Every so often a seed comes out of the roaster pale. The batch around it is the color we were aiming for, and this one bean sat in the same drum, in the same heat, for the same number of minutes, and it is still the color of raw peanut. It is called a quaker, and it is the cheapest chemistry lesson in coffee. A quaker is a seed from a cherry that was picked before it ripened, so it never built up the things that brown. Same heat, same time, no color.
That tells you something precise: browning is not what heat does to a coffee seed. Browning is a reaction between two specific ingredients, and if the seed is short of either one, the heat has nothing to work with. And it points at the second half of the answer, which is the part that surprised me when I went looking for it properly. The brown material those ingredients turn into is not caramel. It is mostly the seed itself.
What is actually in a green coffee seed
Start with the inventory, because the proportions do most of the arguing here.
By dry weight, a green arabica seed is roughly half insoluble structural carbohydrate: cellulose and the cell wall sugars that hold the thing together. After that, the largest pool is protein, at about 11 to 15 percent. That is close to the protein content of bread flour, and it is worth sitting with, because protein is not what anyone thinks of as a coffee component.
Sucrose, the table sugar that gets all the credit, is 6 to 9 percent. Chlorogenic acids, the big family of plant acids that break down into much of what we taste as brightness and, later, as bitterness, run 5 to 8 percent. Trigonelline is 1 to 3 percent. Caffeine is about 1 to 3 percent, and it does not meaningfully change at any roast level, which is its own small correction for another day.
Then the two pools that matter most for color, and they are tiny. Free amino acids, meaning amino acids floating loose rather than bound into protein, come in at 0.3 to 0.6 percent. Glutamic acid is the largest single one, with asparagine, aspartic acid, alanine and proline behind it. And free reducing sugars, meaning glucose and fructose sitting loose in the seed, are 0.2 to 0.5 percent.
Look at those last two numbers next to each other. They are the same size. At the moment a coffee seed starts to brown, it holds about as much loose amino acid as it holds loose sugar, and both are rounding errors against the 6 to 9 percent of sucrose everybody talks about. That seems like it cannot be right, because sucrose outweighs the free amino acids by something like fifteen to one. The reason it is right is the next section, and it is the single most skipped step in the usual explanation.
Sugar cannot react until it breaks in half
The browning reaction at the center of roasting is the Maillard reaction, named for Louis-Camille Maillard, who described it in 1912. The short version is that a sugar and an amino acid join up and then fall apart into hundreds of smaller, smellier things. The longer version is that the sugar's carbonyl group condenses with the amino group, loses a molecule of water, rearranges into a more stable compound, and then that compound dehydrates and fragments into the reactive two-carbonyl molecules that go on to do most of the interesting work.
The detail that matters: only a reducing sugar can start that reaction. A reducing sugar is one with a free carbonyl group available to do the joining. Glucose and fructose are reducing sugars. Sucrose is not. Sucrose is a disaccharide, a glucose and a fructose bonded together, and the bond ties up exactly the group the reaction needs.
So sucrose cannot take part at all until it hydrolyzes, or inverts, into its two halves. Until that happens, the 6 to 9 percent is locked. The reaction runs on the 0.2 to 0.5 percent of loose glucose and fructose, plus whatever sucrose has split so far, plus the sugars that come free as the cell wall polysaccharides break down under heat.
This is why the arithmetic earlier looked wrong and is not. Sucrose is not the thing that reacts. Sucrose is the reservoir that feeds the thing that reacts, and the rate at which it opens is one of the things a roaster is actually controlling.
Two reactions, and they are further apart than they sound
Caramelization is a different reaction, and the difference is not a technicality. Caramelization is sugar breaking down under heat with no amino group involved at all: sugar alone, decomposing and then polymerizing into brown pigment. Maillard needs both a sugar and an amino group. Caramelization needs only heat and sugar.
They also happen at different temperatures, and the gap is large. Work on caramelization in food systems puts its floor above about 120°C, or 248°F, and sucrose specifically does not begin to caramelize until around 170°C, or 338°F. The Maillard reaction proceeds effectively from about 50°C, which is 122°F. That is cooler than the water out of a hot tap. Maillard chemistry is, very slowly, happening in a sack of green coffee sitting on a warehouse floor.
Here I have to reconcile two numbers that look like a contradiction, because a careful reader will notice it. Coffee sources generally give Maillard an onset of 140 to 165°C. That figure and the 50°C figure are both correct and they are answering different questions. 50°C is where the reaction can proceed at all. 140°C is where it proceeds fast enough to see and smell inside a roast that lasts ten minutes. Reaction rate roughly triples to quintuples for every 10°C, so a reaction that is real but glacial at 50°C is transformative at 180°C. When someone says browning starts at 140°C, they mean browning becomes visible at 140°C.
Which leaves the useful version: Maillard runs the entire roast, from before the first minute to the moment of drop. Caramelization switches on near the end, at temperatures you reach around or after first crack. They are not a sequence. They overlap, and for the back half of the roast they are competing for the same loose sugar, with Maillard holding a 120-degree head start.
And by the time sucrose caramelization is available, there is very little sucrose left to caramelize. Measurements across roast levels have sucrose at roughly 3 percent of its original amount in a light roast, under 1 percent at medium, and gone at dark. More than 99 percent of the sugar in green coffee is destroyed during roasting. Most of it was consumed by the reaction that started at 50°C, not the one that started at 170°C.
What the brown actually is
If the sugar is nearly all gone, and the bean is visibly brown, then the brown is made of something. It has a name. Melanoidins.
Melanoidins are the large brown nitrogen-containing polymers that the Maillard reaction ends in, and estimates put them at about a quarter of the mass of a roasted coffee bean. That estimate deserves a caveat, and I would rather give it than not: melanoidins have no single defined structure. The figure is a mass-balance estimate of a high-molecular-weight fraction that has not been fully characterized. Researchers working on it say the structures and mechanisms remain to be established. So treat 25 percent as a good estimate of a category, not a measurement of a compound.
The interesting question is what they are built out of, and there is a piece of work that answers it unusually cleanly. A group at the University of Aveiro took green coffee apart, added single components back in, and roasted the reconstructed beans, which lets you see what each ingredient contributes. Published in the Journal of Agricultural and Food Chemistry in 2012, Fernando Nunes and Manuel Coimbra's in-bean models found that roasting-modified galactomannans, which are cell wall sugars, accounted for 47 percent of the high-molecular-weight brown material, and that proteins supplied about 92 percent of its nitrogen. Stripping out the hot-water-extractable fraction cut melanoidin formation by 76 percent.
And the line that reframes the whole subject, in their own words: sucrose is important for the formation of colored structures but not to the formation of the unknown material. Adding sugar made the bean browner without making more of the polymer.
Structural work supports it from another direction. Comparing green and roasted coffee, the protein bands visible on a gel fade after roasting while very large aggregates appear that exist only in roasted samples. Water-soluble protein drops by roughly half. The protein is not simply being destroyed. A good deal of it is being built into something bigger.
So the brown in your grinder is a polymer assembled largely on the seed's own cell wall sugars and its own protein, with the loose sugars acting as the reagent that starts the reaction and then leaving almost nothing of themselves behind. Sugar is the match. The cell wall and the protein are the log. Caramel barely enters it.
The amino acids do not run out, and that took me a while to understand
There is a puzzle in the measurements. Across a roast taken all the way to a very dark Italian level, total free amino acids fall by only about 42 percent. Meanwhile sucrose falls by more than 99 percent and water-soluble protein falls by about 55 percent. Stranger still, several individual amino acids, including glutamic acid, leucine, phenylalanine and valine, have been measured increasing during roasting.
A pool that is being consumed cannot grow. Unless something is refilling it.
The explanation that fits all three observations is that heat is hydrolyzing protein into its component amino acids at roughly the rate the Maillard reaction is consuming them. The free amino acid pool is not a fixed budget being spent down. It is a flow-through reservoir, drained from one end and topped up from the other, and the 11 to 15 percent of protein is the tank behind it.
I should be straight about the status of that claim. I have not found a paper that states it in one sentence. It is an inference that fits measurements from three separate studies, and I am presenting it as an inference rather than as a finding. If you read it somewhere stated as settled, be slightly suspicious.
What it does explain is why the amino side of the reaction never becomes the bottleneck. The sugar side does. Which puts the limiting step back on sucrose inversion, and makes moisture a more interesting variable than it usually gets credit for, because hydrolysis needs water. Work on water activity in green coffee has found the Maillard rate peaking around 0.70 and falling off above it. The reaction has an optimum rather than simply wanting less water, which is not how drying a batch is usually described.
What any of this changes about a cup you are actually drinking
Three things, and the first one is the one people feel most.
A roasted coffee holds essentially no sugar you can taste. The residual sugars sit well below the threshold at which a tongue detects sweetness, at every roast level. So the sweetness in a good cup is very largely an aroma effect, built from sweet-smelling volatile compounds, some of which raise perceived sweetness at concentrations too low to smell on their own. This part is not new and I am not going to pretend otherwise: Barista Hustle has a good piece on exactly this, and it is the best short treatment I know of. What is worth adding is the consequence for roast level. If sweetness came from caramelized sugar, darker would be sweeter. It is not. Sugar retention falls monotonically toward zero as the roast darkens, and perceived sweetness peaks somewhere in the middle and then drops away.
Second, the two cracks have different causes. First crack is a physical event: water in the seed flashing to steam and fracturing the structure. Second crack, later and louder, is carbon dioxide pressure in a bean that has become brittle. And a large share of that carbon dioxide comes from a side branch of Maillard chemistry called Strecker degradation, in which an amino acid reacts with one of those two-carbonyl intermediates and is stripped of a carbon as CO2. Which gives you a nice, slightly grim fact: the gas that makes fresh coffee bloom in your brewer, and the sound of second crack, are both byproducts of amino acids being taken apart. If you have ever waited a few days for a fresh bag to settle down, that is what you were waiting on.
Third, and this is where I have to concede something about our own bags. We grade roast level with an Agtron reading, which is a measurement of color, and that reading is where the roast level on a coffee's page comes from. Light, medium, medium-dark. It is a real and repeatable number and it is genuinely useful for consistency between batches of the same coffee.
It is also, by everything above, a measurement of the least informative thing in the bean. Color tells you melanoidins formed. It does not tell you which amino acids were spent getting there, whether the sugar inverted early or late, or how long the seed spent in the temperature band that builds the compounds you will actually smell. Two coffees can land on the same Agtron number by different routes and taste noticeably different. We publish the label, not the curve. If you want development time and drop temperature for a specific bag, we do not currently print either, and anyone who tells you a color number captures a roast is overselling a number.
What we do publish is a cup score on every coffee, and tasting notes on the single origins, which are an attempt at the outcome rather than the process. They have their own limits. But a cup score is at least a judgment about the cup, and the cup is the only place any of this chemistry becomes something you care about.
Questions we get about this
Is roasting coffee just caramelizing the sugar?
No, on both halves. The main browning reaction is the Maillard reaction, which needs an amino acid as well as a sugar and proceeds from around 50°C. Sucrose caramelization is a separate reaction that does not begin until about 170°C, by which point more than 99 percent of the sugar is already gone. Caramelization does happen, late and to a small extent. It is not what turns coffee brown.
What is the brown material in roasted coffee?
Melanoidins: large brown nitrogen-containing polymers formed at the end of the Maillard reaction, estimated at roughly a quarter of the roasted bean's mass. Reconstruction experiments found they are built largely on the bean's own cell wall sugars and protein, with protein supplying about 92 percent of their nitrogen, while added sucrose increased color without increasing the polymer itself.
Does sucrose react with the amino acids directly?
No. Sucrose is a non-reducing sugar, which means the chemical group the browning reaction needs is tied up in the bond holding its two halves together. It has to hydrolyze into glucose and fructose first. So sucrose is the reservoir that feeds the reaction rather than a participant in it, and the loose glucose and fructose already in the seed are what the reaction actually starts on.
Do darker roasts have more caffeine or less?
Caffeine is essentially unchanged across roast levels. Measured per gram, differences between light and dark are small and inconsistent. The differences people notice are mostly about density and how coffee is measured, since a darker bean has lost more mass and a scoop of it holds fewer beans.
Does a longer Maillard phase make coffee sweeter?
There is no good evidence for it. The difficulty is that you cannot extend that part of a roast in isolation, because stretching it necessarily lowers the rate of temperature rise, and a slower reaction over a longer time may produce a similar amount of product. Scott Rao has made this argument at length and it is worth reading. Published work comparing roast variables has found development time to be a stronger influence on flavor than the time taken to reach first crack, which is a different claim from longer being better.
Where this came from
The mechanism here is not ours and I am not going to dress it up as a discovery. The reaction chemistry belongs to food science, and the specific pieces I have leaned on hardest are Lothar Kroh's 1994 paper in Food Chemistry on caramelization, which is where the temperature floors come from, and Nunes and Coimbra's 2012 in-bean models in the Journal of Agricultural and Food Chemistry, which is the work that actually shows what melanoidins are assembled from. The free amino acid and protein figures come from a 2024 review by Portillo and Arévalo and from roast-series measurements published by Ali and colleagues in 2025 and Lu and colleagues in 2026. Barista Hustle's explainer and Sweet Maria's coffee library are the two places in coffee writing I would send someone next.
What is ours is smaller and more specific: we roast this, we watch what comes off the cooler, we grade it on a color scale we have just spent a section questioning, and we are the ones who have to decide where to stop. The chemistry is public. The judgment about a particular seed on a particular morning is the part you are actually buying.
If you want to taste the middle of the range
ASCENT is our house blend, a 50/50 pairing of washed Kenya and natural Ethiopia, and it sits in the medium band, which is roughly where perceived sweetness peaks. We prefer it as espresso. It cups at 92.3, the score is on its page, and it ships in 250g, 2lb and 5lb. We roast weekly.
Also worth reading: what a quaker is and why one ends up in a bag, and what a light roast actually is, which covers the roast-level end of this in more detail.
