Evolutionarily informed deep learning methods for predicting relative transcript abundance from DNA sequence.

Deep learning methodologies have revolutionized prediction in many fields and show potential to do the same in molecular biology and genetics. However, applying these methods in their current forms ignores evolutionary dependencies within biological systems and can result in false positives and spurious conclusions. We developed two approaches that account for evolutionary relatedness in machine learning models: (i) gene-family-guided splitting and (ii) ortholog contrasts. The first approach accounts for evolution by constraining model training and testing sets to include different gene families. The second approach uses evolutionarily informed comparisons between orthologous genes to both control for and leverage evolutionary divergence during the training process. The two approaches were explored and validated within the context of mRNA expression level prediction and have the area under the ROC curve (auROC) values ranging from 0.75 to 0.94. Model weight inspections showed biologically interpretable patterns, resulting in the hypothesis that the 3' UTR is more important for fine-tuning mRNA abundance levels while the 5' UTR is more important for large-scale changes.

Unraveling the KNOTTED1 regulatory network in maize meristems.

KNOTTED1 (KN1)-like homeobox (KNOX) transcription factors function in plant meristems, self-renewing structures consisting of stem cells and their immediate daughters. We defined the KN1 cistrome in maize inflorescences and found that KN1 binds to several thousand loci, including 643 genes that are modulated in one or multiple tissues. These KN1 direct targets are strongly enriched for transcription factors (including other homeobox genes) and genes participating in hormonal pathways, most significantly auxin, demonstrating that KN1 plays a key role in orchestrating the upper levels of a hierarchical gene regulatory network that impacts plant meristem identity and function.

A k-mer grammar analysis to uncover maize regulatory architecture.

BACKGROUND: Only a small percentage of the genome sequence is involved in regulation of gene expression, but to biochemically identify this portion is expensive and laborious. In species like maize, with diverse intergenic regions and lots of repetitive elements, this is an especially challenging problem that limits the use of the data from one line to the other. While regulatory regions are rare, they do have characteristic chromatin contexts and sequence organization (the grammar) with which they can be identified. RESULTS: We developed a computational framework to exploit this sequence arrangement. The models learn to classify regulatory regions based on sequence features - k-mers. To do this, we borrowed two approaches from the field of natural language processing: (1) "bag-of-words" which is commonly used for differentially weighting key words in tasks like sentiment analyses, and (2) a vector-space model using word2vec (vector-k-mers), that captures semantic and linguistic relationships between words. We built "bag-of-k-mers" and "vector-k-mers" models that distinguish between regulatory and non-regulatory regions with an average accuracy above 90%. Our "bag-of-k-mers" achieved higher overall accuracy, while the "vector-k-mers" models were more useful in highlighting key groups of sequences within the regulatory regions. CONCLUSIONS: These models now provide powerful tools to annotate regulatory regions in other maize lines beyond the reference, at low cost and with high accuracy.

Identification and Characterization of Maize salmon silks Genes Involved in Insecticidal Maysin Biosynthesis.

The century-old maize (Zea mays) salmon silks mutation has been linked to the absence of maysin. Maysin is a C-glycosyl flavone that, when present in silks, confers natural resistance to the maize earworm (Helicoverpa zea), which is one of the most damaging pests of maize in America. Previous genetic analyses predicted Pericarp Color1 (P1; R2R3-MYB transcription factor) to be epistatic to the sm mutation. Subsequent studies identified two loci as being capable of conferring salmon silks phenotypes, salmon silks1 (sm1) and sm2 Benefitting from available sm1 and sm2 mapping information and from knowledge of the genes regulated by P1, we describe here the molecular identification of the Sm1 and Sm2 gene products. Sm2 encodes a rhamnosyl transferase (UGT91L1) that uses isoorientin and UDP-rhamnose as substrates and converts them to rhamnosylisoorientin. Sm1 encodes a multidomain UDP-rhamnose synthase (RHS1) that converts UDP-glucose into UDP-l-rhamnose. Here, we demonstrate that RHS1 shows unexpected substrate plasticity in converting the glucose moiety in rhamnosylisoorientin to 4-keto-6-deoxy glucose, resulting in maysin. Both Sm1 and Sm2 are direct targets of P1, as demonstrated by chromatin immunoprecipitation experiments. The molecular characterization of Sm1 and Sm2 described here completes the maysin biosynthetic pathway, providing powerful tools for engineering tolerance to maize earworm in maize and other plants.

Core Promoter Plasticity Between Maize Tissues and Genotypes Contrasts with Predominance of Sharp Transcription Initiation Sites.

Core promoters are crucial for gene regulation, providing blueprints for the assembly of transcriptional machinery at transcription start sites (TSSs). Empirically, TSSs define the coordinates of core promoters and other regulatory sequences. Thus, experimental TSS identification provides an essential step in the characterization of promoters and their features. Here, we describe the application of CAGE (cap analysis of gene expression) to identify genome-wide TSSs used in root and shoot tissues of two maize (Zea mays) inbred lines (B73 and Mo17). Our studies indicate that most TSS clusters are sharp in maize, similar to mice, but distinct from Arabidopsis thaliana, Drosophila melanogaster, or zebra fish, in which a majority of genes have broad-shaped TSS clusters. We established that ∼38% of maize promoters are characterized by a broader TATA-motif consensus, and this motif is significantly enriched in genes with sharp TSSs. A noteworthy plasticity in TSS usage between tissues and inbreds was uncovered, with ∼1500 genes showing significantly different dominant TSSs, sometimes affecting protein sequence by providing alternate translation initiation codons. We experimentally characterized instances in which this differential TSS utilization results in protein isoforms with additional domains or targeted to distinct subcellular compartments. These results provide important insights into TSS selection and gene expression in an agronomically important crop.

The Maize TFome--development of a transcription factor open reading frame collection for functional genomics.

Establishing the architecture of the gene regulatory networks (GRNs) responsible for controlling the transcription of all genes in an organism is a natural development that follows elucidation of the genome sequence. Reconstruction of the GRN requires the availability of a series of molecular tools and resources that so far have been limited to a few model organisms. One such resource consists of collections of transcription factor (TF) open reading frames (ORFs) cloned into vectors that facilitate easy expression in plants or microorganisms. In this study, we describe the development of a publicly available maize TF ORF collection (TFome) of 2034 clones corresponding to 2017 unique gene models in recombination-ready vectors that make possible the facile mobilization of the TF sequences into a number of different expression vectors. The collection also includes several hundred co-regulators (CoREGs), which we classified into well-defined families, and for which we propose here a standard nomenclature, as we have previously done for TFs. We describe the strategies employed to overcome the limitations associated with cloning ORFs from a genome that remains incompletely annotated, with a partial full-length cDNA set available, and with many TF/CoREG genes lacking experimental support. In many instances this required the combination of genome-wide expression data with gene synthesis approaches. The strategies developed will be valuable for developing similar resources for other agriculturally important plants. Information on all the clones generated is available through the GRASSIUS knowledgebase (http://grassius.org/).

Large-scale transcriptional profiling of lignified tissues in Tectona grandis.

BACKGROUND: Currently, Tectona grandis is one of the most valuable trees in the world and no transcript dataset related to secondary xylem is available. Considering how important the secondary xylem and sapwood transition from young to mature trees is, little is known about the expression differences between those successional processes and which transcription factors could regulate lignin biosynthesis in this tropical tree. Although MYB transcription factors are one of the largest superfamilies in plants related to secondary metabolism, it has not yet been characterized in teak. These results will open new perspectives for studies of diversity, ecology, breeding and genomic programs aiming to understand deeply the biology of this species. RESULTS: We present a widely expressed gene catalog for T. grandis using Illumina technology and the de novo assembly. A total of 462,260 transcripts were obtained, with 1,502 and 931 genes differentially expressed for stem and branch secondary xylem, respectively, during age transition. Analysis of stem and branch secondary xylem indicates substantial similarity in gene ontologies including carbohydrate enzymes, response to stress, protein binding, and allowed us to find transcription factors and heat-shock proteins differentially expressed. TgMYB1 displays a MYB domain and a predicted coiled-coil (CC) domain, while TgMYB2, TgMYB3 and TgMYB4 showed R2R3-MYB domain and grouped with MYBs from several gymnosperms and flowering plants. TgMYB1, TgMYB4 and TgCES presented higher expression in mature secondary xylem, in contrast with TgMYB2, TgHsp1, TgHsp2, TgHsp3, and TgBi whose expression is higher in young lignified tissues. TgMYB3 is expressed at lower level in secondary xylem. CONCLUSIONS: Expression patterns of MYB transcription factors and heat-shock proteins in lignified tissues are dissimilar when tree development was evaluated, obtaining more expression of TgMYB1 and TgMYB4 in lignified tissues of 60-year-old trees, and more expression in TgHsp1, TgHsp2, TgHsp3 and TgBi in stem secondary xylem of 12-year-old trees. We are opening a door for further functional characterization by reverse genetics and marker-assisted selection with those genes. Investigation of some of the key regulators of lignin biosynthesis in teak, however, could be a valuable step towards understanding how rigidity of teak wood and extractives content are different from most other woods. The obtained transcriptome data represents new sequences of T. grandis deposited in public databases, representing an unprecedented opportunity to discover several related-genes associated with secondary xylem such as transcription factors and stress-related genes in a tropical tree.

A genome-wide regulatory framework identifies maize pericarp color1 controlled genes.

Pericarp Color1 (P1) encodes an R2R3-MYB transcription factor responsible for the accumulation of insecticidal flavones in maize (Zea mays) silks and red phlobaphene pigments in pericarps and other floral tissues, which makes P1 an important visual marker. Using genome-wide expression analyses (RNA sequencing) in pericarps and silks of plants with contrasting P1 alleles combined with chromatin immunoprecipitation coupled with high-throughput sequencing, we show here that the regulatory functions of P1 are much broader than the activation of genes corresponding to enzymes in a branch of flavonoid biosynthesis. P1 modulates the expression of several thousand genes, and ∼1500 of them were identified as putative direct targets of P1. Among them, we identified F2H1, corresponding to a P450 enzyme that converts naringenin into 2-hydroxynaringenin, a key branch point in the P1-controlled pathway and the first step in the formation of insecticidal C-glycosyl flavones. Unexpectedly, the binding of P1 to gene regulatory regions can result in both gene activation and repression. Our results indicate that P1 is the major regulator for a set of genes involved in flavonoid biosynthesis and a minor modulator of the expression of a much larger gene set that includes genes involved in primary metabolism and production of other specialized compounds.

From plant gene regulatory grids to network dynamics.

The regulation of gene expression is the most basic level at which genotypes encoded in DNA can manifest themselves into observable phenotypes. In eukaryotes, gene regulatory networks (GRNs) describe the regulatory web through which transcription factors and microRNAs tightly regulate the spatial and temporal expression of genes. In yeast, Escherichia coli, and animals the study of GRNs has uncovered many of the network properties responsible for creating complex regulatory behavior such as organism growth, development, and response to environmental stimuli. In plants, the study of GRNs is just starting to gain momentum thanks to new high quality genomes and the development of new tools for GRN mapping. Here, we review the latest advancements in the study of plant GRNs and describe the tools and techniques used to produce them. We also discuss the emerging field of network dynamics and the methods currently being developed to measure network dynamics and function in plants.

AGRIS: the Arabidopsis Gene Regulatory Information Server, an update.

The Arabidopsis Gene Regulatory Information Server (AGRIS; http://arabidopsis.med.ohio-state.edu/) provides a comprehensive resource for gene regulatory studies in the model plant Arabidopsis thaliana. Three interlinked databases, AtTFDB, AtcisDB and AtRegNet, furnish comprehensive and updated information on transcription factors (TFs), predicted and experimentally verified cis-regulatory elements (CREs) and their interactions, respectively. In addition to significant contributions in the identification of the entire set of TF-DNA interactions, which are the key to understand the gene regulatory networks that govern Arabidopsis gene expression, tools recently incorporated into AGRIS include the complete set of words length 5-15 present in the Arabidopsis genome and the integration of AtRegNet with visualization tools, such as the recently developed ReIN application. All the information in AGRIS is publicly available and downloadable upon registration.

Emergence of switch-like behavior in a large family of simple biochemical networks.

Bistability plays a central role in the gene regulatory networks (GRNs) controlling many essential biological functions, including cellular differentiation and cell cycle control. However, establishing the network topologies that can exhibit bistability remains a challenge, in part due to the exceedingly large variety of GRNs that exist for even a small number of components. We begin to address this problem by employing chemical reaction network theory in a comprehensive in silico survey to determine the capacity for bistability of more than 40,000 simple networks that can be formed by two transcription factor-coding genes and their associated proteins (assuming only the most elementary biochemical processes). We find that there exist reaction rate constants leading to bistability in ∼90% of these GRN models, including several circuits that do not contain any of the TF cooperativity commonly associated with bistable systems, and the majority of which could only be identified as bistable through an original subnetwork-based analysis. A topological sorting of the two-gene family of networks based on the presence or absence of biochemical reactions reveals eleven minimal bistable networks (i.e., bistable networks that do not contain within them a smaller bistable subnetwork). The large number of previously unknown bistable network topologies suggests that the capacity for switch-like behavior in GRNs arises with relative ease and is not easily lost through network evolution. To highlight the relevance of the systematic application of CRNT to bistable network identification in real biological systems, we integrated publicly available protein-protein interaction, protein-DNA interaction, and gene expression data from Saccharomyces cerevisiae, and identified several GRNs predicted to behave in a bistable fashion.

Assessing the regulatory potential of transposable elements using chromatin accessibility profiles of maize transposons.

Transposable elements (TEs) have the potential to create regulatory variation both through the disruption of existing DNA regulatory elements and through the creation of novel DNA regulatory elements. In a species with a large genome, such as maize, many TEs interspersed with genes create opportunities for significant allelic variation due to TE presence/absence polymorphisms among individuals. We used information on putative regulatory elements in combination with knowledge about TE polymorphisms in maize to identify TE insertions that interrupt existing accessible chromatin regions (ACRs) in B73 as well as examples of polymorphic TEs that contain ACRs among four inbred lines of maize including B73, Mo17, W22, and PH207. The TE insertions in three other assembled maize genomes (Mo17, W22, or PH207) that interrupt ACRs that are present in the B73 genome can trigger changes to the chromatin, suggesting the potential for both genetic and epigenetic influences of these insertions. Nearly 20% of the ACRs located over 2 kb from the nearest gene are located within an annotated TE. These are regions of unmethylated DNA that show evidence for functional importance similar to ACRs that are not present within TEs. Using a large panel of maize genotypes, we tested if there is an association between the presence of TE insertions that interrupt, or carry, an ACR and the expression of nearby genes. While most TE polymorphisms are not associated with expression for nearby genes, the TEs that carry ACRs exhibit enrichment for being associated with higher expression of nearby genes, suggesting that these TEs may contribute novel regulatory elements. These analyses highlight the potential for a subset of TEs to rewire transcriptional responses in eukaryotic genomes.

Reconstructing the maize leaf regulatory network using ChIP-seq data of 104 transcription factors.

The transcription regulatory network inside a eukaryotic cell is defined by the combinatorial actions of transcription factors (TFs). However, TF binding studies in plants are too few in number to produce a general picture of this complex network. In this study, we use large-scale ChIP-seq to reconstruct it in the maize leaf, and train machine-learning models to predict TF binding and co-localization. The resulting network covers 77% of the expressed genes, and shows a scale-free topology and functional modularity like a real-world network. TF binding sequence preferences are conserved within family, while co-binding could be key for their binding specificity. Cross-species comparison shows that core network nodes at the top of the transmission of information being more conserved than those at the bottom. This study reveals the complex and redundant nature of the plant transcription regulatory network, and sheds light on its architecture, organizing principle and evolutionary trajectory.

Author Correction: Widespread long-range cis-regulatory elements in the maize genome.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Widespread long-range cis-regulatory elements in the maize genome.

Genetic mapping studies on crops suggest that agronomic traits can be controlled by gene-distal intergenic loci. Despite the biological importance and the potential agronomic utility of these loci, they remain virtually uncharacterized in all crop species to date. Here, we provide genetic, epigenomic and functional molecular evidence to support the widespread existence of gene-distal (hereafter, distal) loci that act as long-range transcriptional cis-regulatory elements (CREs) in the maize genome. Such loci are enriched for euchromatic features that suggest their regulatory functions. Chromatin loops link together putative CREs with genes and recapitulate genetic interactions. Putative CREs also display elevated transcriptional enhancer activities, as measured by self-transcribing active regulatory region sequencing. These results provide functional support for the widespread existence of CREs that act over large genomic distances to control gene expression.

Genome-Wide TSS Identification in Maize.

Regulation of gene expression is a fundamental biological process that relies on transcription factors (TF) recognizing specific cis motifs in the regulatory regions of the genes that they control. In most eukaryotic organisms, cis-regulatory elements are significantly enriched around the transcription start site (TSS). However, different from other genic features, TSSs need to be experimentally determined, becoming then important components of genome annotations. One of the methods for experimentally determining TSSs at the genome-wide level is CAGE (cap analysis of gene expression). This chapter describes how to prepare a CAGE library for sequencing, starting with RNA extraction, library construction, and quality controls before proceed to sequencing in the Illumina platform. We then describe how to use a computational pipeline to determine, from the alignment of CAGE tags, the genome-wide location of TSSs, followed with statistical approaches required to cluster TSSs that operate as transcriptional units, and to determine core promoter properties such as shape. The analyses described here focus on maize, since its large and yet deficiently annotated genome creates some unique challenges, but with some modifications can be easily adopted for other organisms as well.

A Maize Gene Regulatory Network for Phenolic Metabolism.

The translation of the genotype into phenotype, represented for example by the expression of genes encoding enzymes required for the biosynthesis of phytochemicals that are important for interaction of plants with the environment, is largely carried out by transcription factors (TFs) that recognize specific cis-regulatory elements in the genes that they control. TFs and their target genes are organized in gene regulatory networks (GRNs), and thus uncovering GRN architecture presents an important biological challenge necessary to explain gene regulation. Linking TFs to the genes they control, central to understanding GRNs, can be carried out using gene- or TF-centered approaches. In this study, we employed a gene-centered approach utilizing the yeast one-hybrid assay to generate a network of protein-DNA interactions that participate in the transcriptional control of genes involved in the biosynthesis of maize phenolic compounds including general phenylpropanoids, lignins, and flavonoids. We identified 1100 protein-DNA interactions involving 54 phenolic gene promoters and 568 TFs. A set of 11 TFs recognized 10 or more promoters, suggesting a role in coordinating pathway gene expression. The integration of the gene-centered network with information derived from TF-centered approaches provides a foundation for a phenolics GRN characterized by interlaced feed-forward loops that link developmental regulators with biosynthetic genes.

In silico identification of regulatory elements of GRIN1 genes.

The ionotropic receptor of glutamate activated by N-methyl-D-aspartate (iGluR-NMDA) is a multiheteromeric complex constituted by at least three different types of subunits, encoded by seven different genes. The subunits of iGluR-NMDA have a complex system of regulation of their gene expression. Their expression is specific for each type of neural cell, as well as for the age of the organism. Moreover, there are reports that iGluR-NMDA expression is species-specific. Even though this macromolecular complex is very important in physiology and pathology of the central nervous system, knowledge to date about the regulatory elements controlling expression is scarce. We present the results of an in silico prediction of potential regulatory elements, some of which coincide with the few known experimentally. We also present the important differences regarding the presence and the localization of the regulatory elements among human, rat, and mouse species.

Important biological information uncovered in previously unaligned reads from chromatin immunoprecipitation experiments (ChIP-Seq).

Establishing the architecture of gene regulatory networks (GRNs) relies on chromatin immunoprecipitation followed by massively parallel sequencing (ChIP-Seq) methods that provide genome-wide transcription factor binding sites (TFBSs). ChIP-Seq furnishes millions of short reads that, after alignment, describe the genome-wide binding sites of a particular TF. However, in all organisms investigated an average of 40% of reads fail to align to the corresponding genome, with some datasets having as much as 80% of reads failing to align. We describe here the provenance of previously unaligned reads in ChIP-Seq experiments from animals and plants. We show that a substantial portion corresponds to sequences of bacterial and metazoan origin, irrespective of the ChIP-Seq chromatin source. Unforeseen was the finding that 30%-40% of unaligned reads were actually alignable. To validate these observations, we investigated the characteristics of the previously unaligned reads corresponding to TAL1, a human TF involved in lineage specification of hemopoietic cells. We show that, while unmapped ChIP-Seq read datasets contain foreign DNA sequences, additional TFBSs can be identified from the previously unaligned ChIP-Seq reads. Our results indicate that the re-evaluation of previously unaligned reads from ChIP-Seq experiments will significantly contribute to TF target identification and determination of emerging properties of GRNs.