Pericytes Elicit Resistance to Vemurafenib and Sorafenib Therapy in Thyroid Carcinoma via the TSP-1/TGFβ1 Axis

The Surveillance, Epidemiology, and End Results (SEER) cancer registry reveals that advanced-stage papillary thyroid cancer (PTC) is less tractable to therapy than localized PTC, as implied by increasing mortality rates (1). BRAF^V600E is a potent regulator of the MAPK (e.g., ERK1/2) pathway which is highly mutated in human cancers, and has been identified as a crucial and common oncogene in PTC. This mutation is associated with loss of radioiodine avidity, higher rates of recurrence and metastases, poorer prognosis, and lower survival rates (2–4). Thus effective treatment of advanced PTC through BRAF^V600E would meet an urgent clinical need.

Vemurafenib is the first FDA-approved BRAF^V600E inhibitor for the treatment of BRAF^V600E-positive metastatic melanoma (5). Vemurafenib binds to the ATP binding cassette of mutated BRAF^V600E and inhibits its pathway through ERK1/2. It has been used against metastatic BRAF^V600E-PTC that is refractory to radioiodine, but response has been variable and relapse is common (6). Numerous mechanisms can promote resistance through bypass of pharmacologic BRAF^V600E inhibition, which enables rebound of ERK1/2 (7, 8). Another mechanism of resistance involves upregulation of proangiogenic molecules such as the tyrosine kinases (TK) in the microenvironment (9). BRAF^V600E-PTC growth is influenced by the tumor microenvironment, which is in turn altered by the tumor itself, leading to abnormal extracellular matrix (ECM) deposition and activation of angiogenic pathways (10, 11). Many of the processes involved in thyroid tumor growth and metastasis are mediated by signaling molecules downstream of activated TKs.

Another FDA-approved drug, sorafenib, inhibits both TKs and BRAF intracellular signaling. Sorafenib targets RET (including RET/PTC) to inhibit proangiogenic pathways such as VEGFR2 and PDGFRB (12), and is currently used to treat hepatocellular carcinoma, advanced renal carcinoma, and metastatic thyroid carcinoma (13). However, resistance to sorafenib can develop in the first-line setting in patients with thyroid carcinoma (12, 14, 15), and BRAF^V600E may impact treatment duration and promote resistance mechanisms to TK inhibitors.

Angiogenic factors confer survival advantages and are overexpressed in both tumor cells and blood vessels in PTC (16). The angiogenic microenvironment includes pericytes, which are heterogeneous stromal cell populations that are fundamental to vessel stabilization and maturation and express a range of angiogenic factors (e.g., PDGFRB, VEGF, etc.; ref. 17). Pericytes regulate paracrine communications between tumor cells and microvascular endothelial cells in the thyroid gland (18) and human tumors (17). In the tumor vasculature, pericytes protect endothelial cells from antiangiogenic therapies, and may be players in resistance to vascular and microenvironment targeting drugs. They likewise confer survival advantages to endothelial cells through the secretion of proangiogenic factors (19). However, pericyte depletion using genetic methods has led to more metastasis and enhanced epithelial-to-mesenchymal transition (EMT) in breast cancer models (20).

Thrombospondin-1 (TSP-1; THBS1 gene) is produced by tumor and stromal cell types and plays a fundamental role in regulating the angiogenic microenvironment as well as cell proliferation, adhesion, migration, and invasion, and angiogenesis (21). Its functional domains are crucial for cell–cell or cell–ECM interactions. It is enriched in the thyroid carcinoma microenvironment and plays an important role in tumor aggressiveness (22). TSP-1 is also a key regulator of latent TGFβ activation, the conversion of latent TGFβ to its biologically active form in certain diseases (23). TGFβ is an early tumor suppressor that is also a player in the metastatic switch of tumors, and promotes EMT and metastasis (24). TGFβ-induced SMAD phosphorylation and EMT induction required MAPK pathway activation in murine thyrocytes derived from BRAF^V600E mice, indicating that tumor initiation by BRAF^V600E predisposes murine thyroid cells to TGFβ-induced EMT, through a MAPK-dependent process (25).

The inefficacy of single-agent BRAF^V600E or TKs inhibitors, and the eventual resistance to these agents, even in combination, highlight the need for a better understanding of the tumor microenvironment, including the cross talk between pericytes and tumor cells. Therefore, in this study, we have analyzed the effects of combined vemurafenib plus sorafenib therapy in BRAF^V600E PTC patient-derived cells using cell cultures and in vivo models. Our results demonstrate that pericyte-derived secretomes increase pERK1/2, pAKT, and pSMAD3 levels in thyroid tumor cells to overcome the inhibitory effects of vemurafenib and sorafenib either alone or in combination. Pericyte-derived factors also increased survival of BRAF^V600E tumor cells and refractoriness to tumor cell death. We demonstrate that pericytes are a source of both TSP-1 and TGFβ, and that antagonism of TSP1-dependent activation of latent TGFβ1 overcomes resistance to BRAF^V600E inhibitors or TKI. Together, these data provide evidence. Thus, pericytes elicit resistance to vemurafenib and sorafenib therapy via the TSP-1/TGFβ1 axis, suggesting this axis as a promising new target in overcoming therapy resistance.

We used authenticated (short tandem repeat and DNA sequencing for KTC1; DNA sequencing and RT-PCR for TPC1) KTC1 (BRAF^WT/V600E) and TPC1 (BRAF^WT/WT) human thyroid carcinoma cell lines, and human pericytes (BRAF^WT/WT) were obtained from Promo Cell (18). The use of these cell lines was approved from the committee on microbiological safety [COMS, Beth Israel Deaconess Medical Center (BIDMC), Boston, MA]. KTC1 is a spontaneously immortalized human thyroid carcinoma cell line that harbors BRAF^WT/V600E mutation. It was established from the metastatic pleural effusion from recurrent and radioiodine (RAI)-refractory PTC in a 60-year-old male patient (26) by Dr. J. Kurebayashi (Department of Breast and Thyroid Surgery Kawasaki Medical School Kurashiki, Japan) and provided by Dr. Rebecca E. Schweppe (University of Colorado, Aurora, CO).

For our in vitro assays, we used 10 mmol/L vemurafenib (PLX4032, RG7204, catalog no. S1267) (Selleckchem) dissolved in 100% dimethyl sulfoxide (DMSO, vehicle). Sorafenib tosylate (catalog no.S1040, Selleckchem), a multikinase inhibitor, was dissolved in 100% DMSO (Sigma) according to the manufacturer's instructions to produce 10 mmol/L stock solution. Intermediate doses of vemurafenib or sorafenib were prepared in 100% DMSO and diluted in 0.2% fetal bovine serum (FBS) DMEM to achieve desired final concentrations, maintaining a constant final concentration at 2% DMSO for optimal solubility (see Supplementary Methods). Synergy, subadditive, or additive activity for the combined treatments of vemurafenib plus sorafenib were estimated using GeoGebra Classic and applying Loewe test method according to Tallarida (27) to assess drug synergy and antagonism. Cells were treated for 48 hours in the presence of 0.2% FBS DMEM at final 2% DMSO with: 1, 2.5, 5, or 10 μmol/L of either vemurafenib or sorafenib; or combined therapy with vemurafenib plus sorafenib combining all above doses. Vehicle was used as untreated control (2% DMSO diluted in 0.2% FBS DMEM). Before adding treatments, cells were washed with PBS from 10% FBS DMEM. Quantitative analysis was performed by crystal violet assays (see Supplementary Methods) of adherent cells (magnification: 10×). Vehicle (control) was 2% DMSO diluted in 0.2% FBS DMEM.

Peptide SRI31277 (24) was synthesized by BioMatik and purity confirmed at Southern Research. We reconstituted the peptide in 0.2% FBS DMEM) to achieve the stock concentration of 2.6 mmol/L. SRI31277 was diluted in 0.2% FBS DMEM to achieve final concentration of 1 μmol/L, 2.5 μmol/L, 5 μmol/L, 10 μmol/L, 25 μmol/L, 50 μmol/L, or 100 μmol/L.

Pericytes were seeded at about 90% confluence in 6-well dishes in DMEM growth medium supplemented with 10% FBS. Forty-eight hours following cell seeding, pericytes were treated for 5 hours with 10 μmol/L vemurafenib, 2.5 μmol/L sorafenib, combined therapy with 10 μmol/L vemurafenib plus 2.5 μmol/L sorafenib, or vehicle (2% DMSO) in the presence of 0.2% FBS DMEM growth medium. Following treatment, the 0.2% FBS DMEM cell growth medium enriched by cell-derived secreted protein factors was defined as secretome and was normalized to the same cell growth medium to subtract background; then, it was collected and separated from dead cell debris by short spin. We collected an aliquot of secretome volume for ELISA analysis. In addition, the remaining volume of all four secretomes was used to treat BRAF^WT/V600E-KTC1 and BRAF^WT/WT-TPC1 for 5 hours. At the same time, another condition included BRAF^WT/V600E-KTC1 and BRAF^WT/WT-TPC1 cells (both cell lines were seeded at 90% confluence in the presence of 10% FBS DMEM growth medium the day prior to treatments) directly treated (without pericyte secretome) for 5 hours with 10 μmol/L vemurafenib, 2.5 μmol/L sorafenib, combined therapy with 10 μmol/L vemurafenib plus 2.5 μmol/L sorafenib, or vehicle (2% DMSO) in the presence of 0.2% FBS DMEM growth medium. Also, after secretome collection, adherent pericytes were lysed for protein extraction to perform Western blotting assays. After treatment of the BRAF^WT/V600E-KTC1 and BRAF^WT/WT-TPC1 thyroid tumor cells, we collected an aliquot of secretome volume for ELISA analysis (Supplementary Materials and Methods). Then we lysed the adherent thyroid tumor cells for protein extraction in order to perform Western blotting assays.

mCherry-KTC1 and pericytes were seeded at 1.5 × 10⁵ cells (1:1) per well in 6-well dishes or 3 × 10⁴ cells per well in 24-well dishes. Forty-eight hours after cell seeding, cells were treated for 48 hours with 10 μmol/L vemurafenib, 2.5 μmol/L sorafenib, combined therapy with 10 μmol/L vemurafenib plus 2.5 μmol/L sorafenib, or vehicle (2% DMSO) in the presence of 0.2% FBS DMEM growth medium. After 48 hours, the 0.2% FBS DMEM cell growth medium enriched by cell-derived secreted protein factors was defined as secretome and was normalized to the same cell growth medium to subtract background, then it was collected for ELISA analysis (Supplementary Materials and Methods) and separated from dead cell debris by short spin. Adherent cells were fixed with 10% formalin for 20 minutes at room temperature. Cells were then washed with PBS. Cells were stained with 5 μmol/L Hoechst 33342 (Nexcelom) diluted in PBS for 15 minutes. Total number of cells (cell growth)/well was analyzed by Celigo image cytometer (Nexcelom). Data were plotted as matrix of cell count using both GraphPad Prism 6 and excel software.

Cells were grown in 10-cm dishes in 10% FBS DMEM. They were treated with vehicle (DMSO), vemurafenib, sorafenib, or combined vemurafenib plus sorafenib therapy in the presence of 0.2% FBS DMEM when reached about 90%–100% confluence. Western blotting assays were performed according to standard procedure (11). The intensity of each protein band was normalized to housekeeping protein band (tubulin or actin) and quantified by densitometry analysis (ImageJ software).

All animal work was approved and done in accordance with federal, local, and institutional guidelines (IACUC) at the BIDMC (Boston, MA). Human metastatic KTC1 tumor-derived cells harboring the heterozygous BRAF^V600E mutation and engineered to express luciferase were cultured in 10-cm dishes and grown in DMEM supplemented with 10% FBS, penicillin, streptomycin, and amphotericin at 37°C with 5% CO₂ atmosphere. Prior to implantation, cells were trypsinized, gently centrifuged, and suspended in serum-free DMEM growth medium to achieve a cell suspension concentration ranged between 3.5 × 10⁶ and 5 × 10⁶ cells/10 μL. The cells were kept on ice until implantation. KTC1 cells were orthotopically injected in the right thyroid of 9-week-old male NSG mice (strain name: NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ; stock number: 005557; n = 5 per group) according to our previous experimental procedures (22). Mice were randomly divided into four groups of 5 for the purpose of establishing a timeline of tumor cell growth and response to therapy with vemurafenib, sorafenib, combined therapy (vemurafenib plus sorafenib), or vehicle. Treatments were started 6 weeks after KTC1 tumor cells implantation and performed for 5 weeks.

For in vivo studies, drug suspensions were prepared for vemurafenib (10 mg/mL in 2% hydroxypropylcellulose) and sorafenib (3.75 mg/mL according to Fendrich and colleagues; ref. 28). Freshly prepared drug suspensions were stored at 4°C and used within 48 hours. Mice were dosed once daily with vehicle alone (control), vemurafenib (100 mg/kg), sorafenib (30 mg/kg), or combination of vemurafenib and sorafenib as indicated by oral gavage using a 22G needle.

To determine the association of a select set of 23 genes linked to extracellular matrix functions, pericyte functions, angiogenesis, cell growth, adhesion/migration/invasion, and metastasis pathways with BRAF mutational status, we performed analysis on the genes PTC TCGA data. We downloaded RNA-seq data of PTC from TCGA to analyze 23 genes differentially expressed using 211 BRAF^WT/V600E-PTC, 23 PTC harboring BRAF^WT/V600E and hTERT mutations, and 256 BRAF^WT/WT-PTC samples. After performing analysis on the 23 selected genes, we considered only those with raw P values < 0.05 and fold-change (FC) ≥1.2 or ≤−1.2 as significantly associated with BRAF mutational status. Network analysis was also performed. More details are reported in the Supplementary Methods. Pericyte abundance analysis was assessed by the ssGSEA algorithm using RNAseq expression data from NT and PTC TCGA samples. SSGSEA calculates separate enrichment scores for each pairing of a sample and gene set; each enrichment score represents the degree to which the genes in a particular set are coordinately upregulated or downregulated. We specified the positive signature for pericytes based on the expression of NG2, PDGFRB, αSMA, and CD90 genes. To identify samples enriched with pericytes, we used the genes PECAM1, LYVE1, and CD34 as negative signature. On the basis of score differences, we ranked samples from the highest enrichment of pericyte signature to the least enrichment.

Statistical analysis was carried out using GraphPad Prism 6 software, Microsoft Excel, and GeoGebra Classic statistical tools. χ²test, Student t test, Mann–Whitney test, one-way ANOVA for multiple comparisons tests, and Pearson correlation analysis were used. Data are reported as the averaged value, and error bars represent the SD of the average for each group. Results with P values below 0.05 were considered statistically significant.

We also used virus transduction assays (for gene overexpression or knockdown); gene-regulatory networks/pathway analyses; and TEM (for more details, see the Supplementary Methods).

As a first step toward suppressing PTC cell survival, we assessed anti-BRAF^V600E (vemurafenib) and anti-multi-TK (sorafenib) combined therapy in heterozygous BRAF^WT/V600E (KTC1) or BRAF^WT (TPC1) tumor cells derived from invasive PTC (Fig. 1A). We have previously characterized the PTC microenvironment, which has pericytes (11), crucial components of the vasculature known to express TKs (e.g., PDGFRB), as well as other markers such as NG2 and αSMA (17). Our results confirmed these findings (Fig. 1B), and human pericytes were also negative for mesenchymal and endothelial cell markers (Fig. 1B). Furthermore, transmission electron microscopy (TEM) revealed that human well-differentiated thyroid carcinoma tissue were characterized by endothelial cells and pericytes with a large nucleus and little cytoplasm as compared with the normal thyroid (NT) tissue (Fig. 1C). Pericytes may impact resistance in cancer by regulating TK-dependent angiogenic signaling pathways. We therefore analyzed the expression of two major TKs (PDGFRB and VEGFR2) in PTC-derived cells and human pericytes (Fig. 1D). Pericytes showed 3.7-fold and 5.1-fold changes in phospho(p)-PDGFRB-Y751 levels upon PDGFB treatment, and 1.1-fold and 1.38-fold changes in pVEGFR2-Y1059 levels upon VEGFA treatment when compared with KTC1 and TPC1 cells, respectively (Fig. 1D). PDGFB and VEGFA stimulated phosphorylation of PDGFRB and VEGFR2 in pericytes and BRAF^WT/V600E-PTC, but not in BRAF^WT/WT-PTC, which showed low levels of these receptors (Fig. 1D). PDGFB stimulated VEGFR2 phosphorylation more than VEGFA did, possibly due to potential PDGFRB/VEGFR2 heterodimers; however, further studies are needed to define this phenomenon. To determine whether simultaneous inhibition of BRAF^V600E and TKs was effective against tumor cells, we combined vemurafenib and sorafenib. Previous studies had identified the dose–response curve for vemurafenib (IC₅₀, 50% maximal inhibitory concentration) in thyroid cancer cells (11); to assess the most effective doses of combined vemurafenib plus sorafenib, we treated tumor cells for 48 hours with seven different drug dose combinations (Supplementary Fig. S1). We used isobolographic analysis (Fig. 1E and F) to assess synergy or additivity. Our results showed that 10 μmol/L vemurafenib plus 2.5 μmol/L sorafenib had the highest therapeutic efficacy (synergistic effect) against BRAF^WT/V600E-PTC cells (Fig. 1E), but was subadditive in BRAF^WT/WT-PTC cells (Fig. 1F) and pericytes (Supplementary Fig. S2A). Specifically, the targeted therapy achieved a significantly lower viability (31%, 46%, and 49% by vemurafenib, sorafenib, and combined therapy, respectively) in BRAF^WT/V600E-PTC cells than in pericytes (Supplementary Fig. S2B). Similarly, combined therapy significantly decreased BRAF^WT/V600E-KTC1 cell viability compared with vehicle (60% decrease), vemurafenib (31% decrease), or sorafenib (20% decrease; Supplementary Fig. S1). In BRAF^WT/WT-TPC1 cells, combination therapy yielded a 23% decrease in cell viability compared with vehicle, 18% decrease compared with vemurafenib, and a 52% decrease compared with sorafenib (Supplementary Fig. S3). We also found significant induction of cell death in PTC-derived cells when we used combined therapy, as shown by cell death analysis (Fig. 1G). Combined therapy was significantly more effective than vehicle (21-fold increase), vemurafenib (7.3-fold increase), or sorafenib (10.8-fold increase) in inducing death in BRAF^WT/V600E-KTC1 cells (Fig. 1G). Also, vemurafenib (2.2-fold change increase) or combined therapy (2.1-fold change increase) induced significantly higher rate of cell death in BRAF^WT/V600E-KTC1 than BRAF^WT/WT-TPC1 cells, indicating the higher specificity of vemurafenib in targeting BRAF^WT/V600E PTC cells versus BRAF^WT/WT PTC cells. In contrast, sorafenib was more effective on BRAF^WT/WT-TPC1 cells compared with BRAF^WT/V600E-KTC1 (0.51-fold change).

Furthermore, analysis of cell proliferation by BrdU (5-bromo-2-deoxyuridine) assay showed that targeting BRAF^V600E by vemurafenib and sorafenib significantly inhibited DNA synthesis of PTC cells compared with vehicle or single agents. This effect was likely driven by combined therapy in BRAF^WT/V600E-KTC1 cells, but by sorafenib alone in BRAF^WT/WT-TPC1 cells (Fig. 1H and I). Significantly, high doses of vemurafenib were also required to inhibit melanoma cell viability (29). Since BRAF^V600E is a strong regulator of tumor cell migration (30), we applied a monolayer wound-healing (cell motility) assay to study the effects of cell migration during treatment (Fig. 1J and K). We measured the initial wound at baseline (time zero), treated the cells with vehicle, vemurafenib, sorafenib, or a combination of the two for 7 hours (a time point prior to any observed effects on cell proliferation), and then quantified healing area at 7 hours versus baseline in each condition (Fig. 1J and K). Importantly, BRAF^WT/V600E-KTC1 cells displayed a significant decrease in cell motility upon combination treatment as compared with vehicle at 7 hours (Fig. 1J and K). In addition, vemurafenib treatment caused a decrease in cell motility compared with vehicle, while sorafenib did not (Fig. 1J and K). In summary, combined therapy blocked tumor cell proliferation, increased cell death, and decreased motility in BRAF^WT/V600E-PTC cells, likely via BRAF^V600E inhibition.

We have developed the first interventional preclinical mouse trial of vemurafenib therapy in BRAF^WT/V600E-PTC patient-derived cells (Fig. 2A). Immunocompromised mice were orthotopically implanted with human KTC1 cells derived from recurrent BRAF^V600E-positive PTC, and engineered to express luciferase (Fig. 2B). Orthotopic tumors developed in all mice and were analyzed 6 weeks after injection as baseline (Fig. 2B). Mice were then randomized for treatment with vehicle, vemurafenib, sorafenib, or combined therapy with vemurafenib plus sorafenib. All tumors in the vehicle-treated mice exhibited a 3-fold increase in growth over baseline at week 5 (Fig. 2C). The vemurafenib dosage was similar to that of other studies (31); here we found that only vemurafenib resulted in consistent reduction in tumor growth, likely due to cytostatic effects, with 61% significant reduction in tumor growth at week 4 and 48% reduction at week 5 as compared with vehicle-treated mice (Fig. 2B and C). Therapeutic response to sorafenib was fluctuant and resulted in a smaller reduction (23%) in tumor growth at week 4 and increased reduction (59%) at week 5 as compared with vehicle (Fig. 2B and C). Combined therapy followed the pattern of sorafenib activity and yielded a 5.1-fold increase (36%) in tumor growth at week 4 versus vehicle, while a reduction (43%) in tumor growth occurred at week 5 versus vehicle (Fig. 2B and C). We observed no apparent toxic side effects upon either single-agent treatment or combined therapy. Because inhibition of BRAF^V600E may redifferentiate thyroid tumor cells (32), we analyzed this phenomenon by microSPECT/CT imaging. Vemurafenib-treated mice at 5 weeks posttreatment showed >2-fold increase in ^99mTc uptake, suggesting that targeting BRAF^V600E could block not only tumor growth but also induce thyroid tumor redifferentiation more substantially than sorafenib (Fig. 2D and E).

We have hypothesized that pericytes, which are fundamental to vessel maturation (33), are also fundamental to thyroid tumor cell viability, and limit the efficacy of BRAF^V600E inhibitors and TKI. To test this hypothesis, we developed an experimental model using secretome derived from human pericytes treated for 5 hours with either vehicle, vemurafenib, sorafenib, or combined therapy. Pericytes were grown in medium with low concentration of FBS (i.e., 0.2%) during treatment. The 0.2% FBS cell growth medium enriched by pericyte-derived secreted factors within 5 hours of drug or vehicle treatment was defined as the secretome and was normalized to the same cell growth medium to subtract background. A multiplex ELISA assay (Supplementary Materials and Methods) including the most important cytokines and angiogenic factors showed no changes in secretion levels upon drug treatment (Supplementary Fig. S4). Intracellular TSP-1 protein expression in pericytes was upregulated within 5 hours of drug treatments (2.9-fold change with vemurafenib, 1.6-fold change with sorafenib, and 1.7-fold change with combined therapy vs. vehicle; Fig. 3A). In contrast, levels of secreted TSP-1 either fell or remained unchanged after treatment (12.1% decrease with vemurafenib, unchanged with sorafenib, and 15.3% decrease with combined therapy vs. vehicle; ELISA assay, Supplementary Materials and Methods; Fig. 3B). Levels of secreted TGFβ1 likewise fell or remained unchanged after treatment (17.8% decrease with vemurafenib, unchanged with sorafenib, and 27.3% decrease with combined therapy vs. vehicle; ELISA assay, Supplementary Materials and Methods; Fig. 3B). TSP-1 is a mediator of TGFβ1 activation, which regulates many cell functions through SMAD, ERK1/2, AKT proteins (23, 24). Moreover, we found that TKs levels were affected by the targeted therapy, that is, PDGFRB phosphorylation levels were reduced upon drug treatment (43.6% decrease upon vemurafenib, 70.1% decrease upon sorafenib, and 58.8% decrease upon combined therapy) versus vehicle (Fig. 3A). pVEGFR2 protein decreased upon treatment with vemurafenib (20.0%) or sorafenib (22.4%), however, no substantial changes were observed upon combined therapy (Fig. 3A). Protein expression of prosurvival factors likewise fell in pericytes after drug treatment versus vehicle treatment: (i) pAKT (94.5% decrease upon vemurafenib, 80.2% decrease upon sorafenib and 93.9% decrease upon combined therapy), (ii) pERK1/2 (15.7% decrease upon vemurafenib, 74.8% decrease upon sorafenib and 82.7% decrease upon combined therapy), and (iii) pSMAD3 (7.7% increase upon vemurafenib, 15.6% increase upon sorafenib, and 21.1% decrease upon combined therapy; Fig. 3A).

It is known that BRAF^V600E inhibitors such as vemurafenib selectively inhibit MAPK signaling (e.g., ERK1/2) in BRAF^WT/V600E thyroid tumor cells (32, 34, 35). To investigate the effects of pericytes on PTC-derived KTC1 or TPC1 cells, we treated these thyroid tumor cells for 5 hours with vemurafenib, sorafenib, or combined therapy in the presence or absence of the pericyte-derived conditioned medium (secretome) containing secreted TSP-1 and TGFβ1 (Fig. 3C). Importantly, pericytes and BRAF^WT/V600E-KTC1 cells were substantially responsive to treatment with exogenous recombinant human latent TGFβ1 protein (while response in BRAF^WT/WT-TPC1 cells was less robust), which upregulated pSMAD3 protein levels (Fig. 3D), suggesting the presence of endogenous regulators of latent TGFβ1. When we used a TSP-1 antagonist (i.e., SRI31277) derived from the LSKL sequence of latent TGFβ1 that blocks TSP1-mediated TGFβ1 activation (24) plus latent TGFβ1, we found pSMAD3 protein expression decreased 19% in pericytes versus latent TGFβ1 treatment alone (Fig. 3D).

Because TSP-1 is also a key-player in aggressive anaplastic thyroid carcinoma (ATC) harboring BRAF^V600E (22), we analyzed TSP-1 protein levels in aggressive PTC-derived cells. Interestingly, BRAF^WT/WT-TPC1 cells showed low intracellular TSP-1 protein levels (Supplementary Fig. S5); also, secreted TSP-1 levels (as well as TKs proangiogenic factors, i.e., VEGFR2 or PDGFRB levels) were substantially lower (∼20–50 folds) in BRAF^WT/WT-TPC1 cells compared with BRAF^WT/V600E-KTC1 cells across all treatments (Supplementary Fig. S6A), suggesting expression of these factors might depend on the BRAF^V600E pathway. Indeed, TSP-1 protein expression was downregulated by direct drug treatments (without the presence of pericyte secretome) compared with vehicle in BRAF^WT/V600E-KTC1 cells, i.e., 15.5% by vemurafenib, 32.4% by sorafenib and 51.5% by combined therapy (Fig. 3E). Importantly, direct treatment by combined therapy more effectively downregulated both TSP-1 (51.5% vs. vehicle), and the intracellular signaling effectors pERK1/2 (88% vs. vehicle), pAKT (58.8% vs. vehicle), and pSMAD3 (33.8% vs. vehicle), as well as TKs pVEGFR2 (38.2% vs. vehicle) and pPDGFRB (9.5% vs. vehicle) than single agents in BRAF^WT/V600E-KTC1 cells (Fig. 3E). Vemurafenib treatment upregulated pERK1/2, as expected, in BRAF^WT/WT-TPC1 cells (Supplementary Fig. S5), likely due to paradoxical effects (36).

Because our mouse data suggested that kinase and angiogenesis inhibitors such as vemurafenib or sorafenib elicited cytostatic effects with differing levels of pharmacologic action (Fig. 2B and C), probably due to angiogenic microenvironment-mediated effects, we focused our attention on pericytes, denizens of the tumor microenvironment, which are critical to vessel stabilization and angiogenic endothelial functions (33). To understand the functional role of pericytes in paracrine communication with PTC cells, we assessed the ability of conditioned media (secretome) derived from pericytes to influence thyroid tumor cell intracellular signaling upon vehicle, vemurafenib, sorafenib, or combined treatment. As all pericyte secretome was collected within 5 hours of treatment, it was unlikely to have been produced during deregulation of pathways related to cell death. Across all treatments, the presence of pericyte secretome as compared to no pericyte secretome (Fig. 3E) consistently and substantially increased levels of: (i) pERK1/2 (2.9-fold change with vehicle, 6.9-fold change with vemurafenib, 4.5-fold change with sorafenib, and 4.0-fold change with combined therapy); (ii) pAKT (1.7-fold change with vehicle, 11-fold change with vemurafenib, 3.5-fold change with sorafenib and 8.6-fold change with combined therapy); (iii) pSMAD3 (0.9-fold change with vehicle, 1.2-fold change with vemurafenib, 1.4-fold change with sorafenib and 2-fold change with combined therapy); (iv) TSP-1 (1.6-fold change with vehicle, 1.4-fold change with vemurafenib, 1.9-fold change with sorafenib and 1.6-fold change with combined therapy); and (v) pVEGFR2 (1.02-fold change with vehicle, 1.9-fold change with vemurafenib, 1.6-fold change with sorafenib and 1.1-fold change with combined therapy) in BRAF^WT/V600E-KTC1 cells (Fig. 3E). As a result, pericytes provided significant growth advantages to BRAF^V600E-KTC1 cells, even when treated with vehicle (2.1-fold change compared with BRAF^WT/V600E-KTC1 tumor cells not cocultured with pericytes; Fig. 3F and G). These results may be linked to the ability of BRAF^V600E-KTC1 cells when stimulated by pericyte secretome (5 hours) to significantly increase secretion of TSP-1 (150.2% with vehicle, 112.5% with vemurafenib, 104.9% with sorafenib, and 81.1% with combined therapy) and TGFβ1 (20.7% with vehicle, 83.9% with vemurafenib, 38.8% with sorafenib, and 47.2% with combined therapy) compared with the BRAF^WT/V600E-KTC1 cells without pericyte secretome (Fig. 3H). Pericyte secretome promoted a moderate rebound of pERK1/2 (but not when treated with vemurafenib), pAKT, and pSMAD3 in BRAF^WT/WT-TPC1 cells upon treatment with vehicle or drugs (Supplementary Fig. S5).

Furthermore, we used one of the most efficient short hairpin RNA (shRNA; ref. 22) to knockdown TSP-1 in pericytes (Fig. 3I), downregulating TSP-1 protein levels by more than 50%, and reducing secreted TSP-1 levels even more robustly upon drug treatment (Supplementary Fig. S6B). Importantly, knockdown of TSP-1 (by shTSP-1) in pericytes substantially reduced the capability of the shTSP-1 pericyte secretome (compared with shGFP pericyte secretome, control) to trigger rebound of pERK1/2 (but not with combined therapy), pAKT, and pSMAD3 levels in BRAF^WT/V600E-KTC1 tumor cells (Fig. 3J). Also, secretion of TSP-1 decreased by 12.8% in BRAF^WT/V600E-KTC1 cells in the presence of the combined therapy–treated shTSP-1 pericyte secretome compared with shGFP secretome alone (Supplementary Fig. S6C). ShTSP-1 pericyte secretome did not have substantial additive effects in combination with drug treatments in suppressing pERK1/2 (except with sorafenib, 29%) and pAKT levels, and exerted moderate effects in downregulating pSMAD3 (22.5% with vemurafenib) in BRAF^WT/WT-TPC1 cells (Supplementary Fig. S7), with no changes observed in the secreted levels of TSP-1 (only a moderate decrease with sorafenib, Supplementary Fig. S6D), suggesting that BRAF^WT/WT-PTC cells may have a different TSP-1–regulatory pathway than BRAF^WT/V600E-PTC cells.

Overall, these results demonstrate that pericyte-derived secretome (e.g., TSP-1, TGFβ1) induces the rebound of prosurvival and proangiogenic factors and overcomes the inhibitory effects of targeted therapy in BRAF^WT/V600E-PTC cells, and ultimately contributes to an increase in BRAF^WT/V600E-thyroid tumor cell survival.

The inhibition of TGFβ1 activation is a therapeutic strategy against cancer (24). We used an antagonist (i.e., SRI31277; ref. 24) that blocks TSP1-mediated TGFβ1 activation in the extracellular environment to determine the role of the TSP-1/TGFβ1 pathway in BRAF^WT/V600E-KTC1 cells using pSMAD3 protein expression. We assessed dose–response (IC₅₀) for the SRI31277 peptide in BRAF^WT/V600E-KTC1 or BRAF^WT/WT-TPC1 cells, and in pericytes treated with a matrix of different doses (Supplementary Fig. S8A–S8C). Our results showed that compared with vehicle, 10 μmol/L SRI31277 provided a significant therapeutic effect against PTC cells and pericytes, reducing cell viability by 3.1-fold in BRAF^WT/V600E-KTC1, 4.1-fold in BRAF^WT/WT-TPC1, and 2.6-fold in pericytes (Supplementary Fig. S8A–S8C). SRI31277 upregulated pSMAD3 protein levels in pericytes within 5 hours upon treatment with vehicle (37%) or vemurafenib (4.5%), and downregulated its levels in the presence of sorafenib (4.7%) or combined therapy (27%; Fig. 4A). Also, SRI31277 treatment substantially reduced both secreted TSP-1 (11%) and TGFβ1 (25.5%) in pericytes treated with combined therapy (Fig. 4B). Importantly, in BRAF^WT/V600E-KTC1 cells treated with pericyte secretome the pharmacologic antagonism of TSP-1 by SRI31277 treatment substantially downregulated protein levels of pERK1/2 (23.4% with vehicle, 17.9% with vemurafenib, 37.7% with sorafenib, and 8.2% with combined therapy) and pSMAD3 (30.1% with vehicle, 27.4% with vemurafenib, 14.8% with sorafenib and 24.4% with combined therapy; Fig. 4C, right). Intriguingly, we found that SRI31277 upregulated both pERK1/2 (100% with vehicle, 101% with vemurafenib, 24% with sorafenib, and 205% with combined therapy) and pSMAD3 (34.8% with vehicle, 21.1% with vemurafenib, 40.3% with sorafenib, and 25% with combined therapy) in the absence of pericyte secretome compared with vehicle in BRAF^WT/V600E-KTC1 cells (Fig. 4C, left). SRI31277 less robustly affected pERK1/2 and pSMAD3 levels in BRAF^WT/WT-TPC1 cells treated with pericyte secretome (Supplementary Fig. S9). Secreted levels of both TSP-1 (5% with vehicle, 58.6% with vemurafenib, 26.7% with sorafenib, and 52.3% with combined therapy), and TGFβ1 (17.1% with vehicle, 45.6% with vemurafenib, 27.9% with sorafenib, and 32% with combined therapy conditions) increased in BRAF^WT/V600E-KTC1 cells cultured with pericytes for 48 hours compared with BRAF^WT/V600E-KTC1 cells in single culture (Fig. 4D). Importantly, the pharmacologic antagonism of TSP-1 by SRI31277 combined with BRAF^V600E inhibition (vemurafenib) significantly reduced secretion of TSP-1 and TGFβ1 in BRAF^WT/V600E-KTC1 cells cocultured with pericytes compared to the KTC1 cell coculture without SRI31277 treatment (Fig. 4D). More importantly, the antagonism of TSP-1 by SRI31277 significantly overcame therapeutic resistance of BRAF^WT/V600E-KTC1 cell tumor growth inhibition to either vemurafenib (67.1% reduction vs. no SRI31277), sorafenib (63.1% reduction vs. no SRI31277), and combined therapy (66.5% reduction vs. no SRI31277; Fig. 4E). We measured within 5 hours of treatment, the concentration of endogenous active TGFβ1 by ELISA in pericytes and KTC1 cells (Supplementary Materials and Methods). Active TGFβ1 represented a small fraction of the total TGFβ1 (reported in Fig. 3B and H). Specifically it was 21.8 pg/mL and 14.5 pg/mL with vehicle; 14.4 pg/mL and 6.6 pg/mL with vemurafenib, 22.6 pg/mL and 10.6 pg/mL with sorafenib; and 13.5 pg/mL and 6.8 pg/mL with combined therapy in BRAF^WT/V600E-KTC1 cells and pericytes, respectively. Finally, to overcome SRI31277 antagonism upon TGFβ1 activation by TSP-1, we treated BRAF^WT/V600E-KTC1 cells in pericytes coculture using the recombinant human active TGFβ1 (Fig. 4F). Importantly, cell stimulation by recombinant human active TGFβ1 protein significantly rescued BRAF^WT/V600E-KTC1 tumor cell growth in the coculture with pericytes within 48 hours of treatment with SRI31277 plus sorafenib (1.34-fold, P = 0.0075). After treatment with SRI31277 plus combined therapy, TGFβ1 produced an even more robust rescue effect (8.7-fold, P = 0.0044; Fig. 4F). Stimulation with recombinant human active TGFβ1 was also effective in significantly enabling BRAF^WT/V600E-KTC1 tumor cell growth in the absence of SRI31277 treatment (1.4-, 3.1-, and 2.6-fold change upon vehicle, vemurafenib and combined therapy, respectively).

Taken together, these results indicate BRAF^WT/V600E-KTC1 cells elicit paracrine signals in synergy with pericytes, which contribute to tumor survival (Fig. 4E and F) via the TSP-1/TGFβ1 axis, and that pericytes trigger resistance to the targeted therapy.

Our in vivo mouse results (Fig. 2B and C) did not confirm the unique and synergistic effect of combined therapy with vemurafenib and sorafenib observed in our in vitro models (Fig. 1E–G), indicating the likely elicitation of drug resistance by the tumor microenvironment. Therefore, to assess the potential mechanisms of the apparent resistance to this targeted therapy, we have performed IHC on the orthotopic thyroid tumors. We found luciferase expression substantially decreased in orthotopic mouse tumors (Fig. 2B and C) after 5 weeks of treatment with either vemurafenib, sorafenib, or combined therapy (Fig. 5A). To corroborate these results, we analyzed the expression level of PAX8, a thyroid-specific marker, to identify BRAF^WT/V600E human thyroid tumor cells (11), which matched the downregulation of the luciferase marker in drug-treated mice (Fig. 5A). We also used terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) to quantify cell death. Importantly, we found that vemurafenib moderately increased cell death in a subpopulation of tumor cells (Fig. 5A), whereas sorafenib treatment was less effective, as confirmed by in vitro results using the Annexin V/PI assay (Fig. 1G). Combined therapy with vemurafenib and sorafenib was found to induce cell death more effectively than single agents in a subpopulation of tumor cells (Fig. 5A), indicating that either BRAF^V600E inhibitor (i.e., vemurafenib) or TKI (i.e., sorafenib) abrogates BRAF^WT/V600E-PTC through complex mechanisms of action, ultimately resulting in cytostatic effects likely due to the presence of different clones with intrinsic primary resistance (34). BRAF^WT/V600E stimulates PTC cell proliferation (Fig. 1H), and affects the expression of TSP-1 and TGFβ1-dependent pSMAD3 and pERK1/2 (Figs. 3E and 4C), suggesting that this oncogene promotes thyroid tumor aggressiveness via the TSP-1/TGFβ1 axis. TSP-1 can inhibit VEGF-stimulated VEGFR2 phosphorylation in microvascular endothelial cells and block angiogenesis (37). Because our in vitro data suggested a role for TSP-1 in the paracrine communication between BRAF^WT/V600E-PTC cells and pericytes, we assessed the expression of the TSP-1/TGFβ1 axis and proangiogenic factors in vivo in the angiogenic microenvironment of orthotopic BRAF^WT/V600E-PTC treated with vemurafenib, sorafenib, combined therapy, or vehicle (Fig. 5B). Combined therapy reflected the action of both vemurafenib and sorafenib in downregulating cytosolic TSP-1 and TGFβ1 in tumor cells more than 50% compared with vehicle or single agents. Also, combined therapy reduced protein expression levels of prosurvival factors such as pAKT and tumor growth-related molecules such as pERK1/2. Critically, no inhibition (specifically by vemurafenib and combined therapy) of the TSP-1/TGFβ1 axis was observed in the vascular compartment, including endothelial cells and pericytes (Fig. 5B), suggesting that stromal vascular cells in the microenvironment can elicit drug resistance and provide advantages to PTC growth. With reduction of TSP-1 and TGFβ1 in tumor cells, we also found a substantial decrease (∼2.5-fold change) in pSMAD3 expression involved in the TGFβ1 pathway in orthotopic thyroid tumor cells after treatment with vemurafenib or combined therapy, with no associated changes in the vascular/endothelial compartment (Fig. 5B). In addition, vemurafenib substantially reduced pERK1/2 and pAKT levels in tumor cells but not in the vascular/endothelial compartment (Fig. 5B). In contrast, sorafenib treatment alone was ineffective in suppressing the expression of these intracellular signaling targets in vivo in both tumor and vascular cells (Fig. 5B). Vemurafenib (but not sorafenib) was ineffective at downregulating markers of vascular density (e.g., CD31; Fig. 5C), suggesting that vemurafenib specifically targets BRAF^V600E-thyroid tumor cells but not BRAF^WT-stromal cells, and also that possibly subpopulations of stromal vascular cells elicit resistance to vemurafenib. Therefore, the tumor-associated vascular milieu may contribute to paracrine signaling and sustained BRAF^V6000E-thyroid tumor cell survival; indeed, we found no cell death effects from vemurafenib or combined therapy treatment in either CD31⁺ (vascular endothelial cells), NG2⁺, or PDGFRB⁺ (markers of pericytes; ref. 17) cells in the vascular compartment, including endothelial cells and pericytes (Fig. 5C). Interestingly, sorafenib produced a substantial reduction (∼2-fold change) of viability in the αSMA⁺ cell population (microvessels; Fig. 5C), and all drug treatments substantially decreased levels of proangiogenic factors VEGF and VEGFR2 in the tumor cells; in addition, vemurafenib reduced PDGFRB levels in the tumor cells. Only sorafenib and combined therapy (likely reflecting sorafenib activity) downregulated PDGFB levels in cells of the vascular/endothelial compartment (Fig. 5C). Because BRAF^WT/V600E-PTC expresses high levels of adhesion molecules, which play important roles in the ECM of the tumor microenvironment (10, 22) we also performed trichrome staining of BRAF^WT/V600E-orthotopic PTC that revealed robust collagen deposition (Fig. 5D). Vemurafenib was associated with decreased collagen deposition (Fig. 5D), while sorafenib proved an ineffective mediator. When combined with vemurafenib, sorafenib limited its suppressive efficacy as well.

Collectively, these results indicate that combined therapy suppresses proangiogenic molecules and TSP-1/TGFβ1 expression in thyroid tumor cells but not robustly in the vascular compartment, including pericytes and endothelial cells. This heterogeneous therapeutic response may be linked to the presence of different subpopulations (lineage) of pericytes (NG2⁺, PDGFRB⁺, αSMA⁺ cells) in the tumor microenvironment that might trigger resistance to targeted therapy (i.e., BRAF^V600E and TK inhibitors; Fig. 3E). All information about antibodies used in this study is reported in Supplementary Table S1.

TSP-1 mediates the interaction of tumor cells with the ECM (38), and plays a key role in progression when the BRAF^V600E mutation is present (22). TSP-1 also profoundly influences tumor cell proliferation, adhesion, and migration (22). We therefore validated our TSP-1/TGFβ1 in vitro and mouse findings by assessing the functional interactions of TSP1-dependent regulatory gene networks in clinical samples of BRAF^WT/V600E or BRAF^WT/WTPTC. We used TCGA (The Cancer Genome Atlas samples (39) and applied the Linear model for RNA-seq data (Limma) for moderate T-statistics to identify genes fundamental to tumor growth and microenvironment functions that were differentially expressed in these two tumor groups. We analyzed 23 TCGA genes that are known to regulate ECM or pericyte function, angiogenesis, inflammation, immune response, cell viability and growth, cytoskeleton organization, adhesion/migration/invasion, and metastasis. Nineteen of 23 genes (82.6%) were significantly differentially expressed in BRAF^WT/V600E compared with BRAF^WT/WT PTC; 10 of 19 (52.7%) were upregulated, and 9 of 19 (47.3%) were downregulated (Supplementary Table S2). The set of upregulated genes included TSP-1 (THBS1; Fig. 6A), which significantly increased (1.56-fold change) in BRAF^WT/V600E-PTC (Supplementary Table S2). Furthermore, FN1, COL1A1, ITGA3, TGFβ, and THBS2, etc., which are crucial microenvironment-associated ECM components, were also upregulated with significant fold-change ranging from 1.24 to 11.3. We found similar results when comparing PTC harboring both BRAF^WT/V600E and hTERT mutations (which has been reported in TCGA to cooccur in a very small number of PTC samples; ref. 39) versus either BRAF^WT/V600E-PTC (Supplementary Table 3; Supplementary Fig. S10) or BRAF^WT/WT-PTC (Supplementary Tables 4; Supplementary Fig. S11), indicating that BRAF^V600E is important in the transcriptional regulation of TSP-1 and other genes with functions linked to the tumor microenvironment, including the vascular compartment and associated endothelial cells and pericytes. The significantly differentially expressed genes were used to build a TSP-1 gene regulatory network enrichment using the Cytoscape Genemania algorithm that also included gene coexpression results from TCGA. Each node identified genes and each edge represented functional interactions between genes. We found significant interactions between the upregulated genes in BRAF^WT/V600E-PTC versus BRAF^WT/WT-PTC samples. TSP-1 significantly interacted with all upregulated genes, and more importantly was coexpressed with the TGFβ1 gene (Fig. 6B), suggesting the importance of TSP-1 in the direct regulation of TGFβ1 activation and pathways. We next carried out a TSP-1 pathways enrichment analysis (Fig. 6C), which identified significant pathways crucial for TGFβ signaling, metastasis, inflammation, immune modulation, tumor microenvironment–associated ECM remodeling functions, tumor growth, and VEGF ligand–VEGF receptor interactions, etc. Importantly, many genes involved in these pathways are known to play roles in endothelial cell and pericyte functions in the vascular compartment. We found similar results when we compared PTC harboring both BRAF^WT/V600E and hTERT mutations versus BRAF^WT/WT-PTC (Supplementary Fig. S12). Furthermore, we quantified pericyte abundance (Fig. 6D) using canonical markers such as αSMA, PDGFRB, NG2 (17), and CD90 (THY1) and the single sample Gene Set Enrichment Analysis (ssGSEA) algorithm applied to PTC TCGA data. From a set of 538 samples (59 NT and 479 PTC), we identified the 5% most pericyte-enriched (n = 27, all PTC) and 5% least pericyte-enriched (n = 27, 21 NT and 6 PTC) according to the PTC TCGA data (Fig. 6D). The remaining 90% of samples that included 38 of 59 NT or 446 of 479 PTC (185 BRAF^WT/V600E-PTC, 20 PTC with both BRAF^WT/V600E and hTERT mutations, and 241 BRAF^WT/WT-PTC) ranked in the middle (mediocre) range (defined as “average samples” with neither high nor low pericyte enrichment) of pericyte abundance scores (Fig. 6D). Specifically, NT samples were significantly overrepresented (3.5 folds, P < 0.001) in the low pericytes–enriched group (21 of 27, 77.7%) as compared with PTC samples (6/27, 22.2%). In contrast, 27 of 33 PTC samples (81.8%) showed substantial enrichment (4.5-fold increase) in pericytes while only 6 of 33 samples (18.1%) showed lower enrichment. Of the samples exhibiting a high abundance of pericytes (n = 27), 16 (59.2%) were BRAF^WT/V600E-PTC (P < 0.001, compared with NT), 3 (11.1%) were PTC with both BRAF^WT/V600E and hTERT mutations, and 8 (29.6%) were BRAF^WT/WT-PTC, and none were NT. Of the samples with the least enrichment (n = 27), 5 (18.5%) were BRAF^WT/V600E-PTC and 1 (3.7%) were BRAF^WT/WT-PTC, whereas the vast majority (n = 21) were NT (P < 0.001, compared with BRAF^WT/V600E-PTC with high abundance of pericytes). Therefore high pericyte enrichment aligned with BRAF^WT/V600E-PTC, and was 2-fold or 5.3-fold higher than levels observed in BRAF^WT/WT-PTC or PTC with both BRAF^WT/V600E and hTERT samples, respectively (Fig. 6D). Importantly, intermediate risk of recurrence as assessed by the PTC TCGA clinical database was associated with the 33 PTC samples as follows: 15/33 (45.4%) were BRAF^WT/V600E-PTC, 4/33 (12.1%) were BRAF^WT/WT-PTC, and 1/33 (3%) were PTC with BRAF^WT/V600E and hTERT mutations. Importantly, among the BRAF^WT/V600E-PTC samples with intermediate risk of recurrence, 13 of 15 (86.6%) exhibited a high abundance of pericytes, whereas 2 of 15 (13.3%) showed low pericyte enrichment. Three of 4 (75%) of BRAF^WT/WT-PTC samples with intermediate risk of recurrence showed high pericyte enrichment, and 1 of 4 (25%) showed low pericyte enrichment. No associations were found in PTC samples between BRAF mutational status, the low- or high-risk category, and pericytes abundance score. Overall, our data analysis indicated that pericytes population increased in BRAF^WT/V600EPTC than PTC with other genetic alterations, or even more robust than in NT samples (Fig. 6D). Collectively, our data indicate different activity by targeted therapy with vemurafenib or sorafenib on the regulation of angiogenesis and ECM molecules expression in the BRAF^WT/V600E-KTC1 orthotopic tumor cells and vascular/endothelial compartment (Fig. 6E). BRAF^WT/V600E-PTC cells evoke paracrine regulatory networks to recruit pericytes, which ultimately sustain tumor cell survival. Overall, our findings reveal a new model of resistance to vemurafenib and sorafenib therapy in BRAF^WT/V600E-PTC via the TSP-1/TGFβ1 axis triggered by pericytes (Fig. 6F and G).

Although thyroid cancer mortality rates are lower relative to incidence rates, thyroid cancer mortality has nevertheless increased significantly since the late 1980s (1). Patients diagnosed with thyroid cancer can be treated with radioactive iodine, but a subset of patients fails to respond to this treatment and suffer low survival rates (40), due to the BRAF^V600E mutation. To date, most clinical trials for metastatic thyroid cancer have focused on single agents with low response rates; more effective treatment options for this disease are urgently needed. The prevalence and critical role of genetic mutations in cancer cells have led to targeted molecular therapy. Current therapies target BRAF^V600E, MEK, PI3K, TKs, etc. Because BRAF^V600E is the most frequent oncogene implicated in PTC initiation (32) and aggressiveness (22), targeting this mutation holds great promise for future therapies. Small-molecule kinase inhibitors allow for fewer side effects than traditional chemotherapy; however, most cancers are heterogeneous and have the capacity to develop resistance to targeted therapies (41). Sorafenib was the first targeted therapy approved for patients with advanced differentiated thyroid carcinoma (DTC; ref. 42). It is an oral multikinase inhibitor that is used as first-line treatment for metastatic DTCs. When tumors grow, they generate new blood vessels by angiogenesis to supply adequate nutrients (43). Researchers have therefore focused on antiangiogenic therapy, including disruption of new blood vessels, to suppress tumor growth. Sorafenib has been considered a possible angiogenesis disruptor (44). Here we have shown, however, that thyroid tumor cells harboring the BRAF^V600E mutation elicit resistance to sorafenib and limit its therapeutic efficacy.

Other systemic therapies have been used against DTC in the clinic; we chose to investigate vemurafenib, the first FDA-approved selective oral inhibitor of BRAF^V600E (45). Vemurafenib has been used in patients with metastatic and radioiodine-refractory BRAF^V600E-PTC and continuously administered twice a day in cycles of 28 days (6). This therapy showed antitumor activity with partial response in 10 of 26 patients (38.5%, best overall response). Four patients (15%) died after a median follow-up of 18.8 months. It is unclear whether the similarities of drug targets in DTC could lead to complete cross-resistance and whether sequential treatment would be efficacious. Previous studies revealed that treatment of BRAF^V600E-positive thyroid cancer cells with vemurafenib initially produced a therapeutic response, however, mechanisms of resistance were quickly elicited which led to cell death refractoriness (8, 11, 34, 35). Also, studies on BRAF^V600E human melanoma cells showed resistance to completely suppress ERK1/2 activation even with high doses of vemurafenib (46), similar to our results using the 10 μmol/L dosage. Our study is the first to show data on combined inhibition of both BRAF^V600E and tyrosine kinase as a potential targeted therapeutic option. Combined therapy with vemurafenib and sorafenib showed synergistic effects in BRAF^WT/V600E-PTC cells, whereas it was subadditive in BRAF^WT/WT-PTC cells and pericytes. As a result, this combined inhibition was significantly more effective than vehicle or single agents in inducing higher rates of death in BRAF^WT/V600E-PTC than BRAF^WT/WT-PTC cells. However, our mouse data showed that vemurafenib produced cytostatic effects in orthotopic tumors, whereas combined therapy (likely reflecting sorafenib activity) generated biological fluctuations with tumor inhibition alternating with tumor growth. To understand this phenomenon, we subsequently analyzed the microenvironments of orthotopic BRAF^WT/V600E-PTC, and found these tumors were enriched with heterogeneous populations of pericytes, in particular PDGFRB⁺, αSMA⁺, and NG2⁺. Many genetic alterations, including the BRAF^V600E oncogene, confer a cell-autonomous fitness advantage to tumors by providing independence from growth factors or suppressing cell death responses. In addition, tumor progression is affected by microenvironment-associated factors that cannot be overcome by adopting targeted therapies against only tumor cells per se. The combined vemurafenib plus sorafenib therapy showed synergy effects against BRAF^WT/V600E-PTC cells but not BRAF^WT/WT-PTC cells or pericytes, suggesting biological cooperation between BRAF^V600E and TK pathways. This therapeutic synergy inhibited activation of ERK1/2, AKT, and TKs (e.g., VEGFR2, PDGFRB) to suppress tumor survival and motility.

The depletion of pericytes in transgenic mice under control of NG2 promoter inhibited breast tumor growth and defective tumor vasculature, but in one study was also found to increase metastasis (20). In the context of BRAF^WT/V600E-PTC, pericytes could elicit resistance driven by higher intracellular levels of TSP-1. Intracellular TSP-1 was found to behave differently in pericytes than in BRAF^WT/V600E-PTC cells upon the administration of targeted therapy; that is, pericytes treated with the targeted therapy upregulated intracellular TSP-1, whereas BRAF^WT/V600E-PTC cells downregulated it. This might be due to the fact that thrombospondins, including TSP-1, are ER-resident effectors of adaptive ER stress (47), and can function inside the cell during stress, tissue damage, or active remodeling to augment ER function (47). Here, we found that pericytes secreted TSP-1 and TGFβ1, and induced the rebound of pERK1/2, pAKT and pSMAD3 levels to overcome the inhibitory effects of the targeted therapy in BRAF^WT/V600E-PTC cells. This led to increased BRAF^V600E-PTC cell survival and cell death refractoriness, indicating that accumulated intracellular TSP-1 might drive this effect. These intracellular signaling effectors were suppressed by direct treatment with vemurafenib or sorafenib, and more strongly by combined therapy, suggesting that microenvironment-associated pericytes represent one of the constraints for these targeted therapies. It has been reported that pericytes can communicate with tumor cells and endothelial cells by paracrine communication to ultimately elicit tumor cell survival (19). Combined therapy against BRAF^WT/V600E-PTC cells alone strongly induced cell death and inhibited proliferation, but when these tumor cells were cocultured with pericytes, they showed increased cell survival. More importantly, our in vivo findings showed that combined therapy effectively suppressed the expression of both TSP-1/TGFβ1 and TKs pathways in tumor cells, but not in the vascular compartment. We can argue, therefore, that vascular microenvironment-derived TSP-1 is an important factor in drug resistance in PTC, which functions in a signaling network with intracellular TSP-1, TGFβ1/SMAD, and TKs (i.e., VEGFR2, PDGFRB) for the regulation of tumor growth, as was confirmed in our PTC clinical sample data analysis. Importantly, specific sequences in the thrombospondin type 1 repeats (TSR) are critical for activating latent TGFβ1 through disruption of the LSKL sequence in the latency-associated peptide region of latent TGFβ1 within the mature domain, which reconforms the latent complex to make it accessible to TGFβ signaling receptors (48). Peptides of the LSKL sequence or related analogues such as SRI31277, antagonize TGFβ activation by preventing TSP-1 binding to the latent complex (24). Like TSP-1, TGFβ1 plays an important role in tumorigenesis and angiogenesis. TGFβ signaling can be protumorigenic or tumor suppressive (49, 50). TGFβ acts through membrane receptor kinases to activate SMAD transcription factors for many biological functions (51). To assess the role of this pathway in PTC cells and pericytes, we used the SRI31277 peptide, which antagonizes TSP-1 activity on TGFβ1 activation. This peptide showed significant antitumor effects against multiple myeloma (24). In our study, we analyzed the specificity of the TSP1-driven drug resistance mechanism promoted by pericytes in PTC cells through both an SRI31277-based pharmacologic approach and an shRNA strategy; together these showed that antagonism or knockdown of TSP-1, respectively, downregulated TGFβ1 pathways in PTC cells when these cells were stimulated with pericyte secretome. Ultimately, this disruption of the TSP-1/TGFβ1 axis recovered the ability of the targeted therapy to inhibit BRAF^WT/V600E-PTC intracellular cell signaling, leading to significant decrease in tumor cell growth. Our data also indicated that pericyte populations expressing PDGFRB and NG2 were unchanged by any treatment in our orthotopic mouse model of human BRAF^WT/V600E-PTC, whereas pericytes expressing αSMA were substantially reduced upon administration of sorafenib alone, despite this treatment yielding no significant tumor growth reduction, which suggests that the αSMA⁺ cell subpopulation may be susceptible to induction of cell death by TK inhibitors (e.g., sorafenib). Recruitment of αSMA⁺ pericytes around teratocarcinoma vessels indicated vessel differentiation and higher tumor cell proliferation (52). Our results therefore suggest that diverse pericyte subpopulations play multiple roles in the PTC microenvironment, as shown in other tumor models (17), and may destabilize vessels and promote/sustain BRAF^WT/V600E-PTC cell-to-ECM adhesion for survival, motility and ultimately vascular invasion. Further studies will be needed, however, to explore these possibilities, characterize pericyte subpopulations, and determine the level of activation of pericytes in thyroid carcinoma. In glioblastoma, a contact-dependent interaction with tumor cells switched on the tumor-promoter character of pericytes, inducing their participation in tumor progression (53). Importantly, BRAF^WT/V600E-PTC samples showed higher pericyte enrichment than BRAF^WT/WT-PTC or NT samples, as well as a significant abundance of ECM molecules, which may be linked to the activation of regulatory gene networks and pathways fundamental for tumor growth, angio-invasion, and metastasis. Pericytes express and secrete ECM molecules that are important for cell survival, adhesion, and migration (18), and support processes important to the efficient alignment of endothelial cells required for maturation of capillary structures and delivery of nutrients. Ultimately this process might contribute to BRAF^WT/V600E-PTC cell survival, extra-thyroidal extension, angio-invasion, and metastasis. Because humans and mice have different oral drug bioavailability, physiology, and metabolism, vemurafenib and sorafenib doses used in mice are not comparable with those appropriate to human patients. Vemurafenib was substantially more effective than sorafenib at inhibiting tumor growth in vivo, and caused both cytotoxic and cytostatic effects, as well as downregulation of TSP-1, TGFβ1, SMAD3, ERK1/2 and AKT protein levels in tumor cells, but not in the vascular compartment, including endothelial cells and pericytes. Overall, our results indicate that the TSP-1/TGFβ1 axis is expressed and functional in the PTC microenvironment and might contribute to targeted therapy resistance. Our findings also revealed an abundance of collagen deposition, another key player in migration and angio-invasion, in the orthotopic tumors treated with either sorafenib or combined therapy, but not vemurafenib. Sorafenib primarily acted as an antiangiogenic agent, suppressing VEGFR2 expression in the orthotopic tumor cells, and VEGF and PDGFB protein expression levels in the vascular compartment/endothelial; however, it was ineffective in blocking the likely expression of ECM molecules (e.g., collagen) in the thyroid tumor microenvironment, and maintained high protein levels of pERK1/2, pSMAD3 and pAKT. Ultimately, this effect substantially extended tumor survival compared with vemurafenib treatment, and negatively impacted the therapeutic efficacy of the combined treatment as well. Critically, it has been suggested that TSP-1 may regulate the expression of collagen in response to stress and tissue remodeling, independent of TGFβ1 activation (54). One possibility is that ER stress response may lead to a lack of TSP-1 suppression and cause hyper-secretion of ECM molecules such as collagen and fibronectin. Furthermore, we found these ECM molecules transcriptionally upregulated in BRAF^WT/V600E-PTC compared to BRAF^WT/WT-PTC clinical samples, indicating that the BRAF^V600E oncogene deregulates ECM composition, and not only provides growth advantages to PTC cells but also promotes cell adhesion/migration, leading to intravascular invasion and metastasis (e.g., to the neck lymph node compartments), which is frequent in BRAF^WT/V600E-PTC.

Collectively, our results indicate that the action of the TSP-1/TGFβ1 axis enables drug resistance to BRAF^V600E and TK inhibitors in BRAF^WT/V600E-PTC cells. Our study is limited by a lack of human samples from patients with metastatic BRAF^V600E-PTC treated with BRAF^V600E inhibitors and TKI; however, our results are substantiated by our integrated cell cultures, cell cocultures, and mouse model, as well as the use of TCGA clinical samples of PTC. We cannot exclude the possibility that this resistance mechanism is specific only to sorafenib. Further studies will be needed to assess BRAF^V600E inhibitors in combination with other TK inhibitors such as lenvatinib. Also, as TGFβ1 is known to evoke active immune suppression, future research could assess the effects of pharmacologic antagonism of the TSP-1/TGFβ1 axis on the immune microenvironment of thyroid carcinomas.

In summary, our findings indicate that pericytes may be a double-edged sword in BRAF^WT/V600E-PTC therapy, and secrete soluble factors, such as TSP-1 and TGFβ1, that trigger resistance to BRAF^V600E inhibitors and TKI in BRAF^WT/V600E-PTC. Pharmacologic antagonism of TSP-1 to block TGFβ1 activation may be crucial for the inhibition of tumorigenic effects by TGFβ1 pathway in BRAF^WT/V600E-PTC cells. This might represent a novel therapeutic strategy with translational applications in clinical trials against BRAF^WT/V600E-PTC resistant to targeted therapies. Finally, TSP-1 is a potential biomarker for assessing therapeutic response to BRAF^V600E and TK inhibitors in patients with invasive BRAF^WT/V600E-PTC.

J.E. Murphy-Ullrich has ownership interest (including stock, patents, etc.) in EMD Milipore license for anti-TSP antibody (not used in paper) and a US patent on SRI31277. C. Nucera received expert testimony for NIH as an ad hoc peer-reviewer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Lawler, C. Nucera

Development of methodology: A. Prete, A.S. Lo, Z.A. Antonello, D.M. Vodopivec, J. Clohessy, J. Lawler, C. Nucera, A.M. Dvorak

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Prete, A.S. Lo, P.M. Sadow, Z.A. Antonello, D.M. Vodopivec, S. Ullas, J.N. Sims, J. Clohessy, T. Sciuto, J. Lawler, C. Nucera

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Prete, A.S. Lo, P.M. Sadow, S.S. Bhasin, Z.A. Antonello, D.M. Vodopivec, S. Ullas, J.N. Sims, T. Sciuto, M. Bhasin, J.E. Murphy-Ullrich, S.A. Karumanchi, C. Nucera

Writing, review, and/or revision of the manuscript: A. Prete, A.S. Lo, P.M. Sadow, S.S. Bhasin, D.M. Vodopivec, S. Ullas, J.N. Sims, M. Bhasin, J.E. Murphy-Ullrich, J. Lawler, S.A. Karumanchi, C. Nucera, A.M. Dvorak

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.M. Vodopivec, C. Nucera, A.M. Dvorak

Study supervision: J. Clohessy, C. Nucera

Other (figure preparation): A. Prete, C. Nucera

Other (provided SRI312777): J.E. Murphy-Ullrich

C. Nucera (Principal Investigator, Human Thyroid Cancers Preclinical and Translational Research at the Beth Israel Deaconess Medical Center (BIDMC)/Harvard Medical School) was awarded grants by the National Cancer Institute/NIH (1R21CA165039-01A1 and 1R01CA181183-01A1), the American Thyroid Association (ATA), and ThyCa:Thyroid Cancer Survivors Association Inc. for Thyroid Cancer Research. C. Nucera was also a recipient of the Guido Berlucchi “Young Investigator” research award 2013 (Brescia, Italy) and BIDMC/CAO Grants. SRI31277 production and J.E. Murphy-Ullrich's contributions were supported by NIH grant 1R01CA175012. We are grateful to Prof. Harold F. Dvorak and Miss Elizabeth McGonagle (BIDMC/HMS) for critical reading of our manuscript. We are grateful to Dr. Andrew L. Kung (Memorial Sloan Kettering Cancer Center, NYC, USA) for kindly providing the FUW-Luc-mCherry-puro (luciferase/cherry) plasmid, and to Prof. Judy Lieberman and Dr. Minh T.N. Le (Children's Hospital, Harvard Medical School, Boston, MA) for kindly providing packaging plasmids (psPAX2, pMD2-G, and RTR2). We also thank Miss Nicole Pandell for the technical support for the oral gavages in the mice.

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.