Abstract
Endometriosis affects roughly 10% of women of reproductive age. The defining feature is endometrial-like tissue growing outside the uterus, where it provokes chronic inflammation, fibrosis, and pain. Peritoneal macrophages—normally the cells that clear menstrual debris—become dysfunctional: they accumulate in large numbers, yet their phagocytic capacity is reduced relative to the burden of ectopic tissue, a defect that is multifactorial rather than absolute. Why macrophages settle into these defective states has been unclear. The discovery of trained immunity offers one mechanistic framework. Innate immune cells, including macrophages, can develop a lasting functional memory of prior exposures through epigenetic and metabolic reprogramming. Prior stimulation can produce enhanced responsiveness (trained immunity) or attenuated responses to subsequent challenge (innate immune tolerance); neither outcome necessarily corresponds to a fixed pro- or anti-inflammatory macrophage phenotype. In endometriosis, these two programs have been proposed to exert opposite effects on disease. In a proof-of-concept mouse study, adoptive transfer of BCG-trained macrophages accelerates lesion growth, whereas LPS-tolerized macrophages suppress it—an effect that depends on IL-10 signaling through STAT3 to induce SOCS3 in endometriotic cells. No human study has directly demonstrated trained or tolerized memory in endogenous peritoneal macrophages from patients with endometriosis. Human data are suggestive but retrospective: women with a history of Gram-negative infections (LPS-rich) tend to have milder disease and shorter infertility, while those with Gram-positive infections (which engage TLR2, like BCG) report more pain and more frequent endometriomas; these associations do not demonstrate that microbial products directly imprinted peritoneal macrophages and may be confounded. Single-cell transcriptomics now reveals that endometriotic lesions contain multiple macrophage subsets far beyond the simple M1/M2 binary, and these subsets are shaped by estrogen, extracellular vesicles, DAMPs, and microbial signals. This review draws together the trained immunity and endometriosis literatures to critically evaluate the evidence that pathological macrophage memory contributes to lesion persistence, and to assess whether therapeutically reprogramming this memory—via metabolic modulators like metformin and itaconate derivatives, epigenetic drugs, CD47-SIRPα checkpoint blockade, or nanoparticle-based delivery—could become a non-hormonal treatment strategy; most of these interventions remain untested in endometriosis models.
1 Introduction
Endometriosis is the growth of endometrial-like tissue at ectopic sites—pelvic peritoneum, ovaries, rectovaginal septum—affecting 6–10% of reproductive-aged women (
, ). Chronic pelvic pain, dysmenorrhea, dyspareunia, and infertility are its main clinical burdens ( ). Sampson’s theory of retrograde menstruation explains how endometrial fragments reach the peritoneal cavity—retrograde flow happens in about 90% of menstruating women—but it does not explain why only some develop the disease ( , ).The answer is partly immunological. Macrophages are the most abundant leukocytes in peritoneal fluid and are responsible for clearing menstrual debris (Table 1).
, ). In endometriosis, peritoneal macrophages are present in elevated numbers and show activation markers, yet their phagocytic capacity is reduced, which allows ectopic fragments to survive, implant, and grow ( – ). As lesions establish, macrophage polarization (an activation state driven by the local cytokine milieu) shifts from an early M1-like inflammatory phenotype to a late M2-like tissue-remodeling phenotype that supports angiogenesis, fibrosis, and immune escape ( , , ). This polarization shift is conceptually distinct from innate immune memory: M1/M2 polarization is a transient activation state, whereas trained immunity and tolerance are durable, epigenetically maintained changes in responsiveness to a subsequent stimulus (Table 1
| Feature | Acute activation (M1/M2 polarization) | Trained immunity | Innate immune tolerance |
|---|---|---|---|
| Definition | Reversible activation state driven by the local cytokine milieu | Durable enhancement of responsiveness to a subsequent stimulus | Durable reduction of responsiveness to a subsequent stimulus |
| Trigger | IFN-γ/LPS (M1); IL-4/IL-13/IL-10 (M2) | β-glucan, BCG, heme, oxidized LDL | Repeated low-dose LPS, chronic DAMP exposure, apoptotic cells |
| Persistence after stimulus removal | Resolves (transient) | Weeks to months (epigenetically maintained) | Days to weeks (epigenetically maintained) |
| Molecular substrate | Transcription factor programs (STAT1/NF-κB vs STAT6/PPARγ) | H3K4me3/H3K27ac at inflammatory gene loci; aerobic glycolysis | H3K4me3 at IL10/TGFB; repressive marks at TNF/IL6; itaconate production |
| Relationship to M1/M2 | Is the M1/M2 state | Often M1-like in cytokine output, but not reducible to M1 | Often M2-like/IL-10-producing, but not reducible to M2 |
| Reversibility | Fully reversible by changing cytokines | Reversible in principle by epigenetic/metabolic reprogramming | Reversible in principle; requires re-education |
| Dependence on continued stimulation | Requires ongoing cytokine signal | Persists after stimulus cleared | Persists after stimulus cleared |
Conceptual distinctions among acute macrophage polarization, trained immunity, and innate immune tolerance.
Macrophage polarization reflects a context-dependent activation state, whereas trained immunity and innate immune tolerance are defined by altered responses to subsequent stimulation. M1/M2 markers alone cannot establish innate immune memory.
What locks macrophages into this maladaptive state? Adaptive immune memory—antigen-specific T and B cells—has been studied extensively in endometriosis (
, ) but does not fully explain the durable reprogramming seen in peritoneal macrophages. Over the past fifteen years, the concept of trained immunity has changed the way innate immune memory is understood ( – ). Monocytes, macrophages, and natural killer (NK) cells can develop long-term functional memory of prior inflammatory encounters through epigenetic and metabolic rewiring ( – ). Depending on the stimulus, this memory takes one of two forms: an enhanced inflammatory response (trained immunity sensu stricto) or a dampened, anti-inflammatory state (innate immune tolerance) ( , ).Jeljeli and colleagues showed in 2020 that these two memory states can steer endometriosis in opposite directions in a mouse model (
). By adoptively transferring differentially trained macrophages into mice, they found that BCG-trained macrophages accelerated lesion growth while LPS-tolerized macrophages suppressed it via an IL-10-dependent mechanism ( ). This study provided the first proof-of-concept linking innate immune memory to a gynecological disease, but it used in vitro-trained, adoptively transferred macrophages and does not establish that endogenous peritoneal macrophages in women with endometriosis are themselves trained or tolerized.This review covers the evidence linking trained immunity and tolerance to endometriosis. It starts with the molecular machinery of innate memory—the histone marks, the metabolic switches, and the role of hematopoietic stem cells—before examining how the endometriotic microenvironment engages these same pathways, using estrogen, DAMPs, extracellular vesicles, and microbial products as cues that may train or tolerize peritoneal macrophages. It then works through the consequences for lesion persistence, fibrosis, pain, and fertility, and, throughout, distinguishes direct experimental evidence from indirect support and mechanistic inference, and ends with a survey of therapeutic strategies aimed at reprogramming macrophage memory toward a disease-suppressing state.
1.1 Literature search and evidence selection
For this narrative review, relevant literature was identified through PubMed and Web of Science using combinations of terms relating to endometriosis, macrophages, trained immunity, innate immune tolerance, immunometabolism, epigenetic reprogramming, and therapeutic modulation. Studies were selected for their relevance to the review questions, with priority given to original disease-specific experimental studies and complementary human observations, while mechanistic evidence from other disease settings and pertinent reviews provided context. Evidence was appraised in relation to study design, species, cellular source, direct functional demonstration of memory after restimulation, and relevance to endometriosis. Findings from independent research groups were considered to avoid placing disproportionate interpretative weight on any single study or research team. Because this was a narrative rather than systematic review, no exhaustive or quantitatively reproducible screening process is claimed.
2 Trained immunity: mechanisms and conceptual framework
2.1 Definition and historical context
Netea and colleagues introduced “trained immunity” in 2011 to describe the heightened responsiveness of innate immune cells after a primary challenge with homologous or heterologous stimuli (
). They distinguished it from classical immune priming—a transient state that fades once the infection clears—and from adaptive memory, which requires V(D)J recombination and clonal expansion ( , ). Trained immunity is antigen-nonspecific, operates in monocytes, macrophages, and NK cells, and is maintained by functional reprogramming rather than genetic alteration ( – ).Two observations drove the concept. Plants and invertebrates, which lack adaptive immunity entirely, mount stronger responses to secondary infections. And BCG vaccination in humans reduces all-cause infant mortality by roughly 43%—an effect too large to be explained by tuberculosis prevention alone (
– , ). Monocytes from BCG-vaccinated individuals produce more TNF-α, IL-1β, and IL-6 when stimulated ex vivo with unrelated microbial ligands, and this heightened state persists for months ( , , ).2.2 Acute activation, trained immunity, and tolerance: a conceptual distinction
A recurring source of confusion is the conflation of macrophage polarization with innate immune memory. These are related but not equivalent. Polarization (often summarized as M1 versus M2) describes a reversible activation state imposed by the prevailing cytokine milieu; it largely resolves when the cytokines are withdrawn. Trained immunity and tolerance, by contrast, are defined by an altered response to a subsequent stimulus that is sustained by epigenetic and metabolic reprogramming and can persist for weeks to months after the initial stimulus is removed (Table 1 and illustrated in Figure 1.
, ). A trained cell is not simply an M1 cell, and a tolerized cell is not simply an M2 cell. This distinction matters because the manuscript literature on endometriosis has relied heavily on M1/M2 surface markers, which are transient and context-dependent and cannot by themselves serve as surrogate indicators of trained or tolerized states. The key conceptual distinctions among acute macrophage polarization, trained immunity, and innate immune tolerance are summarized inFigure 1
2.3 Epigenetic reprogramming as the memory substrate
Trained immunity is written into the epigenome—not as permanent genetic changes, but as sustained modifications to chromatin structure and histone marks at the regulatory regions of inflammatory and metabolic genes (
, , ). H3K4me3, a mark of active promoters, is enriched at the TNF, IL6, and IL1B loci in trained monocytes ( , , ). H3K27ac accumulates at distal enhancers, keeping them poised for rapid reactivation ( , ). These marks last for weeks to months and survive cell division ( ). In the β-glucan-trained monocyte system, this enrichment was mapped directly by Saeed and colleagues (2014) as part of the BLUEPRINT Epigenome Project: training enriched H3K4me3 at pro-inflammatory gene promoters, whereas LPS tolerization enriched it at the IL10 promoter instead ( ). The specific H3K4me3 accumulation at TNF, IL6, and IL1B loci in trained monocytes was demonstrated in this study, rather than representing generic inflammatory activation. BCG vaccination alters DNA methylation in hematopoietic stem and progenitor cells (HSPCs), and these changes are passed to myeloid progeny ( , ). Methylation is less dynamic than histone modifications but contributes to the stability of the trained phenotype over months to years ( ). ATAC-seq shows that trained monocytes open chromatin at binding sites for IRF, STAT, AP-1, and C/EBP transcription factors—the master regulators of inflammatory gene programs ( , ). The accessible regions are enriched for latent enhancers (H3K4me1-marked) that gain H3K27ac upon restimulation ( ).2.4 Metabolic rewiring: the engine of innate memory
Epigenetic reprogramming during trained immunity is coupled to metabolic rewiring. Cheng and colleagues (2014) showed that β-glucan training shifts human monocytes from oxidative phosphorylation (OXPHOS) to aerobic glycolysis, and this switch runs through dectin-1→Akt→mTOR→HIF-1α (
). Rapamycin, Akt inhibitors, or HIF-1α blockade each abolished training. Mice with myeloid-specific Hif1a deletion could not mount trained immunity against polymicrobial sepsis ( ). The causal direction runs: metabolic shift → epigenetic change → trained functional state.Several Tricarboxylic acid (TCA) cycle intermediates act as signals connecting metabolism to chromatin (
– ). Succinate accumulates during TCA cycle remodeling and stabilizes HIF-1α by inhibiting prolyl hydroxylases, driving glycolytic gene transcription and IL-1β production; this succinate→HIF-1α→IL-1β axis was established by Tannahill and colleagues (2013), who showed that LPS-stimulated macrophages accumulate succinate and that this succinate promotes IL-1β via HIF-1α ( ). Fumarate inhibits KDM5 histone demethylases, which keeps H3K4me3 marks at pro-inflammatory gene promoters and reinforces training ( ). Acetyl-CoA, generated from glycolysis and fatty acid oxidation, supplies acetyl groups for histone acetyltransferases (HATs), raising H3K27ac at inflammatory enhancers ( , ). Itaconate, produced from citrate by IRG1 (ACOD1), does the opposite: it inhibits succinate dehydrogenase (SDH), alkylates KEAP1 to activate the Nrf2 antioxidant pathway, and suppresses glycolysis ( , ). Cell-permeable derivatives such as 4-octyl itaconate and dimethyl itaconate reduce inflammation in sepsis, autoimmunity, and ischemia-reperfusion models ( , ). α-Ketoglutarate (α-KG), from glutaminolysis, promotes tolerance through TET-mediated DNA demethylation and JmjC-domain histone demethylases ( , ). The balance between succinate/fumarate (pro-training) and itaconate/α-KG (pro-tolerance) is a metabolic rheostat. Which way it tips determines whether a macrophage becomes trained or tolerized.2.5 Central versus peripheral trained immunity
In endometriosis, the duration of any endogenous memory state after withdrawal of estrogenic, microbial or damage-related cues has not been measured. If responsiveness depends on continuing exposure, transient control of local triggers may suffice; if a stable memory state persists, sustained or repeated peripheral reprogramming might be needed. A centrally imprinted HSPC program would also raise the possibility that local peritoneal treatment alone may not provide a durable reset. These competing possibilities require paired baseline, washout, and rechallenge measurements before treatment duration can be rationally defined.
Trained immunity operates at two anatomical levels. Central training happens in the bone marrow, where inflammatory signals reprogram long-lived HSPCs epigenetically and metabolically, biasing their differentiation toward myeloid progeny with enhanced inflammatory capacity (
, – ). IFN-γ (from BCG), IL-1β (from β-glucan), and GM-CSF all trigger this reprogramming ( , , ). Once imprinted, the HSPCs produce trained monocytes and macrophages for months to years ( , ). Johansson and colleagues (2023) showed that both hematopoietic stem cells (HSCs) and self-renewing tissue-resident macrophages acquire durable epigenetic changes that increase chromatin accessibility at myeloid-lineage genes and sustain enhanced myelopoiesis ( ). Daman and colleagues (2024) went further, demonstrating that inflammation-induced HSPC reprogramming can alter systemic immune tone, including anti-tumor responses ( ). Peripheral training occurs locally in tissue-resident macrophages, without bone marrow involvement. Repeated exposure to microbial products (LPS, β-glucan), endogenous DAMPs (oxidized LDL, uric acid, HMGB1), or metabolic signals (hyperglycemia) is sufficient ( , , ). Peripheral training is likely more dynamic and reversible than central training ( ). Central and peripheral training are not mutually exclusive and may interact: a peripherally trained tissue macrophage pool can be continuously replenished and re-educated by centrally reprogrammed HSPCs, so both compartments could, in principle, sustain a memory state within the peritoneal cavity. Whether peritoneal macrophages in endometriosis undergo one, the other, or both is an open question with direct therapeutic implications.2.6 Innate immune tolerance
Innate immune tolerance is trained immunity’s mirror image: a state of reduced responsiveness on secondary stimulation. Pro-inflammatory cytokine output drops; IL-10 and TGF-β rise (Table 2.
, , ). Importantly, tolerance should not be equated with a uniformly anti-inflammatory or inherently beneficial macrophage state: it reflects a quantitatively and qualitatively altered responsiveness to subsequent stimulation whose functional consequences are context-dependent. Depending on the setting, a tolerized macrophage may be tissue-protective (limiting collateral inflammation) or permissive (failing to clear pathogens or ectopic tissue), and these two faces are both relevant to endometriosis. The canonical inducer is repeated low-dose LPS (endotoxin tolerance), but other TLR ligands, sustained DAMP signaling, or prolonged exposure to apoptotic cells can produce the same result ( , ). The same epigenetic machinery is used, but the marks land on different genes. In tolerized macrophages, H3K4me3 accumulates at IL10 and TGFB promoters, while repressive marks (H3K27me3, H3K9me2) cover TNF and IL6 ( , ). The TCA cycle shifts toward itaconate and away from succinate ( , ). The key transcription factors driving tolerance are NF-κB p50/p50 homodimers, BCL-3, and STAT3—the last activated by IL-10 itself ( , ). A side-by-side comparison of trained immunity and innate immune tolerance is provided inTable 2
| Feature | Trained immunity (pro-inflammatory) | Innate immune tolerance (reduced responsiveness) |
|---|---|---|
| Prototypical inducer | BCG, β-glucan (standard dose), oxidized LDL, heme | Repeated low-dose LPS (LPSlow), chronic DAMP exposure, apoptotic cells |
| Primary receptor(s) | Dectin-1 (β-glucan), NOD2 (BCG), TLR2, TLR4 | TLR4 (sustained), TAM receptors (Tyro3/Axl/MerTK) |
| Key signaling pathway | Dectin-1→Syk→Akt→mTOR→HIF-1α | NF-κB p50/p50 homodimers, BCL-3, STAT3 (IL-10→pSTAT3) |
| Dominant histone marks | H3K4me3↑ at TNF, IL6, IL1B promoters; H3K27ac↑ at enhancers | H3K4me3↑ at IL10, TGFB promoters; H3K27me3↑ at TNF, IL6 promoters |
| DNA methylation | Not a uniform diagnostic signature; methylation changes depend on the inducer, cell lineage, and experimental system. | Context-dependent changes; no universal hypermethylation signature defines innate immune tolerance. |
| Metabolic profile | Aerobic glycolysis; ↑ lactate; succinate accumulation; fumarate → KDM5 inhibition | OXPHOS preserved; itaconate production (IRG1/ACOD1); Nrf2 activation; α-KG → TET/JmjC activity |
| Key pro-inflammatory cytokines | TNF-α ↑↑, IL-6 ↑↑, IL-1β ↑↑, CCL2 ↑↑ | TNF-α ↓, IL-6 ↓, IL-1β ↓, CCL2 ↓ |
| Key anti-inflammatory cytokines | IL-10 production is variable and stimulus-dependent; a low IL-10 level is not required to define training. | IL-10/TGF-β responses vary by experimental context and are not sufficient to establish tolerance. |
| Surface markers (context-dependent) | No validated universal surface-marker panel; activation-associated markers may vary by species and stimulus. | No validated universal surface-marker panel; CD206 or other polarization markers do not demonstrate tolerance. |
| Phagocytic capacity | Preserved or enhanced | Variable and context-dependent; not an invariant feature. Reduced phagocytosis in endometriosis involves lesion-encoded checkpoints (CD47-SIRPα, CD200-CD200R), which are inhibitory signals rather than canonical markers of tolerance |
| Effect on endometriosis (mouse model; single study, Jeljeli et al., 2020 ( | ))Accelerated lesion growth (↑ volume, weight, growth kinetics) after adoptive transfer of in vitro-trained macrophages | Suppressed lesion growth (↓ volume, weight, growth kinetics) after adoptive transfer of in vitro-tolerized macrophages; IL-10-dependent |
| Human clinical correlate (retrospective and inferential) | GP infection history → more severe pain, ↑ ovarian endometrioma | GN infection history → shorter infertility, milder disease |
| Therapeutic implication | Target for suppression (metformin, rapamycin, itaconate derivatives) | Potential modulation of responsiveness (e.g., MPL or apoptotic-cell approaches); disease-specific memory effects unproven. β-glucan is not established as a tolerizing agent. |
Molecular, metabolic, and functional characteristics of trained immunity and innate immune tolerance in macrophages.
The characteristics summarized are derived from different experimental systems and may vary with cellular context and stimulation conditions. Surface markers and phagocytic activity should not be considered universal indicators of trained immunity or innate immune tolerance.
3 Macrophage diversity in the endometriotic microenvironment
3.1 Beyond M1/M2: single-cell resolution
Macrophages in endometriosis have long been sorted into the M1/M2 binary (Table 1).
, , ). In this scheme, M1 cells (IFN-γ + LPS) make TNF-α, IL-1β, and IL-6 and are associated with host defense; M2 cells (IL-4, IL-13, IL-10, TGF-β) promote tissue remodeling, angiogenesis, and immune suppression ( , , ). The standard disease narrative is an M1→M2 shift: early lesions are M1-dominant, established ones M2-dominant ( , , ). It is important to recognize that this M1→M2 transition describes a cytokine-driven change in polarization state, not necessarily the establishment of innate immune memory. Whether the shift is accompanied by, or arises from, durable trained or tolerized states is a separate question that requires functional and epigenomic evidence (seeSingle-cell transcriptomics has made this binary untenable (
, , ). Henlon and colleagues (2024) performed single-cell RNA sequencing in an experimental mouse model of endometriosis and found multiple macrophage subpopulations with pro-disease or pro-resolving signatures that do not map cleanly onto M1 or M2 ( ). Genome-wide association study (GWAS) data integrated with scRNA-seq maps endometriosis risk loci predominantly to an M2-like macrophage subset, suggesting a possible contribution of inherited susceptibility to macrophage-related disease pathways ( ). Endometriotic lesions do not contain one macrophage type. They contain a dynamic ecosystem of functionally distinct states, each shaped by local cues. Candidate populations for testing include the mouse lesion-associated Spp1/Trem2/Mmp12/Cd9 scar-associated cells, Folr2/Mrc1/Gas6/Ccl8 tumor-associated-like cells, and Apoe/Saa3/Pid1 lipid-associated, proresolving large peritoneal macrophages reported by Henlon et al. ( ). These are transcriptionally defined candidates—not validated trained or tolerized subsets. Sorted populations would require an initial stimulus–washout–rest–heterologous rechallenge protocol, quantitative cytokine and phagocytosis readouts, chromatin accessibility or histone-mark profiling, and metabolic measurements; persistence and reversibility should be compared across time points.A caveat is warranted: single-cell transcriptomics captures transcriptional state at one moment, not the latent memory of prior exposures that defines trained immunity or tolerance. No scRNA-seq-defined subset has yet been demonstrated to represent a trained or tolerized macrophage, and trained versus tolerized states are unlikely to be identifiable from gene expression alone, because the same transcriptome can arise from either acute activation or durable reprogramming. Resolving this will require functional restimulation assays together with epigenomic and metabolic profiling—ATAC-seq, histone ChIP-seq, and metabolomics or extracellular-flux measurements—applied to prospectively isolated subsets. The relationship between scRNA-seq-defined subsets and the M1/M2 paradigm is therefore indirect: these subsets are best understood as context-dependent activation states whose future responsiveness (training versus tolerance) remains to be defined. Although Henlon et al. also included human stromal-cell and endothelial assays and cross-species comparisons, their macrophage discovery populations were profiled by scRNA-seq in experimental mice and should not be treated as human clinical-stage subtypes (Table 3.
). A detailed summary of macrophage subsets is given inTable 3
| Macrophage subset | Key markers | Anatomic location | Proposed function | Disease association | Species/study type |
|---|---|---|---|---|---|
| Scar-associated macrophages | Mmp12, Cd9, Spp1, Trem2 | Mouse lesions | Fibrosis/matrix-remodeling signature | Pro-disease-associated; no human stage assignment | Mouse scRNA-seq ( | )
| Tumor-associated-like macrophages | Folr2, Mrc1, Gas6, Ccl8 | Mouse lesions | Human stromal-cell COL1A1/TGFB1 and endothelial angiogenesis responses in separate assays | Pro-disease activity in models; not clinical staging | Mouse scRNA-seq; human-cell assays ( | )
| Lipid-associated proresolving LpMs | Apoe, Saa3, Pid1 | Mouse peritoneum | Lipid handling; Apoe mimetic reduced mouse lesion size/fibrosis | Putatively protective; memory untested | Mouse scRNA-seq; Apoe intervention ( | )
| M2-like macrophages (contextual) | Variable; ARG1 not universal in humans | Human/mouse tissue | Repair, angiogenesis, suppression (context-dependent) | No validated stage or memory identity | Heterogeneous reports ( | , , )
| Macrophage–myofibroblast transition | CD68, ACTA2, COL1A1, COL3A1 | Mouse lesion fibrosis | Matrix synthesis; TGFB1/SMAD3-associated | Fibrotic experimental phenotype | Mouse single-cell study ( | )
| Large peritoneal macrophages | CD11b, F4/80hi, ICAM2, GATA6 | Mouse peritoneum | Homeostasis and debris clearance | Phagocytosis context-dependent; not tolerance | Mouse-focused reviews ( | , )
| Small peritoneal macrophages | CD11b, F4/80lo, MHC-IIhi, CCR2 | Mouse peritoneum | Recruited inflammatory macrophages | No fixed M1→M2 transition proven | Mouse-focused reviews ( | , )
Macrophage subsets in the endometriotic microenvironment.
Macrophage subsets are summarized from human and mouse studies using single-cell transcriptomics and other experimental approaches. Species and study types are specified to distinguish findings obtained from different experimental systems. The macrophage populations attributed to Henlon et al. (
) were characterized in a mouse model and should not be treated as validated human subtypes. Rows attributed to Henlon et al. ( ) are limited to populations characterized in the mouse experimental system; macrophage categories drawn from other studies are contextual and should not be interpreted as equivalent single-cell clusters.3.2 Estrogen as a macrophage regulator
Ectopic lesions have high aromatase activity and produce estradiol (E2) locally (
). Macrophages in the peritoneal fluid of women with endometriosis overexpress both ERα and ERβ ( ). Macrophages in endometriotic lesions express estrogen receptors, and estrogen signaling in macrophages—particularly through ERα—has been directly implicated in recruiting and activating macrophages within lesions ( , ). In mice, E2 increases lesional macrophage numbers and upregulates Csf1, Nt3, and TrkB ( ). Greaves and colleagues (2015) showed that E2 orchestrates bidirectional macrophage-nerve crosstalk: neurons exposed to E2 release CCL2, recruiting macrophages; E2-treated macrophages secrete BDNF and NT-3, promoting neurite outgrowth into lesions ( ). Endometriosis is, in this sense, an estrogen-dependent neuroinflammatory disorder.Estrogen receptor signaling may shape macrophage recruitment and activation during lesion development (
, ). Experimental models suggest stage- and context-dependent contributions of ERβ- and ERα-associated pathways, but a universal ERβ-to-ERα switch within a defined 72-hour interval has not been established. Accordingly, associated changes in inflammatory and tissue-remodeling phenotypes should not be interpreted as a fixed temporal sequence in human endometriosis. These interactions may contribute to feedback between local estrogen signaling, macrophage activation, and lesion maintenance, although their timing and directionality require further investigation. Again, this ERβ→ERα and M1→M2 sequence is a description of polarization dynamics; whether these estrogen-driven states additionally imprint durable trained or tolerized memory is not yet known.For comparison, a Mendelian randomization study of endometrial cancer reported associations involving menstrual and pregnancy-related traits and examined metabolic mediators (
). These cancer-specific findings provide broader reproductive-metabolic context but should not be extrapolated to macrophage memory in endometriosis.3.3 Defective phagocytosis
A defining functional defect of endometriosis-associated macrophages is impaired phagocytosis—the very process that should clear ectopic endometrial cells (
, , ). This phagocytic defect is relative rather than absolute and is multifactorial: macrophages remain partially functional, but their clearance capacity is outstripped by the burden of ectopic tissue and is suppressed by multiple lesion-encoded signals. CD47, a “don’t eat me” signal, is upregulated on ectopic endometrial stromal cells and on the small extracellular vesicles (sEVs) they secrete ( , ). CD47 binds SIRPα on macrophages, triggering SHP-1/SHP-2 phosphatase signaling that blocks phagocytosis ( , ). Li and colleagues (2022) showed that a CD47-blocking antibody restored macrophage phagocytosis of endometrial stromal cells and induced apoptosis in those cells, positioning CD47-SIRPα as a checkpoint in endometriosis analogous to cancer ( ). Wang and colleagues (2026) added that lesion-derived sEVs carry CD47 along with epithelial markers (CD133, EpCAM), making them a delivery vehicle for phagocytic suppression in the peritoneal space ( ). Beyond CD47, CD200 on endometrial cells engages CD200R on macrophages, adding a second inhibitory signal ( ). PGE2 from COX-2-overexpressing endometriotic cells suppresses macrophage phagocytic activity ( , ). And IL-10 from tolerized macrophages and Tregs dampens effector functions further ( , ). These checkpoints (CD47-SIRPα, CD200-CD200R) are lesion-encoded inhibitory signals that impair phagocytosis; they are not canonical features of innate immune tolerance and should not be read as evidence that endometriosis-associated macrophages are tolerized.4 Trained immunity versus innate immune tolerance in endometriosis: evidence
The experimental evidence linking trained immunity and tolerance to endometriosis is direct but narrow: it rests on a single mechanistic study that manipulated macrophage memory in vitro and tested its consequences by adoptive transfer into mice. The evidence summarized in this section should be read with this limitation in mind.
4.1 BCG training accelerates lesion growth
Jeljeli and colleagues (2020) trained peritoneal macrophages in vitro with BCG, then adoptively transferred them into mice with induced endometriosis (
). The trained macrophages adopted a pro-inflammatory memory phenotype: CD43, CD206, CCR2, CXCR4, CD80, TLR2, and Ly6C were elevated, and restimulation produced high levels of TNF-α, IL-6, and CCL2 ( ). They showed enhanced glycolysis with increased lactate—the metabolic signature of trained immunity ( ). At the chromatin level, H3K4me3 was enriched at TNF and IL6 promoters ( , ). Mice receiving BCG-trained macrophages developed larger lesions—higher volume, weight, and growth rate—than mice receiving untrained cells ( ). This experiment demonstrates that adoptively transferred, in vitro-trained macrophages are sufficient to accelerate lesion growth in mice; it does not show that endogenous macrophages in human lesions are trained, nor that trained immunity is necessary for lesion persistence. A pro-inflammatory macrophage memory that is beneficial during infection becomes, in endometriosis, a driver of lesion expansion.4.2 LPS-induced tolerance suppresses lesions through IL-10/STAT3/SOCS3
When the same group trained macrophages with repeated low-dose LPS (LPSlow) instead, the outcomes reversed (
). LPSlow-tolerized macrophages expressed less CCR2, CXCR4, CD80, and Ly6C; produced less TNF-α, IL-6, and CCL2; and made more IL-10 ( ). H3K4me3 clustered at the IL10 promoter, matching the chromatin signature of endotoxin tolerance ( , ). Adoptive transfer of these tolerized macrophages suppressed lesion volume, weight, and growth ( ). Silencing Il10 in the LPSlow macrophages eliminated the protective effect, demonstrating that IL-10 is necessary for the suppression observed in this model. Mechanistically, IL-10 from the tolerized cells bound receptors on endometriotic cells, phosphorylated STAT3, and induced SOCS3 and BCL3 ( ). Downstream, this IL-10→pSTAT3→SOCS3 axis reduced expression of CXCR4 (migration/homing), OCT4 and SNAI1 (EMT and stemness), and COL1A1 (fibrosis) ( ).A retrospective clinical observation provides a possible, but indirect, parallel: women reporting previous severe Gram-negative infections had shorter infertility duration and less severe endometriosis than those without such a history (
). Such infections may involve exposure to LPS-containing microbial products, but direct exposure or memory reprogramming of peritoneal macrophages was not demonstrated. Conversely, women with a Gram-positive infection history (potentially involving different microbial ligands and host responses) reported more pelvic pain and more ovarian endometriomas ( ). These are retrospective, self-reported associations from a single cohort. They do not demonstrate that microbial products directly trained or tolerized peritoneal macrophages in these patients, and they are vulnerable to confounding (for example, differences in health-care seeking, antibiotic exposure, or other unmeasured variables). The link between prior infection, peritoneal macrophage exposure to LPS or TLR2 ligands, and subsequent disease severity therefore remains inferential and should be interpreted as hypothesis-generating rather than as direct evidence.Related genetic epidemiology research has examined antibody-response phenotypes and ovarian cancer susceptibility, including an association with genetically predicted anti-polyomavirus 2 IgG seropositivity (
). However, antibody seropositivity is not equivalent to active infection or macrophage training, and these findings cannot validate the retrospective infection-history associations described here.4.3 IL-10: protective and pathogenic
IL-10 is the key mediator of macrophage-driven lesion suppression in this model (
), which is counterintuitive. IL-10 is immunosuppressive and is usually associated with M2 macrophages that aid tumor growth and immune evasion ( , , ). The function of IL-10 in endometriosis is context-dependent and cannot be attributed to its cellular source alone: its biological effect is determined jointly by its source, by the responding cell type, and by the local signaling context ( ). Tolerized macrophage-derived IL-10 acts on endometriotic cells through STAT3→SOCS3 and dampens the fibro-inflammatory phenotype ( ). regulatory T cell (Treg)-derived IL-10—produced in response to CCL25/TECK from endometrial stromal cells—does the opposite: it promotes immune escape, proliferation, and invasion ( ). This split explains why blanket IL-10 blockade is not a straightforward strategy in endometriosis. IL-10 signals through a heterodimeric receptor (IL-10R1/IL-10R2) that activates JAK1/TYK2 and downstream STAT3 ( , ); the same receptor–STAT3 axis can produce opposite outcomes depending on which cell receives the signal and when. In endometriotic epithelial-stromal cells, IL-10-induced STAT3 induces SOCS3 and BCL3, restraining CXCR4, stemness, and fibrotic gene expression ( ); in other immune cells, IL-10 can instead suppress effector responses and promote immune evasion ( , ). Timing also matters: early in disease, IL-10 may limit the initial inflammatory wave, whereas sustained IL-10 later in disease may contribute to the immune-evasive, pro-remodeling milieu. These contextual features—receptor availability, responding cell identity, and temporal dynamics—rather than the source alone, determine whether IL-10 is protective or pathogenic ( , ). Importantly, available endometriosis studies do not establish whether IL10RA (IL-10R1) and IL10RB (IL-10R2) are differentially expressed across endogenous peritoneal macrophage subsets, regulatory T cells, and ectopic epithelial or stromal cells in human lesions. Comparative, cell-type-resolved receptor-expression measurements, paired with IL-10-induced STAT3 phosphorylation and downstream response assays across disease stages, are needed. Thus, differences in IL-10 receptor abundance or signaling competence remain a proposed explanation—not a demonstrated mechanism—for the divergent effects attributed to IL-10 in endometriosis.4.4 Endogenous cues that train peritoneal macrophages
Several signals within the endometriotic microenvironment may contribute to macrophage functional reprogramming, although their capacity to induce durable innate immune memory in endometriosis remains unestablished. Ectopic endometrial cells undergo cyclic bleeding and die, releasing HMGB1, HSPs, ATP, uric acid, and mitochondrial DNA that engage TLR2, TLR4, TLR9, and NLRP3 on peritoneal macrophages (Figure 2. Importantly, the proposed contribution of persistent innate immune memory remains hypothetical in human endometriosis.
, , ). It is plausible, but unproven, that chronic, repetitive DAMP exposure may push macrophages toward tolerance, analogous to repeated low-dose LPS—a mechanism that would be consistent with the phagocytic impairment seen in established disease ( , ). Conversely, acute, high-dose DAMP exposure during menstruation could instead drive inflammatory training. These are hypotheses; no study has directly shown that DAMPs imprint trained or tolerized memory in peritoneal macrophages in endometriosis. Lesion-derived sEVs carry proteins, lipids, mRNA, and microRNAs that transfer to recipient macrophages ( ). Whether sEV cargo induces training or tolerance is not known, but the fact that sEVs carry CD47 means they are well positioned to modulate macrophage function. The gut and reproductive tract microbiome is increasingly linked to endometriosis ( , ). LPS (Gram-negative), lipoteichoic acid (Gram-positive), and β-glucan (fungal) that reach the peritoneal cavity through compromised barriers could provide tonic training or tolerization signals to resident macrophages ( , ). That GN infection history associates with milder disease and GP history with worse disease ( ) is consistent with, but does not establish, a model in which different microbial products imprint macrophages in different directions. Ectopic lesions undergo cyclic microhemorrhage, releasing hemoglobin and heme ( , ). Iron drives oxidative stress through Fenton chemistry, and labile heme has recently been recognized as a bona fide inducer of trained immunity. In contrast to canonical trained-immunity ligands, heme training depends on spleen tyrosine kinase (Syk) signaling acting upstream of c-Jun N-terminal kinase (JNK) and confers long-term skewing of hematopoietic stem and progenitor cells toward myelopoiesis ( ). The potential interactions among local microenvironmental signals, macrophage functional states, and disease-associated outcomes are summarized inFigure 2
In a separate reproductive tract condition, a two-sample Mendelian randomization study investigated genetically proxied skin microbiota and vaginitis (
). This observation illustrates growing interest in microbiota-related gynecological conditions but does not establish microbial exposure or innate immune memory in the endometriotic peritoneum.Local signals, including estrogen, prostaglandin E2 (PGE2), heme, damage-associated molecular patterns (DAMPs), extracellular vesicles, and microbial products, may influence macrophage activation, phagocytosis, cytokine production, and immunometabolism. These changes may contribute to lesion persistence, fibrosis, and neuroinflammation. However, whether such signals induce durable trained immunity or innate immune tolerance in endogenous macrophages during human endometriosis remains uncertain. Dashed arrows indicate hypothetical relationships requiring experimental validation.
5 From macrophage memory to lesion persistence
The links between macrophage memory and the clinical hallmarks of endometriosis—fibrosis, pain, and infertility—are, at present, largely indirect. Direct evidence exists only for lesion growth in the adoptive-transfer model (
); the extension of this framework to fibrosis, pain, and fertility rests on mechanistic inference and should be read as hypothesis-generating rather than established.5.1 Fibrosis and tissue remodeling
Fibrosis drives pain, adhesions, and infertility in endometriosis (
, ). Zhong and colleagues (2025) identified TGFB1/SMAD3-driven macrophage-myofibroblast transition (MMT) as a source of myofibroblasts in lesions ( ). A CD68+/α-SMA+ macrophage population, positioned at a fibrotic terminal state along pseudotime, produces collagen and other ECM proteins. These MMT cells originate from M2 macrophages, and pharmacological TGFB1/SMAD3 inhibition reduces MMT cell numbers, collagen deposition, and lesion volume in mice ( ). LPSlow-tolerized macrophages suppress COL1A1 expression in endometriotic cells ( ), which implies, but does not prove, that macrophage memory states shape the fibrotic program. By extension, it is plausible that trained macrophages, with their sustained output of TNF-α, TGF-β, and PDGF, could promote MMT, whereas tolerized macrophages, with IL-10, could restrain it; these are testable hypotheses rather than demonstrated outcomes. The studies considered in this narrative review do not establish a causal role for persistent macrophage innate immune memory in endometriosis-associated fibrosis; this is a review-level assessment rather than a claim based on an exhaustive systematic search.5.2 Pain and neuroinflammation
Chronic pelvic pain is the cardinal symptom (
, ). The macrophage-nerve crosstalk described above—CCL2→macrophage recruitment, BDNF/NT-3→neurite sprouting—correlates with pain severity ( ). It is plausible, though not yet shown, that trained macrophages, which produce more TNF-α and IL-1β, could amplify this loop, since both cytokines sensitize nociceptors ( , ); this remains speculative. GP infection history, associated with training-type stimuli, correlates with worse pelvic pain in the retrospective analysis discussed above ( ).5.3 Infertility
Endometriosis impairs fertility through distorted anatomy, poor oocyte quality, reduced endometrial receptivity, and an inflammatory peritoneal environment toxic to sperm and embryos (
, ). Macrophages infiltrate the ovary, follicular fluid, and endometrium, and their polarization affects reproductive outcomes ( ). Early disease with M1-dominant macrophages produces excessive inflammatory cytokines that damage gametes ( , ). Late disease with M2 dominance creates a fibrotic uterine environment hostile to implantation ( , ). By analogy with the lesion-growth data, trained macrophages with amplified TNF-α and IL-6 might worsen inflammatory damage to gametes, and tolerized macrophages might create a more permissive environment; however, no direct evidence links trained or tolerized macrophages to fertility outcomes in endometriosis. The clinical observation that GN infection history associates with shorter infertility ( ) is retrospective and does not establish causation.6 Therapeutic reprogramming of macrophage memory
6.1 Rationale
The Jeljeli study showed two things in a mouse model: adoptively transferred macrophage memory can steer endometriosis severity, and that memory can be reprogrammed in vitro before transfer (Table 4.
). This raises the possibility of therapies that do not suppress or eliminate macrophages but instead redirect their functional memory from pathogenic to protective. The goal is to shift macrophages from pro-inflammatory training toward tolerance, using the same epigenetic and metabolic pathways that nature uses to set innate immune responsiveness. It must be emphasized that these strategies remain preclinical and, for the most part, untested in endometriosis. A useful distinction is between interventions that directly target the molecular machinery of trained immunity or tolerance (metabolic and epigenetic modulators) and broader macrophage-directed or anti-endometriotic approaches—such as CD47-SIRPα blockade, anti-fibrotic agents, and nanoparticle-based delivery—that may alter macrophage function or lesion biology without necessarily reprogramming innate immune memory itself. Each strategy is considered below with an indication of its current evidence base, and the distinction is summarized inTable 4
| Evidence strength/strategy | Translation challenges | Off-target effects | Central vs. peripheral |
|---|---|---|---|
| Metabolic mTOR/HIF-1α modulation (metformin/rapamycin): pathway established in trained-immunity models ( | ); macrophage-memory reset in endometriosis unproven.Dose, chronic safety and peritoneal target engagement; metformin observations confounded by PCOS. | Systemic immunosuppression and effects on non-myeloid metabolism. | Systemic exposure may affect both compartments; local peripheral specificity unproven. |
| Itaconate derivatives: anti-inflammatory/metabolic effects in non-endometriosis models ( | , ); endometriosis memory effect hypothetical.Peritoneal exposure, stability and pharmacokinetics undetermined. | Broad electrophilic targets, Nrf2 signaling and non-myeloid activity. | Peripheral delivery is a proposal; central consequences not established. |
| HDAC/BET modulation: chromatin target rationale from other models ( | , ); memory reversal in endometriosis hypothetical.Selectivity, dose and long-term safety in benign chronic disease. | Widespread epigenomic and transcriptional changes. | Systemic distribution could reach central and peripheral cells; preferential local targeting untested. |
| CD47–SIRPα blockade: endometriosis-related in-vitro/ex-vivo phagocytosis evidence ( | ); not a memory-reset study.Localized dosing and demonstration of durable benefit. | Anemia, healthy-cell clearance and potential immune disturbance. | Primarily peripheral lesion/phagocyte targeting; central memory not directly targeted. |
| Nanoparticle delivery: delivery platform, not intrinsically a memory treatment; application to memory reset hypothetical ( | ).Peritoneal retention, uptake, formulation stability, manufacturing. | Off-target uptake and innate immune activation. | Designed for peripheral targeting; marrow spillover requires assessment. |
| TGF-β/SMAD3 inhibition: mouse endometriosis antifibrotic evidence ( | ); not memory-specific.Fibrosis readouts, sustained efficacy and local delivery. | Interference with homeostatic repair and broad TGF-β functions. | Peripheral tissue/lesion intervention; central training not established. |
| Apoptotic cells/PS liposomes: tolerogenic effects described in other disease settings ( | ); not tested for endometriosis memory.Cell-product standardization, dosing and functional rechallenge validation. | Generalized immune suppression and altered clearance. | Likely peripheral if delivered locally; systemic distribution possible. |
| β-glucan: repeated exposure can induce training rather than tolerance ( | ); no evidence of tolerance induction in endometriosis.Do not advance as tolerizing therapy without dose-specific rechallenge evidence. | Potential inflammation amplification. | Both compartments potentially exposed; central vs. peripheral effect depends on route. |
Emerging therapeutic strategies targeting macrophages in endometriosis.
The four columns summarize the strength of supporting evidence, challenges for clinical translation, potential off-target effects, and the anticipated central or peripheral effects of each therapeutic strategy.
6.2 Tolerance induction through controlled innate stimulation
Systemic LPS is too toxic for clinical use, but several alternative approaches to modulate macrophage responsiveness remain experimental. Monophosphoryl lipid A (MPL) is a detoxified LPS derivative used as a vaccine adjuvant. It retains TLR4 agonism without the inflammatory toxicity of LPS (
). Whether MPL can produce the LPSlow macrophage phenotype in endometriosis is unknown; it has not been tested in this setting. β-glucan at standard doses induces training through dectin-1→Syk→mTOR→HIF-1α ( , ). Contrary to a common assumption, continuous β-glucan exposure induces trained immunity rather than tolerance in macrophages ( ), so β-glucan is not an established tolerizing agent, and any repurposing would require evidence of a tolerance-inducing dose that does not currently exist. Infusion of apoptotic cells or phosphatidylserine-expressing liposomes pushes macrophages toward a tolerogenic phenotype via TAM receptors (Tyro3, Axl, MerTK) and downstream TGF-β and IL-10 ( ). This tolerogenic strategy has shown promise in transplantation and inflammatory-disease models but has not been tested in endometriosis.6.3 Metabolic reprogramming strategies
Because metabolic rewiring is central to trained immunity (
– , ), metabolic drugs offer a direct route to macrophage reprogramming. Rapamycin inhibits mTORC1, blocking the glycolytic switch that underlies training ( ). mTOR is hyperactivated in ectopic lesions, where it drives proliferation, angiogenesis, and macrophage recruitment ( ). Metformin activates AMPK, antagonizes mTOR, and pushes metabolism toward OXPHOS; it inhibits trained immunity in vitro and in mice ( ). Some observational and small clinical studies have reported reduced pain or lesion size with metformin in endometriosis, but the epidemiological evidence is weak and is confounded by the fact that metformin is most often prescribed for polycystic ovary syndrome, a condition that itself frequently co-occurs with endometriosis; a causal, disease-modifying effect of metformin on endometriosis has not been established. Metformin has many targets—it also suppresses endometrial stromal cell proliferation—but whether metabolic reprogramming of macrophages contributes to any benefit in patients is unproven. Itaconate and its derivatives (4-octyl itaconate, dimethyl itaconate) inhibit SDH, activate Nrf2, and suppress glycolysis ( , ). They work in sepsis, autoimmunity, and ischemia-reperfusion models ( , ), but have not been tested in endometriosis. Key challenges for translation include potential off-target effects on non-myeloid cells and the difficulty of achieving sustained therapeutic concentrations in the peritoneal cavity. Genetic or pharmacological HIF-1α inhibition blocks trained immunity induction ( ). HIF-1α inhibitors are in cancer trials and could be relevant to endometriosis, but have not been evaluated in endometriosis models.As a methodological parallel rather than disease-specific evidence, an ovarian cancer investigation combined gene-expression analysis, experimental validation, and candidate drug screening to identify targets related to cisplatin resistance (
). Such target-discovery workflows could inform future research designs, but the reported ovarian cancer targets do not demonstrate macrophage-memory reprogramming in endometriosis.6.4 Epigenetic reprogramming approaches
Epigenetic drugs may be able to reverse the chromatin marks that maintain macrophage memory. Histone deacetylase (HDAC) inhibitors—vorinostat, romidepsin, panobinostat—shift histone acetylation and could reverse the hyperacetylated enhancer landscape of trained macrophages (
, ). HDAC inhibitors have anti-fibrotic and anti-inflammatory effects preclinically; several are FDA-approved for hematological cancers ( ). Bromodomain and extraterminal domain (BET) proteins (BRD2–4, BRDT) read acetylated histones and recruit transcriptional machinery to active enhancers ( ). JQ1 and related compounds suppress key inflammatory genes and are effective in inflammatory disease models ( ). Disrupting enhancer-promoter interactions that maintain the trained transcriptome could, in principle, erase pathological macrophage memory. These agents directly target the epigenetic machinery of innate memory, but none has been tested specifically for reprogramming trained macrophages in endometriosis, and their broad epigenomic effects raise safety concerns for a chronic, non-malignant condition. Low-dose DNA methyltransferase (DNMT) inhibitors (decitabine, azacitidine) could reset DNA methylation in trained HSPCs ( , ), though the risk of broad epigenomic disruption makes this approach less attractive for a non-malignant condition.6.5 CD47-SIRPα checkpoint blockade
Magrolimab, an anti-CD47 antibody, improves macrophage-mediated tumor clearance in early-phase cancer trials (
). CD47 blockade restores phagocytosis of endometrial stromal cells in vitro and ex vivo ( ). CD47-SIRPα blockade is best classified as a macrophage-directed phagocytic-restoration strategy rather than a memory-reprogramming strategy: it does not by itself reset the epigenetic or metabolic memory state, though it may indirectly reshape the macrophage population. The challenges for endometriosis include: targeted peritoneal delivery to avoid anemia from systemic erythrocyte clearance; the risk that broad CD47 blockade promotes phagocytosis of healthy CD47-expressing cells and exacerbates autoimmune phenomena, as observed in some cancer trials; potentially pairing with a pro-phagocytic signal like calreticulin exposure; and ensuring that reactivated macrophages do not flip into a hyper-inflammatory, tissue-destructive mode.6.6 Nanoparticle-based macrophage reprogramming
Nanoparticle delivery is a platform for macrophage-directed therapy rather than a memory-reprogramming strategy in its own right. Engineered nanobiologics can already bias myeloid cell function in vivo: trained-immunity-promoting nanobiologics modulate macrophage metabolism and augment checkpoint inhibition in cancer (
), demonstrating that nanoparticle cargo can be delivered to myeloid cells to alter their functional state. In principle, nanoparticles could deliver tolerizing cargo selectively to peritoneal macrophages: lipid nanoparticles carrying IL-10 mRNA could mimic the tolerized secretome without LPS exposure, and poly(lactic-co-glycolic acid) (PLGA) nanoparticles loaded with itaconate derivatives could provide sustained local release. Macrophage-targeted particles decorated with mannose or dectin-1 ligands could exploit the high endocytic capacity of lesion-associated macrophages. These concepts remain hypothetical in endometriosis: no nanoparticle formulation has yet been shown to reprogram macrophage memory in an endometriosis model, and biodistribution, off-target uptake, and immunogenicity remain unresolved.6.7 β-glucan dosing
β-glucan illustrates the dosing problem. Standard immunostimulatory doses drive pro-inflammatory training through dectin-1→Syk→mTOR→HIF-1α and are being exploited in oncology, where trained macrophages enhance checkpoint inhibitor responses (Table 4 and illustrated in Figure 3, distinguishing strategies targeting innate immune memory-associated machinery from broader macrophage-directed and lesion-targeted interventions.
, , – ). In a single adoptive-transfer mouse model, BCG-trained macrophages enhanced lesion growth ( ); whether endogenous training has similar effects in human disease is unknown. The premise that a lower or continuous β-glucan dose would instead induce tolerance is not supported by current evidence: continuous exposure to non-soluble β-glucans induces trained immunity rather than tolerance in macrophages ( ). Whether any dose or schedule of β-glucan can reproducibly induce tolerance in the peritoneal cavity in vivo remains to be demonstrated, and a β-glucan-based regimen in endometriosis carries the risk of inadvertently inducing pro-inflammatory training. Until this is resolved, β-glucan is better regarded as a cautionary example of context-dependent dosing than as a ready-to-test tolerizing agent. If macrophage-memory biomarkers are developed—such as the H3K4me3:H3K27me3 ratio at IL10 versus TNF, the lactate:pyruvate ratio as a glycolysis proxy, or the itaconate:succinate balance—they could guide dosing in future studies. The proposed therapeutic approaches are summarized inFigure 3
7 Integrating trained immunity into endometriosis treatment
7.1 Limitations of current therapies
First-line therapies for endometriosis—combined oral contraceptives, progestins, GnRH agonists, aromatase inhibitors—are all hormonal. They suppress estrogen-driven lesion growth, and they work for symptom management in many patients (
, ). They are also contraceptive, cause hypoestrogenic side effects (bone loss, vasomotor symptoms), and leave the underlying immune dysfunction untouched ( , ). About 30% of patients do not respond ( ). Surgery provides temporary relief, but five-year recurrence rates are 40–50% ( , ). Non-hormonal, disease-modifying treatments are needed.7.2 An immune memory-directed framework
The framework described below is hypothetical and is offered as a conceptual scaffold, not a validated clinical pathway. A treatment strategy built around macrophage memory could work in stages. First, diagnostics. Peritoneal fluid biomarkers—cytokine profiles, metabolite ratios—or single-cell epigenomic analysis could classify a patient’s macrophages along the training-versus-tolerance axis and match them to the right intervention. Such assays do not yet exist, and obtaining peritoneal fluid requires invasive sampling (usually laparoscopy), which limits practical applicability; any biomarker-based classification would also need prospective validation before it could guide treatment, and classifying patients for therapy raises ethical and practical questions about who should be treated and on what evidence. Second, tolerance induction. For patients whose macrophages show a pro-inflammatory memory (severe pain, early disease, GP infection history), metformin, itaconate derivatives, low-dose β-glucan, or MPL could nudge macrophages toward the IL-10-producing, lesion-suppressing state —assuming such agents can be shown to induce tolerance in the peritoneal cavity, which has not yet been demonstrated. Third, phagocytic restoration. For patients with established lesions and impaired clearance, CD47-SIRPα blockade—ideally delivered locally to the peritoneum—could restore phagocytosis and promote regression. Fourth, anti-fibrotic reprogramming. For deep infiltrating endometriosis or extensive adhesions, TGFB1/SMAD3 blockade (to stop MMT) combined with macrophage reprogramming could reduce fibrosis and restore immune surveillance. Fifth, maintenance. Trained immunity lasts months to years (
, ). Intermittent dosing or continuous low-dose metabolic modulators (metformin) may be needed to hold the tolerized state. Peritoneal fluid monitoring or circulating monocyte epigenetics could guide duration. Each of these stages depends on resolving the open questions about memory durability and reversibility identified in the Limitations section below.7.3 Limitations of current evidence
This narrative review distinguishes direct disease-specific intervention evidence from associations, cross-disease mechanistic analogies and proposed therapies. The following appraisal specifies what each evidence stream does and does not establish.
Evidence appraisal—causal lesion growth (direct but limited): One mouse adoptive-transfer study of macrophages trained or tolerized in vitro reported opposing effects on induced lesion growth (
). This establishes experimental sufficiency in that setting, not endogenous memory, necessity, or human causality.Evidence appraisal—endogenous and central memory (not demonstrated): Human peritoneal macrophages have not been shown to retain training or tolerance after washout and rechallenge. HSPC-mediated central training has not been established in endometriosis; extrapolation from other inflammatory diseases remains indirect (
, , – ).Evidence appraisal—duration and reversibility (unknown): No longitudinal human study has defined the persistence, stimulus independence, or reversibility of endometriosis-associated macrophage memory. Treatment duration and the likely benefit of local versus systemic intervention cannot be predicted until these properties are measured.
Evidence appraisal—clinical associations (indirect): Retrospective, self-reported infection histories in the Jeljeli study are vulnerable to confounding and do not establish microbial ligand delivery to peritoneal macrophages or causal memory imprinting (
).Evidence appraisal—fibrosis, pain, infertility and cell subsets (hypothesis-generating): Fibrosis-related signaling, nociceptive crosstalk and reproductive dysfunction have supporting disease biology, but their causal dependence on trained or tolerized macrophage states has not been shown. Mouse scRNA-seq populations and human cross-species comparisons are descriptive and cannot replace functional memory assays (
– , ).Evidence appraisal—therapeutic reprogramming (largely indirect or hypothetical): Metabolic and epigenetic agents have mechanistic support from other experimental systems, whereas CD47 blockade and TGF-β/SMAD3 inhibition may alter lesion biology without necessarily resetting innate immune memory. No macrophage-specific, durable memory-resetting intervention has been validated clinically for endometriosis (
, , ).7.4 Research priorities
A priority is to establish paired human peritoneal macrophage samples and prospective functional restimulation after defined washout periods, with longitudinal follow-up to distinguish persistent memory from continuous local stimulation. Only after durability and reversibility have been demonstrated should treatment intervals, local delivery, or systemic HSPC-directed interventions be evaluated.
Several questions need answers before these ideas can reach the clinic. We do not know whether central trained immunity—HSPC reprogramming—occurs in endometriosis, and if it does, whether it explains the systemic immune changes reported in patients. We do not know whether the macrophage memory states in endometriosis are heritable through cell division; in other contexts, training persists for weeks to months (
), and the durability matters for dosing. We do not know whether a tolerized state can be maintained without compromising host defense in the peritoneal cavity. We do not know which genetic variants (in TLR4, NOD2, DECTIN1, IRG1) affect individual training responses and could serve as pharmacogenomic markers. And we do not know how trained or tolerized macrophages interact with T cells, NK cells, neutrophils, and mast cells in the lesion microenvironment—a gap that matters for predicting the net effect of any reprogramming strategy. Addressing these questions requires, at minimum, a functional assay for innate immune memory in human peritoneal macrophages and a second, independent disease model to corroborate the adoptive-transfer findings.8 Conclusion
The trained immunity field has shown that macrophages carry a functional memory, written into their epigenome and metabolome, of the inflammatory challenges they have faced. Whether and how this memory contributes to endometriosis is an emerging and still unsettled question. Direct evidence from a single mouse study shows that adoptively transferred trained macrophages accelerate, and tolerized macrophages suppress, lesion growth; the extension of this framework to fibrosis, pain, and fertility in human disease remains inferential. The endometriotic microenvironment—the estrogen, the DAMPs, the extracellular vesicles, the microbial signals—provides cues that could, in principle, imprint these states. Recent synthesis of immune changes in the eutopic endometrium further highlights tissue-specific heterogeneity, although it does not establish trained immunity in patients with endometriosis (
). Candidate approaches should therefore be described as research priorities rather than clinically actionable therapies; endometriosis-specific proof of mechanism and an acceptable safety profile remain prerequisites for translational development.If macrophage memory does contribute to lesion persistence, then reprogramming that memory is a plausible disease-modifying strategy that could work alongside or instead of hormonal suppression. However, this field is at an early, largely preclinical stage. Most of the candidate agents discussed here have not been tested in endometriosis models; their optimal dosing, timing, and delivery route are unknown; and the risk-benefit profile for a chronic, non-malignant condition differs substantially from that of cancer. The work ahead is to establish whether trained or tolerized macrophages exist in human lesions, to develop assays that can detect such memory states, and only then to translate the mechanistic insights of trained immunity biology into treatments that are safe, effective, and durable enough for the millions of women who live with this disease.
Statements
Author contributions
XG: Validation, Conceptualization, Investigation, Project administration, Visualization, Writing – original draft, Methodology, Data curation, Formal analysis, Software. LR: Conceptualization, Methodology, Investigation, Software, Project administration, Formal analysis, Writing – original draft, Visualization, Data curation. HH: Methodology, Data curation, Project administration, Conceptualization, Investigation, Formal analysis, Writing – original draft. YD: Supervision, Project administration, Conceptualization, Writing – review & editing, Methodology, Data curation, Formal analysis. TZ: Writing – review & editing, Methodology, Conceptualization, Supervision, Resources, Data curation, Visualization, Formal analysis, Project administration, Validation.
Funding
The author(s) declared that financial support was not received for this work and/or its publication.
Acknowledgments
The authors thank the investigators whose foundational work on trained immunity, macrophage biology, and endometriosis made this review possible.
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.
Abbreviations
ATAC-seq, assay for transposase-accessible chromatin using sequencing; α-KG, α-ketoglutarate; BCG, bacille Calmette-Guérin; BET, bromodomain and extraterminal domain; DAMP, damage-associated molecular pattern; DIE, deep infiltrating endometriosis; DNMT, DNA methyltransferase; E2, estradiol; ECM, extracellular matrix; EMT, epithelial-mesenchymal transition; ER, estrogen receptor; GN, Gram-negative; GP, Gram-positive; GWAS, genome-wide association study; HAT, histone acetyltransferase; HDAC, histone deacetylase; HIF-1α, hypoxia-inducible factor 1-alpha; HSC, hematopoietic stem cell; HSPC, hematopoietic stem and progenitor cell; IRG1, immune-responsive gene 1 (also known as ACOD1); JNK, c-Jun N-terminal kinase; LPS, lipopolysaccharide; MMT, macrophage-myofibroblast transition; MPL, monophosphoryl lipid A; mTOR, mechanistic target of rapamycin; NK, natural killer; Nrf2, nuclear factor erythroid 2-related factor 2; OXPHOS, oxidative phosphorylation; PLGA, poly(lactic-co-glycolic acid); scRNA-seq, single-cell RNA sequencing; sEV, small extracellular vesicle; SIRPα, signal regulatory protein alpha; SOCS3, suppressor of cytokine signaling 3; Syk, spleen tyrosine kinase; TAM (cell), tumor-associated macrophage; TAM receptors, Tyro3/Axl/MerTK receptor family (distinct from tumor-associated macrophages); TCA, tricarboxylic acid; TGF-β, transforming growth factor beta; TLR, Toll-like receptor; TNF, tumor necrosis factor; Treg, regulatory T cell.