Extending the analogy of the body as a factory
A biology-heavy post
The analogy of the body as a factory, or multiple factories, is a useful one. From the outside, our bodies look relatively static and unchanging. But, in reality, we are constantly producing and degrading molecules, intaking and expelling, using fuel and producing waste. We eat, drink, and breathe an incredibly wide variety of substances, almost none of which can be used or even discarded without some form of transformation.
The cycles that transform these substances are factory lines, and subject to all the same constraints and measurements of factory lines in Detroit or Shenzhen: worker productivity, inventory, throughput, unit cost, etc. The workers being enzymes doesn’t change the need to use them productively, and the unit cost being measured in ATP doesn’t change the economic necessity of reducing that so the products are economical to produce.

This might all be relatively obvious to you. But I bring this up because I’ve been reading and thinking a lot about manufacturing recently, especially given that the production of NeutraOat is going to be more difficult and require a lot more hands-on work than Highway Pharmaceutical’s cat drugs. The supply chains and contract factories that work on cat drugs are much more established than those that work on grain products, and cat drugs have such good unit economics that manufacturers can have thoroughly unoptimized factories and still make money1.
As I’ve been reading about manufacturing and the problems that come up manufacturing physical products, it’s made me much more conscious of the problems that our body’s factory lines must face. I want to explore some of those today. A lot of the framework I’ll be using is from Eli Goldratt’s The Goal, which is a great book with a stupid name. I’ll also be using a lot of concepts from the Toyota Production System.
Let’s start with some definitions. When we’re talking about factory lines and their problems, there are basically three types of factory lines:
1. A factory line that produces one product. The real world example would be Henry Ford’s famous Model T line. The biological equivalent would be intestinal goblet cells, which produce almost solely MUC2.
2. A factory line that regularly produces more than one product. The real world example would be Toyota’s multiple car lines. The biological equivalent would be the metabolic pathways, which produce sets of NADH, ATP, and intermediates relatively regularly.
3. A factory line that inconsistently produces a mix of products. The real world example would be a tool maker with a large catalog, like Hitachi. The biological equivalent would be the liver, which has to rapidly produce and degrade enzymes to metabolize whatever broad range of toxins it encounters.
These are arranged in order of complexity of operation. A one-product factory has a consistent set of inputs and outputs, with only variations in quantity. A multiple product factory has relatively predictable variations in quantity, type, and setup. An inconsistently multiple product factory has unpredictable variations in quantity, type, and setup. Let’s start by discussing the problems of a MUC2 one product factory line.
But first, some context: The MUC2 product factory line is devoted to producing mucin 2, which is a member of the mucin protein family (surprise!). MUC2 is produced by goblet cells in the epithelial lining of the lumen of the large intestine. MUC2 combines with some other proteins and a bunch of oligosaccharide side chains to form the mucal lining of the intestinal epithelium. This mucal lining protects against harmful bacteria, provides a home for helpful bacteria, and lubricates the passage of whatever you’re digesting through the intestine.
Because a one product factory, like MUC2, has to worry first and foremost about quantity (i.e. getting inputs in the door and product out the door), the most obvious place for it to optimize first is trying to control the whole chain. Henry Ford famously did this by controlling steel mills at one end and dealerships at the other, all to try to optimize throughput. However, if that’s not an option, the factory line at least has to have very close relationships with both its suppliers and its distributors. In intestinal goblet cells, the equivalent of customer demand is the signal from cytokines and neural impulses asking for more mucin, the equivalent of distribution is granulation cytosis and then peristalsis pushing MUC2 through the microvilli and around the intestine, and the supply signal is handled by amino acid stress responses. The amino acids, especially threonine, are almost always the rate limiting ingredient for MUC2, so if the body has enough threonine, it probably can make MUC22.
The idea of optimizing throughput for a one product factory then necessarily demands the question: what throughput is optimal? The naive answer is “as much as possible”, but that’s not the case. A factory producing as much product as possible isn’t just a waste of money and resources, it also jams up the factory with a lot of slowly degrading product that needs to be disposed of. In the body, overproduction of mucus leads to choking, if in the lungs, or diarrhea, if in the intestines.
So, a factory’s optimal throughput is how much can actually be sold, for a physical factory, or distributed, for a biological factory like a goblet cell. This was another reason for Henry Ford to want to control the Ford dealerships. However, if that’s the factory’s optimal throughput, what rate should each step in the assembly line process goods at? Again, the naive answer is “as high as a rate as possible as needed for the optimal throughput”, and again that’s not the case. There will be, necessarily, a slowest part of assembly, a bottleneck or rate-limiting step3. If you run the steps before the bottleneck as fast as possible, you will just pile up slowly degrading inventory in front of the bottleneck. If you run the steps after the bottleneck as fast as possible, they won’t have anything to do.
Ford knew this, and he actually solved the issue in a very simple way. He physically limited the amount of space each assembly step had to store inventory in process. It was impossible for steps before the bottleneck to run at full speed because they needed to wait for downstream steps to clear the inventory already produced. They were “chained” to the bottleneck. In mucus cells, production is dictated by the clearance and the storage space: goblet cells can only hold a certain amount of granules at a time. Clearance can only be done by sentinel goblet cells, which flush the mucin crypts in case of a threat and wash away the infection, or by bicarbonate in the case of packed mucin. Either way, space is limited.
So, while we want to run the bottleneck at whatever speed is necessary to meet the demand of the end user, including maxing the bottleneck out. The steps before the bottleneck (assuming the bottleneck isn’t the first step) need to only run at the speed necessary to keep the bottleneck at the desired pace. But this then brings up the issue of spare capacity. If the bottleneck’s capacity is much lower than the steps before it or after it, we might be running our steps before the bottleneck at only a fraction of their possible capacity. It’s tempting, then, to economize, and try to reduce the capacity of the steps before and after the bottleneck to exactly the rate of the bottleneck. In an actual factory, this might imply firing workers, while in the body this might entail something like reducing the amount of spare enzyme.
This is a mistake, though. Not only does this ignore the much more obvious solution of trying to increase the capacity of the bottleneck – are you sure you don’t want more product? – but it reduces the ability to handle supply shocks. Let’s take an extreme example, where we’ve successfully reduced all steps to exactly the same rate. Then, literally any step, or multiple steps at once, can become a bottleneck with even a slight impairment in their rate. In a factory, this might be someone going home sick. In a mucus cell, this might be something like a dietary deficiency, leading to deficiency in an enzyme that’s used for some intermediate step. One predictable bottleneck is a lot easier to monitor and improve than a bunch of unpredictable potential bottlenecks.
So, you want to maintain some amount of slack in the system. How much depends on the system, how often it gets impaired, and how important it is to keep the system running vs. keeping it lean. A mission critical system should have lots of slack so it never goes down.
This idea of slack also becomes important depending on rates of error. While overproduction is inefficient, it can also relieve the need to have error-free production. This “safety stock” can be either extra product made during each step of production or a supply kept in storage and used when needed. The latter is preferred in modern manufacturing, both because it requires just a one time, rather than continuous investment in overproduction, and because it makes accounting for how many defective parts you have easier. It does, however, necessitate a storage space for the safety stock off-site and a means of transporting it into the workflow. For mucus cells, the body keeps its mucus stockpile in the colon, which also provides a home for the microbiome and a functional barrier for harmful bacteria.
Of course, the easiest way to address a high error rate is just to have fewer errors. Error fixing and prevention is a key part of manufacturing, both in the body and industrially. For parts that can’t be fixed, or errors caused by a defect in machinery, production needs to be stopped and the defective parts or machinery replaced, preferably as soon as the part is manufactured to avoid wasted time. For erroneous parts that can be fixed, they need to be tagged so that the next step down the line knows to fix them.
So, ideally, every step in the manufacturing line has five separate functions:
1. Check for any errors from the previous step
2. Stop the line if errors are unfixable or egregious
3. Carry out processing
4. Check own work for errors, tag if necessary
5. Transport the product to the next step
Given the lack of intelligence involved at all times in manufacturing inside the body, and often in manufacturing in industry, it’s necessary for these error-checking to be as automated as possible. This can be process-based, checking if the required steps were done, or outcome-based, checking if the intermediate product fits certain specifications.
In the mucus cells, these functions are separated. The error checking is mostly handled by the unfolded protein response and related functions. The processing itself is handled by the Golgi bodies. The process-based checking is handled by both glycosylation checks and polymer assembly. Then vesicles and, ultimately, epithelial cells, handle the transport to the next step. Outcome-based checking is done on a tissue level, checking to make sure the mucus is still functioning properly to coat the intestine. If it isn’t, it needs to be expelled through physiologic responses (e.g. mucus expulsion during sickness).
This almost concludes single-line product manufacturing 101. The last elements we need to cover are just the overall organizational elements, which tie into making production as efficient as possible. First, eliminate transportation as much as possible. While making transportation fast between steps is good, the fastest transportation is no transportation. Transportation should only be as necessary, like if there are two parts of production that have to be physically separated for safety reasons.
This is why mucus is produced, packaged, and released in the epithelia in a few micron radius, with very little transport involved. The only significant segregation in mucus production is in the colon itself. The mucus “stockpile” in the colon involves a two layer barrier, with the loose bacterially-colonized outer barrier segregated from the attached, sterile interior barrier. This keeps the interior barrier safe and ready to use as a stockpile, and allows for rapid flushing of the outer barrier.
Second, make your workers as efficient as possible, including by having more machines than workers. Workers are a continual cost, while machines are a one-time cost, so it’s generally better to have more machines than workers than vice versa. This also includes being flexible about reassigning workers if necessary, or at least letting them work on the step before or after theirs if they have downtime and it won’t cause overproduction. As you might guess, this also occurs in the mucosal cells: goblet cells can work on mucosal granule maturation, production of mucus itself, or discharge, depending on the need.
All of this is relatively easy to understand in a single product line factory. I don’t want to say it’s just common sense, but it does all flow from a central dictate to be as efficient as possible in matching the factory’s supply to the outside demand.
But single product line factories are rare, especially in the body. It’s much more common to repurpose factory lines for multiple products, because the fixed costs of a factory, or a cell line, don’t make sense to rebuild anew for most cases. This introduces the added complication of how to use the same machinery for different products by changing up the setups.
Fortunately, that doesn’t add a lot more conceptual difficulty, although it may add more practical difficulty. Setups, like transportation, are inherently non-productive. The best setup is no setup at all. However, unlike transportation, there is a second best setup, beyond just making setups more efficient. Some setups don’t have to slow down production, if they can be done without interrupting the flow of product. The easiest way to do this is just to have a spare machine that the worker sets up while product is still flowing through another machine in the old setup, but it’s also possible to set up a switch that flips between two setups.
In a machine factory, these are often physical casting dies, metal frames that molten metal can be poured into. Setups can involve fully changing over a die, swapping one in for another, or it can involve just changing around a die to make it appropriate for a different part. In the metabolic pathways, this can involve keeping redundant co-expressed isoenzymes online, like pyruvate kinase, which constantly has both PKM1 (always on) and PKM2 (regulatable) running. Or it can involve toggle switches, like the phosphorylation on/off gates of the pyruvate dehydrogenase complex.
The other complication is that communication between steps gets more complicated in a multi-product factory line. In a one product factory line, each step knows at all times exactly what to expect from the preceding and the following step. Everything from transportation to error checking is made much easier because there’s one thing to check for. In a multiple product factory line, steps either have to be prepared to receive and keep track of any of a number of products from the preceding step, or they have to have a master signal that tells them what product is being produced at any given time so they know what to check for at any given time.
Toyota famously solved this problem with a “kanban” system, essentially a bunch of boxes and notecards before each step. Workers have dedicated boxes for each part they could produce, along with notecards for each part. They not only can find parts when they need them, but also can use notecards to pass to the preceding step, asking for more parts, like a waitress at a diner. This has the added benefit of being a clear, visual representation of what parts each step is using, as well as an easy way to make production more efficient: if there’s one box per part, removing boxes forces workers to produce smaller lot sizes.
In metabolic pathways, the cards are metabolites, which can be directly read by regulatory elements. Meanwhile, the boxes are metabolons, enzyme clusters that also store substrates of intermediates.
A multiple product factory line that produces different products at irregular intervals faces this same setup and unpredictability problems. However, it also faces the issue that keeping readiness for parts and setups is wasteful. It makes much more sense to put setups or the necessary raw materials in a warehouse somewhere.
However, this raises the issue of memory. Whether in a factory or in the body, setups and production techniques have to be stored long term. This requires memory on an institutional level. But, frankly, I don’t have the room to get into this memory topic fully. Memory in the body spans the gamut from short term phosphorylation and metabolite balances to long term epigenetic modification. Bodily memory is inextricably bound to our persistence as an organism rather than a collection of cells, so it’s not a great thing to leave to the end of a blog post.
Alas, that’s going to be how it has to be.
Cat drug manufacturing is much more regulated, of course, which brings its own set of headaches, but that’s a different problem. The game ends up being less “optimize the factory” and more “don’t get the factory shut down”.
Assuming all steps don’t run at exactly the same rate.
All of this is obviously oversimplified so I don’t bore you with details of mucin production. One important thing to note in my oversimplification is that mucin production is a crucial part of keeping the intestines healthy, so while it is a one product line factory, it’s an incredibly crucial one. It’s like a war time factory that can never shut off. So, its controls also represent that.



Yes! The biology returns!