Fragments in cells, writ large

Earlier this year we highlighted work in which a dozen fragments were screened against cells to look for noncovalent binders across the proteome. A new paper in Science by Georg Winter and collaborators at the Austrian Academy of Sciences, Pfizer, and several other organizations ups the game by more than an order of magnitude, and uses machine learning to make predictions about fragments’ cellular destinations and binding partners. (See also Derek Lowe’s post here.)

 

The researchers started with 407 diverse fully functionalized fragments (FFFs), which as we previously discussed consist of a variable fragment coupled to a photoreactive group and an alkyne moiety that can be used to pull down any bound proteins using click chemistry. These were selected from a larger set of ~6000 FFFs available from Enamine. The FFFs were incubated at 50 µM with intact HEK293T cells, followed by ultraviolet crosslinking.

 

Next, cells were lysed and treated with a biotin-azide probe that reacts with the alkyne on the FFFs. Covalently modified proteins were captured on streptavidin resin and proteolytically digested. Tandem mass tag (TMT) proteomics, which we wrote about here, was used to identify captured proteins. Unlike earlier methods, the researchers did not pinpoint the specific fragment binding sites on proteins.

 

In total the researchers found 2667 proteins bound to one or more fragments, of which ~86% had no reported ligands. Both proteins and ligands varied considerably in promiscuity: some proteins bound to more than half of the FFFs, and some fragments bound to hundreds of proteins, while others bound only a few, or none. To look for specific interactions, the researchers focused on proteins bound by fewer than 10 different ligands.

 

Three protein-ligand interactions were analyzed in some detail: the kinase CDK2 (and other CDK family members), the adapter protein DDB1, and the solute carrier protein SLC29A1. In each case the researchers confirmed the results from their chemoproteomic screens. Follow-up studies with related molecules led to more potent derivatives, with a CDK2 inhibitor showing low micromolar activity in a biochemical assay and an SLC29A1 inhibitor showing micromolar activity in a cell-based assay.

 

The researchers also found patterns in their larger data set. Armed with 47,658 protein-ligand interactions, the researchers were able to use machine learning to start to predict which molecular features were associated with binding. They ranked fragments as promiscuous or nonpromiscuous and built a promiscuity model. Molecules with higher lipophilicity and a greater fraction of aromatic carbon atoms tended to be more promiscuous, but the model could correctly categorize compounds as promiscuous even if they had lower ClogP values, or nonpromiscuous even if they had higher ClogP values.

 

Beyond promiscuity, the researchers used machine learning to predict other behavior, such as subcellular localization. A relatively easy case was to predict which molecules would accumulate in lysosomes; these tended to be hydrophobic basic amines. More impressively, the researchers could predict fragments likely to bind to transmembrane transporters, RNA binding proteins, and even intrinsically disordered proteins. And this is just the start: they hope one day to predict “target proteins from an input chemical structure alone.”

 

Perhaps most exciting, all of the data and models are available for free at Ligand Discovery. You can explore the proteins bound across all 407 fragments, input one or more proteins and find ligands, predict whether any given FFF is likely to be promiscuous or not, and even “build a machine learning model on the fly to predict potential interactions.” 

 

Check it out and let us know your experience.

In situ click chemistry on RNA

In templated or in situ reactions, bonds form between two fragments that are brought together in the context of a larger molecule such as a protein. We have written previously about dynamic combinatorial chemistry (DCC), which depends on reversible bond formation where the larger molecule shifts the equilibrium toward the linked fragments. For irreversible bond formation, the larger molecule effectively catalyzes the formation of an inhibitor (or at least a binder). In a recent Angew. Chem. Int. Ed. paper Jyotirmayee Dash and colleagues at the Indian Association for the Cultivation of Science describe an application of the latter that uses RNA as the template.

The target of interest was TAR RNA, a short region of viral RNA essential for HIV replication. We have previously highlighted a few examples of fragment screening against RNA (here, here, and here), including TAR, but most of the hits were weak.

The researchers used azide-alkyne cycloaddition, the quintessential click chemistry reaction. They built a small library of four alkynes (only one of which was fragment-sized) and 11 azides. All of these were incubated together (4 µM of each alkyne and 1 µM of each azide) in the presence of 5 µM biotin-labeled TAR RNA for 72 hours. (The reaction is typically slow at room temperatures unless catalyzed by metal ions.) Magnetic streptavidin-coated beads were then used to capture the RNA and any bound ligands, which were identified by HPLC-MS. Control experiments were run with TAR DNA or a mutant form of TAR RNA lacking an essential bulge. The result was one fairly potent compound (below) that was specific for TAR RNA, as well as a couple other molecules that were both weaker and less specific.

The affinity of compound 3ba for TAR RNA was measured by isothermal titration calorimetry (ITC) as well as by a fluorescence assay, which were in good agreement. Importantly, the ITC data suggested 1:1 binding, which is particularly important given that the ligand contains two 2-aminothiazoles, a moiety that has been called a PrAT for its promiscuous behavior. Finally, the ligand could displace the Tat peptide at low micromolar concentrations, suggesting that it is binding at the biologically relevant site of TAR.

I do have a few quibbles. It would have been interesting if the researchers had reported the affinity of the azide and alkyne themselves to see how much of a boost they got by linking. And since the most potent molecule is not always selected from target-guided synthesis, it would have been interesting to make and test other possible cycloaddition products to see if they missed anything useful.

Still, it is nice to see a submicromolar RNA binder come out of an in situ screen. Targeting RNA with small molecules has recently become trendy, and it will be fun to see how far approaches like these can go.

Fragment-linking on proteins: amide formation

Among fragment-linking approaches, protein-templated reactions have a special appeal. When two fragments that bind next to one another on a protein react, the protein essentially creates its own ligand. The reaction can be reversible, as in dynamic combinatorial chemistry, though the choice of chemistries tends not to be very drug-like, necessitating replacement of the linker for lead development to proceed. Irreversible chemistries, such as cycloadditions (“click chemistry”), give more stable linkers, but these still tend to be unusual. In a recent open-access paper in Angew Chem., Jörg Rademann and collaborators at the Freie Universität Berlin and Philipps-Univesität Marburg have extended the approach to amides – one of the most vanilla chemistries out there.

The researchers used factor Xa as a model protein. Using information in the literature they designed compound 15, found that it was a low nanomolar inhibitor, and obtained a crystal structure of the molecule in complex with the enzyme. Cleaving one of the amide bonds provides two fragments, compounds 16 and 17, each with very weak activity. Put another way, linking these two fragments leads to superadditivity of binding.

Chemists often make amides by “activating” a carboxylic acid and reacting it with an amine. We usually want this to happen quickly, but here the researchers wanted the reaction to only occur when the two fragments bind next to one another on the surface of the protein.

The team made a dozen activated esters of compound 1 and incubated 5 mM of each of these with 0.285 mM of amine 14, in the presence or absence of 14.5 nM factor Xa. (Each combination of reactants would produce compound 15.) After two hours protein was added to the no-protein controls and fluorogenic substrate was added to all the samples, which were tested for enzyme activity. Not surprisingly, inhibition was complete for particularly active esters (such as 4-nitrophenyl ester), whether or not protein had been present during the incubation. Similarly, the presence of protein also didn’t matter for particularly inactive esters (such as methyl ester): inhibition was minimal because very little product 15 formed. But for two Goldilocks esters (phenyl ester and 2,2,2-trifluoroethyl ester), inhibition was greater when protein was present during incubation.

The researchers used mass spectrometry to quantify the amount of compound 15 formed. For the 2,2,2-trifluoromethyl ester, after 1.7 hours the concentration of compound 15 was 10 nM in the presence of factor Xa, but only 0.15 nM in its absence. (The phenyl ester gave less selective results.)

Of course, as these numbers suggest, product formation could be limited by the concentration of the enzyme, possibly complicating detection. More significantly, even relatively unreactive esters could react with the dozens of nucleophiles on proteins, potentially inactivating or denaturing them. Still, this is an intriguing approach, and it will be fun to see whether it works in more challenging systems.

The calm before the click in chitinase

In situ click chemistry is a topic we’ve covered before on Practical Fragments. Essentially, two ligands bind near one another on a target protein and react to form a linked molecule. There are several published examples, but it is not clear why it sometimes works and sometimes doesn’t. A new paper in Proc. Acad. Nat. Sci. USA by a team of Japanese and US researchers led by Satoshi Ōmura and Toshiaki Sunazuka at the Kitasato Institute in Tokyo addresses this question.

The researchers had previously discovered potent inhibitors of an antibacterial target enzyme called Serratia marcescens chitinase B, or SmChiB, using in situ click chemistry. In the presence of SmChiB, azide 2 reacts with alkyne 3 to yield triazole 4, which binds 26-fold more tightly than azide 2:

 

In the new paper, the goal was to use crystallography and computational chemistry to investigate how the reaction proceeds. To avoid azide 2 reacting with alkyne 3 in the crystal and so better visualize starting points, the researchers prepared the closely related alkene 5 mimic of alkyne 3. Unfortunately, due to its (unmeasurably poor) affinity, alkene 5 did not yield a co-crystal structure on its own.

The researchers were able to obtain a co-crystal structure of SmChiB bound to triazole 4 (green carbons below). Surprisingly, a co-crystal structure of azide 2 showed the molecule bound in a quite different orientation. However, a co-crystal structure of the ternary complex of azide 2 (cyan below) and alkene 5 (magenta) bound simultaneously to SmChiB revealed a close overlay of azide 2 with the corresponding fragment in triazole 4. Alkene 5 in the ternary complex adopted two conformations (the electron density is memorably described as resembling “a two-horned goat head”). As shown in the figure below, one of these orientations places the alkene moiety in close proximity to the azide moiety, primed for clicking.

Next, the researchers used this ternary structure to run high-level density functional theory calculations to determine the energetics of the click reaction and compared these with the same reaction run in water. The values were quite similar (if anything, the protein had a slightly higher activation barrier), suggesting that the protein was not directly catalyzing the reaction with specific amino acid side chains. Rather, the reaction was being accelerated simply by the preorganization of the azide and alkyne.

On the one hand, these results aren’t really a surprise: I think most people assumed that in situ chemistry works by bringing the reactants together rather than anything more exotic (with the odd exception). On the other hand, it is nice to see experiment match theory.

More generally, the results help to explain why in situ click chemistry is so challenging. The crystal structure of azide 2 and alkene 5 shows the relevant moieties quite close to each other, yet the reaction is still somewhat inefficient. Finding two fragments that not only bind near one another but are also oriented properly is likely to be a rare event.

Click first, ask questions later

Three years ago Beat Ernst and his colleagues at the University of Basel described using NMR to identify two molecules that bind next to one another on the protein MAG. They then used in situ “click chemistry” to link these together to obtain a more potent binder. In a recent issue of J. Am. Chem. Soc. they have taken a similar approach to the protein E-selectin, but without the in situ part.

The selectins are cell adhesion proteins involved in a variety of biological processes, notably inflammation and tumor metastases. They bind to carbohydrates on the surface of leukocytes, but the affinities of any one selectin for a given carbohydrate tends to be low – often only millimolar. In the current paper, the researchers started with a reasonably potent modified carbohydrate, compound 3.

The researchers performed an NMR screen of 80 fragments in which they looked for increased relaxation of protons in the fragments upon binding to protein. This led to five hits. To determine whether these bound near compound 3, the researchers modified compound 3 with a “spin-label,” a moiety that would increase the rate of relaxation of nearby molecules and so make them detectable (see the previous post for more details). Two of the fragments appeared to bind near compound 3, and the researchers chose to pursue compound 4 – the very same fragment they had pursued previously for MAG.

At this point the researchers replaced the spin label with an alkyne (attached via spacers of various lengths) and added azide groups (again, with various spacers) to compound 4 and attempted to perform in situ click chemistry in the presence of the protein, as they had done previously. Nothing happened. Having come this far, they used more conventional conditions (ie, without the protein present) to make a small library of 20 triazoles and tested these for binding, leading to 5 hits with nanomolar activity, such as compound 43.

Given the high affinity of compound 43 for E-selectin, why didn’t it form in situ? The researchers suggest that:

Given its flat binding site, E-selectin does not act as an effective supramolecular catalyst for the alkyne-azide cycloaddition, because even upon simultaneous binding of first- and second-site ligands their azide- and acetylene-substituted linkers are not sufficiently preorganized to accelerate the cycloaddition reaction.

This is an example of a false negative from in situ fragment assembly. We previously wrote about another case in which in situ click chemistry yielded the less potent of two regioisomers due to trace amounts of contaminating copper. There is something conceptually beautiful about having a protein template the formation of its own inhibitor, but how often does it really work?

Ending on a positive note, the NMR approach described here is an example of linking without the need for protein structure. You just might want to click before you assay.

Way back when I was but a young lad, I worked on dual inhibitors of 5-LO and PAF in the lab of T.Y. Shen at UVa.  They were crazy compounds, but as an undergraduate it was a great experience.  We even did our own assays requiring fresh whole blood.  I loved it and learned to HATE HPLC assays at the same time.  People still seem to be doing stuff like that.  Overall, the field of leukotriene antagonists has been pretty successful, but it is littered with debris. 

Prostaglandin E2 (PGE2) is a key mediator in inflammation, pain, fever, atherosclerosis, and tumorigenesis: a veritable collection of the key therapeutics areas for most pharmaceutical companies.  Arachidonic acid (AA) is acted upon by COX1/2, target for a whole pile of drugs, to generate PGH2.  This is then isomerized to PGE2 by microsomal Prostaglandin E2 synthase (mPGES).  Inhibition of mPGES would be expected to only preclude PGE2 and thus eliminate many of the adverse side effects of COX inhibitors.  As would be expected, mPGES inhibition is not new and several inhibitors are in the clinic.  So, we happen upon this paper: a recent academic effort at a potentially rich field.   The authors begin the explanation of their approach with this: 

To date, there is no real three-dimensional X-ray crystal structure of mPGES-1 in the apo form or with an inhibitor bound with exception of electron crystallographic structure complexed with glutathione in its closed state (PDB code:3DWW).

Their strategy is to replace glutathione with non-peptidomimetics via fragments based upon this pharmacophore model (3DWW): two negative ionizable, one HBD, and one HBA

 They hypothesized their molecule thusly:

The choice of the triazole is to leverage click chemistry to generate it, while the sulfonamide has been shown to a "privileged" fragment in other drugs. Prior to synthesis, the decided to calculate the binding energies (with three decimal place precision!).

The authors then tested the compounds in a biochemical assay.  The first thing that they noticed is that n=2 for the linker is better than n =1 or 3 (50% increase in potency).  Compound 6f (R1=R2=phenyl) was the most potent (1.1 uM).  However at 29 HAC, its LEAN is 0.21 and would be considered lousy.  It can be argued that lipases and such are going to have relatively inefficient compounds due to the highly hydrophobic nature of the active site. [I completely ignored their discussion of the calculations and modeling, because honestly, does it really matter?] Compound 6f showed ~1000x mPGES-1 selectivity over COX1 and no COX2 activity.  They then (much to Dan's delight I am sure) tested for activity with and without detergent.  Not surprisingly, in the presence of detergent, compound 6f lost significant amount (75), but not all of its activity.  The authors very honestly say:

Therefore, compound 6f may be judged as a partial nuisance inhibitor
of mPGES-1 instead of true mPGES-1 inhibitor.

 I would have like them to have tested ALL of their compounds in the presence of detergent to begin with.  This smacks of something that was added in after initial review and honestly really makes the paper uninteresting.  I would really like to know if 6f would still have the outstanding potency from the compounds made, or if some other compound would have aggregated less and thus been of interest.   I would also argue that calling this paper "fragment-based" is a stretch.  As Dan just pointed out, some target classes may just need larger compounds to hit them. 

Protein-templated click chemistry – just add copper

We’ve written previously about protein-templated chemistry (here and here), in which a protein catalyzes the formation of a more potent inhibitor from two lower affinity fragments. Of course, proteins aren’t the only things that can catalyze reactions: copper is well-known to promote the cycloaddition between azides and alkynes. Like peanut butter and chocolate, it turns out that copper in the context of a protein can be even better than either alone, as reported in a recent issue of Angew. Chem. by an international team of researchers from Japan and the US.

The researchers were interested in using in situ click chemistry to discover inhibitors of histone deacetylases (HDACs), and they decided to see if they could use an activity assay to detect the formation of inhibitors formed in situ. They incubated two different hydroxamic-containing alkynes (known HDAC inhibibitors) with 15 different azides in the presence of HDAC8 and looked for enhanced inhibition of the enzyme. Of these 30 combinations, they found a single hit: the reaction of compound 1b with compound 2o (see figure).


However, there were several oddities. First, the linked compound (anti-3) is no more potent than the initial hydoxamic-containing fragment. Second, only the anti isomer was formed, despite the fact that the syn isomer is almost 10-fold more potent. Finally, the yields of anti-3 were much higher than typically observed in these sorts of experiments. This made the researchers suspicious, and after a series of experiments they determined that trace amounts of copper, most likely introduced in the synthesis of 1b, had incorporated into the active site of HDAC8 and were serving to accelerate the reaction. A small amount of copper in the absence of protein was unable to catalyze the reaction, nor was the protein alone when copper was carefully removed.

There are a number of interesting implications from this paper, but one in particular is rather sobering: in situ assembly screening does not necessarily yield the most potent inhibitor. I suspect this is a general feature of kinetically-guided methods of inhibitor discovery, but what do you think?

Click here to link

Fragment linking is a topic we’ve discussed several times. One of the more interesting approaches is template-directed synthesis, in which a protein causes two fragments in close proximity to react with one another (see here for example). In a recent issue of Angew. Chem. Int. Ed., Beat Ernst and colleagues at the University of Basel provide a new variant of this theme, without requiring structural information about the protein.

The researchers were interested in a protein called myelin-associated glycoprotein, or MAG, which blocks axonal regrowth. They started with an NMR screen to determine which members of a fragment library bind to MAG as assessed by a phenomenon known as transverse magnetization decay; essentially, small molecules that bind to a protein behave like large molecules in showing a rapid decay in magnetization, so an increased magnetization decay of fragments in the presence of protein suggests binding. A number of fragments were identified as binders, but the site of binding was not determined.

To find molecules that could be linked, the researchers took a known ligand, the sialic acid derivative 1 (see figure – albeit larger than a fragment), and modified this to contain a spin-label. Spin-labels are small moieties that contain an unpaired electron and, just like large molecules, cause an increase in magnetization decay, but only to molecules within close proximity. The two effects are additive, and thus the researchers could determine which fragments bind to the protein in close proximity to the spin-label-containing derivative of compound 1. In fact, the distance dependence is so pronounced that different protons on the fragment can show different effects, thus indicating which portion of the fragment is close to the spin label (see here for a similar approach using ILOE). In this case, the researchers found that a nitroindole fragment (see figure) had its 5-membered ring positioned closer to the spin label than its 6-membered ring.



Knowing the relative positions of the two ligands, the researchers modified them so they could be linked together. They added functional groups with different linker lengths to create several analogs, replacing the spin label with an alkyne and adding an azide to the nitroindole fragment. They then incubated all the analogs together in the presence of the protein. Analysis of the reaction by HPLC-MS after three days at 37 degrees revealed one prominent product, with a mass consistent with compound 7. Two isomers of this product can be formed, with syn and anti configurations around the triazole, and the researchers synthesized both of them. Interestingly, the anti isomer (shown) had a Kd for MAG of 190 nM, while the syn isomer bound roughly 10-fold more weakly.

Although the ligand efficiency of the final compound is low, sugar-based molecules typically have low ligand efficiencies, and maintaining the same efficiency as starting compound 1 is impressive. However, the ligand efficiency of the final molecule is probably lower than the second-site ligand: the researchers don’t report its affinity, but since it would likely need to be 10 mM or better to be detected its ligand efficiency is probably at least 0.23 kcal/mol/atom.

Still, the final product is sufficiently potent that it could make a useful biological probe. Moreover, the approach is notable in not requiring structure of the protein – a rare and attractive feature for fragment linking.

Click and Clack

A new paper describes how combining fragment classes that selectively react with each other can yield an inhibitor to a protein-protein interaction.

Most of you are probably familiar with the concept of “click chemistry”, exemplified by the Huisgen cycloaddition of azides and acetylenes to form triazoles, and defined earlier this century by K. Barry Sharpless and colleagues. The idea, put very simply, is to react two molecules selectively and reliably to generate a product in high yield. The approach has obvious applications to fragment-based ligand discovery, and in fact Sharpless and co-workers demonstrated that the enzyme acetylcholinesterase can catalyze the formation of an inhibitor with femtomolar potency, starting from two fairly large and potent “fragments,” one containing an azide, the other an alkyne.

The notion of using a protein as a template on which to assemble an inhibitor dates back further than click chemistry. Huc and Lehn used reductive amination between aldehydes and amines in the presence of carbonic anhydrase to capture molecules that interacted more strongly with the protein than did the starting materials, and the technology even formed the basis of a company, Therascope.

In the latest example of this line of research, Roman Manetsch and colleagues find that the antiapoptotic cancer target Bcl-xL is able to catalyze the formation of an inhibitor from fragments sporting thioacids and sulfonyl azides; the inhibitor itself was previously discovered using different fragment-based methods at Abbott. The reaction appears to be relatively fast, and works even in the presence of a small pool of sulfonyl azides.


The paper, published this month in the Journal of the American Chemical Society, is a nice proof-of-concept study, but it does raise practical questions. First, the authors report the use of only six sulfonyl azides and three thioacids. These generally need to be custom-made rather than purchased; how much of a barrier will this represent to other practitioners? Once assembled, how stable will these fragments be in long-term storage? Finally, the product acylsulfonamide is a rather special entity not commonly found in drugs, and with a limited range of bioisosteres (indeed, it was originally employed at Abbott to replace a carboxylic acid). Still, the paper does provide an interesting - if less conventional - method of fragment-based lead discovery.