What is the cost of actual API in developing drugs?

Derek Lowe has been taking a chainsaw (or an axe) to Lexchin and Light's arguments that the actual cost of drug development is 43 million dollars instead of the Tufts $1 billion dollar number. I don't want to recapitulate the argument, so I'll send you to these posts:

• Derek talks about why Lexchin/Light are wrong
• Professor Light responds
• Derek talks about his view (the median viewpoint within pharma?) about the realities of drug discovery and development, and why it costs so much

To a manufacturing guy like me, some of the comments are quite interesting. From clinicaltrialist, the cost of (some) biologics:

I should add, for biologics trials, cost of goods can be material part of the cost. API for antibodies cost about $300/g or more, and typically, you would dose about 1 mg/kg once or week (roughly). That's 100 mg per week, or 5 g per year per patient. Once you add in the fill/finish, labeling and QA costs, it's about $500/g X 5 g = couple of thousand dollars per patient.

$300/gram!!? That's pricer than XPhos! (but not Josiphos!)

Here's a comment about the cost of small molecule API:

API cost can be an issue. The last campaign I ran cost ~$2M. For two batches. That doesn't cover all of Phase II - only part. Plus all the upfront cost of working out the synthesis strategy, impurity profile, polymorph stuff, GMP stuff...etc. This is not as expensive as clinical, but its not insignificant.

The comments on that particular post have really taught me about all the moving parts of a clinical trial. Basically, it seems like you're hiring doctors, patients, hospitals and people to watch over the process. That sounds expensive.