Intra- and Interobserver Reproducibility Assessment of PD-L1 Biomarker in Non-Small Cell Lung Cancer

An artificial intelligence-powered PD-L1 combined positive score (CPS) analyser in urothelial carcinoma alleviating interobserver and intersite variability

AIMS

Immune checkpoint inhibitors targeting programmed death-ligand 1 (PD-L1) have shown promising clinical outcomes in urothelial carcinoma (UC). The combined positive score (CPS) quantifies PD-L1 22C3 expression in UC, but it can vary between pathologists due to the consideration of both immune and tumour cell positivity.

METHODS AND RESULTS

An artificial intelligence (AI)-powered PD-L1 CPS analyser was developed using 1,275,907 cells and 6175.42 mm2 of tissue annotated by pathologists, extracted from 400 PD-L1 22C3-stained whole slide images of UC. We validated the AI model on 543 UC PD-L1 22C3 cases collected from three institutions. There were 446 cases (82.1%) where the CPS results (CPS ≥10 or <10) were in complete agreement between three pathologists, and 486 cases (89.5%) where the AI-powered CPS results matched the consensus of two or more pathologists. In the pathologist's assessment of the CPS, statistically significant differences were noted depending on the source hospital (P = 0.003). Three pathologists reevaluated discrepancy cases with AI-powered CPS results. After using the AI as a guide and revising, the complete agreement increased to 93.9%. The AI model contributed to improving the concordance between pathologists across various factors including hospital, specimen type, pathologic T stage, histologic subtypes, and dominant PD-L1-positive cell type. In the revised results, the evaluation discordance among slides from different hospitals was mitigated.

CONCLUSION

This study suggests that AI models can help pathologists to reduce discrepancies between pathologists in quantifying immunohistochemistry including PD-L1 22C3 CPS, especially when evaluating data from different institutions, such as in a telepathology setting.

Artificial Intelligence and Lung Pathology

This manuscript provides a comprehensive overview of the application of artificial intelligence (AI) in lung pathology, particularly in the diagnosis of lung cancer. It discusses various AI models designed to support pathologists and clinicians. AI models supporting pathologists are to standardize diagnosis, score PD-L1 status, supporting tumor cellularity count, and indicating explainability for pathologic judgements. Several models predict outcomes beyond pathologic diagnosis and predict clinical outcomes like patients' survival and molecular alterations. The manuscript emphasizes the potential of AI to enhance accuracy and efficiency in pathology, while also addressing the challenges and future directions for integrating AI into clinical practice.

Clinical Validation of Artificial Intelligence-Powered PD-L1 Tumor Proportion Score Interpretation for Immune Checkpoint Inhibitor Response Prediction in Non-Small Cell Lung Cancer

PURPOSE

Evaluation of PD-L1 tumor proportion score (TPS) by pathologists has been very impactful but is limited by factors such as intraobserver/interobserver bias and intratumor heterogeneity. We developed an artificial intelligence (AI)-powered analyzer to assess TPS for the prediction of immune checkpoint inhibitor (ICI) response in advanced non-small cell lung cancer (NSCLC).

MATERIALS AND METHODS

The AI analyzer was trained with 393,565 tumor cells annotated by board-certified pathologists for PD-L1 expression in 802 whole-slide images (WSIs) stained by 22C3 pharmDx immunohistochemistry. The clinical performance of the analyzer was validated in an external cohort of 430 WSIs from patients with NSCLC. Three pathologists performed annotations of this external cohort, and their consensus TPS was compared with AI-based TPS.

RESULTS

In comparing PD-L1 TPS assessed by AI analyzer and by pathologists, a significant positive correlation was observed (Spearman coefficient = 0.925; P < .001). The concordance of TPS between AI analyzer and pathologists according to TPS ≥50%, 1%-49%, and <1% was 85.7%, 89.3%, and 52.4%, respectively. In median progression-free survival (PFS), AI-based TPS predicted prognosis in the TPS 1%-49% or TPS <1% group better than the pathologist's reading, with the TPS ≥50% group as a reference (hazard ratio [HR], 1.49 [95% CI, 1.19 to 1.86] v HR, 1.36 [95% CI, 1.08 to 1.71] for TPS 1%-49% group, and HR, 2.38 [95% CI, 1.69 to 3.35] v HR, 1.62 [95% CI, 1.23 to 2.13] for TPS <1% group).

CONCLUSION

PD-L1 TPS assessed by AI analyzer correlates with that of pathologists, with clinical performance also being comparable when referenced to PFS. The AI model can accurately predict tumor response and PFS of ICI in advanced NSCLC via assessment of PD-L1 TPS.

A Deep Learning-Based Assay for Programmed Death Ligand 1 Immunohistochemistry Scoring in Non-Small Cell Lung Carcinoma: Does it Help Pathologists Score?

Several studies have developed various artificial intelligence (AI) models for immunohistochemical analysis of programmed death ligand 1 (PD-L1) in patients with non-small cell lung carcinoma; however, none have focused on specific ways by which AI-assisted systems could help pathologists determine the tumor proportion score (TPS). In this study, we developed an AI model to calculate the TPS of the PD-L1 22C3 assay and evaluated whether and how this AI-assisted system could help pathologists determine the TPS and analyze how AI-assisted systems could affect pathologists' assessment accuracy. We assessed the 4 methods of the AI-assisted system: (1 and 2) pathologists first assessed and then referred to automated AI scoring results (1, positive tumor cell percentage; 2, positive tumor cell percentage and visualized overlay image) for final confirmation, and (3 and 4) pathologists referred to the automated AI scoring results (3, positive tumor cell percentage; 4, positive tumor cell percentage and visualized overlay image) while determining TPS. Mixed-model analysis was used to calculate the odds ratios (ORs) with 95% CI for AI-assisted TPS methods 1 to 4 compared with pathologists' scoring. For all 584 samples of the tissue microarray, the OR for AI-assisted TPS methods 1 to 4 was 0.94 to 1.07 and not statistically significant. Of them, we found 332 discordant cases, on which the pathologists' judgments were inconsistent; the ORs for AI-assisted TPS methods 1, 2, 3, and 4 were 1.28 (1.06-1.54; P = .012), 1.29 (1.06-1.55; P = .010), 1.28 (1.06-1.54; P = .012), and 1.29 (1.06-1.55; P = .010), respectively, which were statistically significant. For discordant cases, the OR for each AI-assisted TPS method compared with the others was 0.99 to 1.01 and not statistically significant. This study emphasized the usefulness of the AI-assisted system for cases in which pathologists had difficulty determining the PD-L1 TPS.

Comparing deep learning and pathologist quantification of cell-level PD-L1 expression in non-small cell lung cancer whole-slide images

Programmed death-ligand 1 (PD-L1) expression is currently used in the clinic to assess eligibility for immune-checkpoint inhibitors via the tumor proportion score (TPS), but its efficacy is limited by high interobserver variability. Multiple papers have presented systems for the automatic quantification of TPS, but none report on the task of determining cell-level PD-L1 expression and often reserve their evaluation to a single PD-L1 monoclonal antibody or clinical center. In this paper, we report on a deep learning algorithm for detecting PD-L1 negative and positive tumor cells at a cellular level and evaluate it on a cell-level reference standard established by six readers on a multi-centric, multi PD-L1 assay dataset. This reference standard also provides for the first time a benchmark for computer vision algorithms. In addition, in line with other papers, we also evaluate our algorithm at slide-level by measuring the agreement between the algorithm and six pathologists on TPS quantification. We find a moderately low interobserver agreement at cell-level level (mean reader-reader F1 score = 0.68) which our algorithm sits slightly under (mean reader-AI F1 score = 0.55), especially for cases from the clinical center not included in the training set. Despite this, we find good AI-pathologist agreement on quantifying TPS compared to the interobserver agreement (mean reader-reader Cohen's kappa = 0.54, 95% CI 0.26-0.81, mean reader-AI kappa = 0.49, 95% CI 0.27-0.72). In conclusion, our deep learning algorithm demonstrates promise in detecting PD-L1 expression at a cellular level and exhibits favorable agreement with pathologists in quantifying the tumor proportion score (TPS). We publicly release our models for use via the Grand-Challenge platform.

Semi-automated approaches for interrogating spatial heterogeneity of tissue samples

Tissues are spatially orchestrated ecosystems composed of heterogeneous cell populations and non-cellular elements. Tissue components' interactions shape the biological processes that govern homeostasis and disease, thus comprehensive insights into tissues' composition are crucial for understanding their biology. Recently, advancements in the spatial biology field enabled the in-depth analyses of tissue architecture at single-cell resolution, while preserving the structural context. The increasing number of biomarkers analyzed, together with whole tissue imaging, generate datasets approaching several hundreds of gigabytes in size, which are rich sources of valuable knowledge but require investments in infrastructure and resources for extracting quantitative information. The analysis of multiplex whole-tissue images requires extensive training and experience in data analysis. Here, we showcase how a set of open-source tools can allow semi-automated image data extraction to study the spatial composition of tissues with a focus on tumor microenvironment (TME). With the use of Lunaphore COMET platform, we interrogated lung cancer specimens where we examined the expression of 20 biomarkers. Subsequently, the tissue composition was interrogated using an in-house optimized nuclei detection algorithm followed by a newly developed image artifact exclusion approach. Thereafter, the data was processed using several publicly available tools, highlighting the compatibility of COMET-derived data with currently available image analysis frameworks. In summary, we showcased an innovative semi-automated workflow that highlights the ease of adoption of multiplex imaging to explore TME composition at single-cell resolution using a simple slide in, data out approach. Our workflow is easily transferrable to various cohorts of specimens to provide a toolset for spatial cellular dissection of the tissue composition.

Concordance of assessments of four PD-L1 immunohistochemical assays in esophageal squamous cell carcinoma (ESCC)

OBJECTIVE

Given real-world limitations in programmed death-ligand 1 (PD-L1) testing, concordance studies between PD-L1 assays are needed. We undertook comparisons of PD-L1 assays (DAKO22C3, Ventana SP263, Ventana SP142, E1L3N) among observers in esophageal squamous cell carcinoma (ESCC) to provide information on the analytical and clinical comparability of four PD-L1 IHC assays.

METHODS

Paraffin embedded samples of 50 cases of esophageal squamous cell carcinoma were obtained, satined with all four PD-L1 assays. PD-L1 was evaluated by 68 pathologists from 19 different hospitals. PD-L1 expression was assessed for combined positive score (CPS).

RESULTS

The expression sensitivity of SP263 was the highest in ESCC, followed by 22C3, E1L3N and SP142. Taking CPS 10 as the critical value, inter-observer concordance for CPS scores among 68 physicians was assessed for the 22C3, SP263, SP142, and E1L3N assays, yielding values of 0.777, 0.790, 0.758, and 0.782, respectively. In the comparison between assays, the overall CPS scores concordance rates between 22C3 and SP263, SP142, and E1L3N were 0.896, 0.833, and 0.853, respectively. 22C3 and SP263 have high concordance, with OPA of 0.896, while E1L3N and SP142 have the highest concordance, with OPA of 0.908.

CONCLUSION

In ESCC, the concordance of PD-L1 evaluation among observers is good, and the immune cell score is still an important factor affecting the concordance of interpretation among observers. Cases near the specific threshold are still the difficult problem of interpretation. SP263 had the highest CPS score of the four assays. SP263 cannot identify all 22C3 positive cases, but had good concordance with 22C3.E1L3N and SP142 showed high concordance.

Weakly Supervised Deep Learning Predicts Immunotherapy Response in Solid Tumors Based on PD-L1 Expression

Programmed death-ligand 1 (PD-L1) IHC is the most commonly used biomarker for immunotherapy response. However, quantification of PD-L1 status in pathology slides is challenging. Neither manual quantification nor a computer-based mimicking of manual readouts is perfectly reproducible, and the predictive performance of both approaches regarding immunotherapy response is limited. In this study, we developed a deep learning (DL) method to predict PD-L1 status directly from raw IHC image data, without explicit intermediary steps such as cell detection or pigment quantification. We trained the weakly supervised model on PD-L1-stained slides from the non-small cell lung cancer (NSCLC)-Memorial Sloan Kettering (MSK) cohort (N = 233) and validated it on the pan-cancer-Vall d'Hebron Institute of Oncology (VHIO) cohort (N = 108). We also investigated the performance of the model to predict response to immune checkpoint inhibitors (ICI) in terms of progression-free survival. In the pan-cancer-VHIO cohort, the performance was compared with tumor proportion score (TPS) and combined positive score (CPS). The DL model showed good performance in predicting PD-L1 expression (TPS ≥ 1%) in both NSCLC-MSK and pan-cancer-VHIO cohort (AUC 0.88 ± 0.06 and 0.80 ± 0.03, respectively). The predicted PD-L1 status showed an improved association with response to ICIs [HR: 1.5 (95% confidence interval: 1-2.3), P = 0.049] compared with TPS [HR: 1.4 (0.96-2.2), P = 0.082] and CPS [HR: 1.2 (0.79-1.9), P = 0.386]. Notably, our explainability analysis showed that the model does not just look at the amount of brown pigment in the IHC slides, but also considers morphologic factors such as lymphocyte conglomerates. Overall, end-to-end weakly supervised DL shows potential for improving patient stratification for cancer immunotherapy by analyzing PD-L1 IHC, holistically integrating morphology and PD-L1 staining intensity.

SIGNIFICANCE

The weakly supervised DL model to predict PD-L1 status from raw IHC data, integrating tumor staining intensity and morphology, enables enhanced patient stratification in cancer immunotherapy compared with traditional pathologist assessment.

PD-L1 Assessment in Needle Core Biopsies of Non-Small Cell Lung Cancer: Interpathologist Agreement and Potential Associated Histopathological Features

OBJECTIVE

Immune checkpoint inhibitors are used in the treatment of non-small cell lung cancer (NSCLC). Programmed cell death-ligand 1 (PD-L1) immunohistochemistry (IHC) assessed by pathologists is subject to interobserver variability. In advanced/metastatic disease and inoperable patients, PD-L1 assessment relies on biopsy specimens, commonly needle core biopsies (NCB). We aimed to determine the interobserver agreement for PD-L1 tumor proportion score (TPS) in NSCLC NCBs and identify histopathological features that may be related to interobserver variability.

MATERIAL AND METHODS

Sixty NSCLC NCBs with PD-L1 IHC were evaluated independently by four pathologists from different institutions. PD-L1 TPS was evaluated in three categories: no/low expression ( < 1%), intermediate expression (1%49%), and high expression (≥50%). Histological tumor type, necrosis, tumor-infiltrating lymphocytes, tumor length/percentage in the biopsy, and crush/squeeze artifact was evaluated.

RESULTS

The statistical analysis of the three PD-L1 TPS categories demonstrated moderate agreement (Fleiss Kappa 0.477) in the no/low category, fair agreement (Fleiss Kappa 0.390) in the intermediate category, and almost perfect agreement (Fleiss Kappa 0.952) in the high category. A significant correlation (p=0.003) was found between the crush/squeeze artifact in NCB and rate of discordant TPS categories. There was no significant correlation between pathologists' agreement in the TPS categories and histological tumor type, tumor length, tumor ratio, necrosis, and tumor-infiltrating lymphocytes.

CONCLUSION

Our results demonstrated moderate agreement among pathologists for the PD-L1 TPS 1% cut-off in NSCLC NCB, which is lower than that reported in resection materials. The presence of crush/squeeze artifact in NCBs is significantly related to the rate of discordant TPS categories, suggesting that PD-L1 assessment of pulmonary NCBs requires an awareness of this artifact.

Gene Expressions and High Lymphocyte Count May Predict Durable Clinical Benefits in Patients with Advanced Non-Small-Cell Lung Cancer Treated with Immune Checkpoint Inhibitors

BACKGROUND

Not all patients with advanced non-small cell lung cancer (NSCLC) benefit from immune checkpoint inhibitors (ICIs). Therefore, we aimed to assess the predictive potential of gene expression profiling (GEP), peripheral immune cell counts, and clinical characteristics.

METHODS

The primary endpoint of this prospective, observational study was a durable clinical benefit (DCB) defined as progression-free survival >6 months. In a subgroup with histological biopsies of sufficient quality (n = 25), GEP was performed using the nCounter® PanCancer IO 360 panel.

RESULTS

DCB was observed in 49% of 123 included patients. High absolute lymphocyte count (ALC) and absence of liver metastases were associated with DCB (OR = 1.95, p = 0.038 and OR = 0.36, p = 0.046, respectively). GEP showed clustering of differentially expressed genes according to DCB, and a strong association between PD-L1 assessed by GEP (CD274) and immunohistochemistry (IHC) was observed (p = 0.00013). The TGF-β, dendritic cell, and myeloid signature scores were higher for patients without DCB, whereas the JAK/STAT loss signature scores were higher for patients with DCB (unadjusted p-values < 0.05).

CONCLUSIONS

ALC above 1.01 × 109/L and absence of liver metastases were significantly associated with DCB in ICI-treated patients with NSCLC. GEP was only feasible in 20% of the patients. GEP-derived signatures may be associated with clinical outcomes, and PD-L1 could be assessed by GEP rather than IHC.

Impact of Decalcification, Cold Ischemia, and Deglycosylation on Performance of Programmed Cell Death Ligand-1 Antibodies With Different Binding Epitopes: Comparison of 7 Clones

Programmed cell death ligand-1 (PD-L1) expression levels in patients' tumors have demonstrated clinical utility across many cancer types and are used to determine treatment eligibility. Several independently developed PD-L1 immunohistochemical (IHC) predictive assays are commercially available and have demonstrated different levels of staining between assays, generating interest in understanding the similarities and differences between assays. Previously, we identified epitopes in the internal and external domains of PD-L1, bound by antibodies in routine clinical use (SP263, SP142, 22C3, and 28-8). Variance in performance of assays utilizing these antibodies, observed following exposure to preanalytical factors such as decalcification, cold ischemia, and duration of fixation, encouraged additional investigation of antibody-binding sites, to understand whether binding site structures/conformations contribute to differential PD-L1 IHC assay staining. We proceeded to further investigate the epitopes on PD-L1 bound by these antibodies, alongside the major clones utilized in laboratory-developed tests (E1L3N, QR1, and 73-10). Characterization of QR1 and 73-10 clones demonstrated that both bind the PD-L1 C-terminal internal domain, similar to SP263/SP142. Our results also demonstrate that under suboptimal decalcification or fixation conditions, the performance of internal domain antibodies is less detrimentally affected than that of external domain antibodies 22C3/28-8. Furthermore, we show that the binding sites of external domain antibodies are susceptible to deglycosylation and conformational structural changes, which directly result in IHC staining reduction or loss. The binding sites of internal domain antibodies were unaffected by deglycosylation or conformational structural change. This study demonstrates that the location and conformation of binding sites, recognized by antibodies employed in PD-L1 diagnostic assays, differ significantly and exhibit differing degrees of robustness. These findings should reinforce the need for vigilance when performing clinical testing with different PD-L1 IHC assays, particularly in the control of cold ischemia and the selection of fixation and decalcification conditions.

Immune Biomarkers in Triple-Negative Breast Cancer: Improving the Predictivity of Current Testing Methods

Triple-negative breast cancer (TNBC) poses a significant challenge in terms of prognosis and disease recurrence. The limited treatment options and the development of resistance to chemotherapy make it particularly difficult to manage these patients. However, recent research has been shifting its focus towards biomarker-based approaches for TNBC, with a particular emphasis on the tumor immune landscape. Immune biomarkers in TNBC are now a subject of great interest due to the presence of tumor-infiltrating lymphocytes (TILs) in these tumors. This characteristic often coincides with the presence of PD-L1 expression on both neoplastic cells and immune cells within the tumor microenvironment. Furthermore, a subset of TNBC harbor mismatch repair deficient (dMMR) TNBC, which is frequently accompanied by microsatellite instability (MSI). All of these immune biomarkers hold actionable potential for guiding patient selection in immunotherapy. To fully capitalize on these opportunities, the identification of additional or complementary biomarkers and the implementation of highly customized testing strategies are of paramount importance in TNBC. In this regard, this article aims to provide an overview of the current state of the art in immune-related biomarkers for TNBC. Specifically, it focuses on the various testing methodologies available and sheds light on the immediate future perspectives for patient selection. By delving into the advancements made in understanding the immune landscape of TNBC, this study aims to contribute to the growing body of knowledge in the field. The ultimate goal is to pave the way for the development of more personalized testing strategies, ultimately improving outcomes for TNBC patients.

Between Generating Noise and Generating Images: Noise in the Correct Frequency Improves the Quality of Synthetic Histopathology Images for Digital Pathology

Artificial intelligence and machine learning techniques have the promise to revolutionize the field of digital pathology. However, these models demand considerable amounts of data, while the availability of unbiased training data is limited. Synthetic images can augment existing datasets, to improve and validate AI algorithms. Yet, controlling the exact distribution of cellular features within them is still challenging. One of the solutions is harnessing conditional generative adversarial networks that take a semantic mask as an input rather than a random noise. Unlike other domains, outlining the exact cellular structure of tissues is hard, and most of the input masks depict regions of cell types. This is also the case for non-small cell lung cancer, the most common type of lung cancer. Deciding whether a patient would receive immunotherapy depends on quantifying regions of stained cells. However, using polygon-based masks introduce inherent artifacts within the synthetic images - due to the mismatch between the polygon size and the single-cell size. In this work, we show that introducing random single-pixel noise with the appropriate spatial frequency into a polygon semantic mask can dramatically improve the quality of the synthetic images. We used our platform to generate synthetic images of immunohistochemistry-treated lung biopsies. We test the quality of the images using a three-fold validation procedure. First, we show that adding the appropriate noise frequency yields 87% of the similarity metrics improvement that is obtained by adding the actual single-cell features. Second, we show that the synthetic images pass the Turing test. Finally, we show that adding these synthetic images to the train set improves AI performance in terms of PD-L1 semantic segmentation performances. Our work suggests a simple and powerful approach for generating synthetic data on demand to unbias limited datasets to improve the algorithms' accuracy and validate their robustness.

Efficacy of PD-1/PD-L1 inhibitors in advanced gastroesophageal cancer based on characteristics: a meta-analysis

Objective: Evaluate the potency of anti-PD-1/PD-L1 antibodies in advanced gastroesophageal cancer patients with different clinical features. Methods: Randomized, controlled trials comparing anti-PD-1/PD-L1 antibodies with chemotherapy in individuals with gastroesophageal cancer were retrieved. Results: 15 trials involving 9194 individuals were included. PD-1/PD-L1 inhibitors significantly improved overall survival (OS) but not progression-free survival. Significantly improved OS was observed in PD-L1 combined positive score ≥1, primary esophageal cancer, primary gastric cancer and Asian patients. Subgroup analysis revealed significant OS benefit achieved for esophageal squamous cell carcinoma, but not for esophageal adenocarcinoma. Conclusion: PD-1/PD-L1 inhibitors improved OS in advanced gastroesophageal carcinoma, especially in patients with esophageal cancer. Race, primary tumor sites and PD-L1 combined positive score can be used to predict the potency of immune checkpoint inhibitors.

Predicting response to immune checkpoint blockade in NSCLC with tumour-only RNA-seq

BACKGROUND

Targeted RNA sequencing (RNA-seq) from FFPE specimens is used clinically in cancer for its ability to estimate gene expression and to detect fusions. Using a cohort of NSCLC patients, we sought to determine whether targeted RNA-seq could be used to measure tumour mutational burden (TMB) and the expression of immune-cell-restricted genes from FFPE specimens and whether these could predict response to immune checkpoint blockade.

METHODS

Using The Cancer Genome Atlas LUAD dataset, we developed a method for determining TMB from tumour-only RNA-seq and showed a correlation with DNA sequencing derived TMB calculated from tumour/normal sample pairs (Spearman correlation = 0.79, 95% CI [0.73, 0.83]. We applied this method to targeted sequencing data from our patient cohort and validated these results against TMB estimates obtained using an orthogonal assay (Spearman correlation = 0.49, 95% CI [0.24, 0.68]).

RESULTS

We observed that the RNA measure of TMB was significantly higher in responders to immune blockade treatment (P = 0.028) and that it was predictive of response (AUC = 0.640 with 95% CI [0.493, 0.786]). By contrast, the expression of immune-cell-restricted genes was uncorrelated with patient outcome.

CONCLUSION

TMB calculated from targeted RNA sequencing has a similar diagnostic ability to TMB generated from targeted DNA sequencing.

Artificial intelligence in cancer immunotherapy: Applications in neoantigen recognition, antibody design and immunotherapy response prediction

Cancer immunotherapy is a method of controlling and eliminating tumors by reactivating the body's cancer-immunity cycle and restoring its antitumor immune response. The increased availability of data, combined with advancements in high-performance computing and innovative artificial intelligence (AI) technology, has resulted in a rise in the use of AI in oncology research. State-of-the-art AI models for functional classification and prediction in immunotherapy research are increasingly used to support laboratory-based experiments. This review offers a glimpse of the current AI applications in immunotherapy, including neoantigen recognition, antibody design, and prediction of immunotherapy response. Advancing in this direction will result in more robust predictive models for developing better targets, drugs, and treatments, and these advancements will eventually make their way into the clinical setting, pushing AI forward in the field of precision oncology.

The Ring Study: an international comparison of PD-L1 diagnostic assays and their interpretation in non-small cell lung cancer, head and neck squamous cell cancer and urothelial cancer

PD-L1 immunohistochemistry has been approved as a diagnostic assay for immunotherapy. However, an international comparison across multiple cancers is lacking. This study aimed to assess the performance of PD-L1 diagnostic assays in non-small cell lung cancer (NSCLC), head and neck squamous cell cancer (HNSCC) and urothelial cancer (UC). The excisional specimens of NSCLC, HNSCC and UC were assayed by Ventana SP263 and scored at three sites in each country, including Australia, Brazil, Korea, Mexico, Russia and Taiwan. All slides were rotated to two other sites for interobserver scoring. The same cohort of NSCLC was assessed with Dako 22C3 pharmDx PD-L1 for comparison. The PD-L1 immunopositivity was scored according to the approved PD-L1 scoring algorithms which were the percentage of PD-L1-expressing tumour cell (TC) and tumour proportion score (TPS) by Ventana SP263 and Dako 22C3 staining, respectively. In NSCLC, the comparison demonstrated the comparability of the SP263 and 22C3 assays (cut-off of 1%, κ=0.71; 25%, κ=0.75; 50%, κ=0.81). The interobserver comparisons showed moderate to almost perfect agreement for SP263 in TC staining at 25% cut-off (NSCLC, κ=0.72 to 0.86; HNSCC, κ=0.60 to 0.82; UC, κ=0.68 to 0.91) and at 50% cut-off for NSCLC (κ=0.64 to 0.90). Regarding the immune cell (IC) scoring in UC, there was a lower correlation (concordance correlation coefficient=0.10 to 0.68) and poor to substantial agreements at the 1%, 5%, 10% and 25% cut-offs (κ= -0.04 to 0.76). The interchangeability of SP263 and 22C3 in NSCLC might be acceptable, especially at the 50% cut-off. In HNSCC, the performance of SP263 is comparable across five countries. In UC, there was low concordance of IC staining, which may affect treatment decisions. Overall, the study showed the reliability and reproducibility of SP263 in NSCLC, HNSCC and UC.

Cell adhesion molecules and immunotherapy in advanced non-small cell lung cancer: Current process and potential application

Advanced non-small cell lung cancer (NSCLC) is a severe disease and still has high mortality rate after conventional treatment (e.g., surgical resection, chemotherapy, radiotherapy and targeted therapy). In NSCLC patients, cancer cells can induce immunosuppression, growth and metastasis by modulating cell adhesion molecules of both cancer cells and immune cells. Therefore, immunotherapy is increasingly concerned due to its promising anti-tumor effect and broader indication, which targets cell adhesion molecules to reverse the process. Among these therapies, immune checkpoint inhibitors (mainly anti-PD-(L)1 and anti-CTLA-4) are most successful and have been adapted as first or second line therapy in advanced NSCLC. However, drug resistance and immune-related adverse reactions restrict its further application. Further understanding of mechanism, adequate biomarkers and novel therapies are necessary to improve therapeutic effect and alleviate adverse effect.

Effect of a standardised heart team protocol versus a guideline-based protocol on revascularisation decision stability in stable complex coronary artery disease: rationale and design of a randomised trial of cardiology specialists using historic cases

INTRODUCTION

A multidisciplinary heart team approach has been recommended by revascularisation guidelines, but how to organise and implement the heart team in a standardised way has not been validated. Inter-team and intra-team decision instability existed in the guideline-based heart team protocol, and our standardised heart team protocol based on a mixed method study may improve decision stability. The objective of this study is to evaluate the effect of the standardised heart team protocol versus the guideline-based protocol on decision-making stability in stable complex coronary artery disease (CAD).

METHODS AND ANALYSIS

Eighty-four eligible interventional cardiologists, cardiac surgeons or non-interventional cardiologists from 26 hospitals in China have been enrolled. They will be randomised to a standardised heart team protocol group or a guideline-based protocol group to make revascularisation decisions for 480 historic cases (from a prospective registry) with stable complex CAD. In the standardised group, we will establish 12 heart teams based on an evidence-based protocol, including specialist selection, specialist training, team composition, team training and a standardised meeting process. In the guideline-based group, we will organise 12 heart teams according to the guideline principles, including team composition and standardised meeting process. The primary outcome is the overall percent agreement in revascularisation decisions between heart teams within a group. To demonstrate the clinical implication of decision-making stability, we will further explore the association between decision stability and 1-year clinical outcomes.

ETHICS AND DISSEMINATION

The study was approved by the Institutional Review Board (IRB) of Fuwai Hospital (No. 2019-1303). All participants have provided informed consent and all patients included as historic cases provided written informed consent at the time of entry to the prospective registry. The results of this trial will be disseminated through manuscript publication and national/international conferences, and reported in the trial registry entry.

TRIAL REGISTRATION NUMBER

NCT05039567.

Immunotherapy targeting inhibitory checkpoints: The role of NK and other innate lymphoid cells

Monoclonal antibodies that target specific ligand-receptor signaling pathways and act as immune checkpoint inhibitors have been designed to remove the brakes in T cells and restore strong and long-term antitumor-immunity. Of note, many of these inhibitory receptors are also expressed by Innate Lymphoid Cells (ILCs), suggesting that also blockade of inhibitory pathways in innate lymphocytes has a role in the response to the treatment with checkpoint inhibitors. ILCs comprise cytotoxic NK cells and "helper" subsets and are important cellular components in the tumor microenvironment. In addition to killing tumor cells, ILCs release inflammatory cytokines, thus contributing to shape adaptive cell activation in the context of immunotherapy. Therefore, ILCs play both a direct and indirect role in the response to checkpoint blockade. Understanding the impact of ILC-mediated response on the treatment outcome would contribute to enhance immunotherapy efficacy, as still numerous patients resist or relapse.

Association of PD-L1 Expression and Other Variables With Benefit From Immune Checkpoint Inhibition in Advanced Gastroesophageal Cancer: Systematic Review and Meta-analysis of 17 Phase 3 Randomized Clinical Trials

Importance

Approval by the US Food and Drug Administration of immune checkpoint inhibition (ICI) for advanced gastroesophageal cancer (aGEC) irrespective of PD-L1 status has generated controversy. Exploratory analyses from individual trials indicate a lack of meaningful benefit from ICI in patients with absent or low PD-L1 expression; however, analysis of a single variable while ignoring others may not consider the instability inherent in exploratory analyses.

Objective

To systematically examine the predictive value of tissue-based PD-L1 status compared with that of other variables for ICI benefit in aGEC to assess its stability.

Data Sources

MEDLINE, Embase, Scopus, Web of Science, Cochrane Central Register (2000-2022).

Study Selection, Data Extraction, and Synthesis

Randomized clinical trials (RCTs) were included of adults with aGEC (adenocarcinoma [AC] or squamous cell carcinoma [SCC]) randomized to anti-PD-1 or PD-L1-containing treatment vs standard of care (SOC). Study screening, data abstraction, and bias assessment were completed independently by 2 reviewers. Of 5752 records screened, 26 were assessed for eligibility; 17 trials were included in the analysis.

Main Outcomes and Measures

The prespecified primary end point was overall survival. The mean hazard ratio (HR) for ICI vs SOC was calculated (random-effects model). Predictive values were quantified by calculating the ratio of mean HRs between 2 levels of each variable.

Results

In all, 17 RCTs (9 first line, 8 after first line) at low risk of bias and 14 predictive variables were included, totaling 11 166 participants (5067 with SCC, 6099 with ACC; 77.6% were male and 22.4% were female; 59.5% of patients were younger than 65 years, 40.5% were 65 years or older). Among patients with SCCs, PD-L1 tumor proportion score (TPS) was the strongest predictor of ICI benefit (HR, 0.60 [95% CI, 0.53-0.68] for high TPS; and HR, 0.84 [95% CI, 0.75-0.95] for low TPS), yielding a predictive value of 41.0% favoring high TPS (vs ≤16.0% for other variables). Among patients with AC, PD-L1 combined positive score (CPS) was the strongest predictor (after microsatellite instability high status) of ICI benefit (HR, 0.73 [95% CI, 0.66-0.81] for high CPS; and HR, 0.95 [95% CI, 0.84-1.07] for low CPS), yielding a predictive value of 29.4% favoring CPS-high (vs ≤12.9% for other variables). Head-to-head analyses of trials containing both levels of a variable and/or having similar design generally yielded consistent results.

Conclusions and Relevance

Tissue-based PD-L1 expression, more than any variable other than microsatellite instability-high, identified varying degrees of benefit from ICI-containing therapy vs SOC among patients with aGEC in 17 RCTs.

High interobserver and intraobserver reproducibility among pathologists assessing PD-L1 CPS across multiple indications

AIMS

A common concern among pathologists scoring PD-L1 immunohistochemical staining is interobserver and intraobserver variability. We assessed interobserver and intraobserver reproducibility of PD-L1 scoring among trained pathologists using combined positive score (CPS; tumour cell and tumour-associated immune cell staining).

METHODS AND RESULTS

Data were collected for 2 years (2017-2019) from 456 pathologists worldwide. Digital training encompassed unique, tumour-specific training and test sets. Samples were stained using PD-L1 IHC 22C3 pharmDx and evaluated at specific CPS cut-offs for gastric cancer (GC), cervical cancer (CC), urothelial cancer (UC), oesophageal cancer (OC), and head and neck squamous cell carcinoma (HNSCC). Pathologists underwent expert-to-peer training and scored 20 blinded samples on day 1 and 25 blinded samples on day 2 (including 15 of the day 1 samples). Interobserver and intraobserver reproducibility were assessed. For GC (120 observers) and CC (32 observers) samples assessed at CPS ≥1, average interobserver agreement was 91.5% and 91.0%, respectively, and average intraobserver agreement was 90.2% and 96.6%, respectively. For UC (139 observers) and OC (52 observers) samples measured at CPS ≥10, average interobserver agreement was 93.4% and 93.7%, respectively, and average intraobserver agreement was 92.0% and 92.5%, respectively. For HNSCC samples (113 observers), average interobserver agreement was 94.1% at CPS ≥1 and 86.5% at CPS ≥20; intraobserver agreement was 94.7% at CPS ≥1 and 90.5% at CPS ≥20.

CONCLUSION

The consistently high interobserver and intraobserver concordance rates support the effectiveness of face-to-face training of many global pathologists for scoring PD-L1 CPS across multiple indications at several specific cut-offs.

Comparison of three validated PD-L1 immunohistochemical assays in urothelial carcinoma of the bladder: interchangeability and issues related to patient selection

Different programmed cell death-ligand 1 (PD-L1) assays and scoring algorithms are being used in the evaluation of PD-L1 expression for the selection of patients for immunotherapy in specific settings of advanced urothelial carcinoma (UC). In this paper, we sought to investigate three approved assays (Ventana SP142 and SP263, and Dako 22C3) in UC with emphasis on implications for patient selection for atezolizumab/pembrolizumab as the first line of treatment. Tumors from 124 patients with invasive UC of the bladder were analyzed using tissue microarrays (TMA). Serial sections were stained with SP263 and SP142 on Ventana Benchmark Ultra and with 22C3 on Dako Autostainer Link 48. Stains were evaluated independently by two observers and scored using the combined positive score (CPS) and tumor infiltrating immune cells (IC) algorithms. Differences in proportions (DP), overall percent agreement (OPA), positive percent agreement (PPA), negative percent agreement (NPA), and Cohen κ were calculated for all comparable cases. Good overall concordance in analytic performance was observed for 22C3 and SP263 with both scoring algorithms; specifically, the highest OPA was observed between 22C3 and SP263 (89.6%) when using CPS. On the other hand, SP142 consistently showed lower positivity rates with high differences in proportions (DP) compared with 22C3 and SP263 with both CPS and IC, and with a low PPA, especially when using the CPS algorithm. In conclusion, 22C3 and SP263 assays show comparable analytical performance while SP142 shows divergent staining results, with important implications for the selection of patients for both pembrolizumab and atezolizumab.

A digital assay for programmed death-ligand 1 (22C3) quantification combined with immune cell recognition algorithms in non-small cell lung cancer

PD-L1 (22C3) checkpoint inhibitor therapy represents a mainstay of modern cancer immunotherapy for non-small cell lung cancer (NSCLC). In vitro diagnostic (IVD) PD-L1 antibody staining is widely used to predict clinical intervention efficacy. However, pathologist interpretation of this assay is cumbersome and variable, resulting in poor positive predictive value concerning patient therapy response. To address this, we developed a digital assay (DA) termed Tissue Insight (TI) 22C3 NSCLC, for the quantification of PD-L1 in NSCLC tissues, including digital recognition of macrophages and lymphocytes. We completed clinical validation of this digital image analysis solution in 66 NSCLC patient samples, followed by concordance studies (comparison of PD-L1 manual and digital scores) in an additional 99 patient samples. We then combined this DA with three distinct immune cell recognition algorithms for detecting tissue macrophages, alveolar macrophages, and lymphocytes to aid in sample interpretation. Our PD-L1 (22C3) DA was successfully validated and had a scoring agreement (digital to manual) higher than the inter-pathologist scoring. Furthermore, the number of algorithm-identified immune cells showed significant correlation when compared with those identified by immunohistochemistry in serial sections stained by double immunofluorescence. Here, we demonstrated that TI 22C3 NSCLC DA yields comparable results to pathologist interpretation while eliminating the intra- and inter-pathologist variability associated with manual scoring while providing characterization of the immune microenvironment, which can aid in clinical treatment decisions.

PD-L1 immunohistochemistry: Clones, cutoffs, and controversies

Cancer immunotherapy has become a major component of oncologic treatment for a growing number of malignancies. Of particular interest to pathology has been monoclonal antibody therapy targeting immune checkpoints, notably programmed cell death (PD-1) and programmed cell death ligand (PD-L1). Targeting of these checkpoints attempt to overcome tumor evasion of the immune system. While PD-L1 testing is currently implemented as a predictive biomarker in multiple indications with the PD-L1 axis blockade, PD-L1 immunohistochemistry has been a complex issue for the pathology laboratory as it requires an understanding of multiple clones, on multiple testing platforms for multiple different malignancies, each with variable scoring criteria and thresholds. This review attempts to summarize the important PD-L1 testing algorithms and test performance for the practicing pathologist who actively reviews PD-L1 immunohistochemistry.

Neoadjuvant Chemo-Immunotherapy for Locally Advanced Non-Small-Cell Lung Cancer: A Review of the Literature

Non-small cell lung cancer accounts for approximately 80–85% of all lung cancers and at present represents the main cause of cancer death among both men and women. To date, surgery represents the cornerstone; nevertheless, around 40% of completely resected patients develop disease recurrence. Therefore, combining neoadjuvant chemo-immunotherapy and surgery might lead to improved survival. Immunotherapy is normally well tolerated, although significant adverse reactions have been reported in certain patients treated with inhibitors of immune checkpoints. In this review, we explore the current literature on the use of neoadjuvant chemo-immunotherapy followed by surgery for treatment of locally advanced non-small-cell lung cancer, with particular attention to the histological aspects, ongoing trials, and the most common surgical approaches. In conclusion, neoadjuvant immunotherapy whether combined or not with chemotherapy reveals a promising survival benefit for patients with advanced non-small-cell lung cancer; nevertheless, more data remain necessary to identify the best candidates for neoadjuvant regimens.

Non-Invasive Measurement Using Deep Learning Algorithm Based on Multi-Source Features Fusion to Predict PD-L1 Expression and Survival in NSCLC

Background

Programmed death-ligand 1 (PD-L1) assessment of lung cancer in immunohistochemical assays was only approved diagnostic biomarker for immunotherapy. But the tumor proportion score (TPS) of PD-L1 was challenging owing to invasive sampling and intertumoral heterogeneity. There was a strong demand for the development of an artificial intelligence (AI) system to measure PD-L1 expression signature (ES) non-invasively.

Methods

We developed an AI system using deep learning (DL), radiomics and combination models based on computed tomography (CT) images of 1,135 non-small cell lung cancer (NSCLC) patients with PD-L1 status. The deep learning feature was obtained through a 3D ResNet as the feature map extractor and the specialized classifier was constructed for the prediction and evaluation tasks. Then, a Cox proportional-hazards model combined with clinical factors and PD-L1 ES was utilized to evaluate prognosis in survival cohort.

Results

The combination model achieved a robust high-performance with area under the receiver operating characteristic curves (AUCs) of 0.950 (95% CI, 0.938–0.960), 0.934 (95% CI, 0.906–0.964), and 0.946 (95% CI, 0.933–0.958), for predicting PD-L1ES <1%, 1–49%, and ≥50% in validation cohort, respectively. Additionally, when combination model was trained on multi-source features the performance of overall survival evaluation (C-index: 0.89) could be superior compared to these of the clinical model alone (C-index: 0.86).

Conclusion

A non-invasive measurement using deep learning was proposed to access PD-L1 expression and survival outcomes of NSCLC. This study also indicated that deep learning model combined with clinical characteristics improved prediction capabilities, which would assist physicians in making rapid decision on clinical treatment options.

Artificial intelligence-assisted system for precision diagnosis of PD-L1 expression in non-small cell lung cancer

Standardized programmed death-ligand 1 (PD-L1) assessment in non-small cell lung cancer (NSCLC) is challenging, owing to inter-observer variability among pathologists and the use of different antibodies. There is a strong demand for the development of an artificial intelligence (AI) system to obtain high-precision scores of PD-L1 expression in clinical diagnostic scenarios. We developed an AI system using whole slide images (WSIs) of the 22c3 assay to automatically assess the tumor proportion score (TPS) of PD-L1 expression based on a deep learning (DL) model of tumor detection. Tests were performed to show the diagnostic ability of the AI system in the 22c3 assay to assist pathologists and the reliability of the application in the SP263 assay. A robust high-performance DL model for automated tumor detection was devised with an accuracy and specificity of 0.9326 and 0.9641, respectively, and a concrete TPS value was obtained after tumor cell segmentation. The TPS comparison test in the 22c3 assay showed strong consistency between the TPS calculated with the AI system and trained pathologists (R = 0.9429-0.9458). AI-assisted diagnosis test confirmed that the repeatability and efficiency of untrained pathologists could be improved using the AI system. The Ventana PD-L1 (SP263) assay showed high consistency in TPS calculations between the AI system and pathologists (R = 0.9787). In conclusion, a high-precision AI system is proposed for the automated TPS assessment of PD-L1 expression in the 22c3 and SP263 assays in NSCLC. Our study also indicates the benefits of using an AI-assisted system to improve diagnostic repeatability and efficiency for pathologists.

Immune Checkpoint Inhibitors in Triple Negative Breast Cancer: The Search for the Optimal Biomarker

Triple negative breast cancer (TNBC) is a high-risk and aggressive malignancy characterized by the absence of estrogen receptors (ER) and progesterone receptors (PR) on the surface of malignant cells, and by the lack of overexpression of human epidermal growth factor 2 (HER2). It has limited therapeutic options compared to other subtypes of breast cancer. There is now a growing body of evidence on the role of immunotherapy in TNBC, however much of the data from clinical trials is conflicting and thus, challenging for clinicians to integrate the data into clinical practice. Landmark phase III trials using immunotherapy in the early-stage neoadjuvant setting concluded that the addition of immunotherapy to chemotherapy improved the pathologic complete response (pCR) rate compared to chemotherapy with placebo while others found no significant improvement in pCR. Phase III trials have investigated the utility of immunotherapy in previously untreated metastatic TNBC, and these studies have similarly arrived at inconsistent conclusions. Some studies showed no benefit while others demonstrated a clinically significant improvement in overall survival in the PD-L1 positive population. It is not yet clear which biomarkers are most useful, and assays for these biomarkers have not been standardized. Given the often serious and severe side effects of immunotherapy, it is important and necessary to identify predictive biomarkers of response and resistance in order to enhance patient selection. In this review, we will discuss both the challenges of traditional biomarkers and the opportunities of emerging biomarkers for patient selection.

Tumour mutational burden: an overview for pathologists

Cancer immunotherapy holds great promise and has shown durable responses in many patients; however, these responses are not uniform in all patients or all tumour streams. There is an ongoing clinical need for objective diagnostic biomarkers to identify patients that will respond to immunotherapies. Tumour mutational burden (TMB) is a diagnostic biomarker that can stratify cancer patients for response to immune checkpoint inhibitor therapies. It is commonly defined as the average number of somatic mutations per megabase in a tumour exome. Here we describe the TMB biomarker, how it is determined, its underlying molecular basis, the relationship to neoantigens and the issues around its clinical use. This overview is directed toward practising pathologists wishing to be informed of this predictive biomarker.

Association of artificial intelligence-powered and manual quantification of programmed death-ligand 1 (PD-L1) expression with outcomes in patients treated with nivolumab ± ipilimumab

Assessment of programmed death ligand 1 (PD-L1) expression by immunohistochemistry (IHC) has emerged as an important predictive biomarker across multiple tumor types. However, manual quantitation of PD-L1 positivity can be difficult and leads to substantial inter-observer variability. Although the development of artificial intelligence (AI) algorithms may mitigate some of the challenges associated with manual assessment and improve the accuracy of PD-L1 expression scoring, use of AI-based approaches to oncology biomarker scoring and drug development has been sparse, primarily due to the lack of large-scale clinical validation studies across multiple cohorts and tumor types. We developed AI-powered algorithms to evaluate PD-L1 expression on tumor cells by IHC and compared it with manual IHC scoring in urothelial carcinoma, non-small cell lung cancer, melanoma, and squamous cell carcinoma of the head and neck (prospectively determined during the phase II and III CheckMate clinical trials). 1,746 slides were retrospectively analyzed, the largest investigation of digital pathology algorithms on clinical trial datasets performed to date. AI-powered quantification of PD-L1 expression on tumor cells identified more PD-L1-positive samples compared with manual scoring at cutoffs of ≥1% and ≥5% in most tumor types. Additionally, similar improvements in response and survival were observed in patients identified as PD-L1-positive compared with PD-L1-negative using both AI-powered and manual methods, while improved associations with survival were observed in patients with certain tumor types identified as PD-L1-positive using AI-powered scoring only. Our study demonstrates the potential for implementation of digital pathology-based methods in future clinical practice to identify more patients who would benefit from treatment with immuno-oncology therapy compared with current guidelines using manual assessment.

Comparison of Semi-Quantitative Scoring and Artificial Intelligence Aided Digital Image Analysis of Chromogenic Immunohistochemistry

Semi-quantitative scoring is a method that is widely used to estimate the quantity of proteins on chromogen-labelled immunohistochemical (IHC) tissue sections. However, it suffers from several disadvantages, including its lack of objectivity and the fact that it is a time-consuming process. Our aim was to test a recently established artificial intelligence (AI)-aided digital image analysis platform, Pathronus, and to compare it to conventional scoring by five observers on chromogenic IHC-stained slides belonging to three experimental groups. Because Pathronus operates on grayscale 0-255 values, we transformed the data to a seven-point scale for use by pathologists and scientists. The accuracy of these methods was evaluated by comparing statistical significance among groups with quantitative fluorescent IHC reference data on subsequent tissue sections. The pairwise inter-rater reliability of the scoring and converted Pathronus data varied from poor to moderate with Cohen’s kappa, and overall agreement was poor within every experimental group using Fleiss’ kappa. Only the original and converted that were obtained from Pathronus original were able to reproduce the statistical significance among the groups that were determined by the reference method. In this study, we present an AI-aided software that can identify cells of interest, differentiate among organelles, protein specific chromogenic labelling, and nuclear counterstaining after an initial training period, providing a feasible and more accurate alternative to semi-quantitative scoring.

The Challenge to the Pathologist of PD-L1 Expression in Tumor Cells of Non-Small-Cell Lung Cancer—An Overview

In recent years, the treatment of non-small-cell lung cancer (NSCLC) has been fundamentally changed by immunotherapy where the immune system is reactivated using anti-programmed cell death protein 1/programmed death ligand 1 (PD1/PD-L1) checkpoint inhibition. With this, the immunohistological detection of PD-L1 has become one of the most important predictive biomarkers, leading pathologists to play a central role in the immuno-oncological therapy decisions. This has brought them the challenge of requiring the knowledge of relevant checkpoint inhibitors (CI), different PD-L1 scores and cut-offs as well as the choice of the right tissues and controls. Their involvement is also required in the careful validation of both clinical trial assays (CTAs) and laboratory developed tests (LDTs), in addition to the internal and external quality assessment and the interpretation and scoring of the staining based on specialist training. After the training of tumor proportion score (TPS) scoring in NSCLC, pathologists show a high level of concordance, with some variation between different cut-offs. Since not all patients benefit from immunotherapy, further research is needed to validate new predictive markers and optimize existing ones. In this context, these studies focus on a combination of PD-L1 and molecular signatures.

Development and validation of a supervised deep learning algorithm for automated whole-slide programmed death-ligand 1 tumour proportion score assessment in non-small cell lung cancer

AIMS

Immunohistochemical programmed death-ligand 1 (PD-L1) staining to predict responsiveness to immunotherapy in patients with advanced non-small cell lung cancer (NSCLC) has several drawbacks: a robust gold standard is lacking, and there is substantial interobserver and intraobserver variance, with up to 20% discordance around cutoff points. The aim of this study was to develop a new deep learning-based PD-L1 tumour proportion score (TPS) algorithm, trained and validated on a routine diagnostic dataset of digitised PD-L1 (22C3, laboratory-developed test)-stained samples.

METHODS AND RESULTS

We designed a fully supervised deep learning algorithm for whole-slide PD-L1 assessment, consisting of four sequential convolutional neural networks (CNNs), using aiforia create software. We included 199 whole slide images (WSIs) of 'routine diagnostic' histology samples from stage IV NSCLC patients, and trained the algorithm by using a training set of 60 representative cases. We validated the algorithm by comparing the algorithm TPS with the reference score in a held-out validation set. The algorithm had similar concordance with the reference score (79%) as the pathologists had with one another (75%). The intraclass coefficient was 0.96 and Cohen's κ coefficient was 0.69 for the algorithm. Around the 1% and 50% cutoff points, concordance was also similar between pathologists and the algorithm.

CONCLUSIONS

We designed a new, deep learning-based PD-L1 TPS algorithm that is similarly able to assess PD-L1 expression in daily routine diagnostic cases as pathologists. Successful validation on routine diagnostic WSIs and detailed visual feedback show that this algorithm meets the requirements for functioning as a 'scoring assistant'.

Immunohistochemistry as predictive and prognostic markers for gastrointestinal malignancies

Biomarkers play a key role in the comprehensive pathologic evaluation of gastrointestinal malignancies. These biomarkers can be predictive, indicating whether a tumor is likely to respond to a particular therapy, or prognostic, providing information about the likely course and outcome of a disease. This review article will discuss available immunohistochemical stains for assessing these markers, including staining rationale, scoring criteria, associated systemic therapies, and pictorial examples. PD-L1, HER2, and mismatch repair status can be evaluated via immunohistochemistry for esophageal, gastric, and colorectal carcinomas. Biomarkers currently play a more limited role in evaluation of pancreatic and small bowel malignancies. Immunohistochemistry can also be used to evaluate biomarker status in gastrointestinal stromal tumors, gastrointestinal malignancies with NTRK gene fusions, and undifferentiated carcinomas with switch-sucrose non-fermentable complex abnormalities.

The Reproducibility of Histopathologic Assessments of Programmed Cell Death-Ligand 1 Using Companion Diagnostics in NSCLC

Introduction

Accurate results on the status of programmed cell death-ligand 1 (PD-L1) rely on not only the quality of immunohistochemistry testing but also the accuracy of the pathologic assessments. We explored the intraobserver and interobserver reproducibility of the interpretations for the companion diagnostics, the Dako PD-L1 22C3 pharmDx kit (Dako North America, Inc, Carpinteria, CA) and the VENTANA PD-L1 (SP263, Ventana Medical Systems, Inc, Tucson, AZ) assay, and the consistency between microscopic and digital interpretations of PD-L1.

Methods

A total of 150 surgical specimens diagnosed as NSCLC from December 2013 to July 2017 were included in this study. Twenty pathologists from different medical centers were enrolled to interpret the results of PD-L1 on the same day. A total of 100 sections were stained with the 22C3 clone and scored for the interobserver reproducibility, 20 cases of which were interpreted twice to assess the intraobserver reproducibility, and 50 cases of which were scanned into digital images to measure the consistency between microscopic and digital interpretations. A total of 44 sections were stained with the SP263 clone and scored for the interobserver reproducibility.

Results

For the intraobserver reproducibility of 22C3, the overall percent agreements were 92.0% and 89.0% for binary tumor evaluation at the cutoffs of 1% and 50%, respectively. The reliability among the pathologists revealed a substantial agreement for 22C3, whereas it revealed a substantial agreement at the cutoff of 1% and moderate agreement at the cutoffs of 25% and 50% for SP263. Microscopic and digital interpretations of PD-L1 revealed good consistency.

Conclusions

Intraobserver and interobserver reproducibility of the interpretations for PD-L1 was high using the 22C3 clone but lower for the SP263 clone. Corresponding training on such assessments, especially on the cases around the specific cutoffs, is essential for markedly improving such reproducibility. Digital imaging could improve the reproducibility of interpretation for PD-L1 among pathologists.

PD-L1 Is Preferentially Expressed in PIT-1 Positive Pituitary Neuroendocrine Tumours

Pituitary neuroendocrine tumours (PitNETs) cause lifelong morbidity, some requiring extensive surgical intervention, radiotherapy, or chemotherapy. A small percentage still cause debilitating disease, resistant to standard treatments, and may benefit from novel therapies. We assessed PD-L1 expression in a large cohort of PitNETs to investigate whether immunotherapy could represent a rational therapeutic choice. Unselected PitNETs undergoing surgical resection were reclassified according to the WHO 2017 system and underwent PD-L1 immunohistochemistry (clone SP263) in tissue microarray format. Membranous expression was scored as 0 (no expression), 1+ (< 50% expression) and 2+ (> 50% expression). A total of 265 PitNETs underwent PD-L1 immunohistochemistry. Prominent non-specific cytoplasmic staining was noted making assessment of true membrane expression difficult. Allowing for this, 40 of 264 (15%) PitNETs demonstrated strong staining (> 50% of neoplastic cells positive). These included 5/10 (50%) somatotrophs, 7/17 (41%) lactotrophs, 2/5 (40%) mammosomatotrophs, 4/8 (50%) mixed somatotroph-lactotrophs, 3/5 (60%) PIT-1 positive plurihormonal tumours with TSH expression, 10/28 (36%) of PIT-1 positive plurihormonal tumours, and 4/10 (40%) of PIT-1 positive tumours with no hormonal expression. Only 2/32 (6%) transcription factor triple negative, hormone negative tumours, 5/113 (4%) of gonadotrophs, and 0/6 thyrotrophs or 0/30 corticotrophs showed significant staining. We conclude that PD-L1 expression is common in somatotrophs, lactotrophs, and PIT-1 positive plurihormonal PitNETs but rare in transcription factor negative, hormone negative PitNETs, gonadotrophs, and corticotrophs. If the therapeutic role of immunotherapy is to be explored in PitNETs, it may be that it is of most benefit in the PD-L1 high subgroup.

SP142 PD-L1 Scoring Shows High Interobserver and Intraobserver Agreement in Triple-negative Breast Carcinoma But Overall Low Percentage Agreement With Other PD-L1 Clones SP263 and 22C3

SP142 programmed cell death ligand 1 (PD-L1) status predicts response to atezolizumab in triple-negative breast carcinoma (TNBC). Prevalence of VENTANA PD-L1 (SP142) Assay positivity, concordance with the VENTANA PD-L1 (SP263) Assay and Dako PD-L1 IHC 22C3 pharmDx assay, and association with clinicopathologic features were assessed in 447 TNBCs. SP142 PD-L1 intraobserver and interobserver agreement was investigated in a subset of 60 TNBCs, with scores enriched around the 1% cutoff. The effect of a 1-hour training video on pretraining and posttraining scores was ascertained. At a 1% cutoff, 34.2% of tumors were SP142 PD-L1 positive. SP142 PD-L1 positivity was significantly associated with tumor-infiltrating lymphocytes (P <0.01), and node negativity (P=0.02), but not with tumor grade (P=0.35), tumor size (P=0.58), or BRCA mutation (P=0.53). Overall percentage agreement (OPA) for intraobserver and interobserver agreement was 95.0% and 93.7%, respectively, among 5 pathologists trained in TNBC SP142 PD-L1 scoring. In 5 TNBC SP142 PD-L1-naive pathologists, significantly higher OPA to the reference score was achieved after video training (posttraining OPA 85.7%, pretraining OPA 81.5%, P<0.05). PD-L1 status at a 1% cutoff was assessed by SP142 and SP263 in 420 cases, and by SP142 and 22C3 in 423 cases, with OPA of 88.1% and 85.8%, respectively. The VENTANA PD-L1 (SP142) Assay is reproducible for classifying TNBC PD-L1 status by trained observers; however, it is not analytically equivalent to the VENTANA PD-L1 (SP263) Assay and Dako PD-L1 IHC 22C3 pharmDx assay.

PD-1/PD-L1 in Cancer: Pathophysiological, Diagnostic and Therapeutic Aspects

Immune evasion is a key strategy adopted by tumor cells to escape the immune system while promoting their survival and metastatic spreading. Indeed, several mechanisms have been developed by tumors to inhibit immune responses. PD-1 is a cell surface inhibitory receptor, which plays a major physiological role in the maintenance of peripheral tolerance. In pathological conditions, activation of the PD-1/PD-Ls signaling pathway may block immune cell activation, a mechanism exploited by tumor cells to evade the antitumor immune control. Targeting the PD-1/PD-L1 axis has represented a major breakthrough in cancer treatment. Indeed, the success of PD-1 blockade immunotherapies represents an unprecedented success in the treatment of different cancer types. To improve the therapeutic efficacy, a deeper understanding of the mechanisms regulating PD-1 expression and signaling in the tumor context is required. We provide an overview of the current knowledge of PD-1 expression on both tumor-infiltrating T and NK cells, summarizing the recent evidence on the stimuli regulating its expression. We also highlight perspectives and limitations of the role of PD-L1 expression as a predictive marker, discuss well-established and novel potential approaches to improve patient selection and clinical outcome and summarize current indications for anti-PD1/PD-L1 immunotherapy.

PD-L1 in Cytological Samples: A Review and a Practical Approach

With a growing number of predictive biomarkers needed to manage patients with non-small cell lung cancer (NSCLC), there has been a paradigm shift in care and handling of diagnostic samples. Among the various testing methods, immunohistochemistry (IHC) is the most cost- effective and widely available. Furthermore, over the past decade immunotherapy has emerged as one of the most promising cancer treatments. In this scenario IHC is the most used testing method available for PDL-1/PD1 immunotherapy. Several monoclonal antibodies targeting programmed death 1 (PD-1)/programmed death ligand-1 (PD-L1) pathways have been integrated into standard-of-care treatments of a wide range of cancer types, once provided evidence of PD-L1 expression in tumor cells by immunohistochemistry (IHC). Since currently available PD-L1 assays have been developed on formalin-fixed paraffin embedded (FFPE) histological specimens, a growing body of research is being dedicated to confirm the feasibility of applying PDL-1 assays also to cytological samples. Albeit promising results have been reported, several important issues still need to be addressed. Among these are the type of cytological samples, pre-analytical issues, cyto-histological correlation, and inter-observer agreement. This review briefly summarizes the knowledge of the role of cytopathology in the analysis of PD-L1 by immunocytochemistry (ICC) and future directions of cytopathology in the immunotherapy setting.

PD-L1 and beyond: Immuno-oncology in cytopathology

Over the past decade, immunotherapy has emerged as one of the most promising cancer treatments. Several monoclonal antibodies targeting the programmed death 1 (PD-1)/ programmed death ligand-1 (PD-L1) pathway have been integrated into standard-of-care treatments for a wide range of cancer types. Although all the available PD-L1 immunohistochemistry (IHC) assays have been developed on formalin-fixed histological specimens, a growing body of research has recently suggested the feasibility of PD-L1 testing on cytological samples. Although promising results have been reported, several important issues still need to be addressed. Among these are pre-analytical issues, cyto-hystological correlation, and inter-observer agreement. This review will briefly summarise the knowledge gaps and future directions of cytopathology in the immuno-oncology scenario.

Programmed cell death-ligand 1 assessment in urothelial carcinoma: prospect and limitation

Programmed cell death protein 1/programmed death-ligand 1 (PD-1/PD-L1) inhibition has revolutionized the treatment paradigm of urothelial carcinoma (UC). Several PD-L1 assays are conducted to formulate appropriate treatment decisions for PD-1/PD-L1 target therapy in UC. However, each assay has its own specific requirement of antibody clones, staining platforms, scoring algorithms, and cutoffs for the determination of PD-L1 status. These prove to be challenging constraints to pathology laboratories and pathologists. Thus, the present article comprehensively demonstrates the scoring algorithm used and differences observed in each assay (22C3, SP142, and SP263). Interestingly, the SP142 score algorithm considers only immune cells and not tumor cells (TCs). It remains controversial whether SP142 expressed only in TCs truly accounts for a negative PD-L1 case. Moreover, the scoring algorithm of each assay is complex and divergent, which can result in inter-observer heterogeneity. In this regard, the development of artificial intelligence for providing assistance to pathologists in obtaining more accurate and objective results has been actively researched. To facilitate efficiency of PD-L1 testing, several previous studies attempted to integrate and harmonize each assay in UC. The performance comparison of the various PD-L1 assays demonstrated in previous studies was encouraging, the exceptional concordance rate reported between 22C3 and SP263. Although these two assays may be used interchangeably, a clinically validated algorithm for each agent must be applied.

The International Association for the Study of Lung Cancer Global Survey on Programmed Death-Ligand 1 Testing for NSCLC

INTRODUCTION

Programmed death-ligand 1 (PD-L1) immunohistochemistry (IHC) is required to determine the eligibility for pembrolizumab monotherapy in advanced NSCLC worldwide and for several other indications depending on the country. Four assays have been approved/ Communauté Européene-In vitro Diagnostic (CV-IVD)-marked, but PD-L1 IHC seems diversely implemented across regions and laboratories with the application of laboratory-developed tests (LDTs).

METHOD

To assess the practice of PD-L1 IHC and identify issues and disparities, the International Association for the Study of Lung Cancer Pathology Committee conducted a global survey for pathologists from January to May 2019, comprising multiple questions on preanalytical, analytical, and postanalytical conditions.

RESULT

A total of 344 pathologists from 64 countries participated with 41% from Europe, 24% from North America, and 18% from Asia. Besides biopsies and resections, cellblocks were used by 75% of the participants and smears by 11%. The clone 22C3 was most often used (69%) followed by SP263 (51%). They were applied as an LDT by 40% and 30% of the users, respectively, and 76% of the participants developed at least one LDT. Half of the participants reported a turnaround time of less than or equal to 2 days, whereas 13% reported that of greater than or equal to 5 days. In addition, quality assurance (QA), formal training for scoring, and standardized reporting were not implemented by 18%, 16%, and 14% of the participants, respectively.

CONCLUSIONS

Heterogeneity in PD-L1 testing is marked across regions and laboratories in terms of antibody clones, IHC assays, samples, turnaround times, and QA measures. The lack of QA, formal training, and standardized reporting stated by a considerable minority identifies a need for additional QA measures and training opportunities.

Validation of multiplex immunofluorescence and digital image analysis for programmed death-ligand 1 expression and immune cell assessment in non-small cell lung cancer: comparison with conventional immunohistochemistry

AIMS

This study aimed to validate the application of combined multiplex immunofluorescence (mIF) and digital image analysis (DIA) in formalin-fixed and paraffin-embedded tissues for the quantitative assessment of programmed death-ligand 1(PD-L1) and immune cells (ICs) in non-small cell lung cancer (NSCLC).

METHODS

Fifty resected samples of NSCLC were sequentially stained with a DNA-tagged mIF (panel including PD-L1, CKpan, CD8, CD68 and 4',6-diamidino-2-phenylindole (DAPI)) and conventional immunohistochemistry (cIHC). The assessment of cell density and consistency of tumour proportion score (TPS) via DIA were compared with those by pathologists.

RESULTS

A strong correlation in the cell population of immune markers was obtained between mIF and cIHC (for PD-L1: R=0.9304, CKpan: R=0.8231, CD8: R=0.9314 and CD68: R=0.8366) within 95% limits of agreement. The continuous TPS calculated using mIF was highly consistent with the IHC staining results which were evaluated by pathologists (R=0.9362). However, in the comparison of TPS using interval variables, a poor agreement was obtained at a cut-off of 1% (κ=0.197), whereas excellent agreement was achieved at cut-offs of 50% (κ=0.908) and 5% (κ=0.823). DIA on mIF showed that PD-L1 commonly colocalised with CD68⁺ macrophages and CD8⁺ cytotoxic cells were closer to PD-L1^-/CK⁺ tumour cells (TCs) than to PD-L1⁺/CK⁺ TCs in spatial distribution.

CONCLUSIONS

A combination of mIF and DIA is useful for the quantification of PD-L1 expression and IC populations in NSCLC. Further validation of TPS at a cut-off of 1% and assay harmonisation is essential for translating this method in a diagnostic setting.

Automated PD-L1 Scoring for Non-Small Cell Lung Carcinoma Using Open-Source Software

PD-L1 expression in non-small cell lung cancer (NSCLC) is predictive of response to immunotherapy, but scoring of PD-L1 immunohistochemistry shows considerable interobserver variability. Automated methods may allow more consistent and expedient PD-L1 scoring. We aimed to assess the technical concordance of PD-L1 scores produced using free open source QuPath software with the manual scores of three pathologists. A classifier for PD-L1 scoring was trained using 30 NSCLC image patches. A separate test set of 207 image patches from 69 NSCLC resection cases was used for comparison of automated and manual scores. Automated and average manual scores showed excellent correlation (concordance correlation coeffecient = 0.925), though automated scoring resulted in significantly more 1–49% scores than manual scoring (p = 0.012). At both 1% and 50% thresholds, automated scores showed a level of concordance with our ‘gold standard’ (the average of three pathologists’ manual scores) similar to that of individual pathologists. Automated scoring showed high sensitivity (95%) but lower specificity (84%) at a 1% threshold, and excellent specificity (100%) but lower sensitivity (71%) at a 50% threshold. We conclude that our automated PD-L1 scoring system for NSCLC has an accuracy similar to that of individual pathologists. The detailed protocol we provide for free open source scoring software and our discussion of the limitations of this technology may facilitate more effective integration of automated scoring into clinical workflows.

PD-L1 as a biomarker of response to immune-checkpoint inhibitors

Immune-checkpoint inhibitors targeting PD-1 or PD-L1 have already substantially improved the outcomes of patients with many types of cancer, although only 20-40% of patients derive benefit from these new therapies. PD-L1, quantified using immunohistochemistry assays, is currently the most widely validated, used and accepted biomarker to guide the selection of patients to receive anti-PD-1 or anti-PD-L1 antibodies. However, many challenges remain in the clinical use of these assays, including the necessity of using different companion diagnostic assays for specific agents, high levels of inter-assay variability in terms of both performance and cut-off points, and a lack of prospective comparisons of how PD-L1⁺ disease diagnosed using each assay relates to clinical outcomes. In this Review, we describe the current role of PD-L1 immunohistochemistry assays used to inform the selection of patients to receive anti-PD-1 or anti-PD-L1 antibodies, we discuss the various technical and clinical challenges associated with these assays, including regulatory issues, and we provide some perspective on how to optimize PD-L1 as a selection biomarker for the future treatment of patients with solid tumours.