Introduction

Medullary thyroid carcinoma (MTC) is an uncommon cancer derived from neuroendocrine C-cells of the thyroid. The prognosis is very variable, with an estimated overall 10-year survival of around 70% [1]. Tumor stage is a strong predictor of survival, being the presence of lymph node metastasis as the most relevant prognostic parameter [2].

In contrast to other neuroendocrine neoplasia, a histological classification to grade these tumors was lacking. However, in 2021, a morphological two-tyer grading system was described and proved to have a strong prognostic value. This system classifies tumors as a high grade when any of these features are present: necrosis, ≥ 5 mitosis/2 mm2, or a Ki-67 proliferation index ≥ 5% [3]. After its initial description in 2021, this system has been independently validated [4], and it is endorsed by the most recent WHO classification of thyroid neoplasms [5].

The most frequent molecular alterations in MTC are RET mutations. These are present in around 50% of sporadic tumors and in virtually all hereditary patients. The second most frequent alterations are RAS mutations that constitute approximately 25% of sporadic cases. Thus, around 25% of patients harbor tumors that are negative for RET and RAS mutations, which are commonly described as “double wild type” [6]. Patients with RET-mutated tumors have been described as having a worse prognosis. In addition, recent studies with large cohorts have found an association between RET mutation and high-grade MTC [7, 8].

Gene expression studies in MTC have proven useful for characterizing altered molecular pathways in tumors with more aggressive mutations, such as RET M918T [9], and for identifying potential therapeutic targets [10]. These studies have also defined molecular subgroups, including a “proliferative-like subtype” characterized by enhanced proliferative activity in juxtaposition to a “mesenchymal-like subtype” characterized by epithelial-to-mesenchymal transition signatures [11]. However, to date, differences in gene expression profiles of MTCs using the new 2021 classification have not been evaluated. Given the prognostic impact of MTC grade, our study explores differences in the expression profile between high- and low-grade MTC.

Methods

Patients and Tissue Specimens

The study was approved by the Hospital Universitario Ramón y Cajal Research Ethics Committee (approval code: 85/22). A total of 30 MTC specimens obtained between 2004 and 2022 were included based on tissue availability. Medical records were reviewed for clinical parameters and follow-up. Disease-free survival (DFS) was defined as the time from the date of radical treatment to tumor relapse or death due to any cause, whatever occurred first. Overall survival (OS) was defined as the time from the date of treatment initiation to death due to any cause. Slides were reviewed by two pathologists (I.R.C. and C.A.-P.), and tumor grade was noted according to the International Medullary Thyroid Carcinoma Grading System [3]. Desmoplasia was defined as “newly formed fibrotic (collagenous) stroma surrounding the invasive epithelial tumor cells not found in the non-neoplastic thyroid parenchyma,” as previously described [12]. The presence of amyloid was confirmed with Congo red staining.

Nucleic Acid Extraction

Nucleic acid extraction was performed after identifying tumor areas containing at least 60% of tumor cells. Macrodissection was performed using formalin-fixed paraffin-embedded (FFPE) blocks. AllPrep DNA/RNA FFPE kit (Qiagen, Valencia, CA, USA) was used to extract DNA and RNA for next-generation sequencing following the manufacturer’s instructions. DNA and RNA were fluorometrically quantified by Qubit dsDNA BR assay kit (Invitrogen, Carlsbad, CA, USA) and Qubit RNA high-sensitivity assay kit (Invitrogen), respectively. Recover All Total Nucleic Acid Isolation Kit (Invitrogen) was used to extract RNA for gene expression analysis following the manufacturer’s instructions. RNA quality was assessed using RNA Screen Tapes on a 2200 TapeStation system (Agilent, Santa Clara, CA, USA).

Next-Generation Sequencing (NGS)

Mutational analysis was performed with the NGS panel Oncomine Focus Assay (OFA, Thermo Fisher Scientific, Waltham, MA, USA). OFA is a panel that screens 52 genes: it detects hotspot mutations in 35 genes (including RET, HRAS, and NRAS), copy number variations (CNVs) in 19 genes, and 23 gene fusions (inter- and intragenic). Libraries were automatically prepared using an Ion Chef Instrument (Thermo Fisher Scientific, Waltham, MA, USA). Subsequently, NGS libraries were sequenced with an Ion S5 using the Ion 530 Chef Kit (Thermo Fisher Scientific). Reads were aligned with the GRCh37-hg19 human reference genome. Potential mutations, copy number alteration, and fusions were called using Ion Reporter™ Software (5.10 version-Thermo Fisher Scientific). These processes were performed following the manufacturer’s instructions.

Gene Expression Analysis

The NanoString nCounter gene expression platform was used to analyze the expression of 760 human mRNAs included in the Tumor Signaling 360 Panel (NanoString Technologies Inc., Seatle, WA, USA) [13]. These mRNAs are involved in the tumor biology, the microenvironment, and the immune response. Each sample requires an input between 100 and 150 ng of total RNA. Fluorescently color-coded reporter probes and biotin-labelled capture probes were hybridized to the mRNA on a thermal cycler overnight and automatically processed and loaded to the NanoString sample cartridge provided in the nCounter Prep Station in accordance with the manufacturer’s protocol.

qRT-PCR was used to validate the expression of selected genes using in-house designed primers (Online Resource 1). After RNA extraction, retrotranscription was performed with the High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific), following the manufacturer’s instructions. Quantitative PCR was performed using the iTaq Universal SYBR Green Supermix Kit (BIORAD), following the manufacturer’s instructions. Data analysis was performed by quantifying the expression levels of the indicated genes, using a relative quantification ΔΔCt method. The reference gene used was GAPDH.

Immunohistochemistry

Immunohistochemistry (IHC) was performed on whole sections of 4 μm thick mounted on positively charged slides. Antibody against DLL3 (clone SP347, ready-to-use, Ventana, Roche) was incubated for 32 min on a Ventana Benchmark XT Immunostainer (Roche, Basel, Switzerland), after pretreatment with the Cell Conditioning 1 (CC1) solution for 64 min at 100 °C. Antibody against Ki67 (clone MIB-1, ready-to-use, Agilent) was incubated on a Dako Omnis (Agilent).

The Ki67 proliferation index was assessed by two pathologists using a multiheaded microscope. Both pathologists agreed on the hot spot area. Subsequently, one pathologist counted manually between 500 and 2000 tumor cells to determine the ratio of positive cells. DLL3 was quantified independently by two pathologists (I.R.C. and I.C.-B.) blinded to the molecular data of each tumor. Each slide had a small cell lung carcinoma (SCLC) with adjacent non-tumoral lung parenchyma that were used as positive and negative controls, respectively. Moreover, additional slides from selected tumors were incubated without the primary antibody. The percentage of tumor cells was evaluated using the 4 × objective lens, following the methodology previously published [14]. Then, the mean percentage of positive tumor cells of each tumor was obtained, and three groups were considered according to previous reports in MTC [15]: 0%, low (1–49%, and high (≥ 50%).

Statistical Analysis

NanoString gene expression results were analyzed using nSolver analysis software (NanoString Technologies Inc.). Identification of differentially expressed genes was conducted on normalized data. This software allowed us to obtain the hierarchical grouping, the scatter diagrams (volcano plots), and statistical classification of the differentially expressed genes, along with false discovery rate corrected p-values.

Overrepresentation analysis was performed using the “genekitr” package in RStudio v4.2.1 (RStudio, Boston, MA, USA) on the differentially expressed genes. These genes were queried in the curated pathways from the Reactome database contained within the Molecular Signatures Database (accession date: June 5, 2024). Numerical variables were summarized as median (range) and categorical variables as frequencies and percentages. The chi-squared or Fisher’s exact test was used to evaluate the association between qualitative variables. Mann–Whitney’s U test was used to evaluate the association between quantitative variables. DFS and OS data were plotted in Kaplan–Meier curves, and the log-rank test was used to compare these parameters. Data were analyzed using the statistical software IBM SPSS v19 (IBM, Armonk, NY, USA) and RStudio was employed for graphical representation. The threshold for statistical significance was set at p < 0.05.

Results

Cohort Characteristics

A total of 30 patients with MTCs were included in the current study. There were 11 (36.7%) patients with high-grade MTCs and 19 (63.3%) with low-grade MTCs. Table 1 shows the clinicopathological characteristics of the cohort according to the MTC grade. The median age was 50.5 years (range: 24–84 years) with a female-male ratio of 2:1. Ten (90.9%) patients with high-grade MTCs had lymph node metastases at diagnosis compared to 6 (31.6%) patients with low-grade MTCs (p = 0.002). In contrast, no patients in the high-grade subgroup were diagnosed at AJCC stage I compared to 8 (42.1%) patients with low-grade MTCs.

Table 1 Clinicopathological features of patients with MTC
Full size table

Regarding pathological features, high-grade MTCs were larger than low-grade MTCs (median: 5 cm vs 2 cm, respectively, p value < 0.01). High-grade MTCs showed a higher proportion of tumors with vascular invasion (p < 0.001), microscopic extrathyroidal extension (p = 0.001), macroscopic extrathyroidal extension (p < 0.001), positive resection margin (p = 0.001), and desmoplasia (p = 0.023). Necrosis represented less than 1% of the tumor area, except in one tumor where it accounted for up to 30% of the area.

Mutational status was analyzed in 24 MTCs including all high-grade MTCs. Cases without this information included five tumors measuring less than 1 cm and one tumor that did not reach the quality parameters. The distribution of mutations was as follows: 16 RET mutations (12 RET p.M918T, 1 RET p.C630I, 1 RET p.C634Y, 1 RET p.A883F, and 1 RET p.D898_E901del), 3 HRAS p.Q61R mutations, and 2 KRAS mutations (1 KRAS p.Q61R and 1 KRAS p.G12V). Three tumors with RET p.M918T mutations harbored additional mutations. One high-grade MTC had a PIK3CA p.E39K. Another high grade had a PIK3CA p.D1045N and a BRAF p.T599I. A low-grade MTC had a JAK1 p.V658I and a DDR2 p.R124W (Online Resource 2). Germline RET status was not known.

Differential Gene Expression Between High-Grade and Low-Grade MTC

Transcriptomic analysis using the NanoString platform was performed in 21 MTCs with available mutational status. Representative tumor areas from 9 high-grade and 12 low-grade MTCs were included. Unsupervised clustering analysis was not able to segregate high-grade from low-grade MTCs (Fig. 1A). There were eleven differentially expressed genes according to grade: EGLN3, EXO1, UBE2T, UBE2C, FOXM1, CENPA, DLL3, CCNA2, SOX2, KIF23, and CDCA5 (Fig. 1B). Moreover, we analyzed transcriptomic differences according to the two morphological elements that give rise to the grade: the presence of necrosis and a high proliferation index defined by a Ki67 ≥ 5% (Online Resource 3). The Venn diagram showed that there were only two genes exclusively overexpressed in the high-grade MTC subset: DLL3, an inhibitory NOTCH pathway ligand associated with neuroendocrine carcinomas [16], and CENPA, a protein-coding gene associated with centromere function (Fig. 1C). In contrast, there were four genes (CCNA2, CDCA5, FOXM1, and UBE2C) differentially expressed among those three conditions.

Fig. 1
figure 1

Expression profile according to MTC grade. A Unsupervised clustering analysis based on molecular pathways did not cluster separately high-grade from low-grade MTC. B Volcano plot showing differentially expressed genes according to grade. C Venn diagram including the genes differently expressed among three morphological conditions: high-grade vs low-grade MTC, proliferation index ≥ 5% vs < 5%, and tumors with necrosis present vs absent. D Top fifteen overrepresented pathways considering the genes differentially expressed between high-grade and low-grade MTC

Full size image

The most significant pathways differentially expressed between high-grade and low-grade MTCs were DNA damage repair (p = 0.013) and p53 signaling (p = 0.013); followed by cell cycle (p = 0.016), apoptosis (p = 0.023), Myc signaling (p = 0.028), and mTOR signaling (p = 0.047). In addition, we conducted an independent overrepresentation analysis considering the eleven differentially expressed genes according to grade. As shown in Fig. 1D, the fifteen most overrepresented pathways from the Reactome database were related to cell cycle progression and DNA repair.

A panel of eight genes was further validated by qRT-PCR in 25 MTCs (11 high grade and 14 low grade) (Fig. 2). This panel included the four genes overexpressed in the three conditions (CCNA2, CDCA5, FOXM1, and UBE2C); EGLN3, the most overexpressed gene in high-grade MTC; SOX2, a cancer stem cells related transcription factor overexpressed in high-grade follicular-derived thyroid carcinomas [17]; the two genes overexpressed exclusively in a high-grade subset (CENPA and DLL3); and ASCL1, not included in the NanoString Tumor Signaling 360 Panel. ASCL1 is a known upstream regulator of DLL3 expression [18]. qRT-PCR validation showed the eight genes to be overexpressed in high-grade MTC compared to low-grade MTC (p < 0.001) (Fig. 2). Moreover, there was a positive correlation between the levels of DLL3 and ASCL1 (r = 0.77, 95% C.I. 0.54–0.89, p = 6.738e-06), and between the levels of SOX2 and ASCL1 (r = 0.73, 95% C.I. 0.46–0.87, p = 3.963e-05) (Online Resource 4).

Fig. 2
figure 2

Panel of genes validated by qRT-PCR. *p < 0.001

Full size image

DLL3 Expression Impact on Prognosis

DLL3 is a protein implicated in neuroendocrine tumorigenesis and is known to be upregulated in SCLC subtype A (A means associated with ASCL1 upregulation) [19]. Since this protein is being explored as a target for conjugated drugs, DLL3 expression by IHC has been extensively studied in SCLC [14]. Therefore, we further quantified DLL3 protein expression by IHC. DLL3 diffusely stains the cytoplasm of tumor cells in MTC with variable intensity (Fig. 3A–F). Interestingly, a perinuclear strong dot-like staining was observed in scattered cells in some tumors (Fig. 3G). The non-tumor thyroid was negative for DLL3 expression, except for granular staining related to colloid content observed in some normal follicles (Fig. 3H). This granular staining was interpreted as an artifact because it was also present, though with weaker intensity, in slides incubated without the primary antibody. The proportion of positive tumor cells was measured, as previously described for SCLC [14]. There was a tendency towards an increased number of high-grade tumors with high expression of DLL3, although it was non-significant (Table 2). The presence of desmoplasia was associated with higher expression of DLL3 (Fig. 3). In tumors with desmoplasia, the mean expression of DLL3 was 55.18 ± 32.03, in contrast to a mean expression of 14.36 ± 21.58 in tumors without desmoplasia (p = 0.01). Nevertheless, considering the 24 tumors with a known mutational status, there were no significant differences in DLL3 expression between RET mutated tumors (50.09 ± 35.85) compared with RET wild type (28.88 ± 30.89) (p = 0.157).

Fig. 3
figure 3

DLL3 immunohistochemical expression in MTC. A Low-grade MTC and B high-grade MTC with low expression of DLL3 (DLL3, 40 ×). C Low-grade MTC with a high cytoplasmic expression of DLL3 of variable intensity among tumor cells (DLL3, 40 ×). D High-grade MTC with high expression of DLL3 and central comedonecrosis (DLL3, 20 ×). E High-grade MTC with prominent desmoplasia (H-E, 10 ×) and F a high expression of DLL3 (DLL3, 10 ×). G Strong perinuclear dot-like staining in scattered cells (DLL3, 40 ×). H Non-specific granular staining in normal follicles (DLL3, 40 ×)

Full size image
Table 2 DLL3 expression according to MTC grading
Full size table

Because we had survival data from our patients, we calculated DFS and OS by taking the cutoff of ≥ 50% positive tumor cells as high expression of DLL3, as previously considered in MTC [14, 15]). MTC grade had an impact on prognosis. After a median follow-up of 63.87 months (range 5.57–220.53 months), high-grade MTC showed a significantly worse prognosis in both, disease-free survival (p < 0.0001) and overall survival (p < 0.00049), compared with low-grade MTC (Fig. 4A, B). DLL3 overexpression, defined as ≥ 50% of positive tumor cells, was also associated with a significantly lower disease-free survival (p = 0.041) (Fig. 4C) and overall survival (p = 0.01) (Fig. 4D). The hazard ratio was not calculated because there were few or no events in the groups with the most favorable prognosis.

Fig. 4
figure 4

Kaplan–Meier survival curves according to grade and DLL3 expression. A Disease-free survival according to grade. B Overall survival according to grade. C Disease-free survival according to DLL3 positive expression. D Overall survival according to DLL3 positive expression

Full size image

Discussion

The MTC grading system is a novel prognostic parameter that identifies patients at a higher risk of relapse and lower overall survival. Going deeper, we have additionally shown that there are transcriptomic differences between high-grade and low-grade MTC.

Using the NanoString Tumor Signaling 360 Panel, we were able to identify eleven genes differentially expressed between high-grade and low-grade MTC. The grade is based on proliferation activity and necrosis. This is consistent with the finding that some of the most differentially expressed pathways are DNA damage repair, cell cycle, and apoptosis. Other pathways overexpressed in high-grade MTC were p53 signaling and Myc signaling. It is interesting that Qu et al. [11] using RNA-seq from MTCs identified a proliferative subtype that upregulated Myc targets and DNA repair genes. This cluster, identified by unsupervised clustering analysis, comprised most patients with relapse in their series, and it partly agrees with those pathways identified by us in high-grade MTC. Therefore, the Myc signaling pathway seems to play a role in aggressive MTC as occurs in SCLC [20]. In addition, the p53 pathway has been described as upregulated in MTCs harboring RET M918T mutation, considered to be an aggressive subgroup [9]. Furthermore, TP53 mutations have recently been identified as an adverse molecular event associated with decreased overall survival in MTC [7]. The fact that we have demonstrated for the first time an association of the upregulation of some genes such as EGLN3, SOX2, and DLL3 with aggressive MTC was due to our novel approach, considering morphological grade to stratify patients instead of RET mutational status [9, 10], as well as from using a panel of genes focused on cancer pathways instead of RNA-seq. Interestingly, all our high-grade RET-mutated MTCs harbored the RET M918T mutation, which was also present in 3 low-grade MTCs.

EGLN3 was the most differentially expressed gene between both conditions. EGLN3 encodes for a prolyl hydroxylase, also known as PHD3. It is involved in a variety of processes such as HIF factors regulation [21], apoptosis induction [22], and p53 protein stabilization [23]. Its prognostic value in cancer is not universal. For example, in clear cell renal carcinoma, where HIF induction has a pathogenic role, EGLN3 is overexpressed. However, its overexpression has been associated with a better recurrence-free survival [24]. In contrast, there are conflicting reports on non-small-cell lung cancer regarding its prognostic value [25, 26]. In our series, EGLN3 overexpression was associated with the presence of necrosis but not with a high proliferation index. Morphological necrosis in MTC is characterized by the presence of debris and apoptotic cell figures, so EGLN3 overexpression could reflect increased apoptosis. EGLN3 is induced by hypoxia and suppressed HIF activity by hydroxylating HIF1α and HIF2α to target them to proteasomal degradation [27]. However, in our panel, HIF1α was not associated with grade, nor with the presence of necrosis. In this regard, HIF1α expression has been previously associated with a lower overall survival and progression-free survival in MTC [28]. This is the first report of EGLN3 associated with a worse prognostic parameter in MTC.

The four genes that were overexpressed in the three morphological conditions (high grade, high proliferation, and the presence of necrosis) are genes required for cell cycle progression: Cyclin A2 (CCNA2), Cell Division Cycle Associated 5 (CDCA5), Forkhead Box M1 (FOXM1), and Ubiquitin Conjugating Enzyme 2C (UBE2C). The upregulation of these genes is expected in conditions with increased proliferation, and they are commonly identified as worse prognostic parameters in many cancer datasets [29, 30]. For example, UBE2C participates in the regulation of cell cycle progression through the M phase. Its overexpression has been associated with a release from the inhibitory signal of mitotic spindle checkpoint and reentry into mitosis. Moreover, it can induce an uncontrolled anaphase-promoting complex/cyclosome (APC/C) activity that led to chromosome missegregation and progressive aneuploidy [31]. UBE2C upregulation has been found in cancers from many origins, including follicular cell–derived thyroid carcinomas [31, 32].

The gene sex-determining region Y-box 2 (SOX2) was also upregulated in high-grade MTC and MTC with necrosis. This gene has a main role in the maintenance of the stem cell phenotype and self-renewal, being one of the Yamanaka factors [33]. In addition, it is considered an oncoprotein that controls several features including uncontrolled proliferation, resistance to apoptosis, cancer stem cell formation, migration, invasion, and epithelial-to-mesenchymal transition. In most cancer types, it is associated with poor prognosis and high tumor grade [34, 35], and it is expressed in anaplastic thyroid carcinoma, a follicular-derived thyroid neoplasia with an ominous prognosis [36]. To our knowledge, this is the first report of an association between SOX2 overexpression and high-grade or aggressive features in MTC.

Two genes were upregulated in high-grade MTC that did not appear in the subsets of proliferation or necrosis: CENPA and DLL3. CENPA encodes for a histone H3-like nucleosomal protein found in centromeric nucleosomes, which is also required for the assembly of kinetochore proteins, so it is implicated in cell cycle progression [37, 38].

DLL3 encodes for Delta-like 3 protein (DLL3), an inhibitory ligand of the Notch signaling pathway. Neuroendocrine carcinomas, such as SCLC, express DLL3 in more than 85% of tumors [14], and in the digestive system, DLL3 is overexpressed in neuroendocrine carcinomas compared with neuroendocrine tumors [39]. DLL3 is a recognized downstream target of achaete-scute homologue 1 (ASCL1), which is a key factor in neuroendocrine cell differentiation. Moreover, ASCL1 upregulation defined the most frequent molecular subtype of SCLC (SCLC-A), which also upregulated DLL3 and SOX2 [19, 40]. In our series, both DLL3 and SOX2 correlate with ASCL1 expression, being the three genes upregulated in high-grade MTC.

Taking into consideration the MTC mutational status, there was no association between the presence of RET mutation and the level of expression of DLL3 by IHC, although there was a trend towards overexpression in RET-mutated tumors. This is an issue that should be examined further in a larger cohort of patients.

The presence of desmoplasia in MTC has been associated with the presence of lymph node metastases, and it has been proposed as a parameter to report to reduce lymph node dissections in patients with low risk of lymph node metastases [12, 41, 42]. Moreover, DLL3 expression has already been associated with the presence of desmoplasia and lymph node metastases [43], as confirmed by our results. Clinicopathological data are lacking to establish a strict comparison between DLL3 expression in the two available MTC cohorts [15, 43] and our cohort. As it happens in our cohort, in both series published, around half of the tumors had high expression of DLL3 (defined as ≥ 50% of positive tumor cells). In addition, the present study is the first one suggesting that DLL3 overexpression is associated with lower disease-free survival and overall survival. This contrasts with results from SCLC, where DLL3 expression has no impact on prognosis [14].

Our results may have relevance for MTC treatment. DLL3 is the target of several therapeutic agents that are currently under research in early development clinical trials in SCLC [44]. These agents include antibody–drug conjugates, T-cell engager molecules, and chimeric antigen receptor therapies. The first antibody–drug conjugate available was rovalpituzumab tenserine, a monoclonal anti-DLL3 antibody linked to a DNA intercalating agent (pyrrolobenzodiazepine) that exerts the cytotoxic action [15]. This conjugate was used in a phase I/II trial in 13 patients with MTC. In this trial, all MTC patients were recruited in stage IV; 6 (46%) expressing high DLL3 and 7 (54%) expressing low DLL3. However, trials with this drug were discontinued [44]. Among T-cell engager molecules, the most clinically advanced molecule is tarlatamab, a compound that bispecifically binds both DLL3 and CD3 leading to T-cell-mediated tumor lysis. It has already completed a phase 1 trial, and it is currently running a phase 2 study in SCLC [44, 45]. We hypothesize that patients with high-grade MTC, who exhibit a tendency towards high expression of DLL3 and an expression profile similar to SCLC-A, may benefit from inclusion in clinical trials using compounds that target DLL3.

Our results are limited by using a restricted gene panel designed to evaluate the tumor biology, microenvironment, and immune response. However, this approach has proven useful in demonstrating that morphological differences corresponding to tumor grade are associated with distinct expression profiles, which may be relevant for treatment strategies. In conclusion, we have identified a set of genes (EGLN3, EXO1, UBE2T, UBE2C, FOXM1, CENPA, DLL3, CCNA2, SOX2, KIF23, and CDCA5) differently expressed between high-grade and low-grade MTC. The main pathways altered were related to cell cycle progression. In addition, we have found that high-grade MTC upregulates ASCL1, DLL3, and SOX2, a gene signature akin to SCLC molecular subgroup A. Taking into consideration our results about the impact of DLL3 expression on prognosis, both desmoplasia and DLL3 overexpression should be explored as predictors of aggressive disease and poor outcomes.