A network-based approach for identifying suitable biomarkers for oral immunotherapy of food allergy

Abstract

Background: Oral immunotherapy (OIT) is a promising therapeutic approach to treat food allergic patients. However, concerns with regards to safety and long-term efficacy of OIT remain. There is a need to identify biomarkers that predict, monitor and/or evaluate the effects of OIT. Here we present a method to select candidate biomarkers for efficacy and safety assessment of OIT using the computational approaches Bayesian networks (BN) and Topological Data Analysis (TDA).

Results: Data were used from fructo-oligosaccharide diet-supported OIT experiments performed in 3 independent cow's milk allergy (CMA) and 2 independent peanut allergy (PNA) experiments in mice. Bioinformatical approaches were used to understand the data structure. The BN predicted the efficacy of OIT in the CMA with 86% and indicated a clear effect of scFOS/lcFOS on allergy parameters. For the PNA model, this BN (trained on CMA data) predicted an efficacy of OIT with 76% accuracy and shows similar effects of the allergen, treatment and diet as compared to the CMA model. The TDA identified clusters of biomarkers closely linked to biologically relevant clinical symptoms and also unrelated and redundant parameters within the network.

Conclusions: Here we provide a promising application of computational approaches to a) compare mechanistic features of two different food allergies during OIT b) determine the biological relevance of candidate biomarkers c) generate new hypotheses to explain why CMA has a different disease pattern than PNA and d) select relevant biomarkers for future studies.

Keywords: Bayesian network analyses; Bioinformatics; Experimental food allergy; Oral immunotherapy; Topological data analyses.

Fig. 1

Experimental timelines of PNA and CMA models. 6-week-old female C3H/HeOuJ mice were randomly allocated to the control- and experimental groups: sham-sensitized control group;, sensitized control group;, FOS supplemented group; oral immunotherapy group; and the oral immunotherapy with FOS supplementation group. Mice were i.g. sensitized to the cow’s milk protein whey or PE (20 mg whey in 0.5 ml or 6 mg PE in 2 ml PBS) with cholera toxin as an adjuvant (15 μg in 0.5 ml PBS). The FOS supplemented diet was provided from D35 to the end of the protocol and OIT with 10 mg whey or 1.5 or 15 mg PE in 0.5 ml PBS was given from D42-D59 (5 oral gavages/week for 3 weeks). Acute allergic symptoms were measured upon i.d. challenge at D64 (10 μg whey or 1 μg PE in 20 μl PBS/ear), mast cell degranulation was measured upon i.g. challenge at D70 (50 mg whey or 15 mg PE in 0.5 ml PBS) and an i.p. challenge (50 μg whey or 100 μg PE in 200 μl PBS) was conducted at D77 to stimulate T cell responses prior to organ collection. At 6 time points throughout the animal experiment (D0, D35, D50, D63, D71 and D78), subgroups of mice from each control- and experimental group were killed by cervical dislocation and blood and organs were collected. PE; peanut extract, CT; cholera toxin, OIT; oral immunotherapy, FOS; fructo-oligosaccharides, i.d.; intradermal, i.g.; intragastric, i.p.; intraperitoneal, LP; lamina propria of small intestine, SCFA; short-chain fatty acids

Fig. 2

Bayesian network trained on CMA data, effects of sensitization and/or OIT. BN depicting the relationships between the analyzed parameters and the animal treatments (sensitization, scFOFS/lcFOS diet, or OIT treatment) using CMA data. Moreover the probability distributions of all BN variables are depicted a) irrespective of animal treatments, b) assuming that all animals were sensitized and c) assuming that all animals were sensitized and received OIT

Fig. 3

Bayesian network trained on CMA data, effects of scFOS/lcFOS diet with or without OIT. BN depicting the relationships between the analyzed parameters and the animal treatments (sensitization, scFOFS/lcFOS diet, OIT treatment) using the CMA data. The probability distributions of all BN variables are depicted assuming that all animals were sensitized and a) received scFOS/lcFOS diet or b) received scFOS/lcFOS diet in combination with OIT

Fig. 4

The mutual nearest neighbors network of the CMA model. Topological network showing the clustering of parameters (dots with same color). The clusters were used to identify the cluster-relationships in CMA. Moreover, the encircled clusters were used to compare the cluster-relationships between CMA and PNA (see Fig. 5)

Fig. 5

The mutual nearest neighbors network of the PNA model. Depicted is the topological network showing the clustering of parameters (dots with same color). The clusters were used to identify the cluster-relationships in CMA. Moreover, the encircled clusters were used to compare the cluster-relationships between PNA and CMA (see Fig. 4)