Automated Serendipity with Self-Optimizing Continuous-Flow Reactors

Article abstract of DOI:10.1002/ejoc.201500980

A self-optimizing approach is used as a tool for targeting known and unknown materials in the continuous reaction of aniline, dimethyl carbonate (DMC) and tetrahydrofuran (THF) in supercritical CO₂ on γ-Al₂O₃. The study led to the formation of methylated anilines or carbamate derivatives and unusual addition products with THF including pyrrolidines and N-alkylated anilines. The identification of these products leads to the development of a plausible mechanism for the reactions. The system not only demonstrates a high flexibility that the self-optimization approach provides, including the ability to optimize for a variety of products by using a single catalyst, but also the ability to discover unexpected and original synthetic reactions. Self-optimizing continuous-flow reactors have been used to investigate the reaction of aniline with THF and DMC. High-selectivity profiles were obtained for the synthesis of methylated anilines or N-phenylpyrrolidines. By combining the three components in a single reaction system, a tandem alkylation reactivity leading to diversely alkylated anilines was discovered.

Full text of DOI:10.1002/ejoc.201500980

SHORT COMMUNICATION
DOI: 10.1002/ejoc.201500980
Automated Serendipity with Self-Optimizing Continuous-Flow Reactors
Zacharias Amara,*[a] Emilia S. Streng,[a] Ryan A. Skilton,[a] Jing Jin,[a]
Michael W. George,[a,b] and Martyn Poliakoff*[a]
Keywords: Self-optimization / Flow chemistry / Aniline / Alkylation / Acid-base catalysis / Supercritical fluids
A self-optimizing approach is used as a tool for targeting
known and unknown materials in the continuous reaction of
aniline, dimethyl carbonate (DMC) and tetrahydrofuran
(THF) in supercritical CO2 on γ-Al2O3. The study led to the
formation of methylated anilines or carbamate derivatives
and unusual addition products with THF including pyrrol-
idines and N-alkylated anilines. The identification of these
products leads to the development of a plausible mechanism
for the reactions. The system not only demonstrates a high
flexibility that the self-optimization approach provides, in-
cluding the ability to optimize for a variety of products by
using a single catalyst, but also the ability to discover unex-
pected and original synthetic reactions.
tion between these three compounds, possibly one of the
first reported examples of a new reaction being discovered
by flow chemistry.[3,6]
We have discussed in some details the different optimiza-
tion strategies,[4,5] and here we focus on their chemical con-
sequences. In general, we used the same range of conditions
Introduction
Currently, there is increasing interest in minimizing the
amount of waste generated in chemical processes.[1] One
source of such waste is the formation of byproducts.[2]
However, the time needed to characterize those byproducts
often discourages all but the most persistent chemists from
doing so. The aim of this Communication is to demonstrate
that recent advances in flow chemistry, particularly the de-
velopment of self-optimizing reactors, not only can acceler-
ate the identification of byproducts but also can lead to the
discovery of unexpected reactions.[3]
for all experiments and
a Super Modified Simplex
(SMSIM) algorithm to modify the reaction parameters.[7]
As SMSIM is a local algorithm, the final result might de-
pend on the starting conditions. We deliberately used errors
in sampling to make sure the optimization did not get stuck
in local minima.
Self-optimizing reactors combine a computer-controlled
continuous reactor with in-line analysis to adjust reaction
conditions automatically to control the outcome of the re-
action,[4] see Figure 1. We have previously demonstrated
that, for a reaction that can give rise to two different prod-
ucts, such a reactor can be optimized to maximize the yield
of either product.[4,5]
Here, we apply a self-optimizing approach to a more
complex system, the three-component reaction of aniline,
dimethyl carbonate (DMC) and tetrahydrofuran (THF), ca-
talysed by γ-alumina and show that a relatively complete
understanding of the various reactions can be obtained in
quite a short time by systematically optimizing the yields
of the different products and byproducts. In particular, we
describe what appears to be a previously unreported reac-
Results and Discussion
The initial rationale for studying the reaction of aniline
with DMC was a modest extension of our previous work
on the catalytic etherification of alcohols in supercritical
CO2.[4a,5,10] However, an immediate problem arose; there
were frequent blockages in the pipework. Therefore, THF
was added as a co-solvent; such approach is well known to
increase the solubility of compounds in non-polar super-
critical fluids.[11] The consequence was that the GC analysis
of the unoptimized reaction by using the CO2/THF solvent
showed quite a complex mixture of products (see Fig-
ure 2a). Apparently, THF was acting as a reactant. There-
fore, for subsequent optimizations, THF was replaced with
toluene, which was expected to be inert under these reaction
conditions, and a high yield of the expected product, N,N-
dimethylaniline (1), could be obtained (see Figure 2c,
Table 1, and Scheme 1).
[a] School of Chemistry, University of Nottingham
University Park, NG7 2RD, United Kingdom
E-mail: zacharias.amara@nottingham.ac.uk
martyn.poliakoff@nottingham.ac.uk
http://www.nottingham.ac.uk/supercritical/beta/mp-entry.html
[b] Department of Chemical and Environmental Engineering,
University of Nottingham Ningbo China
Before starting the first series of optimizations, a tem-
perature-ramp experiment was performed in order to find
the lower limit at which the catalyst was starting to show
some activity. The higher temperature limits were defined
199 Talking East Road, Ningbo 315100, China
Supporting information for this article is available on the
WWW under http://dx.doi.org/10.1002/ejoc.201500980.
Eur. J. Org. Chem. 2015, 6141–6145
© 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
6141

Z. Amara, M. Poliakoff et al.
SHORT COMMUNICATION
Figure 1. Schematic representation of a self-optimizing continuous-flow reactor equipped with four HPLC pumps and a reactor (pre-
heater filled with sand and tubular reactor filled with catalyst). The output of the reactor is analysed by an in-line GC. The data from
the analysis are fed into a computer, which applies the Super Modified Simplex (SMSIM) algorithm[7] to calculate a new set of reaction
parameters (e.g. temperature, pressure, flow rates, etc.), which is expected to give a higher yield of the desired product.[8] The procedure
is repeated iteratively until a maximum yield is obtained.[9]
Figure 2. Optimization of the reaction of aniline with DMC in toluene. (a) GC trace of the initial product mixture (from aniline + DMC
+ THF); (b) trajectory of the optimization for 1 through parameter space (from aniline + DMC); (c) GC trace of the optimized product
mixture for 1 and for 2, respectively (with aniline + DMC); for conditions see Table 1.
Table 1. Conditions optimized for the highest yields of 1–5 and 7–9 for reactions between aniline and DMC and/or THF.
Product Selectivity
[%]
Conversion
[%]
Temperature Aniline[a] flow
DMC flow
[mLmin–1]
THF flow
[mLmin–1]
DMC/aniline THF/aniline
[°C]
[mLmin–1]
molar ratio
molar ratio
1[b]
2[b]
3[c]
Ͼ 95
83Ϯ8
70Ϯ10
45Ϯ5
Ͼ 99
30Ϯ5
20Ϯ5
6
100
85Ϯ10
50Ϯ10
100
100
85Ϯ5
100
345Ϯ5
345Ϯ5
183Ϯ3
318Ϯ5
275Ϯ15
200Ϯ10
235Ϯ5
196
0.17Ϯ0.03
0.25Ϯ0.05
0.2Ϯ0.1
0.16Ϯ0.05
0.15Ϯ0.1
0.2Ϯ0.1
0.17Ϯ0.08
0.16
0.5Ϯ0.1[a]
0.35Ϯ0.05[a]
0.9Ϯ0.1
0.50Ϯ0.05[a]
–
0.43Ϯ0.08
0.5Ϯ0.1
1.05
–
–
–
Ͼ 2.5
1.5–2.2
Ͼ 150
2.2–5
–
Ͼ 50
Ͼ 90
280
–
–
–
4[b,e]
5[d]
7[c]
–
–
0.3Ϯ0.1[a]
0.65Ϯ0.05
0.45Ϯ0.15
0.46
2.1–3.6
Ͼ 150
Ͼ 100
130
8[c]
9[c]
72
[a] 0.275 m solution in toluene. [b] 10 MPa, 1 mLmin–1 CO2. [c] 20 MPa, 1 mLmin–1 CO2. [d] 10–12 MPa, 1.1Ϯ0.1 mLmin–1 CO2. [e]
As a control, the reaction was repeated by using a niobium catalyst, Nb2O5·5H2O, in place of γ-Al2O3; 4 was observed as an additional
product (see Scheme 1); 4 was also produced in a separate experiment on Nb2O5 with a mixture of 1 + MeOH rather than DMC.[15]
either by the equipment limitations or by safety restrictions. get product could be detected and contained within the
This was followed by a catalyst stability test under pro- fixed boundaries (see Exp. Sect.).[5a]
longed reaction time and at fixed conditions. The optimiza-
The highest yields of 1 were obtained with an excess of
tion was then typically started at conditions where the tar- DMC at 345 °C (Table 1). At that temperature, DMC is
6142
www.eurjoc.org
© 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Eur. J. Org. Chem. 2015, 6141–6145

Automated Serendipity with Self-Optimizing Continuous-Flow Reactors
out to be too reactive for isolation from our reaction mix-
tures (Scheme 3). Thus, attempts were made to trap 6 in the
presence of DMC. GC analysis of the product mixture from
the multicomponent system of aniline, THF and DMC
indicated the presence of several additional products in
small amounts. We therefore carried out optimization ex-
periments for each of these, which led to the isolation and
characterization of 7 (30% selectivity), 8 (20% selectivity)
and 9 (6% selectivity). None of these compounds have been
previously reported. They were separated from the reactor
mixtures by chromatography and characterized by NMR
spectroscopy (see Supporting Information). Their forma-
tion can be rationalized by an unusual sequence of nucleo-
philic substitutions (Scheme 3). Intermediate 6, obtained
from a first THF alkylation, is susceptible to both N- and
O-methylation to produce 7 or 8. The isolation of 7 suggests
that O-alkylation is kinetically favored. In addition, 7 can
react with an additional molecule of THF to produce 9. All
of these reactions are likely to be essentially concurrent and
happen under similar conditions. The main limitation in
achieving higher selectivity is the fact that ring closure is
always predominant and 5 is obtained as a major product.
Alkylation is also in competition with the methylation/
carbamoylation processes, and 1, 2 and 3 are observed as
well. Therefore, the optimization focuses on tuning the tem-
perature; even so, optimum conditions for the formation of
products 7–9 are similar (see Scheme 3).
Scheme 1. Catalytic reaction of aniline with DMC in toluene and
scCO2 to give 1–4.
likely to be decomposed into dimethyl ether (DME),[12]
which may well be the actual methylating agent. Conve-
niently, the flow rates of aniline and DMC and hence the
reagent stoichiometry could be manipulated to give a high
yield of N-methylaniline (2) at the same temperature (see
Table 1).[13] Compound 2 is relatively difficult to prepare
by more conventional routes, which usually involve toxic
reagents and generate significant amounts of waste.[14] Simi-
larly, a further optimization for a third reaction product,
namely carbamate 3, led to a 70% selectivity (Table 1). The
formation of 3 occurs at significantly lower temperature,
where DMC is not fully decomposed and the addition of
aniline to DMC is more likely to happen.
In most of the optimizations reported above, we have
used toluene as a co-solvent as this appeared to give the
cleanest route to the desired products, i.e. minimizing the
byproducts. With this in mind, we decided to reinvestigate
the effect of THF as a reagent on γ-Al2O3. To avoid poten-
tial blockages of the reactor, both aniline and THF were
used as solutions in toluene. Flowing aniline and THF
through the reactor without DMC led to the formation of
N-phenylpyrrolidine (5) (Scheme 2). Optimization gave a
Ͼ 98% yield of 5 at ratios of aniline/THF between 1:2.1
and 1:3.6 (see Table 1). Formation of 5 by reaction of anil-
ine and THF was first reported[16] in 1937, and it was then
studied in more detail in 1990 by Hargis et al.[17] who car-
ried the reaction out continuously in the gas phase on a
TiO2 catalyst at low pressure. To the best of our knowledge,
this is the first report of a continuous reaction of aniline
and THF at high pressure and in scCO2 providing a 10-fold
increase in productivity compared to the earlier studies.
Very recently during the preparation of this manuscript,
Korbad and Lee[18] have reported the formation of 5 with
long reaction times in toluene and with stoichiometric
amounts of the Lewis acid AlMe3. Therefore, the reaction
reported here has definite advantages in terms both of re-
duced waste and of higher productivity compared to pre-
vious routes to 5.
Scheme 3. Plausible pathway for the reactions of aniline (in tolu-
ene), DMC and THF in scCO2 yielding 5 and 7–9.
In addition it was found, somewhat surprisingly, that
when 2 was used as the starting material for the reaction
with THF, the major product observed was still 5 (75%)
together with 1 (up to 25%; Scheme 4). N-Demethylation
therefore seems to take place as an additional competitive
pathway under these conditions.[20]
Scheme 4. Proposed demethylation mechanism occurring during
the reaction of 2 with THF, yielding 5 and 1.
Scheme 2. Reaction of aniline with THF in toluene and scCO2 to
give 5.
The reactions were repeated with Me-THF as the start-
A plausible mechanism for the formation of 5 proceeds ing material and γ-Al2O3 as a catalyst yielding quantita-
via the intermediate[19] 6, which, even if it is formed, turned tively 2-methyl-1-phenylpyrrolidine (10) (Scheme 5). How-
Eur. J. Org. Chem. 2015, 6141–6145
© 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.eurjoc.org
6143

Z. Amara, M. Poliakoff et al.
SHORT COMMUNICATION
was controlled by a back-pressure regulator at the outlet of the
system.
ever, the range of optimized conditions were not as broad
as in the case of THF; instead the system located an opti-
mal point at 237 °C, 14.4 MPa, 1.16 mLmin–1 CO2 and a
Optimization for 1, 2 and 4: Ranges allowed: Organics 0.05–
2-MeTHF/aniline ratio of 4.2. Addition of DMC to the re- 2.0 mLmin–1 (0.275 m solutions in toluene); temperature 150–
350 °C; CO2 flow was set at 1 mL min–1 and pressure at 10 MPa.
action mixture with 2-MeTHF, enabled us to isolate and
characterize product 11. It is therefore likely that the attack
Optimization for 3 and 7–9: Ranges allowed: Organics 0.1–
2.0 mLmin–1 (aniline 0.275 m in toluene, THF and DMC pure);
occurs on the less substituted carbon atom, most probably
by an SN2-type mechanism. We also showed that 11 can be temperature 150–300 °C; CO2 flow was set at 1 mL min–1 and pres-
sure at 20 MPa.
an intermediate for the formation of 10, by pumping 11
into the reactor without other reagents and finding 10 as
the sole product.
Optimization for 5, 10 and 11: Ranges allowed: Organics 0.01–
2.0 mLmin–1 (0.275 m solutions in toluene); temperature 150–
300 °C; CO2 flow 0.5–2 mLmin–1; pressure 10–20 MPa.
Supporting Information (see footnote on the first page of this arti-
cle): Further details including operating procedures, analysis and
characterizations of all new compounds.
Acknowledgments
This work was supported by SYNFLOW (FP7/2007-2013), Com-
panhia Brasileira de Metalurgia e Mineração (CBMM) and SIN-
CHEM (FPA 2013-0037). M. W. G. thanks the Royal Society for a
Wolfson Merit Award. We are grateful to R. A. Bourne and X.
Han for their help. We thank M. Guyler, P. Fields, R. Wilson, K.
Hind, D. Litchfield and J. Warren for their technical support.
Scheme 5. Reaction of aniline with DMC and 2-MeTHF in toluene
and scCO2 to give 10 and 11.
[1] P. Anastas, N. Eghbali, Chem. Soc. Rev. 2010, 39, 301–312.
[2] R. A. Sheldon, Green Chem. 2007, 9, 1273–1283.
[3] a) J. Hartwig, J. B. Metternich, N. Nikbin, A. Kirschning, S. V.
Ley, Org. Biomol. Chem. 2014, 12, 3611–3615; b) D. N. Tran,
C. Battilocchio, S.-B. Lou, J. M. Hawkins, S. V. Ley, Chem. Sci.
2015, 6, 1120–1125.
[4] a) A. J. Parrott, R. A. Bourne, G. R. Akien, D. J. Irvine, M.
Poliakoff, Angew. Chem. Int. Ed. 2011, 50, 3788–3792; Angew.
Chem. 2011, 123, 3872; b) J. P. McMullen, M. T. Stone, S. L.
Buchwald, K. F. Jensen, Angew. Chem. Int. Ed. 2010, 49, 7076–
7080; Angew. Chem. 2010, 122, 7230; c) J. P. McMullen, K. F.
Jensen, Org. Process Res. Dev. 2010, 14, 1169–1176; d) V. Sans,
L. Porwol, V. Dragone, L. Cronin, Chem. Sci. 2015, 6, 1258–
1264.
[5] a) R. A. Bourne, R. A. Skilton, A. J. Parrott, D. J. Irvine, M.
Poliakoff, Org. Process Res. Dev. 2011, 15, 932–938; b) D. N.
Jumbam, R. A. Skilton, A. J. Parrott, R. A. Bourne, M. Poliak-
off, J. Flow Chem. 2012, 2, 24–27; c) R. A. Skilton, A. J. Par-
rott, M. W. George, M. Poliakoff, R. A. Bourne, Appl. Spec-
trosc. 2013, 67, 1127–1131.
Conclusions
Here we have shown that self-optimizing reactors enable
chemists to switch the selectivity to different products in a
chemically complex system by using γ-Al2O3 as a catalyst.
The technology is capable of optimizing the yield of as yet
uncharacterized products, ultimately leading to a better un-
derstanding of the chemistry. Hence it can be used for the
development of new chemical transformations, as demon-
strated here with the discovery of an original multi-compo-
nent reaction sequence.
Studying the reactivity and selectivity of organic reac-
tions can be complex and time-consuming, but it is a crucial
step in helping organic chemists to devise better processes.
It has been suggested[21] that developing new chemistry in
continuous systems is one of the major challenges in flow
chemistry. Here we have demonstrated that self-optimizing
reactors can provide an interesting and accelerated solution
to this problem.
[6] K. S. Elvira, X. C. i. Solvas, R. C. R. Wootton, A. J. deMello,
Nat. Chem. 2013, 5, 905–915.
[7] M. W. Routh, P. A. Swartz, M. B. Denton, Anal. Chem. 1977,
49, 1422–1428.
[8] K. Burgemeister, G. Francio, V. H. Gego, L. Greiner, H. Hugl,
W. Leitner, Chem. Eur. J. 2007, 13, 2798–2804.
[9] a) S. Krishnadasan, R. J. C. Brown, A. J. deMello, J. C. de-
Mello, Lab. Chip 2007, 7, 1434–1441; b) M. Rasheed, T. Wirth,
Angew. Chem. Int. Ed. 2011, 50, 357–358; Angew. Chem. 2011,
123, 374.
[10] a) P. N. Gooden, R. A. Bourne, A. J. Parrott, H. S. Bevinakatti,
D. J. Irvine, M. Poliakoff, Org. Process Res. Dev. 2010, 14, 411–
416; b) B. Walsh, J. R. Hyde, P. Licence, M. Poliakoff, Green
Chem. 2005, 7, 456–463; c) W. K. Gray, F. R. Smail, M. G.
Hitzler, S. K. Ross, M. Poliakoff, J. Am. Chem. Soc. 1999, 121,
10711–10718.
Experimental Section
CAUTION! The reactions described here involve high pressures
and require equipment with the appropriate pressure rating.
General: All experiments were carried out by using a self-optimiz-
ing reactor (Figure 1), which has been described in detail previous-
ly.[4a,5a] The conditions to produce the first simplex [(n + 1) vertices,
n variables] were determined by the operator. After that, the mea-
surement points were calculated by the SMSIM algorithm within
the allowed ranges. The result of the reaction was determined by
in-line GLC analysis (programme time 17 min) and the pressure
[11] M. A. McHugh, V. J. Krukonis, Supercritical Fluid Extraction:
Principles and Practice, Butterworth Publishers, Stoneham,
MA, 1986, p. 507.
[12] Y. Fu, H. Zhu, J. Shen, Thermochim. Acta 2005, 434, 88–92.
6144
www.eurjoc.org
© 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Eur. J. Org. Chem. 2015, 6141–6145

Automated Serendipity with Self-Optimizing Continuous-Flow Reactors
[13] The activity of fresh catalyst decreases during intial use but
stabilizes after ca. 2 h. Our optimizations were all carried out
with catalysts in this “aged” condition.
[14] S. Narayanan, K. Deshpande, Appl. Catal. A 2000, 199, 1–31.
[15] For further details see the Supporting information.
[16] Y. K. Yur’ev, G. A. Minkina, J. Gen. Chem. (USSR) 1937, 7,
5399–5401.
[19] a) J. M. Flaniken, C. J. Collins, M. Lanz, B. Singaram, Org.
Lett. 1999, 1, 799–801; b) H. F. Russell, J. B. Bremner, J. Bush-
elle-Edghill, M. R. Lewis, S. R. Thomas, F. Bates II, Tetrahe-
dron Lett. 2007, 48, 1637–1639.
[20] T. Rosenau, A. Hofinger, A. Potthast, P. Kosma, Org. Lett.
2004, 6, 541–544.
[21] S. V. Ley, SCI/RSC Symposium on Continuous Processing and
Flow Chemistry, Horsham, UK, Sept. 2013.
Received: July 24, 2015
[17] D. C. Hargis, R. L. Shubkin, Tetrahedron Lett. 1990, 31, 2991–
2994.
[18] B. L. Korbad, S.-H. Lee, Chem. Commun. 2014, 50, 8985–8988.
Published Online: August 25, 2015
Eur. J. Org. Chem. 2015, 6141–6145
© 2015 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.eurjoc.org
6145