Kell‐Score—A web resource for estimating the immunogenicity of novel Kell blood group variants

High-throughput technologies accelerate the discovery of new blood group variants1 and aim at preventing future clinical complications. Thus, tools for predicting the blood group phenotype from genetic data would help haematologists to interpret the significance of new variants for clinical practice, and furthermore, to identify rare blood group carriers. To date, the Kell blood group system includes 38 antigens as approved by the International Society of Blood Transfusion (ISBT) Red Cell Immunogenetics and Blood Group Terminology Working Party.2, 3 Recently, we analysed the molecular properties of all currently registered KEL alleles that encode full-length variants of the Kell protein as either unexpressed/null (K0) or diminished/mod phenotype (Kmod) or expressing different antigens.4 The insights gained in this study were used to create a catalogue of objective criteria to estimate the potential of a new Kell variant to (i) destabilize the protein, for example, to prevent expression on the erythrocyte surface by hampering stable protein folding, or alternatively, to (ii) create a new antigen on the protein surface that is specifically recognized by immune receptors. With this, a two-score heuristic measure that combines features typical for a specific phenotype was computed for all theoretically possible Kell missense mutations. These measures are not intended to quantify the immunogenicity of Kell variants,5 but rather to create an initial evaluation of the potential molecular effects of newly discovered variants. Here we provide the results of these predictions to the scientific community through a user-friendly online resource that includes an intuitive analysis pipeline to explore and evaluate any number of previously uncharacterized Kell missense variants. Amino acid positions of interest can be simultaneously localized in the 3D protein structural viewer. The web interface is accessible at https://kellscore.ikmb.uni-kiel.de. Based on these Kell variant parameters, we set out to define a weighting scheme for an automated classification/categorization of any protein missense variant. To this end, we designed a two-score heuristic measure by collecting protein amino acid positional and variant features. Simply, any property that is considered typical for a phenotype is adding up to the final score for predicting Kell (i) destabilization or (ii) antigenicity. Therefore, the two prediction scores of an individual variant are each the sum of added characteristic points. Some features are given more weight than others and can be interdepending (Table 1); further details on scoring conditions are given in the Supporting Information Data Sheet (Table S2). A maximum score of five indicates a strong prediction for the given phenotype, and zero denotes a predicted ‘neutral’ variant of no effect. On the webpage, the two scores are displayed graphically as a five-point system for a quick visual estimation of the variant. In the web interface, the user can browse along the protein sequence (UniProt accession: P23246, NCBI reference sequence: NP_000411.1) and simultaneously highlight the 3D position of the selected amino acid in the structural model4, 7 (Figure 1A,B), where it is visualized as ball-and-sticks together with residues within a 7 Å radius. These structural neighbours are also highlighted in the sequence. Known Kell phenotypes are colour-coded as previously applied.4 Destabilization and antigenicity scores can be obtained from a pop-up window by clicking a residue and choosing the variant amino acid of interest, by uploading variants from a local text file or by entering variants in the text box above. In addition, single missense variants that define an ISBT allele can be chosen from a dropdown button. Following selection, a table is generated (Figure 1D) that visualizes (i) a prediction score for each phenotype depicted as five red squares (scores in teal colour indicate default scores of known phenotypes); (ii) relative solvent accessibility (RSA); (iii) the variant type, for example, either of an ISBT variant category including allele name4 or rare or not yet detected in the European population, or a hypothetical mutation as inferred from the (iv) minor allele frequency (MAF) of the European/non-Finnish population as derived from the gnomAD database.8 By selecting the teal icon in the Details column, a text box opens (Figure 1C) to provide specific information about the variant position and the origin of conditional sub-scores that sum up to the final score of maximum five. If the variant is given a default score, for example, a variant with a known ISBT phenotype, conditional subscores are indicated as ‘Notional subscores’. An eye-shaped orange icon in the 3D column activates the protein structural visualization of the position. Every column in the table can be sorted and filtered for prioritizing entries. The web interface was designed as a single-page web application and implemented using the Ember.js (https://emberjs.com/) JavaScript web application framework utilizing design templates of the Bootstrap 5 toolkit (https://getbootstrap.com/). The protein visualization is generated by a Mol* plugin instance of the JavaScript library Mol* (https://github.com/molstar/molstar).7 Communication between visualization and sequence input is realized using the Mol* plugin's application programming interface (API) and contemporary software design patterns like data binding and modifiers. Due to the lack of an experimental high-resolution Kell structure, the method relies on a predicted protein model and is therefore inherently limited in its accuracy regarding 3D structural context and dynamics of individual residues. However, we provide calculated scores also for known Kell antigens and K0 phenotype and find that our predictions match well with their phenotype (Figure 1). To conclude, users of the Kell web resource are provided with a protein sequential and structural viewer that allows them to analyse any single protein mutation in molecular detail and provides an estimate of the molecular consequence of the variant, taking into account the 3D context and all known ISBT phenotypes. In addition, researchers can prioritize variants from any number of entries by ranking destabilization and antigenicity prediction scores. Gabriele Mayr designed and performed the research study, conceptualized the web interface, analysed the data and drafted the manuscript. Marcus Schmöhl designed and implemented the webpage. Maike Bublitz analysed the data and drafted the manuscript. Christoph Gassner analysed the data. Andre Franke conceptualized the web interface. All authors read and approved the manuscript. We are grateful to the IKMB IT department for their support in hosting the website and managing and maintaining the necessary infrastructure. Open Access funding enabled and organized by Projekt DEAL. This study is an outcome of the Deutsche Forschungsgemeinschaft (DFG) Collaborative Research Center 1182 ‘Origin and Function of Metaorganisms’ (https://www.metaorganism-research.com, Project number: 261376515, Projects A2.2 and INF) and received infrastructure support from the DFG Excellence Cluster 2167 ‘Precision Medicine in Chronic Inflammation’ (PMI; Project number: 390884018). The Institute of Translational Medicine of the Private University in the Principality of Liechtenstein is financially supported by the Hans Groeber-Stiftung, Vaduz, Principality of Liechtenstein and the Tarom Foundation, Schaan, Principality of Liechtenstein. Christoph Gassner acts as a consultant to Inno-Train GmbH, Kronberg im Taunus, Germany, a provider of genotyping kits for molecular blood group diagnostics since 1998. CG holds the European and US patents P3545102 and US20190316189 on the ‘Determination of the genotype underlying the S-s-U- phenotype of the MNSs blood group system’. No other conflict of interest is declared by any of the authors. Data are available in the Supporting Information of this article. Table S1. Table S2. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.