Trained immunity is a form of functional memory in the innate immune system in which an initial encounter with a microorganism, microbial product, or inflammatory stimulus changes how cells respond to later challenges. These changes can persist after the initiating stimulus has disappeared and can affect responses to unrelated agents. They are maintained through alterations in cellular metabolism, gene regulation, and sometimes the production of new immune cells, rather than through the rearranged antigen receptors characteristic of adaptive immunity.
The concept broadens the traditional understanding of immunological memory, once associated almost exclusively with lymphocytes. It helps explain how short-lived innate cells can participate in long-lasting changes in host defense and why previous infections or environmental exposures may influence subsequent inflammatory responses. Training is not necessarily beneficial: the same mechanisms that improve antimicrobial defense can also sustain damaging inflammation.
Research on trained immunity connects molecular studies of chromatin and metabolism with investigations of vaccination, blood-cell development, tissue immunity, and chronic disease. A central challenge is distinguishing genuine memory from continuing stimulation, changes in cell composition, and conventional adaptive immune responses.
Conceptual boundaries and historical development
Classical descriptions of immunity divided host defense into a rapid, broadly responsive innate component and a slower adaptive component capable of highly specific memory. In this framework, T cells and B cells retain information about earlier encounters through the selection and persistence of antigen-specific clones. Their receptors acquire diversity through processes including V(D)J recombination. Trained immunity does not require this receptor-generating machinery. Instead, it changes the readiness of cells to activate particular defense programs.
Evidence challenging a strict separation between innate defense and memory accumulated from several directions. Plants and invertebrates showed enhanced resistance after previous exposure despite lacking the vertebrate adaptive immune system. Experimental studies also found that earlier microbial exposures could improve resistance to unrelated infections. In 2011, Mihai Netea and colleagues formulated trained immunity as a framework linking such observations to persistent functional changes in innate immune cells. Subsequent work supplied direct evidence for cellular and molecular mechanisms.
Two studies published in 2012 became important reference points. Jessica Quintin and colleagues showed that exposure to Candida albicans or fungal β-glucan could establish enhanced responses in experimental systems, with a prominent role for monocytes and epigenetic regulation. Johanneke Kleinnijenhuis and colleagues reported that vaccination with bacillus Calmette–Guérin, or BCG, increased human monocyte responses to unrelated microbial stimuli and was associated with changes in histone methylation. These findings helped connect older observations of nonspecific protection with experimentally measurable innate-cell states.
Operationally, training requires a temporally separated sequence: an initial stimulus, an interval during which the acute response subsides, and a later challenge. This distinguishes it from immediate activation and from priming, in which a second signal arrives while the first response remains active. The boundaries are not absolute, however, because immune states change continuously and residual stimulation can be difficult to exclude.
Trained immunity is also distinguished from immune tolerance, especially the reduced inflammatory response that can follow repeated exposure to bacterial endotoxin. Training usually denotes enhanced responsiveness, whereas tolerance denotes diminished responsiveness in particular pathways. Both involve lasting reprogramming, and both can coexist within one cell: inflammatory cytokine production may decline while antimicrobial functions remain preserved. Consequently, “memory” in innate immunity describes altered response potential, not simply a universal increase in immune activity.
Inducing stimuli and recognition pathways
Training begins when an exposure engages receptors and signaling pathways capable of leaving a durable cellular imprint. Important experimental stimuli include fungal β-glucans, BCG, certain microbial components, and endogenous products associated with tissue damage or metabolic disturbance. Their effects depend on concentration, duration, route of exposure, and the prior state of the responding cell. The same class of stimulus can produce training, tolerance, or cell injury under different conditions.
Fungal β-glucans are polysaccharides found in the cell walls of many fungi. Their recognition by Dectin-1, a C-type lectin receptor, activates signaling involving spleen tyrosine kinase and downstream inflammatory pathways. In widely used human monocyte experiments, a brief β-glucan exposure is followed by washing and several days of rest. On later challenge, the cells may produce more tumor necrosis factor and interleukin-6 than untrained controls. The precise response depends on β-glucan structure, preparation, and experimental conditions.
BCG is a live attenuated strain of Mycobacterium bovis used as a vaccine against tuberculosis. It activates several recognition systems rather than a single training pathway. One important component is NOD2, an intracellular receptor that recognizes muramyl dipeptide derived from bacterial peptidoglycan. Studies of human cells have linked NOD2 signaling to BCG-induced reprogramming. Because BCG contains many biologically active components and can persist after administration, its effects are more complex than those of a purified receptor ligand.
Nonmicrobial training has attracted particular interest in chronic inflammatory disease. Oxidatively modified low-density lipoprotein can induce heightened inflammatory responses in cultured monocytes, together with changes in lipid handling and gene regulation. Experimental work with crystals and other sterile inflammatory stimuli likewise indicates that innate memory need not originate in infection. These observations provide a possible connection between repeated metabolic injury and persistent inflammatory behavior, although they do not establish that every such exposure causes durable training in humans.
Signals released by other cells can also transmit or amplify training. Interleukin-1 is a prominent example of a cytokine that links local stimulation to broader changes in immune-cell development and responsiveness. Interferons and other mediators contribute in stimulus-specific settings. Thus, the initiating event may be direct recognition by the cell that later remembers, indirect exposure to an inflammatory environment, or a combination of both. Identifying the initiating receptor alone is therefore insufficient to explain the eventual trained state.
Chromatin remodeling and the persistence of altered responses
A central feature of trained immunity is the modification of chromatin, the complex of DNA and associated proteins that regulates access to genetic information. Innate immune cells generally retain the same underlying DNA sequence after training, but particular genes become easier or harder to activate. This is an instance of epigenetic regulation: a persistent change in cellular state that does not require alteration of the nucleotide sequence.
Much of the evidence concerns chemical modifications of histones, the proteins around which DNA is packaged. Training-associated changes include increased trimethylation of histone H3 at lysine 4, abbreviated H3K4me3, at some promoters, and changes in H3K4me1 and H3K27ac at regulatory regions. These marks are associated with active or activation-ready chromatin in many settings. Their significance depends on location and context; no single histone modification is a universal molecular signature of trained immunity.
Regulatory regions called enhancers are especially important. Some are already accessible before stimulation, whereas others acquire accessibility during the initial response. After stimulation ends, a regulatory region may retain features that allow faster or stronger reactivation. Changes in transcription-factor recruitment and the organization of regulatory complexes can thereby alter later gene expression. Studies published by Sadia Saeed and colleagues in 2014 helped map how human monocyte differentiation, training, and tolerance involve distinct epigenomic programs.
This persistence does not mean that inflammatory genes remain continuously active. A trained cell may return close to its original cytokine-production baseline while retaining an altered capacity to respond. The distinction between baseline activity and response potential is essential: a measurement made before rechallenge may miss functional memory, whereas persistent inflammation without a rest interval may resemble training without demonstrating it.
Chromatin associations must also be separated from causal mechanisms. An activating histone mark may be a consequence of transcription rather than the original carrier of memory. Stronger evidence comes from manipulating the enzymes that deposit or remove marks, disrupting relevant regulatory elements, or following changes over time. Such experiments remain difficult because chromatin regulators often have broad effects on cell survival and differentiation.
Additional layers include DNA methylation, noncoding RNA, and higher-order genome organization. Their contributions vary across cell types and stimuli. Epigenetic memory is therefore best understood as a distributed regulatory state rather than a single molecular switch. In dividing precursor cells, some components of this state can influence descendants; in long-lived tissue cells, persistent changes can remain within the originally exposed population.
Metabolic reprogramming and its connection to gene regulation
Training changes how immune cells acquire and use nutrients. Metabolism is not merely a source of fuel for an otherwise independent immune program: metabolic intermediates also regulate signaling and chromatin-modifying enzymes. This creates a mechanistic connection between the conditions experienced during an initial exposure and the cell’s later response capacity.
One influential finding was that β-glucan training promotes increased glycolysis, including under conditions where oxygen remains available. Work published by Shih-Chin Cheng and colleagues in 2014 linked this shift to the Akt–mTOR–HIF-1α signaling axis. Glycolysis supplies energy rapidly and provides intermediates for biosynthesis. Inhibiting relevant pathways during the initial training phase can reduce subsequent enhancement of cytokine responses, supporting a functional contribution rather than a purely accompanying metabolic change.
The metabolic program is nevertheless stimulus dependent. BCG-induced training has been associated with changes in both glycolysis and mitochondrial oxidative phosphorylation. It is therefore misleading to describe trained cells simply as abandoning mitochondrial respiration. Different trained states can combine greater glucose utilization, altered mitochondrial activity, and changes in amino-acid or lipid metabolism.
The citric acid cycle provides important links to epigenetic regulation. In some β-glucan models, glutamine metabolism contributes to accumulation of fumarate. Fumarate can inhibit particular α-ketoglutarate-dependent enzymes, including histone demethylases, thereby influencing the persistence of activating chromatin marks. This illustrates how an altered metabolite concentration can change the activity of an enzyme controlling gene accessibility.
Lipid metabolism supplies another connection. The mevalonate pathway, which contributes to cholesterol biosynthesis, has been implicated in experimental training. Mevalonate can affect growth-factor signaling and downstream metabolic pathways, linking an anabolic program to enhanced inflammatory responsiveness. Such results do not imply that blood cholesterol concentration is a direct measure of trained immunity: circulating lipid levels, intracellular pathway activity, and immune-cell memory are distinct variables.
Metabolic interventions require careful interpretation. Blocking glycolysis or mitochondrial function can impair viability, proliferation, or the immediate response to the second stimulus, independently of memory formation. Experiments must therefore distinguish effects on induction, maintenance, and recall. Measurements of metabolite abundance also need to be supplemented by flux analysis, because a large intracellular pool does not necessarily indicate rapid pathway activity. Together, these considerations explain why trained immunity is described as immunometabolic reprogramming rather than as one fixed metabolic phenotype.
Bone-marrow progenitors and tissue-resident memory
The lifespan of circulating monocytes presents an apparent problem for long-lasting innate memory. Many monocytes remain in the bloodstream for only days, yet functional changes following some exposures can be detected months later. One explanation is that training affects hematopoietic stem cells and progenitors in the bone marrow, altering the cells they subsequently produce. This is often called central trained immunity, in contrast to reprogramming of mature cells in peripheral tissues.
In 2018, studies led by Ioannis Mitroulis and Eva Kaufmann demonstrated complementary aspects of this process in mice. β-Glucan exposure altered progenitor activity and myeloid-cell production, while BCG exposure reprogrammed hematopoietic stem cells in ways associated with altered macrophage function and protection against experimental mycobacterial infection. These experiments showed that innate memory could involve changes in the production line of immune cells, not just persistence within terminally differentiated cells.
Signals such as interleukin-1 and interferon-γ can influence these progenitor responses in particular experimental systems. Changes may include increased progenitor proliferation, altered metabolism, modified chromatin accessibility, and a shift toward myeloid differentiation. Such effects are not identical to indefinite expansion of stem cells. Sustained inflammatory stimulation can also impair stem-cell function, and the consequences depend on intensity and duration.
Human evidence has supported the relevance of this mechanism while remaining more limited than controlled animal experimentation. A 2020 study by Branko Cirovic and colleagues examined bone-marrow and peripheral-blood responses after BCG vaccination and found changes consistent with reprogramming of hematopoietic progenitors and their descendants. Human studies cannot routinely use the extensive lineage tracing or transplantation procedures available in mice, so they rely on repeated sampling and integrated molecular analyses.
Long-lived tissue-resident cells provide a second route to persistence. Many macrophage populations can maintain themselves locally, allowing an earlier exposure to affect the same cellular population for extended periods. In the brain, microglia can develop altered responses after systemic or local inflammatory events. Mouse studies have demonstrated both training-like enhancement and tolerance-like suppression, depending on the exposure schedule.
The distinction between central and peripheral training is useful but incomplete. Bone-marrow-derived cells enter tissues, tissue inflammation sends signals back to marrow, and local environments can reinforce or overwrite earlier programs. A persistent tissue response may therefore reflect long-lived resident cells, altered recruitment, continuing environmental signals, or several mechanisms together. Establishing which mechanism dominates requires tracking both cellular ancestry and functional state.
Innate memory across cell types and organisms
Monocytes and macrophages are the best-studied examples of trained immunity, but they are not the only cells capable of memory-like behavior. Natural killer cells provide a particularly important comparison. Certain NK-cell populations expand and persist after infection, and cytokine exposure can generate populations with enhanced later responses. Their biology demonstrates that innate lymphocytes can retain information about previous stimulation, although the mechanisms do not always match those of myeloid training.
In mice, cytomegalovirus infection drives expansion of NK cells expressing the activating receptor Ly49H, which recognizes a viral protein. In humans, cytomegalovirus exposure is associated with distinctive NK-cell populations, often including cells expressing NKG2C and showing altered signaling and epigenetic characteristics. Some NK memory is therefore more receptor-selective than the broad heterologous responsiveness usually emphasized in trained monocytes. Cytokine-induced memory-like NK cells, generated experimentally with interleukin-12, interleukin-15, and interleukin-18, illustrate a different route based on prior inflammatory stimulation.
Neutrophils can also show altered function after training stimuli. Their short lifespan makes changes in bone-marrow production and maturation especially relevant. Reports of trained granulocyte responses must distinguish reprogramming of individual mature cells from the release of newly generated cells with different properties. Similar questions arise for dendritic cells, whose altered antigen presentation can influence subsequent adaptive responses as well as innate defense.
Memory-like responses also occur outside professional immune cells. Experimental work on epithelial stem cells, including skin stem cells, has shown that earlier inflammation can alter chromatin and accelerate responses to later tissue injury. Such findings suggest that tissue defense can retain a record of prior damage. However, epithelial inflammatory memory is not automatically equivalent to monocyte training: the cell types, functional outputs, and duration of persistence differ.
Comparative biology provides a broader context. Plants can establish systemic acquired resistance after local infection, while several invertebrates exhibit immune priming that changes survival or pathogen control after subsequent exposure. These organisms lack vertebrate-style B- and T-cell immunity, making them important evidence that useful immune memory does not require rearranged antigen receptors.
The similarities should not be mistaken for proof of one conserved mechanism. Plant defense can involve signaling networks, metabolites, and sometimes persistent epigenetic changes that differ substantially from mammalian pathways. Invertebrate results vary by species and pathogen, and some apparent memory may reflect persistent infection or transferred protective factors. “Innate immune memory” is therefore a broad comparative category, within which trained immunity describes a more specific set of experimentally characterized forms of cellular reprogramming.
Experimental models and standards of evidence
A common laboratory model begins with human monocytes isolated from blood. Cells receive a training stimulus for approximately 24 hours, are washed, and then rest for several days before challenge with a second agent. Depending on the protocol, rechallenge often occurs around day six or seven. Enhanced production of cytokines such as tumor necrosis factor or interleukin-6 is compared with identically cultured cells that did not receive the first stimulus.
This design separates induction from recall, but it introduces important complications. Monocytes differentiate during culture, and serum composition, cell density, adherence, and donor characteristics can affect the outcome. A training stimulus may change survival or differentiation, leaving a different mixture of cells at the time of rechallenge. Increased cytokine concentration could consequently reflect more viable cells rather than stronger responses per cell. Cell counting, viability measurements, phenotyping, and normalization are therefore integral to interpretation.
Functional assessment extends beyond cytokine secretion. Studies measure phagocytosis, microbial killing, reactive oxygen production, migration, and effects on tissue injury or pathogen burden. These outcomes need not change together. Strong inflammatory cytokine release is not necessarily evidence of better antimicrobial protection, and improved pathogen control does not by itself establish that training caused the effect.
Molecular methods help define the underlying state. Flow cytometry identifies cell populations and activation characteristics. RNA sequencing measures transcriptional programs, while chromatin immunoprecipitation assays map histone modifications or protein binding. ATAC-seq measures chromatin accessibility. Single-cell approaches are particularly useful for distinguishing changes within a cell type from shifts in the proportions of different populations. Metabolomics and isotope-tracing experiments assess metabolic changes and pathway flux.
Animal experiments can test mechanisms through receptor deletion, cell depletion, transplantation, and controlled pathogen challenge. Models lacking functional adaptive lymphocytes help establish that a protective effect does not require conventional T- or B-cell memory. Nevertheless, such animals have altered immune development, and results do not transfer automatically to immunologically intact humans.
Strong evidence combines temporal separation, removal or exclusion of the initiating stimulus, a reproducible functional change, and mechanistic support. The strongest causal studies manipulate a proposed memory-maintaining pathway without merely disabling the entire immune response. A major unresolved methodological issue is the absence of a universal biomarker for trained immunity. Different stimuli can produce different molecular states, so no single cytokine assay, histone mark, or blood transcriptional signature currently identifies all forms of training.
Human vaccination and clinical evidence
BCG has been central to human trained-immunity research because it is widely used and can produce effects beyond its intended mycobacterial target. Epidemiological studies and some clinical trials have reported differences in infections or mortality unrelated to tuberculosis after BCG vaccination. These observations motivated mechanistic research, but epidemiological associations alone cannot establish trained immunity: vaccine timing, underlying health, access to care, concurrent vaccines, and other factors can influence outcomes.
Human experimental studies provide a more direct bridge between vaccination and innate reprogramming. Following BCG vaccination, investigators have detected changes in monocyte cytokine responses, metabolism, and chromatin. In a 2018 study led by Rob Arts, healthy volunteers received BCG before an attenuated yellow-fever vaccine challenge. BCG vaccination was associated with lower subsequent vaccine-virus viremia, and interleukin-1-related responses were linked to the observed protection. The controlled challenge supported biological activity but did not establish broad protection against natural viral infections.
Clinical evaluation requires a further step: demonstrating benefit in the population and disease setting of interest. A randomized controlled trial can test whether vaccination changes a prespecified clinical outcome, while associated laboratory studies examine whether trained immunity mediates the effect. Even when an intervention changes an innate immune assay, that change may not be responsible for its clinical outcome. Conversely, a clinical benefit may involve conventional adaptive immunity, changes in pathogen exposure, or other mechanisms.
The COVID-19 pandemic provided a large-scale test of proposals to repurpose BCG for nonspecific protection. Early ecological comparisons between countries were highly vulnerable to differences in population age, epidemic timing, testing, and health systems. Subsequent randomized studies did not establish BCG as effective protection against COVID-19. In particular, the BRACE trial, reported in 2023, did not show a reduction in its principal COVID-19 outcomes among health-care workers. These findings illustrate why mechanistic plausibility and favorable immune measurements cannot substitute for clinical endpoints.
Interpretation also depends on age, prior BCG exposure, vaccine strain, route of administration, interval since vaccination, and the infectious challenge. A result in newborns cannot be assumed to apply to older adults, and an effect on one infection cannot be generalized to all pathogens. Trials of interventions intended to induce trained immunity therefore evaluate specific products and populations rather than a uniform “immune boost.”
BCG remains an established tuberculosis vaccine, while its proposed broader uses constitute separate questions. Trained immunity supplies a framework for investigating those questions; it does not itself demonstrate the efficacy or safety of a new vaccination indication.
Chronic disease, therapeutic investigation, and unresolved questions
Persistent innate reprogramming offers a possible mechanism linking earlier exposures to chronic inflammatory disease. In atherosclerosis, modified lipoproteins and metabolic inflammation can influence monocyte function and blood-cell production. Experimental studies show that such stimuli can induce training-like states, suggesting a route by which transient metabolic disturbances might have effects beyond their immediate duration. Establishing the contribution of this mechanism to human cardiovascular events remains more difficult than demonstrating it in cultured cells or animals.
A notable mouse study published in 2018 by Anette Christ and colleagues found that a Western-type diet produced changes in innate immunity and myeloid progenitors that persisted after the animals returned to a standard diet. The work implicated inflammatory signaling, including the NLRP3 inflammasome, in this response. Because the experiments used a particular genetically susceptible mouse model, they establish a mechanism under defined conditions rather than a direct estimate of risk or persistence in humans.
Related questions arise in obesity, diabetes, recurrent inflammatory disorders, and tissue injury. Studies have identified altered monocyte responses and epigenetic or metabolic features in several disease settings. These findings can be compatible with training, but chronic disease often supplies continuous stimulation. Distinguishing a retained memory of a previous exposure from an ongoing response to current pathology is therefore essential. Similarity to an experimentally trained state does not by itself establish how that state originated.
In cancer, trained immunity is being investigated as a way to modify myeloid-cell behavior and strengthen antitumor responses. Intravesical BCG is an established treatment for selected non-muscle-invasive bladder cancers, where it provokes a complex local immune response. Innate reprogramming may contribute, but it is not the sole established explanation for treatment activity; adaptive responses and interactions within the tumor environment also matter. Experimental strategies involving microbial products, particles, or targeted delivery remain distinct from proven clinical treatments.
Therapeutic investigation has two opposing aims: inducing protective training and suppressing maladaptive memory. Candidate targets include inflammatory signaling pathways, metabolic enzymes, and chromatin regulators. Their broad biological roles make selectivity a major problem. A treatment that reduces harmful inflammatory persistence might also compromise antimicrobial defense or normal hematopoiesis, while a training-inducing intervention might amplify tissue damage in an unsuitable setting.
Several basic questions remain open. The durability and reversibility of different trained states are incompletely mapped, especially in humans. The relative contributions of stem cells, long-lived tissue cells, and repeated environmental stimulation vary across contexts. Sex, age, genetic variation, microbiota, and previous infections may influence both induction and recall. Resolving these questions requires longitudinal studies that connect defined exposures, cell-specific molecular changes, and clinically meaningful outcomes rather than treating heightened inflammation as a sufficient measure of memory.