If plasma exchange helps, is the effect from removal or replacement?

Heterochronic parabiosis shares circulation and another animal's organ clearance/secretion. Young-plasma infusion adds material; therapeutic plasma exchange (TPE) removes and replaces it. Their positive findings are not independent replications of one mechanism. This article asks which controls identify the source of an effect. s1

01Young plasma is unnecessary for one combination

Mehdipour's 2020 mouse experiment replaced approximately half the platelet-rich plasma with saline, commercial 5% mouse albumin and age-matched blood cells. Relative to isochronic young–young or old–old exchange, selected muscle repair, liver fat/fibrosis and hippocampal proliferating-cell measurements improved six days after one intervention. This supports that replacement combination without young plasma, not a specific removed factor or long-term healthspan. s2

Albumin was tested. Adding it to pre-exchange serum from four older people did not rescue cultured muscle-progenitor proliferation; post-exchange serum supported proliferation. However, albumin supplementation promoted neural-progenitor proliferation. Both findings matter: the muscle assay does not support albumin supplementation as sufficient to explain that response, but cannot exclude in-vivo or other-cell contributions. Albumin binding/transport properties, platelet-related material and removal remain incompletely separated. s2

Six wells are not six people; two injury sites are not two independent mice. The muscle figure includes injury-site observations; tissue experiments involved approximately 4–8 mice per group. Results and methods also describe ages differently. Complete animal/donor tables were unavailable for clustered reanalysis. Failure to differ from young controls is not equivalence; Ki67-positive cells do not establish mature new neurons, restored memory or longer life. s2

02The human comparison

Fuentealba's 2025 study compared Sham with three albumin-replacement TPE schedules: two sessions during the first week of each month with or without IVIG, and one session per month. There was no young-plasma or IVIG-only arm. Sampling before sessions 1, 4 and 6 reflects three and five completed treatments, not persistence after completing six or stopping. Identically named visits across schedules need not mean equal calendar time. s3

The title and registry call allocation randomized, but methods describe first-come-first-served assignment, filling shorter schedules before monthly/placebo groups, without a verifiable sequence or concealment method. Primary completion is November 30, 2023; first public registration is August 2, 2024, which cannot establish prospective specification. Masking descriptions also differ. These limits do not establish fabrication; they prevent assuming verified randomized causal effects. s3 s5

The flowchart has 44 allocated and 42 analyzed, the demographic table counts 41, and the registry and released clock matrix contain 40. The matrix has ten people per arm and 35 algorithms, lacking Intrinsic Capacity from the final 36-algorithm version. Linkage checks passed, but only five Sham participants have visit 3. The non-IVIG short-schedule arm has 9, 8 and 10 observed participants at visits 1–3, with seven complete triples. Denominators are not interchangeable. s3 s4

There was no formal power calculation, and functional, cognitive or symptom outcomes were not assessed as predefined endpoints. Missing these outcomes is not evidence of no functional effect. Participants were described as generally healthy and mostly over 50, with one woman in her forties enrolled exceptionally; paper participant A4 was taking rapamycin. That identifier cannot be verified against released matrix labels, so no fabricated drug-exclusion sensitivity analysis is possible. s3

03Two leaps inside “2.61 years younger”

First, scales: original DNAmGrip, DNAmGait and DNAmVO2max estimate strength, speed and oxygen uptake. DNAmTL was developed in kilobases; age-regression residuals do not become years. DNAmFitAge itself is an age output. Multiplying Intrinsic Capacity residuals by −100 to obtain a similar numerical scale does not establish common units. Obtained methods/code do not sufficiently verify conversion of every component to a common direction and year scale. Raw TruAge inputs and IDATs were unavailable; we did not regenerate EAA. s3 s4 s6 s7

Second, the independent unit: the released script compares 35 algorithm-specific group means from the same participants and correlated blood information. Our technical calculation gives means of −2.605 for IVIG and −1.318 for monthly TPE at visit 2, close to headline means; the IVIG–Sham six-comparison Holm P is 0.0000643. Reproducing that number validates neither common age units nor algorithm independence. Final 35/36-algorithm differences and missing inputs including iage_base.rds prevent claiming full pipeline reproduction. s4

04Participant-level comparisons

We froze 35 algorithms × two baseline-to-follow-up comparisons × three active arms against Sham: 210 contrasts, preserving scales. We report Welch differences in individual paired changes and baseline-adjusted differences from follow-up EAA regressed on treatment and baseline EAA. The latter use HC3 variance and residual degrees of freedom for pointwise 95% intervals. Each method receives separate Holm correction over 210 tests. These are conditional complete-case comparisons, not corrections for allocation, calendar time or unmeasured confounding. s4

Figure 1. Displayed algorithms were chosen before calculation. Monthly SystemsAge at visit 3 has a nominal difference of −6.85 (−13.28 to −0.43); it must not disappear because other intervals cross zero. With five Sham participants, it does not survive family adjustment. Every algorithm and both methods are available.

Ten direct-change and nine baseline-adjusted contrasts have unadjusted P<0.05; none meet 0.05 after Holm correction. The latter also include higher Horvath values. Our wider family differs from individual author tests and does not “refute” every result. Wide intervals and mixed directions establish neither zero effect nor consistent, durable whole-body rejuvenation.

Figure 2. Each follow-up uses available complete pairs; connecting means does not create a fixed-cohort trajectory. Missingness can alter composition.

05Does negative correlation identify compensation?

The paper proposes compensation from negatively correlated successive changes. For the released 35-algorithm numerical mean, the non-IVIG short-schedule arm has seven complete triples and Spearman ρ=−0.786. But changes share a middle measurement. With independent visit errors, Cov(e₂−e₁,e₃−e₂)=−Var(e₂); equal error variances give correlation −0.5. Without calibrated repeat-measurement error, this correlation cannot uniquely identify feedback. Nor does the algebra prove the observed pattern is entirely error. s3 s4

Figure 3. Self-chosen, uncalibrated error variances; a mathematical alternative, not biological falsification.

Released cytomics include 36 features and 36 people, one without a clock-group link. All released features remain descriptive. The IVIG arm's mean paired CD4 change is −5.73 percentage points (nine people), versus +1.34 in Sham (ten), without a mechanistic test. Flow cytometry uses isolated, frozen and recovered PBMC; gated frequencies are not exact whole-blood DNA fractions, and overlapping subsets cannot sum to 100%. Immune remodeling and cell-mixture effects on methylation can coexist; correlated omics do not independently establish mediation or improved infection resistance. s3 s4

06Does sustained low concentration establish resetting?

In a constant-volume single-compartment model dC/dt=P−kC, starting equilibrium 1 and P=k, removing half gives C(t)=1−0.5e^(−kt). Repeated removal each half-life keeps concentration low without resetting feedback; stopping allows recovery. Two chosen models also yield the same post-removal 0.5 and subsequent 0.75 observation yet approach different limits, 1 and 0.8. Sparse concentration data cannot uniquely identify production, clearance or lasting feedback. This is a turnover counterexample, not a real-protein half-life, clinical dose or clock fit.

This model does not cover all original findings. Many 2020 antibody-array readings increased six days after mouse exchange and one month after human TPE, supporting the authors' hypothesis that removal releases production from inhibition. Constant production in our model only recovers from below and cannot exceed the original equilibrium; reliable absolute concentrations persistently above it would reject this model for that protein. The arrays report background/internal-control-normalized relative readings, not production flux. Increased production, altered clearance and tissue release require separation. Increased readings must not be erased by “natural recovery,” but do not uniquely establish the proposed autoinductive feedback reset. s2

Figure 4. Dimensionless time: A uses k=P=ln2; B uses k=ln6, P=0.8k. Both start at 0.5 after removal and equal 0.75 at t=1, but have different limits. Actual mechanisms require time courses, flux/tracers and targeted perturbation/rescue.

07Disease outcomes must remain in their population

AMBAR randomized 347 people with mild-to-moderate Alzheimer's disease; 322 received treatment. At 14 months, pooled active arms differed by +3.5 ADCS-ADL points (95% CI 0.4 to 6.6; P=0.03), but −2.1 ADAS-Cog points (−4.4 to 0.2; P=0.06). The functional signal and overall cognitive trend differ; both coprimary outcomes were not conclusively positive. Albumin and some IVIG remained part of treatment. This does not identify removal or establish healthy-person rejuvenation/longevity. Attrition, adverse events and industry funding remain; individual data were not reanalyzed. s8

PLASMA included 18 people: nine randomized in a crossover and nine subsequently treated openly. Safety/feasibility were primary. ADAS-Cog crossed zero, while two caregiver-rated functional scales showed exploratory improvement without multiple-testing correction. Neither selectively reporting the cognitive null nor treating function as confirmed efficacy is justified. s9

A 2026 Norwegian young-donor exchange pilot included 12 people with AD-related mild cognitive impairment. Unblinded, uncontrolled and heterogeneous protocols produced no consistent direction in cognitive, grip, lung-function or imaging trajectories. “Interstitial rejuvenation” explicitly describes an intention: molecular markers did not confirm interstitial exchange. Two grade-4 events were judged unlikely related; three people had generalized urticaria requiring interruption of a session. Feasibility can justify further study; low assessed causality does not mean no serious events or established safety and efficacy. s10

08Comparisons that move attribution forward

On October 7, 2026, the 60-person ExPlas exercised-donor infusion trial (NCT05068830) is active, not recruiting. s13 The 34-person frailty albumin-exchange trial NCT07836127 is not yet recruiting, with a frailty-response primary endpoint one month after final treatment. s12

None of six reviewed registry records has posted results, including completed NCT05004220. s5 s15 The published 2026 pilot’s NCT06234436 record remains UNKNOWN without posted results; the older NCT03353597 also says UNKNOWN. s14 s16 No posted results does not mean no papers; UNKNOWN does not establish ongoing treatment. Registration identifies plans, not benefit.

FDA's December 2024 update warns that young plasma has no demonstrated clinical benefit for ageing/memory uses and lists infectious, allergic and respiratory risks. This concerns young-plasma infusion, not every disease indication for TPE. s11 The June 2025 corrected safety section reports one mild albumin reaction in 240 procedures, while the abstract still reports two events requiring discontinuation and the flowchart gives different withdrawal reasons. Six sessions per completed active participant do not reconcile with 240. Unresolved denominator/version differences prevent converting a low procedure fraction into low individual long-term risk. s3

Specified replacement combinations can change selected tissue and blood measurements; disease trials contain clinical signals worth following. Unique removal mechanisms, durable resetting and ordinary-human healthspan benefit still require direct evidence. The calculation package preserves separate algorithms and chosen models. The same Codex agent researched, translated and self-reviewed this article; there was no independent or professional medical review.