Memory T cells can show a range of persistence within non-lymphoid compartments. Many lymphocytes move freely through the various organs unimpeded before exiting the tissue via the draining lymphatic vessels1,2,3. Recognition of antigen leads to their transient retention4 while physical barriers may slow the return of cells to the circulation5. Finally, a subset of T cells is specialized for purely localized patterns of immune surveillance6,7 and only poorly exits the tissues, if at all8. These tissue-resident memory T (TRM) cells9 have a cell-autonomous limitation in their recirculation capacity10,11,12 and show a superior ability to control localized infections in a number of settings13,14,15. In this Perspective, I detail the transcription networks that define sequestered TRM cells, identifying CD103+CD8+ memory T cells as the key population of memory CD8+ T cells that encompasses all the hallmarks of permanent tissue residency. Finally, I describe how these archetypical TRM cells show an unusual pattern of egress from the lungs and discuss how this impacts the course of respiratory infections, including SARS-CoV-2.

Identifying tissue-resident memory T cells

TRM cells were initially identified as a distinct, sessile T cell subset that coexisted alongside tissue-emigrating T cells8. This was a break from the prevailing understanding of tissue T cells based on early lymphocyte circulation experiments5,16,17. At that time, the widely accepted view was that these T cells were simply recirculating memory cells that either happened to be found in non-lymphoid tissues in large numbers18,19,20 or, alternatively, were trapped by some sort of gating mechanisms or by structural barriers such as the basement membrane that lines epithelia5. The identification of a unique TRM cell subset meant that non-lymphoid tissues contained at least two populations of memory T cells, each with its own distinct phenotype and functional properties. One was a recirculating subset that at the time was thought to comprise effector memory T (TEM) cells17 and the other, the newly identified permanently resident TRM cell population.

A major challenge from that point onwards has been distinguishing non-migrating TRM cells from recirculating memory T cells, largely because of the phenotypic overlap between these populations. For example, TRM cells do not express CC-chemokine receptor 7 (CCR7) — a receptor required for entry into lymphoid tissues and the marker that was originally used to differentiate TEM cells (identified as CCR7-negative) from lymphoid-tissue-constrained central memory T (TCM) cells (identified as CCR7-positive)17. Separately, CD69 had been proposed to be a pan-TRM cell identifier21, yet it is upregulated by both antigen-specific and nonspecific stimuli22 and a substantial fraction of migratory T cells express this molecule once in the tissues23. Compounding the confusion is the extensive heterogeneity seen in both circulating and tissue-resident memory T cell populations24,25,26,27, expanded by a history of natural infection28. Therefore, although combinations of surface markers can cover a range of TRM-like tissue cells, it would be fair to say that to date there remains no unifying phenotypic identifier for this population.

CD103+CD8+ TRM cells: the archetypical TRM cell

Although it has proven difficult to identify TRM cells by definitive phenotypic means, therapeutic and experimental interventions can eliminate all circulating T cells from the blood, leaving long-term tissue residents as the only memory T cells remaining in non-lymphoid compartments. Two approaches have proven particularly useful in this regard. The first exploits T cell responses against a transplantation mismatch to selectively eliminate cells in the circulation29,30,31 whereas the second uses a more versatile cytolytic antibody-based technique for the same purpose27,32,33. Of additional importance is the in vivo infusion of labelling antibodies before tissue analyses to exclude cells that are simply in the vasculature34. This technique eliminates confounding contributions by blood-borne cells and is critical when examining highly vascularized organs such as the lung, although it does not identify TRM cells per se.

One of the striking features of the mouse TRM cells that remain after circulating T cells are depleted from the tissues is the dominance of a population of CD8+ T cells expressing the CD103 (also known as αE integrin) subunit of the αEβ7 integrin complex23,33. CD103+ TRM cells are highly enriched in the environmentally exposed epithelia of the skin, small intestine and female reproductive tract8,10,35. At these epithelial sites, interaction between αEβ7 and its abundantly expressed target ligand E-cadherin36 probably plays a role in cell adhesion and retention. However, CD103+CD8+ TRM cells are also found in non-epithelial tissues such as the brain12,37, and although CD103 has variously been implicated as being important for TRM cell development38,39,40, its expression is not ubiquitous37 and therefore not mandatory for all forms of T cell residency. Nonetheless, tissue-lodged CD103+CD8+ memory T cells are highly resistant to equilibration across parabiotic pairs41, are uniquely spared from elimination by the approaches mentioned above23,30, selectively survive for prolonged periods in transplanted tissues in mice8,33 as well as in humans42,43 and persist independently of antigen recognition15,37. Moreover, CD103+CD8+ memory T cells are usually not found in secondary lymphoid organs15,44 — with one striking exception45,46 to be described in detail below. Thus, although not all TRM cells express CD103, the balance of evidence argues that CD103+CD8+ tissue T cells are true TRM cells, making this an easily identifiable archetypical population and an ideal reductionist means for delineating tissue residency mechanisms.

RUNX3 and TGFβ in CD8+ TRM cell development

Early experiments in mice comparing the transcriptomes of CD103+CD8+ TRM cells isolated from a range of organs with those of their circulating counterparts provided some of the first insights into the transcription networks critical for TRM cell development and survival10,39. Not surprisingly, genes associated with tissue egress were found to be downregulated in TRM cells, including Ccr7 and the genes encoding the sphingosine-1-phosphate receptors S1PR1 and S1PR510,11. Without the downregulation of these receptors, the precursors of TRM cells return to the circulation, thereby dampening TRM cell development11,47. Other genes that come into play are those involved in dealing with local metabolite availability7,48,49 and those that prolong T cell survival23, with both sets of genes necessary to maintain a long-lived sequestered T cell population. Further experiments fleshed out how transcription factors control the various networks, such as the involvement of KLF2, which modulates the expression of S1PR1 and CCR711. Following this, key upstream gene regulators were identified, such as T-bet and EOMES23,50 as well as BLIMP1 and the BLIMP1 homologue HOBIT51,52; of note, BLIMP1 and HOBIT are also involved in the development of innate-like lymphocytes that permanently reside in mouse tissues, such as natural killer cells and natural killer T cells51. Most recently, an overarching transcription factor has come into focus. RUNX3 has been identified as contributing to TRM cell formation, and it directly or indirectly regulates BLIMP1 and KLF2 expression as well as modulating downstream retention components53. This contribution is particularly striking as RUNX3 is a pivotal player in CD8+ T cell development and functionality54,55.

As the network analyses evolved, one commonality to emerge was the involvement of TGFβ in TRM cell development and survival in a range of tissues and organs23,40,56,57,58. TGFβ appears to use a non-canonical signalling pathway59 that controls much of the CD8+ TRM cell gene expression signature60. It has been shown to facilitate tissue entry via selectin upregulation61 and can regulate a broad range of transcription regulators and cytokine-driven survival factors during CD8+ TRM cell development11,23,50. Combined, there is now a wealth of data regarding the tenets of transcriptional control of TRM cell formation, which largely pivots around a TGFβ–RUNX3 axis, at least in the case of the mouse CD8+ TRM cell subset.

TRM cell re-entry into the circulation

Although TRM cells were originally shown to persist in non-lymphoid organs in quasi-perpetuity8, there have been subsequent descriptions of TRM cell egress with the resultant ‘ex-TRM cells’ ultimately being incorporated into the circulation33,41,62. CD8+ TRM cell numbers show an intrinsic decline in organs such as the lung and liver30,41, but not in tissues such as the skin and small intestine, where the cells effectively remain in place for life once lodged8,41. However even for these fixed populations, TRM cells can be forced to leave using in situ antigen stimulation via peptide challenge33,44. Such active dislodgement is not universal, with CD103+CD8+ TRM cells sometimes remaining resident in the tissue even after multiple rounds of cell division initiated by local infection8,63,64. Perhaps tellingly, when CD103+CD8+ TRM cells are selectively dislodged by intervention, the resultant ex-TRM cells appear to adopt a phenotype intermediate between those of upstream resident memory T cells and conventional recirculating memory T cell populations, with a CD103 expression status that is either undefined or reported to be transient33,44. Moreover, when these same cells are directly isolated from non-lymphoid compartments, they are inferior in their capacity to enter the circulation compared to counterparts extracted from lymphoid organs33,65.

It remains difficult to reconcile these conflicting results, but studies on CD8+ TRM cells in the liver and recent revelations regarding the basis for CD4+ T cell residency provide valuable insight that might explain egress variability. Although much more is known about CD8+ TRM cells, there are many examples of CD4+ TRM cell-type counterparts13,27,66. Comparisons make it clear that the two are unrelated in terms of mechanistic underpinnings and they can exhibit quite distinct patterns of tissue residency even in the same organ29,67. As noted above, the archetypical CD103+CD8+ TRM cells use a set of TGFβ-driven transcriptional networks to shut down tissue egress, upregulate survival factors and tailor metabolic pathways. By contrast, few of these networks have been associated with CD4+ TRM cell residency, which instead relies on retention mechanisms variously operating via cell aggregation, antigen-specific T cell activation and chemotactic agents68,69 (Fig. 1). The reason why CD4+ and CD8+ TRM cells are likely to differ at the mechanistic level is the pivotal role RUNX3 plays in TRM cell development and survival53. This transcription factor is repressed in CD4+ T cells by the opposing gene regulator ThPOK (also known as ZBT7B), which is itself a lineage-determining factor70,71. Although natural RUNX3 upregulation can convert CD4+ T cells to an unconventional CD8αα+ intraepithelial regulatory T cell population with CD8+ TRM cell-like qualities72, the intrinsic paucity of RUNX3 expression in conventional CD4+ TRM cells results in low CD103 levels in these cells and more transient tissue residency as a direct consequence of their inability to access the RUNX3-mediated pathways downstream of TGFβ signalling53,73.

Fig. 1: Subtypes of tissue-resident memory T cells based on transcription profiles.
figure 1

The mechanism promoting permanent residency in non-lymphoid tissues for the CD103+CD8+ tissue-resident memory T (TRM) cell population involves a RUNX3-driven transcriptional network that is downstream of TGFβ receptor signalling53,60. This transcription programme is missing in CD4+ TRM cells as a consequence of deficiencies in RUNX3 expression73. Instead, these populations use a combination of cell aggregation and extrinsic chemokine networks for tissue retention68,69. The typical CD103+CD8+ TRM cell transcription programme is also missing in CD103 liver-like TRM cells because of deficiencies in TGFβ engagement65.

Somewhat analogous to their CD4+ tissue-resident T cell counterparts, mouse liver CD8+ TRM cells are also deficient in CD103 expression74. These cells show medium-to-long-term tissue residency74, but not the almost lifelong persistence of TRM cells in organs such as skin and small intestine41,65. Although the liver T cells are fully capable of responding to TGFβ, local requirements negate this capacity, resulting in an immature or less differentiated CD103 TRM cell population (Fig. 1) with an inferior term of residency combined with more flexible reprograming compared to mature CD103+ TRM cell counterparts65. Collectively, the results show that although CD103 TRM cells can reside in tissues for a considerable period, they can exhibit a range of spontaneous egress and reprogramming capabilities because of deficiencies in TGFβ-mediated maturation. Given the heterogeneous nature of tissue-resident T cells, including variable CD103+ T cell content across different organs37 and the known recruitment of recirculating T cells by the peptide stimulation used for TRM cell dislodgement32, it is possible that less differentiated populations analogous to the liver CD103 TRM cells may preferentially contribute to the egress process. Regardless, although some TRM cells can leave the tissues and enter the circulation, the balance of data argues that for the archetypical CD103+CD8+ TRM cells, this process is not constitutive and when it does happen, it usually results in cells that do not phenocopy their direct upstream antecedents.

TRM cells in the lungs

From the discussion above, it can be reasonably argued that because they fully engage the TGFβ–RUNX3 residency programme, mouse CD103+CD8+ tissue T cells fit the original TRM cell definition8; specifically, they form a distinct subset of memory T cells that remains lodged in peripheral compartments in virtual perpetuity. However, there is one organ where the CD103+CD8+ T cells do not abide by this rule, and its uniqueness has important disease implications. Unlike the situation in other tissues, CD103+CD8+ TRM cells in the lung do not require local antigen stimulation for dislodgement45. Also unusually, the egressing memory T cells retain cell surface expression of CD103 post-exit, meaning that the lung-draining lymph nodes are unique in having a substantive subset of memory CD8+ T cells with this marker45,46. Lung CD103+CD8+ TRM cells are fully mature and unremarkable in terms of their TGFβ requirement for development and survival58. They also express the gene signatures associated with tissue residency10,12, including a cluster of TRM cell-associated transcription factors, namely HOBIT, NR4A1, aryl hydrocarbon receptor (AhR) and BHLHE407,51,75,76. In all critical aspects, they resemble TRM cells from other tissues, meaning that their exit from the lung is probably an organ-specific feature rather than due to a cell-intrinsic programme. Such a mechanistic distinction is important, as it would suggest that the egress process would probably capture TRM cells beyond the archetypical CD103+CD8+ subset that was used to define this phenomenon and would do so regardless of where they fall on a maturation and term-of-residency continuum.

Exacting experiments by Stolley and colleagues45 proved that the resultant draining lymph node-resident memory T cells were indeed constitutively derived from upstream lung tissue counterparts, possibly dislodged as a consequence of virus-induced tissue damage77,78 or the interruption of tonic TGFβ signalling needed to retain TRM cells in tissues79. Although the resultant lymph node accumulation offers an additional avenue to maintain regional protection45,80, memory T cell exit helps to explain one of the intriguing conundrums associated with immune protection in the lungs. It has long been known that T cell immunity in the lung wanes over time, with this first reported for respiratory infections with influenza virus and Sendai virus in mice81,82. This decline in lung-based immunity occurs despite virus-specific memory cells persisting in the circulation30,82,83,84. Non-TRM cell-based mechanisms were originally proposed to describe the behaviour of lung T cell populations81,85,86,87, variously confounded by blood-borne cells that are particularly problematic when dealing with this highly vascularized organ34. More recently, it was shown that the waning local immunity correlates with declining lung TRM cell numbers in mouse after influenza virus infection83,84 and in humans after respiratory syncytial virus challenge88. Although other mechanisms have been posited to account for this TRM cell attrition, such as the selective death of lung TRM cells30,84 or the disappearance of structures associated with focal damage67, none exclude concurrent tissue egress. Once lost, lung TRM cells are difficult to replace in the absence of renewed infection owing to the strict antigen recognition requirements for effective lodgement67,83,89, which are optional in many other tissues37 including the upper respiratory tract90. Overall, a range of mechanistic overlays would imply that losing TRM cells over time is important for this organ — for example, to limit ongoing damage to its delicate oxygen-exchange architecture91.

Finally, the natural decay of lung TRM cells stands in stark contrast to what is seen elsewhere in the body, where CD103+CD8+ TRM cell populations can remain tightly contained (Fig. 2). CD103+CD8+ TRM cells show long-term persistence in organs such as the brain, skin and cervicovaginal tissue8,39,92, despite the loss of their CD103 counterparts. The extent to which these spatial and temporal restrictions can operate was dramatically illustrated by experiments that lodged CD103+CD8+ TRM cells in a small patch of skin, thus confining effective protection to just that location while leaving the remainder of the torso under the inferior control of memory cells in the blood8,63. By contrast, lung TRM cell residency is unstable and transient, resulting in surveillance that is increasingly dependent on recirculating populations over time, with a concomitant decline in local T cell immunity.

Fig. 2: Selective and constitutive egress of lung CD103+CD8+ TRM cells.
figure 2

Inflammation associated with infection of tissues such as skin, small intestine and reproductive tract (left panels) and lung (right panels) leads to the recruitment of a variety of CD4+ and CD8+ T cells that combat the invading pathogens (part a). These populations include effector memory T (TEM) cells that continuously recirculate between non-lymphoid organs and blood as well as tissue-resident memory T (TRM) cell precursors (not shown). Following resolution of the infection (part b), most of the recruited T cells exit or die, leaving local immunosurveillance to recirculating TEM cells and the more potent TRM cells. Over time, some TRM cell subsets selectively disappear (part c, left panel), resulting in a resident population highly enriched in long-lived CD103+CD8+ TRM cells that afford long-term local immunity against re-infection (part d, left panel). In the lungs, CD103+CD8+ TRM cells are gradually lost after the infection has resolved and instead accumulate in the proximal draining lymph nodes (part c, right panel) leaving the lower respiratory tract deficient in CD103+CD8+ TRM cells and thus susceptible to re-infection (part d, right panel).

TRM cell lung egress and immunity to SARS-CoV-2

At the time of this writing and nearly three years since the emergence of the SARS-CoV-2 virus in late 201993,94, the COVID-19 pandemic continues to be a major challenge in many parts of the world. Despite reports showing that circulating antiviral T cell immunity can be cross-reactive against emerging variants95, long lived96,97 and associated with better disease outcomes98,99, immunity from combinations of COVID-19 vaccination and SARS-CoV-2 infection has been found to steadily decline100,101. One possible contributor may be that anti-SARS-CoV-2 tissue-resident T cells that are pivotal for immune protection show the same type of numerical decay as reported for mouse CD103+CD8+ TRM cells. Employing strategies that slow TRM cell loss102 could be advantageous, as might approaches that circumvent the lung altogether. The upper respiratory tract, especially the nasal mucosa, is a prime target with respect to the latter possibility as it does not show the TRM cell decline that is intrinsic to the lung90. Alternatively, it may be that TRM cells are actually counterproductive, leading to tissue damage. This is especially poignant because repeated antigen encounters extend the durability of CD103+CD8+ TRM cells in the lung102, yet a recent report found that experiencing successive SARS-CoV-2 infections progressively increases the risk of adverse health outcomes103. In terms of their potential to contribute to tissue damage, TRM cells have an innate immune alarm and recruitment function32,104, and the innate response has been shown to be a key mediator of COVID-19-associated lung pathology105,106.


Overall, TRM cells provide superior protection against tissue-localized infection, primarily because of constraints in their migration capabilities. Despite proving to be long-lived and effective in a range of different infectious diseases, lung TRM cells have an unusual propensity for tissue exit reflected in a decay in local T cell immunity. Such a feature may have evolved to protect this organ against long-term damage or may simply be a by-product of some unique anatomical feature intrinsic to lung function. Given the ability of TRM cells to respond to infection with an immediacy unmatched by the blood-based memory populations, there is a need to focus on their deposition in the different compartments of the respiratory system, especially in settings or sub-regions that support their long-term survival.