Long-term breast cancer response to CDK4/6 inhibition defined by TP53-mediated geroconversion

SUMMARY

Inhibition of CDK4/6 kinases has led to improved outcomes in breast cancer. Nevertheless, only a minority of patients experience long-term disease control. Using a large, clinically annotated cohort of patients with metastatic hormone receptor-positive (HR+) breast cancer, we identify TP53 loss (27.6%) and MDM2 amplification (6.4%) to be associated with lack of long-term disease control. Human breast cancer models reveal that p53 loss does not alter CDK4/6 activity or G1 blockade but instead promotes drug-insensitive p130 phosphorylation by CDK2. The persistence of phospho-p130 prevents DREAM complex assembly, enabling cell-cycle re-entry and tumor progression. Inhibitors of CDK2 can overcome p53 loss, leading to geroconversion and manifestation of senescence phenotypes. Complete inhibition of both CDK4/6 and CDK2 kinases appears to be necessary to facilitate long-term response across genomically diverse HR+ breast cancers.

Graphical abstract

In brief

Kudo et al. demonstrate that TP53 loss in HR+ breast cancer is clinically associated with long-term failure of CDK4/6i due to cellular escape from quiescence via impaired DREAM complex assembly. Selective CDK2 inhibition is found to reinforce p130 dephosphorylation in p53 mutant models and facilitate durable tumor growth suppression.

INTRODUCTION

Inhibition of the G1 checkpoint is an integral requirement for numerous oncologic therapies and has been most effectively achieved in hormone receptor-positive (HR+) breast cancer through selective CDK4/6 kinase inhibitors (CDK4/6i).^1,2 Cells held in G1 by CDK4/6 inhibition can undergo a variety of fates, including apoptosis, senescence, or cell-cycle re-entry.^3,4 Of these, cell-cycle re-entry is a major limitation to drug efficacy as it can restore proliferative capacity and allow mutagenesis that can further enforce resistance. By contrast, eliciting cell states that prevent cell-cycle re-entry can promote tumor clearance and long-term disease control.⁵

In HR+ breast cancer, CDK4/6i are standard of care, promoting relapse-free and overall survival.^6–12 However, there is high variance in the duration of benefit from these therapies ranging from a few months to over 5 years.^13–15 The initial effect of CDK4/6i is to cause a decline in cell-cycle progression factors, closely corresponding to their induction of G1 arrest in laboratory models.^16,17 However, a rebound in Ki67 levels is observed during prolonged therapy, suggesting that some tumors can escape drug-mediated G1 arrest.^18,19 We hypothesized that this capacity for cell-cycle re-entry might cause the lack of long-term disease control.

Herein, we sought to determine the genomic configurations and underlying mechanisms associated with long-term disease control among patients treated with CDK4/6i and thereby identify pharmacologic approaches to elicit this effect in patients.

RESULTS

Somatic variants enriched in cancers failing to have long-term response to CDK4/6i

To identify the genomic patterns associated with clinical outcomes, we analyzed a cohort of 467 patients with metastatic HR+ and human epidermal growth factor receptor 2 (HER2)-negative (HR+/HER2−) breast cancer treated at Memorial Sloan Kettering Cancer Center (MSK) with first-line CDK4/6i (abemaciclib, palbociclib, and ribociclib) and endocrine therapy (Table S1). Tumor-normal sequencing of up to 510 cancer-associated genes was performed via the MSK-IMPACT assay.²⁰ For each gene, we defined oncogenic somatic mutations, structural variants, and copy-number alterations using OncoKB.²¹

First, we sought to identify the alterations in genes associated with longer, intermediate, and short response. This was accomplished by dividing patient time on first-line CDK4/6i into tertiles (n = 156 patients). Patients who discontinued therapy for reasons other than disease progression were excluded from the analysis. We iteratively assessed the enrichment of oncogenic variants in each gene with a Firth-penalized logistic regression, comparing cases with longer response to (i) intermediate response and (ii) short response. The model was adjusted for de novo metastatic status at diagnosis and endocrine therapy partner and corrected for multiple hypothesis testing.

Oncogenic alterations in PPM1D (odds ratio [OR]: 4.47, 95% confidence interval [CI]: 1.69–11.83, p = 0.002, q = 0.04) and TP53 (OR: 2.52, 95% CI: 1.42–4.51, p = 0.0071, q = 0.039) were significantly enriched in the intermediate, compared to long-term response group (Figure 1A). In patients who experienced a short response, TP53 (OR: 5.54, 95% CI: 2.95–10.38, p < 0.001, q < 0.001), MDM2 (OR: 3.35, 95% CI: 1.03–10.89, p = 0.046, q = 0.28), MYC (OR: 3.00, 95% CI: 1.29–6.99, p = 0.015, q = 0.11) and PTEN (OR: 2.84, 95% CI: 1.10–7.33, p = 0.003, q = 0.24) mutations were enriched. TP53 alterations remained significant after adjustment for multiple-hypothesis testing (Figure 1B).

Figure 1. Somatic variants enriched in patients failing to achieve long-term response to CDK4/6i.

(A and B) The association of individual genes with long term disease control (progression free survival, PFS) as compared to (A) inter-mediate and (B) short-term disease control with CDK4/6i and endocrine therapy (ET), based on even tertiles of treatment duration. The colors indicate statistical significance (q < 0.1) and the circles’ size reflects the frequency of alteration in the cohort. All q values are calculated based on the Benjamini and Hochberg method correction of log rank p values.

(C) The results of an elastic net Cox proportional hazard model, whereby risk groups are stratified based on K-means clustering. Here, the optimal number of clusters (n = 3) was selected automatically by employing an Akaike information criterion and again separated patients into short, intermediate, and prolonged time on treatment.

(D–F) The variable importance of the elastic-net Cox proportional hazard model, after 50 runs and 5-fold cross-validation. Selection frequency is defined as the proportion of runs in which each gene achieves a coefficient greater than zero, and the mean hazard ratio is the mean coefficient across all runs. (E and F) The PFS of patients receiving first-line CDK4/6i and ET, harboring tumors with pre-treatment functional alterations in TP53 (E) and MDM2 (F).

(G) The PFS of patients receiving first and second-line CDK4/6i and ET in the Tempus cohort. Survival analyses were performed with univariate and multivariate Cox proportional hazard models, adjusted by treatment class (CDK4/6i + AI or CDK4/6i + SERD).

See also Tables S1–S3.

To elucidate the clinical impact of each gene in the broader context of the clinicogenomic feature space, we implemented an elastic net Cox regression on binary oncogenic variant status of each gene and selected clinical features. Our model identified a “longer response” group (n = 117, 25.1%) from patients with a median progression-free survival (PFS) of 31.3 months (95% CI 24.4–49.3), compared with an “intermediate” (n = 223, 47.8%, median PFS = 15.2 months, 95% CI 12.3–17.7) and “short response” group (n = 127, 27.2%, median PFS = 7.8 months, (95% CI 5.3–9.7) (Figure 1C). The hazard ratio (HR) of the “intermediate” group relative to “longer response” group was 1.68 (95% CI, 1.23–2.24, p = 0.001), while the respective HR of the “short response” group was 3.67 (95% CI, 2.66–5.07, p < 0.001). TP53 and MDM2 were among the most important somatic genes to stratify between these groups, obtaining variable selection frequencies of 1.0 and 0.90 and mean hazard ratios (HR) of 2.10 ± 0.023 (SEM) and 1.40 ± 0.024 (SEM), respectively (Figure 1D).

Of these 467 patients, 129 (27.6%) had pre-CDK4/6i loss-of-function variants in TP53, corresponding to short PFS (median PFS: 8.7 months, 95% CI: 6.6–10.3 months, HR: 2.12, 95% CI: 1.68–2.68; p = 2.64e-10; Figure 1E). Similarly, MDM2 amplifications were present in 30 patients (6.4%) and corresponded to short PFS (median PFS: 10.3 months, 95% CI: 4.8–17.5 months, HR: 1.74, 95% CI: 1.17–2.59; p = 0.006; Figure 1F).

Given the ubiquity of TP53 alterations in ER + MBC, we sought to verify its association in independent clinical cohorts. First, we interrogated an international, real-world dataset derived from both academic and community practices (“Tempus cohort”). Of the 2,820 patients treated with either first or second line CDK4/6i and ET, 766 (27.1%) had pre-treatment loss-of-function variants in TP53, which conferred shorter PFS (median PFS: 11.2 vs. 18.1 months, HR: 1.44, 95% CI: 1.28 to 1.62, p = 1.6e-9; Figure 1G) (Table S2). Next, we reviewed available clinicogenomic data from two large, randomized clinical trials, including a pooled analysis of ribociclib and ET in the metastatic setting (MONALEESA pooled trials)²² as well as a phase 3 registrational clinical trial of adjuvant abemaciclib (MONARCH-E)²³ (Table S3). Progression free survival (PFS) was shorter in cases with TP53 loss-of-function variants identified in the MONALEESA pooled trials (8.6 months, 95% CI 6.6 to 12.7) compared to WT (21.3 months, 19.1 to 24.6). In the MONARCH-E analysis, patients with baseline TP53 loss-of-function (LOF) variants were more likely to experience a recurrence (event rate of 29.1% compared to 17.3%). Taken together, these clinical outcomes amassed from a total of 4,457 patients, demonstrate that TP53 loss-of-function variants portend poorer outcomes across a variety of CDK4/6i agents and treatment contexts.

p53 loss promotes long-term cell outgrowth

The stability of p53 is regulated by MDM2, and overexpressed MDM2 represents a common mechanism of cancer cell inactivation of p53.²⁴ As TP53 LOF mutations and MDM2 amplifications promote inactivation of p53, we asked whether and how p53 loss alters CDK4/6i response using breast cancer cell lines. We assessed the effect of CDK4/6i treatment on phosphorylation of CDK4/6 substrate Rb1 in a CDK4/6i-sensitive, TP53-wild type (WT) model (MCF7 parental) compared to those lacking p53 (p53KO) or overexpressing MDM2 (Figures 2A and S1A). We found that 100 nM abemaciclib treatment for 24 h potently suppressed both p-Rb and levels of its effector E2F in WT, p53KO, and MDM2OE cells. However, this effect was not observed in cells resistant to CDK4/6i due to high CDK6 expression via inactivation of FAT1 (FAT1KO). Accordingly, MCF7 parental, p53KO, and MDM2OE cells accumulated in G1 (>80%) while FAT1-loss cells did not (Figure 2B). Moreover, the drug concentration required to inhibit proliferation was nearly identical for MCF7 parental (IC50 = 9.5 nM), p53KO (13.9 nM), and MDM2OE (14.9 nM) cells (Figure 2C). Similarly, in the CDK4/6i-sensitive cell line HCC1500, p53KO cells showed loss of Rb phosphorylation and G1 arrest comparable to parental cells (Figures S1B and S1C). Previous work has established that HR+ breast cancer cells blocked in G1 can potentially remain effectively inhibited by CDK4/6i for many weeks.²⁵ To investigate whether mutant TP53 modulates this effect, we subjected parental and p53-deficient cells to prolonged drug treatment for over 5 weeks. While the growth of parental cells remained inhibited at 5 weeks, both p53KO and MDM2OE cells resumed proliferation between 4 and 5 weeks, revealing a potential avenue for drug resistance (Figures 2D, S1D, and S1E). Similarly, the growth of p53KO xenograft tumors was unaffected by CDK4/6i treatment after 5 weeks, while the growth of WT tumors remained suppressed (Figure 2E). Moreover, the addition of the antiestrogen fulvestrant to abemaciclib did not prevent the outgrowth of the p53KO cells (Figures S1F and S1G). To further ascertain the sufficiency of TP53 alterations in mediating these effects, we re-introduced WT or TP53 loss-of-function mutations into p53KO cells and investigated their response to CDK4/6i. Similar to parental MCF7 and p53KO models, cells expressing WT, R273H or R280K TP53 all incurred decreases in p-Rb and E2F1 levels by abemaciclib (Figure 2F) and entered G1 arrest after 24 h of CDK4/6i (Figure 2G). The 5-day IC50 values were nearly equivalent across these three cell lines (WT = 29 nM, R273H = 17.3 nM, R280K = 33.9 nM; Figure S1H). However, with long-term CDK4/6i, the growth of cells with WT TP53 remained completely inhibited, whereas those with R273H or R280K TP53 mutants showed outgrowth (Figure 2H). Consistent with the in vitro results, isogenic xenograft tumors harboring pathogenic TP53 mutations developed CDK4/6i resistance while their WT TP53 counterparts remained sensitive (Figure 2I). In accordance, Ki67 levels were markedly reduced in WT treated tumors but not in R280K mutants (Figure S1I). Finally, to assess these phenotypes in a patient-derived model, we established organoids from a patient-derived tumor with HR+/HER2− and WT-TP53 (PDO #6), which proved sensitive to CDK4/6i (Figures 2J and S1J–S1L). Compared to non-targeted sgRNA organoids (NTsgRNA), p53 knockout organoids (p53sgRNA) were resistant to continuous treatment with abemaciclib alone or the combination of abemaciclib with fulvestrant (Figures 2K and S1M). These findings demonstrate that intact p53 is necessary for long-term growth suppression by CDK4/6i in HR+ breast cancer.

Figure 2. p53 loss promotes long-term cell outgrowth.

(A) Immunoblotting of indicated proteins in MCF7 parental, p53KO, MDM2OE, and FAT1KO cells treated for 24 h.

(B) Cell-cycle distribution after 24 h. Student’s t test was performed to compare G1 fraction. The mean value of three replicates (top). Changes in the proportion of cells in G1 phase. Student’s t test (bottom). Data are means ± SD of three biological replicates.

(C) IC50s of abemaciclib on day 4.

(D) Cell viability assay with 50 nM abemaciclib. Data are means ± SEM of four replicates. Two-way ANOVA, Tukey’s.

(E) Relative tumor volume at 5 weeks compared to that on day 0 after treatment with vehicle or ribociclib (25 mg/kg) (n = 5). Data are represented as means of five replicates ±SEM. Student’s t test.

(F) Immunoblotting of indicated proteins in V5-WT, V5-R273H, and V5-R280K treated for 24 h.

(G) Cell-cycle distribution after 24 h. Data are means of three replicates. Student’s t test.

(H) Cell viability with 50 nM abemaciclib. Data are means ± SEM of four replicates. Two-way ANOVA, Tukey’s test.

(I) Tumor volume ratio of V5-WT or V5-R280K cells xenograft tumors treated with 25 mg/kg ribociclib for 29 days compared to day 0. Data are means ± SD of four replicates. Two-tailed unpaired t test.

(J) Patient-derived organoids PDO #6 were originally derived from patient tumors with HR+/HER− and WT-TP53. Scale bar: 400 μM.

(K) Representative images of PDO #6 NT sgRNA or p53sgRNA treated with DMSO, 12.5 nM abemaciclib and combination of abemaciclib and 10 nM fulvestrant on indicated days. Scale bar: 400 μM.

See also Figure S1.

p53 loss enables cell-cycle re-entry and prevents geroconversion

Akin to non-transformed breast epithelial cells in prolonged G1 arrest, HR+ breast cancer cells have a limited capacity to re-enter the cell cycle and can remain arrested even upon drug withdrawal, where they are subject to the execution of a senescence program that remodels chromatin and gene expression.^5,16,26 We assessed if these “downstream” phenotypes might be absent and thus contribute to the outgrowth of CDK4/6i-treated p53KO and MDM2OE cells. We examined the effect of drug withdrawal by culturing cells in 50 nM abemaciclib for 7 days followed by drug washout. Whereas parental cells showed continuous growth inhibition after washout, the p53KO and MDM2OE cultures resumed proliferation and were proficient in colony formation assays compared to MCF7 cells (Figures 3A and 3B). Similarly, TP53 mutant cells retained the ability to generate colonies after drug washout (Figure S2A). To further ascertain the timing of this effect, we performed fluorescence ubiquitin cell cycle indicator (FUCCI) time-lapse imaging to quantify cell-cycle re-entry over time following CDK4/6i washout.²⁷ Cells were labeled by a red fluorescence emitting protein during G1 (Figure S2B). More than 90% of MCF7 cells and p53KO cells were present in G1 after 6 days of treatment. Upon drug washout, parental cells failed to enter S phase over a 75-h time course, while p53KO cells began accumulating in S phase after drug washout (Figures 3C and S2C and Video S1). In addition to cell-cycle re-entry, we examined the cells for features of entry into senescence, including expression of senescence-associated-beta-galactactosidase (SA-β-Gal),²⁸ accumulation of senescence-associated heterochromatin foci (SAHF), and expression of senescence-associated transcripts.^29–31 After 6 days of abemaciclib treatment, ~80% of parental MCF7 cells were positive for SA-β-Gal, as compared to 6%, 20%, and 4% in p53KO, MDM2OE, or FAT1KO cells, respectively (Figures 3D and S2D). Induction of WT but not mutant TP53 restored SA-β-Gal activity to levels observed in MCF7 parental cells (Figure 3E). These findings were also observed in HCC1500 parental and p53KO cells (Figures S2E and S2F). In further confirmation, we utilized pharmacologic inhibition of MDM2 by Nutlin-3a in combination with CDK4/6i and found that this increased p53 expression and SA-β-Gal activity and decreased long-term cell proliferation (Figures S2G–S2I). Moreover, after 14 days of abemaciclib treatment, SA-β-Gal expression was potently induced in NTsgRNA organoids but not p53sgRNA organoids (Figure 3F). Additionally, combined anti-estrogen therapy with CDK4/6i did not increase SA-β-Gal activity in the p53KO cells to the level of parental cells (Figures S1M and S2J). As an earlier measure of progression to senescence, we assessed facultative heterochromatin formation by measuring HP1γ foci by immunofluorescence. We found accumulation of HP1γ foci in CDK4/6i treated parental cells but not in p53KO, MDM2OE, or FAT1KO cells (Figures 3G and 3H). HP1γ foci were induced by re-introduction of WT TP53 (Figures 3I and S2K). Finally, we sought to identify the senescence-associated genes which were up-regulated by CDK4/6i-induced cell-cycle arrest in HR+ breast cancer cells to ascertain if p53KO suppressed their induction. Given the cell-type specificity of senescence-associated transcripts, we developed a NanoString nCounter expression panel (n = 271 genes) based on prior databases (Figure S3A).^29–32 Comparison between senescent-proficient (MCF7 parental) and senescent-deficient (MCF7-FAT1KO; MCF7-CDK6 high) cells nominated IGFBP5, CAV1, and GEM as strongly associated with senescence features in vitro and in vivo (Figures S3B–S3D).³³ These genes were increased in other CDK4/6i sensitive models, HCC1500 cells, and PDX BC6 tumors (Figures S3E–S3G). In comparing MCF7 parental and p53KO treated with CDK4/6i for 21 days, we identified robust increases in CAV1, IGFBP5, and GEM in parental cells but not in p53KO cells (Figures S3H and S3I). In further support, senescence-associated genes and SA-β-Gal staining were not increased in p53KO xenograft tumors compared to MCF7 parental tumors (Figures 3J and 3K).

Figure 3. p53 loss enables cells re-enter cell cycle and prevent geroconversion.

(A) Cells were treated with 50 nM abemaciclib for seven days. Cell viability was measured with continuous drug treatment (red, blue, and green) or after drug withdrawal (pink, light blue, and light green). Data are means ± SEM of four replicates. Linear regression analysis.

(B) Colony formation assay. Cells were treated for 10 days, re-seeded and cultured without drug.

(C) MCF7 and p53KO cell expressing IncuCyte-Cell-Cycle-Green/Red-Lentivirus were treated with 100 nM abemaciclib for six days. After withdrawal of drug, cells were subjected to time-lapse fluorescence microscopy. Representative images at 27 h after drug withdrawal. Scale bar: 100μM (top). Time course change of each cell-cycle phase proportion. Data were analyzed using IncuCyte starting after drug withdrawal (bottom).

(D) The histogram shows counts of SA-β-Gal-positive cells under 50 nM abemaciclib treatment for six days (top). Bars represent the percentage of SA-β-Gal positive cells (bottom). Data are means ± SD of three replicates. Two-way ANOVA, Sidak’s.

(E) Bars show the percentage of SA-β-Gal positive cells on day 5. Data are means ± SD of three replicates. Two-way ANOVA, Sidak’s.

(F) Bars represent the percentage of SA-β-Gal positive cells of each PDO #6 after 16 days. Data are means ± SD of three replicates. Two-way ANOVA, Sidak’s.

(G) Representative images of each cell line treated with 100 nM abemaciclib for eight days and the colocalization of HP1γ foci and DAPI (yellow arrows). Scale bar: 5 μM (left), 25 μM (right).

(H) The percentage of HP1γ foci-positive cells was quantified in three different fields. Data are represented as means ± SD. Two-way ANOVA, Sidak’s.

(I) The percentage of HP1γ foci-positive cells treated for 8 days was quantified in three different fields. Data are represented as represented as mean ± SD. Two-way ANOVA, Sidak’s.

(J) qPCR of indicated genes expression in xenograft tumors with vehicle or 25 mg/kg ribociclib after 5 weeks. The ratios are represented as means ± SD from two tumors. n = 6. Student’s t test.

(K) Representative images of SA-β-Gal staining of xenograft tumors in J. Scale bar: 100 μM. n = 5.

See also Figures S2 and S3 and Video S1.

Taken together, these data point to a halt in progression from G1 arrest to senescence in p53 loss cells treated with CDK4/6i, facilitating their re-entry into the cell cycle.

Loss of p53 suppresses DREAM complex assembly

To identify pathways related to the effects of p53 loss upon breast cancer response to CDK4/6i, we analyzed treated parental and p53KO cells over time by RNA sequencing. MCF7 parental and p53KO cells were treated by 50 nM abemaciclib for 3 days, 14 days, and 21 days. Consistent with the differences in phenotypic responses over time, multiple gene sets were different between these groups at later time points. To isolate differences that might account for the cell-cycle re-entry phenotypes, we focused on targets of E2F and the DREAM complex³⁴ (Figure 4A). The DREAM complex is comprised of p130/p107, E2F4, DP1, and a stable core complex of MuvB-like proteins and serves to repress cell-cycle dependent gene expression during quiescence, thereby maintaining quiescence.^34–37 We examined the effect of p53KO on DREAM complex components after CDK4/6i treatment and found p130 phosphorylation to be discordant between parental and p53KO cells (Figure 4B). The phosphorylation dependent interaction of p130 with E2F4 is central to DREAM complex suppression of cell cycle, suggesting the potential significance of this difference in mediating outcome differences.^37–39 For further confirmation, the association between E2F4 and p130 was examined by immunoprecipitation upon abemaciclib treatment and found to be induced in parental but not p53KO cells (Figure 4C). To corroborate these differences, we examined the potential for E2F4 to bind DREAM complex sites in WT and p53KO models. Using chromatin immunoprecipitation, we found MYBL1 and HMGB3⁴⁰ avidly bound to E2F4 upon CDK4/6i treatment only in parental and not p53KO cells (Figure 4D).

Figure 4. Loss of p53 suppresses DREAM complex assembly.

(A) GESA plot with normalized enrichment score (NES) of REACTOME gene sets. NES was analyzed between MCF7 and p53KO treated with 50 nM abemaciclib at day21. REACTOME-E2F-mediated-regulation-of-DNA-replication (top), REACTOME-Transcription-of-E2F-targets-under-negative-control-by-DREAM-complex (middle) and REACTOME-Transcription-of-E2F-targets-under-negative-control-byp107 RBL1-and-p130 RBL2-in-complex-with-HDAC1 (bottom).

(B) Immunoblotting of DREAM complex components with 50 nM abemaciclib.

(C) Cells were treated with 100 nM abemaciclib for 48 h. Lysates were immunoprecipitated with p130 and IgG antibodies.

(D) Cells were treated with 7 days of abamacilib and tested by ChIP followed by real-time PCR. Bar showed technical duplicate. Unpaired, Student’s t test.

(E) Western blot results of MCF7 cells and p53KO transduced with doxycycline (dox)-inducible HA-tagged p21. The cells were treated with 50 nM abemaciclib or DMSO +/− dox for 48 h.

(F) Dox-inducible HA-p21 MCF7 and p53KO cells were treated with 100 nM abemaciclib and +/− dox for 72 h. Lysates were immunoprecipitated with E2F4 and IgG antibodies.

(G) Time course change of the proportion of G1 phase after drug withdrawal. FUCCI labeled MCF7 were treated with 50 nM abemaciclib and FUCCI labeled p53KO cells were treated with or without dox and with 50 nM abemaciclib for 4 days. The cells were cultured without abemaciclib for 62 h under time-lapse imaging. Two-way ANOVA, Tukey’s.

(H) Cell viability treated with 50 nM abemaciclib or DMSO +/− dox. Data are means ± SEM of four replicates. Two-way ANOVA, Tukey’s.

(I) Flow cytometry analysis showing the percentage of SA-β-Gal positive cells. Data are means ± SD of three replicates. Two-way ANOVA, Sidak’s.

(J) Colony formation assay. Cells were treated with 50 nM abemaciclib ± dox for 11 days and reseeded without drug.

(K) Cell viability of p130KO in dox inducible HA-p21 cells. The cells were treated with 50 nM abemaciclib or DMSO +/− dox. Data are means ± SEM of six replicates. Two-way ANOVA, Tukey’s.

See also Figure S4 and Video S2.

Given the known roles of p53 in regulating p21 expression and the potential for p21 to suppress CDK2 dependent p130 phosphorylation,^40,41 we engineered a doxycycline-inducible p21 into parental MCF7 and p53KO cells to assess for rescue (Figure 4E). The phosphorylation of p130 persisted in p53KO cells treated with abemaciclib, but was reduced upon induction of p21, resembling the pattern of parental cells. Moreover, under stable overexpression of p21, p53KO cells treated with CDK4/6i showed a restored interaction of E2F4 with p130 (Figure 4F), sustained G1 arrest after drug withdrawal (Figure 4G, Video S2), sustained suppression of proliferation (Figure 4H), hallmarks of senescence (Figure 4I), and reduced colony formation (Figure 4J).

Together, these data imply a critical role for p130 and the DREAM complex in mediating the effects of CDK4/6i in p53 WT cells. To further test the centrality of p130, we generated p130 KO cells and compared these to parental cells and found loss of p130 was sufficient to enable long-term outgrowth from CDK4/6i (Figures 4K and S4A). Moreover, fewer senescent cells and more colonies were observed in drug treated p130KOs compared to control non-target gRNA (NT sgRNA) (Figures S4B and S4C). These results suggested that loss of DREAM complex is sufficient to block geroconversion. Given the additional functions of p21 in tumor suppression beyond its effects on CDK2-p130, we assessed whether p21 expression was sufficient to facilitate long term growth suppression in the absence of p130 (Figure S4D). Consistent with its other roles, induction of p21 with CDK4/6 inhibition did support stronger initial growth suppression with or without drug treatment. However, even with overexpression of p21, cells with p130KO eventually escaped growth inhibition upon prolonged CDK4/6i treatment, with kinetics similar to those observed in p53KO cells (Figure 4K). Moreover, the induction of p21 was unable to drive CDK4/6i-treated p130KO cells into senescence (Figures S4E and S4F). Collectively, these results reveal the centrality of the p130 DREAM complex in mediating long-term growth suppression and driving cells into senescence from CDK4/6i.

Combined inhibition of CDK2 and CDK4/6 enables long-term growth suppression

As CDK4/6i leads to effective down-regulation of CDK2 function in p53 WT cells in part through p21 redistribution from CDK4 to CDK2, and p130 can be phosphorylated by CDK2,^41–43 we hypothesized that combined CDK2 and CDK4/6 inhibition could suppress p130 phosphorylation in p53KO cells. Here, we utilized a highly selective and potent CDK2 inhibitor (CDK2i), AZD8421.⁴⁴ Whereas CDK4/6 inhibition alone was sufficient to inhibit both Rb and p130 phosphorylation in parental cells, p53KO cells required the addition of the CDK2i to effectively down-regulate p130 (Figure 5A). Moreover, combined CDK4/6i and CDK2i could block cell-cycle re-entry upon drug withdrawal (Figure 5B). Further, combined inhibition of CDK2 with CDK4/6 enabled long-term growth suppression of p53KO cells (Figures 5C and 5D) and was associated with potent induction of SA-β-Gal expression (Figure 5E).

Figure 5. Combined inhibition of CDK2 and CDK4/6 enables long-term growth suppression.

(A) Immunoblotting of indicated proteins in cells with DMSO, 50 nM abemaciclib, 500 nM AZD8421, or combination for 72 h.

(B) MCF7 and p53KO cells with FUCCI were treated with 50 nM abemaciclib +/− 500 nM AZD8421 for 8 days. Time course change of G1 phase proportion starting after drug withdrawal. Two-tailed, unpaired t test.

(C) Cell viability of the cells treated with drug. Data are means ± SEM of four replicates. Two-way ANOVA, Tukey’s.

(D) Colony formation assay. Cells were treated for 10 days, re-seeded and cultured without drug.

(E) Bar shows the percentage of SA-β-positive cells on day 5. Data are means ± SD of three replicates. Two-way ANOVA, Sidak’s.

(F) Immunoblotting of p53 mutant BC primary cell line #5 treated with drugs for 48h.

(G) Cell viability of PD BC line #5 treated with drugs. Data are means ± SEM of four replicates. Two-way ANOVA, Sidak’s.

(H) The cells were treated with DMSO, 500 nM AZD8421, 50 nM abemaciclib, or combination for 6 days. Confluency of PD BC line #5 was analyzed by Incucyte for 13 days after drug washout. Data are represented as means of duplicates ± SD. Two-way ANOVA, Tukey’s.

(I) Bars indicate the percentage of SA-β-Gal-positive cells in PD BC line #5 cells. Data are means ± SD of three replicates. Two-way ANOVA, Sidak’s.

(J) Relative mRNA expression of MKI67. The ratios are represented as means ± SD from two tumors. n = 6. Student’s t test.

(K) Immunoblotting of indicated antibodies. Lysate was extracted from PDX #5 tumors treated with drugs for 5 days.

See also Figures S5 and S6.

To establish the generality of these effects, we tested additional CDK2 inhibitors (ebvaciclib alone, CDK2/4/6 inhibitor; PF-07104091 and SNS032, CDK2 inhibitors; combined with abemaciclib) as well as siRNA against CDK2 in wild-type TP53 or knockout p53 cells (Figures S5A–S5G and S6A–S6I).^45,46 In all cases, combining CDK2i with CDK4/6i in p53KO cells enabled stronger suppression of phosphorylation of p130 (Figures S5A, S6A, and S6E), stronger suppression of the cancer cell outgrowth during long term treatment (Figures S5B, S6B, and S6D) and increased induction of SA-β-Gal positive cells (Figures S5C, S6C, and S6F). Whereas p53KO cells resumed proliferation and were proficient in colony formation assays compared to MCF7 cell only by CDK4/6i, the combination led to continuous growth inhibition after drug washout (Figures S5D and S5E). The observed results were not restricted to abemaciclib as the CDK4/6i partner as similar findings were observed when CDK2i was given with ribociclib (Figures S5H–S5J).⁴⁷

To extend the relevance of these findings, we tested the combination in patient derived models. The HR+/HER2− patient-derived organoid model, p53 sgRNA PDO #6 showed complete growth inhibition to the combination of abemaciclib and a CDK2i, while manifesting resistance to abemaciclib alone (Figures S5F and S5G). Similar effects were observed in an HR+/HER2− p53 mutant primary breast cancer cell line, PD BC line #5, as combined CDK4/6 and CDK2 inhibition suppressed the phosphorylation of both Rb and p130 (Figures 5F and S6G), long-term growth (Figure 5G), growth after withdrawal of the drug (Figure 5H), and induced SA-β-Gal activity (Figures 5I, S6H, and S6I). Induction of wild-type p53 in this model enabled long-term and complete growth suppression upon abemaciclib treatment, further corroborating the particular role of p53 in this aspect of drug response (Figures S6J–S6L). The effects observed in terms of suppression of markers of proliferation through combined CDK2i with CDK4/6i in p53KO were further manifest in vivo in this model with short term treatment with AZD8421 combined with ribociclib causing marked reduction of MKI67, p130 phosphorylation, E2F1, and cyclin A2 levels compared to the single agents (Figures 5J and 5K). Together, these data reveal that p53KO cells poorly undergo geroconversion in response to CDK4/6i and that this response is significantly embellished through the addition of a CDK2i.

p53 loss is associated with cell-cycle re-entry in patients treated with neoadjuvant CDK4/6i

To assess these findings in human breast cancer samples, we examined the impact of TP53 status upon patients treated with neoadjuvant ribociclib in the FELINE phase 2 clinical trial (NCT02712723). In this trial, two cohorts of patients with early-stage HR+/HER2− breast cancer were treated with a combination of ribociclib and letrozole (an aromatase inhibitor) evaluated at 14 days and 180 days on therapy for changes in Ki67 (Figure 6A).^19,48 Change in Ki67 (dichotomized at 10%) on neoadjuvant endocrine therapy has been associated with improved time to recurrence in prior neoadjuvant endocrine trials.^49,50

Figure 6. p53 loss is associated with cell-cycle re-entry in patients treated with neoadjuvant CDK4/6i.

(A) Sample collection in FELINE trial. Core needle biopsy was performed over the course of treatment: screening (day 0), cycle 1 (day 14), and end of treatment (day 180).

(B) Ki-67 trend is depicted for all patients in the FELINE trial for whom Ki-67 was evaluable at all time points and sequencing of the day 0 tumor was performed. The rates of Ki-67 > 10% at surgery are compared by pre-treatment TP53 status utilizing Fisher’s exact (two-sided) test.

(C) Proliferation score in wild type and mutant TP53 tumors at baseline, on-treatment day14 and EOT.

(D and F) GSEA of DREAM complex genes in WT compared vs. MT at day14 (left) and EOT (right).

(E) GSEA of senescence ribo genes. EOT vs. baseline in WT TP53 (left) and in MT (right) (F) Model for loss of p53 resistant mechanism. In the presence of p53, CDK4/6i inhibited phosphorylation of RB and p130. The cells were arrested in quiescence through DREAM complex, leading to irreversible cell-cycle arrest. Loss-of-function p53 and insufficient p21 enabled CDK2 to retain phosphorylated p130, leading to cell-cycle re-entry and escape from quiescence.

See also Figure S6.

Whole-genome sequencing (WGS) was performed on baseline samples from the ribociclib containing arms of the study (Figure S6M, CONSORT diagram). The characteristics of Ki67% at surgery for this subset (median 0.71, IQR [0.1–6.33]) was similar to that of 70 evaluable cases in the original cohort (median 0.45, IQR [0.11–3.78], Wilcoxon rank-sum test p value = 0.63). Among the 32 patients, 10 (31.3%) harbored a pre-treatment TP53 loss of function variant (Figure S6M). Of these 10 cases, 6 (60%) did not achieve a low (<10%) Ki67 upon surgical resection as compared to those with TP53 WT tumors (n = 22) where only one (4.5%) did not achieve low Ki67 (OR = 26.8, 95% CI 2.4 to 1450, p = 0.014; Figures 6B, S6N, and S6O).

To assess the role of the DREAM complex in these effects, we analyzed single-cell RNA sequencing data¹⁹ (GSE158724) from a subset of patients from this trial. First, an analysis of the change in proliferation score using the single-cell RNA-seq data revealed that TP53 mutant cases showed a rebound at the end of treatment compared to day 14 (Fold change = 7.08, p < 2e-16), while TP53 WT cases maintained a low proliferation score over time (Fold change = 0.85, p = 6.7e-06, Figure 6C). Moreover, gene set enrichment analysis (GSEA) on differentially expressed genes between TP53 WT and mutant cases revealed that DREAM complex related genes were positively correlated with TP53 mutational status (Figure 6D). We examined the time-dependent up-regulation of transcripts associated with stable arrest previously identified by long-term ribociclib in MCF7 cells. These transcripts were found to be increased in wild-type TP53 tumors at the end of treatment but not in mutant TP53 tumors (Figure 6E).

These data provide further support that p53 loss prevents a durable cell-cycle arrest by CDK4/6i in HR+ breast cancer in vivo thereby limiting the full therapeutic potential for this approach.

DISCUSSION

In this study, we investigated the underlying basis for long-term response to CDK4/6 inhibition and identify defective p53 function to be a highly prevalent determinant preventing long-term disease control. While identifying the clinical scenario of high-risk or metastatic HR+ breast cancer defines the group of patients that can benefit from CDK4/6 inhibitors, it fails to establish which patients have a high likelihood of long-term clinical benefit owing to the wide variance in duration of response observed.⁵¹ We and others have previously identified several genetic alterations that mediate de novo resistance to CDK4/6i therapy,^25,33,52 but these mainly constitute uncommon alterations that promote rapid progression. By contrast, most patients on CDK4/6i therapy remain on therapy for >6 months. Therefore, we focused on identifying genetic alterations that separate an “intermediate” response from a durable response. Using these criteria, we identified pre-existing loss-of-function variants in TP53 as under-represented in patients who experienced prolonged response. Moreover, an independent cohort featuring more than 2,800 patients confirmed the robust significance of TP53 in this context. The results from randomized CDK4/6i trials further establish TP53 to be a consistent and significant prognostic factor associated with diminished response to CDK4/6i.^53,54

TP53 is one of the most commonly mutated genes in breast cancer and plays protean roles in tumor maintenance and progression, regulating DNA damage repair, cell cycle, chromatin architecture, metabolism, and death.^55–57 Thus, the ways in which mutant TP53 might be relevant to drug response could be multifactorial. However, our findings on CDK4/6i response point to the specific function of p53 in regulating the DREAM complex in mediating these differences in outcomes. In still further support of a central role for p53 in this effect, oncogenic alterations in genes known to interact with p53, MDM2, and PPM1D, were also enriched in patients who did not achieve a long-term response.^58–60

p53-deficient cells failed to initiate the changes that lead to phenotypes of cancer cell senescence, changes that were seen in the p53 proficient models. Mirroring these results from models, samples from breast cancer patients treated with neoadjuvant ribociclib on the FELINE trial revealed that resumption of the cell cycle after initial G1 arrest was commonly observed in p53 mutant but not wild-type cases. The specific contribution of the senescence program to mediating the efficacy of the CDK4/6i remains uncertain as there may be both tumor suppressive and tumor promoting factors that can be elicited in different contexts.^31,61 Nevertheless, failure-to-induce senescence in this scenario was reflective of a capacity of p53 deficient cells to re-enter the cell cycle and thereby re-initiate proliferation.

The failure of p53 loss cells to elaborate senescence phenotypes suggested that the defect in this context was upstream of the epigenetic and transcriptional re-programming that drives these phenotypes. We examined the gene expression changes in response to CDK4/6i and identified changes in E2F and DREAM complex targets as differential between the p53 proficient and deficient models. Based on the seminal work identifying the function of the DREAM complex in maintaining quiescence and blocking cell-cycle re-entry,^35,36 we analyzed the assembly of this complex and identified p130 as differentially phosphorylated in p53 KO and WT drug-treated cells. The lower levels of p21 expressed in p53 KO cells provided a probable candidate for residual p130 phosphorylation by the CDK2 kinase. Either re-establishment of p21 expression or direct inhibition of CDK2 facilitated p130 de-phosphorylation, blocking cell-cycle re-entry, and enabling durable inhibition of proliferation (Figure 6F).

Our findings highlight the critical function of p53 and the DREAM complex in mediating the antitumor effects of G1-directed anticancer therapeutics and point to a specific context for several highly selective CDK2 inhibitors including AZD8421 now entering clinical trials.^44,45,62 Our data imply that the p53-deficient set of tumors (a highly prevalent subset) may represent a major indication, serving to distinguish them from treatment with the current generation of CDK4/6i and potentially extending the benefit of G1-directed therapies to a significantly wider population of patients.

Limitations of the study

One limitation of this study is that development of biomarkers of stable arrest and senescence in patients has not been possible due to limited on-treatment samples. We have attempted to validate the findings in a small cohort of tumors from the FELINE study.¹⁹ These data suggest increased senescence-associated transcripts in WT-CDK4/6i treated cases. However, the cohort was insufficiently powered to analyze the full spectrum of senescence-associated transcripts and SASP to definitively mark this phenotype. Newer, preferably blood-based, biomarkers that assess more than the initial steps of G1-arrest are needed to mark the distinct cell fates that CDK4/6i induce to effectively stratify patients for treatment escalation.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resourses and reagents should be directed to and will be fulfilled by the lead contact, Sarat Chandarlapaty (ChandarS@mskcc.org).

Materials availability

There are restrictions to the availability of primary cell lines through Material Transfer Agreement requests at Memorial Sloan Kettering Cancer Center. AZD8421 is a proprietary compound provided by AstraZeneca for the study.

Data and code availability

• Raw data reported in this paper will be shared by the lead contact upon request.

• RNA-seq data have been deposited at GEO with accession numbers of GSE248294 and GSE272312.

• No original code has been reported in this manuscript: detailed equations for data analysis have been included in the method details.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODELS AND STUDY PARTICIPANT DETAILS

Cell lines

MCF7 and HCC1500 cell lines were obtained from the American Type Culture Collection (ATCC). HEK293T cells were kindly gifted by Ping Chi’s lab. MCF7 cells and HEK293T cells were maintained in the DMEM/F12 medium. HCC1500 cells were maintained in an RPMI medium. All media were supplemented with 10% FBS, 2 mmol/L L-glutamine, 20 U/mL penicillin, and 20 μg/mL streptomycin. All cell lines were authenticated by short tandem repeat genotyping and tested negative for Mycoplasma contamination. All cells were maintained in humidified incubator at 37°C with 5% carbon dioxide. All cell lines were cultured and collected within 10 passages. STR genotyping of all cell lines used in this study was previously performed to validate cell line identity.

Primary breast cancer cell line

Patient derived xenograft (PDX) tumor for p53 mutant PD BC cell line was obtained from a female breast cancer patient. The tumors were washed with cold PBS containing antibiotics upon arrival and cut into 1–3mm³ pieces. Small pieces were digested using tumor dissociation kit (Miltenyi Biotec, 130-095-929) on gentleMACS Octo Dissociator at 37°C for 1h. Digested tissue suspension was spun down at 600 rcf for 5 min. Supernatant was aspirated before 10 mL TrypLE was added (Gibco, 12563011) and suspension was incubated at 37 for 10 min on thermomixer. The suspension was strained over a 100 mm filter. 10 mL Advanced DMEM/F12 containing 1x Glutamax, 10mMHEPES, and antibiotics was added to rinse off the tube. Step was repeated if necessary. Suspension was spun down at 600 rcf for 5 min. Supernatant was aspirated before 1 mL of ACK lysis buffer was added (Gibco, A1049201) and incubated for 5 min at room temperature. PBS was added to stop the reaction. Suspension was spundown at 600 rcf for 5 min. Cells were resuspended in medium and seeded.

Breast cancer organoids

PDX tumors were obtained from female breast cancer patients. Tumors were washed with cold PBS containing antibiotics upon arrival and cut into 1–3mm3 pieces. Small pieces were digested using tumor dissociation kit (Miltenyi Biotec, 130-095-929) on gentleMACS Octo Dissociator at 37C for 1h. Digested tissue suspension was spun down at 600 rcf for 5 min. Supernatant was aspirated before 10 mL TrypLE was added (Gibco, 12563011) and the suspension was incubated at 37 for 10 min on thermomixer. The suspension was strained over a 100 μm filter. 10 mL of Advanced DMEM/F12 containing 1x Glutamax, 10mMHEPES, and antibiotics was added to rinse off the tube. This step was repeated if necessary. Suspension was spun down at 600 rcf for 5 min. Supernatant was aspirated before adding 1 mL of ACK lysis buffer (Gibco, A1049201) then was incubated for 5 min at room temperature. PBS was added to stop the reaction. The suspension was spun down at 600 rcf for 5 min. The pellet was resuspended in 1 mL of cold Matrigel (Corning, 356231). 40 μL of suspended samples was added to each well of prewarmed 24well plates (Greiner, M9312). Solidified in 37C after 20 min, 400mL warm of organoid media was added to each well. Media was refreshed every 3 or 4 days and organoids were passaged every 9 days. Organoid culture media contains Advanced DMEM-F12 without phenol red (Gibco), 1003 Glutamax (Life teqnologies, 35050-061), 10mM HEPES (Gibco, 15630080), 50x B27 supplement (Gibco, 17504044), 100x Penicillin-Streptomycin(Gibco), 100 μg/mL primocin (Invitorgen, NC9141851), 10% R-Spondin conditioned medium, 10% Noggin conditioned medium, 500 nmol/L A83-01 (Tocris Bioscience, 29-391-0), 1 ng/mL hEGF (PeproTech, AF-100-15), 100 ng/mL Hydrocortisone (Sigma Aldrich, H0888), 5 ng/mL hFGF2 (PeproTech, AF-100-18B), 5 ng/mL hFGF7 (PeproTech, 100-19), 10 ng/mL hFGF10 (PeproTech, AF-100-26), 1 nmol/L estradiol, 1 ng/mL Neuregulin 1 (PeproTech, 100-03), 1.25 mmol/L N-acetylcysteine (Sigma Aldrich, A9165), 10 mol/L nicotinamide (Sigma Aldrich, N0636), 4 μg/mL Heparin (Stemcell technologies) and 5 mmol/L Y-27632 (Enzo Life Science, 50-103-1737).

Mouse models

Female NOD.Cg-Prkdc<scid> Il2rg<tm1Wjl>/SzJ (NSG) mice were obtained from the Jackson Laboratory at age of 6–8 weeks (stock no. 005557). All mouse studies were conducted through Memorial Sloan Kettering Cancer Center and performed in compliance with institutional guideline under an institutional animal care and use committee (IACUC)-approved protocol (MSKCC 12-10-016) and the ARRIVE guidelines. Patient consent for tumor use in animals was obtained under a protocol approved by the Memorial Sloan Kettering Cancer Center IRB (IRB#13-040).

Human patients

A total of 467 female patients with metastatic HR + breast cancer treated with CDK4/6i in combination with ET the first-line setting underwent prospective clinical genomic profiling between April 2014 and September 2022. This study was approved by the Memorial Sloan Kettering Cancer Center Institutional Review Board (IRB) and all patients provided written informed consent for tumor sequencing and review of patient medical records for detailed demographic, pathologic, and treatment information (NCT01775072). The demographic and clinical characteristics of the cohort are presented in Table S1.

We studied the breast cancer samples from the FELINE clinical trial (ClinicalTrials.gov identifier: NCT02712723), which evaluated the addition of CDK inhibition to endocrine therapy in the neoadjuvant setting. FELINE is a randomized, placebo controlled, multicenter investigator-initiated trial. Female patients (n = 120) were randomized equally into three arms: (1) endocrine therapy alone (letrozole plus placebo), (2) intermittent high-dose combination therapy (letrozole plus ribociclib (600 mg d⁻¹, 3 weeks on/1 week off) or (3) continuous lower-dose combination therapy (letrozole plus ribociclib (400 mg d⁻¹)). Patients were treated for six cycles and biopsies were collected at baseline (SC), following treatment initiation at day 14 (CD) and end of treatment at surgery around day 180 (EOT). Immediately after collection, biopsy samples were snap-frozen and embedded in OCT compound. Informed consent was obtained from all patients following protocols approved by the institutional review boards and in accordance with the Declaration of Helsinki. This study was approved by University of Kansas Institutional Review Board (protocol no. CLEE011XUS10T).¹⁹

For a validation cohort, we analyzed the deidentified clinicogenomic Tempus database of patients sequenced, which contained retrospective, observational data originating from hundreds of community sites and academic sites. A total of 2820 female patients with advanced ER+/HER2-breast cancer sequenced with Tempus xT who were treated with CDK4/6i and endocrine therapy in the first-line and second-line (post first-line endocrine monotherapy) were included.^71–73 The demographic and clinical characteristics of the cohort are presented in Table S2.

METHOD DETAILS

SA- β-gal assays

Cells were treated with DMSO, 50 nM abemaciclib or 500 nM evbaciclib for 5–8 days. Cells were harvested using trypsin/EDTA and then stained with the CellEvent Senescence Green Flow Cytometry Assay Kit (Invitrogen, C10840) according to the manufacturer’s instructions. Briefly, cells were fixed with 2% paraformaldehyde for 15 min at room temperature and stained with the CellEvent Senescence Green Probe for 2 h at 37°C without CO2. After incubation, cells were washed with PBS three times and resuspended in FACS buffer (PBS with 2%FBS) for analysis on BD Biosciences LSR Fortessa using a 488-nm laser and 530-nm/30 filter (BD Biosciences). Data analysis was performed with FCS Express 7 (De Novo Software). SA-β-Gal staining was assayed using Senescence β-Galactosidase Staining Kit from Cell Signaling Technologies (#9860) according to manufacturer’s instructions and previously described.⁶¹ For tumor tissue staining, 10μm thickness slides were prepared from OCT compound frozen tissue and stained for overnight with the kit.

Cell-cycle analysis

Cells were treated with DMSO or 100 nM abemaciclib for 24 h. Cells were detached from the cell culture dish with trypsin/EDTA, and then washed with PBS and fixed in 70% ice-cold EtOH overnight. Prior to staining, EtOH was removed, and cells were washed twice with FACS buffer. Cells were then resuspended in staining buffer containing 1,000 μL FACS buffer with 2 μg/mL propidium iodide (Invitrogen) and 100 μg/mL Rnase A (Invitrogen). Cell-cycle profiles were measured with BD Biosciences LSR Fortessa and analyzed with FCS Express 7.

Cloning and plasmids

LentiCRISPRv2 puro backbone was used for generating knockout cell lines. lentiCRISPRv2 puro was a gift from Brett Stringer (Addgene plasmid # 98290). Single-guide RNAs were designed through MIT CRISPR Designer (crispr.mit.edu), and the sequences are as follows: p53-CRISPR, CACCGGAGCGCTGCTCAGATAGCGA; Instructions for using the lentiCRISPRv2 plasmids are as described by the Zhang laboratory (https://media.addgene.org/cms/filer_public/53/09/53091cde-b1ee-47ee-97cf-9b3b05d290f2/lenticrisprv2-and-lentiguide-oligo-cloning-protocol.pdf). Oligos were annealed and ligated with BsmBI-digested lentiviral vector. Then, the ligation system was transformed into Stbl3 bacteria, and plasmids were extracted for sequencing.

For another guide RNA, we obtained sgTP53_3 which was a gift from William Hahn (Addgene plasmid # 78164), and lentiCRISPRv2 hygro which was a gift from Brett Stringer (Addgene plasmid # 98291).

pDONR223_MDM2_WT was a gift from Jesse Boehm & Matthew Meyerson & David Root (Addgene plasmid # 82897) cloned into pLenti PGK Neo DEST (pLenti PGK Neo DEST (w531–1) a gift from Eric Campeau & Paul Kaufman (Addgene plasmid # 19067) by using the Gateway LR Clonase II Enzyme Mix (Invitrogen).

pDONR223_CDKN1A_WT was a gift from Jesse Boehm & William Hahn & David Root (Addgene plasmid # 82201) was cloned into pInducer20 (a gift from Stephen Elledge, Addgene plasmid # 44012) by using the Gateway LR Clonase II Enzyme Mix. As p53 expressing vector, pLenti6/V5-p53_wt p53, pLenti6/V5-p53_R273H and pLenti6/V5-p53_R280K were used. These plasmids were gifts from Bernard Futscher (Addgene plasmid # 22945, # 22934 and # 22933).

tFucci(CA)5 was a gift from Atsushi Miyawaki (Addgene plasmid # 153521).

pDONR223_ V5-p53_wt p53 was generated from pLenti6/V5-p53_wt p53 by Gateway BP Clonase II Enzyme mix (Invitrogen) then was cloned into pInducer20 (a gift from Stephen Elledge, Addgene plasmid # 44012) by using the Gateway LR Clonase II Enzyme Mix.

Lentiviral delivered sgRNA targeting RBL2 were purchased as a DNA construct from Cellecta, All sgRNA sequences used are provided in the key resources table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
CDK6	Cell Signaling Technology	3136S, RRID: AB_10691463
Cyclin A2	Cell Signaling Technology	4656, RRID: AB_2071958
E2F1	Cell Signaling Technology	3742S, RRID: AB_10695245
FAT1	Abcam	#ab190242
pRb S780	Cell Signaling Technology	8180S, RRID: AB_10950972
pRb S807/811	Cell Signaling Technology	8516S, RRID: AB_11178658
Rb	Cell Signaling Technology	9309S, RRID: AB_10696874
GAPDH	Cell Signaling Technology	5174, RRID: AB_10622025
RBL1	Cell Signaling Technology	89798, RRID:AB_2800144
RBL2	Cell Signaling Technology	13610, RRID: AB_2798274
E2F4 (CHIP, WB)	Cell Signaling Technology	40291, RRID: AB_2799174
p53	Cell Signaling Technology	9282, RRID: AB_331476
p21	Cell Signaling Technology	2947, RRID: AB_823586
HA-Tag	Cell Signaling Technology	3724, RRID: AB_1549585
V5-Tag	Cell Signaling Technology	13202, RRID: AB_2687461
MDM2	Cell Signaling Technology	86934, RRID: AB_2784534
Phospho-p130 (S672)	Abcam	#ab76255, RRID: AB_2284799
p130 (IP)	Santa Cruz Biotechnology	c-374521, RRID: AB_10989107
normal mouse IgG	Santa Cruz Biotechnology	sc-2025, RRID: AB_737182
normal rabbit IgG	Cell Signaling Technology	2729, RRID: AB_1031062
MDM2	Millipore	OP46, RRID: AB_437744
E2F4 (IP)	Santa Cruz Biotechnology	sc-398543
b-actin	Cell Signaling Technology	4970, RRID: AB_2223172
CDK4	Cell Signaling Technology	12790, RRID: AB_2631166
CDK2	Cell Signaling Technology	2546S, RRID: AB_10695594
Anti-mouse IgG, HRP-linked	Cell Signaling Technology	7076, RRID: AB_330924
Anti-rabbit IgG, HRP-linked	Cell Signaling Technology	7074, RRID: AB_2099233
HP1γ	Millipore	05–690
Goat anti-Mouse IgG (H + L) Cross-Adsorbed ReadyProbes Secondary Antibody, Alexa Fluor™ 594	Thermo Fisher Scientific	R37121, RRID: AB_2556549
CONFIRM anti-Ki-67 (30-9)	Roche	RRID: AB_2631262
Bacterial and virus strains
One Shot Stbl3 chemically competent E. coli	Thermo Fisher Scientific	Cat# C737303
Biological samples
BC6 PDX	Laboratory of Harikrishna Nakshatri	N/A
PDX #5	MSKCC	N/A
FELINE breast cancer samples	University of Kansas Medical Center	ClinicalTrials.gov identifier: NCT02712723
Chemicals, peptides, and recombinant proteins
Palbociclib	Selleck Chemicals	Cat# S1579
Abemaciclib mesylate	Selleck Chemicals	Cat# S7158
Ribociclib	Selleck Chemicals	Cat# S7440
Nutlin-3a	Selleck Chemicals	S8059
PF-07104091	Selleck Chemicals	S9878
SNS-032	Selleck Chemicals	BMS-387032
PF-06873600	AdooQ BioScience	A16875
ribociclib (LEE011)	Novartis	N/A
AZD8421	AstraZeneca	N/A
Dimethyl sulfoxide	Sigma-Aldrich	Cat# D2650
fulvestrant	Selleck Chemicals	S1191
Critical commercial assays
QIAGEN RNeasy kit	Qiagen	Cat# 74106
CellEvent Senescence Green Flow Cytometry Assay Kit	Invitrogen	C10840
Resazurin	R&D Systems	Cat# AR002
Dynabeads Protein G	Invitrogen	10003D
Premo™ FUCCI Cell Cycle Sensor	Invitrogen	P36238
IncuCyte Cell Cycle Green/Red Lentivirus Reagents	Sarutorius	4779
CellTiter-Glo® 3D Cell Viability Assay	Promega	G9683
SimpleChIP Enzymatic Chromatin IP kit	Cell Signaling Technology	#9003
Gateway LR Clonase II Enzyme Mix	Invitrogen	Cat# 11791020
Gateway BP Clonase II Enzyme Mix	Invitrogen	Cat# 11789020
Deposited data
Raw data of RNAseq	This paper	GEO: GSE248294
Analyzed data of scRNAseq	Griffiths et al., 202119	GEO: GSE158724
Raw data of RNAseq	This paper	GEO: GSE272312
Experimental models: Cell lines
MCF7	ATCC	CRL-12584, RRID:CVCL_0031
HCC1500	ATCC	CRL-2329, RRID:CVCL_1254
HEK293T	ATCC	CRL-3216, RRID:CVCL_0063
MCF7 p53KO #1	This paper	N/A
MCF7 p53KO #2	This paper	N/A
PDO#6	This paper	N/A
NTsgRNA PDO#6	This paper	N/A
p53sgRNA PDO#6	This paper	N/A
PD BC line #5	This paper	N/A
PD BC line #5 V5 WT	This paper	N/A
HCC1500 p53KO #1	This paper	N/A
MCF7 p53KO + V5 WT	This paper	N/A
MCF7 p53KO + V5 R273H	This paper	N/A
MCF7 p53KO + V5 R280K	This paper	N/A
MCF7 MDM2OE	This paper	N/A
MCF7 FAT1KO	Li et al., 201825	N/A
MCF7 CDK6 OE	Li et al., 201825	N/A
MCF7 p21-HA	This paper	N/A
p53KO p21-HA	This paper	N/A
Cas9-MCF7 p21-HA	This paper	N/A
RBL2KO-MCF7 p21-HA	This paper	N/A
MCF7 p130KO#1	This paper	N/A
MCF7 p130KO#2	This paper	N/A
Experimental models: Organisms/strains
NSG mice (female, 6–8 weeks)	Jackson Laboratory	JAX:005557,RRID:IMSR_JAX:005557
Oligonucleotides
MYBL1 ChIP (forward: 5′- ctagctgcgggggagagg-3′, reverse: 5’- cctgacatgttgggggtatcc -3’)	This paper	N/A
HMGB3 ChIP (forward: 5’-ctacttgggtcggctccatt -3’, reverse: 5′- tcctactctgccttctgcca -3′)	This paper	N/A
p53CRISPR sgRNA (CACCGGAGCGCTGCTCAGATAGCGA)	the Zhang laboratory	https://media.addgene.org/cms/filer_public/53/09/53091cde-b1ee-47ee-97cf-9b3b05d290f2/lenticrisprv2-and-lentiguide-oligo-cloning-protocol.pdf
TaqMan Probes (see Table S4)
Recombinant DNA
CRko-sg_RBL2_1 in pRSG16-U6-sg-UbiC-TagRFP-2A-Puro	Cellecta	R# 99328-1P (221011004)
CRko-sg_RBL2_2 in pRSG16-U6-sg-UbiC-TagRFP-2A-Puro	Cellecta	R# 99328-2P (221011005)
Cas9-2A-Blast lentivirus plasmid	Cellecta	# SVC9B-PS
psPAX2	Laboratory of Didier Trono	Addgene# 12260
lentiCRISPR v2 puro	Stringer et al., 201963	Addgene # 98290
sgTP53_3	Hong et al., 201664	Addgene #78164
tFucci(CA)5	Ando et al., 202365	Addgene# 153521
pDONR223_MDM2_WT	Berger et al., 201666	Addgene #82897
pLenti PGK Neo DEST	Campeau et al., 200967	Addgene# 19067
pDONR223_CDKN1A_WT	Kim et al., 201668	Addgene # 82201
pInducer20	Meerbrey et al., 201169	Addgene #44012
pLenti6/V5-p53_wt p53	Junk et al., 200870	Addgene # 22945
pLenti6/V5-p53_R273H	Junk et al., 200870	Addgene # 22934
pLenti6/V5-p53_R280K	Junk et al., 200870	Addgene # 22933
Software and algorithms
FCS Express 7	De Novo Software	RRID:SCR_016431
nSolver 4.0 software	NanoString	RRID:SCR_003420
MIT CRISPR Designer	MIT	crispr.mit.edu
Analysis of RNAseq data	Basepair software	https://www.basepairtech.com/
Image Studio Lite 5.2	LICORbio	N/A
GraphPad Prism10	GraphPad	RRID:SCR_002798
IncuCyte S3 Live Cell Analysis System	Sartorius	RRID:SCR_023147

Lentiviral and retroviral infection and generation of stable cell lines

HEK293T cells were transfected with 4.5μg of lentiviral vector, 4.5 μg of psPAX2 and 1μg of pVSVG with 40μL X-tremeGENE HP (Roche) according to the manufacturer’s protocol. Conditioned medium containing recombinant lentivirus was collected 48hrs after transfection and filtered through non-pyrogenic filters with a pore size of 0.45 μm (Merck Millipore). Samples of these supernatants were applied immediately to target cells together with Polybrene (Sigma-Aldrich) at a final concentration of 8 μg/mL, and supernatants were incubated with cells. After infection, cells were placed in fresh growth medium and cultured as usual. Selection with 2 μg/mL puromycin (Thermo Fisher Scientific) for 7 days, 1 mg/mL G418 (InvivoGen) for 14 days or 200 μg/mL hygromycin (InvivoGen) for 7 days was initiated 48 h after infection for MCF7. Selection with 0.5 μg/mL puromycin was initiated 72h after infection for HCC1500. HEK293T cells were seeded in a 10 cm cell culture dish and transfected with 4.5 μg of Cas9-2A-Blast lentivirus plasmid (Cellecta, # SVC9B-PS) or RBL2 KO lentiviral vector (CRko-sg_RBL2 in pRSG16-U6-sg-UbiC-TagRFP-2A-Puro, CELLECTA), 3.4 μg psPAX2/pCL-Ampho, and 1.1 μg pVSVG with 27 μL X-tremeGENE HP (Roche) according to the manufacturer’s protocol. MCF7 p21-WT-HA dox inducible cells were infected with 4mL of Cas9-2A-Blast lentivirus in presence of 8 μg/ml polybrene (Santa Cruz Biotechnology) and incubated for 12–16 h. Following which the cells were placed in fresh growth medium. Selection with 8 μg/mL blasticidin (Gibco) was initiated 48 h after infection and continued till the control plate (with no Cas9-2A-Blast virus infection). The generated Cas9-MCF7 p21-WT-HA dox inducible cells were infected with 4mL of the CRko-sg_RBL2 viral supernatant, together with polybrene and selection with 2 μg/mL puromycin (Thermo Fisher Scientific) was initiated 48 h after infection.

Cell viability assay

For cell lines, cell viability was measured by Resazurin (R&D Systems) as described previously.²⁵ Briefly, 1,500 cells were seeded to 96-well plate allowed to recover overnight. Cells were treated with drugs at day 0. Resazurin was added to the cells 4 h prior to the measurement at day 3, day 5 and day 7. For long-term growth assay, 300 or 500 cells were seeded in 96-well. Fluorescence in the plate was measured using a microplate reader (SpectraMax M5, Molecular Devices). IC50 was calculated by GraphPad Prism 9.0 by using a sigmoidal regression model. For organoids, cell viability was determined over time course with a microplate reader using CellTiter-Glo 3D Cell Viability Assay (Promega, G9683) according to manufacturer’s directions. The area of organoids was analyzed using the Incucyte S3 (Sartorius).

Immunoblotting

Cells were lysed in RIPA buffer (Thermo Scientific, 89901) supplemented with 1x protease and phosphatase inhibitor cocktail (Thermo Scientific, 78445). Protein concentration was determined using the BCA protein assay kit (Thermo Scientific, 23227) and equalized using RIPA buffer. Cell lysates were subjected to 4–12% Bis-Tris gel (Thermo Scientific, WG1402BOX) electrophoresis, transferred to nitrocellulose membranes and blocked with 5% non-fat milk in TBST for 1 h at room temperature. Membranes were incubated with the indicated antibodies overnight and then with secondary HRP-coupled anti-mouse (CST, #7076) or anti-rabbit antibody (CST, #7074) for 1 h at room temperature and visualized using Konica Minolta Film Processor with Western Lightning Plus (PerkinElmer, NEL104001) or Immobilon Crescendo Western HRP substrate (Millipore, WBLUR0100). For LiCOR, secondary antibodies used were anti-rabbit IgG (whole molecule)–peroxidase antibody produced in goat (Sigma-Aldrich, catalog A4914), anti-mouse IgG (GE Healthcare Bio-Sciences, catalog NXA931), and IRDye-labeled antibodies (LI-COR Biosciences, catalog 926–68071, catalog 926–32211, catalog 926–32210, catalog 926–68070) and chemiluminescent HRP substrate or on an Odyssey Imaging System (LI-COR Biosciences).

Colony formation assay

MCF7 cells were treated with DMSO or 50 nM abemaciclib for 10 days, and 500 or 1000 cells were seeded in 6 well plate after drug withdrawal for colony formation assay. HCC1500 cells were treated with DMSO or 100 nM abemaciclib for 10 days and seeded at 50000 cells in 12 well plate. Cells were cultured without drug for 18–21 days. After removal of medium, 100% methanol were added for fixation and cells were incubated for 20 min. After removal of methanol and the cells were rinsed with H₂O and covered with crystal violet staining solution (0.5% crystal violet in 25% methanol) for 5 min. The cells were washed with H2O twice and then dried up overnight.

Immunoprecipitation

Cells were lysed in IP lysis buffer (Pierce, 87787) supplemented with 1x protease and phosphatase inhibitor cocktail (Thermo Scientific, 78445). The antibody for p130, E2F4, rabbit IgG or mouse IgG were incubated with the Dynabeads Protein G (Invitrogen, 10003D) for 4 h at 4°C. Excess antibody is washed away by placing the tube in a DynaMag magnet and removing the supernatant. Lysate and Dynabeads were incubated overnight with end-over-end agitation at 4°C. Beads were separated by magnet and washed with IP lysis buffer 3 times before proteins were eluted by heating in 2xNuPAGE LDS sample buffer (Invitrogen, NP0008) and 2x sample reducing buffer (Invitrogen, NP0009) with IP lysis buffer for 5 min at 70°C. Samples were subjected to downstream western blotting with 8% Bis-Tris gel (Thermo Scientific, NW00080BOX) electrophoresis.

FUCCI cell cycle time-lapse imaging

MCF7 or p53KO MCF7 cells were treated with 50 nM abemaciclib for 6–10 days, and replated in Lab-Tek II Chambered Coverglass (Thermo scientific, 155382PK) without drug. On the same day, Invitrogen Premo FUCCI Cell Cycle Sensor (Invitrogen, P36238) was added to chamber. After 27 h, images were taken every hour with a Zeiss Axio Observer Z1 inside a heated 37°C chamber with 5% CO2 for 48 h and analyzed by ZEN Blue 3.2 software. HA-p21 dox inducible cell line which stably expressed tFucci (CA)5, were seeded in Lab-Tek II Chambered Coverglass and treated with 50 nM abemaciclib. After 24 h, 0.5 μ g/ml doxycycline was added to chamber to induce p21. After 4 days abemaciclib treatment, medium was refreshed with 0.5 μ g/ml doxycycline and without abemaciclib. The images were taken every 3 h for 72 h.

To quantify the cell cycle phase proportion, MCF7 and p53KO cells were plated in 6cm dish and IncuCyte Cell Cycle Green/Red Lentivirus Reagents (Sarutorius, 4779) and polybrene were added. After 2 μg/mL puromycin selection for one week, the cells with fluorescence were sorted by Aria-3 cell sorter (BD bioscience). Sorted cells were seeded in Corning 96 well plates (Corning, 3595) and treated with 50 nM abemaciclib for 6 days. The medium was refreshed without drug. Images are acquired and analyzed by using a Sartorius IncuCyte S3 with 3 h imaging intervals. For HA-p21 dox inducible cell lines which were expressing tFucci(CA)5, IncuCyte S3 was used directly to analyze cell cycle analysis.

Senescence-associated heterochromatin foci (SAHF) assay

Cells were plated on 4-well Millicell EZ SLIDE glass (EMD Millipore, PEZGS0416). After 7–10 days DMSO or drug treatment, cells were and fixed for 10 min with 4% paraformaldehyde dissolved in PBS. After three times wash in PBS, cells were incubatoed in 1% BSA dissolved in PBST for 30 min. Cells were then incubated with HP1g antibody (anti-HP1 γ antibody, clone 42s2, EMD Millipore, 05–690, 1:5000) at 4°C overnight, prior to washing with PBS and incubation with anti-mouse secondary antibodies (Goat anti-Mouse IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluo 594,1:500) for 1 h at room temperature. After further washing, coverslips were mounted onto slides with Mounting Medium with DAPI (abcam, ab104139) covered with Fisgerbrand Premiun Cover Glasses (Fisher Scientific, 12-548-5M). Images were acquired Leica TCS SP5 Upright microscope and analyzed with the LAS AF software.

CHIP assay

For chromatin immunoprecipitation (ChIP), MCF7 p53KO cells were treated with 100nM abemaciclib or vehicle (DMSO) for 6 days prior to cross-linking and collection. ChIP was performed using the SimpleChIP Enzymatic Chromatin IP kit (Cell Signaling Technology #9003) according to the manufacturer’s protocol. Briefly, 4.0 × 10⁶ MCF7 p50KO cells were used per chromatin IP, which was equivalent to 10 μg of chromatin. IPs were performed using of the following antibodies, as per manufacturer’s protocol: E2F4 (Cell Signaling Technology #40291), Normal Rabbit IgG (Cell Signaling Technology #2729). For ChIP-qPCR analysis, reactions were performed using the SimpleChIP Universal qPCR Master Mix (Cell Signaling Technology #88989) according to manufacturer’s protocol. Signals of ChIP samples were normalized to their respective input signals. ChIP primer sequences are as shown in key resources table.

In vivo tumor experiment

Tumors were obtained from MCF7 parental cells, MCF7- p53KO or MCF7-CDK6 expressing cells xenograft tumors which were previously reported.²⁵ Briefly, NOD.Cg-Prkdc<scid> Il2rg<tm1Wjl>/SzJ (NSG) mice were obtained from the Jackson Laboratory (stock no. 005557). Each mouse was injected with MCF7 parental or CDK6-ovexpressing, subcutaneously 1 week after the implantation of estradiol pellets (25 mg). Each mouse was injected 10 million MCF7 parental, MCF7-p53KO, V5 WT p53 or V5 R280K cells. 0.18mg b-estradiol pellets 3 days prior to cell implantation. Mice were treated at a 5 days on/2 days off schedule for 17 to 35 days with ribociclib at 25 mg/kg (orally). For PDX model, 10 million PD p53 mutant cells were injected subcutaneously, 0.18mg b-estradiol pellets 3 days prior to cell implantation. Mice were treated by 5 days of ribociclib at oral 25 mg/kg (QD) and 5 days of AZD8421 at oral 100 mg/kg (BID).

BC6 PDX tumor was provided by Harikrishna Nakshatri. 0.18mg b-estradiol pellets 3 days prior to tumor implantation. Mice were treated at a 5 days on/2 days off schedule for 25 to 35 days with ribociclib 75 mg/kg (orally).

Quantitative RT-PCR

Quantitative RT-PCR was performed as described previously.²⁵ Briefly, total RNA was extracted using Rneasy Mini Kit (Qiagen, 74106). RNA was reverse-transcribed into cDNA using qScript cDNA SuperMix (Quanta Biosciences, 95048). qPCR reactions were performed with TaqMan PCR Master Mix (Applied Biosystems, 4369016) using a ViiA 7 Real-Time PCR system (Applied Biosystems) along with primers. Samples were run in triplicate, and mRNA levels were normalized to those of RPLP0, PSMC4 or TBP for each reaction. Taqman primers were purchased from Themo Fisher Scientific and included oligonucleotides are listed in Table S4.

RNA sequencing

Cells were treated with 50nM abemaciclib for 0h, 1 day, 3 days, 5 days, 21 days and 29 days. Total RNA from these cells was extracted using RNeasy Kit (Qiagen). Two independent biological replicates of each RNA sample were submitted for library preparation and RNA sequencing. For another experiment, cells were treated with 50 nM abemaciclib for 0h, 3 days and 21 days. Three independent biological replicates of each RNA sample were submitted and were applied to sequencing. RNAs were sequenced by Illumina HiSeq 2×150 bp sequencing in Genewiz (Genewiz, South Plainfield, NJ, USA). Raw sequencing reads were aligned to the human genome (GRCh38.74) using STAR (2.5.1).⁷⁴ Read counts for each transcript was measured using featureCounts⁷⁵ Raw feature counts were normalized and differential expression analysis using DESeq2.⁷⁶ Differential expression rank order was utilized for subsequent Gene Set Enrichment Analysis (GSEA),⁷⁷ performed using the fgsea (v1.8.0) package in R (v 4.2.0). Gene sets queried included those from the Reactome Gene Sets available through the Molecular Signatures Database (MsigDB).

Nanostring nCounter assay

The nCounter platform (NanoString Technologies) was used to analyze RNA samples cell lines. A minimum of 100 ng of total RNA was used to measure the expression of 271 genes of custom nanoString nCounter gene expression panel and 12 housekeeping genes. Expression counts were then normalized using the nSolver 4.0 software (NanoString, Seattle, WA, USA).

MSK cohort analysis

A total of 467 patients with metastatic HR + breast cancer treated with CDK4/6i in combination with ET the first-line setting underwent prospective clinical genomic profiling between April 2014 and September 2022. This study was approved by the Memorial Sloan Kettering Cancer Center Institutional Review Board (IRB) and all patients provided written informed consent for tumor sequencing and review of patient medical records for detailed demographic, pathologic, and treatment information (NCT01775072). All protein-coding exons and selected intronic and regulatory regions of 341–510 cancer-associated genes, all as previously described.^20,55,78 Somatic mutations, DNA copy-number alterations, and structural rearrangements were identified as previously described and all mutations were manually reviewed. For each gene, oncogenic relevance of specific variants was assessed using the latest versions on the OncoKB knowledgebase.²¹ For each patient with genomic sequencing performed on a tumor sample acquired before the initiation of CDK4/6i, the presence or absence of a putatively gain or loss-of-function variant was noted. For each patient with genomic sequencing performed on a tumor sample acquired before the initiation of CDK4/6i and/or ET, the presence or absence of a predicted oncogenic variant in any pretreatment sample was annotated. Genes with predicted oncogenic alterations in <5% of patients were filtered out of subsequent analyses.

We categorized CDK4/6i regimens based on their endocrine therapy backbone to two major categories: 1) CDK4/6i plus aromatase inhibitors, including letrozole, exemestane, or anastrozole and 2) CDK4/6i plus selective estrogen receptor degraders (SERD) including fulvestrant. We divided time on treatment into tertiles to distinguish between short, intermediate and long-term responders. To isolate the effects between patients who failed to achieve long-term response, we first excluded short-term responses (first tertile) and patients in the second tertile who stopped therapy for reasons other than disease progression (such as toxicity, patient preference, insurance issues, etc.). For the short-term vs. long-term responses, we performed a similar analysis, excluding patients in the second tertile.

For each gene, we iteratively performed Firth penalized logistic regression, with achievement of long-term response as the dependent variable and presence of pathogenic variant in pre-treatment sample as the independent variable. This was performed in the univariate setting, as well as adjusting for endocrine therapy partner. We rejected the null hypotheses with a two-sided p = 0.05. Resulting p values were corrected for multiple hypothesis testing with Benjamini and Hochberg method. For select genes achieving significance, we used univariate and multivariate Cox Proportional hazard models adjusted by treatment class (CDK4/6i + AI or CDK4/6i + SERD) to determine the association between genomic alterations in each gene and PFS. We tested the proportionality assumption of the Cox regression model through time-dependency analysis of selected genetic alterations (cox.zph function of the R package survival).

To identify the pattern of genomic alterations associated with longer, intermediate, and short response, we implemented an elastic net Cox regression on binary functional variant status of each gene as well as select clinical features (endocrine therapy partner, de novo metastatic status) using the Oncocast R package.

Risk groups were stratified based on K-means clustering, and the optimal number of clusters (n = 3) was selected automatically by employing an Akaike information criterion. and again separates patients into short, intermediate, and prolonged time on treatment. Our principal aim was variable interpretability over pure predictive accuracy. Therefore, we sought to obtain measures of variable importance after 50 runs and 5-fold cross validation. Selection frequency was employed as the measure of variable importance, and defined as the proportion of runs (from 50 runs and 5-cross fold validation) in which each gene achieves a coefficient greater than zero. All analyses were performed in R version 4.1.0 (R Foundation).

Tempus cohort analysis

Biopsies taken either prior or up to 3 months following CDK4/6i were included. Patients were classified as TP53-mut if corresponding tumor sequencing harbored either: i) an annotated pathogenic or likely pathogenic variant will allele frequency >0.2, ii) homozygous deletion, or iii) heterozygous deletion with paired pathogenic or likely pathogenic germline variant with allele frequency >0.8. Tumors with the following features were removed from the analysis: i) annotated pathogenic or likely pathogenic variant with allele frequency ≤ 0.2, ii) heterozygous deletion not described above, iii) variant of uncertain significance. Multivariate Cox Proportional hazard models adjusted by treatment class (CDK4/6i + AI or CDK4/6i + SERD) and treatment line to determine the association between genomic alterations in each gene and PFS.^79,80

FELINE trial cohort analysis

Sample processing

Representative sections of 135 OCT compound-embedded core tumor biopsies from 65 patients were reviewed by two pathologists with expertise in breast cancer (Qiqi Ye and J.S. Reis-Filho), and the tumor cell content assessed for subsequent tissue microdissection and DNA extraction.

Microdissection and DNA extraction

Eight-μm-thick sections from frozen tumor blocks were stained with nuclear fast red and microdissected using a sterile needle under a stereomicroscope (Olympus SZ61) to enrich tumor content, as previously described.^81,82 Genomic DNA of microdissected tumor and matched normal samples (buffy coat) was extracted using the DNeasy Blood and Tissue Kit (Qiagen) and QIAamp DNA Blood Mini Kit (Qiagen), respectively, according to manufacturers’ instructions.

Whole genome sequencing

DNA from baseline (SC), day 14 (CD), end of treatment (EOT) tumor and matched normal samples were subjected to whole-genome sequencing (WGS) at MSK’s Integrated Genomics Operations (IGO) using validated protocols,^83,84 with a median sequencing coverage depth of 57x (range, 48x-62x). WGS was completed on 52 SC, 48 CD and 35 EOT tumor and matched normal samples from 65 cases.

Analysis of FELINE scRNAseq

FELINE study¹⁹ 10X scRNAseq count data were downloaded from GEO, accession GSE158724. Out of the 110568 cancer cells publicly released, we selected 97492 cells that already passed quality control steps in the original publication, belonging to 24 pts for which TP53 mutation status was available (18 TP53 wt and 6 TP53 mutated). Furthermore, only genes with at least one count across all the cells were used for downstream analysis. scRNAseq data were aggregated by sum of gene counts from cells derived from each patient at each time point, similarly to what previously described.⁸⁵ DESeq() function from Deseq2⁷⁶ v1.42. R software, version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria; http://www.r-project.org/) was used to fit a negative binomial generalized linear model to pseudobulk counts, to test genes for differential expression between TP53 wild type and mutated samples, with the significance cutoff alpha = 0.05.

fgsea() function, with nperm = 100000, from fgsea v1.28.0 R package was used to run a preranked gene set enrichment analysis on the results from the differential gene expression analysis. For each cell, the proliferation score was calculated by means of a ssGSEA analysis on the expression values of a curated and published list of proliferation specific genes.⁸⁶ These genes are: ′MCM7′,′MCM5′,′MCM4′,′MCM3′,′MCM2′,′MCM6′,′CCNB2′,′CCNB1′,′PLK1′,′BUB1′,′MKI67′,′PCNA’,′CCND3′,′CCND1′,′CCND2′,′E2F1′,′CCNE1′,′CCNE2′,′MYBL2’. Briefly for each cell the following steps were performed.

1. All genes were rank transformed, with order(), decreasing = T, R function such that the most highly expressed gene would have the rank of 1.

2. Rank transformed genes were classified as positive or negative if they were present in the proliferation gene set of interest or not, respectively. For both positive and negative genes, the step cumulative density function was calculated as the cumulative sum of the positive, or negative, gene ranks normalized over the sum of the ranks of the positive, or negative, gene ranks.

3. The difference between the step cumulative density function of the genes in the gene set (positive) and the genes not in the gene set (negative) was used to define the enrichment score.

Significance of the differences between TP53 mutation status and time points was assessed by Wilcoxon rank test, with the wilcox.test() R function.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis of differences between samples was performed using two-tailed Student’s t-tests, and p < 0.05 was defined as significant. When comparing various groups two-way ANOVA statistical test was used applying the Sidak’s method or Tukey’s method to correct for multiple comparisons. The analysis was conducted using GraphPad Prism 9. Independent experiments were conducted with a minimum of two biological replicates per condition to allow for statistical comparison. The numbers of biologically independent experiments, the number of individual mice or patient numbers, details of statistical tests performed can be found in the respective figure legends or method section. We showed p values *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 in figures.

Supplementary Material

Supplemental information can be found online at https://doi.org/10.1016/j.ccell.2024.09.009.

Highlights.

• TP53 mutation is associated with lack of long-term disease control on CDK4/6i

• p53 deficiency suppresses DREAM complex in BC cells, enabling cell-cycle re-entry

• Disruption of DREAM complex is sufficient to promote long-term CDK4/6i resistance

• Inhibition of CDK2 overcomes p53 loss, promoting geroconversion and disease control