Stem-Cell Exhaustion, Cellular Rejuvenation and the Translational Frontier of Geroscience: A Structured Critical Evidence Review of Human and Preclinical Interventions Targeting Biological Aging

Table 2: Representative human trials in aging-directed intervention research (registry values retrieved 7 September 2026).

Stem-cell exhaustion and cell-based regenerative interventions

Mesenchymal stromal cells in aging frailty are the largest human cell-therapy dataset directed at an aging phenotype, and the clearest illustration of the reporting problem. The published phase 1 was a non-randomised dose-escalation study in 15 frail patients given allogeneic bone-marrow MSCs at 20, 100 or 200 million cells intravenously, five per dose; the primary outcome, treatment-emergent serious adverse events at one month, occurred in none, and no clinically significant donor-specific immune reaction occurred within six months [26]. These findings derive from that 15-participant phase 1 cohort, not from the 65 participants recorded as actual enrolment for the parent registry record NCT02065245 [12]. Beyond that the record thins abruptly: the phase 2b of the same product enrolled 150 participants, reached primary completion in September 2021 and carries no posted results, and a phase 1/2 on vaccine response (62), an Alzheimer's extension (50) and a Vietnamese phase 1/2 (158) likewise have none [12]. For the three largest MSC-in-frailty studies — 150, 158 and 62 participants — no efficacy effect size is available from the registry, and none was verified from a peer-reviewed publication.

Haematopoietic stem cells supply the field's most striking recent preclinical rejuvenation result and its most instructive retraction. Antibody-mediated depletion of myeloid-biased HSCs rejuvenated aged immunity in mice, with functional challenge by vaccination and by pathogenic Friend retrovirus [27]. In the same compartment, Nature retracted in its entirety a 2010 report that systemic signals regulate ageing and rejuvenation of blood stem-cell niches, three authors requesting retraction after re-examination raised serious concerns with reported data on osteopontin-positive niche cells, the first author declining to sign [28].

Neural, muscle and epithelial compartments are preclinical throughout: neural stem-cell rejuvenation is supported across several niches, within a review that also records conflicting human hippocampal-neurogenesis findings [29]; ageing deregulates muscle satellite cells through cell-intrinsic and niche axes [30]; and aging reduces intestinal stem-cell self-renewal with reduced canonical Wnt signalling, while aged human organoids respond to WNT3a [31]. No human rejuvenation trial was identified in any of these three compartments, so the clinical marketing of stem-cell interventions for neurological and musculoskeletal aging has no corresponding human efficacy evidence.

Induced pluripotent stem cell–derived therapies

iPSC-derived products have reached patients for neurodegenerative disease, not aging, and no iPSC product with an aging indication was identified in the registry. The best-documented efficacy dataset is an open-label, single-centre, uncontrolled phase I/II trial of allogeneic iPSC-derived dopaminergic progenitors in Parkinson's disease (jRCT2090220384, UMIN000033564), in which seven patients were transplanted and six evaluated over 24 months [32]. Doses were 2.1–2.6 × 106 cells per hemisphere in three patients and 5.3–5.5 × 106 in four; adverse events occurred in 7 of 7, totalling 73 events, 72 mild and one moderate, with no serious event requiring hospitalisation or resulting in death. No tumorigenicity signal was detected on any measure applied over 24 months: no tumour-like enlargement, no increased 18F-fluorothymidine accumulation, no appreciable translocator-protein tracer uptake, and in corresponding rat grafts Ki-67 positivity below 1.0% with no overgrowth [32]. This is limited reassurance from seven patients over 24 months, not evidence that the tumorigenicity risk of pluripotency-derived products is resolved. The product was characterised as approximately 60% dopaminergic progenitors and 40% dopaminergic neurons with no TPH2-expressing cells — composition reporting that should be the norm and frequently is not.

Efficacy was modest and mixed: MDS-UPDRS part III OFF improved in 4 of 6 evaluable patients (mean −9.5 points) and total putaminal 18F-DOPA Ki rose 44.7%, with a dose gradient of +7.0% against +63.5%. Against these gains the Unified Dyskinesia Rating Scale total worsened by 12.3 points (+116.4%) in 6 of 7 patients, and potentially tacrolimus-associated adverse events occurred in 3 of 7 (42.9%), including two renal-impairment events [32]. The single-patient autologous precedent is instructive chiefly for its regulatory route, an FDA expanded-access IND [33]; eight further phase 1 or 1/2 iPSC-derived neural programmes were identified, none with an aging indication [12]. iPSC-derived therapy has therefore produced limited 24-month reassurance on tumorigenicity in seven patients; it has not demonstrated efficacy in a controlled design, its benefit carries a large increase in dyskinesia burden, and its immunosuppression requirement imports a quantifiable harm; extrapolation to an aging indication would be unjustified.

Partial epigenetic reprogramming

Preclinical. In vivo partial reprogramming ameliorates hallmarks of aging and prolongs lifespan in premature-aging mice [34], restores vision in glaucoma, optic-nerve-crush and aged-mouse models when OSK is delivered virally with c-Myc omitted [35], and shifts multi-tissue molecular profiles in wild-type mice, with explicitly small group sizes [36]. An inducible-epigenomic-change model further showed that faithful DNA repair itself advances aging at physiological, cognitive and molecular levels — including exdifferentiation, senescence and clock advancement — reversibly by OSK [37]. Two strategies address exposure rather than dose: restricting OSK to stressed and senescent cells through a Cdkn2a-promoter construct [38], and systemic dual adeno-associated virus serotype 9 delivery in 124-week-old mice at approximately 6 × 1013 vector genomes/kg [39]. A narrative review describes lifespan extension in progeroid and very old wild-type mice with frailty-index improvement, and no teratoma formation under prolonged induction [40]; it is used only for contextual synthesis, and no pivotal effect size rests on it. The lifespan, frailty and multi-tissue findings are attributed to their primary reports [34,36,39], none of which displays a complete set of group sizes and exact lifespan values publicly — which is why a review-level summary circulates in place of primary numbers. Two cautions follow: these are small, largely single-laboratory studies, and progeroid gains measure correction of a monogenic defect rather than modification of normative aging.

Human. Human partial reprogramming consists of one trial. Life Biosciences reports that its OSK platform's ER-100 is the first epigenetic-reprogramming therapy cleared to enter human trials, with authorisation to proceed on 15 January 2026 and the phase 1 beginning in early 2026 in open-angle glaucoma and non-arteritic anterior ischaemic optic neuropathy [41]. The registry record (NCT07290244) describes a recruiting phase 1 with 18 estimated participants aged 40–85, primary completion projected May 2027 [12]. This is the field's pivotal human data point, and it contains no efficacy information. Within the evidence retrieved for this review to 7 September 2026, no human efficacy data for partial epigenetic reprogramming were identified in any tissue or at any dose, and no retrieved source supports a claim that reprogramming has rejuvenated a human being. Because the search was not an exhaustive multi-database screen (Section 8.1), this is an absence in the retrieved record rather than a demonstrated absence in the world literature.

Cellular senescence, senolytics and senomorphics

Senescent cells enter an essentially irreversible cell-cycle arrest while remaining viable, accumulate with aging and frailty, and act largely through the SASP, which comprises pro-inflammatory, pro-apoptotic and pro-fibrotic factors in 30–70% of senescent cells; above a threshold, self-amplifying spread is proposed to outpace immune clearance, producing inflammaging [42]. The counterweight is explicit in the hallmark’s synthesis: limited, spatially confined senescence suppresses oncogenesis and improves wound healing, so senolysis is a trade-off rather than a subtraction [1].

Dasatinib plus quercetin shows a consistent pattern across four indications — feasibility met, mechanism partially demonstrated, clinical endpoints not met (Tables 1 and 4) [43,44,45]. The most instructive is an open-label phase 1 in mild Alzheimer's disease with five completers, in which quercetin was not detected in cerebrospinal fluid at all, cognitive and neuroimaging endpoints did not differ from baseline, and cerebrospinal-fluid interleukin-6 and glial fibrillary acidic protein both increased (P = 0.008 and P = 0.028) [46] — a warning against assuming that a peripherally validated senolytic mechanism transfers to the brain.

The most rigorous senolytic trial here is a 20-week open-label randomised phase 2 in postmenopausal women, enriched by T-cell p16INK4a messenger RNA above the 95th percentile of young controls, under IND 145558 [25]. The primary endpoint was missed: CTx at 20 weeks changed by −4.1% against −7.7% in controls, P = 0.611. P1NP rose 16% at two and four weeks before falling at 20 weeks, and in the highest senescence-burden tertile P1NP rose 34% at two weeks and radius bone mineral density rose 2.7% at 20 weeks (P = 0.004). Adverse events occurred in 77% of treated women against 17% of controls, with one withdrawal for QTc prolongation [25]. The National Institute on Aging's summary records no between-group difference in either bone marker by 20 weeks [47].

A phase 2 of fisetin in nursing-home residents was terminated, the registry recording that the data and safety monitoring board and the National Institute on Aging stopped it for futility (NCT04537299), while other fisetin trials remain active [12]. The intravitreal BCL-xL senolytic programme, analysed in Section 4.4, carries an on-target class liability, since BCL-xL inhibition causes thrombocytopenia, which is why proteolysis-targeting chimeras were developed and a local route chosen [48]. No human senomorphic trial was verified; the class is a gap in the human record rather than a body of negative evidence.

Nutrient-sensing pharmacology

mTOR inhibition. The strongest preclinical dataset in geroscience is the multi-site heterogeneous-mouse lifespan programme, in which microencapsulated rapamycin at 42 ppm begun at 20 months extended median lifespan by 11% in males (P = 0.0007) and 15% in females (P < 0.0001), while a three-month exposure gave 11% in males (P = 0.024) but only 4% in females (P = 0.15, not significant) [7]. Two features matter as much as the headline: four comparator compounds on the identical platform were null (17-DMAG +9%, P = 0.12; minocycline +5%, P = 0.99; β-guanidinopropionic acid −1%, P = 0.73; MitoQ +2%, P = 0.91, all in males), and site heterogeneity was large, the male cycled-rapamycin median change ranging from −5% to +41% across three sites [7]. A double-masked randomised placebo-controlled canine trial is under way but embargoed [49].

The human record is thinner by orders of magnitude: a published systematic review included 19 articles from 11,717 deduplicated records, rating risk of bias high in four and of some concern in 11 [17]. The largest randomised rapamycin healthspan trial analysed 114 completers over 48 weeks under a direct-to-consumer sponsor, and its interpretation is dominated by product definition, since the compounded drug achieved approximately one-third the blood concentration of commercial rapamycin at 24 hours; the primary endpoint was not met (visceral adipose tissue, P = 0.942), a lean-tissue subgroup finding in women on 10 mg carried a confidence interval spanning 2.42 to 323.7 and so estimates nothing, and no other significant effects were observed [23].

The decisive human lesson comes from a different agent, and it is taken here from the primary randomised report rather than from commentary. In the phase 2b of the mTORC1 inhibitor RTB101, the proportion of participants with one or more laboratory-confirmed respiratory tract infections was lower on RTB101 10 mg once daily than on pooled placebo (34 of 176, 19%, versus 50 of 180, 28%; odds ratio 0.601, 90% confidence interval 0.391 to 0.922, P = 0.02), while the RTB101-plus-everolimus comparison did not reach significance [14]. FDA then requested a change in primary endpoint before phase 3, on the stated concern that laboratory confirmation of an infection was not relevant to how patients feel and function [16]; in the phase 3, RTB101 10 mg once daily did not reduce clinically symptomatic respiratory illness (134 of 511, 26%, versus 125 of 510, 25%; odds ratio 1.07, 90% confidence interval 0.80 to 1.42, P = 0.65) [14]. The phase 2a predecessor had shown reduced reported infections and improved influenza-vaccination response with low-dose TORC1 inhibition [15], so target engagement was not the failure point.

Other nutrient-sensing agents contribute one methodological lesson each (Table 1). Metformin altered skeletal-muscle transcriptome adaptations to resistance training — an interference signal, not a benefit [50] — and over two years improved a frailty index (P = 0.0222) while remaining null on the Fried criteria in the same randomised population 51: a discordance between instruments rather than a drug failure. The publication does not state a registration number; the registry record NCT02570672 is consistent with, not confirmed identical to, the published trial, and is described that way throughout [12]. No registry record was retrieved for the Targeting Aging with Metformin trial, and no facts about it are asserted [12]. Nicotinamide riboside is safe to 3,000 mg/day but unproven on any clinical endpoint, its first Tier 1 test reporting from December 2027 [12,52,53]. Caloric restriction at 25% prescribed, achieving 11.9%, lowered weight and lipids [54] yet produced discordant clocks in one randomised sample — principal-component PhenoAge and GrimAge null, DunedinPACE d = −0.29 and −0.25 — with the investigators stating that effects were small and a conclusive test requires long-term disease and mortality follow-up [55,56]. Spermidine was tested at approximately 0.9 mg/day, orders of magnitude below rodent regimens; the primary cognitive outcome and its P value were not captured in the evidence set assembled for this review, so no efficacy conclusion is drawn, and an exposure at that level would not test the mechanism in any case [57]; and taurine's founding premise was refuted by longitudinal cross-species data showing that circulating taurine often increased or remained constant with age, although trials proceeded and results are unposted [12,58,59].

Systemic milieu, telomeres, and immune and thymic rejuvenation

Plasma. Therapeutic plasma exchange with 5% albumin replacement was studied in individuals aged 46–77 with proteomics of 507 proteins; complement C3 and C1q fell by roughly half immediately after a procedure but returned to baseline about a month later, with no significant lasting effects across three rounds [60]. The study is uncontrolled, small and heterogeneous and carries a published correction; its title claims biological-age reduction while its strongest findings are proteomic. FDA stated in February 2019 that establishments were offering young-donor plasma infusions at up to thousands of dollars per infusion for indications ranging from normal aging to dementia, and that administration outside the recognised Circular of Information requires an active IND [61].

Telomeres. Telomerase gene therapy in aged mice was reported to delay aging without increasing cancer [62], but Mendelian randomisation in 379,758 UK Biobank participants found genetically determined telomere length associated with age-related outcomes in both directions, so "longer is better" is not defensible [63]. The human translational record here is dominated by an integrity failure and a governance failure. A 2022 report of intranasal and injectable gene therapy for healthy life extension was retracted after the sponsoring university requested it following internal review of data discrepancies; the article discloses that the work was fully funded by a commercial sponsor owning the pending patent, with an author on that sponsor's board, and reporting on the retraction records that in 2021 the company sent six dementia patients to Mexico for injections of the experimental treatment [64,65]. Separately, a registered AAV-hTERT study for the treatment of aging in Colombia (NCT04133649) was last verified in December 2019, remains listed as recruiting, and carries blank enrolment, allocation, masking and primary-purpose fields — an activity styled as a trial that satisfies none of the ISSCR recommendations discussed in Section 5.3 [12].

Immune and thymic rejuvenation. The basis for the claim that epigenetic age has been reversed in humans is a single uncontrolled study of nine men aged 51–65 given growth hormone, dehydroepiandrosterone and metformin, with no control group, no stated blinding and no preregistration [66]. Reported outcomes were a mean epigenetic age change of approximately −1.5 years after one year and improved thymic fat-free fraction in seven of nine participants, against unusual baseline clock offsets (Levine PhenoAge −17.5 ± 0.98 years) [66]. The design cannot support the claim built upon it: nine uncontrolled participants, a multi-drug protocol in which growth hormone is the agent most plausibly responsible for the imaging change, no preregistration, and baseline offsets that make regression toward expected values a live alternative. The extension study lists completion in December 2025 with no results [12]. Thymic biology otherwise remains preclinical for aging purposes, allogeneic thymus transplantation for congenital athymia being the only existing replacement therapy and iPSC-derived thymic epithelial cells the proposed path [67]; a registry search returned three thymus-regeneration records, two terminated [12].

Extracellular vesicles and adjacent modalities

Extracellular vesicles combine substantial preclinical rationale, no aging efficacy evidence, an explicit regulatory prohibition on marketing, and a large commercial marketplace; as of September 2022, the landscape comprised seven published clinical studies and 14 ongoing trials in non-aging indications [21]. The central methodological finding should govern all EV reporting: there is no consensus on MSC-exosome dose, because doses were reported variously as micrograms, particle numbers, or numbers of parent MSCs [21]. None is a potency assay and none is interconvertible, so two preparations nominally given at "the same dose" may differ in activity by an unbounded factor. The regulatory position is unambiguous: FDA stated in a public safety alert posted 9 December 2019 that exosomes used to treat diseases in humans are regulated as drugs and biological products subject to premarket review, and that there are currently no FDA-approved exosome products, citing multiple reports of serious adverse events in patients treated with unapproved exosome products [68]. Despite this, 6.68% of 1,480 United States businesses selling purported stem-cell treatments marketed exosome products as of 31 March 2021 [69].

Adjacent modalities. Aging-directed gene therapy in humans is confined to the ocular OSK phase 1 (NCT07290244), the patient-paid AAV-hTERT record (NCT04133649) and a 12-participant AAV-follistatin study (NCT07443826); mitochondrial transplantation exists only in acute injury [12], and a mitochondria-targeted antioxidant was null in the reference lifespan platform [7]. Faecal microbiota transplantation extended healthspan and lifespan in two progeroid mouse models [70], yet no human trial with an aging or frailty endpoint was identified, although dysbiosis remains a formal hallmark [1]. Xenotransplantation reached formal trials after FDA approved the first clinical trial of kidneys from genetically modified pigs [12,71], but addresses organ supply rather than aging biology. Finally, the one artificial-intelligence-discovered drug with randomised human data in an age-related disease, a TNIK inhibitor in idiopathic pulmonary fibrosis, was powered for safety rather than efficacy: its primary endpoint was the percentage of patients with at least one treatment-emergent adverse event, dose-limiting hepatotoxicity produced liver-related discontinuations of 22.2% and 17% in the two highest-exposure arms, and the forced vital capacity difference favouring active treatment at 12 weeks was exploratory [72].

Biomarkers, clocks and the endpoint problem

The first multi-tissue methylation age predictor yielded 353 clock CpGs and was proposed as a surrogate marker for evaluating rejuvenation therapies [73]; second-generation measures instead predict time to death or functional decline, and all clocks fall on a spectrum between chronological and biological components that has proved difficult to separate [74,75]. Three findings determine how clocks should be treated as endpoints: no consensus exists on validating biomarkers of aging before clinical translation [10]; technical noise materially limits reliability, which is why principal-component versions were developed [8,9]; and three well-known clocks disagreed about the same randomised intervention in the same participants [55]. The defensible position is narrow: epigenetic clocks are validated predictors in observational settings and are not validated surrogate endpoints for intervention efficacy. Their appearance as primary or co-primary endpoints in registered trials — including a topical rapamycin study titled around re-setting the epigenetic clock (NCT04608448) and plasma-exchange, alpha-ketoglutarate and taurine studies with biological-age endpoints [12] — is a choice no regulator has accepted and the field's own methodological literature does not support [10,76].

The construct with the strongest claim to regulator-legibility is different. The WHO ICD-11 process withdrew "old age" as a category title and index listing, replacing it with "ageing associated decline in intrinsic capacity" and replacing "pathological" with "biological" in the associated extension-code definition, with accompanying commentary stating that ageing is not a pathological process and that old age is not a disease [77]; the 2026 expert consensus submitted to FDA converges on the same construct [4]. That convergence is not regulatory acceptance and no such claim is made here, but it is the most coherent endpoint proposal currently on the table, and empirical support for its enrichment component already exists in the senolytic bone trial's tertile findings [25].

Null, failed and retracted evidence

Because null results are frequently omitted from secondary accounts, they are consolidated here and in (Table 4) with the prominence given to positive findings. Two publications were verified as retracted [28,64,65], and one foundational biomarker claim was refuted by longitudinal cross-species analysis [58,59]. Primary endpoints were missed in a senolytic bone trial [25], a rapamycin healthspan trial [23], and a phase 3 mTOR-inhibitor trial [14]; an intravitreal senolytic phase 2b was discontinued with results never posted, and one trial stopped for futility [12]; and a senolytic phase 2 in knee osteoarthritis was reported in the trade press to have failed against placebo, an account recorded but not treated as trial evidence [78]. Caloric restriction was null on two of three epigenetic clocks [55]; metformin was null on one of two frailty instruments in the same trial while significant on the other, so it is recorded here as mixed rather than negative [51] and interfered with resistance-training adaptation in another [50]; four compounds were null on the reference mouse lifespan platform [7]; dyskinesia burden more than doubled in the best-documented iPSC efficacy dataset [32].