Analysis of Histomorphological/molecular Association and Immune Checkpoint Regulators in Epithelioid Glioblastoma and Pleomorphic Xanthoastrocytoma: Are These Tumors Potential Candidate for Immune Checkpoint Blockade

for immunotherapy. Diaminobenzidine was used as a chromogen. The presence and absence of markers including intensity, staining pattern and distribution was noted.


Introduction
Epithelioid glioblastoma is a rare provisional variant of IDH-wildtype GBs introduced in the 2016 WHO classi cation [1]. It is predominantly composed of solid aggregates or cohesive sheets of small to medium sized melanoma-like cell with variable "rhabdoid" features [2,3]. Diagnosis is often challenging as there are no speci c immunohistochemical (IHC) and molecular markers, with only few small series been reported till date. [3,4,5.6]. Most eGB's are recognized as de-novo tumors, however a few cases preceded by or concomitant with lower grade precursors have been reported [7,8,9]. Approximately 16.6-93 % of eGB's harbor BRAF V600E mutation, which is infrequent in other types of GBs [1,5,10]. eGB and PXA, especially A-PXA sometimes show overlapping cytological(discohesive epithelioid pattern and perivascular/intercellular lymphocyte in ltration) and molecular (BRAF V600E mutation) features thus making the distinction between these subtypes occasionally di cult [2,4,7,8]. There is a recent case report of eGB arising in a pre-existing PXA and studies documenting molecular overlap between the two entities [11, 2, 12,, 13].
Despite advances in therapy, prognosis for patients with GB remains dismal [14,15]. Recently in eld of neurooncology, there is a great interest to understand and evaluate the therapeutic potential of immunotherapies targeted towards activating and enhancing endogenous host immune responses including vaccines, adoptive T cell strategies and modulators of immune checkpoint regulators such as programmed death 1(PD-1), programmed death ligand 1(PDL-1) and cytotoxic T lymphocyte antigen 4(CTLA-4) [16,17]. Immune-based treatments exert an indirect anti-tumor effect by generating potent, tumor-targeting immune responses and enhancing T cell functions [18]. Immune checkpoint inhibition has been shown to produce prolonged clinical remissions in subsets of patients with advanced solid organ malignancies like melanoma and small cell lung carcinoma [19]. Among brain tumors, many clinical trials using immune checkpoint blockers (ICB) are ongoing in patients with GB [20]. Owing to their rarity, there is no detailed study on status of PDL-1 and CTLA-4 expression in eGB's and PXA's.
This study was undertaken with the aim of analyzing the clinical, histopathologic and molecular pro le of eGB's and PXAs to establish similarities and differences in the two groups. In addition PDL-1 expression, CTLA-4 expression and T-cell in ltration was assessed to identify genetic subgroups that could bene t from the targeted therapy and determine their prognostic signi cance.

Case selection
Histologically con rmed cases of eGB and PXA diagnosed over a 3-year period (2016-2018) were retrieved from the archives of four tertiary care hospitals.
Approval was obtained from institutional ethics committee for conducting experiments on human patient samples. Cases having adequate tissue in para n blocks were identi ed. Fourteen eGB's and twenty-one PXAs (six A-PXA and 15 PXA) were included in the study. Eighteen 1° GB's were also taken to compare with there molecular and immune pro le. Histopathological features were reviewed by two independent pathologists (VS, SM) according to the current WHO classi cation of CNS tumors. Patient demographics, tumor location and size, imaging ndings, histopathological ndings and clinical outcome were noted.
Endothelial cells, cortical neurons and in ltrating in ammatory cells were generally positive and served as internal positive controls. In cases with inhomogeneous immunoreaction, areas with the highest staining were scored. Tumours with > 1% of cells showing PDL-1 membrane staining of any intensity were considered positive. PDL-1 tumor proportion score (TPS) was calculated as a percentage of total tumor component showing focal/complete membranous staining of any intensity and was scored as weak (1-5%), moderate (> 5-25%), or strong (> 25%). Cytoplasmic, brillary and stromal PDL-1 staining was also observed occasionally but these cases were not considered positive, as the signi cance of nonmembranous PDL-1 staining remains unclear. The CTLA-4 intensity of positive T cells were recorded as negative(0), weak(1-5%), moderate (6-25%) and strong (> 25%) based on percentage of lymphocytes showing positivity. For tumor in ltrating lymphocytes (TILs)-CD3 and CD8, the percentage of strongly positive cells over all nucleated cells was calculated using a hotspot approach. The whole tumor sections were scanned at a low power magni cation (4x objective), and the areas of highest density of CD3 and CD 8 positive cells in direct contact with tumor cells at 400x visual eld (40x objective × 10x ocular) were enumerated. Intra-vascular, perivascular and TILs in areas around zones of necrosis were excluded during counting.

Fluorescence In Situ Hybridization
Dual-probe uorescence in situ hybridization (FISH) assay was performed on para n-embedded sections for assessment of EGFR ampli cation. Signals were scored in at least 200 non-overlapping, intact nuclei. Sections from non-neoplastic cortical tissue obtained from epilepsy surgery specimens were used as a control for each probe pair. Locus-speci c probes paired with centromere probes for chromosomes 7 (CEP7, Vysis, Downers Grove, IL) was used for EGFR assay. EGFR ampli cation was considered when > 10% of tumor cells showed either an EGFR: CEP7 ratio of > 2 or innumerable tight clusters of signals of the locus probe [6].

Mutation analysis of BRAF V600E
DNA was extracted from 10-micron sections of formalin-xed para n embedded (FFPE) tissue using the Recover all nucleic acid extraction kit using manufacturer's protocol (Invitrogen, Carlsbad, CA). DNA was quantitatively and qualitatively examined using Qubit (Invitrogen, Carlsbad, CA) and agarose gel. For all the cases, 50 ηg of total DNA was PCR ampli ed for detection of mutation in BRAF (codon 600). All PCR reactions were performed in a total of 20 micro litre reaction mixture using the Taq DNA polymerase (Invitrogen, Carlsbad, CA) with the following conditions: Denaturation at 95°C (10 min) followed by 42 cycles of denaturation at 95°C (30 s), annealing at 60°C (35 s) and elongation at 72°C (90 s). Bidirectional sequencing was performed using the Big Dye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Courtaboeuf, France) using the ABI 3500xL sequencer (Applied Biosystems, Foster City, CA) [6].

Statistical analysis
All statistical analyses were performed using IBM SPSS version 23 software (SPSS Inc., Chicago, IL).
Associations among clinicopathological factors were tested using Fisher's exact test or non-parametric t test (Mann-Whitney test). Survival analysis was performed using the Kaplan-Meier method and the logrank test. All statistical tests were 2-sided and p < 0.05 was considered statistically signi cant.

Results
Fourteen eGB's and twenty-one PXAs (including six A-PXA) were available for the study. There were 23 male and 12 female patients. Thirteen (37.2%) were pediatric and twenty-two (62.8%) were adult. Mean ages for PXA and eGB patients were 21.2 (range 11-42) and 30.9 (range 7-50) years, respectively. Most PXAs and eGB's occurred in the cerebral hemispheres while one each was seen in the cerebellum. The neuroradiological images showed similar nonspeci c features that were peripheral smooth, well demarcated, contrast enhancing and localized super cial cerebral masses with perilesional edema. PXA in addition often had single and/or multiple cysts.
The median follow-up interval for all patients was 26 months (range 2-48months). 4 A-PXAs (66.67%) and 4 PXA (26.67%) recurred; the median interval to recurrence for the former was 14 months while for the latter was 28.5 months. Amongst the recurrence one case each of A-PXA (case 19) and PXA (case 21) progressed to eGB and one PXA progressed to A-PXA (Case 30). All the patients with eGB except two died of disease within 2 years of diagnosis, respectively. However, the two eGB patients still remaining alive at last follow-up (Cases 12 &13) with survival times already exceeding 2 years (26 and 32 months at last follow-up), suggests that long survival times are seen in this GB subset.
Clinical, histopathological, immunohistochemical and molecular ndings are summarized in Fig. 1a PXA cases were characterized by spindled cells forming fascicles, scattered bizarre mono-and multinucleated cells with or without xanthomatous change, perivascular lymphocte cu ng, a dense pericellular reticulin network, and numerous eosinophilic granular bodies (EGB's) (Fig. 2a). A-PXAs in addition showed ve or more mitosis, microvascular proliferation(MVP) and necrosis. However 4 of 6 cases of A-PXA (66.67%) also showed focal areas of epithelioid pattern of tumor cells, which lacked xanthomatous change, EGB's and reticulin ber deposition (Fig. 2b). Two PXA/A-PXA (Case 19 and 21) cases, which transformed to eGB on follow-up, lost the pleomorphism and degenerative changes such as EGB's, reticulin ber deposit and intracellular xanthomatous changes. The eGB's were composed almost exclusively of epithelioid cells, frequent mitoses and areas of necrosis (Fig. 2c)  without and one PXA (Case 21) with BRAF V600E mutation acquired CDKN2A homozygous deletion at recurrence and transformed to eGB on follow up (Fig. 2d). None of the 1° GB's showed BRAF V600E mutation. There was 100% concordance between BRAF V600E mutation detection by IHC and Sanger sequencing) (Fig. 2e,f,i,j). Cases with positive BRAF mutation were associated with longer median progression free survival (PFS) of 27 months vs 16.5 months in those with absent mutation (P-value = 0.023), however overall survival (OS) analysis did not reach was statistical signi cance (P-value = 0.125) (Fig. 2g,h). None of the PXA or A-PXA cases showed EGFR ampli cation. However 14.2% eGBs showed EGFR ampli cation in contrast to 1° GB's which showed ampli cation in 66.66%, thus pointing towards the fact that eGBs are distinct from 1° GB's on molecular bases as well.

Immune pro le
Immune pro le ndings are summarized in Fig. 3.
Quantitative analysis of tumor-in ltrating lymphocytes CD3 and CD8 expression was seen in 92.9%-100% cases of eGB, PXA and A-PXA, however only 66.7% and 55.6% cases of 1°GB's showed positivity for these markers (P-value = 0.001 each) (Fig. 5d) (Fig. 5e-f)). In addition we noted that the proportion of tumor cells expressing PDL-1 signi cantly correlated with increasing density of TIL's/100 tumor cells in all tumor subtypes (Spearman correlation co-e cient r = 0.29, p = 0.03)

Discussion
Recently studies have shown that sometimes PXAs particularly anaplastic PXAs show considerable clinical, radiological, histological and molecularly overlap with eGBs resulting in di culties when attempting to segregate these entities [2,3,7]. Though eGB occurs in de novo fashion in the majority of cases, some cases arose from lower-grade gliomas as well [7,8,9,11,22]. The fact that most of these lower-grade lesions documented so far were PXA and that both tumors commonly exhibit BRAF V600E mutation re-enforce the possibility that eGB and PXA are related [2,11,23]. Two independent studies in addition revealed a striking similarity of genetic alterations (BRAF V600E, TERT promoter mutations and CDKN2A/B homozygous deletions) in anaplastic PXAs and eGBs [24,25]. Further, eGBs have shown PXA or astrocytoma like areas with presence of BRAF mutation in both components [5,8,26], thus suggesting that BRAF mutation might possibly be an early event to both tumors and additional genetic alterations such as TERT-p, LSAMP or CDKN2A, is essential for progression of PXA to eGB. [5,13,27]. Supporting this hypothesis were our two cases of PXA and A-PXA that progressed to eGB after acquiring CDKN2A homozygous deletion. Tanaka et al also recently reported a case of eGB developing within the tumor bed of a PXA, thirteen years after initial resection [11].
A recent study by Kurshonov et al documented considerable molecular and clinical heterogeneity within the eGBs by global DNA methylation and CNV analyses and established three distinct subsets: a PXA subset, with a high percentage of BRAF V600E mutations, but a relatively low percentage of TERT promoter mutations; an adult IDH-wild-type GB subset, with a relatively low percentage of BRAF V600E mutations, but a high percentage of TERT promoter mutations; and a pediatric RTK1 subset not harboring either mutation. They proposed that it is likely that the "epithelioid" GB phenotype represents a mere histologic pattern rather than a variant or separate entity [12].
In the current study PXA/A-PXA and e-GB exhibited some common ndings on MRI: (i) contrast-enhanced solid and/or cystic; (ii) well circumscribed with no or little peritumoral edema; (iii) leptomeningeal involvement and dissemination. Histologically areas resembling PXA with focal lack of cytological uniformity along with spindled cells forming fascicles, few bizarre multinucleated cells, accompanied with in ltrating lymphocytes and perivascular lymphocytic cu ng were identi ed in three of fourteen (21.4%) eGB's. However it was relatively inconspicuous in comparison to the adjacent, more abundant epithelioid rhabdoid looking tumor component as it constituted around 5-10% of the entire tumor raising the possibility that at least in few eGB's, a precursor PXA component may be overgrown by the more malignant epithelioid component or was not sampled in the resection specimen. Mitotic gures, MVP and necrosis were found in both eGBs and A-PXA. Further, no reliable IHC or molecular differences were observed between the two glioma subtypes. In addition in our series A-PXA behaved clinically similar to eGB's, with recurrences and death of a patient within 26 months of initial resection. All these ndings suggest that eGB and PXA particularly A-PXA are either the same entity or highly related. eGBs differs from conventional GB's in that its course is often complicated by early recurrence, intratumoral hemorrhage and leptomeningeal spread and in addition it does not show EGFR ampli cation, IDH1 gene mutation or PTEN deletion, but instead, about 16.6-93% of these cases harbor BRAF V600E mutation [3,8], a nding that remained true in our study as well.
In our study 100% concordance was found between Sanger sequencing and IHC results for BRAF V600E mutation, thus immunostaining could be used as a subsitute and cost-effective method for veri cation of the mutation. However any negative or low staining cases may be selected to undergo genetic analysis based on other clinical and histopathologic features. Other studies have also documented a high concordance between IHC and ARMs/Sanger sequencing with senstivity and speci city of IHC staining being 97-100% [28.29].
With the success of Immune checkpoint blockade (ICB) in various solid organ malignancies, the potential candidature of brain tumors for ICB is now being actively explored. Recent studies have reported PDL-1 and CTLA4 expression in GBs, which are targetable by prospective immunotherapies. Quanti cation of PDL-1 expression on tumor cells and CTLA-4 on T lymphocytes by IHC is widely being used as a predictive biomarker for ICB response [17,30]. Adaptive upregulation of PDL-1 as an immune escape mechanism is usually seen in tumors with increased CTL in ltrates [31]. A recent clinical trial reported lack of response to ICB in a small cohort of recurrent GBs despite PDL-1 expression and attributed the failure to the scarcity of intra-tumoral T-lymphocytes [32]. Hereby we used a simplistic approach of IHC for PDL-1, CTLA-4, CD3 and CD8 to identify whether eGB's and PXA are potential candidates for ICB.
In our small cohort the frequency and TPS of PDL-1 expression in PXA and eGB's was found to be higher than 1°GB though the difference could not meet statistical signi cance p-value = 0.097 and 0.342). The frequency of PDL-1 expression was 71.4% in eGB, 66.67% in APXA and 60% in PXA as compared to 1°GB (44.4%). Although overall positivity rate reported in adult gliomas is 44.72%, a wide range of PDL-1 expression (6.1 to 100%) has been noted across different studies [33]. Contrary to the studies on gliomas by Garber et al and Bergoff et al who documented high frequency of PDL-1 expression in only grade IV tumors, we observed its higher expression in PXA (grade II) and A-PXA (Grade III) cases [34,35]. Although limited, the available studies present considerable differences in correlating PDL-1 expression with the patient outcome. In meta-analysis of 1052 patients, Xue et al. demonstrated high/positive PDL-1 expression to be associated with poor OS, which was also documented by Nduom et al [36,37]. However Berghoff et al and Heiland et al observed no association of PDL-1 with OS [35,38]. In present study also we could not elucidate any association between PDL-1 positivity and OS or PFS in tumor subtypes. Further, in all tumor subtypes, there was no association of PDL-1 with any speci c genetic alteration (Fig. 2).
Both frequency and intensity of CTLA-4 expression was found to be signi cantly higher in eGB and PXA cases as compared to 1°GB respectively (P-value = 0.001 and 0.0008). Though CTLA-4 is a well studied immune checkpoint protein, its expression in glioma and its effects on prognosis was yet not examined until recently Liu et al in his study documented that CTLA-4 have a positive correlation with PD-1 and TIL's and its higher expression are observed in higher grades of glioma and resulted in poor OS [39]. Though no association of CTLA-4 with overall survival was documented in our study (P-value = 0.1286), a variety of clinical trials targeting CTLA-4 have demonstrated promising bene ts in patients with glioma therefore comprehensive analysis of CTLA-4 expression will still be required to identify its enrichment criteria in gliomas [40].
Frequency and density of TIL's expression was signi cantly higher in PXA (100% for both CD3 and CD8) and eGB (CD3-100%, CD8-92.9%) cases as compared to 1°GB (CD3-66.7%, CD8-55.6%) (P-value = 0.001 each, P-value = 0.0004 and 0.0001). Our ndings were concordant with previous study by Berghoff et al and Kim et al who reported CD3 expression in 66.7 to 89.6% and CD8 expression in 44.4 to 77.6% GB's [35,41]. Though there was statistically no signi cant difference between low and high-density groups of TIL with respect to overall survival in our study (p-value = 0.21), few studies have observed prolonged survival with increased TIL [41,42]. However Hans documented that CD8 + TILs alone cannot effectively predict patient outcome in GB whereas high CD4 + TIL levels in combination with low CD8 + TIL levels were associated with unfavourable prognosis [43].
We acknowledge that the power of survival analyses in our series is limited by the sample size and the retrospective study design, and our results surely need con rmation in larger and prospectively collected cohorts. These observations together offer some circumstantial evidence that PXA and eGB's are more immunogenic thus reinforcing exploitation of PD-1/PDL-1 and CTLA-4 immune checkpoints for immune evasion.

Conclusion
This comprehensive analysis highlights homogenous molecular pro le of eGBs and PXA, suggesting that they are closely related. In addition increased TIL densities and upregulation of PDL-1 and CTLA4 in these tumors suggests potential candidature for immunotherapy. Clinical trials with speci c immune checkpoint inhibitors are warranted and it remains to be explored whether PDL-1/TIL expression patterns, patient's age or molecular alterations will correlate with response to such treatments.

Declarations
Acknowledgement: All staff of Neuropathology and Neurosurgery Department of AIIMS, New Delhi for support in accomplishment for this work. Dr. Ajay Garg contributed by providing radiological data