Research ArticleSKIN DISORDERS

Hair eruption initiates and commensal skin microbiota aggravate adverse events of anti-EGFR therapy

See allHide authors and affiliations

Science Translational Medicine  11 Dec 2019:
Vol. 11, Issue 522, eaax2693
DOI: 10.1126/scitranslmed.aax2693

The skinny on a cancer drug side effect

Therapeutics targeting the epidermal growth factor receptor (EGFR) are used for many cancer types, but they have notable side effects, including a potentially severe and disfiguring skin rash. Klufa et al. discovered that the inhibition of EGFR interferes with the ability of skin stem cells to reestablish a secure barrier after it is broken in the process of hair eruption. This loss of skin barrier integrity permits microorganisms to invade the skin and trigger inflammation similar to that which occurs in atopic dermatitis. The authors also identified a pathway that could be targeted to protect the skin without restoring EGFR activity directly.


Epidermal growth factor receptor (EGFR)–targeted anticancer therapy induces stigmatizing skin toxicities affecting patients’ quality of life and therapy adherence. The lack of mechanistic details underlying these adverse events hampers their management. We found that EGFR/ERK signaling is required in LRIG1-positive stem cells during de novo hair eruption to secure barrier integrity and prevent the invasion of commensal microbiota and inflammatory skin disease. EGFR-deficient epidermis is permissive for microbiota outgrowth and displays an atopic-like TH2-dominated signature. The opening of the follicular ostia during hair eruption allows invasion of commensal microbiota into the hair follicle, initiating an additional TH1 and TH17 response culminating in chronic folliculitis. Restoration of epidermal ERK signaling via prophylactic FGF7 treatment or transgenic SOS expression rescues the barrier defect in the absence of EGFR, highlighting a therapeutic anchor point. These data reveal that commensal skin microbiota provoke atopic-like inflammatory skin diseases by invading into the follicular opening of erupting hair.


Epidermal growth factor receptor (EGFR) overexpression and activation mutations are commonly found in solid tumors, rendering it an attractive molecular target for anticancer therapy (1). Consequently, EGFR inhibitors (EGFR-Is) have been successfully developed as one of the prototypic targeted antineoplastic therapies and are frequently implemented in current anticancer treatment regimens for non–small cell lung, colorectal, and head and neck cancers (2).

The cutaneous adverse events occurring in EGFR-I–treated patients with cancer reflect the central function of EGFR in the skin. A stigmatizing papulopustular skin rash occurs on the face and upper trunk in 60 to 90% of patients after the first week of therapy, and its severity correlates with treatment response (3). Staphylococcus aureus superinfections, pruritus, dry skin, alopecia, and hair alterations manifest themselves after prolonged treatment and may appear at various body sites (3). Although not life-threatening, these adverse effects decrease patients’ therapy adherence, resulting in dose reduction or even cessation of otherwise successful EGFR-I therapy, thereby drastically narrowing its therapeutic efficacy (4).

The lack of mechanistic details about etiology and pathogenesis of the skin inflammation limits the therapeutic options for supportive care. Currently, broad-spectrum antibiotics, corticosteroids, moisturizers, sun protection, and prevention of mechanical stress can alleviate some of the symptoms (4). However, none of these therapies completely prevent skin toxicities.

Several genetic mouse models demonstrated the importance of EGFR signaling during homeostasis and tumorigenesis in different organs like brain, bone, heart, skin, and several epithelial tissues (510). Mice lacking EGFR in the epidermis display defects in hair follicle development, cycling, and hair morphology followed by severe skin inflammation, similar to mice lacking the metalloproteinase ADAM17, a sheddase for tumor necrosis factor–α (TNFα), Notch, and EGFR ligands (1114).

EGFR controls skin homeostasis by a complex pleiotropic sequence of events. The most dominant phenotypic hallmarks occurring in mice lacking EGFR in the epidermis reflect the situation observed in EGFR-I–treated patients and in patients carrying loss-of-function mutations of EGFR and ADAM17 (15, 16). In the absence of EGFR signaling, barrier defects, along with up-regulation of chemokines and cytokines, are detected in the skin together with massive infiltration of immune cells (11, 13). In adult mice lacking epidermal ADAM17, an S. aureus–driven dysbiosis is responsible for the development of late-stage eczema (14, 17).

All these studies have characterized the chronic and late stages of the skin inflammation but failed to identify why the dysbiosis develops in the absence of epidermal ADAM17 and EGFR signaling. In this study, we aimed at identifying the cause, the initial trigger, and the underlying molecular mechanisms causing the complex skin inflammation induced by lack of EGFR signaling, with the aim of developing a feasible therapy for EGFR-I–treated patients with cancer.


Skin barrier defect coincides with the onset of skin inflammation at the hair follicles in keratinocyte-specific EGFR-deficient mice

Mice with constitutive deletion of EGFR in the epidermis with K5-cre (EGFRΔep) develop severe skin inflammation, and the majority of mutants die within the first weeks after birth (fig. S1A) (11). Using a tamoxifen-inducible K5-creER line (EGFRΔepER mice), we determined that only deletion before postnatal day 5 (P5) resulted in lethality as observed in EGFRΔep mice, whereas EGFR deletion after P5 did not affect the life span of the mutants (Fig. 1A). A time course analysis investigating epidermal barrier integrity by transepidermal water loss (TEWL) revealed that EGFRΔep mice were born with an intact barrier and developed a progressive barrier defect around P8 (Fig. 1B). Protein screens comparing skin from mice at P5 and P8, representing intact barrier and open barrier, respectively, confirmed increased expression of C-C motif chemokine ligand 2 (CCL2) at P5 (fig. S1B) (11). However, cytokines/chemokines previously described to be up-regulated in the inflamed skin of EGFRΔep mice were not up-regulated before epidermal barrier disruption, and increased CXCL2 and thymic stromal lymphopoietin (TSLP) were detectable only at P8 (fig. S1B). This indicates that except for CCL2, EGFRΔep keratinocytes (KCs) do not express proinflammatory cytokines before barrier disruption, excluding their direct involvement in skin barrier breakdown.

Fig. 1 Skin barrier defect and skin inflammation concur with hair development in KC-specific EGFR-deficient mice.

(A) Kaplan-Meier plot of EGFRΔepER mice treated with tamoxifen between P2 and P4 and after P5. (B) Time course analysis of transepidermal water loss (TEWL) in WT and EGFRΔep mice (n ≥ 4). ctrl, control. (C and D) Flow cytometric analysis of epidermal cell suspensions at P5, P8, and P21 of WT and EGFRΔep mice gated on CD45+ cells (C) and on MHC-IIhigh cells within the CD45/MHC-II+ population; n ≥ 3. (E) LCs (langerin, green) and their activation status (MHC-II, red) in epidermal sheets from WT and EGFRΔep mice at the indicated time points and quantification thereof (n = 2 per time point). The inset depicts a resting LC (WT P8) and an activated LC (EGFRΔep P8). (F) Epidermal sheets stained for LCs (MHC-II, red) and hair follicles epithelial cell adhesion molecule (EPCAM; green) from WT and EGFRΔep mice at P13.The arrowheads point to MHC-IIhigh LCs colocalizing with EPCAM-positive hair follicles. The inset represents an enlarged example of a hair follicle. Scale bar, 100 μm. (G) Epidermal tail sheet showing activated LC (MHC-IIhigh, red) distribution in and around the pilosebaceous unit in EGFRΔep mice at P14 (examples indicated by arrowheads). Scale bar, 500 μm. Data in (B), (C), and (D) are presented as means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001 as determined by Student’s t test.

Flow cytometric analysis of EGFRΔep epidermal cell suspensions revealed an influx of neutrophils and αβ/γδ T cells and a decrease in dendritic epidermal T cells (DETCs) at P20 (fig. S1C). At P5 and P8, however, no change in the epidermal immune cell composition, which consisted of Langerhans cells (LCs) and DETCs, could be detected (Fig. 1, C to E, and fig. S1D).

LCs are the first sensors of any kind of threat (18). Starting from P8, activated LCs (MHC-IIhigh) appeared, and LCs decreased at P21, indicating the start of the inflammation (Fig. 1, D and E). Moreover, activated LCs were localized around the hair follicles [identified with epithelial cell adhesion molecule (EPCAM) or Hoechst] as shown in epidermal sheets, but not in the dermis (Fig. 1, F and G, and fig. S1E). These results indicate that the hair follicle is the epidermal structure where the inflammation is initiated.

Lrig1+ stem cells secure barrier integrity during hair eruption via EGFR signaling

The skin barrier breakdown in EGFRΔep mice at P8 coincides with hair eruption, which is delayed in mutant mice compared with wild-type (WT) mice, in which it occurs around P5 (fig. S1A). To investigate whether hair eruption is responsible for the initiation of skin inflammation, we treated WT mice before P5, during anagen (4 weeks old) and telogen (7 weeks old) hair stages, or in adulthood (>3 months old) daily with vehicle or the EGFR-I erlotinib for 1, 2, or 4 weeks (Fig. 2, A to C, and fig. S2A). Erlotinib did not induce barrier defects or inflammation in adult mice irrespective of the hair cycle stage (Fig. 2, A to C, and fig. S2A). However, treatment before hair eruption at P4 induced early lethality, scaly skin, growth retardation, LC activation, immune influx, KC MHC-II up-regulation, and hair defects (Fig. 2D and fig. S2, B to E).

Fig. 2 EGFR in Lrig1+ stem cells secures skin barrier integrity during hair eruption.

(A to C) Adult or P4 mice were treated daily for the indicated duration (1 to 4 weeks) with vehicle or erlotinib. (A) TEWL measurement, (B) flow cytometric analysis of epidermal cell suspensions showing neutrophil recruitment, and (C) quantification by Luminex multiplex assay of TNFα, CCL2, and IL-17A proteins in cutaneous lysates (n ≥ 3). (D) Kaplan-Meier plot of mice treated daily with vehicle or erlotinib starting at P4. *P < 0.05 as determined by log-rank (Mantel-Cox) test. (E) TEWL time course and pictures of back skin of WT and hairless (hr/hr) mice treated daily with vehicle and erlotinib after waxing and tape stripping (TS; n ≥ 3). Black asterisks indicate comparisons of WT mice between vehicle and erlotinib treatment; red asterisks indicate comparisons between WT mice and hairless mice treated with erlotinib. (F) Quantification by Luminex multiplex assay of TNFα and CCL2 proteins in cutaneous lysates at indicated time points after waxing and TS (n ≥ 3). (G) TEWL time course of EGFRΔep nu/nu mice (n = 3). Black asterisks indicate comparisons of WT with EGFRΔep mice; red asterisks indicate comparisons of EGFRΔep with EGFRΔep nu/nu mice. (H) Relative wound size of full-thickness punch biopsy wounds on WT and EGFRΔep (left graph) and tamoxifen-treated EGFRf/f and EGFRΔepER mice (right graph). (I) EGFRΔLGR5, EGFRΔLRIG1, and EGFRΔLRIG1/LGR5 mice were injected with tamoxifen (P0 and P2), and MHC-IIhigh cells were quantified from epidermal tail sheets in adult mice. Graphical illustrations depict cre specificity (bold red line). Scale bar, 100 μm. The inset shows enlarged area as indicated. (J) TEWL was measured in the specific cre lines. Data in (A) to (C), (F), (I), and (J) are shown as means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001 as determined by one-way ANOVA with Tukey’s post hoc test. Data in (E), (G), and (H) are shown as means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001 as determined by Student’s t test for each time point.

We next removed the hair of adult mice by waxing and tape stripping (TS) during erlotinib/vehicle treatment (Fig. 2, E and F). Erlotinib-treated mice displayed a stronger barrier defect after waxing and TS [day 0 (d0)] and a slower barrier repair (d3 and d5) compared with vehicle treatment (Fig. 2E). The barrier defect in erlotinib-treated mice then further increased after de novo hair eruption (d5) and persisted throughout the experiment (d7 to d14), accompanied by hair growth retardation (Fig. 2E). Cytokine/chemokine expression (TNFα and CCL2) in erlotinib-treated mice followed a similar pattern, with higher expression at d3 (CCL2), undetectable amounts at d7, and a second rise of TNFα and CCL2 at d14 (Fig. 2F).

Hairless (hr/hr) mice have an intact immune system and develop skin inflammation when crossed into an EGFRΔep background (11). The reason for this is that hair morphogenesis and eruption are normal in hr/hr mice, but they gradually lose their hair starting from the first hair cycle and go bald because they are unable to reinitiate hair growth during the subsequent cycles. After the initial TS-induced epidermal disruption, barrier function normalized and remained intact in adult bald hr/hr erlotinib-treated mice, as opposed to mice with hair (Fig. 2E). This demonstrates that in the absence of de novo hair eruption, no skin inflammation develops upon EGFR inhibition. Waxing without TS did not induce barrier defects (d0 to d5) but mirrored the above observation that the barrier defect parallels the hair eruption in erlotinib-treated mice (fig. S2F). We had previously shown that the skin inflammation occurs independently of B and T cells by analyzing EGFRΔep mice in a Rag2–knockout (KO) background (11). Crossing of EGFRΔep mice with nude (nu/nu) mice, which show delayed and attenuated hair growth, revealed that in the absence of hair eruption at P8, there was no marked barrier disruption until P16, when the first hair appeared in nu/nu EGFRΔep mice (Fig. 2G).

These results establish that opening of the follicular ostia during hair eruption initiates barrier defects in the absence of EGFR. Therefore, we hypothesized that a “microwound” inflicted to the epidermis during hair eruption might not close fast enough without EGFR, thus triggering barrier breakdown and inflammation. Accordingly, we observed delayed wound healing in EGFRΔep and EGFRΔepER mice (Fig. 2H).

To identify the skin stem cell population responsible for barrier breakdown, we used the LGR5-creER, LRIG1-creER, and LGR5-creER LRIG1-creER mouse lines to delete EGFR after birth (1921). These mouse strains lack EGFR in stem cells of the bulb region (LGR5-creER), the junctional zone (LRIG1-creER), or both (LGR5-creER LRIG1-creER), while EGFR is still expressed in the interfollicular epidermis (fig. S2G). Barrier defect, immune infiltrate, and MHC-II up-regulation on LCs and KCs in the epidermis could be detected in EGFRΔLRIG1 and EGFRΔLRIG1/LGR5, but not in EGFRΔLGR5 mice, indicating that during hair eruption, EGFR signaling in LRIG1+ cells maintains the barrier integrity, thereby preventing skin inflammation (Fig. 2, I and J, and fig. S2H).

Commensal skin microbiota amplify the barrier defect and skin inflammation

Antibiotic treatment ameliorates the skin rash in patients receiving EGFR-I, and EGFR-I–treated KCs have a reduced antimicrobial peptide response (11, 22). Consequently, we hypothesized that commensal bacteria contribute to the inflammation.

We therefore measured the quantity of bacteria on the skin of EGFRΔep mice during the initial barrier breakdown (Fig. 3A). EGFRΔep mice displayed up to eight times more colony-forming units (CFUs) on their back skin as compared with cage mate controls (Fig. 3A). In addition, we could localize Gram-positive bacteria in the hair follicles of EGFRΔep mice (Fig. 3B).

Fig. 3 The outgrowth of the commensal skin microbiota amplifies the barrier defect and contributes to the skin inflammation.

(A) Blood agar plates and quantification of bacterial colony-forming units (CFUs) from skin swabs of EGFRΔep mice relative to WT cage mates (n = 4). *P < 0.05 as determined by paired Student’s t test. (B) Gram staining of skin sections from WT and EGFRΔep mice (three hair follicles are shown) at P18. Enlarged picture represents Gram-positive colonies (arrowheads) around the hair shaft. Scale bars,100 μm or 50 μm (insets). (C) Staphylococcus amplicon sequence variants (ASVs) and (D) diversity of microbiota determined by 16S rRNA gene amplicon sequencing of skin swabs from EGFRΔep mice and WT controls from separate cages at the indicated time points after birth (w, weeks; m, months). (E) Serum titers of S. aureus–specific IgG1 antibodies of WT and EGFRΔep mice at indicated time points (n = 3 to 4). (F) Kaplan-Meier plot of WT and EGFRΔep mice treated with or without antibiotics (Abx) (n ≥ 12). (G) Phenotype of mice from indicated genotypes and treatments. Enlargement shows scaly skin of EGFRΔep mice (representative images of n ≥ 3). (H) Epidermal sheets from WT and EGFRΔep mice stained for langerin (green) and MHC-II (red) with and without Abx treatment. The inset represents enlargement of the indicated area showing resting LCs (WT and WT Abx), MHC-II–positive KCs, and activated LCs (EGFRΔep and EGFRΔep Abx). (I) TEWL, (J) Ki67+ KCs (n ≥ 4), IFE, interfollicular epidermis, (K) flow cytometric analysis of epidermal single cells, (L) TNFα serum protein of WT and EGFRΔep mice under the indicated conditions (n ≥ 3). (M) Kaplan-Meier plot and (N) quantification of cutaneous cytokines/chemokines at 3 weeks of age for WT mice treated topically with vehicle or erlotinib starting from P4 under conventional (CV) and germ-free (GF) conditions. Data in (D), (E), (I) to (L), and (N) are presented as means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001 as determined by one-way ANOVA with Tukey’s post hoc test. Data in (F) and (M): *P < 0.05 and **P < 0.01 as determined by log-rank (Mantel-Cox) test.

S. aureus superinfections are a common adverse event in EGFR-I–treated patients (23). However, 16S ribosomal RNA (rRNA) gene amplicon sequencing of back skin swabs and selective cultivation revealed no changes in skin microbiota composition and diversity (Shannon index) compared with WT controls at 2 weeks of age (Fig. 3, C and D). Thereafter, however, a drastic loss of bacterial diversity occurs, concomitant with the outgrowth of the commensal Staphylococcus xylosus on 3-week-old EGFRΔep mice (Fig. 3, C and D, and fig. S3, A and B). The pathogenic dysbiosis dominated by S. aureus only occurs during the chronic phase of the inflammation in adult EGFRΔep mice (Fig. 3C and fig. S3A). In contrast, 5-month-old WT cage mate controls, although exposed to S. aureus, did not produce S. aureus–specific immunoglobulin G1s (IgG1s) in contrast to 5-month-old EGFRΔep mice (Fig. 3E). No S. aureus or S. aureus–specific IgG1s are detectable in young EGFRΔep mice (P20; Fig. 3E and fig. S3A).

To investigate whether microbiota are responsible for triggering inflammation, mice were treated with a broad-spectrum antibiotic (cefazolin) starting before barrier breakdown, which effectively eliminated bacteria from the back skin (fig. S3A). Antibiotic therapy could prevent early lethality, scaly skin, and KC-specific MHC-II expression (Fig. 3, F to H). However, LC activation, barrier defect, and KC hyperproliferation, although ameliorated, were still prominent in antibiotic-treated EGFRΔep mice (Fig. 3, H to J). Antibiotic treatment diminished additional inflammatory parameters such as activation of KCs, efflux of local immune populations, influx of neutrophils, γδ/αβ T cells, and increase in serum TNFα (Fig. 3, K and L). Likewise, most cytokines/chemokines, except CCL2 and TSLP, were reduced by antibiotic therapy (fig. S3C). To support this finding, germ-free (GF) mice were topically treated with erlotinib starting at P4 and analyzed at P20 (Fig. 3, M and N). This resulted in growth retardation in the conventional and GF environment, phenocopying EGFRΔep mice (fig. S3D). However, early lethality was ameliorated in erlotinib-treated GF mice (Fig. 3M). Similar to what was observed with antibiotic treatment, erlotinib-induced TNFα and interleukin-4 (IL-4) were still up-regulated in GF mice (Fig. 3N and fig. S3C). CXCL2 and IL-17A were not up-regulated under GF conditions or antibiotic treatment, whereas CCL2 and TSLP were up-regulated independently of microbiota (Fig. 3N and fig. S3C). In summary, our results show that antibiotic therapy in the initial phase markedly reduces inflammation, whereas the skin barrier remains open and T helper 2 (TH2) chemokines/cytokines are still expressed.

Over time, the prolonged inflammation in EGFRΔep mice causes complete hair loss (alopecia) and wrinkled scaly skin with lesions (11). Barrier disruption still persists after the onset of alopecia, and S. aureus is detectable throughout the lesioned skin, with the highest density in the superficial pustules (fig. S3E). Therefore, we investigated the effect of antibiotics on the chronic inflammation in adult EGFRΔep mice. Two weeks after antibiotic treatment, the skin appeared whitish, smooth, and a new hair cycle was initiated (fig. S3F). As opposed to the prophylactic antibiotic application, the barrier function was restored (fig. S3G). Active hair follicles could be detected, together with new cutaneous fat deposition and a reduction in epidermal thickening (fig. S3H). KC hyperproliferation was reduced, and serum TNFα was normalized (fig. S3, I and J). This demonstrates that commensal microbiota invade the skin barrier during hair eruption and an S. aureus–dominated dysbiosis prevents barrier recovery later on.

We next treated WT mice daily with vehicle/erlotinib and challenged them with S. aureus for 7 days, which was followed by a 2-week regeneration phase (fig. S3K). Erlotinib-treated mice displayed stronger inflammation, as measured by weight loss during the acute bacterial challenge (fig. S3K). Whereas the recovery phase was comparable to vehicle control, erlotinib-treated mice did not gain weight concomitant with de novo hair growth and displayed barrier defects together with increased IgE titers even 2 weeks after regeneration (fig. S3, K to M). These data indicate that the barrier defect and TH2 response are microbiota independent and that the outgrowth of commensal microbiota and invasion into the hair follicle amplify these defects and aggravate the folliculitis, which then culminate in dysbiosis and chronic inflammation.

Epidermal barrier integrity during hair eruption is crucially dependent on the EGFR–extracellular signal–regulated kinase axis

We next investigated the cell-autonomous mechanism causing the barrier breakdown during hair eruption in the absence of EGFR. Because patients treated with MEK (mitogen-activated protein kinase kinase) inhibitors display a similar papulopustular skin rash as EGFR-I–treated patients, we hypothesized that the extracellular signal–regulated kinase (ERK) pathway may be affected (24).

We therefore crossed EGFRΔep mice with mice displaying a hyperactive ERK pathway (K5-SOS transgenic mice) (25, 26). ERK phosphorylation was restored in EGFRΔep K5-SOS mice (fig. S4A). EGFRΔep K5-SOS mice had an intact skin barrier, with normalized epidermal thickness and no early lethality (Fig. 4, A to C). Skin sections stained for CD45/MHC-II, filaggrin, and loricrin showed an intact epidermal barrier and reduced inflammation in EGFRΔep K5-SOS mice (Fig. 4D). The scaly skin phenotype disappeared (fig. S4B). KC-specific MHC-II up-regulation at 3 weeks of age was still prominent, whereas LC activation and emigration, neutrophil influx, DETC efflux, and recruitment of αβ/γδ T cells were ameliorated (Fig. 4E and fig. S4, C and D). These results demonstrate that restoring the ERK activity in the absence of EGFR prevents the barrier defect, thereby reducing the inflammatory response.

Fig. 4 EGFR maintains barrier integrity during hair eruption via the epidermal ERK cascade.

(A) Analysis of TEWL, (B) epidermal thickness, and (C) Kaplan-Meier plot (n = 7 to 32) of WT, EGFRΔep, and EGFRΔep K5-SOS mice treated with or without antibiotics. Shown as means ± SEM. ***P < 0.01 as determined by log-rank (Mantel-Cox) test. (D) Skin sections of WT, EGFRΔep, and EGFRΔep K5-SOS mice at 3 weeks of age. Filaggrin (green), β4 integrin (red in the left panel or green in the middle panel; basal membrane, arrowheads), and loricrin (red) staining indicate the skin barrier in WT and EGFRΔep K5-SOS mice. CD45 (red) and MHC-II (green) staining indicates inflammation; a representative pustular eruption is shown by an asterisk (right). One representative experiment is shown out of three. Scale bars, 100 μm. (E) Characterization of immune infiltrate in epidermal cell suspensions of the respective genotype and treatment. (F) Principal components analysis and (G) heat map of a selection of significantly (P < 0.05) deregulated structural (left) and inflammatory (right) genes from RNAseq data of skin from 3-week-old WT, EGFRΔep, EGFRΔep K5-SOS, and EGFRΔep mice treated with antibiotics. Data in (A), (B), and (E) are shown as means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001 as determined by one-way ANOVA with Tukey’s post hoc test.

To mechanistically dissect the cell-autonomous versus the microbiota effects, we performed RNA sequencing (RNAseq) analysis on total skin of 3-week-old WT, EGFRΔep, EGFRΔep K5-SOS, and antibiotic-treated EGFRΔep mice. Principal components analysis revealed the superiority of EGFRΔep K5-SOS to antibiotic treatment with regard to both clustering and proximity to the WT situation (Fig. 4F). Furthermore, selected KC differentiation and barrier proteins were reexpressed in EGFRΔep K5-SOS but not in antibiotic-treated mice (Fig. 4G). Common inflammatory genes were ameliorated in both rescue models, indicating the overall reduced inflammation (Fig. 4G).

Skin inflammation consists of microbiota-dependent and microbiota-independent arms

To identify the contribution of microbiota to the inflammation, we grouped cytokines/chemokines, and their receptors were up-regulated in EGFRΔep mice and down-modulated by antibiotic treatment (Fig. 5A). Among them, TH1/17 cytokines/chemokines but also TH2 cytokines (Il4) and the Il36 family were prominent (Fig. 5A).

Fig. 5 Microbiota-dependent and microbiota-independent inflammatory signatures can be distinguished in the skin of EGFRΔep mice.

(A and B) Heat map of dysregulated cytokines, chemokines, and their receptors in EGFRΔep mouse skin that are rescued (A) or unaffected (B) by antibiotic therapy but restored in EGFRΔep K5-SOS mice. (C) Enrichment plot of FC-ε signaling Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enriched in EGFRΔep mice under antibiotic treatment as compared with EGFRΔep K5-SOS. (D and E) Mast cell (avidin, green; CD45, red) immunostaining (D) and quantification (E). (F) IgE serum titers of mice of indicated genotype and treatment (n ≥ 3). Data in (E) and (F) are shown as means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001 as determined by one-way ANOVA with Tukey’s post hoc test.

To identify the inflammatory signature linked to the barrier defect, we next grouped up-regulated cytokines/chemokines and their receptors, which are rescued in the EGFRΔep K5-SOS mice but not with antibiotics (Fig. 5B). The only two cytokines that were expressed in EGFRΔep mice under antibiotic therapy were Tslp and Il18, as well as their receptors (Crlf2 and Il18r1). Several TH2-associated chemokines remained unchanged by antibiotics but rescued by an intact barrier (for example, Ccl17; Fig. 5B and fig. S5A). Protein quantification corroborated our RNA analysis for TSLP and CCL2 (fig. S5B). Although some microbiota-induced changes could be observed in EGFRΔep K5-SOS mice (for example, MHC-II on KCs; Fig. 4E), IL-17A expression vanished as it did with the antibiotic treatment alone, confirming the restricted bacterial entry (fig. S5B).

The comparison of antibiotic-treated EGFRΔep to EGFRΔep K5-SOS mice revealed an enrichment in chemokine signaling, FC-ε signaling, and FC-γ–mediated phagocytosis pathways, indicating microbiota-independent regulation (Fig. 5C and fig. S5A). This prompted us to investigate mast cell accumulation and IgE titers in EGFRΔep K5-SOS mice, which were both normalized, whereas these TH2 hallmarks were unchanged by antibiotics (Fig. 5, D to F).

Skin barrier defects, TH2 cytokines, increased IgE titers, and S. aureus susceptibility are all hallmarks of atopic dermatitis (AD) (27). We therefore compared our mouse models with a human AD expression signature (28), which confirmed their relationship (fig. S5C). Moreover, the identified genes were either rescued by the intact barrier (SOS expression) or by the antibiotic therapy or by both (fig. S5C).

In summary, the skin inflammation induced by EGFR inhibition consists of a bacterial-induced inflammatory arm dominated by TH1/17 cytokines and a TH2-driven response not affected by microbiota. Moreover, restoring epidermal barrier integrity suppresses these cutaneous TH2-biased immune responses and simultaneously dampens bacterial-induced TH1/17 responses.

Human patients with defective EGFR signaling display atopic hallmarks

We next matched RNAseq data from a patient with a null mutation in EGFR to a human AD gene signature (15, 28) and found a strong correlation between the most up-regulated (red) and down-regulated (blue) genes (Fig. 6A). We found that two EGFR ligands, epiregulin (EREG) and betacellulin (BTC), were down-regulated in patients with AD (Fig. 6A).

Fig. 6 Atopic hallmarks are present in human patients with defective EGFR signaling.

(A) Correlation plot of log fold change (logFC) comparing human atopic dermatitis (AD) gene signatures (89ADGES) and a human EGFR-mutant patient signature. Plot shows up-regulated genes in both conditions (red dots) and down-regulated genes in both conditions (blue dots). EGFR ligands (Ereg and Btc) are highlighted in red. (B and C) Skin biopsies of patients with SCC stained for filaggrin (B) and TSLP (C) before and after cetuximab therapy. The arrowheads indicate filaggrin expression. A patient with AD was used as a TSLP-positive control. (D) Gram staining after cetuximab treatment. The insets show a magnification of indicated areas. Scale bar, 50 μm. (E) Imaging mass cytometry of skin samples before and after cetuximab treatment. Colors are as indicated in the figure. The image was split into three for better visualization. Scale bars, 100 μm. FSP-1, fibroblast-specific protein 1; LAG3, lymphocyte-activation gene 3.

We next analyzed the skin tissue of two patients with inoperable squamous cell carcinoma (SCC) before and during cetuximab (EGFR-I) treatment (fig. S6, A and B). Filaggrin expression was reduced during EGFR-I treatment (Fig. 6B and fig. S6C). TSLP is a hallmark AD cytokine, inducing TH2 responses (29). TSLP expression in EGFR-I–treated patients was comparable to that in a patient with AD (Fig. 6C). In addition, Gram staining of EGFR-I–treated patient skin revealed the presence of Gram-positive bacteria (Fig. 6D and fig. S6D). Imaging mass cytometry and Giemsa staining confirmed that the folliculitis is characterized by a mixed immune infiltrate, similar to EGFRΔep mice, consisting of T cells (CD8, CD3, and Lag3), macrophages (CD163 and CD68), mast cells (Giemsa), neutrophils (CD15), and strong MHC-II (human leukocyte antigen–DR isotype) expression (Fig. 6E and fig. S6, A, B, and E). We therefore conclude that the inflammation occurring in mice and humans with null mutations in EGFR or in EGFR-I–treated patients is similar to TH2-driven atopic-like inflammation.

Fibroblast growth factor 7 restores ERK signaling independent of EGFR to secure epidermal barrier integrity

We next hypothesized that ERK activation independent of EGFR might represent a feasible therapeutic option for EGFR-I–associated adverse events. Fibroblast growth factor 7 (FGF7), also known as KC growth factor, is a promising candidate because its receptor is dominantly expressed on KCs and its signaling pathways are similar to EGFR (30).

Subcutaneous injection of FGF7 in EGFRΔep mice starting before barrier disruption (P8) resulted in ERK phosphorylation and expression of its target early growth response protein 1 (EGR1) (Fig. 7A) (31). FGF7 treatment effectively prevented immune infiltrate and MHC-II up-regulation (Fig. 7, B and C, and fig. S7, A and B). The epidermal barrier remained intact; the epidermal thickening was prevented, and filaggrin was expressed at the FGF7 application site (Fig. 7, D to F). Furthermore, FGF7 but not phosphate-buffered saline (PBS) treatment minimized or completely prevented the formation of pustular eruptions (Fig. 7G). In contrast, in a reactive setting, FGF7 was not able to reverse barrier defects, rescue hair growth, and suppress inflammation in adult EGFRΔep mice (fig. S7, C and D). To exclude that FGF7 treatment might induce tumor growth, we treated EGFRΔep K5-SOS mice daily with subcutaneous FGF7 for 2 weeks but did not observe papilloma formation (fig. S7, E to G).

Fig. 7 FGF7 treatment activates epidermal ERK signaling and prevents barrier disruption and inflammation.

(A) Skin sections of WT and EGFRΔep mice at 3 weeks of age treated with PBS or rmFGF7 and stained for EGR1 and phospho-ERK1/2. Scale bars, 100 μm. (B and C) Quantification of immune infiltrate using antibodies against (B) CD45 and (C) MHC-II. A minimum of five microscopic fields per mouse were counted. (D) Filaggrin (green) and β4 integrin (red, basal membrane) show the barrier (arrowheads). One representative experiment is shown. Scale bar, 100 μm. (E) TEWL and (F) epidermal thickness of EGFRΔep mice treated with PBS or rmFGF7. (G) Volume of pustular eruptions was measured on skin sections of mice treated with PBS or FGF7 as indicated. Each dot represents the volume of a respective pustule in an independent skin section. No detectable pustules per skin section were set to 0 (as in WT mice). The graph represents the summary of two independent experiments. (H) Schematic overview of the phenotype of all mouse models used with respect to the bacterial status and barrier integrity as indicated. Data in (B), (C), and (E) to (G) are shown as means ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001 as determined by one-way ANOVA with Tukey’s post hoc test.

In summary, these results provide evidence that FGF7 is able to activate ERK independent of EGFR in vivo. This represents preclinical evidence for the prophylactic potential of FGF7 to manage skin toxicities induced by EGFR-I.


Here, we provide evidence that the similarity of the skin inflammation between EGFR-I and RAS/RAF/MEK/ERK inhibitors is due to the dominant role of EGFR in controlling the ERK signaling cascade during hair shaft penetration in LRIG1+ hair follicle stem cells. Epidermal overexpression of SOS rescued the barrier defects and the consequent inflammatory cascade induced by epidermal EGFR deletion. Thus, we mechanistically clarified the chain of events causing skin toxicities in patients with cancer treated with EGFR-Is.

With the investigated mouse models, we can attribute the majority of the TH2 signature to the barrier breakdown, which occurs at the time of hair eruption and allows skin bacteria to enter the hair shaft (Fig. 7H). The commensal microbiota then exacerbate some responses (for example, TNFα and the TH2 cytokines IL-4 and IL-33) and additionally initiate TH1/TH17-dominated responses, which are responsible for the early death of EGFRΔep mice. In the clinic, antibiotics in EGFR-I–treated patients only reduce rash severity but hardly affect overall rash incidence (4). On the basis of our findings, we can now speculate that antibiotics diminish commensal microbiota, thereby reducing rash severity, and that the overall rash occurrence is due to the barrier defect and concomitant TH2 responses around the erupting hair shaft. This, in turn, opens the way for feasible and efficient treatment options, which target TH2 cytokines such as TSLP, IL-4, or TH2 chemokines such as CCL17 thymus and activation regulated chemokine (TARC), as already done or planned in AD (32).

It is intriguing that the cutaneous transcriptional profile together with the general inflammatory hallmarks observed in EGFRΔep mice and a patient with an EGFR null mutation resemble the signatures observed in human AD. AD is described as a TH2-dominated mixed immune reaction combined with barrier defects, dry and itchy skin, IgE accumulation, and S. aureus superinfections (32).

It would therefore be interesting to investigate the use of EGFR ligands to treat patients with AD. We identified two EGFR ligands, EREG and BTC, as down-regulated in human AD. This indicates that EGFR signaling is reduced in AD and suggests that EGFR activation might be beneficial to ameliorate AD symptoms. The observation that EREG KO mice develop spontaneous dermatitis and EGF ameliorates AD in mouse models supports this possibility (3335).

Filaggrin mutations account for the majority of AD susceptibility in humans (32). Diminished filaggrin expression might predispose the skin of EGFR-I–treated patients to AD-like flares and might be responsible for the inflammation initiated by new hair eruption. In the clinic, the skin rash appears in most patients about 1 to 2 weeks after EGFR-I treatment initiation specifically on the face and the upper trunk region, which harbor the highest density of hair follicles (36). Conversely, regions with terminally differentiated hair on arms, legs, and head remain largely unaffected during the initial rash in EGFR-I–treated patients (3). This suggests that the face and upper trunk regions harbor a certain fraction of actively cycling and erupting hair follicles, which are then disrupted during the early EGFR-I treatment phase. Clinical observations support our hypothesis: (i) Several case studies reported that EGFR-I–associated skin rash spared previously irradiated skin, which is devoid of active hair follicles due to the irradiation, and (ii) EGFR-I–induced skin rash might be triggered by hair regrowth after chemotherapy, which is often given in combination with EGFR-I (3739).

Here, we provide evidence that barrier defects and their resulting TH2 signature are the initial trigger and that the subsequent bacterial invasion is mediated by the commensal microbiota. We can now speculate that EGFR-I treatment first causes an overexaggerated immune reaction and, in a second step, misdirects responses to bacteria due to barrier defects and their resulting IgE response. This might favor the early outgrowth of certain commensal bacterial species such as S. xylosus and eventually results in an almost complete takeover of the pathobiont S. aureus as a secondary event during folliculitis.

Recent data from ADAM17sox9 mice described the bacterial dysbiosis dominated by S. aureus as the driver of the chronic skin inflammation (14). We now identified the cause of the initial sterile barrier disruption (hair eruption), its mechanism (dysfunctional ERK cascade), and the responsible stem cell population (LRIG1+ cells) and showed that the vast majority of the resulting TH2-dominated inflammation is microbiota independent.

Our findings also provide a feasible therapeutic anchor point for management of the rash incidence and prevention of S. aureus colonization. Prophylactic application of FGF7 restored epidermal ERK activation, corrected KC differentiation, and prevented barrier defects during hair eruption. Recombinant FGF7 (palifermin) therapy is already in clinical use to ameliorate the symptoms of radiation-induced mucositis in patients with leukemia receiving myeloablative radiotherapy and is therefore a valid treatment option for the management of EGFR-I–induced skin rash (40). Furthermore, we did not observe any influence of FGF7 treatment on tumor induction using the oncogenic K5-SOS mouse model in an EGFR-deficient background. These results were independently confirmed in xenograft models of head and neck and colorectal carcinoma. FGF7 treatment had no effect on tumor growth and did not counteract the efficacy of EGFR-I alone or in combination with chemotherapy (41).

However, results obtained so far rely on experimental mouse models, which await clinical testing before translation of our findings to patients. Consequently, the applicability of FGF7 treatment in patients with cancer has to be carefully evaluated before implementation into treatment regimens, and further studies will be required to determine the clinical safety and relevance as a supportive care treatment.

Our data demonstrate that EGFR is the epidermal master regulator of the ERK signaling cascade, a finding that converges with a common mechanism of skin inflammation described in patients harboring ADAM17 and EGFR mutations and patients with cancer treated with EGFR-Is, MEK/ERK inhibitors, and multikinase inhibitors. We determined that the barrier integrity during hair eruption is maintained by LRIG1+ hair follicle stem cells. Consequently, hair eruption in the absence of EGFR signaling induces a sterile TH2-dominated response and allows bacterial invasion through the follicular ostia, which results in a mixed AD-like immune response. Last, we could translate our mechanistic findings into a preclinical therapeutic approach by applying FGF7, which prevents barrier defects independent of EGFR. Management of cutaneous side effects of targeted cancer therapies should allow more aggressive treatment regimens and combination therapies without dose adaptations or cessation, and it should improve patients’ quality of life. In addition, it may be possible to extrapolate our findings to AD, improving our understanding of disease ontogeny and therapeutic possibilities by highlighting barrier defects as the initiating event.


Study design

The objective of this study was to elucidate the mechanistic details responsible for the cutaneous side effects of EGFR-I treatment during targeted cancer therapy. We then used our findings to propose a feasible supportive care therapy. All experiments were designed to use the smallest number of mice that allowed us to perform adequately powered statistical analyses. Mice were randomly allocated to treatment and experimental groups, and the number of biological replicates for each experiment is specified in the figure legends. Each experiment was independently repeated at least twice. Outlier detection and removal were not included in the study design. Blinding was not possible during treatment because of the visual phenotype of the skin inflammation but was used whenever possible during quantification and assessment of sample material. Mice were treated with pharmacological inhibitors, recombinant protein, systemic antibiotics, and tamoxifen and were subjected to hair removal in certain experiments. Specific scientific questions were addressed by comparing GF mice with mice raised in a conventional environment. At the indicated time points, the skin microbiome was characterized, and TEWL was determined. Human and murine tissue samples were further subjected to downstream analysis including flow cytometric analysis, histology, immunofluorescence analysis, immunohistochemistry, multiplex immunoassays, and RNAseq. One experimental setup included an S. aureus infection model. Primary data are provided in data file S1.


EGFRΔep and EGFRΔepER mice were generated as previously described (11). K5-SOS transgenic mice have been previously described (26). Hairless mice (hr/hr; Skh-1) were purchased from Charles River Laboratories, and athymic nude mice were purchased from Harlan Sprague Dawley Inc. (acquired by Envigo Biosciences in 2015). The aforementioned mice were bred and maintained in the facilities of the Medical University of Vienna in accordance with institutional policies and federal guidelines. All mice had access to food and water ad libitum. Animal experimental procedures were approved by the Animal Experimental Ethics Committee of the Medical University of Vienna and the Austrian Federal Ministry of Science and Research (animal license numbers: GZ 66.009/124-BrGT/2003, GZ 66.009/109-BrGT/2003, GZ BMWF-66.009/0073-II/10b/2010, GZ BMWF-66.009/0074-II/10b/2010, GZ BMWFW-66.009/0200-WF/II/3b/2014, and GZ BMWFW-66.009/0199-WF/II/3b/2014).

GF C57BL/6 mice were kept under sterile conditions in Trexler-type plastic isolators, exposed to 12-hour light/12-hour dark cycles, and supplied with autoclaved tap water and 50-kilogray irradiated sterile pellets (breeding diet: Altromin 1414) ad libitum. Axenicity was assessed every 2 weeks by confirming the absence of bacteria, molds, and yeast by aerobic and anaerobic cultivation of mouse feces and swabs from the isolators in VL (Viande-Levure), Sabouraud dextrose, and meat-peptone broth and subsequent plating on blood, Sabouraud, and VL agar plates. Conventional specific pathogen–free C57BL/6 mice (WT) were kept in individually ventilated cages (Tecniplast), exposed to 12-hour light/12-hour dark cycles, and fed with the same sterile diet as their GF counterparts. Animal experiments were approved by the committee for protection and use of experimental animals of the Institute of Microbiology of the Czech Academy of Science v.v.i. (approval ID 117/2013).

Tamoxifen treatment

K5-CreERT transgenic EGFRf/f mice were intraperitoneally injected with 1 mg of tamoxifen per 25 g body weight on five consecutive days as previously described (11). After initial deletion, mice received tamoxifen twice a week for maintenance. Before the wound healing analysis, mice additionally received topical administration of 4-hydroxy-tamoxifen (Sigma-Aldrich; 4 mg 4-OH tamoxifen dissolved in 0.2 ml of acetone) every other day for a total of 2 weeks before full-thickness punch wounds were applied. Lrig1creER and LGR5creER mice were subcutaneously injected with tamoxifen (1 mg/25 g body weight) on P0 and P2.

Wound healing assay

Punch wounds (5-mm full thickness) were placed on the dorsal skin of 3.5-week-old EGFRΔep mice and tamoxifen-treated EGFRΔepER mice and their respective littermate controls. Wound closure was assessed, and skin biopsies were taken at the indicated time points.

Erlotinib treatment (conventional environment)

Animals received daily intraperitoneal injections of erlotinib (Apollo Scientific Ltd.) dissolved in 0.5% methyl cellulose (Sigma-Aldrich) or vehicle alone at a concentration of 50 mg/kg per day unless otherwise stated. To avoid cross contamination of erlotinib, all animals in each cage were treated with either erlotinib or vehicle.

Erlotinib treatment (GF environment)

For experiments under GF conditions, erlotinib administration was adapted to account for the technical and experimental restrictions and limitations of the isolators. Therefore, several topical erlotinib administration protocols were tested to identify the dose equivalence that resulted in comparable inflammatory hallmarks to the mice intraperitoneally injected with erlotinib (50 mg/kg per day) under conventional housing. Consequently, erlotinib (LC Laboratories) was dissolved in dimethyl sulfoxide (DMSO) at a concentration of 50 mg/ml, sterile filtered, and applied topically daily (50 μl until treatment day 8 and 100 μl from treatment day 9 until sample collection to achieve 1 g/kg per day of topical erlotinib) on the dorsal skin of GF and conventionally housed mice starting at P4. Control mice received topical DMSO. To avoid cross contamination of erlotinib, all animals in each cage were treated with either erlotinib or vehicle.

Hair removal and TS

Vehicle and erlotinib treatment was initiated 2 days before and continued daily throughout the experiment. For hair removal, mice were anesthetized, dorsal skin was shaved with an electrical animal razor (Aesculap GT608), and hairs were removed with cold wax according to the manufacturer’s instruction (Veet). Where indicated, mice were tape stripped 15 times with commercially available adhesive tape (Tesafilm).

Antibiotic treatment

Mutant mice were cohoused with their respective controls, and mice were treated with cefazolin (0.5 g/liter) (Astro Pharma) supplemented with drinking water ad libitum. Cefazolin was freshly prepared twice a week and protected from light. Mouse cages were replaced every other day for increased cleanness.

FGF7 treatment

EGFRΔep mice were injected subcutaneously daily in the neck with recombinant murine FGF7 (BioLegend) or PBS at a concentration of 2.5 μg/day per mouse starting P8. For treatment of adult EGFRΔep and EGFRΔep K5-SOS mice, the dose was increased to 10 μg of recombinant FGF7, which was subcutaneously injected daily for 14 days.

TEWL measurement

TEWL was measured with a Tewameter TM 300 probe attached to the MDD4 display device (Courage+Khazaka) according to the manufacturer’s recommendations.

Preparation of cutaneous protein lysates and Luminex protein quantification

Dorsal skin biopsies were taken and immediately snap frozen and stored at −80°C for further use. For homogenization, skin biopsies were added to radioimmunoprecipitation assay lysis buffer, supplemented with Protease Inhibitor Cocktail (Roche), and homogenized in Precellys tubes containing ceramic beads (VWR), using a Precellys 24 homogenizer (Bertin; 2× 30 s at 6000 rpm followed by 30 s on ice after each cycle). Skin lysates were transferred to Eppendorf tubes and centrifuged at 14,000g for 15 min at 4°C to remove cell debris. The supernatant was transferred to a new Eppendorf tube and subsequently snap frozen and stored at −80°C. For protein quantification, cutaneous lysates were thawed on ice and subjected to Bradford protein quantification according to the manufacturer’s protocol (Bio-Rad). Seventy to 100 μg of total protein was used for each assay. For quantification of serum proteins, 50 μl of murine serum was used. Multiplex Luminex assays (Thermo Fisher Scientific) were performed according to the manufacturer’s recommendations and measured on a Luminex MAGPIX System using the xPONENT Software.

Preparation of epidermal single-cell suspensions and flow cytometry analysis

Mice were euthanized, and mouse ears were split into the dorsal and ventral side and placed on 0.8% trypsin (Gibco) for 45 min at 37°C to allow the separation of the epidermis and dermis. The epidermis was cut into small pieces, further digested in deoxyribonuclease I (250 μg/ml) for 30 min at 37°C, washed, and filtered using a 70-μm cell strainer.

Single-cell suspensions were subsequently blocked with FC Block (BD Pharmingen) and stained with indicated fluorescently labeled antibodies (BioLegend) at 4°C for 30 min. Before flow cytometric analysis, SYTOX Blue Dead Cell Stain (Invitrogen) was added according to the manufacturer’s recommendations to identify dead cells. Cells were recorded using an LSR II flow cytometer (BD Biosciences) and analyzed using FlowJo software 7.6.4.

Preparation of epidermal sheets

To separate the epidermis from the dermis, split mouse ears were floated at 37°C with the dermal side facing down on 3.5% ammonium thiocyanate for 25 min, and mouse tail sheets were floated on 20 mM EDTA for 2.5 hours. They were subsequently fixed with 4% paraformaldehyde (PFA) for 30 min at room temperature and further subjected to immunofluorescence staining.

Histological analysis and immunofluorescence microscopy

Dorsal skin was harvested and fixed in 4% PFA, embedded in a paraffin block, and cut into 5-μm sections. After dewaxing and rehydration, sections were stained with hematoxylin/eosin or Giemsa according to standard procedures. For visualization of bacteria, skin sections were stained using the Gram staining kit for tissue (Sigma-Aldrich). For immunohistochemistry assays, fixed skin samples were subjected to antigen retrieval using 2100 Retriever (BioVendor) according to the manufacturer’s protocol, blocked with hydrogen peroxide and goat serum, and subsequently stained with indicated primary antibodies. SignalStain detection reagent (Cell Signaling Technology) was used for visualization according to the manufacturer’s instructions. For immunofluorescence staining of cryosections, dorsal skin was embedded in OCT (Sakura), immediately frozen, cut into 5-μm sections, and postfixed with 4% PFA for 30 min at room temperature. Subsequently, skin sections or epidermal sheets were blocked with 5% goat serum, 2% bovine serum albumin (BSA) tris-buffered saline–Tween (TBS-T) for 1 hour and incubated with primary antibodies diluted in 5% goat serum, 2% BSA TBS-T at 4°C overnight. Subsequently, slides were rinsed and incubated with an appropriate secondary antibody and Hoechst (Sigma-Aldrich) for 2 hours in a dark humidified slide chamber. Tissue sections were mounted, and pictures were taken using a Nikon Eclipse 80i microscope.

16S rRNA gene amplicon sequencing of cutaneous swabs

For 16S rRNA gene sequencing, WT mice were analyzed at the indicated time points without any EGFRΔep mice present in the litter to avoid contamination with the microbiome of the mutant mice. Eurotubo Collection swabs (Deltalab) were prewetted in 1 ml of sterile PBS, and a defined area of dorsal skin was sampled (vigorously swabbed for 10 s). Negative control swabs were prewetted and exposed to the room environment. The QIAamp DNA Microbiome Kit (QIAGEN), which includes a step for depletion of host cell DNA, was used for DNA extraction according to the manufacturer’s guidelines.

Amplicon sequencing of 16S rRNA genes was performed as described previously (42). Sequences were amplified and barcoded using a two-step polymerase chain reaction (PCR) approach. In the first step, 16S rRNA genes were amplified using degenerate primers that target most bacteria and archaea (H_341F 5′-GCTATGCGCGAGCTGCCCTACGGGNGGCWGCAG and H_785R 5′-GCTATGCGCGAGCTGCGACTACHVGGGTATCTAATCC; both primers contain a universal “HEAD” sequence as target for a barcoding primer in the second PCR) (42). Blank nucleic acid extractions and negative (water only) PCRs were included as controls. All first-step PCRs were prepared in triplicate (20-μl volume) containing 1× DreamTaq buffer (Thermo Scientific), 0.2 mM deoxynucleotide triphosphate mix (Thermo Scientific), 1-U DreamTaq (Thermo Scientific), BSA (0.2 mg/ml) (Thermo Scientific), 1 μM each forward and reverse primer mix, and 1 μl of template. Thermal cycling conditions were 95°C for 3 min; 25 cycles of 95°C for 30 s, 52°C for 30 s, 72°C for 1 min; final extension at 72°C for 7 min. After confirmation of product formation by gel electrophoresis, replicate PCRs were pooled and cleaned using the Zymogen DNA Clean and Concentrator kit (Zymo Research Corp.). Amplicons were eluted in 30 μl of nuclease-free water. Second-step barcoding PCRs (50-μl volume) contained 1 μl of the cleaned first-step PCR product as template and were subjected to thermal cycling conditions of 95°C for 3 min; 5 cycles of 95°C for 30 s, 52°C for 30 s, 72°C for 1 min; final extension at 72°C for 7 min. PCR products were again checked by gel electrophoresis and cleaned as previously described. DNA was quantified using Quant-iT PicoGreen double-stranded DNA (dsDNA) kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. Barcoded amplicons from different samples were pooled at equivalent copy numbers (2 × 1010) and paired-end sequenced (Microsynth AG) on an Illumina MiSeq sequencer system [2 x 300 base pairs (bp)]. Sequencing results were processed into library-specific paired-end reads as outlined previously (42). Paired-end reads were assembled using fastq-join, clustered into chimera-filtered operational taxonomic units (OTUs) (97% identity) with UPARSE, and taxonomic classification was determined with the Ribosomal Database Project (RDP) classifier implemented in Mothur.

16S rRNA gene amplicon sequence variants (ASVs) at single-nucleotide resolution were determined from paired-end reads using dada2 with the filterAndTrim (truncQ = 2, minLen = 250), learnErrors(), derepFastq(), dada(), and mergePairs() functions, and taxonomic classification was determined with the RDP classifier implemented in Mothur. ASVs classified at the genus level (>80% confidence) as Staphylococcus were placed into a reference tree of Staphylococcus strains using RAxML-EPA. The reference tree was constructed using RAxML from 74 type strain 16S rRNA sequences obtained from the RDP and aligned with the SINA aligner. The presence of contaminant sequences particularly afflicting low-biomass samples is well documented (43, 44). Therefore, a conservative curation of identified OTUs was performed. First, we retained OTUs with greater than 10 read counts in any one sample. Second, OTUs with greater read counts in negative controls (DNA extraction and PCR controls) were removed. Third, to account for potential cross contamination occurring during sequencing, OTUs in other barcoded samples with read counts three orders of magnitude higher were examined as possible candidate cross-contaminating sequences.

For alpha-diversity analyses, ASV tables were imported into the R software environment (R Core Team, 2015) and processed within the package phyloseq (45). The datasets were subsampled at the depth of the smallest library. Rarefied ASV tables were then used to calculate alpha-diversity indices (Shannon diversity, Simpson, and inverse Simpson).

Species (for example, Staphylococcus-type species) within a species group (S. aureus/Staphylococcus argenteus versus S. xylosus/ Staphylococcus saprophyticus) had identical 16S rRNA gene sequences and, thus, could not be distinguished by 16S rRNA sequencing. Therefore, parallel cultivation-based analyses using Staphylococcus identification plates (SAID, bioMérieux) revealed the identity of the species within each species group.

Bacterial cultivation of murine dorsal skin swabs

For analysis of young WT mice, only litters without EGFRΔep mice present were selected and sampled to avoid contamination with the mutant microbiome (indicated as “separate cage”) unless otherwise stated (indicated as “cage mate”). Therefore, a defined area of dorsal mouse skin was sampled (vigorously swabbed for 10 s) using PBS-prewetted compact dry swabs (HyServe), and an aliquot was plated on Staphylococcus identification plates (SAID) and blood agar (bioMérieux), respectively. Plates were incubated as recommended by the manufacturer to allow quantification of the cutaneous bacterial load and the presence of S. aureus. Pictures of the agar plates were taken after development of visible colonies as recommended.

S. aureus culture and epicutaneous skin infection

The methicillin-resistant S. aureus strain USA300 was cultured as described previously (46). Briefly, frozen S. aureus USA300 was streaked on tryptone soy agar broth (Oxoid) 2 days before infection and incubated overnight at 37°C. On the following day, a single S. aureus colony was picked to inoculate 10 ml of tryptone soy broth and incubated overnight at 37°C on a shaker at 180 rpm. On the day of infection, the overnight culture was diluted 1:300 in 30 ml of ice-cold tryptone soy broth in an Erlenmeyer flask and cultured 2 to 2.5 hours at 37°C and 180 rpm until reaching an OD600 (optical density at 600 nm) between 0.5 and 0.8. The bacterial suspension was then cooled on ice for 10 min and centrifuged for 5 min at 1800g at 4°C. The pellet was resuspended in sterile PBS at 109 CFUs/ml and kept on ice until infection (an aliquot was used for serial dilutions plated for titer determination). For epicutaneous skin infection of mice, age-matched 8- to 10-week-old female C57BL/6J mice were anesthetized by intraperitoneal injection of a combination of Ketasol and Rompun in NaCl and shaved on the back, flanks, and abdomen with a clipper (Oster), followed by five times TS of the back skin. Subsequently, a patch of sterile gauze (about 1 cm by 1.2 cm) inoculated with 100 μl of S. aureus suspension (108 CFUs) was placed onto the tape-stripped skin site, where it was fixed with adhesive transparent film (Tegaderm 16002, 3M) (47). The body weight of the mice was assessed daily. On day 21, mice were anesthetized, and TEWL at the site of infection was assessed. Blood was collected into Z-Gel tubes (Sarstedt) for serum preparation, and mice were euthanized. For assessment of the contribution of EGFR to the immune response in this infection model, mice were intraperitoneally injected with erlotinib or vehicle (50 μg/kg per day) daily, starting 2 days before infection.

Quantification of murine serum IgE and S. aureus–specific IgG antibodies

Total mouse serum IgE was determined as previously described (48). Briefly, MaxiSorp enzyme-linked immunosorbent assay (ELISA) plates (Nunc) were coated with purified rat anti-mouse IgE (clone R35-72, BD Biosciences) at 2 μg/ml overnight at 4°C, followed by blocking with PBS 1% BSA for 2 hours at room temperature. Mouse sera were diluted 1:5 in PBS 1% BSA and incubated on the blocked plates simultaneously with serial dilutions of purified mouse IgE (as standard; BD Pharmingen) overnight at 4°C. Bound IgE was detected with 3, 3′, 5, 5′tetramethyl benzidine (TMB) substrate (Sigma-Aldrich) after incubation of the wells for 1 hour at room temperature with biotinylated rat anti-mouse IgE (clone R35-118, BD Pharmingen), diluted 1:1000 in PBS with 1% BSA, followed by 20 min with avidin–horseradish peroxidase (eBioscience), diluted 1:250 in PBS with 1% BSA.

For detection of IgG1 antibodies specific for S. aureus–secreted proteins, an overnight S. aureus USA300 culture was prepared as described above, diluted 1:100 in fresh tryptic soy broth, and grown for 8 hours at 37°C at 180 rpm. The suspension was then centrifuged for 10 min at 1800g at 4°C. The supernatant was then filtered using a 0.45-μm pore size Stericup filter (Millipore) and 10 times concentrated using spin concentrators with 10 K MWCO (Pierce) and subsequently dialyzed two times overnight against PBS at 4°C using 10 K MWCO slyde-a-lyzer cassettes (Pierce). The processed S. aureus supernatant (SASN) was lastly stored at −80°C. For the ELISA, MaxiSorp plates were coated overnight at 4°C with SASN diluted 1:250 in PBS. After blocking as described above, serial serum dilutions (in PBS) were loaded onto the plates and incubated overnight at 4°C. Bound IgG1 was detected using biotinylated rat anti-mouse IgG1 (clone A85-1, BD Pharmingen) diluted 1:1000 in PBS 1% BSA, otherwise as described above for total IgE. For calculating the SASN-specific antibody titers, the serum dilution that gave half-maximal signal of a reference serum (a pool of immune sera from S. aureus–infected mice) was plotted (48).

RNAseq analysis of cutaneous biopsies

Mutant mice were cohoused with their respective littermate controls, and skin biopsies were immediately immersed in RNA later (Life Technologies) and stored at −80°C until further processing. For RNA isolation, TRIzol reagent was used as recommended, and subsequently, RNA was subjected to an additional column-based purification (RNeasy Kit, QIAGEN), snap frozen, and stored at −80°C. Quality control of RNA samples was performed using RNA 6000 Nano Kit on a 2100 Bioanalyzer (Agilent). Sequencing libraries were prepared at the “Core Facility Genomics,” Medical University of Vienna, using the NEBNext Ultra Directional RNA Library Prep Kit for Illumina according to the manufacturer’s protocols (New England Biolabs). Libraries were QC checked on a Bioanalyzer 2100 (Agilent) using a High Sensitivity DNA Kit for correct insert size and quantitated using Qubit dsDNA HS Assay (Invitrogen). Pooled libraries had an average length of 330 to 360 bp and were sequenced on a NextSeq500 instrument (Illumina) in 1 × 75 bp sequencing mode.

RNA counts were read into DESeq2 package and preprocessed according to the instructions of the package provider. Differentially expressed genes and logFCs were determined with LIMMA using a linear model with the blocking factor “mouse” and the factor “treatment/KO.” P values were adjusted for multiple testing according to Benjamini Hochberg.

Data acquisition and gene set enrichment analysis

Datasets E_GEOD-54162 (GSE54162) were downloaded from ArrayExpress and preprocessed using the ArrayExpress and lumi packages in R (49, 50). LogFCs were calculated using the LIMMA package. P values were adjusted for multiple testing according to Benjamini Hochberg. Homolog matching between mouse and human genes was done using the biomaRt package of R (51). For signature acquisition, the 89ADGES signature (logFCs and P values) was extracted from the supplementary materials of (28).

Gene set enrichment analysis (GSEA) was done using the GSEA executable provided by the Broad Institute with the ranking metric Difference_of_Classes of the log2-transformed data and gene sets C2 and C5 provided by the Molecular Signature Database (52, 53). A false discovery rate of <0.25 was considered to be significant.

Patient skin biopsies

Skin biopsies were obtained from two patients from the Department of Dermatology Rudolphstiftung Hospital (Vienna, Austria) who received cetuximab for the treatment of inoperable SCC. In both patients, clinically involved skin was photographed, and biopsies were taken after the diagnosis of a “cetuximab-induced papulopustular skin rash.” In both patients, leftover and/or tumor-adjacent tissue obtained before cetuximab initiation served as a pretreatment control. Neither of the patients received systemic antibiotic treatment for at least 4 weeks before biopsy collection. The patients gave consent for the retrospective use of their data and samples.

Imaging mass cytometry and imaging acquisition

Antibody labeling, sample treatment, antibody staining, and image acquisition were done as previously described with slight deviations (54). Briefly, antibody staining without RNA staining was performed as follows. Samples were treated 2 × 5 min in fresh Xylol, followed by 2 × 1 min in 100% fresh ethanol, and then air dried. Samples were treated with 10% H2O2 for 10 min; then, liquid was removed, and slides were immediately submerged in preheated target retrieval buffer (RNAscope target retrieval buffer, Advanced Cell Diagnostics) for 15 min at 98°C. Samples were then quickly dipped into ddH2O, then into fresh 100% ethanol, and then air dried. A hydrophobic barrier was drawn around the area of interest (ImmEdge Hydrophobic Barrier PAP Pen), and samples were treated with protease III solution (RNAscope kit, Advanced Cell Diagnostics) for 30 min at 40°C. Slides were then submerged briefly in ddH2O and transferred to TBS (pH 7.5). An antibody master mix in TBS-T (0.1%) was prepared, and samples were stained at 4°C in a wet chamber overnight (see table S1 for antibodies and concentrations). The next day, slides were washed for 5 min in TBS and then stained for 5 min in a 1:1000 dilution of 500 μM MaxPar Intercalator-Ir (Fluidigm) in PBS. Slides were washed for 5 min in PBS, dried under airflow, and stored at room temperature until measurements.

Image acquisition was performed on a Helios (Fluidigm) coupled with a Hyperion Imaging System (Fluidigm) at 200 Hz. Raw data (.mcd files) were converted to multipage tiff files using HistoCAT++ (55). False color images of individual channels were prepared in FIJI open source software with a Gaussian blur filtering step (sigma = 0.5).


Statistical comparisons were performed using GraphPad Prism 5.02 software. Heat maps of RNAseq data were calculated with Microsoft Excel 2010. Statistical significance was determined using Student’s unpaired two-tailed t test for comparisons of two groups and one-way analysis of variance (ANOVA), followed by Tukey’s post hoc test for multiple comparisons. Kaplan-Meier survival curves were generated, and log-rank (Mantel-Cox) test was used to assess statistical significance using Prism software. Experiments were repeated independently at least two times with similar results. Dot plots depict biological replicates unless otherwise stated. Data are represented as means ± SEM; n describes the number of biological replicates.


Fig. S1. Cutaneous cytokine expression and immune infiltrate follow hair eruption in EGFRΔep mice.

Fig. S2. Hair eruption initiates barrier defect and inflammation.

Fig. S3. The bacterial composition affects the phenotype of EGFRΔep and of erlotinib-treated mice.

Fig. S4. Transgenic K5-SOS expression reduces inflammation in EGFRΔep mice.

Fig. S5. The atopic signatures of EGFRΔep mice can be distinguished in bacteria-dependent and bacteria-independent settings.

Fig. S6. Patients with SCC treated with cetuximab display reduced filaggrin expression, mixed immune infiltrate, and bacterial colonization.

Fig. S7. Prophylactic, but not reactive, FGF7 prevents inflammation in EGFRΔep mice without inducing papilloma formation.

Table S1. Resources.

Data file S1. Original data.


Acknowledgments: We are grateful to D. Gubi, E. Berger, M. Jarguz, and A. Hladik for the excellent technical assistance. We thank M. Holcmann for the scientific input and critical reading of the manuscript. The Core Facility Genomics at the Medical University of Vienna is acknowledged for carrying out the RNAseq analysis and the Joint Microbiome Facility for the microbiome analysis. We are grateful to M. Hammer and the staff of the Department of Biomedical Research of the Medical University of Vienna for maintaining our mouse colonies. Funding: This work was supported by grants from the Austrian Science Fund (FWF, PhD program W1212 “Inflammation and Immunity” to M.S., P27129 to T.B., I2320-B22 to A.L., and P31113-B30 to P.S.), the Ministry of Health of the Czech Republic (15-30782A to D. Srutkova), and the Marie Skłodowska-Curie Individual Fellowship of the European Commission (H2020-MSCA-IF-2014 655153 to P.S.). M.S.’s research is funded by the WWTF and a European Research Council (ERC) grant (ERC-2015-AdG TNT-Tumors 694883). D. Schulz was supported by the Forschungskredit of the University of Zurich grant no. FK-17-115. B.B.’s research is funded by an SNSF Assistant Professorship grant and by the ERC under the European Union’s Seventh Framework Program (FP/2007-2013)/ERC Grant Agreement no. 336921. Author contributions: J.K. and T.B. designed the experiments, conducted the experiments, and wrote the paper. B.H., C.H., P.S., B.L., D. Srutkova, D. Schulz, and T.M. conducted the experiments. I.V., K.R., B.B., H.K., S.K., and A.L. designed and supervised the experiments. M.S. and T.B. conceived and supervised the whole project, wrote the paper, and provided the funding. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The RNAseq data for this study have been deposited in the Gene Expression Omnibus (GEO) database (GSE119376). 16S rRNA sequencing data have been deposited into BioProject PRJNA48899 under SRA accession SRP159613. All data associated with this study are present in the paper or the Supplementary Materials. Original data are provided in data file S1.

Stay Connected to Science Translational Medicine

Navigate This Article