Sorafenib and Quinacrine Target Anti-Apoptotic Protein MCL1: A Poor Prognostic Marker in Anaplastic Thyroid Cancer (ATC)

Anaplastic thyroid cancer (ATC) is a rare but highly aggressive form of thyroid cancer with a median survival of 3–5 months that remains unchanged despite increased understanding of the etiology of the disease (1). Unfortunately, refractoriness to chemo- and radiotherapy is a common feature and development of novel therapeutics is essential to improve the prognosis of ATC patients. Recent insight into genetic lesions that drive the progression of ATC may help shape concepts of novel strategies of targeted therapeutic approaches. Currently, the proto-oncogenes RET, RAS, and BRAF are some of the best described targets in thyroid cancer (2, 3) and specifically activating BRAF V600E mutations have been reported to occur in approximately 25% of ATCs (4, 5).

Given the frequency of activating mutations of the BRAF oncogene in ATC, it is perhaps not surprising that the multikinase inhibitor sorafenib (Nexavar), an approved drug for the treatment of advanced renal carcinoma (6), unresectable hepatocellular carcinoma (7), and progressive radioactive iodine-refractory differentiated thyroid carcinoma (8), has sparked clinical interest in ATC. Sorafenib targets BRAF and CRAF, in addition to several other tyrosine kinases, suggesting that at least a subpopulation of ATC patients might respond to sorafenib. However, sorafenib has shown limited activity in the reported clinical trials of ATC to date (9, 10). One phase II study of sorafenib in patients with advanced ATC indicated activity but at low frequency in a similar manner as fosbretabulin, a vascular disrupting agent (10). It is becoming increasingly clear that sorafenib may trigger toxicities in thyroid cancer patients that frequently result in dose reduction (11). Thus, treatment with sorafenib alone may be insufficient to evoke a strong antitumor response in ATC patients and incorporation of additional targeted therapeutics that exhibit low toxicity into sorafenib protocols may be required to improve outcome.

Additional molecular changes occur in ATC cells that may contribute to disease aggressiveness include aberrant activation of NFκB signaling. Imbalanced activation of NFκB may possibly contribute to the treatment-refractory proinflammatory and metastatic phenotype of ATC. Indeed, the expression of RELA/p65 was found to be increased in ATC tissues compared with that of normal thyroid (12). Several inhibitors of NFκB signaling such as dehydroxymethylepoxyquinomicin (DHMEQ), triptolide, imatinib, and bortezomib have shown promising results in preclinical experiments with ATC cells (13–16).

The acridine derivative quinacrine, used historically for malaria treatment, is a potent inhibitor of NFκB signaling (17), and is currently being evaluated in phase II cancer clinical trials (18). Its extensive use during the Second World War by over three million soldiers makes it a well-studied drug with a safety profile based on extensive epidemiologic data. Moreover, as quinacrine is currently used for the treatment of giardiasis or lupus and is very affordable (∼$30 USD/month of therapy), it is a good candidate compound for repositioning to target malignancies with oncogenic activation of NFκB signaling. We recently reported the effectiveness of quinacrine with other standard-of-care therapies in liver and colon cancer (19, 20). Quinacrine was found to effectively target NFκB and inhibit MCL1 expression in colorectal cancer cells. In addition, we have previously shown that sorafenib inhibits both JAK/STAT3 and NFκB signaling that also results in the downregulation of MCL1 (21, 22).

Herein, we show that quinacrine combines favorably with sorafenib in an additive to synergistic manner and generates a strong antitumor response in an orthotopic mice model of ATC without significant toxicity. Treating ATC cells with the sorafenib/quinacrine drug combination dramatically reduced the levels of antiapoptotic MCL1 and triggered MCL1–dependent cell death. MCL1 protein is overexpressed in a subset of ATC patient specimens compared with non-neoplastic thyroid. Furthermore, gene set enrichment analysis of meta-data indicates hyperactivation of NFκB signaling in ATCs. These findings provide a rationale for future clinical trials of the drug combination quinacrine/sorafenib in aggressive thyroid cancers.

Detailed Materials and Methods are provided as Supplementary Information.

For details about the cell lines and reagents used in this study, see ref. 21.

Twelve ATCs and ten “normal” (non-neoplastic) thyroid patient formalin-fixed paraffin-embedded (FFPE) tissue specimens were obtained from the Department of Pathology's tissue bank/Penn State Hershey Cancer Institute (Hershey, PA). The following antibodies were used for immunohistochemical detection: anti-phosphotyrosine-STAT3 (Y705), anti-NFκB-p65, and anti-MCL1. Slides were scored and the expression parameters were correlated with clinical outcome.

Gene expression data were collected through Oncomine/Gene Expression Omnibus (GEO) and previously published reports containing complete curated expression profiles of resected clinical specimens of indicated tissues. The gene expression profiles were computationally analyzed for statistically significant differences with respect to the expression of four gene sets of putative NFκB target genes using the Gene Set Enrichment Analysis (GSEA; http://www.broadinstitute.org/gsea/index.jsp; ref. 23).

Total RNA was extracted using the RNeasy Mini Kit and converted to cDNA using the First Strand Synthesis kit. For qPCR analysis, the SYBR select master mix was used together with the primers for MCL1, STAT3, NFκB-p65/RelA.

The siGENOME Non-Targeting siRNA #1 and MCL1 were transfected and cells were collected for dose response by CellTiter-Glo assay. The same batch of cells was plated for measuring knockdown efficiency of MCL1.

Immunoblotting was done as described previously in ref. 21.

Sub-G₁ and cell survival analysis was done as described previously in ref. 21.

A midline cervical incision was made and the underlying submandibular glands were retracted laterally. Gentle retraction of the midline strap muscles was performed to reveal the thyroid glands adjacent to the trachea and visible underneath a deep layer of semitransparent strap muscles. Direct injection of the thyroid was done, the submandibular glands were returned to the original location, and the skin closed in a single layer (24). Mice were treated with sorafenib, doxorubicin, and quinacrine as described previously (21), and followed for tumor burden.

Mouse FFPE sections were analyzed by IHC, as described previously, for clinical specimens above. Antibodies were used for the detection of VEGF, CD34 and Ki67 are listed in Supplementary Information. The detection of apoptotic cells was done by the TUNEL kit.

C57BL6/J mice were injected with doses of quinacine (2 times) × 1 dose oral gavage, sorafenib (5 times), whereas doxorubicin (2.5 times) of that used in the orthotopic model. At day 7, mice were sacrificed, whole blood was isolated by cardiac puncture for hematology and serum chemistry, and tissues were isolated for histologic assessment.

All comparisons were made relative to untreated controls, and statistically significant differences are indicated as *P < 0.05 and **P < 0.005. We used CalcuSyn software to determine synergy. Overall survival was determined using Kaplan–Meier method. Cox multiregression analysis was performed on patient parameters relating to the ATC specimens stained by IHC for pStat3, NFκB-p65, and MCL1.

Targeted therapy acts against specific molecules that are preferentially present in malignant cells and required for cancer cell survival and growth. We have previously shown that both sorafenib and quinacrine can target cell death signaling through modulation of NFκB and JAK-STAT signaling in cancer cells (19–22). Although these pathways are frequently activated in cancer (25, 26), it has until recently been largely unknown to what extent these pathways are also aberrantly activated in ATC. To address target availability in ATC, we compared the expression of NFκB-p65 (RelA), STAT3, and MCL1 mRNA in a panel of human cancer cell lines and resected ATC specimens to that in “normal” (non-neoplastic) thyroid by qRT-PCR (Supplementary Fig. S1). The genomic profile of the ATC cell lines used in our study have been established and verified previously (27–29).

Detectable expression of MCL1, RELA, and STAT3 mRNA was found in all five human ATC cell lines tested (Supplementary Fig. S1A) tested by qRT-PCR. Expression of MCL1 and RELA varied from average to average-high in comparison with the prostate cancer cell lines and the hepatoma cell line HepG2. Expression of STAT3 was variable among the ATC cells, and, interestingly STAT3 mRNA was found underexpressed in the 16T and 21T cell lines. In general, the mRNA expression levels of these oncogenes was lower in ATC cells than the expression found in the untransformed fibroblast cell lines MRC5 and HFF (Supplementary Fig. S1A), only the prostate cancer cell line DU145 expressed more MCL1, RELA, and STAT3 mRNA than the fibroblasts. Thus, a certain level of variability with respect to expression of these genes is present in ATC cells.

To better address whether the expression of these genes may be upregulated by carcinogenesis, we hypothesized that a better comparison would be between resected ATC and non-neoplastic thyroid (Supplementary Fig. S1B–S1D). We observed that the mRNA of MCL1 and RelA were expressed in both normal thyroid and ATC tissues, although the sample size was limited due to specimen availability. Interestingly, RelA was underexpressed at the mRNA level in our samples, whereas expression data from one ATC sample indicated that potentially a subpopulation of ATCs may express high levels of MCL1 mRNA (Supplementary Fig. S1B). However, STAT3 mRNA expression was low or unquantifiable in ATC in comparison with that in normal thyroid tissue (Supplementary Fig. S1D).

However, as both NFκB and JAK-STAT signaling can be significantly regulated at the posttranscriptional level, we aimed to address this by assessing the protein expression in the normal and ATC tissues. We performed IHC for NFκB-p65, MCL1, and phosphorylated (active) STAT3 (STAT3pY) on resected formalin-fixed paraffin-embedded (FFPE) patient tissues. Our initial analysis of the stained tissues did not reveal significantly different overall expression (number of positive cells) of NFκB-p65 or STAT3pY between normal thyroid and ATC (Supplementary Fig. S2A and S2B). In fact, some normal thyroid tissues showed higher levels of STAT3pY (Supplementary Fig. S2B) consistent with our qPCR data showing lower expression of STAT3 mRNA (Supplementary Fig. S1D). STAT3pY staining was found predominantly in vascular endothelial cells and occasionally in thyroid follicular cells (Supplementary Fig. S2C). Similar findings with respect to STAT3 have been reported and STAT3 has been proposed to function as a tumor suppressor in thyroid tumorigenesis by suppressing HIF-1α signaling, aerobic glycolysis, and tumor growth (30).

We assessed whether NFκB-p65 staining (in cytoplasm and nucleus) correlated with overall survival (OS) in ATC patients. Neither “low” (below median) nor “high” (above median) overall expression of NFκB-p65 correlated with OS (P = 0.4936; Supplementary Fig. S3A). However, in the case of NFκB-p65, we found that ATC cells would predominantly express NFκB-p65 in the cytoplasm or nucleus and strong nuclear staining was only observed in ATC tissues (Fig. 1A). This finding is consistent with NFκB-p65/RelA being latent in the cytoplasm but translocated into the nucleus to modulate transcription of target genes once activated. Thus, we hypothesized that ATCs with aberrantly activated NFκB signaling may show primarily nuclear staining of NFκB-p65 as a consequence of activation (12). Four out of 11 ATC patients were classified as having predominately nuclear staining for NFκB-p65 and showed an overall poorer prognosis with median survival of 1.4 months compared with 7.9 months (HR, 7.8; 95% CI, 1.24–48.67) for patients with predominantly cytoplasmic staining (Fig. 1B). Nuclear staining for NFκB-p65 was not the only predictor (nor independent) of survival. ATC patients with “small” tumors (greatest dimension <10 cm) and/or who were subjected to surgery as part of their treatment had better OS as compared with patients who had “large” tumors (greatest dimension >10 cm) and who did not receive surgery (Supplementary Fig. S3B and S3C). Although tumor stage (IV-B vs. IV-C) showed a trend towards influencing OS it was not statistically significant (P = 0.0690) in the analyzed cohort (Supplementary Fig. S3D). A potential explanation for the latter finding may be that our cohort was of modest size even by ATC standards and increased sample size would likely confirm tumor stage to prognosticate survival.

ATCs appeared to stain more frequently for the NFκB target MCL1 compared with normal thyroid tissue (Fig. 1C and D). Three of 4 ATCs with nuclear NFκB-p65 also had an expression (% Mcl-1+ cells) above group median (Fig. 1D). Furthermore, by Kaplan–Meier estimates, we observed a trend where patients with “high” (above median) MCL1 appeared to have decreased OS compared with patients with “low” (below median) MCL1 (Fig. 1E) although this trend was not statistically significant when subjected to multivariate survival analysis using Cox regression model (P = 0.120).

To validate our findings with respect to increased NFκB-p65 activation in ATC compared with normal thyroid tissue, we performed meta-analysis of previously published gene expression data on ATC. Three studies allowed us access to expression profiles for data mining of a sufficiently large set of genes (31–33). Indeed, GSEA of the expression profiles previously published (33) revealed that normal thyroid and ATC separated in distinct phenotypic expression profiles (Supplementary Fig. S4A). We investigated whether gene sets with activation of NFκB signaling were enriched in clinical ATCs. Subsequently, we matched two gene sets (HALLMARK_INFLAMMATORY_RESPONSES and TNF_SIGNALING_VIA_NFKB) present in the molecular signatures database (MSigDB) with the expression profiles of normal thyroid and ATC. Increased overlap between genes expressed in the ATC expression profile was observed with both of the gene sets (FDR q < 0.00001 and nominal P < 0.01; Fig. 2A and B). In addition, we assessed whether sets of genes containing the RelA annotation motif GGGAMTTYCC within 2 kb of their transcription start site were increasingly expressed in ATC. Such genes were increasingly expressed in clinical ATC specimens compared with normal thyroid (FDR q < 0.0001 and nominal P < 0.01; Fig. 2C). We performed GSEA for gene sets relating to NFκB activity for ATC-curated gene expression profiles published by Salvatore and colleagues (32) and Hebrant and colleagues (ref. 31; Supplementary Fig. S4B and S4D). Although no enriched expression of NFκB-dependent genes was found with statistical certainty (i.e., FDR q < 0.25 and P < 0.05) on the curated expression profiles, the ATC expression profiles from both studies contained a larger number of genes linked to NFκB activity than either normal thyroid tissues or papillary thyroid cancers. We did not find MCL1 overexpression in ATC compared with neither normal thyroid tissues nor papillary thyroid cancer (data not shown). In contrast, data from the established expression profile from Hebrant and colleagues (31) indicated reduced mRNA expression of MCL1 compared to normal tissues and papillary thyroid cancer (data not shown). MCL1 protein expression has been shown to be regulated both at the transcriptional and posttranslational level (35) and thus they may not always correlate in cancer.

We have previously assessed sorafenib (S) and quinacrine (Q)-dependent targeting of the antiapoptotic protein MCL1 to sensitize tumor cells to programmed cell death (apoptosis). Targeting of MCL1 by S and Q was previously shown by us to overcome TRAIL- and chemoresistance in cancer cells (19–22). We hypothesized that MCL1 may be a bona fide drug target in ATC.

Q and S displayed single agent antitumor activity in human ATC cells. Dose–response relationships for S (Fig. 3A) and Q (Fig. 3B) were established in the ATC cell lines THJ-16T, THJ-21T, and THJ-29T (Table 1). The THJ-16T cell line appeared to be significantly more resistant to S compared with the THJ-21T and THJ-29T cell lines. In contrast, the same cell lines showed smaller differences with respect to sensitivity to Q (Fig. 3B; Table 1). The antitumor response appeared to be at least partially explained by apoptosis in the ATC cell population. Increased sub-G₁ DNA was observed following short-term treatment with both S (S10) and Q (Q20) alone (Fig. 3C). An additive effect was observed by the combination of S and Q (S10Q20). We further characterized the dose–response characteristics of the combination of S and Q in THJ-29T and THJ-16T cells using the bioluminescent CellTiter-Glo cell survival assay (Fig. 3D). A combinatorial effect (reduced fluorescence indicating reduced survival) was observed with S and Q. We assessed the nature of the S and Q combinatorial drug response in the sorafenib-refractory ATC cell line THJ-16T. We used the theorem of Chou–Talalay (36) to generate combination indices (CI) for the drug combination of S/Q in ATC (Fig. 3E). The best outcome as judged by CIs generated in our cell survival assay was “strong synergism” indicating that Q may sensitize sorafenib-refractory ATC cells to cell death. Taken together, our data indicate that the drug combination S/Q triggers apoptosis and antitumor responses that are synergistic in ATC cells.

We previously showed that treatment of 8505C with increasing concentrations of sorafenib (S) that the drug ameliorates the expression of several prosurvival proteins (21). Among others, phosphorylation of STAT3, MEK, and ERK was inhibited. In addition, expression of prosurvival protein MCL1, but not Bcl-XL, was inhibited by drug S. Western blot analysis showed that low dose treatment of 8505C cells with Q (1 μmol/L) and S (0.3–3.2 μmol/L) inhibits phosphorylation of STAT3 at the prosurvival phosphorylation site of Y705 (Fig. 4A). At higher doses, the drug combination triggers a robust proapoptotic response in 8505C cells associated with an essentially complete eradication of MCL1 and full-length PARP (Fig. 4B). Expression of mutated p53 was not affected by the drug combination in 8505C cells. Indeed, previous data show that Q kills cancer cells independent of p53 (20). Q and S/Q inhibited the expression of NFκB-p65 (RelA) in 8505C and THJ-16T cells (Fig. 4C) supporting the notion that the NFκB pathway is a target in ATC cells. To functionally validate our findings with regards to p65/RelA and STAT3 expressions we used siRNA to target p65/RelA and total STAT3 in 8505C cells prior to treatment with sorafenib and quinacrine. Neither scrambled (Scr) siRNA nor siRNA targeting STAT3 had a significant impact on the cell viability dose–response relationship of sorafenib or quinacrine in 8505C cells (Fig. 4F and G). However, targeting of p65/RelA caused an approximately 2-fold increased sensitization of 8505C cells to sorafenib compared with cells subjected to treatment with Scr siRNA (IC₅₀: 5.15 μmol/L; 95% CI, 4.14–6.41 μmol/L and IC₅₀: 10.49 μmol/L, 95% CI, 7.56–14.5 μmol/L respectively; Fig. 4F). siRNA-dependent targeting of p65/RelA did not have any significant impact on the survival dose–response relationship following treatment with quinacrine (Fig. 4G). Thus, our Western blot protein data indicate that the drug combination of S/Q targets several prosurvival pathways and the antiapoptotic protein MCL1 in ATC cells in vitro. However, targeting of NFκB signaling may in particular sensitize ATC cells to sorafenib perhaps indicating that quinacrine may trigger synergism as a result of its capacity to target NFκB-dependent transcription.

To assess the extent to which MCL1 may be required for the therapeutic efficacy of the drug combination S/Q, we targeted the expression of MCL1 by siRNA gene silencing and assessed how this impacted the dose–response characteristic of S and Q in 8505C cells (Fig. 5A–D). Knockdown of MCL1 sensitized 8505C cells to S by 1.4-fold but did not affect the IC₅₀ to Q (Fig. 5C–E). Analyzing the combinatorial effect of S and Q in 8505C cells with and without targeted MCL1 expression showed that intact MCL1 expression is required to produce a potent combinatorial drug response. Synergism was observed at an IC_S80/Q60 combinatorial dose, whereas IC_S80/Q50 and IC_S80/Q30 produced near additive effects in the 8505C cell line (Fig. 5F). In contrast, MCL1 knockdown produced an antagonistic relationship between S and Q at the corresponding effective dose levels (Fig. 5G). We also found that Q synergistically sensitized 8505C cells to S at a concentration as low as 125 nmol/L (CI = 0.88) and this drug synergism was ameliorated completely by MCL1 depletion (CI = 1.03; data not shown). We propose a model where quinacrine-dependent inhibition of MCL1 expression contributes to the combinatorial efficacy of S/Q (Supplementary Fig. S5). This may favor targeting ATC with high NFκB and MCL1.

To test the efficacy of drug combination S/Q in vivo, we assessed to what extent ATC cells grew as tumors in immunodeficient (Nu/Nu) mice. ATC cells were labeled with fluorescent cell permeable dye CellVue Maroon and injected orthotopically into recipient mice (Supplementary Fig. S6A). At one week after injection, mice were subjected to NIR imaging to visualize fluorescently labeled ATC cells and document the extent of tumor growth. Nu/Nu mice are significantly less sensitive to DNA damage compared with immunodeficient SCID mice that lack the Ku70 subunit of DNA-PK, and thus were the preferred mouse model. Orthotopic models are considered clinically more relevant compared with subcutaneous models and therefore we used an orthotopic in vivo model (Supplementary Fig. S6B). Injected mice developed tumors in their thyroid gland within 7 days and started to show lethargy at day 12 (Fig. 6A). Efforts to assess tumor burden over time in these mice by noninvasive NIR imaging proved to be difficult as the thyroid is localized behind the salivary gland and so NIR imaging was used as a postmortem imaging methodology to verify ATC tumor growth from labeled implanted cells (Fig. 6B and C and Supplementary Fig. S6C). Immunohistochemical data indicated the 8505C tumor xenograft had high expression of MCL1 (Supplementary Fig. S6D).

Median survival for untreated mice was 22 days (Table 2). Although S, Q, and doxorubicin did increase median survival (26, 32, and 31 days, respectively) these trends did not reach statistical significance when compared with the median survival of untreated (control) mice (Fig. 6D–F; Table 2). These data are consistent with the clinical experience where sorafenib and doxorubicin provide a limited impact on OS for ATC patients. In contrast, we found the sorafenib and quinacrine (S/Q) combination significantly increases the survival of mice carrying 8505C xenografts in their thyroid gland compared with untreated mice (median survival 75.5 days, P = 0.0138; Gehan–Breslow–Wilcoxon test; Fig. 6G; Table 2). Analysis of FFPE thyroid tumors indicated increased TUNEL positivity and reduced expression of the proliferation marker Ki-67 following treatment with the different modalities (Fig. 6H; ref. 19). In summary, although sorafenib, quinacrine, and doxorubicin fail to produce a significant survival advantage in this ATC orthotopic mouse model, the combination of sorafenib and quinacrine increases the median survival 3-fold compared with that of untreated (control) mice without any apparent toxicity.

Although the safety profiles for sorafenib and quinacrine are known in human subjects, the potential for the drug combination to trigger toxicity in an additive/synergistic manner in vivo is largely unknown. We addressed this through a short-term (7 days) dose-escalation study in mice followed by a histopathologic assessment and blood correlatives following treatment. No lethality or lethargy was observed in the mice following exposure the supertherapeutic doses of S, Q, doxorubicin, and S/Q. However animals subjected to S/Q displayed a transient 15% weight loss nadir at day 2 following initiation of treatment (data not shown). All of the S/Q-treated animals recovered from this by day 5. Analysis of serum chemistry and hematology indicated no toxicity of S/Q compared with untreated (control) mice (Supplementary Figs. S7 and S8). In contrast, mice subjected to treatment with the first-line chemotherapeutic doxorubicin displayed a reduction in the number of RBCs and hemoglobin (Hgb) compared with mice treated with S/Q (Supplementary Fig. S8B and S8C). Furthermore, doxorubicin-treated mice had an increased level of platelets in their blood compared with control and S/Q-treated mice (Supplementary Fig. S8E). Histopathologic evaluation of the bone marrow, gastrointestinal (GI) tract, heart, pancreas, liver, and kidney suggested that doxorubicin-treated mice more frequently displayed signs of bone marrow and GI toxicity than the other treatment groups (Table 3). Such signs included prominent congested sinusoids in the bone marrow and villous blunting, crypt dilation, crypt hyperplasia, and inflammation of lamina propria in the duodenum. Although doxorubicin is known to cause cumulative-dose cardiotoxicity, we did not find any histopathologic evidence of this in our short-term study. Histopathologic signs of toxicity from the combination of S/Q were consistent with what was observed with treatment with either compound alone and did not increase in severity following the combination (Table 3 and data not shown). The only unique histopathologic manifestation that we were able to link to S/Q exposure was the appearance of a single small focal chronic cortical infarct with a volume of <0.01% in the kidneys of 2 of 5 mice subjected to S/Q (Table 3), which was not frequent enough to be statistically significant. Thus, the S/Q combination is well tolerated even at doses that greatly exceed those used to evoke potent antitumor responses in vivo.

Preclinical data indicates that targeting of tumor angiogenesis and VEGF signaling can trigger antitumor responses in models for ATC (37). Preclinical data and two clinical case studies also indicate that receptor tyrosine kinase (RTK) inhibitors that target VEGF signaling may hold promise for the treatment of ATC (38, 39). To address whether the combination of sorafenib and quinacrine have potential to modulate tumor angiogenesis, we performed IHC for CD34 (a marker for the presence of small vessel endothelium) and VEGF (a proangiogenic marker and effector molecule) on FFPE sections from 8505C xenografts from Nu/Nu mice subjected to treatment with S/Q (Supplementary Fig. S9). The number of CD34-positive vessels was dramatically reduced by S and the combination of S/Q (Supplementary Fig. S9A and S9B). Moreover, S alone and in combination with Q exhibited a potent inhibition of VEGF expression in the tumor xenografts. These data are consistent with sorafenib's antiangiogenic properties in addition to BRAF targeting. Taken together, our data indicate that the combination of sorafenib and quinacrine inhibits angiogenesis in addition to the targeting of antiapoptotic molecules in ATC cells.

Despite advances in understanding the molecular pathogenesis of ATC the disease remains associated with a poor prognosis. Increased efforts are being undertaken to define “actionable” genetic changes and develop improved treatment protocols. The multikinase inhibitor sorafenib has shown promising results in some solid malignancies but data from recent ATC clinical trials indicate that sorafenib has limited impact on overall survival in ATC (9, 10). Our studies show that sorafenib combines favorably with quinacrine to trigger a combinatorial response in ATC to significantly improve survival of mice implanted orthotopically with ATC cells relative to untreated (control), single agent, or doxorubicin-treated mice (Fig. 6). This S/Q drug combination targets NFκB-p65 and MCL1 in vitro (Figs. 4 and 5). Both NFκB-p65 and MCL1 are overexpressed in a subset of ATCs (Fig. 1). Furthermore, mouse dose-escalation toxicology data suggests that the S/Q drug combination is safe at the doses tested (Table 3; Supplementary Figs. S7 and S8). Further assessment of the drug combination in animal models for locoregional toxicities following radiation delivery to the head and neck may be necessary. ATC patients frequently undergo radiotherapy and it is currently unknown whether the drug combination of S/Q would have a negative therapeutic index in this context. However, our findings provide a foundation for evaluation of the S/Q drug combination in clinical trials.

It is somewhat surprising that benefit from sorafenib is limited despite 25% of ATCs showing BRAF V600E mutations as sorafenib targets multiple pathways with activated tyrosine kinases such as VEGFR and PDGFR signaling (40, 41). This observation could be due to the fact ATCs also show mutations in β-catenin that generates hyperactivated Wnt signaling (42) and that ATC cells undergo EMT and show evidence of increased stemness compared with more differentiated forms of thyroid cancer (43). EMT is associated with refractoriness to therapy and more aggressive behavior in several other malignancies. Moreover, tumor suppressor p53 is frequently lost and mutated in ATC and there is increased inflammatory NFκB signaling and enhanced activation of matrix metalloproteinases (MMP-9) which may promote the invasiveness of ATC (44, 45), features supported by our GSEA of meta-data (Fig. 2 and Supplementary Fig. S4). It is therefore unlikely that targeting a single molecular facet in ATC would yield a significant improvement of patient outcomes.

Interestingly, quinacrine is able to trigger antitumor responses independent of mutated p53 through inhibition of NFκB and MCL1 (19, 20). Quinacrine can also synergize with the antiangiogenic drugs such as cediranib that targets VEGF signaling (44) by blocking autophagy triggered by receptor tyrosine kinase inhibitors suggesting that quinacrine can modulate the activity of antiangiogenic agents. Sorafenib displays antiangiogenic activity in our in vivo model and represses VEGF expression including when combined with quinacrine (Supplementary Fig. S9), unlike single-agent quinacrine (Supplementary Fig. S9A–S9C). Thus, the combinatorial efficacy, multiple pathway targeting, safety, and low expense of quinacrine make the combination of sorafenib plus quinacrine an attractive option to further test in the clinic as ATC treatment.

We have previously shown that quinacrine targets NFκB/RelA and antiapoptotic MCL1 in cancer cells (19, 20). MCL1 is an antiapoptotic Bcl-2 family member and overexpression of MCL1 negatively impacts the responses to some therapeutics in patients with malignancies such as CLL (45). We show MCL1 is overexpressed in some ATCs and expression is inhibited through treatment with the drug combination of sorafenib and quinacrine in vitro (Figs. 1B, D, and E and 4B). Quinacrine has previously been shown to inhibit the transcriptional activity of NFκB/RelA and in addition to reactivate wild-type p53 in renal carcinoma cells (17). Quinacrine can inhibit expression of NFκB-p65 (Fig. 4C). This may be of relevance as our IHC data indicate that nuclear expression of NFκB-p65 is linked to poor prognosis in ATC patients (Fig. 1A and C). Three out of 4 ATCs with “high” MCL1 expression also showed nuclear expression of NFκB-p65 (Fig. 1D). MCL1 is a downstream target gene of NFκB/RelA (46) but we failed to substantiate significantly increased expression of MCL1 mRNA in clinical ATCs through data mining of expression profiles from previously published studies (31). However, several studies indicate that expression of MCL1 protein is also regulated posttranscriptionally and its mRNA/protein has a short half-life (35). It is possible that NFκB-p65 contributes to the MCL1 mRNA pool and subsequent protein synthesis in ATCs. Subsequently, lesions with nuclear NFκB-p65 may be selectively sensitive to quinacrine treatment. We hypothesize quinacrine disrupts the expression of MCL1 through inhibition of NFκB/RELA.

We observed a trend suggesting a link between MCL1 overexpression, tumor size, and OS (Fig. 1E). Evidence suggests that apoptosis signaling indeed may be impaired in ATC. The inhibitor of apoptosis protein (IAP) Survivin is overexpressed in ATCs compared with papillary thyroid cancers (31, 33) and Survivin expression has been proposed as a marker for dedifferentiation in thyroid carcinogenesis (49). Interestingly, expression of antiapoptotic Bcl-2 appears to be lost in ATC compared with well-differentiated papillary and poorly differentiated thyroid cancers (50). MCL1 overexpression has been documented in a number of solid tumors and small-molecule compounds targeting MCL1 directly are being developed for therapy, but the role of MCL1 in ATC remains largely unclear. MCL1 shows limited background expression in the follicular epithelium of normal thyroid (51) and MCL1 expression is frequently observed in undifferentiated follicular and papillary tumors with low expression of Bax (52). Perhaps, not surprisingly, indirect targeting of MCL1 with sorafenib may sensitize ATC cells to stimuli that trigger programmed cell death (21). In the absence of additional data it is tempting to speculate that MCL1 may be linked to the pathogenesis of ATC and may represent a novel “actionable” scaffold for targeted therapeutic intervention in a subset of ATC patients. Thus, MCL1 may be a bona fide target in ATC beyond the application of S/Q treatment.

In contrast to immunohistochemical data on prostate cancer (25), overexpression of STAT3pY was not apparent in ATC tissue (Supplementary Fig. S2B and S2C). At least in some cases more frequent staining for STAT3pY was seen in the non-neoplastic thyroid compared with ATC. ATCs may potentially also express lower levels of STAT3 mRNA compared with non-neoplastic thyroid as evident from our qRT-PCR analysis (Supplementary Fig. S1D). As we had limited specimens future studies should repeat this analysis with a bigger sample size.

STAT3 has been suggested to be a tumor suppressor in the thyroid as papillary thyroid tumors that emerge in mice as a result of BRAFV600E mutation are significantly smaller when they retained the STAT3 gene (30). The same study demonstrated an inverse correlation between STAT3pY expression, tumor size, and metastasis. STAT3pY was frequently detected in endothelial cells of papillary thyroid cancer which is consistent with what we observe in non-neoplastic thyroid and ATC (Supplementary Fig. S2C and data not shown). It seems unlikely that STAT3pY is a rate-limiting therapeutic target in ATC. However, it may be plausible that the S/Q drug combination targeting STAT3 signaling in the tumor vasculature of ATC's may contribute to the overall improved response observed in our orthotopic ATC model (Fig. 6). Additional studies will have to address this therapeutic possibility more directly.

In summary, we show that the combination of sorafenib and quinacrine can trigger synergistic MCL1–dependent drug responses in ATC cells in vitro. This combination appears to have limited toxicity in mice and triggers potent antitumor responses in an orthotopic model for ATC. On the basis of our data we propose that sorafenib and quinacrine should be explored as a therapeutic modality for aggressive thyroid cancers in future clinical trials.

No potential conflicts of interest were disclosed.

Conception and design: J. Abdulghani, D. Goldenberg, N.K. Finnberg, W.S. El-Deiry

Development of methodology: J. Abdulghani, P. Gokare, J.-N. Gallant, T. Whitcomb, D. Goldenberg, N.K. Finnberg, W.S. El-Deiry

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P. Gokare, D.T. Dicker, T. Whitcomb, J. Derr, J. Liu, N.K. Finnberg, W.S. El-Deiry

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Abdulghani, P. Gokare, T.K. Cooper, J. Liao, J. Derr, J. Liu, N.K. Finnberg, W.S. El-Deiry

Writing, review, and/or revision of the manuscript: J. Abdulghani, P. Gokare, J.-N. Gallant, J. Derr, J. Liu, D. Goldenberg, N.K. Finnberg, W.S. El-Deiry

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Abdulghani, P. Gokare, D.T. Dicker, J. Derr, N.K. Finnberg, W.S. El-Deiry

Study supervision: D. Goldenberg, N.K. Finnberg, W.S. El-Deiry

Other (mouse surgeries): J.-N. Gallant

Other (provided training and expertise for procedures using animals):T. Whitcomb

Other (interpreted the immunohistochemial stains for the most biomarkers. Reviewed the manuscript and provided opinions): J. Liu

Other (co-corresponding author and co-principal investigator on study): N.K. Finnberg

The authors would like to thank Dr. Henry Crist, Trey Bruggeman, and Dr. Nicole C. Williams in identifying suitable patient FFPE specimens for analysis and critiquing IHC staining of said specimens.

This work was supported by grants to W.S. El-Deiry and a grant from Thyroid Cancer Survivors Inc. (ThyCa) and the American Thyroid Association (ATA; to N.K. Finnberg, grant #143317).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.