Fragmenting natural products – sometimes PAINfully

Many drugs have their origins in natural products. But as any synthetic organic chemist will tell you, natural products often have complex architectures that can take years of effort and dozens of chemists to make in the lab. Thus, many of the compounds made in industry look quite different from natural products, particularly in the past few decades. High failure rates in drug discovery have led folks to return to natural products or similar compounds, such as those from diversity oriented synthesis (DOS). In a recent issue of Nature Chemistry, Herbert Waldmann and colleagues at the Max-Planck Institute in Dortmund examine whether natural products can serve as starting points for new fragments.

The researchers started by computationally deconstructing 183,769 natural products into 751,577 component fragments. After various filters (size, lipophilicity, reactivity, etc.) they arrived at 110,485 fragments sorted by similarity into 2000 clusters. The resulting fragments differ in their overall calculated properties from commercial fragments. This is all highly reminiscent of the Emerald (nee deCODE) “fragments of life”, though surprisingly that work is not referenced.

One challenge of designing new fragments is that you may not be able to buy them. In this case, nearly half of the clusters did have a compound that could be purchased – though perhaps this somewhat defeats the purpose of trying to explore novel chemical space. At any rate, 193 fragments were either bought or synthesized. These were tested in functional assays against p38a MAP kinase and several protein phosphatases. A number of hits were identified, and in the case of p38a, nine kinase-fragment co-crystal structures were solved. Some of these were similar to previously reported fragments, but others were more unusual. Together with the crystal structures, these fragments provide new ideas for a well-studied target.

Looking at the structures of some of the phosphatase inhibitors, however, I started to worry. One strong point of the paper is that it is very complete: the chemical structures of all 193 tested fragments are provided in the supplementary information. Unfortunately, the list contains some truly dreadful members; 17 of the worst are shown here, with the nasty bits shown in red. All of these are PAINS that will nonspecifically interfere with many different assays.

Compounds 15, 44, 49, 159, 166, 173, 174, and 175 are catechols; compounds 89 and 151 (yes, they are the same molecule – guess they really liked this one), 165, 166, 167, and 168 are quinones; compounds 55, 89/151, and 166 are hydroquinones; compound 20 is a Michael acceptor; compound 76 is an epoxide; and compound 184 is a redox cycler. In other words, these fragments are a depressing example of life imitating art (or at least satire).

To be blunt: none of these molecules should appear in a screening library today.

I don’t want to pick on these researchers; it is after all laudable that they fully disclosed the structures of their molecules.

However, I am concerned that other people may build libraries containing some of these fragments, or worse, that opportunistic vendors will start selling “natural-product derived fragments.” Indeed, most of these molecules are commercially available. It is disappointing that so many nuisance compounds would find their way into research published in a Nature family journal, and I think it is important to call it out. Only by publicizing the problems that can arise will people be made aware of the dangers.

FAK This

FAK, also known as PTK2, is a well known oncology target.  Current known inhibitors can be broken into three different binding classes:   Cpds I-IV are hinge binders, the chloropyramine targets the FAK-VEGF interface, and Y15 targets the Y397 site. Cpds V and VI were recently reported as novel allosteric inhibitors of FAK.

In this paper, a group led by researchers at Merck Serono, report their discovery of a new core from an "accelerated knowledge-based fragment growing approach".  

They used a commercially available fragment library (defined as:  MW, <200 solubility="">1 μg/mL; number of hydrogen bond donors and acceptors, ≤3) was screened against the immobilized kinase domain of FAK by SPR, which allowed them to determine kinetics for most of the fragments.  Compound I (Magenta) was found to be a 43 μm inhibitor.  The X-ray structure showed it to make excellent contacts with the protein.  Addition of the spinach shown in green, afforded an order of magnitude increase in potency.  In order to facilitate better elaboration, they chose to use 7-azaindole as the scaffold.

Then, going through traditional SAR and medchem, they end up with this table.  The best compound is a single digit nanomolar inhibitor with cell-based activity. One important aspect of this work is that the lead series can induce a rare helical loop DFG conformation.  In their conclusion, they state 

it was easier to improve kinase inhibition than kinase selectivity.

The first thing that stands out here is the solubility limit.  For a 250 Da fragment, 1μg/mL corresponds to 4 μM.  To me that sounds incredibly low; is it a typo?  They don't mention who those fragments are from.  I would love to know out of sheer curiousity.   Secondly, although they talk about their accelerated fragment growing approach, they don't actually explain what they mean by that?  To the best of my reading, I don't think they have introduced anything novel here.  

Fragment linking for LDHA: Ariad’s turn

Last year we highlighted a paper from AstraZeneca in which researchers there used a fragment-linking approach to tackle an enzyme important for cancer metabolism, lactate dehydrogenase A (LDHA). Turns out they weren’t alone – researchers at Ariad had also been working on the same target, as Stephan Zech reported at FBLD 2012. They have now published some of this work in J. Med. Chem.

Anna Kohlmann and colleagues at Ariad started with a fairly small library, just 735 fragments from Maybridge. These were screened using STD-NMR at 2-3 mM per fragment, resulting in 38 hits, about half of which contained carboxylic acids – not surprising given that the substrate and cofactor are both negatively charged. Most of the fragments could be competed by the cofactor NADH, and although they bound too weakly to show any inhibition in an enzymatic assay, they did show binding by SPR. Crystal soaking led to a co-crystal structure of compound 1, which binds in the substrate and part of the cofactor site (where the nictotinamide moiety of NADH normally binds).

Fragment growing led to compounds 2 and 5, both with enhanced affinity. Interestingly, crystallography revealed that compound 5 binds in a distant part of the cofactor binding site, where the adenosine moiety of NADH normally binds. Elaboration of this molecule didn’t do much for affinity but did suggest a linking strategy, resulting in molecules such as compound 9, with nanomolar potency and detectable cell-based activity.

Apropos to Darwin Day, this is an interesting example of convergent evolution: two companies applying fragment-linking to discover molecules that bear some similarity to one another (Ariad compound 8 in blue, AstraZeneca compound 26 in red).

Near the end of the paper, the researchers also carefully investigated some of the other previously reported “inhibitors” of LDHA and found that they are in fact aggregators. This is not surprising given their structures, which look like something that might appear in an April Fool’s post. Unfortunately these molecules were reported in prominent journals such as Chem. Biol. and Proc. Nat. Acad. Sci. USA; the later, published in 2010, has already been cited at least 100 times. Publicly revealing them to be artifacts is a beautiful example of the self-correcting nature of science. I hope we’ll see more of it.

Beware correlation inflation

Drug discovery today is replete with rules and metrics: the Rule of 5, the Rule of 3, (though perhaps not 1), not to mention ligand efficiency and friends. The hope is that these encapsulate physical trends that will guide drug hunters towards better compounds. However, there is a danger that rules will become strait-jackets; plenty of drugs, after all, lie well outside the Rule of 5 (Ro5). In a paper recently published in J. Comput. Aided Mol. Des., Peter Kenny (of FBDD-Lit fame) and Carlos Montanari argue that the correlations underlying many rules may not be as robust as they appear. The article is full of the trenchant prose we’ve come to expect of Kenny, so I’ll quote liberally.

The background:

Those who have followed the drug discovery literature over the last decade or so will have become aware of a publication genre that can be described as ‘retrospective data analysis of large proprietary data sets’ or, more succinctly, as ‘Ro5 envy’.

The problem:

Although data analysts frequently tout the statistical significance of the trends that their analysis has revealed, weak trends can be statistically significant without being remotely interesting.

This is especially likely to occur when data are “binned” into a smaller number of categories before being analyzed, thereby hiding variation and making correlations appear stronger than they really are. Since many published analyses use proprietary, unavailable data, Kenny and Montanari constructed model “noisy” data sets and looked for correlations in the primary data and the binned data. They found that correlations in the binned data were inflated. Perhaps counter-intuitively, the effect actually gets more pronounced the larger the data set.

Having described the problem, Kenny and Montanari go on to question some recent high-profile papers correlating, for example, lipophilicity with pharmacological promiscuity, or the percentage of sp3-hybridized carbons (Fsp3) with solubility (see also here). In the latter case, all the data were publicly available, and a reanalysis with the primary data as opposed to binned data caused the correlation coefficient (r) to drop from 0.972 to 0.247!

Graphical representation of data comes under heavy scrutiny too. In particular, the common practice of subdividing data points into small numbers of categories (often red, yellow, and green) can make these categories appear discrete when the underlying data are better described as a continuum.

The overall message is that weak correlations may lead to misguided strategies:

To restrict values of properties such as lipophilicity more stringently than is justified by trends in the data is to deny one’s own drug-hunting teams room to maneuver while yielding the initiative to hungrier, more agile competitors.

There is something to this, though acting on it is not without risk. As the old saying goes, nobody gets fired for buying IBM. Most drug discovery efforts fail, but if you fail making conventional compounds, you’re less likely to come under fire than if you fail by doing something outside the accepted norm.

But whatever you do, it’s worth remembering:

The human liver remains an effective antidote to the hubris of the drug designer.