Understanding Pregnancy Centered Medications: Characterizing the Interactions of a Series of Sulfonylurea Analogs and the ATP Binding Cassette Transporter Proteins, P-Glycoprotein and Breast Cancer Resistance Protein

.................................................................................................................. ii ACKNOWLEDGMENTS ............................................................................................ iv PREFACE ..................................................................................................................... vi TABLE OF CONTENTS ............................................................................................. vii LIST OF TABLES ......................................................................................................... x LIST OF FIGURES ...................................................................................................... xi CHAPTER 1 .................................................................................................................. 1 Manuscript Submission Statement .................................................................................

x

INTRODUCTION
Glyburide (generic, glibenclamide) is a small molecule sulfonylurea used for the treatment of hyperglycemia. Glyburide is the most prescribed small molecule for the first line of defense or the treatment of Type II Diabetes due to its very attractive pharmacological profile. 1 The sulfonylurea compounds were first discovered by Janbon et al.; when researching sulfonamides in patients with typhoid fever, the team noted hypoglycemic results, leading to introduction of sulfonylureas to treat hyperglycemia into the US commercial market in 1955. 2 The sulfonylureas have evolved over two generations, though each generation shares the same core sulfonylurea backbone as presented in Figure 1. parameters, with the latter generation having an increased safety and efficacy. 3 The improvements to the physical-chemical and PK/PD parameters are due to the structural changes at the para position on the central aryl ring (R 1 ) and functional group R 2 attached to the remote urea position, as detailed in Figure 1. The first-generation sulfonylureas have smaller molecular weights, with more polar and hydrophilic substituents. The second-generation sulfonylureas have larger molecular weights, containing more non-polar lipophilic substituents. This increased lipophilic character allows for easier membrane permeability, and therefore an increased potency. This is an important point to discuss as the target protein, the sulfonylurea receptor (SUR1), is membrane bound protein, and having molecules with an increased hydrophobicity allows for the molecules to penetrate the membrane more readily, reaching the receptor more efficiently. 4 The commercial sulfonylurea structures are presented in   As expected, the similarity of the core structure of the first and second-generation sulfonylureas allow for a similar response to reducing hyperglycemia. 5 This can be explained in the structure activity relationship research performed for the development of the sulfonylureas, as demonstrated by an increase in binding affinity for the sulfonylurea receptor 1 (SUR1). The SUR1 receptor is an active transporter that has been shown to have multiple binding sites, A and B, with the more potent secondgeneration molecules believed to bind to both the A and B sites. For reference, the molecules tolbutamide (1 st gen) and glyburide (2 nd gen) and the representative binding sites of SUR1 are presented in Figure 4. 6 This increased structure activity relationship (SAR) work also presents as a longer half-life, higher lipophilicity, and increased plasma protein binding adding to the increase in efficacy and safety. 7 Glisoxepide Glyburide Glipizide Glimepiride Gliquidone Figure 4. Biding Site of SUR1 protein and 1 st and 2 nd generation SUs structural map. 8

SULFONYLUREAS FOR TYPE II DIABETES
The sulfonylurea chemical class has been widely used in the treatment of hyperglycemia for more than sixty years and is now a popular choice for a first line of treatment in diabetes care. Specifically, the sulfonylureas have become an integral treatment option for those patients who have not maintained or achieved adequate glycemic control through diet and exercise alone. 9,10,11 The sulfonylureas work by binding to the SUR1 receptor, which triggers the cascade of two signal pathways that release insulin in the pancreatic β-cell. The first pathway is the ATP-sensitive potassium channel (KATP) pathway and is the link between glucose metabolism and the stimulation of insulin production. The second pathway is a glucose signaling pathway and augments the insulin release due to the increase in the Ca 2+ concentration. 12 In this respect, the sulfonylureas are insulin secratogues, meaning that the mode of action is similar to that of insulin, stimulating the ATP sensitive Potassium (K ATP ) channels within the pancreatic βcells. 13 In summary, this action stimulates the insulin release via the occupation of the β-cell membrane SUR1. 14,15,16 In response to an increase in blood glucose concentrations, the enhanced rate of glucose metabolism causes the changes in concentrations of the adenosine diphosphate (ADP) and adenosine triphosphate (ATP), driving the closure of the pore. The depolarization of the cell membrane is due to a decrease in the K+ permeability into the cell and an opening of the Ca 2+ channels. This enhancement of the Ca+2 influx across the cell membrane is the final step of the cascade and triggers the release of insulin. 17 The dual pathway mechanism of insulin release in the pancreatic beta cells is depicted in Figure 5.

STATES
As part of normal function, the body needs to regulate insulin production as a direct response to the increase or decrease in blood sugar levels after food consumption. There are two states of blood sugar levels, hyperglycemia and hypoglycemia, which represent too much and too little blood sure, respectively. Hyperglycemia can be a chronic condition in which the body's inability to produce any or enough insulin causes elevated levels of glucose in the blood, and are known as Type I and Type II Diabetes, respectively. 19 Both types of Diabetes are characterized by an insulin resistance and a resultant insulin deficiency. In general terms, Type I diabetes is an autoimmune disorder leading to the destruction of pancreatic beta cells and hence the body's inability to produce insulin; and Type II Diabetes is primarily a problem of progressively impaired glucose regulation due to the dysfunction of the pancreatic beta cells and general insulin resistance. 20,21 . Also, Type II diabetes is influenced by many genetic and environmental factors, such as obesity and overweight contributing to the insulin resistance, compounding the issue. 22  This inability to control blood glucose levels occurs in pregnant women and is referred to as pregnancy induced gestational diabetes mellitus (GDM). The definition of GDM is carbohydrate intolerance resulting in hyperglycemia, with the onset or recognition occurring during pregnancy. 25 GDM is a well characterized disease affecting a large portion of the population and has been widely linked to the weight gain associated with pregnancy. 26 Although the exact cause of GDM is not known, there are some theories that may explain the condition and why it occurs. During pregnancy, the placenta supplies the fetus with nutrient and water, and produces hormones to maintain the pregnancy throughout gestation. Some of these hormones produces a contra-insulin effect, blocking the normal function of insulin. As the gestational period goes on, the placenta grows, produces more hormones, and the increase of insulin resistance becomes greater. Under normal circumstances, and normal pancreatic function, there would simply be an increase in insulin production to compensate for the state of insulin resistance. However, when the production of insulin is somehow blocked by the effect of the placental hormones, GDM can occur. GDM occurs in up to 20% of all pregnancy globally, depending on country, population demographic and lifestyle, though approximately 30-40% requiring pharmacological treatment. 27,28,29,30 Type II and GDM are characterized by insulin resistance and a subsequent insulin deficiency, and numerous studies have demonstrated that the rate of occurrence within a population of GDM to Type II Diabetes are similar. 31 However, the one real difference between Type II and GDM is that pregnancy is a state of increasing insulin resistance, as the need to provide a constant supply of nutrients to the fetus. This pathological increase in insulin resistance, diminished sensitivity and impaired insulin secretion due to various conditions of pregnancy, which runs counter to the need of the developing fetus and produces a serious condition of hyperglycemia in the mother. 32 So, the goal of any Diabetes-related management during pregnancy is to maintain blood glucose as close to normal as possible.

CHALLENGES OF GDM TREATMENT TO MOTHER AND FETUS
As noted, GDM occurs in a significant number of pregnancies globally, and is considered a risk factor for Type II Diabetes for the mother. There is a great deal of evidence that proves the treatment of GDM is far better than not treating GDM. In fact, common conditions are drastically reduced through treatment options, saving the mother and fetus from hypertension, preeclampsia, urinary tract infections, development of Type II diabetes later in life in mothers and child, as well as macrosomia, neonatal hypoglycemia, and childhood obesity. 33 Leaving these conditions untreated will potentially cause harm to the mother and fetus, and requires chronic maintenance through proper diet, exercise and pharmacological treatment. 34 Therefore, to be a successful medication for GDM, the need to control blood glucose in a similar manner to insulin, while maintaining the safety of the mother and fetus paramount. Prescribing drugs in pregnancy represents an unusual situation that must be constantly monitored. Drugs that will help the mother can deform or kill the fetus. 35 This is well characterized that there are a small group of drugs that are known to cause adverse events or birth defects in humans, but the safety or harm to the fetus must not be exaggerated. Today, the importance of monitoring the pregnant woman and fetus for harm is well understood, but initially, the placenta was thought to be a barrier, protecting the fetus from harm and the mother's blood, and delivering only what's needed, for example essential nutrients. The Thalidomide disaster of the 1950's demonstrated that the placenta was a very leaky barrier, and that precautions must be taken when treating pregnant women with medicines, as it is now understood that many substances will cross the placenta. 36 Treatment conditions of the mother quickly accounts for how much harm will be done to the fetus, and that can be determined by the rate and extent of the of the drug concentration that crosses the placenta, and at what gestational stage. 37 For those patients that cannot manage the hyperglycemia with diet and exercise alone, the preferred pharmacological treatment option for diabetes has been insulin. The same approach is taken for GDM, with insulin as the first line of treatment due to its high efficacy, safety, and the fact that the higher molecular weight insulin cannot cross the placenta barrier. 38,39 Insulin treatment however, suffers from compliance issues and in some countries the cost is prohibitive. 40 Specific to pregnant women, the insulin route of administration and the schedule of treatments present numerous difficulties. 41 Due to the similarities in GDM and Type II conditions, and the ease of treatment for oral medications, doctors have explored treating GDM with the small molecule sulfonylureas. 42,43 In order for the sulfonylureas to be effective in the treatment of diabetes, the patient needs to have retained some level of pancreatic insulin-releasing function. 44 This holds true for GDM based on the similarities of the pathophysiology with Type II, the sulfonylurea compounds are an excellent choice for treating either condition. 45,46 In fact, there is a growing acceptance of Glyburide as a treatment option for GDM, based on the attractive pharmacological response and low potential adverse side effects (i.e., neonatal hypoglycemia), making the sulfonylurea(s) a viable option based on compliance, and overall mother and fetal outcomes. 47,48,49 Considerable data exists suggesting that Glyburide is a safe alternative to insulin for treatment of GDM. 50  Glyburide does not cross the placenta to any appreciable extent, and Glyburide will leave the placental barrier against the concentration gradient. 56,57,58,59,60 This represents an ideal situation for treating pregnant women and has garnered much excitement.

TRANSPORTER STORY
The placental transport studies have demonstrated that Glyburide is a substrate for active transport by the ATP Binding Cassette (ABC) transporter proteins specifically Pglycoprotein (Pgp), Breast Cancer Resistance Protein (BCRP), and the Multiple Resistance Protein (MRP) family. 61, 62,63, 64 , 65 The ABC transporter proteins are part of a larger class of drug transporters, with a primary function to transport nutrients and endogenous substrates (sugars, amino acids, vitamins), and to protect the body from dietary and environmental toxins. 66 As the transporters are designed to transport substrates of a wide physiologic background, drugs that are similar in structure to the physiological substrate can also be transported. It is in this functionality and the ability to transport a broad range of substrates that the drug transporters play a significant role in the bioavailability, efficacy and PK of most drugs. 67 Transport of drugs can be classified in a variety of ways, with the main characterizations being directional (efflux vs influx), and through energy vs non-energy dependent (active or passive) means. For the context of the work reported here, we will focus on the efflux by active transport systems, which have been further classified into two main categories, primary and secondary. The primary systems utilize ATP hydrolysis to drive the transport and include the ABC transporters. The secondary systems are ones that utilize multiple driving forces such as ion concertation gradients, and electric potential difference across the cell membrane, an example is the Solute Carrier transporters (SLC). 68,69 The ABC proteins of the human genome comprise a family of 49 proteins, and based on their amino acid sequence, are divided into roughly 7 sub-families. As ABC transporters serve a protection function, they are found in most tissues at barrier locations throughout the body, such as brain, testes, heart, liver, kidney, and gastrointestinal tract. Transporter proteins are membrane-bound proteins whose primary function is to facilitate the flux of molecules in and out of cells, protecting from toxins. 70,71,72 The ABC drug transporters have a diverse substrate pool; as such, many molecules that bear structural similarities to the endogenous compounds may be recognized and transported. This diverse set of substrates covers molecules that are nonpolar, weakly amphiphilic, and encompass a diverse group of compounds from anticancer drugs to natural products. 73,74,75 As illustrated in Figure 4, all ABC Transporter proteins share similar structural components, consisting of 4 different domains: two transmembrane and two cytosolic domains. The transmembrane domains (TMD) consist of alpha helices (in groups of [5][6] embedded in the membrane bilayer, with the ability to recognize and transport a broad range of substrates by changing conformations. The nucleoside binding domains (NBD) is the location of the binding of ATP that drives the transport mechanism, containing the Walker A and B motifs for this function. 76 A complete transporter structure is   considered a homo-dimer or a fully functional transporter, example of which are Pgp, MRP, and OATP. The exception to this requires the two smaller monomer structures to pair together to form a homodimer, an example of this is BCRP.

PROBLEM STATEMENT AND SUMMARY OF REMAINING SECTIONS OF THESIS
The ABC transporters described above in Figure 6 hold the key to the unique placental transport of Glyburide. As discussed, the oral hypoglycemic sulfonylurea molecule Glyburide has been shown to be actively transported through a pronounced interaction with the ATP binding cassette (ABC) active transporter proteins. This unique placental transport behavior is largely due to the primary transport resulting in the efflux of Glyburide by Pgp and BCRP, minimizing the effect of the drug on the fetus.
Specifically, the small molecule Glyburide has been shown to not cross the placental barrier to any appreciable effect; and, Glyburide will leave the fetal compartment against the concentration gradient (active efflux). 78 The research presented here will focus on understanding this unique placental transport to design medications for pregnant women that are safe and effective, without presenting harm to the mother or fetus.
There is a great deal of research detailing the mechanism of action of the sulfonylureas, the PK/PD of the sulfonylureas during pregnancy, and even the specific ABC transporter proteins responsible for the unique placental transport behavior. However, there has been no research to date to explain the nature of the interaction with the ABC transporter proteins to better design pregnancy centered drugs. The research presented here will look to explain the interactions of a series of sulfonylurea analogs and two of the main ABC proteins responsible for active drug transport, BCRP and Pgp. The work described in this body of work was performed in cell-based transport assays using cell lines that overexpress Pgp or BCRP, and molecular modeling software packages to build structure activity models based on the transport activity.
In the coming chapters we will explore the interaction of the sulfonylureas with the two transporters, individually, building quantitative structure activity relationships (QSAR), and then complete the body of work with molecular modeling highlighting the molecular descriptors/features that drive the affinity to one transporter or the other. In Chapter 2 we will evaluate the sulfonylurea analogs using Madin-Darby Kidney Cells (MDCK) that overexpress Pgp and BCRP proteins, ultimately building 2D, 3D-QSAR and pharmacophore models using the molecular dynamics modeling software VLifeMDS® to explain the molecule-protein interaction. And finally, in Chapter 3, we will detail and explain the similarities and differences between the two molecule-protein interactions, using the molecular modeling software Cresset Group Forge®. The outcome of this research will be to characterize the molecular features driving the interactions with the ABC transporter proteins, in the hope that this knowledge will help design novel pregnancy centered medications. This active transport occurs at various locations throughout the body but has been identified as responsible for unique transport behavior of Glyburide at the placenta barrier. Glyburide has been shown to not cross the placental barrier to any appreciable effect, and will leave the placental barrier against the concentration gradient, attributed to the active transport by the ABC proteins. Harnessing the unique placental transfer by the ABC proteins and understanding what molecular features drive the drug-protein interaction, creates the possibility of designing medications specifically for pregnant women. Glyburide is a second-generation sulfonylurea hypoglycemic agent, widely used in the treatment of Type II diabetes. 88,89,90 The sulfonylurea molecule class are insulin secratogues, stimulating the ATP sensitive Potassium (K ATP ) channels within the pancreatic β-islet cells. 91 Stimulation of the K ATP channel triggers the insulin release via the occupation of the ABC sulfonylurea receptor (SUR1) in the β-cell membrane. 92,93 In order for this mechanism to work, patients need to have retained some level of pancreatic β-cell function, as shown in the cascade of events presented in Figure 7. In general terms, Type II Diabetes is primarily a condition of progressively impaired glucose regulation due to the dysfunction of the pancreatic beta cells and general insulin resistance. 15 Chlorpropamide 11%) and Glyburide will leave the placenta barrier against the concentration gradient. 107 As previously mentioned, this unique placental transport behavior has been demonstrated to be ABC transporter mediated. Evaluation of this unique transport has been performed in vitro, and repeatedly demonstrated in cell lines and in placenta-like models (vesicles, perfused placenta, etc.), confirming Glyburide as a substrate for active transport, and demonstrating that the mechanism of placenta transport can be studied in vitro. 108,109,110,111 Understanding the sulfonylurea and ABC transporter interactions and the unique placental transport serves as an example for the design of pregnancy center drugs. To further investigate this unique interaction of the sulfonylureas with the ABC transporter proteins, Glyburide and a series of sulfonylurea analogs were tested in in vitro cell-based transport assays using a Madin-Darby kidney cell (MDCK) lines overexpressing either the Pgp or BCRP proteins. From the measured transporter activity, we developed both 2D and 3D quantitative structure activity relationship (QSAR) models to account for the Pgp and BCRP substrate activity for the sulfonylurea compounds. To accompany this work, we have performed a two-step computational study to generate the 3D pharmacophore to aid in describing the parts of the molecule responsible for the Pgp and/or BCRP activity.

MATERIALS and METHODS
A set of 78 sulfonylurea compounds were acquired for this study, selected via a compound similarity search against Glyburide in both the Pfizer internal and publicly available external databases. The compounds for the study were further selected based on the Tanimoto searching criteria of >70% similarity, which allowed for the selection of molecules based on the structural similarities to Glyburide. The sulfonylurea analogs were divided into two groups, the commercial sulfonylureas (molecules [1][2][3][4][5][6][7][8][9][10][11][12][13][14] and the sulfonylurea analog series (molecules . The structures of all molecules studied are presented in Appendix I and II.

CELL CULTURES
The supplements, in 75cm 2 tissue culture T-flasks were incubated at 37°C with 95% and 5% CO 2 and passaged every 4 days after achieving 90% confluence. The cells were then harvested with trypsin and plated for a density of 2 x 10 6 cells/cm in Falcon/BD 96 well insert plates with a 1µm pore polyethylene terephthalate filter. Seeded inserts were then placed into prefilled Falcon/BD feeder trays containing 37mL growth medium and incubated at 37°C with 95% and 5% CO 2 for 4 days. Immediately prior to performing the assay, each well of the plates were assessed for uniform barrier functionality of the polarized cell monolayers. The integrity of the MDCK-MDR or MDCK-BCRP cell monolayers were evaluated by measuring the trans-epithelial electrical resistance, and only those cell monolayers with a measurement with of at least 320Ωcm 2 were used. 120

TRANSPORT ASSAY PROCEDURES
The cell transport assays are the most direct assay for performing drug transfer studies across a cell monolayer, and to determine a transporters function. The transport assays have been standardized in that the cell lines are widely shared between research labs or are commercially available. All transport assays were performed as per standard procedures, and as previously described. 121,122,123,124 Briefly, the assays were performed

LC-MS ANALYSIS
The LC-MS analysis was conducted according to internal procedures, and as previously

DATA ANALYSIS
For these studies, the apparent permeability (P app ) was calculated for each compound according to equation 1:

Equation 1
Where the area is the surface area of the cell monolayer (0.625cm 2 ), C D (0) is the concentration of compound in the donor chamber, t is the time, M t is the mass of the compound, and is the flux of compound across the cell monolayer. The P app was calculated in both apical to basolateral and basolateral to apical directions to determine the efflux ratio as shown in Equation 2.

Equation 2
where A→B and B→A denote the direction of transport. In order for a compound to be characterized as a Pgp substrate, the value of the efflux ratio (ER) would exceed a value of 2.5. The value of 2.5 was determined through internal data analysis and has been demonstrated to minimize the false-positive and false-negative results in the assay as tested by the appropriate internal control molecules for each cell line. The efflux ratio is a way to rank order and predict ABC transporter activity, with the larger efflux ratio meaning substrate activity. and used as is.

COMPUTER MODELING
All computational studies were performed on an HP, Intel i5 processor running Windows 7 Professional Office, with VLifeMDS® software (version 4.6). 128 In order to minimize variability and difficulty interpreting the results, it is important to establish a statistically significant correlation between the molecular descriptors and the biological activity. This correlation starts with the molecular alignment of the molecules across the test, training, and validation data sets. 129 Due to the well documented Glyburide ABC substrate activity, and the common sulfonylurea backbone of the molecules in the present study, Glyburide was used as the template molecule.
As the x-ray crystal structures of the compounds or the ABC transporter proteins were not available, the 2D structures were obtained from Chemaxon® Marvin Sketch and parameters were left as recommended by the software. The Figure 4 shows the overlay of the MMFF94 energy minimized molecules.
The total energy of the molecular conformation was calculated using the MMFF94 relationship presented in Equation 3: where, E B = bond stretching energy E A = angle bending energy E BA = bond stretching and angle bending energy E OOP = out of plane bending energy E T = torsion energy E vdW = van der Waals energy E ELEC = electrostatic energy As shown in the energy equation, the MMFF94 energy calculation has multiple energy terms to capture all potential molecular motions. The total energy of the system is calculated as the sum of individual energy terms defined for a force field. 136 Figure 10. Molecular Overlay of MMFF94 energy minimized molecules.
After energy minimization, the molecules were loaded into the VLifeMDS® software and aligned to a set conformation using the VLifeEngine® module. To correctly overlay the molecules for quality QSAR modeling, the alignment was performed using the modules' atom-based approach. This approach allowed for the selection of the central arylsulfonylurea section of the Glyburide template molecule. Figure 9 depicts the molecule overlay of the energy minimized molecules, with Figure 11 showing the sulfonylurea backbone chosen for molecule conformer alignment template (highlighted in red).  The initial evaluation of the commercial sulfonylureas was run at a standard concentration, and all compounds were run in at least triplicate. Traditionally, transport assays evaluating the sulfonylureas have been performed with concentrations ranging from 1-500µM, depending on the substrate or inhibitor function. 142 Accounting for this range, our studies were performed at a nominal concentration of 2 µM for the sulfonylurea analogs due to ease of screening and to not saturate the transport mechanism for Pgp or BCRP activity. Also, to make sure our cell monolayers were performing as  As shown in the bar graphs of Figure 13, the lower concentrations of the compounds demonstrated the active transport out of the cells from the basolateral to apical. In the lower concentration ranges (2-20 µM) the rate of efflux exceeded the rate of absorption by the following values: approximately 6-8 times for Glyburide, 5-6 times for Glimepiride, 3-4 times for Gliquidone, and 5-7 times for Glisoxepide. At the highest concentration tested (100µM), the transport was nonlinear, and the transport started to balance out, especially for the Glyburide assay. This is an effect of the inhibitory nature of the compounds at the 100µM concentration, as discussed in literature. 144 These findings demonstrate the transport is a carrier mediated process from the basolateral to apical compartment for each of the compounds represented. These results were expected based on the known Pgp-Glyburide interaction, and the molecular similarities of the four compounds. Also, the non-Pgp activity of Gliclazide was expected as the only reported Gliclazide Pgp activity found in literature was at very high concentrations. 145 The bulk of the experimental work reported in this paper was performed to evaluate a large series of sulfonylurea analogs as substrates of the ABC transporters (Pgp and/or BCRP), and to use this information to build 2D and 3D QSAR models. Based on the initial experiments, and internal procedures, all the sulfonylurea analogs cell transport assays were performed at 2µM and measured in triplicate. It is well understood that the work reported describes local models for both Pgp and BCRP, as all the compounds used in the study were sulfonylureas, though containing structurally diverse R 1 and R 2 groups   Developing QSAR models is an iterative approach, and hundreds of modeling simulations were performed to evaluate the necessary parameters and build the most representative models defining each sulfonylurea and ABC transporter activity. The models presented here represent the most appropriate to define the sulfonylurea-protein interactions, and the most predictive, and will be discussed individually here.

QSAR DISCUSSION
A QSAR model is a regression or classification model that relates a set of predictor variables (x) to the potency of a response variable (y). The QSAR model is defined by the two and three-dimensional descriptors that make up the space surrounding the molecule, and are important tools to describe the correlation of a biological activity to the molecular features responsible for the activity. To explain the interaction of the sulfonylureas with the Pgp and BCRP proteins, we developed both 2D and 3D QSAR models that will be described in more depth here.
The 2D QSAR models require calculating molecular descriptors and matching the molecular descriptors to a biological activity. The 2D modeling and descriptors are the most widely used, based on their simple nature of employing a direct math algorithm that is highly reproducible and requires minimal compute time consumption. 146 In contrast, the 3D QSAR models are more in-depth and require more computational time to complete the analysis. In the VLifeMDS® software, there are approximately 1300 molecular descriptors used for the 2D QSAR models, whereas the 3D QSAR models use steric (S) and electrostatic (E) descriptors that specify the region where the structural feature variation between the test and training set of compounds leads to an increase or decrease in activity. 147  variables that are carried out by the software. The fundamental difference between the forward and backward selection methods is the use of descriptors (backward) or no descriptors (forward) to build the model. 148 The Pgp and BCRP models with the greatest predictive ability were achieved with MLR or kNN regression analysis and will be described here in more depth.
The kNN method adopts a nearest neighbor principle for generating the relationship between the molecular descriptors and a given biological activity. The basic principle of the kNN classification model is that the compounds are assigned to a class membership of its nearest neighbors in a common rectangular grid, taking into account the weighted similarities between a compound and its nearest neighbors. 149,150 The kNN statistical methodology is represented by points of the grid in the form of where: y is the dependent variable (activity), and the points and values represent descriptors, or steric/electrostatic spaces on the grid.
Also included in the VLifeMDS® software are numerous regression methods to calculate the best for the data, with the Multiple Linear Regression (MLR) methodology demonstrated the greatest predictive ability for the sulfonylurea and ABC transporter protein interactions. 151 The MLR methodology relates the dependent variable (activity) to a number of independent variables (molecular descriptors) through linear equations. The MLR analysis estimates the values of regression by applying least squares curve fitting methods. 152,153 The MLR equation takes the form: where: Y is the dependent variable (activity), b's are the regression coefficients of the corresponding x's (molecular descriptors), and c is the regression constant (intercept).
Simply stated, the points and values represent molecular descriptors surrounding the molecule on the grid in 2-dimensional space or the steric and electrostatic descriptors of The predictive ability of the both the 2D and 3D models was evaluated by the crossvalidation (q 2 ) term, employing the leave one out (LOO) methodology. The LOO principle is a method that computes the statistics for the left-out value, allowing for faster computation time. 154,155 Simply, in the LOO principle, one data point is selected from the test set, and the model is built around the remaining points, with the model cross- The predictive ability of all the QSAR models was confirmed by the predicted r 2 (pred_r 2 ) and the external validation test data set. One final point used in the evaluation is the standard error of estimation for the cross-validated q 2 and the predicted r 2 , with a low standard error signifying that the models are statistically significant. In summary, the VLifeMDS® software program calculates the best model based on the squared correlation coefficient (r 2 ) which defines the linearity, the cross validated coefficient (q 2 ) which is a measure of quality of the fit, and the predicted r 2 value. Finally, all the models generated for each protein in both the 2D and 3D space were evaluated for acceptable performance against linearity, cross-validation, standard error, and the predictability of the model to identify substrates not in the training or test set.
Many models succeeded in one or two of the criteria, but the models chosen to describe the behavior had the best balance of high linearity and cross-validation values, low standard error, and high ability to predict substrate behavior.

Pgp 2D QSAR DISCUSSION
Of the 200 models run, the optimum model to describe the sulfonylurea-Pgp interaction within the 2D space was obtained in Model 58, using the kNN methodology with sphere exclusion. 156 The sphere exclusion technique represents a simple clustering method whereby molecules are clustered together based on a defined similarity score, continuing until all molecules are grouped. 157 Model 58 parameters are shown in Table 2 and proved very robust and an excellent cross validation (q 2 ), with the predicted r 2 and externally validated r 2 very close in value, but not over-predicting. This is an important point as over-predicting models can lead to false predictions, allowing the model to incorrectly label molecules as substrates/non-substrates for activity. The external predictability of Model 58 was determined by the predicted r 2 (pred_r 2 ) value, which was 0.8150, meaning that the model has a prediction rate of 81.5%. This prediction rate is right in line with the observed Pgp activity data for Glyburide, Glimepiride, Glipizide, Glisoxepide, and Gliquidone with the model ability to select 4 out of 5 molecules' activity. Table 1 summarizes the kNN QSAR Model 58. Graphing the predicted vs. the experimental Pgp activity is also a good way to assess model performance. Model 58 shows good linear correlation between the two data sets, as presented in Figure 16.   Table 3 Molecules used in study with experimental, predicted, and residual activity.
To further evaluate the model describing the sulfonylurea and Pgp interaction, the nearest neighbor algorithm was performed at three, six, or nine nearest neighbors for the calculations. Though all the models presented similar results, the kNN models with the increased nearest neighbor values did not outperform the simpler Model 58 with only three nearest neighbor parameters. In fact, as shown in Table 4, Model 58 with the n = 3 parameter outperformed the n = 6 and n= 9 nearest neighbor models with better Predicted r 2 and Ext Val r 2 values, in a head to head comparison. The three most representative kNN models using the 3, 6, and 9 nearest neighbors are presented in Table 4  This was relatively unexpected as usually the more neighbors, the more rigorous the calculations due to the larger fields of interaction, and hence a better model. However, the structural similarity of the sulfonylurea analogs made the selection of a larger nearest neighbor factor an insignificant point. As the longer and more extensive calculations for increased nearest neighbor parameters did not prove to be a better modeling assumption, further modeling attempts will be performed as simple as possible, increasing the throughput without sacrificing robustness. As you presented in Table 4 above, the nearest neighbor calculations for the n = 3, 6, or 9 values were close but turned out to be not that as predictive as the Model 58 (n = 3). Comparing the models, we see that the cross validation was better in the n = 6 and 9 models (0.7151 to 0.7306 to 0.7210); however, the n = 3 model beat the others in both the pred_r 2 (0.8150 compared to 0.7961 and 0.6312), and external validation r 2 (0.8288 compared to 0.7706 and 0.7880).

BCRP 2D QSAR DISCUSSION
The optimum model to describe the sulfonylurea-BCRP interaction within the 2D space was obtained in Model 89, using the MLR methodology, with the sphere exclusion technique. 158 Model 89 parameters are shown in Table 3, proved very robust and an excellent cross validation (q 2 ), with the predicted r 2 and externally validated r 2 also very close to in value, but not over-predicting, similar to the model for Pgp. The external predictability of Model 89 was determined by the predicted r 2 (pred_r 2 ) value, which was 0.7711, meaning that the model has a prediction rate of 77.11%. Graphing the predicted verse the experimental Pgp activity also demonstrated the linear correlation between the two and is presented in Figure 17.   Table 6.  Figure   18.  descriptors and the definitions are presented in Table 8.  160 Model 10 parameters are presented in Table 4 and  Plotting the predicted Pgp activity against the experimental activity also provides an assessment of the model. As shown in Figure 19, the predicted vs experimental Pgp activity for the training and test sets showed excellent linearity. There was one outlier in the test set (molecule 77) that had a very high Papp value (12.4) but was predicted to be low due to the 2-methyltetrahydrofuran ligand attached to the benzamido group.

BCRP 3D QSAR DISCUSSION
The optimum model to describe the sulfonylurea-BCRP interaction with the 3D space was obtained in Model 65, using the MLR methodology, also using the sphere exclusion technique, and with Gasteiger Marsili energy minimization calculations. 161   Plotting the predicted BCRP activity against the actual experimental activity also provides an assessment of the model. As shown in Figure 22 the predicted vs actual BCRP activity for the training and test sets showed excellent linearity.  The equation above shows the contributions for the various descriptors contributing to the explanation of the BCRP activity. Those descriptors having a positive value shows that an increase in that descriptor increases the BCRP activity. Conversely, there were three descriptors that shown an inversely proportional relationship in that decreasing the value of that descriptor will increase the BCRP activity. From the model it is observed that the electrostatic fields E_714 and E_3205 need more negative ligands to increase activity, with a negative coefficient on both the cyclohexyl moiety and on the carbonyl of the amide, respectively. Also, the steric descriptor S_1060 shows that the cyclohexyl ring needs smaller group at that position. The field points model is shown in Figure 24. The plot of steric and electrostatic field points indicates the region of local fields around the aligned molecules. 163 The colored spheres (blue and green) represent the fields, steric

PHARMACOPHORE MODELING DISCUSSION
A pharmacophore model is a 3D representation of the steric and electrostatic features needed to ensure the optimal molecular interactions with the desired biological activity. 164 Pharmacophore identification studies were performed using the MolSign module of the VLifeMDS® v4.6 software package. In general, the pharmacophore models consist of  Like the QSAR modeling efforts, the first and second-generation sulfonylureas were analyzed against the low energy conformer Glyburide as the template. As expected, only 5 of the sulfonylureas overlaid precisely to the core conformer structure. The first-generation sulfonylureas are lacking the benzamido ligand of the scaffold and therefore only partially fit the conformer; with Glimepiride, Glisoxepide, Gliquidone, Glipizide, and Gliclazide, all fit the scaffold overlay.
For the pharmacophore evaluation, the 45 sulfonylurea analogs demonstrating Pgp activity, and the 25 sulfonylureas demonstrating BCRP activity were used to build the respective models. In the VLifeMDS® MolSign software module, Glyburide was loaded as the reference molecule. There are three parameters to optimize that influence the pharmacophore generation: primary feature count, tolerance limit and maximum distance allowed. Briefly, the primary feature count is the number of features in the pharmacophore; the tolerance limit is set to between 10-30% and accounts for variability in the pharmacophore features; and the max distance allowed determines how far apart a feature can be from another (max distance is 15Å). For the pharmacophore generation, the parameters where optimized to capture the best feature coverage of the molecules in the study. For both the reported Glyburide pharmacophores, the feature count was set to 5, the tolerance limit was set to 10%, and the max distance was set to 10Å.
These parameters were optimized to maximize the pharmacophore model activity to capture as many molecular features as possible without dampening the sensitivity. Too few or too many features will not allow for a robust model by including too many or too few molecules in the model. Meaning, allowing for too wide criteria on the tolerance may include features that are not shared across the entire molecule set; on the feature count, limiting the feature count to a reasonable value (i.e., 5) forces the algorithm to pick the best features to define the activity without selecting the entire molecule; and limiting the distance allows for feature resolution across the entire molecule. The impact of these assumptions is that we are building a local model maximizing the activity of the sulfonylureas against Glyburide, to best explain the transport behavior of Glyburide in the ABC transporters Pgp and BCRP.
From the pharmacophore generation, the sulfonylurea-Pgp and sulfonylurea-BCRP pharmacophores each contained 5 key features describing the steric and electrostatics of the protein-sulfonylurea activity. The key pharmacophore features were aromatic rings, hydrogen donors, and hydrogen acceptors. Though these features were similar across both ABC transporter models, there were important differences concerning the activities and therefore the pharmacophore models will be discussed separately.

SULFONYLUREA-Pgp PHARMACOPHORE
The sulfonylurea-Pgp pharmacophore contained 5 features, 2 aromatic rings, 1 Hydrogen donor, and 2 Hydrogen acceptors, as pictured in Figure 26.  As expected, the 5 second generation sulfonylureas fit the pharmacophore template well, being similar in size and structure to Glyburide. The first-generation sulfonylureas however are much smaller in size and do not contain the ligands attached at the para position of the arylsulfonylurea core structure. For reference, the core sulfonylurea structure is presented in Figure 26, with the para position of the arylsulfonylurea ring denoted as R 1 circled in blue.

SUFONYLUREA-BCRP PHARMACOPHORE
Similar to the Pgp pharmacophore, the sulfonylurea-BCRP pharmacophore contained 5 features, but differed in the types and location of the features. The 2 aromatic rings, 1 Hydrogen donor, 1 Hydrogen acceptors, and 1 hydrophobic as pictured in Figure 27. The

PHARMACOPHORE MODELING DISCUSSION
The value of the pharmacophore model is the highlight of the ligands at certain points on the molecule structure that drive the structure activity relationship. This is possible to accomplish as all of the molecules in the study have been normalized to the most active molecule (Glyburide), providing the means to identify the structural changes that are believed to be implicated in the observed pharmacological activity.
The two pharmacophore models generated showed many similarities and some interesting differences. Both models shared the two aromatic rings, but differed in the This combination of hydrophilic and hydrophobic character matches well with the understanding that these ABC proteins rely on these broad moieties instead of very specific chemical structures. 165 Clearly from both models, the activity in both the Pgp and BCRP models was driven by the arylsulfonylurea and the substituted benzamido ring. The non-substrates of both proteins were lacking at least the benzamido ligand and the steric and electrostatic character that it brings.
As expected, a large set of the sulfonylurea analogs that do not have Pgp or BCRP activity, also contain some or all of these features. However, for those molecules that do not have all the key feature points for either pharmacophore, the activity is obviously lacking. For those molecules that share the key features, but still lack activity, there were a few noticeable differences. These molecules also had other substituents that interfered with the activity, bulky groups, or electron withdrawing/adding.
As there is no crystal structure of Pgp or BCRP currently available, the pharmacophore model allows for a mapping of the key features on the molecule that relate to an increase in activity. The specificity of the pharmacophore model will be lacking as a true binding pocket definition of size and electrostatics for either protein is not known. However, the pharmacophore models for both Pgp and BCRP represent the key molecular features (steric/electrostatic character) of the ligands on the molecule Glyburide that explain the ABC transporter activity.

GENERAL DISCUSSION
The oral hypoglycemic sulfonylurea molecule Glyburide has been shown to be actively Care was taken to select the experimental concentrations that resulted in relevant and comparable results between the two studies. Of special note are the three sulfonylureas Glimepiride, Glisoxepide and Gliquidone, that were identified as substrates of Pgp and BCRP and have not been previously reported. Our results also confirm Glyburide as a substrate of Pgp and BCRP, as previously reported. 166 The first reported carrier mediated transport of Glyburide was carried out by Goldstein et al demonstrating that Glyburide was a substrate of Pgp. 167  The second part of our research looked at the transport mediated activity of the series of sulfonylurea analogs at a standard concentration of 2µM. The concentration was held constant so as to identify the structure activity relationship and build three-dimensional pharmacophore and QSAR models to explain the nature and character of the Pgp-and BCRP-sulfonylurea interactions. The cell assays were performed for all available sulfonylurea analogs, and the apparent permeability and subsequent ER were calculated.
The ER were then converted to log(ER) to use as the biological activity in the two and three-dimensional QSAR models. The 45 compounds that showed Pgp activity and the 25 compounds that showed BCRP activity were used to build the respective pharmacophore and QSAR models detailing the molecular attributes of the interaction.
The pharmacophore models demonstrated that the two main sections of the molecule, the benzamido and aryl-sulfonylurea moieties, are key to explaining the interaction with both Pgp and BCRP. In fact, in our study the molecules that lacked either substituent showed no Pgp/BCRP activity. The pharmacophore model for each protein detailed 5 features responsible for activity, with the models consisting of aromatics, 2 hydrogen acceptors, 1 hydrogen donor, and 1 hydrophobic group. The five-feature count was determined to be the most appropriate value to use based on trial and error with the software program.
There was a balance of too few/too many features in the pharmacophore and still have a model that represents the key points across the entirety of the molecule. The Pgp pharmacophore consisted of two aromatic rings, two hydrogen acceptors and one hydrogen donor. Meanwhile, the BCRP pharmacophore shared the aromatic and one hydrogen acceptor but differed in that there was only one hydrogen donator, with the added feature of hydrophobic character on the benzamido ligand. The Pgp and BCRP pharmacophore models are presented in Figure 29 as simple diagrams.  Figure 30 with some example ligands that increase of decrease activity based on our work. Figure 30. Core structure of sulfonylurea with the potential ligands that can increase or decrease the activity for both Pgp and BCRP. Note: light blue highlights benzamido ligand; dark blue highlights arylsulfonylurea; blue circle denotes where bonding occurs for ligands; and blue semi-circle denotes where bonding occurs for ligands.
The SAR of the sulfonylureas with the SUR1 receptor was first reported by Meyer et al in 1999 and described the ligands on Glyburide that are required for SUR1 activity.
Meyer reported that removing the cyclohexyl or benzamido ligand marketed reduced SUR1 activity. In fact, swapping the larger ring structures for smaller, less bulky groups (i.e., methyl group) had the same effect. 170 This was also witnessed in our work, the sulfonylureas that more closely resembled the Glyburide structure, containing the benzamido and aryl-sulfonylurea portions, were more likely to be substrates of Pgp and BCRP. Similarly, if the sulfonylurea or analogs were missing one or the other benzamido of aryl-sulfonylurea portion, then they were not recognized as Pgp or BCRP substrates.
Further, as demonstrated in the pharmacophore models, strategic placement of the hydrogen donating/accepting, aromatics, hydrophobic, or bulky groups increased the Pgp and BCRP activity.
Knowing that the sulfonylureas work through binding to the SUR1 protein (ABCC8), which is also an ABC The second-generation sulfonylureas were designed to interact with both binding pockets of the SUR1 protein, and in doing so were considerably more potent than the preceding generation. This increased potency across the generations is the same impact as seen in our work, in fact it mimics it very closely. Simply, those analogs that did not have both the arylsulfonylurea and benzamido ligands did not exhibit Pgp or BCRP activity.
Building on the similarity of the SUR1 and Pgp transporter proteins theory, Bessadock et al have reported the Glyburide pharmacophore based on the 3-dimensional alignment of Glyburide with vinblastine, a known Pgp inhibitor. Their pharmacophore models called out five key features: the two aromatic rings, two hydrogen donor groups (NH and NH proximal to S), and one electron donor group (C=O) and is presented in Figure 30. Our work is very much in line with these findings and confirms the Bessadock model. Also, this model is representative of the hypothesized molecule features of the SUR1, which was expected based on the proximity on the ABC membrane transporter phylogenic tree. 172 Figure 32. Key features in Glyburide-Pgp interaction, reproduced from Bessadock. Key: green is aromatic red is hydrogen donating, blue is hydrogen accepting groups.
We have found that pharmacophore model presented in Figure 31 accounts for many of the same molecular features as reported in our research. We confirm the necessity of the two aromatic rings, the two hydrogen donators, and the hydrogen acceptor. These features are confirmed by the Pgp substrate activity of the 45 sulfonylurea molecules used to build the pharmacophore.

CONCLUSION
Pgp and BCRP are the most studied ABC transporter proteins and have been found to be largely responsible for the multidrug resistance phenomenon in cancer therapies, but characterization of either has been difficult due to the lack of the membrane crystal structure and general substrate promiscuity. Numerous groups are working towards generating the crystal structure of the ABC transporters, but that work is still years to come. 173,174 The lack of a defined substrate binding pocket for either Pgp or BCRP proteins has made it difficult to develop one single QSAR or pharmacophore model describing the spatial and structural features responsible for activity. 175  Computational QSAR models describing the Pgp and BCRP activity for a series of sulfonylurea analogs were derived with statistical significance and predictive capabilities by using the 2-dimensional and 3-dimensional molecular descriptors presented in the VLifeMDS® suite. The predicative ability of these models observed for the training, test, and validation sets of molecules make these models useful for designing new compounds to help explain sulfonylurea-Pgp and sulfonylurea-BCRP activity. Further, this is the first set of QSAR studies performed to explain the ABC protein interactions of the hypoglycemic sulfonylureas.

COMPARISON OF PGP AND BCRP ACTIVITY FOR A SERIES OF SULFONYLUREAS BY ACTIVITY CLIFF ANALYIS
by Samuel D Bell 7 ; Brian Bronk 8 ; William Euler 9 is submitted to Molecular Pharmaceutics

COMPARISON OF PGP AND BCRP ACTIVITY FOR A SERIES OF SULFONYLUREAS BY ACTIVITY CLIFF ANALYIS
ABSTRACT Activity Atlas software was used to describe the protein activity between a series of sulfonylureas and two of the ATP Binding Cassette (ABC) transporter proteins found to be largely responsible for the active transport of Glyburide, Pgp and BCRP. The Activity Cliff Analysis has provided a platform to understand the interactions that are important to the guide the drug-protein interaction, with the potential to better design medications for targeted delivery, for instance in pregnant women. The Activity Cliff Analysis is based on the key features of average shape, hydrophobic region, and electrostatic patterns of the active compound, and were mined and mapped to detail the differences in molecular features driving the ABC protein activity to either Pgp or BCRP, specifically.
As described in our previous work, Activity Atlas was used on a large series of sulfonylurea analogs, with the objective of investigating and understanding the molecular features that underlie the ABC protein activity in the hopes of better design of pregnancy centered medications. As expected there were many similarities in the molecular features driving the protein-sulfonylurea activity; but there were also many appreciable differences as demonstrated by the analysis of the hydrophobic, electrostatic and shape molecular descriptors. When coupled to the 3D QSAR data, the Activity Atlas method is particularly useful to visualize and decipher structure activity relationships.

INTRODUCTION
Glyburide is a second-generation sulfonylurea used in the treatment of Type II Diabetes and has been shown to be especially effective in patients that retain some level of insulin production from the pancreatic beta cells. As outlined in the Chapter 1 Introduction section, researchers began looking to Glyburide as a treatment option for gestational diabetes mellitus (GDM). The similarities in the pathophysiology of type II and GDM make the oral hypoglycemic agents a smart choice, as compared to insulin.
In fact, researchers have run numerous clinical studies demonstrating that Glyburide is as safe and efficacious in pregnant women as insulin, without many of the insulin related difficulties.
It is in these studies of Glyburide with GDM that researchers have determined a unique placental transport activity. Specifically, Glyburide has been demonstrated to not cross the placental barrier to any appreciable effect; and, Glyburide will leave the fetal compartment against the concentration gradient. 176,177 Numerous studies have been performed to understand the mechanism responsible, and it has been found that Glyburide is actively transported by the ATP binding cassette (ABC) transporter proteins, specifically Pgp, BCRP and the MRPs. 178,179,180,181,182,183 The ABC proteins are responsible for the active transport of endogenous and exogenous molecules at the barrier membranes throughout the body and have been shown to be important players in the pharmacokinetic profiles and disposition of many drugs.
The documented interactions of Glyburide with the ABC transporter proteins have prompted numerous studies evaluating Glyburide, Glipizide, and Gliclazide, with similar results. 184 In these studies Glyburide, Glipizide, and Gliclazide have been shown to act as substrates or inhibitors of the ATP transporter proteins, depending on the concentration dosed. 185,186,187 Figure 32 shows the molecular structure of the three sulfonylureas most studied with the ABC transporters. Figure 33. Glyburide, Glipizide, and Gliclazide structures.
Many studies have been also performed to understand the placental transport, pharmacokinetics/pharmacodynamics, or even to determine the proteins responsible for the transport of Glyburide. Our research is the first investigation to study the interaction of the sulfonylurea hypoglycemic agents and the ABC transporter proteins.
In this investigation, a series of sulfonylurea analogs were studied for activity against Pgp and BCRP in cell-based transport assays. From the transport assay data, we used Cressets Activity Atlas modeling capabilities to identify and describe the key molecular features of the sulfonylurea analogs driving the interaction with either the Pgp or BCRP. Understanding the differences in the molecular targets could have a profound impact on tailored drug delivery and the future of medicine. Specifically, these studies may contribute to the design of medications for pregnant women that are safe and efficacious, without presenting any harm to the fetus.

DATA SET
A total of 78 sulfonylurea analogs were available to be evaluated in the Pgp and BCRP cell-based transport assays. For the Pgp assay, all 78 compounds were tested, with a subset of 53 compounds tested in the BCRP assays. The transport assays were performed in Madin-Darby Canine Kidney cells that were transfected with overexpressing Pgp or BCRP genes. The transport assays were validated with positive and negative transport control compounds, and the cell-monolayers determined acceptable for use by measuring the trans-epithelial resistance, as discussed in the preceding chapters. As our research and published articles have shown, Glyburide has been reported as a substrate for both Pgp and BCRP, and as a substrate and/or inhibitor for other ABC transporters and was therefore used as the reference molecule in the studies. As a crystal structure of the sulfonylureas with either protein has not been reported, the lowest energy conformation was used in these studies as described by Lins et al. 188 The lowest energy conformations were determined with by Chemaxon® and ChemDraw® software using the MMFF94 energy minimization calculation algorithms. 189,190 The sulfonylurea analogs were tested at a constant concentration (2µM) as per internal procedure, with each measurement performed in at least triplicate to assure adequate statistical control. The data generated from each cell-based assay was then used to build 2D and 3D quantitative structure activity models (QSAR) and a pharmacophore model with the molecule design software suite from VLifeMDS. The 2D and 3D QSAR models demonstrated excellent predictability, but for the work presented here, we will focus on the 3D data set and modeling. For both cell-cased assays, the 3D protein models demonstrated excellent predictability with the Pgp model demonstrating a linearity value of r 2 = 0.8304 and cross-validation of q 2 =0.7501; and the BCRP model demonstrating a linearity value of r 2 = 0.9083, with a crossvalidation value of q 2 = 0.7789.

CONFORMATION HUNT AND ALIGNMENT OF MOLECULES
The first step of the modeling efforts was to generate the most stable conformations and to overlay the molecules to a template. As Glyburide is well characterized and known to interact with Pgp and BCRP, we used Glyburide as the template molecule for conformer hunting and molecular alignments. The conformation hunt was carried out with Cresset Forge® software package using the MMFF94 energy minimized structure of Glyburide, as we have previously reported. 191 The remaining sulfonylurea analogs were aligned by the maximum common substructure, using the "very accurate but slow" configuration setting to maximize success. As with all 3D model generation, the molecular alignment is the most critical first step and if not performed correctly can lead to incorrect modeling practices. After completion of the alignment process, visual inspection of the all the molecular alignments was performed to make sure there were no disparities between molecules.
The molecules were evaluated for maximal alignment scores against Glyburide and the molecules that were suboptimal, were adjusted and rescored. For example, manipulating the phenyl ring of the benzamido substructure proved to be the most common fix, and maximized the substructure similarity scores to Glyburide.

ACTIVITY ATLAS MODELING
The Activity Atlas modeling is part of the Cresset Forge® molecular dynamics software package and is routinely used for the design and discovery of new molecules.
The Activity Atlas modeling suite performs three types of analyses that are key to the understanding the activity for the molecules against a specific target, and are defined as the Average of the Actives, Activity Cliff Summary, and the Regions Explored Analysis. 192 For our research, the Regions Explored analysis was not performed as it is an assessment of what regions have been explored on the molecule, disregarding the biological activity. For the research performed here, we will discuss the Average of Actives and Activity Cliff Analysis, as they will have more insight on the activity of the sulfonylureas.

ACTIVITY CLIFF MODELING
As described in the literature references, Activity Analysis modeling can pinpoint the critical region of the structure activity relationships (SAR), providing a visual summary of the activity cliff for each of the data sets. 193 An Activity Cliff is defined as a pair or series of structurally similar compounds that have a large difference in potency or activity. 194 This is especially pertinent for sets of molecules run in different assays, allowing the data to be combined and compared to further characterize the molecules, the associated biological activity and critical differences in SAR.
As discussed by Cheesewright et al., describing how the molecule fits into a binding pocket requires the ability to define the properties near the molecular surface, and not simply as a collection of atoms. 195 Normally, this modeling effort would be too calculation-intensive with long computer run times. The benefit of the Cresset software to simplify and this process and condense the complex three-dimensional fields (for example, electrostatic and van der Waals) down to local extrema, called field points. These field points are defined for key molecular features, such as H + donor, H + acceptor, positive/negative ionic character, hydrophobic character, etc. The field points are then grouped into field patterns, with these field patterns able to compare molecules directly, representing a summary of the 3D properties of each molecule.
In order to evaluate the molecular features against a given biological activity, a 3D lattice of grid points is created covering the entire volume surrounding the molecule.
For each point or each pair of points, a coefficient is then created to define the space.
The calculation for each coefficient is calculated by: Co eff = (disparity-minDisparity) * ∆Field xyz * weight Where Co eff is the grid point

Disparity is the value between two molecules
MinDisparity is a minimum threshold value ∆Field xyz is the field difference at this point described for this molecular pair Weight is the product of the molecule and alignment weights The calculation for the coefficient represented in the above equation helps pinpoint the critical regions of the structure activity relationship by looking at each grid point in relation to the others.

AVERAGE OF ACTIVES MODELING
The Weight is the product of the molecule and alignment weights

RESULTS
The results of the Activity Cliff Analysis show distinct difference between the sulfonylurea analogs interactions with ABC transporters, Pgp and BCRP. To ensure robust calculations, the molecules were energy minimized using the MMFF94 energy field, and then conformer generation was performed to ensure the molecular structure disparity was minimized to eliminate variability and allow for the modeling to be governed only by the biological activity of the aligned structures.
In performing the comparative analysis differentiating the model systems, we compared the BCRP activity data against the Pgp activity data ( Figure 2). As shown in the Figure, there is good correlation between the Pgp and BCRP activity, with a linearity score of 0.84 This correlation shows that the nature of transport for the two proteins is similar, with more sulfonylurea analogs substrate of Pgp than that of BCRP, as shown by the activity. This proved a very interesting point as the understanding was that Glyburide (and by assumption, the sulfonylurea class) would show more substrate activity with BCRP, as is reported in literature. [1][2][3][4][5][6][7][8] The activity map detailing the Pgp activity to that of BCRP is shown in Figure 34. Glyburide are similar in structure, molecular weight, and ligand placement, and therefore will behave in a like manner, as expected. The third region is those molecules that have increased Pgp/BCRP activity as compared to Glyburide, which can be due to change in the physical-chemical properties of the molecules, such as molecular weight, lipophilicity, etc.
As previously reported by our lab, the pharmacophore of the sulfonylurea-Pgp and the sulfonylurea-BCRP systems demonstrated different feature points, as shown in Figure   3. A simple definition of the pharmacophore model is a representation of the steric and electronic features that are necessary for the interaction with a protein or a biological response. For both of the pharmacophores, the feature points were spread throughout the entire molecule, outlining the areas of interest. As shown in Figure 3, there are four feature points that are similar in both pharmacophores, with two distinct differences. The differences in the Pgp and BCRP pharmacophore models were found to be at the amide and benzamido ring moieties, respectively. For reference, the fivepoint pharmacophores describing the Pgp and BCRP protein interactions with the sulfonylureas are presented in Figure 35. The 3D QSAR modeling of drug-protein interactions is also very important to help understand the similarities and differences of the two protein models. As we have previously demonstrated, the 3D QSAR models identify the molecular features responsible for the activity in the space surrounding the molecules. As shown in   Starting with the Activity Cliff Summary maps, we will describe the molecular features needed for each interaction to improve the selectivity of the sulfonylurea analogs for Pgp or BCRP. As shown in the activity summary maps in Figure 5, there is a good deal of overlap in the electrostatic, hydrophobic, and shape molecular descriptors. However, there is also differences specific to each Pgp and BCRP model The electrostatics average summary explains where an increase in negative (red) or positive charge (blue) will increase activity. As presented in Figure 6, the electrostatics summary details the differences in the sulfonylurea interactions needed for the Pgp and BCRP proteins. As shown in Figure 37. The hydrophobics average summary explains where an increase in hydrophobic character will increase or decrease the activity and is described as favorable (green) or unfavorable (purple) hydrophobics. As shown in Figure  The shape summary explains where bulky groups are needed to increase the activity and is described as favorable (green) or unfavorable (purple) shape. As shown above in Figure

DISCUSSION
The ABC transporter proteins are responsible for the active transport of substrates into and out of cells. Specifically, ABC transporter proteins have gained much notoriety for being responsible for the multidrug resistance (MDR) phenomenon in various cancer treatments. 198 Much work has been done over the years to identify the features of a molecules that are responsible for inducing interactions with the ABC efflux transporters. 199 Expanding on that research, many groups have looked to develop global models that will identify the universal features of molecules that are substrates or inhibitors. 200 However, this has proven very elusive as the broad substrate specificity of each of the transporters proteins allows for much overlap for the substrate recognition. From a biological standpoint, the overlap of substrate recognition is needed and very much an inherent part of the body's redundant system of protection. 201 202 With this is mind, we endeavored to describe the interaction of a single class of compounds, the sulfonylureas, with two prominent ABC transporters, Pgp and BCRP.
In order to decipher the underlying structure activity relationship of the sulfonylurea and ABC transporter proteins, the molecules' transporter activity was examined by the Activity Analysis Cliffs for the electrostatic, hydrophobic, and shape descriptors. The first step in this process was to generate the lowest energy conformers, and then overlay the molecules on the template molecule, Glyburide. This allowed for the activity to be solely a function of the molecular structure, and for the generation of field points surrounding the molecules. These field points were then used for the visualization and calculation of the Activity Cliff data describing each sulfonylureaprotein interaction.
Results of the Activity Cliff analysis study detail many similarities governing the activity and interaction between the sulfonylureas and the ABC transporters, Pgp and BCRP. To start, the central aryl ring serves as a linker to the benzamido and sulfonylurea ligands and does not contribute to the transporter activity calculation.
There is electrostatic activity in both models surrounding the sulfonylurea and carbonyl group. Also, there is a good deal of overlap between the models on the favorable shape on activity surrounding the cyclohexyl and benzamido ligands. The hydrophobic summary also details very similar activity surrounding the cyclohexyl and benzamido ligands of both models.  Figure 41. The Activity Atlas Modeling has proven to be a valuable tool for deciphering the structure activity relationships defining the interactions of sulfonylurea analogs with Pgp and BCRP transporter proteins. Using the Activity Atlas modeling we were able to leverage the three-dimensional insight for a series of sulfonylureas and the associated Pgp/BCRP activity, based on electrostatic, hydrophobic, and shape descriptors. This led to identifying key molecular features that were missed in our previous modeling attempts. The Activity Atlas modeling specifies the need for bulky steric groups, an increase in hydrophobic character, and more positively charged groups on/near the cyclohexyl ring. Also, the Activity Atlas model specifies the need for increased hydrophobic character and bulky steric groups on the benzamido moiety.
Finally, and representing the difference in activity between the two proteins, the space immediately surrounding the sulfonylurea and amide groups require different charges to drive the activity, with an increase in negative charge increasing the Pgp activity, and an increase in positive charge increasing the BCRP activity.

CONCLUSION
This is the first summary and comparison of a large series of sulfonylurea analogs activity with the ABC transporter proteins, Pgp and BCRP. Based on the work previously reported on Glyburide and its unique placental transport activity, we sought to define the molecular properties responsible for interaction with the ABC transporters in the hopes of learning how to use this information to better design pregnancy centered medications. The sulfonylurea analogs tested in our work show a higher propensity for Pgp activity as compared to that of BCRP. This means that in our cell-based transport assays, the same molecule had higher activity in the Pgp model than in the BCRP model. As the cell-based assays are very sensitive to drug concentration effects, we maintained drug concentrations to that were carefully selected to not elicit a false positive or negative response.
Applying the understanding from the activity cliff analysis, the 3D QSAR models generated were condensed down to a simple map of the critical points driving the structure activity relationship for three descriptors: electrostatics, hydrophobics and shape. The Activity Cliff Analysis detailed key similarities and differences governing the molecular features driving activity in the Pgp and BCRP protein models. Though the electrostatic, hydrophobic, and shape activity cliffs were shared across molecules in both models, there are important differences in that could be used to drive the sulfonylurea interaction to Pgp or BCRP. In this regard, the Activity Atlas modeling proved to be a valuable tool as previously undiscovered features were characterized and shown to be of key importance to define the sulfonylurea-Pgp/BCRP activity.

Designing Pregnancy Centered Medications: Insights and Recommendations
The first three chapters detail the evaluation of a series of sulfonylurea analogs in cellbased transport assays to determine if they are substrates for the ATP binding cassette transporter proteins, Pgp and BCRP. The cell lines used in the study were Madin-Darby Canine Kidney cells, stably transfected for Pgp or BCRP gene overexpression, described and characterized in literature and summarized in Chapters 1 and 2.
Glyburide is the most prescribed oral sulfonylurea medication and has been shown to be actively transported by the ABC transporter proteins, specifically Pgp and BCRP, but also by various other transporters. Our work examined the interaction of the sulfonylurea analogs with Pgp and BCRP, and determined the molecular features driving the specific protein activity. We confirmed that the sulfonylureas are substrates of both Pgp and BCRP and built pharmacophore and QSAR models explaining the Pgp and BCRP activities in the two and three-dimensional space.
The study of Glyburide and the ABC transporters has presented a unique and exciting opportunity to learn about tailoring drug delivery for pregnant women. As discussed previously, Glyburide has demonstrated unique placental transport due to the active transport by the ABC transporters, minimally crossing the placenta and leaving the placenta against the concentration gradient. The treatment of Gestational Diabetes Mellitus with Glyburide is an interesting case for drug delivery to the receptor (pancreatic β-cells), allowing the mother to maintain and control insulin levels for proper function, while limiting the exposure of the drug to the fetus (through placental transport). Numerous studies have been performed to evaluate the pharmacokinetics and pharmacodynamics of Glyburide during pregnancy detailing: the fetal exposure, umbilical cord concentrations, the safety compared to insulin, and placental transport assays. However, there has not been extensive research to understand the molecular properties responsible for driving the interaction with the ABC transporters responsible for the placental transport, for instance, with Pgp instead of BCRP.
Our research demonstrated that the sulfonylurea molecules need to have specific physical-chemical parameters to have ABC transporter activity, in Pgp and BCRP.
These parameters, among many others, speak to the size, shape, and lipophilicity of the molecule, governed by the placement of ligands at the appropriate locations on the molecule. As discussed in chapters 2 and 3, the sulfonylurea analogs with similar or larger molecular weight or molecular volume as compared to Glyburide were more likely to be substrates for both Pgp and BCRP. Conversely, sulfonylurea analogs that were much smaller in molecular weight and volume, or were missing the arylsulfonylurea and benzamido core structure, were not substrates of either Pgp or BCRP. We found that these results were consistent with the structure activity relationship of the first and second-generation sulfonylureas, with the smaller, firstgeneration sulfonylureas being weak substrates of the SUR1 receptor. The main difference between the fist and second-generation sulfonylureas lies in the "core" structure found to increase potency of the second-generation sulfonylureas. This required core structure is presented in Figure 42, with the benzamido ligand highlighted in light blue, and the arylsulfonylurea highlighted in dark blue.  Based on this work presented in chapters 1-3, we present some recommendations to design medications for pregnant women based on these learnings from our Glyburide case study. First, and assumed, the molecules need to have appropriate safety and PK/PD parameters, for example, high protein binding (>99%), high bioavailability (80-100%), moderate half-life (4 to 10hrs), to afford predictable and well-behaved ADMET (absorption, distribution, metabolism, excretion, toxicology) properties.
Similarly, important physical-chemical properties would warrant the molecules to be at least the same size (M w of ~500g/mol), volume (molecular vol 424.74), and molecular shape (low energy conformer in Figure 44) to interact with binding pockets of the target receptor (in this case, SUR1), as well as the placental ABC transporter proteins. Specifically, those molecules fitting the Glyburide pharmacophores need an increase in the bulky and hydrophobic ligands on the ends of the molecule, with strategically placed electrostatic groups (negative or positive) in the central core to enhance the ABC transporter activity. Designing molecules that incorporate these molecular features, aligned against the Glyburide pharmacophore for Pgp or BCRP, will increase the respective activity. The preceding are all learnings from our research to direct the sulfonylurea molecules to interact with one protein or the other. For the purpose of keeping drugs out of the placenta, interactions with more than one ABC transporter protein would provide a potentially superior solution. For conditions requiring treatment in transporter rich areas, for example the blood brain barrier, this approach will not work due to the need to cross the tight junctions containing the numerous transporters designed for redundant protection. However, for those disease states that have receptor targets outside of the area protected by the ABC proteins, this approach may be appropriate. Therefore, to design pregnancy centered medications, we need to maintain the fetal protection by designing the molecules that interact with numerous placental transporters. This would allow for protection of the fetus by the natural and redundant system of the ABC Transporter proteins, inherent in pregnant women and ever increasing over the course of gestation. By design, the molecules would interact with more than one ABC transporter, allowing for redundant protection of the fetus if an issue arose with one of the transporters in the mother, for example, Pgp deficiency or mutation in the BCRP protein reducing the activity.
To better define this idea, future research efforts will focus on evaluating the substrate activity of the sulfonylurea analogs in other ABC transporter proteins known to be present in the fetal-placental barrier.
Following the work presented in this thesis, the two and three-dimensional QSAR models would be built for the new sulfonylurea-ABC protein interactions, and the respective pharmacophores generated. Aligning the pharmacophore models of each ABC transporter against the lowest energy conformation of Glyburide will allow for the overlap the key points from each molecule to the overall key pharmacophore features. Then, comparing the numerous models against each-other in the Activity Analysis software will help describe the needed features for each interaction. And finally, this concept will be tested in a more "global mode", using our sulfonylurea activity models to predict and describe the key pharmacophore features and activities of other molecular classes.
As you can imagine, the treatment options for pregnant women is severely limited due to the fact that most drugs will cross the placenta and present a safety risk to the fetus.
Therefore, designing drugs not to cross the placenta would allow doctors to treat pregnant women's conditions without harming the fetus. This would expand the treatment options and positively impact the lives and conditions of an entire subset of the population currently overlooked, with the potential to revolutionize treatment options.