Selective CDK7 Inhibition Suppresses Cell Cycle Progression and MYC Signaling While Enhancing Apoptosis in Therapy-resistant Estrogen Receptor–positive Breast Cancer

Estrogen receptor (ER)-positive (ER⁺) breast cancer represents the most common subtype of breast cancer and accounts for most breast cancer–related deaths. Despite advancements in treating metastatic ER⁺ breast cancer with the combination of endocrine therapy plus a CDK4/6 inhibitor (CDK4/6i), nearly all patients ultimately develop treatment resistance. A key challenge in treating metastatic disease after progression on endocrine therapy and a CDK4/6i is the existence of multiple mechanisms of resistance (1, 2).

We and other groups have shown that the ER ligand-binding domain (LBD) mutations are the most common genetic mechanism of acquired resistance to endocrine treatment in metastatic ER⁺ breast cancer (3, 4). Notably, the Y537S ER mutation, an ER LBD mutation that supports estrogen-independent breast cancer growth, was found clinically to be enriched after treatment with the combination of the selective ER degrader (SERD), fulvestrant, and the selective CDK4/6i, palbociclib (5), suggesting its potential role in driving acquired resistance to this combination therapy. In a prior study, we performed a non-biased genome-wide CRISPR/Cas9 gRNA knockout (KO) screen and identified CDK7 as a potential vulnerability in ER⁺ breast cancer cells, particularly in the presence of mutant ER under hormone-deprived (HD) conditions, indicating that CDK7 inhibition could potentially overcome treatment resistance engendered by the ER mutations (6).

CDK7 has regulatory roles in cell cycle and transcription. In addition, CDK7 is overexpressed in several cancers compared with normal tissue, including ER⁺ breast cancer (7–10), and therefore is a candidate therapeutic target. In complex with cyclin H1 and MAT1, CDK7 serves as a CDK-activating kinase (CAK) and phosphorylates the cell cycle–associated CDKs, including CDK4, CDK6, CDK2, and CDK1 (11, 12). CDK7 also promotes RNA transcription by phosphorylating the carboxyl-terminal repeat domain of the RNA polymerase II (RNA-polII) large subunit Rpb1, which is regulated by sequential phosphorylation on different residues within a heptapeptide repeat unit (13, 14). More specifically, CDK7 was shown to phosphorylate RNA-polII at S5 and S7 promoting the release of RNA-polII from the mediator, allowing the promoter escape of transcription (13, 14). In addition, CDK7 indirectly increases the phosphorylation of RNA-polII at other sites. In this regard, the CDK7-mediated activation of CDK9 leads to RNA-polII phosphorylation at S2 and promotes rapid extension of transcripts (15). Pertinent to ER⁺ breast cancer, CDK7 also phosphorylates ER at S118 (16), a site critical for ER transactivation. Encouragingly, in an early phase clinical trial the combination of an oral CDK7 inhibitor (CDK7i), samuraciclib, with fulvestrant, demonstrated a favorable toxicity profile and promising antitumor activity in patients with metastatic ER⁺ breast cancer who had previously received treatment with a CDK4/6i and endocrine therapy (17).

The recent development of CDK7i prompted preclinical investigation into their effects in multiple cancer types. Studies employing THZ1, a covalent CDK7i, revealed selective transcriptional activity by preferentially reducing the expression of oncogenes, such as MYC, that rely on continuous transcription driven by superenhancers (18–21). In our previous work, we demonstrated potent growth inhibition by THZ1 in ER⁺ breast cancer cell lines and xenografts harboring the Y537S ER mutation (6). However, more recent studies indicated that THZ1’s effects are not solely attributed to CDK7 inhibition but are also influenced by potent CDK12/13 inhibitory activity (22). Currently, the specific on-target effects of selective CDK7 inhibition in ER⁺ breast cancer remain uncertain. In addition, it remains unclear which of the effects of CDK7 inhibition, including disruption of cell cycle progression, transcriptional initiation, and ER phosphorylation, contribute to the antitumor activity of CDK7i in ER⁺ breast cancer. Motivated by the need to identify: (i) pharmacodynamic endpoints to test the clinical efficacy of CDK7i and (ii) potential biomarkers for the selection of the specific subgroup of patients with ER⁺ breast cancer most likely to benefit from novel CDK7i, we sought to investigate the antitumor activity and on-target effects of selective CDK7 inhibition in therapy-resistant ER⁺ breast cancer.

MCF7 and T47D cells with a doxycycline (DOX)-inducible Y537S or D538G (DOX-Y537S or DOX-D538G) mutation in ER (6), were grown in DMEM and RPMI, respectively [supplemented with 10% FBS, 10 μg/mL penicillin-streptomycin-glutamine (PSG), and 500 μg/mL Geneticin], hereafter referred to as full media (FM) conditions. DOX-inducible ER-mutant MCF7 and T47D cells (MCF7 DOX-Y537S, MCF7 DOX-D538G, and T47D DOX-Y537S) were treated with 1 μg/mL DOX in FM for 3 days to induce the expression of the mutation and subsequently maintained in phenol-red-free DMEM or RPMI supplemented with 10% charcoal-stripped FBS, PSG, and Geneticin, hereafter referred to as HD conditions. MCF7 cells with a stable TALEN knock-in Y537S (KI-Y537S) or D538G (KI-D538G) ER mutation (6) were grown in DMEM supplemented with 10% FBS and PSG. Palbociclib-sensitive (PalboS) and palbociclib-resistant (PalboR) T47D cells (23) and MCF7 cells (24) were grown in DMEM supplemented with 10% FBS and PSG, with or without palbociclib 100 nmol/L for the PalboS cells, and with palbociclib 1 μmol/L for the PalboR cells. Cells were authenticated by short tandem repeat profiling (Bio-Synthesis) and regularly tested for Mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza) according to the manufacturer's instructions.

SY-1365 was a generous gift from Syros [for the qPCR study SY-1365 (catalog # HY-128587) was purchased from MedChem Express]. Samuraciclib (CT7001 hydrochloride) was purchased from MedChem Express. Palbociclib (catalog # S1579) and fulvestrant (catalog # 1191) for in vitro and in vivo studies were purchased from Selleckchem.

Cells were plated in triplicate in 24- or 96-well plates and after 24 hours (day 0) treated with a CDK7i, SY-1365 or samuraciclib, palbociclib, or vehicle (DMSO). At indicated timepoints, cells were counted using the Celigo image Cytometer (Nexcelom). Hoechst (0.4 μg/mL) was used for nuclear staining and propidium iodide (4 μg/mL) was used to stain dead cells. For the growth studies, the number of live cells at different timepoints was normalized to the number of live cells counted at day 0. For the IC₅₀ studies, the number of live cells after drug treatments was normalized to the number of cells treated with DMSO at day 5 and the IC₅₀ values were calculated in Graphpad Prism (RRID:SCR_000306) using the log(inhibitor) versus response – Variable slope (four parameters) function.

T47D and MCF7 [wild-type (WT)-ER or DOX-Y537S mutant] were cultured in FM and then treated for 15 minutes with DMSO or with 50 nmol/L SY-1365 for 15 minutes, 30 minutes, 45 minutes, 1 hour, 3 hours, and 6 hours. Cell lysates were harvested using buffer TRI Reagent (Zymo Research). RNA was harvested using a RNeasy Mini kit (Qiagen). Reverse transcription was performed using High Capacity cDNA Reverse Transcritpion Kit (Thermo Fisher Scientific). A total of 50 ng of RNA was used per sample to perform qPCR using a SsoAdvanced Universal SYBR Green Supermix (Bio-Rad). A CFX96 Connect Real-Time thermal cycler (Bio-Rad) was used. Expression levels were calculated with the ΔΔCt method using GAPDH measurements as control. Experiments were conducted in triplicates and repeated three independent times. Primers for MYC, E2F1, and GAPDH are the following:

• MYC Forward: TGCTCCATGAGGAGACAC;

• MYC Reverse: GTGGCCCGTTAAATAAGCTG;

• E2F1 Forward: GGCTGGACCTGGAAACTGAC;

• E2F1 Reverse: CTGCCACTCTGGCAGTGCA;

• GAPDH Forward: CGAGATCCCTCCAAAATCAA;

• GAPDH Reverse: TTCACACCCATGACGAACAT.

One-way ANOVA with Tukey multiple comparisons test was used to determine statistical significance of the different comparisons.

Cells were plated in triplicate in 96-well plates. After 24 hours, cells were starved in FBS-free medium for 24 hours for cell cycle synchronization and then treated with the CDK7i SY-1365 or samuraciclib, palbociclib, or DMSO for 48 hours in their growth media. The cell cycle assay was performed using the Click-iT EdU Alexa Fluor 488 HCS Assay (Thermo Fisher Scientific) according to the manufacturer's instructions. Newly synthesized DNA was labeled with EdU (5-ethynyl-2′-deoxyuridine), positive cells were counted, and the cell cycle phases were estimated using the Celigo image cytometer.

Cells were lysed in RIPA buffer (Boston BioProducts), supplemented with protease and phosphatase inhibitors (Sigma-Aldrich) and sonicated for 1 minute. Protein concentration was determined by bicinchoninic acid (BCA) assay (Thermo Fisher Scientific), samples were subjected to SDS-PAGE by using NuPAGE 4 to 12% Bis-Tris gels (Life Technologies) and blotted on Trans-Blot Turbo Midi Nitrocellulose membranes (Bio-Rad Laboratories). The list of the antibodies with their relative dilutions is reported in the Supplementary Table S8.

DOX-Y537S MCF7 cells were engineered to constitutively express the CDK7 C312S mutation (MCF7 WT-C/S, MCF7 DOX-Y537S-C/S) with constructs kindly provided by Dr. Nathanael Gray's lab. Briefly, the C312S-mutant CDK7 cDNA (22) was cloned into a pLVX-EF1A-IRES-Puro plasmid (Clontech, #631988). The resulting plasmid was packaged into viral particles in HEK293T cells, using Lenti-XTM HTX Packaging System (Clontech, #631247 and #631249).

TP53 was KO in DOX-Y537S MCF7 cells by lentiviral transduction using the lentiCRISPRv2 (Addgene; # 52961) transfer vector and pMD2.G (Addgene; #12259) and psPAX (Addgene; #12260) lentiviral packaging vectors. Transduced cells were selected and maintained in puromycin (Puro; 1 mg/mL) selective culture media. The following single-guide (sg)RNA sequences were used:

• sgTP53_1 Forward: CATTGTTCAATATCGTCCG;

• sgTP53_1 Reverse: TCGCTATCTGAGCAGCGCT.

• sgTP53_2 Forward: CATGGGCGGCATGAACCGG;

• sgTP53_2 Reverse: CCGGTTCATGCCGCCCATG.

The human CRISPR/Cas9 KO Pooled Library (H3; Addgene #133914) was used. This library targets more than 18,000 annotated genes in the human genome, with six sgRNAs per gene on average for a total of 117,587 sgRNAs, and 3,842 sgRNAs targeting controls such as AAVS1, ROSA26, and CCR5. The library was amplified and sequenced for quality controls. The packaging into lentiviruses and the infection of PalboS and PalboR T47D cells were performed following the manufacturers’ instructions. Two hundred million PalboS and PalboR cells were infected using a 0.1 multiplicity of infection to ensure that most of the cells received only one viral construct. To enhance virus infection, cells were maintained in their growth media supplemented with 8 μg/mL of polybrene for 48 hours. After transduction, cells were selected for 96 hours in their growth media supplement with 2 μg/mL of Puro. The Illumina library preparation was performed from genomic DNA extracted from 30 million PalboS and PalboR cells using the Blood & Cell Culture DNA Kit (Qiagen). The library was constructed by three steps PCR, modifying the Screen Protocol of the Addgene Pooled library #1000000132 following the conditions reported in the Supplementary Table S9. The PCR product was purified by 2% agarose gel and Gel Purification Kit (Qiagen). The sgRNA library distribution in the cell populations was analyzed by Illumina NS500 Single-End 25 bp sequencing of 50 μL of PCR product. Cells were then randomized (day 0) to be treated with DMSO or palbociclib for 10 doublings. To this purpose, 30 million cells were plated and grown in FM containing Puro and DMSO or Puro and palbociclib 100 nmol/L for PalboS cells, and Puro and palbociclib 1 μmol/L for PalboR cells. To limit sgRNA selection by sampling, 30 million cells were replated every time that the cells were passaged. After 10 doublings, the genomic DNA was extracted from 30 million cells and the resulting libraries were sequenced by Illumina NS500 Single-End 25 bp sequencing. Each screen condition was conducted in duplicate randomizing the infected PalboS or PalboR T47D cells into two arms/condition. The day 0 libraries of each screen served as controls to identify positively or negatively selected genes or pathways. Sequencing results from replicates were then concatenated and analyzed using MAGeCK-VISPR and MAGeCKFlute (refs. 25, 26; https://github.com/liulab-dfci/MAGeCK). Non-essential sgRNAs were filtered out running MAGeCK-VISPR. MAGeCKFlute was run on the MAGeCK-VISPR output after we filtered out 748 control sgRNAs and 217 pan-essential genes (27). A list of 26 sgRNAs was also manually filtered out from the final list of CRISPR KO genes (Supplementary Table S9).

All mice were maintained in accordance with local guidelines and therapeutic interventions approved by the Animal Care and Use Committees of Dana-Farber Cancer Institute. For patient-derived xenograft (PDX) studies, patient consent for tumor implantation in nude mice was obtained under an Institutional Review Board (IRB)-approved protocol (Dana-Farber/Harvard Cancer Center IRB protocol 93-085) and with patient consent. These PDX were published previously (6). Tumor samples from the ER-WT PDX1415 and the Y537S ER-mutant PDX1526 were dipped in 50% matrigel and implanted into the cleared fourth mammary fat pads of ovariectomized NOD-SCID-IL2Rgc–/– mice (Jackson Laboratories), without estradiol (E2) supplements for the PDX1526 model and supplemented with 0.18 mg (60 days release) E2 pellets for the PDX1415 model. When tumors reached 150–200 mm³, mice were randomized into four arms: control (drug vehicle; N = 5), fulvestrant 5 mg/mouse/week subcutaneously (N = 6), CDK7i SY-1365 30 mg/kg/twice a week intravenously (N = 5), and fulvestrant 5 mg/mouse + SY-1365 30 mg/kg (N = 4). The dose of SY-1365 was selected on the basis of previous pharmacokinetic studies in mice (28). For the samuraciclib study, when tumors from the PDX1526 reached 150–200 mm³, mice were randomized into four arms: control (drug vehicle; N = 6), fulvestrant 5 mg/mouse/week subcutaneously (N = 6), samuraciclib 30 mg/kg/every day orally (N = 5), and fulvestrant 5 mg/mouse + samuraciclib 30 mg/kg (N = 5). Fulvestrant vehicle was 10% ethanol and castor oil, while SY-1365 vehicle was 20% captisol pH 4–6 with 1M HCL, and samuraciclib vehicle was 10% DMSO in PBS. Tumor volume was measured at least once a week. Tumor volume growth rates were analyzed using a linear mixed effects model that modeled log tumor volume as a function of separate fixed effect slopes of days per treatment and patient-specific random intercepts using the lme4 (v1.1-35.1) package (29) and the Kenward–Rodgers method to estimate degrees of freedom. Pairwise and interaction contrasts comparing slopes between treatments were tested using the emmeans (v1.9.3) package (https://cran.r-project.org/web/packages/emmeans/index.html). Tests for drug synergy were formulated as interaction contrasts comparing the effect of the combined treatment compared with vehicle with the summed individual effects of both single-agent treatments compared with vehicle. Pairwise comparisons were adjusted for multiple testing using the Tukey method for comparing a family of estimates. After 28 days mice were euthanized; tumors were harvested 24 hours after the last dose of drug treatment. For each mouse, a half of the tumor was snapped frozen for DNA and RNA extraction and the other half was formalin-fixed and paraffin-embedded (FFPE) for immunohistochemistry (IHC) staining.

Dual IHC staining of Ki67 (Biocare Medical, CRM325, RRID: AB_2721189), ER (Thermo Fisher Scientific RM9101S0, RRID: AB_149902), p-ER (S118; SAB 11072, RRID: AB_895302), p-CDK1 (T161; Cell Signaling Technology, CST9114, RRID: AB_2074652), p-CDK2 (T160; Cell Signaling Technology, CST2561, RRID: AB_2078685), and c-Myc (Abcam, ab32072, RRID: AB_731658) was conducted on 4-μm FFPE sections, using both Bond Polymer Refine Kit and Bond Polymer Refine Red kit in Leica Bond RX system. The slides were deparaffinized and heat-mediated antigen retrieval was performed with EDTA buffer (pH 9.0). The IHC staining was performed using the antibodies and the incubation conditions reported in Supplementary Table S8. Antigen-antibody reaction was visualized with 3,3′-diaminobenzidine (DAB) chromogen. Omission of the primary antibody was used as a negative control. Whole slide images were acquired from stained slides using a Vectra 3.0 Automated Quantitative Pathology Imaging System (Akoya Biosciences) and analyzed using Halo Image Analysis platform (Indica Labs). Image annotations were performed by one research pathologist. Areas containing invasive carcinoma were included in image analysis.

Terminal deoxynucleotidal transferase–mediated biotin–deoxyuridine triphosphate nick‐end labeling (TUNEL) staining was performed using the In Situ Cell Death Detection Kit, Fluorescein (Roche, catalog # 11684795910) on 4-μm FFPE sections of the mice tumors. All non-necrotic areas of the slides were used to determine the number of TUNEL positive cells, using the QuPath software with integrated tool for cell detection analysis. DAPI staining was used to count all the nuclei of the cells in the selected areas, while GFP staining was used to count all the TUNEL positive cells. The percentage of positive cells is represented by the number of positive cells (GFP positive) over the number of the total cells (DAPI positive) in the area/slide. At least two tumors were used for each group of analysis.

Genomic DNA was extracted from frozen pulverized PDX tumors using Blood & Cell Culture DNA Kit (Qiagen). Quality controls, library preparation, and whole-exome sequencing (WES) were performed at Novogene Corporation Inc. Sequencing libraries were generated using Agilent SureSelectXT2 Homo Sapiens All Exon V6 Kit (Agilent Technologies) following manufacturer's recommendations. WES was performed by Illumina Paired-End 150 bp sequencing.

Total RNA was isolated using TRIzol (Life Technologies) and RNeasy Mini Kit (Qiagen). For all the cell line studies, the RNA extraction and sequencing was done in at least duplicates.

RNA sequencing (RNA-seq) libraries were prepared using NEBNext Ultra II non-directional RNA Library Prep kit (New England Biolabs) and sequenced by Illumina Novaseq6000 Paired-End150 bp sequencing at Novogene.

Somatic variants in tumor-only mode were called using mutect2 (gatk4.1.7.0) and a panel of germline variants from the 1000 Genomes Project (https://www.internationalgenome.org/1000-genomes-project-publications). To automate sample processing, the bcbio-nextgen somatic variant2 pipeline was implemented (http://doi.org/10.5281/zenodo.4041990). For variant annotations snpEff, vcfanno, gnomAD, and COSMIC were used as a part of bcbio-nextgen workflow [snpEff (30): https://pcingola.github.io/SnpEff/; vcfanno (31); GnomAD (32): https://gnomad.broadinstitute.org/]. OpenCRAVAT (33) was used to annotate on-target somatic variants with CHASMplus, chasmplus_BRCA, PolyPhen2, FATHMM, COSMIC, COSMIC Gene, and cancer_hotspots annotations. A total of 13,061 and 15,288 somatic variants passed quality filters in the two PDX models (PDX1415 and PDX1526, respectively). After data cleaning, variant selection was based on: (i) HIGH or MODERATE impact (HIGH: stop gain variants, frameshifts, and splice site variants; MODERATE: mostly missense variants) according to snpEff classification (https://useast.ensembl.org/info/genome/variation/prediction/predicted_data.html) and without using COMSIC filters; (ii) having PASS in the filter field (variant allele frequency≥10%; passed technical filters); and (iii) variants located in the genes included in a custom breast cancer gene panel (34). This filter resulted in 56 variants in the two models, which are visualized in Fig. 2A. To visualize the oncoprint, ComplexHeatmap (35) package in R (https://www.R-project.org/) was used.

Sequenced reads were aligned to the hg19 reference genome assembly and samples were analyzed using the pipeline VIPER (36).

Gene set enrichment analysis was performed by GSEA_4.1.0 on gene lists from DESeq2 testing output ranked by log₂FC x [−log₁₀(P-adj)] values. Only the Hallmark pathways that are significantly [FDR < 0.25, gene set enrichment analysis (GSEA) weighted Kolmogorov–Smirnov test] positively or negatively enriched by normalized enrichment score (NES) are shown.

Gene ontology analysis was performed using the Hallmark dataset and the Compute Overlap function on the online Broad GSEA Application.

Cells were lysed in TPER buffer (Thermo Fisher Scientific) supplemented with 0.5 mol/L NaCl and protease and phosphatase inhibitors (Sigma-Aldrich), and mixed at 4°C for 30 minutes. Lysates were clarified by centrifugation for 15 minutes at 14,000 × g, 4°C, and supernatants were transferred to fresh tubes followed by protein concentration determination by BCA assay (Thermo Fisher Scientific).

The concentration of the lysates was adjusted to 0.5 mg/mL by diluting the samples in a solution of 2.5% 2-mercaptoethanol and SDS buffer containing TPER buffer, 0.25 mol/L Tris (pH 6.8), 8% SDS, 4% glycerol, and Bromophenol Blue (Sigma-Aldrich). Lysates were denatured at 100°C for 8 minutes and centrifuged for 2 minutes, at 14,000 × g, room temperature. Lysates were spotted in an array format and processed as described previously (37) at the Antibody-Based Proteomic Core (Baylor College of Medicine, Houston, TX). Data quality was evaluated by manual inspection and comparison with control samples. Data were normalized using total protein and negative controls. Individual normalized fluorescence values were filtered out if below 200, and technical replicates were filtered out if more than 75% of their proteins had normalized fluorescence values below 200. Each sample was assessed in biological and technical triplicate; the average of the normalized technical replicates per biological replicate for each protein was used for downstream analyses. Normalized data are reported in Supplementary Table S2. Differential protein expression was tested using an unpaired Welch t test between treatment groups for each protein. P values were adjusted for multiple testing using the Benjamini–Hochberg procedure (38).

The multi-commodity network flow optimization tool Simultaneous Analysis of Multiple Networks (SAMNet; ref. 39) was used to integrate reverse phase protein array (RPPA) and RNA-seq differential expression results using a weighted protein–protein interaction network. RPPA proteins were chosen as source nodes, RNA-seq genes were chosen as sink nodes, and each treatment contrast (MCF7 WT SY-1365 over MCF7 WT Veh, MCF7 DOX-Y537S SY-1365 over MCF7 DOX-Y537S Veh, MCF7 PalboS SY-1365 over MCF7 PalboS Veh, MCF7 PalboR SY-1365 over MCF7 PalboR Veh, T47D PalboS SY-1365 over T47D PalboS Veh, T47D PalboR SY-1365 over T47D PalboR Veh) was assigned its own commodity flow. Gene and protein lists taken from differential expression testing results were filtered (RPPA P_adj ≤ 0.05; RNA-seq P_adj ≤ 0.001, RNA-seq |log₂FC| ≥ 1) to limit network input sizes. The protein–protein interaction network was taken from iRefIndex17 (40), modified to retain interactions present in iRefIndex14. Additional nodes for phosphorylated proteins present in the RPPA source lists were added to the interaction network as performed in previous studies, with edges of cost = 0.25 connecting them to their corresponding total protein (41). SAMNet was run with manually selected hyperparameters (number of edge/weight randomizations = 100, node weight parameter gamma = 10, edge penalty epsilon = 1e-3), and its minimizing network flow solution was filtered to remove non-source/sink nodes that had an estimated robustness ≤ 0.4 or specificity ≥ 0.4. The filtered network solution was grouped into Louvain clusters (42) at a resolution of 1.0. Gene set analysis was run on node lists from the filtered network solution and its Louvain subclusters that had at least 25 nodes using Enrichr (43) via the enrichR R package on Hallmark gene sets. P values were adjusted for multiple testing using the Benjamini–Hochberg procedure.

Gene Expression Omnibus accession number for all RNA-seq data is GSE230362.

To investigate the on-target activity of CDK7 inhibition, we studied the effects of SY-1365 (also known as mevociclib), a selective and potent covalent CDK7i (44), in breast cancer cells with WT-ER and isogenic breast cancer cells with DOX-inducible expression of the Y537S ER mutation. We first verified the on-target activity of SY-1365 by engineering MCF7 cells to express a C312S CDK7 mutation. This mutation blocks the covalent binding of SY-1365 to CDK7 without affecting CDK7 activity. The expression of the C312S CDK7 mutation led to an approximate 100-fold shift in the IC₅₀ of SY-1365 in MCF7 cells with and without the DOX-induced expression of the ER Y537S mutation, providing evidence of a window of an on-target growth inhibitory concentration (Fig. 1A; Supplementary Table S1). In contrast, the IC₅₀ of the same cells treated with THZ1 remained unchanged upon expression of the C312S CDK7 mutation (Fig. 1B), supporting the off-target effects of THZ1 as reported recently (22).

With on-target effects on cell growth confirmed, we proceeded to evaluate the impact of SY-1365 on cell growth in various cell lines, including MCF7 cells with knocked-in expression of the Y537S and D538G ER mutations, as well as MCF7 and T47D cells with and without DOX-induced expression of the Y537S and D538G ER mutations. In all these model cell lines, we observed a dose-dependent effect of SY-1365, and this effect was comparable between cells with mutant and WT-ER (Fig. 1C; Supplementary Fig. S1A; Supplementary Table S1). Furthermore, similar effects were observed with a second selective CDK7i, samuraciclib (Supplementary Fig. S1B; Supplementary Table S1).

To gain insights into the molecular effects of SY-1365, we examined known CDK7 targets in a dose- and time-dependent manner. Treatment with SY-1365 at a concentration of 50 nmol/L or higher and as early as 6 hours led to the repression of p-S2 and p-S7 of RNA-polII with a less pronounced effect on p-S5 (Fig. 1D). However, this effect was not seen with lower concentrations, and with the 50 nmol/L concentration, phosphorylation at these sites started to increase after 48 hours. Conversely, inhibition of p-CDK1 (T161), p-CDK2 (T160), and p-ER (S118) and c-Myc was observed at 24 hours, persisted at 48 hours, and was evident with a SY-1365 concentration as low as 10 nmol/L. Importantly, expression of the C312S CDK7 mutation rescued the suppressive effects of 50 nmol/L SY-1365 on p-CDK1 (T161) and p-CDK2 (T160; Supplementary Fig. S1C), supporting the on-target effect of this dose.

To gain a broader view of changes in protein expression and phosphorylation, we employed a RPPA that quantified a total of 230 proteins and phospho-proteins (Fig. 1E; Supplementary Table S2). Several significant protein changes were related to the cell cycle, such as decreased expression of phospho-Rb (S807/811) and total Rb, Aurora A, and c-Myc. In addition, there was a significant decrease in proteins related to epithelial–mesenchymal transition (EMT) in ER-WT cells. Notably, we observed an increase in p53 phosphorylation at S15 in response to SY-1365 treatment. S15 phosphorylation of p53 was shown to impair the ability of MDM2 to inhibit p53-mediated transcription and p53 degradation (45). This finding may explain the increase in p53 and cleaved caspase-7 levels detected in the RPPA data in the presence of WT-ER and mutant ER cells (Fig. 1E). In keeping with these findings, previous reports in other cancer types have highlighted the role of CDK7 inhibition in mediating p53-induced apoptosis (46–49). In support of the association between CDK7 and p53, as shown in Fig. 1C, T47D cells that harbor a pathogenic TP53 mutation (L194F; ref. 50) were less sensitive to SY-1365 compared with MCF7 cells. Furthermore, silencing of TP53 in MCF7 cells resulted in a close to 10-fold shift in the IC₅₀ of SY-1365 in the absence and presence of the Y537S ER mutation (Supplementary Fig. S1D; Supplementary Table S1). This shift remained within the on-target effect of SY-1365, suggesting that while part of the antitumor activity of CDK7 inhibition is mediated by p53, the presence of p53 is not required for on-target activity.

To delineate the global transcriptional effects of CDK7 inhibition we performed RNA-seq experiments. For these experiments, we used a concentration of 50 nmol/L because we determined that this concentration yielded on-target gene expression changes; in the cells expressing the C312S CDK7 mutation, treatment with SY-1365 at a concentration of 50 nmol/L resulted in changes in the expression of 22 genes only, suggestive of marginal off-target transcriptional effects. In contrast, with a concentration of 150 nmol/L in the presence of the C312S CDK7 mutation, the expression of 700 genes was significantly modulated, indicating that this concentration results in substantial off-target effects (Supplementary Fig. S1E). In MCF7 cells with conditional expression of the Y537S ER mutation (without expression of the C312S CDK7 mutation), 50 nmol/L SY-1365 treatment resulted in 250 and 614 genes significantly upregulated and downregulated, respectively, in the presence of WT-ER, and 193 and 534 genes upregulated and downregulated, respectively, in the presence of the Y537S mutation as early as 6 hours (Supplementary Fig. S1F–S1I). The transcriptional analysis revealed a decrease in the expression of genes that are MYC targets and cell-cycle related (E2F and G₂–M) at 6 and 24 hours in the presence of WT or mutant ER (Fig. 1F; Supplementary Fig. S1F–S1I; Supplementary Table S3). Genes related to apoptosis and the p53 pathway were upregulated in WT-ER cells, whereas genes related to estrogen response were decreased only in the presence of the Y537S ER mutation. Similar transcriptional effects were detected in MCF7 cells in which the Y537S ER mutation was knocked-in and expressed under the endogenous ER promoter (Supplementary Fig. S2A–S2C; Supplementary Table S3).

Given the short half-life of MYC mRNA (51), we tested whether the suppression of c-Myc was due to a transcriptional effect by analyzing the levels of MYC nascent RNA at early timepoints. In both MCF7 (Supplementary Fig. S3A) and T47D (Supplementary Fig. S3B) cells with and without induction of the Y537S ER mutation, we observed decreased nascent RNA levels of MYC after 45 to 60 minutes of treatment with 50 nmol/L SY-1365, providing evidence for the role of CDK7 inhibition in mediating transcriptional downregulation of MYC. In contrast, nascent RNA levels of E2F1 decreased only after 3 hours (Supplementary Fig. S3C and S3D). The transcription of E2F1 is cell cycle dependent and regulated by positive feedback (52). Taken together, the early transcriptional downregulation of MYC versus the later downregulation of E2F1 suggests that MYC transcription may be a direct target of CDK7 inhibition, while decrease in E2F1 transcription is likely due to a secondary effect resulting from the cell cycle arrest mediated by CDK7 inhibition. However, the possibility that the downregulation of MYC transcription is an indirect consequence, such as CDK7i effects on other transcription factors or regulators of MYC transcription, were not ruled out.

To validate the functional implications of the proteomic and transcriptomic findings, which consistently indicated that CDK7 inhibition impacts the cell cycle, we conducted cell cycle analyses. In keeping with the observed decrease in CDK2 and CDK1 phosphorylation and the role of CDK7 in the phosphorylation of CDK4 and CDK6, SY-1365 treatment resulted in significant accumulation of WT-ER and Y537S ER-mutant cells in G₀–G₁ and G₂–M with concomitant reduction of the S-phase. In contrast, treatment with palbociclib, a CDK4/6i, resulted primarily in a G₀–G₁ arrest (Fig. 1G and H; Supplementary Table S4). Importantly, similar effects were observed with samuraciclib (Supplementary Fig. S3E). Taken together, these in vitro studies showed that the on-target effects of CDK7 inhibition at early timepoints lead to cell cycle inhibition, inhibition of c-Myc, and upregulation of apoptosis in the presence of WT and mutant ER.

We expanded our investigation of CDK7 inhibition to in vivo studies using ER⁺ breast cancer PDX models. We examined two genetically distinct models derived from patients with heavily pretreated metastatic ER⁺ breast cancer (Fig. 2A; Supplementary Fig. S4A; Supplementary Tables S5 and S6). PDX1415 originated from a liver metastasis of a patient who received prior treatments with an aromatase inhibitor, fulvestrant, capecitabine, taxol, eribulin, and carboplatin-gemcitabine. PDX1526 was derived from a chest wall metastasis harboring a Y537S ESR1 mutation and MYC copy-number gain from a patient who had prior treatments with an aromatase inhibitor, everolimus, fulvestrant, abemaciclib, and capecitabine. WES showed that this PDX model retained the Y537S ESR1 mutation and MYC copy-number gain. As expected with the presence of an activating ER LBD mutation, this PDX displayed resistance to estrogen deprivation and grew in oophorectomized mice without estradiol (E2) supplements. In addition, the transcriptomes of the two models were distinct. Consistent with the presence of the ESR1 mutation, the PDX1526 was enriched in genes of estrogen response compared with PDX1415. Moreover, MYC expression (log₂FC = 6.33, FDR < 0.001) and MYC targets, such as ID1 (log₂FC = 3.22, FDR < 0.001) and MCM family members (log₂FC 0.61–1.42, FDR < 0.001), were upregulated in PDX1526 compared with PDX1415 (Fig. 2B and C; Supplementary Fig. S4B; Supplementary Table S3).

The two PDX models displayed different sensitivities to treatment with fulvestrant and SY-1365 (Fig. 2D and E). PDX1415 was resistant to fulvestrant and SY-1365 as single agents but sensitive to the combination of fulvestrant and SY-1365 with a significant interaction test (P < 0.001). PDX1526 was resistant to fulvestrant but sensitive to SY-1365 as a single agent and to the combination of SY-1365 plus fulvestrant. We also tested the PDX1526 model for samuraciclib alone and in combination with fulvestrant. Of note, in this experiment this model was sensitive to fulvestrant alone, underscoring the heterogeneity of fragments from a single PDX model. This model was sensitive to samuraciclib alone and there was enhanced activity with the combination of samuraciclib plus fulvestrant (Fig. 2F). Importantly, SY-1365, samuraciclib and the combinations of SY-1365 or samuraciclib with fulvestrant were well tolerated in both models (Supplementary Fig. S4C–S4E).

To gain insights to the mechanisms of action of CDK7 inhibition alone and in combination with fulvestrant, we performed detailed analyses of the tumors harvested after 28 days of treatment (Fig. 3; Supplementary Fig. S5). In both models, the changes in Ki67 supported the tumor volume measurements (Fig. 3A and B; Supplementary Fig. S5A and S5B). In keeping with the ligand-independent S118 phosphorylation of ER [p-ER (S118)] engendered by the ER mutations (6, 53), ER phosphorylation was detected in the Y537S ER-mutant PDX1526 model even though these tumors grew in estrogen-deprived conditions (Fig. 3C and D). In the PDX1526 model, fulvestrant treatment alone resulted in decreased ER expression, but p-ER (S118) was not significantly decreased. In contrast, SY-1365 treatment resulted in decreased p-ER (S118) but increased total ER protein expression coupled with transcriptional upregulation of ER as evidenced by increased ESR1 mRNA levels (Fig. 3C–G). Combination treatment with fulvestrant and SY-1365 led to a decrease in expression of ER and p-ER (S118; Fig. 3C–F). In support of increased ER transcription induced by SY-1365 treatment, we again observed an increase in total ER protein and mRNA levels and a decrease in p-ER (S118) levels in response to SY-1365 alone in the ER-WT PDX1415 model (Supplementary Fig. S5C–S5E). In contrast to the ER-mutant PDX1526 model, the ER-WT PDX1415 model was sensitive to fulvestrant alone, and in response to fulvestrant alone or fulvestrant plus SY-1365, there was a near complete loss of total ER and p-ER (S118) levels (Supplementary Fig. S5D and S5E). Taken together, these results provide in vivo evidence for the modulation of ER in response to CDK7 inhibition and a rationale for the combination of a SERD with a CDK7i in ER⁺ breast cancer.

Similar to the on-target effects observed with CDK7 inhibition in the cell lines, SY-1365 treatment in the PDX1526 model led to a decrease in p-CDK1 (T161), p-CDK2 (T160), and c-Myc levels (Fig. 3H–J). Fulvestrant alone had similar effects, but the magnitude of these effects was lower compared with SY-1365. This is consistent with the decreased effects on Ki67 and tumor volumes observed with fulvestrant treatment alone compared with SY-1365. The combination of SY-1365 and fulvestrant potentiated the inhibition of p-CDK1 (T161), p-CDK2 (T160), and c-Myc levels, in line with the enhanced suppression in tumor growth compared with each drug alone. In addition, the combination treatment increased apoptosis, which is consistent with the transcriptomic and proteomic analysis of the cell lines (Fig. 3K).

The PDX1415 model, which was resistant to single agent SY-1365, had low c-Myc levels at baseline and after treatment (Supplementary Fig. S5F). Moreover, p-CDK1 (T161) or p-CDK2 (T160) were not suppressed by SY-1365 treatment (Supplementary Fig. S5G and S5H). However, SY-1365 in combination with fulvestrant resulted in a significant decrease in p-CDK2 (T160; Supplementary Fig. S5H) without an increase in treatment-induced apoptosis in this model (Supplementary Fig. S5I). Thus, the greater impact on CDK2 phosphorylation and ER levels induced by the combination of SY-1365 and fulvestrant provides a putative explanation for the enhanced inhibition on tumor growth observed with the combination treatment in this model.

In keeping with the IHC findings, the transcriptomic analysis of the PDX1526 model showed that treatment with SY-1365 alone and in combination with fulvestrant decreased the expression of genes involved in the cell cycle and MYC signaling (Fig. 3L). In both PDX models, SY-1365 treatment increased the expression of estrogen response genes, consistent with the increase in ER expression (Fig. 3L; Supplementary Fig. S5J). In contrast, fulvestrant decreased the expression of estrogen response genes, and the combination of fulvestrant and SY-1365 led to a trend toward decreased expression of estrogen response genes compared with the vehicle control arm.

In summary, the in vivo studies revealed that SY-1365–mediated inhibition of CDK1 and CDK2 phosphorylation as well as inhibition of c-Myc expression, are likely important for the antitumor activity. These effects were augmented when combined with fulvestrant, supporting the combination of a CDK7i with a SERD. Moreover, in both PDX models, SY-1365 resulted in increased ER transcripts, total ER protein, and ER signaling, likely in response to the inhibition of ER phosphorylation. This effect was counteracted by the addition of fulvestrant, providing a second mechanism of the enhanced activity detected with the combination of fulvestrant and SY-1365 (Fig. 3M).

A significant challenge in metastatic ER⁺ breast cancer is the development of resistance following treatment with a CDK4/6i, particularly due to the diverse mechanisms of resistance, such as RB1 mutations, increased cyclin E1 expression, and elevated PI3Kinase signaling (23, 54, 55). To identify potential therapeutic vulnerabilities in CDK4/6i resistance, we performed genome-wide CRISPR/Cas9 KO screens in T47D cells with acquired resistance to palbociclib (PalboR), which acquired an RB1 loss (23). In addition, we performed screens in T47D cells sensitive to palbociclib (PalboS) treated with vehicle control or palbociclib 100 nmol/L (Fig. 4A).

We initially assessed the correlation between two replicate library screens and confirmed the reproducibility of the results (Supplementary Fig. S6A–S6C). Genes were called essential for growth or suppressive of growth based on β-scores of < −0.5 or ≥ 0.5, respectively. In the PalboS cells with or without palbociclib treatment, we found that ESR1 (β-scores: −2.118 and −2.281, respectively, P < 0.001), CCND1 (β-score: −1.595 and −1.819, respectively, P < 0.001), CDK4 (β-score: −2.392 and −1.856, respectively, P < 0.001) and other genes known to be important in ER⁺ breast cancer cell growth were among the essential genes (Fig. 4B and C; Supplementary Table S7). Interestingly, CDK6 (β-score: −0.433, P: 0.038), CCND3 (β-score: −0.325, P: 0.107), and CCNE1 (β-score: −0.306, P: 0.581) were not essential in vehicle control conditions but gained essentiality after treatment with palbociclib (CDK6: β-score: −2.083, P < 0.001; CCND3: β-score: −1.666, P < 0.001; CCNE1: β-score: −1.860, P < 0.001). As expected, silencing of the tumor suppressors RB1 and PTEN increased cell growth, and this effect was further enhanced by palbociclib treatment, indicating that the efficacy of palbociclib is likely dependent on intact PTEN in addition to the known RB1 dependency.

In the PalboR cells, CCND1, CDK4, and CDK6 were non-essential (CCND1: β-score: −0.628, P: 0.065; CDK4: β-score: 0.308, P: 0.687; CDK6: β-score: 0.077, P: 0.868; Fig. 4C and D; Supplementary Table S7). In contrast, CDK1, CDK2, and PIK3CA remained essential (CDK1: β-score: −2.099, P < 0.001; CDK2: β-score: −1.552, P < 0.001; PIK3CA: β-score: −2.411, P < 0.001). Strikingly, CDK7 was essential in both the PalboS and PalboR (Fig. 4B–D). Furthermore, pathway analysis revealed that the genes essential in PalboR cells were enriched for genes involved in cell cycle pathways and MYC targets (Fig. 4E). Importantly, these pathways were also found to be enriched in the transcriptomic analysis of ER⁺ breast cancer biopsies from patients resistant versus sensitive to palbociclib in combination with endocrine therapy, providing clinical relevance to our findings (56). In summary, our results demonstrate that CDK7 and pathways impacted by CDK7 remain essential after the acquisition of resistance to palbociclib.

To follow up on the results of the CRISPR/Cas9 KO screen, we assessed the antitumor activity of SY-1365 in two ER⁺ PalboR model cell lines. These models have different mechanisms of acquired resistance to palbociclib (Supplementary Fig. S6D–S6F; Supplementary Table S3). The T47D PalboR model is characterized by a genetic loss of RB1 (23), while the MCF7 PalboR model retains low expression of Rb, but shows increased expression of receptor tyrosine kinases [EGFR, p-EGFR (Y1173), and p-HER2 (Y1248)] and the c-Fos transcription factor of the AP1 complex (Supplementary Fig. S6D–S6F), which have been implicated in endocrine resistance (57). In line with the CRISPR/Cas9 KO screen, the T47D and MCF7 PalboR cells were both sensitive to SY-1365 in a dose-dependent manner with IC₅₀ values comparable between the PalboR and their PalboS counterparts (Fig. 5A and B; Supplementary Table S1). In addition, SY-1365 at a concentration of 50 nmol/L inhibited the cell growth of PalboS and PalboR cell models (Supplementary Fig. S6G–S6J). While palbociclib had no impact on the cell cycle in the PalboR cells, SY-1365 increased the percentage of cells in G₂–M-phase, and to a lesser degree, increased the percentage of cells in G₀–G₁-phase, coupled with a decrease in the S-phase (Fig. 5C–F; Supplementary Table S4). Supporting these results, the PalboR cells were also sensitive to samuraciclib (Supplementary Fig. S7A–S7C).

Proteomic and transcriptomic analyses (Fig. 5G and H) revealed that one of the most significant effects of SY-1365 in both MCF7 and T47D PalboR cells was the inhibition of key transcriptional targets of MYC (Fig. 5G). As an example, SY-1365 downregulated the MYC transcriptional target RRM2 (58), the regulatory subunit that is the rate-limiting factor for ribonucleoside reductase crucial for nucleotide synthesis (Fig. 5H; Supplementary Fig. S8A–S8D). Importantly, RRM2 gains essentiality after palbociclib treatment and remains essential in PalboR cells (Fig. 4B–D), indicating that nucleotide synthesis is a potential vulnerability during palbociclib treatment, and inhibition of nucleotide synthesis may contribute to the activity of CDK7 inhibition in PalboR. In the MCF7 PalboR cells, SY-1365 treatment increased the mRNA expression of genes related to the p53 pathway and apoptosis. At the protein level, we observed increased p53, p53 (S15), and p21^Waf1/Cip1 levels (Fig. 5H). As expected, in the T47D PalboS and PalboR cells, which harbor a pathogenic TP53 mutation (L194F; ref. 50), we did not detect upregulation of transcripts or proteins related to p53 in response to SY-1365 treatment. In summary, in models with different genotypes and putative mechanisms of palbociclib, we observed sensitivity to CDK7 inhibition. These results indicate that CDK7 is a potential target for overcoming CDK4/6i resistance driven by diverse mechanisms.

To gain insights into the effects of CDK7 inhibition and leverage the multiple models and the multi-omic analyses of CDK7 perturbation, we utilized the SaMNet algorithm. SaMNet allows the interpretation of diverse assays over multiple experiments. This algorithm uses a constrained optimization approach to integrate mRNA and protein expression from different datasets, together with an existing interactome dataset (39). Through this analysis, we identified an interaction network with 12 distinct clusters (Fig. 6A). Pathway analysis showed that these clusters are enriched in genes/proteins involved in cell cycle, p53 signaling, and apoptosis, and additional pathways such as EMT and apical junctions. Importantly, this analysis highlights tight interactions between these pathways. In addition, the integrative analysis identified the E2F and G₂–M cell cycle pathways as the top overrepresented pathways in the network, reinforcing the observation that these pathways play key roles in the response to CDK7 perturbation (Supplementary Fig. S8E).

Selective CDK7i are emerging as a promising therapeutic strategy for several cancers, including ER⁺ breast cancer (17). CDK7 is a master regulator of both the cell cycle and transcription. In addition, CDK7 plays a role in ER activation through S118 ER phosphorylation. In this study, we conducted comprehensive in vitro and in vivo studies using models of WT and mutant ER, as well as models of resistance to CDK4/6 inhibition, to elucidate the key functions of CDK7 driving the antitumor activity in ER⁺ breast cancer.

Our findings indicate that a key mechanism of action of selective CDK7 inhibition in ER⁺ breast cancer is cell cycle suppression. In PalboS cells that are characterized by an intact cyclin D1/CDK4/6/Rb axis, CDK7 inhibition blocks cell progression to the S-phase and leads to the accumulation of ER⁺ breast cancer cells in the G₀–G₁-phase, implying effects on CDK4/6 and CDK2. CDK7 inhibition also resulted in an accumulation of cells in the G₂–M-phase, providing functional evidence for an impact on CDK1 activity. This latter effect is augmented in CDK4/6i-resistant cells in which the cyclin D1/CDK4/6/Rb axis is disrupted. This is evidenced by the increased accumulation of the palbociclib-resistant cells in G₂–M induced by CDK7 inhibition. Our in vivo studies further support these findings. In a PDX model of ER⁺ breast cancer sensitive to SY-1365 monotherapy, CDK7 inhibition suppressed CDK1 and CDK2 phosphorylation. In a second PDX model that did not respond to SY-1365 as monotherapy, the reduction in CDK1 and CDK2 phosphorylation in response to SY-1365 was limited. Nonetheless, in this model, adding fulvestrant to SY-1365 lowered p-CDK2 (T160) and significantly suppressed tumor growth, highlighting the role of cell cycle inhibition in the response to SY-1365 and the rational for the combination of a CDK7i with a SERD. Moreover, our integrative analysis identified the E2F and G₂–M pathways as the dominant pathways affected by CDK7 inhibition.

We investigated the effect of CDK7 inhibition on c-Myc expression, since previous studies indicated that MYC silencing is a key endpoint of CDK7 inhibition (59, 60). Our results demonstrated that CDK7 inhibition decreased MYC signaling and protein expression of c-Myc in ER⁺ breast cancer cell lines. In a PDX model with relatively high expression of MYC, we detected downregulation of c-Myc protein levels in response to CDK7 inhibition. In addition, RNA-seq analyses of the PDX tumors also showed decreased expression of genes that are MYC transcriptional targets. This effect is likely due to a transcriptional effect since we detected decreased nascent MYC RNA levels at early timepoints after CDK7 inhibition; however, we did not rule out the possibility that the effects on MYC transcription are not indirect effects of CDK7 inhibition.

c-Myc is an oncogene and transcription factor that affects multiple processes such as cell proliferation, survival, differentiation, metabolism, and senescence (61). Highly relevant to the role of CDK7 as a CAK, c-Myc is a key regulator of the cell cycle. Thus, the downstream effects of MYC silencing through CDK7 inhibition likely reinforce the CDK7i-mediated inhibition of CDK2 phosphorylation and contribute to the downregulation of the expression of genes related to the cell cycle. The cell cycle effects resulting from c-Myc suppression and the effects on the cell cycle through the CAK activity of CDK7 are tightly interconnected, and most likely both contribute to the activity of SY-1365 that we observed in preclinical models. However, because c-Myc and the cell cycle are intricately connected, it is difficult to differentiate between the downstream effects of CDK7 inhibition that are mediated by the inhibition of c-Myc versus the inhibition of CDK phosphorylation and to determine what is the contribution of each of these effects.

CDK7 phosphorylates ER at S118 (16), and as expected, SY-1365 suppressed ER S118 phosphorylation in ER⁺ breast cancer cells with and without the expression of the Y537S ER mutation. Longer-term in vivo studies with PDX models, with and without an ER mutation, also demonstrated SY-1365 suppression of S118 ER phosphorylation. However, in these in vivo studies, we also observed a significant increase in total ER protein levels with SY-1365 monotherapy in the presence of WT and mutant ER that was associated with increased ER transcription and ER transcriptional targets, which likely represents a feedback response. These effects were mitigated when fulvestrant was combined with SY-1365. These observations highlight the importance of combining a SERD together with CDK7 inhibition in ER⁺ breast cancer and provide a mechanistic explanation for the additive effect that we observed with this combination in two PDX models (Fig. 3M).

In summary, our comprehensive analyses of the downstream effects of CDK7 inhibition using a selective CDK7i, SY-1365, and careful attention to the on-target activity of SY-1365, revealed that the main consequences of CDK7 inhibition are the inhibition of the CAK function of CDK7 and the transcriptional silencing of MYC (Fig. 6B). These results align with a recent study of CDK7 inhibition in multiple myeloma (62). These two effects are tightly interconnected, and this dual activity likely accounts for the potency of CDK7i. Furthermore, CDK7 emerges as a key vulnerability in palbociclib-resistant cells that have acquired RB1 loss, and we demonstrated the antitumor activity of CDK7 inhibition in models of palbociclib resistance with or without RB1 loss. Finally, we show that CDK7 inhibition triggers p53 activity and apoptosis, and intact p53 is required, at least in part, for the efficacy of CDK7i. These results imply that CDK7i has cytotoxic activity, which could be an important advantage over the cytostatic activity of ET and CDK4/6i. These results are also consistent with the exploratory biomarker analysis of the clinical trial with samuraciclib in patients with metastatic ER⁺ breast cancer. In this trial, the clinical benefit rate was higher in patients with WT TP53 compared with patients with mutant TP53 (17). Previous studies also showed that CDK7 inhibition can mediate p53-induced apoptosis. However, in these published studies, evidence of increased apoptosis was seen with either high concentrations of the CDK7i (48) or in combination with a second drug (49). This raises the possibility that this effect may be off target and requires additional investigation.

Taken together, these results support the continued development of CDK7i in combination with a SERD in ER⁺ breast cancer, particularly after the development of resistance to palbociclib. In addition, the utilization of high MYC signaling and WT-TP53 as a biomarker to select patients with ER⁺ breast cancer sensitive to this combination warrants further investigation.

C. Guarducci reports other support from Associazione Italiana Ricerca sul Cancro and American-Italian Cancer Foundation during the conduct of the study. C. De Angelis reports grants from Gilead, Daiichi Sankyo, and Novartis, as well as personal fees from Novartis, Roche, Lilly, Seagen, GSK, and Gilead outside the submitted work; in addition, C. De Angelis has a patent for PCT/US21/70543 pending. L. Malorni reports grants and personal fees from Pfizer and Novartis, as well as personal fees from Seagen, Roche, and Menarini Stemline outside the submitted work. J.S. Bergholz is currently an employee at Tango Therapeutics. J.J. Zhao reports non-financial support from Geode Therapeutics and Crimson Biopharm outside the submitted work. E. Lim reports participation in advisory board for Ellipses, Novartis, Pfizer, AstraZeneca, Gilead, MSD, and Lilly (all paid to institution); payment or honoraria for lectures, presentations, and speakers bureaus from Ellipses, Novartis, Pfizer, AstraZeneca, Gilead, MSD, and Lilly; support for attending meetings and/or travel from AstraZeneca, Novartis, Gilead, and Lilly; and research funding from Pfizer and Ellipses. R. Schiff reports grants from Breast Cancer Research Foundation and Puma Biotechnology Inc during the conduct of the study. R. Schiff also reports grants from Gilead Sciences and Puma Biotechnology Inc; personal fees from Daiichi Sankyo, MacroGenics, Binaytara Foundation, UpToDate, and UTSA/SABCS; non-financial support from Seagen and AstraZeneca; and other support from Dava Oncology, LP outside the submitted work. In addition, R. Schiff has a patent for PCT/US21/70543 (methods for breast cancer treatment and prediction of therapeutic response) pending. G.I. Shapiro reports grants and personal fees from Merck KGaA/EMD-Serono; grants from Tango Therapeutics, Bristol Myers Squibb, Pfizer, and Eli Lilly; and personal fees from Bicycle Therapeutics, Boehringer Ingelheim, Concarlo Holdings, Zentalis, Kymera Therapeutics, Janssen, Xinthera, Syros, ImmunoMet, and Blueprint Medicines outside the submitted work. In addition, G.I. Shapiro has a patent (dosage regimen for sapacitabine and seliciclib) issued to self and Cyclacel Pharmaceuticals and a patent (compositions and methods for predicting response and resistance to CDK4/6 inhibition) issued to self and Liam Cornell. R. Jeselsohn reports grants from NIH RO1 and Dunkin Donuts Breakthrough during the conduct of the study; R. Jeselsohn also reports grants and personal fees from Pfizer and Lilly, as well as personal fees from GE Health and Novartis outside the submitted work. No disclosures were reported by the other authors.

C. Guarducci: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. A. Nardone: Conceptualization, data curation, formal analysis, validation, investigation, writing–original draft, writing–review and editing. D. Russo: Formal analysis, writing–review and editing. Z. Nagy: Investigation, writing–review and editing. C. Heraud: Investigation, writing–review and editing. A. Grinshpun: Investigation, writing–review and editing. Q. Zhang: Investigation, writing–review and editing. A. Freelander: Data curation, formal analysis, investigation, writing–review and editing. M.J. Leventhal: Investigation, writing–review and editing. A. Feit: Data curation, formal analysis, writing–review and editing. G. Cohen Feit: Investigation, writing–review and editing. A. Feiglin: Formal analysis, writing–review and editing. W. Liu: Investigation, writing–review and editing. F. Hermida-Prado: Formal analysis, investigation, writing–review and editing. N. Kesten: Investigation, visualization, writing–review and editing. W. Ma: Formal analysis, investigation, writing–review and editing. C. De Angelis: Investigation, visualization, writing–review and editing. A. Morlando: Formal analysis, writing–review and editing. M. O'Donnell: Investigation, writing–review and editing. S. Naumenko: Formal analysis, visualization, writing–review and editing. S. Huang: Investigation, writing–review and editing. Q.D. Nguyen: Visualization, writing–review and editing. Y. Huang: Investigation, visualization, writing–review and editing. L. Malorni: Visualization, writing–review and editing. J.S. Bergholz: Visualization, writing–review and editing. J.J. Zhao: Visualization, writing–review and editing. E. Fraenkel: Visualization, writing–review and editing. E. Lim: Conceptualization, visualization, writing–review and editing. R. Schiff: Conceptualization, resources, supervision, funding acquisition, investigation, visualization, writing–original draft, writing–review and editing. G.I. Shapiro: Conceptualization, investigation, visualization, writing–review and editing. R. Jeselsohn: Conceptualization, resources, supervision, funding acquisition, investigation, writing–original draft, writing–review and editing.

This work was conducted with support from: NIH 5RO1CA237414-02, BARR Award Dana-Farber Cancer Institute, and Duncan Donuts Drives Cancer Discovery Award to R. Jeselsohn; American-Italian Cancer Foundation Post-Doctoral Research Fellowship 2018–2020 and an Associazione Italiana Ricerca sul Cancro fellowship and Fondazione CR Firenze #19288 year 2017 to C. Guarducci; 5U01CA253547 and the Barbara Weedon Fellowship to M.J. Leventhal; NCI CA210057 to J.J. Zhao; CPRIT Core Facility Award (RP170005) and P30 Cancer Center Support Grant (NCI-CA125123), NIH S10 instrument award (S10OD028648-01) to RPPA core facility.