Deoxysphingolipids

During early type 2 diabetes, C18 deoxysphingolipids accumulate in the sensory-neuron unit before terminal loss. Selectively reducing that family to its nondiabetic exposure range, starting before injury, would avert at least 20% of five-year net peptidergic skin-terminal loss in a baseline-describable group accounting for at least 10% of reference-care incident DPN cases.

The population comprises adults diagnosed with type 2 diabetes within five years, without clinical DPN. The connected unit includes a peptidergic neuron's dorsal-root-ganglion soma, peripheral axon and skin endings, the satellite glia surrounding that soma, and the nonmyelinating Schwann cells supporting that axon. An effect confined to myelinated fibers, vessels, brain or retina does not establish this claim. Inherited sensory neuropathy establishes human biological capability; it does not establish common diabetic exposure or coverage.

The molecular family is 1-deoxysphinganine and 1-deoxysphingosine with a C18 backbone, and their N-acyl derivatives. For operational measurement, include acyl chains C12–C26 with zero to six double bonds, resolving chain length, double-bond position and hydroxylation rather than combining isomers. These are alanine-derived, C1-hydroxyl-deficient sphingolipids. Glycine-derived deoxymethyl species, canonical ceramides and downstream oxidized clearance products are outside this proposition. In shorthand, m18:0/24:1 has a saturated, singly hydroxylated C18 base and a monounsaturated C24 acyl chain. It is a deoxydihydroceramide despite its acyl-chain double bond.

The intervention concerns this defined family together, not whichever species happens to correlate best with injury. Species can contribute unequally, interact or buffer one another. A negative result for one C24 species does not resolve the family claim. Conversely, an effect of removing the family does not show that every member is toxic.

Allowed baseline groups are the whole eligible population, fasting triglycerides ≥150 mg/dL, HbA1c ≥7.5%, fasting plasma serine below the matched nondiabetic 5th percentile, or plasma summed C18 deoxysphingolipids above the matched nondiabetic 95th percentile. Local exposure above that percentile may also define a group if measured at baseline. Groups cannot be selected by later denervation or response. The same group and connected cells must satisfy all four requirements.

R1 defines persistent local accumulation: at least one specified species has a group-mean local exposure above the age-, sex- and BMI-matched nondiabetic 95th percentile for at least 90 cumulative days in year one, while the implicated structures are intact before losses attributed to the route. Include every baseline participant and early loss, rather than recruiting only later survivors. Measure membrane-associated species as mol per mol membrane lipid in identified cellular and organelle membranes, and freely available aqueous bases as mol/L. Blood hydrolysis totals are neither measurement. The timing and percentile are explicit definitions of persistent abnormal accumulation, not known toxicity thresholds.

For each person, five-year net loss L is terminal disappearances minus newly established terminals, divided by baseline terminal count. A replacement that subsequently disappears contributes to both counts. Follow actual structures with stable peptidergic identity, not CGRP staining alone. 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; the full materiality threshold is B≥0.20. Coverage C is the qualifying group's share of 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.

Source assessment

Best judgment: about 20%. Uncertainty range (sensitivity): 1–66%. The conditional product is 19.2%. The main uncertainty is whether common early type 2 diabetes produces a local lipid exposure whose selective removal preserves human peptidergic terminals. These are subjective probabilities of the complete causal proposition, not treatment effects or confidence intervals. Reviewed 14 September 2026. Probability method · Collection index.

Biological logic and the selective comparison

Serine palmitoyltransferase normally joins serine to an acyl substrate. Using alanine generates the deoxy backbone. Reduced intracellular serine relative to alanine can promote this reaction, but plasma serine is not the enzyme's substrate environment. Increased production, import, altered acylation and impaired clearance are alternative ways to increase local exposure. Persistent accumulation does not require production to exceed disposal forever: a higher steady concentration can have balanced fluxes.

These molecules cannot enter the ordinary sphingosine-1-phosphate degradation route, but they are not chemically indestructible. Oxidative clearance exists. N-acylation, redistribution and desaturation change both location and toxicity. A useful balance therefore follows species through production, transport, interconversion and clearance. Alecu et al., 2017.

Define Q as an ideal adult-onset, molecularly selective export-and-capture intervention in the connected neuronal and glial unit. At each site it lowers each family member only when its exposure exceeds the corresponding matched nondiabetic 95th-percentile trajectory, capping it at that upper-normal boundary; lower exposures remain unchanged. Removed material enters an inert external reservoir with its mass accounted for. The same rule operates throughout five years. Molecular identity and exposure determine removal, independently of structural outcome. This is an idealization of selective lipid handling, not an assertion that serine, myriocin or an ELOVL1 inhibitor already implements it.

Q has no direct effect on canonical sphingolipid synthesis, serine availability, insulin signaling, vascular supply or other lipid traffic. Other initiating metabolic exposures receive identical reference care. Changes caused by removing the deoxy species—including restored canonical-lipid metabolism, membrane remodeling, calcium, organelle function, glial support and replacement—may occur. Preserving canonical concentrations by an additional clamp would incorrectly remove possible downstream mediation. Independent molecular capture and restoration of the original local deoxy trajectory distinguish the intended effect from transport or carrier artifacts.

The claim does not require a particular mitochondrial pore, endoplasmic-reticulum lesion, apoptosis program or clearance enzyme. Those are alternative mediators. Nor is serine deficiency compulsory: handling or import can create the same exposure. The priced structural effect is the net effect of Q, including any benefit lost by reducing a protective family member.

Conditional probabilities

Every probability after R1 assumes all preceding claims are true in the same permitted group. Retain only groups satisfying all earlier claims. The conjunction expresses local accumulation, a causal structural contribution, its magnitude and its population coverage.

See the claims and probabilities table above.

R1 is more likely than not because diabetic lipid measurements, nutrient perturbation and human serine genetics converge on an exposure route. Its range remains wide because no study has measured the specified early human cellular exposure. R2 gives substantial conditional weight to inherited human neuropathy, sensory-neuron perturbations and structural animal rescue; its lower bound allows serine and myriocin benefits to arise mainly through other molecules. It does not charge the absence of local accumulation twice. R3 remains uncertain because most effects concern sensation, myelinated axons or repair, and the available structural studies do not estimate selective five-year preservation. R4 allows a common metabolic subgroup but discounts unknown case coverage and the distinction between obesity, diabetes and neuropathy.

The best inputs [65,70,65,65]% give 0.65×0.70×0.65×0.65=0.1922375. About 20% is the supported precision; a distinction between 19% and 22% would not be defensible.

Evidence for and against the proposition

Human structural associations are suggestive but cross-sectional. Fridman and colleagues studied 75 people: 19 lean, 19 obese without diabetes, 18 obese with type 2 diabetes without clinical neuropathy, and 19 with diabetic neuropathy. Plasma summed deoxydihydroceramides were 89.4 nM in DPN versus 52.0 nM in lean controls, but the two diabetic groups did not differ (p=0.8127). Within the 56 obese participants, including both diabetic groups, the lipid–distal-fiber correlation was r=−0.40, approximately 95% CI −0.60 to −0.15 with 55 available biopsies. The triglyceride-adjusted correlation was −0.35, p=0.010. This continuous association survives removing lean controls and deserves weight despite the binary null. It remains unadjusted for several relevant differences, including age, and does not identify local exposure or direction. Mean distal fiber density was already low in the clinically neuropathy-free diabetic group, 5.6 versus 3.0 fibers/mm in DPN. Fridman et al., 2021.

That study detected N-acyl deoxydihydroceramides rather than free deoxy bases. Its 89.4 nM plasma total cannot be equated to 89.4 nM free base applied to cultured neurons. A separate study of 75 people with idiopathic small-fiber neuropathy found higher blood deoxysphingolipids in those below the cohort's median fiber density, 0.36 versus 0.29 µM. Diabetes was excluded, and groups were defined by the structural outcome. The whole-cohort regression retained deoxysphingolipids and sex, with adjusted R²=0.22. This supports a broader human association, not a prospective diabetic subgroup or a causal intervention. Kreß et al., 2025.

A small human tissue comparison is particularly relevant to the compartment question. Dohrn and colleagues found elevated plasma deoxy bases in 39 people with diabetic neuropathy, without correlation to clinical severity. Sural-nerve sphingoid-base profiles did not differ between six diabetic-neuropathy and three inflammatory-neuropathy biopsies. This is a small, established-disease comparison against another neuropathy, measured after collapsing N-acyl species into bases. It neither demonstrates normal diabetic nerves nor excludes a focal neuronal species excess. The full methods and numerical tissue estimates were not accessible, limiting a stronger interpretation. Dohrn et al., 2015.

Human genetic and prospective serine evidence strengthens an upstream route, without identifying this mediator. A 2026 preprint found PHGDH and PSPH signals and Mendelian-randomization estimates of 0.85–0.93 odds of neuropathy per genetically predicted SD higher circulating serine across six exposure datasets. PHGDH colocalization probability was 0.92. The outcome was broadly coded neuropathy among people with type 2 diabetes, not incident molecularly tracked C-fiber loss. Serine also supplies canonical sphingolipids, phosphatidylserine and other metabolism; genetic association and colocalization do not establish mediation by deoxy species. Fridman et al., 2026 preprint.

In the Diabetes Prevention Program Outcomes Study, 1,947 participants had baseline metabolomics and later complication assessment. Higher baseline serine was associated with less neuropathy around 15 years later, OR 0.68 (95% CI 0.56–0.84). However, the neuropathy endpoint was abnormal 10-g monofilament detection, and 42% of affected participants had never developed diabetes. This is useful temporal evidence for serine-related susceptibility, not direct prevention evidence for the present molecular and fiber-specific proposition. Perng et al., 2025.

Inherited disease supplies a qualified human intervention anchor. In the randomized HSAN1 trial, nine participants received 400 mg/kg/day serine and nine placebo for one year. Average disease duration was about 25 years. The binary progression comparison was negative: 1/9 versus 2/9 progressed by more than one CMTNS point, Fisher p=1.0. The continuous clinical-score contrast favored serine by −1.5 points (95% CI −2.8 to −0.1). Distal skin innervation increased by a median 8 versus 0 fibers/mm² of skin surface, p=0.014; baseline distal biopsies were mostly denervated. That is a repair signal in established inherited disease. Thigh reinnervation and several functional measures did not favor serine convincingly, and secondary comparisons were not multiplicity-adjusted. Fridman et al., 2019.

Serine reduced hydrolyzed plasma deoxysphinganine by 60%, versus a 9% increase with placebo, but it also changes precursor supply for canonical lipids. The trial is not selective deoxy removal. Its registry reports week-48 skin-density levels, which must not be mistaken for changes. No randomized DPN serine trial with structural results was identified in the literature and registry search. An additional registered inherited-neuropathy trial has no posted results; that absence is not a negative treatment result. NCT06113055.

Patient-derived sensory neurons provide complementary evidence: three HSAN1 donors were compared with four control lines derived from three people, not four independent controls. Late axon/myelin deterioration and impaired outgrowth were serine-responsive; 10 mM serine increased replated neurite growth about 2.5-fold. It simultaneously restored gangliosides. These were nonisogenic lines, differentiated repeatedly and cocultured with rat Schwann cells. The model demonstrates inherited human neuronal capability and an alternative canonical-lipid mediator, not selective diabetic peptidergic rescue. Clark et al., 2021.

Diabetic animal experiments separate functional benefit from structural preservation. Othman and colleagues used both immediate and delayed dietary serine in streptozotocin-diabetic rats. Preventive treatment increased sensory conduction from 23.8 to 30.3 m/s and partly restored myelinated-axon diameter distributions. Glucose and triglycerides remained elevated. Crucially, skin fiber density was not reduced by diabetes in this experiment, so that endpoint could not demonstrate prevention of diabetic terminal loss. Thermal responses and soma-size abnormalities were not rescued. Othman et al., 2015.

Handzlik and colleagues found that serine/glycine withdrawal plus high-fat feeding accelerated sensory deficits. Myriocin preserved epidermal and corneal innervation after six months; the skin analysis included 10 low-fat, 7 high-fat, 12 restricted-high-fat and 8 restricted-high-fat-plus-myriocin mice. It also changed canonical sphingolipids, other hepatic lipids and weight. Whole-sciatic-nerve deoxy levels did not track the neuropathy phenotype. In a separate db/db experiment, myriocin increased epidermal fiber density in 10 versus 9 mice despite limited effects on paw-skin deoxylipids. These are structural positives for a broad metabolic intervention whose mediator remains unresolved. Handzlik et al., 2023.

The same paper's serine-only db/db arm measured sensation and lipids, not skin fiber density. Paw summed deoxydihydroceramides fell to 0.525 of control and m18:0/24:1 to 0.592, with eight mice per lipid group. Functional improvement cannot silently supply the missing structural endpoint. In another db/db study, three months of serine shifted small unmyelinated-axon diameter distributions without changing average axon number; after six months, ultrastructural degeneration remained despite biochemical and sensory improvement. The late images are not a precise equivalence test excluding 20% preservation. Xia et al., 2023.

Species composition matters, but a single toxin has not been established for human diabetes. Majcher and colleagues found that ELOVL1/CERS2 manipulations reduced very-long-chain deoxy species and rescued cell injury. In embryonic chicken DRG cultures, 50–100 nM added free deoxysphinganine reduced neurites, with rescue by an ELOVL1 inhibitor. This is a lower-dose neuronal result than many micromolar culture studies, but donor-level replication, protein binding and local membrane exposure still limit human transfer. The intervention also changes canonical acyl-chain metabolism. Majcher et al., 2025.

Their source tables show ELOVL1 knockdown reducing m18:0/24:1 to 0.416 of control while summed saturated-base deoxydihydroceramides rose to 1.083, alongside an 83-percentage-point viability rescue. Thus total pool reduction is not necessary for protection in that preparation. Conversely, fatty-acid addbacks can restore toxicity without restoring that exact C24:1 species. A later SH-SY5Y study independently implicated very-long-chain acyl synthesis using 1 µM deoxysphinganine and TECR/CERS2 perturbations. It also documented altered canonical ceramides and slower basal growth in the modified lines. These studies support composition-dependent cellular toxicity, not a uniquely identified common-DPN toxin. Byrnes et al., 2026.

Ideal experiments that would resolve each claim

These are definitions of decisive comparisons under unlimited measurement and selective-intervention capability. They are not proposed studies. Use representative independent humans from the stated early-diabetes population, with stable terminal-subtype tracking and matched nondiabetic exposure measurements. Assign Q or reference care at baseline and continue for five years. Verify molecular selectivity, local target engagement, reservoir mass balance and absence of direct changes to other initiating pathways. Restoration of the original deoxy exposure and an independent capture method distinguish the lipid effect from intervention artifacts. Downstream changes remain allowed as specified above. Measure all permitted groups without defining one by its treatment response.

The following decision rules apply to each candidate group. A row is YES if at least one group retained from its predecessors satisfies it, and NO only if all retained groups fail it. For finite data, resolve a threshold claim only when an appropriate simultaneous uncertainty interval for the relevant group comparisons lies wholly on its stated side. An interval crossing the boundary, failed engagement, loss of subtype identity or missing early losses is unresolved, not NO. Exact equality belongs to the stated ≥ side, although finite noisy data at equality may remain unresolved. A nonpositive L0 makes relative preservation undefined and does not establish a qualifying loss-prevention effect.

R1 — local accumulation, no predecessors. Repeatedly measure the specified intact species in identified neuronal and glial compartments from baseline through year one, mapping them to the terminals followed. Establish species-specific membrane mole fractions or free-base concentrations, and their matched nondiabetic distributions, without relying on hydrolysis totals. Track exposure before attributed structural loss and include early losses in the original cohort. YES requires at least one permitted group/species/connected-compartment combination to meet the >95th-percentile boundary for ≥90 cumulative days. NO means no permitted combination meets it, with adequate temporal and spatial coverage; for example, a clear elevation lasting only 30 days fails this persistent-accumulation definition. Borderline duration or concentration remains unresolved. This is a descriptive exposure comparison; it does not establish that diabetes caused the elevation or that the lipid causes injury. Human plasma and animal whole-tissue results only partially approximate it.

R2 — a structural contribution, assuming R1. In the same R1-qualified group, compare Q with reference care from baseline for five years. Measure actual disappearances and successful replacements per person's baseline peptidergic terminal count, then estimate B using group means. Verify that Q achieves its molecular exposure target in the implicated cells throughout the risk period, with all off-target and restoration conditions above. YES is B≥0.05; NO is B<0.05. For example, L0=0.30 and L1=0.28 gives B=0.067 and satisfies this row. A sufficiently precise upper bound below 0.05 establishes NO even if there is a smaller effect. Pain relief, better conduction, soma survival or altered CGRP expression without structural preservation leaves this row unresolved. The HSAN1 trial and myriocin experiments approximate parts of this comparison but do not implement Q in the target population.

R3 — material magnitude, assuming R1–R2. Use the same randomized Q comparison, qualifying group and structural estimand; do not switch to a more responsive post-treatment subset. YES is B≥0.20, and NO is 0.05≤B<0.20 under the assumed predecessors. At L0=0.30, L1=0.24 reaches the boundary, whereas L1=0.28 satisfies R2 but fails R3. Precision must distinguish 20% relative preservation, not merely detect a treatment difference. An upper bound below 0.20 resolves NO for this row; an interval spanning 0.20 remains unresolved. Reinnervation from a near-zero baseline in established HSAN1 and qualitative mouse ultrastructure do not estimate this magnitude.

R4 — coverage, assuming R1–R3. In a population-representative reference-care cohort with the same baseline eligibility, measure all allowed group markers and ascertain five-year incident DPN consistently. For groups independently satisfying R1–R3, estimate C=P(group membership | reference-care incident DPN). YES requires at least one such group with C≥0.10; NO requires all such groups to have C<0.10 with adequate precision. A rare group containing 3% of incident cases fails even with large structural rescue. The share among all diabetic adults, among the recruited trial participants, or among remaining cases after Q is the wrong denominator. If several groups qualify, retain their individual definitions rather than constructing an outcome-selected union. Current inherited-disease and cross-sectional diabetic samples do not resolve this quantity.

Sensitivity, possible updates and boundaries

The skeptical joint interpretation uses [35,30,30,30]% and gives 0.945%: blood changes rarely identify the specified persistent local accumulation; serine's structural effects mostly arise through other substrates; any deoxy effect is small or uncommon. The favorable interpretation uses [90,90,90,90]% and gives 65.61%: local exposure substantially overlaps sensory-neuron toxicity, the animal structural effects are predominantly deoxy-mediated, and a common baseline metabolic group shares that route. These assumptions can move together. The 1–66% range is sensitivity across these defensible interpretations, about 65 percentage points wide, not a calibrated distribution or confidence interval.

Two asymmetric alternatives make the source of uncertainty clearer. If serine biology is real but mainly mediated by canonical-lipid restoration independent of deoxy removal, [65,30,50,60]% gives 5.85%. If deoxy toxicity is substantial but concentrated in a rare group, [65,85,80,25]% gives 11.05%. Neither biochemical plausibility nor large selected-sample rescue establishes common case coverage.

Hypothetically, evidence of selective human structural preservation at the measured local exposure could move R2 from 70% to 90%, raising the whole probability to 24.7% with other judgments fixed. Effective local removal yielding B<0.05 could move it to 30%, lowering the probability to 8.2%. Evidence also resolving magnitude or prevalence should update those rows together. Merging R1–R2 into their 45.5% joint probability leaves the headline unchanged. Additional obligatory discounts for a mitochondrial lesion or serine limitation would price a narrower proposition.

The species catalog, persistence definition, allowed baseline groups and 5%/20%/10% thresholds affect scope. Relaxing an outcome or coverage criterion can enlarge the truth set with Q fixed, but current data do not quantify a new probability. Changing the exposure cap also changes Q and requires a fresh assessment. Missing selective human structural evidence is the principal limitation; another blood association does not close it. The claim overlaps canonical ceramide injury, Schwann support, ER stress and protective lipid deficiency. Shared serine or myriocin results cannot count as independent confirmations of those mechanisms.