Molecular Residual Disease in Breast Cancer: Detection and Therapeutic Interception

Over the past several decades, there have been great strides in improving outcomes in breast cancer. Screening has led to earlier diagnoses, and advances in surgery, radiation, and medical therapeutics have led to improved overall outcomes. Despite these advances, breast cancer remains the second leading cause of cancer-related death in women in the United States and a leading cause of cancer-related death in women globally (1). This discrepancy is in part due to the vast heterogeneity of disease behavior and varying risks of recurrence. After surgery, adjuvant therapies are generally applied to minimize this risk; these measures may include radiation, chemotherapy, endocrine therapy, immunotherapy, HER2-directed therapies, cyclin-dependent kinase 4/6 (CDK 4/6) inhibition, and/or PARP inhibition. An important limitation surrounding the decision to use adjuvant therapy, however, is the fact that it is based upon extrapolation of a patient's recurrence risk at a single point in time, at the time of resection. There is not a standard method to monitor for recurrence or persistence of microscopic disease after surgery and/or adjuvant therapy (2). As a result, some patients who may benefit from additional adjuvant therapy are undertreated, and similarly some patients who may not need systemic therapy are overtreated. This unmet clinical need has created a burgeoning field in solid tumor oncology: the detection and management of molecular residual disease (MRD) after curative treatment.

Adjuvant treatment decisions are based on breast cancer subtypes, along with clinical, pathologic, and/or genomic risk factors assessed at the time of diagnosis and/or surgery (3–7). Patients may receive various therapies, all of which are given based on metrics assessed at the time of surgery. Subsequent surveillance consists of annual mammograms in any remaining breast tissue and otherwise, no blood-based or radiographic studies are undertaken unless there is a clinical finding that prompts imaging (8–10). The tumor markers CA 15–3 (CA 27.29) and carcinoembryonic antigen (CEA) are not exclusively produced by cancer cells and have low sensitivity (40.3%–88.3%) and specificity (46.2%–47.3%), which are further decreased in low disease states, and their half-lives are days to weeks (11, 12).

There is an unmet need for better assessment of MRD given concern over both overtreatment and undertreatment in the adjuvant setting, and new assays allow for the sensitive detection of circulating tumor DNA (ctDNA) and strongly predict disease recurrence (13–15). At present, clinical trials are underway to determine whether intervening in the setting of detectable MRD improves the risk of disease recurrence. Ideally, patients with MRD would receive targeted care to minimize this high recurrence risk. On the other hand, many patients may receive prolonged adjuvant therapy, with the accompanying risks of side effects, despite having been cured. Similarly, MRD has not yet been used as a tool for therapy deescalation; this is a goal for future applications. As such, there is considerable interest in the detection of MRD via ctDNA and the potential interventions to help achieve cure or deep control prior to overt metastatic breast cancer (MBC); these data will be discussed in this review article.

The concept of molecular/minimal residual disease was initially established in hematologic malignancies such as leukemia, where the cancer resides in the bloodstream and bone marrow, and detection is central to diagnosis and evaluation for treatment response. After a patient receives induction chemotherapy, a bone marrow biopsy is performed to evaluate for persistent leukemia, or MRD, which would indicate a need for escalated therapy (16–18). Similarly, detectable MRD after consolidation chemotherapy would also mandate more aggressive therapy. Thus, MRD serves as a risk stratification tool in hematologic malignancies and is used to inform the decision of whether to escalate therapy. Although early technology limitations had previously not allowed for the application of this concept in solid tumors, new advances now allow us to study MRD via ctDNA in this disease type, including breast cancer.

Unlike leukemia, breast cancer does not reside in a continuous compartment like blood or bone marrow, and the area of disease is resected entirely. Blood does, however, serve as a reservoir in which evidence of persistent or recurrent breast cancer may reside. As cells die, they shed small amounts of DNA, called cell-free DNA (cfDNA). If there is tumor DNA within the cfDNA content, this is called ctDNA. The half-life of ctDNA is under 2 hours, which allows it to function as an up-to-date measure of tumor activity (19). Thus, ctDNA detection after tumor resection and/or adjuvant therapy is suggestive of incomplete cure and risk of recurrence (Fig. 1).

Multiple studies have since shown that detection of ctDNA after surgery, either immediately or on subsequent serial sampling, is predictive of early relapse and an overall poorer prognosis; furthermore, ctDNA detection has shown a median lead time of 7.9 to 18.9 months prior to clinical recurrence of breast cancer (20–24). With the lack of routine imaging in breast cancer, the lead time over radiologic relapse has not been established in most studies. Of note, detection of higher levels of circulating tumor cells and specific disseminated tumor cells may also be classified as MRD, though these have unique limitations, including a need for more tumor content and invasive analysis, respectively. Thus, while these are additional tools for MRD evaluation, their detection is currently less clinically feasible than ctDNA at this time, and they will not be discussed in detail in this review (25, 26).

Although ctDNA as a concept has been known for many years, detection in early breast cancer has only recently been unfolding (27–30). One major issue is the overall ctDNA content is partially proportional to the tumor burden. While in MBC, ctDNA may make up almost the entire cfDNA content, levels in localized breast cancer may be substantially lower, frequently at 0.1% or lower, and the content in MRD is even smaller (13, 28). The detection of MRD would likely be clinically meaningful only if it could be identified prior to radiographic and/or clinical evidence of disease recurrence. This fact mandates exquisite sensitivity at very low allele frequencies.

Broadly speaking, there are two approaches to ctDNA evaluation in MRD: tumor-informed and tumor-agnostic detection (Fig. 2).

• (i) Tumor-agnostic detection: in the advanced cancer setting, Foundation One CDx and Guardant 360 are both FDA-approved ctDNA assays that use targeted sequencing of known cancer-associated mutations via hybrid capture and next-generation sequencing (NGS). These assays are ideal when the ctDNA content is large enough for targeted sequencing and/or whole-exome sequencing (WES) and are most effective in metastatic disease (31). These advanced cancer genotyping assays do not have a role in the MRD setting, given the need for sufficient sensitivity and specificity for potential clinical use. More recent tumor-agnostic assays have been designed for use in the low-ctDNA MRD setting. The Guardant Reveal assay uses cancer-specific methylation signatures alongside genomic data to identify tumor-agnostic MRD in ctDNA, and in colon cancer has a reported sensitivity of 55% to 91%, where higher sensitivity is observed after serial sampling, and a specificity of 100% (32, 33). This assay assesses both genomic and epigenomic regions and is currently being studied in the Observation of ResiduAl Cancer with Liquid biopsy Evaluation (ORACLE) study (NCT05059444), which includes 1,000 patients with localized cancer after curative treatment; the study includes 11 solid tumors, including breast cancer, and is scheduled to complete in 2027 (34).

• (ii) Tumor-informed detection: tumor-informed assays increase sensitivity in a low ctDNA context by assessing for a known set of tumor-associated alterations (31, 35). Instead of initially obtaining a comprehensive genomic snapshot, these assays prioritize a “yes/no” answer of detectability. A tumor-informed approach involves taking an excised tumor and performing ultradeep sequencing to identify alleles specific to that exact tumor and distinct from other cfDNA. Once these alleles are identified, individualized primers or probes are designed on the basis of the tumor-specific mutations and used in subsequent MRD blood testing. Sensitivity is further increased by serial sampling and by evaluating for multiple alleles (13, 14, 20). The Natera Signatera assay has received FDA breakthrough designation and uses a tumor-informed PCR-based approach. A patient's primary tumor undergoes ultradeep WES, and personalized primers are developed to target up to 16 variants. This assay can evaluate for single-nucleotide variants and short indels, and when more than 16 are identified, prioritization is based upon variant allele frequency (VAF) and context of the mutation in the setting of the cancer. To distinguish somatic clonal variants from germline mutations, the patient's white blood cells also undergo matched WES (Fig. 2C). This technique also allows for distinguishing mutations due to clonal hematopoiesis of indeterminate potential (CHIP), a key confounder in ctDNA analysis (36).

• Coombes and colleagues used the Signatera assay to serially evaluate 49 patients with early-stage breast cancer after surgery and adjuvant therapy. Out of the 18 patients who relapsed, 16 had detectable ctDNA with a sensitivity of 55.6%, increased to 88.9% on serial testing. All presented with MRD prior to clinical progression, with a median lead time of 8.9 months (0.5–24 months). At the time of publication, there were 31 patients who had not relapsed, and none had evidence of MRD (specificity 100%; ref. 13).

• (iii) Combined approaches: combining tumor-informed detection with tumor-agnostic deep sequencing may capitalize on the strengths of each assay. As proof of concept, Garcia-Murillas used digital PCR (dPCR) to detect early breast cancer recurrence, and subsequent plasma-based sequencing in five patients predicted the genomic makeup of the metastatic disease with better accuracy than genomic profiling of the primary tumor. Given that dPCR detects only a target allele, this was used as an initial detection tool, and once patients had detectable MRD, the plasma subsequently underwent high-depth targeted capture-based massively parallel sequencing (24).

There are other technologies currently in development.

• (i) The residual disease and recurrence (RaDaR) assay: The RaDaR assay is a tumor-informed PCR-based assay where personalized primers and matched buffy coat genotyping are used to distinguish clonal hematopoiesis of indeterminate potential (CHIP) from tumor mutations. This assay was validated in a retrospective pilot study of early-stage breast cancer, where the assay detected ctDNA as low as 0.0029%, which strongly associated with distant recurrence, with 12 of 13 distant recurrences with positive ctDNA (92%; ref. 37). In a separate retrospective study of early-stage breast cancer, MRD was detected in 17 of 17 (100%) patients that relapsed and in 0 of 5 patients that did not. These above studies have been presented as abstracts in major oncology meetings (37, 38). A large number of variants were sequenced, with a median of 41 (range 10–61) at a depth of 100,000× per locus. Interestingly, patients with brain-only metastases also had MRD detected on this assay; detection of central nervous system (CNS) disease has been a historic limitation of plasma-based assays (39, 40). VAF was very low, ranging from 0.0007% to 1.3%. Median lead time was just over 1 year, with a maximum of 2 years (38).

• In the phase II Oxel study evaluating capecitabine, nivolumab, or the two drugs in combination as adjuvant therapy for triple negative breast cancer (TNBC), patients underwent MRD analysis via RaDaR, and 12 of 33 (36%) of patients were found to have MRD. Three patients had MRD clearance on serial testing, and one of these three recurred. The remaining 9 patients with MRD also experienced recurrence. Twenty-one patients were MRD negative at baseline, and one patient's samples turned positive after 6 weeks. Of the 12 patients who had ctDNA measured and had distant recurrence, 10 were MRD positive at baseline and an additional patient's sampling turned positive after 6 weeks. Baseline MRD was positive at a VAF range of 0.0012% to 3.6% and was associated with inferior distant disease free survival (DFS) at a median of 4.0 months versus not reached (P < 0.0001) and overall survival (OS; median not reached for either group; P = 0.0048; ref. 41).

• (ii) Invitae Personalized Cancer Monitoring: Invitae PCM is a PCR-based tumor-informed assay that reports a flexible design, wherein between 18 and 50 variants are tracked after amplicon design. The assay has been studied in stage II–III breast cancer and shown to be associated with a median lead-time to clinical relapse of 13.7 months with detection of ctDNA in 10/13 patients that experienced relapse and no patients that had not relapsed by a median follow-up of 52.7 months (42). This assay is currently being used in the TRAK-ER trial (NCT04985266), which is evaluating the impact of fulvestrant and palbociclib in patients with ER⁺/HER2⁻ breast cancer, detectable ctDNA, and no overt metastatic disease.

• (iii) NeXT Personal: This assay combines a tumor-informed approach as well as surveillance for a panel of known clinically significant variants. This assay reports detection of stage II–IV cancer down to a limit of detection of 1 part per million, around 1×10⁻⁶ allele fraction (43). This platform uses tumor and normal whole-genome sequences (WGS), rather than WES, to design the bespoke ctDNA panels. The latter is made up of up to 1,800 somatic tumor variants in order to maximize sensitivity. In testing in healthy donor plasma, the specificity of the assay was reported to be 100% (44). A notable distinction of this assay is that the tumor-informed approach uses WGS rather than WES.

• (iv) CAPP-Seq: Other assays under development utilize hybrid capture-based next-generation sequencing (NGS). For example, the Cancer Personalized Profiling by deep-Sequencing (CAPP-Seq) assay uses biotinylated probes against known mutations from The Cancer Genome Atlas (TCGA) and COSMIC and reports a detection limit of 0.02%. By also incorporating barcoding and error correction via polishing to account for artifacts, they report a limit of 0.001% (45). This technique has been validated in lung cancer, diffuse large B-cell lymphoma, and bladder cancer but not breast cancer at the time of this review's publication. It is commercially in use via the Avenio system (46). A similar technique using targeted error correction sequencing (TEC-Seq) was developed by Phallen and colleagues and also uses hybrid capture as the assay's central mechanism (47).

• (v) MAESTRO: Minor allele enriched sequencing through recognition oligonucleotides (MAESTRO) utilizes massively parallel mutation enrichment and short probe panels in duplex sequencing to assess for a larger number of mutations (up to 10,000) at very low frequencies using a technique that minimizes excess sequencing of normal DNA. By using short probes in this hybridization-based capture technique, this approach reports a need for up to 100-fold fewer reads. Ostensibly, this approach would be effective in WES- or WGS-based tumor-informed MRD detection. This technique has been retrospectively assessed in samples from patients with breast cancer and shown increased MRD detection, given the feasibility of tracking a larger number of unique variants with an improved signal-to-noise ratio (48).

• (vi) PhasED-Seq: Kurtz and colleagues describe a technique for utilizing two or more mutations that occur in cis, termed phased variants, in a technique named PhasED-Seq. The purpose of this technique is to improve sensitivity and specificity at very low tumor fractions. This assay also uses WGS rather than WES and has been studied primarily in B-cell lymphomas, where phased variants are more commonly described, but this technique has been applied to six solid tumors, including one breast cancer (31, 49). In its application in solid tumors, the assay is used in a tumor-informed approach, where tumors undergo upfront WGS from paired tumor/normal samples to first identify potentially trackable phased variants, and a bespoke panel is subsequently designed. In these six solid tumors, multiple phased variants were identified, and the background rate of noncandidate variants was found to be lower than in the case of SNVs, and 6 additional samples at very low tumor fractions were found to have MRD where prior SNV-based assays had returned negative MRD results (49).

• (vii) Grail Galleri: Epigenetics is becoming an area of increased focus in cfDNA assays, particularly in the setting of very low ctDNA content, with interest regarding the efficacy of testing for methylation signatures in patients with little to no ctDNA. This approach is the cornerstone technology for the Grail Galleri assay, which is currently evaluating a generalized cancer screening test on the basis of cfDNA methylation patterns in patients without known cancer (NCT05155605; ref. 50).

Multiple studies have shown that MRD detection precedes clinical metastatic recurrence with the lead time increasing as the sensitivities of the assays increase (Table 1). In 2015, one of the first studies to show the clinical implications of MRD in breast cancer demonstrated that ctDNA detection after surgery predicted relapse with a median lead time of 7.9 months. WES of the primary tumor was used to design digital PCR (dPCR) probes, which bound to tumor-specific targets in plasma sampling to detect analytes at a VAF of 0.1%. On the basis of massive parallel sequencing on the primary breast tumor, at least one somatic mutation was noted in 43 of the 55 (78%) patients prospectively studied. Detectable MRD was associated with an HR of 25.1 [95% confidence interval (95% CI), 4.08–130.5; P < 0.0001]. Sensitivity was initially 50% and increased to 80% on serial testing, while specificity was around 96% (20). In a subsequent prospective, multicenter validation study in 101 women with early-stage breast cancer, ctDNA detection on serial plasma samples predicted relapse with a median lead time of 10.7 months (95% CI, 8.1−19.1 months) at a sensitivity of 75% and specificity of 92%, with the exception of brain-only metastases. Brain-only metastases were detected via ctDNA in only 1 of 6 patients (17%), suggesting that the blood–brain barrier may disrupt the otherwise continuous repository of cfDNA in nonprotected parts of the body (24). Olsson and colleagues used ctDNA-based cancer rearrangements to retrospectively show 86% of patients with eventual metastases had detectable ctDNA prior to recurrence, with an average lead time of 11 months (range, 0–37 months) and a serial sensitivity of 93% and specificity of 100% (23). Parsons and colleagues used a tumor-informed approach to serially evaluate for up to 488 mutations and found MRD detection at 1 year in stage 0–III breast cancer was strongly associated with distant recurrence (HR 20.8), with a median lead time of 18.9 months (range 3.4–39.2; ref. 14). Lipsyc-Sharf and colleagues prospectively showed that in patients with high-risk hormone receptor positive (HR+) breast cancer in the late adjuvant setting, ctDNA was detectable a median of 1 year before clinical metastasis (15).

Studies have also shown that residual disease after neoadjuvant therapy is associated with higher risk of recurrence. Radovich and colleagues showed that ctDNA and/or circulating tumor cell (CTC) detection after neoadjuvant chemotherapy in TNBC was associated with inferior DFS (HR, 2.67; 95% CI, 1.28–5.57; P = 0.009) and inferior OS (HR, 4.16; 95% CI, 1.66–10.42; P = 0.002; ref. 21). Riva and colleagues designed dPCR probes against TP53 mutations and found that persistent ctDNA positivity after 1 cycle of therapy was associated with worse DFS (P = 0.001) and OS (P = 0.006; ref. 51). Furthermore, in the adaptive I-SPY 2 trial, Magbanua and colleagues reported evidence that lack of ctDNA clearance was strongly associated with recurrence risk (HR, 10.4; 95% CI, 2.3–46.6; ref. 52).

Multiple studies are now underway evaluating interventions to change the above natural history. The c-TRAK-TN trial is a multi-center phase II trial evaluating the efficacy of pembrolizumab in the setting of MRD after treatment for early-stage TNBC. This study used tumor-informed sequencing and serial sampling at 3-month intervals for 12 months. Among 208 patients enrolled, 161 entered ctDNA surveillance, and 45 were entered into the pembrolizumab intervention. Of note, over half of patients with MRD in fact had metastatic disease on staging scans at the time of ctDNA detection, which highlights an important area of variation and limitation in MRD analyses; breast cancer does not undergo regular radiographic surveillance besides in residual breast tissue, so the distinction between clinical and radiographic recurrence remains to be further assessed. Among a small number of patients (n = 5) who started pembrolizumab for detectable MRD, none experienced prolonged ctDNA clearance, and 4 recurred; further study is underway (53).

The important question remains as to whether therapeutic intervention impacts the poor prognosis associated with MRD. Multiple studies in the metastatic setting have shown changes in ctDNA alleles can predict response (35, 54–60). As such, there are multiple observational and interventional clinical trials currently underway in the adjuvant treatment setting using MRD. Several major interventional trials are highlighted below based on drug class:

• (i) CDK 4/6 inhibitors: the ongoing LEADER trial (NCT03285412) at our institution is a phase II study currently evaluating the impact of adding ribociclib to adjuvant endocrine therapy in patients with resected early-stage HR⁺ breast cancer who have detectable MRD (Fig. 3). The DARE study is similarly evaluating the effect of adding palbociclib to patients with MRD (NCT04567420). Trak-ER (NCT04985266) is a two-part trial where patients with high-risk clinical features and MRD are randomized to standard endocrine therapy versus palbociclib + fulvestrant.

• (ii) PARP inhibitors: the ZEST trial (NCT04915755) is evaluating the efficacy of adjuvant niraparib in stage I–III HER2⁻ disease; patients with HR⁺ breast cancer must also have a BRCA mutation, while in the TNBC arm, patients need only to be MRD⁺.

• (iii) Antibody–drug conjugates: the ASPRIA study (NCT04434040) is assessing the addition of the programmed death cell ligand 1 (PD-L1) inhibitor atezolizumab in combination with the antibody–drug conjugate (ADC) sacituzumab govitecan in patients with TNBC and positive MRD.

• (iv) Immunotherapy: the c-TRAK-TN trial (NCT03145961) is evaluating pembrolizumab given in the setting of high risk TNBC, as described above. The ASPRIA trial above is combining immunotherapy with an ADC, and the PERSEVERE trial below has multiple arms, which include the use of immunotherapy.

• (v) Basket Trial: the PERSEVERE (NCT04849364) trial is using ctDNA to inform postneoadjuvant therapy in patients with MRD⁺ TNBC and pathologic residual disease. Patients may receive standard of care capecitabine or the PARP inhibitor talazoparib, the PD-L1 inhibitor atezolizumab, or the PI3K inhibitor inavolisib.

The list of these ongoing clinical trials is summarized in Table 2. The variety of trials in this setting reflects the promise of this new technology and area of needed clinical study.

Although there are advantages offered by MRD detection, there remain notable limitations providers must consider. First, most patients with breast cancer will be receiving adjuvant therapy regardless of MRD detection. Thus, the presence of MRD after surgery is likely most informative when a patient is already receiving or has received standard of care adjuvant therapy. It is important to establish via clinical trial whether modification of our adjuvant approach improves outcomes in this setting. Second, technical errors are potential pitfalls. In order to limit error and background noise, assays may undergo interventions such as barcoding, polishing, and free radical elimination (31). To evaluate for CHIP, some assays are now performing preemptive WES of white blood cells, while other assays instead rely on knowing the most common CHIP mutations and interpreting these findings with caution. MRD detection may also be less sensitive when patients are on active chemotherapy, or if recurrence is local rather than distant. Further study is also required to assess the sensitivity of plasma-based ctDNA analysis in patients with CNS-only recurrence, given anatomic limitations of the blood–brain barrier. Studies are also continuing to distinguish the lead times of clinical versus radiographic recurrence, given breast cancer does not undergo systemic radiographic surveillance after curative surgery.

While the above are technical limitations, there also remain patient-level limitations. Most importantly, there is no evidence-based therapy that improves outcomes for patients with breast cancer. These studies are currently underway (Table 2). Until an effective intervention is identified, the knowledge that a patient has MRD, a known indicator of increased risk of recurrence, risks being a source of stress without a clear therapeutic benefit. We also do not yet know when one should test for MRD, and given we know sequential testing improves sensitivity and specificity, how often (56).

Furthermore, increasingly sensitive assays may allow for future studies of therapy deescalation. Evidence suggests that patients with breast cancer do not all require the same level of intensive adjuvant therapy (61). A reliable test for molecular evidence of disease could potentially allow for serial monitoring in the setting of deescalated adjuvant therapy in otherwise low risk patient populations. In colorectal cancer, the DYNAMIC study showed ctDNA-guided chemotherapy deescalation led to patients receiving less chemotherapy with no difference in 2-year recurrence free survival (62). A key piece to underline in this trial is the fact that adjuvant chemotherapy for stage II colorectal cancer has not definitively been linked to improved OS, which is distinct from breast cancer, where adjuvant therapy in all three receptor-based subtypes has shown survival benefit (62–64). Thus, the threshold to study deescalation may be initially higher in breast cancer and mandate an exquisitely sensitive concurrent MRD assay.

There are multiple areas of potential clinical impact in the MRD space. By definition, the term MRD itself describes residual disease; thus this describes a state where there is detectable disease after definitive treatment. That being said, these assays have also shown prognostic significance in the neoadjuvant space, and there are also ongoing studies of assays similar to the tumor-agnostic MRD assays in the early detection space. Each clinical application holds potential to markedly impact patient care. In the adjuvant space, there would be great benefit in distinguishing patient groups at greatest risk of recurrence, as well as at lowest risk, to potentially further refine the adjuvant treatment landscape to be even more personalized to each patient's individual risk.

The evolving field of MRD is an exciting and potentially practice-changing space. Breast cancer is a unique disease in that in some cases it can lie dormant for many years and recur as metastatic disease many years after curative intervention. There remains a need for real-time monitoring for recurrence, and increasingly sensitive ctDNA assays are affording providers that opportunity. As these continue to improve, we in parallel are seeking to learn about the clinical actionability of MRD. As new information is identified through clinical trials, there is hope for a tool that allows for real-time clinical escalation and deescalation in contexts personalized to each patient.

A.J. Medford reports personal fees from Illumina, Natera, and Guardant Health outside the submitted work. L.M. Spring reports consultant/advisory board for Novartis, Puma, G1 Therapeutics, Daiichi Pharma, and AstraZeneca, and institutional research support from Phillips, Merck, Genentech, Gilead, and Eli Lilly. S.A. Hurvitz reports grants from Ambrx, Arvinas, AstraZeneca, Bayer, Cytomx, Dantari, G1 Therapeutics, Greenwich Life Sciences Inc., Eli Lilly, LOXO Oncology, Macrogenics, Radius, Sanofi, and Zymeworks outside the submitted work; and grants and other support from Daiichi-Sankyo, Genentech/Roche, Gilead, Novartis, OBI Pharma, Orinove, Orum, Pfizer, Phoenix Molecular Design, Pieris, PUMA, and SeaGen. N.C. Turner reports advisory board honoraria from AstraZeneca, Lilly, Pfizer, Roche/Genentech, Novartis, GlaxoSmithKline, Repare Therapeutics, Relay Therapeutics, Zentalis, Gilead, Inivata, Guardant, and Exact Sciences; and research funding from Astra Zeneca, Pfizer, Roche/Genentech, Merck Sharp & Dohme, Guardant Health, Invitae, Inivata, Personalis, and Natera. A. Bardia reports personal fees from Pfizer, Novartis, Genentech, Merck, Radius Health, Immunomedics/Gilead, Sanofi, Daiichi Pharma/AstraZeneca, Phillips, Eli Lilly, and Foundation Medicine during the conduct of the study; and grants from Genentech, Novartis, Pfizer, Merck, Sanofi, Radius Health, Immunomedics/Gilead, Daiichi Pharma/Astra Zeneca, and Eli Lilly outside the submitted work. No disclosures were reported by the other author.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.