Assessment of Programmed Death–ligand 1 (PD-L1) Immunohistochemical Expression on Cytology Specimens in Non–Small Cell Lung Carcinoma

A Comparative Study With Paired Surgical Specimens

Andrea Hernandez, DO; Tamar C. Brandler, MD; Fang Zhou, MD; Andre L. Moreira, MD, PhD; Nina Schatz-Siemers, DO; Aylin Simsir, MD


Am J Clin Pathol. 2019;151(4):403-415. 

In This Article


With increased focus on targetable tumor biomarkers such as PD-L1, treating clinicians are increasingly aiming to gain more information from smaller material obtained by minimally invasive techniques such as cytology sampling, which are often the sole available specimens in patients with advanced lung cancer.[7] Our study adds to the limited available literature comparing PD-L1 IHC testing in cytology and surgical pathology material Table 6. A recent study of 41 paired NSCLC cases evaluated PD-L1 IHC expression in cytology smears and cell blocks in comparison to surgical pathology biopsy specimens.[2] After the exclusion of cytology samples displaying low tumor cellularity, the authors showed a high concordance between paired specimens, particularly using smear preparations (36/37 smears and 31/38 cell blocks showed concordant expression). A study of 86 paired lung cancer cases showed a high correlation coefficient (R2 = 0.87- 0.89) and overall agreement of 85% and 94% when the cutoff for positivity was 1% and 50% PD-L1 expression, respectively.[13] Other investigators have shown indirect evidence of similar PD-L1 expression between cytology and surgical pathology specimens by bulk case comparison.[14,15]

Our study compared PD-L1 expression in matched NSCLC samples using the Dako 22C3 pharmDx antibody, the US Food and Drug Administration–approved companion diagnostic assay for treatment with pembrolizumab. Tumors displaying 50% or more PD-L1 expression and no EFGR or ALK aberrations qualify the patient for first-line therapy with this agent, while tumors with 1% to 49% PD-L1 expression and disease progression on platinum-based chemotherapy qualify the patient for second-line therapy.[16] Paired agreement in this study was evaluated by following clinically relevant PD-L1 TPS tiers: negative (<1%), low expression (≥1%-49%), and high expression (≥50%).

Agreement was achieved in 67% (35/52) of matched cases (moderate agreement; κ = 0.51; 95% CI, 0.42–0.60) when comparing PD-L1 testing in cytology with surgical pathology samples. Most discordant cases (11/17) were negative on cytology yet had low expression (7/11) or high expression (4/11) on the corresponding surgical specimen. Variability in the level of positive expression between paired samples was also a key reason for discordant results (5/17). This variability may pose a therapeutic dilemma in terms of whether the patient qualifies for first- or second-line pembrolizumab treatment.

Many factors play a role in the successful adaptation of IHC biomarkers originally validated on histologic specimens to cytologic preparations.[2] These include antibody clone used, platform selected, cell fixative used, cytology preparation (cell block vs smears), staining of background cells, and the subjective nature of the IHC evaluation.[17,18] With regard to assessment of PD-L1 IHC, the cellularity of the sample, heterogeneous expression of PD-L1 within the same tumor, and type of pathology specimen (surgical/cytology vs biopsy/excision) are additional variables that should be considered.[19,20]

In our study, eight (47%) of the discordant cases had cytology samples with fewer than 100 tumor cells, with only slight agreement with the corresponding surgical specimens (κ = 0.19; 95% CI, 0.04–0.35). Cases with 100 or more tumor cells had higher agreement (κ = 0.63; 95% CI, 0.53–0.73). While low tumor cell count is less than optimal and may lead to a false-negative PD-L1 expression, our study showed that cases with a tumor cell count less than 100 may display positive PD-L1 expression and therefore may not be the sole limiting factor. Skov and Skov[13] previously reported no statistical changes in paired results after exclusion of cytology specimens with fewer than 100 tumor cells.

Intratumoral heterogeneity has been reported to have an impact on the detection of targetable biomarkers,[21] with studies showing marked variability in PD-L1 staining within a single tumor.[19,20] Five study cases showed varied positive (low or high) expression levels between surgical and cytology specimens. This variation is likely attributable to tumor heterogeneity. One of the cases in our study Image 4 displayed high positive PD-L1 expression on both surgical pathology and cytology and low expression on a second cytology specimen. Upon review of the PD-L1 stained surgical pathology slide from this case, heterogeneous PD-L1 expression within the tumor was indeed noted.

Image 4.

Intratumoral heterogeneity in an excisional specimen of a patient diagnosed with lung adenocarcinoma is reflected in two pleural fluid specimens (sample 27). A, Cell block prepared from pleural fluid 1 (H&E, ×40). B, Programmed death–ligand 1 (PD-L1)–stained cell block section with 1% to 49% expression (×40). C, Cell block prepared from pleural fluid 2 (sample 44) (H&E, ×40). D, PD-L1–stained cell block section with tumor cells showing 50% or more expression. E, Excisional lymph node specimen (H&E, ×20). F, PD-L1–stained concurrent surgical excision showing heterogeneous PD-L1 expression among tumor cells with reported PD-L1 of 50% or more.

Tumor heterogeneity has also been investigated as a possible reason for discordant PD-L1 expression between small surgical pathology biopsy specimens and excisional samples.[22,23] Similarly, agreement was found to be higher in our study when comparison was made between cytology and paired surgical pathology excisional samples (κ = 0.78; 95% CI, 0.63–0.92) vs cytology with paired smaller surgical biopsy specimens (κ = 0.41; 95% CI, 0.31–0.52). Furthermore, as previously mentioned, most discordant cases in our study underrepresent PD-L1 surgical pathology scores. This trend was further explored to show that in paired cases with negative cytology PD-L1 staining and positive surgical pathology PD-L1 staining, more than half (6/11) of the surgical cases had PD-L1 levels of 10% or less.

Timing of sample acquisition was also examined as a possible source of disagreement between paired samples, as prior studies have reported altered, either diminished or inducible, PD-L1 expression levels in chemotherapy- and/or radiotherapy-treated tumors.[24,25] Investigation of paired samples showed that four (23.5%) of 17 discordant study cases were not acquired concurrently, with the four patients receiving treatment prior to the acquisition of the cytology sample. When evaluating concordant study cases for temporal lag in cytology acquisition, eight (23%) of 35 patients received treatment, with the samples acquired subsequent to the surgical pathology/treatment-naive material. Discordance in our study was likely not attributable to treatment prior to PD-L1 testing on cytology material, with moderate agreement reached in both patients who were previously treated (κ = 0.50; 95% CI, 0.31 to 0.69) and those who were not (κ = 0.51; 95% CI, 0.41 to 0.61).

PD-L1 staining in effusion specimens may pose a diagnostic challenge due to PD-L1 nonspecific expression in macrophages, B cells, natural killer cells, and dendritic cells.[17] This difficulty is augmented when tumor cells appear as discohesive cells among macrophages Image 5. In addition, to date, PD-L1 IHC assays have not been validated for decalcified tissue,[12] and staining after decalcification should be avoided. One surgical pathology case in our study group consisted of metastatic adenocarcinoma to the rib. In adherence with the Dako manual,[12] the rib specimen included in our study did not undergo decalcification prior to PD-L1 testing.

Image 5.

Macrophages as a pitfall for programmed death–ligand 1 (PD-L1) interpretation in cytologic specimens. A, Cell block prepared from pleural fluid showing a tumor cell (center) among many scattered macrophages (H&E, ×40). B, PD-L1–stained cell block section revealed additional tumor cells on deeper sectioning showing strong membranous expression for PD-L1. Background macrophages also show membranous staining for PD-L1 at varying intensities (×40).

Last, of the different cytology samples in our cohort, those procured directly from the lung (either by CT-guided or EBUS-guided FNA) showed substantial agreement (κ = 0.74; 95% CI, 0.61–0.88) with PD-L1 expression in paired surgical specimens. Cytology samples obtained from lymph nodes and pleural effusions showed only fair agreement (κ = 0.34; 95% CI, 0.16–0.52 and κ = 0.39; 95% CI, 0.25–0.53, respectively). These results provide evidence that cytology material obtained directly from the lung may serve as the best sample for PD-L1 biomarker testing.

Limitations of our study include the relatively small sample size, which should be followed up with larger paired PD-L1 expression studies, limited number of squamous cell carcinoma cases for PD-L1 analysis, and the use of only one type of PD-L1 assay.

In conclusion, our study confirms that cytology cell blocks can be used for PD-L1 IHC evaluation. Discordance with surgical specimens may occur. Most of our discordant cases were negative on cytology with positive expression on the corresponding surgical specimen. Positive expression on cytology appears to be reliable compared with paired surgical specimens as illustrated in our study, whereas a patient with a negative PD-L1 cytology result may benefit from PD-L1 testing on repeat cytology or surgical material. Differences in scoring of surgical pathology and cytology samples are likely attributed to intratumoral heterogeneity and cell block cellularity. While cases with a tumor cell count less than 100 may lead to missing some PD-L1 positive cases, cases with fewer than 100 tumor cells that do show positive PD-L1 staining have proven to be reliable. Surgical pathology excisional samples showed greater agreement with cytology material than was observed with surgical pathology biopsy specimens, reinforcing prior reports that smaller diagnostic material (surgical pathology or cytology) may lead to discordant PD-L1 expression levels compared with larger excisional tissue. PD-L1 IHC on cytology cell blocks should be considered in the evaluation of PD-L1 biomarker testing as a reliable method. Further studies with larger cohorts may shed light on how to best use these specimens for PD-L1 analysis to broaden the accessibility of testing for patients with NSCLC.