A multicenter, phase II trial of GC1118, a novel anti‐EGFR antibody, for recurrent glioblastoma patients with EGFR amplification

Abstract Background We evaluated the therapeutic efficacy of GC1118, a novel anti‐epidermal growth factor receptor (EGFR) monoclonal antibody, in recurrent glioblastoma (GBM) patients with EGFR amplification. Methods This study was a multicenter, open‐label, single‐arm phase II trial. Recurrent GBM patients with EGFR amplification were eligible: EGFR amplification was determined using fluorescence in situ hybridization analysis when a sample had both the EGFR/CEP7 ratio of ≥2 and a tight cluster EGFR signal in ≥10% of recorded cells. GC1118 was administered intravenously at a dose of 4 mg/kg once weekly. The primary endpoint was the 6‐month progression‐free survival rate (PFS6). Next‐generation sequencing was performed to investigate the molecular biomarkers related to the response to GC1118. Results Between April 2018 and December 2020, 21 patients were enrolled in the study and received GC1118 treatment. Eighteen patients were eligible for efficacy analysis. The PFS6 was 5.6% (95% confidence interval, 0.3%–25.8%, Wilson method). The median progression‐free survival was 1.7 months (range: 28 days–7.2 months) and median overall survival was 5.7 months (range: 2–22.0 months). GC1118 was well tolerated except skin toxicities. Skin rash was the most frequent adverse event and four patients experienced Grade 3 skin‐related toxicity. Genomic analysis revealed that the immune‐related signatures were upregulated in patients with tumor regression. Conclusion This study did not meet the primary endpoint (PFS6); however, we found that immune signatures were significantly upregulated in the tumors with regression upon GC1118 therapy, which signifies the potential of immune‐mediated antitumor efficacy of GC1118.


| INTRODUCTION
Glioblastoma (GBM) is one of the most devastating malignancies with a median survival of 15 months. 1 Most patients eventually succumb to recurrent disease despite intensive care; however, none of the current treatment has shown clinically meaningful efficacy for recurrent tumors. 2 Many studies explored the genomic landscape of GBMs; alterations of the epidermal growth factor receptor (EGFR) are found in ~60% of GBM patients, including mutation and copy number amplification. These alterations usually result in constitutive activation of EGFR signaling, for example, EGFRvIII, the most common EGFR mutation in GBM, activates the tyrosine kinase without ligand binding. 3,4 Therefore, given tumor specificity and frequency, EGFR has been considered a compelling therapeutic target for GBM.
Numerous anti-EGFR agents have been evaluated in GBMs. 5 However, all failed to show survival benefit including EGFR tyrosine kinase inhibitors and EGFRvIIItargeting peptide vaccine. 6,7 Several reasons have been suggested to account for these failures-genetic heterogeneity spanning inter-tumor and intra-tumor scales, signaling redundancy, the blood-brain barrier, etc. [8][9][10][11] Importantly, the profile of EGFR alteration in GBM differs from other solid cancers with sensitivity to EGFR tyrosine kinase inhibitors. EGFR mutations found in lung cancer usually affect the intracellular kinase domain, while EGFR mutations of GBMs affect the extracellular domain and are found in the context of EGFR amplification.
Regarding this GBM-specific characteristic, an anti-EGFR antibody can be an ideal option for GBMs. While previous studies with anti-EGFR antibodies were disappointing, 12,13 the initial success of depatuxizumabmafodotin, an antibody-toxin conjugate targeting the EGFR, was encouraging. 14 GC1118, a novel anti-EGFR antibody, may have selective advantages to GBMs over other anti-EGFR antibodies; first, GC1118 has a distinct binding epitope. It recognizes a unique and critical EGFR epitope for EGF binding which does not overlap with those of other anti-EGFR antibodies. 15 Second, GC1118 has superior inhibitory activity against high-affinity ligands, 15 which are dominant in GBMs. Third, it can pass through the blood-brain barrier and even brain-tumor barrier as shown in a previous in vivo study. 16 GC1118 has already demonstrated potential antitumor efficacy in colorectal and gastric cancers. 17,18 We also observed a comparable antitumor effect of GC1118 in in vitro and in vivo GBM models. 16 Moreover, this efficacy was associated with high EGFR amplification. Based on these findings, we expected that GC1118 would be beneficial for GBM patients with EGFR amplification.
In this study, we aimed to evaluate the antitumor effect of GC1118 against GBM patients with EGFR amplification. We also conducted genomic analyses to verify the molecular correlates to clinical response.

| Study design and patients
This phase II, open-label, single-arm study was conducted at Samsung Medical Center and Seoul National University. The primary endpoint was the 6-month progressionfree survival rate (PFS6). Secondary objectives included progression-free survival, overall survival, radiographic response rate, and safety.
Eligible patients were ≥19 years of age with life expectancy of ≥3 months, Karnofsky performance status score ≥70, and adequate bone marrow and end-organ function. We only included the recurrent GBM patients, who progressed following the initial Stupp regimen. 19 EGFR amplification was confirmed by fluorescence in situ hybridization (FISH) analysis at each center-a sample was defined as amplified if the EGFR/CEP7 ratio was ≥2 and a tight cluster EGFR signal (at least 15 gene copy number) was found in ≥10% of cells. Patients who had received any EGFR targeting agents including small molecules or monoclonal antibodies were excluded.
This study (NCT03618667) was in compliance with the Declaration of Helsinki and guidelines on Good Clinical Practice. Ethics approval was obtained from local institutional review boards of each hospital (Samsung Conclusion: This study did not meet the primary endpoint (PFS6); however, we found that immune signatures were significantly upregulated in the tumors with regression upon GC1118 therapy, which signifies the potential of immunemediated antitumor efficacy of GC1118.

| Study procedures
The recommended dose for phase II trial was determined to be 4 mg/kg weekly according to the first in human trial. 20 Patients were treated with GC1118 on Days 1,8,15, and 22 of a 28-day cycle. GC1118 treatment continued up to six cycles until the disease progressed or unacceptable toxicities occurred.
Follow-up included a weekly physical, neurologic examination, complete blood counts, and a chemical battery every 2 weeks. Brain imaging (MRI) was performed every 8 weeks. Treatment response was evaluated according to RANO criteria. 21 We measured two diameters of each target lesion found in T1 contrast-enhanced images; i , the maximum length across and i , the corresponding perpendicular one to i . We compute the sum of products of two diameters lesions) and compare this metric to that of baseline image. Radiographic response of existing lesion(s) was defined as follows; partial response (PR) as ≤−50%, stable disease (SD) as ≥−50% and <25%, and progression of disease (PD) as ≥25%.
Treatment was interrupted for Common Terminology Criteria for Adverse Events (AEs) version 4.03 Grade 3 drug-related non-hematologic toxicity (except alopecia, nausea, vomiting, and fatigue). Treatment was resumed at the physician's discretion and a dose reduction was permitted up to 2 mg/kg (1 mg/kg at each decision). Patients who experienced three or more sequential interruptions in treatment were permanently excluded from the study.

| Statistical analysis
All patients who received at least one dose of GC1118 are eligible for safety analysis. For efficacy analysis, patients who were treated by at least one dose of GC1118 and had available tumor assessment were included.
The primary endpoint was PFS6. We applied the binomial test to compare the PFS6 of this study to previous results from the available relevant literatures (10% [range: 11%-20%]). [22][23][24][25][26] To detect an improvement from 10% to 35%, with the power 0.8 and alpha 0.05, 18 subjects were required according to A'Hern's method. 27 Sample size was determined to be 18 subjects, and 23 subjects were required assuming a 20% dropout rate.
All statistical analyses were performed using R version 3.6.3 (http://www.R-proje ct.org). 28 Continuous variables are presented as median values and ranges or mean values with standard deviations (s.d.). p ≤ 0.05 was used as a threshold for statistical significance.

| Biomarker analysis
Tumor tissue from initial diagnosis was collected and subjected to next-generation sequencing. For the majority of patients, fresh frozen tumor tissue was available except six patients; for these patients, archival formalin-fixed paraffin-embedded tissue was used.
Whole-exome sequencing data were processed accordingly as described in previous literature. [29][30][31][32] Briefly, somatic mutations were detected by MuTect and copy number variations were estimated using ngCGH and ABSOLUTE algorithms. 33,34 RNA-sequencing data were processed for read counts and structural variation. We used the R package DEGseq for read count normalization (reads per kilobase of transcript per million read) and differential gene expression analysis. 35 To detect EGFRvIII, GSNAP was used. 36 For downstream analysis, pre-ranked gene set enrichment analysis was performed using "fgsea." 37 We used CIBERSORTx for deconvolution analysis. 38 A more detailed method is provided in supplementary method.
For safety analysis, we included the patients who received at least one dose of GC1118 (n = 21). Among these (n = 21), three patients discontinued GC1118 treatment after one cycle without tumor assessment: two patients refused to take further treatment after two doses of GC1118 due to a drug-related adverse effect (skin rash); another patient experienced rapid clinical deterioration immediately after the first drug injection. Consequently, 18 patients were eligible for efficacy analysis ( Figure S1).
The median progression-free survival was 1.7 months (range: 0.9-7.2 months) and median overall survival was 5.7 months (range: 2-22.0 months). Objective response rate was 5.6% (1 PR) and disease-control rate was 22.2% (1 PR and 3 SD) by RANO criteria (Figure 1). We observed four cases with tumor regression-two of them (SNUH-003 and SMC-0002) showed a regression rate more than 50%, but one patient was defined as PD due to development of a new lesion (Figure 1).
A single patient exhibited a notable response to GC1118; patient SMC-0002 showed regression of a contrast-enhancing lesion following GC1118 treatment ( Figure S2); however, the patient experienced a Grade 3 skin rash that necessitated dose reduction. GC1118 was reduced to 2 mg/kg, and the patient eventually developed a recurrent tumor after discontinuation of therapy. Interestingly, the epicenter of the recurrent tumor was located aside from the previously responding lesion ( Figure S2).

| Toxicity
GC1118 was generally well tolerated except for skin toxicity; none of patients experienced a Grade 4 treatmentrelated AE. Treatment-related AE profiles for all patients (n = 21) are summarized in Table 2.
Skin toxicity was the most frequent AE: 76.2% of patients (16/21) experienced skin rash of at least Grade 1. Grade 2 skin toxicity was found in nine patients, and Grade 3 skin rash was noted in four patients; these four patients needed treatment interruption including dose reduction or skipped dosages.
The main cause of dose reduction was skin toxicity; GC1118 was reduced to 3 mg/kg in six patients and 2 mg/ kg in two patients. Two patients (SMC-0002 and SNUH-008) were hospitalized because of skin rash attributed to GC1118. SMC-0002 patient maintained the treatment with dose reduction and intermittent dose omission, while SNUH-008 patient was dropped out due to three consecutively missed drug dosages.

| Genomic landscape of study cohort
Genomic data were available for 20 patients, including all patients in the efficacy analysis group (Figure 2). The median of the log2 value of EGFR copy number was 1.7 (range: 0.2-3.7); five tumors were non-amplified according to this analysis. As chromosome seven gain is common in GBMs, we applied ABSOLUTE to estimate the absolute copy number while adjusting the tumor purity and ploidy. 34 Accordingly, most tumors showed high absolute copy number with two exceptional cases (mean ± s.d., 12.7 ± 3.2). This implied 90% agreement between the two methods, FISH and whole exome sequencing. Gain-of-function EGFR mutations were found in 11 patients. Most of them were derived from the extracellular domain (81.8%, n = 9/11). Four patients co-expressed EGFRvIII and gain-of-function EGFR mutation. A289 was the hot spot where most variations occurred (63.6%, n = 7/11). A previous study suggested that A289 missense mutations were associated with the clinical response to depatuxizumab-mafodotin, potentially by modifying the receptor sensitivity. 39 However, we did not find any significant relationship between A289 mutations status and clinical outcome in our study. In contrary, tumors with A289 mutation showed more progression (85.7%, n = 6/7) compared to tumors without A289 mutation (40%, n = 4/10), which had no prognostic implication ( Figure S3).
To comprehend the molecular characteristics underlying clinical response, we investigated transcriptomic profiles. As only a single patient met the primary endpoint, we focused on whether existing tumors experienced regression during therapy. We found four patients with tumor regression during therapy (SMC-0002, SMC-0014, SNUH-003, and SNUH-007) and compared them to the remaining patients to identify the genomic signatures associated with response to GC1118.
GC1118 shows preferential inhibitory effect against highaffinity EGFR ligands, thereby exerting a more profound therapeutic effect in a subset of tumors. 15 We compared the mRNA expression of EGFR ligands. As anticipated, highaffinity ligands were dominant in GBM tumors. We did not find any difference in terms of ligand expression between tumors with distinct responses ( Figure S4).
We conducted pre-ranked gene set enrichment analysis using differentially expressed genes between patients with tumor regression and without regression. Immune-related pathways were enriched in the tumor-regression group, while pathways involved in intercellular communication (e.g., synapses) were downregulated ( Figure S5). This F I G U R E 1 Radiographic response of existing lesion(s) following GC1118 treatment. We measured the two diameters of target lesion(s) on T1 post-contrast images and computed products of perpendicular diameters. We compared the sum of products with baseline to evaluate the radiographic response and presented the best response in each case. Overall response was assessed by RANO criteria and presented with different colors (black-PD; gray-SD; white-PR). *indicates development of new lesion including leptomeningeal seeding. All patients marked with * showed a decrease or stable state of target lesion(s); however, they were assessed as having disease progression owing to developing new lesion(s) remote from the original target lesion(s). PD, progression of disease; PR, partial response; SD, stable disease. result was further supported by single-sample gene set enrichment analysis: genetic signatures involved in antigen processing during the adaptive immune response were upregulated in the tumor-regression group ( Figure S5). To understand the functional implications of enriched immune pathways in the tumor-regression group, we deconvoluted the bulk RNA-seq data into various cellular compositions with LM22 reference using CIBERSORTx. γδ-T cells were significantly more abundant in the tumorregression group (p-value = 0.008, Wilcoxon signed-rank test). Natural killer (NK) cells, regardless of activation status, were also more frequent in responding tumors despite statistical insignificance ( Figure S5).

| DISCUSSION
GC1118 is a novel anti-EGFR monoclonal antibody, which specifically binds to EGFR and inhibits the downstream cascades of the EGFR pathway. 15 Previously, we identified EGFR amplification as a potential biomarker to predict the clinical response to GC1118 in an in vivo study. 16 We adopted this result to set a phase II clinical trial, but failed to demonstrate significant clinical improvement. PFS6 was 5.6% and only one patient completed the entire course of GC1118.
The major difference between the experimental condition of the previous in vivo study and the clinical trial arises from the timely acquisition of specimens for genomic analysis. Unlike the preclinical study, we conducted genomic analysis using primary tumors while we treated the patients for their recurrent disease. Timely obtaining tumor specimens is important to make accurate molecular diagnoses, but usually impracticable in case of brain tumors.
We hypothesized that tumor evolution might contribute to this failure. We defined the temporal interval-time interval between primary tumor acquisition and GC1118 initiation-and compared it between two patients' group stratified by tumor regression. Interestingly, tumorregression group showed longer intervals compared to the rest (mean ± s.d., 593 ± 428 vs. 357 ± 140, respectively, pvalue = 0.40, Wilcoxon rank-sum test, two-sided).
Not only physical temporal interval, but also therapeutics may influence the evolutionary trajectory by affecting genomic integrity. 40 We investigated the prior treatment history of patients and found that patients in the nonregression group were far more heavily treated. Half of patients in non-regression group had at least two distinct prior treatment regimens (42.9% [N = 6/14]), while all patients in tumor-regression group were treated by standard care alone. Shortly, patients in the non-regression group were more heavily treated during a shorter period. This F I G U R E 2 Genomic profiles of study cohort. An oncoplot depicting the genomic profiles of the study cohort. We focused on several key drivers of GBMs to analyze the mutation and copy number variation. Patients were ordered by radiographic response of existing lesion. Copy numbers derived from "GISTIC" are presented. EGFR "hot spot" variation (A289) is marked with an *. GBM, glioblastoma; EGFR, epidermal growth factor receptor; mMGMT, MGMT methylation status; NA, not available. may suggest innate treatment resistance or the potential of therapy-induced tumor evolution.
Previously, many studies depicted the significance of therapeutics on the genomic evolution of malignant gliomas. 41 Accordingly, prior intense treatment history might have a significant effect on the tumor genome, thereby boosting the shift in genomic landscape of tumors with EGFR dominancy, as shown in our study.
In the context of tumor evolution under therapy, EGFR is prone to experience clonal replacement. 40 The mutational switching phenomenon was also highlighted in several GBM key drivers including EGFR. However, the functional significance of this clonal replacement in EGFR alteration has not been fully elucidated.
A recent study by the GLASS consortium found that the classical subtype, enriched with EGFR alteration, was the most plastic to subtype switching upon relapse. 42 Loss of EGFR amplification was significantly associated with a shift in cell state composition, which involves mesenchymal transition. All these findings suggest that the functional dominance of EGFR alteration may be more vulnerable to change during treatment. However, it is noteworthy that a subset of classical tumors still maintains their transcriptomic profiles as well as genomic dominance of EGFR.
There are additional reasons besides tumor evolution that can account for the failure of this trial. Although we adopted the strict eligibility criteria based on FISH to select the patients with EGFR amplification, the threshold of high EGFR amplification might not be sufficient. In two cases, copy number profiles of tumors were not matched, which signifies 90% agreement of two distinct methods. This disagreement is acceptable according to literatures as whole exome sequencing normalizes copy number across the tissue sample instead of on a cell-by-cell basis as with FISH. 43 However, this discrepancy may indicate the intratumor heterogeneity regarding EGFR amplification. In other words, our criteria did not necessarily select the patients with sufficient amount of tumor cells with EGFR amplification to derive clinical benefit from GC1118.
Coexisting EGFR alterations may be potential confounders that alter the drug efficacy in this trial. A previous study found that missense mutations of the EGFR extracellular domain contributed to the receptor hypersensitivity to ligands, especially low-affinity EGFR ligands. 13 Many tumors in our cohort showed EGFR mutations, mostly affecting the extracellular domain. We found more progressive disease in tumors with an A289 missense mutation, although this trend was not translated into clinical implication. If EGFR is hypersensitized for low-affinity ligands because of this mutation, GC1118's superior inhibitory effect against high-affinity ligands must be weakened. We assumed that this might account for the worse treatment response observed in tumors with A289 mutation, however, we could not draw a concrete conclusion due to the limited number of cases.
Interestingly, we identified that the tumor-regression group was enriched with immune-related pathways by transcriptomic analysis. In a previous study, we suggested that antibody-dependent cellular cytotoxicity might play an important role in eliciting antitumor effects of GC1118 in in vivo models. 16 Antibody-dependent cellular cytotoxicity driven by anti-EGFR antibodies can induce cross talk among immune cells, especially NK cells and dendritic cells; this cross talk can prime antitumor cellular immunity. [44][45][46][47] A synergistic effect against tumor cells by combining anti-EGFR antibody and immunotherapy has been validated in other solid tumors, thus, further supporting the immunologic effect of anti-EGFR antibodies. 48 Moreover, EGFR alteration has been proposed to have immunologic roles in GBMs; EGFR mutation governs the vascular and immune microenvironments by mediating the trans-differentiation of glioma stem cells into pericytes. 49 Our cases do not exactly fit in this model, but it is noteworthy that EGFR alteration may influence the immune landscape of gliomas and provide a stratification scheme to find patients who are sensitive to immunotherapy. A recent study also depicted the potential of EGFR amplification as a surrogate marker for resistance to immunotherapy in GBMs. 50 γδ-T cell and NK cells were upregulated in the tumorregression group according to deconvolution analysis. These cells are involved in the innate immune response and function as cytotoxic lymphocytes. Although γδ-T cells have a dual effect regarding cancer progression, they can directly kill the tumor cells by diverse mechanisms including antibody-dependent cellular cytotoxicity or indirectly induce an antitumor effect by interacting with multiple immune counterparts such as B cells, dendritic cells, and NK cells. 51 All these findings suggest that GC1118 may exert its antitumor effect via immunemediated manner, at least partially.
Collectively, our study fail to show a survival benefit of GC1118 against GBMs with EGFR amplification. Although we compile another failure story with anti-EGFR drug, we have learned several lessons from this study. First, as the EGFR pathway is vulnerable to tumor evolution, timely sampling is mandatory to target EGFR axis. We need to consider combinatorial treatment to overcome the emergence of new clones escaping anti-EGFR treatment. In our study, two out of four patients with tumor regression experienced disease progression due to developing a new lesion or a recurrent lesion aside from the primary one. These findings indicate that anti-EGFR drug alone cannot control the entire tumor cells.
We observed that immune-related genomic signatures were upregulated in tumors with regression. This finding is similar in the context to a recent study, which suggests a hazardous effect of EGFR amplification in response to immunotherapy. These collectively suggest a potential synergism of combining GC1118 and immunotherapy. Albeit speculative, this hypothesis is worth of being explored and should be investigated in the future prospective study.

FUNDING INFORMATION
This study was supported by the GC Biopharma Corp. The funding source had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the article.

CONFLICT OF INTEREST STATEMENT
None of the authors have potential conflicts of interest with this work to declare.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author, S.H.L. The data are not publicly available due to institutional policy.