# 33. Persistent senescence-like secretory signaling contributes to terminal loss

**Best judgment: about 10%; uncertainty range (sensitivity): 0.1–62%.** Persistent secretory states can impair peripheral nerve repair, and structural intervention evidence now extends beyond pain behavior. Whether the same process precedes and materially contributes to early human diabetic terminal loss remains unresolved. These are literature-informed subjective probabilities, not an empirically calibrated treatment success rate. Reviewed 14 September 2026. [Probability method](probability_method.md) · [Collection index](README.md).

## The causal claim and its scope

In adults with type 2 diabetes diagnosed within five years and no clinical DPN at entry, a specified persistent cellular state generates a local secretory exposure before loss of the connected peptidergic C-fiber arbor. Neutralizing that source's specified extracellular products from entry would avert at least **20% of five-year net physical terminal loss** in a baseline-defined group accounting for at least **10% of incident DPN under reference care**.

The assessed branch is secretory signaling. Intrinsic dysfunction of a stressed neuron or Schwann cell can matter without this branch being true. A universal intervention that “reverses senescence” has no single molecular meaning; this assessment instead specifies both the source state and the products intercepted. Neither the state definition nor the intervention selects cells according to whether they later prove harmful.

**State and compartments.** The eligible sources are the connected sensory neuron, its satellite glia, its ensheathing nonmyelinating Schwann cells, and fibroblasts or keratinocytes at its distal skin interface. A cell satisfies the operational senescence-like state when all of the following persist for at least 14 consecutive days:

- p16 or p21 protein exceeds the matched nondiabetic 95th percentile and twice the positive reference median.
- Senescence-associated β-galactosidase activity measured at pH 6 exceeds the same two reference boundaries.
- It contains at least three colocalized γH2AX/53BP1 nuclear damage foci, also exceeding the reference 95th percentile.

For an assay with a zero reference median, twice the validated detection limit replaces twice the median. These are chosen measurement boundaries, not established toxicity thresholds. They do not require a postmitotic neuron to cease dividing, and they do not establish irreversible cell-cycle arrest. The composite deliberately excludes a transcriptomic score alone.

**R1's occurrence boundary:** at least 10% of one listed cell class, averaging each person's cell fraction equally, satisfies this state during the first year. It must precede the connected arbor's first ≥5% net physical deficit lasting 30 days.

**R2's exposure boundary:** before that deficit and during the first year, at least two proteins from **IL-1α, IL-1β, IL-6, TNF, HMGB1, CCL2, CXCL1, CXCL8, MMP2, MMP9 and MMP12** exceed their matched nondiabetic 95th percentile and twice their positive median locally for at least seven cumulative days. The same zero-median rule applies. At least 20% of each excess local exposure must originate from the R1 state cells, established by source tracing. The recipient compartment is the connected neuron or its satellite/Remak interface, including the terminal interface. Plasma concentration does not substitute for local exposure.

**Q — source-specific extracellular interception.** An ideal controller neutralizes at least 95% of those listed products as they leave operational state cells, including their extracellular vesicle-associated or matrix-bound forms, throughout five years. It does not kill those cells, alter p16/p21 or intracellular stress programs directly, or remove the same proteins from other sources. It intercepts useful as well as detrimental products from the specified source. Other upstream diabetic exposures remain comparable to contemporary diabetes care without Q; downstream inflammation, repair, neuronal activity, feedback and cell survival can change freely. Lost protective signaling and indirect cell loss count against benefit.

**Groups and endpoints.** The finite candidate groups are the whole eligible population, males, females, baseline BMI ≥30 kg/m², baseline HbA1c ≥7.5%, age ≥65 years, and the group with ≥10% of one eligible cell class satisfying the composite state during a 30-day baseline observation. No new intersections or future responder groups are admitted.

Track stable peptidergic identity and physical anatomy in fixed distal-leg territories. Let L₀ and L_Q be the person-weighted mean five-year net losses under reference care and Q, each relative to baseline terminal amount. Net loss includes destruction minus replacement, and any subsequent loss of replacements. With L₀ > 0, define **B = (L₀ − L_Q)/L₀** and **C = P(group | incident DPN under reference care)**. New bilateral length-dependent nerve signs with objective small-fiber loss or abnormal conduction, excluding another cause, define incident DPN.

A claim at each step means that at least one candidate group satisfies the entire prefix through that step. Groups failing a predecessor cannot supply a later success. Thus the final conjunction concerns the same group, connected cells and time course; a complete NO excludes every remaining candidate group.

## Reasoning from the biology

DNA damage, lysosomal changes and cell-cycle inhibitors can accompany chronic stress. None specifies the cell's extracellular output. RNA abundance must pass through translation, secretion, extracellular retention, clearance and delivery before establishing an exposure at the axon. Elevated IL-6 in blood and IL6 RNA in a DRG cell do not measure that sequence.

The relevant path is:

`persistent source state → specified local secretory exposure → altered destruction or replacement → net terminal deficit`.

Cytokine signaling, matrix proteolysis, altered glial support and axonal execution are alternative downstream mediators. No particular one is mandatory. A state could impair intrinsic Schwann-cell repair while its secretions remain neutral; that would support an intrinsic-state hypothesis but fail this secretory claim. Conversely, secretory injury need not kill the source cell.

There are plausible reversals. p21 may help a stressed postmitotic neuron survive. Inflammatory recruitment can aid debris removal and repair. Removing a source can remove trophic support as well as cytokines. A pain-producing neuron can disappear after a senolytic while the animal becomes less pain-sensitive: analgesia and preserved innervation can move in opposite directions.

For scale, if reference net loss is 30% of baseline terminals, 20% preservation requires six percentage points. A cell-killing treatment that directly costs four points would need ten points of other preservation to meet that target. This is accounting, not a measured diabetic loss rate. Q avoids direct killing, but its indirect costs remain in the endpoint. [Calculations](../../../findings/damage_calibration_senescence_quantitative.md).

## Evidence that moves the judgment

**Human painful-DPN signatures are positive at the donor level, but do not establish the source state or timing.** The iPain study applied a SenMayo gene set to a painful-DPN DRG dataset. Its released, age-filtered comparison contains **four DPN and six nondiabetic donors**, not a longitudinal diabetes-without-DPN cohort. Recalculation gives a mean score difference of **1.68 standardized-score units**, approximate Welch interval **0.23–3.12**, and **AUROC 0.958** (23 of 24 cross-group donor pairs; donor-bootstrap interval 0.75–1.00). The exact two-sided label-permutation P is 0.0095.

The DPN donors are 5.3 years older on average. A small age-adjusted model retains the difference, but cannot remove the joint contrasts in diabetes, established neuropathy and pain. The gene set neither demonstrates simultaneous p16/p21, lysosomal and DNA-damage criteria nor identifies local protein release. [Techameena et al., 2024](https://doi.org/10.1038/s41467-024-52052-8); [donor calculations](../../../findings/damage_calibration_senescence_quantitative.md).

The same paper's other-pain comparison has **50 ganglion samples with identifiers indicating 34 donors**, including five donors contributing both pain states. A donor-weighted, age-adjusted comparison gives **+0.114 score units (0.001–0.226)**; the five within-donor contrasts average +0.117, exact sign-flip P = 0.438. The reported extremely small generated-cell P value cannot make thousands of model-generated cells independent humans. These analyses retain a pain association without establishing a precise causal effect.

A separate public sural-nerve analysis of **77 established-DPN donors**, including 27 regenerator and 22 degenerator extremes, found **AUROC 0.517** for the bulk SenMayo score; the standardized extreme-group difference was 0.060. This weak discrimination limits that proxy in that tissue and comparison. It does not show that local secretion, a minority source state or early timing is absent. [Analysis and endpoint interpretation](../../../findings/aug13_recent_dpn_omics_triage.md).

**Direct pain experiments do not resolve structural protection.** iPain's senolytic experiments followed mouse nerve injury; marker reduction and analgesia were not measurements of human terminal preservation. Donovan subsequently found p16/p21/IL6-associated states after injury and with aging. The electrophysiology comprised **82 neurons from 26 mice**; cell-level analyses alone overstate the independent n. In the available source data, high firing poorly distinguished p16-positive cells (AUROC 0.542) or p21-positive cells (0.508). IL-6 increased evoked firing, which is a functional effect rather than an axon-loss endpoint.

After five days of ABT263, the apoptotic-marker fraction rose from **1.37% to 3.02%**, with six mice per arm; this snapshot does not quantify cumulative neuronal loss. Its four human donors, two young and two aged, were not a DPN prevention cohort. Human IL6 staining and mouse plasma ELISA do not establish source-specific local secretion. [Donovan et al., 2025](https://doi.org/10.1038/s41593-025-01954-x); [source-data analysis](../../../model_effort/integrative_model_v1/review/new_senolysis_function.md).

**Schwann-cell experiments supply structural and secretory evidence in repair.** Fuentes-Flores studied aged nerves and chronically denervated mouse stumps. ABT263 and p16-3MR/GCV treatment improved axonal regeneration after reconnection. Reanalysis using mice as units found proximal regeneration ratios of **1.41, 1.74 and 1.62** across three contrasts with **4/4, 3/5 and 5/6 mice**, respectively. Across 17 exploratory structural/functional tests, none survived Holm correction; the smallest adjusted P was 0.0696. Small samples leave meaningful positive effects plausible, rather than proving absence.

Conditioned medium from doxorubicin-treated Schwann cells impaired embryonic DRG outgrowth, directly supporting a transferable extracellular effect. However, the preparation involved **80 nM doxorubicin for 24 hours followed by nine days after washout**, and concentrated medium at up to **0.6 mg protein/mL**. The three-to-four reaggregates per condition do not establish independent donor or medium-preparation n. A mixed-medium result, normalized outgrowth 0.804, is compatible with an additive dose interpretation of 0.813; it does not require an emergent switch. These exposures and severed nerves differ sharply from intact early human diabetes. Global p16-cell removal is also not uniquely Schwann-specific. [Fuentes-Flores et al., 2023](https://doi.org/10.15252/emmm.202317907); [mouse analysis](../../../model_effort/integrative_model_v1/review/new_repair_senolysis_findings.md); [medium-dose analysis](../../../findings/sep12_schwann_public_comparison.md).

**A 2026 chemotherapy preprint strengthens the structural branch.** Malachowski and colleagues report that paclitaxel produces p16-associated dermal fibroblast states and loss of epidermal nerve profiles. Both global senescent-cell removal and a PDGFRα-lineage/p16 intersection improved structural outcomes. More informatively, deleting MK2 in the PDGFRα lineage preserved epidermal fibers while senescence-associated staining persisted. The corresponding structural comparison has **4/4 vehicle and 5/5 paclitaxel mice** across Cre controls.

This helps separate secretory regulation from eliminating the source cell. It is still not Q: MK2 was deleted in all targeted lineage cells, not only composite-state cells, and controls other outputs besides the listed secretome. The principal paclitaxel regimen was **50 mg/kg intravenously twice** in female mice, with skin assessed around day 9. Pooled single-cell sequencing used five mice per condition but one pool per condition. Human comparisons involved three controls and three CIPN cases for bulk RNA, with a later histology comparison of three and six; they lacked chemotherapy-exposed unaffected controls.

Drug prevention used **CDD2231**, whereas delayed pharmacological rescue used **ATI450**. Late global senolytic treatment also improved epidermal profiles. These are chemotherapy injury and repair results, not five-year diabetic prevention or specifically peptidergic terminal outcomes. Sarm1 knockout preservation with persistent fibroblast markers is compatible with downstream axon execution but does not by itself exclude parallel injury paths. Full text and legends were accessible; underlying figure values were not recovered, so no precise rescued fraction is inferred. [Malachowski et al., 2026 preprint](https://doi.org/10.64898/2026.08.04.742055).

**Human senolytic engagement is weaker than the original skin claim suggests.** Hickson's open-label diabetic-kidney-disease study used three days of dasatinib plus quercetin and repeat biopsies on day 14. Its correction withdrew the claimed significant skin-marker reduction after reanalyzing paired raw observations. Nine released skin pairs give p16 change **−0.40 cells/mm (−1.15 to +0.35)** and p21 change **−0.54 (−1.21 to +0.13)**. Corrected log-scale individual P values are 0.164 and 0.092; the corrected combined skin test is 0.118.

The results permit reduced markers, but do not establish the size of skin-state removal. Adipose and circulating-marker findings are not direct nerve engagement. Participants had advanced kidney disease, there was no concurrent control, and neither innervation nor incident DPN was measured. The correction is a concrete warning against treating marker responses as resolved prevention predictions. [Hickson et al., 2019](https://doi.org/10.1016/j.ebiom.2019.08.069); [2020 corrigendum and released observations](https://pmc.ncbi.nlm.nih.gov/articles/PMC6994619/).

**Broader state restoration remains distinguishable.** Partial OSKM reprogramming improved regeneration in aged mice after sciatic crush, but expression was systemic, not restricted to Schwann cells or secretion. Doxycycline induction was two days per week; structural assessment followed two or four weeks. Main methods and results differ on aged-mouse age, and the general “≥3 biological replicates” statement does not identify every animal denominator. This supports a broader repair possibility without isolating Q. [Wang et al., 2025](https://doi.org/10.1002/advs.202511019).

A 2026 conference abstract reports that imeglimin or NMN reduces high-glucose-associated β-galactosidase staining in an immortalized Schwann line, but supplies no independent n, complete dose/time specification or axonal structural endpoint. It adds plausibility to a metabolic trigger, not a completed causal chain. [Kato et al., 2026](https://doi.org/10.2337/db26-1435-P).

## Probabilities of the claims

Each conditional assumes that at least one group satisfies every predecessor; it asks whether any such group also meets the new requirement. The conditions are not independent marginal guesses. R3 tests a ≥5% contribution and R4 the additional ≥20% magnitude boundary; they do not charge the existence of structural causation twice.

| Claim | Exact additional requirement | Best conditional probability | Uncertainty range (sensitivity) | Main uncertainty |
|---|---|---:|---:|---|
| R1 | The composite state meets the occurrence and pre-loss timing boundary. | 60% | 25–90% | Human signatures and aged/injured states support occurrence, but not the exact early composite or order. |
| R2 | State cells supply the specified local secretory exposure. | 70% | 35–95% | Conditioned-medium and MK2 findings support an extracellular route; human local protein delivery and panel coverage are unmeasured. |
| R3 | Q yields B ≥0.05 in a preceding-qualifying group. | 55% | 20–90% | Structural repair/CIPN positives versus target specificity, diabetic transfer and protective secretory functions. |
| R4 | At least one such group has B ≥0.20. | 65% | 30–90% | Conditional on a real ≥5% contribution, whether the route accounts for a material fraction amid parallel injury. |
| R5 | At least one group satisfying R1–R4 has C ≥0.10. | 65% | 25–90% | Age and metabolic stress make broad exposure plausible; strong effects could be confined to uncommon states. |

The central product is **0.60 × 0.70 × 0.55 × 0.65 × 0.65 = 0.0975975**, reported as about 10%. A separate whole-claim assessment is “plausible but substantially less likely than not”: direct non-diabetic structural evidence prevents dismissal, while no study establishes the state, secretion, timing and material human prevention together. That supports the product's broad neighborhood, not its decimal precision. Five-point row differences are not empirically resolved.

## Ideal experiments that would resolve each claim

With unlimited measurement and intervention capability, follow the eligible population from the baseline observation through five years. Map the same connected cells, extracellular exposures and physical arbors without measurement injury. Use individually randomized Q versus reference care with verified source-specific interception, equal human weighting and complete observation. A separate matched nondiabetic cohort supplies reference distributions. This is an ideal truth test, not a proposed practical study.

**R1 — source state and time order.** Observe every component of the composite in the same cells over time and calculate the defined person-weighted fractions before the connected arbor's sustained deficit. YES: at least one candidate group crosses the R1 boundary. NO: none does. A high transcript score, one positive marker or state first appearing after denervation cannot produce YES.

**R2 — source and recipient exposure.** Among R1-qualifying groups, label each source cell's extracellular products and measure their local concentration/time profiles at connected recipient interfaces. YES: at least one group meets the two-protein, seven-day, magnitude and ≥20% source-attribution boundary. NO: every R1-qualifying group fails it. Intracellular synthesis without delivery is a negative for this exact exposure claim, even if intrinsic dysfunction exists.

**R3 — a nontrivial structural contribution.** Among groups satisfying R1–R2, compare true mean L₀ and L_Q with interception achieved throughout the defined sites and period. YES: B ≥0.05 for at least one group with positive L₀. NO: all such groups have B <0.05, or no positive reference loss. Pain relief without preserved physical anatomy cannot satisfy this claim.

**R4 — materiality.** Use the same structural comparison restricted to groups satisfying R1–R3. YES: at least one has B ≥0.20. NO: every one has B <0.20. Extra growth that subsequently disappears contributes no enduring net preservation.

**R5 — population coverage.** Measure incident DPN in the reference-care population, applying the original baseline group definitions. YES: at least one R1–R4 group accounts for ≥10% of those cases. NO: all such groups account for <10%. Case coverage is not the fraction of cultured cells responding.

Failed interception, incomplete source tracing, uncertain identity or follow-up do not establish these NO outcomes; they leave the corresponding claim unresolved.

## Uncertainty, alternative interpretations and updates

The following are complete, jointly defensible interpretations of the same bounded proposition. No empirical weights are assigned.

| Interpretation | R1–R5 conditional percentages | Whole probability |
|---|---|---:|
| Preferred | 60, 70, 55, 65, 65 | 9.76% |
| Lower: early composite uncommon; local products weak; repair results poorly transfer | 25, 35, 20, 30, 25 | 0.13% |
| Upper: early state and exposure common; structural secretory effects transfer and dominate locally | 90, 95, 90, 90, 90 | 62.33% |
| Strong local effect, uncommon case coverage | 65, 85, 80, 85, 25 | 9.39% |
| Common marker state, mostly small structural contribution | 85, 70, 40, 35, 85 | 7.08% |

The **0.1–62% sensitivity range**, roughly 62 percentage points wide, is not a confidence interval or a distribution over a hidden probability. Shared interpretation of the injury studies moves several rows together. The rare-effect and common-small-effect scenarios show why prevalence and magnitude cannot be inferred from the same signature. Sampling intervals for donor scores or skin markers address different uncertainties and cannot be relabeled as intervals on this hypothesis.

**Hypothetical upward evidence:** early-human source mapping plus a selective structural comparison supporting substantial benefit across a common baseline group could justify [85,90,80,80,80]%, giving **39.2%**. **Hypothetical downward evidence:** preserved composite states and secretion but little structural effect under verified Q, with any benefit restricted to rare groups, could justify [55,65,20,35,50]%, giving **1.3%**.

Holding other judgments fixed, moving R3 from 55% to 20% or 85% gives **3.5% or 15.1%**; certainty about R3 alone gives **17.7%**. That is an arithmetic ceiling under frozen other judgments, not expected research value. Real evidence can move multiple rows.

The state-marker, duration, protein-panel and 20% materiality boundaries define this proposition. Relaxing to one marker, adding unlisted secreted products or pricing arbitrary state reversal would change the hypothesis, not merely shrink its uncertainty. Existing data do not quantify how those changes affect human case coverage. There is no comparable set of prospectively resolved human structural predictions here from which to estimate a calibrated translational base rate.

## Overlap and boundary

This source-specific route can act through [inflammatory signaling](19_tnf_tlr4_inflammation.md), [skin repair](32_skin_repair_environment.md) or [SARM1 execution](27_sarm1_axon_destruction.md). Their interventions and required exposures differ, so none of these complete numerical propositions automatically implies another. [Epigenetic memory](40_epigenetic_memory.md) could stabilize the state but is not required. These probabilities cannot be summed into independent fractions of DPN.