New pharmaceutical compound

High oral bioavailability is desired

Good oral absorption is the first requirement

Interest in predicting human intestinal absorption (HIA) [1]

Effect on the central nervous system (CNS)?

Several mechanisms regulating drug permeability to the brain

Blood-brain barrier (BBB) is the most important

High BBB permeability indication for CNS effect

Common measure: log 𝐵𝐵 = log(𝐶𝑏𝑟𝑎𝑖𝑛

𝐶𝑏𝑙𝑜𝑜𝑑) [2]

Time (min)

-20000

30000

80000

130000

180000

230000

280000

330000

0 2 4 6 8 10 12 14 16 18 20

35

30

25

20

15

10

5

0

Inte

nsi

ty (

mV

)

Acetylsalicylic acid

-10000

0

10000

20000

30000

40000

50000

0 10 20 30 40 50 60

6

5

4

3

2

1

0

Time (min)

Inte

nsi

ty (

mV

)

Halothane

-50000

-30000

-10000

10000

30000

50000

0 10 20 30 40 50 60 70 80 90

10

8

6

4

2

0

Time (min)

Inte

nsi

ty (

mV

)

Ibuprofen

-100000

0

100000

200000

300000

400000

500000

600000

0 10 20 30 40 50 60

70

60

50

40

30

20

10

0

Time (min)

Inte

nsi

ty (

mV

)

Amobarbital

Time (min)

-40000

160000

360000

560000

760000

960000

0 2 4 6 8 10 12 14 16 18 20

100

80

60

40

20

0

Inte

nsi

ty (

mV

)

Acetaminophen

-50000

0

50000

100000

150000

200000

250000

300000

0 10 20 30 40 50 60 70 80 90

35

30

25

20

15

10

5

0

Time (min)

Inte

nsi

ty (

mV

)

Indomethacin

Synthesis of a phosphocholine based lipid for improved

micellar LC based in vitro predictions of human intestinal

absorption and blood-brain barrier partitioning

M. De Vrieze1, P. Janssens2, R. Szucs3, J. Van der Eycken2, F. Lynen1

1 Separation Science Group, Department of Organic and Macromolecular Chemistry, Ghent University, Krijgslaan 281 S4-bis, B-9000 Ghent 2 Laboratory for Organic and Bioorganic Synthesis, Department of Organic and Macromolecular Chemistry, Ghent University, Krijgslaan 281 S4, B-9000 Ghent 3 Pfizer Global R&D, Sandwich CT13 9NJ, Kent, United Kingdom

INTRODUCTION

The log BB model performed better compared to HIA

prediction, although data provided by MLC with miltefosine as

surfactant contributed in a positive way to both models.

This approach shows potential as an alternative or

complementary MLC strategy to predict in vivo behavior.

Additional research, using a variety of (phospho)lipids as

surfactant for MLC, might be very interesting, since this could

better mimic the composition of biological membranes.

REFERENCES

Figure 1: Drug interactions in MLC. Above the

CMC, retention depends on interactions with the

modified stationary phase and with the micelles

present in the mobile phase.

[1] Hou TJ, Wang JM, Zhang W, Xu XJ (2007) J Chem Inf Model 47 (1):208-218.

[2] Fu XC, Wang GP, Shan HL, Liang WQ, Gao JQ (2008) Eur J Pharm Biopharm 70 (2):462-466.

[3] Molero-Monfort M, Escuder-Gilabert L, Villanueva-Camañas RM, Sagrado S, Medina-Hernández MJ (2001) J Chrom B 753

:225-236

[4] Zhang QH, Horst R, Geralt M, Ma XQ, Hong WX, Finn MG, Stevens RC, Wuthrich K (2008) J Am Chem Soc 130 (23):7357-

7363.

RESULTS & DISCUSSION

CONCLUSION

EXPERIMENTAL

MLC

The retention factors (k) of the compounds were determined

Several molecular descriptors were added to the model Total molar charge Molecular weight Molar refractivity

Molar volume Parachor Polarizability

Log P Log D 7.4 Intrinsic aqueous solubility

pH solubility profile Plasma protein binding Ames test mutagenic index

Human intestinal absorption Polar surface area Hydrogen bond acceptor

Hydrogen bond donor Molecular surface area

Partial least squares (PLS) regression

Correlation coefficient (R) between actual (in vivo) and predicted values

Selecting the most relevant descriptors: monitor effect on the leave-one-

out cross-validation (LOOCV) regression coefficients upon systematic

removal and/or reinsertion of all descriptors from the models

Both models: remove Molar refractivity Molar volume

Log P Log D 7.4

Ames test mutagenic index Hydrogen bond donor

HIA and log BB

0.01 M miltefosine was dissolved in the mobile phase. The pH was adjusted

with a phosphate buffer at pH 7.4. The osmotic pressure was reproduced by

addition of NaCl (9.20 g/L). Column & flow rate: GraceSmart C18 column (3

µm, 150 mm x 2.1 mm) at 37 °C, flow rate 0.2 ml/min.

In vitro HIA and log BB prediction using

Micellar liquid chromatography (MLC)

RPLC; surfactant above critical micellar

concentration (CMC) in mobile phase

Secondary equilibrium (Figure 1) [3]

Stationary phase bulk solvent

Micelles bulk solvent

Retention time + descriptors model

Thus far in MLC: SDS, Brij35, CTAB were

used as surfactants.

Not comparable to membrane lipids

Miltefosine (Figure 2) is presented here

as an alternative MLC-surfactant

Model construction based on log k + computed descriptors

HIA prediction experimental HIA values

Log BB prediction experimental log BB values

Finally: model evaluation

Figure 2: Synthesis of miltefosine.

Synthesis of miltefosine

The synthesis route (Figure 2) was slightly modified from a previously

reported procedure by Zhang et al. [4]. HPLC-TOF-MS, 1H-NMR and 13C-

NMR were used for structure confirmation.

The results from the PLS and LOOCV regressions before and after

elimination of superfluous molecular descriptors are presented in Table 1. The

large difference in correlation coefficient before optimization is an indication of

overfitting in the model. By removing unnecessary descriptors, the overfitting

was reduced a lot. The final correlation coefficient of HIA (0.7175; based on

36 compounds) was lower than that for log BB (0.7849; based on 48

compounds). For both predictions, data provided by MLC with miltefosine

proved to contribute in a positive way.

The correlation between actual and predicted HIA and log BB values is

illustrated in Figure 4 before and after optimization. Although there are a few

outsiders, the predicted values for most compounds are close to the actual (in

vivo) determined values.

Figure 3: Chromatograms for some compounds using MLC with miltefosine as surfactant

To illustrate the retention behavior in purely aqueous MLC with 0.01 M

miltefosine, some chromatograms are presented in Figure 3.

HIA Log BB

R(PLS) 0.8237 0.8827

R(LOOCV) 0.3666 0.5298

HIA Log BB

R(PLS) 0.7991 0.8484

R(LOOCV) 0.7175 0.7849

Before optimization

After optimization

Table 1: Correlation coefficients between actual

and predicted HIA and log BB values using PLS

and LOOCV before and after optimization of

molecular descriptors.

40 50 60 70 80 90 10040

50

60

70

80

90

100

110

40 50 60 70 80 90 10030

40

50

60

70

80

90

100

110

120

130

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2-2

-1.5

-1

-0.5

0

0.5

1

1.5

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

3

Actual HIA

R = 0.3666

3

2.5

2

1.5

1

0.5

0

-0.5

-1

-1.5

-2

Pre

dic

ted

log B

B

Actual log BB

R = 0.5298

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Actual log BB

R = 0.7849

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Actual HIA

R = 0.7175

40 50 60 70 80 90 100

HIA Log BBB

efo

reo

pti

miz

ati

on

Aft

ero

pti

miz

ati

on

1.5

1

0.5

0

-0.5

-1

-1.5

-2

Pre

dic

ted

log B

B

110

100

90

80

70

60

50

40

Pre

dic

ted

HIA

130

120

110

100

90

80

70

60

50

40

30

Pre

dic

ted

HIA

40 50 60 70 80 90 100

Figure 4: Visual representation of the correlation between actual

and predicted log BB values using the LOOCV method before and

after elimination of superfluous molecular descriptors.

Prediction of HIA and log BB values

Each PLS regression leads to an equation, generally written as Y = b0 + b1 X1

+ b2 X2 + … + bn Xn, where Y can be HIA or log BB, and X1 … Xn are the

molecular descriptors. The coefficients (b0 … bn) for the two models are

divers, reflecting the difference between HIA and BBB permeation.