Gen9 Vanishes

Earlier this week one of my colleagues had gotten a somewhat ominous email  from the CEO of Gen9 titled "Special Gen9 Announcement", which led off by saying that their holiday shutdown would be followed with a "corporate restructuring period" during which "Gen9 will not be accepting orders". The next day came an article from Scott Kirsner detailing the effective shutdown of Gen9 and sale of its assets to Ginkgo Bioworks for an undisclosed amount of cash and stock.  Interestingly, Kirsner reports that only 10 Gen9 employees will make the transition and that most of the Gen9 staff was laid off in mid-December.  It is surprising that no gossip of the cutbacks seemed to enter my radar, given a number of personal connections to the company (CEO Kevin Munnelly was a colleague at Millennium; several members of the Gen9 business group were ex-Codon or ex-Infinity and we had done limited business with Gen9)

Gen9 represents the third (not second, as Kirsner stated) attempt by George Church and Joe Jacobson and various collaorators to launch a company in the synthetic biology space.  First there was engeneOs, which was focused on applications.  Pieces of engeneOS, particularly much of the intellectual estate, went into Codon Devices, where I spent two years.  Codon tried to both tackle building interesting things with synthetic biology and revolutionizing the gene synthesis process itself.  Even after Codon failed, some of the ideas from it were being incubated in Franco Cerrina's lab at Boston University (Franco, a very kind soul and brilliant scientist, had been a key adviser to Codon).  Franco was tragically found dead of natural causes in his laboratory, but Jacobson and Church went on to found Gen9, which would focus on gene synthesis.  Pieces of Codon's intellectual property estate in applications ended up going into Eleven Biosciences and Celexion, both of which are now defunct.  Now Gen9 has ended its run as an independent company.  Gen9 was also part of a trio of new generation synthetic biology companies which launched around the same time that promised to radically alter the cost equation of gene synthesis: Cambrian Genomics is gone also leaving only Twist Bioscience.  Why all the trouble?

I think to really understand why these sorts of synthetic biology plays are so hard, you really need to look at the point-of-view of a customer.  I particularly think this way having previously been on the other side of the fence and really finding people like the current me very frustrating.  Companies like Gen9 tried to convince scientists to abandon conventional cloning in favor of "clone-by-phone".  Customers always have other options: they can not do a project, use conventional PCR, cloning and/or mutagenesis, assemble their own genes from oligos or use an existing gene synthesis vendor.

Ultimately, the challenge is that customers of gene synthesis want their genes delivered quickly, cheaply, unconstrainedly, correctly, flexibly and assuredly.  Speed is important because gene synthesis is a means to an experimental end; time spent waiting for synthesis is unwanted delay. Cost matters most for large projects.  Mutant libraries in particular are going to be more likely to be executed by conventional mutagenesis unless the cost of design and synthesis is competitive, or if the downstream assay has cost or capacity constraints in which a tightly designed library has overall cost advantages.  By unconstrained I mean the ability to design anything, regardless of GC extremes or hairpins or any particular sequence features.  For most projects that are being designed, the customers want that design and that design only, though there is a growing market for "mostly correct" designs.  Customers want flexibility in delivery formats; everybody has their own customized vector in which they want their genes delivered. Finally, if a company doesn't deliver on time nearly every time, then you'll be loathe to continue to use them.  And of each of these is really hard.

Cost is a real bear.  First, Gen9 located in the intellectually rich but costly Boston area.  I don't know the current favorite figure, but perhaps $400K per employee per year is the cost of employing anyone here.  That's not just salary, but salary and benefits and office space and basic consumables.  To operate continuously in the face of vacations and sick days and the many functional requirements, you can't operate a business like this with a skeleton staff.  I once guesstimated (around the time Codon shut down) that 30 was a ballpark figure for the minimum number of employees such a gene synthesis company could have.  Maybe you could get down to 20, but maybe not.  So that yields a fixed burn rate of $8-12M per year before you've ordered a single instrument or any reagents to make genes. My estimate isn't allowing much if anything for further R&D.   If you're going to cut the cost of gene synthesis, that running cost needs to amortized across all of your business, which means you need to have a lot of business.  Put another way, if you are burning $12M a year and making 100 megabases a month, then a penny a base is going to overhead.  Gen9 was trying to sell genes for 5-10 cents a base or so; that overhead could be a serious drag on that effort.

So to get costs down you really need to scale in a hurry; the more megabases (or gigabases) you can churn out the more you can spread that overhead cost around.  Now the challenge becomes finding customers who consistently want huge amounts of DNA continuously.  Ideally a small number of very large customers with very big orders, because each additional customer requires a certain amount of staff time to maintain the relationship, and each order requires a certain amount of staff time to vet.

To maximize your ability to find such customers and to deliver to them, you want a process that can make anything.  Anything! If you tell customers you can't build X or Y, then they may downsize or cancel the project.  Worse, if you say you can make X but can't , you'll lose credibility.  Synthetic biology has advanced a lot since Codon; approaches such as Gibson cloning have replaced the Type IIS restriction enzyme schemes we used for large construction, which greatly relaxes certain constraints.  But the gene synthesis options I am aware of still have severe limitations on extremes of GC content, and have other things they dislike (such as long runs of G).  

Even with these advances, building really large constructs (say tens of kilobases) still tends to be something these companies treat very gingerly, with good reason.  What if one piece won't build?What if the entire construct is unstable?  Builds are still some sort of hierarchical process; you can't do a 256-piece Gibson reaction (to be less facetious, I think even 8-way reactions are not strictly reliable).  Recombinational assembly in yeast is another possibility, but slow and requiring yet another set of skills, implying some more expensive skilled employees.

Checking synthetic constructs for correctness remains a challenge.  Ideally they'd be correct implicitly, and a key bit of the intellectual property passed from engeneOS to Codon to Gen9 was around error correction.  But you'll still need to sequence.  Sequencing costs have plummeted, but unfortunately not in a particularly useful way.  Base production costs have plummeted; the cost of library construction has fallen very gradually.  So if you want to verify ten kilobase constructs, you may be paying far more to get a construct ready to sequence than to actually sequence it.  Instruments such as Illumina are inherently batch instruments (as are the methods for inexpensive oligo synthesis), so you need to continuously have large batches of projects moving through the factory or the synthesis and sequencing cost equations get ugly in a hurry.  You also need to deal with constructs that aren't suitable for short reads.  I discovered belatedly at Codon that we had taken on one construct we could not verify; it contained exact repeats much longer than the Sanger reads we were using for validation.  If you take on long repeats, then you need to figure out how to use long, noisy reads to validate those and encounter yet another realm of batching rules, library construction and downstream informatics processing.

Servicing all possible customer vectors certainly makes your offering more attractive, but carries significant risk.  What if the vector doesn't work well in your lab?  What if it is unstable?  Perhaps it is a single copy vector?  Is it what your customer really says it is, particularly down to every base?  You can, of course, verify the customer vector, but that adds cost.  Plus the cost of tracking each customer's custom vector through your factory.  

Drop the ball on any one of these, and that "assuredly" goes out the window.  Allow yourself some cushion to help with batching or for missteps, and now your delivery time is stretched out.  It doesn't help that for the smaller, easier stuff there are many companies delivering such constructs with fast turnaround times.  This is again where scale of order comes to play.  If you have a single 1 kilobase gene and have a choice between two week delivery at 20 cents a base or 4 week delivery at 5 cents a base, will $200 vs. $50 be worth saving two weeks?  Perhaps.  But if you have 100 such genes to order, then $20K vs $5K is much more likely to get your attention.  Unfortunately, there are far more people thinking of small projects impatiently than large projects with a little patience, and besides those small projects don't fit well into the mass business model.

That's one reason I've been surprised that these newer companies haven't linked up with an established player such as IDT or MWG of Sigma-Aldrich or such.  Such a marriage wouldn't be easy, but might have the opportunity to batch together many smaller orders into the sizes that a Gen9-style company might find palatable.  Of course, the established company is taking some risk, and furthermore would need to be a reliable partner in encouraging some customers to divert their orders into a lower cost, slightly slower turnaround service.

It is interesting that Ginkgo has decided to take on Gen9's assets, given that Ginkgo has large supply agreements with Twist.  Apparently Ginkgo thinks they can profitably shave time off synthesis orders.  The issues I've outlined above are somewhat ameliorated when synthesis supplier and customer are in the same organization, but only somewhat.  For example, Ginkgo will know their own vectors inside-and-out, so that issue is tamped down.  But the inherent problems of scaling will remain, though perhaps Ginkgo is willing to trade time for cost.  Ginkgo also buys an insurance policy and bargaining leverage versus Twist, which appears to now have a monopoly in the open market for ultra-low cost gene synthesis.

Undoubtedly there will be additional startups in gene synthesis space.  I do remain a huge believer that radically altering DNA synthesis costs would transform biology, but as I noted before radically will mean 1-2 logs from where Gen9 and Twist were pricing (and perhaps, at least for Gen9, losing money on every base but trying to make it up in volume).  George Church and others want to write entire mammalian genomes; that would be $30M a genome at $0.01/basepair.  Oxford Nanopore has ideas for novel synthesis approaches and I've heard other noise.  But the challenging business pitfalls I've outlined will remain.  Perhaps future companies will look more like Synthetic Genomics, which largely sells the tools for customers to build their own genes.  That doesn't solve all the problems, but shifts many of them onto the customers.  It's no longer "clone by phone", but still a much wider space of "designer biology" than previously possible. 

Perhaps ultimately these gene synthesis capabilities will need to be regarded as infrastructure to be centrally funded rather than left to businesses likely to founder in a difficult business environment.  That's a bit of a cost-shifting exercise, but perhaps only with steady funding over very long periods can gene synthesis costs be brought down substantially.

It's always disappointing when companies such as Gen9 fail, as the biotech world gets a little less interesting each time.  Luckily, there is a regular stream of new and bold ventures to replace them.    

[2017-01-24 -- fixed misspelling of Franco's surname, which I greatly regret]