The Asthma Universe: A Literature-Derived Network Synthesis

1. Introduction

1.1. Burden and Clinical Heterogeneity of Asthma

Asthma is among the most prevalent chronic diseases worldwide, affecting an estimated 300 million individuals and causing approximately 1,000 deaths each day. Most of the deaths occur in low- and middle-income countries, and are preventable [1]. The incidence of the pathology is increasing in children among 0 and 9 years old [2]. Despite substantial therapeutic advances over the past two decades, including the widespread adoption of inhaled corticosteroids and the introduction of targeted biologic therapies, a sizeable proportion of patients continue to experience poor disease control, frequent exacerbations, and progressive loss of lung function [3,4]. The disease further imposes a considerable socioeconomic cost through direct healthcare utilization, lost productivity, and long-term disability [5,6,7].

Clinically, asthma is defined by variable airflow limitation, airway hyperresponsiveness, and chronic airway inflammation [8,9], with the history of typical respiratory symptoms that vary over time and can be worsened by exposure to specific triggers [10]. This apparent simplicity, however, conceals a remarkable degree of biological and clinical heterogeneity. Age of onset, trigger profile, inflammatory signature, comorbidity pattern, and therapeutic responsiveness differ markedly across patients, leading to the progressive recognition that asthma is not a single disease but rather an umbrella category comprising multiple endotypes and phenotypes [11,12,13,14].

1.2. Endotypes, Phenotypes, and the T2 Paradigm

Asthma is a complex disease made up of a number of disease variants with different underlying pathophysiologies. Two main inflammatory endotypes have been identified, which are defined as distinct disease entities with specific mechanisms within the asthma syndrome [15,16]. Type 2 (T2)-high asthma, driven by Th2 lymphocytes, type 2 innate lymphoid cells (ILC2), and the canonical cytokines interleukin (IL)-4, IL-5, and IL-13, accounts for the majority of allergic and eosinophilic forms and is the substrate against which most currently approved biologics are directed [17,18,19]. Non-T2 (T2-low) asthma, characterized by neutrophilic or paucigranulocytic inflammation, involvement of Th1 and Th17 pathways, and interleukin-17 and interleukin-6 signaling, remains poorly responsive to conventional therapies and constitutes a major unmet clinical need [16,20,21,22].

Within each endotype, several clinical phenotypes have been delineated, including allergic asthma, late-onset eosinophilic asthma, aspirin-exacerbated respiratory disease, exercise-induced asthma, obesity-associated asthma, and neutrophilic asthma of the adult smoker [12,23,24,25,26,27]. Severe asthma, defined by the need for high-intensity treatment to maintain control or by persistent uncontrolled disease despite such treatment, represents a particularly heterogeneous category, with its own constellation of molecular drivers, structural changes, and comorbidities [28,29,30]. Pediatric asthma displays additional specificities related to immune system maturation, respiratory viral infections, microbiome development, and environmental exposures during critical developmental windows [31,32,33,34,35].

1.3. Beyond the Airway: Systemic, Environmental, and Ecological Dimensions

Alongside the classical airway-centered view of asthma pathogenesis, an expanding body of evidence has established that asthma is embedded in a much broader network of systemic, environmental, and ecological determinants. Epithelial barrier dysfunction has emerged as a unifying upstream mechanism linking environmental exposures, alarmin release (thymic stromal lymphopoietin, IL-25, IL-33), and downstream type 2 and non-type 2 inflammatory cascades [36,37,38,39]. Dysbiosis of the airway and gut microbiome modulates immune development and inflammatory tone from early life onwards, with measurable consequences on asthma inception, severity, and exacerbation rate [40,41,42,43,44]. Obesity and the metabolic syndrome interact with pulmonary physiology, adipokine signaling, and systemic low-grade inflammation, shaping a distinct obese-asthma phenotype with its own therapeutic challenges [45,46,47,48].

Environmental exposures, including ambient air pollution (particulate matter, ozone, diesel exhaust particles), indoor allergens, tobacco smoke, and occupational agents, act both as direct triggers and as modulators of epigenetic programming, with demonstrable effects on prevalence, severity, and long-term trajectory [49,50,51,52,53,54]. Psychological and neurological comorbidities, including anxiety, depression, altered autonomic regulation, and alterations in neuroimmune circuits, further contribute to symptom burden and treatment adherence [55,56,57,58,59,60]. Finally, asthma frequently coexists with other atopic and non-atopic conditions, such as allergic rhinitis, chronic rhinosinusitis with nasal polyps, atopic dermatitis, gastro-esophageal reflux, obstructive sleep apnea, cardiovascular disease and other pathologies, in configurations that point towards shared immunological and structural substrates and crosstalks rather than coincidental co-occurrence [61,62,63,64,65,66].

1.4. The Limits of Compartmentalized Reviews and the Rationale for a Network-Based Synthesis

The expansion of molecular and clinical knowledge has been accompanied by a proliferation of narrative and systematic reviews, each typically focused on a specific axis of the disease: T2 cytokines and type 2 immunity [17,18,67,68,69,70], epithelial alarmins and barrier function [37,38], eosinophilic inflammation [71,72], neutrophilic inflammation [21], airway remodeling [8,73], microbiome [74,75], air pollution [49,50], obesity [12], pediatric asthma [31,32], severe asthma [4,28], or individual biologic targets [76,77]. While these reviews have been essential in consolidating domain-specific knowledge, their very specialization could fragment the biology of asthma into parallel storylines that rarely meet one another explicitly in the text. A reader interested in how, for instance, microbiome dysbiosis interacts with epithelial barrier dysfunction and contemporary downstream T2 cytokines is forced to reconstruct those connections manually across several reviews. Moreover, there is the possibility of connecting more precisely the molecular patterns, the arising of the pathophysiological networks, the consequent emerging symptoms, diseases and the connections with the environmental factors. This operates at a systemic level, integrating organ-focused perspectives into a whole-organism view.

Network medicine offers a complementary lens for this problem. Originally formalized in the context of human disease networks and interactome analyses [78,79], the approach represents biological knowledge as graphs in which nodes are entities (cells, molecules, processes, clinical phenotypes) and edges are the mechanistic or statistical relationships between them. Applied to a complex disease such as asthma, a network-based representation can make the connectivity structure of the disease explicit, highlight highly connected hubs that integrate multiple axes of pathogenesis, reveal clusters that correspond to biologically meaningful subsystems, and expose gaps in the literature where expected connections are absent or underreported.

The value of such a synthesis is not primarily computational. It is conceptual: by forcing every statement in the literature to be cast as an explicit node-edge pair, the approach exposes implicit assumptions, reconciles terminological inconsistencies across subfields, and provides a shared scaffold on which future experimental and clinical work can be positioned.

1.5. Objectives of the Present Work

The aim of the present scoping review is to systematically map the current state of knowledge on asthma pathophysiology, comorbidity, and pathogenic mechanisms as represented in recent review and systematic review articles, and to synthesize this knowledge in the form of an explicit network. Specifically, the objectives are:

To identify the biological entities (cells, molecules, pathways, structural components, environmental factors, comorbidities, and clinical phenotypes) consistently reported in the recent asthma literature, and to determine which of these are most robustly supported across independent sources.

To extract, normalize, and catalog the mechanistic connections reported between these entities, distinguishing pathogenic, protective, and mixed roles, and attributing each connection to the source or sources that report it.

To construct a literature-derived mechanistic network of asthma and to describe its overall structure, including the identification of high-connectivity hubs and the emergent thematic clusters.

To provide, as companion outputs, structured catalogs of uncertain or contested findings and of currently available therapeutic options, as they map onto the network.

The systemic network of asthma is multi-level, enabling clear connections between molecular mechanisms and emerging phenomena, and it is semantic: each connection has a direction and a type (pathogenic, protective, therapeutic, or mixed), and each node has a category (e.g., severe asthma, T2-high, T2-low). The semantic visualization enables direct usability of the network, supporting immediate learning by the reader.

The review adopts a scoping approach, following the Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) [80]. The scoping format is appropriate because the primary objective is to map the breadth of the evidence and its internal connectivity rather than to answer a single focused clinical question or to synthesize quantitative outcomes [81,82].

2. Materials and Methods

This scoping review was conducted and is reported in accordance with the PRISMA-ScR guidelines [80]. A completed PRISMA-ScR checklist is submitted with the manuscript as non-published material, in line with the journal policy. The methodological framework follows the approach originally proposed by Arksey and O'Malley [81] and subsequently refined by Levac and colleagues [83] and by the Joanna Briggs Institute [82], which is the reference standard for scoping reviews.

2.1. Protocol and Registration

Methodological decisions, eligibility criteria, data charting framework, and filtering rules were defined a priori at the outset of the work and consistently applied throughout the extraction and synthesis process; they are fully reported in the present Methods section to ensure transparency and reproducibility. The review protocol was not prospectively registered in a public registry such as the Open Science Framework or INPLASY. The absence of prospective protocol registration is acknowledged as a methodological limitation (Section 4.9). To offset this limitation, the complete dataset (nodes, edges, uncertain findings, and therapies) has been deposited on the Zenodo public repository under a CC BY 4.0 licence [84] (DOI 10.5281/zenodo.20832882).

2.2. Review Questions and Objectives

The review was structured around three core questions:

Q1. Which biological, clinical, and environmental entities are reported as components of asthma pathophysiology in the recent review literature, and with what level of consistency across independent sources?

Q2. Which mechanistic connections between these entities are described, and which roles (pathogenic, protective, mixed, therapeutic) are attributed to them?

Q3. What is the overall and systemic connectivity structure that emerges when these entities and connections are assembled into a single network, and which functional clusters can be identified within it?

2.3. Eligibility Criteria

Sources of evidence were eligible for inclusion if they met all the following criteria:

Publication type: review articles or systematic reviews (with or without meta-analysis).

Topic: asthma pathophysiology, immune mechanisms, endotypes or phenotypes, comorbidities, environmental triggers, clinical manifestations, or related pathogenic and protective mechanisms.

Publication window: 2021–2026, to reflect the most current understanding of asthma biology and to capture the impact of recent conceptual shifts (e.g., expansion of the alarmin axis, recognition of the epithelial barrier hypothesis, refinement of the T2 paradigm).

Language: English.

Accessibility: full text available in open access or through institutional access; sources for which full text could not be obtained despite these channels were excluded and are documented in the PRISMA flow diagram.

Evidence base: sources providing mechanistic or clinical evidence derived from human studies. Findings based exclusively on animal or cell-line models were excluded from data extraction. This exclusion is methodologically central and was enforced consistently throughout the extraction process to ensure that the resulting network reflects mechanisms with demonstrated human relevance, while acknowledging that much foundational asthma research originates from preclinical models.

2.4. Information Sources

Five electronic bibliographic databases were searched: PubMed/MEDLINE, Scopus, Web of Science, Embase, and Cochrane. Multi-database coverage was adopted to mitigate the known platform-specific retrieval biases of single-source searches and to ensure broad representation of the biomedical, clinical, and evidence-synthesis literature on asthma pathophysiology. Searches were conducted between September 2025 and March 2026. The date of the most recent search was 18 February 2026.

2.5. Search Strategy

Given the breadth of asthma pathophysiology and the exploratory nature of the review questions, an iterative block-search strategy was adopted, rather than a single compound Boolean query. This approach is explicitly supported by scoping review methodology [81,83] and was selected because it enables progressive refinement of the search as new thematic areas emerge from the reading of included sources.

Seventeen queries were executed, each filtered by article type (Review OR Systematic Review) and by publication date (2021–2026). The queries were designed to cover the main axes of asthma biology and clinical presentation. The full list of queries and the number of hits returned is reported in Table 1. Four additional string-based queries served as scoping probes during the search and their hits were fully included in the seventeen reported queries; they are therefore not listed separately to avoid double-counting.

2.6. Selection of Sources of Evidence

The selection process is summarized in the PRISMA 2020 flow diagram (Figure 1).

Title and abstract screening. The 15,623 records identified across the five databases were combined and screened by title and abstract against the eligibility criteria. Records whose focus on asthma pathophysiology, immune mechanisms, endotypes, comorbidities, or environmental determinants was clear from the title and abstract were retained; records duplicative of more recent and comprehensive sources within the same sub-topic, clearly outside the scope (e.g., epidemiological descriptive studies without mechanistic content, commentary or editorial pieces without primary review content, or narrowly clinical case series), or not review/systematic review in nature despite the filter were excluded. A total of 363 records were retained as potentially relevant after this step.

Full-text retrieval. Of the 363 records sought, 92 could not be retrieved in full text because they were not available in open access and no institutional subscription was available. These sources are listed in Online Resource 3 (ESM_3.xlsx), with their PubMed identifiers, to enable readers with institutional access to verify the non-inclusion and to consult them independently. The remaining 271 full-text reports were assessed for eligibility.

Full-text eligibility assessment. Twenty reports were excluded after full-text reading, of which 18 were identified as duplicates of already-included sources (partial index overlap between thematically adjacent queries that had escaped title-abstract deduplication) and 2 were off-topic after full reading. Aggregate counts by exclusion reason are reported in Online Resource 4 (ESM_4.xlsx); individual bibliographic entries for these excluded records were not retained during the review workflow, a limitation acknowledged in Section 4.9.

2.7. Data Charting Process

Data were charted into a structured multi-sheet spreadsheet that was iteratively adapted during the first forty source readings and then frozen. Each entry was manually verified by the author against the source article, with per-article controls of entry counts and systematic safeguards against extraction drift or errors. A large language model (Claude Opus 4.6, Anthropic) was used as a methodological collaborator for pattern recognition, synonym normalization, edge-list generation, internal consistency checks across sheets, and generation of the interactive HTML network visualization. The specific role of the language model is described in Section 2.13.

The charting framework comprises four linked sheets in the main database (Asthma_Network_Database.xlsx), plus a separate companion file (Asthma_Therapies_Database.xlsx) for therapeutic data, which answers a clinical rather than a mechanistic question. The file Asthma_Therapies_Database.xlsx contains a Therapies sheet of 767 entries with attributes including therapeutic category (e.g., biologic, inhaled corticosteroid, bronchodilator, non-pharmacological) and reported efficacy.

Nodes. Every biological, clinical, or environmental entity mentioned in each source was extracted as a node, with attributes including role (pathogenic, protective, mixed, therapeutic), biological level (1–4; see Section 2.8), applicable asthma subtype flags, the specific citation quoted from the source (with the original bibliographic reference number as it appears in the source, when provided), node priority (A/B/C; see Section 2.9), and source study.

Connections. Every mechanistic relationship described in each source between two or more entities was extracted as a narrative pathway (e.g., "IL-13 → SPDEF expression → ↑MUC5AC → Goblet cell hyperplasia"), with the same attributes as for nodes plus the directional structure of the pathway.

Uncertain findings. Findings reported with explicit caveats, contradictions between sources, or statements of ongoing controversy were extracted into a dedicated sheet rather than being included as edges in the network. These uncertain findings are complementary to the network: the network captures where the literature converges, whereas the Uncertain findings sheet captures where it diverges. This separation was a deliberate methodological choice: incorporating contradictory or unresolved findings as network edges would have introduced noise, blurring the signal of well-supported mechanisms. By keeping them in a parallel sheet, the productive frontier of the field remains fully documented and queryable, without compromising the interpretability of the network itself.

Summary. Database statistics and controlled vocabularies.

2.8. Data Items and Biological Level Framework

Each node was assigned a biological level reflecting its position in the pathophysiological hierarchy of asthma (Table 2). When a node spanned two domains (for example, an environmental agent acting through molecular mechanisms), the level associated with its dominant role in the network was retained.

The asthma subtype flags are Boolean fields indicating whether the entity was described in the cited source as applicable to the given subtype: T2-high/Allergic/Eosinophilic; T2-low/Neutrophilic; Severe asthma; Pediatric/Childhood; General (cross-phenotype). A single entity can be true for multiple subtypes; if an entity was described as applicable to both T2-high and T2-low asthma, it was reassigned to the General flag to avoid double-counting across mutually exclusive subtype categories.

Nodes originally extracted with variant names were normalized to canonical names through manual curation, using the shortest and most widely adopted form as canonical (e.g., "IL-13 (Interleukin-13)" and "IL-13" were merged into "IL-13"). Simplifications and assumptions made during this normalization step are documented in the project repository.

2.9. Node Priority Classification

To address the risk of network inflation — the inclusion of excessive numbers of nodes that reduce analytical interpretability without adding explanatory power — all nodes were classified into three priority tiers based on hierarchically applied criteria.

Priority A — Core network nodes. A node was assigned Priority A if it met at least one of the following:

Multi-study confirmation: the node appeared in three or more independent source articles with concordant biological role.

Hub structural role: the node functions as a convergence point for multiple pathophysiological modules, connecting at least two distinct biological domains (for example, innate immunity and remodeling, or environmental exposure and adaptive immunity).

Priority B — Secondary nodes. A node was assigned Priority B if it met at least one of: two-study confirmation with concordant biological role; bridge function connecting two otherwise disconnected modules even when supported by fewer than three studies (for example, a miRNA linking epithelial damage to ILC2 activation); or clinically relevant entity with documented association to disease severity, exacerbation risk, or treatment response.

Priority C — Peripheral nodes. A node was assigned Priority C if it met any of: intracellular intermediate within a well-characterized cascade without independent cross-module connections; isoform or sub-entity of a higher-order entity already represented (for example, individual collagen types I/III/V subsumed by ECM); single-source linear intermediary with a single unidirectional connection and no further ramification; purely technical or biomarker entity without direct pathophysiological role as a network effector.

Priority C nodes were retained in the full database for methodological transparency and for possible re-evaluation should future evidence accumulate, but were excluded from the final network figure.

2.10. Critical Appraisal of Individual Sources

Formal critical appraisal of the included sources was not performed, consistent with PRISMA-ScR guidance that makes appraisal optional for scoping reviews whose objective is to map the breadth of the evidence rather than to synthesize effect estimates [80,82]. The rationale for this decision is threefold. First, the aim of this review is descriptive: to map what the literature reports, not to grade its strength. Second, the included sources are themselves almost entirely secondary (reviews and systematic reviews), many of which already apply critical appraisal to their primary sources; a second layer of appraisal would be methodologically redundant. Even for narrative reviews without their own critical appraisal, the multi-source convergence requirement (≥3 independent sources per Priority A node) functions as a quality threshold that idiosyncratic claims cannot pass. Third, the network synthesis makes the strength of evidence operationally visible through the number of independent sources supporting each node and connection, which functions as an implicit, transparent appraisal signal that the reader can interpret directly from the data.

2.11. Synthesis of Results

The synthesis proceeded in three stages.

Stage 1. Entity normalization. Synonyms, abbreviations, and formatting variants were harmonized using a controlled vocabulary. For example, variants such as IL-13 / IL-13 (Interleukin-13), TH2 cells / Th2 cells, or ASM / Airway smooth muscle were fused into a single canonical label.

Stage 2. Edge construction. Narrative connections in the Connections sheet were algorithmically decomposed into directed edges (Source → Target). Where a pathway description contained intermediate entities not meeting inclusion criteria, the edge was preserved between the two nearest included entities (intermediate compression).

Stage 3. Network construction and descriptive metrics. The final network figure includes:

Nodes: all Priority A nodes (multi-study confirmation ≥3 independent sources, or hub structural role connecting distinct biological domains), together with the subset of Priority B nodes whose bridge function or clinical relevance justifies inclusion despite fewer than three supporting studies. Priority C nodes are retained in the full database (Section 2.9) but excluded from the network figure.

Edges: all relationships connecting two included nodes, reported in at least one included study.

This combination balances robustness of node identification — through multi-source convergence, which strengthens reliability and safeguards against idiosyncratic findings — with preservation of the mechanistic richness of the edge set. It also improves interpretability and visual readability, avoids network inflation, and keeps the architecture navigable without sacrificing the diversity of relationships reported in the literature. The threshold of one study for edges, as opposed to three for nodes, reflects the combinatorial nature of the edge space (the number of possible directed pairs among the 265 retained nodes is approximately 70,000): requiring multiple independent confirmations for each edge would have systematically eliminated genuine mechanistic patterns reported in specialized sources. The number of independent studies reporting each edge is retained as an attribute of the edge, allowing the reader to operationally distinguish a "core" robust sub-network from the full literature-derived network.

For each node, degree was computed as the number of incident edges in the final network. Thematic clusters were identified by neighborhood inspection around high-degree hubs, using the connectivity structure and role distributions as cues. A complementary community-detection algorithm was applied to the final network as a cross-validation step, identifying 15 fine-grained communities whose mapping onto the 10 narrative clusters is described in Section 3.5. Advanced centrality measures (betweenness, eigenvector, closeness) were not applied in this release and are identified as the natural next methodological step.

2.12. Contradiction Management

When an included source reported a finding that directly contradicted an established node or connection already in the database (for example, opposite directionality of regulation or opposite functional role under comparable conditions), the affected entity was transferred to the Uncertain findings sheet with an explicit description of the conflict, citations from all involved studies, and explanatory notes on the nature of the contradiction (for example, endotype-specific, context-dependent, dose-dependent). Nodes and connections supported by two or more concordant studies and no discordant study were designated validated network elements.

2.13. Use of Generative Artificial Intelligence

In accordance with the Springer Nature policy on the disclosure of generative artificial intelligence, the role of large language model assistance in the preparation of this work is reported here. A large language model (Claude Opus 4.6, Anthropic) was used as a methodological collaborator during the data charting and synthesis phases, specifically for: (i) assisting the extraction of entities and connections from included sources through iterative prompting and verification; (ii) constructing and refining the synonym dictionary for node normalization; (iii) formally verifying the edge list from narrative connections; (iv) assisting with the linguistic refinement of the manuscript; (v) generating and iteratively refining the interactive HTML network visualization code (D3.js-based) used to explore the final 265-node / 1,632-edge network. All inclusion decisions, eligibility judgments, charting choices, normalization rules, and network-construction thresholds were defined and approved by the authors. The language model was not used to screen records for eligibility, to perform critical appraisal, or to take final decisions on any content of the manuscript. All AI-assisted outputs were reviewed and verified by the authors, who take full responsibility for the content of this publication.

3. Results

3.1. Overview of the Included Literature

Of the 251 sources included in the scoping review, all were review articles or systematic reviews published between 2021 and 2026 that addressed asthma pathophysiology from at least one of the following perspectives: immune mechanisms and endotypes, epithelial and structural biology, comorbidities, environmental determinants, microbiome, obesity and metabolism, pediatric specificities, severe asthma, neuroimmune crosstalk, or therapeutic targets. Ten sources qualified as cornerstone integrative reviews that individually contributed entities or connections across more than four thematic axes of the network [3,8,18,28,29,33,36,37,74,85]. Twenty-seven sources were specifically focused on pediatric asthma or asthma inception across the life course, and 32 on severe, difficult-to-treat, or refractory asthma. Sixty-seven sources dealt wholly or primarily with environmental and climate determinants, 34 on microbiome and gut-lung axis, 26 on obesity and metabolic determinants, and 24 on neurological, psychological, or neuroimmune aspects of asthma.

Additional thematic clusters of included reviews focused on epithelial-barrier and mucosal mechanisms [86,87], T-cell heterogeneity [70,88], eosinophilic and mast-cell biology [89,90,91], atopic dermatitis as a comorbid axis [92,93,94,95,96], and air-pollution determinants [49,97,98,99,100].

3.2. The Asthma Network: Global Structure

After application of the inclusion criteria described in Section 2.11 (nodes of A-priority; nodes of B-priority; edges connecting two A- or B-priority nodes reported in at least one included study), the resulting network comprises 265 nodes and 1,632 edges (1,558 unidirectional and 74 bidirectional). The complete asthma network is shown in Figure 2. The full interactive version, with on-demand filtering by cluster, role, and subtype, is provided as Online Resource 1 (asthma_network.html, file name ESM_1.html).

Of 1,632 edges, 1,447 (88.7%) were classified as pathogenic, 158 (9.7%) as protective, 24 (1.5%) as mixed (i.e., reported with both pathogenic and protective effects across sources or contexts), and 3 (0.2%) as therapeutic. The prevalence of pathogenic edges reflects the dominant framing of the reviewed literature, which tends to report mechanistic connections in terms of disease-promoting pathways; protective edges predominantly concerned microbiota components, short-chain fatty acids, regulatory T cells, regulatory cytokines (IL-10, IL-37, TGF-β in specific contexts), breastfeeding, dietary components, and exercise-related pathways [25,43,101,102,103,104,105].

Of note, by examining the network, it is clear that protective mechanisms operate at multiple levels, reflecting a distributed, system-wide compensatory response that is partly mediated by protective environmental agents [106,107].

Edges are distributed heterogeneously across asthma subtypes (note that an edge may belong to more than one subtype if the supporting source describes the mechanism in several clinical contexts). T2-high/allergic/eosinophilic phenotypes accounted for 517 edges (31.7%), non-T2/neutrophilic for 186 (11.4%), severe asthma for 252 (15.4%), pediatric asthma for 285 (17.5%), and general or cross-phenotype for 1,085 edges (66.5%). The T2-high footprint in the network was approximately three times larger than the T2-low footprint, mirroring the long-standing asymmetry in the asthma literature, which has concentrated mechanistic attention on the type 2 axis and has relatively less well-characterized the non-type 2 pathways [6,17,20,21,22].

3.3. Hub Nodes: The Most Connected Entities

The 30 highest-degree nodes in the network represent the entities that integrate the largest number of mechanistic connections across the scoped literature. The hub list is provided in Online Resource 5 (ESM_5.xlsx). The hub distribution is consistent with a scale-free-like topology in which a small number of highly connected integrative nodes coexist with a large tail of sparsely connected peripheral entities [78,79]. The three leading hubs — eosinophils, asthma exacerbations, and IL-13 — operate at two different levels of biological description (cellular/molecular and clinical level), indicating that the network naturally integrates mechanistic and clinical tiers rather than separating them. Airway hyperresponsiveness (AHR) and FEV1 decline emerged as the principal functional-outcome hubs, confirming that the reviewed literature consistently routes mechanistic narratives towards these two physiological endpoints [3,9,108]. Airway barrier dysfunction ranked eighth, reflecting the recent conceptual shift towards an epithelium-centered view of asthma pathogenesis [37,38,109,110].

3.4. Thematic Clusters Emerging from the Network

Inspection of the neighborhood structure around high-degree hubs identified ten thematic clusters that cover the entire connectivity of the network. Each cluster groups nodes share dense internal connectivity and biological coherence, while maintaining inter-cluster edges that define the overall integrated architecture of the network.

The interactive HTML visualization (Online Resource 1) displays 15 fine-grained communities identified by unsupervised community-detection. For the present narrative synthesis, we group these into 10 thematic clusters guided by biological coherence: high-degree hubs were used as anchors and neighborhoods inspected for thematic affinity. The mapping between the visual network communities and the narrative clusters is provided in the next paragraphs of this section. Two visual communities ("Innate regulatory" and "Epigenetic & signaling") do not map onto a single narrative cluster but distribute across multiple ones (notably Clusters 2, 3, and 4), reflecting the cross-cutting nature of regulatory and signaling nodes in the network.

The ten clusters are presented in an upstream-to-downstream order following the epithelium-to-outcome axis described above: from the host-environment interface (Cluster 1) and the two immune cascades it activates (Clusters 2 and 3), through the structural consequences (Cluster 4) and systemic modulators (Clusters 5-8), to the comorbid extensions (Cluster 9) and the clinical manifestations (Cluster 10).

Cluster 1 — Epithelial barrier and alarmin axis. It integrates the airway epithelium, its structural constituents (tight junction proteins, E-cadherin), damage-associated molecular patterns (HMGB1, ATP, uric acid), and the three canonical alarmins (TSLP, IL-25, IL-33), together with their receptors and the epithelial-derived cytokines IL-1β, IL-6, and GM-CSF. Epithelial dysfunction is reported as the upstream initiator of most downstream inflammatory cascades in the network [18,37,38,39,108,109,110,111,112,113,114]. Edges from this cluster fan out densely into the T2 adaptive axis (Cluster 2), into the non-T2/neutrophilic axis (Cluster 3), and, critically, into both the environmental (Cluster 7) and microbiome (Cluster 5) clusters, indicating that the epithelium operates as the principal point of integration between host and environment [86,87].

Cluster 2 — Type 2 adaptive and effector immunity. It is the densest and most centrally located cluster in the network. It encompasses TH2 cells, ILC2, the canonical cytokines IL-4, IL-5, IL-9, and IL-13, immunoglobulin E and its high-affinity receptor, and the downstream effector cells (eosinophils, mast cells, basophils) together with their mediators (histamine, cysteinyl leukotrienes, prostaglandin D2, major basic protein, eosinophil peroxidase, eosinophil cationic protein, eosinophil extracellular traps) [17,18,36,67,68,71,89,115]. The heterogeneity of T-cell subsets contributing to type 2 immunity beyond the canonical TH2 framework — including TFH13, pathogenic TH2A, and TH2/TH17 dual-positive cells — has been increasingly characterized [69,70,88], and the involvement of mast cells in this cluster extends to mast cell activation syndromes that overlap with severe asthma phenotypes [91]. Within this cluster, eosinophils function as the integrative super-hub, connecting upstream T2 signaling with structural and functional downstream effects (airway remodeling, AHR, exacerbations, FEV1 decline), with novel perspectives on eosinophil biology in severe asthma continuing to emerge [90]. The cluster extends into the comorbidity cluster via allergic rhinitis, chronic rhinosinusitis with nasal polyps, eosinophilic esophagitis, and atopic dermatitis [66,116,117]. The therapeutic monoclonal antibodies targeting IgE, type 2 cytokines, and alarmin pathways — discussed in detail in Section 3.5 — also operate primarily within this cluster [118].

Cluster 3 — Non-type 2 and neutrophilic pathways. It organizes around neutrophils, TH17 cells, IL-17 family cytokines (IL-17A, IL-17F, IL-22), IL-6, IL-8 (CXCL8), IL-1β, and IL-23. Additional nodes include NETs, NLRP3 inflammasome activation, pyroptosis, and M1 macrophages [8,16,20,21,22,119,120,121]. It has fewer edges than Cluster 2 (257 vs 471 edges), but disproportionate connectivity to severe asthma and steroid-resistance nodes, consistent with the clinical observation that non-T2 asthma is both under-represented in the mechanistic literature and over-represented in treatment-refractory phenotypes. Strong edges to obesity (Cluster 6), tobacco smoke, and gut-lung axis components indicate that non-T2 inflammation is embedded in a broader systemic context rather than being an intrinsically airway-confined phenomenon. Macrophage polarization toward an M1 phenotype under air-pollution exposure provides a mechanistic bridge to Cluster 7, where particulate matter drives non-T2 inflammation through alveolar macrophage activation [122].

Cluster 4 — Airway remodeling and structural alterations. It groups airway smooth muscle (ASM), fibroblasts, myofibroblasts, extracellular matrix components (collagens, fibronectin, MMP-9, TIMPs), angiogenesis, goblet cell metaplasia, MUC5AC, epithelial-mesenchymal transition, basement membrane thickening, and subepithelial fibrosis [8,38,73,123,124,125,126,127]. This cluster is both the downstream convergence point of the T2 and non-T2 inflammatory axes and the substrate of long-term functional decline (FEV1 loss, fixed obstruction, ACO development). Edge density between this cluster and severe asthma is especially high, reflecting the consensus that structural remodeling is the principal biological substrate of irreversible airflow limitation [4,28,128]. Small airways dysfunction emerges as a specialized sub-hub [123,124]. Mechanosensitive calcium-regulatory pathways in ASM and the dynamics of immune senescence in chronic remodeling are additional layers recently characterized within this cluster [129,130].

Cluster 5 — Microbiome, gut-lung axis, and mucosal immunology. It comprises airway and gut microbiome composition, short-chain fatty acids (butyrate, propionate), secretory IgA, mucosal tolerance, gut-lung axis signaling, and early-life modifiers (C-section delivery, breastfeeding, formula feeding, antibiotic exposure, farm exposure, hygiene hypothesis constructs) [40,41,43,74,75,131,132,133,134,135,136]. The role of secretory IgA in mucosal tolerance is increasingly recognized as a major determinant of the protective output of this cluster [137], and host genetics interact with microbiome and environmental exposures through pathways that are still being mapped [138]. Probiotic-targeted interventions have emerged as a candidate therapeutic axis based on the dysbiosis-inflammation-oxidative stress link [139,140]. The gut mycobiome adds a parallel dimension, with airway mycosis contributing to non-canonical type 2 responses [141] and biofilm-mediated immune dysregulation representing a poorly characterized but mechanistically relevant phenomenon [142]. This cluster is a major source of protective edges (84 of 158 protective edges, ~53%). Respiratory viral infections, particularly rhinovirus and RSV, sit at the interface between this cluster and the exacerbation node [143,144,145,146,147]. The extensive recent literature on the gut-lung axis converges on dysbiosis-driven immune dysregulation as a unifying framework, with implications extending to the gut-lung-vascular axis in asthma-cardiovascular comorbidity [148,149,150,151].

Cluster 6 — Obesity and metabolic modifiers. Obesity and metabolic syndrome appear as a macro-hub coordinating adipokines (leptin, adiponectin), insulin resistance, type 2 diabetes, dyslipidemia, metabolic inflammation markers (CRP, IL-6), M1 macrophage polarization, and reduced responsiveness to inhaled corticosteroids [12,47,48,152,153,154,155,156,157,158]. The adipokine signaling axis — leptin acting as pro-inflammatory mediator, adiponectin as counter-regulator — is increasingly recognized as a node-specific therapeutic target [159,160], while lipid and sterol metabolism contribute independent layers of immune regulation within this cluster [161,162]. This cluster is characterized by a distinctive edge pattern: most connections are to non-T2/neutrophilic nodes (Cluster 3) rather than to the classic T2 axis (Cluster 2), and strong edges to mechanical-physiological nodes (dysanapsis, reduced lung compliance, EIB) and to the environmental cluster via air pollution synergy. This supports interpretation of obesity as a cross-cutting modifier acting on non-allergic pathways rather than a simple additive risk factor. In pediatric obese asthma, comorbidities cluster differently from adult obese asthma, suggesting a distinct sub-phenotype that requires dedicated stratification [163,164].

Cluster 7 — Environmental and ecological determinants. The widest cluster by range of entities, including outdoor air pollutants (PM2.5, PM10, NO2, ozone, diesel exhaust particles), indoor air pollutants (volatile organic compounds, bisphenol A, cleaning agents), tobacco smoke (active, second-hand, prenatal), e-cigarettes, pollen (amplified by climate and thunderstorm asthma), occupational agents, housing typologies, extreme weather events, climate change variables, and social determinants of health [49,50,51,52,53,165,166,167,168,169,170,171]. Outdoor air pollution acts both on incidence and on exacerbation rate, with consistent associations across geographic and demographic settings [99,172], and its impact on pediatric and childhood asthma is particularly well documented [98,100,173,174]. Indoor air pollution — volatile organic compounds, particulate matter, and biomass combustion products — exerts an independent burden, especially in vulnerable groups and in school environments [175,176,177,178,179]. Edges from this cluster dominantly target epithelial barrier dysfunction (Cluster 1) and epigenetic modifications, which then reverberate throughout the network [180]. Gene-environment interactions (GSTM1, CDHR3, 17q12-21, NRF2, aryl hydrocarbon receptor signaling) constitute an internal bridge to the structural/inflammatory clusters [181,182,183]. Beyond pollution, extreme weather events and climate change variables increasingly contribute to acute exacerbations [184,185], while housing typologies modulate chronic exposure burden [186], and physical activity-pollution interactions add a further layer of complexity in youth populations [187]. Air pollution also intersects with the alveolar compartment [188], with broader dysbiosis–lung disease pathways [44] and with multiple chemical sensitivity, a distinct but related entity in which environmental exposures drive systemic neuroimmune dysregulation [189,190]. Social determinants of health — environmental justice inequities, neighborhood-level exposure differences, and violence-related chronic stress — interact with biological pathways and shape both prevalence and severity [54,60,191]. Resolution mechanisms of pollution-induced inflammation represent a complementary therapeutic dimension still under characterization [192].

Cluster 8 — Neuroimmune and psychological axis. It groups airway neuroendocrine cells, neuropeptides (substance P, CGRP, VIP), neurotransmitters (acetylcholine, serotonin, dopamine, GABA, histamine acting neurally), the lung-brain axis, vagal afferent signaling, blood-brain barrier disruption, systemic low-grade neuroinflammation, anxiety and depression, alexithymia, breathlessness perception dissociated from bronchoconstriction, and circadian rhythm disruption [55,57,193,194,195,196,197,198,199,200,201]. Dysfunctional breathing patterns, particularly in pediatric populations, add a further behavioral layer to this cluster [202], and psychological interventions are emerging as candidate disease-modifying tools for adolescents with asthma [203]. Although the number of edges in this cluster is lower than in Clusters 1–7, the edge density relative to the number of nodes is comparable, indicating a well-delineated subsystem. The cluster shows a distinctive pattern of bidirectional connections to exacerbations and to symptom nodes (dyspnea, cough), suggesting that neuroimmune circuits modulate perception and behavioral response to airway inflammation in addition to participating in the inflammation itself. The mechanistic interface between neuroimmune signaling and airway remodeling represents a further axis of integration within this cluster [204].

Cluster 9 — Comorbidities and systemic associations. It organizes the major asthma comorbidities into a single integrated component: allergic rhinitis, chronic rhinosinusitis with or without nasal polyposis, atopic dermatitis, eosinophilic esophagitis, gastroesophageal reflux, obstructive sleep apnea, obesity (linking to Cluster 6), cardiovascular disease and hypertension, chronic kidney disease, osteoporosis, type 2 diabetes, anxiety and depression (linking to Cluster 8), asthma-COPD overlap, and food allergy [61,62,63,64,65,205,206,207,208,209,210]. In pediatric populations, the comorbidity profile differs from adult asthma and includes both allergic and non-allergic conditions clustering together [211,212]. Atopic dermatitis represents a particularly tight node, both as a co-existing T2 disease and as the upstream step of the atopic march, with distinct adult and pediatric manifestations [92,93,94,95]. Allergic rhinitis [213], cardiovascular comorbidity through coronary artery disease and IL-33/ST2-driven heart failure [214,215,216], and osteoporosis as a long-term skeletal complication [217] further illustrate the systemic reach of the cluster. The key observation is that comorbidities are not peripheral to the main network but deeply integrated: their edges terminate predominantly on core mechanistic nodes (T2 cytokines, epithelial barrier, airway barrier dysfunction, chronic low-grade inflammation, microbiome dysbiosis), not on separate clinical endpoints. This supports the reframing of asthma comorbidities as shared-mechanism conditions rather than coincidental co-occurrences. This integrated topology supports the case for moving beyond compartmentalized disease frameworks toward a systems-level representation, in which asthma and its comorbidities are approached as interconnected expressions of a shared mechanistic substrate. The network synthesis presented here provides the operational scaffold for such an integrated view.

Cluster 10 — Clinical expression and functional outcomes. Groups the clinical-functional manifestations: AHR, bronchoconstriction, FEV1 decline and airflow obstruction, exacerbations, wheezing, cough, dyspnea, chest tightness, sleep disturbance, exercise-induced bronchoconstriction, and patient-centered outcomes (fatigue, work impairment, quality of life) [3,9,25,108,218,219,220,221,222]. Receives the largest number of incoming edges (approximately 19% of all edges in the network). Its internal structure mirrors the heterogeneity of asthma presentations: pediatric-specific nodes (recurrent wheeze, early-onset phenotypes) and adult-specific nodes (fixed airflow limitation, late-onset eosinophilic phenotype, airway remodeling-dominant phenotypes) form partially separate sub-clusters within this final common pathway.

3.5. Companion Catalogues: Uncertain Findings and Therapies

The Uncertain Findings sheet of the master dataset contains 624 entries catalogued as reported with explicit caveats, inter-study contradictions, or ongoing debate. Additionally, 24 edges in the network (1.5%) were classified as mixed, representing entities or connections for which the reviewed literature reports both pathogenic and protective effects. These are discussed in Section 4.3.

The Therapies companion file (Asthma_Therapies_Database.xlsx) contains 767 entries kept separate from the network proper, consistent with the scoping decision to focus the network on pathogenic and protective biology. Each therapy entry is linked to one or more network nodes that it targets. Six therapeutic categories emerged: (1) inhaled corticosteroids and bronchodilators; (2) biologics directed at the T2 axis (anti-IgE: omalizumab; anti-IL-5/IL-5R: mepolizumab, reslizumab, benralizumab; anti-IL-4Rα: dupilumab; anti-TSLP: tezepelumab) [76,77,223,224,225,226,227,228]; (3) emerging non-T2-targeted therapies (anti-IL-6, anti-IL-17, JAK inhibitors, anti-alarmin combinations); (4) non-pharmacological interventions, including breathing techniques, pulmonary and exercise rehabilitation, dietary interventions targeting type 2 inflammation [229], broader nutritional approaches based on bioactive compounds [230,231], mineral and micronutrient supplementation including zinc and vitamin D [232,233,234], probiotics and prebiotics, and weight reduction in obese asthma; (5) comorbidity-directed treatments (CPAP for OSA-associated asthma, proton pump inhibitors for GERD-associated asthma, dupilumab for concomitant atopic dermatitis, allergen immunotherapy for allergic rhinitis-asthma); (6) treatable-traits-based personalized approaches [29,235,236,237]. Notably, approximately 72% of all therapeutic entities in the catalogue converge on nodes within Clusters 1 and 2 (epithelial/T2), with non-T2 and cross-cutting clusters substantially underserved by the current pharmacological pipeline.

3.6. Putting It All Together: An Integrated Systems-Level View of Asthma

Several structural observations follow directly from this integrated representation, independent of any further analysis. The network is connected: every entity in it links, directly or indirectly, to every other, indicating that the contemporary asthma literature, despite its surface fragmentation, can be assembled into a single coherent map. The 30 highest-degree hubs span four biological levels (cells, cytokines, structural alterations, clinical endpoints) and seven of the ten thematic clusters, indicating that asthma pathogenesis is integrated across scales rather than confined to one molecular axis. The principal organizing axis is epithelium-to-outcome rather than T2-vs-non-T2: the epithelial-alarmin cluster operates upstream as the integrator of environmental signals, the clinical-functional cluster operates downstream as the convergence point of all upstream pathways, and the T2 and non-T2 cascades appear as parallel transmission channels between these two poles. Comorbidities, microbiome, metabolism, and environmental determinants connect to core mechanistic nodes in depth rather than peripherally, indicating that they are constitutive components of asthma pathobiology rather than coincidental modifiers.

A further structural observation, made visible only by the integrated representation, concerns the existence of cross-cutting modules: computational community-detection on the network identified two fine-grained clusters — "Innate regulatory" and "Epigenetic & signaling" — that do not map onto a single thematic narrative cluster but rather distribute across the type 2 (Cluster 2), non-type 2 (Cluster 3), and remodeling (Cluster 4) axes. This pattern reflects the cross-cutting role of immune-regulatory cells (Tregs, M1/M2 macrophages, dendritic cells, IL-10) and of intracellular signaling-epigenetic machinery (NF-κB, STAT3, HDAC, MAPK, 17q12-21) as integrators across the canonical inflammatory subtypes. The detection of such modules — entities whose biological function is precisely that of operating across compartments — is a representative example of an emergent property accessible to the network synthesis approach but invisible, by construction, to compartmentalized reviews of a single axis.

4. Discussion

4.1. Principal Findings

This scoping review and network synthesis of 251 recent review-level sources produced an integrated map of asthma pathophysiology comprising 265 robustly reported entities and 1,632 mechanistic connections, organized around 30 highly connected hubs and ten coherent thematic clusters. Three principal findings stand out.

First, the asthma literature of the last five years, despite its apparent fragmentation, is internally coherent and can be assembled into a single connected graph without isolated components: every identified entity connects, directly or indirectly, to every other. This is neither trivial nor self-evident, given the historical separation between molecular immunology, epithelial biology, environmental health, microbiology, metabolism, and clinical phenotyping.

Second, the topology of the resulting network is compatible with a scale-free-like architecture [78,79], in which a small number of super-hubs integrate most of the connectivity while a long tail of specialized entities contributes local mechanistic detail. The top hubs do not belong to a single biological level: cells, cytokines, structural alterations, functional outcomes, environmental factors, and systemic conditions coexist in the top 30, demonstrating that the reviewed literature consistently crosses biological scales when describing asthma.

Third, the clustering structure departs from the canonical T2/non-T2 dichotomy in an important way. While Clusters 2 (T2 adaptive-effector) and 3 (non-T2/neutrophilic) emerge as distinct components, they do not function as mutually exclusive descriptions of the disease. The dominant flow pattern of the network supports an epithelium-to-outcome organizing axis: the clinical-functional cluster (Cluster 10) receives approximately 90% of its inter-cluster connections; the epithelial-alarmin cluster (Cluster 1) operates as a central integrator at the host-environment interface; and the T2 cluster functions as the dominant transmission channel, projecting 76% of its inter-cluster edges outward, with the densest projection toward the clinical-functional cluster.

Examined directly, T2 and non-T2 exchange only 10 reciprocal edges in the entire network, none of which are inhibitory. If the two cascades were truly mutually exclusive, the network would display inhibitory crosstalk between them; instead, the two clusters do not inhibit one another but ignore one another topologically. The two cascades therefore behave as biologically separable pathways sharing an upstream interface (the epithelium) and a downstream convergence point (the clinical-functional cluster), but operating without reciprocal interference along the way. This is consistent with the increasing clinical recognition of mixed-inflammation phenotypes, in which T2 and non-T2 inflammation coexist within the same patient.

Such absence of inhibitory exchange is not visible from compartmentalized reviews of either axis alone, which describe each cascade internally but cannot represent the structural relationship between them; the integrated network makes this absence measurable rather than inferred, updating the canonical dichotomic view of asthma without contradicting the underlying biology of either cascade.

4.2. The Emergent Architecture of Asthma: An Integrated View

The network makes explicit a view that the recent literature has approached but rarely formalized: asthma is organized as a multi-axis system in which at least six functionally distinct but mechanistically interconnected subsystems converge on a shared airway substrate. These subsystems are (i) the epithelial-alarmin interface with the environment; (ii) the adaptive-effector type 2 arm; (iii) the innate and non-T2 arm; (iv) the structural-remodeling arm; (v) the ecological arm comprising microbiome, early-life determinants, and mucosal immunity; (vi) the systemic-metabolic arm integrating obesity, metabolism, and systemic inflammation. A seventh subsystem, the neuroimmune axis, modulates and is modulated by the others. An eighth, the comorbidity axis, shares mechanistic substrates with all the others and is therefore best understood as an integrated component of asthma pathogenesis rather than a set of associated diseases [61,62,63,64].

The connectivity pattern suggests three general principles of asthma organization that are visible only in a network representation.

Principle 1 — The epithelium is the entry point, not only a damaged tissue. Every major external input (pollutants, allergens, viruses, microbiome shifts, tobacco smoke, ozone, diesel particles, occupational agents) reaches the rest of the network through the epithelial cluster. Airway barrier dysfunction ranks seventh among hubs despite being a relatively new conceptual entity, and the three alarmins (TSLP, IL-25, IL-33) and their downstream amplifier ILC2 are all in the top 30. This is consistent with the epithelial barrier hypothesis articulated in recent years [18,37,38,109,110] and reframes the epithelium from a passive target to the principal translator between environment and immune response.

Principle 2 — T2 and non-T2 are parallel channels, not opposed categories. As shown above, the two clusters exchange minimal direct connectivity (10 reciprocal edges, none inhibitory) but share common upstream inputs from the epithelial-alarmin cluster and common downstream outputs into the clinical-functional cluster. Several hubs (IL-33, IL-6, NF-κB, TNF-α, neutrophils in severe T2 asthma, eosinophils in some non-T2 contexts, steroid resistance) participate in both, and so-called paucigranulocytic asthma, historically the leftover phenotype, emerges in the network as a state of low activity on both channels rather than as a distinct endotype [16,22,23]. Mixed-type inflammation is the rule rather than the exception when both channels are simultaneously active.

Principle 3 — Comorbidities, environment, microbiome, and metabolism are not modifiers: they are constitutive. Their edges terminate on core mechanistic nodes in depth, not superficially. Removing these clusters from the network would reduce the degree of most top hubs by 30–50% and would disconnect major downstream outcomes. Interpreting the obese-asthma phenotype, the air-pollution-associated asthma, or the early-life microbiome-associated asthma as simple modifiers of an underlying asthma substrate underestimates their structural role.

Taken together, these three principles support a reading of asthma as a network-level adaptive reorganization of an integrated airway-immune-epithelial-metabolic-ecological system under environmental and systemic pressure, rather than a disease with one dominant mechanism plus contextual modifiers. This reading is compatible with, and gives an operational scaffold to, the treatable-traits framework [29,235,236] and with the growing consensus that asthma endotypes are overlapping functional states on a continuum rather than discrete categories [13,14,238].

4.3. Principal Uncertainties and Unresolved Findings

The 624 uncertain findings catalogued in the dataset represent the productive frontier of the field, where contemporary asthma research is still actively debating, and where future studies are most likely to reshape current understanding. Seven principal clusters of uncertainty emerge with particular clinical and scientific salience.

(i) Bidirectional and context-dependent mediators. Twenty-four edges were classified as mixed-role, reflecting explicitly contradictory reports of pathogenic versus protective effects across sources. Four specific mediators recur across these edges. Prostaglandin E2 and the eicosanoid axis exert opposing effects on airway smooth muscle tone, eosinophil recruitment, and airway remodeling depending on receptor subtype expression (EP2/EP4 mediate bronchodilation; DP2/CRTH2 are pro-inflammatory) [166,239,240]. Interferon signaling shows context-dependent roles in asthma exacerbations, viral defence, and Th2/Th17 balance: defective type I interferon responses during rhinovirus exacerbations are well documented, but whether augmenting IFN signaling is therapeutically beneficial depends critically on disease phase and dominant endotype [40,143,144]. Serotonin and other neuroimmune mediators (dopamine, GABA, substance P) are reported with opposite effects on ASM, eosinophils, and Th2 polarization, reflecting the recent emergence of the neuroimmune cluster as a relevant target of asthma biology [61,193,198]. IL-13 on the epithelium is classified as mixed because IL-13 induces goblet cell metaplasia and mucus hypersecretion while also promoting epithelial repair and barrier-related gene expression in specific contexts [17,37,226,241]; this has direct clinical implications for long-term use of anti-IL-13 and anti-IL-4Rα biologics.

(ii) Directionality of the obesity-asthma association. Whether obesity causes asthma, asthma promotes weight gain, or both share upstream determinants remains incompletely resolved, particularly across the pediatric and adult divide. Pediatric obese asthma and adult-onset obese asthma appear to involve distinct mechanistic signatures [31,242,243,244], suggesting that the "obese-asthma phenotype" is itself heterogeneous. The directionality question has direct therapeutic consequences: whether weight reduction should be considered a disease-modifying intervention or a symptomatic adjuvant depends on which direction of causation predominates.

(iii) Aetiology and boundaries of paucigranulocytic asthma. Paucigranulocytic asthma, historically the leftover phenotype, remains ill-defined mechanistically [16,22]. The network shows it as a state of low activity on both T2 and non-T2 channels rather than a distinct endotype. Whether this represents a genuinely distinct biological entity, a transient inter-phenotype state, or a classification artifact is unresolved, with direct implications for biologic selection and clinical trial stratification.

(iv) Bidirectional microbial taxa. Although the network captures Haemophilus influenzae, Moraxella catarrhalis, and Streptococcus pneumoniae predominantly as pathogenic drivers of exacerbations, type 2 cytokine induction, and neutrophilic inflammation [40,41,104], several included reviews note that the same taxa can act as protective commensals or as neutral bystanders in specific contexts. The network surfaces this ambiguity directly: intranasal pneumococcal conjugate vaccine (but not polysaccharide vaccine) expands Tregs in lymph nodes, lungs and spleen, suggesting that controlled exposure to S. pneumoniae antigens has immunoregulatory potential [104]; an inverse co-occurrence between Moraxella and Corynebacterium [75] suggests context-dependent rather than intrinsically pathogenic roles. Candida and other gut fungi show a similar pattern, predominantly associated with IL-4, ILC2, and IgE elevations [245,246,247], but counter-regulated by specific Lactobacillus strains producing fungicidal compounds [101]. The resolution of these context-dependent roles requires longitudinal, strain-resolved, and niche-aware microbiome studies, with strict control for age, antibiotic exposure, dose, and disease phase [75].

(v) Causality and interpretation of airway microbiome changes. Whether airway microbiome alterations reflect consequences of reduced airflow, disease biology, or prescribed therapies (selective pressures), versus active contribution to phenotypic heterogeneity, remains unresolved [75]. Unidirectional causality is unlikely in chronic disease; longitudinal and interventional studies, not cross-sectional association studies, are needed to disentangle this.

(vi) Biologic response prediction, remission criteria, and long-term safety. No universally accepted criteria for clinical remission on biologics currently exist: at least four distinct definitions (Menzies-Gow, ACAAI, SANI, SEPAR) coexist with varying thresholds for ACT/ACQ, lung function, inflammatory markers, and duration [237]. Significant improvements are observed in the placebo arms of pivotal trials, with reported placebo remission rates of 7.7% in QUEST and 21.9% in NAVIGATOR [223,237]. Moreover, the safety of biologics other than omalizumab during pregnancy remains poorly characterized: although the EXPECT registry (n=250) supports the gestational safety of omalizumab, human data on mepolizumab, reslizumab, benralizumab, dupilumab, and tezepelumab in pregnancy are essentially absent [196,223]. Long-term safety signals require continued monitoring, including a documented decline in mepolizumab-associated FEV1 improvement over 200 weeks [224], theoretical concerns about anti-tumorigenic eosinophil depletion [19], and the cardiovascular signal observed in the EXCELS cohort for omalizumab [65]. Together these gaps represent a critical limitation given the chronic nature of asthma.

Together, these six clusters of uncertainty define where the next wave of asthma research is likely to be most informative. The network-based representation makes them not only visible but queryable and updatable as new evidence accrues.

(vii) Exercise as protective hub versus acute trigger. The network captures exercise as one of the most consistently protective nodes, with 9 of 10 edges classified as protective, targeting AHR, airway remodeling, eosinophils, FeNO, FEV1, exacerbations, anxiety, depression, sleep, and breathlessness, with documented reductions in ICS requirement after sustained aerobic training [5,25,105]. Yet exercise can acutely trigger bronchoconstriction through hyperpnea-induced epithelial dehydration, mast cell degranulation, and release of histamine and cysteinyl leukotrienes [12,25]. The two roles operate on different time scales: acutely, exercise can precipitate EIB in patients with elevated baseline AHR; chronically, regular training reduces that same baseline AHR. The unresolved question is how to stratify EIB-susceptible patients against the much larger group for whom exercise should be actively encouraged as a disease-modifying intervention — a stratification that the integrated network shows is decisively skewed toward the protective role across the reviewed evidence.

4.4. Clinical and Translational Implications

Five implications follow from the network analysis. Hub-informed stratification: the hub list suggests a stratification strategy based on which hubs are dominantly active in a given patient, complementary to T2/non-T2 but going beyond it (e.g., epithelial-alarmin-dominant, obesity-metabolic-dominant, microbiome/early-life-dominant, remodeling-dominant, or mixed profiles). Therapy gap in non-T2 and cross-cutting clusters: approximately 72% of current therapies converge on T2 and epithelial clusters, while non-T2 asthma, remodeling-dominant asthma, obese asthma, and neuroimmune-driven symptom burden remain substantially underserved [8,16,20,22,46]. Treatable traits as a network-compatible framework: the treatable-traits approach [29,235,248] can be interpreted in network terms as the clinical recognition of active sub-clusters in the individual patient and the targeting of their respective hubs. Comorbidity as co-pathogenesis: the integrated position of the comorbidity cluster argues against managing asthma and its comorbidities in parallel silos; allergic rhinitis, chronic rhinosinusitis, atopic dermatitis, gastroesophageal reflux, obstructive sleep apnea, obesity, anxiety and depression share mechanistic nodes with core asthma biology and should be screened and treated as part of the same clinical problem. Systemic therapeutic targets: the network architecture invites the identification of therapeutic targets that operate not by blocking a single component of a molecular chain but as systemic modulators of the entire network, including comorbidities and inter-axis bridges. The cross-cutting modules ("Innate regulatory" and "Epigenetic & signaling") identified by community-detection in Section 3.5 are natural candidates for such targets, as their position across multiple inflammatory subtypes makes them potential leverage points for multi-axis intervention.

4.5. Pediatric Specificities, Life-Course, and Novel Mechanisms

The pediatric footprint in the network (285 pediatric edges) shows distinctive connectivity: respiratory viruses, early-life microbiome exposures, maternal and perinatal factors, environmental exposures during critical developmental windows, and atopic march are disproportionately represented. Preschool wheeze endotypes [33,249], pediatric severe asthma [31], obesity-related pediatric asthma [242,250], and national consensus on pediatric asthma management [251] each constitute well-defined sub-cluster signatures. Comorbidity-based endotype mapping has been independently proposed as a methodological precedent for the network approach used here [252]. The life-course perspective [35,128] integrates these pediatric-specific features with adult outcomes, supporting the view that trajectories, rather than cross-sectional phenotypes, should guide longitudinal management. Recently characterized mechanistic nodes — pyroptosis [253,254], ferroptosis [255,256,257], NLRP3 inflammasome activation, autophagy dysfunction, mitochondrial dysfunction [258], sterol and lipid metabolism, metabolomics-defined asthma signatures [159,259,260], extracellular vesicles [261], complement activation [262], and circadian rhythm regulation [263] — appear across multiple clusters as already-connected mechanistic bridges.

4.6. Strengths and Limitations

The main strengths of this work are: (i) Originality of the integrated representation. To our knowledge, this is the first literature-derived mechanistic network of human asthma to assemble, on a single connectivity scaffold and with explicit source attribution, the molecular, cellular, tissue, organ, clinical, environmental, microbial, metabolic, and neuroimmune dimensions of the disease. Existing reviews of asthma biology either concentrate on one mechanistic axis (e.g., type 2 cytokines, epithelial barrier, microbiome) or compile narrative syntheses without producing a structured queryable object. The present network spans all reviewed axes and renders them simultaneously interrogable.

(ii) Semantic richness of the connectivity. Each edge carries explicit metadata that goes beyond co-occurrence: direction (→ or ↔), functional role (pathogenic, protective, mixed, or therapeutic), asthma subtype attribution, and source-traceable citation. To our knowledge, no prior asthma network synthesis has combined these four semantic dimensions; this transforms the network from a topological diagram into an interpretable structured representation in which the meaning of every connection, not just its presence, is queryable.

(iii) Multilevel architecture. Every node carries an explicit biological level (molecular/cellular, pathophysiological/systemic, clinical, environmental), so that connections crossing scales — from environmental exposure to molecular signaling to tissue remodeling to clinical outcome — are made structurally visible. This contrasts with conventional network representations that operate within a single biological scale.

(iv) Scale and breadth of the scoped literature. The 251 included sources span the full spectrum of asthma biology over the most recent five-year period (2021–2026), capturing the contemporary reconceptualization of the disease (epithelial barrier hypothesis, alarmin axis expansion, comorbidity reframing, gut-lung axis, neuroimmune circuits).

(v) Strict exclusion of exclusively animal or cell-line evidence. The network reflects mechanisms with demonstrated human relevance — a methodological choice consistent with the documented translational gaps observed across multiple anti-cytokine biologic classes that succeeded in murine models but failed in human trials.

(vi) Companion catalogues of uncertain findings and therapies. The dedicated repository of 624 uncertain entries and 767 therapy entries makes the productive frontier of the field, and the current pharmacological landscape, simultaneously visible and queryable alongside the consolidated network. The complete dataset is publicly deposited on Zenodo, ensuring full independent verifiability and reuse.Four limitations apply: (1) No prospective protocol registration. Methodological decisions were defined a priori and consistently applied throughout the workflow, but not registered in OSF, INPLASY, or PROSPERO. To partially offset this limitation, the complete dataset has been deposited on the Zenodo public repository. (2) Incomplete full-text access. Ninety-two potentially relevant sources were not retrievable in full text and were therefore excluded (Online Resource 3). Although the proportion is non-trivial (approximately 25% of sources sought for retrieval), the records are individually traceable via their PubMed identifiers, allowing readers with institutional access to consult them independently. (3) Review-derived inheritance. The network inherits the biases of the reviewed literature: well-funded topics are over-represented, while emerging or clinically descriptive topics — particularly those covering symptoms and the modalities of their expression — are comparatively under-represented. (4) Descriptive metrics only. Node priority and thematic clusters were defined by inclusion rules and neighborhood inspection around high-degree hubs; formal community detection algorithms [78,79] and quantitative centrality measures beyond degree were not applied in this release and constitute the natural next methodological step. Individual bibliographic entries for the 20 records excluded at full-text review (18 duplicates and 2 off-topic) were not retained during the workflow; aggregate counts are reported in Online Resource 4.

5. Conclusions

This scoping review introduces, to our knowledge, the first literature-derived mechanistic network of human asthma to assemble, on a single connectivity scaffold and with explicit source attribution, the molecular, cellular, structural, clinical, environmental, microbial, metabolic, neuroimmune, and comorbidity dimensions of the disease, extracted from 251 peer-reviewed reviews and systematic reviews of the last five years. Three features distinguish this representation: it is integrated, mapping every reviewed axis onto a single graph rather than treating them in isolation; it is semantic, in that each of its 1,632 connections carries an explicit direction, functional role (pathogenic, protective, mixed, therapeutic), and source attribution; and it is multilevel, with every node anchored to a biological scale ranging from molecular signaling to environmental exposure. The resulting representation makes contemporary asthma biology, in its full reviewed extent, structurally visible and queryable.

Three conceptual implications follow. First, asthma emerges as a network-level adaptive reorganization of an integrated airway-immune-epithelial-metabolic-ecological system under environmental and systemic pressure, rather than a disease with one dominant mechanism plus contextual modifiers. Second, the canonical T2/non-T2 dichotomy is updated rather than overturned: the two cascades are biologically separable but not mutually exclusive, sharing a common upstream interface (the epithelium) and a common downstream convergence point (the clinical-functional cluster), with mixed-inflammation phenotypes becoming the expected rather than the exceptional case. Third, comorbidities, microbiome, metabolism, and environmental determinants are constitutive components of asthma pathobiology, not peripheral modifiers — a reframing that has direct clinical consequences for stratification, comorbidity screening, and intervention prioritization.

For researchers, the network provides a quantitative baseline against which more sophisticated network-medicine analyses — including algorithmic community detection, centrality measures, robustness analysis, and integration with molecular interactomes — can be built. The companion catalogues of uncertain findings and therapies map the productive frontier of the field and the areas where therapeutic innovation would address the largest unmet need. For clinicians, the cluster-based representation complements and operationalizes the treatable-traits framework, suggests a hub-informed stratification strategy that goes beyond the T2/non-T2 dichotomy, and argues for the integrated management of asthma and its comorbidities as a single co-pathogenetic problem. The explicit identification of mixed-role findings and of context-dependent effects provides a practical lens through which apparently contradictory literature reports can be reconciled.

Three priorities emerge as future directions: (i) the extension of the current descriptive network with formal computational network-medicine methods, to quantify hub robustness, identify modules with algorithmic rigor, and perform network-based drug repurposing analyses; (ii) the integration of the literature-derived network with primary datasets — omics profiles, biomarker-stratified clinical cohorts, real-world data from biologics registries — to move from a review-derived to a hybrid literature-plus-data network; (iii) the application of the same framework to other multi-factorial chronic diseases where literature fragmentation is analogous (chronic obstructive pulmonary disease, obesity-related disease, systemic autoimmune conditions), to test whether the principles identified here generalize beyond asthma.

Representing asthma as a network of human-relevant mechanistic evidence does not replace existing conceptual frameworks: it renders them simultaneously visible, comparable, and progressively improvable. The shift it operates — from compartmentalized review to integrated representation — is a small methodological step that opens a new vantage point on a disease whose complexity has long resisted single-axis description.