functional response. Therefore, we performed functional interaction network analysis and displayed distances based on physical, colocalization and co-expression values. The most highly expressed genes circled in red were tightly associated by physical interactions (pink lines) and outliers joined by pathway association (blue lines) (Fig. 3D).
Analysis of MDA-MB-231RES cells, revealed only 5/21 of the genes were most highly induced by chemotherapy and RAD51i (Fig. 3E). These genes were functionally associated with inhibition of non-homologous end joining, cluster I-p73, cluster II-PARP1, cluster III-cABL1 and no highly expressed genes in cluster IV (Fig. 3E). Functional interaction network analysis revealed that RAD51 was again centrally located in the network and closely associated with cABL1 and BRCA1 (Fig. 3F). Notably RAD51 was also induced in response to chemotherapy alone in both cell lines and as it is centrally located in each network it is a rationale target in TNBC tumours that often display chemoresistance heterogeneity.
Gene ontology molecular functions analysis of the 21 genes revealed that the MDA-MB-231SENS continued to rely upon both homologous recombination and non-homologus end joining as two of the top 10 pathways while MDA-MB-231RES relied upon only homologous recombination (Figs. 3G/H). Supporting protein analysis of MDA-MB-231RES revealed that DNA-PK did not display the classical phosphorylation in response to chemotherapy as seen in MDA-MB-231SENS, although total DNA-PK protein expression was sustained (Fig. 3I). This suggested non-functional or ablated non-homologous end joining activity. We also observed associated increases in 53BP1 protein expression in MDA-MB-231RES across all treatment conditions supporting activation of DNA-PK and non-homologous end joining activity (Fig. 3I). RAD51 protein expression was not enhanced and there was an associated loss of BRCA1 expression, suggesting suppression of homologous recombination. However, we also observed a compensation by enhanced BRCA2 expression in MDA-MB-231RES cells. In addition, we observed an induction of p73 with associated repression of p53 activation, providing a switch in DNA damage sensing to p73.