Intrinsic limits on axon regrowth

In adults with established type 2 diabetic peripheral neuropathy (DPN), reducing neuronal M1 muscarinic signaling, PTEN activity, or both would increase durable terminal replacement enough to reduce five-year net denervation by at least 20% in a baseline-defined group comprising at least 10% of the stated population.

This is a repair/progression exception, not prevention of incident DPN. The reference population has definite DPN for at least 12 months, residual distal-leg epidermal innervation of at least one fiber/mm, and surviving sensory parent neurons connected to the region studied. Completely lost neurons are outside this proposition. The target is physical terminals of human peptidergic C-fibers: unmyelinated sensory axons with a peptide-producing identity established before treatment. A change in peptide or PGP9.5 staining alone does not create a terminal.

The finite group list is: the whole reference population; baseline epidermal density 1–5 fibers/mm; DPN duration 1–5 years; HbA1c at least 8%; body mass index at least 30 kg/m²; or age below 65 years. Groups can overlap. There must be one group and one intervention branch satisfying every requirement; evidence from different responsive groups cannot be assembled into a successful chain. Group membership cannot depend on treatment response. The 10% denominator is this established-DPN population, rather than incident cases.

Q begins at enrollment and continues for five years. It acts only in the surviving sensory neurons and their axons, with three alternatives: reduce M1 receptor coupling to its intracellular G-protein signal by 80%; reduce PTEN lipid-phosphatase activity by 50%; or apply both changes. These fractions refer to each neuron's untreated activity at the same time and location. M1 is a receptor for acetylcholine; PTEN removes a phosphate from the growth-promoting membrane signal PIP3. Actual target activity, rather than transcript abundance or administered dose, defines engagement. This is a bounded manipulation of two named brakes, not an unlimited increase in every growth pathway.

Q changes neither glycemia nor insulin delivery, and directly manipulates no keratinocytes, immune cells, Schwann cells or vessels. Downstream neuronal metabolism, growth and communication with those cells can respond naturally. The comparison supplies identical vehicles and other care. Suppression below a nondiabetic activity level is allowed: this hypothesis concerns a brake that becomes consequential when repair demand rises, not necessarily a diabetes-induced excess of that brake. Current topical drugs and constitutive knockouts approximate parts of Q; neither establishes its human dose or cellular selectivity.

For each donor, follow the same baseline territory and identify physical terminal entries, subsequent disappearances, and connections to surviving parents. Count separate terminal units by unchanged anatomical rules, without rewarding fragmentation or brighter staining. Let N0 be baseline abundance and N5 abundance at five years. Mean reference loss L0 = mean(N0 − N5 under reference) must be positive. Benefit B is the Q-minus-reference difference in final abundance divided by L0. Gains can exceed the original loss; absence of reference progression supplies no evidence for slowing it.

The replacement condition also requires the excess number of surviving terminals first formed after treatment to be at least 20% of L0. Thus protection of pre-existing terminals alone cannot satisfy the proposition. Growth and protection may both contribute; these counts do not claim an independently identified molecular mediation fraction. Endpoint assessment must account for deaths and unobserved territories without assigning them zero loss or selecting treatment-dependent survivors. If their possible outcomes change which side of a boundary the result occupies, that requirement remains unresolved.

Source assessment

Best judgment: about 16%; sensitivity range: 0.8–59%. This is a probability that the complete causal claim below is true, not the expected treatment benefit. Human treatment data support retained capacity to increase epidermal innervation. The largest uncertainty is whether a defined change inside surviving neurons produces lasting, functional peptidergic replacements that keep pace with continuing diabetic injury.

Biological logic and alternatives

Terminal abundance reflects both disappearance and replacement. Surviving neurons can retain growth machinery while normal signaling limits its deployment. M1 antagonism can release metabolic and growth responses; partial PTEN suppression can increase PIP3-dependent growth. Neither route requires every case to share the same upstream metabolic defect.

New branches must enter appropriate skin territory, remain connected, acquire sensory function, and survive. Extra branches in a nerve trunk, a growth-marker increase, or a single larger biopsy count does not by itself demonstrate useful replacement. Extrinsic support, loss of parent axons, inadequate energy delivery, continuing destruction, or inappropriate branching can each prevent the proposed benefit. A topical treatment acting through skin or glia may help patients while failing this neuron-specific proposition.

Conditional probabilities

Each row after the first assumes that the complete preceding prefix holds for at least one of the same group–Q combinations. A later row asks whether such a combination extends through that row. It is not a probability averaged over different successful groups. Percentages are subjective conditional judgments.

See the claims and probabilities table above.

R1

Neuronal perturbations and human biopsy responses support capacity; human target engagement, subtype and cellular attribution are missing.

R2

Some animal structural effects persist, but human birth-dated functional replacement has not been measured.

R3

Six-month abundance can improve without a sustained balance against ongoing loss.

R4

Residual innervation is common enough to make coverage plausible; responsive subgroup prevalence is unmeasured.

Best inputs [65,60,60,70]% give 0.65 × 0.60 × 0.60 × 0.70 = 16.38%. Lower [35,25,25,35]% and upper [90,85,85,90]% scenarios give 0.765625% and 58.5225%. These are coherent skeptical and favorable interpretations, not statistical confidence limits, a distribution of treatment effects, or a variance estimate. The breadth chiefly reflects neuronal attribution and the conversion of extra entries into sustained replacement. Merging persistence and materiality into one conditional at 36% preserves the best joint probability; adding row labels is not evidence against the mechanism.

Evidence that changes the assessment

A controlled human structural signal, with an imperfect mechanistic comparator. Sivadasan and colleagues randomized 58 adults with established T2D neuropathy to 24 weeks of topical pirenzepine or placebo; 57 entered the modified intention-to-treat analysis. Reported pooled ankle intraepidermal nerve fiber density (IENFD) changes were +1.8 versus −0.6 fibers/mm, with a baseline-adjusted between-group P=.012. Biopsies were centrally counted blind. Other efficacy measures did not differ in that analysis. This supports a local structural response, but it neither identifies the neuronal compartment nor distinguishes new growth from survival. Primary trial.

Interpretation is limited by 12 treatment discontinuations among 46 active recipients versus none among 12 placebo recipients, incomplete reporting of outcome-specific biopsy denominators, and a vehicle difference. The public protocol's formulation table lists 365 versus 438.2 mg emollient per 2 mL, a 16.7% reduction with active treatment, but equal ethanol doses. That conflicts with the article's explanation of higher ethanol in the active formulation. The documented emollient difference, rash and subsequent introduction of moisturizer complicate drug-specific attribution; their net effect on innervation is unknown. They do not establish that the structural contrast is false. The protocol specifies a within-group IENFD test; the reported randomized contrast is the relevant comparison but remains exploratory. Trial and public supplemental protocol, Table 3 and section 13.

Quality-of-life improvement in the per-protocol subset does not establish functional recovery in the randomized population: the reported treatment-minus-placebo changes are −10.4 points in that subset but +3.5 points in the broader analysis, where a reduction is favorable. HbA1c was not collected in this trial. Stable treatment and safety glucose measurements do not prove matched cumulative glycemia.

Oxybutynin is not a successful mature-fiber replication. Its 20-week trial randomized 51 participants. Reconstructing the 37 readable primary biopsy pairs from the public figure gives a treatment-minus-placebo change of +0.833 fibers/mm, 95% CI −0.723 to +2.390, P=.282. Baseline IENFD was substantially lower in the active group; baseline-adjusted comparisons were also nonsignificant. A secondary sprouting comparison was +1.049 profiles/mm (0.214–1.885), nominal P=.0155. This is provisional evidence for growth-related profiles, without proof of later maturation; it is not another significant mature-IENFD trial. Casselini et al., 2024; figure-based quantitative analysis (background note outside this collection).

The positive 24-week pirenzepine study is NCT04005287. The separate completed 12-week study NCT04786340, also enrolling 58, still has no posted results in the public registry checked for this assessment. It supplies no efficacy replication and cannot be treated as negative biology. 24-week record; 12-week record.

M1 experiments establish an intrinsic growth-control mechanism in rodents. Calcutt and colleagues found enhanced adult sensory-neuron outgrowth with structurally different M1 antagonists, receptor deletion, and downstream AMPK manipulations. M1-deleted mice were protected from diabetic IENFD loss; pharmacological treatment reversed established indices, and some benefits persisted for 5–9 weeks after withdrawal. Relevant in-vivo groups were generally 3–12 animals, not the dozens of neurons counted in some culture panels. The knockout was not restricted to adult peptidergic neurons, and ordinary staining did not track the lifetime of individual new terminals. M1 messenger RNA did not significantly increase in diabetic ganglia, consistent with release of a normal brake rather than correction of necessary receptor overexpression. Calcutt et al., 2017.

PTEN supports plasticity, with a distinction between extra axons and improved repair. A chronic diabetic mouse study reports benefit from combined local PTEN knockdown and insulin, including epidermal innervation, without correcting glycemia. Its abstract describes comparisons of component interventions, so it is not simply a study with no single-component arms. However, the full primary efficacy tables were not recovered here; the available abstract does not establish a neuron-only, PTEN-only structural effect at Q's dose. Schwann-cell effects and added insulin are explicit alternative contributors. Pham et al., 2023.

In a separate sensory-neuron PTEN deletion study, intact mouse nerves developed extra myelinated and unmyelinated axon clusters and higher sensory action-potential amplitudes, while recovery after nerve crush changed only modestly or not at all. The accessible preprint's skin comparisons used four animals per group and one-sided P=.041–.047; the corresponding two-sided values are .082–.094. This does not erase the nerve-trunk phenotype. It limits confidence in the skin-specific result and shows why more axons cannot automatically be read as better injury recovery. Developmental deletion, nondiabetic animals, and prominently nonpeptidergic PTEN expression differ from Q. The final 2026 abstract retains the central distinction; detailed final tables were unavailable. Eaton et al., final article; accessible primary preprint.

Human regeneration studies demonstrate a deficit without locating its cause. After similar capsaicin denervation, Polydefkis and colleagues reported regrowth rates of 0.177 fibers/mm/day in controls versus 0.074 in diabetes, retaining a difference after baseline-density adjustment; rates were 0.10 without neuropathy and 0.04 with neuropathy. There were 31 controls and 19 diabetic participants, including eight with neuropathy. This supports impaired replacement under a standardized injury, not an M1/PTEN mechanism. Polydefkis et al., 2004.

The longer Khoshnoodi study included 36 T2D, 11 T1D and ten control participants. T2D participants were older; the adjusted control-minus-T2D initial-rate contrast was 0.044 fibers/mm/day, with an interval spanning −0.008 to 0.097. Late net reinnervation was incomplete, but a plateau can result from slowed entry or continued disappearance. Fiber counts alone do not identify these processes. Neither the initially neuropathy-free population nor experimental axotomy directly measures five-year progression in established disease. Khoshnoodi et al., 2019.

Quantitative support (background note outside this collection) contains the summary-data checks, formulation comparison and limits on recovered denominators. No pooled drug-effect estimate is used: the interventions, populations and structural measurements are not interchangeable.

Ideal experiments that would resolve each claim

These are truth tests for the stated proposition. The common ideal comparison follows independent human donors in each listed baseline group, with Q, matched reference and target-engagement measurements in identified neurons. It measures entries and losses repeatedly through five years, retains baseline subtype identity, and separates newly formed terminal units from rediscovered or newly stained old ones. The participant is the inference unit. Animal or culture results can support a component but cannot decide the exact human proposition.

For each row, YES means the true joint quantities meet the inclusive boundaries in at least one combination satisfying the earlier rows; NO means every such combination fails at least one boundary. Finite evidence must resolve those quantities with sufficient joint precision. Intervals crossing a boundary, missing cellular exposure, unsuitable subtype identification or unresolved follow-up remain unresolved. Failure of an earlier prefix does not empirically estimate a later conditional.

1 — Neuronal release and new entry. Compare each of the three Q branches against the same person's reference counterfactual under otherwise matched diabetes, skin support and care. Verify 80% M1 signaling suppression, 50% PTEN suppression, or both in the relevant neuronal compartments, with no direct support-cell perturbation. Count the excess number of unique, physically connected new terminal entries during year one relative to baseline abundance. YES requires at least 5%; NO requires less than 5% for every candidate combination despite verified engagement. Neuronal target expression without response does not satisfy this row. Human topical IENFD supplies partial structural support; rodent receptor and PTEN perturbations supply cellular plausibility. Neither provides the exact selective human comparison.

2 — Persistence and sensory function. Follow the birth-dated extra entries from row 1 for at least 90 days each, extending observation beyond year one when necessary. Confirm anatomical continuity and stimulus-evoked conduction from their peripheral receptive territory into the parent axon. Compare the number meeting both criteria with the corresponding reference number. YES requires an excess of at least 5% of baseline abundance; NO requires an upper bound below 5%. Pain relief, aggregate sensory potentials, transient sprouts and marker positivity alone are unresolved. Post-withdrawal rodent structural effects bear on persistence; the human sprout result has no linked later maturity measurement.

3 — Material replacement during continuing injury. Continue the same comparison for five years. Directly measure total surviving abundance and the subset first formed after enrollment. YES requires positive reference net loss, B≥20%, and a Q-minus-reference excess in the surviving newly formed subset of at least 20% of L0. Both magnitude conditions are necessary. Protection confined to old terminals, early growth later erased, or a precise benefit below either boundary is NO for this row. A truly nonpositive reference loss fails this relative-preservation proposition; imprecision about a near-zero reference loss leaves it unresolved. The pirenzepine trial supplies six-month total counts; it does not measure either five-year quantity. Death and missing structural outcomes must not manufacture a benefit.

4 — Population coverage. Measure the prevalence of the successful baseline group in a representative sample of the defined established-DPN population and estimate rows 1–3 for that group with the same Q branch. YES requires its prevalence to reach 10%; NO requires every successful combination to remain below 10%. A post-treatment responder fraction, the prevalence of M1 expression, or the proportion of people with low IENFD alone is unresolved. The existing small selected trials establish neither this denominator nor a reproducible responsive subgroup.

What would move the judgment

A hypothetical vehicle-matched human structural replication that also identifies neuronal engagement, functional new terminals and persistence would move several rows together. Inputs [90,85,80,85]% would give 52.02%. Replication of the same six-month count alone would justify a smaller change, because neuronal localization and five-year persistence would remain open.

Conversely, a hypothetical result showing that the topical effect is entirely extrinsic and that engaged neuronal suppression produces only short-lived or misdirected sprouts would move the inputs to [30,35,30,40]%, giving 1.26%. An underpowered null or the absence of public trial results would not justify that update.

The current record supports retained growth capacity more strongly than this complete repair claim. It does not distinguish a five-point probability difference reliably. The related skin repair environment, trophic support, energy and transport hypotheses can participate in the same response. Their different interventions and populations prevent a simple probability ordering or addition.