An End-to-End Platform for Digital Pathology Using Hyperspectral Autofluorescence Microscopy and Deep Learning-Based Virtual Histology.

Conventional histopathology involves expensive and labor-intensive processes that often consume tissue samples, rendering them unavailable for other analyses. We present a novel end-to-end workflow for pathology powered by hyperspectral microscopy and deep learning. First, we developed a custom hyperspectral microscope to nondestructively image the autofluorescence of unstained tissue sections. We then trained a deep learning model to use autofluorescence to generate virtual histologic stains, which avoids the cost and variability of chemical staining procedures and conserves tissue samples. We showed that the virtual images reproduce the histologic features present in the real-stained images using a randomized nonalcoholic steatohepatitis (NASH) scoring comparison study, where both real and virtual stains are scored by pathologists (D.T., A.D.B., R.K.P.). The test showed moderate-to-good concordance between pathologists' scoring on corresponding real and virtual stains. Finally, we developed deep learning-based models for automated NASH Clinical Research Network score prediction. We showed that the end-to-end automated pathology platform is comparable with an independent panel of pathologists for NASH Clinical Research Network scoring when evaluated against the expert pathologist consensus scores. This study provides proof of concept for this virtual staining strategy, which could improve cost, efficiency, and reliability in pathology and enable novel approaches to spatial biology research.

Comparative analysis of machine learning approaches to classify tumor mutation burden in lung adenocarcinoma using histopathology images.

Both histologic subtypes and tumor mutation burden (TMB) represent important biomarkers in lung cancer, with implications for patient prognosis and treatment decisions. Typically, TMB is evaluated by comprehensive genomic profiling but this requires use of finite tissue specimens and costly, time-consuming laboratory processes. Histologic subtype classification represents an established component of lung adenocarcinoma histopathology, but can be challenging and is associated with substantial inter-pathologist variability. Here we developed a deep learning system to both classify histologic patterns in lung adenocarcinoma and predict TMB status using de-identified Hematoxylin and Eosin (H&E) stained whole slide images. We first trained a convolutional neural network to map histologic features across whole slide images of lung cancer resection specimens. On evaluation using an external data source, this model achieved patch-level area under the receiver operating characteristic curve (AUC) of 0.78-0.98 across nine histologic features. We then integrated the output of this model with clinico-demographic data to develop an interpretable model for TMB classification. The resulting end-to-end system was evaluated on 172 held out cases from TCGA, achieving an AUC of 0.71 (95% CI 0.63-0.80). The benefit of using histologic features in predicting TMB is highlighted by the significant improvement this approach offers over using the clinical features alone (AUC of 0.63 [95% CI 0.53-0.72], p = 0.002). Furthermore, we found that our histologic subtype-based approach achieved performance similar to that of a weakly supervised approach (AUC of 0.72 [95% CI 0.64-0.80]). Together these results underscore that incorporating histologic patterns in biomarker prediction for lung cancer provides informative signals, and that interpretable approaches utilizing these patterns perform comparably with less interpretable, weakly supervised approaches.

Expansion of RiPP biosynthetic space through integration of pan-genomics and machine learning uncovers a novel class of lanthipeptides.

Microbial natural products constitute a wide variety of chemical compounds, many which can have antibiotic, antiviral, or anticancer properties that make them interesting for clinical purposes. Natural product classes include polyketides (PKs), nonribosomal peptides (NRPs), and ribosomally synthesized and post-translationally modified peptides (RiPPs). While variants of biosynthetic gene clusters (BGCs) for known classes of natural products are easy to identify in genome sequences, BGCs for new compound classes escape attention. In particular, evidence is accumulating that for RiPPs, subclasses known thus far may only represent the tip of an iceberg. Here, we present decRiPPter (Data-driven Exploratory Class-independent RiPP TrackER), a RiPP genome mining algorithm aimed at the discovery of novel RiPP classes. DecRiPPter combines a Support Vector Machine (SVM) that identifies candidate RiPP precursors with pan-genomic analyses to identify which of these are encoded within operon-like structures that are part of the accessory genome of a genus. Subsequently, it prioritizes such regions based on the presence of new enzymology and based on patterns of gene cluster and precursor peptide conservation across species. We then applied decRiPPter to mine 1,295 Streptomyces genomes, which led to the identification of 42 new candidate RiPP families that could not be found by existing programs. One of these was studied further and elucidated as a representative of a novel subfamily of lanthipeptides, which we designate class V. The 2D structure of the new RiPP, which we name pristinin A3 (1), was solved using nuclear magnetic resonance (NMR), tandem mass spectrometry (MS/MS) data, and chemical labeling. Two previously unidentified modifying enzymes are proposed to create the hallmark lanthionine bridges. Taken together, our work highlights how novel natural product families can be discovered by methods going beyond sequence similarity searches to integrate multiple pathway discovery criteria.

Genetic interaction mapping informs integrative structure determination of protein complexes.

Determining structures of protein complexes is crucial for understanding cellular functions. Here, we describe an integrative structure determination approach that relies on in vivo measurements of genetic interactions. We construct phenotypic profiles for point mutations crossed against gene deletions or exposed to environmental perturbations, followed by converting similarities between two profiles into an upper bound on the distance between the mutated residues. We determine the structure of the yeast histone H3-H4 complex based on ~500,000 genetic interactions of 350 mutants. We then apply the method to subunits Rpb1-Rpb2 of yeast RNA polymerase II and subunits RpoB-RpoC of bacterial RNA polymerase. The accuracy is comparable to that based on chemical cross-links; using restraints from both genetic interactions and cross-links further improves model accuracy and precision. The approach provides an efficient means to augment integrative structure determination with in vivo observations.

High-quality MS/MS spectrum prediction for data-dependent and data-independent acquisition data analysis.

Peptide fragmentation spectra are routinely predicted in the interpretation of mass-spectrometry-based proteomics data. However, the generation of fragment ions has not been understood well enough for scientists to estimate fragment ion intensities accurately. Here, we demonstrate that machine learning can predict peptide fragmentation patterns in mass spectrometers with accuracy within the uncertainty of measurement. Moreover, analysis of our models reveals that peptide fragmentation depends on long-range interactions within a peptide sequence. We illustrate the utility of our models by applying them to the analysis of both data-dependent and data-independent acquisition datasets. In the former case, we observe a q-value-dependent increase in the total number of peptide identifications. In the latter case, we confirm that the use of predicted tandem mass spectrometry spectra is nearly equivalent to the use of spectra from experimental libraries.

Assessing Exhaustiveness of Stochastic Sampling for Integrative Modeling of Macromolecular Structures.

Modeling of macromolecular structures involves structural sampling guided by a scoring function, resulting in an ensemble of good-scoring models. By necessity, the sampling is often stochastic, and must be exhaustive at a precision sufficient for accurate modeling and assessment of model uncertainty. Therefore, the very first step in analyzing the ensemble is an estimation of the highest precision at which the sampling is exhaustive. Here, we present an objective and automated method for this task. As a proxy for sampling exhaustiveness, we evaluate whether two independently and stochastically generated sets of models are sufficiently similar. The protocol includes testing 1) convergence of the model score, 2) whether model scores for the two samples were drawn from the same parent distribution, 3) whether each structural cluster includes models from each sample proportionally to its size, and 4) whether there is sufficient structural similarity between the two model samples in each cluster. The evaluation also provides the sampling precision, defined as the smallest clustering threshold that satisfies the third, most stringent test. We validate the protocol with the aid of enumerated good-scoring models for five illustrative cases of binary protein complexes. Passing the proposed four tests is necessary, but not sufficient for thorough sampling. The protocol is general in nature and can be applied to the stochastic sampling of any set of models, not just structural models. In addition, the tests can be used to stop stochastic sampling as soon as exhaustiveness at desired precision is reached, thereby improving sampling efficiency; they may also help in selecting a model representation that is sufficiently detailed to be informative, yet also sufficiently coarse for sampling to be exhaustive.

Immunoproteasome functions explained by divergence in cleavage specificity and regulation.

The immunoproteasome (iP) has been proposed to perform specialized roles in MHC class I antigen presentation, cytokine modulation, and T cell differentiation and has emerged as a promising therapeutic target for autoimmune disorders and cancer. However, divergence in function between the iP and the constitutive proteasome (cP) has been unclear. A global peptide library-based screening strategy revealed that the proteasomes have overlapping but distinct substrate specificities. Differing iP specificity alters the quantity of production of certain MHC I epitopes but does not appear to be preferentially suited for antigen presentation. Furthermore, iP specificity was found to have likely arisen through genetic drift from the ancestral cP. Specificity differences were exploited to develop isoform-selective substrates. Cellular profiling using these substrates revealed that divergence in regulation of the iP balances its relative contribution to proteasome capacity in immune cells, resulting in selective recovery from inhibition. These findings have implications for iP-targeted therapeutic development.

The proteasome-interacting Ecm29 protein disassembles the 26S proteasome in response to oxidative stress.

Oxidative stress has been implicated in multiple human neurological and other disorders. Proteasomes are multi-subunit proteases critical for the removal of oxidatively damaged proteins. To understand stress-associated human pathologies, it is important to uncover the molecular events underlying the regulation of proteasomes upon oxidative stress. To this end, we investigated H2O2 stress-induced molecular changes of the human 26S proteasome and determined that stress-induced 26S proteasome disassembly is conserved from yeast to human. Moreover, we developed and employed a new proteomic approach, XAP (in vivo cross-linking-assisted affinity purification), coupled with stable isotope labeling with amino acids in cell culture (SILAC)-based quantitative MS, to capture and quantify several weakly bound proteasome-interacting proteins and examine their roles in stress-mediated proteasomal remodeling. Our results indicate that the adapter protein Ecm29 is the main proteasome-interacting protein responsible for stress-triggered remodeling of the 26S proteasome in human cells. Importantly, using a disuccinimidyl sulfoxide-based cross-linking MS platform, we mapped the interactions of Ecm29 within itself and with proteasome subunits and determined the architecture of the Ecm29-proteasome complex with integrative structure modeling. These results enabled us to propose a structural model in which Ecm29 intrudes on the interaction between the 20S core particle and the 19S regulatory particle in the 26S proteasome, disrupting the proteasome structure in response to oxidative stress.

Molecular Details Underlying Dynamic Structures and Regulation of the Human 26S Proteasome.

The 26S proteasome is the macromolecular machine responsible for ATP/ubiquitin dependent degradation. As aberration in proteasomal degradation has been implicated in many human diseases, structural analysis of the human 26S proteasome complex is essential to advance our understanding of its action and regulation mechanisms. In recent years, cross-linking mass spectrometry (XL-MS) has emerged as a powerful tool for elucidating structural topologies of large protein assemblies, with its unique capability of studying protein complexes in cells. To facilitate the identification of cross-linked peptides, we have previously developed a robust amine reactive sulfoxide-containing MS-cleavable cross-linker, disuccinimidyl sulfoxide (DSSO). To better understand the structure and regulation of the human 26S proteasome, we have established new DSSO-based in vivo and in vitro XL-MS workflows by coupling with HB-tag based affinity purification to comprehensively examine protein-protein interactions within the 26S proteasome. In total, we have identified 447 unique lysine-to-lysine linkages delineating 67 interprotein and 26 intraprotein interactions, representing the largest cross-link dataset for proteasome complexes. In combination with EM maps and computational modeling, the architecture of the 26S proteasome was determined to infer its structural dynamics. In particular, three proteasome subunits Rpn1, Rpn6, and Rpt6 displayed multiple conformations that have not been previously reported. Additionally, cross-links between proteasome subunits and 15 proteasome interacting proteins including 9 known and 6 novel ones have been determined to demonstrate their physical interactions at the amino acid level. Our results have provided new insights on the dynamics of the 26S human proteasome and the methodologies presented here can be applied to study other protein complexes.

CryptoSite: Expanding the Druggable Proteome by Characterization and Prediction of Cryptic Binding Sites.

Many proteins have small-molecule binding pockets that are not easily detectable in the ligand-free structures. These cryptic sites require a conformational change to become apparent; a cryptic site can therefore be defined as a site that forms a pocket in a holo structure, but not in the apo structure. Because many proteins appear to lack druggable pockets, understanding and accurately identifying cryptic sites could expand the set of drug targets. Previously, cryptic sites were identified experimentally by fragment-based ligand discovery and computationally by long molecular dynamics simulations and fragment docking. Here, we begin by constructing a set of structurally defined apo-holo pairs with cryptic sites. Next, we comprehensively characterize the cryptic sites in terms of their sequence, structure, and dynamics attributes. We find that cryptic sites tend to be as conserved in evolution as traditional binding pockets but are less hydrophobic and more flexible. Relying on this characterization, we use machine learning to predict cryptic sites with relatively high accuracy (for our benchmark, the true positive and false positive rates are 73% and 29%, respectively). We then predict cryptic sites in the entire structurally characterized human proteome (11,201 structures, covering 23% of all residues in the proteome). CryptoSite increases the size of the potentially "druggable" human proteome from ~40% to ~78% of disease-associated proteins. Finally, to demonstrate the utility of our approach in practice, we experimentally validate a cryptic site in protein tyrosine phosphatase 1B using a covalent ligand and NMR spectroscopy. The CryptoSite Web server is available at http://salilab.org/cryptosite.

Architecture of the Human and Yeast General Transcription and DNA Repair Factor TFIIH.

TFIIH is essential for both RNA polymerase II transcription and DNA repair, and mutations in TFIIH can result in human disease. Here, we determine the molecular architecture of human and yeast TFIIH by an integrative approach using chemical crosslinking/mass spectrometry (CXMS) data, biochemical analyses, and previously published electron microscopy maps. We identified four new conserved "topological regions" that function as hubs for TFIIH assembly and more than 35 conserved topological features within TFIIH, illuminating a network of interactions involved in TFIIH assembly and regulation of its activities. We show that one of these conserved regions, the p62/Tfb1 Anchor region, directly interacts with the DNA helicase subunit XPD/Rad3 in native TFIIH and is required for the integrity and function of TFIIH. We also reveal the structural basis for defects in patients with xeroderma pigmentosum and trichothiodystrophy, with mutations found at the interface between the p62 Anchor region and the XPD subunit.

IMG-ABC: A Knowledge Base To Fuel Discovery of Biosynthetic Gene Clusters and Novel Secondary Metabolites.

UNLABELLED: In the discovery of secondary metabolites, analysis of sequence data is a promising exploration path that remains largely underutilized due to the lack of computational platforms that enable such a systematic approach on a large scale. In this work, we present IMG-ABC (https://img.jgi.doe.gov/abc), an atlas of biosynthetic gene clusters within the Integrated Microbial Genomes (IMG) system, which is aimed at harnessing the power of "big" genomic data for discovering small molecules. IMG-ABC relies on IMG's comprehensive integrated structural and functional genomic data for the analysis of biosynthetic gene clusters (BCs) and associated secondary metabolites (SMs). SMs and BCs serve as the two main classes of objects in IMG-ABC, each with a rich collection of attributes. A unique feature of IMG-ABC is the incorporation of both experimentally validated and computationally predicted BCs in genomes as well as metagenomes, thus identifying BCs in uncultured populations and rare taxa. We demonstrate the strength of IMG-ABC's focused integrated analysis tools in enabling the exploration of microbial secondary metabolism on a global scale, through the discovery of phenazine-producing clusters for the first time in Alphaproteobacteria. IMG-ABC strives to fill the long-existent void of resources for computational exploration of the secondary metabolism universe; its underlying scalable framework enables traversal of uncovered phylogenetic and chemical structure space, serving as a doorway to a new era in the discovery of novel molecules. IMPORTANCE: IMG-ABC is the largest publicly available database of predicted and experimental biosynthetic gene clusters and the secondary metabolites they produce. The system also includes powerful search and analysis tools that are integrated with IMG's extensive genomic/metagenomic data and analysis tool kits. As new research on biosynthetic gene clusters and secondary metabolites is published and more genomes are sequenced, IMG-ABC will continue to expand, with the goal of becoming an essential component of any bioinformatic exploration of the secondary metabolism world.

Scoring Large-Scale Affinity Purification Mass Spectrometry Datasets with MiST.

High-throughput Affinity Purification Mass Spectrometry (AP-MS) experiments can identify a large number of protein interactions, but only a fraction of these interactions are biologically relevant. Here, we describe a comprehensive computational strategy to process raw AP-MS data, perform quality controls, and prioritize biologically relevant bait-prey pairs in a set of replicated AP-MS experiments with Mass spectrometry interaction STatistics (MiST). The MiST score is a linear combination of prey quantity (abundance), abundance invariability across repeated experiments (reproducibility), and prey uniqueness relative to other baits (specificity). We describe how to run the full MiST analysis pipeline in an R environment and discuss a number of configurable options that allow the lay user to convert any large-scale AP-MS data into an interpretable, biologically relevant protein-protein interaction network.

A systematic computational analysis of biosynthetic gene cluster evolution: lessons for engineering biosynthesis.

Bacterial secondary metabolites are widely used as antibiotics, anticancer drugs, insecticides and food additives. Attempts to engineer their biosynthetic gene clusters (BGCs) to produce unnatural metabolites with improved properties are often frustrated by the unpredictability and complexity of the enzymes that synthesize these molecules, suggesting that genetic changes within BGCs are limited by specific constraints. Here, by performing a systematic computational analysis of BGC evolution, we derive evidence for three findings that shed light on the ways in which, despite these constraints, nature successfully invents new molecules: 1) BGCs for complex molecules often evolve through the successive merger of smaller sub-clusters, which function as independent evolutionary entities. 2) An important subset of polyketide synthases and nonribosomal peptide synthetases evolve by concerted evolution, which generates sets of sequence-homogenized domains that may hold promise for engineering efforts since they exhibit a high degree of functional interoperability, 3) Individual BGC families evolve in distinct ways, suggesting that design strategies should take into account family-specific functional constraints. These findings suggest novel strategies for using synthetic biology to rationally engineer biosynthetic pathways.

Covalent docking of large libraries for the discovery of chemical probes.

Chemical probes that form a covalent bond with a protein target often show enhanced selectivity, potency and utility for biological studies. Despite these advantages, protein-reactive compounds are usually avoided in high-throughput screening campaigns. Here we describe a general method (DOCKovalent) for screening large virtual libraries of electrophilic small molecules. We apply this method prospectively to discover reversible covalent fragments that target distinct protein nucleophiles, including the catalytic serine of AmpC ß-lactamase and noncatalytic cysteines in RSK2, MSK1 and JAK3 kinases. We identify submicromolar to low-nanomolar hits with high ligand efficiency, cellular activity and selectivity, including what are to our knowledge the first reported reversible covalent inhibitors of JAK3. Crystal structures of inhibitor complexes with AmpC and RSK2 confirm the docking predictions and guide further optimization. As covalent virtual screening may have broad utility for the rapid discovery of chemical probes, we have made the method freely available through an automated web server (http://covalent.docking.org/).

Affinity purification-mass spectrometry and network analysis to understand protein-protein interactions.

By determining protein-protein interactions in normal, diseased and infected cells, we can improve our understanding of cellular systems and their reaction to various perturbations. In this protocol, we discuss how to use data obtained in affinity purification-mass spectrometry (AP-MS) experiments to generate meaningful interaction networks and effective figures. We begin with an overview of common epitope tagging, expression and AP practices, followed by liquid chromatography-MS (LC-MS) data collection. We then provide a detailed procedure covering a pipeline approach to (i) pre-processing the data by filtering against contaminant lists such as the Contaminant Repository for Affinity Purification (CRAPome) and normalization using the spectral index (SIN) or normalized spectral abundance factor (NSAF); (ii) scoring via methods such as MiST, SAInt and CompPASS; and (iii) testing the resulting scores. Data formats familiar to MS practitioners are then transformed to those most useful for network-based analyses. The protocol also explores methods available in Cytoscape to visualize and analyze these types of interaction data. The scoring pipeline can take anywhere from 1 d to 1 week, depending on one's familiarity with the tools and data peculiarities. Similarly, the network analysis and visualization protocol in Cytoscape takes 2-4 h to complete with the provided sample data, but we recommend taking days or even weeks to explore one's data and find the right questions.

Discovery and characterization of gut microbiota decarboxylases that can produce the neurotransmitter tryptamine.

Several recent studies describe the influence of the gut microbiota on host brain and behavior. However, the mechanisms responsible for microbiota-nervous system interactions are largely unknown. Using a combination of genetics, biochemistry, and crystallography, we identify and characterize two phylogenetically distinct enzymes found in the human microbiome that decarboxylate tryptophan to form the ß-arylamine neurotransmitter tryptamine. Although this enzymatic activity is exceedingly rare among bacteria more broadly, analysis of the Human Microbiome Project data demonstrate that at least 10% of the human population harbors at least one bacterium encoding a tryptophan decarboxylase in their gut community. Our results uncover a previously unrecognized enzymatic activity that can give rise to host-modulatory compounds and suggests a potential direct mechanism by which gut microbiota can influence host physiology, including behavior.

A systematic analysis of biosynthetic gene clusters in the human microbiome reveals a common family of antibiotics.

In complex biological systems, small molecules often mediate microbe-microbe and microbe-host interactions. Using a systematic approach, we identified 3,118 small-molecule biosynthetic gene clusters (BGCs) in genomes of human-associated bacteria and studied their representation in 752 metagenomic samples from the NIH Human Microbiome Project. Remarkably, we discovered that BGCs for a class of antibiotics in clinical trials, thiopeptides, are widely distributed in genomes and metagenomes of the human microbiota. We purified and solved the structure of a thiopeptide antibiotic, lactocillin, from a prominent member of the vaginal microbiota. We demonstrate that lactocillin has potent antibacterial activity against a range of Gram-positive vaginal pathogens, and we show that lactocillin and other thiopeptide BGCs are expressed in vivo by analyzing human metatranscriptomic sequencing data. Our findings illustrate the widespread distribution of small-molecule-encoding BGCs in the human microbiome, and they demonstrate the bacterial production of drug-like molecules in humans. PAPERCLIP:

Molecular architecture of the 40S⋠eIF1⋠eIF3 translation initiation complex.

Eukaryotic translation initiation requires the recruitment of the large, multiprotein eIF3 complex to the 40S ribosomal subunit. We present X-ray structures of all major components of the minimal, six-subunit Saccharomyces cerevisiae eIF3 core. These structures, together with electron microscopy reconstructions, cross-linking coupled to mass spectrometry, and integrative structure modeling, allowed us to position and orient all eIF3 components on the 40S⋅eIF1 complex, revealing an extended, modular arrangement of eIF3 subunits. Yeast eIF3 engages 40S in a clamp-like manner, fully encircling 40S to position key initiation factors on opposite ends of the mRNA channel, providing a platform for the recruitment, assembly, and regulation of the translation initiation machinery. The structures of eIF3 components reported here also have implications for understanding the architecture of the mammalian 43S preinitiation complex and the complex of eIF3, 40S, and the hepatitis C internal ribosomal entry site RNA.