Scale-up of a Vilsmeier formylation reaction: Use of HEL Auto-MATE and simulation techniques for rapid and safe transfer to pilot plant from laboratory

Article abstract of DOI:10.1021/op0155211

The application of reaction calorimetry and process modelling to allow for the rapid and safe scale-up of a Vilsmeier formylation reaction to the pilot plant will be described. This transformation was a key step in the preparation of the backbone amide linker (the so-called "BAL" handle) for solid-phase chemistry. In particular, use was made of Auto-MATE equipment from Hazard Evaluation Laboratories (HEL) and "Reaction Simulator" software to derive a thermokinetic model which allowed us to simulate heat-flow data on-scale. The model was then refined using a HEL SIMULAR 1-L calorimeter, and a direct comparison of the data showed there to be a 20% error in the enthalpy data gathered from the smaller Auto-MATE. The use of a preformed Vilsmeier reagent and dichloromethane as a reaction solvent gave a "square-wave" profile typical of a feed-controlled reaction. These conditions were successfully scaled to a 50-L pilot-plant vessel.

Full text of DOI:10.1021/op0155211

Organic Process Research & Development 2002, 6, 311−316
Scale-Up of a Vilsmeier Formylation Reaction: Use of HEL Auto-MATE and
Simulation Techniques for Rapid and Safe Transfer to Pilot Plant from
Laboratory
Ulrich C. Dyer, David A. Henderson, Mark B. Mitchell,* and Peter D. Tiffin
Chemical Synthesis Department, Roche DiscoVery Welwyn, Broadwater Road,
Welwyn Garden City, Hertfordshire, AL7 3AY, UK
Abstract:
Scheme 1
The application of reaction calorimetry and process modelling
to allow for the rapid and safe scale-up of a Vilsmeier
formylation reaction to the pilot plant will be described. This
transformation was a key step in the preparation of the
backbone amide linker (the so-called “BAL” handle) for solid-
phase chemistry. In particular, use was made of Auto-MATE
equipment from Hazard Evaluation Laboratories (HEL) and
“Reaction Simulator” software to derive a thermokinetic model
which allowed us to simulate heat-flow data on-scale. The model
was then refined using a HEL SIMULAR 1-L calorimeter, and
a direct comparison of the data showed there to be a 20% error
in the enthalpy data gathered from the smaller Auto-MATE.
The use of a preformed Vilsmeier reagent and dichloromethane
as a reaction solvent gave a “square-wave” profile typical of a
feed-controlled reaction. These conditions were successfully
scaled to a 50-L pilot-plant vessel.
Auto-MATE and an “in-house” software package2 for
delineating safe conditions for the pilot plant.
In looking to scale up the process rapidly we focused on
the ease of isolation of 3 from the crude mixture of aldehydes
generated in the Vilsmeier reaction and its separation by
recrystallisation.
Derivation of a Solvent-Based Procedure. In seeking
to control the heat of reaction we felt that a solvent-based
procedure would be advantageous over the neat process
described in the literature. Using Auto-MATE we simulta-
neously evaluated four reaction protocols under isothermal
conditions at -5 °C (Table 1).
Run 1 provided us with an evaluation of the neat literature
conditions and as expected showed heat accumulation. The
mixture also became quite viscous and stirred with difficulty.
In run 2 we explored the use of N,N-dimethylformamide as
Introduction
The so-called “BAL” (backbone amide linker) handle (1)
is a useful reagent for solid-phase chemistry and is widely
used within Roche both locally and globally. Small laboratory
batches had previously been synthesised within our depart-
1
ment using the literature procedure involving a hazardous
Vilsmeier reaction as the first stage. With a requirement for
larger amounts it was felt that to scale the synthesis to our
pilot plant some modifications were required.
In this report we discuss the use of the HEL Auto-MATE
reaction calorimeter and reaction simulation software to
rapidly address the problems we envisaged in the scale-up
of Stage 1 (Scheme 1) to a 50-L pilot-plant vessel.
(
2) The reactor model dynamically recalculates the overall heat-transfer
coefficient U through evaluation of reactor outside (h ) and inside (h ) film
coefficients coupled to a knowledge of the reactor thermal conductivity
k/l). In addition a correction term Ucorr is included to account for other
o
i
Discussion
(
General Overview. Our objective was to transfer this
chemistry from the laboratory to the pilot plant in an
expedient and safe manner. The volume requirements for
this material were likely to be very low, and we anticipated
the preparation of only one or two batches. Our primary goal
to make use of reaction calorimetry, in particular the HEL
extraneous factors (e.g., fouling). Ucorr is in turn evaluated from a simple
heating or cooling curve collected in the target reactor. The program
maintains a database of reactor properties, heat transfer fluid properties
and solvent properties. The major process solvent is generally chosen to
model the process; however, for mixed-solvent systems Wilson plots may
be needed to select the most appropriate solvent for modelling. The reaction
model can include up to five consecutive reactions with 26 independent
chemical species. Reactions are modelled using a standard power law
protocol. The powers correspond to the molecularity of the fundamental
reaction steps in the model. This is clearly distinct from the empirical order
of reaction, which generally relates to the rate-determining step (rds) of
the mechanism. The powers can be user-defined so that multistep processes
can, when appropriate, be truncated into one reaction. In this case the powers
would usually correspond to the rds for the truncated reaction sequence.
For more information about Reaction Simulator Software contact Dr. Mark
B. Mitchell by e-mail at Mark.Mitchell@Pfizer.com.
*
To whom correspondence should be sent. Present address: Pfizer Global
Research and Development-La Jolla, 10777 Science Center Drive, San Diego,
CA 92121-1111, U.S.A.
§
The initial spike in the power profile was an artifact due to imperfect PID
loop response.
(1) Albericio, F.; Kneib-Cordonier, N.; Biancalana, S.; Gera, L.; Masada, R.
I.; Hudson, D.; Baranay, G. J. Org. Chem. 1990, 55, 3730.
1
0.1021/op0155211 CCC: $22.00 © 2002 American Chemical Society
Vol. 6, No. 3, 2002 / Organic Process Research & Development
•
311
Published on Web 04/24/2002

Table 1: First Auto-MATE experiment assessing solvent
effect
addition
run
vessel contents
mixture
comments
accumulation,
viscous and unstirrable
accumulation,
viscous and unstirrable
1
2
3
4
ArOH/POCl
ArOH/DMF
ArOH/POCl
ArOH/POCl
3
DMF
POCl3
3
/DCM DMF in DCM accumulative,
controlable and stirrable
accumulative and a
trifluorotoluene two-phase mixture
3
/
DMF in
trifluorotoluene
reaction solvent, while run 3 used dichloromethane and
finally run 4 used R,R,R-trifluorotolune, proposed as a useful
alternative to DCM. No other solvents were considered at
this stage. From this study we felt that only DCM was
suitable as a candidate solvent for further scale-up. Under
the conditions evaluated in Table 1 we observed that the
exotherm was accumulative and not completely feed-
controlled as desired. It should be noted that the above
experiment involved the formation of the Vilsmeier inter-
mediate reagent “in situ”. We next sought to evaluate the
use of preformed reagent and the effects of temperature.
Refinement of the Dichloromethane Process. From this
experiment it was clear that the preferred conditions for scale-
Figure 1. Comparisons of calorimetric power profiles.
up involved the use of preformed Vilsmeier reagent (Table
2
2, run 3). From the “square wave” form of the power trace
(
Figure 1) it was clear that the reaction of the Vilsmeier
reagent with the phenol was fast and proceeded in an
essentially feed-controlled manner. By way of contrast the
rate of formation of the Vilsmeier reagent must be slow
compared to the addition rate to account for the induction
and accumulation observed when using “in situ” generated
reagent. The use of Auto-MATE allowed us to screen
reaction conditions and to select those most suitable for scale-
up in only seven experiments.
Figure 2. Process of kinetic data extraction.
Scheme 2
We now wished to use the data generated from the Auto-
MATE experiments to simulate temperature profiles on scale-
up using “Reaction Simulator” software. This, we hoped,
would further refine the process and generate useful informa-
tion for a HAZOP.
Reaction Modelling and Simulation. The objective for
reaction modelling was to derive kinetic data for the crucial
first stage (Vilsmeier reaction) of the process, as obtained
through fitting of empirical calorimetric profiles to a
proposed reaction model. Armed with both the kinetic and
thermal data, it would be possible to simulate the thermal
behaviour of the process upon scale-up to a pilot-plant
reactor.
aromatic substitution process rather than the initially formed
phosphorylated species (5) (Scheme 2).
It is generally accepted that the mechanism of Vilsmeier
formylation proceeds through the intermediacy3 of chloro-
iminium ion (6), which acts as the electrophile in the
,4
Kinetic data was extracted from the power data by use
of an in-house software package “Reaction Simulator”. The
process of kinetic data extraction is illustrated in Figure 2,
and effectively involves the systematic variation of trial
enthalpy and rate constant values to minimise the deviation
(
3) When many potential mechanisms exist each, should be modelled in turn
to see which fits the data best (see ref 10, for example). In this case the
mechanism of Vilsmeier formation is well accepted to proceed through
the indicated pathway. It was not deemed necessary to model other potential
routes, a decision which seems justified in the light of the comparable
activation data found from this work when compared to the literature
between the empirical and simulated power trace (q
r
). Note
in this example only q will be used in the so-called objective
r
function fitting; however, the process could also be per-
formed with, or in combination with, product distribution
(ref 7).
(4) Jugie, G.; Smith, J. A. S. J. Chem. Soc., Perkin Trans. II 1975, 925.
312
•
Vol. 6, No. 3, 2002 / Organic Process Research & Development

Table 2: Temperature dependence and effect of preforming the Vilsmeier reagent
temp
°C
vessel
contents
run
addition mixture
comments
1
2
3
-5
5
-5
ArOH/POCl
ArOH/POCl
ArOH/DCM
3
/DCM
/DCM
DMF/DCM
DMF/DCM
DMF/POCl /DCM
3
peak heat output at end of feed; exotherm accumulates; ∆H ) -122 kJ/mol
peak heat output at end of feed; exotherm accumulates; ∆H ) -115 kJ/mol
constant heat output;. feed-controlled; ∆H ) -113 kJ/mol
3
Figure 3. Simplex regression and comparison of experimental
and simulated power profiles for Reaction 3.
Figure 4. Simplex regression and comparison of experimental
and simulated power profiles for in situ at -5 °C allowing
and concentration data. For more complex chemistry models
the use of a multidata objective function is essential. This
approach to data extraction is also used in commercial
simulation packages such as Batchcad and ProSim. For the
case of “Reaction Simulator” the objective function was
optimised using a modified simplex algorithm.
The thermo-kinetic data for reaction 3 (Scheme 1) was
initially extracted, by simplex regression of initial trial data
against the power data for the preformed Vilsmeier reaction
1 2
derivation of enthapies for reactions 1 and 2 (∆H and ∆H ).
Activation enthalpy and entropy values could also be
obtained from an Eyring plot. The derived parameters are
presented in Table 3.
Using the rate data we were able to show significant
temperature dependence for the formation of the Vilsmeier
reagent. At low temperatures (below 15 °C) there is signifi-
cant accumulation. An increase in reaction temperature
results in an increase in rate of reaction, and at 25 °C
the process is essentially feed-controlled as illustrated in
Figure 6.
5
6
(Figure 1, yellow trace). Both k
3
3
and ∆H were variable in
the simplex, and the simplex progress, optimized values, and
fit comparisons are illustrated in Figure 3.
The process was now repeated for the in situ example at
Figure 7 illustrates that the rate-determining step for lower
temperatures is clearly that for the chloride displacement
reaction. The activation enthalpy and entropy for this rate-
limiting step (Table 1, reaction 2) is comparable to that
-
3 3
5 °C, keeping the above values for ∆H and k constant,
and allowing k , k , ∆H , and ∆H to be variable in the
1
2
1
2
regression. The results are depicted in Figure 4.
Finally, the analysis was performed for the in situ example
7
previously reported by use of a spectroscopic method to
at 5 °C, holding the obtained enthalpy values ∆H
and ∆H constant, whilst allowing k , k , and k
variable. The outcome can be found in Figure 5.
1
, ∆H
to be
2
,
q
-1
q
follow the overall reaction kinetics ∆H ) 66 kJ mol , ∆S
-1
-1
3
1
2
3
)
-128 J K mol )
8
Most importantly though an approximate rate constant
Approximate rate constants were now available for all
three steps in the reaction, and since rate constant data was
available at two temperature points, the Arrhenius terms
could be calculated using a standard logarithmic plot. This
would clearly only give a rough idea of the temperature
dependence since only two data points were available.
-1
-1
for reaction 3 (k ) 0.23 mol L sec ) was derived, and
3
(7) Alunni, S.; Linda, P.; Marino, G.; Santini, S.; Savelli, G. J. Chem. Soc.,
Perkin Trans. II 1972, 2070.
8) It has been commented that rate constants exceeding 0.1 mol L s-1 reflect
-1
(
feed-controlled reactions and, as such, should be more accurately described
as in excess of the limiting value where heat output rate becomes dominated
by factors other than kinetics. Whilst we agree in part with this view, it is
well established from the literature that rate constants significantly greater
than this can be derived using mathematical regression methods (see ref
10).
(
(
5) BATCHCAD is a registered trademark of GSE Systems Inc., U.S.A.
6) ProSim is a registered trademark of ProSim S.A., France.
Vol. 6, No. 3, 2002 / Organic Process Research & Development
•
313

Figure 5. Simplex regression and comparison of experimental and simulated power profiles for in situ at 5 °C, allowing derivation
all three rate constants (k , k , and k ).
1
2
3
Figure 6. Simulated power profiles of formation of Vilsmeier reagent at different temperatures.
Table 3: Arrhenius terms and activation ethalpies and
entropies
q
q
A
E
a
∆H
∆S
mol- L s-1
1
kJ mol
-1
kJ mol
-1
J K-1 mol-1
reaction
1
2
3
0.9179
5.74
54.55
11.83
3.39
52.20
9.48
-288
-142
-257
7
4.09 × 10
38.49
this would allow us to model the heat flow of this reaction
on-scale.
Calorimetric Studies in SIMULAR and Pilot Plant
Simulation. Having chosen to scale the route involving
preformation of the Vilsmeier reagent we wanted to char-
acterise the process in our 1l SIMULAR calorimeter. We
wished to primarily compare the quality of calorimetric data
obtained from our well established SIMULAR system with
data obtained from the small-scale Auto-MATE. The manu-
facturers (HEL) themselves acknowledge that Auto-MATE
is less accurate than SIMULAR and do not recommend trans-
Figure 7. Temperature dependence of rate of phosphorylation
and chloride displacement.
There was a difference between Auto-MATE and SIMU-
LAR collected data. For reaction of Vilsmeier reagent with
the phenol Auto-MATE appears to have over-estimated the
reaction enthalpy (-113 kJ/mol vs -93 kJ/mol) by around
9
fer of a process purely on the basis of Auto-MATE data.
The data for reaction of the preformed Vilsmeier reagent
with the phenol is presented in Figure 8.
2
0%, although the reaction profile was the expected “square
(
9) Simms, C.; Singh, J. Org. Process Res. DeV. 2000, 4, 554.
10) Bollyn, M.; Van den Bergh, A.; Wright, A. Chem.-Anlagen Verfahren
996, 29, 95.
wave” form. The reaction was simulated as before, and these
data are also presented in Figure 8.
(
1
314
•
Vol. 6, No. 3, 2002 / Organic Process Research & Development

Figure 10. Comparison of real and simulated temperature
profiles from Vilsmeier reaction in a 50-L pilot-plant vessel.
phenol, when performed under power-compensation mode,
in Auto-MATE in comparison to that for SIMULAR.
With the data generated we felt confident enough to safely
scale the process to our 50-L pilot-plant vessel. Once the
Vilsmeier reagent had been preformed in a separate vessel,
it was added to a solution of 3,5-dimethoxyphenol in dichlo-
romethane over approximately 45 min under isoperibolic
conditions. The temperature profile was taken from the vessel
chart recorder and compared to that predicted using the
SIMULAR model data (Figure 10).
Figure 8. SIMULAR experimental and simulated power and
energy profiles.
The MTSR for both the experimental and simulated data
was around 14 °C, and in that respect the data seemed to fit
very well. The temperature profiles, although, did not appear
to fit as well, with a slight skew of the experimental data
away from the simulated data. This was probably due to the
nature of the manually controlled feed on the pilot-plant
vessel, making it difficult to get precise control over the
addition rate. “Reaction Simulator” assumed a constant
addition rate.
Figure 9. Comparison of Auto-MATE and SIMULAR simu-
lated Vilsmeier reaction data under isoperiobolic conditions in
Conclusions
5
0-L Pfaudler vessel.
HEL Auto-MATE allowed us to rapidly screen reaction
conditions and to successfully select a suitable solvent.
Thermokinetic data was obtained from three parallel Auto-
MATE experiments. The so-obtained data allowed us to
derive rate constants for each step in the mechanism, and
these were then used to approximately simulate temperature
profiles and addition regimes on-scale for both the preforma-
tion of Vilsmeier reagent and its subsequent reaction with
3,5-dimethoxyphenol in DCM. The model derived for the
reaction was then refined in one SIMULAR experiment, and
a 20% error in the enthalpy calculation in Auto-MATE was
observed. This error did impact the extracted kinetic data;
however, it did not significantly affect the thermal modelling
of the process since the isoperibolic MTSR value differed
at most by only 2 °C. The process was successfully scaled
to a 50-L Pfaudler pilot-plant vessel, and the observed
experimental temperature profile showed a good correlation
with the final computer-simulated model.
With data available from both the Auto-MATE and
SIMULAR models for the Vilsmeier reaction we were able
to predict thermal behaviour of the process in a 50-L Pfaudler
glass-lined steel reactor. The “reaction simulator” enabled
us to generate a range of temperature and addition profiles,
but for brevity only that which was chosen following a
HAZOP discussion is shown.
The reaction would be performed under isoperibolic
(constant jacket temperature) conditions, and hence the
simulations were also performed for an isoperibolic reaction.
Figure 9 illustrates the difference between the Auto-MATE
and SIMULAR-based models.
The maximum temperature of synthetic reaction (MTSR)
is overestimated in the Auto-MATE model by approximately
2
°C. This discrepancy can be traced back to the 20% error
in the ∆H value for the addition of Vilsmeier reagent to the
Vol. 6, No. 3, 2002 / Organic Process Research & Development
•
315

Experimental Section
was filtered through a “Nutsche” pressure filter. The filter
cake was washed with water (3 × 5.0 L). The damp cake
was dried in a vacuum oven at 40 °C, to constant weight
(53 h) furnishing 4.92 kg of product. NMR analysis revealed
the material to contain ca. 60% desired para isomer.
This isomer mixture was charged to a 50-L glass-lined
vessel followed by addition of ethanol (39.0 L). The stirred
mixture was then heated to reflux and allowed to cool
naturally overnight. The suspension was cooled to ca. 10
°C and filtered. The filter cake was washed with cold ethanol
(2 × 4.9 L) and dried to constant weight (24 h) in a vacuum
oven at 30 °C, to give 2.72 kg product.
Preparation of 4-Formyl-3,5-dimethoxyphenol (3). 3,5-
Dimethoxyphenol (2) (4.86 kg) was charged to a 50-L glass-
lined vessel followed by dichloromethane (10.0 L). The
resulting endotherm lowered the reactor temperature to
around 4 °C. The solution was then stirred at ca. 5 °C for
about 3 h.
To a separate 50-L glass-lined vessel was charged
phosphorus oxychloride (7.24 kg) followed by dichlo-
romethane (10.0 L). The solution was cooled to ca. 10 °C,
and a solution of DMF (2.69 L) in dichloromethane (10.0
L) was added over 45 min, maintaining the reactor temper-
ature between 9 and 16 °C. When the addition was complete,
the mixture was allowed to stir for 30 min.
NMR showed that the material still contained ca. 7%
unwanted ortho isomer.
The Vilsmeier solution was then added gradually to the
phenol solution over 40 min, maintaining the reactor tem-
perature between 1 and 15° C. The reaction mixture was
allowed to stir for 1.5 h at 5 °C before being quenched by
running slowly into precooled R.O. Water (50 L) in a glass-
lined 250-L vessel over 35 min. The reactor temperature was
kept between 5 and 17 °C. Dichloromethane (20 L) was
added, and the layers were separated. The upper aqueous
layer was then re-extracted with dichloromethane (25 L). The
upper aqueous layer was neutralized with 50% aqueous
potassium hydroxide (13.5 L) to pH 6, and the resultant slurry
The crude product was slurried in ethanol (10.0 L) in a
20-L flask for 2 h and filtered. The filter cake was washed
with ethanol and dried in vacuo overnight to give 4-formyl-
3,5-dimethoxyphenol (1.81 kg) as a white solid.
Acknowledgment
We thank David Bilcock, Roy Lawrence, and Graham
Cole for assistance with the pilot-plant work.
Received for review December 10, 2001.
OP0155211
316
•
Vol. 6, No. 3, 2002 / Organic Process Research & Development