Controlling the Course of a Two-Way Switchable Pd-Catalyzed Process by means of Empirical Multivariate Models

Article abstract of DOI:10.1002/cctc.201500234

A two-way switchable Pd-catalyzed process that can pursue two different mechanisms, namely hydro- and methoxy-deiodination was discovered, developed, and optimized by means of statistical experimental design, multivariate modelling, and response surface methodology. The investigation revealed that the two-way switchable process might be controlled either by ligand alone, or by ligand and other process variables in combination. The present study represents the first example of a catalytic system that provides either hydro- or methoxy-deiodination selectively by simply fine-tuning the experimental variables.

Full text of DOI:10.1002/cctc.201500234

DOI: 10.1002/cctc.201500234
Full Papers
Controlling the Course of a Two-Way Switchable Pd-
Catalyzed Process by means of Empirical Multivariate
Models
Alexander H. Sandtorv and Hans-RenØ Bjørsvik*[a]
A two-way switchable Pd-catalyzed process that can pursue
two different mechanisms, namely hydro- and methoxy-deiodi-
nation was discovered, developed, and optimized by means of
statistical experimental design, multivariate modelling, and re-
sponse surface methodology. The investigation revealed that
the two-way switchable process might be controlled either by
ligand alone, or by ligand and other process variables in com-
bination. The present study represents the first example of
a catalytic system that provides either hydro- or methoxy-deio-
dination selectively by simply fine-tuning the experimental var-
iables.
Introduction
Hydrodehalogenation is an essential transformation in organic
synthesis.[1] In this context, palladium has been shown to be
an efficient catalyst,[2] and several variations have been dis-
closed recently.[3] If an alcohol is present in the reaction mix-
ture, a competitive ipso-substitution-type dehalogenation pro-
cess can take place, a coupling between the aryl halide and
the alcohol to provide an alkoxydehalogenation product.[4] Di-
verse approaches have been disclosed for pursuing alkoxyde-
halogenation including treatment with alkoxide,[5] Cu-catalyzed
reactions,[6] Ullmann-type coupling reactions,[7,8] and Pd-cata-
lyzed coupling processes.[9,10]
Scheme 1. A general catalytic process proceeding through a hydrodeiodina-
tion (X=I) or a methoxydeiodination (R=H) mechanism. Step (a) oxidative
addition, step (b) coordination of the alcohol reagent to Pd, step (c) depro-
tonation by means of base B,[11] step (d) reductive elimination producing the
ArÀOR ether, step (e) b-hydride elimination, step (f) isomerization, and
step (g) reductive elimination affording the hydrodehalogenated product
ArÀH.
A general mechanism that includes both the hydrodeiodina-
tion and methoxydeiodination pathways is outlined in
Scheme 1. The b-hydride elimination step (e) determines the
main product distribution between alkyl aryl ether (1) or aryl
(2). The common strategy to control the selectivity of this step
has been ligand design and engineering with the aim of im-
proving the selectivity of the CÀO coupling.[4,10] Sophisticated
and dedicated catalytic systems have been synthesized for the
combined goal of high selectivity and yield.
Recently, we revealed a switchable Pd-catalyzed process that
could operate either as a Suzuki coupling reaction or as a hy-
drodeiodination reaction working on diiodoimidazoles in the
absence of a base.[13] The attained results and potential of this
process encouraged us to undertake the investigation dis-
closed herein, which includes the exploration of the general
validity and scope of the method, exploiting haloarenes in-
stead of diiodoimidazoles as substrates. Owing to the basic
chemical character, we realized that this investigation was
a multivariate challenge, which could only be solved by means
of statistical experimental design,[14] multivariate regression,[15]
and/or other methods for multivariate data analysis.[16]
In such cases, tuning of the process, or experimental vari-
ables has in general been treated very superficially or by the
obscure and incomplete OVAT (one-variable-at-a-time) strat-
egy[12] in attempts to optimize the process.
[a] Dr. A. H. Sandtorv, Prof. Dr. H.-R. Bjørsvik
Department of Chemistry
University of Bergen
AllØgaten 41, B-5007 Bergen (Norway)
Fax: (+47)55-58-94-90
E-mail: hans.bjorsvik@kj.uib.no
Results and Discussion
Supporting Information for this article is available on the WWW under
http://dx.doi.org/10.1002/cctc.201500234.
Pre-experimental design
This publication is part of a Special Issue on Palladium Catalysis. To view
the complete issue, visit:
http://onlinelibrary.wiley.com/doi/10.1002/cctc.v7.14/issuetoc
To obtain an outline of the present multivariate challenge,
a pre-experimental design study was performed. Each experi-
ChemCatChem 2015, 7, 2196 – 2205
2196
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
3a, a product that was not observed during our previous
study.[13] From this investigation, we identified KOH as the opti-
mal base.
The quantity of the base influenced the reaction per-
formance and KOH provided the highest yields of products 2a
and 3a (Figure 2a). Furthermore, we discovered that a reaction
Figure 1. Ishikawa cause-and-effect diagram—mapping of the experimental
variables for the two Pd-catalyzed processes hydrodeiodination and meth-
oxydeiodination.
mental variable that was identified was successively included
in an Ishikawa cause-and-effect diagram,[17] as shown in
Figure 1.
After a review of the process, we came to the conclusion
that ten different experimental variables could potentially influ-
ence both the performance of the reaction and the direction
of the two-way switchable process.
Six of these experimental variables (displayed in green in the
Ishikawa diagram of Figure 1) were chosen for further investi-
gation (see the Supporting Information for a discussion of
these variables). The other variables were kept at fixed, but
carefully designated experimental levels.
Introductory experiments
At the outset, 3-iodo-4-nitrotoluene (1a) was used as a model
Figure 2. a) Investigation of the effect of base on the ipso-type-substitution/
dehydrogenation processes. y1 =conversion based on starting material 1a,
y2 =selectivity towards the hydrodehalogenation product 2a, y3 =yield of
the hydrodehalogenation product 2a, y4 =selectivity towards the product
3a, y5 =yield of methoxydehalogenation product 2a. All responses were
measured by GC–MS. b) Influence of water as cosolvent in methanol as reac-
tion medium. General procedure: 3-nitro-4-iodotoluene 1a (0.19 mmol),
ligand L3 (0.15 mol%) and KOH [for (a) as described and for (b) 0.5 equiv.],
Degassed methanol (5 mL) and Pd(OAc)2 (0.15 mol%), 1008C for 90 min.
Analyzed by GC–MS. See the Supporting Information for details.
substrate.
A series of screening experiments (System I,
Scheme 2) were performed to investigate variations of the cat-
alytic system and the influence of various solvents. However,
a maximum yield of only ꢀ3% of the hydrodeiodinated prod-
uct 2a was obtained in this screening study.
Further screening included a base (System II, Scheme 2), and
revealed that the presence of a base was essential for the hy-
drodeiodination reaction (see the Supporting Information for
details). We also obtained the methoxydeiodinated compound
medium composed of methanol, with or without water as a co-
solvent, demonstrated a substantial effect on the reaction out-
come; leading to either 2a or 3a (see Figure 2b). Increased
quantities of water in the reaction medium completely sup-
pressed the methoxydeiodination reaction.
We also examined the manner in which the base should be
added to the reaction mixture. Addition of KOH as a solid or in
solution resulted in a noticeable effect on the reaction yield.
The reaction profiles of the hydrodeiodination and methoxy-
deiodination under these different addition protocols are
shown in Figure 3. The reaction profiles clearly show that the
methoxydeiodination product 3b degrades when the reaction
time is longer than 100 min. Furthermore, when KOH was
added as a solution, there was a noticeable increase in the
yield of both products compared to when KOH was added as
a solid. We attribute this effect to the fact that KOH is a highly
hygroscopic substance and the solid base had absorbed some
water.
Scheme 2. System I: Introductory screening experiments including variation
of solvent and Pd catalyst. System II: Screening of the solvents and bases on
the Pd-catalyzed ipso-substitution-type dehalogenation processes.
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2197
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
using a multivariate approach, namely by performing a princi-
pal component analysis[16] of the physical, chemical, and struc-
tural properties of a larger ligand library and selecting a subset
from this.[18] However, such an approach would have been la-
borious and far outside the scope of this study; therefore, we
decided to perform
a ligand library.
a simple experimental screening of
The deactivated iodoarene 1b (Table 1) and the activated io-
doarene 1c (Table 2) were used as model substrates.
Table 1. Investigation of the ipso-type-substitution deiodination reaction
with a deactivated arene.
Figure 3. Reaction profile for the hydrodeiodination/methoxydeiodination
reaction with KOH added either as a solid or as a solution.
Ligand screening
Response[a]
A ligand screening, including a test library composed of twelve
phosphorous ligands (Figure 4) with different steric and elec-
tronic properties, was investigated. An alternative and more
exhaustive ligand screening could have been carried out by
Entry
Ligand
Conv.
[%]
Sel. 2b
Sel. 3b
Sel. uni
[%]
[%]
[%]
1
2
3
4
5
6
7
8
L1
L2
L3
L4
L5
L6
L7
L8
42
35
86
100
33
69
25
100
27
39
28
54
65
32
100
72
41
55
87
60
78
55
55
57
46
35
68
0
15
59
45
10
40
11
45
45
20
–
–
–
–
13
–
–
3
–
11
–
9
L9
10
11
12
13
L10
L11
L12
no ligand
25
21
–
23
[a] Conv.=Conversion based on substrate 1b. Sel. 2b=Selectivity to-
wards product 2b. Sel. 3b=Selectivity towards product 3b. Sel. uni=Se-
lectivity towards an unidentified byproduct. Responses were measured
by GC.
The attempt to perform the hydrodeionation of the deacti-
vated arene 1b, provided the expected product 2b with selec-
tivity in the range of 41–100% by using the ligands L1–L12.
Furthermore, for a few of the ligands the methoxydeiodinated
product 3b was produced with selectivities in the range of
10–68% (Table 1). The activated arene 1c provided only small
quantities of the hydrodeiodinated product 2c. An Ullmann-
type reaction took place to provide biphenyl 4 (13–17%) when
the ligands L6 or L8 were used. These experiments show that
the electronic structure of the arene substrate was imperative
for the performance and success of the studied switchable
two-way process (hydrodeiodination and methoxydeiodina-
tion), and that arenes elaborated with electron-donating
groups were poor substrates.
Statistical experimental design[14] and modelling[15]
Based on the results from the experimental screening de-
scribed above, we decided to continue the investigation by
Figure 4. Phosphine ligands and Pd–NHC catalysts.
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2198
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
DFFD
zfflfflfflffl}|fflfflfflffl{
Table 2. Investigation of the ipso-type-substitution deiodination reaction
with an activated arene.
MFFD ¼ ½1 x1x2x3x4 x1x2x1x3x1x4x2x3x2x4 Á Á Á
x3x4x1x2x3x1x2x4x1x3x44x2x3x44x1x2x3x44Š
ð3Þ
Using the model matrix MFFD, one can estimate the model pa-
rameters that describe the correlation between the experimen-
tal variables and their various interaction terms and the out-
come of the two distinct products 2b and 3b: yjL =f(x1,…,x4),
j2[3, 5], L2[L3, L4, L6], where the model terms are outlined in
Equation (1). By means of gas chromatographic analyses, the
responses (y1,…,y5) were measured for each experiment. The
numerical values for the responses y3 and y5, yield of 2b and
3b, respectively, were multivariate correlated with the model
matrix MFFD of Equation (3). The numerical values of the regres-
sion coefficients were calculated (listed in Table 4) by a multiple
linear regression (MLR)[15] method by means of the computer
software SAS[21] and MATLAB.[22] The stem plot of Figure 5
shows the experiments that were carried out at low (À1),
center (0), and high (+1) settings of the experimental vari-
ables. It is evident from Figure 5 that the three distinct catalyt-
ic systems involving the ligands L3, L4, and L6 operated for
the hydrodeiodination. However, the catalytic system compris-
ing the L3 ligand displayed a pronounced property, it was ca-
Response[a]
Sel. 2c
Entry
Ligand
Conv.
[%]
Sel. 4
[%]
[%]
1
2
3
4
5
6
7
8
L1
L2
L3
L4
L5
L6
L7
L8
0
0
0
0
0
17
0
13
0
–
–
–
–
–
0
–
0
–
100
100
100
100
–
–
–
–
–
100
–
100
–
–
–
9
L9
10
11
12
13
L10
L11
L12
no ligand
5
20
22
26
–
–
[a] Conv.=Conversion based on substrate 1c. Sel. 2c=Selectivity to-
wards product 2c. Sel. 4=Selectivity towards product 4. Responses were
measured by GC.
using a small sub-set of the P ligands (L3, L4, and L6, Figure 2)
for further reaction discovery. In addition to the notable effect
of the catalytic system, it was evident that changing the exper-
imental conditions significantly affected both the performance
and outcome of the Pd-catalyzed processes. The investigation
was continued by using a full factorial design,[19] including
some experiments in the center of the experimental domain
(2k +c=24 +2=18). This design included the following experi-
mental variables: x1 =reaction temperature [8C], x2 =volume of
cosolvent [mL], x3 =volume of solvent [mL], and x4 =quantity
of base [equiv.]. The design matrix DFFD is provided in standard
order[20] (Table 3, entries 1–18), but was performed in random
order in the laboratory. To facilitate the calculation of the re-
gression coefficients, Equation (1), the experimental variables
x1,.., x4 were scaled according to Equation (2).
4
3
4
X
X X
y ¼ fðx1; x2; x3; x4Þ ¼ b0 þ
bixi þ
bijxixjþ
i¼1
i¼1 j¼2
ð1Þ
ð2Þ
2
3
4
X X X
bijkxixjxkþb1234x1x2x3x4;i < j < k
i¼1 j¼2 k¼3
Â
Ã
1
zi À zi;L þ 2 Â ðzi;H À zi;LÞ
zi;H À zi;L þ 2 Â ðzi;H À zi;LÞ
xi ¼
Â
à ; i ¼ 1; :::; 4
1
Figure 5. Screening of experimental conditions and three various ligands.
Each ligand is explored at three various experimental levels. The form of the
responses going from: “all at low level” via “all at center level”, to “all at
high level” indicates that some of the experimental variables influence the
response with quadratic terms. The experimental variables xk (definition
[unit]), levels [À1, 0, +1]: x1 (reaction temperature [8C]), [90, 100, 110], x2
(volume of cosolvent [mL]), [0.10, 1.00, 1.90], x3 (volume of solvent [mL]),
[2.00, 2.50, 3.00], x4 (quantity of base [equiv.]), [0.1, 0.3, 0.5].
in which zi is the corresponding experimental variables in real
units. zi,L and zi,H represent the selected low and high experi-
mental values expressed in real units of the experimental vari-
able i=1,…,4. On the basis of the design matrix DFFD, a model
matrix MFFD was created according to Equation (3).
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2199
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Table 3. Experimental results (responses) from a statistical experimental design involving the three ligands L3, L4, and L6 as ligands in the Pd-catalyzed
two-way switchable process that follows either a hydrodeiodination mechanism or a methoxydeiodination mechanism.
Entry
Experimental
variables[a]
Responses[b]
L3
y3
L4
y3
L6
y4
x1
À1
x2
x3
À1
x4
À1
y1
y2
y4
y5
y1
43
y2
63
y4
y5
y1
31
y2
69
y3
21
y5
1
2
3
4
5
6
6[c]
7
À1
À1
+1
+1
À1
À1
À1
+1
+1
À1
À1
+1
+1
À1
À1
+1
+1
0
25 100 25 100 24
27 37 16
31
0.1
10
0.1
0
0.1
18
3
–
9
0.1
11
21
0
9
11
50
0
20
50
39
–
+1
À1
+1
À1
+1
+1
À1
+1
À1
+1
À1
+1
À1
+1
À1
+1
0
À1
À1
À1
+1
+1
+1
+1
+1
À1
À1
À1
À1
+1
+1
+1
+1
0
À1
À1
À1
À1
À1
À1
À1
À1
+1
+1
+1
+1
+1
+1
+1
+1
0
83
3
43 36
58 49
100 100 100
46 96 44
100 100 100
47 83
100 100 100
0
4
0
0
2
0
8
0
–
6
0
0
0
0
0
0
0
0
0
0
0
–
–
–
–
–
–
–
–
100 100 100
100
100 100 100
0
24
0
11
13
0
0
7
0
9
11
0
100
3
76 23
100 28
89 71
87 73
5
5
0
0.1
30
28
80
84
7
78
17
54
3
22
18
47
9
52
83
79
33
77
83
65
74
69
38
88
39 17
27
100
–
34
97
–
9
97
–
66
0
–
3
–
–
33
–
82
–
100
83 65
39
100 41
100
38
100 18
95 45
100
39 22
78 65
79 62
96 32
83 64
88 73
83 54
85 63
83 57
100 38
93 82
7
27 18
100 100 100
14
34
5
66
8
9
17 13
0
0
0
0
0
0
0
0
0
0
0
–
–
–
–
–
–
–
–
100 100 100
0.1
0
10
0
0
5
0
8
0
0
0
81
0
11
36
0
30
11
73
0
42
0
69
0
0
15
0
21
0
16
37
10
11
10
–
–
–
–
–
–
–
–
100
58
0
31
100
68
0
66
82
80
–
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
81 100
0
3
8
0
62 14
100 100 100
0
5
0
0
2
0
4
0
0
0
41
33
88
62
49
–
41 100
33 100
88 100
62 100
49 100
–
–
–
–
–
–
–
–
22
9
37 100
50
30
61
49
–
–
–
–
–
–
–
34
18
20
–
–
–
–
–
–
–
22 18
21 17
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
À0.5
+0.5
0
0
0
0
0
0
4
1
–
–
–
–
–
–
–
–
17 13
12 10
11
11
17 12
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
–
À0.5
+0.5
0
0
17
15
À0.5
+0.5
0
0
0
0
À0.5
+0.5
0
7
0
6
0
–
–
[a] Experimental variables xk (Definition [unit]), levels [À1, 0, +1]: x1 (reaction temperature [8C]), [90, 100, 110], x2 (volume of co-solvent [mL]), [0.10, 1.00,
1.90], x3 (volume of solvent [mL]), [2.00, 2.50, 3.00], x4 (quantity of base [equiv.]), [0.1, 0.3, 0.5]. [b] y1 =conversion based on starting material 1b, y2 =selec-
tivity towards the hydrodeiodination product 2b, y3 =yield of the hydrodeiodination product 2b, y4 =selectivity towards the product 3b, y5 =yield of me-
thoxydeionination product. All responses were measured by means of GC-MS. [c] Replicate of experiment 6, the experiment that provided the highest out-
come in the factorial experimental design part. This experiment was repeated a third time, but the base was added as a solution with the identical quanti-
ty to provide quantitative yield of methoxydeiodinated product.
pable of facilitating both of the
two distinct mechanisms, hydro-
Table 4. Linear models describing the influence of the various experimental variables x1,…,x4 on the responses
deiodination and methoxydeio-
dination.
y3 and y5 studied with three various catalytic systems involving the ligands L3, L4, and L6.
L3
L3
L4
L4
L6
L6
y3
y5
y3
y5
y3
y5
The two empirical models
y3L3 =f(x1, x2, x3, x4) and y5L3 g(x1,
x2, x3, x4), outlined in Table 4
were used in an attempt to fur-
ther optimize the outcome of
the processes. Unfortunately,
the follow-up optimization ex-
periments failed. Figure 5 indi-
cates that the reason may be
due to non-modelled quadratic
effects.
b0
8.3333
4.0625
À2.4375
À3.6875
À4.0625
0.6875
À0.5625
À0.9375
3.1875
3.8125
2.0625
0.0625
0.6875
30.6111
14.0000
À9.0000
6.6250
À7.2500
À2.0000
3.6250
55.0556
33.7500
6.5000
À1.5000
À12.1250
1.7500
1.7778
30.9444
23.8750
1.2500
0.7500
À21.1250
À0.8750
À2.3750
À20.7500
2.5000
3.3750
2.6250
À3.6250
À3.7500
À3.5000
0.6250
13.9611
2.7688
b1
À2.0000
À1.0000
À0.2500
À2.0000
1.0000
0.2500
2.0000
0.7500
1.0000
b2
À5.3687
b3
3.7438
b4
5.1062
À0.2438
1.6187
b12
b13
b14
b23
b24
b34
b123
b124
b134
b234
b1234
À5.0000
À4.2500
0.8750
6.9813
1.6250
2.0000
5.8750
À0.8750
1.5000
2.3750
À5.6250
5.6250
À2.1250
À1.2438
À2.6312
1.2562
À1.3687
À2.2562
3.3812
À1.0062
À0.8813
0.2500
À2.3750
2.1250
À3.2500
À3.3750
À0.8750
À0.3750
0.2500
À0.7500
À1.0000
À0.2500
À0.7500
0.7500
À1.0625
À4.0625
À0.9375
To describe these non-linear
relationships, a more extensive
design is required. The experi-
mental DFFD (Table 3, entries 1–
18) was thus extended to incor-
porate the necessary experi-
ments (Table 3, entries 19–26).
À2.5000
R2
R2
RMSEP
RSD
0.896
0.117
3.245
3.442
0.745
À1.171
11.639
12.345
0.996
0.970
2.172
2.304
0.977
0.801
0.629
0.667
0.963
0.685
7.230
7.669
0.540
À2.912
10.952
11.616
Adj
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2200
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
The collection of the experiments (entries 1–26, which form
a new design matrix DCCD) constitute a central composite
design (CCD)[23] that also allows to estimate the model of Equa-
tion (4).
We decided to further investigate the CCD only for the cata-
lytic system involving the L3 ligand. By means of the scaled ex-
perimental variables x1,..,x4 the model matrix MCCD was created
according to Equation (5). Two empirical models were derived,
y3L3 =f(x1, x2, x3, x4) and y5L3 g(x1, x2, x3, x4), by using the multiple
linear regression (MLR) method. The estimated model parame-
ters are listed in Table 5. The subsequent step involved pruning
L3
L3
Table 5. Quadratic models describing y3 and y5
.
Full model
L3
L3
y3
y5
a0
10.7650
6.6085
0.5733
À0.6388
À0.3381
2.0600
0.8100
À1.6850
3.9400
3.6850
1.9350
À8.5068
5.4932
9.4932
À9.3561
0.731
60.1060
14.5450
a1
a2
À9.3030
a3
6.2424
a4
À6.2927
À2.0000
3.6250
À4.2500
1.6250
a12
a13
a14
a23
a24
a34
a11
a22
a33
a44
R2
0.2500
Figure 6. CND plots evaluating the full quadratic models of the a) hydro-
deiodination and b) methoxydeiodination pathways.
À2.3750
À36.307
25.693
11.693
vestigated experimental domain. Therefore, we used the
model to produce the response surface iso-contour projection
in the extrapolated region, which is the region defined by the
red lines in Figure 7, (the region above and at the right hand
side of the red drawn lines in Figure 8). To evaluate the predic-
tion for the extrapolated region, we conducted the experi-
ments indicated by the red circle in Figure 8, which predict
ypred ꢀ65% yield. The experiment conducted at the accompa-
nying experimental conditions (see Table 6) provided an
almost quantitative yield, which shows that the response sur-
face has a significantly steeper ascent in the extrapolated area.
À34.778
0.871
R2
0.389
0.706
Adj
RMSEP
RSD
4.254
4.428
10.542
10.973
of non-significant regression parameters, based on the cumula-
tive normal probability (CND)[24] plots of Figure 6. A new re-
gression was then performed by using this model form to
obtain the models listed Table 5.
4
3
4
4
X
X X
X
y ¼ fðx1; x2; x3; x4Þ ¼ b0 þ
bixi þ
bijxixj þ
bixi2
Scope of the switchable process
i¼1
i¼1 j¼2
i¼1
A thorough investigation of the scope of the switchable pro-
cess was performed by investigation of arenes with electron-
withdrawing groups. To confirm the findings from above, we
also studied some arenes bearing electron-donating groups
(Table 7). The hydrodeiodination process proceeded smoothly
with electron-withdrawing groups and the method exhibited
excellent functional-group tolerance. Carboxylic acid 2j was
hydrodeiodinated smoothly, whereas the reaction with the cor-
responding ortho-substituted carboxylic acid 1i was unsuccess-
ful. This may be due to formation of a 5-membered palladacy-
cle that includes the carboxylic group. The corresponding
ortho-substituted carbaldehyde 1m provided the methoxylat-
ed product 2m. Mixed haloarenes selectively provided the hy-
drodeiodination product in the presence of a bromine sub-
stituent (entry 6) and in one case chloroiodoarene 1g afforded
the hydrodihalogenation product 1h. Unfortunately, this was
ð4Þ
ð5Þ
DCCD
zfflfflfflffl}|fflfflfflffl{
MCCD ¼ ½1 x1x2x3x4 x1x2x1x3x1x4x2x3 Á Á Á
x2x4x3x4x1x1x2x2x3x3x4x4Š
The final models involving the L3 ligand (Table 5) were used to
produce response surfaces in terms of iso-contour projection
(Figure 7a,b and 8), which subsequently were used to predict
the optimized conditions of y3L3 and y5L3, respectively.
The derived model describing the methoxydehalogenation
(y5L3) is reasonable from a modelling point of view and could
be used to directly predict optimized conditions (optimized
conditions can be found within the boundaries of the experi-
mental domain). The model for hydrodeiodination, however,
suggested that a high yield might be attained outside the in-
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2201
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Figure 7. iso-Contour projections of the re-
sponse surfaces (contour lines shows the
yield) for the a) hydrodeiodination and b) me-
thoxydeiodination pathways of the two-way
switchable Pd-catalyzed process. Each path-
way is composed by a plot of a subplot col-
lection that is composed of a 5”5 matrix of
2D iso-contour maps. Each of the 2D plots
displays the volume of solvent [mL] (x3) as the
abscissa and the quantity of the base [equiv.]
(x4) as ordinate. The locations of the 2D plot
in the horizontal direction (the outer frame)
view the experimental variable volume of co-
solvent [mL] (x2) at five discrete experimental
levels, and in the vertical direction (also at
five discrete experimental levels) view the ex-
perimental variable x1, that is reaction tem-
perature [8C].
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2202
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
Table 7. Scope of the Pd-catalyzed process that follows either a hydro-
deiodination mechanism or a methoxydeiodination mechanism.
Figure 8. iso-Contour projections of the response surface (contour lines
show the yield) for the hydrodeiodination pathway of the Pd-catalyzed
switchable process. The model predicts higher reaction outcome when the
variables x1, x2, and x3 are all extrapolated outside the investigated region
(the red lines shows a small part of the experimental domain). The red circle
shows the tested experiment, predicted to be ypred =65%.
Table 6. Optimized conditions for the hydrodeiodination and methoxy-
deiodination reactions
Experimental value[a]
Pred.[b]
[%]
Expe.[b]
[%]
x1
x2
x3
x4
[c]
[c]
y3
y5
+2
+1
+2
À1
+2
+1
+1
À1
65
100
100
100
[a] Experimental variables xk (Definition [unit]), x1 (reaction temperature
[8C]), x2 (volume of co-solvent [mL]), x3 (volume of solvent [mL]), x4 (quan-
tity of base [equiv.]), [b] Pred.=predicted by the model and Expe.=ex-
perimental result obtained in experiment. [c] y3 =yield of the hydrodeio-
dination product 2b, y5 =yield of methoxydeionination product.
a special case, and further experiments showed that other
chloroarenes proceeded with only low yields.
[a] Yield based on GC. [b] Reaction time 180 min. [c] Yield based on
1H NMR spectroscopy and checked again authentic samples by GC–MS.
Conclusions
By means of empirical modelling and response surface meth-
odology we were able to establish, develop, and optimize
a palladium-catalyzed switchable process that can perform
either a hydrodeiodination or a methoxydeiodination reaction.
In contrast to commonplace investigations in catalyzed synthe-
sis, which involves the design of dedicated and sophisticated
catalysts to achieve high selectivities and yields, we have es-
tablished a method that involves fine-tuning of the experimen-
tal and process variables.
cal models that describe a two-way switchable process, which
allowed us to selectively pursue either hydrodeiodination or
methoxydeiodination. Both directions of the process afforded
good selectivity and high yields. The hydrodeiodination pro-
cess operated with electron-withdrawing groups and exhibited
excellent functional-group tolerance. The methoxydeiodination
process operated well with iodonitrobenzenes.
By means of a simple ligand screening we identified the
ligand L3 to be versatile and efficient. By using experimental
design and multivariate regression we have established empiri-
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2203
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
insoluble and no reaction took place. The results are displayed in
Figure 2b.
Experimental Section
General
All chemicals and reagents were purchased from Sigma–Aldrich,
Norway (ꢁ97%) and were used without further purification. Dis-
tilled and deionized water was used as a cosolvent and solvents
(Sigma–Aldrich) were used as reagent-grade quality. The obtained
reaction products were compared to authentic samples for identifi-
cation.
The effect of a solid base versus an aqueous solution of
base
A microwave reactor tube was charged with nitro-2-iodobenzene
(0.19 mmol), ligand
3 (5 mol%), TBAB (3 mol%), and KOH
(0.5 equiv.). KOH was added either as 1) solid, followed by water
(0.1 mL) and methanol (3 mL); or 2) as a solution followed by
methanol (3 mL). In both cases, a freshly prepared solution of
Pd(OAc)2 in methanol (0.15 mol%) was added through a syringe.
The reactor tube was placed in the microwave cavity and heated
at 110 8C for 90 min. The post-reaction mixture was passed through
a plug of silica and analyzed by GC–MS. The results are displayed
in Figure 3.
GC–MS analyses were performed on a capillary gas chromatograph
equipped with a fused silica column (25 m, 0.20 mm i.d., 0.33 mm
film thickness) at a helium pressure of 200 kPa, splitless/split injec-
tor and flame ionization detector. 1H NMR spectra were recorded
on a NMR spectrometer operating at 400 MHz. 13C NMR spectra
were recorded on a NMR spectrometer operating at 150 MHz.
Chemical shifts were referenced to the deuterated solvent used in
the experiment.
The microwave-assisted experiments were performed by using
a Biotage Initiator Sixty EXP Microwave System, which operates at
0–400 W at 2.45 GHz, in the temperature range of 40–2508C, and
a pressure range of 0–20 bar (2 MPa, 290 psi). The reactor-tube
volume was 2.5 mL.
The effect of the ligand
A microwave reactor tube was charged with either nitro-2-iodo-
benzene (0.19 mmol) or 2-iodoanisole (0.19 mmol), TBAB (3 mol%),
KOH (2 equiv.), and the appropriate ligand (5 mol%). Ligand series
are displayed in Figure 2. The reactor tube was sealed and flushed
with argon through the septa. A mixture of water and methanol
(1:4, 5 mL) was then added. A freshly prepared solution of
Pd(OAc)2 (0.15 mol%) in methanol was added through a syringe.
The tube was placed in the microwave cavity and heated at 1008C
for a period of 90 min. The post-reaction mixture was passed
through a plug of silica and analyzed by GC–MS. The results are
displayed in Table 1 and 2.
Experimental procedure for the screening of solvents, cata-
lysts, and ligands
A microwave reactor tube was charged with 3-nitro-4-iodotoluene
(0.19 mmol), catalyst, and ligand (0.15 mol%). The reactor tube was
then sealed and carefully flushed with argon through the septa.
Degassed solvent (5 mL) and the Pd catalyst was used, a fresh solu-
tion of Pd(OAc)2 (0.15 mol%) was added through a syringe. The re-
actor tube was placed in the microwave cavity and heated at
1008C for a period of 60 min. The post-reaction mixture was
passed through a plug of silica and analyzed by GC–MS. The yield
for each experiment can be found in Table 1 in the Supporting In-
formation.
Procedure for experiments utilized in the experimental
design
A microwave reactor tube was charged with nitro-2-iodobenzene
(0.19 mmol), TBAB (3 mol%), KOH (0.1–0.5 equiv.), and one of the li-
gands 3, 4, or 6 (5 mol%), before the reactor tube was sealed and
flushed with argon through the septa. Water (0.1–1.9 mL) and
methanol (2–3 mL) were added through a syringe, followed by
a freshly prepared solution of Pd(OAc)2 (0.15 mol%) in methanol.
The tube was placed in the microwave cavity and heated at 90–
110 8C, for 90 min. The post-reaction mixture was passed through
a plug of silica and analyzed by GC–MS. The design matrix contain-
ing the setting of each single experiment is displayed in Table 3.
Experimental procedure for base screening
A microwave reactor tube was charged with 3-nitro-4-iodotoluene
(0.19 mmol), XPhos-ligand (0.15 mol%) and the base (2 equiv.). The
tube was sealed and flushed with argon through the septa. De-
gassed solvent was then added (5 mL), followed by a freshly pre-
pared solution in the studied solvent of Pd(OAc)2 (0.15 mol%), by
means of a syringe. The reactor tube was placed in the microwave
cavity and heated at 1008C for 90 min. The post-reaction mixture
was passed through a plug of silica and analyzed by GC–MS. See
Table 2 in the Supporting Information for details.
Optimized procedure for the methoxydeiodination reaction
A microwave reactor tube was charged with nitro-2-iodobenzene
(0.19 mmol), ligand 3 (5 mol%), and TBAB (3 mol%). Fresh solu-
tions of KOH (0.1 equiv., 0.1 mL) in water and Pd(OAc)2
(0.15 mol%) in methanol were added through a syringe, followed
by methanol (3.0 mL). The tube was placed in the microwave
cavity and heated at 110 8C for 90 min. The post-reaction mixture
was passed through a plug of silica and analyzed by GC–MS, see
Table 6 for details.
The effect of a cosolvent
A microwave reactor tube was charged with 3-nitro-4-iodotoluene
(0.19 mmol), XPhos-ligand (0.15 mol%), TBAB (3 mol%), and KOH
(2 equiv.). The reactor tube was then sealed and flushed with
argon through the septa. A mixture of water and methanol (see
Figure 1b for details) was added, followed by a freshly prepared
solution of Pd(OAc)2 (0.15 mol%) in methanol through a syringe.
The tube was placed in the microwave cavity and heated at 1008C
for 90 min. The post-reaction mixture was passed through a plug
of silica and analyzed by GC–MS. For experiments where the
amount of water was higher than 70%, the starting material was
Optimized procedure for the hydrodeiodination reaction
A
microwave reactor tube was charged with haloarene
(0.19 mmol), ligand (5 mol%), KOH (0.5 equiv.), and TBAB
(3 mol%). Methanol (3.5 mL) was added through a syringe, fol-
3
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2204
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Full Papers
[15] a) N. R. Draper, H. Smith, Applied Regression Analysis, 3rd ed., Wiley, New
York, 1998; b) D. C. Montgomery, E. A. Peck, Introduction to linear regres-
sion analysis, Wiley, New York, 1982.
[16] a) I. T. Jolliffe, Principal Component Analysis, Springer Series on Statistics,
Springer, New York, 2002; b) E. R. Malinowski, Factor Analysis in Chemis-
try, 3rd ed., Wiley, New York, 2002, pp. 1–432.
lowed by fresh solution of KOH (0.5 equiv.) and a solution of
Pd(OAc)2 (0.15 mol%) in methanol. Finally water (2.8 mL) was
added. The tube was placed in the microwave cavity and heated
to 1208C for 90 min. The post-reaction mixture was passed
through a plug of silica and analyzed by GC–MS, see Table 6 for
details.
[17] K. Ishikawa, Guide to Quality Control, 2nd ed., Asian Productivity Organi-
zation, Tokyo, 1982.
[18] a) H.-R. Bjørsvik, U. M. Hansen, R. Carlson, Acta Chem. Scand. 1997, 51,
733; b) N. Fey, A. G. Orpen, J. N. Harvey, Coord. Chem. Rev. 2009, 253,
704–722.
Acknowledgements
[19] A full factorial design (FFD) is a statistical experimental design that is
constituted by all possible combinations of a set of experimental vari-
ables (factors). FFD is the most fool-proof design approach that sup-
ports both continuous variables (e.g. reaction temperature, concentra-
tion, etc.) and categorical variables (type of catalyst, base, etc.). In FFD,
an experimental run is conducted at every combination of the selected
experimental level. The number of experiments (in the design matrix D)
is determined by the number of experimental levels and variables. For
example, a full factorial design composed of three variables (k=3) that
is investigated at two experimental levels, requires 2k =23 =8 experi-
ments. In addition, some experiments (c) in the center of the experi-
mental design are included.
[20] Standard order of a design matrix D for a full factorial design with k
variables at two experimental levels is created in the following way.
The first (x1) column of the D matrix starts with À1 and alternates in
sign for all 2k runs. The second (x2) column starts with À1 repeated two
times, and then two repeats of +1 until all 2k places are filled for
column (x2). The third column (x3) starts with À1 that is repeated four
times, then four repeats of +1 and so on. General: xi, i=1,…,k (vari-
ables - columns) starts with 2iÀ1 repeats of À1 followed by 2iÀ1 repeats
of +1.
[21] For example, see: a) Base SAS 9.2 Procedures Guide, N. C. Cary, SAS Insti-
tute Inc, 2009; b) SAS/GRAPH 9.2: Statistical Graphics Procedures Guide,
2nd ed., N. C. Cary, SAS Institute Inc. 2010.
[22] For example, see: a) Using MATLAB, version 6, The Matwork, Inc., M. A.
Natick, 2000; b) Using MATLAB Graphics, version 6, The Matwork, Inc.,
M. A. Natick, 2000; c) D. Hanselman, B. Littlefield, Mastering MATLAB: A
Comprehensive Tutorial and Reference, Prentice-Hall Inc., Upper Saddle
River, NJ, 1996.
[23] A central composite design (CCD) is composed by a factorial design in-
cluding some center experiments and a group of star points (star
design). The composite design allows the calculation of model parame-
ters that can account for the curvature of a response surface. Although
the distance from the center (0) of the factorial design point is Æ1
(scaled units) for each variable (factor), the distance from the center to
A.H.S. gratefully acknowledges the Department of Chemistry at
the University of Bergen for funding his research fellowship.
Keywords: homogenous catalysis · multivariate regression
analysis
· palladium · reaction mechanisms · statistical
experimental design
[1] F. Alonso, I. P. Beletskaya, M. Yus, Chem. Rev. 2002, 102, 4009–4091.
[2] a) B. Wei, T. S. A. Hor, J. Mol. Catal. A 1998, 132, 223–229; b) J. Uenishi,
R. Kawahama, Y. Shiga, O. Yonemitsu, Tetrahedron Lett. 1996, 37, 6759–
6762; c) P. P. Cellier, J.-F. Spindler, M. Taillefer, H.-J. Cristau, Tetrahedron
Lett. 2003, 44, 7191–7195; d) C. Amorim, X. Wang, M. A. Keane, Chin. J.
Catal. 2011, 32, 746–755; e) T. Hennebel, J. Benner, P. Clauwaert, L. Van-
haecke, P. Aelterman, R. Callebaut, N. Boon, W. Verstraete, Biotechnol.
Lett. 2011, 33, 89–95; f) A. Pyo, S. Kim, M. R. Kumar, A. Byeun, M. S.
Eomb, M. S. Han, S. Lee, Tetrahedron Lett. 2013, 54, 5207–5210; g) G.
Chelucci, S. Figus, J. Mol. Catal. A 2014, 393, 191–209; h) G. Chelucci,
Chem. Commun. 2014, 50, 4069–4072.
[3] For example, see: a) C. A. Marques, M. Selva, P. Tundo, J. Org. Chem.
1993, 58, 5256–5260; b) G. V. Lowry, M. Reinhard, Environ. Sci. Technol.
1999, 33, 1905–1910; c) M. S. Viciu, G. A. Grasa, S. P. Nolan, Organome-
tallics 2001, 20, 3607–3612; d) J. Moon, S. Lee, J. Organomet. Chem.
2009, 694, 473–477; e) K. A. Cannon, M. E. Geuther, C. K. Kelly, S. Lin,
M. H. R. MacArthur, Organometallics 2011, 30, 4067–4073; f) H. Min, S.
Lee, M. Park, J. Hwang, H. M. Jung, S. Lee, J. Organomet. Chem. 2014,
755, 7–11; g) E. M. Zahran, N. M. Bedford, M. A. Nguyen, Y.-J. Chang,
B. S. Guiton, R. R. Naik, L. G. Bachas, M. R. Knecht, J. Am. Chem. Soc.
2014, 136, 32–35.
[4] a) G. Mann, J. F. Hartwig, J. Am. Chem. Soc. 1996, 118, 13109–13110;
b) A. V. Vorogushin, X. Huang, S. L. Buchwald, J. Am. Chem. Soc. 2005,
127, 8146–8149.
[5] a) J. A. Zoltewicz, T. M. Oestreich, A. A. Sale, J. Am. Chem. Soc. 1975, 97,
5889–5896; b) A. G. Osborne, L. A. D. Miller, J. Chem. Soc. Perkin Trans.
1 1993, 181–184.
[6] M. A. Keegstra, L. Brandsma, Recl. Trav. Chim. Pays-Bas 1991, 110, 299–
300.
[7] J. Lindley, Tetrahedron 1984, 40, 1433–1456.
a star point possess a value of jaj ¼
6
1. CCDs make use of in total five
experimental levels, namely Æ1, Æa, and 0 (scaled units). The CCD
used herein is composed by (2k +2k+c, k=4, c=2)=26 experiments.
[24] The cumulative normal probability (CND) plot is a graphical tool that
can be used to evaluate whether or not a data vector is nearly statisti-
cally normal distributed. The data vector are plotted against a theoreti-
cal normal distribution in such a way that the data points should form
an approximate straight line. Departures from this straight line indicate
departures from normality. Thus, when the vector of regression coeffi-
cients is evaluated in a CND plot, the data point that follows the
straight line will be non-significant coefficients (normal distributed
noise). The data point that falls off-line represent significant contribu-
ting models terms and should thus be included in the final model. See:
a) C. Daniel, Technometrics 1959, 1, 311 –341; b) A. Birnbauma, Techno-
metrics 1959, 1, 343–357.
[8] S. V. Ley, A. W. Thomas, Angew. Chem. Int. Ed. 2003, 42, 5400–5449;
Angew. Chem. 2003, 115, 5558–5607.
[9] a) K. E. Torraca, X. Huang, C. Parrish, S. L. Buchwald, J. Am. Chem. Soc.
2001, 123, 10770–10771; b) S. Kuwabe, K. E. Torraca, S. L. Buchwald, J.
Am. Chem. Soc. 2001, 123, 12202–12206.
[10] a) S. Gowrisankar, A. G. Sergeev, P. Anbarasan, A. Spannenberg, H. Neu-
mann, M. Beller, J. Am. Chem. Soc. 2010, 132, 11592–11598; b) S. Gowri-
sankar, H. Neumann, M. Beller, Chem. Eur. J. 2012, 18, 2498–2502.
[11] Z. Ahmadi, J. S. McIndoe, Chem. Commun. 2013, 49, 11488–11490.
[12] R. Carlson, J. Carlson, Design and optimizing in organic synthesis, Data-
handling in Science and Technology, Vol. 24, Elsevier, Amsterdam, 2005,
pp. 21–23.
[13] A. H. Sandtorv, H.-R. Bjørsvik, Adv. Synth. Catal. 2013, 355, 3231–3243.
[14] a) G. E. P. Box, W. G. Hunter, J. S. Hunter, Statistics for experimenters. An
introduction to design, data analysis, and model building, Wiley, New
York, 1978; b) G. E. P. Box, N. R. Draper, Empirical model-building and re-
sponse surfaces, Wiley, New York, 1987.
Received: March 5, 2015
Published online on May 26, 2015
ChemCatChem 2015, 7, 2196 – 2205
www.chemcatchem.org
2205
ꢀ 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim