Skip to main content
Frontiers in Endocrinology logoLink to Frontiers in Endocrinology
. 2026 Nov 2;17:2009751. doi: 10.3389/fendo.2026.2009751

Gut bacterial extracellular vesicles in glucocorticoid-associated skeletal complications: evidence boundaries and translational challenges

Zhe Liu 1,†, Xiuli Wu 1,†, Shengjie Zhang 1, Longfei Li 1, Lijing Zhao 1, Huimin Guo 1, Nan Sun 1, Jinju Li 1, Baoxia Sun 1,*, Xiaoxia Yan 1,*
PMCID: PMC13773003  PMID: 43042294

Abstract

Long-term or high-dose glucocorticoid (GC) therapy causes major skeletal complications, particularly glucocorticoid-induced osteoporosis (GIOP) and glucocorticoid-associated osteonecrosis of the femoral head (GIONFH). Bacterial extracellular vesicles (BEVs) produced by gut-resident bacteria are candidate inter-organ messengers, but the GC-specific evidence base is extremely limited. In an updated structured narrative search executed on 11–13 October 2026 and bounded to publications through 15 September 2026, one eligible in vivo dataset was identified that jointly incorporated GC exposure, administration of bacterially derived vesicles, and skeletal phenotyping: cultured Lactobacillus animalis-derived BEVs attenuated GC-induced osteonecrosis in mice. Within our operational intestinal-associated bacterial-source scope, 14 primary reports were retained: one directly evaluated GC-associated GIONFH and 13 addressed non-GC skeletal conditions. The core count is bounded by this source criterion and the documented search, rather than representing all bacterial OMV studies with skeletal outcomes. We classify the literature along two independent axes—vesicle source (bacterial versus host) and skeletal context (GC versus non-GC)—while keeping hypothesis-generating mechanisms, non-vesicular microbial products, non-skeletal vesicle evidence, and source-unresolved studies outside the matrix. We also distinguish community-derived gut microbiota-derived extracellular vesicles (GMEVs), cultured commensal-derived BEVs (cBEVs), and engineered BEVs, because these preparations are not biologically interchangeable. The anchor study is critically appraised, and GIOP is separated from GIONFH rather than treated as a single disease entity. Current direct bacterial-vesicle evidence supports proof of concept for GIONFH only; no eligible direct GC–bacterial-vesicle dataset for GIOP was identified within this search and source framework, and no eligible human study directly linking source-resolved circulating GMEVs to GC-associated skeletal outcomes was identified within this search framework. The principal translational barriers are source attribution; isolation and analytical standardization; biomarker analytical and clinical validation; biodistribution, PK–PD, and therapeutic delivery; safety, immunogenicity, and scalable manufacturing; and interspecies and human validation. Gut bacterial EVs should therefore be regarded as candidate mediators of a GC-associated gut–immune–bone–vascular axis rather than established drivers of skeletal injury. Immediate priorities are independent replication, testing of human-prevalent donor taxa, and prospective source-resolved clinical cohorts.

Keywords: bacterial extracellular vesicles, bone metabolism, glucocorticoid-induced osteoporosis, glucocorticoids, gut microbiota-derived extracellular vesicles, gut–bone axis, osteonecrosis of the femoral head, source attribution

1. Introduction

Glucocorticoids remain foundational therapy for autoimmune disease, allergy, and transplant rejection, yet chronic or high-dose exposure is a leading cause of iatrogenic skeletal harm. GC treatment increases fracture risk in a manner influenced by cumulative exposure, dose, baseline susceptibility, age and fracture ascertainment; estimates are not directly comparable across treatment populations (1–3). GIONFH remains among the most disabling skeletal complications and can require arthroplasty in patients of working age (4, 5). Current management mitigates but does not reverse these outcomes (6, 7). Contemporary guidance emphasizes early fracture-risk assessment, bone mineral density and vertebral imaging, and risk-adapted anti-osteoporosis therapy in patients receiving prolonged glucocorticoids (8). Glucocorticoid excess is also a prototypical endocrine perturbation of skeletal remodeling, acting directly on osteoblast, osteocyte, and osteoclast lineages while altering systemic metabolic, immune, and gastrointestinal signals. This endocrine context provides the rationale for examining whether intestinal-associated bacterial extracellular vesicles contribute to GC-associated skeletal complications beyond the canonical suppression of bone formation and stimulation of resorption.

The “gut–bone axis” has reframed metabolic bone disease as a system in which microbial metabolites, immune signaling, and endocrine outputs from the gut influence skeletal homeostasis (9–13). Within this framework, GMEVs have gained attention as a previously underappreciated communication channel (14–17). GMEVs carry lipopolysaccharide (LPS), peptidoglycan fragments, lipoteichoic acid, porins, adhesins, and small non-coding RNAs (18–20), and vesicles of gut bacterial origin can be detected beyond the intestine, indicating that translocation into the circulation is possible (21–23).

Despite this conceptual appeal, the GC-specific bacterial-vesicle literature is sparse. The 2022 Science Advances study by Chen and colleagues provides the field’s most direct anchor: GC exposure depleted intestinal L. animalis and its EVs, and oral replenishment of cultured L. animalis-derived BEVs alleviated GC-induced osteonecrosis of the femoral head (24). This dataset defines a proof-of-concept link between GC exposure, a gut commensal-derived vesicle preparation, and skeletal/vascular phenotypes; it does not establish that naturally occurring community-derived GMEVs are causal mediators in humans.

Three questions motivate the synthesis that follows. First, what has actually been demonstrated about gut bacterial EVs as inter-organ signals under GC exposure? Second, does the evidence differ between GIOP and GIONFH? Third, what source-attribution, mechanistic, and translational barriers must be overcome before bacterial-vesicle biomarkers or therapeutics can be evaluated clinically?

How this review differs from existing syntheses. Several recent reviews cover adjacent ground. Liu et al. proposed three gut–bone axes (traditional, bacterial-EV-based, and intestinal-organoid-EV-based) as an integrative framework for osteoporosis (14); Cheung et al. surveyed host-derived and gut-microbiota-derived EVs as nanocarriers for osteoporosis and osteoarthritis (16); Liang et al. reviewed oral and gut bacterial EVs in bone health with emphasis on the oral–gut–brain–bone axis (25); Yin et al. compared mammalian, plant, and microbiota-derived vesicles as therapeutics across bone diseases (17); Han et al. traced microbiota-derived EVs from host–microbe interaction to therapeutic application (15); and Parrotta et al. synthesized experimental and bioinformatic evidence for EVs in osteonecrosis of the femoral head, where the dominant vesicle sources are host cells (26). Two 2026 Bone Research reviews additionally synthesize the diet–gut microbiome–osteoporosis literature (27) and the broader mechanisms of steroid-induced osteonecrosis (28), respectively, but neither isolates the directness of bacterial-vesicle evidence under GC exposure. The present review therefore narrows the question to GC-associated skeletal injury and separates evidence by two independent dimensions—vesicle source and disease context—while holding non-vesicular, non-skeletal, source-unresolved, and purely hypothesis-generating evidence explicitly outside the matrix. Its organizing contribution is not another catalogue of pathways, but a critical delineation of where GC-specific bacterial-vesicle evidence actually ends. A side-by-side comparison is provided in Supplementary Table S1. The operational core is restricted to intestinal-microbiota-associated bacterial EV preparations; gastric- and oral-source OMV studies are considered adjacent rather than being silently merged into the core (Section 2.3; Supplementary Table S2C).

2. Literature search and evidence classification

Because the central claim of this review concerns how directly the available evidence addresses the GC–bacterial-vesicle–bone question, we state explicitly how studies were identified, how they were appraised, and how individual preparations and claims were assigned to the nonordinal evidence-relevance matrix. The direct GC-focused scope concerns exogenous pharmacologic exposure; endogenous hypercortisolism is discussed only as biological context and does not enter the direct evidence set without a qualifying study.

2.1. Search strategy and screening

This article is a critical narrative review supported by a structured, documented literature-identification, screening, and evidence-classification process, rather than a systematic review or meta-analysis. One author conducted the final manual screening and study-level evidence verification. No protocol was registered; duplicate independent screening, quantitative pooling, and formal study-wide risk-of-bias scoring were not performed. Database-specific strategies, eligibility criteria, core-study extractions, and prioritized-record dispositions are provided in Supplementary Tables S2A-S2C. Search counts are reported for traceability, not as a PRISMA study-selection flow or as evidence of exhaustive systematic-review coverage.

PubMed, Web of Science Core Collection, and Scopus were searched for publications first formally published on or before 15 September 2026. Web of Science and Scopus searches were executed on 11 October 2026 and PubMed searches on 13 October 2026. The database-specific P1–P4 strategies combined glucocorticoid, bacterial/microbiota, extracellular-vesicle, and skeletal concept blocks in complementary ways (Supplementary Table S2A). No lower publication-date eligibility limit was imposed, although indexing coverage varied by database. The searches retrieved 11,708 source records; 16 first formally published after the cutoff and three preprint-only records were removed before de-duplication. Of the remaining 11,689 source records, 2,511 duplicate entries were merged, leaving 9,178 unique publication records. All 1,537 records in the P1–P3 union underwent manual title/abstract screening by one author. The other 7,641 P4-only records were retained as an auxiliary omission-check pool and were not exhaustively screened one by one. Targeted searches on GC-associated intestinal changes and bacterial-versus-host EV attribution, together with reference-list and citation checking, were also used to inform contextual discussion and identify potential omissions.

From the P1–P3 screening work set, 41 prioritized candidate records underwent further author-performed eligibility assessment based on potential relevance to intact bacterial EVs, skeletal outcomes, and the GC-focused question; their individual dispositions and available full-text status are documented in Supplementary Table S2C. One author reviewed the complete texts and verified the S2B study-level extractions for all 14 core studies. Of the 41 prioritized records, 14 were retained as core studies, 18 were assigned to adjacent/contextual evidence, and nine were held outside the matrix. These are dispositions within the prioritized subset, not a claim that all 9,178 deduplicated records received full-text review. The single-reviewer process, English-abstract restriction, and incomplete manual coverage of the P4-only set limit search sensitivity and are addressed in the limitations.

Core inclusion was restricted to published primary in vitro, animal, or human studies testing isolated intact bacterial EVs (or an EV-enriched preparation with credible bacterial contribution) and reporting an experimental skeletal or bone-cell endpoint. To maintain the review’s intestinal-microbiota focus, eligible producers were intestinal-community-derived preparations or cultured bacterial taxa with an established intestinal ecological association, 7including probiotic chassis and intestinally associated opportunists; cultivation or such an association alone was not treated as proof that administered EVs originated from the intestine in vivo. Predominantly gastric- or oral-source bacterial OMV studies were considered adjacent evidence rather than core intestinal-associated preparations. This was an operational anatomical/source boundary, not an exclusion based on pathogenicity or study quality (Rule 1, Section 2.3). Contextual reviews were cited but were not counted as core studies; records without an English-language abstract were excluded at the record-eligibility stage. The resulting operational core comprises 14 reports: one direct GC-associated GIONFH study and 13 non-GC skeletal studies. Four reports identified or confirmed in the eligibility-closure exercise addressed P. mirabilis OMVs (29), B. pseudolongum sEVs (30), engineered EcN BEVs displaying BMP-2/CXCR4 (31), and bone-targeted PM-OMVs (32). The latter report contains inconsistent producer nomenclature (published as ‘Pseudomonas mirabilis’, whereas its cited ATCC 12453 strain is catalogued as Proteus mirabilis), and source attribution is accordingly flagged rather than silently corrected. A 2026 H. pylori OMV osteoporosis study (33) was retained as adjacent gastric-source evidence under the same anatomical boundary. No claim is made that the 14-report set exhausts all skeletal bacterial-OMV studies regardless of anatomical source.

2.2. Definitions: BEV, GMEV, and three categories that are not interchangeable

For each vesicle-containing preparation we record three independent attributes: (i) how it was obtained (directly from an intestinal/community sample or from a defined bacterial culture), (ii) the evidence for bacterial identity and intestinal anatomical provenance, and (iii) whether the resulting EV was modified or loaded by engineering. These descriptors are not mutually exclusive. In particular, an engineered BEV may originate from a cultured gut-associated strain.

  1. Community-derived EV-enriched preparations: fractions isolated from intestinal content, mucosa or a related specimen. A preparation may have evidence of bacterial contribution without species-level attribution; enrichment does not exclude co-isolated host EVs or soluble material.

  2. Cultured strain-derived BEVs (cBEVs): vesicles obtained from a separately cultivated, taxonomically identified bacterial strain under specified growth and isolation conditions. Producer identity can be established for this culture preparation, but its presence in human circulation or gut-lumen origin in vivo is not thereby demonstrated.

  3. Engineering status (a separate attribute): a vesicle preparation may be native, surface-modified, loaded with a specified exogenous cargo, or incorporated in a combined formulation. Engineering does not change the bacterial source or the GC/non-GC classification of the relevant experimental arm (Sections 5.2.2, 7.3 and 8.5).

The distinction is not pedantic. MISEV2023 states that “different species, strains, and growth conditions affect EV heterogeneity on multiple levels, including function,” and that “for most bacterial species, studies of the influence of culture conditions on the yield and composition of bacterial EVs are in their infancy”; it therefore recommends that culture details be reported, including explicitly that authors “report bacterial growth phase at harvest” (34). Medium composition, oxygenation, culture format (planktonic versus biofilm), and harvest phase can all alter BEV yield, cargo, and immunogenicity (19, 20, 35). Consequently, a cBEV preparation is best understood as the vesicles that one strain makes under one set of laboratory conditions. Throughout this review we therefore name the category whenever a study is cited, and we treat cBEV evidence as an indirect proxy for in vivo GMEV biology rather than as evidence about the gut lumen itself.

Four distinct questions must be addressed separately: (i) is the recovered material a vesicle-enriched preparation with credible bacterial contribution; (ii) is the bacterial producer identified and, independently, is the intestinal anatomical origin established; (iii) is the proposed cargo demonstrated to be of bacterial origin and protected within intact EVs; and (iv) is that cargo necessary for the skeletal phenotype? Host-style miRNA identifiers alone cannot establish bacterial transcription, and whole-vesicle efficacy does not establish cargo necessity.

2.3. Evidence-relevance matrix, decision rules, and what falls outside it

Qualifying vesicle-related skeletal claims were placed into a nonordinal two-axis evidence matrix defined by (i) vesicle producer (bacterial versus host) and (ii) model context (GC-associated skeletal disease versus non-GC skeletal disease). These yield evidence categories A–D (Table 1); they are positions on a relevance matrix, not a rank order of quality. A publication may contribute more than one preparation/claim to different analytic strata, while retaining one publication-level S2B identifier for indexing. Table 2 is a condensed display of bacterial-EV skeletal studies rather than a claim that all experimental arms in a paper share identical properties. The population counted in Table 2 is the defined intestinal-associated bacterial-source core; it is not intended as a census of all bacterial skeletal OMV studies.

Table 1.

Evidence-relevance matrix used to assign categories.

Vesicle source GC-related skeletal context Non-GC skeletal context
Vesicle source: bacterial/gut-derived (identified) category A — direct GC + bacterial vesicle + bone evidence category B — bacterially derived vesicle evidence from non-GC bone models (cross-disease extrapolation)
Vesicle source: host-derived (identified) category C — host-EV evidence within GC bone disease (cross-source extrapolation) category D — host-EV evidence in a non-GC skeletal context (cross-source + cross-disease extrapolation)
Not assigned to a matrix category (held outside the matrix, and the reason why):
Category Content Why it is not assigned to a matrix category
Adjacent mechanistic evidence Non-vesicular microbial products (SCFAs; purified surface proteins such as Amuc_1473); non-skeletal vesicle evidence (e.g. oncological EV-RNA detection) No intact bacterial vesicle was studied in a skeletal context, or no skeletal context was studied at all; such evidence informs hypotheses but cannot test a vesicle-mediated mechanism
Unresolved-source studies Vesicle effects reported without distinguishing bacterial from host vesicles (Rule 4) Cannot be assigned to either row of the matrix; may be cited for context but cannot support a bacterial-vesicle claim
Hypothesis-generating mechanisms Mechanisms inferred from general EV biology, osteoimmunology, vascular biology, or cargo-function analogy without a qualifying vesicle study in the relevant skeletal context Plausibility is not equivalent to matrix-eligible evidence; these claims are discussed as testable hypotheses and are not assigned category D

Evidence categories represent relevance and directness to the GC–bacterial-vesicle–bone question, not methodological quality; the latter is appraised descriptively in Table 3. Categories B and C represent two distinct forms of extrapolation and are not ordered relative to each other. Hypothesis-generating claims that do not satisfy the matrix entry conditions are held outside the category system. Decision rules 1–4 are defined in Section 2.3.

Table 2.

Core evidence matrix of bacterially derived vesicle studies in skeletal contexts.

S2B Study Bacterial source/EV context Skeletal model GC exposure Cargo/effector Key skeletal finding Evidence category/section 5
S2B-01 Chen et al., 2022 (24) L. animalis; cultured commensal-derived cBEVs GC-induced GIONFH; male C57BL/6J mice Yes Whole-vesicle proteins; causal effector unresolved Femoral-head osteonecrosis attenuated; angiogenesis/osteogenesis increased; apoptosis reduced category A; 5.1
S2B-02 Liu et al., 2021 (58) Child gut microbiota EV-enriched fractions + cultured A. muciniphila arm; community/cultured strata OVX osteoporosis; female C57BL/6 mice No Not resolved Vesicle-associated bone tracing signal was observed and preserved bone mass, microarchitecture, and strength category B; 5.2.1
S2B-03 Lu et al., 2026 (59) R. intestinalis; cultured commensal-derived REVs OVX osteoporosis; rats + hBMSCs No Histidylleucine; intravesicular provenance unresolved Improved BMD and osteogenic differentiation; histidylleucine was functionally relevant category B; 5.2.1
S2B-04 Ji et al., 2026 (38) A. muciniphila; engineered BEVs with SDSSD bone-targeting peptide OVX osteoporosis; female C57BL/6 mice No miR-21-5p reported; bacterial origin not established Enhanced bone accumulation; improved osteogenesis and reduced osteoclastogenesis category B, engineered; 5.2.2
S2B-05 Liu et al., 2023 (62) EcN; engineered BEVs with hCXCR4 surface display OVX osteoporosis; female C57BL/6 mice No SOST siRNA Bone-targeted delivery, SOST silencing, and improved distal-femur bone parameters category B, engineered; 5.2.2
S2B-06 Liu et al., 2023 (63) L. rhamnosus GG; engineered bone-targeted EVs OVX osteoporosis; female C57BL/6 mice No Reported miRNA profile; bacterial origin unresolved Femoral accumulation and improved bone formation with altered osteoclastogenic responses category B, engineered; 5.2.2
S2B-07 Kong et al., 2025 (64) EcN; engineered BEV-DCS OVX osteoporosis; female C57BL/6J mice No FRAT peptide; DC-STAMP surface targeting Targeted pre-osteoclast delivery and reduced bone loss category B, engineered; 5.2.2
S2B-08 Zhang et al., 2025 (60) L. salivarius; cultured commensal-derived EVs ORX osteoporosis; male rats No Broad protein cargo; causal effector unresolved Vesicle-associated bone tracing signal was observed and mitigated androgen-deficiency osteoporosis category B; 5.2.1
S2B-09 Wang et al., 2025 (61) L. reuteri + L. paracasei; cultured probiotic-derived BEVs OVX osteoporosis; female C57BL/6 mice No Reported miRNA profiles; bacterial origin unresolved Femoral accumulation and improved bone mass/osteogenic responses; combination performed better in tested model category B; 5.2.1
S2B-10 Zhou et al., 2025 (65) EcN; synthetic-biology engineered BEVs-BP Osteoporotic fracture + bone organoid models No pBMP-2-VEGF plasmid; IL-4 is separate hydrogel cargo Enhanced osteogenesis/angiogenesis; composite IL-4/BEVs-BP@GelMA accelerated fracture repair category B, engineered; 5.2.2
S2B-11 Wang et al., 2022 (29) P. mirabilis (intestinal-associated opportunistic species); cultured OMVs, not intestine-isolated EVs OVX osteoporosis; mice; osteoclast/BMM assays; CIA model No Bacterial cargo unresolved; miR-96-5p/Abca1 is a host-cell response axis Reduced osteoclastogenesis/bone resorption and attenuated OVX bone loss category B; 5.2.1
S2B-12 Lei et al., 2026 (30) B. pseudolongum; cultured sEVs Cold-exposure bone loss; mice; BMSCs/BMMs No No single causal cargo resolved Bone uptake; enhanced osteogenesis/mineralization; suppressed pyroptosis; associated with preservation of bone mass category B; 5.2.1
S2B-13 Liu et al., 2024 (31) EcN; synthetic-biology engineered BEVs-BC OVX osteoporosis; female C57BL/6 mice; BMSCs No Surface-displayed BMP-2 + CXCR4 fused to ClyA Bone targeting; improved BMD/BV/TV/BS/TV/Tb.N and dynamic bone formation category B, engineered; 5.2.2
S2B-14 Lin et al., 2025 (32) PM-OMVs, bone-targeted; producer described as “Pseudomonas mirabilis” in source paper, but ATCC 12453 catalogued as Proteus mirabilis (source-name discrepancy) OVX osteoporosis; female C57BL/6 mice; BMSCs/osteoclast assays No Endogenous miRNA cargo not source-resolved; SDSSD is exogenous targeting modification Reported skeletal improvements with targeted PM-OMV formulation; source-name and some preparation labels in primary report require caution category B, engineered; 5.2.2

The matrix includes the fourteen core studies in the current study-level evidence extraction set. Only studies in which intact bacterially derived vesicles were administered in a skeletal context are included. Category A denotes direct GC + bacterial vesicle + skeletal evidence; Category B denotes non-GC skeletal bacterial-vesicle evidence. Engineered BEVs are explicitly marked as an orthogonal translational attribute and are not treated as equivalent to endogenous GMEVs. Community-derived preparations retain their source-resolution limitations. The H. pylori OMV study identified during supplementary checking is treated as adjacent broader bacterial-OMV evidence and is not included in this core matrix. Core set: 14 publication-level reports (S2B-01–S2B-14; category A, n=1; category B, n=13), selected under Rule 1. Gastric/oral OMV studies outside the operational intestinal-associated boundary remain adjacent; S2B-14 is flagged for unresolved taxonomic/preparation labeling inconsistencies.

Evidence category assignment describes source and disease relevance, not methodological rigor, risk of bias or certainty. Category A may have serious design limitations, while an experiment in category D may be technically strong. We therefore keep evidence relevance (Tables 1, 2) separate from a descriptive methodological and internal-validity reporting assessment (Table 3). The latter uses reported, not reported and unclear designations without converting nonreporting into an automatic high-risk-of-bias judgment; it is not a formal SYRCLE score (34, 36).

Table 3.

Descriptive appraisal of four representative bacterial-EV skeletal studies.

Panel A: Preparation, intervention, endpoints and evidential limitations
Study Vesicle category and source attribution Separation and characterization Model, dose, route Biodistribution evidence Functional validation Major limitation or uncertainty
Chen 2022 (24) Cultured L. animalis EVs; serum/gut taxonomic assignment not fully species-specific Centrifugation, 0.22-µm filtration, UF, OptiPrep; TEM/NTA MPS-associated GIONFH, male mice; oral L. animalis EVs 30 µg once/week ×6 weeks (early/preventive); late-stage regimen assessed separately (SI Supplementary Figure S8). Femoral-head vesicle-associated fluorescence; intact EV delivery not demonstrated µCT 10/group and selected tissue assays 5/group; SI Supplementary Figure S8 late-stage bone endpoints 7–8/group, no notable protection. Single murine preventive/early-intervention result; late-stage benefit not shown; producer relevance, intact trafficking, and necessary cargo unproven.
Liu 2021 (58) Child gut community EV-rich fractions; separate cultured Akk EV preparation UF + OptiPrep; TEM, DLS, NTA, BCA OVX mice: oral CGM-/Akk-EVs 1.2×10¹0 particles (~200 µg protein)/dose, twice/week ×2 months in efficacy arm; separate i.v. and tracing regimens. Bone-associated dye signal; intact delivery not established Oral skeletal µCT n=8/group and osteoclast n=5 (SI Supplementary Figure S8); i.v. bone outcomes n=7–9 (S9). Separate 16S Supplementary Tables S2, S3. 16S Supplementary Tables S2, S3 describe recipient and donor community composition (12/6 samples), NOT EV efficacy n or species-specific cargo; no GC model.
Lu 2026 (59) Cultured Roseburia intestinalis REVs Clarification, ultracentrifugation, EXODUS; TEM/NTA OVX/dysbiosis rat arms; in 4-group antibiotic model REVs 10 mg/kg or free histidylleucine 1 mg/kg i.p. every 3 days ×5; separate hBMSC arms. DiR-REV localization described as oral and i.p. in different primary-text passages; original-report discrepancy unresolved. BMD, osteogenesis; free histidylleucine also active No GC model; oral/i.p. DiR-route conflict; free dipeptide response does not prove vesicle packaging/necessity.
Ji 2026 (38) Cultured Akk EVs modified with SDSSD; miR provenance uncertain UF/ultracentrifugation; TEM/NTA/zeta; targeting confirmation OVX mouse; 50 µg EVs/100 µL tail vein once/week ×8 weeks (efficacy). Distinct 4-h biodistribution study uses 200 µg/200 µL. Enhanced bone-associated label at 4 h In vivo µCT/serum/histology 5 mice/group; cell assays often N = 3; supplement supplies safety and inhibitor-efficiency readouts. Engineered vs native BEV evidence only; bacterial miR-21-5p biogenesis/cargo protection unproven; SI peptide naming needs reconciliation.
Panel B: Internal-validity reporting by domain
Study Random allocation Blinded assessment Unit/n Sample-size basis Exclusions/attrition Vesicle-specific controls/interpretation
Chen 2022 (24) Reported randomized mouse allocation; allocation-concealment details not ascertained Not blinded during experiments; blinded during data analyses (SI Methods p.2) Independent mice/cells: µCT 10/group; tissue 5/group; selected tracer 3/group Prospective power calculation not ascertained Pre-treatment exclusion by weight (<18/>26 g) or poor condition; postallocation losses not ascertained Vehicle/live-microbe and tracing comparisons; free-dye-only control and intact-vesicle delivery not established
Liu 2021 (58) EV-efficacy-arm randomization method not ascertained Blinded efficacy or histology assessment not ascertained Oral EV µCT n=8, osteoclast n=5; i.v. bone n=7–9; tracing n=3/group. 16S S2: 12, S3: 6 separate specimens Prospective power calculation not ascertained EV-arm outcome-specific losses/animal exclusions not ascertained Community and cultured Akk EV arms; GW4869 comparison; 16S taxonomy not EV efficacy or source-protected cargo proof
Lu 2026 (59) Reported random assignment for 16S/OVX and 4-group antibiotic-dysbiosis rat arms; other arms separately qualified Reporting Summary: author states “Not applicable”; assessor blinding not demonstrated Rat dysbiosis n=20, four arms; skeletal µCT n=5/group; selected IF n=3/group; distinct hBMSC replicates n=5/group described as literature-/standard-based; numerical a priori power calculation not shown Reporting Summary states no data excluded; full per-arm animal flow not ascertained REVs vs free histidylleucine; DiR localization route discordant (oral vs i.p.); cargo necessity not established
Ji 2026 (38) Reported: 20 female mice randomized across 4 arms (n=5/group) Blinded animal outcome assessments not ascertained Mouse µCT/serum/histology n=5/group; cell figures N = 3 with independence of units unresolved Prospective power calculation not ascertained Postrandomization attrition/exclusions by endpoint not ascertained Native AKK-EVs vs engineered BT-AKK-EVs; cell miR inhibitor tests; bacterial origin/protection not established
Panel C: Outcome-reporting completeness and source verification
Study Original supplements inspected Reported outcomes/denominators Remaining completeness limitation
Chen 2022 (24) Supplementary Methods; Supplementary Figures S1-S8; separate Supplementary Tables S1-S3 (xlsx/zip). Early-stage positive µCT/tissue endpoints (n by assay); SI Supplementary Figure S8 late-stage ONFH bone outcomes n=7–8/group without notable protection. No prospective outcome protocol or complete postallocation animal-flow record identified; differential outcomes and dye-only comparator unresolved.
Liu 2021 (58) Supporting Information PDF: Supplementary Figures S1-S9 and Supplementary Tables; separate 16S taxonomic workbooks S2/S3. Oral EV µCT n=8/group, osteoclast n=5/group (S8); i.v. skeletal n=7–9/group (S9); 16S S2 4×3 and S3 2×3 specimens are NOT efficacy n. No verified prospectively prespecified outcomes; EV-arm allocation/blinding/attrition remain not ascertained. Taxonomic profiles do not validate EV cargo necessity.
Lu 2026 (59) Publisher Supplementary Information and Nature Portfolio Reporting Summary (earlier direct check). Dysbiosis rat µCT n=5/group, selected IF n=3/group; author reported random assignment, no data exclusion, masking “Not applicable.” Route of DiR biodistribution reported inconsistently (oral vs i.p.); full selective-reporting appraisal and quantitative power calculation unavailable.
Ji 2026 (38) Supplementary Material 1 DOCX: primer table; Supplementary Figures S1-S8 (MALDI-TOF, organ safety, cell viability, Masson, miR inhibitor efficiency). Four mouse arms, n=5/group; several in-vitro plots N = 3; supplement verifies selected chemistry/safety and inhibitor-efficiency assays. No source-verified a priori power calculation, assay blinding, per-arm attrition or independent N = 3-unit detail; peptide naming and bacterial miRNA provenance unresolved.

Panel A: preparation, intervention, endpoints and evidential limitations; Panel B: internal-validity reporting by domain; Panel C: outcome-reporting completeness and source verification.

“Not ascertained” indicates that an inspected source did not establish the domain; it must not be interpreted as evidence that the procedure was not performed or as a formal high-risk-of-bias rating. “Author reports not applicable” preserves the original claim and does not verify adequate masking. Group sizes and regimens are specific to experiments or outcomes, not universal across a paper. Chen 2022 reports unblinded experimental conduct but blinded data analysis in its Supplementary Methods. Lu 2026 contains an unresolved original-report discrepancy about the oral versus intraperitoneal DiR-REV biodistribution route.

Original Supplementary Methods/Figures were directly inspected for Chen (24), Liu (58) and Ji (38); Liu’s separately published 16S taxonomic Supplementary Tables S2, S3 and Chen’s Supplementary Tables S1–S3 were also available for source checking. Lu (59) Supplementary Information and Reporting Summary were assessed in the preceding source-verification pass. The outcome inventory is descriptive; without prespecified protocols, it cannot establish the absence of selective outcome reporting. Reporting appraisal is informed by MISEV2023 and selected SYRCLE concepts and is not a formal scored SYRCLE assessment.

Categories B and C should not be compared ordinally: category B retains bacterial source identity but lacks a GC skeletal model, whereas category C retains the GC skeletal setting but concerns host-derived EVs. These are different extrapolation directions. ‘B’ does not signify stronger evidence than ‘C’.

The following preparation-/claim-level decision rules were used. A publication is counted once in S2B, but each relevant EV preparation, intervention arm and mechanistic claim is interpreted on its own terms:

  • Rule 1 (core intestinal-associated bacterial-EV skeletal inclusion). The operational core set addresses bacterial vesicles of intestinal-microbiota relevance, not all bacterial OMVs with skeletal effects. A qualifying experimental arm must (i) administer or test an isolated, intact bacterial-EV or bacterial-vesicle-enriched preparation with credible bacterial contribution; (ii) report an experimental skeletal or bone-cell endpoint; and (iii) use a directly intestinal-community-derived preparation or a cultured bacterial taxon with an established intestinal ecological association (including probiotic chassis or intestinally associated opportunistic species). For the non-GC core, eligible disease contexts were systemic noninfectious bone loss (including sex-steroid deficiency and cold exposure), osteoporosis-related fracture repair, or bone-cell experiments directly linked to those settings. Studies centered on periodontal regeneration, infection-related bone defects, or chemotherapy-associated MSC senescence without a corresponding qualifying skeletal disease model were considered adjacent or outside the disease-focused core, as appropriate. The identity of a cultured donor may support taxon-level preparation attribution but does not establish intestinal anatomical origin of the administered EVs in vivo. Predominantly gastric- or oral-source OMVs are discussed as adjacent evidence; pathogen status alone is not an exclusion criterion. Live-bacterium-only interventions, free proteins/metabolites, and unresolved bacterial-versus-host EV preparations do not qualify as bacterial-EV core evidence. Source-name discrepancies and combination interventions are explicitly flagged rather than normalized or attributed to BEVs alone (Rules 2–4; Supplementary Tables S2B–D). The broader A–D conceptual matrix separately accommodates appropriately characterized host-EV skeletal studies.

  • Rule 2 (preparation and claim as analytic units). Matrix assignments are made for the vesicle preparation or experimental claim, not automatically for an entire paper. A cultured-strain arm, community-derived EV-enriched arm, native BEV and engineered BEV must each retain its own origin, engineering and outcome labels. Multiple arms from one publication do not increase the count of unique studies.

  • Rule 3 (mixed and combination interventions). When intact BEVs, live bacteria, soluble cargo or biomaterial carriers occur in separate or combined study arms, only the BEV-containing arm informs the BEV matrix. If a skeletal benefit was demonstrated with a BEV-plus-hydrogel/IL-4 combination, it is attributed to that combination, not to BEVs alone. A free-cargo comparator may demonstrate free-cargo sufficiency without proving vesicular encapsulation or necessity.

  • Rule 4 (graded source attribution). A cultured, identity-verified producer provides strain-level attribution for its recovered preparation but not evidence of intestinal anatomical provenance in vivo. Directly recovered community EV-enriched fractions can be described as community-associated only when bacterial contribution is supported by source-appropriate controls; without species-specific markers they are not species-resolved. Material that cannot distinguish bacterial EVs from host EVs or co-isolated non-vesicular contaminants is coded unresolved-source and cannot support species-specific BEV causation. Cargo origin, intravesicular localization and functional necessity are separately coded as established, unresolved or not assessed. See Supplementary Table S2D for operational thresholds and examples.

Outside the matrix. Four bodies of evidence are deliberately not assigned a category, because forcing them into the two-axis matrix would misrepresent what they measured: (i) non-vesicular microbial products, including short-chain fatty acids and purified surface proteins; (ii) non-skeletal vesicle evidence, however rigorous; (iii) unresolved-source studies; and (iv) hypothesis-generating mechanisms inferred from general EV biology, osteoimmunology, vascular biology, or cargo-function analogy without a qualifying vesicle study in the relevant skeletal context. Each is labeled explicitly wherever it is cited.

2.4. Status of quantitative claims about the literature

Statements that an eligible study was “the only” one identified, or that “no eligible study” was identified, describe the output of the operational intestinal-associated source framework and are bounded by the specified databases, search terms, screening depth, and cutoff date. They should not be read as absolute claims about the published record, particularly in a field where preprints and very recent online-first articles may not yet be indexed.

3. Gut bacterial EVs as candidate inter-organ messengers

3.1. Biogenesis, composition, and a note on RNA nomenclature

GMEVs are bilayer-enclosed vesicles, typically 30–400 nm, released by virtually all intestinal bacteria (18, 19). Gram-negative species primarily bud outer-membrane vesicles from the outer membrane, whereas Gram-positive species—which lack an outer membrane—release cytoplasmic-membrane-derived vesicles following peptidoglycan remodeling; the mechanisms governing 7vesiculation in Gram-positive commensals remain comparatively poorly characterized (18, 20). Cargo includes LPS, peptidoglycan, lipoteichoic acid, porins, adhesins, enzymes, and small non-coding RNAs, and confers the capacity to modulate host immune, metabolic, and signaling pathways (18, 19, 35). The compositional breadth of these vesicles is also their principal analytical liability: because BEVs are heterogeneous by species, strain, and growth condition (34), and because the same preparation can contain several vesicle subpopulations (37), a report that “GMEVs do X” is uninterpretable unless the vesicle population is specified.

RNA nomenclature. Throughout this review we use bacterial small RNAs, bacterial regulatory RNAs, or miRNA-like small RNAs to describe RNA carried by naturally produced GMEVs. Canonical miRNA designations (e.g., miR-146a, miR-155, miR-21-5p) name host-encoded sequences; applying them to bacterial vesicles implies a bacterial origin that, as discussed in Section 8.1, is seldom demonstrated. We therefore reserve miR-X nomenclature for host miRNAs measured in host-derived vesicles, or for exogenous cargo deliberately loaded into engineered BEVs. One consequence is worth flagging at the outset, because it recurs: a recent study reporting that miR-21-5p is enriched within Akkermansia muciniphila BEVs (38) uses a mammalian nomenclature for a putatively bacterial vesicle cargo, and the bacterial origin of that sequence was not established—precisely the attribution problem discussed in Section 8.1.

3.2. Crossing the intestinal barrier and reaching the skeleton

For GMEVs to participate in gut–bone communication, they must cross the intestinal barrier and enter the systemic circulation. Under physiological conditions the mucus layer and tight-junction complexes restrict translocation (39), but several independent observations indicate that translocation occurs: LPS-positive BEVs have been traced along the gut–portal vein–liver axis (21); bacterial membrane vesicles and phages have been detected in blood after oral administration of a commensal strain (22); and commensal BEVs can deliver bioactive protein to distal organs (23). Bacterial EVs also act at sites far from the gut in other disease contexts (40, 41), and GMEVs constitute a molecularly distinct entity from the parent microbiota rather than a passive reflection of bacterial abundance (42).

Surface determinants of BEVs—including phosphatidylserine, outer-membrane protein A, and flagellin—can engage host pattern-recognition receptors such as TLRs and trigger inflammatory programs in innate immune cells in vitro (43–45). These observations support the plausibility of host-cell recognition and uptake. However, preferential uptake of gut bacterial EVs by bone-resident cells or bone microvascular endothelial cells (BMECs), and any dependence of that uptake on GC exposure, has not been directly established. Such target-cell specificity is therefore hypothesis-generating evidence outside the matrix; the translocation observations themselves are adjacent non-skeletal evidence.

4. Glucocorticoid-induced remodeling of the microbiota–EV ecosystem

4.1. GC-induced gut dysbiosis

Chronic GC therapy reshapes gut microbiota composition and function. Reported changes include reduced α-diversity, altered Firmicutes/Bacteroidetes ratios, depletion of commensals such as Lactobacillus and Bifidobacterium, expansion of pathobionts, and dysregulation of short-chain fatty acid (SCFA) and secondary bile-acid metabolism (46–48). Because SCFAs themselves regulate systemic bone mass (12), metabolite depletion and vesicle remodeling are difficult to separate experimentally: an intervention that restores SCFA production will also alter the vesiculating community. Altered microbial composition has likewise been reported in steroid-associated osteonecrosis, supporting a link between dysbiosis and this specific skeletal outcome (48, 49). Although the GC–dysbiosis relationship is supported by both animal models and human observational data, the causal chain connecting dysbiosis to GMEV alteration and then to skeletal phenotype in patients remains unclosed.

4.2. GC-dependent alterations in GMEV quantity and composition

Direct evidence here is narrow, and its limits are worth stating precisely. In the murine GC model of Chen and colleagues, GC exposure was accompanied by depletion of L. animalis-derived material in intestinal content and serum, together with a shift of the EV pool toward a composition associated with a pro-resorptive and pro-apoptotic skeletal phenotype (24). Two qualifications follow from the primary report itself: the model used a single GC regimen (methylprednisolone 20 mg/kg/day for the first 3 days of each week over 3 weeks) in male mice only, and the L. animalis signal was quantified by qRT-PCR using primers that could not be rendered species-specific (Section 5.1). That the GMEV pool is dynamically regulated by host physiology is independently supported by reports of diurnal variation and diet-responsive remodeling (50, 51).

The broader assertion that GC administration generally increases total fecal and serum GMEV concentration and shifts vesicle size profiles toward larger diameters rests on the same single dataset (24). We therefore treat it as a model-specific observation requiring replication rather than as an established property of GC exposure. It is category A only as a finding within that specific murine model; extrapolation to other GC regimens, species, or patients lies outside the demonstrated evidence and should not be presented as a quantitative generalization.

4.3. Context-dependent intestinal barrier effects and systemic bacterial-vesicle exposure

The effect of glucocorticoids on the intestinal barrier is context dependent rather than uniformly disruptive. In epithelial and inflammatory models, glucocorticoids can strengthen barrier function, increase transepithelial resistance, or counter cytokine-induced permeability changes through regulation of tight-junction proteins (52, 53). By contrast, chronic stress/corticosterone models implicate epithelial glucocorticoid-receptor signaling in increased permeability and altered claudin regulation (54, 55), and barrier impairment has been reported in GC-induced osteoporosis models in which microbiota and barrier function appear jointly involved (46). These apparently divergent findings indicate dependence on dose, duration, inflammatory context, and model system. Accordingly, increased systemic exposure to bacterial vesicles under GC treatment should be treated as a testable hypothesis rather than a general consequence of GC exposure.

We identified no eligible GC-exposed cohort directly measuring circulating source-resolved GMEV abundance alongside barrier-injury markers. That proposed relationship is hypothesis-generating and requires dedicated longitudinal studies that co-measure vesicle source, permeability or enterocyte-injury markers, and skeletal outcomes.

4.4. Three-way interaction among GC, GMEVs, and immune cells

GCs exert direct immunosuppressive effects on myeloid and lymphoid compartments. Concurrently, BEVs bearing LPS and other pathogen-associated molecular patterns (PAMPs) can engage TLR-dependent signaling and drive pro-inflammatory programs in innate immune cells (43, 44, 56). Because inflammatory cytokines and pathological bone resorption are tightly coupled (56, 57), an interaction between GC immunosuppression and BEV-driven innate signaling is biologically plausible. However, the supporting data derive from non-skeletal cell systems and ex vivo models. No eligible study in the defined search directly quantified this balance within bone tissue under GC exposure; its direction, concentration dependence, time course, and cell-type specificity remain unknown. These propositions are hypothesis-generating and lie outside the evidence-relevance matrix. Figure 1 distinguishes the proposed endogenous GC–gut–GMEV–bone axis from the intervention-associated observations in the murine GIONFH study.

Figure 1.

Infographic with two main sections: Panel A shows the proposed endogenous pathway linking pharmacologic glucocorticoid (GC) exposure to gut dysbiosis, altered gut barrier effects, possible changes in gut microbe extracellular vesicle (GMEV) exposure, and uncertain skeletal effects, noting the sequence is unvalidated. Panel B summarizes observed results from a mouse model showing that oral cultured L. animalis BEVs improved angiogenic and osteogenic measures, reduced apoptosis, and attenuated femoral-head osteonecrosis in early interventions, but did not protect in late-stage disease, with a caution that endogeneity of GMEV loss remains unproven.

Separation of the hypothesized endogenous GC–gut–GMEV–bone axis from intervention-associated findings in a single murine GIONFH study. (A) Pharmacologic GC exposure may alter the gut community and intestinal barrier in context-dependent ways; a causal sequence linking those changes to source-resolved circulating GMEVs and skeletal injury remains unproven. Dashed arrows denote hypothetical associations, not established mechanisms. (B) In Chen et al. (2022), oral administration of EVs from cultured Lactobacillus animalis in a male-mouse GC-associated osteonecrosis model was associated with improvements in angiogenic and osteogenic measurements, reduced apoptosis-related readouts, and attenuation of femoral-head osteonecrosis in an early/preventive treatment paradigm. The tested later-stage arm did not show notable bone protection. These parallel readouts neither establish an ordered causal mediator sequence nor demonstrate that loss of endogenous GMEVs is necessary for disease. Within the specified intestinal-associated source and literature-search framework (formal publication cutoff: 15 September 2026), 14 core reports were retained (one GC-associated GIONFH study and 13 non-GC skeletal studies); the latter do not provide direct evidence for GIOP or GIONFH. Solid arrows indicate intervention-associated observations only.

5. Evidence for the GC–bacterial-vesicle–bone axis: matrix-based assessment

We organize the fourteen core studies within the four-cell matrix defined in Section 2 and Table 1. Evidence categories indicate relevance and directness rather than methodological quality, and categories B and C are not ordered relative to each other. Evidence that does not meet matrix entry conditions is labeled as adjacent, unresolved-source, or hypothesis-generating rather than being forced into a category. Table 2 provides the study-level core evidence matrix, while Table 3 gives a structured methodological and reporting appraisal of four representative studies (preparation, internal-validity reporting, and outcome-reporting completeness); Table 4 summarizes major claims and their evidence boundaries.

Table 4.

Claim-by-claim evidence summary for bacterial vesicles in glucocorticoid-associated skeletal injury.

Claim Current evidence Evidence category Appropriate interpretation Key gap
Bacterial vesicles attenuate GC-induced osteonecrosis Chen et al. (2022) showed that L. animalis EVs reduced osteonecrosis in one mouse GC model (24) category A A preventive/early-stage L. animalis EV intervention attenuated osteonecrosis in one GC-exposed mouse model; late-stage oral intervention did not show notable protective effects (Chen Supplementary Figure S8). Independent replication, stage-specific efficacy, prospective power/animal-flow reporting, necessity tests, and human source-resolved evidence.
Bacterial vesicles modulate skeletal outcomes in non-GC models Thirteen non-GC skeletal studies provide complementary evidence for gut-associated or cultured bacterial vesicle effects on bone (29–32, 38, 58–65). category B Bacterial vesicles modulate skeletal homeostasis in non-GC models; engineered studies additionally support targeting and cargo-delivery feasibility. Direct validation in GC models; source-resolved human GMEV evidence
Engineered bacterial vesicles can be targeted to bone Seven engineered BEV studies demonstrate bone targeting and/or programmable cargo delivery in OVX osteoporosis or osteoporotic fracture models (31, 32, 38, 62–65). category B; engineered Engineering can confer targeting and therapeutic cargo delivery, but these products are not endogenous GMEVs and were not tested in GC skeletal models. Engineered BEVs tested under GC exposure and disease-specific efficacy
Host-derived EVs modify GC skeletal phenotypes Multiple host-EV studies in GC osteoporosis and GC osteonecrosis (66, 69–72) category C EVs are plausible bone-directed delivery vehicles in GC contexts; source and cargo are host-derived Any bacterial-vesicle equivalent
Bacterial vesicle small RNAs suppress osteoblast Wnt/β-catenin signaling Mechanistic inference from general EV and osteoblast biology Hypothesis-generating (outside matrix) Plausible by analogy; not demonstrated; no bacterial effector identified Direct GC + bacterial vesicle + osteoblast experiments
Bacterial vesicles promote osteoclastogenesis PAMP–TLR biology and osteoimmunology analogy (43, 56) Hypothesis-generating (outside matrix) TLR-dependent osteoimmune signaling is plausible, but the GC-context bacterial-vesicle effect is untested GC-context osteoclast assays
Bacterial vesicles induce osteocyte SOST expression Osteocyte-biology inference Hypothesis-generating (outside matrix) No direct evidence that gut bacterial EVs induce osteocyte SOST under GC exposure GC-context osteocyte assays
Bacterial vesicles affect bone microvascular endothelial cells GIONFH vascular pathology plus limited cell models; host-EV bone–vessel coupling data (69) Hypothesis-generating for bacterial-vesicle mechanism; category C only for the host-EV analogy Host-EV data support analogy for bone–vessel coupling; bacterial-vesicle action on BMECs remains inferred Direct GC + BMEC experiments
Source-resolved GMEV signatures can diagnose GC skeletal injury Oncology liquid-biopsy methodology for EV-associated RNA (76, 77) Adjacent (non-skeletal) EV-cargo detection is technically feasible; bacterial contribution unresolved Source-resolved assays and GC-cohort validation
Non-vesicular microbial products modulate GC bone pathology Amuc_1473 protein in skeletal homeostasis; butyrate in GC-associated osteonecrosis (74, 75) Adjacent (non-vesicular) Microbial molecules influence bone, but these studies do not test vesicle-mediated mechanisms Experiments that isolate the vesicle fraction
Bacterial vesicles enhance anti-resorptive drug efficacy None Conceptual No evidence supports synergy or an altered safety profile Basic preclinical dose-finding before hypothesis framing

categories A–D denote matrix position and directness to the GC–bacterial-vesicle–bone question, not methodological quality. categories B and C are not ordered relative to each other. Claims marked adjacent, unresolved-source, or hypothesis-generating are held outside the matrix by design.

5.1. Category A: direct GC + bacterial vesicle + bone evidence (GIONFH only)

The single category A study is Chen et al. (24). In male C57BL/6J mice exposed to methylprednisolone, oral administration of cultured L. animalis-derived BEVs attenuated GC-induced osteonecrosis, accompanied by more favorable angiogenic and osteogenic read-outs and lower apoptosis measures. Vesicles were purified by filtration, ultrafiltration, and OptiPrep density-gradient centrifugation, and were characterized by transmission electron microscopy and nanoparticle tracking; skeletal assessment included micro-CT, vessel volume/CD31, osteocalcin, and TUNEL. Within the limits of the search described in Section 2, this was the only identified eligible dataset that simultaneously satisfied GC exposure, delivery of bacterially derived vesicles, and skeletal phenotyping. Importantly, the direct phenotype is GIONFH/osteonecrosis; no corresponding eligible category-A bacterial-vesicle dataset was identified for GIOP.

The anchor study is a murine intervention proof of concept rather than a definitive endogenous GMEV mechanism. Its principal limitations include limited human relevance of the donor taxon, cultured-vesicle rather than source-resolved natural community exposure, uncertainty about the fraction of an administered fluorescently labeled preparation that reaches bone as intact EVs, a preventive/early-intervention rather than established-disease reversal setting, and unidentified necessary effector(s) and responding cells. The original Supplementary Methods confirm that researchers were aware of group assignment during experiments but blinded to group assignment during data analysis; thus this was not a fully blinded experiment (24; Table 3).

Importantly, Chen et al.’s Supplementary Figure S8 reports no notable protective effect of oral L. animalis or L. animalis EVs when given at a later stage of MPS-induced ONFH (bone microstructure n = 7–8/group). The preclinical signal therefore supports timing-dependent protection, not reversal of established ONFH. The Supplementary Methods also distinguish biologically independent experimental units from technical replicates, while detailed allocation concealment, a prospective power calculation, complete postallocation animal flow, and free-dye exclusion of vesicle-tracer artifacts could not be confirmed. These reporting limits are not, by themselves, proof that a method was absent.

Taken together, the anchor study demonstrates that a GC-sensitive, bone-beneficial cultured bacterial-vesicle preparation is experimentally testable in one murine GIONFH model. It does not establish which molecule acts, which cell responds, whether naturally occurring community-derived GMEVs mediate the effect, and, given the late-stage negative supplementary findings, it does not demonstrate reversal of established disease, or whether the responsible organism is relevant to the human gut. The result should therefore be interpreted as proof of concept for GIONFH rather than as direct evidence for GIOP or for a generalized human GC–GMEV axis.

5.2. Category B: bacterial vesicle evidence from non-GC bone disorders

5.2.1. Nonengineered community-enriched and cultured bacterial EVs

Liu et al. (2021) showed that EV-enriched preparations derived from child gut microbiota entered bone and preserved bone mass and strength in ovariectomized mice, with a parallel cultured Akkermansia muciniphila arm (58). The community-derived preparation cannot be assigned to a single bacterial taxon, and no specific effector cargo was resolved; the study therefore supports gut microbiota-associated bacterial-vesicle activity in bone without establishing a GC-specific mechanism.

Lu et al. (2026) reported that Roseburia intestinalis-derived EVs promoted osteogenic differentiation and mineralization in vitro and improved bone mineral density in ovariectomized rats, with histidylleucine identified as a functionally relevant metabolite (59). The study is strengthened by measurement of the donor taxon in patients with osteoporosis and in animals. However, free histidylleucine reproduced the osteogenic effect, so the data establish vesicle-preparation sufficiency rather than proving that vesicle packaging is necessary; whether histidylleucine is intravesicular cargo remains unresolved.

Wang et al. (2022) showed that Proteus mirabilis-derived OMVs inhibited osteoclast formation and bone resorption and attenuated OVX-induced bone loss in mice (29). The study used intact bacterial OMVs characterized by TEM and nanoparticle tracking. The mechanistic read-out centered on a host-cell miR-96-5p/Abca1 axis and mitochondrial apoptosis; no single bacterial cargo was established. Because the donor is an intestinal/gut-associated bacterium but the experiment used cultured OMVs rather than endogenous circulating GMEVs, this is non-GC cultured-BEV evidence rather than direct GIOP/GIONFH evidence.

Lei et al. (2026) identified Bifidobacterium pseudolongum-derived small EVs as a gut–bone mediator in a cold-exposure model of bone loss (30). B.p-sEVs were isolated from anaerobic culture, characterized by TEM and particle-size analysis, and shown to enter bone and enhance BMSC osteogenesis while suppressing pyroptosis. No GC exposure was used, and no single causal vesicle cargo was established; the study therefore expands the non-GC gut bacterial EV evidence base without supporting a GC-specific mechanism.

Zhang et al. (2025) demonstrated that Lactobacillus salivarius-derived EVs were transported to bone and mitigated orchiectomy-induced osteoporosis in rats (60). The study also linked reduced fecal L. salivarius abundance to male osteoporosis and showed complementary microbiota and bacterial interventions. Because the vesicles were produced from a defined cultured commensal strain and no GC exposure was used, this is non-GC cultured-BEV evidence rather than GIOP or GIONFH evidence; proteomic profiling did not identify a single causal EV cargo.

Wang et al. (2025) evaluated vesicles derived from Lactobacillus reuteri and Lacticaseibacillus paracasei, separately and in combination, in ovariectomized mice (61). Both preparations accumulated in femur and promoted osteogenic responses, while the combined regimen produced greater bone-protective effects in the tested model. The combination arm introduces a co-delivery limitation, and the reported host-style miRNA sequences were not source-resolved to bacterial biosynthesis; accordingly, the study supports non-GC BEV feasibility rather than a defined bacterial-RNA mechanism.

The remaining seven studies tested engineered BEVs as delivery platforms.

5.2.2. Engineered BEVs: therapeutic-platform evidence

Ji et al. (2026) surface-engineered A. muciniphila BEVs with the SDSSD bone-targeting peptide and demonstrated enhanced bone accumulation and attenuation of bone loss in ovariectomized mice (38). The targeting phenotype was conferred by engineering; the reported miR-21-5p effector used host miRNA nomenclature, and its bacterial origin was not established. The study therefore demonstrates bone-targeted BEV delivery rather than endogenous GMEV biology.

Liu et al. (2023) engineered Escherichia coli Nissle 1917 BEVs to display hCXCR4 and loaded SOST siRNA, producing a bone-targeted BEV system that silenced SOST and improved distal-femur bone parameters in ovariectomized mice (62). This provides direct evidence for programmable cargo delivery and skeletal targeting, but the model was non-GC and the product was recombinant/cultured rather than endogenous GMEV.

A second Liu et al. (2023) study engineered Lactobacillus rhamnosus GG EVs with the SDSSD bone-targeting peptide and demonstrated femoral accumulation, improved bone parameters, and modulation of osteogenic and osteoclastogenic responses in ovariectomized mice (63). The study supports targeting and delivery feasibility, whereas the bacterial provenance of the reported host-style miRNA sequences remains unresolved.

Kong et al. (2025) engineered EcN BEVs displaying DC-STAMP and loaded the osteoclastogenesis-inhibitory FRAT peptide. FRAT@BEV-DCS showed targeted intracellular delivery to pre-osteoclasts and reduced bone loss in ovariectomized mice (64). This is a clear example of engineered BEV targeting of a defined therapeutic peptide, but it remains a non-GC osteoporosis study.

Zhou et al. (2025) developed synthetic-biology-derived EcN BEVs carrying a pBMP-2-VEGF plasmid and incorporated them with IL-4 in GelMA hydrogel for osteoporotic fracture repair (65). BEVs-BP promoted osteogenic and angiogenic responses, whereas the principal in vivo fracture-healing experiment used the combined IL-4/BEVs-BP@GelMA intervention. Thus, the in vivo effect cannot be attributed to BEVs-BP alone without qualification.

Liu et al. (2024) used synthetic biology to display BMP-2 and CXCR4 on EcN BEVs (BEVs-BC) and demonstrated bone targeting, enhanced osteogenic responses, and prevention of OVX osteoporosis (31). The intervention was entirely non-GC and engineered; BMP-2/CXCR4 were deliberately displayed through the ClyA scaffold, so the study supports therapeutic-platform feasibility rather than endogenous GMEV biology.

Lin et al. (2025) reported SDSSD-modified PM-OMVs with enhanced skeletal localization and improved OVX bone outcomes (32). The source article calls the producer “Pseudomonas mirabilis”, whereas the cited ATCC 12453 is catalogued as Proteus mirabilis; the actual experimental strain was not independently verified. Accordingly, this report is retained provisionally as engineered cultured bacterial-OMV platform evidence with a producer-label discrepancy, not as species-confirmed evidence of endogenous intestinal GMEVs.

Across these thirteen non-GC studies, preparation type shapes what can reasonably be inferred. Community-enriched fractions and vesicles from cultured gut-associated taxa provide preliminary evidence of skeletal activity, whereas engineered systems test targeting and cargo-delivery strategies, sometimes within combined interventions. Neither line of work establishes that endogenous GMEVs share these properties or that bacterial EVs improve GC-related bone disease.

Category B therefore informs biological plausibility and platform design, but independent studies under GC exposure are still needed, both to replicate the GIONFH finding and to test a dedicated GIOP model.

5.3. Category C: host-derived EV analogy in GC-associated bone disease

A substantial GC-bone EV literature derives from host-cell sources. Dexamethasone-stimulated M1-macrophage exosomal miR-1a-3p promotes adipogenic differentiation of bone marrow mesenchymal stem cells (BMSCs) in GC-associated osteonecrosis of the femoral head (66). BMSC-derived exosomes exert osteogenic effects in steroid-induced osteonecrosis models (67), and siRNA-encapsulated MSC-derived exosomes inhibit osteonecrosis in related models (68). Bone-targeting delivery of platelet-lysate exosomes ameliorates GIOP by enhancing bone–vessel coupling (69). Human umbilical-cord MSC-derived exosomes promote osteogenesis in GIOP through PI3K/AKT-mediated inhibition of ferroptosis (70); Wharton’s-jelly MSC exosomes reduce osteocyte apoptosis in GC-induced osteonecrosis via the miR-21–PTEN–AKT axis (71); and EVs enriched in miR-127-5p counteract steroid-induced osteonecrosis pathology through the Hippo pathway (72). Related work has engineered exosome-based nanoplatforms that reprogram macrophage polarization in osteonecrosis (73). For reference, a recent synthesis of EV evidence in osteonecrosis of the femoral head draws almost exclusively on this host-cell literature (26).

Together, these findings demonstrate that EVs function as bone-directed delivery vehicles in GC contexts and that vesicle-based intervention can modify GC skeletal phenotypes. They cannot establish that GMEVs fulfill equivalent roles: the vesicle source in every case is a host cell, and the cargo is host-encoded. The analogy is useful for delivery engineering (Section 7.3) and should not be cited as mechanistic support for bacterial vesicles.

5.4. Hypothesis-generating mechanisms outside the matrix

Host-EV experiments, osteoimmunology and non-GC bacterial-EV models raise specific questions that have yet to be tested in a GC-associated skeletal setting. The following are experimental tests, not pathways inferred to be active:

  • Does an isolated, source-resolved bacterial-EV preparation alter osteoblast Wnt/β-catenin or osteocyte SOST responses during pharmacologic GC exposure when compared with EV-depleted medium and equivalent free cargo?

  • Can a putative bacterial small RNA be verified as bacterial in sequence and biosynthetic origin, protected inside intact vesicles, delivered to target skeletal cells, and necessary for the observed phenotype? Host miRNA nomenclature or a free-cargo effect alone is insufficient.

  • Do bacterial-EV preparations alter osteoclast differentiation, RANKL/OPG signaling or PAMP-dependent innate responses under GC exposure, and are these effects retained after particle, soluble-contaminant and endotoxin normalization?

  • What fraction of vesicle-associated tracer signal represents intact particles in bone, and which recipient cells exhibit cargo release and functional target engagement rather than label transfer?

  • Are changes in macrophage activation or Treg/Th17 balance sufficient to influence GIOP-specific systemic fragility or GIONFH-specific focal osteonecrosis, or do they simply parallel generalized inflammation?

Resolving these questions will require source-verified preparations, particle-specific controls and disease-relevant skeletal endpoints.

5.5. Adjacent mechanistic evidence and unresolved-source studies

Four kinds of evidence inform the wider discussion without meeting the core bacterial-EV skeletal inclusion criteria: non-vesicular microbial products, non-skeletal EV findings, source-unresolved studies, and gastric-source bacterial OMVs.

Non-vesicular microbial products (adjacent). The A. muciniphila EV-enriched protein Amuc_1473 has been linked to skeletal homeostasis and counteracts multifactorial osteoporosis (74), and gut microbiota-derived butyric acid alleviates GC-associated osteonecrosis by modulating inflammatory cytokines in BMSCs (75). Both are genuinely relevant to microbial regulation of bone, but experiments using purified Amuc_1473 or free butyrate do not constitute intact-vesicle evidence. Administering a purified protein or a free metabolite does not test a vesicle-mediated mechanism, and neither can be counted as primary GMEV–bone evidence. We have accordingly excluded Amuc_1473 from Table 2 (whose scope is limited to preparations in which intact vesicles were administered) and treat both findings as adjacent.

Non-skeletal vesicle evidence (adjacent). Oncology studies establish the technical feasibility of measuring EV-associated RNA in plasma (76–79). They do not resolve the microbial source of the circulating vesicles or provide skeletal outcomes, so their relevance here is methodological rather than direct.

Unresolved-source studies. Some microbiota–EV reports do not establish whether the recovered vesicles are bacterial or host-derived. Under Rule 4, these studies remain outside the matrix. Microbiota transplantation followed by an EV readout, for example, does not by itself assign a bacterial source to the isolated particles. Such findings may inform study design, but they cannot be cited as direct evidence of bacterial-vesicle activity.

Broader bacterial-OMV evidence (adjacent). A 2026 study of Helicobacter pylori-derived OMVs showed bone localization, suppression of osteogenic differentiation, promotion of osteoclast differentiation, and aggravation of OVX osteoporosis (33). Because H. pylori is a gastric pathogen rather than a gut-resident commensal under our operational definition, and the model was non-GC osteoporosis, we retain this study as adjacent broader bacterial-OMV evidence rather than as core GMEV evidence. Its pathogenic direction also reinforces that bacterial vesicle effects are source- and context-dependent rather than uniformly osteoprotective.

6. Disease-specific mechanisms: GIOP versus GIONFH

GIOP and GIONFH share GC exposure as an upstream trigger, but their core pathophysiology diverges: GIOP is characterized by a systemic remodeling imbalance, whereas GIONFH additionally involves failure of bone–vascular coupling. We treat them as two major skeletal manifestations within the broader spectrum of glucocorticoid-associated skeletal injury. The two manifestations are grouped here for mechanistic comparison only; they retain distinct diagnostic criteria, natural histories, and therapeutic algorithms, and no unified disease entity is proposed.

6.1. GIOP: systemic bone-remodeling imbalance

GIOP follows a time-dependent remodeling disturbance: an early increase in bone resorption may accompany initiation of GC exposure, whereas sustained suppression of osteoblast function and bone formation, osteocyte injury and impaired bone quality become prominent with continued exposure, contributing to fragility (2, 3, 7). Vascular perturbation, apoptosis and oxidative stress may also contribute; they are not exclusive to GIONFH. No qualifying study has yet demonstrated a direct bacterial-EV effect in a GC-induced osteoporosis model. The 13 category B studies (S2B-02–S2B-14; Table 2) tested non-GC skeletal settings and support cross-disease plausibility only, not direct GIOP benefit. As a distinct, non-vesicular microbiome-directed comparator, a viable A. muciniphila formulation containing salidroside improved skeletal mineralization in dexamethasone-challenged zebrafish; this result cannot be assigned to bacterial EVs and does not change the count of direct GC–BEV studies (80).

  1. Osteoblast lineage. Host BMSC/MSC-derived EV studies (category C) indicate that EV cargo modulates Wnt/β-catenin, Runx2, and Osterix (66, 68, 70). Direct bacterial-vesicle evidence on osteoblasts under GC conditions is absent; the proposed bacterial-EV mechanism is hypothesis-generating rather than assigned to a matrix category.

  2. Osteoclast lineage. PAMP-bearing BEVs can activate innate immune cells through TLR-dependent pathways (43, 44), and inflammatory signaling is tightly coupled to pathological bone resorption (56). Whether gut bacterial EVs promote osteoclastogenesis via TLR/MyD88/NF-κB or alter the RANKL/OPG balance under GC exposure is plausible but untested and is therefore hypothesis-generating.

  3. Osteocyte lineage. L. animalis-derived BEVs reduced skeletal apoptosis in the anchor GIONFH model (category A) (24), and Wharton’s-jelly MSC exosomes reduced osteocyte apoptosis in GC-induced osteonecrosis (category C) (71). Neither finding constitutes direct bacterial-vesicle evidence for GIOP; proposed regulation of osteocyte SOST by gut bacterial EVs remains hypothesis-generating.

6.2. GIONFH: bone–vascular coupling failure

Vascular compromise—including endothelial dysfunction, ischemia, and impaired repair—is a prominent component of GIONFH pathophysiology alongside lipid-metabolism disturbance, osteocyte death, remodeling imbalance, and mechanical loading (4, 5, 81). The category A dataset reports improvements in femoral-head angiogenic and osteogenic measurements and reduced apoptosis-related readouts after oral cultured L. animalis-derived BEV intervention in one murine GC model (24). It does not establish that endogenous GMEVs cause bone–vascular coupling failure, nor quantify the relative causal contributions of the shared processes in GIONFH. Candidate mechanisms for bacterial-vesicle-induced BMEC injury—such as endoplasmic-reticulum stress, JNK activation, VEGF-A suppression, and endothelial apoptosis—remain hypothesis-generating because they have not been directly tested under GC + bacterial vesicle + skeletal conditions.

The juxtaposition of the two manifestations suggests, but does not establish, different dominant clinical end points despite overlapping osteocyte injury, oxidative stress, apoptosis, vascular dysfunction and abnormal remodeling: systemic fracture susceptibility in GIOP versus focal necrosis and potential subchondral failure in GIONFH. This divergence is a hypothesis. Crucially, the single category A dataset (S2B-01; Table 2, row 1) pertains to GIONFH, not GIOP, while the thirteen category B studies (S2B-02–S2B-14) did not model GC-induced skeletal disease and therefore cannot be used as direct GIONFH evidence. We identified no eligible study directly comparing bacterial-vesicle cargo or target-cell specificity across GIOP and GIONFH within one experiment (Figure 2). Direct human evidence for either manifestation is currently unavailable.

Figure 2.

Flowchart summarizes pharmacologic glucocorticoid exposure effects leading to shared cellular and tissue responses including osteoblast and osteocyte dysfunction, oxidative stress, vascular injury, remodeling imbalance, and impaired repair, which diverge into GIOP with systemic remodeling imbalance and generalized bone fragility, and GIONFH with focal osteonecrosis and possible subchondral collapse.

Shared tissue and cellular responses to pharmacologic glucocorticoid exposure and distinct skeletal manifestations of GIOP and GIONFH. GC-associated injury may involve osteoblast and osteocyte dysfunction, oxidative stress and apoptosis, vascular compromise, remodeling imbalance, and impaired repair, with overlapping contributions in both conditions. GIOP is predominantly characterized by systemic remodeling disturbance and generalized skeletal fragility; resorption may increase early, whereas suppression of bone formation persists with continued GC exposure. GIONFH involves focal femoral-head osteonecrosis and vascular injury; subchondral collapse may occur but is not inevitable. The schematic does not assign these processes to endogenous GMEVs, claim their relative causal weights are known, or portray the proposed progression as an inevitable sequence.

6.3. Osteoimmunity and the gut–immune–bone–vascular network

Gut bacterial EVs may shape the bone immune microenvironment by modulating macrophage polarization, the Treg/Th17 axis, and systemic inflammation; BEV-driven innate immune activation has been documented in non-skeletal settings (43, 44), while inflammatory bone resorption and microbiota–immune–endocrine crosstalk are documented separately (56, 57). Because most supporting observations derive from non-GC models or ex vivo systems, their application to GC-associated skeletal injury is hypothesis-generating rather than assigned to a matrix category. We therefore use the gut–immune–bone–vascular network as an integrative scaffold in which bacterial vesicles are candidate signal nodes, not confirmed central mediators.

7. Translational opportunities

Translational claims about GMEVs fall into three levels of support: demonstrated (direct experimental evidence in GC-bone contexts), preclinical extrapolation (evidence from non-GC skeletal or related-disease models), and conceptual (plausible but unsupported by direct evidence). Evidence gaps themselves are addressed in Section 8.

7.1. Candidate biomarkers: rationale, feasibility, and outstanding validation needs

Circulating GMEV signatures could, in principle, reveal microbial or host-response information that is not captured by conventional bone-turnover markers. Their appeal lies partly in the possibility of repeated, minimally invasive sampling. Oncology studies show that EV-associated RNAs can be measured in plasma (76–79), but this provides technical precedent rather than skeletal validation. Within the core evidence set, S2B-03 and S2B-08 include human observations on donor taxa or gut microbial abundance; neither links a source-resolved circulating GMEV signature to GC-associated skeletal outcomes.

The clinically relevant question is whether such a signature can improve the detection or prediction of GIOP or GIONFH beyond established risk factors. Answering it will require prospective GC-exposed cohorts; the necessary analytical and outcome-validation safeguards are discussed in Section 8.3 (37, 42).

7.2. Probiotic and dietary interventions

Modulation of the gut microbiota—and by extension the GMEV pool—through probiotics, prebiotics, or diet is currently preclinical extrapolation. In animals, specific Lactobacillus strains prevent bone loss in a clinically relevant oral GC model (82), and microbiota-dependent bone protection by probiotics has been demonstrated in sex-steroid deficiency (83, 84). Human evidence is inconsistent: a controlled randomized trial of daily Lactobacillus acidophilus supplementation in postmenopausal women did not demonstrate clinically meaningful improvement in bone mineral density (85); a systematic review likewise reported no consistent effect of probiotic supplements on BMD or bone turnover markers (86); whereas a separate systematic review and meta-analysis reported small but statistically significant benefits (87). We read this discrepancy as a caution rather than as a verdict: effects, where present, are modest and strain-, dose-, and population-dependent, and murine probiotic findings should not be extrapolated to GC-treated patients. Whether any such benefit is mediated by GMEVs remains open.

7.3. Engineered BEVs and bone-targeted delivery

Seven studies in the core set tested engineered BEVs in non-GC skeletal models (S2B-04–S2B-07, S2B-10, S2B-13 and S2B-14; Table 2). Their constructs and skeletal findings are described in Section 5.2.2. Taken together, they show that cultured bacterial vesicles can be modified for bone targeting or delivery of selected cargo, although some outcomes arise from combined formulations. The proposed therapeutic material is the purified, cell-free vesicle preparation, not the live or genetically modified producer strain. Engineering also changes the exposure and safety questions: a modified BEV cannot be assumed to behave like an endogenous GMEV. None of the engineered constructs in this core set has been tested in a GC bone model.

The next experiments need to resolve cargo loading and release, bone-targeting specificity, in vivo stability, PK–PD relationships, repeat-dose immunogenicity and manufacturing consistency. The pBMP-2-VEGF plasmid carried in S2B-10 is one example of the additional controls required for an engineered cargo; regulatory oversight of genetically modified bacterial products also remains relevant (35, 88).

Conceptual extension: adjunctive use with anti-resorptive therapy. Combining GMEVs with bisphosphonates or denosumab to enhance efficacy or reduce drug exposure has been suggested. No direct synergy evidence was identified in the eligible set, and we found no eligible study assessing whether vesicle-mediated immunomodulation modifies either drug class’s adverse-event profile. Because both the mechanism and the safety rationale are unestablished, we mention this only to mark it as a question that requires basic preclinical dose-finding before it can be framed as a hypothesis worth testing. A separate non-BEV precedent is a DSPE-lipid-coated formulation combining viable L. rhamnosus GG with alendronate, reported to improve OVX-associated osteoporosis outcomes; this does not demonstrate alendronate loading into isolated BEVs or synergy with a BEV intervention (89).

8. Translational barriers

8.1. Source attribution: the foundational technical barrier

Establishing that a circulating EV originates from a specific intestinal bacterium is the most fundamental obstacle to clinical GMEV research, and it underlies several of the limitations noted above. MISEV2023 states directly that “when obtaining bacterial EVs from in vivo and environmental sources, consider that host EVs or EVs from non-target species are likely present,” and adds a warning that bears on this review’s central question: non-specific isolation methods can co-isolate non-vesicular material including pili, flagellae, phage, and protein, lipoprotein, and nucleoprotein complexes, while LPS may be present in non-vesicular forms such as micelles or complexes with LPS-binding protein, requiring appropriate controls (34). In other words, both the vesicle and the “bacterial signal” can be misattributed unless attribution is demonstrated rather than assumed.

A complete attribution chain comprises: (i) taxonomic assignment via 16S rRNA gene sequencing, metagenomics, or strain-specific membrane proteins; (ii) contamination control to exclude co-isolated host EVs, food- and plant-derived EVs, fungal vesicles, and environmental particles; (iii) host-EV discrimination using bacterial markers (LPS, lipoteichoic acid, peptidoglycan, bacterial lipids) versus canonical host markers (CD9/CD63/CD81, Alix, TSG101), noting that validated affinity reagents exist for only a limited number of bacterial species (34); (iv) orthogonal biophysical separation to resolve the overlapping size and density ranges of bacterial and host vesicles (37); (v) validation by in vitro culture of the candidate strain with comparative proteomic, lipidomic, and RNA profiling; and (vi) functional cargo attribution linking a specific cargo activity back to the cultured donor strain.

Progress is being made: GMEVs have been shown to constitute a molecularly distinct entity from the parent microbiota (42), orthogonal separation and biochemical characterization protocols for BEVs in human body fluids have been established (37), and antibiotic-conjugated probes that selectively label outer-membrane vesicles now permit direct visualization of bacterial vesicles in blood (90). The field nonetheless lacks consensus standards, and until they exist, any claim that a given RNA, protein, or lipid is “GMEV-derived” should be treated as provisional.

8.2. Isolation, purification, and standardization

Ultracentrifugation, density-gradient ultracentrifugation, size-exclusion chromatography, and immunoaffinity capture each balance yield, purity, and vesicle integrity differently, and each enriches a partially different vesicle subpopulation (37); MISEV2023 notes that different separation methods “may enrich or deplete subtypes of bacterial EVs” (34). Because GMEVs overlap with host EVs in size, density, and—absent specific probes—surface markers, the choice of method can determine which vesicles are recovered and therefore what a study concludes. This methodological heterogeneity, rather than biological disagreement, plausibly accounts for part of the apparent inconsistency in the literature. Minimal reporting standards for particle-size distribution, particle concentration, protein-to-particle ratios, and cargo quantification are prerequisites for comparability, and BEV studies should adapt the MISEV2023 template to bacterial vesicles—including reporting culture conditions and growth phase at harvest, which are currently omitted from most skeletal studies (34).

8.3. Biomarker analytical and clinical validation

The main obstacle is not simply detecting a vesicle-associated signal, but showing that it is bacterial in origin and performs reliably as a clinical assay. None of the core studies establishes the sensitivity, specificity, reproducibility, longitudinal stability or prespecified decision threshold of a source-resolved circulating GMEV test. Human donor-taxon observations in S2B-03 and S2B-08 cannot substitute for vesicle-level validation. Assay performance must first be demonstrated, followed by prospective testing with predefined sampling windows, blinded skeletal outcome assessment and independent replication to establish predictive value beyond existing measures.

8.4. Biodistribution, PK–PD, and therapeutic delivery

Biodistribution and pharmacokinetic–pharmacodynamic characterization remain essential for the engineered-BEV pathway in Section 7.3. Several core studies demonstrate bone accumulation after systemic or oral administration, but the evidence is heterogeneous in route, dose, labeling strategy, and endpoint; S2B-04 includes an explicit biodistribution/PK comparison, whereas S2B-02 and S2B-08 demonstrate bone localization without establishing a complete quantitative exposure profile. For the category A anchor (S2B-01; Table 2, row 1), fluorescence was detected in the femoral head, but the fraction of administered dose reaching bone, the intestinal transit route, and clearance kinetics were not determined. Future studies therefore need quantitative biodistribution, tissue exposure, release kinetics, dose–response relationships, and PK–PD linkage before bone delivery can be treated as an established therapeutic property.

8.5. Safety, immunogenicity, and scalable manufacturing

Safety considerations differ by donor-cell envelope architecture and by whether vesicles are native or engineered. Gram-negative bacterial vesicles carry LPS and other PAMPs that may provoke systemic inflammation; Gram-positive-derived vesicles—including lactic acid bacteria represented in the core evidence set—raise distinct immunogenicity concerns through lipoteichoic acid, peptidoglycan-associated components, and other PAMPs, and cannot be assumed safe by analogy (18, 20, 35). Surface modification and exogenous cargo loading may further alter immune recognition and pharmacology. Long-term toxicity, repeat-dose immunogenicity, and batch-to-batch consistency remain insufficiently characterized, while GMP-compliant scale-up requires control of producer strain, culture conditions, vesicle composition, potency, and contaminants (34, 35, 88).

8.6. Interspecies and human validation

The final translational barrier is disease- and species-specific validation. The core evidence set contains one category A murine GIONFH study and thirteen category B non-GC skeletal studies; no category A GIOP study and no direct human GMEV–skeletal outcome study were identified. Although S2B-03 and S2B-08 include human observational components, these do not establish causality, circulating GMEV source attribution, or clinical prediction. Humanized microbiota mice, gut–bone organoid co-cultures, and prospective longitudinal GC cohorts can bridge experimental and clinical evidence, but each has limitations: humanized mice retain murine immunity and bone biology, whereas organoids lack systemic, pharmacokinetic, and mechanical inputs. Disease-specific validation should therefore distinguish GIOP from GIONFH and test the biomarker and therapeutic pathways separately.

9. A proposed experimental roadmap

The seven priorities below address gaps that cannot be resolved by a single experiment. They are interdependent rather than sequential clinical stages, with biomarker validation and engineered-BEV development proceeding along separate but related paths (Figure 3).

Figure 3.

Infographic outlining validation prerequisites, candidate biomarker and engineered therapeutic pathways, disease endpoints, and seven research priorities for GC–BEV studies. Visuals represent vesicle discrimination, experimental steps, bone outcomes, and numbered research themes, emphasizing the need for both human and interspecies validation.

Proposed parallel translational pathways for gut-associated bacterial extracellular vesicles in GC-related skeletal disorders. Shared prerequisites include discrimination of bacterial from host vesicle signals, orthogonal vesicle characterization, reproducible preparation, and separate GIOP and GIONFH clinical endpoints. The candidate biomarker arm requires source-resolved circulating GMEV measurements, analytical validation, prospective GC-exposed cohorts, and demonstration of clinical predictive value beyond established risk factors. The distinct therapeutic arm concerns purified, cell-free engineered BEV candidates requiring verified cargo and targeting, biodistribution and PK–PD evaluation, safety/manufacturing characterization, and disease-specific efficacy testing. The seven Section 9 research priorities are interdependent, cross-cutting checkpoints rather than seven completed or sequential clinical stages. Neither pathway has established human clinical validity or efficacy; connecting arrows represent proposed development relationships only.

  1. Establish gold-standard source attribution. Develop bacteria-specific surface-marker panels and adopt single-vesicle analytical techniques, reporting the full attribution chain of Section 8.1 and the culture parameters specified by MISEV2023.

  2. Generate direct GC + bacterial-vesicle + bone evidence. Replicate the anchor GIONFH finding in independent laboratories and with additional strains, including taxa prevalent in the human gut. In parallel, establish a direct GIOP model so that GC-specific bacterial-vesicle evidence is no longer confined to osteonecrosis.

  3. Resolve vesicle cargo–target-cell relationships. Identify the effector molecule and the responding cell type, and compare proteomic, small-RNA, and lipid profiles across GIOP and GIONFH within a single experiment.

  4. Separate prevention from treatment. Determine whether bacterial vesicles modify established disease, not only its induction, since the anchor study used a prophylactic paradigm and did not establish reversal of established disease.

  5. Develop bone-targeted BEV delivery. Evaluate engineered BEVs for targeting efficiency, biodistribution and PK–PD (Section 8.4), immunogenicity and manufacturing feasibility (Section 8.5), with explicit comparison against host-EV benchmarks.

  6. Validate candidate biomarkers prospectively. Embed nested case–control studies within GC-treated cohorts, measuring source-resolved GMEV signatures alongside barrier-injury markers such as I-FABP and skeletal endpoints, with analytical validation and clinical validation as specified in Section 8.3.

  7. Build multi-omics integration models. Combine metagenomics, metabolomics, GMEV-omics, and host transcriptomics to construct a gut–immune–bone–vascular regulatory network that can be perturbed and re-measured, while retaining the source-attribution and human-validation requirements of Sections 8.1 and 8.6.

10. Discussion

10.1. Principal findings and interpretation

The most direct experimental evidence in glucocorticoid-associated skeletal injury comes from Chen et al., who showed in one murine osteonecrosis model that GC exposure depleted L. animalis-associated signals and that replenishment with cultured L. animalis-derived BEVs attenuated femoral-head osteonecrosis with concurrent angiogenic, osteogenic and apoptosis-related changes (24). Supplementary Figure S8 did not show notable benefit from late-stage treatment; the positive intervention evidence is therefore better described as preventive/early-stage proof of concept for GIONFH, not direct evidence for GIOP, and it does not establish a generalized or replicated human gut bacterial-EV mechanism. Beyond this anchor, thirteen non-GC skeletal studies using community-enriched, cultured intestinal-associated, or engineered bacterial-vesicle preparations report bone-associated labeling or changes in skeletal outcomes (29–32, 38, 58–65). These studies broaden biological and translational plausibility but remain outside the direct GC-specific evidence category.

10.2. Current controversies and research gaps

Several interpretive problems cut across the evidence. Vesicles obtained from cultured gut-associated bacteria can be traced to a laboratory producer, but they should not be equated with naturally circulating intestinal GMEVs. GC effects on the intestinal barrier vary by model, and host-style miRNA names do not establish bacterial cargo origin. The absence of notable protection in Chen et al.’s tested late-stage arm also limits claims about treatment of established disease. Source boundaries remain consequential: gastric- and oral-source OMVs were considered adjacent, gut-associated opportunists were eligible, and the reported producer identity in S2B-14 remains unresolved.

Progress will depend on connecting observations that have so far been studied separately. In GC-treated cohorts, source-resolved circulating GMEV signals should be measured alongside intestinal barrier markers and prospectively ascertained skeletal outcomes. Experimental comparisons of GIOP and GIONFH could then ask whether the relevant donor taxa, vesicle cargo and responding bone or vascular cells differ between phenotypes. These questions cannot be answered convincingly without source controls, quantitative biodistribution and PK–PD information, and preparation-specific safety data.

10.3. Limitations of this review

This critical narrative review has limitations that affect the interpretation of its negative and exclusivity claims. One author screened 1,537 P1–P3 title/abstract records, assessed 41 priority candidates and verified the full texts and extractions of all 14 core studies, without an independent second screener. The 7,641 P4-only records were used for targeted omission checking rather than individually screened, and the 41 dispositions are not a full-text flow for all 9,178 deduplicated records. Searches covered PubMed, Web of Science Core Collection and Scopus, excluded records lacking an English abstract, and used a publication cutoff of 15 September 2026. Selection or extraction errors and missed studies therefore remain possible.

The appraisal of study methods was descriptive, not a formal study-wide risk-of-bias assessment. Table 3 draws on four source-checked reports, including the Supplementary Files for Chen, Liu and Ji, Liu’s 16S taxonomic workbooks and Lu’s publisher reporting information; without prospective protocols, selective outcome reporting cannot be fully assessed. Direct GC-associated bacterial-vesicle skeletal evidence is confined to a single murine GIONFH study. The 14-report core is also limited by an operational intestinal-associated source definition: the gastric-source H. pylori OMV study was kept as adjacent evidence, whereas cultured gut-associated P. mirabilis OMVs were eligible. For S2B-14, the reported name “Pseudomonas mirabilis” conflicts with the Proteus mirabilis attribution of the cited ATCC 12453 strain, and the experimental strain has not been independently verified. These qualifications should temper inferences about evidence gaps, generalizability and causation.

10.4. Clinical and translational implications

The immediate clinical implication is caution rather than therapeutic substitution. Existing evidence-based management of GIOP—including fracture-risk assessment, bone mineral density and vertebral imaging, and risk-adapted pharmacologic therapy—should remain the standard of care (8). Bacterial EVs are not ready to replace established anti-resorptive or anabolic treatment. Their nearer-term translational value lies in clarifying GC-associated gut–bone mechanisms, developing source-resolved biomarkers, and identifying adjunctive strategies that can be tested only after reproducibility, biodistribution, dosing, and safety are established. The experimental priorities outlined in Section 9 identify interdependent research priorities, from independent replication of the murine GIONFH observation to source-resolved prospective human validation, rather than seven sequential clinical-development stages.

11. Conclusions

Gut bacterial EVs are plausible components of a glucocorticoid-associated gut–immune–bone–vascular network, but the evidence remains at an early stage. Direct support is confined to one murine GIONFH proof-of-concept dataset using cultured L. animalis-derived BEVs; no eligible GC-associated bacterial-vesicle GIOP study was identified within the defined source-and-search framework, and no eligible source-resolved human study linking circulating GMEVs to GC-associated skeletal outcomes was identified in the documented search. The most defensible priorities are independent replication, rigorous source attribution, testing of human-prevalent donor taxa, and prospective clinical validation. Until those conditions are met, bacterial EVs should be regarded as candidate mediators and research tools rather than established drivers, biomarkers, or therapeutic targets of glucocorticoid-associated skeletal disease.

Funding Statement

The author(s) declared that financial support was received for this work and/or its publication. This work was supported by the 2022 National Young Qihuang Scholars Training Program (Grant No. [2022]256), the Henan Provincial Science and Technology Research Program (Grant Nos. 232102310423 and SBGJ202502045), and research grants from the Henan Provincial Administration of Traditional Chinese Medicine (Grant Nos. GZY-KJS-2025-104, 2022ZYZD14, 2024ZY2115, 2024ZY2108, and 2025ZY2022). The funders had no role in study design, data collection, data analysis or interpretation, manuscript preparation, or the decision to submit the work for publication.

Edited by: Subhashis Pal, SRM Medical College Hospital and Research Centre, India

Reviewed by: Da Sun, Wenzhou University, China

Ke You, University of Macau, Macao SAR, China

BEV, bacterial extracellular vesicle; BMEC, bone microvascular endothelial cell; BMSC, bone marrow mesenchymal stromal cell; cBEV, cultured strain-derived bacterial extracellular vesicle; GC, glucocorticoid; GIOP, glucocorticoid-induced osteoporosis; GIONFH, glucocorticoid-associated osteonecrosis of the femoral head; GMEV, gut microbiota-derived extracellular vesicle; GMP, good manufacturing practice; PAMP, pathogen-associated molecular pattern; PK–PD, pharmacokinetic–pharmacodynamic.

Author contributions

ZL: Conceptualization, Methodology, Investigation, Writing – original draft, Writing – review & editing. XW: Conceptualization, Methodology, Investigation, Writing – review & editing. SZ: Investigation, Writing – review & editing. LL: Investigation, Writing – review & editing. LZ: Investigation, Writing – review & editing. HG: Investigation, Writing – review & editing. NS: Visualization, Writing – review & editing. JL: Visualization, Writing – review & editing. BS: Supervision, Writing – review & editing. XY: Supervision, Writing – review & editing.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fendo.2026.2009751/full#supplementary-material

Table1.docx (62KB, docx)

References

  • 1. Weinstein RS. Clinical practice. Glucocorticoid-induced bone disease. N Engl J Med. (2011) 365:62–70. doi:  10.1056/NEJMcp1012926 [DOI] [PubMed] [Google Scholar]
  • 2. Chotiyarnwong P, McCloskey EV. Pathogenesis of glucocorticoid-induced osteoporosis and options for treatment. Nat Rev Endocrinol. (2020) 16:437–47. doi:  10.1038/s41574-020-0341-0 [DOI] [PubMed] [Google Scholar]
  • 3. Canalis E, Mazziotti G, Giustina A, Bilezikian JP. Glucocorticoid-induced osteoporosis: pathophysiology and therapy. Osteoporos Int. (2007) 18:1319–28. doi:  10.1007/s00198-007-0394-0 [DOI] [PubMed] [Google Scholar]
  • 4. Zhang J, Cao J, Liu Y, Zhao H. Advances in the pathogenesis of steroid-associated osteonecrosis of the femoral head. Biomolecules. (2024) 14:667. doi:  10.3390/biom14060667 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Li L, Zhao S, Leng Z, Chen S, Shi Y, Shi L, et al. Pathological mechanisms and related markers of steroid-induced osteonecrosis of the femoral head. Ann Med. (2024) 56:2416070. doi:  10.1080/07853890.2024.2416070 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. den Uyl D, Bultink IE, Lems WF. Advances in glucocorticoid-induced osteoporosis. Curr Rheumatol Rep. (2011) 13:233–40. doi:  10.1007/s11926-011-0173-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. Chen M, Fu W, Xu H, Liu CJ. Pathogenic mechanisms of glucocorticoid-induced osteoporosis. Cytokine Growth Factor Rev. (2023) 70:54–66. doi:  10.1016/j.cytogfr.2023.03.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Humphrey MB, Russell L, Danila MI, Fink HA, Guyatt G, Cannon M, et al. 2022 american college of rheumatology guideline for the prevention and treatment of glucocorticoid-induced osteoporosis. Arthritis Care Res (Hoboken). (2023) 75:2405–19. doi:  10.1002/acr.25240 [DOI] [PubMed] [Google Scholar]
  • 9. Lyu Z, Hu Y, Guo Y, Liu D. Modulation of bone remodeling by the gut microbiota: a new therapy for osteoporosis. Bone Res. (2023) 11:31. doi:  10.1038/s41413-023-00264-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Sjogren K, Engdahl C, Henning P, Lerner UH, Tremaroli V, Lagerquist MK, et al. The gut microbiota regulates bone mass in mice. J Bone Miner Res. (2012) 27:1357–67. doi:  10.1002/jbmr.1588 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Yan J, Herzog JW, Tsang K, Brennan CA, Bower MA, Garrett WS, et al. Gut microbiota induce IGF-1 and promote bone formation and growth. Proc Natl Acad Sci USA. (2016) 114:E475–84. doi:  10.1073/pnas.1607235113 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12. Lucas S, Omata Y, Hofmann J, Böttcher M, Iljazovic A, Sarter K, et al. Short-chain fatty acids regulate systemic bone mass and protect from pathological bone loss. Nat Commun. (2018) 9:55. doi:  10.1038/s41467-017-02490-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Shi H, Huang L, Zhang JH, Shen C, Zhang N, Lv C, et al. Gut microbiota regulates brain-bone axis to influence osteoporosis pathogenesis and treatment. Res (Wash D C). (2026) 9:1178. doi:  10.34133/research.1178 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Liu H, Li R, Yang H, Situ B, Wang G, Xu K, et al. Extracellular vesicles in gut-bone axis: novel insights and therapeutic opportunities for osteoporosis. Small Sci. (2025) 5:2400474. doi:  10.1002/smsc.202400474 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Han Y, Guo Y, Xu W, Zhang Y, Hao H, Song Y, et al. Gut microbiota-derived extracellular vesicles: novel perspectives from host-microbe interactions to therapeutic applications. J Physiol Biochem. (2026) 82:32. doi:  10.1007/s13105-026-01175-1 [DOI] [PubMed] [Google Scholar]
  • 16. Cheung KCP, Jiao M, Xingxuan C, Wei J. Extracellular vesicles derived from host and gut microbiota as promising nanocarriers for targeted therapy in osteoporosis and osteoarthritis. Front Pharmacol. (2022) 13:1051134. doi:  10.3389/fphar.2022.1051134 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Yin G, Wang H, Liu Y, Wang Z, Lin X, Cheng Q, et al. Mammalian, plant, and gut microbiota-derived extracellular vesicles as emerging therapeutics for bone diseases. Front Bioeng Biotechnol. (2026) 14:1825617. doi:  10.3389/fbioe.2026.1825617 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Avila-Calderon ED, Ruiz-Palma MDS, Aguilera-Arreola MG, Velázquez-Guadarrama N, Ruiz EA, Gomez-Lunar Z, et al. Outer membrane vesicles of Gram-negative bacteria: an outlook on biogenesis. Front Microbiol. (2021) 12:557902. doi:  10.3389/fmicb.2021.557902 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Sartorio MG, Pardue EJ, Feldman MF, Haurat MF. Bacterial outer membrane vesicles: from discovery to applications. Annu Rev Microbiol. (2021) 75:609–30. doi:  10.1146/annurev-micro-052821-031444 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. Liang X, Dai N, Sheng K, Lu H, Wang J, Chen L, et al. Gut bacterial extracellular vesicles: important players in regulating intestinal microenvironment. Gut Microbes. (2022) 14:2134689. doi:  10.1080/19490976.2022.2134689 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Jain H, Kumar A, Almousa S, Mishra S, Langsten KL, Kim S, et al. Characterisation of LPS+ bacterial extracellular vesicles along the gut-hepatic portal vein-liver axis. J Extracell Vesicles. (2024) 13:e12474. doi:  10.1002/jev2.12474 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Champagne-Jorgensen K, Jose TA, Stanisz AM, Mian MF, Hynes AP, Bienenstock J. Bacterial membrane vesicles and phages in blood after consumption of Lacticaseibacillus rhamnosus JB-1. Gut Microbes. (2021) 13:1993583. doi:  10.1080/19490976.2021.1993583 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Jones EJ, Parker A, Juodeikis R, Blackshaw LA, Brion A, Carding SR. Special delivery - extracellular vesicles released by commensal gut bacteria deliver bioactive protein to distal organs. Extracell Vesicles Circ Nucl Acids. (2025) 6:791–806. doi:  10.20517/evcna.2025.32 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24. Chen CY, Rao SS, Yue T, Tan YJ, Yin H, Chen LJ, et al. Glucocorticoid-induced loss of beneficial gut bacterial extracellular vesicles is associated with the pathogenesis of osteonecrosis. Sci Adv. (2022) 8:eabg8335. doi:  10.1126/sciadv.abg8335 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25. Liang P, Chen X, Su Z, Luo Y, Wang T, Li J, et al. The role of oral and gut microbiota in bone health: insights from bacterial extracellular vesicles. Microorganisms. (2025) 13:2254. doi:  10.3390/microorganisms13102254 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Parrotta EI, Benedetto GL, Cuda G, Longo UG, Carnevale A, Galasso O, et al. Extracellular vesicles in osteonecrosis of the femoral head: an integrated review of experimental and bioinformatic evidence. J Pers Med. (2026) 16:208. doi:  10.3390/jpm16040208 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Wang Y, Xu M, Dong Q, Lin X, Zhang Q, Shi H, et al. The gut-bone axis: impact of diet on gut microbiome and osteoporosis. Bone Res. (2026) 14:61. doi:  10.1038/s41413-026-00550-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Zhou R, Bian Y, Cai X, Sun H, Lv Z, Xu Y, et al. Advances in the mechanism for steroid-induced osteonecrosis of the femoral head. Bone Res. (2026) 14:23. doi:  10.1038/s41413-025-00477-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Wang T, Mo L, Ou J, Fang Q, Wu H, Wu Y, et al. Proteus mirabilis Vesicles Induce Mitochondrial Apoptosis by Regulating miR96-5p/Abca1 to Inhibit Osteoclastogenesis and Bone Loss. Front Immunol. (2022) 13:833040. doi:  10.3389/fimmu.2022.833040 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30. Lei LM, Lin X, Xu F, Li FXZ, Shan SK, Zheng MH, et al. Cold exposure-induced loss of beneficial gut bacterial sEV is associated with the pathogenesis of bone loss through pyroptosis pathway. J Nanobiotechnology. (2026) 24:844. doi:  10.1186/s12951-026-04719-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Liu H, Song P, Zhang H, Zhou F, Ji N, Wang M, et al. Synthetic biology-based bacterial extracellular vesicles displaying BMP-2 and CXCR4 to ameliorate osteoporosis. J Extracell Vesicles. (2024) 13:e12429. doi:  10.1002/jev2.12429 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32. Lin S, Chen C, Zheng Y, Wu B, Wu W. Engineered pseudomonas mirabilis-derived outer membrane vesicles targeting bone microenvironment to improve osteoporosis. Biomedicines. (2025) 13:847. doi:  10.3390/biomedicines13040847 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33. Zhao J, Liu J, Huang S, Ye S, Zhao S. Helicobacter pylori-secreted outer membrane vesicles induce bone loss and accelerate osteoporosis pathogenesis in mice by regulating gut microbiota and autophagy. Microbiol Res. (2026) 307:128485. doi:  10.1016/j.micres.2026.128485 [DOI] [PubMed] [Google Scholar]
  • 34. Welsh JA, Goberdhan DCI, O'Driscoll L, Buzas EI, Blenkiron C, Bussolati B, et al. Minimal information for studies of extracellular vesicles (MISEV2023): from basic to advanced approaches. J Extracell Vesicles. (2024) 13:e12404. doi:  10.1002/jev2.12404 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35. Qian D, Xu P, Wang X, Du C, Zhao X, Xu J. Bacterial extracellular vesicles for gut microbiome-host communication and drug development. Acta Pharm Sin B. (2025) 15:1816–40. doi:  10.1016/j.apsb.2025.03.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36. Hooijmans CR, Rovers MM, de Vries RBM, Leenaars M, Ritskes-Hoitinga M, Langendam MW. SYRCLE's risk of bias tool for animal studies. BMC Med Res Methodol. (2014) 14:43. doi:  10.1186/1471-2288-14-43 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37. Tulkens J, De Wever O, Hendrix A. Analyzing bacterial extracellular vesicles in human body fluids by orthogonal biophysical separation and biochemical characterization. Nat Protoc. (2020) 15:40–67. doi:  10.1038/s41596-019-0236-5 [DOI] [PubMed] [Google Scholar]
  • 38. Ji N, Zhao J, Mao Y, Zhang J, Chen X, Hu Y, et al. Engineered Akkermansia muciniphila extracellular vesicles for targeted delivery of miR-21-5p alleviate postmenopausal osteoporosis via PI3K-AKT pathway. J Nanobiotechnology. (2026) 24:624. doi:  10.1186/s12951-026-04513-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39. Konig J, Wells J, Cani PD, García-Ródenas CL, MacDonald T, Mercenier A, et al. Human intestinal barrier function in health and disease. Clin Transl Gastroenterol. (2016) 7:e196. doi:  10.1038/ctg.2016.54 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Chronopoulos A, Kalluri R. Emerging role of bacterial extracellular vesicles in cancer. Oncogene. (2020) 39:6951–60. doi:  10.1038/s41388-020-01509-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41. Ishizaka A, Koga M, Hayashi T, Ishii KJ, Yamamoto H, Yotsuyanagi H, et al. Gut microbiota-derived bacterial extracellular vesicles in COVID-19: their signature and immunological impact. J Extracell Vesicles. (2026) 15:e70341. doi:  10.1002/jev2.70341 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42. Kaisanlahti A, Turunen J, Hekkala J, Mishra S, Karikka S, Amatya SB, et al. Gut microbiota-derived extracellular vesicles form a distinct entity from gut microbiota. mSystems. (2025) 10:e0031125. doi:  10.1128/msystems.00311-25 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43. Collins SM, Brown AC. Bacterial outer membrane vesicles as antibiotic delivery vehicles. Front Immunol. (2021) 12:733064. doi:  10.3389/fimmu.2021.733064 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44. Song M, Zhou W, Fan J, Jia C, Xiong W, Wei H, et al. Diarrheal microbiota-derived extracellular vesicles drive intestinal homeostasis dysfunction via miR-125b/NF-kB-mediated macrophage polarization. Gut Microbes. (2025) 17:2541036. doi:  10.1080/19490976.2025.2541036 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45. Li C, Peng J, Xiao L, Wu H, Chen J, Chen N. Pseudomonas aeruginosa-derived extracellular vesicles enhance macrophage aerobic glycolysis that fuels inflammation. Front Microbiol. (2025) 16:1619101. doi:  10.3389/fmicb.2025.1619101 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46. Schepper JD, Collins F, Rios-Arce ND, Kang HJ, Schaefer L, Gardinier JD, et al. Involvement of the gut microbiota and barrier function in glucocorticoid-induced osteoporosis. J Bone Miner Res. (2020) 35:801–14. doi:  10.1002/jbmr.3947 [DOI] [PubMed] [Google Scholar]
  • 47. Huang C, Zhao Y, Qing L, Tang J, Min C, Wu P. Exploring the alteration of gut microbiota in the rats with steroid-induce osteonecrosis of the femoral head. BMC Microbiol. (2025) 25:725. doi:  10.1186/s12866-025-03897-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48. Zheng QY, Tao Y, Geng L, Ren P, Ni M, Zhang GQ. Non-traumatic osteonecrosis of the femoral head induced by steroid and alcohol exposure is associated with intestinal flora alterations and metabolomic profiles. J Orthop Surg Res. (2024) 19:296. doi:  10.1186/s13018-024-04713-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49. Ferrini A, He M, Mont MA, Goodman SB, Parvizi J. Osteonecrosis of the femoral head: a dysbiotic condition? J Arthroplasty. (2026) 41:2755–9. doi:  10.1016/j.arth.2025.11.044 [DOI] [PubMed] [Google Scholar]
  • 50. Tan J, Taitz JJ, Ni D, Potier-Villette C, Pinget G, Pulpitel T, et al. Gut microbiota-derived extracellular vesicles exhibit diurnal regulation and activate hepatic gluconeogenesis. Mol Metab. (2025) 98:102180. doi:  10.1016/j.molmet.2025.102180 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51. Tan J, Ni D, Taitz J, Pinget GV, Read M, Senior A, et al. Dietary protein increases T-cell-independent sIgA production through changes in gut microbiota-derived extracellular vesicles. Nat Commun. (2022) 13:4336. doi:  10.1038/s41467-022-31761-y [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52. Fischer A, Gluth M, Weege F, Pape UF, Wiedenmann B, Baumgart DC, et al. Glucocorticoids regulate barrier function and claudin expression in intestinal epithelial cells via MKP-1. Am J Physiol Gastrointest Liver Physiol. (2014) 306:G218–28. doi:  10.1152/ajpgi.00095.2013 [DOI] [PubMed] [Google Scholar]
  • 53. Boivin MA, Ye D, Kennedy JC, Al-Sadi R, Shepela C, Ma TY. Mechanism of glucocorticoid regulation of the intestinal tight junction barrier. Am J Physiol Gastrointest Liver Physiol. (2007) 292:G590–8. doi:  10.1152/ajpgi.00252.2006 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54. Shukla PK, Meena AS, Pierre JF, Rao R. Central role of intestinal epithelial glucocorticoid receptor in alcohol- and corticosterone-induced gut permeability and systemic response. FASEB J. (2022) 36:e22061. doi:  10.1096/fj.202101424R [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55. Zheng G, Wu SP, Hu Y, Smith DE, Wiley JW, Hong S, et al. Chronic stress and intestinal barrier dysfunction: glucocorticoid receptor and transcription repressor HES1 regulate tight junction protein Claudin-1 promoter. Sci Rep. (2017) 7:4502. doi:  10.1038/s41598-017-04755-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56. Terkawi MA, Matsumae G, Shimizu T, Takahashi D, Kadoya K, Iwasaki N. Interplay between inflammation and pathological bone resorption: insights into recent mechanisms and pathways in related diseases for future perspectives. Int J Mol Sci. (2022) 23:1786. doi:  10.3390/ijms23031786 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57. Xie X, Chen X, Wang Z, Chen Y, Li J. The role of gut microbiota-immune-endocrine crosstalk in the pathogenesis of osteoporosis. Front Immunol. (2026) 17:1813653. doi:  10.3389/fimmu.2026.1813653 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 58. Liu JH, Chen CY, Liu ZZ, Luo Z, Rao S, Jin L, et al. Extracellular vesicles from child gut microbiota enter into bone to preserve bone mass and strength. Adv Sci (Weinh). (2021) 8:2004831. doi:  10.1002/advs.202004831 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59. Lu J, Liu Y, Chen J, Yu S, Yang S, Huang C, et al. Roseburia intestinalis-derived extracellular vesicles alleviate osteoporosis through gut microbiota-regulated histidylleucine via Akt/FOXO1 signaling. Commun Biol. (2026) 9:1155. doi:  10.1038/s42003-026-10401-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60. Zhang W, Xie J, Wang Z, Zhong Y, Liu L, Liu J, et al. Androgen deficiency-induced loss of Lactobacillus salivarius extracellular vesicles is associated with the pathogenesis of osteoporosis. Microbiol Res. (2025) 293:128047. doi:  10.1016/j.micres.2025.128047 [DOI] [PubMed] [Google Scholar]
  • 61. Wang S, Hu Y, Song P, Lin Q, Tang X, Gao Q, et al. Harnessing extracellular vesicles from Lactobacillus reuteri and Lactobacillus paracasei for synergistic osteoporosis therapy. Compos Part B Eng. (2025) 297:112255. doi:  10.1016/j.compositesb.2025.11225542574925 [DOI] [Google Scholar]
  • 62. Liu H, Zhang H, Wang S, Cui J, Weng W, Liu X, et al. Bone-targeted bioengineered bacterial extracellular vesicles delivering siRNA to ameliorate osteoporosis. Compos Part B Eng. (2023) 255:110610. doi:  10.1016/j.compositesb.2023.11061042574925 [DOI] [Google Scholar]
  • 63. Liu H, Wu Y, Wang F, Wang S, Ji N, Wang M, et al. Bone-targeted engineered bacterial extracellular vesicles delivering miRNA to treat osteoporosis. Compos Part B Eng. (2023) 267:111047. doi:  10.1016/j.compositesb.2023.11104742574925 [DOI] [Google Scholar]
  • 64. Kong X, Liu H, Chen S, Liu Z, Chen Q, Li X, et al. Bioengineered bacterial extracellular vesicles for targeted delivery of an osteoclastogenesis-inhibitory peptide to alleviate osteoporosis. J Control Release. (2025) 382:113751. doi:  10.1016/j.jconrel.2025.113751 [DOI] [PubMed] [Google Scholar]
  • 65. Zhou G, Zhou Q, Li R, Sheng S, Gao Q, Zhou D, et al. Synthetically engineered bacterial extracellular vesicles and IL-4-encapsulated hydrogels sequentially promote osteoporotic fracture repair. ACS Nano. (2025) 19:16064–83. doi:  10.1021/acsnano.5c03106 [DOI] [PubMed] [Google Scholar]
  • 66. Duan P, Yu YL, Cheng YN, Nie MH, Yang Q, Xia LH, et al. Exosomal miR-1a-3p derived from glucocorticoid-stimulated M1 macrophages promotes the adipogenic differentiation of BMSCs in glucocorticoid-associated osteonecrosis of the femoral head by targeting Cebpz. J Nanobiotechnology. (2024) 22:648. doi:  10.1186/s12951-024-02923-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 67. Fang S, Li Y, Chen P. Osteogenic effect of bone marrow mesenchymal stem cell-derived exosomes on steroid-induced osteonecrosis of the femoral head. Drug Des Devel Ther. (2019) 13:45–55. doi:  10.2147/DDDT.S178698 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68. Zhang C, Su Y, Ding H, Yin J, Zhu Z, Song W. Mesenchymal stem cells-derived and siRNAs-encapsulated exosomes inhibit osteonecrosis of the femoral head. J Cell Mol Med. (2020) 24:9605–12. doi:  10.1111/jcmm.15395 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 69. Zheng G, Ma HW, Xiang GH, He GL, Cai HC, Dai ZH, et al. Bone-targeting delivery of platelet lysate exosomes ameliorates glucocorticoid-induced osteoporosis by enhancing bone-vessel coupling. J Nanobiotechnology. (2022) 20:220. doi:  10.1186/s12951-022-01400-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 70. Zhao ZM, Ding JM, Li Y, Wang DC, Kuang MJ. Human umbilical cord mesenchymal stem cell-derived exosomes promote osteogenesis in glucocorticoid-induced osteoporosis through PI3K/AKT signaling pathway-mediated ferroptosis inhibition. Stem Cells Transl Med. (2025) 14:szae096. doi:  10.1093/stcltm/szae096 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 71. Kuang MJ, Huang Y, Zhao XG, Zhang R, Ma JX, Wang DC, et al. Exosomes derived from Wharton's jelly of human umbilical cord mesenchymal stem cells reduce osteocyte apoptosis in glucocorticoid-induced osteonecrosis of the femoral head in rats via the miR-21-PTEN-AKT signalling pathway. Int J Biol Sci. (2019) 15:1861–70. doi:  10.7150/ijbs.32262 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 72. Li S, Guo P, Wang L, Zhang Y, Li S, Liu C, et al. miR-127-5p-enriched extracellular vesicles modulate the Hippo pathway to counteract steroid-induced osteonecrosis pathology of the femoral head. Free Radic Biol Med. (2025) 238:90–104. doi:  10.1016/j.freeradbiomed.2025.06.031 [DOI] [PubMed] [Google Scholar]
  • 73. Chen P, Wang R, Fang S. Metabolic glycoengineered exosome-A2M nanoplatform reprograms macrophage polarization and orchestrates bone regeneration in ONFH. Cell Death Discov. (2025) 11:510. doi:  10.1038/s41420-025-02690-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 74. Rao SS, Zeng HJ, Wang Z, Hong CG, Tan YJ, Duan YX, et al. Amuc_1473 links gut microbes to skeletal homeostasis and counteracts multifactorial osteoporosis. Adv Sci (Weinh). (2026) 13:e23067. doi:  10.1002/advs.202523067 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 75. He S, Chen H, Xi H, Sun G, Du B, Liu X, et al. Gut microbiota-derived butyric acid alleviates glucocorticoid-associated osteonecrosis of the femoral head via modulating inflammatory cytokines in bone marrow mesenchymal stem cells. Mediators Inflammation. (2025) 2025:8742817. doi:  10.1155/mi/8742817 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 76. Drula R, Ott LF, Berindan-Neagoe I, Pantel K, Calin GA. MicroRNAs from liquid biopsy derived extracellular vesicles: recent advances in detection and characterization methods. Cancers (Basel). (2020) 12:2009. doi:  10.3390/cancers12082009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 77. LeBleu VS, Kalluri R. Exosomes as a multicomponent biomarker platform in cancer. Trends Cancer. (2020) 6:767–74. doi:  10.1016/j.trecan.2020.03.007 [DOI] [PubMed] [Google Scholar]
  • 78. Urabe F, Kosaka N, Ito K, Kimura T, Egawa S, Ochiya T. Extracellular vesicles as biomarkers and therapeutic targets for cancer. Am J Physiol Cell Physiol. (2020) 318:C29–39. doi:  10.1152/ajpcell.00280.2019 [DOI] [PubMed] [Google Scholar]
  • 79. Ozawa PMM, Jucoski TS, Vieira E, Carvalho T, Malheiros D, Ribeiro EMSF. Liquid biopsy for breast cancer using extracellular vesicles and cell-free microRNAs as biomarkers. Transl Res. (2020) 223:40–60. doi:  10.1016/j.trsl.2020.04.002 [DOI] [PubMed] [Google Scholar]
  • 80. Ma J, Ruan J, Yi J, Zhang S, Ma Y, Chen R, et al. Delivery system based on Akkermansia muciniphila loaded salidroside for the treatment of osteoporosis in zebrafish model. Adv Healthc Mater. (2026) 15:e04437. doi:  10.1002/adhm.202504437 [DOI] [PubMed] [Google Scholar]
  • 81. Yang Y, Jian Y, Liu Y, Ma M, Guo J, Xu B, et al. Mitochondrial maintenance as a novel target for treating steroid-induced osteonecrosis of femoral head: a narrative review. EFORT Open Rev. (2024) 9:1013–22. doi:  10.1530/EOR-24-0023 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 82. Chargo NJ, Schepper JD, Rios-Arce N, Kang HJ, Gardinier JD, Parameswaran N, et al. Lactobacillus reuteri 6475 prevents bone loss in a clinically relevant oral model of glucocorticoid-induced osteoporosis in male CD-1 mice. JBMR Plus. (2023) 7:e10805. doi:  10.1002/jbm4.10805 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 83. Li JY, Chassaing B, Tyagi AM, Vaccaro C, Luo T, Adams J, et al. Sex steroid deficiency-associated bone loss is microbiota dependent and prevented by probiotics. J Clin Invest. (2016) 126:2049–63. doi:  10.1172/JCI86062 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 84. Abboud M, Papandreou D. Gut microbiome, probiotics and bone: an updated mini review. Open Access Maced J Med Sci. (2019) 7:473–6. doi:  10.3889/oamjms.2019.047 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 85. Harahap IA, Moszak M, Czlapka-Matyasik M, Skrypnik K, Bogdański P, Suliburska J. Effects of daily probiotic supplementation with Lactobacillus acidophilus on calcium status, bone metabolism biomarkers, and bone mineral density in postmenopausal women: a controlled and randomized clinical study. Front Nutr. (2024) 11:1401920. doi:  10.3389/fnut.2024.1401920 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 86. Harini JN, Gayathri G, Mahadevan S, Ilangovan R. Effects of probiotic supplements on bone mineral density and bone turnover markers in postmenopausal women: a systematic review. Clin Nutr ESPEN. (2025) 69:503–15. doi:  10.1016/j.clnesp.2025.07.1117 [DOI] [PubMed] [Google Scholar]
  • 87. Wang F, Wei W, Liu PJ. Effects of probiotic supplementation on bone health in postmenopausal women: a systematic review and meta-analysis. Front Endocrinol (Lausanne). (2024) 15:1487998. doi:  10.3389/fendo.2024.1487998 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 88. Guo JM, Huang ZJ, Wang QJ, Wang M, Ming Y, Chen W, et al. Opportunities and challenges of bacterial extracellular vesicles in regenerative medicine. J Nanobiotechnology. (2025) 23:4. doi:  10.1186/s12951-024-02935-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 89. Mehvish HB, Zhang S, Liang Y, Sun D, Qiu X, Qu K, et al. Enhanced osteoporosis treatment via nano drug coating encapsulating Lactobacillus rhamnosus GG. ACS Appl Mater Interfaces. (2025) 17:5326–39. doi:  10.1021/acsami.4c17985 [DOI] [PubMed] [Google Scholar]
  • 90. Li QB, Ou ZH, Lin JD, Tang D, He B, Wu Y, et al. Specific labeling of outer membrane vesicles with antibiotic-conjugated probe reveals early bacterial infections in blood. Nat Commun. (2025) 16:3535. doi:  10.1038/s41467-025-58676-8 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Table1.docx (62KB, docx)

Articles from Frontiers in Endocrinology are provided here courtesy of Frontiers Media SA

RESOURCES