Quantitative Structure–Activity Relationships (QSAR)

Together with a selection of other analogs from our laboratories, totaling over 100, and a number from the literature, a database of over 200 artemisinin analogs was assembled. It was of interest not only from the perspective of drug design but ultimately for understanding and eliminating neurotoxicity by design that computer-aided techniques of 3D-quantita- tive structure–activity relationships were employed. In this regard, comparative molecular field analysis (CoMFA)27_{seemed well suited to our problem. The database molecules were}

quite rigid and easily overlapped, and in vitro antimalarial and neurotoxicity data were readily available. While this database could be easily aligned and spreadsheets constructed for analysis, the problem of multiple flexible side chains was not solved. For example, glo- bal minima could be obtained for these analogs and they could be aligned based on the inflexible tetracyclic template (Figure 9.8). The side chains in this instance adopt optimal FIGURE 9.4

Putative molecular mechanism of action of artemisinin.

FIGURE 9.5

Total synthesis of (+)-artemisinin.

©2000 by CRC Press LLC

FIGURE 9.6

Analogs synthesized from the total synthetic manifold.

FIGURE 9.7

Deoxoartemisinin analogs from artemisinin analogs.

FIGURE 9.8

Standard alignment of 202 artemisinin analogs used in the CoMFA model development.

©2000 by CRC Press LLC



orientations in space to minimize steric and electrostatic interactions, however, the core tet- racycle is still apparent from the figure. In addition, the oxygen atoms of the common per- oxy moiety are labeled for reference.

We did conduct CoMFA analysis on this conformational hypothesis, but, also in consider- ation of Meshnick’s hemin hypothesis, we docked and minimized each analog to hemin. The resulting steric interactions between the hemin template and the side chains on the artemisi- nin analogs forced the side chains away from the underside where the peroxide was electro- statically attracted to the iron core of the porphyrin. Once hemin was removed from each aggregate complex, the database was reassembled using the same alignment rule (Figure 9.9). Note that hemin was included in this alignment for visual reference but was not included in the CoMFA analysis. CoMFA analysis was then conducted on this dock hypoth- esis database. It was hoped that a comparison of these models might reveal the validity of the Meshnick hypothesis and allow us to pick the most realistic model before proceeding further. As shown inTable 9.1, the full database of 202 compounds gave a weak correlation coeffi- cient of 0.79, and a somewhat surprising cross validation result with a q2 _{of 0.64. The standard}

error of 0.71 (log units!) supported the moderate quality of this model. When three outliers were removed from the 202 standard database, a slight improvement of the r2 _{value was}

noted. However, the same 199 database with the dock hypothesis led to a substantial improve- ment in r2_{, s, and F values. The q}2 _{was now a more reliable 0.65. This result alone provides}

strong support for the dock hypothesis; however, it was of continuing interest on our part to improve this model further. We noted in our database a difficulty we have had in dealing with the input of bioassay data on approximately 40 racemic analogs. In the past, we have assumed that only one enantiomer was bioactive, but that assumption is debatable based on other stud- ies. Thus, we removed all of the racemates. To our surprise and pleasure, the dock hypothesis, trimmed data set having only optically pure compounds (160 Dock) gave a great enhance- ment in all statistical measures for this model, with r2 _{now approaching unity and s diminish-}

ing below 0.4. Enhancement in magnitude of F-test and cross-validated r2 _(q2_{) are evident.} FIGURE 9.9

Alignment of 202 artemisinin analogs based on docking to hemin.

©2000 by CRC Press LLC

Having established a reliable model and provided support for the mechanistic hypotheses of Meshnick, we set about an inspection of the output derived from these CoMFA models. Color coded contour plots in three dimensions indicate regions in which the 3D-QSAR model predicts either enhanced or decreased potency. Contour plots are constructed from electrostatic data points as well as steric data points. Thus, two sets of field data are avail- able. The red and green contours provide steric SAR and the yellow and blue contours cor- respond to electrostatic SAR. More specifically, red denotes regions where greater steric bulk is predicted to lower activity while imposition into the green regions is predicted to enhance potency. For the electrostatic contours, yellow contours attract electronegative atoms, and blue attracts electron deficient atoms.

While the n = 160 database provided the best statistical model, visually there was little dif- ference between n = 199 and n = 160 dock databases. This is not of major significance as both models contain a large number of molecular examples and, therefore, might be expected to offer similar visual clues after deletion of a relatively small percentage of the data. However, the quality of predictions of these visually comparable models is not the same.

We have shown the contours for standard (Figure 9.10) and dock (Figure 9.11) databases. A comparison of these contour plots is meaningful. Without any information regarding the possible involvement of hemin in the MOA, the standard CoMFA contour plots reveal a large red (sterically forbidden) plate-shaped contour on the peroxy face of artemisinin. In this instance, the side chains project in any favorable minima, including the area occupied by hemin. Likewise, smaller green contours are positioned about C-9 and C-3 and the “equa- tor” of the molecule. In contrast, the dock database shows attenuated red contours in the

TABLE 9.1

Comparative Molecular Field Analyses of Artemisinin Databases — Statistical Results of Partial Least Squares (PLS) Analyses

No. of

Compounds Probe Atom r2 _{s F q}2

Optimum No. of Components 202 2Å/C.3 0.79 0.71 149.62 0.64 5 199 2Å/C.3 0.83 0.65 184.97 0.69 5 199 (Dock) 2Å/C.3 0.93 0.42 244.61 0.65 6 160 2Å/C.3 0.89 0.55 251.11 0.74 5 160 (Dock) 2Å/C.3 0.96 0.35 317.56 0.74 7 FIGURE 9.10

CoMFA contour maps for the standard alignment database (n = 199, 2Å/C.3).

©2000 by CRC Press LLC



steric plots and an enhanced yellow electrostatic contour adjacent to the peroxy group, but missing yellow contours about the lactone and C-13 oxygen atoms. In the steric contours of the dock model, the green contours now predominate, again about the equatorial zone. Both models support Meshnick’s hypothesis in a complementary fashion.