The role of new technologies in defining a manufacturing process for ppara agonist ly518674

Article abstract of DOI:10.1021/op8002486

The impact of several new technologies on the development of a manufacturing process for LY518674 is described. Extensive use of process analytical technology (PAT) throughout development, both at laboratory and pilot-plant scale, enabled data-rich experi

Full text of DOI:10.1021/op8002486

Organic Process Research & Development 2009, 13, 131–143
Full Papers
The Role of New Technologies in Defining a Manufacturing Process for PPARr
Agonist LY518674†
Mark D. Argentine, Timothy M. Braden, Jeffrey Czarnik, Edward W. Conder, Steven E. Dunlap, Jared W. Fennell,
Mark A. LaPack, Roger R. Rothhaar, R. Brian Scherer, Christopher R. Schmid,‡ Jeffrey T. Vicenzi, Jeffrey G. Wei, and
John A. Werner*
Chemical Product Research and DeVelopment, Lilly Research Laboratories, A DiVision of Eli Lilly and Company, Lilly
Corporate Center, Indianapolis, Indiana 46285, U.S.A.
Robert T. Roginski
EigenVector Research, Inc., 3905 West Eaglerock DriVe, Wenatchee, Washington 98801, U.S.A.
Abstract:
ensuring the patient receives medicines of the highest
quality.1,2
The impact of several new technologies on the development of a
manufacturing process for LY518674 is described. Extensive use
of process analytical technology (PAT) throughout development,
both at laboratory and pilot-plant scale, enabled data-rich experi-
ments, shortened development cycle times, and obviated the
requirement of PAT for process control at larger scale. In situ
ReactIR was used to develop a kinetic model for a one-pot
preparation of a semicarbazide intermediate. Parallel crystallizers
fitted with online focused-beam reflectance measurement (FBRM)
and particle vision and measurement (PVM) probes were used
in the development of several challenging crystallization
processes. Application of the process knowledge afforded
by these technologies, combined with the principles of
Quality by Design, resulted in excellent purity control
throughout the four-step process. A single, 5-min, MS-
friendly method capable of separating over 30 components
was developed using a combination of chromatography
modeling software, sub-2 µm column technology, and
higher-pressure LC equipment. The method was used
across all four processing steps, greatly facilitating impurity
tracking, and reducing assay time and solvent use by 85%
and 93%, respectively.
This contribution highlights the impact of several technolo-
gies on both productivity and process understanding during the
development of a manufacturing process for PPARR agonist
LY518674 (1). Process analytical technology (PAT) was used
extensively, both in the laboratory and in the pilot plant, with
the objective of designing quality into each processing step
through increased process understanding.3 The desired result
is a robust process that consistently delivers active pharmaceuti-
cal ingredient (API) with the appropriate quality attributes4
without the requirement of PAT for process control in
manufacturing.
Applications of the technologies listed below are exemplified
in the present paper, which focuses on development of process
and impurity control strategies to enable the preparation of API
for registration-phase studies.
• ultra high-pressure chromatography and sub-2 µm column
technology for rapid assays and impurity tracking.
• chromatography modeling software for method optimiza-
tion in the separation of more than 30 components.
• in situ ReactIR for kinetic modeling of a KOCN reaction
profile.
• online vapor-phase MS for monitoring residual oxalyl
chloride during a pilot plant distillation operation.
• parallel crystallizers fitted with in situ focused-beam
reflectance measurement (FBRM) and particle vision and
measurement (PVM) probes for expedited crystallization
development.
Introduction
New technologies are playing an increasingly important role
in the development of chemical processes within the pharma-
ceutical industry. Process analytical technology, automation,
parallel reactions, continuous processing, and fast, high-resolu-
tion assays are enabling data-rich experiments that enhance
process knowledge in shorter development time frames. Com-
bined with the principles of Quality by Design, this knowledge
results in more robust manufacturing processes, thereby
(1) Pharmaceutical cGMPs for the 21st Century - A Risk-Based Approach.
Final Report. Department of Health and Human Services, U.S. Food
and Drug Administration: Washington, D.C., Fall (September) 2004;
see http://www.fda.gov/cder/gmp/gmp2004/GMP_finalreport2004.htm.
(2) See ICH Guidelines Q8, Annex to Q8, and Q9; Available online at:
http://www.ich.org/cache/compo/363-272-1.html.
(3) Guidance for Industry: PAT - A Framework for InnoVatiVe Pharma-
ceutical DeVelopment, Manufacturing, and Quality Assurance. Depart-
ment of Health and Human Services, U.S. Food and Drug Adminis-
tration: Washington, D.C., September 2005. See http://www.fda.gov/
cder/guidance/6419fnl.pdf.
(4) Ganzer, W. P., Materna, J. A., Mitchell, M. B., Wall, L. K. Current
Thoughts on Critical Process Parameters and API Synthesis. Phar-
maceutical Technology; July 2, 2005. Available online at http://
pharmtech.findpharma.com/pharmtech/content/printContentPopup.
jsp?id)170114.
† This paper is dedicated to the memory of our colleague, Dr. Christopher
Schmid (1959-2007). His integrity, passion for science, and faith in God were
an inspiration to us all.
* To whom correspondence should be addressed. Telephone: (317) 276-9385.
E-mail: jwerner@lilly.com.
‡ Chris Schmid passed away 26 December 2007.
10.1021/op8002486 CCC: $40.75
Published on Web 02/03/2009
2009 American Chemical Society
Vol. 13, No. 2, 2009 / Organic Process Research & Development
•
131

Scheme 1. Initial route selected for developmenta
LY518674 (1),5 is a highly potent and selective agonist of
peroxisome proliferator-activated receptor alpha (PPARR).6,7 It
was recently evaluated in phase II clinical studies in patients
with dyslipidemia and hypercholesterolemia at doses of 10-100
µg/day.8
The starting point for our research was the route selected
for further development by Dr. Christopher Schmid and his
colleagues (Scheme 1).9 The convergent synthesis involves the
coupling of semicarbazide 3 and carboxylic acid 5 to give
acylsemicarbazide 7. Semicarbazide 3 was prepared regiospe-
cifically in three steps from commercially available 4-methyl-
benzaldehyde in 69% overall yield. Carboxylic acid 5 was
obtained in one step by monoalkylation of the dianion of
4-hydroxyphenylbutyric acid (4).10 Treatment of acylsemicar-
bazide 7 with 10-camphorsulfonic acid (CSA) afforded triaz-
olone 8 via an acid-mediated cyclization. Purification by
treatment with Amberlyst 15, followed by crystallization,
afforded 8 in 50-55% yield from 5. Hydrolysis of the ester
with aqueous base completed the synthesis of LY518674 (1)
in 36% overall yield from 2.
a Reagents and conditions: (a) 50 psig H2, 5% Pt/C, THF, 50 °C, 8 h, 85%;
(b) TMS-NCO (1.5 equiv), IPA, rt, 16 h, 86%; (c) MsOH (1.1 equiv), CH2Cl2,
reflux, 16 h, 95%; (d) NaOEt (2 equiv), EtOH, EtOAc, reflux, 0.5 h; add 4,
reflux 0.5 h; add EtO2CCBr(CH3)2 (3 equiv), reflux 1 h; add NaOEt (1 equiv)
over 1 h, reflux 0.5 h; acidify with H3PO4, remove EtOH and crystallize, 95%;
(e) oxalyl chloride (1.15 equiv), DMF (0.05 equiv), EtOAc, rt, 30 min; (f) pyridine
(2.3 equiv), EtOAc, 0 °C; (g) CSA (1.1 equiv), EtOAc, reflux, 6 h; (h) Amberlyst
15, EtOAc, reflux, 1 h; MTBE (recrystallization), 50-55% overall from 5; (i)
NaOH, H2O, toluene, rt, 4 h; HCl, EtOAc, distill; crystallize from EtOAc, 95%.
(5) (a) Wang, X.; Barr, R. J.; Bean, J. S.; Kauffman, R. F.; Mayhugh,
D. R.; Montrose-Rafizadeh, C.; Renner, J.; Saeed, A.; Singh, J.; Zink,
R. W.; Mantlo, N. B. Abstracts of Papers, MEDI-363; 224th ACS
National Meeting, Boston, MA, U.S.A., August 18-22, 2002,
American Chemical Society: Washington, D.C., 2002. (b) Mantlo,
N. B.; Collado Cano, I.; Dominianni, S. J.; Etgen, G. J., Jr.; Garcia-
Paredes, C.; Johnston, R. D.; Letourneau, M. E.; Martinelli, M. J.;
Mayhugh, D. R.; Saeed, A.; Thompson, R. C.; Wang, X.; Coffey,
D. S.; Schmid, C. R.; Vicenzi, J. T.; Xu, Y. Peroxisome Proliferator
Activated Receptor Alpha Agonists. WO/2002/0238553, 2002. (c) For
a discussion of the SAR, see: Xu, Y.; Mayhugh, D.; Saeed, A.; Wang,
X.; Thompson, R. C.; Dominianni, S. J.; Kauffman, R. F.; Singh, J.;
Bean, J. S.; Bensch, W. R.; Barr, R. J.; Osborne, J.; Montrose-
Rafizadeh, C.; Zink, R. W.; Yumibe, N. P.; Huang, N.; Luffer-Atlas,
D.; Rungta, D.; Maise, D. E.; Mantlo, N. B. J. Med. Chem. 2003, 46,
5121–5124. (d) For a discussion of the pharmacology, see: Singh, J. P.;
Kauffman, R.; Bensch, W.; Wang, G.; McClelland, P.; Bean, J.;
Montrose, C.; Mantlo, N.; Wagle, A. Mol. Pharmacol. 2005, 68, 763–
768.
Results and Discussion
Development of a robust impurity control strategy for the
active pharmaceutical ingredient (API) was one of our primary
objectives.11 Characterization of API from earlier campaigns
using a combination of HPLC, LC-MS, and LC-NMR
resulted in the identification of compounds 9-14 (Scheme 2),
which were then synthesized (see Supporting Information for
experimental procedures). Their low rejection in the final
crystallization step required us to understand their origin in order
to reliably control them at levels below 0.1%.12
A single, fast HPLC method was desired to track the
formation and fate of starting materials, intermediates, and
process-related impurities. With a single method, tracking of
impurities across multiple steps is simplified, easily monitoring
impurities being formed or removed as the synthesis progresses.
Use of HPLC mobile-phase systems compatible with both UV
and MS detection was also desired for efficient, information-
rich data generation. However, this goal is often challenging
based upon the number of components to separate and the
potentially wide range of polarities associated with them.
(6) “Peroxisome proliferator-activated receptors (PPARs) are members
of the nuclear hormone receptor superfamily of ligand-dependent
transcription factors. [They] participate in a broad spectrum of
biological processes, including cell differentiation, energy balance, lipid
metabolism, insulin sensitivity, bone formation, inflammation and
tissue remodeling.” Quote taken from: Guan, Y.; Zhang, Y.; Breyer,
M. D. Drug News Perspect. 2002, 15, 147–154.
(7) For a recent review of PPARR in the pathogenesis of metabolic
syndrome in relation to diabetes and atherosclerosis, see: (a) Staels,
B. Ther. Res. 2006, 27, 1347–1358. See the following for specific
therapeutic areas: (b) Human metabolic syndrome: Azhar, S.; Kelley,
G.FutureLipidol.2007,2,31–53.(c)Immunosuppressiveagents:Cunard,
R. Curr. Opin. InVest. Drugs 2005, 6, 467–472. (d) Diabetic
nephropathy: Varghese, Z.; Moorhead, J. F.; Ruan, X. Z. Kidney Int.
2006, 69, 1490–1491.
(8) Nissen, S. E.; Nicholls, S. J.; Wolski, K.; Howey, D. C.; McErlean,
E.; Wang, M.-D.; Gomez, E. V.; Russo, J. M. J. Am. Med. Assoc.
2007, 297, 1362–1373.
(9) Braden, T. M.; Coffey, D. S.; Doecke, C. W.; LeTourneau, M. E.;
Martinelli, M. J.; Meyer, C. L.; Miller, R. D.; Pawlak, J. M.; Pedersen,
S. W.; Schmid, C. R.; Shaw, B. W.; Staszak, M. A.; Vicenzi, J. T.
Org. Process Res. DeV. 2007, 11, 431–440.
(11) A brief summary of this work has been disclosed previously: Fennell,
J. W.; Dunlap, S. E.; Metzler, R.; Vicenzi, J. T.; Werner, J. A. Abstracts
of Papers; ORGN-009. 234th ACS National Meeting, Boston, MA,
United States, August 19-23, 2007; American Chemical Society:
Washington, D.C., 2007.
(12) API impurity levels were as follows: 9 (0.26%), 10 (0.05%) 11 (0.18%)
12 (0.18%), 13 (0.10), and 14 (0.12%). The HPLC method for this
analysis is described in the Supporting Information.
(10) For the preparation of 4, see: (a) Schmid, C. R.; Beck, C. A.; Cronin,
J. S.; Staszak, M. A. Org. Process Res. DeV. 2004, 8, 670-673. For
an alternative approach using HBr, see: (b) Delhaye, L.; Diker, K.;
Donck, T.; Merschaert, A. Green Chem. 2006, 8, 181–182.
132
•
Vol. 13, No. 2, 2009 / Organic Process Research & Development

Scheme 2. Impurities found in early lots of 1
Figure 1. Comparison of the optimized DryLab prediction (top)
with the actual separation (bottom) of a 36-component mixture
on a Zorbax SB-C18 (2.1 mm × 150 mm, 1.8 µm) at 40 °C and
0.4 mL/min.
Application of Advanced HPLC Technology for “Fast-
Assay” Development. The initial analytical method was
developed on a Zorbax RX-C18 column (4.6 mm × 250 mm,
5 µm particles), aided dramatically by availability of pure
samples for most components along with a chromatographic
software modeling tool (DryLab 2000)13 to rapidly optimize
the separation. However, each sample required 30 min per
injection at a flow rate of 1.5 mL/min, plus time for equilibra-
tion. The method conditions were later adapted to a Zorbax
SB-C18 phase, a similar C-18 phase that offered broader
availability of particle sizes and column dimensions for future
efficiency gains. Use of the Zorbax SB-C18 4.6 mm × 250
mm column with 5 µm particles afforded similar separation,
although it still required over 20 min per injection despite an
increase in flow rate to 2 mL/min.
Advances in the use of smaller-particle columns, as well as
higher-pressure LC instrumentation, were explored to further
improve analysis time without reducing separation efficiency.
Two parallel paths were pursued for evaluation of smaller-
particle columns: chromatographic modeling and column
scaling.
DryLab separation modeling software was applied success-
fully to the sub-2 µm column format to simplify the daunting
task of direct method development and optimization for a
mixture with over 30 components consisting of starting materi-
als, intermediates, and process-related impurities. Optimization
was performed using data from two linear gradients (10 and
30 min) at 40 °C. As seen in Figure 1, the predicted and actual
separations compare quite well.14 A labeled chromatogram
indicating the identity of each component is available in the
Supporting Information.
Figure 2. Comparison of separation performance with three
different particle-size columns on an Agilent 1200 SL system.
column-volume gradient programs. Scaling to the 3.5 µm
column was performed with a drop in flow rate to maintain an
operating pressure comparable to that of the 5 µm column.
Scaling to the 1.8 µm column also required a reduced flow rate
but took advantage of the higher operating pressure available
with the higher-pressure LC system employed for the analysis
(maximum pressure of 600 bar). The impact of high-efficiency
sub-2 µm column technology is illustrated in Figure 2, compar-
ing the chromatographic separation obtained initially by using
the 5 µm particle-size column to separations using the 3.5 and
1.8 µm particle-size columns. As appropriate, gradients and
injection volumes were scaled for equivalent column volumes
and load capacities. Reduction in run time by approximately
one-third was easily achieved while maintaining, and often
enhancing, component separations.
The speed of analysis was further enhanced by mapping the
separation to an ACQUITY BEH-C18 column (1.7 µm packing)
that could operate at even higher pressures (∼1000 bar) using
the Waters ACQUITY UPLC. As shown in Figure 3, acceptable
separation was achieved in less than 10 min with a 150 mm
column. In fact, a 5-min analysis time could be achieved without
a dramatic loss in resolution by reducing the column length to
Scaling from the original 4.6 mm × 250 mm, 5 µm column
to smaller-particle columns was performed by mathematically
scaling the gradient based upon the changes in flow rate and
column length to maintain comparable linear velocity and
(13) Molnar Institut fu¨r angewandte Chromatographie, Schneeglo¨ckchen-
strasse 47, D-10407 Berlin, Germany; http://www.molnar-institut.com/
cd/indexe.htm.
(14) Details of this work have recently been reported: Scherer, R. B.;
Argentine, M. D.; Werner, J. A. Utilization of Technological AdVances
in HPLC to ImproVe Separations and Cycle Times, Abstract #551;
HPLC 2006, San Francisco, CA, 2006.
Vol. 13, No. 2, 2009 / Organic Process Research & Development
•
133

Scheme 3. One-pot preparation and purification of
semicarbazide 16a
Figure 3. Optimized separation: comparison of 100 mm (top)
and 150 mm (bottom) ACQUITY BEH-C18 columns (1.7 µm
particle size, 2.1 mm i.d.) on a Waters ACQUITY UPLC system.
a Reagents and conditions for the optimized process: (a) KOCN (0.98 equiv),
n-BuOH/H2O (1:10 v/v), 20 °C, 8 h, 92% (not isolated, 16/17, 80:20); (b) HCl
(0.3 equiv), 20 °C, 3 h then filtered, 59% (16/17, 99.9:0.1).
100 mm, which also allowed an increase in flow rate. The result
was a single, fast, MS-friendly method for reaction monitoring
and impurity tracking with a greater than 6-fold improvement
in assay time versus the initial method (see Supporting Informa-
tion for method conditions).
Fast LC assays have a dramatic impact on both productivity
and solvent use. For example, analysis time was decreased by
85% (42 h vs 6.3 h) for a nine-reaction experiment where each
reaction was sampled at seven time points.15 In addition, solvent
use was decreased by 93%, due to the combined effects of lower
flow rate and shorter analysis time, reducing both cost and
environmental impact.
Development of a “One-Pot Process” To Prepare Semi-
carbazide 16. Application of in Situ ReactIR, FBRM, and
PVM. A one-pot preparation and purification of semicarbazide
16 (Scheme 3) was developed to replace the three-step sequence
outlined in Scheme 1. An earlier attempt9 to develop a one-pot
process starting with 4-methylbenzylhydrazine (15)16 was
unsuccessful due to an inability to separate 16 from its
regioisomer 17. Direct recrystallization from either IPA or EtOH
gave no significant enrichment, and the mesylate salt of 16 was
found to be unstable in alcoholic solvents at elevated temper-
atures. However, we have found that simple addition of aqueous
HCl preferentially dissolves 17, allowing 16 to be isolated
directly by filtration (vide infra).
The initial process consisted of: (1) adding aqueous KOCN
to a solution of 15, (2) treatment of the resulting mixture of
semicarbazides with a substoichiometric amount of aqueous HCl
to preferentially dissolve 17 as its HCl salt, and (3) isolation of
the desired product 16 by filtration. This process afforded 16
in 65% yield with 1.0-1.5% of 17, and 0.5-3% each of 18,
19, and 20. Hydrazone 18 and semicarbazone 19 were
controlled by the rigorous exclusion of oxygen to prevent the
facile oxidation of 15 to 2. Attempts to improve the isomeric
ratio (which remained constant throughout the reaction) by
varying the temperature17 or the reaction solvent were not
successful. This process was selected for further development
due to its operational simplicity over the three-step process
shown in Scheme 1.
There were three major development objectives for this step.
First, the product partially floated, filling the reaction flask with
solid. This is unacceptable in a manufacturing setting. Second,
residual cyanate levels needed to be controlled since the
presence of KOCN during the HCl addition resulted in the
formation of 20, an insoluble byproduct. Finally, a robust
purification process was required to control the level of 17
consistently below 0.5%. The key role that process analytical
technology (PAT) played in successfully addressing each of
these issues is discussed below.
Microscopic analysis of the “floating” solid indicated that it
consisted of microcrystalline agglomerates. Slowing the addition
rate of aqueous KOCN and seeding the reaction mixture helped,
but did not eliminate the problem. On the basis of observing
that addition of approximately 10% IPA appeared to break up
some of the particles, 18 water-miscible Class 2/3 solvents18
were screened as reaction cosolvents in an attempt to improve
crystallization performance. The best cosolvents were then
compared using an automated Mettler Toledo MultiMax parallel
reactor, where the distribution of particle chord lengths could
be monitored using a Lasentec FBRM probe.19 The dramatic
influence of the cosolvent is illustrated in Figure 4, which shows
significantly fewer, but larger, crystals when the cosolvent is
(17) The ratios of 16:17 were: 82:18, 80:20, 76:24 at 5, 20, and 50 °C,
respectively.
(18) ICH Harmonised Tripartite Guideline. Impurities: Guideline For
Residual SolVents QC3 (R3), revised November 2005. See: http://
www.ich.org/cache/compo/363-272-1.html#Q3C.
(15) See the Supporting Information for an illustration of the use of a data
export method that enabled the preparation of nine multicomponent
reaction profiles in less than 30 min from the raw data in 63
chromatograms.
(16) For preparation of the benzyl hydrazine from the corresponding
chloride, see: Finneman, J. I.; Fishbein, J. C. J. Am. Chem. Soc. 1995,
117, 4228–4239.
(19) For useful discussions on the use of process analytical technology on
crystallization processes, see: (a) Yu, L. X.; Lionberger, R. A.; Raw,
A. S.; D’Costa, R.; Wu, H.; Hussain, A. S. AdV. Drug DeliVery ReV.
2004, 56, 349–369. (b) Birch, M.; Fussell, S. J.; Higginson, P. D.;
McDowall, N.; Marziano, I. Org. Process Res. DeV. 2005, 9, 360–
364. (c) Barrett, P.; Smith, B.; Worlitschek, J.; Bracken, V.; O’Sullivan,
B.; O’Grady, D. Org. Process Res. DeV. 2005, 9, 348–355.
134
•
Vol. 13, No. 2, 2009 / Organic Process Research & Development

Figure 5. Cyanate concentration as determined by IR versus
elapsed time for kinetics experiment. The blue outline represents
the subset of data used for kinetic modeling. The gap in the
data line is due to a spectral measurement missed by the
instrument.
For developing the kinetic model, the reaction was carried
out with a 2 mol % excess of cyanate, as described in the
Experimental Section. IR data was collected throughout the
course of the reaction. Following preprocessing, the spectral
signal for cyanate was converted to concentration using the
calibration model (Figure 5). The following three assumptions
were made during our analysis. First, the reaction between
cyanate and 15 is not measurably affected by equilibrium
consideration, given the very low solubilities of 16 and 17 in
the solvent system utilized.21 Second, the reaction follows
second-order kinetics overall between cyanate and 15, and first-
order in each reactant.22 And third, hydrolysis of cyanate has
only a small effect that does not significantly impact the overall
findings.23 Given these simplifying yet reasonable assumptions,
we note that the concentration profile following cyanate addition
can be used to extract kinetics information.
In order to extract the rate constant from the data, we take
advantage of the self-similar nature of second-order kinetic
profiles; any point in the profile can be chosen as an arbitrary
initial condition, and the remaining points are calculated on the
basis of that choice. (See Supporting Information for further
discussion.) The measurements used for kinetic analysis include
the first point following the peak cyanate concentration as the
initial condition and all subsequent measurements having
concentration values greater than or equal to the arbitrary cutoff
value of roughly 0.01 M. We are assuming a concentration
profile that obeys the following relationship.24
Figure 4. Comparison of the FBRM profiles for the preparation
of the semicarbazide mixture in H2O (top) and 9% n-BuOH in
H2O (bottom). Both reactions were seeded.
present. 1-Butanol (6-9% v/v)20 was found to give large, dense
crystals that rapidly settled out of solution and were easily
filtered. In fact, the two isomers crystallized independently with
different habits, the desired isomer 16 as rectangular plates and
the undesired isomer 17 as bundles of rods. By contrast, small
crystals remained throughout the course of the reaction when
only water was used. The beneficial effect of a cosolvent does
not appear to result from increased solubility, as both com-
pounds are sparingly soluble in the reaction medium. (The
solubility of 16 in H2O and 1:10 n-BuOH/H2O at 22 °C is 2.9
mg/mL and 3.9 mg/mL, respectively. The solubility of 17 in
H2O and 1:10 n-BuOH/H2O at 22 °C is 1.1 mg/mL and 1.3
mg/mL, respectively.) Instead, it is possible that the cosolvent
serves as a wetting agent that prevents the small crystals from
agglomerating, thereby permitting larger-crystal growth to occur.
The second development objective, the control of residual
KOCN to minimize the formation of 20 upon HCl addition,
was also successfully addressed via PAT. In this case, online
IR was used to measure in situ cyanate concentrations, which
were then used in the development of a kinetic model for
determining reaction completion. This approach provided a
wealth of process knowledge with only a few experiments.
(21) Baker and Gilbert report that equilibrium is reached when 97% of
hydrazine cyanate has reacted to semicarbazide in H2O at 25 °C.
However, in the present system, precipitation of the products would
be expected to drive the reaction to completion: Baker, E. M.; Gilbert,
E. C. J. Am. Chem. Soc. 1942, 64, 2777–2780.
(22) Williams, A.; Jencks, W. P. J. Chem. Soc., Perkin Trans. 2 1974,
1753–1759.
(23) Kinetics for the aqueous hydrolysis of cyanate have been well
characterized. For a recent reference with a summary of the literature,
see: DeMartini, N.; Murzin, D. Y.; Forssen, M.; Hupa, M. Ind. Eng.
Chem. Res. 2004, 43, 4815–4821.
(20) Lower amounts of cosolvent resulted in agglomerates of smaller-sized
crystals, while higher amounts of cosolvent (e.g., 15% v/v) resulted
in a sticky solid and a biphasic filtrate.
(24) Levenspiel, O. The Chemical Reactor Omnibook, Oregon State
University Bookstores: Corvalis, OR, 1996; p 2.3.S. and subsequent
manipulation in the Supporting Information.
Vol. 13, No. 2, 2009 / Organic Process Research & Development
•
135

Scheme 4. Equilibria for the separation of 16 and 17
[A]
M - 1
)
, F ) exp(k[A]0(M - 1)t)
[A]0 MF - 1
Here, species A refers to cyanate, and the initial concentration
is obtained directly from the processed IR signal. The value of
M, which is the molar ratio of 15 to cyanate at the initial point,
is calculated from the initial molar ratio and conversion of
cyanate. With a cost function defined as the sum of squares of
the residuals between the kinetic model and the measured data,
a function minimization routine25 generated a rate constant of
90 M-1 h-1 as optimal (Figure 6). Extending the optimization
out further had minimal impact on the value of the rate constant;
including all data points to the end of the measurement period
resulted in an optimal value of k of 88 M-1 h-1.
Table 1. Predicted reaction times for cyanate conversion to
reach >95% with KOCN addition times of 0.25 h, 0.5 h, and
1 h (20 °C, 2 mol % excess of 15)
OCN-
conversion (%) (tfill ) 0.25 h) (tfill ) 0.5 h) (tfill ) 1 h)
time (h)
time (h)
time (h)
95.0
99.0
99.5
99.9
0.70
1.95
2.79
5.14
0.90
2.16
3.00
5.35
1.30
2.59
3.43
5.78
extending the addition time from 0.25 to 1 h in order to achieve
the improved crystal growth mentioned earlier.
Figure 6. Fit of experimental data to the kinetic model. The
The ability to model the time for reaction completion using
only fill time, temperature, and stoichiometry allows the level
of cyanate to be controlled to less than 0.5% prior to the addition
of HCl, thereby maintaining the levels of biscarboxamide 20
below this value. Importantly, this can be now accomplished
in manufacturing without the need for an in-process control
assay for cyanate or online monitoring.
The final objective for this step was to develop a robust
purification. Once again, PAT played an important role, with
data from both FBRM and particle vision and measurement
(PVM) probes providing key insights for improving the process.
As mentioned above, 16 was purified by treatment of the
semicarbazide mixture with aqueous HCl. The equilibria
involved in this process are shown in Scheme 4. The preferential
dissolution of isomer 17 is the result of two factors, its higher
basicity (the pKa for ammonium salts 22 and 21 are 3.27 and
2.89, respectively, in MeOH/H2O at 25 °C) and a 50% greater
solubility of 22 compared to 21. However, the initial process
was not robust. Levels of 17 varied from 0.1-1.2% and were
strongly dependent upon the washing protocol employed, even
when the HCl stoichiometry was increased.
rate constant was determined by nonlinear least-squares.
The reaction scheme depicted for this step, beginning with
the semibatch period where the cyanate solution is added at a
constant rate over a predetermined period of time followed by
a constant-volume batch reaction is readily modeled (see
Supporting Information for development of the model). In
practice, cyanate will be the limiting reagent, but reversing the
roles for the present experiment affords more data to use for
fitting to a kinetic model without the concern of running up
against limit of quantitation considerations. Once the rate
equations for cyanate and 15 are made dimensionless and the
initial conditions are simplified through variable substitution,
only the following three adjustable parameters remain: M, as
defined above; ꢀ, the ratio of the initial reaction volume to that
of the cyanate solution to be added; and R, the product of (1)
the rate constant, (2) the time for the cyanate addition (units
consistent with that of the rate constant), and (3) the concentra-
tion of cyanate in the dosing solution.
In practice, KOCN will be the limiting reagent (with 2 mol
% excess 15), and ꢀ will be very close to 8. The time for cyanate
conversion to reach completion as predicted by the model is
shown in Table 1 for three different cyanate addition times (tfill).
Under these conditions, cyanate is essentially consumed (99.9%
conversion) within 5-6 h. Little penalty is observed in
In situ monitoring using a Lasentec FBRM probe during
the HCl addition enabled us to quickly determine the origin of
the variability. As 5 N HCl was added to the slurry, a decrease
in the number of larger particles was observed with the
dissolution of 17. Unexpectedly, a sharp increase occurred in
the number of small particles (Figure 7, red line), along with
the simultaneous appearance of numerous, hair-like crystals
(25) MATLAB function fminbnd: This algorithm uses golden section
search and parabolic interpolation.
136
•
Vol. 13, No. 2, 2009 / Organic Process Research & Development

extracted directly from the aqueous EtOH upon adjustment of
the pH to less than 2, or it could be separated from the neutral
organic byproducts by adjustment of the pH to 7-9, followed
by extraction into toluene upon acidification. The latter process
was preferred because it provided both an inherently safer and
a more robust process. Removal of the potentially polymerizable
ethyl methacrylate avoided the safety issues associated with its
concentration during the crystallization process. In addition,
ethyl methacrylate and unreacted EBIB solubilized the product
and led to variability in the crystallization yield. The new
process using toluene afforded two additional advantages. It
allowed both impurities 10 and 4 to be quantitatively removed
by an aqueous extraction containing 5-7 mol % NaHCO3,
thereby eliminating API impurity 11. It also enabled the
development of a robust crystallization of 5 from toluene/
heptane with improved impurity rejection on a multikilogram
scale (4 × 18 kg lots, 86 ( 1% yield, 0.2 ( 0.1% total related
substances (TRS)).
Development of the Process to Prepare Triazolone 8.
Application of DoE, UPLC, and FBRM. Preparation, puri-
fication, and isolation of triazolone 8 was one of the primary
quality control points for the entire synthetic route. As shown
in Scheme 1, it was prepared by first coupling acid chloride 6
with semicarbazide 16 to give acylsemicarbazide 7. This product
was carried forward without isolation, since it was formed in
nearly quantitative yield, and all attempts to filter the crystalline
product were unsuccessful.26 Addition of CSA at reflux affected
the cyclization to 8, as described by Schmid and co-workers,9
albeit with several byproducts that could not be rejected by
crystallization (vide infra).
The three primary development objectives for this sequence
were as follows: first, replacement of DMF as a catalyst in the
acid chloride formation, as it gives rise to API impurity 9;
second, improvement of the impurity control strategy, eliminat-
ing the need for treatment with Amberlyst 15 resin prior to
crystallization; and finally, development of a controlled, robust
crystallization process.
DMF was replaced with pyridine in the acid chloride
formation. Carson and Reist27 have shown that pyridine is a
highly effective catalyst for the formation of acid chlorides,
while suppressing anhydride formation. We compared the
activity of DMF and pyridine as catalysts in the conversion of
5 to 6 via oxalyl chloride in toluene. Pyridine was found to
give essentially identical results over the range tested (0.1-5.0
mol %), with two quality benefits: elimination of the problematic
chloroiminium species formed with DMF, and lighter-colored
reaction mixtures (light-yellow vs yellow-brown). Symmetrical
anhydride was not detected with either catalyst. A catalyst
loading of 0.1 mol % pyridine was selected and the process
was optimized using a fractional factorial experimental design
varying the following three parameters: oxalyl chloride stoi-
chiometry, reaction concentration, and temperature (see Sup-
porting Information). Excess oxalyl chloride was removed by
Figure 7. Comparison of FBRM profiles for reactions con-
ducted at two different concentrations. Photo insert shows the
hair-like crystals of 22, which were only observed at the higher
concentration.
in the PVM (see photo insert in Figure 7). The small crystals
were not visible to the naked eye. They decreased in number
as the acid was added but remained throughout the equilibration
step. Apparently, as 17 was protonated, the solution concentra-
tion of 22 exceeded its solubility, and it began precipitating.
As the addition proceeded, the reaction mixture became more
dilute allowing some of the precipitated 22 to redissolve. This
explains the cause for the variability observed in the product
purity and its dependence on the washing protocol. The isomer
was being removed not only by expressing the filtrate from the
filter cake but also by redissolution of precipitated 22 that was
contained within the cake. Once understood, the variability was
readily eliminated by simply increasing the amount of solvent
from 11 mL/g of 15 to 16 mL/g; thereby ensuring that 22
remained in solution (Figure 7, blue line). This modification
significantly improved the process robustness, enabling the level
of 17 to be controlled consistently below 0.2% at both laboratory
and pilot-plant scale, albeit with a slight decrease in yield to
59% due to increased product loss to filtrate.
Development of the Monoalkylation Process to Prepare
Carboxylic Acid 5. Carboxylic acid 5 was prepared by
monoalkylation of the dianion of 4 with ethyl 2-bromoisobu-
tyrate (EBIB), as shown in Scheme 1. Our two primary
objectives were (1) development of a control strategy for the
four API impurities (11-14) that originated from this step and
(2) development of an improved isolation/purification process.
Fortunately, the levels of 12-14 could all be controlled to less
than 0.1% by setting appropriate specifications on the alkylating
agent EBIB. The other API impurity, phenol 11, which
originated from the carry-though and forward processing of
unreacted starting material 4, was reduced to less than 0.1%
via the modified workup described below.
The alkylation reaction was conducted as described by
Schmid and co-workers.9 This process gave a reaction profile
consisting of the desired acid 5 and 1-2% each of 10 and 4,
as well as excess alkylating agent and ethyl methacrylate from
the dehydrobromination of EBIB. In addition, the crystallization
of 5 from acidic aqueous EtOH was no longer acceptable due
to poor impurity rejection and a tendency for the product to
oil.
(26) Treatment of acylsemicarbazide 7 with aqueous NaOH (20 h at 40
°C followed by 24 h at 100 °C), using conditions similar to those
described in ref 5c, affords LY518674 (1) in a single step (77% yield).
However, isolation and purification of 8, followed by saponification
to give 1, provides a better impurity-control strategy than the one-
step process.
The isolation was modified by quenching the reaction into
a mixture of toluene and aqueous HCl. The product could be
(27) Carson, J.; Reist, E. J. J. Org. Chem. 1958, 23, 1492–1496.
Vol. 13, No. 2, 2009 / Organic Process Research & Development
•
137

Scheme 5. Impurities derived from 7
Scheme 6. Control of impurities
distillation, using online vapor-phase MS analysis of the
distillate to monitor the removal during pilot-plant production.
Acylsemicarbazide 7 was prepared in nearly quantitative
yield by addition of the resulting acid chloride solution to a
slurry of semicarbazide 16 in toluene containing pyridine, which
served as both a base and an acylation catalyst. Acid was then
added directly to the crude reaction mixture to effect the
cyclization.
6). The 4-methylbenzyl hydrazine 26, liberated from 7, 16, and
25 reacted with CSA to give hydrazone 27, which was readily
removed by extraction. The only components stable to the
ethanolysis conditions were the desired product 8 and oxadia-
zolone 24, which was readily rejected by crystallization. This
control strategy is so efficient that any amount of compounds
1, 5, 7, 16, 24, and/or 25 can be present without any impact to
the quality of triazolone 8 obtained from the crystallization.
The use of EtOH in the presence of H2SO4 gave rise to three,
potentially genotoxic impurities that were readily controlled.
The ethyl sulfonate ester derived from CSA was rejected in
the crystallization. Urethane (23), a known animal carcinogen
formed from ethanolysis of 7 and 16, was removed in both the
aqueous washes and the crystallization step. Finally, monoethyl
sulfate (EtOSO3H), formed by esterification of H2SO4,29 was
removed in the aqueous NaHCO3 wash. Although the latter
could be eliminated by the use of anhydrous HCl, the use of a
liquid acid was deemed preferable for safety and simplicity in
handling at large scale.
Use of toluene as the reaction solvent for the three-step
sequence permits direct crystallization of triazolone 8 after
aqueous workup, upon concentration and addition of n-heptane,
without the need for a solvent exchange. Due to the strong
temperature dependence of the solubility of 8 in 1:2 toluene/
heptane,30 we felt a more robust crystallization could be
achieved by an initial antisolvent addition followed by cooling.
After concentration of the toluene solution, n-heptane was added
at 45 °C giving a 65:45 ratio (v/v) of toluene/heptane. The
solution was seeded at 37 °C (∼30% supersaturation), and the
remaining heptane was added over 2.5 h, resulting in a 33:67
ratio (v/v) of toluene/heptane. Data from the Lasentec FBRM
probe (Figure 8), indicate that most of the crystal growth
occurred during the antisolvent addition. These data also show
that no change in the distribution of particle chord lengths
occurred as the slurry was cooled from 37 to 10 °C over 3 h.
The product precipitated as clear, rectangular plates that filtered
rapidly. Data from the FBRM probe also showed that no
10-Camphorsulphonic acid was found by Schmid to be
preferred for promoting the cyclization of 7 to triazolone 8.9
An excess of CSA (1.7 equiv) was used to keep the reaction
rate from slowing dramatically, as the imine nitrogen in the
product (pKa ≈ 2.8) was protonated under the strongly acidic
conditions. Typically the reaction was complete after 3 h at
100 °C or 17 h at 70 °C. Similar product profiles were obtained
at both temperatures: triazolone 8 (60-70%), hydrazide 25
(10-15%), API 1 (5-10%), carboxylic acid 5 (5-10%), and
oxadiazolone 24 (2%), along with several smaller impurities
(Scheme 5). The 5-min UPLC assay described earlier dramati-
cally aided the development of this reaction sequence, as the
origin and fate of over 30 components needed to be understood.
Hydrazide 25 was the most problematic impurity. Previously,
it had been removed by adsorption on Amberlyst 15 resin, as
it was not rejected by crystallization and was poorly removed
by acidic extraction. Furthermore, it was not possible to reduce
the levels of this impurity, as cleavage of the carboxamide group
in 7 is surprisingly facile, occurring under both acidic and basic
conditions and at temperatures above 100 °C. We suspect this
lability arises from anchimeric assistance by the oxygen of the
internal amide group. Unfortunately, attempts to verify this
hypothesis via NMR detection of the putative O-acyl intermedi-
ate were unsuccessful.
The critical discovery for impurity control in this process
was somewhat counterintuitive, namely, addition of a strong
acid (HCl or H2SO4) with EtOH to the crude reaction mixture
and heating at reflux.28 This converted numerous impurities that
could not be rejected by crystallization into a smaller number
of components that were readily rejected by either extraction
or crystallization. API 1 was esterified to give 8. All amide-
containing species and unreacted 5 were converted into diethyl
ester 28 which was readily rejected by crystallization (Scheme
(29) A stock solution of 2.5 M H2SO4 in EtOH exists as an 80:20 mixture
of EtOSO3H and H2SO4, based upon integration of the quartets at 4.2
ppm and 3.8 ppm, respectively. The equilibrium ratio is a function of
the amount of water present. Diethyl sulfate (q, 4.3 ppm) is not formed
under these conditions, and when spiked into the solution it is
converted to the monoethyl sulfate. Lit. for 1H NMR shift for
HO3SO13CH2CH3 4.25 ppm (dq): Sen, A.; Benvenuto, M.; Lin, M.;
Hutson, A.; Basickes, N. J. Am. Chem. Soc. 1994, 116, 998–1003.
(30) Plots of the solubility vs temperature and solubility vs percent of
n-heptane are included in the Supporting Information.
(28) It is well-known that ethyl hydrogen sulfate (EtOSO3H) is formed
upon reaction of H2SO4 with EtOH. For kinetic studies, see: (a)
Theodore, S.; Sai, P. S. T. Can. J. Chem. Eng. 2001, 79, 54–64. (b)
Chen, L.; Johnson, B. D.; Grinberg, N.; Bicker, G. R.; Ellison, D. K.
J. Liq. Chromatogr. Relat. Technol. 1998, 21, 1259–1272. (c) Clark,
D. J.; Williams, G. J. Chem. Soc. 1957, 4218–4221. (d) Deno, N.;
Newman, M. J. Am. Chem. Soc. 1950, 72, 3852–3856.
138
•
Vol. 13, No. 2, 2009 / Organic Process Research & Development

Figure 9. Influence of the crystallization cooling rate on the
MSZW at a concentration of 1 g LY518674 per 13 mL of
EtOAc.
Figure 8. In situ FBRM data collected during crystallization
of 8.
less than 4% product would be lost to the filtrate. Levels of
HOAc and EtOH were typically 0.69% (w/w) and 0.03% (w/
w), respectively and were very consistent.
attrition of the crystals occurred upon prolonged stirring (8 h),
but only the first two hours of data are included in Figure 8.
The robustness of the control strategy was demonstrated in
three, 7 kg pilot-plant lots that afforded triazolone 8 in 64 (
1% yield and with total related substance levels of 0.33 ( 0.02%
for the three-step sequence.
Knowledge of the metastable zone width (MSZW), the
difference between the dissolution and spontaneous nucleation
points as a function of a solubility-influencing parameter such
as temperature and at given substrate load, is critical to design
a robust crystallization process.34 This system possesses a broad
MSZW and a steep solubility/temperature dependence with a
high supersaturation level (471%), even in the midpoint of
MSZW. Seeding, therefore, became indispensable for such a
system to start a heterogeneous nucleation at the desired
supersaturation ratio to prevent very heavy, yet uncontrollable,
nucleation from happening. The MSZW is also a function of
the cooling rate as shown in Figure 9, increasing significantly
at rates above 0.4 °C/min. The optimum crystallization condi-
tions consisted of seeding (1% w/w load) at 60 °C (30%
supersaturation) followed by cooling at a rate of 0.2 °C/min to
5 °C. Control of the crystallization process was also essential
for obtaining reproducibly low levels (<0.2%) of residual
EtOAc in the API, which varied from 0.5-1.5% at cooling rates
above 0.5 °C/min and if the crystallization process was not
seeded.
Implementation of the process at 0.1-4.5 kg scale afforded
LY518674 (1) in 95.4 ( 0.6% yield from 8. The crystallization
process proved to be quite robust, giving comparable product
from three lots, both in terms of its chemical purity (0.18 (
0.02% TRS) and its physical properties (d90 (128-160 µm),
d50 (38-58 µm), d10 (12-19 µm)). Importantly, the high
product quality obtained directly from the initial crystallization
obviates the need for a recrystallization, thereby eliminating any
additional risk of worker exposure to the highly potent API.
Development of the API Crystallization. Development of
the final step was straightforward, saponification of 8 with
aqueous NaOH at room temperature followed by crystallization
of LY518674 (1) from EtOAc (Scheme 1). It was found that a
biphasic hydrolysis, as described in the original protocol, was
not required, so toluene was removed. However, addition of
the EtOAc prior to neutralization was found to be highly
advantageous, keeping the product in solution throughout the
acidification process. LY518674 (1) contains three ionizable
groups with pKa values of 9.0 (triazolone ring (N-4)), 3.8
(carboxylic acid), and 2.8 (triazolone ring (N-1)).31 The
monoanion is only sparingly soluble and tended to form a gum
at around pH 7 if the organic solvent was not present.
Ethyl acetate, an ICH Class 3 solvent,32 is uniquely suited
for use in both the extraction and the crystallization. The EtOH
released by the hydrolysis of both EtOAc and 8 dramatically
increased the solubility of the product in the organic layer,
thereby minimizing processing volumes.33 Yet upon removal
of water and EtOH by distillation, LY518674 was isolated in
96% yield via a seeded, temperature-controlled crystallization
process.
Technology played a key role in our ability to rapidly
optimize the API crystallization process. The Crystal16 auto-
mated solubility station by Avantium Technologies was used
to determine the impact of water, EtOH, and HOAc on product
solubility. It was found that stopping the distillation with 13
volumes of solvent and less than 0.4% H2O ensured that
spontaneous nucleation would not occur above 70 °C and that
Conclusions
Several new technologies were used during the development
of a manufacturable route for PPARR agonist LY518674 (1).
Their use allowed for a higher level of process knowledge to
be obtained during development while simultaneously increas-
ing productivity. As a result, a control strategy was developed
in which all six API impurities (9-14) were controlled to below
(31) The pKa values were measured by potentiometric titrations at three
concentrations of dioxane water and extrapolating to pure water using
a Yasuda-Shedlovsky plot. Baertschi, S.; et al. Unpublished results,
Eli Lilly and Company.
(32) Residual levels up to 0.5% are acceptable for Class 3 solvents without
justification (ref 18).
(33) At 23 °C, the solubility of 1 is 3.7 mg/mL in anhydrous EtOAc vs
38.1 mg/mL in water-saturated EtOAc and 100 mg/mL in water-
saturated EtOAc containing 2 wt % EtOH (see ref 9). Acetic acid is
also formed during the hydrolysis of EtOAc but does not significantly
impact the solubility of 1 in EtOAc.
(34) Mersmann, A., Ed. Crystallization Technology Handbook, 2nd ed.;
Marcel Dekker: New York, 2001.
Vol. 13, No. 2, 2009 / Organic Process Research & Development
•
139

the ICH reporting threshold of 0.05%.35 Furthermore, the
acquired process knowledge enabled quality to be designed into
each step such that the total impurity load for each synthetic
intermediate was less than 0.5%.
The combination of sub-2 µm column technology and the
higher pressures accessible with the Waters ACQUITY UPLC
greatly increased analytical efficiency by allowing over 30
components to be tracked with a single, 5-min, MS-friendly
assay throughout the four-step process. Further productivity
gains were achieved with the use of chromatography-modeling
software to aid method development and data export methods
to expedite the conversion of chromatography data into mul-
ticomponent reaction profiles.15
Process analytical technology (PAT), such as online IR,
PVM, FBRM, and MS played a key role in acquiring the
process knowledge needed for development of the control
strategy and in allowing comparison of the process performance
at laboratory and pilot-plant scale. Importantly, the knowledge
gained will permit subsequent scaleup to occur without the
requirement of PAT for process control, except in instances
where doing so will limit worker exposure (e.g., monitoring of
drying operations by online MS).
min to remove oxygen prior to adding 15 (33.10 g, 191.7 mmol,
1.0 equiv), after which it was deoxygenated for an additional 5
min.
A solution of KOCN (16.20 g, 98% assay, 195.7 mmol, 1.02
equiv) in deoxygenated H2O (61 mL) was added over 60 min.
Ten minutes into the addition, the reaction was seeded with 16
(0.300 g, 1.67 mmol). When the addition was complete, the
lines were rinsed with deoxygenated H2O (19.4 mL), and the
slurry was stirred overnight at 20 °C.
Note: A 2 mol % excess of KOCN was used to avoid
concerns of approaching the limit of quantitation throughout
the reaction, whereas it is the limiting reagent under the standard
conditions. In addition, the n-BuOH:H2O ratio was reduced from
9:1 to 6.3:1.0 (v/v) to ensure miscibility of the two solvents
throughout the reaction.36
Development of the API Crystallization Process. The
research was conducted using an Avantium Technologies
Crystal16 automated solubility station. This is a medium-
throughput screening tool having a 4 × 4 array of 16 1.5 mL
microreactor vials, permitting the rapid acquisition of informa-
tion with minimal material requirements. Each microreactor is
equipped with a light-emitting diode (LED) and photosensor
for measuring turbidity in transmission (through the sample)
to determine dissolution point. The Crystal16 automated
solubility station was used to acquire solubility information and
to define the dependence of the MSZW on the cooling rate.
A Mettler Toledo MultiMaxIR automated parallel reactor
system equipped with a focused-beam reflectance measurement
(FBRM) probe was used for larger-scale crystallization devel-
opment for better simulation of process equipment and
parameters.
1-[(4-Methylphenyl)methyl]hydrazinecarboxamide (16).
Caution! Aqueous solutions of KOCN are subject to autocata-
lytic hydrolysis and should not be stored.37 All solutions were
prepared immediately prior to use and were maintained at 20
°C in a jacketed vessel at all times. Note: It is important to
deoxygenate all solutions prior to use and to conduct the reaction
under an inert atmosphere to prevent oxidative degradation of
15.38
Experimental Section
General. All reactions were run under a nitrogen atmo-
sphere, unless otherwise specified. Reagents were used as
received unless otherwise noted. Proton NMR spectra were
obtained at 400 MHz and carbon NMR spectra were obtained
at 100.6 MHz. NMR chemical shifts are reported in δ units
referenced to residual proton signals in the deuterated solvent.
All yields are corrected for chemical purity of both the limiting
reagent and the product (i.e., yield ) (weight of product ×
purity/MW of product)/(weight of limiting reagent × purity/
MW of limiting reagent) × 100). If the purity of a product is
not specified, it is greater than 99%. Analytical methods used
for reaction monitoring and purity determination are described
in the Supporting Information.
Development of the Kinetic Model for Formation of 16.
The reaction was conducted in a Mettler AutoChem RC1e
reaction calorimeter with an 800-mL glass vessel equipped with
a ReactIR 4000 FTIR with a SiComp attenuated total reflectance
probe sensor. Calibration was achieved using freshly prepared
standard solutions of KOCN in an n-BuOH:H2O solvent system
matching that used in the reaction at 20 °C. The cyanate profile
was generated by integrating over the spectral range of
2088-2254 cm-1 after spectral preprocessing to remove the
underlying water absorbance. The resulting calibration plot
exhibits a significant departure from linearity but is amply fit
by a quadratic function. See the Supporting Information for
details on the calibration and determination of the limits of
detection and quantitation. IR data were collected throughout
the course of the reaction. Following preprocessing, the spectral
signal for cyanate was converted to concentration using the
calibration model.
(4-Methylphenyl)hydrazine hydrochloride (11.5 kg, 66.6
mol, 1.02 equiv)16 was added to a deoxygenated solution
(36) (a) For a discussion of the liquid-liquid-solid equilibria for the ternary
system of n-BuOH/water/KCl, see: (b) Gomis, V.; Ruiz, F.; Asensi,
J. C.; Saquete, M. D. J. Chem. Eng. Data 1996, 41, 188–191.
(37) (a) For a detailed investigation of an explosion of 30% aqueous KOCN
that occurred in a vented drum, see: Pinsky, M. L.; Vickery, T. P.;
Freeman, K. P. J. Loss PreV. Process Ind. 1990, 3, 345–348. (b) A
major factor in the drum explosion was the heating of the aqueous
mixture to 50 °C in order to dissolve all of the solids, prior to storage.
However, calorimetry experiments demonstrated that a 30% aqueous
solution of KOCN is unstable even at room temperature, reaching its
maximum decomposition rate after 30 h at 25 °C due to autocatalysis
by HCO3-. See also: Urben, P. G., Ed. Bretherick’s Handbook of
ReactiVe Chemical Hazards, 6th ed.; Butterworth-Heinemann: Oxford,
Boston, 1999; Vol. 1, pp 201-202.
(38) Oxygen must be excluded from all process streams to minimize
formation of the insoluble hydrazone 18, resulting from reaction of
4-methylbenzaldehyde (from the oxidation of 15) with 16, and
semicarbazone 19, resulting from its subsequent reaction with KOCN.
Solutions were deoxygenated by purging them with N2 (1.18 ft3 N2
L-1 of solution) to an O2 level of approximately 0.4 ppm. Initial O2
levels ranged from about 9 ppm in the laboratory to 3.4 ppm in the
pilot plant. Typically, 0.5 h was required for the deoxygenation of
laboratory runs, and approximately 1 h was required for deoxygenation
in the pilot plant.
Water (455 mL) and 1-butanol (33.5 mL) were mixed.
Argon was sparged through the solution for approximately 20
(35) ICH Harmonised Tripartite Guideline. Impurities in New Drug
Substances Q3A (R2), revised October 2006. Available on-line at:
http://www.ich.org/cache/compo/363-272-1.html#Q3A(R).
140
•
Vol. 13, No. 2, 2009 / Organic Process Research & Development

containing n-BuOH (3.94 kg, 17.21 L) and H2O (154 L) at 20
°C. The solution was transferred through a 0.45 µm in-line
filter,39 and the line was rinsed with deoxygenated H2O (6.0
L). A deoxygenated aqueous solution of KOCN was prepared
by adding KOCN (5.60 kg, 94.3% assay,40 65.1 mol, 1.0 equiv)
to H2O (21.0 L) with stirring at 20 °C and purging the resulting
solution with N2.
The aqueous KOCN was added to the hydrazine solution at
20 °C in two portions in order to saturate the reaction mixture
with product prior to seeding. The first portion (0.12 equiv)
was added over 10 min, and the reaction was stirred for 0.75 h.
The reaction was then seeded with a slurry of 16 (0.1 kg, 0.56
mol, 0.008 equiv) in n-BuOH (0.10 L) and H2O (0.40 L). The
container was rinsed with H2O (0.80 L). The remaining aqueous
KOCN was added at a rate of 0.016 equiv/min until 60% of
the total KOCN had been added. The rate was slowed to 0.0046
equiv/min during the remainder of the addition, giving a total
addition time of 2 h.41 The combination of solvent composition
and addition rate permits the two isomeric products to crystallize
independently. The line was rinsed with deoxygenated H2O (6.0
L), and the reaction slurry was stirred for 12 h at 20 °C.42 The
reaction is quantitative giving a slurry containing an 80:20
mixture of the desired semicarbazide 16 and the isomeric 17,
respectively.
2 equiv) at 25 °C. The resulting slurry was heated at reflux
(76.8 °C) for 1 h. Ethyl 2-bromoisobutyrate (31.8 kg, 163.0
mol, 2.25 equiv) was added. An additional charge of NaOEt
(23.6 kg of 21% NaOEt in EtOH, 72.8 mol, 1 equiv) was added
over 1 h.45 When less than 1% starting material remained by
HPLC, the reaction mixture was cooled to 25 °C and transferred
into a mixture of aqueous 0.5 N HCl (3.72 kg 35% HCl, 35.7
mol, 0.5 equiv + 68 kg of H2O) and toluene (65 L) at 22 °C.
The pH was adjusted from 6.55 to 8.39 with the addition of
50% NaOH with stirring. After phase separation, the aqueous
phase was washed with toluene (65 L). Toluene (130 L) was
added, and the pH of the aqueous phase was adjusted from 8.77
to 1.99 by the addition of HCl (13.4 kg of 35% HCl, 128.6
mol, 1.77 equiv) with stirring at 23 °C. The organic phase was
washed sequentially with H2O (66 L) and 0.078 M NaHCO3
(0.44 kg, 5.24 mol, 0.07 equiv in 67 L of H2O).46 The organic
layer was concentrated to approximately 6 volumes by vacuum
distillation. After cooling to 20 °C, heptane (40 L) was added,
followed by seed crystals of 5 (0.19 kg, 0.64 mol, 0.09 equiv)
and the resulting slurry was stirred for 0.5 h. Additional heptane
(175-180 L) was added with stirring over at least 3 h. The
resulting slurry was stirred at 20 °C for 2 h and at 2 °C for 2 h
prior to filtration. The wet cake was washed with 6:1 (v/v)
heptane/toluene (2 × 35 L). Acid 5 (18.64 kg, 99.0% assay,
99.76% purity, 86.3% yield) was obtained as an off-white
powder after vacuum drying for approximately 12 h at 45 °C/
100 mmHg, mp 70.6-71.5 °C. Anal. Calcd for C16H22O5: C,
65.29; H, 7.53; O, 27.18. Found: C, 65.18; H, 7.37. 1H NMR
(DMSO-d6: δ 1.14 (t, J ) 7.3 Hz, 3H), 1.47 (s, 6H), 1.73
(quintet, J ) 7.5 Hz, 2H), 2.17 (t, J ) 7.5 Hz, 2H), 2.47-2.51
(m, 2H), 4.13 (q, J ) 7.1 Hz, 2H), 6.69-6.70 (m, 2H),
7.04-7.07 (m, 2H), 12.01 (s, 1H, D2O exch). 13C NMR
(DMSO-d6: δ 14.3, 25.5 (2C), 26.8, 33.5, 34.0, 61.4, 79.0, 119.3
(2C), 129.5 (2C), 135.6, 153.6, 173.7, 174.7. IR (KBr): 3443,
3072, 3003, 2977, 2926, 1733, 1704, 1610, 1582, 642, 591
cm-1. HRMS (m/z, M + 1): Calcd for C16H23O5: 295.1540.
Found: 295.1532.
Hydrochloric acid (3.89 L of 5.04 M, 0.30 equiv) was added
to the slurry (pH 8.83) at 20 °C to dissolve the undesired isomer.
The transfer line was rinsed with H2O (7.3 L), and the reaction
mixture was stirred for 3 h at 20 °C (final pH 2.31). The slurry
was filtered, and the wet cake was washed with 35 °C H2O (3
× 27.5 L). The product was dried overnight at 60 °C/100
mmHg to give 7.045 kg of 16 (58.6% yield, 98.4% assay, 0.07%
H2O, 0.11% 17, 0.10% 20)43 as a white solid, mp 152.5-155.2
°C. Anal. Calcd for C9H13N3O: C, 60.32; H, 7.31; N, 23.45; O,
1
8.93. Found: C, 60.31; H, 7.20; N, 23.38. H NMR (DMSO-
d6: δ 2.25 (s, 3H), 4.20 (s, 2H), 4.18 (s, 2H), 6.10 (br s, 2H),
7.10 (m, 4H). 13C NMR (DMSO-d6: δ 21.1, 52.3, 128.3 (2C),
129.4 (2C), 135.4, 136.5, 160.5. IR (KBr): 3424, 3333, 3045,
3023, 2978, 2361, 2333, 1907, 1653, 1576, 903, 846 cm-1.
HRMS (m/z M + 1): Calcd for C9H14N3O: 202.0951. Found:
202.0946.
4-(2-Ethoxy-1,1-dimethyl-2-oxoethoxy)benzenebutanoic
Acid (5). 4-Hydroxyphenylbutanoic acid (13.06 kg, 72.48 mol,
1 equiv)10 and EtOAc (6.6 kg, 74.9 mol, 1.03 equiv)44 were
added to NaOEt (46.88 kg of 21% NaOEt in EtOH, 144.7 mol,
2-[4-[3-[2,5-Dihydro-1-[(4-methylphenyl)methyl]-5-oxo-
1H-1,2,4-triazol-3-yl]propyl]phenoxy]-2-methylpropanoic
Acid, Ethyl Ester (8). Caution! PPARR agonist LY518674
(1) and its ester derivative 8 are highly potent compounds5,8
and should be handled with extreme caution to prevent worker
exposure.
Oxalyl chloride (2.25 kg, 17.7 mol, 1.20 equiv) was added
to a solution of 5 (4.33 kg, 14.71 mol, 1.00 equiv), pyridine
(1.2 mL, 0.015 mol, 0.0010 equiv), and toluene (25 L) at 22
°C over 10 min. The lines were rinsed with toluene (1 L) and
the reaction was stirred for at least 1.5 h until less than 0.5%
of the starting acid remained.47 The reaction was concentrated
to approximately 8-10 L by vacuum distillation (∼60 °C/
100-150 mmHg) to remove excess oxalyl chloride.
(39) In-line filtration removes any hydrazone impurity that formed due to
oxidative degradation of the starting hydrazine.
(40) The KOCN also contains 0.6% K2CO3 (w/w). The assay for KOCN
was determined by acid-base titrations using vendor-provided proce-
dures. The assay is a two-part procedure. Part 1 involves titration of
residual potassium carbonate (K2CO3) vs H2SO4. Part 2 involves
heating a cyanate solution with H2SO4, followed by a back-titration
of excess (unreacted) H2SO4 with sodium hydroxide. Final results are
calculated by subtracting the amount of K2CO3 from the titration results
in Part 2.
(41) The KOCN solution may be added at a constant rate over 1 h.
(42) The reaction time can be determined using the kinetic model described
in this paper. Typically, the reaction should be stirred at least 6 h to
ensure complete conversion of KOCN.
(43) See Supporting Information for HPLC and UPLC purity methods,
including retention times and relative response factors for individual
components.
(45) Some of the NaOEt was consumed by dehydrobromination of the
alkylating agent.
(46) The NaHCO3 typically removes 1.5-2% residual 4, 1.5-2% of diacid
10, and 3-5% 5. Using a slight excess eliminates the need for an
in-process assay.
(47) Samples were quenched into MeOH containing 1% pyridine (v/v) and
analyzed by the UPLC purity method. See the Supporting Information
for method conditions.
(44) EtOAc was added to scavenge any hydroxide in order to prevent ester
hydrolysis during the reaction.
Vol. 13, No. 2, 2009 / Organic Process Research & Development
•
141

The resulting acid chloride solution was added to a slurry
of semicarbazide 16 (2.64 kg, 98.4% assay, 14.49 mol, 1.0
equiv) in pyridine (1.51 kg, 19.1 mol, 1.30 equiv) and toluene
(34 L) over 0.5 h while maintaining the temperature below 40
°C. The lines were rinsed with toluene (4 L) and the reaction
was stirred at 25 °C until the reaction was complete (at least
1 h).48
Sodium hydroxide (25.7 kg of 1 N NaOH, 25 mol, 2.19
equiv) was added to 8 (4.99 kg, 11.40 mol, 1.0 equiv). The
resulting pale-yellow solution was stirred at 25 °C for 6 h until
<0.1% of 8 remained by HPLC. Ethyl acetate (60 L) was
added, and the resulting mixture was stirred for at least 3 h
(pH 9.97 at end of stir time). The pH was adjusted to 1.75 by
addition of 5 N HCl (5.30 kg, 24.7 mol, 0.99 equiv). The
organic phase was washed with H2O containing 2% EtOH by
weight (20 kg H2O + 0.4 kg EtOH).33 The organic phase was
polish filtered through a 20 µm cartridge filter, and the line was
rinsed with EtOAc (2 L). The filtered solution was concentrated
to about 12 L (2.3 volumes) by vacuum distillation at 70-80
°C/740 mmHg. Additional EtOAc (25 L) was added, and the
distillation was continued to the same end point. After additional
EtOAc (48 L) was added and the temperature was adjusted to
60 °C, the solution was seeded with 1 (0.050 kg, 0.12 mol,
0.01 equiv). The mixture was stirred at 60 °C for 1 h before
cooling to 5 °C at a rate of 0.2 °C/min over 4.6 h. The slurry
was filtered after stirring for 1 h. The cake was washed with
two portions of 5 °C EtOAc (15 L each), the first with stirring
of the slurry, and the second without stirring. PPARR agonist
1 (4.49 kg, 99.6% assay, 99.85% purity, 96% yield) was
obtained as a white powder after vacuum drying for ap-
proximately 18 h at 60 °C/100 mmHg, mp 130.7-131.2 °C.
Anal. Calcd for C23H27N3O4: C, 67.46; H, 6.65; N, 10.26; O,
15.63. Found: C, 67.55; H, 6.75; N, 10.29. 1H NMR (DMSO-
d6: δ 1.48 (s, 6H), 1.82 (m, 2H), 2.27 (s, 3H), 2.38 (m, 3H),
2.50 (m, 2H), 4.74 (s, 2H), 6.76 (m, 2H), 7.07 (m, 2H), 7.14
(s, 4H), 11.46 (s, 1H, D2O exch), 12.99 (s, 1H, D2O exch). 13C
NMR (DMSO-d6: δ 21.1, 25.5 (2C), 26.0, 28.2, 33.8, 47.5, 78.7,
119.0 (2C), 127.9 (2C), 129.4 (2C), 129.5 (2C), 134.9, 135.0,
136.9, 146.4, 153.9, 154.73, 175.6. IR (KBr): 3085, 3033, 3013,
2994, 2949, 2930, 3500, 2500, 1731, 1683, 1611, 1589, 1507,
1472, 1437, 1264, 1232, 1180, 1147, 833, 808, 642 cm-1. UV
(MeOH): 205 nm (ε 21690), 220 nm (ε 19819), 275 nm (ε
891). HRMS (m/z, M - 1): Calcd for C23H26N3O4: 408.1929.
Found: 408.1927.
(()-10-Camphorsulfonic acid (6.83 kg, 29.40 mol, 2.00
equiv) was added, and the reaction mixture was heated to 100
°C and stirred for 3 h. The reaction mixture was cooled to 80
°C, and a 2.5 M of H2SO4 in EtOH (7.57 kg of 2.5 M solution,
19.7 mol, 1.34 equiv)49 was added over 10-15 min. Caution!
Care should be exercised in handling this solution, which is
corrosive and contains significant levels of EtOSO3H.28,29 The
reaction was heated at reflux (∼80 °C) for about 2 h, until 25
has decreased to <0.5% by area relative to triazolone 8. After
cooling to room temperature, the reaction mixture was trans-
ferred over 0.5 h into a solution of NaHCO3 (6.55 kg, 78.0
mol) in H2O (27 L) at a rate to control gas evolution and to
maintain the pH in the quench mixture above 5.5. The phases
were separated and the organic phase was washed with H2O
(27 L). After phase separation, the organic phase was concen-
trated to ∼13 L (3 volumes) by vacuum distillation 36-52 °C/
100-150 mmHg. After the addition of heptane (5.5 L, 3.76
kg), the crystallization was seeded with 8 (0.34 kg, 0.77 mol,
0.052 equiv) at 37 °C, and the mixture was stirred for 10 min.
Heptane (12.59 kg, 18.4 L) was added slowly over 2.5 h with
stirring, while maintaining the temperature at 37 °C. When the
addition was complete, the slurry was cooled with stirring to
10 °C over 3 h with a cooling rate of ∼8 °C/h and was stirred
for an additional 1 h. The solid was filtered and was washed,
with stirring, with a 10 °C solution of 3:1 n-heptane (15 L)/
toluene (5 L). This was followed by two cake washes (16 L
each), without stirring, with a 10 °C solution of 3:1 n-heptane/
toluene. Triazolone ester 8 (4.405 kg, 99.27% assay, 99.69%
purity, 63.6% yield) was obtained as a white crystalline solid
after vacuum drying for approximately 15 h at 50 °C/100
mmHg, mp 95.5-97.1 °C. Anal. Calcd for C25H31N3O4: C,
68.63; H, 7.14; N, 9.60; O, 14.63. Found: C, 68.43; H, 7.14;
N, 9.57. 1H NMR (DMSO-d6: δ 1.14 (t, J ) 7.1 Hz, 3H), 1.47
(s, 6H), 1.79 (quintet, J ) 7.6 Hz, 2H), 2.24 (s, 3H), 2.34 (m,
2H), 2.48 (m, 2H), 4.13 (q, J ) 7.1 Hz, 2H), 4.70 (s, 2H),
6.67-6.70 (m, 2H), 7.02-7.05 (m, 2H), 7.08-7.12 (m, 4H),
11.41 (s, 1H, D2O exch). 13C NMR (DMSO-d6: δ 14.3, 21.1,
25.5 (2C), 26.0, 28.1, 33.8, 47.48, 61.4, 79.0, 119.3 (2C), 127.9
(2C), 129.4 (2C), 129.5 (2C), 135.0, 135.3, 136.9, 146.4, 153.6,
154.7, 173.7. IR (KBr): 3436, 3097, 2990, 2921, 1726, 1698,
1611, 1510, 814, 756, 740, 722 cm-1. HRMS (m/z, M + 1):
Calcd for C25H32N3O4: 438.2387. Found: 438.2377.
CCDC 702044-702048 contains the supplementary crystal-
lographic data for this paper, which includes data for com-
pounds: 1, 5, 8, 16, and 17. These data can be obtained free of
charge via http://www.ccdc.cam.ac.uk/cgi-bin/catreq.cgi, by
emailing data_request@ccdc.cam.ac.uk, or by contacting The
Cambridge Crystallographic Data Centre, 12, Union Road,
Cambridge CB2 1EZ, U.K.; Fax: +44 1223 336033.
Acknowledgment
The development and scaleup of a multistep process involves
a large number of talented scientists and technical personnel.
We express our gratitude to the following individuals: Matthew
Allgeier, Dr. Todd Gillespie, Maryam Mostafavi, Jean-Claude
Garay, Dr. Thomas Zennie, and Lisa Zollars for providing
analytical support and assistance with the characterization of
API impurities; Dr. P. Y. Chen, Michael Heller, John Howell,
Jeffrey Lewis, Robert Metzler, Richard Miller, Gary Richardson,
Otis Williams, and the API Operations staff for their technical
assistance in process scaleup; Robert Waggoner for handling
all procurement activities for the program; Amanda McDaniel
for assistance with the kinetic modeling studies; Jeff Bieszki
and John Simms of Waters Corporation for their technical
2-[4-[3-[2,5-Dihydro-1-[(4-methylphenyl)methyl]-5-oxo-
1H-1,2,4-triazol-3-yl]propyl]phenoxy]-2-methylpropanoic Acid
(LY518674, 1). Caution! PPARR agonist LY518674 (1) and
its ester derivative 8 are highly potent compounds5,8 and should
be handled with extreme caution to prevent worker exposure.
(48) Analytical data for compound 7 is included in the Supporting
Information.
(49) A stock solution was prepared by the slow addition of 98% H2SO4
(6.51 kg, 65.0 mol) to absolute 2B-3 EtOH (23 L, 394 mol) and stored
at room temperature.
142
•
Vol. 13, No. 2, 2009 / Organic Process Research & Development

assistance with the Waters ACQUITY UPLC system; Dr.
Gregory Stephenson for X-ray crystal structure analysis; and
Dr. Alain Merschaert for his helpful technical discussions.
Finally, we thank the management team of Chemical Product
Research and Development at Eli Lilly and Company for
providing the state-of-the-art development research equipment
used on this project.
factorial design used for process optimization leading to 6; (5)
HPLC and UPLC analytical methods and relative response
factors for several components; (6) solubility data used during
crystallization development for triazolone 8; (7) spectral char-
acterization and experimental procedures for the synthesis of
the following compounds: API impurities (9-14) and their
synthetic precursors, nonisolated intermediate 7, and 17-22,
24, 25, 27, and 28. This material is available free of charge via
the Internet at http://pubs.acs.org.
Supporting Information Available
(1) Outline for using an export method for rapid reaction
profile generation; (2) additional details related to development
of the kinetic model for reaction of KOCN with hydrazine 15;
(3) comparison of the reaction profiles using pyridine and DMF
as catalysts for acid chloride formation; (4) the fractional
Received for review September 30, 2008.
OP8002486
Vol. 13, No. 2, 2009 / Organic Process Research & Development
•
143