Identification of borderline thyroid tumors by gene expression array analysis




A subset of follicular lesions of the thyroid is encapsulated similar to follicular adenomas but with partial nuclear features suggestive of papillary thyroid carcinoma (PTC), raising the possibility of biologically borderline tumors.


Gene expression profiling and advanced significance analyses were performed on 50 histologically unequivocal benign and malignant tumors, and a list of 61 differentially expressed genes was generated. By using this 61-gene list, unsupervised hierarchical and K-means cluster analyses were performed on 40 additional tumors, including 15 histologically borderline tumors, 11 benign tumors, and 14 PTCs.


Analysis revealed 3 distinct tumor groups—benign, malignant, and intermediate. Tumors in the intermediate group (n = 15) were mostly histologic borderline tumors and had an expression profile overlapping with the benign and malignant groups. Twenty-seven genes were expressed differentially between the benign and intermediate groups, including the cyclic AMP response element-binding protein/p300-interactivator with glutamic acid/aspartic acid-rich carboxy-terminal domain 1 or CITED1 gene and the fibroblast growth factor receptor 2 or FGFR2 gene. Fourteen genes were expressed differentially between the intermediate group and malignant tumors, notably overexpression of the met proto-oncogene and of the high-mobility group adenine/thymine-hook 2 or HMGA2 gene in malignancies. Mutations of the v-raf murine sarcoma viral oncogene homolog B1 or BRAF gene were identified in 4 of 14 malignant tumors but not in benign or intermediate tumors. Patients who had either histologically or molecularly borderline tumors did not have metastasis or recurrences.


Gene expression profiling supported the finding that encapsulated thyroid follicular lesions with partial nuclear features of PTC are biologically borderline tumors that are distinct molecularly from benign and malignant tumors. Cancer 2009. © 2009 American Cancer Society.

Thyroid nodules are common in the US population and occur in >60% of individuals, and their incidence steadily is increasing mainly because of the increased detection of smaller, asymptomatic nodules.1, 2 Although the majority of these nodules are benign, a significant number undergo surgical excision. On pathologic review of thyroid tumors, clear-cut benign or malignant diagnoses often can be rendered; however, follicular lesions of the thyroid can pose a diagnostic challenge.

A particular diagnostic dilemma is presented in a subset of encapsulated follicular lesions with partial nuclear features of papillary thyroid carcinoma (PTC) (occasional nuclear grooves, focal nuclear clearing, and overlapping nuclei) and with histologic features that fail to place them reliably in either the benign category or the malignant category. In our experience, these tumors represent approximately 10% of all follicular-patterned lesions observed at surgical pathology.3 The difficulty of classifying this group of tumors is exemplified further by several studies in which poor interobserver agreement was demonstrated among expert endocrine pathologists ranging from 39% to 58% when they reviewed follicular-patterned lesions of the thyroid.4-8 This diagnostic difficulty in classifying such borderline tumors with standard terminology led Williams to propose the term well differentiated tumor of uncertain malignant potential (WDT-UMP) as a separate diagnostic category.9

The use of immunohistochemical and molecular studies to aid in differentiating benign from malignant thyroid tumors has been studied extensively. Although multiple immunohistochemical markers (cytokeratin 19 [CK19], human mesothelial cell 1 [HBME1], galectin-3, cyclic adenosine monophosphate response element-binding protein/p300-interacting 1 [CITED1]) have been evaluated for their usefulness in diagnosing PTC, variable sensitivity and specificity have been reported among studies.10-13 The use of these markers has been studied in a few so-called WDT-UMPs with variable results. Some investigators reported that staining patterns of CK19, galectin-3, HBME1, and CITED1 were similar to the patterns observed in benign lesions, others reported patterns that were similar to those observed in follicular variants of PTC (FVPTCs), yet other investigators observed that the staining intensity of these and other markers was intermediate between the intensity observed benign and malignant tumors.14-16 Limited molecular studies have been undertaken to further characterize the biologic nature of these lesions, although rearranged in transformation (RET)/PTC rearrangement has been evaluated with variable results.17

Microarray gene profiling has shown promise in the diagnosis of thyroid tumors. Several groups, including ours, have been able to use this platform to accurately discriminate benign tumors from malignant tumors.18-21 During such analyses, we recently encountered 5 cases that constituted a third group of tumors that were distinct molecularly from benign and malignant tumors in cluster analysis.22 Histologic review indicated that 4 of those 5 tumors were encapsulated follicular lesions with partial nuclear features of PTC. Given this information, we undertook the current study to further define this intermediate group of borderline thyroid tumors on a molecular basis, and we attempted to identify their molecular signatures.


Tumor Samples

Tissue samples were collected at time of surgery, snap-frozen in liquid nitrogen, and stored at −80°C. Representative slides for all tumors were reviewed by 2 pathologists (T.S. and Y.-T.C.). In total, 90 thyroid tumor samples, including 16 PTCs, 22 FVPTCs, 15 hyperplastic nodules, 22 follicular adenomas (FAs), and 15 histologically borderline tumors, were analyzed in this study. Borderline tumors were defined as encapsulated lesions with follicular architecture in which the morphologic features of PTC were qualitatively incomplete and that did not demonstrate evidence of capsular and/or vascular invasion. The incomplete features of PTC were widespread in the lesions that we analyzed in this study and did not represent focal findings in an otherwise benign nodule. An example of such a case is shown in Figure 1. In essence, these lesions could be classified as WDT-UMP, as proposed by Williams. The officially reported final diagnosis of the 15 borderline tumors, all of which were rendered before we started the current study, was 7 FAs and 8 FVPTCs. This study was approved by our Institutional Review Board.

Figure 1.

This photomicrograph shows a histologically borderline lesion. Note the follicular architecture with focal nuclear clearing and occasional nuclear grooves.

RNA Isolation and GeneChip Hybridization

RNA was extracted using an RNeasy Mini kit (Qiagen, Valencia, Calif) from frozen tissue following the manufacturer's protocol. RNA purity was confirmed by spectrophotometry. Total RNA was reverse transcribed to complementary DNA (cDNA) according to the manufacturer's protocol (NuGEN Ovation RNA Amplification System V2; NuGEN Technologies, Inc., San Carlos, Calif) and then labeled with biotin. Next, cDNA was hybridized to gene chips for microarray analysis using either GeneChip U95A or GeneChip U133A (Affymetrix, Santa Clara, Calif).

Microarray Data Analysis

ArrayAssist (version 5.2.2; Stratagene, Inc., La Jolla, Calif) was used for gene chip analysis. Interchip and intrachip normalization was carried out by using robust multichip analysis (RMA). After RMA, hybridization signals underwent variance stabilization, log transformation, and baseline transformation. Advanced significance analysis was performed on 50 U95A Gene Chips, including 10 hyperplastic nodules, 16 FAs, 13 FVPTCs, and 11 PTCs. This formed the training set. Gene expression of benign tumors was compared with that of malignant tumors. After Benjamini-Hochberg correction for false-discovery, gene probe sets with significant differential expression (≥2-fold with P<.05) were identified. Then, this probe list was converted to correspond to probes on the U133A Gene Chips (array comparison software; available at: accessed May 1, 2009). The remaining 40 tumor samples, all of which were analyzed with U133A Gene Chips, formed the test set. The test set was then assessed using unsupervised hierarchical cluster analysis and K-means cluster analysis with both 2-group and 3-group cluster algorithms. Genes that were expressed differentially between borderline tumors and benign and malignant tumors were identified further with advanced significance analysis.

Detection of BRAF Mutations

All 40 tumors that formed the test group were analyzed for v-raf murine sarcoma viral oncogene homolog B1 (BRAF) mutations in which glutamate was substituted for valine at codon 600. One microgram of RNA was reverse transcribed in a 20-μL reaction, and a 1 μL aliquot of cDNA was used for polymerase chain reaction (PCR) analysis. The following PCR primers were used: forward primer, 5′-TGCTTGCTCTGATAGGAAAATG-3′; reverse primer, 5′-GACTTTCTAGTAACTCA-GCAG-3′. Amplification was carried out for 35 cycles (at 94°C for 15 seconds, at 60°C for 1 minute, and at 72°C for 1 minute). All PCR products were observed by electrophoresis on a 2% agarose gel and were purified using a PCR purification kit (Qiagen). BRAF mutations were detected by direct sequencing of PCR products. All sequencing was performed bidirectionally using the Big Dye Terminator cycle-sequencing kit and the Applied Biosystems Automated 3730 DNA Analyzer (Applied Biosystems, Foster City, Calif).


Differentiation of Benign and Malignant Tumors

The training set consisted of 50 tumors, including 26 unequivocal benign tumors (16 FAs and 10 hyperplastic nodules) and 24 unequivocal malignancies (11 PTCs and 13 FVPTCs). In total, 66 probe sets corresponding to 61 genes displayed significant differential expression between benign tumors and malignant tumors. Thirty-one genes had up-regulated expression in malignancies compared with benign tumors, and 30 genes were down-regulated (Table 1).

Table 1. Sixty-One Genes Differentially Expressed Between Benign, Borderline, and/or Malignant Thyroid Lesions
Gene NameGene SymbolFold Change*P
  • GTP indicates guanine triphosphate; Cpb, cyclic adenosine monophosphate response element-binding protein; EGF, epidermal growth factor; Sec7, a guanine-nucleotide-exchange factor (also called ARNO3 and cytohesion 3); AT, adenine and thymine; TIMP, tissue inhibitor of metalloproteinase; cAMP, cyclic adenosine monophosphate.

  • *

    Fold change is shown relative to benign lesions.

Differentially expressed between benign and borderline/malignant lesions   
 Ankyrin 2, neuronalANK2−2.70.0281
 Rho GTPase activating protein 6ARHGAP6−2.42.0329
 Cadherin 16, kidney-specific cadherinCDH16−2.28.0185
 Collagen, type IX, alpha 3COL9A3−5.87.0160
 Chondroitin beta 1,4ChGn−3.72.0111
 Dual-specificity phosphatase 4DUSP4+3.69.0206
 EGF-containing fibulin-likeEFEMP1−2.60.0464
 engulfment and cell motility 1ELMO1−2.60.0261
 Fibroblast growth factor receptor 2FGFR2−2.13.0343
 Fibronectin leucine rich transmembrane protein 1FLRT1−2.10.0252
 Glycine amidinotransferaseGATM−2.41.0482
 V-kit Hardy-Zuckerman 4 felineKIT−3.85.0039
 Low-density lipoproteinLRP4+5.89.016
 Matrilin 2MATN2−3.38.0127
 Slit homolog 1 (Drosophila)SLIT1+3.35.0258
 Spectrin, alpha, nonerythrocytic 1SPTAN1+2.66.0160
 Transcription factor CP2-like 1TFCP2L1−3.54.0029
 Phosphoinositide-binding proteinPIP3-E−3.14.0343
 Pleckstrin and Sec7 domainPSD3+2.40.0169
 Pleckstrin and Sec7 domainPSD3+2.55.0214
 Tensin 3TNS3−2.41.0029
 Tetraspanin 12TSPAN12−2.35.0047
 T-cell lymphoma invasion and metastasis 1TIAM1+3.91.0160
Differentially expressed between malignant and borderline/benign lesions   
 Deiodinase, iodothyronine, type 1DIO1−4.47.0321
 Deltex 4 homolog (Drosophila)DTX4+3.68.0111
 Uridine diphosphate-N-acetyl-alpha-D-galactosamineGALNT7+2.07.0213
 High-mobility group AT-hook 2HMGA2+3.56.0204
 Insulin-like growth factor binding protein 6IGFBP6+3.18.0160
 Met proto-oncogeneMET+2.35.0182
 Protein SPROS1+3.97.0089
 Syndecan 4SDC4−3.26.0049
 Serpin peptidase inhibitor, clade ASERPINA1+5.64.0252
 Serpin peptidase inhibitor, clade ASERPINA1+4.81.0233
 Solute carrier family 4SLC4A4−4.03.0034
 TIMP metallopeptidase inhibitor 1TIMP1+2.72.0446
 Uridine phosphorylase 1UPP1+2.25.0127
 T-cell lymphoma invasion and metastasis 1TIAM1+3.91.0160
Differentially expressed only between benign and malignant lesions   
 Chromosome 11 open reading frame 17C11orf17−2.12.0239
 Calpain 3CAPN3+2.00.0263
 Calpain 3CAPN3+2.10.0410
 Creatine kinase, brainCKB−2.46.0189
 Cysteine and glycine-rich protein 2CSRP2−2.41.0189
 Death-associated protein kinase 2DAPK2+2.23.0322
 Dipeptidyl-peptidase 4DPP4+2.83.0127
 Dipeptidyl-peptidase 4DPP4+2.51.0117
 Homogentisate 1,2-dioxygenaseHGD−3.17.0149
 Myosin, heavy chain 10MYH10+2.73.0214
 Phosphonoformate immunoassociated protein 5PFAAP5+2.59.0189
 Phosphonoformate immunoassociated protein 5PFAAP5+2.28.0258
 Placental growth factorPGF−2.22.0301
 Myosin, heavy chain 10MYH10+2.73.0214
 PBX/knotted 1 homeobox 2PKNOX2−2.31.0455
 Protein kinase, cAMP-dependentPRKACB−2.20.0241
 Glytaminyl-peptide cyclotransferaseQPCT+3.43.0136
 RAB27A, member RAS oncogeneRAB27A+2.41.0111
 RAB27A, member RAS oncogeneRAB27A+2.08.0063
 Retinoid X receptor, gammaRXRG+2.57.0261
 Solute carrier family25SLC25A15−2.75.0261

Unsupervised Hierarchical Cluster Analysis

An independent set of 40 test samples also were characterized using the gene list that was generated by the training set. The test set included 15 histologically borderline tumors as well as a second group of unequivocally benign (n = 11) and malignant (n = 14) tumors, including 6 FAs, 5 hyperplastic nodules, 9 FVPTCs, and 5 PTCs. In an unsupervised hierarchical cluster analysis, all benign tumors again were separated from malignant tumors as expected (Fig. 2). In addition to these 2 groups, a third intermediate group was identified. This group contained 15 tumors, and most (10 tumors) were histologically borderline tumors. Three FVPTCs and 2 FAs also were identified in this group. Of the 5 remaining borderline tumors, 4 tumors clustered with the benign group, and 1 tumor clustered with the malignant group. It is noteworthy that these tumors were the most peripheral nodes in these 2 groups, indicating an expression profile closer to that observed in the intermediate group (Fig. 2).

Figure 2.

This graphic was generated by an unsupervised hierarchical cluster analysis. FA indicates follicular adenoma; HYP, hyperplastic lesion; BOR, borderline tumor; FVPTC, follicular variant of papillary thyroid carcinoma; PTC, papillary thyroid carcinoma.

K-Means Cluster Analysis

To help elucidate the differences in gene expression between the 3 groups of tumors, this test set also was subjected to K-means cluster analysis using both 2-group and 3-group cluster algorithms. In the 2-group cluster algorithm, tumors were separated into 2 groups based on expression of the 61 genes of interest. This algorithm distinguished benign and malignant tumors with 93% sensitivity and 82% specificity (Fig. 3). Borderline tumors were divided: Four tumors (27%) were grouped with the benign tumors, and 11 tumors (73%) were grouped with the malignant tumors.

Figure 3.

This graphic was generated a 2-group K-means cluster analysis. FA indicates follicular adenoma; BOR, borderline tumor; FVPTC, follicular variant of papillary thyroid carcinoma; PTC, papillary thyroid carcinoma; HYP, hyperplastic lesion.

In the 3-group cluster algorithm, tumors were separated into 3 designated groups based on their expression profile. With this algorithm, malignant tumors primarily formed 1 group (with 1 borderline tumor), benign tumors formed a second group (with 4 borderline tumors), and a third group included 10 borderline tumors, 2 FAs, and 3 FVPTCs (Fig. 4). The 2 FAs were grouped previously with the malignancies in the 2-group algorithm (FA-3 and FA-4), and 1 of the 3 FVPTCs that was grouped with the borderline tumors (FVPTC-3) previously had clustered with the benign tumors in the 2-group algorithm.

Figure 4.

Three-group K-means cluster analysis identified 3 distinct groups: malignant (left), benign (center), and intermediate (right). BOR indicates borderline tumor; FVPTC, follicular variant of papillary thyroid carcinoma; PTC, papillary thyroid carcinoma; FA, follicular adenoma; HYP, hyperplastic lesion.

Correlation With Final Clinical Diagnosis and Patient Follow-Up

Of the 15 borderline tumors that were included in this study, 7 tumors were diagnosed officially and reported as FA (47%), and 8 tumors (53%) were FVPTCs. Upon review of the 2-group K-means cluster analysis of these 15 borderline tumors, a correlation between the reported diagnosis and the cluster group was observed in only 6 of 15 tumors (40%), underscoring the diagnostic dilemmas that pathologists face with these tumors (Table 2). In addition, no borderline tumors were associated with lymph node metastasis or distant metastasis. Of the 9 patients with histologically borderline tumors who were followed, 6 patients were diagnosed officially with FVPTC, and none developed a recurrence after surgery (mean follow-up, 1.7 years; range, 2 months to 4.4 years) based on thyroglobulin level, ultrasound studies, or a combination of both methods. Similarly, among the 3 patients with FVPTC in the molecularly intermediate group (FVPTC-1, FVPTC-2, and FVPTC-3), none had lymph node metastasis, extranodal extension, or recurrent disease at a follow-up of 23 months, 23 months, and 25 months, respectively.

Table 2. Comparison of Final Diagnosis With 2-Group Clustering of Borderline Tumors
SamplePathologic DiagnosisCluster GroupConcordance
  • BOR indicates borderline tumor; FVPTC, follicular variant of papillary thyroid carcinoma; FA, follicular adenoma.

  • *

    With 3-group clustering, these tumors fell into the intermediate category.

BOR 2FABenignYes
BOR 3FAMalignant*No
BOR 4FVPTCMalignant*Yes
BOR 5FAMalignant*No
BOR 6FAMalignant*No
BOR 7FVPTCMalignant*Yes
BOR 8FVPTCMalignant*Yes
BOR 9FAMalignant*No
BOR 10FAMalignantNo
BOR 11FVPTCMalignant*Yes
BOR 12FVPTCMalignant*Yes
BOR 13FAMalignant*No

Gene Signature of Borderline Tumors

To identify gene expression profiles that distinguish borderline tumors from either benign tumors or malignant tumors, another advanced significance analysis was carried out. Twenty-seven of 61 genes had significant differential expression between benign tumors and borderline tumors, whereas 14 of 61 genes had significant differential expression between malignant tumors and borderline tumors. Only 1 of these genes (T-cell lymphoma invasion and metastasis 1 or TIAM1) overlapped between the 2 comparisons (Fig. 5). Of the 27 genes that distinguished benign tumors from borderline tumors, 8 genes had up-regulated expression in borderline tumors, including CITED1, and 19 genes were down-regulated including, fibroblast growth factor receptor 2 (FGFR-2) (Table 1). Of the 14 genes that distinguished malignant tumors from borderline tumors, 11 genes were up-regulated in malignant tumors, including the met proto-oncogene (MET) and the high-mobility group adenine/thymine-hook 2 gene (HMGA2); whereas 2 genes were relatively down-regulated, notably, deiodinase 1 (DIO1) (Table 1).

Figure 5.

This Venn diagram illustrates differentially expressed genes relating 61 genes to benign, borderline, and malignant tumors.

Mutational Analysis of v-raf Murine Sarcoma Viral Oncogene Homolog B1

BRAF mutational analysis was performed on all tumors in the test set. BRAF mutations were identified in 4 of 14 malignant tumors (29%) (Table 3). No borderline tumors or benign tumors had BRAF mutations.

Table 3. Mutations of v-Raf Murine Sarcoma Viral Oncogene Homolog B1 in Tumors
DiagnosisNo. With BRAF Mutation (%)
  1. BRAF indicates v-raf murine sarcoma viral oncogene homolog B1; PTC, papillary thyroid carcinoma; FVPTC, follicular variant of papillary thyroid carcinoma; BOR, borderline tumor; FA, follicular adenoma.

PTC, n=53 (60)
FVPTC, n=91 (11)
BOR, n=150 (0)
FA, n=60 (0)
Hyperplastic nodule, n=50 (0)


Encapsulated follicular lesions with cytologic atypia remain a diagnostic challenge for pathologists. In the current study, we used molecular profiling to identify a third category of thyroid tumors that, based on gene expression data, is likely to be premalignant. This third category of encapsulated follicular tumors with cytologic atypia typically has not fit into previously proposed benign or malignant classification schemes using standard histology, immunohistochemistry, or mutation analysis. In this study, we observed that the majority of histologically borderline tumors (66.7%) will fall into an intermediate group, and few share gene expression similarities with benign tumors (26.7%) or malignant tumors (6.7%; K-means cluster analysis) (Fig. 4).

Many genes that were expressed differentially between benign tumors and malignant tumors in our training set were classic markers of PTC, as documented previously by gene profiling studies, including CITED1; dipeptidyl-peptidase 4 (DPP4); FGFR2; and serpin peptidase inhibitor, clade A (SERPINA1).15, 20, 23 It is noteworthy that borderline tumors, like malignant tumors, already had up-regulated gene expression of CITED1 and pleckstrin and Sec7 domain 3 (PSD3) and down-regulated gene expression of FGFR2 relative to benign tumors (Table 1). These genes and others listed in Table 1 are potential markers of early tumorigenesis. In contrast, some genes that were altered consistently in malignant tumors were unchanged in the borderline group. For instance, DIO1, a differentiation marker that consistently is lost in PTC, was retained in this borderline group. Conversely, MET, SERPINA1, tissue inhibitor of metalloproteinase 1 (TIMP1), and HMGA2, genes that often are activated or overexpressed in PTC, had lower expression in the borderline group relative to the malignant group. These genes potentially may represent gene expression changes that are involved in the later stages of cancer development. These findings support our proposal that the histologically borderline tumors are premalignant and still lack the complete phenotype of PTC. The contribution of the genes identified in this study in the progression of these lesions and their potential role in carcinogenesis need to be evaluated further with additional studies.

The results of BRAF mutation analysis also were in agreement with previous reports in the literature,24 with mutations identified in 29% of malignancies. To date, BRAF mutations have not been identified in benign lesions or in borderline encapsulated follicular tumors,3, 25 similar to findings in this study. Several studies recently demonstrated that BRAF mutations are associated with more aggressive tumors,26, 27 suggesting that borderline tumors are more likely to be indolent tumors. The finding that BRAF mutation is more frequent in classic PTC than in FVPTC24, 28, 29 also supports the notion that, for FVPTCs derived from FAs, as we propose in the current report, BRAF either is uninvolved in carcinogenesis or is involved only as a late event. In addition, because of its higher frequency in classic PTC versus FVPTC, BRAF mutational analysis remains of limited usefulness in the diagnostic evaluation of these lesions.

Of the 15 histologically defined borderline tumors in this study, 10 were clustered in an intermediate group, separate from benign and malignant clusters (Fig. 3). It is noteworthy that not all borderline tumors were separated into this third group: One tumor was clustered with malignant tumors, and 4 tumors were clustered with benign tumors. Conversely, 3 histologically unequivocal FVPTCs and 2 FAs were identified in the molecularly intermediate group. Given our notion that FAs, borderline tumors, and FVPTCs probably are 3 entities in a biologic continuum, this imperfect correlation between the histologic classification and molecular prediction is fully expected.

To date and to the best of our knowledge, only 1 other study in the literature has attempted to classify thyroid tumors of uncertain malignant potential using gene expression profiling with microarray techniques.25 Those authors provide a pathologic scoring system (using histopathology and immunohistochemistry) correlated with the gene expression profile to separate tumors of uncertain malignancy (T-UM) into benign or malignant groups of tumors. Because the objective of that recent study was to identify which T-UMs actually were malignant on the basis of gene profiling, the possibility of a premalignant category was not specified. Nonetheless, the authors do acknowledge the likelihood that thyroid cancer tumorigenesis probably is a spectrum of disease, beginning with benign tumors and ending with clear-cut malignancy. Moreover, they point out that they were able to find several genes that are unique to the gene-expression signature of T-UM compared with benign or malignant tumors. This is in accordance with our findings and supports the notion of a unique class of tumors.

The idea that thyroid cancer tumorigenesis is preceded by a preneoplastic or premalignant condition has been addressed previously in the literature.17, 30, 31 Pennelli et al. suggested that the progression of thyroid carcinogenesis begins with nuclear atypia of low-grade chromatin clearing; followed by high-grade thyroid intrafollicular neoplasia with nuclear grooves, pseudoinclusions, and overlap; and followed finally by the appearance of intrafollicular papillary carcinoma.30 Fusco et al. described the presence of RET/PTC rearrangements in follicular tumors with borderline morphologic features for PTC and suggested that these tumors indeed may represent premalignant disease.17 Cytologic alterations suggestive of PTC are well known in Hashimoto thyroiditis and have been suggested as possible precursors to malignancy, and immunophenotypic changes characteristic of PTC and RET/PTC rearrangements have been demonstrated in some cases.31-35 Gene expression profiling of borderline tumors in the current study provides additional support for the idea of progression from non-neoplastic disease and FA to carcinoma.

The use of immunohistochemical markers has been studied in a few WDT-UMPs with variable results. Papotti et al. studied the expression of galectin-3 and HBME1 in 13 WDT-UMPs and noted some degree of staining with either antibody in 12 of 13 tumors, concluding that this provided evidence supporting a relation between PTCs and these lesions.14 Immunohistochemical staining for HBME1, Galectin-3, and CK19 (data not shown) in the histologically borderline tumors that we studied revealed heterogeneous staining patterns. This variability, again, may reflect the biologically borderline nature of these tumors.

Unfortunately, part of the problem with standard diagnostic tools is an inherent need by clinicians to separate tumors into benign or malignant categories for patient and clinician satisfaction. Partially for that reason, the term WDT-UMP proposed by Williams and by Rosai has not been embraced in practice and certainly is not in use at most institutions.9, 36 Consequently, the majority of borderline tumors, as in the current study, probably are diagnosed as FVPTCs because of pathologists' general preference to err on the side of over-diagnosis for potential legal concerns. Then, these patients are subjected, perhaps unnecessarily, to completion thyroidectomies, central neck dissections, and even radioactive iodine therapy. With the current 2-tiered classification (benign and malignant), our 2-K means cluster would place 73% of histologically borderline tumors in the malignant category. This certainly does not correlate with the clinical behavior of these tumors.

Several groups have reviewed the outcome of patients with encapsulated PTC, including both classic PTC and FVPTC.37-39 Liu et al. reviewed the outcome data from 42 patients with encapsulated, noninvasive FVPTCs who had a median 10-year follow-up and reported that no patients had recurrences and that none had lymph node metastasis.37 Vickery identified 10 patients who had encapsulated papillary cancers; in those patients, none had a recurrence, and only 1 patient had developed lymph node metastasis at a median follow-up of 15 years.38 Evans identified 7 patients who had encapsulated PTC and reported no recurrences or distant metastases at a median follow-up of 13.5 years.39 The number of studies that specifically have investigated tumors with borderline features is limited, although no tumor recurrences have been reported.17 Likewise, none of the patients with borderline tumors in the current study had lymph node metastasis, and none of those with clinical follow-up developed recurrent disease or distant metastasis.

Although limited follow-up data were available on the histologically borderline tumors in our study, all the tumors had an indolent clinical course, suggesting a good prognosis for patients with these tumors. It seems reasonable to suspect that the molecular classification may be closer to the true clinical nature of these tumors. However, this hypothesis would need to be proven by the analysis of a larger series of such borderline tumors with careful long-term clinical follow-up to truly define the nature of these lesions. This type of study would be worth pursuing.

The terminology regarding these lesions remains controversial. The difficulty in both diagnosis and classification of these lesions is reflected in the poor interobserver agreement among expert endocrine pathologists. The data presented here provide evidence that histologically borderline tumors represent a molecularly distinct group of tumors that need to be evaluated further with additional molecular studies and longer follow-up.

Conflict of Interest Disclosures

The authors made no disclosures.