Langerhans cells are generated by two distinct PU.1-dependent transcriptional networks

Transcription factor expression in LCs

To better define the transcriptional network controlling LC differentiation, the expression of the transcription factors PU.1, ID2, IRF4, and IRF8 was analyzed by flow cytometry. PU.1 and ID2 expression was monitored using our reporter strains where an IRES (internal ribosome entry site)-eGFP cassette has been inserted into the 3′ untranslated region of the Sfpi1 and Id2 genes, generating PU.1^GFP and ID2^GFP reporters, respectively (Nutt et al., 2005; Jackson et al., 2011). PU.1 and ID2 were both constitutively expressed in epidermal LCs, and the expression of the latter increased on migration to the skin draining LNs (Fig. 1 A).

Homozygous knockin mice harboring eGFP fused to the C terminus of the IRF8 protein were used to follow IRF8^GFP expression (unpublished data). IRF8 was relatively lowly expressed in epidermal LC but strongly up-regulated in the migratory LCs in LNs (Fig. 1 A). Similarly, intracellular staining for IRF4 revealed that LCs did not express IRF4 in the epidermis; however, IRF4 expression was up-regulated in the LN LCs (Fig. 1 A).

PU.1 and ID2, but not IRF4 and IRF8, are required for the generation of LC

To rigorously assess the function of PU.1, ID2, IRF4, and IRF8 in LC differentiation, mice harboring floxed alleles of each gene were crossed to Itgax^cre mouse, whereby Cre-mediated deletion is under the control of the CD11c promoter (Caton et al., 2007). This led to the generation of four new mouse strains: Itgax^creSfpi1^fl/fl, Itgax^creIrf4^fl/fl, Itgax^creIrf8^fl/fl, and Itgax^creId2^fl/fl, hereafter named PU.1cKO, IRF4cKO, IRF8cKO, and Id2cKO, respectively.

The presence of LCs in the epidermis of each of the mouse strains was followed by flow cytometry. This showed an almost complete loss of epidermal CD11c⁺MHCII⁺ LCs in PU.1cKO and Id2cKO mice (Fig. 1 B). LC loss was also apparent after immunohistochemical staining for Langerin in epidermal sheets from adult mouse (Fig. 1 C). Deletion of either IRF4 or IRF8 in CD11c-expressing cells had no effect on the frequency of LCs found in the epidermis, a finding consistent with our observation that these transcription factors were very lowly expressed in epidermal LCs (Fig. 1, A and B). To further corroborate these findings, LC cells were enumerated from epidermal sheets of Irf4^−/− and Irf8^−/− mice. Consistently with our observations using cKO mice, neither factor was required for the generation of normal numbers of LCs (unpublished data). Although our finding contrasts with a previous study using Irf8^−/− mice that found a twofold reduction in the frequency of LCs (Schiavoni et al., 2004), our results are consistent with the recent findings on human harboring mutations in IRF8 disrupting its interaction with the DNA (Hambleton et al., 2011) and unpublished observation in the mouse that LC generation is largely independent of IRF8 (Hashimoto et al., 2011).

As expected, the frequency of migratory DCs was severely reduced in the peripheral LNs of PU.1cKO and Id2cKO mice (Fig. 1 D). In the case of PU.1, while all migratory DCs were reduced in number, the majority of the remaining Langerin⁺ cells found in the skin-draining LN coexpressed CD103, which marks a subset of migrating dermal DCs (Bursch et al., 2007; Fig. 1 E–G). In keeping with previous findings, both the CD103⁺Langerin⁺ dermal DCs and CD103⁻Langerin⁺ LC compartments were severely reduced in the absence of ID2 (Hacker et al., 2003; Ginhoux et al., 2009; Fig. 1, E–G).

Ablation of IRF8 did not strongly impair the egress of LCs from the skin, as the frequency of migratory LCs found in the LN mirrored the wild-type controls (Fig. 1 D). Importantly, migratory CD103⁺ dermal DCs were absent from the LNs of IRF8cKO, consistent with previous studies and confirming that CD11c-driven deletion was efficient at this locus (Fig. 1 E). In keeping with the up-regulation in IRF4 expression upon migration, the frequency of migratory LCs in peripheral LNs was greatly reduced in IRF4cKO (Fig. 1, D–G). Collectively, these data highlight the pivotal and DC-intrinsic role of PU.1 and ID2 for the development and or maintenance of epidermal LC.

In vitro generated LCs

To systematically investigate the role of different transcription factors during LC differentiation, an in vitro system that supports LC development was needed. TGF-β1 is critical for LC differentiation in vivo (Borkowski et al., 1996, 1997; Becker et al., 2011) and has been shown to cooperate with GM-CSF to drive the differentiation of CD14⁺ human monocytes into LCs (Geissmann et al., 1998). Consistently, Becker et al. (2011) have recently described an in vitro system, where the addition of TGF-β to BM precursors cultured in the presence of MCSF and GM-CSF could readily generate LC-like cells. To examine this differentiation system in more detail, BM cells were cultured with GM-CSF and TGF-β for 3 d. Examination of the CD11c⁺MHCII⁺ component of the resulting cultures showed the presence of four phenotypically distinct populations based on the expression of DEC205 and EPCAM, markers commonly used to delineate LCs (Fig. 2 A). While all four fractions arose from a TGF-β responsive precursor, which had undergone multiple cell divisions, their relative abundance was dependent on the TGF-β concentration (Fig. S1, A–C). Furthermore, GM-CSF was also found to be essential in this culture system, as DEC205⁺EPCAM⁺ DCs could not be generated from BM precursors cultivated in the presence of MCSF alone or supplemented with TGF-β (Fig. S1 D).

To better understand the relationship between these different populations induced by GM-CSF and TGF-β, the four fractions were sorted by flow cytometry and recultured for a further 24 h before reanalysis (Fig. 2 B). Fraction I, defined as DEC205⁻EPCAM⁻ retained multipotent potential as these cells generated all four subsets. Fraction II (DEC205⁺EPCAM⁻) failed to generate any other population and is likely to be terminally differentiated. Fraction III (DEC205⁻EPCAM⁺), differentiated predominantly into the DEC205⁺EPCAM⁺ fraction IV, while fraction IV cells maintained their identity but showed a further up-regulation of DEC205 (Fig. 2 B).

To characterize the lineage identity of these four different fractions, cells were sorted according to EPCAM and DEC205 cell surface expression. Cytospins of each population were stained with hematoxylin and eosin and analyzed for cellular morphology. Consistent with the above finding that fraction I maintained some developmental plasticity, this fraction retained a variety of cells in size and shape, while fraction II was highly enriched for cells with macrophage morphology. Fractions III and IV were more homogeneously small in size and presented small cytoplasmic protrusions (Fig. 2 C). Although Langerin expression is known to be relatively low in this culture system (Becker et al., 2011), we found that the mRNA for Cd207 (encoding Langerin) was exclusively found in Fraction IV, suggesting that this fraction contained cells with a more mature phenotype than those found in fraction III (Fig. 2 D). To further define the nature of these in vitro generated LC-like cells, CD103 expression was measured by flow cytometry. CD103 is a marker commonly used to distinguish LCs (Langerin⁺CD103⁻) from CD103⁺ DCs (Langerin⁺CD103⁺), although GM-CSF is known to induce CD103 expression in non-CD103⁺ DC subsets in vitro (Edelson et al., 2011; Jackson et al., 2011; Sathe et al., 2011). Despite CD103 expression being higher in fraction IV (Fig. 2 E), these cells lacked the transcriptional signature of CD103⁺ DCs, as the expression of the key transcription factor BATF3, and the chemokine receptor XCR1 was much lower than in conventional CD103⁺ and CD8⁺ DCs (Fig. 2 E; Hildner et al., 2008; Edelson et al., 2010; Crozat et al., 2011; Bachem et al., 2012). In keeping with their expression in LCs in vivo, the mean fluorescence index of ID2^GFP and IRF8^GFP increased in fractions III and IV, whereas PU.1 expression was uniform in all populations (Fig. 2 G).

In DCs, the TGF-β–induced response is mediated by the transcription factor RUNX3 (Fainaru et al., 2004). Although all cells were exposed to TGF-β, qPCR analysis revealed that Runx3 expression was restricted to fractions III and IV, confirming that these cells arose from a TGF-β responsive BM precursor (Fig. 2 H). In contrast, Klf4 (Krüppel-like factor 4), a transcription factor essential for macrophage function (Liao et al., 2011), was highly enriched in fraction II, whereas Irf4 showed a reciprocal expression pattern. (Fig. 2 H). Collectively, these observations highlight the fact that TGF-β–driven DEC205⁺EPCAM⁺ cells express a transcriptional profile and morphological features consistent with those of LCs.

In vitro generated LCs require PU.1 but not ID2

We next examined whether in vitro BM-derived LCs rely on the same transcriptional machinery as LCs generated in vivo by culturing BM cells from cKO mice cultured in the presence of GM-CSF and TGF-β. 3 d after stimulation, cells were harvested and stained for their surface expression of CD11c, MHCII, DEC205, and EPCAM. Generation of LCs in vitro was observed from IRF4cKO (not depicted), Irf4^−/− (IRF4 KO), and IRF8cKO BM, albeit for the latter to a lesser extent than from wild-type BM (Fig. 3 A). In contrast, PU.1 was required for the formation of DEC205⁺EPCAM⁺ BM-derived LCs. Surprisingly deletion of ID2 in CD11c-expressing cells did not impair the in vitro generation of BM-derived LCs (Fig. 3 A), in striking contrast to the loss seen in vivo (Fig. 1, B and D). Using a PCR-based assay, we confirmed the efficient Cre-mediated excision of the targeted PU.1, IRF8, and ID2 cKO alleles occurred in the CD11c⁺MHCII⁺ cells generated in these cultures (Fig. 3 B). These observations suggest that the in vitro BM-derived LCs represent a distinct subset that is dependent on PU.1 expression but independent of ID2.

PU.1 is required for LC repopulation of the skin under inflammatory conditions

Experimental evidence has pointed to a potential dual origin of the LC network. After UV exposure, epidermal LCs are severely reduced. The resulting inflammation leads to the recruitment of blood-borne monocytes to the injured skin where they differentiate into LCs (Ginhoux et al., 2006). Although replenishment of the LC network is dependent on MCSFR-driven signaling, differentiation is independent of IL34 (Greter et al., 2012; Wang et al., 2012).

Following from this, we propose that as inflammatory and steady-state LCs originate from distinct precursors, the transcriptional machinery guiding their differentiation is likely to be different. Before testing this hypothesis, it was necessary to first ascertain that blood monocytes subsets were not affected by the loss of PU.1 or ID2 in CD11c-expressing cells. Circulating CD11b⁺MCSFR⁺ cells can be divided into Ly6C^lowCD11c⁺CX3CR1^high resident monocytes and Ly6C^highCD11c⁻CX3CR1^low inflammatory monocytes (Geissmann et al., 2003; Fig. 4 A), with Itgax^cre activity being restricted to resident monocytes (Caton et al., 2007; Gaiser et al., 2012). Both populations were found in normal numbers in PU.1cKO and ID2cKO mice (Fig. 4, B and C), demonstrating that any Cre-mediated gene deletion in monocytes did not impact on their frequency and thus potential to respond to the inflammatory signals that induced LC repopulation. To further characterize those monocytes, expression of PU.1^GFP and ID2^GFP within those populations was monitored by flow cytometry. Both populations expressed low levels of ID2 compared with splenic DCs, whereas PU.1 expression in inflammatory monocytes was similar to the levels of splenic cDCs, while resident monocytes expressed higher PU.1 (Fig. 4 D).

We then tested whether either PU.1 or ID2 were required for the differentiation of BM-derived LCs after skin injury induced by UV exposure. To this end, lethally irradiated C57BL/6 Ly5.1⁺ mice were reconstituted with wild-type, PU.1cKO, or ID2cKO BM (all Ly5.2⁺, Fig. 5 A). 6 wk after reconstitution, mice were exposed to UV light for 30 min to deplete host LCs. 3 wk after UV exposure, the recruitment of donor Ly5.2⁺ cells into the epidermis was monitored by flow cytometry. As expected, BM-derived Ly5.2⁺ LCs were generated in Ly5.1⁺ mice reconstituted with wild-type BM, whereas host Ly5.1⁺ LCs were severely reduced (Fig. 5 B). Analysis of the ears, trunk epidermis, and LNs of mice reconstituted with Ly5.2⁺ wild-type, PU.1cKO, or ID2cKO BM revealed a similar the degree of chimerism regardless of donor genotype (Fig. 5 C). Mice reconstituted with Ly5.2⁺ PU.1cKO BM failed to differentiate into CD11c⁺MHCII⁺ epidermal LCs, highlighting the pivotal role of PU.1 in driving their differentiation from a BM-derived precursor (Fig. 5 D). In contrast, LCs were generated in Ly5.1⁺ mice reconstituted with Ly5.2⁺ ID2cKO BM (Fig. 5 D). These results perfectly mirrored the results generated using the in vitro LC differentiation system and demonstrate that in contrast to steady-state LC where ID2 is indispensable, inflammatory BM-derived LCs are insensitive to ID2 loss.

Consistent with the above findings, Langerin⁺ cells in peripheral LNs were severely reduced in mice reconstituted with PU.1cKO, whereas they were only slightly reduced in mice reconstituted with ID2cKO BM (Fig. 5 E). Analysis of the cell surface markers expressed by migratory DCs revealed that the vast majority of Langerin⁺-expressing cells derived from ID2cKO are CD103⁻ and therefore represent LCs (Fig. 5 E). In contrast, mice reconstituted with Ly5.2⁺ PU.1cKO BM contained Langerin⁺CD103⁺ dermal DCs and few epidermal LCs (Fig. 5 E).

To determine whether the LCs generated from ID2cKO BM could persist longer-term, we repeated the UV-induced skin injury model described above (Fig. 5 A) and analyzed the LC repopulation 6 wk after UV treatment. ID2cKO-derived LCs (Fig. 5 F), which had efficiently excised the targeted ID2 allele (Fig. 5 G), were still present at the time point, indicative of sustained LC reconstitution.

To gain insights into differential requirement of LCs generated after UV induced skin injury for PU.1 and ID2, we reconstituted lethally irradiated Ly5.1⁺ mice with either PU.1^GFP or Id2^GFP BM (both Ly5.2⁺). 6 wk after reconstitution, mice were exposed to UV, and 3 wk later the epidermis was isolated and analyzed by flow cytometry. Donor-derived cells were able to differentiate into three distinct subsets based on the surface expression of the markers CD11c and MHCII (Fig. 6 A). Both CD11c⁻MHCII⁻ and CD11c⁺MHCII⁻ cells lacked PU.1^GFP and expressed ID2^GFP constitutively, whereas the CD11c⁺MHCII⁺ LCs expressed high levels of PU.1 and ID2. Thus, despite the induction of ID2 expression in BM-derived LCs, ID2 deletion did not affect LC differentiation in this setting, whereas PU.1 was found essential (Fig. 5, D–G).

Recently, it has been reported that during inflammation two distinct types of LCs seed the epidermis: a short-term LC, which persisted for up to 4 wk and was ID2 independent, and a long-term ID2-dependent LC (Seré et al., 2012). In that study, the short-term LC subset exhibited a distinct phenotype characterized by the lack of Langerin expression. In contrast the ID2cKO LCs that were present in the epidermis 3 wk after UV treatment in our experiments expressed Langerin in a manner indistinguishable from wild-type LCs (Fig. 6 B). As we have shown that all CD11c⁺MHCII⁺ cells expressed high PU.1^GFP (Fig. 6 A) and that MHCII and Langerin expression were exactly coincident (Fig. 6 B), we conclude that in our model all epidermal LCs were CD11c⁺MHCII⁺Langerin⁺ and PU.1^GFP+. Conversely the CD11c⁺MHCII⁻Langerin⁻ cells lacked PU.1^GFP (Fig. 6) and instead expressed TCR-β or -γδ (Fig. 6 C).

These observations highlight that the LC network relies on two distinct transcriptional pathways, both of which are PU.1 dependent. In addition, these results demonstrate that in vitro BM-derived LCs can be used as a surrogate system to study the molecular mechanisms involved in inflammation-induced LC differentiation in vivo, as LC differentiation in both models is PU.1 sensitive and ID2 insensitive.

PU.1 regulates RUNX3 expression in LCs

To further determine how PU.1 affects the differentiation of BM-derived LCs, we focused our analysis on the transcription factor RUNX3, which is pivotal in mediating LC differentiation in response to TGF-β (Fainaru et al., 2004). Real-time RT-PCR analysis showed that Runx3 expression was strongly reduced in PU.1-deficient DC generated in vitro (Fig. 7 A). To determine whether PU.1 was able to directly regulate Runx3 expression, BM cells from PU.1cKO were cultured in the presence of GM-CSF and TGF-β, and transduced with retroviral vectors expressing either GFP alone or with PU.1. Strikingly, RUNX3 expression was significantly increased in CD11c⁺MHCII⁺ DCs ectopically expressing PU.1 compared with DCs expressing GFP alone (Fig. 7 B). LC numbers were increased in vitro when PU.1cKO cells were transduced with PU.1, suggesting that PU.1 controls Runx3 expression in LCs (Fig. 7 C).

To examine whether PU.1 could regulate Runx3 expression by direct binding to its regulatory elements, PU.1-bound DNA regions in myeloid cells were retrieved from the online available database (Ghisletti et al., 2010). PU.1 bound to several intronic and 3′ regions of the Runx3 gene, most of which contained the PU.1 consensus core motif (GGAA, Fig. 7 D). In addition, in silico analysis of Runx3 promoter showed other potential PU.1 binding sites. To test whether PU.1 directly binds in the Runx3 gene in LCs, PU.1-bound chromatin was immunoprecipitated from wild-type cells cultured in the presence of GM-CSF ± TGF-β. Consistent with the ChIPseq data from macrophages, PU.1 was constitutively bound to the first intron of Runx3. Interestingly, although no binding of PU.1 to the Runx3 promoter was observed in BM cells cultured in GM-CSF alone, the addition of TGF-β to the media induced the recruitment of PU.1 to Runx3 promoter and also increased its binding to the intronic region (Fig. 7 E). Collectively, these observations suggest that PU.1 directly regulates Runx3 expression in a TGF-β–dependent manner and thereby promotes LC differentiation.

RUNX3 expression rescues LC differentiation in the absence of PU.1

To test if the requirement of PU.1 for LC generation could be circumvented by constitutive expression of RUNX3, PU.1cKO BM cells were cultured in GM-CSF ± TGF-β and transduced with retroviral vectors expressing either GFP alone or with RUNX3. Strikingly, restoring RUNX3 expression could bypass the requirement of PU.1 to drive LC differentiation. Retroviral transduction of RUNX3 into wild-type BM precursors also increased the numbers of LCs generated by twofold when compared with control cultures, suggesting that the expression of RUNX3 is normally limiting for LC differentiation in vitro (Fig. 8 A).

A cardinal feature of the RUNX proteins is their highly conserved Runt domain, which is essential for DNA binding and protein–protein interaction (Ito, 2004). Alignment of the sequence coding for the Runt domains of RUNX1 and RUNX3 showed very high homology. The critical residues for RUNX1 DNA binding (arrows) were almost exclusively retained in the predicted DNA binding domain (box, Fig. 8 B), and were highly conserved in RUNX3 (Michaud et al., 2002; Li et al., 2003). Therefore, based on these similarities, a single point mutation within the last DNA contacting residue of RUNX3 (RUNX3^R196Q) was introduced and enabled to address the functionality of such mutant in promoting LC differentiation. Constitutive expression of RUNX3^R196Q failed to promote enhanced LC differentiation in wild-type cells or to rescue the deficiency of PU.1cKO to differentiate into LCs (Fig. 8 C). Collectively, these results highlight the critical role for RUNX3 DNA binding in promoting LC differentiation and support the hypothesis that RUNX3 levels are a limiting factor for LC differentiation.