Corante

About this Author
DBL%20Hendrix%20small.png College chemistry, 1983

Derek Lowe The 2002 Model

Dbl%20new%20portrait%20B%26W.png After 10 years of blogging. . .

Derek Lowe, an Arkansan by birth, got his BA from Hendrix College and his PhD in organic chemistry from Duke before spending time in Germany on a Humboldt Fellowship on his post-doc. He's worked for several major pharmaceutical companies since 1989 on drug discovery projects against schizophrenia, Alzheimer's, diabetes, osteoporosis and other diseases. To contact Derek email him directly: derekb.lowe@gmail.com Twitter: Dereklowe

Chemistry and Drug Data: Drugbank
Emolecules
ChemSpider
Chempedia Lab
Synthetic Pages
Organic Chemistry Portal
PubChem
Not Voodoo
DailyMed
Druglib
Clinicaltrials.gov

Chemistry and Pharma Blogs:
Org Prep Daily
The Haystack
Kilomentor
A New Merck, Reviewed
Liberal Arts Chemistry
Electron Pusher
All Things Metathesis
C&E News Blogs
Chemiotics II
Chemical Space
Noel O'Blog
In Vivo Blog
Terra Sigilatta
BBSRC/Douglas Kell
ChemBark
Realizations in Biostatistics
Chemjobber
Pharmalot
ChemSpider Blog
Pharmagossip
Med-Chemist
Organic Chem - Education & Industry
Pharma Strategy Blog
No Name No Slogan
Practical Fragments
SimBioSys
The Curious Wavefunction
Natural Product Man
Fragment Literature
Chemistry World Blog
Synthetic Nature
Chemistry Blog
Synthesizing Ideas
Business|Bytes|Genes|Molecules
Eye on FDA
Chemical Forums
Depth-First
Symyx Blog
Sceptical Chymist
Lamentations on Chemistry
Computational Organic Chemistry
Mining Drugs
Henry Rzepa


Science Blogs and News:
Bad Science
The Loom
Uncertain Principles
Fierce Biotech
Blogs for Industry
Omics! Omics!
Young Female Scientist
Notional Slurry
Nobel Intent
SciTech Daily
Science Blog
FuturePundit
Aetiology
Gene Expression (I)
Gene Expression (II)
Sciencebase
Pharyngula
Adventures in Ethics and Science
Transterrestrial Musings
Slashdot Science
Cosmic Variance
Biology News Net


Medical Blogs
DB's Medical Rants
Science-Based Medicine
GruntDoc
Respectful Insolence
Diabetes Mine


Economics and Business
Marginal Revolution
The Volokh Conspiracy
Knowledge Problem


Politics / Current Events
Virginia Postrel
Instapundit
Belmont Club
Mickey Kaus


Belles Lettres
Uncouth Reflections
Arts and Letters Daily
In the Pipeline: Don't miss Derek Lowe's excellent commentary on drug discovery and the pharma industry in general at In the Pipeline

In the Pipeline

« Are Genes Patentable Or Not? | Main | TRIM21: A Cure For the Common Cold? Maybe Not. . . »

November 2, 2010

Good Old Medicinal Chemistry: What Can You Get Away With?

Email This Entry

Posted by Derek

Medicinal chemists spend an awful lot of time working with SAR, structure-activity relationship(s). That's how we think: hmm, what happens if I put a chloro there? If I make that ring one size larger? If I flip that stereocenter/add a nitrogen/tie back that chain? Ideally, you pick up on a trend that you can exploit to give you a better compound, but the problem is, no SAR trend lasts forever. Methyl's good, ethyl's fine, anything bigger falls off the cliff - that sort of thing.

Activity "cliffs" of this sort are the subject of a paper earlier this year in the Journal of Chemical Information and Modeling. (For some earlier approaches to this same type of question, see here, here, here, and especially here).This group (from Germany) looked over several public SAR databases and used a new algorithm to extract "matched molecular pairs", which are compounds that differ only at one point in their structure. And what they were looking for wasn't the orderly progressions; they were after the changes that tended to suddenly change the activity of a compound by at least 100-fold. Were there, they wondered, functional group shifts that have a greater or lesser chance of doing that, over a wide range of targets and compound classes?

It looks like there are, and they're the transformations that you might well imagine. Messing around with a carboxyl group, for example, seems rarely to be a neutral event. Carboxylates are so relentlessly polar and hydrogen-bonding that your SAR is probably going to love 'em or hate 'em. The next two liveliest groups were carbonyls (in general) and amines. Of less interest (but equally believable) is the transformation from methyl to bulky alkyl (or vice versa, which is the direction I'd recommend people try to go if at all possible - other things being equal, no one should grease up their compounds unless there's absolutely no choice).

Well, it needs no ghost come from the grave to tell us this, either. How about any surprises? Adding a secondary hydroxyl group was surprisingly silent, compared to what you might picture. And switching from secondary to tertiary amines (just with methyl groups) is a much less conservative switch than you might imagine, with several huge activity shifts across different target classes. Introduction of methyl ethers rarely affected things much one way or another, and that might account for the low tendency of dimethylamine-to-morpholine doing anything. Small halogens on aryl rings (fluorine, chlorine) had low potential to cause big shifts, with ortho-chloros showing no examples of that happening at all. Oddly (at least to me) was the fact that morpholine-to-alkylpiperazine showed almost no big changes, either.

But it has to be emphasized that these are (1) averages and (2) averages over a large (but not gigantic) data set. For example, one of the "no changes at all" transformations is a favorite med-chem isostere, thiophene for phenyl. And that's true - most of the time, that does nothing. But I've seen two examples in my career when that one actually caused a big change in activity, so it's rare, but not impossible. That's the thing that makes med-chem so enjoyable and so frustrating at the same time. It's full of things (like actually discovering a drug) that are rare, but not quite impossible.

Comments (19) + TrackBacks (0) | Category: Life in the Drug Labs


COMMENTS

1. anon the II on November 2, 2010 10:34 AM writes...

When I think of molecules binding in pockets, I tend to think of those big discontinuities as walls rather than cliffs. It's easier to hang things over a cliff than it is to put them through a wall.
Its amazing how many analogies we come up with trying to think through an SAR.

Permalink to Comment

2. Wavefunction on November 2, 2010 11:43 AM writes...

I am always a little wary of such studies. Sometimes they will use fancy machine learning and data mining techniques to come up with a result that's bleedingly obvious to an experienced medicinal chemist. But as you indicate, they can also produce some more interesting, non-intuitive stuff.

Permalink to Comment

3. You're Pfizered on November 2, 2010 11:46 AM writes...

And the magic of putting an F somewhere in a molecule for ADME purposes and having the binding/activity properties change in a way you weren't expecting.

Permalink to Comment

4. milkshake on November 2, 2010 2:28 PM writes...

I always wanted to put an extra SF5- group into my molecule. Apparently they are metabolically-stable, just like CF3 but bigger.

Another metabolically-stable piece with physical properties similar to benzene ring is carborane. (Enter evil laugh here).

Modifying carboxyl: I have seen examples of successful swap of CO2H for a primary carboxamide CON2H. Even if the activity in biochemical activity took a hit by this change, the in vivo activity actually improved (since acidic compounds often have ridiculously high plasma protein binding)

Permalink to Comment

5. CMCguy on November 2, 2010 2:33 PM writes...

Nice synopsis Derek and to #2 Wavefunctions point I think such studies actually do re-enforce the value of experience as good general guidance however trust that such knowledge comes with keen awareness than often can be the compound(s) that don't fit the expected patterns that lead to most interesting avenues.

Permalink to Comment

6. will on November 2, 2010 2:37 PM writes...

re secondary and tertiary amines - is there are big difference in the barrier to inversion for the two?

Permalink to Comment

7. sjb on November 2, 2010 4:24 PM writes...

Isn't there a bit of a market for carboranes with BCNT, though?

Permalink to Comment

8. HK on November 2, 2010 5:09 PM writes...

Isn't there some benefit to studies that can predict an outcome that would otherwise take half a lifetime of industrial experience to deduce?

Permalink to Comment

9. Anonymous on November 2, 2010 5:55 PM writes...

When I have used matched molecular pairs (or bioisoster replacement as it is more commonly called) I have found that the activity difference on average correlates well with LogP,

Permalink to Comment

10. Pete on November 3, 2010 4:13 AM writes...

Derek, I need to correct you on one (very minor) point. It is not necessary for matched molecular pairs to differ at only a single point in the structure. Matched molecular pairs (MMPs) are pairs of molecules that have a specific relationship to each other and if you’re interested in the effect simultaneously introducing chloro substituents at both meta positions of a phenyl subsitituent then you could address this question with MMPs. Matched molecular pair analysis (MMPA) is particularly useful for checking the MedChem folklore to see if there’s any data to support opinions which are often stated as fact.

Anonymous 9, there is a difference between MMPs and bioisostere replacements. The two members of the MMP have a specific structural relationship with respect to each other but that relationship does not have to be bioisosteric. For example, Derek has referred to the effects of carboxylates. On the subject of bioiososteres you may be interested in the discussion on (ester) bioisoteres that is currently going on in the Medicinal Chemistry & Drug Discovery LinkedIn group.

Wavefunction, The data analysis is usually very simple in that you average the effect of the structural change on the property of interest and check that that it is significant. It is a data-mining technique (it uses real data) and I’m not sure where the ‘fancy machine learning’ comes in. I like to think about MMPA as different type of local QSAR model.

I have found MMPA to be useful in integrating potency SAR with measured physicochemical properties such as solubility and plasma protein binding. It’s easier to refer you to an article rather than attempt to reproduce everything here so I’ve put a link at the end of my comments. The article presents MMPA for the effect on plasma protein binding of substituting tetrazole for carboxylate. One of the more interesting results from MMPA has been the observation that N-methylation of secondary amides typically increases aqueous solubility. The article shows that this is dependent on structural context and no significant effect was observed for cyclic amides.

Here’s the article (BMCL 19, 850-853):

http://dx.doi.org/10.1016/j.bmcl.2008.12.003

Permalink to Comment

11. Simon on November 3, 2010 7:50 AM writes...

Why should the existence of "cliffs" come as a 'surprise'? - Considering that, e.g., steric mismatch is penalized by steep curves of van der Waals forces, this behavior is evident.

However, with 'data mining' carried out by informaticians (rather than 'real' chemists), the revelations of cliffs may come as a surprise? But not to any moderately-skilled medicinal chemist, one would guess...

Permalink to Comment

12. Wavefunction on November 3, 2010 8:22 AM writes...

10: I am not talking about this study in particular which seems to be much more chemically oriented. But as someone mentioned before, there are also other studies mainly done by informaticians (not chemists) who use a lot of black boxes to divine intuitive understanding. Neural networks especially come to mind.

Permalink to Comment

13. cliffintokyo on November 3, 2010 8:59 PM writes...

Some useful SAR construction pointers, in the article and from Derek.
I am (still) waiting for the day when we can confidently decide not to synthesize an analog on the basis of modeling and SAR considerations.

Permalink to Comment

14. Pete on November 4, 2010 2:17 AM writes...

In response to Cliff's comment, we were trying to reduce binding of the anionic compounds to plasma protein. The matched molecular pair analysis was performed to explore the effect of changing carboxylate to tetrazole. The results of the analysis were so clear that we decided not to make tetrazoles. However, I have to admit that this is one most unambigous matched molecular pair relationships that I have ever observed. The glycogen phosphorylase program was the the first AZ project to get the matched molecular pair treatment and served as a test bed for the approach in the early days (most of the work described in the article dates back to 2002/3).

The tetrazole/carboxylate protein binding relationship is not something that you can address with logP because it is likely that these functional groups bind to albumin in their anionic form.

Permalink to Comment

15. LeeH on November 4, 2010 8:25 AM writes...

I, along with a few others in the industry (notably people at the former Upjohn facility in Kalamazoo) have been thinking about this method for over 15 years, and I've implemented and used the method for the last 10 years or so.

The thing to remember about this method is that it is quite different from most other data mining or QSARish method out there. Using most methods, one looks for trends and central tendencies. When you use matched molecular pair analysis to look at activities that involve some kind of interaction with a binding site, you actually do the opposite. You're looking for the exceptions. Changing a hydrogen to a chlorine may not give you a nice bump in activity most of the time, but in some cases it might. It's those cases that you want to find.

Yeah, I know, now the medicinal chemists are saying "but those cases are immediately obvious. We don't need a fancy computational method." That's true for simple cases (like single atom changes), and where you are looking at a fairly limited number of compounds. It's not the case where the changes are more subtle, or the effect not so extreme, or where you have too many compounds to eyeball (like HTS data). Using matched pair analysis in these situations can reveal hints for changes that might be useful in the context of lead optimization.

Note that in the case of properties like solubility or membrane permeability, the same structural change may lead to the same shift in property, so that more consistent trends may be present.

Permalink to Comment

16. Cliff on November 5, 2010 2:14 AM writes...

#14
Thanks Pete, for your nice illustration in response to my overgeneralisation.
Its a good example of med chem in practice, and should help to get folks thinking about using molecular pairing.

Permalink to Comment

17. James Choi on November 6, 2010 5:58 PM writes...

Hi Pete,
This is just another example of Derek professing his opinion on an area of chemistry that he has no clue on. He professes to be a Medicinal chemist but is just a mixer of chemicals. Doesnt seem to have a clue of medicinal chemistry

Permalink to Comment

18. James Choi on November 6, 2010 6:00 PM writes...

Hi Pete,
This is just another example of Derek professing his opinion on an area of chemistry that he has no clue on. He claims to be a Medicinal chemist but is just a mixer of chemicals. Doesnt seem to have a clue of medicinal chemistry. Once must certainly wonder if theyre handing out Ph.D's on the street at his alma mater.

Permalink to Comment

19. Got QSAR? on November 8, 2010 2:56 PM writes...

"Ideally, you pick up on a trend that you can exploit to give you a better compound, but the problem is, no SAR trend lasts forever."

Hmmm. Where do I begin here? Trends don't last forever or you didn't realize the entire trend from the beginning? Maybe in your Methyl - Ethyl - Propyl - cliff example the trend was increased hydrophobicity with limited size leads to increased activity? Maybe if you plotted a hydrophobicity descriptor with a steric descriptor your trend would have continued? Maybe you would have made -CF3.

In my experience trends fail because we are using the wrong descriptors or aren't using the correct ones.

Medicinal chemistry *IS* finding trends. Serendipity (and molecular modeling for that matter) only gets you so far.

Permalink to Comment

POST A COMMENT




Remember Me?



EMAIL THIS ENTRY TO A FRIEND

Email this entry to:

Your email address:

Message (optional):




RELATED ENTRIES
Molecular Printing of Drug Molecules. Say What?
Pfizer Walks Again By Night
Gitcher SF5 Groups Right Here
Changing A Broken Science System
One and Done
The Latest Protein-Protein Compounds
Professor Fukuyama's Solvent Peaks
Novartis Gets Out of RNAi