Single-sequence protein structure prediction using a language model and deep learning.

AlphaFold2 and related computational systems predict protein structure using deep learning and co-evolutionary relationships encoded in multiple sequence alignments (MSAs). Despite high prediction accuracy achieved by these systems, challenges remain in (1) prediction of orphan and rapidly evolving proteins for which an MSA cannot be generated; (2) rapid exploration of designed structures; and (3) understanding the rules governing spontaneous polypeptide folding in solution. Here we report development of an end-to-end differentiable recurrent geometric network (RGN) that uses a protein language model (AminoBERT) to learn latent structural information from unaligned proteins. A linked geometric module compactly represents Cα backbone geometry in a translationally and rotationally invariant way. On average, RGN2 outperforms AlphaFold2 and RoseTTAFold on orphan proteins and classes of designed proteins while achieving up to a 106-fold reduction in compute time. These findings demonstrate the practical and theoretical strengths of protein language models relative to MSAs in structure prediction.

Low-N protein engineering with data-efficient deep learning.

Protein engineering has enormous academic and industrial potential. However, it is limited by the lack of experimental assays that are consistent with the design goal and sufficiently high throughput to find rare, enhanced variants. Here we introduce a machine learning-guided paradigm that can use as few as 24 functionally assayed mutant sequences to build an accurate virtual fitness landscape and screen ten million sequences via in silico directed evolution. As demonstrated in two dissimilar proteins, GFP from Aequorea victoria (avGFP) and E. coli strain TEM-1 ß-lactamase, top candidates from a single round are diverse and as active as engineered mutants obtained from previous high-throughput efforts. By distilling information from natural protein sequence landscapes, our model learns a latent representation of 'unnaturalness', which helps to guide search away from nonfunctional sequence neighborhoods. Subsequent low-N supervision then identifies improvements to the activity of interest. In sum, our approach enables efficient use of resource-intensive high-fidelity assays without sacrificing throughput, and helps to accelerate engineered proteins into the fermenter, field and clinic.

Unified rational protein engineering with sequence-based deep representation learning.

Rational protein engineering requires a holistic understanding of protein function. Here, we apply deep learning to unlabeled amino-acid sequences to distill the fundamental features of a protein into a statistical representation that is semantically rich and structurally, evolutionarily and biophysically grounded. We show that the simplest models built on top of this unified representation (UniRep) are broadly applicable and generalize to unseen regions of sequence space. Our data-driven approach predicts the stability of natural and de novo designed proteins, and the quantitative function of molecularly diverse mutants, competitively with the state-of-the-art methods. UniRep further enables two orders of magnitude efficiency improvement in a protein engineering task. UniRep is a versatile summary of fundamental protein features that can be applied across protein engineering informatics.

The G-Box Transcriptional Regulatory Code in Arabidopsis.

Plants have significantly more transcription factor (TF) families than animals and fungi, and plant TF families tend to contain more genes; these expansions are linked to adaptation to environmental stressors. Many TF family members bind to similar or identical sequence motifs, such as G-boxes (CACGTG), so it is difficult to predict regulatory relationships. We determined that the flanking sequences near G-boxes help determine in vitro specificity but that this is insufficient to predict the transcription pattern of genes near G-boxes. Therefore, we constructed a gene regulatory network that identifies the set of bZIPs and bHLHs that are most predictive of the expression of genes downstream of perfect G-boxes. This network accurately predicts transcriptional patterns and reconstructs known regulatory subnetworks. Finally, we present Ara-BOX-cis (araboxcis.org), a Web site that provides interactive visualizations of the G-box regulatory network, a useful resource for generating predictions for gene regulatory relations.

The evening complex coordinates environmental and endogenous signals in Arabidopsis.

Plants maximize their fitness by adjusting their growth and development in response to signals such as light and temperature. The circadian clock provides a mechanism for plants to anticipate events such as sunrise and adjust their transcriptional programmes. However, the underlying mechanisms by which plants coordinate environmental signals with endogenous pathways are not fully understood. Using RNA-sequencing and chromatin immunoprecipitation sequencing experiments, we show that the evening complex (EC) of the circadian clock plays a major role in directly coordinating the expression of hundreds of key regulators of photosynthesis, the circadian clock, phytohormone signalling, growth and response to the environment. We find that the ability of the EC to bind targets genome-wide depends on temperature. In addition, co-occurrence of phytochrome B (phyB) at multiple sites where the EC is bound provides a mechanism for integrating environmental information. Hence, our results show that the EC plays a central role in coordinating endogenous and environmental signals in Arabidopsis.

Tradict enables accurate prediction of eukaryotic transcriptional states from 100 marker genes.

Transcript levels are a critical determinant of the proteome and hence cellular function. Because the transcriptome is an outcome of the interactions between genes and their products, it may be accurately represented by a subset of transcript abundances. We develop a method, Tradict (transcriptome predict), capable of learning and using the expression measurements of a small subset of 100 marker genes to predict transcriptome-wide gene abundances and the expression of a comprehensive, but interpretable list of transcriptional programs that represent the major biological processes and pathways of the cell. By analyzing over 23,000 publicly available RNA-Seq data sets, we show that Tradict is robust to noise and accurate. Coupled with targeted RNA sequencing, Tradict may therefore enable simultaneous transcriptome-wide screening and mechanistic investigation at large scales.

Pseudomonas syringae Type III Effector HopBB1 Promotes Host Transcriptional Repressor Degradation to Regulate Phytohormone Responses and Virulence.

Independently evolved pathogen effectors from three branches of life (ascomycete, eubacteria, and oomycete) converge onto the Arabidopsis TCP14 transcription factor to manipulate host defense. However, the mechanistic basis for defense control via TCP14 regulation is unknown. We demonstrate that TCP14 regulates the plant immune system by transcriptionally repressing a subset of the jasmonic acid (JA) hormone signaling outputs. A previously unstudied Pseudomonas syringae (Psy) type III effector, HopBB1, interacts with TCP14 and targets it to the SCFCOI1 degradation complex by connecting it to the JA signaling repressor JAZ3. Consequently, HopBB1 de-represses the TCP14-regulated subset of JA response genes and promotes pathogen virulence. Thus, HopBB1 fine-tunes host phytohormone crosstalk by precisely manipulating part of the JA regulon to avoid pleiotropic host responses while promoting pathogen proliferation.

Phytochromes function as thermosensors in Arabidopsis.

Plants are responsive to temperature, and some species can distinguish differences of 1°C. In Arabidopsis, warmer temperature accelerates flowering and increases elongation growth (thermomorphogenesis). However, the mechanisms of temperature perception are largely unknown. We describe a major thermosensory role for the phytochromes (red light receptors) during the night. Phytochrome null plants display a constitutive warm-temperature response, and consistent with this, we show in this background that the warm-temperature transcriptome becomes derepressed at low temperatures. We found that phytochrome B (phyB) directly associates with the promoters of key target genes in a temperature-dependent manner. The rate of phyB inactivation is proportional to temperature in the dark, enabling phytochromes to function as thermal timers that integrate temperature information over the course of the night.

Learning Microbial Interaction Networks from Metagenomic Count Data.

Many microbes associate with higher eukaryotes and impact their vitality. To engineer microbiomes for host benefit, we must understand the rules of community assembly and maintenance that, in large part, demand an understanding of the direct interactions among community members. Toward this end, we have developed a Poisson-multivariate normal hierarchical model to learn direct interactions from the count-based output of standard metagenomics sequencing experiments. Our model controls for confounding predictors at the Poisson layer and captures direct taxon-taxon interactions at the multivariate normal layer using an ℓ1 penalized precision matrix. We show in a synthetic experiment that our method handily outperforms state-of-the-art methods such as SparCC and the graphical lasso (glasso). In a real in planta perturbation experiment of a nine-member bacterial community, we show our model, but not SparCC or glasso, correctly resolves a direct interaction structure among three community members that associates with Arabidopsis thaliana roots. We conclude that our method provides a structured, accurate, and distributionally reasonable way of modeling correlated count-based random variables and capturing direct interactions among them.

Variable suites of non-effector genes are co-regulated in the type III secretion virulence regulon across the Pseudomonas syringae phylogeny.

Pseudomonas syringae is a phylogenetically diverse species of Gram-negative bacterial plant pathogens responsible for crop diseases around the world. The HrpL sigma factor drives expression of the major P. syringae virulence regulon. HrpL controls expression of the genes encoding the structural and functional components of the type III secretion system (T3SS) and the type three secreted effector proteins (T3E) that are collectively essential for virulence. HrpL also regulates expression of an under-explored suite of non-type III effector genes (non-T3E), including toxin production systems and operons not previously associated with virulence. We implemented and refined genome-wide transcriptional analysis methods using cDNA-derived high-throughput sequencing (RNA-seq) data to characterize the HrpL regulon from six isolates of P. syringae spanning the diversity of the species. Our transcriptomes, mapped onto both complete and draft genomes, significantly extend earlier studies. We confirmed HrpL-regulation for a majority of previously defined T3E genes in these six strains. We identified two new T3E families from P. syringae pv. oryzae 1_6, a strain within the relatively underexplored phylogenetic Multi-Locus Sequence Typing (MLST) group IV. The HrpL regulons varied among strains in gene number and content across both their T3E and non-T3E gene suites. Strains within MLST group II consistently express the lowest number of HrpL-regulated genes. We identified events leading to recruitment into, and loss from, the HrpL regulon. These included gene gain and loss, and loss of HrpL regulation caused by group-specific cis element mutations in otherwise conserved genes. Novel non-T3E HrpL-regulated genes include an operon that we show is required for full virulence of P. syringae pv. phaseolicola 1448A on French bean. We highlight the power of integrating genomic, transcriptomic, and phylogenetic information to drive concise functional experimentation and to derive better insight into the evolution of virulence across an evolutionarily diverse pathogen species.

The molecular basis of host specialization in bean pathovars of Pseudomonas syringae.

Biotrophic phytopathogens are typically limited to their adapted host range. In recent decades, investigations have teased apart the general molecular basis of intraspecific variation for innate immunity of plants, typically involving receptor proteins that enable perception of pathogen-associated molecular patterns or avirulence elicitors from the pathogen as triggers for defense induction. However, general consensus concerning evolutionary and molecular factors that alter host range across closely related phytopathogen isolates has been more elusive. Here, through genome comparisons and genetic manipulations, we investigate the underlying mechanisms that structure host range across closely related strains of Pseudomonas syringae isolated from different legume hosts. Although type III secretion-independent virulence factors are conserved across these three strains, we find that the presence of two genes encoding type III effectors (hopC1 and hopM1) and the absence of another (avrB2) potentially contribute to host range differences between pathovars glycinea and phaseolicola. These findings reinforce the idea that a complex genetic basis underlies host range evolution in plant pathogens. This complexity is present even in host-microbe interactions featuring relatively little divergence among both hosts and their adapted pathogens.

Repeated treatment with S. aureus superantigens expands the survival rate of Ehrlich ascites tumor bearing mice.

The bacterial superantigen Staphylococcal enterotoxin-A (SEA), produced by some strains of Staphylococcus aureus, causes proliferation of cytotoxic T-lymphocytes and cytokine production in vivo. SEA has been shown to be highly efficient for antibody-targeted superantegen immunotherapy for different tumor models. A candidate B-cell superantigen that has received considerable attention these days is staphylococcal protein-A (PA). It has been shown to possess multiple immunological responses. The anti-tumor property of PA is well documented in the literature in various transplantable tumors of rats and mice. In the present study, we have shown that the T-cell superantigen SEA and B-cell superantigen PA induce immunomodulatory and anti-tumor activity which is strongly protentiated by PA + SEA co-administration. Combination treatment with PA and SEA prolongs the immune response in vivo, limits the development of immunological unresponsiveness and promotes maximum anti-tumor effects to tumor carrying animals, as compared with PA or SEA alone. The immune response after combined therapy is characterized by substantially augmented IFN-gamma, TNF-alpha, Nitric oxide and strong CTL activity. Our data demonstrate that combined PA + SEA therapy induces long-term survival of the animals, carrying the Ehrlich ascites tumor.

Superantigen-induced apoptotic death of tumor cells is mediated by cytotoxic lymphocytes, cytokines, and nitric oxide.

The bacterial superantigen staphylococcal enterotoxin A (SEA) is a potent inducer of CTL activity and cytokine production in vivo. Protein A (PA) of Staphylococcal aureus has been found to have diverse biological response modifying properties and to possess antitumor, antitoxic and antiparasitic effects. In this study we examined the anti-tumor effect of these two superantigens used separately as well as in combination in mice carrying the Ehrlich ascites tumor. With combined treatment, DNA cell cycle analysis of tumor cells showed a significant (P < 0.05) percentage of tumor cell death. Levels of the soluble mediators TNF-alpha, IFN-gamma and IL-1 as well as NO were elevated. Additionally, CD4(+) and CD8(+) specific T cells in spleen, thymus and PBMC in tumor carrying mice were increased (P < 0.01). Our data altogether suggests that enhanced tumor cell death is caused by the increased CTL activity, cytokine and nitric oxide levels, in response to the combined effect of SEA + PA.