Long non-coding RNAs (lncRNAs) constitute the majority of transcripts in mammalian genomes and yet, their functions remain largely unknown. We systematically suppressed 285 lncRNAs in human dermal fibroblasts and quantified cellular growth, morphological changes, and transcriptomic responses using Capped Analysis of Gene Expression (CAGE). The resulting transcriptomic profiles recapitulated the observed cellular phenotypes, yielding specific roles for over 40% of analyzed lncRNAs in regulating distinct biological pathways, transcriptional machinery, alternative promoter activity and architecture usage. Overall, combining cellular and molecular profiling provided a powerful approach to unravel the distinct functions of lncRNAs, which we highlight with specific functional roles for ZNF213-AS1 and lnc-KHDC3L-2.

Mammalian genomes encode tens of thousands of noncoding RNAs. Most noncoding transcripts exhibit nuclear localization and several have been shown to play a role in the regulation of gene expression and chromatin remodelling. To investigate the function of such RNAs, methods to massively map the genomic interacting sites of multiple transcripts have been developed. However, they still present some limitations. Here, we introduce RNA And DNA Interacting Complexes Ligated and sequenced (RADICL-seq), a technology that maps genome-wide RNA-chromatin interactions in intact nuclei. RADICL-seq is a proximity ligation-based methodology that reduces the bias for nascent transcription, while increasing genomic coverage and unique mapping rate efficiency compared to existing methods. RADICL-seq identifies distinct patterns of genome occupancy for different classes of transcripts as well as cell type-specific RNA-chromatin interactions, and emphasizes the role of transcription in the establishment of chromatin structure.

Aside from their roles in the cytoplasm, RNA-interference components have been reported to localize also in the nucleus of human cells. In particular, AGO1 associates with active chromatin and appears to influence global gene expression. However, the mechanistic aspects remain elusive. Here, we identify AGO1 as a paraspeckle component that in combination with the NEAT1 lncRNA maintains 3D genome architecture. We demonstrate that AGO1 interacts with NEAT1 lncRNA and its depletion affects NEAT1 expression and the formation of paraspeckles. By Hi-C analysis in AGO1 knockdown cells, we observed global changes in chromatin organization, including TADs configuration, and A/B compartment mixing. Consistently, distinct groups of genes located within the differential interacting loci showed altered expression upon AGO1 depletion. NEAT1 knockout cells displayed similar changes in TADs and higher-order A/B compartmentalization. We propose that AGO1 in association with NEAT1 lncRNA can act as a scaffold that bridges chromatin and nuclear bodies to regulate genome organization and gene expression in human cells.

GM-CSF produced by autoreactive CD4 positive T helper cells is involved in the pathogenesis of autoimmune diseases, such as Multiple Sclerosis. However, the molecular regulators that establish and maintain the features of GM-CSF positive CD4 T cells are unknown. In order to identify these regulators, we isolated human GM-CSF producing CD4 T cells from human peripheral blood by using a cytokine capture assay. We compared these cells to the corresponding GM-CSF negative fraction, and furthermore, we studied naïve CD4 T cells, memory CD4 T cells and bulk CD4 T cells from the same individuals as additional control cell populations. As a result, we provide a rich resource of integrated chromatin accessibility (ATAC-seq) and transcriptome (RNA-seq) data from these primary human CD4 T cell subsets, and we show that the identified signatures are associated with human autoimmune disease, especially Multiple Sclerosis. By combining information about mRNA expression, DNA accessibility and predicted transcription factor binding, we reconstructed directed gene regulatory networks connecting transcription factors to their targets, which comprise putative key regulators of human GM-CSF positive CD4 T cells as well as memory CD4 T cells. Our results suggest potential therapeutic targets to be investigated in the future in human autoimmune disease.

In eukaryotes, mRNA abundance is often a poor proxy for protein abundance. Despite this, the majority of methods used to dissect function in mammalian biology and for biomarker discovery in complex diseases involve manipulation or measurement of mRNA. The discrepancy between mRNA and protein abundance is likely due to several factors, including differences in the rates of translation and degradation between proteins and cell-types, unequal contribution of individual splice variants to the production of a given protein and cell-type specific differences in splice variant use. Here we performed experimental and computational time-series analysis of RNA-seq and mass-spectrometry of three key immune cell-types in human and mice and constructed mathematical mixed time-delayed splice variant models to predict protein abundances. These models had median correlations to protein abundance measurements of 0.79-0.94, which is a significant increase from the previously reported 0.21 on human protein atlas data, and out-performed less complicated models without the usage of multiple splice variants and time-delay in cross-validation tests. We showed the importance of our models for biomarker discovery by re-analysing RNA-seq data from five different complex diseases, which led to the prediction of new disease proteins that were validated in multiple sclerosis. Our findings suggest that similar protein abundance models may be created for the most critical cell-types in the human body.

Disease modules in molecular interaction maps have been useful for characterizing diseases. Yet biological networks, commonly used to define such modules are incomplete and biased toward some well-studied disease genes. Here we ask whether disease-relevant modules of genes can be discovered without assuming the prior knowledge of a biological network. To this end we train a deep auto-encoder on a large transcriptional data-set. Our hypothesis is that such modules could be discovered in the deep representations within the auto-encoder when trained to capture the variance in the input-output map of the transcriptional profiles. Using a three-layer deep auto-encoder we find a statistically significant enrichment of GWAS relevant genes in the third layer, and to a successively lesser degree in the second and first layers respectively. In contrast, we found an opposite gradient where a modular protein-protein interaction signal was strongest in the first layer but then vanishing smoothly deeper in the network. We conclude that a data-driven discovery approach, without assuming a particular biological network, is sufficient to discover groups of disease-related genes.

We show how complexity theory can be introduced in machine learning to help bring together apparently disparate areas of current research. We show that this new approach requires less training data and is more generalizable as it shows greater resilience to random attacks. We investigate the shape of the discrete algorithmic space when performing regression or classification using a loss function parametrized by algorithmic complexity, demonstrating that the property of differentiation is not necessary to achieve results similar to those obtained using differentiable programming approaches such as deep learning. In doing so we use examples which enable the two approaches to be compared (small, given the computational power required for estimations of algorithmic complexity). We find and report that (i) machine learning can successfully be performed on a non-smooth surface using algorithmic complexity; (ii) that parameter solutions can be found using an algorithmic-probability classifier, establishing a bridge between a fundamentally discrete theory of computability and a fundamentally continuous mathematical theory of optimization methods; (iii) a formulation of an algorithmically directed search technique in non-smooth manifolds can be defined and conducted; (iv) exploitation techniques and numerical methods for algorithmic search to navigate these discrete non-differentiable spaces can be performed; in application of the (a) identification of generative rules from data observations; (b) solutions to image classification problems more resilient against pixel attacks compared to neural networks; (c) identification of equation parameters from a small data-set in the presence of noise in continuous ODE system problem, (d) classification of Boolean NK networks by (1) network topology, (2) underlying Boolean function, and (3) number of incoming edges.

The concept of information has emerged as a language in its own right, bridging several disciplines that analyze natural phenomena and man-made systems. Integrated information has been introduced as a metric to quantify the amount of information generated by a system beyond the information generated by its elements. Yet, this intriguing notion comes with the price of being prohibitively expensive to calculate, since the calculations require an exponential number of sub-divisions of a system. Here we introduce a novel framework to connect algorithmic randomness and integrated information and a numerical method for estimating integrated information using a perturbation test rooted in algorithmic information dynamics. This method quantifies the change in program size of a system when subjected to a perturbation. The intuition behind is that if an object is random then random perturbations have little to no effect to what happens when a shorter program but when an object has the ability to move in both directions (towards or away from randomness) it will be shown to be better integrated as a measure of sophistication telling apart randomness and simplicity from structure. We show that an object with a high integrated information value is also more compressible, and is, therefore, more sensitive to perturbations. We find that such a perturbation test quantifying compression sensitivity provides a system with a means to extract explanations--causal accounts--of its own behaviour. Our technique can reduce the number of calculations to arrive at some bounds or estimations, as the algorithmic perturbation test guides an efficient search for estimating integrated information. Our work sets the stage for a systematic exploration of connections between algorithmic complexity and integrated information at the level of both theory and practice.

mtDNA recombination following hybridization is rarely found in animals and was never until now reported in reef-building corals. Here we report unexpected topological incongruence among mitochondrial markers as evidence of mitochondrial introgression in the phylogenetic history of Stylophora species distributed along broad geographic ranges. Our analyses include specimens from the Indo-Pacific, the Indian Ocean and the full latitudinal (2000 km) and environmental gradient(21°C-33°C) of the Red Sea (N=827). The analysis of Stylophora lineages in the framework of the mitogenome phylogenies of the family Pocilloporidae, coupled with analyses of recombination, shows the first evidence of asymmetric patterns of introgressive hybridization associated to mitochondrial recombination in this genus. Hybridization likely occurred between an ancestral lineage restricted to the Red Sea/Gulf of Aden basins and migrants from the Indo-Pacific/Indian Ocean that reached the Gulf of Aden. The resulting hybrid lives in sympatry with the descendants of the parental Red Sea lineage, from which it inherited most of its mtDNA (except the recombinant region that includes the nd6, atp6 and mtORF genes) and expanded its range into the hottest region of the Arabian Gulf, where it is scarcely found. Noticeably, across the Red Sea, both lineages exhibit striking differences in terms of phylogeographic patterns, clades-morphospecies association, and zooxanthellae composition. Our data suggest that the early colonization of the Red Sea by the ancestral lineage, which involved overcoming multiple habitat changes and extreme temperatures, resulted in changes in mitochondrial proteins, which led to its successful adaptation to the novel environmental conditions.

The Mos1-mediated Single-Copy Insertion (MosSCI) method is widely used to establish stable Caenorhabditis elegans transgenic strains. Cloning MosSCI targeting plasmids can be cumbersome because it requires assembling multiple genetic elements including a promoter, a 3′UTR and gene fragments. Recently, Schwartz and Jorgensen developed the SapTrap method for the one-step assembly of plasmids containing components of the CRISPR/Cas9 system for C. elegans (Schwartz and Jorgensen 2016 Genetics, 202:1277-1288). Here, we report on the adaptation of the SapTrap method for the efficient and modular assembly of a promoter, 3′UTR and either 2 or 3 gene fragments in a MosSCI targeting vector in a single reaction. We generated a toolkit that includes several fluorescent tags, components of the ePDZ/LOV optogenetic system and regulatory elements that control gene expression in the C. elegans germline. As a proof of principle, we generated a collection of strains that fluorescently label the endoplasmic reticulum and mitochondria in the hermaphrodite germline and that enable the light-stimulated recruitment of mitochondria to centrosomes in the one-cell worm embryo. The method described here offers a flexible and efficient method for assembly of custom MosSCI targeting vectors.