Cutting the cost of chiral production

Sarah Houlton reviews some of the newly developed production paths to chirals that manufacturers hope will improve yields, cut solvent use and lower costs.

In 2003, chiral compounds generated revenues of US$7.7bn. This figure is expected to grow by 11% a year between now and 2009. A little over half - 54% - came from the chiral pool or was made by the resolution of racemates. A further 35% was made by chemocatalysis, and 11% by biocatalysis.Much synthetic effort is focused on developing new and better ways of making chiral compounds, both in academia and industry, and Scientific Update's recent Chiral Europe conference, held in Cambridge, included numerous fascinating presentations about chiral synthetic methods - and a couple of talks that gave an insight into the problems of process development for complex chiral compounds.

As Clark Landis of the University of Wisconsin explained, there are a number of practical concerns that pose challenges in catalytic asymmetric synthesis. 'Green' process ideals include routes with high atom efficiency, catalysts that have high activity and are long lived, and processes that result in minimal, simple separations, which will often mean making enantiopure compounds rather than ones that will have to be separated by resolution.

A good synthesis will also go from simplicity to complexity in just a few steps, and give high chemo-, regio-, stereo- and enantioselectivity. Catalyst families that are selective for broad classes of substrates are ideal, as are catalyst families that are readily extended into diverse collections, e.g. by instant ligand modification.

Cheaper catalysts would also be advantageous - many of the transition metals that frequently feature in chiral catalysts are prodigiously expensive. 'We would like to use iron rather than rhodium!' Landis said. 'The aim is to transform cheap, poor catalysts into good ones for asymmetric hydrogenations, for example. Can we better emulate metalloenzyme environments in organotransition metal catalysts?'

When the crystal structure of a metal-containing enzyme catalyst is investigated, it is clear that there are many critical interactions that lie outside the metal co-ordination sphere, and the catalyst won't work without them. 'How can we create much more rich catalytic environments around the metals?' he asked.

By taking a modelling approach, Landis said, chiral diphosphine ligands can be designed. First, chiral diphosphines with relevant functionality are created. They are then attached to a transition metal - alkene complex with a complementary functional group. 'The DuPHOS template consistently showed promise,' he said. 'It puts the functional groups in the outer coordination sphere.' The biggest problem is synthesising them - making such richly functionalised ligands can prove difficult, as the myriad of functional groups involved are often incompatible, and the normal synthetic routes for making chiral phospholanes does not allow functional groups to be incorporated easily.

An alternative, which gives a similar shaped ligand with a simpler synthesis, are 3,4-diazaphospholanes. However, when they are exposed to the transition metal, the mixture goes 'gooey', so a benzene ring was added to make the structure more rigid. The synthesis is straightforward, and they've made a diverse array of diazaphospholanes. These can then be coupled with amino acids to produce chiral ligands. His group has also been looking at attaching the diazaphospholane ligands to solid supports, making a range of ligands in parallel without a need for chromatographic purification. The polymer-bound ligands can then be screened to identify targets for solution synthesis.

plant scale

A couple of industrial chemists spoke about the synthesis of real-world pharmaceutical ingredients at scale. Nathan Yee from Boehringer Ingelheim in the US described the development process of the novel hepatitis C protease inhibitor BILN 2061. More than 200 million people around the world are infected with the hepatitis C virus, and at least two-thirds of those infected will develop the chronic form of the disease. This is a progressive condition that leads to end-stage liver disease, including cirrhosis and liver cancer. The current standard treatment, Rebetron, is a combination of interferon A and oral ribavirin, and gives a 50% sustained response rate, but costs around $17,000 a year per patient.

A potential target for therapeutic intervention is hepatitis C virus NS3 protease, and in in vitro assays it was found to be inhibited by a hexapeptide, a lead compound that can be used as a basis for rational drug design. This led to the clinical candidate BILN 2061, and the development team was brought in to develop both a rapid route to supply kilo quantities for toxicology and Phase Ia trials. They also needed a practical route that was more cost-effective, and could be used as a basis for commercial manufacturing. The route needs to be highly reproducible, with no need for chromatographic separation; create products with very low levels of organic and inorganic impurities, and its solid state properties need to be strictly controlled.

Retrosynthetic analysis of the target molecule provides four fragments (see figure 1). Fragment 1, containing the cyclopropane unit, was made as a racemate, which then had to be separated into its enantiomers by resolution. The initial synthesis, while it gave rapid entry into the racemic ethyl ester of the fragment, had only modest diastereo-selectivity, gave a low yield, and used cryogenic chemistry. An optimisation programme led to a process that, while enzymatic resolution was still needed, required no product isolation until the final step. It can be scaled up to 100kg.

The process developed to make fragment 3, the cyclopropyl ester, was not asymmetric either, but it works well on a large scale - up to 100kg - and can be resolved enzymatically to separate the isomers.

The quinolone unit proved more problematic. The first step, the Friedel Crafts acylation of meta-anisidine, gave a mixture of products and just a 35% yield of the desired product on a 300g scale with the acyl group para to the methoxy unit - the literature had reported 40%. It was difficult to scale up, because the product decomposes and emulsions form during work-up, but meta-anisidine was an attractive starting material because of its low cost.

They landed on the alternative of using the Sugasawa reaction, which involves the electrophilic addition of acetonitrile in the presence of boron trichloride and aluminium trichloride instead. This gave the desired acylated product in yields of 42-47% on a multi-kilo scale. 'It's not a great yield, but it enabled us to move on,' Yee told the conference.

The heterocyclic part of the fragment was made separately and then coupled to the amine in a reaction with an 89% yield; ring closure gave the required quinolone unit in 75% isolated yield. Some problems were experienced on scale-up of this reaction, notably a low purity. This could be improved by using DME as solvent instead of THF; the synthesis proved reproducible on a multikilo stage.

molecule assembly

The remaining problem involved the assembly of all the fragments to give the target molecule. 'It was important to assemble it quickly, and find the assembly that minimised the number of moves and the total cost,' Yee said. The first route was designed to give rapid access to the first kilo of product for toxicity studies, with the aim of moving to a better synthesis for larger scale production. Fragments 1 and 2 were coupled by HOBT-catalysed bond formation, and this was then protected with p-nitrobenzoic acid in a Mitsunobu reaction, and then fragment 3 was attached. The ring was closed using a metathesis reaction which, Yee explained, was a nice reaction from an academic standpoint, but posed problems as a process because it makes a number of different dimers and oligomers.

Finally, the p-nitrobenzoic acid ester protecting group was removed, prior to coupling the quinoline fraction with a further Mitsonobu reaction. 'The reaction works, but it's not that great, and Mitsonobu reactions at scale aren't ideal,' he said. An alternative procedure proved more effective - going through the brosylate salt. This gave better yields, and required no chromatography.

project refinement

The project is still ongoing, and the team has now identified two scaleable routes to the product. They are continuing to look for more efficient syntheses of the building blocks, plus better catalysts and shorter reaction times for the ring closing metathesis reaction.

Another tale of process development was related by Stuart Mickel, who works for Novartis Process Research in Basel, Switzerland. Discodermolide is a sponge-derived poly-ketide, and preclinical data indicated that it was an extremely exciting lead structure for the development of novel anticancer agents - it acts as a microtubule stabilising agent.

The problem with discodermolide, however, is that natural sources provide at best several milligrams - it is present at just 2ppm in a sponge discovered by a manned submersible 50m below the surface of the Caribbean - and a biotechnological route proved impossible. So a practical synthesis that could deliver, in the first instance, gram quantities was essential, but ideally it would also have the potential to be developed later on into an industrial synthesis.

'It is the most complex molecule that Novartis process research has ever been asked to make at a large scale,' Mickel told the conference.

Discodermolide posed a huge synthetic challenge. It has 13 chiral centres and a trisubstituted cis double bond, which is particularly hard to make. Of the five published total syntheses of the molecule, that of the Smith group involved a high pressure reaction that can't be scaled up; Myles' route is not stereoselective; Schreiber's uses ozone and mercury (II) chloride; Marshall's involves low molecular weight acetylenes and tin chemistry; and Paterson's uses both selenium and boron chemistry.

hybrid route

The Novartis team decided that the best route was a hybrid involving elements of the Smith and Paterson routes - and the longest linear chain in the process is a 26-step synthetic sequence! Some of the more notable steps in the sequence include the second reaction, a lithium borohydride reduction of a methyl ester to an alcohol - at a 55kg scale.

The more obvious choice, lithium aluminium hydride gave an 85% yield, but the work-up includes a very slow filtration that took more than 24 hours. The borohydride alternative gave a marginally better yield, but the work-up was much more sensible - a quench with acetic acid followed by extraction - which was complete in less than three hours. Around 70m3 of hydrogen was produced as a by-product!

The fourth reaction, a boron triflate mediated aldol coupling, involved an unstable aldehyde that had to be used immediately after it had been formed. By this stage the scale was down to 25kg, and while the process only gave a yield of around 50% it was completely stereoselective and very tolerant of temperature effects.

high yields

Reaction nine, a second boron triflate aldol, worked extremely well in the plant and the product crystallised out of the reaction mixture in 85% yield. A number of side products were isolated, but these could be minimised by keeping the temperature below -5°C. Even though this gives a lower yield, it was much better to minimise the side products as they were extremely difficult to separate out.

It proved impossible to raise the yield of the lithium borohydride amide reduction in reaction 11 above 60%, even though the literature claims 90%. A number of side products were also formed and chromatography was required.

The fourteenth reaction was a Suzuki coupling of two of the fragments. This used a very expensive palladium catalyst that is tricky to make and requires a large excess of one coupling partner. The product is isolated in 70% yield from the reaction mixture by recrystallisation.

final assembly

Reaction 25 - the final assembly prior to deprotection - was described by Mickel as 'probably the most difficult reaction I have ever had to scale up'. This final Paterson boron aldol reaction gave yields between 30 and 60% at a 20g scale, and in the initial 50g reactions the yield was just 23%. Tweaking the conditions for the final run gave an HPLC yield of 60%, but around 20% vanishes during solvent evaporation. Using a peroxide work-up loses a further 20%. The solution was to eliminate the work-up completely, replacing it with chromatographic separation on a reverse phase silica column - hardly ideal, but it raised the yield to 50-55%.

The scale of the chromatography is astounding. A total of 700g of reaction mixture from two runs was diluted with 368kg of acetonitrile, TBME and water in a ratio of 85:15:10. This was applied to 20kg of RP-18 silica gel on a 120 x 30cm column. This was eluted with a further 1060kg of the solvent mixture, followed by a further 150kg of acetonitrile and TBME in a 1:1 mixture. Fractions of 20kg were collected and combined, they were evaporated to 10% of their original volume, the TBME was extracted and it was finally evaporated to give 150g of pure aldol product, free from epimers.

The product has dropped out of clinical trials, but had the project continued, Mickel says, the plan was to scale up by a factor of 10. 'If it went into production, then the scale-up would have been by a factor of thousands - that's oceans of solvent!' he said. Clearly this would not be a practical synthesis at large scale.

The final reaction was to remove three silicon-containing protecting groups. Even this posed problems, as the product cyclises and oils out, smearing itself around the reactor. This was, however, easy to recycle, simply by washing the reactor walls down with methanol. Despite the fact that the compound dropped out of trials because of toxicity problems, the statistics of the product the team made are pretty impressive. In the 20 months the project ran, they made as much discodermolide as would be produced by 3,000kg of the sponge, and the 120g of discodermolide required a total of 33 tonnes of solvents. In the earlier stages of the drug's development, access to large quantities of this extremely rare molecule took priority and, obviously, at a production scale this would not be practical.

Achieving the ideals of atom efficiency, low solvent use and high yields in the synthesis of complex chiral molecules will remain a priority for process development chemists as they search for cheap, effective, practical syntheses of their target molecules.