The atopic dermatitis blood signature is characterized by increases in inflammatory and cardiovascular risk proteins

Beyond classic “allergic”/atopic comorbidities, atopic dermatitis (AD) emerges as systemic disease with increased cardiovascular risk. To better define serum inflammatory and cardiovascular risk proteins, we used an OLINK high-throughput proteomic assay to analyze moderate-to-severe AD (n = 59) compared to psoriasis (n = 22) and healthy controls (n = 18). Compared to controls, 10 proteins were increased in serum of both diseases, including Th1 (IFN-γ, CXCL9, TNF-β) and Th17 (CCL20) markers. 48 proteins each were uniquely upregulated in AD and psoriasis. Consistent with skin expression, AD serum showed up-regulation of Th2 (IL-13, CCL17, eotaxin-1/CCL11, CCL13, CCL4, IL-10), Th1 (CXCL10, CXCL11) and Th1/Th17/Th22 (IL-12/IL-23p40) responses. Surprisingly, some markers of atherosclerosis (fractalkine/CX3CL1, CCL8, M-CSF, HGF), T-cell development/activation (CD40L, IL-7, CCL25, IL-2RB, IL-15RA, CD6) and angiogenesis (VEGF-A) were significantly increased only in AD. Multiple inflammatory pathways showed stronger enrichment in AD than psoriasis. Several atherosclerosis mediators in serum (e.g. E-selectin, PI3/elafin, CCL7, IL-16) correlated with SCORAD, but not BMI. Also, AD inflammatory mediators (e.g. MMP12, IL-12/IL-23p40, CXCL9, CCL22, PI3/Elafin) correlated between blood and lesional as well as non-lesional skin. Overall, the AD blood signature was largely different compared to psoriasis, with dysregulation of inflammatory and cardiovascular risk markers, strongly supporting its systemic nature beyond atopic/allergic association.


Results
The proteomic blood signature of AD is largely different from psoriasis. Using the OLINK high-throughput proteomic platform, we assessed a panel of 257 immunological and cardiovascular risk proteins in serum of moderate-to-severe AD and psoriasis patients in comparison to healthy control subjects ( Table 1). The proximity extension technology used in our study is potentially superior to conventional multiplex immunoassays, since only correctly matched antibody pairs give a signal, yielding higher specificity and sensitivity 40,41 . The body-mass-index (BMI) was similar in AD and controls, while higher in psoriasis patients. Among the 257 investigated proteins, only 11 were significantly upregulated in both diseases when compared to controls (Fig. 1a). The majority of markers were exclusively upregulated in either AD (n = 45) or psoriasis (n = 53), and only a minority were downregulated (Fig. 1a, Supplementary Table S1). When results were adjusted for BMI (Fig. 1b, Supplementary Table S2) and for the presence of asthma and the cardiovascular risk factors such as arterial hypertension, diabetes mellitus, and hypercholesterolemia (Supplementary Figure S1, Supplementary  Table S3), the observed differences between AD and psoriasis, compared to controls, were largely preserved. Mutually upregulated proteins included markers of Th1 (IFN-γ, CXCL9, TNF-β/lymphotoxin) and Th17 (CCL20, IL-17C) immune responses, the CD4 T-cell chemoattractant IL-16, and the differentiation/proliferation factor IL-20 (Fig. 1c).
Correlation with skin disease suggests skin-blood-interaction. To investigate whether the extent of AD skin disease influences levels of blood markers, we correlated serum proteins with the clinical AD measures SCORAD (SCORing Atopic Dermatitis, Fig. 2a) and body surface area (BSA) involvement (Fig. 2b). The endothelial cell activation marker E-selectin, the Th17 marker PI3/Elafin, the CCR1/CCR2/CCR3 chemokine CCL7, and the CD4 + cell chemoattractant IL-16 showed the most significant correlations for both SCORAD and BSA (Fig. 2a,b). Those serum markers were only correlated with skin scores, but not with body-mass index (BMI) (Fig. 2c-j). Leptin, a measure of total body fat, only correlated with BMI, but not with SCORAD or BSA ( Fig. 2k-l), as expected. A complete list of correlation coefficients can be found in Supplementary Table S4.
Mutual regulation of blood and skin markers. We next compared blood protein levels in AD with a robust meta-analysis derived atopic dermatitis (MADAD) skin transcriptome that integrates several published cohorts 47 . Markers that were upregulated in both the MADAD transcriptome and the blood were mostly chemokines, as depicted in Fig. 3.
Correlations of blood inflammatory mediators were found with both lesional and non-lesional AD skin. As mostly chemokines showed concomitant regulation between lesional skin and serum of AD patients (Fig. 3), we further compared levels of immune markers in skin and blood of our patients. We quantified a panel of inflammatory molecules in respective lesional and non-lesional biopsies of our AD cohort using RT-PCR, and correlated them with serum OLINK levels (Fig. 4, Supplementary Figure S2). We found significant positive correlations for the general inflammation marker MMP12, the Th1/Th17 cytokine IL-12/IL-23p40, as well as Th1 (CXCL9), Th2 (CCL22) and Th17 (PI3/Elafin) markers (Fig. 4). Notably, these correlations were consistently found not only with lesional, but also with non-lesional skin (Fig. 4). IL-17A in skin was only upregulated in a subset of patients (Supplementary Figure S2i-j). When assessing skin with detectable IL-17, significant correlations of lesional (r = 0.53, p = 0.004), but not non-lesional skin were obtained with blood (r = 0.28, p = 0.29, data not shown).

Multiple inflammatory pathways show a stronger enrichment in AD than in psoriasis.
We also conducted an enrichment analysis by using various published pathways (Kyoto Encyclopedia of Genes and Genomes/KEGG, Reactome Pathway Database, BioCarta, Pathway Interaction Database/PID) to compare AD and psoriasis blood proteomic profiles (Supplementary Table S5) [48][49][50][51][52] . Selected pathways significantly enriched in the serum of AD and/or psoriasis (in comparison to controls) are depicted in

Discussion
While several studies have described a limited set of biomarkers being increased in the blood of AD patients [31][32][33] , the current study is the first to investigate a broad array of immune and cardiovascular risk proteins in serum of moderate-to-severe AD patients, compared to psoriasis and controls. A large bulk of evidence in psoriasis suggests an increase in cardiovascular risk factors and associated cardiovascular comorbidity [53][54][55] , with similar data only recently emerging for AD 5,39 . The concept of increased cardiovascular risk and disease is supported by several epidemiological cohort studies 10,18,19,56 , case-control studies 13,15,16 and population based surveys 12 . Using coronary computed tomography angiography, AD patients showed increases in coronary artery disease compared to healthy controls 16 . However, some other AD studies showed only marginal or no increased cardiovascular manifestations 11,14,57 , adding to the controversy as to whether AD is an independent risk factor for cardiovascular disease, mandating further investigation.
Recent studies in blood of severe AD patients showed significantly increased T-cell activation 25,30 and increases in serum cytokines 31,33 that correlated with clinical severity. Our study is the first to define inflammatory and cardiovascular risk proteins that are commonly upregulated in AD and psoriasis. These markers are largely associated with Th1 (IFN-γ, CXCL9, TNF-β, lymphotoxin-α), but also Th17 (CCL20) responses. Consistently, these immune axes are also upregulated in chronic skin lesions of both diseases 58 . IL-20 is involved in epidermal hyperplasia and inhibits keratinocyte differentiation 59 , but is also expressed in atherosclerotic plaques, and was shown to promote atherosclerosis in a mouse model 60 . IL-16, a chemo-attractant for CD4 + T-helper cells and myeloid cells, is also expressed in atherosclerotic plaques, but may have a plaque-stabilizing effect 61,62 . These factors, together with an environment of chronic/Th1-triggered inflammation 4,63 , with various degrees of Th17 activation 26,64,65 , suggest a potentially pro-atherogenic milieu in the blood of both psoriasis and AD patients that may directly impact endothelial cells 66 .
While we observed increases of several specific inflammatory and cardiovascular risk proteins only in psoriasis, including markers of coagulation, angiogenesis, endothelial activation and lipid metabolism, consistent with its well-established systemic inflammatory nature 67-69 , we also observed a large set of markers exclusively upregulated in AD. Consistent with skin data 4,5,58,70 , we found increases of Th1 (CXCL10, CXCL11), Th2 (IL-13, CCL13, CCL17, CCL11, IL-10), Th17/Th22 (S100A12) and Th1/Th17/Th22 (IL-12/IL-23p40) associated products in serum of AD patients. The fact that blood levels of the inflammatory marker MMP12 and several mediators from all T-helper-cell axes (CXCL9-Th1, CCL22-Th2, PI3-Th17, IL-12p40-Th1/Th17/Th22) correlated not only with lesional, but also with non-lesional skin, point to the critical role of systemic inflammation in immune abnormalities that AD already harbors in non-lesional skin. This systemic inflammation consists of a strong adaptive component, evidenced by increases in multiple factors involved in T-cell development and activation, such as CD40L, IL-7, CCL25, IL-2RB, IL-15RA, and CD6. IL-2RB is of special interest as it is part of the high-affinity IL-2 receptor, which is involved in transduction of mitogenic signals from IL-2.
IL-17C, produced by keratinocytes, endothelial cells, and leukocytes in the skin, can induce anti-microbial peptides in synergy with IL-22, TNF-α, and IL-1β 71 . In atherosclerotic plaques, smooth muscle cell-derived IL-17C plays a pro-atherogenic role by supporting recruitment of Th17 cells 72 . IL-17C and TNF-β, elevated in AD, have overlapping signaling pathways with IL-17A/F and TNF-α, which are thought to contribute to atherosclerosis in psoriasis 21,23 . Thus, synergistic effects on endothelial cells 73,74 and other cell types need to be considered. TNFSF14/LIGHT, a pro-inflammatory cytokine associated with atherosclerosis 40 , plays crucial roles in T-cell homing into inflamed tissues 75 , and in the induction of matrix metalloproteinases/MMPs in macrophages 76,77 . Several MMPs, which are involved in tissue remodeling including atherosclerosis 78 , were also increased in AD serum (MMP-1, MMP-12, MMP-10).
When integrating many of these markers by using established lists of inflammatory pathways, we found enrichment of multiple pathways in AD to a much higher degree than in psoriasis. Enriched pathways included those also triggered by other atopic conditions, and particularly asthma (i.e. IL-4 immune signaling), also evidenced by the efficacy of specific Th2 targeting-strategies in both AD and asthma 79,80 . Many more immune pathways showed stronger enrichment in AD compared to psoriasis (e.g. cytokine-cytokine receptor interaction, chemokine signaling pathway, cytokines and inflammatory response, dendritic cell pathway, Th1/Th2 differentiation). These findings together with our past flow cytometry studies 30 , support a stronger systemic inflammation in AD compared to psoriasis. Atherosclerosis is mediated by local inflammatory mediators including chemokines and their receptors, that are involved in the recruitment of inflammatory cells to the intima as an essential step in plaque development 81 . Such mediators were numerously increased in our AD cohort, including CCL4, CCL17, CCL28, CXCL5, CXCL10, and CX3CL1/fractalkine. CX3CL1/fractalkine is produced by endothelial cells, and is a strong chemoattractant for monocytes and lymphocytes, mediating their extravasation 82 . CCL4 and its receptor CCR5 have recently been demonstrated to play diverse roles in the inflammatory events underlying cardiovascular diseases and diabetes mellitus 83 . CXCL5 is increased in atherosclerosis, mediating a protective role in a mouse model by modulating macrophage activation 84 . CCL28 is chemotactic to T-cells, B-cells, and eosinophils to mucosal effector sites, and is increased in asthma 85 . CCL17 has been shown to drive atherosclerosis by restraining regulatory T-cell homeostasis 86 , and CXCL10 is associated with the severity of coronary artery disease 87 .
Several growth factors associated with atherosclerosis were also increased, such as the vascular growth factor VEGF-A 88 . Hepatocyte growth factor/HGF, produced by mesenchymal cells, is a biomarker of macroangiopathy 89 , and circulating HGF levels have been positively associated with stroke 90 . CD137, a co-stimulatory molecule expressed on activated T-cells, B-cells, DCs, as well as endothelial cells 91 , increases atherosclerosis in an ApoE(-/-) mouse model via leukocyte recruitment and inflammation 92 .
Taken together, we found multiple factors to be uniquely increased in AD that might be contributors of a pro-atherogenic burden in this disease. In line with recent publications 10, 15,18,19 showing increases in BMI in North American and Asian children and adults with AD, our cohort was also overweight. However, inflammatory mediators involved in atherosclerosis development (E-selectin, CCL7, IL16, PI3/elafin) were significantly correlated with AD severity/SCORAD, but not with BMI, strongly suggesting the contribution of cutaneous disease to cardiovascular morbidity.
Also, while associations with cardiovascular outcomes were reported in US and Asian studies 10,15,93 , only small increases in angina pectoris, arterial hypertension and peripheral arterial disease risks were found in a German cohort 14 . This European cohort also did not show increases in genetic risk factors for cardiovascular disease in AD 14 . One might speculate that varying degrees of cardiovascular disease across cohorts results from varying decades of chronic disease rather than due to shared genetic risks, but this assumption needs verification in the cohorts with robust increases in cardiovascular risk.
Our study poses several limitations. The subjects investigated (both healthy and AD populations) were overweight, potentially contributing to increases in inflammatory markers, and profiles might be different in a lean population. Another limitation is that while AD and healthy subjects were matched for BMI and age, psoriasis patients had higher BMIs, although our data were corrected for BMI (and also for other cardiovascular risk factors such as asthma, arterial hypertension, hypercholesterolemia, and diabetes mellitus). Larger future studies should be performed that will also include lean patients and control populations. In sum, we have characterized a blood AD signature that is profoundly different from psoriasis. This profile helps to better understand cardiovascular risk in AD, and might also aid in identifying biomarkers to monitor therapeutic responses. Targeted therapeutic blockade of specific immune axes, e.g. Th17/IL-23 in psoriasis or Th2 in AD, is needed to assess the contribution of polar cytokine activation to overall systemic inflammation, and its effect on cardiovascular comorbidity and biomarkers. These studies should also assess whether biomarkers are modifiable risk factors responsive to treatment, as suggested by their decline with cyclosporine A treatment in severe chronic AD 33 .

Methods
Patients and samples. A cohort of 59 patients with moderate-to-severe AD (31 male and 28 female patients; mean age 40.5 years, range 18-72 years; mean SCORAD 54.1, range 34.5-89; mean blood eosinophils 6.91%, SD 4.94; median total serum IgE 2,412kU/L, IQR 5,028) was included in this study (Table 1, Supplementary Table S6). 20 patients (33.9%) reported to have mild asthma, of which 11 (18.6%) were on asthma treatment (inhaler); 9 patients (16.1%) reported seasonal allergies. 11 AD patients suffered from one or more of the following cardiovascular risk factors: Arterial hypertension (n = 8), diabetes mellitus (n = 4), or hypercholesterolemia (n = 7). Serum samples (n = 59) with corresponding lesional (n = 58) and nonlesional (n = 53) skin punch biopsies were collected. All methods were carried out in accordance with relevant guidelines and regulations. All experimental protocols were approved by the local institutional review boards (IRB of The Rockefeller University and the Icahn School of Medicine at Mount Sinai, both New York, NY). All patients gave written informed consent prior to inclusion. Lesional biopsies were obtained from chronic lesions, and non-lesional biopsies were taken from uninvolved skin in proximity but at least 4 cm away from lesional samples. 22 serum samples from moderate-to-severe psoriasis patients (18 male and 4 female patients; mean age 46.8 years, range 20-67 years; mean PASI 27.6, range 11.5-45.9) and 18 from healthy control subjects (12 male and 6 female subjects; mean age 41.3 years, 24-55  (Table 1, Supplementary Table S6). Washout periods prior to study inclusion were 4 weeks for systemic therapy (cyclosporine, oral steroids, azathioprine, mycophenolate mofetil and all other systemic immunosuppressants) and 2 weeks for phototherapy and topical corticosteroids for both psoriasis and atopic dermatitis. None of the subjects had any defined history of cardiovascular disease. OLINK multiplex assay. Serum samples were collected, centrifuged, and stored at −80 °C until further processing. Aliquots were analyzed with an OLINK Proseek ® multiplex assay 40,94 , a proximity extension assay (PEA) technology with oligonucleotide-labeled antibody probe pairs that bind to their respective targets 41 . Upon binding of antibody pairs to their respective targets, DNA reporter molecules bound to the antibodies gave rise to new DNA amplicons with each ID-barcoding their respective antigens. The amplicons were subsequently quantified using a Fluidigm BioMark TM HD real-time PCR platform 40 . Serum was analyzed using Inflammation I, cardiovascular disease/CVD II, and CVD III multiplex panels, which contain a broad array of established and exploratory markers 40 . OLINK data per subject are given in Supplementary Table S6. Skin RT-PCR. Real time-PCR was performed on AD-related genes as described previously 95,96 . Results were normalized to the housekeeping gene hARP and log2-transformed for analysis. Primers and probes used are listed in Supplementary Table S7.
Statistical analysis. Statistical analysis was carried out using R-language (R-project.org) and packages available through the Bioconductor Project (www.bioconductor.org).
Quantification of Protein levels. Quality control of OLINK chip data was carried out using their standard quality control pipeline (QC) 40,94 . A minor number of samples in each panel were excluded after this quality control procedure. A small batch effect (explaining 8.4% of the variance) was observed corresponding to the processing times. This plate effect was estimated through a linear model using a set of intra-plate control samples that were repeated in each plate. We corroborate that the batch was successfully adjusted for the set of 7 intra-plate samples (showed in duplicates in Supplementary Table S6) presented in the statistical analysis for this paper.
Analysis of Protein profiles. Protein expression profiles were modeled using linear models for high-throughput data on R's limma framework. The model included Disease as a factor and covariates Age, Gender and BMI. Comparison between groups were estimated using an empirical Bayesian method 97 available in the limma package; this uses the variance across all genes to estimate per gene variance. After model estimation using residual maximum likelihood algorithms, hypothesis testing was conducted for comparisons of interest using contrasts under the general framework for linear models in the limma package. P-values from the moderated t-test were adjusted for multiple hypotheses using the Benjamini-Hochberg procedure, which controls the FDR (false discovery rate). A sensitivity analysis including asthma and cardiovascular (CVD) risk outcomes as a covariate was also carried out. Covariates with missing values were imputed as the mean (age, BMI) and as "not present" for binary comorbidities. Sensitivity analysis showed no departure from the attained conclusions due to imputation. Comparison of protein profiles among groups was carried out using linear models for high-throughput data on R's limma framework. Protein annotations, as well as Gene-Protein relationships were obtained by using UniPro IDs and R's AnnotationDbi package.
RT-PCR data. Ct values were derived by normalizing Ct values to an endogenous control gene (hARP). Values under the limit of detection (LOD) were substituted by the 20% of the minimum value above the LOD. Data was log2-tranformed prior to analysis.
Correlation between skin mRNA and protein profiles was evaluated using Pearson and Spearman correlation coefficients on log2-transformed levels. Data is presented in scatterplots with estimated linear regression and 95% confidence interval.
Pathway Enrichment Analysis. Gene set over-representation analysis was performed using XGR software 98 based on functional category including KEGG 99 , BioCarta 100 , REACTOME 101 , PID 51 and MSigDB 102 . The significance of the overlaps was assessed using FDR < 0.05. Data availability. All data generated or analyzed during this study are included in this published article (and its Supplementary Information Files).