CNS Physicochemical Property Space Shaped ... - ACS Publications

E-mail: [email protected], [email protected]. Cite this:J. Med. Chem. 2017, 60, 14, 5943-5954. Biography. Zoran Rankovic is Director of ...
3 downloads 0 Views 6MB Size
Download PDF
Perspective pubs.acs.org/jmc

CNS Physicochemical Property Space Shaped by a Diverse Set of Molecules with Experimentally Determined Exposure in the Mouse Brain Miniperspective Zoran Rankovic*,† Downloaded via UNIV OF TOLEDO on September 28, 2018 at 08:31:42 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

Eli Lilly and Company, 893 South Delaware Street, Indianapolis, Indiana 46285, United States S Supporting Information *

ABSTRACT: Understanding the “limits and boundaries” of the central nervous system (CNS) property space is a critical aspect of modern CNS drug design. Medicinal chemists are often guided by the physicochemical properties of marketed CNS drugs, which are heavily biased toward “traditional” aminergic targets and commonly described as small lipophilic amines. This miniperspective describes the statistical analysis of the calculated physicochemical properties for a diverse set of ligands for mostly “nontraditional” CNS targets and classified as either “brain penetrant” or “peripherally restricted” on the basis of the experimental mouse brain exposure. The results suggested that (a) the physicochemical property space conducive to brain exposure is larger than the one defined by the marketed CNS drugs and (b) the most critical brain exposure determinants are descriptors of the molecular size and hydrogen bond capacity. These findings led to a modified version of the CNS MPO scoring algorithm, termed CNS MPO.v2.

■

INTRODUCTION A landmark study by Lipinski et al.1 of the physicochemical properties associated with the oral bioavailability of drugs and advanced clinical candidates, codified in a simple mnemonic known as the “rule of 5” (Ro5), was a major milestone in the evolution of medicinal chemistry from a largely trial-and-error effort to the modern scientific endeavor driven by robust prediction methods. Analogous to the now widely accepted Ro5 in the design of oral drugs, the physicochemical properties needed for brain penetration have been extensively studied by a number of researchers in an attempt to define the attributes of successful CNS drugs and drug candidates using a variety of approaches.2−4 Many of these studies have focused on the physicochemical properties of marketed CNS drugs, which were found to be generally smaller, more lipophilic, with fewer hydrogen bond donors (HBDs) and a lower topological polar surface area (TPSA) compared with oral non-CNS drugs.4 Indeed, the median values derived from an analysis of marketed CNS drugs offer useful guidance when defining a desirable CNS candidate profile: cLogP = 2.8; cLogD = 1.7; HBD = 1; TPSA = 44.8 Å2; pKa = 8.4; RB = 4.5; MW = 305.3 Da.5 Obviously, these numbers are median values, and there are many marketed drugs with properties at either end of the range. More importantly, the current CNS pharmacopeia is heavily dominated by drugs that modulate monoamine GPCRs, transporters, and ion channels. Of 119 drugs included in the analysis described above,5 34 are aminergic GPCR ligands, 24 are ion channel blockers, and 11 are aminergic transporter inhibitors. This situation is likely to change as CNS drug © 2017 American Chemical Society

discovery efforts across the pharmaceutical industry have been rapidly refocusing toward emerging CNS therapeutic areas, such as neurodegeneration and oncology. Consequently, there are ever-increasing numbers of “nontraditional” CNS targets currently being evaluated in the clinic, and many of these targets are characterized by larger and more polar ligands, e.g., proteases, kinases, and phosphodiesterases.6−8 The median values of the CNS drug properties will likely change as these new approaches begin delivering drugs to the market. To elucidate the molecular properties that influence the extent of CNS exposure, an analysis of the physicochemical parameters was carried out for a large and diverse set of compounds with experimentally determined mouse unbound brain and plasma concentrations (Cu,b and Cu,p, respectively), and Kp,uu (Cu,b/Cu,p).9 The properties of this set were also compared to oral drugs marketed for CNS and peripheral (nonCNS) disorders. It is proposed in this study that the property space of the CNS penetrant molecules is slightly larger and more forgiving than the one defined by the currently marketed CNS drugs. This work endeavors to provide guidance for the design of CNS penetrant molecules, which may be particularly helpful to medicinal chemists working on “nontraditional” CNS targets. Received: October 6, 2016 Published: April 7, 2017 5943

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

■

Perspective

RESULTS AND DISCUSSION Data Sets. To examine the effect that molecular properties have on brain exposure, a diverse mix of marketed drugs, literature standards, and compounds from Eli Lilly’s internal projects with experimental brain exposure assessment (BEA) data (2.2 μmol/kg, iv) was split into two sets: a brain penetrant (BP) subset (BEA-BP, compounds with Cu,b > 10 nM and Kp,uu > 0.3; N = 260) and a peripherally restricted (PR) subset (BEA-PR, Cu,b < 10 nM and Kp,uu < 0.3; N = 308). The calculated physicochemical properties of the two sets were analyzed and statistically compared with each other, as well as a collection of CNS drugs (N = 324) and oral non-CNS drugs (N = 845). For the purpose of this study, it was important that the compounds in the brain-penetrant BEA-BP set covered a considerably greater target diversity than the CNS drugs set (Figure 1). For example, 89% of the CNS drugs in

BEA-PR compounds, were assessed and compared using eight calculated fundamental physicochemical properties that are widely recognized across the medicinal chemistry community as key parameters in drug design:2−5 (a) the calculated logarithm of the octanol/water partition coefficient (cLogP); (b) the calculated logarithm of the octanol/water partition coefficient at physiological pH 7.4 (cLogD); (c) the most basic center (pKa); (d) the molecular weight (MW); (e) the number of hydrogen bond donors (HBD); (f) the number of hydrogen bond acceptors (HBA); (g) the topological polar surface area (TPSA); and (h) the number of rotatable bonds (RB), as described in Methods. The range and distribution of these physicochemical properties for each of the four sets of compounds are depicted in Figure 2 (for an explanation of Box Plot graphics see Figure 1 in Supporting Information). The property analysis revealed no statistically significant difference in lipophilicity among the two BEA sets and the CNS drug set (two-sided Student’s t test, p > 0.05), indicating that, in terms of brain exposure, this property is not a major differentiator. The median cLogP value for the BEA-BP set was 3.3 compared to 3.2 for the BEA-PR set (p = 0.1677), whereas the median cLogD value was 3.0 for the BEA-BP set and 2.6 for the BEA-PR set (p = 0.1135). The cLogP values varied from 1.8 (10th percentile) to 4.6 (90th percentile) for the molecules in BEA-BP set and from 1.4 (10th percentile) to 5.6 (90th percentile) in BEA-PR set. Similarly, narrow and overlapping ranges were found for cLogD values that varied from 1.2 (10th percentile) to 4.3 (90th percentile) and from 0.9 (10th percentile) to 4.7 (90th percentile) in the BEA-BP and BEA-PR sets, respectively. As previously reported,4 the CNS set had an overall higher lipophilicity than the non-CNS drugs set. The median cLogP and cLogD values for the CNS drugs were 3.1 and 2.1, respectively, compared to 2.3 and 1.0 for the non-CNS drug set, respectively (both values were significantly different between the two drug sets; p < 0.0001). The non-CNS drugs showed a broader range of cLogP values, from −1.0 (10th percentile) to 5.1 (90th percentile), and cLogD values, from −2.3 (10th percentile) to 4.2 (90th percentile), compared to the CNS drugs with cLogP values ranging from 1.1 (10th percentile) to 4.8 (90th percentile) and cLogD values from −0.2 (10th percentile) to 3.8 (90th percentile). The one log unit difference between the median cLogP and cLogD values for the CNS drugs reflects the fact that the majority of CNS drugs contain a basic group (69% in this set). This is a consequence of the significant bias in this set toward the structures designed to modulate targets from aminergic GPCR (48%), aminergic transporter (15%), and ion channel (28%) gene families, all of which are characterized by the presence of a basic amino group as one of the key pharmacophore elements. Indeed, the CNS drug set showed a significantly higher median pKa value (8.2) with a range from −1.7 to 9.8 (10th and 90th percentile), compared to non-CNS drugs with a median pKa of 2.7 (p < 0.0001) and a broader range of values, from −4.4 to 9.5 (10th and 90th percentile). In contrast, no significant difference was observed between the two BEA sets, with a median pKa value of 5.9 for the BEA-BP set and 5.0 for the BEA-PR set (p = 0.3547). Interestingly, when nonbasic molecules (pKa < 7) were excluded from the analysis, a reversed picture had emerged with a statistically significant but small difference between the median pKa values for the basic subsets of the BEA-BP and BEA-PR sets (8.3 and 8.7, respectively; p = 0.0024) and no difference between the basic subsets of CNS drugs and oral

Figure 1. Target family distribution: (a) CNS drug set; (b) BEA-BP set (experimental Cu,b > 10 nM and Kp,uu > 0.3); (c) structure clustering analysis of the two sets separate and combined.

this set modulate targets from only three gene families, i.e., aminergic GPCRs (46%), aminergic transporters (15%), and ion channels (28%). In contrast, these “traditional” CNS target families are considerably less represented in the BEA-BP set, which also includes a diverse range of more “contemporary” targets (92 in total) from gene families, such as peptidicGPCRs (7%), kinase (14%), protease (5%), and other enzymes (22%). The two sets of structures also show extensive internal diversity and cover a remarkably different structural space with very little overlap. For example, sphere exclusion analysis10 revealed 168 clusters in the CNS drug set of 324 structures. The largest cluster contained no more than 20 members, and there were 123 singletons (Figure 1c). The BEA-BP set (260 structures) showed similar internal diversity and had 163 clusters. The largest had no more than 11 members, and there was a total of 122 singletons. Furthermore, the two sets when combined (N = 584) showed very little overlap, with the total number of clusters and singletons almost the same as the sum of the corresponding clusters in the individual sets (Table 1c). In other words, the two sets of structures could be described as very distant and large islands. Property Analysis. The four sets of molecular structures described above, CNS drugs, non-CNS drugs, BEA-BP, and 5944

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

Perspective

Figure 2. Physicochemical property range and the distribution across the four sets of structures: the compounds classified on the basis of experimental brain exposure assessment (BEA) data as brain penetrant (BEA-BP, green) or peripherally restricted (BEA-PR, red) compounds, CNS drugs (blue), and oral non-CNS drugs (amber): P10, 10th percentile; Q1, 25th percentile; Q3, 75th percentile; P90, 90th percentile.

12 compounds (4.6%) in the BEA-BP set that contained a carboxylic group or related bioisostere likely to be negatively charged at a physiological pH. Similarly, of 324 structures in our CNS set, there were only 5 drugs (1.5%) that contained a carboxylic acid group. It is important to note that all carboxylate-containing compounds from both sets are, in fact, zwitterions, having at least one basic amino group. Several αamino acid-like compounds are known to be actively transported across the BBB, as exemplified by gabapentin or L-DOPA, which are transported via the L-type amino acid transporter 1 (LAT-1).15 Others are believed to cross the BBB as overall neutral molecules by a passive diffusion mechanism.16,17 However, it should be noted that active transport of α-amino acid-like compounds is very rare and not well

non-CNS drugs (Figure 3). In any case, the presence of a basic group is clearly not a prerequisite for brain penetrant molecules. It also appears that, in regard to CNS exposure, a basic pKa, as cLogP and cLogD, is not an important differentiating parameter except at high values when a vast majority of the molecules are protonated at the physiological pH; e.g., pKa > 10. In fact, considering the safety liabilities associated with lipophilic basic amines, including hERG block11 and phospholipidosis,12 as well as the risk of P-gp recognition,13 one could see basic groups as an undesirable feature in the CNS drug design that should be avoided whenever possible. In contrast to basic amines, carboxylic acids are generally associated with poor brain exposure due to a combination of multiple factors, including high plasma protein binding, poor passive permeability, and P-gp recognition.14 There were only 5945

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

Perspective

ordered region of the lipid bilayer obeys an extended power law decay, indicating that a larger portion of free volume is accessible to smaller penetrant molecules.20 Interestingly, in our study, the median MW of the brain-penetrant BEA-BP set (389.0) was not significantly different from that of non-CNS drugs (354.4; p = 0.6593) and was considerably higher than the median MW value of the CNS drugs (301.0; p < 0.0001). This finding suggested that although the MW is an important differentiating parameter between brain-penetrant and peripherally restricted molecules, in the context of brain exposure, the optimal MW range may be considerably higher than the one defined by the marketed CNS drugs. The hydrogen bond capacity, as described by HBD and HBA counts, is one of the most critical considerations in drug design. This is true especially for the HBD count, whose mean value for oral drugs launched over the past century remained unchanged, indicating the importance of this property for oral drugs.19 In the context of CNS drug design, the HBD count is arguably the most critical physicochemical parameter.2−4 In this analysis, the CNS drugs and the compounds in the brainpenetrant BEA-BP set had a minimal number of HBDs, with a median value of 1 and a narrow range of 0−2 (10th to 90th percentile). In contrast, the BEA-PR set and the oral non-CNS drugs had a statistically higher median HBD value (2; p < 0.0001), suggesting that HBD is an important CNS differentiating parameter (Figure 2). The bar chart in Figure 4 depicts the considerable impact that an increasing HBD count has on CNS exposure and shows that compounds with HBD > 2 are highly unlikely to be brain penetrant. The significant impact of the HBD count on CNS exposure is most likely the consequence of an additive combination of two detrimental effects generally associated with hydrogen bonding, namely, poor passive permeability and increased risk of interactions with efflux transporters. Studies using artificial membranes (PAMPA) suggested that an increased HBD count is associated with lower passive permeability, a phenomenon attributed to the desolvation of the associated hydrogen-bound water molecules, which is required for membrane permeability.22 This effect is often further exaggerated by biological membranes, where an increased H-bonding potential also increases the risk of P-gp recognition.23 The comparison of HBA counts for the four compound sets showed a statistically significant difference between the BEA-BP set (median HBA = 6) and the BEA-PR (HBA = 7) set (p < 0.0001), as well as between the CNS (median HBA = 4) and non-CNS (median HBA = 6) drugs (p < 0.0001), suggesting that this is indeed an important CNS differentiating parameter. It is also important to note that the median HBA count for the brain-penetrant BEABP set (6) was significantly higher than for the CNS drug set (p < 0.0001), which indicated that the HBA range consistent with

Figure 3. pKa range and the distribution across the basic (pKa > 7) subsets of the four sets BEA-BP (green), BEA-PR (red), CNS drugs (blue), and oral non-CNS drugs (amber): P10, 10th percentile; Q1, 25th percentile; Q3, 75th percentile; P90:, 90th percentile.

understood phenomenon and consequently difficult to translate into a rational CNS drug design. The MW comparison showed that CNS drugs are smaller than oral non-CNS drugs with median values of 301.0 and 354.4 Da, respectively (p < 0.0001). A very similar and statistically significant difference (p < 0.0001) was found between BEA sets, with the median MW value of 389.0 Da for the BEA-BP set and 446.0 Da for the BEA-PR set. The impact that increasing MW has on CNS exposure is depicted in the bar chart in Figure 4, which shows that compounds with MW > 400 Da are significantly less likely to be brain penetrant. A similar effect was observed with properties such as molecular volume (MV) and number of heavy atoms (HA), further emphasizing the importance of molecular size as a CNS discriminating parameter (Supporting Information). These findings are consistent with previous literature reports. For example, an examination of a large set of compounds with MW < 600 Da showed a strong influence of MW on P-gp risk and passive permeability.5,18 In this study, only 14.9% of compounds with MW between 250 and 300 Da were found to be P-gp substrates, whereas 72.2% of compounds in the 500−550 Da bin were characterized as P-gp substrates. In a similar fashion, the portion of compounds with favorable passive permeability was found to be the largest within a low MW subset. The size dependency of cell membrane permeation can be rationalized by a free-volume theory for diffusion in liquid or hard spheres developed by Cohen and Turnbull in the late 1950s, which suggests that translational diffusion of solute occurs when statistical redistribution of free volume opens up a void of a critical size in the immediate vicinity of the solute.19 The size distribution of the free volume pockets in the more

Figure 4. Distribution of MW and HBD count across the combined BEA sets (N = 584), BEA-BP (green) and BEA-PR (red). 5946

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

Perspective

the CNS exposure was greater than the one defined by the currently marketed CNS drugs (median HBA = 4); (Figure 2). It is probably not surprising that the TPSA, another measure of the hydrogen-bonding potential, displayed across the four compound sets a very similar pattern to the HBD and HBA count (Figure 2). The median TPSA of the BEA-BP set (67.4 Å2) was significantly lower than the median value of the BEAPR set (86.4 Å2; p < 0.0001). The median TPSA value for the CNS drugs (40.5 Å2) was lower compared to the non-CNS drug set (83.8 Å2; p < 0.0001). Again, similar to the HBA and HBA count, the median TPSA of the brain-penetrant BEA-BP set was statistically higher compared to the CNS drugs (67.4 and 40.5 Å2, respectively; p < 0.0001), suggesting that a broader TPSA range is available to medicinal chemists designing brainpenetrant molecules compared to the range that one would consider solely on the basis of currently marketed CNS drugs. Just as with the presence of a basic group, the comparatively low median HBA and TPSA values of the CNS drugs set reflect the common pharmacophore of the ligands designed for aminergic GPCRs, transporters, and ion channels that are extensively represented in the CNS drug set (89%). It is worth noting at this point that the TPSA and/or HBD count reduction is one of the most frequently reported and successful strategies used in the optimization of brain exposure.2,3 Finally, the median RB count comparison suggested that CNS drug molecules are slightly more rigid compared to oral non-CNS drugs (4 and 5, respectively; p < 0.0001), which may be related to the overall smaller size of the CNS drugs. No difference was observed between the median RB counts of the BEA-BP and BEA-PR sets (Figure 2). The results of our analysis suggest that in the context of brain exposure, molecular size and hydrogen bond capacity are the most critical differentiating parameters. Because MW can be considered a surrogate for other physicochemical properties,32 including the hydrogen bond potential, the combined BEA sets were dissected to establish if the effects of these two parameters on brain exposure are independent from each other (Figure 5a). Although the sample is relatively small (especially at the lower margins), one may still conclude, if focused on the core and most densely populated space (TPSA of 60−80 Å2 and MW of 300−500 Da), that as the values of each of the two properties increase, the likelihood of brain penetration decreases. For example, in the 80 Å2 TPSA bin, 4 of 5 compounds in the MW 200 Da cluster are brain penetrants, whereas in the MW 500 Da cluster, only 2 of 19 compounds are brain penetrants. Similarly, in the MW = 400 Da bin, 6 of 7 compounds in the 40 Å2 TPSA cluster are brain penetrants, while the same profile was shared by only 6 of 23 compounds in the TPSA > 120 Å2 cluster. Analogous patterns were found in relation to the HBD count and MW (Figure 5b). These trends in the BEA data set suggested that the MW and TPSA/HBD effects on the brain exposure are independent of each other. This conclusion was further asserted by recursive partitioning analysis, which is a statistical method for multivariable analysis that creates a decision tree that strives to correctly classify members of the population (in this case the BEA set) by splitting it into subpopulations (BEA-BP or BEA-PR) based on several dichotomous independent variables.24 This analysis showed that the most important differentiation parameters between the BEA-BA and BEA-PR sets are TPSA (65%), MW (21%), and HBD (13%; Figure 3 in Supporting Information).

Figure 5. CNS property space, showing distribution of the key CNS differentiating parameters across the combined BEA-BP (green) and BEA-PR (red) sets (trellis view): (a) TPSA and MW; (b) HBD count and MW.

These findings are in agreement with the good separation in the MW-TPSA space that was found between the analyzed sets. For example, only 19% of the compounds from the BEA-PR set fell within the CNS property space as defined by the Q3 (75th percentile) values for MW (430 Da) and TPSA (85 Å2), which contrasts with 63% of the brain-penetrant BEA-BP set and 87% of the CNS drugs (Figure 6). Multiparameter Evaluation. A good understanding of the general risks and benefits associated with individual physicochemical parameters and their effect on molecular pharmacokinetic and pharmacodynamic properties is a critical aspect of modern CNS drug design. Often, however, no single physicochemical parameter can be used to fully explain or predict the intricate pharmacokinetic properties related to brain exposure; rather, more complex multivariate models that integrate multiple descriptors simultaneously are required. Wager and colleagues at Pfizer have recently reported a multiparameter scoring approach designed specifically for CNS drug discovery, termed CNS multiparameter optimization (CNS MPO).25 To avoid hard cutoffs, the algorithm generates desirability scores T0 (0−1) for six physicochemical properties (cLogP, cLogD, MW, TPSA, HBD count, and pKa of the most basic center), which are then summed to produce an overall desirability score (CNS MPO = ∑T0(cLogP, cLogD, MW, TPSA, HBD, pKa), ranging from 0 (undesirable) to 6 (highly desirable). The authors showed that the majority of the CNS drugs in their set (74%, N = 119) have CNS MPO scores of ≥4. The CNS MPO values for the four sets were calculated utilizing the same algorithm as described by Wager et al.25 (Figure 7a). As expected, the analysis showed that the median CNS MPO score of the brain-penetrant BEA-BP set was statistically higher than the BEA-PR set: 4.5 and 3.9, 5947

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

Perspective

Figure 6. Distribution across the MW and TPSA property space of (a) CNS drugs, (b) BEA-BP set, and (c) the BEA-PR set. Horizontal line marks MW = 430 Da. Vertical line marks TPSA = 85 Å2.

Moreover, only 9% of CNS drugs in our set were found to have the MPOMW,HBD score of 10 are considerably less likely to cross the BBB by a passive diffusion mechanism.2,3 It is also well documented that these attributes strongly influence compound developability properties, including solubility, metabolism, and toxicology, and therefore need to be closely monitored and controlled during lead optimization process.21 Consequently, the author sought retaining these parameters in the CNS MPO algorithm while still modifying it in a way that reflects the conclusions from the BEA data analysis. This approach led to a removal of the cLogD score from the original algorithm to simultaneously reduce the contribution of lipophilic parameters (as well as to eliminate an apparent overlap with already featured cLogP and pKa) and to increase the weight of the HBD score. Changes to the original preferred parameter ranges were also tried but did not produce improvements. Thus, the modified algorithm was based on already described desirability scores for five fundamental physicochemical parameters, with the HBD score multiplied by 2: CNS MPO.v2 = ∑T0(cLogP, MW, TPSA, pKa, 2 HBD). Gratifyingly, just as for the simplified algorithms, the median CNS MPO.v2 scores were found to be significantly different between the BEA-BP and BEA-PR sets (4.9 and 3.9, respectively; p < 0.0001), as well as between the CNS and non-CNS drugs sets (4.7 and 4.2, respectively; p < 0.0001), as depicted in Figure 7d. It was also interesting to find that there was no significant difference between the BEA-BP and CNS drugs sets (4.9 and 4.7, respectively) and BEA-PR and nonCNS drug sets (3.9 and 4.2, respectively; p = 0.7602). The improved performance of the CNS MPO.v2 algorithm was driven primarily by the positive shift in the BEA-BP and CNS drugs scores, as shown in Figure 8. The average difference between the two scores (CNS MPO.v2 − CNS MPO) was positive for the BEA-BP (0.3) and CNS drugs (0.1), whereas the BEA-PR set was neutral (0) and the non-CNS set was negative (−0.2). Ultimately, the modified algorithm showed improved separation between the brain-penetrant and peripherally restricted compounds. For example, 83% of compounds in

respectively (p < 0.0001). Interestingly, the CNS MPO median values for the CNS drugs and non-CNS drugs were not significantly different (4.6 and 4.4, respectively). It is also worth noting that the median values as well as the range (10th to 90th percentile) of the CNS MPO scores for the CNS drugs overlapped almost perfectly with the BEA-BP set, whereas the values for the non-CNS drugs and BEA-PR sets were significantly different, p < 0.0001 (Figure 7a). The author was also interested in examining how the CNS MPO scoring may change if the algorithm was modified to reflect the findings from the property analysis of the BEA data sets. As discussed above, the results of the BEA data analysis suggested that, in the context of brain exposure, the MW, TPSA, and HBD count were the most critical physicochemical properties, while cLogP, cLogD, and pKa were important but not CNS discriminating parameters. In line with these findings, one of the initially explored CNS MPO algorithm modifications was the exclusion of the “CNS nondiscriminating” cLogP, cLogD, and pKa parameters from the algorithm. The resulting sum of the desirability scores for the remaining MW, HBD, and TPSA parameters was then multiplied by 2 to produce MPOMW,HBD,TPSA scores within a suitable range (0−6) and to enable a direct comparison with the original CNS MPO algorithm (Figure 7b). Indeed, it was satisfying to find that the simplified algorithm retained the ability to distinguish between the two BEA sets, with median MPOMW,HBD,TPSA scores for the BEA-BP and BEAPR sets of 4.9 and 3.7, respectively (p < 0.0001). The greater difference between the median MPOMW,HBD,TPSA scores compared to the median CNS MPO values for the two BEA sets (1.2 and 0.6 units, respectively) is translated into an improved separation between the brain-penetrant and peripherally restricted compounds in the combined BEA set. For example, 81% of compounds with a MPOMW,HBD,TPSA score of >5 were brain penetrant (Figure 7f), whereas only 18% were brain penetrant if the score was 5 were brain penetrant (Figure 7g), whereas 32% were brain penetrant if the score was 5 were brain penetrant (Figure 7h), whereas only 20% were brain penetrant if the score was 5 were brain penetrant (Figure 7e), while 33% were brain penetrant if the score was 10 nM and Kp,uu > 0.3; N = 260) or as peripherally restricted (BEA-PR; Cu,b < 10 nM and Kp,uu < 0.3; N = 308) molecules. In addition, the properties of the two BEA sets were also contrasted with the properties of large and carefully curated sets of CNS (N = 324) and nonCNS (N = 845) drugs. Our findings and potential property ranges to consider when designing brain-penetrant molecules are summarized in Table 1. 5950

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

Perspective

Table 1. Physicochemical Properties of Brain-Penetrant BEA-BP Set, Peripherally Restricted BEA-PR Set, and CNS Drugsa BEA-BP

a

BEA-PR

CNS drugs

property

median

P10

P90

median

P10

P90

median

P10

P90

cLogP cLogD pKa HBD HBA TPSA (Å2) MW (Da) RB CNS MPO CNS MPO.v2

3.3 3 5.9 1* 6* 67.4* 389.0* 5 4.5* 4.9*

1.8 1.2 −0.8 0 3 32.8 300.0 2 3.3 3.8

4.6 4.3 8.8 2 8 93.5 474.4 7 5.5 5.7

3.2 2.6 5 2* 7* 86.4* 446.0* 5 3.9* 3.9*

1.4 0.9 −1.0 1 5 64.4 373.8 4 2.8 2.7

5.6 4.7 9.2 3 9 115.2 535.6 9 5.1 4.9

3.1 1.9 7.9 1 4 40.5 301.0 4 4.6 4.7

1.1 −0.2 −1.7 0 2 10.4 207.8 1 3.1 3.4

4.8 3.8 9.7 2 6 77.9 412.9 7 5.7 5.7

P1 = 10th percentile; P90 = 90th percentile. *Statistically significant (p < 0.05) difference between the median values for the two BEA sets.

In this context, it is somewhat comforting to find that although MW proved an important differentiator between the brain-penetrant and peripherally restricted sets (median MW = 389 and 446 Da, respectively; p < 0.0001), its upper margin may not be as low as suggested by the median MW of the CNS drugs (301 Da). In addition to molecular size, the hydrogen bond capacity descriptors were also found to be critical CNS differentiators. The median values of TPSA, HBD, and HBA counts were all determined to be significantly different (p < 0.0001) between the BEA-BP and BEA-PR sets (Table 1). Similar to the MW, the analysis of the BEA-BP set points toward significantly higher upper margins of the HBA and TPSA optimal ranges compared to the ones implied by the properties of marketed CNS drugs. Interestingly, a very similar divergence with respect to the current CNS drugs property space can be observed in an analysis of recent CNS clinical candidates at Pfizer.26 As for the BEA-BP set (median MW = 389 Da, and TPSA = 67.4 Å2), the median MW and TPSA values of Pfizer’s CNS candidates were considerably higher (397.9 Da and 80.8 Å2), and the median HBD count was identical to the CNS drugs: median MW = 301 Da, TPSA = 40.5, HBD = 1. In a similar fashion to the BEA-BP compounds, the median cLogD value of the Pfizer’s candidates was also found to be higher (3.0) when compared to the CNS drugs, a consequence of a lower median pKa (4.1). The only difference from the trend observed in the BEA-BP set was the reduced median cLogP value (2.2) of the Pfizer’s candidates set, which the authors argued was a result of an increased focus on control of this property during the optimization process and successful implementation of the CNS MPO principles in drug design. Indeed, the CNS property space is likely to continue evolving in this direction, driven by the contemporary medicinal chemistry practices (e.g., control cLogP and pKa) and challenges of the emerging targets that are often characterized by larger and more polar binding sites. One of the outcomes of this study are potential upper limit guidelines for CNS-critical physicochemical parameters that one could define on the basis of their 90th percentile values (P90) for the BEA-BP set (Table 1), as follows: MW ≤ 470 Da; TPSA ≤ 90 Å; and HBD ≤ 2. These can be useful guidelines to consider when designing brain penetrant molecules for CNS drug discovery programs, especially for challenging targets. However, it is important to realize that although this study reveals a somewhat larger CNS property space compared to the one outlined by the marketed CNS drugs, it also clearly shows that molecules in a lower property space have greater chance of success. For example, in our BEA set only 33% of compounds

with one H-bond donor and 25% with two H-bond donors are brain penetrant if MW > 400 Da (Figure 5). In contrast, among compounds with MW < 400 Da a total of 78% with one Hbond donor and 53% with two H-bond donors are brain penetrant. Consequently, the above findings are hoped to offer encouragement and guidance for drug discovery efforts toward the more challenging CNS targets but not as an excuse for property inflation38 and lessening the vigor in pursuit of high quality molecules. Judicious selection of lead compounds and an effective application of multivariate optimization principles, including CNS MPO and efficiency metrics such as ligand efficiency (LE)39 and lipophilicity ligand efficiency (LLE),31 will remain critical to the success of the CNS drug discovery programs. Considering the impact of molecular size on the brain exposure, one may find that monitoring and optimizing LE is especially important in the CNS drug discovery. It is also important to note that potential species difference is clearly the main limitation when applying learnings from the mouse BEA study into drug design. The extent to which rodent CNS pharmacokinetics parameters are translatable to human is still a matter of debate,40 mainly due to the limited availability of human data. Studies reported in the literature suggest that for nontransporter substrates Kp,uu is generally preserved across species; i.e., rodent Kp,uu is widely used to predict human Kp,uu.41 In addition, comparison of human and mouse in vitro P-gp efflux ratio suggests that species difference is possible but a relatively rare occurrence.42 Compounds found to be P-gp substrates in rodents are likely to also be substrates in higher species.43 However, recent quantitative proteomics studies uncovered a significant interspecies difference in the abundance of BBB transporters, e.g., P-gp being more abundant at the mouse BBB, and BCRP more abundant at the human BBB.44 In any case, while the discussion on impact of species difference on the brain exposure continues, CNS drug discovery teams have little option but to continue using rodent data to progress their projects into the clinic.2,3 Importantly, the notable examples of successful translation of rodent CNS pharmacokinetics and pharmacodynamics to clinic relayed primarily on drug unbound concentrations.45,46 In summary, the results of our statistical analyses of the experimental mouse brain exposure data indicated that the physicochemical property space conducive to CNS exposure is larger and slightly more forgiving than the one delineated by the currently marketed CNS drugs. Furthermore, while controlling molecular lipophilicity and pKa is important to minimize pharmacokinetic and toxicological risks, the molecular size and hydrogen bond capacity expressed by the MW and 5951

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

Perspective

method developed by Ertl.54 The physicochemical data of the different compounds sets were presented and compared using box plots. Statistical analyses were carried out by JMP 12.1.0 and visualized with TIBCO Spotfire 6.5.3. Recursive partitioning analysis was carried out by JMP 12.1.0. Differences between the four sets of structures were statistically evaluated by a Tukey Cramer multiple comparison test using the ANOVA option and the comparison circles in SpotFire (TIBCO Spotfire 6.5.3) and, where stated, by Student’s t test for continuous variables (JMP 12.1.0). Statistical significance is defined by a p-value of less than 0.05.

HBD/TPSA parameters are the most critical CNS-differentiating properties. These findings led to the design of the CNS MPO.v2 algorithm, which offered an improved separation between the brain penetrant and peripherally restricted molecules in Eli Lilly’s database.

■

METHODS

Data Sets. The primary objective of data collection was compiling the largest possible data sets while ensuring a high level of data quality. The combined data set analyzed in this study contains experimental and calculated data for 1737 compounds: brain penetrant BEA-BP set (N = 260); peripherally restricted BEA-PR set (N = 308); CNS drugs (N = 324); and oral non-CNS drugs (N = 845). Brain Exposure Assessment (BEA) Data Set. A total of 568 compounds with experimentally determined mouse CNS pharmacokinetic (PK) parameters were extracted from Ely Lilly’s corporate database. The PK data were generated following the same experimental procedure. Briefly, the concentration unbound in plasma (Cu,p) and brain (Cu,b) values were obtained from total concentrations measured in the mouse (brain and plasma harvested 5 min after a single iv dose of 2.17 μmol/kg) and fraction unbound in plasma ( f u,p) and brain ( f u,b) obtained from the LC/MS analysis of the brain homogenate, as previously reported.46 In order to obtain an accurate Kp,uu value at the measurement time point, the system should be at steady state. Hence, classical Kp,uu measurements typically involve constant drug infusion over several hours before plasma and brain tissue are collected. This is a laborious and low throughput assay; very few compounds have Kp,uu values determined by his method. Consequently, an assumption made in this study is that the equilibrium is achieved by the 5 min time point. Indeed, it has been previously shown that for 90% of compounds in the two BEA sets the Kp,uu values obtained at 5 and 60 min were within the experimental variability.47 The compounds were classified on the basis of their Cu,b levels as low (C u,b < 10 nM) and high (C u,b > 10 nM). Another pharmacokinetics parameter used for the classification in this analysis was unbound brain-to-plasma ratio Kp,uu (Kp,uu = Cu,b/Cu,p), classified as high when Kp,uu > 0.3 and low when Kp,uu < 0.3. While the 0.3 value is a universally accepted Kp,uu cutoff, the relatively high 10 nM cutoff value for Cu,b adopted as an additional classification parameter in this analysis is designed to exclude a small subset of compounds with “gray zone” profiles, e.g., high Kp,uu but low Cu,b or low Kp,uu but high Cu,b. This more stringent classification provides greater separation between the two BEA sets and therefore a better chance of identifying the most critical physicochemical determinates of the brain exposure. On the basis of these experimental data, the BEA set was split into two subsets: the BEA-BP set, consisting of compounds with Cu,b > 10 nM and Kp,uu > 0.3 values (N = 260), and the BEA-PR set made of compounds with Cu,b < 10 nM and Kp,uu < 0.3 (N = 308). The structural diversity of the CNS drugs and BEA-BP sets was examined and compared using a leader clustering method based on a fixed cluster radius (20%).10 Oral Non-CNS Drugs and CNS Drugs Sets. Initial collection of approved drugs was extracted from Investigational Drugs Database (IDdb),48 Pharmaprojects,49 and ChemBl50 databases. Using IDdb and Orange Book51 information and occasional SciFinder52 literature searches, a total of 1169 unique single agent oral drugs were identified, of which 324 were assign to the “CNS drugs” and 845 to “non-CNS drugs” category. The “CNS drugs” assignment was manual and particularly rigorous. For example, drugs for the allergy, pain, and muscle relaxation indications, which may also function through peripherally expressed targets, were assigned “CNS drugs” category only if there was a clear evidence of the CNS exposure. CNS drugs that are known to be actively transported into the brain, such as LDOPA and gabapentin,15 were excluded. Computational Analysis and Physicochemical Properties. Calculated physicochemical properties were obtained using standard commercial packages: ChemAxon53 for cLogP, cLogD at pH7.4, and the most basic pKa calculations. TPSA was calculated using the

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jmedchem.6b01469. Box Plot graphics explained (Figure 1); calculated physicochemical properties for BEA-BP, BEA-PR, CNS drugs, and non-CNS drugs sets (Table 1); distribution of the molecular volume (MV) and heavy atom (HA) count across the two combined BEA sets (Figure 2); recursive partitioning analysis of the physicochemical properties for the two combined BEA sets (Figure 3); Chemaxon, BioBite, and Prism calculated CNS MPO scores (Figure 4); SMILES for structures in the CNS drugs set (Table 2); SMILES for structures in the non-CNS drugs set (Table 3) (PDF)

■

AUTHOR INFORMATION

Corresponding Author

*Phone: +1 (901) 5953678. E-mail: [email protected], [email protected]. ORCID

Zoran Rankovic: 0000-0001-6866-4290 Present Address †

Chemical Biology and Therapeutics, St. Jude Children’s Research Hospital, 262 Danny Thomas Place, Memphis, TN 38105, U.S. Notes

The author declares no competing financial interest. Biography Zoran Rankovic is Director of CBT Chemistry Centers at St. Jude Children’s Research Hospital in Memphis, TN. Before joining St. Jude in late 2016, he was a research fellow at Eli Lilly in Indianapolis, IN, and medicinal chemistry director at Merck, Schering-Plough, and Organon UK. He started his industrial career at Organon in 1995, the same year he earned his Ph.D. in Organic Chemistry from the University of Leeds (U.K.), under the guidance of Professor Ronald Grigg. During his career Zoran has been fortunate to be able to contribute to and lead teams that delivered multiple clinical candidates for a range of CNS indications, including neurooncology, neurodegeneration, psychiatry, and pain.

■

ACKNOWLEDGMENTS The author thanks Richard Morphy, Michal Vieth, and Ian Watson for helpful conversations and their feedback on this manuscript, and Jibo Wang for providing an easy to use high throughput CNS MPO score calculation tool.

■

ABBREVIATIONS USED BBB, blood−brain barrier; CNS, central nervous system; GPCR, G-protein-coupled receptor; hERG, human ether-a5952

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

Perspective

spirotetracyclic zwitterionic dual H1/5-HT2A receptor antagonists for the treatment of sleep disorders. J. Med. Chem. 2010, 53, 7778−7795. (17) Mihalic, J. T.; Fan, F.; Chen, X.; Chen, X.; Fu, Y.; Motani, A.; Liang, L.; Lindstrom, M.; Tang, L.; Chen, L.-L.; Jaen, J.; Dai, K.; Li, L. Discovery of a novel melanin concentrating hormone receptor 1 (MCHR1) antagonist with reduced hERG inhibition. Bioorg. Med. Chem. Lett. 2012, 22, 3781−3785. (18) Pettersson, M.; Hou, X.; Kuhn, M.; Wager, T. T.; Kauffman, G. W.; Verhoest, P. R. Quantitative assessment of the impact of fluorine substitution on p-glycoprotein (P-gp) mediated efflux, permeability, lipophilicity, and metabolic stability. J. Med. Chem. 2016, 59, 5284− 5296. (19) Cohen, M. H.; Turnbull, D. Molecular transport in liquids and glasses. J. Chem. Phys. 1959, 31, 1164−1168. (20) Marrink, S. J.; Berendsen, H. J. C. Permeation process of small molecules across lipid membranes studied by molecular dynamics simulations. J. Phys. Chem. 1996, 100, 16729−16738. (21) Leeson, P. D.; Davis, A. M. Time-related differences in the physical property profiles of oral drugs. J. Med. Chem. 2004, 47, 6338− 6348. (22) Veber, D. F.; Johnson, S. R.; Cheng, H.-Y.; Smith, B. R.; Ward, K. W.; Kopple, K. D. Molecular properties that influence the oral bioavailability of drug candidates. J. Med. Chem. 2002, 45, 2615−2623. (23) Hitchcock, S. A. Structural modifications that alter the Pglycoprotein efflux properties of compounds. J. Med. Chem. 2012, 55, 4877−4895. (24) Breiman, Leo Classification and Regression Trees; Chapman & Hall/CRC: Boca Raton, FL, 1984. (25) Wager, T. T.; Hou, X.; Verhoest, P. R.; Villalobos, A. Moving beyond rules: the development of a central nervous system multiparameter optimization (CNS MPO) approach to enable alignment of druglike properties. ACS Chem. Neurosci. 2010, 1, 435−439. (26) Wager, T. T.; Hou, X.; Verhoest, P. R.; Villalobos, A. Central nervous system multiparameter optimization desirability: Application in drug discovery. ACS Chem. Neurosci. 2016, 7, 767−775. (27) Gunaydin, H. Probabilistic approach to generating MPOs and its application as a scoring function for CNS drugs. ACS Med. Chem. Lett. 2016, 7, 89−93. (28) Neil-Dwyer, G.; Bartlett, J.; McAinsh, J.; Cruickshank, J. M. β3Adrenoreceptor blockers and the blood-brain barrier. Br. J. clin. Pharmac. 1981, 11, 549−553. (29) Kartalija, M.; Kaye, K.; Tureen, J. H.; Liu, Q.; Tauber, M. G.; Elliott, B. R.; Sande, M. E. Treatment of experimental cryptococcal meningitis with fluconazole: Impact of dose and addition of flucytosine on mycologic and pathophysiologic outcome. J. Infect. Dis. 1996, 173, 1216−1221. (30) Minedermann, T.; Zimmerli, W.; Gratzil, O. Rifampicin concentrations in various compartments of the human brain: a novel method for determining drug levels in the cerebral extracellular space. Antimicrob. Agents Chemother. 1998, 42, 2626−2629. (31) Leeson, P. D.; Springthorpe, B. The influence of drug-like concepts on decision-making in medicinal chemistry. Nat. Rev. Drug Discovery 2007, 6, 881−890. (32) Reichel, A. Addressing central nervous system (CNS) penetration in drug discovery: Basics and implications of the evolving new concept. Chem. Biodiversity 2009, 6, 2030−2049. (33) Smith, D. A.; Di, L.; Kerns, E. H. The effect of plasma protein binding on in vivo efficacy: misconceptions in drug discovery. Nat. Rev. Drug Discovery 2010, 9, 929−939. (34) Van de Waterbeemd, H.; Camenisch, G.; Folkers, G.; Chretien, J. R.; Raevsky, O. A. Estimation of blood-brain barrier crossing of drugs using molecular size and shape, and H-bonding descriptors. J. Drug Targeting 1998, 6, 151−165. (35) Leo, A.; Hansch, C.; Elkins, D. Partition coefficients and their uses. Chem. Rev. 1971, 71, 525−616. (36) Hansch, C.; Steward, A. R.; Anderson, S. M.; Bentley, D. L. Parabolic dependence of drug action upon lipophilic character as revealed by a study of hypnotics. J. Med. Chem. 1968, 11, 1−11.

go-go-related gene; LAT1, L-type amino acid transporter 1; LE, ligand efficiency; MPO, multiparameter optimization; MV, molecular volume; P-gp, P-glycoprotein; log P, logarithm of the octanol/water partition coefficient; cLogP, calculated log P; log D, logarithm of the octanol/water distribution coefficient at a given pH; cLogD, calculated log D; MW, molecular weight; HBD, hydrogen bond donor; HBA, hydrogen bond acceptor; PSA, polar surface area; TPSA, topological polar surface area; RB, rotatable bond

■

REFERENCES

(1) Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeney, P. J. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Delivery Rev. 1997, 23, 3−26. (2) Rankovic, Z. CNS Drug design: balancing physicochemical properties for optimal brain exposure. J. Med. Chem. 2015, 58, 2584− 2608. (3) Hitchcock, S. A.; Pennington, L. D. Structure−brain exposure relationships. J. Med. Chem. 2006, 49, 7559−7583. (4) Ghose, A. K.; Herbertz, T.; Hudkins, R. L.; Dorsey, B. D.; Mallamo, J. P. Knowledge-based, central nervous system (CNS) lead selection and lead optimization for CNS drug discovery. ACS Chem. Neurosci. 2012, 3, 50−68. (5) Wager, T. T.; Chandrasekaran, R. Y.; Hou, X.; Troutman, M. D.; Verhoest, P. R.; Villalobos, A.; Will, Y. Defining desirable central nervous system drug space through the alignment of molecular properties, in vitro ADME, and safety attributes. ACS Chem. Neurosci. 2010, 1, 420−434. (6) Vieth, M.; Siegel, M. G.; Higgs, R. E.; Watson, I. A.; Robertson, D. H.; Savin, K. A.; Durst, G. L.; Hipskind, P. A. Characteristic physical properties and structural fragments of marketed oral drugs. J. Med. Chem. 2004, 47, 224−232. (7) Morphy, R. The influence of target family and functional activity on the physicochemical properties of pre-clinical compounds. J. Med. Chem. 2006, 49, 2969−2978. (8) Heffron, T. P. Small molecule kinase inhibitors for the treatment of brain cancer. J. Med. Chem. 2016, 59, 10030−10066. (9) Reichel, A. Addressing central nervous system (CNS) penetration in drug discovery: Basics and implications of the evolving new concept. Chem. Biodiversity 2009, 6, 2030−2049. (10) Hodes, L.; Feldman, A. Clustering a large number of compounds. 3. The limits of classification. J. Chem. Inf. Comput. Sci. 1991, 31, 347−350. (11) Jamieson, C.; Moir, E. M.; Rankovic, Z.; Wishart, G. Medicinal chemistry of hERG optimizations: Highlights and hang-ups. J. Med. Chem. 2006, 49, 5029−5050. (12) Meanwell, N. A. Improving drug candidates by design: a focus on physicochemical properties as a means of improving compound disposition and safety. Chem. Res. Toxicol. 2011, 24, 1420−1456. (13) Didziapetris, R.; Japertas, P.; Avdeef, A.; Petrauskas, A. Classification analysis of P-glycoprotein substrate specificity. J. Drug Targeting 2003, 11, 391−406. (14) Giacomini, K. M.; Huang, S.-M.; Tweedie, D. J.; Benet, L. Z.; Brouwer, K. L. R.; Chu, X.; Dahlin, A.; Evers, R.; Fischer, V.; Hillgren, K. M.; Hoffmaster, K. A.; Ishikawa, T.; Keppler, D.; Kim, R. B.; Lee, C. A.; Niemi, M.; Polli, J. W.; Sugiyama, Y.; Swaan, P. W.; Ware, J. A.; Wright, S. H.; Yee, S. W.; Zamek-Gliszczynski, M. J.; Zhang, L. Membrane transporters in drug development. Nat. Rev. Drug Discovery 2010, 9, 215−236. (15) Kageyama, T.; Nakamura, M.; Matsuo, A.; Yamasaki, Y.; Takakura, Y.; Hashida, M.; Kanai, Y.; Naito, M.; Tsuruo, T.; Minato, N.; Shimohama, S. The 4F2hc/LAT1 complex transports L-DOPA across the blood-brain barrier. Brain Res. 2000, 879, 115−121. (16) Gianotti, M.; Botta, M.; Brough, S.; Carletti, R.; Castiglioni, E.; Corti, C.; Dal-Cin, M.; Fratte, S. D.; Korajac, D.; Lovric, M.; Merlo, G.; Mesic, M.; Pavone, F.; Piccoli, L.; Rast, S.; Roscic, M.; Sava, A.; Smehil, M.; Stasi, L.; Togninelli, A.; Wigglesworth, M. J. Novel 5953

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954

Journal of Medicinal Chemistry

Perspective

(37) Hughes, J. D.; Blagg, J.; Price, D. A.; Bailey, S.; DeCrescenzo, G. A.; Devraj, R. V.; Ellsworth, R. V. D. E.; Fobian, Y. M.; Gibbs, M. E.; Gilles, R. W.; Greene, N.; Huang, E.; Krieger-Burke, E. H. T.; Loesel, J.; Wager, T.; Whiteley, L.; Zhang, Y. Physiochemical drug properties associated with in vivo toxicological outcomes. Bioorg. Med. Chem. Lett. 2008, 18, 4872−4875. (38) Leeson, P. D. Molecular inflation, attrition and the rule of five. Adv. Drug Delivery Rev. 2016, 101, 22−33. (39) Hopkins, A. L.; Groom, C. R.; Alex, A. Ligand efficiency: a useful metric for lead selection. Drug Discovery Today 2004, 9, 430− 431. (40) Nicolas, J.-M. Species differences and impact of disease state on BBB. In Blood−Brain Barrier in Drug Discovery: Optimizing Brain Exposure of CNS Drugs and Minimizing Brain Side Effects; Di, L., Kerns, E. H., Eds.; Wiley: Hoboken, NJ, 2015, DOI: 10.1002/ 9781118788523.ch4. (41) Di, L.; Rong, H.; Feng, B. Demystifying brain penetration in central nervous system drug discovery. J. Med. Chem. 2013, 56, 2−12. (42) Feng, B.; Mills, J. B.; Davidson, R. E.; Mireles, R. J.; Janiszewski, J. S.; Troutman, M. D.; de Morais, S. M. In vitro P-glycoprotein assays to predict the in vivo interactions of P-glycoprotein with drugs in the central nervous system. Drug Metab. Dispos. 2008, 36, 268−275. (43) Syvanen, S.; Lindhe, O.; Palner, M.; Kornum, B. R.; Rahman, O.; Langstrom, B.; Knudsen, G. M.; Hammarlund-Udenaes, M. Species differences in blood-brain barrier transport of three positron emission tomography radioligands with emphasis on P-glycoprotein transport. Drug Metab. Dispos. 2009, 37, 635−643. (44) Uchida, Y.; Ohtsuki, S.; Katsukura, Y.; Ikeda, C.; Suzuki, T.; Kamiie, J.; Terasaki, T. Quantitative targeted absolute proteomics of human blood−brain barrier transporters and receptors. J. Neurochem. 2011, 117, 333−345. (45) Kalvass, J. C.; Olson, E. R.; Cassidy, M. P.; Selley, D. E.; Pollack, G. M. Pharmacokinetics and pharmacodynamics of seven opioids in Pglycoprotein-competent mice: assessment of unbound brain EC50,u and correlation of in vitro, preclinical, and clinical data. J. Pharmacol. Exp. Ther. 2007, 323, 346−355. (46) Raub, T. J.; Lutzke, B. S.; Andrus, P. K.; Sawada, G. A.; Staton, B. A. Early preclinical evaluation of brain exposure in support of hit identification and lead optimization. In Optimizing the “Drug-like” Properties of Leads in Drug Discovery, 1st ed.; Borchardt, R. T., Kerns, E. H., Hageman, M. J., Thakker, D. R., Stevens, J. L., Eds.; Springer: New York, 2006; pp 355−410, DOI: 10.1007/978-0-387-44961-6_16. (47) Dolgikh, E.; Watson, I.; Desai, P. V.; Sawada, G. A.; Morton, S.; Jones, T. M.; Raub, T. J. QSAR Model of unbound brain-to-plasma partition coefficient, Kp,uu,brain: Incorporating P-glycoprotein efflux as a variable. J. Chem. Inf. Model. 2016, 56, 2225−2233. (48) https://cortellis.thomsonreuterslifesciences.com/ngg/login.do (accessed February 11, 2016). (49) http://www.pjbpubs.com/pharmaprojects/index (accessed February 11, 2016). (50) ChEMBL. https://www.ebi.ac.uk/chembl (accessed February 12, 2016). (51) Drugs@FDA: FDA Approved Drug Products.https://www. accessdata.fda.gov/scripts/cder/daf (accessed February 12, 2016). (52) CAS SciFinder. http://cas.org/prdicts/scifinder (accessed December 15, 2016). (53) ChemAxon. http://www.chemaxon.com (accessed December 15, 2016). (54) Ertl, P.; Rohde, B.; Selzer, P. Fast calculation of molecular polar surface area as a sum of fragment-based contributions and its application to the prediction of drug transport properties. J. Med. Chem. 2000, 43, 3714−3717.

5954

DOI: 10.1021/acs.jmedchem.6b01469 J. Med. Chem. 2017, 60, 5943−5954