All the human NSCLC tissue samples used in the present study were obtained from the University of California, San Francisco Thoracic Oncology tissue bank, with approval from the Committee on Human Research (approval number: H8714-11647-10). The patient samples were collected between February 2011 and October 2016. There were 24 men and 13 women in the study cohort, with a mean age of 57.7±13.2 years (age range, 27–82 years). The patients only received erlotinib treatment during the study. When progression occurred, the patients were switched to other treatments such as chemotherapy. All specimens were snap-frozen in liquid nitrogen immediately following resection, and then stored at −170°C until further use. Formalin-fixed paraffin-embedded (FFPE) samples from the same patient were fixed in PBS buffer with 10% formalin for 24–48 h at room temperature and embedded in paraffin. Formalin-fixed paraffin-embedded (FFPE) samples (3 µm thick) were used for immunohistochemistry staining.

Total RNA was isolated by using the RNeasy kit (Qiagen GmbH) according to the manufacturer's protocol. Genomic DNA contamination was eliminated using DNase I, and RT was performed using 500 ng RNA and the iScript cDNA synthesis kit (Bio-Rad Laboratories, Inc.) using random hexamer according to the manufacturer's protocol. The cDNA was analyzed using RT-qPCR n a 7500 Real-Time PCR machine (Thermo Fisher Scientific, Inc.) using SYBR-Green qPCR Master Mixes (Thermo Fisher Scientific, Inc.). The following thermocycling conditions were used: Initial denaturation 96°C for 5 min, followed by 40 cycles at 96°C for 15 sec and 60°C for 1 min. Gene expression was normalized to GAPDH using the 2^−ΔΔCq method (47). The following primer sequences were used: GLI1 forward, 5′-CTCCCGAAGGACAGGTATGTAAC-3′ and GLI1 reverse, 5′-CCCTACTCTTTAGGCACTAGAGTTG-3′, GAPDH forward, 5′-ACAACAGCCTCAAGATCATCAG-3′ and GAPDH reverse, 5′-TCTTCTGGGTGGCAGTGATG-3′ (48).

The UltraVision LP Detection System HRP DAB kit (cat. no. TL-060-HD; Thermo Fisher Scientific, Inc.) was used according to the manufacturer's protocol. All steps were carried out at room temperature except for the primary antibody incubation. Briefly, FFPE slides were deparaffinized in xylene and rehydrated in a descending ethanol series from 100–70%. Heat-mediated antigen retrieval was performed by boiling the slides for 20 min in a citrate buffer (pH 6.0). Slides were quenched in UltraVision Hydrogen Peroxide Block for 10 min at room temperature, washed three times in PBS, and incubated with UltraVision Protein Block for 5 min at room temperature. After washing once with PBS, the slides were incubated with rabbit anti-human GLI1 (1:100 diluted in PBS; Santa Cruz Biotechnology; cat. no. sc-20687) overnight at 4°C. The slides were washed three times with PBS and incubated with Primary Antibody Enhancer for 10 min at room temperature, followed by an incubation with HRP Polymer for 15 min at room temperature. The slides were re-washed three times with PBS and the color was developed following incubation with 1:30 dilution of DAB in DAB Quanto Substrate for 3 min at room temperature. The slides were subsequently washed four times with distilled water and counterstained in Mayer's hematoxylin (Thermo Fisher Scientific, Inc.) for 1 min at room temperature, washed in running tap water for 10 min and mounted. Images from representative fields were obtained using an Olympus BX43 light microscope (magnification, ×200; Olympus Corporation) and examined for positive nuclear staining. IHC staining was blindly scored by two independent pathologists. IHC score was determined using staining intensity for the majority (≥50%) of the tumor cells as shown in Fig. 5A: No staining, 0; light yellow staining, 1; yellowish/brown staining, 2; strong brown staining, 3.

Cell lysates were prepared by adding 300 ml of M-PER Mammalian Protein Extraction Reagent (Thermo Fisher Scientific, Inc.) to 6-well plates. After shaking gently for 5 min at room temperature, the cell lysates were centrifuged at 14,000 × g for 5 min at 4°C. Total protein concentration was determined using a Bradford protein assay (Bio-Rad Laboratories, Inc.), 40 µg protein/lane was separated via 10% SDS-PAGE and the separated proteins were subsequently transferred onto polyvinylidene membranes (Sigma-Aldrich; Merck KGaA). The membranes were blocked with 5% non-fat dry milk for 30 min at room temperature, washed twice with TBST, and sliced right above the 50 kDa pre-stained molecular weight markers (Bio-Rad Laboratories, Inc.). The membranes were incubated with primary antibodies at room temperature for 2 h (top half: Anti-GLI1 antibody, 1:150 dilution in 5% BSA, Santa Cruz Biotechnology, sc-20687; bottom half: Anti-β-actin antibody, 1:500 dilution in 5% BSA, Sigma-Aldrich; Merck KGaA, A5441). Membranes were washed three times with TBST and subsequently incubated with secondary antibodies at room temperature for 1 h (top half: Goat anti-mouse IgG, 1:2,000 in 5% BSA, Santa Cruz Biotechnology, sc-2005; bottom half: HRP-conjugated rabbit anti-mouse IgG, 1:100,000 in 5% BSA, Sigma-Aldrich, A9044). Membranes were re-washed three times with TBST and incubated with Immobilon ECL Ultra Western HRP Substrate (Sigma-Aldrich; Merck KGaA) for 5 min at room temperature. The substrate solution was drained, and the membranes were exposed to X-ray film.

The following human lung cancer cell lines were purchased from American Type Culture Collection: A549 (cat. no. CCL-185), A427 (cat. no. HTB-53), H1299 (cat. no. CRL-5803), H2170 (cat. no. CRL-5928), H1703 (cat. no. CRL-5889), H522 (cat. no. CRL-5810), H838 (cat. no. CRL-5844), H322 (cat. no. CRL-5806), H1650 (cat. no. CRL-5883), H1975 (cat. no. CRL-5908), H820 (cat. no. HTB-181), H441 (cat. no. HTB-174), H460 (cat. no. HTB-177), H1666 (cat. no. CRL-5885), HCC2935 (cat. no. CRL-2869). All the cell lines were cultured in RPMI-1640 supplemented with 10% fetal bovine serum, penicillin (100 IU/ml) and streptomycin (100 µg/ml) (all purchased from Thermo Fisher Scientific, Inc.) at 37°C in a humidified incubator with 5% CO₂.

For each well of the 6-well plates, 3×10⁵ cells were plated in fresh media without antibiotics for 24 h prior to transfection. Cells were transfected with either 2 mg pcDNA3.1 vector containing GLI1 cDNA or pcDNA3.1 vector control (Thermo Fisher Scientific, Inc.) using Lipofectamine^® 2000 transfection reagent (Thermo Fisher Scientific, Inc.) according to the manufacturer's protocol. For RNA interference, cells in 6-well plates were transfected with 100 pmole of synthesized GLI1 small interfering (si)RNA (siRNA-1, Assay ID 107670; siRNA-2, Assay ID 115641, Thermo Fisher Scientific, Inc.) or scrambled siRNA using Lipofectamine^® 2000 transfection reagent (Thermo Fisher Scientific, Inc.) according to the manufacturer's protocol. After the transfection, the cells were cultured for 48 h before further analysis with drug treatment.

Cells were seeded in 96-well plates at a density of 500–1,000 cells/well, and the medium was changed daily. Logarithmically growing cells were treated with increasing doses (0.01 µM to 1 mM) of erlotinib and/or with GLI inhibitor (0.01-10.00 µM), or DMSO control for 3 days. Cells were subsequently assessed for viability using CellTiter-Glo Luminescent Cell Viability Assay reagent (Promega Corporation) according to manufacturer's instructions. Luminescence was measured using a GloMax-96 Microplate Luminometer (Promega Corporation), and the percent of cell survival was calculated with the DMSO treated cells set as 100%. GraphPad Prism v6.0 (GraphPad Software, Inc.) was used to generate dose-response curves and IC₅₀ values.

A549 or H2170 cells were treated with 0.01–10 µM of erlotinib alone, GLI inhibitor alone, or the 1:1 combination of the two drugs for 72 h, and assayed for cell viability. The CalcuSyn software version 2.0 (Biosoft) was used to calculate the CI to identify synergistic, additive, or antagonistic drug interactions for the combination treatment as indicated in Fig. 4, where CI <1 indicates synergism, CI, 1 indicates additive effects, and CI >1 indicates antagonism. Synergism was further divided into moderate synergism (CI, 0.7–0.9), synergism (CI, 0.3–0.7) and strong synergism (CI, 0.1–0.3).

The comparisons of IC₅₀ values or cell viabilities under different experimental conditions were analyzed using the unpaired Student's t test for two groups, or by one-way ANOVA followed by Tukey's post hoc test for multiple groups. The data from three experiments were averaged and plotted as the mean ± standard deviation. The associations between GLI1 mRNA levels and IC₅₀ for erlotinib were assessed using Pearson's correlation coefficient. The Cox proportional hazards model was used to evaluate the risks in progression-free survival (PFS) for patients with different GLI1 IHC scores. The Kaplan-Meier plot was used to illustrate the differences in PFS between the low-risk group (low GLI1 IHC staining) and high-risk group (high GLI1 IHC staining). The log-rank test was used to evaluate the differences in PFS between dichotomous groups defined by patient characteristics. The Fisher's exact test was used to analyze the association between GLI expression levels and patient characteristics. P≤0.05 was considered to indicate a statistically significant difference.

To investigate the association between gene mutations and erlotinib sensitivity, NSCLC cell lines with different EGFR and KRAS mutations as reported by COSMIC database (49,50) were examined (Fig. 1). The HCC2935 and H1650 cell lines contain the sensitizing exon 19 deletion in EGFR, while the H820 and H1975 cell lines contained both sensitizing EGFR mutation (E19del and L858R, respectively) and the T790M resistant mutation. A total of 4 cell lines with mutations in the downstream signaling protein, RAS were also investigated: A549 with KRAS G12S, H1299 with NRAS Q61K, A427 with KRAS G12D, and H441 with KRAS G12V. It was hypothesized that the cell lines containing sensitizing EGFR mutations would have a lower IC₅₀ compared with that with EGFR T790M or RAS activating mutations.

Unexpectedly, although the cell lines with a double mutation in EGFR or those with RAS mutations had on average higher IC₅₀ values, there was a wide range of variability in the individual cell lines (Fig. 1). Both the H820 and H1975 cell lines contained the EGFR T790M mutation, as well as either the sensitizing E19del or L858R mutation; however, there was a significant difference in the IC₅₀ values (P<0.01). In addition, for the three cell lines containing KRAS mutations at the same codon, the IC₅₀ of H441 (KRAS G12V) was significantly lower compared with that in A549 (KRAS G12S; P<0.001) and A427 (KRAS G12D; P<0.001). Furthermore, the H1650 cell line contained the sensitizing EGFR E19del mutation; however, its IC₅₀ was not lower compared with that in the H820 (E19del and T790M) and H441 (KRAS G12V) cell lines. These results indicated that the analysis of EGFR and RAS hotspot mutations is not sufficient to predict TKI sensitivity in cell lines, and it is likely that additional genetic or epigenetic modifications modulate their response to TKI.

During the investigation into GLI1 function, it was found in a preliminary study that the A549 and A427 cell lines had high mRNA expression levels GLI1, whereas it was hardly detectable in H441 cells (data not shown). To investigate whether the mRNA expression level of GLI1 was correlated with erlotinib sensitivity, 15 NSCLC cell lines with a variety of mutation backgrounds (Table I) were examined. Among them, 4 cell lines contained EGFR mutations, 7 contained an activating mutation in the RAS family, and 4 did not contain either of those mutations (51–57).

As shown in Fig. 2, a wide range of IC₅₀ values were observed among the different cell lines. The highest IC₅₀ values (H522 and H1703) were ~25 fold higher compared with that for the lowest one (HCC2935). The Pearson's correlation analysis showed that the mRNA expression levels of GLI1 was significantly correlated with IC₅₀ for erlotinib (Pearson's correlation co-efficient r=0.747; P<0.01).

To investigate the function of GLI1 expression in erlotinib sensitivity, A427 cells, which is one of the cell lines with a high mRNA expression level of GLI1, were transfected with GLI1 siRNA-1, GLI1 siRNA-2, or scrambled control siRNA, respectively for 2 days, followed by a treatment with 25 µM erlotinib for 3 days. A total of two GLI1 siRNAs were used to ensure that the observed phenomena were due the silencing of GLI1 instead of the off-target effect a certain siRNA. As shown in Fig. 3A, the downregulation of GLI1 by siRNA sensitized A427 cells to erlotinib treatment (P<0.03 and P<0.008 for siRNA-1 and siRNA-2, respectively).

Subsequently, the effect of GLI1 overexpression on erlotinib sensitivity was assessed. H1299 cells, which has a moderate level of GLI1 mRNA expression as shown in Fig. 2, were transfected with GLI1 expression vector or a control vector for 2 days, followed by treatment with 30 mM erlotinib for 3 days. As shown in Fig. 3B, upon GLI1 overexpression, the cells exhibited increased viability in the presence of 30 µM erlotinib.

Taken together, the results suggested that GLI1 modulates erlotinib sensitivity in lung cancer cells, and that high GLI1 mRNA levels may be a critical and independent mechanism to confer resistance to erlotinib in lung cancer.

Subsequently, two NSCLC cell lines, H2170 and A549, were treated with both a small molecule GLI inhibitor (58) and different concentrations of erlotinib for 72 h. The combination treatment synergistically suppressed proliferation of both the H2170 and A549 cell lines (Fig. 4). CalcuSyn analysis showed that the CI for the two compounds was 0.173 and 0.231 for H2170 and A549, respectively, suggesting a strong synergism of the two drugs in both cell lines.

To investigate the clinical relevance of GLI1 as a predictive biomarker for EGFR-TKI therapy, a retrospective study was conducted in a cohort of 37 lung cancer patients that received erlotinib treatment. More than half of the patients progressed within 6 months of erlotinib treatment. Only one patient was progression-free for 36 months. The median follow-up time was 4 months, with a range of 0.13 to 36 months. FFPE tissue specimens were collected in the surgeries before erlotinib treatment and were analyzed by IHC staining. The slides were scored according to the standard protocol (Fig. 5A, with panel a-d showing representative staining of score 0, 1, 2 and 3, respectively). The mean IHC score (± standard deviation) for the cohort was 0.95 (±0.76). By using a Cox proportional hazards model, it was calculated that for every 1-point increase in the IHC score, patients were 2.1 (95% confidence interval, 1.30–3.32) times more likely to progress or die.

Subsequently, each case was asigned to either GLI1-low (IHC score ≤1) or GLI-high (IHC score >1) groups. Kaplan-Meier curves for progression-free survival (PFS) of GLI1-low (28 cases) and GLI1-high (9 cases) groups of patients with lung cancer receiving erlotinib treatment are presented in Fig. 5B, which revealed a significant difference in PFS by the log-rank test (P=0.0021). The data indicated that PFS in patients receiving erlotinib treatment was significantly improved for those with a low GLI1 expression compared with those with a high GLI1 expression. No statistically significant differences were observed between the dichotomous groups defined by other clinicopathological characteristics such as sex, smoker, age or tumor stage (Table II). Notably, more GLI1-high cases were observed in the squamous cell carcinoma subgroup in comparison with the adenocarcinoma subgroup (P=0.034), although the two pathological subgroups did not show a significant difference in PFS. When the PFS of the adenocarcinomas (25 samples) was analyzed, the association between high GLI1 expression and poor PFS was also observed (P=0.0020). The squamous cell carcinomas (9 samples) did not show a significant difference in PFS between the GLI1-high and GLI1-low groups, probably due to the small number of cases. In summary, the result suggests that GLI1 may be a predictive biomarker for patients treated with EGFR-TKI.