Activation of the JAK/STAT Pathway leads to BRAF inhibitor Resistance in BRAF^V600E Positive Thyroid Carcinoma

Abstract

A subset of thyroid cancers, recurrent differentiated thyroid cancers and anaplastic thyroid cancer (ATC), are difficult to treat by thyroidectomy and systemic therapy. A common mutation in thyroid cancer, BRAF^V600E, has targetable treatment options; however, the results have been disappointing in thyroid cancers compared to BRAF^V600E melanoma, as thyroid cancers quickly become resistant to BRAF^V600E inhibitor (BRAFi). Here, we studied the molecular pathway that is induced in BRAF^V600E thyroid cancer cells and patient derived tumor samples in response to BRAFi, vemurafenib, using RNA-sequencing and molecular analysis. Both inducible response to BRAFi and acquired BRAFi resistance in BRAF^V600E thyroid cancer cells showed significant activation of the JAK/STAT pathway. Functional analyses revealed that the combination of BRAFi and inhibitors of JAK/STAT pathway controlled BRAF^V600E thyroid cancer cell growth. The Cancer Genome Atlas data analysis demonstrated that potent activation of the JAK/STAT signaling was associated with shorter recurrence rate in differentiated thyroid cancer patients. Analysis of tumor RNA expression in poorly differentiated thyroid cancer and ATC patients also support that enhanced activity of JAK/STAT signaling pathway is correlated with worse prognosis. Our study demonstrates that JAK/STAT pathway is activated as BRAF^V600E thyroid cancer cells develop resistance to BRAFi and that this pathway is a potential target for anticancer activity and to overcome drug resistance that commonly develops to treatment with BRAFi in thyroid cancer.

Introduction:

The majority of thyroid cancers originate from thyroid follicular epithelial cells (1), and as tumors acquire a rising mutational burden and epigenetic changes, they can progress from well-differentiated thyroid cancer (WDTC) to poorly-differentiated thyroid cancer (PDTC) and, finally, to anaplastic thyroid cancer (ATC) (2,3). The most prevent subtype of WDTC is papillary thyroid cancer (PTC), comprising approximately 80% of the cases(4). The second most frequently diagnosed thyroid cancer type is follicular thyroid cancer (FTC), comprising about 10–15% of all thyroid cancer cases. The mortality of WDTC is relatively low as most tumors are either indolent or treatable by surgery and radioactive iodine treatment. However, the relative 5-year survival rate drops dramatically to about 50% and 70% for FTC and PTC patients who presented metastasis, respectively. Moreover, patients with PDTC and ATC have disease-specific mortality rates that range from 40 – 100%, with most refractory to iodine therapy and generally exhibiting poor response to systemic therapy (2,5).

BRAF^V600E mutations occur in approximately 50% of WDTCs, a similar rate to melanoma, 40% or more of ATCs (3,6,7), and with much lower frequency in other solid tumors. BRAF^V600E mutations render the BRAF protein, a member of the RAS family of proteins, to be constitutively active and induce oncogenic changes in the cell by activating downstream mitogen-activated protein kinase (MAPK) pathways. Chemical inhibitors with considerable activity against BRAF^V600E, including vemurafenib, demonstrated efficacy in melanoma by prolonging progression-free survival (8–10) and overall survival (8,10). However, targeted therapies for BRAF mutant thyroid cancers have shown limited response in clinical trials (11,12).

Although not all melanoma patients respond to BRAF inhibitor (BRAFi), with objective response rates of 48–53% (10,13), the objective response rate to BRAFi in thyroid cancer is lower, with partial response rate of 29–38.5% (14,15). A number of studies have shown that intrinsic resistance to BRAFi in thyroid cancer is mediated by feedback reactivation of the MAPK pathway (16), as well as alternative activation of other signaling pathways, such as phosphoinositide-3-kinase/Akt/mammalian target of rapamycin (PI3K/Akt/mTOR) as seen in melanoma (17) and thyroid cancer (18), or increased HER2/HER3 signaling through HER3 transcription activation (19). Tumors from a subset of patients who initially responded to BRAFi and later relapsed after developing drug resistance, were found to acquire secondary RAS mutations, including KRAS^G12V, NRAS^Q61K, and NRAS^G13D (20,21). These findings demonstrate the challenges with BRAFi monotherapy. Recently, the FDA approved combination treatments with inhibitors of BRAF and MEK for metastatic, BRAF-mutant ATC (22). In the phase 2 clinical trial, 69% of the patients responded to the combination therapy of dabrafenib, a BRAFi, and trametinib, a MEK inhibitor (MEKi), but long-term outcomes are still undetermined. Potential problems of this dual therapy include the development of resistance to kinase inhibitors, commonly seen in solid tumors (23), as both BRAFi and MEKi act largely on the same pathway. Studies have shown that melanomas treated with combination of BRAFi and MEKi can still acquire resistance by activating alternative pathways (24).

Here, we aimed to uncover molecular pathways that underlie resistance to BRAFi in thyroid cancer. We created BRAF^V600E thyroid cancer cell lines that were more resistant to vemurafenib than parental thyroid cancer cell lines and identified the main downstream pathway perturbed by BRAFi using transcriptome analysis. The transcriptional pathway modulated by the BRAFi also was closely associated with the state of aggressiveness in thyroid cancers when analyzed using publicly available data, including PDTC and ATC in cBioPortal and PTC in the Cancer Genome Atlas (TCGA) clinical and molecular data. Our functional experiments suggest therapeutic leads that may demonstrate enhanced susceptibility to BRAFi in combination. Taken together, the results of our work uncover a common pathway that is involved in oncogenic processes and BRAFi resistance in thyroid cancer.

Materials and Methods:

Sample collection and regulatory approval

Studies were conducted in accordance with the Declaration of Helsinki. Patients with a diagnosis of PTC were consented pre-operatively for study enrollment between the years 2018–2019 under Weill Cornell Medicine Institutional Review Board approval. Clinical characteristics of the patients and tumors were collected via chart review. Fresh tumor specimens were taken by experienced pathologists. The BRAF mutational status was confirmed independently by next-generation sequencing as standard of care at Weill Cornell Medicine as previously described (25).

Cell Culture

The 8505C cell line was obtained from DSMZ (#ACC 219), and BRAF^V600E mutation was confirmed by Sanger sequencing (Supplementary Figure S1). The SW1736 and WRO (also known as WRO82–1) cell lines were generously provided by Dr. Fagin’s laboratory at MSKCC and Dr. Juillard’s laboratory at UCLA, respectively (26). It has been known in the field that there are 2 distinct WRO cell lines: BRAF mutant and wild type (27,28). The WRO cell line used in our study had a BRAF^V600E mutation (Supplementary Figure S1) and was authenticated by the short tandem repeat anlysis (ATCC) as identical to the WRO cell lines, which were reported in 2 publications (Supplementary Figure S2) (27,29). The high levels of thyroid-specific transcription factors, PAX8, TTF1, and TTF2, in the WRO cell line further support that the cell line was of human thyroid cancer origin (Supplementary Figure S3) (27,29). The 8505C cell line exhibited minimal expression of PAX8, which is a characteristic of dedifferentiated ATC. Two human BRAF^V600E mutant thyroid carcinoma cell lines (8505C, WRO), and their respective vemurafenib-resistant clones (8505C-Res-1, WRO-Res-6) were grown at 37°C under 5% CO₂. All cell lines were grown in RPMI-1640 media supplemented with L-glutamine, 10% fetal bovine serum (FBS) and antibiotic and antimycotic solution (ThermoFisher). For vemurafenib-resistant clones additionally contained 1 μM vemurafenib (Selleck Chemicals, #S1267). Cell lines were confirmed negative for Mycoplasma prior to the study with the kit (Lonza, #LT07–118,) following the protocol provided by the manufacturer. Cell lines were maintained from cryopreserved stocks made at low passage numbers and all cells used in experiments were of a passage number <30. Fresh tumor tissues were cut into 1–2 mm pieces using surgical blade and plated in 6 well dish in RPMI-1640 media supplemented with 10% FBS and antibiotic solution.

RNA-sequencing (RNA-seq) analysis

Duplicates of the following RNA samples were processed for RNA-sequencing analysis: Parental 8505C treated with 2 μM vemurafenib or 0.1% dimethylsulfoxide (DMSO), 8505C-Res-1 treated with 2 μM vemurafenib or 0.1% DMSO, Parental WRO treated with 2 μM vemurafenib or 0.1% DMSO, and WRO-Res-6 treated with 2 μM vemurafenib or 0.1% DMSO. All treatments were for 24 hours. RNA was extracted using the RNeasy Mini Kit (Qiagen), and was submitted to the Genomics Research Core facility (Weill Cornell Medicine) for quality control, library constructing and sequencing. Total RNA integrity and concentration were checked using the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA) and the NanoDrop system (ThermoFisher). Using Illumina TruSeq RNA Sample Library Preparation v2 kit (Illumina, San Diego, CA), The poly-A containing messenger RNA was purified using oligo-dT attached magnetic beads. Purified mRNA were fragmented into small pieces using divalent cations under elevated temperature. The cleaved RNA fragments were copied into first strand cDNA using reverse transcriptase and random primers. Second strand cDNA synthesis followed, using DNA Polymerase I and RNase H. The cDNA fragments then went through an end repair process, the addition of a single ‘A’ base, and then ligation of the adapters. The products were then purified and enriched with PCR to create the final cDNA library. The normalized cDNA libraries were pooled and sequenced on Illumina HiSeq2500/4000 sequencers with single-end 50 cycles.

Sequencing reads were trimmed to remove low-quality bases and adapter sequences using cutadapt v1.9.1 (30) and were mapped to the human GRCh37 reference genome using STAR v2.5.2b (31). The abundances of transcript were estimated in Fragments Per Kilobase of exon per Million mapped reads (FPKM) using Cufflinks v2.1.1(32). After log₂ transformation of the FPKM values, differentially expressed genes were identified using the limma packaged in R. A cutoff of adjusted P < 0.05 and |log₂(FC)|> 1 was used to identify differentially regulated genes. Only protein-coding genes were used for selection.

Cell Viability

Vemurafenib, Fludarabine (STAT1 activation inhibitor) (Selleck Chemicals, #S1491) and AZD1480 (JAK2 inhibitor) (Selleck Chemicals, #S2162) were dissolved in DMSO, and stock solutions were stored at −80°C. 8505C, WRO, and their respective vemurafenib-resistant clonal cells were seeded in triplicate at a density of 10,000 cells per well, within 200 μl of phenol-free RPMI media (ThermoFisher) per well in 96-well flat-bottomed culture dishes. Cells were exposed to experimental drug regimens at the indicated concentrations after 24 hours and then incubated for 72 hours. Cell viability and apoptisis were assessed using the Vybrant ^® MTT Cell Proliferation Assay Kit (ThermoFisher #V13154) and Annexin V apoptosis detection kit (ThermoFisher #88–8007-72) according to manufacturer’s specifications. Data was collected using Tecan Infinite M1000 Pro Microplate reader for viability assessment.

Flow cytometry

Trypsinized 8505C and WRO cells were washed once with phosphate-buffered saline (PBS) and stained with the anti-human ICAM1-APC/Fire 750 (BioLegend, #353122), anti-human HLA-A2-APC (BioLegend, #343307), or anti-human IFNgR1-PE (Biolegend, #308703) antibodies to measure protein expression on each cell line. Cells were exposed to vemurafenib or IFNγ (Peprotech, #300–02) alone for 72 hours. Live cells were gated based on calcein blue, AM staining (ThermoFisher, #C1429). All antibody incubations for flow cytometry were performed for 20 minutes at 4°C in PBS supplemented with 2% FBS. For apopotosis assay, APC-cojugated Annevin V and propidium iodide (PI) were used for staining. Positively stained cells were determined by gating based on fluorescence minus one (FMO) control. Data was acquired using Gallios flow cytometry (Beckman Coulter) and analyzed using FlowJo software (BD).

CRISPR-Cas9 gene editing

IFNGR1- and STAT1-knockout cell lines were generated using the Alt-R CRISPR-Cas9 System (Integrated DNA Technologies, Inc.) according to the manufacturer’s instructions. crRNA guide sequences (AA and AB) against two exons of each target were used: IFNGR1 crRNA-AA: 5’-rArCrA rUrGrA rArCrC rCrUrA rUrCrG; IFNGR1 crRNA-AB: 5’-rCrCrG rArArA rCrUrA rCrCrU rGrUrU; STAT1 rRNA-AA: 5’-rUrGrU rGrArU rArGrG rGrUrC rArUrG; STAT1 rRNA-AB: 5’-rCrCrA rCrUrA rGrUrU rCrArU rCrArU. crRNA and tracrRNA oligos were annealed in equimolar concentrations (44 μM) by heating at 95°C for 5 minutes, followed by gradual cooling to room temperature. Each guide RNA duplex (final concentration of 22 μM) was then incubated with Cas9 nuclease (final concentration of 18 μM) at room temperature for 15 minutes to form the ribonucleoprotein (RNP) complex. 8505C cells were mixed individually with RNP-AA and RNP-AB complexes into a 10 μl Neon tip and electroporated (1450 V/10 ms/3 pulses) using the Neon^™ Transfection System (Invitrogen). For each electroporation, 22 pmol gRNA-AA, 22 pmol gRNA-AB and 18 pmol Cas9 nuclease were used per 5 × 10⁴ target cells. After electroporation, cells were immediately transferred to a 6-well plate containing 2 mL of pre-warmed culture medium and incubated in a humidified 37°C, 5% CO₂ incubator.

Quantitative PCR

Cells were digested with trypsin EDTA, resuspended, washed with PBS, and then lysed using the RNeasy Mini Kit (Qiagen). The volume of each RNA sample containing 1 μg of RNA was determined via NanoDrop (ThermoFisher) and this 1μg was then diluted into 10 μL of RNase free water and reverse transcribed in a 20 μL reaction using standard reagents and procedures (ThermoFisher, #18064014). For real-time quantitative PCR (qPCR), these cDNA samples were amplified with TaqMan^™ probes (ThermoFisher) for β-actin (Hs01060665), GAPDH (Hs99999905), and IRF-1 (Hs00971965) in quadruplicate using Roche LightCycler 480 II. Sybr green-based qRT-PCR assay was used with STAT1, IFNGR1, β-actin primers. Samples were normalized to housekeeping genes, β-actin or GAPDH, to calculate fold-change using the ΔΔCt method (33).

Immunoblotting

Cells were lysed in ice-cold NP-40 lysis buffer (Santa Cruz Biotechnology, #sc-24948) with phosphatase and protease inhibitors added, and protein concentrations were measured using the BCA Protein Assay Kit (ThermoFisher, #23227). Lysates were run on 7.5% gradient SDS-polyacrylamide gels and transferred using the methanol-based wet-transfer technique onto 0.45 μm pore-size nitrocellulose membranes (Bio-Rad Laboratories). The membranes were blocked with 5% phosphoBLOCKER (Cell Biolabs, # AKR-103) and incubated overnight at 4°C for primary antibody binding and with 5% nonfat dry milk reconstituted in Tris-Buffered Saline (TBS) for one hour at room temperature for secondary antibody binding. Blots were developed on standard radiographic film using either Pierce ECL, SuperSignal^™ West Pico PLUS, or West Femto Chemiluminescent Substrates (ThermoFisher). All antibodies were obtained from Cell Signaling Technology. Primary antibodies against phosphorylated p44/42 MAPK (ERK1/2)(Thr202/Tyr204) (#9101), phosphorylated STAT1 (Tyr701) (#9167), phosphorylated STAT3 (Tyr705) (#9131), p44/41 MAPK (ERK1/2) (#9102), STAT1(#9172), and STAT3 (#9132) were used at 1:1000 dilutions. Secondary anti-rabbit antibody (#7074) or anti-mouse antibody (#7076) used at 1:5000 dilution for each primary antibody as appropriate.

Database Query

We queried publicly available data for different subtypes of thyroid cancers using cBioPortal (3) and the TCGA (34). First, we used probes for interferon regulatory factor 1 (IRF1) and signal transducer and activator of transcription 1 (STAT1) expression in PDTC and ATC (3). We compared overall survival for the top 50% and bottom 50% of expression by generating Kaplan Meier curves and comparing overall survival using the log-rank test. Next, we used the TCGA RNA expression data to assess mRNA expression of IRF1 and signal transducer and STAT1 in PTC (34). The mRNA Z-score that was computed by the TCGA is the relative expression of an individual gene in a tumor sample to the gene’s expression distribution in a reference population or, when available, normal adjacent tissue. Clinicopathological data associated with each PTC sample were downloaded from the data portal (http://cancergenome.nih.gov). We evaluated the relationship of IRF1 and STAT1 in PTC patients as well as for differences in overall survival in patients with overexpression of IRF1 and STAT1. RNA-sequencing data based on cell lines were analyzed using the Qiagen Ingenuity Pathway Analysis (IPA) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database algorithms.

Statistical Analysis

Comparisons between different groups were analyzed using Student’s unpaired t-test or one-way ANOVA test, with results expressed as the mean ± standard error of the mean. Disease-free and overall survival analyses were generated using the Kaplan-Meier method, with curves compared using log-rank tests. Pearson’s correlation coefficient was determined using bivariate analysis. A two-tailed P value of less than 0.05 was considered statistically significant. Analysis was completed on Prism 8.0 (Graph Pad).

Data Availability Statement

The data generated in this study are available within the article and its supplementrary data files. The raw RNA-sequencing and processed files are deposited in GEO with accession number GSE178267. Data analyzed from the TCGA (https://www.cancer.gov/tcga) and cBioPortal (http://www.cbioportal.org) are publicly available.

Results:

To identify genes and pathways regulated in response to vemurafenib (Vem), a BRAFi, in BRAF-mutant thyroid cancer cells, we analyzed genome-wide changes in RNA expression by RNA-sequencing (RNA-seq) that occurred after 24-hour treatment with 2 μM of Vem. Well-established thyroid cancer cell lines, 8505C and WRO, were verified for the BRAF^V600E mutation status by PCR amplification and Sanger sequencing. 8505C and WRO lines retained BRAF^V600E mutation (Supplementary Fig. S1), as previously reported (27,29). Both 8505C and WRO cells showed relative resistance to Vem with half maximal inhibitory concentrations (IC50) of 3.3 μM and 5.3 μM, respectively (Supplementary Fig. S4). Short duration of Vem treatment induced RNA expression changes in 2,119 genes with at least 2-fold Fragments Per Kilobase of exon per Million mapped reads (FPKM) difference in 8505C cells compared to a vehicle control (DMSO) treated cells as visualized by volcano plot (Fig. 1A and Supplementary Table 1). WRO cells upon Vem treatment resulted in 2,325 dysregulated genes relative to DMSO-treated WRO cells (Fig. 1C and Supplementary Table 2).

Figure 1. BRAF inhibitor, vemurafenib, activate JAK/STAT-signaling pathway in BRAF^V600E thyroid cancer cells.

(A, D) Differential analysis of gene activity in 8505C (A) and WRO (D) cells after Vem treatment (2 μM, 24 hours) displayed as volcano plots. Significantly differentially regulated “HALLMARK_INTERFERON_GAMMA_RESPONSE” genes (239 genes compiled from gene set Molecular Signature Database M5913 and M15615) are marked on the plots, with red dots and blue dots indicating upregulation and downregulation after Vem treatmemt, respectively. Each condition had 2 biological replicates. (B, E) KEGG pathway analysis was used to predict the pathway that was significantly altered after Vem treatment (2 μM, 24 hours) compared to vehicle control-treated (DMSO, 0.1%) 8505C (B) and WRO (E) cells (Absolute fold change > 1.5 was used in the analysis). The FPKM and log2 fold change values were calculated and statistically analyzed using RNA-sequencing data. Gene ratio on the x-axis indicates the number of genes annotated to a pathway within the list of differential genes. The adjusted P-value is shown as a heatmap scale bar. (C, F) Ingenuity pathway analysis (IPA) with differentially expressed genes predicted upstream activators and inhibitors in 8505C (C), WRO (F) cells in response to Vem treatment. Altered genes with absolute fold change > 1.5, P value <0.05 were used. (G, I) Flow cytometry histogram plots showing HLA-A2 or ICAM-1 expression in 8505C (G) and WRO (I) cells after Vem (2 μM) or IFNγ (1×10⁴ pg/ml) treatment for 72 hours. Fluorescence minus one (FMO) control was used to gate (indicated with solid lines) for positively stained cells. The percentages of ICAM-1-positive cells are presented in the histogram plots. (H, J) Quantitation of mean fluorescence intensity (MFI) of HLA-A2 or ICAM-1 expression in 8505C (H) and WRO (J) cells after Vem (0.5 or 2 μM, 72 hours) treatment were analyzed relative to DMSO (D)-treated control cells. Data represents mean ± sem (n = 5–8 per group; ***, P < 0.001, ****, P < 0.0001 by Student’s t-test).

Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was performed to identify specific pathways that were significantly altered in BRAF-mutant thyroid cancer cells after Vem treatment. Modulated pathways included a wide range of genes/pathways with known function in adhesion (cell adhesion, focal adhesion), PI3K-Akt signaling pathway, ligand-receptor interaction (ECM, axon guidance), cell cycle, and inflammatory responses (inflammatory mediator regulation of thermo-transient receptor potential (TRP) channels, TGF-β signaling pathway) (Fig. 1B and E). To further refine the upstream activators that invoke these gene alterations, we utilized Ingenuity Pathway Analysis (IPA) and identified the inflammatory cytokine IFNγ as one of the strongest activators in both cell lines (Fig. 1C and F). Many IFNγ-downstream target genes were differentially regulated, with more genes showing upregulation after Vem treatment (Fig. 1A and D). We found 31 upregulated and 17 downregulated IFNγ-response genes in 8505C. In WRO cells, 51 and 15 IFNγ-response genes were upregulated and downregulated, respectively. In response to Vem treatment, RNA expression of genes directly associated with MHC class I antigen processing and presentation (HLA, β2M, TAP1) and inflammatory mediators (IL6, IRF1, and ICAM-1) were significantly elevated in both 8505C and WRO cells. Next, we quantified the surface expression level changes of human leukocyte antigen serotype A*02 (HLA-A2) and intercellular adhesion molecule-1 (ICAM-1) by flow cytometry. In 8505C cells, Vem treatment at 0.5 μM and 2 μM increased the expression of HLA-A2 by approximately 2-fold in mean fluorescence intensity (MFI) (Fig. 1G and H). Exposure to IFNγ also induced HLA-A2 expression significantly, verifying that HLA expression is regulatable in BRAF-mutant thyroid cancer cells similar to the prior observations in PTC (35) and melanoma cells (36). The ICAM-1 MFI was elevated 2-fold, and the percentages of ICAM-1-positive cells were increased from 72% to 98%. WRO cells exhibited a similar pattern, with both upregulation of HLA-A2 and ICAM-1 levels after Vem treatment (Fig. 1I and J). Previously, we have shown that ICAM-1 expression was inducible in both cell lines by IFNγ (37). These findings support that the IFNγ signaling pathway was significantly activated in BRAF-mutant thyroid cancer cell lines in response to Vem treatment.

Next, we established BRAF-mutant thyroid cancer cell clones with BRAFi resistance that were adapted to long-term exposure of high dose of Vem to select for BRAFi resistant clones. We selected resistant clones in vitro to mimic advanced thyroid tumors with acquired resistance to molecular targeted therapy after continuous treatment to Vem. 8505C and WRO cells were treated with Vem at a dose close to their IC50s for a week, and the drug concentration was increased approximately 2-fold weekly until the final concentration reached 20 μM, approximately a 5-fold increase of IC50s (Fig. 2A). Individual colonies were selected and further expanded to establish resistant clones of 8505C and WRO. Among about 10 clones selected from each cell line, we chose three to four clones that continued to expand well in culture for further analysis (Fig. 2B, C, and Supplementary Fig. S5). Phosphorylation levels of the ERK protein, downstream signaling molecule of BRAF, were compared between parental and resistant clones to estimate sensitivity to Vem. Both 8505C and WRO parental cells exhibited marked reduction of phosphorylated ERK (p-ERK) after 2 μM of Vem treatment for 72 hours (Fig. 2B). On the contrary, Vem-resistant 8505C clones exhibited higher levels of p-ERK with less significant changes in the total amount of ERK (t-ERK) upon Vem treatment. Compared to Vem-resistant 8505C clones, selected Vem-resistant WRO clones were more responsive to Vem. Two clones with the least sensitivity to Vem, Vem-resistant 8505C clone 1 (Res-1) and WRO clone 6 (Res-6), were selected for downstream analysis.

Figure 2. Vemurafenib-resistant BRAF^V600E thyroid cancer clones exhibit higher p-ERK level upon vemurafenib treatment.

(A) Schematics of generation of clonally-selected Vem-resistant 8505C and WRO clones. (B) Representative western blot images show the level of p-ERK, t-ERK, and GAPDH in resistant 8505C and WRO clones relative to parental cells after Vem treatment (2μM, 72hr). Each band intensity in immunoblots were quantified using ImageJ software. p-ERK to t-ERK levels were normalized to the GAPDH expression in each experimental condition (The normalized p-ERK level in untreated, parental cells is set to 1). The experiment was repeated at least three times. (C) MTT analysis shows that 8505C-Res-1 clone is more resistant to Vem than parental 8505C. (n = 9 per group). (D) Representative western blot images of Vem-resistant Res-1 clone vs. parental 8505C upon Vem treatment (0.5 or 2 μM; 24 or 72 hr). (E) Cell viability was assessed by MTT in Vem-resistant Res-6 clone and parental WRO cells after different concentrations of Vem treatment. (F) Representative western blot images of Vem-resistant Res-6 clone vs. parental WRO upon Vem treatment (0.5 or 2 μM; 24 or 72 hr).

Treatment of these resistant clones of 8505C and WRO with variable concentration of Vem supported that both 8505C-Res-1 and WRO-Res-6 had higher IC50 against Vem (Fig. 2C and 2E). Compared to parental cells, there was an approximately 4–7-fold increase in the IC50 with 8505C-Res-1 and WRO-Res-6 cells estimated as 23.3 μM and 19.6 μM, respectively. Other Vem-resistant clones that were not included for downstream analysis also exhibited IC50 over 10 μM. 8505C-Res-1 clone was more refractory to Vem-induced apoptosis than parental 8505C, which contributes to the difference in overall cell survival (Supplementary Fig. S6). The resistant clones, 8505C-Res-1 and WRO-Res-6 cells, were accompanied by more persistent p-ERK levels after Vem treatment with two different concentrations for 24 and 72 hours relative to parental cells (Fig. 2D and 2F).

Two genetically distinct BRAF-mutant thyroid cancer lines with a shared BRAF^V600E mutation and enhanced BRAFi resistance provided a valuable resource to find a common dysregulated pathway that may contribute to the resistance to BRAFi. Differential gene expression analysis showed that expression level of 1,625 genes were significantly different between 8505C-Res-1 and parental 8505C cells (Fig. 3A and Supplementary Table 3). KEGG analysis revealed that 8505C-Res-1 retained activated gene expression signatures that were similarly observed in transiently Vem-treated parental 8505C cells (Fig. 3B). In 8505C cells, PI3K-Akt signaling pathway and neuroactive ligand-receptor interaction were the most significant pathways altered in response to transient treatment with BRAFi or acquired BRAFi resistance. Comparison of regulated transcripts by acute Vem treatment and Vem-resistant clone showed that many differentially expressed genes were overlapping (Fig. 3C). In addition, IFNγ, an activator predicted in 8505C upon Vem treatment was also predicted to be the top activator in 8505C-Res-1 cells, supporting a shared pathway downstream of acute response and chronic resistance to vemurafenib (Supplementary Fig. S7).

Figure 3. Vemurafenib-resistant BRAF^V600E thyroid cancer clones retain altered JAK/STAT signaling signatures.

(A, D) Volcano plots show change of gene expression comparing Vem-resistant clones versus parental 8505C (A) and WRO (D) cells. Log2 fold change of FPKM is plotted on the x-axis, and −log10 of adjusted P-value is shown on the y-axis. Upregulated and downregulated IFNγ target genes are marked by red and blue dots, respectively, on the volcano plots. (B, E) KEGG pathway analysis was performed to predict major pathways that was significantly altered in Vem-resistant clones relative to parental 8505C (B) and WRO (E) cell lines. All cell lines were treated with DMSO at 0.1% for 24 hours. (C, F) Differentially expressed genes identified by acute treatment of vemurafenib were compared with those genes revealed in Vem-resistant clones for 8505C (C) and WRO (F) cell lines. (G, I) Representative flow cytometry plots exhibit the level of HLA-A2 and ICAM-1 expression in parental (Par) and Vem-resistant clone 1 (Vem-R1) 8505C cells (G) or parental and Vem resistant clone 6 (Vem-R6) WRO (I) cells. Vem (2) indicates treatment with Vem (2 μM, 72 hr). (H, J) MFI fold change of HLA-A2 and ICAM-1 expression in Par 8505C and Vem-R1 8505C clone (H), and Par WRO and Vem-R6 WRO clone (J). Expression at the basal level with DMSO treatment was compared with treatment with Vem (0.5 μM or 2 μM, 72hr). Data represents mean ± sem (n = 5–8 per group; ***, p < 0.001, ****, p < 0.0001 by Student’s t-test).

WRO-Res-6 cells exhibited much fewer differentially expressed genes when compared to parental WRO cells (Fig. 3D and Supplementary Table 4). Distinct pathways, such as cytokine-cytokine receptor interaction and cAMP signaling pathway, were enriched relative to parental cells briefly treated with BRAFi (Fig. 3E). WRO-Res-6 cells were associated more with suppression of IFNγ downstream genes rather than activation as shown in BRAFi-resistant 8505C cells (Fig. 3A and D). IPA analysis revealed that WRO-Vem-R6 cells were predicted to have a strong activation of suppressor of cytokine signaling 1 (SOCS1), which takes part in negative feedback loop to attenuate IFNγ signaling (Supplementary Fig. S7). Even though the direction of differentially expressed genes in WRO-Res-6 cells diverge from those regulated by brief treatment of vemurafenib, about half of those differentially expressed genes in WRO-Res-6 cells were commonly regulated by acute response to vemurafenib (Fig. 3F). Taken together, dysregulation of IFNγ downstream signaling is central to brief and chronic response to vemurafenib in both 8505C and WRO cells.

Although BRAFi-resistant 8505C and WRO clones appear to have unique molecular signatures, the expression of two major IFNγ downstream gene targets, HLA-A2 and ICAM-1, were distinguished by elevated expression in resistant clones compared to parental cells (Fig. 3G and I, Supplementary Tables 3 and 4). 8505C-Res-1 and WRO-Res-6 clones had approximately 4.5-fold and 7-fold increase in the MFI of HLA-A2 and ICAM-1 relative to wild-type counterparts, respectively (Fig. 3H and J). Treatment with Vem induced HLA-A2 expression over 10-fold in both BRAFi-resistant clones relative to parental cells, indicative of the potentiated IFNγ signaling pathway in the resistant cells. The ICAM-1 expression was significantly higher in BRAFi-resistant clones, with more pronounced elevation in 8505C-Res-1 cells relative to WRO-Res-6 cells.

To investigate the downstream signaling pathway of IFNγ, we analyzed the activities of STAT proteins in these cell lines. Following the receptor engagement with IFNγ, the IFNγ receptors and recruited Janus kinase (JAK) complexes are phosphorylated, and create docking and activation sites for STATs (38). STAT1 and STAT3 are the major STATs activated by the JAK activation, which stimulate expression of IFN-stimulated genes. We treated parental and Vem resistant clones of 8505C and WRO with JAK/STAT pathway inhibitors, alone or in combination with Vem. The additional drug treatments included a AZD1480 (AZD), an ATC-competitive JAK2 inhibitor, and fludarabine (Flu), a STAT1 inhibitor, (Flu). Similar to the results in Fig. 2B and D, the p-ERK1/2 level remained high in Vem-resistant clones compared to parental thyroid cancer cancer cells after Vem treatment whereas the total ERK level was less affected (Fig. 4A and 4B). Monotreatment with AZD or Flu did not change the p-ERK1/2 levels in parental 8505C cells, although p-ERK1/2 signals were increased slightly in the WRO cells by the same treatment. However, because the ERK activity of both Vem-resistant clones was not significantly altered by the AZD or Flu, our data implicate that the JAK/STAT signaling blockade do not play a critical role in augmenting the MAPK/ERK response.

Figure 4. BRAF-mutant thyroid cancer cells activated JAK/STAT signaling pathway upon BRAFi exposure.

(A-B) Representative western blots of p-ERK1/2, p-STAT1, and p-STAT3 are shown for Par and Vem-R clones in 8505C (A) and WRO (B) cells. Cells were treated with Vem (2 μM), JAK2 inhibitor AZD1480 (AZD) (1 μM), or STAT1 inhibitor Fludarabine (Flu) (0.25 μM) either alone or in combination for 6 hours. Each band intensity in immunoblots were quantified using ImageJ software. p-ERK to t-ERK levels were normalized to the GAPDH expression in each experimental condition (The normalized p-ERK level in untreated, parental cells is set to 1). Data are representative of at least 3 independent biological replicates. (C-D) MFI fold change of HLA-A2 expression in 8505C (C) and WRO (D) parental (light gray) and vemurafenib-resistant (dark gray) cell lines after mono- and combination therapy with vemurafenib and fludarabine.

Strikingly, we found that both 8505C-Vem-Res-1 and WRO-Vem-Res-6 cells exhibit significantly elevated level of p-STAT1 and p-STAT3 compared to parental cells with barely detectable expression at a basal condition (Fig. 4A and B). A transient Vem treatment of parental cells also increased the p-STAT levels, which extent was more pronounced in WRO than the 8505C. The phosphorylated STAT1 and STAT3 levels were markedly reduced in cells treated with AZD alone or in combination with Vem. Meanwhile, Flu treatment was less effective in blocking the p-STAT1 (< 50%) in parental and Vem-resistant thyroid cancer cell lines. These data show that upstream inhibition of JAK2 via AZD is more effective than Flu in inhibiting phosphorylation of the STATs in BRAF-mutant thyroid cancer cell lines. Overall, Vem-resistant BRAF-mutant thyroid cancer cells shared a robust activation of STAT1 and STAT3.

We next quantified surface expression of HLA-A2, a marker downstream of the JAK/STAT pathway, using flow cytometry after Vem-resistant cells were treated with Vem alone or in combination with Flu. As shown in Fig. 3G and I, the HLA-A2 expression levels were higher in 8505C-Vem-R1 and WRO-Vem-R6 clones than parental cells. In both parental and Vem-resistant clones, HLA-A2 expression increased further when treated only with Vem (Fig. 4C and D). Combination treatment with Vem and Flu decreased the HLA-A2 expression level significantly to a similar level as untreated Vem-resistant cells. This finding suggests that the Flu treatment was effective in decreasing HLA-A2 expression, a functional demonstration of blockade of JAK/STAT signaling, despite less pronounced effect on inhibiting phosphorylation of STAT1.

In order to determine whether blockade of JAK/STAT-signalng pathway sensitizes BRAF^V600E thyroid cancer cells to Vem, we next analyzed cell growth of 8505C, WRO, and SW1736, a BRAF^V600E ATC cell line, after treatment with chemical inhibitors of JAK/STAT signal mediators in combination with Vem. A high concentration of AZD inhibited 8505C cell growth to the similar level as Vem, but the combination treatment with Vem and AZD was not markedly different from monotreatment with either Vem or AZD (Fig. 5A). SW1736 and WRO cell growth was also decreased significantly in the combination treatment with Vem and AZD. The combined addition of Vem and Flu was also effective in controlling 8505C cell growth with Flu showing dose-dependent effect in monotreatment. (Fig. 5B). SW1736 cells displayed sensitivity to Flu mono- and combination treatment with Vem. Flu treatment was also effective in controlling WRO cell growth, even though the extent of inhibition by the combination treatment was not greater than Vem monotreatment. Next, we have examined whether endogenous JAK/STAT pathway inhibition in 8505C cells changed 8505C cell susceptibility to the Vem. CRISPR/Cas9-mediated depletion of STAT1 diminished the STAT1 expression by 50% (Fig 5C), which reduction was also verified by immunoblot analysis (Fig. 5D). Next, we targeted IFNGR1, which is one of the surface receptor with IFNGR2, which is preassociated with JAK proteins. More than 50% of IFNGR1 expression decrease were achieved by the CRISPR/Cas9-mediated depletion when examined at the mRNA and surface protein levels (Fig. 5E–F). Depletion of the STAT1 enhanced 8505C cell apoptosis and overall death upon Vem treatment compared to wildtype 8505C cells (Fig. 5G–H). The endogenous IFNGR1 expression did not significantly change 8505C cell sensitivity to Vem. Overall, our data demonstrate that blockade of the JAK/STAT signaling inhibited significant cell growth in BRAF mutant thyroid cancer cell lines.

Figure 5. Inhibition of the JAK/STAT pathway sensitizes BRAF-mutant thyroid cancer cell lines to vemurafenib.

(A-B) Cell viability of BRAF-mutant thyroid cancer cell lines, 8505C, SW1736, and WRO, after JAK/STAT pathway inhibition by increasing concentrations of AZD1480 (A) or Fludarabine (B), as monotherapy or in combination with vemurafenib (2 μM). Data from each experimental condition are normalized to vehicle (DMSO)-treatment group with no drug addition, and vehicle-treatment group is set to 100%. Data represents mean ± SEM (*p<0.05; **p<0.01; ***p<0.001; ****p<0.0001 by one-way Anova test). (C-D) qRT-PCR (C) and representative Western blot (D) results for STAT1 expression in 8505C cells expressing sg-STAT1 relative to no treatment (n = 6 for each condition; **p<0.01 by Student’s t-test for (C)). (E-F) qRT-PCR (E) and representative flow cytometry histogram (F) results for IFNGR1 expression in 8505C cells expressing sg-IFNGR11 relative to no treatment (n = 6 for each condition, ****p<0.0001 by Student’s t-test (E); n = 4 per group (F)). (G) Representative flow cytometry results for Annexin V and propidium iodide staining in wild-type, sg-STAT1, and sg-IFNGR11 expressing 8505C cells after 24 hr. treatment of Vem (5 μM). (H) Percentages of cell death were plotted by combining both apopototic and necrotic cell populations (n = 6 for each condition, *p<0.05 by 2-tailed Student’s t-test and one-way Anova analysis).

IRF1, a pro-inflammatory transcription factor and downstream target of the JAK1/2-STAT1 pathway, was predicted to be activated after Vem treatment (Fig. 1C). To investigate a potential connection between IRF1 expression with upstream JAK/STAT pathway genes in thyroid cancer, we analyzed STAT1 and IRF1 expression in previously published data from Memorial Sloan Ketterling Cancer Center (MSKCC) on PDTC and ATC (3). We found a strong positive correlation between STAT1 and IRF1 expression (Fig. 6A), suggesting a shared regulatory pathway. Next, we analyzed IRF1 mRNA levels of parental and Vem resistant 8505C cells after treatment with Vem and AZD, alone or in combination by qPCR. In parental and Vem-Res-1 8505C cells, IRF1 expression increased more than 2-fold after treatment with Vem alone compared to untreated controls (Fig. 6B). The addition of AZD with Vem decreased IRF1 expression significantly in both parental and Vem resistant 8505C cells. These data support that Vem treatment activates IRF1 via JAK1/2.

Figure 6. IRF1 and STAT1 expression are correlated with worse clinical outcomes in thyroid cancers.

(A) IRF1 and STAT1 mRNA expression is positively correlated in PDTC and ATC (3). P significance was determined by Spearman test (n = 37). (B) qPCR analysis of normalized IRF1 mRNA relative to GAPDH level of 8505C-Par and 8505C-Vem-R1 after Vem and Azd treatment (**p<0.01; ***p<0.001; one-way Anova test). (C) Correlation analysis of the mRNA level of IRF1 and STAT1 in 388 PTC patient tumor samples using the published TCGA study (34). (D) qPCR analysis of measuring IRF1 level relative to β-actin in fresh tumor culture after 24 hours of treatment with Vem (2 μM). Each sample was normalized to the vehicle control (DMSO, 0.1%)-treated condition. BRAF^V600E tumor samples showed significant upregulation when compared to BRAF^WT tumor samples (****, p < 0.0001 by unpaired t-test). (E) Overexpression of IRF1 is associated with worse overall survival in a PDTC and ATC study from MSKCC (3). mRNA level quantified with the probe 23875 for IRF1 is plotted. P value determined using log-rank test (n = 17 and 18 for top 50% and bottom 50% expression, respectively). (F) Disease-free survival analysis of the 388 PTC patients with IRF1 mRNA of z score >= 1.5 relative to unaltered level (34). (G) Survival curve is plotted for PDTC and ATC patients with different STAT1 expression level using probe 209969. (H) Disease-free survival analysis of the PTC patients with STAT1 mRNA z score >= 1.5. P value was determined using a log-rank test.

We next assessed STAT1 and IRF1 expression in PTCs, less aggressive thyroid cancer subtypes, using published TCGA data on 388 PTC patient samples (34). Similarly, we found a strong positive correlation between STAT1 and IRF1 expression, implicating a common pathway regulating both STAT1 and IRF1 in PTC (Fig. 6C). We have used 4 freshly isolated, patient-derived thyroid tumor tissues for drug treatment and analysis (Fig. 6D). Our patient-derived cohort included 3 PTCs and one poorly differentiated thyroid cancer sample (Supplementary Table 5). Two of the 4 samples had the BRAF^V600E mutation, and both mutated BRAF samples had significant upregulation of IRF1 compared to the wild type BRAF samples. These data show that a wide range of BRAF^V600E thyroid cancer pathological subtypes elicits JAK/STAT activation upon Vem treatment.

To further elucidate the clinical relevance of IRF1 expression, as an indicator for JAK/STAT signaling activation in thyroid cancer patients, we performed overall survival analysis of the PDTC and ATC patients (Fig. 6E) (3) and disease-free survival analysis in 388 PTC patients (Fig. 6F) (34). Two studies used different experimental approaches to quantify and assess gene expression levels in thyroid cancer (the former study used microarray analysis and the latter used RNA-sequencing). In PDTC and ATC patients, tumor samples with the top 50% of IRF1 expression showed worse overall survival compared to the bottom 50% of expression (Fig. 5E). Elevated IRF1 level was also significantly associated with shorter disease-free survival rate relative to unaltered IRF1 expression in PTC patients (Fig. 6F).

Next, we correlated the STAT1 expression with the clinical data of thyroid cancer patients. Increased expression of STAT1 had worse overall survival compared to the bottom 50% of expression in PDTC and ATC patients (Fig. 6G). We found that the difference in survival was likely driven by the histology that is associated with disparate expression levels of IRF1 and STAT1; the ATC showed higher expression of these genes when compared to PDTC (median PDTC 4.9 vs. ATC 7.8 for IRF1; median PDTC 7.8 vs. ATC 10.5 for STAT1). In PTC, elevated STAT1 expression was significantly associated with shorter disease-free survival rate (Fig. 6H). These findings suggest that activation of the JAK/STAT pathway may be associated with tumor recurrence and worse prognosis.

Discussion:

Thyroid cancer shares many similarities with melanoma, including prevalent mutation in the BRAF oncogene; however, BRAFi has shown much greater efficacy in melanoma compared to thyroid cancer (10,13–15). Studies have alluded to intrinsic pathways leading to resistance to BRAFi in thyroid cancer (19,39), and BRAFi therapy for patients with PTC has also led to resistance (20,21). We aimed to ascertain novel pathways that lead to BRAFi resistance in thyroid cancers, using BRAF mutated thyroid cancer cell lines with Vem resistant clones, patient-derived thyroid tumor samples and publicly available expression data from PTC, PDTC, and ATC.

Specifically, we found that Vem treatment activates the JAK/STAT pathway with quantitative differences in expression of downstream targets of the pathway, including ICAM-1, HLA-A2, and IRF1. Furthermore, Vem resistant clones acquired dysregulated JAK/STAT-signaling compared to the parental cell lines. Treatment with pharmacological inhibitors or genetic disruption targeting JAK/STAT pathway inhibited significant cell growth of BRAF^V600E thyroid cancer cell lines; the cell growth control effect was near the level achieved by Vem treatment, supporting for the role of JAK/STAT pathway in sustaining the growth of BRAF mutant thyroid cancer cells.

As the JAK/STAT pathway was shown to be activated in resistant clones and after BRAFi treatment, additional targeting of this pathway may be beneficial for combination treatment. Combined treatment with Vem and inhibitors of JAK/STAT pathway exhibited a larger effect on 8505C and SW17336 cell survival relative to monotherapy when lower concentration of the drug was used. The benefit of combination treatment was less clear in WRO cells; this may be due to lower activation of the JAK/STAT pathway observed in parental WRO cells compared to 8505C cells at a basal condition and upon treatment with Vem. On the contrary, WRO-Vem-R6 cells retained a robust level of P-STAT1 and were predicted to have a strong activation of suppressor of cytokine signaling 1 (SOCS1), which takes part in negative feedback loop to attenuate JAK/STAT signaling (Supplementary Fig. S5). This difference may explain why the response of WRO-Vem-R6 cells was less robust to Vem or inhibitors of JAK/STAT pathway than was seen in 8505C-Vem-R1 cells. Nonetheless, converging on the JAK/STAT signaling pathway to acquire resistance to Vem in two independently selected Vem-resistant clones and in response to Vem treatment in different histologies of thyroid cancer cells support the crucial survival and escape mechanisms provided by the JAK/STAT signaling in thyroid cancer.

Our study reveals that activation of JAK/STAT signaling pathway is an integral pathway that thyroid cancer cells employ in response to Vem treatment. We found that a strong JAK/STAT activation provided a growth/survival advantage in thyroid cancer cell lines to evade the cytotoxic effects of Vem. Furthermore, our analysis of the previously reported MSKCC study demonstrated that ATC and PDTC patients with tumors with a signature of an activated JAK/STAT signaling pathway- upregulated STAT1 or IRF1 expression- had worse overall survival. Additionally, PTC patients in the TCGA cohort with elevated JAK/STAT response had earlier recurrence than PTC patients with basal expression, supporting for the tumor growth advantage with activated JAK/STAT signaling. Importantly, a recently published study by Lee et al. demonstrated that 2 out of 6 BRAF^V600E thyroid cancers that dedifferentiated to more aggressive histologies following the BRAFi therapy featured missense mutations in the JAK1 and JAK2 (40). Taken together, our data support that dysregulation of the JAK/STAT pathway may be one of the mechanisms that underlie both the BRAFi resistance and thyroid cancer pathogenesis.

Further studies to maximize effectiveness of BRAFi therapy by inhibiting JAK/STAT signaling in expanded panel of human thyroid tumors and preclinical thyroid cancer models will be important to validate our findings and rationally design therapeutic modalities relevant to cancer patients. Our findings in the thyroid cancer are further supported by melanoma literature demonstrating that inhibition of JAK1 signaling provides benefit and overcome BRAFi resistance (41). Future works to examine how the combination treatment targeting the BRAF and JAK/STAT pathways affects the thyroid tumor-immune microenvironment in vivo will be important given that thyroid cancers are known to be infiltrated with immune cells and the JAK/STAT signaling also plays an important role in immune regulation.

In conclusion, thyroid cancer cells quickly acquire resistance to BRAFi and an important mechanism contributing to this resistance appears to be through the JAK-STAT activation. Dual inhibition of BRAF and the JAK-STAT pathway is a potential therapeutic treatment for anticancer activity and to overcome drug resistance to BRAFi in thyroid cancer.

Implications:

Dual inhibition of BRAF and JAK/STAT signaling pathway is a potential therapeutic treatment for anticancer activity and to overcome drug resistance to BRAFi in thyroid cancer.