Continuous Consecutive Reactions with Inter‐Reaction Solvent Exchange by Membrane Separation

Abstract Pharmaceutical production typically involves multiple reaction steps with separations between successive reactions. Two processes which complicate the transition from batch to continuous operation in multistep synthesis are solvent exchange (especially high‐boiling‐ to low‐boiling‐point solvent), and catalyst separation. Demonstrated here is membrane separation as an enabling platform for undertaking these processes during continuous operation. Two consecutive reactions are performed in different solvents, with catalyst separation and inter‐reaction solvent exchange achieved by continuous flow membrane units. A Heck coupling reaction is performed in N,N‐dimethylformamide (DMF) in a continuous membrane reactor which retains the catalyst. The Heck reaction product undergoes solvent exchange in a counter‐current membrane system where DMF is continuously replaced by ethanol. After exchange the product dissolved in ethanol passes through a column packed with an iron catalyst, and undergoes reduction (>99 % yield).

constant for each catalyst loading was determined using the isolation method.
Experiments at each catalyst loading were performed in duplicate and the average value of the kinetic constant was used. An exponential curve fitting was applied in order to describe kinetic constant as a function of the catalyst concentration. This equation was further used in a mathematical model to describe the PFR-m-CSTR performance. Conversion over time at different catalyst loadings and the fitted curve for kinetic constant as a function of catalyst concentration is presented in Figure S1. The rate of the Heck reaction is somewhat low compared to some reaction times in the literature within the minute range [3] . This work replicated the conditions developed by Caron et al [2] where the batch process time is 10 hours, suggesting the slow reaction confirmed by the reaction kinetics in Figure S1. Generally bromobenzenes have relatively low reactivity and require higher temperatures and longer residence times than commonly reported alternatives such as iodobenzenes. For example a study on Heck coupling of aryl halides to alkenes under segmented flow conditions has shown that the best conversion achieved at 40 min residence time and using 10 mol% catalyst at 130 0 C was 65% for p-bromonitrobenzene, while for most bromobenzenes the conversion was within the range of 20-30% under the same conditions [4] . The conversion achieved at 90 0 C in our continuous reactor is >95% which required longer residence times. •

Reactor set-up -PFR-m-CSTR
The set-up consisted of two reactors in series. A constant flow of the feed solution was first passed through a U-shaped PFR, placed in a heating chamber maintained at 90 0 C. The PFR was made of 316 SS ½'' tube with a length of 0.64 m (total volume of 60 mL). The outlet of the PFR was directly connected to the inlet of the m-CSTR.
The m-CSTR [5] was made of 316 SS, could operate under high pressure (69 bar), and hold circular flat sheet membranes with an effective area of 51 cm 2 . The m-CSTR was operated in a bottom-to-top permeation mode and contained a magnetic stirrer in the feed/retentate chamber. This is to ensure that any dissolved gas
After that, and always under an N 2 atmosphere, ~ 262 mL of anhydrous DMF were added and the solution was mixed using a magnetic stirrer. Then, 32 mL of ethyl acrylate (10 equivalents, 1 mol.L -1 ) and 5.9 mL of triethylamine (1.4 equivalents, 0.14 mol.L -1 ) were added to the flask and mixed. The flask was then connected to the system as a feed solution and kept under an N 2 blanket (~ 0.5 bar overpressure).
More feed solution was prepared throughout the running of the system by using the procedure described above but with 10 times lower catalyst and ligand concentrations. Catalyst loadings were varied throughout the continuous run in order to increase productivity and decrease residence time.

Membrane reactor stability study
In order to evaluate the membrane performance and stability over a prolonged period of operation whilst minimising the amount of catalyst used a 1100 hour long run was performed using catalyst loadings in the range 0.05-1mol%.  [2] , the postreaction mixture (100 mL) was allowed to cool down and diluted with toluene (100 mL). The solution was washed with 1M HCl (100 mL) and water (2 x 100 mL). The organic phase was dried over Na 2 SO 4 and the solvent removed in vacuo to yield a brown oil. The residue was triturated with hexane to provide the title compound. The mother liquor was concentrated, and the residue again triturated with hexane to provide further portions of the title compound. Total yield was ~72%. Isolated product was analysed by 1 H NMR spectroscopy. 1  . Data are consistent with that reported previously [2] . The isolated product was used for preparing artificial solutions for the kinetic studies and the preliminary continuous experiments on the second reaction.
The crude product from the continuous reaction was also analysed by 1 H NMR spectroscopy, and was found to contain <20% impurities, consisting mainly of the starting 1-Bromo-4-chloro-2-nitrobenzene (1) (excluding the triethylamine hydrobromide) ( Figure S3).

Figure S3
1 H NMR spectra of the isolated crude product from the continuous Heck reaction.

Reduction reaction kinetic studies
The kinetic studies were performed as series of batch experiments on a reaction The reaction is very fast and small samples 0.1 mL were taken frequently at about 2-5 min intervals to determine the reaction kinetic. Samples were diluted with 0.4 mL ethanol and analysed by GC. Typically the reaction got to completion within ~1 hour.

Continuous reduction reaction
A stainless steel column (240 x10 mm) was filled with 69 g iron powder (325 mesh, density 7.86 g.mL -1 ), void fraction ~0.47. The column was placed into a heating chamber, maintained at 85°C. The product stream from the membrane cascade was mixed with 0.59M NH 4 Cl aqueous solution in ratio 3.5:1 and pumped into the column via an HPLC pump. The pump flow rate was varied to match the flow rate from the cascade.

Product isolation
The postreaction mixture was filtered through Celite©, and the ethanol removed in vacuo. The residue was diluted with toluene, extracted with water, dried over previously [2] .
The crude product from the continuous reaction was also analysed by 1 H NMR spectroscopy, and was found to contain <20% impurities (excluding the triethylamine hydrobromide, that was carried trough from the first reaction), consisting of the  Table S1 FigureS5 Schematic representation of the membrane screening rig. Duramem 150, although exhibiting low permeance, showed the best rejection for the product and was chosen for further study in the membrane cascade.

Experimental setupmembrane cascade
Schematic representation of the membrane cascade is shown in Figure 1, to 0.1 L the response time will decrease from ~100h to ~45h without sacrificing the solvent exchange or the product recovery ( Figure S8). Alternatively the product dilution could be reduced by using a membrane with the same rejection but higher permeance. For example a membrane with a permeance 4 times higher than the current membrane would completely eliminate product dilution and even improve the solvent exchange ratio retaining the same product recovery (Figure S9).
Optimisation studies on the membrane cascade are beyond the scope of the current work, but these estimates indicate that the cascade performance can be further improved.  Table S1 the side product triethylamine hydrobromide is also well retained by the membranes. Interestingly the salt is much better retained by the 24% PBI membrane than the main product (71% vs. 44% rejection). This is an advantage and could be used to design a continuous membrane purification unit similar to the one reported earlier [9] which is able to retain the impurity and permeates the desired product through. Finally the product from the reduction reaction can be crystallised in a continuous MSMPR crystallization unit with integrated nanofiltration membrane recycle for enhanced yield and purity [10] . Conceptual design of a multiple reaction continuous process with different membrane units embedded is presented in Figure   S10. although small separation (less than 1% difference) seemed to occur for simplicity it was not taken into account; iii) it was difficult to quantify the osmotic pressure with so many compounds present in the postreaction mixture, instead the apparent permeance and rejection determined during the membrane selection experiments (Table S1) Recovery stage

Heck reaction
The model describing PFR-m-CSTR system is based on the mass-balance were determined as following: 9% substrate, 24% product and 54% salt. This experiment together with the kinetic studies was carried out in order to be able to make a reasonable prediction of the PFR-m-CSTR reactor performance. Most importantly our previous studies [11] have shown that side product (salt) solubility could be a major issue during the continuous experiment causing salt accumulation and consequent precipitation, reactor clogging and over pressurising. The salt solubility at 90 0 C in DMF was also determined at ~0.56 mol.L -1 . Using equation 38 two substrate concentrations (0.1 and 0.2 mol.L -1 ) were theoretically evaluated to verify whether the salt concentration in the m-CSTR reactor will remain below the solubility limit during continuous run. To simulate the extreme case it was assumed instantaneous 100% conversion of substrate to product and salt. As can be seen from Figure S11 for both substrate concentrations the salt in the reactor remains below its solubility limit and it was safe to perform the experiment.