Clival chordoma research
Home › Diet, supplements and claims

Metabolic switching and aggressive behaviour

Agent report for Kya, 15 September 2026, written after her partner asked in em-research: "When a tumour switches metabolism is there any research showing more aggressive behaviour?" The worry behind the question is specific and fair: that a high-fat or ketogenic diet pushes the tumour onto ketones and fatty acids, and that the act of switching is itself a step towards a nastier tumour. Working frame: probable clival chordoma, chondrosarcoma the main imaging lookalike, dedifferentiated or poorly differentiated subtypes still possible, no histology, 51 kg. Every PMID below was verified through NCBI E-utilities on 15 September 2026. No survival figures for the patient, by rule.

Evidence levels. L1 randomised trial, meta-analysis or guideline. L2 prospective non-randomised trial or large cohort. L3 retrospective human series. L4 case reports or very small series. L5 laboratory, animal or cell work. Where a study is L5 I say whether it was done in vivo (in a living animal) or in vitro (in a dish), because for this question that distinction carries most of the weight.

Thread 24 covered whether these tumours can burn ketones. This thread is the next question: if they do, does it make them worse. It does not repeat Vargas-Lopez 2024 (chondrosarcoma ketolysis), Sperry 2020 (glioblastoma fatty acid oxidation, unrestricted ketogenic diet shortened survival in some models) or ERGO2 (Voss 2020, negative randomised trial); those are cited and stand.

Tier A. The short answer to his question

Yes. There is a real body of evidence, mostly in animals, that the metabolic state a tumour is pushed into changes how it behaves, and that several of those pushes make it behave worse. But the evidence is about particular fuels and particular stresses, not about "switching" as a general act, and the strongest findings point at two things her partner has already half-identified: fat, especially one specific saturated fat, and metabolic stress from starvation, hypoxia or acid.

Four points, in plain terms.

  1. The single most directly relevant result is about dietary fat, not about ketosis. In mice, a short spell on a palm-oil-rich diet made oral cancers and melanoma permanently more metastatic, and the effect was still there generations of tumour transplants later, after the fat was long gone. Palmitic acid did it. Oleic acid and linoleic acid did not. That is an intervention, in a living animal, with a lasting result (Pascual 2021, Nature, PMID 34759321, L5 in vivo).
  2. Feeding a tumour ketones directly has made tumours grow faster in a mouse, in at least one experiment. Beta-hydroxybutyrate given to breast cancer xenografts increased tumour growth about 2.5-fold (Bonuccelli 2010, Cell Cycle, PMID 20818174, L5 in vivo). And a high-fat ketogenic diet raised blood acetoacetate and grew BRAF-mutant melanoma faster (Xia 2017, Cell Metab, PMID 28089569, L5 in vivo). Both are mutation-specific or model-specific. Neither has been done in chordoma.
  3. Metabolic stress does select and drive nastier behaviour, and this is the part of her partner's intuition that is best supported. Starving cells of glucose selected for cells with KRAS pathway mutations (Yun 2009, Science, PMID 19661383, L5 in vitro). Cutting off a tumour's blood supply, which starves and suffocates it, made tumours invade and metastasise more in mouse models (Paez-Ribes 2009 and Ebos 2009, both Cancer Cell, PMIDs 19249680 and 19249681, L5 in vivo). Low oxygen turns on EMT and stem-like programmes (Yang 2008, Nat Cell Biol, PMID 18297062; Heddleston 2009, Cell Cycle, PMID 19770585, both L5). Acid made tumours invade, and neutralising it stopped them (Estrella 2013, Cancer Res, PMID 23288510, L5 in vivo).
  4. And it cuts the other way too, which is why this is not a simple answer. Ketogenic diets slowed melanoma in mice regardless of the tumour's genetics or its metabolic flexibility (Weber 2022, Cancer Metab, PMID 35851093, L5 in vivo). Beta-hydroxybutyrate suppressed colorectal cancer in mice (Dmitrieva-Posocco 2022, Nature, PMID 35477756, L5 in vivo). A ketogenic diet made a PI3K inhibitor work much better (Hopkins 2018, Nature, PMID 30051890, L5 in vivo), although in the leukaemia model in that same paper the ketogenic diet on its own accelerated the disease.

So the honest summary is: fuel state changes behaviour, the direction depends on the tumour, and nobody has run the experiment in chordoma. What survives all of it as a practical point is the fat one, and it argues for the type of fat in the diet rather than against fat as a calorie source.

Tier B. Evidence that the switch CAUSES more aggressive behaviour

These are intervention experiments: something was changed, and the tumour behaved differently. All L5. I have marked in vivo or in vitro for each, because a result in a living animal is worth far more here than a result in a dish.

B1. Dietary fat, and one fat in particular

Pascual 2017 (Nature, PMID 27974793, L5 in vivo and in vitro, with a human correlation). A subpopulation of CD44-bright cells in human oral carcinomas expressing high levels of the fatty acid receptor CD36 were the only cells able to start a metastasis. Palmitic acid, or a high-fat diet, specifically boosted the metastatic potential of those CD36-positive cells, and did so in a CD36-dependent way. Blocking CD36 with antibodies caused almost complete inhibition of metastasis in orthotopic mouse models, with no side effects reported. Clinically, the presence of CD36-positive metastasis-initiating cells correlated with poor prognosis across several carcinoma types. The authors' conclusion is the sentence her partner should read: metastasis-initiating cells "particularly rely on dietary lipids to promote metastasis".

Pascual 2021 (Nature, PMID 34759321, L5 in vivo) is the follow-up and the more alarming paper. Dietary palmitic acid, but not oleic acid and not linoleic acid, promoted metastasis in oral carcinomas and melanoma in mice. Tumours from mice fed a short-term palm-oil-rich diet, or cells briefly exposed to palmitic acid in a dish, stayed highly metastatic even after being serially transplanted without any further exposure to palmitic acid. The memory required CD36 and was carried by a stable chromatin mark, H3K4 trimethylation deposited by Set1A/COMPASS, and it worked through a neural programme that stimulates intratumoural Schwann cells and innervation. So this is not a transient metabolic preference. A dietary metabolite installed a durable, heritable, pro-metastatic state.

Two honest limits on Pascual. First, these are carcinomas and melanoma, epithelial and neural-crest tumours with an EMT and metastasis biology that chordoma does not obviously share; chordoma is locally destructive far more than it is metastatic. Second, the exposures were high-palmitate palm-oil diets in mice, not a British kitchen. But the mechanism is a fatty acid receptor and a histone mark, both of which chordoma cells have, and the 2026 chordoma work in Tier D2 shows chordoma does run a fatty acid programme when stressed.

B2. Ketones fed directly to a tumour

Bonuccelli 2010 (Cell Cycle, PMID 20818174, L5 in vivo). In MDA-MB-231 breast cancer xenografts, administration of 3-hydroxybutyrate (beta-hydroxybutyrate) increased tumour growth by about 2.5-fold, with no measurable increase in tumour blood vessel formation. Both beta-hydroxybutyrate and L-lactate acted as chemoattractants for cancer cells, and L-lactate stimulated lung metastases by about 10-fold without increasing primary tumour growth. This is the "reverse Warburg" group, and the model is an extreme one (metabolite given to a xenograft), but it is the clearest single demonstration that a ketone body can be growth-promoting in a live animal.

Xia 2017 (Cell Metab, PMID 28089569, L5 in vivo). The ketone body acetoacetate selectively enhances BRAF V600E-dependent MEK1 activation. A high-fat ketogenic diet raised serum acetoacetate and increased the growth of BRAF V600E human melanoma cells in xenograft mice, and lowering circulating acetoacetate with hypolipidaemic agents, or blocking the acetoacetate to BRAF interaction, attenuated that growth. The authors explicitly frame the lesson as diet needing to be matched to a tumour's mutation profile, which they call "precision diets". Relevance to the patient: this effect requires a BRAF V600E mutation. Chordoma is not a BRAF V600E disease. It is cited here as proof of principle that a ketone body can act as a signalling molecule that switches on a growth pathway, not just as a fuel.

B3. Metabolic stress selecting or driving aggressive cells

Yun 2009 (Science, PMID 19661383, L5 in vitro). Colorectal cancer cells with KRAS or BRAF mutations upregulate GLUT1, take up glucose faster, and survive low glucose. When wild-type KRAS cells were put in a low-glucose environment, very few survived; most survivors expressed high GLUT1, and 4 per cent of those survivors had acquired KRAS mutations that their parent cells did not have. This is the cleanest published demonstration that nutrient deprivation can select for cells with a more aggressive driver mutation. It is in a dish, in a colorectal line, and the analogy to a dietary intervention in a person is loose. But it is the mechanism her partner is worried about, shown working.

Paez-Ribes 2009 (Cancer Cell, PMID 19249680, L5 in vivo) and Ebos 2009 (Cancer Cell, PMID 19249681, L5 in vivo), published back to back. Anti-angiogenic drugs targeting the VEGF pathway shrank tumours in mouse models of pancreatic neuroendocrine carcinoma and glioblastoma but elicited adaptation and progression to greater malignancy, with heightened invasiveness and in some cases increased lymphatic and distant metastasis; genetic deletion of Vegf-A did the same, so it was not an off-target drug effect. Ebos showed sunitinib accelerated metastatic growth and decreased overall survival in mice in several metastasis assays, including when given before tumour cells were even implanted, suggesting conditioning of the host tissues. These two papers are the reason oncologists take "starving a tumour makes it meaner" seriously: it is the best-documented instance of a metabolic and microenvironmental squeeze backfiring in vivo.

Hypoxia driving EMT and stemness. Yang 2008 (Nat Cell Biol, PMID 18297062, L5, in vitro and in vivo with human tissue correlation): HIF-1alpha binds the TWIST promoter directly and drives EMT and metastatic phenotypes; silencing TWIST reversed it; co-expression of HIF-1alpha, TWIST and Snail in head and neck primaries correlated with metastasis and worst prognosis. Heddleston 2009 (Cell Cycle, PMID 19770585, L5): hypoxia promoted self-renewal in both stem and non-stem glioblastoma populations, pushed non-stem cells towards a stem-like phenotype with upregulation of OCT4, NANOG and c-MYC, and forced expression of non-degradable HIF-2alpha increased the tumour-forming potential of the non-stem population. This is dedifferentiation driven by a microenvironmental metabolic stress, which is the closest thing in the literature to her partner's specific fear.

Acid. Estrella 2013 (Cancer Res, PMID 23288510, L5 in vivo). Using intravital microscopy, the regions of greatest tumour invasion corresponded to the areas of lowest extracellular pH, and tumours did not invade into tissue with normal or near-normal pH. Oral sodium bicarbonate raised peritumoral pH and inhibited tumour growth and local invasion. Invasive-edge cells expressed GLUT1 and the sodium-hydrogen exchanger. Fermentative metabolism generates the acid; the acid enables the invasion.

B4. Oxidative metabolism and the cells that survive treatment

This is the version of the question that matters most for the patient, because it is about radiotherapy and drug resistance rather than about diet.

LeBleu 2014 (Nat Cell Biol, PMID 25241037, L5 in vivo and in vitro, with human tissue correlation). Migratory and invasive cancer cells specifically favour mitochondrial respiration and increased ATP production, using PGC-1alpha to raise oxidative phosphorylation, mitochondrial biogenesis and oxygen consumption. In human invasive breast cancers, PGC-1alpha expression in invasive cells correlated strongly with distant metastasis. Silencing PGC-1alpha suspended invasive potential and attenuated metastasis without affecting proliferation, primary tumour growth or the EMT programme. That last clause is important: this is a metastasis-specific metabolic requirement, not general growth.

Farge 2017 (Cancer Discov, PMID 28416471, L5 in vivo, patient-derived xenografts, with a human signature). Acute myeloid leukaemia cells surviving cytarabine were not the quiescent stem cells everyone expected. They had high reactive oxygen species, increased mitochondrial mass, active polarised mitochondria, increased fatty acid oxidation and upregulated CD36, in other words a high-OXPHOS state. High-OXPHOS human lines were chemoresistant in vivo, low-OXPHOS lines were not, and the high-OXPHOS gene signature predicted treatment response in both xenografts and patients. Blocking mitochondrial translation, electron transfer or fatty acid oxidation shifted cells to low OXPHOS and markedly improved the effect of chemotherapy.

Viale 2014 (Nature, PMID 25119024, L5 in vivo). In an inducible KRAS pancreatic cancer model, the dormant cells that survived switching the oncogene off, and that were responsible for relapse, had cancer stem cell features and relied on oxidative phosphorylation, with high expression of mitochondrial, autophagy and lysosome genes and reduced dependence on glycolysis. They were highly sensitive to OXPHOS inhibitors, which suppressed recurrence.

Sharma 2010 (Cell, PMID 20371346, L5 in vitro) is the origin of the persister concept: a small reversibly drug-tolerant subpopulation with more than 100-fold reduced drug sensitivity, maintained by IGF-1 receptor signalling and an altered chromatin state requiring the demethylase KDM5A, entered and left transiently. Not a metabolic paper as such, but it is the framework into which the OXPHOS-persister findings fit.

What B4 adds up to. The state that survives treatment is repeatedly the mitochondrial, fat-burning, slow one. That is an argument for asking the oncology team about metabolic vulnerabilities of residual disease after treatment. It is not evidence that eating fat creates those cells. The stress in every one of these experiments was a drug, not a dinner.

Tier C. Evidence that it only CORRELATES: human tissue series

These are human studies. They are higher up the evidence ladder for "is this real in people" and lower down for "does it cause it". None of them involved changing anyone's diet.

C1. Chordoma, and this is the most relevant human evidence that exists

Zhang Q 2024 (Nat Commun, PMID 39333076, L3, 187 skull base chordoma tumours). Integrated genomics, transcriptomics, proteomics and phosphoproteomics. The findings that matter here:

So in the largest molecular series of skull base chordoma ever assembled, a mitochondrial or oxidative signature travels with the more aggressive tumours. It is an association. The direction of causation is unknown, and the obvious alternative reading is that whatever makes a chordoma unstable and aggressive also makes it need more mitochondria, rather than the mitochondria making it aggressive.

Yin H 2024 (Cell Rep Med, PMID 39368483, L3, 102 chordoma patients, with in vitro and in vivo validation). Tandem mass tag proteomics and clustering split chordoma into three molecular subtypes: bone-microenvironment-dominant, mesenchymal-derived, and mesenchymal-to-epithelial-transition-mediated. The three subtypes had discrete clinical prognosis and distinct biological attributes: osteoclastogenesis and immunogenicity, oxidative phosphorylation, and receptor tyrosine kinase activation respectively, with proposed matched treatments of denosumab, S-Gboxin (an OXPHOS inhibitor) and anlotinib. Each showed positive treatment effect in its subtype in vitro and in vivo.

This is the single most useful fact in this thread for the patient's team. There is a published, proteomically defined chordoma subtype whose defining biology is oxidative phosphorylation, and the authors propose an OXPHOS inhibitor for it. That makes the metabolic state of her particular tumour a question with a possible answer once there is tissue, and it is worth asking whether the operating unit or the national MDT can run or access this kind of proteomic subtyping on her specimen. It is also, note, the mirror image of her partner's worry: in that paper, OXPHOS-high is a treatment target, not a doom signal.

He 2024 (Spine, PMID 38251727, L3, small: 32 chordoma samples against 14 nucleus pulposus controls). HIF-1alpha was expressed in all chordoma samples and significantly higher than in normal nucleus pulposus (p = 0.004). On multivariate Cox analysis, high HIF-1alpha expression was associated with shorter continuous disease-free survival (hazard ratio about 3.4, p = 0.021) and shorter overall survival (hazard ratio about 4.5, p = 0.012), with expression in both cytoplasm and nucleus a stronger predictor. Relative effects only, in a series of 32 spinal tumours, and it says nothing about any individual. The point for this thread is only that the hypoxia response is on in chordoma and tracks with worse behaviour, exactly as it does in the glioma and head and neck literature.

Zhang TL 2024 (Adv Sci, PMID 39207055, L5 with human correlation). Single-cell and spatial work in chordoma: the hypoxic microenvironment reprogrammes cancer-associated fibroblasts into an endoplasmic-reticulum-stress subtype via transcriptional upregulation of IER2, which then drives a pro-tumour paracrine GMFG signal onto ITGB1 on the tumour cells; ITGB1 inhibition attenuated malignant progression. High IER2 and ITGB1 expression correlated with aggressive phenotype and poor survival. So in chordoma specifically, low oxygen has been shown to remodel the supporting cells into a state that pushes the tumour along. That is metabolic stress causing aggressiveness, in chordoma, though via the stroma rather than via the tumour's own fuel choice.

Jian 2018 (Am J Transl Res, PMID 29636883, L5 in vitro). Brachyury, the chordoma-defining transcription factor, directly upregulates carbonic anhydrase IX, and CA9 was responsible for paclitaxel resistance in a primary chordoma line; brachyury and CA9 levels correlated in chordoma tissue. CA9 is the classic hypoxia and acid-handling enzyme. This links the tumour's own identity gene to its pH and hypoxia adaptation and to drug resistance.

C2. Chondrosarcoma

Annovazzi 2019 (Ann Nucl Med, PMID 31396797, L3, 95 patients with chondroid lesions). FDG PET SUVmax tracked histological grade: best cutoff 2.6 for enchondroma versus low-grade chondrosarcoma (sensitivity 0.68, specificity 0.86), 3.7 for low-grade versus intermediate or high-grade (0.83 and 0.84), and 7.7 to separate intermediate or high-grade chondrosarcoma from dedifferentiated chondrosarcoma (sensitivity 0.92, specificity 0.90). Note also, and this is independently useful for the biopsy discussion, that preoperative biopsy agreed with the final grade in only 78.3 per cent of cases overall and in just 58.6 per cent for intermediate and high-grade tumours.

What that means and does not mean. Glucose avidity rises steeply with dedifferentiation in cartilage tumours, and a dedifferentiated chondrosarcoma is a glucose-hungry tumour, not a ketone-hungry one. It is a correlation with grade, and it is the aggressive component that is metabolically active, so this is a marker of the dedifferentiated clone rather than evidence that glucose caused it. Practically it is an argument for asking whether an FDG PET would add information about whether there is a high-grade or dedifferentiated component, which is a question already worth putting to the team for other reasons.

C3. The wider human picture

Across tumour types, metabolic gene signatures, FDG avidity and mitochondrial markers correlate with grade, invasion and worse outcome, and reviews describe tumours moving between glycolysis and oxidative phosphorylation according to what is available rather than being locked into either. There is also a well-documented human counter-example inside melanoma: Vazquez 2013 (Cancer Cell, PMID 23416000, L5 with human tissue) found PGC-1alpha-high melanomas have greater mitochondrial capacity and resistance to oxidative stress, while Luo 2016 (Nature, PMID 27580028, L5) found PGC-1alpha suppresses melanoma metastasis through a separate ID2-TCF4-integrin axis, with elevated PGC-1alpha inversely correlated with vertical growth in human specimens. Same coactivator, opposite sign to LeBleu's breast cancer result. The field genuinely does not have one rule.

Tier D. Counter-evidence, honestly

If this thread only carried Tier B it would be misleading.

D1. Forcing a metabolic switch that made tumours less aggressive

Weber 2022 (Cancer Metab, PMID 35851093, L5 in vivo) is the most direct rebuttal of the "switching makes it worse" thesis, because it deliberately tested metabolic flexibility. Mice bearing BRAF-mutant, NRAS-mutant and wild-type human melanoma xenografts, plus immunocompetent mice with highly metastatic melanoma allografts, were fed control or ketogenic diets. Ketogenic diets reduced tumour growth regardless of the tumour's genetic background, its metabolic signature or the host immune status, and reduced metastasis in the immunocompetent model. The title says it: ketogenic diets slowed melanoma growth in vivo regardless of tumour genetics and metabolic plasticity.

Dmitrieva-Posocco 2022 (Nature, PMID 35477756, L5 in vivo). A dietary screen in autochthonous mouse models of colorectal cancer found ketogenic diets had a strong tumour-inhibitory effect, reproduced by beta-hydroxybutyrate alone, which reduced colonic crypt cell proliferation and potently suppressed intestinal tumour growth, acting through the receptor Hcar2 and the regulator Hopx. Human organoid and single-cell data from patient biopsies were consistent. This is the exact same molecule as Bonuccelli's 2.5-fold growth increase, in a different tissue, doing the opposite. Beta-hydroxybutyrate is not simply food; it is a signalling molecule, and what it signals depends on the receptor the tissue carries.

Hopkins 2018 (Nature, PMID 30051890, L5 in vivo). PI3K inhibitors cause a glucose and insulin rebound that reactivates the very pathway they block. Preventing that feedback with an SGLT2 inhibitor or with a ketogenic diet greatly improved the efficacy to toxicity ratio of PI3K inhibitors across patient-derived xenografts of endometrial and bladder cancer, and allograft models of PIK3CA-mutant breast cancer and MLL-AF9 acute myeloid leukaemia. And the caution, in the authors' own words: "treatment with the ketogenic diet alone had variable effects in different tumor models", and in the AML model "the ketogenic diet alone accelerated disease progression suggesting that this diet may be detrimental for some cancer patients when used in isolation." Both halves of that paper matter. Keto as an adjunct to a drug whose resistance mechanism is insulin: helpful. Keto on its own: model-dependent, and in one model harmful.

Lien 2021 (Nature, PMID 34671163, L5 in vivo) explains why the answers keep disagreeing, and it is the most intellectually satisfying paper in this thread. Calorie restriction and a ketogenic diet both lower glucose and insulin, but only calorie restriction inhibited growth of the tumour allografts tested. The difference was lipids: calorie restriction lowered plasma and tumour lipid levels, and tumours then needed stearoyl-CoA desaturase to make their own monounsaturated fats; calorie restriction also impaired tumour SCD activity, creating a mismatch between unsaturated and saturated fatty acids that slowed growth. A ketogenic diet impaired SCD activity too, but supplied so much fat that the tumour's unsaturated to saturated ratio stayed fine. Changing the fat composition of the ketogenic diet to raise tumour saturated fat did slow growth. Conclusion: what matters is not glucose or ketones but whether the diet creates a mismatch between the fats the tumour can make and the fats it is handed.

Read Lien next to Pascual and the practical shape of the whole question changes: the composition of the fat may matter more than the presence of ketones.

D2. And a chordoma-specific paper that cuts both ways

Wei 2026 (bioRxiv preprint, PMID 42182510, L5 in vitro and in vivo, not peer reviewed, flagged as a preprint). In chordoma: recurrence correlated with coordinated upregulation of monounsaturated and polyunsaturated fatty acids and a low PUFA to MUFA ratio, producing a lipid-peroxidation-resistant state that protected against DNA damage and cell death. Single-cell metabolic profiling found a fatty-acid-biosynthesis-high subpopulation coupled to stemness. Radiotherapy tolerance was linked to elevated FASN and lipid droplet expansion, MUFA loading alone phenocopied radiotherapy tolerance, and lipid droplets accumulated after irradiation via ROS, ER stress, SREBP1 and FASN. Pharmacological FASN blockade restored radiosensitivity in vitro and in vivo in a patient-derived xenograft.

This is the only paper in existence linking fat metabolism, stemness and treatment resistance in chordoma, and it is a preprint. It supports her partner's instinct in the most specific way available: in chordoma cells, loading monounsaturated fat was enough to make them behave like radiation-tolerant cells. It also supports Lien: the species of fat is the variable. It is not a dietary study, the MUFA loading was done to cells, and no animal in it was fed anything. It should not be used to make a diet decision. It should be used to ask her radiotherapy team a question.

Tier E. What exists and what does not, for chordoma and chondrosarcoma

Said plainly, because the gaps are as important as the findings.

What exists.

What does not exist. None of the following has been published, as of 15 September 2026:

So the specific fear, that a ketogenic diet will convert a conventional chordoma into a dedifferentiated one, has no supporting evidence and also no refuting evidence. It is untested. What can be said is that the known drivers of dedifferentiation in these tumours are genetic and epigenetic events (chromosome instability, 1p36 and 9p21 loss, TP53, SMARCB1 loss), and no dietary trigger has been identified for any of them.

Tier E2. Dietary fat itself, separate from ketosis

Her partner's question deserves this split out, because it is the one place the evidence is strong and it is about fat rather than about ketones.

Yes. Dietary fat has in vivo evidence of promoting metastasis, independent of ketosis, and it is specific to palmitic acid. Pascual 2017 (PMID 27974793) and Pascual 2021 (PMID 34759321), both Nature, both L5 in vivo:

Neither experiment was in ketosis. The mice were not ketogenic; they were on a high-palmitate diet. This is the cleanest piece of evidence in the whole thread, and it points not at ketosis and not at fat as a calorie source but at one particular saturated fat.

Where palmitic acid comes from in a real diet: palm oil and palm kernel oil (and therefore many processed foods, biscuits, spreads and ice cream), coconut oil to a lesser degree, butter, cream, cheese, lard, beef fat, and processed meat. Where the alternatives come from: olive oil and rapeseed oil (oleic acid), avocado, nuts, oily fish. The honest caveat is that mouse palm-oil diets are far more palmitate-dense than a normal human diet, and no human trial has tested palmitate restriction as a cancer intervention. The cost of preferring olive oil to palm oil is zero, which is the only reason this is worth acting on at all.

Tier F. What this does and does not mean for the patient's food

The established need, which nothing in this thread touches, is calories and protein before surgery. ESPEN's practical guideline on clinical nutrition in cancer (Muscaritoli 2021, Clin Nutr, PMID 33946039, L1 guideline) and ESPEN's surgery guideline (Weimann 2021, Clin Nutr, PMID 34242915, L1 guideline) both point the same way: identify and treat malnutrition, do not restrict intake, and optimise nutritional state before a major operation. At 51 kg and facing skull base surgery, the intervention with actual evidence behind it is eating enough. That remains the priority and this thread does not qualify it.

Separate the three things that keep getting run together.

  1. High fat as a way to get calories in. Evidenced, sensible, and the reason the butter, cream, cheese and oils are in the plan. Keep.
  2. Deliberate ketosis as tumour treatment. Unproven in people, failed in the only randomised brain tumour trial, contradicted mechanistically in chondrosarcoma, and in two animal models actively harmful (unrestricted keto shortening survival in some glioblastoma models, Sperry 2020; keto alone accelerating AML, Hopkins 2018). This thread does not change the position taken in thread 24: keep the fat, drop the goal.
  3. The composition of the fat. New, and the one genuinely actionable point in this thread. On Pascual and on Lien, the species of fat plausibly matters more than the presence of ketones. A shift in the mix towards olive oil, rapeseed oil, avocado, nuts and oily fish, and away from palm oil and palm-oil-containing processed food, costs nothing, does not reduce calories, and is the only change this thread supports. Butter, cream and cheese are palmitate sources, and they are also the things she is actually managing to eat, so this is a gentle rebalancing and not a removal. Calories still win any conflict.

What I would not do on the strength of this thread. I would not stop the high-fat diet. I would not cut butter or cream while she is underweight. I would not add any supplement or oil on the basis of a mouse experiment. And I would not treat the Wei preprint as a reason to change food, because it did not feed anything to anyone.

Three questions this thread generates for the team, which is where it should go.

  1. Once there is tissue, can the specimen be characterised metabolically, including the possibility of the published proteomic subtyping in which one chordoma subtype is defined by oxidative phosphorylation (Yin 2024)? Is chromosome 1q status being assessed, given the association with worse outcome in the 187-tumour series (Zhang Q 2024)?
  2. Is there a role for FDG PET in assessing whether there is a high-grade or dedifferentiated component, given that SUVmax separates high-grade from dedifferentiated chondrosarcoma with sensitivity 0.92 and specificity 0.90 (Annovazzi 2019), and that biopsy agreed with final grade in only 58.6 per cent of intermediate and high-grade cases in that series?
  3. Is the radiotherapy team aware of the fatty acid synthesis and lipid droplet work on radioresistance in chordoma, and does it have any bearing on dose, fractionation or on what is worth measuring in the specimen (Wei 2026, preprint)?

And the one honest closing line for her partner. The worry is not silly and it is not resolved. The mechanism he is describing is real in animals, it has been demonstrated in chordoma stroma with hypoxia and in chordoma cells with fat loading, and nobody has tested a diet against it. But the evidence that exists points at the kind of fat rather than at ketones, and at drugs and radiotherapy rather than at dinner, as the things that actually push a tumour into its resistant state. The decision in front of the family this month is surgery, tissue and radiotherapy. Food is the supporting act, and its job is to get her to the operation strong.

Sources

All PMIDs verified through NCBI E-utilities on 15 September 2026.

Intervention evidence that a metabolic state or fuel increases aggressiveness (all L5).

Human correlative evidence.

Counter-evidence.

Chordoma fat metabolism and radioresistance.

Nutrition guidelines.

Cited from thread 24, not repeated here.

Related repo threads. Chordoma metabolism and dedifferentiation: thread 21. Ketone utilisation and immunogenicity: thread 24. Outcome determinants: thread 19. Chordoma cells and dedifferentiation: thread 16.

Research thread 25 · source file em-research/25-metabolic-switching-and-aggressiveness.md · last edited 15 September 2026