Knockout of the Arp2/3 complex in epidermis causes a psoriasis-like disease hallmarked by hyperactivation of transcription factor Nrf2

The actin-related protein (Arp2/3) complex consists of two actin-related proteins (Arp2 and Arp3; also known as Actr2 and Actr3, respectively) and five additional actin-related protein complex subunits (Arpc1-5) (Goley and Welch, 2006). The Arp2/3 complex catalyses the polymerization of monomeric actin [globular (G)-actin] into branched filaments [filamentous (F)-actin] thereby generating branched F-actin networks and mechanical forces in the cell (Goley and Welch, 2006; Rotty et al., 2013).

The genome of most eukaryotes contains a single gene encoding each Arp2/3-complex subunit (Goley and Welch, 2006). Mice and humans have instead two functional Arp3, Arpc1 and Arpc5 loci, which give rise to paralogues with diverging sequences and slightly different functional properties (Abella et al., 2016). Biochemical reconstitution of the Arp2/3 complex revealed that Arpc2 and Arpc4 are crucial for the structural integrity of the complex, whereas the other subunits have only variable effects on actin nucleation (Gournier et al., 2001). Accordingly, knockdown of either Arpc2 or Arpc4 leads to the downregulation of the Arp2/3 complex in a variety of cell types (Goley and Welch, 2006). For this reason, Arpc2 and Arpc4 are referred to as core subunits, whereas Arp2, Arp3, Arpc1, Arpc3 and Arpc5 are regarded as peripheral subunits.

The Arp2/3 complex regulates many actin-dependent processes, such as cell migration, endocytosis, vesicle trafficking, organelle remodelling, and cell-cell and cell-matrix adhesion (Goley and Welch, 2006; Leyton-Puig et al., 2017; Rotty et al., 2013). In agreement with its central role in the cell, germline knockout of the Arp2/3 complex and that of its activators causes early embryonic lethality in most multicellular animal models (Stradal et al., 2004). Thus, whether and how the Arp2/3 complex affects developmental and homeostatic events remain two outstanding and largely unexplored issues. Although conditional knockout technologies can bypass embryonic lethality, targeting of the peripheral subunit Arpc3 has only produced an ‘Arpc3-less’ Arp2/3 complex and minor alterations in the central nervous system (Kim et al., 2013b, 2015; Zuchero et al., 2015) and small intestine (Zhou et al., 2015). Surprisingly, knockout of Arpc3 in the epidermis resulted in neonatal lethality and defective epidermal permeability barrier (EPB) despite a largely normal tissue architecture (Zhou et al., 2013).

Epidermis is a stratified self-renewing epithelium fulfilling essential protective functions (Simpson et al., 2011; Watt, 2014). The most abundant cells in the epidermis are keratinocytes, which self-organize in tightly juxtaposed layers. Strong cell-cell adhesion and firm binding between keratinocytes and the underlying basement membrane confer the necessary robustness on to the epidermis (Simpson et al., 2011; Watt, 2014). The keratinocytes bound to the basement membrane form the basal layer, which proliferates and can undergo either symmetric or asymmetric division, the former ensuring regenerative potential and the latter constant renewal of the epidermis (Sotiropoulou and Blanpain, 2012; Watt, 2014). The daughter cells positioned above the basal layer enter the spinous layer, where they exit the cell cycle and begin a complex differentiation process that reinforces intercellular junctions (Simpson et al., 2011). The upper granular layer consists of flattened keratinocytes characterized by a water-impermeable cornified envelope located beneath the plasma membrane. Finally, the outermost cornified layer contains dead squamous cells tightly linked together and devoid of internal organelles, which are constantly shed off (Segre, 2006). The keratinocyte differentiation programme controlling epidermal morphogenesis and homeostasis involves profound signalling-regulated rearrangements of both the composition and the architecture of the (actin) cytoskeleton (Simpson et al., 2011).

A master regulator of epidermal homeostasis is nuclear factor (erythroid-derived 2)-like 2 (NFE2L2 or Nrf2), a member of the cap-and-collar family of transcription factors (Sykiotis and Bohmann, 2010). Nrf2 binds to an antioxidant response element located in the regulatory regions of its target genes, most of which promote detoxification from reactive oxygen species and toxic compounds (Gorrini et al., 2013; Sykiotis and Bohmann, 2010). Not surprisingly, compelling evidence associates deregulation of Nrf2 expression and/or activity with a number of pathologies (Schäfer and Werner, 2015; Sykiotis and Bohmann, 2010). In homeostasis, the levels of Nrf2 are controlled by a Keap1-containing cullin 3-based E3 ligase complex that binds and ubiquitylates Nrf2 in the cytosol thereby marking it for proteasome-mediated degradation (Sykiotis and Bohmann, 2010). In stressed cells, Keap1 no longer ubiquitylates Nrf2, which can accumulate and translocate into the nucleus where it promotes gene transcription (Sykiotis and Bohmann, 2010).

Here, we report that ablation of the Arp2/3 complex in mouse epidermis causes a severe psoriasis-like disease hallmarked by Nrf2 hyperactivation and decreased F-actin levels. Unexpectedly, we find that the Arp2/3 complex modulates Nrf2-dependent gene transcription in keratinocytes by regulating Nrf2 localization through a mechanism that involves binding of Nrf2 to actin filaments. The finding that the Arp2/3 complex couples actin-based regulation of keratinocytes' shape and transcriptome, which can influence epidermal morphogenesis and homeostasis, opens a new view on its mode of action. Furthermore, the downregulation of Arpc4 in both human and mouse psoriatic epidermis suggests a possible role for the Arp2/3 complex in psoriasis pathogenesis.

We exploited a mutant (flox, f) allele of the gene encoding the Arp2/3-complex core subunit Arpc4 (Fig. 1A) to obtain homozygous Arpc4^f/f mice on a pure B6 genetic background (Fig. 1B). As homozygous germline deletion of Arpc4 was embryonic lethal prior to embryonic day (E) 7.5 (not shown), transgenic Arpc4^f/wt mice expressing Cre recombinase downstream of the keratin 14 (K14) promoter were crossed with the Arpc4^f/f mice to ablate the expression of Arpc4 in the hair follicles and basal cells of the epidermis (Fig. 1C). The resulting offspring was composed of all four genotypes in the expected Mendelian ratio, including epidermis-specific Arpc4 knockout (hereafter referred to as Arpc4 eKO) mice (Fig. 1C). The Arpc4 eKO pups could be easily recognized by the macroscopic appearance of the skin, which displayed uneven thickening accompanied by alopecic areas (Fig. 1D, Movie 1). The expression of Arpc4 and Arpc1A were dramatically decreased in these epidermal regions (Fig. S1A). Keratinocytes isolated from the abnormal areas of the Arpc4 eKO pups were not only devoid of Arpc4 but also lacked the entire Arp2/3 complex (Fig. S1D), thus confirming that the core subunits are essential for the stability of the Arp2/3 complex.

Initial skin lesions were detected microscopically in the Arpc4 eKO mice at day 1 after birth [postnatal day (P) 1]. Local thickening of the epidermis, mainly the cornified layer, and the presence of scattered ‘ghost cells’ (anucleated cells) were particularly obvious in the head region (Fig. 1E). These initial lesions developed progressively into macroscopic psoriasis-like plaques: body and extremities of the Arpc4 eKO mice had dry, unevenly thickened and poorly furred skin at P5 (Fig. 1D). Furthermore, microscopic analyses revealed abnormal thickening of the cornified layer (hyperkeratosis), presence of nuclei in the cornified layer (parakeratosis) accompanied by microabscesses, and hyperplasia of the epidermal squamous cells (acanthosis) at P7 (Fig. 1E). In addition, hair follicles in the psoriasis-like lesions were rare and often lacked the shaft and the sebaceous gland (Fig. 1E, Fig. S1C). Inflammatory infiltrations mainly consisting of lymphocytic cells were observed in dermis (dermatitis) (Fig. 1E, Fig. S1Ci,Ciii). The psoriasis-like lesions progressed further over time and acquired an ichthyosis-like appearance (Fig. 1E). Ultrastructural analyses confirmed these observations and further showed that the Arpc4 eKO mice have no evident signs of compromised epidermal integrity (Fig. 1F). Consistent with this, the Arpc4 eKO pups mounted a fully functional outside-in EPB (Fig. S1D) and the EPB regulators YAP/TAZ (Zhou et al., 2013) attained normal activity both in vivo and in isolated keratinocytes (Fig. S2A,B). The presence of inside-out EPB could not be reliably probed because epidermal lesions are microscopic in Arpc4 eKO newborns. Thus, we cannot exclude the possibility of a mild inside-out EPB defect. In any case, the Arpc4 eKO animals had to be euthanized by P21 owing to the severity of the skin lesions.

Molecular characterization of the Arpc4 eKO skin with well-established differentiation markers [cornified envelope regulator transglutaminase 1 (Matsuki et al., 1998), cornified envelope component involucrin (Koch et al., 2000), suprabasal keratin 1 (Roth et al., 2012) and basal and stem-cell keratin 15 (Bose et al., 2013)] showed that ablation of the Arp2/3 complex perturbs the differentiation programme of keratinocytes (Fig. S1E). In addition, skin lesions showed increased Ki67 positivity in both the basal and the suprabasal epidermal layers (not shown).

Although the psoriasis-like lesions were of variable size and usually more severe in the dorsal than in the ventral trunk, extremities and head (Movie 1), the disease showed full penetrance. The patchy nature of the phenotype was somewhat surprising as K14-Cre-mediated recombination gives a mosaic pattern at day E15 and occurs in most of the epidermis after birth (Huelsken et al., 2001). However, flox alleles that remain largely mosaic after birth have been previously reported (Kobielak et al., 2003; Raymond et al., 2005). As lesions were located in the areas of the epidermis with no Arp2/3 complex expression (Fig. S1A) and the Arpc4 eHet mice were normal (not shown), mosaic recombination of the Arpc4 flox allele likely explains the above findings. Overall, these observations indicate that ablation of the Arp2/3 complex in the mouse epidermis perturbs epidermal cell differentiation and causes a psoriasis-like disease.

To characterize this phenotype at the molecular level, we determined the global gene expression profile of the epidermis of wild-type and Arpc4 eKO siblings. Comparative analysis of RNA sequencing (RNA-seq) data identified 141 differentially expressed transcripts, most of which (131) were upregulated upon knockout of Arpc4 (Fig. 2A, Table S1). RT-qPCR analysis of independent epidermal samples supported the validity of this gene signature (Fig. 2B). Ingenuity pathway analysis (IPA) allowed us to extract system-level pathophysiological information from the list of Arpc4-responsive genes. The category ‘Dermatological Diseases and Conditions’ ranked first among the IPA-annotated diseases and disorders and most of the other top categories were related to events often observed in psoriasis, such as inflammatory response (Fig. 2C, Table S2). In agreement with the anatomopathological analysis of the skin of the Arpc4 eKO mice, psoriasis was the most enriched ‘Disease or Functions’ annotation belonging to the ‘Dermatological Diseases and Conditions’ (Fig. 2D, Table S2). In summary, the gene expression profiles of the epidermis of the wild-type and the Arpc4 eKO mice confirm the diagnosis based on anatomopathological criteria.

The RNA-seq data revealed an unexpected connection between the function of the Arp2/3 complex and that of Nrf2: among the genes showing a significantly altered expression in wild-type versus Arpc4 eKO mice, 65 were Nrf2-responsive genes (Table S1). Instead, only one target (Parp8) of G-actin-sensitive myocardin-related transcription factors (MRTFs) (Esnault et al., 2014; Posern and Treisman, 2006) was significantly affected. Thus, the G-actin pool controlling MRTF activity is apparently independent of the Arp2/3 complex.

Remarkably, mice with enhanced Nrf2 activity in keratinocytes present severe hyperkeratosis and acanthosis of the epidermis (Schäfer et al., 2012) resembling the initial abnormalities observed in the Arpc4 eKO mice. This and the observation that most differentially expressed genes are upregulated upon knockout of Arpc4 suggest that the Arp2/3 complex might restrain Nrf2 activity in keratinocytes. Consistent with this hypothesis, the epidermis of the Arpc4 eKO mice showed increased nuclear Nrf2 levels at P4 (Fig. 3A). The number of Nrf2-positive interfollicular keratinocytes declined with age in the normal epidermis, but Nrf2 expression remained ubiquitous in the psoriasis-like lesions (Fig. 3B). Of note, neither the mRNA [852±68 (n=3) and 686±52 (n=3) reads in wild-type and eKO epidermises, respectively; P=0.12, unpaired two-tailed t-test] nor the protein levels of Keap1 were downregulated in the Arpc4 eKO mice (Fig. 3B). This rules out that the Arp2/3 complex inhibits Nrf2 by controlling Keap1 expression. Most importantly, Nrf2-target genes identified by RNA-seq [Sprr2, keratin 6 (Krt6) and S100a9], as well as the genuine Nrf2-target gene Nqo1, attained increased protein levels in the affected epidermal areas (Fig. 3B). Thus, Nrf2 is hyperactive in the epidermal regions devoid of the Arp2/3 complex.

To prove that ablation of Arpc4 in basal keratinocytes increases Nrf2 activity, we isolated wild-type and Arpc4 KO keratinocytes. As expected, the knockout cells lacked both lamellipodia and ruffles and showed a global reduction in the F-actin levels, stress fibres included (Fig. S3A). Of note, reduced F-actin levels could also be observed in the psoriatic epidermis of the Arpc4 eKO mice (Fig. S3B). Phosphorylated Nrf2 levels were higher in the Arpc4 KO than the wild-type keratinocytes (Fig. 4A), suggestive of increased Nrf2 activity. Compared with wild-type keratinocytes, the Arpc4 KO keratinocytes and those overexpressing Nrf2 (see also Fig. 7B) displayed a very similar, mainly upregulated, gene expression pattern (Fig. 4B,C, Table S3). Furthermore, silencing of Nrf2 in the Arpc4 KO keratinocytes attenuated the expression of Nrf2-responsive genes identified by RNA-seq (Fig. 4D,E), but did not rescue the actin cytoskeleton (Fig. S3C). Collectively, these results show that the Arp2/3 complex regulates Nrf2 activity in keratinocytes.

To elucidate the underlying molecular mechanism, we expressed EGFP-tagged Arpc4 and empty EGFP in two independent populations of Arpc4 KO keratinocytes to generate rescued and knockout isogenic lines, respectively (Fig. 5A). Molecular characterization of these keratinocytes indicated that the introduction of EGFP-Arpc4 in the Arpc4 KO cells was sufficient to rescue the endogenous Arp2/3-complex subunits, whereas it did not affect either Nrf2 or Keap1 expression (Fig. 5A). Morphological characterization of the isogenic EGFP- and EGFP-Arpc4-expressing Arpc4 KO keratinocytes showed that the former are smaller, have reduced F-actin levels and lack lamellipodia, and often display bleb-like structures and fewer stress fibres (Fig. 5B, Fig. S4A). This latter observation is in line with the filaments assembled by the Arp2/3 complex being incorporated into stress fibres (Hotulainen and Lappalainen, 2006). Fractionation experiments confirmed biochemically that F-actin is decreased in the absence of the Arp2/3 complex (Fig. 5C,D, Fig. S4B). Consistent with the well-established localization of the Arp2/3 complex (Rotty et al., 2013), EGFP-Arpc4 decorated actin-rich lamellipodia and dotty cytoplasmic structures likely corresponding to vesicles and vesicular compartments (Fig. 5B, Fig. S4A).

The knockout keratinocytes showed more than a two-fold increase in the nuclear versus cytosolic Nrf2 ratio compared with that of the rescued ones (Fig. 6A,B), which correlated with a higher expression of a battery of genuine Nrf2-responsive genes, including the direct targets Sprr2, Sprr1 and Epgn (Papp et al., 2012; Schäfer et al., 2012, 2014) (Fig. 6C). Therefore, regulation of Nrf2 localization and activity by the Arp2/3 complex is cell-autonomous in keratinocytes.

We failed to show that the Arp2/3 complex interacts with Nrf2 (not shown), but found that phosphorylation of Nrf2 at Ser40 increases in keratinocytes treated with the Arp2/3 complex inhibitor CK-666 (Nolen et al., 2009) (Fig. 7A). Notably, phosphorylation of Ser40 promotes nuclear accumulation and activation of Nrf2 (Huang et al., 2002). Consequently, we hypothesized that the Arp2/3 complex could control Nrf2 through the polymerization of actin filaments and explored this possibility further. As five different commercial anti-Nrf2 antibodies cross-react with additional proteins in keratinocytes, we carried out single-cell analyses exploiting a keratinocyte line expressing EGFP-tagged Nrf2 at low levels (Fig. 7B, Fig. S5A). Upon CK-666 treatment, the number of cells having a higher EGFP-Nrf2 signal in the nucleus than in the cytosol doubled (Fig. 7C,D). The observed cell-to-cell variability is most likely inherent to the polyclonal nature of keratinocytes. Live-cell imaging confirmed that pharmacological inhibition of the Arp2/3 complex leads to nuclear accumulation of EGFP-Nrf2 (Fig. S5B, Movie 2). Similar to CK-666, actin depolymerization induced by latrunculin A caused nuclear accumulation of EGFP-Nrf2 in keratinocytes (Fig. 7E,F). By contrast, inhibition of formin-induced actin polymerization by SMIFH2 (Isogai et al., 2015b) had no significant effects (Fig. 7E,F). The sum of these results points towards a specific role for the actin filaments nucleated by the Arp2/3 complex in the control of Nrf2 localization.

In agreement with the above results, CK-666 treatment also stimulated the expression of Nrf2-target genes in both wild-type and EGFP-Nrf2-expressing keratinocytes (Fig. S5C). This functionally links Arp2/3-complex-mediated actin polymerization and Nrf2 activity and also support the physiological relevance of this new Nrf2-regulatory mechanism.

Surprisingly, EGFP-Nrf2 partially colocalized with F-actin in lamellipodia and along stress fibres (Fig. 7G). This localization pattern was clearly perturbed by CK-666 and latrunculin A, but not SMIFH2 (Fig. S6). Thus, the presence of Nrf2 on F-actin structures seems to be largely dependent on the actin nucleation activity of the Arp2/3 complex. Co-sedimentation assays showed that purified full-length GST-Nrf2 (Fig. 8A) binds directly to F-actin at concentrations that mimic the low steady-state levels of Nrf2 in the cell (Kim et al., 2011) (Fig. 8B). Congruently, the equilibrium dissociation constant of the Nrf2-F-actin interaction lies in the high nanomolar range (Fig. 8C). In vivo, however, only the actin filaments nucleated by the Arp2/3 complex are permissive for the binding of Nrf2. It is noteworthy that the initial branched geometry of these filaments excludes the actin-binding proteins that decorate linear actin networks (Michelot and Drubin, 2011) and may outcompete Nrf2.

In summary, binding of Nrf2 to F-actin indicates that Nrf2 regulation by the Arp2/3 complex involves actin nucleation.

The phenotype of the Arpc4 eKO mice suggests that the Arp2/3 complex might be involved in the pathogenesis of psoriasis. However, population-based studies in humans have not yet revealed an association between ARPC4 and psoriasis. Hence, we first analysed ARPC4 expression and distribution in the skin of healthy donors by immunohistochemistry. Human epidermis showed a rather uniform positivity for ARPC4 both in the basal and supra-basal layers (Fig. 9A, Fig. S7). This expression pattern strongly resembles that of mouse Arpc4 (Fig. S1A). Notably, ARPC4 was particularly abundant within the nucleus in roughly half of the analysed healthy skin samples, as well as in asymptomatic psoriatic skin (Fig. S7). This suggests that nuclear ARPC4 is not predisposing to or protecting from psoriasis. Next, healthy skin was compared with both asymptomatic and lesional skin of psoriatic patients. We found that ARPC4 expression and distribution are similar in the epidermis of asymptomatic skin of psoriatic patients and healthy donors (Fig. 9A,B, Fig. S7). Nevertheless, reduced basal ARPC4 levels were observed in two asymptomatic psoriatic skin samples (Fig. S7), thus raising the possibility that downregulation of the ARP2/3 complex is pathogenic in human epidermis. More importantly, ARPC4 immunoreactivity was significantly decreased in the epidermis of lesional psoriatic skin, especially in the basal proliferative layer (Fig. 9A,B, Fig. S7).

To corroborate the link between reduced Arpc4 expression and psoriasis, we used the imiquimod (IMQ)-induced psoriasiform mouse model, which closely resembles the human psoriatic lesions in terms of phenotypic and histological characteristics, including epidermal hyperplasia (acanthosis) and the presence of inflammatory infiltrates in the dermis (van der Fits et al., 2009). As expected, B6 mice treated with IMQ showed (1) a remarkably increased epidermal thickness (198.45±48.03 µm and 40.06±13.04 µm in IMQ-treated and control mice, respectively; P<0.001), (2) thickening of the cornified layer (71.85±34.56 µm versus 29.22±17.09 µm; P<0.001), (3) perturbed Krt10 expression (indicative of impaired terminal differentiation; Fig. 9C), (4) enhanced proliferation (hallmarked by Ki67-positive keratinocytes; Fig. 9C), and (5) widespread inflammatory infiltrate (not shown), compared with control mice. In addition to these psoriasis-like manifestations, Arpc4 expression was diminished in the IMQ-treated epidermis, especially in the basal layer keratinocytes (Fig. 9C).

Therefore, the association between low Arpc4 expression and psoriasis appears to be conserved in mice and humans.

Here, we report that the Arpc4 epidermal knockout mice are Arp2/3-complex null and develop severe psoriasis-like lesions hallmarked by ubiquitous Nrf2 expression and upregulation of Nrf2-target genes. Furthermore, we discovered that the Arp2/3 complex inhibits Nrf2 activity through a cell-autonomous mechanism involving interaction between Nrf2 and F-actin. The downregulation of Arpc4 in human and mouse psoriatic skin suggests that the function of the Arp2/3 complex in the epidermis is conserved. Most importantly, our findings ascribe an unforeseen gene transcription-regulatory role to the actin network generated by the Arp2/3 complex in health and disease.

We propose a sequestration model to explain how the Arp2/3 complex harnesses the activity of Nrf2 in keratinocytes: actin nucleation by the Arp2/3 complex results in the formation of branched actin filaments that can be subsequently remodelled and incorporated into linear actin structures. The filaments generated by the Arp2/3 complex sequester part of the cytosolic Nrf2 pool and restrain the activity of Nrf2. Genetic ablation and pharmacological inhibition of the Arp2/3 complex relieve Nrf2, which can translocate into the nucleus where it increases the transcription of genes involved the psoriatic response. Notably, we provide robust experimental evidence in support of the tenets of this model.

First, the marked reduction in F-actin observed in the Arpc4 eKO mice (Fig. S3B) shows that the Arp2/3 complex is active in the ‘unchallenged’ epidermis and can effectively participate in Nrf2 regulation. Thus, this new actin-based Nrf2-regulatory mechanism may functionally connect the reduction in F-actin levels with the rise in Nrf2 activity during keratinocyte differentiation (Schäfer and Werner, 2015; Vaezi et al., 2002). It also suggests that the knockout of Arpc3 does not phenocopy that of Arpc4 because the ‘Arpc3-less’ Arp2/3 complex ensures normal epidermal F-actin levels (Zhou et al., 2013). Instead, the EPB defect and the perinatal lethality observed in the Arpc3 eKO mice likely stem from spurious dominant effects exerted by mislocalized ‘Arpc3-less’ Arp2/3 complex on tight junctions and YAP/TAZ (Zhou et al., 2013).

Second, genetics and chemical biology show that the Arp2/3 complex assembles actin filaments endowed with unique Nrf2-regulatory features. The knockout of Arpc4 revealed that the Arp2/3 complex regulates both the branched and the linear F-actin networks (Fig. 5B, Fig. S3A) to which Nrf2 can associate (Fig. 7G, Fig. 8, Fig. S6). Nevertheless, mice lacking either Fmn1 or mDia1 (Diaph1), formins that polymerize linear actin filaments in epithelial cells (Isogai et al., 2015c; Kobielak et al., 2004), do not have skin defects (Eisenmann et al., 2007; Wynshaw-Boris et al., 1997). Furthermore, the pan-formin inhibitor SMIFH2 failed to affect the localization of EGFP-Nrf2 (Fig. 7E,F, Fig. S6), the phosphorylation of Nrf2 at Ser40 (Fig. 7A) and the expression of Nrf2-target genes in keratinocytes (not shown). Hence, remodelling of Nrf2-bound branched actin filaments likely explains the localization of Nrf2 on linear F-actin structures. Most importantly, these observations show that the initial branched geometry of the actin filaments nucleated by the Arp2/3 complex is a key Nrf2-regulatory feature. Remarkably, we also discovered that the Arp2/3 complex restrains Nrf2 phosphorylation on Ser40 (Fig. 4A, Fig. 7A), which favours nuclear accumulation and activation of Nrf2 (Huang et al., 2002). Thus, Nrf2 regulation by the Arp2/3 complex involves both binding to F-actin and a post-translational modification. In this regard, it has been proposed that Keap1 forms a complex with F-actin and Nrf2 thereby promoting ubiquitylation and proteasome-mediated degradation of Nrf2 in the cytosol (Kang et al., 2004). However, this is highly unlikely because the same region of Keap1 interacts with actin and Nrf2 (Kang et al., 2004) and cellular actin is abundant enough to fully outcompete Nrf2. Not surprisingly, subsequent studies have excluded the possibility that the actin cytoskeleton allows Keap1 to control Nrf2 (Velichkova and Hasson, 2005). In any case, we found that Keap1 is diffused and uniformly distributed in keratinocytes expressing EGFP-Nrf2 (Fig. S8). This latter observation is consistent with the cell type-specific localization of Keap1 at actin-positive structures (Velichkova et al., 2002). Therefore, the Arp2/3 complex and Keap1 regulate Nrf2 through entirely distinct mechanisms. As neither Nrf2 nor Keap1 levels vary upon knockout of Arpc4 in keratinocytes (Fig. 5), it also appears that the Arp2/3 complex and Keap1 act independently on Nrf2.

Third, the knockout of Arpc4 resulted in decreased F-actin levels both in vivo and in isolated keratinocytes (Fig. S3B, Fig. 5D). Although this involved both branched and linear F-actin structures (Fig. 5B, Fig. S3A), the knockout of Arpc4 had negligible effects on the expression of MRTF-target genes (Fig. 2A, Fig. 4B). Intriguingly, we also found that the expression of Nrf2-target genes is insensitive to formin inhibition (not shown). Therefore, keratinocytes have distinct F-actin and G-actin pools regulated by either the Arp2/3 complex or formins, as well as independent mechanisms to sense and respond to the changes in either F-actin network.

Fourth, knockout of Arpc4 in basal keratinocytes causes a psoriasis-like disease hallmarked by Nrf2 hyperactivation (Figs 1 and 2, Fig. S1) and downregulation of Arpc4 occurs in both human psoriatic skin and the IMQ-induced psoriasis mouse model (Fig. 9). Moreover, elevated expression of Nrf2 has been recently observed in psoriasis and suggested to cause pathogenic hyperproliferation of keratinocytes (Yang et al., 2017). Hence, the phenotype of the Arpc4 eKO mice could be partly due to Nrf2 hyperactivation enhancing keratinocyte proliferation in vivo. The similarities between the early phenotype of the Arpc4 eKO mice and that of mice expressing constitutively active Nrf2 in basal keratinocytes support this view (Schäfer et al., 2012). As only the knockout of Arpc4 causes a psoriasis-like disease in mice, Nrf2-independent regulation of gene transcription and actin dynamics by the Arp2/3 complex also has a key pathophysiological role. The use of the Nrf2 activator dimethyl fumarate (DMF) to treat moderate-to-severe plaque psoriasis does not contradict the pathogenic role of Nrf2 in psoriasis. Actually, the anti-inflammatory and immune-modulatory properties of DMF might not require Nrf2 (Schulze-Topphoff et al., 2016) and could counteract cell proliferation driven by the further activation of Nrf2 in psoriatic skin. Alternatively, the beneficial anti-inflammatory and cytoprotective functions of Nrf2 may prevail over its pathogenic proliferative role in the epidermis only above a certain expression level and/or in full-blown psoriatic lesions.

In any case, downregulation of the Arp2/3 complex in keratinocytes qualifies it as one of the elusive epidermal cell-intrinsic psoriasis triggers. In line with this idea, the psoriasis-like disease in the Arpc4 eKO mice is not due to a generic pro-inflammatory response in keratinocytes. In fact, Arpc4 ablation did not increase the levels of active, phosphorylated NF-κB in isolated keratinocytes (Fig. 4A) nor did it produce upregulation and enrichment of NF-κB-target genes in the epidermis (Fig. 2A). Similarly, we also ruled out early disruption of the basement membrane in the Arpc4 eKO pups as the primary cause of the psoriasis-like disease (Fig. S9). Given that the Arp2/3 complex is essential for embryogenesis, we propose that epigenetic events and/or somatic mutations determine Arp2/3 complex levels in the epidermis. Arpc4 downregulation by IMQ in mouse epidermis (Fig. 9C) supports this view.

In summary, we have discovered that the Arp2/3 complex affects epidermal morphogenesis and homeostasis by coupling actin-based regulation of keratinocytes' shape and transcriptome. These unforeseen findings pave the way for a deeper understanding of the cellular and organismal functions of the Arp2/3 complex in health and disease.

Arpc4^{tm1a(EUCOMM)Wtsi)} mice (knockout-first mice with conditional potential, promoterless cassette) were obtained from the Wellcome Trust Sanger Institute. These mice were backcrossed with Flp recombinase transgenic C57Bl/6NRj mice to remove the cassette flanked by Frt sites thereby creating conditional Arpc4^wt/f mice. Arpc4^wt/f mice were backcrossed on a C57Bl/6 background at least ten times, before intercrossing Arpc4^wt/f mice to obtain homozygous Arpc4^f/f animals. Arpc4^f/f mice were phenotypically identical to wild-type B6 mice and bred well. Heterozygote K14-Cre(neo) transgenic B6 mice (Margadant et al., 2009) were backcrossed at least ten times with C57Bl/6 mice before breeding with homozygous Arpc4^f/f mice. Analysed mouse cohorts consisted of both sexes. Genotyping of the Arpc4 alleles was performed with the wild-type primers Arpc4_39529_F (AAGCCTTGCCCGAGATAATG) and Arpc4_39529_R (AAGCAAAGCCAGTCCCTCAC), positions of which are depicted in Fig. 1A.

Eight-week-old female C57BL/6 mice (Harlan Laboratories, San Pietro al Natisone, Italy) were divided randomly into control (n=6) and IMQ-treated (n=12) groups. Whereas control mice received a control cream (Vaseline; Walgreens Pharmacy), IMQ-treated mice received on the shaved back skin a daily topical dose of 50 mg of 5% IMQ cream (Aldara, Meda AB, Solna, Sweden). On day 5, skin biopsies of the treated area were collected with a 8-mm biopsy puncher, as previously described (Palombo et al., 2016). Epidermal and scale thickness, and cell infiltrate number were analysed as parameters of skin acanthosis and inflammation (not shown). The significance of differences between experimental groups (mice treated with control cream versus mice treated with IMQ) was calculated by unpaired Student's t-test (with P<0.05 considered significant).

Keratinocytes were isolated from back skin epidermis of Arpc4^f/f and Arpc4^f/f::K14-Cre(neo) littermates up to P4. In brief, skins were removed, put on sterile 3MM paper and trypsinized (EDTA-free 0.25% trypsin) for 16 h at 4°C to separate the epidermis from the dermis. Epidermises were minced, and cells were detached by stirring on ice for 1 h. Cell suspensions were filtered and seeded on dishes coated with 10 μg cm⁻² collagen IV (Becton Dickinson) and cultured in EpiLife medium with 40 μM CaCl₂ and EpiLife-defined growth supplement (EDGS).

Primary keratinocytes grown in EpiLife medium containing 40 μM CaCl₂ and EDGS were immortalized with SV40 large T antigen as previously described (Mertens et al., 2005). Immortalized Arpc4^f/f keratinocyte cultures were subsequently treated with Adeno-cre virus produced in HEK293A cells to knock out Arpc4. Rescue of the Arpc4^f/f::K14-Cre(neo) (line #1) and Arpc4^−/− (line #2) populations devoid of Arpc4 was performed by retroviral transduction with pMX encoding either EGFP or human ARPC4 tagged with EGFP at the C terminus. After two or three rounds of transduction, EGFP-positive keratinocytes were sorted. EGFP-Nrf2-expressing keratinocytes were generated from an immortalized Arpc4^f/f keratinocyte line by retroviral transduction with pMX encoding human NRF2 tagged with EGFP at the N terminus. Low passage (<18) keratinocytes were used for experiments.

Nrf2 knockdown (KD) keratinocytes were obtained by transducing Arpc4 KO cells with lentiviruses encoding shRNA TRCN0000012129 (#1) and shRNA TRCN0000054658 (#2) targeting mouse NFE2L2. Control KD Arpc4 KO keratinocytes were obtained by transduction with empty pLKO-derived lentiviral particles prepared as previously described (Leyton-Puig et al., 2017). Low passage (<18) keratinocytes were used for experiments.

Keratinocytes were divided using TrypLE Express enzyme (Invitrogen) and trypsin inhibitor (1 mg ml⁻¹; Sigma) and plated in CELLnTEC Keratinocyte medium on collagen I-coated dishes. After 24 h, EpiLife medium containing 40 μM CaCl₂ and EDGS supplement was added. Only mycoplasma-free cells were used in all experiments.

Back skins (or decapitated bodies, P1) were fixed in EAF (4% formaldehyde, 5% glacial acetic acid) and embedded in paraffin. Sections (5 µm) were deparaffinized and rehydrated, and subjected to antigen retrieval and inactivation of endogenous peroxidase prior to the addition of primary antibodies. A two-way staining protocol [biotin/horseradish peroxidase (HRP)] was used employing 3,3′-diaminobenzidine (DAB) as a substrate (details are provided in Table S4). All sections were counterstained with Haematoxylin, washed and mounted with Entellan. Investigators were blind to group allocations during both experiments and subsequent analyses.

Tissues were fixed in Karnovsky's fixative (2% paraformaldehyde, 2.5% glutaraldehyde in 0.1 M cacodylate buffer at pH 7.2). Post-fixation was performed with 1% osmium tetroxide in 0.1 M cacodylate buffer at pH 7.2. After washing, tissue blocks were stained en bloc with Ultrastain 1 (Leica), followed by ethanol dehydration series. Finally, tissue was embedded in a mixture of DDSA/NMA/Embed-812 (EMS), sectioned and stained with Ultrastain 2 (Leica) and analysed with a Tecnai 12G2 electron microscope (FEI). Investigators were blind to group allocations during the experiments and subsequent analyses.

Newborn mice were euthanized by intraperitoneal administration of a lethal dose of sodium pentobarbital (Pharmacy of the Faculty of Veterinarian Medicine, Utrecht University; 2 mg/g body weight). The tail of each pup was clipped and stored for PCR-based genotypic analysis. Carcasses were rinsed in ice-cold PBS and dehydrated by immersion in 25%, 50%, 75% and 100% methanol for 2 min each at 4°C. Next, they were progressively rehydrated by immersion in 100%, 75%, 50% and 25% methanol for 2 min each at 4°C and finally in PBS. After immersion in 0.1% Toluidine Blue for 2 min on ice, carcasses were briefly washed in PBS and immediately photographed. PCR-based genotypic analyses were conducted only after completion of the assay. Investigators were blind to group allocations during the experiments and subsequent analyses.

Cell lysates, protein quantification and SDS-PAGE were performed as previously described (Beli et al., 2008; Galovic et al., 2011; Isogai et al., 2015a, 2016). Fractionations were run on 4-15% gradient gels (TGX, Bio-Rad) using the Laemmli buffer system and blotted on nitrocellulose (pore size 0.2 µm; Pall). Blots were routinely blocked with 5% bovine serum albumin (BSA) for 1 h and then, depending on the manufacturer, incubated with primary antibodies for 1 h or overnight. All secondary HPR-conjugated antibodies were from Bio-Rad. Primary antibodies are listed in Table S4.

Keratinocytes were plated on collagen I (2 µg ml⁻¹)-coated coverslips and treated 48 h later. Cells were fixed as previously described (Isogai et al., 2015c) and images acquired sequentially on a CLSM Leica TCS SP5 using identical settings throughout each experiment. For confocal time-lapse video microscopy, cells were plated on collagen I-coated glass-bottom Cellview dishes (Greiner) and imaged on a CLSM Leica TCS SP5 microscope equipped with a humidified climate chamber with 5% CO₂ at 37°C (63×, 1.45 N.A, Argon laser 5-7%, 1.4 A.U., line scan mode, accumulation 4, gain 40%, 2 frames min⁻¹). Movies were assembled from (512×512 pixels) time series and processed using ImageJ.

Automated analysis compressed confocal (1024×1024 pixel) z-stacks of non-permeabilized EGFP-Nrf2-expressing keratinocytes counterstained with DAPI was carried out using Cell profiler (Broad Institute) (Jones et al., 2008) and a custom-modified ‘Human cytoplasm-nucleus translocation assay’ pipeline.

Semi-quantitative analysis of ARPC4 immunoreactivity in human skin samples was performed as follows: images were processed using ImageJ plugin ‘IHC ToolBox’ (http://imagej.nih.gov/ij/plugins/ihc-toolbox/index.html) using either pre-set or custom-made H-DAB models to extract the unmixed DAB (brown) signal. The resulting DAB images were inspected manually and only those ensuring a satisfactory removal of the counterstaining were analysed. For each image, two full-thickness epidermal regions of interest were randomly selected and mean intensity measured with the ImageJ ‘Measure’ function. Values were averaged and then converted from the original 0-255 (black-white) scale into a 0-1 (black-white) scale. Each value (x) was then transformed using the 1-x formula for plotting according to a more intuitive 0-1 (white-black) scale.

To obtain cytosol-enriched and nuclear-enriched fractions, keratinocytes (approximately 16×10⁶ cells, 80% confluent cultures) were washed twice with ice-cold PBS supplemented with calcium (0.884 mM) and magnesium (0.492 mM) (Lonza, BE17-513F), scraped with a rubber policeman and spun at 900 g for 5 min at 4°C. The pellet was taken up in 150 µl hypertonic buffer [100 mM PIPES pH 6.8, 1 mM EGTA, 1 mM MgCl₂ supplemented with phosphatase and protease inhibitor cocktail (Roche)], re-suspended gently and left at room temperature for 5 min. After addition of 0.5% Triton X-100 and gentle re-suspension, samples were spun and centrifuged at 900 g for 5 min at 4°C. The resulting supernatant is referred to as cytosol. The pellet was washed with 10 bed volumes of hypertonic buffer without Triton X-100 and then re-suspended in RIPA buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS) supplemented with phosphatase and protease inhibitor cocktail (Roche), sonicated three times for 10 s in a Branson sonicator bath (75% intensity) and centrifuged for 30 min at 15,000 rpm (21,130 g) at 4°C. The final supernatant is referred to as nucleus.

To obtain G-actin and F-actin fractions, the G-actin/F-actin in vivo assay kit (Cytoskeleton) was used. Briefly, keratinocytes (approximately 3×10⁶ cells, 80% confluent cultures) were lysed in 400 µl of the LAS2 as per the manufacturer's instructions. After removing cell debris and unbroken cells, protein concentration was measured and lysates diluted in LAS2 buffer to obtain a final concentration of 1 mg ml⁻¹. Next, 150 µl of each lysate were fractionated.

Muscle actin (Cytoskeleton) was recycled as previously described (Hertzog and Carlier, 2005) and then used for the F-actin co-sedimentation assays using an established protocol (Hertzog and Carlier, 2005). Full-length GST-Nrf2 was expressed in Escherichia coli and purified using previously described procedures (Beli et al., 2008; Galovic et al., 2011; Innocenti et al., 2004, 2005) with the exception that cells were grown at 30°C prior to induction.

Tissues were homogenized in TRIzol reagent (Ambion Life Technologies) using a polytron (DI 18 Disperser, IKA) as per the manufacturer's protocol. Typically, 1 and 0.5 ml of TRIzol reagent were used per 50-100 mg of epidermis and 1×10⁶ cells, respectively.

Total RNA was extracted using TRIzol reagent as per the manufacturer's protocol. Briefly, 0.2 volumes of chloroform (Chloroform stab./Amylene, Biosolve) were added to the TRIzol homogenate and tube(s) (Falcon, 15 ml) were shaken vigorously. The tubes were incubated for 2-3 min at room temperature and centrifuged (Hettich, rotanta 46 RS) at 4500 g or 1 h at 4°C. Approximately 70% of the upper aqueous phase was transferred to a clean 15 ml tube and 0.5 volumes of isopropanol (Sigma-Aldrich) were added. The tube(s) were then incubated overnight at −20°C and centrifuged at 4500 g for 30 min at 4°C. The supernatant was removed and the pellet was washed twice with 80% ethanol (Sigma-Aldrich). The total RNA pellet was air-dried for 8 min and dissolved in an appropriate volume of nuclease-free water (Ambion Life Technologies) and quantified using a Nanodrop UV-VIS Spectrophotometer. The total RNA was further purified using MinElute Cleanup Kit (Qiagen) according to the manufacturer's instructions. Quality and quantity of the total RNA was assessed by the 2100 Bioanalyzer using a Nano chip (Agilent). Total RNA samples having RNA integrity number (RIN)>8 were subjected to library generation.

Strand-specific libraries were generated using the TruSeq Stranded mRNA sample preparation kit (Illumina, RS-122-2101/2) according to the manufacturer's instructions (Illumina, 15031047 Rev. E). Briefly, polyadenylated RNA from 1000 ng intact total RNA was purified using oligo-dT beads. Following purification, RNA was fragmented, random primed and reverse transcribed using SuperScript II Reverse Transcriptase (Invitrogen, 18064-014) with the addition of actinomycin D. Second strand synthesis was performed using Polymerase I and RNaseH with replacement of dTTP for dUTP. The generated cDNA fragments were 3′-end adenylated and ligated to Illumina Paired-end sequencing adapters and subsequently amplified by 12 cycles of PCR. The libraries were analysed on a 2100 Bioanalyzer using a 7500 chip (Agilent), diluted and pooled equimolar into a 10-plex, 10 nM sequencing pool.

The epidermal and keratinocyte libraries were sequenced with 51-base and 65-base single reads, respectively, on a HiSeq2500 using V4 chemistry (Illumina). Reads were aligned with TopHat (Kim et al., 2013a; Trapnell et al., 2009) (version 2.0.12), which allows for exon-exon-junctions against the mouse build 38. Read counts were generated using a custom script based on the union mode of the HTSseq-count (Anders et al., 2015). Only reads that mapped uniquely to the transcriptome were used for counting. All samples were merged into one dataset and genes that have zero expression across all samples were removed from the dataset. Identification of differentially expressed genes was performed using the R package Limma (Ritchie et al., 2015). In Fig. 2, wild-type samples were compared with knockout samples and a given gene was considered differentially expressed only when P<0.05 in all three comparisons. In order to create a robust list, a log₂ fold change of 1.5 was set as a minimum. Nrf2-target genes were retrieved from the NRF2ome website publicly available online and, after removal of duplicated entries, complemented by mining Nrf2-target genes from published literature. The MRTF signature (921 genes) was previously published (Esnault et al., 2014). Investigators were blind to group allocations during the experiments. Reads were normalized to 10 million reads per sample to plot expression levels in the heat maps. Hierarchical clustering based on complete linkage and using the Euclidean distance was performed on both genes (rows) and samples (columns) of the data. The order of the samples and genes is based on the results of the clustering analyses.

Total RNA was isolated from keratinocytes and skin epidermis using GeneJET RNA purification kit (Thermo Scientific) and TRIzol, respectively. For keratinocytes, complementary DNA synthesis was performed using 1-2 µg of mRNA with High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) and oligo-dT (25 ng ml⁻¹) according to the manufacturer's instructions. For the epidermis, the mix was also supplemented with random primers (1:100 from the supplied 10× RT stock solution). Real-time qPCR reactions were set up using 1-2 ng of cDNA as a template and gene-specific primers (600 nM) in a StepOnePlusTM Real-Time PCR system (Applied Biosystems). All reactions produced single amplicons (100-200 bp), which allowed us to equate one threshold cycle difference using Hprt as reference housekeeping gene (Isogai et al., 2015b). Biological duplicates were assayed in triplicate and data were normalized with respect to the central value of the control. RT-qPCR primers are listed in Table S5.

Skin biopsies were taken from lesional skin of adult patients affected by chronic plaque psoriasis (n=10). Biopsies were also obtained from normal skin of healthy subjects undergoing plastic surgery (n=5). For six psoriatic patients, a skin biopsy from asymptomatic normal-appearing skin distant from the lesions was also taken.

For immunohistochemistry, skin samples were fixed in 10% formalin and embedded in paraffin. Five-µm-thick sections were dewaxed and rehydrated. After quenching endogenous peroxidase, achieving antigen retrieval, and blocking non-specific binding sites, sections were incubated overnight at 4°C with goat anti-Arpc4 antibodies. Biotinylated anti-goat antibodies and staining kits were from Vector Laboratories. Signal was developed using DAB (3,3′-diaminobenzidine) (DAKO) followed by counterstaining with Haematoxylin (Vector Laboratories). Controls omitting the primary antibodies gave virtually no staining (not shown).

GraphPad Prism (version 6.oh) was used to carry out all statistical analyses and to plot results as indicated in the figure legends.

Animal studies were performed with approval of the local Animals Ethics Committee (DEC) and according to Dutch legislation. IMQ treatments were performed in compliance with a protocol approved by the Italian ISS (Istituto Superiore di Sanità) (23/12/2013, IDI protocol N. SA-IDI-CA-1). Human studies were performed according to the Declaration of Helsinki with regard to scientific use, and approved by the ethical committee of the Fondazione ‘Luigi Maria Monti’ – Istituto Dermopatico dell'Immacolata (IDI)-IRCCS (Rome, Italy) (23/12/2015, Protocol N. 103/CE/2015). Patients were enrolled in the study after written informed consent.

We thank J. Neefjes (Leiden University) for the EGFP-Nrf2 plasmid, A. Berns (NKI) and A. Sonnenberg (NKI) for critical reading of the manuscript and Ron Kerkhoven for supervising the work of the NKI Genomics Core Facility.