RNA Sequencing Identifies Multiple Fusion Transcripts, Differentially Expressed Genes, and Reduced Expression of Immune Function Genes in BRAF (V600E) Mutant vs BRAF Wild-Type Papillary Thyroid Carcinoma

Abstract

Context:

The BRAF V600E mutation (BRAF-MUT) confers an aggressive phenotype in papillary thyroid carcinoma, but unidentified additional genomic abnormalities may be required for full phenotypic expression.

Objective:

RNA sequencing (RNA-Seq) was performed to identify genes differentially expressed between BRAF-MUT and BRAF wild-type (BRAF-WT) tumors and to correlate changes to patient clinical status.

Design:

BRAF-MUT and BRAF-WT tumors were identified in patients with T1N0 and T2–3N1 tumors evaluated in a referral medical center. Gene expression levels were determined (RNA-Seq) and fusion transcripts were detected. Multiplexed capture/detection and digital counting of mRNA transcripts (nCounter, NanoString Technologies) validated RNA-Seq data for immune system-related genes.

Patients:

BRAF-MUT patients included nine women, three men; nine were TNM stage I and three were stage III. Three (25%) had tumor infiltrating lymphocytes. BRAF-WT included five women, three men; all were stage I, and five (62.5%) had tumor infiltrating lymphocytes.

Results:

RNA-Seq identified 560 of 13 085 genes differentially expressed between BRAF-MUT and BRAF-WT tumors. Approximately 10% of these genes were related to MetaCore immune function pathways; 51 were underexpressed in BRAF-MUT tumors, whereas 4 (HLAG, CXCL14, TIMP1, IL1RAP) were overexpressed. The four most differentially overexpressed immune genes in BRAF-WT tumors (IL1B; CCL19; CCL21; CXCR4) correlated with lymphocyte infiltration. nCounter confirmed the RNA-Seq expression level data. Eleven different high-confidence fusion transcripts were detected (four interchromosomal; seven intrachromosomal) in 13 of 20 tumors. All in-frame fusions were validated by RT-PCR.

Conclusion:

BRAF-MUT papillary thyroid cancers have reduced expression of immune/inflammatory response genes compared with BRAF-WT tumors and correlate with lymphocyte infiltration. In contrast, HLA-G and CXCL14 are overexpressed in BRAF-MUT tumors. Sixty-five percent of tumors had between one and three fusion transcripts. Functional studies will be required to determine the potential role of these newly identified genomic abnormalities in contributing to the aggressiveness of BRAF-MUT and BRAF-WT tumors.

Thyroid cancer has almost tripled in incidence in the past decade in the United States (1). Differentiated thyroid cancer (DTC) consists of papillary (PTC), follicular, and Hürthle cell histotypes. Most DTCs (85%–90%) are stage I, with a 99% survival rate, but in stage IV, 5-year survival is <50%. Although somatic mutations are present in about 70% of DTCs, the cellular events responsible for tumor behavior are complex, involving genes/proteins that have not yet been well characterized. Of the mutations associated with PTC, BRAF V600E is the most common (40%–50%) and has been associated with a more aggressive phenotype (2, 3). Nevertheless, significant clinical overlap exists between tumors that are BRAF V600E (BRAF-MUT) mutant and those with wild-type BRAF (BRAF-WT), suggesting that additional factors play a role in determining tumor behavior. Rapid advances in transcriptome analysis technologies allow exploration of other concomitant genetic alterations that might drive tumor aggressiveness and response to therapy (2–8). In the current study, we performed RNA sequencing (RNA-Seq) (9) to identify additional genetic abnormalities present in patients with PTC and to identify genes that are differentially expressed between BRAF-MUT and BRAF-WT tumors, and which therefore might contribute to the heterogeneity of phenotypic responses (8).

Materials and Methods

Patients

Patients undergoing thyroid surgery at Mayo Clinic were invited to participate in an Institutional Review Board–approved protocol that permits collection of blood and tissue (tumor and normal thyroid) for studies of the genetic pathogenesis of thyroid malignancies. All samples were snap frozen in liquid nitrogen within 30 minutes of resection and stored at −80°C. Patient clinical data were transferred from the Mayo Clinic Electronic Medical Record to our Thyroid Tumor Registry.

For the current study, we identified patients with PTC with several TNM subtypes. We included patients who had smaller tumors and no lymph nodes (T1N0) as well as patients with larger tumors with nodal involvement (T2–3 N1a–b). We then determined tumor BRAF mutation status from DNA extracted from paraffin blocks. The presence of 10% or more of mutant alleles in a tumor was used as the cutoff for classifying it as BRAF-MUT; those with 0% mutant alleles were deemed BRAF-WT. Histology was reviewed and the degree of lymphocyte infiltration was determined by a single endocrine pathologist (M.R.). Two patients had follicular variant of PTC (7). Background lymphocytic thyroiditis in uninvolved thyroid tissue was rated as absent, mild (scattered/lymphoid aggregates), moderate (easily identifiable lymphoid aggregates), or marked (large confluent lymphoid aggregates involving most of the tissue). Chronic inflammation involving the tumor was graded as 0 (no inflammation); 1 (10% or less); 2 (11%–30%); or 3 (≥30%) of tumor involved by chronic inflammation. An immune infiltration score was developed to summarize lymphocyte infiltration data. Initially, tumors with “absent” infiltration in uninvolved thyroid tissue were assigned a score of 0; those with “mild” were scored 1, “moderate” 2, and “marked” 3. These scores were then added to those assigned for tumor infiltrating lymphocytes (0+ to 3+), such that a sample with marked infiltration of nontumor tissue and 3+ inflammation of the tumor received an aggregate immune infiltration score of 6, whereas a tumor with no evidence of infiltration of either tumor or adjacent tissue received a score of 0, and so forth.

Genotyping thyroid tumors

We cut 10-μm-thick slices from formalin-fixed paraffin-embedded archival tissue blocks and fixed them to glass microscopy slides. For each block, an additional consecutive 2-μm-thick slice of tissue was stained with hematoxylin and eosin and used as a template for identification of normal and benign tissue on the corresponding unstained slides.

For nucleic acid extraction, slides that contained clearly delineated tumorous regions were manually microdissected after paraffin removal through xylene baths to selectively harvest tumor tissue. A few slides were covered in their entirety by tumor or a through and through admixture of tumor and other tissues. These slides were extracted in toto using the PicoPure Kit (Life Technologies). The tissue removed from the slides was immediately placed in PicoPure buffer and incubated at 65°C for 24 hours. The proteinase K contained in this buffer was subsequently inactivated after incubation by heating to 95°C for 10 minutes. The extracts were used directly for PCR detection of mutant BRAF (∼1799 T>A, p. V600E), as previously described (10). Estimates of the relative amounts of BRAF mutant DNA vs BRAF-WT DNA were made by comparison of the crossing points of mutant and BRAF-WT from the same specimen.

Thyroid gene counts

Total RNA was extracted from fresh-frozen tumor samples. Indexed Tru-Seq libraries were prepared and sequenced to a median depth of 88 M aligned tags using 2 × 50 paired end sequencing on the Illumina HiSeq2000 platform. Read statistics are given in Supplemental Table 1, published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org. Briefly, median read depth was 142 M counts (range 44 M to 259 M), with median alignment efficiency of 78% (range 57%–82%). Via frequency distribution plots, we determined the lower limit of expression to be a count of 8, and the median expression of genes with 10 or more counts was 17 284 transcripts (range 15 721 to 18 989). The expression levels of genes and exons from the RNA-Seq data were calculated by an in-house assembled RNA-Seq analytic pipeline. Briefly, the reads were aligned to both the genome and the exon junctions using bowtie (11), within the TopHat tool package (12). The exon and gene level counts were summarized using the HTSeq method (http://www-huber.embl.de/users/anders/HTSeq/doc/overview.html) using gene and exon definitions obtained from the UCSC Golden Path hg19 database (RefGene file). For the current study, only the read counts are used for the gene/exon quantity.

Differential expression was determined using the edgeR package in the R programming language, version 2.14.0 (13). We excluded a gene from analysis if less than four patients had counts per million over two, and used a prior N of 2 in our estimation of tagwise dispersion. Gene counts data have been deposited in GEO (GSE48953).

Validation

Validation of RNA-Seq gene counts data was carried out using multiplexed capture/detection and digital counting of mRNA transcripts for immune-related genes (NanoString Immunology CodeSet; GXA-HIM1). Seventeen of the 20 samples had sufficient RNA for NanoString validation (100 ng). Within the CodeSet, there are two sequence-specific probes for each transcript: a 3′-biotinylated capture probe and a 5′-reporter probe tagged with a fluorescent barcode. Following 12 to 30 hours of hybridization at 65°C, the NanoString Prep Station robot removed excess capture and reporter probes and immobilized the RNA on a streptavidin-coated cartridge using the biotinylated capture probes. The digital analyzer, equipped with charge-coupled device camera, performed data collection by counting fluorescent barcodes and aligning them to gene targets, providing a tally of target RNA molecules present in each sample. Normalization to spike in controls and total gene counts was performed with nSolver. Differential expression was determined using edgeR.

Fusion transcripts

Fusion transcripts were detected using the SnowShoes-FTD method (14). This algorithm has been validated in both cell lines and primary tumor samples (14, 15). High-confidence fusion transcripts are nominated based on five or more discordant read pairs (the two ends from the paired-end sequencing reads mapped to different genes), and one or more split reads (in which one end encompasses the fusion junction and the other maps to either of the fusion partners). We have previously determined that the false detection rate for such high-confidence transcripts is <5%, as assessed by RT-PCR amplification and sequencing of candidate transcripts (14, 15). A selected subset of candidate fusion transcripts was validated using RT-PCR, with 100% validation, as described in the text. PCR primers for fusion transcript validation are given in Supplemental Table 2.

Results

Clinical characteristics

Twenty patients (12 BRAF-MUT; 8 BRAF-WT) had fresh frozen tumor DNA to determine BRAF status and RNA of sufficient quantity and quality to perform deep sequencing, as shown in Figure 1A. Gender and tumor size were similar, whereas the BRAF-MUT patients tended to have more N1, stage III, and multifocal disease (Table 1). Five of 12 (42%) BRAF-MUT vs 6/8 (75%) of BRAF-WT patients had chronic lymphocytic infiltration of nontumor tissue (P = .2 by Fisher's exact test). Three of 12 (25%) BRAF-MUT vs 5/8 (62%) BRAF-WT patients had tumor infiltrating lymphocytes (P = .17 by Fisher's exact test). Individual characteristics of the 20 patients are described in Table 2.

Figure 1.

Schematic representation of the analytical approach used to characterize immune function gene expression patterns in BRAF-MUT vs WT tumors. RNA-Seq analysis was carried out on 20 PTC samples to identify 560 differentially expressed genes, as shown in (A). These genes were subjected to MetaCore process pathway analysis. Seven statistically significant pathways were identified (P < .001), as shown in (B). Six of these were related to immune function (bold in B). These six process pathways comprised 67 genes, of which 55 were differentially expressed in the RNA-Seq data at P < .01. Orthogonal validation was carried out using a 540 gene immune function NanoString CodeSet, of which 94 genes were determined to be differentially expressed at P < .01. NK, natural killer; ECM, extracellular matrix.

Table 1.

Clinical Characteristics of the 20 Patients Included in the Study

	BRAF V600E (no., %)	BRAF (WT) (no., %)
Age at surgery	54.5 (31.3–81.9)	40.1 (33.2–55.5)
Gender
Male	3 (25)	3 (37.5)
Female	9 (75)	5 (62.5)
Tumor size (T)
T1a	0 (0)	0 (0)
T1b	6 (50)	5 (62.5)
T2	3 (25)	2 (25)
T3	3 (25)	1 (12.5)
T4	0 (0)	0 (0)
Lymph node status (N)
N0	4 (33.3)	6 (75)
N1a	7 (58.3)	1 (12.5)
N1b	1 (8.3)	1 (12.5)
Metastasis (M)
M0	12 (100)	8 (100)
M1	0 (0)	0 (0)
Stage
I	9 (75)	8 (100)
II	0 (0)	0 (0)
III	3 (25)	0 (0)
IV	0 (0)	0 (0)
Multifocality
Unifocal	7 (58.3)	7 (87.5)
Multifocal unilateral	1 (8.3)	1 (12.5)
Multifocal bilateral	4 (33.3)	0 (0)
Histological variant
Classic papillary	12 (100)	6 (75)
Follicular variant	0 (0)	2 (25)
Lymphocytic thyroiditis in tumor-adjacent tissue	5 (42)	6 (75)
Tumor-infiltrating lymphocytes	3 (25)	5 (62)
MACIS score	5.4 (3.82–7.69)	4.2 (3.46–5.38)
Total	12	8

Abbreviation: MACIS, Metastasis, Age, Completeness of resection, Invasion, Size.

Continuous variables are reported in median (range); categorical variables are reported in number (percentage).

Table 2.

Patient Clinical Characteristics and Fusion Transcripts Status

NGS Sample ID	Age at Surgery/Gender	TNM	Histology	Tumor-Adjacent Lymphocytic Thyroiditis/Tumor-Infiltrating Lymphocytes	Fusion Transcripts
BRAF-MUT
NGS_7	41/F	T3N1aM0	PTC	Absent/0
NGS_4	63/F	T1bN1aM0	PTC	Absent/0	RHOBTB2 → PEBP4
NGS_1	81/F	T3N1bM0	PTC	Absent/0
NGS_16	39/M	T2N1aM0	PTC	Absent/0
NGS_11	79/M	T1bN0M0	PTC	Absent/0	KIAA1267 → ARL17B
					METTL10 → FAM53B
					PPIP5K1 → CATSPER2
NGS_14	51/M	T1bN1aM0	PTC	Mild/2+	HEPHL1 → PANX1
NGS_18	31/F	T3N1aM0	PTC	Absent/0	HEPHL1 → PANX1
					METTL10 → FAM53B
					MYH14 → CLU
NGS_19	61/F	T1bN0M0	PTC	Mild/1+	KIAA1267 → ARL17B
					PPIP5K1 → CATSPER2
NGS_2	38/F	T2N1aM0	PTC	Moderate/2+	KIAA1267 → ARL17B
NGS_3	67/F	T1bN0M0	PTC	Moderate/0
NGS_9	57/F	T1bN0M0	PTC	Absent/0	HEPHL1 → PANX1
NGS_5	35/F	T2N1aM0	PTC	Mild/0
BRAF-WT
NGS_12	35/F	T1bN0M0	FV	Marked/2+	HEPHL1 → PANX1
NGS_17	41/F	T1bN0M0	FV	Absent/0	HEPHL1 → PANX1
					RHOBTB2 → PEBP4
					CKLF → CMTM4
NGS_6	55/M	T1bN0M0	PTC	Mild/1+	ETV6 → NTRK3
NGS_8	44/F	T1bN1aM0	PTC	Moderate/1+
NGS_10	38/M	T2N0M0	PTC	Absent/0
NGS_13	33/F	T3N1bM0	PTC	Mild/1	PPIP5K1 → CATSPER2
					ZNF252 → TMED10
					MKRN1 → BRAF
NGS_15	43/M	T2N0M0	PTC	Mild/0	PPIP5K1 → CATSPER2
NGS_20	34/F	T1bN0M0	PTC	Mild/1+	TG → EEF1A1

Abbreviations: FV, follicular variant of PTC; NCS, Next Generation Sequencing. NGS numbers were assigned to each sample, as described in the text. Fusion transcripts are described in the text. Grading scales for tumor-adjacent and tumor-invading lymphocytes are given in Materials and Methods.

Gene expression

A total of 13 085 genes were interrogated using edgeR, of which 560 were differentially expressed (P < .01) (Figure 1A and Supplemental Table 3) between BRAF-MUT and BRAF-WT. MetaCore (Thomson Reuters) process pathway analysis was carried out to identify biological processes associated with these genes, as shown in Figure 1A. The top 10 process pathways are identified in Figure 1B. Seven process pathways were significant at P < .001 (Figure 1B), and six of these were immune/inflammation-related pathways (bold font in Figure 1B). These six pathways included 67 immune function genes, 55 of which differed between BRAF-MUT and BRAF-WT tumors (Supplemental Table 4). Of particular interest was the dramatic lower expression of immune and inflammatory response genes in the BRAF-MUT vs BRAF-WT tumors. Figure 2 depicts the 55 differentially expressed RNA-Seq immune function gene counts, some of which have been identified by others in PTC and some of which have been shown to be dysregulated in other malignancies. Only 4 of the 55 genes were expressed at a higher level in BRAF-MUT than BRAF-WT tumors; these genes are HLAG, CXCL14, TIMP1, and IL1RAP. Cluster analysis (Supplemental Figure 1) indicates the degree to which expression of these 55 immune function genes is associated with tumors with mutant and wild-type BRAF.

Figure 2.

Expression of 55 immune function genes from RNA-Seq gene count data. We identified 55 differentially expressed immune function genes associated with six statistically significant MetaCore process pathways (Figure 1). Median log2 gene count data were used to calculate fold change (BRAF-MUT gene counts − BRAF-WT gene counts).

NanoString was used to validate the RNA-Seq data for immune genes. Ninety-four of 540 immune genes were differentially expressed, with 67 higher in BRAF-WT and 27 higher in BRAF-MUT tumors (P < .01) (Supplemental Table 5). Of the 55 immune function genes identified as differentially expressed in the RNA-Seq data (bold in Supplemental Table 4), 43 were also evaluated by NanoString. All 43 genes differed significantly (P < .01) on the NanoString panel (bold italic in Supplemental Table 5). In all but four instances (FN1; IL1RAP; TGFBR1; CFI), the gene was underexpressed in BRAF-MUT vs BRAF-WT tumors (Figure 3). The Spearman rank correlation coefficient mean R = 0.861 ± 0.019 (range 0.83–0.89) for RNA-Seq vs NanoString established the validity of the RNA-Seq data for these samples. These data confirm that BRAF-MUT tumors have less expression of immune and inflammatory response genes than BRAF-WT PTCs.

Figure 3.

Orthogonal validation of differentially expressed immune function genes. Of the 55 differentially expressed genes shown in Figure 2, 43 were interrogated on the NanoString immune function CodeSet. The median RNA-Seq gene counts for each of these genes are shown in gray-filled bars, whereas the median NanoString gene counts are shown in black-filled bars.

For exploratory purposes, we developed an immune function gene score based on expression of four immune function genes: IL1B, CCL19, CCL21, and CXCR4. These were among the most significantly differentially expressed immune function genes (based on fold change and P value) and exhibited a high degree of correlation (Spearman log rank correlations >0.7 and P < .001 for all pairwise comparisons). Individual tumors were scored based on median expression, with a score of 1 assigned for each gene that was expressed above median and 0 for each gene expressed below median. Thus, a tumor that expressed all four genes above median received a score of 4, whereas a tumor that expressed all four genes below median was scored as 0, and so forth. Gene expression scores were compared with immune infiltration scores (as described in Materials and Methods). Scores for individual tumors are given in Supplemental Figure 2A. We observed that tumors with gene scores ≥3 had median immune infiltration scores of 2, whereas tumors with gene scores ≤2 had median immune infiltration scores of 0 (P = .009 by Mann-Whitney U test). We conclude that increased expression for immune function genes is related, at least for some subset of genes, to a higher degree of lymphocyte infiltration.

In parallel, we examined the distribution of immune infiltration scores in PTCs with wild-type and mutant BRAF, as shown in Supplemental Figure 2, B and C. Although there is some overlap between the immune scores in the two groups, the fraction of BRAF-MUT tumors with immune infiltration score = 0 (7/12) was notably greater than that observed in wild-type tumors (2/8), and there was a statistically significant trend for the BRAF-MUT tumors to have lower immune infiltration scores than those assigned to the wild-type controls (P = .05 by Mann-Whitney U test), consistent with our observation that expression of immune function genes is generally lower in BRAF-MUT tumors.

Given that the BRAF-MUT is thought to result in a high level of activation of the MAPK pathway, we initially asked if genes associated with MAPK signaling were differentially expressed in BRAF-MUT vs BRAF-WT tumors. We identified eight unique GO_BP terms associated with MAPK signaling (GO:0008349, GO:0005078, GO:0051019, GO:0048273, GO:0031434, GO:0031435, GO:0008545, GO:0004707); 61 genes linked to these GO terms were expressed in the PTC samples. All 61 transcripts were more abundant in the BRAF samples (fold change >0). However, only one gene, IGBP1, was differentially expressed at >2-fold with P < .05. Thus, BRAF mutational status does not appear to be associated with significant changes in expression of MAPK signaling genes. This is perhaps not surprising, because MAPK pathway activity is generally controlled by posttranslational modification rather than changes in gene expression. We attempted to develop several alternative approaches to score TNM status and to correlate these scores with BRAF mutational status or MAPK gene expression profiles. However, the sample size proved too small to develop models with sufficient statistical power to make any definitive conclusions.

Thyroid fusion transcripts

The SnowShoes-FTD workflow (15) was used to detect 11 high-confidence fusion transcripts, defined as those having two or more discordant reads (fusion encompassing) and one or more split reads (fusion spanning). Table 2 illustrates each patient by TNM, histologic and BRAF status, and their associated fusion transcript(s). As shown in Supplemental Table 6, four of these chimeric transcripts were interchromosomal and therefore presumed to arise as a result of genomic rearrangement. Among the remaining seven intra-chromosomal fusions, two (HEPHL1 → PANX1 and METTL10 → FAM53B) are derived from fusion partners that are on the same strand and in the same orientation and may arise from transcriptional read-through, deletion, or translocation. The remaining five intrachromosomal fusions would require translocation or translocation with inversion and are therefore likely to arise from chromosomal rearrangement.

Previous analyses (14) have shown that the false detection rate for high-confidence fusion transcripts of the sort listed in Supplemental Table 6 is on the order of <5%. We elected to validate a subset of candidate fusions, those that were predicted to encode in-frame chimeric fusion proteins (CKLF → CMTM4, ETV6 → NTRK3, MKRN1 → BRAF, and PPIP5K1 → CATSPER2). As shown in Figure 4, all four fusion transcripts were validated using PCR primers generated from the SnowShoes output (Supplemental Table 2) followed by Sanger sequencing of the resultant PCR product. Notably, we validated recurrence of the PPIP5K1 → CATSPER2 chimera in four tumors, with a second isoform of this fusion transcript confirmed in two of the tumors. No fusion transcripts tested failed to validate in any sample in which they were identified by RNA-Seq analysis.

Figure 4.

RT-PCR validation of in-frame fusion transcripts. PCR primers were designed from the SnowShoes output (sequence given in Supplemental Table 2). These primers were used for RT-PCR amplification of candidate chimeric transcripts. PCR products of the predicted size were detected in all cases. These products were extracted from the gels and subjected to Sanger sequence analysis to confirm their identities.

The predicted products of each of the fusion transcripts are given in Supplemental Table 7. One fusion transcript (RHOBTB2 → PEBP4) appears to involve a “promoter swap” in which the 5′-untranslated region (UTR) of RHOBTB2 is fused into 5′UTR of PEBP4 to give rise to a transcript that encodes full-length PEBP4 protein. Two of the fusion transcripts involve 3′UTR swaps: METTL10 → FAM53B in which the 3′UTR of METTL10 is fused into FAM53B to give rise to full-length METTL10 and ZNF252 → TMED10 in which the 3′UTR of ZNF252 is fused into TMED10 to give rise to full-length ZNF252. Fusion transcripts that arise due to promoter or 3′UTR swaps are likely to be regulated by mechanisms that differ from those of the parental genes.

Four fusion transcripts are in-frame and predicted to encode chimeric proteins (CKLF → CMTM4, ETV6 → NTRK3, MKRN1 → BRAF, and PPIP5K1 → CATSPER2). Four transcripts are fused out-of-frame and predicted to give rise to C-terminal truncation mutants containing various lengths of the coding sequences of the upstream fusion partner (HEPHL1 → PANX1, KIAA1267 → ARL17B, MYH14 → CLU, and TG → EEF1A1). Both chimeric proteins and C-terminal deletion mutants may have unique functions compared with the parental genes. Moreover, expression of the putative C-terminal deletion products could be affected by nonsense-mediated degradation or by translational regulatory processes that impinge on the 3′UTRs, which are presumably altered due to the fusion event. Given the central role of BRAF in PTC, it is relevant to note that the MKRN1 → BRAF in-frame fusion is predicted to encode amino acids 1–257 of MKRN1 (NP 001138597.1-length 329 aa) fused to amino acids 381–766 of BRAF (NP 004324.2-length 766 aa). The 386-amino-acid BRAF coding sequences include the entire kinase domain. By comparison with a previously described KIAA1549 → BRAF fusion product, the MKRN1 → BRAF fusion protein would be expected to have kinase activity and might be transforming (16).

Although the number of tumors analyzed was relatively small, several fusion transcripts appeared to be recurrent, expressed in more than one tumor (shown in Supplemental Table 6). HEPHL1 → PANX1 was expressed in 5 of 20 tumors and METTL10 → FAM53B in 2 of 20 tumors; however, these transcripts may arise due to transcriptional read-through, as discussed above. KIAA1267 → ARL17B, PPIP5K1 → CATSPER2, and RHOBTB2 → PEBP4 are likely to arise from translocation and are expressed in 3/20, 4/20, and 2/20 samples, respectively. Tumors with or without BRAF mutation appeared to express similar numbers of fusion transcripts. Among the recurrent transcripts, KIAA1267 → ARL17B and METTL10 → FAM53B were expressed exclusively in tumors with BRAF mutations, but the significance of this distribution is difficult to assess, given the small number of samples.

Discussion

In the current study, we performed RNA-Seq of BRAF-MUT and BRAF-WT PTC to determine the extent to which genes were differentially expressed. We discovered not only differences at the level of hundreds of individual genes, but also preferential lower expression of many gene ontology pathways in the BRAF-MUT tumors. Of interest was the notable six- to eight-fold down-regulation compared with up-regulation in the immune and inflammatory MetaCore pathways in the mutant tumors (Figures 2 and 3). We, therefore, validated the RNA-Seq data using the NanoString nCounter Gene Expression CodeSet for 540 human immunology genes, confirming higher overall immune function gene expression in the BRAF-WT tumors (Supplemental Table 5).

Many of the chemokines and their receptors were expressed at a lower level in BRAF-MUT tumors (Figures 2 and 3), probably due to a greater presence of an inflammatory infiltrate in both nontumor and tumor tissues in the BRAF-WT patients. Furthermore, normal thyrocytes, which are induced to express the RET/PTC1 oncogene, also have induction of inflammatory molecules such as CCL2, CXCL12, CXCR4, IL1B (17). A number of reports also describe increased expression of MET, FN1, TIMP1 CITED1, CHI3L1, SERPINA1, DPP4, and SFTPB in PTC (18–21), all of which we found expressed at a higher level in BRAF-MUT tumors. Giordano et al (4) reported that BRAF-MUT tumors overexpressed TM7SF4 (which encodes a dendritic cell protein), TMPRSS6, FN1, MET and underexpressed the oncogene VAV3, all consistent with our results (Supplemental Table 3). They also noted underexpression of TIMP1 in RAS mutant-positive tumors (4), whereas we found increased TIMP1 gene counts in BRAF-MUT tumors. The most highly significant gene overexpressed in BRAF-MUT tumors in our study was KLK7 (Supplemental Table 3). Kim et al (22) showed this gene to be overexpressed in five patients with PTC, four of whom had BRAF-MUT tumors. KLK7 has been shown to induce epithelial-to-mesenchymal transition in prostate cancer cells (23) and bears further study in PTC. We were surprised to find that CXCR4 gene counts were fewer in our BRAF-MUT tumors, because Torregrossa et al (24) found higher CXCR4 mRNA and protein in BRAF-MUT tumors. However, in their study the odds ratio of higher CXCR4 immunohistochemical staining was increased in PTC patient tumors with lymphocytic thyroiditis (24); we had more patients with lymphocytic thyroiditis in our BRAF-WT cohort.

One notable exception to apparent reduced expression of immune function genes in BRAF-MUT tumors was the Major Histocompatibility Complex, Class I, G (HLA-G) gene, a nonclassical HLA class 1 molecule, which was overexpressed in the BRAF-MUT tumors (Figure 2). HLA-G is expressed not only in infiltrating inflammatory cells but also in PTC tumor cells (25), which may explain the higher expression of HLA-G in BRAF-MUT tumors in the current study despite less lymphocytic infiltration. HLA-G evokes a broad spectrum of immunosuppressive functions including inhibition of cytotoxic CD8+ T cells and natural killer cells, altering dendritic cell function, and shifting from Th1 to Th2 T cells (26, 27). The overall environment favors an escape from immune surveillance of BRAF-MUT tumors. In breast cancer, HLA-G was inversely related to tumor infiltrating lymphocytes and was associated with a higher recurrence rate (28). In PTC, Nunes et al (25) showed increased HLA-G protein expression in patients with lymph node metastases and capsular invasion. Soluble HLA-G is also higher in patients with PTC and lymphocytic thyroiditis compared with healthy subjects (29) and could possibly serve as a biomarker of disease activity. It is tempting to speculate that mutational activation of BRAF may be linked to induction of HLA-G, which may, in turn, result in immune suppression; however, our data do not preclude alternative hypotheses.

Other immune pathway gene counts overexpressed in BRAF-MUT vs BRAF-WT tumors include CXCL14, shown to be associated with BRAF mutation (4, 30) and lymph node metastasis (30); TIMP1, which can activate AKT signaling and is up-regulated in BRAF-MUT PTCs (31); and IL1RAP (interleukin-1 receptor accessory protein), a gene not described in thyroid cancer but that is dysregulated in acute myelogenous leukemia stem cells (32). Like HLA-G, CXCL14 is expressed not only in inflammatory cells but also in PTC tumor cells (30).

Fusion proteins are commonly associated with hematologic malignancies. They are recognized much less frequently in solid tumors, although DTCs have traditionally been recognized as expressing several such proteins (33, 34). The functional significance of the currently identified fusion transcripts is, for the most part, unknown. A microarray of 528 known fusion genes (35) identified none in 12 PTCs that corresponded to the findings in our study, and there are no articles identified in PubMed for 8 of the 11 fusions. In contrast, ETV6-NTRK3 has been found in a variety of nonthyroidal tumors including secretory breast cancer, congenital fibrosarcomas, mesoblastic nephromas, and acute myelogenous leukemia (33, 36). This fusion protein may influence the PI3K/AKT and RAS-MAPK pathways (37) and could therefore contribute to oncogenic properties involved in thyroid cancer. The KIAA1267-ARL17 fusion, which truncated the coding structure, was reported in five cases of acute myeloid leukemia (38), and the METTL10-FAM53B fusion was identified in a lung cancer metastasis (39). It is notable that seven of the genes involved in the fusions reported here have identified oncogenic or tumor suppressive functions including PANX (40, 41), FAM53B (42, 43), RHOBTB2 (44, 45), PEBP4 (46), MKRN1 (47, 48), BRAF (2–7), and clusterin (49, 50). BRAF is well known as a major driver in PTC, and one or more of these other genes, as fusion transcripts, may contribute either positively or negatively to the malignant behavior of these thyroid tumors.

There is extensive literature on the roles of autoimmunity and inflammation in PTC (8, 51–53). Dvorkin et al (54) recently showed that patients with Hashimoto's thyroiditis and PTC have a more favorable outcome, and Kim et al (55) have reported a lower frequency of lymphocytic thyroiditis in BRAF-MUT PTC patients. The current study demonstrates substantial differences in the expression of immune process pathways between BRAF-MUT vs BRAF-WT tumors, as well as in the frequency of autoimmune thyroiditis. It is possible, indeed likely, that these immune and inflammation differences contribute to the phenotypic expression of patients with BRAF-MUT tumors that have more aggressive clinical courses. Whether the BRAF mutation directly alters expression of immune pathways or functions independently (but perhaps synergistically) requires further investigation.

In summary, immune/inflammatory pathways are expressed at a lower level in BRAF-MUT tumors due to either suppression of immune/inflammation genes in BRAF-MUT tumors or increased expression by the tumor or by infiltrating lymphocytes in BRAF-WT tumors. Moreover, up-regulation of HLA-G and CXCL14 in PTC epithelial cells may be directly responsible for impaired immune surveillance in BRAF-MUT compared with BRAF-WT tumors. Finally, a number of newly identified fusion transcripts may contribute to the biology of PTC. A critical clinical question arises from the observation that PTCs with BRAF mutations are heterogeneous with respect to immune function gene expression and lymphocyte infiltration. In the future, it will be important to determine if the immune status of such tumors has prognostic significance.