Rapid visualisation of bioavailability potential using simple radar plots

1
Rapid visualisation of bioavailability potential
using simple radar plots
Dr Tim Ritchie Global Discovery
Chemistry Novartis Institute for Medical
Sciences London, UK
2
Visualisation of molecular properties important
for oral absorption
"Good "
"Bad"

• What are logP, MW, PSA, WS, and nrotb?
• What are 'good' and 'bad' values for each?
• Can we predict drug behaviour using the radar
  plot?
• Are there exceptions?

3
LogP a partition between polar and non-polar
enviroments
Hydrophobic
Naphthalene will prefer the octanol
layer high logP (3.3)
Glucose will prefer the water layer low
logP (-2.2)
Hydrophilic
octanol
water
4
LogP
The combination of hydrophobic and hydrophilic
regions in a molecule contributes to the overall
logP value
Hydrophobic
Hydrophilic
Glivec clogP 4.5
Prilosec clogP 2.6
5
The MDL Comprehensive Medicinal Chemistry (CMC)
database

• The mean clogP in the CMC drugs database is 2.51
• LogP is important in membrane permeation (lipid
  bilayers)
• Poor absorption/permeation is more likely when
  logP is over 5 (a 'Rule-of-5' from Lipinski et
  al, Adv Drug Del Rev 1997, 23, 3).

6
MW Molecular weight
The molecular weight of a compound is a measure
of its overall size.
Water 18.0 g/mol
Clozaril 326.8
Glivec 493.6
Sandimmun 1202.6
7
MR (Molar refractivity) vs. MW (Molecular weight)
8
MW Molecular weight

• The mean MW in the CMC is 348.9 g/mol
• Poor absorption/permeation is more likely when MW
  is over 500 ('Rule-of-5' from Lipinski et al, Adv
  Drug Del Rev 1997, 23, 3).

9
PSA Polar Surface Area
Solvent-accessible surface around molecule
Polar surface area focusses on hydrophilic groups
that can form H-bonds

• The Novartis PSA calculation algorithm requires
  only the 2-D structure and is very fast (Ertl et
  al, J Med Chem 2000, 43, 3714)
• This approach has been adopted by many
  pharmaceutical companies and software vendors

10
Polar surface area vs. hydrogen bonding potential
Donors N-H O-H Acceptors 'O''N'
Polar surface area
Sum of H-bond donors acceptors
11
PSA Polar Surface Area

• The mean PSA in the CMC database is 75.1 Å2
• Polar surface area is known to strongly influence
  membrane permeability
• PSA values gt130 Å2 result in lower oral
  bioavailability

From DE Clark, J Pharm Sci 1999, 88, 807.
12
WS Water solubility score

• Orally administered drugs have to dissolve in the
  gastro-intestinal tract in order to be absorbed!
• Low aqueous solubility may
• delay or limit oral absorption
• make formulation difficult
• limit maximum possible dose in man

• The Novartis water solubility algorithm gives a
  qualitative ranking between 1 and 5
• 1 very good solubility gt 1 g/L
• 2 good solubility 0.1 - 1 g/L
• 3 medium solubility 0.01 - 0.1 g/L
• 4 poor solubility 0.001 - 0.01 g/L
• 5 very poor solubility lt 0.001 g/L

13
WS Water solubility score

• The mean WS in the CMC is 2.5
• A score of 1, 2, or 3 is considered acceptable

Medium
Medium
Medium
High
High
High
Low
Low
Low
14
nrotb number of rotatable bonds

• The number of rotatable bonds in a molecule gives
  an indication of its flexibility.

Clozaril has only one rotatable bond...
Glivec has seven rotatable bonds...
15
nrotb number of rotatable bonds

• The mean nrotb in the CMC is 5.56
• Permeability is significantly reduced if nrotb is
  gt10

From DF Veber et al, J Med Chem2002, 45, 2615.
16
How to display the data?

• We have selected five molecule properties
  important for oral absorption/bioavailability
• We can calculate or measure them
• We have defined limits for each property based on
  evidence from in-house and literature studies
• Can we view all this information in a useful way?

17
The radar plot

• In this example, the average values for the five
  properties in the CMC database are used

• Ideally, the blue pentagon should fall well
  within the green area

18
Graphical vs. tabular data
19
Can we predict good and bad drug molecules?
The vast majority of marketed oral drugs fall
within the radar plot limits
Glivec
Viagra
Seroxat
20
Can we predict good and bad drug molecules?
Three examples of past development candidates
that failed - potent and selective but...
No oral absorption in human volunteers
Terminated due to "technical formulation issues"
Low oral bioavailability in rats
21
Which compound has more chance of success?
10 nM combichem hit
6000 nM combichem hit
22
What about more than one compound?

• Properties for two compounds can be displayed,
  e.g.
• Different hits or leads
• An initial (blue) and an optimised (cyan) lead
• How about thousands of compounds?

23
What about more than one compound?

• Two Novartis combichem libraries (3,000 compounds
  in each)
• The average values for the entire library are
  used for the plot
• The Standard Deviation is shown as a pale blue
  band
• Giving an idea of diversity
• Virtual libraries or vendor collections can be
  screened

24
What about 'real' drugs?
a - Novartis drugs Diovan, Lamisil, Glivec,
Zometa, Femara and Zelmac b - Blockbusters
2001 Lipitor, Prilosec, Norvasc, Celebrex,
Vioxx, Seroxat, Ogastro, Prozac, Zocor, and
Zyprexa Pharma company pipelines?
25
Can we predict good and bad drug molecules?
'Good' plots do not always mean 'good drugs'
Resibufogenin A toxin from toad venom!
2,4,5-T A component of Agent Orange!
26
The radar plot cannot predict biological activity!

• Compounds that fit within the good property space
  should exhibit reasonable oral absorption.
• BUT many substances fit into the 'good' property
  space, and
• suffer from metabolism, efflux, etc
• are toxic
• bind to undesirable targets
• The radar plot cannot predict biological activity!

27
Can we predict good and bad drug molecules?
'Bad' plots do not always mean 'bad drugs'
Sandimmun/Neoral Heroic formulation effort!
28
Compounds that break the limits are 'higher risk'

• Sandimmun is an exception
• Compounds that fall outside 'good' space are more
  likely to suffer from complications e.g.
• Low or variable bioavailability
• Formulation difficulties
• High protein binding / accumulation
• Delays in development or worse

29
In summary - a tool to visualise important
molecule properties
"Bad"
"Good "

• Single or multiple compounds can be evaluated
• The majority of oral drugs fall within the limits
• Compounds that break limits may suffer
  complications

30
Novartis web-based in silico profiling page

• The prototype radar plots were created in Excel
  using calculated data and a simple macro
• The radar plot is now available as a Java-based
  web tool on the Novartis Intranet
• Input is either an ISIS/Draw structurean SD
  fileSMILES codescompound IDs
• Each plot is generated in about 2 seconds

31
Acknowledgements

• Thanks to Peter Ertl Paul Selzer Jörg
  Mühlbacher Terry Hart Bernd Rohde