VLS with High Resolution Models Antagonist Agonist Selectivity

In contrast to the dynamic conformational changes involved during in vivo agonist recognition and receptor activation, a single static receptor conformation represents the ligand binding pocket during virtual screening. As a con sequence of this rigid receptor representation, induced fit effects are overlooked, and ligands often dock incorrectly to alternate crystallized or modeled conformations of the same receptor. This phenomenon is referred to as the cross-docking problem. Previous estimates based upon a general protein/ ligand benchmark set concluded that at most, 50% of ligands cross-dock correctly to a single receptor conformation [50]. This problem may be heightened for GPCRs, which exhibit multiple ligand- dependent conformational states. Even provided with a highly accurate crystallographic conformation, the extent to which a single structure will facilitate recognition of other diverse ligand types is unclear. The ligand types selected and the overall VLS yield then depend critically on the nature of the receptor conformation employed for screening.

VLS with Antagonist-Selective Models The crystal structures of bRho with retinal, P2AR with carazolol, P2AR with timolol, P1AR with cyanopindo-lol, and AA2aR with ZM241385 correspond to an inactive receptor conformation adapted for antagonist or inverse agonist binding. Homology models constructed from these templates may then prove more effective at retrieving antagonist than agonist compounds [51]. We have also recently shown that the P2AR/carazolol structure provides high enrichment factors for antagonists, while failing to retrieve agonists [52] , To evaluate docking performance, we constructed a protonated all, atom model from the PDB-deposited coordinates and conducted VLS trials with 1000 (1K) and 14,000 (14K) ligand test sets composed of only small molecule GPCR ligands. For the 1K test set, the PDB - based P2AR model provided an enrichment factor of 50.9 for antagonist compounds and a hit rate of 80% for the top-scoring 1% of the database (10 compounds). Similarly, an enrichment factor of 36.5 and hit rate of 38.6% were obtained for the top-scoring 1% (140 compounds) of the 14K test set (Table 15.2). High yields of known antagonist and inverse agonist compounds in the hit lists indicated that the PDB-based model is able to retrieve a number of structurally distinct small molecules -80% of the known antagonist and inverse agonist compounds contained in the 1K test set were retrieved in the top-scoring 10%. This demonstrates that a single GPCR conformation can provide excellent antagonist enrichment for a diverse set of compounds, even in the absence of empirical restraints. Recently, Topiol et al. also reported qualitatively successful antagonist VLS results for the P2AR crystal structure using a proprietary compound database, though values for enrichment factors or yields were not provided [53]. Our quantitative results are consistent with their findings and are extended to a stringent test set that is restricted to only known GPCR ligands.

VLS with Agonist-Selective Models Previously, agonist selectivity in VLS has been demonstrated for a limited number of bRho-based GPCR homology models. In work by Bissantz and colleagues, manual rotations of TM6 were introduced to approximate receptor activation, as indicated by experimental

TABLE 15.2 Summary of VLS Results for Several 02AR Models

P2AR Model, 14K Test Set

PDB-Based P2AR Model, No ECL2, IK Test Set

P2AR Agonist Model, TM5 1-A Shift,

IK Test Set

P2AR Agonist Model, TM5 1-A Shift, 14K Test Set

P2AR Agonist Model, No ECL2, IK Test Set

Ideal Scores, IK Test Set

Ideal Scores, 14K Test Set

Antagonists

EF/hit rate

6.4/10

50.9/80

36.5/38.6

50.9/80

6.4/10

1.4/1.4

6.4/10

63.6/100

94.6/100

(1%)

EF/hit rate

2.7/4.2

14.6/22.9

10.7/11.3

13.3/20.8

1.3/2.1

1.5/1.6

1.3/2.1

19.9/31.2

20.0/21.1

(5%)

EF/hit rate

3.3/5.3

8.0/12.6

6.4/6.8

6.7/10.5

2.0/3.2

1.4/1.4

1.3/2.1

10.0/15.8

10.0/10.6

(10%)

Yield (10%)

33.3

80

64.2

66.7

20.0

13.5

13.3

100

100

Agonists

EF/hit rate

0/0

0/0

1.6/2.1

0/0

50.9/80

38.4/52.9

31.8/50

63.6/100

72.6/100

(1%)

EF/hit rate

5.3/8.3

0/0

3.5/4.9

0/0

18.6/29.2

12.7/17.4

14.6/23

19.9/31.2

20.0/27.6

(5%)

EF/hit rate

4.0/6.3

2.7/4.2

3.3/4.5

2.0/3.2

9.4/14.7

7.4/10.2

7.4/11.6

10.0/15.8

10.0/13.8

(10%)

Yield (10%)

40

26.7

32.6

20

93.3

74.1

73.3

100

100

Two test sets were evaluated: the IK test set is composed of 954 compounds (15 antagonists/15 agonists), while the 14K test set is composed of 14,006 compounds (148 antagonists/193 agonists). EF, enrichment factor.

Two test sets were evaluated: the IK test set is composed of 954 compounds (15 antagonists/15 agonists), while the 14K test set is composed of 14,006 compounds (148 antagonists/193 agonists). EF, enrichment factor.

evidence from electron paramagnetic resonance (EPR) and fluorescence spec-troscopy studies. This was performed to generate bRho-based homology models of the agonist-bound dopamine D3, 8-opioid, and p2AR receptors [51]. These models were successful in retrieving full agonists from a decoy set when combined with a scoring scheme that produced consensus hit lists from the results of multiple docking and scoring functions. However, the combination of docking and scoring functions used to generate the hit list was optimized separately for each receptor to yield the best hit rates, and a single, general scoring scheme independent of receptor type was not identified. Additionally, a systematic exploration of the optimum TM6 rotation was not conducted.

In our previous work (described above) with TM5 flexible models of p2AR, we found that TM5 shifts toward the ligand binding pocket facilitated agonist docking and binding affinity calculations for a select group of compounds. To investigate whether manipulation of TM5 could generate agonist-bound models suitable for predicting binding geometries and relative binding energies for a much larger variety of ligands, we recently tested a series of TM5 shifted models in a VLS application [52] . In this protocol for agonist-bound model generation, TM5 shifts were systematically introduced in 0.5-A increments, the ligand binding pocket side chains were optimized with the full agonist isoproterenol, and the best receptor conformation was selected by VLS enrichment (Fig. 15.3). This strategy is simply an extension of LGM, where the generation of alternate receptor conformations is modified to incorporate rigid body shifts of the TM domains. While our present strategy focuses on shifts of TM5, future work could incorporate additional shifts, tilts, or rotations of TM5, TM6, or TM7.

A set of seven hypothetical agonist-bound models was evaluated by VLS, and a 1-A shift of TM5 toward the ligand binding pocket provided the best enrichment factors and hit rates for known agonists. Examination of the binding pocket conformation for this model finds that the TM5 shift allows formation of two hydrogen bonds between the para-hydroxyl of the catechol-amine ring, and Ser203 and Ser207 while maintaining the hydrogen bond between the ligand amine and Asp113, similar to the 2-A tilted model obtained by flexible refinement of TM5 as described above (see Section 15.2.3). Screening with this single pocket conformation achieved agonist enrichment factors of 50.9, 18.6, and 9.4 for cutoffs at the top-scoring 1%, 5%, and 10% of database molecules, respectively (Table 15.2). Inspection of the hit list found that the top-scoring 1% of the database contained a diverse set of ligand chemotypes, including both catecholamine (norepinephrine, dobutamine, dopexamine) and non-catecholamine (salmeterol, pirbuterol, sibenadet, formoterol, and fenoterol) ligands, as well as full (norepinephrine) and partial (salbutamol) agonists. The retrieval of a diverse set of agonist compounds with a single receptor conformation validates this approach for constructing other agonist-selective receptor models for use in VLS.

0 0

Post a comment