# 01. Excess polyol flux and impaired sorbitol disposal

**Preventing excess glucose-to-sorbitol metabolism could materially preserve human peptidergic C-fiber terminals: about 20% probability, with a 2–55% scenario-sensitivity range.** The reproducible central product is 21.1%. The main uncertainty is whether the measured human biochemical change causes enough loss in this particular fiber population, before clinical neuropathy. Literature reviewed 14 September 2026.

## The causal proposition

In adults with type 2 diabetes diagnosed within five years, without confirmed distal polyneuropathy and with initially preserved distal-leg skin innervation, increased aldose-reductase conversion of glucose to sorbitol produces a persistent metabolite or cofactor disturbance in sensory axons or their connected support cells. Preventing that excess from the start of observation would avert at least 20% of five-year net peptidergic terminal loss in an independently defined subgroup containing at least 10% of incident DPN cases under reference care.

The reference is the same diabetes care and initiating glucose, lipid and other exposures without pathway manipulation. The ideal intervention clamps the **glucose-reduction reaction** of aldose reductase to its matched nondiabetic rate in the affected nerve unit. It preserves the enzyme's non-glucose aldehyde-processing functions. This is a biochemical idealization, not a claim that an existing inhibitor has that selectivity. Downstream redox state, local perfusion, glial support and axon maintenance may respond.

For an explicit, bounded subgroup question, eligible baseline groups are the whole population, HbA1c ≥7.5%, local nerve-unit sorbitol above the matched nondiabetic 95th percentile, or local sorbitol-dehydrogenase activity below the matched 5th percentile. The biochemical rules refer to axons, Remak Schwann cells surrounding unmyelinated axons, or their local microvascular endothelium; each compartment is evaluated separately. Membership is fixed before intervention, without using later injury or treatment response. At least one **same group and connected compartment** must satisfy the entire chain. A rare inherited SORD neuropathy cannot supply the required case coverage.

Let L0 and L1 be five-year net terminal loss under reference care and the intervention, each expressed as a fraction of baseline terminal number in a fixed distal-skin territory. Loss counts disappearance minus successful replacement, using stable subtype identity rather than reduced peptide staining. Define preservation B = (mean L0 − mean L1)/mean L0 within the subgroup, requiring mean L0 > 0. The complete claim requires B ≥0.20 and reference-case coverage C ≥0.10. Coverage is the fraction of incident cases in that baseline subgroup, not the fraction of residual cases after prevention. Incident DPN requires new bilateral length-dependent sensory signs and objective small-fiber loss or abnormal nerve conduction, excluding an alternate neuropathy. These are chosen definitions, not measured attributable fractions. Type 1 diabetes and established-disease repair supply evidence under different scopes.

## How the mechanism could work

Aldose reductase consumes NADPH while reducing glucose to sorbitol. Sorbitol dehydrogenase, encoded by SORD, consumes NAD+ while converting sorbitol to fructose. A larger sorbitol pool can reflect greater formation, slower disposal or both. It does not measure flux. Disposal can catch up at a higher pool size; a persistent pool does not require formation to exceed disposal indefinitely.

The allowed mediating branches are excess sorbitol/fructose pools and their osmotic or carbonyl consequences, reduced NADPH/NADP+ availability, or increased NADH/NAD+ associated with polyol turnover. They are alternatives, not mandatory serial steps. A redox effect can occur without gross swelling. Myo-inositol depletion is not required. If the initial reaction occurs in a supporting cell, its consequence must reach the identified sensory axon; that bridge belongs in structural causation.

Human placebo nerve sorbitol of 0.0314 nmol/mg wet tissue corresponds to about 39 micromolar if tissue contains 0.8 mL water/g. That is only 0.013% of a 300 mOsm/L reference concentration. A large diffuse osmotic load is implausible at that average; a much smaller cellular volume could concentrate the pool. Neither localization nor cofactor turnover is measured by this calculation. [Human concentration](https://doi.org/10.2337/diacare.27.10.2369); [calculation and localization sensitivity](../../../findings/damage_calibration_polyol_quantitative.md).

## Conditional probabilities

Each probability is conditional on the evidence and all preceding claims being true. The cumulative event retains only baseline groups satisfying every preceding claim; a different group cannot be substituted at each step. Thus the last row completes the proposition. The low and high values belong to coordinated skeptical and favorable interpretations, not independent error bars.

| Claim | Additional proposition | Best probability | Sensitivity range | Why this neighborhood and breadth |
|---|---|---:|---:|---|
| R1 | At least one allowed group has glucose-driven aldose-reductase flux above its matched nondiabetic rate in a connected nerve compartment before prospective terminal loss. | 85% | 65–95% | Human inhibitor engagement and glucose-exposed human Schwann cells support the reaction; direct adult pre-loss cell-specific flux is absent. |
| R2 | In a group satisfying R1, that flux produces a metabolite/cofactor excursion outside the matched nondiabetic 95% reference interval for at least 90 cumulative days in the first year, beginning before prospective loss. | 85% | 65–95% | Human sorbitol/fructose pools support a persistent biochemical consequence; cell localization, duration and cofactor measurements are missing. No injury is assumed. |
| R3 | In a group satisfying R1–R2, suppressing the excess reaction averts at least 5% of five-year net peptidergic terminal loss through at least one named metabolite/cofactor branch. | 60% | 30–80% | Human structural intervention signals and diabetic-animal small-fiber rescue support causation. The human subtype, timing and mediation remain extrapolations. |
| R4 | In a group satisfying R1–R3, preservation reaches B ≥0.20. | 65% | 40–85% | Conditional on small-fiber causation, a material effect is plausible. Large trials make dominance less plausible but do not measure this structural effect. |
| R5 | At least one group satisfying R1–R4 contains C ≥0.10 of reference-care incident DPN cases. | 75% | 45–90% | Hyperglycemia and elevated polyols are not rare exposures; prevalence of the group with a material effect is nevertheless unmeasured. |

The 5% boundary in R3 separates a small structural contribution from the 20% materiality boundary. Since R4 implies that smaller boundary, it adds no restriction to the completed hypothesis. The reference interval and 90-day definition distinguish exposure from presumed toxicity. Changing those definitions or the allowed subgroups changes the priced proposition.

Base: 0.85 × 0.85 × 0.60 × 0.65 × 0.75 = **0.21133125**. Skeptical inputs [65,65,30,40,45]% give **2.3%**; favorable inputs [95,95,80,85,90]% give **55.2%**. Reader-facing rounding is about 20% and 2–55%. Merging R3 and R4 gives a conditional 39%, preserving exactly the same joint belief. A separate whole-claim assessment is “plausible, but less likely than not”: human biochemical engagement is much better established than prevention of this fiber's loss. That agrees with the product.

## Evidence and its limits

**Human biochemical exposure.** Dyck studied 21 people with diabetic neuropathy, four diabetic people without neuropathy and 11 controls. Endoneurial sugars were elevated; sorbitol correlated inversely with myelinated-fiber number (P=.003), while myo-inositol was not depleted. This is predominantly established disease, with too few unaffected diabetic donors to establish a pre-loss trajectory. [Dyck 1988](https://doi.org/10.1056/NEJM198809013190904).

Bril and Buchanan randomized 101 patients; 93 nerve specimens were analyzed (placebo/5 mg/20 mg: 33/31/29). Twelve weeks of ranirestat reduced nerve sorbitol 65.2% and 83.5%, respectively. This is direct nerve engagement, not proof of engagement in peptidergic axoplasm or of structural benefit. [Primary trial](https://doi.org/10.2337/diacare.27.10.2369).

**Human structural intervention.** Sima's randomized double-blind study used 250 mg/day sorbinil for one year in 16 people with neuropathy, ten on active treatment. It reported 41.8 ±8.0% lower nerve sorbitol and a 33% increase in myelinated-fiber density in the treated group (P=.04). Regenerating myelinated fibers also increased. These are consequential observations, but the accessible abstract does not supply a reconstructable randomized difference and confidence interval. Small myelinated fibers are not C-fibers. [Sima 1988](https://pubmed.ncbi.nlm.nih.gov/3136331/).

Greene's 52-week randomized zenarestat study measured serial contralateral biopsies and nerve chemistry. Its positive small-myelinated-fiber result came from a secondary analysis of doses achieving >80% sorbitol suppression. Achieved suppression is not randomized; neither a universal toxic threshold nor a large prespecified structural effect follows. The full numerical tables and contents of the linked 2001 erratum were not recovered. They limit the weight and precision assigned to this positive line. [Primary report](https://doi.org/10.1212/WNL.53.3.580), [erratum](https://doi.org/10.1212/WNL.56.8.1124-a).

**Larger and discordant intervention results.** In 549 randomized participants, 83% with type 2 diabetes and mean neuropathy duration 4.9 years, 10/20/40 mg ranirestat for 52 weeks produced adjusted sensory-conduction-sum differences of approximately 0.38/1.38/1.33 m/s versus placebo; none passed the multiplicity-adjusted comparison. Symptoms and quantitative sensory testing did not differ. [Bril 2009](https://pmc.ncbi.nlm.nih.gov/articles/PMC2699746/).

A separate 73-person, 26-week study found a sensory-conduction-sum benefit: 7.28 versus 1.92 m/s, difference **5.36 m/s**, P=.006. Adjustment for baseline HbA1c imbalance retained it, but total clinical score did not improve versus placebo. This is a sum across three nerves, not 5.36 m/s in each. [Japanese trial](https://pmc.ncbi.nlm.nih.gov/articles/PMC4736957/).

At 24 months, an 800-person trial found peroneal-conduction changes of 0.49/0.95/0.90 m/s for placebo/40 mg/80 mg. Neither active dose met its efficacy target; 633 completed follow-up. In a separate 557-person, 52-week study, 40 mg produced a 0.52 m/s tibial-conduction benefit but only a −0.19-point clinical-score difference. These support a small physiological action, not consistent broad clinical reversal. They contain no structural equivalence test against 20% terminal preservation. [24-month trial](https://pubmed.ncbi.nlm.nih.gov/26313450/), [52-week trial](https://pmc.ncbi.nlm.nih.gov/articles/PMC6400176/).

**Biological capability and transfer.** Biallelic SORD disease in 45 affected people from 38 families establishes that severe disposal failure can cause human axonal neuropathy, predominantly motor. It does not establish the diabetic dose or sufficient subgroup prevalence. [Cortese 2020](https://doi.org/10.1038/s41588-020-0615-4).

Human stem-cell-derived Schwann cells accumulated sorbitol under 5 versus 30 mM glucose over 72 hours; the tested sensory neurons did not show that response. Replicate cultures are not independent donors, and systemic bupropion rescue does not isolate a polyol mediator. [Majd 2023, targeted assays and methods](https://pmc.ncbi.nlm.nih.gov/articles/PMC10249419/).

Ten STZ-diabetic rats versus ten controls had elevated nerve sugars at 27 days without detected nerve-trunk fiber-density differences. This supports early chemistry, not absence of all terminal injury. [Temporal study](https://pubmed.ncbi.nlm.nih.gov/10580857/). In ob/ob mice, fidarestat 16 mg/kg/day from ages 5 to 11 weeks alleviated intraepidermal fiber loss. The primary abstract reports approximately 78% loss in untreated mice but does not provide structural arm means, uncertainty or independent animal count; those were not recovered from the full text. The severe leptin-deficient phenotype differs from ordinary human type 2 diabetes. [Drel 2006](https://pubmed.ncbi.nlm.nih.gov/17130477/).

The studies differ in endpoints and the strength of their causal comparisons. Biochemistry can change without structure, and structure can change without detectable clinical recovery. Neither possibility makes every clinical null uninformative. Large trials lower confidence in a common dominant pathway; structural observations prevent treating the whole pathway as disproven.

## Ideal experiments that would resolve each claim

These define truth conditions, not proposed work or evidence. Assume direct, noninjurious longitudinal measurement of adult human nerve compartments, stable identification and tracking of peptidergic terminals and replacements, and the substrate-selective reaction clamp above. Sample independent humans representatively, retain baseline subgroup membership, and observe reference clinical incidence for five years. Other initiating exposures remain comparable; downstream consequences may change. Each claim concerns groups satisfying its predecessors. NO requires excluding every remaining allowed group and branch, not merely one unsuccessful comparison.

**R1 — early glucose-driven flux.** Before counted terminal loss, measure isotope-resolved glucose-to-sorbitol formation per unit viable cell volume and time in axons, Remak cells and adjacent endothelium. Brief randomized matched glucose-exposure comparisons, with hormones and other substrates controlled, identify glucose dependence; repeated unmanipulated measurements establish the natural rate. YES means a rate above the matched nondiabetic rate in at least one allowed group, with the increase attributable to glucose before loss. NO means no positive glucose-attributable excess in any allowed connected compartment/group. An interval spanning zero, unresolved tracer recycling or post-denervation sampling remains unresolved. This is a directional rate claim: finite measurements cannot prove an exact zero. Human inhibitors establish reaction-dependent pools and human-cell perturbation supplies glucose direction; neither provides this complete adult time-resolved comparison. R1 does not resolve persistence or injury.

**R2 — a persistent biochemical excursion.** Assuming R1, measure intracellular sorbitol/fructose in mol/L and NADPH/NADP+ and NADH/NAD+ ratios through the first year. Compare the reaction clamp with reference care and use isotope mass balance to separate formation, disposal and exchange. YES requires at least one named pool/ratio outside its matched nondiabetic central 95% interval for at least 90 cumulative days, starting before loss, and removal of that excursion by the clamp. NO requires every branch to fail that duration/boundary or to be unrelated to the excess reaction. Incomplete compartment measurements, duration intervals crossing 90 days or insufficient clamp specificity are unresolved. A high stable pool can pass even when formation equals disposal. Human nerve sugars support the pool branch; sustained adult cell-specific cofactor trajectories remain absent. Passing R2 says nothing about toxicity.

**R3 — at least a small structural effect through the defined chemistry.** Assuming R1–R2, randomize the five-year reaction clamp versus reference care before prospective loss. Verify isotope flux suppression while retaining non-glucose aldehyde metabolism. Restore the measured sorbitol/fructose or cofactor trajectories in separate arms, singly and jointly, to determine whether one or their combination mediates preservation; these are the only allowed branches. A distinct manipulation of the same reaction checks intervention artifacts. Track actual terminals and replacements, rather than CGRP expression, pain or conduction. YES is B ≥0.05 attributable to the named chemistry; NO is B <0.05 or exclusion of mediation through every branch. For example, 30% reference loss versus 28% intervention loss gives B=6.7% and passes R3. Confidence bounds wholly on one side of 5% can decide a finite comparison; crossing bounds, failed engagement or uncertain subtype identity cannot. Human myelinated-fiber observations and mouse skin rescue supply parts of this comparison; none resolves it in early human peptidergic fibers.

**R4 — material magnitude.** Assuming R1–R3, use the same randomized structural contrast and baseline groups to estimate B against the **20%** boundary. Thirty percent loss versus 24% or less passes; 30% versus 28% passes R3 but fails R4. YES is B ≥0.20; NO is B <0.20 for every group surviving R3. Finite bounds crossing 20%, or reference mean loss indistinguishable from zero, are unresolved. Evidence that the reaction changes structure is not charged again: this claim concerns magnitude. Existing trials provide no human peptidergic estimate at this boundary.

**R5 — case coverage.** Assuming R1–R4, estimate how many incident cases in the representative reference-care population belonged at baseline to each surviving group. YES requires C ≥0.10 for at least one; NO requires C <0.10 for all. For example, 1,500 of 10,000 reference incident cases passes and 500 fails; adequate precision must exclude the 10% boundary. Overlapping groups cannot be added unless their union itself satisfies the preceding causal claims, which is not assumed. Sampling only high-sorbitol patients, incomplete incidence ascertainment or intervals crossing 10% remains unresolved. Common hyperglycemia supports plausibility, but no cited study supplies this coverage among initially unaffected adults. Passing R5 does not establish an effective or tolerable drug.

## What drives uncertainty and would change the judgment

The skeptical scenario combines mostly supporting-cell chemistry, little human subtype-specific preservation and a limited materially affected group. The favorable scenario interprets structural trials as a reproducible metabolic effect that also reaches early C-fibers in a common biochemical subgroup. Both permit small conduction benefits. Small-study bias, missing compartment/time measurements and case coverage drive the span; another precise conduction estimate cannot remove them. No distributional weights are assigned, so the span is neither a confidence/credible interval nor a variance.

Hypothetically, a selective human structural contrast supporting the fiber and mediator could raise R3 from 60% to 85%, taking the full product to **29.9%** if other inputs stayed fixed. Precise non-preservation despite engagement could lower R3 to 25%, taking it to **8.8%**. Actual evidence could change several inputs. Relaxing the 20% or 10% threshold broadens the proposition; it does not create evidence for it.

The named redox and carbonyl branches overlap [oxidative/nitrosative injury](39_oxidative_nitrosative_injury.md), [carbonyl stress](05_methylglyoxal_carbonyl_stress.md) and [PARP](28_parp_nad_depletion.md). None is a compulsory separate mediator, and their probabilities should not be multiplied into this assessment.