Delta-like Protein 3 Prevalence in Small Cell Lung Cancer and DLL3 (SP347) Assay Characteristics

A validation cohort of 1503 cases of formalin-fixed (10% neutral buffered formalin), paraffin-embedded SCLC whole-section samples was procured from an internal tissue bank at Ventana Medical Systems, Inc, and the following commercial tissue banks: Asterand Bioscience; Avaden BioSciences; Boca Biolistics; Conversant Bio; GLAS; ILSbio; Impath; PrecisionMed, Inc; TriMetis Life Sciences; and US Biomax, Inc. The cohort included primary and metastatic SCLC samples and consisted of resections, core needle biopsies, and fine-needle aspirates. Review of hematoxylin-eosin–stained slides from cases in the validation cohort confirmed morphologic features consistent with small cell carcinoma. These include small to medium-sized cells with minimal cytoplasm, indistinct nucleoli, and nuclear molding.

The discovery cohort used for immunoreactivity assessments included the following: (1) Cut slides (4 μm) from 2 tissue microarrays (TMAs) of normal tissue: one array (FD808i-1) containing 24 types of normal human tissue representing 72 individual cases, and another array (999l, Food and Drug Administration) containing 32 types of normal human tissue representing 78 individual cases procured from US Biomax; (2) Cut slides (4 μm) also procured from a tumor tissue array (808i-2, Food and Drug Administration) containing 54 cases of carcinoma from multiple locations plus 18 cases with 6 types of normal tissue adjacent to tumor (single cores per case) representing 72 individual cases, also from US Biomax; (3) Cut slides (4 μm) from an array (LUC1504) of SCLC tissues containing 150 cores including 5 normal/benign lung tissue and 70 cases of SCLC tissue procured from Pantomics.

Four-micrometer-thick tissue sections were cut from the cohort of cases and mounted on positively charged glass slides. The slides from both validation and discovery cohorts were stained with VENTANA DLL3 (SP347) Assay (P/N 790-7016, Ventana) along with OptiView DAB IHC Detection Kit (P/N 760-700, Ventana) on the BenchMark ULTRA automated staining platform using the recommended staining conditions of the DLL3 (SP347) assay package insert. Rabbit monoclonal negative control Ig (P/N 790-4795) was used as the negative reagent control. The samples were counterstained with hematoxylin II (P/N 790-2208, Ventana) and bluing reagent (P/N 760-2037, Ventana). Recommended tissue controls were used based on the DLL3 (SP347) assay package insert. For each case in the cohort and for the TMAs, one slide was stained with hematoxylin-eosin using either the VENTANA SYMPHONY system or Sakura Tissue-Tek Prisma & Film Automated Slide Stainer & Coverslipper according to the manufacturer's instruction.

Negative reagent control and DLL3 (SP347) assay–stained slides were evaluated for nonspecific background staining and cross-reactivity on a scale of 0 to 3 in increments of 0.5. Background staining was considered staining of any types of cells focally/diffusely in the negative reagent control and stroma/nonneoplastic cells focally/diffusely in the DLL3 (SP347) assay. Background staining greater than 0.5 was considered obtrusive to interpretation of specific staining, and those samples were excluded from analysis. The percentage tumor cell staining was assessed in the DLL3 (SP347)–stained slides. In addition, overall intensity of tumor cell staining on a scale of 0 to 3 in increments of 1 was also assessed in the DLL3 (SP347)–stained slides. Strong cytoplasmic and/or membranous staining of tumor cells was scored as 3, moderate staining as 2, and weak staining as 1. Absence of staining was given a score of 0. Reactivity was defined as samples that had more than 0% DLL3 staining. In addition, the SCLC samples were assessed to identify different staining characteristics.

Nonspecific background staining was captured and scored for all tested tissue arrays using a scale of 0 to 3 in increments of 0.5. Background staining greater than 0.5 was considered obtrusive to interpretation of specific staining, and those samples were deemed unacceptable. Reactive cores were defined as those tissue cores that stained positive (DLL3 stain intensity or percentage DLL3 cell staining >0).

Neoplastic tissue arrays were scored for specific percentage tumor cell staining. In addition, the overall intensity of tumor cell staining was captured on a scale of 0 to 3 in increments of 1 (intensity scores were given in the same fashion as the scores given for the prevalence study as described above).

Normal, nonneoplastic tissue arrays were scored for DLL3 stain intensity captured on a scale of 0 to 3 in increments of 1. Strong cytoplasmic and/or membranous staining of tissue cores was scored as 3, moderate staining as 2, and weak staining as 1. Absence of staining was given a score of 0.

In the validation cohort of 1503 SCLC cases, 141 samples (9.4%) were deemed not evaluable or not acceptable (eg, no tissue/tumor present, crush, necrosis, or high background), and these samples were excluded from the analysis of DLL3 staining and reactivity. Of the 1362 evaluable samples, reactivity was observed in 1040 (76.4%) (Table 1). In addition, the samples exhibited a wide range of staining intensities (Figure 1) and percentage tumor cell staining (bimodal distribution of percentage staining, with 80% of the cases either falling below 25% or above 75% tumor cell staining) (Figure 2).

Upon examination of the SCLC specimens, the following staining characteristics were identified: partial and complete membrane/cytoplasmic staining, granular staining pattern, and homogeneous or heterogeneous staining pattern (Figure 3, A through L).

In the discovery TMAs, 29 samples were deemed not evaluable (eg, no tissue/tumor present, artifacts, or edge artifacts), and these cases were not used in the analysis of reactivity. Samples exhibited a range of staining intensity from weak or moderate to strong in both normal and neoplastic tissues and a range of percentage tumor cell staining in neoplastic tissues.

Reactivity was observed in 24 of 156 evaluable cores (15.8%) in the nonneoplastic TMAs (Table 2). The reactive nonneoplastic cores were composed primarily of normal tissues with a neural or neuroendocrine component. Reactivity was observed in 16 of 65 evaluable cores (24.6%) (1 core was normal tissue adjacent to tumor) in the carcinoma TMA (Table 3) and 80 of 143 evaluable cores (55.9%) in the SCLC TMA.

This is the first study in the literature, to our knowledge, to use the VENTANA DLL3 (SP347) Assay to examine the prevalence of DLL3 protein expression within a cohort of SCLC. In the validation cohort of cases, a high percentage (76.4%) had reactivity to our antibody, exemplified by observable staining with the OptiView detection kit. In addition, the validation cohort of cases showed a bimodal distribution of percentage of cells staining and a wide range of staining intensities.

Three studies of DLL3 expression–level data^8,10,11  showed similar prevalence results. However, all 3 of these studies used a mouse monoclonal antibody, in contrast to the SP347 rabbit monoclonal antibody, and their analysis of positivity had different definitions. In the first study describing DLL3 prevalence in SCLC, the authors used an H-score of 100 or higher as a positive cutoff and found that 120 (72%) of 167 treatment-naïve and 17 (85%) of 20 recurrent and treatment-refractory SCLC cases were positive for DLL3 expression.⁸  In the second and third studies, positivity was defined as 1% or more of tumor cells staining. The second study showed that of 48 patients with SCLC or large cell neuroendocrine carcinoma, 42 (88%) had positive staining.¹⁰  Lastly, in a Japanese cohort of 63 patients with SCLC, 52 (83%) were deemed positive for DLL3 expression.¹¹  In summary, DLL3 (SP384) prevalence data are consistent with the data in the literature using other antibodies, showing that there are high rates of DLL3 protein expression in SCLC.

Additionally, reactivity was observed in a small percentage of nonneoplastic and neoplastic tissues, and was significantly higher within discovery SCLC TMAs in our study. This is consistent with a study by Saunders et al⁸  that showed low prevalence of DLL3 in normal, nonneoplastic tissues. Also, this study demonstrated that the reactive tissues were primarily of neural or neuroendocrine origin, which corresponds with the known characteristics of DLL3 in mRNA studies.^8,12 

Herein, we present data that show the prevalence of DLL3 expression in a cohort of 1503 cases and also show that the VENTANA DLL3 (SP347) Assay is reactive with recognizable staining patterns that can be used for determining the level of DLL3 expression in SCLC.