Expression of Glioma-associated oncogene homolog 1 as biomarker with sonidegib in advanced basal cell carcinoma

The pivotal BOLT (Basal cell carcinoma Outcomes with LDE225 [sonidegib] Treatment) study established the durable efficacy and manageable toxicity of sonidegib 200 mg once daily (QD) through 42 months in patients with advanced basal cell carcinoma (BCC). This secondary analysis used expression of Glioma-associated oncogene homolog 1 (GLI1) as a biomarker to assess the extent of Hedgehog pathway inhibition by sonidegib in patients with locally advanced BCC (laBCC) and metastatic BCC (mBCC). The study enrolled 230 patients, 79 and 151 receiving sonidegib 200 and 800 mg QD, respectively. At week 17, GLI1 expression was reduced from baseline by a median percentage (95% confidence interval) of 88.7% (54.6%–93.0%) and 97.0% (77.5%–98.9%) for aggressive laBCC, 97.5% (80.3%–98.8%) and 95.0% (80.7%–97.5%) for nonaggressive laBCC, and 99.1% (96.4%–99.6%) and 99.3% (95.9%–99.9%) for mBCC in the 200 and 800 mg groups, respectively. Substantial repression of GLI1 was observed in patient subgroups stratified by age, sex, BCC cytological subtype, Eastern Cooperative Oncology Group performance status, lesion site, baseline number of BCCs, and prior radiotherapy. Results support further studies on the inhibition of Hedgehog pathway genes by sonidegib in patients with laBCC and mBCC.


INTRODUCTION
Basal cell carcinoma (BCC) is the most common skin malignancy, affecting more than 10 million new patients annually worldwide and with an incidence that increases by approximately 1% each year [1][2][3]. Most BCCs are treated with surgery, with a favorable prognosis [4]. In cases of locally advanced BCC (laBCC) where surgery is contraindicated, inhibition of the Hedgehog pathway is one of the few approved and recommended treatment options [4,5].
Most BCCs exhibit constitutive activation of the Hedgehog pathway due to mutations in pathway members, most often Patched 1 (PTCH1) and Smoothened (SMO) [6][7][8]. Expression of the transcription factor Gliomaassociated oncogene homolog 1 (GLI1) is a biomarker for Hedgehog pathway activation [9]. Sonidegib, a Hedgehog pathway inhibitor (HHI), blocks pathway signaling by selective inhibition of SMO [10]. It is approved in the US, EU, Switzerland, and Australia for the treatment of adult patients with laBCC not amenable to curative surgery or radiotherapy [11][12][13][14]. In Switzerland and Australia, sonidegib is also approved for the treatment of metastatic BCC (mBCC) [13,14].

Patient disposition, demographics, and clinical characteristics
Overall, 230 patients were enrolled in the study, of whom 79 and 151 were randomized to sonidegib 200 and 800 mg QD, respectively (Supplementary Figure 1). At the time of data cutoff for biomarker analysis (6 months), 39 (49.4%) patients in the 200 mg group and 46 (30.5%) patients in the 800 mg group were still receiving treatment. The biomarker population included 67 and 83 patients randomized to sonidegib 200 and 800 mg QD, respectively. AEs were the most common reason for discontinuation, reported for 16 (20.3%) and 48 (31.8%) discontinuing patients in the 200 and 800 mg group, respectively.
Patient demographics and clinical characteristics were similar between the biomarker and intent-totreat (ITT) populations (Supplementary Table 1). The biomarker population was 62.7% and 65.1% male with mean (standard deviation [SD]) age of 65.6 (15.6) and 64.0 (14.9) years for the 200 and 800 mg groups, respectively. The ITT population was 60.8% and 63.6% male with mean (SD) age of 65.6 (15.7) and 63.6 (14.6) years for the 200 and 800 mg groups, respectively. Patients with laBCC comprised 91.0% and 91.6% of the biomarker population, and 83.5% and 84.8% of the ITT population for the 200 and 800 mg groups, respectively.

DISCUSSION
Results from this secondary analysis suggest that sonidegib treatment led to substantial reductions in GLI1 levels from baseline across doses, BCC subtypes, examined time points, and demographic and baseline disease characteristics in patients with advanced BCC. Marked reductions in GLI1 levels from baseline were observed in patients who achieved disease control with sonidegib, consistent with Hedgehog pathway inhibition.
The Hedgehog pathway plays a key role in regulating cell proliferation and differentiation during development and is involved in the maintenance and repair of many adult tissues including the skin, hair, muscles, and nervous system [19]. Upon activation of the pathway, a signaling ligand of the Hedgehog family binds the transmembrane receptor PTCH1 [20]. This causes the G-protein coupled receptor SMO-sonidegib's target in the pathway-to dissociate from PTCH1 and migrate to the primary cilium, where it releases the cytoplasmic sequestration of the GLI family of transcription factors by Suppressor of Fused through a multistep signaling cascade [20]. Noncanonical pathways of GLI1 activation have also been reported, including by the Ras and p53 families of tumor suppressors [21]. Aberrant GLI1 activation promotes tumor growth, migration, and angiogenesis [21].
Despite the overall durable efficacy of sonidegib treatment observed in the BOLT study, a subset of patients develop resistance to HHIs that can be either primary or acquired after initial response to treatment [22]. Moreover, although GLI1 suppression is dose-and exposuredependent, a reduction of GLI1 expression did not always correlate with tumor response in the phase 1 efficacy and safety study for sonidegib, most likely due to limited sample size, indicating resistance may develop despite GLI1 inhibition [10]. The strong-but not completeinhibition of GLI1 expression in this biomarker analysis is consistent with the possibility that patients with low GLI1 inhibition are resistant to sonidegib. However, there was no significant correlation between sonidegib efficacy and the strength of GLI1 inhibition. This is possibly due to the study not being sufficiently powered to confirm this correlation. Additionally, inhibition of GLI1 in the different patient subgroups examined was uniformly strong, and none of the examined demographic and baseline clinical characteristics showed strong correlation with sonidegib resistance.
Animal and cell-line cancer models studied the impact of sonidegib on Hedgehog pathway activity. In human primary glioblastoma initiating cells, sonidegib reduces levels of GLI1, GLI2, PTCH1, and PTCH2 messenger ribonucleic acid (RNA) and protein, as well as GLI1 and GLI2 translocation into the nucleus [23]. Epistasis testing using sonidegib in combination with direct GLI1 and GLI2 inhibition by RNA interference revealed the effect of sonidegib on Hedgehog pathway downstream targets, including upregulation of the cell death mediators, Fas, Death receptor (DR)4, and DR5, and downregulation of the cell proliferation mediators B-cell lymphoma 2 and Platelet-derived growth factor receptor A [23]. In human renal cell carcinoma lines, sonidegib inhibited GLI1 and GLI2 and reduced cell proliferation in combination with everolimus or sunitinib, whereas direct inhibition of GLI1 and GLI2 in combination with everolimus or sunitinib had no impact on proliferation, suggesting that other targets downstream of SMO may play a role in tumor response [24]. This analysis found no significant association between the magnitude of GLI1 inhibition and time to onset of grade ≥ 2 CK elevation-suggesting that Oncotarget 3478 www.oncotarget.com CK elevation is not influenced by the extent of GLI1 inhibition. CK elevation is a common treatment-emergent AE observed in patients treated with Hedgehog inhibitors. While the exact mechanism responsible for muscle spasms and increased CK levels in patients receiving HHIs is not completely understood, muscle spasms are considered to be linked with paradoxical activation of the SMO/calcium/ AMP-activated protein kinase axis, and the inhibition of SMO signaling leads to an influx of calcium into muscle cells [25]. The pivotal clinical studies of vismodegib (Erivedge ® , Genentech, South San Francisco, CA), an HHI indicated for the treatment of advanced BCC, did not  include CK monitoring; however, muscle spasms were reported in 71.2% of patients [26]. All currently approved HHIs have the same mechanism of action and similar safety profiles; it is thought that CK elevation and muscle spasms are HHI class-effect AEs.
Limitations of the current study include the lack of statistical power to assess specific biomarker-related hypotheses. All biomarker analysis should be considered exploratory and hypothesis-generating, and all P-values were nominal or were adjusted for multiplicity only at the biomarker level within a specific model or analysis. GLI1 was the only examined biomarker, and data on the expression of other prominent members of the pathway were not collected. Additionally, GLI1 expression as related to reduction in tumor mass was not directly evaluated in this analysis. Subsequently, decreases in tumor size may have contributed to reductions in GLI1 levels measured following treatment.
In summary, the reduction of GLI1 expression is consistent with potent inhibition of the Hedgehog pathway by sonidegib in patients with laBCC and mBCC. These results support further clinical studies on the impact of sonidegib on Hedgehog pathway biomarkers.

BOLT study design
This randomized, double-blind, adaptive phase 2 multicenter study adhered to the ethical principles of the Declaration of Helsinki and received approval for its protocol and all amendments from an Independent Ethics Committee or Institutional Review Board at each study site. Written informed consent was obtained from all patients before any study-specific procedures.
Study design is described in detail elsewhere and is briefly summarized here and in Supplementary Figure 2. The study enrolled men and women age ≥ 18 years with a histologically confirmed diagnosis of mBCC or laBCC (not amenable to radiation therapy, curative surgery, or other local therapies) and a World Health Organization (WHO) performance status ≤ 2. Patients were randomized 1:2 to receive sonidegib 200 or 800 mg QD until disease progression, unacceptable toxicity, withdrawal of consent, study termination, or death.

BOLT efficacy assessments
Tumor response was evaluated with Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 for patients with mBCC. For laBCC, the standard RECIST 1.1 criteria are inadequate, since posttreatment morphological changes such as ulceration, cyst formation, and scarring may confound tumor response evaluation. A modified (m)RECIST protocol was developed to assess tumor response in laBCC, integrating central histological review, one-dimensional localized soft-tissue magnetic resonance imaging (MRI) along the lesion's longest diameter per RECIST 1.1, and bidimensional color photography measurements per WHO guidelines.
The primary efficacy endpoint was ORR per central review. Secondary efficacy assessments included TTR and best overall response (either CR, PR, StDis, progressive disease, or unknown). Best overall response was assessed by central review according to mRECIST in patients with laBCC and RECIST 1.1 in patients with mBCC. mRECIST criteria includes a combination of MRI scans, color photography, and baseline and follow-up histopathology data for patients to assess tumor response.

BOLT safety assessments
Safety assessments included AEs monitored throughout the study, coded using the medical Dictionary for Regulatory Activities version 19.0, and assessed for toxicity using the National Cancer Institute Common Terminology Criteria for Adverse Events version 4.03. CK levels were assessed prior to starting treatment or within 72 hours of the first dose, then weekly during the first 2 months, and every 4 weeks thereafter while on study treatment.

Biomarker assessments
Biopsies were collected from accessible lesions at Screening, weeks 9 and 17 predose, and within 21 days after the last dose of study drug. For patients with multiple lesions, no specific lesion was designated for biopsy, and any lesion could be used. Presence of tumor tissue in biopsy samples was histologically confirmed prior to biomarker analysis.
GLI1 expression at Screening was used as baseline measurement. Gene expression at all examined time points was assessed using reverse transcriptase polymerase chain reaction, in terms of the number of reaction cycles needed to reach the threshold amount of product. Expression of GLI1 at each time point was normalized to the housekeeping gene ubiquitin C, and fold and percent change from baseline of normalized GLI1 expression were computed.
Percent change from baseline in GLI1 expression was stratified by type of BCC (aggressive and nonaggressive laBCC and mBCC), age, sex, number of lesions at baseline, cytology subtype, primary site of cancer, prior radiotherapy, and Eastern Cooperative Oncology Group performance status. Association between GLI1 expression and select efficacy and safety assessments was evaluated, including best overall response, TTR, and time to onset of grade ≥ 2 CK elevation.

Statistical analyses
The ITT population included all randomized patients. The subset of the ITT population with valid Oncotarget 3481 www.oncotarget.com biomarker samples comprised the biomarker population used for all biomarker analyses. Data collected up to 6 months after the last patient randomization date were included in the analysis. All statistical evaluations were exploratory since the study was not powered to assess specific hypotheses regarding change in GLI1 expression. Since inhibition of the Hedgehog pathway with sonidegib should result in GLI1 inhibition, postbaseline records with > 100% change in GLI1 levels from baseline were considered outliers and were excluded from analyses and models.
Visual assessments of strip plots determined the relationship between baseline GLI1 levels or changes from baseline in GLI1 levels and clinical efficacy outcomes. Percent change in GLI1 levels was summarized using descriptive statistics (mean, 95% CI, median, 95% distribution-free CI, SD, quartiles, and range). Longitudinal analyses of GLI1 expression were performed using a linear mixed model, including fold change in GLI1 expression from baseline as a response variable and visit, treatment group, and visit-by-treatment interaction as covariates. Median percent change from baseline and 95% CI for each dose group and time point were computed based on model-based least squares mean fold change estimates.
Cox proportional hazards models determined associations between baseline GLI1 levels or changes from baseline in GLI1 levels, and TTR and time to onset of grade ≥ 2 CK elevation. Kaplan-Meier curves and logrank test-based P-values for the difference in time-toonset curves were produced for groups defined by dose level and categorized GLI1 levels. Normalized baseline GLI1 levels were categorized as low or high based on the median (both doses combined). Percent change in GLI1 from baseline at week 9 or 17 was categorized as low (greater inhibition) or high (lesser inhibition) based on the third quartile (both doses combined).

Author contributions
Reinhard Dummer, Ralf Gutzmer, Nicholas Squittieri and John Lear critically reviewed and edited the manuscript. Li Liu did the biostatistics of data and critically reviewed and edited the manuscript.

ACKNOWLEDGMENTS
This study was sponsored and funded by Novartis and Sun Pharmaceutical Industries, Inc. Writing and editorial support for manuscript preparation were provided by Ginny Feltzin, PhD, and Zehra Gundogan, VMD, of AlphaBioCom, LLC, and funded by Sun Pharmaceutical Industries, Inc. All authors met the International Council of Medical Journal Editors criteria and received neither honoraria nor payment for authorship.