Proceedings From the ASCO/College of American Pathologists Immune Checkpoint Inhibitor Predictive Biomarker Summit

PURPOSE

Immune checkpoint inhibition (ICI) therapy represents one of the great advances in the field of oncology, highlighted by the Nobel Prize in 2018. Multiple predictive biomarkers for ICI benefit have been proposed. These include assessment of programmed death ligand-1 expression by immunohistochemistry, and determination of mutational genotype (microsatellite instability or mismatch repair deficiency or tumor mutational burden) as a reflection of neoantigen expression. However, deployment of these assays has been challenging for oncologists and pathologists alike.

METHODS

To address these issues, ASCO and the College of American Pathologists convened a virtual Predictive Factor Summit from September 14 to 15, 2021. Representatives from the academic community, US Food and Drug Administration, Centers for Medicare and Medicaid Services, National Institutes of Health, health insurance organizations, pharmaceutical companies, in vitro diagnostics manufacturers, and patient advocate organizations presented state-of-the-art predictive factors for ICI, associated problems, and possible solutions.

RESULTS

The Summit provided an overview of the challenges and opportunities for improvement in assay execution, interpretation, and clinical applications of programmed death ligand-1, microsatellite instability-high or mismatch repair deficient, and tumor mutational burden-high for ICI therapies, as well as issues related to regulation, reimbursement, and next-generation ICI biomarker development.

CONCLUSION

The Summit concluded with a plan to generate a joint ASCO/College of American Pathologists strategy for consideration of future research in each of these areas to improve tumor biomarker tests for ICI therapy.

INTRODUCTION

Immune checkpoint inhibitor (ICI) therapy for cancer has been a remarkable breakthrough, extending the lives of thousands of patients, for which the Nobel Prize was awarded in 2018.¹ However, the recognition of benefits from ICI for a variety of cancers was associated with the identification of immune-related toxicities consistent with the mechanism of action of these drugs.² Serious adverse events, mostly related to autoimmune phenomenology, are typically observed in 15%-20% or more of treated patients, with drug-related mortalities of 1%-2%, although these have declined with increasing experience.² Therefore, identification of patients most likely or very unlikely to benefit from ICI therapies is of paramount importance to maximize the potential benefit for the former and avoid needless exposure to toxicities in the latter groups.

CONTEXT

• Key Objective

• ASCO and the College of American Pathologists convened a virtual Immune Checkpoint Inhibitor Predictive Factor Summit to address issues related to programmed death ligand-1 immunohistochemical staining, microsatellite instability-high/mismatch repair deficient tumor status, and tumor mutational burden-high status.

• Knowledge Generated

• Experts in the field reviewed principles, regulation, College of American Pathologists oversight, and reimbursement principles for tumor biomarker tests, followed by state-of-the-art summaries of analytical and clinical topics for each of the predictive factors.

• Relevance

• The Summit concluded with visions of the future and consensus to pursue further initiatives to improve selection of patients most likely to respond to immune checkpoint inhibition. We anticipate that these initiatives will clarify precision in standard of care for the exciting field of immuno-onocology.

Clinicians now routinely use predictive biomarkers to guide patient selection for US Food and Drug Administration (FDA)–approved anti–programmed death receptor-1 (anti–PD-1) and anti–programmed death ligand-1 (anti–PD-L1; anti–PD-[L]1) cancer therapies. FDA-approved biomarkers for clinical benefit from anti–PD-(L)1 therapies include immunohistochemistry (IHC) for expression of PD-L1 on tumor cells and/or immune tumor-infiltrating cells, microsatellite instability-high (MSI-H)/mismatch repair deficient (dMMR) tumor status, and tumor mutational burden-high status (TMB-H). Although each of these biomarkers has shown some association with response to ICI therapy, there are major issues regarding their application in the clinical care of patients with malignancy, including the selection of specific assays for each marker and their performance characteristics regarding the identification of groups of patients who should or should not be offered ICI. Furthermore, regulatory and reimbursement issues for both assays and treatments have confounded these concerns, especially related to clinical applications. For example, Appendix Figure A1 displays the complexity of available assays for specific tumor types and approved ICI, as used in a single institution's pathology laboratory.

As a result of the uncertain analytical validity and predictive value of these factors, resulting from a variety of issues discussed below, confusion has arisen among clinicians and pathologists. To address these issues, ASCO and the College of American Pathologists (CAP) convened the virtual Immune Checkpoint Inhibitor Predictive Factor Summit, held from September 14 to 15, 2021, the proceedings of which are presented in this document. Leaders from the oncology and pathology communities, pharmaceutical and diagnostics industries, the FDA, the Centers for Medicare/Medicaid Services (CMS), private health insurance organizations, the National Cancer Institute (NCI), and patient advocacy organizations were invited to participate (Appendix Table A1). The meeting was presented in five segments: (1) review of the basic principles of biomarker development, validation, regulation, and reimbursement, (2) overview of PD-L1 IHC, (3) overview of microsatellite instability-high (MSI)/dMMR testing, (4) overview of tumor mutational burden-high (TMB) testing, and (5) future opportunities and concerns (Appendix Table A1). This proceedings document summarizes background information, current state of the art, opportunities for improvement, and next steps. Videos of the entire Summit are available in the official website of the College of American Pathologists.³

TUMOR BIOMARKER DEVELOPMENT, QUALITY ASSURANCE, AND REGULATORY OVERVIEW

Tumor Biomarker Principles

Previously proposed elements necessary to apply a tumor biomarker test (TBT) in clinical practice were reviewed extensively. These included the distinction between a tumor biomarker and a TBT, noting that one or more tests may exist for a given tumor biomarker. Further discussion focused on the intended uses of predictive factors for benefit or resistance, which may lead to opting in or out of standard of care, respectively, from a therapy. As reviewed elsewhere, development of an assay for a TBT from discovery through validation to clinical commercial availability is a long and complex process during which analytical validity, clinical validity, and clinical utility must be established (Fig 1).^4-8

FIG 1.

Overview of biomarker development in the United States. PMA, premarket approval.

Approaches to establishing analytical validity of a specific assay are relatively standardized in the field of laboratory medicine. However, assays on the basis of pathologists' interpretation of immunochemical stains, particularly when semiquantitative cell thresholds are used, present challenges to achieving reliable interobserver reproducibility. Two research strategies to evaluate tumor biomarker clinical utility might be pursued: prospective randomized clinical trials, in which the TBT clinical utility is the primary objective of the study,^6,7 and retrospective analysis of archived study participants sample from prospective randomized clinical trials, in which the TBT was not used for treatment selection.⁹

Many stakeholders are involved in the decision to use a TBT to make clinical decisions: clinicians, pathologists, diagnostic and pharmaceutical companies, regulatory and expert guidelines bodies, third-party payers, and of course patients themselves. Each of these stakeholders needs to consider both the analytical validity and clinical utility of the TBT. These decisions are confounded by two separate pathways for clinical application of a TBT: (1) FDA approval/clearance or (2) generation of a high-complexity laboratory developed test (LDT) within a laboratory that is approved according to the Clinical and Laboratory Improvement Act of 1988 (Clinical and Laboratory Improvement Amendments, commonly referred to as CLIA).^10,11

At present, a complex system has evolved in which specific TBT within a biomarker class are approved for specific ICI agents for specific, different cancers. Germane to the Summit, the tumor-immune interaction is complex, and biomarker and technology development are progressing simultaneously. Therefore, confusion has arisen regarding which biomarker, and which test for that biomarker, should be used for patients with different types of malignancies for whom different specific ICI agents are being considered.

Laboratory Quality Overview

Laboratory quality management is multifaceted, spanning 12 essential elements: organization, personnel, equipment, purchasing and inventory, process control, information management, documents and records, organizational management, assessment, process improvement, customer service, and facilities and safety.¹² Proficiency testing (PT) is part of the assessment of ongoing laboratory quality monitoring and is conducted via a CMS-approved PT program, enabling the end-to-end assessment of a laboratory's performance of the analytical process. PT programs allow laboratories and regulators to detect drift in testing and thus an external assessment of laboratory and assay performance. Given the complexity of biomarker tests, personnel must meet educational requirements established by CLIA and undergo competency assessment at least annually.

Successful laboratory performance in PT and/or alternative performance assessment is mandatory for each analyte processed in the clinical laboratory. Following a PT failure, laboratories must perform a root cause analysis to determine the cause of the failure as well as implement corrective and preventative actions. Continued failure in these PT programs may result in a notification from CLIA to discontinue testing until adequate performance for clinical care is restored and maintained.

The CAP has PT programs that address TBTs for ICI. In the CAP IHC PD-L1 PT program for non–small-cell lung cancer (NSCLC), an unstained tissue microarray (TMA) slide containing 10 cores of tissue is sent to participating laboratories. Currently, since there are no FDA-approved or FDA-cleared computer-assisted scoring and interpretation of PD-L1 IHC tests, all are scored and interpreted by trained pathologists using several different scoring systems (Tables 1 and 2). The results are graded by 80% consensus and each participating laboratory is provided with its results in the context of the other laboratories, which are anonymized. It is important to note that although 80% consensus is the standard for grading, some programs do have a higher requirement, such as for human epidermal growth factor receptor 2 (HER2) in breast cancer. Further investigation into the benefit of raising the grading standard for PD-L1 could be considered.

TABLE 1.

FDA-Approved or FDA-Cleared Immunohistochemistry Assays for Programmed Death Ligand-1

TABLE 2.

Scoring Systems for PD-L1 Immunohistochemistry

The CAP MSI PT program consists of three tissue specimens with cross-correlation to the CAP mismatch repair (MMR) PT program using IHC testing (Table 3). The 2021 results indicate a very high degree of concordance in MSI testing (> 99.0% concordance). The CAP TMB PT program launched in 2022, with laboratories receiving one DNA sample with 50% tumor plus a matched normal sample, tested with next-generation sequencing (NGS; Table 4).

TABLE 3.

FDA-Approved or FDA-Cleared Mismatch Repair and MSI Assays

TABLE 4.

FDA-Approved or FDA-Cleared Next-Generation Sequencing Assays for Tumor Mutational Burden

Regulatory Overview

The FDA has regulated in vitro diagnostics (IVDs) for clinical use since 1976. There are three risk-based classifications of IVDs, and the intended use determines the type of submission (Fig 1). The major elements of an IVD submission include the intended use/indications for use, device description, internal controls, preanalytical steps, analytical validation/performance, clinical validation/performance, instrument(s)/software (if applicable), and labeling. The clinical validation study design depends on the indications for use. A reasonable assurance of safety and effectiveness regarding analytical validity and clinical performance characteristics must be demonstrated for approval. Outside of the companion diagnostic (CDx) program described below, FDA has not traditionally insisted on what is now termed clinical utility for approval or clearance of a diagnostic. Rather, this consideration of clinical utility is often left to individual clinician judgment, guidelines bodies, or third-party payers.

All US laboratories reporting test results used for patient diagnosis or care must comply with CLIA regulatory standards, established and administered within CMS. Laboratories are subject to routine on-site inspections every 2 years, having a choice of a CMS-approved accreditor, such as the CAP Laboratory Accreditation Program, to conduct the inspections.

In addition to CLIA certification, the test performance specifications must either be verified or validated before a laboratory can report patient results using a new biomarker test. For FDA-approved unmodified tests, the analytical studies such as accuracy, analytic precision, reportable range, and reference range must be verified. For LDTs (for use in a single laboratory only), which include any modification of an FDA-approved test, CLIA requires that the laboratory must establish the accuracy, precision, analytical sensitivity and specificity, reportable test result range, reference intervals, and any other performance characteristic required for test performance. LDTs are not included in the following discussions or tables in this publication.

CDx provide information that is essential for the safe and effective use of a corresponding therapeutic product. The use of a CDx is stated in IVD labeling and that of the corresponding therapeutic product. PD-L1 by IHC is the most common FDA-approved CDx test with 14 indications, compared with three for MSI/dMMR and one for TMB (at the time of the Summit). By contrast, a diagnostic that identifies a biomarker-defined subset of patients who are likely to respond well to a drug, helping to determine risk benefit for patients, but is not a prerequisite for receiving the drug, is often referred to as complementary diagnostic. PD-L1 IHC testing for anti–PD-(L)1 therapy is considered a complementary diagnostic in certain cancers, since expression of PD-L1 is dynamic and ICI benefit has been observed in patients whose tumor biopsies were deemed marker-negative.

Although well-intentioned, the FDA CDx initiative was originally designed to approve one specific TBT for one specific therapeutic agent against a target biomarker. This approach has led to some confusion in the ICI field, with the proliferation of several commercial tests for the same ICI biomarker. Alternatively, a group or class of TBTs that evaluate a target biomarker against which a group or class of therapeutics has activity could be considered. The latter approach has been used, and vetted by ASCO/CAP committees, for HER2 and estrogen and progesterone receptors regarding anti-HER2 and endocrine therapies, respectively, in breast cancer.^13-16 The final guidance on oncology therapeutic group labeling was issued by FDA in April 2020 whereby labeling claims may be broadened to include the use of a group of oncology therapeutic products rather than single agents.¹⁷

Panel participants in the Summit discussed that the group/class labeling of biomarkers for therapeutics may help to overcome some of the challenges previously experienced in the CDx space for ICI, primarily the slow pace of regulatory approval for one test/one drug/one indication in relation to the accelerated growth of medical knowledge.

Reimbursement Issues

Although reimbursement is often used to refer to payer decisions about medical necessity of a test or treatment, reimbursement more accurately refers to payment. A robust discussion focused on the challenges for reimbursement decisions for TBTs. As described above, some TBTs may be FDA-approved or FDA-cleared as a CDx with a corresponding therapeutic, others may be FDA-approved or FDA-cleared as a biomarker test independent of a therapeutic.¹⁸ Representatives of CMS, private third-party payers, and patient advocacy organizations discussed the importance of understanding the intended use of an assay and if that use has clinical utility to determine medical necessity. Insurance covers only medically necessary care. Payers seek a high level of evidence to cover a test or treatment as they have responsibility for their members' health and the affordability of care.

Regardless, emphasis was placed on the need for high levels of evidence to support payment decisions, although the complexity of contracting among various entities involved was raised as a confounding issue in precise reimbursement structures. Private payers and CMS may rely on FDA approvals, internal technology assessment panels,¹⁹ or guidance from professional societies such as those developed by the National Comprehensive Cancer Network or jointly by ASCO and the CAP for decisions regarding coverage.

CONSIDERATIONS FOR INDIVIDUAL ICI BIOMARKERS

PD-L1 IHC

Biological background.

The PD-1 inhibitory receptor and its major ligand PD-L1 comprise an immunosuppressive pathway that is targeted by anti–PD-(L)1 therapies. At present, several such agents, which are all monoclonal antibodies (mAbs), are approved by the FDA for use in a variety of cancers on the basis of PD-L1 positivity, and four IHC assays for PD-L1 are approved as companion or complementary diagnostics (Table 1). Multiple small molecule inhibitors of this pathway are in preclinical development, but none is approved at the time of the Summit.²⁰

Analytical issues.

Assays for PD-L1 are performed either as FDA-approved assays using one of four available mAbs: SP142 and SP263 (Roche Tissue Diagnostics/Ventana Medical Systems Inc, Tucson, AZ²¹), and 22C3 pharmDx and 28-8 pharmDx (Dako, Agilent, North America Inc, Santa Clara, CA²²; Table 1), or one of these or other antibodies as LDTs. Of note, the antibody/autostainer pairing completes the FDA-approved assay. If a laboratory modifies the FDA-approved assay in any way, including using a different autostainer or on a different tumor type, then the assay is treated as an LDT and must be validated as such.

Although there is substantial evidence that PD-L1 IHC tests are clinically useful, there are challenges in performing PD-L1 IHC. Both approved commercial assays and LDTs have variable limits of detection (LOD). Autostainers have been shown to vary by ± 2 standard deviations from run to run.^23,24 Further contributing to variable PD-L1 IHC test results, elements such as cold ischemia time, time of tissue fixation, antigen retrieval protocols, etc contribute to the overall analytical variability even within FDA-approved assays. Another problem is the dynamic nature of PD-L1 expression. Clearly, PD-L1 expression can change over time and across anatomic sites,²⁵ and can be measured differently according to the cell type being scored. These issues likely contribute to poor reproducibility and the growing complexity of PD-L1 expression as a predictive factor for ICI efficacy.

At present, PD-L1 IHC is scored visually as a tumor proportion score (TPS) for mAb 22C3 (or tumor score for mAb 28-8), an immune cell score (ICS or, specific to mAb SP142, immune cell [IC]), or a combined positive score (CPS; Table 2, Fig 2). Each of these scoring systems may have very different associations with outcomes, and each may differ according to the cancer's tissue of origin.²⁶ Two statistically powered, multi-institutional studies assessed PD-L1 IHC assay reproducibility.^27,28 Both studies found that TPS is read with high concordance; however, ICS scores were not highly concordant, even with additional training. The 22C3, 28-8, and SP263 assays were generally equivalent, while the SP142 assay had consistently lower PD-L1 scores for both tumor cells and ICs.

FIG 2.

Examples illustrating challenges of PD-L1 staining. (A) PD-L1 IHC image from a NSCLC, with TC membranous staining (red box) and IC staining (blue box). Image was scanned from a whole slide of a NSCLC stained with monoclonal antibody 22C3, approximating 20× objective magnification. This tissue could be scored tumor proportion score > 50%. However, the image illustrates how scoring might differ if the pathologist only estimates the proportion of ICs instead of precisely counting each one, since some macrophages (blue arrow) show PD-L1 expression but are harder to count than those in the blue box. By contrast, use of a scoring system in which the precise number of cells is counted (not estimated) for a combined positive score or IC score is time-consuming and likely to vary between readers. (B) PD-L1 IHC image from a NSCLC (20× objective). In this case, the staining appears to be predominantly of macrophages mixed with TCs, in which the dendritic processes of the macrophages are intercalating the TCs. It is difficult to determine how many macrophages are present in this image, which is likely to produce discordant interobserver results. IC, immune cell; IHC, immunohistochemistry; NSCLC, non–small-cell lung cancer; PD-L1, programmed death ligand-1; TC, tumor cell.

The manufacturer-specified method for CPS after mAb staining involves manually counting cell types. The results from the 2021 CAP IHC PD-L1 PT program for NSCLC indicate substantial agreement in TPS and high agreement for CPS, but very poor agreement for ICS. The lack of reproducibility for ICS may be related to the visual estimations involved and more than 95% of respondents estimated or eyeballed CPS and IC scores, rather than explicitly counting individual cells and calculating a score, which is the method included in FDA-approved instructions. Surveys of pathologists indicate that most do not follow this approach and instead provide a visual estimation, possibly leading to additional inaccuracy.^29,30 Taken together, these observations suggest that TPS read by pathologists may be sufficiently reproducible to be used as a clinical test. By contrast, it was noted that CPS and ICS are poorly reproducible, largely because of the challenges in accurately differentiating ICs from other native cell types³¹ (and unpublished data).

In summary, PD-L1 expression has varying definitions and positivity thresholds across clinical trials, and the relevance of any biomarker information can change during the different phases of drug development, depending on the context (for example, adjuvant v metastatic), whether the drug is to be given as a single agent or in combination, and if so with what, and other variables. At least part of this variability can be explained by the relatively complex nature of visualizing PD-L1 staining patterns/intensity, and the complexity of the scoring systems.

Clinical validation.

One of the greatest impacts of anti–PD-(L)1 therapies has been in NSCLC. No biomarkers were used for patient selection in the early CHECKMATE 003 study, and pretreated patients with advanced NSCLC receiving nivolumab had a 5-year overall survival (OS) of 16%.³² However, during the development of pembrolizumab in NSCLC, the importance of PD-L1 as a predictive biomarker for ICI emerged, suggesting that PD-L1 expression should determine what therapy patients with advanced-stage NSCLC without targetable driver mutations should be offered.^33-36 For example, in the KEYNOTE-024 trial, patients with advanced metastatic NSCLC with a TPS ≥ 50% who were treated with pembrolizumab had a 5-year OS of 31.9% compared with 16.3% for those who received chemotherapy only.³⁷ In a subsequent study (Keynote-042), patients with metastatic NSCLC with PD-L1 staining of ≥ 1% were randomly assigned to either single-agent pembrolizumab or to combination chemotherapy.³⁸ Although all patients in the pembrolizumab arm, compared with the chemotherapy arm, had improved OS, the hazard radio for OS declined from 0.68 to 0.75 to 0.80 for patients with ≥ 50%, ≥ 20%, and ≥ 1% PDL1 staining, respectively, suggesting a stepwise decline in benefit for those with lower PD-L1 expression.³⁸

PD-L1 IHC has not been as potent a predictive marker for ICI benefit in other cancer types.^39,40 This variability in results may be a function of several issues, including the use of different mAbs, different staining techniques, scoring of different cell types, and selection of different positivity cutoffs among the various trials. Furthermore, as noted, it is likely that one size will not fit all when applying these biomarkers to cancers of different origins. Recognition and delineation of whether it is or it is not important to provide analytical and scoring systems specific to the cancer's tissue of origin must be clarified.

Potential resolutions to problems and looking to the future.

Using a single biomarker such as PD-L1 IHC was a good starting point for ICI optimization. However, clearly there needs to be better standardization of the analytical issues inherent to PD-L1 analysis. For example, rather than using a single positive and negative control, a cell line array with a range of PD-L1 expression levels, or antigenic peptides coupled to glass beads, might be more beneficial for biochemistry standardization and improve reproducibility.²³ For laboratories that do not quantify images, standardization arrays could be used to monitor run-to-run reproducibility of threshold (LOD) spots. Laboratories performing digital quantification should run two to three repeats of the validated assay with each run to generate a reference data set against which to normalize investigational results. Although such efforts have been reported,²⁴ broad prospective testing and proof of clinical utility of doing so are lacking, and further studies should be performed to generate data supporting intrainstrumental and cross-institutional consistency of the algorithms. Using a single positive cell line, lymph node, or tonsil as a positive control specimen may not be sufficient.

Automated imaging systems, if they could be developed for PD-L1, could reduce the variability and bias in scoring these assays, and there is some literature to support this potential future solution.²⁹ Furthermore, since PD-L1 expression is dynamic, performing real-time biopsies for analysis rather than using tissue biopsies from months or even years in the past should improve the predictive value of this marker. Although this approach is intuitively superior, studies are ongoing to investigate this hypothesis. Importantly, the specific cell type (tumor and/or infiltrating leukocytes) needs to be scored using molecular features rather than visual morphologic interpretation, preferably using standardized automatic methods.

To properly address many of these issues, it was strongly suggested that pharmaceutical companies should be encouraged to provide tissue specimens for prospective-retrospective studies.

To conclude, one speaker opined that, in respect to NSCLC, “No one who can tolerate it (ICI) should not get immunotherapy.” Both the speaker and the panel acknowledged, however, that this only pertains because of the vagaries of PD-L1 and other biomarker determinations, and that such an approach may disappear with more standardized TBT evaluations.

MSI/dMMR Testing and Determination of TMB

Overall biological background.

In theory, ICI should be more effective against cancers that differ substantially from their normal cell counterparts by virtue of developing a large number of mutations encoding neoantigens, eliciting a robust immune response. Thus, testing the cancer for such tumor-specific antigens could provide an effective predictive factor for ICI. At present, there are only limited data from multiparameter protein or transcriptional analyses to directly assess neoantigen expression. Therefore, assessments of correlates of tumor neoantigen presentation, such as DNA mismatch repair phenotypes or elevated TMB, have been used to do so. Although dMMR is a genetic mechanism that leads to MSI-H and elevated TMB via genomic instability, there are numerous other environmental exposure etiologies that lead to elevated TMB and neoantigenicity, such as smoking in NSCLC and exposure to ultraviolet light in melanoma. Nonetheless, in theory, both biomarkers more practically quantify genetic instability.

MMR IHC/MSI

Biological background for MMR/MSI.

The MMR proteins MutL homolog 1 (MLH1), postmeiotic segregation increased (PMS)1 homolog 2 (PMS2), MutS homolog 2 (MSH2), and MutS homolog 6 (MSH6) are involved in a critically important DNA repair pathway, and the loss of one or more of these proteins leads to MSI.⁴¹ Patients with a hereditary propensity for MSI-H/dMMR tumors because of germline mutation in MMR genes (Lynch syndrome) comprise a unique genomic subset that is at risk for developing colorectal and other cancers at a young age.⁴² MSI-H/dMMR tumors also arise sporadically, either through MLH1 promoter methylation (most commonly) or biallelic inactivation of an MMR gene limited to the tumor (ie, so called Lynch-like syndrome).

Analytic validity.

MMR status is determined by IHC, and MSI can be determined by polymerase chain reaction (PCR) or NGS. Compared with PD-L1 and TMB, many pathologists have extensive experience with MMR IHC and MSI PCR, as these tests have long been used to screen patients for Lynch syndrome. Until 2017, all MMR IHC assays were LDTs with many laboratories performing this testing having long, successful track records.

FDA first granted approval in 2017 for pembrolizumab in patients with advanced MSI-H/dMMR cancers based solely on this biomarker independent of tissue of origin. To date, there are FDA approvals for the treatment of MSI-H/dMMR metastatic colorectal cancer with pembrolizumab (first-line), and with nivolumab alone or combined with ipilimumab (second-/third-line; Table 3).^43-45 Dostarlimab has an FDA approval for MSI-H/dMMR metastatic endometrial cancer⁴⁶ and as a tumor agnostic strategy for use in patients with MSI-H/dMMR solid tumors. Ventana Medical Systems/Roche Tissue Diagnostics subsequently received FDA clearance for an MMR IHC assay coupled with BRAF V600E IHC for the purpose of identifying dMMR colon cancers as a screen for Lynch syndrome, and in 2021, the same MMR IHC assay without the BRAF component received FDA approval as a CDx to select dMMR patients for dostarlimab-gxly (FDA approvals available).^47-49

There are a few issues laboratories may encounter when performing MMR IHC. Because loss of expression of both MSH2 and MSH6 is rare, it is very difficult to source material for assay validation. False negatives because of poor tissue fixation, cautery, and low cellularity can lead to diagnostic confusion. Although MSI by PCR may have fewer technical challenges than IHC, it requires both tumor and normal DNA, and fewer laboratories are resourced to perform this testing. MSI by NGS is a new technology and testing still needs to be carefully validated in multiple tumor types (ie, the microsatellite panel selected to determine MSI in one tumor may not perform as well in another tumor type).

In the CAP MMR PT program, half of the laboratories performed MMR IHC on the basis of the Ventana clones used in these FDA-approved/FDA-cleared assays. Of note, the Ventana MLH1 M1 clone and PMS2 A16-4 clone are demonstrably inferior to other clones, producing frequent artifactual dot-like nuclear signal with the former and high cytoplasmic background with the latter in MLH1-/PMS2-deficient cancers.^50-52 Both of these patterns are often falsely interpreted as intact MMR. Other concerns include use of nonformalin fixatives, prolonged cold ischemia time, incomplete fixation, selection of less successful mAb clones, and failure to recognize uncommon abnormal patterns of MMR protein expression. For example, MMR IHC generally performs better on biopsy than resection material, as tissues tend to be more promptly and uniformly fixed.⁵³ The CAP has moved to a 10-core TMA-based survey in 2022, using genetically modified cells lines, so as to be able to challenge all the patterns of MMR expression.

MSI-H assessment by PCR or NGS is through interrogation of loci within the genome that are particularly prone to replication errors because of MMR defects (ie, microsatellites—typically mononucleotide and dinucleotide repeats). There is one FDA-approved CDx assay (FoundationOne CDx [F1CDx], Foundation Medicine Inc, Cambridge MA) specifically for pembrolizumab, and three FDA-cleared assays (OncoMate MSI Dx, Promega Corp, Madison WI; MSK-IMPACT, Memorial Sloan Kettering Cancer Institute, NYC, NY; and PGDx elio, Personal Genome Diagnostics, Baltimore, MD) for MSI on the market (Table 3). These three FDA-cleared tests (OncoMate, MSK-IMPACT, and PGDx elio) are cleared for tumor profiling only and not prescriptive CDxs, implying that the manufacturers did not provide any data for clinical validation of the device for ICI therapy.

Clinical validity.

Data from clinical trials have shown that patients with advanced MSI-H/dMMR tumors, compared with tumors that are microsatellite stable, are more likely to benefit from immunotherapy.^43-46,54,55 The pan-cancer incidence of MSI-H/dMMR is relatively low (approximately 4% of all patients with metastatic cancer will have MSI-H/dMMR tumors), but this biomarker has been shown to be a powerful predictor of ICI response, especially in colorectal cancers (over 50% of patients with MSI-H tumors will benefit from ICI).^44,54,56,57 By contrast, patients with MSI-low tumors are less likely to, although the exact rate of benefit varies by tumor type.⁵⁶ Taken together, these data suggest that MSI-H is specific for benefit from ICI (in other words, the positive predictive value is quite high). However, because of a relatively low sensitivity, other biomarkers of ICI activity should be evaluated in a patient with a negative result.

Although preclinical data support an association of MSI-H with expression of PD-L1 and other immune checkpoint molecules in colon cancer,^58,59 comparative clinical data are sparse. Data comparing the two MSI testing approaches (IHC v genetic) are also sparse. Furthermore, there is no consensus regarding whether these assays provide different predictive accuracies among different kinds of cancers or different ICI therapeutics.

Potential Resolutions to Problems and Looking to the Future

There are numerous ongoing efforts to improve MMR/MSI assay performance and utilization. The CAP Center Guideline: MMR and MSI Testing in Patients Being Considered for Checkpoint Inhibitor Therapy makes recommendations regarding the preference of MMR IHC or MSI PCR (or their equivalence) on a tumor-by-tumor basis.^60-62 It also emphasizes the importance of validating MSI NGS in multiple tumor types and provides guidance on how to deal with indeterminate and discordant MMR/MSI results. In addition, predictive marker immunohistochemistry performed on cytology specimens is subject to increased scrutiny. The CAP All Common Checklist item COM.01520, which requires PT or alternative assessment for predictive markers, was revised in 2021 to extend this requirement to include immunocytology assays.⁶²

The CAP has also enhanced the Anatomic Pathology Checklist.⁶³ Within this, item ANP.22969, which pertains to immunohistochemistry and in situ hybridization for all predictive markers, requires all laboratories to document the cold ischemia time and length of formalin fixation time in pathology reports. The Canadian Biomarker Quality Assurance program has launched several image-based online modules to challenge pathologist's biomarker interpretation skills, including a recently released MMR challenge.⁶⁴ The CAP is working to develop similar interpretation-centric programs.

Materials for IHC controls and validations are critical. Array Science ⁶⁵ has recently produced a cell line–based control in which each of the cores in a 4-core TMA has been specifically knocked out for one of the DNA MMR genes. The Anatomic Pathology Patient Interest Association recently published a review on IHC validations for rare antigens, which included a useful table of vendors of TMA blocks and slides that might be useful for validation purposes.⁶⁶ Moreover, the CAP is changing the format of its IHC MMR PT Program from a whole section (only one tumor challenge per mailing) to a 10-core, TMA-based program.

Tumor Mutational Burden

Biological background.

Similar to MSI, TMB is considered a proxy for tumor immunogenicity since, if transcribed and translated, more mutations should lead to higher numbers of neoantigens, provoking a stronger immune response.

Analytical validity.

There are a number of assays to assess TMB, and at least one ICI therapy, pembrolizumab, received agnostic (independent of tissue of origin) FDA approval to treat TMB-H (defined as TMB ≥ 10 mutations/megabases [Mb]) advanced solid cancers (Table 4).⁶⁷

One method to estimate TMB is by performing NGS.⁶⁸ However, NGS is a technology, not a specific assay, and technical differences between assays can lead to different TMB results. The utility of TMB was established using whole-exome sequencing with paired tumor and normal tissues.⁶⁷ However, this approach is costly and time-consuming to deploy clinically.

For practical purposes, targeted panels consisting of several hundred genes are used to estimate TMB. This approach raises concerns, as panels vary in size, gene content, technical sensitivity, and bioinformatic algorithms, among other factors. Variability arises from a number of sources. The size of a panel is directly correlated with the statistical confidence in a TMB estimate. Because a typical 500-gene panel is much smaller than the exome, confidence intervals are wide for TMB, particularly at lower cutoffs.

Bioinformatic platforms and algorithms also substantially affect TMB analysis.⁶⁹ Different bioinformatic tools operate with different sensitivities and parameters, such that a complex variant can be counted differently (or missed altogether). Filtration of different variants (pathogenic or germline, for example) is necessary, but different approaches lead to different biases. Methodologies to address sampling error, selection bias, and noise vary across laboratories. Given this widespread variability, true standardization across assays may not be feasible at this time.

Biological variability also affects the estimate of TMB. The distribution and immunologic consequences of different mutations may vary across cancer types, or even among patients harboring cancers arising from the same tissue type.^70,71 Different neoplastic processes may be associated with mutations in different genes (which may or may not be included in a given panel), or with different mutational patterns. Population disparities have also been noted: White patients are over-represented in cancer databases, sometimes leading to TMB inflation in patients with non-White ancestry.^68,72

Although ≥ 10 mutations/Mb has been used to determine high TMB using the TMB-H biomarker, cutoff determination is controversial. This TMB-H cutoff was defined using multiple approaches to balance enrichment and sensitivity. The Friends of Cancer Research (Friends) has led the TMB Harmonization Project, intended to standardize the approach and achieve consensus around a cutoff.⁶⁸ However, there are data suggesting a continuous increase in ICI response rates from approximately 20% to approximately 40% in patients within the TMB range of 5 to ≥ 20 mutations/Mb (unpublished data presented at the conference⁷³). Indeed, response rates of ≥ 10% are reported even in patients with < 5 mutations/Mb. In other words, a cutoff of ≥ 10 mutations/Mb might exclude a modest number of patients who could benefit from anti–PD-1 therapy.

In summary, preanalytic, analytic, biologic, and bioinformatic factors contribute variability to TMB estimates (Fig 3). Taken together, these factors may limit the concept that high TMB equates with high levels of antigenicity.

FIG 3.

Factors affecting TMB estimate. TMB, tumor mutational burden.

Clinical Validity.

Despite the analytical issues discussed above, there is evidence that patients with TMB-H compared with TMB-low tumors have an increased likelihood of benefit from immunotherapy.^74,75 However, the consensus of the Panel was that TMB is a less reliable biomarker than MSI for prediction of ICI activity. Nonetheless, the FDA has approved or cleared three assays for TMB (Table 4). The FoundationOne CDx (F1CDx) test was approved by the FDA as a CDx for pembrolizumab in patients with solid tumors having a TMB ≥ 10 mutations/Mb on the basis of the KEYNOTE-158 pembrolizumab study in 10 tumor types.^59,68,75,76 Activities to satisfy several postmarketing requirements are ongoing, including data from other tumor types, pediatric patients, follow-up data, and treatment response assessment using alternative TMB cutoff values. Two other tests (Omics Core, NantHealth Inc, Morrisville, NC; and PGDx elio Tissue Complete, Personal Genome Diagnostics) have been FDA-cleared for only TMB tumor profiling but are not prescriptive CDxs, implying that the manufacturers did not provide any data for clinical validation of the device for ICI therapy.

Further concerns are related to pan-cancer TMB use to select ICI. Because the KEYNOTE-158 study included patients who had exhausted all other treatment options, very few patients with common tumor types (breast, colon, prostate, and lung) were enrolled.⁷⁷ However, after the Summit, a retrospective study of data combined from 12 clinical trials was published addressing TMB performed by whole-exome sequencing in formalin-fixed, paraffin-embedded tissue from 1772 patients with 24 different tumor types, including many of the more common malignancies.⁷⁸ The results from this study suggested that TMB ≥ 175 mutations/exome was predictive of outcomes to pembrolizumab. The response to pembrolizumab was 31.4.% (95% CI, 27.1 to 36.0) compared with 9.5% (95% CI, 8.0 to 11.2) for patients with high versus low (≥ 175 v < 175) TMB regardless of tumor type. The incidence of TMB ≥ 175 varied widely among the different tissues of origin. Furthermore, responses to pembrolizumab on the basis of elevated TMB have been reported in patients with breast cancers⁷⁹ and colon cancers.⁸⁰

Potential Resolutions to Problems and Looking to the Future for Biomarker Tests Reflecting Tumor Neoantigen Burden

Because of the challenges inherent in selecting cutoffs for continuous variables such as TMB, especially since TMB values vary by cancer type,⁸¹ further elucidation of clinically relevant cutoffs in different cancers is necessary. The TMB Harmonization Project has provided best practices and recommendations to reduce variability in TMB testing, and a free-source calibration tool is available to conduct calibrations analyses and improve reproducibility.^68,81

FUTURE DIRECTIONS: DEVELOPMENT OF NOVEL ICI PREDICTIVE FACTOR ASSAYS

In addition to the discussions regarding the currently approved PD-L1 IHC, MSI-H/dMMR, and TMB-high biomarkers for anti–PD-(L)1 treatment eligibility, several other promising biomarker initiatives were discussed (Appendix Table A2). Gene expression profiling tests to generate a predictive score for ICI benefit are being investigated, with a focus on interferon-gamma–related gene expression signatures. For example, RNAseq can define an elevated T-cell–inflamed gene expression profile in the tumor microenvironment (TME).⁸² Tumors positive for an inflamed TME were enriched for single-agent anti–PD-1 response across multiple cancer types.⁸³ Of note, tumor mutational load and inflammatory gene expression signature appear to be independent variables.⁸² Indeed, a clinical trial has been conducted to address the differential efficacy of pembrolizumab-based combination therapy on the basis of two analytically validated, next-generation biomarkers (T-cell–inflamed gene expression profile and TMB; ClinicalTrials.gov identifier: NCT03516981).

Additional factors in the TME are being explored. For example, CXCL10 transcripts associated with interferon-gamma and the T-cell–inflamed phenotype have been correlated with anti–PD-1 efficacy in melanoma.⁸⁴ Research continues into factors underlying interpatient heterogeneity in the T-cell–inflamed TME, including tumor cell–intrinsic differences, microbiota variations, and host germline DNA differences.

As additional immune checkpoint molecules are being targeted in clinical trials, the development of specifically relevant biomarkers has gained importance. Lymphocyte-activation gene-3 (LAG-3) and T-cell immunoglobulin and immunoreceptor tyrosine-based inhibitory motif domain are immune checkpoints expressed by T cells that are now being targeted in the clinic with blocking mAbs. In a phase III study of patients with advanced melanoma, relatlimab (a human immunoglobulin G4 mAb directed against LAG-3) combined with nivolumab demonstrated clinically meaningful improvement in progression-free survival compared with nivolumab alone, leading to FDA approval for this regimen without a qualifying biomarker in 2022.⁸⁵ IHC tests have been developed for both LAG-3 and T-cell immunoglobulin and immunoreceptor tyrosine-based inhibitory motif domain, and ongoing studies aim to correlate test results with therapeutic outcomes in patients receiving these therapies.

Increasingly, many trials are studying immunotherapy in the neoadjuvant setting, and end points commonly include pathologic response (% residual viable tumor) in the complete surgical specimen. This strategy has led to a proliferation of scoring systems to discriminate degrees of pathologic response after the neoadjuvant treatment.⁸⁶ Shifting between approaches is cumbersome and may make cross-trial comparisons difficult. To address this issue, the Society for Immunotherapy of Cancer has sponsored PATHdata, a collaborative multistakeholder working group developing recommendations for standardized pathologic response data collection and reporting across tumor types.

Most of the approaches to determine residual disease after neoadjuvant therapy have been performed using light microscopy with hematoxylin and eosin staining. Future investigations will include machine learning to potentially automate readings to further standardize such assessments.

As noted, the main targets for ICI are neoantigens expressed in the tumor. Specific efforts to identify and quantitate them have been reported.⁸⁷ In addition to identification of individual mutated proteins, advances in multiplex immunofluorescence IHC provide an opportunity for tissue sparing and may lead to more comprehensive characterization of the tumor including the immune TME.⁸⁷

Intratumoral heterogeneity presents challenges to the application of single gene/protein biomarkers.⁸⁸ However, liquid biopsy technology, in particular circulating tumor DNA (ctDNA), has evolved very quickly.⁸⁹ Liquid biopsies hold potential for evaluating TMB and tracking subclonal genetic alterations representing the patient's entire tumor burden, rather than a single tumor biopsy site. In this regard, plasma collection for ctDNA profiling is being used in randomized clinical studies to assess associations with patient outcomes, possibly yielding a useful biomarker for patient stratification. For example, in the phase III JAVELIN Renal 101 trial, certain protein-altering mutations identified in tissue predicted outcomes of patients with renal cell carcinoma, both in the experimental arm (avelumab [anti–PD-L1] plus axitinib) and the control arm (sunitinib monotherapy).⁹⁰ Although not reported in this study, detection in ctDNA of many of these mutations represents a potential application of the liquid biopsy concept. Additionally, interrogation of ctDNA dynamic changes before and during treatment with ICI has been reported in preliminary investigations to provide information predictive of response to ICI.⁹¹

Another form of liquid biopsy involves determination of circulating leukocyte patterns and phenotypic profiles. Highly preliminary but provocative studies have suggested that elevated pretreatment levels of regulatory T-cells (Treg), elevated neutrophil/lymphocyte ratios, loss of T-cell CD27 and CD28, expression of leukocyte Tim-3 and CD57, and/or elevated human leukocyte antigens-DR isotope on monocytes or dendritic cells may all portend resistance to ICI therapies.^92-95 These observations require substantial validation in future clinical trials.

The Cancer Immune Monitoring and Analysis Centers and Cancer Immunologic Data Commons network was established by the NCI with the Partnership for Accelerating Cancer Therapies.⁹⁶ The Network addressed scientific challenges stemming from the complex interplay between the cancer and the host immune system. This interplay is further confounded by individual laboratory research methods, interlaboratory variability in assay performance and reproducibility, tumor and interpatient heterogeneity, and the small size of patient cohorts in early-phase clinical trials. Cancer Immune Monitoring and Analysis Center laboratories provide comprehensive analysis of NCI-sponsored clinical trial specimens, with assays categorized on the basis of the priority for implementation in trials. The Cancer Immunologic Data Commons supports a centralized database for biomarker and associated clinical data integration, including bioinformatics tools for correlative analysis within and across trials. Comprehensive analyses of multiple clinical trials have been conducted, demonstrating the success of this collaborative model. It is hoped that the strategies and resources developed by this network will help inform the discovery of next-generation immunotherapy biomarkers and provide a model for biomarker data integration that will be needed to unify test development and clinical correlations across the immunotherapy research community.

Given immune system complexity, a combination approach to predictive biomarkers for ICI might include PD-L1 expression plus other factors, such as MSI/dMMR and/or TMB, and even completely unrelated markers, such as tumor-infiltrating lymphocyte subsets, host microbiome, and germline HLA, may help address the limitations of PD-L1 as a standalone marker. As biomarker test results are the basis for potentially life-changing decisions, improvement is certainly needed. An affordable, accessible integrated panel of biomarkers may result in better response prediction and ultimately improved medical management. These considerations raise the concept of the importance of looking at combined data sets and the need for collaboration to confirm these markers.

Moreover, future technological advancements will lead to more uniform reference standards, which will be critical to align assay performance across sites and facilitate the success of multicenter clinical trials. Transparency in reporting of biomarker studies, particularly data from subgroups, would enhance data interpretation. Wider access to raw data from published studies would enable meta-analyses and possible refinement of cutoff values. The field would also benefit from additional development and adoption of standards for clinical validity and clinical utility, as these are lacking for many diagnostic tests that are designed to provide prediction of benefit from targeted therapeutics, leading to more precise application of these treatment modalities.

In conclusion, the data presented and resulting discussions at this Summit exemplify the biomarker maturation process, where more is learned regarding the limitations of a test as it is exposed to increasing biological variability in both clinical trials and standard clinical practice. It is apparent that the current tests have significant utility but also multiple limitations and biases. Although these issues confound clear understanding of when to use a predictive test and which test to use, they reflect the complexity of cancer and immune system biology and the novelty of ICI, rather than failures of the biomarkers themselves. The scoring systems have become increasingly complex, involve multiple estimations, and in many cases are no longer binary. There is also variability in how the assays are performed, with different mAb clones, microsatellite targets, and quantification methods. Improved reference materials that are well characterized and designed to address inherent test variability and test the LODs are needed, to ensure adequate assessment of test performance. To achieve this, the concept of reference materials may need to expand from traditional tissue-based standards to other materials such as genetically modified cultured cell lines. Furthermore, it is hoped that automated reading systems may complement or replace visual scoring, further standardizing assessment of in vitro assays.

What cannot be lost is that the goal of precision immunotherapy is to recommend the most appropriate treatment for each patient. Although the PD-L1 IHC, MSI-H/dMMR, and TMB-H biomarkers have brought the field closer to this goal, all could be improved technically, and their clinical utility better understood.

The following are considered some areas for improvement for each of the three assays:

• Pursue a group biomarker rather than individual CDx test indications

• Standardize assays among laboratories with PT and quality assurance and control strategies

• Develop reliable internal standards and well-characterized reference materials

• Conduct more prospective-retrospective, or even prospective randomized trials to investigate remaining questions around different cutoffs predictive of clinical outcomes for different tumor types. Doing so will require that companies and academics establish collaborative groups in a precompetitive setting to generate data sets with sufficient power to address the important questions remaining.

Engagement of stakeholders across the entire research community will be needed to bring much-needed clarity and new advances in biomarker development for ICI therapy, which has emerged as a major treatment modality and a new standard of care in oncology.

APPENDIX

TABLE A1.

Summit Structure, Agenda, and Participants

TABLE A2.

Novel Approaches for the Potential Development of Biomarkers for Immune Checkpoint Inhibition Therapies

FIG A1.

Ordering guide for PD-L1 within Auroral Health/ACL Laboratories/Great Lakes Pathologists (Courtesy of Dr Frank Zuehl). PD-L1, programmed death ligand-1.

AUTHOR CONTRIBUTIONS

Conception and design: Roy S. Herbst, Jonathan L. Myles, Suzanne L. Topalian, Sophia L. Yohe, Andrew M. Bellizzi, Upal Basu Roy, Robin H. Edwards, Ehab A. El-Gabry, Julia Elvin, Lisa M. McShane, Matthew Oberley, Reena Philip, Jason N. Rosenbaum, Eric H. Rubin, Shimere W. Sherwood, Magdalena Thurin, Patricia Vasalos, Jordan Laser

Administrative support: Georganne Bradshaw, Shimere W. Sherwood, Patricia Vasalos

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).