A better drug for Alzheimer's?
Alzheimers disease (AD) is a fatal disorder that robs victims of their most precious organ: the brain. By slowly destroying the complex web of neuronal connections that support cognitive processes such as thought and memory, AD destroys its victims personalities, leaving them unable to recall the past or process the present. The duration of the disease, from onset of mild dementia to death, averages 8 to 10 years. According to the Alzheimers Association, 10% of people over 65 years of age and nearly 50% of those over age 85 have AD (1). More than 4 million Americans have the disease, including former president Ronald Reagan. There is no known cure.
Caring for an individual with AD is extremely challenging. In addition to memory loss, patients may exhibit agitation, combativeness, hallucinations, depression, sleeplessness, and wandering. Although physicians can treat some of these behaviors using antipsychotics, anxiolytics, and antidepressants, they have few choices for treating the underlying degeneration of cognitive functioning. Therefore, in the absence of a cure, drug treatments that slow or delay the degeneration of cognitive function can improve the quality of life for patients and caregivers. However, such therapeutic agents constitute a distressingly short list.
The drugs that are currently available for use in AD were developed from an understanding of the role of the cholinergic system in cognitive function. Decreases in the markers of cortical cholinergic innervation were correlated early and consistently with AD and have led to the cholinergic hypothesis of memory dysfunction in aged patients, who show reduced activity of acetylcholinesterase (AChE), choline acetyl transferase, and high-affinity choline uptake, reflective of a damaged cholinergic system (2). Moreover, numerous studies in various animal models support the role of acetylcholine (ACh) in cognition and learning (3, 4). These observations have led to the idea that the symptoms of dementia seen in AD are associated with cholinergic loss. Thus, increasing the amount of ACh in the nervous system by inhibiting its degradation by AChE would ameliorate these symptoms.
Indeed, early clinical trials with the AChE inhibitor physostigmine suggested an improvement in cognitive measures in patients (5). The compounds narrow therapeutic window and short duration of action, however, limited its therapeutic use. More recently, the positive effects of the three currently marketed AChE inhibitors, which bolster the cholinergic system in a subpopulation of AD patients, have lent greater support to this hypothesis (see box, FDA-approved treatments for Alzheimers disease). Two of the drugs are only indicated for mild to moderate Alzheimers disease and do not slow disease progression. The third, rivastigmine, was just approved in the United States and has yet to become an established treatment for AD.
Discovery of DMP 543
More than 15 years ago, DuPonts medicinal discovery department initiated a program of identifying agents to tweak the cholinergic system with a novel approach. We sought agents that might enhance cholinergic function, not by increasing the duration of ACh released at the synapse, as esterase inhibitors do, but by enhancing endogenous stimulus-produced release of ACh at neuron termini. We thought that this approach would result in higher levels of the neurotransmitter solely when its release is triggered by excitation of the cholinergic neuron. The benefits of this strategy would be a lower probability of ACh overload toxicitymore likely with cholin es ter ase inhibitionand an elimination of the distortion of temporal patterns of cholinergic transmission that occurs with direct cholinergic agonists.
The cortical and hippocampal regions of rat brain are important areas of study in cognitive processes. To identify molecules that might enhance stimulus-induced ACh release, we used a rat cortical slice assay (10). This test allowed us to determine whether agents would induce the enhanced release of ACh when the cholinergic component of this system was stimulated by elevated potassium ion concentrations.
As a starting point for identifying agents that would enhance stimulus-induced ACh release, we began screening compounds from our proprietary library. We selected molecules from this collection that had some structural similarity to those with known cognition-enhancing properties, with the idea that part of this activity might be due to neurotransmitter release enhancement. In our in-house library, there was a molecule that possessed a lactam heterocycle also found in piracetam (1) and its analogues.
These nootropic agents were shown to have cognition-enhancing activity in various animal models (11, 12). The lack of a clear mechanism of action of these agents provided the possibility that enhanced ACh release might indeed be a property contrib uting to the activity of this series.
As a result of our strategy, we discovered early in our testing that bicyclic compound 2 possessed activity in our release enhancement assay.
When additional members of this series were evaluated, 3 (DuP 996), the 4-pyridyl isomer of 2, exhibited an acceptable level of ACh release-enhancing properties.
After performing an extensive evaluation in various animal models, we believed that this agent possessed properties that supported its development as a potential cognition enhancer for use in the treatment of AD (13, 14). However, after progressing into Phase III clinical trials, the agent failed to provide the statistically significant efficacy needed to justify its submission for a New Drug Application, and development was halted.
Throughout the development of DuP 996, work continued toward understanding the pharmacodynamic, pharmacokinetic, and distribution properties of this compound. In particular, we became more concerned with the penetration and distribution of the compound in the brain, as well as the mechanism responsible for its ability to induce release of ACh. By the time DuP 996 failed in the clinic, we had already identified a new series of molecules that addressed deficiencies that could have been responsible for the lack of efficacy of DuP 996 in humans.
With more thorough pharmacological evaluation of DuP 996, we identified three properties that we thought should be improved in a second-generation compound: potency, plasma half-life, and brain penetration. To improve potency, we examined a large variety of molecules with mono-, bi-, tri-, and tetracyclic cores to which pyridyl groups were attached (15). Among these systems, the tricyclic anthrone 4 showed considerable improvement in potency in our slice assay.
To improve plasma half-life, we focused on diminishing the tendency of the pyridyl pendants to undergo N-oxidation (16). Pyridine N-oxides were major metabolites of DuP 996 in rats, dogs, and humans and were inactive as release enhan cers. The goal of improving plasma half-life was met by incorporating fluorine substituents adjacent to the pyridine ring nitrogen (17). Fluorine diminishes the basicity of the pyridine nitrogen by about 5 orders of magnitude and eliminates oxidation at this location as a metabolic pathway. Indeed, agents possessing this type of substituent did have plasma half-lives significantly greater than that of DuP 996. Furthermore, the incorporation of the anthrone ring system and the addition of fluorine substituents increased the lipophilicity of our second-generation compound over that of DuP 996 by a factor of ~30, and apparently contributed to the greater brain penetration of our successor development candidate, DMP 543 (5) (18).
Mechanistic studies of DuP 996 and related compounds also identified a possible site of action for its neurotransmitter release-enhancing properties. When these compounds were evaluated in rat hippocampal CA1 neurons at concentrations that cause ACh release, electrophysiological parameters that control neuronal firing were affected. In particular, DuP 996 was shown to cause significant blockage of certain M-type potassium channels in voltage-clamped rat CA1 hippocampal neurons. The concentration-dependent curve for blockage was strikingly similar to that of ACh release enhancement. This electrophysiological evidence, along with that obtained with other analogues, strongly suggested that the ability of these agents to block potassium M-currents is at least partly responsible for their enhancement of ACh release (19).
We had succeeded in identifying a second-generation ACh release enhancer with improved potency, longer plasma duration, and greater brain penetration. These refinements, along with a better understanding of how the compound works, supported a second clinical evaluation of this approach as a means of bolstering the compromised cholinergic system of AD patients. However, before large-scale preclinical development of this new drug candidate could begin, a viable process for its preparation was required.
Process chemistry of DMP 543
Most scientists will agree that a drug that cannot be produced on sufficient scale to fulfill patient requirements is of little worth. The drug will remain a laboratory curiosity if we cannot safely prepare it without following all governmental regulations, avoiding patent infringement, and producing an acceptable profit or other benefit to the company. Between laboratory discovery and the pharmacists shelf are many channels, and the failure to navigate any one of them will result in the loss of the candidate drug. Most of these channels are usually addressed by the chemical process R&D (CPR&D) group in close collaboration with other departments that contribute to the drugs development.
Our CPR&D division is generally involved early in a drugs development. Typically, we conduct process research before the actual nomination of the compound to drug candidate status, to minimize the time needed to supply the drug to our internal customers. There are two broad mandates within CPR&D: Supply sufficient drug to support the developmental program, and improve the production process until it is transferred to an outside manufacturer. It is critical to continue to produce pure drug substances for ongoing clinical studies, safety assessment, formulation, pharmacokinetics, and pharmacy R&D.
The production process almost always has to be modified to meet these needs as the production scale continues to increase. The necessity to balance these requirements, along with the constraints listed earlier, leads to what Michael Pierce of our group calls doing chemistry with one arm tied behind your back.
CPR&Ds first look at the chemistry to produce DMP 543 came only 28 days before the due date to fulfill safety assessment needs. When time lines are particularly short and the required amount is sufficiently small, we are often content to move into the laboratory to start preparation with little or no chemical modification of the existing process. Because even a few weeks of additional exclusive sales during the patent life of a blockbuster drug may result in additional revenues of tens of millions of dollars, delays in development must be ruthlessly eliminated. At this point, we often use a variety of techniques, such as rotary evaporation or column chromatography, which become very difficult to engineer during further scale-up to kilogram levels. It is important to note that although we can cheat on many facets of a good process, such as cost of starting materials, purification methods, or inexpensive waste disposal, safety cannot be compromised. Reactions with the potential to produce violent decompositions or toxic byproducts are not used or are conducted in specialized equipment usually located in facilities dedicated to this sort of work (see box, Thermal analysis for hazardous chemistry).
The Discovery Chemistry Department developed the synthetic scheme to prepare the first useful quantities of DMP 543 (Figure 1) (17, 18, 20). As is typical for many compounds transferred to CPR&D, our first preparation closely followed this synthesis. The scheme consisted of the benzoyl peroxide-catalyzed benzylic chlorination of the fluoropicoline 6 followed by its conversion to the corresponding iodide with sodium iodide. This reactive iodide was used to bis-alkylate anthrone, using sodium hydride as a base. Finally, the crude drug was purified by a combination of column chromatography and recrystallization.
The value in reproducing the known procedure is that we start with a familiar chemical foundation for our future reaction modifications, we can validate our analytical methods with authentic samples, and we gain a little breathing room for process development before the next supply deadline falls due. As with most central nervous system pharmaceuticals, the dosage of DMP 543 was projected to be low, 110 mg/day. Such low doses mean that bulk drug preparations need not be large, particularly during early clinical studies, and it permits us to prepare the first lot in an ordinary research laboratory using prosaic laboratory glassware, such as 5-L vessels. Thus, after two weeks of work, we produced our first bulk preparation of 300 g of DMP 543.
Several process concerns became apparent during these early weeks. During our explorations of recrystallization conditions as a means of increasing purity, we found that we could produce many different crystalline forms of DMP 543. Much as calcium carbonate can crystallize into different crystal forms or polymorphs depending on reaction conditions, DMP 543 had the propensity to form, at last count, as many as 17 distinct polymorphs (see box,
Polymorphs and pharmaceuticals). This unusually large number of polymorphs is due to the variety of conformations that the pyridinyl side groups may adopt to fit into different crystal lattices. Michael Maurin of our pharmacy R&D department could differentiate these various forms by a combination of differential scanning cal orimetry and X-ray diffraction spectroscopy. Adding to the problem was that seemingly identical recrystallization conditions did not always produce the same polymorph or combination of polymorphs, and some crystal forms could transform into others, particularly when heated.
The reproducibility of polymorph selection was established when DMP 543 was recrystallized from a solution of ethyl acetateheptane. This was a convenient solvent system because the preceding process step used the same combination to purify DMP 543 by column chromatography. By increasing the concentration of the drug in the solvent mixture and varying the ratio of the solvents, we could efficiently recrystallize DMP 543.
To our surprise, the first process lot consisted of yet another previously unseen polymorph. We were unwilling to send our internal customers a polymorph that we were not sure we could reproduce, no matter how chemically pure it was. Evidently, the deciding factors for producing the target polymorphamong the various reaction conditions such as concentration, solvent, rate of cooling, seeding or lack of it, and impurity profileoperate within narrowly defined limits. This was not the robust process we required.
We decided to reexamine single solvents for recrystallization. Changing from two solvents to one would eliminate irreproducibility of solvent composition during recrystallization caused by differential evaporation rates; such variations could easily influence polymorph selection. A single solvent would also possess other advantages in process chemistry, such as simplifying solvent recycling. Isopropanol was the best candidate with respect to polymorph reproducibility; it also had a good solubilitytemperature relationship toward DMP 543, allowing high recovery of crystals after filtration and drying. This new crystalline form was examined by Pharmacy R&D and approved. However, as we were to learn later, the polymorph story of DMP 543 had more chapters.
The other factor we wished to modify before the next bulk drug delivery, due 4 months after our first preparation, was the identity of the anthrone alkylating agent. Because anthrone can yield monoalkylated and C,O-dialkylated products as well the C,C-dialkylated product DMP 543, the selection of the specific functionality on the methyl group was critical to avoid burdensome impurities and lowered yields. The product of the first step, the chloromethyl analogue 7, was inactive as an alkylating agent. We could prepare the bromide, but it produced an impure, crude drug. However, the iodomethyl analogue was, as we expected, an excellent candidate with regard to reactivity. It was prepared by the N-chlorosuccinimide (NCS) chlorination of 6 via a Finkelstein reaction with NaI in acetone (21).
Although the chlorinated mixture contained the monochloride 7, as well as unreacted starting material 6 and dichlorinated products, only the monochloride exchanged halogen with NaI to form the iodide 8. All of the halomethylpicolines were thermally unstable oils, but the iodide was particularly sensitive. Once it was purified and isolated, we reacted it immediately with anthrone to minimize losses due to decomposition. Of course, we could prepare and purify the iodide quickly in the laboratory, but almost all manipulations require significantly more time to complete on a larger scale, and we anticipated unacceptable losses to decomposition if we retained this procedure.
One option was to create the iodide intermediate in situ by combining the mixture containing 7 with anthrone anion in the presence of NaI. A catalytic amount of NaI would be sufficient, because the reaction of sodium anthrone with 8 would regenerate iodide ion. One object of our research was to identify stable intermediates of DMP 543. This would expand process flexibility by permitting storage of the intermediate or procuring it from contract manufacturers. Another trait of the ideal alkylating molecule would be crystallinity to facilitate handling on a large scale and, presumably, the ability to separate via recrystallization the chlorination-derived impurities that had formerly been carried through to the crude bulk drug.
Among such likely candidates were the alkyl or aryl sulfonates. The presence of sulfonates will often impart crystallinity to the oily parent molecule. This would add another step to our otherwise succinct synthesis, because the best precursor to a sulfonate would be the benzyl alcohol. As a prologue to the solution of this problem, Jianguo Yin of CPR&D had noted that the chlorinated mixture produced in the first step could be hydrolyzed by potassium carbonate in refluxing water to produce the benzyl alcohol 9 while leaving the chlorination reaction impurities untouched (Figure 2). Happily, the presence of the hydroxyl group lent sufficient water solubility to 9 that workup and purification became simple. The bulk of the impurities settled into an oily, dark layer upon cooling of the alkaline hydrolysis reaction mixture and were easily drawn off. The last vestiges of impurities were removed by heptane washes, and pure 9 was isolated in 6570% yield by subsequent ethyl acetate extraction and evaporation. Alternatively, it could be isolated by crystallization by concentration of the ethyl acetate solution and dilution with heptane.
These procedures allowed us to obtain an additional functional group transformation for little additional cost or bother basically, we had changed the workup of the chlorination step to include alkaline hydrolysis to produce alcohol 9. This insightful observation became the shortcut we needed to add the extra step of a sulfonate intermediate to the sequence. The p-toluenesulfonate was easily formed but was oily. The methanesulfonate (mesylate), 10, proved to be the ideal candidate. It was formed in 9095% yield by the addition of methanesulfonyl chloride to an ethyl acetate solution of 9 that also contained triethylamine to capture the hydrochloric acid generated. Following aqueous workup and crystallization by adding heptane, highly crystalline 10 was obtained.
The mesylate was, as expected, a stable entity, and it could be shipped without hazard or fear of decomposition. This was fortunate, because the second preparation was launched in our process facility in Deepwater, NJ, but was finished at the Merck Frosst research center in Dorval, Quebec, because the drug was destined for clinical trials outside the United States.
For this campaign, 10 was converted to the iodide by adding NaI, similar to the conversion of chloride 7. This transformation was necessary because the mesylate was a poor alkylating agent. The benzyl iodide was prepared as a much purer intermediate and, as anticipated, this produced a significantly cleaner reaction mixture. However, the product still required two recrystallizations sandwiched around a chromatography, followed by a final isopropanol recrystallization to attain 98.5% purity, the minimum required for clinical supply purposes. The final recrystallization also established the desired polymorph. Unfortunately, the crystals derived from the second recrystallization were of a shape not previously notedgranular and sugarlikeand this shape carried through after the isopropanol isolation as well. To our mounting concern, the melting point of this lot was higher than expected: 170 °C instead of 160 °C. Microscopy revealed large rounded crystals in place of the former bars and needles (Figure 3). X-ray diffraction spectroscopy confirmed our fears: The fickle nature of the crystallinity of DMP 543 had reasserted itself, and we had yet another polymorph.
Three additional recrystallizations of this lot from isopropanol were seeded with the 160 °C melting polymorph of DMP 543 (polymorph A), but they all produced the new polymorph B. Evidently, the laboratory was contaminated by seeds of polymorph B. Remarkably, recrystallization of DMP 543, seeded or not, at our process research laboratories in Deepwater, now only produced polymorph B, graphically demonstrating the pervasive nature of the seed, which seemingly leaped the >300 mi between Dorval and Deepwater in several days.
During the intervening years, we have never been able to prepare polymorph A again, although existing stocks retain their identity. This is a classic example of disappearing polymorphs, the sudden inability to reproduce a familiar crystal form, and it demonstrates how rapid the diffusion of a crystal seed can be (22). On examination of polymorph B, Pharmacy R&D subsequently accepted this crystal form as the preferred one; thermal analysis indicated that it is the most stable polymorph observed to date. In retrospect, it is fortunate that this polymorphic change occurred before further development of the project. The appearance of new polymorphs after commercialization is a process chemists nightmare. After 5 years, we are reasonably confident of our current polymorphs stability, but there is no guarantee that one day an even more stable arrangement of DMP 543 molecules will not appear.
We had a relatively long 5 months before our third scheduled delivery of 200 g, which was destined to support additional clinical studies. Our team had expanded to two labora- tories, and we were determined to establish a process that could be brought into the pilot plant if demand continued to increase. Several problems remained:
Our search to replace CCl4 led to several rewarding results. There are a number of potential ways to chlorinate picolines (23, 24). Some of these methods incorporate picoline N-oxides as reactive precursors; but in our case, the influence of the electron-withdrawing fluorine atom on the pyridine ring would inhibit such oxidations. In addition, we decided to forego the hazards of chlorine gas at a subpilot plant scale. We chose to continue to use NCS as the chloride source. A literature search revealed how infrequently NCS is used in nonchlorinated solvents, with only two examples for benzylic chlorination (25, 26). Acetonitrile was identified as an excellent alternative solvent for this reaction. Curiously, some reactions reached completion much faster than we expected: 2 h versus 8 h for a typical reaction.
We assumed that water had found its way into some of the reactions. Any moisture present would react with NCS, or the chlorine derived from it, to form hydrochloric acid, which is known to catalyze the analogous benzylic chlorination of toluene (27, 28). Ultimately, we found it more convenient to add 0.03 equiv of acetic acid, which reproducibly increased the chlorination rate to attain reaction completion at 2 h. The mechanism of this rate acceleration is not clear.
Thus, the solvent switch not only permitted the use of a much-preferred solvent, but also led to a fortuitous observation that shortened the duration of the reaction by 75%. As an added bonus, light was no longer needed, removing the last barrier to scale-up in our plant.
The bis-alkylation was improved in several distinct steps. Initially, we sought a more convenient base for a viable process. Screening various bases led us to lithium tert-butoxide, which was conveniently available in THF solution. This eliminated the handling problems of charging reactive powders and the inherently hazardous nature of NaH when used on large scale.
To understand the reasons behind the significant quantity of byproducts, the major impurities required identification. A side product of anthrone synthesis is bianthrone 11.
Although our commercial stocks did contain detectable amounts of this compound, the crude reaction mixtures consistently contained higher levels than could be traced to the starting material. This observation could be explained by the formation of additional bianthrone via a radical coupling induced by the presence of iodine as anthrone was converted to the enolate by base (2931). This observation suggested that we minimize the exposure of anthrone to NaI in order to reduce the formation of bianthrone.
The other major impurity was olefin 12. It most likely arose from the deprotonation of either 8 or 10, followed by SN2 alkylation onto another molecule of 8 or 10 and elimination of a molecule of acid. This finding is analogous to the reaction of 4-(chlorometh yl)pyridine with NaH (32). For a positive identification, we could make this olefin in 43% yield by treating 10 with potassium t-butoxide and NaI in THF. This suggested that we minimize the exposure of the alkylating agent to base to reduce the appearance of the olefin. The elucidation of the two major impurities led directly to the solution of the purification problem.
We carried out our optimized bis-alkylation process by preparing the anthrone anion in a separate pot, in which anthrone and 3 equiv of base were mixed. We subsequently transferred this solution to a mixture of 10 and 0.5 equiv of NaI. Solution analysis after 12 h indicated 8792% yields of DMP 543. We were extremely satisfied to find that our expectations of experimental changes were entirely fulfilled: The formation of the formerly major impurities 11 and 12 was now largely suppressed (<1% by HPLC). Yet, we were still unsuccessful in raising the purity of the resulting mixture to the point that recrystallization would result in an acceptably pure drug. To eliminate the tedious column chromatography, we developed a series of steps that further reduced the stubborn impurities. Following an initial isopropanol recrystallization, reslurrying the crude DMP 543 in refluxing cyclohexane removed the bulk of the remaining byproducts. Treatment initially with alumina and then with activated carbon now enabled recrystallization from isopropanol to produce an acceptable bulk drug. We used this procedure to produce 284 g of DMP 543 of >99.5 wt% purity, well within our clinical requirements.
We hope that we have successfully conveyed the challenge, urgency, frustration, and ultimate satisfaction that stemmed from our research and development of DMP 543. The route to a marketable pharmaceutical is mined with many pitfalls that ultimately trap most drug candidates. Although this can often be disappointing, the discoveries and intellectual pursuit are what make the journey worthwhile. We remained encouraged throughout, bolstered by the knowledge that this work may ultimately lead to treatments for devastating diseases such as Alzheimers.
The authors thank Robert E. Waltermire, Philip Ma, and Pat N. Confalone for their intellectual contributions and encouragement. In particular, the suggestions, support, and guidance of Joseph Fortunak are appreciated. The contributions of Yide Xing, Edward Gorko, and Ken Lynam are gratefully acknowledged. The remaining researchers are mentioned in the text. We thank Catharine Foris for helpful discussions on polymorphism. Our company became the DuPont Merck Pharmaceutical Co. in 1991 and was co-owned by Merck and DuPont until 1998. Robert Zamboni of Merck Frosst generously allowed our group to use his facilities for several preparations during 1995.
Jaan A. Pesti is a principal research scientist in chemical development R&D at DuPont Pharmaceuticals Co. (Experimental Station, E336/147, Wilmington, DE 19880-0336; 302-695-3189; firstname.lastname@example.org). He joined DuPont in 1983 after a postdoctoral appointment in synthetic organic chemistry at the University of California, Berkeley, and has worked in the chemical process group at the Experimental Station and at the Chambers Works in Deepwater, NJ. He received a B.A. degree in chemistry from Long Island University, Brooklyn, NY, and a Ph.D. in organic chemistry from Columbia University. He optimizes manufacturing processes for DuPonts pharmaceutical candidates and is involved in most other aspects of bringing a drug to commercialization. He is a cofounder of Balticum Organicum Syntheticum 2000, the first of a series of conferences on organic synthesis held in the Baltic countries.
Robert J. Chorvat is associate director, Chemical and Physical Sciences, at DuPont Pharmaceuticals Co. (Experimental Station, E353/136K, Wilmington, DE 19880-0353; email@example.com). He has been with DuPont since 1987 and has held supervisory positions in cardiovascular and central nervous system disease research. Previously, he was with G. D. Searle & Co. He is the author and co-author of more than 35 papers; he has written several book chapters on pharmaceutical agents and is the co-inventor on 34 patents and patent applications. His current responsibilities include competitive surveillance, screening library diversification, and information management. He received his B.S. degree in chemistry from Illinois Benedictine College (now University; Lisle, IL) and his Ph.D. in organic chemistry from the Illinois Institute of Technology, Chicago.
George F. Huhn is president of Decision Sight Inc., a consulting company that provides innovative decision support solutions and software applications for quantifying, visualizing, and optimizing management decisions. Before founding Decision Sight, he served as vice president of OmniAlret.com, a wireless financial information services company, and as managing consultant for the K. W. Tunnell Co. He also worked as a senior research scientist for Chemical Development R&D at DuPont and DuPont Merck Pharmaceutical. He is the author or co-author of numerous papers and articles and is the co-inventor on several patents. He holds a B.S. degree in chemistry from Drexel University and an Executive Masters of Science degree in the management of technology from the Wharton School and the University of Pennsylvania.