Therapy surviving breast cancer cells enter senescence before repopulating

To replicate the clinically observed therapy response of tumors and to reproducibly establish treatment surviving breast cancer cells, we utilized an in vitro model previously developed in our laboratory for enriching drug tolerant persister cells following chemotherapy [37]. Four breast cancer cell lines, representing different molecular subtypes and characteristics were treated with high-dose doxorubicin (DOX), selected to kill > 90% of the cells (120 nM for MCF7, 70 nM for T47D, 150 nM for MDA-MB-231 and 200 nM for Hs578T, derived from Fig. 2B) via apoptosis (Figure S2), leaving only a small fraction of the population alive (Fig. 1A). Only approximately 7.25% MCF7, 8.2% T47D, 1.85% MDA-MB-231 and 0.29% Hs578T of the cells were able to evade apoptosis, while exhibiting prominent morphological and molecular characteristics of senescence. Most of the cells and their nuclei were enlarged, the cell bodies became flattened, and stained positively for senescence-associated ß-galactosidase (SA-ß-Gal) substrate X-gal (Fig. 1B). We also observed the overexpression of the cyclin-dependent kinase inhibitor protein CDKN1A (p21) [38] and the downregulation of the nuclear lamina building block LMNB1 [39], both widely used markers for the identifying senescent cells (Fig. 1C). Additionally, the cells exhibited extensive DNA damage, increased mitochondrial and lysosomal mass, and developed a single, fused nucleolus (Fig. 1D). Surprisingly, despite senescence being considered irreversible, a small number of cells (0.00001 - 0.00004%) managed to escape TIS in each cell line, leading to repopulation in all experiments. In these REPOP cells, the senescent phenotype was reversed, and proliferation rapidly resumed, suggesting that in breast cancer cells TIS is only transient.

Fig. 1

Chemotherapy-surviving cells transition into a transient senescent state. A Representative growth kinetics of cell cultures following 5-day doxorubicin (DOX) treatment. DOX was administered on day 0, and the medium was refreshed on day 5. The minimum cell counts are highlighted on the curves (red). B X-Gal staining of control (CTR), therapy-induced senescent (TIS), and repopulated (REPOP) cells, accompanied by quantification of staining intensity.C Western blot analysis of senescence marker CDKN1A and LMNB1 protein expression in CTR, TIS, and REPOP cells, with quantification of relative protein levels. D Fluorescence microscopy detection of DNA damage (γ-H2AX), senescence-associated β-galactosidase (SA-β-Gal) activity, mitochondria, lysosomes, and nucleoli in CTR and TIS cells. Scale bar: 20 µm

TIS breast cancer cells possess a unique pattern of drug resistance and sensitivity

Since chemotherapy is administered to patients in repeated cycles, we evaluated how TIS cells respond to a second dose of DOX 12 days after the initial treatment (Fig. 2A). Surprisingly, these cells demonstrated significant resistance to the repeated treatment. However, once the cells escaped senescence and repopulated, the DOX-resistant phenotype was reversed (Fig. 2B). Additionally, when REPOP cells were re-treated with DOX, they re-entered senescence and regained drug resistance, indicating that even after re-sensitization, TIS provides substantial protection against repeated treatments. To investigate cross-resistance to other drugs, we screened 63 FDA-approved compounds with various mechanisms of action and mapped the resistance/sensitivity profile of TIS cells. Resistance to antimetabolites was a common feature, as 5-azacytidine, 5-fluoro-2’-deoxycytidine, cladribine, cytarabine, clofarabine and troxacitabine were all ineffective at killing TIS cells (Fig. 2C, Table S2). Inhibition of topoisomerase II (mitoxantrone, pixantrone, voreloxin) and polo-like kinase 1 (volasertib) was also ineffective, as was DNA alkylation by agents such as chlormethine, chlorambucil, and melphalan. The TIS phenotype also showed resistance to the neddylation inhibitor pevonedistat, the farnesyltransferase inhibitor tipifarnib and the FLT3 inhibitor gilteritinib. However, surprisingly, TIS cells remained sensitive to quizartinib, another FLT3 inhibitor. In contrast, certain compounds were effective across all cell states, including CTR, TIS and REPOP. Given the reliance of breast cancer cells on histone deacetylation mediated epigenetic regulation, the response to HDAC inhibitors (belinostat, HDAC- 42, panobinostat, pracinostat, ricolinostat, romidepsin, vorinostat) was similar between CTR and TIS cells, as was the response to the PI3K inhibitor duvelisib and the proteasome inhibitors bortezomib and ixazomib. However, TIS cells were universally resistant to another proteasome targeting drug, carfilzomib. This unique drug resistance/sensitivity profile cannot be attributed solely to the lack of proliferation, as other compounds targeting rapidly dividing cells were still able to kill TIS cells. For example, while TIS cells were resistant to dinaciclib, the multi-CDK inhibitors AT7519 and SB-1317 which require an active cell cycle, were toxic to both proliferating and non-proliferating cells. Similarly, Bruton's tyrosine kinase inhibitor ibrutinib was toxic to both CTR and TIS cells across all cell lines. Multi-tyrosine kinase inhibitors generally showed toxicity to both non-senescent and TIS cells with masitinib, nintedanib, sorafenib and sunitinib killing nearly all cell types in at least three cell lines, while TIS cells remained completely resistant to crenolanib. Overall, 17 compounds were non-toxic at the tested concentrations, but of the remaining 46, TIS cells exhibited resistance to 23 drugs in at least three cell lines, indicating that TIS represents a transient yet significant phenotype of drug resistance.

Fig. 2

Therapy-induced senescent (TIS) cells exhibit resistance to a broad spectrum of compounds with diverse mechanisms of action. A Schematic representation of the experimental workflow, including senescence induction and subsequent drug screening assays. B Doxorubicin (DOX) sensitivity profiles of control (CTR, black), TIS (red), repopulated (REPOP, gray), and re-induced senescent (re-TIS, orange) cells. C Heatmap illustrating drug sensitivity to 46 FDA-approved anticancer compounds across the four breast cancer cell states. Red indicates at least a threefold increase in resistance of TIS cells relative to CTR cells, while green denotes no significant difference in sensitivity between CTR and TIS cells. D Schematic summary of all cell states (CTR, TIS, REPOP and re-TIS) investigated in the cytotoxicity experiments. E Dose–response analysis of CTR (black), TIS (red), REPOP (gray), and re-TIS (orange) cells treated with 8 selected compounds. F Sensitivity profiles of CTR (black), TIS (red), REPOP (gray), and re-TIS (orange) cells treated with clinically relevant breast cancer therapies

To demonstrate that TIS alone can protect cancer cells repeatedly from subsequent rounds of treatment without the emergence of any other drug resistance mechanism, REPOP cancer cells were re-treated with DOX to induce TIS again (re-TIS, Fig. 2D). The cells were then exposed to 15 selected compounds to assess drug resistance (Fig. 2E, Figure S3A). The re-TIS cells displayed almost identical resistance profiles to original TIS cells. Compounds that were ineffective against TIS cells in all cell lines but were not tested on re-TIS cells shown in Figure S3B. Meanwhile drugs that exhibited limited efficacy in killing TIS cells in three or two out of four cell lines are detailed in Figure S3 C and D, respectively.

Finally, we tested drugs routinely used in breast cancer treatment in the same manner (Fig. 2F). The antimetabolite gemcitabine, along with the taxanes paclitaxel and docetaxel — cornerstones of breast cancer therapy for decades — effectively eliminated both CTR and REPOP cells, but were ineffective against the TIS and re-TIS phenotype.

TIS breast cancer cells are only partially sensitive to Bcl-2 inhibition

To further investigate the TIS phenotype we tested navitoclax, a well-known Bcl-2 inhibitor, which earned wide recognition as a senolytic compound. While we confirmed the senolytic activity of navitoclax in cytotoxicity assays (Fig. 3A), long-term treatment produced ambiguous results (Fig. 3B, Figure S4 A). Given the ongoing debate over the specificity and selectivity of Bcl-2 inhibitors, we conducted an experiment combining DOX and navitoclax during senescence induction to investigate whether depleting the TIS cells could prevent repopulation. Interestingly, navitoclax had to be continuously administered from the onset of the treatment to significantly reduce relapse in our assay, suggesting that BCL2 overexpression may be an early response to DOX rather than a senescence-specific alteration. However, the outcomes varied widely, ranging from complete eradication of cells to no observable effect. In a clinically relevant mouse model of triple-negative breast cancer (TNBC, Fig. 3C), a single dose of pegylated liposomal DOX (DOXIL) induced complete tumor response, with no detectable or palpable tumors for 40–60 days, mirroring our in vitro findings (Fig. 3D). Despite this, navitoclax failed to extend survival or demonstrate any disease-modifying effect in this model (Fig. 3E). Senescent cells are hypothesized to evade apoptosis by overexpressing Bcl-2, an anti-apoptotic factor that inhibits pro-apoptotic proteins BAX and BAK. Therefore, inhibiting Bcl-2 should selectively induce cell death in senescent cells. To test this hypothesis, we applied several senolytic drug treatments in our cellular assay. TIS cells responded differently depending on the potency and selectivity of the tested molecules (Fig. 3F, Table S3). The selective Bcl-2 inhibitor venetoclax (from Fig. 2C) exhibited equal toxicity toward both CTR and TIS cells, while navitoclax, ABT-737 and A-1331852 — inhibiting Bcl-2/Bcl-XL/Bcl-w, Bcl-2/Bcl-XL and Bcl-XL, respectively, — were selectively toxic to TIS cells. This observation suggests that bypassing apoptosis in TIS could be driven by Bcl-XL rather than Bcl-2. While this explanation appeared plausible, repeated monitoring of Bcl-2 and Bcl-XL during and after DOX treatment revealed no correlation between Bcl inhibitor efficacy and protein expression levels (Figure S4B). Similarly, the localization and/or heterogeneity of these proteins did not support the link between Bcl-2 inhibition and viability (Figure S4C). Moreover, the mRNA expression levels of BCL2 and BCL2L1 were comparable across CTR, TIS, and REPOP cells (Figure S4D). Although compounds like dasatinib, fisetin, quercetin and piperlongumine have been reported to selectively kill senescent cells via different targets, TIS cells in our assays displayed resistance to these agents. To further support that apoptosis can be induced in TIS cells, we performed live-cell monitoring with Annexin V staining during belinostat treatment (Figure S4E,F). However, apoptotic pathways in TIS cells remained largely unchanged (Figure S4G,H).

Fig. 3

Therapy-induced senescence (TIS) breast cancer cells exhibit partial sensitivity to senolytic treatment. A Navitoclax sensitivity profiles of control (CTR, black), TIS (red), repopulated (REPOP, gray), and re-induced senescent (re-TIS, orange) cells. B Crystal violet staining of breast cancer cell cultures treated with doxorubicin (DOX) alone or in combination with Navitoclax (DOX + Navitoclax). C Schematic representation of the experimental design comparing the effects of DOXIL versus DOXIL + Navitoclax treatment in an in vivo model. D Representative tumor growth curve of Brca1-/-;p53-/- tumors treated with the maximum tolerated dose of DOXIL. E Kaplan–Meier survival analysis of Brca1-/-;p53-/- tumor-bearing mice treated with DOXIL (n = 7) or DOXIL + Navitoclax (n = 13). Statistical analysis shows p = 0.3888. F Sensitivity analysis of CTR (black) and TIS (red) cells treated with six different senolytic compounds

The TIS phenotype is transient and transcriptomically distinct compared to CTR and REPOP cells

To characterize the TIS cells across all 4 breast cancer lines, we analyzed the differentially expressed gene (DEG) sets using bulk RNA sequencing (Fig. 4A). Normalized gene expression data revealed distinct differences between TIS samples and CTR and REPOP cells (Fig. 4B). This was further supported by PCA analysis, which clearly separated TIS cells from both CTR and REPOP cells, indicating that all cell lines responded similarly to DOX and developed the TIS phenotype in a consistent manner (Fig. 4C). DEG analysis revealed that while there were notable differences between the four cell lines, all TIS cells share 929 DEGs compared to CTR cells, regardless of their origin (Fig. 4D and E). Strikingly, the vast majority of these genes (896) were overexpressed, with only 33 (3.5%) being downregulated. In contrast, during the transition from TIS to REPOP, the expression of 722 genes changed, with only 16 (2.2%) being upregulated while the expression of 706 genes significantly decreased (Fig. 4F and G). By comparing DEGs between CTR and REPOP cells, we found only 1 gene that remained overexpressed and none that remained downregulated across all four cell lines, indicating that the original gene expression pattern was largely re-established in the REPOP cells (Fig. 4H and I). To confirm that the TIS state is transcriptionally transient, we analyzed the direction of gene signature changes. The transition from CTR to TIS induced a dramatic overexpression of genes, while the transition from TIS to REPOP showed the opposite effect (Figure S5). Furthermore, when we followed the expression of 13 genes that were significantly upregulated in TIS cells, we observed that in REPOP cultures these same genes exhibited the reverse trend, becoming downregulated. We further validated the RNA sequencing results using qPCR (Fig. 4J, Figure S6), expanding our analysis to include 11 selected genes. This set comprised senescence markers (CDKN1A, LMNB1, MKI67), apoptosis regulators (BCL2, BCL2L1), a highly overexpressed gene (ITGB6), and drug targets (ACTA2, GSDMC, KRT6A, PDE1A, PSMA8), all of which showed strong concordance with our RNA-seq data in all four breast cancer cell lines.

Fig. 4

Therapy-induced senescence (TIS) cells display a unique gene expression profile. A Workflow of bulk RNA sequencing. Samples were collected from untreated control (CTR) cells on Day 0. Doxorubicin (DOX) treatment was applied (red arrow) for 5 days, and RNA was isolated from TIS cells on Day 12, 7 days post-drug removal. REPOP cells were harvested between Days 24–36 when culture confluency reached 80–90%. B Box plots representing global gene expression differences between TIS and CTR/REPOP cells across all cell lines. C Principal component analysis (PCA) of the gene expression data showing in the case of all four cell lines clustering of CTR and REPOP cells, which exhibit transcriptomic similarity, while TIS cells form a distinct cluster, highlighting their unique gene expression profile. D, E UpSet plots of differentially expressed genes (DEGs) from CTR vs. TIS comparisons, illustrating upregulated (D) and downregulated (E) genes. Set sizes are indicated, with red columns representing DEGs shared across all cell lines. F, G UpSet plots of upregulated (F) and downregulated (G) DEGs from TIS vs. REPOP comparisons, with set sizes and shared DEGs (red columns) shown. H, I UpSet plots of upregulated (H) and downregulated (I) DEGs from CTR vs. REPOP comparisons, with set sizes indicated. J Relative quantification of BCL2L1, CDKN1A, and ITGB6 expression in MCF7, T47D, MDA-MB-231, and Hs578T breast cancer cell lines under CTR, TIS, and REPOP conditions using qRT-PCR. The bar plot illustrates the relative mRNA expression levels (2^-ΔΔCt) of BCL2L1, CDKN1A, and ITGB6 across the four breast cancer cell lines under the three experimental conditions: CTR (red), TIS (green), and REPOP (blue), with biological replicates distinguished by solid fill (Rep1) and diagonal hatching (Rep2,///). Bars represent mean expression levels, with error bars indicating standard deviation, and individual technical replicates (n = 4 per condition) displayed as light gray diamond markers. The qRT-PCR measurements were performed in two biological replicates per condition, ensuring reproducibility. The y-axis is set to a symlog scale to accommodate the broad range of expression levels, and gene labels are rotated for clarity. K, L Pathway analysis based on RNA-seq data from MCF7, T47D, MDA-MB-231, and Hs578T cells, highlighting upregulated and downregulated pathways in CTR vs. TIS (K) and TIS vs. REPOP (L) comparisons. Top and bottom 10 pathways are shown based on the normalised enrichment score (NES). M Venn diagram of mRNAs significantly upregulated (log2 fold change > 1, adjusted p-value < 0.05) in the CTR-to-TIS transition and simultaneously downregulated (log2 fold change < 1, p-value < 0.05) in the TIS-to-REPOP transition. N Heatmap of scaled gene expression changes in genes associated with resistance to tested compounds. Expression values were centered and scaled for each gene across all samples

Given the significant similarities among TIS cells, across the four tested cell lines, we conducted gene set enrichment analysis on the mRNA expression data to identify common characteristics of the TIS phenotype (Fig. 4K). The analysis revealed that TIS cells: (1) are non-proliferative, as 5 out of 6 proliferation-related gene sets (G2M Checkpoint, E2F Targets, MYC Targets V1 and V2, Mitotic Spindle) from the “hallmark” sets of the Molecular Signatures Database (MSigDB), were significantly downregulated, with the exception of the p53 pathway, (2) rely on KRAS signaling, as gene sets comprising both up- and downregulated genes related to KRAS activation (KRAS Signaling UP and DN) were significantly upregulated, and (3) exhibit significantly reduced DNA repair capacity, as the DNA Repair gene set was downregulated. Additionally, all 7 immune-related hallmark gene sets (Allograft Rejection, Coagulation, Complement, Interferon alpha Response, Interferon gamma response, IL6-JAK-STAT3 Signaling, Inflammatory Response) were altered, suggesting that TIS may have a significant immune modulatory effect. After escaping TIS, all of these changes were reversed (Fig. 4L).

Senescence was also induced in healthy human foreskin fibroblast (HFF) cells using DOX to compare TIS-related changes in malignant and non-cancerous cells (Figure S7). While the treatment caused significant DNA damage in HFF cells (Figure S7 A), induced TIS in virtually all surviving cells (Figure S7B), the difference between CTR and TIS cells in the PCA analysis were less pronounced than it was observed in the case of breast cancer cell lines (Figure S7 C). Additionally, the distribution of up- and downregulated genes was more balanced in the HFF cells (Figures S7D). A comparison of hallmark pathways of MSigDB in the four cancer cell lines and in HFF cells revealed 11 shared gene sets (KRAS signaling up, KRAS signaling down, Coagulation, IL6-JAK-STAT Signaling, TNFa Signaling via NFκB, Unfolded protein response, Myc targets v2, Mtorc1 signaling, Mitotic Spindle, E2F targets, G2M checkpoint). However, one of these, TNFa Signaling via NFκB, showed changes in the opposite direction in non-malignant HFF cells (Figure S7E). A similar analysis focusing on biological processes (KEGG) identified five shared gene sets between TIS HFF and breast cancer cells that were altered in the same manner (Neuroactive ligand receptor interaction, Retinol metabolism, Cytokine cytokine receptor interaction, DNA replication, Cell Cycle), while the other 15 most differentially expressed pathways were dissimilar (Figure S7 F,G). Taken together, the analysis revealed that while HFF cells share some common TIS-related pathways with breast cancer cells, the response is less pronounced in non-malignant cells, with a more balanced gene regulation and some key pathways, such as TNFa signaling via NFκB, showing opposite trends. Additionally, significant differences were observed in 15 key biological processes between TIS in HFF and breast cancer cells.

To find the core gene set exclusive to the TIS cells, we identified genes that were transiently overexpressed only in TIS cells across all four cell lines (Fig. 4M). A total of 316 mRNAs which were upregulated in TIS but downregulated once the cells resumed proliferation (Figure S8 A). This TIS-specific gene set suggests that TIS cells secrete elevated levels of cytokines, likely due to the Senescent-associated Secretory Phenotype (SASP), rely on the JAK/STAT signaling pathway and retinol metabolism, and express a wide array of ABC transporters (Figure S8B).

While the transcriptomes of CTR and REPOP cells were highly similar, indicating that the REPOP cells indeed resemble their parental lines, we could identify residual gene expression changes that may serve as a molecular memory of the cells recovering from TIS. We selected the genes that were overexpressed (log2 FC > 1, adjusted p-value < 0.05) in TIS cells of the breast cancer cell lines and maintained these changes in the REPOP cells when compared to the CTR. As mentioned earlier, there was only one gene (IFIT1) that fulfilled these stringent conditions, but 22 genes fell into this category in at least 3 of the studied cell lines (Figure S8C). MDA-MB-231 cells appear to be outliers in this respect, but out of the 22, 12 genes from this group show similar changes in these cells, only with less pronounced expression changes or lower significance (Table S4). GO biological function analysis (Figure S8D) indicates that the genes in this category are related to interleukin and interferon signaling, indicating long-term changes in immunogenecity after TIS.

To determine whether TIS cells actively modulate their immune environment, we conducted a sensitive cytokine expression analysis, focusing on ten well-characterized immunomodulatory cytokines (Figure S9 A,B). In both MCF7 and T47D TIS cells, mRNA expression levels of IL2, IL2Rα, IL3, IL6, IL10, IL13, TNFα, IFNγ, and CSF2 were elevated compared to CTR cells, whereas IL4 expression remained negligible across both cell lines and cell states. Notably, IL2Rα, IL6, and IL13 exhibited robust upregulation in TIS cells from both cell lines, with 18-, 70-, and sevenfold increases in MCF7 and 11-, 32-, and fivefold increases in T47D, respectively. While T47D TIS cells displayed a pronounced increase in IL10 expression (19-fold), MCF7 TIS cells showed only a modest elevation (1.7-fold).

Functionally, several of these cytokines are well-documented mediators of immune suppression and tumor progression. IL6 plays a key role in senescence induction and tumor promotion by regulating myeloid-derived suppressor cells (MDSCs), which suppress T-cell-mediated antitumor immunity [40]. IL10 is a potent immunosuppressive cytokine with immune-regulatory and angiogenic properties, facilitating tumor cell survival, proliferation, and metastasis by inhibiting antitumor immune responses [41]. IL13 has been implicated in indirect immunosuppression through the promotion of tumor-associated macrophage (TAM) differentiation, leading to TGF-β secretion and the establishment of a tumor-promoting microenvironment [42].

Interestingly, the drug resistance observed in TIS cells can be explained only in a few cases with known mechanisms. We examined the expression levels of the usual drug resistance factors, such as efflux transporters, DNA repair mechanisms and genes which are targets of the screened compounds or known to inactivate drugs or counter their effects (Fig. 4N, Figure S10A). CDK1, 2, 4, 5, 7 and 9 are inhibited by AT7519, SB-1317 and dinaciclib, but the expression profile of these genes did not correlate with either resistance or sensitivity to CDK inhibitors. The loss or reduced expression of FDPS gene should have sensitized cells to tipifarnib [43], but TIS cells were showed resistance. Overexpression of the antiapoptotic Bcl-2 could explain TIS cell tolerance to many drugs and their sensitivity to venetoclax [44], but no such link was found. Increased expression of RRM1, 2 and 2B is often detected in DOX, gemcitabine and docetaxel resistant tumors [45], yet these genes were rather downregulated than overexpressed in TIS cells. SAMHD1 and DCK have been reported to mediate wide scale resistance to antimetabolites such as cytarabine, clofarabine and cladribine [46,47,48], but we found increased expression only of SAMHD1 and only in two cell lines in TIS cells. Investigating the expression of targets for biological therapies produced surprisingly mixed results. Targets of crenolanib/giltertinib/quizartinib (FLT3/PDGFRA, KIT), gefitinib (EGFR), ibrutinib (BTK), idasanutlin (MDM2), bortezomib/ixazomib (PSMB5) and nintedanib (FGFR1) were not upregulated in TIS cells. On the other hand, IDH2, the target of enasidenib, was downregulated in TIS cells, but not resulted in resistance as CTR and TIS cell were similarly sensitive to the drug. Similarly, despite TIS cells overexpressed NEDD8, the main mediator of neddylation, the cells were still resistant to the NEDD8 inhibitor pevonedistat. In two cases, the resistance against Selinexor and volasertib in TIS cells can be explained by the deregulation of their specific targets XPO1 and PLK1, respectively. Out of three investigated DNMTs, 1 and 3B were almost always downregulated in TIS cells, while decreased expression of 3A was found in two cell lines, suggesting a possible mechanism to reduce efficiency of DNMT inhibitors like 5-azacytidine and 5-fluoro-2’-deoxycytidine [49, 50]. Downregulation of TOP2A, which could be a key resistance mechanism against DOX, mitoxantrone, pixantrone and voreloxin, was detected in three cell lines. ABCB1, ABCC1, ABCC3 and ABCG2 drug transporters are known to recognize and expel a wide range of structurally and mechanistically diverse compounds [51]. ABCC1 and C3 were expressed cell line specifically, but ABCB1 – in accordance with our previous finding [37] – and ABCG2 were overexpressed in TIS cells in all cell lines. Surprisingly, due to the wide substrate specificity of ABCB1, TIS cells should have been resistant to belinostat, bortezomib, gefitinib, sorafenib and sunitinib too, a feature TIS cells clearly not possess, therefore we tested whether ABCB1 inhibition with tariquidar has any effect on DOX sensitivity (Figure S10B). This experiment proved that despite ABCB1 is significantly overexpressed in TIS in all cell lines, it's not enough to protect cells from drug treatment. Despite our extensive gene expression analysis, no discernible alterations were found in the classical resistance pathways or drug target genes that could account for the unique drug resistance and sensitivity profile observed in TIS cells, suggesting that the mechanisms underlying TIS-associated drug responses may involve non-canonical or context-dependent regulatory networks yet to be fully understood.

Single-cell transcriptome profiling reveals TIS cells as a unique and distinct cell population

To further characterize the transcriptional changes and study the dynamics of TIS, as well as the molecular basis of potential resistance and escape mechanisms, we performed single-cell RNA sequencing (scRNA-seq) of MCF7 and T47D breast cancer cell lines (Fig. 5A and B). To ensure the quality and usability of our scRNA-seq data, we first assessed the datasets by analyzing replicates and comparing differentially expressed genes from pseudo-bulk analysis to bulk RNA-seq. We observed minimal to no batch effect among the replicates and high concordance between bulk and pseudo-bulk comparisons, confirming the reliability of our scRNA-seq datasets (Figure S10C and D). As expected from our bulk RNA-seq analysis, unsupervised clustering followed by uniform manifold approximation and projection (UMAP) of single-cell gene expression profiles revealed a distinctly different transcriptional state for TIS cells compared to the CTR in both cell lines. In contrast, REPOP cells clustered closely with the CTR population and, in the case of T47D cells, were nearly identical to the original CTR population, underscoring the reversible and transient nature of the TIS state. Interestingly, the replicates of MCF7 REPOP samples exhibited substantial variability, even though they were collected at the same time post-treatment, indicating that the escape from senescence may occur with distinct and variable kinetics. Importantly, when the 316 shared genes, previously identified from the bulk RNA-seq (Fig. 4N), were projected onto single cell RNA sequencing results from MCF7 and T47D, the pattern specifically identified TIS cells (Figure S10E and F) confirming that the discovered gene set is TIS-specific. Integrated analysis and joint normalization of MCF7 and T47D cell lines highlighted both cell type-specific and senescence-related transcriptomic alterations (Fig. 5C). This analysis demonstrates that while the inherent gene expression profile of each cell type remains the most defining feature, the TIS cell state introduces a secondary, yet dominant and distinct, dimension. We observed no overlap between TIS and CTR populations on the different UMAPs and did not find batch effect among the replicates (Figure S11 A-D).

Fig. 5

Therapy-induced senescence (TIS) cells that escape senescence drive repopulation after chemotherapy. A UMAP projection of single-cell transcriptomes from the MCF7 breast cancer cell line, illustrating distinct clustering of control (CTR, dark brown), therapy-induced senescent (TIS, scarlet), and repopulating (REPOP, light brown) cell populations. The separation of these clusters indicates transcriptionally distinct states, with TIS cells forming a well-defined cluster distinct from CTR and REPOP populations. The REPOP population shows a partial transcriptional shift toward CTR, reflecting its reversal from the TIS state. B UMAP projection of single-cell transcriptomes from the T47D breast cancer cell line, similarly showing distinct clustering of CTR (dark green), TIS (crimson), and REPOP (light green) cell populations. The clustering pattern resembles that observed in MCF7 cells, with a well-separated TIS population and REPOP cells positioned between TIS and CTR clusters, suggesting partial transcriptional reversion. C Integrated UMAP analysis of MCF7 and T47D cell lines, combining data from both models to highlight cell type-specific transcriptomic profiles. Control populations (MCF7 CTR in dark brown; T47D CTR in dark green) form distinct clusters, whereas TIS populations from both cell lines (scarlet for MCF7; crimson for T47D) also exhibit clear separation from their respective CTR counterparts. The integration further reveals that despite the transcriptional similarities in senescence-associated programs, MCF7 and T47D maintain cell-line-specific transcriptomic differences, as reflected in their segregated distributions. D Feature plots showing the expression of key senescence and proliferation-related markers in individual MCF7 cells. CDKN1A (p21) is upregulated in TIS cells (top left panel), confirming cell cycle arrest. Conversely, markers associated with proliferation, including LMNB1, TOP2A, and MKI67, are downregulated (top right and bottom panels), consistent with the senescent phenotype. The presence of scattered high-expressing cells within the TIS population suggests the existence of ‘escaper’ subpopulations that may retain some proliferative capacity. E Feature plots of CDKN1A, LMNB1, TOP2A, and MKI67 expression in T47D cells, showing a similar transcriptional profile to MCF7 cells. CDKN1A is significantly upregulated in TIS cells, while LMNB1, TOP2A, and MKI67 are markedly downregulated. As in MCF7, a fraction of TIS cells display non-uniform expression of these markers, indicating potential escape from the senescent state. F UMAP projection of MCF7 cells overlaid with cell cycle phase information (G1: green; S: blue; G2/M: red). Pie charts illustrate the proportional distribution of cells in each phase across the CTR, TIS, and REPOP populations. TIS cells predominantly reside in the G1 phase (79.7%), reflecting irreversible cell cycle arrest, whereas REPOP cells show an increased proportion of cycling cells, particularly in the G2/M phase, indicating their proliferative re-entry. G UMAP projection of T47D cells colored by cell cycle phases, with corresponding pie charts depicting phase distributions in CTR, TIS, and REPOP populations. TIS cells in T47D exhibit a similar G1 arrest phenotype as observed in MCF7, while REPOP cells regain a more balanced cell cycle distribution, mirroring their recovery from senescence. H Gene set enrichment analysis (GSEA) of differentially expressed genes in MCF7 TIS cells compared to CTR, highlighting significant enrichment of senescence-associated pathways (e.g., inflammatory response, DNA damage signaling) and suppression of proliferation-associated pathways. Normalized enrichment scores (NES) are shown, with positive values indicating upregulated pathways and negative values indicating suppressed pathways. I GSEA results for T47D TIS cells, demonstrating pathway-level alterations similar to those observed in MCF7, with enrichment of senescence-associated programs and suppression of cell cycle progression. J Schematic representation of the senescence induction and reversion process. Cells undergo therapy-induced senescence (TIS) following exposure to doxorubicin (DOX). Over time, a subset of TIS cells escape growth arrest and re-enter the cell cycle, forming the REPOP population. This process involves transcriptional reprogramming, with a balance between senescent, proliferating, and apoptotic fates. K UMAP trajectory analysis of MCF7 cells depicting the transition from CTR to TIS (bottom panel) and from TIS to REPOP (top panel). Cells are colored based on pseudotime, capturing the progressive shift in transcriptional states. TIS cells form a distinct branch, while REPOP cells demonstrate convergence toward CTR, reflecting their transcriptional plasticity. L UMAP trajectory analysis of T47D cells, analogous to MCF7. The bottom panel illustrates the transition from CTR to TIS, while the top panel depicts the transition from TIS to REPOP. The REPOP population becomes almost identical to the CTR cluster, exhibiting highly overlapping transcriptional profiles, even more profoundly than in MCF7. This supports a model of senescence escape and complete proliferative recovery. M Heatmap of gene expression changes in a curated set of genes associated with drug resistance, senescence regulation, and cell cycle control across CTR, TIS, and REPOP states in both MCF7 and T47D cell lines. Genes exhibit dynamic expression patterns, with key senescence markers upregulated in TIS and downregulated upon REPOP transition, while drug resistance-associated genes show variable trends between cell lines. This highlights the complex interplay between senescence, proliferation, and therapy resistance

Senescence factors arresting cell cycle measured by scRNA-seq

By visualizing the expression levels of the CDKN1A, a cell cycle inhibition and known senescence marker gene, in individual cells, we observed its significantly elevated expression in the TIS populations (Fig. 5D and E, top left panels). Along with the mostly undetectable expression of the nuclear lamina component LMNB1 gene (Fig. 5D and E, bottom right panels), another frequently used marker of senescence, these findings strongly indicate a robust cell cycle arrest in the TIS cell population, which is not observed in the CTR or REPOP cells. The expression of these genes showed a strong correlation with the previously observed protein levels (Fig. 1C). Detailed analysis of additional cell cycle markers, such as TOP2A and MKI67, further corroborates the presence of non-dividing cells within the senescent population (Fig. 5D and E, top right and bottom left panels). However, we identified a subpopulation of cells, termed'escapers,'which, while simultaneously expressing growth activators and suppressors, may re-enter the cell cycle and form the basis for repopulation, as indicated by the dotted-line rectangles in Fig. 5D and E. In alignment with this observation, we projected cell cycle-specific gene expression patterns onto all cells to calculate the ratios of the G1 (gap), S (synthesis), G2, and M (mitosis) cell cycle phases among the three experimental conditions (CTR, TIS, and REPOP) in both cell lines (Fig. 5F and G). In the CTR populations of both cell lines, we observed approximately equal ratios of cells in the G1, S, and G2/M phases. This distribution shifted dramatically towards the G1 phase in senescent cells, with 79.7% and 66% in MCF7 and T47D cell lines, respectively. This further supports the notion that senescent cells are in cell cycle arrest, potentially serving as a crucial mechanism for evading the effects of various drugs. After repopulation, the cell cycle balance was almost completely restored to resemble the CTR population.

Single cell transcriptomics identifies activated and suppressed pathways related to TIS cell state

Single-cell transcriptomics has provided a detailed view of the molecular landscape in TIS cells, revealing specific alterations that may endow these cells with unique properties to survive or resist drug treatments. In the MCF7 cell line (Fig. 5H), three key DNA repair pathways — Homologous Recombination, Base Excision Repair and Fanconi Anemia Pathway — were significantly suppressed (Figure S12 A,C,E,G,I). This downregulation suggests a reduced capacity to maintain genomic stability, contributing to TIS cell survival despite DNA damage. ATP-dependent Chromatin Remodeling was also suppressed, limiting chromatin changes for DNA repair and gene regulation during senescence. Both Cell Cycle and DNA Replication pathways were suppressed, likely keeping TIS cells in growth arrest and enabling them to evade the cytotoxic effects of chemotherapeutics targeting dividing cells. Conversely, pathways like Estrogen Signaling, ECM-Receptor Interaction, and Cell Adhesion Molecules were activated, suggesting increased cellular signaling and adhesion, potentially altering the tumor microenvironment to support TIS cell survival.

In the T47D cell line (Fig. 5I), similar to MCF7 cells, all three previously mentioned DNA repair pathways, as well as Mismatch Repair and Nucleotide Excision Repair, were suppressed (Figure S12B,D,F,H,J). In addition to these pathways, Spliceosome and One-Carbon Pool by Folate pathways were also suppressed, indicating a broader disruption in cellular metabolism and mRNA processing, which may further stabilize the TIS state.

These findings underscore the complexity and adaptability of TIS cells, revealing reduced DNA repair capabilities and significant alterations in signaling pathways. This molecular reprogramming likely supports the maintenance of the TIS cell state and the resistance of the cells to a variety of therapeutic agents.

Trajectory analysis reveals the reversible nature of TIS

To better understand cell state transitions and transcriptomic changes, we performed trajectory analysis on CTR, TIS and REPOP cells (Fig. 5J). Differential gene expression analysis was conducted on MCF7 and T47D (CTR vs. TIS). The overlapping genes from both comparisons were then cross-referenced with the trajectory analysis results. All 93 genes were found to be differentially expressed. This analysis confirmed that the CTR and REPOP stages are closely related in terms of their transcriptomic profiles (Fig. 5K and L). Furthermore, trajectory analysis indicated that release from cell cycle arrest is critical for the transition from TIS to REPOP. However, these trajectories could be heavily influenced by the expression of genes directly related to the cell cycle which could falsely separate TIS cells from the others. Therefore, the analysis was performed again with the exclusion of these genes, but the results were the same (Figure S13). These findings emphasize the reversible nature of TIS, highlighting the potential for cells to re-enter the cell cycle and proliferate once the arrest is lifted.

Gene expression patterns cannot explain the drug resistance of TIS cells

By investigating the expression of the previously selected drug resistance genes we found no common mechanism that could explain the significantly increased drug tolerance of TIS cells from both cell lines (Fig. 5M). Similar to the results obtained from bulk RNA sequencing, there are genes that could seemingly play a role in drug resistance in one of the cell lines, but comparing their expression levels to the other line or REPOP cells disproves it. For example, TOP2A downregulation could be the reason why MCF7 cells are resistant to topoisomerase II inhibitors, however T47D cells are also resistant to topo II poisons, while overexpressing TOP2A. Conversely, decreased NEDD8 levels can cause pevonedistat resista