Preparation of saxagliptin, a novel DPP-IV inhibitor

Article abstract of DOI:10.1021/op900226j

The commercial-scale synthesis of the DPP-IV inhibitor, saxagliptin (1), is described from the two unnatural amino acid derivatives 2 and 3. After the deprotection of 3, the core of 1 is formed by the amide coupling of amino acid 2 and methanoprolinamide

Full text of DOI:10.1021/op900226j

Organic Process Research & Development 2009, 13, 1169–1176
Preparation of Saxagliptin, a Novel DPP-IV Inhibitor
Scott A. Savage,* Gregory S. Jones, Sergei Kolotuchin, Shelly Ann Ramrattan, Truc Vu, and Robert E. Waltermire
Bristol Myers Squibb Co., Research and DeVelopment, One Squibb DriVe, New Brunswick, New Jersey 08903, U.S.A.
Scheme 1. Retrosynthetic analysis
Abstract:
The commercial-scale synthesis of the DPP-IV inhibitor, saxaglip-
tin (1), is described from the two unnatural amino acid derivatives
2 and 3. After the deprotection of 3, the core of 1 is formed by the
amide coupling of amino acid 2 and methanoprolinamide 4.
Subsequent dehydration of the primary amide and deprotection
of the amine affords saxagliptin, 1. While acid salts of saxagliptin
have proven to be stable in solution, synthesis of the desired free
base monohydrate was challenging due to the thermodynamically
favorable conversion of the free amine to the six-membered cyclic
amidine 9. Significant process modifications were made late in
development to enhance process robustness in preparation for the
transition to commercial manufacturing. The impetus and ration-
ale for those changes are explained herein.
3 have been prepared on commercial scale by a number of
contract manufacturers.6 While the strategy of early drug
deliveries was rapid syntheses to support preclinical activities
and phase I clinical trials, as saxagliptin entered phase II, a
greater emphasis was placed on defining and demonstrating a
commercially viable synthetic process. Prior to process valida-
tion at the commercial manufacturing facility, further optimiza-
tion was performed to increase process robustness and decrease
EHS (environmental health and safety) impact. This report
describes the final processes demonstrated during the validation
campaign and highlights some of the key considerations during
development.
Introduction
A recent epidemiological study suggests that the total number
of people with diabetes worldwide is expected to double to 366
million by the year 2030.1 This epidemic has spurred much
research into alternative diabetes treatments exploiting the
incretin effect, including the inhibition of dipeptidyl peptidase
IV (DPP-IV).2 DPP-IV is the primary enzyme responsible for
degradation of incretins, such as glucagon-like peptide-1 (GLP-
1), which is a hormone responsible for the glucose-dependent
stimulation of insulin in the human body. DPP-IV inhibitors
serve as effective glucose regulators primarily by increasing
the endogenous concentration of GLP-1. The first of this new
class of pharmaceuticals was approved by the FDA in 2006.3
DPP-IV inhibitors have been clinically proven to lower blood
glucose levels, increase glucose tolerance, and improve insulin
response in patients with type 2 diabetes mellitus.2,4 Saxagliptin
(Onglyza, Bristol-Myers Squibb, AstraZeneca) is a highly
potent, orally available reversible dipeptidyl peptidase IV (DPP-
IV) inhibitor recently approved by the FDA for the treatment
of type 2 diabetes mellitus.5
Results and Discussion
Scheme 1 shows the retrosynthetic plan for the preparation
of saxagliptin, 1, from the two proposed regulatory starting
materials 2 and 3. It is readily apparent that the cornerstone of
this synthesis is the amide coupling. While many suitable amide-
or peptide-coupling protocols are known,7 our approach had to
balance the typical hurdles found in commercial syntheses: API
quality, cost of goods, yield, processing time, and process
robustness. In addition, a change in the API final form from
the benzoate salt to the free base monohydrate 1 required
extensive changes to the final step of the synthesis. Due to the
thermodynamically favorable cyclization of the free base
monohydrate 1 form of the API to amidine 9, detailed studies
of the unit operations of the final step of the synthesis were
required to define a robust process.
Deprotection of Methanoprolinamide. Hydrochloric acid
was initially chosen for the removal of the Boc group to convert
the methanoprolinamide 3 to 4. The hygroscopic nature of the
resulting HCl salt rendered it impractical to handle on larger
scales. While many acids proved adequate for this conversion,
methanesulfonic acid (MSA) provided a crystalline salt with
The initial route was a 15-step, convergent synthesis5a having
the retrosynthetic plan shown in Scheme 1. Compounds 2 and
* scott.savage@bms.com.
(1) Wild, S.; Roglic, G.; Green, A.; Sicree, R.; King, H. Diabetes Care
2004, 27, 1047–1053.
(2) (a) Idris, I; Donnelly, R. Diabetes, Obes. Metab. 2007, 9, 153–165.
(b) von Geldern, T. W.; Trevillyan, J. M. Drug DeV. Res. 2006, 67,
627–642.
(3) FDA News Release P06-169, 2006.
(6) (a) Vu, T. C.; Brzozowski, D. B.; Fox, R.; Godfrey, J. D.; Hanson,
R. L.; Kolotuchin, S. V.; Mazzullo, J. A.; Patel, R. N.; Wang, J.; Wong,
K.; Yu, J.; Zhu, J.; Magnin, D. R.; Augeri, D. J.; Hamann, L. G.
Methods and Compounds Producing Dipeptidyl Petidase IV Inhibitors
and Intermediates Thereof. Int. Pub. Num. WO2004052850, 2004. (b)
Hanson, R. L.; Goldberg, S. L.; Brzozowski, D. B.; Tully, T. P.;
Cazzulino, D.; Parker, W. L.; Lyngberg, O. K.; Vu, T. C.; Wong,
M. K.; Patel, R. N. AdV. Synth. Catal. 2007, 349, 1369–1378.
(7) Montalbetti, C. A. G. N.; Falque, V. Tetrahedron 2005, 61, 10827–
10852.
(4) Langley, A. K.; Suffoletta, T. J.; Jennings, H. R. Pharmacotherapy
2007, 27, 1163–1180.
(5) (a) Augeri, D. J.; Robl, J. A.; Betebenner, D. A.; Magnin, D. R.;
Khanna, A.; Robertson, J. G.; Wang, A.; Simpkins, L. M.; Taunk, P.;
Huang, Q.; Han, S.-P.; Abboa-Offei, B.; Cap, M.; Xin, L.; Tao, L.;
Tozzo, E.; Welzel, G. E.; Egan, D. M.; Marcinkeviciene, J.; Chang,
S. Y.; Biller, S. A.; Kirby, M. S.; Parker, R. A.; Hammann, L. G.
J. Med. Chem. 2005, 48, 5025–5037. (b) Bristol-Myers Squibb and
Astra Zeneca News Release, July 31st, 2009.
10.1021/op900226j CCC: $40.75  2009 American Chemical Society
Published on Web 10/16/2009
Vol. 13, No. 6, 2009 / Organic Process Research & Development
•
1169

tries.11 The original synthesis of saxagliptin utilized 1-[3-
(dimethylamino)propyl]-3-ethylcarbodiimide hydrochloride (EDC)
and 1-hydroxybenzotriazole hydrate (HOBt) for the coupling
to form amide 5. In an effort to identify a lower-cost alternative
to EDC, several alternative conditions were evaluated.12 While
carbodiimides and methanesulfonyl chloride proved effective,
many other typical coupling reagents failed to yield clean
conversions to amide 5. The first pilot-plant campaign utilized
methanesulfonyl chloride (MsCl), where the intermediate mixed
anhydride was initially formed as a solution in ethyl acetate in
the presence of diisopropylethylamine (DIPEA). Once formed,
methanoprolinamide 4 and catalytic HOBt were added to
complete the coupling.6a Optimization of both solvent selection
and workup conditions for this process provided solution yields
of 5 in 95-99%.
As saxagliptin proceeded through development, it became
apparent that the economics of scale would drive down the cost
of EDC into the competitive realm.13 Therefore, we fully
developed both the MsCl and EDC routes for the coupling
reaction to allow a complete, comparative evaluation. Ulti-
mately, the exceptional yield and robustness of the EDC/HOBt
process outweighed the small impact on the cost of API
production and was selected for the commercial route. It is
worth noting that the poor chromophores of 2 and 4, along with
the incompatibility of the activated ester to HPLC conditions,
required the coupling reaction to be monitored by the increase
in the concentration of the product, amide 5, as opposed to the
more typical percent conversion of starting materials to product.
The EDC-activated ester of 2 proved to be somewhat unstable
when monitored by react-IR and required an equivalent of HOBt
hydrate to provide optimal yield. The solution yield of amide
5 was found to be directly proportional to the equivalents of
HOBt charged. For example, 0.5 equiv of HOBt generated an
approximately 50% solution yield. This instability also pre-
vented the replacement of HOBt with less reactive alternative
reagents, such as 5-nitro-2-hydroxypyridine,14 as the EDC-
activated ester decomposed at a faster rate than it reacted with
either the HOBt alternative or amine 4.
Scheme 2. Deprotection of methanoprolinamide 3
excellent physical properties for both isolation and storage
(Scheme 2). A broad solvent screen was conducted and revealed
that alcoholic solvents afforded the highest-quality MSA salt
4. Isopropanol was ultimately selected, as it provided high
solubility of the starting material 3, excellent product quality,
and an essentially quantitative yield of 4. Our preference was
to utilize nonalcoholic solvents in combination with MSA, to
avoid the formation of genotoxic methanesulfonate esters.8
However, the purification and yield provided by isopropanol
outweighed the liabilities of using this combination, as extremely
low levels of isopropyl methanesulfonate (IPMS) were observed
in the product 4.9 In fact, levels of IPMS in 4 prepared on
multikilogram (multikg) scales were typically <5 ppm (limit
of detection), despite the presence of ∼1000 ppm in the filtrate
solution. In addition, tolerance studies demonstrated that the
coupling process could withstand >1000 ppm of IPMS in MSA
salt 4 and still produce 7 with no detectable IMPS found by
gas chromatography.
In manufacturing, a mixture of isopropanol and 3 was heated
to 60 °C prior to a controlled addition of MSA. This provided
a nearly addition-controlled reaction during the MSA charge,10
resulting in facile control of off-gassing on scale and preventing
the isobutylene byproduct from overwhelming the plant’s
thermal oxidizer. Analysis of the off-gas indicated that the
expected 1 equivalent of CO2 was accompanied by 0.15
equivalent of isobutylene. As a result of the low water content
of the isopropanol, no tert-butanol was formed, and the majority
(0.85 equiv) of the tert-butyl cation reacted with isopropanol
to form isopropyl tert-butyl ether. Due to the extremely low
solubility of product 4 in isopropanol, the crystallization
consistently initiated prior to the completion of the MSA charge.
The poor chromophore of the deprotected product and the
heterogeneity of the slurry formed during the reaction made it
difficult to use traditional in-process monitoring by HPLC area
percent conversion. Thus, the conversion of 3 to 4 was
monitored by quantitation of the residual starting material
remaining in the supernatant. Upon reaction completion, typi-
cally 3 hours at 60 °C, the slurry was cooled to 20 °C and the
product isolated on a filter dryer or centrifuge, with isopropanol
used as the wash solvent. Using this process, over 1.7 MT (15
batches) has been prepared to date with batch scales up to 205
kg and an average isolated yield of 94%.
Due to the instability of the activated ester of 2, a general
trend during development of the process was that coupling
conditions that afforded a faster rate of reaction generally gave
higher yields of 5. All orders of addition were evaluated, and
interestingly, it was found that the best process was to charge
all four solids (2, 4, EDC, and HOBt) to the reactor prior to
the sequential addition of solvent and base. Acetonitrile was
found to be the preferred solvent for this reaction and required
only 1.1 L/kg of total solids. This small amount of solvent was
feasible because the use of 2.2 equivalents of DIPEA reacts to
(11) Carey, J. S.; Laffan, D.; Thomson, C.; Williams, M. T. Org. Biol.
Chem. 2006, 4, 2337–2347.
Amide Coupling. Commercial-scale amide couplings are
widespread in the pharmaceutical and biotechnology indus-
(12) Carbodiimides and sulfonyl chlorides (mesyl and tosyl) were the only
two groups of coupling agents found suitable for this conversion.
Examples of failed reagents are: chloroformates, piv chloride, POCl3-
DMF, 2-chloro-4,6-dimethoxy-1,3,5-triazine, N,N′-carbonyldiimida-
zole. Specifically, mixed anhydrides suffered from nonselective
reactivity with amine 4.
(8) Glowienke, S.; Frieauff, W.; Allmendinger, T.; Martus, H.; Suter, W.;
Mueller, L. Mutat. Res. 2005, 581, 23–24.
(9) The combined mother liquor and cake washes had <1000 ppm of IPMS
on scale.
(10) Lab-scale gas evolution studies showed that no significant gas was
evolved until after ∼15% of the MSA was added and that, if the MSA
addition was spread over 1 h or longer, ∼90% of the reaction would
occur during the acid charge. This allowed a high level of control
over the evolution of carbon dioxide and isobutylene.
(13) While kilogram quantities typically cost ∼$2000/kg, manufacturing-
scale quantities typically run an order of magnitude less at $100 to
$200 per kg.
(14) Dunn, P. J.; Hoffmann, W.; Kang, Y.; Mitchell, J. C.; Snowden, M. J.
Org. Process Res. DeV. 2005, 9, 956–961.
1170
•
Vol. 13, No. 6, 2009 / Organic Process Research & Development

Scheme 3
Initially, trifluoroacetic anhydride and pyridine were the
preferred reagents for this reaction. Pyridine played a dual role
as both an initial activating agent for the trifluoroacetic
anhydride and, subsequently, as a base for the trifluoroacetic
acid byproduct. The reaction was quite robust using these
reagents and easily converted >99.5% of amide 5 to the
corresponding nitrile 6 at pilot-plant scales of >100 kg.
However, due to concerns over the odor and toxicity of pyridine
at manufacturing scale, we investigated alternative bases. The
identification of a direct substitute base was not as straightfor-
ward as initially expected, especially as pyridine participates
in the mechanism as an activating agent.16 Table 1 summarizes
a number of bases, or classes of base, that were evaluated and
rejected on the basis of deficiencies in reactivity or general
environmental, health, and safety concerns. Typical tertiary alkyl
amines (DIPEA, triethylamine, N-methylmorpholine, etc.) are
oxidized by trifluoroacetic anhydride via a hydrogen transfer
mechanism, with generation of the toxic byproduct fluoral.17
In addition, 2- and 4-alkyl pyridines are reactive with trifluo-
roacetic anhydride.18 One of the more interesting replacements
considered was potassium trifluoroacetate, as this salt has been
shown to form a dimer with trifluoracetic acid.19 An initial
hypothesis was that this phenomenon might allow potassium
trifluoroacetate to replace pyridine by acting as a buffer for the
trifluoroacetic acid generated during the dehydration. Unfortu-
nately, this system had no tolerance for the slight excess of
TFAA, 10 mol %, required for complete conversion of the
amide and resulted in acylation of the carbamate nitrogen and
loss of the Boc protecting group. While numerous organic and
inorganic bases were considered and screened, 3- and 5-
substituted pyridines were the only class of bases that satisfied
reactivity requirements. Both 3-methoxypyridine and ethyl
nicotinate were found to be effective replacements for pyridine,
but ethyl nicotinate was chosen for its superior toxicity profile20
(very limited toxicity data were available for 3-methoxypyri-
dine) and nonoffending odor.
generate the free amines of 4 and EDC, which are both oils
under these conditions. Even at scales >100 kg, the solids rapidly
dissolve during the DIPEA addition and are completely in
solution within minutes of the addition. While the sequential
addition of acetonitrile and DIPEA was demonstrated multiple
times on approximately 100 kg scale, the following, more robust
process was ultimately developed. A mixture of DIPEA,
acetonitrile (3 L/kg of 2), and a small amount of EtOAc (1
L/kg), which provides a homogeneous solution, was added to
the four solids. Since 2 and EDC react immediately upon
addition of solvent, premixing the acetonitrile, ethyl acetate,
and DIPEA allowed for immediate dissolution of HOBt and
the MSA salt 4, thereby precluding degradation of the EDC
activated ester prior to the DIPEA charge. After addition of
the DIPEA solution, the coupling reaction was >90% complete
after approximately 30 min. Aging the reaction for a total of
2-3 h gave consistent solution yields of >95% of amide 5.
Typical yield losses during the subsequent aqueous work up
were <5%.
Nitrile Formation. Amide 5 is an amorphous glass which
required telescoping the coupling process leading to 5 into the
subsequent dehydration step to form the corresponding nitrile
7 (Scheme 3). After the coupling reaction, ethyl acetate was
added to allow an acidic aqueous workup to remove the basic
byproducts of the process. Not surprisingly, the residual HOBt
remaining in the organics reacts with the trifluoroacetic anhy-
dride (TFAA) used for the amide dehydration, forming the
corresponding trifluoroacetate triazole. While this species is still
an effective dehydrating agent, it proved to have a higher affinity
toward acylation of the adamantane alcohol. This expected side
reaction made complete conversion of the amide to nitrile 6
difficult in the presence of large amounts of residual HOBt.
Taking advantage of the acidic nature of HOBt, two aqueous
potassium bicarbonate washes were utilized to remove 90% of
the HOBt used for the coupling reaction.15 This process
modificationeliminatedtheissueofincompleteamidedehydration.
Because of its lack of nucleophilicity, ethyl nicotinate appears
to have no reactivity with trifluoroacetic anhydride by 1H NMR
and does not play a role as an activating agent in the dehydration
of the amide. With TFAA as the dehydrating reagent (as
opposed to the TFAA-pyridine complex formed in the original
process) the typical conversion of amide 5 to nitrile 6 was
approximately 98%. Even up to 2.3 equiv of TFAA did not
drive the reaction to complete conversion, and larger amounts
were not compatible with the Boc protected amine. The switch
from pyridine to ethyl nicotinate resulted in an approximately
2% yield loss due to lower conversion, but ethyl nicotinate is
far more friendly from an EHS perspective.
The apparently simple change from pyridine to ethyl
nicotinate required an extensive change in the workup. Pilot-
plant batches utilizing pyridine only required a single water
(16) Anthoni, U.; Christensen, D.; Christophersen, C.; Nielsen, P. H. Acta
Chem. Scand. 1995, 49, 203–6.
(17) Schreiber, S. L. Tetrahedron Lett. 1980, 21, 1027.
(18) Kiwase, M.; Teshima, M.; Saito, S.; Tani, S. Heterocycles 1998, 48,
2103.
(15) Potassium bicarbonate allowed for higher concentrations of salts than
sodium bicarbonate and therefore reduced product loss to the aqueous
layer. Though potassium carbonate washes were also effective, they
were not chosen to avoid the possibility of epimerization during the
subsequent distillation.
(19) Clark, J. H.; Emsley, J. J. Chem. Soc., Dalton Trans. 1974, 11, 1125–
9.
(20) LD 50: >2005 mg/kg (rat), cutaneous irritation: nonirritant (rabbit),
ocular irritation: nonirritant (rabbit). See the MSDS at Seppic (Seppic,
Inc., Fairfield, NJ).
Vol. 13, No. 6, 2009 / Organic Process Research & Development
•
1171

Table 1. Pyridine alternatives for dehydration reaction
successful
dehydration
TFAA
compatible
base
comments
tertiary amines
potassium trifloroacetate
phosphates
yes
yes
no
no
redox reaction with TFAA17
no tolerance of excess TFAA
heterogeneous conditions
acylation of alkyl pyridine
odor concerns
yes
yes
no
2- or 4-alkyl pyridines
3- or 3,5-alkyl pyridines
2,2′-bipyridine
na
yes
NA
NA
no
no
no
yes
yes
yes
yes
yes
yes
yes
yes
mutagen
quinoline
mutagen
dimethylaminopyridine
3-acetoxypyridine
3-cyanopyridine
3-methoxypyridine
ethyl nicotinate
low reaction conversion
poor reaction conversion
poor reaction conversion
some lots had strong odor
excellent toxicity data
yes
yes
Scheme 4
wash with a small hydrochloric acid charge to protonate the
excess pyridine. Since the trifluoroacetic acid salt of pyridine
has a favorable partition coefficient between water and ethyl
acetate, >90% of the pyridine was removed by this single
aqueous wash. The trifluoroacetic acid salt of ethyl nicotinate
proved to be more difficult to remove, as it is freely soluble in
ethyl acetate. We decided that the best approach would be to
first remove the TFA byproduct with another base. A salt screen
of inorganic bases and tertiary amines with higher pKa’s than
that of ethyl nicotinate revealed that tetramethylethylenediamine
(TMEDA) forms a water-soluble bis-TFA salt that is completely
insoluble in ethyl acetate. As a result, the final procedure for
the ethyl nicotinate process incorporated a water and TMEDA
charge after the dehydration reaction and a subsequent phase
split. Hydrochloric acid (2 N) was then charged to extract the
corresponding ethyl nicotinate hydrochloride salt, which pos-
sesses a far more favorable partition coefficient between water
and ethyl acetate than its TFA analogue. This low pH wash
(pH 2-2.5) effectively removed approximately 90% of the ethyl
nicotinate while avoiding loss of the Boc protecting group.
The pyridine and ethyl nicotinate processes both resulted in
trifluoroacetylation of the adamantane alcohol, requiring a
subsequent hydrolysis. To avoid any possible racemization of
the nitrile-bearing stereogenic center,21 aqueous potassium
carbonate was utilized for the hydrolysis of 6 to alcohol 7. A
small amount of methanol was added to improve the phase
transfer of the 25 wt % potassium carbonate into the ethyl
acetate layer. After approximately 1 h at 40 °C, the hydrolysis
was complete, and there was no detectable epimerization. The
reaction was then cooled to 15 °C in preparation for the
subsequent phase split. While an ethyl acetate/heptane crystal-
lization was feasible, an isopropanol/water crystallization proved
to be far superior at removing a number of polar byproducts
formed in this three-step telescope as well as trace impurities
present in the two starting materials. After an efficient solvent
exchange from ethyl acetate to isopropanol, water was added
to initiate crystallization and to adjust to a final solvent
composition of approximately 75% water and 25% isopropanol.
The average overall yield for the conversion of 2 to 7 using
the ethyl nicotinate-based dehydration on scale-up to 280 kg
was 78%, and the purity was >99.7%, with the only significant
impurity being ∼0.15% of the intermediate amide 5. To date,
10 batches and over 1.2 MT have been prepared with the
described process.
Deprotection of Primary Amine. The final step in the
preparation of the drug substance was the deprotection of 7
(Scheme 4). As expected, many acids facilitated this Boc
deprotection. In addition to providing a clean, fast reaction,
hydrochloric acid was ultimately chosen due to the preferential
characteristics of the chloride salts formed upon neutralization.
Both potassium and sodium chloride were easily purged, and
just as importantly, these helped to lower the concentration of
1 in the aqueous waste by increasing the ionic strength of the
aqueous phase. A solvent screen showed that both acetonitrile
and isopropanol gave exceptionally clean reaction profiles, but
the ICH Q3C class 2 status of acetonitrile made the selection
of isopropanol obvious. Somewhat surprisingly, the use of
methanol gave small amounts of the corresponding carboxylic
acid, presumably due to imidate formation and subsequent
hydrolysis. Initial pilot-plant campaigns used 2 L of isopropanol
per kilogram of 7 and 1.4 equiv of 2 N HCl. A final process
improvement was to switch to a mixture of 1 L of isopropanol
and 1 L of water per kg of 7, in conjunction with using
concentrated HCl. The reduction in isopropanol was important,
as it reduced the amount of water in the organic phase after
extraction of the free amine 1. This was critical in reducing the
cycle time of the subsequent drying distillation, and thus limited
cyclic amidine formation. HCl salt 8 was quite stable under
the acidic deprotection conditions of 60 to 70 °C, as the product
is much less prone to cyclization under acidic conditions. Similar
to the synthesis of methanoproline 4, concentrated hydrochloric
acid (1.4 equiv) was added to a 65 °C solution of 7 to allow
nearly addition control of gas evolution. In addition to the
standard one equivalent of CO2, this process formed 0.4 equiv
of isobutylene and 0.6 equiv of tert-butanol. Unlike the
(21) Epimerization of the nitrile stereogenic center was observed at pH
>12 under warm, alcoholic conditions.
1172
•
Vol. 13, No. 6, 2009 / Organic Process Research & Development

deprotection of 3, the presence of water in this reaction
prevented the formation of any detectable isopropyl tert-butyl
ether. At the end of the reaction, additional water was added at
65 °C to prevent formation of an unstirrable slurry of the HCl
salt 8 upon cooling.
Monohydrate Formation and Crystallization. Early clini-
cal deliveries of the active drug were of the benzoate salt. A
change in the final form was required due to formulation
requirements. The free base monohydrate 1 was selected as the
preferred form for commercial development. While 1 is stable
as a solid, handling it in solution was initially challenging due
to the more facile conversion of the free base to the cyclic
amidine 9. Detailed process-enabling kinetic studies to under-
stand and control this undesired conversion were performed and
will be the topic of a future publication.
Concentrated sodium hydroxide (50 wt %) was utilized to
neutralize the first equivalent of hydrochloric acid. The final
pH adjustment was carried out using 25 wt % potassium
carbonate to allow a buffered titration of the remaining 0.4 equiv
of HCl. Both of these base additions were carried out in the
presence of methylene chloride. Despite the general undesir-
ability of using an ICH Q3C class 2 solvent in an API step,
methylene chloride was chosen as it was the only process-
friendly solvent identified to efficiently extract 1 from an
aqueous solution. In addition, the product is quite stable in this
solvent and tolerates distillation under atmospheric conditions.
As has been made apparent by the use of concentrated acids
and bases, minimal water is used for this process to minimize
product loss to the aqueous layer.22 In addition, 1.25 kg of
sodium chloride per kg of starting material was added to further
reduce losses in the extraction and to reduce the amount of water
present in the organic extract.
The methylene chloride solution of 1 was distilled under
atmospheric conditions to remove water down to <0.4 equiv.
This prevented crystallization of monohydrate 1 upon subse-
quent addition of ethyl acetate. This operation carries the greatest
risk of forming the undesired cyclic amidine 9. The small
change in reaction solvent mentioned earlier, using 1 L of IPA
per kilogram (kg) of 7 rather than 2 L per kg, had a profound
effect on this distillation. The reduction in isopropanol lowered
the initial water content of the organic phase enough to cut the
methylene chloride consumption by approximately 50% and
significantly reduced the cycle time and thus the formation of
cyclic amidine 9. After a polish filtration to remove residual
solid salts, a controlled water addition was performed to induce
crystallization of the monohydrate 1. A constant volume
distillation, charging fresh ethyl acetate at approximately the
distillation rate, was utilized to complete the solvent switch from
methylene chloride to ethyl acetate. While this type of distil-
lation is known to reduce solvent usage,23 it was chosen here
primarily to minimize the adherence of solids to the walls of
the reactor during the distillation and continued crystallization.
Although this is an atypical crystallization protocol for an API,
the nature of the formulation allowed for a broad range of
acceptable particle sizes. A small amount of water was added
postdistillation to ensure enough water was present to achieve
Figure 1. Tolerance study of cyclic amidine impurity in the
crystallization and cake wash.
the target 1.5-2.0% water. In addition to assuring proper
hydration of the product, the addition of water enhanced removal
of the ∼0.4% cyclic amidine 9 formed during the process, and
minimized yield losses. For these same reasons, the cake wash
was composed of 2% water in ethyl acetate. Figure 1 displays
results of a tolerance study in which various amounts of the
cyclic amidine impurity 9 was spiked into precrystallization
process streams in order to test the ability of the crystallization
and cake wash to reject this impurity. As demonstrated in the
graph, the crystallization readily purges that impurity at levels
above 3%, as compared to a typical in-process level of 0.5%.
The typical yield for this process on scale was 86-90%, with
>99.9% purity and <0.1% cyclic amidine. This final procedure
was significantly greener due to waste reduction (70% less in
total waste, 60% less methylene chloride) and yield improve-
ment compared with the original synthesis of the monohydrate.
Drying the Monohydrate. Drying of organic hydrates has
been discussed in the literature,24 and in this case drying of the
monohydrate free base 1 required an in-depth understanding
of the thermodynamics in order to avoid formation of a known,
undesired anhydrous crystalline form. To avoid concerns over
the kinetics of form conversions, relative humidity conditions
were established under which the desired monohydrate was the
thermodynamically preferred crystal form. The relative humidity
required to maintain the monohydrate was primarily established
by monitoring the form conversion of a physical mixture of
the monohydrate and anhydrous forms in slurries with varying
water content or activity. Those experiments were supplemented
by dynamic vapor sorption (DVS), drying experiments, and
solubility measurement of both forms as a function of water
content. All of these investigations suggested that the relative
humidity should be maintained above approximately 4.5% in
order to preserve the monohydrate indefinitely. For operational
reasons, a decision was made to monitor the dew point rather
than the relative humidity during drying. At the 40 °C dryer
jacket temperature, 4.5% relative humidity corresponds to a dew
point of approximately -8 °C. Several options for providing
the required humidity were considered. Because the azeotrope
between water and ethyl acetate (∼3.6 wt % water at 25 mmHg,
(24) (a) Cypes, S. H.; Wenslow, R. M.; Thomas, S. M.; Chen, A. M.;
Dorwart, J. G.; Corte, J. R.; Kaba, M. Org. Process Res. DeV. 2004,
8, 576–582. (b) Variankaval, N.; Lee, C.; Xu, J; Calabria, R.; Tsou,
N.; Ball, R. Org. Process Res. DeV. 2007, 11, 229–236.
(22) The water solubility of the monohydrate is 19 mg/mL.
(23) Gentilcore, M. J. Chem. Eng. Proc. 2002, 56–59.
Vol. 13, No. 6, 2009 / Organic Process Research & Development
•
1173

conversion, and ensured the preservation of the desired mono-
hydrate while maintaining an acceptable overall drying cycle
time.
Conclusions
From the purchased unnatural amino acid derivatives 2 and
3, the synthesis of the DPP-IV inhibitor saxagliptin involves
five chemical transformations and three isolations. Significant
hurdles, such as analytical challenges due to poor UV chro-
mophores and a thermodynamically favorable cyclization, were
overcome to provide a robust and efficient synthesis of
saxagliptin. In addition, vital changes were made for environ-
mental and safety concerns, including replacement of pyridine
with the more benign base, ethyl nicotinate. The key step in
this convergent synthesis is the amide coupling, which links
the adamantane and proline-based cores. In the optimized
commercial process, the overall yield from the proposed starting
materials is approximately 65% and represents a significant
improvement over earlier process versions. The safe imple-
mentation and robustness of the processes have been validated
by multiple process demonstrations in both pilot and manufac-
turing facilities. Specifically, nine batches (>600 kg) of saxa-
gliptin, 1, have been prepared by the described processes, all
having >99.9% purity, and the correct monohydrate form.
Figure 2. Pilot-plant drying trends.
6.3 wt % water at 250 mmHg)25 is greater than the solubility
of water in ethyl acetate (3.3 wt %) a water-wet ethyl acetate
cake wash alone does not provide sufficient excess water to
allow for a robust drying process across multiple pressures.
Although a humid nitrogen sweep did provide the requisite
humidity, this approach required the use of a portable reactor
and was ultimately replaced with the following simpler ap-
proach. After the cake was washed with 2 wt % water in ethyl
acetate and deliquored, a small addition of water (0.04 kg/kg
1) was charged to the agitated wetcake just prior to drying. A
key to the success of this procedure was that the addition of
water raised the concentration beyond the azeotropic composi-
tion during vacuum drying. This promoted the preferential
removal of ethyl acetate during drying and allowed for a robust
drying process on pilot- and manufacturing scale that consis-
tently delivered the desired monohydrate. The low solubility
of 1 in water also made this operation feasible, without concern
of changing the physical properties of the wet cake.
The dew point in the dryer vent line and the cake temperature
provided insight into the drying process on scale. Figure 1 shows
the trends of a pilot-plant batch that was dried in a filter dryer
with 40 °C on the jacket and 40 mbar pressure. Evaporative
cooling initially lowered the cake temperature as ethyl acetate
was rapidly removed from the powder. As the ethyl acetate
was depleted, the cake warmed until it essentially reached
water’s boiling point under these conditions. The dew point
followed a similar trend since water was driven from the cake
as it warmed. The plateaus in batch temperature and dew point
represent the portion of drying where the excess water was being
removed from the cake. Once the surface water was dried from
the cake, the batch temperature then rose and the dew point
fell, giving independent confirmation that the dryer was ready
to be sampled for analytical verification of drying completeness.
In this example, the dew point fell precipitously once the excess
water was removed from the cake. This occurred because there
was a slight nitrogen sweep into the headspace of the dryer,
corresponding to about 1 dryer volume turnover every 30 min.
In the absence of such a sweep, or similar leak from the
atmosphere, the dew point above the dry cake will remain at
roughly the same value as during the water drying period. As
demonstrated in Figure 2, this approach ensured adequate
humidity, well beyond the thermodynamic limit for form
Experimental Section
L-cis-4,5-Methanoprolineamide Methanesulfonic Acid
Salt (4). Isopropyl alcohol (808 kg) and 3 (135 kg, 600 mol)
were charged to a glass-lined reactor. An additional isopropyl
alcohol charge (39 kg) was added via “spray balls” to remove
solids from the walls. The slurry was then heated to 60 °C,
where most if not all of the solids dissolved. Methanesulfonic
acid (74.54 kg, 776 mol) was charged while maintaining the
temperature between 55 and 65 °C. This continuous feed was
spread over 30 min to control gas evolution. Isopropanol (10
kg) was used to rinse the methanesulfonic acid charge lines.
The product began to crystallize out of solution at this point.
The reaction was held at 60 °C for 3 h. The reaction was
sampled to ensure there was <3 mg/mL of 3 remaining. The
slurry was cooled to 20 °C over one hour and held for an
additional hour. The product was then filtered using four loads
on a centrifuge. Each cake load was washed with isopropanol
(50 kg). The solids were dried at 50 °C in a conical dryer to
give 4 as an off-white product (124 kg, 94%). 1H NMR (400
MHz, DMSO-d6) δ 7.94 (s, 1H), 7.67 (s, 1H), 4.46 (dd, J1 )
7.8 Hz, J2 ) 3.0 Hz, 1H), 3.31 (ddd, J1 ) 7.0 Hz, J2 ) 6.0 Hz,
J3 ) 2.5 Hz, 1H), 2.62-2.52 (m, 1H), 2.38 (s, 3H), 2.12 (dd,
J1 ) 10.9 Hz, J2 ) 2.8 Hz, 1H), 1.75 (ddd, J1 ) 10.9 Hz, J2 )
9.1 Hz, J3 ) 5.3 Hz, 1H), 0.84 (m, 1H), 0.61 (ddd, J1 ) 7.3
Hz, J2 ) 5.0 Hz, J3 ) 2.5 Hz, 1H) 13C NMR (100 MHz,
MeOD) δ 171.43, 60.33, 38.17, 36.61, 31.58, 17.05, 8.84.
Analysis Calculated for C7H14N2O5S: C, 35.29; H, 5.92; N,
11.76; S, 13.46. Found: C, 35.12; H, 5.86; N, 11.68; S, 13.55.
(S)-N-Boc-3-hydroxyadamantylglycine-L-cis-4,5-metha-
noprolinamide (5). Acetonitrile (318 kg), N,N-diisopropyl-
ethylamine (118 kg, 913 mol), and ethyl acetate (122 kg) were
charged to a glass-lined reactor to prepare a homogeneous
solution. To a separate glass-lined reactor were charged 2 (135
(25) Horsley, L. H. Azeotropic Data III; Advances in Chemistry Series,
116, American Chemical Society: Washington D.C., 1973; p 21.
1174
•
Vol. 13, No. 6, 2009 / Organic Process Research & Development

kg, 415 mol), 4 (97 kg, 436 mol), HOBt (62 kg, 405 mol), and
EDC (87 kg, 454 mol). The DIPEA solution was then added
to the solids, followed by an ethyl acetate rinse (45 kg). The
jacket of the reactor was maintained at 25 °C during and post
addition, allowing the exotherm of the base addition to take
the reaction to ∼40 °C. The resulting solution was held for
3 h. Ethyl acetate (1094 kg), 2 N hydrochloric acid (217 kg,
415 mol), water (208 kg), and brine (483 kg) were charged
and allowed to agitate for 20 min. After settling for 30 min,
the lower aqueous layer was removed (1162 kg). Potassium
bicarbonate solution (20 wt %, 770 kg) was charged and stirred
for 20 min. After settling for 30 min, the lower aqueous layer
was removed. The potassium bicarbonate wash was repeated,
and the aqueous waste was combined (1750 kg). A sample was
taken for HPLC analysis to determine the solution yield of 5
(373 mol, 90%). The jacket of the reactor was set to 90 °C,
and vacuum was pulled to ∼350 mbar. The ethyl acetate
solution was distilled down to ∼540 L. Once the target volume
was reached, ethyl acetate (1500 kg) was charged at a rate such
that the volume in the reactor remained constant. At the end of
the distillation, the solution was diluted with an additional ethyl
acetate (973 kg) charge. A Karl Fisher test was performed to
ensure the water content of the organics was <0.05% and was
suitable for proceeding to the subsequent dehydration reaction.
Analytical data were consistent with previously reported data.5
(S)-N-Boc-3-hydroxyadamantylglycine-L-cis-4,5-metha-
noprolinenitrile (7). Ethyl nicotinate (198 kg, 1.3 kmol) was
charged to the dry solution of 6 above and was cooled to -10
°C. Trifluoroacetic anhydride (165 kg, 788 mol) was added over
30 min to keep the temperature of the reaction <10 °C. After
holding for 30 min, a sample was taken to confirm reaction
completion by HPLC. Water (945 kg) and tetramethylethyl-
enediamine (69 kg, 590 mol) were charged. The heterogeneous
solution was then warmed to 15 to 20 °C and allowed to
separate. The lower aqueous layer (1110 kg) was removed and
the organics were recooled to 5 °C. Hydrochloric acid (2 N,
683 kg, 1.3 kmol) was added slowly to keep the batch
temperature <10 °C. The mixture was agitated for 20 min prior
to allowing the layers to separate for 1 h. The bottom aqueous
layer (1067 kg) was removed. Aqueous potassium carbonate
(25 wt %, 688 kg, 1.2 kmol) was then charged, followed by
methanol (213 kg). The mixture was warmed to 40 °C and held
for 1 h. Once the reaction was determined complete by HPLC,
the reactor was cooled to 15 °C. After settling for 30 min, the
bottom aqueous layer (606 kg) was removed. Hydrochloric acid
(2 N, 6 kg) was added to adjust the pH to 6-8 and brine (483
kg) was added. After settling for 30 min, the bottom aqueous
layer (600 kg) was removed. The reactor jacket was set to 80
°C and pressure to 300 mbar. The organics were distilled down
to 540 L. Once the target volume was reached, isopropanol (810
kg) was added at a rate equal to the distillation rate. A sample
for GC analysis was taken to ensure that the ratio of isopropanol
to ethyl acetate was >95:5 wt %. The solution was then further
concentrated to 450 L, keeping the temperature 35-50 °C. Once
the target volume was reached, the vacuum distillation was
stopped and water (300 kg) was added at 40-50 °C. The batch
was slowly cooled to 40 °C, and 7 (1.4 kg) was added to initiate
nucleation. Water (600 kg) was added continuously over 2 h,
maintaining a batch temperature of 40 °C. The resulting slurry
was cooled to 15 °C and held for 1 h. The batch was filtered
on a filter dryer and washed with a 75% water, 25% isopropanol
mixture (2 × 250 kg). The solids were dried at 50 °C to give
7 as a white solid (134 kg, 78%). 1H NMR (400 MHz, CDCl3-
d6) δ 5.30 (d, J ) 9.9 Hz, 1H), 5.02 (dd, J1 ) 10.6 Hz, J2 )
2.3 Hz, 1H), 4.44 (d, J ) 9.9 Hz, 1H), 3.86-3.79 (m, 1H),
2.55 (ddd, J1 ) 16.4 Hz, J2 ) 10.6 Hz, J3 ) 5.8 Hz, 1H), 2.35
(dd, J1 ) 13.6 Hz, J2 ) 2.3 Hz, 1H), 2.26-2.20 (m, 2H), 1.87
(ddd, J1 ) 13.2 Hz, J2 ) 6.5 Hz, J3 ) 6.5 Hz, 1H), 1.77 (dt,
J1 ) 11.6 Hz, J2 ) 2.3 Hz, 1H) 1.75-1.42 (m, 12H), 1.41 (s,
9H), 1.05 (dd, J1 ) 6.8 Hz, J2 ) 4.3 Hz, 2H) 13C NMR (100
MHz, CDCl3) δ 169.99, 155.84, 119.28, 80.00, 68.61, 58.66,
46.32, 45.12, 44.42, 44.31, 41.22, 38.04, 37.56, 37.06, 35.22,
30.51, 30.25, 30.23, 28.39, 17.83, 13.57. MS (FAB) m/z 416
[M + H]+.
(S)-3-Hydroxyadamantylglycine-L-cis-4,5-methanopro-
linenitrile (1). To a glass-lined reactor was charged 7 (82.5
kg, 199 mol), isopropanol (64.8 kg), water (82.5 kg), and 37%
hydrochloric acid (3.9 kg, 39.7 mol). The batch was heated to
65 °C, and hydrochloric acid (23.5 kg, 238 mol) was added
over 30 min, followed by a water flush (16.5 kg). After 90 min
at 65 °C, water (165 kg) was charged, and an HPLC sample
was taken to determine reaction completion. The batch was
cooled to 25 °C, and methylene chloride (660 kg) was added.
Sodium hydroxide (10 N, 24.8 kg, 199 mol) was charged with
a water (16.5 kg) line flush. Potassium carbonate (24 wt %,
43.9 kg, 79 mol) was added to adjust to pH 9, followed by a
water flush (16.5 kg). Sodium chloride (103 kg, 1.8 kmol) was
charged and allowed to agitate for 30 min. The layers were
allowed to separate, and the lower organic layer was transferred
to a clean, glass-lined reactor. An atmospheric distillation was
performed to reduce the total volume to 250 L. Ethyl acetate
(74 kg) was added, and the solution was filtered through a 1
µm line filter. Ethyl acetate (74 kg) was used to chase the reactor
and transfer lines. Water (2.2 kg) was then added, and the
solution was aged for 30 min to induce crystallization. Another
water charge (2.2 kg) was performed, followed by an additional
20 min age. Water (17.5 kg) was charged, and the pressure
was set to 300 mbar. The jacket of the reactor was set to 60
°C, and distillation was initiated at 20 °C. Ethyl acetate (520
kg) was charged at a rate to keep the volume in the reactor
constant during the distillation. The pressure was reduced
periodically during the distillation to maintain the target 15-30
°C batch temperature. The final pressure set point was 100 mbar.
After completion, water (6.2 kg) was charged, and the batch
was cooled to 5 °C. The slurry was filtered on a filter dryer
and washed with wet ethyl acetate (6.2 kg water, 148 kg
EtOAc). The monohydrate 1 was dried at 40 °C. The dew point
in the dryer was monitored by a dew point hygrometer, Vaisala
HMP360 series probe. A small water charge (2.8 kg) was added
during drying to ensure that the dew point in the dryer remained
>-8 °C. After the cake temperature had risen above 20 °C, a
dry nitrogen purge equivalent to two dryer volume turnovers
per hour was initiated. The nitrogen purge primarily served to
clear the dryer headspace of water vapor following removal of
surface water from the drug substance. By this mechanism the
dew point measurement dropped precipitously once the cake
Vol. 13, No. 6, 2009 / Organic Process Research & Development
•
1175

was dry, and served as an independent signal to the cake
temperature that the cake was dry. Monohydrate 1 was isolated
as a white solid (58.2 kg, 88%). 1H NMR (400 MHz, CD2Cl2-
d6) δ 5.25 (dd, J1 ) J2 ) 1.0 Hz, 1H), 4.93 (dd, J1 ) 10.6 Hz,
J2 ) 2.3 Hz, 1H), 3.55-3.50 (m, 1H), 3,35 (s, 1H), 2.45 (ddd,
J1 ) 16.1 Hz, J2 ) 10.9 Hz, J3 ) 5.6 Hz, 1H), 2.25 (dd, J1 )
13.6 Hz, J2 ) 2.5 Hz, 1H), 2.18-2.10 (m, 2H), 1.83-1.42 (m,
15H), 1.40-1.27 (m, 3H) 1.0-0.87 (m, 2H) 13C NMR (100
MHz, CD2Cl2) δ 173.43, 120.15, 68.83, 60.90, 46.57, 45.51,
45.08, 45.01, 41.62, 38.15, 37.92, 37.35, 35.88, 30.98, 30.93,
30.80, 18.00, 13.69. MS (FAB) m/z 316 [M + H]+.
John Mazzullo, and Jurong Yu. We thank Raymond Scaringe
and Shawn Yin for their close collaboration on the identification
of relative humidity conditions for the form-controlled drying
of monohydrate 1, Yun Ye and Xiaozhong Liang for the
development of analytical methods and campaign support, and
the process safety group (Francis J Okuniewicz and Simon
Leung). We especially thank Jeff Reinhardt and the rest of our
operations group for their partnership during multiple pilot-plant
campaigns and with the transfer of technology to manufacturing,
and Process R&D management for their support.
Acknowledgment
We are grateful for the support of our many project team
collaborators, including Elias Mattas, James Vernille, Nian Liu,
Received for review August 27, 2009.
OP900226J
1176
•
Vol. 13, No. 6, 2009 / Organic Process Research & Development