Genomic and transcriptomic hallmarks of poorly differentiated and anaplastic thyroid cancers

Authors

Iñigo Landa

Tihana Ibrahimpasic

Laura Boucai

Rileen Sinha

Jeffrey A. Knauf

Ronak H. Shah

Snjezana Dogan

Julio C. Ricarte-Filho

Gnana P. Krishnamoorthy

Bin Xu

Nikolaus Schultz

Michael F. Berger

Chris Sander

Barry S. Taylor

Ronald Ghossein

Ian Ganly

James A. Fagin

Doi

10.1172/JCI85271

PMID: 26878173 · DOI: 10.1172/JCI85271 · Journal: Journal of Clinical Investigation (2016)

TL;DR

Landa, Ibrahimpasic, and colleagues at MSKCC performed targeted deep sequencing (MSK-IMPACT, 341 cancer genes) on 117 patient-derived advanced thyroid tumors — 84 poorly differentiated (PDTC) and 33 anaplastic thyroid cancers (ATC) — and transcriptomic profiling of 37 of them, then compared the results against the TCGA papillary thyroid carcinoma cohort (PMID:25417114). ATCs had a higher mutation burden than PDTCs and a markedly higher prevalence of TP53 (73% vs 8%), TERT promoter, PI3K/AKT/mTOR, SWI/SNF, and histone-methyltransferase mutations. BRAF and RAS were the predominant drivers and split PDTC by histopathologic definition (Turin → RAS-like; MSKCC criteria → BRAF-like) and by metastatic tropism (BRAF → nodal; RAS → distant). EIF1AX mutations were enriched in advanced disease (~10%) and showed extreme co-occurrence with RAS (odds ratio 58.3, P < 0.001). The TCGA-derived BRAF-RAS score (BRS) tracked with driver mutation in PDTCs but ATCs were uniformly BRAF-like irrespective of driver, consistent with a stepwise model in which advanced thyroid cancers arise from differentiated precursors through accumulation of additional genetic hits.

Cohort & data

  • Cohort size: 117 patient-derived advanced thyroid tumors — 84 PDTC and 33 ATC. Median age 58 (PDTC) and 66 (ATC); female:male 1.5:1 (PDTC) and 1.2:1 (ATC). 92/117 primary, 25/117 nodal or distant metastases. Paired normal tissue available for 106/117 (78 PDTC, 28 ATC).
  • Reference comparator: TCGA papillary thyroid carcinoma cohort (n = 390/401 depending on analysis) from the thca_tcga_pub study (PMID:25417114).
  • Sequencing assay: MSK-IMPACT targeted-capture NGS on the 341-gene IMPACT341 panel; mean depth 584× tumor / 236× normal (739× for ATCs, where deep coverage was needed because of low purity).
  • Tumor purity: median 72% (PDTC), 42% (ATC) — ATCs are extensively infiltrated by tumor-associated macrophages.
  • Transcriptomics & CNAs: 37 fresh-frozen tumors (17 PDTC + 20 ATC) were profiled on the Affymetrix U133 Plus 2.0 array and on the Agilent SurePrint G3 1×1M array-CGH platform; expression data deposited under GEO accession GSE76039.
  • Sample preservation: 80 FFPE / 37 frozen overall.
  • Dataset: thyroid_mskcc_2016 (studyId in cBioPortal).
  • PDTC definitions used in stratification: Turin proposal (architectural high-grade features + solid/nested/insular growth + absent PTC nuclear features + mitosis ≥3/10 HPF or necrosis) vs. MSKCC criteria (≥5 mitoses/10 HPF and/or necrosis with follicular-cell differentiation, irrespective of growth pattern).

Key findings

  • Mutation burden (median ± IQR mutations across the 341-gene panel): ATC 6 ± 5, PDTC 2 ± 3, PTC 1 ± 1 (each pairwise P < 1×10⁻⁴). Within PDTC, higher burden tracked with older age (47 vs 58 vs 64 years, P < 1×10⁻³), tumor size > 4 cm (36% vs 43% vs 71%, P = 0.04), distant metastasis (8% vs 29% vs 57%, P = 2×10⁻³), and worse overall survival (logrank P = 0.01).
  • BRAF V600E in 33% of PDTC and 45% of ATC; NRAS/HRAS/KRAS in 28% (PDTC) and 24% (ATC), mutually exclusive with BRAF and gene fusions.
  • PDTC histologic definition splits by driver: 92% of RAS-mutant PDTCs met the Turin definition; 81% of BRAF-mutant PDTCs were classified as PDTC only by MSKCC criteria (mitosis + necrosis irrespective of growth pattern). BRAF-mutant PDTCs were smaller, more often nodal-metastatic, and overrepresented in females (P = 0.005); RAS-mutant PDTCs were larger and tended toward distant metastasis.
  • TERT promoter mutations show stepwise enrichment along disease progression: 9% PTC → 40% PDTC → 73% ATC. Of 117 advanced tumors, 49 carried C228T (c.-124G>A) and 8 carried C250T (c.-146G>A). TERT promoter mutations were subclonal in PTC but clonal in PDTC/ATC. Significant co-occurrence with BRAF/RAS in advanced disease (OR 3.4, P = 0.004), consistent with the GABPA-binding-element mechanism of TERT promoter activation requiring MAPK-activated ETS factors.
  • TERT and outcome in ATC: TERT-mutant ATC patients had markedly shorter survival (median 147 vs 732 days, P = 0.03), most pronounced in cancers also carrying BRAF or RAS mutations. TERT-mutant PDTCs developed more distant metastases (56% vs 20%, P = 0.01).
  • EIF1AX mutated in 11% of PDTC and 9% of ATC (vs 1% of PTC). 14/15 EIF1AX-mutant tumors also carried RAS mutations (OR 58.3; P < 0.001). Mutations clustered in N-terminal residues (also seen in uveal melanoma) or at a thyroid-specific p.A113splice site between exons 5 and 6 producing a 12-aa in-frame deletion via cryptic splice acceptor usage. EIF1AX mutations associated with larger tumors and shorter PDTC survival (logrank P = 0.048).
  • TP53 mutated in 73% of ATC vs 8% of PDTC (P < 1×10⁻⁴) — a key dichotomy distinguishing the two entities.
  • PI3K/AKT/mTOR pathway mutations (PIK3CA, PTEN, PIK3C2G, PIK3CG, PIK3C3, PIK3R1, PIK3R2, AKT3, TSC1, TSC2, MTOR) in 39% ATC vs 11% PDTC (P = 1×10⁻³). PIK3CA (18% ATC) and PTEN (15% ATC) were notably enriched; all 5 PIK3CA helical-domain (E542K/E545K) mutations were in ATC, while the lone kinase-domain H1047R was in PDTC. All 3 NF1-mutant ATCs also had truncating PTEN alterations (P = 2×10⁻³).
  • SWI/SNF complex (ARID1A, ARID1B, ARID2, ARID5B, SMARCB1, PBRM1, ATRX) mutated in 36% ATC vs 6% PDTC (P = 1×10⁻⁴) — first report of SWI/SNF disruption in advanced thyroid cancer; mutations were largely mutually exclusive within the complex.
  • Histone methyltransferases (KMT2A, KMT2C, KMT2D, SETD2) mutated in 24% ATC vs 7% PDTC (P = 0.02). Other epigenetic-regulator alterations seen in CREBBP, EP300, BCOR, BCL6.
  • DNA mismatch-repair (MSH2, MSH6, MLH1) mutations in 12% ATC vs 2% PDTC; MMR-mutant tumors showed a hypermutator phenotype (median 16.5 vs 5 mutations in MMR-mutant vs WT ATC, P = 1×10⁻³).
  • ATM mutated in 7% PDTC and 9% ATC, associated with higher mutation burden in both (PDTC P = 0.04, ATC P = 7×10⁻³).
  • No support for prior reports of frequent WNT-pathway mutations in ATC: only 1 missense CTNNB1 p.L347P (not the previously reported exon-3 hotspot), 2 ungermlined-paired AXIN1 variants, and 1 truncating APC p.Q1529X. The cohort also failed to support roles for apoptosis, Hedgehog, homologous recombination, immune-response, JAK-STAT, polycomb, ubiquitination, or TGFβ pathway mutations in PDTC/ATC.
  • Gene fusions present in 14% of PDTC, absent in ATC, and mutually exclusive with BRAF/RAS/TSHR/STK11. Detected: 5 RET/PTC (RET with CCDC6 or NCOA4), 3 PAX8–PPARG, 3 ALK fusions (STRN, EML4, and a novel CCDC149 partner). PDTCs harboring fusions were younger (49 vs 58 years, P = 0.04).
  • Novel ATC fusion: a single ATC carried a t(15;19) NUTM1–BRD4 in-frame fusion (NUT exons 1–2, BRD4 exons 14–20) — patient was a 34-year-old woman alive 10 years post-diagnosis (clinical outlier).
  • Recurrent arm-level CNAs more frequent in advanced disease: losses of 1p, 8p, 13q, 15q, 17p, 22q; gains of 1q and 20q. ATCs had higher prevalence of 8p loss, 17p loss, and 20q gain (each P < 2×10⁻⁴ vs PDTC). 22q loss was strongly enriched in RAS-mutant PDTC vs BRAF-mutant PDTC (P = 1×10⁻³). 1p, 13q, 15q losses enriched in PDTCs without known driver mutations. Outcome correlates: PDTC 1q gain → worse survival (logrank P = 0.06); ATC 13q loss → P = 0.02; ATC 20q gain → P = 0.06.
  • BRAF-RAS score (BRS) computed from 67 of the 71 TCGA BRS genes: all 13 BRAF-V600E PDTCs and ATCs were BRAF-like; RAS-mutant PDTCs were RAS-like, but RAS-mutant ATCs were paradoxically BRAF-like (Mann-Whitney P = 3×10⁻³), indicating that high MAPK transcriptional output is a property of ATCs irrespective of driver.
  • Thyroid differentiation score (TDS): PDTCs and PTCs had similar TDS, whereas ATCs had profoundly suppressed TDS for TG, TSHR, TPO, PAX8, SLC26A4, DIO1, DUOX2; no thyroid expressed THRB, DUOX1, SLC5A5 (NIS), or SLC5A8. TDS-BRS correlation preserved in PDTC (r = 0.72, P < 0.01) but lost in ATC (r = -0.43, P = 0.06).
  • M2 macrophage signature (78-gene set) cleanly separated ATCs from PDTCs by unsupervised clustering, confirming the histologic observation of TAM infiltration in ATC.
  • Comparison with whole-exome ATC study (Kunstman et al., PMID 25576899): IMPACT detected mutations at higher frequency than WES (TP53 73% vs 27%; BRAF 45% vs 27%; PIK3CA 18% vs 9%; PTEN 15% vs 0%), attributed to the deeper coverage (739× vs 264×). IMPACT additionally detected SWI/SNF and HMT mutations missed by WES; IMPACT did not include RASAL1, USH2A, HECTD1, MLH3, MSH5.

Genes & alterations

  • BRAF — V600E in 33% PDTC and 45% ATC. PDTC-MSK histology strongly enriched for BRAF; BRAF-mutant PDTCs are smaller and more often nodal-metastatic.
  • NRAS, HRAS, KRAS — collectively in 28% PDTC and 24% ATC; mutually exclusive with BRAF and gene fusions; PDTC-Turin histology strongly enriched for RAS; RAS-mutant PDTCs trend to distant metastasis.
  • TERT — promoter C228T or C250T in 40% PDTC and 73% ATC; clonal in advanced disease vs subclonal in PTC; co-occurs with BRAF/RAS (OR 3.4, P = 0.004); markedly worse survival in TERT-mutant ATC.
  • TP53 — mutated in 73% ATC vs 8% PDTC (P < 1×10⁻⁴); a defining event of the PDTC→ATC transition.
  • EIF1AX — N-terminal hotspot or thyroid-specific p.A113splice mutation; 11% PDTC, 9% ATC; near-perfect co-occurrence with RAS (OR 58.3, P < 0.001); shorter survival in PDTC (logrank P = 0.048).
  • PIK3CA, PTEN, PIK3C2G, PIK3CG, PIK3C3, PIK3R1, PIK3R2, AKT3, TSC1, TSC2, MTOR — PI3K/AKT/mTOR pathway disrupted in 39% ATC vs 11% PDTC (P = 1×10⁻³). PIK3CA helical-domain mutations restricted to ATC.
  • NF1 — truncating mutations in 3 BRAF/RAS-WT ATCs; all co-occurred with PTEN truncation (P = 2×10⁻³).
  • ARID1A, ARID1B, ARID2, ARID5B, SMARCB1, PBRM1, ATRX — SWI/SNF subunits mutated in 36% ATC vs 6% PDTC (P = 1×10⁻⁴), generally mutually exclusive within the complex.
  • KMT2A, KMT2C, KMT2D, SETD2 — HMTs mutated in 24% ATC vs 7% PDTC (P = 0.02).
  • MSH2, MSH6, MLH1 — MMR mutations in 12% ATC vs 2% PDTC; associated with hypermutator phenotype.
  • ATM — 7% PDTC, 9% ATC; co-segregates with higher mutation burden.
  • RB1, NF2, MEN1 — infrequent truncating mutations.
  • CDKN1B, CDKN2C, CDKN2A, ERBB2, PTCH1, DAXX — single-case mutations replicating findings from the prior WES ATC study.
  • TSHR, STK11 — low-frequency mutations in both PDTC and ATC.
  • CTNNB1, AXIN1, APC — only one CTNNB1 p.L347P (non-hotspot), one APC truncation, and two unverified AXIN1 variants — the study explicitly fails to replicate prior reports of frequent WNT-pathway alterations in ATC.
  • Gene fusions: RET fused with CCDC6 or NCOA4 (5 PDTCs); PAX8–PPARG (3 PDTCs); ALK fused with STRN, EML4, or novel CCDC149 (3 PDTCs); NUTM1–BRD4 in 1 ATC. NTRK1/3 fusions could not be assessed because their introns were not covered by the IMPACT panel.
  • Other low-frequency hits (≥2 ATC or ≥3 PDTC): DIS3, FAT1, POLE, RBM10, RAD54L, RECQL4, SF3B1. RTKs other than RET include EPHA3 (3 ATC-only), EGFR, FLT1 (VEGFR1), FLT4 (VEGFR3), KDR (VEGFR2). All four NOTCH1–NOTCH4 family members were mutated. Other epigenetic regulators: CREBBP, EP300, BCOR, BCL6.

Clinical implications

  • Histology-driver-tropism axis in PDTC: Application of either Turin or MSKCC PDTC criteria predicts driver biology — Turin → RAS-like with distant-metastasis tropism; MSKCC → BRAF-like with locoregional-nodal tropism. The authors argue this is directly actionable for risk stratification and possibly for therapy selection (RAS- vs BRAF-targeted strategies).
  • Risk stratification by TERT and EIF1AX: TERT promoter status (and especially TERT + BRAF/RAS double-mutant ATCs) identifies an extreme-risk subgroup; EIF1AX mutation independently predicts shorter PDTC survival (logrank P = 0.048).
  • Therapeutic targeting rationales: The 39% prevalence of PI3K/AKT/mTOR pathway alterations in ATC supports the use of pathway inhibitors and is consistent with the prior single-case report of an everolimus responder (Wagle et al., NEJM 2014, referenced). BRAF-V600E in 45% of ATCs supports BRAF-inhibitor strategies (cf. Rosove et al., NEJM 2013). The novel SWI/SNF and HMT enrichments in ATC suggest exploration of chromatin-targeted agents.
  • Subclonal early-disease screening: The discovery in well-differentiated thyroid tumors of subclonal mutations of genes that this study shows are enriched in advanced disease (TERT, TP53, SWI/SNF, HMTs) should raise concern that these PTCs may be poised to progress; deep sequencing of indolent-appearing thyroid tumors may have prognostic value.
  • Diagnostic platform: The study argues — based on the 2- to 3-fold higher mutation-detection frequencies vs WES — that deep targeted sequencing (≥500×) is the assay of choice for ATC because of its low tumor purity from extensive macrophage infiltration.

Limitations & open questions

  • Targeted 341-gene panel does not interrogate the full exome; genes such as RASAL1, USH2A, HECTD1, MLH3, MSH5 (the latter two in MMR) are not covered. Authors explicitly acknowledge this drawback.
  • Intronic coverage by IMPACT excludes NTRK1 and NTRK3 fusions, leaving a known thyroid-fusion class undetected.
  • Low ATC tumor purity (median 42%) limits homozygous CNA detection; the analysis therefore restricts itself to arm-level events at conservative log-ratio thresholds.
  • For 11/117 tumors no paired normal was available, requiring use of pooled normals and manual review; for 23 tumors purity could not be confidently estimated.
  • The cohort failed to reproduce prior reports of high CTNNB1 mutation frequency in ATC; whether this is a true biological discrepancy or a function of prior over-calling on Sanger sequencing in low-purity samples remains unresolved.
  • The mechanistic basis for the EIF1AX–RAS co-occurrence (and its specificity to thyroid for the p.A113splice variant) is unknown; the authors flag this as the highest-priority follow-up.
  • Why ATCs are uniformly BRAF-like by BRS irrespective of driver mutation is not mechanistically resolved — proposed explanations include greater chromatin-modifier disruption, parallel-pathway activation, and TAM-driven signaling.
  • Survival outcomes for individual mutations in ATC are based on small numbers (33 ATCs), and several P values cluster near 0.05.
  • No drug treatments were administered or tracked as part of this study; all therapeutic implications are inferred.

Citations from this paper used in the wiki

  • “Compared to PDTCs, ATCs had a greater mutation burden, including a higher frequency of mutations in TP53, TERT promoter, PI3K/AKT/mTOR pathway effectors, SWI/SNF subunits, and histone methyltransferases.” (Abstract, p. 1052)
  • “Together, 40% of PDTCs and 73% of ATCs harbored TERT promoter mutations (49/117 C228T [c.-124G>A]; 8/117 C250T [c.-146G>A]) as compared with 9% of PTCs from TCGA.” (p. 1055)
  • “TERT mutations co-occurred with BRAF/RAS mutations in PDTCs and ATCs combined (P = 4 × 10⁻³).” (p. 1055)
  • “11% of PDTCs and 9% of ATCs harbored EIF1AX mutations…strongly associated with RAS (14/15, P < 1 × 10⁻⁴).” (p. 1055)
  • “TP53 mutations, although highly prevalent in ATCs, were relatively rare in PDTCs (73% vs. 8%, P < 1 × 10⁻⁴).” (p. 1055)
  • “Genes encoding components of the SWI/SNF chromatin remodeling complex were mutated in 36% of ATCs and 6% of PDTCs (P = 1 × 10⁻⁴). This is the first report of mutations in ARID1A, ARID1B, ARID2, ARID5B, SMARCB1, PBRM1, and ATRX genes in advanced thyroid tumors.” (p. 1055)
  • “Ninety-two percent of RAS mutations were found in PDTCs fulfilling the Turin definition…By contrast, 81% of BRAF mutations were found in PDTCs defined based only on MSKCC criteria.” (pp. 1053–1055)
  • “all 13 BRAFV600E-mutated PDTCs and ATCs were BRAF-like. However, although RAS-mutant PDTCs were strongly RAS-like, RAS-mutant ATCs were BRAF-like (P = 3 × 10⁻³).” (p. 1059)
  • “Survival of ATC patients harboring TERT promoter mutations was markedly diminished (732 vs. 147 days, P = 0.03).” (p. 1055)
  • “MSK-IMPACT…targeted sequencing of 341 cancer genes…was performed in all 117 tumors.” (p. 1053)
  • “Average depth of coverage was 584× for tumors and 236× for paired normal tissues…Coverage for ATCs was 739×.” (p. 1053)
  • “expression data…have been deposited in NCBI’s Gene Expression Omnibus (GEO)…GEO Series accession number GSE76039.” (p. 1063)

This page was processed by crosslinker on 2026-05-14.