bbeaR: an R package and framework for epitope-specific antibody profiling

Abstract

Summary

Analysis of epitope-specific antibody repertoires has provided novel insights into the pathogenesis of inflammatory disorders, especially allergies. A novel multiplex immunoassay, termed Bead-Based Epitope Assay (BBEA), was developed to quantify levels of epitope-specific immunoglobulins, including IgE, IgG, IgA and IgD isotypes. bbeaR is an open-source R package, developed for the BBEA, provides a framework to import, process and normalize .csv data files exported from the Luminex reader, evaluate various quality control metrics, analyze differential epitope-binding antibodies with linear modeling, visualize results and map epitopes’ amino acid sequences to their respective primary protein structures. bbeaR enables streamlined and reproducible analysis of epitope-specific antibody profiles.

Availability and implementation

bbeaR is open-source and freely available from GitHub as an R package: https://github.com/msuprun/bbeaR; vignettes included.

Supplementary information

Supplementary data are available at Bioinformatics online.

1 Introduction

Antibodies are part of the adaptive humoral immunity and offer essential protection against a variety of pathogens. When antibodies recognize a specific site on an antigenic protein, i.e. an epitope, an immune complex is formed, allowing the immune system to fight the pathogen (Lu et al., 2018; Sela-Culang et al., 2013). The variable domain of an antibody recognizes a wide range of molecules, while the constant domain defines its isotype (i.e. IgM, IgD, IgG, IgA and IgE) and effector functions (Lu et al., 2018). In the case of viral or bacterial infection, IgG and IgM antibodies could neutralize or help clear the pathogen (Forthal, 2014). In allergy, where an innocuous protein is recognized as an antigen, cross-linked IgE activates mast cells and basophils, which readily release inflammatory molecules such as histamine, heparin and tryptase, leading to allergic reactions (Burton and Oettgen, 2011). Quantifying epitope-specific antibodies has many applications, including (i) diagnostics, where the amount of IgE specific to different epitopes can indicate clinical reactivity and severity of allergy (Sackesen et al., 2019; Santos et al., 2020), Iii) prognostics, where isotypes induced during immunotherapy can identify responders (Suarez-Farinas et al., 2019; Sugimoto et al., 2016), (iii) insights into the natural disease evolution and pathogenesis (Suprun et al., 2020).

Bead-Based Epitope Assay (BBEA) is a Luminex-based high-throughput assay developed to measure the levels of many epitope-specific antibodies simultaneously and outperforms peptide microarrays (Suprun et al., 2019). Here, we present an open-source R package, bbeaR, which facilitates the management, preprocessing and quality control of BBEA data (Fig. 1). It creates a data structure that eases the differential antibodies analysis using limma or linear modeling and contains custom visualizations. bbeaR comes with a vignette with a detailed walkthrough of two examples, showcasing the package utility. While the framework is described using an example of food allergies, it can be applied to any disease or experimental condition where antibody responses are of interest. To the authors’ knowledge, bbeaR is the only R package currently available providing comprehensive pipeline for import, quality control and analytics.

Fig. 1.

Functionality schematic of bbeaR for data import, normalization, quality control (QC), analysis and visualizations

2 Materials and methods

BBEA is run on 96-well microplates, allowing the measurement of many plasma/serum epitope-specific antibodies on many patients simultaneously (Breen, 2017). We recommend that experiments are randomized using PlateDesigner (Suprun and Suárez-Fariñas, 2019), to avoid confounding by technical factors, such as plate or well position. PlateDesigner’s experimental metadata can be uploaded directly to the Luminex reader, avoiding data entry errors. The assay’s readout, exported from Luminex as a comma-separated values (.csv) file for each plate, is a starting point of the bbeaR pipeline.

Csv files from each plate are imported simultaneously, extracting counts and Median Fluorescence Intensity (MFI) tables, along with metadata about the plate run. The MFI is provided for each Luminex bead, i.e. an analyte, which uniquely corresponds to an individual epitope, and annotations are mapped. The counts undergo a first round of quality controls, where samples or antibodies can be removed if they do not reach sufficient reads, e.g. at least 25 counts per analyte (Bjornstal et al., 2011). Then, for each epitope-specific antibody j and sample i, the MFI signal is normalized as

$$\begin{array}{c}{Y}_{ij}={ \hspace{0.17em}\text{log}\hspace{0.17em}}_2\left(MF{I}_{ij}+\text{offset}\right)\\ n{\text{MFI}}_{ij}=\{\begin{array}{cc}0,& Y_{ij}<\frac{{\Sigma }_{k\epsilon BG}{Y}_{kj}}{N_{BG}},\\ Y_{ij}-\frac{{\Sigma }_{k\epsilon BG}{Y}_{kj}}{N_{BG}},& \text{otherwise}\end{array}\end{array}$$

where BG are background wells. The distribution of nMFI is evaluated using Cullen-Frey plots and histograms, and agreement between technical replicates assessed by plate-layout plots, coefficients of variation (CV) and intra-class correlation coefficients (ICC). If an experiment consists of several microplate runs, plate effect can be identified using principal component analysis (PCA) and the nMFI can be adjusted for plate and well effects using a limma model (Silver et al., 2009; Smyth, 2004; Suprun et al., 2019) or other approaches like SVA/ComBat (Leek et al., 2012).

The differential epitope-specific antibodies can be identified by comparing patient groups, using either classical linear models or limma (Silver et al., 2009; Smyth, 2004). We have implemented custom net circle visualization plots that present fold-changes and P-values across antibodies ordered by the epitopes’ position on the protein. Additionally, amino acid sequences of the epitopes are mapped to the protein using the topology plot and can be quite useful when a 3D crystal structure is not available. This plot extracts protein characteristics from the UniProt database, and displays additional information, i.e. sites of disulfide bridges, glycosylation and enzymatic cleavage.

3 Usage example

The features of bbeaR are illustrated in two detailed examples, provided as vignettes: (i) comparison of egg epitope-specific IgE antibodies in egg-allergic children and controls (Suprun et al., 2020) and (ii) immunomodulation of milk epitope-specific IgEs in patients undergoing oral immunotherapy (Suarez-Farinas et al., 2019) (Supplementary Files S1 and S2). While the egg experiment was performed on one microplate, the milk experiment was performed on four plates and required a plate effect adjustment. A brief walkthrough of the milk allergy project is outlined below.

Levels of IgE specific to 66 allergenic milk epitopes were measured before and after treatment in 47 milk-allergic patients undergoing two years of immunotherapy with either omalizumab or placebo adjuvant (Suarez-Farinas et al., 2019). First, to make sure the MFI values are valid, it is important to have sufficient numbers of analyte (that represent epitopes) counts. In all of the plates, minimum counts were above 75, which is considered reliable (Bjornstal et al., 2011). nMFI values are stored in ExpressionSet object along with epitope annotation and patients’ phenotypic or clinical information, allowing for a seamless use of commonly used -omic methods. PCA visualization indicates the presence of a plate effect, which can be quantified by principal variance component analysis (Bushel, 2019; Li et al., 2009) to account for ∼5% of the total variability. Since the samples were randomly allocated to plates, the plate effect was adjusted using linear models. The distribution of the nMFIs is investigated through the Cullen–Frey plot (Delignette-Muller and Dutang, 2015), which is shown to be close to log-normal. As such, parametric models using the nMFI can be used in downstream analyses. Reliability across technical replicates are measured by CV and ICC. After the reliability is ascertained, the replicates are averaged. Changes from baseline to the end of treatment in placebo and omalizumab groups are modeled with limma and show decreased levels of more epitope-specific IgEs in the placebo arm, visualized by the net circle plots.

4 Conclusion

bbeaR provides a flexible functionality for analyzing epitope-specific antibodies quantified with the BBEA or other Luminex-based platforms. After data loading and pre-processing, the workflow is similar to that of microarray and RNA-seq pipelines, allowing users to implement other tools developed for these high-throughput technologies. bbeaR allows seamless and reproducible profiling of epitope-specific antibodies in R.