Innate and adaptive cytokines associated with TH1/17/22, but not TH2, are overexpressed in rosacea
In order to identify potential early instigators of rosacea skin inflammation, we analysed skin biopsies from rosacea patients taken either during acute flare-ups or during stabilised chronic inflammation (Suppl. Table 1) and compared them to skin from healthy donors.
First, we confirmed overexpression of the cathelicidin transcript hCAP18 in rosacea (Suppl. Fig. 1a), as described in a previous report.13 We then analysed mRNA expression levels of selected innate and adaptive cytokines involved in the pathogenesis of different inflammatory skin diseases to identify unique expression patterns. Among innate cytokines, we observed a significant increase in TNF, IL1B, IL8, IL12B, IL23A, and IL10 but not IL36B in rosacea skin lesions as compared to skin from healthy donors (Suppl. Fig. 1a and Fig. 1a). Furthermore, we could observe a trend increase in IL6 expression, and bimodal distribution for IFNA2 and IFNB1. Adaptive T cell-derived cytokines IFNG, IL17A, IL17F, and IL22 were significantly overexpressed in lesional rosacea skin while IL4 and IL13 were either undetectable or not increased (Suppl. Fig. 1b and Fig.1b). These results are in line with previously published data22 indicating a predominant TH1/TH17-related inflammation in rosacea. When comparing acute flare-ups with stabilized chronic lesions of rosacea in order to identify potentially unique patterns, we found a selective and significant increase in expression of type I IFNs IFNA2 and IFNB1 in flare-ups (Fig.1a). Interestingly, whereas type I IFNs were selectively upregulated in rosacea patients during acute flares, the type I IFN response gene MX1 was upregulated in all rosacea samples as compared to healthy skin (Suppl. Fig. 1c and Fig. 1c) suggesting that type I IFNs are transiently produced early during acute flare-ups of rosacea. All other cytokines revealed comparable expression levels between acute flare-ups and chronic stabilized lesions though there was a tendency for an increased expression of TNF in more chronic rosacea lesions (Fig.1a and 1b).
Taken together, these findings indicate the following: first, rosacea is characterised by predominant TH1/TH17 signatures, and second, rosacea is driven by early overexpression of type I IFN during acute flare-ups, which then is subsequently relayed by TNF and a chronic inflammation.
Type I interferon overexpression correlates with pDC infiltration in flares of rosacea
As type I IFNs are preferentially expressed by pDCs, we investigated their presence in rosacea by staining paraffin-embedded sections for CD123 (IL3RA). Lymphoid cells with strong CD123 expression represent bona-fide pDCs as shown by co-expression of BDCA2 and absence of CD11c by microscopy (Fig. 2a). While largely absent in facial skin of healthy donors, we found high numbers of pDCs within the dermal infiltrate of rosacea (Fig 2b). When plotting the percentages of CD123+ pDCs (Fig. 2b) along with IFNA2 (Fig.1a) and IFNB1 expression (not shown) we found statistically significant monotonic positive correlation between the two (Fig. 2d; IFNA2: r=0.7363, p=0.0011; IFNB1: r=0.7951, p=0.0002), indicating that pDCs represent the principal source of type I IFNs during flare-ups of rosacea.
Type I interferon expression is critically dependent on pDCs in a mouse model of rosacea
To investigate this functionally, we took advantage of a previously described in vivo model of rosacea.13 Briefly, the cathelicidin antimicrobial peptide LL-37, which is overexpressed in skin of rosacea patients (Suppl. Fig 1; and13), is injected intradermally four consecutive times over a period of 48h. We found that, upon injection of LL-37, pDCs are rapidly recruited into the skin and accumulate over time (Fig. 3a). The pDC accumulation significantly correlated with gene expression of Ifna1 (r=0.7056, p>0.0001) and Ifnb1 (r=0.6784, p<0.0001; Fig 3b). Moreover, depletion of pDCs by anti-Pdca1 antibodies completely abolished Ifna1 expression and largely abrogated Ifnb1 expression in the skin (Fig. 3c). Congruently, induction of downstream type I IFN stimulated genes Mx2 was significantly reduced and entirely abolished for Isg15 and Irf7 (Suppl. Fig. 2a). We obtained similar results using a different pDC depletion system, the BDCA2-DTR transgenic mouse (Suppl. Fig. 2), confirming in vivo that pDCs are the principal source of type I IFNs. Interestingly, blocking type I IFN signalling by an anti-Ifnar1 antibody not only blocked type I IFN signature genes but also significantly reduced the numbers of pDCs infiltrating the skin (Fig. 3d). These results indicate that the continuous overexpression of LL-37 in rosacea increases type I IFN production by pDCs, which further amplifies their skin infiltration and sustains in situ production of type I IFNs. Indeed, Cxcr3-family chemokines Cxcl9, Cxcl10, and Cxcl11, which mediate pDC migration into the skin, were induced by LL37, and their expression was highly dependent on pDCs and Ifnar signalling (Suppl. Fig 3).
Type I interferon blockade and pDC depletion abolish select TH1/TH22 signatures
Next, we sought to analyse the functional role of pDC-derived type I IFNs in driving pathogenic inflammation in rosacea. First, we investigated if the in situ cytokine expression profile in the mouse model indeed reflects human rosacea. The expression of pro-inflammatory innate cytokines, including TH1- polarising cytokines Il12a and Tnf, TH17-/22-polarising cytokines Il1b, Il23a, and Il6, as well as Il10 were significantly induced as compared to controls showing a similar expression profile as in rosacea (Fig.4a). Overexpression of Tnf, Il1b, Il23a, and Il6 was critically dependent on pDCs and type I IFN signalling, while induction of Il36b and Il10 expression was independent of either. All adaptive cytokines tested, including Il17a, Il17f and Il22, Il4 and Il13, and Ifng, were significantly induced upon LL-37 injection. However, only the overexpression of Il17f and Il22, but not of TH2 and TH1 family cytokines, was dependent on pDCs and type I IFN signalling (Fig. 4b). These findings indicate that LL-37 overexpression in rosacea activates skin-infiltrating pDCs to produce type I IFN, which is required to drive a TH17-/22-related inflammation.
Cathelicidin peptides found in rosacea display differential potencies to bind and complex DNA and to induce type I interferons
It is widely established that LL-37 is able to activate pDCs to produce large amounts of type I IFNs by internalising extracellular nucleic acids. In rosacea, however, cathelicidin is processed into several additional C-terminal peptides besides LL-37, such as FR-29, FA-29, and DI-27 (Suppl. Fig. 4; and13). These, despite considerable overlap (Suppl. Fig. 4b) display significant differences in terms of hydrophobic surfaces (Suppl. Fig. 4a-c), charge, and percentage of predicted alpha-helical structures (Suppl. Fig. 4d). Thus, we wondered whether these peptides had a differential ability to activate pDCs. We isolated human pDCs and stimulated them with complexes of DNA and the different cathelicidin derived peptides. When complexed with DNA, FR-29 showed significantly increased IFNα production by pDCs as compared to LL-37 for the same molar concentration, (Fig.5a). Neither FA-29 nor DI-27 were able to stimulate detectable amounts of IFNα from pDCs in vitro. When reducing the concentration of peptides below the stimulatory capability of LL-37 to 3μM, FR-29 retained significant – albeit reduced – stimulatory capacity. To analyse the underlying mechanism of differential IFNα production, we first assessed the DNA-binding capacity of these peptides by picogreen fluorescence quenching. Indeed, FR-29 bound and condensed DNA more potently than LL-37 (Suppl. Fig. 5). Conversely, FA-29 and DI-27 had almost indiscernible DNA binding activity. To further assess the capacity of FR-29 and LL-37 to internalise DNA into pDCs, we fluorescently tagged purified, undigested DNA, and complexed it with the cathelicidin peptides LL-37 and FR-29. Whereas pDCs were unable to uptake DNA in the absence of cathelicidin peptides, LL-37 and FR-29 both showed uptake of fluorescently labelled DNA by ca. 10% of cells (Fig.5b). However, the relative uptake of DNA was markedly more prominent when complexed with FR-29 than with LL-37 as exemplified by the increased fluorescence intensity. Taken together, these data indicate that several fragments found in rosacea such as FA-29 and DI-27 cathelicidin peptides fail to activate pDCs, whereas FR-29 has a significantly more potent effect than LL-37 due to increased affinity to nucleic acids.
To further evaluate the in vivo relevance of these findings, we analysed Ifnα gene expression in the rosacea mouse model. Indeed, intradermal injection of FR-29 induced significantly more potent Ifna1 production than LL-37 (Fig.5c). pDC infiltration of the skin was comparable between FR-29 and LL-37 suggesting that enhanced Ifna1 production upon FR-29 injection is due to a differential activatory capacity of the cathelicidin peptides rather than increased skin infiltration by pDCs (Fig.5d).
Kallikrein 5 recapitulates pDC-dependent type I interferon expression
Cathelicidin pro-peptide processing is required for LL-37 activity in the skin. Kallikrein 5 (KLK5), which is overexpressed in rosacea,10,13 mediates processing of cathelicidin into the active LL-37 peptide. To determine whether KLK5 overexpression is sufficient to instigate activation of the LL-37/pDC/type I IFN axis, we took advantage of a recently described transgenic murine model expressing human KLK5 in the epidermis under the K5-promoter (Tg.KLK5; and23). First, we confirmed in vitro and in vivo that human KLK5 is able to cleave Cramp pro-peptide, the murine analogue of human cathelicidin. In vitro, human KLK5 effectively processed murine Cramp pro-peptide as shown by western blot (Fig. 6a). In addition, spontaneously after birth, we found significantly increased expression of Cramp pro-peptide and increased presence of the cleaved active form of Cramp in the skin of Tg.KLK5 mice as compared to wild type littermates (Fig. 6b). These findings indicate that constitutive overexpression of KLK5 in the skin leads to accumulation of active cathelicidin peptides similar as seen in rosacea. Next, we investigated the presence of pDCs in skin of these mice. Seven days after birth, pDCs accumulated significantly and spontaneously in Tg.KLK5 mice but not control mice (Fig. 6c), and were paralleled by type I IFN overexpression (Fig. 6d). Whereas type I IFNs were upregulated early in the skin and declined over time, TH17-related cytokines Il22, Il17a, and Il17f were significantly overexpressed only at a later time point indicating that type I IFNs might drive TH17 response in Tg.KLK5 mice. Depletion of pDCs from birth led to loss of induction of type I IFN in the skin, confirming pDCs as the principal source of type I IFN in the Tg.KLK5 mouse model (Fig. 6e). Furthermore, blockade of type I IFN signalling significantly inhibited expression of TH17/TH22 cytokines Il22 and Il17f, in line with data obtained in the LL-37 intradermal injection model (Fig. 6f). Taken together, these findings indicate that KLK5 overexpression in rosacea increases cleavage of cathelicidin in the skin. In turn, accumulating active cathelicidin peptides attract and activate pDCs resulting in a type I IFN-driven TH17/TH22 immune response characteristic for rosacea.
Cathelicidin peptides kill bacteria associated with rosacea leading to activation of pDCs
It has been shown that complexes of host DNA and antimicrobial peptides such as LL-37 play an important role in the pathogenesis of several chronic inflammatory diseases such as psoriasis, lupus erythematosus, or atherosclerosis.16,24,25 Because rosacea severity has been linked to dysbiotic communities of commensal skin bacteria, in particular Bacillus oleronius and Staphylococcus epidermidis, we wondered whether commensal microbial DNA rather than host DNA could play a pathogenic role in driving inflammation in rosacea. First, we investigated killing of different skin and non-skin associated bacteria by LL-37. B. oleronius was the most susceptible, followed by the other skin-associated bacteria S. epidermidis and C. acnes (Fig. 7a and b). Gut and lung-associated bacteria, though still subject to killing, were far less susceptible, indicating a preferential killing of skin commensals. When comparing the different rosacea-associated cathelicidin peptides, we found that also FR-29 and FA-29 but not DI-27 were able to kill all bacteria tested (Suppl. Fig. 6). Similar to its ability to condense DNA and activate pDCs, FR-29 showed a higher potency than LL-37 in killing B.oleronius and S.epidermidis, while the latter was slightly more potent or at least equally efficient in killing the other strains of bacteria tested (Suppl. Fig. 6). These data indicate that the different cathelicidin peptides found in skin of rosacea patients show a slightly different spectrum of antimicrobial activity. Moreover, FA-29, which is not able to bind and condense DNA, efficiently killed several strains of bacteria at similar concentrations as LL-37. Thus, the ability of cathelicidin peptides to condense DNA and activate endosomal TLRs is not directly linked to their antimicrobial capacity to kill bacteria.
Next, we tested if LL-37-killed B.oleronius could serve as DNA source and activate pDCs in vitro. We isolated human pDCs from healthy donors and cultured them in presence of live bacteria with or without LL-37. Whereas B.oleronius alone was not able to activate pDCs, the addition of LL-37 allowed for potent activation of pDCs leading to important production of IFNα (Fig. 7c). In accordance with our previous results, FR-29 was able to induce significantly more IFNα by pDCs than LL-37 when in the presence of B.oleronius, whereas FA-29 and DI-27 displayed poor capacity to activate pDCs (Fig. 7d).
To further investigate the in vivo relevance of these findings, we took advantage of the rosacea mouse model and injected heat-killed B.oleronius alone or pre-incubated with LL-37. Interestingly, B.oleronius alone was not able to induce type I IFN expression within 24h. Pre-incubated with LL-37 however, B.oleronius induced significantly more type I IFN expression than LL-37 alone (Fig. 7e). These finding indicate that an increased bacterial load alone is not sufficient to rapidly engage the type I IFN pathway but that bacterial killing by cathelicidin peptides is required. As B.oleronius further increased type I IFN expression within the skin upon LL-37 injections, we wondered if commensal skin bacteria are actually a prerequisite to activate pDCs and induce IFNα production in our rosacea model. Thus, we pre-treated mice with a topically applied wide-spectrum antibiotics for 2 days, prior to LL-37 injections. We found that type I IFN expression was almost completely abolished upon antibiotic treatment of the skin (Fig. 7f), whereas injection of B.oleronius with LL-37 was sufficient to restore and enhance type I IFN even further.
Thus, skin commensal bacteria are required but, by themselves, not sufficient to rapidly activate pDCs and induce type I IFN in rosacea. On the other hand, B.oleronius is sufficient for cathelicidin-driven overexpression of type I IFN in the skin. Taken together, these findings indicate a functional role for dysbiotic communities of commensal bacteria in triggering a pathogenic immune response and flare-ups of rosacea.
Type I interferon and IL22 co-operate in the induction of pathogenic angiogenesis
One of the hallmarks of rosacea is recurrent episodes of flushing and eventually fixed centro-facial erythema, which is caused by neoangiogenesis and telangiectasia. Therefore, we wondered if the KLK5‒LL-37‒pDC‒IFNα‒TH17/22 axis could play a pathogenic role in driving neoangiogenesis in rosacea. First, we assessed the number of dermal ECs as well as their proliferative status in the rosacea mouse model. LL-37 injections significantly increased the number of ECs (Fig. 8a) and the percentage of proliferating ECs (Fig. 8b) as compared to control injections. As FR-29 injections further augmented numbers and proliferation of ECs, we questioned if the pro-angiogenic effect of cathelicidin peptides14 might be through activation of the IFNα-TH17/22 pathway. Indeed, blockade of type I IFN signalling significantly reduced both the overall numbers and percentage of proliferating ECs (Fig. 8c). To further dissect the mechanism, we sought to determine the effect of IFNα on primary ECs, which are dependent on growth factors for survival and proliferation in culture. While EC growth factors maintained viability and markedly increased cells numbers in vitro, IFNα did not have any discernible direct effect on EC viability and proliferation (Suppl. Fig. 7a-c). So, we tested if type I IFN-induced downstream TH17/22 cytokines were instead able to promote EC proliferation. Neither IL17 nor IL22 alone had a direct effect on EC viability. However, ECs cultured in the presence of both IFNα and IL22 showed maintained viability and significantly increased cell numbers. Interestingly, this effect was specific for IL22 as IL17A did not augment EC viability when combined with IFNα (Fig. 8d and Suppl. Fig. 7c and 7d). As IFNα is required for IL22-driven EC proliferation, we wondered if IFNα mediated its effect via upregulation of the IL22-receptor on ECs. The expression of IL22RA1 – the receptor subunit specific for IL22 – was absent on unstimulated ECs but greatly and significantly induced upon stimulation with IFNα (Fig. 8e). On the other hand, IL17RA and IL17RB, which form the active receptor for IL17, as well as IL10RB, the second subunit of the IL22 receptor, were constitutively expressed and not altered by addition of IFNα or growth factors. These results show that IFNα renders ECs susceptible to IL22 through upregulation of its cognate receptor, thereby enabling IL22-driven proliferation. To verify this in vivo, we assessed EC numbers and proliferation upon IL22-blockade in the rosacea mouse model. Indeed, inhibition of IL22 significantly reduced the numbers and percentage of proliferating ECs in the skin (Fig. 8h) similar to type I IFN blockade (Fig 8c). LL-37 injections led to neoangiogenesis and formation of visible microvessels, as assessed by in vivo videocapillaroscopy, which resembled images taken from human rosacea patients (Suppl. Fig. 7e). These microvessels were significantly reduced upon Ifnar blockade and inhibition of Il22 (Fig. 8g). In summary, overexpression of cathelicidins induces vascular EC proliferation and neoangiogenesis in rosacea via induction of pDC-derived type I IFN, which in turns leads to upregulation of IL22 and expression of the IL22-receptor on EC thereby enabling IL22-driven neoangiogenesis.