Methods

Study design and sample collection

We included adult patients (> 18 years) with the diagnosis of atopic dermatitis based on Hanifin and Rajka criteria made at least 12 months prior to the inclusion, who were indicated for systemic therapy with dupilumab 300 mg s.c. The Ethical Committee of the Charité Universitätsmedizin Berlin approved this non-interventional study (EA1/112/19).
After obtaining informed consent, we gathered data on the clinical presentation of AD, current symptoms before the initiation of the therapy, and six months afterward, along with patient reported outcomes.
In addition, serum, and inter-scapular skin swabs from patients with moderate to severe atopic dermatitis, before and 6 months after initiating systemic therapy, were acquired. The microbiome sampling location was chosen based on the relative stability of microbiota in this anatomic location, precise and straightforward identification of the sampling area, and sufficient surface for biomaterial collection. Serum was prepared as described previously14.

Serum proteomic screening

Analysis of 440 proteins in sera of patients with moderate to severe atopic dermatitis before systemic therapy and six months after initiating dupilumab was done using Quantibody Human Cytokine Array Q440 chip (RayBiotech, GA, USA). Fluorescent protein arrays were scanned using a PowerScanner (Tecan Group AG). Array microphotographs were quantified using Protein Array Analyzer for ImageJ15.

Protein biomarker measurement

Human serum samples were analyzed using ELISA kits provided by R&D Systems, Minneapolis, USA (human CCL17, DY364; human CCL13, DY327; human CCL27, DY376; CCL22, DY336; IL22, DY782; IL11, DY218; CD40L, DY617; E-selectin, DY724; BDNF, DY248; Notch1, DY5317; CD25s, DY223; ADAM8, DY1031; FGF1, DY232; CFD, DY1824 human), following the manufacturer’s protocol.

Serum miRNA extraction, profiling and validation

Serum miRNA isolation, library preparation and sequencing were done following manufacturers instructions and are described in detail in the supplement. The differentially expressed miRNAs were further validated using reverse transcription quantitative real-time PCR (RT-qPCR) as described before14 and in detail in the supplement.

Quantification of selected skin microbiota

We used the data from next generation sequencing (NGS) platform Illumina to identify the main bacterial species informative of the patient’s skin status. Based on the results from our previously published paper13, we decided to further investigate 3 bacterial species in detail, using qPCR analysis.
Cutibacterium acnes (ATCC 6919), Staphylococcus epidermidis (ATCC 12228) and Staphylococcus aureus (ATCC 29213) cultures were grown as described elsewhere. Skin microbiota were quantified using RT-qPCR as described in the supplement.

Data availability

The sequencing data presented in this study were be deposited in the European Nucleotide Archive (ENA) under the accession number: PRJEB59318.

Classification model

The supervised machine learning classification model was performed using a random forest algorithm with the help of the caret package16. All samples with complete observations (without missing values) were included in the training set and divided into good (super and high responders) and low responders groups. Features for the final random forest model were selected based on significant differences in each biomarker. The model fitness was calculated using 10 sets of 10-fold repeated cross-validation.

Statistics

Mann Whitney U test was used for comparing values between unpaired observations, with Holm’s p-value correction for multiple comparisons. Paired data were analyzed using paired Wilcoxon’s test with Holm’s correction where appropriate. Next-generation sequencing-derived data were analyzed using the Wald test with Benjamini-Hochberg FDR to correct for multiple comparisons. P values < 0.05 were considered significant.