The research, led by Huanbo Cheng, Zhian Huang, Lifei Hu, and Yuanming Ba, focused deliberately on the classical heartland of gouqi cultivation: Ningxia, Gansu, and Qinghai Provinces. Ningxia alone has more than 600 years of cultivation history and remains the authentic producing region against which all others are judged. The team collected 90 authenticated batches of dried ripe fruits of Lycium barbarum L., harvested in September and October of 2024 from commercial plantations in Ningxia (25 samples), Qinghai (25 samples), and two cities in Gansu, Jiuquan and Baiyin (20 samples each). All samples were taxonomically verified by Prof. Keli Chen of Hubei University of Chinese Medicine. The fresh berries were gently dehydrated with convective airflow at 50 degrees Celsius until their moisture content dropped below 13 percent, mimicking standard commercial processing. Rather than selecting a single cloned cultivar, the researchers deliberately sampled locally dominant cultivars from multiple orchards in each region, a choice that reflects real agricultural practice but introduces cultivar-related variation the analysis had to overcome.

The chemical analysis proceeded on two parallel fronts. The first concerned functional metabolites, the bioactive compounds responsible for gouqi’s health reputation. Using ultra-high performance liquid chromatography coupled with a photodiode array detector, the team quantified seven phenolic compounds: 4-hydroxybenzoic acid, chlorogenic acid, scopoletin, 4-coumaric acid, ferulic acid, rutin, and kaempferol. Betaine, a signature low-molecular-weight constituent linked to hepatoprotective and antihyperglycemic effects, and ascorbic acid 2-beta-glucoside (AA-2βG), a water-soluble vitamin C derivative, were measured using Chinese Pharmacopeia methods and national drug standards. Colorimetric assays quantified polysaccharides, total flavonoids, and total phenolics, the macromolecular and bulk antioxidant fractions. Method validation was rigorous: all calibration curves exceeded an R-squared of 0.9992, chromatographic resolution was greater than 1.5 for the seven phenolics, and precision, repeatability, and stability relative standard deviations all stayed within accepted limits, with recoveries between roughly 98 and 103 percent.

The second front targeted something that has long been overlooked in gouqi quality research: the inorganic elements within the fruit. Previous studies paid close attention to color, shape, taste, and organic actives, but mineral composition remained underexplored. Using inductively coupled plasma mass spectrometry, the researchers measured 38 elements, including 17 rare-earth elements such as lanthanum, cerium, neodymium, and ytterbium, plus trace actinides like thorium and uranium. Sample preparation involved microwave-assisted digestion of powdered berries in nitric acid, with a staged temperature program reaching 180 degrees Celsius. Internal standards containing germanium, indium, rhodium, and bismuth were introduced online to monitor instrumental drift, and a pooled quality-control sample, prepared by combining equal aliquots of all 90 samples, was analyzed after every 15 test samples to track batch consistency. Validation showed R-squared values above 0.9994 for all 38 elements, relative standard deviations below 3 percent, and recoveries between 92.6 and 105.6 percent.

The element results revealed dramatic geographic contrasts. Potassium, sodium, magnesium, and calcium dominated at gram-per-kilogram levels, with Gansu Baiyin and Ningxia samples showing significantly higher potassium, sodium, and magnesium, while calcium peaked in Qinghai at 0.87 grams per kilogram. Sixteen elements, including iron, aluminum, zinc, manganese, copper, strontium, and lithium, were found at milligram levels. Baiyin and Ningxia samples consistently showed elevated aluminum, manganese, iron, barium, and cobalt, which the authors attribute to the regions’ geographic proximity and similar soils. Jiuquan stood out for its lithium, at 3.35 milligrams per kilogram, roughly five times higher than neighboring regions, while showing the lowest barium, iron, and aluminum. Qinghai samples had the lowest silver, nickel, and arsenic but the highest uranium and cadmium. The rare-earth elements, which coexist in minerals due to their similar atomic structures, grouped tightly together across all regions, with Baiyin samples generally showing the highest concentrations and Jiuquan the lowest. Cerium, lanthanum, neodymium, and yttrium were the most abundant rare earths overall.

The metabolite results told a more nuanced story. Qinghai samples contained the highest rutin content, at 252.21 micrograms per gram, possibly driven by the intense ultraviolet radiation at high altitude, and the lowest kaempferol. Polysaccharides were significantly lower in both Gansu regions. Ningxia samples led in betaine at 9,608 micrograms per gram, followed closely by neighboring Baiyin. AA-2βG was significantly depleted in Jiuquan. Yet when the researchers applied unsupervised chemometrics, hierarchical cluster analysis and principal component analysis, to the 12 functional metabolites alone, the four regions blurred into overlapping clusters. A supervised orthogonal partial-least-squares discriminant analysis fared better when the two Gansu cities were merged: with R-squared X of 0.657, R-squared Y of 0.772, and Q-squared of 0.721, the model separated Ningxia, Gansu, and Qinghai with confidence, achieving an area under the receiver operating characteristic curve above 0.98. Still, the two Gansu subregions remained stubbornly indistinguishable by metabolites alone.

The inorganic elements told an entirely different story. Hierarchical clustering cleanly divided all 90 batches into four regions, and principal component analysis with six components explained 88.11 percent of the cumulative variance, separating Baiyin, Ningxia, and Qinghai distinctly. The supervised OPLS-DA model on elements achieved remarkable performance, with R-squared X of 0.810, R-squared Y of 0.910, and Q-squared of 0.888, and 200-permutation tests confirmed no overfitting. Sixteen elements emerged as key discriminators with variable importance in projection scores above 1, including lithium, nickel, lead, zinc, magnesium, chromium, copper, cobalt, praseodymium, neodymium, scandium, strontium, aluminum, lanthanum, uranium, and gadolinium. The ROC analysis yielded a perfect area under the curve of 1, indicating flawless classification of the elemental fingerprints.

To push accuracy further, the team deployed four machine learning classifiers spanning different algorithmic paradigms: a backpropagation neural network representing shallow learning, a one-dimensional convolutional neural network as a lightweight deep feature extractor, a support vector machine using kernel mapping, and a random forest as an ensemble method. Hyperparameters were tuned by grid search under fivefold cross-validation, and the Kennard-Stone algorithm selected a representative 20 percent test set from raw, unscaled data before any preprocessing, ensuring strict prevention of data leakage. On functional metabolites, the models struggled, with test-set accuracies ranging from 72.22 percent for the backpropagation network to 88.89 percent for support vector machine and random forest, and confusion matrices showing that samples from the two Gansu cities were misclassified in every model, likely due to their shared provincial climate. But on inorganic elements, all four models achieved 100 percent accuracy on both training and test datasets, with stable AUC values of 1.0000 and no misclassifications whatsoever.

Beyond classification, the study uncovered intriguing correlations between elements and bioactives. Across 1,225 pairs analyzed, 842 showed positive correlations and 642 were statistically significant. Rare-earth elements showed strong positive correlations with 4-hydroxybenzoic acid, AA-2βG, and polysaccharides, but negative correlations with 4-coumaric acid and kaempferol. Aluminum and iron correlated strongly and positively with rare-earth elements, while lithium showed an extremely significant negative correlation with them. The authors hypothesize that low-dose rare-earth elements, known to promote seed germination, enhance photosynthesis, and improve disease resistance, may influence the accumulation of polysaccharides, betaine, and flavonoids through modulation of calcium-reactive oxygen species signaling, phenylalanine ammonia-lyase activity, or photosynthetic efficiency, though they caution that these ion-mediated mechanisms remain speculative and require controlled experiments to establish causality.

The findings carry clear practical weight. China produced 421,600 tons of gouqi berries in 2021, with exports exceeding 10,000 tons and an average export price of 7.21 dollars per kilogram, and quality and price vary substantially by region. A reliable, scientifically grounded authentication method could protect the premium value of Ningxia gouqi and support geographical certification systems. The study’s innovation lies in being the first to incorporate rare-earth elements into gouqi provenance discrimination, exploiting their role as plant growth regulators and their potential interactions with organic ligands in plant tissues. The authors caution that the models reflect internal validation only, with 90 batches from a single harvest season, and that generalizability to independent external samples from different years or seasons remains to be established. They plan to continue collecting multi-batch, multi-season samples for prospective validation. Even so, the message is clear: when it comes to tracing the origin of one of the world’s most celebrated superfoods, the minerals absorbed from the soil speak with far greater precision than the fruits’ celebrated chemistry, and machine learning can decode that message flawlessly.