Coffee fermentation: yeast took the cupping score from 80.3 to 82

Walk past a wet mill at seven in the morning and the tank looks like nothing is happening. Water, a bed of pale de-pulped beans, a skin of foam along one edge. The step carries a name that promises activity, and the trade explains it in one line: it is there to get the coffee mucilage off the bean. Two open access studies went looking at what is alive in that water, and the answer runs well past mucilage.

What washed coffee processing asks the tank to do

De-pulped ripe cherries go underwater for 6 to 72 hours before drying. That window comes from Schwan and Wheals in 2004 and still anchors the papers measuring wet process coffee today. The stated purpose is narrow. The fermentation is performed mainly to remove the mucilage layers still stuck to the beans, in the words of the team that weighed them.

That is the job description, not the job. The water is not sterile, and the thing being dissolved off the seed is sugar.

Who is actually in the water

A group working in Yunnan sequenced the whole community through a washed ferment of Coffea arabica from Puer City, sampling at 0, 12, 24 and 36 hours. Proteobacteria held between 54.36 and 61.60% of bacterial abundance in every sample. Firmicutes swung far wider, 8.57 to 41.25%, and the genus-level counts show why.

At the start Achromobacter and Trichocoleus sat almost level, at 12.95 and 12.88%. By 36 hours Achromobacter had climbed to 24.18% and Trichocoleus had fallen to 0.28%. Weissella went from 0.15 to 33.74%, Leuconostoc from nothing at all to 6.39%. Both of those are lactic acid bacteria and both arrived late. What the tank holds is a succession, not a standing population, which is one reason washed coffee processing tastes unlike the other routes through washed, natural and honey.

The counts climb and the sugar budget empties

The second study is a controlled one. Forty kilos of Coffea arabica var. Bourbon travelled from Zentveld's farm at Newrybar, New South Wales, to UNSW Sydney, where it fermented 36 hours at room temperature, mixed every 12. Total yeast rose from 5.5 to 7.3 log CFU per ml. Hanseniaspora uvarum went 5.1 to 7.1, Pichia kudriavzevii 4.8 to 6.5. Aerobic bacteria ran 5.1 to 7.9, lactic acid bacteria 5.3 to 7.3.

The coffee mucilage tells the other half. Sucrose dropped from 13.03 to 2.01 grams per 100 grams, glucose from 19.41 to 7.06, fructose from 27.02 to 4.05. Pitch a yeast starter culture in and the fall gets steeper: under H. uvarum, glucose was not detectable by 36 hours. Glycerol and lactic acid showed up where there had been none, 0.90 and 0.81 grams in the spontaneous run against 2.10 and 1.73 with both yeasts working together. Coffee fermentation is a sugar budget being spent, and those metabolites are the receipt. It is the same accounting that turns 6.23 kilos of cherry into a kilo of green coffee, measured a stage earlier.

What three Q-Graders made of it

Three testers holding Q-Grader certification scored the roasted result. The ferment pitched with both yeasts took the highest overall mark, 82.0, and that one is flagged significant at p below 0.05. The others are not. The published results put P. kudriavzevii at 81.5, the spontaneous control at 80.3 and H. uvarum at 79.8, and state plainly that the differences between those three are not statistically significant.

Which makes the honest reading a narrow one. One panel of three, one farm, one variety. The chemistry is where the authors commit further, reporting 2 to 3 fold higher concentrations of glycerol, alcohols, aldehydes, esters and organic acids in the inoculated beans.

Where the measurement stops

Both papers are candid about their own blind spots. The Sydney team ran plate counts and qPCR side by side and found the plates underestimated the population by roughly one log: a plate shows only cells that will still grow, while qPCR counts the DNA of stressed and dead ones too. Neither study reports pH at any time point, so the familiar line about the tank turning sharply acidic gets no support here.

The Yunnan work detected 2,548 metabolites and found 143 significantly changed between the start and 36 hours, 115 of them down and 28 up. Its fungal picture moves just as much: Cystofilobasidium ranged from 7.54 to 53.41%, Hanseniaspora from 3.45 to 26.80%, Aspergillus from 10.76% at 12 hours down to 0.90% at the end. One region, one variety, one tank. Specialty coffee varies more than that.

Roast date is printed on almost every bag. The fermentation window is on almost none. Ask a roaster how long the lot sat in the tank, and whether anyone steered it or left it to whatever was already in the water.