Machine learning analyses of antibody somatic mutations predict immunoglobulin light chain toxicity

Maura Garofalo,Antonio Lanzavecchia,Luca Piccoli,Jacopo Sgrignani,Marco Prunotto,Margherita Romeo,Andrea Cavalli,Olivier Michielin,Luca Varani,Mathilde Foglierini,Maria Monica Barzago,Milos Matkovic,Luisa Diomede,Raoul De Gasparo,Sara Ravasio

doi:10.1038/s41467-021-23880-9

Abstract

In systemic light chain amyloidosis (AL), pathogenic monoclonal immunoglobulin light chains (LC) form toxic aggregates and amyloid fibrils in target organs. Prompt diagnosis is crucial to avoid permanent organ damage, but delayed diagnosis is common because symptoms usually appear only after strong organ involvement. Here we present LICTOR, a machine learning approach predicting LC toxicity in AL, based on the distribution of somatic mutations acquired during clonal selection. LICTOR achieves a specificity and a sensitivity of 0.82 and 0.76, respectively, with an area under the receiver operating characteristic curve (AUC) of 0.87. Tested on an independent set of 12 LCs sequences with known clinical phenotypes, LICTOR achieves a prediction accuracy of 83%. Furthermore, we are able to abolish the toxic phenotype of an LC by in silico reverting two germline-specific somatic mutations identified by LICTOR, and by experimentally assessing the loss of in vivo toxicity in a Caenorhabditis elegans model. Therefore, LICTOR represents a promising strategy for AL diagnosis and reducing high mortality rates in AL.

Highlights

In systemic light chain amyloidosis (AL), pathogenic monoclonal immunoglobulin light chains (LC) form toxic aggregates and amyloid fibrils in target organs
To exclude possible bias induced by the use of a group of nox sequences having an artificially low level of somatic mutations (SMs), we randomly selected 1000 LC sequences from a healthy donor repertoire[25] and compared the probability distributions of the number of SMs (PDSM) between the three groups
We observed similar PDSM between the nox and hdnox groups, while the PDSM of tox and hdnox, as well as, tox and nox were significantly different. This result supports nox sequences as a bona fide group of LCs (Supplementary Fig. 1). These findings suggest that SMs are key determinants of the toxicity of LCs and, can be used as predictor variables to develop LC toxicity prediction tools

Summary

Introduction

In systemic light chain amyloidosis (AL), pathogenic monoclonal immunoglobulin light chains (LC) form toxic aggregates and amyloid fibrils in target organs. We present LICTOR, a machine learning approach predicting LC toxicity in AL, based on the distribution of somatic mutations acquired during clonal selection. Predicting the onset of AL is highly challenging, as each patient carries a different pathogenic LC sequence resulting from a unique rearrangement of variable (V) and joining (J) immunoglobulin genes and a unique set of somatic mutations (SMs) acquired during B cell affinity maturation[9] (Fig. 1a). Machine learning has been used in different areas of medicine, such as diagnosis[10,11,12], prognosis[13,14], drug discovery[15,16] and drug sensitivity prediction[17,18,19]. The high diversity of LC sequences accountable for AL development and the possibility of accessing databases of pathogenic and non-pathogenic LC sequences prompted us to use a machine-learning-based strategy to devise a predictor of LC toxicity in AL named LICTOR (λ-LIght-Chain TOxicity predictoR)

Methods

Results

Conclusion