What really happens during extraction? There are some things we know, and some things that are still a little muddled. This week I did a few experiments in hopes of tightening up our understanding of how extraction works.
What we know about extraction:
We know that water is an excellent solvent.
Water is sometimes called the “universal solvent” because it dissolves more substances than any other liquid. Its molecular structure (two hydrogen atoms attached to an oxygen atom, H2O) creates an uneven distribution of electron density, making it polar. This polarity makes it easy for other molecules to break up and dissolve in water. The partial positive and negative charges on opposite sides of the H2O molecule mean that it’s attracted to a wide variety of molecules, easily dissolving their structures. Hot water is an even more efficient solvent because the extra energy from heat makes the polarized molecules vibrate faster, making them even more attractive (insert joke here).
We know what compounds inside the coffee bean are water soluble.
These soluble compounds include fruit acids, caffeine, lipids, melanoidins, carbohydrates, and plant fiber. Each of these extracts into the brew liquid at different times and impart different flavors to the resulting cup. Only 30% of a coffee bean is water soluble, and only about 20% of these solubles taste good in a brew—the rest can add bitter or papery off-flavors to the coffee liquor. Luckily for us, the desirable solubles extract first and the unpleasant ones put up a little more of a fight before dissolving into water.
We know that different compounds come out during different phases of the brew.
Fruit acids and caffeine dissolve the easiest, washing into the brew first. When isolated, this portion of the brew can taste bright and thin. Next come the lipids and fats, which release from the coffee cell as an emulsion and enter the brew liquid that way. These have a huge impact on perceived mouthfeel, but a paper filter will prevent most of these lipids from entering the cup. Last to dissolve are melanoidins and carbohydrates or plant matter. Melanoidins are a byproduct of the Maillard effect and are responsible for the brown color of coffee beans and coffee liquid. Carbohydrates and plant fibers make up more than half of the total compounds in a coffee bean, but again not all are water soluble. These last two are not always delicious, adding bitter, papery, or earthy flavors, but they are also vitally important to creating balance in the final cup.
There’s a simple experiment that can help illustrate when different compounds get dissolved into a brew. If you haven’t done this yet (and you have an espresso machine at your disposal), go try it right now!
Prepare a shot, and instead of pouring into a single vessel, prepare up to 6 demitasses, rocks glasses, or spoons. As the shot begins dripping, allow it to run into the first vessel. After 5-10 seconds (depending on how many vessels are being used), switch out the cups and let the espresso fall into a fresh cup. Continue this process, filling the cups in order, until the espresso has reached the end of its dial.
When placed in order, the different phases of espresso extraction will look radically different, going from dark and sludgy to light and thin. The difference in taste will also be considerable. I did a quick version of this experiment with just 3 vessels—it’s difficult to get enough liquid into each cup with more vessels than this—and found some interesting readings.
The graph shows that extraction starts off incredibly high at the beginning of the shot (I switched vessels after the first 5g of espresso liquid had pulled), and then drops drastically in the second and third phases of the espresso.
I began to wonder: what would happen if we did a similar experiment with other brew methods? Espresso is an extremely compressed form of extraction; what would it look like to measure TDS and taste differences in extraction in a more prolonged brew method?
I wanted to know:
At what rate are compounds being extracted from the beans into the brew?
For this, I needed to know the rate of extraction into a brew, total extraction over total brew time, and how extraction changed proportionally to total beverage mass as the brew progressed.
This, of course, meant data collection. I began by doing a V60 pour over, first by dialing the coffee and taking sensory notes as well as TDS and extraction readings of said dial for a baseline reading. My recipe called for 22.22g of coffee and 400g of water in a 5:00 minute brew with a 30 second bloom. I chose a long brew ratio, thinking that extending the brew process might allow me to collect more data.
Every 30 seconds, I moved the V60 dripper over to pour into a new carafe. I also added 50g doses of water during the 30 second intervals. This created 8 carafes full of a small amount of brewed liquid representative of every 30 second stage of the brew.
This technique created certain logistical and consistency issues that I did my best to mitigate. Moving the dripper from carafe to carafe created much more agitation than during my dial. I was also only able to control how much water I added at every interval, not how much water passed through the dripper and into the carafes every 30 seconds. This meant that there were varying amounts of liquid in each carafe, with the most in the last one—I poured the last 50g water dose between the 3:30-4:00 minute mark, and then simply allowed all the remaining brew liquid to drain into the carafe for my final sample.
The results were fascinating, both from a sensorial perspective and based on the refractometer readings. In the graph we see that an astounding amount of extraction happened just in the first 30 seconds, exclusively during bloom. Extraction continued to increase until the 1:00 minute mark, but then fell sharply for the remainder of the brew.
In the cup we noticed significant changes from batch to batch . As expected, the first minute or so of brew was full of bright soaring acids—the fruit acids that dissolve out of the coffee fairly easily. As extraction continued we began to taste brown sugar, molasses, and toffee notes. After the 2:00 minute mark, the coffee began to become herbal, bitter, and a little astringent, before becoming very watery and thin in the final minute of brewing.
What I find most interesting about the table above is that the number of descriptors per brew phase closely matches the graph line showing TDS—there’s a distinct peak in flavor notes at the 1:00 minute mark where TDS was highest, and the number of flavors perceived by my cuppers decreased continually after that.
It became clear that I needed the brew weights of each 30 second phase in order to fully understand the rate at which extraction was happening in the total brew. I did two more pour overs and weighed the output of each phase, taking TDS readings and collating information to determine total brew weight, percentage of total brew weight that each phase represented, and was even able to calculate TDS for the total brew based on these numbers.
This second graph shows how total TDS of the final beverage changes over time. Here it becomes clear that although rate of extraction decreases significantly after the first 90 seconds of brewing, over all TDS steadily increases throughout brew time. Although the first few phases have very high TDS readings, the brew weight of these first 90 seconds of brewing form such a small part of total brew volume that their contribution is diluted into the fully brewed cup.
All of these show that the more than half of the total solids are already in the beverage by the 90 second mark, even though only 20% of beverage mass has flowed out.
This experiment was informative, but I wanted to know more. How would extraction change over time if all the water was in contact with all the coffee throughout the total brew time? This called for a full immersion brew experiment.
I started with a French press. I figured I could use syringes to dip below the layer of crust, and then use the same syringe filters we usually use for espresso to filter the brew. This proved surprisingly challenging. Worst of all, the data this experiment provided didn’t seem very useful. Like the pour over experiment, there was an increase in TDS extraction from 0:30 to 1:00 minute, but from there on out the TDS readings plateaued completely. There was no change in the readings from the 1:00 mark all the way to the end of brewing at 4:00 minutes.
And yet, when we tasted each 30 seconds of brew, we noticed drastic changes in flavor development from cup to cup. The samples from the 1:00 of brewing were sour and thin, while samples from the middle of the brew were incredibly floral, with a hint of alcohol that made us think of floral extract. How were these drastic changes in flavor happening if TDS was remaining totally static from the 1:00 minute mark onward?
I cast about for an alternative full immersion brew method that might allow more flexibility and landed on a Clever dripper. Instead of moving the brew device from carafe to carafe as I did with the V60, I chose to use eight drippers to brew simultaneously, and stop them 30 seconds apart. I stopped the first brew at 0:30 seconds and let the total brew water drain out, then stopped the second brew at the 1:00 mark and let the total brew water drain out, continuing the pattern until I had eight brews. An added benefit of this method was that I had plenty of brew liquid to allow my cuppers to taste and analyze the different phases of extraction.
As is often the case with these kinds of experiments, the data still didn’t give me the results that I wanted. Both full immersion brew methods created a short ramp up in TDS that plateaued very early in the brew. The clever showed slightly more variance, but even so, the variance between the first phase of brewing at 0:30 seconds and the last at 4:00 was only 0.29 TDS—the coffee had already extracted to 1.06 within the first 30 seconds of brewing.
There were differences in flavor between the eight brews, as evidenced by our sensory notes in the table below. Since the coffee took an extra minute or more to drain through the Clever dripper, the first few brews don’t display as drastic a difference as some of my other experiments. Another interesting development was that while my cuppers loved the original dial, they were much less favorable to the eighth cup on the table—which was brewed to the same parameters and had the same TDS readings as the dial. I think this has to do with the biases created from tasting all previous cups, and wonder if the results would have been different had I randomized the order.
What can we learn from these experiments?
The data I collected drove home some important points:
The majority of extraction happens very early on in the brew, and usually peaks at the 1:00 mark.
This is not really news—we already knew that most of the good stuff (fruit acids and caffeine) came out very early on in extraction, followed by lipids and fats, which add more mouthfeel and may offer the beginnings of brown sugar notes that melanoidins continue to bring until becoming bitter. Plant fibers and carbohydrates are the last to emerge from the bean and into the brew, and too many of these compounds could ruin the cup. However, some bitterness can help balance out the flavors, so having some of these in your brew is useful.
Despite all this information, I was still surprised to see just how much is extracted within just the first 30 seconds for brewing. During sensory analysis of these experiments, one of my colleagues looked at me, eyes wide, and said “Did you realize that we were really brewing a concentrate and then diluting it?” Although a slight oversimplification, he’s definitely onto something. The backbone of a cup of the brewed coffee comes out almost immediately, in a concentrated, imbalanced, aggressively tart form. The rest of the brew really serves to provide structure and balance to those intense first drops.
TDS readings can’t account for the deliciousness of a brew.
The full immersion brews showed that TDS reached a plateau after the first minute. The pour over experiment, which had the benefit of adding fresh, hot water to the grounds, had a distinct peak at the one minute mark and a dramatic fall from there. Both these brew methods indicate that by 1:00 of brewing, most of the solid compounds that are going to get into the brew have already migrated there.
And yet, different phases of brewing tasted radically different—even if they had the exact same TDS reading.
What accounts for this incredible difference in flavor from phase to phase? One answer is volatile aromatic compounds, which can be measured somewhat successfully with Nuclear Magnetic Resonance Spectroscopy. However, coffee has so many aromatic compounds (over 800 at last count) that we can’t measure them all just yet.
There’s an important lesson here: numbers can’t actually tell us everything we need to know about our brewed coffee. It’s easy to fall back on refractometer readings to dial in a coffee—get it within an “ideal range” of TDS and percent extraction and leave it at that. These findings prove what we’ve always known: dialing should be a sensory experience, not just a technological one. If we rely solely on numbers to determine our brewing parameters, we’ll never really do the coffee justice. Coffee is a craft after all. Science is important, even essential, to understanding our product and the limits we can take it to, but it simply can’t replace the human, sensory component that each producer, roaster, and barista offers to the brew.
There are myriad tools baristas can use to make their jobs easier and their extraction more precise, but none of them can do what an educated palate can. The next time you find yourself in front of a brewing device—be it espresso, pour over, or full immersion—remember: the most important calibration tool at your disposal is your own palate. Trust it to tell you what the coffee has to offer and what you can do to get it there.