Chordoma cells and dedifferentiation

Agent report for Kya, 7 September 2026, for her partner's questions. Every study claim carries a PMID verified through NCBI E-utilities on 7 September 2026; figures are from abstracts (web search was unavailable, full texts not read). Clinical reasoning is labelled as such. The patient's histology is pending, so this is about chordoma in general, not her tumour. No survival figures, by rule.

Citation corrections against the brief: the Hung 2020 dedifferentiated series is in Am J Surg Pathol, not Mod Pathol; the Duan 2022 single-cell paper is in Cancer Immunol Immunother, not Cancer Res; Bai 2023 is an RNA expression study, not a second whole-genome study; the Zenonos group has no dedicated dedifferentiation paper, their relevant work is the 1p36/9p21 prognostic panel.

1. What the cells are

Origin. Chordoma arises from remnants of the notochord, the embryonic midline rod that organises spine formation and normally disappears except within the intervertebral discs. Remnants persist in the clivus and sacrum, hence the two commonest sites. Yamaguchi's autopsy work suggests many chordomas grow out of a benign intraosseous precursor, the benign notochordal cell tumour (BNCT) (Yamaguchi 2004, Am J Surg Pathol, PMID 15166667; Yamaguchi 2005, Mod Pathol, PMID 15803192); two clival cases with a BNCT component next to chordoma exist (PMID 31518983).

Physaliphorous cells. Conventional chordoma is lobules of large cells with bubbly vacuolated cytoplasm ("physaliphorous", bubble-bearing) in a mucin-rich myxoid matrix, with smaller epithelioid cells in cords between them. The cells co-express epithelial markers (cytokeratins, EMA) and S100, an unusual mix that reflects the notochord's character (Meis 1987, PMID 2440324; Walcott 2012, Lancet Oncol, PMID 22300861).

Brachyury (TBXT), the lineage marker. Brachyury is the transcription factor that builds the notochord. It is expressed in the embryonic notochord and in all 53 chordomas tested, in chondroid and chordoid areas alike, and not in cartilage tumours (Vujovic 2006, J Pathol, PMID 16538613); it separates skull base chordoma from chondrosarcoma, the main lookalike in the patient's case (Oakley 2008, Mod Pathol, PMID 18820665). It is also a driver: silencing it arrests chordoma growth, and it sits at the top of a network of about 160 target genes (Nelson 2012, J Pathol, PMID 22847733). Genetics agree. Four chordoma families carried a duplication of a piece of 6q27 containing only the T gene (Yang 2009, Nat Genet, PMID 19801981); a common T variant, rs2305089, gives an allelic odds ratio near 6 for sporadic chordoma (Pillay 2012, Nat Genet, PMID 23064415); somatic T duplications occur in up to 27 percent of sporadic tumours and copy the germline architecture (Tarpey 2017, Nat Commun, PMID 29026114). Chordoma is a notochord programme that never switched off.

Conventional versus chondroid; WHO 2020. Chondroid chordoma has cartilage-like areas but is brachyury positive. Conventional and chondroid share the same driver losses (PTEN locus lost in 59.5 percent, CDKN2A in 66.7 percent of 42 tumours, no prognostic difference; Yang 2020, PMID 32547095), and the fifth-edition WHO classification (2020) folds chondroid into conventional chordoma NOS, alongside two other types: dedifferentiated and the new poorly differentiated chordoma defined by SMARCB1 loss (Anderson and Doyle 2021, Histopathology, PMID 33438273; PMID 35713256).

2. Genomics of conventional chordoma, and what it says about tempo

The Sanger sequencing of 104 sporadic chordomas found T duplication in up to 27 percent, PI3K pathway mutations (PIK3CA, PTEN and relatives) in 16 percent, LYST inactivated in 10 percent, and a low point mutation burden overall (Tarpey 2017, PMID 29026114). Whole-genome sequencing of 80 skull base chordomas found PBRM1 (a SWI/SNF chromatin remodelling gene) in 12.5 percent and homozygous CDKN2A/2B deletion as the commonest events, and "despite the low mutation rate, extensive somatic copy number alterations frequently occur, most of which are clonal and showed highly concordant profiles between paired primary and recurrence/metastasis samples" (Bai 2021, Nat Commun, PMID 33536423). RNA profiling splits chordoma into a chromatin-remodelling-loss subtype (PBRM1, SETD2) and an EMT/Sonic Hedgehog subtype (Bai 2023, Clin Cancer Res, PMID 36260525); a 2026 copy-number classification of 103 skull base tumours confirms a chromosome 9 (CDKN2A) loss cluster enriched for cell-cycle genes (PMID 42284679). Homozygous 9p21 (CDKN2A) deletion and 1p36 deletion independently predict early progression after surgery and after radiotherapy in a prospective clival series of 105 (Zenonos 2018, J Neurosurg, PMID 29905508), the basis of a three-tier risk grouping used to decide on radiotherapy (PMID 34745826).

Implication (clinical reasoning). A cancer with few mutations, driven by an over-active developmental switch plus loss of one or two brakes, divides slowly and changes slowly. Measured growth agrees: untreated skull base chordomas grew a mean 8 percent of volume per month with wide scatter (Wang 2020, PMID 31926359); a Japanese series split tumours at 3 mm a year into slow and rapid growers (Morimoto 2019, PMID 31240525). Slow does not mean harmless; it means the tumour can reach centimetres before it touches anything that hurts.

3. Dedifferentiated chordoma

Definition. Biphasic: conventional (or chondroid) chordoma next to a high-grade sarcoma, usually undifferentiated pleomorphic sarcoma or fibrosarcoma, sometimes osteosarcoma or rhabdomyosarcoma, with a sharp border. In the sarcoma part the cells have lost cytokeratin and brachyury; they no longer look like notochord (Meis 1987, Am J Surg Pathol, PMID 2440324; Hung 2020, Am J Surg Pathol, PMID 32427623).

Frequency. Under 5 percent (PMID 27052403); under 1 percent of more than 1,000 chordomas at MGH (Hung 2020); 2 to 8 percent in a 2025 sequencing paper (PMID 40802536).

Timing. Three routes: de novo, at recurrence, or after radiotherapy. Hanna's four de novo cases came with the note that it is "more commonly diagnosed in recurrences and following radiotherapy" (Hanna 2008, J Bone Joint Surg Br, PMID 18450635); Meis's three were all de novo and unirradiated (PMID 2440324). The pooled analysis of 130 cases (14 SEER, 116 literature) found 74 de novo, 39 secondary, 18 unknown, with no difference between de novo and secondary in age, site, size, metastasis at diagnosis, relapse or mortality (PMID 36804481). Radiotherapy is one route, not the main one.

Molecular events. TP53 is the signature. Hung found TP53 mutations in the dedifferentiated component in 3 of 4 sequenced tumours (PMID 32427623). A 2025 study with matched primary conventional and recurrent dedifferentiated tissue from one patient saw mutations rise from 211 to 430, new TP53 and BRCA1 mutations, and loss of 6q including TBXT (explaining the lost brachyury stain); across cases and literature TP53 was mutated in 9 of 23 dedifferentiated or poorly differentiated tumours versus 5 of 445 conventional (PMID 41420192). TP53 appears in both de novo and secondary cases (PMID 36804481). An MSKCC series also reported frequent SMARCB1 alteration (PMID 31338785), possibly overlap with poorly differentiated chordoma before the 2020 definitions.

Behaviour. The sarcoma component dictates outcome, and smaller dedifferentiated areas do better (Hanna 2008). Metastasis is the defining difference: 36 percent metastatic at presentation versus 3 percent for conventional, 57 percent versus 5 percent distant metastasis after treatment, and much shorter time to local failure after radiotherapy (Nachwalter 2019, J Neurooncol, PMID 31338785); 6 of 7 with follow-up metastasised in Hung's series. Every series describes outcomes as worse and shorter than conventional chordoma; in the 58-study pooled analysis surgery was the only factor that held in multivariable analysis, with resection type, chemotherapy and proportion of dedifferentiated tissue mattering in single-variable analyses (Liu 2021, Front Oncol, PMID 34490087). One caution against fatalism: in Meis's third case the osteosarcoma component never reappeared and all recurrences and lung metastases were conventional chordoma (PMID 2440324).

4. Poorly differentiated chordoma

A different tumour, often confused with dedifferentiated. Not biphasic: uniform sheets of epithelioid to rhabdoid cells with prominent nucleoli, mitoses and necrosis, which keep brachyury and cytokeratin but have lost SMARCB1 (INI1), a core SWI/SNF subunit (Mobley 2010, Acta Neuropathol, PMID 21057957; Hasselblatt 2016, Acta Neuropathol, PMID 27067307). Mobley's four cases (three clival) all lacked INI1, all expressed brachyury, three of four had 22q deletion at SMARCB1; ten typical chordomas kept INI1. The MGH series of 19: median age 11 (range 1 to 29), 53 percent skull base and clivus, 32 percent cervical spine, 16 percent sacrum, all brachyury and cytokeratin positive (Shih 2018, Mod Pathol, PMID 29483606). Adult extra-axial cases are exceptional (PMID 39577663). SMARCB1 loss appears to lock SWI/SNF in a growth rather than differentiation configuration, with CDKN2A also lost in the cell lines (Walhart 2023, PMID 36657718).

Tazemetostat. SMARCB1 loss creates dependence on EZH2, so the EZH2 inhibitor tazemetostat (licensed for INI1-negative epithelioid sarcoma, PMID 33035459) is the rational drug. A metastatic poorly differentiated chordoma had a durable response after tazemetostat then radiotherapy (Gounder 2019, Clin Cancer Res, PMID 30642912); children with INI1-deficient tumours including this one responded in phase 1 (PMID 29141919) and it was the Pediatric MATCH arm for SMARCB1 loss (PMID 37228094). It also worked in a PBRM1-mutant conventional chordoma xenograft (PMID 35326637). Relevant to the patient only if her tumour is INI1 negative, unusual at 34, which is exactly why the stain is routine.

[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.]

5. Is it a mixture? Heterogeneity and metabolism

Yes, at several levels. Single-cell RNA sequencing of six chordomas (33,737 cells): tumour cells 48 percent, fibroblasts 21 percent, macrophages 14 percent, T and NK cells 12 percent; the tumour cells formed six subclusters with epithelial-like matrix, stem-cell and immunosuppressive programmes, TGF-beta the main axis (Duan 2022, Cancer Immunol Immunother, PMID 35084549). A 90,691-cell skull base study found a stem-like, cathepsin-L-marked cluster that "tended to be distributed in the inferior part of the tumor" and is linked to radioresistance, plus a partial-EMT signature linked to invasion (Zhang 2022, Cell Discov, PMID 36127333). Spatial transcriptomics shows the stroma is active: inflammatory fibroblasts, and in recurrent tumours hypoxic, ER-stressed fibroblasts, track with invasion (PMID 39611243; PMID 37772937); immune content ranges from heavily infiltrated to "immune desert" (PMID 41586579). At the DNA level, though, copy number changes are clonal and concordant between primary and recurrence (Bai 2021, PMID 33536423): transcriptionally diverse, genomically one family, unless a dedifferentiated clone has emerged.

Metabolism and FDG-PET. Conventional chordoma is a low-to-moderate glucose user: mean SUVmax 5.8 (SD 3.7) in 23 biopsy-proven tumours, all with heterogeneous uptake (PMID 33521875); less FDG-avid than pituitary adenoma among sellar masses (PMID 28427982); in seven residual skull base chordomas only one of eight sites was visible on FDG but seven of nine on the hypoxia tracer FMISO, so hypoxic without being strongly glycolytic (PMID 22391104). No study compares SUV in dedifferentiated against conventional chordoma; that dataset does not exist. By analogy, dedifferentiated components of cartilage tumours are far more FDG-avid than their low-grade parents (SUVmax 16 in one dedifferentiated chondrosarcoma, PMID 27476566; FDG grading of chondroid tumours, PMID 31396797), and chordoma cell lines run GLUT1, hexokinase 2 and LDHA glycolytic programmes in the laboratory (PMID 32096180). A focal hot spot inside a cold chordoma would therefore raise concern for a high-grade component, but that is analogy, not chordoma data.

Why it matters for biopsy and surgery. Two spinal chordomas read as conventional on CT-guided trocar biopsy proved dedifferentiated at resection; the sarcoma was a distinct region with a sharp demarcation visible on T2 MRI (PMID 27052403). Cytology cannot reliably separate the two (PMID 40879100). Lessons: a biopsy samples one place, so the report should say where the needle went; any area that looks different on imaging (lower T2, brisker enhancement, restricted diffusion, a PET hot spot) is the place to sample or to be sure the resection includes; the radioresistant stem-like cells may sit in a predictable part of the tumour (PMID 36127333). Resection specimens, not biopsies, define subtype.

6. Why a clival chordoma can be missed on MRI

• Marrow hides it. The clivus converts with age from red marrow (low to intermediate T1) to fatty marrow (bright T1); in adults, low T1 signal in the clivus is the marker of disease (Neelakantan 2014, Clin Radiol, PMID 25168701). Chordoma is intermediate to low on T1, very bright on T2, and enhances heterogeneously (Erdem 2003, Radiographics, PMID 12853676; PMID 33521875). In a young woman whose clival marrow is still partly red, on a brain protocol, a small lesion can be indistinguishable from marrow. Three quarters of chordomas also hold small T1-bright foci of haemorrhage or mucin (17 of 22, PMID 33521875) that can pass for fat.
• Thin slices and contrast. Brain MRI uses 5 mm slices centred on the brain; the clivus is at the edge of the field. Dedicated skull base imaging means thin-slice T2, pre- and post-contrast fat-suppressed T1, and CT for bone (Santegoeds 2018, Curr Radiol Rep, PMID 29629241). Without contrast, a small chordoma and a benign remnant look the same.
• Ecchordosis physaliphora (EP) and BNCT. EP is a notochordal remnant on the dorsal clivus in about 2 percent of autopsies; of 41 patients referred with suspected clival chordoma, 8 (19.5 percent) had EP: circumscribed, T1 low, T2 high, no enhancement, often a T2-dark stalk (Park 2017, Neurosurg Rev, PMID 27206421). BNCT was found in 0.76 percent of 916 posterior fossa scans; the authors note such lesions "may be easily overlooked" and many "do not fit neatly" into either category (Golden 2014, PMID 23464492). A UK series proposes managing EP and BNCT together as benign notochordal cell tumour under surveillance (Sooltangos 2022, PMID 35832978); 58 benign-appearing clival lesions stayed stable over a median of nearly 3 years even with minor enhancement or erosion (PMID 38691870). Histologically EP and chordoma can be identical (PMID 34712372), so the distinction rests on enhancement, bone destruction, size and behaviour over time. The patient's lesion (3.8 cm, destructive, enhancing, encasing vessels) is not in this category; the relevance is to the November 2025 scan, where a small non-enhancing, non-destructive lesion could reasonably have been called a remnant or marrow. Whether that scan showed anything is a question for a skull base neuroradiologist with the images.
• Documented misreads both ways. A clival chondroid chordoma in a 15-year-old with three years of diplopia and headache was reported on CT as sinusitis and polyps (PMID 16333198); ectopic pituitary tumours, aneurysmal bone cysts and carotid dissections have been misdiagnosed as chordoma (PMIDs 40073404, 26605267, 26622598). Only tissue settles it.

Table: the three WHO 2020 chordoma types

Sources: PMIDs 33438273, 16538613, 29026114, 33536423, 2440324, 32427623, 41420192, 31338785, 21057957, 29483606, 27052403, 40802536.

7. Plain-language summary for her partner

A chordoma is made of cells left over from the notochord, the rod that organised the spine when the patient was an embryo. Those cells were supposed to disappear. A few stayed behind in the clivus, and at some point one of them kept a developmental switch called brachyury turned on and lost a brake or two. That is nearly the whole genetic story: chordoma has very few mutations compared with most cancers. That is why it grows slowly, why it can sit for years, and why ordinary chemotherapy, which works on fast-dividing cells, does little.

Slow is not the same as safe. A slow tumour in the clivus can reach several centimetres before it touches a nerve, and on a brain scan without thin slices and contrast a small one can look like normal bone marrow or like a harmless remnant that radiologists are taught to leave alone. So the 2025 scan may have shown something that was reasonably read as nothing. That is a question for a skull base radiologist with the actual images.

"Dedifferentiation" means part of the tumour has stopped being chordoma. Under the microscope there are two tumours side by side: ordinary chordoma, and a high-grade sarcoma that has lost the notochord markers and usually carries a new mutation in a gene called TP53. It happens in fewer than 5 in 100 chordomas. It can be there from the start, appear when a tumour comes back, or follow radiotherapy. When it is present, spread to lungs and bones becomes common rather than rare and everything moves faster. There is a separate, mostly childhood type called poorly differentiated chordoma, defined by loss of a protein called INI1, which has a targeted drug. Pathologists stain for brachyury and INI1 as routine, so both questions are answered from the tissue.

Is it a mixture? Yes. Roughly half the cells in a chordoma are tumour cells and the rest are fibroblasts and immune cells, and the tumour cells come in several states, including a stem-like, radiation-resistant group that tends to sit in one part of the tumour. Ordinary chordoma uses little glucose on PET and is patchy. Nobody has measured a dedifferentiated area on PET in chordoma, but in related bone tumours the high-grade part lights up far more. All of this is why a single needle can miss the worst part, and why the report should say exactly where the sample came from and whether it matched the imaging.