Results

Cohort characteristics and clinical response stratification

75 patients were included in this observational study. Of those, 25 patients were not fully analyzed as summarized in Figure 1A.
The final dataset consisted of 50 AD patients (mean age 48.8 years) observed in two time-points (before and 6 months after introducing dupilumab therapy) as well as 39 healthy individuals (mean age 27.2 years) and 15 psoriasis patients serving as controls (mean age 53.9 years). The sex distribution was comparable between the groups (Figure 1B-D).
Three groups of AD patients based on their response to dupilumab were built: 1) low responders if their SCORAD or body surface area (BSA) was reduced by less than 75% from baseline, 2) high responders if the SCORAD or BSA was reduced by more than 75% but less than 90%, 3) super responders with a 90% SCORAD reduction or involvement of only ≤ 2% of total BSA. (Figure 1A, F). 15 (30%) patients were considered low responders, 21 (42%) high responders, and 14 (28%) super responders.
The responder status was associated with significant differences in SCORAD. Median SCORAD values after dupilumab therapy for low, high, and super responders were 39.4 (IQR 14.9), 17.9 (IQR 14.5), and 3.85 (IQR 4.875), respectively. The SCORAD decreased proportionally in each group and were 73% ± 34%, 30% ± 14%, and 11% ± 12% of their initial levels in each respective group (Figure 1F). Importantly, there were no differences in SCORAD values before initiating systemic therapy in the respective groups (median SCORAD = 55.9, Kruskal-Wallis rank sum test 3.06, p-value = 0.22, Figure S1).
A similar trend was seen in pruritus, measured on a visual analog scale (VAS, range from 0 to 10). Initially, there was no difference in median pruritus score between groups (low responders 7.6 (IQR 1.5); high responders 6.1 (IQR 3.3); super responders 6.8 (IQR 1.75), Kruskal-Wallis test p = 0.19). The values of pruritus decreased significantly after 6 months of therapy with dupilumab and were 2.3 (IQR 5.2); 1.6 (IQR 1.7); 0.65 (IQR 0.875) in each respective group (Figure 1F).
BSA (0 - 100%) decreased from 34.87 ± 25.61 in low responders; 48.62 ± 25.16 in high responders; 32.86 ± 21.1 in super responders to become 23.93 ± 15.38; 7.88 ± 4.98; 1 ± 0.88 in each respective group (Figure 1F).
Dermatology Life Quality Index (DLQI, ranging 0 - 30 points) followed the same trend (Figure 1F).
Taken together, our results indicate that the groups were comparable regarding their initial severity of AD, and showed significant changes in response to the systemic therapy.

Candidate biomarker identification through screening

We initially performed a proteomic screening with 440 proteins using a microarray and analyzed samples from AD patients. We identified 27 proteomic candidates (|Hedges G| > 0.9, Figure 2A-B). To determine the specificity of the proteomic markers, samples from psoriasis patients and healthy individuals were also studied.
To assess the importance of miRNA as biomarkers, we performed screening from 4 AD patients before treatment, 6 months after therapy, and as controls, from 6 healthy individuals. When unsupervised clustering was performed, AD samples clustered separately from healthy individuals as shown in the dendrogram (Figure 2C).
The analysis of the miRNA patterns before and upon dupilumab therapy largely overlapped, while a clear separation from healthy individuals was determined (Figure 2D). Based on the differential expression analysis (with the DESeq2 package for R17), we selected the following miRNAs as candidate biomarkers for further study: hsa-miR-29a-3p, hsa-miR-25-3p, and hsa-miR-378a-3p (Figure 2E). We also included hsa-miR-451a, based on a literature search18, for further validation by RT-qPCR on the sera of AD and psoriasis patients as well as healthy individuals.
Patients suffering from AD have shown higher colonization rates withS. aureus in lesional, but also non-lesional skin19. NGS data on 7 AD samples and 7 healthy individuals revealed a high abundance of C. acnes , S. aureus , and S. epidermidis . Although the relative abundance ofC. acnes and S. epidermidis was comparable among the samples, S. aureus colonization was significantly increased in AD patients. Therefore, we restricted our further analysis to these three main actors and relativized the amount of S. aureus to stable members of the healthy skin microbiota (C. acnes and S. epidermidis ).

Serum proteomic profiles and their course upon Th2 targeted treatment

To validate the screening results, we measured a panel of dysregulated proteins on the entire cohort and performed correlations with the extent of the clinical response upon treatment.
CCL17 (one of the best-described biomarkers of AD) was significantly decreased after 6 months of treatment regardless of the responder status. We also observed a decrease in the chemokines CCL13, CCL22, CCL27, and E-Selectin and an increase in BDNF upon dupilumab treatment (Figure 3A, D, Figure S2A, E, F, I).
To verify if the identified biomarkers reflect the clinical response, we stratified patients according to their treatment outcomes. In low responders, BDNF and ADAM8 increased after therapy, and no alteration was observed in well-responding patients (Figure S2C, E). By contrast, CCL22 and CCL13 did not change significantly in low responders but decreased in high and super responders (Figure S2A and I).
Individual biomarker patterns among the responders were observed before initiating systemic treatment. Super responders had higher levels of Notch1, CD25s, IL11, and lower levels of FGF1 when compared to high responders, (Figure S3).
We observed differences in expression levels of several protein biomarkers in serum between AD and psoriasis patients as well as healthy individuals. BDNF, CCL13, CD25s, CCL17 and E-selectin were exclusively dysregulated in AD patients, when compared to healthy, but also to psoriasis patients (3A, D, I, Figure S2E, I). In addition, CCL22 and CFD were less expressed in healthy individuals compared to AD patients (Figure S2A, D). ADAM8, CD40L, IL22 were lower in psoriasis patients (Figure S2C, G, J).
In summary, CCL17, CCL13, and E-selectin correlated positively with SCORAD, pruritus, and BSA. By contrast, BDNF levels correlated negatively with BSA in AD patients (Figure 3 and Figure S5), indicating their usefulness as a severity-oriented biomarker panel.

Serum miRNA pattern in AD and their alteration upon Th2 targeted therapy

Interestingly, we observed significantly lower expression of all investigated miRNA before therapy when compared to healthy individuals. After therapy with dupilumab, this difference was less prominent in hsa-miR-29a-3p, hsa-miR-25-3p, and hsa-miR-378a-3p. We did not detect any significant differences in the measured miRNA from AD patients before and after therapy nor between AD and psoriasis patients (Figure 4). These results suggest that differences in miRNA profile are rather reflecting the disease as such, than its severity.

Skin microbial composition in AD patients and its alteration upon systemic Th2 therapy

The relative abundance of selected bacteria was determined by RT-qPCR and assessed in relation to clinical symptoms. We observed a relatively stable abundance of S. epidermidis throughout the course of therapy (Figure 5D). On the other hand, the ratio of S. aureus toS. epidermidis decreased significantly after systemic therapy. A similar finding was seen in the ratio of S. aureus to C. acnes (Figure 5A). Importantly, the ratio of C. acnes toS. epidermidis remained constant during the observed period (Figure 5A). These changes in bacterial DNA ratio are dependent on the overall decrease in the amount of measured S. aureus DNA and a slight increase in C. acnes proportion after initiating systemic therapy. The ratio of the measured bacterial DNA of S. aureus toC. acnes correlated to SCORAD and BSA indicating its association with the clinical status (Figure 5E). Significant differences were also seen in the ratio of these bacteria in the IGA score, with the highest values in S. aureus to C. acnes ratio observed in grade 4 and lowest in grades 0-1 (Figure 5B). There were no significant differences in the bacterial ratio values before the initiation of dupilumab between the low, high, and super responders (Figure 5C), suggesting that the baseline bacterial skin composition does not seem to associate with treatment outcome in this setting.

Integrative analysis of biomarker composites in good responders

Next, we performed biomarker pattern analysis concerning the treatment effects on severity. Principal component analysis was performed on the whole cohort with the most informative biomarkers. The groups showed a large overlap with the highest differences observed between AD patients before therapy and healthy individuals. AD patients (after therapy) resembled healthy individuals more closely, while psoriasis patients presented in between (Figure 6A). The largest differences among the groups were depicted in Figure 6B.
We investigated the baseline biomarker profiles of AD patients concerning their treatment response. We observed that CCL17, E-selectin, CD25s, and Notch1 consistently changed in all responder groups i.e., they individually showed a consistent pattern in high and super responders (either increasing or decreasing), but the size of the effect was limited. As Notch1, CD25s, IL22, FGF1, CCL27, and CCL17 were strongly correlated they could not be used for further predictive classification modeling (Figure S4A-B).
To increase biomarker sensitivity and to prevent highly correlated variables from distorting random forest accuracy, we subsequently calculated biomarker ratios to one another to form composite biomarkers and evaluated their fitness to predict response to dupilumab (low vs. high and super responders). We observed that the baseline values of Notch1 to CD25s ratio and CCL17 to E-selectin ratio were the best-performing predictors of low response to dupilumab after 6 months of therapy (Figure 6C-D). Subsequently, we analyzed the predictive capability of these two promising composite biomarkers in a random forest classifier. The area under the curve of this prediction model was 0.72 indicating higher than the random probability to correctly predict therapy outcome based on the serum levels of four proteins before initiating systemic therapy (Figure 6E).
As the skin microbial composition before treatment did not show significant differences among the responder groups nor IGA response scores, it did not present predictive capabilities regarding dupilumab therapy outcomes in atopic dermatitis patients after 6 months (data not shown).