The research team, led by Xuechun Wang and Jian-Lin Wu, took a traditional Liu-Pao tea product as its starting material and fermented it further with one of three single strains: the golden flower fungus Eurotium cristatum (strain 8730), and two strains of the probiotic bacterium Bacillus coagulans, designated No. 10059 and JZ-1. Fermentation ran for seven days at 50 degrees Celsius, with leaf piles 80 to 100 centimeters high and moisture contents between roughly 10 and 25 percent. The choice of organisms was deliberate. E. cristatum is famed for the golden flower fungus aroma of brick teas, producing volatile compounds such as methyl salicylate and linalool oxide, while B. coagulans secretes proteases and lipases that can break down proteins and fats into flavor-active fragments. What remained unclear before this study was precisely how each strain drives the formation and transformation of the metabolites behind taste.

The first line of evidence came from human panels. Six trained tasters, working blind with randomly coded samples across three independent sessions, scored the teas on sweetness, umami, bitterness, sourness, astringency, and aged flavor using a ten-point scale under a protocol approved by an ethics committee. The unfermented control scored 5.50 for sweetness and 4.25 for bitterness. After fermentation with B. coagulans No. 10059, bitterness and astringency fell to 3.00 and 3.75 respectively, while the JZ-1 strain produced a more mellow profile with a noticeable cheese-like note. But the standout was E. cristatum: sweetness climbed to 7.50, umami jumped from 1.50 to 3.75, and bitterness dropped to 2.75, yielding the most pronounced mellow-sweet character of all the samples, accompanied by the fungus’s signature fungal aroma.

Human palates can be subjective, so the team turned to an SA-402B electronic tongue equipped with six lipid-membrane sensors that quantify sourness, bitterness, astringency, umami, saltiness, and sweetness, along with aftertaste attributes. The instrument confirmed the panel’s verdicts with striking precision. Sweetness values rose from 3.654 in the traditional tea to 4.102 and 4.302 in the two B. coagulans samples, and reached 5.989 in the E. cristatum tea, the highest of all. Bitterness moved in the opposite direction, falling from 3.484 to 2.500 in the fungal sample. Umami intensity increased across all fermented teas, from 2.776 in the control to 4.687 in the E. cristatum product. Sourness, saltiness, and astringency aftertaste registered below their tasteless thresholds and were excluded, leaving sweetness as the single most discriminative taste parameter between strains.

To trace these sensory shifts back to chemistry, the researchers quantified the tea’s major constituents. Total protein dropped significantly in the B. coagulans teas, most notably with JZ-1, consistent with the bacterium’s proteolytic enzymes, but rose significantly in the E. cristatum sample, likely reflecting fungal protein secretion and the conversion of insoluble proteins into soluble forms and peptides. Total polyphenols, the most abundant chemical class in the tea, fell significantly in the bacterial fermentations, from about 362 milligrams per gram to roughly 297, while remaining unchanged in the fungal sample. Flavonoids, a polyphenol subclass, trended upward with E. cristatum, and soluble sugar showed no significant change in any group. These bulk measurements hinted that each microbe was running a fundamentally different biochemical program on the same tea base.

The centerpiece of the study was a feature-based molecular networking approach, or FBMN, run on the GNPS platform and coupled to ultra-high-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry. After preprocessing the raw data with MS-DIAL, the team generated a molecular network of 18,342 mass spectral nodes, each representing a detected chemical feature, clustered by spectral similarity and visualized in Cytoscape with pie charts showing each compound’s abundance across samples. The strategy allowed the researchers to see at a glance which metabolites were associated with which fermentation strain, dramatically accelerating the discovery of microbial-derived compounds that conventional targeted analysis would have missed.

In total, 358 compounds were tentatively identified, including 249 flavonoids, 21 peptides, 6 amino acids, 32 lipids, and 50 alkaloids or nucleosides, with 10 compounds flagged as potentially novel. Among the flavonoids were catechins, flavonol glycosides, biflavonoids, and a distinctive group called flavan-3-ol B-ring fission analogs, or FBRFAs, including teadenol A, planchol A, and fuzhuanin A. The network also revealed a cluster of ethyl-pyrrolidinone-substituted flavonoid alkaloids, the puerins, and led to the characterization of a potential new compound, eps-teadenol, whose fragmentation pattern matched that of known pyrrolidinone-substituted flavan-3-ols. From this cluster, ten previously unreported eps-FBRFAs were tentatively annotated. On the peptide side, the network yielded 16 linear peptides, including the pentapeptide Gly-Pro-Phe-Pro-Leu/Ile, and five indole-containing cyclic dipeptides such as neoechinulin A and echinulin, three of which had never been reported in Liu-Pao tea.

Statistical modeling then separated the strain-specific signatures. Orthogonal partial least squares discriminant analysis identified 71 differential compounds in the E. cristatum tea, 33 up-regulated and 38 down-regulated. Nine FBRFAs and three eps-FBRFAs rose significantly, and the prenylated cyclic peptides neoechinulin A, echinulin, and tardioxopiperazine A surged from trace levels to marked accumulation, a hallmark of fungal secondary metabolism. Critically, non-ester-type catechins such as epicatechin and epigallocatechin increased dramatically while their gallate-esterified counterparts, including EGCG and ECG, declined, suggesting that fungal enzymes hydrolyze esterified catechins into their non-esterified forms. In the B. coagulans teas, the story was proteolytic: 38 differential compounds emerged in the JZ-1 fermentation, 15 of them up-regulated, including four amino acids and eleven peptides, with representative peptides such as Pro-Glu-Val reaching their highest levels in the JZ-1 group.

Linking chemistry back to the palate, redundancy analysis showed that the first two axes explained over 93 percent of the variance in flavor attributes, with sweetness aligning strongly with non-esterified catechins, FBRFAs, and eps-FBRFAs, while bitterness tracked with the ester-type catechins. This fits established taste science: non-ester catechins like EGC and EC contribute to sweet aftertaste, whereas gallated catechins are among the most bitter and astringent compounds in tea. The linear peptides enriched by B. coagulans correlated positively with aged flavor, plausibly explaining the bacterial teas’ deeper aged character. Pushing the mechanistic question further, the team docked twelve FBRFAs and eps-FBRFAs into the Venus flytrap domains of the human sweet taste receptor TAS1R2/TAS1R3, the same pocket that recognizes sucralose. All twelve bound with predicted energies between minus 7.1 and minus 9.0 kilocalories per mole, and eleven contacted key residues including Tyr-103, Asp-142, and Asp-278. The authors are careful to note that docking alone does not prove receptor activation, and functional validation remains to be done.

Finally, the study proposed a biosynthetic route for the eps-FBRFAs, a class reported here for the first time in Liu-Pao tea. The pathway begins with the de-esterification of gallated catechins such as EGCG into non-ester forms, followed by cyclic peroxide formation, meta-cleavage, reduction, oxidation, and lactonization to yield FBRFAs. In parallel, theanine undergoes Strecker degradation to aldehyde intermediates that cyclize into ethyl pyrrolidone, which then conjugates with FBRFAs at the C-8 position. Supporting this route, theanine levels fell most sharply in the E. cristatum fermentation. Because eps-FBRFA enrichment appeared only in the fungal samples despite identical fermentation conditions, the strain itself appears essential to the transformation. Together, the findings give tea producers a molecular toolkit: E. cristatum for sweetness and reduced bitterness, B. coagulans for aged flavor depth, and a validated analytical pipeline for engineering the next generation of fermented teas.

Subject of Research: Microbial fermentation and flavor metabolites in Liu-Pao dark tea

Article Title: Integrated sensory evaluation and feature-based molecular networking elucidate flavor improvement and associated metabolites in dark tea fermented by Eurotium cristatum and Bacillus coagulans

Article References: Integrated sensory evaluation and feature-based molecular networking elucidate flavor improvement and associated metabolites in dark tea fermented by Eurotium cristatum and Bacillus coagulans. (n.d.). https://doi.org/10.1016/j.fochx.2026.104544

Image Credits: AI Generated

DOI: 10.1016/j.fochx.2026.104544

Keywords: dark tea, Liu-Pao tea, Eurotium cristatum, Bacillus coagulans, fermentation, metabolomics, molecular networking, flavor chemistry, catechins, peptides, sweet taste receptor, food science

The heart of the device is MspA, a porin from the soil bacterium Mycobacterium smegmatis that has already earned a storied reputation in DNA sequencing. MspA forms a stable, conical pore roughly one nanometre wide at its constriction, a scale at which a single small molecule can obstruct the flow of ions and produce a characteristic electrical signature. When a voltage is applied across a membrane containing the pore, analytes that wander into the aperture transiently block or modulate the ionic current, and the resulting blips carry fingerprints of the molecule’s size, shape, charge and chemistry.

The problem has always been versatility. Nanopores are exquisitely selective, which is a virtue for detecting one target molecule but a liability when the target is unknown. Different chemical classes interact with the pore environment in different ways: saccharides are neutral and hard to trap, amino acids span a dramatic range of charge and hydrophobicity, and nucleotides carry dense negative charges that make them rush through too quickly to be read. Earlier engineered pores succeeded with individual analyte classes—discriminating all twenty proteinogenic amino acids in one design, or distinguishing monosaccharides in another—but no single pore had tackled all of them at once.

The new work solves this by grafting a chemical adapter into the pore. The team modified MspA with a maleimido-C2-FPBA moiety, a benzaboronic acid derivative attached through a short linker to a defined site inside the pore lumen. Boronic acids are famous in chemical biology for forming reversible covalent complexes with cis-diols, the pairing of hydroxyl groups found abundantly on sugars. But the adapter does more than catch carbohydrates. Inspired by iminoboronate chemistry, the reversible interaction between boronic acids and nitrogen-containing functional groups allows the adapter to transiently capture amines as well, slowing the passage of amino acids and peptides so that their signals can be recorded.

The result is a pore that no longer lets any of these small molecules simply tumble through. Instead, each analyte is repeatedly captured, held and released at the adapter site, producing long trains of current fluctuations rather than a single fleeting blip. Amino acids produce residence events whose depths and durations reflect their side chains; nucleoside monophosphates generate distinct blockade patterns shaped by their base and phosphate; saccharides, bound through their diols, yield slow, stuttering signals; and peptides produce composite signatures reflecting both their terminal amines and their residue composition. Crucially, all four classes can be present in the same solution and still be told apart, because their event statistics cluster in separate regions of feature space.

Distinguishing those clusters is where machine learning enters. The researchers extracted features from thousands of individual events—mean and variance of the blockade current, dwell times, recurrence rates and higher-order statistics of the fluctuation patterns—and trained classifiers on labelled mixtures. Using a few-shot learning strategy, which requires only a small number of labelled examples per analyte, the system assigned unseen events to the correct analyte with 98.7% accuracy across the four chemical classes. The approach is notable for its data efficiency: instead of demanding enormous training sets, the engineered pore’s physically distinct binding chemistry produces such characteristic signals that a handful of reference measurements per molecule suffices.

The most striking demonstration involves glycopeptides, molecules that marry a peptide backbone to one or more covalently attached glycans. Glycopeptides are central to biology—most secreted and membrane proteins carry them—and their analysis is a major bottleneck in proteomics, typically requiring enzymatic release of the sugars and elaborate liquid chromatography tandem mass spectrometry workflows. Because the FPBA adapter engages the cis-diols of the glycan while the pore constriction senses the peptide, the sensor reads both parts of the molecule in one event. The team showed that the device could determine the compositional makeup of native, unlabelled glycopeptides, distinguishing variants that differ in their sugar content without any prior chemical derivatization.

Experts in bioanalytical chemistry will recognize how much engineering subtlety underlies this apparent simplicity. The site-specific attachment of the adapter required a genetically introduced handle in the MspA protein, and the linker length and chemistry had to be tuned so that analytes of wildly different sizes—from a single amino acid of around one hundred daltons to glycopeptides several times larger—could all access and interact with the boronic acid. The ionic current readout itself is straightforward, but converting raw fluctuation trains into reliable chemical identities demands rigorous control of pH, salt concentration and voltage, conditions the authors establish and characterize in the study.

The broader significance lies in what a practical four-class nanopore sensor could enable. A benchtop or even portable device that identifies metabolites, nucleotides, sugars and peptide fragments from a crude mixture would be transformative for point-of-care diagnostics, metabolic flux studies, quality control in biopharmaceutical manufacturing and the emerging field of glycomics. It also complements the rapid progress in nanopore peptide sequencing, where related engineered pores have recently been used to read stepwise-shortened peptides, and with nanopore single-molecule chemistry more generally, which now extends far beyond its nucleic-acid origins. A universal small-molecule reader, in other words, no longer seems fanciful.

Challenges remain before the technology migrates from research lab to routine use. Real biological samples contain hundreds of analytes, some present at vanishingly low concentrations, and the classifier’s performance on such dense, unbalanced mixtures will need to be demonstrated. Sensor lifetime, throughput and the standardization of pore fabrication will also matter for adoption. But the conceptual hurdle has been cleared: a single chemically versatile protein pore, read by a modest machine-learning model, can simultaneously name molecules from four different chemical worlds. As nanopore sensing matures, the humble ionic-current trace is quietly becoming one of the most information-dense signals in all of analytical science.

Subject of Research: An engineered MspA nanopore sensor that simultaneously identifies amino acids, nucleoside monophosphates, saccharides and peptides, including native glycopeptides, using machine-learning classification.

Article Title: A versatile nanopore identifies four classes of analytes simultaneously

Article References: A versatile nanopore identifies four classes of analytes simultaneously. (2026). Nature Biotechnology. https://doi.org/10.1038/s41587-026-03322-x

Image Credits: AI Generated

DOI: 10.1038/s41587-026-03322-x

Keywords: nanopore sensing, MspA, biosensor, machine learning, amino acids, saccharides, peptides, nucleoside monophosphates, glycopeptides, single-molecule detection, bioanalytical chemistry, chemical biology