AI is a viable alternative to high throughput screening: a 318-target study

High throughput screening (HTS) is routinely used to identify bioactive small molecules. This requires physical compounds, which limits coverage of accessible chemical space. Computational approaches combined with vast on-demand chemical libraries can access far greater chemical space, provided that the predictive accuracy is sufficient to identify useful molecules. Through the largest and most diverse virtual HTS campaign reported to date, comprising 318 individual projects, we demonstrate that our AtomNet® convolutional neural network successfully finds novel hits across every major therapeutic area and protein class. We address historical limitations of computational screening by demonstrating success for target proteins without known binders, high-quality X-ray crystal structures, or manual cherry-picking of compounds. We show that the molecules selected by the AtomNet® model are novel drug-like scaffolds rather than minor modifications to known bioactive compounds. Our empirical results suggest that computational methods can substantially replace HTS as the first step of small-molecule drug discovery.

Pyrazinamide triggers degradation of its target aspartate decarboxylase

Pyrazinamide is a sterilizing first-line tuberculosis drug. Genetic, metabolomic and biophysical analyses previously demonstrated that pyrazinoic acid, the bioactive form of the prodrug pyrazinamide (PZA), interrupts biosynthesis of coenzyme A in Mycobacterium tuberculosis by binding to aspartate decarboxylase PanD. While most drugs act by inhibiting protein function upon target binding, we find here that pyrazinoic acid is only a weak enzyme inhibitor. We show that binding of pyrazinoic acid to PanD triggers degradation of the protein by the caseinolytic protease ClpC1-ClpP. Thus, the old tuberculosis drug pyrazinamide exerts antibacterial activity by acting as a target degrader, a mechanism of action that has recently emerged as a successful strategy in drug discovery across disease indications. Our findings provide the basis for the rational discovery of next generation PZA.

It has been shown that the bioactive component of pyrazinamide, pyrazinoic acid (POA), blocks coenzyme A biosynthesis in M. tuberculosis by binding to the aspartate decarboxylase PanD. Here the authors show that pyrazinamide triggers degradation of PanD by stimulating its degradation by the caseinolytic protease Clp.

Development of high throughput screening methods for inhibitors of ClpC1P1P2 from Mycobacteria tuberculosis

Tuberculosis affects about 100 million people worldwide and causes nearly 2 million deaths annually. It has been estimated that one third of all humans is infected with latent Mycobacterium tuberculosis (Mtb). Moreover, Mtb has become increasingly resistant to available antibiotics. Consequently, it is important to identify and characterize new therapeutic targets in Mtb and to synthesize selective inhibitors. ClpP1, ClpP2 and their associated regulatory ATPases, ClpX and ClpC1 are required for the growth of Mtb and for its virulence during murine infection and are highly attractive drug targets, especially since they are not present in the cytosol of mammalian cells, and they differ markedly from the mitochondrial ClpP complex. The importance of these proteins in Mtb is emphasized by the existence of several natural antibiotics targeting this system. In order to find new inhibitors of ClpC1P1P2 system, we developed an assay based on the ATP-dependent degradation of a fluorescent protein substrate. The hits obtained were further characterized with a set of secondary assays to identify precise targets within a complex. A large library of compounds was screened and led to the identification of a ClpC1 ATPase inhibitor demonstrating that this approach can be used in future searches for anti-TB agents.

An Antibacterial β-Lactone Kills Mycobacterium tuberculosis by Disrupting Mycolic Acid Biosynthesis

The spread of antibiotic resistance is a major challenge for the treatment of Mycobacterium tuberculosis infections. In addition, the efficacy of drugs is often limited by the restricted permeability of the mycomembrane. Frontline antibiotics inhibit mycomembrane biosynthesis, leading to rapid cell death. Inspired by this mechanism, we exploited β-lactones as putative mycolic acid mimics to block serine hydrolases involved in their biosynthesis. Among a collection of β-lactones, we found one hit with potent anti-mycobacterial and bactericidal activity. Chemical proteomics using an alkynylated probe identified Pks13 and Ag85 serine hydrolases as major targets. Validation through enzyme assays and customized ¹³ C metabolite profiling showed that both targets are functionally impaired by the β-lactone. Co-administration with front-line antibiotics enhanced the potency against M. tuberculosis by more than 100-fold, thus demonstrating the therapeutic potential of targeting mycomembrane biosynthesis serine hydrolases.

Acyldepsipeptide antibiotics kill mycobacteria by preventing the physiological functions of the ClpP1P2 protease

The Clp protease complex in Mycobacterium tuberculosis is unusual in its composition, functional importance and activation mechanism. Whilst most bacterial species contain a single ClpP protein that is dispensable for normal growth, mycobacteria have two ClpPs, ClpP1 and ClpP2, which are essential for viability and together form the ClpP1P2 tetradecamer. Acyldepsipeptide antibiotics of the ADEP class inhibit the growth of Gram-positive firmicutes by activating ClpP and causing unregulated protein degradation. Here we show that, in contrast, mycobacteria are killed by ADEP through inhibition of ClpP function. Although ADEPs can stimulate purified M. tuberculosis ClpP1P2 to degrade larger peptides and unstructured proteins, this effect is weaker than for ClpP from other bacteria and depends on the presence of an additional activating factor (e.g. the dipeptide benzyloxycarbonyl-leucyl-leucine in vitro) to form the active ClpP1P2 tetradecamer. The cell division protein FtsZ, which is a particularly sensitive target for ADEP-activated ClpP in firmicutes, is not degraded in mycobacteria. Depletion of the ClpP1P2 level in a conditional Mycobacterium bovis BCG mutant enhanced killing by ADEP unlike in other bacteria. In summary, ADEPs kill mycobacteria by preventing interaction of ClpP1P2 with the regulatory ATPases, ClpX or ClpC1, thus inhibiting essential ATP-dependent protein degradation.

Structure and Functional Properties of the Active Form of the Proteolytic Complex, ClpP1P2, from Mycobacterium tuberculosis

The ClpP protease complex and its regulatory ATPases, ClpC1 and ClpX, inMycobacterium tuberculosis(Mtb) are essential and, therefore, promising drug targets. TheMtbClpP protease consists of two heptameric rings, one composed of ClpP1 and the other of ClpP2 subunits. Formation of the enzymatically active ClpP1P2 complex requires binding of N-blocked dipeptide activators. We have found a new potent activator, benzoyl-leucine-leucine (Bz-LL), that binds with higher affinity and promotes 3-4-fold higher peptidase activity than previous activators. Bz-LL-activated ClpP1P2 specifically stimulates the ATPase activity ofMtbClpC1 and ClpX. The ClpC1P1P2 and ClpXP1P2 complexes exhibit 2-3-fold enhanced ATPase activity, peptide cleavage, and ATP-dependent protein degradation. The crystal structure of ClpP1P2 with bound Bz-LL was determined at a resolution of 3.07 Å and with benzyloxycarbonyl-Leu-Leu (Z-LL) bound at 2.9 Å. Bz-LL was present in all 14 active sites, whereas Z-LL density was not resolved. Surprisingly, Bz-LL adopts opposite orientations in ClpP1 and ClpP2. In ClpP1, Bz-LL binds with the C-terminal leucine side chain in the S1 pocket. One C-terminal oxygen is close to the catalytic serine, whereas the other contacts backbone amides in the oxyanion hole. In ClpP2, Bz-LL binds with the benzoyl group in the S1 pocket, and the peptide hydrogen bonded between parallel β-strands. The ClpP2 axial loops are extended, forming an open axial channel as has been observed with bound ADEP antibiotics. Thus occupancy of the active sites of ClpP allosterically alters sites on the surfaces thereby affecting the association of ClpP1 and ClpP2 rings, interactions with regulatory ATPases, and entry of protein substrates.

Cleavage Specificity of Mycobacterium tuberculosis ClpP1P2 Protease and Identification of Novel Peptide Substrates and Boronate Inhibitors with Anti-bacterial Activity

The ClpP1P2 protease complex is essential for viability in Mycobacteria tuberculosis and is an attractive drug target. Using a fluorogenic tripeptide library (Ac-X3X2X1-aminomethylcoumarin) and by determining specificity constants (kcat/Km), we show that ClpP1P2 prefers Met ≫ Leu > Phe > Ala in the X1 position, basic residues or Trp in the X2 position, and Pro ≫ Ala > Trp in the X3 position. We identified peptide substrates that are hydrolyzed up to 1000 times faster than the standard ClpP substrate. These positional preferences were consistent with cleavage sites in the protein GFPssrA by ClpXP1P2. Studies of ClpP1P2 with inactive ClpP1 or ClpP2 indicated that ClpP1 was responsible for nearly all the peptidase activity, whereas both ClpP1 and ClpP2 contributed to protein degradation. Substrate-based peptide boronates were synthesized that inhibit ClpP1P2 peptidase activity in the submicromolar range. Some of them inhibited the growth of Mtb cells in the low micromolar range indicating that cleavage specificity of Mtb ClpP1P2 can be used to design novel anti-bacterial agents.

Lassomycin, a ribosomally synthesized cyclic peptide, kills mycobacterium tuberculosis by targeting the ATP-dependent protease ClpC1P1P2

Languishing antibiotic discovery and flourishing antibiotic resistance have prompted the development of alternative untapped sources for antibiotic discovery, including previously uncultured bacteria. Here, we screen extracts from uncultured species against Mycobacterium tuberculosis and identify lassomycin, an antibiotic that exhibits potent bactericidal activity against both growing and dormant mycobacteria, including drug-resistant forms of M. tuberculosis, but little activity against other bacteria or mammalian cells. Lassomycin is a highly basic, ribosomally encoded cyclic peptide with an unusual structural fold that only partially resembles that of other lasso peptides. We show that lassomycin binds to a highly acidic region of the ClpC1 ATPase complex and markedly stimulates its ATPase activity without stimulating ClpP1P2-catalyzed protein breakdown, which is essential for viability of mycobacteria. This mechanism, uncoupling ATPase from proteolytic activity, accounts for the bactericidal activity of lassomycin.

The active ClpP protease from M. tuberculosis is a complex composed of a heptameric ClpP1 and a ClpP2 ring

Mycobacterium tuberculosis (Mtb) contains two clpP genes, both of which are essential for viability. We expressed and purified Mtb ClpP1 and ClpP2 separately. Although each formed a tetradecameric structure and was processed, they lacked proteolytic activity. We could, however, reconstitute an active, mixed ClpP1P2 complex after identifying N-blocked dipeptides that stimulate dramatically (>1000-fold) ClpP1P2 activity against certain peptides and proteins. These activators function cooperatively to induce the dissociation of ClpP1 and ClpP2 tetradecamers into heptameric rings, which then re-associate to form the active ClpP1P2 2-ring mixed complex. No analogous small molecule-induced enzyme activation mechanism involving dissociation and re-association of multimeric rings has been described. ClpP1P2 possesses chymotrypsin and caspase-like activities, and ClpP1 and ClpP2 differ in cleavage preferences. The regulatory ATPase ClpC1 was purified and shown to increase hydrolysis of proteins by ClpP1P2, but not peptides. ClpC1 did not activate ClpP1 or ClpP2 homotetradecamers and stimulated ClpP1P2 only when both ATP and a dipeptide activator were present. ClpP1P2 activity, its unusual activation mechanism and ClpC1 ATPase represent attractive drug targets to combat tuberculosis.

Structure-based design of estrogen receptor-beta selective ligands

We present the structure-based optimization of a series of estrogen receptor-beta (ERbeta) selective ligands. X-ray cocrystal structures of these ligands complexed to both ERalpha and ERbeta are described. We also discuss how molecular modeling was used to take advantage of subtle differences between the two binding cavities in order to optimize selectivity for ERbeta over ERalpha. Quantum chemical calculations are utilized to gain insight into the mechanism of selectivity enhancement. Despite only two relatively conservative residue substitutions in the ligand binding pocket, the most selective compounds have greater than 100-fold selectivity for ERbeta relative to ERalpha when measured using a competitive radioligand binding assay.

Two distinct proteolytic processes in the generation of a major histocompatibility complex class I-presented peptide

Although cellular proteins degraded by proteasomes are the source of most antigenic peptides presented on major histocompatibility complex class I molecules, it is unknown whether the eight- to nine-residue peptides that fit in the binding groove of class I molecules are directly produced by proteasomes alone in vivo. If the eight-residue peptide SIINFEKL from chicken ovalbumin is extended by one or several residues at its C terminus and microinjected into cells or expressed from a minigene, it is processed and presented on major histocompatibility complex class I. However, processing and presentation are inhibited by proteasome inhibitors, such as lactacystin. In contrast, when SIINFEKL is extended by 2 to 25 residues at its N terminus, its presentation is not blocked by proteasome inhibitors. N-terminal processing also can occur when the extended peptide is cotranslationally inserted into the endoplasmic reticulum. Thus, two different proteolytic steps in the generation of an chicken ovalbumin-presented peptide can be distinguished. Cleavage by the proteasome defines the proper C terminus, whereas distinct peptidase(s) in the cytosol or endoplasmic reticulum may generate the appropriate N terminus from extended peptides.

Lactacystin and clasto-lactacystin beta-lactone modify multiple proteasome beta-subunits and inhibit intracellular protein degradation and major histocompatibility complex class I antigen presentation

The antibiotic lactacystin was reported to covalently modify beta-subunit X of the mammalian 20 S proteasome and inhibit several of its peptidase activities. However, we demonstrate that [3H]lactacystin treatment modifies all the proteasome's catalytic beta-subunits. Lactacystin and its more potent derivative beta-lactone irreversibly inhibit protein breakdown and the chymotryptic, tryptic, and peptidylglutamyl activities of purified 20 S and 26 S particles, although at different rates. Exposure to these agents for 1 to 2 h reduced the degradation of short- and long-lived proteins in four different mammalian cell lines. Unlike peptide aldehyde inhibitors, lactacystin and the beta-lactone do not inhibit lysosomal degradation of an endocytosed protein. These agents block class I antigen presentation of a model protein, ovalbumin (synthesized endogenously or loaded exogenously), but do not affect presentation of the peptide epitope SIINFEKL, which does not require proteolysis for presentation. Generation of most peptides required for formation of stable class I heterodimers is also inhibited. Because these agents inhibited protein breakdown and antigen presentation similarly in interferon-gamma-treated cells (where proteasomes contain LMP2 and LMP7 subunits in place of X and Y), all beta-subunits must be affected similarly. These findings confirm our prior conclusions that proteasomes catalyze the bulk of protein breakdown in mammalian cells and generate the majority of class I-bound epitopes for immune recognition.