Increased heterogeneity in expression of genes associated with cancer progression and drug resistance.

Fluctuations in the number of regulatory molecules and differences in timings of molecular events can generate variation in gene expression among genetically identical cells in the same environmental condition. This variation, termed as expression noise, can create differences in metabolic state and cellular functions, leading to phenotypic heterogeneity. Expression noise and phenotypic heterogeneity have been recognized as important contributors to intra-tumor heterogeneity, and have been associated with cancer growth, progression, and therapy resistance. However, how expression noise changes with cancer progression in actual cancer patients has remained poorly explored. Such an analysis, through identification of genes with increasing expression noise, can provide valuable insights into generation of intra-tumor heterogeneity, and could have important implications for understanding immune-suppression, drug tolerance and therapy resistance. In this work, we performed a genome-wide identification of changes in gene expression noise with cancer progression using single-cell RNA-seq data of lung adenocarcinoma patients at different stages of cancer. We identified 37 genes in epithelial cells that showed an increasing noise trend with cancer progression, many of which were also associated with cancer growth, EMT and therapy resistance. We found that expression of several of these genes was positively associated with expression of mitochondrial genes, suggesting an important role of mitochondria in generation of heterogeneity. In addition, we uncovered substantial differences in sample-specific noise profiles which could have implications for personalized prognosis and treatment.

Transcription factor binding process is the primary driver of noise in gene expression.

Noise in expression of individual genes gives rise to variations in activity of cellular pathways and generates heterogeneity in cellular phenotypes. Phenotypic heterogeneity has important implications for antibiotic persistence, mutation penetrance, cancer growth and therapy resistance. Specific molecular features such as the presence of the TATA box sequence and the promoter nucleosome occupancy have been associated with noise. However, the relative importance of these features in noise regulation is unclear and how well these features can predict noise has not yet been assessed. Here through an integrated statistical model of gene expression noise in yeast we found that the number of regulating transcription factors (TFs) of a gene was a key predictor of noise, whereas presence of the TATA box and the promoter nucleosome occupancy had poor predictive power. With an increase in the number of regulatory TFs, there was a rise in the number of cooperatively binding TFs. In addition, an increased number of regulatory TFs meant more overlaps in TF binding sites, resulting in competition between TFs for binding to the same region of the promoter. Through modeling of TF binding to promoter and application of stochastic simulations, we demonstrated that competition and cooperation among TFs could increase noise. Thus, our work uncovers a process of noise regulation that arises out of the dynamics of gene regulation and is not dependent on any specific transcription factor or specific promoter sequence.

EpICC: A Bayesian neural network model with uncertainty correction for a more accurate classification of cancer.

Accurate classification of cancers into their types and subtypes holds the key for choosing the right treatment strategy and can greatly impact patient well-being. However, existence of large-scale variations in the molecular processes driving even a single type of cancer can make accurate classification a challenging problem. Therefore, improved and robust methods for classification are absolutely critical. Although deep learning-based methods for cancer classification have been proposed earlier, they all provide point estimates for predictions without any measure of confidence and thus, can fall short in real-world applications where key decisions are to be made based on the predictions of the classifier. Here we report a Bayesian neural network-based model for classification of cancer types as well as sub-types from transcriptomic data. This model reported a measure of confidence with each prediction through analysis of epistemic uncertainty. We incorporated an uncertainty correction step with the Bayesian network-based model to greatly enhance prediction accuracy of cancer types (> 97% accuracy) and sub-types (> 80%). Our work suggests that reporting uncertainty measure with each classification can enable more accurate and informed decision-making that can be highly valuable in clinical settings.

Designer bacterial cell factories for improved production of commercially valuable non-ribosomal peptides.

Non-ribosomal peptides have gained significant attention as secondary metabolites of high commercial importance. This group houses a diverse range of bioactive compounds, ranging from biosurfactants to antimicrobial and cytotoxic agents. However, low yield of synthesis by bacteria and excessive losses during purification hinders the industrial-scale production of non-ribosomal peptides, and subsequently limits their widespread applicability. While isolation of efficient producer strains and optimization of bioprocesses have been extensively used to enhance yield, further improvement can be made by optimization of the microbial strain using the tools and techniques of metabolic engineering, synthetic biology, systems biology, and adaptive laboratory evolution. These techniques, which directly target the genome of producer strains, aim to redirect carbon and nitrogen fluxes of the metabolic network towards the desired product, bypass the feedback inhibition and repression mechanisms that limit the maximum productivity of the strain, and even extend the substrate range of the cell for synthesis of the target product. The present review takes a comprehensive look into the biosynthesis of bacterial NRPs, how the same is regulated by the cell, and dives deep into the strategies that have been undertaken for enhancing the yield of NRPs, while also providing a perspective on other potential strategies that can allow for further yield improvement. Furthermore, this review provides the reader with a holistic perspective on the design of cellular factories of NRP production, starting from general techniques performed in the laboratory to the computational techniques that help a biochemical engineer model and subsequently strategize the architectural plan.

Single cell functional genomics reveals the importance of mitochondria in cell-to-cell phenotypic variation.

Mutations frequently have outcomes that differ across individuals, even when these individuals are genetically identical and share a common environment. Moreover, individual microbial and mammalian cells can vary substantially in their proliferation rates, stress tolerance, and drug resistance, with important implications for the treatment of infections and cancer. To investigate the causes of cell-to-cell variation in proliferation, we used a high-throughput automated microscopy assay to quantify the impact of deleting >1500 genes in yeast. Mutations affecting mitochondria were particularly variable in their outcome. In both mutant and wild-type cells mitochondrial membrane potential - but not amount - varied substantially across individual cells and predicted cell-to-cell variation in proliferation, mutation outcome, stress tolerance, and resistance to a clinically used anti-fungal drug. These results suggest an important role for cell-to-cell variation in the state of an organelle in single cell phenotypic variation.