Abstract

Background

   Therapy-induced senescence (TIS) is considered a permanent cell cycle
   arrest following DNA-damaging treatments; however, its irreversibility
   has recently been challenged. Here, we demonstrate that escape from TIS
   is universal across breast cancer cells. Moreover, TIS provides a
   reversible drug resistance mechanism that ensures the survival of the
   population, and could contribute to relapse.

Methods

   TIS was induced in four different breast cancer cell line with
   high-dose chemotherapy and cultured until cells escaped TIS. Parental,
   TIS and repopulating cells were analyzed by bulk and single-cell RNA
   sequencing and surface proteomics. A genetically engineered mouse model
   of triple-negative breast cancer was used to prove why current
   senolytics cannot overcome TIS in tumors.

Results

   Screening the toxicity of a diverse panel of FDA-approved anticancer
   drugs revealed that TIS meditates resistance to half of these
   compounds, despite their distinct mechanism of action. Bulk and
   single-cell RNA sequencing, along with surface proteome analysis,
   showed that while parental and repopulating cells are almost identical,
   TIS cells are significantly different from both, highlighting their
   transient nature. Furthermore, investigating dozens of known drug
   resistance mechanisms offered no explanation for this unique drug
   resistance pattern. Additionally, TIS cells expressed a gene set
   associated with immune evasion and a potential KRAS-driven escape
   mechanism from TIS.

Conclusion

   Our results reveal that TIS, as a transient drug resistance mechanism,
   could contribute to overcome the immune response and to relapse by
   reverting to a proliferative stage.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12943-025-02310-0.

   Keywords: Therapy-induced senescence, Drug resistance, Breast cancer,
   Senescence escape, Tumor relapse

Background

   Cancer claims 10 million lives annually, largely due to the emergence
   of drug resistance [[54]1]. Drug resistant cancer cells employ various
   strategies, such as inactivating [[55]2] or effluxing drug molecules
   [[56]3], enhancing DNA repair to counteract DNA damage [[57]4], evading
   apoptosis by distorting apoptotic pathways [[58]5], reducing target
   proteins expression [[59]6, [60]7] or temporarily arresting the cell
   cycle to avoid drugs that target rapidly dividing cells [[61]8]. Recent
   discoveries have further expanded the repertoire of resistance
   mechanisms: disruption in the tumor microenvironment can promote drug
   resistance [[62]9] and treatment induced epigenetic changes give rise
   to drug-tolerant persister cells [[63]10]. Tumor heterogeneity, where
   different cell populations within the tumor exhibit diverse molecular
   characteristics and drug sensitivities, also promotes resistance
   [[64]11] as does the ability to evade immunosurveillance [[65]12].
   However, despite these discoveries, attempts to translate these
   findings into clinical practice have largely failed, leading to the
   sobering realization that the contribution of each individual mechanism
   to overall drug resistance may be more limited than previously thought.

   When cellular damage occurs, the subsequent response is contingent upon
   the severity of the injury. In cases of mild damage, cells may activate
   repair mechanisms to restore homeostasis. However, if the damage
   exceeds the cell's reparative capacity, programmed cell death pathways,
   such as apoptosis, or uncontrolled cell death via necrosis, become
   inevitable outcomes [[66]13]. However, in the narrow range between
   reversible and irreversible damage there exists a third option, called
   cellular senescence. Senescence was first described by Hayflick as a
   response to cellular aging. They observed that fetal fibroblasts could
   undergo only a limited amount of cell divisions before irreversibly
   losing their ability to replicate, a state termed replicative
   senescence [[67]14]. It was discovered that telomere shortening, caused
   by continuous cell division, triggers senescence in cells, which is why
   cancer cells frequently overexpress telomerase to avoid senescence
   [[68]15–[69]19]. In addition to aging, cellular senescence can be
   induced by various stress factors such as DNA damage, abnormal cell
   replication or oxidative stress [[70]20]. Senescence serves as a
   crucial protective mechanism for damaged cells; while these cells lose
   their ability to replicate, they can still contribute to tissue
   homeostasis through processes like remodeling, repair, and regeneration
   in certain contexts [[71]21].

   Senescence is an enigmatic stress response mainly due to its many
   interconnected subtypes [[72]21]. Replicative senescence, DNA-damage
   induced senescence, oncogene-induced senescence (OIS), oxidative
   stress-induced senescence, therapy-induced senescence (TIS) and
   epigenetically induced senescence all lead to similar outcomes, yet the
   molecular pathways driving these changes can vary significantly.
   Despite these differences, the irreversibility of senescence is so
   widely accepted that it has become a desired outcome in cancer therapy.
   If treatment cannot eliminate every tumor cell, it can still inflict
   enough damage to induce permanent cell cycle arrest, which effectively
   achieves the same goal [[73]22, [74]23]. While senescence is widely
   regarded as a permanent state, emerging studies suggest that cancer
   cells can escape senescence more frequently than previously thought.
   For example, lung cancer cells lacking p53 and p16 can re-enter cell
   cycle after TIS by overexpressing Cdc2/Cdk1 affecting approximately 1
   in a million cells [[75]24]. Additionally, in locally advanced
   non-small cell lung cancer, cells that escape TIS are often polyploid,
   a transient state in which senescence escape is more likely to occur
   [[76]25]. In B-cell lymphoma and lymphoblastic leukemia cells released
   from TIS by inducible alterations in H3 K9 trimethylation or p53
   expression exhibited increased stemness, enhanced Wnt-signaling and
   leukemia-initiating properties [[77]26]. Furthermore, lung, colon, and
   breast cancer cells have been shown to escape TIS in vitro and form
   tumors in vivo when injected into mice [[78]27].

   Here, we demonstrate that chemotherapy induces senescence in four
   distinct breast cancer cell lines, followed by a delayed escape from
   the TIS state, with cells re-entering the cell cycle weeks later.
   Remarkably, during TIS, cancer cells exhibit resistance to a wide range
   of drugs with diverse structures and mechanisms of action. Through bulk
   and single-cell RNA sequencing (scRNA-seq), we identify a novel TIS
   transcriptome shared across all cell lines, revealing that TIS cells
   are fundamentally distinct from both control (CTR) and repopulated
   (REPOP) cells. We propose that TIS is not only a reversible state but
   also a critical mechanism of drug resistance, with profound
   implications for cancer treatment and therapy design.

Methods

Cell lines

   The human mammary carcinoma cells lines (MCF7, T47D, MDA-MB-231,
   Hs578T) were obtained from the National Cancer Institute’s
   Developmental Therapeutics Program (National Institutes of Health,
   Bethesda, MD, USA). Cells were cultured in Roswell Park Memorial
   Institute (RPMI) medium (Thermo Fisher Scientific, Massachusetts, USA),
   supplemented with 10% fetal bovine serum (FBS, Thermo Fisher
   Scientific, Massachusetts, USA), 5 mmol/L glutamine (Euroclone, Pero
   MI, Italy), and 50 units/mL penicillin and streptomycin (Capricorn
   Scientific, Ebsdorfergrund, Germany). All cell lines were cultured at
   37°C with 5% CO[2].

   Human foreskin fibroblast (HFF) cell line was obtained from the
   National Cancer Institute’s Developmental Therapeutics Program. Cells
   were cultured in Dulbecco’s modified Eagle’s medium (DMEM-F12, Thermo
   Fisher Scientific, Massachusetts, USA) supplemented with 10% fetal
   bovine serum (FBS, Thermo Fisher Scientific, Massachusetts, USA), 1%
   L-glutamine (Thermo Fisher Scientific, Massachusetts, USA), 0.1%
   gentamicin (50 mg/mL, Thermo Fisher Scientific, Massachusetts, USA),
   and 16 ng/mL fibroblast growth factor 2 (Peprotech, London, UK). Cells
   were kept at 37°C and 5% CO[2].

Drugs

   The complete list of drugs can be seen in the Supplemental Information.

TIS induction

   For TIS induction cells were treated with 120 nM (MCF7), 70 nM (T47D),
   150 nM (MDA-MB-231), 200 nM (Hs578T), 210 nM (HFF) DOX and incubated
   for 5 days. After five days of incubation, treated cells were washed
   with pre-warmed PBS and the medium was renewed. MCF7, T47D, MDA-MB-231
   and Hs578T plates remained in the incubator for seven days to ensure
   that the cells had become senescent. HFF plates remained in the
   incubator for 19 days with medium change every week. REPOP cells are
   those that survived DOX treatment, underwent TIS, expressed all
   canonical senescence markers, but later re-entered the cell cycle and
   lost TIS-associated attributes. These cells were analyzed when they
   reached ~ 80% confluency. re-TIS cells were established by replating
   REPOP cells and re-exposing them to DOX to induce TIS again. Visual
   representation of TIS induction and different cell states are
   summarized on Figure S1.

SA-β-Gal staining

   Cells were plated into 96-well tissue culture plates. SA-β-Gal staining
   was performed using the Senescence β-Galactosidase Staining Kit (Cell
   Signaling Technology, Danvers, MA, USA) according to the manufacturer’s
   instructions.

In vitro cytotoxicity assay

   10.000 cells/well were plated in 96-well plates. TIS induction was
   performed as described above. After TIS induction cells were counted in
   three wells and the results were averaged, after which the appropriate
   number of CTR cells was plated. The following day, both TIS and CTR
   cells were treated with a given concentration range of different drugs
   and incubated for five days. Drug testing on re-TIS cells was conducted
   under comparable conditions. A total of 10.000 REPOP cells per well
   were seeded onto 96-well plates, and TIS was induced by treatment with
   DOX over a period of five days. Following this incubation, the treated
   cells were washed with pre-warmed PBS, and the culture medium was
   replenished. The cells were then maintained in culture for an
   additional seven days. On Day 12, the cells were exposed to the
   selected compounds for further analysis. Viability was assessed using
   the PrestoBlue® assay (Thermo Fisher Scientific, Massachusetts, USA),
   according to the manufacturer’s instructions. Cells were incubated for
   1.5 h with 5% PrestoBlue® Cell Viability Reagent diluted in PBS at
   37°C/5% CO2. The fluorescence signal was measured
   spectrophotometrically using an EnSpire microplate reader (Perkin
   Elmer, Waltham, Massachusetts, USA). Data were normalized to untreated
   cells, curves were fitted by the GraphPad Prism version 8.0.1 for
   Windows (GraphPad Software, Boston, Massachusetts USA) using the
   sigmoidal dose–response model.

Immunocytochemistry and fluorescent staining

   Cells were seeded into glass-bottomed 8-well slides (ibidi, Gräfelfing,
   Germany) at 20.000 cells/well density. TIS induction was performed as
   described above. CTR cells were seeded at 20.000 cells/well density.

Immunocytochemistry

   TIS and CTR cells were fixed with 4% PFA for 15 min at room temperature
   and washed twice with PBS. Complete blocking solution (0.5% BSA-PBS,
   0.1% Triton X-100, 5% goat serum, and 1% fish gelatin) was used for one
   hour at room temperature. Samples were incubated overnight at 4°C with
   γ-H2A.X, Bcl-2 and Bcl-XL (Thermo Fisher Scientific, Waltham, MA, USA)
   primary antibodies. After incubation, the cells were washed with PBS,
   and the secondary antibody Alexa Fluor 488 (Thermo Fisher Scientific,
   Waltham, MA, USA) was added in complete blocking solution. Nuclei were
   labeled with DAPI (Dojindo EU, Munich, Germany). Cells were observed
   with a Zeiss LSM-710 confocal microscopy at 40 × magnification.

Fluorescent staining

   Fluorescent staining with Nucleolus Bright Red, Cellular Senescence
   Detection Kit-SPiDER-βGal, MitoBright Red, LysoTracker Red (Dojindo EU,
   Munich, Germany) was performed according to the manufacturer’s
   instructions. Cells were observed with a Zeiss LSM-710 confocal
   microscopy at 40× magnification.

Crystal violet staining

   Cells were seeded into 6-well plates at 100.000 cells/well density.
   After five days of DOX treatment cells were washed with pre-warmed PBS
   and the medium was renewed. The day after the navitoclax treatment
   started, it was given three times a week for two weeks at an IC30
   concentration.

   For crystal violet staining a stock solution of 0.5% (w/w) crystal
   violet (Merck, Darmstadt, Germany) was prepared in 25% methanol, from
   which a 10 × dilution working solution was prepared in 25% methanol.
   Cells were washed twice with PBS. The 6-well plates were placed on ice
   and the cells were fixed with ice-cold 100% methanol for 5–10 min.
   After fixation 1 ml crystal violet working solution was added to the
   cells at room temperature for 5–10 min. Cells were washed several times
   with distilled water and then dried overnight.

Western blotting

   Cells were lysed in a lysis buffer (50 mM HEPES, pH 7.3, 150 mM NaCl,
   10% glicerin, 1% Triton X-100, 1 mM EDTA, 1.5 mM MgCl2) supplemented
   with protease inhibitors. Samples were separated by sodium dodecyl
   sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) then transferred
   to polyvinylidene difluoride (PVDF) membrane (Bio-Rad, Hercules,
   California, USA). The membranes were incubated with CDKN1A (p21), LMNB1
   or Lamin B1 (Cell Signaling Technology, Danvers, MA, USA), Bcl-2
   (Thermo Fisher, Waltham, MA, USA) and Bcl-XL (Proteintech, Rosemont,
   USA) primary antibodies at 4°C overnight and then incubated with
   HRP-conjugated secondary antibodies (List of all antibodies used in
   this study can be found in the Supplementary Material 1). Signals were
   detected using an ECL chemiluminescent detection system (WesternBright
   ECL kit, Advansta, San Jose, USA) and a Chemidoc MP device (Bio-Rad,
   Hercules, California, USA). The relative expression level of each
   protein was determined by densitometric analysis using Image Lab
   software (Bio-Rad, Hercules, California, USA).

Statistical analysis of in vitro experiments

   All experiments were repeated with a minimum of three biological
   replicates. For statistical analysis one-way ANOVA or unpaired t-test
   was performed on GraphPad Prism version 8.0.1 (GraphPad Software,
   Boston, Massachusetts USA). n.s. not significant; ^∗ p ≤ 0.05, ^∗∗ p ≤
   0.01, ^∗∗∗ p ≤ 0.001, ^∗∗∗∗ p ≤ 0.0001.

In vivo experiments

   All animal protocols were approved by the Hungarian Animal Health and
   Animal Welfare Directorate according to the EU’s most recent
   directives. All surgical procedures were performed according to the
   Committee on the Care and Use of Laboratory Animals of the Council on
   Animal Care at the Department of Experimental Pharmacology, National
   Institute of Oncology, Budapest, Hungary (PEI/001/1738–3/2015 and
   PE/EA/1461–7/2020).

   1 mm^3 cubes of Brca1^−/−;p53^−/−FVB mouse mammary tumors (a kind gift
   from Sven Rottenberg, NKI) were transplanted orthotopically into the
   mammary fat pad of wild type female FVB mice (Department of
   Experimental Pharmacology, National Institute of Oncology, Budapest,
   Hungary) under anesthesia (20 mg/kg zolazepam, 12.5 mg/kg xylazine,
   3 mg/kg butorphanol, 20 mg/kg tiletamine) [[79]28]. The tumor size was
   monitored at least 3 times per week by caliper measurements after the
   tumors became palpable. Tumor volume was calculated using the V =
   (length × width^2)/2 formula. When the volume of the tumors reached
   ∼200 mm^3, DOXIL treatment was initiated using the maximum tolerable
   dose (MTD, 6 mg/kg iv respectively). Two days after the DOXIL
   treatment, Navitoclax treatment was initiated, administered three times
   a week for two weeks at a dose of 50 mg/kg. Animals were sacrificed
   when the tumor volume reached ∼2000 mm^3.

   Kaplan–Meier survival curves were compared by Log-rank (Mantel-Cox)
   test.

RNA isolation and transcriptome analysis

   MCF7, T47D, MDA-MB-231 and Hs578T TIS cells were harvested at day 12,
   HFF TIS cells were harvested at day 24. CTR and REPOP cells were
   harvested at ~ 80% confluency. Cells were homogenized in TRIzol™
   Reagent (Thermo Fisher Scientific, Massachusetts, USA). Total RNA was
   extracted from samples using Direct-zol® MiniPrep kit (Zymo Research,
   Irvine, California, USA) following the manufacturer's instructions. To
   prevent DNA contamination, an in-column DNAse I treatment was
   performed.

   The prepared total RNA samples were sent to Xenovea Ltd. (Szeged,
   Hungary) for transcriptome analysis. The RNA concentration was
   determined by using the Qubit RNA HS Assay Kit on the Qubit 3.0
   Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA). Quality CTR
   was assessed by Labchip GX Touch HT instrument on DNA 5K/RNA/CZE Chip
   (Perkin Elmer, Waltham, MA, USA) with RNA Pico Sensitivity Assay
   Reagents (Perkin Elmer, Waltham, MA, USA). NextFlex PolyA beads 2.0 kit
   and NextFlex Rapid Directional RNA-seq Kit 2.0 with UDIs (Perkin Elmer,
   Waltham, MA, USA) were used for mRNA capture and strand-specific
   library preparation. The library quantities were measured by Quant-iT
   1 × dsDNA HS Assay kit (Thermo Fisher, Waltham, MA, USA) with Fluostar
   Omega (BMG Labtech, Ortenberg, Germany). The fragment size distribution
   of the libraries was determined by capillary electrophoresis on Labchip
   GX Touch Nucleic Acid Analyzer on XMark HT chip by using DNA NGS 3K
   Assay kit (Perkin Elmer, Waltham, MA, USA). Pooled libraries were
   sequenced with 50 M 150 bp paired-end reads on NovaSeq 6000 platform
   (Illumina, San Diego, CA, USA).

   The raw reads were preprocessed with Fastq Toolkit (v2.2.5) on
   Basespace Sequence Hub (Illumina), to remove low quality bases (< Q30)
   on 5'and 3'ends, trimming adapters (using sequences
   AGATCGGAAGAGCACACGTCTGAACTCCAGTCA,AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT),
   and filter reads with mean read quality less than Q30. Splice aware
   aligner STAR (v2.7.10) was used to map the filtered, trimmed reads to
   genome assembly hg38. Gene counts were obtained by Subread's (v2.0.3)
   Feature counts function. Post-processing on the count matrix was
   executed using the R language (v4.3.3) [R core team]. The numeric
   matrix of read counts were preprocessed by edgeR (v4.0.16): genes with
   sufficiently large counts were retained after which the library sizes
   were recalculated and normalized. Next limma (v3.58.1) was used to
   transform the recalculated count data to log2-counts per million
   (logCPM) and to estimate the mean–variance trend for linear modeling.
   Throughout the rest of the analysis, the term'gene expression'refers to
   the logCPM values. Differentially expressed genes were calculated by
   the standard limma pipeline. First, a linear model was fitted for each
   gene using weighted least squares. Next, contrasts of interest (i.e.
   TIS vs CTR, REPOP vs TIS or REPOP vs CTR) were estimated. Finally,
   empirical Bayes smoothing of standard errors were applied.
   Differentially expressed genes were selected based on their log2 fold
   change (LFC) and Benjamini–Hochberg false discovery rate adjusted
   p-value (adj.p.val). A gene was designated as significantly
   differentially expressed if |LFC|> = 1 and adj.p.val < 0.05.

   Overall similarities and differences in gene expressions were
   visualized by boxplots, heatmaps and principal component analysis (PCA)
   plots. Differentially expressed genes between two samples were
   visualized by a volcano plot. With more than two comparisons, Venn
   diagrams or upset graphs were used for visualization.

   Gene set enrichment analysis (GSEA) was conducted by the fgsea
   (v1.28.0) package.

   R Markdown (v2.28) code to reproduce results in Figs. [80]4 B-I, L-M,
   O, S3 C-G and S4 A is available at [81]https://github.com/lovircsa/TIS.

Fig. 4.

   [82]Fig. 4
   [83]Open in a new tab

   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

Single-cell RNA sequencing

   MCF7 and T47D CTR, TIS (at day 12) and REPOP cells were collected by
   trypsinization, and 15.000 cells/samples were used. Two biological
   replicates of single-cell suspensions were prepared using Scipio
   Bioscience’s reversible hydrogel technology RevGel-seq™ with the
   instrument-free 3’ scRNA-seq benchtop kit Asteria™, allowing for
   coupling of solid polymer-barcoded beads and cells in a homogenous
   phase. Cell capture was followed by cell lysis, mRNA capture on
   barcoded beads, reverse transcription, PCR amplification and cDNA
   sequencing. The samples were sequenced to yield approx. 35.000 raw
   reads per analyzed cell.​​

   Single-cell RNA sequencing (scRNA-seq) data were processed by aligning
   reads to the Human_Scipio_2022_A genome using the Cytonaut platform
   (v2.1.0). The resulting output files — features.tsv.gz,
   barcodes.tsv.gz, and matrix.mtx.gz — were imported into R for further
   analysis using the Seurat package (v5.1.0). To eliminate potential
   doublets, each sample was analyzed individually. Low-quality or dying
   cells often exhibit extensive mitochondrial contamination. Therefore,
   the PercentageFeatureSet() function was used to calculate mitochondrial
   indicators, and abnormal cells were subsequently removed based on a
   specified cutoff, which was determined as follows. Cells expressing
   fewer than 500 genes or fewer than 800 unique molecular identifiers
   (UMIs) were removed, along with those with over 10% mitochondrial
   content. The remaining cells were normalized and scaled using Seurat’s
   NormalizeData and ScaleData functions. Variable genes were identified
   using FindVariableFeatures, followed by principal component analysis
   (PCA). The optimal number of principal components (PCs) was determined
   using the ElbowPlot method, with PCs selected for each sample as
   follows: MCF7_CTR_1: PCs 1–15, MCF7_CTR_2: PCs 1–15, MCF7_TIS_1: PCs
   1–17, MCF7_TIS_2: PCs 1–19, MCF7_REPOP_1: PCs 1–17, MCF7_REPOP_2: PCs
   1–15, T47D_CTR_1: PCs 1–16, T47D_CTR_2: PCs 1–17, T47D_TIS_1: PCs 1–12,
   T47D_TIS_2: PCs 1–15, T47D_REPOP_1: PCs 1–17, T47D_REPOP_2: PCs 1–17.
   Cell clustering was conducted using the FindClusters function with a
   resolution of 0.8, and dimensionality reduction was performed using the
   RunUMAP function to generate UMAP embeddings. To identify and remove
   doublets, the DoubletFinder package (v2.0.4) was used. For each sample,
   pK (a tunable parameter representing the neighborhood size used for
   k-nearest neighbor clustering in DoubletFinder) selection was performed
   using the paramSweep function, and the optimal pK value was determined
   using the find.pK function, selecting the value corresponding to the
   maximum BCmetric score. The expected doublet rate was set to 7.5%, with
   an adjusted expected doublet count calculated using the modelHomotypic
   function to correct for homotypic doublet formation. The doublet
   classification step was performed using the doubletFinder function,
   with the following parameters: pN = 0.25, pK = sample-specific value,
   and nExp = adjusted expected doublet count. Following classification,
   cells labeled as "Doublet" were removed from each dataset.

MCF7 and T47D sample processing for UMAP and downstream analysis

   Following the removal of doublets, cells from MCF7 and T47D samples
   were processed separately, with all cells from each sample type merged
   into a single Seurat object. Further filtering was applied to remove
   cells with more than 10% mitochondrial RNA, more than 30,000 UMIs,
   fewer than 500 unique genes, or fewer than 1,000 UMIs. Standard Seurat
   processing was performed, starting with NormalizeData (using the
   LogNormalize method and a scale factor of 10,000). Variable genes were
   identified with FindVariableFeatures. The data was then scaled using
   ScaleData, and principal components were computed via RunPCA. Batch
   effect correction was not applied in this analysis. The samples
   exhibited high consistency, and applying batch correction could have
   undesirably diminished the biological information. Clustering analysis
   was conducted using FindNeighbors (first 15 principal components for
   MCF7 and first 20 principal components for T47D) and FindClusters, with
   clustering resolutions set to 0.3 for MCF7 and 0.5 for T47D samples.
   Finally, RunUMAP was used to calculate UMAP embeddings based on the
   first 15 principal components for MCF7 and the first 20 principal
   components for T47D. To assess cell cycle stages, CellCycleScoring in
   Seurat was used to calculate G2/M and S phase scores. Differential gene
   expression was analyzed using the Seurat function FindMarkers. For each
   comparison, ident.1 was set to the sample of interest, ident.2 was set
   to the background or control sample, and the Wilcoxon rank sum test
   ("wilcox") was used as the differential test. To account for multiple
   testing, adjusted p-values were calculated using the Bonferroni
   correction based on all genes in the dataset. A threshold of 0.05 was
   applied to the adjusted p-values (padj) to determine statistical
   significance. The gene set enrichment analysis (GSEA) was performed
   using clusterProfiler package (v4.6.2). For the combined analysis of
   MCF7 and T47D samples, cells from both datasets were merged into a
   single Seurat object. The same Seurat processing pipeline was applied,
   utilizing the first 15 principal components for both the FindNeighbors
   and RunUMAP steps. Clustering was performed with a resolution parameter
   set to 0.5.

Trajectory analysis

   Monocle3 (v1.3.1) begins by projecting cells into a low-dimensional
   space that represents their transcriptional states, using UMAP. Next,
   it clusters similar cells through the Louvain community detection
   algorithm, which identifies distinct groups. These groups are then
   merged into larger "supergroups". Finally, Monocle 3 maps the
   developmental trajectories of individual cells within these
   supergroups, pinpointing key locations where cells branch or converge.
   Two distinct trajectories were generated for both MCF7 and T47D cell
   lines: one from the control population to TIS (therapy-induced
   senescence), and the other from TIS to repopulation. Genes that changed
   along these trajectories were identified, and the top 200 genes from
   each trajectory were selected. The top 200 genes from MCF7 control to
   TIS were then compared with the top 200 from T47D control to TIS. A
   similar comparison was performed for the TIS to repopulation
   trajectory. Next, differential gene expression was analysed for both
   MCF7 control vs. TIS and T47D control vs. TIS. The gene lists from both
   comparisons were overlapped, and this overlap was further compared with
   the overlapping trajectory genes. The final gene list included 92
   genes, with CDKN1A added manually. Pseudobulk counts were generated for
   MCF7 and T47D, and TMM-normalized FPKM counts were produced. These were
   then cross-referenced with the earlier gene lists. After Z-score
   normalization, the gene expression data was visualized using a heatmap.

Surface and secreted proteins/peptides characterization

   To identify the targetable proteome on the surface of the CTR, TIS (at
   day 12) and REPOP MCF7 cells, we used a high-throughput surface
   biotinylation method similarly as described in the previous works
   [[84]29, [85]30]. Parental and repopulated cells were cultured until
   they reached ~ 80% confluency before labeling, while TIS cells were
   processed on day 7 after doxorubicin removal. First, the culture medium
   was discarded, and the stage-specific cells were washed with pre-warmed
   PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na[2]HPO[4] and 1.8 mM KH[2]PO[4];
   pH 7.4) three times. The surface proteins of the cells were labeled by
   2 mM Sulfo-NHS-SS-biotin at room temperature in PBS (at pH 8.0) for 20
   min. We have systematically optimized these labeling conditions (PBS
   buffer, pH = 8.0, room temperature, incubation time) to maximize the
   biotinylation of cell surface proteins while maintaining cell
   viability. Over 99% of cells remained intact, as confirmed by trypan
   blue staining and propidium iodide uptake assays, consistent with our
   previous studies [[86]29, [87]30]. Furthermore, the labeled
   transmembrane protein segments were verified to be topologically
   correct — i.e., labeling occurred exclusively on extracellular
   domains — based on data from the UniTmp database [[88]31]. This
   confirms that the membrane-impermeable agent did not cross the cell
   membrane, a point also reinforced by a recent study [[89]32]. The
   biotinylation process was stopped by Tris buffered saline (TBS: 25 mM
   Tris base, 150 mM NaCl, pH 7.2), the solutions were discarded and the
   cells were washed again three times with TBS. The cells were scraped
   into an ice-cold hypotonic lysis buffer (20 mM Tris–HCl, 10 mM KCl, 20
   mM sucrose, 10 mM iodoacetamide (IA), pH 7.4). The cells were lysed on
   ice manually using a plastic micro pestle and a 1 mL syringe with
   26-gauge ½ inch needle. A micro pestle was applied at least 40 times
   with a 180° rotation in both directions within an Eppendorf tube.
   Subsequently, cell lysis was facilitated using a ½ inch, 26-gauge
   needle with a 1 ml syringe, where the solution was aspirated up and
   down 40 times to ensure thorough disruption. The remained intact cells
   and cell debris and nuclei were pelleted at 1700 × g for 5 min at 4°C,
   and the supernatant was transferred into a 10.4 mL polycarbonate tube
   and centrifuged at 100,000 × g for 1 h at 4°C using a 70.1 Ti fixed
   rotor (Beckman Coulter). Biotinylated proteins were enriched in the
   pellet fractions, and resuspended in the 10-times diluted lysis buffer
   and homogenized by 25 strokes with a Potter–Elvehjem PTFE pestle in a
   glass tube on ice, finally stored at − 20°C. The protein concentration
   of the membrane preparations was measured by the Lowry method.

   Membrane preparations with same protein content were solubilized in the
   presence of 0.1% (w/v) Rapigest SF Surfactant and the solutions were
   supplemented with 1.25 mM iodoacetamide and 1.25 mM 2,2′-Thiodiethanol.
   Denatured proteins were digested overnight (~ 16 h) at 37°C with
   proteomics grade trypsin, in a 1:50 (w/w) enzyme-to-protein ratio.
   Digestion was stopped by heat inactivation at 95°C for 10 min,
   thereafter the biotinylated surface peptides were pulled down
   neutravidin agarose resin for 1 h at room temperature. Non-specific
   peptides were removed by several washing steps. The biotinylated
   peptides were eluted by 10 mM Dithiothreitol in 50 mM NH[4]HCO[3]buffer
   using two consecutive incubations of 30 min, each at 37°C. The
   fractions were combined and alkylated with 25 mM iodoacetamide in dark
   at 37°C for 45 min. The solutions were dried in a pre-heated vacuum
   concentrator, then the peptides desalted with a reversed-phase C18 spin
   column as described in a previous work [[90]30]. The samples were dried
   again and stored at − 20°C until further use.

Mass spectrometry

   All measurements were carried out on a Waters ACQUITY UPLC M-Class LC
   system (Waters, Milford, MA, United States) coupled with an Orbitrap
   Exploris 240 mass spectrometer (Thermo Fisher Scientific, Waltham, MA,
   United States). Symmetry C18 (100 Å, 5 µm, 180 µm × 20 mm) trap column
   was used for trapping and desalting the samples. Chromatographic
   separation of peptides was accomplished on an ACQUITY UPLC M-Class
   Peptide BEH C18 analytical column (130 Å, 1.7 µm, 75 µm × 250 mm) at
   45°C by gradient elution. Water (solvent A) and acetonitrile (solvent
   B), both containing 0.1% formic acid were used as mobile phases at a
   flow rate of 200 nL/min. The sample temperature was maintained at 5°C.
   The mass spectrometer was operated using the equipped Nanospray Flex
   Ion Source. Data were collected using the data-dependent acquisition
   (DDA) method with MS1 scan between 360 and 2200 Th using 60,000
   resolution, while ddMS2 scans with isolation windows of 2 Th were
   collected at 30,000 resolution keeping a 3 s cycle time. Data
   acquisition was performed using XcaliburTM 4.6 (Thermo Fisher
   Scientific, Waltham, MA, United States). Raw LC–MS data files were
   processed using Fragpipe v22.0 [[91]33]. Uniprot Human reference
   proteome assuming 2 missed cleavage sites was used (20,575 proteins)
   for protein identification. An initial open search (using default
   settings) was performed to identify detectable peptide modifications.
   The final quantitative analysis was performed assuming Met oxidation,
   3-(carbamidomethylthio)propanoyl Lys (effect of biotin labelling) and
   pyro Glu as variables and carbamidomethyl Cys as fixed modification.
   Protein quantification was performed within Fragpipe using IonQuant
   with default settings, and enabling match between runs. A contaminant
   database was created using a database search of LC–MS data collected
   from a labeled and enriched digest of cell medium using the same search
   settings and Uniprot Bovine reference proteome database. Contaminant
   proteins and proteins with only one identified peptide were removed
   from the dataset before statistical analysis. Statistical analysis and
   visualization of proteomics data was performed in Perseus 1.6.15
   [[92]34] and Instantclue v0.12.2 [[93]35]. Protein and peptide
   intensity data were log2 transformed and median normalized before
   statistical analysis. Differential expression analysis (DEA) was
   performed in Perseus using the built in Student’s t-test with
   permutation-based false discovery rate (FDR) estimation. A limit of FDR
   < 0.05 was applied using the Significance Analysis of Microarrays (SAM)
   method with a s0 = 0.1 value (approximate weight of mean difference in
   FDR calculation) using 250 randomizations [[94]36]. In order to confirm
   protein level changes, DEA was also performed on individual peptides
   using the same approach. To confirm cell surface/extracellular presence
   of proteins the number of biotin labelled peptides were calculated for
   each identified protein. Proteomics raw data, protein and peptide level
   DEA results are available online (see ‘[95]Data and code availability’
   section).

qRT-PCR validation

   Total RNA was extracted using the Qiagen RNeasy Mini Kit, and 1 µg of
   total RNA was reverse transcribed with the High-Capacity cDNA Reverse
   Transcription Kit (Thermo Fisher Scientific) following the
   manufacturer’s protocol. Quantitative SYBR Green-based qRT-PCR assays
   were performed in quadruplicates using the PCRBIO HS Taq Mix Red on a
   QuantStudio 7 Flex Real-Time PCR System (Applied Biosystems). Gene
   expression data were analyzed using the ΔΔCt method. ACTB was used as
   the housekeeping gene for internal control and normalization. Primer
   sequences used for relative quantification can be found in
   Supplementary Table S1.

Results

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
   [[96]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. [97]2B) via apoptosis (Figure S2), leaving only a
   small fraction of the population alive (Fig. [98]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. [99]1B). We also observed the
   overexpression of the cyclin-dependent kinase inhibitor protein CDKN1A
   (p21) [[100]38] and the downregulation of the nuclear lamina building
   block LMNB1 [[101]39], both widely used markers for the identifying
   senescent cells (Fig. [102]1C). Additionally, the cells exhibited
   extensive DNA damage, increased mitochondrial and lysosomal mass, and
   developed a single, fused nucleolus (Fig. [103]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. 2.

   [104]Fig. 2
   [105]Open in a new tab

   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

Fig. 1.

   [106]Fig. 1
   [107]Open in a new tab

   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. [108]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. [109]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. [110]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.

   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. [111]2D). The cells were then exposed to
   15 selected compounds to assess drug resistance (Fig. [112]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. [113]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. [114]3A), long-term treatment
   produced ambiguous results (Fig. [115]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. [116]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. [117]3D). Despite
   this, navitoclax failed to extend survival or demonstrate any
   disease-modifying effect in this model (Fig. [118]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. [119]3F, Table S3). The selective Bcl-2 inhibitor venetoclax
   (from Fig. [120]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.

   [121]Fig. 3
   [122]Open in a new tab

   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. [123]4A). Normalized gene expression data revealed
   distinct differences between TIS samples and CTR and REPOP cells
   (Fig. [124]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. [125]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. [126]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. [127]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. [128]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. [129]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.

   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. [130]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. [131]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. [132]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 [[133]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 [[134]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 [[135]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. [136]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 [[137]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 [[138]44], but no such link was found. Increased expression
   of RRM1, 2 and 2B is often detected in DOX, gemcitabine and docetaxel
   resistant tumors [[139]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 [[140]46–[141]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 [[142]49, [143]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 [[144]51]. ABCC1 and C3 were expressed cell line
   specifically, but ABCB1 – in accordance with our previous finding
   [[145]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. [146]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. [147]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. [148]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.

   [149]Fig. 5
   [150]Open in a new tab

   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. [151]5D and E, top left panels). Along with the mostly
   undetectable expression of the nuclear lamina component LMNB1 gene
   (Fig. [152]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. [153]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. [154]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. [155]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. [156]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. [157]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. [158]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. [159]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. [160]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. [161]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 resistance in T47D TIS cells, but
   MCF7 TIS cells are also insensitive to the drug while significantly
   overexpressing NEDD8 compared to CTR and REPOP cells. These findings,
   along with the bulk RNA-seq results, suggest that TIS cells exhibit an
   inherent drug resistant phenotype that is not directly linked to single
   genes or mechanisms.

Targeting the top overexpressed genes in TIS cells revealed no exploitable
sensitivity

   Identifying compounds or strategies capable of selectively eliminating
   TIS cells represents a transformative step toward developing novel
   therapeutic options for cancer treatment. To achieve this, we first
   pinpointed the most overexpressed genes in TIS cells across all four
   cell lines using our bulk RNA-seq data. We then selected the top five
   genes for which commercially available inhibitors were accessible
   (Fig. [162]6A, Table S5). Previous studies have shown that statins
   effectively reduce Keratin 6A (KRT6A) gene promoter activity, leading
   to their use as a first-line treatment for pachyonychia congenita, a
   rare skin disorder linked to KRT6A [[163]52]. Based on this, we treated
   CTR and senescent cells with simvastatin; however, no difference in
   sensitivity was observed (Fig. [164]6B). The proteasome 20S subunit
   alpha 8 (PSMA8), a proteasomal protein sensitive to MG132, a potent
   proteasome inhibitor, was also tested to determine if MG132 selectively
   targets senescent cells. However, the response was consistent in both
   CTR and senescent cells, similar to results with other
   proteasome-targeting drugs previously tested [[165]53]. Additionally,
   although thymoquinone (TQ) has been reported to repress
   phosphodiesterase 1A (PDE1A) expression [[166]54], our assay
   demonstrated that TQ induced comparable cytotoxicity in both CTR and
   senescent cells. We also investigated gasdermin C (GSDMC), which
   induces pyroptosis upon cleavage by caspase-8, as it was highly
   expressed in TIS cells. We tested the effects of dimethyl
   2-oxoglutarate (DM-αKG) and actinomycin D, compounds known to elevate
   caspase-8 levels and trigger GSDMC cleavage [[167]55, [168]56].
   Regrettably, DM-αKG was non-toxic, while TIS cells exhibited resistance
   to actinomycin D. Lastly, we inhibited acyl-CoA synthetase medium chain
   family member 2A (ACSM2A) using 4-Methylsalicylic acid [[169]57], but
   it showed no toxicity in either cell type.

Fig. 6.

   [170]Fig. 6
   [171]Open in a new tab

   TIS cells exhibit a unique proteomic profile. A Expression levels of
   the top five proteins, for which inhibitors are available, in control
   (CTR, red), TIS (green), and repopulated (REPOP, blue) cells. B
   Sensitivity of CTR (black) and TIS (red) cells to six different
   inhibitors targeting the proteins identified in panel (A). C
   Hierarchical clustering analysis of Z-score normalized proteomic
   profiles from MCF7 CTR, TIS, and REPOP samples, demonstrating distinct
   clustering of each state. Proteins with significantly different
   expression levels (FDR < 0.05) in different comparisons are marked on
   the right. D, E Volcano plots illustrating protein expression changes
   during the CTR-to-TIS transition (D) and the TIS-to-REPOP transition
   (E). Significance limits (FDR < 0.05) are shown by dotted blue lines,
   significantly increased or decreased proteins are colored red or blue,
   respectively. Number of proteins with significant change are shown at
   the top corners. F Volcano plot showing protein expression differences
   between CTR and REPOP states. Significance limits (FDR < 0.05) are
   shown by dotted blue lines, significantly increased or decreased
   proteins are colored red or blue, respectively. Number of proteins with
   significant change are shown at the top corners. G, H Signaling pathway
   enrichment analysis of differentially expressed proteins, highlighting
   altered pathways in the TIS vs. CTR (G) and TIS vs. REPOP (H)
   comparisons. I Heatmap of proteins uniquely overexpressed in TIS cells,
   underscoring their distinct proteomic signature

   Due to we based our treatment approach on mRNA expression, the protein
   levels of each target were measured (Figure S14). While no consistent
   upregulation of these was observed across all cell lines, certain cases
   exhibited increased expression upon TIS induction. Specifically, ACSM2A
   and GSDMC were overexpressed in MDA-MB-231, PSMA8 in MCF7, and PDE1A in
   both MCF7 and T47D cells. Despite these expression changes, none led to
   collateral sensitivity in TIS cells, reinforcing the hypothesis that
   TIS itself confers a protective advantage to cancer cells, independent
   of drug target protein overexpression.

   Taken together, these findings reveal that despite targeting a spectrum
   of highly expressed senescence-associated genes with carefully selected
   inhibitors, none of the tested compounds demonstrated the selective
   cytotoxicity required to effectively eliminate senescent cells –
   highlighting the significant challenge of overcoming therapy-induced
   senescence in cancer treatment.

Proteomics analysis of TIS cells highlights SASP and a plausible escape
mechanism

   Recognizing the limited potential of traditional therapeutic
   approaches, we advanced toward a novel strategy centered on immune and
   cell surface/extracellular targets. Using a specific biotin labeling
   technique to pull down membrane and extracellularly secreted proteins –
   developed by Langó et al. [[172]29, [173]30] (Figure S15) – on MCF7
   CTR, TIS, and REPOP cells, we aimed to identify TIS-specific proteome
   expression variations, uncovering new targets for selective inhibitors
   or immunotherapy. Clustering of 6, 8 and 8 biological replicates of
   CTR, TIS and REPOP samples, respectively, revealed that, while CTR and
   REPOP cells exhibit nearly identical proteomes, TIS cells display a
   distinct protein profile (Fig. [174]6C). Specifically, 420 proteins
   were differentially expressed by TIS cells compared to CTR, and 370
   compared to REPOP samples, whereas only 92 proteins varied between CTR
   and REPOP cells (Fig. [175]6D, E and F). From these distinct protein
   sets, two pathways – spliceosome and RAS signaling – emerged as
   recurrent features, aligning with findings from our transcriptome
   analysis described above (Fig. [176]6G,H). Although the role of
   spliceosome in cellular senescence [[177]58–[178]60] and SASP [[179]61]
   is a rapidly developing area of research, the large-scale presence of
   splicing proteins on the cell surface and/or secreted form is
   unprecedented. Interestingly, 85 out of the 95 differently expressed
   proteins were earlier detected in extracellular vesicles secreted by
   senescent tumor cells [[180]62]. Furthermore, RAS overexpression has
   previously been associated with senescence, primarily as an oncogenic
   signal triggering [[181]63], rather than as part of TIS dynamics. KRAS
   protein was overexpressed in MCF7 TIS cells (Fig. [182]6I), despite its
   typically marginal expression in breast cancer cells and tumors
   [[183]64]. This finding aligns with our bulk RNA-seq results, which
   revealed substantial changes in both the up- and downregulated gene
   sets associated with KRAS expression (Fig. [184]4K, L). Unexpectedly,
   TIS cells did not exhibit elevated KRAS mRNA expression compared to CTR
   and REPOP cells (Figure S16A), nor did they show increased KRAS protein
   levels (Figure S16B). It was hypothesized that the differential
   regulation of KRAS-related genes could still result from an oncogenic
   mutation induced by high-dose DOX treatment. However, deep sequencing
   of KRAS confirmed that both CTR and TIS cells in both cell lines
   retained the non-mutated wild-type gene (Figure S16 C,D). The lack of
   KRAS involvement in breast cancer cells was further supported by
   treating CTR and TIS cells with RMC-6236, a pan-KRAS inhibitor, which
   did not elicit significant sensitivity in either condition (Figure
   S16E). These findings strongly suggest that KRAS pathway activation in
   TIS cells occurs through a non-canonical mechanism.

   The analysis of proteomics results further indicated extensive
   secretion of SASP-related proteins (Fig. [185]6I). Of the 95 proteins
   overexpressed exclusively in TIS cells, 41 were associated with
   senescence and 17 with alternative splicing and the spliceosome (Table
   S6 A and B). Furthermore, we found that FBL [[186]65], NME2 [[187]66],
   FLNA [[188]67], ENO1 [[189]68, [190]69] and ANXA2 [[191]70, [192]71]
   were described in a recently identified set of SASP proteins, found in
   irradiated, RAS-transformed and atazanavir treated senescent
   fibroblasts, further strengthening the similarity between TIS in cancer
   cells and TIS and OIS in healthy cells [[193]65]. These findings
   underscore the distinct molecular landscape of TIS cells, with KRAS
   overexpression, SASP protein secretion, and unique associations with
   senescence and splicing pathways, highlighting potential new avenues
   for therapeutic targeting.

   We also observed the translocation of high mobility group box 1 (HMGB1)
   protein from the nucleus to the plasma membrane – a recently discovered
   Damage-associated molecular pattern (DAMP) signal originating from
   senescent cells [[194]72–[195]76] as well as the secretion of growth
   differentiation factor 15 (GDF15) a newly recognized marker of
   senescence and aging [[196]65, [197]77]. Validation of the proteomic
   data was performed using individual peptide sequence analysis (Figure
   S17 A,B).

   Interestingly, partially overlapping with the senescence markers, a
   group of proteins related to the spliceosome were also identified in
   our assay (Table S6B). Each of these proteins either plays a role in
   constitutive and/or alternative splicing or forms part of the
   spliceosome, indicating a significant involvement of RNA splicing in
   establishing and maintaining the TIS phenotype.

   The additional 52 proteins had not previously been identified as
   contributors to senescence, suggesting that some may play roles in the
   maintenance or regulation of TIS, or even serve as biomarkers of this
   phenotype. The mRNA expression levels of the 95 protein encoding genes
   were investigated in bulk and scRNA-seq results (Figure S18 A and B).
   Notably, ribosomal proteins were enriched within this group, supporting
   prior findings that link senescence with ribosome biogenesis defects
   [[198]78].

   Consistent with our RNA-seq results, no single resistance mechanism or
   mechanism group was identified that could fully account for the
   heightened drug tolerance observed in TIS cells. In addition to
   transcriptional alterations, therapy-induced senescence may be
   accompanied by widespread chromatin remodeling, which could contribute
   to its persistence and heterogeneity.

Discussion

   Therapy-induced senescence (TIS) has recently emerged as a promising
   strategy in cancer treatment [[199]79, [200]23]. The concept of
   inducing cytostasis in cancer cells, rather than eliminating them, has
   gained attention for its potential to expand therapeutic options across
   malignancies by shifting away from solely relying on high doses of
   toxic drugs and/or radiation — which carry severe side effects and the
   risk of drug resistance — TIS offers an approach to permanently arrest
   cell proliferation in tumors [[201]22]. However, our findings suggest
   that TIS may function less as an ideal therapeutic endpoint and more as
   a robust resistance mechanism, underscoring the need to reconsider its
   implications in treatment strategies.

   To effectively compare studies and findings, a clear definition of TIS
   in cancer cells is essential. Senescence is widely regarded as an
   irreversible growth arrest in both healthy and malignant cells
   [[202]80]. However, recent studies suggest that some subpopulations of
   senescent cancer cells can evade this arrest, regaining proliferative
   capacity through escape or reversion mechanisms [[203]24, [204]81].

   The primary challenges in TIS-studies and defining senescence in cancer
   cells is the absence of specific markers [[205]82, [206]38] and the
   diversity of treatment protocols, encompassing a range of drugs and
   treatment duration to induce senescence.

   Senescence is generally considered irreversible only if the inducing
   stimulus persists for at least 4 days [[207]83, [208]84]; cancer cells
   treated for less than 96 h can readily escape TIS. For example, early
   studies showed that H1299 lung cancer cells could evade TIS after 1–3
   days of camptothecin treatment, with escape observed approximately
   20–25 days post-treatment — similar to our observations. Notably, p16-
   and p53-negative H1299 cells were used, as p53 and p16 are known to
   limit senescence reversibility [[209]24]. In another model, a 2 h DOX
   treatment induced TIS in MCF7 cells, and a small subset escaped
   long-term arrest despite wild-type p53 expression, forming clonal
   outgrowths within 2–3 weeks [[210]85]. More recent studies similarly
   observed TIS escape in various cell lines, but proliferation resumed
   significantly earlier, 5–7 days after treatment removal, with treatment
   durations not exceeding 24 h [[211]27, [212]86]. Saleh et al. conducted
   a series of studies on TIS escape of H460, A549, and HCT116 BTG1-RFP
   cells, using various drugs. In most cases, cells regained proliferative
   capacity approximately 7 days after treatment [[213]86, [214]27].

   Tóth et al. investigated TIS induction in MCF7 and MDA-MB-231 cell
   lines using bromodeoxyuridine, gemcitabine, and palbociclib.
   Remarkably, despite drug treatments lasting up to 1–2 weeks,
   TIS-escaped cells began appearing within 7 days upon finishing the
   treatment [[215]87]. The early escape from TIS raises the question
   about whether fully developed TIS was achieved or if cells entered a
   transient growth arrest [[216]88] or a reversible phase of senescence
   due to short-term or lower-dose treatments. A study published in 2023
   systematically examined the duration dependency of TIS induction,
   demonstrating that MCF7 cells commit irreversibly to TIS only after
   continuous MEK and CDK4/6 inhibition for over 4 days [[217]89]. Our
   experimental design ensured the senescent phenotype was established by
   applying high-dose DOX for 120 h, with surviving cells analyzed one
   week post treatment. The observed repopulation time (> 30 days) closely
   mirrored relapse times in a clinically relevant mouse model of triple
   negative breast cancer following complete response to DOX [[218]90,
   [219]91].

   Using multiple markers and features to confirm the presence of the
   senescent phenotype is not only recommended but essential due to the
   complexity of senescence. In our system, both the initially selected
   markers and additional findings supported our hypothesis that the
   surviving cells were in TIS. For example, the TIS cell surfaceome
   revealed extensive secretion of different factors, including numerous
   SASP proteins, HMGB1 translocation, and ribosomal elements. Alongside
   with the established senescence markers like CDKN1A and LMNB1, our
   transcriptome analysis identified a distinctive expression pattern,
   comprising genes upregulated across all cell lines exclusively during
   TIS, which may serve as a specific TIS signature in breast cancer. This
   signature includes genes known for their roles in cellular senescence
   regulation such as matrix metallopeptidase 12 (MMP12)
   [[220]92–[221]94], cytochrome P450 family 1 subfamily A member 1
   (CYP1A1) [[222]95, [223]96], baculoviral IAP repeat containing 3
   (BIRC3) [[224]97–[225]99], collagen type IV alpha 3 chain (COL4A3)
   [[226]100], and neurotrophic receptor tyrosine kinase 2 (NTRK2)
   [[227]101], as well as genes associated with aging, like C–C motif
   chemokine ligand 26 (CCL26) [[228]102], glutamate ionotropic receptor
   delta type subunit 2 (GRID2) [[229]103], carbonic anhydrase 10 (CA10)
   [[230]104, [231]105], and C-type lectin domain family 12 member A
   (CLEC12A) [[232]106]. Interestingly, many genes in this set are not
   previously linked to senescence, offering a novel pool of potential TIS
   markers. Additionally, genes with functions that protect against
   senescence, such as reelin (RELN) [[233]107–[234]109], were also
   present, potentially indicating a molecular toolkit supporting cell
   survival and repopulation.

   Our surfaceome analysis revealed a substantial secretion of
   spliceosome-associated proteins. Although spliceosome involvement in
   cellular senescence [[235]58–[236]60] and the senescence-associated
   secretory phenotype (SASP) [[237]61] is an emerging field, no direct
   evidence currently links spliceosome components to intercellular
   communication during cellular senescence. However, recent studies
   suggest a potential role for spliceosome components in mediating
   intercellular communication, particularly under stress conditions in
   cancer. Under cellular stress induced by chemotherapeutic agents,
   γ-irradiation, or hypoxia, spliceosome elements-including heterogeneous
   nuclear ribonucleoproteins (hnRNPs), members of the family of
   serine/arginine (SR)-rich proteins, and small nuclear RNAs (snRNAs) are
   known to translocate from the nucleus to the cytoplasm and aggregate in
   stress granules. This redistribution, notably of overexpressed factors
   like serine/arginine-rich splicing factors 1 and 3 (SRSF1, SRSF3), has
   been observed under various stressors. Furthermore, spliceosome
   components can be secreted in extracellular vesicles from stressed
   tumor cells. When taken up by neighboring tumor cells, these splicing
   factors may play a pivotal role in modulating molecular processes
   within the recipient cells, potentially enhancing their resilience to
   chemotherapy [[238]110]. Our findings suggest that the secretion and
   subsequent internalization of spliceosome components may confer
   adaptive advantages under stress conditions within the context of
   cellular senescence.

   When cells enter senescence, they inherently become more resistant to
   apoptosis [[239]111], which has led to the use of antiapoptotic
   inhibitors as senolytics. However, our findings suggest that targeting
   these pathways in TIS cancer cells may be unproductive. Given that TIS
   cells are non-proliferative and supposedly overexpress antiapoptotic
   proteins from the Bcl-2 family, one would expect them to show higher
   tolerance to most anticancer agents – yet, this is not the case.
   Apoptosis resistance in senescent cells was first observed in healthy
   fibroblasts under serum deprivation [[240]112], which prompted the
   application of Bcl-2 inhibitors as senolytics against non-malignant
   senescent cells [[241]113]. Although some reports suggest that
   antiapoptotic inhibitors might target senescent cancer cells, our
   experiments with venetoclax, navitoclax, ABT-737 and A1331852 yielded
   mixed results. Venetoclax showed no selectivity for TIS cells, while
   the other three compounds displayed some TIS-specific cytotoxicity;
   however, the outcomes in long-term assays were inconsistent
   (Fig. [242]3). Strikingly, these results appeared unrelated to Bcl-2 or
   Bcl-XL expression, as both proteins were already overexpressed in the
   parental cell lines (Figure S2). This raises the question of whether
   antiapoptotic inhibitors truly act as senolytics by targeting Bcl-2/XL
   in TIS cancer cells or through an alternative, yet unidentified
   mechanism.

   Although an apoptosis-resistant phenotype would theoretically render
   TIS cells resistant to a broad spectrum of drugs [[243]114], our
   findings did not support this expectation. For example, while HDAC
   inhibition has been proposed as a senolytic approach in various cancer
   types [[244]115], it was equally toxic to both CTR and TIS cells in our
   screening (Fig. [245]2) and is known to induce cell death via apoptosis
   [[246]116]. Similarly, apoptosis-inducing proteasome inhibitors
   [[247]117] also proved toxic to both CTR and TIS cells in our
   experiments, challenging the notion that TIS cells are resistant to
   apoptosis. Tyrosine kinase inhibitors, which also kill cancer cells by
   inducing apoptosis [[248]118], showed no selective resistance in TIS
   cells either. These findings strongly argue against apoptosis
   resistance in TIS cancer cells.

   The hallmark of senescent cells is their complete lack of proliferation
   [[249]38]. It is widely accepted that slowly or non-proliferating cells
   tolerate significantly higher concentrations of chemotherapeutics
   compared to rapidly dividing populations [[250]119]; thus, the drug
   resistance profile of TIS cells could theoretically be attributed to
   their long-term exit from the cell cycle. However, several findings
   challenge this interpretation. Drugs that typically rely on cell
   proliferation, such as AT7519 [[251]120], ibrutinib [[252]121],
   masitinib, sorafenib and sunitinib [[253]122] should theoretically have
   no effect on non-dividing cells, yet they effectively killed TIS cells.
   Conversely, compounds without cell cycle specificity or with documented
   efficacy against quiescent cells – such as carfilzomib [[254]123,
   [255]124], homoharringtonine [[256]125] and melphalan [[257]126],
   failed to harm TIS cells. This unexpected combination of effects
   underscores the unique chemoprotection provided by TIS and draws a
   parallel with drug-tolerant persister cells, as both arise following
   high drug exposure without involving known resistance mechanisms
   [[258]127].

   One of the most intriguing findings in our study is the recurring
   pattern of immune-modulatory gene and protein expression. While some
   reports suggest that senescent cells are immunogenic and cleared by the
   immune system [[259]128, [260]129], others indicate that these cells
   evade the immune response [[261]130, [262]131]. Our results demonstrate
   that TIS cells can significantly influence immune cell behavior through
   secreted factors and altered gene expression. Notably, we observed
   suppression of the Cytokine-Cytokine Receptor Interaction pathway,
   which is essential for immune cell communication and response. This
   downregulation may enable TIS cells to evade immune surveillance,
   reducing their chances of immune clearance. Combined with the
   activation of Neuroactive Ligand-Receptor Interaction pathways, these
   changes suggest that TIS cells undergo substantial alterations in
   environmental interactions, potentially creating a supportive niche for
   their long-term survival – even in a controlled culture setting where
   tumor microenvironmental factors are absent, but would otherwise be
   crucial. The activation of the Cytosolic DNA-Sensing Pathway, along
   with the Helicobacter pylori and Staphylococcus aureus Infection
   pathways, underscores an altered immune response in TIS cells. These
   pathways may represent adaptive mechanisms through which TIS cells
   manage chronic inflammatory signals or potential bacterial challenges,
   aiding their survival in a hostile microenvironment. Certain genes
   within these pathways are known to suppress or attenuate T cell
   responses. For instance, TIS cells in all tested cell lines
   overexpressed galectin 9 (LGALS9), a secreted galectin, recently shown
   to induce Gal-9/TIM-3 mediated apoptosis in T cells, thereby limiting
   their expansion and essentially acting as an immune inhibitor
   [[263]132, [264]133]. C-X-C motif chemokine ligand 12 (CXCL12) is a
   cytokine which, if secreted in high concentration by senescent cells in
   the tumor microenvironment (TME), can create a “cytokine shield” that
   inhibits CD8+ T cell infiltration by downregulating C-X-C chemokine
   receptor type 4 (CXCR4) and impairing directional migration [[265]134].
   Triggering receptor expressed on myeloid cells 1 (TREM1) was found to
   suppress the antitumor immune response, and its inhibition
   significantly delays melanoma growth. TREM1 was also universally
   overexpressed in TIS cells across all four cell lines [[266]135].
   Metastasis-associated in colon cancer 1 (MACC1), an established
   metastasis regulator and marker, facilitates immune evasion in
   malignancies through various processes, primarily by regulating the
   expression of immune cell-modifying surface molecules [[267]136]. The
   presence of these factors alone suggests a profound alteration of the
   local immune landscape and response, but TIS cells overexpress not only
   these, but many additional modulators, underscoring the complexity of
   targeting these cells through immune mechanisms.

   Modulated immune reactivity may be an even more important consequence
   of TIS as it seems at first glance. REPOP cells are highly similar to
   the CTR lines in terms of drug tolerance and gene expression patterns,
   as shown by both the bulk and single cell sequencing results. However,
   comparing parental and repopulating cells and identifying the remaining
   gene expression changes may reveal mechanisms which enable repopulating
   cells to escape TIS with a higher efficiency upon repeated treatment.
   By comparing the transcriptomic data of CTR and REPOP cells, we found a
   signature of 23 genes, that were overexpressed in the malignant cells
   during TIS and remained highly expressed in the REPOP cells. This gene
   set contained several members of the IFIT (Interferon Induced proteins
   with Tetratricopeptide repeats) and the OAS (2-5A synthetase) family,
   which are primarily involved in the defense against viral infections.
   However, recent results indicate both protein family’s involvement in
   other pathologies, like cancer [[268]137]. Several genes among the
   identified signature have already been connected to cancers or cellular
   senescence. Importantly, specific representatives of the residual
   signature genes (ISG15, IFI27, IFI6, RSAD2) appear to aid cancer cells
   in evading immunosurveillance and inflammatory cells by inhibiting
   apoptotic processes [[269]138]. IFIT1, which was significantly
   overexpressed in all four malignant TIS cell types, has been shown to
   be significantly upregulated in residual breast cancer tumors
   [[270]139]. It was also shown to be part of an IFN-related DNA damage
   resistance signature in breast cancer, together with ISG15, another
   gene in the signature [[271]140]. IFIT3 was part of a SASP-related
   prognostic signature to assess therapeutic effect and prognosis in AML
   [[272]141], indicating its role in the regulation of cellular
   senescence. Its high expression level was connected to increased
   resistance of different therapeutics in oral squamous cell carcinoma
   [[273]142] and was identified as a hub gene in triple negative breast
   cancer [[274]143]. IFITM1 was found to be overexpressed in SCLC tumors
   and was significantly increased in cisplatin-resistant SCLC tissues
   [[275]144]. IFITM1 inhibition was also suggested to enhance the
   sensitivity to tamoxifen in ER-positive breast cancer cells [[276]145].
   OAS1 overexpression influenced survival and immune cell infiltration in
   patients with lung adenocarcinoma [[277]146] and OAS2 was shown to
   participate in a sustained senescence response of human primary
   melanocytes upon repeated UVB exposure [[278]147]. Genes implicated in
   the modulation of tumor microenvironment and immunomodulation include
   CLIC5 [[279]148], RSAD2 [[280]149] and OASL [[281]150]. Collectively,
   this shows that TIS initiates long-term changes in malignant cells that
   remain even after returning to the proliferating state. These residual
   gene expression changes may facilitate immune evasion and
   anti-apoptotic processes, rendering the repopulating cells more
   resistant to future anti-cancer treatments.

   Afifi et al. recently proposed that MCF7 cells could escape TIS through
   sustained overexpression of the MYC oncogene, demonstrating that MYC
   degradation makes senescence irreversible [[282]89]. Although we did
   not observe MYC upregulation in TIS cells across any of the cell lines
   (data not shown), the hypothesis that a proto-oncogenic signal or
   pathway could compromise the irreversibility of senescence and enable
   certain cells to evade TIS warranted further investigation. As
   non-canonical KRAS pathway activation was observed in TIS cells
   throughout our study, from bulk RNA-seq data to surfaceome analysis, it
   could be an escape mechanism similar to MYC overexpression. Seemingly,
   KRAS protein itself is not required in TIS cells, but its pathway
   remains essential. The KRAS signaling pathway is well known for
   promoting cancer cell survival and immune evasion in tumors, and some
   studies suggest that it may also support repopulation under certain
   conditions. For example, KRAS overexpressing hematopoietic stem cells
   are less prone to irradiation-mediated damage and repopulate quicker
   after treatment, leading to accelerated hematologic recovery in a mouse
   model due to increased Erk1/2 phosphorylation and cyclin-dependent
   kinase 4 and 6 (Cdk4/6) activation [[283]151]. The amplification of
   KRAS also renders lung cancer cells resistant to crizotinib, a MET
   inhibitor, by hypersensitizing them to ligand stimulation [[284]152].
   Since we also observed similar changes in TIS cells, we hypothesize
   that non-canonically activated KRAS pathway could act as a “doorstop”
   for senescence, inhibiting full commitment to irreversible TIS.

Conclusion

   In summary, our study redefines therapy-induced senescence (TIS) in
   cancer treatment by uncovering its complex adaptive mechanisms and
   resistance traits. Although TIS has emerged as a promising strategy to
   halt tumor progression, our findings suggest it functions more as a
   dynamic resistance state than a permanent growth arrest. Transcriptomic
   analyses revealed extensive shifts in gene expression within TIS cells,
   yet these changes do not reveal any definitive resistance mechanism
   based on current knowledge, underscoring the unique, elusive nature of
   TIS resistance.

   The pronounced shift in the transcriptome profile within TIS cells
   suggests the potential involvement of epigenetic modulation or
   reprogramming in TIS formation and maintenance. Additionally,
   disparities between mRNA and protein levels imply that
   post-translational modifications (PTMs) may play a critical role in
   establishing and stabilizing senescence. Our data also show that TIS
   cells suppress immune signaling pathways, such as the Cytokine-Cytokine
   Receptor Interaction pathway, while activating immune-evasive and
   survival pathways, including increased secretion of galectin 9, CXCL12,
   and TREM1. These immune-modulatory factors likely help TIS cells evade
   immune clearance, fostering their long-term persistence.

   These findings establish TIS not only as a novel mechanism of drug
   resistance but also as a formidable immune barrier. Growing evidence
   suggests that chemotherapy’s primary contribution to tumor reduction
   extends beyond direct cytotoxic effects, as it also enhances tumor
   immunogenicity, facilitating immune-mediated clearance. However, our
   results demonstrate that TIS cells actively evade immune elimination
   through a repertoire of secreted and surface-expressed factors,
   effectively shielding themselves from immune attack. Notably, these
   same immune evasion mechanisms may constitute exploitable therapeutic
   vulnerabilities. Targeting these pathways could prolong relapse-free
   survival and improve overall patient outcomes by enabling the immune
   system to eliminate TIS cells post-chemotherapy, thereby increasing the
   likelihood of durable remission and potentially achieving a cure.

   These findings collectively highlight the limitations of TIS as a
   therapeutic endpoint. Rather than serving as a stable state, TIS equips
   cancer cells with adaptive mechanisms that complicate treatment. The
   potential involvement of epigenetic reprogramming and PTMs in TIS opens
   new avenues for investigation, suggesting that interventions targeting
   protein-level modifications or epigenetic alterations may be crucial
   for overcoming TIS-driven resistance. Future studies integrating
   epigenomic profiling, chromatin activity analysis, and transcription
   factor mapping will be essential to fully elucidate the regulatory
   landscape of TIS and its potential vulnerabilities. Understanding how
   TIS transitions to an active resistance mechanism could offer promising
   strategies for developing targeted treatments capable of dismantling
   this evasive phenotype.

Supplementary Information

   [285]Supplementary Material 1^ (9.4MB, pdf)
   [286]Supplementary Material 2^ (1.1MB, pdf)

Acknowledgements