Introduction
Rheumatoid arthritis (RA) is a chronic systemic autoimmune disease characterized by symmetric inflammatory polyarthritis, synovial hyperplasia, cartilage loss, focal bone erosion, and variable extra-articular involvement. Global analyses indicate that RA remains a growing cause of disability, with prevalence increasing as populations age and survive longer with chronic disease.1 The biological problem is not simply excess inflammation. RA develops through a sequence that includes loss of tolerance to post-translationally modified proteins, autoantibody maturation, mucosal and systemic immune activation, recruitment of leukocytes into the synovium, and establishment of a self-sustaining immune-stromal niche.2–5 Once this niche is established, synovial macrophages, T and B lymphocytes, neutrophils, endothelial cells, osteoclast-lineage cells, and fibroblast-like synoviocytes (FLSs) exchange cytokines, chemokines, growth factors, metabolites, extracellular vesicles, and matrix-derived danger signals.
The result is a tissue-level circuit that can persist even when one inflammatory mediator is therapeutically blocked.6–8
Disease-modifying antirheumatic drugs (DMARDs) remain the standard of care, and contemporary RA management is based on treat-to-target principles with methotrexate as a common anchor drug and escalation to biologic or targeted synthetic DMARDs when required.9,10 These therapies have substantially improved remission and structural outcomes, but primary non-response, secondary loss of response, intolerance, infection risk, organ toxicity, reproductive considerations, cardiovascular risk signals for selected agents, and disparities in access continue to motivate adjunctive pharmacological research. Any TCM-derived intervention must therefore be evaluated against modern treat-to-target care, including methotrexate (MTX), biologic DMARDs (bDMARDs), and targeted synthetic DMARDs (tsDMARDs), rather than against untreated disease.
Traditional Chinese medicine (TCM) has been used for inflammatory arthritides through botanical extracts, multi-herb formulations, and purified or semi-purified constituents. Modern pharmacology has identified structurally diverse molecules with anti-inflammatory or immunoregulatory actions, including triptolide and celastrol from Tripterygium wilfordii, sinomenine from Sinomenium acutum, berberine from Coptis species and related botanicals, paeoniflorin from Paeonia lactiflora, tanshinone IIA from Salvia miltiorrhiza, and icariin from Epimedium species. The term “TCM-derived” describes the historical or pharmacognostic source, not chemical exclusivity: curcumin, quercetin, resveratrol, and several other compounds occur broadly in plants and foods.
This distinction matters because traditional use, botanical extraction, isolated-compound pharmacology, and clinical efficacy are different levels of evidence.11,12
Recent reviews have comprehensively catalogued plant-derived compounds, extracts, inflammatory pathways, and pharmacokinetic limitations in RA and arthritis.13,14 The present review asks a narrower and more translational question. Rather than treating pathway breadth as mechanistic proof, we appraise whether a chemically defined intervention reaches a relevant compartment at a plausible exposure, whether a proximal target or causal pathway has been demonstrated, whether the model represents established disease, whether safety permits chronic use, and whether the clinical intervention is actually the same product as the preclinical molecule. The review therefore integrates five evidence dimensions—chemical identity, model relevance, exposure, proximal mechanism, and replication—and explicitly separates standardized botanical products from isolated constituents and metabolites.
Table 1 summarizes this evidence hierarchy, whereas Table 2 focuses on exposure, formulation, and human translation. Figure 1 provides an evidence-aware mechanistic overview. This exposure-aware, causal-mechanistic framework is the principal distinction from recent compound-by-pathway inventories.
Table 1Evidence Hierarchy for 15 Representative TCM-Derived Constituents in Rheumatoid ArthritisConstituentChemical Class/Intervention TypeRepresentative RA SystemsRoute/Exposure ContextProximal Target EvidenceKey Limitation/Toxicity FlagHighest Evidence TierTriptolideDiterpene triepoxide; isolated moleculeRA-FLSs; CIA/AIA; chronic arthritisCell/animal studies; systemic exposure constrained by narrow therapeutic windowMainly downstream/associative; direct causal target engagement in RA not establishedReproductive, hepatic, renal, hematologic and immune toxicity; TwHF product efficacy is not attributable to triptolide aloneAnimal therapeutic/human evidence applies to TwHF product, not isolated triptolideCelastrolQuinone-methide triterpenoid; isolated moleculeMacrophage-like systems; AIA/CFA; FLS/osteoclast studiesPredominantly preclinical; formulation-dependent exposureElectrophilic covalent reactivity supports plausible proximal chemistry, but RA-selective target engagement remains incompletePromiscuous thiol reactivity and narrow safety windowAnimal therapeuticSinomenineMorphinan alkaloid; molecule distinct from marketed hydrochloride productRA-FLSs; AIA; neutrophil studies; human biomarker/clinical reportsOral/clinical product use; exposure–response incompletely definedTLR4–MyD88–NF-κB and NET findings mainly associativeProduct–molecule non-equivalence; target identity not establishedHuman clinical/biomarkerBerberineIsoquinoline alkaloid; isolated moleculeCIA; gut microbiota models; T-cell assaysOral parent drug: low systemic but high intestinal exposureGut cortistatin/microbiota model is exposure-coherent; direct joint targets not establishedPoor systemic bioavailability; metabolite/microbiota dependenceAnimal therapeuticPaeoniflorinMonoterpene glycoside; isolated molecule distinct from total glucosides of peonyAIA/CIA; RA-FLSsPredominantly oral preclinical; extensive metabolismTLR4/circRNA claims mainly associativeLimited human PK/PD and target engagement; product–constituent distinctionAnimal therapeuticCurcuminDiarylheptanoid polyphenol; multiple non-equivalent formulationsRA-FLSs; CIA/AIA; small RCTsOral exposure highly formulation dependent; free parent usually lowDownstream NF-κB/Nrf2 and related readouts mostly associativeFormulation heterogeneity; low/variable bioavailabilitySmall RCTsArtesunateSemi-synthetic artemisinin derivativeRA-FLSs; CIA; osteoclast/bone modelsPreclinical systemic dosing; anti-RA clinical exposure undefinedNF-κB/PI3K–AKT/p62–Nrf2 findings mainly associativeLong-term RA dosing and safety not establishedAnimal therapeuticBaicaleinFlavone; isolated moleculeRA-FLSs; experimental arthritisPreclinical cell/animal exposurePI3K–AKT–mTOR changes mainly associativeActive conjugates and achievable joint exposure uncertainAnimal/cellQuercetinFlavonol; broadly distributed phytochemicalRA-FLSs; Th17; osteoclast assays; CIAVariable oral exposure; extensive metabolismIntegrated Th17–FLS–RANKL/osteoclast evidence, but proximal target unresolvedNon-TCM specificity; low/variable parent exposureAnimal therapeuticIcariinPrenylated flavonol glycosideCIA/AIA; macrophage/synoviocyte assaysPreclinical oral/systemic exposure; active metabolites likely relevantmiR-223-3p/NLRP3 and ferroptosis readouts mostly associativeFerroptosis claims require rescue; human target engagement absentAnimal therapeuticTanshinone IIAAbietane diterpenoid quinoneRA-FLSs; osteoclast cultures; CIAPreclinical; formulation-dependent exposureLDHC–ROS and apoptosis studies suggest proximal hypotheses but require replicationLimited human RA evidenceAnimal therapeuticGinsenosidesTriterpenoid saponins/metabolites; constituent classAIA/CIA; immune-cell assaysParent saponins undergo extensive metabolism, including compound K formationPPARγ/NF-κB/STAT3/Th17-Treg signals mainly associativeDifferent ginsenosides/metabolites are not interchangeableAnimal therapeuticAndrographolideLabdane diterpenoid lactoneRA-FLSs; CIAPreclinical systemic exposureElectrophilic chemistry plausible; HIF-1α/NF-κB effects mainly downstreamTarget selectivity and chronic exposure unclearAnimal therapeuticEmodinAnthraquinoneNeutrophil/macrophage–FLS systems; arthritis modelsPreclinical; systemic exposure and metabolites variableNET formation/apoptosis findings need cytotoxicity controlsGastrointestinal/hepatic safety; PK uncertaintyAnimal/cellNorisoboldineAporphine alkaloidOsteoclast cultures; CIAPreclinicalRANKL-driven osteoclast inhibition; proximal target incompleteLimited human dataAnimal therapeuticNotes: Evidence tiers identify the highest level of evidence for the specific intervention named in the first column. “Associative” indicates that downstream pathway changes were reported without sufficient target-engagement, necessity, or rescue evidence. Standardized products, purified constituents, metabolites, and formulation-enhanced preparations are treated as distinct interventions.Abbreviations: RA-FLSs, rheumatoid arthritis fibroblast-like synoviocytes; CIA, collagen-induced arthritis; AIA, adjuvant-induced arthritis; PK/PD, pharmacokinetics/pharmacodynamics; RCT, randomized controlled trial.Table 2Translational PK/PD and Clinical-Product Considerations for the Best-Studied InterventionsInterventionClinical/Product IdentityRepresentative Human Regimen/EvidenceExposure/PK IssueStructural Outcome EvidenceKey Safety/Interpretive IssueStandardized TwHF productBotanical extract/glycoside product; not isolated triptolide/celastrolTRIFRA: 207 patients; TwHF 20 mg three times daily, MTX 12.5 mg weekly, or combination for 24 weeks;15 2023 consensus synthesized 19 RCTs/1795 participants16Product composition and batch standardization determine exposure; constituent attribution is invalid without compositional PKClinical disease-activity benefit shown; robust long-term radiographic/structural evidence remains limitedReproductive toxicity, hepatic/renal risk, hematologic and GI adverse effects; caution in pregnancy/gestation planningSinomenine/sinomenine hydrochloridePurified alkaloid versus marketed hydrochloride productClinical and meta-analytic signals exist,17,18 but trial quality, formulation and background therapy varyHuman exposure–response and proximal target engagement remain incompletely definedStructural protection not established to modern DMARD standardsProduct–molecule non-equivalence; combination safety requires prospective studyCurcumin formulationsConventional curcumin, nanomicelles, and bioavailability-enhanced matrices are pharmacokinetically distinctExamples include 36-patient placebo-controlled trial of 250 or 500 mg twice daily for 90 days19 and other small formulation-specific trials20–23Free parent exposure is low and strongly formulation dependent; conjugates/metabolites may dominate24–27No convincing evidence of prevention of radiographic progressionSmall samples, heterogeneous formulations/background DMARDs; do not pool conceptually as one “curcumin” interventionBerberineIsolated alkaloid; oral parent drugHuman RA efficacy evidence limited relative to preclinical literatureExtensive intestinal first-pass elimination; low plasma parent concentration with high gut exposure28No established human structural benefitDirect synovial mechanism may be implausible after oral dosing; gut-restricted mechanism is more exposure-consistentPaeoniflorin/total glucosides of peonyIsolated glycoside versus standardized multicomponent productClinical product evidence should not be assigned automatically to paeoniflorinExtensive metabolism; limited human synovial exposure dataNot established for isolated paeoniflorinProduct–constituent confusion and weak target engagementTriptolide/celastrolPurified TwHF constituentsNo evidence equivalent to standardized TwHF RCTsExposure constrained by toxicity; formulation and tissue distribution remain importantPreclinical structural/inflammatory signals onlyNarrow therapeutic window; cannot inherit clinical efficacy claims from TwHF extractNotes: Human regimens are included only when directly supported by cited clinical sources.
Absence of a value indicates that clinically interpretable exposure or structural evidence is not established; no unverified concentration has been imputed.Figure 1Integrated immunological mechanisms of TCM-derived bioactive constituents in rheumatoid arthritis. Key signaling nodes to which representative compounds converge are inflammatory, stress-response, osteoclastogenic, and angiogenic. The cellular effects include macrophages, adaptive lymphocytes, fibroblast-like synoviocytes, osteoclasts, neutrophils, dendritic cells and endothelial cells.
Arrows show the proposed directionality of the relationship, NOT the strength of the causal evidence. The majority of links are based mainly on isolated-cell assays or on models of arthritis in animals; human synovial validation and randomized clinical evidence are considerably less. The arrows indicate the sequential progression from representative TCM-derived bioactive constituents to convergent molecular targets, subsequent immune–stromal remodeling, and ultimately disease-level effects.
The symbols “↓” and “↑” indicate decreased/inhibited and increased/enhanced effects, respectively.A four-column infographic linking TCM constituents to molecular nodes, immune remodeling and disease effects.Four-column infographic summarizing links from representative constituents to molecular nodes, immune stromal remodeling and disease level effects in rheumatoid arthritis. 1 Representative constituents. Terpenoids slash diterpenoids: triptolide, celastrol, tanshinone IIA, artesunate.
Alkaloids: sinomenine, berberine, norisoboldine. Polyphenols slash flavonoids: curcumin, quercetin, baicalein. Glycosides slash saponins: paeoniflorin, icariin, ginsenosides.
2 Convergent molecular nodes. NF-kappaB slash MAPK. JAK slash STAT.
PI3K-AKT-mTOR. ROS-Nrf2. NLRP3-caspase-1.
RANKL-NFATc1. HIF-1alpha-VEGF. gut microbiota slash metabolites. 3 Immune-stromal remodeling.
Macrophages slash monocytes: down arrow inflammatory states; up arrow pro-resolving programs; down arrow TNF, IL-1beta, IL-6 and GM-CSF. T- and B-cell responses: down arrow Th17 slash Tfh and IL-17; up arrow Treg; limited direct evidence for autoantibody suppression. Fibroblast-like synoviocytes: down arrow proliferation, migration, invasion, MMPs and RANKL; up arrow apoptosis or return toward quiescence.
Osteoclast lineage: down arrow RANKL signaling, differentiation and resorption. Neutrophils, DCs and endothelium: down arrow NETs, costimulation and angiogenesis left parenthesis emerging or compound-specific evidence right parenthesis. Stress and cell-death programs: down arrow ROS, pyroptosis and ferroptosis; restored autophagy and mitochondrial quality.
4 Disease-level effects. Reduced synovitis and pannus. Preserved cartilage and matrix.
Reduced focal bone erosion. Improved pain, mobility and physical function. Arrows connect the four columns from left to right.
Evidence hierarchy: isolated-cell assays less than animal arthritis models less than human tissue validation less than randomized clinical evidence. Most mechanisms remain preclinical and require exposure-target engagement confirmation. TCM, traditional Chinese medicine; FLS, fibroblast-like synoviocyte; NET, neutrophil extracellular trap; DC, dendritic cell; MMP, matrix metalloproteinase.Abbreviations: DC, dendritic cell; FLS, fibroblast-like synoviocyte; MMP, matrix metalloproteinase; NET, neutrophil extracellular trap; TCM, traditional Chinese medicine.
Literature-Search Framework and Evidence Appraisal
This article is a critical narrative review, not a systematic review or meta-analysis. We searched PubMed and Web of Science for English-language publications from 1 January 2010 through 31 July 2026 and also retained earlier landmark studies when they were required to interpret clinical products, pharmacokinetics, or canonical mechanisms. The search combined RA terms with TCM/natural-product terms and compound-specific terms.
A representative Boolean strategy was: (“rheumatoid arthritis” OR synovitis OR “fibroblast-like synoviocyte” OR osteoclast OR macrophage OR Th17 OR Treg OR “neutrophil extracellular trap”) AND (“traditional Chinese medicine” OR phytochemical OR “natural product” OR “bioactive constituent” OR the individual compound name) AND (mechanism OR pharmacokinetic* OR exposure OR safety OR clinical OR trial). Searches for translational evidence additionally combined each compound/product with (“target engagement” OR CETSA OR chemoproteomic* OR rescue OR CRISPR OR plasma OR synovial OR metabolite OR randomized). Reference lists of key reviews and included clinical studies were also examined.
Records were screened by title/abstract and then by full text for relevance to RA pharmacology, exposure, safety, or clinical translation. Because the search was designed prospectively as a narrative review rather than a PRISMA workflow, an exact numerical screening yield was not prospectively recorded; we therefore do not report a reconstructed count that could imply false precision. This limitation is now stated explicitly.
The 15 core entities were selected using explicit inclusion logic. Eligible entities had to be chemically defined constituents or clearly defined constituent classes with recurrent RA-relevant evidence and sufficient information to support appraisal across at least two of the following domains: validated arthritis models or primary human RA systems; pharmacokinetic/exposure information; safety or therapeutic-window data; proximal mechanistic evidence; or human clinical/biomarker evidence. We also sought representation across major chemical classes and RA-relevant compartments rather than selecting compounds solely because they were frequently mentioned.
Multi-component proprietary products and standardized botanical preparations were excluded from the 15-molecule core set and are discussed separately because product efficacy cannot be assigned to one marker constituent. Thus, Tripterygium glycosides/TwHF extracts, sinomenine hydrochloride products, and total glucosides of peony are treated as clinical products rather than as interchangeable evidence for triptolide/celastrol, sinomenine base, or paeoniflorin. Resveratrol and other broadly studied nutraceuticals were not included in the 15-entity core because the purpose was not to provide an exhaustive phytochemical catalogue.
Evidence was graded qualitatively according to chemical characterization, model relevance, exposure/route, proximal target evidence, and replication. Network pharmacology, docking, bulk enrichment, isolated terminal cytokine changes, LC3-II alone, marker-only ferroptosis readouts, or M1/M2 marker shifts were not treated as causal mechanistic proof.
Immunopathological Framework of Rheumatoid Arthritis
Loss of Tolerance, Autoantibodies, and the Preclinical Phase
RA autoimmunity can precede clinical synovitis by years. In genetically susceptible individuals, mucosal and systemic exposures may promote post-translationally modified antigens and maturation of anti-citrullinated protein antibodies (ACPAs) and rheumatoid factor.3–5 For pharmacological interpretation, the important point is that most TCM-derived compound studies begin after induction of experimental arthritis and therefore test suppression of established inflammation rather than prevention of tolerance loss. Changes in total immunoglobulin, rheumatoid factor, or anti-collagen antibodies in collagen-induced arthritis (CIA) should not be equated with correction of human ACPA fine specificity, germinal-center biology, or synovial B-cell pathotypes.29–34
Direct evidence for effects on autoreactive B-cell clones, plasmablast survival, ACPA glycosylation, germinal centers, or antigen presentation by human RA B cells remains limited. This gap should remain explicit rather than being replaced by broad claims of “immune balance”.
The Synovial Immune-Stromal Ecosystem
The pharmacologically relevant synovial compartments are inflammatory and resolving macrophage states, pathogenic RA-FLSs, T- and B-cell programs, neutrophils, endothelium, osteoclast-lineage cells, and their metabolic interactions. Human single-cell and spatial studies show that these states are heterogeneous and tissue-imprinted, making binary M1/M2 labels and acute cytokine-stimulated cultures inadequate surrogates for the full disease state.35–41
RA-FLSs are particularly important because lining and sublining populations contribute to cartilage invasion, matrix degradation, leukocyte recruitment, cytokine production, and RANKL-dependent osteoclastogenesis. A pharmacological study should therefore distinguish non-specific cytotoxicity from durable reprogramming of pathogenic fibroblast states and, where possible, use patient-derived cells, organoids, or multicellular synovial systems.42–44
Adaptive immune cells remain relevant because Th17-associated programs, T peripheral helper/T follicular helper cells, B cells, and plasma cells can reinforce myeloid and stromal activation. However, splenic Th17/Treg percentages or antibody titres in CIA do not by themselves demonstrate restoration of human immune tolerance. The pharmacologically relevant question is whether an intervention changes a disease-relevant cellular state at achievable exposure.40,45
Bone Erosion, Angiogenesis, and Immunometabolic Stress
Structural protection requires more than suppression of paw swelling. RANKL-driven osteoclastogenesis, RA-FLS-mediated cartilage invasion, and impaired bone repair are the key structural compartments. Studies claiming disease modification should therefore include direct bone/cartilage outcomes such as micro-computed tomography, histomorphometry, erosion scoring, or validated imaging, rather than relying only on inflammatory biomarkers.32,33,46
Hypoxia, glycolytic reprogramming, mitochondrial stress, ROS, inflammasome activation, angiogenesis, and autophagy intersect with these cellular compartments. They are pharmacologically relevant because HIF-1α, Nrf2, NLRP3, and PI3K–AKT–mTOR recur across compound studies, but changes in these nodes may be downstream consequences rather than direct targets. Figure 2 summarizes the RA compartments used in this review as the biological frame for evaluating compound evidence.47–49Figure 2Immunopathological framework of rheumatoid arthritis.
Rheumatoid arthritis (RA) has a series of steps that are working in concert: Loss of immune tolerance, chronic inflammation of the synovium and gradual destruction of the joints. (1) Preclinical phase: Genetic susceptibility and mucosal/systemic exposures, leading to post-translational protein modifications, autoreactive T- and B-cell activation, autoantibody production, immune-complex formation and early activation of osteoclasts. (2) Synovial immune–stromal ecosystem: pathogenic fibroblast-like synoviocytes and inflammatory macrophages create bi-directional signaling loops to maintain synovial heterogeneity, matrix degradation, recruitment of leukocytes, and production of cytokines.
Other cells that influence lymphoid, myeloid, and fibroid pathotypes are T cells, B cells, plasma cells, dendritic cells, mast cells, and endothelial cells. (3) Structural damage: RANKL-dependent osteoclastogenesis, synovial cartilage invasion, defective bone formation, angiogenesis, hypoxia, metabolic reprograming, oxidative stress, inflammasome activation, and impaired autophagy all contribute to persist synovitis and irreversible structural damage. Arrows show a proposed biological relationship and not equal degree of causal evidence.A three-section infographic outlining preclinical RA, chronic synovitis and joint damage pathways.Three-section infographic titled, Immunopathological Framework of Rheumatoid Arthritis.
A) Loss of tolerance and the preclinical phase. Heading, Environmental slash mucosal risk factors. Items: Periodontal inflammation, Lung inflammation, Obesity, Smoking.
Heading, Post-translational modifications of self-proteins. Items: Citrullination with label, Cit; Carbamylation with label, Carb; Acetylation with label, Ac. Heading, Antigen presentation by APCs.
Labels: Self-antigen left parenthesis peptide right parenthesis; Dendritic cell left parenthesis macrophage right parenthesis; MHC II; T cell. Next labels: Expansion of autoreactive T cells; Autoreactive B cells. Heading, Autoantibody production.
Items: ACPAs left parenthesis anti-citrullinated protein antibodies right parenthesis; Rheumatoid factor left parenthesis anti-Fc IgM slash IgG right parenthesis. Heading, Immune-complex formation. Bottom label, Preclinical RA.
Sentence: Autoimmunity without clinical synovitis left parenthesis may precede arthritis by years right parenthesis. B) The synovial immune-stromal ecosystem. Label, Synovial lining layer.
Labels: Resident FLSs left parenthesis homeostatic right parenthesis; Activated slash invasive FLSs left parenthesis pathogenic phenotype right parenthesis; Inflammatory macrophages slash monocytes. Central label: Chronic inflammation left parenthesis bidirectional crosstalk right parenthesis. Heading, Supporting immune and stromal cells.
Items: Th17 T cells with IL-17 and IL-22; Tregs cells with IL-10 and TGF-beta; B cells slash plasma cells; Dendritic cells; Mast cells; Endothelial cells with Angiogenic factors. Heading, Key effector programs. Boxes: Pro-inflammatory cytokines listing TNF, IL-6, IL-1; Autoantibodies slash immune complexes; Tissue remodeling listing MMPs, RANKL.
Bottom label, Chronic synovitis. Sentence: Sustained immune-stromal interactions drive persistent inflammation and joint tissue damage. C) Bone erosion, angiogenesis and immunometabolic stress.
Joint label: Synovial pannus left parenthesis inflamed synovium right parenthesis. Three labeled pathways: Osteoclast activation and bone erosion with RANKL and labels, Osteoclast, Bone erosion; Angiogenesis with VEGF, FGF and label, New blood vessel formation; Immunometabolic stress listing Hypoxia left parenthesis HIF-1alpha right parenthesis, Mitochondrial dysfunction, Glycolytic switch. Bottom bar label, Sustained synovitis and tissue damage.
Sentence: Bone destruction, cartilage loss and progressive joint dysfunction.Abbreviations: ACPA, anti-citrullinated protein antibody; FLS, fibroblast-like synoviocyte; MMP, matrix metalloproteinase; NLRP3, NOD-, LRR- and pyrin domain-containing protein 3; RA, rheumatoid arthritis; RANKL, receptor activator of nuclear factor-κB ligand; RF, rheumatoid factor; ROS, reactive oxygen species; VEGF, vascular endothelial growth factor.
Mechanisms of Representative TCM-Derived Bioactive Constituents
Regulation of Monocytes and Synovial Macrophage Programs
Sinomenine has been associated with reduced TNF, IL-1β, and IL-6 production, altered circulating monocyte/macrophage subsets, inhibition of TLR4–MyD88–NF-κB signaling in RA-FLSs, and reduced NET formation in adjuvant-induced arthritis.17,50,51 These observations support activity across innate-immune compartments, but they do not identify a single proximal molecular target and should therefore be classified mainly as associative pathway evidence. Human target engagement and exposure–response relationships remain insufficiently defined.
Celastrol, a quinone-methide triterpenoid from T. wilfordii, has been reported to reduce ROS, NF-κB activation, NLRP3 inflammasome assembly, and IL-1β/IL-18 release in macrophage-like systems and experimental arthritis.52 The sequence is biologically coherent, but celastrol is electrophilic and can modify multiple protein thiols. Consequently, downstream NLRP3/NF-κB changes should not be interpreted as evidence of selective target engagement without chemoproteomic target identification, occupancy measurements, and genetic or pharmacological rescue. This chemical promiscuity may contribute to both potency and toxicity.
Oral berberine is associated with reduced Th17 responses, altered gut microbial/metabolite programs, and increased intestinal cortistatin in CIA.53 Because systemic parent-drug exposure is low whereas intestinal exposure is high, a gut-restricted or microbiota-mediated mechanism is more pharmacologically plausible than direct joint action after oral dosing. High-concentration cultured-cell studies may still be useful for hypothesis generation, but they should not be treated as exposure-matched proof of the in vivo mechanism.54
Curcumin, paeoniflorin, baicalein, tanshinone IIA, andrographolide, and ginsenosides have also been reported to reduce inflammatory cytokines or macrophage-associated histological signals in experimental arthritis. Many studies infer “M1-to-M2 polarization” from CD86/CD206, iNOS/arginase-1, or related markers. In human RA these marker sets do not define stable binary macrophage populations, so such findings are best described as marker shifts unless supported by single-cell, spatial, lineage, and functional validation.55,56
Modulation of T-Cell and B-Cell Responses
For TCM-derived compounds, a Th17/Treg rebalancing has been frequently reported. Berberine has been observed to decrease Th17 differentiation and Th17 related responses in CIA, and cortistatin derived from the gut has been suggested as a possible intermediate.53 Icariin has been shown to inhibit experimental arthritis via the miR-223-3p/NLRP3 signaling pathway57 while other preclinical studies have implicated icariin-derived compounds with general inflammatory and osteoclast pathways. Paeoniflorin and ginsenoside preparations have also been linked to modifications in Th17/Treg ratios and a decrease in IL-6/STAT3 signaling.
Biologically, these observations make sense, as TGF-β and IL-2 promote regulatory phenotypes, while IL-6, IL-1β, and IL-23 promote Th17 programs, and metabolic stress promotes Th17 programs. However, splenic percentage changes do not define tolerance restoration and many studies do not differentiate between the stable regulatory T cells and the transiently activated effector cells that express regulatory markers.
Triptolide is a general immunosuppressive agent that can suppress T-cell activation, cytokine transcription and inflammatory gene expression. Its high potency is mechanistically important, but is clinically problematic because transcriptional inhibition is not limited to the pathogenic lymphocytes. Straightforward translation is complicated by reproductive, hepatic, renal, hematologic and immune toxicity.
Evidence of a selective effect of celastrol or curcumin on pathogenic T peripheral helper cells, synovial tissue-resident T cells or autoreactive clones is also limited.
Evidence for B-cells is not as strong as the T-cell literature. A decrease in RF or anti-collagen antibody levels in animal models could be due to an overall dampening of inflammation, a change in antigen burden or a defect in plasma-cell responses. There are very few studies that address dynamics of germinal centers, affinity maturation, fine specificity of ACPA, Fc glycosylation, and antigen presentation by human RA B cells.
Therefore, the term correction of autoimmunity should not be used unless tested upstream B-cell biology. This is an important opportunity for future research as B-cell rich synovial pathotypes may behave differently from myeloid- or fibroblast-dominant disease.
Suppression or Reprogramming of Fibroblast-Like Synoviocytes
RA-FLSs display tumor-like features, including resistance to apoptosis, enhanced migration and invasion, abnormal metabolism, and production of MMPs, RANKL, cytokines, and chemokines. Triptolide inhibits FLS proliferation and inflammatory mediator production, and it can interfere with angiogenic cascades.58 Artesunate suppresses TNF-induced NF-κB and PI3K-AKT signaling in FLSs and reduces migration and invasion in experimental systems.59,60 Andrographolide inhibits FLS migration and invasion through HIF-1α-related mechanisms.61 Baicalein induces apoptosis and suppresses proliferation and epithelial-mesenchymal-transition-like programs through PI3K-AKT-mTOR inhibition.62 Tanshinone IIA also promotes apoptosis in RA-FLSs.63
These results suggest that the FLSs can be viewed as a convergent pharmacological compartment, yet three interpretive issues arise again. First, acute cytokine stimulated FLS cultures do not mimic the stable epigenetic and positionally identities of lining and sublining fibroblasts. Second, selective inhibition of invasion or of cytokine release can be confused with cytotoxicity.
Anti-proliferative, pro-apoptotic, and non-cytotoxic signaling effects need to be separated in dose response experiments, therefore. Third, changes in pathways, measured post cell death, do not necessarily constitute evidence of a proximal mechanism. To differentiate between true state reprogramming and a non-specific suppression, rescue experiments, target-depletion studies, real-time migration assays and human synovial organoids are required.
Paeoniflorin has been reported to suppress inflammatory responses in RA-FLSs through TLR4-related signaling and a circRNA–miRNA regulatory axis.64,65 Because ceRNA mechanisms are especially vulnerable to stoichiometric over-interpretation, these findings are treated here as hypothesis-generating rather than established causal mechanisms. The more clinically relevant test is whether paeoniflorin can alter patient-derived RA-FLS phenotypes at free parent/metabolite exposures achievable in plasma or synovial tissue.
Inhibition of Osteoclastogenesis and Structural Damage
Osteoclast inhibition is essential for any claim of structural protection. Quercetin reduces IL-17-stimulated RANKL production in RA-FLSs, suppresses Th17 differentiation, and inhibits osteoclast formation, providing a mechanistically integrated link between adaptive immunity, stroma, and bone.66 Norisoboldine inhibits osteoclast differentiation and ameliorates bone erosion in CIA.67 Tanshinone IIA suppresses osteoclastogenesis through an LDHC-ROS-related mechanism,68 while artesunate has been reported to preserve bone through p62-Nrf2 signaling.59 Icariin has additional activity on the miR-223-3p/NLRP3 axis and on ferroptosis-associated synoviocyte survival,57,69 but these findings should not be treated as direct evidence of erosion prevention.70
These studies are best when they include the use of tartrate-resistant acid phosphatase staining, resorption-pit assays, micro-computed tomography, histomorphometry, and molecular perturbation. Use of serum bone-turnover markers alone is not adequate. In addition, it is critical to distinguish direct effects on the osteoclast precursors from indirect effects that may occur as a result of decreased systemic inflammation or decreased FLS RANKL.
Human validation should include osteoclastogenesis by patient monocytes, co-culture with RA-FLSs or B cells, and unbound concentrations achievable in the clinic.
To achieve structural benefit, there might be a need to control both osteoclasts and FLS-mediated cartilage damage. While an NFATc1 blocking compound that is too toxic for chronic use is not likely to be a disease-modifying therapy, a modest anti-osteoclast effect could be useful if safe, orally active, unidirectional, and additive to the effects of methotrexate or a biologic. The current animal studies do not include this risk-benefit/combination context.
Neutrophils, Dendritic Cells, Endothelial Cells, and Angiogenesis
Neutrophils are abundant in RA synovial fluid and can release proteases, ROS, alarmins, and NETs. NETs provide citrullinated antigens and may amplify autoantibody-driven inflammation. Triptolide suppresses neutrophil recruitment, cytokine release, and inflammatory activity in chronic arthritis.71 Sinomenine and emodin have also been reported to reduce NET formation or promote resolution-associated neutrophil apoptosis.51,72 These findings are relevant, but NET assays require methodological care: extracellular DNA, citrullinated histone staining, and myeloperoxidase-DNA complexes can reflect different forms of cell death.
Demonstrating PAD4 dependence and excluding cytotoxic lysis would strengthen mechanistic claims.73,74
There is relatively little information on dendritic-cell research. While there are a number of constituents that in vitro decrease maturation markers or antigen-presenting capacity, few directly investigate presentation of RA-relevant autoantigens or tolerogenic function. This is not a trivial oversight, as dendritic cells link events occurring in the mucosa and adaptive autoimmunity.
Hypoxia, VEGF, inflammatory growth factors and chemokines are responsible for synovial angiogenesis. Triptolide inhibits an angiogenic cascade in experimental RA58 and the other agents, artesunate, curcumin and tanshinone IIA have been associated with reduced HIF-1α/VEGF signaling. Anti-angiogenic activity may be beneficial in preventing access by leukocytes and preventing the formation of the pannus, however chronic suppression of the vascular system may have adverse effects on repair.
Perfusion, endothelial phenotype and tissue oxygenation should therefore be measured in studies and the presence of angiogenesis inferred from the amount of VEGF expression.
Oxidative Stress, Mitochondrial Quality Control, Autophagy, Pyroptosis, and Ferroptosis
Pyroptosis and NLRP3–caspase-1 signaling provide a plausible link among ROS, myeloid activation, and IL-1-family cytokines. Celastrol and icariin studies report changes in this pathway,52,57 but causal attribution requires target engagement and perturbation. Ferroptosis evidence is weaker: marker changes in lipid peroxidation, GPX4, or system Xc− are not sufficient without rescue by canonical ferroptosis inhibitors, iron dependence, and exclusion of apoptosis or necroptosis.69,75,76 Accordingly, ferroptosis is treated as an emerging hypothesis rather than an established anti-RA mechanism.
Autophagy may have protective effects by eliminating damaged organelles, and may also promote the survival of pathogenic FLSs. Berberine has been related to the autophagy-related regulation of IL-21 signaling and Th17/Treg balance, while icariin and other flavonoids have been related to PI3K-AKT-mTOR-ULK1 pathways. LC3-II is not enough on its own to interpret – accumulation of LC3-II can be due to the increase of autophagosomes, or autophagosomes may not be degraded.
A combination of tandem reporters, lysosomal inhibitors, and genetic perturbation are essential.77
Pyroptosis and NLRP3-caspase-1 signaling provide a plausible link among ROS, macrophage activation, and IL-1 family cytokines. Celastrol has direct preclinical evidence in this domain,52 while icariin and other compounds have been reported to reduce NLRP3 activation.57 Ferroptosis has more recently been implicated in FLS and chondrocyte injury, and icariin has been reported to modulate ferroptosis-related markers.69 Yet much of the ferroptosis literature is marker-based. A rigorous claim requires rescue by canonical inhibitors, dependence on GPX4 or system Xc−, demonstration of iron-dependent lipid peroxidation, and exclusion of apoptosis or necroptosis.75,76
Gut-Joint Axis and Systemic Immunometabolism
The gut microbiota can impact on RA via barrier integrity, microbial metabolites, molecular mimicry and modulation of T-cell and myeloid programs. One of the most convincing of these compounds studied was berberine, which was shown to have low systemic exposure, high exposure in the gut, and effects on gut cortistatin and the gut microbial community in a coherent pharmacological model.53 Other constituents, such as curcumin, ginsenosides and polysaccharide derived products might also have the ability to alter the microbial communities or the metabolism of bile acids and short-chain-fatty-acids.78,79
Animal facility, diet, antibiotics, sequencing pipeline, DMARD exposure and geography are significant sources of microbiome findings confounding. Changing the relative abundance of the microbiota is not sufficient to prove causality-further fecal-transfer and/or gnotobiotic experiments, combined with metabolite rescue and host loss-of-function experiments are needed to demonstrate that microbiota changes underlie efficacy.
Convergent Signaling Networks: Useful Map, Insufficient Proof
Seven recurrent signaling modules appear across the literature: NF-κB/MAPK, JAK/STAT, PI3K–AKT–mTOR, ROS–Nrf2, NLRP3–caspase-1, RANKL–NFATc1, and HIF-1α–VEGF. These modules are useful for organizing biological readouts, but they do not form a linear causal chain and should not be counted as separate direct targets simply because multiple downstream proteins change after treatment. Figure 1 summarizes the principal compartments and signaling nodes while explicitly separating proposed directionality from causal certainty.
Table 1 labels evidence as associative unless proximal target engagement, necessity, or rescue has been demonstrated.
Upstream shift of mechanistic studies. Chemoproteomics can be used for electrophilic compounds, like celastrol and andrographolide; Thermal proteome profiling and cellular target engagement can be used to identify targets that are non-covalent; CRISPR perturbation can be used to test necessity; and rescue with constitutively active proteins and with drug resistant proteins can be used to test pathway dependence. Target modulation should then be investigated using pharmacokinetic-pharmacodynamic modelling to determine whether or not clinically plausible exposure allows for target modulation.
In the absence of this sequence, “multi-target” may lose its meaning and become a term to describe poor definition of target rather than a true therapeutic benefit.80
Clinical Evidence, Product Identity, Pharmacokinetics, Safety, and Delivery
Clinical Products are Not Interchangeable with Isolated Molecules
The strongest human evidence in this field concerns standardized T. wilfordii Hook F (TwHF) preparations rather than purified triptolide or celastrol. In the multicentre TRIFRA trial, 207 patients with active RA were randomized to MTX 12.5 mg weekly, TwHF 20 mg three times daily, or the combination for 24 weeks; ACR50 responses were 46.4%, 55.1%, and 76.8%, respectively.15 These data establish clinical activity of the standardized botanical intervention but cannot be attributed to a single constituent. The 2023 International Consensus Guidance, based on 19 randomized trials involving 1795 participants, similarly evaluates TwHF as a product and emphasizes reproductive, hepatic, and renal safety considerations.16 Therefore, standardized TwHF extract/glycosides, purified triptolide, and purified celastrol are treated as distinct interventions throughout this review.81,82
The same product–constituent rule applies to sinomenine and paeoniflorin. Marketed sinomenine hydrochloride preparations and total glucosides of peony are formulated products with clinical use in China; their efficacy or safety should not be assigned automatically to sinomenine base or paeoniflorin alone. Curcumin provides a second major example of formulation dependence: clinical studies have used chemically and pharmacokinetically different preparations, including conventional curcumin, nanomicelles, and bioavailability-enhanced matrices.
A 36-patient double-blind trial tested 250 or 500 mg twice daily for 90 days, whereas other studies used different formulations and doses.19–23,83 These trials are signals of formulation-specific adjunctive activity, not evidence for a single interchangeable entity called “curcumin”. Table 2 summarizes these translational distinctions. Small clinical studies have also examined resveratrol and quercetin in RA, but these findings remain intervention-specific and do not establish efficacy for other isolated constituents.84,85
Human studies remain much thinner than the preclinical literature and should be interpreted using conventional clinical-trial criteria: randomization and blinding, sample size, comparator, background DMARD therapy, formulation identity, duration, validated disease-activity outcomes, structural/imaging endpoints, and adverse events. Small adjunctive studies can justify larger trials but are not comparable in evidential weight to the TwHF extract trials and do not support replacement of MTX, bDMARDs, or tsDMARDs. Structural protection and combination safety remain unproven for nearly all isolated compounds.
Exposure, Metabolism, Formulation, and PK/PD
Exposure–response mismatch is a central translational problem. Curcumin shows low and highly formulation-dependent systemic exposure, with conjugated metabolites often dominating over free parent compound.24–27 Berberine undergoes extensive intestinal first-pass elimination and has low plasma parent-drug levels despite high gastrointestinal exposure.28 Ginsenosides can be converted by the microbiota to metabolites such as compound K, while paeoniflorin and icariin also undergo extensive biotransformation. Consequently, a 20–100 μM parent-compound cell experiment may be mechanistically informative but clinically irrelevant if free parent drug at that concentration is not achievable at the target tissue.
Table 2 therefore separates nominal administered dose from the more important question of systemic or tissue exposure.
Translational pharmacology should measure the free parent compound and major metabolites in plasma, synovial fluid, and synovial tissue; establish temporal correlation between the exposure and modulation of the target(s); and assess free active molecular species in human cells. Nanoparticles, liposomes, phospholipid complexes, hydrogels, and joint targeted delivery are ways to increase exposure, but these raise a new medicinal product, for which efficacy and toxicity are not known from the parent herb. Delivery systems should be compared to exposure matched free compound and tested for carrier related immunogenicity or accumulation.
Standardization of botanicals is also crucial. Efficacy and Toxicity are subject to change depending on species authentication, plant part used for cultivation, extraction solvent, marker compounds, impurity profiles and batch-to-batch variability. Removal or control of toxic constituents is of prime importance for T. wilfordii products.
Only reporting mass of “extract” will not allow for reproducibility or mechanistic attributions.
Safety and Drug–Drug Interactions
Natural origin does not imply a favorable therapeutic index. TwHF products, triptolide, and celastrol raise concerns involving reproductive, hepatic, renal, hematologic, gastrointestinal, and immune toxicity; reproductive risk is specifically emphasized in the international TwHF guidance.16,86 Emodin and related anthraquinones require gastrointestinal and hepatic caution, while multiple alkaloids, flavonoids, and botanical constituents can affect metabolic enzymes or transporters. For chronic RA treatment, the clinically relevant question is not whether an agent suppresses inflammation at a tolerated experimental dose, but whether efficacy is separated from toxicity across repeated exposure and in combination with standard DMARDs.87,88
Combination studies should consider the pharmacokinetic and pharmacodynamic interactions with methotrexate, glucocorticoids, biologics and JAK inhibitors. Prospective monitoring of hepatic enzymes, blood counts, renal function, reproductive toxicity, QT effects (where applicable) and infection endpoints should be considered. Clinical attractiveness does not only depend on the ability of a compound to reduce the dose of a classical drug in an animal model, but also on the ability to enhance the benefit-risk ratio for the combination.
Methodological Limitations and Recurrent Failure Modes
Firstly, many studies involve giving a preventative dose before or at the same time as the arthritis is induced. This design is an interference design aimed at testing interference with model initiation NOT treatment of established RA. Objective arthritis should be reached before beginning therapeutic dosing and methotrexate or another suitable positive control should be used as a benchmark.
Secondly, animal models contain only part of the picture of RA. CIA is incredibly useful for Adaptive Immunity and erosive disease but requires collagen immunization, while AIA is extremely innate and adjuvant driven and serum transfer models do not require adaptive priming. Model replication can identify if a compound is working upstream on the tolerance or downstream on the effecter inflammation or is working generally through toxicity.
Thirdly, doses are sometimes not related to pharmacokinetics. There is broad suppression and suppression in pathways is sometimes under-controlled in high concentrations. Only administered dose should be reported in studies, purity, solubility, free concentration, stability, cytotoxicity, and human equivalent exposure should be reported.
Fourthly, pathway inference is often circular. Phosphorylated p65 decreases and NF-κB is identified as the mechanism in a compound that yields reduction of arthritis. This does not rule out the possibility that decreased tissue damage secondarily results in decreased NF-κB.
Experiments of necessity and sufficiency are necessary. This is equally true of JAK-STAT, PI3K-AKT-mTOR, Nrf2, NLRP3 and ferroptosis claims.
Fifth, publication and selection bias could be causing an overestimate of consistency. Less noticeable are the negative compounds, failed programs of replication, and toxicity restricted programs. Most studies are single center, involve a single sex and are inadequate in pre-registering endpoints and reporting on randomization and blinding.
Such limitations should be taken into consideration before attempting to quantitatively rank compounds or pathways.
Future Directions
Substantive progress requires a shift from broad anti-inflammatory description to causal, exposure-aware immunology. The following priorities would materially strengthen the field: Use primary human RA synovial tissue, matched blood, synovial organoids, explants, and multicellular co-cultures to test whether compounds reprogram disease-relevant macrophage and fibroblast states rather than merely suppress acute cytokine responses.Integrate single-cell RNA sequencing, spatial transcriptomics, imaging mass cytometry, and proteomics with orthogonal functional assays. Molecular signatures should be linked to cell abundance, localization, migration, efferocytosis, matrix invasion, and osteoclast resorption.Establish proximal targets using chemoproteomics, target-engagement assays, CRISPR perturbation, and rescue experiments.
Multi-target activity should be demonstrated, not assumed from downstream pathway breadth.Match in vitro concentrations to unbound human exposure and test major circulating or microbiota-derived metabolites. Joint tissue distribution and PK/PD relationships should be measured directly.Evaluate established disease, both sexes, aged animals, relevant comorbidities, and combination therapy with methotrexate or biologics. Structural outcomes and infection/toxicity endpoints should accompany arthritis scores.Standardize botanical taxonomy, extraction, chemical fingerprints, active-constituent ranges, impurities, and stability.
Isolated compounds and standardized extracts should be developed as distinct products.Design biomarker-stratified Phase I/II trials with transparent randomization, allocation concealment, validated clinical outcomes, and pre-specified safety monitoring. Myeloid-, lymphoid-, and fibroblast-dominant synovial pathotypes may provide rational enrichment strategies.Treat prevention as a separate research question. Studies in ACPA-positive at-risk individuals would require exceptionally strong safety data and evidence that an intervention changes immune maturation rather than only inflammatory symptoms.
Discussion
The evidence synthesis supports three conclusions. First, the literature contains reproducible biological activity across macrophage, RA-FLS, osteoclast, T-cell, neutrophil, endothelial, and metabolic compartments, but downstream pathway breadth is common and therefore has low discriminative value by itself. Second, translational credibility rises sharply when chemical identity, clinically plausible exposure, and causal target engagement are aligned; this is still uncommon for isolated TCM-derived constituents.
Third, product identity matters: standardized TwHF extracts/glycosides, sinomenine hydrochloride preparations, total glucosides of peony, purified constituents, metabolites, and formulation-enhanced products are different interventions and should not share efficacy claims by association.
This framework also clarifies why an extensive preclinical literature has produced relatively few clinically mature interventions. Preventive dosing in inducible arthritis models, high nominal in vitro concentrations, pathway readouts measured after non-specific cytotoxicity, and lack of rescue or target-engagement experiments can all generate apparently coherent mechanistic narratives without establishing a clinically actionable mechanism. The problem is not that multi-compartment pharmacology is implausible; rather, it remains difficult to distinguish therapeutically useful polypharmacology from chemically driven promiscuity or downstream stress-response suppression.
Among current translational paths, standardized TwHF is the best-supported clinically studied botanical product, although safety and reproductive risk limit generalization. Berberine provides a distinct, exposure-consistent strategy in which gut-restricted or microbiota-mediated action may be more plausible than direct synovial exposure. Selected RA-FLS/osteoclast-directed leads may also be worth developing if structural benefit, target engagement, and a chronic therapeutic window can be shown.
In contrast, high-micromolar parent-compound cell studies without corresponding human exposure represent the weakest translational evidence. These distinctions are more informative than ranking compounds by the number of pathways they are reported to alter.
Accordingly, the next generation of studies should be designed around a product-specific chain of evidence: verified chemistry → unbound exposure at the relevant compartment → proximal target engagement → causal perturbation/rescue → disease-relevant cellular and structural outcomes → combination safety against treat-to-target background therapy. Applying this sequence would allow the field to identify which TCM-derived interventions are credible adjuncts or drug-development leads and which remain primarily pharmacological hypotheses.
Conclusion
TCM-derived bioactive constituents provide useful pharmacological leads for RA, but their evidence hierarchy is highly uneven. The strongest translational path is standardized TwHF as a clinically studied botanical product; a second plausible path is exposure-consistent, gut-restricted berberine pharmacology; and a third is the development of selected RA-FLS/osteoclast-directed molecules with demonstrable target engagement and structural benefit. The weakest evidence consists of high-μM parent-compound cell studies that lack human exposure matching, proximal target validation, or rescue experiments.
For most isolated compounds, structural protection, long-term combination safety, and clinically relevant PK/PD remain unproven. Future work should prioritize four operational requirements: chemically defined interventions, exposure-matched pharmacology, causal target validation in human RA-relevant systems, and adequately powered add-on trials with structural and safety endpoints. Until those requirements are met, TCM-derived compounds should be regarded as mechanistic leads or potential adjuncts rather than replacements for guideline-directed DMARD therapy.