Canonical ceramides

In adults diagnosed with type 2 diabetes within five years and without clinical DPN, excess canonical ceramides accumulate in peptidergic sensory neurons or their associated nonmyelinating Schwann cells before terminal loss. At least one of the chemically defined exposure corrections below, begun before injury, would reduce five-year net peptidergic skin-terminal loss by at least 20% in a baseline-describable group accounting for at least 10% of reference-care incident DPN cases.

This is a local excess-ceramide claim. It does not claim that every ceramide is harmful, that high plasma total ceramide measures nerve exposure, or that inhibiting all sphingolipid synthesis is beneficial. Canonical ceramides can be needed for normal membrane maintenance, while a particular species at a particular site can impair signaling or membrane function.

The molecular catalog comprises ceramides with a normal sphingosine backbone, d18:1 with its 4E double bond, and a nonhydroxylated C14, C16, C18, C20, C22, C24 or C26 N-acyl chain that is saturated or monounsaturated. Resolve acyl double-bond position and geometry when identifying a species. Deoxyceramides, dihydroceramide precursors, glycosylceramides, sphingomyelins and synthetic short-chain C2/C6 analogues are outside this catalog. Their measurements can inform metabolism without substituting for the specified exposure.

To make the possible composition effects bounded, consider these candidate corrections: lower one chemically resolved catalog species, all catalog species with C14–C18 acyl chains, all with C20–C26 chains, or the whole catalog. These are alternatives, not additional requirements or independent probabilities. A correction is fixed by molecular identity before its outcome is assessed. The claim does not permit an outcome-fitted weighted score or an arbitrary mixture selected from responders.

For each candidate, the connected compartment is either the peptidergic neuron from its dorsal-root-ganglion soma to its skin terminals, or the Remak Schwann cells supporting that same peripheral axon. An effect confined to unrelated myelinating Schwann cells, retinal vessels, liver or brain does not satisfy the proposition. The same molecular correction, compartment and baseline group must satisfy every row.

Allowed groups are the whole eligible population, fasting triglycerides ≥150 mg/dL, HbA1c ≥7.5%, or a baseline local exposure to a catalog species above the matched nondiabetic 95th percentile. R1 requires a group's mean local exposure to at least one species targeted by its candidate correction to exceed the age-, sex- and BMI-matched nondiabetic 95th percentile for at least 90 cumulative days in year one, before losses attributed to that route. Include all baseline participants and early losses; do not condition inclusion on future survival or denervation. Quantify each species as mol per mol membrane lipid in identified cellular and organelle membranes. Bulk blood concentration, total tissue mass and a ceramide-antibody signal do not resolve this exposure.

For each person, L is terminal disappearances minus newly established terminals over five years, divided by baseline terminal count. A new terminal that later disappears contributes to both counts. Stable peptidergic identity and actual structural tracking distinguish axon disappearance from changing CGRP expression. With positive reference-care mean loss L0 and intervention mean loss L1, relative preservation is B=(L0−L1)/L0. A smaller structural contribution is B≥0.05 and material preservation is B≥0.20. Coverage C is the qualifying group's share of all reference-care incident DPN cases, with C≥0.10. Incident DPN requires new bilateral length-dependent sensory signs with objective small-fiber loss or abnormal nerve conduction, excluding other causes. These thresholds define the proposition; they are not established biological constants.

Source assessment

Best judgment: about 10%. Uncertainty range (sensitivity): 0.4–55%. The conditional product is 10.8%. The main uncertainty is whether a defined local ceramide excess, rather than circulating lipid transport or loss of other sphingolipids, causes material human terminal loss. These are subjective causal probabilities, not effect sizes or confidence intervals. Reviewed 14 September 2026. Probability method · Collection index.

From lipid production to selective structural causation

Ceramides can arise from de novo synthesis, sphingomyelin hydrolysis or salvage of sphingosine. Conversion into complex sphingolipids, hydrolysis by ceramidases, and transport determine their local abundance. A higher pool does not necessarily imply increased production, and sustained accumulation can reach a new steady state with balanced production and removal.

A lipoprotein-borne plasma ceramide is not automatically available at a neuronal mitochondrial or plasma membrane. Conversely, local hydrolysis can generate a transient membrane domain without a high blood concentration. Backbone, acyl chain, membrane leaflet and surrounding lipids can change the effect. Loss of myelin lipid mass can coexist with a small local ceramide excess, but the former does not establish the latter.

For a candidate correction, define Q as an ideal adult-onset molecular export-and-capture intervention restricted to its specified cells. It caps each targeted species' local exposure at the corresponding nondiabetic 95th-percentile trajectory, leaving lower exposures and untargeted species directly unchanged. Removed molecules enter an inert external reservoir with mass accounted for. The cap operates throughout five years. It recognizes chemical identity and exposure, not presumed toxicity.

Q does not directly alter fatty-acid delivery, glucose, vascular supply, serine availability, ceramide synthases, sphingomyelinases or other lipid classes. Reference care and other initiating exposures are comparable. Downstream changes caused by removing the targeted ceramide—including insulin signaling, other sphingolipid pools, membrane remodeling, organelle responses and axonal replacement—may occur. Holding those downstream concentrations fixed would remove possible mediation. The net structural effect includes any loss of useful signaling or membrane support from the correction.

This differs from myriocin, which inhibits the entry to canonical and deoxy sphingolipid synthesis, and from fumonisin B1, which changes ceramide synthesis and its substrates. Increasing ceramidase activity also creates sphingosine and potentially sphingosine-1-phosphate; that is not a chemically isolated ceramide-removal experiment. Independent capture and restoration of the original local ceramide trajectory are needed to distinguish Q from these intervention effects. Apoptosis, mitochondrial fission, ER stress and loss of glial support are possible mediators rather than obligatory extra probabilities.

Conditional probability assessment

Every row after R1 assumes all preceding claims hold for the same permitted candidate correction, cell compartment and baseline group. Retain only combinations satisfying all predecessors. The conjunction includes the exposure, structural causation, material magnitude and case coverage.

See the claims and probabilities table above.

R1 is near even odds because ceramide metabolism can change in diabetes, but human circulating results disagree in direction and the required pre-loss membrane exposure is unmeasured. R2 gives conditional weight to Schwann-cell interventions and broad structural animal rescue. Its range is particularly wide because synthetic analogues and upstream synthesis inhibitors do not identify the catalog species' local structural effect. R3 discounts the unmeasured material magnitude after accepting a smaller effect. R4 gives some weight to common metabolic exposures while acknowledging that the causal compartment and candidate correction could occur in few incident cases.

Best inputs [55,55,55,65]% give 0.55×0.55×0.55×0.65=0.10814375. The evidence does not reliably distinguish a 10% from a 15% judgment. These alternatives are not an intervention ranking.

Evidence that changes the assessment

The most cited human positive is a small concurrent symptom study. Hammad and colleagues selected 80 people with type 1 diabetes, 19 with and 61 without neuropathy symptoms, with generally normal conventional lipid profiles. Neuropathy classification used a questionnaire score or selected sensory responses; the clinical examination was explicitly excluded. At sampling, diabetes duration averaged about 18–19 years. Models adjusted for mean HbA1c, baseline disease cohort and concurrent LDL, HDL and triglycerides. The 19 events supported six predictor coefficients including the lipid marker, about 3.2 events per predictor. This is not an early type 2 incident-DPN comparison. Hammad et al., 2017.

Per SD higher log-transformed plasma concentration, reported odds ratios were 2.44 (95% CI 1.10–5.38) for C22:0, 2.79 (1.18–6.60) for C24:0 and 3.59 (1.34–9.62) for C26:0. C16:0 was 0.78 (0.41–1.48) and C24:1 was 1.67 (0.75–3.74). These findings do not support a universal “short chains harmful, long chains protective” rule in human neuropathy. They also do not establish the reverse rule: there was no direct causal chain-length comparison.

The study reported 78 adjusted marker associations without multiplicity correction. Its three canonical-species p values, 0.028, 0.019 and 0.011, do not survive a Bonferroni threshold of 0.05/78=0.000641. This does not prove zero association; it prevents calling the selected positives independently confirmed. The reported odds ratios are not discrimination statistics. Under an illustrative normal-distribution assumption using the printed group means and SDs, C24:0 and C26:0 would give AUROCs of only 0.65 and 0.69. Those are assumption-based calculations, not observed or validated AUROCs. Individual data are needed for actual discrimination and multivariable checking.

Type 2 plasma evidence points in a different direction. In ADDITION-Denmark, Rumora and colleagues profiled 991 metabolites in 9 lean controls, 49 type 2 participants without DPN and 48 with DPN. Plasma ceramides and sphingomyelins were lower in diabetes, with more pronounced changes in the neuropathy group. Most diabetic participants used statins and had lower total cholesterol than controls. The DPN-specific selected sphingolipid was a hydroxylated glycosylceramide, outside the present ceramide catalog. The reported 0.88 multimetabolite classifier AUROC was not a ceramide effect or a prospective structural prediction. Changes in plasma lipoproteins, treatment and complex lipid synthesis remain competing explanations. Rumora et al., 2021.

A separate study compared 44 severely obese participants with neuropathy and 44 without, matched for glycemic status, plus 43 lean controls. Ceramide and dihydroceramide pathways contributed to plasma separation, but diacylglycerols were the leading lipid pathway. This provides a useful obesity-matched association beyond lean-versus-diabetic comparisons. It does not measure the local molecular target or show that lowering it preserves terminals. The accessible primary text did not supply a validated, standalone ceramide discriminator or selective causal effect. Guo et al., 2022.

Direct human nerve evidence is too small and chemically coarse to settle R1. Dohrn and colleagues compared sphingoid-base profiles from six diabetic-neuropathy and three inflammatory-neuropathy sural biopsies, finding no reported difference. The comparator was CIDP, not healthy nerve; collapsing N-acyl species into bases prevents a canonical-chain-specific test. This cannot establish equivalence or exclude a focal neuronal membrane change. Full numerical tissue results were not accessible. Dohrn et al., 2015.

Schwann-cell perturbation supports a partial lipid-dependent injury mechanism. Suzuki and colleagues exposed immortalized mouse and primary rat Schwann cells to palmitate for 48 hours. Myriocin and fumonisin B1 reduced ceramide accumulation and partially reduced cell death, yet did not suppress palmitate-induced caspase-3 activation. Synthetic C2-ceramide caused death without activating caspase-3. Thus a ceramide-associated, caspase-independent component could coexist with ceramide-independent palmitate injury. The experiment does not make caspase-3 activation a necessary step for this claim. Suzuki et al., 2011.

The full paper could not be accessed through the publisher or an author repository. Independent preparation counts, exact dose–response values and quantitative species measurements therefore remain unavailable in this assessment. The primary abstract supports the qualitative intervention result, not a precise rescue fraction. C2-ceramide's short acyl chain gives it different delivery and membrane behavior from native C14–C26 species, so its toxicity cannot establish the human local dose.

Animal structural rescue exists, but mediator identity is unresolved. Handzlik and colleagues found that myriocin increased epidermal fiber density in db/db mice treated from six weeks of age for eight weeks, with 10 vehicle and 9 treated animals. It did not correct body weight or hyperglycemia, but strongly lowered hepatic canonical sphingolipids and had limited effects on paw-skin ceramides/deoxylipids. In a separate serine/glycine-restricted high-fat experiment, myriocin preserved skin and corneal innervation while changing several lipid classes and weight. Whole-sciatic-nerve ceramide levels did not track the neuropathy phenotype. This is favorable structural evidence for perturbing sphingolipid metabolism, with weak localization and no isolation of canonical ceramides. Handzlik et al., 2023.

Normal ceramide synthesis can protect nerve structure. Whole-body CERS2-deficient mice developed vesiculation and detachment of inner myelin layers in about 20% of peripheral axons, alongside major CNS and systemic abnormalities. This warns against interpreting complete synthesis loss as selective removal of an excess. It concerns broad developmental and adult lipid deficiency, not the upper-normal cap defined here, and primarily myelinated structures. Imgrund et al., 2009.

Likewise, early db/db nerve lipidomics found loss of myelin-enriched galactosylceramides, sulfatides and plasmalogens before advanced injury, with 4–5 mice per genotype in the early comparisons. Those are complex-lipid deficiencies, not direct measurements of the canonical ceramide exposure specified here. A lipidomic observation must retain its exact chemical class. Palavicini et al., 2020.

Ideal experiments that would resolve each claim

These comparisons define what would settle the claims with unlimited measurement and intervention capability; they are not proposed studies. Recruit representative independent humans from the stated early type 2 population, measure baseline group markers, identify the connected neuronal/Schwann compartments and follow stable peptidergic terminals for five years. Compare each chemically fixed candidate Q with reference care, beginning at baseline. Verify local species exposure, capture specificity and mass balance throughout; use independent capture chemistry and restoration of the original exposure to identify intervention artifacts. Other initiating exposures remain comparable and downstream changes may occur as defined above.

A row is YES if at least one candidate/group/compartment combination retained from its predecessors satisfies it. It is NO only if every retained combination fails. Use simultaneous intervals appropriate to the actual candidate family and group comparisons, rather than selecting nominal positives from many tests. A finite interval crossing a boundary remains unresolved. Inadequate engagement, missing early losses, nonpositive reference mean loss or loss of stable terminal identity cannot establish preservation. Equality belongs to the ≥ side; finite noisy data at exact equality may remain unresolved.

R1 — exposure, no predecessors. Measure intact catalog species as local membrane mole fractions, resolving backbone, chain, double-bond identity and cell/organellar location from baseline through year one. Link those structures to the terminals followed. Establish the matched nondiabetic exposure distributions with the same measurement. YES requires the defined >95th-percentile elevation for ≥90 cumulative days before attributed losses in at least one permitted combination. NO requires adequate spatial and temporal coverage showing no combination meets both boundaries; for example, a clear elevation lasting only 20 days fails the persistent-accumulation proposition. Borderline level/duration or bulk-only measurement is unresolved. This is a descriptive diabetic-context exposure claim, not proof that hyperglycemia produced it or that the exposure is injurious. The available human plasma and sural-base studies only partially approximate this measurement.

R2 — structural causation, assuming R1. Compare Q and reference care in the same R1-qualified population and cellular route, starting before the attributed losses. Directly count disappearances and successful replacements, calculate each person's L, and estimate B from group means. YES is B≥0.05; NO is B<0.05 with sufficiently precise measurement across all retained candidates. For example, reducing L from 0.30 to 0.28 gives B=0.067 and meets this row. A narrow upper bound below 0.05 establishes NO even if a smaller effect exists. Changes in caspase activity, pain, conduction, myelin appearance or CGRP expression alone are unresolved for this structural claim. Suzuki's cell deaths and Handzlik's broad synthesis-inhibitor rescue do not identify this exact comparison.

R3 — material magnitude, assuming R1–R2. Use the same candidate, cell compartment, group and randomized structural comparison, without selecting responders. YES is B≥0.20; NO, under the predecessors, is 0.05≤B<0.20. L0=0.30 and L1=0.24 reaches the threshold; L1=0.28 satisfies R2 but fails R3. Precision must distinguish this relative preservation boundary, not merely separate treatment means. An interval spanning 0.20 remains unresolved. None of the cited studies estimates this selective human five-year magnitude.

R4 — population coverage, assuming R1–R3. In a representative reference-care cohort with the same eligibility, ascertain baseline permitted-group membership and five-year incident DPN. For independently qualified groups estimate C=P(group | reference-care incident DPN), retaining the original case definition and denominator. YES requires a qualifying group with C≥0.10; NO requires every qualifying group to have C<0.10 with adequate precision. A group with large preservation but only 2% of reference incident cases fails. Do not substitute its share of recruited participants, all diabetic adults or residual treated cases. Imprecise coverage or a group constructed from treatment response is unresolved. Existing selected symptom cohorts and rare inherited lipid disorders do not supply this measurement.

Sensitivity and what would change the judgment

The coordinated skeptical interpretation uses [25,20,25,30]% and gives 0.375%: most plasma associations reflect systemic transport, local ceramide elevation is uncommon, and broad inhibitor rescue is mediated by other products or too small to meet materiality. The favorable interpretation uses [85,85,85,90]% and gives 55.27%: bulk/plasma discordance masks a common local species excess, Schwann-cell and animal structural effects transfer to human terminals, and a baseline metabolic group captures at least 10% of cases. The row ranges express these interpretation changes. They are not independent distributions, and their approximately 55-percentage-point whole-claim span is not a confidence interval.

Two additional interpretations help separate the uncertainties. Treating the plasma signal as mostly unrelated to local exposure, while retaining the best later conditionals, gives [30,55,55,65]%=5.9%. Accepting common local accumulation and a smaller structural effect, but doubting 20% materiality, gives [75,75,25,65]%=9.1%. Opposite plasma associations do not force biological absence, and cellular toxicity does not establish the magnitude.

Hypothetically moving R2 from 55% to 85% after a selective human structural comparison raises the complete probability to 16.7%, with other judgments fixed. Effective local correction yielding B<0.05 could move R2 to 20%, lowering it to 3.9%. Evidence establishing magnitude or coverage should update those judgments too. Merging R1–R2 into their 30.25% joint probability preserves the headline; adding mandatory apoptosis or mitochondrial-fission rows would change the proposition rather than refine the same arithmetic.

The main unresolved evidence is a matched human local exposure and selective structural contrast. Neither an unadjusted plasma association nor a genetic instrument that mainly changes LDL particle transport identifies it. Molecular catalog and candidate-correction choices define scope. Relaxing structural or coverage thresholds with Q fixed enlarges the truth set; changing the exposure cap changes Q and requires reassessment.

This mechanism can overlap fatty-acid delivery, insulin signaling, Schwann support, ER stress or mitochondrial dynamics. It remains distinct from deoxysphingolipids and protective lipid deficiency. Shared myriocin rescue cannot be counted independently for these alternatives, and their probabilities must not be added.