Development of a robust ring-closing metathesis reaction in the synthesis of SB-462795, a cathepsin K inhibitor

Article abstract of DOI:10.1021/op700288p

The development of a robust and high-yielding ring-closing metathesis (RCM) reaction and its demonstration on multikilogram scale are described. A detailed understanding of the impact of impurities on the RCM reaction was achieved using a variety of chemical and statistical methods. Specifically, individual impurities were evaluated in spiking studies to identify those that negatively affected the RCM reaction. Projection methods (PCA and PLS) were applied to historical data to identify the main sources of variation in starting material quality and determine the main detrimental impurities that impeded the RCM reaction. The synthesis of the starting material was then modified to adequately control these key impurities, which in turn ensured a robust RCM process. Finally, the robustness of the RCM reaction was assessed using a probability-based approach.

Full text of DOI:10.1021/op700288p

Organic Process Research & Development 2008, 12, 226–234
Development of a Robust Ring-Closing Metathesis Reaction in the Synthesis of
SB-462795, a Cathepsin K Inhibitor
Huan Wang,*,† Steven N. Goodman,† Qunying Dai,† Gregory W. Stockdale,‡ and William M. Clark, Jr.†
Chemical DeVelopment and Statistical Sciences, GlaxoSmithKline, 709 Swedeland Road, King of Prussia,
PennsylVania 19406, U.S.A.
Scheme 1. Synthesis of SB-462795 via the RCM
disconnection
Abstract:
The development of a robust and high-yielding ring-closing
metathesis (RCM) reaction and its demonstration on multikilo-
gram scale are described. A detailed understanding of the impact
of impurities on the RCM reaction was achieved using a variety
of chemical and statistical methods. Specifically, individual impuri-
ties were evaluated in spiking studies to identify those that
negatively affected the RCM reaction. Projection methods (PCA
and PLS) were applied to historical data to identify the main
sources of variation in starting material quality and determine
the main detrimental impurities that impeded the RCM reaction.
The synthesis of the starting material was then modified to
adequately control these key impurities, which in turn ensured a
robust RCM process. Finally, the robustness of the RCM reaction
was assessed using a probability-based approach.
development for the treatment of osteoporosis and osteoarthritis,3
clearly exemplifies this point. The formation of the tetrahy-
droazepine ring in 2 is effected by RCM of diene 1 using
Hoveyda’s second-generation catalyst4 in toluene. With pure
diene 1, this RCM reaction can be performed with very low
loadings (0.1–0.2 mol %) of the catalyst; however, the reaction
conversion suffers when 1 is of reduced purity. Therefore,
significant effort was directed at identifying the impurities which
reduced conversion and at controlling their content in 1 in order
to afford a robust RCM reaction.
Introduction
During the past decade, the ring-closing metathesis (RCM)
reaction has emerged as a powerful tool for C-C bond
formation.1 Not only has this methodology been widely used
in academic research, but it has been extended to include large-
scale synthesis in a pharmaceutical setting.2 The sensitivity of
the RCM reaction to impurities is a key challenge for developing
a robust industrial process, since the efficiency of the reaction
must be balanced by the cost and practicality of controlling the
purity of the substrate on large scale. Consequently, understand-
ing the impact of key impurities can greatly influence the
reliability of the RCM reaction.
Results and Discussion
Synthesis of Diene 1. The synthesis of diene 1 is detailed
in Scheme 2.5 Tosylate 3 is converted to epoxide 4 in toluene
using DBU as base, and excess DBU is neutralized with
aqueous citric acid. After aqueous washes and a solvent
The RCM transformation depicted in Scheme 1, utilized in
the synthesis of SB-462795, a potent cathepsin K inhibitor under
(3) (a) Kumar, S.; Dare, L.; Vasko-Moser, J. A.; James, I. E.; Blake, S. M.;
Rickard, D. J.; Hwang, S.-M.; Tomaszek, T.; Yamashita, D. S.;
Marquis, R. W.; Oh, H.; Jeong, J. U.; Veber, D. F.; Gowen, M.; Lark,
M. W.; Stroup, G. Bone 2007, 40, 122. (b) Yamashita, D. S.; Marquis,
R. W.; Xie, R.; Nidamarthy, S. D.; Oh, H.-J.; Jeong, J. U.; Erhard,
K. F.; Ward, K.W.; Roethke, T. J.; Smith, B. R.; Cheng, H.-Y.; Geng,
X.; Lin, F.; Offen., P. H.; Wang, B.; Nevins, N.; Head, M. S.;
Haltiwanger, R. C.; Narducci Sarjeant, A. A.; Liable-Sands, L. M.;
Zhao, B.; Smith, W. W.; Janson, C. A.; Gao, E.; Tomaszek, T.;
McQueney, M.; James, I. E.; Gress, C. J.; Zembryki, D. L.; Lark,
M. W.; Veber, D. F. J. Med. Chem. 2006, 49, 1597. (c) Cummings,
M. D.; Marquis, R. W.; Ru, Y; Thompson, S. K.; Veber, D. F.;
Yamashita, D. S. PCT Int. Appl. WO 2001070232 A1, 2001.
(4) Garber, S. B.; Kingsbury, J. S.; Gray, B. L.; Hoveyda, A. H. J. Am.
Chem. Soc. 2000, 122, 8168.
* To whom correspondence should be addressed. Tel: 610-270-5362. Fax:
610-270-4022. E-mail: huan.2.wang@gsk.com.
† Chemical Development, GlaxoSmithKline.
‡ Statistical Sciences, GlaxoSmithKline.
(1) For general reviews on ring-closing metathesis: (a) Hoveyda, A. H.;
Zhugralin, A. R. Nature 2007, 450, 243. (b) Schrock, R. R. Angew.
Chem., Int. Ed. 2006, 45, 3748. (c) Grubbs, R. H. Angew. Chem., Int.
Ed. 2006, 45, 3760. (d) Grubbs, R. H. Handbook of Metathesis; Wiley:
Weinheim, 2003 and references cited therein.
(2) (a) Yee, N. K.; Farina, V.; Houpis, I. N.; Haddad, N.; Frutos, R. P.;
Gallou, F.; Wang, X.-j.; Wei, X.; Simpson, R. D.; Feng, X.; Fuchs,
V.; Xu, Y.; Tan, J.; Zhang, L.; Xu, J.; Smith-Keenan, L. L.; Vitous,
J.; Ridges, M. D.; Spinelli, E. M.; Johnson, M.; Donsbach, K.; Nicola,
T.; Brenner, M.; Winter, E.; Kreye, P.; Samstag, W. J. Org. Chem.
2006, 71, 7133. (b) Nicola, T.; Brenner, M.; Donsbach, K.; Kreye, P.
Org. Process Res. DeV. 2005, 9, 513.
(5) The synthesis of tosylate 3 and sulfonamide 5, as well as a discussion
of BTPP-catalyzed epoxide ring-openings, will be disclosed in separate
publications.
226
•
Vol. 12, No. 2, 2008 / Organic Process Research & Development
10.1021/op700288p CCC: $40.75
2008 American Chemical Society
Published on Web 02/27/2008

Scheme 2. Synthesis of diene 1
Scheme 3. Formation of secondary amine 6
high levels of purity by HPLC (>98% PAR). However, despite
extensive efforts to optimize the process conditions and
consistently prepare the product with high purity, different
batches of diene 1 thus prepared gave extremely inconsistent
results in the subsequent RCM reaction, with conversions
ranging from 2% to 100%. The capricious RCM performance
of diene 1 prompted a series of detailed studies to identify the
impurities responsible for this variability.
Identification of Impurities that Negatively Impact RCM.
Spiking Experiments. In order to identify the impurities that
negatively impacted the RCM reaction, an extensive series of
spiking experiments was conducted. All known compounds
which were a part of the diene preparation, either as input
reagent, solvent, or byproduct, were charged in substoichio-
metric amounts to individual RCM reactions using a batch of
diene 1 that gave complete conversion in RCM with 0.2 mol
% catalyst. Highlighted results from these studies are sum-
marized in Scheme 4.
While many suspected agents had no impact on the metath-
esis of 1, five impurities were shown to significantly retard the
RCM process. Not surprisingly, these impurities all contain
either a basic functionality (BTPP, DBU, and secondary amine
6) or a chelating structure (urea 8 and 2-propyl ester 9, vide
infra). It is worth noting that secondary amine 6 and 2-propyl
ester 9 also underwent RCM themselves under the reaction
conditions, albeit with much lower efficiency: in the presence
of 5 mol % of Hoveyda’s catalyst in reflux toluene, the RCM
of amine 6 and ester 9 reached 14% and 53% conversion,
respectively. More interestingly, urea 8 did not undergo RCM
when exposed to Hoveyda’s catalyst, even though the reaction
between the two seemed facile as 2-isopropoxystyrene was
readily generated. Presumably, the urea carbonyl in the resultant
alkylidene complex preferentially chelated to the metal center
and inhibited the coordination of a second olefin as well as the
ring closure. An attempt to isolate this alkylidene complex,
however, was not met with success.
exchange from toluene to 2-propanol, epoxide 4 is coupled with
sulfonamide 5 in 2-propanol in the presence of a catalytic
amount of phosphazene base P1-tert-butyl-tris-(tetramethylene)
(BTPP). The reaction is complete in 3–5 h with 90–91% crude
peak area ratio (PAR) of 1 by HPLC. Aqueous citric acid is
then added to solubilize BTPP in the aqueous solution, and a
cooled, seeded crystallization delivers the desired product diene
1 in 80% yield.
The formation of the desired product 1 was accompanied
by the generation of secondary amine 6, which increased to
3–5% over prolonged reaction times. This was especially true
when high amounts of BTPP were used at high temperatures.
Presumably, as the reaction progressed and the concentration
of sulfonamide 5 decreased, deprotonation of the product
hydroxyl by BTPP became a competitive side reaction. The
resultant alkoxide could attack the pyridine ring, leading to
irreversible loss of SO2 and pyridyl migration to the carbinol
(Scheme 3).
Design of experiments (DoE)6 was used to assess the
relationship between the reaction conditions and the reaction
profile of the coupling between 4 and 5. The resultant model
indicated that increased concentration would permit lower BTPP
loading and minimize formation of 6, which was verified by
scale-up of these conditions to multigram scales. This procedure
was used to prepare over 20 batches of 1 which all showed
Statistical Analysis of Batches. With five possible culprits
identified from spiking experiments, it was necessary to
determine the extent to which these impurities were present in
different batches of 1. The impurity levels in each batch7 are
tabulated with its corresponding performance in the RCM
reaction, and a graphical depiction of representative data is
shown in Figure 1. Although spiking experiments showed that
(6) For detailed discussion on DoE methodologies, see: (a) Box, G. E. P.;
Hunter, J. S.; Hunter, W. G. Statistics for Experimenters: Design,
InnoVation and DiscoVery, 2nd ed.; Wiley: New York, 2005. (b)
Anderson, M. J.; Whitcomb, P. J. DoE Simplified: Practical Tools
for EffectiVe Experimentation; Productivity: OR, 2000.
(7) Due to the large difference in UV response for these impurities, all
laboratory batches of diene 1 were analyzed using LC-MS, and
impurity contents were calculated using calibrated mass signals.
Vol. 12, No. 2, 2008 / Organic Process Research & Development
•
227

Scheme 4. Spiking studies: Effect of impurities on RCM
conversion
Figure 1. Correlation of impurity content of 1 and performance
in the RCM reaction (0.2 mol % catalyst).
the stronger impact it has on that principal component. For
instance, urea 8 and ester 9 have the largest p1 values, indicating
that their amounts have the largest influence on the first principal
component (t1). Similarly, the second principal component t2
is mainly influenced by the amounts of urea 8 and 2-propyl
ester 9 as well. Even without any correlation to RCM
performance, this indicated that urea 8 and 2-propyl ester 9
accounted for most of the batch-to-batch variation in purity and
the content level of these two impurities needed to be controlled
and reduced.
While PCA only implicates potential sources of variation
in starting material purity, PLS further correlates this variation
to the conversion of 1 in the RCM reaction by building a
multivariate model between the two. The coefficients of the
model are then translated back to the originally measured
impurity levels to indicate the magnitude of impact from each
on RCM conversion.
The PLS model contains two principal components. As is
expected, the first principal component (Figure 3a) shows that
all four impurities have negative effects on the RCM conversion,
indicating that the overall purity of the starting material is of
utmost importance to the reaction conversion. After the first
principal component, the dominant influence of urea 8 and
2-propyl ester 9 is again shown by their large negative
coefficients in the second principal component (Figure 3b). It
suggests that these two impurities negatively impact the RCM
performance of 1 even after the overall purity of the starting
material has been taken into consideration, thus again implicat-
ing them as the two most detrimental impurities.
The PCA and PLS models not only show that impurities 8
and 9 account for the largest source of variation in the starting
material quality, they also confirm a strong negative correlation
between their content level and RCM performance. This
historical data analysis alone does not prove a causal relation-
ship, and neither method of analysis overcomes the issue that
the data are heavily correlated. However, the fact that they both
arrived at the same conclusion, which was also consistent with
the spiking study findings, did provide confidence that focusing
our efforts on controlling these two impurities would provide
substantial benefit.
DBU negatively impacted the RCM reaction, it was never
detected in any of the batches and is therefore omitted from
the graph.
In order to identify the impurities that caused the most batch-
to-batch variation, the highly correlated data set was analyzed
by two projection methods: principal component analysis (PCA)
and projection to latent structure (PLS).8 PCA transforms
complex data sets onto simpler coordinates called principal
components. By this method, two principal components were
identified which accounted for 99% of the variation in the
impurity content in different batches of diene 1. The score plot
in Figure 2a compares the different batches (labeled with batch
numbers) in the new coordinate system, with “score” being the
coordinates of these batches on the two principal components
(t1 and t2). Since points close to the origin represent low total
impurity levels and all four impurities retard RCM, it is not
surprising that only the batches close to the origin gave high
conversion in RCM (>80% conversion in RCM with 0.2 mol
% catalyst, coded green) while the batches farther from the
origin performed poorly (<80% conversion, coded red).
On the other hand, the loading plot in Figure 2b showed the
makeup of the two principal components to illustrate how the
originally measured impurity levels “loaded into” the two
principal components. The higher loading (p) an impurity has,
Removal of Urea 8, 2-Propyl Ester 9 and Improved Diene
Preparation. RemoVal of Urea 8. Urea 8 is a known impurity
found in almost all batches of sulfonamide 5, which is prepared
by the coupling of 2-pyridinesulfonyl chloride (10) with
(8) For detailed discussion of projection methods and their applications,
see: Multi- and MegaVariate Data Analysis; Umetrics: Umea, Sweden,
2006.
228
•
Vol. 12, No. 2, 2008 / Organic Process Research & Development

Figure 2. PCA analysis of impurity profile of all batches (R2 ) 0.99, Q2
) 0.88): (a) the score plot; and (b) the loading plot.
X
cum
Figure 3. PLS model correlation of impurity content with RCM conversion at 0.2 mol % catalyst (R2X ) 0.927, R2Y ) 0.905, Q2
) 0.854): coefficient plots of (a) first principal component; and (b) second principal component.
cum
present in the reaction mixture during the coupling.9 This
hypothesis was supported by the observation that only a trace
amount of urea 8 was detected in the reaction mixture when
the CH2Cl2 solution of sulfonyl chloride 10 was concentrated
under vacuum prior to the coupling with 11 (Table 1, entry 1),
whereas 4 mol % of 8 was observed when the coupling was
carried out using the CH2Cl2 solution of 10 directly after the
K2CO3 wash (entry 2). Carbon dioxide was further implicated
as the carbon source of urea 8 by the observed formation of a
large amount of 8 when dry ice was added to a mixture of amine
11 and Et3N prior to the addition of sulfonyl chloride 10 in the
coupling reaction (entry 3). One plausible mechanism illustrating
the role of CO2 in the formation of 8 is depicted in Scheme 6.
With the likely origin of 8 identified, the problem was easily
solved by replacing K2CO3 with NaOH in the chlorination
Scheme 5. Preparation of sulfonamide 5
methallylamine 11 (Scheme 5). Historically, the amount of 8
in sulfonamide 5 was observed to be <1% by HPLC, although
calibration using a pure standard of urea 8 indicated that some
batches of 5 contained as much as 5–7 wt% of this impurity.
It was postulated that a potential source of the one-carbon
unit in urea 8 was carbon dioxide, which formed during the
aqueous K2CO3 wash after the chlorination reaction, remained
solubilized in the CH2Cl2 solution of 10 and therefore was
(9) It was also postulated that co-solvent CH2Cl2 could form phosgene
(or a phosgene equivalent) during the oxidation of mercaptopyridine,
which would then be trapped by amine 11 during the subsequent
coupling step to form urea 8. However, this proposal was invalidated
as urea 8 was observed in sulfonamide 5 even when the chlorination
was conducted in solvents such as toluene or chlorotoluene.
Vol. 12, No. 2, 2008 / Organic Process Research & Development
•
229

Table 1. Quantities of urea 8 formed under different
diene was optimized for throughput, yield and process ef-
ficiency. The reaction was carried out with urea-free sulfona-
mide 5. Upon completion of the reaction, a seeded cooling
crystallization using heptane as the antisolvent then provided
an easily filterable slurry, producing large, uniform crystals with
low cake resistance upon filtration. Figure 4 illustrates that the
product isolated by this method (Figure 4b) appears to be more
consistent in size and aspect ratio in relation to the previous
isolation procedure, which used an aqueous citric acid quench
and did not control particle formation (Figure 4a). The diene
thus prepared was of consistently high purity, typically contain-
ing only negligible amounts of BTPP and secondary amine 6,
and no urea 8 or 2-propyl ester 9.
reaction conditions
form of 10 used in
entry
coupling
additive
urea 8/mol%
1
2
3
oila
solutionb
oila
-
0.05
4
27
-
CO2(s)
a After aq K2CO3 workup, sulfonyl chloride 10 was concentrated to an oil prior
to coupling with amine 11. b After aq K2CO3 workup, sulfonyl choloride 10 was
used directly as a CH2Cl2 solution in the coupling.
Scheme 6. Possible mechanism for the formation of 8 with
CO2
RCM Performance of Representative Batch and Speci-
fication Setting. RCM Performance of RepresentatiVe Batch.
With the procedure for diene 1 finalized and representative
batches available, the RCM procedure was optimized using
Response Surface Methodology (RSM).11 A central composite
design with temperature (60–90 °C), reaction volume (5–10
volumes) and catalyst loading (0.1–0.3 mol %) revealed that
reaction volume had no impact on the reaction.12 The conversion
was determined by reaction temperature and catalyst loading,
with complete RCM achieved at high temperature and high
catalyst loading (Figure 5a). The experimental space where the
reaction would reach complete conversion is indicated by the
yellow region in the overlay plot (Figure 5b).
The model also provided a basis for assessing the robustness
of the reaction. Although the long-term consistency of the RCM
reaction cannot be demonstrated by a limited number of
experiments, the probability of its achieving a desired conversion
can be assessed using a Bayesian reliability approach based on
the RSM model data.13 Therefore, a probability function was
established for reaching specific conversions at a given catalyst
loading and reaction temperature. The probability of achieving
complete RCM conversion (>95%) and, thus, the robustness
of the process under different reaction conditions were then
calculated for each catalyst loading/temperature combination
across the experimental region (Figure 6). At low temperature
and/or low catalyst loading, the probability of complete RCM
conversion is very low, as indicated by the green color coding.
The probability rises with higher catalyst loading and higher
temperature, with the calculated 50% probability contour almost
identical to the edge of the yellow region in the overlay plot in
Figure 5b, which corresponds to the predicted mean of RCM
conversion. With high temperature and sufficient catalyst
loading, over 90% probability can be reached within this
experimental space. Therefore, a catalyst loading of 0.25 mol
% and a temperature of 80 °C were selected as the final reaction
conditions for the RCM, with a calculated probability of >95%
for achieving complete RCM conversion.
workup. The sulfonyl chloride solution thus obtained produced
only a trace amount of urea 8 in the coupling reaction mixture,
and no urea was detected in the isolated solid of sulfonam-
ide 5.
The acidity of sulfonamide 5 also allowed it to be readily
deprotonated by alkaline solutions and represented a convenient
handle for its purification. Thus, sulfonamide 5 (contaminated
with urea 8) was treated with aqueous NaOH to furnish an
aqueous solution of its sodium salt, leaving solid urea 8 to be
filtered off. Sulfonamide 5 was recovered after simple reacidi-
fication with HCl. This procedure allowed us to rework large
quantities of contaminated material, affording pure sulfonamide
5 which contained undetectable quantities of the urea impurity.
RemoVal of 2-Propyl Ester 9. 2-Propanol adduct 9 was
identified relatively late during development work. In fact, it
was hardly detectable in early laboratory batches of diene 1.
Unfortunately, it became a significant problem during a 1.5 kg
scale-up experiment, at which point it was produced in ∼2.2
mol %.10
Interestingly, 2-propyl ester 9 was not observed in solution
during the coupling reaction between sulfonamide 5 and epoxide
4, but only formed after the quench with aqueous citric acid.
Since a neutral water quench still led to the formation of
appreciable quantities of 9, a nonaqueous workup was inves-
tigated. The acid quench was initially included to protonate the
BTPP catalyst and prevent it from oiling out when the product
precipitated. However, the discovery that BTPP was miscible
with heptane led to the development of an anhydrous isolation
using heptane as the antisolvent. Using the modified procedure,
pure product containing very low levels of BTPP and no
significant quantities of 2-propyl ester 9 was routinely obtained.
ImproVed Diene Preparation. Having identified the two main
impurities and methods for their removal, the synthesis of the
Specification Setting. Another aspect of robustness is toler-
ance of the reaction of impurities in the starting material.
(11) For detailed discussion on RSM and their applications, see: Anderson,
M. J.; Whitcomb, P. J. RSM Simplified: Optimizing Processes Using
Response Surface Methods for Design of Experiments; Productivity:
New York, 2005.
(12) No cross-metathesis product was observed even when the reaction
was run at the highest concentration.
(10) This batch of diene gave 52% conversion with 0.2 mol % catalyst
(13) (a) Peterson, J. J. J. Qual. Technol. 2004, 36, 139. (b) Miro-Quesada,
G.; Del Castillo, E.; Peterson, J. J. J. Appl. Stat. 2004, 31, 291.
and 92%-95% conversion with 0.5 mol % catalyst.
230
•
Vol. 12, No. 2, 2008 / Organic Process Research & Development

Figure 4. Comparison of product crystals: material isolated from the citric acid procedure (a) versus material produced using a
controlled crystallization with heptane as antisolvent (b).
Figure 5. (a) Response surface model and (b) overlay plot of RCM conversion.
Figure 6. Probability of reaching complete conversion in RCM using a representative batch of 1.
Although historical data were available (vide supra), they could
not be used to set univariate specifications due to the heavy
correlation in the data. An orthogonal design had to be used
for this purpose, and the robustness of the specifications could
then be assessed by model-based probability of a successful
RCM reaction using this material.
Therefore, a two-level factorial DoE was performed where
different amounts of the four main impurities (urea 8: 0–0.05
Vol. 12, No. 2, 2008 / Organic Process Research & Development
•
231

Figure 7. Probability of achieving complete RCM using diene 1 with different purity profile.
mol %; BTPP: 0–0.01 mol %; amine 6: 0–0.05 mol %; 2-propyl
ester 9: 0–0.2 mol %) were spiked into a representative batch
of 1.14 The RCM reaction was then carried out under the
standard conditions and the reaction conversion was modelled
against the impurity levels. Consistent with findings from
spiking experiments and PLS interpretations, all four impurities
showed a statistically significant, negative impact on the RCM
conversion.
reaction consistently gave >99% conversion in 20–50 g jacketed
laboratory reactor experiments. Scale-up of the reaction cul-
minated in a plant run, where the reaction was carried out on
a 6.7 kg scale. With 0.25 mol % catalyst loading, the reaction
proceeded smoothly in five volumes of toluene, reaching
completion within two hours at 80 °C. A simple cool-down
and filtration then afforded the cyclized product in 90% yield,
with a residual ruthenium content of 121 ppm.16
Again with an empirical model available, the probability of
complete RCM conversion was calculated for different levels
of impurity contents in the starting material. The reliability plot
is shown in Figure 7: The probability of achieving complete
conversion is very high (∼100%) with the unspiked batch (the
panel at the lower left corner), but decreases with higher
amounts of spiked impurities. When all four impurities were
spiked (the upper right corner in the figure), the probability of
reaching complete conversion is essentially zero. Based on the
calculated probability results, specifications of the four impuri-
ties could then be set: Urea 8 <0.01 mol %, BTPP <0.007
mol %, amine 6 <0.01 mol % and 2-propyl ester 9 <0.15 mol
%.15 Under the standard RCM conditions, a batch of 1
containing all four impurities at these levels would have >90%
probability of reaching completion in the RCM reaction.
Scale-Up Results of the RCM Reaction. With robustness of
the RCM reaction significantly increased from improved starting
material purity and optimized reaction conditions, scale-up of
the reaction was conducted. Using diene prepared by the
modified procedure under optimized RCM conditions, the
Conclusion
The development of a robust and high-yielding synthesis of
2 using RCM is described. The work is highlighted by: (1)
spiking studies to identify individual impurities which negatively
affect the RCM reaction; (2) historical data analysis using
projection methods to identify the main sources of variation in
starting material quality and determine the main detrimental
impurities in starting material, which led to (3) subsequent
modification of the starting material synthesis to ensure diene
1 of high purity, which in turn ensured a robust RCM process;
and (4) a probability-based approach in response surface
optimization and reaction robustness assessment.
Experimental Section
General. Unless otherwise indicated, all reactions were
conducted under a nitrogen atmosphere. Solvents and reagents
(16) The residual ruthenium was effectively purged through the subsequent
four chemical transformations and the final particle formation, with
all intermediates being solids and isolated through crystallization. RCM
product 2 with as much as 427 ppm of Ru has been used in multikilo
campaigns and no ruthenium (<1 ppm) was detected in any batches
of the final API. Most of the purging is believed to occur during the
hydrogenation of 2, where Ru can be adsorbed by the charcoal support
of Pd/C catalyst. On laboratory scale, hydrogenated product with less
than 1 ppm of Ru was obtained from a batch of 2 that contained 359
ppm of Ru.
(14) Impurity content in this batch: urea 8: not detected; BTPP: 0.006 mol
%; amine 6: 0.03 mol %; 2-propyl ester 9: not detected.
(15) Including the impurities in the starting material, the specifications
should be: urea 8: <0.01 mol %; BTPP: <0.013 mol %; amine 6:
<0.04 mol %; 2-propyl ester 9: <0.15 mol %.
232
•
Vol. 12, No. 2, 2008 / Organic Process Research & Development

were obtained from commercial sources and used without
further purification. Hoveyda’s catalyst was purchased from
Materia Inc. in Pasadena, California. All DoE data were
processed using Design-Expert Version 7.0.3, PCA and PLS
models were analyzed using SIMCA-P+ 11, reliability calcula-
tions were carried out in SAS 9.1.
combined organic layer was dried and concentrated. The product
was purified by flash column chromatography: 2.67 g, 5.32
mmol, 79% yield. ES-MS: m/z: 502 (M + H+), 524 (M+Na+).
1H NMR (CDCl3, 300 MHz) δ 8.66 (dm, 1H, J ) 4.8 Hz),
8.08 (dt, 1H, J ) 7.9, 1.0 Hz), 7.95 (td, 1H, J ) 7.7, 1.7 Hz),
7.89 (dm, 1H, J ) 7.8 Hz), 7.52 (m, 4H), 6.43 (d, 1H, J ) 9.1
Hz), 6.05 (m, 1H), 5.81 (m, 1H), 5.47 (dm, 1H, J ) 17, 1.4
Hz), 5.32 (dt, 1H, J ) 10.4, 1.4 Hz), 5.23 (dd, 1H, J ) 12.5,
6.3 Hz), 5.05 (d, 1H, J ) 1.5 Hz), 5.02 (dm, 1H, J ) 6.6 Hz),
4.72 (m, 1H), 4.28 (m, 2H), 3.79 (dd, 1H, J ) 15.2, 10.4 Hz),
3.56 (dd, 1H, J ) 15.2, 2.7 Hz), 1.34 (d, 9H, J ) 6.2 Hz). 13C
NMR (CDCl3, 100 MHz) δ 169.02, 165.86, 157.97, 149.41,
138.64, 138.09, 137.12, 136.45, 131.53, 130.16, 129.87, 129.54,
127.37, 126.94, 123.25, 117.16, 116.05, 71.29, 68.92, 55.73,
52.99, 48.10, 21.76, 21.59, 17.59.
1,3,4,5-Tetradeoxy-3-(1,3-dioxo-1,3-dihydro-2H-isoindol-2-
yl)-1-{[(1R)-1-methyl-2-propen-1-yl]amino}-2-O-2-pyridinyl-L-
threo-pent-4-enitol (6). Diene 1 (1.0 g, 2.27 mmol) and Cs2CO3
(0.775 g, 2.38 mmol) were heated in acetonitrile (5 mL) to 80
°C and stirred overnight. Upon cooling, the slurry was filtered
through celite to remove solid residues, and the solvent removed
under vacuum. The product was purified by flash column
chromatography: 0.60 g, 70% yield. ES-MS: m/z: 378 (M +
1
H+). H NMR (CDCl3, 300 MHz) δ 7.88 (ddd, 1H, J ) 5.0,
2.1, 0.8 Hz), 7.70 (m, 2H), 7.63 (m, 2H), 7.33 (ddd, 1H, J )
9.1, 7.1, 2.0 Hz), 6.61 (ddd, 1H, J ) 7.2, 5.0, 0.9 Hz), 6.54
(dm, 2H, J ) 9.2 Hz), 6.45 (m, 1H), 6.05 (m, 1H), 5.67 (m,
1H), 5.48 (dm, 1H, J ) 17 Hz), 5.32 (dd, 1H, J ) 9.8, 1.4
Hz), 5.16 (t, 1H, J ) 9.4 Hz), 5.01 (m, 2H), 3.13 (m, 2H), 2.81
(dd, 1H, J ) 13, 6.2 Hz), 1.13 (d, 3H, J ) 6.5 Hz). 13C NMR
(CDCl3, 100 MHz) δ 167.83, 163.11, 146.53, 142.64, 138.44,
133.53, 131.83, 131.69, 122.83, 120.73, 116.71, 114.10, 111.01,
72.99, 56.47, 56.41, 47.87, 21.32.
N,N′-Bis[(1R)-1-methyl-2-propen-1-yl]urea (8). [(1R)-1-
Methyl-2-propen-1-yl]amine hydrochloride (4.0 g, 37.4 mmol)
was stirred in acetonitrile (24 mL) at room temperature. To this
slurry was added triethylamine (5.73 mL, 41.1 mmol), then CDI
(3.64 g, 22.4 mmol). After stirring the slurry overnight, the
solvent was removed under vacuum, and the residue partitioned
between EtOAc and water. Upon separation of the layers, the
organic layer was washed sequentially with sat. aq. NH4Cl, sat.
aq. NaHCO3, brine, then dried over Na2SO4. The desiccant was
filtered and solvent removed under vacuum to provide a slightly
yellow solid (4 g). Pure 8 was obtained by treating the crude
solid with hot TBME (25–30 mL), diluting with hexanes,
cooling slowly to 5 °C and filtering on a Buchner funnel: 2.376
g, 76% yield. ES-MS: m/z: 169 (M + H+). 1H NMR (CDCl3,
300 MHz) δ 5.84 (m, 2H), 5.04–5.20 (m, 4H), 4.29 (m, 2H),
1.22 (dd, 6H, J ) 6.8, 4.3 Hz). 13C NMR (CDCl3, 100 MHz)
δ 157.63, 140.86, 113.21, 47.39, 20.93.
1,3,4,5-Tetradeoxy-3-{[(2-{[(1-methylethyl)oxy]carbonyl}-
phenyl)carbonyl]amino}-1-[[(1R)-1-methyl-2-propen-1-yl](2-py-
ridinylsulfonyl)amino]-L-threo-pent-4-enitol (9). Diene 1 (2.99
g, 6.77 mmol) was suspended in EtOH (20 mL) and N2H4
hydrate (1.0 mL, 20.6 mmol, 3.0 equiv) was added at room
temperature. The mixture turned clear within a few minutes. It
was stirred at room temperature overnight and turned into a
white suspension. NaOH (1 N, 20 mL) was added and the
mixture was extracted with CH2Cl2 (20 mL × 2). The organic
layer was dried and concentrated. The crude aminoalcohol was
dissolved in CH2Cl2 (20 mL). Isopropyl hydrogen phthalate17
(1.48 g, 7.11 mmol, 1.05 equiv), EDCI (1.50 g, 7.82 mmol,
1.16 equiv) and HOOBt (57.7 mg, 0.354 mmol, 0.05 equiv)
were added at 0 °C. The mixture was stirred at 0 °C for 3 h
and diluted with HCl (1 N, 20 mL). The layers were separated,
and the aqueous layer was extracted with CH2Cl2 (20 mL). The
General Procedure for Spiking Experiments. In a vial,
diene 1 (100–200 mg) and Hoveyda’s catalyst (0.5 mol %) were
dissolved in toluene (1–2 mL) under nitrogen, and the impurity
(0.5–0.8 mol %) was added to the reaction mixture. The
resultant solution was heated to reflux for 3–5 h and the reaction
conversion was determined by HPLC.
1,3,4,5-Tetradeoxy-3-(1,3-dioxo-1,3-dihydro-2H-isoindol-2-
yl)-1-[[(1R)-1-methyl-2-propen-1-yl](2-pyridinylsulfonyl)amino]-
L-threo-pent-4-enitol (1). In a glass-lined reactor, a slurry of
tosylate 3 (8.5 kg, 21.2 mol) and toluene (34 L) was heated to
56 °C under nitrogen, and DBU (3.3 kg, 21.7 mol) was added
via pressure vessel to maintain the temperature between 55 and
65 °C. The starting material dissolved during the addition and
the mixture was aged at this temperature for 60 min. Upon
completion of the reaction, the mixture was cooled to 25 °C
and 10% w/w aq. citric acid (0.42 L water with 45 g citric acid
monohydrate) was added. After stirring for 10 min, water (25.5
L) was added, the mixture stirred and the layers allowed to
separate. The organic layer was then washed twice with water
(2 × 8.5 L). The solution of crude 4 was then transferred to a
second reactor via an inline filter to remove a small layer of
film, and the transfer line then rinsed with toluene (8.5 L). To
this second reactor was charged 5 (4.5 kg, 21.2 mol) and the
solvent partially removed by vacuum distillation to reach 15 L
total volume (36.0 L toluene removed). 2-Propanol (25.5 L)
was charged and the system vacuum distilled to achieve 15 L
total volume (25.5 L removed). The total solvent volume was
adjusted to 3.4 volumes using 2-propanol (15.0 L added), BTPP
(0.7 kg, 2.12 mol) was added and the mixture heated to 80 °C
over ∼30 min. After aging for 3.5 h, the reaction was complete,
and heptane was added (17.0 L) while maintaining the tem-
perature g60 °C. The temperature was then adjusted to 53 °C
and seeds of 1 (4.5 g in 2:1 heptane:2-propanol) were charged
to the reactor. The slurry was held at 52 °C for 60 min, then
cooled to ∼10 °C at 0.5 °C/min. After holding for 30 min, the
slurry was isolated by centrifugal filtration, washed with cold
heptane/2-propanol (2:1 v/v, 2 × 8.5 L). The product was dried
under vacuum at 45 °C to afford 1: 7.25 kg (16.4 mol, 78%),
>99.9% pure by HPLC. 1H NMR (CDCl3, 400 MHz) δ 8.62
(dm, 1H, J ) 4.0 Hz), 8.02 (d, 1H, J ) 7.9 Hz), 7.93 (td, 1H,
J ) 7.7, 1.6 Hz), 7.83 (m, 2H), 7.70 (m, 2H), 7.50 (dm, 1H, J
) 0.9 Hz), 6.29 (m, 1H), 5.81 (m, 1H), 5.39 (d, 1H, J ) 17.1
Hz), 5.30 (dd, 1H, J ) 10.2, 0.5 Hz), 5.13 (m, 2H), 4.89 (bs,
(17) Loev, B.; Kormendy, M. J. Org. Chem. 1962, 27, 1703.
Vol. 12, No. 2, 2008 / Organic Process Research & Development
•
233

1H), 4.68 (t, 1H, J ) 9.0 Hz), 4.60 (dt, 1H, J ) 9.2, 2.6 Hz),
4.39 (m, 1H), 3.56 (dd, 1H, J ) 15.1, 2.6 Hz), 3.45 (dd, 1H, J
) 15.3, 9.2 Hz), 1.30 (d, 3H, J ) 6.9 Hz). 13C NMR (CDCl3,
100 MHz) δ 168.6, 158.6, 150.1, 138.9, 137.9, 134.3, 132.7,
132.4, 127.2, 123.7, 123.4, 121.3, 117.7, 69.2, 58.1, 56.2, 49.4,
18.1.
(td, 1H, J ) 7.8, 1.6 Hz), 7.75 (m, 2H), 7.66 (m, 2H), 7.60 (m,
1H), 5.52 (m, 1H), 5.20 (m, 2H), 5.02 (br, 1H), 4.63 (m, 1H),
4.42 (dd, 1H, J ) 9.6, 4.4 Hz), 4.21 (d, 1H, J ) 16 Hz), 3.87
(dd, 1H, J ) 16, 4.6 Hz), 1.31 (d, 3H, J ) 6.9 Hz). 13C NMR
(CDCl3, 100 MHz) δ 19.65, 48.29, 52.77, 54.75, 71.54, 122.29,
123.01, 126.68, 126.76, 130.95, 131.60, 133.85, 138.62, 150.62,
157.20, 168.01.
2-[(3S,4S,7R)-3-Hydroxy-7-methyl-1-(2-pyridinylsulfonyl)-
2,3,4,7-tetrahydro-1H-azepin-4-yl]-1H-isoindole-1,3(2H)-di-
one (2). In a glass-lined reactor, diene 1 (6.74 kg, 15.3 mol)
and Hoveyda’s catalyst (24 g, 0.038 mol, 0.25 mol %) were
dissolved in toluene (34 L), and the mixture was sparged with
nitrogen (3.4 L/min) for 1 h. With continued nitrogen sparging,
the mixture was heated first to 60 °C over 20 min, then to 80
°C over 45 min. The reaction mixture was stirred at 80 °C for
2 h, and HPLC showed the reaction was complete. The mixture
was heated to 92 °C and the nitrogen sparging was discontinued.
The reaction mixture was cooled to 5 °C over 3 h and held at
5 °C overnight. The product was collected by centrifugal
filtration, rinsed with cold toluene (13.5 L), and dried under
vacuum at 50 °C: 5.70 kg (13.8 mol, 90%), 99.45% purity by
HPLC PAR. Ru content: 121 ppm. 1H NMR (CDCl3, 400 MHz)
δ 8.81 (dm, 1H, J ) 4.0 Hz), 8.05 (d, 1H, J ) 7.8 Hz), 7.97
Acknowledgment
We thank Dr. Hayao Matsuhashi and Ms. Jessica Bailey for
their initial efforts in developing this chemistry, Mr. Christopher
Werner for his contributions to the robust crystallization of 1,
Dr. Grant Spoors for helpful discussion in the mechanism of
urea formation, Mr. Tom Wrzosek for helpful discussion in
PCA/PLS modeling, and Dr. John Peterson for discussion in
the application of Bayesian predictive approach to process
optimization.
Received for review December 14, 2007.
OP700288P
234
•
Vol. 12, No. 2, 2008 / Organic Process Research & Development