Clinical Efficacy and Molecular Response Correlates of the WEE1 Inhibitor Adavosertib Combined with Cisplatin in Patients with Metastatic Triple-Negative Breast Cancer

Patients with metastatic triple-negative breast cancer (mTNBC) have few treatment options, and a median overall survival (OS) of 1–1.5 years after diagnosis (1). In the TNT and TBCRC009 trials, median progression-free survival (PFS) was 3 months on first- or second-line platinum agents (2, 3). Although median PFS increased to around 7 months in the atezolizumab and nab-paclitaxel arm of the IMpassion130 study (4), this regimen is only approved for patients with PD-L1–positive mTNBC, a small subset of patients with mTNBC (5), and most patients experience progression after a few months (5). Therefore, more effective treatment strategies are needed.

One potential therapeutic target is WEE1, a tyrosine kinase that arrests the cell cycle at the G₂ checkpoint by inhibiting cyclin-dependent kinase 1 (CDK1) and preventing entry into mitosis, thus providing an opportunity for DNA damage repair (6). WEE1 may be particularly important in p53-deficient cancers, such as mTNBC, in which TP53 mutations occur in 80% of tumors (7), because loss-of-function mutations in TP53 leave cells completely dependent on the ATR-CHK1-WEE1 axis for G₂ checkpoint control and also prevent cell-cycle arrest at the G₁ checkpoint, allowing entry into the S phase of DNA synthesis prior to repair of damaged DNA (6). WEE1 also slows cell-cycle progression through the S phase by inhibiting CDK2, stabilizing replication forks, and preventing DNA double-strand breaks (6, 8). WEE1 inhibition may cause replication stress in a p53-independent manner (8) and sensitize tumors with p53 deficiency or high genomic instability to DNA-damaging therapies (9, 10).

Adavosertib (AZD1775) is a potent and selective WEE1 inhibitor (6, 11). Preclinical studies have demonstrated that adavosertib sensitizes cells to DNA-damaging chemotherapies independent of TP53 status (12, 13). A phase I dose-escalation study of adavosertib in combination with gemcitabine, cisplatin, or carboplatin showed safety and efficacy with a numerical but not significant improvement in response rate among patients with advanced solid tumors with versus without TP53 mutations (6). Three phase II studies of adavosertib with chemotherapy demonstrated encouraging antitumor efficacy in patients with TP53-mutated or TP53-unselected ovarian cancer who were sensitive or refractory/resistant to platinum-based therapy (14–16), including significant improvements in PFS and OS as compared with chemotherapy alone (15). Subsequent trials of adavosertib in combination with docetaxel and cisplatin, gemcitabine and radiation, or irinotecan showed favorable clinical activity in patients with TP53-unselected locally advanced head/neck squamous cell carcinoma (8), pancreatic cancer (17), and relapsed refractory pediatric solid tumors (18), respectively. Single-agent adavosertib also demonstrated high activity in recurrent uterine serous carcinoma, which has frequent TP53 mutations and increased replication stress (19).

Here we report the results of a phase II study assessing the objective response rate (ORR) of adavosertib and cisplatin as first- or second-line therapy for mTNBC. We also evaluted molecular correlates of response, including genomic, transcriptomic, and tumor microenvironment characteristics.

We conducted a single-arm, two-stage phase II study of adavosertib, a WEE1 inhibitor, with cisplatin for patients with mTNBC (estrogen receptor <1%, progesterone receptor <1%, and HER2 negative by the American Society of Clinical Oncology/College of American Pathologists guidelines) treated with 0–1 prior lines of chemotherapy in the metastatic setting. Participants could not have received prior platinum chemotherapy or a prior WEE1 inhibitor. Eligible patients included those with a history of treated brain metastases, defined as no ongoing requirement for corticosteroids and no evidence of progression for ≥1 month after treatment as ascertained by clinical examination and imaging. Patients were included regardless of tumor TP53 status based on the range of mechanisms by which adavosertib may induce cytotoxicity, coupled with the promising preclinical and clinical data in TP53-unselected cancers (12, 13, 17, 18).

Treatment consisted of one cycle of cisplatin monotherapy (75 mg/m² i.v.) followed 21 days later by adavosertib 200 mg oral twice daily for five doses over 2.5 days from days 1–3 in combination with cisplatin 75 mg/m² i.v. every 21 days. The dosing schedule for adavosertib was selected on the basis of the relatively short half-life observed in the phase I clinical trial (6), as well as preclinical data suggesting that multiple doses increase combinatorial efficacy with chemotherapy without affecting tolerability (20). AstraZeneca provided funding for the study and provided adavosertib. All patients with biopsy-accessible disease underwent paired site-matched research biopsies 5–48 hours after the monotherapy cisplatin dose on cycle 1 day 1 and 5–8 hours ± 24 hours after the last dose of adavosertib on cycle 2 day 3.

Patients underwent restaging scans every 6 weeks, and response was evaluated by RECIST 1.1. All patients were enrolled at Dana-Farber Cancer Institute (DFCI, Boston, MA). The DFCI institutional review board approved the study, and participants provided written informed consent before study entry. The study was monitored by the Data Safety Monitoring Committee at Dana-Farber/Harvard Cancer Center and conducted in accordance with the ethical guidelines outlined by the Belmont Report.

Exploratory analyses evaluated the association of response with genomic, transcriptomic, and immunofluorescence profiles of on-treatment tumors (Supplementary Materials and Methods S1). Targeted panel sequencing was available on archival tumor specimens from 30 enrolled patients (81%; refs. 21, 22). To enrich for functional genomic alterations, only nonsynonymous mutations and significant copy-number alterations were included. Whole-transcriptome sequencing was performed on 45 on-study tumor biopsies from 26 treated patients on cisplatin monotherapy and/or on cisplatin plus WEE1 inhibitor combination therapy (76%; Supplementary Materials and Methods S2). Gene pathway expression was evaluated with gene set enrichment analysis (GSEA) using the hallmark gene sets from the Molecular Signatures Database (MSigDB; ref. 23), and tumor immune cell composition was assessed with CIBERSORTx (24). For these analyses, RECIST responses were categorized by the clinical benefit rate (CBR), defined as the proportion of patients achieving stable disease ≥6 months or complete or partial response. The clinical benefit group included 1 patient with an unconfirmed partial response, who had new lesions on repeat imaging.

To investigate the tumor immune microenvironment, six paired site-matched pre- and post-WEE1 inhibitor biopsies were evaluated with cyclic immunofluorescence (CyCIF; refs. 25–27). CyCIF was also used to measure WEE1 inhibitor–induced changes in cell-cycle state and DNA damage–related markers, including decreased phopho-CDK1/2 as an indication of target engagement (6, 28). Formal statistical tests for significance were not performed in the CyCIF cohort given the power limitations of the small sample size (n = 6).

The primary endpoint of the study was ORR, defined as the proportion of patients with complete or partial responses per RECIST 1.1. The study used a Simon optimal two-stage design that had 90% power to detect the difference between null (20%) and alternative (40%) ORRs with a one-sided type I error of 0.1. The null hypothesis was based on the ORR of 20% previously observed with first- or second-line platinum therapy in mTNBC without BRCA1/2 mutations (3). In the first stage, 17 patients were enrolled. Because there were at least four responses, the study continued to stage 2, where another 20 patients were enrolled. An ORR >30% would identify the regimen as worthy of further study. The analysis population consisted of all patients who initiated protocol therapy. Secondary endpoints included PFS, defined as the time from study enrollment until disease progression or death, whichever occurred first, and OS. Patients alive and without disease progression were censored at date of last disease evaluation. PFS was assessed by the Kaplan–Meier method.

Continuous molecular variables were compared between response groups (clinical benefit vs. none) with the nonparametric Wilcoxon rank-sum test, and proportion of patients with somatic alterations were compared with two-sided Fisher exact tests. Kaplan–Meier analyses evaluated the association of molecular correlates with survival. Significance testing for differences in PFS and OS were calculated using the log-rank test. All comparisons were two-sided with an alpha level of 0.05. The FDR for GSEA was controlled with the Benjamini–Hochberg method with a threshold of q < 0.05. All statistical analyses were performed in SAS 9.4 and R 3.5.1.

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

From January 26, 2017 to December 11, 2018, 37 patients enrolled in this study; 3 withdrew consent prior to therapy initiation, and 34 initiated study treatment (Supplementary Fig. S1). The median follow-up at the time of data cutoff was 32 months, and no patients remained on study therapy. Baseline characteristics are shown in Table 1. The median age was 56 years (range, 22–69 years), and only 2 patients (6%) had known BRCA1/2 mutations. Most patients (59%) had received no prior therapy for mTNBC, and the remainder had received one prior line of chemotherapy. In the early-stage setting or as first-line therapy for mTNBC, 91% of patients had been treated with taxanes and cyclophosphamide, and 79% had been treated with anthracyclines. Only 11% of patients had received prior immune checkpoint inhibitors. The median time from adjuvant therapy completion to metastatic cancer diagnosis was 12.8 months, and 50% (15/30) of the participants without de novo metastatic disease had a disease-free interval < 12 months. Most patients had visceral metastases, including liver (53%), lung (53%), and both liver and lung (32%). Among the 30 patients with targeted genomic sequencing, TP53 and ROS1 were the most frequently mutated genes with nonsynonymous mutations in 28 (93%) and 5 (17%) patients, respectively (Supplementary Fig. S2). Without matched normal samples, ROS1 variants may represent germline variants, which have previously been reported at a similar frequency in breast cancer (29).

The ORR was 26% (9/34; 95% CI, 13–44) overall (Fig. 1A), 30% (6/20; 95% CI, 12–54) in the first-line cohort, and 21% (3/14; 95% CI, 5–51) in the second-line cohort (Table 2). All responses were confirmed with subsequent repeat imaging scans. Of the three complete responders, two had lymph nodes as target lesions, which are considered normal when the short axis measures < 10 mm and so do not require a 100% reduction in lesion size to be classified as a complete response (30). Two of the complete responders had no prior chemotherapy for metastatic disease, and the third had de novo metastatic disease (Table 1). None were BRCA1/2 carriers. Of the two complete responders with targeted panel sequencing results, notable DNA repair alterations include both having TP53 mutations and one having an ATM mutation (Fig. 2A).

Median PFS was 4.9 months (95% CI, 2.3–5.7) overall (Fig. 1B), 3.8 months (95% CI, 1.3–5.6) in the first-line cohort, and 5.0 months (95% CI, 1.3–6.8) in the second-line cohort (Table 2). The CBR was 32% (95% CI, 17–51) overall, 30% (95% CI, 12–54) in the first-line cohort, and 36% (95% CI, 13–65) in the second-line cohort (Table 2). At the time of data cutoff, 29% of patients were still in survival follow-up. With the available survival data, median OS was 12.5 months (95% CI, 7.5–14.9) overall (Fig. 1C), 12.5 months (95% CI, 9.7–14.0) in the first-line cohort, and 16.7 months (95% CI, 2.3–24.4) in the second-line cohort.

All patients who initiated protocol therapy (n = 34) were evaluated for safety outcomes. Adverse events (AE) related to study therapy occurred in 31 (91%) patients, while grade 3–5 AEs occurred in 18 (53%) patients. The most common treatment-related AEs regardless of grade were nausea (50%), diarrhea (35%), anemia (29%), and neutropenia (29%; Table 3). The most common grade 3–4 treatment-related AEs were diarrhea (21%), neutropenia (18%), and anemia (12%; Table 3). One death occurred from culture-negative sepsis, possibly related to study therapy. This patient was hospitalized for abdominal pain, duodenitis, and lactic acidosis with mild transient neutropenia to absolute neutrophil count of 700, responsive to filgrastim. She subsequently developed atrial fibrillation with rapid ventricular rate complicated by persistent hypotension requiring vasopressors and respiratory failure requiring intubation.

Targeted panel sequencing of archival tumor biopsies from 30 enrolled patients showed no clear association of response with frequently mutated breast cancer driver genes (Fig. 2A; ref. 31): no single gene nonsynonymous mutation or deleterious copy-number change was significantly associated with clinical benefit (Supplementary Results; Supplementary Fig. S3), including TP53 and BRCA1/2 alterations (Supplementary Figs. S4 and S5B). This was also true of MSigDB hallmark DNA repair, E2F, and G₂-M pathway alterations: (23) none were associated with PFS or OS by the Kaplan–Meier method (Supplementary Figs. S5C and S6). However, tumors with MSigDB G₂-M pathway alterations had slightly worse PFS in unadjusted analyses (P = 0.04; Supplementary Fig. S6C).

GSEA comparing biopsies before and after WEE1 inhibitor therapy (n = 26 patients) revealed that DNA repair and cell-cycle pathways, specifically E2F and G₂-M (23), were enriched in 25 cisplatin biopsies compared with 20 cisplatin plus WEE1 inhibitor biopsies [q values <0.001; normalized enrichment scores (NES) 2.22, 2.64, and 1.81, for DNA repair, E2F and G₂-M, respectively], suggesting that adavosertib downregulates DNA repair and cell-cycle genes. Upregulation of the IFNα response pathway following WEE1 inhibitor therapy (NES −1.72, q < 0.01) was also observed. Conversely, GSEA comparing biopsies from patients with clinical benefit versus without demonstrated no differences in DNA repair or E2F/G2M cell-cycle pathways (Fig. 2B). An evaluation of previously reported WEE1 inhibitor gene expression correlates (6, 32, 33), including CCNE1, (34) showed that only EGR1 exhibited differential expression (P = 0.04) involving higher expression in patients with no clinical benefit prior to WEE1 inhibitor therapy (Supplementary Fig. S7).

However, GSEA demonstrated upregulation of immune-related pathways and downregulation of the p53 pathway in patients with clinical benefit versus without (Fig. 2B). The allograft rejection pathway was the most enriched gene set in cisplatin tumor biopsies from patients with (n = 8) versus without (n = 16) clinical benefit, while the IFNα response, IFNγ response, and allograft rejection pathways were the first, second, and fourth most enriched gene sets in tumors from patients with (n = 5) versus without (n = 14) clinical benefit after WEE1 inhibitor therapy, all with NES >1.8 and FDR q values <0.01 (Fig. 2B). A comparison of genes within the allograft rejection pathway revealed that antigen presentation genes, including HLA-DQA1, HLA-DRA, CD74 (HLA-DR), HLA-DOA, HLA-DMA, and HCLS1, had higher expression in responders than nonresponders in cisplatin biopsies (Mann–Whitney P <0.05 without FDR correction; Supplementary Fig. S8A), while T-cell infiltration genes, including CD8A, CD4, CXCL9, and CSK, had higher expression in responders than nonresponders in cisplatin WEE1 inhibitor biopsies (Mann–Whitney P <0.05 without FDR correction; Supplementary Fig. S8B).

Tumor infiltrating lymphocytes (TIL) proportions inferred from RNA-sequencing data did not differ by clinical benefit group at either biopsy timepoint (Supplementary Fig. S9). While TILs increased after WEE1 inhibitor treatment in 4/5 patients with clinical benefit, this difference was not significant (paired Wilcoxon P = 0.1; Fig. 2C). Memory CD4⁺ T cells were significantly increased after WEE1 inhibitor treatment in 18 patients with paired biopsies (paired Wilcoxon P = 0.01) with no difference by clinical benefit (Supplementary Fig. S10). Antitumor M1 macrophages trended toward being higher in post-WEE1 inhibitor biopsies from patients with clinical benefit (Wilcoxon P = 0.09; Supplementary Fig. S11).

Tumor immunostaining studies were undertaken to explore the effects of WEE1 inhibitor therapy on immune infiltration and DNA damage markers. Stromal TILs measured using the International Immuno-Oncology Biomarker Working Group criteria (35) in six available paired biopsies demonstrated a nonsignificant increase from a median of 1% to 7.5% following WEE1 inhibitor treatment (paired Wilcoxon P = 0.1; Supplementary Fig. S12). IHC for DNA damage markers showed positive RAD51 and negative phospho-histone H3 (pHH3) staining in biopsies following cisplatin that switched to negative RAD51 and positive pHH3 staining in some biopsies after cisplatin WEE1 inhibitor treatment, as well as positive geminin and γ-H2AX staining in both biopsies (Supplementary Fig. S13; Supplementary Table S1). While reduced RAD51 suggested suppressed homologous recombination, the small number of paired biopsy samples with high-quality IHC data cannot establish whether or not RAD51 is a predictive biomarker.

Using CyCIF, we stained tumor sections with 24 different tumor, stromal, and immune cell markers to characterize six of the paired site-matched pre- and post-WEE1 inhibitor biopsies (Supplementary Table S2). Mean marker fluorescence intensity in single cells across all samples revealed changes between monotherapy and combination therapy (Fig. 3A). CyCIF also demonstrated the expected changes in cell cycle and DNA damage markers with WEE1 inhibitor therapy. In tumor cells, defined as pan-cytokeratin-positive (panCK⁺) cells, CyCIF showed that phospho-CDK1/2/3/5, the target for WEE1, decreased on average by 56% in combination therapy versus monotherapy biopsies (Fig. 3B; Supplementary Fig. S14; Supplementary Table S3). The number of positive pHH3 tumor cells increased by an average of 28-fold from monotherapy to combination therapy biopsies (Fig. 3B; Supplementary Fig. S14; Supplementary Table S3), suggesting increased entrance into mitosis following WEE1 inhibitor therapy. These data serve as pharmacodynamics evidence of WEE1 engagement by adavosertib (6, 28).

CyCIF analysis of immune cells demonstrated increased intratumoral T-cell infiltration following WEE1 inhibitor therapy in patients having more favorable responses. The percent of panCK-negative cells positive for T-cell markers (CD3⁺CD4⁺ or CD3⁺CD8⁺) increased from monotherapy to combination therapy biopsies in 5/6 patients (Fig. 3B). In addition, CD3⁺CD4⁺ and CD3⁺CD8⁺ T-cell numbers exhibited larger increases in patients with partial response or stable disease versus progressive disease: the number of CD3⁺ CD4⁺ and CD3⁺CD8⁺ cells increased by an average fold change of 12.4 and 13.6 versus 4.0 and 3.0, respectively (Supplementary Table S3). Conversely, there was no clear change in the expression of macrophage markers CD68 and CD163 across biopsy timepoints or response groups (Fig. 3B and C; Supplementary Table S3).

This phase II study combining adavosertib and cisplatin in patients with mTNBC, the first to our knowledge, demonstrated an ORR of 26%, which missed the prespecified ORR cutoff of 30%, and a median PFS of 4.9 months. This median PFS compares favorably to prior trials of first-line platinum chemotherapy in this patient population, such as the TNT and TBCRC009 trials that showed a median PFS of 3 months for first- or second-line platinum therapy (2, 3). Notably, our cohort included many patients with early relapse, so the median disease-free interval of 13 months in the present trial was shorter than the median disease-free interval of 21 months in the historical control TBCRC009 trial by which our study was powered (3), raising the question of whether adavosertib had benefit beyond cisplatin in our more treatment-refractory population. In contrast to some prior studies (6, 12), we did not observe differential response rates by TP53 alteration status perhaps due to our use of combination therapy or the small number of tumors without a TP53 alteration. Despite a previous report of responses to adavosertib monotherapy in some patients with germline BRCA mutations (28), genomic alterations in DNA repair and cell cycle pathways, including BRCA1/2 alterations, did not correlate with response, consistent with results from the randomized NCI mpact trial (36). Future studies of WEE1 inhibitors with more comprehensive genomic assays will likely elucidate the role of genomic instability in response to WEE1 inhibition in clinical mTNBC cohorts.

CyCIF on paired site-matched pre- and post-WEE1 inhibitor biopsies demonstrated decreased phosphorylation of CDK1/2/3/5 in patients treated with combination therapy, as well as increased staining for the mitotic marker pHH3. These anticipated changes in cell-cycle state provide pharmacodynamics evidence of successful target engagement. Likewise, gene expression analyses of the pre- and post-WEE1 inhibitor biopsies found that DNA repair and cell-cycle pathways were downregulated following WEE1 inhibitor treatment, consistent with the mechanism of action of adavosertib.

The gene expression results also showed enriched p53 gene pathway expression in patients with no clinical benefit but did not reveal an association of DNA repair or other cell-cycle genes with response. In line with prior reports demonstrating that EGR1 expression decreases with higher doses of WEE1 inhibitor therapy (6, 32). EGR1 was found to have higher pre-WEE1 inhibitor expression in patients with no clinical benefit, suggesting that adavosertib may not have adequately suppressed EGR1 levels in these nonresponding tumors. Our results did not detect a role for other previously reported WEE1 inhibitor response correlates, such as CCNE1 and G₁–S regulatory genes (33, 34), perhaps due to the greater heterogeneity of clinical samples compared with the preclinical experimental conditions of these prior reports.

However, our gene expression data showed upregulation of immune response pathways in patients with clinical benefit both before and after WEE1 inhibitor treatment. This was corroborated by the multiplex CyCIF finding of greater T-cell infiltration following WEE1 inhibitor therapy in tumors with more favorable responses. Further studies will be needed to elucidate the mechanism of increased T-cell infiltration following WEE1 inhibition. On the basis of our results with pHH3 immunofluorescence and γ-H2AX IHC, WEE1 inhibition facilitated entry into mitosis with ongoing evidence of DNA damage. These biological changes may be associated with micronucleation and ultimate activation of the cGAS/STING pathway, which is expected to result in a type I IFN response and the production of cytokines that drive cytotoxic T-cell infiltration and activation (37, 38). These results also raise the question of whether the favorable response of tumors with enriched immune pathway gene expression is specific to adavosertib and cisplatin, more generalizable to a number of different therapies, or indicative of better overall prognosis. Beyond requiring validation in future larger data sets of WEE1 inhibitor-treated tumors, these findings highlight the need to evaluate the role of immune pathways in response to other nonimmune therapies. If confirmed in larger prospective studies, the additional finding of IFNα gene expression enrichment following WEE1 inhibitor therapy raises the questions of whether PD-L1 expression may be induced in adaptation and whether WEE1 inhibitors may synergize with immunotherapy. A clinical trial evaluating the combination of adavosertib with the PD-L1 inhibitor durvalumab is ongoing (NCT02617277).

The AEs observed in this study were similar to those observed in previous studies combining adavosertib with chemotherapy, consisting most commonly of bone marrow and gastrointestinal toxicity (6, 8, 14, 17, 18). The frequency of serious AEs ≥ grade 3 was also similar in the prior combination studies, occurring in about half of all participants (6, 8, 14, 17, 18), although a prior possible treatment-related death had not previously been reported. The first selective WEE1 kinase degrader, generated by linking adavosertib to the cereblon-binding ligand, pomalidomide, recently demonstrated G₂–M accumulation at lower doses than adavosertib in cancer cell lines (39), raising the question of whether such novel WEE1 kinase degraders will show lower dose-limiting AEs and thereby greater efficacy in clinical studies.

A strength of this study is its deep multimodal molecular characterization of paired site-matched biopsies pre- and post-WEE1 inhibitor. However, the study is limited by its small sample size that curtails the power of correlative analyses to find significant differences between response and biopsy groups. Furthermore, its nonrandomized design may have introduced biases, such as a healthier enrolled study population than historical control studies of platinum monotherapy used to estimate the null hypothesis and response rate cutoff. In addition, the use of archival biopsies for the targeted next-generation sequencing assessments raises the possibility that our lack of genomic correlates of response, including TP53 alterations, may be due to an incomplete assessment of the mutation status of the tumors at the time of protocol therapy rather than a true absence of an association. Finally, our ability to isolate the effect of adavosertib in our paired biopsy comparisons may have been affected by the slightly different timing of the biopsies in relation to cisplatin, and differential gene expression between these two timepoints may partially be due to different exposure periods to cisplatin.

In conclusion, among patients with mTNBC treated with 0–1 lines of prior therapy, the combination of adavosertib and cisplatin did not meet the prespecified efficacy cutoff of ORR >30% required to show a statistically significant improvement in response rate compared with cisplatin monotherapy as defined by this study. This result may be due to small sample size, nonrandomized study design, or true lack of benefit. Similar to prior combination studies of WEE1 inhibitors and chemotherapy, about half of patients experienced serious AEs (6, 8, 14, 17, 18). Ultimately, the recent approval of immune checkpoint inhibitors for patients with treatment-naïve PD-L1–positive mTNBC makes WEE1 inhibitor combinations most relevant in PD-L1–negative populations and second-line treatment settings. Alternatively, the combination of WEE1 inhibition with chemotherapy and PD-1/L1 inhibitors in the first-line mTNBC setting may be important if future studies confirm our finding of an augmented immune response in WEE1 inhibitor-treated tumor biopsies and show that the association of enriched immune signaling in responders is specific to WEE1 inhibition rather than general treatment response.

T.E. Keenan reports grants from AstraZeneca during the conduct of the study and grants from Bristol Myers Squibb, Merck, Immunomedics, and Cyclacel outside the submitted work. T. Vallius reports grants from Finnish Medical Foundation, Relander Foundation, Turku University Foundation, Maud Kuistila Memorial Foundation, Cancer Society of Southwest Finland, and Finnish Society of Oncology during the conduct of the study. J.L. Guerriero reports grants and personal fees from GlaxoSmithKline; grants from Eli Lilly and Pfizer; and personal fees from Codagenix, Verseau, and Kymera outside the submitted work. M.X. He reports grants from NIH and National Science Foundation during the conduct of the study and personal fees from Amplify Medicines outside the submitted work. R.E. Godin reports other from AstraZeneca during the conduct of the study and other from AstraZeneca outside the submitted work. P.K. Sorger reports personal fees from RareCyte Inc and Nanostring and personal fees and other from Glencoe Software during the conduct of the study, as well as personal fees from Applied Biomatch outside the submitted work. A. D'Andrea reports nonfinancial support from AstraZeneca; personal fees from Bayer AG, LAV Global Management Company Limited, L.E.K. Consulting, Patheon Pharmaceuticals, Pfizer, and Tango Therapeutics; personal fees and other from Cedilla Therapeutics, Cyteir, Ideaya, Inc., and Impact Therapeutics; other from Epizyme, Inc. and GlaxoSmithKline; grants and other from Lilly Oncology; and grants and personal fees from Merck KGAa/EMD Serono outside the submitted work. E.P. Winer reports grants from AstraZeneca, Jounce, Leap, Lilly, Novartis, Seattle Genetics, Syros, and Zymeworks; personal fees from Carrick Therapeutis, G1 Therapeutics, GSK; and grants and personal fees from Genentech/Roche and Genomic Health outside the submitted work. E.A. Mittendorf reports personal fees from Merck, Genomic Health, Roche/Genentech, and Sellas LifeSciences; other from Glaxo Smithkline; and personal fees from Physician Education Resource outside the submitted work. E.M. Van Allen reports personal fees from Tango Therapeutics, Genome Medical, Invitae, Monte Rosa Therapeutics, Manifold Bio, Illumina, and Enara Bio and grants from Novartis and BMS outside the submitted work, as well as a patent for Institutional patents filed on chromatin mutations and immunotherapy response, and methods for clinical interpretation pending. G.I. Shapiro reports grants from AstraZeneca during the conduct of the study, as well as grants and personal fees from Eli Lilly, Merck KGaA/EMD-Serono, Sierra Oncology, and Pfizer; grants from Merck & Co.; and personal fees from G1 Therapeutics, Roche, Bicycle Therapeutics, Fusion Pharmaceuticals, Cybrexa Therapeutics, Astex, Almac, Ipsen, Bayer, Angiex, Daiichi Sankyo, Seattle Genetics, Boehringer Ingelheim, ImmunoMet, Asana, Artios, Atrin, Concarlo Holdings, Syros, Zentalis, and CytomX Therapeutics outside the submitted work; in addition, G.I. Shapiro has a patent for Dosage regimen for sapacitabine and seliciclib issued to Cyclacel Pharmaceuticals and a patent for Compositions and methods for predicting response and resistance to CDK4/6 inhibition pending to Liam Cornell and G.I. Shapiro. S.M. Tolaney reports grants from AstraZeneca during the conduct of the study, as well as personal fees from AstraZeneca, Puma, Daiichi Sankyo, Silverback Therapeutics, G1 Therapeutics, Athenex, OncoPep, Kyowa Kirin Pharmaceuticals, Samsung Bioepsis Inc., CytomX, Gilead, Mersana Therapeutics, and Certara; grants and personal fees from Eli Lilly, Merck, Novartis, Nektar, Pfizer, Genentech/Roche, Immunomedics, BMS, Eisai, Nanostring, Sanofi, Odonate, and Seattle Genetics; and grants from Exelixis and Cyclacel outside the submitted work. No disclosures were reported by the other authors.

T.E. Keenan: Conceptualization, data curation, software, formal analysis, methodology, writing-original draft, writing-review and editing. T. Li: Data curation, software, formal analysis, methodology. T. Vallius: Data curation, software, formal analysis, methodology, writing-original draft, writing-review and editing. J.L. Guerriero: Data curation, software, formal analysis, methodology, writing-original draft, writing-review and editing. N. Tayob: Data curation, software, supervision, methodology. B. Kochupurakkal: Data curation, software, formal analysis, methodology. J. Davis: Data curation, software, methodology. R. Pastorello: Data curation. R.K. Tahara: Conceptualization. L. Anderson: Data curation. J. Conway: Software, methodology, writing-review and editing. M.X. He: Software, methodology, writing-review and editing. E. Shannon: Data curation. R.E. Godin: Funding acquisition, writing-review and editing. P.K. Sorger: Resources, supervision, writing-review and editing. A. D'Andrea: Resources, supervision. B. Overmoyer: Data curation. E. Winer: Resources, supervision, funding acquisition. E.A. Mittendorf: Resources, supervision, funding acquisition, writing-review and editing. E.M. Van Allen: Conceptualization, resources, supervision, funding acquisition, writing-original draft, writing-review and editing. G.I. Shapiro: Conceptualization, resources, supervision, funding acquisition, writing-original draft, writing-review and editing. S.M. Tolaney: Conceptualization, resources, supervision, funding acquisition, writing-original draft, writing-review and editing.

This work was supported by AstraZeneca, the Mehlman Family Fund, the Harvard Ludwig Center, NCI grant U54-CA225088, and the NCI Specialized Program of Research Excellence (SPORE) Grant 1P50CA168504. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH/NCI. T.E. Keenan acknowledges grant support from T32CA009172. T. Vallius acknowledges grant support from the Finnish Medical Foundation, Relander Foundation, Turku University Foundation, Maud Kuistila Memorial Foundation, the Finnish Society of Oncology, and the Cancer Society of Southwest Finland. E.A. Mittendorf acknowledges the Rob and Karen Hale Distinguished Chair in Surgical Oncology for support. The authors thank Kate Bifolck for her editorial support to this work. The authors also acknowledge the DFCI Oncology Data Retrieval System (OncDRS) for the aggregation, management, and delivery of the genomic research data used in this project.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.