Talk given 2016-03 | CUP XVI | Santa Fe, NM https://www.eyesopen.com/events/2016/03/cup-16 matched-pairs and internal H-bonds

1. Matched Pairs - Closing the Circle

2. Secondment to work with Andrew Grant, Alderley Park (2002-2003)
Shared office with Pete Kenny – talked a lot
….about a method to bring the most interesEng pairs of molecules to the top
those for which the smallest structural change give rise to the biggest change in, say, affinity.
For any [matched]-pair of molecules i and j:
My Matched-Pair Journey
where ΔAij is the ra,o of the affini,es,
Sij is the similarity of the two molecules
From manual to automaEzed Medicinal Chemistry
Cheminforma5cs (OEChem)

3. Cheminforma5cs – Tools
AstraZeneca has a long history of developing various Breaking-Bonds/R-group tools,
in parEcular matched-molecular pair tools:
WizePairZ – to exploit MMPs with unspecified cores in Medicinal Chemistry
Periscope – to encode and search Markush structures in chemical patents
FOG analysis – key compound predicEon in patents using frequency of R-groups
Scaffold Hopping – a tool for scaffold hopping by fragment replacement in 3D
SAR spoZer – web-based tool to find “interesEng” pairs of molecules
Avesome – to exploit MMPs with specified cores in Medicinal Chemistry
Matsy – using Matched Molecular Series as a tool to opEmize biological acEvity
…
+ 10

4. Cheminforma5cs – Impact
Used in prospecEve designs, facilitaEng development of candidate drugs
Johansson, A. et al “Discovery of …AZD1979…” J.Med.Chem ASAP, 2016
I did that pic!
Fame and glory
J
Several independent publicaEons, and JMedChem twiZer background
Boström, J. et al “Oxadiazoles in Medicinal Chemistry” J.Med.Chem 2012, 55, 1817-1830
Binding Affinity: 80 nM
Log D: 3
HLM Clint: 20 µl/min/mg
Binding Affinity: 40 nM
Log D: 2
HLM CLint: 10 µl/min/mg
1,3,4-oxadiazolshowed beZer properEes than 1,2,4-oxadiazoles
in terms of lipophilicityes, solubility, HLM Clint and hERG
O
O N
N
R1R2
O
N N
O
R1R2

5. Social behaviors
Drug design not always that raEonal:
para-subsEtuEon, amide-bond/Suzuki couplings, …
Boström & Brown ”Stuck in a rut with old chemistry”, Drug Discovery Today, 2016, Guest Editorial, ASAP
Brown & Boström ”Analysis of Past and Present SyntheEc Methodologies on Medicinal Chemistry: Where Have All the New ReacEons Gone?” J.Med.Chem, 2016, ASAP
Brown & Boström ”Understanding Our Love Affair with p-Chlorophenyl: Present Day ImplicaEons from Historical Biases of Reagent SelecEon” J.Med.Chem, 2015, 58, 2390.
Russell Crowe as John
Nash in “A BeauEful
Mind” can see paZerns
where there are no
paZerns.
By applying hypothesis-
based approaches –
conducEng experiments
with testable predicEons,
knowledge can supersede
storytelling

6. Matched Pairs – Closing The Circle
What’s the deal with
medicinal chemists,
internal H-bonds
and ring-closing?
Seinfeld
drug designers like rings
Behaviors – para-subsEtuEon, amide-bond/Suzuki couplings, and…
Boström & Brown ”Stuck in a rut with old chemistry”, Drug Discovery Today, 2016, Guest Editorial, ASAP
Brown & Boström ”Analysis of Past and Present SyntheEc Methodologies on Medicinal Chemistry: Where Have All the New ReacEons Gone?” J.Med.Chem, 2016, ASAP
Brown & Boström ”Understanding Our Love Affair with p-Chlorophenyl: Present Day ImplicaEons from Historical Biases of Reagent SelecEon” J.Med.Chem, 2015, 58, 2390.

7. Rings Present in Drug Databases
1.3%
98.7%
ChEMBL 0.9%
99.1%
AstraZeneca
any ring-size and type (aliphaEc/aromaEc), ring-processing using the OEChem funcEon OEFindRingAtomsAndBonds
to determine which atoms and bonds are members of one or more rings (and which are acyclic)
Taylor, et al ”Rings in Drugs” J. Med. Chem., 2014, 57, 5845 (frameworks, frequencies)
4.8%
95.2%
Known Drugs*
Compounds without Ring(s)
Compounds with Ring(s)
Most compounds include one or more rings
– drug designers like rings

8. First Hypothesis
“whenever a drug designer sees an internal hydrogen he want to ring-close it”
See this
Make that
survey a bunch of drug designers!

9. Thoughts About Ring-Closures
Survey submiZed to Medicinal Chemists at three AstraZeneca sites: in Mölndal, Cambridge, and Boston

10. Survey Says – 80 % “yes”
(N=41)
9
32
0
None said
“no”!
“if a drug designer see an internal hydrogen they want to close it” TRUE
N=41
Richard
Dawson

11. “The main reason for ring-closing is to gain potency”
reduce entropic cost of binding by freezing the bioacEve conformaEon results in a more potent compound?
See this
Make that
Second Hypothesis
survey a bunch of drug designers…again!

12. Thoughts About Potency…
Survey submiZed to Medicinal Chemists at three AstraZeneca sites: in Mölndal, Cambridge, and Boston
…and what happens with lipophilicity?
(lipophilicity is a cardinal property in drug discovery –
it can affect many other properEes relevant in lead opEmizaEon)

13. Survey Says (Part 1)
Around half (53%) say “same”, very few (8%)
answer “no “, and a third (38%) answers “yes “
“same potency”
got most votes
me wrong“In general do you
think ring-closed
compounds are more potent?

14. “Closing the ring reduces entropic penalty upon binding which is
why dG becomes more favorable compared to open-ring.”
“You have to be lucky if the ring-closed compound is more potent.
But if it is that is great.”
“I think these are a case-by-case basis and no generaliza,on can be made.
I think they are definitely worth making to explore.
Some,mes it works some,mes it doesn't.”
“Only when the right/ac?ve conforma?on is locked by the ring closure a
more potent compound is seen.”
“It could be more potent – experiment is king!”
Survey Comments – Potency

15. Around half (57%) say “yes”, very few (10%) answer
“no “, and a third (32%) answers “same“
people say “yes”
lipophilicity increases
Survey Says (Part 2)
“In general do you
think ring-closed
compounds are more lipophilic?

16. “Same lipophilicity since the H-bond in the open molecule only bonds internally
giving the overall H-bond capacity to be roughly the same.”
“In terms of intramolecular HB they are masked and can not be counted as normal.
So no major change in lipophilicity.”
“It depends doesn't it?”
Survey Comments – Lipophilicity

17. “it may result in a scaffold which will have a different posi,on with regard to future
IP filings”
“Selec,vity may differ a lot since the internal H-bonded compound may bind another
target in an open non-H-bonded conforma,on.”
“[yes I’d ring-close] as well as I'd like to do the opposite”
Other [Important] Comments
There are other aspects (IP, Selec5vity, SAR, …)

18. Ring-Closing an Autonomic Response?
– is it generally effecEve or ineffecEve?

19. Big-data mining approach
using Matched-Molecular Pair technology
(Python/OEChem)
…and potency (ChEMBL) data
…and lipophilicity (IBIS) data
…some other parameters

20. • “Molecules that differ only by a small structural change”
(remainder of the molecule is exactly the same).
For example
• Osen easier to predict differences in a value of a property than predicEng the value itself
(one reason is cancellaEon of errors).
• The observaEon of an effect across several chemical series increases our confidence that
the effect is real -> general “rules of thumbs” -> future design
• Large amounts of data across are available either publicly (e.g. ChEMBL) or internally within
pharmaceuEcal companies.
• Data greedy – one limitaEon is that it can only make predicEons about structural features
that have precedent in the given assay.
Matched-Pairs – What is it?
Prazosin Terazosin
Solubility: 1.1 mg/ml Solubility: 28.1 mg/ml
* Water solubility for Terazosin increases significantly when going from furan to THF. As a result, improved bioavailability
(90% vs 57%) and half-life (2–3 ,mes) afford longer dura,on of ac,on, and allow once-daily administra,on.
Giordanetto, Boström and Tyrchan “Follow-on drugs: how far should chemists look?” Drug Discovery Today, 2011, 16 (15-16),722-732

21. Matched-Pairs – What to expect?
Two most common R groups are: hydrogen [*-H], methyl [*-C] (using “Matsy”)
This matched-pair is ”random”.
Potency
200k MP’s
pIC50’s from >5000 assays)
Cohen's d effect size: 0.03
Mean: 0.03 (pIC50’s: 5.88-5.84)
magic methyl…?
A parEcular matched pair transformaEon is equally likely to increase acEvity as decrease it.
no effect

22. Two most common R groups are: hydrogen [*-H], methyl [*-C]
Lipophilicity
There’s enrichment, since we expect 6250 for each if random
Cohen's d effect size: 0.24
Mean: 0.28 (logD’s: 2.25 - 1.97)
Matched-Pairs – What to expect?
small effect magic hydrogen!

23. Summary so far…
Most compounds (>95%) include one or more rings
There’s a wide interest (>80%) in designing/making “rings-closed” analogs
Matched molecular pairs is a powerful method for mining data
• Especially for phys-chem properEes (logD)
• Not as successful for biological acEviEes (normal distribuEons around zero)

24. Matched Pairs – Closing The Circle
cheminforma5cs
using SMARTS
Check distances in low-
energy conforma5ons
2D
3D
Methods to identify internal H-bonds ring-closure pairs (2D or 3D)

25. Workflow to Detect 2D ‘Closure-Pairs’
Record molecules that hits SMARTS, tag as “open”
The number of rings counted (also used to speed up search)
[O;H0:1]=,-*~*~*~[N;H,H2:1] aVeryGeneralSMARTS
Cl
NH
R3
R1 "
R2 "
“open”
“fragments”
search if
fragments match
+1 contain
addi,onal ring
Cl
NH
N
N
N
checks if patch distance of
connecEon points are correct
all unmatched atoms
required to be in a ring*
à A closed-case
All “open” molecules are fragmented and connecEon points monitored,
as well as the path distances between connecEon points.
“close”
Step 1
3 rings
4 rings
N
R4
*any ring allowed, but restricEons on num donors/acceptors
database
database
Py/OEChemTk (José BaEsta and Magnus Norrby) – ComputaEonally very expensive calculaEons
Step 2
Step 3

26. Workflow to Detect 3D ‘Closure-Pairs’
Take all “open”-cases [internal H-bonds] in the ring-closure pairs from 2D
Expand into mulE-conformaEonal ensembles (OMEGA-Tk)
Check for internal H-bond distances (<3.5Å) among low-energy conformaEons
Virtually all retrieved by the 3D approach – jusEfying the 2D approach*
* electrostaEc ‘collapse’ of the ligand in vacuo
Boström et al “ConformaEonal energy penalEes
of protein-bound ligands”, 1998, J.Comput-
Aided Mol. Design 12, 383-383
The importance of using solvaEon models…
electrostaEcs sEll tricky to get right.

27. Potency Database – ChEMBL v20
Addi5onal closure-pair criteria
• Closure-pairs must be tested in the same assay
• RestricEons on pairs:
• ΔH-donors ≤ 2
• ΔH-acceptors ≤ 2
• ΔH-acceptors and H-donors ≤ 3
• Δformal charge == 0
DB Details…
• Only use pIC50 data for binding assays “B”
• 63 940 targets
• 752 228 acEviEes (could be extended with 50%; 429 347 Ki values)
• 323 055 disEnct compounds
• 271 440 (15 ≤ heavy atom count ≤ 40)
• 83 711 with a possible internal H-bond (hits SMARTS)
>60k papers: 94% from Bioorg. Med. Chem. LeI., J. Med. Chem., J. Nat. Prod., Bioorg. Med. Chem., Eur. J. Med. Chem., Med. Chem. Res.

28. pIC50 (ChEMBL) Closure-Pairs
Normally distributed around zero - equally likely to gain potency as loosing potency.
Δ pIC50
”close” more potent ”open” more potent
N = 2200
(850 uniq)
Counts
pIC50 (”open”) – pIC50 (”closed”) = 0.05 (6.33-6-28)
Sama analysis on internal data more or less iden,cal – no “file-drawer” effect

29. Not all internal H-bonds are equal
SMARTS defini5ons – adding context
[O;H0:1]=[C]-[A,a;X4]~[A,a;X4]-[O;H:1] ACCcarbonyl_LinkerSp3_DONhydroxy_Dist5
[O;H0:1]=[C]-[A,a;X4]~[A,a;X4]-[N;H,H2:1] ACCcarbonyl_LinkerSp3_DONnitrogen_Dist5
[O;H0:1]=[C]~[A,a;X3]-[O;H:1] ACCcarbonyl_LinkerSp2_DONhydroxy_Dist4
[O;H0:1]=[C]~[A,a;X3]-[N;H,H2:1] ACCcarbonyl_LinkerSp2_DONnitrogen_Dist4
Acceptors (5): carbonyl, heterocyclic N, hydroxyl, S=O, alkoxy
Donors (2): hydroxl, nitrogen
Linker-type (2): sp2, sp3 (aliphaEc or aromaEc)
Linker-length (3): 4, 5, 6
[O;H0:1]=[C]~[A,a;X4]-[O;H:1]
ACC:carbonyl
Linker: sp3
DON: hydroxy
Dist4:
(easy to customize)
60 SMARTS combinaEons
strong poor good

30. Some types of ring-closures increases potency on average?
Calculate cohen’s effect size for all SMARTS (incl pairs>20)
Closure-Pairs – Potency
SMARTS Mean d2 err_d d pairs
ACCmetoxy_LinkerSp2_DONhydroxy_Dist4 -0.57 0.61 0.78 0.13 107
ACChydroxy_LinkerSp2_DONnitrogen_Dist4 0.36 0.37 0.61 0.11 124
ACChydroxy_LinkerSp3_DONhydroxy_Dist5 0.45 0.25 0.50 0.20 34
ACCmetoxy_LinkerSp2_DONnitrogen_Dist4 -0.23 0.25 0.50 0.15 63
ACChydroxy_LinkerSp3_DONnitrogen_Dist5 -0.32 0.23 0.48 0.11 113
ACCmetoxy_LinkerSp3_DONnitrogen_Dist5 -0.29 0.21 0.46 0.10 120
ACCcarbonyl_LinkerSp2_DONhydroxy_Dist5 0.32 0.20 0.45 0.15 55
ACCmetoxy_LinkerSp3_DONhydroxy_Dist5 0.29 0.18 0.42 0.16 47
ACChydroxy_LinkerSp3_DONhydroxy_Dist4 0.24 0.13 0.37 0.09 155
ACCcarbonyl_LinkerSp3_DONhydroxy_Dist6 0.11 0.09 0.29 0.18 35
ACCcarbonyl_LinkerSp2_DONnitrogen_Dist5 0.12 0.09 0.29 0.18 36
ACChydroxy_LinkerSp2_DONhydroxy_Dist4 -0.08 0.07 0.25 0.11 98
ACChydroxy_LinkerSp3_DONnitrogen_Dist4 -0.12 0.07 0.27 0.05 382
ACCmetoxy_LinkerSp3_DONhydroxy_Dist4 0.10 0.06 0.25 0.08 162
ACCmetoxy_LinkerSp3_DONnitrogen_Dist4 -0.08 0.05 0.23 0.05 466
Signal!?
…but only 11 uniq pairs
…but only 20 uniq pairs

31. There Are Many Ways to Make a Ring
Internal H-bond
CHEMBL1143672: Synthesis and pharmacology of site-specific cocaine abuse
treatment agents: restricted rotaEon analogues of methylphenidate.
CHEMBL1203942 CHEMBL904
pIC50: 6.6 pIC50: 5.7
“The results suggest that the conforma,on of methylphenidate (MP) in which the carbonyl
group of the methyl ester is H-bonded to the piperidinyl N−H may be the bioac,ve form of
the molecule.” J. Med. Chem. (2007) 50:2718-2731
Displacement of [3H]nisoxeEne from norepinephrine transporter (NET) in Sprague-Dawley rat corEcal Essue at 5 uM
Ring-Closed
(6-ring)
less potent
agents [ACCcarbonyl_LinkerSp3_DONnitrogen_Dist5]

32. CHEMBL1129655: A comparaEve molecular field analysis study of N-benzylpiperidines
as acetylcholinesterase inhibitors.
CHEMBL333909
CHEMBL56079
pIC50: 6.1 pIC50: 7.0
“compounds have been found to inhibit the metabolic breakdown of the neurotransmijer
acetylcholine (ACh) by the enzyme acetylcholinesterase (AChE) and hence alleviate memory
deficits in pa,ents with Alzheimer’s Disease by poten,a,ng cholinergic transmission.”
J. Med. Chem. (1996) 39:380-387
(5-ring)
more potent
There Are Many Ways to Make a Ring
Internal H-bond Ring-Closed
[ACCpyr_LinkerSp2_DONnitrogen_Dist4]

33. IdenEficaEon of a New Chemical Class of Potent Angiogenesis Inhibitors Based
on ConformaEonal ConsideraEons and Database Searching
CHEMBL101253 CHEMBL101683
equipotent
ΔpIC50: -0.1
There Are Many Ways to Open a Ring
VEGF tyrosine kinase receptors KDR and Flt-1 are targets in anEcancer
“resulted in a compound with the same level of potency and selec,vity”
Furet et al Bioorg.Med. Chem. LeZ. 13, (2003) 2967–2971
design
[ACCcarbonyl_LinkerSp2_DONnitrogen_Dist5]

34. Closure-Pairs – logD
lipophilicity (logD) increases with 0.6, on average, upon ring-closure
Cohen's d effect size: 0.64
Mean: 0.6 (logD’s: 2.4 - 1.8)
medium effect
Calculated lipophilicity (clogP) very similar profile: 3.3 -2.4
N= 2400

35. Closure-Pairs – logD
All types of ring-closures increases lipophilicity on average,
some more than others
SMARTS Mean Cohens d2 err_d d no pairs
ACCcarbonyl_LinkerSp3_DONnitrogen_Dist4 1.20 1.49 1.22 0.38 21
ACCcarbonyl_LinkerSp3_DONnitrogen_Dist5 1.06 0.95 0.97 0.27 32
ACChydroxy_LinkerSp3_DONnitrogen_Dist5 0.78 0.91 0.95 0.13 135
ACCpyriLinkerSp3_DONnitrogen_Dist5 0.86 0.89 0.95 0.24 39
ACCpyriLinkerSp2_DONnitrogen_Dist5 0.67 0.82 0.91 0.28 27
ACCmetoxy_LinkerSp3_DONnitrogen_Dist5 0.75 0.82 0.91 0.12 152
ACCmetoxy_LinkerSp3_DONnitrogen_Dist4 0.63 0.59 0.77 0.06 567
ACChydroxy_LinkerSp3_DONnitrogen_Dist4 0.56 0.56 0.74 0.06 471
ACChydroxy_LinkerSp3_DONhydroxy_Dist4 0.47 0.49 0.70 0.12 125
ACCmetoxy_LinkerSp2_DONnitrogen_Dist5 0.39 0.47 0.69 0.26 24
ACChydroxy_LinkerSp3_DONhydroxy_Dist5 0.46 0.43 0.65 0.20 39
ACCpyriLinkerSp2_DONnitrogen_Dist4 0.37 0.40 0.63 0.13 89
ACCmetoxy_LinkerSp3_DONhydroxy_Dist4 0.44 0.40 0.63 0.08 256
ACCcarbonyl_LinkerSp3_DONhydroxy_Dist6 0.44 0.38 0.62 0.14 82
ACCmetoxy_LinkerSp3_DONhydroxy_Dist5 0.35 0.37 0.61 0.13 86
ACCcarbonyl_LinkerSp2_DONnitrogen_Dist5 0.24 0.26 0.51 0.15 64
0.60 0.64
(no pairs>20)
Signal!
…all uniq pairs

36. average Caco-2 (”closed”) – average Caco-2 (”open”) = -3.0 (within experimental error!)
N = 60Permability
SCHEMBL3813720
Caco-2 PappAB (10-6cm/s): 33.5
SCHEMBL9832404
Caco-2 PappAB (10-6cm/s): 4.0
SCHEMBL4122434
Caco-2 PappAB (10-6cm/s): 21.9
SCHEMBL4123304
Caco-2 PappAB (10-6cm/s): 1.0
GVK2154080
Caco-2 PappAB (10-6cm/s): 4.0 SCHEMBL4981036
Caco-2 PappAB (10-6cm/s): 1.1
same
Two cases where ”open” form more permeable
No trend observed, lipophilicity effect not reflected in permabiliEes
Need more examples to draw conclusions…

37. What about Solubility?
Average pSolubility (”closed”) – average pSolubility (”open”) = 0.2
N>1000
exp Aq Sol on AZ pairs
”open” more soluble”close” more soluble
Solubility (Aq) decrease with 0.2, on average, upon ring-closure
Δ pSolubility
Counts

38. Protein-ligand complexes (X-ray)
HET
PDB
UniProt
PDB dump: 11886 ligands (all-against-all)
15 ≤ heavy atom count ≤ 40
75 pairs
remove duplicates
check if bind to same target (UniProt ID)
present in ≤ 10 pdb’s (reagents, ATP, ADP, etc)
6 pairs (might have lost a few in translaEon)

39. Protein-ligand complexes – Example
L. Zhao et al. Bioorg. Med. Chem. LeZ. 20 (2010) 7216–7221. The respecEve PDB access codes are 3PA3, 3PA4 and 3PA5
equipotent IC50
(nM) at CHK1
Design, synthesis and SAR of thienopyridines as potent CHK1 inhibitors
Tanimoto Shape : 0.97 Tanimoto Shape : 0.97

40. Summary
+80% of AZ drug designers want to make/design ring-closed analogs
when spo…ng an internal H-bond
There are many (perceived) reasons FOR and AGAINST ring-closures
• IP, potency, SAR, lipophilicity, solubility, syntheEc feasibility, permeability, curiosity,…
New OEChem matched-pair code run on experimental data reveals that,
• it is on average equally likely to gain potency as loosing potency
• lipophilicity increases in general with nearly a log unit
• some paZerns give bigger effects
Lots of interesEng scienEfic quesEons...
• The data-set is great source for new ideas
Yes it can be an effec5ve design strategy…
but consider gain vs pain