Statistical DoE approach to the removal of palladium from active pharmaceutical ingredients (APIs) by functionalized silica adsorbents

Article abstract of DOI:10.1021/op5000336

The influence of four parameters (temperature, scavenging time, amount of scavenger, and concentration of palladium in the solution) on the efficiency of Pd removal from a cross-coupling reaction, using a commercially available Pd scavenger, SPM32, was studied. The DoE-based method employed yielded more information than is readily attainable from standard adsorption isotherms and kinetics experiments. The optimal regime of scavenging was identified; intuitive and nonintuitive effects of temperature, scavenging time, and scavenger amounts were highlighted; and a mathematical model quantifying predicted Pd removal from the synthetic intermediate was built.

Full text of DOI:10.1021/op5000336

Article
pubs.acs.org/OPRD
Statistical DoE Approach to the Removal of Palladium from Active
Pharmaceutical Ingredients (APIs) by Functionalized Silica
Adsorbents
Jan Recho,* Richard J. G. Black, Christopher North, James E. Ward, and Robin D. Wilkes
PhosphonicS Ltd., 44c Western Avenue, Milton Park, Abingdon, OX14 4RU, United Kingdom
S
* Supporting Information
ABSTRACT: The influence of four parameters (temperature, scavenging time, amount of scavenger, and concentration of
palladium in the solution) on the efficiency of Pd removal from a cross-coupling reaction, using a commercially available Pd
scavenger, SPM32, was studied. The DoE-based method employed yielded more information than is readily attainable from
standard adsorption isotherms and kinetics experiments. The optimal regime of scavenging was identified; intuitive and
nonintuitive effects of temperature, scavenging time, and scavenger amounts were highlighted; and a mathematical model
quantifying predicted Pd removal from the synthetic intermediate was built.
that bind strongly to metal residues. Removing the catalyst with
high affinity metal scavengers containing functional groups
which can effectively compete to remove the metal from the
API is often then the only viable solution.
1. INTRODUCTION
In recent years, the use of transition metal-mediated reactions
has become common in large scale active pharmaceutical
ingredient (API) synthesis due to the wide array of organic
chemistry transformations that these metals can catalyze.1,2
Among transition metals, palladium plays a prominent role in
modern organic chemistry, with commonly used Pd-catalyzed
reactions including hydrogenations, carbon−carbon cross-
couplings (Sonogashira, Suzuki, Heck, Stille, Hiyama),
carbon−heteroatom couplings (Buchwald-Hartwig), functional
group deprotections and cyclization reactions. While this
versatile methodology has provided organic chemists with
exciting new possibilities, it has also presented a new challenge
of removing the traces of Pd left in solution after the reaction.3,4
The toxicity of Pd has led major regulatory and advisory bodies
(EMEA,5 ICH6) to issue tight guidelines, with the EMEA, for
example, advising an oral concentration limit of 5 mg/kg, and a
parenteral concentration limit of 0.5 mg/kg for Pd in drugs.
The presence of additional Pd residues in synthetic
intermediates can also affect the subsequent steps in a synthesis,
catalyzing the formation of unwanted side products and
impurities.7
Functionalized silicas are well-known as efficient metal
scavengers and can be used in both small scale and large
scale operations,16−19 either in batch or in continuous
(cartridge) format.16,20 Silicas are stable in all of the common
organic solvents, as well as in aqueous solutions (provided pH
< 9); they do not swell, generate low back pressure (if any) in a
cartridge, and their high porosity provides ample surface area
for optimal contact between the scavenger functional groups
and the API solution.21 However, while the use of function-
alized silicas as scavengers to solve Pd removal issues is ever-
increasing, to the best of our knowledge very few quantitative
studies have been conducted on their scale-up and to
understand the parameters affecting Pd removal. One such
study of note has been carried out by Girgis et al.,22 although
the focus was less on characterization of the scavenging
operation itself, but more on the transition from a batch mode
of operation to a continuous-flow, cartridge-based mode of
operation.16
This paper describes a systematic study of the influence of
several key parameters on the scavenging of Pd from an organic
product solution using functionalized silica scavengers, with the
aim of building a mathematical model to quantitatively describe
the scavenging operation. While this paper provides valuable
insights into which parameters should be carefully monitored
by process chemists when scaling up a Pd scavenging operation,
it should be noted that by their very nature, such operations are
very dependent upon the type of scavenger used, the nature
and speciation of the metal, and the nature of the product
solution to be scavenged. The relative importance of each
parameter might be expected to change from one product
solution to the other, and from scavenger to scavenger. The
To address these regulations, several strategies are used by
process chemists to remove Pd impurities.4,8 In cases where Pd
is not tightly bound to the API, or where the Pd species present
can be easily removed, a simple solution can sometimes be
found, such as an aqueous wash, distillation, crystallization,9 or
adsorption onto activated charcoal.10 Other methods to work
around the problem include using alternative synthetic routes11
or heterogeneous catalysis,12−14 but both of these options
present additional consequences. In many cases, the metal is
tightly bound to the API, and removing it presented a
significant challenge, prompting long and costly studies.15 This
is especially true for densely functionalized synthetic inter-
mediates made for particular applications within the pharma-
ceutical industry. Indeed, some of the most common functional
groups that provide biological affinity or performance tend to
feature combinations of heteroatoms and electron-rich moieties
Received: January 29, 2014
Published: April 14, 2014
© 2014 American Chemical Society
626
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
general methodology employed, however, is intended to be
adaptable to other solid phase scavengers, other metals, and
other product solutions.
threshold of 15% Pd retained in the product solution after
workup was chosen. (3) The solution had to strongly retain Pd
so that a metal scavenger would be required for its effective
purification. To assess this criterion, a kinetics test was
conducted on candidate product solutions using 2 mol equiv
of the scavenger, SPM32. Those from which more than 60% of
the Pd was removed within the first hour were rejected on the
basis that they were insufficiently powerful to compete with the
scavenger for Pd.
A number of product solutions were trialed against these
three requirements, either by using the crude product of a well-
known Pd-catalyzed reaction (Suzuki, Heck, amino acid N-
debenzylation) or by directly dissolving a molecular species
bearing potential Pd chelating functionalities in an appropriate
solvent (see Supporting Information, section 6, for details).
The crude reaction product of the Buchwald-Hartwig cross-
coupling reaction to give N-aryl-α-amino acid 3 met these
criteria and was chosen as the reaction partner for this DoE
study (Scheme 1).
2. RESULTS AND DISCUSSION
Building a mathematical model to represent Pd adsorption onto
a scavenger required several steps: (1) selection of one
scavenger23 and a model product solution from which to
scavenge the Pdthe model solution should realistically reflect
the characteristics of a synthetic intermediate; (2) character-
ization of the interaction between the product solution and the
scavenger following an accepted process; (3) study of the
scavenging process and identification of the key parameters that
affect scavenging, which will be incorporated into the study; (4)
identification of a suitable Design of Experiment (DoE) design
space: the ranges over which each of the previously identified
parameters will be studied, allowing a mathematical model to
be built that can accurately model Pd adsorption from the
chosen API-like molecule, depending on the variations of the
chosen parameters.
Scheme 1. Buchwald-Hartwig Reaction, the Crude Product
of Which Was Used as a Model Solution for This Study
2.1. Development of a Suitable API Intermediate/
Scavenger System Mimic. This study focused on Pd
contamination of API-like products because Pd is widely used
in organic synthesis,1,2 and its removal is therefore an important
challenge for the pharmaceutical industry to overcome.
PhosphonicS SPM32 scavenger was chosen for this study, as
it is used in both academia24 and within the pharmaceutical
industry and has been incorporated into the purification of a
number of APIs by both pharmaceutical and biotechnology
companies,25 including at the multikilogram scale.16 Although
other options are available and used at scale,26 SPM32 is one of
the most efficient scavengers for Pd removal from APIs,
consistently achieving suitable Pd removal and proving robust
under most operating conditions. The differentiating factor of
PhosphonicS technology is the presence of additional
heteroatom functional groups within the linker to the surface
of the silica support. For SPM32, this appears as a dialkyl
sulfide group providing an extra ligand binding site for Pd
(Figure 1), in addition to the terminal thiol functionality which
is present in competitors’ materials.
All of the reaction components are readily available, a high
amount of Pd (typically 400−500 mg/kg) was consistently
retained in the product solution, and the product was
sufficiently competitive for Pd in the above-mentioned kinetics
test. Furthermore, N-aryl-α-amino acids are a synthetically
valuable group of building blocks used in the preparation of a
number of biologically important molecules.28
The experimental procedure for this reaction is described in
the Supporting Information. A two-step filtration of the crude
reaction solution, first though a 10 μm pore sized sintered
funnel, then through a 0.1 μm nylon filter to remove any
suspended matter (mainly cesium carbonate, which is only
sparingly soluble in toluene), results in a clear solution. This
provided a study medium which was found to contain 426 mg/
kg of Pd. This concentration can be easily tuned by
concentration or simple dilution in toluene.
Figure 1. PhosphonicS SPM32 Pd scavenger.
2.3. Initial Characterization of the System. An accepted
procedure to characterize the adsorption of a metal from the
liquid phase onto a solid surface is to study the kinetic and
thermodynamic properties of the adsorption under controlled
conditions.22,29−31
Scavenger screening studies within the pharmaceutical
industry have regularly found SPM32 and other multifunctional
silica scavengers to be more efficient for Pd removal and a
computational study has recently been conducted to better
understand the importance of the thioether unit in this
regard.27 The functional group loading (FGL) of the SPM32
scavenger comprises between 0.8 and 1.1 mmol of functional
group per gram of silica, with the FGL of the particular batch of
scavenger used in this study measured at 0.93 mmol/g (see
section 2, Supporting Information).
In order to pick an appropriate molecule featuring API-like
functionality for the study, the following characteristics were
deemed necessary: (1) The chemical composition of the
solution had to be viewed as a realistic model of a synthetic
intermediate by pharmaceutical chemists. (2) The concen-
tration of retained Pd in solution had to be sufficiently high that
it could be easily tuned to fit the study’s needsa lower
An adsorption isotherm describes the equilibrium reached
during the adsorption process at a constant temperature. Three
useful pieces of information are thus generated: the efficiency of
metal removal from the liquid phase as a function of the
quantity of scavenger present; the overall potential of the
scavenger for removing metal from solution (i.e., the final
concentration of metal in solution at equilibrium); and an
indication of the maximum loading of metal onto the scavenger
using an excess of metal with respect to the scavenger.
An isotherm curve describing the equilibrium state for the
interaction of Pd scavenger SPM32 with a toluene solution of
N-aryl-α-amino acid 3 was generated at 60 °C and is shown in
Figure 2.
627
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
Figure 2. Isotherm curve for the scavenging of the crude Buchwald-Hartwig product with PhosphonicS SPM32 scavenger (60 °C, 24 h contact
time). Pd concentration before scavenging co(Pd) = 393 mg/kg.
Figure 3. Kinetics curve for the scavenging of the Buchwald-Hartwig product with PhosphonicS SPM32 scavenger (60 °C, 3 ME of SPM32 vs Pd,
300 rpm agitation speed). Pd concentration before scavenging co(Pd) = 393 mg/kg.
It was decided that 24 h was sufficient to allow equilibrium of
the adsorption process to be reached (see kinetics profile in
Figure 3). The convex curve of this graph represents a
“favorable isotherm” whereby a large quantity of Pd is retained
on the scavenger as an excess of scavenger is introduced. The
residual Pd in the product solution could be reduced to below
the limit of detection of the analytical method, proving that
SPM32 is an appropriate choice for this product solution.
Additionally, up to 60 g of Pd can be loaded per kg of
scavenger: a process chemist wishing to recover part of the cost
of the Pd by sending loaded scavenger to a metal refiner would
get a good economic return in this instance, increasing overall
process efficiency.
A kinetics profile of the adsorption reaction further assesses
the compatibility of the scavenger with the product solution
being treated. The kinetics curve for the interaction of 3 mol
equiv of SPM32 with the toluene solution of N-aryl-α-amino
acid 3 at 60 °C is shown in Figure 3. The exponential-like
decrease of the Pd concentration in solution demonstrates a
favorable reaction rate with 84% of the metal being removed
within 4 h of treatment.
important parameters would be required to build a picture for
the optimal scavenging conditions.
The limits of using an isotherms/kinetics assessment to
understand the behaviour of Pd removal using solid phase
scavengers are quickly reached. If a single isotherm is measured
at a single temperature, only a very limited picture of the
thermodynamics of the scavenging reaction will be obtained.
Better practice is to perform the isotherm at several
temperatures, typically 3, in order to have an understanding
of the effect of temperature on the Pd removal. In addition, a
kinetics curve describes the system at one temperature only and
at one ratio of scavenger to metal only.
The number of experiments required to create these two
curves can also be optimized, i.e. with a similar number of
experiments, one could obtain far more information about the
system’s behaviour.
An alternative approach is to use statistical DoE, a well-
known and widely used method32−34 to obtain the largest
possible amount of information on a particular process from the
minimum number of experiments. Furthermore, it also provides
a clearer picture of how the studied parameters influence the
variation in performance.
Both the reaction rate (kinetics) and the equilibrium position
(isotherm) of this metal adsorption process provided useful
data for judging the potential of the scavenging process, but
adopting this approach limits the data available to one
temperature, 60 °C, and in the case of reaction rate, one
single ratio of scavenger to metal (in this case 3 ME). Lots of
additional experiments at various temperatures for each of the
2.3. Design and Implementation of the DoE Experi-
ments. 2.3.a. Design Space Definition. A DoE-based
methodology was used to study the influence on the overall
efficiency of a scavenging operation of not only the two
parameters that are most commonly studiedmolar equiv-
alents (ME) and time (t)but also two further parameters that
experience has shown to be critical to scavenging operations
628
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
the initial concentration of Pd in the solution (before
scavenging), co(Pd), and the temperature, T, at which the
process is run. The ranges over which these four continuous
parameters were studied defines the experimental space, and
these are shown in Table 1.
Table 2. Pd Concentration of the Five API-Like Product
Solutions
targeted Buchwald-Hartwig reaction product
solution
measured Pd concentration
(mg/kg)
100 mg/kg Pd Buchwald-Hartwig solution
200 mg/kg Pd Buchwald-Hartwig solution
300 mg/kg Pd Buchwald-Hartwig solution
400 mg/kg Pd Buchwald-Hartwig solution
500 mg/kg Pd Buchwald-Hartwig solution
100.0
202.4
314.8
411.4
507.3
Table 1. Definition of the Experimental Space of the DoE
a
Study
parameter studied
Temperature - T
range
points measured
30 °C →
90 °C
30 °C, 45 °C, 60 °C, 75 °C, 90
°C
Molar equivalent of
0.6 → 3.0
0.6, 1.2, 1.8, 2.4, 3.0
Design of the Experimental Matrix. The 29 experiments
optimization matrix was designed in collaboration with Dr.
Andrei Zlota of The Zlota Co. LLC using the DoE Fusion PRO
software (model robustness option).35 The matrix included a
duplicated center point, and two additional replicates allowing
for the estimation of experimental error. All experiments were
carried out on the same day, using the same batch of SPM32
and the same batch of Buchwald-Hartwig reaction product
solutions, to minimize any uncontrolled variations in the
experimental conditions. Detailed experimental conditions can
be found in the Supporting Information.
The parameters’ values for each of the experiments in the
matrix are shown on the left columns in Table 3, with the
results on the right columns, in terms of residual Pd
concentrationcf(Pd)and percentage of Pd scavenged%
Pd.
A noteworthy result is that entry #9 yields not only the
lowest residual amount of Pd in solution, cf(Pd), but also the
highest percentage of Pd scavenged. In contrast, entry #16
yields the highest residual amounts of Pd in solution, but this
does not correspond to the lowest percentage of Pd removal.
Depending on the application the scavenger is used for
(purification of the solution, or recovery of Pd), the optimized
values of all 4 parameters may be different.
The residual Pd concentration, cf(Pd), will be focused on, as
it is the key result, because a careful control of this number is
most often needed during API purification. The variability
observed in Table 3 for the process result cf(Pd), with
concentrations between 21.7 mg/kg and 423.9 mg/kg, shows a
good signal/noise ratio. The 21.7 mg/kg lower limit was
deemed suitable in order to ensure precision in the measure-
ments: indeed, under 20 mg/kg, the error on Pd concentration
measurements rises over 5%.
scavenger to Pd - ME
Contact time between
scavenger and solution - t
20 min →
180 min
20 min, 60 min, 100 min, 140
min, 180 min
Concentration of Pd in the 100 mg/kg → 100 mg/kg, 200 mg/kg, 300
product solution - co(Pd)
500 mg/kg
mg/kg, 400 mg/kg, 500
mg/kg
a
Upper and lower limits were chosen based on feasibility and effect of
the studied parameter on the scavenging process.
The initial concentration of Pd in the solution before
scavenging, co(Pd), may appear to be a surprising choice, as it is
generally not a tunable parameter: Pd must be removed from
the product solution regardless of co(Pd). However, a good
understanding of the effect of initial concentration of Pd allows
the provision of a better cost-to-benefits model at an early stage
of the project.
As shown in Table 1, the experimental matrix was designed
with data points measured at five regular intervals throughout
the chosen range.
Temperature, T, was varied between 30 and 90 °C. The
boiling point of toluene is 110 °C, and the upper limit of 90 °C
was chosen to allow a wide error margin. The lower limit of 30
°C is the minimum temperature that could be reproducibly
achieved without purposely cooling the experiments during the
scavenging process.
Molar equivalent of scavenger, ME, was varied between 0.6
and 3.0, with these limits being based upon the isotherm curve
(Figure 2)varying the ME between these values ensures a
large variation in the residual levels of Pd, which can be easily
measured with small relative errors.
Reaction time, t, was varied between 20 and 180 min. The
kinetics curve (Figure 3) shows that 85% of the Pd is removed
after 180 min, using this as the upper limit ensures that
significant levels of Pd remain in the solution, simplifying trace
analysis. A minimum of 20 min ensures that human error on
small time measurements will have no significant impact.
Concentration of Pd in the solution, co(Pd), was varied
between 100 mg/kg and 500 mg/kg, which is the range in
which the use of metal scavenger is typically the most cost
efficient option. It would pose no particular experimental issue
to extend this range if a project required it.
2.3.b. DoE Study. Description of the Experimental
Procedure for the DoE Experiments. All five Buchwald-
Hartwig product solutions at five levels of Pd concentration
were made from the same batch of crude Buchwald-Hartwig
product, as described in the Supporting Information, to which
additional solvent (toluene) was added or removed under
vacuum to reach the desired Pd concentration. Metal content
analyses were run using inductively coupled plasma−optical
emission spectroscopy (ICP-OES), and are shown in Table 2.
2.4. Quantifying the Influence of the Four Parame-
ters−Data Analysis. A coded variable is assigned to each
parameter, which is proportional to each parameter, but varies
between −1 and +1 (Table 4).
A mathematical quadratic model linking the residual Pd
concentration, cf(Pd), to all four parameters was devised using
powerful quality-by-design software,36 and found to fit the data
well. The mathematical model obtained is shown in Table 5.
The error between this mathematical model and the data used
to build it was found to be minimal (see Table 6). More
detailed information on the design of the DoE study and the
statistical analysis of the results may be found in the Supporting
Information.
This model gives the equation of a response surface. For each
point on this response surface, the value of cf(Pd) corresponds
to the values of the four parameters. Projections of this
response surface give a clearer picture of the behaviour of the
scavenger when two parameters are varied. For example, Figure
629
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
Table 3. DoE Experimental Matrix for the Study of the
Influence of Temperature (T), Scavenging Time (t), Molar
equiv of Scavenger vs Pd (ME), and Initial Pd Concentration
(co(Pd)) on Scavenging Efficiency
Table 5. Mathematical Model Obtained, in Coded Variables
coded variable model
A = co(Pd)
B = T
C = time
D = ME
reaction conditions − 20 g of
cf(Pd) =
solution treated
results of the experiments
residual Pd Pd
+114.2618
c0(Pd)
(mg/
kg)
+109.0803 (A)
−31.6922 (B)
T
t
ME
concentration cf(Pd) removal
entry
(°C) (min) (equiv)
(mg/kg)
(%)
−26.6544 (C)
−79.3941 (D)
+21.6016 (B)2
+33.8644 (C)2
−19.3548 (A * B)
−6.2749 (A * C)
−57.1664 (A * D)
−14.2782 (B * D)
−6.7416 (C * D)
#1
#2
#3
#4
#5
#6
#7
#8
100
400
100
100
300
300
500
400
300
500
500
200
100
200
200
500
500
200
300
100
100
500
400
400
500
300
100
100
500
30
45
30
90
60
30
90
45
90
30
30
75
60
45
45
60
30
75
60
30
30
90
75
75
30
90
90
30
90
20
140
100
180
100
180
180
60
0.6
1.2
3.0
0.6
1.8
0.6
0.6
2.4
3.0
3.0
1.8
2.4
3.0
1.2
2.4
0.6
0.6
1.2
1.8
0.6
3.0
0.6
2.4
1.2
3.0
0.6
3
94.1
239.8
36.3
6%
42%
64%
29%
67%
25%
30%
58%
93%
67%
28%
84%
53%
44%
60%
16%
17%
46%
68%
4%
70.9
103.9
237.3
357.0
173.5
21.7
#9
180
180
20
#10
#11
#12
#13
#14
#15
#16
#17
#18
#19
#20
#21
#22
#23
#24
#25
#26
#27
#28
#29
170.0
364.5
31.6
Table 6. Error Analysis of the Mathematical Model Obtained
140
20
error statistic name
statistical value
R2
0.9961
0.9935
46.6
adjusted R2
140
60
114.1
80.9
20
423.9
419.8
108.4
100.5
96.0
100
60
100
20
180
180
140
60
29.1
71%
30%
83%
44%
39%
23%
69%
51%
77%
354.6
69.2
230.9
311.7
244.0
31.5
20
20
20
180
20
1.8
3.0
49.5
118.4
Table 4. Relationship between the Coded and Natural
a
Variables
coded
range
parameter studied
temperature - T
natural range
Figure 4. Variations of cf(Pd) with ME and co(Pd), when temperature
and time are fixed at 60 °C and 180 min.
30 to 90 °C
−1 to 1
−1 to 1
−1 to 1
molar equivalent of scavenger (to Pd) - ME 0.6 to 3.0
contact time between scavenger and solution 20 to 180 min
- t
combinations) can be sorted from that with the biggest effect
on the scavenging efficiency, to that with the least, as
represented in the Pareto chart (Figure 5). This ranking allows
assessment of which parameter change would have the most
important effect on a scavenging operation.
concentration of Pd in the product solution - 100 to 500
co(Pd)
−1 to 1
mg/kg
a
Points were measured for coded parameter values of −1, −0.5, 0, 0.5,
and 1.
Variations in the initial Pd concentration co(Pd) have the
biggest effect on the final Pd residual concentration cf(Pd); as
would be expected, the amount of scavenger required to reduce
Pd to low levels increases with Pd input. The initial
concentration of Pd in solution co(Pd) is not necessarily a
parameter that can be tuned, but understanding the effects of its
variations on Pd removal improves the process understanding.
The next important parameter is the molar equivalent (ME) of
scavenger with respect to Pd, which therefore remains an
efficient variable to alter in order to improve Pd removal.
Scavenging time and temperature appear less important,
accounting for an impact on scavenging of roughly a third of
4 is the projection of the response surface defined by T = 60 °C
and t = 180 min.
As expected, it shows a decrease in residual Pd when ME is
increased (more scavenger added) and when co(Pd) is
decreased (less Pd initially in solution). The dark blue area
represents residual Pd concentrations below 28 mg/kg. For
example, the model predicts that at 60 °C and 180 min of
contact time, 1.0 ME of scavenger should be enough to treat a
solution containing 100 mg/kg of Pd to below 28 mg/kg.
2.4.a. Ranking the Most Efficient Parameters. Based on the
data shown in Table 3, the parameters studied (and their
630
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
Figure 5. Pareto chart - Parameters in order of decreasing importance (left to right).
Figure 6. Interaction between c0(Pd) and ME. The red line represents the influence of co(Pd) on the result cf(Pd) at ME = 3. The blue shows the
corresponding effect with ME = 0.6, T = 60 °C, and t = 100 min.
the effect of co(Pd). They represent the easiest parameters to
manipulate in a scavenging process.
2.4.b. Effect Interactions. Interactions between parameters
can highlight where a parameter change might be the most
efficient. For example, Figure 6 shows the interaction between
co(Pd) and the ME of scavenger.
The nonparallel nature of two lines shows that increasing the
ME will not have the same magnitude of effect at high or low
values of co(Pd). At low values of co(Pd), an increase in ME will
have an effect on cf(Pd) that will be significantly different to its
effect at high co(Pd) values.
Parameter A*D, a two-factor interaction which represents a
simultaneous variation of co(Pd) and ME, has the third biggest
effect on the result. An important A*D interaction means that
varying factors A and D at the same time does not result in an
effect that is simply the sum of the effects of varying A alone,
and D alone. In this case, it means an increase in ME will have a
different impact on cf(Pd) depending on whether one works at
high or low co(Pd).
The possibility of controlling the reaction with a given
parameter should therefore be carefully considered to achieve
631
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
Figure 7. Response surface contour plot. Variation of cf(Pd) with ME and co(Pd). T = 60 °C and t = 100 min.
Figure 8. Response surface contour plot. Variation of cf(Pd) with ME and co(Pd). T = 90 °C and t = 180 min.
the optimum effect, depending on the initial concentration of
Pd co(Pd), it may or may not be worth varying the ME. Even
more than simply highlighting parameters which will have only
a marginal effect on scavenging efficiency, fitting a mathematical
632
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
model to the data can also show areas where varying parameter
levels will have a negative effect on scavenging.
Alternatively, the nature of the interaction between the
scavenger and the Pd species might change, from mono- to
bidentate for example.
One point to highlight is that a DoE experimental matrix,
while requiring roughly the same amount of measurements as
plotting isotherms and kinetics at three temperatures, will yield
far more valuable data, and lead to a better understanding of the
scavenging process.
2.5. Verification Experiments. The mathematical model
obtained provides an easy way of predicting the behaviour of
this product solution/scavenger system, within the design
space.
To examine the validity of this model, two sets of verification
experiments were carried out. One set was performed on the
same 20 g scale and with the same apparatus as the 29
experiments that were used to build the model, and tests the
validity of the model itself. The second set of verification
experiments was carried out at a slightly larger scale, to provide
an initial assessment of the impact of scale-up on the reliability
of the model developed.
The verification experiments were carried out under the
following conditions: (1) T = 60 °C, t = 100 min, co(Pd) = 200
mg/kg, 2.6 ME (predicted residual from the mathematical
model: cf(Pd) = 24 mg/kg); (2) T = 60 °C, t = 100 min, co(Pd)
= 300 mg/kg, 2.8 ME (predicted residual from the
mathematical model: cf(Pd) = 45 mg/kg).
The scavenging parameters for the verification experiments
were chosen for several reasons. One, they are mild scavenging
conditions: between the extremes of temperature, not requiring
a long contact time, and not using large amounts of scavenger.
In other words, in a working domain that would be appealing to
the industrial chemist.
Two, the predicted residual Pd concentration cf(Pd) falls in a
range allowing for more accurate measurements. It was noted
that the relative error on cf(Pd) measurements increases above
5% on measurements of Pd concentrations below 20 mg/kg.
The 50 mg/kg upper limit was set in order to demonstrate
the validity of the model in an area where the conditions still
define a meaningful scavenger operation.
2.5.a. Model Validity. Different batches of product solution
were prepared and used for the verification experiments; hence,
a slight variation of co(Pd) can be observed compared to
previous experiments. The results of the two small scale
experiments are shown in Table 7.
2.4.c. Temperature and Time: A Nonevident Effect on
Residual Pd. It would not be unreasonable to assume that
increasing the contact time, t, between a scavenger and a
product solution leads to lower residual Pd levels. Also, the
effect of the temperature on scavenging efficiency is often
assumed to be monotonic, either detrimental or beneficial
depending on the type of scavenger considered. When the
effect of temperature is studied by traditional methods, i.e., by
plotting isotherms at three temperatures, its impact on Pd
removal is assessed only at the reaction equilibrium. If instead
kinetics are plotted at three temperatures, then the effect of
temperature on Pd removal can only be assessed at one specific
ME of scavenger. A large number of such “isotherm/kinetics at
three temperatures” studies have been carried out within the
field of metal adsorption.22,29−31
This study highlights the fact that the temperature and time
parameters can actually be correlated with other parameters in
the scavenging process. Under certain conditions, an increase in
contact time, t, and temperature, T, can actually lead to
complex and possibly detrimental effects on the scavenging
efficiency. This would not be detected by running three
isotherms at three different temperatures, except if they
happened by chance to be run under exactly the right
conditions. More generally, running “one factor at a time”
investigations would not detect such interactions.
Figure 7 shows the dependence of cf(Pd) on the two most
important parameters of this study: co(Pd) and ME, when time
and temperature are set respectively at 60 °C and 100 min. The
red area of the graph is where treatment with SPM32 leads to
poor results: cf(Pd) remains over the 50 mg/kg threshold,
whereas in the white area, cf(Pd) is below 50 mg/kg.
Considered in isolation, it shows an expected resultit is
possible to reduce the residual Pd to below 50 mg/kg in a low
concentration solution (co(Pd) = 100 mg/kg) with small
quantities of scavenger (0.6 ME). For more concentrated
solutions, higher ME of scavenger are required, until for co(Pd)
> 350 mg/kg it becomes impossible to reduce Pd residuals to
below 50 mg/kg (at this reaction time and temperature and
within the parameter ranges studied here).
Figure 8 is a plot of the same parameters as Figure 7, but at a
higher temperature and longer reaction time. An interesting
effect is observed when Figure 7 is compared with Figure 8.
Under the harsher scavenging conditions described in Figure
8, it becomes possible to reduce the Pd content of solutions of
the highest co(Pd) (500 mg/kg) down to below 50 mg/kg (top
right corner). Under these conditions, however, the quantity of
scavenger needed to obtain the threshold cf(Pd) of 50 mg/kg is
increased to approximately 1.2 ME (bottom left corner of the
graph). This is a significant increase in ME compared with
Figure 7, where time and temperature were lower.
Table 7. Results of the Small Scale Verification Experiments
result:
cf(Pd)
(mg/kg)
expected cf(Pd) value
according to model
(mg/kg)
temp. time
co(Pd)
(°C) (min) (mg/kg) ME
60
60
100
100
196.4
291.5
2.61
2.80
32.1
39.8
24.1
45.5
A number of chemical explanations can be proposed for this
phenomenon. It is not the purpose of this paper to prove or
disprove them, but some suggestions can be submitted for
consideration. It is possible, for example, that a higher
temperature leads to a change in the speciation of the dissolved
Pd, resulting in the formation of a species that is more difficult
to scavenge. Another possible explanation could be that
prolonged contact time between the product solution and the
scavenger could lead to another dissolved molecule (e.g., the
product or a byproduct) coordinating to the scavenger, thus
using up active sites that then are unavailable for Pd chelation.
At 95% confidence, each verification experiment leads to
results that fully fit the range at stake. The values lie within the
2σ range, deemed highly appropriate for this study.
2.5.b. Impact of a Scale Increase. The validity of the model
to a moderate scale-up was tested by increasing the scale to 100
g of product solution (5 times larger).
For these larger scale experiments, an alternative apparatus
setup was used in order to better represent what could be
encountered in an industrial application. A pitch-bladed
overhead stirrer (blade diameter: 4.0 cm, blade height: 1.2
633
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
Table 8. Scale-up Verification Experiment for 200 mg/kg a Product Solution
temp.
(oC)
time
(min)
co(Pd)
(mg/kg)
result: cf(Pd)
(mg/kg)
expected cf(Pd) value according to model
ME
(mg/kg)
Experiment 1 (analysis repeated
twice)
60
100
193.1
2.60
2.60
2.60
29.1
29.3
31.9
23.4
23.4
23.3
Experiment 2 (analysis repeated
twice)
2.60
31.6
30.5
1.5
23.3
Mean
Std. dev. σ
Table 9. Scale-up Verification Experiment for a 300 mg/kg Product Solution
temp.
(°C)
time
(min)
co(Pd)
(mg/kg)
result: cf(Pd)
(mg/kg)
expected cf(Pd) value according to model
ME
(mg/kg)
experiment 1 (analysis repeated
twice)
60
100
291.5
2.80
2.80
2.80
41.2
39.6
40.0
45.5
45.5
45.6
experiment 2 (analysis repeated
twice)
2.80
39.8
40.2
0.7
45.6
Mean
Std. dev. σ
cm) was used instead of the shaker stage used at small scale.
The agitation speed was set to 300 rpm, and the reaction vessel
was a flanged round-bottomed glass flask (diameter 85 cm). At
this agitation speed, a good suspension of the SPM32 scavenger
in the product solution was observed. The scavenging operation
was heated using a heating mantle equipped with a feedback
temperature probe.
To increase the reliability and the reproducibility of the
analytical data, each experiment was repeated twice, and for
each repeat, the analysis was also repeated. The results are
presented in Table 8 and Table 9.
These results are very similar for both the 20 g and the 100 g
scales, and the model was verified successfully at 95%
confidence. It therefore appears that the moderate scale
increase, including the change of conditions (from a vial to a
round-bottom flask, and from shaking to overhead stirring) has
only a marginal influence on the reaction, and therefore does
not significantly impact on the accuracy of the predictions made
by the model.
which can generate a good amount of process knowledge with
significantly fewer experiments.
While there are many parameters that can be varied to
improve the effectiveness of Pd removal from a pharmaceutical
intermediate, an understanding of the importance of those
parameters can lead to additional process benefits, such as the
removal of Pd to lower levels or a more economic use of the
metal scavenger. On scale-up, the latter benefit in particular has
key importance as chemists try to develop economic
manufacturing processes.
A systematic study is also able to highlight effects that would
have been otherwise missed, or wrongly attributed. A rise in
temperature, typically, can be very beneficial under certain
conditions, but this study suggests that the same parameter can
become detrimental under other sets of conditions. Equally, an
increase in ME of scavenger might have less of an impact under
some conditions compared to others.
The validity of this model to a moderate increase in scale was
also tested, with modifications including not only a change in
scale, but also an alteration of the stirring method (pitched
blade overhead stirrer) to better reflect equipment used
industrially. The model was found to remain valid, with only
minor changes in cf(Pd) observed compared to small scale
experiments.
These characteristics not only are dependent upon the
reaction used, the nature of the metal, and the product solution,
but also change considerably from one scavenger to another. It
was not the intention of this study to propose a set of general
trends, applicable to any scavenging operation. It is hoped,
however, to present chemists with an efficient method,
adaptable to any scavenging operation, which allows an
increased understanding and prediction of the behaviour of
one given scavenging operation, aimed towards minimizing cost
and time at scale. Using such a mathematical method, a process
chemist would have the capacity to quickly adapt to changes in
the process, with no additional experiments after the initial
matrix other than one or two verification experiments.
In the example presented here, the original concentration
co(Pd) of Pd in the Buchwald-Hartwig product solution was
shown to have the most statistically significant effect on
For meaningful scale-up predictions, however, further scale-
up investigations remain necessary.
CONCLUSIONS
■
A method has been presented which is aimed at understanding
the important parameters in a metal scavenging operation by
the study of a Pd-catalyzed Buchwald-Hartwig reaction.
Metal scavenging is a complex multivariate process; suitable
multivariate analysis, such as statistical design of experiment,
can be a powerful tool to quickly generate such process
understanding. To the best of our knowledge, this is the first
factorial DoE study on a metal scavenging process, and the
benefits of the method are clearly illustrated. Optimized
conditions under which to scavenge Pd from an API model
solution can easily be identified.
A matrix of 29 experiments might be considered to be a
significant commitment of resources, which, while beneficial in
terms of developing a process model, may not appear attractive
to the process chemist. However, chemists should consider
screening designs and fractional factorial screening designs,
634
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635

Organic Process Research & Development
Article
(18) Grosjean, C.; Henderson, A. P.; Her
Knowles, J. P.; Whiting, A.; Wright, A. R. Org. Process Res. Dev. 2009,
13, 434.
(19) Salome, C.; Salome-Grosjean, E.; Park, K. D.; Morieux, P.;
Swendiman, R.; DeMarco, E.; Stables, J. P.; Kohn, H. J. Med. Chem.
2010, 53, 1288.
(20) Brown, J.; Chighine, A.; Colucci, M. A.; Galaffu, N.; Hirst, S. C.;
Seymour, H. M.; Shiers, J. J.; Wilkes, R. D.; Williams, J. G.; Wilson, J.
R. H. Tetrahedron Lett. 2008, 49, 4968.
(21) Jal, P. K.; Patel, S.; Mishra, B. K. Talanta 2004, 62, 1005.
(22) Girgis, M. J.; Kuczynski, L. E.; Berberena, S. M.; Boyd, C. A.;
Kubinski, P. L.; Scherholz, M. L.; Drinkwater, D. E.; Shen, X.; Babiak,
S.; Lefebvre, B. G. Org. Process Res. Dev. 2008, 12, 1209.
(23) The possibility of including several scavengers in this study was
considered, but was found to be impractical. The fact that there is no
continuity in the nature of scavengers makes them difficult to study by
DoE. For a continuously defined numerical parameter (e.g., temper-
ature), one can easily define upper and lower limits for the design
space (e.g., 30 to 80 °C) and a midpoint (e.g., 55 °C). There is no easy
way to do this for a categorical parameter such as ″nature of
scavenger”. Ways around this problem can be found by sorting
scavengers based on one of their characteristics, such as the empirically
measured efficiency of the scavenger, for example. However, this
method introduces a source of potentially large error as this
characteristic is only measured under one set of conditions, and is
assumed to remain unchanged when conditions vary across the design
space. Another drawback is that it significantly increases the number of
experiments required to study a system.
́
ault, D.; Ilyashenko, G.;
scavenging efficiency, closely followed by the molar equivalents
of SPM32 scavenger used and an interaction of these two
factors.
ASSOCIATED CONTENT
■
S
* Supporting Information
Synthesis of the Buchwald-Hartwig stream. Analysis of the
functional group loading (FGL) of PhosphonicS SPM32
scavenger. Typical procedure for an experiment in the
experimental matrix. ICP-OES analysis of the Pd content in
samples. Design and statistical analysis of the DoE experiment.
Screening of Pd-containing solutions for good Pd retention.
This material is available free of charge via the Internet at
http://pubs.acs.org.
AUTHOR INFORMATION
■
Corresponding Author
*E-mail: jan.recho@phosphonics.com.
Notes
The authors declare no competing financial interest.
ACKNOWLEDGMENTS
■
We would like to thank Dr. Andrei A. Zlota of The Zlota Co.
LLC for fruitful discussion and guidance regarding the setup of
the DoE experiments described in this paper and the
interpretation of their results.
(24) (a) Salome, C.; Salome-Grosjean, E.; Stables, J. P.; Kohn, H. J.
Med. Chem. 2012, 53, 3756. (b) Kohn, H. L.; Salome, C. WO 2010/
148300, 2010.
(25) (a) Ford, R.; Mather, A.; Mete, A.; Wiley, K.; Bull, R. J. US
2011/0190309, 2011. (b) Charrier, J. D.; MacCormick, S.; Storck, P.-
H.; Pinder, J.; O’Donnel, M. E.; Knegtel, R. M. A.; Young, S. C. Y.;
Kay, D.; Reaper, P. M.; Durrant, S. J.; Twin, H. C.; Davis, C. J. WO
2012/138938, 2012. (c) McGee, P.; Garnett, I.; Juan, E.; Manage, A.;
Carniaux, J.-F. US 2012/0165520, 2012.
(26) http://www.phosphonics.com/.
(27) Mondal, B.; Wilkes, R. D.; Percy, J. M.; Tuttle, T.; Black, R. J.
G.; North, C. Dalton Trans. 2014, 43, 469.
(28) Narendar, N.; Velmathi, S. Tetrahedron Lett. 2009, 50, 5159.
(29) Wang, L.; Zhang, J.; Zhao, R.; Li, Y.; Li, C.; Zhang, C. Bioresour.
Technol. 2010, 101, 5808.
(30) Dinu, M. V.; Dragan, E. S. Chem. Eng. J. 2010, 160, 157.
(31) Boparai, H. K.; Joseph, M.; O’Carroll, D. M. J. Hazard. Mater.
2011, 186, 458.
(32) Owen, M. R.; Luscombe, C.; Lai, L.-W.; Godbert, S.; Crookes,
D. L.; Emiabata-Smith, D. Org. Process Res. Dev. 2001, 5, 308.
(33) Hersmis, M. C.; Spiering, A. J. H.; Waterval, R. J. M.; Meuldijk,
J.; Vekemans, J. A. J. M.; Hulshof, L. A. Org. Process Res. Dev. 2001, 5,
54.
(34) Alimardanov, A. R.; Barrila, M. T.; Busch, F. R.; Carey, J. J.;
Couturier, M. A.; Cui, C. Org. Process Res. Dev. 2004, 8, 834.
(35) More information may be found at http://www.linkedin.com/
pub/andrei-zlota/3/1a4/8b4.
REFERENCES
■
(1) Magano, J.; Dunetz, J. R. Chem. Rev. 2011, 111, 2177.
(2) Magano, J.; Dunetz, J. R. Transition Metal-Catalyzed Couplings in
Process Chemistry: Case Studies from the Pharmaceutical Industry; Wiley-
VCH, 2013.
(3) Barbaras, D.; Brozio, J.; Johannsen, I.; Allmendinger, T. Org.
Process Res. Dev. 2009, 13, 1068.
(4) Bien, J. T.; Lane, G. C.; Oberholzer, M. R. Organometallics in
Process Chemistry; Topics in Organometallic Chemistry; Springer,
2004; Volume 6, p 263.
(5) Guideline on the specification limits for residues of metal catalysts of
metal reagents, EMEA, Doc. Ref. EMEA/CHMP/SWP/4446/2000,
2008.
(6) Guideline for elemental impurities, Q3D, ICH, Draft consensus
guide - step 2b version, 2013.
(7) Hermann, J. C.; Chen, Y.; Wartchow, C.; Menke, J.; Gao, L.;
Gleason, S. K.; Haynes, N. E.; Scott, N.; Petersen, A.; Gabriel, S.; Vu,
B.; George, K. M.; Narayanan, A.; Li, S. H.; Qian, H.; Beatini, N.; Niu,
L.; Gan, Q. F. ACS Med. Chem. Lett. 2013, 4, 197.
(8) Garrett, C. E.; Prasad, K. Adv. Synth. Catal. 2004, 346, 889.
(9) Manley, P. W.; Acemoglu, M.; Marterer, W.; Pachinger, W. Org.
Process Res. Dev. 2003, 7, 436.
(10) Prashad, M.; Hu, B.; Har, D.; Repic, O.; Blacklock, T. J.;
Acemoglu, M. Adv. Synth. Catal. 2001, 343, 461.
(11) Carrera, D. E.; PeiJue, S.; Safina, B. S.; Li, J.; Angelaud, R. Org.
Process Res. Dev. 2013, 17, 138.
(36) Fusion Product Development; S-Matrix Corporation; Eureka,
California, USA; http://smatrix.com.
(12) Yin, L.; Liebscher, J. Chem. Rev. 2007, 107, 133.
(13) Al-Hashimi, M.; Sullivan, A. C.; Wilson, J. R. H. J. Mol. Catal., A:
Chem. 2007, 273, 298.
(14) Al-Hashimi, M.; Qazia, A.; Sullivan, A. C.; Wilson, J. R. H. J.
Mol. Catal., A: Chem. 2007, 278, 160.
(15) Bullock, K. M.; Mitchell, M. B.; Toczko, J. F. Org. Process Res.
Dev. 2008, 12, 896.
(16) Reginato, G.; Sadler, P.; Wilkes, R. D. Org. Process Res. Dev.
2011, 15, 1396.
(17) Galaffu, N.; Man, S. P.; Wilkes, R. D.; Wilson, J. R. H. Org.
Process Res. Dev. 2007, 11, 406.
635
dx.doi.org/10.1021/op5000336 | Org. Process Res. Dev. 2014, 18, 626−635