The NCI-MATCH trial: Lessons for precision oncology

Abstract

The NCI-MATCH (Molecular Analysis for Therapy Choice) trial (NCT02465060) was launched in 2015 as a genomically-driven, signal-seeking precision medicine platform trial — largely for patients with treatment-refractory, malignant solid tumors. Having completed in 2023, it remains one of the largest tumor-agnostic, precision oncology trials undertaken to date. Nearly 6000 patients underwent screening and molecular testing, with a total of 1473 (inclusive of continued accrual from standard next-generation sequencing) being assigned to one of 38 substudies. Each substudy was a Phase II trial of a therapy matched to a genomic alteration, with a primary endpoint of objective tumor response by RECIST criteria. Here, we summarize the outcomes of the initial 27 substudies in NCI-MATCH, which met its signal-seeking objective with 7/27 positive substudies (25.9%). In this Perspective, we discuss key aspects of the design and operational conduct of the trial, highlighting important lessons for future precision medicine studies.

Introduction

It has long been known that certain malignancies are primarily driven by ‘driver’ mutations and that inhibition of the affected pathways can lead to a substantial anti-tumor response and survival advantage. A well-established example is chronic myelogenous leukemia, driven by a translocation of chromosome 9 and 22 which results in formation of the BCR/ABL fusion oncogene. The use of tyrosine kinase inhibitors against this driver has reduced the annual mortality by a factor of 10 – from 10–20% annually to 1–2% annually (1). Other notable examples include EGFR mutations in a subset of lung cancers and breast cancers driven by overexpression of HER2, to name but a few (2) — leading to the idea of precision oncology, whereby a treatment is targeted to specific molecular driver.

It has also become clear that some driver mutations can occur across different tumor histologies but crucially, these can confer susceptibility to targeted therapies in some tumor sites but not in others. An early example of this complexity was provided by drugs directed to the V600E mutation in the BRAF gene; while high response rates (40% or more) (3) were observed in melanoma patients with the aberration, colorectal cancer patients were almost uniformly resistant (4). With the development of massively parallel high-throughput genomic sequencing (next-generation sequencing, NGS) more than a decade ago (5), the broader picture of DNA aberrations across cancers began to emerge, along with myriad potential therapeutic opportunities.

With these considerations in mind, the ECOG-ACRIN Cancer Research Group collaborated with the National Cancer Institute (NCI) to design a trial that would systematically evaluate the activity of the many emerging targeted drugs across a range of different cancer diagnoses. Planning began in 2013 and the NCI-MATCH trial launched in 2015 — at that time representing the largest tumor-histology agnostic, genomically-driven clinical trial yet undertaken (6,7). At the time of its planning, numerous studies attested to the variability of results obtained from different next-generation sequencing (NGS) platforms (8,9), prompting implementation of a uniform assay (Oncomine^™, Thermo-Fisher) in four credentialed laboratories, to be applied across all samples (10). Updates to the assay platform were made to keep pace with scientific developments in the field, including designated immunohistochemical biomarkers for patient selection (11), all the while maintaining a consistent platform for the screened population. When the screening reached the target accrual of 6000 in July 2017, the trial was continued using commercial and academic laboratory testing, until it closed in January 2023. In this Perspective, we summarize the key features of the trial design and conduct, as well as some key challenges and how we addressed these along the way. We discuss the outcomes and limitations, and what they mean for patients and clinicians. The results provide a perspective on the interpretation of the NCI-MATCH trial, and its implications for the design of successor studies.

Trial Design and Conduct

The design of NCI-MATCH (Fig 1) has been outlined in detail (6,7). Each target molecular aberration had to be supported as a driver of tumor growth, and the paired drug demonstrated to inhibit either the driver itself, or the growth pathway activated by it. Each list of qualifying molecular aberrations for a gene was defined through database searches for new experimental or clinical data, and was updated regularly as the trial proceeded.

Figure 1. NCI-MATCH Platform Trial Design.

The trial was implemented in three parts, sequentially: a pilot phase (795 patients) that led to multiple modifications; a screening accrual phase (~6000 patients); and a continuation phase whereby patients were recruited from a network of academic and commercial laboratories. In the initial 6000-patient screening period, eligibility was assessed prior to obtaining a dedicated biopsy, while during the continuation period, tumor samples were assayed as part of standard clinical practice, and eligibility assessed when a candidate mutation was identified. In either case, eligible patients with qualifying tumor molecular aberrations were assigned to a therapeutic substudy, to receive treatment directed to their molecular profile. As a further quality control measure, for patients tested at designated labs, confirmation sequencing was conducted by the NCI-MATCH Central Labs and only patients whose tumor genomic profile was confirmed in this manner were included for efficacy endpoints. Each substudy was a separate Phase II trial constituting a drug-genomic driver pair. Patients were assigned to substudies with the assistance of a decision tool (MATCHBox), overseen for appropriateness by a team of medical oncologists, laboratory scientists, and bioinformaticists from the trial leadership that reviewed every patient. In each substudy, the initial aim was to accrue 35 patients, assuming 31 would be eligible and start protocol treatment (analyzable); there was provision to increase accrual to 70 in selected arms. The primary endpoint for each substudy was objective response rate (ORR), defined as rate of complete or partial response, as assessed by RECIST guidelines (12).

Pilot Phase of Trial Initiation

The NCI-MATCH trial incorporated an initial pilot phase of accrual to assess the distribution of genomic aberrations (given limited sequencing data at the time for metastatic cancers), and to evaluate the performance of all elements needed to provide therapeutic options to patients across a wide geographic area. In the initial three months, the trial had 10 treatment arms, and the central network comprising four laboratories – harmonized to run assays on an identical platform – was responsible for all the sequencing (6). Unanticipated enthusiasm in the research community for this trial revealed a number of practical needs — including higher tissue processing and sequencing throughput to enable acceptable turnaround times, and a support desk and broad educational program for clinical providers. The pilot also revealed the need for a substantially higher total screening accrual, since estimates of prevalence based on The Cancer Genome Atlas data (from less advanced cancers) were incorrect by a factor of two or more, and also revealed the need for a greater number of treatment arms, to provide options for patients with less prevalent actionable mutations or alterations. Similar needs had been identified in the earlier SHIVA trial (13), conducted with a small number of treatment arms across multiple institutions in France — indicating the importance of large numbers of therapeutic options in future precision oncology trials.

As had been hoped at the outset, most of the accrued patients were not from the four most common tumor types (breast, lung, colon and prostate cancers); over 60% were from less common or rare histologic types [Table 1] (6). Furthermore, new biopsy specimens of metastatic disease were successfully acquired (7), with the aim of capturing new genomic changes that may have occurred since earlier collection of primary tumor tissue (6).

Table 1.

Total accrual by Disease to NCI-MATCH.

	Screening Cohort N = 6390	Outside Assay N = 762
Colorectal Cancer	963 (15.1%)	76 (10%)
Breast	764 (12%)	85 (11.2%)
Ovarian	610 (9.5%)	38 (5.0%)
NSCLC	485 (7.6%)	53 (7.0%)
Pancreas	413 (6.5%)	25 (3.3%)
Uterine Cancer	402 (6.3%)	70 (9.2%)
Liver and Hepatobiliary Cancer	290 (4.5%)	39 (5.1%)
Sarcoma	288 (4.5%)	29 (3.8%)
Head and Neck	239 (3.7%)	34 (4.5%)
Neuroendocrine Cancer	214 (3.3%)	11 (1.4%)
Gastroesophageal Cancer	211 (3.3%)	34 (4.5%)
Prostate Cancer	157 (2.5%)	33 (4.3%)
Bladder/Urothelial	108 (1.7%)	17 (2.2%)
Cervical Cancer	103 (1.6%)	9 (1.2%)
CNS	103 (1.6%)	106 (13.9%)
Small Cell Lung Cancer	90 (1.4%)	4 (0.5%)
Melanoma	85 (1.3%)	14 (1.8%)
Kidney	83 (1.3%)	12 (1.6%)
Lymphoma	55 (0.9%)	1 (0.1%)
Mesothelioma	55 (0.9%)	2 (0.3%)
Anal Cancer	52 (0.8%)	6 (0.8%)
Myeloma	1 (<0.1%)	1 (0.1%)
Other	619 (9.7%)	63 (8.3%)

Screening of 6000 patients with a Dedicated Assay

Following the pilot phase, 24 available treatment arms (subsequently expanded to 38 in total) were approved, and a revised goal for 6000 patients to be screened was set. With the increase in available therapies, the treatment assignment rate increased from ~8% to 17.8%, despite each arm of the trial excluding patients if it was already known (based on phase 2 or 3 data) that the drug was either active or inactive in that patient’s cancer type (7). The proportion of patients with an actionable mutation (one for which any targeted therapy was available within NCI-MATCH or outside the trial) was over twice this rate, at 37.6% — indicating that nearly two of five patients with advanced cancer may have a candidate treatment revealed by next-generation sequencing (NGS) of their tumor. Given the continuing development of targeted drugs and therapeutic protocols, this finding in an unselected population may be a starting point for future trials designed to investigate and establish the efficacy of molecularly targeted therapies and their contribution to patient outcome. The proportion of screened patients who were actually registered for treatment was 12.4%, representing 70% of all patients assigned to a treatment arm; interim disease progression or consent withdrawal are believed to be the major contributors to the fractional registration.

Trial Continuation beyond Central Molecular Screening

At the conclusion of the biopsy screening portion of the trial in July 2017 (two years ahead of schedule), 5961 patients had been enrolled for molecular profiling, and 11 treatment substudies had completed full accrual. By that time, the sequencing landscape had also changed, and greater methodological consistency resulted in high reproducibility of genomic findings across a variety of sequencing platforms (9). Also, there was a desire to speed up the identification of patients with rare targetable tumor mutations. To this end, NCI-MATCH expanded its reach by engaging a network of academic and commercial labs (termed the NCI-MATCH Designated Laboratory Network), that performed NGS assays as routine care at sites participating in the trial. After a careful vetting process, a total of 30 laboratories (12 commercial and 18 academic), were approved to identify patients for this next phase of the trial, resulting in the treatment of an additional 512 patients. In addition, evidence accumulating external to the trial did not show a major difference between molecular abnormalities found at initial diagnosis and those found in metastatic disease (2). Accordingly, the trial was continued using the archived specimens that are usually analyzed in genomic laboratories, without a requirement for a new biopsy. This approach has now been implemented for all current NCI-sponsored solid tumor precision medicine trials.

Outcomes and Evaluation of NCI-MATCH

Was the trial feasible?

Feasibility was a concern at the outset of the trial. Thus, the statistical plan included monitoring for insufficient accrual and lack of activity among the arms, as well as stopping rules built into the individual substudies. Concerns about accrual were swept away promptly: registration of 6000 patients in just 15 months was unmatched in the history of ECOG-ACRIN therapeutic studies. Several other genomically-targeted therapy studies conducted in the same time-frame also support feasibility. One example is the SHIVA trial, conducted in France, with 10 treatment arms (13). Though there were limitations in terms of the availability of therapies in the SHIVA trial, it accrued 741 patients in 21 months. Another trial in France that required re-biopsy (MOSCATO-01) accrued 1035 patients over 51 months (14). Hyman et al reported on a study of neratinib in 141 patients with a variety of cancers whose tumors harbored mutations in ERRB2 and ERRB3 (15). A large trial of genomically-defined maintenance therapy in colorectal cancer (MODUL) has been completed with a total accrual of 824 patients (16). The VIKTORY trial in Korea accrued 772 patients with gastric cancer in a period of 52 months (17). These and other trials attest to the ability of the oncology community, researchers, and patients together, to develop and implement precision oncology trials. Response data are now available for on a substantial proportion (27/38) of the NCI-MATCH treatment arms (discussed below), and eight arms have been closed based on very low prevalence of the targeted aberration.

Did NCI-MATCH achieve its signal-seeking goal?

The goal of NCI-MATCH was to understand the activity of molecularly targeted therapy applied to cancer gene-defined subsets across different tumor histologies, and to document response — regardless of tumor histologic type. The purpose was signal-seeking: a specific target number of positive Phase II studies was not defined, nor was it attempted to define a concept of ‘tumor-agnostic activity’ in which an agent could be defined as broadly active in cancers with the targeted molecular alteration. Sample size and power calculations stipulated that, for a substudy with 31 analyzable patients, five or more responses (partial or complete) would be required, i.e., objective response rate (ORR) ≥ 5/31 (16%). Given the admixture of multiple tumor types to be accrued in each arm, this ORR was considered indicative of activity across tumor types and worthy of further investigation in a tumor-type agnostic fashion.

A summary of the outcomes of the 27 substudies reported to date is shown in Table 2. Seven of the 27 arms (25.9%) met the pre-specified criterion for positivity. Other arms had lower response rates, some of which may support future development with combinations, or with single agents in specific tumor types. In fact, at least one response was observed in 22/27 (81.5%) of the substudies. Across all the treatment arms reported to date, the overall response rate among evaluable patients was 79/765 (10.3%). In terms of breadth of activity, two of the arms – nivolumab in mismatch repair-deficient tumors and dabrafenib plus trametinib in BRAF V600 mutant tumors (37, 21) – identified activity across a broad range of malignancies. These results contributed to tumor type-agnostic registration for nivolumab, and an accelerated approval for the combination of dabrafenib plus trametinib in patients with BRAF V600-mutant tumors. These FDA approvals provide strong support for the impact of genomically-driven trials for patients, especially those with less common diseases. It should be noted that despite the modest impact of early inhibitors, signals were also obtained in select genotypes with PI3K-PTEN-AKT pathway aberrations. The AKT inhibitors capivasertib and ipatasertib had almost identical response rates in AKT E17-mutated cancers, a mutually validating result (34,40). Copanlisib in PIK3CA-mutated tumors also provided a signal of activity, with a response rate of 16% and 38% of patients free of progression at 6 months (38). These results also provide additional impetus for the study of mutation-specific agents directed to these targets.

Table 2.

Outcomes of the initial 27 substudies (of 38 total) in NCI-MATCH.

Arm	Molecular Aberration	Treatment	N Enrolled	N Evaluable†	Number of Responses (%)	6-month PFS	Ref	Met Endpoint?*
A	EGFR activating mutations	afatinib	19	14	1 (7.1%)	8.9%	18	No
B	HER2 activating mutations	afatinib	40	37	1 (2.7%)	12.0%	19	No
F	ALK fusions	crizotinib	5	4	2 (50.0%)	25%	20	Yes
G	ROS1 fusions	crizotinib	4	4	1 (25.0%)	50%	20	No
H	BRAF V600E or V600K mutations	Dabrafenib/trametinib	35	29	11 (37.9%)	68.4%	21	Yes
I	PIK3CA mutation without RAS mutation or PTEN loss	taselisib	70	61	0.0%	19.9%	22	No
J	HER2 amplification	Trastuzumab/pertuzumab	35	25	3 (12%)	25.3%	23	No
K2	FGFR mutation/fusion	erdafitinib	35	21	3 (14.3%)	36.8%	24	Yes
M	TSC1 or TSC2 Mutations	TAK-228	49	34	5 (14.7%)	28.7%	25	No
N	PTEN aberration, with + expression on IHC	GSK2636771	24	22	0.0%	4.8%	26	No
P	PTEN loss by IHC	GSK2636771	35	32	0.0%	3.3%	26	No
Q	HER2 amplification	ado-trastuzumab emtansine	38	36	2 (5.6%)	23.6%	27	No
R	BRAF fusions/non-V600 mutations	trametinib	35	32	1 (3.0%)	17%	28	No
S1	NF1 mutation	trametinib	50	46	2 (4.3%)	20.5%	29	No
S2	GNAQ or GNA11 mutation	trametinib	4	4	1 (25%)	50%	29	No
T	SMO or PTCH1 mutations	vismodegib	34	22	2 (9.1%)	22.4%	30	No
U	NF2 mutation	Defactinib	35	30	1 (3.3%)	22.8%	31	No
V	C-kit mutations	Sunitinib	10	8	2 (25%)	25%	32	No
W	FGFR pathway aberrations	AZD4547	52	48	4 (8.3%)	15.0%	33	No
Y	AKT mutations	capivasertib	35	35	10 (28.6%)	50.0%	34	Yes
Z1A	NRAS mutations	binimetinib	53	47	1 (2.1%)	29.2%	35	No
Z1B	CCND1/2/3 amp and Rb +	palbociclib	40	32	0.0%	16.0%	36	No
Z1D	dMMR status	nivolumab	47	42	15 (35.7%)	51.3%	37	Yes
Z1F	PIK3CA	copanlisib	35	25	4 (16.0%)	38%	38	Yes
Z1H	PTEN mut without PTEN protein loss	copanlisib	35	23	1 (4.3%)	14.3%	39	No
Z1K	AKT mutation	Ipatasertib	35	26	6 (23.1%)	52.4%	40	Yes
Z1L	BRAF Fusions or Non-V600E, Non-V600K BRAF Mutations	Ulixertinib	35	26	0.0%	5%	41	No

^†Eligible, treated and variant confirmed by central laboratory testing.

^*A substudy with 31 or more analyzable patients was to be called positive if the null hypothesis of objective response rate (ORR) ≤ 5% could be rejected at the one-sided type 1 error rate of 1.8%; if there were fewer than 31 analyzable patients, a type I error of 5.0% was used. This requires five or more responses (partial or complete) for a substudy with 31 analyzable patients, i.e., ORR > 5/31 (16%).

Therefore, even in a heavily pretreated population, the goal of identifying signals was met, although the majority of the arms did not meet the activity threshold, and none of the positive arms had response rates in the range of highly active targeted single agents, e.g. imatinib in gastrointestinal stromal tumor (GIST) (42). Consequently, NCI-MATCH led to additional questions about what factors influence response to a given agent when the targeted mutation is present. Illuminating such complexity may be possible upon completion of additional sequencing (whole exome, RNAseq, and others) and circulating tumor DNA (ctDNA) analyses of pretreatment and progression samples — and may encourage combination studies at earlier stages of patients’ treatment courses.

Limitations of NCI-MATCH

Design Limitations.

This trial does not provide an evaluation of the efficacy of using genomics to target molecular abnormalities in patients with metastatic cancer. The ‘match rate’ refers only to patients who had the index molecular variant when an appropriate substudy was available. At the time of screening, the appropriate substudy for any given patient might not have been available, and thus the patient could not be accrued. In regard to response, NCI-MATCH was a signal-seeking trial, with response assessed in a tumor-agnostic way — and as such the analysis could not provide response rates for specific tumor types as, by design, it was not powered to accrue enough patients to do so.

Operational Limitations.

Even with a central institutional review board, the time it took to open a substudy was long, sometimes approaching a year. Various scenarios contributed, including issues with drug supply or change in formulation, awaiting phase I results, protocol preparation, education of site staff and individual site review, and stress on the clinical and ethical protocol evaluation system. In some cases, not all patients accrued could be evaluated for the primary endpoint, due to lack of tissue availability for confirmatory sequencing by the NCI-MATCH assay. In future, this confirmatory step should not be necessary, given the high reproducibility of the assay.

Diversity of Accrual.

The racial/ethnic composition of the patients accrued to NCI-MATCH did not mirror distributions within the US population at large. Among all registered patients, 9.3% were Black, 5.6% were Hispanic and 3.9% were Asian. All these frequencies appear to be lower than catchment area population representation, although a formal analysis has not been conducted and comparator data are difficult to collect, due to the large number of sites (n=367) that accrued at least one patient. Proportions of different groups were almost identical in both the initial screening phase and subsequent periods of accrual, consistent with similar profiles of participating oncologists. In the absence of detailed information on social determinants of health for participants in the NCI-MATCH accrual, there remain questions as to how geographically and socioeconomically representative the trial population is. More detailed reporting of social determinants of health, together with pro-active outreach to underserved populations, is needed to provide access for a diverse population to precision oncology trials.

Implications of the NCI-MATCH Trial

This trial was established during a period of controversy over the value of genomically-driven clinical trials, whether or not such trials could be accomplished by the National Clinical Trials Network, and whether any patient benefit would be realized. The results have highlighted specific topics to be addressed, based upon the lessons learned (Box 1). These fall into four main categories: the definition of rare tumors, the relative contributions of tissue of origin and molecular subtype to outcomes, next steps to design genomic cancer clinical trials, and understanding of co-mutations and the tumor microenvironment.

Box 1. Key lessons from NCI-MATCH to guide future precision medicine trials.

1. Pro-active outreach is needed to assure optimal patient (and provider) diversity.

2. Rare tumors are an area of unmet need that can be met (at least in part) with genomic trials; but novel trial designs and regulatory approaches are needed.

3. A clinically-relevant definition of ‘driver mutation’ will be helped by rigorous criteria for evidence to support the matching of therapy to mutation, enabling greater therapeutic activity.

4. Circumvention of resistance mechanisms will be helped by intervention earlier in the disease course and progress will be accelerated by combination approaches involving targeted therapies and immunotherapies.

5. Trial design should encompass as many therapeutic options for as many molecular aberrations as possible, so as to have an impact commensurate with the collective effort required.

Defining Rare Tumors

Rare tumors have typically been specified by histology or tissue of origin, and are conventionally considered those with a prevalence of less than 15 per 100,000 population (43). By this definition, all but 11 tumor types are classified as rare. Pediatric tumors also need to be considered by these criteria, and the initiation of the Pediatric-MATCH clinical trial (NCT03155620) recognizes this need (44); a retrospective review of all screened cases applied the WHO International Classification of Diseases for Oncology (ICD-O-3), and this provided greater diagnostic specificity compared to the MedDRA disease coding captured during the study (unpublished). In addition, the advent of broadly-available sequencing suggests two further rare tumor groups: unusual genomic profiles within a particular tumor type, and unusual genomic drivers across tumor types. Certain genomic aberrations are closely associated with some rare tumor subtypes, but may also be found sporadically in other tumor types, e.g. neurotrophic tyrosine receptor kinase (NTRK) fusions (45) (NTRK inhibitors have received disease-agnostic approval).

A question of therapeutic interest has long been whether the histologic tumor type or the genomic driver should be the more important consideration in choosing a treatment. Regardless of how a ‘rare tumor’ is defined, it is evident that a precision medicine approach is required for the 25% of adults who have a tumor type that is below a prevalence of 15/100,000, and for additional genomic and histopathological subsets of cancer. We emphasize the precision approach since it has both immediate and long-term implications for cancer diagnosis and treatment: immediate, in that many therapies are already approved, or are in development for specific genomic aberrations. These rare tumors provide a proof of principle: that outcomes are better when a vulnerability can be identified, even when we may not have all the tools at our disposal to effect a cure. But as importantly, the classification of these rare tumors opens for patients the possibility of clinical research options that would otherwise not be available, permits the design of a series of trials to build upon positive results, and may facilitate regulatory decision-making. Further, the NCI-MATCH trial shows that such treatment can be delivered in community as well as academic settings.

The finding that 38% of the accrual to the NCI-MATCH trial was in rare or uncommon cancers (defined histologically) raises issues relevant to treatment of these patients (46). Overall, the frequency of ‘actionable’ genomic aberrations is similar among rare versus common cancers, as is the degree of benefit from interventions (47). Further, the use of more extensive analyses such as whole-genome sequencing may identify actionable aberrations in as many as 62% of a large sample (including rare and common cancers) (48). Therefore, a strong rationale exists to examine genomic characteristics of all rare tumors, given that they have fewer treatment options to begin with.

How are the successes in this setting to be made widely available (or commercialized, which generally amounts to the same thing) to patients with rare tumors? The FDA Oncology Center of Excellence has provided extensive guidance to address this issue, with a recent focus on the opportunities to bring real-world evidence to bear on trials in rare tumors (). In genomic subgroups in which a response signal has been observed, how will it be feasible to develop studies to render this signal more effective, if not curative? Large-scale deep-sequencing analyses of samples from patients treated within substudies of NCI-MATCH are ongoing, and will be hypothesis-generating as to molecular characteristics that influence response (49). In future trials, such comprehensive analysis methods (that are available at major centers) would need to be incorporated into the clinical trial procedures within a practical time-frame, using tissues or other samples that can be acquired, transported, stored and analyzed uniformly.

Impact of Tissue of Origin vs Molecular Subtype in Response

Though not the primary goal of NCI-MATCH, the trial has made several contributions to this dialogue in precision medicine. One of the most revealing and impactful results emerged from the BRAF substudy (Arm H in Table 2), in which patients harboring a BRAF V600 mutation were offered a combination of dabrafenib (a BRAF inhibitor) and trametinib (a MEK inhibitor) as dual inhibition of the MAP kinase signaling pathway. As noted previously, RAF-directed therapy has proven highly effective in some diseases (such as melanoma) (3), but is almost inactive in colon cancer with a BRAF V600 mutation (4). In NCI-MATCH, tumor types such as these (as well as lung and thyroid cancers), in which the effect of combined therapy was already known, were excluded from the substudy. Patients in the initial 35-patient cohort harboured a broad range of histologies and the response rate (38%) was substantial (21). These results strongly suggest that response to inhibiting this pathway is indeed disease-agnostic. With additional disease-specific sensitivities observed in a parallel Novartis trial (50), the FDA was approached by Novartis and ECOG-ACRIN for a disease-agnostic indication, which was approved under Accelerated Approval provisions in June 2022.

These findings contrast with those of capivasertib and ipatasertib (arms Y and Z1K, respectively) in AKT E17K-mutated cancers, where lower response rates (around 20%) were identified in the substudies for each agent. In both treatment arms, the aberration was identified most frequently in women’s cancers, including breast, endometrial, ovarian, and cervical cancers. Furthermore, responses were largely confined to these tissue types, suggesting that both tissue of origin and mutation are required for response in these circumstances. Future trials will be needed to further address this issue.

Relevance to Genomic Cancer Trials

Beyond demonstration of feasibility, one must ask how these results inform the development of future studies directed to patients with genomically-defined cancers. The characterization of molecular changes in individual cancers has led the field of precision medicine and has greatly changed the outcomes of treatment for specific patients. One need only look at the falling death rates from lung cancer (where EGFR and ALK-directed therapies have had a clear impact) (51), or the improved outcomes for patients with metastatic melanoma (once it became possible to target mutant BRAF) (52), to appreciate that effective molecular medicine has changed standards of care in oncology. It may also be pointed out that the incremental survival benefits observed in patients with metastatic melanoma harboring mutant BRAF, translates to higher cure rates when brought forward to the adjuvant treatment setting (53). In the NCI-MATCH trial, in all but three arms, the intervention tested was a single agent. The limitations of single agent therapies in this trial echo what has been observed for many years with traditional chemotherapy and anti-HER2 therapy in breast cancer, for example. In this context, the overall response rate of 10.3% observed in NCI- MATCH (across all arms) may be viewed as meaningful for future research in the area. The implications of these results are as much strategic for therapeutics as they are specific to trial design. Sequencing of cancer tissues is needed to identify potential vulnerabilities in the tumor, and this trial provides an impetus to direct treatment to these vulnerabilities as early as possible in a patient’s course. It is recognized that single agent treatments will have limited effects, and that rational combinations to overcome resistance should be explored early, especially in potentially curative settings.

This was not a trial in which therapy directed to resistant disease could be explored. It does, however, set the stage for future studies of this nature, and a trial called ComboMATCH (NCT05564377), also to be coordinated by ECOG-ACRIN, will investigate combinations of therapies in similar molecular subtypes of cancer (54). The combinations to be tested in ComboMATCH will require in vivo evidence of efficacy for the combination in well-characterized relevant tumor models and will be restricted to targeted therapies. Given that many of the frequently co-occurring mutations are currently considered undruggable, research on combinations with immune therapies and other modalities should be a high priority.

Co-Mutations and the Tumor Microenvironment

A major finding of the molecular analysis of NCI-MATCH (and other sequencing studies of advanced cancers (47,48)) was that most patients whose tumors had a qualifying mutation also had at least one co-occurring mutation that was known – based on preclinical evidence – to contribute to drug resistance. Some of these have already been outlined (7), and additional sequencing and ctDNA analysis is underway to characterize more completely the tumors of treated patients. Most of these additional mutations (for example in TP53, KRAS, p16, MYC) are currently undruggable, although candidate molecules are in development. Nearly all arms of NCI-MATCH were targeting oncogenic alterations that are known to be truncal. This circumstance facilitated reliance on the Designated Laboratory Network and use of archival tumor specimens as a source for sequencing. However, many of the co-occurring alterations were subclonal, and so possibly acquired in response to prior therapy (55). For trials aiming to overcome resistance mediated by these co-occurring alterations, repeat tumor biopsies or ctDNA analysis of mutation profiles continue to be important considerations.

Alternative approaches to addressing resistance in these subclones should also be explored. Hahn et al have recently characterized cancer targets as either ‘intrinsic’ to the cancer itself (such as oncogenes, as well as epigenetic, metabolic, transcriptional or signaling dysregulation, and DNA damage response aberration) or ‘extrinsic’, involving cellular components of the tumor microenvironment (such as immune cells, cancer-associated fibroblasts, and blood vessels) – all of which contribute to tumor growth and progression, and are therefore plausible targets (56). The combination of targeted therapies with agents directed at the tumor microenvironment is yielding regimens with markedly enhanced activity. The multi-kinase inhibitor cabozantinib, combined with the anti-PD-1 antibody nivolumab in kidney cancer, is a salient recent example (57) — as is vemurafenib (BRAF inhibitor) with rituximab (anti-CD20) in hairy-cell leukemia (58), with many more trials in prospect. Precision cancer studies should include collection of appropriate specimens for research into which molecular or patient characteristics contribute to response or resistance to combined targeted and immunological treatments. A precision medicine trial that would address such combinations directed to genomically-defined subsets is a current need with considerable therapeutic potential.

Conclusion

In addition to its impact in clinical research, the NCI-MATCH trial emphasizes the importance of tumor DNA sequencing as part of the standard evaluation of most patients with cancer, given that more than one-third of patients will be shown to have a cancer for which the outcome can be improved with targeted therapy. Since the disease type alone cannot predict the molecular characteristics of individual tumors, implementing such interventions is not currently possible without genomic sequence information.

There are often questions as to the value of routine sequencing; we emphasize that NCI-MATCH was not set up to address this question. Relating the number of responses to the number screened is not meaningful; while some 17% of those screened could be assigned to a treatment arm, there was a total of 38% with driver mutations for whom a treatment could have been available. Assignment to treatment in NCI-MATCH depended on whether the substudy was open or closed in general, and for the patient’s tumor type in particular — as some tumors were more likely to have a given genomic variant, and the numbers of such tumors in a given substudy were restricted to allow a broader recruitment of tumor types, as required by study design. In addition, patients with tumors for which a treatment was known to be effective were not eligible, since the purpose of the trial was to address what was unknown. For all these reasons, while the value of a genomic screening policy is not represented by any proportion of responders to numbers screened in NCI-MATCH, the results clearly support availability of NGS to patients with advanced cancer.

Nevertheless, it is abundantly clear with emerging data that diseases such as melanoma, lung cancer, colon cancer, and gynecologic malignancies, as well as some rarer cancers, all now have expanded treatment options resulting from available detailed sequence analyses. Furthermore, the increasing technological capacity of sequencing and other emerging genomic characterization tools to identify evolution of sub-populations of tumor cells over time holds the promise of early approaches to target resistant clones. Although not addressed by NCI-MATCH, germline sequencing of cancer patients is also receiving considerable attention, and has been implemented in Pediatric MATCH (44,59).

These results have implications for the critical elements of future platform trial designs: rigorous molecular characterization and objective assignment of patients to treatment arms; tumor sample acquisition to enable retrospective additional sequencing and immune system evaluation to define better who will respond; pathology review of tumor specimens from treated patients as a key quality factor; and sufficient numbers of treatment arms to warrant the efforts both in the cooperative groups that run the trial, and in the community, where the resources for trial activation and management compete with other research priorities. Put another way, to be motivated to open a complex trial, oncologists must have confidence that their patients will benefit through availability of novel therapies that would be otherwise inaccessible.