Expression of CD70 (CD27L) Is Associated With Epithelioid and Sarcomatous Features in IDH-Wild-Type Glioblastoma

Case Selection and Tumor Classification

Archival tissues were procured from the National Institutes of Health/National Cancer Institute (NIH/NCI) in Bethesda, Maryland (2000–2016). All patients were deceased at the time of evaluation and were exempt from Institutional Review Board review. At the time of the original diagnosis, cases were classified according to the 2007 WHO Classification of Central Nervous System Tumors (23). Using an integrated diagnostic approach that incorporates histology, immunohistochemistry, and molecular data (where available), cases were re-reviewed and assigned to the newest diagnostic categories outlined in the 2016 WHO classification system (24) (see below for interpretation of immunohistochemistry). Fluorescence in situ hybridization (FISH) for 1p/19q co-deletion was retrospectively included as part of the original case work-up in our lab.

The IDH-wildtype variants gliosarcoma and epithelioid glioblastoma (E-GBM) were diagnosed based on previously reported histologic features (25–7). Briefly, cases of gliosarcoma contained a biphasic population of neoplastic cells consisting of a malignant glial component (GFAP+) and a sarcomatous component composed of overtly neoplastic cells enmeshed in a dense pericellular reticulin meshwork; these 2 areas were often, but not always, regionally distinct. E-GBM contained a monotonous population of round, occasionally discohesive cells with eosinophilic cytoplasm, eccentric nuclei, and prominent nucleoli. No eosinophilic granular bodies or Rosenthal fibers were noted. Focally, areas of spindled cells were occasionally present in an otherwise predominant population of epithelioid cells.

Tissue Microarrays (TMA)

Hematoxylin and eosin (H&E) sections from formalin-fixed paraffin-embedded (FFPE) tissue blocks were evaluated for adequate tumor content. Due to the marked morphologic, immunophenotypic, and molecular heterogeneity in diffuse gliomas, 4-mm tissue cores were used for construction. Tissue cores were extracted from donor paraffin blocks and manually aligned in a melted paraffin bath using a pre-defined grid. Each block contained 28 cores and a 2-mm control tissue marker (liver, kidney, or lymph node) that also served as a spatial reference. H&E-stained sections were then reviewed to confirm representative tumor sampling and accurate core placement.

Immunohistochemistry and Staining Evaluation

Immunohistochemical staining was performed on 4-µm-thick tissue sections with a Ventana BenchMark Ultra (IDH1 R132H, ATRX, p53, H3K27M, EGFR, INI1, GFAP, CD68, CD163, HLA-DR, Ki-67; Ventana Medical Systems, Tucson, AZ), Ventana XT (CD27), or Leica Bond (CD70; Leica Microsystems, Bannockburn, IL) automated immunostainer using manufacturer recommended antigen retrieval. Antibody names, clones, dilutions, source, and expected cellular localization are listed in Table 1. For CD70 staining, a range of dilutions (1:500–1:2000) was tested, with a final dilution of 1:500 used for evaluation. Cases were deemed as having true expression when they exhibited crisp membranous staining or unequivocal strong cytoplasmic staining in tumor cells that was stronger than adjacent non-neoplastic tissue (neuropil, RBCs, vessel media, and adventitia). While a subset of activated lymphocytes express CD70 under normal conditions, absence of staining in adjacent immune cells provided an internal control.

Immunoreactivity for CD70 was assessed using both quantitative and semi-quantitative methods. For semi-quantitative analysis, expression in tumor cells was scored as follows: 0 (negative); 1 (<5% of tumor cells); 2 (5–50% of tumor cells); or 3 (>50% of tumor cells). The cut-off used here (5%) is based on the commonly reported staining threshold used for PD-L1 expression, another co-stimulatory molecule (28). Tumor cells showing weak granular cytoplasmic staining (and a corresponding absence of moderate or strong membranous staining) in any number of tumor cells were given a score of 1. Quantitative methodology is described below. Full tissue sections from positive TMA cases (n = 18) were subsequently stained with anti-CD70 to assess the distribution and heterogeneity of staining. Evaluation of morphologic and immunohistochemical features allowed separation of expression on tumor cells rather than macrophages/microglia to interrogate the association of CD70-positivity and TAM.

IDH mutation status was determined primarily by immunohistochemistry. For designation as IDH-wild-type, a cutoff of 10% was used, where the probability of a variant IDH1 or IDH2 mutation was calculated using a predictive model for IDH mutations in grade II–IV astrocytomas and oligodendrogliomas (29). Cases not meeting the 10% cutoff were designated as not otherwise specified. All gliosarcoma and epithelioid glioblastoma cases were designated as wildtype for IDH; these histologic variants did not show reactivity with anti-IDH R132H. All other surrogate markers were interpreted based on their reported sensitivity and specificity (see Ref. [30] for review). For p53, staining was scored as positive (predictive of p53 mutation) when ≥10% of tumor cells demonstrated moderate or strong nuclear immunoreactivity (31). EGFR was scored as positive when tumor cells showed moderate to strong (2–3+) membranous staining; EGFR immunohistochemistry was mainly used in the context of negative staining (i.e. EGFR amplification is virtually never seen with negative EGFR immunohistochemistry, and has a reported negative predictive value of 0.99–1) (32).

Digital Imaging and Automated Analysis

TMA slides and select case slides from the original paraffin blocks were digitally scanned and converted to whole slide images using the Aperio XT system (Leica Biosystems, Buffalo Grove, IL) at 40× magnification and viewed with Aperio Imagescope software (version 11.2). Tumor areas on TMAs and fresh-cut full sections were digitally annotated with regions of interest for analysis. Staining was quantified using the positive pixel count v9 (ppc; CD70, CD27, and CD163) and nuclear v9 (p53) algorithms and reported as the percentage of positive staining of the tumor area analyzed (irrespective of cellular localization for ppc). Bright-field images were extracted from Imagescope and captured using the Nuance multispectral imaging system (PerkinElmer, Waltham, MA).

TCGA Datasets

Expression levels and clinical data from TCGA (http://cancergenome.nih.gov) and Vital (33) cohort datasets were obtained through the GlioVis data portal for visualization and analysis of brain tumor expression datasets (http://gliovis.bioinfo.cnio.es, accessed May 29, 2017) (34). Using RNAseq and the Affymetrix U133A Array (HG-U133A) data, CD70 mRNA expression levels were acquired from 447 TCGA GBM samples and correlated with overall survival (Affymetrix) and gene expression-based subtype (ie neural, proneural, mesenchymal, and classical) (RNAseq). For generation of survival curves in the recurrent, non-G-CIMP dataset, the median mRNA cutoff (5.32285) was used for defining high and low expression of CD70. Mutation frequency from the TCGA was obtained from cBioPortal (http://www.cbioportal.org, accessed May 29, 2017) (35, 36).

Statistical Analysis

Differences among groups were determined using a Kruskal–Wallis test. Individual 2-group tests were performed using a Wilcoxon rank sum test. In the cases in which several 2-group comparisons were made after using a Kruskal–Wallis test to overall the overall difference among groups, the Hochberg method was used to adjust the p values for multiple comparisons (37). All p values are 2-tailed. Survival analysis was performed using the Kaplan-Meier estimation and log-rank test. For TMA data, survival was calculated (in months) from the date of tumor detection by imaging (overall survival) and the time (months) from the last date of radiotherapy (post-radiation survival) until the date of death. No censored events were included in the TMA analyses; this limited bias is being reported but is not expected to materially impact the interpretation of the findings due to the relatively short survival of patients with glioblastoma following treatment. Analysis of tissue microarray and TCGA data was performed using SAS version 9.2 (SAS Institute, Inc., Cary, NC) and illustrated in Microsoft Excel. A p value <0.05 was considered statistically significant.

Clinicopathologic Features

A total of 214 patients, including 12 paired patient samples (primary and recurrent tissue from the same patient), were included in the initial study cohort. Of these, 9 cases (7 diffuse midline gliomas and 2 IDH-wild-type glioblastomas) showed diminished immunoreactivity with a number of the antibodies tested. In our experience, staining for ATRX (HPA001906) routinely shows decreased immunoreactivity in the presence of tissue artifacts (e.g. prolonged fixation). As a consequence, cases showing loss of normal internal positive control (vessel endothelial cells, non-neoplastic cells such as neurons and glia, etc.) with anti-ATRX were excluded from further analysis. A total of 205 tumor tissues were suitable for evaluation. Patient demographics and clinical characteristics are outlined in Table 2. Classification of tumors according to the 2016 WHO Classification of Central Nervous System Tumors are shown in Table 3. The majority of cases (90.1%) were assigned to the following 2016 WHO diagnostic categories: anaplastic astrocytoma, IDH mutant (n = 23); glioblastoma, IDH mutant (n = 15); glioblastoma, IDH-wild-type (n = 96); glioblastoma, not otherwise specified (n = 29); diffuse midline glioma, H3 K27M mutant (n = 16). A subset of evaluable diffuse midline gliomas were negative for H3K27M by immunohistochemistry (n = 5).

A spectrum of staining for CD70 was observed, ranging from focal, weak and cytoplasmic to strong and diffuse cell surface (membranous) expression. Eighteen cases (8.8% of high-grade gliomas) exhibited moderate-to-strong staining (present in >5% of tumor cells and scored as 2 or 3). As a group, these high-expressing cases showed similar histologic and clinical features (see below) and were grouped together for subsequent analysis and illustration. Forty-eight cases (23.4% of all tumors) showed rare weak positive or diffuse faint cytoplasmic staining in tumor cells (scored as 1). Quantitative evaluation of CD70 staining in high-expressing tumors (score 2 to 3) showed a higher median expression of 20.2% (range 4%–68%) compared to a median expression of 0.07% (range 0.0004 to –7.3%) in weak (score 1) or negative (score 0) cases (p < 0.01). Cases with moderate-to-strong (score 2 to 3) CD70 staining exhibited the following characteristics: high-grade histology (WHO grades III and IV, 18/18), history of radiotherapy (13/14), IDH R132H wild-type (17/18), and retained expression of ATRX (14/14) (see Table 4 for complete tumor immunophenotypes).

CD70 Is Expressed in the IDH-Wild-type Variants Gliosarcoma and Epithelioid Glioblastoma

A predominant histologic pattern emerged in cases with moderate-to-strong expression, characterized by either focal or diffuse epithelioid cellular morphology (n = 4/4 E-GBM tested; Fig. 1) or sarcomatous transformation (n = 10/16 gliosarcomas tested; Fig. 2). E-GBM is a newly described, but rare, variant of IDH-wildtype glioblastoma with distinct morphologic and molecular features. In our cohort, 4 tumors met histologic criteria for this designation, as it has been recently defined (24) (Fig. 1A–C); BRAF V600E mutation was negative in 3 of these cases. The single case that was V600E positive showed a strong cytoplasmic pattern of staining with focal membranous immunoreactivity, compared to the crisp and diffuse membranous staining observed in the remaining three V600E negative E-GBM cases. Staining for cytokeratins (CAM5.2) was focally present in a majority of CD70-positive tumors, both in gliosarcoma (n = 5) and E-GBM (n = 3) (Fig. 1E). By comparison, only 1 out of 4 CD70-negative gliosarcomas showed immunoreactivity with CAM5.2. Expression of CD70 in gliosarcoma was often restricted to reticulin-rich, GFAP-poor areas (Fig. 2D, I). In our study, all but one gliosarcoma were secondary (ie prior diagnosis of high-grade glioma with subsequent radiation; n = 10/14 with available history). A case of primary gliosarcoma (i.e. arising de novo with no history of high-grade glioma and radiation prior to gliosarcoma diagnosis) was included and showed rare staining for CD70 in <5% of tumor cells. No radiation-induced gliosarcomas (i.e. arising post-radiation without a history of high-grade glioma) were evaluated in this study. No tumors, including those containing an epithelioid morphology, showed evidence of SMARCB1 (hSNF5/INI1) gene alterations (i.e. INI1 expression retained on immunohistochemistry) (Fig. 1F). CD70 expression in an anaplastic oligodendroglioma (IDH-mutant and 1p/19q co-deleted) and a diffuse midline glioma, H3 K27M-mutant tumor was also identified (Fig. 3G–J). Representative images of CD70-negative cases are shown in Supplementary Data Figure S1.

Analysis of CD70 expression across the groups using the 2016 WHO classification showed disproportionate, but non-significant, expression in IDH-wild-type glioblastoma (p = 0.11) (Fig. 4A). Stratification by IDH-wild-type variants showed a significantly higher percentage of CD70 expression in gliosarcoma and E-GBM compared to IDH-wild-type glioblastomas with classic histologic features, as well as to the remaining diagnostic entities (p < 0.05; Fig. 4B). In a separate dataset (Vital cohort), mRNA expression was shown to be higher in gliosarcoma (n = 2) compared to glioblastoma samples (n = 26) (Fig. 4C). Additionally, data from the TCGA datasets showed CD70 mRNA expression to be significantly higher in mesenchymal tumors compared to neural, proneural, and classical gene expression-based subtypes (based on subtype signatures by Verhaak et al. (38)) (p < 0.001; Fig. 4D). A significant correlation between CD70 mRNA and protein expression has previously been described (21).

Expression of CD70 Is Associated With Prior Radiotherapy and Improved Survival in Secondary Gliosarcoma

Of the samples that were obtained after radiation treatment and with available history (n = 14), all but 1 case demonstrated moderate-to-strong CD70 expression (Table 5). Survival analysis showed significantly prolonged post-radiation survival in cases of gliosarcoma with moderate-to-high (2–3) expression of CD70 compared to CD70 low or negative (0–1) gliosarcoma (p < 0.05; Fig. 5A). Overall survival in gliosarcomas based on CD70 expression showed a trend towards increased survival in CD70-positive tumors (groups 2–3 vs groups 0–1, p = 0.062; Fig. 5B). The remaining comparisons showed no significant differences in post-radiation or overall survival. Comparison of paired primary and recurrent tumor samples included CD70-positive (n = 3) and CD70-negative (n = 9) post-radiation samples. In these groups, all treatment naïve tumors (primary diagnosis, n = 12) were negative or showed rare weak cytoplasmic CD70 staining. Hence, 3 paired cases showed new-onset moderate-to-strong expression of CD70 at recurrence after radiotherapy. The time from the last dose of radiation to CD70 staining at recurrence ranged from 6 to 58 months; this was similar to cases without post-radiation CD70 expression (4–50 months). Similar to our findings, analysis of TCGA data showed shorter overall survival in recurrent non-G-CIMP tumors with low CD70 mRNA expression (n = 6, median 17.3) compared to high CD70 mRNA expression (n = 6, median 51.05) (p < 0.01; Fig. 5C).

CD163+ Tumor-Associated Macrophages Are Significantly Increased in HGG With Moderate-to-Strong Expression of CD70 Without an Accompanying Increase in CD27+ TIL

Expression in all tumors with moderate-to-strong CD70 expression was coupled with a significant increase in CD163+ TAM (p < 0.05; Fig. 6). Staining for CD68 and HLA-DR showed a similar distribution of expression to CD163 (Supplementary Data Fig. S2). Additional comparisons of TAM, including CD70+/− gliosarcomas, were not significant. There was no significant increase in CD27+ TIL in all group comparisons.