The JDRF has spent a lot of effort and money on projects that address both Smart Insulin and Artificial Pancreases. Most of us are very familiar with the concept of Artificial Pancreases, but fewer are aware of the smart insulins. In brief, quoting from the JDRF page on Smart Insulin:
‘Smart’ insulins or glucose-responsive insulins are being designed to only turn on when they’re needed and off when they’re not. These insulins could make hypos history and help ensure perfect glucose control throughout any given day.
Iin 2008 when Todd Zion, a chemical engineer at MIT, began to experiment during his doctorate with chemically modifying insulin so it would automatically react to changing blood glucose levels, interest grew. Merck now owns the company he set up and started clinical trials in 2015.
The JDRF is also funding John Fossey at the University of Birmingham in the UK, who has taken a different approach creating balls of gel that contain insulin and work like a bath bomb.
The JDRF likes to use the following image to describe the concept:
Where glucose in the blood will unlock the insulin molecules when they reach a certain concentration.
Whilst we know this is some distance away from coming to fruition, there is research going back to 1978 that has considered glucose responsive approaches.
Whilst we talk about Smart Insulin in a different breath from artificial pancreases, it is a chemical or biological closed loop. Platforms like OpenAPS and the 670G and successors are electronic closed loops so I thought it might be an interesting experiment to look at what the electronic one does, how that might be managed in the Smart Insulin, and if there is anything that raises eyebrows in relation to this thought experiment.
Closed Loop Systems
Ultimately, Smart Insulin and Bionic Pancreases have a lot in common. Both use a signalling mechanism to determine when to deliver more insulin and when to deliver less insulin, and then do so. Technically, Smart Insulin should already be in the blood, so its time of reaction to signals plus action should be considerably quicker than interstitial fluid being monitored, and exogenous insulin delivered from an external source via a cannula. Both can be described as Closed Loop Systems.
But experience of the closed loop approaches that currently work raise some interesting questions of Smart Insulins. The key driver behind the development of all closed loop systems to date has been safety. How do you ensure the user is safe? Then it’s how to make it work effectively.
First up, the open source closed loops all rely on user input to provide key data that determines insulin delivery, which includes insulin sensitivity, insulin carb ratios and background insulin requirements for starters. OpenAPS-Autotune then allows the system to learn about the user’s insulin needs and adjust things accordingly over time, whilst Autosens manages this on an intraday basis. Commercial platforms like iLet and Bigfoot take the user’s weight and then titrate from an estimate based on that.
We know from experience that not everyone has the same values for these things, and that they can vary quite a bit by day, season, health, illness and a myriad of other aspects.
What does this mean?
It means that in an algorithm-based closed loop, for a glucose level that is 20 points above the target, in person A, the loop will deliver a units and in person B, b units to get the glucose level back to target. In theory the same algorithm will deliver the insulin at the same time, and both A and B should get back to target in a similar timeframe.
How would this work for a Smart Insulin? It has no concept of how many units the person would need to move from T+20 to T. It simply releases insulin at a rate when the glucose level is above it’s threshold, and let’s assume that is T. The other assumption is that it releases glucose at a rate that varies starting at r when the level is T+1 and xr when the glucose level is T+X, with some upper bound on what X might be. How that rate varies might be achieved using quanta above T or might be a more sliding scale above T until we reach X. Both of these have implications for how they might work in people.
In practice this would mean that someone with a greater insulin sensitivity will see faster action from a smart insulin than someone with lesser insulin sensitivity, and the person with lesser insulin sensitivity will also see greater post-prandial rises and more time above the threshold level, for the same amount of insulin.
It also means that someone who is less insulin sensitive will need more insulin, more regularly, in order to achieve a blood glucose reduction of the same level as a more sensitive person (which is also true of refilling pens, pump carts, etc., now). Essentially, it will be necessary to have a higher concentration within the blood to get the same effect.
And what of the off button? What happens when you exercise, for example? Or if your insulin sensitivity changes fast, as in during or post exercise? At the moment, on an algorithmic system, you can inform that you want it to target a higher glucose level and then leave it be whilst it modifies its behaviour based on sensed changes. How does a Smart Insulin handle this?
What about the insulin embedded in these products itself? What would it be? Regular Human insulin (which we know doesn’t absorb well subcutaneously, but when applied via IV, is very effective, very fast)? Or would some other formulation be needed to work in the context of the mechanism that manages release?
If we look at a few studies [1, 2, 3] of artificial regular human insulin when used intravenously, we find that it has a half life of somewhere between 10 and 15 minutes, as opposed to the often assumed 5-8 minutes that we see with naturally produced insulin. With this level of half-life, insulin release threshold and maximum effect (essentially, concentration) become critical. It doesn’t take too much either way for hypoglycaemia to occur unexpectedly, and with current forms of insulin, you might as well use “regular” as when administered intravenously, there is little difference between it and faster acting insulins, which are designed to reduce absorption time when administered subcutaneously rather than have a faster action when in the blood.
Another question for me is the accuracy or otherwise of the medium which senses glucose. Whilst we know that the best CGM systems have a MARD of 9%, and finger pricking devices, up to 15%, plus the error grid percentiles, what does this mean for something in the body? Is a nearly 10% error good enough when you can’t directly take over control? What about insulin delivery? In a mechanical system, it’s not always easy to determine exactly what of the 0.1u or 0.05u that was released has been absorbed, but you know what’s been released. When the release mechanism is handled in a chemical fashion, it becomes even more important to manage the accuracy of dose.
I guess the other thing that you have significant control over with a mechanical solution is the ability to down regulate and to manage degradation of components. What happens in a Smart Insulin if it is accidentally used beyond the recommend time period and the control mechanism no longer functions as it should? There is no calibration mechanism, so you could incur faster or slower release of insulin, with the associated “side effects”.
Many of these questions are discussed in the recent review of Glucose Responsive Insulin undertaken by Jianhai Yang &, Zhiqiang Cao for the JDRF, but they are all pertinent and difficult to solve using chemical engineering techniques and present the key challenges.
But, so what?
The key here is that many of the issues that we have been trying to address in closed loop physical/algorithmic/mechanical systems also have to be addressed in Smart Insulin. The key one being, how do you do this safely?
When you consider the reference model for oref0 in OpenAPS or what Medtronic have done with the 670G to ensure it is safe for users and get it past the FDA, you realise that in order to make smart insulin effective you have to recreate a similar framework in a totally new paradigm. You also have to consider that in the Smart Insulin model, there is much less ability to customise the behaviour of what you’re using, or intervene if things go wrong, so you need something that works for most people, most of the time with as few risks as possible.
So while Smart Cells and Merck may have completed a phase 1 clinical trial on Smart Insulin, I suspect we aren’t going to see Smart Insulins coming to the market any time soon. If I had to place a bet, I’d say that the mythical “ten years out” model is one to be adhered to, and that it may well be “ten years +”, and that’s if those working on it can get the products doing what they need to. Given that the JDRF is also funding stem cell repurposing and beta cell encapsulation, it’s very obvious that none of these are absolute bankers in terms of a better treatment option.
Fundamentally then, until there’s a cure, closed loop is the end game. Biological and chemical closed loops are by far the hardest to achieve consistently and safely, and that means the electronic version will be here for a while.
Interesting analysis as always, Tim. One small comment:
In this bit, “In practice this would mean that someone with a greater insulin sensitivity will see faster action from a smart insulin than someone with lesser insulin sensitivity, and the person with greater insulin sensitivity will also see greater post-prandial rises and more time above the threshold level.”
Should the “greater insulin sensitivity” be “lesser” in that second part? Seems like it, but maybe I’m misunderstanding.
Thanks for the thought-provoking piece.
Thanks! And yes, good catch Eric!
So this insulin will never be released to the public.