Understanding allergic multimorbidity within the non-eosinophilic interactome

Abstract

Background

The mechanisms explaining multimorbidity between asthma, dermatitis and rhinitis (allergic multimorbidity) are not well known. We investigated these mechanisms and their specificity in distinct cell types by means of an interactome-based analysis of expression data.

Methods

Genes associated to the diseases were identified using data mining approaches, and their multimorbidity mechanisms in distinct cell types were characterized by means of an in silico analysis of the topology of the human interactome.

Results

We characterized specific pathomechanisms for multimorbidities between asthma, dermatitis and rhinitis for distinct emergent non-eosinophilic cell types. We observed differential roles for cytokine signaling, TLR-mediated signaling and metabolic pathways for multimorbidities across distinct cell types. Furthermore, we also identified individual genes potentially associated to multimorbidity mechanisms.

Conclusions

Our results support the existence of differentiated multimorbidity mechanisms between asthma, dermatitis and rhinitis at cell type level, as well as mechanisms common to distinct cell types. These results will help understanding the biology underlying allergic multimorbidity, assisting in the design of new clinical studies.

About the Authors:

Daniel Aguilar

Roles Conceptualization, Investigation, Methodology, Software, Supervision, Writing – original draft, Writing – review & editing

* E-mail: [email protected]

Affiliations Biomedical Research Networking Center in Hepatic and Digestive Diseases (CIBEREHD), Instituto de Salud Carlos III, Barcelona, Spain, ISGlobal, Barcelona Institute for Global Health, Barcelona, Spain, 6AM Data Mining, Barcelona, Spain

ORCID logo http://orcid.org/0000-0001-7520-4508

Nathanael Lemonnier

Roles Methodology, Writing – review & editing

Affiliation: Institute for Advanced Biosciences, Inserm U 1209 CNRS UMR 5309 Université Grenoble Alpes, Site Santé, Allée des Alpes, La Tronche, France

ORCID logo http://orcid.org/0000-0002-3994-8697

Gerard H. Koppelman

Roles Writing – review & editing

Affiliations University of Groningen, University Medical Center Groningen, Beatrix Children’s Hospital, Department of Pediatric Pulmonology and Pediatric Allergology, Groningen, Netherlands, University of Groningen, University Medical Center Groningen, GRIAC Research Institute

Erik Melén

Roles Writing – review & editing

Affiliation: Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden

Baldo Oliva

Roles Methodology, Writing – review & editing

Affiliation: Structural Bioinformatics Group, Research Programme on Biomedical Informatics, Department of Experimental and Health Sciences, Universitat Pompeu Fabra, Barcelona, Spain

Mariona Pinart

Roles Writing – review & editing

Affiliation: ISGlobal, Barcelona Institute for Global Health, Barcelona, Spain

Stefano Guerra

Roles Conceptualization, Supervision, Writing – review & editing

Affiliations ISGlobal, Barcelona Institute for Global Health, Barcelona, Spain, Asthma and Airway Disease Research Center, University of Arizona, Tucson, Arizona, United States of America

Jean Bousquet

Roles Supervision, Writing – review & editing

Affiliations Hopital Arnaud de Villeneuve University Hospital, Montpellier, France, Charité, Universitätsmedizin Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Comprehensive Allergy Center, Department of Dermatology and Allergy, Berlin, Germany

Josep M. Anto

Roles Conceptualization, Funding acquisition, Supervision, Writing – review & editing

Affiliation: ISGlobal, Barcelona Institute for Global Health, Barcelona, Spain

Introduction

Mapping diseases onto molecular interaction networks (such as the protein-protein interaction network, also known as the interactome), has contributed to the elucidation of disease mechanisms and the identification of new disease-associated genes [1, 2]. Evidence suggests that disease-associated genes are not randomly distributed within the interactome, but instead they work coordinately forming connected communities linked to disease phenotypes [1, 3–5]. Furthermore, genes expressed in a particular tissue tend to form a well-localized subnetwork, and the partition of the complete interactome into tissue-specific subnetworks has important implications for the understanding of disease mechanisms [6]. Gene activity is often dependent on tissue context, and human diseases arise from the complex interplay of tissue and cell-lineage-specific processes [7, 8]. Disease-associated genes are usually tissue-specific and their interaction patterns with other genes change in diseased tissues as compared to healthy ones [9]. These observations make elucidating the context-specific role of genes in pathophysiological processes particularly challenging [10, 11]. Exploiting tissue-specific information has provided valuable clues on tissue-specific gene functions [12].

The computational analysis of tissue-specific cellular networks helps to understand the tissue-specific mechanisms of diseases, and how those mechanisms interplay with one another. Authors have long hypothesized that perturbations of cellular networks are key to many phenotypic and pathophenotypic outcomes [1, 4, 13–16]. Because of this, co-morbid and multi-morbid phenotypes are expected to share tissue-specific causative mechanisms [12, 13]. Studies have found that multimorbidity between metabolic diseases can be explained by shared cellular mechanisms [17], and that multimorbidities do not necessarily imply that the involved diseases are linked through shared genes [16, 18–20].

In a previous work, we uncovered significant patterns of network connectivity between the cellular networks associated to asthma (A), dermatitis (D) and rhinitis (R) [21], which supported the idea that A, D and R form a multimorbidity cluster due to shared genes [22, 23] and pathomechanisms [24–26]. While eosinophils have been singled out as prominent mediators in a number of inflammatory diseases [27–30] and multimorbidities [31–34], many other cell types (e.g. macrophages, monocytes/dendritic cells, lymphocytes), are involved in complex and heterogeneous diseases such as A, D and R [35–37]. Yet, a cell-type-based interactome analysis of the allergic multimorbidity has not been reported to the best of our knowledge. In this study, we use the interactome and expression data to investigate the mechanisms of multimorbidity between A, D and R at a cell-type-specific level, focusing on emergent non-eosinophilic allergy-mediating cell types across distinct tissues. Our results provide new insights could provide valuable information to improve prevention and treatment of these diseases.

Methods

Methods are described in detail in S1 Text.

Data sources

Gene-disease associations.

We built the sets of genes associated to A, D and R by integrating data from four sources: (1) The Comparative Toxicogenomics Database [38], which provides highly reliable gene-disease associations characterized through various experimental procedures combined with a process of expert curation of the literature and other databases (e.g. OMIM [39]). (2) The DisGeNet catalog, that contains curated gene-disease associations extracted from literature [40]. (3) UniProt-derived gene-disease associations, extracted from the Involvement in disease section of the Uniprot Knowledgebase [41]. (4) The Phenotype-Genotype Integrator database, that integrates information various NCBI genomic databases with association data from the National Human Genome Research Institute GWAS Catalog [42]. This is the only data source containing solely GWAS-derived gene associations [43]. Genes associated to a disease d (any of A, D or R) will be hereinafter referred to as d-associated genes.

The interactome.

We built the functional interaction network (hereinafter called the interactome for brevity) by combining data from: (1) The Reactome Functional Interaction Network (v. 022717) [44], which includes not only protein-protein interactions but also gene expression interaction, metabolic interactions and signal transduction. (2) The STRING interaction network (v.10.5) [45].

Cell-type-specific gene expression.

Gene expression levels were obtained from the human gene expression atlas available at ArrayExpress under accession number E-MTAB-62 [46]. This is a cell-type-wide compendium of high-quality microarray-derived expression data that has been previously used in other network-based analysis of gene expression [47–49] and has been incorporated into a number of biomedical software packages [50–52]. We filtered the data to remove redundancies and samples subjected to particular treatments or environmental factors (see S1 Text). We then centered and standardized the expression level of each gene as:where Eg,c is the expression level of the gene g in cell type c, Mg is the median expression level the gene g across all cell types, and MADg is the median absolute deviation of the expression levels of gene g across all cell types. This made the expression levels comparable between genes [53, 54].

We defined a gene to be cell-type-specific if its absolute normalized expression level eg,c was at least 1.5 larger than the interquartile range (IQR) of its normalized expression across all cell types [6, 12, 55, 56]. Genes specific to a cell type c (any of our cell types of interest) will be hereinafter referred to as c-specific genes.

Cellular pathways

Cellular pathways were downloaded from Reactome database in the UniProt2Reactome format files [44]. Pathway-associated genes either without expression data or not present in the interactome were not considered. Disease-related cellular pathways (e.g. Constitutive Signaling by Aberrant PI3K in Cancer) were not considered. Reactome is a collection of pathways built in a hierarchical manner, where larger pathways are subdivided into smaller pathways with more specific functionalities. This implies a trade-off between the specificity in the representation of cellular functions and the average number of genes per pathway [57]. To minimize the overlap between pathways in order to avoid redundancies that could negatively affect our analysis [58], we calculated the pairwise overlap between pathways at distinct levels of the Reactome hierarchy using the Sorensen-Dice method [59–61]. If two pathways had an overlap of > 50% genes, the one with the lowest number of associated genes was removed from the set. We chose pathways of at depth 3 of the hierarchy because it provided a mean overlap < 1% while annotating 4,809 genes (this is 87,9% of the total genes annotated in the database, all levels considered). Genes associated to a pathway p (any of our pathways of interest) will be hereinafter referred to as p-associated genes.

Pathway annotation in our previous study of A, D and R were extracted from BioCarta [62]. There is not a perfect equivalence between cellular pathways from BioCarta and Reactome databases, so in order to compare our results to those from our previous whole-organism multimorbidity study [21] we performed an association test to identify which BioCarta pathways significantly overlapped with Reactome pathways (Fisher’s Exact test, adjusted P <0.05; S1 Table). P-values in this study were adjusted by the Benjamini-Hochberg method for false discovery (FDR) control [63].

Cell-type-specific networks

In order to generate the specific network for any cell type c, we selected all edges from the interactome connecting c-specific genes [6, 64]. Because of the interactome-based nature of our analysis, those genes not present in the interactome or not present in the expression dataset were removed from the analysis. The statistical significance of the number of d-associated genes present in each c-specific network was calculated by means of a Fisher’s Exact test (adjusted P <0.05).

Quantifying cell-type-specific multimorbidity

In order to obtain a quantitative measure of the extent to which A, D and R multimorbidity is manifested in distinct cell types, we designed an interactome-based approach (workflow in Fig 1; illustrated with an example in S1 Fig). Briefly, we scored all genes specific to a given cell type according to their connectivity (or their "closeness") to known disease-associated genes, under the rationale that the malfunction of one (or more) of the disease-associated genes is likely to perturb the function of the neighboring genes, eventually disrupting a cellular mechanism and giving rise to a diseased phenotype [5, 65–68]. In other words, we scored each gene in each cell type according to its contribution to the manifestation of A, D and R. Then, we selected the set of top-scoring genes (called S; Scd being the top-scoring genes for disease d in cell type c). Finally, for each cell type we calculated the overlap between the sets of top-scoring genes for AD, AR, DR and ADR. This overlap was called the Multimorbidity Score (MS; MSTd1,d2 being the Multimorbidity Score for diseases d1 and d2 in cell type c). The process is described in detail in S1 Text.

[Figure omitted. See PDF.]

Fig 1. Workflow for Quantifying cell-type-specific multimorbidity section.

Only multimorbidity between two diseases is shown. Numbered circles indicate the steps of in the section Quantifying cell-type-specific multimorbidity in Methods.

https://doi.org/10.1371/journal.pone.0224448.g001

Characterizing cell-type-specific multimorbidity mechanisms

After having quantitatively scored the multimorbidity between diseases in different cell types, we wished to identify the actual cellular mechanisms involved in the manifestation of the multimorbidities. To do so, we designed a method to measure the perturbation that a disease can exert over a cellular pathway in a given cell type. The starting point is the set of top-scoring genes Scd calculated in the previous section. We identified the set of cellular pathways present in cell type c, and then scored how perturbed they were by the manifestation of disease d using Scd (workflow in Fig 2; illustrated with an example in S2 Fig). This score was called the Perturbation Score (PS; PScp,d being the perturbation experimented by pathway p during the manifestation of disease d on cell type c). Under the assumption that any disease can be viewed as the product of perturbed cellular mechanisms (i.e. cellular pathways), and that multimorbidity is known to arise as those perturbed mechanisms are shared by distinct diseases [12, 13, 69, 70], we selected as candidate mechanisms for multimorbidity those pathways that were significantly perturbed in more than one disease in the same cell type. The process is described in detail in S1 Text.

[Figure omitted. See PDF.]

Fig 2. Workflow for Characterizing cell-type-specific multimorbidity mechanisms section.

Only multimorbidity between two diseases is shown. Numbered circles indicate the steps of in the section Characterizing cell-type-specific multimorbidity mechanisms in Methods.

https://doi.org/10.1371/journal.pone.0224448.g002

Identifying cell-type-specific candidates to multimorbidity

Lastly, we wished to identify individual genes that might constitute candidates to multimorbidity. In the Quantifying cell-type-specific multimorbidity section we had identified the sets of genes more susceptible to be perturbed by a disease in a cell type (Scd). We identified as multimorbidity candidates those genes simultaneously belonging to > = 2 of those sets (i.e. susceptible to be perturbed by two diseases in the same cell type) for AD, AR, DR multimorbidities, and > = 3 in the case of ADR multimorbidity. In addition, we numerically scored the contribution of each gene g to multimorbidity (MSg,cd1,d2 being the Multimorbidity Score for gene c with respect to diseases d1 and d2 in cell type c). This process is detailed in Text S1.

Results

Gene-disease associations

The number of genes associated to A, D and R with representation in the interactome and expression data was 98, 62 and 10, respectively. The complete list of genes is shown in Table 1 (see S2 Table and Gene-disease associations in the Methods section for data sources). Three genes were associated with A, D and R: IL13, platelet-activating factor acetylhydrolase PLA2G7 and LRRC32, a signal peptide cleavage essential for surface expression of a regulatory T cell surface protein. The complete list of all disease-associated genes in each cell type is provided in S2 Table.

[Figure omitted. See PDF.]

Table 1. Gene-disease associations.

https://doi.org/10.1371/journal.pone.0224448.t001

Cell-type-specific gene expression and the cell-type-specific networks

The complete interactome contained 15,332 genes (nodes) and 394,317 interactions (edges). The total number of cell types was 60, classified into 15 distinct tissues. The total number of genes with expression data was 8,461 (of which 7,486 were present in the interactome). Table 2 shows the number of genes specific to each cell-type-specific network and its statistical significance (an extended version of the table with p-values is provided as S3 Table). The number of genes present in a cell-type-specific network is lower than the number of cell-type-specific genes because we only considered directly connected cell-type-specific gene pairs. In other words, for a cell type c, any c-specific gene not connected to other c-specific gene was not a part of the c-specific network. The cell type with the most specific genes was hematopoietic stem cell with 1,156 specific genes. The cell type with the least specific genes was blood-derived monocyte with 132 genes. The complete list of tissues, cell types and cell-type-specific genes is available at S2 Table.

[Figure omitted. See PDF.]

Table 2. Number of disease-associated genes on cell-type-specific networks.

https://doi.org/10.1371/journal.pone.0224448.t002

Cellular pathways

The number of pathways in Reactome database was 519 after filtering, with an average pairwise overlap of 0.01%. Overall, 6,989 genes were associated to at least one pathway. On average, ~37% of genes on cell-type-specific networks were associated to at least one pathway. The fraction of pathway-associated genes present in each cell type is shown in S4 Table. The list of genes associated to each pathway in each cell-type-specific network is provided in S5 Table. The connectivity Ccp of the pathways is shown in S6 Table. As an example, Fig 3 shows cellular pathway (Regulation of TLR by endogenous ligand) mapped onto a cell-type specific network (CD19+ B cell).

[Figure omitted. See PDF.]

Fig 3. Pathway Toll Like Receptor 4 TLR4 Cascade on the CD19+ B cell specific network.

(A) Complete view of the largest component of the network. Pathway-associated genes and their interactions are shown in orange. (B) Zoom to the pathway-associated genes and their closest neighbors only. Pathway-associated genes and their connections are shown in orange. (C) Top-scoring asthma genes (see Methods) are shown with blue borders. (D) Top-scoring dermatitis genes are shown with blue borders. (E) Top-scoring rhinitis genes are shown with blue borders. The fraction of pathway genes within the top-scoring gene sets is only significant for dermatitis and rhinitis. (F-H) Distribution of random Perturbation Score (PS) for A, D and R, respectively. An arrow represents the real PS. Pathways whose PS is significantly larger than random expectation (P < 0.05, panels G and H) are denoted as perturbed in the respective disease.

https://doi.org/10.1371/journal.pone.0224448.g003

Quantification of cell-type-specific multimorbidity

The Multimorbidity Score (MS) quantitatively measured the multimorbidity between A, D and R specific to different cell types (Table 3). S2 Table contains the number of top-scoring genes for each disease on each cell-type-specific network (|Scd|, see Methods). Of the 60 cell-type-specific networks, 12 were associated to a single disease and were not considered for further multimorbidity analysis. Inspection of Table 3 shows 14 cell types associated to ADR multimorbidity because their MS value is > 0 for all combinations of the three diseases (the strength of the association given by the MS value, ranging from 0 to 1). The cell types include monocytes-macrophages, T cells and plasma cells, as well as skin endothelial cells and esophageal epithelial cells. These 14 cell types will be subject to scrutiny in the following sections. S7 Table provides a combined overview of the results of Tables 2 and 3, containing the cell types with a significant number of A-, D- or R-associated genes as well as those cell types with nonzero MS.

[Figure omitted. See PDF.]

Table 3. Cell-type-specific multimorbidities between asthma, dermatitis and rhinitis.

https://doi.org/10.1371/journal.pone.0224448.t003

Cell-type-specific multimorbidity mechanisms

Table 4 shows the pathways identified as candidate mechanisms for multimorbidity in the 14 cell types where MS for ADR is >0 (Table 3), where pathways in the Cytokine signaling in immune system category roughly correspond to the pathways activated in the type-2 asthmatic response (particularly, IL4 and IL13 signaling [71, 72]). S8 Table shows candidate mechanisms in all other cell types (which are restricted to AD multimorbidity except for one pathway in primary bone marrow CD34+ stem cells, associated to AR multimorbidity). It is noteworthy that some cell types do not present any significant mechanism for multimorbidity despite being associated to multimorbidity in Table 3 (namely, epidermis/dermis, and primary microvascular endothelial cells, not associated to any pathway). Other cell types are strongly associated to ADR multimorbidity while not being associated to any mechanism for ADR multimorbidity. This is the case of CD14+ monocytes, for which only a mechanism mediation AD multimorbidity (NOD1/2 signaling pathway) was found. The reason for these observations is that, on average, only ~37% of genes in a given cell type are annotated to a least one pathway (S4 Table). Thus, a large number of non-annotated genes might be still contributing to multimorbidity. The cellular pathways perturbed in each individual disease and cell type (i.e PScpd significant at P < 0.05, see Methods) are provided in S9 Table.

[Figure omitted. See PDF.]

Table 4. Cellular pathways associated to multimorbidity between asthma, dermatitis and rhinitis.

https://doi.org/10.1371/journal.pone.0224448.t004

Candidate multimorbidity genes

Table 5 shows the 30 top-scoring candidate genes for multimorbidity (and S10 Table contains the full collection of candidate genes). The score assigned to multimorbidity (columns AD, AR, DR, ADR in Table 5 and S10 Table) can be read as the importance of the gene as mediator for multimorbidity. As expected, many of the top-scoring candidates are associated to immune system pathways. It is noteworthy that some genes may be associated to pathways which are, in fact, not characterized as multimorbidity mechanisms. For instance, Table 5 shows IL13 gene as a strong ADR multimorbidity candidate in esophageal epithelium. This gene is annotated as belonging to the Interleukin-10 signaling an Interleukin-4 and 13 signaling pathways. However, neither pathway was characterized as a mechanism of multimorbidity for esophageal epithelium in Table 4, because their perturbation score PSTpd did not reach statistical significance. Genes in Table 5 show a higher score, on average, for AR than for AD multimorbidity (P = 0.01482; paired Wilcoxon-Mann-Whitney test), implying a more closely-knit biological mechanism for AR than for AD multimorbidity. The same was observed for AD vs DR (P = 1.02·10−3; paired Wilcoxon-Mann-Whitney test) but not for AR vs DR. This observation was also true when comparing scores of the whole set of predicted genes in S10 Table. Comparisons are shown in S11 Table. Genes which are not known be associated to any of the diseases under study (i.e. they are not present in Table 1) but were characterized as candidates for multimorbidity are particularly interesting candidates for experimental characterization. There are 100 genes of this kind, and 21 of them are candidates for ADR multimorbidity. Table 6 shows the 30 top-scoring ones.

[Figure omitted. See PDF.]

Table 5. Candidate genes associated to multimorbidity between A, D and R.

https://doi.org/10.1371/journal.pone.0224448.t005

[Figure omitted. See PDF.]

Table 6. Candidate genes associated to multimorbidity between A, D and R, and not associated to any of the diseases.

https://doi.org/10.1371/journal.pone.0224448.t006

Discussion

In this study, we have performed an interactome-based analysis of expression data to characterize specific mechanisms for multimorbidity between asthma (A), dermatitis (D) and rhinitis (R) in distinct 14 non-eosinophilic cell types and 15 tissues. We observed differential roles for cytokine signaling, particularly associated with type 2 inflammation, TLR-mediated signaling and metabolic pathways for multimorbidities across distinct cell types. Furthermore, we also identified individual genes potentially associated to multimorbidity mechanisms.

Strengths

Interactome-based computational analysis provide a global view of the increasing complexity of disease-gene association data, and the relationships among diseases, genes and functions [73]. By employing an expression compendium that incorporates information on multiple heterogeneous gene expression experiments, we were able to identify cell-type-specific mechanisms that underlie the multimorbidity between A, D and R, focusing on 14 cell types that are emerging as major components in these complex diseases in 15 distinct tissues. Although eosinophils are an important cell type in A [30, 74], we focused on other important yet no so well-studied cell types in connection to ADR multimorbidity.

Our approach characterizes the mechanisms of multimorbidity not only by analyzing the contributions of individual genes, but also their interrelationship and their connectivity to other genes within the interactome. This is relevant because molecular causes of multimorbidity are not restricted to shared genes, but involve a cascade of common perturbed cellular mechanisms without which the whole mechanisms of multimorbidity cannot be properly characterized. Although the statistical analysis of the overlap between sets of genes has been widely employed to uncover disease-disease and disease-pathway associations, the limited knowledge of disease-associated genes and lack of annotation data have hampered its results [75, 76]. More recent approaches incorporating interactome-derived data provided a substantial improvement to characterize multimorbidity [20, 65, 76, 77]. Our approach can detect multimorbidity even if no shared genes are involved by identifying the cell-type-specific mechanisms associated to multimorbidity. In this respect, and because cellular pathways represent a curated set of gene functions which may be only partially present in some cell types, our method allows not only to statistically quantify if a pathway can be considered as a specific multimorbidity mechanism in a cell type, but also the discovery of particular genes involved in the multimorbidity process. Finally, our method is fully scalable approach, making it possible to study and characterize the etiology critical for multimorbidity between large groups of diseases. The findings of this in silico study are hypothesis-generating and are intended to guide new experiments on cell-type-specific allergic multimorbidity. Consequently, they should be confirmed by proper mechanistic and genetic studies.

Weaknesses

As usual in differential expression studies, we are considering the gene expression level as a proxy for the gene activity. However, these two characteristics do not always match. For instance, a gene can be significantly over-expressed in a certain tissue or cell type and yet, at the same time, its product can be rendered inactive through a post-translational modification (e.g. phosphorylation). Our methodology does not capture those cases. Similarly, the time-dependent gene expression patterns are not captured in our study, which only considers an interactome static in time.

Lack of data availability also limited our analysis. Eosinophils are not a part of the expression compendium used in this study. However, to the best of our knowledge, no cell-type- or tissue-wide expression compendium resolving eosinophils as an individual cell type exists. This is why we chose to focus our attention in other cell types, important yet no so well-studied in connection to ADR multimorbidity. Furthermore, our dataset reflects only expression levels in healthy individuals because no cell-type-wide expression compendium in subjects with ADR multimorbidity exists.

Another limitation of our study is data completeness. The intersection of expression and interactome data sources yields a low coverage of the complete genome. Although this is a common limitation (and authors have argued that the current coverage of the human interactome does not limit its successful application to the investigation of disease mechanisms [5, 16]), some data loss is unavoidable: for instance, a protein such as filaggrin (FLG), commonly associated to multimorbidity between A and D [78], was not present in our expression dataset and could not be incorporated to the study. Also, our expression dataset contains data primarily from adult subjects. Thus, it is unclear if our results can be generalized to other age groups like young children or elderly people. However, we believe that gain in knowledge largely compensates these limitations.

As for disease-gene associations, we are including gene-disease associations partially derived from GWAS studies, whose reliability has been questioned [79–81]. Additionally, the current human interactome is highly biased toward highly studied genes (a category that includes many disease-associated genes), representing only a very small densely connected fraction of the full interactome [82–86]. This bias might be larger than expected and may have an impact on the biological conclusions extracted from the studies of the interactome [87]. However, non-biased interactomes have a much lower coverage, which makes them unsuitable for some topology-based studies [87]. We tried to address this effect by building null models which take into account the degree of the original genes. It should be also noted that there are numerous factors, other than genetic ones, that determine multimorbidity, some of which are environmental, lifestyle-related or treatment-induced. Finally, different mutations on the same gene can have different pathological effects on its gene products [88]. We considered all disease-associated mutations to have an effect on gene activity that, in turn, has a molecular impact on the interactome.

Quantification of cell-type-specific multimorbidity

The MS measure treats multimorbidity symmetrically with respect to the diseases being compared, meaning that it numerically reflects the mutual influence that the manifestation of one disease exerts over the other disease in a cell type. It can be interpreted as a measure of the degree to which a multimorbidity is present and specific to a certain cell type (regardless the fact that systemic mechanisms may be playing a role in multimorbidity as well, a case which is not captured by our method). Lower MS values imply that the specific mechanisms of the diseases are largely detached from each other in the corresponding cell type: the perturbation caused by the manifestation of one disease d1 will be less likely to travel throughout the network and perturb the mechanisms that give rise to disease d2. At MS = 0 there is no multimorbidity between the diseases in the corresponding cell type (although multimorbidity may be present as a more systemic process). At MS = 1, the mechanisms of both diseases are identical in that cell type. We find an example of this in the primary microvascular endothelial cells, where MS = 1 for the DR multimorbidity. The implication of this is that not only the gene sets associated to D and R are identical in this cell type, but also that the gene sets influenced (or perturbed) by the malfunction of those genes are also identical, thus rendering both diseases the same disease in mechanistic terms for this cell type. Our methodology identifies cell-type-specific interactomes that are not exclusive of a single cell type: some parts of the interactome can be shared by more two or more (usually related) cell types.

MS revealed that all cell types with a significant number of disease-associated genes in at least one disease also display some degree of multimorbidity. For instance, genes associated to A and D are significantly associated to the monocyte cell-type-specific network (Table 2), which also displays a MS > 0 across all multimorbidities (AD, AR, DR, ADR; Table 3). The reverse, however, is not necessarily true: primary microvascular endothelial cells displayed high MS values despite not showing any significant gene association. The reason lies in the use of interactome data, which takes into account the interconnectivity amongst genes as well as their number, allowing for the identification of multimorbidities that would go unnoticed in a standard association analysis. In this line, it is also of note that a significant number of disease-associated genes in a cell type does not necessarily imply a stronger MS. For instance, macrophages and monocyte-derived macrophages have a significant number of disease-associated genes for A and D, and yet their MS value for AD multimorbidity are 0.17 and 0.15, respectively. As another example, CD14+ T cells show large MS values for all multimorbidities despite the fact that no statistical association was found neither with D- nor with R-associated genes in this cell type.

Cell-type-specific multimorbidity mechanisms

Cytokine signaling, critical to the induction of the type 2 response, seems to be the main mechanism behind AD multimorbidity, and it is present in a number of distinct cell types, blood-related or not (Table 4, S8 Table). IL4 and IL13 have long been known to be amongst the cytokines secreted by Th2 cells in response to allergen-induced IgE synthesis in A, and the existence of an underlying IL4- and IL13-mediated pathomechanism for this multimorbidity has been suggested by a number of observations, for instance the response to similar treatments (e.g. dupilumab, a human monoclonal antibody that inhibits this type of signaling) [3]. IL10-associated signaling, a regulator of other proinflammatory cytokines [89], was also found as a contributing mechanism for AD multimorbidity across many cell types, as was IL1-associated signaling. IL1 is a known inflammatory marker associated to D and bronchial A [90], amongst other diseases with inflammatory components. Interestingly, a role for IL1 as a mediator in multimorbidities has already been hinted, as IL1 blocking therapies have proven effective against conditions encountered as comorbidities in patients with rheumatic diseases [91, 92]. We have to point out, however, that the definition of a pathway (as a functionally annotated gene set) should be taken into account when analyzing those results. For instance, the pathway Antigen Processing and Cross-Presentation is associated to AR multimorbidity in erythrocytes (Table 4). This contradicts evidence on MHC presence in human nonnucleated cells [93] because of the definition of the pathway in Reactome database, that includes genes also annotated in TLR-mediated pathways.

On the other hand, innate immune response mediated by toll-like receptors (TLRs) seems to be the key mechanism for multimorbidities implicating R. The TLR family of genes is important in barrier homeostasis and in the activation of the innate immune system [94], and there are evidences of its involvement in R [95–97]. Although the link between A and R is well established (the "United Airways" concept [98, 99]), there is limited knowledge about the mechanistic interplay between A and R [100, 101]. AR multimorbidity seems also largely restricted to a few blood-related cell types: CD19+ B cells, monocytes and erythrocytes. Genetic studies have linked the TLR6-TLR1 locus to a role in the development of R [102], and changes in TLR1 have been reported in asthmatic patients [17, 103], but no direct association between A and R is known. Similarly, changes in TLR2 and TLR4 expression are known to disturb the skin barrier in D [104]. According to our observations, TLR4-mediated cascade might play an important role in R-associated multimorbidities in blood-related cell types.

Esophageal epithelium cells seem to be also associated to AR multimorbidity by means of IL1 signaling pathway and genes such as IL-13 and IL-33. It is known that chronic eosinophilic inflammation of the esophagus is associated with tissue remodeling and fibrosis that shares many traits with A [105, 106]. Patients with eosinophilic esophagitis often present multimorbid conditions that include A and D [107]. It is also noteworthy the role of metabolism of proteoglycans in CD19+ B cells for this multimorbidity. In this sense, our results indicate that structurally similar proteoglycans neurocan (NCAN) and versican (VCAN) are related to this mechanism. Although no evidence linking these two genes to A or R is known, VCAN encodes an extracellular matrix protein that has been associated with A in murine models and with bronchiolitis in humans [108, 109].

We observed that cells of the skin epidermis/dermis, and primary microvascular endothelial cells were not significantly associated to any pathway. A number of reasons explain this observation: first, as already noted in the Results section, annotated pathways only cover approximately one-third of all genes in our cell-type-specific networks, leaving room for yet-unannotated mechanisms to play a critical role in multimorbidity. Second, our approach identifies significantly perturbed pathways, implying that some pathways may be perturbed without reaching the statistical significance cutoff of α = 0.05. Finally, our study only reflects cell-type-specific mechanisms, not excluding the existence of systemic mechanisms that may have relevant impact in a number of cell types. The fact that no pathway was characterized for these cell types, however, does not preclude the existence of individual candidate genes which might be playing a role in multimorbidity in them (see next section).

Cell-type-specific candidate genes

We identified a number of individual genes as potentially associated to multimorbidity (Tables 5 and 6; S10 Table). The identification of candidate genes complements the characterization of mechanisms of multimorbidity based on pathway annotation. For instance, interleukin 1 receptor-like 1 (IL1RL1) is amongst the top-scoring candidates for ADR multimorbidity in primary microvascular endothelial skin-derived cells, yet it is not associated to any pathway in this cell type (and, thus, its contribution would have been lost had we focused solely on pathway-annotated multimorbidity mechanisms). A candidate gene in a particular cell type may belong to a pathway not identified as a mechanism in that cell type. This is the case, for example, of the IL13 gene, a high-scoring candidate gene in esophageal epithelium for ADR multimorbidity. This gene belongs to two pathways: Interleukin-4 and 13 signaling and Interleukin-10 signaling, and yet none of the two pathways is identified as a significant mechanism for this cell type and multimorbidity (because when considering all the pathway-associated genes, neither pathways is found to be significantly perturbed). Thus, we can conclude that IL13 may play and important role as a multimorbidity mediator. Our results also provide valuable information of cell-type-specificity of candidate genes. For instance, IL4 and IL5, two of the main inflammatory cytokines, in are associated to monocytes but not to macrophages (S10 Table), in agreement with previous observations [110].

The only non-cytokine-related gene in the top 10 positions of Table 5 is the PLA2G7 gene, which controls inflammation though the inactivation of platelet-activating factor (PAF), a potent phospholipid-derived mediator of inflammation that is secreted by many immune cells and controls vascular permeability. Although no study associating PLA2G7 to ADR multimorbidity exists, it is a strong candidate if we take into account the wide range of actions of PAF (considered a universal biological regulator [111]) that in turn associates PLA2G7 to a number of inflammatory conditions other than A, D or R [112–114]. The "United Airways" concept, introduced in the previous section, is also supported by our results: on average, mechanisms mediating between A and R are more closely-knit (represented by higher average scores) than mechanisms mediating A and D, although A and D share more disease-associated genes (Table 1).

Some of the highest-scoring candidate genes were not even associated to any of the diseases of interest (Table 6) illustrating the potential of our approach to characterize yet-undescribed molecular mechanisms of multimorbidity. We predict interleukin receptors IL22RA1 and IL20RA to play an important role in the ADR multimorbidity in the esophageal epithelium. To date, IL22RA1 had been only associated to inflammatory responses in airway epithelia by genetic studies, and IL20RA to psoriasis [115, 116]. Also, the functional nature of genes in Table 6 is also much more diverse than that of Table 5. This is strongly suggestive of a research bias towards already-known cytokine-related mediators when it comes to the study of these allergic diseases, overlooking other functional groups. For instance, the second highest-scoring gene in Table 6 is butyrylcholinesterase BCHE, a poorly-studied detoxifying enzyme that has been proposed as a marker to identify and prognose systemic inflammation [117, 118] and that has only marginally associated to allergic diseases. BCHE is highlighted by our method as a mediator in ADR multimorbidity in skin. Table 6 also shows that the role of proteoglycans seems to be restricted to AR multimorbidity only through neurocan (NCAN) and chondroitin sulfate proteoglycan 5 (CSPG5). Although proteoglycans are known to influence the remodelling of nasal mucosa in R [119], no evidence exists linking them to allergic multimorbidity. However, our results indicate that TLRs are characteristically associated to multimorbidity involving R (Table 4), so there may be an interesting link between TLRs and proteoglycans in relation to AR multimorbidity, since it is known that chondroitin sulfate proteoglycans have the ability to bind TLRs and activating macrophages [120].

Comparison to our previous study

In our previous in silico study of multimorbidity between A, D and R, we explored multimorbidity at whole organism level [21]. In this study we incorporated additional data that reflects the spatial cell-type-specific nature of the diseases and their multimorbidity. This presents a key opportunity to better understand the mechanisms of diseases, since cell-type-specific data provides a more accurate picture of multimorbidity. We incorporated changes in the methodology as well. It remains focused on exploiting the topology of the interactome, but adopting a more complex approach that measures the role of pathways not only in terms of their direct interactions to individual disease-associated genes, but in terms of their global connectivity to those genes within a specific network.

Methodologically, differences in the gene-disease data sources used in both studies have an impact in the characterization of disease-associated genes. Also, availability of expression data limited the number of genes present in the study. For instance, thymic stromal lymphopoietin (TSLP), found to be associated to A, D and R in our previous study, is absent in this study because it was not present in the expression compendium. Also, pathway annotation in our previous study was extracted from BioCarta database, which is no longer updated, which made us chose Reactome database instead.

One of the main findings in our previous study was the significant role of eosinophilic-mediated pathways in AD multimorbidity (BioCarta pathways CCR3 signaling in Eosinophils and The Role of Eosinophils in the Chemokine Network of Allergy were identified with a high score). Because eosinophils were not included in the present study, the Reactome equivalents of those pathways (S1 Table) are not present in our results, confirming that our observations can be linked to mechanisms mediated by other cell types. IL10 signaling pathway, a relevant mechanism in AD multimorbidity across many cell types, was also identified amongst the highest-scoring pathways in our previous study (under the BioCarta denomination Regulation of hematopoisesis by cytokines). Our previous study also linked IL4-mediated, GATA3-mediated mechanisms and 4–1BB-dependent immune responses to ADR multimorbidity. GATA3-mediated mechanisms are represented in our dataset by interleukin pathways in the Cytokine signaling in immune system category, and, from a cell-type-specific point of view, these processes seem more relevant in AD multimorbidity (despite the fact that IL1, IL4 and IL13 signaling in particular also contribute to ADR multimorbidity in some cell types). Aside from differences in pathway annotation of genes, this could reflect a more systemic role for these pathways in ADR multimorbidity. However, 4–1BB-dependent immune response (represented by Toll-like receptor cascades in our dataset) is clearly associated to AR, DR and ADR multimorbidity in a number blood-derived cell types. In all, we believe that our results are complementary to those of our previous study since they focus on the cell-type-specific mechanisms of multimorbidity instead of global (or systemic) ones.

Conclusions

We designed an in silico approach that integrated current public expression and network interaction databases and applied an interactome-based analysis to uncover the cell-type-specific pathophysiological mechanisms of multimorbidity between A, D and R. We observed that interleukin-mediated signaling is present in all multimorbidities involving asthma but not rhinitis, while rhinitis-associated multimorbidities have a strong TLR-mediated component. IL1 signaling is the only type-2 pathway candidate for AR multimorbidity, found in esophageal epithelium. We also generated a collection of genes potentially linked to cell-type-specific multimorbidity, some of which were not previously associated to any of the diseases. Our results provide a better understanding of the pathophysiological mechanisms triggering ADR multimorbidity, assisting in the design of new mechanistic and clinical studies.

Supporting information

[Figure omitted. See PDF.]

S1 Fig. Illustration of the process to calculate cell-type-specific multimorbidity.

This toy example uses a simplified network of the cell type c, where we will measure the multimorbidity score MS for diseases d1 and d2. The numbers circled in grey correspond to the numbered steps in the section Calculating cell-type-specific multimorbidity of Methods. (A) Genes associated to dis1 (6, orange border) are given an initial score of 1, while all other genes are given a score of 0. (B) The NetScore algorithm scores all genes in the network according to their connectivity to the D-associated genes (blue gradient). Genes in closer proximity to dis1-associated genes get higher scores. (C) The top-scoring genes are selected (in blue). Disease dis1 has 13 top-scoring genes (Scdis1). (D) Genes associated to dis2 (5, in orange border) are given an initial score of 1, while all other genes are given a score of 0. (E) The NetScore algorithm scores all genes according to their connectivity to the dis2-associated genes (blue gradient). (F) The top-scoring genes are selected (in blue). Disease dis2 has 47 top-scoring genes (Scdis2). (G) There is 1 gene common to both top-scoring sets (in blue). The Multimorbidity Score (MS) of the diseases is calculated as the Sorensen-Dice overlap between their top-scoring gene sets. In this case, MScdis1,dis2 is (2 · 1) / (6 + 47) = 0.038. A permutation test over 103 iterations will establish if MScdis1,dis2 is statistically significant (P < 0.05).

https://doi.org/10.1371/journal.pone.0224448.s001

(PNG)

S2 Fig. Illustration of the process to characterize cell-type-specific multimorbidity mechanisms.

This example uses the network of S1 Fig (225 genes). The pathway P has a total of annotated 20 genes, of which 9 are in the network (shown in orange border). (A) The 13 top-scoring genes for disease d1 (Sc d1; see S1C Fig) are shown in blue, and there are 3 pathway genes within this set. Thus, the perturbation score PSc d1,P is (9/20) / (13/225) = 7.79. For the sake of the example, we will assume that this value is significantly larger than random expectation (P < 0.05). (B) The 47 top-scoring genes for disease d2 (S cd2; see S1F Fig) are shown in blue. There are 7 pathway genes within the Scd2 set. Thus, the perturbation score PScd2,P is (9/20) / (47/225) = 2.15. For the sake of the example, we will assume that this value is significantly larger than random expectation as well (P < 0.05). Consequently, because pathway P is significantly associated to (or perturbed by) diseases d1 and d2, we assume that it is part of the mechanism of multimorbidity between dis1 and dis2 in cell type c.

https://doi.org/10.1371/journal.pone.0224448.s002

(PNG)

S1 Table. Association between Reactome pathways and BioCarta pathways.

Only significant associations are shown. LOR: Log Odds Ratio.

https://doi.org/10.1371/journal.pone.0224448.s003

(XLS)

S2 Table. List of cell-type-specific genes.

This table contains: 1) the database sources of diease-associated genes; 2) the complete list of cell types and tissues (including those without disease-associated genes, discarded in this study); 3) the list of all cell-type-specific genes.

https://doi.org/10.1371/journal.pone.0224448.s004

(XLS)

S3 Table. Fraction of disease-associated genes in each cell type.

Statistical significance was calculated by means of a Fisher’s Exact Test.

https://doi.org/10.1371/journal.pone.0224448.s005

(XLS)

S4 Table. Fraction of pathway-associated genes present in each cell type.

https://doi.org/10.1371/journal.pone.0224448.s006

(XLS)

S5 Table. List of genes associated to each pathway in each cell-type-specific network.

https://doi.org/10.1371/journal.pone.0224448.s007

(XLS)

S6 Table. The connectivity Ccp of the pathways.

https://doi.org/10.1371/journal.pone.0224448.s008

(XLS)

S7 Table. Summary of Tables 2 and 3.

The column n diseases contains the number of diseases (A, D, R) with a significant number of associated genes from Table 2 (values are highlighted in blue gradient). The column n MS > 0 contains the number of combinations of diseases (AD, AR, DR, ADR) with nonzero MS from Table 3 (values are highlighted in red gradient). The column n MS > 0.50 contains the number of combinations of diseases (AD, AR, DR, ADR) with MS > 0.50 (also from Table 3, highlighted in red gradient).

https://doi.org/10.1371/journal.pone.0224448.s009

(XLS)

S8 Table. Cellular pathways associated to multimorbidity between asthma, dermatitis and rhinitis.

Red cells: multimorbidity between A and D. Orange cells: multimorbidity between A and R. Light blue cells: multimorbidity between D and R. Dark blue cells: multimorbidity between A, D and R. Only cell types not present in Table 4 in the manuscript are shown.

https://doi.org/10.1371/journal.pone.0224448.s010

(XLS)

S9 Table. Pathways associated to diseases in the cell-type-specific networks.

A: asthma. D: dermatitis. R: rhinitis. Only significant associations (P < 0.05) are shown.

https://doi.org/10.1371/journal.pone.0224448.s011

(XLS)

S10 Table. Complete list of candidate genes for multimorbidity.

Colors and dots are as in Tables 5 and 6 in the manuscript. Pathway associations with a grey background mean that the pathway was not associated to the corresponding cell type (see Table 4, S8 Table).

https://doi.org/10.1371/journal.pone.0224448.s012

(XLS)

S11 Table. Comparison of multimorbidity scores.

Scores for AD, AR and DR multimorbidities from Table 5 (30 top-scoring genes) and S10 Table (all genes) are pairwisely compared by means on a Wilcoxo-Mann-Whitney paired test.

https://doi.org/10.1371/journal.pone.0224448.s013

(XLS)

S1 Text. Supplementary Methods.

https://doi.org/10.1371/journal.pone.0224448.s014

(PDF)

Acknowledgments

We thank Judith García-Aymerich, PhD and Emre Guney, PhD for fruitful discussions.

Citation: Aguilar D, Lemonnier N, Koppelman GH, Melén E, Oliva B, Pinart M, et al. (2019) Understanding allergic multimorbidity within the non-eosinophilic interactome. PLoS ONE 14(11): e0224448. https://doi.org/10.1371/journal.pone.0224448

References

1. Barabási AL, Gulbahce N, Loscalzo J. Network medicine: a network-based approach to human disease. Nat Rev Genet. 2011 Jan;12(1): 56–68. pmid:21164525

2. Ghiassian SD, Menche J, Chasman DI, Giulianini F, Wang R, Ricchiuto P et al. Endophenotype Network Models: Common Core of Complex Diseases. Sci Rep. 2016 Jun 9;6: 27414. pmid:27278246

3. Gandhi TK, Zhong J, Mathivanan S, Karthick L, Chandrika KN, Mohan SS et al. Analysis of the human protein interactome and comparison with yeast, worm and fly interaction datasets. Nat Genet. 2006 Mar;38(3): 285–93. pmid:16501559

4. Sharma A, Menche J, Huang CC, Ort T, Zhou X, Kitsak M et al. A disease module in the interactome explains disease heterogeneity, drug response and captures novel pathways and genes in asthma. Hum Mol Genet. 2015 Jun 1;24(11): 3005–20. pmid:25586491

5. Santolini M, Barabási AL. Predicting perturbation patterns from the topology of biological networks. Proc Natl Acad Sci U S A. 2018 Jul 3;115(27): E6375–E6383. pmid:29925605

6. Kitsak M, Sharma A, Menche J, Guney E, Ghiassian SD, Loscalzo J, Barabási AL. Tissue Specificity of Human Disease Module. Sci Rep. 2016 Oct 17;6: 35241. pmid:27748412

7. Winter EE, Goodstadt L, Ponting CP. Elevated rates of protein secretion, evolution, and disease among tissue-specific genes. Genome Res. 2004 Jan;14(1): 54–61. pmid:14707169

8. Lage K, Hansen NT, Karlberg EO, Eklund AC, Roque FS, Donahoe PK et al. A large-scale analysis of tissue-specific pathology and gene expression of human disease genes and complexes. Proc Natl Acad Sci U S A. 2008 Dec 30;105(52): 20870–5. pmid:19104045

9. Barshir R, Shwartz O, Smoly IY, Yeger-Lotem E. Comparative analysis of human tissue interactomes reveals factors leading to tissue-specific manifestation of hereditary diseases. PLoS Comput Biol. 2014 Jun 12;10(6): e1003632. pmid:24921629

10. Shen-Orr SS, Tibshirani R, Khatri P, Bodian DL, Staedtler F, Perry NM et al. Cell type-specific gene expression differences in complex tissues. Nat Methods. 2010 Apr;7(4): 287–9. pmid:20208531

11. Pandey AK, Lu L, Wang X, Homayouni R, Williams RW. Functionally enigmatic genes: a case study of the brain ignorome. PLoS One. 2014 Feb 11;9(2): e88889. pmid:24523945

12. Greene CS, Krishnan A, Wong AK, Ricciotti E, Zelaya RA, Himmelstein DS et al. Understanding multicellular function and disease with human tissue-specific networks. Nat Genet. 2015 Jun;47(6): 569–76. pmid:25915600

13. Goh KI, Cusick ME, Valle D, Childs B, Vidal M, Barabási AL. The human disease network. Proc Natl Acad Sci U S A. 2007 May 22;104(21): 8685–90. pmid:17502601

14. Schuster-Böckler B, Bateman A. Protein interactions in human genetic diseases. Genome Biol. 2008 Jan 16;9(1): R9. pmid:18199329

15. Zhong Q, Simonis N, Li QR, Charloteaux B, Heuze F, Klitgord N et al. Edgetic perturbation models of human inherited disorders. Mol Syst Biol. 2009;5: 321. pmid:19888216

16. Menche J, Sharma A, Kitsak M, Ghiassian SD, Vidal M, Loscalzo J et al. Disease networks. Uncovering disease-disease relationships through the incomplete interactome. Science. 2015 Feb 20;347(6224): 1257601. pmid:25700523

17. Lee WI, Yao TC, Yeh KW, Chen LC, Ou LS, Huang JL; PATCH Study Group. Stronger Toll-like receptor 1/2, 4, and 7/8 but less 9 responses in peripheral blood mononuclear cells in non-infectious exacerbated asthmatic children. Immunobiology. 2013 Feb;218(2): 192–200. pmid:22727330

18. Park J, Lee DS, Christakis NA, Barabási AL. The impact of cellular networks on disease comorbidity. Mol Syst Biol. 2009;5: 262. pmid:19357641

19. Gomez-Cabrero D, Menche J, Vargas C, Cano I, Maier D, Barabási AL et al. From comorbidities of chronic obstructive pulmonary disease to identification of shared molecular mechanisms by data integration. BMC Bioinformatics. 2016 Nov 22;17(Suppl 15): 441. pmid:28185567

20. Rubio-Perez C, Guney E, Aguilar D, Piñero J, Garcia-Garcia J, Iadarola B et al. Genetic and functional characterization of disease associations explains comorbidity. Sci Rep. 2017 Jul 24;7(1): 6207. pmid:28740175

21. Aguilar D, Pinart M, Koppelman GH, Saeys Y, Nawijn MC, Postma DS et al. Computational analysis of multimorbidity between asthma, eczema and rhinitis. PLoS One. 2017 Jun 9;12(6): e0179125. pmid:28598986

22. Ferreira MA, Vonk JM, Baurecht H, Marenholz I, Tian C, Hoffman JD et al. Shared genetic origin of asthma, hay fever and eczema elucidates allergic disease biology. Nat Genet. 2017 Dec;49(12): 1752–1757. pmid:29083406

23. Demenais F, Margaritte-Jeannin P, Barnes KC, Cookson WOC, Altmüller J, Ang W et al. Multiancestry association study identifies new asthma risk loci that colocalize with immune-cell enhancer marks. Nat Genet. 2018 Jan;50(1): 42–53. pmid:29273806

24. Bousquet J, Anto J, Auffray C, Akdis M, Cambon-Thomsen A, Keil T, et al. MeDALL (Mechanisms of the Development of ALLergy): an integrated approach from phenotypes to systems medicine. Allergy 2011; 66: 596–604. pmid:21261657

25. Pinart M, Benet M, Annesi-Maesano I, von Berg A, Berdel D, Carlsen KC et al. Comorbidity of eczema, rhinitis, and asthma in IgE-sensitised and non-IgE-sensitised children in MeDALL: a population-based cohort study. Lancet Respir Med. 2014; 2: 131–40. pmid:24503268

26. Garcia-Aymerich J, Benet M, Saeys Y, Pinart M, Basagaña X, Smit HA et al. Phenotyping asthma, rhinitis, and eczema in MeDALL population-based birth cohorts: an allergic comorbidity cluster. Allergy 2015; 70: 973–84. pmid:25932997

27. Celestin J, Frieri M. Eosinophilic disorders in various diseases. Curr Allergy Asthma Rep. 2012 Feb;12(1): 18–24. pmid:22160831

28. Furuta GT, Atkins FD, Lee NA, Lee JJ. Changing roles of eosinophils in health and disease. Ann Allergy Asthma Immunol. 2014 Jul;113(1): 3–8. pmid:24795292

29. Werfel T, Allam JP, Biedermann T, Eyerich K, Gilles S, Guttman-Yassky E et al. Cellular and molecular immunologic mechanisms in patients with atopic dermatitis. J Allergy Clin Immunol. 2016 Aug;138(2): 336–49. pmid:27497276

30. Lambrecht BN, Hammad H. The immunology of asthma. Nat Immunol. 2015 Jan;16(1): 45–56 pmid:25521684

31. Olze H, Zuberbier T. Comorbidities between nose and skin allergy. Curr Opin Allergy Clin Immunol. 2011 Oct;11(5): 457–63. pmid:21822129

32. Kita H. Eosinophils: multifunctional and distinctive properties. Int Arch Allergy Immunol. 2013;161 Suppl 2: 3–9.

33. Incorvaia C, Masieri S, Cavaliere C, Makri E, Sposato B, Frati F. Asthma associated to rhinitis. J Biol Regul Homeost Agents. 2018 Jan-Feb;32(1 Suppl. 1): 67–71. pmid:29552876

34. Poddighe D, Brambilla I, Licari A, Marseglia GL. Pediatric rhinosinusitis and asthma. Respir Med. 2018 Aug;141: 94–99. pmid:30053979

35. Boulay ME, Boulet LP. The relationships between atopy, rhinitis and asthma: pathophysiological considerations. Curr Opin Allergy Clin Immunol. 2003 Feb;3(1): 51–5. pmid:12582315

36. Yu S, Kim HY, Chang YJ, DeKruyff RH, Umetsu DT. Innate lymphoid cells and asthma. J Allergy Clin Immunol. 2014 Apr;133(4): 943–50. pmid:24679467

37. Gieseck RL 3rd, Wilson MS, Wynn TA. Type 2 immunity in tissue repair and fibrosis. Nat Rev Immunol. 2018 Jan;18(1): 62–76. pmid:28853443

38. Davis AP, Grondin CJ, Johnson RJ, Sciaky D, McMorran R, Wiegers J et al. The Comparative Toxicogenomics Database: update 2019. Nucleic Acids Res. 2019 Jan 8;47(D1): D948–D954 pmid:30247620

39. Amberger JS, Hamosh A. Searching Online Mendelian Inheritance in Man (OMIM): A Knowledgebase of Human Genes and Genetic Phenotypes. Curr Protoc Bioinformatics. 2017 Jun 27;58: 1.2.1–1.2.12.

40. Piñero J, Bravo À, Queralt-Rosinach N, Gutiérrez-Sacristán A, Deu-Pons J, Centeno E et al. DisGeNET: a comprehensive platform integrating information on human disease-associated genes and variants. Nucleic Acids Res. 2017 Jan 4;45(D1): D833–D839. pmid:27924018

41. The UniProt Consortium. UniProt: the universal protein knowledgebase. Nucleic Acids Res. 2017 Jan 4;45(D1): D158–D169. pmid:27899622

42. Ramos EM, Hoffman D, Junkins HA, Maglott D, Phan L, Sherry ST et al. Phenotype-Genotype Integrator (PheGenI): synthesizing genome-wide association study (GWAS) data with existing genomic resources. Eur J Hum Genet. 2014 Jan;22(1): 144–7. pmid:23695286

43. Lehne B, Lewis CM, Schlitt T. From SNPs to genes: disease association at the gene level. PLoS One. 2011;6(6): e20133. pmid:21738570

44. Fabregat A, Jupe S, Matthews L, Sidiropoulos K, Gillespie M, Garapati P et al. The Reactome Pathway Knowledgebase. Nucleic Acids Res. 2018 Jan 4;46(D1): D649–D655. pmid:29145629

45. Szklarczyk D, Morris JH, Cook H, Kuhn M, Wyder S, Simonovic M et al. The STRING database in 2017: quality-controlled protein-protein association networks, made broadly accessible. Nucleic Acids Res. 2017 Jan 4;45(D1): D362–D368. pmid:27924014

46. Lukk M, Kapushesky M, Nikkilä J, Parkinson H, Goncalves A, Huber W et al. A global map of human gene expression. Nat Biotechnol. 2010 Apr;28(4): 322–4. pmid:20379172

47. Bezginov A, Clark GW, Charlebois RL, Dar VU, Tillier ER. Coevolution reveals a network of human proteins originating with multicellularity. Mol Biol Evol. 2013 Feb;30(2): 332–46. pmid:22977115

48. Faisal A, Peltonen J, Georgii E, Rung J, Kaski S. Toward computational cumulative biology by combining models of biological datasets. PLoS One. 2014 Nov 26;9(11): e113053. pmid:25427176

49. Isik Z, Baldow C, Cannistraci CV, Schroeder M. Drug target prioritization by perturbed gene expression and network information. Sci Rep. 2015 Nov 30;5: 17417. pmid:26615774

50. Engreitz J, Daigle B Jr, Marshall J, Altman R (2010). “Independent component analysis: mining microarray data for fundamental human gene expression modules.” J Biomed Inform, 43(6): 932–944. pmid:20619355

51. Grapov D, Wanichthanarak K, Fiehn O. MetaMapR: pathway independent metabolomic network analysis incorporating unknowns. Bioinformatics. 2015 Aug 15;31(16): 2757–60. pmid:25847005

52. Nelms BD, Waldron L, Barrera LA, Weflen AW, Goettel JA, Guo G et al. CellMapper: rapid and accurate inference of gene expression in difficult-to-isolate cell types. Genome Biol. 2016 Sep 29;17(1): 201. pmid:27687735

53. Tomlins SA, Rhodes DR, Perner S, Dhanasekaran SM, Mehra R, Sun XW et al. Recurrent fusion of TMPRSS2 and ETS transcription factor genes in prostate cancer. Science. 2005 Oct 28;310(5748): 644–8. pmid:16254181

54. Rhodes DR, Kalyana-Sundaram S, Mahavisno V, Varambally R, Yu J, Briggs BB et al. Oncomine 3.0: genes, pathways, and networks in a collection of 18,000 cancer gene expression profiles. Neoplasia. 2007 Feb;9(2): 166–80. pmid:17356713

55. Kryuchkova-Mostacci N, Robinson-Rechavi M. A benchmark of gene expression tissue-specificity metrics. Brief Bioinform. 2017 Mar 1;18(2): 205–214. pmid:26891983

56. Sonawane AR, Platig J, Fagny M, Chen CY, Paulson JN, Lopes-Ramos CM et al. Understanding Tissue-Specific Gene Regulation. Cell Rep. 2017 Oct 24;21(4): 1077–1088. pmid:29069589

57. Khatri P, Sirota M, Butte AJ. Ten years of pathway analysis: current approaches and outstanding challenges. PLoS Comput Biol. 2012;8(2): e1002375. pmid:22383865

58. Stoney R, Robertson DL, Nenadic G, Schwartz JM. Mapping biological process relationships and disease perturbations within a pathway network. NPJ Syst Biol Appl. 2018 Jun 11;4: 22. pmid:29900005

59. Dice Lee R. "Measures of the Amount of Ecologic Association Between Species". Ecology. 1945; 26 (3): 297–302.

60. Sørensen T. "A method of establishing groups of equal amplitude in plant sociology based on similarity of species and its application to analyses of the vegetation on Danish commons". Kongelige Danske Videnskabernes Selskab. 1948; 5 (4): 1–34.

61. Fuxman Bass JI, Diallo A, Nelson J, Soto JM, Myers CL, Walhout AJ. Using networks to measure similarity between genes: association index selection. Nat Methods. 2013 Dec;10(12): 1169–76. pmid:24296474

62. Nishimura D. Biocarta. Biotech Software & Internet Report. 2001 Vol. 2, No. 3.

63. Benjamini Yoav; Hochberg Yosef "Controlling the false discovery rate: a practical and powerful approach to multiple testing". Journal of the Royal Statistical Society, Series B. 1995; 57 (1): 289–300.

64. Riba M, Garcia Manteiga JM, Bošnjak B, Cittaro D, Mikolka P, Le C et al. Revealing the acute asthma ignorome: characterization and validation of uninvestigated gene networks. Sci Rep. 2016 Apr 21;6: 24647. pmid:27097888

65. Glaab E, Baudot A, Krasnogor N, Schneider R, Valencia A. EnrichNet: network-based gene set enrichment analysis. Bioinformatics. 2012 Sep 15;28(18): i451–i457. pmid:22962466

66. Ko Y, Cho M, Lee JS, Kim J. Identification of disease comorbidity through hidden molecular mechanisms. Sci Rep. 2016 Dec 19;6: 39433. pmid:27991583

67. Guney E, Oliva B. Exploiting protein-protein interaction networks for genome-wide disease-gene prioritization. PLoS One. 2012;7(9): e43557. pmid:23028459

68. Guney E, Oliva B. Analysis of the robustness of network-based disease-gene prioritization methods reveals redundancy in the human interactome and functional diversity of disease-genes. PLoS One. 2014 Apr 14;9(4): e94686. pmid:24733074

69. Lee DS, Park J, Kay KA, Christakis NA, Oltvai ZN, Barabási AL. The implications of human metabolic network topology for disease comorbidity. Proc Natl Acad Sci U S A. 2008 Jul 22;105(29): 9880–5. pmid:18599447

70. McCormack T, Frings O, Alexeyenko A, Sonnhammer EL. Statistical assessment of crosstalk enrichment between gene groups in biological networks. PLoS One. 2013;8(1): e54945. pmid:23372799

71. Fajt ML, Wenzel SE. Asthma phenotypes and the use of biologic medications in asthma and allergic disease: the next steps toward personalized care. J Allergy Clin Immunol. 2015 Feb;135(2): 299–310. pmid:25662302

72. Thomson NC. Novel approaches to the management of noneosinophilic asthma. Ther Adv Respir Dis. 2016 Jun;10(3): 211–34. pmid:26929306

73. Goh KI, Choi IG. Exploring the human diseasome: the human disease network. Brief Funct Genomics. 2012 Nov;11(6): 533–42. pmid:23063808

74. Prussin C, Metcalfe DD. 5. IgE, mast cells, basophils, and eosinophils. J Allergy Clin Immunol. 2006 Feb;117(2 Suppl Mini-Primer): S450–6. pmid:16455345

75. Pan K-H, Lih C-J, Cohen SN (2005) Effects of threshold choice on biological conclusions reached during analysis of gene expression by DNA microarrays. Proceedings of the National Academy of Sciences of the United States of America 102: 8961–8965. pmid:15951424

76. Aguirre-Plans J, Piñero J, Menche J, Sanz F, Furlong LI, Schmidt HHHW, Oliva B, Guney E. Proximal Pathway Enrichment Analysis for Targeting Comorbid Diseases via Network Endopharmacology. Pharmaceuticals (Basel). 2018 Jun 22;11(3).

77. Ogris C, Guala D, Helleday T, Sonnhammer EL. A novel method for crosstalk analysis of biological networks: improving accuracy of pathway annotation. Nucleic Acids Res. 2017 Jan 25;45(2): e8. pmid:27664219

78. Spergel JM. From atopic dermatitis to asthma: the atopic march. Ann Allergy Asthma Immunol. 2010 Aug;105(2): 99–106. pmid:20674819

79. Kraft P, Zeggini E, Ioannidis JP. Replication in genome-wide association studies. Stat Sci. 2009 Nov 1;24(4): 561–573. pmid:20454541

80. Wang K, Bucan M, Grant SF, Schellenberg G, Hakonarson H. Strategies for genetic studies of complex diseases. Cell. 2010 Aug 6; 142(3): 351–3. pmid:20691891

81. Schork AJ, Thompson WK, Pham P, Torkamani A, Roddey JC, Sullivan PF et al. All SNPs are not created equal: genome-wide association studies reveal a consistent pattern of enrichment among functionally annotated SNPs. PLoS Genet. 2013 Apr; 9(4): e1003449. pmid:23637621

82. Das J, Yu H. HINT: High-quality protein interactomes and their applications in understanding human disease. BMC Syst Biol. 2012 Jul 30;6: 92. pmid:22846459

83. Gillis J, Pavlidis P. "Guilt by association" is the exception rather than the rule in gene networks. PLoS Comput Biol. 2012;8(3): e1002444. pmid:22479173

84. Gillis J, Ballouz S, Pavlidis P. Bias tradeoffs in the creation and analysis of protein-protein interaction networks. J Proteomics. 2014 Apr 4;100:44–54. pmid:24480284

85. Vidal M. How much of the human protein interactome remains to be mapped? Sci Signal. 2016 May 10;9(427): eg7. pmid:27165778

86. Cornish AJ, David A, Sternberg MJE. PhenoRank: reducing study bias in gene prioritization through simulation. Bioinformatics. 2018 Jun 15;34(12): 2087–2095. pmid:29360927

87. Schaefer MH, Serrano L, Andrade-Navarro MA. Correcting for the study bias associated with protein-protein interaction measurements reveals differences between protein degree distributions from different cancer types. Front Genet. 2015 Aug 4;6:260. pmid:26300911

88. Dudley AM, Janse DM, Tanay A, Shamir R, Church GM. A global view of pleiotropy and phenotypically derived gene function in yeast. Mol Syst Biol. 2005;1:2005.0001. pmid:16729036

89. de Waal Malefyt R, Haanen J, Spits H, Roncarolo MG, te Velde A, Figdor C et al. Interleukin 10 (IL-10) and viral IL-10 strongly reduce antigen-specific human T cell proliferation by diminishing the antigen-presenting capacity of monocytes via downregulation of class II major histocompatibility complex expression. J Exp Med. 1991 Oct 1;174(4): 915–24. pmid:1655948

90. Krause K, Metz M, Makris M, Zuberbier T, Maurer M. The role of interleukin-1 in allergy-related disorders. Curr Opin Allergy Clin Immunol. 2012 Oct;12(5): 477–84. pmid:22885885

91. Cavalli G, Dinarello CA. Treating rheumatological diseases and co-morbidities with interleukin-1 blocking therapies. Rheumatology (Oxford). 2015 Dec;54(12): 2134–44.

92. Ruscitti P, Cipriani P, Liakouli V, Carubbi F, Berardicurti O, Di Benedetto P, Ciccia F et al. The Emerging Role of IL-1 Inhibition in Patients Affected by Rheumatoid Arthritis and Diabetes. Rev Recent Clin Trials. 2018;13(3):210–214. pmid:29542422

93. Nombela I, Ortega-Villaizan MDM. Nucleated red blood cells: Immune cell mediators of the antiviral response. PLoS Pathog. 2018 Apr 26;14(4):e1006910. pmid:29698529

94. Takeda K, Akira S. Toll-like receptors. Curr Protoc Immunol. 2015 Apr 1;109:14.12.1–10

95. Aryan Z, Rezaei N. Toll-like receptors as targets for allergen immunotherapy. Curr Opin Allergy Clin Immunol. 2015 Dec;15(6): 568–74. pmid:26418475

96. Radman M, Golshiri A, Shamsizadeh A, Zainodini N, Bagheri V, Arababadi MK et al Toll-like receptor 4 plays significant roles during allergic rhinitis. Allergol Immunopathol (Madr). 2015 Jul-Aug;43(4): 416–20.

97. Renkonen J, Toppila-Salmi S, Joenväärä S, Mattila P, Parviainen V, Hagström J et al. Expression of Toll-like receptors in nasal epithelium in allergic rhinitis. APMIS. 2015 Aug;123(8): 716–25. pmid:26061394

98. Compalati E, Ridolo E, Passalacqua G, Braido F, Villa E, Canonica GW. The link between allergic rhinitis and asthma: the united airways disease. Expert Rev Clin Immunol. 2010 May;6(3): 413–23. pmid:20441427

99. Yii ACA, Tay TR, Choo XN, Koh MSY, Tee AKH, Wang DY. Precision medicine in united airways disease: A "treatable traits" approach. Allergy. 2018 Oct;73(10): 1964–1978. pmid:29869791

100. Oka A, Hirano T, Yamaji Y, Ito K, Oishi K, Edakuni N et al. Determinants of Incomplete Asthma Control in Patients with Allergic Rhinitis and Asthma. J Allergy Clin Immunol Pract. 2017 Jan–Feb;5(1): 160–164. pmid:27707660

101. Wise SK, Lin SY, Toskala E, Orlandi RR, Akdis CA, Alt JA et al. International Consensus Statement on Allergy and Rhinology: Allergic Rhinitis. Int Forum Allergy Rhinol. 2018 Feb;8(2): 108–352. pmid:29438602

102. Nilsson D, Henmyr V, Hallden C, Sall T, Kull I, Wickman M et al. Replication of genomewide associations with allergic sensitization and allergic rhinitis. Allergy 2014;69:1506–1514. pmid:25066275

103. Lun SW, Wong CK, Ko FW, Hui DS, Lam CW. Expression and functional analysis of toll-like receptors of peripheral blood cells in asthmatic patients: implication for immunopathological mechanism in asthma. J Clin Immunol. 2009 May;29(3): 330–42. pmid:19067129

104. Panzer R, Blobel C, Fölster-Holst R, Proksch E. TLR2 and TLR4 expression in atopic dermatitis, contact dermatitis and psoriasis. Exp Dermatol. 2014 May;23(5): 364–6. pmid:24661005

105. Atkins D, Furuta GT, Liacouras CA, Spergel JM. Eosinophilic esophagitis phenotypes: Ready for prime time? Pediatr Allergy Immunol. 2017 Jun;28(4): 312–319. pmid:28339136

106. Nhu QM, Aceves SS. Tissue Remodeling in Chronic Eosinophilic Esophageal Inflammation: Parallels in Asthma and Therapeutic Perspectives. Front Med (Lausanne). 2017 Aug 7;4:128.

107. Davis BP, Rothenberg ME. Mechanisms of Disease of Eosinophilic Esophagitis. Annu Rev Pathol. 2016 May 23;11:365–93. pmid:26925500

108. Reeves SR, Kaber G, Sheih A, Cheng G, Aronica MA, Merrilees MJ et al. Subepithelial Accumulation of Versican in a Cockroach Antigen-Induced Murine Model of Allergic Asthma. J Histochem Cytochem. 2016 Jun;64(6): 364–80. pmid:27126823

109. Pasanen A, Karjalainen MK, Bont L, Piippo-Savolainen E, Ruotsalainen M, Goksör E et al. Genome-Wide Association Study of Polymorphisms Predisposing to Bronchiolitis. Sci Rep. 2017 Jan 31;7:41653. pmid:28139761

110. Arango Duque G, Descoteaux A. Macrophage cytokines: involvement in immunity and infectious diseases. Front Immunol. 2014 Oct 7;5:491. pmid:25339958

111. Demopoulos CA, Pinckard RN, Hanahan DJ. Platelet-activating factor. Evidence for 1-O-alkyl-2-acetyl-sn-glyceryl-3-phosphorylcholine as the active component (a new class of lipid chemical mediators). J Biol Chem. 1979 Oct 10;254(19):9355–8. pmid:489536

112. Kald B, Smedh K, Olaison G, Sjödahl R, Tagesson C. Platelet-activating factor acetylhydrolase activity in intestinal mucosa and plasma of patients with Crohn’s disease. Digestion. 1996 Nov-Dec;57(6):472–7. pmid:8913710

113. Travers J, Pei Y, Morin SM, Hood AF. Antiinflammatory activity of the platelet-activating factor receptor antagonist A-85783. Arch Dermatol Res. 1998 Oct;290(10):569–73. pmid:9836508

114. Gill P, Jindal NL, Jagdis A, Vadas P. Platelets in the immune response: Revisiting platelet-activating factor in anaphylaxis. J Allergy Clin Immunol. 2015 Jun;135(6):1424–32. pmid:26051949

115. Zissler UM, Chaker AM, Effner R, Ulrich M, Guerth F, Piontek G et al. Interleukin-4 and interferon-γ orchestrate an epithelial polarization in the airways. Mucosal Immunol. 2016 Jul;9(4): 917–26. pmid:26577568

116. Kingo K, Mössner R, Rätsep R, Raud K, Krüger U, Silm H et al. Association analysis of IL20RA and IL20RB genes in psoriasis. Genes Immun. 2008 Jul;9(5): 445–51. pmid:18480827

117. Zivkovic AR, Schmidt K, Sigl A, Decker SO, Brenner T, Hofer S. Reduced serum butyrylcholinesterase activity indicates severe systemic inflammation in critically ill patients. Mediators Inflamm. 2015;2015:274607. pmid:25762852

118. Delacour H, Dedome E, Courcelle S, Hary B, Ceppa F. Butyrylcholinesterase deficiency. Ann Biol Clin (Paris). 2016 Jun 1;74(3):279–85.

119. Kim TH, Lee JY, Lee HM, Lee SH, Cho WS, Ju YH et al. Remodelling of nasal mucosa in mild and severe persistent allergic rhinitis with special reference to the distribution of collagen, proteoglycans, and lymphaticvessels. Clin Exp Allergy. 2010 Dec;40(12):1742–54. pmid:20860724

120. Stephenson EL, Mishra MK, Moussienko D, Laflamme N, Rivest S, Ling CC et al. Chondroitin sulfate proteoglycans as novel drivers of leucocyte infiltration in multiple sclerosis. Brain. 2018 Apr 1;141(4):1094–1110. pmid:29506186