Protein aggregation is a pathological hallmark of more than 50 human diseases and a major problem for biotechnology. Methods have been proposed to predict aggregation from sequence, but these have been trained and evaluated on small and biased experimental datasets. Here we directly address this data shortage by experimentally quantifying the aggregation of >100,000 protein sequences. This unprecedented dataset reveals the limited performance of existing computational methods and allows us to train CANYA, a convolution-attention hybrid neural network that accurately predicts aggregation from sequence. We adapt genomic neural network interpretability analyses to reveal CANYA’s decision-making process and learned grammar. Our results illustrate the power of massive experimental analysis of random sequence-spaces and provide an interpretable and robust neural network model to predict aggregation.

The transcription factor CREB3L1 is expressed in a wide variety of tissues including cartilage, pancreas, and bone. It is located in the endoplasmic reticulum and upon stimulation is transported to the Golgi where is proteolytically cleaved. Then, the N-terminal domain translocates to the nucleus to activate gene expression. In thyroid follicular cells, CREB3L1 is a downstream effector of thyrotropin (TSH), promoting the expression of proteins of the secretory pathway along with an expansion of the Golgi volume. Here, we analyzed the role of CREB3L1 as a TSH-dependent transcriptional regulator of the expression of the sodium/iodide symporter (NIS), a major thyroid protein that mediates iodide uptake. We show that overexpression and inhibition of CREB3L1 induce an increase and decrease in the NIS protein and mRNA levels, respectively. This, in turn, impacts on NIS-mediated iodide uptake. Furthermore, CREB3L1 knockdown hampers the increase the TSH-induced NIS expression levels. Finally, the ability of CREB3L1 to regulate the promoter activity of the NIS-coding gene (Slc5a5) was confirmed. Taken together, our findings highlight the role of CREB3L1 in maintaining the homeostasis of thyroid follicular cells, regulating the adaptation of the secretory pathway as well as the synthesis of thyroid-specific proteins in response to TSH stimulation.

Congenital iodide transport defect is an uncommon autosomal recessive disorder caused by loss-of-function variants in the sodium iodide symporter (NIS)-coding SLC5A5 gene and leading to dyshormonogenic congenital hypothyroidism. Here, we conducted a targeted next-generation sequencing assessment of congenital hypothyroidism-causative genes in a cohort of nine unrelated pediatric patients suspected of having a congenital iodide transport defect based on the absence of ^99mTc-pertechnetate accumulation in a eutopic thyroid gland. Although, unexpectedly, we could not detect pathogenic SLC5A5 gene variants, we identified two novel compound heterozygous TG gene variants (p.Q29* and c.177-2A>C), three novel heterozygous TG gene variants (p.F1542Vfs*20, p.Y2563C, and p.S523P), and a novel heterozygous DUOX2 gene variant (p.E1496Dfs*51). Splicing minigene reporter-based in vitro assays revealed that the variant c.177-2A>C affected normal TG pre-mRNA splicing, leading to the frameshift variant p.T59Sfs*17. The frameshift TG variants p.T59Sfs*17 and p.F1542Vfs*20, but not the DUOX2 variant p.E1496Dfs*51, were predicted to undergo nonsense-mediated decay. Moreover, functional in vitro expression assays revealed that the variant p.Y2563C reduced the secretion of the TG protein. Our investigation revealed unexpected findings regarding the genetics of congenital iodide transport defects, supporting the existence of yet to be discovered mechanisms involved in thyroid hormonogenesis.

Protein aggregation is a pathological hallmark of more than fifty human diseases and a major problem for biotechnology. Methods have been proposed to predict aggregation from sequence, but these have been trained and evaluated on small and biased experimental datasets. Here we directly address this data shortage by experimentally quantifying the amyloid nucleation of >100,000 protein sequences. This unprecedented dataset reveals the limited performance of existing computational methods and allows us to train CANYA, a convolution-attention hybrid neural network that accurately predicts amyloid nucleation from sequence. We adapt genomic neural network interpretability analyses to reveal CANYA’s decision-making process and learned grammar. Our results illustrate the power of massive experimental analysis of random sequence-spaces and provide an interpretable and robust neural network model to predict amyloid nucleation.

Background. The sodium/iodide symporter (NIS) mediates active iodide accumulation in the thyroid follicular cell. Autosomal recessive iodide transport defect (ITD)-causing loss-of-function NIS variants lead to dyshormonogenic congenital hypothyroidism due to deficient iodide accumulation for thyroid hormonogenesis. Here, we aimed to identify, and if so to functionally characterize, novel ITD-causing NIS pathogenic variants in a patient diagnosed with severe dyshormonogenic congenital hypothyroidism due to a defect in iodide accumulation in the thyroid follicular cell, as suggested by non-detectable radioiodide accumulation in a normally located thyroid gland, as well as in salivary glands. Methods. The proposita NIS-coding SLC5A5 gene was sequenced using Sanger sequencing. In silico analysis and functional in vitro characterization of the novel NIS variants was performed. Results. Sanger sequencing revealed novel compound heterozygous SLC5A5 gene variants (c.970-3C>A and c.1106A>T, p.D369V). In silico analysis suggested that c.970-3C>A disrupts the canonical splice acceptor site located in intron 7. Splicing minigene reporter assay revealed that c.970-3C>A causes exon 8 skipping during NIS pre-mRNA splicing leading to the NIS pathogenic variant p.Y324Hfs*148. Moreover, in silico analysis indicated p.D369V as pathogenic. Functional in vitro studies demonstrated that D369V NIS does not mediate iodide accumulation, as D369V causes NIS to be retained in the endoplasmic reticulum. Mechanistically, we propose an intramolecular ionic interaction involving the β carboxyl group of D369 and the guanidinium group of R130, located in transmembrane segment 4. Of note, an Asp residue at position 369—which is highly conserved in SLC5A family members—is required for functional NIS expression at the plasma membrane. Conclusions: We uncovered a critical intramolecular interaction between R130 and D369 required for NIS maturation and plasma membrane expression. Moreover, we identified the first intronic variant causing aberrant NIS pre-mRNA splicing, thus expanding the mutational landscape in the SLC5A5 gene leading to dyshormonogenic congenital hypothyroidism.