Working Paper · Nutritional Biochemistry · Oncology
How mitochondrial dysfunction may drive the cancer cell’s signature behaviours — and why treating it as primarily a genetic disease has stalled progress for two decades.
Cancer is almost universally framed as a genetic disease: accumulating somatic mutations corrupt the machinery of growth control, and treatment is directed at correcting or killing those specific mutant cells. The evidence for this framing is real — virtually every cancer carries identifiable mutations. But the genetic model has not translated into proportional improvements in survival, and a quiet alternative has gathered substantial experimental support: that the primary lesion in most cancers is mitochondrial dysfunction, and that the mutations we observe are downstream consequences of the energy crisis that follows, not upstream causes. This paper examines both views fairly, presents the mechanistic evidence for the metabolic model, and explores what accepting it even partially would mean for treatment. The conclusion is not that genetics is irrelevant — some cancers are predominantly genetic — but that treating every cancer as a genetic disease, while ignoring its universal metabolic phenotype, may explain why the therapeutic needle has barely moved.
The dominant framework for understanding cancer comes from Hanahan and Weinberg, who proposed in 2000 that all cancers, regardless of tissue origin or genetic makeup, acquire the same six capabilities during their development.1 This was updated in 2011 to ten hallmarks, with reprogrammed energy metabolism — the Warburg effect — finally included as a named feature.2
Cancer cells generate their own mitogenic signals rather than waiting for external cues, bypassing the normal requirement for growth-factor stimulation.
Normal tissue homeostasis is enforced partly by anti-proliferative signals; cancer cells rewire to ignore or override them, most commonly via loss of RB pathway function.
Apoptosis is the fail-safe mechanism that eliminates damaged or aberrant cells. Cancer cells disable it, most often through p53 loss or BCL-2 overexpression, allowing damaged cells to survive and divide.
Normal cells have a finite replication capacity enforced by telomere shortening; cancer cells reactivate telomerase (or use alternative lengthening mechanisms) to divide indefinitely.
Tumors beyond a few millimeters must induce new blood vessel formation to receive oxygen and nutrients; they do so by upregulating VEGF and other angiogenic factors while suppressing inhibitors.
The shift from localized to invasive disease, and ultimately to distant metastasis, involves suppression of cell-adhesion molecules and upregulation of matrix metalloproteinases enabling migration.
Aerobic glycolysis (the Warburg effect) and glutamine dependence are now recognized as hallmarks, as is the cancer cell’s ability to avoid immune destruction through checkpoint upregulation.
The high mutational burden of cancer cells reflects an acquired defect in DNA-repair fidelity. Chronic inflammation in the tumor microenvironment supplies growth factors and promotes invasion.
What the hallmarks describe is more important than their catalogue. Every one of these features converges on a common metabolic output. Self-sufficiency in growth signals means upregulating the same nutrient-sensing and proliferation pathways — PI3K/AKT/mTOR — that also govern glucose and glutamine uptake. Evasion of apoptosis involves loss of cytochrome c release from mitochondria, which is itself tied to cardiolipin integrity (see Part III). Limitless replication requires massive anabolic substrate. Angiogenesis is induced partly by hypoxia-inducible factor (HIF-1α), which is itself activated when mitochondria fail to consume oxygen efficiently. The hallmarks read, from a metabolic perspective, less like independent features and more like downstream consequences of a single underlying disruption.
Large-scale genomic analyses — most prominently The Cancer Genome Atlas — have documented the frequency of pathway-level alterations across thousands of tumors:3
These numbers are striking, but their interpretation matters enormously. The observation that nearly every cancer alters the RTK/RAS/PI3K axis does not, by itself, tell us whether those alterations are the cause or a consequence of the disease. Consider what these pathways do: RTK/RAS/PI3K signaling promotes glucose uptake, glycolysis, and biosynthetic precursor generation — exactly what a cell with impaired oxidative phosphorylation would need to upregulate to survive. p53 normally triggers apoptosis in cells with DNA damage or energetic stress; losing it lets energetically compromised cells escape death. RB normally restrains entry into the cell cycle; losing it enables replication even when cellular conditions are suboptimal.
The metabolic interpretation is not that these mutations are unimportant — they clearly facilitate cancer progression. It is that the same metabolic pressure (mitochondrial dysfunction, energy deficit) might be the upstream event that creates selection pressure for cells that acquire these mutations, rather than the mutations themselves being the originating lesion.
Some cancers are straightforwardly genetic in their origin. BRCA1/2 mutations in breast and ovarian cancer, APC mutations in familial adenomatous polyposis, RB1 loss in retinoblastoma, and BCR-ABL in chronic myeloid leukemia are genuine driver events — heritable, mechanistically understood, and in the case of CML, successfully targeted (imatinib). The extraordinary success of imatinib — converting CML from a near-uniformly fatal disease to one manageable with a daily pill — proves that targeting a specific oncogenic mutation can be curative, at least when a single driver dominates. Pediatric cancers as a group tend to have simpler mutational landscapes and are more tractable to genetic approaches. The claim this paper examines is not that all cancers are purely metabolic, but that the metabolic dimension is consistently underweighted relative to the genetic one.
Whatever their tissue of origin, whatever mutations they carry, cancer cells share a metabolic fingerprint that has been recognized since Otto Warburg’s observations in the 1920s and confirmed by modern metabolomics and imaging:
What makes this phenotype so striking is its consistency. A lung adenocarcinoma and a colon adenocarcinoma may share almost no mutations, but they share this metabolic profile almost entirely. This universality is exactly what a single upstream cause — mitochondrial dysfunction — would predict.
If cancer were primarily a genetic disease, we would expect to find a small set of driver genes — mutated in most cancers, necessary and (in combination) sufficient to cause cancer, and targetable to produce reliable cures. The actual picture is more complicated. The Cancer Genome Atlas found that while pathway-level alterations are nearly universal, the specific genes mutated within those pathways vary enormously between patients and cancer types. The number of meaningful driver mutations per cancer is typically 2–8, but they are drawn from a long tail of possible genes — researchers have identified hundreds of potential cancer driver genes, with no single gene mutated across a majority of cancer types.3
This matters for treatment. If the specific mutation varies, targeted therapies aimed at specific mutations will work only in the patients who carry that mutation. The history of oncology since the genomic revolution has been one of spectacular successes in narrow populations (imatinib in BCR-ABL–positive CML; EGFR inhibitors in EGFR-mutant lung cancer; BRAF inhibitors in V600E-mutant melanoma) and broad frustration elsewhere. Overall cancer mortality — adjusted for age and population — has improved only modestly over the past two decades for most solid tumor types, despite enormous investment.4 Survival gains are real and meaningful for specific cancers, but the broad promise of precision oncology has not fully materialized.
Cancer survival statistics are genuinely complicated. Age-adjusted incidence and mortality have improved for some cancers (colorectal, cervical, some childhood cancers) due to screening, vaccination, and genuine treatment advances. The claim here is specifically about mortality from solid tumors after diagnosis, where the picture is more equivocal. Immunotherapy (checkpoint inhibitors) is a legitimate exception, with durable responses in a subset of patients — notably, its mechanism (restoring immune recognition) is not dependent on identifying specific driver mutations in each patient, which is consistent with targeting a universal cancer vulnerability rather than an individual genetic lesion.
Some of the most revealing evidence against a purely genetic model comes from experiments designed to support it. Several studies have introduced known cancer-causing nuclear DNA (with confirmed oncogenic mutations) into cells with normal mitochondria, and conversely transferred normal nuclei into cells with cancerous mitochondria. The results are difficult to explain under strict genetic primacy:5
None of these experiments are definitive on their own, and critics raise valid concerns about artifact and interpretation. But the pattern they describe — that the mitochondrial environment shapes tumorigenic capacity more strongly than the nuclear genotype — is not easily accommodated within a purely genetic framework.
The canonical cancer research models — the systems through which most experimental oncology is conducted — have systematic limitations that are relevant to the treatment gap:
Mitochondrial abnormalities in cancer are not subtle. Across malignancy types, electron microscopy and biochemical analyses consistently show:6
Roskelley and colleagues documented over decades that cytochrome oxidase (Complex IV) activity is deficient in nearly all highly malignant cancers examined — a finding that predates the molecular biology era but remains significant.7 Complex IV is the terminal electron acceptor in the ETC and the site at which oxygen is consumed. Its deficiency means the cell cannot efficiently complete oxidative phosphorylation regardless of how much glucose and oxygen are available.
More recent work has identified two enzymes as consistently altered across malignancies: GAPDH (glyceraldehyde-3-phosphate dehydrogenase) is elevated, reflecting upregulation of glycolysis; and β-F1-ATPase (the catalytic subunit of ATP synthase, Complex V) is reduced. This is mechanistically important: β-F1-ATPase is not only required for oxidative phosphorylation but also for the intrinsic apoptotic pathway — its reduction simultaneously impairs mitochondrial ATP production and apoptotic signaling, which would contribute to evasion of cell death.8
Of all the molecular abnormalities in cancer mitochondria, cardiolipin dysregulation may be the most mechanistically important and the most underappreciated in mainstream oncology discussion.9
Cardiolipin (CL) is a unique phospholipid found almost exclusively in the inner mitochondrial membrane (IMM), where it constitutes approximately 20% of total lipid. Its structure — a dimeric phosphoglycerol backbone with four fatty acid chains, almost entirely long-chain polyunsaturated fatty acids (primarily linoleic acid, C18:2, in humans) — is essential for the biophysical properties of the IMM. No short-chain saturated or monounsaturated fatty acid can substitute; the specific geometry of linoleic acid enables the tight membrane curvature of cristae and the proximity of ETC complexes required for efficient electron transfer.9
Cardiolipin performs several specific functions that are disrupted in cancer:
In cancer, cardiolipin is consistently abnormal: total CL content is reduced, the fatty acid composition is shifted (often with less linoleic acid and more of other species), and CL is oxidized at higher rates. The remodeling enzymes that maintain CL composition — particularly Tafazzin (TAZ) — are dysregulated.9
The requirement for linoleic acid in cardiolipin might seem to argue against the seed oils paper’s claim that excess dietary linoleic acid is harmful. It does not. The two arguments are complementary, not contradictory, and the connection runs exactly where you would expect it to.
The seed oils paper does not argue that linoleic acid in cardiolipin is the problem. It argues that excess dietary LA creates an oxidative burden through its metabolites (OXLAMs, 4-HNE). And cardiolipin is precisely the structure those metabolites damage. The inner mitochondrial membrane is lined with LA in cardiolipin, sitting at the site of maximal ROS generation. More dietary LA means more LA incorporated into cardiolipin, means more 4-HNE precursor at the exact location where it can inactivate PDH and KGDH (the two enzymes Humphries and Szweda showed are selectively adducted by 4-HNE). The pathway runs: excess dietary LA → elevated oxLAMs and 4-HNE → cardiolipin oxidation at the IMM → disruption of CL’s structural and signaling roles → ETC dysfunction → impaired oxidative phosphorylation → the Warburg shift that Seyfried identifies as the cancer cell’s defining metabolic state.
The body needs some LA for normal CL synthesis — estimates suggest 2–4% of calories provides adequate LA for all physiological purposes including CL maintenance. The modern seed-oil exposure (7–10% of calories, a several-fold increase from pre-20th-century baselines) represents an excess that overloads the very membrane it supplies, creating the oxidative conditions under which CL is damaged rather than maintained. This is not a contradiction: it is the specific molecular mechanism linking dietary fat quality to mitochondrial dysfunction and, by Seyfried’s argument, to cancer risk.
An important and somewhat counterintuitive observation is that some cancer cells consume oxygen without producing ATP — their mitochondria are uncoupled. This is distinct from the cold-induced uncoupling mediated by uncoupling protein 1 (UCP1) in brown adipose tissue. In cancer, the uncoupling appears to reflect structural membrane damage (altered CL, altered IMM lipid composition) that allows protons to leak back across the inner membrane without passing through ATP synthase, dissipating the proton gradient as heat rather than ATP.
The clinical implication is a measurable correlate of malignancy: more malignant tumors produce more heat per unit mass, reflecting greater uncoupling. This heat signature — long observed empirically — is mechanistically explained by the degree of IMM disruption.10 It also creates a perverse dynamic: the cancer cell is consuming oxygen (so it does not appear anoxic), but generating little ATP from that oxygen consumption, which forces compensatory glycolysis and glutamine fermentation to maintain energy supply.
Otto Warburg identified in the 1920s that cancer cells produce lactic acid even in the presence of oxygen — the phenomenon now called aerobic glycolysis or the Warburg effect. From a pure ATP-yield perspective this seems irrational: fermentation produces ~2 ATP per glucose, while full oxidative phosphorylation produces ~30. Why would a highly proliferative cell use an energy pathway 15 times less efficient?
The answer is that a cell with damaged mitochondria cannot use oxidative phosphorylation efficiently, regardless of substrate or oxygen availability. Glycolysis becomes the default not by choice but by necessity. Additionally, glycolysis is not only an ATP source — it supplies carbon skeletons for biosynthesis (ribose for nucleotides via the pentose phosphate pathway, pyruvate for amino acids, glycolytic intermediates for lipid synthesis) that a rapidly dividing cell requires. From this perspective, the Warburg shift is metabolically coherent for a cell that can no longer run oxidative phosphorylation.
Lactate dehydrogenase isn't the only enzyme cancer cells lean on to regenerate NAD⁺ from NADH under fermentation-heavy metabolism. NQO1 (NAD(P)H quinone oxidoreductase 1) does the same job by reducing a quinone substrate instead of pyruvate, and is characteristically overexpressed in tumors for that reason. The same enzyme is why several bioreductive chemotherapy drugs are built as quinones — they get converted into DNA-alkylating or redox-cycling species specifically inside the NQO1-high cells that most need the salvage pathway, turning a survival adaptation into a selective liability. See the Black Seed supplement page for the enzyme mechanism in the other direction — the same NQO1 chemistry recycling NAD⁺ and quinones in normal cells at physiological turnover.
Glucose alone cannot sustain cancer cell proliferation. Most cancer cells are also addicted to glutamine — the most abundant amino acid in plasma — which they consume at rates far exceeding their protein synthesis needs. Glutamine fermentation in cancer proceeds through a specific pathway first characterized in detail by Moreadith and Lehninger:11
Tomitsuka and colleagues provided direct evidence that cancer cells have an active fumarate reductase reaction — running the succinate-to-fumarate step in reverse compared to the normal TCA cycle direction.12 This reductive TCA activity is consistent with a partially dysfunctional ETC that cannot fully oxidize substrates in the forward direction, but can still extract energy from partial reductive cycling. It also generates succinate, which has signaling properties (activating HIF-1α, suppressing prolyl hydroxylases) that promote the angiogenic and inflammatory phenotype of tumors.
A critical prediction of the metabolic model is that cancer cells, despite their altered metabolism, maintain intracellular ATP levels within a viable range — neither so low as to trigger cell death through ATP depletion, nor so high as to produce Donnan equilibrium failure. This ATP homeostasis prediction has been confirmed: when tumor cells (particularly ascites tumor cells) were compared with normal liver, kidney, and embryo cells, their total ATP status was approximately equivalent, despite the dramatically different fuel sources and pathway utilization.13
This finding is important for two reasons. First, it demonstrates that cancer cells are not simply energy-depleted — they are energy-adequate through fermentation, which sustains their proliferative capacity. Second, it implies that the cancer cell’s metabolic reprogramming is a successful adaptive response to mitochondrial dysfunction, not a failure state. The cell is doing what it needs to do to survive under impaired conditions.
The corollary is that as mitochondria become progressively more dysfunctional, the cell increasingly diverts ATP away from “expensive” cellular maintenance processes — including DNA repair, transcriptional fidelity, and the energy-intensive processes of chromatin remodeling — and toward the minimal requirements for survival and replication. This is the mechanistic basis for the mutation accumulation argument developed in Part V.
The genome is not passively stable. Maintaining it requires continuous, ATP-demanding work: base-excision repair, nucleotide-excision repair, mismatch repair, double-strand break repair, and the proofreading activity of DNA polymerases all consume energy. Chromatin compaction and decompaction for transcription factor access, histone modification, and the spindle checkpoint that ensures accurate chromosome segregation are similarly expensive. Under normal conditions, cells allocate substantial ATP toward these fidelity-maintaining processes because the cost of not doing so — mutation accumulation — is evolutionarily catastrophic.
A cell in chronic energy deficit, compensating for mitochondrial dysfunction through glycolysis and glutamine fermentation, operates under a different ATP budget. Fermentation provides enough ATP to survive and replicate, but the total yield is dramatically lower than oxidative phosphorylation would provide. Under this constraint, ATP is rationed to the processes most essential for immediate survival, and genomic maintenance is a long-term investment that a stressed cell will deprioritize.14
This model explains something that the purely genetic model struggles with: why so many different cancers end up with mutations in the same pathways. If the initial event were a random mutation in any of hundreds of possible driver genes, the subsequent evolution of the cancer would depend on the specific mutation — and we would expect the signaling pathway profiles to be diverse. Instead, they converge on RTK/RAS/PI3K (which drives glucose uptake and glycolysis), p53 (which would eliminate energetically compromised cells), and RB (which would restrain replication in energetically stressed cells). These are exactly the mutations that a glycolytic, energy-deprived cell would be selected for.
The conventional view treats oncogene upregulation as the disease: a mutated RAS or amplified Myc drives proliferation pathologically. The metabolic model offers an alternative reading: oncogene upregulation is a compensatory response by a cell that needs to increase glycolytic capacity to compensate for failing oxidative phosphorylation.
Myc upregulation increases glucose transporter expression, glycolytic enzyme expression, and glutamine uptake — all exactly what a cell with impaired mitochondria needs to maintain ATP supply. HIF-1α upregulation (normally induced by hypoxia) achieves the same shift toward glycolysis and is constitutively active in many cancers even at normal oxygen tensions — a finding that makes sense if the cell’s mitochondria are functionally hypoxic regardless of ambient oxygen. RAS activation promotes glucose uptake through PI3K/AKT/mTOR signaling, also increasing glycolytic flux.5
Under this interpretation, the oncogenes are not the disease itself but the cell’s (maladaptive) solution to the disease. This has a therapeutic implication: inhibiting an oncogene in a cell that depends on it for ATP supply may not cure the cancer so much as create selection pressure for cells that find alternative routes to the same metabolic end.
The argument that mutations are downstream of metabolic dysfunction requires accepting a specific causal chain that is difficult to establish experimentally — partly because mitochondrial dysfunction and mutation accumulation are likely to be mutually reinforcing rather than purely sequential. Some oncogenic mutations (Myc amplification, certain Ras mutations, IDH1/2 mutations) demonstrably alter mitochondrial function — they are upstream of the metabolic shift in those specific contexts, not downstream of it. IDH1/2 mutations, for example, produce the oncometabolite 2-hydroxyglutarate, which inhibits α-ketoglutarate-dependent dioxygenases and directly alters the mitochondrial metabolic environment. In these cases, the genetic event is clearly driving the metabolic change. The metabolic model does not need to claim this never happens; it claims only that it is not the dominant pattern across most cancers.
The xenograft model — implanting human cancer cells into immunodeficient mice and testing whether treatments shrink the resulting tumors — remains the primary preclinical screening tool for oncology drugs. Its failure rate at predicting human clinical outcomes is high by any measure: estimates of the concordance between xenograft results and Phase II/III trial outcomes in solid tumors range from roughly 5–8%.15
Several specific problems arise from a metabolic perspective. Human and mouse cells have fundamentally different metabolic rates (mice have ~7x the mass-specific metabolic rate of humans), different glucose regulatory dynamics, and different telomere biology. Implanting human cells into this environment alters their metabolic phenotype in ways that are difficult to characterize and unlikely to match the original tumor. The immunodeficient host eliminates the tumor microenvironment's immune component — precisely the component that checkpoint immunotherapy, currently one of the most effective new treatments, depends on. And the implanted tumor grows from a single cell line or biopsy, rather than through the spontaneous process of malignant progression in an intact organism.
Tumors arising spontaneously in non-human animals — particularly domestic dogs, which share much of the human environmental exposome, have a similar body size and lifespan relative to mice, and develop epithelial cancers at meaningful rates — provide a more faithful model system. Spontaneous canine osteosarcoma, lymphoma, and mammary tumors have been studied as comparative oncology models and show similar genomic and metabolic features to their human counterparts. Dogs receive conventional veterinary oncology treatment, allowing parallel assessment of treatment efficacy in a species whose tumors developed naturally in an intact immune and metabolic environment.16
The metabolic model specifically predicts that interventions targeting the fermentable fuel supply — glucose and glutamine restriction, ketogenic diets, calorie restriction — should work across species if the mechanism is fundamental. This testable prediction in spontaneous animal tumors is underexplored relative to its potential.
The dominant modalities of cancer treatment — cytotoxic chemotherapy, radiation, targeted therapy — share a conceptual approach: identify something that cancer cells need (DNA replication, a specific signaling protein, rapid division) and interfere with it. The problem is that cancer cells, operating under high mutational pressure and metabolic plasticity, are extraordinarily good at finding alternative routes. Targeted therapies face near-universal acquired resistance; even imatinib, one of the most successful targeted therapies, requires indefinite administration and produces resistance in aggressive disease.
Metabolic therapies aim at something more fundamental: the fuels themselves. Cancer cells cannot easily evolve around the absence of glucose and glutamine, because these are not pathway components that can be re-routed — they are the carbon and nitrogen sources without which the cell simply cannot build the biomass required for replication. This is why the metabolic approach has a theoretical advantage over targeted therapies in terms of resistance potential.
Calorie restriction (CR) — reducing total energy intake without malnutrition — is the most consistently effective anti-cancer intervention in animal models. It extends lifespan, reduces spontaneous tumor incidence, and slows tumor growth across dozens of cancer types and animal species.17 The mechanism is metabolic: CR lowers blood glucose, reduces insulin and IGF-1 signaling, and switches the organism from a glucose-dependent to a fat-dependent metabolic mode — exactly the fuel shift that cancer cells, dependent on fermentable substrates, cannot easily follow.
Several drugs effectively mimic aspects of calorie restriction and have shown preclinical anti-cancer activity through this mechanism:
Calorie restriction as an adjunct to cancer treatment is not validated in randomized trials. The animal data, while extensive, involves rodents whose baseline metabolic rate and carcinogenesis biology differ from humans. The ketogenic diet as a cancer therapy has case reports and small trials showing safety and feasibility, but no large randomized controlled trials demonstrating survival benefit. This is partly a funding problem (dietary interventions are not patentable) and partly a genuine evidence gap. The mechanistic rationale is strong; the clinical evidence is preliminary. Describing CR-mimetics as “a better half-ass treatment” — while colorful — understates both their mechanistic grounding and their potential: metabolic restriction is not an alternative to conventional treatment but a plausible, underinvestigated adjunct to it.
If the metabolic model is substantially correct, it suggests a reorientation of treatment strategy rather than a replacement of existing tools:
A third metabolic lever, largely outside the Seyfried framework but consistent with it, concerns iron. Cancer cells are characteristically iron-avid: they upregulate transferrin receptor 1 (TfR1), reduce ferritin expression, and create a net iron import phenotype that exceeds their mitochondrial needs. The reason is that iron is essential for ribonucleotide reductase (RNR) — the rate-limiting enzyme that converts ribonucleotides to the deoxyribonucleotides required for DNA replication. Rapidly proliferating cells with high RNR demand import iron aggressively and are more exposed to its depletion.
The Zacharski FeAST trial (VA Cooperative Study 410) randomized 1,277 patients with peripheral arterial disease to phlebotomy-induced iron reduction (targeting serum ferritin ~25 ng/mL) or control.18 Cancer was a secondary endpoint. The results were striking: iron reduction was associated with a 35% reduction in new cancer diagnoses (hazard ratio ≈ 0.65, 95% CI 0.43–0.97) and a >60% reduction in cancer-specific mortality (HR ≈ 0.39, P = .003). These are effect sizes larger than most oncology interventions achieve in randomized trials against established disease.
There is also a direct connection to the mitochondrial dysfunction model developed in Part III: cancer cells with impaired ETC already generate elevated H⊂2;O⊂2;. Elevated iron converts this hydrogen peroxide into the hydroxyl radical (the most reactive oxygen species) via Fenton chemistry — which then oxidizes cardiolipin, further impairing the ETC. This creates a self-amplifying loop: mitochondrial damage raises ROS, excess iron converts that ROS into radical species, which damage cardiolipin and deepen the ETC defect. Reducing iron load interrupts this cycle at the catalyst step.
Strongfor the mechanistic rationale (TfR1 upregulation, RNR iron dependence, Fenton chemistry at damaged ETC) and the randomized trial signal. Mixedfor the cancer endpoint specifically — it was a secondary analysis in a PAD population; replication in a cancer-prevention design is needed before this can be treated as established.
Scope
Where this paper agrees with mainstream oncology