How a Cell Builds a Protein

Every red blood cell in your body is packed with about 270 million molecules of hemoglobin, the protein that picks up oxygen in your lungs and drops it off everywhere else. Your bone marrow makes more than two million new red cells every second, so it is building hemoglobin at a staggering rate. The instructions for it are written in your DNA.

Each hemoglobin molecule is four chains of smaller units called amino acids: two alpha chains and two beta chains. The beta chain is 146 amino acids long, and its recipe is a stretch of DNA on chromosome 11 called the HBB gene. This page follows that one gene, letter by letter, from the DNA to a finished, folded protein. Every sequence you'll see is the real human one.

Here is the whole trip in a single cell, drawn far from scale. Drag the slider to move through it and follow the gene's copy as it leaves the nucleus:

Three molecules carry the instructions in turn. The DNA holds the permanent copy and never leaves the nucleus. A working copy of one gene, the messenger RNA or mRNA, carries it out to the rest of the cell. There a machine called the ribosome reads the message and strings amino acids together in the order it spells, and the chain folds up into a protein. Biologists call this one-way flow, from DNA to RNA to protein, the central dogma.

Each step raises a question. How can a string of four chemical letters describe a chain of twenty kinds of amino acid? How does the cell find one gene among twenty thousand? And how does a molecule with no eyes read a message? The answers start with the shape of DNA.

Four letters

DNA is two long strands wound around each other. Each strand has a backbone of sugar and phosphate, and hanging off every sugar is one of four bases: adenine (A), thymine (T), guanine (G), and cytosine (C). The bases of the two strands meet in the middle, in pairs. Here are the first 21 letters of the HBB gene's code. Drag to turn it, and use the slider to take the twist out:

Twisted up, the molecule is the famous double helix, 2 nanometers across, with a full turn every 10.5 base pairs. Take the twist out and it turns into a ladder: the backbones are the rails and the base pairs are the rungs. The twist matters for how DNA packs and how proteins grip it, but the information is all in the rungs. Notice that every rung is the same length, and that green always meets red, and orange always meets blue.

That rule comes from the shapes of the bases. A and G are big, with two rings of atoms; T and C are small, with one. Along the edge that faces the partner, each base carries hydrogen atoms that can be shared (donors) and atoms that can accept them (acceptors). Choose a base for each strand and see which pairs fit between the two backbones:

Two big bases overlap; two small ones leave a gap. Only a big base with a small one spans the space between the backbones, and of those, only two combinations line up a donor against an acceptor at every point. A with T makes two hydrogen bonds, and G with C makes three. A with C, or G with T, have the right size but put like against like: an H against an H, a dot against a dot.

That one rule is what lets DNA be copied. Pull the two strands apart, and each loose base can only accept its own partner from the cell's pool of free bases. Slide the fork along to unzip this piece of the gene and let new bases move in:

Behind the fork the single double strand has become two, and they spell the same thing. Each copy keeps one old strand, drawn with a solid backbone, and gains one new one, because the old strand served as the pattern, or template. In a real cell, an enzyme called DNA polymerase does the fetching and joining, working from many forks along each chromosome at once.

The same pairing rule runs through everything that follows. A cell uses it to copy its genes when it divides, to make working copies of single genes, and, in the end, to read them.

Genes

Your cells carry about 3.1 billion base pairs in each set of chromosomes. Printed at one letter per millimeter, they would stretch about 3,100 kilometers. Somewhere in there is the beta-globin recipe. Drag the zoom slider from the whole of chromosome 11 down to single letters, and watch for the gene:

Out at the scale of the chromosome, the gene is invisible: chromosome 11 alone holds 135 million base pairs, and HBB is about 1,600 of them, near one tip. Closer in, it turns out to have neighbors. The two HBG genes make the globin chains a fetus uses, HBE1 one used by the early embryo, and HBD a minor adult chain. HBBP1 is a broken copy that makes nothing. Closer still, the gene itself has parts. The thick blocks are the stretches that end up in the message, called exons, and the thin lines between them are introns, which are copied and then thrown away. The thickest parts are the ones that spell out amino acids.

Look at the two strands at the bottom of the zoom. The start of the code sits 50 letters past the place where copying begins, and copying begins just after a short stretch marked promoter. A gene, then, is more than its code. It is a region with a landing site for the copying machinery, a stretch to copy, and signals along the way that say where to cut and where to stop.

The cell finds that landing site with proteins that grip particular short sequences. In many genes, this one included, the first to arrive is the TATA-binding protein, which grips a sequence close to TATAAAA about 30 letters before the start. Slide it along the beta-globin promoter and watch how many of its seven letters match:

Most places match two or three letters, which is what random DNA does, and the protein slides on. At position −32 the letters read CATAAAA, six of seven, and it grips. Gripping bends the DNA sharply, and other proteins gather on the bend, including RNA polymerase, the machine that will copy the gene. They set it down about 30 letters further on, facing the gene. Strands of DNA have a direction, written 5′ to 3′ after the way the sugars point, and the promoter's orientation decides which way the polymerase runs and therefore which strand it reads.The beta-globin promoter has more such boxes further upstream, CCAAT and CACCC. Single-letter changes in the CACCC and TATA-like boxes turn up in people with mild beta-thalassemia, in whom the gene is read less often than normal.

Real cells add a great deal on top of this: enhancer sequences thousands of letters away, proteins that switch the gene on only in developing red cells, and DNA packed so tightly that a gene has to be unwound before anything can land on it. The core idea survives all of that. A short sequence near the gene says "start here, go this way."

Copying the gene

RNA polymerase now does what the copying fork did, on a small scale. It pries open a short stretch of the double helix, reads one strand, and builds a matching strand of RNA, a close chemical cousin of DNA. RNA uses the base uracil (U) where DNA has T; it pairs with A just the same. Drag the slider to let the polymerase make the first letters of the beta-globin message:

Inside the polymerase about 14 letters of DNA are held open as a bubble. Only the bottom, template strand is read. Each new RNA letter is the partner of the template letter across from it, so the RNA comes out with the same sequence as the top strand, U for T. That top strand is called the coding strand for exactly that reason. The newest eight letters of RNA stay paired to the template, and then the RNA peels away and the DNA closes up behind the polymerase.

The first 50 letters aren't code at all. The message carries a lead-in, and only at letter 51 does AUG appear, where the ribosome will later start reading.

How long does this take? Measured in living human cells, RNA polymerase moves at roughly 2 to 5 thousand letters a minute once it is under way . That makes beta-globin quick. Choose a speed and compare it with two much longer genes:

At 3 thousand letters a minute, about 50 a second, beta-globin's 1,608 letters take about half a minute. The gene for titin, the largest protein in muscle, takes over an hour and a half. Dystrophin, a muscle protein whose gene stretches over 2.2 million letters, takes about half a day. Measured in cultured human muscle cells, the polymerase on dystrophin made 2.4 thousand letters a minute, so one copy takes about 16 hours . Most of those letters are introns, which brings us to what happens to the copy.

Cutting out the introns

In bacteria the RNA is ready to read the moment it's made. In our cells the first copy, the pre-mRNA, still contains the introns, and a large machine called the spliceosome cuts them out and joins the exons end to end. Each intron starts with GU and ends with AG, part of how the spliceosome recognizes it. Drag the slider to splice the beta-globin copy:

At the start, more than half of the copy is intron; when both are gone, the 1,608-letter copy has become a message of 628 letters, and the darker coding part, 444 letters, now runs without a break. Beta-globin is a small, tidy gene. A typical human gene has about 11 exons, and its introns average about 7,000 letters, so the cut-out parts usually outweigh the message many times over .

Splicing has a payoff. Many genes can be spliced in more than one way, skipping or keeping particular exons, so one gene can make several related proteins. The message also gets a chemical cap on its front end and a tail of a couple of hundred A's on its back end, which protect it and help it get read. Then it is exported through a pore in the nuclear envelope, and the cell faces its real puzzle: turning four letters into twenty amino acids.

The genetic code

Proteins are built from twenty kinds of amino acid, and the message has only four letters. One letter per amino acid can't work, so the cell must read words. Step up the word length and count how many different words there are:

Two letters give 16 words, four short. Three letters give 64, more than enough, and that is what cells use: the message is read in non-overlapping three-letter words called codons. With 64 codons for 20 amino acids there are spares, and the spares aren't wasted. Most amino acids have several codons, and three codons mean stop.

Here is the whole code. Choose the three letters of a codon, or click one in the table, and look at its neighbors in the same box:

Eight of the sixteen boxes are a single band: for CUU, CUG, or any CU codon, the answer is leucine, whatever the third letter. In most other boxes the third letter only splits the box in two. That is why the code is called redundant, and why so many changes in the third letter of a codon change nothing in the protein. Only AUG (methionine) and UGG (tryptophan) stand alone.

The table has other order in it. Codons with U in the middle all code for oily amino acids, so a change in the first or third letter there usually swaps one oily amino acid for another. And AUG has a second job: it is where reading starts. The three dark codons, UAA, UAG, and UGA, are where it stops.The table was worked out between 1961 and 1966, starting when Marshall Nirenberg and Heinrich Matthaei fed a string of nothing but U's to a cell extract and got back a chain of nothing but phenylalanine.

There are no spaces in a message, so where reading starts decides how it is split into codons. Move the starting letter and watch the chain below the beta-globin message:

From letter 51 the triplets read AUG GUG CAU CUG ACU, and the chain begins methionine, valine, histidine, leucine, threonine: beta-globin. Start one letter later and the same message reads UGG UGC AUC UGA, three amino acids and a stop. Two letters later gives five and a stop. Three letters later is back in step, minus the first amino acid. Each message has three possible reading frames, and only one is the protein.

The ribosome finds the right one by starting at the front of the message and sliding along until it meets the first AUG. In beta-globin there is no AUG in the 50-letter lead-in, so the first one is the real start.

Adapters

A codon can't grab an amino acid by itself: there's nothing about the shape of GAG that fits glutamate. The link is made by a family of small RNA molecules called transfer RNAs, or tRNAs. Each is about 76 letters long and folds back on itself into four short double-stranded arms. Drag to turn this one, and use the slider to fold its arms into place:

On paper a tRNA is drawn as a flat cloverleaf. Folded, its arms stack end to end into an L. At the tip of one arm is the anticodon, three unpaired letters that can pair with a codon. At the tip of the other, about 7 nanometers away, the amino acid is attached. Nothing in the tRNA touches both at once. The connection between a codon and its amino acid is made earlier, by a set of enzymes, one for each amino acid, that recognize their own tRNAs and load each with the right amino acid.These loading enzymes, aminoacyl-tRNA synthetases, check their work: several of them have a second pocket that clips off an amino acid that was loaded by mistake. So the code really lives in those enzymes.

The anticodon pairs with the codon the same way DNA strands pair, running in the opposite direction. Choose a codon and offer it different tRNAs:

A perfect match pairs all three letters the way DNA does. But a tRNA with the anticodon UUC also accepts GAG: the first two letters pair normally, and at the third a G sits across from a U, a looser fit that still holds. Francis Crick predicted this sloppiness at the third position in 1966 and called it wobble . Some tRNAs carry a modified base, inosine, that wobbles with three different letters, so one tRNA can read CUU, CUC, and CUA.

Wobble explains the shape of the codon table. The third letter matters least because it is the one checked least strictly, and a cell gets by with fewer kinds of tRNA than there are codons. The first two letters, though, must pair perfectly, and that is where the ribosome comes in.

The ribosome

The ribosome is where codon meets tRNA. It is built from two parts, a small subunit and a large one, each a tangle of RNA and dozens of proteins, and the message runs through the gap between them. Inside are three slots for tRNAs, side by side along the message, named A, P, and E. Turn the model, pull the subunits apart, and point at the parts to name them:

The message enters on the right and is read moving left to right, 5′ to 3′. The A site is where a new tRNA arrives with its amino acid, the P site holds the tRNA carrying the chain built so far, and the E site holds an emptied tRNA on its way out. The chain leaves through a tunnel in the large subunit that holds a few dozen amino acids, so the first part of a protein is already out in the cell while the rest is being made.

Step through three rounds of reading at the start of the beta-globin message. Watch where the chain is attached at each step:

Each round has the same four steps. A tRNA whose anticodon pairs with the codon in the A site settles in; the wrong ones are turned away. The large subunit then moves the whole chain off the tRNA in the P site and joins it to the amino acid on the new tRNA, making a peptide bond. Next the ribosome shifts exactly three letters along the message, which moves the tRNAs from A to P and from P to E, and leaves the A site empty over the next codon. The empty tRNA drops out of the E site to be reloaded.

The peptide bond itself is made by RNA. When the structure of the large subunit was solved at atomic detail in 2000, there was no protein anywhere near the spot where bonds form, only ribosomal RNA . Reading ends when a stop codon reaches the A site. No tRNA matches it; a protein called a release factor fits there instead and cuts the finished chain loose.

A ribosome doesn't need the message to itself. As soon as one has moved off the start, another can begin. Set how often a new ribosome starts on the beta-globin message and let it run:

Each ribosome takes about 26 seconds to make one beta-globin chain at 5.6 codons a second , and its chain grows longer as it moves toward the stop. With a new ribosome every 5 seconds, about five read the message at once, roughly what was seen when polysomes were first found on globin messages , and a finished chain comes off every 5 seconds. Start them every second and they jam: a ribosome covers about 30 letters, so they queue nose to tail and the output is set by how fast they can clear the start. A message being read by many ribosomes at once is called a polysome, and it is one reason a single gene copy can turn out so much protein.

Speed and mistakes

The hardest part of the ribosome's job is the check in the A site. Dozens of kinds of tRNA float in the cell, so for any codon nearly every tRNA that bumps into the A site is the wrong one. Some differ from the right one at a single letter and pair almost as well. The ribosome tells them apart by holding each arrival a moment and seeing whether it stays. Here is a toy version of that check, with rates chosen to land in the real range. Make the ribosome wait longer before it accepts a tRNA, and compare speed with mistakes:

With no wait at all, the ribosome is fast and useless: most of what it accepts is wrong. Each extra millisecond of checking cuts the error rate by a large factor, because a near miss lets go much faster than the right tRNA does, while the speed falls only gently. Wait too long and the right tRNAs start letting go too, and reading drags. Real ribosomes sit in the shaded band: in bacteria up to about 17 amino acids a second, in mammalian cells about 3 to 6, with one wrong amino acid in every 1,000 to 10,000 .

The real ribosome does better than a single wait. It checks twice, with an energy-spending step in between that a wrong tRNA can't undo, an idea called kinetic proofreading that John Hopfield worked out in 1974 . Copying DNA uses a proofreading step too and then adds a repair crew, which is how it gets down to about one error in a billion letters or better .

One mistake in 10,000 sounds tiny. Whether it is depends on how long the protein is. Choose an error rate per amino acid and see what fraction of chains come out perfect:

At one error in 10,000, about 98.5% of beta-globin chains are perfect, and for a typical protein of a few hundred amino acids the figure is still near 96%. For titin, the largest human protein at about 34,000 amino acids, only about 3% come out without a single wrong amino acid. Most of those mistakes are harmless, because a single wrong amino acid usually doesn't change how a protein folds, and cells break down the ones that come out badly. The errors that matter more are the ones written into the gene, because every copy then carries them.

Changed letters

A change in the DNA of a gene is a mutation. Unlike a ribosome's slip, it is copied into every message and every chain. The simplest kind swaps one letter for another. Pick a letter of the beta-globin code, by number or by clicking it, and choose what to change it to:

Most changes do one of three things. A silent change, like turning the third letter of codon 3 from CAT to CAC, leaves the protein exactly as it was, thanks to the redundant code. A missense change swaps one amino acid. The most famous is the sickle-cell change: GAG becomes GTG, and a glutamate becomes a valine. In 1956 Vernon Ingram showed that this one amino acid is the entire difference between normal and sickle hemoglobin . A different change at the same codon, GAG to AAG, gives hemoglobin C, a much milder condition. A nonsense change turns a codon into a stop. At codon 40, CAG to TAG ends the chain after 39 amino acids. It is a common cause of beta-thalassemia around the Mediterranean, especially in Sardinia .

Adding or removing letters is worse, because the ribosome keeps counting in threes from the start. Choose an edit and a place for it, and compare the two chains:

Delete one letter and every codon after the gap is read one letter off. The amino acids that follow are nonsense, and within a few dozen codons a stop turns up by chance. That kind of change is called a frameshift. Deleting three letters removes one amino acid and leaves the rest in frame. In 1961 Crick, Brenner, and colleagues used exactly this behavior, with deletions and insertions in a virus gene, to show that the code is read in threes . The loss of four letters at codons 41 and 42 of beta-globin is one of the most common causes of beta-thalassemia in southern China and Southeast Asia ; the shifted chain stops after 59 amino acids.

How often does a random single-letter change do harm? The code itself decides, and we can simply count. Choose which letter of the codon gets hit:

Of the 1,323 possible single-letter changes, about a quarter are silent, 73% are missense, and fewer than 3% are nonsense. Split by letter, the pattern from the codon table jumps out: changes in the third letter are silent more than two times in three, while changes in the second letter are never silent at all. Whether a missense change matters depends on where in the folded protein the new amino acid lands, which is where the chain goes next.

Folding

A chain of amino acids is not yet a working protein. It has to fold into the right shape, and for most small proteins nobody does the folding for it: the sequence alone decides. The main push comes from water. Several of the twenty amino acids have oily side chains, and water packs around oily surfaces awkwardly, so oily side chains end up pressed together, out of its way. Here is a made-up chain of 22 beads, 9 of them oily, jostled the way water jostles everything. Let it run, then turn the oily pull off and start again:

With the pull on, the chain thrashes for a moment and then collapses into a ball, with the oily beads in the middle and the others on the outside facing the water. Turn the pull off and the same chain stays a loose, wandering string. Real proteins fold the same way at heart, though with many more forces at work: hydrogen bonds, charges attracting and repelling, the stiffness of the chain. Small proteins can fold in a few microseconds, while others take seconds .

Two shapes turn up everywhere inside folded proteins, and one dominates hemoglobin: the alpha helix, where the chain winds into a tight spiral held by hydrogen bonds between each amino acid and the one four further along. Here are amino acids 3 to 18 of the beta chain, which form its first helix. Coil them up and look at where the oily side chains end up:

Stretched, the chain is over 5 nanometers long. Coiled, it is a rod about 2.3 nanometers long with 3.6 amino acids per turn, and the dashed lines show each amino acid bonded to the one four further on. Look along the rod and most of the oily side chains line up on one side, the side that packs against the rest of the protein. The beta chain has eight such helices, labeled A to H, folded around a pocket that holds the iron-containing heme group that binds oxygen.

Amino acid 6 sits on the other side, facing the water. Normally it is a glutamate, which carries a charge and is happy there. Switch on the sickle-cell valine and that spot becomes oily too: a small greasy patch on the outside of the finished protein.

One letter, whole cells

An oily patch on the outside of a protein has nowhere to hide, except against another protein. When sickle hemoglobin gives up its oxygen, its shape shifts slightly, and the valine patch on one molecule fits into an oily pocket on its neighbor. The molecules then stack into long, stiff fibers. Choose the hemoglobin, then lower the oxygen, as happens when blood passes through small vessels in working muscle:

Normal hemoglobin stays dissolved at any oxygen level. Sickle hemoglobin does too while it carries oxygen, but as the oxygen drops, its molecules join into fibers that stretch and stiffen the whole cell into the crescent that gives the disease its name . Stiff cells jam in small vessels and break apart early, which causes the pain and anemia of sickle cell disease. All of that follows from one letter, an A changed to a T in the seventh codon of a 1,608-letter gene: the cell copied the change faithfully, the ribosome read it faithfully, and the chain folded exactly as its sequence told it to.

The same machinery runs for each of your roughly 20,000 protein-coding genes, with different promoters deciding which ones a cell reads and how often. A red-cell precursor reads the globin genes almost to the exclusion of everything else; a nerve cell never reads them at all. Much is still being worked out: how a cell sets the pace of each gene, how a long chain finds its fold among countless wrong ones, and why some mistakes in folding lead to disease decades later. The four steps on this page, copy, cut, read, and fold, are the part that every living cell shares.

References

1. C. N. Tennyson, H. J. Klamut and R. G. Worton, “The human dystrophin gene requires 16 hours to be transcribed and is cotranscriptionally spliced”, Nature Genetics 9 (1995), 184-190.
2. I. Jonkers and J. T. Lis, “Getting up to speed with transcription elongation by RNA polymerase II”, Nature Reviews Molecular Cell Biology 16 (2015), 167-177.
3. A. Piovesan et al., “Human protein-coding genes and gene feature statistics in 2019”, BMC Research Notes 12 (2019), 315.
4. F. H. C. Crick, “Codon-anticodon pairing: the wobble hypothesis”, Journal of Molecular Biology 19 (1966), 548-555.
5. P. Nissen, J. Hansen, N. Ban, P. B. Moore and T. A. Steitz, “The structural basis of ribosome activity in peptide bond synthesis”, Science 289 (2000), 920-930.
6. N. T. Ingolia, L. F. Lareau and J. S. Weissman, “Ribosome profiling of mouse embryonic stem cells reveals the complexity and dynamics of mammalian proteomes”, Cell 147 (2011), 789-802.
7. J. R. Warner, P. M. Knopf and A. Rich, “A multiple ribosomal structure in protein synthesis”, Proceedings of the National Academy of Sciences USA 49 (1963), 122-129.
8. H. S. Zaher and R. Green, “Fidelity at the molecular level: lessons from protein synthesis”, Cell 136 (2009), 746-762.
9. J. J. Hopfield, “Kinetic proofreading: a new mechanism for reducing errors in biosynthetic processes requiring high specificity”, Proceedings of the National Academy of Sciences USA 71 (1974), 4135-4139.
10. V. M. Ingram, “A specific chemical difference between the globins of normal human and sickle-cell anaemia haemoglobin”, Nature 178 (1956), 792-794.
11. R. Galanello and R. Origa, “Beta-thalassemia”, Orphanet Journal of Rare Diseases 5 (2010), 11.
12. F. H. C. Crick, L. Barnett, S. Brenner and R. J. Watts-Tobin, “General nature of the genetic code for proteins”, Nature 192 (1961), 1227-1232.
13. J. Kubelka, J. Hofrichter and W. A. Eaton, “The protein folding ‘speed limit’”, Current Opinion in Structural Biology 14 (2004), 76-88.
14. P. Sundd, M. T. Gladwin and E. M. Novelli, “Pathophysiology of sickle cell disease”, Annual Review of Pathology 14 (2019), 263-292.
15. Sequences: NCBI RefSeq NM_000518.5 (HBB messenger RNA) and NG_059281.1 (HBB gene); gene positions from NCBI Gene, GRCh38. The genetic code is NCBI translation table 1.