RAGE signaling

In adults with type 2 diabetes diagnosed within five years, without confirmed distal polyneuropathy and with initially preserved distal-leg innervation, increased local engagement of the receptor for advanced glycation end products (RAGE) produces persistent signaling that contributes to terminal loss. Adult-onset interruption of that signaling, beginning before prospective loss, would avert at least 20% of five-year net peptidergic C-fiber terminal loss in a baseline subgroup containing at least 10% of incident DPN cases under reference care.

The candidate compartments are sensory neurons, their Remak Schwann cells, nerve-associated macrophages, and the adjacent microvascular endothelium. A compartment must connect to the tracked distal sensory axons. The allowed ligand families are AGE-modified proteins, HMGB1, S100B and S100A8/A9; one or a combination can supply the exposure. This is a receptor hypothesis, not a claim that every AGE effect requires RAGE. Matrix glycation and direct carbonyl injury have separate routes.

The ideal intervention disables intracellular RAGE signaling in one specified compartment throughout follow-up, preserving ligand binding and other receptors. It does not sequester ligands from their other receptors or alter systemic glucose treatment. It removes protective as well as damaging RAGE signals in that compartment: the net structural result determines the claim. This adult, cell-selective manipulation is an idealization, not an existing drug. Direct neuronal and indirect support-cell routes are alternatives; a positive macrophage result need not imply a neuronal-autonomous effect.

The bounded baseline subgroup family comprises the whole eligible population, HbA1c ≥7.5%, or local occupancy of RAGE by an allowed ligand family above the matched nondiabetic 95th percentile in a specified compartment. Membership is determined before manipulation, never by future response. The same group, compartment and ligand route must survive every requirement. Unrestricted searches for an unnamed responsive subgroup are outside this proposition.

Net loss is terminal disappearance minus successful replacement over five years, divided by baseline terminal number in a fixed skin territory. Stable subtype identification must distinguish loss from reduced peptide expression. Let B = (mean reference loss − mean intervention loss)/mean reference loss, requiring positive mean reference loss. The full hypothesis requires B ≥0.20. Case coverage C is the fraction of incident cases under reference care belonging to the qualifying baseline group; C must be ≥0.10. Incident DPN means new bilateral length-dependent sensory signs with objective small-fiber loss or abnormal nerve conduction, excluding another neuropathy. These thresholds define the question; they are not observed attributable fractions. Pain relief, type 1 models and repair after crush have different endpoints or scopes.

Source assessment

About 10% probability; uncertainty range (sensitivity) 0.3–43%. The central product is 9.4%. RAGE can alter nerve inflammation, sensation and regeneration. The main uncertainty is whether its net effect in early human diabetes destroys enough peptidergic terminals to meet the structural claim. Literature reviewed 14 September 2026.

Biological logic

Ligand production, clearance and binding determine receptor occupancy. Tissue staining cannot distinguish extracellular ligand available to a receptor from intracellular or fixed material. Receptor expression likewise does not establish signaling. A cell can have active RAGE without that activity causing structural loss.

The defined activity readouts are receptor-dependent nuclear NF-κB, phosphorylated ERK1/2, phosphorylated STAT3 or GTP-bound RAC1. These are alternative signaling outputs. For this proposition, persistent activity means an output above its matched nondiabetic 95th percentile for at least 90 cumulative days in year one, with the excess removed by the selective RAGE interruption. This operational boundary does not presume toxicity. Effects on inflammation, perfusion, trophic signaling and axonal transport may follow and are allowed to change.

The structural sign is not guaranteed. NF-κB and ERK can support growth, and injury-recruited macrophages can assist regeneration. Preventing inflammatory injury could preserve axons while blocking a useful repair signal could remove that benefit. Net terminal maintenance, including replacement, is therefore the appropriate outcome; neither lower inflammatory staining nor less pain settles it.

Conditional probabilities

Each row assumes the preceding claims really hold in at least one common eligible group and connected compartment. The event after each row retains only groups satisfying all preceding rows. The complete product is a chain-rule judgment, without assuming independence. R3 measures a small structural effect; R4 adds magnitude and R5 adds coverage.

See the claims and probabilities table above.

R1

Human vascular ligand/receptor localization and diabetic mouse chemistry support exposure. Occupancy, early timing and precise ligand identity are poorly measured.

R2

Receptor perturbations identify signaling in mice and cultures. Given actual local engagement, activation is plausible; human duration and competing receptor contributions remain uncertain.

R3

Experimental injury and support-cell effects establish capacity, but small-fiber protection is inconsistent and RAGE can promote repair. This is the main causal transfer uncertainty.

R4

Once a structural effect exists, amplification could make it material. Concurrent injury routes and lost protective signaling could limit its size.

R5

The ligands occur in common metabolic/inflammatory states, but prevalence of a materially RAGE-dependent subgroup is unknown.

The 5% boundary is a chosen small-effect definition nested inside 20%; it adds no restriction to the completed claim. A 30% reference loss versus 28% with interruption gives B=6.7%, passing R3 and failing R4. This prevents counting structural causation twice.

Base inputs [75,80,40,60,65]% give 9.36%. Coordinated skeptical inputs [45,50,15,30,30]% give 0.304%; favorable inputs [90,95,70,85,85]% give 43.242%. Merging R3 and R4 gives a conditional 24% and leaves the product unchanged. Independently considering the full proposition suggests a possible but unlikely material human structural cause: multiple experimental routes exist, without a human structural intervention anchor. That agrees with a judgment near one in ten, not a precisely measured 9.36%.

Evidence for and against

Human localization is not early causal exposure. Bierhaus studied ten people with DPN and eight controls, six of whom had Charcot–Marie–Tooth neuropathy. Half the diabetic samples came from severe peripheral vascular disease amputations. RAGE, carboxymethyllysine, activated NF-κB and IL-6 colocalized predominantly in vascular/supporting structures. These are highly selected established cases with unmatched alternative pathology. In the accompanying six-month diabetic-mouse experiment, RAGE deletion improved sensation without reported preservation of plantar PGP9.5-positive fibers. The structural comparison had only three healthy and five diabetic mice per genotype. Figure 6A's caption and bar colors disagree about control/diabetic assignment, so its bars do not support a reliable reconstructed protection interval. The reported dissociation matters, but does not establish a precise zero. Bierhaus 2004, original Figure 6.

A separate human skin study included 44 people with type 2 diabetes or prediabetes: 29 contributed immunostaining/fiber-density analyses and a separate 15 contributed RNA measurements. Across vascular RAGE staining grades I–IV, median intraepidermal density was 6.7, 4.2, 1.9 and 2.3 fibers/mm (P=.0037). Nephropathy and dyslipidemia also differed with severity. This supports a tissue association, not receptor occupancy, temporality or mediation. The observations do not require accepting every mechanistic citation in that paper. Human skin study, 2014.

A positive structural paper was retracted. The 2008 experimental diabetic-neuropathy paper by Toth and colleagues is excluded. The notice identifies a fabricated immunoblot and other figures drawn from older, unrepresentative cohorts. Its claimed serial structural protection supplies no positive weight here. This is separate from the surviving Bierhaus and human skin observations. Publisher retraction.

Culture establishes a possible injury route at a selected dose. Vincent exposed embryonic-day-15 rat sensory neurons to bovine-brain S100, 5 μg/mL. At 24 hours, DNA-fragmentation labeling rose markedly; soluble RAGE or a receptor-blocking antibody reduced it. Experiments were repeated on three culture occasions, not three specified independent human donors. Basal medium already contained 25 mM glucose; the high-glucose condition added 20 mM, totaling 45 mM. RAGE blockade did not rescue that glucose-induced injury, although α-lipoic acid did. Thus S100 can cause receptor-dependent injury in this preparation, without establishing that the same route mediates glucose injury or occurs at human nerve ligand concentrations. The S100 preparation is not a defined circulating human ligand mixture. Vincent 2007, methods and Figure 6.

Macrophage and large-fiber protection do not resolve the terminal endpoint. At eight weeks of STZ diabetes, Osonoi found protection of conduction, neuronal size and retrograde transport with global or bone-marrow RAGE deletion. Original Table 1 has six mice per group: diabetic wild-type and knockout skin densities were 56.5±7.6 and 59.7±7.1 fibers/mm, versus 67.4±4.9 in nondiabetic wild type (SD). The knockout difference is +3.2 fibers/mm, approximate unadjusted 95% interval −6.3 to +12.7. The authors reported no significant small-fiber impairment under their multiple-comparison analysis; that does not make the observed diabetic reduction zero. Neither decisive terminal rescue nor precise exclusion follows. Marrow transplantation also changes more than a single resident macrophage lineage. Osonoi 2022; numerical support (background note outside this collection).

Repair can change the sign. In diabetic mice after sciatic crush, Juranek reported improved myelinated-fiber regeneration with global or marrow RAGE deficiency. Digitized day-21 global-knockout versus wild-type density is approximately 7,518 versus 6,423 fibers/mm², 17% higher, with nine mice/group. This is not 17% of diabetic terminal loss prevented. Axon caliber, remyelination and tissue area affect density. Juranek 2013.

Conversely, RAGE/ligand blockade impaired regeneration after nondiabetic crush, including receptor-blocking antibody and neuronal/macrophage dominant-negative manipulations. Effects on day-21 myelinated density were large, although histological groups were small, approximately 3–6 mice. These receptor-specific comparisons make a generic “all RAGE activity damages nerves” model untenable. Rong, ligand/receptor blockade, cellular RAGE perturbations. Adult rat sensory cultures also showed approximately doubled neurotrophin-supported outgrowth with glycated albumin, S100B or HMGB1, and 40–90% inhibition by RAGE suppression. Full dose and independent-preparation details were not recovered for that study, limiting a quantitative comparison with the embryonic injury assay. Saleh 2013.

Later studies do not supply the missing human structural test. Azeliragon 10–30 mg/kg relieved mechanical hypersensitivity in STZ mice at three hours; a single-dose effect disappeared by 24 hours. Repeated 30 mg/kg dosing prolonged analgesia without changing glucose. These are behavioral outcomes; the accessible report does not provide a terminal-preservation comparison or recoverable independent n. Ma 2023. Six-month STZ studies of DIAPH1 deletion and combined AGER/DIAPH1 deletion measured predominantly myelinated structure and conduction; double deletion protected motor but not sensory conduction. Similar protocols and matching wild-type values/plot patterns prevent counting their controls as independent replication. There is no same-experiment AGER-only arm establishing epistasis. DIAPH1 study, double deletion.

Searches through the review date did not recover a randomized human RAGE-selective DPN structural trial or a genetic perturbation that identifies this exact causal effect. Biomarker correlations and trials in other diseases cannot fill that comparison. Missing accessible full details of the adult culture and azeliragon studies restrict quantitative interpretation, rather than establishing biological absence.

Ideal experiments that would resolve each claim

These are conceptual truth definitions. Assume noninjurious measurements in connected adult human nerve compartments, stable longitudinal identification of peptidergic terminals and replacements, and the compartment-selective RAGE signaling interruption defined above. Use independent humans, representative sampling and randomized intervention allocation. Begin before prospective loss and retain the same baseline groups. Hold diabetes care and other initiating metabolic exposures comparable; permit downstream inflammatory, vascular and trophic effects to change. Measure receptor engagement and signaling specificity directly, and restore intracellular RAGE signaling in a separate arm to distinguish its effect from an intervention artifact. NO requires excluding all allowed surviving groups/compartments, not just one negative experiment.

R1 — ligand access and occupancy. With no predecessor assumed, measure the fraction of RAGE molecules occupied by each allowed ligand family in the four candidate compartments, alongside matched nondiabetic reference trajectories, before tracked terminal loss. Identify extracellular bound ligand rather than total tissue staining. YES means an occupancy increase above the matched reference in at least one baseline group before loss; NO means no positive increase for any allowed family/group/compartment. Unknown binding identity, unmatched clearance or an interval spanning zero remains unresolved. A finite study cannot distinguish every arbitrarily small increase from exact zero. Existing human staining supplies location, not this temporal occupancy contrast. No signaling or injury is inferred from passing R1.

R2 — persistent receptor-dependent activity. Assuming R1, follow nuclear NF-κB, phosphorylated ERK1/2 and STAT3, and GTP-bound RAC1 in the same cells through year one. Compare randomized RAGE signaling interruption with intact signaling while retaining ligand exposure and other receptor functions. YES requires at least one named output above its matched nondiabetic 95th percentile for ≥90 cumulative days beginning before loss, with removal of that excess by RAGE interruption. NO means none satisfies duration and receptor dependence. Incomplete silencing, unmeasured branches or timing/precision crossing the boundary remains unresolved. This establishes activity rather than its structural sign. Mouse receptor perturbations supply causal dependence; early adult human trajectories are absent.

R3 — small net structural causation. Assuming R1–R2, maintain the specified interruption for five years versus reference care and directly count persistence, disappearance and successful replacement of identified human terminals. Verify cell selectivity and use signaling restoration or an independent receptor-specific intervention to exclude artifacts. YES is B ≥0.05 in a surviving group; NO is B <0.05 in every such group, including net harm from impaired repair. An estimate of 6.7% with bounds above 5% can pass; an estimate near zero with bounds below 5% can fail. Bounds crossing 5%, failed target engagement or peptide-staining-only changes remain unresolved. This estimates the total effect of the defined RAGE intervention; the optional downstream routes need not all be harmful. Existing animal interventions address parts of causation, without the early human subtype comparison.

R4 — material magnitude. Assuming R1–R3, estimate B in the same randomized comparison against 20%. Thirty percent reference loss versus 24% intervention loss passes exactly; 30% versus 28% fails despite passing R3. YES is B ≥0.20 for a surviving group; NO is B <0.20 for all. Finite intervals crossing 20% or a reference loss denominator insufficiently separated from zero are unresolved. This adds effect size, not another test of whether RAGE matters at all. No existing study measures this exact human quantity.

R5 — population coverage. Assuming R1–R4, measure baseline subgroup membership and incident DPN under representative reference care for five years. YES means at least one materially protected group's share of reference incident cases is ≥10%; NO means every share is <10%. For example, 1,500 of 10,000 reference cases passes, 500 fails, with adequate precision around the boundary. Overlapping groups are not added unless the union satisfies the earlier effect requirements. Selected biopsy cohorts, incomplete incidence follow-up or bounds crossing 10% remain unresolved. Current human localization studies establish neither the qualifying group's prevalence nor this case fraction.

Why the range remains broad

The skeptical interpretation places most measured ligands/activity after injury or in vascular disease, transfers little of embryonic toxicity to adult terminals, and expects protective signaling to offset benefit. The favorable interpretation gives causal weight to diabetic repair and macrophage perturbations, assumes their effects extend to early C-fiber maintenance, and places that state in a common ligand-high group. Those coordinated interpretations justify the endpoints of the range. Neither requires treating the small mouse null as definitive or counting the retracted positive.

These are literature-informed subjective probabilities, not empirically calibrated forecasts. The range is sensitivity to defensible interpretations of the same proposition, not a confidence interval or a distribution over probabilities. There is no defensible numerical variance. Changing ligand scope, compartment selectivity, duration or the 20%/10% thresholds changes the proposition and requires a separate elicitation.

Hypothetically, a selective early human structural contrast could raise R3 to 75%, taking the full product to 17.6% if other judgments stayed fixed. A precise engaged non-preservation contrast could lower R3 to 15%, taking it to 3.5%. Actual evidence could change several rows. A better pain response alone would not warrant the first update. The overlap with TNF/TLR4 signaling, transport, vascular delivery and oxidant injury does not make all of those compulsory serial discounts.