|Home | About | Journals | Submit | Contact Us | Français|
Although, much progress has been made in our understanding of DC ontogeny and function, the transcriptional regulation of DC lineage commitment and functional specialization in vivo is poorly understood. We performed a comprehensive comparative analysis of CD8+, CD103+, CD11b+, and plasmacytoid DC subsets and the recently identified Macrophage DC precursors and Common DC precursors across the entire immune system. Here we characterize candidate transcriptional activators involved in myeloid progenitor commitment to the DC lineage and predicted regulators of DC functional diversity in tissues. We identify a molecular signature that distinguishes tissue DC from macrophages. We also identify a transcriptional program expressed specifically during steady-state tissue DC migration to the draining lymph nodes that may control tolerance to self-tissue antigens.
The Immunological Genome Project (ImmGen) is a consortium of immunologists and computational biologists from multiple institutes who have united to create an exhaustive database of gene expression and regulatory gene networks across the entire murine hematopoietic lineage using the same rigorously controlled data generation pipeline. Utilizing this extensive database, we sought to define the transcriptional profile and regulatory networks that control the dendritic cell (DC) lineage development homeostasis and function.
Discovered only fifty years ago, DC are the most recent addition to the hematopoietic cell lineage1. DC represent a small population of hematopoietic cells that share properties with tissue macrophages (MF), including their localization in most tissues, their ability to sample extracellular antigens, sense environmental injuries and contribute to the induction of tissue immune response1. However, in contrast to MF whose main role is to scavenge damaged cells or pathogenic microbes and promote tissue repair, the main function of DC is to initiate antigen specific adaptive immune responses against foreign antigens that breach the tissues 2 as well as maintain tolerance to self-antigens 3. The unique role of DC in adaptive immunity relies on their ability to process and present self and foreign antigens in the form of MHC-class II- and MHC class I-peptide complexes on the cell surface4,5 together with a superior ability to migrate to the tissue draining lymph nodes (LN) 6 and co-localize with T and B lymphocytes 7. This makes DC uniquely poised to control the induction of an antigen-specific immune response. Controversies, however, still exist as to the overall distinction between DC and MF due to partially overlapping phenotypse and functions and consequently the exact contribution of MF and DC to tissue immune responses remains debated 8,9
DC consist of distinct subsets with different abilities to process antigens, respond to environmental stimuli and engage distinct effector lymphocytes 10. Classical DC (cDC) form the predominant DC subset and are further subdivided into lymphoid tissue resident CD8+ cDC and CD8− cDC 11. Lymphoid tissue resident cDC subsets are functionally specialized and CD8+ cDC excel in the cross-presentation of cell associated antigens to CD8+ T cells, whereas CD8− cDC are the most potent at stimulating CD4+ T cells. The second major subset of DC is called plasmacytoid DC (pDC). pDC are uniquely potent at producing large amounts of the antiviral interferon alpha cytokine and initiate T cell immunity against viral antigens 12. Non-lymphoid tissue DC also include two cDC subsets, the CD103+ cDC and CD11b+ cDC 13. Similar to lymphoid tissue CD8+ cDC, non-lymphoid CD103+ cDC are efficient cross-presenters of cell-associated antigens and are the most potent at stimulating CD8+ T cells10, but may also facilitate the induction of T regulatory cells in the intestine 14.
The successive steps that lead to DC lineage commitment in the bone marrow are starting to be characterized. A myeloid precursors called macrophage and DC precursor (MDP) 15 was recently identified and shown to give rise to monocytes and to the common DC precursor (CDP) 16. CDP is a clonogenic precursors that has lost monocyte/macrophage differential potential and gives rise exclusively to pDC and cDC 17,18. CDP also produce pre-cDC, a circulating cDC restricted progenitor that has lost pDC differentiation potential 16 and home to tissues to differentiate locally into lymphoid tissue resident CD8+ and CD8− cDC16 and non lymphoid tissue resident cDC 19. Although, much progress has been made in our understanding of DC ontogeny and function, the transcriptional regulation of DC lineage commitment, diversification and functional specialization in vivo as well as the relationship between lymphoid and non-lymphoid tissue DC remain poorly understood. These questions remain unanswered due, in part, to the limited data available to perform comprehensive, comparative analysis both vertically and horizontally across the immune system.
This study deciphers the transcriptional network of the bone marrow derived DC precursors, the lymphoid tissue and non-lymphoid tissue DC as well as non-lymphoid tissue DC in a migratory state. The results of this study help characterize a DC-specific signature that distinguishes cDC from MF in tissues. Our study also identifies the lineage relationship between different tissue DC subsets as well as the predicted regulators of tissue DC diversity. Our results also uncover a common transcriptional program expressed by all non-lymphoid tissue cDC that migrated to the draining lymph nodes, regardless of their tissue or lineage origin.
We characterized 26 distinct DC populations isolated from primary lymphoid tissues, secondary lymphoid tissues and non-lymphoid tissues based on cell surface expression thought to represent discrete DC subsets with specialized immune function in vivo13 (Table 1). Each subset was sorted to high purity according to the ImmGen standard operating protocol. CD8+ and CD8− cDC and pDC were isolated from the spleen, thymus and LN, CD103+ and CD11b+ cDC were purified from the lung, liver, small intestine and kidney and the epidermal LC were isolated from the epidermis. Tissue migratory CD103+ and CD11b+ DC were isolated from tissue draining LN. Granulocyte macrophage precursors (GMP), macrophage DC precursors (MDP) and common DC precursors (CDP) were purified from the bone marrow, whereas circulating monocytes were isolated from the blood. Cell populations were double sorted based on the cell surface markers described in Table 1 to reach more than 99% purity. The final cytometric sorts (10,000 to 30,000 cells) were performed directly in Trizol, frozen after 2 minutes, and sent to the ImmGen core team in Boston. RNA was prepared from the Trizol lysate and hybridized to the microarray as described in 21. Expression profiling data were generated on Affymetrix ST1.0 microarrays per ImmGen pipeline, with data generation and quality control as detailed in 21. The purified DC subsets were isolated from laboratories located in New York (NY) and Boston (MA). One population of spleen DC (population 1, sorted in NYC, NY) was sorted based on MHC II and CD11c expression and lack of F4/80 and B220 expression and found to be identical to spleen DC (population 2, sorted in Boston, MA) purified based mainly on CD11c expression (Supplementary Note 2). To ascertain whether site or batch effects may confound the signals, CD8+ and CD8− spleen cDC were sorted independently at the two different locations (NY and MA). The data showed excellent correlation within each subset, with little evidence for lab-specific influences, in relation to the differences between CD8− and CD8+ cDC subsets (Supplementary Fig. 1).
Myeloid lineage commitment to the mononuclear phagocyte lineage is determined at the stage of the MDP, at which point erythroid, megakaryocyte, lymphoid and granulocyte fates have been precluded 15,16,22. DC commitment occurs during MDP transition to CDP, with the loss of monocyte potential 16, whereas commitment to cDC occurs at the pre-cDC stage with the loss of pDC potential 17,18.
We probed the regulator expression pattern along the myeloid-DC lineage tree to search for transcriptional activators and repressors that correlated with each differentiation step, and thus identified groups of regulators induced at different stages during DC differentiation (see materials and methods). The first group was up-regulated (fold change [FC]≥1.5) specifically during myeloid precursors commitment to the MDP potentially influencing global DC and MF development (Fig. 1a top). This group included the transcription factors Sox4 and Taf4b known to play a role, respectively, in cell fate and initiation of transcription. The second group of transcripts was up-regulated (FC≥1.5) during GMP-MDP transition to CDP but not during GMP-MDP transition to circulating monocytes (Fig.1a. bottom, left and middle panel) potentially promoting lineage differentiation to DC over monocytes. This group of regulators included the regulators Irf8, Bcl11a, Runx2, Klf8 and the zinc finger protein Zbtb46. A third group of regulator transcripts was down-regulated (FC≤0.67) during GMP-MDP transition and contained the TGF-β induced homeobox gene Tgif, the transcription factor Tcfec and the E3 ubiquitine ligase Trim13. To search for regulators that might contribute to DC lineage diversification, we examined the expression of candidate gene regulators during CDP differentiation to either pDC, lymphoid tissue CD8+ cDC or lymphoid tissue CD8− cDC (Supplementary Fig. 2, Fig. 1c). CDP differentiation into pDC was associated with the down-regulation of Id2, Zbtb46 and the TGF-β regulator Cited2 suggesting that induction of these factors likely contribute to lineage commitment to cDC differentiation. In contrast, CDP differentiation to cDC was associated with the down-regulation of Irf8, Tcf4, Runx2 and up-regulation of Batf3, Bcl6 and Ciita transcription factors. We further identified regulator transcripts up-regulated (FC≥1.5) in either CD8+ or CD8− cDC. These transcripts included expected genes such as the CD8+CD103+-specific transcription factor Irf819,23 as well as novel genes like hox factor Pbx1 gene transcript shown to function during definitive hematopoiesis in the fetal liver24.
Several transcription factors identified in our analysis have been shown to control DC development. For example, pDC and cDC differentiation is dependent on the zinc finger protein Ikaros25, the cytokine Flt3 ligand and its receptor Flt3 (ref 26), whose expression is partly controlled by Pu.1(Sfpi1) 27 and the transcription factor Stat3, which is activated upon Flt3 signaling and mediates Flt3 ligand-dependent DC differentiation28. Factors that regulate DC diversification are also starting to be identified. The transcription factors Tcf4 (E2-2), Spib, and Irf8 have been shown to control pDC differentiation 29 whereas Bcl6 has also been shown to control cDC but not pDC30 development in the spleen. The transcription factors Batf3 and Irf8, the inhibitor of DNA (Id) 2 and the mammalian target of rapamycin (Mtor)control the development of CD8+ and CD103+ cDC, while CD8− cDC differentiation is controlled by the transcription factors Irf2, Irf48 and Notch220, a factor that also controls the differentiation of intestinal CD103+CD11b+ cDC20. Consistent with our finding that Zbtb46 is expressed during transition to CDP and down-regulated during CDP differentiation into pDC, two studies published this week showed that Zbtb46 transcription factor is restricted to cDC committed precursors and tissue cDC and is absent from pDC, monocytes, macrophages and other myeloid and lymphoid cells31,32. Zbtb46 over-expression in bone marrow progenitor cells inhibited granulocyte potential and promoted cDC development strongly suggesting that Zbtb46 helps enforce cDC lineage commitment32 and may serve as a useful marker to distinguish between cDCs and other tissue phagocytes31,32.
Altogether, these results provide a map of known regulators but also unknown potential regulators that accompany key cellular checkpoints of DCpoiesis and help identify the molecular cues that control monocyte/DC lineage differentiation as well as DC diversification in vivo.
One of the major challenges to understanding the exact contribution of cDC versus MF in tissue immunity has been the lack of specific phenotypic markers to define tissue cDC. We first ascertained whether cDC and MF sorted based on published markers clustered as one or separate populations by performing a principal component analysis (PCA) of the top 15% most variable genes expressed by different steady-state leukocytes isolated from the same organ in the steady-state (Supplementary Fig. 1d). These results revealed that MF and cDC form distinct populations at the transcriptome level.
We then asked whether cDC expressed a set of transcripts that are present in all cDC subsets, but absent from MF. As non-lymphoid tissue CD11b+ cDC likely form a heterogeneous population8,9, we excluded non-lymphoid tissue CD11b+ cDC from the comparative analysis and asked whether lymphoid tissue CD8+ cDC and CD8− cDC and non-lymphoid tissue CD103+ cDC shared specific cDC transcripts that were absent from four prototypical MF populations profiled by the ImmGen group as described in methods. Twenty-four transcripts were differentially expressed (FC≥2; t-test false discovery rate [FDR]≤0.05) in cDC and absent from MF as determined by expression values below QC95, or the value at which the transcript has a 95% chance of being expressed (Fig. 2a, Table 2). This group of transcripts form the “core cDC signature” and includes the chemokine receptor Ccr7 shown to control cDC migration to the draining lymph nodes33, the zinc finger protein Zbtb46, which is first up-regulated at the CDP stage (Fig. 1a), and, Flt3 that encodes the receptor for the cytokine Flt3 ligand receptor known to control DC differentiation and homeostasis22,34. Although many of the genes transcripts enriched in cDC and absent in MF were also found in other hematopoietic cell populations, Zbtb46, Flt3, as well as Pvrl1 and Anpep (Cd13) were significantly up-regulated in cDC compared to all other hematopoietic cell subsets (Fig. S3; Table S1). Strikingly, many transcripts had no identified role in cDC biology, such as Kit, the receptor for kit ligand (stem cell factor) known for its role in hematopoiesis as well as mast cell differentiation35, and Btla (CD272), an Immunoglobulin superfamily member that attenuates B cell and T cell receptor-mediated signaling 36. Btla was specifically up-regulated in CD8+ CD103+ cDC populations in comparison to CD8− cDCs (Fig. 2a-b, Fig. S3, Table S1). Several transcripts were up-regulated (FC≥2; t-test FDR≤0.05) in cDC compared to MF including multiple class II genes and Dpp4 (CD26) (Fig. 2a, Table 3), whose role in DC function remains unclear. Using flow cytometry, we confirmed that Flt3, c-Kit, BTLA and CD26 were expressed at the protein level on spleen CD8+, spleen CD8− cDC as well as non-lymphoid tissue CD103+ cDC and absent from red pulp MF, lung alveolar MF, peritoneal MF and microglia (Fig. 2b). Altogether these results identify a core DC signature that helps distinguish tissue DC from MF in tissues.
DC subsets are classified based on distinct cell surface markers and different subsets exist in lymphoid and non-lymphoid tissues. To understand the relationship between these different DC populations, we performed a PCA of the top 15% most variable genes of lymphoid tissue pDC (spleen, skin draining LN and mesenteric LN pDC), lymphoid tissue cDC (LN, spleen, thymic CD8+ cDC; LN, spleen CD8−CD4+ cDC; spleen CD8-CD4-CD11b+cDC) and non-lymphoid tissue CD103+ cDC (lung, liver, and small intestine) populations. The main principal components identified three distinct DC clusters (Fig. 3a). One cluster was formed by lymphoid tissue CD8+ and non-lymphoid tissue CD103+ cDC (Fig. 3a). A second cluster was formed by lymphoid tissue CD8− cDC, whereas a third cluster was formed by pDC (Fig. 3a). We used these clusters to define specific gene expression signatures. The pDC cluster expressed 93 gene transcripts absent from other cDC, whereas the two cDC clusters expressed 125 gene transcripts absent from pDC including Zbtb46, Pvrl1, and Anpep (Cd13) (Fig. 3b; Supplementary Fig. 4; Supplementary Table 1). Further analysis of the two cDC subsets revealed 28 gene transcripts shared by CD8+ and CD103+ cDC and absent from other cDC (Fig. 3b; Supplementary Fig. 3-4; Supplementary Table 1). CD8+ and CD103+ cDC specific gene transcripts included Tlr3, the chemokine receptor Xcr1. In agreement with their unique TLR3 expression, CD8+ DC and CD103+CD11b− DC share a superior ability to respond to TLR3 ligand adjuvant 37-40. In addition, recent data revealed that CD8+ DC are also the only lymphoid tissue DC subset that produces interferon lambda in response to the TLR3 ligand Poly I:C 41 and similar results were found for lung CD103+CD11b− DC (Helft and Merad unpublished). Remarkably, Xcr1, which controls CD8+ T effector cell differentiation in mice and human 42,43, is expressed only in CD8+CD103+ cDC across the entire hematopoietic cell lineage (Supplementary Fig. 4c, Supplementary Table 1). Lymphoid tissue CD8− cDC populations co-expressed core cDC gene transcripts together with several monocyte and MF gene transcripts (Supplementary Fig. S4, Supplementary Table 1), which is consistent with recent results from our laboratory showing that CD8− cDC consist of two subsets that differentially expressed MF related genes20. However, some genes including Dscam (down syndrome cell adhesion molecule) were uniquely expressed by CD8− cDC and absent from macrophages and monocytes (Supplementary Fig. 4, Supplementary Table 1). Dscam is a molecule with enormous molecular diversity which plays a role in axon guidance 44 and also functions as an immuno-receptor able to recognize diverse pathogens in drosophila 45
To dissect the gene architecture program of these subsets of DC, we searched among the 334 fine modules of strongly co-expressed genes and predicted regulators identified for the entire ImmGen compendium (http://www.immgen.org/ModsRegs/modules.html) to identify those that were significantly up-regulated in specific DC subsets in comparison to the rest of the ImmGen samples (material and methods) (Fig. 3c-e). Module 150 was significantly up-regulated in pDC (p=4.77*10-11) and predicted regulators of this module included Irf8, Stat2, Runx2, and Tsc22d1 (Fig. 3c), which were also expressed during CDP commitment to pDC (Fig. 1b). Module F156 was significantly up-regulated in cDC (p=7.01*10-35) and enriched in core cDC genes (p=4.18*10-10; hypergeometric test) such as Zbtb46 and Pvrl1. Predicted regulators of this module included Batf3 and Relb (Fig. 3d), which were also up-regulated during DC commitment to cDC (Fig. 1b). Module 152 was significantly up-regulated in CD8+ CD103+ cDC (p=1.34*10-25) and enriched in CD8+ CD103+ DC transcript signature (p<*10-13, hypergeometric test) such as Tlr3, Xcr1 and Fzd1 (Fig. 3e). Irf8 and Pbx1 also identified during CDP commitment to CD8+ cDC (Fig. 1b) were predicted regulators of this module (Fig. 3e). Module 154 was significantly up-regulated in CD8− cDC (p=1.08*10-15) and in intestinal CD103+ CD11b+ cDC, a non-lymphoid tissue cDC subset recently shown to share development properties with lymphoid tissue CD8− cDC20 (Supplementary Fig. 5). These modules together with the core gene signature identify novel genes as well as potential regulators of DC functional specialization in vivo.
Non-lymphoid tissue CD11b+ cDC remain the least well characterized cDC subset both ontogenically and functionally. The small intestine is populated by three phenotypically distinct cDC subsets that differentially express the integrins CD103 and CD11b. CD103+CD11b− cDC and CD103+CD11b+ cDC are derived from the CDP and pre-DC 46,47, require FLT3 ligand for their development 47, migrate efficiently to the draining LN 48 and are thought to represent cDC 49. In contrast, the CD103− CD11b+ subset derives from circulating monocytes 46,47, develops independently of FLT3 ligand, requires CSF-1R ligand for its development 47 migrates poorly to the draining LN 48 and is thought to relate more closely to MF than to cDC 49.
To examine whether the core cDC signature identified above is differentially expressed by these subsets, we purified CD103+CD11b−, CD103+CD11b+ and CD103−CD11b+ small intestine (SI) cDC subsets as well as lung, liver and kidney CD11b+ cDC, and reran a PCA with the rest of the cDC subsets and with MF isolated from the spleen, lung, brain and peritoneum. CD11b+ cDC subsets were distributed across the PCA between the cDC and MF (Fig. 4a) and expressed a variable number of cDC core genes (Fig. 4b) indicating that non-lymphoid tissue CD11b+ cDC as currently defined, represent a heterogenous population.
CD103+CD11b− SI cDC clustered with the CD8+ CD103+ cDC, whereas lamina propria CD103+CD11b+ SI cDC clustered near lymphoid CD8− cDC and did not express CD8+ CD103+ cDC unique transcripts (Fig. 4a, Supplementary Fig. 6). Accordingly, CD103+CD11b− SI cDC expressed all CD8+CD103+ cDC specific transcripts identified (Fig. 3), including Fzd1 a Wnt receptor signaling which controls β-catenin activation and translocation to the nucleus (Supplementary Fig. 4, 6). Fzd1 was expressed specifically in CD103+CD11b− SI cDC and was absent from CD103+CD11b+ cDC and CD103−CD11b+ SI cDC (Supplementary Fig. 6). β-catenin activation and translocation to the nucleus can control the ability of DCs to promote T cell tolerance in the intestine50. It will be important to examine the contribution of Fzd1 and CD103−CD11b+ SI cDC to T cell immune modulation in the gut.
In contrast, the CD103−CD11b+ cDC clustered close to MF and away from other DC (Fig. 4a). Consistent with the PCA results, we found that CD103+CD11b− and CD103+CD11b+ SI DC expressed 100% of the core cDC genes as well as the cDC specific proteins c-Kit, Flt3, BTLA and CD26 on the cell surface (Fig. 4b-d, Fig. S6) suggesting that these two subsets belong to the DC lineage. In contrast, CD103−CD11b+ cDC clustered close to MF and away from cDC, they expressed only 40% of the core cDC genes and lacked the cDC proteins Kit, Flt3, BTLA and CD26 on the cell surface (Fig. 4b-d; Supplementary Fig. 6) suggesting that small intestine CD103−CD11b+ cDC belong to the MF lineage. Accordingly, a focused analysis of MF-associated transcripts indicated that the CD103-CD11b+ cDC SI population clustered with MF (Gautier et al., in preparation). Altogether, these results establish that the current DC phenotypic definition, which is based on MHC class II and CD11c expression, is not sufficient to identify tissue DC; and the use of the cDC gene signature provides a new means to distinguish CD11b+ cDC from MF in non-lymphoid tissues.
Tissue draining LN contain blood-derived DC that include pDC, CD8+ cDC and CD8− cDC also called LN resident DC as well as non-lymphoid tissue CD103+ cDC and CD11b+ cDC that have migrated from the drained tissue also called “tissue migratory cDC” 6. The mechanisms that control non-lymphoid tissue cDC migration and function in the draining LN in response to tissue injury or tissue immunization are starting to be unraveled; however, far less is known about the gene program that controls cDCs ability to leave peripheral tissues and migrate to the draining LN or the gene regulators that control migratory cDC immune function in the non-inflamed state 6. Here we analyzed the transcriptional program of tissue cDC prior to their migration to the draining LN (parent DC population) and upon migration to the LN as well as LN resident cDCs. Strikingly, we found that migratory cDC segregated together irrespective of their cellular or tissue origin and away from parent cDC populations that populate the drained tissue (Fig. 5a) and expressed a very similar transcriptional program (Fig. 5b-e). CD11b+ migratory cDC clustered together with CD103+ migratory cDC suggesting that among tissue CD11b+ cDC those that migrated in the steady state may represent the “bonafide” cDC. In addition, we found that in contrast to tissue CD11b+ cDC, which expressed moderate amounts of the DC specific gene Flt3, migratory CD11b+ cDC always expressed high Flt3 levels. Specifically, epidermal Langerhans cells (LC), which develop independently of Flt3 and Flt3L 19 and express very low amounts of Flt3 in tissue, dramatically up-regulated Flt3 transcripts once they reach the LN (Supplementary Fig. 7), suggesting that Flt3 plays a critical role in the homeostasis or function of steady state migratory cDC.
We also found that tissue migratory cDC up-regulated some gene transcripts dedicated to the dampening of immune responses (Fig. 6). As dampening genes can also be up-regulated in response to injury, we further compared steady state migratory cDC with Poly I:C activated cDC (Fig. 6a). As expected, Poly I:C activated and steady state tissue migratory cDC up-regulated the co-stimulatory gene transcript Cd40, previously reported on steady state migratory LC51 (Fig. 6a); however, steady state migratory cDC failed to up-regulate inflammatory cytokine transcripts (Fig. 6b) and expressed higher levels of immunomodulatory transcripts compared to Poly:I:C activated cDC (Fig. 6a). Immunomodulatory transcripts up-regulated in steady state migratory cDC included genes known to suppress T cell function either directly such as Cd274 (Pd-l1) 52, or via the production or activation of immunosuppressive cytokines such as the TGF-β activating integrin Itgb853. Other up-regulated transcripts encoded proteins known to reduce DC activation and cytokine production including Socs2, a TLR–responsive gene which regulates DC cytokine release via STAT3 modulation54, Pias3, also known to modulate STAT3 phosphorylation and NF-κB expression55, and Cd200, a protein known to reduce pro-inflammatory DC activation upon binding to its receptor, which is also expressed on DC56. Furthermore, steady state migratory cDC up-regulated transcripts important in reducing DC survival including cell death receptor Fas 57,58. Using flow cytometry and immunofluorescence analysis, we confirmed protein expression of Fas, Cd200, Pd-l1, and Pias3 and the co-stimulatory molecule Cd40 in steady state tissue migratory cDC (Fig. 6c-d). Based on these data we will speculate that cDC that leave the non-lymphoid tissues in the steady state upregulate a transcriptional immunomodulatory program that may prevent the induction of adaptive immune response to self- tissue antigens. The functional relevance of the immunomodulatory migratory DC signature needs to be confirmed experimentally.
This study provides the first comprehensive comparative analysis of DC precursors and tissue DC transcriptome across the entire immune system. The results of this study help identify: a transcriptional network that accompanies DC lineage commitment and diversification; a DC-specific signature that distinguishes cDC from MF in tissues; the relationship between lymphoid and non-lymphoid tissue DC subsets and predicted regulators of DC diversity; and a transcriptional immunomodulatory program expressed specifically during steady state tissue DC migration to the draining LN.
To gain knowledge of the transcriptional network that controls myeloid commitment to the DC lineage, we analyzed the transcriptional network associated with three key DC differentiation checkpoints: CMP ➔MDP, MDP ➔ CDP and CDP ➔ pDC CD8+ cDC or CD8− cDC. This analysis identified a group of transcriptional activators that include Zbtb46, Runx2, Bcl11a and Klf8 that rose specifically during MDP commitment to CDP but not monocytes suggesting their potential key role in driving myeloid commitment to the DC restricted precursors and away from monocytes in vivo. We also characterized the transcriptional networks that accompany CDP differentiation into pDC, CD8+ cDC and CD8− cDC subsets and identify several gene candidates that may drive DC lineage diversification in vivo.
One of the main controversies in the DC literature is the distinct contribution of cDC versus MF to tissue immunity. This confusion is partly a consequence of the paucity of markers available to distinguish between these two cell types leading researchers to use promiscuous markers such as MHC class II, CD11c and F4/80 to assess cDC or MF specific function9. We identified a core cDC gene signature shared by lymphoid tissue CD8− and CD8+ cDC, and non-lymphoid tissue CD103+ cDC and absent from tissue MF. cDC-specific transcripts included the gene coding for Zbtb46, Flt3, Kit and CCR7. The identification of the Kit transcript as part of the cDC gene specific signature is surprising, as Kit and its ligand have never been shown to play an intrinsic role in cDC development in vivo. Future studies will be needed to identify the role, if any, of Kit in DC differentiation, function and homeostasis in vivo. Importantly, the use of the cDC gene signature helped delineate the heterogeneity of non-lymphoid tissue CD11b+ cDC and identify a contaminating MF population, which could not have been detected using phenotypical markers currently used to define DC populations in vivo.
We also established that among cDC, lymphoid tissue CD8+ cDC and non-lymphoid tissue CD103+ cDC shared a gene signature, regardless of the tissue environment in which they reside. CD8+ CD103+ cDC gene signature was absent from the rest of DC including pDC, CD8− cDC and CD103− cDC in these same tissues. These results establish CD8+ CD103+ cDC as a distinct lineage subset and identify the gene regulators that may drive their differentiation, homeostasis and function. Using the novel algorithm termed Ontogenet developed for the ImmGen dataset (Jojic et al, in preparation), we identified modules of strongly co-expressed genes that were specifically and differentially expressed in each DC subset. Specifically we show that modules F150, F156 and F152 are up-regulated in pDC, cDC and CD8+ CD103+ cDC respectively and identify candidate regulatory programs that predict their expression pattern and therefore may drive DC functional specialization in vivo.
Strikingly, we found that regardless of tissue or cellular origin, non-lymphoid tissue CD103+ cDC and CD11b+ cDC as well as epidermal LC that migrate to the draining LN in the steady state up-regulate a shared gene signature. Some of the highest up-regulated transcripts are implicated in DC production of immunosupressive cytokine, the dampening of DC activation and reduction of DC survival that is known to lead to the dampening of T cell activation. These results are consistent with the potential role of steady state migratory cDC in the induction or maintenance of T regulatory response3 and identify candidate molecules that may participate in the control tolerance to self-antigens in vivo.
The results of this study provide a comprehensive characterization of the DC lineage transcriptional network and should help the development of novel genetic tools such as inducible gene regulation in vivo and lineage tracing of genetically marked defined myeloid precursor populations to further comprehend the developmental complexity of the phagocyte system. Moreover, the availability of the ImmGen datasets will now permit further investigations into DC gene expression networks and help unravel the transcriptional program that control DC function in the steady and injured state.
All cells analyzed in this study were obtained from six-week-old male C57BL/6J mice purchased from Jackson Laboratory with exception of LC and Mig LC which were isolated from Langerin EGFP C57BL/6J mice 59. All mice were housed in specific pathogen–free facilities at the Mount Sinai School of Medicine facility. Experimental procedures performed in mice were approved by the Animal Care and Use Committee of the Mount Sinai School of Medicine.
All cells were purified using the ImmGen Standardized Sorting protocol and antibodies listed on http://www.ImmGen.org. Sorting was performed at the Mount Sinai Flow Cytometry Shared Resource Facility using the Aria II (BD) or Influx (BD). Marker combination used to sort specific populations are available on http://www.ImmGen.org. Multiparameter analysis of stained cell suspensions were performed on LSRII (BD) and analyzed with FlowJo software (Tree Star, Inc). Monoclonal antibodies (mAbs) specific to mouse CD8 (clone 53-6.7), CD4 (clone L3T4), CD45 (clone 30-F11), CD11c (clone N418), CD11b (clone M1/70), I-A/I-E (clone M5/114.15.2), CD103 (clone 2E7), CD117/c-Kit (clone 2B8), CD135/Flt3 (clone A2F10), CD26 (clone H1940112), BTLA/CD272 (clone 6F7), CD40 (clone HM40-3), CD95/Fas (clone Jo2), CD200 (clone OX90), PD-L1/CD274 (clone MIH5), and the corresponding isotype controls were purchased from eBioscience or BD.
Viable sorted cells isolated according to MHCII and CD11c expression were sorted, Cytospun onto glass slides, and dried overnight. Slides were fixed for 1 hr with 4% paraformaldehyde in PBS, blocked with 10% goat serum in 0.1% Triton/0.1% BSA in PBS for 1 hr, then stained for 48 hrs at 4°C followed by 1 hr incubation at room temperature (RT) with the goat anti- mouse Pias3 mAb (Sigma, clone P0117) at 1:2000 dilution. Secondary goat anti-mouse Alexa Fluor594 (Invitrogen) was added to slides for 1 hr at RT. Slides were mounted with DAPI in Fluoro-gel with Tris buffer (Electron Microscopy Sciences). Images were acquired at 63× using the Zeiss Axioplan2IE with a Zeiss AxioCam MRc and analyzed using the Zeiss AxioVision software.
RNA was prepared from sorted cell populations from C57BL/6J mice using Trizol reagent as described60. RNA was amplified and hybridized on the Affymetrix Mouse Gene 1.0 ST array according to the manufacturer's procedures. Raw data for all populations were preprocessed and normalized using the RMA algorithm 61 implemented in the “Expression File Creator” module in the GenePattern suite 62. All datasets have been deposited at the National Center for Biotechnology Information/Gene Expression Omnibus under accession number GSE15907 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE15907). RNA processing and microarray analysis with the Affymetrix MoGene 1.0 ST array was prepared according to ImmGen standard operating procedures (http://www.immgen.org/Protocols/ImmGen).
Dataset was filtered for regulators 63 with a coefficient of variation (CV) <0.5 within population replicates. MDP regulators were selected for transcripts expressed 1.5 fold greater than GMP and CDP. CDP regulators were filtered for transcripts expressed 1.5 greater or 0.67 fold less than the precursors MDP and/or GMP. The “MDP and/or GMP” classification allows inclusion of transcripts that may be increasing in the MDP population as well and thus still informative. These were further filtered for expression at least 1.5 fold greater or 0.67 fold less than the nearest neighbor monocytes to find up-regulated or down regulated transcripts important in the DC lineage alone. Gene lists were input into the GenePattern module ExpressCluster (http://cbdm.hms.harvard.edu/resources.html) to identify patterns of expression across the GMP, MDP, CDP and monocyte populations.
Dataset was filtered for regulators with a CV <0.5 within these populations 63. Differentiated DC subsets were compared to their nearest developmental neighbor using a two-way analysis of variance Student's t-test on normalized expression data corrected with the Benjamini and Hochberg false discovery rate <0.05 and further selected for transcripts up-regulated by at least 1.5 fold within each organ. Thus, pDC were compared to CD8+/CD8− cDC and CD8+ cDC were compared to CD8− cDC within spleen, MLN, and SDLN DC and filtered for transcripts up-regulated at least 1.5 fold in each comparison. pDC and cDC were then compared to their precursor CDP to identify transcripts decreased at least 1.5 fold from the precursor. Transcripts up-regulated or expressed at the same level as CDP (red arrow) whereas transcripts downregulated compared to the CDP (blue arrows) are shown.
Differentially expressed transcripts were calculated using an unpaired two-way analysis of variance Student's t-test on normalized expression data. The t-test was controlled for multiple hypothesis using Benjamini and Hochberg false discover rate <0.05. The dataset was then filtered for those probes for which the fold change of any single population mean of the inclusion group over any single population mean of the exclusion group was ≥2 to create signatures of up-regulated transcripts. The dataset was further filtered for transcripts in which, the exclusion populations had an expression value less than the QC value of 95, or the level at which each population would have a predicted 95% certainty of expressing the gene. These gene signatures were also analyzed for variance across all steady-state leukocyte populations using the ANOVA command “aov “in R. Data were log transformed before analysis. Post-hoc pairwise Student's t- tests were run on each population corrected for multiple hypotheses testing using the Bonferroni adjustment. Data are provided in Table S1.
Dataset was filtered for transcripts with a CV <0.5 within population replicates. For creation of the migratory DC signature, the migratory LC, lung CD103+ cDC, and lung CD11b+ cDC samples were directly compared to their tissue resident equivalents and selected for transcripts with a FC≥2 which satisfied the Student's t-test p <0.05. To remove any potential signature that could be created by the lymph node environment, the remaining transcripts were compared in the migratory vs resident SDLN populations again selecting for transcripts at least 2 fold increased which satisfied the t-test p <0.05. This same method was applied to transcripts with a FC ≤ 0.5 which satisfied the t-test p <0.05 to identify a down-regulated signature. This analysis was performed in the GenePattern module Multiplot 64.
The expression data normalization was done as part of the ImmGen pipeline, March 2011 release. Data were log2 transformed. For gene symbols represented on the array with more than one probeset, only the probeset with the highest mean expression was retained. Of those, only the 7996 probesets displaying a standard deviation higher than 0.5 across the entire dataset were used for the clustering.
Clustering was performed by Super Paramagnetic Clustering 65 with default parameters, resulting in 80 stable clusters. The remaining unclustered genes were grouped into a separate cluster (C81). Those are referred to in the text as coarse modules C1-C81. Each coarse cluster was further clustered by hierarchical clustering into more fine clusters, resulting in 334 fine modules, referred to in the text as fine modules F1-F334. The expression of each gene was standardized by subtraction of the mean and division by its standard deviation across all dataset. Replicates were averaged. Mean expression of each module was projected on the tree. Expression values are color coded from minimal (blue) to maximal (red).
A novel algorithm termed Ontogenet was developed for the ImmGen dataset (Jojic et al, in preparation). Ontogenet finds a regulatory program for each coarse and fine module, based on regulators expression and the structure of the lineage tree. One sided two-sample Kolmogorov-Smirnov test was applied to the mean expression of each of the ImmGen fine modules to identify modules with significantly induced expression in specific cell groups. The cell groups were pDC, cDC, CD8- DC and CD8+/CD103+. Background for each was the rest of the ImmGen samples. Benjamini Hochberg FDR <= 0.05 was applied to the p-value table of all four groups across all fine modules. Hypergeometric test for two groups was used to estimate the enrichment of ImmGen fine modules for the four gene signatures. Benjamini Hochberg FDR <= 0.05 was applied to the p-value table of all four groups across all fine modules.
Signature transcripts were clustered and visualized with “HeatMap Viewer” or the “Hierarchical Clustering” tool using Gene Pattern 64. For hierarchical clustering, data were log scaled, centered around the mean, and clustered using Pearson correlation as a measure and pairwise complete-linkage clustering as a linkage type. Data were centered on rows before visualization. Principal component analysis (PCA) was performed using the Immgen PopulationDistances PCA program (http://cbdm.hms.harvard.edu/resources.html). Where indicated, the PCA program was used to identify the 15% most differentially expressed genes among subsets by filtering based on a variation of analysis using the geometric standard deviation of populations to weight genes that vary in multiple populations. Data were log transformed, gene and subset normalized, and filtered for transcripts with a CV<0.5 in each set of sample replicates before visualization. Fold change vs fold change and fold change vs t test p-value were visualized using the “Multiplot” module from Gene Pattern 64. Graphs of individual genes were created utilizing Prism Software.
We thank all of our colleagues in the ImmGen and especially the technical team including Micho Painter, Jeff Ericson and Scott Davis for their critical contribution. We are extremely grateful to Christophe Benoist for his critical contribution to the design and writing of the manuscript. The full list of the ImmGen Consortium participants can be found on the ImmGen website. We also thank eBioscience and Affymetrix for material support of the ImmGen Project. ImmGen is funded by R24 AI072073 from NIH/NIAID, spearheaded by Christophe Benoist. Additional support for the present body of work was funded by AI080884 and HL086899 to MM; JDRF172010770 and DP2DK083052-01 to B.D.B; DK074500 and AI045757 to S.J.T. HL69438, DK056638, HL097819 and HL097700 to P.S.F and U54CA149145 to V.J.