Research ArticleCELL MIGRATION

Salivary gland macrophages and tissue-resident CD8+ T cells cooperate for homeostatic organ surveillance

See allHide authors and affiliations

Science Immunology  03 Apr 2020:
Vol. 5, Issue 46, eaaz4371
DOI: 10.1126/sciimmunol.aaz4371

Salivary Gland Surveillance

Pathogen sensing in tissues is critical to generating rapid immune responses. Within these tissues, macrophages and resident memory CD8+ T cells (TRM) work together to detect pathogens, and Stolp et al. use intravital imaging of submandibular salivary glands (SMG) to show that TRM follow tissue macrophage topology in a dynamic manner. Macrophage depletion is associated with reduced TRM motility and a diminished clustering in response to inflammatory chemokines. However, although SMG TRM respond to chemoattractants, autonomous motility is also observed and is mediated by friction and insertion of cellular protrusions into microenvironmental gaps. These findings demonstrate that SMG TRM can use different motility modes in proximity to tissue macrophages to patrol the tissue microenvironment.


It is well established that tissue macrophages and tissue-resident memory CD8+ T cells (TRM) play important roles for pathogen sensing and rapid protection of barrier tissues. In contrast, the mechanisms by which these two cell types cooperate for homeostatic organ surveillance after clearance of infections is poorly understood. Here, we used intravital imaging to show that TRM dynamically followed tissue macrophage topology in noninflamed murine submandibular salivary glands (SMGs). Depletion of tissue macrophages interfered with SMG TRM motility and caused a reduction of interepithelial T cell crossing. In the absence of macrophages, SMG TRM failed to cluster in response to local inflammatory chemokines. A detailed analysis of the SMG microarchitecture uncovered discontinuous attachment of tissue macrophages to neighboring epithelial cells, with occasional macrophage protrusions bridging adjacent acini and ducts. When dissecting the molecular mechanisms that drive homeostatic SMG TRM motility, we found that these cells exhibit a wide range of migration modes: In addition to chemokine- and adhesion receptor–driven motility, resting SMG TRM displayed a remarkable capacity for autonomous motility in the absence of chemoattractants and adhesive ligands. Autonomous SMG TRM motility was mediated by friction and insertion of protrusions into gaps offered by the surrounding microenvironment. In sum, SMG TRM display a unique continuum of migration modes, which are supported in vivo by tissue macrophages to allow homeostatic patrolling of the complex SMG architecture.


During viral infections, effector CD8+ T cells (TEFF) generated in reactive lymphoid tissue disseminate into nonlymphoid tissues (NLTs) including the gut, lung, genitourinary tract, and skin. TEFF recruitment into NLTs is mediated by inflammatory chemokines and integrin ligands that direct these cells to kill infected cells (13). After clearance of viral antigens, TEFF differentiate into CCR7+ central memory T cells (TCM) and continue to patrol lymphoid organs, or they differentiate into effector memory T cells that lack CCR7 and CD62L expression and recirculate through NLTs (4). In addition, a subset of TEFF differentiates into tissue-resident memory T cells (TRM), which stably reside in NLTs and, to a minor extent, in lymphoid tissue, as a nonrecirculating, self-renewing population. TRM continuously patrol NLTs in search of cognate peptide–major histocompatibility complex on local cells and act as “first-line” sentinels to eliminate infected cells and to trigger an organ-wide alert status through cytokine secretion upon pathogen re-encounter (3, 59).

The steady-state patrolling behavior of epidermal TRM after clearance of infections has been extensively studied in mouse models. Epidermal TRM display characteristically elongated, dendritic shapes and move with speeds of 1 to 2 μm/min in proximity to the extracellular matrix (ECM)–rich basement membrane (BM) separating epidermis from dermis, i.e., in plane with the bottom keratinocyte layer (10, 11). A similar pattern was recently confirmed in explanted human skin (12). Inhibition of Gαi receptor signaling induces epidermal TRM to round up and stop migrating, suggesting that similar to naïve T cells (TN), chemoattractants are essential for their homeostatic search strategy (13). In addition, epidermal CD8+ T cells follow local chemokine signals to accumulate around infected cells upon pathogen reencounter (14). Recent work has shown that such CD8+ T cell accumulation is a critical step for cooperative elimination of infected stromal cells through repeated cytotoxic attacks (15).

TRM are also present in exocrine glands of the head and neck region, including submandibular salivary glands (SMGs). Salivary glands are targeted by several bacteria and viruses, including human β- and γ-herpesviruses, which can cause disease, mostly in immunocompromised individuals (16, 17). Similar to the skin and gut, SMG contains epithelial tissue basally anchored onto connective tissue. However, although the skin and gut consist of “layered” connective and epithelial tissue, the SMG epithelium displays a more complex three-dimensional (3D) “arborized” geometry. SMG acini secrete saliva into intermediate and collecting ducts, from where it is channeled into the oral cavity via the Wharton’s duct (WD). The glandular epithelium is separated by a BM from the supporting interstitium containing blood and lymphatic vasculature, fibroblasts, and tissue macrophages, which are dispersed throughout tissue (18). In tissue sections, most CD8+ TRM in SMGs are localized within the abundant acini and ducts, implying a mechanism that allows TEFF arriving in interstitial venules to cross the BM underneath the epithelial compartment and develop into memory T cells (19, 20). The cellular dynamics of homeostatic tissue surveillance by SMG TRM and their interactions with other local cell types have not been examined to date.

Here, we used intravital two-photon microscopy (2PM) to uncover the high baseline motility of resting SMGs TRM of 6 to 7 μm/min. TRM followed tissue macrophage topology during SMG surveillance, and tissue macrophage depletion disrupted their patrolling behavior. Super-resolution microscopy of SMGs revealed discontinuous attachment of tissue macrophages to the surrounding tightly packed epithelium, offering paths of least resistance to migrating TRM. When examining the molecular mechanism driving steady-state TRM motility, we did not find compelling evidence for a decisive chemokine- or integrin-driven contribution. Instead, physical confinement alone was sufficient to trigger SMG TRM migration. Mechanistically, the autonomous motility of resting SMG TRM was mediated by friction- and protrusion insertion–driven migration modes, which also left cells responsive to extrinsic chemokine signals. Together, our data suggest a continuum of extrinsic factor–driven and cell-intrinsic TRM motility modes, which exploit topographic features of tissue macrophages to scan the complex 3D architecture of exocrine glands.


Systemic viral infection leads to the establishment of TRM in salivary glands

We used a viral infection model for a comparative analysis of CD8+ T cell populations in lymphoid tissue and SMG (fig. S1A). Systemic infection with lymphocytic choriomeningitis virus (LCMV)–ovalbumin (OVA), a replication-competent, attenuated LCMV mutant expressing OVA as model antigen (21), led to transient and low viral titers in spleens on day 3 postinfection (p.i.) that remained below the detection limit in peripheral lymph nodes (PLNs) and SMGs (fig. S1B). Adoptively transferred green fluorescent protein (GFP)+ OT-I (OVA-TCR-1) CD8+ T cell receptor (TCR) transgenic (tg) T cells (which recognize the OVA257–264 peptide in the context of H2-Kb) (22) underwent a prototypic expansion-contraction kinetic in the spleen and PLN over the course of 30 days (fig. S1, C and D). Despite the lack of detectable viral titers, OT-I T cells accumulated in SMGs from day 6 p.i. onward, with a stable population maintained until at least day 30 p.i. (fig. S1, C and D). By day 30 p.i., OT-I T cells isolated from SMGs but not PLN or spleen showed increased expression of CD103 and CD69, while losing the KLRG1+ (killer cell lectin-like receptor subfamily G member 1) population present on day 6 p.i. (fig. S1, E and F). These data suggest that most SMG CD8+ T cells had developed into bona fide TRM at day 30 p.i. (23). Memory OT-I CD8+ T cells isolated from PLN were about 65% CD62L+ CD44high TCM, with the remaining population being CD62L CD44high memory T cells. To take this heterogeneity into account, we refer to memory T cells isolated from PLN as TPLN-M.

Homeostatic SMG TRM migration is characterized by dynamic cell shape changes

We next determined the localization of TPLN-M and TRM in their target organs during the memory phase after LCMV-OVA infection (Fig. 1A). Immunofluorescence sections showed that most GFP+ OT-I T cells in PLN and SMG were dispersed evenly throughout tissue at day 30 p.i. (Fig. 1B). In PLN, OT-I TPLN-M cells localized mainly with smooth muscle actin (SMA)+ fibroblastic reticular cells, whereas most OT-I TRM in SMG were within or adjacent to EpCAM+ (epithelial cell adhesion molecule) acini and ducts (Fig. 1B). We developed a sequential surgery method to visualize homeostatic tissue surveillance of TPLN-M and TRM in PLN and SMG, respectively, in the same host by 2PM (24). TPLN-M displayed characteristic amoeboid shapes and moved with speeds similar to values reported for TN (11.8 ± 4.0 μm/min, median ± SD) (Fig. 1, C and F to H, and movie S1). Compared with TPLN-M, SMG TRM displayed more pronounced shape changes, with several protrusions probing the microenvironment during migration, at times with thin and elongated cell bodies (Fig. 1, D to G, and movie S2). Although TPLN-M and TRM covered large distances throughout the observation period of intravital imaging sequences (20 to 60 min), both populations differed in their speed and arrest coefficients, i.e., percentage of track segments with speeds <2.5 μm/min. Thus, SMG TRM moved slower than TPLN-M (Fig. 1H) and had higher arrest coefficients (Fig. 1I). Nonetheless, SMG TRM retained a relatively high motility coefficient, which is a proxy of a cell’s ability to scan the environment during random migration, of >15 μm2/min (Fig. 1J). Accordingly, their median speed of 6.8 ± 3.4 μm/min was notably higher than values reported for epidermal TRM (1 to 2 μm/min) (10), with some cells achieving speeds of >12 μm/min. Both TPLN-M and SMG TRM retained a fast response to antigenic stimulation, as systemic administration of cognate peptide resulted in immediate arrest and secretion of interferon-γ (IFN-γ) (Fig. 1, J and K).

Fig. 1 Dynamic motility parameters of memory CD8+ T cells in PLN versus SMG.

(A) Experimental layout for CD8+ T cell analysis in SMG and PLN. i.v., intravenously; i.p., intraperitoneally; IF, immunofluorescence. (B) Immunofluorescent sections of GFP+ OT-I T cells in PLN and SMG in memory phase (≥day 30 p.i.). Scale bars, 100 μm (left) and 20 μm (right). (C) Time-lapse 2PM image sequences showing OT-I CD8+ TPLN-M cell motility in PLN in memory phase (≥day 30 p.i.). High endothelial venule. (D and E) Time-lapse 2PM image sequences showing OT-I CD8+ TRM motility in SMG in memory phase (≥day 30 p.i.). Arrowheads indicate protrusions (D), and the arrow indicates squeezing behavior (E) of OT-I CD8+ TRM. Scale bars, 10 μm (C to E). Time in minutes:seconds. (F) Time-coded shapes of exemplary TPLN-M and TRM tracks. (G) Shape factor distribution of TPLN-M and TRM with exemplary cell shapes. (H) Speed frequency distribution of OT-I CD8+ T cells in PLN and SMG. Arrows indicate median values (in micrometers per minute). (I) Arrest coefficient frequency distribution of OT-I CD8+ T cells in PLN and SMG (cutoff, <2.5 μm/min). (J) Mean displacement versus time of OT-I TPLN-M (left) and TRM (right) before and after OVA257–264 injection with motility coefficients (in square micrometers per minute). (K) IFN-γ expression in OT-I TPLN-M and TRM 24 hours after OVA257–264 injection (means ± SD). (L and M) Speeds (L) and meandering index (M) of P14 CD8+ T cells in PLN and SMG in memory phase (≥day 30 p.i.) of LCMV Armstrong infection. (N) Speeds of P14 CD8+ T cells SMG TRM after injection of irrelevant (OVA257–264) or cognate (gp33–41) peptide. Red lines indicate median. (O) Speeds of OT-I CD8+ TRM in LG in memory phase (≥day 30 p.i.) after LCMV-OVA infection. Data in (G) are from two to three independent experiments and three mice total for each group. Data in (H) and (I) are pooled from five to six mice from four independent experiments with at least 194 tracks analyzed per organ. Data in (J) are pooled from three to four mice from two independent experiments. Data in (K) show one of two independent experiments. Data in (L) and (M) are from four to seven mice and those in (N) are from one to four mice in one to three independent experiments. Data in (G), (I), and (M) were analyzed with Mann-Whitney U test, and data in (H), (L), and (N) were analyzed with Student’s t test. ***P < 0.001, *P < 0.05.

To examine a distinct CD8+ T cell population, we transferred GFP+ P14 CD8+ TCR tg T cells (which recognize the LCMV epitope glycoprotein (gp)33–41 in the context of H2-Db) (25) and performed 2PM of PLN and SMG at >30 days p.i. with the LCMV Armstrong strain. We measured similar speeds and meandering indices for GFP+ P14 TPLN-M and TRM as with OT-I CD8+ T cells, both before and after cognate peptide administration (Fig. 1, L to N). Furthermore, GFP+ OT-I T cells patrolled the structurally comparable lacrimal gland (LG) in the same speed range as in SMG (7.6 ± 4.3 μm/min, median ± SD; Fig. 1O). Collectively, these data suggest that migration parameters of TRM patrolling exocrine glands during homeostasis are independent of TCR specificity and may reflect tissue properties.

TRM colocalize with tissue macrophages in salivary and LGs

To explore the microenvironmental context of exocrine gland TRM migration, we used a CD11c–yellow fluorescent protein (YFP) reporter strain that preferentially labels CD64+ SMG macrophages (fig. S2A), as previously described (18). In tissue sections, CD11c-YFP+ cells were also positive for the macrophage marker Iba-1, whereas some Iba-1+ cells were CD11c-YFPlow/negative, indicating that most but not all tissue macrophages were labeled in CD11c-YFP mice (fig. S2B). Confocal analysis of thick tissue sections showed that CD11c-YFP+ tissue macrophages extended numerous protrusions from their cell bodies throughout the SMG tissue and were located within EpCAM+ ducts and acini, as well as SMA+ perivascular structures of the interstitium (fig. S2C). We further investigated the spatial arrangement of macrophage protrusions with regard to epithelial BM markers. The laminin ligand CD49f (α6) was prominent on the basal side of acini and, to a lesser extent, on ducts, which were identified by the presence of the tight junction protein ZO-1 on the luminal side. In some cases, tissue macrophage protrusions appeared to project between adjacent acini and ducts (fig. S2D).

To assess the spatial relationship between tissue macrophages and TRM, we transferred GFP+ OT-I T cells into CD11c-YFP recipients 1 day before infection with LCMV-OVA and analyzed tissue sections by confocal microscopy in memory phase (≥day 30 p.i.). We observed a notable spatial proximity of TRM and tissue macrophages in SMGs, with about 70% of OT-I T cells directly in contact with CD11c-YFP+ cells (Fig. 2, A and B, and movie S3). A comparable association of TRM and tissue macrophages was observed in LGs after LCMV-OVA infection (Fig. 2C). The close spatial association between SMG tissue macrophages and TRM was confirmed by correlative light and electron microscopy imaging, with both cell membranes adjacent to each other (Fig. 2D). Electron microscopy images also highlighted the compact tissue structure of acinar and ductal epithelium linked by tight junctions and surrounded by a dense ECM (Fig. 2E). Occasionally, we observed small TRM clusters around tissue macrophages (Fig. 2F). CXCR3−/− OT-I TRM failed to accumulate at tissue macrophage clusters, suggesting the existence of local CXCL9/CXCL10 “hotspots” at these sites (Fig. 2F). Similarly, CD3+ T cells colocalized with CD68+ macrophages in human parotid gland sections, both as dispersed individual cells and in clusters (Fig. 2, G and H). Together, TRM displayed a prominent colocalization with tissue macrophages in noninflamed salivary glands across species.

Fig. 2 TRM colocalize with tissue macrophages in SMG and LG.

(A) Immunofluorescent section showing localization of SMG TRM adjacent to tissue macrophages (arrows). Scale bars, 1 mm (left), 100 μm (middle), and 20 μm (right). (B) Percentage of SMG TRM adjacent to tissue macrophages. Data are pooled from 105 fields of view (FOVs) with a total of 3270 TRM and shown as a box-and-whiskers graph with 2.5 to 97.5 percentiles. (C) Immunofluorescent LG section in memory phase (≥30 days p.i. with LCMV-OVA) showing GFP+ OT-I TRM adjacent to CD11c-YFP+ tissue macrophages (indicated by yellow arrowheads). Scale bars, 1 mm (left), 100 μm (middle), and 20 μm (right). (D) Correlative light and electron microscopy sections (left, confocal image; middle and right, TEM image) showing close spatial association of SMG TRM and tissue macrophages. M, tissue macrophages; E, epithelial cell; ME, myoepithelial cell. Scale bars, 5 μm (left), 2 μm (middle), and 1 μm (right). (E) TEM images showing attachment of epithelial cells to ECM (top) and through intercellular junctions (white arrows; bottom). Scale bars, 800 nm. (F) Immunofluorescent section of WT and CXCR3−/− OT-I T cells and macrophages. Magnified image shows association of CXCR3−/− OT-I TRM to tissue macrophages (arrows). Scale bars, 100 μm (left) and 20 μm (right). (G) Dispersed T cells (brown) in human parotid salivary gland associate with macrophage cell bodies or thin protrusions (red), indicated by arrowheads. Scale bar, 20 μm. (H) Example of colocalization of CD68+ macrophage (red) and CD3+ T cell clusters (brown) in human parotid salivary gland. Scale bar, 50 μm.

Migrating TRM follow tissue macrophage topology during SMG surveillance

The spatial proximity of TRM to SMG macrophages in tissue sections raised the question whether patrolling TRM migrate alongside macrophages. 2PM imaging of GFP+ TRM in LCMV-OVA memory phase CD11c-YFP recipients confirmed that TRM crawled along CD11c-YFP+ macrophages during most of the observation period, with TRM shapes often closely matching the underlying macrophage topology. This was particularly evident along thin macrophage protrusions, which TRM often followed (Fig. 3A and Movie 1). At the same time, TRM protrusions occasionally detached from macrophages, apparently scanning the surrounding environment. Accordingly, we identified occasional TRM track segments that were not associated with tissue macrophages, with a minor reduction in TRM speeds (7.0 ± 5.3 μm/min with macrophages versus 6.1 ± 4.8 μm/min without macrophages; P < 0.001).

Fig. 3 Macrophage depletion disrupts TRM patrolling.

(A) 2PM time-lapse image sequence showing overlap of OT-I TRM tracks with tissue macrophages in SMG in memory phase (≥day 30 p.i.). Scale bar, 20 μm. Time in minutes:seconds. The right panels show the time accumulated overlays of images with or without OT-I TRM. (B) 2PM time-lapse image sequence of TRM in DTx-treated CD11c-YFP or CD11c-DTR → Ubi-GFP chimeras. Magenta lines indicate outlines of acini, and white segmented lines indicate cell tracks. Scale bars, 50 μm (overview) and 20 μm (inset). Time in minutes:seconds. (C) Example TRM tracks in the presence or absence of macrophage. Scale bar, 10 μm. (D) Frequency distribution of TRM speeds in DTx-treated CD11c-YFP or CD11c-DTR bone marrow chimera. Arrows indicate median (in micrometers per second). Data are pooled from two to four independent experiments with four to six mice total and analyzed with Mann-Whitney U test. ***P < 0.001. (E) Track analysis outline. Top: U-turns (red) describe tracks reversing direction while excluding continuous turns. Bottom: Synthetic tracks were generated to assess dwell time in an 80-μm-diameter sphere (black). One example track is shown for control (light blue) and macrophage-depleted (dark blue) condition. (F) Percentage of tracks making a U-turn. Bars indicate 95% confidence intervals. (G) In silico dwell times for TRM tracks in 80-μm-diameter spheres based on measured track parameters.

Movie 1 Intravital imaging of SMG TRM migration along tissue-resident macrophages.

Time series (left) and time accumulated overlay (right), including xy and xz projections. TRM (green) frequently move along thin protrusions of CD11c+ tissue-resident macrophages (blue). Blood vessels are shown in red. Major ticks 50 and 20 μm (zoom). Time in minutes:seconds.

Depletion of tissue macrophages disrupts TRM patrolling within and between acini and ducts

Our observations prompted us to examine TRM motility in the absence of tissue macrophages. To this end, we generated bone marrow chimera by reconstituting C57BL/6 or Ubi-GFP mice with control CD11c-YFP or CD11c-DTR (diptheria toxin receptor) bone marrow. At 6 weeks of reconstitution, we adoptively transferred GFP+ or DsRed+ OT-I T cells, followed by LCMV-OVA infection. In some experiments, we directly transferred OT-I T cells into CD11c-DTR mice and infected mice with LCMV-OVA. Both approaches allowed us to deplete CD11c+ macrophages by diphtheria toxin (DTx) treatment in the memory phase without affecting the unfolding of the adaptive immune response.

2PM imaging in DTx-treated mice revealed that SMG TRM patrolling behavior was disrupted when macrophages were depleted (Fig. 3B). TRM motility was decreased, reflected by less displacement (Fig. 3C) and slower speeds (Fig. 3D). In addition, we observed cells that returned and migrated back the same path within acini and ducts after macrophage depletion (Fig. 3B and movie S4). To quantify this behavior, we developed a method to specifically retrieve U-turns from track parameters (Fig. 3E). This analysis confirmed that the percentage of T cell tracks showing U-turns was doubled in DTx-treated CD11c-DTR SMG from 8.1 to 16.7% of tracks (Fig. 3F). We observed a similar impact of macrophage depletion on TRM speeds in LG (7.6 ± 4.3 and 5.5 ± 3.2 μm/min in control and macrophage-depleted LG; P < 0.001), with a 2.5-fold increase in U-turns (fig. S3, A and B).

We next asked how impaired motility affects organ surveillance. We generated tracks in silico from the datasets obtained by 2PM imaging of DTx- and control-treated SMG and assessed the average TRM dwell time in a sphere with a diameter of 80 μm as a surrogate epithelial structure (Fig. 3E). This analysis uncovered a nearly threefold increased sphere dwell time from 24 ± 1.8 min for control SMG to 69 ± 6.5 min (median ± SEM) for DTx-treated CD11c-DTR SMG (Fig. 3G). A similar finding was made for control and macrophage-depleted LG (fig. S3C). Together, macrophage depletion disrupted TRM patrolling and increased the propensity of TRM to make U-turns.

Absence of macrophages reduces TRM movements into and out of epithelium

We investigated whether lack of macrophages may also affect TRM transitions into and out of the epithelium as part of the impaired motility pattern, as suggested by movie S4. To address this point, we reconstituted irradiated Ubi-GFP mice expressing GFPs in all cells with CD11c-YFP bone marrow before transfer of DsRed+ OT-I T cells and systemic LCMV-OVA infection. We found that in these chimera, acini and ducts of surgically prepared SMGs were GFPbright and readily identifiable by their glandular shapes, whereas connective tissue was GFPlow. Using case-by-case 3D rendering of 2PM image sequences in memory phase (≥30 days p.i. with LCMV-OVA), we observed that DsRed+ TRM were not restricted to individual epithelial structures but occasionally crossed between adjacent acini or between epithelial and connective tissue compartments in a bidirectional manner along macrophage protrusions (Fig. 4A and movie S5). We confirmed this observation in a mouse model expressing membrane tomato and CD11c-YFP (Fig. 4B). In total, 75% of TRM transits (n = 42) into and out of epithelial structures occurred along macrophage protrusions (Fig. 4C). Given that not all tissue macrophages are YFP+ (fig. S2B), the actual percentage of macrophage-assisted transitions may still be higher. DTx treatment of CD11c-DTR SMGs reduced, but did not abolish, TRM transit into or out of acini and ducts. In total, we observed 55 TRM crossing events into or out of acini in CD11c-YFP versus 12 events in CD11c-DTR chimera SMGs. These data corresponded to a 77% fewer crossing events per hour track duration (Fig. 4D). Because this defect is more pronounced than the relatively mild decrease in TRM speeds after macrophage depletion (Fig. 3D), it is unlikely to be fully explained by decreased motility. Accordingly, when we normalized TRM transitions to migrated distance, we continued to observe impaired epithelial crossing in the absence of macrophages (Fig. 4E). Reduced TRM crossing into and out of epithelial structures was also observed when we prolonged DTx treatment for 5 days (movie S6).

Fig. 4 Absence of macrophages reduces TRM movement into and out of epithelium.

(A and B) 2PM time-lapse image sequences of TRM crawling along a macrophage to enter acini in Ubi-GFP (A) and membrane tomato/membrane GFP (mT/mG) (B) bone marrow recipients. Epithelial signal was manually masked to show an isolated acinus in zoomed panels. Dashed white line indicates area displayed in xz view, and arrow indicates TRM-macrophage contact. Scale bars, 50 μm (A, overview), 20 μm (A, insert), 20 μm (B, overview), and 10 μm (C, insert). Time in minutes:seconds. (C) Percentage of TRM transitions into or out of acini and ducts in CD11c-YFP → Ubi-GFP chimeras (n = 42) with and without contact to macrophages. (D and E) 2PM time-lapse image sequence of CD11c-YFP → Ubi-GFP and DTx-treated CD11c-DTR → Ubi-GFP chimeras was analyzed for TRM crossing events (leaving or entering acini). (D) shows average transitions per hour track duration, and (E) depicts transitions per 1000-μm total distance migrated. Data points represent individual image sequences. Line indicates means. (F) Experimental layout for analysis of TRM response to local chemokine. CXCL10 was injected with a fluorescent tracer (Qdots655) for 4 hours to allow TRM accumulation. Integrin-blocking mAbs prevent recruitment of circulating T cells. (G) TRM per square centimeter at sites of CXCL10 injection in the presence or absence of macrophages. Numbers indicate means ± SD. Data in (D) and (F) are pooled from two to four independent experiments with four to six mice total. Data in (D) were analyzed with Mann-Whitney U test, and data in (F) were analyzed with Wilcoxon rank test. *P < 0.05.

Clustering of SMG TRM is impaired after macrophage depletion

We set out to examine the impact of macrophage-TRM cooperativity in a local recall response. We infected CD11c-YFP or CD11c-DTR mice with LCMV-OVA 1 day after transfer of GFP+ OT-I T cells and depleted macrophages after 4 weeks using DTx (fig. S4A). One day later, we rechallenged mice with an mCherry- and OVA-expressing mouse cytomegalovirus (MCMV-OVAmCherry) (26, 27) administered locally via retrograde WD injection. After 48 hours, we analyzed SMG tissue sections for the presence and intensity of viral foci identified by mCherry+ signal. This analysis confirmed the strictly local reinfection because no mCherry signal was observed in the adjacent sublingual gland (fig. S4B). We observed close proximity of OT-I T cells and CD11c-YFP+ cells with MCMV-infected cells after SMGs infection (fig. S4C), with macrophages occasionally engulfing infected cells (fig. S4D). This observation is in line with the well-described macrophage core function of engulfing apoptotic cells (efferocytosis) (2831). Accordingly, we observed massively increased numbers of infected cell foci in macrophage-depleted SMGs after WD infection with MCMV-OVAmCherry as compared with SMGs containing tissue macrophages, irrespective of the presence of TRM (fig. S4E). In the absence of tissue macrophages, TRM partially suppressed viral replication as assessed by decreased mCherry intensity in viral foci (fig. S4F). Given the increased viral load after DTx treatment, tissue macrophage depletion did not permit the direct assessment of the functional impact of reduced TRM migration patterns in the presence and absence of these cells.

We therefore designed an experiment to examine TRM cluster formation in response to exogenously added inflammatory chemokine as a prerequisite for efficient elimination of infected cells (15). We treated LCMV-OVA–immunized CD11c-YFP and CD11c-DTR BM chimera mice with DTx, followed 1 day later by local injection of the CXCR3 ligand CXCL10 and quantum dots into SMG (Fig. 4F). We also administered anti-α4 and LFA-1 blocking monoclonal antibodies (mAbs), which inhibit recruitment of circulating T cells to SMG (32). At 4 hours after CXCL10 administration, we isolated SMG and quantified TRM enrichment in thick confocal SMG sections according to the area marked by the coinjected fluorescent quantum dots. CXCR3−/− OT-I T cells did not show accumulation in CXCL10 injection sites, supporting the specificity of chemokine-triggered clustering (49 to 63 cells/cm2 > 500 μm versus 60 to 68 cells/cm2 < 500 μm from injection site; range from two SMG). Wild-type (WT) OT-I TRM were twofold enriched at CXCL10 injection sites, suggesting that these cells had followed a CXCL10 gradient or became retained during their surveillance path (Fig. 4G). In contrast, local accumulation of TRM was lost when macrophages had been depleted, although TRM numbers outside the site of chemokine injection remained comparable with macrophage-containing SMGs (Fig. 4G). These data suggested that presence of SMG macrophages permitted TRM clustering in response to local inflammatory chemokines.

Macrophages show discontinuous attachment to neighboring cells within SMG

We examined macrophage embedding within salivary gland tissue in more detail. In immunofluorescent sections, most macrophage protrusions contained phosphotyrosine-positive patches (Fig. 5, A and B), suggesting the presence of focal adhesion–like structures at these sites. We investigated how these local signals correlated with macrophage attachment in situ. We applied the super-resolution shadow imaging microscopy (SUSHI) technique, which was originally developed to visualize the complex topology of the extracellular space (ECS) in living brain slices (33). SUSHI can also be used to follow dynamic changes in ECS in response to a hyperosmotic challenge, which leads to cell shrinkage and ECS widening in brain tissue. We adapted SUSHI imaging to acutely sliced SMG sections, which were superfused with the cell-impermeable fluorescent dye calcein to examine macrophage anchorage to adjacent epithelium (Fig. 5C). Steady-state imaging revealed that the interstitium contained more ECS as compared with the tightly packed epithelium (Fig. 5, D and E). We reasoned that SUSHI in combination with hyperosmotic challenge could be applied to explore attachment between neighboring cells. Performing time-lapse ECS imaging, we gradually increased the osmolarity to induce cell shrinkage, which led to a strong increase in ECS in the interstitium (movie S7). In turn, interepithelial junctions remained relatively stable and only mildly increased their spacing under osmotic challenge, reflecting the presence of adherens and tight junctions known to link epithelial cells (Fig. 5F). In contrast, hyperosmolarity induced intraepithelial CD11c-YFP+ macrophages detachment from the adjacent epithelium (Fig. 5, G and H). This finding confirms previous observations that tissue macrophages do not form continuous adhesive contacts with the epithelium, unlike the extensive cell-to-cell contacts between acinar epithelial cells (34).

Fig. 5 SMG macrophages show discontinuous attachment to neighboring cells.

(A) Confocal SMG section showing pTyr signal in tissue macrophages. Left panel shows macrophage/pTyr/4′,6-diamidino-2-phenylindole (DAPI) signal, and right panels show three consecutive z-stacks (spacing, 1 μm) of macrophage/pTyr signal. Scale bars, 3 μm. (B) Quantification of pTyr+ protrusions. (C) Experimental layout of SUSHI of SMG slices. (D) Example of SUSHI image for determination of ECS. E, epithelium; BV, blood vessel; LUT, look-up table. Scale bars, 10 μm. (E) Overview of ECS signal with SMG epithelium and CD11c-YFP+ tissue macrophages. Scale bar, 10 μm. (F) Example of epithelial attachment before and after hyperosmotic challenge. Arrows show interepithelial junctions. Scale bar, 5 μm. (G) Examples of macrophage detachment before and after hyperosmotic challenge. Arrowheads indicate detachment. Scale bars, 5 μm. (H) Quantification of gap size between macrophage and epithelium before and after hyperosmotic challenge. (I) Confocal image of SMG section with macrophage protrusions traversing a BM below an epithelial acinus (indicated by arrow). Scale bars, 20 μm (overview) and 5 μm (insert). (J) Electron microscopy image of macrophages creating a discontinuation of the BM of an acinus (indicated by an arrow). Numbers mark two neighboring macrophages. Arrowheads indicate lack of tight adhesion between macrophages and neighboring cells. Scale bar, 2 μm. All images are representative of at least two independent experiments. Data in (H) were analyzed using a paired t test. ***P < 0.001.

For an examination of macrophage protrusions and epithelial BM, we analyzed laminin-stained tissue sections from immunized CD11c-YFP mice. We observed, in some cases, macrophage protrusions penetrating between adjacent acini, or between epithelium and connective tissue, thus bridging adjacent compartments separated by BM (Fig. 5I and movie S8). Using correlative confocal and transmission electron microscopy (TEM), we validated that some macrophage protrusions transversed BM (Fig. 5J). Together, our data uncover discontinuous attachment of tissue macrophages to neighboring cells and occasional incursions of macrophage protrusions across the epithelial BM, creating paths of least resistance for migrating TRM. These observations offer an explanation for TRM migration in accordance with tissue macrophage topography.

In vivo motility of SMG TRM remains largely intact upon integrin and chemokine receptor blockade

Our data established a correlation between macrophage topography and TRM migration patterns in noninflamed salivary and LGs. Next, we set out to dissect the molecular mechanisms that drive homeostatic TRM motility in SMG, focusing on well-described canonical chemoattractant- and integrin-signaling pathways. Given the expression of promigratory chemokines and adhesion receptors including intercellular adhesion molecule–1 (ICAM-1) on tissue macrophages (35), we examined their influence on TRM migration parameters. Integrins mediate attachment and force transmission through engagement of their ligands expressed by many cell types including macrophages. SMG TRM express α1, α4, αE, αL, β1, β2, and β7 integrins and low levels of αV (figs. S1 and S5A). To assess their involvement in SMG TRM immune surveillance, we administered a mix of integrin-blocking mAbs against the major lymphocyte integrin αL (CD11a/CD18 and LFA-1), the E-cadherin receptor αE (CD103), and α4 (VLA-4 and α4β7) to LCMV-OVA memory phase mice containing TPLN-M and TRM (Fig. 6A). We confirmed that mAbs were saturating surface integrins at the time point analyzed (fig. S5B). We then followed OT-I T cell motility in PLN and SMG on ≥day 30 p.i., using dual surgery 2PM of the same recipient mouse as above. Integrin blockade lowered TPLN-M speeds from 11.7 to 8.8 μm/min (fig. S5C), similar to the decreased cell speeds of CD18-deficient TN in lymphoid stroma (36). In contrast, TRM speeds and crawling along tissue macrophages remained unaltered by this treatment (Fig. 6, B and C, and movie S9). We did not detect Mac-1 (CD11b/CD18) expression on SMG TRM by flow cytometry, and addition of anti-Mac1 mAb to the integrin-blocking mix did not decrease TRM speeds or guidance by tissue macrophages (7.1 ± 2.9 μm/min). Similarly, inclusion of blocking mAbs against α1 and αV, together with αL, α4, and αE, had no impact on TRM speeds or association with macrophages (6.5 ± 2.7 μm/min).

Fig. 6 Robust in vivo SMG TRM motility after inhibition of Gαi and integrins.

(A) Experimental layout. (B) OT-I TPLN-M and TRM speeds after combined anti-αL, -α4, and -αE integrin mAb (αItg) inhibition. Arrows indicate median values (in micrometers per minute). (C) 2PM image of TRM–tissue macrophage colocalization in αItg-treated SMG. Arrows indicate T cell–tissue macrophage contacts. Scale bar, 20 μm. (D) OT-I TRM speeds in SMG after WD administration of RAD or RGD peptide. (E) WT and CXCR3−/− OT-I TRM speeds in SMG in memory phase (≥day 30 p.i.). Arrows indicate median values (in micrometers per minute). (F) OT-I TPLN-M and TRM speeds after systemic treatment with active PTx or inactive (mutant) PTx (PTxmut). Arrows indicate median values (in micrometers per minute). (G) 2PM image of TRM–tissue macrophage colocalization in PTx-treated SMG. Arrows indicate T cell–tissue macrophage contacts. Scale bar, 20 μm. (H) Flow cytometry plot of mixed TRM and macrophages. (I) Quantification of cluster formation as shown in (H). Data in (B), (D), (E), and (F) are pooled from two to five independent experiments with a total of two to seven mice with at least 111 tracks per condition and analyzed with unpaired Student’s t test. Data in (I) are pooled from two independent experiments and analyzed using unpaired Student’s t test. ***P < 0.001.

Poor surface saturation of blocking anti-β1 mAbs on OT-I T cells preempted our assessment of the role of β1 integrins for TRM motility by this approach. As an alternative, we directly administered the β1-blocking peptide RGD or the control peptide RAD through the WD into SMG and followed its impact on TRM motility parameters by 2PM. The WD channels saliva from SMG into the oral cavity and can be used to administer reagents or pathogens through retrograde duct cannulation (37). Control experiments using WD administration of OVA257–264 peptide led to instantaneous arrest of TRM similar to systemic injection, suggesting efficient peptide permeation of SMG by this route. WD injection of either peptide slightly lowered TRM speeds; yet, we did not observe an impact on RGD administration on TRM motility parameters as compared with control peptide (Fig. 6D). This reflected low β1 integrin levels in the interface between macrophages and TRM (fig. S5D). Furthermore, E-cadherin levels on macrophages and TRM were barely detectable in SMG tissue sections, arguing against a role for this cadherin in mediating close spatial association with tissue macrophages (fig. S5E).

Cytokine-driven chemoattractant production plays a key role for T cell trafficking. Because the CXCR3 ligands CXCL9 and CXCL10 play a role in TRM clustering in SMG (Fig. 2F), we cotransferred WT and CXCR3−/− OT-I T cells 1 day before LCMV-OVA infection. Consistent with a recent report (32), we found that the absence of CXCR3 did not impair TRM formation in SMG after viral infection. Nonclustered CXCR3−/− OT-I TRM showed no differences in speeds as compared with WT TRM (Fig. 6E), and lack of CXCR3 did not prevent TRM patrolling along tissue macrophages (Fig. 2F). These data argue against a role for CXCR3 in mediating baseline homeostatic motility of SMG TRM, despite its function in facilitating local clustering in response to proinflammatory CXCR3 ligands (Fig. 4F).

To comprehensively assess a function for potential chemoattractants, we inhibited Gαi signaling by systemic pertussis toxin (PTx) treatment (38) and performed 2PM analysis of OT-I T cell motility parameters on ≥day 30 after LCMV-OVA infection. To control for inhibitor efficacy, we took advantage of the dual surgery of PLN and SMG in the same recipient. Systemic PTx administration slowed TPLN-M down from 11.3 μm/min in control to 8.6 μm/min in PTx-treated recipients (fig. S5F), resembling observations made with PTx-treated TN in PLN (39). Speeds were also decreased in SMG TRM (from 6.6 to 5.5 μm/min) by PTx treatment (Fig. 6F), suggesting a role for chemoattractants in mediating high TRM speeds. Nonetheless, we observed a robust residual motility and continued TRM crawling along tissue macrophages in the presence of PTx (Fig. 6G and movie S10). Furthermore, PTx treatment had essentially no impact on U-turn frequency (1.15-fold increase as compared with PTxmut), in contrast to the absence of macrophages (Fig. 3F). These data suggest that although Gαi-coupled receptors contribute to SMG TRM motility, they are not required for TRM association with tissue macrophages. Last, we interfered with matrix metalloproteinase (MMP) activity using the broad MMP-9, MMP-1, MMP-2, MMP-14, and MMP-7 inhibitor marimastat as previously described (40). MMP inhibition did not reduce TRM migration speeds (6.9 ± 3.0 μm/min). Together, with the exception of a minor effect by PTx, the in vivo inhibitor treatments examined here did not alter TRM motility and close spatial proximity to tissue macrophages.

To directly assess intercellular adhesion, we coincubated freshly isolated tissue macrophages and TRM ex vivo and analyzed cluster formation by flow cytometry (Fig. 6H). As a positive control, we preincubated macrophages with cognate OVA257–264 peptide. Although addition of OVA257–264 to tissue macrophages induced detectable binding to TRM, the baseline association between both populations remained low (Fig. 6, H and I). In sum, within the technical limitations of our experimental approach, our in vivo and in vitro observations did not identify specific molecules that provide strong adhesion of resting SMG TRM to tissue macrophages. Our data rather suggested that TRM association to macrophages occurred preferentially in the context of the SMG microanatomy. Our data do not exclude the presence of unidentified adhesion receptors mediating TRM association to tissue macrophages in vivo.

SMG TRM motility is induced by confinement and can be tuned by extrinsic factors

To address the molecular mechanisms underlying homeostatic TRM motility, we used under-agarose assays that allow one to precisely control environmental factors and provide confinement required for T cell motility (Fig. 7A) (41). To benchmark our system, we transferred TN on CCL21- and ICAM-1–coated plates as surrogate lymphoid tissue microenvironment. We observed high chemokinetic TN motility with similar speeds as measured in vivo (13.3 ± 5.9 μm/min) (Fig. 7, B and C, and movie S11) (36, 41). Similarly, TRM showed a high motility (11.4 ± 3.0 μm/min) when migrating on CXCL10 + CXCL12– and ICAM-1–coated plates (Fig. 7C and movie S12). These observations show that SMG TRM respond to inflammatory chemokines and adhesion molecules with high speeds.

Fig. 7 Confinement induces autonomous SMG TRM motility through friction.

(A) Experimental layout of under-agarose assay. Arrows indicate F-actin flow. (B) Representative TN (n = 75) and TRM (n = 58) tracks in the presence of chemokine and ICAM-1. (C) Speeds of TN and TRM. Data are presented as Tukey box-and-whiskers plot. (D) Time-lapse image sequence showing TRM motility among immotile TN. TRM displacement shown by segmented line. Scale bar, 20 μm. Time in minutes:seconds. (E) Time-lapse image sequence in under-agarose plates coated with HSA showing TPLN-M (top) and TRM (bottom) motility. Cell displacement is shown by segmented lines. Scale bar, 10 μm. Time in minutes:seconds. (F) Representative TN (n = 75), TPLN-M (n = 226), and TRM (n = 379) tracks. (G) TN, TPLN-M, and TRM speeds in under-agarose plates coated with HSA. Numbers indicate the percentage of tracks >3 μm/min (boxed). Lines indicate median. (H) Meandering index of TN, TPLN-M, and TRM tracks. (I) Example image sequences showing TRM in transient contact with macrophages under agarose on fibronectin-coated plates. TRM displacement is shown by the segmented line. Scale bar, 50 μm. Time in minutes:seconds. (J) TRM-macrophage contact duration for individual tracks. (K) Time-lapse image sequence showing SMG (green) epidermal TRM (red) under agarose on HSA. SMG TRM displacement is shown by segmented lines. Scale bar, 20 μm. Time in minutes:seconds. (L) Speeds of SMG and epidermal TRM in under-agarose plates coated with HSA. (M) TRM speeds after treatment with PTx, RGD peptide, and anti-Mac1 mAb, or in the presence of EDTA. Numbers indicate the percentage of tracks >3 μm/min (boxed). Lines indicate median, and dashed line represents median speed of control. (N) Image sequence of TRM protrusions in the presence of EDTA. Scale bar, 10 μm. (O) Representative TRM tracks in the presence of EDTA (n = 75). (P) Mean displacement over time of TRM tracks. Numbers indicate motility coefficients (in square micrometers per minute). Data in (C), (G), (H), (M), and (P) were pooled from at least two independent experiments each. Statistical analysis was performed with unpaired t test (C) or Kruskal-Wallis with Dunn’s multiple comparison in (G), (H), and (M) (as compared with “TRM”). **P < 0.01, ***P < 0.001.

Because we found that in vivo motility of resting TRM was largely refractory to inhibition by PTx and integrins, we examined T cell displacement on plates coated with fatty acid–free human serum albumin (HSA) and thus free of chemoattractants and specific adhesion ligands. In line with previous findings (41), TN and TPLN-M remained essentially immobile throughout the observation period (Fig. 7, D to F, and movies S13 and S14). Under these conditions, only 12% of TN and 31% of TPLN-M migrated faster than 3 μm/min and showed low directionality (Fig. 7, G and H). Most SMG TRM showed robust intrinsic motility on HSA-coated plates despite the absence of chemoattractants and adhesion molecules (Fig. 7, D to F, and movies S13 and S14). Almost 70% of SMG TRM migrated faster than 3 μm/min with high directionality, with their median speed of 5.5 μm/min approaching values observed in vivo (Fig. 7, G and H). High temporal resolution imaging revealed that migratory TRM often formed several protrusions along the leading edge that appeared to probe the environment, followed by rapid displacement of the cell body along one of the protrusions (movie S15).

Next, we performed under-agarose assays in the presence of SMG tissue macrophages to assess their influence on TRM motility. On the few occasions when motile TRM contacted coplated tissue macrophages, these contacts were mostly transient (Fig. 7, I and J). Furthermore, TRM did not crawl along macrophage protrusions as observed in vivo (movie S16). These observations support the notion that TRM-macrophage association occurred preferentially in the SMG microenvironment.

We examined whether spontaneous motility was a common feature of all TRM populations. We isolated epidermal OT-I T cells from >30-day LCMV-OVA–infected mice, which had been recruited to DNFB (dinitrofluorobenzene) + OVA257–264–treated skin during the expansion phase. In line with the reported sensitivity to in vivo PTx treatment (13), epidermal TRM did not show spontaneous motility in the absence of chemokine (Fig. 7, K and L, and movie S17), although these cells remained responsive to exogenous chemokines (5.6 ± 4.2 μm/min in the presence of CXCL10 and CXCL12). Thus, confinement alone was sufficient to induce spontaneous SMG TRM migration, a behavior that, to the best of our knowledge, had not been previously reported for resting T cells. Their speeds were increased in the presence of chemokines and adhesion molecules, suggesting that extrinsic promigratory factors tune intrinsic cell motility.

Friction mediates SMG TRM migration in the absence of chemokines and ICAM-1

We set out to characterize the requirements for autonomous SMG TRM motility using under-agarose assays. Reflecting the absence of chemoattractants and integrin ligands, PTx treatment or addition of the β1-blocking peptide RGD did not affect TRM speeds in this setting (Fig. 7M). Although Mac-1 binds weakly to serum albumin (42), addition of anti–Mac-1 mAb did not cause a reduction in TRM speeds (Fig. 7M). These observations suggested a friction-based migration mechanism (43). Friction is the resisting force when two elements slide against each other and may be composed of a number of fundamental forces. Although the nature of the weak interactions between TRM and migratory surface causing friction are not defined, we hypothesized that these might, in part, involve bivalent cations. Chelation of bivalent cations by EDTA caused a strong decline of TRM speeds under agarose (Fig. 7M). High temporal resolution imaging showed that despite the lack of translocation in the presence of EDTA, SMG TRM continued to probe the environment via transient protrusion formation, essentially “running on the spot” (Fig. 7, N and O, and movie S18, left). This behavior precipitated a loss in the motility coefficient (Fig. 7P). In sum, our data suggest that bivalent cation-dependent friction between SMG TRM and the confining 2D surfaces generated sufficient traction for translocation in the absence of considerable surface binding.

SMG TRM insert protrusions between adjacent structures for translocation

In addition to friction-based migration, protrusion insertion has emerged in recent years as a complementary mechanism to allow cell migration without specific adhesions (43). The continuous probing of TRM in the presence of EDTA (Fig. 7N) provided an opportunity to test whether topographic features of the environment such as narrow intercellular spaces may rescue cell motility by permitting insertion of pseudopods as mechanical “footholds” (44, 45). As a surrogate approach to reintroducing a “2.5D” environmental geometry in under-agarose assays, we cotransferred a surplus of TN together with TRM and performed time-lapse imaging in the presence of EDTA and in the absence of chemoattractants and adhesion molecules (Fig. 8A). TRM localized within TN clusters frequently showed lateral displacement despite the presence of EDTA (movies S18, right, and S19). Under these conditions, TRM displacement occurred through insertion of protrusions between adjacent TN and subsequent translocation of the cell body accompanied by dynamic cell shape changes (Fig. 8B). TRM within TN clusters were faster than isolated TRM (5.8 ± 3.0 and 2.3 ± 1.7 μm/min, respectively), displayed higher directionality, and resembled in cell shape and speeds TRM migrating in vivo (Fig. 8, C to E). Once TRM had traversed TN clusters, they returned to their probing behavior without efficient translocation, indicating a close interdependence on physical contact and motility (movie S18, right).

Fig. 8 TRM insert protrusions for cell displacement in the absence of external chemoattractants and friction.

(A) Experimental layout. Arrows indicate protrusion direction. (B) Image sequences of TRM within TN clusters in the presence of EDTA. Arrowheads show membrane protrusions, and segmented lines indicate cell track. Scale bars, 10 μm. Time in minutes:seconds. (C) Graphical representation of TRM inside TN cluster (i) or dispersed (ii). (D) TRM track speeds according to their location. Numbers indicate the percentage of tracks >3 μm/min (boxed). Lines indicate median. (E) Meandering index of TRM tracks sorted according to their location. Lines indicate median. (F) Image sequences of TRM alone (top) and with 7-μm polystyrene beads (bottom) in the presence of EDTA. Arrowheads show membrane protrusions, and segmented lines indicate cell track. Scale bars, 10 μm. Time in minutes:seconds. (G) TRM track speeds according to their association with or without beads. Numbers indicate the percentage of tracks >3 μm/min (boxed). Lines indicate median. (H) Meandering index of TRM tracks sorted according to their location. (I and J) Proposed model of tissue macrophage–assisted TRM immune surveillance in SMG. (I) Confined SMG TRM display distinct migration modes. Ex vivo confined SMG TRM respond to chemoattractants and adhesion molecules, while displaying autonomous motility through friction. Furthermore, SMG TRM adapt to environmental topology to insert protrusions and move between adjacent structures that are not tightly connected. (J) By associating with protrusion-forming tissue macrophages, which lack extensive contacts with surrounding tissue cells, TRM exploit their distinct motility modes for organ surveillance. Arrows indicate protrusion direction. M, tissue macrophage. Lines indicate median. Data in (D), (E), (G), and (H) are pooled from four to five independent experiments. Statistical analysis was performed with Mann-Whitney U test. ***P < 0.001.

We then examined whether potential residual molecular interactions between TN and TRM might act as drivers of migration. We therefore transferred uncoated polystyrene beads with TRM in under-agarose assays. These beads replaced TN as surrogate 2.5D structures and allowed us to examine protrusion insertion in the absence of potential adhesive interactions. In this setting, TRM recapitulated the behavior observed within TN clusters, showing effective cell displacement only when in contact with clusters of beads for protrusion insertion (Fig. 8F and movie S20). TRM speeds increased to 6.4 ± 1.9 μm/min and became more directional when in contact with beads, whereas isolated TRM showed no displacement (Fig. 8, G and H). This was also seen when beads were passivated with pluronic acid, which prevents any unspecific residual adhesion. We further observed that TRM moved around dense bead areas, in line with a search for permissive gaps for locomotion (movie S20). In sum, our data uncover the remarkable ability of resting SMG TRM to migrate by adapting to topographic features of the environment through protrusion insertion and shape deformation, even in the absence of considerable friction, chemoattractants, and adhesion receptors (Fig. 8I). At the same time, these cells retain responsiveness to chemokine signals, follow macrophage topology to short-cut SMG epithelial barriers, and accumulate at local inflammatory hot spots (Fig. 8J).


After clearing pathogens, TRM display a remarkable capacity to patrol heterogeneous tissues without impairing vital organ functions (59). Their scanning behavior evolved because T cells are major histocompatibility complex–restricted and hence need to physically probe membrane surfaces of immotile stromal cells. In this study, we examined how these cells achieve this feat in the complex arborized epithelial structure of SMGs during homeostatic immune surveillance. Our main finding is that TRM preferentially moved along tissue macrophages, and depletion of macrophages impaired TRM patrolling. These observations assign a new accessory role to tissue macrophages in addition to their core functions for tissue homeostasis and sentinels of infection. Our data suggest two nonexclusive options to explain macrophage guidance of TRM: first, through unidentified specific adhesive interaction(s) independent of ICAM-1 and other canonical adhesion molecules such as CD103 and, second, by offering paths of least resistance within the exocrine gland microenvironment for protrusion insertion by autonomously moving T cells. Our data provide support for the second option without discarding the first one. Thus, although SMG TRM respond to exogenous cues from chemoattractant and adhesion molecules, confinement alone suffices to trigger their friction- and protrusion insertion–based motility. The continuum of intrinsic motility and integration of external factors may permit TRM to patrol these exocrine glands in homeostasis and rapidly respond to inflammatory stimuli.

Macrophages and T cells closely cooperate during the onset of inflammation, the effector phase, and contraction through antigen presentation, cytokine secretion, and effector functions such as phagocytosis. Yet, little is known whether and how these two cell types collaborate for surveillance of NLT during homeostasis. Tissue macrophages are best characterized for their core function of maintenance or restoration of tissue homeostasis by engulfing apoptotic cells, clearing debris, initiation of repair, and cloaking of microlesions (2831, 46). Furthermore, tissue macrophages serve as sentinels of infection, leading to cytokine secretion and leukocyte recruitment (5, 47, 48). In recent years, several nonphagocytic and nonsentinel functions were assigned to macrophages, because it was recognized that core functions of parenchymal parts of organs became outsourced to accessory cells. Accessory macrophage functions include blood vessel and mammary duct morphogenesis, hematopoietic stem cell maintenance, pancreatic cell specification, lipid metabolism, relay of long-distance signals during zebrafish patterning, and electric conduction in the heart (49, 50). Our data suggest a distinct accessory function, which is to facilitate TRM patrolling within and between acini and ducts of arborized secretory epithelium. Our initial assumption was that specific adhesion receptors drive T cell association with tissue macrophages, whereas chemoattractants fuel their high baseline motility. It was therefore startling that we were unable to identify molecules mediating strong adhesive contacts between salivary gland macrophages and TRM. Our data do not rule out the presence of specific adhesive and/or promigratory interactions between TRM and tissue macrophages in situ. For instance, low TRM binding to tissue macrophages in vitro may be owing to altered gene expression patterns after macrophage isolation (51). In addition, we have not examined talin-deficient T cells lacking functional integrins, because these cells are unable to attach to endothelium for extravasation. Poor surface mAb saturation preempted a complete analysis of CD44 for SMG TRM motility (52). Last, PTx treatment induced a minor reduction in TRM speeds in vivo. Yet, PTx treatment had essentially no impact on U-turn frequency and movement along tissue macrophages. In line with this, recent observations suggest that guidance and adhesion do not necessarily correlate because TN migrate along the PLN stromal network in the absence of LFA-1 and CCR7 (41). Last, the fact that isolated TRM can move between polystyrene beads or other lymphocytes for efficient translocation suggests that any structure with permissive gaps can serve for topological guidance. Nonetheless, owing to a lack of a suitable experimental system, our data do not provide unambiguous evidence for or against specific molecular interactions between TRM and macrophages in vivo. Therefore, we cannot definitely clarify to which extent specific interactions and microenvironmental architecture contribute to macrophage-TRM association in situ.

The canonical model of leukocyte migration postulates chemoattractant-stimulated F-actin polymerization at the leading edge (53). The resulting retrograde F-actin flow, in turn, generates traction and cell body translocation via an integrin “clutch” that binds to adhesion receptors of the ECM or on the surface of neighboring cells. Our observations confirm that similar to TN and TPLN-M, ex vivo confined epidermal TRM do not migrate in the absence of integrin ligands or chemoattractants. Spontaneous motility under 2D confinement appears to constitute a distinctive hallmark of SMG TRM not shared by other resting T cells. Low adhesiveness under confinement induces spontaneous amoeboid motility via cortical contractility in adherent mesenchymal cell lines (54, 55), suggesting that TRM may use a similar mechanism for autonomous migration in vitro and in vivo. Yet, it remains currently unknown how this unique motility program is imprinted in SMG TRM and whether it is shared by tissue-resident cells from other exocrine glands. TRM regained the capability to translocate in the presence of EDTA when narrow spaces are created by immotile neighboring cells or beads that lack strong adhesion to each other. Our data suggest that TRM protrusions insert into gaps of the 3D environment akin to cogs of a cogwheel and transmit the necessary force for translocation through retrograde actin flow along irregularly shaped surfaces, even in the absence of adhesive interactions. The flexible anchorage of macrophage protrusions between epithelial cells may facilitate the insertion of F-actin–rich pseudopods by TRM before squeezing of the nucleus as the biggest organelle (56). This migration mode preserves tissue integrity and is energetically favorable by avoiding ECM degradation (57).

Our local CXCL10 deposition experiment suggests that macrophages facilitate local TRM accumulation at sites where inflammatory chemokines are produced. This resembles observations made in skin infection models where CXCR3 promotes CD8+ T cell accumulation at sites of viral replication necessary for efficient elimination of infected cells (2, 3). A recent study by Förster and colleagues has uncovered an unexpectedly low killing rate of cytotoxic T cells against viral-infected stromal cells (58). Thus, effective stromal cell elimination requires cooperativity through repeated cytotoxic attacks by multiple CD8+ T cells. Conceivably, the promigratory accessory function of tissue macrophages described here helps to cluster a quorum of TRM for successful stromal cell killing. Furthermore, unlike the monoclonal TRM population created in our experimental setting, not all TRM recognize the same pathogen under physiological conditions. This imposes a requirement for T cells to scan local sites of pathogen reemergence and to form clusters for timely elimination of fast-replicating microbes. In sum, our data assign a previously unnoticed cooperativity between tissue-resident innate and adaptive immune cell populations. These findings uncover a noticeable capacity of SMG TRM to integrate a continuum of intrinsic and external signals, friction, and 3D structures for efficient motility, providing these cells with maximal flexibility for NLT surveillance. We propose that such a mode of tissue patrolling is ideally adapted to the arborized epithelial architecture of exocrine glands by permitting homeostatic surveillance while maintaining responsiveness to local inflammatory cues.


Study design

The objective of this study was to assess the mechanism of homeostatic CD8+ TRM surveillance in SMG. We developed a procedure for intravital imaging of SMG to visualize dynamic TRM motility in the memory phase after a systemic virus infection, in the presence or absence of tissue macrophages and blocking reagents. This was completed by in vitro motility assays of isolated SMG TRM under confinement. Detailed descriptions of the experimental procedures can be found below and in the Supplementary Materials.


OT-I TCR (22) and P14 TCR tg mice (25) were backcrossed to tg(UBC-GFP)30Scha “Ubi-GFP” (59) or hCD2-dsRed (60) mice. Ubi-GFP (GFP+) OT-I mice backcrossed to CXCR3−/− mice have been described (61). tg(Itgax-Venus)1Mnz CD11c-YFP (62) and tg(Itgax-DTR/EGFP)57Lan CD11c-DTR mice were used as recipients or bone marrow donors for lethally irradiated C57BL/6 or Ubi-GFP mice. C57BL/6 mice were purchased from Janvier (AD Horst). All mice were maintained at the Department of Clinical Research animal facility of the University of Bern, at the Theodor Kocher Institute and the University of Fribourg. All animal work has been approved by the Cantonal Committees for Animal Experimentation and conducted according to federal guidelines.

T cell transfer and viral infections

CD8+ T cells were negatively isolated from the spleen and peripheral and mesenteric lymph nodes of GFP+ or dsRed+ OT-I or GFP+ P14 mice, using the EasySep Mouse CD8+ T cell Isolation Kit (STEMCELL Technologies). CD8+ T cell purity was confirmed to be >95% by flow cytometry before cell transfer. A total of 104 OT-I T cells were intravenously transferred into recipient mice 24 hours before intraperitoneal infection with 105 plaque-forming units (pfu) of LCMV-OVA (21). For epidermal TRM, we injected 5 × 104 OT-I as above, and 15 μl of 0.3% DNFB (in 4:1 acetone/oil) was applied to the right flank on day 3 p.i., followed by 500 ng of SIINFEKL peptide at days 4 and 5 p.i.

Flow cytometry analysis

PLN and spleen were harvested and passed through cell strainers (70 μm; Bioswisstec). For analysis of SMG and LG, organs were minced and treated with collagenase II (2 U/μl; Worthington Biochemical), bovine deoxyribonuclease I (2 U/μl; Calbiochem), and, for intracellular stainings of cytokines, brefeldin A (5 μg/ml; B6542, Sigma-Aldrich) in complete medium RPMI [RPMI 1640, 10% fetal bovine serum (FBS), 1% Hepes, 1% penicillin-streptomycin, 2 mM l-glutamine/1 mM sodium pyruvate] for 30 min at 37°C; passed through a 70-μm cell strainer; and washed with phosphate-buffered saline (PBS)/5 mM EDTA. As gating strategy, at least 105 cells in the lymphocyte FSC/SSC (forward scatter/side scatter) gate were acquired using a FACSCalibur (BD Bioscience), LSR II (BD Bioscience), LSR II SORP Upgrade (BD Bioscience), or Attune NxT Flow Cytometer (Thermo Fisher Scientific). Total cell counts were obtained by measuring single-cell suspensions in PKH26 reference microbeads (Sigma-Aldrich). Gating for CD103+ and KLRG1+ was set according to isotype controls. For CD69 staining, positive and negative gates were set according to distinguishable populations, and fluorescence minus one was subtracted from the final percentage of CD69+ cells as background. The gating strategy for SMG macrophages is depicted in fig. S2A. The mAbs used are listed in the Supplementary Materials.

Two-photon intravital microscopy

2PM intravital imaging of the popliteal lymph node and the SMG was performed as previously described (24, 63). In most experiments, mice were operated twice (for PLN and SMG) in alternating order to directly compare behavior of cells in different organs of the same recipient. Before imaging, blood vessels were labeled by intravenous injection of 400 to 600 μg of 10-kDa Cascade Blue dextran or 70-kDa Texas Red dextran. Surgical exposure of the LG was essentially performed as for the SMG, with the mouse fixed on its left flank onto the custom-built SMG imaging stage and a 10 mm by 5 mm piece of skin excised between the right ear and eye of the mouse. Inhibitor treatment and image analysis are described in the Supplementary Materials.

Under-agarose assays

Single-cell suspensions of SMG and PLN from LCMV-OVA–infected C57BL/6 mice were stained with allophycoyanin-conjugated anti-KLRG1 mAb and sorted for GFP+ or DsRed+ KLRG1 TRM and TPLN-M, respectively. Under-agarose assays were performed as described in (41) and the Supplementary Materials.

Correlative confocal and transmission electron microscopy

CD11c-YFP mice were perfused with PBS and SMG were fixed in situ by left ventricle injection of 1.5% glutaraldehyde/2% paraformaldehyde in 0.1 M sodium cacodylate buffer (pH 7.4). SMG were harvested and immersed in the same solution for 16 hours. Fixed samples were cryoprotected in 30% sucrose before embedding in optimum cutting temperature and freezing. Thirty-micrometer sections were processed for confocal imaging. After confocal image acquisition, coverslips were gently removed, and sections adherent to the slide were processed for TEM as previously described (64).

Super-resolution shadow imaging

SMG were embedded in 4% low gelling agarose (Sigma-Aldrich), cut in 300-μm-thick transversal slices, and submerged in cold complete RPMI medium containing 10% FCS (HyClone) before entering the imaging chamber of a custom-built 3D-STED (stimulated emission depletion microscopy) microscopy setup (33). CD11c-YFP+ macrophages were identified at a depth of 20 to 30 μm below the surface and imaged in STED mode [excitation, 485 nm; depletion, 597 nm; objective HC PL APO 63×/1.30 numerical aperture (NA); Leica]. The medium was exchanged in the chamber with the complete RPMI containing 400 μM calcein dye, which was allowed for 20 to 30 min to disperse throughout the ECS of the tissue. Subsequently, we acquired a SUSHI image to identify a region of interest around macrophages. We performed a hyperosmolar challenge by exchanging the chamber solution with high osmolar solution (350 mosM/liter) and acquired time-lapse images to track changes in ECS topology with a 20-min interval between image frames.

Chemokine-driven TRM accumulation

CD11c-YFP or CD11c-DTR bone marrow chimera received 104 GFP+ OT-I T cells and were infected the day after with 105 pfu of LCMV-OVA. After ≥30 days p.i., mice were anesthetized 1 day after intraperitoneal injection of DTx (4 ng/g). To block immigration of cells from blood, we treated mice with integrin-blocking antibodies anti-αL (FD441.8) and anti-α4 (PS/2) (50 μg per mouse; nanotools). Using thin glass capillaries, we injected 2 μl of a 1:1 mix of mCXLC10 (466-CR-010, R&D Systems) and Qdots655 (0.16 μM; Q2152MP, Thermo Fisher Scientific) into exposed SMG for a final mCXCL10 amount of 0.5 μg per site of injection. After 4 hours, SMG were harvested for vibratome sectioning. Mosaic images of 100-μm-thick sections were taken, and lobes with the highest Qdot signal were analyzed by transforming the 3D image into extended 2D image (ImageJ). TRM density in the surrounding area and injection area (defined as octagon with a diameter of 500 μm) was calculated using Imaris 8.4.1.

Statistical analysis

Two-tailed, unpaired Student’s t test, Mann-Whitney U test, one-way analysis of variance (ANOVA) with Dunnett’s multiple comparisons test, Kruskal-Wallis test, or a Wilcoxon rank test was used to determine statistical significance (Prism, GraphPad). Significance was set at P < 0.05.


Materials and Methods

Fig. S1. Systemic viral infection leads to memory CD8+ T cell populations in secondary lymphoid organs and SMG for comparative analysis.

Fig. S2. Characterization of CD11c-YFP+ macrophages.

Fig. S3. LG TRM motility parameters after macrophage depletion.

Fig. S4. Tissue macrophage–mediated efferocytosis during viral infection of SMG.

Fig. S5. Analysis of integrins and TPLN-M blockade in vivo.

Movie S1. Intravital imaging of TPLN-M migration in PLN.

Movie S2. Intravital imaging of TRM migration in SMG.

Movie S3. Confocal image of SMG TRM colocalization with tissue macrophages.

Movie S4. Intravital imaging of TRM migration in the absence of macrophages.

Movie S5. Intravital imaging of TRM migration along a macrophage to enter acini.

Movie S6. Intravital imaging of TRM migration in the absence of macrophages after 5 days of DTx treatment.

Movie S7. Time-lapse SUSHI of SMG section during hyperosmotic challenge.

Movie S8. Confocal image of tissue macrophage protrusions and basement membranes in SMG.

Movie S9. Intravital imaging of TRM migration and association with macrophages after blocking of αL, α4, and αE integrins.

Movie S10. Intravital imaging of TRM migration and association with macrophages after treatment with PTx.

Movie S11. Time-lapse imaging of TN migration under in vitro confinement on CCL21/ICAM-1–coated plates.

Movie S12. Time-lapse imaging of TRM migration under in vitro confinement on CXL10 + CXCL12/ICAM-1–coated plates.

Movie S13. Time-lapse imaging of TRM and TN migration under in vitro confinement on HSA-coated plates.

Movie S14. Time-lapse imaging of TRM and TPLN-M migration under in vitro confinement.

Movie S15. High temporal resolution analysis of TRM motility under in vitro confinement.

Movie S16. Time-lapse imaging of TRM and tissue macrophages under in vitro confinement.

Movie S17. Time-lapse imaging of SMG and epidermal TRM under in vitro confinement.

Movie S18. Time-lapse imaging of TRM migration in the presence of EDTA.

Movie S19. Time-lapse imaging of TRM migration inside TN cluster in the presence of EDTA.

Movie S20. Time-lapse imaging of TRM migration along polystyrene beads in the presence of EDTA.

Data S1. Raw data for all figure graphs (Excel).

References (6567)


Acknowledgments: We thank M. Thelen (IRB, Bellinzona) for support with confocal imaging. This work benefitted from optical setups of the Microscopy Imaging Center of the University of Bern and of the BioImaging platform of the University of Fribourg. Funding: This work was funded by Swiss National Foundation (SNF) project grants 31003A_135649, 31003A_153457, and 31003A_172994 (to J.V.S.); Leopoldina fellowship LPDS 2011-16 (to B.S.); the Deutsche Forschungsgemeinschaft (German research foundation, DFG)—project number 240245660—SFB1129 (project 8) (to B.S. and O.T.F.) and SFB900 (to K.A.K.); and the Novartis foundation fellowship 16C193 (to F.T.). P.G. and J.S. acknowledge support of the Spanish Ministry of Economy and Competitiveness, “Centro de Excelencia Severo Ochoa 2013-2017,” and support of the CERCA Programme/Generalitat de Catalunya. Author contributions: B.S., F.T., and X.F. performed most experiments with support by L.M.A. and N.R. L.M.A. and V.V.G.K.I. carried out SUSHI imaging under the supervision of U.V.N. P.G. carried out computational analysis under the supervision of J.S. A.R. and F.M. performed correlative electron microscopy of SMG sections under the supervision of M.I. N.P., K.A.K., F.B., D.M., and O.T.F. provided vital material and support. S.M.S.J., M.S.D., and C.S. analyzed human SMG sections. B.S., F.T., X.F., L.M.A., N.R., and J.V.S. designed experiments, performed statistical analysis, and wrote the manuscript with input from all coauthors. Competing interests: D.M. is an inventor on patent application (EP3218504A1) submitted by the University of Geneva that covers “Tri-segmented arenaviruses” as vaccine vectors. All other authors declare that they have no competing interests. Data and materials availability: The data that support the findings of this study are listed in the figures, the Supplementary Materials, and the raw data file. Questions and requests for noncommercially available reagents and mouse strains can be made to J.V.S.

Stay Connected to Science Immunology

Navigate This Article