Tissue-Agnostic Drug Development: A New Path to Drug Approval

The landscape of cancer therapeutics has rapidly evolved over the past decade. The use of molecular genetics and biomarker-driven strategies rather than classic histology based on organ of origin have reinforced the concept of precision oncology. Better understanding of how to employ basket-type clinical trials has made clinical trial options available to patients with a variety of cancers, especially those with rare cancers, and has informed tumor-agnostic drug development. The recent approvals by the FDA have further heightened interest in target-driven as opposed to histology-driven drug development (https://www.fda.gov/drugs/resources-information-approved-drugs/hematologyoncology-cancer-approvals-safety-notifications; Fig. 1). The accelerated approval of pembrolizumab for the treatment of unresectable or metastatic, microsatellite instability–high (MSI-H) or deficient mismatch repair (dMMR) solid tumors marked a milestone in the development of therapeutics, adding the concept of tissue-agnostic drug development. Subsequently, larotrectinib and entrectinib were approved by the FDA for the treatment of adult and pediatric patients with neurotrophic tyrosine kinase (NTRK) fusion–positive solid tumors. In addition, pembrolizumab recently obtained approval in patients with tumor mutational burden–high (TMB-H) solid tumors. Herein, we focus on the history of tissue-agnostic drug development, its rationale, and the need to understand target modulation in the context of histology. Tissue-agnostic approvals have been granted in Europe, Japan, Australia, and other regions (although there may be differences in applications that have been submitted and approved across specific countries or regions).

Therapies targeting the immune system have shown dramatic benefit in the treatment of multiple solid tumors. These therapies induce an immune response against tumor antigens by modulating cells present in the tumor microenvironment (extrinsic), essentially normal host immune cells. This differs from the previous approaches that were designed against molecular targets present in cancer cells (intrinsic). Both strategies have had their successes and failures, and combined approaches are being tested in clinical trials.

The MMR system is crucial in rectifying mismatched nucleotides and could become deficient if there are mutations in MMR proteins or methylation of 1 of 4 genes: MLH1, MSH2, MSH6, or PMS2 (1). There are 3 mechanisms by which these genes can become inactivated: hereditary or Lynch syndrome (biallelic pathologic germline mutations can cause constitutional MMR deficiency), somatic or sporadic mutations via epigenetic inactivation, and promoter methylation. Mismatch repair deficiency causes high rates of mutations and errors in DNA, including in short repetitive sequences or microsatellites, resulting in MSI. MSI-H or dMMR is variable across tumor types. In a cohort of 12,019 patients with more than 30 tumor types, using liquid biopsy, Le and colleagues demonstrated that dMMR was present in numerous solid tumors, with more than 5% of patients with adenocarcinomas harboring microsatellite instability (2). Although incidence varied by cancer, 8% of patients with stage I to III cancers and 4% of patients with stage IV across tumor types were dMMR, representing approximately 60,000 cases annually in the United States. They also identified functional mutation-associated neoantigens or tumor-specific T cells in the peripheral blood that increased in number after anti–PD-1 treatment. MSI-H tumors were also shown to be enriched with activated cytotoxic lymphocytes, and there was a higher rate of tumor cells close to activated intraepithelial lymphocytes undergoing apoptotic cell death.

An early response in a patient with MSI-H colorectal cancer, along with an understanding of the biology of these tumors, formed the basis for conducting clinical trials of checkpoint inhibition in patients with MSI-H/dMMR tumors. To support the accelerated approval of pembrolizumab, Merck submitted data from patients enrolled in 1 of 5 multicohort, single-arm clinical trials (KEYNOTE-016, KEYNOTE-164, KEYNOTE-012, KEYNOTE-028, and KEYNOTE-158). These trials demonstrated efficacy of pembrolizumab treatment in MSI-H or dMMR solid tumors (locally advanced or metastatic). Across the 5 trials, 149 patients with more than 15 cancer types—colorectal cancer (60%) being the majority—were assessed. Pembrolizumab was given either 10 mg/kg every 2 weeks or a flat dose of 200 mg every 3 weeks up to 24 months. Objective responses (ORR) were observed in 36% of patients with colorectal cancer, and ORR was 46% in other tumor types, conferring an overall ORR of 39.6%. Notably, 7% experienced complete response (CR), and the majority (78%) had a response lasting more than 6 months. These data resulted in the FDA granting accelerated approval in May 2017 to pembrolizumab for the treatment of patients with MSI-H or dMMR solid tumors; this became the first tumor-agnostic drug approval as a treatment for patients with cancer (Fig. 1). More recently, Marabelle and colleagues (3) reported an efficacy update of 233 patients in the non–colorectal cancer cohort from KEYNOTE-158 after a median follow-up of 13.4 months, with more than 27 different tumor types; ORR was 34.3%. Although responses appear to occur across tumor types, it is premature to determine whether tumor histology or location, prior treatment, or other factors may influence the response rate (higher or lower) in specific tumor types. In Marabelle and colleagues' report (3), one potential exception was observed in primary brain cancer, with 0 responses out of 13 patients (3). Importantly, prior treatment with temozolomide in brain cancer may result in MSH-6 mutations (and hypermutant status), although these findings may be subclonal, and therefore these tumors may be less likely to respond to checkpoint inhibition (4, 5). Responses to checkpoint inhibition have been described in patients with gliomas and constitutional MMR deficiency (6–8). Current labeling for pembrolizumab recommends testing for these markers in the primary brain tumor specimens before initiation of temozolomide chemotherapy. Further data will be needed to assess for responses in patients with de novo MSI-H/dMMR brain tumors (e.g., in patients with Lynch syndrome) to assess whether these patients may be resistant to treatment; however, de novo MSI-H/dMMR gliomas are rare.

The benefit of pembrolizumab in MSI-H tumors is further supported by the results of KN-177, a trial that randomized 307 patients with untreated metastatic MSI-H/dMMR colorectal cancer to receive either pembrolizumab or chemotherapy. Pembrolizumab improved progression-free survival versus chemotherapy [HR = 0.60; 95% confidence interval (CI), 0.45–0.80; one-sided P = 0.0002; ref. 1]. Labeled warnings include immune-related adverse reactions such as pneumonitis, colitis, hepatitis, endocrinopathies, and nephritis, and infusion-related reactions. The most common adverse reactions described in labeling (≥20%) for pembrolizumab as a single agent include fatigue, musculoskeletal pain, decreased appetite, pruritus, diarrhea, nausea, rash, pyrexia, cough, dyspnea, constipation, pain, and abdominal pain.

Pembrolizumab received approval in TMB-H solid tumors in June 2020 (Fig. 1). TMB is defined as the number of somatic, coding, base substitutions (synonymous and nonsynonymous) and short insertions or deletions (indels) per megabase of tumor genome examined. Multiple reports have shown that TMB confers an increased probability of responding to immunotherapy via increasing the chance of a tumor expressing a neoantigen that can serve as a target for the immune system (9, 10). Apart from the scientific question regarding what defines “high” versus “low” TMB, there are multiple considerations in arriving at an accurate assessment of TMB. Differences in testing across different platforms include using whole-exome sequencing (WES) versus a targeted panel, size of the targeted panel, and an algorithm that includes synonymous mutations or not, among others.

Even considering differences in testing, TMB remains a continuous variable, and values for TMB vary between different malignancies and may vary between primary and metastatic lesions of the same cancer. Fernandez and colleagues (9) compared TMB values obtained by WES of more than 3,534 primary tumors in the Cancer Genome Atlas (TCGA) and 696 metastatic tumors from the Weill Cornell Medicine Advanced Cohort. The range of TMB varied across tumor types: TMB in prostate cancer ranged from 0.03 to 14.13 mutations/megabase (Mut/Mb), whereas TMB in bladder cancer ranged from 0.04 to 99.68 Mut/Mb. Before the KEYNOTE-158 study, there were other reports of TMB being a prognostic or predictive marker of response to immune checkpoint therapy (10).

Pembrolizumab labeling for the TMB-H indication describes the results from the multicenter, multicohort, nonrandomized, open-label KEYNOTE-158 trial (NCT02628067; ref. 11). In the patient population evaluable for efficacy analysis, 102 patients across 10 different tumor types were identified as having ≥10 Mut/Mb and were included in the TMB-H group. The most common tumor types included small-cell lung, cervical, endometrial, anal, and vulvar cancers. Four percent experienced CR, and 25% had partial response, conferring an ORR of 29%. Notably, the median duration of response (DOR) was not reached in the TMB-H group at the time of publication. Responses were durable, with 50% having a duration of response of ≥24 months. Treatment-related side effects in this report were similar to those of previous immune checkpoint studies, with 10% having treatment-related serious adverse events and 15% experiencing grade 3 to 5 treatment-related adverse events.

Although not described in labeling, the FDA review also assessed tumor response information in 2,234 patients (1,772 received pembrolizumab and 462 chemotherapy) enrolled in 12 trials who had TMB assessed by WES (12). Among patients treated with pembrolizumab, 433 were TMB-H (with WES of ≥175 mut/exome approximate to 10 mut/Mb per Foundation One) with an independently assessed ORR of 31% versus 10% who were not TMB-H. The WES analyses included tumor types such as head and neck squamous cell carcinomas, urothelial cancer, gastric cancer, and triple-negative breast cancer that were minimally studied in KEYNOTE-158. In addition to the analyses of ORR, the FDA also reviewed supportive results of exploratory post hoc analyses of progression-free survival and overall survival by TMB when pembrolizumab was compared with chemotherapy in 3 randomized trials [in non–small cell lung cancer (NSCLC), urothelial cancer, and gastric cancer].

KEYNOTE-158 used a cutoff of ≥10 Mut/Mb using the FoundationOne CDx panel to define TMB-H. This cutoff appeared to identify patients in the dataset with a likelihood of responding to pembrolizumab therapy; however, occasional responses were also reported in the non–TMB-H group in this trial. Employing a single cutoff of TMB-H across different tumor types has advantages of standardizing eligibility and allowing participation of patients with rare tumors for which large TMB datasets may not be readily available to define cutoffs. A single cutoff in essence seeks to identify a lower limit or floor for TMB, acknowledging that some differences in immunogenicity (or response rate) might exist across tumor types due to characteristics of different mutations that may be derived from various sources (UV exposure, smoking, etc.). It is difficult, however, to identify reasons for differences in response rates across tumor types for TMB (or other biomarkers), as some differences are expected by chance, and there may be other reasons for differences (e.g., patients in some tumor types may be more heavily pretreated).

An exploratory analysis in the subgroup of 32 patients enrolled in KEYNOTE-158 whose cancer had TMB ≥10 Mut/Mb and <13 Mut/Mb showed an ORR of 13% (95% CI, 4%, 29%), which was lower than the response rate in the overall TMB-H population included in the trial. If treatment effects are higher as TMB increases, differences in the distribution of TMB (a continuous variable) in different tumors may in part explain why the treatment effect in some tumor types has been reported to be larger than others at the cutoff of 10 Mut/Mb. In addition, 1 of 14 patients with TMB-H anal cancer in KEYNOTE-158 responded to pembrolizumab, whereas 9 of 84 in the non–TMB-H group responded; however, median TMB was higher in responders versus nonresponders in anal cancer, with an overall median TMB cutoff of <10.

The FDA approved the TMB-H application, considering unmet need of patients without satisfactory available therapy and the totality of data submitted in the application, which was an extension of the risk/benefit profile of pembrolizumab in numerous other settings. Although the FDA acknowledged uncertainties regarding the approval, the application was granted accelerated approval with a requirement for Merck to collect additional data to further assess the effects of pembrolizumab in different tumor types and at different TMB cutoffs. Specifically, Merck is investigating the use of pembrolizumab in additional patients at TMBs greater than 10 Mut/Mb, between 10 and 13 Mut/Mb, and greater than 13 Mut/Mb to further assess whether modifications to the indication are needed, whether continued marketing for the TMB indication is appropriate, and whether any disease-specific information (or cutoffs) or limitations should be further described in labeling.

Another histology-independent approval was in targeting NTRK rearrangements. The tropomyosin receptor kinase family constitutes 3 proteins, high-affinity nerve growth factor receptor (TRKA), BDNF/NT-3 growth factor receptor (TRKB), and NT-3 growth factor receptor (TRKC), that are encoded by 3 genes: NTRK1, NTRK2, and NTRK3, respectively (13). Since the initial discovery of NTRK fusions in colorectal cancer and papillary thyroid cancer (PTC), multiple tumor types have been identified to harbor NTRK fusions. NTRK fusions have been found to be the sole driver in reported cases of some rare malignancies, such as mammary analogue secretory carcinomas (MASC) of the salivary glands (42.9%–100% of reported cases), secretory breast carcinomas (66.7%–100%), infantile fibrosarcomas (90%–100%), and congenital mesoblastic nephroma (41.5%–92.9%). Overall, the incidence of NTRK fusions ranges from less than 1% to 5% in common cancers.

Approval of larotrectinib, an orally bioavailable, pan-TRK inhibitor, was based on the results of 3 phase I/II trials: LOXO-TRK-14001 (NCT02122913), NAVIGATE (NCT02576431), and SCOUT (NCT02637687). The former two studies enrolled patients 12 years and older, whereas the SCOUT trial was in the pediatric population (14). Initial pooled analysis of the 3 trials included 55 patients across 17 different tumor types. The most common tumor types were salivary gland (22%), non-gastrointestinal stromal tumor soft-tissue sarcomas (20%), infantile fibrosarcoma (13%), thyroid cancer (9%), and 7% each of colorectal cancer, NSCLC, and melanoma. ORR per independent review was 75%, and CR was observed in 22% of patients. Neurotoxicity and hepatotoxicity are labeled warnings, and the most common adverse reactions (≥20%) were fatigue, nausea, dizziness, vomiting, increased aspartate aminotransferase, increased alanine aminotransferase, cough, constipation, and diarrhea. These results led to FDA approval in November 2018 of larotrectinib for the treatment of pediatric and adult patients with NTRK fusion–positive solid tumors (Fig. 1). Hong and colleagues (13) presented an updated pooled analysis of these 3 trials of patients recruited between May 2014 and February 2019. A total of 159 patients were treated, and the reported investigator-assessed ORR and CR were 79% and 16%, respectively. Median reported DOR (Kaplan–Meier) was 35.2 months (95% CI, 22.8–not reached).

Entrectinib, a tyrosine kinase inhibitor that inhibits ROS1, ALK, and TRKA/B/C, obtained approval for NTRK fusion–positive tumors in August 2019 (Fig. 1). Three single-arm trials (ALKA, STARTRK-1, and STARTRK-2) comprised 54 adult and pediatric patients across 10 different tumor types (including >19 histologies; ref. 15). The most common tumor types included were sarcoma, lung and breast cancers, and MASC. The ORR per independent central review was 57%, with CR of 7%. Labeled warnings include congestive heart failure, central nervous system toxicity, fractures, and visual adverse events. The most common adverse reactions (≥20%) were fatigue, constipation, dysgeusia, edema, dizziness, diarrhea, nausea, dysesthesia, dyspnea, myalgia, cognitive impairment, increased weight, cough, vomiting, pyrexia, arthralgia, and vision disorders. A report of an updated pooled analysis of these three trials that included 74 patients recruited until October 2018 described an ORR and CR per independent review of 64% and 7%, respectively (16). Median DOR was 12.9 months.

Development of NTRK inhibitors has been remarkable for the simultaneous development in adult and pediatric populations. The SCOUT trial of larotrectinib enrolled patients as young as 1 month, and entrectinib trials included patients ≥12 years of age. The use of larotrectinib in younger children was facilitated via the development of an oral solution formulation (in addition to capsules used in adults) and by the identification of a safe and effective dose in children with a body surface area less than 1 m² (i.e., 100 mg/m² twice daily).

The low incidence of NTRK fusions in common cancers poses challenges in identifying patients (requiring widespread use of genomic sequencing), need for physician education (ordering analyses optimized to detect these fusions), and limited physician experience in managing patients on TRK inhibitors. Guidelines by organizations, such as the European Society of Medical Oncology, have been proposed to assist in the diagnosis and treatment of patients with NTRK fusion–positive tumors (12, 13).

Both larotrectinib and entrectinib have post-marketing requirements to verify and describe the effects of these drugs in a larger group of patients. They also have post-marketing commitments to support the use of in vitro companion diagnostic devices to select patients for treatment with these drugs.

The tissue-agnostic approvals of pembrolizumab, entrectinib, and larotrectinib have demonstrated proof of principle that is now being investigated in platform trials or basket trials of other clinical targets. Examples of such studies include government-run studies such as the National Cancer Institute–Molecular Analysis for Therapy Choice (NCI-MATCH), professional society-run studies such as the ASCO Targeted Agent and Profiling Utilization Registry (TAPUR), and industry studies.

NCI-MATCH is a U.S. trial that initially used central diagnostic testing and is investigating the use of multiple different arms based on the molecular profile results. NCI-MATCH has reported on 5,954 patients with tumor specimens with refractory malignancies at 1,117 accrual sites (17). A potentially actionable alteration was identified by the investigators in 37.6% of patients who were profiled. A total of 17.8% were assigned to a treatment (some patients were excluded based on eligibility criteria, and others did not have an available subprotocol). In addition to the report describing actionable alterations, NCI-MATCH has also reported on antitumor activity of targeted therapy in individual cohorts including AKT1^E17K-mutated tumors, BRAF^V600E-mutated tumors (excepting certain tumor types such as colorectal cancer, thyroid cancer, and melanoma), BRAF non-V600–mutated tumors, tumors with FGFR aberrations, HER2-amplified tumors, and non–colorectal cancer MMR-deficient tumors (18).

Whereas NCI-MATCH has studied both commercially available and non–commercially available drug products, the ASCO-TAPUR study is a nonrandomized, open-label basket study evaluating commercially available targeted agents (in settings where the drugs are not approved for their malignancy) across many sites in the United States. To date, TAPUR has reported on results in both tumor-specific and broader tumor settings (19).

In addition to NCI-MATCH and TAPUR, other studies are assessing the antitumor effects of drugs targeting ALK, RET, BRAF, RAS, or HER2 or in patients with NRG-1 fusions. Although studies may be assessing for the effects of drugs in these and other biomarkers across tumor types, in some cases such trials have fostered FDA approval in specific tumor types. One example of this approach was in the FDA approvals of dabrafenib and trametinib for patients with BRAF^V600E mutation–positive locally advanced or metastatic anaplastic thyroid cancer. The study that supported approval was a multicohort study that used a Bayesian hierarchical model to assess the effects of dabrafenib and trametinib in multiple cohorts of rare cancers (20). The overall response rate was 61% (95% CI, 39%, 80%) among 23 patients with BRAF^V600E-positive anaplastic thyroid cancer. Although based on few patients, the FDA decision factored in the positive benefit–risk assessment in randomized trials in other settings such as melanoma with a similar response rate, as well as the high response rate in an ultra-rare clinical setting with substantial morbidity.

A more recent example of tumor-specific FDA approval based on data from trials that have enrolled patients with different tumor types is for pralsetinib and selpercatinib for RET fusion–positive NSCLC and thyroid cancer and RET mutation–positive medullary thyroid cancer. In contrast to NTRK-rearranged tumors, in which only fusion genes were deemed targetable, selective RET inhibitors have demonstrated antitumor effects in patients with RET mutation–positive medullary thyroid cancer.

Advances in technology have revealed oncogenic drivers across different tumor types. Moreover, the same oncogenic driver can be present in many histologically distinct tumors. Basket trial designs have been used to assess agents targeting such drivers; however, presence of a genetic abnormality in a tumor and its targeting by a therapeutic does not necessarily correlate with clinical benefit. For example, targeting BRAF^V600E mutations in melanoma resulted in high response rates, but lack of benefit was observed in patients with colorectal cancers (21). Understanding of the role of the target in a given histology and the consequence of target modulation can inform preclinical and clinical development; an understanding of biological underpinnings of a therapy across tumor types can also be helpful to determine whether a tissue-agnostic approach should be pursued. For BRAF^V600E-positive colorectal cancer, reactivation of MAPK signaling as a mechanism for nonresponsiveness to BRAF inhibition was identified in a nonclinical model after nonresponsiveness to single-agent BRAF inhibition was identified in the clinic (21).

Tumors may evolve over time, and using archival samples to determine the presence of targetable alterations may be suboptimal. An example of loss of biomarker expression has been identified with respect to HER2 expression in gastric cancer (22). The labeling of fam-trastuzumab deruxtecan-nxki recommends reassessment of HER2 status in patients with gastric cancer if it is feasible to obtain a new tumor specimen after prior trastuzumab-based therapy and before treatment with fam-trastuzumab deruxtecan-nxki. As such, decisions regarding molecular testing will need to factor in tumor-specific considerations (e.g., handling of tissue, sensitivity in a tumor type), availability of tissue, whether an additional biopsy is feasible or necessary, and factors related to the molecular test. The ability to detect genetic alterations in circulating tumor or free DNA, so-called liquid or plasma biopsy, may have obvious advantages, assuming acceptable performance characteristics. Importantly, different platforms may have different performance characteristics with respect to different analytes. It is feasible that certain platforms may be better for patients with different tumor types, as the prevalence of different aberrations will vary among tumor types.

With respect to FDA current approvals, ongoing uncertainties are being further addressed in the post-market setting. Some uncertainties may exist because of limitations in the number of patients with a tumor type (e.g., no patients or a handful of patients) or given that the prevalence of mutation-positive patients enrolled in a trial may not mimic the prevalence in the real world. To date, all approvals have been accelerated approvals with post-marketing commitments to support development of an in vitro diagnostic (except TMB, for which a device is approved) and post-marketing requirements to obtain additional clinical data in more patients. These data or data from other sources may be useful to assess whether any future labeling changes would better inform the use of drugs in these biomarker-positive settings. For example, such data will provide additional information with respect to response in different tumor types and, for TMB, will provide additional data about the cutoff point (e.g., 10 versus 13 Mut/Mb) or whether continued marketing of the indication is warranted.

Recent years have seen tremendous growth in our knowledge of cancer biology and host response, resulting in expansion of the number of drugs approved to treat patients with different cancer types. Tumor-agnostic development of agents has been a major step forward, especially in providing treatment options for patients with rare tumors. The appropriate development plan, however, will be unique for each drug and should be carefully considered to ensure the appropriate balance between speed, development risk, and uncertainties with respect to risks and benefits.

S. Kummar reports other support from Boehringer Ingelheim, Springworks Therapeutics, Bayer, Genome & Company, HarbourBiomed, Seattle Genetics, Mundibiopharma, PathomIQ, Cadila Pharmaceuticals, Arxeon, and Gilead during the conduct of the study. No disclosures were reported by the other authors.