The epithelial barriers of the skin, airways and intestines in mammals form essential interfaces that constantly sense and respond to external and internal signals1,2,3. Adaptation to environmental exposures is a conserved property of barrier tissues, and the cell types that compose these barriers must balance metabolic functions with host defence4,5. In simpler metazoans that lack specialized immune cells, this function falls solely on epithelial cells, whereas in more cellularly complex mammalian tissues, immune cells may mediate this process5.

For a barrier tissue to optimally access information derived from a previous immune event to inform a present or future memory response, that information can be stored in locally accessible cell types that are residents or permanent residents and maintain appropriate qualitative features (Fig. 1). Deviations in memory storage or retrieval can predispose a tissue to pathological consequences: insufficient memory leads to increased infections; excessive memory retrieval drives chronic inflammation; and malignancy potentially arises from both insufficient and excessive memory6,7. The appropriately named adaptive immune system has key roles in promoting antigen-specific memory through encoding receptors — T cell receptors (TCRs) and B cell receptors (BCRs) — specific for microorganisms, environmental antigens and self-antigens in lymphocyte-bearing organisms8,9.

Fig. 1: Cell type residence and permanence in barrier tissues.
figure 1

The cell types in a tissue at any given moment may be there short term (transient cells), have the capacity to reside long term (resident cells) or be essential constituents (permanent resident cells). The fundamental permanent resident unit of a barrier tissue consists of epithelial stem cells and stromal cells (fibroblasts and endothelial cells). Other permanent resident cells include macrophages and sensory neurons. We acknowledge that there are specific cases in which cell subsets that are typically transient (such as monocyte-derived mononuclear phagocytes) can acquire the characteristics of resident cells (such as Langerhans cells or macrophages) based on environmental perturbation and niche availability. Cell types that are non-essential to the fundamental tissue unit (such as plasma cells) can also exhibit characteristics of permanent residence. Moreover, microorganisms can be permanent residents, residents or transient visitors. Abiotic stimuli, such as nutritional components, can vary in ‘residence’ based on the frequency and duration of environmental exposure. BRM cell, tissue-resident memory B cell; IEL, intraepithelial lymphocyte; ILC, innate lymphoid cell; TEM cell, effector memory T cell; TPM cell, peripheral memory T cell; Treg cell, regulatory T cell; TRM cell, tissue-resident memory T cell.

However, antigen-specific adaptive immune memory is just one of the myriad ways in which tissues and organisms can adapt during immune events10. Indeed, the properties of memory are now being identified within other haematopoietic cells, such as innate immune cells, as well as more recently in the tissue parenchyma10,11,12,13,14,15,16. Here, we use the term memory to refer to a defined response to an initiating trigger that has an altered baseline, sensitivity, rapidity or maximum, and that persists until a secondary challenge11 (Fig. 2a). Immunological, inflammatory and neuronal memory are all subject to independent environmental or host factors, including temperature, metabolism, hormones and circadian cycles, that may diminish or enhance a secondary response.

Fig. 2: Properties of tissue inflammatory memory storage.
figure 2

a | The properties of memory in response to a stimulus include increases in the baseline, sensitivity, rapidity or maximum. b | The discrete components of tissue inflammatory memory storage include specificity, quantity, quality (that is, cell state), durability and distribution among various cell types. Cooperativity may exist across similar or distinct cell types, with each cell type being defined through its own discrete components. Part a adapted with permission from ref.11, Elsevier.

The division of the adaptive immune system into cell types (such as αβ and γδ T cells), cell subsets (such as CD4+ and CD8+ T cells) and cell state descriptors (such as type 1, type 2, type 17, circulating and resident-memory cells) has been invaluable in providing cellular mechanisms for the observed phenomena of tissue immunity17,18. Comprehensively identifying the sets of gene modules that define the types, subsets and states of the main parenchymal, stromal, neuronal and immune cell lineages in barrier tissues will afford an unprecedented view of tissue immunity19. To date, this has been technically challenging given the lack of discrete markers for the prospective isolation of epithelial cells, stromal cells and some immune cells by subset and state, limiting our understanding to broad gene expression patterns within these subsets. However, recent advances in large-scale single-cell RNA sequencing (scRNA-seq) and epigenetic profiling are now enabling inquiries into the properties and adaptations of discrete cell subsets, as well as the relationships between them20,21,22 (Box 1).

In this Review, we focus on the distribution and storage of inflammatory memory across the cell types and subsets that form barrier tissues, highlighting recent mechanistic insights, the potential functional consequences of encoding memory and how it may be reinforced. We propose a framework to guide future experiments aimed at understanding cooperative storage of memory across cell lineages and its impact on tissue adaptation and maladaptation. Elucidating the mechanisms of collective inflammatory memory may eventually allow for new approaches to programme and reprogramme memory for infectious and inflammatory disease (Box 2).

Components of inflammatory memory

We use the term inflammatory memory to describe the broad responses that encompass protective immunological memory and account for protective, neutral and deleterious secondary responses, regardless of cell type. Our working definition expands on the latest thinking in the field, including the definitions of immunological memory and distinctions between adaptation and memory provided in two recent perspective articles10,13.

We formalize tissue inflammatory memory storage into five measurable discrete components that are connected through cooperativity (Fig. 2b). The first component of tissue inflammatory memory is specificity, which refers to the recognition of an initiating stimulus, and can range from unique receptor–ligand recognition, or recognition of context-specific cues, to complete promiscuity. The second component is quantity, which generally refers to an increase in the frequency of responding cells. The third is quality, which describes the polarization of responding cells towards a specified cell state dictated by one to several genes or gene modules, or the activity of their products. The fourth component is durability, which is a measure of the time period of increased quality or quantity of a response owing to a combination of cellular and epigenetic factors. The fifth component describes its distribution, which encompasses the cell lineages, types and subsets that show intrinsic alterations in the first four components in a tissue. Finally, these five components are linked by a sixth, cooperativity, which refers to the factors that operate and communicate between cells to promote collective memory retention and recall.

Specificity, quantity, quality and durability

Immunological memory is present across all kingdoms of life. Bacteria use CRISPR arrays to restrict viral infections, plants use sensitization or priming strategies to defend local and distal tissues, Drosophila fruit flies deploy RNA interference and mammals rely on antigen-specific T cells and B cells12,23,24,25. Much of the recent interest in immunological memory has focused not on its functional properties but, rather, on whether the cellular and molecular mechanisms used should be classified as immunological memory within and across species25. Yet the unifying principle of these mechanisms is that prior exposure to a specific environmental agent induces a change to a measurable parameter (such as specificity) of the host response that persists, often in the absence of the initiating trigger11,13,25. Protective immunity may result from complex cell state programmes26, or even a single gene product being more robustly produced, as is the case for interferon-induced transmembrane protein 3 (IFITM3) and infection with influenza virus27.

Two of the main routes to memory storage are through increasing the number of specific cells present (that is, quantity) or altering the response characteristics of these cells (that is, quality; such as polarization towards type 1 T helper cell, type 2 T helper cell or type 17 T helper cell states)9 (Fig. 2). These shifts are related to clonal expansion and epigenetic alterations (Box 3), respectively12,24, and involve changes to how inputs are sensed in the tissue and to subsequent outputs. However, it is important to note that changes in quantity and quality are not mutually exclusive, and both are found in hallmark examples of adaptive and innate immune memory. Thinking beyond the concept that memory is stored within adaptive or innate immune cells is essential for us to more broadly consider the functional roles of inflammatory memory in tissue adaptation25,28,29,30. It is also important to consider that the time period (durability) over which memory is useful or detrimental, and whether it is an adaptation or memory, will vary based on the specific context (such as microbial or allergen exposure in daily, weekly, monthly or yearly intervals). Finally, it will be important to define whether the presence of the initiating trigger is required to maintain adaptation or whether memory persists in its absence10.

Distribution and cooperativity

Quantitatively and/or qualitatively enhanced responses may be distributed across multiple cell types in the relevant tissue for rapid inflammatory memory14,31,32. Work over the past decade has elucidated discrete lymphocyte subsets that can ‘remember’ specific immune events and persist long term in tissues18,33,34,35,36. This lymphocyte storage of memory occurs in the context of the basic mammalian barrier tissue unit of epithelial cells (the parenchyma), supported by stromal cells (fibroblast and endothelial cells), neurons and haematopoietic cells (macrophages)5,14,37 (Fig. 1).

Cell types other than lymphocytes can also remember immune events. Specific examples include how macrophages can become ‘trained’ to adapt to inflammatory challenges38, how fibroblasts and endothelial cells can be primed by inflammatory cytokines39,40,41 and, most recently, how inflammation shapes epithelial barriers through its actions on the stem cells that give rise to them15,16. But how do parenchymal, stromal, neuronal and haematopoietic cells participate in the distribution and storage of tissue inflammatory memory, both independently and cooperatively10,15,16,28,38,39,40,42? In an attempt to integrate key concepts of inflammatory memory distribution and storage, we describe select examples for each cell type, beginning with those cell subsets for which there is more robust experimental evidence and ending with emergent findings on the distribution of memory. We conclude with more speculative views on the cooperative storage of memories and highlight the need for experimental approaches to test the physiological relevance of inflammatory memory cell states in tissues.

Tissue leukocytes in memory storage

Evidence for the existence and functional roles of tissue-resident memory T cells revitalized the study of tissue-resident immune cells, a field originally pioneered through the study of intraepithelial lymphocytes17,33,35,43,44,45. The initial work on tissue-resident memory cells helped to identify functional markers, such as CD69 and CD103, that are enriched within tissue-resident CD8+ T cell subsets36,46,47. These markers have prompted more recent work characterizing the transcriptional networks that establish tissue-residency programmes and their functional contributions to tissue immunity18,36,46. Efforts continue to extend this paradigm to other lymphocyte subsets36,48.

B cells and plasma cells

The production of antigen-specific antibodies by memory B cells and terminally differentiated plasma cells serves to protect epithelia from bacterial colonization and subsequent invasion26,49. If an antibody can neutralize an environmental agent before epithelial colonization, it may completely restrict the need to call in the next layer of lymphocytes and molecular mechanisms of defence26 (Fig. 3). These antibodies may be synthesized locally and/or in lymphoid organs50.

Fig. 3: Inflammatory memory alters circuits in barrier tissues.
figure 3

The flow of information through a naive tissue (part a) or a hypothetical experienced tissue (part b), in which memory has been established, with key regulators at each step highlighted on the side of the vertical flow. Arrows indicate the predominant direction of information flow. By establishing memory in specific cell subsets, such as antibody-producing plasma cells, resident memory CD8+ T cells and epithelial progenitors, the typical flow of information can be ‘inverted’ in a tissue that has inflammatory memory. BRM cell, tissue-resident memory B cell; TH cell, T helper cell; TPM cell, peripheral memory T cell; Treg cell, regulatory T cell; TRM cell, tissue-resident memory T cell.

Recent work identified discrete subsets of tissue-resident antibody-producing cells. After pulmonary infection of mice with influenza virus, a subset of lung memory B cells expressing CD69 and CXCR3 could persist long term and promote faster viral clearance upon reinfection by producing IgA and IgG51. Furthermore, the process of selecting B cells that are adept at neutralizing viral escape preferentially occurs in long-lived germinal centres within the lungs52, and antigen re-encounter, rather than non-specific inflammation, was shown to be important for establishing resident memory B cells53. Together, these studies provide evidence that the presence of memory B cells in the lung parenchyma leads to enhanced quantity, quality and rapidity of responses, resulting in enhanced protective immunological memory. However, some tissues, such as the lower female genital tract, may rely on rapidly recruited CXCR3+ memory B cells, rather than resident plasma cells, for protective immunological memory54. Although the duration of resident memory B cell responses remains to be determined, solid-organ transplant studies in humans support the potential for plasma cells to preserve lifelong memory in the intestine55.

Whether antibody-mediated B cell memory is productive or detrimental varies according to context. Recent theory and studies suggest that antibody-mediated protection strategies may actually support niche establishment of microorganisms that express certain antigenic determinants by selecting for preferential retention56,57. Paradoxically, the presence of pathogen-specific memory B cells in the lungs may even allow for influenza A virus to infect cell types other than epithelial cells, illustrating how the virus may exploit host defence strategies58. The presumed antimicrobial functions of specific induced antibody mechanisms require careful consideration in light of host–microorganism co-evolution59.

T cells

If a microbial pathogen evades antibody detection and successfully infiltrates an epithelial barrier, conventional T cell-mediated effector programmes provide the next layer of specific control. After education in secondary lymphoid organs, conventional T cells gain the ability to migrate to their target tissue and extravasate from the vascular bed into the parenchyma to establish tissue residency31,32,60,61,62. These steps of tissue homing are required for circulating cells to become tissue residents and are therefore distinct from the developmental determination of permanent resident cell types, such as parenchymal, stromal and neuronal cells. The literature on tissue-resident memory lymphocytes has been extensively covered in recent reviews36,46,47.

Experiments using mouse models for infection-independent transplantation of activated CD8+ T cells directly into tissues, and the depletion of vaccination-induced circulating memory cells, have highlighted the functions of resident memory cells in enhancing tissue-restricted infection control36. Pioneering studies demonstrated that antigen-specific CD8+ T cells transferred directly into the skin markedly suppressed replication of herpes simplex virus, and, in a parabiotic mouse model of vaccinia virus skin infection, resident memory T cells provided 300-fold better protection than central memory T cells33,35. Transfer of lung memory CD4+ T cells into naive recipient mice also afforded complete protection from influenza virus-induced mortality34. Collectively, these findings illustrate that enhancing the quantity of antigen-specific, tissue-resident CD8+ or CD4+ T cells may mediate more effective host protection.

Some fundamental questions that arise from these observations include what are the qualitative effector mechanisms used by these cells; how many cells are required for immunity; and how are they retained long term? The answers to these questions may be interdependent, as distinct effector mechanisms can act directly on specific target cells (such as through cytotoxicity) or affect a tissue more broadly (such as through interferon release)17,45, and this may explain how antigen-specific stimulation can lead to heterologous immunity to unrelated pathogens63,64. Tonic expression of several co-inhibitory receptors may restrain the functional activation of expressed cytotoxic and cytokine gene modules, as suggested by scRNA-seq of human colonic intraepithelial CD8+ T cells in healthy individuals and in patients with ulcerative colitis65. How effector capacity is intrinsically and/or extrinsically modulated in tissue-resident CD8+ T cells will be of interest to explore further46,47.

To understand what constitutes effective protective immunological memory, it is crucial to define the relative number of tissue-resident memory T cells required to effectively survey permanent resident cell types (such as epithelia, fibroblasts and neurons)66,67. Quantitative microscopy illustrated that a lymphocytic choriomeningitis virus (LCMV)-induced T cell in the small intestine may only need to scan six target cells, relative to the 80 targets it was ‘responsible’ for by previous estimates and that, at least after LCMV infection, 90% of memory CD8+ T cells in barrier tissues are resident memory T cells32. The use of quantitative models to determine contact-dependent and contact-independent mechanisms across distinct lymphocyte subsets and environmental challenges will be essential to more accurately quantify surveillance capacity67.

It is currently difficult to generalize about the requirements for the best-known key extrinsic ‘residence’ factors for T cells — namely, antigen, transforming growth factor-β (TGFβ) and IL-15 — as even seemingly related tissues, such as the upper respiratory tract and the lungs, appear to have different requirements68. Local antigen is not required for resident memory T cell programming, but it can amplify numbers of tissue-resident memory T cells by approximately 50-fold (refs53,69,70,71). As TGFβ and IL-15 are trans-presented growth factors that maintain tissue-resident memory T cells62,72, understanding the relative contributions of distinct cell sources of these cytokines will be key73,74 (see Memory niches). Furthermore, the eliciting type of inflammation and/or antigen strength may play a role in the capacity to form CD8+ and/or CD4+ resident memory cells in the lungs75,76. It will also be important to determine how metabolism intersects with these cues, as recent work shows the importance of lipid metabolism and extracellular ATP in shaping fitness and persistence of tissue-resident memory T cells77,78,79. Other work has explored residency in human tissues, leveraging unique diseases and treatments to clarify specific mechanisms of tissue-resident memory T cells80,81.

CD4+ T cells in tissues and CD8+ tissue-resident memory T cell subsets follow different rules. We speculate that non-migratory tissue-residency programmes may help to confine cytotoxic programmes to sites of previous inflammatory exposure, whereas when inflammation subsides, CD4+ T helper cells may patrol new sites33,82,83,84,85,86. However, forkhead box P3 (FOXP3)+CD4+ regulatory T cells have been shown to exploit mechanisms for long-term tissue retention yet ‘memory-less’ suppressor function87,88,89. Future work is needed to elucidate how spatio-temporal aspects of tissue memory circuits relate to organismal memory90.

These studies on tissue-resident memory T cells illustrate how the presence of newly recruited cell subsets can fundamentally invert the typical circuits within a naive tissue from the innate immune system to the adaptive immune system (Fig. 3). Recent work on CD8+ T cells resident in the uterus and skin suggests that these cells autonomously dominate the secondary recall response relative to recruited effector T cells91. In summary, after establishment of inflammatory memory, the key sensor of an experienced tissue becomes the TCR, instead of the Toll-like receptor (TLR), and activates antiviral immunity via the output of B cells, dendritic cells, natural killer cells, keratinocytes3,63,64 and, potentially, other resident cell subsets.

Innate lymphoid cells and innate-like T cells

The qualitative components of resident memory T cells, such as the production of cytokines in the absence of lymph node (re-)education, are often properties of the anticipatory developmental programmes of innate lymphoid cells (ILCs) and innate-like lymphocytes48,92,93. Furthermore, antigen-specific resident memory T cells, similarly to ILCs, can activate a generalized state of tissue alarm or alertness through the production of interferon-γ (IFNγ) acting on other resident cell subsets63,64,91,92. What, then, is the role of antigen-specific memory in tissues, when the qualities appear similar to those developmentally programmed into resident lymphocytes?

The distribution of memory into antigen-specific cell subsets may serve to allow for wider sensing of environmental exposures through TCRs and BCRs, together with more tailored cell state outputs than those by innate lymphocytes and/or innate-like lymphocytes4,94. Consistent with this idea, Tcrd–/– and Tcra–/– mice have a hyper-inflammatory phenotype in the intestinal tract, perhaps from the chronic compensatory activation of ILCs17,94. The extent to which antigen-specific tissue-resident memory cells can replace or extend the homeostatic roles of ILCs is an exciting area of investigation92,93,95,96. The progressive loss, or increased specialization towards different subsets, of ILCs and innate-like T cells with increasing immunological events in tissues may replace their generalized tissue maintenance and repair functions in favour of host defence92,97,98,99.

Macrophages and dendritic cells

Work over the past few years has shown that macrophage and dendritic cell subsets share core developmental cell-type identity programmes as well as tissue-specific cell state adaptations that are driven by their local environment5,100,101. At the same time, investigators have developed the concept of trained immunity, whereby a primary exposure can functionally reprogramme myeloid cells to a secondary challenge with a similar or distinct pathogen-associated molecular pattern or microorganism12,38,102. The unifying aspect of tissue-specific identity and trained immunity is that epigenetic transcriptional response modules integrate with cell type and subset gene regulatory modules to drive the emergence of specific states that are adapted to that tissue and/or microbial challenge10,103,104,105,106 (Box 3). Although the importance of fetal-derived macrophage subsets relative to recruited monocytes in retaining tissue-resident memory remains to be determined, foundational work in this area highlights that myeloid progenitors in the bone marrow also become functionally reprogrammed107,108,109,110.

The concept of trained immunity was largely inspired by two parallel, but related, experimental approaches. Original experiments with Candida albicans infections in mice suggested that a non-lethal dose of C. albicans primes organisms and protects against a future, otherwise lethal, infection in a cytokine-dependent and macrophage-dependent manner28. However, priming with a pathogen or a pathogen-associated molecular pattern does not always lead to an enhanced response, as is the case for lipopolysaccharide (LPS) tolerance42. Studying the transcriptional and epigenetic response of LPS tolerance has highlighted that, whereas some pro-inflammatory mediators are suppressed, antimicrobial effector molecules are primed, which illustrates the layered control of specific effector transcriptional modules in macrophages42. Intriguingly, exposure to β-glucan derived from C. albicans can reverse some aspects of LPS tolerance111, suggesting contextual control over the formation and erasure of memory.

Early studies on the role of trained immunity in tissues suggest that IFNγ from T cells can directly prime alveolar macrophages, which correlate with enhanced bacterial protection after viral exposure107. Current work is also examining how environmentally responsive transcription factors (such as PPARγ) and cytokine-responsive transcription factors (such as STAT6) cooperatively modify macrophage cell states112. It will be of interest to determine how macrophage-mediated trained immunity operates within the broader context of tissue immunity.

Non-leukocytes in memory storage

Whereas tissue residency is an acquired cell state of an effector lymphocyte or monocyte, it is a developmental property of epithelial, stromal and neuronal cell types46,113,114. Across all four cell lineages, if a cell (or its progeny) can persist between the initial immunological event and subsequent recall, then it may have the capacity to meaningfully contribute to collective inflammatory memory. Of course, there may also be instances of bystander effects whereby alterations occur in a cell lineage without impacting subsequent recall. Here, we consider recent insights into epithelia-retained inflammatory memory, which poses an intriguing problem regarding how long-term storage of information could be achieved within a rapidly regenerating compartment, as well as recent insights into the contribution of stromal and neuronal cells to inflammatory memory.

Epithelial cells

The architecture of stratified (skin), pseudo-stratified (airways) and monolayer (lungs and gut) epithelia are distinct, yet all are composed of specialized epithelial cell subsets that collectively perform tissue-essential functions. The production of these cell types is regulated at the level of multipotent epithelial progenitor and stem cells113,115,116,117. This design is crucial for the epithelial development, turnover (in the order of weeks to months) and response to induced demands5,6,7,118. If immunological events are to be remembered by epithelia directly, the progenitor compartment would be the prime candidate for storage, as memory assigned to terminally differentiated cells could be rapidly lost.

Groundbreaking insights came from the analysis of dietary regulation of progenitor cell function in the intestine. In this study, mice given a high-fat diet showed increased stemness of both intestinal stem cells and progenitor cells, such that the stem cells became independent of their normal requirement for a Paneth cell niche in organoid-forming assays119. Furthermore, fatty acid-sensing transcription factor PPARδ was seen to drive elements of the WNT pathway and enhance stemness and tumorigenicity of intestinal progenitors. The enhanced stemness of organoids grown from the epithelia of mice given a high-fat diet persisted across multiple passages, suggesting differences in the output of mature epithelial cells and a potential cellular memory component of stemness that was retained ex vivo119.

Another pioneering study in this area highlighted that a brief period of psoriasis-like inflammation could fundamentally alter skin epithelial stem cells to more rapidly repair a subsequent wound at that site14,16. The contributions of skin-resident macrophages or T cells were formally excluded from this process and, instead, the AIM2–caspase 1–IL-1β axis seemed to be essential for implanting and recalling this memory at a later date16. The inflammatory memory state in progenitor cells persisted for at least 180 days, showing durability of the response.

Identifying non-immune cellular mechanisms that might sustain human inflammatory disease at barrier tissues, another recent study uncovered striking changes in epithelial cellular diversity and mature functional cell types by performing scRNA-seq of samples from patients with chronic rhinosinusitis15. Intriguingly, this study found that alterations in mature cells were driven by the type 2 cytokines IL-4 and IL-13 acting directly on basal airway epithelial progenitor cells, shifting their cell state and restraining their differentiation capacity. Some of these inferred type 2 cytokine signatures determined in vivo could be propagated through long-term in vitro culture and expansion of sorted human basal progenitor cells, suggesting that these cells could serve as repositories for allergic inflammatory memory15.

Taken together, these studies indicate that memory of immune events can be integrated by epithelial progenitors through diverse sensors, altering their recall responses and functional outcomes across distinct barrier tissues and species15,16,119 (Fig. 3). These studies have yet to address, however, how memory may be preferentially retained within heterogeneous progenitor cell subsets120. Furthermore, it remains to be determined how inflammatory memory retained by epithelial cells affects stress-induced ligands, which interact with innate-like T cells, or chemokines that recruit conventional T cells, and potentiate cooperation45,47,121. Notably, these studies of inflammatory memory by epithelial stem cells reveal putative transcription factors on the basis of increases in the accessibility of specific chromatin regions and draw parallels with the mechanisms of transcriptional maintenance observed in myeloid and lymphoid immunological memory12,28,47,103,106,122,123. It is tempting to speculate that genetically encoded susceptibility to disease may modify the epigenetic landscape within disease-relevant cell states at baseline to alter predisposition65.

Stromal cells

Stromal cells include fibroblasts and vascular and lymphatic endothelial cells. They are essential constituents of barrier tissues that provide structure, conduits for cell migration and developmental cues to surrounding cell types31,124. Furthermore, stromal cells have key immunological functions, are pathognomonic in various settings and can lead to the establishment of fibrotic disease125. Of note, experiments performed using type I interferons provided the initial evidence illustrating cytokine-induced priming of subsequent responses, and the underlying mechanisms were further investigated, in large part, with fibroblasts126,127.

A recent review39 elegantly unifies recent experimental evidence of how stromal cells can also exhibit characteristics of inflammatory memory, and we distil select elements here. Most studies have investigated synovial fibroblasts, isolated from inflamed or non-inflamed joints, and their enhanced responsiveness after priming to stimuli such as tumour necrosis factor and LPS40,128,129. Intriguingly, synovial fibroblasts, unlike macrophages or endothelial cells, did not show LPS tolerance, illustrating how distinct cell types may differentially regulate identical genes to facilitate a future stereotyped or learned response40,128,129,130. Extending preliminary insights obtained from scRNA-seq studies to determine whether specific human disease-associated fibroblast subsets preferentially retain more complex memory cell states will be of great interest65,131.


Although the cell bodies of neurons do not reside within epithelial barrier tissues, their processes densely innervate them2,132. Recent work showed that autonomic and peripheral sensory neurons can exert substantial regulation of immunological processes at barrier tissues through directly interfacing with specific immune cell subsets2,133,134 (Fig. 3).

One study highlighted that a specific subset of heat-sensing neurons in the skin drives IL-23 production from dermal dendritic cells, leading to IL-17 production from dermal γδ T cells, and promoting psoriasis-like inflammation132. Intriguingly, this pathway can be triggered solely by activating neurons optogenetically and is beneficial for host defence during C. albicans skin infection135. Further work on understanding painful bacterial infections emphasized important roles for pain-sensing neurons in directly detecting bacterial toxins and regulating the inflammatory response to intradermal infections136,137. There has also been particular interest in studying ILCs, where investigators have identified several distinct means of neuroimmune interaction, effectively placing a new sensory and regulatory layer above the typical innate-like functions of ILCs at barrier tissues133,138.

Is there a memory of painful exposures that can be recalled to inform future immunological encounters? Although the aforementioned circuits have not been explicitly tested for memory formation, recent evidence suggests the possibility that activating cell bodies or peripheral axons of neurons can influence both avoidance behaviour and anticipatory peripheral immune responses, even in the absence of other inflammatory cues135,139,140.

Distribution towards cooperative memory

Cooperativity is a common phenomenon in biology, acting as a crucial regulator of interactions at the molecular, cellular and species levels. Importantly, cooperation can lead to the emergence of higher-order functions that would not have been possible for the individual component4,5,56,141. This specialization is well appreciated as a facilitator of cell type and tissue evolution in metazoans. However, our study of immunological memory has largely focused on the individual components, rather than on their collaborative actions (multiple cell subsets working independently) or cooperative actions (multiple cell subsets working together). Here, based on recent evidence suggesting that multiple cell types distribute memory storage, we propose that inflammatory memory can be retained in cooperative cell circuits (Fig. 4). These circuits may help to reinforce cell states and/or promote cell residence in tissues.

Fig. 4: Distribution and cooperation in tissue memory.
figure 4

a | A schematic illustrating that stromal, neuronal, epithelial and haematopoietic cells can contain distinct combinations of cell state modules that define their cell type, typical role in tissue maintenance, intrinsic immune capacity and inflammatory memory capacity. The presented values reflect hypothetical relative distributions. b | Potential distributed inflammatory memory states for a system comprising two cell types: (line 1) no memory; (lines 2, 3) one cell type, or the other, has inflammatory memory; (lines 4, 5) one cell type, or the other, has inflammatory memory that partially influences the other cell type; (line 6) both cell types collaborate in forming inflammatory memory; or (line 7) both cell types, through cooperation, can mutually reinforce one another.

Intercellular circuits

Looking beyond the intrinsic functions of each cell’s state, we ask how distinct lineages might collaborate or cooperate towards the larger goal of barrier tissue adaptation (Fig. 4). Epithelial cells have long been appreciated as producers of cytokines and chemokines during barrier tissue challenges9,49. In particular, during type 2 immunity, epithelial cells can release the instructive cytokines thymic stromal lymphopoietin (TSLP), IL-33 and IL-25 (ref.49). Importantly, IL-33 expression by basal cells isolated from human donors with chronic obstructive pulmonary disease can be sustained long term142. Together, this work has highlighted that information flow occurs from the epithelium to lymphocytes and myeloid cells.

Only recently has it been appreciated that immune effector cytokines can act directly on permanent residents (that is, stem cells) of the epithelial barrier14. In pioneering work, ILC3-derived IL-22 was shown to influence tissue stem cells in the intestine and promote tissue regeneration143. More recent work in human ulcerative colitis links CD4+ T cell-derived IL-22 with enterocyte fate65. Although IL-22 is generally beneficial owing to its pro-regenerative activity, it may occasionally be detrimental, depending on its cell source and inflammatory context94,143,144, shifting epithelial cell states away from normal metabolic function and towards tumorigenesis94,145.

Direct interactions also facilitate information flow. Such interactions have been revealed in greater detail, for example, through the application of scRNA-seq to generate an atlas of the mouse small intestine, highlighting which specific subsets of stem cells and differentiated cells express MHC class II molecules116,120,146. Expression of MHC class II and antigen presentation machinery was increased in cycling intestinal stem cells, reduced in enterocytes and absent in secretory cell subsets120. By testing lineage-defining cytokines from T helper cell subsets, the authors found that IL-17 was associated with an increase in transit amplifying cells, IL-13 with increased tuft cells and IL-10 with increased stem cell self-renewal120.

Recent work also identified discrete circuits between epithelial instructive cytokines and lymphocyte effector cytokines in type 2 immunity147,148,149. This work showed that specialized secretory epithelial cells known as tuft cells are responsible for the production of IL-25. Helminth infection enhances tuft cell IL-25 production, leading to an increased abundance of both IL-25+ tuft cells and IL-13+ ILC2s147. Follow-up work clarified that dietary-derived or parasite-derived succinate can also activate this circuit148,149. It will be interesting to determine the points at which homeostatic and induced circuits exhibit long-term adaptation or memory responses in one, or both, of these cell types.

Memory niches

Although we know some of the cytokines, chemokines, growth factors and adhesion molecules that are important for the maintenance of resident memory T cells and B cells, the cellular sources of these factors remain incompletely defined. Furthermore, the way in which inflammation shapes their expression by distinct cell lineages has not been fully explored. We highlight some vignettes into how niches are formed in barrier tissues and speculate about how memory in niche cells of any lineage may alter the size or type of memory that is retained.

IL-15, which is a key maintenance factor for intraepithelial lymphocytes, seems to be synthesized by specific subsets of hair follicle-proximal keratinocytes150,151,152. IL-15 maintains both innate-like T cells and subsequent waves of recruited resident memory CD8+ T cells46,62,72. This creates another interesting problem: how are niches reallocated to accommodate newly recruited T cells? Early work in this area highlighted how, upon inflammation, the network pattern of dendritic epidermal T cells and Langerhans cells present in mouse epidermis at steady state becomes altered and more complex through their replacement by resident memory CD8+ T cells153. Similarly, a recent study into coeliac disease illustrated how chronic inflammation can reshape the epithelial niche for innate-like γδ T cells, which are usually sustained by butyrophilin-like 3 and butyrophilin-like 8 molecules in homeostatic conditions, leading to the entry of gluten-reactive, IFNγ-producing γδ T cells154.

The total storage capacity of tissues for distinct antigen-specific resident memory T cell populations has not been formally explored, but early studies suggest that the niche may be capable of accommodating several waves of recruitment without displacement155. Moreover, researchers are starting to consider how inflammatory cues during subsequent waves of tissue damage either help to retain and reinforce or displace and reprogramme existing memory niches. Indeed, extracellular ATP was shown to preferentially deplete bystander resident memory T cells in the tissue to free up niches for new infection-relevant specificities79. Stromal cells will likely play a central role in niche formation156. Exaggerated niche structures known as ectopic lymphoid follicles can form in the lungs via type I interferons, leading to CXCL13-mediated support of germinal centre reactions157. Studies exploring how cellular circuits regulate relative cell numbers during homeostasis and disease will help to inform exploration of this important topic in tissues, with single-cell approaches enabling this line of inquiry37,65.

It will be interesting to determine whether ILCs and innate-like T cells serve as essential early settlers in a tissue to preserve future niches for adaptive lymphocytes, in addition to their sentinel role92. Competition for niches between embryonic or adult-derived macrophages may also be relevant in trained immunity. Whereas, in some cases, cells of the same type may compete for niches, distinct cell types can also form cooperative arrangements to maintain memory108. In some barrier tissues, myeloid cells appear to be instrumental for the preservation of memory within the sub-epithelial compartment, for example, through production of the chemokine CCL5, which retains clusters of resident memory CD4+ T cells in the genital mucosa158. There is also evidence for cooperation between closely related cell subsets, whereby the presence of IFNγ-producing CD4+ T cells helps CD8+ T cells enter the epithelium159. Even the earliest work on resident memory CD8+ T cells illustrated the potential for cooperativity between CD4+ and CD8+ T cells in local immunity33.

Beyond internal cues, tissue-resident cells may also integrate cues from the external environment into their intrinsic cell state programmes160. Some cell subsets even require an environmentally derived ligand for persistence. For example, aryl hydrocarbon receptor ligands, such as those derived from cruciferous vegetables in the intestine, are essential to maintain both intraepithelial innate-like T cells and some conventional CD8+ T cell subsets161. Similarly, microorganism-specific T cells adapt their residency programme to balance effector and immunoregulatory functions in response to their microbial environment93,162,163. Comparing wild or ‘pet store’ mice with specific pathogen-free mice will be essential to determine how environmental factors are stored as transient adaptations or long-term memory164. Using TCR-transgenic systems and swapping antigens across bacterial species illustrated that microbial specificity, rather than antigen, dictates the resultant T cell effector state165,166. More broadly, abundant products of colonic microbial fermentation — that is, short-chain fatty acids — have been shown to promote the production of peripheral regulatory T cells167 and also the antimicrobial function of macrophages168. How diet, leukocytes and microorganisms intersect to impact the quantity, quality and distribution of memory will be of interest to explore further169.

Challenges and future perspectives

The concept of innate and adaptive immunity has been, and will continue to be, essential to establish a framework for how immune responses are initiated, maintained and remembered9,170. Here, we extend recent theory and data to suggest that most, if not all, cells in tissues can adapt to, and potentially remember, immunological events12,25,28,30. The process of inflammatory memory is fundamentally concerned with promoting tissue adaptation to environmental exposures during homeostasis, maintenance and disease settings. This process draws from any of the cell types, subsets and states that may be available at the time of exposure, and the specific molecular mechanisms that each individual cell has within its repertoire. Even molecular structures associated with classical adaptive responses, such as the TCR and the BCR, have hard-wired ‘innate’ functions that are deployed in advance of, or in parallel with, their ‘adaptive’ properties171,172.

One additional area for further investigation relates to the durability of tissue memory, and this will require rigorous definitions of what constitutes short-term and long-term adaptation relative to memory responses10. In simple terms, durability reflects a combination of factors, including the stability of inherited memory within a cell, the lifespan of that cell and the ability of that cell to propagate its memory to progeny within a barrier tissue. We highlight that, by necessity, the two longest-lived cell types in a barrier tissue must be its adult epithelial stem cells and underlying stroma. Furthermore, the dense innervation of barrier tissues, in conjunction with the long-lived nature of sensory and autonomic neurons, raises the possibility that neuron-encoded memory might be able to influence subsequent immune responses in tissues2. Determining the mechanisms of inflammatory memory will allow for experimental systems to rigorously test cell states for their potential relevance to tissue adaptation and maladaptation. In particular, it will be of interest to determine the duration, distribution and interaction of these mechanisms across cell lineages, relative contributions to specific responses and how they are shaped by host and environmental factors15,29. Improving the throughput and cost-effectiveness of techniques to map tissues will provide new opportunities to programme and reprogramme inflammatory memory formation prophylactically and during disease.

Does a particular type of inflammation preferentially allow residence and maintenance for the same polarized T cell type? It remains to be seen how the diversity of immunological experience in a tissue influences the specificity, quantity, quality, durability and plasticity of memory storage. This will likely be dictated by whether memory is truly cell intrinsic or stored in cell-level or tissue-level cooperative circuits, as well as the overall capacity of the tissue for information flow. Furthermore, how the increased specificity of responses to immune events following memory formation intersects with the repair and regenerative capacity of barrier tissues remains to be systematically explored. Although there is an overt emphasis here on local immunity, it is also important to remember that redistributing memory through migration to new sites within the same tissue, or between tissues, is essential for overall organismal health.