I was on J. Craig Venter’s patio one evening, where he was entertaining a DARPA program manager, where he casually pitched an idea. He wanted to ‘3D-print’ insulin in space. This was, of course, the typical academic ‘solution looking for a problem’. In this case the hypothetical was that some astronaut got really sick and suddenly needed a drug (presumably a biologic). Mission control would upload a DNA sequence that would ‘code for’ the drug; a DNA synthesizer would get to work, make the sequence, then feed it to an in vitro expression system and out would pop the drug, saving the astronaut.*
As I worked on insulin in grad school, I risked my job as a lowly postdoc, chiming in and pointedly telling him (in front of the program manager) why it wouldn’t work. Some of my coworkers speculated that I might not show up the next day, but I survived to tell the tale. Since the idea of homebrewing insulin seems to keep coming up, I think it’s worth putting up why insulin is hard. Now, doing it in space in an in vitro system is in my opinion, far, far more difficult than doing it here on terra firma, so the point of this is not to say that it’s impossible, but to illuminate the difficulties that lie ahead in making DIY insulin, and to strongly suggest planning ahead before working on this. I can’t say I’m the most knoweldgeable person, but I’m making myself available for free consultation for anyone DIYer wants to pursue this problem.
In the words of a co-grad student of mine, who went on to work at Eli Lilly in their insulin department, “insulin is a bitch of a molecule”. And she’s an impressive and thorough scientist who spent four gruelling years of grad school tracking down a very subtle instrument artefact that pervaded a series of experiments in the lab – patience is a virtue she had in spades.
How does the body make insulin?
I once received a proposal to biosynthesize insulin where the designers took a lot of time optimizing the codons and choosing and designing the fusion block “easy optimizations”, while the construct looked like: [fusion]-[pubmed sequence]. This sequence is doomed to failure as it completely misses addressing the really hard part of making insulin. As a window into the hard parts, I’ll briefly go over how insulin is made in the body. Please refer to this image that I created about a decade ago to help in understanding this sequence of events.
- The DNA is translated to RNA and loaded into the ribosome. That’s the ‘easy part’.
- The beginning of the sequence contains a hydrophobic stretch of amino acids (‘signal sequence’). When this protrudes from the ribosome, it recruits something called the signal recognition particle, which pauses expression drags the ribosome to the endoplasmic reticulum.
- The translated insulin molecule is shoved into the endoplasmic reticulum amino acid by amino acid.
- The signal sequence is snipped off by a protease. The remaining sequence is called ‘proinsulin’
- In the endoplasmic reticulum, the proinsulin sequence is folded and disulfide bonds are forged to match the appropriate cysteine amino acids. These disulfide bonds are going to be important later. The proinsulin is then exported from the endoplasmic reticulum to secretory vesicles.
- In the secretory vesicle, the pH is gradually lowered activating a series of “prohormone convertase” enzymes (PC 1/3 and PC2) which trim out a middle section from the proinsulin sequence (that’s the C peptide). The remaining sections of insulin are now the B and A sequences – and they’re held together by disulfide bonds. Without the disulfides, B and A will float away from each other.
- We’re not done yet! There’s still these amino acids that were recognized by PC that are dangling off the end of the B chain that need to be removed by an enzyme, carboxypeptidase E (CPE).
So you should be able to see: If you take the raw pubmed sequence of insulin and try to build it, you will have the following grave sequence errors: 1) hydrophobic leader sequence, 2) C-peptide (plus connecting dibasic residues). Any strategy to make insulin using synthetic biology will require a strategy to either make the A and B chains separately, and stitch them together, or to build it with a C peptide and remove the C peptide. Unfortunately, ‘clean-cut’ proteases are hard to find: Typically the proteases used for this sort of thing are Ek, Xa, HRV3c, TEV and Thrombin; Only Ek and Xa, and thrombin make clean cleavages (don’t use Xa, it’s expensive), and none of them leave clean C-termini.
Did you make insulin?
If the plan is to make insulin, even as a proof of concept, you’ll have to figure out how to assess if you’ve actually made it.
The most common, inexpensive way to check to see if you have made a protein is by SDS-PAGE gel. However, the effective range of standard SDS page tends to start around 10 kDa, and insulin at around 6kDa is much smaller than that. If you follow the ‘standard’ SDS-PAGE procedures, then you’ll probably thoughtlessly put DTT or 2ME into the gel and/or loading buffer, which will reduce the insulin disulfide bonds, and turn it into two smaller peptides, which will be even harder to detect! In order to reliably detect insulin, I routinely made by hand and ran custom 10-25% PAGE gradient gels.
Insulin doesn’t stain very well with coomassie blue. Coomassie staining works by sequestering the dye in hydrophobic regions of the protein, which are somewhat inaccessible in insulin due to the disulfide bonds, and is enhanced by the presence of basic amino acids (of which there are only two in insulin). A reducing gel will of course liberate the disulfide bonds and give more access to the hydrophobic amino acids, but now you’ve got two, smaller peptides, so detection will be even harder. In my experience, insulin doesn’t stain with silver staining at all (I don’t know why). In order to reliably detect insulin, I had to use western blotting.
I would suggest mass spec (which is the cheapest of the easy methods, but an order of magnitude more expensive than gels), keeping in mind that mass spec will not report on whether or not the disulfide bonds are properly formed.
How it’s made commercially
Originally Genentech made insulin by producing the A chain and B chains separately in E. coli. After purifying, cleaving from the fusion protein, re-purifying by HPLC, the two chains are stitched together in oxidizing conditions to form the disulfide bonds. Here, there is a problem. You can choose to make the proteins in highly denaturing conditions, in which case the disulfide bonds will be scrambled and sorting out the correctly disulfided-bonded insulins will be challenging. Or you can do it in protein folding conditions, in which case you strongly favor correctly disulfide-bonded insulin. That sounds great, but because it’s a biomolecular reaction, it’s slow if the A and B chains dilute – and if it’s concentrated, then the B chain undergoes an autocatalytic precipitation and the reaction yields are severely compromised. The genentech strategy is no longer used.
There are two contemporary, competing strategies for making insulin. Eli Lilly makes a proinsulin in E. coli with a fusion partner, with disulfide bonds reduced, folds it ex-vivo (this is a difficult process requiring exquisite fine-tuning of temperature, pH, and oxidation) and then treats it with enzymes to liberate insulin. Novo Nordisk makes a pseudo-proinsulin in yeast, where it’s folded inside the cells, cleaved using internal yeast enzymes. Finally, a solid-phase-peptide-synthesized section of insulin is added by reverse proteolysis (that’s running a protease backwards!!), followed by chemical deprotection. If this sounds complicated, it is. While this may work at scale, DIYers should not do it the Novo Nordisk way. Consider doing it like Lily, but make a solid plan. Or come up with something totally new.
Modifying Insulin
Lastly, I give one caveat to modifying insulin. Insulin cross-reacts with the IGF-1 (insulin-like growth factor) pathway. Like all growth factors, overstimulation of the corresponding receptor (IGFR) leads to overgrowth of cells, or in short – cancer. In pharmaceutical practice, every single insulin analog must be tested against IGF activity. Long-term surveillance of insulin analogs (such as inhalable insulin) is hyper-sensitive to the appearance of cancer. If the plan is to make an insulin analog, then stick to ones that are literature-tested.
Conclusion
A lot of biohackers (even PhD level biohackers ensconced in academia) are basically script kiddies – cut and paste these bits of DNA and cross your fingers and hope the result you expect pops out. There’s nothing wrong with being a script kiddie, that works sometimes, and well enough, but there are going to be some problems intractable at that level of understanding. Insulin is one of these. Be prepared to invest a lot of time and thought on the problem.
*there is of course a much better choice than insulin for this hypothetical – an scientist in Antarctica was diagnosed with breast cancer in the middle of the winter and a special, dangerous, delivery of chemotherapeutics had to be airdropped. Since cancer, unlike diabetes, is a disease that can have a sudden onset, one could imagine treating a cancer patient with frontline biologics that are not stocked (there are about 20-30 on the market or about to hit the market) to save space on the space vessel, and ‘print out’ the appropriate biologic on-demand.