A Transient Pseudosenescent Secretome Promotes Tumor Growth after Antiangiogenic Therapy Withdrawal

SUMMARY

VEGF receptor tyrosine kinase inhibitors (VEGFR TKIs) approved to treat multiple cancer types can promote metastatic disease in certain limited preclinical settings. Here, we show that stopping VEGFR TKI treatment after resistance can lead to rebound tumor growth that is driven by cellular changes resembling senescence-associated secretory phenotypes (SASPs) known to promote cancer progression. A SASP-mimicking antiangiogenic therapy-induced secretome (ATIS) was found to persist during short withdrawal periods, and blockade of known SASP regulators, including mTOR and IL-6, could blunt rebound effects. Critically, senescence hallmarks ultimately reversed after long drug withdrawal periods, suggesting that the transition to a permanent growth-arrested senescent state was incomplete and the hijacking of SASP machinery ultimately transient. These findings may account for the highly diverse and reversible cytokine changes observed in VEGF inhibitor-treated patients, and suggest senescence-targeted therapies (“senotherapeutics”)—particularly those that block SASP regulation—may improve outcomes in patients after VEGFR TKI failure.

In Brief

Mastri et al. show that resistance to antiangiogenic therapy can induce senescence-mimicking secretory phenotypes that contribute to tumor promotion when treatment is stopped. Therapeutic targeting of senescence secretory regulators such as IL-6 and mTOR could improve outcomes after treatment failure.

Graphical Abstract

INTRODUCTION

Angiogenesis inhibitors targeting the vascular endothelial growth factor (VEGF) pathway are currently approved to treat multiple late-stage metastatic cancer types; however, we and others have previously raised the possibility that treatment can accelerate metastasis in certain preclinical settings (Jain, 2014; Ebos et al., 2009a; Pàez-Ribes et al., 2009). This effect may offset or diminish clear benefits of therapy and explain instances where progression-free or overall survival curves collapse or crossover after extended treatment (Oza et al., 2015; Haemmerle et al., 2016; Bruix et al., 2015). While preclinical studies have uncovered several mechanisms to explain the invasive and metastatic consequences of angiogenesis inhibition (reviewed in Ebos, 2015), assessing the clinical impact remains a challenge (Blagoev et al., 2013). It is possible that the most deleterious consequence of this effect would arise when therapy is stopped (Ebos and Pili, 2012; Ebos and Kerbel, 2011). Antiangiogenic treatment cessation can occur in perioperative clinical trial treatment settings (i.e., neoadjuvant and adjuvant) (Ebos and Kerbel, 2011) and during established “drug holiday” periods for certain VEGF receptor tyrosine kinase inhibitors (VEGFR TKIs) (e.g., sunitinib). However, stopping therapy occurs most commonly in patients who experience progressive disease or toxicity (e.g., up to 60% in some clinical settings with renal cell carcinoma [RCC] patients [Riechelmann et al., 2008; Bex et al., 2017]). Clinically, tumor “flares” in patients have been reported after halting VEGFR TKI or VEGF antibody (bevacizumab) treatment (Powles et al., 2013; Ebos and Kerbel, 2011; Kuczynski et al., 2013), and studies in mice have shown that antiangiogenic treatment withdrawal may drive regrowth via mechanisms involving focal adhesion kinase (FAK)-regulated platelet extravasation (Haemmerle et al., 2016), metabolomic alterations in lipid synthesis (Sounni et al., 2014), and compromised vascular integrity (Yang et al., 2016).

Recent preclinical studies have shown that disruption of the VEGF pathway by gene knockout (Foersch et al., 2015) or therapeutic inhibition (Hasan et al., 2011; Andrae et al., 2012; Zhu et al., 2013) can induce cellular senescence. Senescence is a stress-induced cellular response typically associated with permanent growth arrest that serves to limit replication of aged/damaged cells and facilitate tissue remodeling/repair (Pérez-Mancera et al., 2014). While senescent cells can impede tumor progression, they can also paradoxically promote tumor growth via senescence-associated secretory phenotypes (SASPs) that include cytokines and growth factors capable of activating auto/paracrine tumor/stromal interactions that facilitate metastatic spread (Ewald et al., 2010; Tchkonia et al., 2013; Coppé et al., 2008, 2010; Lecot et al., 2016). Numerous cancer treatments can induce senescence, and antiangiogenic therapy is well-known to induce a broad array of circulating cytokine changes in patients, many of which overlap with common SASP proteins (Jain et al., 2009). However, the durability of this phenotype following treatment resistance and treatment withdrawal has not been examined, nor has senescence been linked to the known metastasis-promoting effects of antiangiogenic therapy (Mastri et al., 2016).

Here, we investigated the role of senescence in tumor growth rebounds following antiangiogenic therapy resistance and withdrawal. To examine this, we generated several VEGFR TKI-resistant cell lines from clinically relevant surgical metastatic models in mice and evaluated the impact of stopping treatment both in vitro and in vivo. VEGFR TKI resistance led to the emergence of multiple senescence hallmarks in all cells tested, including an antiangiogenic therapy-induced secretome (ATIS) that mimicked the SASP. The ATIS could persist for short periods after stopping treatment and coincided with potent primary and metastatic growth rebounds in vivo. Selective inhibition of known SASP regulators mammalian target of rapamycin (mTOR) and, to a lesser extent, interleukin-6 (IL-6), was found to blunt such withdrawal-mediated rebounds. However, senescence markers were highly variable among cells and eventually reversed after long drug withdrawal periods, suggesting that either transition to senescence (and permanent growth arrest) was incomplete, or that VEGF pathway inhibition induces an atypical (i.e., “escapable”) pseudosenescent phenotype reported to occur with other cancer drugs (Saleh et al., 2018; Chakradeo et al., 2016). Taken together, these studies indicate that therapeutic targeting of secretory profiles associated with cellular aging may improve outcomes in patients following antiangiogenic treatment interruption.

RESULTS

Generation of VEGFR TKI-Resistant Cells in Ortho-Surgical Metastasis Models

Preclinical models examining antiangiogenic treatment resistance rarely faithfully represent treatment-resistant metastatic progression as it presents in patients after surgery (reviewed in Mastri et al., 2016). We therefore derived several VEGFR TKI-resistant cell lines from metastatic lesions spontaneously arising after the removal of primary orthotopic-implanted tumors in mice (Benzekry et al., 2016). These “ortho-surgical” metastasis models included cancer cells from kidney (human SN12-PM6-N and mouse RENCA), breast (human LM2–4 and mouse 4T1), and melanoma (human MeWo) implanted orthotopically into mice receiving continuous treatment with sunitinib (Su) or axitinib (Ax). Following selection of metastatic Su- and Ax-resistant cells (SuR and AxR, respectively) in vivo, cells were adapted in vitro with sustained drug treatment or with treatment withdrawn. In our studies, short-term withdrawal (ST-W) (24–48 hr) or long-term withdrawal (LT-W) (2–6 months) conditions were tested to examine changes in metastatic and senescent cell phenotypes (see Figure 1A for schematic). Representative pre- and post-surgery bioluminescence images (BLIs) are shown for 4T1^SuR (Figure 1B) and RENCA^SuR (Figure S1A) cell selection (see Table S1 and STAR Methods for cell selection and ortho-surgical metastasis model details).

Figure 1. Short-Term Antiangiogenic Treatment Withdrawal following Resistance Accelerates In Vivo Tumor Growth and Metastasis.

(A) Schematic of ortho-surgical model used to generate VEGFR TKI treatment-resistant spontaneous Met cells to study the impact of ST-W and LT-W periods (24–48 hr and 2–6 months, respectively).

(B) Representative example of mouse breast 4T1^SuR cell generation with PT growth/resection and post-surgical Met shown in BALB/c mice by bioluminescence imaging (BLI).

(C–E) Orthotopic tumor growth responses following repeated SuR cell Met variant selection and drug exposure during re-implantation into treatment-naive mice. (C) Human breast LM2–4^SuR cells (SCID; n = 5–12 mice, 2 cycles), (D) human melanoma MeWo^SuR (nu/nu; n = 4, 2 cycles), and (E) mouse breast 4T1^SuR (BALB/c; n = 3–5, 1 cycle).

(F–I) In vivo-derived (and in vitro-maintained) SuR cells implanted orthotopically to monitor PT growth progression (left), final tumor weight at resection (left, inset), and post-surgical survival (right) after ST-W. (F) LM2–4^SuR (SCID; n = 4), (G) MeWo^SuR (SCID; n = 4–5), (H) 4T1^SuR (BALB/c; n = 5), and (I) SN12-PM6-N^SuR (SCID; n = 6–11; only final kidney weight and survival shown).

Veh, vehicle; Su, sunitinib; Ax, axitinib; P, parental; SuR, sunitinib-resistant; AxR, axitinib-resistant; ST-W, short-term withdrawal; LT-W, long-term withdrawal; PT, primary tumor; Tx, treatment; Met, metastasis; +/−, with/without treatment. Su (60 mg/kg/day); Ax (100 mg/kg/day). PT burden was assessed by caliber measurement and/or excised tumor or kidney weight. Overall Survival was based on Kaplan-Meier. Quantitative data are shown as mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001, compared to animals implanted with parental cell lines or untreated resistant cells.

See also Figure S1.

Re-implanted VEGFR TKI-Resistant Tumors Remain Treatment Sensitive In Vivo

In our first studies, we examined the tumor-independent impact of treatment interruption using a re-implantation model. SuR or AxR cell variants were re-implanted orthotopically into treatment-naive mice (e.g., to mimic host-treatment cessation), and pre-surgical primary tumor (PT) growth and post-surgical metastatic (Met) progression evaluated with or without continuous treatment (Ebos et al., 2014). Importantly, mice received treatment immediately after implantation to ensure tumor cells had uninterrupted in vitro and in vivo drug exposure. We found Su treatment significantly reduced PT growth in mice bearing newly implanted LM2–4^SuR and MeWo^SuR tumors, despite over 2 years of constant Su drug exposure during two in vivo selection cycles (Figures 1C and 1D). Similar Su treatment re-sensitivity was observed in BALB/c mice bearing 4T1^SuR and RENCA^SuR orthotopic tumors after one cycle (Figures 1E and S1B, respectively). Together, these results confirm prior studies showing VEGFR TKI treatment re-sensitization following treatment interruptions in vivo (Hammers et al., 2010; Zhang et al., 2011; Kuczynski et al., 2013), and demonstrate the extreme durability of this phenomenon with Met cells continuously exposed to drug.

Rapid Acceleration of Local and Metastatic Growth Follows Acute Treatment Withdrawal

We next tested the tumor-dependent impact of treatment interruptions by removing drug and examining treatment initiation in vivo. To do this, SuR/AxR cells were re-implanted orthotopically into treatment-naive mice and drug withheld (representing ST-W conditions). Treatment withdrawal led to significant presurgical PT growth acceleration and decreased postsurgical survival in LM2–4^SuR/ST-W (Figure 1F), MeWo^SuR/ST-W (Figure 1G), 4T1^SuR/ST-W (Figure 1H), and 4T1^AxR/ST-W (Figure S1C) conditions compared to parental (P) cells. Confirmatory results were obtained in ectopic (subcutaneous) SuR and AxR RENCA models in BALB/c(Figures S1D and S1E, respectively; see STAR Methods for details). Treatment withdrawal was next tested in a variant of SN12-PM6 cells, a human kidney tumor model previously shown to be intrinsically Su-insensitive (Ebos et al., 2014) (see Table S1 and STAR Methods for SN12-PM6-N^SuR generation details). ST-W conditions were also found to increase excised kidney weight and significantly decrease overall survival after nephrectomy in the SN12-PM6-N^SuR cell model (Figure 1I). Together, these results demonstrate that Met cells derived from VEGFR TKI-resistant tumors acquire a tumor-promoting phenotype that is revealed during short periods of treatment withdrawal.

Withdrawal-Mediated Tumor Growth Can Be Reversible

We next compared short- and long-term treatment removal conditions in tumor-bearing mice to test whether withdrawal-mediated tumor growth promotion stems from permanent cellular changes. Using combined analysis of pre- and post-surgical datasets to differentiate PT and Met effects (Ebos et al., 2014) (see Figure 2A for schematic and STAR Methods for details), we found ST-W variant cells (implanted into SCID or BALB/c mice) had significantly increased PT growth and decreased survival compared to LT-W variants in three of five models tested (Figures 2A and S2A). These results were confirmed using an intravenous (i.v.) experimental Met model where we observed increased human LM2–4^SuR cells in mouse lungs 24 hr after i.v. implantation (representing ST-W conditions) compared to LM2–4^P-implanted mice (Figure 2B). In contrast, LT-W cell variants were found to have decreased cells in mouse lungs (Figure 2B) and improved overall survival (Figure 2C) compared to ST-W conditions in the LM2–4^SuR model. Notably, SN12-PM6-N^SuR cells showed no such reversibility in LT-W variants, indicating that intrinsically treatment-insensitive cells may acquire tumor-promoting phenotypes independent of drug presence or absence (Figure S2B). Together, these results identify a potent tumor-dependent Met-promoting phenotype in VEGFR TKI-resistant cells in several cancer models that is magnified during acute periods of treatment withdrawal, but can be reversed following protracted therapy removal in a model-dependent manner.

Figure 2. Long-Term Treatment Withdrawal Can Reduce Tumor Growth-Promoting Phenotypes.

(A) Schematic showing combined analysis of ortho-surgical models for differences in (presurgical) PT and (postsurgical) Met growth for ST-W and LT-W conditions normalized to parental cells (top left). Comparisons of SuR/ST-W (red dots), AxR/ST-W (blue dots), and SuR/LT-W or AxR/LT-W (gray dots) cells from LM2–4^SuR (SCID; n = 3–6), MeWo^SuR (SCID; n = 4), and 4T1^AxR (BALB/c; n = 5) shown (all panels). Crossed lines represent the SD of LT-W (gray cross) and ST-W (red or blue cross)-treated data derived from comparisons of PT weight data (vertical dotted line) and OS data (horizontal dotted line). p values for PT burden (Student’s t test) and survival (Kaplan-Meier log-rank) are listed for comparisons between ST-W and LT-W cell lines, with significant values in bold.

(B) Immunofluorescence staining (top) and quantification (bottom) of human vimentin⁺ LM2–4 cell variants in mouse lung tissue 24 hr after i.v. implantation of P, SuR/ST-W, and SuR/LT-W variants (SCID; n = 8–9).

(C) Survival analysis of SCID mice implanted with ST-W or LT-W variants of LM2–4^SuR (SCID; n = 9–10) cells.

SuR, sunitinib-resistant; AxR, axitinib-resistant; PT, primary tumor; Met, metastasis; OS, overall survival; ST-W, short-term withdrawal; LT-W, long-term withdrawal; i.v., intravenous. Su (60 mg/kg/day); Ax (100 mg/kg/day). PT burden was assessed by excised tumor or kidney weight. OS was based on Kaplan-Meier. Quantitative data shown as mean ± SD. *p < 0.05 and ****p < 0.0001.

See also Figure S2.

VEGFR TKI Resistance Induces Transient Pseudosenescent Phenotypes

We next tested whether withdrawal-mediated tumor growth may be linked to cellular senescence. Senescence hallmarks include tumor suppressor pathways associated with growth arrest (e.g., p53/p21 and p16/Rb), morphological changes (including hypertrophy), and changes in senescence-associated β-galactosidase (SA-β-gal), among several others (Malaquin et al., 2016). While we found Su treatment could increase p53 levels in treatment-naive LM2–4^P cells (similar to published reports Sun et al., 2012 and Zhu et al., 2013); neither p53 nor p16 (not shown) were found to be consistently upregulated in SuR variants (Figure 3A). Instead, p21 was elevated in several SuR and AxR tumor cell lines (Figures 3A and 3B; see Data S1 for blotting replicates) and in transformed mouse endothelial (PY4.1^SuR) and fibroblast (3T3^SuR) cells generated by in vitro selection (Figure 3B; see Figure S3A for cell derivation schematic). Interestingly, p21 elevations maintained or increased during ST-W conditions, were decreased during LT-W conditions for five of six cell lines. Next, we found SA-β-gal expression increased in several SuR cells; however, similar to prior studies by Andrae et al. (2012), these were highly variable across cell lines (Figures 3C and S3B). More consistently, we observed cell size increases in all SuR and AxR cells tested (Figure 3D; statistics shown in Figure S3C). One possible senescence-independent cause of cell size increase, such as hypertetraploidy (Lu et al., 2010), was ruled out by karyotyping (Figure S3D). Next, cell cycle analysis showed G1 arrest following Su treatment (in LM2–4^P cells) and after resistance (in LM2–4^SuR cells) (Figure 3E), which could persist during ST-W but not LT-W conditions (Figure 3E; Figure S3E for statistics). This G1 arrest may explain the static/low senescence-like proliferative state of drug-treated SuR cells (measured by MTS assay) (Figures 3F and S3F). Notably, incremental decreases of Su for 7 days (5–0 μM) led to a rebound in proliferation to pre-resistant levels, confirming a lack of permanent cell cycle arrest typically associated with senescence. Finally, we removed Su completely from media in vitro and observed proliferation rebounds after 3–7 days, depending on the cell line (Figure 3G; LM2–4 and RENCA models, respectively). Taken together, these results confirm that antiangiogenic treatment can induce multiple markers of cellular senescence in several VEGFR TKI treatment-resistant tumor and non-tumor cell populations. However, senescence marker induction was highly varied among cell lines and reversed following long-term treatment withdrawal (Figure 3H). These results indicate that antiangiogenic treatment induces an atypical transient pseudosenescent phenotype lacking permanent growth arrest.

Figure 3. Senescence Hallmarks after VEGFR TKI Resistance Are Model Dependent and Reversible.

(A) Western blot analysis for cell-cycle regulatory proteins in LM2–4 SuR cell variants (n = 3).

(B) Representative blots (left) and quantification (right) of p21 protein levels in SuR and AxR cell variants by western blot analysis (n = 3).

(C) Senescence-associated β-galactosidase-positive cell quantification for LM2–4, RENCA, and 3T3 SuR cell variants (left) with representative RENCA SuR cell variants pictured (right) (n = 3).

(D) Cell size measurement for SuR and AxR cell variants. Representative pictures and cell area histograms for LM2–4 SuR cell variants (top) and area quantification for each cell (bottom) are shown (n > 2,500 cells).

(E) Cell-cycle analysis for LM2–4 SuR cell variants (left) with heatmap emphasizing the effect on G1 and G2 during resistance and after short- and long-term therapy withdrawal (right) (n = 3).

(F) Proliferation assay for LM2–4 and RENCA SuR cell variants after Su treatment at different concentrations (n = 5–10).

(G) Proliferation assay for LM2–4 and RENCA SuR cell variants after Su withdrawal for different number of days (n = 6).

(H) Heatmap showing the senescent characteristics acquired by SuR and AxR cell lines before and after LT-W (compared to parental controls).

SuR, sunitinib-resistant; AxR, axitinib-resistant; W, withdrawal; ST, short term; LT, long term; SA-β-gal, senescence-associated β-galactosidase; prolif., proliferation; GA, growth arrest; NT, not tested. For (H), closed circles are positive data, and open circles are negative data for senescent characteristics. ST-W equaled 2 days (for A–C and E) and 7 days (for F). For (A)–(C) and (E), Su (5 μM) or Ax (0.5 μM) treatment was for 2 days in P cells. Quantitative data are shown as mean ± SD, except for cell size (mean ± SEM). *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001, compared to parental cell lines or untreated resistant cells, unless noted otherwise. All western blot images have been adjusted equally for brightness. Replicates and unadjusted uncropped images are shown in Data S1. See also Figure S3.

Pseudosenescent Cells Can Drive Withdrawal-Mediated Rebound Tumor Growth

We next tested whether senescence marker-positive VEGFR TKI-resistant cells can contribute to tumor cell growth after treatment withdrawal. Using a C₁₂FDG fluorescent substrate of SA-β-gal to identify senescent cells (Debacq-Chainiaux et al., 2009; Morelli et al., 2015), we found significant increases in 4T1^AxR cells compared to parental controls (Figure 4A). Using flow cytometry, we sorted 4T1^AxR cells into C₁₂FDG⁺ (high expressing) and C₁₂FDG⁻ (low expressing) cell populations for in vitro and in vivo experiments (Figure 4B). In vitro, C₁₂FDG⁺ cells had an increased proliferation rate after 4 days compared to C₁₂FDG⁻ cells as measured by cell counting and MTS assay (Figures 4C and 4D, respectively). In vivo, orthotopically implanted C₁₂FDG⁺ cells showed significant increases in PT growth followed by decreased survival following tumor resection at day 14, compared to C₁₂FDG⁻ tumors (Figure 4E, left and right, respectively; survival did not reach significance). Together, these data indicate that sub-populations of VEGFR TKI-resistant cells enriched for senescence marker positivity can contribute to tumor growth following acute withdrawal periods.

Figure 4. Senescence Marker-Positive VEGFR TKI-Resistant Cells Contribute to Withdrawal-Mediated Rebound Growth.

(A) 4T1^P and 4T1^AxR cells were incubated with C₁₂FDG and positive cells quantified (n = 3).

(B) C₁₂FDG-incubated 4T1^AxR cells were sorted into C₁₂FDG⁺ and C₁₂FDG⁻ cell populations for in vitro and in vivo testing.

(C and D) In vitro comparison of cell proliferation at day 4 shown by (C) cell counting (right; representative images shown in left) and (D) MTS assay (n = 4–5).

(E) Comparison of 4T1^AxR C₁₂FDG⁺ and 4T1^AxR C₁₂FDG⁻ PT growth following orthotopic implantation of in BALB/c (left; n = 27), and OS following surgical resection (right; n = 14–15; surgery on day 14).

P, parental; AxR, axitinib-resistant; OS, overall survival; PT, primary tumor. Quantitative data are shown as mean ± SD or SEM for in vivo experiment; *p < 0.05 and **p < 0.01, compared to 4T1^AxR C₁₂FDG⁻ cells, unless noted otherwise.

An ATIS Mimics SASPs and Persists during Acute Withdrawal Periods

Next, we undertook transcriptomic and protein analysis in SuR cells following treatment withdrawal and examined secretory phenotypes. First, whole-genome expression profiles were analyzed in three SuR variants (LM2–4, RENCA, and 3T3) to identify genes reversibly or irreversibly changed (Figure 5A; representative example shown in Figure 5B). We then identified genes associated with secretory proteins using a Gene Ontology (GO) database for gene products outside and unattached to the cell (extracellular region, GO:0005576; Mastri et al., 2018). All SuR cells were found to have increased numbers of secretory-associated genes that remained elevated during ST-W conditions, but were then reduced after LT-W (Figure 5C; ST-W for RENCA and 3T3 only). We next compared SuR secretory transcriptomes to senescence-associated and/or therapy-induced secretomes using gene set enrichment analysis (GSEA) based on published profiles related to inflammation, nuclear factor-κB (NF-κB) pathway signaling, oncogene-induced senescence (OIS), DNA-damage secretory responses (DDSRs), senescence-associated inflammatory responses (SIRs), senescence-messaging secretome (SMS), therapy-induced secretomes (TISs), and acute stress-associated phenotypes (ASAPs) (Coppé et al., 2008, 2010; Rodier et al., 2009; Bent et al., 2016; Kuilman et al., 2008; Obenauf et al., 2015; Acosta et al., 2008; Kuilman and Peeper, 2009; Pribluda et al., 2013). Weighted rank scores showed enrichment for secretory gene sets associated with inflammation, senescence, and cancer therapies (Figure 5D; see Data S2 for details); yet these data suggested a gene signature for SuR cells that is distinct. We therefore developed a composite secretory signature composed of secretory genes significantly increased in SuR cell lines and of key SASP genes (Coppé et al., 2010). To do this, we first identified a preliminary target list common to at least two SuR cell lines, and then expanded this list by RNA (Figure S4A) and protein analysis (Figures S4B and S4C) to create an enriched 33-gene ATIS panel (Figure 5E, left; see Data S3 and STAR Methods for details). GO enrichment analysis identified ATIS genes to be significantly represented in several biological processes, including those associated with angiogenesis, migration, proliferation, and senescence (Figure 5E, right). Next, we obtained confirmations of ATIS signature enrichment by examining whole-genome expression data from two published preclinical and clinical studies involving Su-treatment (NCBI GEO database). In the first study, Diaz-Montero et al. (2016) used a RCC patient-derived xenograft (PDX) implantation model to compare tumors sensitive or resistant to Su treatment. In these data, we found ATIS signatures were enriched in both human (tumor) and mouse (host) resistant samples compared to sensitive counterparts (Figure 5F). In the second study, Braga et al. (2017) compared tumor samples from breast cancer (BC) patients treated with Su for 15 days. In these data, we found ATIS signatures were enriched in the non-responding patient cohorts compared to those patients who responded to Su treatment (Figure 5G; see Figure S4D for further secretory profile comparisons). Finally, to further validate specific ATIS members known to play a key role in SASP expression/regulation, we evaluated levels of IL-6, IL-1α, and IL-1β in cell lysates from LM2–4 SuR variants following short- and long-term treatment withdrawal. We found IL-6, IL-1α, and IL-1β protein increases in SuR variants significantly increased further in ST-W conditions, but then decreased when drug was removed long-term (Figure 5H). Taken together, these results identify a SASP-mimicking VEGFR TKI resistance-induced secretory profile in tumor/non-tumor cell populations that can persist during short periods of treatment withdrawal and, importantly, reverse or diminish following protracted periods of drug removal.

Figure 5. Identification of an ATIS following Resistance.

(A) Schematic for identification of reversible and irreversible gene changes after VEGFR TKI resistance/withdrawal (left) with a representative volcano plot showing gene expression changes that reverse after ST (green) and LT (blue) withdrawal (right; n = 3; 3T3 SuR cells shown as example; irreversible gene expression changes shown in red).

(B) Heatmaps showing reversible/irreversible gene expression changes after therapy withdrawal (n = 3; 3T3 cell variants shown as example).

(C) Number of upregulated genes (based on GO database for extracellular region term; GO:0005576) in SuR cell variants compared to parental (n = 3; LM2–4, 3T3, and RENCA cell variants shown).

(D) SuR cell variants (n = 3) were compared to known inflammatory pathways and published secretomes induced by oncogenes, cancer therapy, and senescence using GSEA. Note: gene set information is summarized in Data S2.

(E) Identification of 33 ATIS gene members in SuR cell variants (left); GO analysis of ATIS signature for significantly enriched biological process terms (right).

(F and G) GSEA analysis of the ATIS signature in published datasets involving Su treatment in preclinical and clinical studies. (F) Preclinical: enrichment of ATIS in tumor and host tissues derived from a Su-resistant RCC PDX model (n = 4; GEO:GSE76068). Tumor samples obtained pre-treatment (pre-Tx), were compared to samples from sensitive and resistant time points. (G) Clinical: enrichment of ATIS in locally advanced human BC patients responsive or non-responsive to su-treatment, compared to pre-treatment levels (n = 5–7; GEO:GSE58837).

(H) Protein quantification by ELISA in lysates of LM2–4 SuR cell variants for IL-6, IL-1α, and IL-1β (n = 4).

P, parental; SuR, sunitinib-resistant; ST-W, short-term withdrawal; LT-W, long-term withdrawal; NT, not tested; GO, Gene Ontology; GSEA, gene set enrichment analysis; Su, sunitinib; RCC, renal cell carcinoma; BC, breast cancer; PDX, patient-derived xenograft; see STAR Methods for details on use of published data obtained from GEO database. ST-W equaled 2 days; P cells treated with Su (5 μM) for 2 days. Quantitative data are shown as mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001, compared to parental cell lines, unless otherwise noted. See also Figure S4.

Therapeutic Inhibition of SASP Regulators Can Limit Withdrawal-Mediated Tumor Growth

We next sought to determine whether therapeutic inhibition of key SASP proteins or regulators shared with the ATIS could impact withdrawal-mediated tumor growth. First, we examined IL-6, as it is an SASP protein with several effector roles in tumor promotion and inhibition of IL-6 has recently been linked to improved VEGF treatment strategies in mice (Zhu et al., 2015; Eichten et al., 2016). IL-6 expression was found to be significantly upregulated in seven of nine SuR/AxR tumor and non-tumor cell lines tested (Figure S5A). We therefore evaluated IL-6 inhibitors in two SuR models in ST-W conditions (as described in Figures 1F–1I). In the first model, anti-mouse IL-6 (α-mIL-6) antibody was administered to BALB/c mice immediately after orthotopic implantation of 4T1 SuR and P variants (Figure 6A). Significant reductions were observed in PT growth in 4T1^SuR/ST-W, but not 4T1^P, tumors compared to untreated controls, indicating the role of IL-6 on PT growth is increased in ST-W conditions. In a second model, α-mIL-6 and anti-human IL-6 (α-hIL-6) antibodies were administered to SCID mice immediately after orthotopic implantation of LM2–4 SuR and P cell variants to compare the impact of tumor (human) and host (mouse) IL-6 rebound growth before and after surgery (Figure 6B; Figures S5B and S5C for full results). IL-6 blockade was similarly found to inhibit SuR/ST-W tumor growth (compared to P controls); however, significance was only reached with α-mIL-6 antibody treatment (Figure 6B). These results indicate that host-mediated IL-6 may be activated via paracrine interactions with SuR tumor variants and, in turn, contribute to withdrawal-mediated rebound growth. In support of this hypothesis, we observed elevations in mouse (host) IL-6 in SCID bearing LM2–4^P tumors (compared to non-tumor-bearing mice), that was increased further in LM2–4^SuR models (Figure S5D). In our next studies, we examined whether mTOR may regulate ATIS proteins in a manner similar to its regulation of SASP proteins (such as IL-6 and IL-1α) via NF-κB (Herranz et al., 2015; Laberge et al., 2015; Salminen et al., 2012). We found that activity of NF-κB (measured by phospho-p65) and mTOR (measured by phospho-S6) increased after Su treatment and resistance (Figure 6C; blotting replicates in Data S1); and treatment with the mTOR inhibitor rapamycin could decrease p-S6 and p-p65 (Figure 6D; blotting replicates in Data S1). Next, we found that SuR-mediated elevations in IL-6 and IL-1α were significantly reduced after rapamycin treatment in cell lysates, suggesting the ATIS and SASP share mTOR-regulated components (Figure 6E). Notably, we found ATIS (and SASP) proteins CCL2 and IL-1β were also increased in SuR and SuR/ST-W conditions (Figure S5E). Interestingly, rapamycin blocked CCL2 but not IL-1β, which is consistent with prior reports of mTOR regulation being limited to some, but not all, SASPs (Laberge et al., 2015). In vivo, we observed that rapamycin treatment could significantly reduce withdrawal-mediated rebound PT growth and increase postsurgical overall survival in SCID mice implanted with LM2–4^SuR cells in ST-W conditions (Figure 6F). Importantly, the growth-inhibitory effects of rapamycin treatment in orthotopically implanted 4T1^AxR C₁₂FDG⁺ tumors was significantly improved over 4T1^AxR C₁₂FDG⁻ tumors, when compared to untreated controls in ST-W conditions (Figures 6G and 6H). Following experiment termination at day 21, we found significant increases in lung Met nodules in mice bearing C₁₂FDG⁺ tumors (compared to C₁₂FDG⁻ tumors), which were significantly reduced after rapamycin treatment (Figure 6I). Together, these results identify IL-6 is an ATIS member contributing, at least in part, to withdrawal-mediated rebound tumor growth. They also demonstrate that targeting mTOR may be an effective strategy to suppress multiple ATIS growth-promoting proteins simultaneously (including IL-6, IL-1α, and CCL2).

Figure 6. Therapeutic Inhibition of IL-6 and mTOR Suppresses VEGFR TKI Withdrawal-Mediated Rebound Tumor Growth.

(A) Tumor volume comparison 23 days after orthotopic implantation of 4T1^P and 4T1^SuR cells (representing ST-W conditions) followed by treatment with α-mIL-6 antibody (BALB/c; n = 11–12).

(B) Combined analysis of differences in (presurgical) PT and (postsurgical) survival in an ortho-surgical model comparing α-mIL-6 (black dots) and α-hIL-6 (gray dots) treatment after LM2–4^P (SCID; n = 5) and LM2–4^SuR/ST-W (SCID; n = 10–18) cell implantation. Crossed lines represent the SD of α-mIL-6 (black cross) and α-hIL-6 (gray cross)-treated data derived from comparisons of PT burden data (vertical dotted line) and median survival data (horizontal dotted line). P values for PT burden (Student’s t test) and survival (Kaplan-Meier log-rank) are listed for comparisons between P and SuR cell lines, with significant values in bold.

(C and D) Quantification of phosphorylation levels of S6 and NF-κB-p65 proteins via western blot analysis for (C) LM2–4 SuR cell variants (n = 3) and (D) after 12.5 nM rapamycin treatment for 48 hr (n = 3).

(E) ELISA quantification of secreted IL-6 and IL-1α levels in conditioned media of LM2–4 SuR cell variants 48 hr after 12.5 nM rapamycin treatment (n = 4).

(F) Comparison of PT burden (left) and survival (right) in mice implanted orthotopically with LM2–4^SuR cells (representing ST-W conditions) and treated with rapamycin (SCID; n = 15).

(G) Comparison of 4T1^AxR/ST-W C₁₂FDG⁺ and 4T1^AxR/ST-W C₁₂FDG⁻ PT growth following orthotopic implantation in BALB/c mice (n = 6) and treated with rapamycin (starting at day 10).

(H and I) Comparison of C₁₂FDG⁺ and C₁₂FDG⁻ tumor responses to rapamycin treatment (compared to respective vehicle-treated control) based on (H) PT growth and (I) lung Met nodules at experiment termination (day 21).

Veh, vehicle; P, parental; SuR, sunitinib-resistant; AxR, axitinib-resistant; ST-W, short-term withdrawal; LT-W, long-term withdrawal; PT, primary tumor; Met, metastasis; Tx, treatment; Rap, rapamycin. One-sample Student’s t test was used for (D). Quantitative data are shown as mean ± SD. Treatment was stopped at day of surgery for (B), whereas it was continued until endpoint for (F). α-mIL-6 (100 μg/mouse/3 days); α-hIL-6 (400 μg/mouse/3 days). ST-W equaled 2 days in vitro; P cells treated with Su (5 μM) for 2 days. Overall Survival was based on Kaplan-Meier. Rapamycin (5 or 10 mg/kg/day). *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001, compared to untreated animals or parental cell lines, unless otherwise noted. Replicates and unadjusted uncropped images are shown in Data S1. See also Figure S5.

DISCUSSION

The implications of stopping antiangiogenic therapy on disease progression remain understudied. Halting a VEGF pathway inhibitor treatment may have a positive impact, such as the eventual re-sensitization to therapy (Kuczynski et al., 2013), or a negative impact, namely, by increasing the potential for rebound growth and thus limiting the overall benefits of therapy (Ebos and Kerbel, 2011). It is also possible that both outcomes occur simultaneously. To examine this, we developed surgical models of anti-angiogenic treatment-resistant metastatic disease to study the impact of treatment cessation. Our results identified a “pseudosenescent” phenotype in tumor and non-tumor cells capable of promoting tumor growth during acute periods of therapy withdrawal. Transcriptomic and proteomic analysis identified an ATIS profile that, similar to SASPs, is regulated by mTOR signaling and proteins such as IL-6. Importantly, induction of senescent cell phenotypes was highly varied and transient, indicating that senescence program activation is either incomplete or atypical, and can eventually diminish. These results link rebound tumor growth after antiangiogenic treatment to the paradoxical tumor-promoting secretory profiles associated with cellular aging.

The concept of rebound growth after antiangiogenic treatment withdrawal remains controversial (Ebos and Kerbel, 2011). Previous publications in both mice (Mancuso et al., 2006) and humans (Griffioen et al., 2012) have demonstrated that revascularization and endothelial cell proliferation can occur in tumors when VEGFR TKI treatment is halted (Ebos and Pili, 2012). However, retrospective analyses of patients having TKI or antibody (bevacizumab) treatments stopped have shown no growth acceleration (Miles et al., 2011; Blagoev et al., 2013). A challenge is that comparisons between humans and mice are confounded by a multitude of differences. Patients with refractory metastatic disease often are co-treated with chemotherapy and have therapy stopped for different reasons (i.e., toxicity, progression, or both combined)—which are conditions rarely replicated in animal studies (Ebos et al., 2014). To date, preclinical studies have mostly assessed treatment cessation in non-resistant and non-metastatic systems that exclude surgery, with the sustainability (reversibility) of any negative effects still unknown (Bagri et al., 2010). For this reason, our studies involving postsurgical metastatic treatment-resistant disease comparing ST-W and LT-W periods represent an important clinically relevant approach. We identified a “window” immediately after withdrawal that may account for growth-promoting “rebound” effects. Importantly, our findings confirm prior reports that re-implantation and re-treatment of selected resistant tumor cells can lead to tumor growth suppression in treatment naive mice (Hammers et al., 2010; Zhang et al., 2011). These results show the extreme durability of this effect, as selected cells over a 2-year period of in vivo/in vitro selection can still remain sensitive to treatment. This may resemble, in part, observations in patients who have progressed on antiangiogenic therapy and continue to receive benefits from treatment. For instance, Burotto et al. (2014) have shown that metastatic RCC patients receiving continued Su-treatment beyond response evaluation criteria in solid tumors (RECIST)-defined progression may have a higher chance of beneficial outcome via disease stabilization. Based on our studies, it is possible that the benefits of treatment beyond progression may also be attributed to suppression of growth-promoting phenotypes kept at bay by treatment.

Related to this, our findings raise key questions about escape from treatment-induced senescence. While we provide critical confirmation of prior studies linking VEGF pathway inhibition to increases in senescence markers (Foersch et al., 2015; Hasan et al., 2011; Zhu et al., 2013; Morelli et al., 2015, 2017), we used several drugs and cell lines to show these markers are highly varied, model-dependent, and can diminish after long periods of drug withdrawal. These findings are consistent with increasing evidence showing treatment-induced senescence may be atypical (i.e., non-permanent) in several instances (Saleh et al., 2018). Senescence is increasingly considered on a spectrum with multiple effectors present at different “early” and “late” stages (Salama et al., 2014). For this reason, early-stage reversible/uncommitted senescence phenotypes have been variously called “premature,” “pseudosenescent,” “short-lived,” “senescence-like,” or “accelerated” (Kim et al., 2014; Demaria et al., 2014; Gewirtz et al., 2008), and may be distinct from senescence induced by replication or oncogenes (reviewed in Ewald et al., 2010; Gewirtz et al., 2008; Saleh et al., 2018; Chakradeo et al., 2016). For instance, doxorubicin can induce a limited senescence in endothelial cells (Bent et al., 2016), and there are several additional examples as well (Georgilis and Gil, 2016; Bent et al., 2016; Obenauf et al., 2015; Chakradeo et al., 2016). Importantly, our results also compliment prior studies showing Su-induction of senescence or senescence-related regulators, including NF-κB (Sanchez et al., 2013) and IL-6 (Zhu et al., 2015), SA-β-gal (Andrae et al., 2012), p53 (Zhu et al., 2013), and mTOR (Elgendy et al., 2017; Jiménez-Valerio et al., 2016). Interestingly, p21 was most commonly upregulated after VEGFR TKI resistance in our models and seemed to occur independent of p53 and p16. Such p21-dependent senescent phenotypes have been reported previously in certain conditions (Muñoz-Espín et al., 2013) and highlight the diversity (and complexity) involved in cellular progression toward senescence induced by therapy.

However, perhaps the most important issue related to transient treatment-induced senescent cell characteristics is the impact on tumor-promoting SASPs. Circulating secreted proteins are well known to be induced by multiple cancer therapies (Ebos and Pili, 2012), although a connection to SASPs in most cases remain unclear. While the SASP seems to be relatively conserved in mammals, it is highly context dependent and variable among cell types and senescence-inducing stimuli (Malaquin et al., 2016; Mastri et al., 2018). In our study, ATIS was found to be broadly consistent with, but not precisely identical to, secretory gene sets associated with inflammation, senescence, and cancer therapies (Pribluda et al., 2013; Sun and Nelson, 2012). The ATIS shares multiple common SASP proteins and, critically, shares common regulatory elements that may be therapeutically targeted to reverse tumor-promoting effects (Tchkonia et al., 2013). For example, mTOR is a key post-translational SASP regulator via pathways involving mitogen-activated protein kinase (MAPK)/p38 (Herranz et al., 2015) and IL-1α expression (Laberge et al., 2015), the latter of which can regulate NF-κB-mediated SASP secretion. Our results show that mTOR can also exert control over the ATIS, as SuR tumor cell variants were found to have increased activity of mTOR/NF-κB, and ATIS proteins (such as IL-6, IL-1α, and CCL2) could be decreased by rapamycin treatment. Importantly, the tumor-inhibitory effects of rapamycin on withdrawal-mediated growth were significantly increased in C₁₂FDG⁺ tumor models, thereby introducing a mechanism to explain why antiangiogenic treatment resistance may sensitize cells to mTOR inhibition (Elgendy et al., 2017; Jiménez-Valerio et al., 2016). Indeed, mTOR targeted therapies have worked better in metastatic RCC (mRCC) patients after initial treatment with VEGFR TKIs. In a phase II non-inferiority trial (RECORD-3), 471 previously untreated mRCC patients were given first-line Su or everolimus (an mTOR inhibitor) and then, after disease progression, crossed over to receive the alternative agent. Outcomes with the Su-everolimus regimen were significantly improved compared to the everolimus-Su regimen (2.8 and 7.1 months, for progression-free survival and overall survival, respectively) (Motzer et al., 2014). Together, our findings suggest mTOR-mediated secretory factors induced by Su could sensitize everolimus treatment effects. If so, then shortening Su washout periods before drug switching (2-week minimum for RECORD-3) may further improve therapeutic outcomes. Further analysis of shortened intervals between treatments could test this hypothesis.

Finally, an important implication of a SASP-related ATIS protein signature is that it may explain the array of tumor and host-mediated secretory changes induced by antiangiogenic therapy. Numerous studies have investigated plasma proteins as surrogate biomarkers of biological effect (i.e., drug hits target) or an anti-tumor effect (i.e., drug influences tumor growth) (Ebos et al., 2009b). While we and others have identified VEGF pathway inhibitor-induced proteins as highly dose-dependent, no single source mechanism has been identified. It is possible that the SASP-mimicking ATIS may explain these phenomena as many common SASP and ATIS markers are well-known to be induced by antiangiogenic therapies (i.e., IL-6, IL-8, SDF-1α, and VEGF, to name only a few) (Duda, 2012). While the ATIS may be a potential predictor of poor outcomes in cancer patients independent of treatment (see Figure S6), it may be possible that ATIS activation can occur independent of cancer itself. For instance, we recently showed that antiangiogenic treatments in non-tumor-bearing mice can produce systemic host secretory changes (or “therasomes”) consistent among several treatment types, including those in the same class (i.e., other VEGFR TKIs), different modes of action (i.e., antibodies), or even with different targets entirely (i.e., PD-1 pathway inhibitors) (Mastri et al., 2018). A re-analysis of these published gene sets showed that ATIS gene signatures can be enriched in cancer-free settings after Su and Ax treatment (see Figure S4D). Together, these results suggest that global secretory changes in response to therapy may derive, at least in part, from mechanisms typically associated with senescence. Future studies are needed to determine whether ATIS gene signatures in non-tumor tissue may represent a new biomarker for antiangiogenic treatment and treatment resistance in patients.

Taken together, our results demonstrate that antiangiogenic treatment resistance can transiently hijack numerous secretory profiles associated with cellular aging. Senotherapies aimed at targeting SASP-mimicking ATIS components or regulators may be effective in prolonging sensitivity to VEGF pathway inhibitors and blunting tumor-promoting effects when treatment is stopped.

STAR★METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by Lead Contact, John M.L. Ebos (John.Ebos@RoswellPark.org).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell lines

Cells used in this study include: Human breast carcinoma LM2–4 cells (a metastatic derivative of MDA-MB-231 from R. Kerbel, Sunnybrook Research Institute, SRI), mouse mammary carcinoma 4T1 cells (from A. Gudkov, Roswell Park Comprehensive Cancer Center, RPCCC), human kidney SN12-PM6-N cells (a metastatic variant of SN12-PM6 cells from R. Kerbel, SRI, derived in nude mice), mouse kidney RENCA (from R. Pili, RPCCC), human melanoma MeWo cells (from R. Kerbel. SRI), immortalized mouse fibroblasts 3T3 (from I. Gelman, RPCCC), and immortalized endothelial cells PY4.1 (from D. Dumont, SRI). All cells were maintained in DMEM (Corning cellgro; 10–013-CV) except for RENCA and 4T1 (RPMI; Corning cellgro; 10–040-CV). All growth media were supplemented with 5% v/v FBS (Corning cellgro; 35–010-CV). Cells were incubated at 37°C and 5% CO₂ in a humidified incubator. All untreated parental cell lines were authenticated by STR profile comparison to the ATCC or DSMZ cell database (for human cells, where available) or were confirmed for species origin (for mouse cells) (DDC Medical, USA). All drug-resistant cells were authenticated by STR profile comparison to their corresponding parental cells to eliminate any trans-contamination during selection process.

Drugs and Doses Used

SU11248/Sunitinib malate (Sutent^®©, Pfizer), AG013736/Axitinib (Inlyta^®©, Pfizer), and rapamycin (R-5000, LC Labs) were prepared as follows. In vivo: sunitinib (Su) and rapamycin were suspended in vehicle formulation containing carboxymethylcellulose sodium (USP, 0.5% w/v), NaCl (USP, 1.8% w/v), Tween-80 (NF, 0.4% w/v), benzyl alcohol (NF, 0.9% w/v), and reverse osmosis deionized water (added to final volume) and adjusted to pH 6 as described in Ebos et al. (2007). Axitinib (Ax) was suspended in vehicle formulation containing carboxymethylcellulose sodium (USP, 0.5% w/v) and reverse osmosis deionized water (added to final volume). All treatments were administered by oral gavage at 60 mg/kg/day (Su), 100mg/kg/day (Ax), and 5 or 10 mg/kg/day (rapamycin) as described elsewhere (Ebos et al., 2007; Wagner et al., 2012). In vitro: Su was dissolved in water, and Ax and rapamycin were dissolved in DMSO as recommended by the manufacturer. Neutralizing IL-6 antibodies for human (Pfizer PF-04236921) and for mouse (BioXcell BE0046/MP5–20F3) were diluted in PBS and administered by intraperitoneal injection at 100 μg/mouse/3 days (anti-mouse IL-6; α-mIL-6) and 400 μg/mouse/3 days (anti-human IL-6; α-hIL-6).

Mouse tumor models

Study approval

Animal tumor model studies were performed in strict accordance with the recommendations in the Guide for Care and Use of Laboratory Animals of the National Institutes of Health and according to guidelines of the IACUC at Roswell Park Cancer Institute (Protocol: 1227M).

Ortho-surgical models of metastasis

The optimization and use of animal models of breast, melanoma, and kidney metastasis following orthotopic primary tumor implantation and surgical resection (‘ortho-surgical’) have been extensively detailed elsewhere (Ebos et al., 2014). Briefly, LM2–4 (1×10⁶ cells in 100 μL DMEM), 4T1 (5×10⁴ cells in 100 μL DMEM), RENCA (4×10⁴ cells in 2.5 μL RPMI and 2.5 μL matrigel), SN12-PM6-N (1×10⁶ cells in 5ul DMEM), MeWo (2×10⁶ cells in 100 μL DPBS) were implanted orthotopically into the right inguinal mammary fat pad (right flank), left kidney (subcapsular space), or dermis (right flank) of 6–8 week old female SCID or BALB/c mice depending on the model (Ebos et al., 2014). Primary breast and melanoma tumor size was assessed regularly with Vernier calipers using the formula width²(length × 0.5) and, for tumor cells expressing luciferase, animals were monitored bi-weekly for bioluminescence (BL) to quantify tumor growth (Ebos et al., 2009a). Surgical removal of breast and melanoma tumors, as well as nephrectomy of tumor-bearing kidneys, followed guidelines described previously (Ebos et al., 2014; Benzekry et al., 2016; Tracz et al., 2014) at predetermined optimal surgery times (Ebos et al., 2014) in order to minimize potential for invasion of primary tumor to adjacent organs and maximize establishment of metastatic disease.

Ectopic tumor implantation

RENCA (1×10⁵ cells in 50 μL RPMI and 50 μL matrigel) were implanted in the subcutaneous space (right flank) of 6–8-week old female BALB/c mice. Tumor growth was monitored with Vernier calipers using the formula width²(length×0.5) and BLI.

Experimental metastasis assays

LM2–4 (1×10⁶ cells in 100 μL DPBS) and SN12-PM6-N (1×10⁶ cells in 100 μL DPBS) were injected directly into the tail vein of 6–8 week old female SCID as previously described (Ebos et al., 2009a).

Treatment-resistant cell derivation and maintenance

Derivation of SuR or AxR cell variants from spontaneous metastatic lesions formed in the ortho-surgical model included the following general procedures. Su or Ax was administered to mice after orthotopic implantation and continued after surgical resection of the primary tumor (LM2–4, 4T1, MeWo) or kidney (SN12-PM6-N, RENCA) until metastatic endpoint (except day of surgery). At endpoint, organs containing metastatic lesions were minced, enzymatically digested (tumor dissociation solution; Miltenyi Biotics; 130–095-929), and filtered to obtain single-cell suspension (40 μm cell strainer; BD Falcon; 352340). Cells were then placed in media (DMEM or RPMI supplemented with 5% v/v FBS, 1000 IU/ml penicillin, and 1000 μg/ml streptomycin) with Su (1 μM, 2.5 μM, and 5 μM) or Ax (0.5 μM or 1 μM). One week later, antibiotics were removed and the highest tolerable Su or Ax concentration was used for continued in vitro maintenance (2–6 months) prior to re-implantation. Resistant cells were re-implanted in vivo and selected in vitro 1–2 times depending on the cell line. Long-term treatment withdrawal (LT-W) cell variants were maintained without drug in parallel with SuR or AxR variant selection. 3T3^SuR and PY4.1^SuR variants were selected in vitro via incremental weekly increases of Su (250nM) until concentration of 5mM was reached. After 2 months, cells were split into two groups containing either drug (SuR) and or having had drug withdrawn (SuR/LT-W). Cells were then cultured for an additional 2 months. Short-term treatment withdrawal (ST-W) cell variants were maintained without drug for only 24–48 hours in vitro or after implantation until endpoint in vivo.

Additional information for selection of each SuR or AxR cell line used in this study is presented in Table S1. Specific variations depending on the cell line are as follows:

Human cells

LM2–4 is a luciferase expressing metastatic human MDA-MB-231 breast cancer-cell line derived after multiple rounds of in vivo lung metastasis selection in SCID mice in an ortho-surgical model (described previously; Ebos et al., 2014; Munoz et al., 2006). Human kidney SN12-PM6-N cells are a variant of SN12-PM6 cells (Ebos et al., 2014) selected from a metastatic lung lesion in nu/nu after tumor-bearing kidney removal. Metastatic variants of human MeWo cells were not previously selected prior to this study. In all models, selection of SuR cell variants from spontaneous (postsurgical) metastatic lesions was conducted in nu/nu mice due to toxicity of long-term (more than 30–40 day) VEGFR TKI treatment in SCID mice (described previously; Tang et al., 2010). Cell variant selection was performed over 1 round (SN12-PM6-N^SuR) or 2 rounds (LM2–4^SuR and MeWo^SuR) of selection with drug treatment exposure in vivo and in vitro extending over a period of 6 months to 2 years (see Table S1). Due to the variable nature of spontaneous metastasis following surgery and duration of Su treatment efficacy (e.g., a combined 389 days in vivo for LM2–4), SuR variants were derived from organs where metastasis gave the greatest chance of variant selection (i.e., lung, kidney, lymph node, etc.). See Table S1 for details. In studies described in Figures 1F, 1G, 1I, 2, and 6) SuR cells were used in SCID mice.

Mouse cells

Luciferase-expressing mouse kidney RENCA and mouse mammary carcinoma 4T1 cells were implanted orthotopically into BALB/c mice and treated with Su or Ax. RENCA^SuR, RENCA^AxR, 4T1^SuR, and 4T1^AxR variants used for analysis were obtained over 1 round of selection (see Table S1).

Methodological considerations for ortho-surgical metastatic models

For all ortho-surgical metastasis models, we followed detailed experimental criteria to control for variable disease progression and model standardization as previously described (Ebos et al., 2014). Briefly, these included the following considerations:

Inclusion/exclusion criteria

Upon surgical removal of primary tumor, if tumor invasion was noted – i.e., growth into peritoneal space (breast, melanoma) or presentation of a non-encapsulated kidney tumor (kidney) - mouse was excluded from study if complete removal of all visible tumor was not possible (described elsewhere; Benzekry et al., 2016). Further, if tumor was not present at any time before and after surgery (determined by BLI or visible macroscopically), mice were excluded from study to eliminate potential for false positives or negative bias to results (Ebos et al., 2014).

Randomization

Randomization procedures for in vivo experiments were performed as previously described (Suresh, 2011). Briefly, mice were randomized either prior to implantation (in studies where treatment started immediately after implantation) or by primary tumor size to ensure equal tumor burden (for studies where treatment was initiated after tumor growth was measurable).

Surgery timing optimization

Surgical timing of primary tumor removal was optimized to ensure primary tumor (as measured by volume, BLI, and duration of growth) could sufficiently metastasize (and exclude surgical cure) but also not progress to the point of local primary tumor invasion at time of surgery (which would confound measurement of spontaneous systemic metastatic disease). For all models, prior optimization of time points was used as described previously (Ebos et al., 2014).

Evaluation of end-stage metastatic disease distribution

Animals were sacrificed according to institutional guidelines, with metastatic endpoints including signs of distress, labored breathing, weight loss, or, if present, measurable metastatic tumor growth at or near institutional size limit guidelines.

Note: All in vivo experiments described in this study involved orthotopic implantation with the exception of studies involving RENCA^SuR and RENCA^AxR (Figures S1D and S1E) which evaluated ectopic (subcutaneous) tumor growth following ST-W conditions. This exception was made because the time period between orthotopic RENCA cell implantation and nephrectomy is relatively short (20 days) and at time of surgery tumor burden is low, making quantification of differences challenging. In this instance, ectopic implantation allowed for implantation of more cells, longer growth duration, and statistical assessment of primary tumor growth differences.

METHOD DETAILS

Senescence-Associated β-galactosidase assays

Cytochemical fixed-cell assay

Parental or resistant cells were plated in 12-well plates in their corresponding growth media at 1000 cells/well. The next day, the cells were washed and media were added with or without Su (5 μM) or Ax (0.5 μM) treatment for 2 days. Cells were then stained for SA-β-gal activity using the Senescence β-Galactosidase Staining kit following the manufacturer’s instructions (Cell Signaling Technology; 9860). For quantification, 10 images were taken for each replicate for a total of 30 for each experimental condition tested. Cells on each image were visually scored for colorimetric changes and then results pooled for analysis.

Fluorescent live-cell assay and sorting

4T1 P and AxR cell variants were grown in corresponding growth media and, at approximately 90% confluency, media containing 100nM Bafilomycin A1 (Invivogen; 88899–55-2) was placed on cells for 1 hour (to alkalinize the lysosomes to increase pH and expose SA-β-gal). 5-dodecanoylaminofluorescein di-β-D-galactopyranoside (C₁₂FDG), a fluorogenic substrate for βgal activity (Setareh Biotech; 7188), was then added for 1 hour at a final concentration of 33 μM, as previously described (Debacq-Chainiaux et al., 2009). Cells were then washed 2x with PBS, trypsinized, and prepared for processing with a LSR II flow cytometer for studies involving P and AxR cell comparison or FACSAria cell sorter (Becton Dickinson) for studies involving AxR variation separation into C₁₂FDG⁺ and C₁₂FDG⁻ cell populations (Morelli et al., 2015). For in vivo studies, C₁₂FDG⁺ and C₁₂FDG⁻ cells (5×10⁴ cells in 100 μL DMEM) were implanted orthotopically into the right inguinal mammary fat pad (right flank) of 6–8-week old female BALB/c mice as described elsewhere (Ebos et al., 2014). In vitro experimental conditions are described below. Studies involving Ax treatment and C₁₂FDG have been done previously (Morelli et al., 2015). Note: all C₁₂FDG experiments used 4T1 AxR cell variants only. SuR cell variants were not suitable due to Su-induced autofluorescence (described elsewhere; Gotink et al., 2011) at the same wavelength (green) as C₁₂FDG (data not shown).

Cell proliferation assay

Cell proliferation was assessed using the CellTiter 96 Aqueous Non-Radioactive cell proliferation (MTS) assay (Promega; G1112). Cells were plated in 96-well plates in their corresponding growth media (5×10³ and 1×10³ cells/well for human and mouse cell lines, respectively). The next day, cells were washed and treated with Su. Su-containing media was replaced every two days for dose-response studies (during a 7-day period; as described in Figure 3F) or removed daily for time-response experiments (3 days for human cell lines and 7 days for mouse cell lines, as described in Figure 3G). At experiment endpoint, MTS was added to the cells and, after 2 hours, optical density measured at wavelength of 490nm. For cell proliferation assays using 4T1^AxR FDG sorted cells (Figures 4C and 4D), cells were 1) plated in 6 well plates and images were taken every day for 4 days prior to visual counting and analysis using ImageJ (quantified using 5 separate images), or 2) cells were plated in 96-well plates (1×10³ cells/well in growth media without Ax) and MTS was performed 4 days later, as described above.

Cell size analysis

Cells were plated in their corresponding media and trypsinized when 90% confluency was reached. After 3 washes, cells were resuspended in PBS and processed with ImageStream (Amnis Corporation). Data was analyzed with the IDEAS image analysis software (Amnis Corporation) and cell area was obtained for each cell.

Cell cycle analysis

Cells were plated in their corresponding growth media and the following day 5 μM Su was added or removed, as indicated. Two days later, cells were trypsinized, washed with PBS and fixed in 70% EtOH. DAPI was added 15 minutes before cells were processed with an LSR II flow cytometer (Becton Dickinson). Data was collected with FACSDiva software (Becton Dickinson) and analyzed with ModFit software (Verity Software House).

In situ immunostaining

After intravenous implantation, as described previously, lung tissues were excised 24 hours later, immediately immersed in Cryo-embedding compound (Ted Pella, Inc; 27300), quickly frozen over dry ice (CO₂), and stored at −80°C until use. Forty μm thick cross sections were fixed in acetone:ethanol mixture (3:1 ratio), blocked with 2% BSA in PBS, and stained with rabbit anti-human Vimentin (Abcam; ab16700) and rat anti-mouse PECAM-1 (Dianova; DIA-310) diluted 1:100 v/v in 1% w/v BSA in PBS, followed by a FITC conjugated goat anti-rabbit IgG (BD PharMingen; 554020) and Cy3 conjugated goat anti-rat IgG (Invitrogen; A10522) diluted 1:500 v/v in 1% w/v BSA in PBS. Sections were mounted using Vectashield Hard Set Mounting medium for fluorescence (Vector; H-1400) supplemented with DAPI (Invitrogen; D3571) at 1:100 v/v dilution. Images were taken in 10 random fields per section (3 sections per animal) using a Zeiss AxioImager A2 epifluorescence microscope at 200x magnification. The percent area of Vimentin was measured by ImageJ software.

Karyotyping

To obtain metaphase spreads to quantify chromosome numbers, LM2–4 SuR cells were submitted to Pathology Resource Network - SKY/FISH facility at Roswell Park Cancer Institute for chromosome count.

RNA isolation

Cells were plated in their corresponding growth media and after one day 5 μM Su was added or removed, as indicated. Two days later cells were harvested and total RNA isolated using a QIAshredder (QIAGEN; 79654) and RNase mini kit (QIAGEN; 74104). Genomic DNA was digested using DNase I (QIAGEN; 79254) as described in the on-column DNase digestion protocol from QIAGEN. RNA concentration was determined using nanodrop 2000c (Thermo Scientific).

Endpoint PCR

One microgram RNA was used for reverse transcription using iScript cDNA synthesis kit (Bio-Rad; 170–8891) according to the manufacturer’s instructions. Endpoint PCR was performed using Taq polymerase as recommended by the manufacturer (Promega; M7502). Thermocycling parameters were: 2 min at 95°C, 30 cycles at 95°C for 30 s, 55°C for 30 s, and 72°C for 30 s, followed by a final extension of 5 min at 72°C. Each PCR amplification product was verified by agarose gel electrophoresis and imaged with ChemiDoc System (Bio-Rad) after incubation with ethidium bromide (Bio-Rad; 1610433). B2M was used as the housekeeping gene. Oligonucleotides were synthesized and purchased by Integrated DNA Technologies (IDT). Primer sequences are listed in Table S2.

Proteome profiler array

Cells were lysed with lysis buffer 17 (R&D Systems; 895943) supplemented with 10 μg/ml aprotinin, 10 μg/ml leupeptin, and 10 μg/ml pepstatin, whereas mouse cells were lysed with lysis buffer I (20mM Tris (pH7.5), 127mM NaCl, 10% Glycerol, 1% v/v NP40 (Igepal), 100mM NaF, 1mM Na₃VO₄) supplemented with 1mM PMSF, 10 μg/ml aprotinin, and 10 μg/ml leupeptin. Total protein levels were quantified with DC protein assay (Bio-Rad; 500–0112). 200 μg of total human protein samples or 300 μg of total mouse protein samples were analyzed respectively with a Human XL Cytokine Array Kit (R&D Systems; ARY022) or Mouse Angiogenesis Array kit (R&D Systems; ARY015) as recommended by the manufacturer. Membranes were exposed to X-ray films, which were imaged (digitized) with ChemiDoc System (Bio-Rad) and analyzed with Image Lab Software (Bio-Rad).

Western blot analysis

Cells were lysed with lysis buffer II (50mM Tris (pH8), 2% w/v SDS, 5mM EDTA, 3mM EGTA, 25mM NaF, 1mM Na₃VO₄) supplemented with 1mM PMSF, 10 μg/ml aprotinin, and 10 μg/ml leupeptin. Cell homogenates were sonicated for 2 s, clarified by centrifugation and total protein concentration was quantified with DC protein assay. Proteins samples were diluted to the appropriate dilution and mixed with 1/5 volume of 5x SDS-PAGE sample buffer (250mM Tris pH6.8, 10% w/v SDS, 25% v/v glycerol, 500mM DTT, and bromophenol blue). Proteins (30–45 μg per lane) were resolved by SDS-PAGE, electrotransferred to Immobilon-P membrane, and incubated with a primary antibody diluted as recommended by the manufacturer. Membranes were then probed with a horseradish peroxidase-conjugated secondary antibody (Promega V8051, W4011, and W4021; R&D HAF005) and protein signals were developed using the Pierce ECL western blotting substrate (Thermo Scientific; 32106) or the SuperSignal West Femto Maximum Sensitive Substrate (Thermo Scientific; 34095). X-ray films were imaged (digitized) with ChemiDoc System and analyzed with Image Lab Software. Primary antibodies were purchased from Santa Cruz biotechnology (p21, sc-6246; p53, sc-6243; GAPDH, sc-25778), Cell signaling (phospho-S6 Ser235/236, 2211, S6, 2317; phospho-p65, 3033; p65, 8242), and Sigma Aldrich (β-actin, A5441).

ELISA analysis

Conditioned media was collected, and cells were counted for later normalization. Separately, cells were lysed with lysis buffer I and total protein concentration quantified with DC protein assay. Concentrations of IL-1α, IL-1β, IL-6, and CCL2 present in cell lysate and conditioned medium were measured using Human IL-1α/IL-1F1 DuoSet ELISA (R&D Systems; DY200), Human IL-1β/IL-1F2 DuoSet ELISA (R&D Systems; DY201), Human IL-6 DuoSet ELISA (R&D Systems; DY206), Human CCL2/MCP-1 DuoSet ELISA (R&D Systems; DY279), and Mouse IL-6 ELISA MAX Deluxe (Biolegend; 431304). Protein concentrations are expressed as pg per mg total protein (for intracellular proteins) or pg per million cells (for secreted proteins).

Whole transcriptome expression analysis

Expression profiling was performed from the Genomics shared resource at Roswell Park Comprehensive Cancer Center. Initially, 500 ng total RNA was converted to cDNA, followed by in vitro transcription to generate biotin labeled cRNA using the Ambion Illumina TotalPrep RNA Amplification Kit (Ambion, Inc.) as per manufacturer’s instructions. 750 ng of the labeled probes were then mixed with hybridization reagents and hybridized overnight at 58°C to the HumanHT-12v4 Expression BeadChip (Illumina; BD-103–0204) for human samples or the MouseWG-6v2 Expression BeadChip (Illumina; BD-201–0202) for mouse samples. Following washing and staining with Cy3-streptavidin conjugate, the BeadChips were imaged using the Illumina iScan Reader to measure fluorescence intensity at each probe. The intensity of the signal corresponds to the quantity of the respective mRNA in the original sample. Analysis of the whole transcriptome expression was performed as follows: 1) The background corrected gene expression levels were extracted from BeadChip using Illumina’s Genome Studio (v2011.1) gene expression module (v1.9.0). The log2 transformed expression levels were quantile normalized using Lumi module in the R-based Bioconductor package. 2) For data quality control, we excluded the genes with detection p value greater than 0.05 (i.e., indistinguishable from the background noise) and genes that passed this filtering were used for differential expression analysis. The Limma program was used to calculate the level of differential gene expression, as previously described (Mastri et al., 2018). Briefly, a linear model was fit to the data and selected contrasts of condition (i.e., case versus control) were performed. A list of differentially expressed genes with p < 0.05 and ≥ 1.5 fold-change was obtained. Volcano plots and heatmaps were created using graphpad Prism version 6.07 and the JAVA based Morpheus program (https://software.broadinstitute.org/morpheus), respectively.

Gene set enrichment analysis

For comparative pathway and gene set correlations, gene set enrichment analysis (GSEA) (Subramanian et al., 2005) was applied to normalized microarray expression data in order to identify overrepresented predefined gene sets using gene sets for secretomes associated with senescence or induced by therapy (see Data S2). Gene-set permutation type with 1000 random permutations was run to obtain a rank gene list. Normalized enrichment score (NES) and nominal (NOM) p values, as defined by the GSEA method, were used to obtain weighted rank scores using the equation: −log(NOM p value) * NES.

Gene ontology analysis

Differentially expressed genes with gene products located in the extracellular region (GO:00005576) were identified using the gene ontology databases (Mastri et al., 2018). The identified genes were then analyzed for GO biological process terms and significant enriched terms were plotted based on p values as determined by the binomial statistic.

Identification of ATIS

A preliminary antiangiogenic therapy-induced secretome (ATIS) was identified by comparing secretomes of the SuR cells and SASP gene sets. ATIS was assembled by 25 genes that are represented in at least two SuR cell lines or in one SuR cell lines and SASP gene set from Coppé et al. (2010). Eight additional genes were added to create an enriched ATIS which were identified by PCR and protein profiler array based on genes and proteins associated with senescence. For a list of genes in identified gene sets see Data S3.

Survival analysis for ccRCC patients

Kaplan-Meier survival analysis for ccRCC patients are in whole or in part based on data generated by The Cancer Genome Atlas (TCGA) Research Network (https://cancergenome.nih.gov). Normalized RNA sequencing data downloaded from cBioportal (http://www.cbioportal.org/, TCGA Provisional; RNA-Seq V2) was used to stratify patients based on high or low tumoral expression levels of the gene of interest. For ATIS, patients were stratified based on the number of ATIS genes with high expression in tumors (more or less than 17 genes). Data were available for 525 out of 533 ccRCC patients.

Confirmation of ATIS signature enrichment in published GEO datasets

Previously published clinical and preclinical data derived from studies involving Su-treatment and tumor analysis were obtained from the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/).

Tumor models

Clinical studies

Whole transcriptome expression data were obtained from GEO with accession number GSE58837. In this study by Braga et al., patients with untreated locally advanced or operable breast cancer were treated with Su for 15 days and RNA samples from before and after treatment were used for Affymetrix microarray whole transcriptome expression analysis (Braga et al., 2017). GSEA analysis compared all pre-treatment samples to post-treatment samples. Treatment samples were stratified into responders/non-responders based on criteria as described by the authors (Braga et al., 2017).

Preclinical studies

Human and mouse specific microarray data derived from tumors were obtained from GEO with accession number GSE76068. In this study by Diaz-Montero et al. (2016), an RCC patient sample was implanted into immunocompromised mice subcutaneously and treated with 40mg/kg/day Su. Tumor samples were obtained at 3 time periods: 1) pre-treatment, 2) Su-treated (sensitive), and 3) Su-treated (resistant). Tumors samples were analyzed using Illumina human and mouse expression BeadChips to distinguish whole transcriptome expression data for tumor (human) and mouse (host) (Diaz-Montero et al., 2016).

Non-tumor models

Preclinical studies

Mouse specific microarray data derived from the lungs of non-tumor bearing mice were obtained from GEO with accession number GSE112466 (Mastri et al., 2018). In this study by Mastri et al., BALB/c mice were treated for 7 days with Ax (100mg/kg/day) or Su (60 or 120mg/kg/day). Tissue samples were analyzed using Illumina mouse expression BeadChips to evaluate whole transcriptome expression data changes in a tumor-independent model (Mastri et al., 2018).

QUANTIFICATION AND STATISTICAL ANALYSIS

Results were subjected to statistical analysis using the GraphPad Prism software package v.6.07 (GraphPad software Inc., San Diego, CA) and R v.3.3.2 through RStudio v.1.0.136 (Integrated Development for R; RStudio, Inc., Boston, MA URL https://www.rstudio.com/). Overall survival was summarized using the Kaplan-Meier method with the association between treatment group and survival evaluated using the two-sided log-rank test. Plots for combined pre- and post-surgical summary analysis (Figures 2A, 6B, and S2A) were conducted as described previously (Ebos et al., 2014). Briefly, this included: Pre-surgical primary tumor effects: Resected tumor or tumor-bearing kidney weights normalized to control animals. Post-surgical survival effects: Median survival based on Kaplan-Meier analysis. Results (except overall survival) are expressed as mean ± standard deviation (SD) or standard error of mean (SEM), as indicated. Comparisons between two groups were made with Student’s two-tailed unpaired t test, whereas one-way ANOVA was used for comparison of more than two groups. Comparisons for tumor volume and bioluminescence measurements were done for specified time point. A minimum significance level of 0.05 was used for all analyses.

DATA AND SOFTWARE AVAILABILITY

Whole transcriptome expression data

Complete microarray data were deposited in GEO (Accession numbers GSE122819, GSE122820, and GSE122821).

Supplementary Material

SUPPLEMENTAL INFORMATION

Supplemental Information includes six figures, two tables, and three data files and can be found with this article online at https://doi.org/10.1016/j.celrep.2018.12.017.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Human monoclonal anti-human IL-6	Pfizer (MTA)	PF-04236921
Rat monoclonal anti-mouse IL-6 (clone MP5-20F3)	BioXcell	Cat#BE0046; RRID: AB_1107709
Rabbit monoclonal anti-human Vimentin	Abcam	Cat#ab16700; RRID: AB_443435
Rat monoclonal anti-mouse PECAM-1 (clone SZ31)	Dianova	Cat#DIA-310; RRID: AB_2631039
Goat polyclonal anti-rabbit IgG; FITC conjugated	BD Biosciences	Cat#554020; RRID: AB_395212
Goat polyclonal anti-rat IgG; Cy3 conjugated	Invitrogen	Cat#A10522; RRID: AB_2534031
Mouse monoclonal anti-mouse p21 (F-5)	Santa Cruz Biotechnology	Cat#sc-6246; RRID: AB_628073
Rabbit polyclonal anti-human p53 (FL-393)	Santa Cruz Biotechnology	Cat#sc-6243; RRID: AB_653753
Rabbit polyclonal anti-human GAPDH (FL-335)	Santa Cruz Biotechnology	Cat#sc-25778; RRID: AB_10167668
Rabbit polyclonal anti-human phospho-S6 (Ser235/236)	Cell Signaling Technology	Cat#2211; RRID: AB_331679
Mouse monoclonal anti-human S6 (clone 54D2)	Cell Signaling Technology	Cat#2317; RRID: AB_2238583
Rabbit monoclonal anti-human phospho-p65 (Ser 536) (clone 93H1)	Cell Signaling Technology	Cat#3033; RRID: AB_331284
Rabbit monoclonal anti-human p65 (clone D14E12)	Cell Signaling Technology	Cat#8242; RRID: AB_10859369
Mouse monoclonal anti-β-actin (clone AC-15)	Sigma-Aldrich	Cat#A5441; RRID: AB_476744
Donkey anti-goat IgG; HRP conjugated	Promega	Cat#V8051; RRID: AB_430838
Goat anti-rabbit IgG (H+L); HRP conjugated	Promega	Cat#W4011; RRID: AB_430833
Goat anti-mouse IgG(H+L); HRP conjugated	Promega	Cat#W4021; RRID: AB_430834
Goat anti-rat IgG; HRP conjugated	R and D Systems	Cat#HAF005; RRID: AB_1512258
Chemicals, Peptides, and Recombinant Proteins
Sunitinib malate	Pfizer (MTA)	CAS: 341031-54-7
Axitinib	Pfizer (MTA)	CAS: 319460-85-0
Rapamycin	LC Labs	Cat#R-5000; CAS: 53123-88-9
Bafilomycin A1	Invivogen	Cat#88899-55-2; CAS: 88899-55-2
5-dodecanoylaminofluorescein di-β-D-galactopyranoside (C12FDG)	Setareh Biotech	Cat#7188; CAS: 138777-25-0
Critical Commercial Assays
Senescence β-galactosidase staining kit	Cell Signaling Technology	Cat#9860
Human XL Cytokine Array kit	R and D Systems	Cat#ARY022
Mouse Angiogenesis Array kit	R and D Systems	Cat#ARY015
Human IL-1α/IL-1F1 DuoSet ELISA	R and D Systems	Cat#DY200
Human IL-1β/IL-1F2 DuoSet ELISA	R and D Systems	Cat#DY201
Human IL-6 DuoSet ELISA	R and D Systems	Cat#DY206
Human CCL2/MCP-1 DuoSet ELISA	R and D Systems	Cat#DY279
Mouse IL-6 ELISA MAX Deluxe	Biolegend	Cat#431304
Deposited Data
Renal clear cell carcinoma (TCGA; KIRC; Provisional)	TCGA Research Network; cBioPortal for cancer genomics	https://cancergenome.nih.gov http://www.cbioportal.org
Expression data from breast cancer tumors treated with a single window of the antiangiogenic agent sunitinib	Braga et al., 2017	GEO: GSE58837
Gene expression changes during development of sunitinib resistance in renal cell carcinoma patient derived xenografts	Diaz-Montero et al., 2016	GEO: GSE76068
Tumor-independent host gene expression changes induced by tyrosine kinase inhibitors	Mastri et al., 2018	GEO: GSE112466
Gene expression changes in human breast cancer cells (LM2-4) after sunitinib resistance and long-term withdrawal	This paper	GEO: GSE122819
Gene expression changes in mouse fibroblasts (3T3) after sunitinib resistance, and short/long-term withdrawal	This paper	GEO: GSE122820
Gene expression changes in mouse kidney cancer cells (RENCA) after sunitinib resistance, and short/long-term withdrawal	This paper	GEO: GSE122821
Experimental Models: Cell Lines
LM2-4	Laboratory of R. Kerbel	N/A
LM2-4 SuR	This paper	N/A
LM2-4 SuR/LT-W	This paper	N/A
SN12-PM6	Laboratory of R. Kerbel	N/A
SN12-PM6-N	This paper	N/A
SN12-PM6-N SuR	This paper	N/A
SN12-PM6-N SuR/LT-W	This paper	N/A
MeWo	Laboratory of R. Kerbel	RRID: CVCL_0445
MeWo SuR	This paper	N/A
MeWo SuR/LT-W	This paper	N/A
4T1	Laboratory of A. Gudkov	RRID: CVCL_0125
4T1 SuR	This paper	N/A
4T1 SuR/LT-W	This paper	N/A
4T1 AxR	This paper	N/A
4T1 AxR/LT-W	This paper	N/A
RENCA	Laboratory of R. Pili	RRID: CVCL_2174
RENCA SuR	This paper	N/A
RENCA SuR/LT-W	This paper	N/A
RENCA AxR	This paper	N/A
RENCA AxR/LT-W	This paper	N/A
3T3	Laboratory of I. Gelman	RRID: CVCL_0594
3T3 SuR	This paper	N/A
3T3 SuR/LT-W	This paper	N/A
PY4.1	Laboratory of D. Dumont	RRID: CVCL_A336
PY4.1 SuR	This paper	N/A
PY4.1 SuR/LT-W	This paper	N/A
Experimental Models: Organisms/Strains
SCID mouse	Roswell Park Comprehensive Cancer Center; Laboratory animal shared resource (LASR)	N/A
Balb/cAnNCrl mouse	Charles River Labs	Strain code: 028; RRID: IMSR_CRL:547
Oligonucleotides
Primers, see Table S2	This paper; Integrated DNA Technology (IDT)	N/A
Software and Algorithms
Illumina Genome Studio v2011.1	Illumina	N/A
R v.3.3.2	The Comprehensive R Archive Network	N/A
RStudio v.1.0.136	Integrated Development for R	N/A
Lumi v.3.30.0		http://bioconductor.org/packages/release/bioc/html/lumi.html
Limma v.3.34.9		http://bioconductor.org/packages/release/bioc/html/limma.html
GraphPad Prism v6.07	GraphPad Software	N/A
JAVA based Morhpeus	Broad Institute	https://software.broadinstitute.org/morpheus/
GSEA	Broad Institute	http://software.broadinstitute.org/gsea/index.jsp
Other
HumanHT-12v4 Expression BeadChip	Illumina	Cat#BD-103-0204
MouseWG-6v2 Expression BeadChip	Illumina	Cat#BD-201-0202

Highlights.

• Stopping antiangiogenic treatment after resistance leads to rebound tumor growth

• Antiangiogenic drug resistance can induce transient pseudosenescent secretory phenotypes

• Targeting SASP regulators like IL-6 and mTOR blunt withdrawal-mediated tumor growth