Kepner-Tregoe decision analysis as a tool to aid route selection. Part 2. Application to AZD7545, a PDK inhibitor

Article abstract of DOI:10.1021/op800033c

Kepner-Tregoe decision analysis was formally used as an aid to route selection, as outlined in the preceding paper. Over 40 paper routes were assessed for suitability for both immediate and longer term manufacture of AZD7545, a compound in the early stages of development. Eight routes were then investigated in full in the laboratory, and a further four in part, over a period of 3-4 months. From this exercise, the preferred long-term manufacturing route was identified before the first pilot scale manufacture had been completed. This route selection exercise worked well in this case where a large number of potential routes had to be considered using limited resources. It was also an effective means of bringing some long-term manufacturing issues to the fore at an early stage in development.

Full text of DOI:10.1021/op800033c

Organic Process Research & Development 2008, 12, 1044–1059
Full Papers
Kepner-Tregoe Decision Analysis as a Tool To Aid Route Selection. Part 2.
Application to AZD7545, a PDK Inhibitor
Jonathan D. Moseley,* David Brown, Catherine R. Firkin, Shelley L. Jenkin, Bharti Patel, and Evan W. Snape
AstraZeneca, Process Research and DeVelopment, AVlon Works, SeVern Road, Hallen, Bristol BS10 7ZE, UK
Abstract:
contractor to produce the first kilogram for initial clinical trials.
Both routes had significant disadvantages for long-term manu-
facture whilst being adequate in the short term.
Kepner-Tregoe decision analysis was formally used as an aid to
route selection, as outlined in the preceding paper. Over 40 paper
routes were assessed for suitability for both immediate and longer
term manufacture of AZD7545, a compound in the early stages
of development. Eight routes were then investigated in full in the
laboratory, and a further four in part, over a period of 3-4
months. From this exercise, the preferred long-term manufactur-
ing route was identified before the first pilot scale manufacture
had been completed. This route selection exercise worked well in
this case where a large number of potential routes had to be
considered using limited resources. It was also an effective means
of bringing some long-term manufacturing issues to the fore at
an early stage in development.
A major drawback in both cases was the introduction early
in the synthesis of the expensive contributory raw material,
chiral acid (2). This had been required in the synthesis of ATP-
sensitive potassium channel openers (for the treatment of urinary
incontinence) on a previous Zeneca project,6 for which several
enzymatic resolutions had been developed.7 Chiral acid (2) was
commercially available for this project on large scale by an
enzymatic resolution subsequently developed by Lonza.8 This
building block was expected to contribute 99% of the total raw
materials costs for the first pilot plant scale manufacture (10
kg @ .£10k/kg, 2001 data), reducing to an estimated 50% at
full scale manufacture (∼£1000/kg at tonne scale). (Cost
contributions remained high at full scale in part because all the
other raw materials were especially cheap; the KTDA process
is therefore essentially unchanged). This was particularly critical
as the expected API dosage was high (300 mg/day), resulting
in a borderline cost of goods issue. Late introduction of chiral
acid (2) would therefore clearly be advantageous. Late introduc-
tion was not necessary to preserve the chiral integrity of the
enantiomeric centre since as a result of its quaternary nature it
resisted racemisation. However, this did preclude the option to
racemise and recycle the unwanted enantiomer in either the
Lonza resolution process or later in the synthesis of AZD7545.
Consideration of the structure of AZD7545 reveals that it
can be assembled in numerous ways (Scheme 1). To have
investigated all of these potential routes in the laboratory was
impractical as it would have required a high resource commit-
ment. It was therefore necessary to assess the routes prior to
laboratory work so that resource could be focused on the most
promising ones. KTDA seemed an ideal method to decide this.
Introduction
AZD7545 was a pyruvate dehydrogenase kinase (PDK)
inhibitor1 designed to provide an oral treatment for type II
diabetes2 and was the first of a number of compounds to enter
development from AstraZeneca’s PDK project.3 It was also the
first compound in AstraZeneca upon which the Kepner-Tregoe
decision analysis (KTDA) tool was formally used as an aid to
route selection, as outlined in the previous paper.4 Consequently,
this tool provided much useful guidance on the route selection
choice for AZD7545 whilst providing valuable learning with
which to modify the process. Several of these improvements
were incorporated for later compounds, as exemplified by the
following paper.5
Kepner-Tregoe Decision Analysis
Decision Statement. The KTDA process has been outlined
in the previous paper.4 The important first task is to define a
decision statement, which was as follows: “To identify routes
most likely to succeed for full scale manufacture, for investiga-
tion in the short term (3-4 months), paying particular attention
At the time AZD7545 (1) entered development in December,
2000, there were two synthetic routes available for consider-
ation: the one used by our medicinal chemistry colleagues to
produce ∼200 g and a modification in use by a tactical
* To
whom
correspondence
should
be
addressed.
E-mail:
(6) Ohnmacht, C. J.; Russell, K.; Empfield, J. R.; Frank, C. A.; Gibson,
K. H.; Mayhugh, D. R.; McLaren, F. M.; Shapiro, H. S.; Brown, F. J.;
Trainor, D. A.; Ceccarelli, C.; Lin, M. M.; Masek, B. B.; Forst, J. M.;
Harris, R. J.; Hulsizer, J. M.; Lewis, J. J.; Silverman, S. M.; Smith,
R. W.; Warwick, P. J.; Kau, S. T.; Yochim, C.; Dryoff, M. C.;
Kirkland, M.; Neilson, K. L. J. Med. Chem. 1996, 39, 4592–4601.
(7) Crosby, J.; Holt, R. A.; Pittam, J. D. PCT Int. Appl. WO 97/38124
A2.
(8) Shaw, N. M.; Naughton, A.; Robins, K.; Tinschert, A.; Schimid, E.;
Hischier, M.-L.; Venetz, V.; Werlan, J.; Zimmermann, T.; Brieden,
W.; de Riedmatten, P.; Roduit, J.-P.; Zimmermann, B.; Neumu¨ller,
R. Org. Process Res. DeV. 2002, 6, 497–504.
jonathan.moseley@astrazeneca.com.
(1) Butlin, R. J.; Nowak, T.; Burrows, J. N.; Block, M. H. PCT Int. Appl.
WO 1999/9962506 A1.
(2) Mayers, R. M.; Leighton, B.; Kilgour, E. Biochem. Soc. Trans. 2005,
33, 367–370.
(3) Morell, J. A.; Orme, J.; Butlin, R. J.; Roche, T. E.; Mayers, R. M.;
Kilgour, E. Biochem. Soc. Trans. 2003, 31, 1168–1170.
(4) Parker, J. S.; Moseley, J. D. Org. Process Res. DeV. 2008, 12, 1041–
1043.
(5) Parker, J. S.; Bower, F.; Murray, P. M.; Patel, B.; Talavera, P. Org.
Process Res. DeV. 2008, 12, 1060–1077.
1044
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development
10.1021/op800033c CCC: $40.75
2008 American Chemical Society
Published on Web 10/28/2008

Scheme 1
· Accommodation (6)
· Chemical feasibility (10)
· Cost of goods (8)
to feasibility, cost and environmental issues in the long term.”
This reflected our desire to find a suitable long-term manufac-
turing route but recognized that we only had a short time for
such investigations. We would therefore focus on those we
thought most likely to give us a long-term manufacturing
option.
· Environment (8)
· Number of stages (4)
The same objective can appear in both “must” and “want”
categories as different levels of stringency can be required. For
example, for environmental issues, a “must” requirement might
be to have no stoichiometric heavy metals in the process;
assuming catalytic heavy metals were acceptable below a certain
threshold value, a “want” requirement would then be to reduce
the loading as much as possible.
Develop Objectives and Identify Musts and Wants. The
objectives chosen from those listed previously4 and identified
as “must” criteria (with their pass criteria) were as follows:
· Safety, health and environment, taken as three separate
but similar factors, in that no compromises could be
made in these areas. Therefore no show-stoppers or
major issues were permitted.
· Chemical feasibility, by which we meant the likelihood
of success of any proposed chemical transformation.
Some literature precedent, even if only by analogy,
must be available.
· Raw materials, which were required to be available on
a 10 kg scale within 6 months. This was necessary for
the project to progress to its next milestone with a 10
kg manufacture in the pilot plant, since however good
a proposed new route might be, it would be no use if
the project failed to deliver its clinical targets on time.
Having identified the “want” criteria, a weighting was
applied to them as a score out of 10. The team agreed that
chemical feasibility was the most important, which received a
weighting of 10, while cost and environmental factors were
scored highly at 8. Accommodation was significant but not over-
riding and so received a 6, and number of steps was given a 4,
because chemical feasibility was felt to cover similar aspects
(but see the conclusions for further discussion on the interplay
between these two parameters). It is not worth scoring “wants”
much below 4, because their contribution to the overall
scores is low and serves only to average out the scores, thus
lessening the discrimination observed between good and poor
routes.
These were identified as “must” criteria, which were essential
to every route to be explored. Intellectual property (IP) was
not included as a “must” in this first case study, although of
course IP issues were taken into account when considering
potential routes. This exemplifies in part how the KTDA
exercise on AZD7545 served as a test-bed for later projects,
and IP was formally included in following PDK compounds,
for example.5
The quantification for the “wants” was as follows (also given
in more detail in Table 1):
· Ease of accommodation into existing plant assets: a
default score of 5 was given, with 1 point subtracted
for each stage requiring special accommodation from
The “want” criteria were identified as follows (with weight-
ings shown in parentheses):
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1045

Table 1. Scoring system for Wants
Want
scoring system
accommodation
default score 5
-1 for each stage requiring special accommodation from standard pilot plant
vessels (e.g., cryogenic, hydrogenation or pressure vessels)
chemical feasibility
default score normalised to 5 for best route
known or estimated yield for each stage multiplied by confidence as assessed
by rating below; total for all stages multiplied together and normalised to
score of 5 for best route, all others pro rata
score 10 for exact literature or AstraZeneca precedent
score 8 for close literature precedent
score 6 for a “better than even chance”
score 4 for significant doubts, but some precedent
score 2 for “one obscure paper suggests this is possible”
costs of goods
environment
default score normalised to 5 for best route
known or estimated yields for all stages after introduction of chiral acid mul-
tiplied together and normalised to score of 5 for best route, all others pro rata
default score 5
-1 for each stage requiring special abatement or disposal (e.g., catalytic
heavy metals, chlorination, iodine waste, oxidation)
number of stages
score 5 for 3 steps
score 4 for 4 steps
score 3 for 5 steps
score 2 for 6 steps
score 1 for 7 steps
score 0 for 8 or more steps
standard pilot plant vessels (e.g., cryogenic, hydrogena-
tion or pressure vessels).
The next step in the process was to evaluate all of the route
ideas that had been generated previously, screening them
through the “musts” and scoring the successful ones against
the “wants”. This data was easily gathered, calculated and
tracked in a spreadsheet, a portion of which is shown in Table
2. Similar routes were gathered into clusters to help highlight
where common issues might lie. For example, if one particular
reaction step common to a cluster of routes could not be shown
to work in practice, this would eliminate all of these routes,
irrespective of their other virtues. On the other hand, if a
speculative reaction was proven to be possible, this might open
up several route options in a cluster of related routes and
possibly indicate where the more successful strategies might
lie. Generally, it was more helpful if the key reactions that
needed investigation were near the start of each reaction
sequence; less chemistry had to be undertaken before a definitive
yes or no answer could then be determined.
The scores from the “want” criteria ranged from 70-144
(for those routes that were scored). The medicinal chemistry
route (Scheme 2, route 1.1) scored 95, which was towards the
bottom end of the distribution, and was deemed to fail the must
environmental and raw materials requirements. The second route
(2.1) that the tactical contractor had quickly developed scored
the highest initially with 144. By contrast, the final route that
was chosen for later manufactures (route 3.4) scored only 130
initially, although this was towards the top end of the range
and enough to be worth investigating. It is worth noting that
more developed routes generally tend to score slightly low
overall compared to “paper” routes, usually because their flaws
are known (although they tend to score better for chemical
feasibility); for undeveloped routes, many problems are often
unseen until encountered in practice (or to use a military axiom,
“no strategy of war survives the first day of action”). Some
routes that obviously failed to meet several “must” criteria were
simply not scored at all, on the assumption that they would be
too low scoring even if the “must” criteria could be resolved;
hence lower scores do not appear in the table. However, those
· Chemical feasibility: for each stage, the known or
expected literature yield was multiplied by the confidence
that that yield could be achieved. So a known reaction
would have a specific yield with a 100% confidence
assigned, whereas a similar but not identical reaction might
have a 60-80% confidence assigned to it. A more
speculative reaction might score only 20-40% (more
details are given in Table 1). This assessment was
performed for every stage in the sequence and multiplied
together to give an overall composite figure representative
of yield and the confidence in achieving it for a given
route. As with the cost of goods above, the highest scoring
figure was given 5 points, with the others normalised
against it accordingly. This procedure required consider-
able effort, although it is worth noting that the same
reaction steps occurred at different points in many
sequences, whilst the figures only had to be calculated
once for each such reaction stage.
· Cost of goods: this was scored solely on the cost of
the chiral acid (2). The combined estimated/known
yield of all stages after the introduction of 2 was
determined and normalised against the highest scoring
yield, which was given 5 points, with the others scored
pro rata accordingly. This apparently simple measure
of cost was deemed appropriate due to the overwhelm-
ing contribution to the cost of goods from the chiral
acid (and was further justified by the trivial cost of other
raw materials).
· Environment: a default score of 5 was given, with 1
point subtracted for each environmental concern that
could not be avoided.
· Number of stages: a default score of 5 was given for a
three stage synthesis (not including the final crystal-
lization), with 1 point deducted for every additional
stage (but no negative scores were assigned beyond 8
steps).
1046
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1047

Scheme 2
failing only one or two “must” criteria were scored if it was
thought that resolving their “musts” might then reveal a
promising route option.
the tactical contractor’s route 2.1 was particularly valuable for
reassessing the group 2 routes. Negative results were often more
helpful, because they could be used to eliminate routes
definitively, whereas positive results invariably were not quite
as positive as hoped, leaving a potential decision unresolved.
Either way limited resources could be efficiently focused on
those reactions which most needed to be proven or disproven.
So far, this initial assessment had been a paper exercise. The
results obtained meant that resource could be allocated to the
most promising routes for further investigation in the laboratory.
These will be discussed in their groups below, focusing mainly
on the key steps in each sequence. However, some comments
must be made on the initial route ideas before moving on to
how the proposed route options fared against the K-T criteria
chosen to assess them.
Ideas for route options had been generated or gathered from
a number of sources, most heavily within the project team and
the local department, but also across the department at other
sites, both nationally and internationally. Ideas from medicinal
chemistry colleagues, which they had not had time to explore,
were also taken into account, and a number of academics were
consulted. However, most of the new ideas involved alternative
syntheses of the expensive chiral acid unit, since this would
have the greatest impact on the project; the challenge of
AZD7545 itself was more one of efficiently combining the
many possible subunits.
In total, the route generation exercise resulted in over 40
potential routes, which were grouped according to overall
similarity or key chemical concept in the formation of AZD7545.
Current resources were judged to allow 3-4 routes to be
explored in detail within 3-4 months, after which time a
decision about longer term manufacture (i.e., >10 kg) would
need to be made. The default position would be further
development of the 1 kg manufacturing route option ongoing
at the tactical contractor, which was progressing well (route
2.1); any new route would have to be better than this. In fact,
about 8 routes were investigated, and a number of other route
options were briefly assessed by targeting reactions on key steps,
which meant that overall 12 routes were assessed by laboratory
work in some measure in the set time. It will be seen in the
discussion below that the choice of which routes to work on
was not driven completely by the KTDA scores; scientific
experience and “chemical intuition” was still allowed to play a
part so that the chemist’s subjective judgment still had a role.
However, KTDA provided the overall framework to guide the
decision-making process.
Results and Discussion
Initial Routes (Group 1). The medicinal chemistry route
(Scheme 2, route 1.1) and a variation upon which the tactical
contractor initially started (Scheme 3, route 1.2) scored only
95 and 98, respectively. Both of these scores were towards the
bottom end of the distribution, and both were deemed to fail
the “must” environmental requirements, mainly for the sto-
ichiometric use of toxic metals in the Ullmann-type9-11 sulfide
coupling step (5a + 6 to give 7). This was also a relatively
high temperature reaction with a long reaction time, producing
a moderate yield (51%) from a very dark reaction mixture after
a difficult and dilute workup. Problems with the workup were
not covered by the KTDA process, since they could not
generally be anticipated in advance, and it was expected that
ongoing process development would eventually provide a
satisfactory workup procedure in most cases. Therefore, work-
ups did not constitute show-stoppers in the KTDA process
unless they were regarded as unworkable at laboratory scale.
The iodide 4 was available in research quantities but would
probably not have been available on the scale required for a 10
As the laboratory work progressed and provided more
accurate information about the chemical feasibility and yields,
the values in the spread-sheet could be modified. In this way,
the table existed as a living document, allowing it to further
guide decision making and resource allocation in real project
time. Given the similarity of several routes, a single piece of
data could affect a number of routes. Furthermore, data from
(9) Hassan, J.; Se´vignon, M.; Gozzi, C.; Schulz, E.; Lemaire, M. Chem.
ReV. 2002, 102, 1359–1469.
(10) Kunz, K.; Scholz, U.; Ganzer, D. Synlett 2003, 2428–2439.
(11) Beletskaya, I. P.; Cheprakov, A. V. Coord. Chem. ReV. 2004, 248,
2337–2364.
1048
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

Scheme 3
Scheme 4
Kilo-Scale Route Variations (Group 2). The second major
group of routes were variations on the one developed by the
tactical contractor for the first 1 kg scale manufacture (Scheme
5, route 2.1). This route failed the environmental criteria11 with
the continuing use of stoichiometric Cu2O (and dilute workup
procedure). Furthermore, reduction of the thiocyanate 10 tended
to result in significant quantities of the disulfide 14b, up to 1:1
with sulfide 14a in the early stages, which reduced the yield
and complicated the workup. However, a major improvement
was the step reorder so that protected chiral acid 3a was
introduced at the penultimate step to give protected sulfide
anilide 13a, which required only the final high-yielding oxida-
tion of the sulfide to sulfone (with concomitant TMS-depro-
tection) to complete the synthesis. This considerably improved
the cost score and the yield somewhat. If the environmental
issue could be addressed, this would be a promising route.
Literature precedent suggested that the key coupling step of
thiocyanate 10 with bromobenzamide 12 to give 15 might be
achievable with either Pd or other catalytic metals9 or possibly
without a catalyst at all.12 This gave minor variations 2.2 to 2.4
(Scheme 5), which were deemed satisfactory on the environ-
mental issues, and with much improved scores for cost and
environment gave the other high scoring routes. Given that only
the reagents of the key coupling step were different, they were
effectively treated as one route option for investigation, since
any solution would give a viable result. In practice, a low level
of reaction was seen with several combinations of Pd sources,
bases and solvents, but none of them were sufficiently promising
to be worth investigating further, and these routes were parked
as other options looked more promising. This was a good
example of focusing on one key step in a cluster of several
potentially good routes.
kg manufacture and would have required an additional step from
2-chloroaniline (9) (and hence scored 0 for raw materials). The
4-mercaptobenzoic acid (6) was available but relatively expen-
sive in absolute terms, although cheap compared to chiral acid
2. Route 1.2 (Scheme 3) substituted the thiocyanate 10 (which
did have to be prepared from 9) for the iodide 4, to give
alternative coupling partner 5b. This avoided the minor problem
of iodine in the waste stream. It also avoided the thiophenol 6
by using the cheap acid 11 from which an earlier introduction
of the dimethyl-substituted amide group gave benzamide 12.
This combination (5b + 12 to give 13) was somewhat
preferential in the main coupling step (70% yield), although
the workup was still challenging. Finally, this route was also
more convergent in adding the dimethylamide functionality in
the subunit 12, rather than leaving this trivial step to the last.
Finally, and ignoring the use of stoichiometric heavy metals
that failed the “must” criteria, a reason for the low “want” scores
was the early introduction of the chiral acid 2 in both cases.
This was achieved through the bis-TMS-protected intermediate
(3b) and activation as the acid chloride to give volatile 3a, which
could be isolated after distillation or more usually stored
temporarily as a dichloromethane solution (Scheme 4). (The
unprotected acid chloride 3c was used successfully in later work
for the final route chosen.) This added considerable cost to both
routes, thus lowering their scores, and the less than complete
conversion here (80% isolated yield), followed by other
moderate yields in the key coupling step, further reduced the
overall efficiency (although it will be recognized that medicinal
chemistry routes are often designed for the facile synthesis of
multiple analogues, not to be efficient for individual target
molecules).
An even better potential alternative to the step order
of routes 2.1-2.4 was to add the protected chiral acid 3a
as the very last step, giving route 2.5 with a score of 134
(Scheme 6, last steps only shown). However, this required
the oxidation of the sulfide 15 to precede the final amide
coupling reaction with 2, which we knew would be
unlikely to proceed selectively in the presence of the
unprotected aniline group (i.e., without undergoing N-
(12) Sawyer, J. S.; Schmittling, E. A.; Palkowitz, J. A.; Smith, W. J. J.
Org. Chem. 1998, 63, 6338–6343.
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1049

Scheme 5
Scheme 6
would give sulfide aniline 15, which would be completed via
13a/b to give AZD7545 as in the other related routes (Scheme
5). The low predicted chemical feasibility, along with some less
desirable chemistries and the extra step, gave this route a score
too low to consider (93), and no work on it was progres-
sed.
Alternative Leaving Group Variations (Group 3). An-
other major group of higher scoring routes that passed the
“must” criteria were those based on alternative leaving groups
for the key coupling reaction with thiocyanate 10. Rather than
substitute the bromide in bromobenzamide 12, diazo, tosyl, nitro
and fluoro groups were proposed as the leaving groups in
benzamides 20-23 (Scheme 8). The last part of each sequence
was identical to those in Scheme 5 (from 15 onwards), but each
earlier fragment required different steps to set up the appropriate
leaving group, so they were clustered separately. The respective
diazo- (20) and tosyl benzamides (21) could in principle be
prepared easily in 2-3 steps. There was good literature
precedent for the coupling step with 20,13 although with 21 it
was much less encouraging, but both route scores were too low
to consider (102 and 119, respectively). Substitution of the nitro
group in 22 is also possible, although generally only with other
strong electron-withdrawing groups present,14 which gave this
route (3.3) the second highest score (143). However, no coupling
reaction between thiocyanate 10 and the known 4-nitrobenza-
mide (22)15 (prepared in one step from 4-nitrobenzoic acid),
with or without base, could be observed, and so this route failed
oxide formation at the aniline). Although the literature
was far from encouraging, this was a good example of
where a quick investigation of a single step was worth
the investment of time for a potentially big gain. Using
samples of 15 prepared from other routes, attempts to
oxidize with either peracetic acid or oxone showed that
there was no prospect of succeeding with this alternative
route. Since it failed on chemical feasibility, it was quickly
dropped. Only the analogous route to 2.1 was scored for
this step reorder; if it had been successful, attention would
then have focused on the analogous methodologies
represented by routes 2.2-2.4.
A final route in this cluster (Scheme 7, route 2.6) substituted
the thiocyanate 10 for the commercially available 4-nitrothiophe-
nol (17) to give the coupled product 18. Assuming the coupling
reaction could be realized, reduction of the nitro group would
give 19, the des-chloro analogue of 15. Speculative chlorination
of 19, for which poly-chlorination might have been an issue,
Scheme 7
1050
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

Scheme 8
Scheme 9
the feasibility test. Limited laboratory work was justified in this
case by the high score and the fact that the key step was
accessible as the second in the sequence. The route via the
known 4-fluoro-dimethylbenzamide (23)16 was not scored at
this stage but prepared later after the success of route 3.4 had
been established (vide infra). Using similar conditions from
those used for bromobenzamide 12 and thiocyanate 10 (Cu2O,
DMF or NMP, >80 °C), an isolated crude yield of 46% was
eventually obtained when tried, but by then route 3.4 was
looking very promising.
cal, so for this and the other reasons above route 3.5 from 27
was neither scored nor investigated.
An even better choice with which we were more confident
of success used commercially available 4-fluorobenzonitrile
(24). This route (Scheme 9, route 3.4) had a good but initially
unexceptional score of 130, reduced by some concerns over
chemical feasibility and a rather linear reaction sequence. But
it held the promise of proceeding by a nonmetal containing SNAr
reaction,17 which improved both the environmental factors11,12
and the feasibility scores. This was successful at the first attempt,
to give an LC conversion of 92% after only 1.5 h in NMP at
80 °C. Hydrolysis of the nitrile 25 was required to give acid
26, which was derivatised to amide 15, initially by EDCI
coupling but later with CDI. The sequence proceeded in linear
fashion but in simple steps so that the feasibility for these steps
was eventually high. In common with the other group 3 routes,
the cost criteria benefited from adding unprotected chiral acid
chloride 3c in the penultimate step to give 13b, which as noted
above had been established as the last possible point. With
improvements to the coupling step, this developed from a
promising route to the final route of choice, which will be
discussed in more detail in a subsequent paper.
A possibly more advantageous analogue of 24 appeared to
be 1-fluoro-4-nitrobenzene (27), which might be expected to
be an even better coupling partner with thiol 14a in the SNAr
reaction (route 3.5).12 However, this would have required
reduction of the nitro group to the amine and then diazotization
or amide formation and then Beckman rearrangement to
introduce the correct carboxylate/amide functionality, either of
which options required too many extra steps and uncertain
chemistry. Furthermore, the original aniline nitrogen in 28 would
have required “protection” with the chiral acid portion to
differentiate it from the other incipient aniline disguised as the
nitro group. This was too early in the sequence to be economi-
“Quetiapine”-Based Routes (Group 4). Another group of
routes essentially reversed the previous disconnection around
the central sulfide, by using a sulfur nucleophile bearing the
carboxylate (6) or amide functionality (29) to displace by SNAr
reaction a suitably substituted nitro-bearing compound 30
(Scheme 10). The nitro group would then be reduced to the
Scheme 10
aniline, with or without prior chlorination, to provide key sulfide
aniline 15. This route was based on analogous chemistry from
a successful ortho-substituted example derived from an unre-
lated AstraZeneca compound (Quetiapine).18 In fact only the
first route, in which 30 had the correct substitution pattern (i.e.,
X ) Y ) Cl), scored highly enough (134) due to the similarity
with Quetiapine chemistry to merit consideration. However, it
was soon realized that there was insufficient discrimination
between substitution at the para and ortho positions of the
dichloride (30), and the preferred 4-fluoro-substituted compound
(46, X ) F, see Scheme 15) was not available at the time on
the scale required. The other routes in this group required extra
steps and relied on the uncertain regioselective chlorination
noted for route 2.6 (Scheme 7), which was reflected in their
lower scores (97-112). Attempts at selective chlorinations on
other routes were similarly unsuccessful, and so this group was
not investigated further.
4-Sulfobenzoic Acid/Tosyl Chloride-Based Routes (Group
5). Another large group of routes started from either readily
available 4-sulfobenzoic acid potassium salt (31) or tosyl
chloride (32). These routes had a number of common themes,
generally relying on an aromatic Friedel-Crafts acylation of
the sulfonyl chloride with a simple chlorobenzene, aniline or
nitrobenzene (as a masked aniline) (9, 33a-d) as the key step.
(13) Petrillo, G.; Novi, M.; Garbarino, G.; Dell’Erba, C. Tetrahedron 1986,
42, 4007–4016.
(14) For example, see: (a) Robert, J.; Anouti, M.; Bosser, G.; Parrain, J.-
L.; Paris, J. J. Chem. Soc., Perkin Trans. 1995, 2, 1639–1644. (b)
Zlotin, S. G.; Kislitsin, P. G.; Samet, A. V.; Serebryakov, E. A.;
Konyushkin, L. D.; Semenov, V. V.; Buchanan, A. C.; Gakh, A. A.
J. Org. Chem. 2000, 65, 8430–8438.
(15) Wenker, H. J. Am. Chem. Soc. 1938, 60, 1081.
(16) Schiemenz, G. P.; Stein, G. Tetrahedron 1970, 26, 200–2026.
(17) Tickner, A. M.; Huang, G. K.; Gombatz, K.; Mills, R. J.; Novack,
V.; Webb, K. S. Synth. Commun. 1995, 25, 2497–2505.
(18) Murray, P. M.; Vaz, L.-M.; Ainge, D.; Harada, K.; Nishino, S.; Yoshii,
K. PCT Int. Appl. US 2007/0203336 A1.
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1051

An advantage of these routes was that the sulfur atom was at
the correct oxidation state, saving a step over the previously
discussed routes above. The 4-sulfobenzoic acid was the
preferred starting material, since whilst tosyl chloride avoided
any difunctionalisation issues between the sulfonic and benzoic
acids in 31, the relatively inert methyl group would require
oxidation at some stage to the benzoic acid. This meant
swapping an easy oxidation (sulfide to sulfone) for a more
difficult one, such that other features of these routes needed to
be desirable to make them good options.
range of conditions between 80 and 120 °C in a systematic
fashion by factorial experimental design (DoE strategy), but
the original FeCl3 was the only Lewis acid that gave clean
product 34b (in ∼60% conversion). Work on other softer Lewis
acids had not started when these routes were abandoned.
Introduction of the aniline functionality at the correct position
of 34b was another challenge that required solving. Use of
benzophenone imine with 10 mol % Pd and the dppf ligand
with sodium pentoxide as base in toluene gave complete
conversion to two products, which proved to be an unresolvable
mixture of the desired mono- to disubstituted benzophenone
imines (36a and 36b) in about 2:1 ratio (Scheme 12).21 The
problem appeared to be one of too great solubility, since once
one benzophenone imine group had been introduced, the
solubility of increased so much that the product 36a reacted
again in preference to the far less soluble dichloride 34b. Other
solvent (DMA and THF) and base (Cs2CO3) combinations were
tried but to no avail. Obviously the use of benzophenone imine
had very poor atom efficiency for the introduction of a single
nitrogen atom. Forcing conditions in sealed tubes using either
NaNH2 or NH3 as a more efficient source were also attempted
but failed to give the desired 37, as did using other sources of
protected nitrogen equivalents such as acetamide and phthal-
imide groups.
Oxidation of the dichlorobenzyl sulfone 34b to the benzoic
acid sulfone 38 (continuation of route 5.6) was attempted with
5 mol% RuCl3 and NaOCl under phase transfer conditions22
and gave about 5% reaction, essentially only one turnover of
the catalyst (Scheme 12). Speculative use of aqueous nitric acid
failed to give any reaction, possibly due to solubility issues.
However, stoichiometric use of KMnO4 in 1 M NaOH gave
complete conversion in 1 h at 100 °C to a new product that
was assumed to be benzoic acid 38.23 Alternatively, KMnO4
in aqueous tBuOH with 2.0 equiv of KHCO3 gave complete
and clean conversion after 24 h at 100 °C (aqueous tBuOH
was found to aid solubility). The MnO2 byproduct was easily
filtered off, but the reaction was dilute and the workup
problematic. The solubility of 38 was low, which hampered
efforts to achieve conversion to the dimethyl amide 40 under
standard conditions, and the result was difficult to confirm.
However, further confirmation that benzoic acid 38 had been
prepared was made by almost quantitative conversion to the
methyl ester 39, which had much greater, albeit still low,
solubility in organic solvents. Once again, the full suite of
solvents, bases and conditions was applied to 39 to introduce a
nitrogen equivalent through the previously successful benzophe-
none imine chemistry,21 which failed to work in this case despite
increased organic solubility.
In addition, the environmental burden of using stoichiometric
Lewis acids as acylating agents was undesirable, although
catalytic Lewis acids would be much more acceptable. These
routes mostly scored in the middle range (100-120), although
their short reaction sequences (generally four steps) and easily
accessible, low-cost starting materials made them attractive (the
highest scoring ones were generally based on 31 as the starting
material). Given this large route cluster with some medium-
high scores, it was felt some investigations were justified where
the key Friedel-Craft acylation fell at the start of a sequence,
such as shown in Scheme 11. For example, acylation of various
Scheme 11
chlorobenzenes (9, 33a-d) with tosyl chloride (32) was quickly
investigated using FeCl3.19 Whilst the protected acetamide 33c
gave the amidine dimer 35 and interaction of FeCl3 with the
unprotected aniline 9 resulted in a significant exothermic event
(route 5.4, score 111), reaction with dichlorobenzene (33b) gave
the desired sulfone 34b20 product in 45% crystallized yield after
only 2 h at 100 °C (route 5.6, score 104). With only 5 mol %
of Lewis acid, the reaction proceeded more slowly and failed
to reach completion after extended period, even with fresh
charges of FeCl3. Reinvestigation of the “protected” aniline 33c
with 32 in the presence of the more potent AlCl3 gave no
reaction in dichloromethane but reacted preferentially with
chlorobenzene (33a) when this was used as a solvent instead,
to give complete conversion to 34a but in a 2:1 mixture of
isomers (assumed para:meta). Repeating the reaction of 32 in
nitrobenzene (33d) gave no reaction with 33c after 18 h at 100
°C. Since reactions of tosyl chloride and dichlorobenzene were
too slow with FeCl3 and too fast with AlCl3, further Lewis acids
(BF3, CoF2, CuI, MnF2, PBr3, TiCl4) were investigated under a
Conversion of the commercially available 31 to dimethyl
amide 41 proceeded with standard reagents (SOCl2, catalytic
DMF, toluene, then dimethylamine ·HCl) in 80% isolated yield
to leave the sulfonyl chloride group intact (Scheme 12).
Coupling this amide with AlCl3 in 20 vol of dichlorobenzene
(33b) revealed several new peaks after 4 h, including one
(21) Wolfe, J. P.; Åhman, J.; Sadighi, J. P.; Singer, R. A.; Buchwald, S. L.
Tetrahedron Lett. 1997, 38, 6367–6370.
(22) Sasson, Y.; Zappi, G. D.; Neumann, R. J. Org. Chem. 1986, 51, 2880–
2883.
(19) Huismann, J. U.S. Patent 2,224,964, 1940.
(20) Zilberman, J.; Ioffe, D.; Gozlan, I. Synthesis 1992, 659–660.
(23) Fatiadi, A. J. Synthesis 1987, 85–127.
1052
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

Scheme 12
corresponding to the previously tentatively identified dimethy-
lamide sulfone 40 (route 5.9, score 129). This thus established
a connectivity between the two original starting materials, 31
and 32. However, no further work was conducted due to the
lack of time.
which was obviously undesirable, whilst leaving the aniline
unprotected was judged to be unfeasible.
Scheme 13
In summary, all the Friedel-Craft-based routes offered (on
paper at least) short syntheses from cheap commercially
available starting materials, and the highest scoring ones justified
the effort invested in them. The key acylation step was
accessible at the start of the syntheses, and the benzoic oxidation
proved easier to achieve than anticipated. Furthermore, a high
level of connectivity was shown between the various routes,
including from sulfobenzoic acid 31, which allowed a large area
of overlapping reaction space to be investigated efficiently.
However, all of these routes broke down due to the failure to
introduce any worthwhile nitrogen atom selectivity into the
aromatic ring, largely as a result of the low solubility of the
dichloride 34b, which was an otherwise accessible and stable
intermediate.
Grignard-Based Routes (Groups 6 and 7). The key step
in these proposed routes relied upon a Grignard coupling with
a sulfonyl chloride, for which there was some precedent.24 The
Grignard reagent could be either the magnesium Grignard of
bromobenzamide (42b) or tolyl magnesium bromide (42a),
which would require oxidation of the methyl group in product
44 (Scheme 13; cf. 34a/b in Scheme 11); however, this latter
transformation had been proven from related work above (see
group 5 route discussion). Grignard attack was required on the
sulfonyl chloride functionality in the coupling partner, which
effectively also required the protected aniline functionality to
be compatible with the Grignard reagent. Without resorting to
a protection-deprotection sequence, the obvious protecting
group was the chiral acid fragment, giving compound 43 (which
coupled with 42b would give 1 directly) (Scheme 13, group 6
routes). However, this required introducing the chiral acid at
the start of the sequence to protect the aniline functionality,
The group 7 routes simply reversed this strategy by making
the Grignard reagent of the protected aniline 45 and coupling
that with tosyl chloride (32) to also give 44 (Scheme 14, group
7 routes). Again, the use of the unprotected aniline was judged
to be unfeasible, whilst use of the chiral acid as the protecting
group so early in the synthesis was undesirable. All of these
routes scored well on environmental and plant accommodation
factors (AstraZeneca has suitable pilot plant facilities and
experience for cryogenic Grignard-type reactions) and were
moderately short sequences. However, those routes that passed
the “must” criteria for feasibility did not score highly in this
area under “wants”, and their cost scores were also moderate.
Scheme 14
Given the concerns over their feasibility and their moderately
low overall scores (103-116 for those scored), no laboratory
work was undertaken on them. However, it should be noted
that this work largely preceded publication of the Knochel-type
(24) Gilman, H.; Fothergill, R. E. J. Am. Chem. Soc. 1929, 51, 3501–3508.
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1053

Grignard chemistry, which might now make both group 6 and
7 routes viable alternatives.
gave the best route (9.1) a medium score of 118 initially.
However, given that this cluster of routes (9.1-9.4) was
distinctly different from other clusters and the raw materials
were cheap and commercially available, and if the regioselec-
tivity issues could be investigated early on in the process (which
we thought they could), then some investment of effort would
be justified. We decided to investigate them further on this basis.
Sulfinic Acid Routes (Group 8). These three routes had as
their key step the formation of the sulfinic acid anion (48a/b),
which was to be used as the nucleophile in an SNAr displace-
ment of 2-chloro-4-fluoronitrobenzene (46), available by halo-
gen exchange25 from cheap and commercially available 2,4-
dichloronitrobenzene (30) (Scheme 15). Oxidation and dimethyl-
amide formation from 49a, followed by reduction of the nitro
group and addition of the chiral acid gave route 8.1 or, with
these steps paired the opposite way round, route 8.3. Route 8.2
used the preformed dimethylamide sulfinate anion (48b from
41) to give another variation (49b). Although these routes passed
all the “must” criteria and scored well on cost, overall they
scored towards the bottom end of the range (70-101), mainly
due to a concern over their chemical feasibility, although
environmental issues and longer reaction sequences also
contributed. Given the low K-T scores, they were not investi-
gated even though the key steps were near the beginning of
A simple amide-ester exchange of the dimethyl ester (52a)
with Me2NH/ethanol in DMF could not be made to work under
standard or forcing conditions, and resulted only in ester
exchange to the mono and diethyl esters (52b/c) (Scheme 16).
Treatment with a large excess of aqueous Me2NH solution in
NMP gave a product tentatively identified as the desired amide
ester (53). However, this did not coelute on LC with the
previous sample made (prepared by methanol quench of the
acid chloride derived from the amide acid 56a; see below).
Monohydrolysis from diester 52a to acid ester 54 was also
unsuccessful, giving mainly the completely hydrolysed diacid
51. Further work on these approaches was stopped at this point,
due to the timing of the project. Given the problems with the
solubility of these compounds and the inability to address the
regioisomers issues, these routes did not look as promising as
first hoped.
Scheme 15
Scheme 16
the synthesis; only a limited number of routes could be
investigated, and the larger group 5 routes were chosen instead.
Symmetrical Bis-aryl Sulfone Routes (Group 9). These
routes were initially inspired by the idea of using the com-
mercially available antibacterial agent, Dapsone,26 (50) as a
starting material for AZD7545 (a late suggestion, not incorpo-
rated into the initial KTDA process, was to rely on a statistical
monodiazotisation and then displace the diazo group with a
halide, followed by a Pd-catalysed carbonylation).27 Even so,
some other related symmetrical difunctional molecules were
also commercially available on large scale and at low cost,
which looked like attractive targets, particularly the 4,4′-
sulfonylbis(methyl) benzoate (52a) and its parent diacid, 4,4′-
sulfonyldibenzoic acid (51), although the diester was only
available in 90% strength. If these could be differentially
functionalized, then the routes might be viable. The K-T analysis
The diacid 51 was prepared initially on laboratory scale by
complete saponification of the diester 52a but was also
potentially available on scale from a commercial source.
Attempted formation of the monoamide acid with thionyl
chloride and dimethylamine was unsuccessful. However, using
EDCI coupling of Me2NH in NMP with diacid 51 gave a largely
statistical mixture of unreacted diacid, desired amide-acid (56a)
and doubly reacted diamide 56b in ∼3:4:2 ratio, respectively
(Scheme 17). The isolated yield of 56a was only 30% due to
problems with the purification and relatively high losses to the
aqueous NMP liquors. Conversion to the primary amide 56c
under standard conditions (SOCl2, NH4OH, NMP) appeared to
be nearly quantitative, but again high losses to the liquors
resulted in a yield of only 35%. Various standard reagents to
achieve the Hofmann rearrangement all failed. For example,
alkaline NaOCl at 80 °C gave diacid 51, presumably by base
(25) Yoshida, Y.; Kimura, Y.; Tomoi, M. Chem. Lett. 1990, 769–772.
(26) The Merck Index, 14th ed.; O’Neil, M. J., Ed.; Merck and Co., Inc.:
Whitehouse Station, NJ, 2006.
(27) Earley, W. G.; Oh, T.; Overman, L. E. Tetrahedron Lett. 1988, 3785.
1054
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

Scheme 17
hydrolysis;28 bromine in NaOH gave no reaction;29 as did
NaOBr generated in situ and other reagents. However, the use
of BzMe3NBr3 did give a 35% yield of the desired aniline 57.30
An initial attempt to introduce chlorine using NCS in acetoni-
trile31 gave a 3:2 mixture of products by LC, later confirmed
by spectroscopy as the desired mono- and undesired dichloro-
amines (16 and 58), in an overall 30% isolated yield. Addition
of the chiral acid portion to aniline 57 to give 59 might have
tempered the reactivity sufficiently to give the desired monochlo-
ride (i.e., 1), but similar chemistry had proven more difficult
than expected when attempted by our medicinal chemistry
colleagues. Furthermore, the route selection phase of the project
had run out of time by this stage. Total time spent on these
stages, which had been plagued by solubility and isolation
issues, was about 3-4 months. Overall, whilst there was some
scope for improving these stages, unresolved regioisomer ratios
combined with low solubility and poor yields throughout meant
that considerable effort would be required to resolve these issues.
Consequently, no further work was conducted on these route
options.
rated into the analysis without having to restart the whole
procedure. However, since this was the first example, some
learning points were made:
•The “must” criteria were generally easy to decide and assess,
as they have to provide clear-cut yes/no answers. However,
the failure to include IP was an oversight (although it had
subconsciously been taken into account), and this was
rectified in all following examples.5
•Cost of goods. The approximation of using solely the cost
of chiral acid 2, in combination with the expected yields
through the syntheses, worked well in this case where cost
of goods was a major issue. However, we were fortunate
that the cost of chiral acid was such a dominant factor and
could be used in this way. In the more typical case,
determining the cost is likely to be much more complicated
(although it is likely that a small number of raw materials
and reagents will account for much of the cost so that
worthwhile approximations can still be made). In other cases,
the dosage may be low or the therapeutic gain high (e.g.,
for oncology projects), such that the cost of goods can largely
be ignored.
•Chemical feasibility. The calculation to try and assess the
yield and confidence of achieving it was the only parameter
that was too complicated in this KTDA exercise. Chemical
feasibility had been given the highest weighting in the
“wants” criteria and so was judged worth assessing in this
way. In retrospect, this procedure was too complicated and
time-consuming and not always of value, depending on what
the data was based on. Known figures based on exact or
similar reactions were probably worthwhile; those where a
more subjective judgment was required (even though this
was performed as a team exercise) were probably of much
less value. In future projects it was decided to use a simple
tally of the number of stages and apply a higher weighting
to this to compensate. This procedure has been found to be
Conclusions
In summary, KTDA worked well in this first case to focus
limited resources on the investigation of potential new manu-
facturing routes where a large number of potential route options
existed. The use of the spreadsheet (Table 2) as a living
document kept resources appropriately focused on promising
avenues of investigation, and allowed new data to be incorpo-
(28) Harvey, I. W.; McFarlane, M. D.; Moody, D. J.; Smith, D. M. J. Chem.
Soc., Perkin Trans. 1988, 1, 1939–1943.
(29) Lalezari, I.; Lalezari, P. J. Med. Chem. 1989, 32, 2352–2357.
(30) Kajigaeshi, S.; Asano, K.; Fujisaki, S.; Kakinami, T.; Okamoto, T.
Chem. Lett. 1989, 463–464.
(31) Nickson, T. E.; Roche-Dolson, C. A. Synthesis 1985, 669–670.
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1055

Table 3. Final KTDA data on 04 May 2001
musts
chemical
feasibility materials
route
no.
route description
(name)
raw
go/no initial initial
final
score
change in
score
go
score
rank
final ranka
Initial Routes
1.1
medicinal chemistry route
initial contractor’s route
1
1
0
1
no
no
95
98
34
31
98
98
+3
0
n/a
n/a
1.2
Campaign 2 Route Variations
2.1
2.2
2.3
2.4
2.5
2.6
contractor’s campaign 2 route
1
1
1
1
0
1
1
1
1
1
1
1
yes
yes
yes
yes
no
144
139
124
132
134
93
1
3
144
139
124
132
58
0
2
3
with catalytic Pd
with catalytic other metal
nonmetal route
with oxidation first
nitro thiol route
0
0
9
5
6
4
35
0
4
n/a
n/r
-78
yes
93
0
Alternative Leaving Group Variations
3.1
3.2
3.3
3.4
diazonium LG
tosylate LG
nitro LG
1
1
0
1
1
1
1
1
yes
yes
no
102
119
143
130
25
12
2
106
119
114
146
+4
0
-29
+16
n/r
n/r
n/a
1
nitrile route
yes
7
Selected Other Route Options
4.1
5.4
5.6
5.9
6.2
7.2
8.3
9.1
quetiapine-type best route
1
0
0
1
1
1
1
1
1
1
no
no
yes
no
134
111
104
129
103
107
70
4
16
23
8
24
19
38
13
124
100
111
112
105
108
71
-10
-11
+7
n/a
n/a
n/r
n/a
n/r
n/r
n/a
n/r
tosyl chloride and aniline
typical tosyl chloride based route
sulfobenzoic acid and aniline
tolyl grignard with Protected Amide
protected aniline Grignard
sulfinic acid (lowest scored route)
best bis-aryl sulfone route
1
0
-17
+2
1
yes
yes
??
1
+1
??
1
+1
yes
118
114
-4
a n/a ) not applicable; n/r ) not ranked.
a better guide for the level of effort required,5 given that
certainty cannot be known in any case.
solubility issues, the latter of which were not (and perhaps could
not have been) foreseen. Thus the group 5 routes failed the
“must” chemical feasibility criteria, which the group 9 routes
passed in principle, but in practice the lack of beneficial
solubility removed the possibility to resolve the failings of the
regioselectivity. In both cases though, investigation was still
felt to be justified because the key steps lay at the beginning of
the synthesis and represented alternative chemical approaches
to the same target molecule.
It is noteworthy that the eventual route chosen for long-
term manufacture (3.4) was the only route that significantly
improved its score (to 146), whilst most other routes either failed
“must” criteria or tended to decline in their scores. The
contractor’s second route (2.1) did maintain its place, which
indicated this too was a viable long-term alternative.
Finally, regarding the time requirements of the process (other
than literature searching, which would have been performed
anyway), the team of six invested about 4 h for the initial
meeting. After this, the lead chemist probably invested a couple
of days to process the data. The spreadsheets had to be
composed from scratch in this first instance, but once prepared,
the monitoring process was simple, and later projects benefited
from the initial development effort. This approach has been
reused about half a dozen times locally (in full) and in
abbreviated form more widely. It is more useful on projects
with a large number of route options; the effort involved (which
must be admitted is substantial) is repaid less on more
constrained targets where the route options may be more
obviously limited (e.g., by competitor IP, raw materials avail-
ability, structure of the molecule). However, there is also value
These modifications were applied to the following PDKi
project, along with other factors specific to those compounds.5
In retrospect, we felt that the KTDA process had led us to
the correct routes. Leaving aside the “must” criteria, which some
“paper” routes clearly failed when investigated in the laboratory,
the summary of “wants” scores were generally a good guide.
For example, both of the first two medicinal chemistry routes
(1.1-1.2) scored low against process chemistry criteria used
for the KTDA. However, what was essentially a step reorder
in the tactical contractor’s improved route (2.1), maintained its
high score throughout the process, justifying its use for the 10
kg campaign and consideration for future manufacture. Varia-
tions on this (routes 2.2-2.6) were clearly justified, as were
the similar group 3 routes. The eventual best route (3.4) chosen
for the following pilot plant scale manufacture did end up with
the highest score, albeit essentially equivalent to the kilo-scale
route (2.1). However, other aspects such as robustness and ease
of workup identified route 3.4 as significantly better overall.
These factors were not scored by KTDA (the ease of workup
for example could not have been known), but KTDA did high-
light this route as one of the better ones, and it is gratifying
that this route was genuinely the highest scoring at the end of
the process based on the initial parameters chosen. Other routes
(such as the promising route 3.3) could not be made to work
and so dropped out of the process at an early stage.
The scores for the sulfobenzoic acid/tosyl chloride routes
(group 5) and the symmetrical bis-aryl sulfone routes (group
9) suggested they were not really good enough to investigate,
and they did not improve over the period, as shown in Table 3.
Both groups of routes failed largely due to regiochemistry and
1056
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

in getting the team to accept ownership of the decision-making
process, so there are additional benefits in all cases.
on this scale in the laboratory). In a separate reactor vessel,
sodium bromide (30.0 g, 291 mmol, 0.75 equiv) was added to
methanol (150 mL, 3.0 vol) and cooled to 0 °C with stirring.
Bromine (65.8 g, 412 mmol, 1.05 equiv) was added slowly to
the second reaction vessel with stirring. (Caution: a 20 K
exotherm was observed on this scale in the laboratory.) Both
reaction mixtures were recooled to 0 °C, and the bromine/
methanol solution was added slowly to the first reaction mixture
(thiocyanate/aniline) over 1.5 h with stirring, maintaining the
temperature at 0-5 °C. The combined reaction mixture was
allowed to warm to 20 °C over 2-3 h. Water (1000 mL, 20
vol) was added slowly with stirring, resulting in an 8 K
exotherm, and a dense precipitate about halfway through the
addition. The mixture was stirred for 1.5 h, and then solid
sodium bicarbonate (70 g, 833 mmol, 2.1 equiv) was added in
portions (Caution: vigorous effervescence; CO2 released). Once
the effervescence had subsided, toluene (500 mL) was added,
and the mixture was stirred for 30 min and then filtered to
remove excess sodium bicarbonate. Methanol was distilled from
the solution under partial vacuum (40 °C at 250 mmHg) until
500 mL of distillate had been collected. Further toluene (500
mL) was added, and the distillation was repeated, collecting
another 500 mL of distillate. The lower aqueous phase was
removed, and water was added, stirred for 20-30 min, and then
separated off. The residual toluene phase was concentrated to
3 vol (150 mL) by distillation, whereupon a solid crystallized.
The reaction liquors were cooled to 0 °C, and the solid was
isolated by filtration and dried to give the title compound as an
off-white to pale yellow solid in typically 75-90% yield and
95% purity (as determined by HPLC).
Preparation of 4-Bromobenzamide (12). [Note: the me-
dicinal chemistry route and early work on the KTDA were
conducted using oxalyl chloride instead of thionyl chloride, but
the latter became the preferred method during the KTDA route
selection process. Ironically, oxalyl chloride was preferred again
for larger scale manufactures]. 4-Bromobenzoic acid (11) (50.0
g, 249 mmol), toluene (417 mL), and NMP (2.1 mL, 21.6 mmol
0.087 equiv) were charged to a nitrogen-flushed vessel and
heated to 60 °C with stirring. Thionyl chloride (31.1 g, 261
mmol, 1.05 equiv) was added dropwise over 15-30 min
(caution: gas evolution), and the reaction mixture stirred at this
temperature for 6 h and then cooled to 0 °C for 30 min.
Dimethylamine in ethanol (89 mL of a 33% solution (5.6 M),
498 mmol, 2.00 equiv) was added smoothly over 1 h maintain-
ing the temperature below 5 °C and then stirred for 2 h before
warming to RT. The reaction mixture was quenched with
aqueous HCl (560 mL of a 1.0 M solution), carefully at first,
and then mixed thoroughly for 30 min once the addition of
HCl was complete before settling. The lower aqueous phase
was separated off, and the organic phase was washed with water
(250 mL). After removal of the water wash, the reaction liquors
were heated to 40 °C, and the volume was reduced to 3 vol
(∼150-200 mL) by distillation of toluene at reduced pressure.
The reaction mixture was cooled again to 0 °C, and after 2 h at
this temperature, the crystalline product was isolated by filtration
and dried in a vacuum oven at RT to give the known compound
12 as a low melting waxy solid in typically 75% yield and
>95% purity as determined by HPLC.
Summary. Overall, the use of KTDA for route assessment
led to the selection of a new route for the pilot plant campaign
(10 kg) and longer term manufacture (>100 kg), which will
be discussed in more detail in a subsequent paper. In conclusion,
the route selection exercise worked well in this case where a
large number of potential routes were to be considered using
limited resources. It was also an effective means of bringing
some long-term manufacturing issues to the fore at an early
stage in development.
Experimental Section
General Note. Because of the nature of this project, it is
neither practical nor profitable to reproduce here the full
experimental detail for each reaction attempted. The experi-
mental procedures that follow are therefore representative of
those reaction steps that were reproduced more often and on
larger scale, generally from the group 2 and 3 route options. In
a similar vein, physical and spectral data is not included since
products of various qualities (and in some cases uncertain
identities) were taken through reaction sequences to prove a
particular point. If that reaction sequence was then discarded,
time did not permit complete characterizations to be obtained.
Full data is available for the more thoroughly investigated group
2 and 3 routes, and this will be reproduced in full in a following
manuscript. Full experimental details can in any case be found
for the early (group 1) routes in the medicinal chemistry patent.1
Procedures are given for the most reliable processes on large
laboratory scale at an early phase in the KTDA process.
General Procedures. Reaction mixtures and products were
analysed by reverse phase HPLC on Hewlett-Packard 1050 or
1100 instruments, generally with an eluent of acetonitrile/water
mixtures at a flow rate of 1.0-1.5 mL/min and a wavelength
of 230-254 nm. 1H and 13C NMR spectra were recorded on a
Varian Inova 400 spectrometer at 400 and 100.6 MHz,
respectively, with chemical shifts given in ppm relative to TMS
at δ ) 0. Electrospray (ES+) mass spectra were determined on
a Micromass LCT with time-of-flight. Analytical TLC was
carried out on commercially prepared plates coated with 0.25
mm of self-indicating Merck Kieselgel 60 F254 and visualised
by UV light at 254 nm. Preparative scale silica gel flash
chromatography was carried out by standard procedures using
Merck Kieselgel 60 (230-400 mesh).
Preparation of Thiocyanate (10). [Important note: the
following description is given as typical of a laboratory-scale
preparation of thiocyanate 10 at the beginning of the project
during the route exploration stage and was safe in our hands
on this scale. There is a significant interaction between
bromine and alcohols as noted in Bretherick,32 and the
procedure reported here should (preferably) not be repro-
duced at all and certainly not be scaled up further. A safe
alternative procedure will be reported in a planned later
manuscript]. 2-Chloroaniline (9) (50.0 g, 392 mmol) was
dissolved in methanol (500 mL, 10 vol) and cooled to 0 °C,
and sodium thiocyanate (63.6 g, 784 mmol, 2.0 equiv) added
in two portions with stirring (a 2-3 K exotherm was observed
(32) Bretherick’s Handbook of ReactiVe Chemical Hazards, 6th ed.; Urban,
P. G., Ed.; Butterworth-Heinemann: Oxford, 1999; Vol. 1.
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1057

Preparation of Sulfide Aniline (15). Thiocyanate (10) (4.0
g, 21.6 mmol) was dissolved in DMF (80 mL, 20 vol) with
stirring. In a separate vessel, sodium sulfide nonahydrate
(Na2S·9H2O) (6.24 g, 26.0 mmol, 1.20 equiv) was dissolved
in water (20 mL, 5.0 vol) with stirring (Caution: an 18 K
exotherm was observed on this scale in the laboratory). This
alkaline solution was added to the first reaction mixture over
10 min, resulting in a 2-3 K exotherm on this scale. The
reaction mixture was heated to 50 °C for 2 h and then
concentrated to 10 volumes (40 mL) by distillation to drive off
the residual water (as determined by Karl Fischer measurement).
The distillate was replaced with fresh DMF (40 mL, 10 vol).
Copper(I) oxide (0.31 g, 2.16 mmol, 0.10 equiv) and 4-bro-
mobenzamide (12) (4.94 g, 21.6 mmol, 1.00 equiv) were added,
followed by further DMF (80 mL, 20 vol), and the reaction
mixture heated to 150 °C for 6 h. If >75% sulfide aniline was
detected by HPLC, with <5% of 12 and dimeric 14b, the
reaction mixture was cooled to 40 °C and filtered through a
pad of Celite (if not, heating continued until these criteria were
met). Water (540 mL, 135 vol) was added slowly to the dark
brown filtrate over 30 min (an 18 K exotherm was observed
on this scale in the laboratory), and the reaction liquors were
then reheated to 50 °C for 1 h, before cooling back to 0 °C for
2 h, during which time a brown solid had formed. The product
was isolated by filtration and dried in a vacuum oven at 40 °C
to give the title compound 15 as a brown solid in typically
60-70% yield and ∼90% purity as determined by HPLC. This
subsequently had to be recrystallised from ethyl acetate for
further use.
Preparation of TMS-Protected Chiral Acid Chloride (3a).
Chiral acid 2 (45.0 g, 285 mmol), dichloromethane (675 mL),
and DMF (0.56 mL, 7.2 mmol 0.025 equiv) were charged to a
nitrogen-flushed vessel and stirred for 10 min. Bis-TMS urea
(75.7 g, 370 mmol, 1.30 equiv) was added in portions over 15
min, and the reaction mixture was stirred at RT for 4-8 h until
>95% conversion of the di-TMS-protected intermediate had
formed (by GC). The reaction mixture was filtered to remove
the insoluble urea precipitate that had formed, and the resulting
solid was washed twice with dichloromethane (298 mL each).
The washes were added to the reaction liquors and charged to
a second nitrogen-inerted vessel and cooled to 0 °C. In a
separate vessel, oxalyl chloride (29.8 mL, 342 mmol, 1.20
equiv) was dissolved in dichloromethane (90 mL) and then
added to the reaction mixture over 15-30 min. The reaction
mixture was stirred at 0 °C for 2 h, allowed to warm to RT
over 2 h, and stirred at RT for 2 h. If the reaction was >95%
complete by GC, the mixture was filtered to remove any residual
urea that had formed and stored cold in a sealed vessel.
Alternatively, 3a could be isolated neat after distillation under
reduced vacuum at 175-250 mbar and 30 °C (to remove
dichloromethane but avoid loss of the volatile product) as a
pale yellow oil. The strength of the dichloromethane solution
was assayed against a purified sample. The typical overall yield
was 80%.
protected acid chloride (3a) (105 g in 2000 mL of dichlo-
romethane, 422 mmol, 1.30 equiv), prepared shortly before
required use, was added over 1 h keeping the temperature below
5 °C. The reaction mixture was allowed to warm to RT over
6 h, and then dichloromethane was distilled off, collecting ∼920
mL distillate. Aqueous HCl (1300 mL of a 2 M solution) was
added, and the reaction mixture was stirred for 30 min. The
lower organic phase was separated off, and further water was
added (1300 mL) and stirred for 30 min. The organic phase
was separated off again and concentrated to dryness to give
the title compound as a sticky solid/gum in typically 75-80%
yield, which was used without further purification.
Preparation of Crude AZD7545 (1). Glacial acetic acid
(150 mL, 1.5 vol) was used to wash crude TMS-protected
sulfide anilide (13a) (325 mmol scale) into a reaction vessel
containing further glacial acetic acid (400 mL, 4.0 vol).
Hydrogen peroxide (35 wt % solution, 307 mL) was added,
and the reaction mixture was heated to 90 °C for 3 h. Water
(1330 mL, 13.3 vol) was then added over 30 mnutes, and the
reaction mixture was cooled to RT. Dichloromethane was added
(2050 mL, 20.5 vol), and the reaction mixture was stirred for
30 min before separating off the lower organic phase. This was
washed with sodium bicarbonate solution (2050 mL) for 30
min, separated, and washed with water (2050 mL) for 30 min
before being separated off again. The organic phase was
concentrated to dryness to give crude AZD7545 as a sticky
solid/gum in typically 75-80% yield and >90% purity as
determined by HPLC.
Preparation of Friedel-Crafts Product (34b).19 Iron(III)
chloride (8.6 g, 53.2 mmol, 1.05 equiv) was added to a dried
nitrogen-purged reaction vessel fitted with a condenser, me-
chanical stirrer and acidic gas scrubbing apparatus. Tosyl
chloride (32) (10.2 g, 56.0 mmol, 1.10 equiv) was added under
a nitrogen purge, followed by dichlorobenzene (33b) (7.5 g,
51.0 mmol). The reaction mixture was heated to 100 °C with
mechanical stirring, during which time acidic gases were
released. After 2 h, all of 33b had been consumed. The reaction
mixture was cooled slightly, aqueous sodium hydroxide was
added (1 N, 25 mL), and the mixture reheated to 100 °C briefly
to destroy any excess 32 remaining. The reaction mixture was
cooled to RT, ethyl acetate (100 mL) and water (75 mL) were
added, and extractionwas attempted, which had a poor phase
separation. Sodium chloride (20 g) was added, and the mixture
was shaken vigorously, but the separation did not improve, so
the mixture was filtered through a pad of Celite, which gave
two clear phases. The organic phase was washed with dilute
brine (50 mL), separated, dried over MgSO4, and concentrated
to dryness under reduced pressure to give a yellow-brown oil
that crystallized on standing. (15.6 g). The crude product was
recrystallised from hot MTBE (100 mL) and iso-hexane (60
mL) to give the title compound 34b as a yellow-brown solid
(4.8 g, 31%).20 A second crop was obtained from the mother
liquors (2.2 g, 14%), giving an overall yield of recrystallised
34b of 7.0 g (45%).
Preparation of Crude TMS-Protected Sulfide Anilide
(13a). Sulfide aniline (15) (100 g, 326 mmol) was dissolved in
dichloromethane (220 mL, 2.2 vol) and triethylamine (131 mL,
95.3 g, 942 mmol, 2.90 equiv) and cooled to 0 °C. TMS-
Preparation of Benzoic Acid (38). Friedel-Crafts product
(34b) (6.03 g, 20.0 mmol), potassium permanganate (15.8 g,
100 mmol, 5.0 equiv) and potassium bicarbonate (4.0 g, 40.1
mmol, 2.0 equiv) were charged to a reaction flask followed by
1058
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

tert-butanol (22 mL) and water (22 mL). The resultant purple
solution was heated to 100 °C (caution: vigorous effervescence
initially) for 24 h to give a colourless solution and a sticky dark
solid. The reaction mixture was cooled to RT, sand (25 g) was
added, and the mixture was filtered. The residual sand/MnO2
solid was rinsed with tert-butanol/water (1:1, 50 mL), and the
combined filtrates were filtered through a small pad of Celite
to remove final traces of MnO2 to give two clear phases. This
reaction solution was used to investigate several alternative
workup procedures involving acid and base washes, extractions,
and crystallisations. The overall yield of the title compound 38
obtained as a solid was 5.8 g (88%) by these various techniques.
Acknowledgment
J.D.M. thanks Danny Levin for helpful discussions at the
start of this project and Kepner-Tregoe for their support in
presenting this case history.
Received for review February 14, 2008.
OP800033C
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1059