Tissue clonality of dendritic cell subsets and emergency DCpoiesis revealed by multicolor fate mapping of DC progenitors

See allHide authors and affiliations

Science Immunology  01 Mar 2019:
Vol. 4, Issue 33, eaaw1941
DOI: 10.1126/sciimmunol.aaw1941

Dendritic cell dynamics

Conventional dendritic cells (cDCs) are critical to innate immunity and orchestrating adaptive T cell responses. cDCs originate from a common precursor and can be delineated into different subtypes. Cabeza-Cabrerizo et al. use multicolor fate mapping in mice to show that precursor cDCs enter tissue, differentiate into a single subtype, and proliferate as clones of sister cDCs under steady-state conditions. Viral infection causes a rapid influx of cDCs into infected tissue, and these cells differentiate into tissue-resident cDCs and dilute preexisting cDC clones. These results provide insight into cDC dynamics in tissues and how infection can cause rapid changes in cDC population frequencies.


Conventional dendritic cells (cDCs) are found in all tissues and play a key role in immune surveillance. They comprise two major subsets, cDC1 and cDC2, both derived from circulating precursors of cDCs (pre-cDCs), which exited the bone marrow. We show that, in the steady-state mouse, pre-cDCs entering tissues proliferate to give rise to differentiated cDCs, which themselves have residual proliferative capacity. We use multicolor fate mapping of cDC progenitors to show that this results in clones of sister cDCs, most of which comprise a single cDC1 or cDC2 subtype, suggestive of pre-cDC commitment. Upon infection, a surge in the influx of pre-cDCs into the affected tissue dilutes clones and increases cDC numbers. Our results indicate that tissue cDCs can be organized in a patchwork of closely positioned sister cells of the same subset whose coexistence is perturbed by local infection, when the bone marrow provides additional pre-cDCs to meet increased tissue demand.


Conventional dendritic cells (cDCs) are leukocytes that play a key role in innate immunity, as well as the initiation and regulation of T cell responses (1). They comprise two broad subtypes, cDC1s and cDC2s, that form a network of immune sentinel cells in most tissues of mice and humans (2). cDC1 and cDC2 originate from cDC-committed hematopoietic progenitors in the bone marrow (BM) known as the common (but, more accurately, the conventional) DC progenitor (CDPs) (37). CDPs give rise to pre-cDCs that exit the BM via the blood to seed lymphoid and nonlymphoid tissues (68). Individual pre-cDCs were originally envisaged to be bipotential and generate both cDC1 and cDC2 (4, 9). More recently, cDC1 or cDC2 subset specification was shown to be able to occur, at least under certain circumstances, during the CDP to pre-cDC transition to give rise to committed pre-cDC1 or pre-cDC2 (1012). It remains unclear to what extent tissues are seeded by committed pre-cDCs versus uncommitted pre-cDCs. Similarly, it is not known to which degree pre-cDCs arriving in nonlymphoid tissues have proliferative capacity (3) and can undergo local expansion to give rise to clones of differentiated cDC1 or cDC2 occupying a defined tissue territory.

The replacement of “old” tissue cDCs with “new” ones is thought to occur at a high rate as the half-life of cDCs in most tissues is 3 to 6 days (8). This steady-state cDC renewal is likely controlled at the level of the generation of pre-cDCs, their BM exit rate, their tissue seeding rate, and the rate of pre-cDC proliferation and differentiation into cDC1 and cDC2. There is indirect evidence to indicate that these parameters are not immutable and that the global generation of cDC (cDCpoiesis) can increase upon loss of cDCs in the periphery (13) or in response to infection or tissue injury (1422). Although all these studies suggest that cDC numbers in tissues can be regulated by local demand, it is unclear whether this requires changes in putative local proliferation of pre-cDCs or, alternatively, communication with the BM and recruitment of additional precursors.

Multicolor fate mapping of cell precursors allows for analysis of single-color cell clusters, which, in turn, informs on the clonal relationship of cells in tissues (23). Here, we use this approach to analyze the distribution and clonal architecture of cDCs in mice in which individual CDPs/pre-cDCs are fate-mapped with one of four possible fluorophores. We show that peripheral tissues, such as the lung and small intestine (SI), contain clones of sister cDCs that arise through local proliferation of immigrant pre-cDCs and their progeny. Most of these clones consist of a single cDC subtype, consistent with early fate specification at the level of the CDP to pre-cDC transition. Upon lung infection with influenza A virus, cDC numbers increase through an acute influx of pre-cDCs from BM, which dilutes preexisting cDC clones. Our findings offer deeper understanding of the organization and dynamics of cDCs in tissues and reveal an axis of emergency immune surveillance that could be manipulated to increase immunity in vaccination or immunotherapy.


Pre-cDCs and their progeny retain proliferative capacity in peripheral tissues

To analyze the cell cycle status of pre-cDCs and cDCs in tissues, we stained cells in the BM, spleen, mesenteric lymph node (mesLN), SI, and lung for both DNA content and phosphorylated histone H3 (phospho-H3). This technique allows us to identify cells in four different cell cycle phases, namely, G0/G1, S, G2, and M (Fig. 1A). Consistent with their high turnover rate, a large fraction of BM CDPs and pre-cDCs was found in S/G2/M phases at any given time (Fig. 1B). That contrasted to pre-cDCs and fully differentiated cDCs in the spleen, LN, lung, and SI, in which the fraction of S/G2/M cells was much smaller, although not negligible, as previously observed (Fig. 1, D, F, H, and J) (8, 24, 25). Because the DNA/phospho-H3 staining does not distinguish cells that are still cycling (G1) from resting cells that have exited the cell cycle (G0), we carried out a separate analysis for Ki67, which is expressed in the G1, S, G2, and M phases but not in G0. As expected, 100% of BM CDPs and 80% of BM pre-cDCs were Ki67+, consistent with cell cycle commitment (Fig. 1A and fig. S1A). In the spleen and mesLNs, 80 and 60% of pre-cDCs were Ki67+, whereas in the lung and SI, this figure was around 40% (Fig. 1, E, G, I, and K). Differentiated cDCs in the spleen and mesLNs were also found to be Ki67+: 20% of splenic cDCs and 40 to 60% of LN cDCs. Similarly, we found that 30 to 40% of cDC1 and the two intestinal subtypes of cDC2 [CD103+ CD11b+ (double positive) cDC2 and CD103 CD11b+ cDC2] were Ki67+ (Fig. 1I and fig. S1B) (5, 26). In the lung, cDC1 and cDC2 stained for Ki67 at similar frequencies to cDCs in the SI (Fig. 1K and fig. S1B). Overall, these data suggest that a substantial fraction of pre-cDCs that have entered peripheral tissues and of their cDC progeny remains in the G1 phase of the cell cycle and has not therefore become postmitotic.

Fig. 1 Pre-cDCs and cDCs actively cycle in lymphoid and nonlymphoid organs.

(A) DNA content histogram (top) or dot plot with DNA content and phospho-H3 (bottom) of BM CDPs from one mouse as a representative example of cell cycle analysis. (B, D, F, H, and J) Percentage of cells in G0/G1, S, G2, or M phases of the cell cycle determined as in (A) in the BM (B), spleen (D), mesLN (F), SI (H), and lung (J). (C, E, G, I, and K) Percentage of Ki67+ cells in the BM (C), spleen (E), mesLN (G), SI (I), and lung (K). Data in (B, D, F, H, and J) are mean values from six C57BL/6 mice. Data in (C), (E), (G), (I), and (K) are compiled from 12 C57BL/6 mice analyzed in two independent experiments. Error bars correspond to variation across mice using SD. Cells are gated as indicated in Materials and Methods.

Clec9aConfetti mice allow for multicolor labeling of cDCs in tissues

The above observations suggested the possibility that at least some tissue cDCs might be organized in patches of sister cells that arise through local cell division. To assess this, we resorted to multicolor fate mapping of cDC precursors. We crossed Clec9aCre mice (27) to Rosa26Confetti mice (progeny henceforth called Clec9aConfetti; fig. S2A) (23). Both CDPs and pre-cDCs express dendritic cell NK lectin group receptor-1 (DNGR-1), encoded by the Clec9a gene, and mice expressing Cre recombinase under the control of the Clec9a locus have been used to trace the cDC lineage in vivo (27). We expected that, in Clec9aConfetti mice, cDC precursors would become stochastically labeled with one of four fluorescent proteins [cyan fluorescent protein (CFP), green fluorescent protein (GFP), yellow fluorescent protein (YFP), and red fluorescent protein (RFP)] and transfer the fluorophore to daughter cells, allowing tracing of cDC clones. We validated Clec9aConfetti mice by flow cytometric spectral analysis (which allows separation of closely related fluorophores, including GFP and YFP) of the spleen, mesLN, SI, and lung cell suspensions. In all organs, we found labeled cDC1 and cDC2 (fig. S2, B to D). The percentage of labeled cells in Clec9aConfetti mice was lower than previously observed using Clec9aCre crossed to a Rosa26YFP single fluorophore fate reporter strain (fig. S2B) (27), likely due to the complexity of the RosaConfetti locus, which reduces the efficiency of Cre-mediated recombination (28). This, together with the transience of Cre expression, leads to incomplete penetrance of the reporting event and causes labeling of only a fraction of cDCs, as previously reported (27). In contrast to cDCs, the frequency of labeled CD64+ cells, generally considered to correspond to monocytes and macrophages (29, 30), was very low (fig. S2B), as expected (27), although in the lung, where CD64+ cells vastly outnumber CD64 cDCs, they contributed to a larger fraction of all labeled cells (fig. S2C). As noted (23), the expression of the four fluorescent proteins in the Confetti reporter cassette was unequal, with clear underrepresentation of GFP+ cDCs (fig. S2, C and D). cDC1s, but not cDC2s, are DNGR-1+ and hence express Cre recombinase after differentiation and become preferentially labeled in Clec9a-Cre–based reporter mice (27), including RosaConfetti (fig. S2, C and D). This also means that cDC1 can continue to recombine the inverted loxP sites of the RosaConfetti locus, switching back and forth between expression of GFP and YFP or of RFP and CFP. Consistent with that notion, GFP+YFP+ or CFP+RFP+ double-positive cells were disproportionately more abundant among cDC1 than cDC2 subsets (fig. S2, C and D). Despite these limitations, the flow cytometric spectral analysis demonstrates that CDP/pre-cDCs in Clec9aConfetti mice can be stochastically labeled with different fluorophores that are transmitted to daughter cells resident in tissues.

Imaging of Clec9aConfetti mice reveals cDC distribution in three dimensions

To visualize the spatial arrangement of labeled cDCs and assess possible clustering by fluorophore, we developed a protocol to fix and clarify tissues while preserving the Confetti fluorescent proteins (Fig. 2A). The method includes agarose embedding and vibratome sectioning, allowing cutting of 300-μm sections, a thickness necessary to circumvent the scarcity of labeled cDCs in nonlymphoid tissues and to image enough cells for the analysis of clusters (Fig. 2A). It is also compatible with antibody (Ab) staining of the sections (see below). Large tissue volumes (1020 μm by 680 μm in 100 z-steps of 3 μm for the SI and 1360 μm by 1360 μm in 40 z-steps of 5 μm for the lung) were then imaged by confocal microscopy using total spectrum acquisition (lambda mode scanning), followed by spectral unmixing to discriminate all fluorophores (Fig. 2A).

Fig. 2 Spectral imaging of organs from Clec9aConfettireveals the presence of single-color cDC clusters.

(A) Workflow of tissue processing, staining, and imaging of Clec9aConfetti mouse organs. (B) 3D projection of a 300-μm spleen section stained with anti-CD169. Confetti surfaces were generated with Imaris software to reduce autofluorescence. Zooming into T cell areas (1) was used to visualize Confetti+ cDCs (2). (C) 3D projection of a mesLN with surfaces as in (B). Square depicts selected zoom in area displayed at the bottom. (D) 3D projection of a 300-μm vibratome section of the SI from a Clec9aConfetti mouse stained for E-cadherin to delineate the epithelium. (E) 3D projection of a 300-μm vibratome section of the lung from a Clec9aConfetti mouse. Autofluorescence channel is displayed to visualize the lung structure.

Validation of the system was initially carried out using lymphoid organs such as the spleen and mesLN. It revealed an intricate network of Confetti+ cells predominantly localized to T cell areas, as expected (Fig. 2, B and C). However, the large number and high density of labeled cells in these organs precluded analysis of clustering by fluorescent protein (Fig. 2, B and C). We therefore focused on nonlymphoid tissues such as the SI and lung, where Confetti+ cells were easily detected but sufficiently sparse to allow cluster analysis (Fig. 2, D and E). Visual inspection of images from the SI and lung of Clec9aConfetti mice revealed that Confetti+ cells were often found in discrete single-color clusters within individual villi or around airways, respectively (Fig. 2, D and E, and movies S1 and S2). Labeling was largely lost in Clec9aConfetti mice crossed to Flt3l−/− mice (fig. S3, A and B, and movies S3 and S4), which lack cDCs but not MACROPHAGEs (31, 32), confirming that Confetti+ cells were bona fide cDCs. This was true even for the lungs, where a considerable number of Confetti+ cells had been found to be CD64+ by flow cytometry (fig. S2D). To clarify the identity of these CD64-expressing cells, we carried out a separate flow cytometric analysis of Clec9atdTomato mice deficient in Flt3L. As expected, the frequency of Tomato-labeled CD64 cells was reduced in Flt3l−/− mice (fig. S3C), consistent with a decrease in total cDC numbers (fig. S3D). In contrast, the frequency of CD64+ cells, irrespective of CD11c and CD11b expression, was not reduced in Flt3l−/− mice, which, again, is as expected (fig S3D). However, the frequency of Tomato-labeled CD64+ cells was reduced in Flt3l−/− mice (fig. S3D), intimating that they represent atypical cDCs that express CD64, as previously suggested (27). Nevertheless, to avoid ambiguity, CD64+ cells were excluded from further analysis.

ClusterQuant 3D analysis indicates single-color cDC clustering in Clec9aConfetti mice

To quantify flurophore-based clustering of cDCs in the SI and lung, we developed a three-dimensional (3D) version of the ClusterQuant software previously used for analysis of cell clusters in single tissue planes (33). Briefly (Fig. 3A), the workflow involves different steps: (1) separation of 3D confocal images into individual z planes, corresponding to the optical slices used for image acquisition, (2) manual segmentation and annotation of cDCs in each plane, and (3) computation of 3D Voronoi polyhedrons using the x, y, and z cell coordinates. Each polyhedron contains all voxels closer to the centroid of that Confetti+ cell than to the centroids of all other cells and is used to compute neighbor and proximity relationships in subsequent analysis steps (Fig. 3, B and C, and movies S5 and S6). To aid tessellation and generate shapes that approximate cell volumes, in step (ii), it is necessary to draw borders along anatomic barriers (crypts and airways). We also randomly inserted dummy cells (displayed as gray cells) into the spaces around Confetti+ cells, helping to remove artificially large Voronoi polyhedrons, avoiding false-positive cluster formation and serving as a randomized control for the analysis (fig. S4, A, B, and E; see below). Each Voronoi polyhedron was then assigned a color based on the fluorescent protein expressed by the annotated underlying cell (i.e., green Voronoi for a GFP cell). Double-positive cells were assigned either purple (RFP+ CFP+) or orange (GFP+ YFP+). Adjacent Voronoi polyhedrons bearing the same color were considered a cluster (i.e., a group of two or more cells of the same color). The colors (but not the positions) of the Voronoi polyhedrons, representing individual cDCs, were then randomized in a step (4) through a Monte Carlo simulation with 10,000 possible realizations. Comparison of the original (O) versus simulated (S) images was carried out by extracting relevant parameters such as number of clusters, cells per cluster, and cluster compactness (Fig. 3A) to determine the probability that the observed clusters could have arisen by chance.

Fig. 3 Analysis of single-color cDC clusters using ClusterQuant 3D.

(A) Workflow of plane separation (1), manual cell annotation (2), 3D Voronoi generation, Monte Carlo (MC) randomization (4), and analysis using ClusterQuant 3D. (B) Original images from the SI (top) and the lung (bottom) were annotated using ClusterQuant 3D software and converted to Voronoi polyhedrons (middle), which were then randomized through a Monte Carlo simulation (right). CD64 staining was used to exclude CD64+ cells from the analysis. Dashed lines indicate the structure of the SI (villi) or lung (airways), determined from CD11b staining or autofluorescence channel, respectively. Colors represent the different Confetti fluorescent proteins. Note the scarcity of double-labeled cells, especially in lung.

In all cases, the ClusterQuant 3D analysis revealed a pattern of single-color cDC clustering in the SI and lung that was not reproduced in randomized scenarios (Fig. 4). The difference between O and S images was statistically significant regardless of whether data were analyzed as number of clusters relative to total cell number, fraction of total cells in clusters (Fig. 4, A and B), or number of clusters per unit volume of tissue (fig. S4, C and D). This was true for both the SI and lung and was observed when images from all mice were pooled together (Fig. 4, A and B) or when images were grouped per mouse (Fig. 4, C and D). Confirming the validity of the 3D ClusterQuant algorithm, analysis of the randomly placed dummy cells showed that they were not significantly clustered (fig. S4E).

Fig. 4 SI and lung cDCs are organized in spatially restricted clusters of sister cells.

(A) From left to right: Number of clusters of two or more cells normalized to a number of cells, percentage of number of cells in clusters, and cluster compactness from 24 images of SI from four mice. Each point represents one image; observed (O, orange) compared with simulations (S, gray). (B) As in (A) from 25 lung images from five mice. (C) Data from (A) grouped per mouse. Lines link the observed and simulated scenarios associated with each mouse. Colors correspond to individual mice (observed and simulation). (D) Data from (B) grouped per mouse as for (C). (E and F) Proportion of clusters of the indicated size in observed scenario (orange) normalized to the simulated scenario (dashed lines). Data are from SI (E) or lung (F) of a representative mouse. Other mice are shown in fig. S4. (G) Proportion of all clusters analyzed of size 1 (gray) or ≥2 in the SI (orange) or lung (blue). (H) Comparison of proportion of clusters of size 2 to 10 in the SI (orange) and in the lung (blue). In all cases, CD64 staining was used to exclude CD64+ cells from the analysis. Statistical analysis in (B) to (E) used a paired t test, and in (F) and (G), used a χ2 test.

In additional analyses using statistical data binning with χ2 testing, we examined the size distribution of clusters and normalized them to the value expected to be obtained by chance, which was calculated from the simulated scenarios. We found that there was a greater proportion of large cDC single-color clusters than expected by chance alone (Fig. 4, E and F, and fig. S4, F and G). In some cases, we observed very large clusters (>15 cells), which are exceedingly improbable in a randomized scenario. However, cDCs were not always in observable clusters and, in more than half of the cases, appeared as isolated cells (Fig. 4G). We also found that clusters in the lung were significantly larger than in the SI, indicative of tissue differences (Fig. 4, G and H). We conclude that at least some cDCs in the SI and lung and, possibly, other tissues of Clec9aConfetti mice, form single-color clusters, which represent clones of sister cells.

Single-color cDC clusters are predominantly composed of a single cDC subset

We next asked whether cDC clones comprise multiple subsets (i.e., arise from uncommitted pre-cDCs) or a single subset (i.e., originate from committed pre-cDC1s or pre-cDC2s or from dividing cDC1 and cDC2s) by analyzing cluster composition. We used staining with Abs against CD11b and CD103 to allow discrimination between cDC1 and cDC2 and Ab against CD64 to exclude CD64+ cells (Fig. 5, A to C, and movies S7 and S8).

Fig. 5 Single-color cDC clusters are predominantly composed of a single cDC subset.

(A) 3D projection of a Clec9Confetti image of the SI (left) or lung (right) stained for CD103, CD11b, and CD64. (B and C) Single z optical slices depicting individual cells from the SI (B) or lung (C) of Clec9Confetti mice; individual channels are shown on the right side of the merged image. (D) Pie charts representing percentage of pure clusters in the SI (orange, 80%) or lung (blue, 70%) were analyzed. Gray bar indicates mixed clusters. Data are pooled from all images, and n indicates the number of clusters analyzed. (E and F) Analysis of the proportion of pure clusters of cluster size 2 to 7 found in the SI (E, orange) or lung (F, blue) (o) versus the expected null distribution assuming random mixing (r, gray). Data are pooled from all images. Error bars correspond to variation across mice using SD. Statistical analysis was carried out using a Fisher’s exact test.

The distribution of fluorescent protein expression among cDC subsets in tissue sections of Clec9aConfetti mice as assessed by microscopy (fig. S5A) was concordant with that observed in cell suspensions analyzed by spectral flow cytometry (fig. S2, C and D), cross-validating both experimental approaches. Cell composition analysis from ClusterQuant 3D analysis revealed that single-color clusters in both the SI and lung often consisted of a single cDC subset (i.e., “pure”) although “mixed” clusters comprising cDC1 and cDC2 could also be observed (Fig. 5D and fig. S5B). Pure clusters accounted for 80% of all clusters in the SI and 70% of clusters in the lung (Fig. 5D). The probability of a single-color cluster being pure by chance was calculated from the cluster size and the proportion of cDC1 and cDC2 cells in that tissue and revealed a far greater proportion of pure clusters than would be expected in a random scenario (Fig. 5, E and F). Together, these data suggest that, during homeostasis, tissue cDC clones of sister cells are predominantly generated by local proliferation of incoming precommitted cDCs and of their differentiated cDC progeny.

Single-color cDC clusters are lost during infection

Next, we asked whether cDC single-color clustering (and, inferentially, cDC clonality) in tissues holds true under inflammatory conditions when an increase in cDC numbers is often observed (16, 1820). Clec9aConfetti mice were intranasally infected with influenza A virus (strain X31), which, over the course of a week, caused a large increase in lung cDC1 and cDC2 numbers, as assessed by flow cytometry (Fig. 6A). This was accompanied by large-scale infiltration of innate immune cells into the lungs but did not affect the specificity of the genetic labeling of cDCs in Clec9aCre reporter mice (fig. S6, A to C). Images of infected lungs that were stained for viral proteins showed that the infection was patchy, with viral replication being confined to discrete foci (fig. S6A and movie S9), as reported (34). At 7 days postinfection (dpi), Confetti+ and CD64+ cells accumulated in much greater numbers around virus-containing foci than in uninfected regions (Fig. 6, B and C). CD64+ cell infiltration was therefore used as a surrogate to discriminate high (infected) and low (uninfected) infiltrated areas (Fig. 6C). 3D ClusterQuant analysis revealed that the cDC single-color clustering that had been observed under steady-state conditions was largely lost from highly infiltrated areas after infection, which no longer showed a statistically significant difference between O and S scenarios irrespective of whether the data were pooled or analyzed by individual mouse (Fig. 6, D and E, fig. S7A, and movie S10). Low infiltrated areas retained some clusters of single-color cDC in some of the mice but this did not reach statistical significance when all mice were considered together (Fig. 6, F and G, fig. S7B, and movie S11).

Fig. 6 Influenza A virus infection dilutes single-color cDC clustering.

(A) Number of cDC1s (left) or cDC2s (right) in the lungs of Clec9aeYFP mice infected with influenza A virus (magenta) or non-infected (N.i., gray). (B) 3D projection of a Clec9Confetti lung section 7 dpi with influenza A virus. (C) Quantification of Confetti+ cells, CD64+ cells, and X31 particles from images with high (magenta) and low (purple) infiltration of cells. (D) Number of clusters of two or more cells normalized to number of cells and percentage of cells in clusters from 20 high infiltrated lung images from five mice. Each point represents one image; observed (magenta) compared with simulations (gray). (E) Data in (D) grouped per mouse. Lines link the observed and simulated scenarios associated with each mouse. Colors correspond to individual mice (observed and simulations). (F) Data as in (D) from 19 low infiltrated lung images from four mice. (G) Data in (F) grouped per mouse. (H and I) Proportion of clusters of the indicated size in original (O) normalized to the simulated (S, dashed lines) scenario from the high (H) or low (I) infiltrated areas of a representative mouse. (J) Proportion of clusters of size 1 (gray) or ≥2 in non-infected lungs (blue) versus. high (magenta) and low (purple) infiltrated areas from lungs of infected mice. n.s., not significant. (K) Comparison of proportion of clusters of size 2 to 10 in non-infected lungs (blue; Fig. 2) and high (magenta) and low (purple) infiltrated areas in the lungs of infected mice. Data correspond to the pool of all mice analyzed. Statistical analysis used an unpaired t test (A) in two independent experiments with three mice per group, paired t test (C to G), and χ2 test (H to K).

We extended the analysis by comparing cluster size distribution between uninfected lungs and areas of low or high infiltration in infected mice (Fig. 6, H to K). This confirmed that the data fitted a random scenario (Fig. 6, H and I), although some low infiltrated areas in some mice still displayed statistically significant formation of clusters (fig. S8). Despite the latter, overall, there was a slight increase in the frequency of cDCs that were not in a cluster compared with the situation in uninfected mice (Fig. 6J). We found statistically significant differences at the level of larger clusters, which could be preferentially found in lungs from uninfected mice (Fig. 6K). Together, these results show that the single-color clustering of cDCs in lung is significantly reduced after influenza A virus challenge, not only in areas with active infection but also, to a lesser extent, in areas away from infectious foci.

Infection-driven loss of single-color cDC clustering correlates with influx of pre-cDCs

The above data suggested that infection-driven increases in local cDC numbers were not likely to be a result of increased local clonal expansion. Flow cytometric analysis of cell cycle status showed that infection led, if anything, to a decrease in the frequency of cDC1 and cDC2 in S/G2/M cell cycle phases and a reduction in percent cells positive for Ki67 (Fig. 7A and fig. S9A). Pre-cDC in the lung and pre-cDC and CDP in the BM did not show any major changes in Ki67 positivity (fig. S9, A and B), although more of them were found to be in S phase at day 7 after infection compared with uninfected mice (Fig. 7, A and B). Therefore, the increase in cDC numbers in the lungs of mice after infection with influenza A virus (Fig. 6A) does not seem to be due to an increase in proliferation of cDCs or their immediate lung precursors.

Fig. 7 Influenza A virus infection increases lung cDC numbers by recruiting BM progenitors.

(A and B) Percentage of the lung pre-cDCs and cDCs (A) and of the BM CDPs and pre-cDCs (B) in G0/G1, S, G2, or M phases of the cell cycle determined as in Fig 1. Data are mean values from six non-infected (top) or six C57BL/6 mice 7 dpi with influenza virus (bottom). (C to F) Numbers of BM CDPs (D) or pre-cDCs in the lung (C), BM (E), or blood (F) in non-infected (gray), or influenza A virus–infected C57BL/6 mice (magenta) at 3 or 7 dpi. Each dot represents one mouse of six per group from one representative experiment. (G) Relative mean percentage of cells from peripheral blood of 41 patients before and after natural infection with influenza A virus obtained from microarray data using CIBERSORT. Tregs, regulatory T cells; NK cells, natural killer cells. (H) Percentage of activated blood DCs in individual patients from (G). (I) Expression of SEMA4D, CD3E, CD79A, and CD79B in peripheral blood from patients before (pre, gray) or in the first 48 hours of symptoms after natural infection with influenza A virus (post, magenta). Statistical analysis in (C) to (F) was based on an unpaired t test in a single experiment with six mice per group. Statistics in (H) and (I) used a paired t test.

To ask whether it could be attributable to changes in the dynamics of lung seeding by pre-cDC, mice were examined at different time after infection. At day 7 after challenge, influenza virus–infected mice had many more pre-cDCs in their lungs than uninfected controls (Fig. 7C). We used markers used to distinguish pre-cDC1s and pre-cDC2s (10, 11) (with the caveat that such markers have not been validated in the lung) and found that the infection-associated increase in lung pre-cDCs predominantly represented uncommitted cells (fig. S9C), consistent with the fact that both cDC1 and cDC2 increased equally (Fig. 6A). In contrast to the lung, there was loss of cDC progenitors from BM, which reached their lowest numbers at 3 to 4 dpi and recovered by day 7 (Fig. 7, D and E). As in the lung, there was no major bias in terms of pre-cDC subset distribution in the BM (fig. S9D).

These results suggested that high demand for cDCs induced by infection is met by rapid mobilization of cDC progenitors from the BM. Consistent with that notion, there was a clear increase in total number of pre-cDCs circulating in peripheral blood of mice at 7 dpi, when the increase in lung pre-cDCs and cDCs became apparent (Fig. 7F), with no differences in pre-cDC1/pre-cDC2 commitment (fig. S9E). To extend these data to humans, we carried out an analysis of transcriptome datasets from peripheral blood of patients before or after infection with influenza A virus (35) using the CIBERSORT algorithm (36). This revealed a marked increase in the blood frequency of DCs, annotated in CIBERSORT as “activated DCs” (Fig. 7, G and H), consistent with a demand-driven increase in DCpoiesis. To assess whether pre-cDC numbers were also increased in human blood, as in mice, we assessed levels of transcripts for SEMA4D, a recently identified marker for CDPs (37). SEMA4D transcripts were increased after infection (Fig. 7I). Although SEMA4D can also be expressed by T cells and B cells, T and B cell markers (CD3E or CD79A/B) were either unchanged or decreased in the same datasets after influenza virus infection (Fig. 7I). Together, these data suggest that, in mice and, possibly, humans, influenza virus infection leads to local demand for cDCs that is met not by increased local clonal expansion of pre-cDC and their progeny but by communicating need to the BM, resulting in an efflux of pre-cDCs into blood and influx into lungs.


The parameters underlying cDCpoiesis remain poorly understood. Here, we use a genetic model coupled to image analysis and 3D cluster quantification to analyze how cDC precursors seed tissues at the single-cell level in the absence of cell transfer. We reveal that pre-cDCs in the steady state enter peripheral tissues and can divide locally before differentiating into cDCs, which display residual proliferative capacity. This leads to formation of discrete clones of cDCs that remain in close proximity. We find that these sisters are predominantly composed of either cDC1 or cDC2, possibly providing in vivo corroboration for the notion that cDC subset commitment can occur at the pre-cDC level (10, 11). Last, we demonstrate that cDC generation is an elastic process that responds to external tissue demand by exporting pre-cDCs from the BM at times of need. The rapid influx of such pre-cDCs into tissues and reduction in local cell division likely lead to accelerated intermingling of clones and correlate with loss of single-color clustering in multicolor fate-mapping mice.

Quantification of images from tissues of Clec9aConfetti mice revealed that 45% of cDCs in the lung and 35% of cDCs in the SI are in single-color clusters of two or more cells. It is remarkable that single-color clusters can be detected at all. Because cDCs are motile, especially when activated, it is likely that clustering of clones dilutes over time. Motility will additionally cause cDC clones to intermingle, and ongoing cDC death will lead to cluster dissolution. In addition, the fact that cDC1 can switch fluorescent protein expression because of continued Cre expression can also break up single-color clusters (although, in some cases, this can be accounted for in the analysis as the half-life of the proteins is sufficiently long to result in double labeling). Last, the use of dummy cells to aid with segmentation in the analysis may inadvertently lead to artificial separation of single-color clusters. For all these reasons, the degree of cDC single-color clustering that we observe in tissues is likely to be an underestimate of the true extent of tissue cDC clonality, and many of the isolated cells in tissues may therefore, in fact, have previously been part of a single-color cluster. The issues discussed above (e.g., cDC migration) will disproportionately affect large clones, causing a bias toward detection of small single-color clusters and isolated cells. The finding of some large clones suggests, therefore, the existence of tissue niches that allow for greater pre-cDC/cDC local proliferation and/or that prolong cDC half-life (25). Such niches may be more common in the lung than in the SI as we noted a clear tendency for single-color cDC clusters in that organ to be larger. Why clonality does not result in observable patches of cDC1 versus cDC2 in tissues can be understood from the fact that large clones are relatively rare and that we visualize only a fraction of the tissue cDC network because of incomplete penetrance of the Cre-mediated recombination event. The superimposition of the mosaic pieces composed of cDC1 and cDC2 clones and single cells is expected to result in spatial mixing of the subsets, also explaining why we do not see obvious single-color clusters in regions where cDCs are too abundant, such as T cell areas of LNs and spleen.

Experiments with cells in vitro, sorted ex vivo, or transferred as bulk populations into mice have suggested that individual pre-cDCs can be either precommitted to generate cDC1 or cDC2 or can be bipotent and generate either cDC subset (10, 11). Here, we revisited this question in Clec9aConfetti mice where the composition of the Confetti+ clones is analyzed in an unperturbed state. We show that cDC clusters in lung and SI are predominantly pure cDC1 or cDC2 consistent with the possibility that they arise primarily from committed pre-cDC1s and pre-cDC2s. We also find that cDC1s and cDC2s retain residual capacity to proliferate after differentiation, which could contribute to the purity of the cDC clusters observed irrespective of pre-cDC commitment. The fact that cDC1s continue to express Cre could, in theory, introduce a bias toward detecting pure clusters of cDC1 expressing two fluorescent proteins surrounded by cDC2 expressing one of the two proteins. However, this was not observable in our images and the rarity of doubly labeled cDC1 events cannot explain the extremely high fraction of pure clusters detected. It should be noted that not all clusters are pure and that some mixed cDC ones are found in both the SI and lung, possibly providing in vivo evidence for the existence of uncommitted pre-cDCs capable of seeding tissues and giving rise to sister cells of different fates.

Infection or inflammation is often accompanied by a local increase in macrophage and cDC numbers, which can be met through local proliferation or increased precursor recruitment. Tissue macrophages are self-renewing and can proliferate more rapidly in response to injury-induced signals and cytokines (3840). Furthermore, blood monocytes can enter tissues upon demand and differentiate into cells that resemble tissue macrophages (41). Monocytes can also differentiate into cells that have cDC features but are distinct from those that arise from regular cDCpoiesis (42). Therefore, increased demand for cDCs in the periphery can only be met through increased generation of cDC precursors in the BM or by greater mobilization and/or proliferation of pre-cDCs in tissues. We show that influenza A virus infection causes an increase in lung cDC numbers, which is not accompanied by increased local proliferation of pre-cDC or their differentiated progeny. Rather, it is accompanied by efflux of pre-cDCs from the BM, which increase in number and frequency in peripheral blood and enter tissues, leading to increased seeding. In Clec9aConfetti mice, this acute influx of pre-cDCs into the lungs results in dissolution of preexisting single-color cDC clusters, a process that might be exacerbated by increased cDC mobility leading to increased intermingling of preexisting clones and migration to mediastinal LNs. Differently from a previous report (20), we did not observe a selective increase in cDC2 in influenza virus–infected lungs and, consistent with that observation, pre-cDC in the BM or lung did not display any infection-induced changes in phenotype suggestive of a bias toward cDC1 or cDC2 commitment (10, 11). Whether this relates to the strain of influenza virus, the severity of the infection or the fact that the previous study (20) did not necessarily discriminate bona fide cDC2 from monocyte-derived cells remains to be established. Overall, our findings reveal that infected tissues can communicate to the BM their need for increased cDC, resulting in a release of pre-cDCs into blood and transient loss of CDPs from their site of origin. By analogy, with the increased release of granulocytes into the circulation during infection or inflammation (43, 44), the fact that cDC generation responds to demand might aptly be termed “emergency DCpoiesis.” The signals that regulate emergency DCpoiesis are likely to constitute useful targets for immunotherapeutic approaches to cancer, infectious disease, and autoimmunity.



Clec9a-Cre, ROSA26-YFP (45), ROSA26-Confetti (23), Flt3l−/− (Taconic Biosciences), and C57BL/6J mice were bred at The Francis Crick Institute under specific pathogen-free conditions. All genetically modified mouse lines were backcrossed to C57BL/6J. Six- to 20-week-old mice were used in all experiments unless otherwise specified. All animal experiments were performed in accordance with National and Institutional Guidelines for Animal Care.

Infection with influenza virus

Mice were anaesthetized via inhalation of isoflurane. Mice were intranasally infected with 35,000 TCID50 of influenza A X31 (H3N2) in 30 μl of phosphate-buffered saline (PBS). Mice were monitored daily for weight loss and signs of infection and euthanized at 1, 3, and 7 dpi.

Flow cytometry

Up to 4 million cells (see the isolation protocol in Supplementary Methods) were preincubated with blocking anti-CD16/32 in fluorescence-activated cell sorting (FACS) buffer for 10 min at 4°C and then stained for 20 min at 4°C with staining cocktail in FACS buffer in the presence of anti-CD16/32. DAPI (4′,6-diamidino-2-phenylindole) was used to exclude dead cells, except for when cells were fixed. In the latter case, dead cells were excluded by LIVE/DEAD fixable blue or aqua dye (Invitrogen). For Ki67, DNA content (FxCycle, Invitrogen) and phospho-H3 staining cells were fixed and permeabilized using Foxp3 Fix/Perm buffer set (00-5523-00, eBioscience). For DNA content examination, samples were collected on a LSR Fortessa flow cytometer (BD Biosciences) or a SP6800 Spectral Analyzer (Sony) and analyzed using FlowJo 9 or 10 software (TreeStar Inc.). Gating strategies are shown in figs. S10 to S12, including the discrimination of the four Confetti fluorophores by spectral analysis. Abs used for flow cytometry are listed in table S2.


Mice were perfused with 20 ml of PBS and 10 ml of Antigenfix (Diapath). One milliliter of melted 2% low–melting point agarose (Invitrogen) in PBS was inserted through the trachea. The lungs, spleen, and LNs were then removed and fixed overnight in Antigenfix at 4°C. The proximal SI (mostly duodenum) was also removed, and contents were flushed with ice-cold PBS and Antigenfix and fixed for 2 hours at 4°C. SIs were subsequently cut longitudinally, rolled into a “Swiss roll,” and fixed overnight in Antigenfix at 4°C. After fixation, tissues were transferred to 30% sucrose in PBS for 24 hours at 4°C. Lungs, spleen, and SI were embedded in 4% agarose in PBS, and 300-μm sections were cut using a Leica VT1200S vibratome and collected in PBS. LNs were unsectioned and imaged as whole mounts. Tissue sections were stained as explained in Supplementary Methods. Abs used in microscopy are listed in table S3. Sections were mounted on slides in the clearing solution RapiClear 1.47 (SunJin Lab) for image acquisition, whereas LNs were clarified instead using uDISCO (46) and mounted on a metallic glass-bottom dish. Images were acquired on a LSM 880 inverted confocal microscope as explained in the Supplementary Materials. Images and movies were generated after adjusting channels using Imaris software.

Cluster analysis

To analyze the clustering of cells in three dimensions, we used a 3D version of ClusterQuant 2D (33), as described in detail in Supplementary Methods. For cluster composition, we used a probability-based method based on the ratio of cDC1:cDC2 in lung and SI, as explained in Supplementary Methods and fig.S5 (C and D).

Human influenza dataset analysis

Human microarray data from influenza A virus–infected cohort [Gene Expression Omnibus GSE68310 (35)] were analyzed using CIBERSORT (36). For gene expression data, samples were normalized and compared in a paired basis. Baseline was compared with first onset of symptoms.


Statistical analyses were performed using GraphPad Prism software (GraphPad), MATLAB (Mathworks), or RStudio. Statistical test used is specified in each figure legend.


Supplementary Methods

Fig. S1. Staining for Ki67 in cDCs and their progenitors.

Fig. S2. Flow cytometric analysis of Clec9aConfetti organs.

Fig. S3. Analysis of Clec9aCre-based fate-mapping mice on a Flt3l−/− background.

Fig. S4. ClusterQuant 3D analysis of single-color cDC clustering in Clec9aConfetti mice.

Fig. S5. Analysis of Confetti+ cDC subsets.

Fig. S6. Quantification of infected cells and cell infiltration during influenza A virus infection.

Fig. S7. Voronoi diagrams of analyzed high and low infiltrated areas from infected Clec9aConfetti mice.

Fig. S8. Cluster size distribution in high and low infiltrated areas of lungs from infected Clec9aConfetti mice.

Fig. S9. Quantification of cDC proliferation and pre-cDC composition during influenza A virus infection.

Fig. S10. Flow cytometry gating strategy for CDPs.

Fig. S11. Flow cytometry gating strategy for cDC subsets and macrophages.

Fig. S12. Flow cytometry gating strategy for Confetti+ cells in Clec9aConfetti mice.

Table S1. Excel file with raw data used to generate all graphs that have n < 25.

Table S2. Abs used in flow cytometry.

Table S3. Abs used in confocal microscopy.

Movie S1. SI from a Clec9aConfetti mouse.

Movie S2. Lung from a Clec9aConfetti mouse.

Movie S3. SI from a Clec9aConfetti mouse crossed to a Flt3l−/− background.

Movie S4. Lung from a Clec9aConfetti mouse crossed to a Flt3l−/− background.

Movie S5. Representative annotated image of the SI of a Clec9aConfetti mouse.

Movie S6. Representative annotated imaged of the lung of a Clec9aConfetti mouse.

Movie S7. SI from a Clec9aConfetti mouse stained for cDC subsets.

Movie S8. Lung from a Clec9aConfetti mouse stained for cDC subsets.

Movie S9. Image an area of the lung of a Clec9aConfetti 1 dpi with influenza A virus and stained for viral proteins.

Movie S10. Representative annotated imaged of a high infiltrated area of the lung of a Clec9aConfetti 7 dpi with influenza A virus.

Movie S11. Representative annotated imaged of a low infiltrated area of the lung of a Clec9aConfetti 7 dpi with influenza A virus.

References (47, 48)


Acknowledgments: We thank members of the Immunobiology Laboratory for helpful discussions and suggestions. We thank the Crick Biological Resources, Flow Cytometry, and Microscopy Facilities for assistance, as well as M. Daly (Laboratory of Molecular Biology, Cambridge) and R. Bongaerts (Sony) for facilitating access to spectral flow cytometry. We thank A. Wack for kindly sharing reagents and expertise. Funding: This work was supported by The Francis Crick Institute, which receives core funding from Cancer Research UK (FC001136), the UK Medical Research Council (FC001136), the Wellcome Trust (FC001136), and by ERC Advanced Investigator (grants AdG 268670 and 786674), as well as, in part, by the Intramural Research Program of NIAID, NIH. M.C.-C. and J.v.B. were supported by Boehringer Ingelheim Fonds. B.U.S. is member of the DFG funded Collaborative Research Center 914 (project A11). Author contributions: M.C.-C., J.v.B., and C.R.e.S. designed experiments, analyzed data, and wrote the manuscript. M.C.-C. and J.v.B. conducted experiments with assistance from S.W., D.H., R.P.J., P.C., N.R., B.F., S.A., E.B., J.v.R., E.S., and F.K. for data analysis and study design. M.C.-C. and R.P.J. did the statistical analysis. H.C., M.B., M.G., and R.N.G. provided key reagents and advice. B.U.S. and N.R. helped with generating mice. C.R.e.S. supervised the project. All authors reviewed and edited the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.

Stay Connected to Science Immunology

Navigate This Article