Cellular processes are carried out by many interacting genes and their study and optimization requires multiple levers by which they can be independently controlled. The most common method is via a genetically-encoded sensor that responds to a small molecule (an "inducible system"). However, these sensors are often suboptimal, exhibiting high background expression and low dynamic range. Further, using multiple sensors in one cell is limited by cross-talk and the taxing of cellular resources. Here, we have developed a directed evolution strategy to simultaneously select for less background, high dynamic range, increased sensitivity, and low crosstalk. Libraries of the regulatory protein and output promoter are built based on random and rationally-guided mutations. This is applied to generate a set of 12 high-performance sensors, which exhibit >100-fold induction with low background and crossreactivity. These are combined to build a single "sensor array" and inserted into the genomes of E. coli MG1655 (wild-type), DH10B (cloning), and BL21 (protein expression). These "Marionette" strains allow for the independent control of gene expression using 2,4-diacetylphophloroglucinol (DAPG), cuminic acid (Cuma), 3-oxohexanoyl-homoserine lactone (OC6), vanillic acid (Van), isopropyl -D-1thiogalactopyranoside (IPTG), anhydrotetracycline (aTc), L-arabinose (Ara), choline chloride (Cho), naringenin (Nar), 3,4-dihydroxybenzoic acid (DHBA), sodium salicylate (Sal), and 3hydroxytetradecanoyl-homoserine lactone (OHC14). Advances in biology are often tied to new methods that use external stimuli to control the levels of gene expression 1-3. Pioneered in the early 1980s, so-called inducible systems were developed that allow genes to be turned on by adding a small molecule inducer to the growth media 4. These consist of a protein transcription factor (e.g., LacI) whose binding to a DNA operator in a promoter is controlled by the inducer (e.g., IPTG). Initially co-opted from natural regulatory networks, over the years many versions were designed to improve performance. In the 1990s, additional systems were developed that responded to other inducers, notably arabinose and aTc, which became common tools in the field. In 1997, Lutz and Bujard published a seminal paper that combined three (IPTG, arabinose, aTc) that could be easily interchanged on a two-plasmid system 5. Its organizational simplicity, compatibility, and quantified response functions were revolutionary. Beyond providing a new tool to biologists to control multiple genes with independent "strings," it facilitated researchers with quantitative backgrounds to enter biology 6-7. Armed with the new ability to control two genes with precision, physicists and engineers built the first synthetic genetic circuits, performed single molecule experiments inside cells, deconstructed the origins of noise in gene expression, determined how enzyme balancing impacts metabolic flux, elucidated rules underlying the assembly of molecular machines, and built synthetic symbiotic microbial commu...

Traditional natural products discovery using a combination of live/ dead screening followed by iterative bioassay-guided fractionation affords no information about compound structure or mode of action until late in the discovery process. This leads to high rates of rediscovery and low probabilities of finding compounds with unique biological and/or chemical properties. By integrating imagebased phenotypic screening in HeLa cells with high-resolution untargeted metabolomics analysis, we have developed a new platform, termed Compound Activity Mapping, that is capable of directly predicting the identities and modes of action of bioactive constituents for any complex natural product extract library. This new tool can be used to rapidly identify novel bioactive constituents and provide predictions of compound modes of action directly from primary screening data. This approach inverts the natural products discovery process from the existing "grind and find" model to a targeted, hypothesis-driven discovery model where the chemical features and biological function of bioactive metabolites are known early in the screening workflow, and lead compounds can be rationally selected based on biological and/or chemical novelty. We demonstrate the utility of the Compound Activity Mapping platform by combining 10,977 mass spectral features and 58,032 biological measurements from a library of 234 natural products extracts and integrating these two datasets to identify 13 clusters of fractions containing 11 known compound families and four new compounds. Using Compound Activity Mapping we discovered the quinocinnolinomycins, a new family of natural products with a unique carbon skeleton that cause endoplasmic reticulum stress.metabolomics | natural products | image-based screening | bioactive small molecules | informatics

Cyclic peptide natural products contain a variety of conserved, nonproteinogenic structural elements such as d-amino acids and amide N-methylation. In addition, many cyclic peptides incorporate γ-amino acids and other elements derived from polyketide synthases. We hypothesized that the position and orientation of these extended backbone elements impact the ADME properties of these hybrid molecules, especially their ability to cross cell membranes and avoid metabolic degradation. Here we report the synthesis of cyclic hexapeptide diastereomers containing γ-amino acids (e.g., statines) and systematically investigate their structure-permeability relationships. These compounds were much more water-soluble and, in many cases, were both more membrane permeable and more stable to liver microsomes than a similar non-statine-containing derivative. Permeability correlated well with the extent of intramolecular hydrogen bonding observed in the solution structures determined in the low-dielectric solvent CDCl3, and one compound showed an oral bioavailability of 21% in rat. Thus, the incorporation of γ-amino acids offers a route to increase backbone diversity and improve ADME properties in cyclic peptide scaffolds.

Cytological profiling (CP) is an unbiased image-based screening technique that uses automated microscopy and image analysis to profile compounds based on numerous quantifiable phenotypic features. We used CP to evaluate a library of nearly 500 compounds with documented mechanisms of action (MOAs) spanning a wide range of biological pathways. We developed informatics techniques for generating dosage-independent phenotypic "fingerprints" for each compound, and for quantifying the likelihood that a compound's CP fingerprint corresponds to its annotated MOA. We identified groups of features that distinguish classes with closely related phenotypes, such as microtubule poisons vs. HSP90 inhibitors, and DNA synthesis vs. proteasome inhibitors. We tested several cases in which cytological profiles indicated novel mechanisms, including a tyrphostin kinase inhibitor involved in mitochondrial uncoupling, novel microtubule poisons, and a nominal PPAR-gamma ligand that acts as a proteasome inhibitor, using independent biochemical assays to confirm the MOAs predicted by the CP signatures. We also applied maximal-information statistics to identify correlations between cytological features and kinase inhibitory activities by combining the CP fingerprints of 24 kinase inhibitors with published data on their specificities against a diverse panel of kinases. The resulting analysis suggests a strategy for probing the biological functions of specific kinases by compiling cytological data from inhibitors of varying specificities.

Collagenases are the principal enzymes responsible for the degradation of collagens during embryonic development, wound healing, and cancer metastasis. However, the mechanism by which these enzymes disrupt the highly chemically and structurally stable collagen triple helix remains incompletely understood. We used a single-molecule magnetic tweezers assay to characterize the cleavage of heterotrimeric collagen I by both the human collagenase matrix metalloproteinase-1 (MMP-1) and collagenase from Clostridium histolyticum. We observe that the application of 16 pN of force causes an 8-fold increase in collagen proteolysis rates by MMP-1 but does not affect cleavage rates by Clostridium collagenase. Quantitative analysis of these data allows us to infer the structural changes in collagen associated with proteolytic cleavage by both enzymes. Our data support a model in which MMP-1 cuts a transient, stretched conformation of its recognition site. In contrast, our findings suggest that Clostridium collagenase is able to cleave the fully wound collagen triple helix, accounting for its lack of force sensitivity and low sequence specificity. We observe that the cleavage of heterotrimeric collagen is less force sensitive than the proteolysis of a homotrimeric collagen model peptide, consistent with studies suggesting that the MMP-1 recognition site in heterotrimeric collagen I is partially unwound at equilibrium.