eFigure 1. Schematic Layout (A) and H&E Staining (B) With Lung Tumor (Blue), Tissue Controls (Red), and Cell Line Controls (Green)
eFigure 2. Representative Images From YTMA 337 of DAB and QIF Immunostaining Including Positive Controls, Placenta, and H820 Cell Line, Negative Control Ramos, and Positive Tumor Staining
eFigure 3. Signal to Noise Curves Across Range of Concentrations Tested for Four PD-L1 Antibodies in Tumor Cases
eFigure 4. Concordance of Four PD-L1 Antibodies as Determined by QIF Score Comparison in YTMA337 in Tumor Cases (A) and Cell Lines (B)
eFigure 5. Schematic Layout of 15-Spot PD-L1 Reference TMA (A) and DAB Staining at Low Power Objective and QIF Staining at High Power Objective of 15-Spot Reference PD-L1 IHC TMA for PD-L1 Antibodies (B)
eFigure 6. Concordance of Four PD-L1 Antibodies as Determined by QIF in Immunostaining (A) and Six PD-L1 Antibodies as Determined by DAB Quantified Membrane Intensity Staining in 15-Spot PD-L1 IHC Reference TMA (B
eFigure 7. 28_8 Discordance Between 28_8 Staining on YTMA 337 Over Four Independent Runs and Three Independent Operator (A), Concordance of SP142 Staining as Assessed by Two Independent Operators Over Two Independent Runs in Tumor (Left) and Cell Line Cores (Right) (B), and 28_8 QIF Staining Regressed to SP142 QIF Results Obtained From YTMA 337 c) Representative Images From 28_8 DAB and QIF Staining (C)
eMaterial. Supplement Methods and Data
Customize your JAMA Network experience by selecting one or more topics from the list below.
Gaule P, Smithy JW, Toki M, et al. A Quantitative Comparison of Antibodies to Programmed Cell Death 1 Ligand 1. JAMA Oncol. 2017;3(2):256–259. doi:10.1001/jamaoncol.2016.3015
Copyright 2017 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.
Do the 4 drug-matched companion diagnostic antibodies for PD-1 axis therapies all produce the same results?
In this study, 3 key components of diagnostic tests were assessed: the primary antibody, assay-specific variables related to the staining platform, and immunohistochemical assessment of tissue. All antibodies tested were concordant.
Discordance of the companion diagnostic test is not attributable to the antibody but rather to inherent tumor heterogeneity or assay- or platform-specific variables.
Assessment of PD-L1 (programmed cell death 1 ligand 1) expression by immunohistochemical analysis has been used as a predictive diagnostic test to identify responders and guide treatment in trials of the PD-1 (programmed cell death 1) axis inhibitors. The definition of PD-L1 positive lacks standardization, and prediction of response by immunohistochemical analysis is additionally limited by the subjective nature of this technique.
To examine whether PD-L1 antibody reagents are interchangeable by quantitatively comparing the expression of the PD-L1 protein.
Design, Setting, and Participants
In this immunohistochemistry standardization study, 30 randomly selected cases of lung cancer resected from January 1, 2008, through December 31, 2009, were obtained from Yale Pathology Archives with a range of expression of PD-L1. To test for protein measurement, rather than clinical utility, a PD-L1 index tissue microarray, including cell line and tissue controls, was used. The results were then validated on a commercially available, genetically defined PD-L1 engineered cell line array with a range of controlled protein-expressing cell lines using 6 monoclonal antibodies (SP142, E1L3N, 9A11, SP263, 22c3, and 28-8). Protein levels were measured by quantitative immunofluorescence and quantitative chromogenic assessment. Data analysis was performed from September 2015 through May 2016.
Concordance between 4 antibodies revealed regression for tumor tissue cores (R2 = 0.42-0.91) and cell line cores (R2 = 0.83-0.97) by quantitative immunofluorescence in the PD-L1 index tissue microarray. All 6 antibodies had high levels of concordance (R2 = 0.76-0.99) when using chromogenic staining in isogenic cell lines.
Conclusions and Relevance
Because the antibodies are highly concordant, these results suggest that assays based on the use of these antibodies could yield concordant results. They further suggest that previously described differences in PD-L1 expression in tissue are independent of the antibody used and likely attributable to tumor heterogeneity, assay- or platform-specific variables, or other factors.
Assessment of PD-L1 (programmed cell death 1 ligand 1) expression by immunohistochemical (IHC) analysis has been used as a predictive diagnostic test to identify responders and guide treatment in trials of the PD-1 (programmed cell death 1) axis inhibitors nivolumab, atezolizumab, durvalumab, and pembrolizumab in non–small cell lung cancer (NSCLC).1-4 The PD-L1 IHC 22C3 PharmDx kit (Dako North America) was approved by the US Food and Drug Administration (FDA) as a companion diagnostic for pembrolizumab in NSCLC, whereas the PD-L1 28-8 PharmDx kit (Dako North America) and the PD-L1 SP142 Ventana test (Ventana Medical Systems Inc) were approved as complementary diagnostics for nivolumab and atezolizumab, repectively.
The definition of PD-L1 positive lacks standardization, and prediction of response by IHC analysis is additionally limited by the subjective nature of the technique. Variable cutoffs for defining positive cases across trials have been used when measuring tumor and/or immune cells in the stroma.4-6 Although some PD-L1 antibodies have been rigorously validated in the published literature, including 28-8 and E1L3N, others are less documented, and specific epitope sequences remain proprietary.7,8
Significant differences in case classification have been observed with 2 validated PD-L1 antibodies, indicating that discordance was a function of tissue heterogeneity or variability among the antibodies.9 To examine the effect of epitope targeting or potential nonspecific binding, we developed a tissue microarray with a range of positive and negative specimens, including tumor, normal tissue, and cell lines (eFigure 1 in the Supplement). We analyzed 6 PD-L1 monoclonal antibodies (intracellular and extracellular domain specific) to determine the concordance among antibodies.
Thirty randomly selected cases of lung cancer resected from January 1, 2008, through December 31, 2009, were obtained from Yale Pathology Archives with a range of expression of PD-L1 as assessed in previous studies.9,10 Tissue and cell line blocks, processed identically, were prepared in a tissue microarray format that contained 0.6-mm representative formalin-fixed, paraffin-embedded cores in 2-fold redundancy. Gene editing technology (Horizon Discovery Plc) was used to develop a genetically defined 15-spot cell line microarray (CLMA) PD-L1 IHC reference standard with a range of controlled protein expression levels (negative, low, medium, and high protein-expressing cell lines). Individual cell lines in the 15-spot CLMA were extensively characterized and verified using molecular assays, IHC analysis, and quantitative digital histologic analysis. All patients had signed consent or waiver of consent forms for tissue use, and the study was approved by the Yale Human Investigation Committee. All data were deidentified.
Data analysis was performed from September 2015 through May 2016. PD-L1 expression was evaluated by chromogenic IHC analysis and quantitative immunofluorescence (QIF) using 6 monoclonal antibodies raised against PD-L1 as summarized in the Table. Antibodies were titrated at a range of concentrations, and optimal assay concentration was determined using an algorithm that uses signal to noise ratio and dynamic range. We successfully performed QIF for 4 of the 6 antibodies using automated quantitative analysis as previously described.9 Chromogenic IHC analysis was quantified using the Aperio Positive Pixel Count based on the intensity of membrane staining of respective PD-L1 antibodies using a modified HER2 algorithm.11 A detailed description of these methods and immunostaining examples are included in eFigure 2 and eFigure 3 in the Supplement.
Correlation among the antibodies was measured by linear regression. Tumor cores had a lower concordance than cell lines, with the lowest being E1L3N, which had a correlation coefficient of 0.42, 0.68, and 0.58 to SP263, 9A11, and SP142, respectively (eFigure 4A in the Supplement). In cell lines, the concordance increased to 0.83, 0.90 and 0.86 for SP263, 9A11, and SP142, respectively (eFigure 4B in the Supplement). To confirm this concordance, we tested the 4 antibodies in the 15-spot CLMA (eFigure 5 in the Supplement contains a schematic of this array and representative immunostaining images). Regression among the antibodies tested on the 15-spot CLMA had high concordance, with R2 values of 0.89 and 0.94 (eFigure 6A in the Supplement), confirming the results found in the Yale index tumor microarray. The antibody 28-8 results were reproducible only under 3,3′-diaminobenzidine staining conditions (eFigure 7 in the Supplement). The antibody 22c3 was also tested using the FDA-approved platform and included in the study analysis. For all chromogenic assessment, we used the Aperio Pixel Counter Software and assessed 3,3′-diaminobenzidine staining of the 15-spot CLMA. All staining was performed in a single run on the Bond Rx (Leica Biosystems Inc) as described in the eMaterial in the supplement except 22c3. With this method, all 6 antibodies (E1l3N, SP142, 9A11, SP263, 28-8, and 22c3) had high levels of concordance among one another, with R2 values of 0.76 to 0.99 in agreement with QIF results (eFigure 6B in the Supplement).
There are currently multiple assays under investigation or approved as companion or complementary diagnostic tests for PD-L1. It is concerning that the FDA clearance of an assay on the basis of assay performance appears to have become more important than the accurate and reproducible measurement of the target. As a result, at least 4 separate antibodies have been included in assays that are part of separate FDA submissions, creating a challenge for pathologists who may need to perform 4 different assays rather than assess PD-L1 expression. Imagine a situation where a pathologist required separate assays for the dozen or so drugs that target the estrogen receptor in breast cancer.
Lung cancer cell lines when tested with the 4 antibodies tested by QIF had high concordance for PD-L1 binding. The Abcam 28-8 antibody elicited reproducible results only when tested on the isogenic cell lines. This finding suggests that 28-8 recognizes PD-L1 in a manner similar to the other antibodies tested. Furthermore, reproducibility in our hands was only seen within a single vial, despite the purchase of 4 vials from the same manufacturer’s lot. The E1L3N antibody appears to be the only outlier, with R2 values as low as 0.42 on tumor specimens. However, high concordance in cell lines and 3,3′-diaminobenzidine assessments and extensive previous validation of E1L3N support the hypothesis that low regressions are a function of tumor heterogeneity. The possibility that E1L3N recognizes a PD-L1 variant or another B7 family member cannot be definitively excluded.12 Overall, this result indicates the need for a tissue series of the tumor type under investigation to be included in antibody validation.
Limitations of this study include the relatively small patient series used, but high levels of concordance and reproducibility among the antibodies suggest that this does not adversely affect the results, except as noted above. Another limitation was our inability to compare 28-8 and 22c3 by QIF. Finally, this work does not define the positive-negative threshold for PD-L1 or the reproducibility of assessment around that threshold. Future work will address the issue of assay threshold reproducibility.
This study provides good evidence that most antibodies used for PD-L1 studies are highly similar in their ability to bind PD-L1. A previous study9 found approximately 25% discordance between E1L3N and SP142 on tissue specimens quantified in more than 500 fields of view. In that work, the authors concluded, “Objective determination of PD-L1 protein levels in NSCLC reveals heterogeneity within tumors and prominent inter-assay variability.”9(p 46) This new data allow us to suggest that the discordance seen in that study is likely attributable to the high levels of heterogeneity of expression of PD-L1 that has also been described in other works9,10,13 rather than antibody-based interassay variability. This result suggests that assays based on the use of the antibodies tested should be concordant, barring the substantial effects of the heterogeneity of PD-L1 expression or subjective scoring systems or other assay- or platform-specific variables.
Corresponding Author: David L. Rimm, MD, PhD, Department of Pathology, Yale University School of Medicine, 310 Cedar St, PO Box 208023, BML 116, New Haven, CT 06520 (firstname.lastname@example.org).
Accepted for Publication: June 3, 2016.
Published Online: August 18, 2016. doi:10.1001/jamaoncol.2016.3015
Author Contributions: Dr Rimm had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Rimm.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Gaule, Rehman, Patell-Socha, Morrill, Rimm.
Critical revision of the manuscript for important intellectual content: Smithy, Toki, Rehman, Cougot, Collin, Neumeister, Rimm.
Statistical analysis: Gaule, Smithy, Toki, Rehman.
Obtaining funding: Rimm.
Administrative, technical, or material support: Smithy, Toki, Patell-Socha, Cougot, Collin, Morrill, Neumeister, Rimm.
Study supervision: Neumeister, Rimm.
Conflict of Interest Disclosures: Dr Rimm reported serving as a paid consultant or adviser to Genoptix/Novartis, Applied Cellular Diagnostics, BMS, Amgen, Optrascan, Biocept, Clearsight, Perkin Elmer, and Metamark Genetics. No other disclosures were reported.
Funding/Support: This work was supported by grant P50CA196530 from the Yale SPORE in Lung Cancer, grant P30CA016359 from the Yale Cancer Center, the Breast Cancer Research Foundation, and a sponsored research agreement from Genoptix.
Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Create a personal account or sign in to: