Subtle CXCR3-Dependent Chemotaxis of CTLs within Infected Tissue Allows Efficient Target Localization

This article is featured in In This Issue, p.5097

A striking characteristic of the cells of the immune system is their ability to travel throughout the organism to guard peripheral tissues and to prevent the spread of infection. First, APCs acquire Ag within peripheral tissues and travel to draining lymph nodes to present the collected Ags to naive T cells (1, 2). Ag-specific T cells that are activated by APC encounter then initiate expression of effector molecules, such as perforin and granzymes, and enter the blood stream via the efferent lymph. Extravasation of such activated T cells preferentially takes place at sites of infection, through regulation of adhesive interactions with the vascular endothelium (3). Several studies have examined the interaction between CTLs and target cells at sites of infection (4–6). In another recent study by Hickman et al. (7), it was shown that expression of the chemokine receptor CXCR3 on T cells assists in the localization, killing, and control of vaccinia virus (VV) infection. However, the exact search strategy by which CXCR3 expression helps T cells to detect virus-infected targets has not yet been elucidated. On the one hand, this could be obtained by chemokinesis (i.e., a general increase of cellular motility that occurs upon encounter of specific chemokine ligands). For example, such chemokinetic effects occur for CXCR7-expressing T cells within lymph nodes (8). On the other hand, CTLs could migrate directionally toward sites of infection by CXCR3-mediated chemotaxis, a process that should also reduce the time with which established sites are reached. A large number of studies have described the migration patterns of T and B lymphocytes within lymph nodes (reviewed in Ref. 9), and evidence for such directed migration has been obtained in several of them (10–14). Within peripheral tissues (i.e., subsequent to extravasation), it was recently proposed that the observed CD8⁺ T cell clustering around malaria-infected hepatocytes is consistent with directed migration (15), but direct evidence for chemotaxis was not obtained. On the contrary, other recent literature suggests that CTLs in pancreas and brain adopt random search strategies (16, 17).

In this study, we develop and use a skin infection intravital imaging setup in which the epidermis of recipient animals is infected with HSV-1. Imaging of virus-specific and bystander CTLs within the epidermis of infected mice indicates that both T cell populations migrate through the epidermis without any directional preference that can be appreciated by direct observation. Importantly though, quantitative analysis of the data and subsequent computer simulations based on these experimental data demonstrate that there is a small preference to migrate toward sites of infection. As expected, and on the basis of the recent findings by Hickman et al. (7), this preference depends on CXCR3 expression on T cells and can be influenced by disturbance of CXCR3 ligands. Furthermore, this preference translates to a marked increase in the efficiency with which effector T cells accumulate at the sites of infection. Thus, our results reveal how subtle chemotaxis of cytotoxic T cells within the periphery allows the efficient localization of clusters of pathogen-infected cells during an antiviral immune response.

C57BL/6, C57BL/6 actin-GFP, and C57BL/6 OTI TCR transgenic mice were obtained from Charles River Laboratories and housed in the animal facility of the Netherlands Cancer Institute (NKI). C57BL/6 gBT TCR transgenic mice (18) were a gift from Dr. F. Carbone (University of Melbourne, Australia). All animal experiments were approved by the Experimental Animal Committee of the NKI, in accordance with national guidelines.

Abs against MHC II, CD4, CD11b, CD11c, Ter-119, CD49b, CD45R, Gr-1, and CD8 were obtained from BD Biosciences. MHC-tetramers loaded with SIINFEKL (Ova_257–264) or SSIEFARL (gB_498–505) peptides were produced by UV-induced ligand exchange (19) and labeled with PE or APC (Invitrogen).

DNA vaccines were generated by the introduction of the indicated fusion genes in pVAX. Tetanus Toxin Fragment C (TTFC)-Ova and TTFC-gB DNA sequences were designed for optimized codon usage, and encoded domain I of TTFC and the indicated T cell epitope, separated by a GVQI peptide linker. HSV_TOM and HSV_TOM-OVA viruses were constructed by the insertion of a CMV immediate-early promoter-tomato (or tomato-Ova_257–264) gene cassette into the intergenic region between the UL26 and UL27 genes of the HSV-1 strain KOS, as described previously (20).

For the measurement of T cell responses, 25 μl of peripheral blood was collected in heparin-coated vials (Microvette; Omnilabo) at the indicated days after transfer. After removal of erythrocytes by NH₄Cl treatment, cells were stained with the indicated Abs and analyzed by flow cytometry. Dead cells were excluded by DAPI staining. Cell sorting was performed on a FACSAria cell sorter using FACSDiva software (version 5.0.1; both from BD), or a MoFlo high-speed sorter using Summit software (version 3.1; both from Beckman Coulter). Analyses were performed on a CyAn_ADP using Summit software (version 4.3; both from Beckman Coulter).

Spleens from GFP-gBT or GFP-OTI mice were isolated and homogenized by mincing through cell strainers (BD Falcon). After RBC lysis, CD8⁺ T cells were isolated using MHC II⁺, CD4⁺, CD11b⁺, CD11c⁺, Ter-119⁺, CD49b⁺, CD45R⁺, and Gr-1⁺ staining as exclusion criteria. C57BL/6 mice were injected i.v. with 2.0 × 10⁵ sorted naive CD8⁺ T cells. Intradermal DNA vaccination with either TTFC-Ova or TTFC-gB was performed on the shaved hind leg of anesthetized mice, as described previously (21). A droplet of 30 μg DNA in 15 μl PBS was applied onto the skin and injected using a sterile disposable 11-needle bar mounted on a rotary tattoo device. Needle depth was adjusted to 0.5 mm, and the needle bar oscillated at 100 Hz for a 60-second tattoo.

Indicated HSV strains were grown in Vero cells as described previously (22), and intraepithelial HSV-1 infection was completed on the shaved flank of anesthetized mice. A droplet of 20 μl virus-infected Vero cell supernatant containing ∼2.5 × 10⁵ PFU was applied onto the skin and introduced as described for intradermal vaccination, but with needle depth adjusted to 0.05 mm, for a 20-second tattoo. At the indicated time points postinfection, mice were sacrificed by cervical dislocation; the skin of the infected flank was collected and divided into two separate specimens that were further treated independently. HSV titers (Artus HSV-1/2 TM PCR Kit; Qiagen) were analyzed in triplicate on each half sample, according to protocol. HSV infection was visualized by incubation with rabbit polyclonal anti HSV Ab (Thermo Scientific), followed by PowerVision Poly-HRP-Goat anti Rabbit staining (ImmunoLogic) and DAB reaction (Dako Cytomation), according to protocol.

Twenty-four hours after HSV infection, mice were anesthetized using a mixture of air and isoflurane (Baxter) and placed in a custom-built chamber with the infected flank gently placed against a coverslip at the bottom side of the chamber. Images were acquired using an inverted Leica TCS SP2 confocal scanning microscope (Leica Microsystem) equipped with Diode and Argon lasers and enclosed in a custom-built environmental chamber that was maintained at 37°C using heated air. Images were acquired using a 20×/0.7 NA dry objective unless indicated otherwise. GFP was excited at 488 nm wavelength and tandem dimer tomato at 561 nm. Typical voxel dimensions were 0.9–2.0 μm laterally × 1.5–2.5 μm axially. Three-dimensional stacks (typical size: ∼500 × ∼500 × ∼30 μm) were captured every minute for a period of up to 2 h. Raw imaging data were processed with Imaris (Bitplane). Imaris Spots module was used to track objects and to calculate cell coordinates (mean positions) over time (representative track in Supplemental Video 4). The resulting tracks were manually checked for correctness, and any track part that was uncertain from the viewpoint of a human observer was excluded. Although this approach results in the splitting of long tracks into short ones, it does not influence the predominantly step-based approaches that are used for subsequent analysis (23).

The cell coordinate data that resulted from tracking were corrected for artifacts as described previously (23). In brief, presumed static auto-fluorescent hair follicles or the infection region itself were used for tissue drift correction, and accuracy of drift correction was checked by visual inspection of videos. Movement steps close to the borders of the image volume were discarded from analysis, because they would appear to move parallel to those borders because of tracking errors (note that an advantage of two-dimensional [2D] over three-dimensional [3D] analyses is that a much smaller number of movement steps need to be discarded for this reason). Finally, imprecision in the axial resolution was either left uncorrected (data not shown) or corrected by multiplying the mean axial speed of cells by a factor (varying between 1.2 and 2.95 depending on the experiment) that made it approximately equal to their mean lateral speed. Because of the flat structure of the epidermis, it is possible that axial speeds are in fact much slower than lateral speeds. Omitting the axial speed correction would entail that the lateral movements are much more important than the axial movements in determining arrival at microlesions (i.e., that migration is effectively 2D). Importantly though, the observed directional preference of CTLs toward infected microlesions (see below) was observed both with and without the correction for axial speed.

After correction for these artifacts, the following parameters were calculated from the cell coordinates: 1) cell speeds (during individual movement steps), 2) turning angles (i.e., the angle between subsequent movement steps of a cell or between longer time intervals for a single cell), 3) distances to infection (i.e., distance between the center of mass of a cell and the location of the most nearby voxel that was assigned to the infection when using a segmentation threshold following application of a Gaussian filter to smoothen the images), 4) angles to infection (i.e., the angle between the direction of migration of a cell and the direction toward the nearest voxel assigned to the infection), and 5) displacement rates toward infection (calculated from the projection of a movement step onto the vector that represents the shortest route toward the nearest voxel assigned to the infection). Note that all the studied migration parameters are averages measured over 1-min time frames.

Cell motion was modeled by describing it as a series of discrete steps in space. Simulations were performed in two or three dimensions, and either as directed or random simulations. For directed simulations, the experimental distributions of speed, angle to infection, and turning angle were used after merging the data from four to five experiments that were performed under the same conditions. Measured pairs of speed and angle to infection were maintained during directional simulations. In random simulations, the angles to infection were chosen from a distribution as expected for random migration, i.e., a uniform distribution in 2D and a sine distribution in 3D. The distributions of speed and angle to infection were sorted according to the distance to infection at which the measurement was taken. Subsequently, the speed/angle-to-infection pairs were binned into distance categories containing 100 measurements each. Each of the distance bins would also be accompanied by a turning angle distribution (containing all the turning angles experimentally measured in that distance bin). In the simulations, a microlesion was modeled as a circle (in 2D) or sphere (in 3D) with radius r. Each in silico cell was initialized at the indicated distance from the simulated infection, and the initial movement step of the cell was chosen at random from the 100 speed/angle-to-infection combinations in the appropriate distance category. In 2D simulations, each combination is associated with two possible movement steps, and one of them would be chosen at random. In 3D simulations, the possible movement steps for each combination can be described by a circle in 3D space, and a random position on that circle was chosen. The cell was then moved to the new position defined by the chosen step. Subsequent movement steps were again chosen from the allowed speed/angle-to-infection pairs. However, if the movement steps were to be chosen randomly from the pairs, a turning angle distribution without any persistence would ensue. In particular, this would lead to a turning angle distribution close to a uniform distribution for 2D simulations and close to a sine distribution for 3D simulations (23). Therefore, we built in the extra restriction that a chosen movement step was accepted with a probability that would ensure a similar turning angle distribution in the simulations as in the experimental data. To achieve that, the turning angle distribution of each distance category was discretized into 18 bins of 10 degrees wide, and the proportion p_i of each turning angle bin was recorded for all distance categories (where i represents the turning angle bin number from 0 to 18). For 2D simulations, we took the proportion of the turning angle bin in which the step under consideration would fall (p_i) as the acceptance probability for that step. For 3D simulations, the proportion of each turning angle bin was first weighted by the factor $wi=1/∫α=absin(α)dα=1/(cos(a)−cos(b))$⁠, where a and b are the lower and upper limit of a turning angle bin (e.g., for the first bin a = 0 and b = 10 degrees). Next, the weighted and normalized proportion $piwi∑ipiwi$was used as acceptance probability for the step. This normalized weighting of proportions ensured that speed-angle-to-infection combinations were sampled in a manner that resulted in a correct turning angle distribution. When a movement step was not accepted, another movement step would be chosen at random until one was accepted.

We performed two types of 3D simulations: one in which CTL migration was restricted at the exterior side, and one in which it was in addition restricted at the interior side. These restrictions were implemented by discarding movement steps that would result in a crossing of the interior or exterior side during the sampling of movement steps.

2D simulations of long-term (up to 12 h) T cell accumulation were performed. Infection growth was described as linear, radial growth following experimental measurements of infection sizes, and directed and random cell migration were simulated as described above. For bystander OTI cells, we considered the possibility that T cells reaching the infection from that moment could leave the simulated region (in the axial direction, i.e., by leaving the epidermis). In these simulations with leaving, T cells that had earlier reached the infection, left with probability δ at each time step.

A linear mixed effects model was constructed to determine whether viral load differed between treatments over the week after inoculation. Joint Wald tests were performed on linear and quadratic coefficients, and Bonferroni-adjusted significance level for the two-sided test was set at 0.017.

Because of persistent migration of T cells at short time intervals, subsequent movement steps are not independent. Therefore, statistics were applied to steps that were at least 10 min apart, because for skin T cells the correlation in movement direction was approximately lost after this time interval (see Fig. 2C). Two-tailed Mann–Whitney tests (using the software package R) were performed to study whether cell migration differs from a random walk, and to compare migration parameters in different conditions. Spearman’s rho between distance from infection and the number of cells at that distance was calculated to study whether spatial distributions exhibited a consistent decline with distance from infection, as expected for chemoattraction toward foci of infection. A p value <0.05 was considered significant.

To detect HSV-1–infected cells within an otherwise normal tissue in vivo, we generated HSV-1 recombinant viruses that either express the tandem dimer tomato fluorescent protein alone (HSV_TOM), or in combination with the Ova_257–264 CD8⁺ T cell epitope (HSV_TOM-OVA). When C57BL/6 mice were locally infected in the skin with either recombinant HSV-1 strain, multiple small foci composed of dozens of infected keratinocytes could be detected in the skin by immunohistochemistry and intravital imaging within 24 h postinfection (Fig. 1A, 1B). Consistent with prior data (16), infected microlesions were confined to the epidermal layer (Fig. 1A). Over a period of 3–4 d, these HSV-induced lesions fused together to form vesicles (Supplemental Fig. 1A) that were also noticeable by visual inspection (Fig. 1C).

Control of HSV-1 infection depends on T cell immunity, as shown by the fact that infection of T cell–deficient mice is lethal (20). To determine whether CD8⁺ T cells also have a role in the control of infection in our recombinant HSV model, we transferred GFP-transgenic gBT I.1 TCR transgenic T cells and boosted their numbers by vaccination before infection with HSV_TOM (Supplemental Fig. 1B, 1C). Importantly, skin viral titers in HSV_TOM-infected mice harboring vaccine-activated GFP-gBT cells were markedly reduced (p < 0.001 relative to controls; Fig. 1D). T cell control of epithelial HSV infection was also apparent by macroscopic appearance (Fig. 1C).

Analysis of viral titers at different time points postinfection demonstrated that HSV-1 specific effector T cells displayed the most pronounced effect on HSV infection between 24 and 48 h of virus exposure (Fig. 1D), and subsequent imaging was performed at 24 h postinfection, before the occurrence of blisters and scabs that impede intravital imaging.

Imaging of HSV_TOM-infected skin demonstrated that effector T cell infiltration occurred for both HSV-specific and bystander (OTI) T cells (Fig. 2A, 2B; Supplemental Videos 1–3). To subsequently visualize the movement of effector T cells within the epidermis, infected skin areas infiltrated with GFP⁺CD8⁺ T cells were imaged at 1-min intervals for up to 1 h (without any measurable phototoxicity, as measured by speed over time; Supplemental Fig. 2). Whereas both gBT and OTI T cells actively moved around HSV_TOM-infected microlesions with a persistence time of 5–10 min (Supplemental Videos 1–3; Fig. 2C), only gBT cells arrested in close proximity to sites of infection, as demonstrated by decreased speed and loss of persistent migration when comparing steps close to (0–25 μm) and far from (>25 μm) the infection (Fig. 2D, 2E; gBT + HSV_TOM: p = 0.000 [speed close versus far]; p = 0.000 [turning angles close versus far]; Mann–Whitney U test; 3D analysis in Supplemental Fig. 3A, 3B). This T cell arrest was due to Ag recognition (24), as OTI T cells in the skin of mice infected with HSV_TOM-OVA also decreased their speeds close to the site of infection (OTI + HSV_TOM-OVA: p = 0.000 [speed close versus far]; p = 0.000 [turning angles close versus far]; Mann–Whitney U test), whereas OTI bystander T cells even slightly sped up close to HSV_TOM infections (OTI + HSV_TOM: p = 0.002 [speed close versus far] and p = 0.64 [turning angles close versus far]; Mann–Whitney U test). Note that Ag-specific T cell arrest took place at up to a few dozen micrometers away from the estimated infection border, suggesting that T cells can detect Ag expressed by newly infected cells before the fluorescent signal becomes apparent.

To determine by which migration strategy effector T cells reach foci of infection after tissue entry, we quantified the angle to infection (i.e., the angle between the measured travel direction and the shortest route toward the infection) for each movement step. The distribution of all the measured angles to infection was close to a uniform distribution (Fig. 2F). However, despite this highly erratic migration of T cells, there was a small bias toward acute angles (<90 degrees), suggesting that T cells had a small preference to migrate toward the infection. To study whether this effect was dependent on the distance from infection, average angles to infection were plotted as a function of distance to the infection (23). At the vicinity of the infection (0–25 μm), angles to infection were not significantly different from 90 degrees (Fig. 2G; gBT + HSV_TOM: p = 0.51; OTI + HSV_TOM: p = 0.32; OTI + HSV_TOM-OVA: p = 0.59; Mann–Whitney U test; 3D analysis in Supplemental Fig. 3C), implying random migration. In contrast, migration angles of T cells located further away from the infection (>25μm) were significantly smaller than 90 degrees (gBT + HSV_TOM: p = 0.003; OTI + HSV_TOM: p < 0.001; OTI + HSV_TOM-OVA: p = 0.003; Mann–Whitney U test). Thus, while T cells around HSV-infected lesions do migrate in all directions, a small but highly significant preference for directions that bring the cell closer to the site of infection is observed. Note that the angular preference at different distances is not a precise, fixed cellular behavior. Rather, the highly erratic migration that is superimposed over the preferential migration (Fig. 2F) gives rise to fluctuations even when many measurements are averaged as in Fig. 2G. The distance-dependent preferential migration was similar for Ag-specific and bystander T cells (Fig. 2G and Supplemental Fig. 3C; differences between mean angles to infection were nonsignificant for all comparisons; Mann–Whitney U test). Thus, the attracting component in T cell migration is independent of cognate Ag. Finally, when T cell migration was analyzed as the average displacement rate of cells toward the infection as a function of distance from infection (Fig. 2H; 3D analysis in Supplemental Fig. 3D), a distance-dependent pattern of attraction toward the site of infection was observed for all three conditions tested. Specifically, whereas close to the infection (up to 25 μm away) displacement rates toward infection were not different from 0 μm/min (gBT + HSV_TOM: p = 0.60; OTI + HSV_TOM: p = 0.42; OTI + HSV_TOM-OVA: p = 0.57; Mann–Whitney U test), further away from the infection (>25 μm away) an average displacement rate of 0.3–0.4 μm/min—significantly different from zero (gBT + HSV_TOM: p = 0.002; OTI + HSV_TOM: p < 0.001; OTI + HSV_TOM-OVA: p = 0.027; Mann–Whitney U test)—was observed. In summary, effector T cells in the epidermis migrate in all directions, but with a small, Ag-independent preference for movement toward infected areas.

From our cell migration analysis, it is apparent that the strength of attraction (the deviation in angles to infection from 90 degrees) varies between experiments. We considered the possibility that the size of HSV foci may affect the strength of the chemoattractant gradient, such that larger foci of infection may attract CTLs more strongly. We therefore quantified the size of the infection at the beginning of each experiment and measured the strength of directed migration, by calculating the mean of angles to infection up to a distance of 200 μm. The relationship between these two measures shows that effector CD8⁺ T cells are more strongly attracted toward large than toward small microlesions (Fig. 3A), although there appears to be a saturation effect for large infections.

Since microlesion size influences attraction strength (Fig. 3A), and attraction strength influences effector T cell arrival at microlesions, we predicted that for large foci of infection a larger fraction of CTLs should be located close to the microlesion than for small foci. We quantified the spatial distribution of effector T cells at the beginning of each experiment by calculating the median of their distances to the infection. As predicted, for large foci, CTLs were located at shorter median distances than for small infections (Fig. 3B). These data indicate that the volume of the infected microlesion correlates with the strength of effector T cell attraction, and—presumably because of this —the spatial distribution of these cells around the lesion. Note that these data are also consistent with a scenario in which T cells arriving early at infection foci secrete chemokines upon target encounter, thereby reinforcing the attraction of yet more effector T cells (in a similar fashion, specific CD8⁺ T cells clustering around malaria-infected hepatocytes were proposed to locally secrete factors that result in the recruitment of additional effector T cells (15).

Because the size of the infection affects the behavior of surrounding T cells, only infections with a minimum volume of 50,000 μm³ (as estimated from the image voxels assigned to infected cells) were included. Note that small infections below this threshold were primarily observed for HSV_TOM-OVA, suggesting that this modified virus grows out less efficiently than HSV_TOM.

Based on recent data by Hickman et al. (7) in VV infection, the chemokine receptor CXCR3 is the primary candidate for mediating the subtle chemotaxis that we detected from the cell migration analysis. To test whether CXCR3 and its ligands are indeed responsible for chemotaxis, we transferred GFP-OTI-CXCR3^−/− cells (CXCR3^−/− cells; Fig. 4A) into C57BL/6 mice, which were then vaccinated with SIINFEKL and infected with HSV_TOM. After HSV_TOM administration, effector T cells from CXCR3^−/− donor origin were found at the site of infection, albeit at slightly lower frequencies than wild type cells were. Consistent with the results by Hickman et al. (7) for VV infection, this result suggests that CXCR3 expression is generally dispensable for T cell activation and tissue extravasation. For wild type OTI T cells infiltrating the area of infection, a strong accumulation around sites of infection and a consistent decline of T cell numbers with increasing distances of infection are expected, whereas an approximately flat spatial distribution would be expected for CXCR3^−/− T cells if signaling through this receptor is responsible for the observed directional migration. In line with this expectation, there was a significant negative correlation (p = 0.001) between T cell numbers and distance from infection for wild type OTI T cells, and a nonsignificant correlation for CXCR3^−/− OTI T cells (Fig.4B, 4C). Furthermore, whereas control T cells showed directed migration at long distances from the infection, no significant directed migration could be detected for CXCR3-deficient T cells (Fig. 4D, 4E). The small (nonsignificant) directed migration component that might still be present could suggest that other chemokine receptors also have a limited role in attracting T cells toward infection foci. However, signaling via CXCR3 clearly explains the bulk of the directional migration observed.

As an independent test for the involvement of CXCR3 that does not rely on genetic deficiency of the receptor, we explored the effect of local delivery of CXCL10. When the pre-existing chemokine gradient was disrupted by local delivery of CXCL10 by tattoo application into an area that was larger than that in which the T cells were localized, the consistent pattern of T cell migration toward foci of infection was no longer observed. Specifically, T cells around individual foci revealed either migration away from or toward foci or migration in an approximately random pattern, consistent with the notion that T cells were attracted to newly created CXCL10 hotspots (Fig. 5). Collectively, these data show that the directed migration displayed by CTLs in proximity of foci of infection is largely mediated by CXCR3 and its ligands.

The observed directional preference of CTLs toward sites of infection is relatively small (Fig. 2G, 2H; Supplemental Fig. 3), and the variation among all measured angles to infection is large (Fig. 2F; Supplemental Fig. 4, lower panels), thereby explaining why directional migration is not apparent from visual inspection of in vivo imaging data (Supplemental Videos 1–3). To determine to what extent T cell arrival at sites of infection is influenced by this small directional preference, we generated realistic, long-lasting tracks based on experimental data, using a computational approach similar to that in previously published work (10, 25, 26). We reconstructed cellular migration paths using the experimentally measured distributions of speed, angle to infection, and turning angle. Furthermore, because the distance to infection at which cells are located determines the typical values found for these three parameters, we chose combinations of speed and migration angles from data points that were binned with respect to their distance from the infection. Note that we did not explicitly incorporate full cellular arrest because of synapse formation with target cells in the simulations, but the in silico cells drastically slow down close to the infection (i.e., for HSV-specific T cells, see Fig. 2D).

During each simulation, in silico CTLs started at a predefined distance away from the circular (or spherical) infection, and combinations of speed and migration angles were drawn at random from the experimental distributions found around that distance. The resulting migration step determined the position, and therefore the distance to the infection, of the next time step. This procedure was repeated until a cell arrived in proximity of the infection (using a distance threshold of 50 μm) or left the space for which we had experimental data (i.e., reached distances too far away from the infection). The tracks generated (e.g., Fig. 6A) matched the experimental data closely in terms of their distributions of speed, angles to infection, and turning angles (Supplemental Fig. 4). In addition to these directed simulations, we also performed random simulations in which we substituted the experimentally found distribution of angles to infection with a random distribution, while keeping the persistence and speed of cells according to the experimental measurements (Fig. 6B).

Both specific and bystander cells starting at a distance of up to ∼150 μm arrived with almost 100% efficiency close to the infection when directed simulations were performed (Fig. 6C). In contrast, for randomly moving cells, regardless of their specificity, the fraction of cells that reached the target was only ∼30%. Furthermore, the cells that did arrive at the infection by random migration required approximately twice the amount of time as directional cells (Fig. 6D). The differences between directed and random simulations were observed when simulations were performed in 2D (Fig. 6) in one-side-restricted or two-side-restricted 3D (Fig. 7) simulations. In conclusion, our 2D and 3D simulations predict that the subtle chemotaxis of CTLs toward sites of infection are highly important for the rapid arrival of the majority of CTLs that are present in the tissue.

In the above experiments, we used short-term imaging data to generate simulations that showed the long-term consequences of the observed bias in angles to infection (Figs. 6, 7). These simulations rely on two essential assumptions: first, that the T cell migration parameters observed during imaging at 24 h postinfection hold true for longer periods, and second, that the T cell migration parameters measured are the major factors in driving T cell accumulation, rather than T cell division, death, or exit. To test whether in vivo T cell accumulation followed the pattern predicted by our simulations, we performed long-term–long-interval in vivo imaging experiments (lasting up to 12 h), monitoring both the spatial distribution of gBT T cells and infection growth (Fig. 8A–C). Analysis of the still images showed a clear accumulation of T cells close to the infection over time (Fig. 8A).

We then simulated CTL migration toward the infection, using the distance distribution observed at the first still time point of imaging (∼17 h) as a starting distribution; in addition, we simulated new cells entering the simulated field at a random position in space and time, such that the same amount of cells was present at the final time point in our simulations as in the imaging data set. Finally, we took into account the infection growth observed during imaging. The spatial distributions at the end of the directed simulations (up to 12 h) matched the experimentally observed spatial distributions well (Fig. 8D, blue dots). In contrast, the match between the spatial distribution observed in random simulations and that observed in vivo was poor (Fig. 8D, red dots), even in a variant of these simulations in which we incorporated full arrest when cells reached the simulated infection (Fig.8D, green dots). Finally, the low level of accumulation observed in random simulations is mainly caused by cellular entry into the epidermis close to the infection, rather than by intraepidermal migration followed by arrest. On the basis of these data, we conclude that the accumulation observed in vivo requires directional migration, and it cannot be explained by random migration combined with arrest close to the infection.

To explore the same issue for T cells that do not encounter cognate Ag, we subsequently performed the same imaging and simulation procedure for OTI bystander cells, which continue migrating when entering the lesion. In this case, directed migration alone consistently overpredicted the accumulation of effector cells close to the infection (Fig. 8E). Extension of the simulations with a probability to leave the infiltrated area after arrival at the infected site (e.g., by exit of the imaging field toward the dermis) improved the correspondence between predicted and observed spatial distribution. This result suggests that T cells that fail to detect cognate Ag may have an increased probability to leave the infiltrated area after a period of unfruitful search. Direct quantification of such bystander leaving during dynamic imaging is precluded in our setup because the thin image volume leads to many fake entry and exit events, which would strongly dominate true events.

The main finding of the current study is that effector CD8⁺ T cells in skin epithelium approach clusters of infected cells by subtle chemotaxis. This directed migration can be described as a small preference for movements toward the microlesion in an otherwise highly irregular trajectory, and in fact the T cells move away from the sites of infection almost as often as they move toward it.

The observed CXCR3-mediated attraction of T cells by clusters of infected cells implies the existence of a chemoattractant gradient, and the finding that the strength of T cell attraction correlates with lesion size is also consistent with this notion. If T cells are guided to sites of infection by a gradient, why is their migration pattern so erratic? One explanation is that T cell migration has to occur through a bed of tightly linked keratinocytes, thereby severely limiting the number of potential trajectories. This, however, does not seem to explain why T cells also frequently move away from the site of infection. Conceivably, the gradient that is sensed by the cells is locally stochastic and shallow. A shallow chemotactic gradient can make it difficult for small cells such as T lymphocytes to respond to the signal because of the minimal difference in signal strength sensed by the front and back end of the cell (27). However, neutrophils have a similar size as T cells and seem to be able to migrate directionally in a much more pronounced manner, both intravascularly and within peripheral tissues (28–30). Thus, it might also be a general property of T cells that random and persistent trajectories are adjusted only slightly upon encountering chemoattractant gradients, possibly to allow the scanning of a larger area of the peripheral tissue.

Upon HSV infection, both specific and bystander effector T cells are found within the infected area (Fig. 2A), consistent with prior data from other groups regarding the behavior of effector T cells in tumors and infected tissues (7, 24, 31, 32). The presence of substantial numbers of nonspecific T cells could reduce the efficiency of cognate CTL-target cell contacts by steric hindrance. For the similar case of T cell scanning of dendritic cells within lymph nodes, we recently demonstrated that such competition can be alleviated if bystander T cells lose their sensitivity to chemokine upon reaching the APC (33). By the same token, in the case of skin infection, this ‘logistic’ problem could be reduced when effector T cells that do not encounter their cognate Ag were to leave the infected area, a concept that is supported by our in silico simulations of bystander cells.

Two observations suggest that the chemoattraction we describe here is biologically meaningful. First, the pattern of T cell accumulation that is measured experimentally is reproduced in computer simulations in which the observed directional migration is incorporated, but not in simulations in which accumulation is driven by a combination of random T cell migration and T cell arrest close to the infection. Second, the observed chemotaxis influences T cell arrival on a time scale that is relevant to the kinetics of HSV-1 spreading within the infected skin (according to our in vivo measurements by a radial growth of approximately three cell layers, in 5 h). To our knowledge, this study presents the first direct evidence of chemoattraction toward regions of interest of effector T cells within a peripheral tissue. Prior studies in other peripheral tissues did not detect directional migration (16, 17), or postulated its occurrence on the basis of indirect observations (7, 15). It may be interesting to reanalyze such existing data by the strategies used here, to evaluate whether similar subtle migration behaviors are hidden in time-lapse imaging data sets.

We thank the Flow Cytometry Facility (NKI) for cell sorting, the Digital Microscopy Facility (NKI) and Kees Jalink for excellent technical support and advice, Ruud van Mierlo for daily assistance, and Feline Dijkgraaf and Johannes Textor for discussions.

The authors have no financial conflicts of interest.