Insulin

Insulin has been first at almost everything a protein can be first at. It was the first protein whose complete sequence was determined, the first for which a precursor was found, one of the first to have its three-dimensional structure solved, and the first medicine ever made by genetic engineering. Each of those steps rested on the one before it, and the last of them turned structural knowledge into a practical result: insulins that are absorbed two to three times faster because a single amino acid was changed[1].

This page follows that chain of work. It is about the chemistry of the molecule and how it came to be understood, not about the treatment of diabetes.

Human insulin at a glance
A chainGIVEQCCTSICSLYQLENYCN — 21 residues
B chainFVNQHLCGSHLVEALYLVCGERGFFYTPKT — 30 residues
Disulfide bondsA7–B7 and A20–B19 between the chains; A6–A11 within the A chain
Molecular weight5807.6 Da, average; calculated from the two chains, less six hydrogens for the three disulfides
Made byBeta cells of the pancreatic islets, as a single-chain precursor
In solutionAssociates into dimers, and into zinc-containing hexamers

An extract that worked

By 1889 it was known that removing the pancreas gave dogs a severe and fatal diabetes. Many attempts followed to treat the disease by feeding pancreas or injecting extracts of it. The Toronto authors open their paper with the obstacle as they understood it: that the active principle was destroyed by the digestive enzymes also present in such extracts, and that attempts had therefore been made to eliminate those enzymes[2].

The Toronto group's first move was to get round that problem rather than solve it. They tied off the pancreatic ducts of dogs; over seven to ten weeks the enzyme-producing tissue degenerated while the islets did not, and an ice-cold extract of what remained lowered blood sugar in diabetic dogs. Extracts of liver and spleen did nothing, and the activity was destroyed by boiling or by pancreatic juice. Foetal calf pancreas, which has not yet developed its proteolytic enzymes, gave a potent extract too[2].

Their 1922 paper reports seven patients treated at Toronto General Hospital. The case set out in detail is a 14-year-old boy, admitted in December 1921, whose first injections in January produced only a 25% fall in blood sugar and no clinical benefit. Daily injections of a better preparation, made by Collip from whole glands, changed that: sugar in the urine fell from over 100 g a day to between 7.5 and 45 g, acetone disappeared, and the boy became "brighter, more active". When the injections were stopped for ten days, the sugar and acetone came back; restarting reversed it again[2].

Two things in that paper are worth noticing. The word insulin does not appear in it: the authors write about an "extract" and an "active principle". And they are careful rather than triumphant, describing the improvement as "of a temporary nature" and warning that "without careful control severe toxic reactions may be encountered"[2].

Two chains and three bridges

Human insulin drawn as two rows of lettered boxes. The A chain of 21 residues sits above the B chain of 30. A bracket above the A chain marks the disulfide between cysteines 6 and 11, and two diagonal lines join cysteine 7 of the A chain to cysteine 7 of the B chain, and cysteine 20 of the A chain to cysteine 19 of the B chain.
Insulin has two chains joined by two disulfide bonds between them, and a third disulfide inside the A chain. Sequences from Mathieu and colleagues (2017)[3]; the connectivity of the three bonds from Ryle, Sanger, Smith and Kitai (1955)[4]; the alanine at B30 in pig insulin from Adams and colleagues (1969)[5].

Knowing that insulin is built from two chains, and knowing the order of residues in each, still leaves a question: which cysteine is bonded to which? Insulin has six, and they could in principle be paired in several ways.

The obvious approach, breaking the protein into fragments and seeing which cysteines travel together, ran into a problem that is worth understanding, because it recurs whenever cysteine-containing peptides are handled. Disulfide bonds exchange partners. Under the conditions used to cut the protein up, the bonds rearrange at random, and the fragments then tell you nothing about the original arrangement. In concentrated hydrochloric acid the results were simply uninterpretable[4].

Ryle, Sanger, Smith and Kitai therefore spent much of their 1955 paper studying the rearrangement itself: that it is catalysed by thiols at neutral pH and can be partly blocked with N-ethylmaleimide, that the acid-catalysed version is inhibited by thiol compounds and runs more slowly in sulfuric than in hydrochloric acid[4]. Having found conditions where the exchange was slow enough, they could trust the fragments. The answer was A20–B19, A7–B7, and an A6–A11 bond inside the A chain, which they presented as completing the structure of insulin[4].

One detail connects this page to another. The authors noticed that the ring closed by the A6–A11 bond is the same size as the disulfide ring in oxytocin and vasopressin, and suggested it might have some structural or biological significance[4].

The shape, and the zinc

Insulin crystals had been known since 1925 and had been examined by X-rays since 1935, but the structure itself took until 1969. Dorothy Hodgkin's group in Oxford measured some 70 000 reflections on a four-circle diffractometer over 18 months and calculated the electron density of pig insulin at 2.8 Å[5].

The picture explained several things at once. The A chain folds into a compact unit with two short near-helical stretches, and the B chain wraps around it, with three turns of α-helix in the middle and extended ends. Both interchain disulfides sit at the ends of that helix. The A6–A11 bond is buried inside the molecule, which is why chemical syntheses that close it first have an easier time[5].

Insulin does not stay a single molecule. Two of them pair up through the ends of their B chains, forming an antiparallel sheet between residues B23 and B28 with the aromatic rings of PheB24, PheB25 and TyrB26 packed together. Three such dimers then arrange themselves around two zinc ions, each zinc held by three HisB10 residues and three water molecules. That hexamer is what sits in the crystal[5], and, as it turned out twenty years later, what sits in a vial of conventional soluble insulin.

The authors were cautious about what the structure explained biologically, writing that they could as yet say little about activity[5]. The practical payoff came later, and from a direction they had not been aiming at.

One chain before two

A two-chain protein raises an awkward question for the cell. Does it make the two chains separately and join them, or make one chain and cut it? Joining two chains correctly is difficult in the test tube, which made the second possibility attractive, but earlier searches had concluded that any precursor amounted to less than a tenth of pancreatic insulin[6].

Steiner and Oyer found one by changing the material. They used a human islet cell tumour, which is almost entirely beta cells and free of the surrounding digestive tissue, so a precursor would not be destroyed during the work. Incubating it with tritiated amino acids and separating the products by size gave, alongside insulin, a larger protein of about 10 800 Da[6].

Several results identified it. It bound anti-insulin serum, which the separated A and B chains do not. As little as 0.1 µg of trypsin converted it in half an hour into something the size of insulin. Splitting its disulfides released no separate chains before that trypsin treatment, but did afterwards. Its free N-terminus was phenylalanine, like the B chain[6].

Together these pointed to a single chain beginning with the B sequence, ending with the A sequence, and carrying an extra stretch between them. The authors put it exactly that way, as a possibility rather than a conclusion, and the paper's title calls it a "probable precursor"[6]. The connecting stretch is now known as the C-peptide.

Insulin from bacteria

Until the 1980s, insulin came from the pancreases of cattle and pigs, collected from slaughterhouses. The supply was tied to the meat trade rather than to medical need, and between 1970 and 1975 the number of glands available in the United States fell sharply and stayed down[7].

The 1979 experiment that pointed to a way out is elegant in its logic. Genes for the A and B chains were synthesised chemically, cloned separately, and fused to the gene for the bacterial enzyme β-galactosidase, joined to it through a methionine. Neither insulin chain contains methionine, so cyanogen bromide, which cuts after methionine, released the chain cleanly from its carrier. The fusion protein made up around a fifth of all the protein in the cells and was conveniently insoluble[8].

The two chains were made in separate bacterial strains and then combined in a test tube, by the method of S-sulfonated derivatives worked out a decade earlier[8]. It is worth noting whose method that was: Panayotis Katsoyannis, whose name also appears on the 1953 synthesis of oxytocin, the first peptide hormone ever made in a laboratory[9]. The same chemistry of protecting, deprotecting and reoxidising cysteines carried across from one hormone to the other.

What the 1979 paper demonstrated was expression and reconstitution, detected by radioimmunoassay at microgram scale. It was a proof of principle, not a production process[8].

Turning that into a medicine took three more years, and most of the difficulty was not chemical. Research with recombinant organisms was then limited to 10-litre cultures, a figure set at the 1975 Asilomar conference as the largest volume that could be handled conveniently in a laboratory. Manufacturing needed far more. A first request for an exception was made in June 1979 and recommended for approval that September, allowing 150-litre vessels; permission for 2000 litres followed in 1980. Clinical trials began in the United Kingdom in July 1980, and the application to the US regulator was filed in May 1982[7].

Proving that the product really was human insulin took twelve separate tests, of which the account singles out three as decisive: a radioreceptor assay, a radioimmunoassay, and high-performance liquid chromatography[7]. Recombinant human insulin was approved during 1982 in the United Kingdom, the Netherlands, West Germany and the United States, the first medicine of any kind made by this technology[7]. This account comes from the vice-president of research at the company that developed it[7].

Chemistry had got there first, in a sense: insulin was being made synthetically during the 1960s, well before recombinant production[10]. Those syntheses were done a chain at a time, which left the same difficulty the Genentech group met in 1979 — joining two chains correctly once you have them.

Designing a faster insulin

Five rows showing the last ten residues of the insulin B chain. Human insulin ends in proline, lysine, threonine. Insulin lispro has lysine and proline swapped; insulin aspart has aspartic acid in place of proline at position 28; insulin glulisine has glutamic acid at position 29 and a further change at B3; insulin glargine has two extra arginine residues added at the end.
Most of the changes that make an insulin faster or slower sit at the end of the B chain, the part that holds two insulin molecules together. Sequences and substitutions from Mathieu and colleagues (2017)[3]; the reasoning behind targeting this region from Brange and colleagues (1988)[1].

Injected insulin has a timing problem. In the neutral solutions used for treatment it exists mostly as zinc-containing hexamers, and a hexamer cannot cross into the blood until it comes apart. Absorption from under the skin is therefore too slow to match the rise in blood sugar at a meal[1].

In 1988 a group at Novo, working with Guy Dodson from the Oxford structure work, attacked this directly. Their reasoning used the crystal structure as a map. They knew which residues form the contact between two insulin molecules — B8, B9, B12, B13, B16 and B23 to B28 — and they knew that several of those also lie in the region that binds the insulin receptor. So they chose targets from the first list that were on the edge of the second: B9, B10, B12, B26, B27 and B28. Into those positions they put negative charges, placed to repel carboxyl groups already present across the interface[1].

The distances explain the results. B9Asp was the most effective single change, because in the dimer the two B9 and two B13 carboxyl groups come within about 4 Å of each other. B28Asp worked better than a change at B27 because B28 approaches a glutamate on the neighbouring molecule to within 3 to 4 Å, while B27 sits around 10 Å from the nearest charge. Trying to use bulk instead of charge, by replacing valine B12 with isoleucine, barely worked: that mutant still crystallised as a hexamer[1].

The best of these analogues were essentially single molecules at the concentrations used in a vial, kept most of their biological activity, and were absorbed two to three times faster after injection into pigs[1]. The authors were careful to say that immunogenicity and growth-factor effects still had to be examined before the new insulins could be judged[1].

The analogues in use today come from this line of work. Insulin lispro simply swaps the order of proline and lysine at B28 and B29, which destabilises the hexamer so that it breaks into dimers and monomers quickly. Insulin aspart replaces proline B28 with aspartic acid — one of the substitutions in the 1988 paper. Insulin glulisine changes asparagine B3 to lysine and lysine B29 to glutamic acid, and is formulated with polysorbate rather than zinc[3].

The same structural thinking, run backwards, gives insulins that last longer. Insulin glargine carries two extra arginines at the end of the B chain and glycine in place of asparagine at A21, which moves its isoelectric point to about pH 6.7: it is soluble in the acidic vial and precipitates in the tissue, releasing slowly, with a duration of 22 to 24 hours. Insulin detemir has threonine B30 removed and a 14-carbon fatty acid attached to lysine B29, which makes it self-associate at the injection site and bind reversibly to albumin. Insulin degludec also uses a fatty acid, but is held back by forming long multihexamer chains from which zinc must be released before absorption can proceed[3].

Before any of this, the problem was approached by formulation rather than by chemistry: adding zinc or the protein protamine to insulin produces lumps that dissolve slowly. Those preparations work, but their timing varies from injection to injection, partly because the suspension has to be mixed before use[3].

In short

  • The 1922 Toronto paper never uses the word insulin, reports seven patients, and describes the improvement as temporary. The chemical obstacle it got around was that the pancreas digests its own hormone.
  • Insulin is an A chain of 21 residues and a B chain of 30, joined by disulfides at A7–B7 and A20–B19, with a third bond inside the A chain at A6–A11.
  • Establishing that connectivity meant first studying how disulfide bonds swap partners during hydrolysis, and finding conditions where they do not.
  • The 1969 crystal structure showed the B chain wrapped around the A chain, two molecules pairing through the ends of their B chains, and three such pairs closing around two zinc ions.
  • Insulin is made as one chain, not two: a precursor of about 10 800 Da that trypsin converts to insulin, reported in 1967 as a "probable precursor".
  • Recombinant human insulin, approved in 1982, was the first medicine made by genetic engineering; the 1979 experiment behind it produced microgram amounts as a proof of principle.
  • Knowing which residues hold two insulin molecules together made it possible to design insulins that come apart faster, and are absorbed two to three times more quickly, by adding a single charge at the right place.

References

  1. Brange J, Ribel U, Hansen JF, Dodson G, Hansen MT, Havelund S, Melberg SG, Norris F, Norris K, Snel L, Sørensen AR, Voigt HO (1988). Monomeric insulins obtained by protein engineering and their medical implications. Nature 333(6174):679–682. DOI: 10.1038/333679a0
  2. Banting FG, Best CH, Collip JB, Campbell WR, Fletcher AA (1922). Pancreatic extracts in the treatment of diabetes mellitus: preliminary report. Canadian Medical Association Journal 12(3):141–146.
  3. Mathieu C, Gillard P, Benhalima K (2017). Insulin analogues in type 1 diabetes mellitus: getting better all the time. Nature Reviews Endocrinology 13(7):385–399. DOI: 10.1038/nrendo.2017.39
  4. Ryle AP, Sanger F, Smith LF, Kitai R (1955). The disulphide bonds of insulin. Biochemical Journal 60(4):541–556. DOI: 10.1042/bj0600541
  5. Adams MJ, Blundell TL, Dodson EJ, Dodson GG, Vijayan M, Baker EN, Harding MM, Hodgkin DC, Rimmer B, Sheat S (1969). Structure of rhombohedral 2 zinc insulin crystals. Nature 224(5218):491–495. DOI: 10.1038/224491a0
  6. Steiner DF, Oyer PE (1967). The biosynthesis of insulin and a probable precursor of insulin by a human islet cell adenoma. Proceedings of the National Academy of Sciences of the United States of America 57(2):473–480. DOI: 10.1073/pnas.57.2.473
  7. Johnson IS (1983). Human insulin from recombinant DNA technology. Science 219(4585):632–637. DOI: 10.1126/science.6337396
  8. Goeddel DV, Kleid DG, Bolivar F, Heyneker HL, Yansura DG, Crea R, Hirose T, Kraszewski A, Itakura K, Riggs AD (1979). Expression in Escherichia coli of chemically synthesized genes for human insulin. Proceedings of the National Academy of Sciences of the United States of America 76(1):106–110. DOI: 10.1073/pnas.76.1.106
  9. du Vigneaud V, Ressler C, Swan JM, Roberts CW, Katsoyannis PG, Gordon S (1953). The synthesis of an octapeptide amide with the hormonal activity of oxytocin. Journal of the American Chemical Society 75(19):4879–4880. DOI: 10.1021/ja01115a553
  10. Mathieu C, Martens PJ, Vangoitsenhoven R (2021). One hundred years of insulin therapy. Nature Reviews Endocrinology 17(12):715–725. DOI: 10.1038/s41574-021-00542-w