Methods for time series prediction and classification of gene regulatory networks (GRNs) from gene expression data have been treated separately so far. The recent emergence of attention-based recurrent neural network (RNN) models boosted the interpretability of RNN parameters, making them appealing for the understanding of gene interactions. In this work, we generated synthetic time series gene expression data from a range of archetypal GRNs and we relied on a dual attention RNN to predict the gene temporal dynamics. We show that the prediction is extremely accurate for GRNs with different architectures. Next, we focused on the attention mechanism of the RNN and, using tools from graph theory, we found that its graph properties allow one to hierarchically distinguish different architectures of the GRN. We show that the GRN responded differently to the addition of noise in the prediction by the RNN and we related the noise response to the analysis of the attention mechanism. In conclusion, this work provides a way to understand and exploit the attention mechanism of RNNs and it paves the way to RNN-based methods for time series prediction and inference of GRNs from gene expression data.

Human serine hydroxymethyltransferase (SHMT) regulates the serine-glycine one carbon metabolism and plays a role in cancer metabolic reprogramming. Two SHMT isozymes are acting in the cell: SHMT1 encoding the cytoplasmic isozyme, and SHMT2 encoding the mitochondrial one. Here we present a molecular model built on experimental data reporting the interaction between SHMT1 protein and SHMT2 mRNA, recently discovered in lung cancer cells. Using a stochastic dynamic model, we show that RNA moieties dynamically regulate serine and glycine concentration, shaping the system behaviour. For the first time we observe an active functional role of the RNA in the regulation of the serine-glycine metabolism and availability, which unravels a complex layer of regulation that cancer cells exploit to fine tune amino acids availability according to their metabolic needs. The quantitative model, complemented by an experimental validation in the lung adenocarcinoma cell line H1299, exploits RNA molecules as metabolic switches of the SHMT1 activity. Our results pave the way for the development of RNA-based molecules able to unbalance serine metabolism in cancer cells.

It is not well understood why severe acute respiratory syndrome (SARS)-CoV-2 spreads much faster than other β-coronaviruses such as SARS-CoV and Middle East respiratory syndrome (MERS)-CoV. In a previous publication, we predicted the binding of the N-terminal domain (NTD) of SARS-CoV-2 spike to sialic acids (SAs). Here, we experimentally validate this interaction and present simulations that reveal a second possible interaction between SAs and the spike protein via a binding site located in the receptor-binding domain (RBD). The predictions from molecular-dynamics simulations and the previously-published 2D-Zernike binding-site recognition approach were validated through flow-induced dispersion analysis (FIDA)─which reveals the capability of the SARS-CoV-2 spike to bind to SA-containing (glyco)lipid vesicles, and flow-cytometry measurements─which show that spike binding is strongly decreased upon inhibition of SA expression on the membranes of angiotensin converting enzyme-2 (ACE2)-expressing HEK cells. Our analyses reveal that the SA binding of the NTD and RBD strongly enhances the infection-inducing ACE2 binding. Altogether, our work provides in silico, in vitro, and cellular evidence that the SARS-CoV-2 virus utilizes a two-receptor (SA and ACE2) strategy. This allows the SARS-CoV-2 spike to use SA moieties on the cell membrane as a binding anchor, which increases the residence time of the virus on the cell surface and aids in the binding of the main receptor, ACE2, via 2D diffusion.

Solubility is a requirement for many cellular processes. Loss of solubility and aggregation can lead to the partial or complete abrogation of protein function. Thus, understanding the relationship between protein evolution and aggregation is an important goal. Here, we analysed two deep mutational scanning experiments to investigate the role of protein aggregation in molecular evolution. In one data set, mutants of a protein involved in RNA biogenesis and processing, human TAR DNA binding protein 43 (TDP-43), were expressed in S. cerevisiae. In the other data set, mutants of a bacterial enzyme that controls resistance to penicillins and cephalosporins, TEM-1 beta-lactamase, were expressed in E. coli under the selective pressure of an antibiotic treatment. We found that aggregation differentiates the effects of mutations in the two different cellular contexts. Specifically, aggregation was found to be associated with increased cell fitness in the case of TDP-43 mutations, as it protects the host from aberrant interactions. By contrast, in the case of TEM-1 beta-lactamase mutations, aggregation is linked to a decreased cell fitness due to inactivation of protein function. Our study shows that aggregation is an important context-dependent constraint of molecular evolution and opens up new avenues to investigate the role of aggregation in the cell.

Specific elements of viral genomes regulate interactions within host cells. Here, we calculated the secondary structure content of >2000 coronaviruses and computed >100000 human protein interactions with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). The genomic regions display different degrees of conservation. SARS-CoV-2 domain encompassing nucleotides 22500 – 23000 is conserved both at the sequence and structural level. The regions upstream and downstream, however, vary significantly. This part codes for the Spike S protein that interacts with the human receptor angiotensin-converting enzyme 2 (ACE2). Thus, variability of Spike S may be connected to different levels of viral entry in human cells within the population. Our predictions indicate that the 5’ end of SARS-CoV-2 is highly structured and interacts with several human proteins. The binding proteins are involved in viral RNA processing such as double-stranded RNA specific editases and ATP-dependent RNA-helicases and have strong propensity to form stress granules and phase-separated complexes. We propose that these proteins, also implicated in viral infections such as HIV, are selectively recruited by SARS-CoV-2 genome to alter transcriptional and post-transcriptional regulation of host cells and to promote viral replication.

The Cystic Fibrosis Transmembrane Conductance Regulator (CFTR) is a large multi-spanning membrane protein that is susceptible to misfolding and aggregation. We have identified here the region responsible for this instability. Temperature-induced aggregation of C-terminally truncated versions of CFTR demonstrated that all truncations up to the second transmembrane domain (TMD2), including the R region, largely resisted aggregation. Limited proteolysis identified a folded structure that was prone to aggregation and consisted of TMD2 and at least part of the Regulatory Region R. Only when both TM7 (TransMembrane helix 7) and TM8 were present, TMD2 fragments became as aggregation-sensitive as wild-type CFTR, in line with increased thermo-instability of late CFTR nascent chains and in silico prediction of aggregation propensity. In accord, isolated TMD2 was degraded faster in cells than isolated TMD1. We conclude that TMD2 extended at its N-terminus with part of the R region forms a protease-resistant structure that induces heat instability in CFTR and may be responsible for its limited intracellular stability.