Current Topics in Medicinal Chemistry, 2007, 7, 557-568
`
`557
`
`Discovery of JANUVIA™ (Sitagliptin), a Selective Dipeptidyl Peptidase IV
`Inhibitor for the Treatment of Type2 Diabetes
`
`Nancy A. Thornberry and Ann E. Weber*
`
`Departments of Metabolic Disorders and Medicinal Chemistry, Merck Research Laboratories, P.O. Box 2000, Rahway,
`NJ 07065
`
`Abstract: The emergence of glucagon-like peptide 1 (GLP-1) as a well validated approach to the treatment of type 2
`diabetes and preclinical validation of dipeptidyl peptidase IV (DPP-4) inhibition as an alternate, oral approach to GLP-1
`therapy prompted the initiation of a DPP-4 inhibitor program at Merck in 1999. DPP-4 inhibitors threo- and allo-isoleucyl
`thiazolidide were in-licensed to jump start the program; however, development was discontinued due to profound toxicity
`in rat and dog safety studies. The observation that both compounds inhibit the related proline peptidases DPP8 and DPP9
`led to the hypothesis that inhibition of DPP8 and/or DPP9 could evoke severe toxicities in preclinical species. Indeed, the
`observed toxicities were recapitulated with a selective dual DPP8/9 inhibitor but not with an inhibitor selective for DPP-4.
`Thus, medicinal chemistry efforts focused on identifying a highly selective DPP-4 inibitor for clinical development. Initial
`work in an a-amino acid series related to isoleucyl thiazolidide was discontinued due to lack of selectivity; however, SAR
`studies on two screening leads led to the identification of a highly selective b-amino acid piperazine series. In an effort to
`stabilize the piperazine moiety, which was extensively metabolized in vivo, a series of bicyclic derivatives were prepared,
`culminating in the identification of a potent and selective triazolopiperazine series. Unlike their monocyclic counterparts,
`these analogs typically showed excellent pharmacokinetic properties in preclinical species. Optimization of this series led
`to the discovery of JANUVIA™ (sitagliptin), a highly selective DPP-4 inhibitor for the treatment of type 2 diabetes.
`
`INTRODUCTION
`
`TARGET SELECTION
`
`The pathogenesis of type 2 diabetes (T2DM) involves a
`set of three primary defects: insulin resistance, b cell dys-
`function, and hepatic glucose overproduction. These defects
`are the principal targets of both current and future therapy.
`Currently available classes of oral antihyperglycemic agents
`include PPARg agonists, sulfonyureas/meglitinides, and
`biguanides. These agents are used either in monotherapy or,
`increasingly, in combinations to lower glucose levels.
`Despite the availability of a range of agents for T2DM, many
`diabetic patients fail to achieve or to maintain glycemic
`targets. In addition, current therapies have limited durability
`and/or are associated with significant side effects (GI
`intolerance, hypoglycemia, weight gain, lactic acidosis and
`edema). Thus, there remain critical unmet medical needs in
`the treatment of this disorder. With an increasing understan-
`ding of the molecular pathways involved in glucose control,
`a range of new potential targets have emerged for treatment
`of the key areas of pathogenesis. In particular, there has been
`increased emphasis on new therapies that increase the
`circulating concentrations of insulin in a glucose dependent
`manner, most notably, glucagon-like peptide 1 (GLP-1)
`analogs [1] and dipeptidyl peptidase 4 (DPP-4) inhibitors
`[2]. In this review we describe Merck’s DPP-4 inhibitor
`program, which was initiated in 1999 and culminated with
`the discovery of JANUVIA™ (sitagliptin), a potent and
`highly selective inhibitor of DPP-4 that shows excellent
`promise for the treatment of T2DM.
`
`*Address correspondence to this author at Merck Research Laboratories,
`P.O. Box 2000, Rahway, NJ 07065; Tel: (732) 594-5796; E-mail:
`ann_weber@merck.com
`
`Over the last decade, GLP-1 receptor agonism has
`emerged as one of the best validated approaches for the
`treatment of T2DM. For example, in 1997 it was reported
`that continuous infusion of GLP-1 to diabetic humans
`resulted in normalization of both postprandial and fasting
`glucose [3]. More recently, sub-chronic (6 wk) continuous
`infusion of GLP-1 was shown to result in profound and
`significant decreases in fasting plasma glucose (14.1 to 10.1
`mM) and HbA1c (9.2 to 7.9 %) [4]. It is generally accepted
`that the key mechanisms responsible for glucose lowering by
`GLP-1 receptor agonism are: (i) stimulation of glucose-
`dependent insulin biosynthesis and secretion, (ii) glucose-
`dependent inhibition of glucagon release, and (iii) delayed
`gastric emptying.
`Also in the 1990s, it became increasingly clear that GLP1
`in vivo, specifically via
`was very
`tightly regulated
`proteolysis at the N-terminus to produce an inactive peptide,
`and that the key enzyme responsible for this inactivation was
`DPP-4, a proline specific dipeptidyl aminopeptidase [5,6].
`These findings resulted in the initiation of programs at
`several companies
`to
`identify DPP-4 resistant GLP-1
`analogs, and also led to the testing of DPP-4 inhibitors in
`animal models of diabetes, where increased levels of GLP-1,
`enhanced insulin secretion, and improved glucose tolerance
`were observed [7-9].
`The human validation of GLP-1, together with preclinical
`validation of DPP-4 inhibition as an alternate oral approach
`to GLP-1, prompted Merck to initiate a project on this
`enzyme in 1999. Our enthusiasm for this mechanism was
`based primarily on the view that this approach would have at
`least three potential advantages over currently available
`agents. First, because GLP-1 stimulates insulin release in a
`
` 1568-0266/07 $50.00+.00
`
`© 2007 Bentham Science Publishers Ltd.
`
`Merck Exhibit 2114, Page 1
`Mylan Pharmaceuticals Inc. v. Merck Sharp & Dohme Corp.
`IPR2020-00040
`
`

`

`558 Current Topics in Medicinal Chemistry, 2007, Vol. 7, No. 6
`
`Thornberry and Weber
`
`Probiodrug had evaluated the safety of L-threo-isoleucyl
`thiazolidide in 4-week toxicity studies in rats and dogs [20].
`In rats, toxicities were limited to the presence of lung
`histiocytosis and thrombocytopenia at relatively high doses
`(77.5 and 698 mg/kg, respectively). In dogs, acute central
`nervous system (CNS) toxicities, characterized by ataxia,
`seizures, convulsions, and tremor, were observed at 75
`mg/kg, and bloody diarrhea was also observed at 225 mg/kg
`upon acute dosing. No additional toxicities were noted in
`these 4 week studies. However, in subsequent chronic
`toxicity studies at MRL, upon 5-6 weeks of treatment with
`this compound in dogs, mortality and profound toxicities
`occurred at doses ‡ 25 mg/kg/day. These toxicities included
`anemia,
`thrombocytopenia, splenomegaly, and multiple
`organ pathology mainly affecting the lymphoid system and
`gastrointestinal tract.
`As noted above, we also in-licensed the allo isomer of
`isoleucyl thiazolidide, which, when compared to the threo
`isomer, has virtually identical affinity to DPP-4, similar
`pharmacokinetic and metabolic profiles, and similar in vivo
`efficacy in an oral glucose tolerance test in diet induced
`obese mice [20]. In parallel with the chronic toxicity studies
`with L-threo-isoleucyl thiazolidide, this compound was
`evaluated in an acute tolerability study in dogs and in 4-week
`toxicity studies in rats. As with the threo isomer, bloody
`diarrhea was observed in dogs, but the allo isomer was > 10-
`fold more toxic when compared on either a dose level or
`plasma exposure basis. In the rat studies, lung histiocytosis
`and thrombocytopenia were observed as had been seen with
`the threo compound, with the allo compound toxic at > 10-
`fold lower dosage. In addition, the other profound toxicities
`that were observed in dogs with the threo compound (e.g.,
`anemia, splenomegaly, and mortality with multiple organ
`pathology) were observed with the allo compound in rats. As
`a result of these findings, development of both compounds
`was discontinued in early 2001.
`
`DPP8/9 TOXICITY STUDIES
`
`The toxicities observed with the threo and allo comp-
`ounds deepened our concern about the potential safety of this
`mechanism. However, the finding that the allo isomer was
`approximately 10-fold more toxic in rats and dogs, despite
`having comparable pharmacodynamic activity and pharma-
`cokinetics in both species, suggested that these toxicities
`were likely not due to DPP-4 inhibition, but instead were
`potentially due
`to off-target activity. In
`this regard,
`subsequent to the initiation of our program, it had become
`increasingly clear that DPP-4 was a member of larger family
`of ‘DPP-4 activity- and/or structure-homologues’ (DASH)
`proteins, enzymes that are unified by their common post-
`proline cleaving serine dipeptidyl peptidase mechanism [21].
`Enzymes that had recently been described included quiescent
`cell proline dipeptidase (QPP) (aka DPP7) [22], DPP8 [23],
`DPP9 [24], and fibroblast activation protein (FAP) [25]. As
`the functions of these enzymes were unknown, determining
`the selectivity of our inhibitors was a key element of our
`medicinal chemistry program, and thus counterscreens for
`these enzymes were developed.
`The selectivity of the allo and threo compounds was
`determined in the DASH family counterscreens, as well as in
`
`strictly glucose-dependent manner, little or no risk of
`hypoglycemia was anticipated. Second, no weight gain was
`expected with DPP-4 inhibitors. Finally, rodent studies with
`GLP-1 analogs had demonstrated a role for this peptide in
`the regulation of b -cell mass [10]; if these findings translated
`to the clinic, there was the potential that DPP-4 inhibitors
`could have long-term beneficial effects on b-cell function.
`At the onset of this program, there were also several
`concerns regarding potential safety issues for this class.
`DPP-4 is a type II membrane bound cell surface protein that
`is ubiquitously expressed, and like many other cell surface
`molecules, DPP-4 had been implicated in a wide range of
`biological functions. Two potential issues were of most
`concern: first, DPP-4 is identical to the T cell activation
`marker CD26, and data in model systems suggested a
`potential co-stimulatory role for this enzyme in T cell
`activation [11]. Moreover, there were reports that some DPP-
`4 inhibitors (Lys [Z(NO2)] pyrrolidide and related comp-
`ounds) had several effects on immune cells, including
`inhibition of proliferation [12]. Second, DPP-4 had been
`shown to cleave a number of immunoregulatory, endocrine,
`and neurological peptides in vitro [13]. While some comfort
`was later provided by the report that DPP-4 deficient mice
`develop normally, and are healthy [14], a finding that we
`subsequently confirmed [15], we were acutely aware of the
`potential for mechanism-based toxicities, and, when pos-
`sible, exploited opportunities to address these issues as the
`medicinal chemistry program progressed, as described
`below.
`
`PROBIODRUG LICENSING EXPERIENCE
`
`When we initiated our internal screening and medicinal
`chemistry program, two compounds were already advancing
`through human clinical
`trials, Probiodrug’s
`isoleucyl
`thiazolidide (1) and NVP-DPP728 (3) from Novartis (Fig. 1)
`[16,17]. Thus, in order to “jump start” our internal program,
`in late 2000 we elected to in-license L-threo-isoleucyl
`thiazolidide (P32/98) and its allo stereoisomer (L-allo-
`isoleucyl thiazolidide, 2). In single dose pharmacodynamic
`studies, P32/98 had been shown to be well tolerated,
`increased active GLP-1, and reduced glycemic excursion
`following food or glucose intake in normal volunteers [18].
`In addition, Probiodrug reported enhanced insulin secretion
`and improved glucose tolerance in single dose studies in a
`small number of diabetic patients [19].
`
`M e
`
`O
`
`Me
`
`O
`
`Me
`
`Me
`
`N
`
`NH2
`
`S
`
`N
`
`NH2
`
`S
`
`allo-Ile-thiazolidide (2)
`
`O
`
`CN
`
`N
`
`HN
`
`NH
`
`N
`
`P32/98 (1)
`
`NC
`
`NVP-DPP728 (3)
`Fig (1). Early DPP-4 inhibitors.
`
`Merck Exhibit 2114, Page 2
`Mylan Pharmaceuticals Inc. v. Merck Sharp & Dohme Corp.
`IPR2020-00040
`
`

`

`Discovery of JANUVIA™ (Sitagliptin), a Selective Dipeptidyl Peptidase IV
`
`Current Topics in Medicinal Chemistry, 2007, Vol. 7, No. 6 559
`
`an in-house panel of other proteases and by MDS Pharma
`Services (PanLabs) in a panel of 170 receptor and enzyme
`assays. No significant activity (IC50 < 100 m M) was observed
`in any of the in-house protease and PanLabs assays, with the
`exception of the sigma s 1 receptor for the allo compound (Ki
`= 42 mM). However, for DPP-4 related dipeptidyl peptidases,
`inhibition was not only observed for DPP-4, but also for the
`closely related dipeptidyl peptidases, QPP, DPP8, and DPP9.
`Both the allo and threo isomers showed comparable QPP
`inhibition activity (IC50 = 18 mM and 14 mM, respectively);
`however, the potency for inhibition of DPP8 and DPP9
`differed by 5- to 10-fold, with the allo isomer being more
`potent in each case (220 nM vs. 2200 nM for DPP8 and 320
`nM vs. 1600 nM for DPP9) [20]. Since these differences in
`inhibition of DPP8/DPP9 were consistent with the observed
`differences in dose necessary to produce toxicity, we
`hypothesized that inhibition of DPP8 and/or DPP9 was
`responsible for the observed toxicities of both compounds in
`preclinical species.
`To obtain evidence that DPP8/9 inhibition was respon-
`sible for the toxicities observed with the allo and threo
`isomers, DPP-4, QPP, and DPP8/9 selective compounds 4, 5,
`and 6 (Fig. 2), respectively, were identified and evaluated in
`2 week rat toxicity studies and in acute dog tolerability
`studies [20]. The results from these studies showed a
`remarkable similarity between the effects produced by the
`DPP8/DPP9 selective inhibitor and the allo compound
`
`(Table 1). In rats, the DPP8/9 inhibitor produced alopecia,
`thrombocytopenia, reticulocytopenia, enlarged spleen, multi-
`organ histopathological changes, and mortality. In dogs, the
`DPP8/9 inhibitor produced gastrointestinal toxicity. The
`QPP inhibitor produced reticulocytopenia in rats only, and
`no toxicities were noted in either species for the selective
`DPP-4 inhibitor. These results provided compelling evidence
`that inhibition of DPP8/9, but not selective DPP-4 inhibition,
`is associated with multi-organ toxicities in preclinical
`species.
`There were two major reasons that we had a high level of
`confidence in this conclusion. First, at least two structurally
`distinct compounds that inhibit DPP8/9 showed remarkably
`similar toxicities in rats and dogs. As noted above, the
`DPP8/9 inhibitor is highly selective over all other proline
`specific enzymes, and inhibition of the allo compound is
`limited to DPP-4, DPP8/9, and weak inhibition of QPP. We
`also showed that the DPP8/9 inhibitors produce similar
`toxicities in DPP-4-deficient mice and wild type mice,
`establishing that the observed toxicities were not due to
`inhibition of DPP-4 [20]. Second, the degree of toxicity
`observed with the allo and threo compounds correlated with
`their affinity for DPP8/9.
`With the finding that DPP8/9 could produce a variety of
`toxicities in vivo, we hypothesized that some of immune
`effects that had been observed with Lys[Z(NO2)]–pyrrolidide
`and related compounds [13] could instead be due to
`
`F
`
`F
`
`NH2
`
`O
`
`N
`
`N
`
`N
`
`N
`
`CF3
`
`4, DPP-4 selective
`DPP-4 IC50 = 27 nM
`
`M e
`
`Me
`
`O
`
`N
`
`NH2
`
`5, DPP8/9 selective
`DPP8 IC 50 = 38 nM
`DPP9 IC 50 = 55 nM
`
`O
`
`NS
`O
`
`O
`
`N
`
`NH2
`
`S
`
`I
`
`6, QPP selective
`QPP IC 50 = 19 nM
`
`Fig. (2). DPP-4, DPP8/9 and QPP selective inhibitors used in comparative toxicity studies.
`
`Table 1. Comparative Toxicity Studies in Rats (2 Weeks of Treatment at Doses of 10, 30, 100 mg/kg/Day) and Dogs (Single Dose, 10
`mg/kg PO) with Selective Inhibitors 4, 5, and 6. A Check ((cid:214)) Indicates the Toxicity was Observed. Historical Data from
`threo and allo Isoleucyl Thiazolidide Safety Studies are Shown for Comparison
`
`Species
`
`Toxicity
`
`DPP-4 Selective (4)
`
`DPP8/9 Selective (5)
`
`QPP Selective (6)
`
`threo-Ile thia (1)
`
`allo-Ile thia (2)
`
`Rat
`
`Alopecia
`
`Thrombocytopenia
`
`Anemia
`
`Reticulocytopenia
`
`Splenomegaly
`
`Mortality
`
`Dog
`
`Bloody diarrhea
`
`anot determined
`
`(cid:214) (cid:214)
`
`(cid:214) (cid:214) (cid:214)
`
`(cid:214) (cid:214)
`
`n.d.a
`n.d.
`(cid:214) (cid:214)
`
`(cid:214) (cid:214)
`
`(cid:214) (cid:214) (cid:214)
`
`Merck Exhibit 2114, Page 3
`Mylan Pharmaceuticals Inc. v. Merck Sharp & Dohme Corp.
`IPR2020-00040
`
`(cid:214)
`

`

`560 Current Topics in Medicinal Chemistry, 2007, Vol. 7, No. 6
`
`Thornberry and Weber
`
`inhibition of DPP-8/9, and we proceeded to assess the
`selectivity of these compounds. We found that they have
`greater intrinsic potency against DPP8 and DPP9 than DPP-4
`[20]. Moreover, we discovered that the DPP8/9 selective
`inhibitor, but not the selective DPP-4 inhibitor, attenuated
`proliferation and IL-2 release in human in vitro models of T
`cell activation [20]. These results strongly suggest that
`proteolytic activity is not required for the putative co-
`stimulatory function of DPP-4/CD26, and that immunolo-
`gical effects previously observed with several DPP-4
`inhibitors compounds in preclinical models were likely due
`to off-target inhibition of DPP8/9. This result provided a
`greater level of confidence that inhibition of DPP-4 would
`not result in compromised immune function.
`While the significance to human safety is unknown, the
`finding that DPP8/9 inhibition produces profound toxicity in
`preclinical species, and is also likely responsible for effects
`on immune function that have been previously attributed to
`DPP-4, prompted us to refocus our program on the discovery
`of highly selective DPP-4 inhibitors.
`
`INITIAL MEDICINAL CHEMISTRY EFFORTS:
`CYCLOHEXYLGLYCINE LEAD
`
`We initiated our medicinal chemistry program prior to
`the discovery of the preclinical toxicity associated with
`inhibition of DPP8/9. Thus, we focused initially on identify-
`ing a “Best in Class” compound by improving upon the
`potency of isoleucyl thiazolidide (IC50 = 420 nM in our
`hands) and the short half-life of NVP-DPP728, which we
`found to be only ~15 min in rats. We thought the latter issue
`might be due to chemical instability. This compound, like
`many of the most potent DPP-4 inhibitors, contained a
`reactive electrophile (a nitrile in this case) that was believed,
`and a close analog later shown, to form a covalent bond with
`the active site serine [26]. This electrophile is six atoms
`away from an amine, perfectly set up to cyclize. In order to
`improve upon the half-life, we chose from the onset to focus
`on structures which lacked this electrophile, even though we
`knew these structures had been generally shown to be less
`potent.
`While we were waiting for results from our internal
`screening efforts, we initiated SAR studies based on the
`known a-amino acid derived inhibitors. The most potent
`inhibitor reported in the literature that did not contain an
`electrophile was cyclohexylglycyl thiazolidide (7, Table 1),
`discovered by chemists at Ferring [27]. With an IC50 of 89
`nM in our hands, this compound was already 4-fold more
`potent than Probiodrug’s related clinical candidate. In order
`to further improve the potency and identify proprietary
`compounds, substitution on the cyclohexyl ring was explored
`[28,29]. In particular, amides, carbamates and sulfonamides
`at the 4-position on the cyclohexyl ring provided compounds
`such as 8, 9 and 10 (Table 2) with improved potency and
`oral bioavailabilities in rats of 81%, 76% and 46%,
`respectively [27].
`Once the toxicity of the thiazolidide derivatives emerged,
`but before we had assays in hand for DPP8 and DPP9, we
`wondered whether the thiazolidine ring might be responsible
`for the observed toxicity. If the thiazolidine ring opened in
`
`Table 2.
`
`4-Substituted Cyclohexylglycine Analogs
`
`R
`
`O
`
`N
`
`NH2
`
`X
`
`R
`
`H
`
`(3,4-di-F-Ph)CONH-
`
`PhCH2OCONH-
`
`(4-CF3O-Ph)SO2NH-
`
`(3,4-di-F-Ph)CONH-
`
`PhCH2OCONH-
`
`(4-CF3O-Ph)SO2NH-
`
`X
`
`S
`
`S
`
`S
`
`S
`
`CH2
`
`CH2
`
`CH2
`
`(3,4-di-F-Ph)CONH-
`
`(S)-CHF
`
`PhCH2OCONH-
`
`(S)-CHF
`
`(4-CF3O-Ph)SO2NH-
`
`(S)-CHF
`
`DPP-4
`
`IC50 (nM)
`
`89
`
`54
`
`25
`
`22
`
`190
`
`94
`
`89
`
`54
`
`56
`
`36
`
`Compound
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`vivo, a free thiol would be revealed, which could be reactive
`and lead to toxicity. To eliminate this possibility, we shifted
`our focus to non-sulfur containing heterocycles. The simple
`pyrrolidide derivatives were generally less potent (compare,
`for example, compounds 11 vs. 8, 12 vs. 9, and 13 vs. 10,
`Table 2); however, the addition of fluorine to the ring
`resulted compounds with increased potency. Indeed the (3S)-
`3-fluoropyrrolidine amides 14, 15, and 16 (Table 2) are
`nearly equi-potent to the corresponding thiazolidide analogs
`[30].
`The 4-trifluoromethoxybenzenesulfonamide derivative
`16 has excellent pharmacokinetic properties across species,
`with a half-life of 4 h, 12 h and 5 h in rats, dogs and rhesus
`monkeys, respectively, and oral bioavailability of 37% to
`89% [30]. It was profiled extensively as a potential
`preclinical development candidate. Only reticulocytopenia at
`100 mg/kg/day was observed in a 2-week toxicity study in
`rats, and it was clean in acute tolerability studies in dogs
`[Lankas, G.; unpublished results]. Despite an attractive
`profile, further work on this compound was discontinued due
`to unacceptable levels of DPP8 and DPP9 inhibition (IC50 =
`1400 nM and 1700 nM, respectively).
`Once the potential toxicity associated with inhibition of
`the DPP8 and DPP9 enzymes was discovered, our goal was
`to identify an inhibitor with a >1000-fold window for DPP-4
`inhibition over inhibition of these enzymes. A clue for how
`to achieve specificity came from a pair of positional scan-
`ning libraries that were developed for these aminopeptidases
`[31]. Each library consisted of dipeptidyl aminomethyl-
`coumarin substrates. In the first “P1” sublibrary, each well
`contained a spatially addressed amino acid at the P1 position
`coupled to an isokinetic mixture of amino acids at P2. In the
`
`Merck Exhibit 2114, Page 4
`Mylan Pharmaceuticals Inc. v. Merck Sharp & Dohme Corp.
`IPR2020-00040
`
`

`

`Discovery of JANUVIA™ (Sitagliptin), a Selective Dipeptidyl Peptidase IV
`
`Current Topics in Medicinal Chemistry, 2007, Vol. 7, No. 6 561
`
`O
`
`N
`
`H
`
`NH2
`
`HN
`
`O
`
`S O
`
`R
`
`17, R = COOH
`DPP-4 IC 50 = 15 nM
`DPP8 IC50 = 2800 nM
`
`18, R = CF3CH2SO2NH-
`DPP-4 IC50 = 2.6 nM
`DPP8 IC50 = 15,000 nM
`
`Fig. (4). DPP-4 inhibitors containing acidic functionality.
`
`This compound was >100-fold selective over both DPP8 and
`DPP9. Potency could be enhanced by appending a second
`phenyl ring. The 4-fluorobiphenyl derivative 20 is a 64 nM
`DPP-4 inhibitor with excellent selectivity over both DPP8
`and DPP9, though both 19 and 20 show decreased selectivity
`over QPP [32]. By incorporating polar functionality in place
`of the methyl group to give N,N-dimethylamide 21, both
`potency and selectivity over QPP were improved [33].
`inhibitor 21 has excellent pharmacokinetic
`While
`properties across species, its efficacy in an oral glucose
`tolerance test in mice was less than anticipated based on the
`exposures obtained and its inhibition of mouse DPP-4. This
`observation was attributed to its high plasma protein binding
`[33]. Replacement of the terminal phenyl ring with a
`heterocycle provided compounds with reduced serum shift,
`leading to triazolopyridine 22 Fig. (5) which was chosen for
`extensive preclinical evaluation [34]
`FROM SCREENING HITS TO b-AMINO ACID LEAD
`While medicinal chemistry efforts in the a-amino acid
`series were ongoing, screening of the Merck sample
`
`second “P2” sublibrary, each well contained a spatially
`addressed amino acid at the P2 position coupled to an
`isokinetic mixture of amino acids at P1. Results for cleavage
`of the libraries by DPP-4 and DPP8 are summarized in Fig.
`(3). While both enzymes showed a strong preference for
`cleavage of substrates with a proline at the P1 position, DPP-
`4 was much more promiscuous at P2. In particular, dipeptides
`containing acidic amino acids such as glutamic acid were
`readily cleaved by DPP-4 whereas the rate of cleavage of
`these dipeptides by DPP8 was greatly reduced. We reasoned
`that incorporation of such acidic functionality at the P2
`position of our inhibitors could provide analogs with
`improved selectivity.
`A variety of acidic derivatives were prepared. The two
`most selective compounds in this series, 17 [E. Parmee,
`unpublished results] and 18 [28], are illustrated in Fig. (4).
`Both are >100-fold selective for DPP-4 over DPP8; how-
`ever, both suffer from poor oral bioavailability in rats (<1%).
`Further optimization did not yield selective compounds with
`improved pharmacokinetic properties and this series was put
`on hold.
`a-AMINO ACID SERIES REVISITED
`The a-amino acid series was ultimately re-examined
`following the discovery of sitagliptin. Because the allo
`isomer of isoleucyl thiazolidide was a more potent inhibitor
`of DPP8 and DPP9 than the threo isomer, we wondered
`whether we could improve selectivity by incorporating a
`“threo” bias into this series, an approach that was not
`possible with the symmetrical cyclohexylglycine derivatives.
`As shown in Fig. (5), when the ethyl sidechain of isoleucyl
`thiazolide was replaced with phenyl to provide b-methyl
`phenylalanine analog 19, potency decreased by ~2-fold but
`selectivity was greatly improved [J. Xu, unpublished results].
`
`(a)
`
`(b)
`
`Fig. (3). Cleavage of peptide scanning libraries by (a) DPP-4 and (b) DPP8.
`
`Merck Exhibit 2114, Page 5
`Mylan Pharmaceuticals Inc. v. Merck Sharp & Dohme Corp.
`IPR2020-00040
`
`

`

`562 Current Topics in Medicinal Chemistry, 2007, Vol. 7, No. 6
`
`Thornberry and Weber
`
`19, R = Ph
`DPP-4 IC50 = 970 nM
`DPP8 IC5 0 > 100,000 nM
`QPP IC5 0 = 12,000 nM
`
`21, R = CONMe2
`DPP-4 IC5 0 = 12 nM
`DPP8 IC50 > 100,000 nM
`QPP IC50 = 45,000 nM
`
`Me
`
`O
`
`R
`
`N
`
`NH2
`
`S
`
`R
`
`O
`
`N
`
`NH2
`
`F
`
`Me
`
`N
`
`Me
`
`O
`
`O
`
`NH2
`
`N
`
`F
`
`F
`
`1, R = Et
`DPP-4 IC5 0 = 420 nM
`DPP8 IC 50 = 2200 nM
`QPP IC 50 = 14,000 nM
`
`20, R = Me
`DPP-4 IC5 0 = 64 nM
`DPP8 IC 50 = 88,000 nM
`QPP IC 50 = 2700 nM
`
`22
`DPP-4 IC5 0 = 8.8 nM
`DPP8 IC 50 > 100,000 nM
`QPP IC 50 > 100,000 nM
`
`F
`
`N
`
`N N
`
`N
`
`NH
`
`NH
`
`O
`
`Cl
`
`O
`
`NHSO2Me
`
`NN
`
`O
`
`HN
`
`O
`
`N
`
`CH3
`
`23, IC5 0 = 1700 nM
`
`HN
`
`O
`
`N
`
`NH2
`
`O
`
`24, IC5 0 = 1900 nM
`
`NH
`
`N
`
`O
`
`N
`
`25, IC5 0 = 11,000 nM
`
`H2 N
`
`Fig. (6). DPP-4 inhibitor screening hits.
`
`Fig (5). b-Substituted a
`
`-amino acid derived DPP-4 inhibitors.
`
`collection was completed. Surprisingly few hits were
`identified. Only three leads, xanthine derivative 23, proline
`amide 24, and piperizine 25, were deemed suitable for
`follow-up (Fig. 6). Xanthine lead 23 and several related
`micromolar hits originated from a commercially available
`screening library. An additional library of compounds was
`prepared in-house for initial optimization; however, no
`increase in potency was seen so this lead was not further
`pursued. It is interesting to note that other pharmaceutical
`companies have subsequently reported structurally similar
`DPP-4 inhibitors [35,36], highlighting a major issue associa-
`ted with screening non-proprietary compound libraries.
`Proline amide screening hit 24 was originally prepared
`for our thrombin inhibitor program, and was in fact a much
`more potent inhibitor of that enzyme than of DPP-4 (IC50s =
`52 nM and 1900 nM, respectively). Initially we believed,
`naively, that the proline moiety mapped to the thiazolidine or
`pyrrolidine ring in the a-amino acid series. Thus, we quickly
`b-amino acyl group with isoleucine and
`replaced the
`cyclohexylglycine side chains, but this led to a 2- to 3-fold
`decrease in activity. Because the proline amide moiety could
`be replaced by a thiazolidine ring without significant loss of
`activity, much of the early left-hand side SAR was
`conducted in this truncated series [37]. Shortening or
`lengthening the distance between the left-hand side phenyl
`ring and
`the amino group
`led
`to compounds with
`dramatically decreased potency, as did replacement of the
`phenyl ring with cyclopenyl, cyclohexyl, or heterocyclic
`groups. In fact the only change that was not detrimental to
`potency was the addition of fluorine to the phenyl ring. In
`particular, the 2-fluoro derivative was 3-fold more potent,
`and the 2,5-difluoro analog showed a 10-fold improvement.
`The 2,4,5-trifluoro derivative was the most potent compound
`prepared in this truncated series (IC50 = 120 nM).
`
`Merck Exhibit 2114, Page 6
`Mylan Pharmaceuticals Inc. v. Merck Sharp & Dohme Corp.
`IPR2020-00040
`
`

`

`Discovery of JANUVIA™ (Sitagliptin), a Selective Dipeptidyl Peptidase IV
`
`Current Topics in Medicinal Chemistry, 2007, Vol. 7, No. 6 563
`
`Early SAR studies on the right hand side of the molecule
`showed
`that
`the amide could be
`replaced by
`the
`corresponding acid without loss of potency. An acetic acid
`scan on the phenyl ring provided 4-phenylacetic acid
`derivative 26 (Fig. 7, IC50 = 510 nM) [38], the first
`subnanomolar compound in this series, which also proved to
`be devoid of thrombin activity (IC50 > 500,000 nM). As we
`saw in the truncated series, the addition of a 2-fluoro
`substituent to the phenyl ring led to an increase in potency.
`Conversion of the acetic acid moiety in the resultant analog
`27 to lactic acid gave 28, a 12 nM inhibitor. More lipophilic
`groups in place of methyl on the acid moiety led to a further
`potency enhancement, with the isopropyl moiety being
`preferred. The addition of a second fluorine to the phenyl
`ring provided 29, a subnanomolar inhibitor.
`Compound 29 had a number of very desirable features.
`First, it represented a truly unique class of inhibitors so we
`anticipated no patent issues. Second, it possessed excellent
`selectivity (‡ 100,000-fold over all related peptidases
`assayed, including DPP8 and DPP9). Unfortunately, oral
`bioavailability in this series was particularly poor. In rats this
`was traced to low oral absorption.
`While work on the proline amide series was proceeding,
`piperazine lead 25, an 11 mM DPP-4 inhibitor, was also
`being explored. This hit was originally prepared for our
`melanocortin 4 receptor (MC4-R) agonist program; however,
`it proved devoid of MC4-R activity. When SAR studies on
`the right hand side of the molecule failed to lead to
`substantial improvements in potency, attention then turned to
`the
`left hand side. The phenethylamine moiety was
`reminiscent of the proline amide lead, so fluorination of the
`phenyl ring, which increased potency in that series, was
`examined. The 3,4-difluorophenyl analog (30, Fig. 8)
`showed a 2-fold increase in potency [39]. Conversion of the
`side chain, essentially a reduced phenylalanine derivative, to
`the corresponding phenylalanine amide gave an inactive
`compound; however, homophenylalanine 31 showed a 100-
`
`fold increase in potency. The 2-fluoro derivative containing
`the more active (R)-benzyl group (32) is a 14 nM DPP-4
`inhibitor. The entire right hand side moiety could be
`removed to give piperazine 33, a much lower molecular
`weight analog, with only a 10-fold loss of potency.
`Optimization of the phenyl ring paralleled that of the proline
`amide series and provided trifluorophenyl derivative 34, the
`most potent compound in this series. Like the proline amide
`series, oral bioavailability of this series was poor, even for
`the low molecular weight compounds such as 33. In this
`case, the problem was traced to extensive metabolism on the
`piperazine ring.
`
`IMPROVEMENTS IN PHARMACOKINETIC PRO-
`PERTIES: THE FUSED PIPERAZINES
`
`In order to stabilize the piperazine ring toward metabo-
`lism, we elected to focus on bicyclic piperazine replace-
`ments. Initial work was done in the simplified, unsubstituted
`piperazine series. Parent compound 35 (Fig. 9) was
`converted to the corresponding triazolopiperidine 36, leading
`to an increase in potency [40]. Ethyl analog 37 was slightly
`more potent, and more importantly, completely stable to
`metabolism in hepatocytes. Despite the improved metabolic
`stability, this compound still had poor oral bioavailability in
`rats (2%), though oral bioavailability in dogs was more
`acceptable (33%). In rats, the low %F was due to low,
`variable absorption. Hepatic extraction was low (10-20%),
`reflecting the increased hepatic stability observed in vitro.
`The solution to improved absorption proved to be quite
`simple, though entirely empirical. We discovered that
`replacing the ethyl side chain with a trifluoromethyl group
`led to an improvement in oral absorption. The oral
`bioavailability of trifluoromethyl analog 38 in rats was 44%.
`With this discovery in hand, all that remained was to adjust
`the fluoro substituents to improve potency. This provided
`difluoro- and trifluorophenyl derivatives 4 and 39 (Fig. 10).
`
`CO2 H
`
`O
`
`CO2H
`
`Me
`
`HN
`
`O
`
`N
`
`NH2
`
`O
`
`27, IC 50 = 54 nM
`
`HN
`
`O
`
`N
`
`NH2
`
`O
`
`28, IC 50 = 12 nM
`
`F
`
`F
`
`CO2H
`
`O
`
`CO2H
`
`HN
`
`O
`
`N
`
`NH2
`
`O
`
`26, IC 50 = 510 nM
`
`HN
`
`O
`
`N
`
`NH2
`
`O
`
`F
`
`F
`
`29, IC5 0 = 0.4 nM
`Fig. (7). Optimization of the b-amino acid proline amide series.
`
`Merck Exhibit 2114, Page 7
`Mylan Pharmaceuticals Inc. v. Merck Sharp & Dohme Corp.
`IPR2020-00040
`
`

`

`564 Current Topics in Medicinal Chemistry, 2007, Vol. 7, No. 6
`
`Thornberry and Weber
`
`NHSO2M e
`
`NHSO2Me
`
`