Clival chordoma research
Home › Understanding the tumour

Chordoma metabolism and dedifferentiation

Agent report for Kya, 10 September 2026, for a family question. Working frame: probable clival chordoma in a 34-year-old woman, 51 kg (down from 58 to 60), histology pending, dedifferentiated or poorly differentiated chordoma among the possibilities. Every PMID was verified through NCBI E-utilities on 10 September 2026; figures come from abstracts unless stated. No survival figures, by rule. Clinical reasoning is labelled.

Evidence levels. L1 randomised trial, meta-analysis or guideline. L2 prospective non-randomised trial or large cohort. L3 retrospective series. L4 case reports or very small series. L5 laboratory, animal or cell-line work, and expert reasoning.

Tier A. Basics, in plain English

What a chordoma cell runs on. A conventional chordoma cell is a notochord cell that never switched off its embryonic programme. It divides slowly: in 19 skull base chordomas the Ki-67 labelling index (the share of cells preparing to divide) was mostly under 6 percent, and tumours above 6 percent or with visible mitoses grew measurably faster on serial MRI (Holton 2000, Neuropathol Appl Neurobiol, PMID 11123715, L3). Mean Ki-67 was 5.6 percent at first surgery and 10.2 percent at recurrence (Sakai 2007, Brain Tumor Pathol, PMID 18095132, L3). Untreated skull base chordomas grew a mean 8 percent of their volume per month with wide scatter (Wang 2020, World Neurosurg, PMID 31926359, L3). It uses glucose modestly, not greedily: mean SUVmax 5.8 on FDG-PET in 23 proven chordomas, with patchy uptake (Olson 2021, Skeletal Radiol, PMID 33521875, L3); in seven residual skull base chordomas only one of eight sites lit up on FDG while seven of nine lit up on the hypoxia tracer FMISO, so these tumours are short of oxygen without being strongly sugar-hungry (Mammar 2012, Int J Radiat Oncol Biol Phys, PMID 22391104, L4). So a conventional chordoma is a low-flame tumour: slow, hypoxic, fed mainly by ordinary blood supply.

What "dedifferentiated" and "poorly differentiated" mean. Two different things.

What "sarcoma-like behaviour" means concretely. Faster division, necrosis from outgrowing the blood supply, and spread through the bloodstream to lungs, bone and liver rather than only creeping locally. In the largest dedifferentiated series, 36 percent had metastases at presentation versus 3 percent for conventional, and 57 percent versus 5 percent developed distant metastases after treatment, with local control after radiotherapy lost much sooner (Nachwalter 2019, J Neurooncol, PMID 31338785, L3). Treatment shifts towards sarcoma chemotherapy (anthracycline-based) alongside surgery and radiotherapy; in a pooled analysis of 130 cases, surgery was the one factor that held in multivariable analysis (Liu 2021, Front Oncol, PMID 34490087, L3).

Does a chordoma "break out" of slow growth? Sometimes, and it is documented three ways: sarcoma present from the start (de novo), sarcoma appearing at a recurrence, or sarcoma appearing after radiotherapy (Hanna 2008, J Bone Joint Surg Br, PMID 18450635, L4; Ikeda 1997, Radiat Med, PMID 9192436, L4). A pooled analysis of 130 cases found 74 de novo and 39 secondary, with no difference between the two in age, site, size, metastasis at diagnosis or relapse (Vuong 2023, World Neurosurg, PMID 36804481, L3). It is a change in the cell's genes, not in the patient's diet: a matched pair of conventional primary and dedifferentiated recurrence from one patient went from 211 to 430 mutations, gaining TP53 and BRCA1 mutations and losing the chromosome region carrying brachyury (Baluszek 2025, Acta Neuropathol Commun, PMID 41420192, L4).

Fuel versus signals: does eating feed the tumour? No, not in any way that food choices can control. Cancer cells grow because oncogenic signals (here brachyury, PI3K/AKT/mTOR, loss of CDKN2A, and in transformed tumours TP53) tell them to divide regardless of need; nutrient uptake follows those signals, which is why proliferating cells switch to glycolysis even with plenty of oxygen, the Warburg effect (Vander Heiden 2009, Science, PMID 19460998, L5 review; Hanahan 2022, Cancer Discov, PMID 35022204, L5 review). Blood glucose is held in a narrow range by the liver and pancreas whatever a person eats; eating less does not starve the tumour, it starves the muscles. Cachexia works the other way round: tumour-driven inflammation and catabolism break down host muscle and fat, and adequate nutrition is the mainstay of managing it (Baracos 2018, Nat Rev Dis Primers, PMID 29345251, L5 review). The ESPEN cancer nutrition guidelines recommend against diets that restrict energy in patients who are malnourished or at risk, and find insufficient evidence for ketogenic diets in cancer (Arends 2017, Clin Nutr, PMID 27637832; Muscaritoli 2021, Clin Nutr, PMID 33946039, L1 guideline). At 51 kg and losing weight, the priority is protein and calories, not restriction (clinical reasoning from L1 guidance).

"Sugar feeds cancer" examined. True at the cell level (most cancers take up glucose), false as advice: dietary sugar does not raise blood glucose in a way a tumour can exploit, and no trial shows that cutting sugar slows an established tumour. The one place glucose and insulin demonstrably matter is drug-induced: PI3K inhibitors raise blood glucose and insulin, which blunts the drug, and a ketogenic diet restored the drug's effect in mice (Hopkins 2018, Nature, PMID 30051890, L5). That is a signal (insulin) rather than a fuel, and it applies to a drug class the patient is not on.

Tier B. Current thinking, the papers

Proliferation and prognosis. Ki-67 above about 5 to 6 percent marks faster growth and earlier progression in skull base series (Holton 2000, PMID 11123715; Zhang 2019, World Neurosurg, PMID 31125781, L3). No study reports Ki-67 ranges for dedifferentiated versus conventional chordoma; the sarcoma component is by definition mitotically active.

Glycolysis and PET. Conventional chordoma is a low-to-moderate FDG user, less avid than pituitary adenoma among sellar masses (Olson 2021, PMID 33521875; Tosaka 2017, World Neurosurg, PMID 28427982, L3). Chordoma cells run the standard glycolytic enzymes GLUT1, hexokinase 2 and LDHA, and glycolysis and growth fell together when an oncogenic non-coding RNA was knocked down (Wang 2020, Eur Rev Med Pharmacol Sci, PMID 32096180, L5). No study compares SUV in dedifferentiated against conventional chordoma. The only chordoma data point: after carbon-ion therapy of 18 sacral chordomas, FDG-PET detected the rising uptake in the one tumour that later proved dedifferentiated (Evangelisti 2019, Eur Rev Med Pharmacol Sci, PMID 31115029, L3). By analogy from cartilage tumours, an SUVmax cutoff of 13.3 was 89 percent sensitive and 100 percent specific for dedifferentiated chondrosarcoma in 73 biopsied tumours (Gundavda 2022, Nucl Med Commun, PMID 34678831, L3).

Hypoxia and HIF-1alpha. HIF-1alpha, VEGF and MMP-2 were expressed in most of 35 sacral chordomas and tracked with vessel density (Li 2012, Oncol Lett, PMID 22783431, L3); in 32 spinal chordomas high HIF-1alpha predicted earlier recurrence or metastasis, hazard ratio near 3 (He 2024, Spine, PMID 38251727, L3, contested in correspondence). FMISO-PET confirms hypoxic tissue in residual skull base tumours (Mammar 2012, PMID 22391104, L4), and a susceptibility-mapping MRI measure correlated with Ki-67 in seven skull base chordomas (Fenech 2026, Magn Reson Med, PMID 41286542, L4). Hypoxia drives radioresistance and invasion.

Brachyury's programmes. Brachyury sits at the top of a network of about 160 target genes controlling cell cycle and extracellular matrix in chordoma cells; silencing it arrests growth (Nelson 2012, J Pathol, PMID 22847733, L5). Brachyury copy number gain (12 of 27 tumours) tracks with higher expression, which tracks with PI3K/AKT pathway activity and with earlier progression; blocking PI3K lowered brachyury and stopped growth in U-CH2 cells (Otani 2018, J Neurosurg, PMID 28753115, L3 and L5). A direct brachyury-to-glycolysis link in chordoma was searched for and not found; the link runs through PI3K/AKT/mTOR, the pathway that couples growth signals to nutrient uptake.

PI3K/AKT/mTOR. In a tissue array of 50 chordomas, phosphorylated AKT was present in 92 percent, phosphorylated 4E-BP1 in 96 percent, phosphorylated mTOR in 27 percent and phosphorylated p70S6K in 62 percent; PTEN was lost in 16 percent (Presneau 2009, Br J Cancer, PMID 19401700, L3). Sacral chordoma lines show hyperactive AKT and mTORC1 with PTEN loss, and rapamycin suppressed their growth (Han 2009, Clin Cancer Res, PMID 19276265, L5); sequencing of 104 chordomas found PI3K pathway mutations in 16 percent (Tarpey 2017, Nat Commun, PMID 29026114, L3).

IDH. Chordoma does not carry IDH1/2 mutations; intracranial chondrosarcoma frequently does, so the mutation is a diagnostic clue for the main lookalike (Arai 2012, Brain Tumor Pathol, PMID 22323113, L3). Separately, wild-type IDH1 is highly expressed in chordoma and inhibiting it collapsed the cells' antioxidant (glutathione and NADPH) supply and shrank two xenografts (Pun 2026, Acta Neuropathol, PMID 42493667, L5).

Lipids and other fuels. A 2026 preprint links recurrence and radioresistance to fatty-acid synthesis (FASN) and lipid droplets, and FASN blockade restored radiosensitivity in cells and xenografts (Wei 2026, bioRxiv, PMID 42182510, L5, preprint, not peer reviewed). Proteomics of 102 chordomas found three subtypes, one marked by oxidative phosphorylation (Yin 2024, Cell Rep Med, PMID 39368483, L3). No glutamine-dependence study in chordoma was found.

SMARCB1 loss and EZH2. SWI/SNF (which contains SMARCB1) and Polycomb (which contains EZH2) are opposing chromatin regulators; losing SMARCB1 leaves EZH2 unopposed, locking cells in a growth state (Wilson 2010, Cancer Cell, PMID 20951942, L5). In poorly differentiated chordoma SMARCB1 loss drives progression, with CDKN2A also lost in the cell lines (Walhart 2023, Am J Pathol, PMID 36657718, L5); the entity behaves far worse than conventional chordoma (Hasselblatt 2016, Acta Neuropathol, PMID 27067307, L3).

What transformation acquires. TP53 is the signature: mutated in 3 of 4 sequenced dedifferentiated components (Hung 2020, PMID 32427623) and in 9 of 23 dedifferentiated or poorly differentiated tumours versus 5 of 445 conventional (Baluszek 2025, PMID 41420192, L3). CDKN2A loss is the commonest alteration acquired during clinical progression of chordoma generally, with p16 protein lost in 53 percent of 384 samples (Cottone 2020, J Pathol Clin Res, PMID 31916407, L3), and homozygous 9p21 (CDKN2A) and 1p36 deletions predict early progression in a prospective clival series of 105 (Zenonos 2019, J Neurosurg, PMID 29905508, L2). PBRM1 (11 to 12.5 percent) and EGFR (9 percent) mutations appear in whole-genome and panel series (Bai 2021, Nat Commun, PMID 33536423; Chan 2025, PMID 40802536, L3), and in dedifferentiated tumours CDKN2A/B and EGFR were the commonest hits (2 of 7 each; Chan 2025).

Tier C. Outside the box: every idea, its evidence, and the gap

Rank order is by human evidence, best first. None is standard of care before histology; each needs the specific marker it targets.

  1. Regional hyperthermia with chemotherapy (sarcoma, L1). In 329 patients with high-risk soft tissue sarcoma, adding regional hyperthermia to neoadjuvant doxorubicin, ifosfamide and etoposide improved local progression-free survival (HR 0.65) and survival (HR 0.73) (Issels 2018, JAMA Oncol, PMID 29450452). Gap: limb and trunk sarcomas; no skull base or chordoma data; heating the clivus next to the brainstem and carotids is not established. Relevant only if a dedifferentiated component is confirmed and a sarcoma protocol is chosen.

  2. mTOR inhibitors: imatinib plus sirolimus or everolimus (L2). First reported as responses to imatinib plus sirolimus (Stacchiotti 2009, Ann Oncol, PMID 19570961, L4). The phase 2 of imatinib plus everolimus in 43 progressing patients gave 9 partial responses by Choi criteria but 1 by RECIST, toxicity stopped treatment in 30 percent, and response tracked with how many tumour cells had active mTOR effectors (Stacchiotti 2018, Cancer, PMID 30216418, L2). A pooled analysis of 12 phase 2 trials (328 patients, combined response rate 4.7 percent) found it the only regimen with significantly longer progression-free survival than the rest, HR 0.50 (Cote 2026, Eur J Cancer, PMID 42302352, L2). Gap: conventional chordoma, advanced disease, no randomised control; the activity is stabilisation, not shrinkage.

  3. Tazemetostat (EZH2 inhibitor) for INI1-deficient tumours (L2). Licensed for INI1-negative epithelioid sarcoma after a phase 2 basket (Gounder 2020, Lancet Oncol, PMID 33035459, L2). One metastatic poorly differentiated chordoma had a durable response after tazemetostat then radiotherapy (Gounder 2019, Clin Cancer Res, PMID 30642912, L4). Gap: applies only if the INI1 stain is negative, which is unusual at 34; that stain is routine and is the question to ask.

    [Correction 7 Oct 2026: tazemetostat was withdrawn from all markets in March 2026 (Ipsen, after SYMPHONY-1 showed second blood cancers in 5.7% vs 0); trials and compassionate access stopped. PMID 42599162; research 12 A3. Read anything above or below about it as history, not an option.]

  4. Immune checkpoint inhibitors (L2). In the AcSé basket, 34 of 97 ultra-rare sarcoma patients had chordoma; response rate at week 12 across the cohort was 6.2 percent (Blay 2023, Lancet Oncol, PMID 37429302, L2). An INI1-deficient paediatric chordoma responded to nivolumab, with T cells recognising a brachyury peptide (Williamson 2021, NPJ Precis Oncol, PMID 34931022, L4). PD-L1 is scarce on chordoma cells (scores under 1 in 93 percent; Chan 2025) and sits mainly on infiltrating immune cells (Mathios 2015, J Neurooncol, PMID 25349132, L5); MD Anderson enrols by PD-L1 or mutational burden (Banu 2024, Neurosurg Focus, PMID 38691867, L5). Gap: low response rate unselected; the selecting markers are not settled.

  5. CDK4/6 inhibition for CDKN2A loss (L2). Palbociclib in 28 patients with p16 or CDKN2A loss: disease control in 39 percent (11 stable), no objective responses, well tolerated, and the immunohistochemical responder signature did not predict benefit (Teleanu 2025, ESMO Open, PMID 40627883, L2). Xenograft support in Passeri 2022 (Front Oncol, PMID 36505864, L5). Gap: stabilisation only; combinations proposed.

  6. Brachyury vaccines (L1 and L2). GI-6301 (yeast-brachyury) plus radiotherapy versus placebo plus radiotherapy in 24 patients: one partial response in each arm, HR 1.02, stopped early for futility (DeMaria 2021, Oncologist, PMID 33594772, L1). BN-Brachyury plus radiotherapy in 29 patients: 2 durable partial responses, 81 percent stable, target not met (Cote 2024, Cancer, PMID 38985843, L2). Gap: safe and immunogenic, not yet shown to shrink tumours.

  7. EGFR inhibitors, afatinib (L5 to L2). Afatinib was the only EGFR inhibitor active across a chordoma cell-line panel and also degrades brachyury (Magnaghi 2018, Mol Cancer Ther, PMID 29237806, L5). The European phase 2 of afatinib in chordoma (NCT03083678) is not found as a published paper in PubMed at this search; treat any quoted results as unverified.

  8. Metformin (L1 elsewhere, L5 here). No chordoma study found. In sarcoma cell lines metformin alone barely affected viability but sensitised cells to glycolysis inhibition (Issaq 2014, Cell Cycle, PMID 24553119, L5). The definitive test elsewhere was negative: 3,649 breast cancer patients randomised to metformin or placebo, HR 1.01 for invasive disease-free survival (Goodwin 2022, JAMA, PMID 35608580, L1). Gap: no reason to expect chordoma to differ.

  9. Glycolysis inhibitors: 2-deoxyglucose and dichloroacetate (L2 safety, L5 efficacy). 2-DG reduced viability in some sarcoma lines depending on their glycolytic dependence (Issaq 2014, PMID 24553119, L5). In a phase 1 with docetaxel (34 patients) it was tolerable at 63 mg/kg, with hyperglycaemia in all and QTc prolongation in 22 percent; 1 partial response (Raez 2013, Cancer Chemother Pharmacol, PMID 23228990, L2). Dichloroacetate has a phase 1 in recurrent brain tumours (Dunbar 2014, Invest New Drugs, PMID 24297161, L2, abstract not read here) and animal sarcoma work only. Gap: no chordoma data; no efficacy signal in humans.

  10. Ketogenic and fasting-mimicking diets as adjuncts (L1, mixed). ERGO: ketogenic diet alone in 20 recurrent glioblastoma patients was feasible but "probably has no significant clinical activity" (Rieger 2014, Int J Oncol, PMID 24728273, L2). ERGO2: 50 patients randomised to calorie-restricted ketogenic diet plus fasting with reirradiation or usual diet; no difference in the 6-month progression-free rate, mean weight loss 2.1 kg (Voss 2020, Int J Radiat Oncol Biol Phys, PMID 32619561, L1). DIRECT: 131 breast cancer patients (BMI over 18 required) randomised to a fasting-mimicking diet around neoadjuvant chemotherapy; more radiological responses (OR 3.2), no toxicity difference (de Groot 2020, Nat Commun, PMID 32576828, L1). A ketogenic RCT in 80 breast cancer patients on chemotherapy reported smaller tumours and downstaging in locally advanced disease but no response difference in metastatic disease (Khodabakhshi 2021, Clin Nutr, PMID 32703721, L1, criticised in correspondence). Meta-analyses: ketogenic diets lower glucose but not IGF-1 or quality of life across 8 RCTs (Salido-Bueno 2024, Nutr Bull, PMID 38923748, L1) and reduce body weight and fat mass across 10 controlled trials (Zhao 2022, Nutrients, PMID 36235844, L1). A review of peri-chemotherapy fasting found weight loss in every study, all in well-nourished patients, and called claims of benefit premature (Bozzetti 2022, Clin Nutr, PMID 35552051, L5 review). Gap and harm: every positive trial excluded underweight patients; chordoma is not chemotherapy-treated in the conventional form; at 51 kg and falling, the expected effect is more weight and muscle loss with no demonstrated tumour effect. Not recommended on current evidence (clinical reasoning from L1 data and ESPEN guidance).

  11. IDH1 inhibition and FASN inhibition (L5). Two 2026 laboratory leads (Pun 2026, PMID 42493667; Wei 2026 preprint, PMID 42182510). Gap: no human data; the FASN paper is a preprint.

  12. PET-guided biopsy of the hottest region (L3 by analogy). In breast cancer the core biopsy site missed the highest-uptake area in 14 percent of tumours, more often over 3 cm (Koolen 2013, Eur J Radiol, PMID 23998705, L3). Two spinal chordomas read as conventional on needle biopsy proved dedifferentiated at resection (Kato 2016, Eur Spine J, PMID 27052403, L4). Gap: no chordoma PET-biopsy study. The practical ask: target any region that looks different on MRI (low T2, brisk enhancement, restricted diffusion) or is hottest on PET, and record where the needle went (clinical reasoning).

What this means for the week

Sources: PMIDs listed inline; the biopsy and subtype background is in research thread 16.

Research thread 21 · source file em-research/21-chordoma-metabolism-dedifferentiation.md · last edited 07 October 2026