Geochemical differences and sources of mineral elements in soil and coffee beans from pu’er and Baoshan, China

Abstract

Background

Yunnan is a major coffee region in China. This study explores mineral sources and bioavailability to identify geochemical indicators for origin traceability.

Methods

We analyzed 85 soil, 52 coffee beans, and 4 bedrock samples from Pu’er and Baoshan, Yunnan. Eleven elements (Cr, Cu, Mn, Zn, Ca, K, Mg, Fe, Pb, Mo, As) were measured by coupled plasma mass spectrometry, along with soil pH and organic matter. Enrichment factors, bioconcentration factors, principal component analysis, and partial least squares regression were applied to assess elemental mobility, sources, and traceability.

Results

K, Ca, Mg, Cu, and Mo showed high bioavailability with greater uptake in Pu’er. Lead, As, Mn, Zn, Fe, and Cu accumulated more in Pu’er beans, while Cr was enriched in Baoshan. Chromium, Mo, and As derived from human activities, whereas Mg, Ca, and K from weathering rock. Elemental sources in Pu’er involved weathering, industry, and agriculture; in Baoshan, coal combustion, mining, and agriculture dominated. Three latent variables were extracted, with higher explanatory power in Pu’er (Xp = 78.07%, Yp = 38.69%) than Baoshan (Xb = 46.26%, Yb = 29.49%). The LV1 axis clearly distinguished origins, identifying Cr, Cu, Ca, and Mg as key soil indicators for coffee traceability.

Conclusion

All toxic elements were below safety limits. Pu’er showed higher Pb and As accumulation, while Baoshan had more Zn and Mn. Natural weathering mainly contributed to Mg, Ca, and K; human activities influenced Cr, Mo, and As. Partial least squares regression effectively distinguished origins, highlighting Cr, Cu, Ca, and Mg as effective markers for distinguishing coffee origins.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12870-026-08998-6.

Introduction

Coffee, renowned for its pleasant aroma and flavor, has become one of the most beloved beverages worldwide. With a history spanning over a century, it has evolved into an integral part of human daily life [1]. As China’s coffee market continues to expand, increasing attention has been directed toward improving coffee yield and quality by exploring the interactions between coffee plants and their cultivation environments [2, 3]. Yunnan Province, located near the Tropic of Cancer, offers ideal geographical and climatic conditions for coffee cultivation. Currently, Yunnan contributes approximately 98%–99% of China’s total coffee production, predominantly of the Arabica variety (Coffea arabica L.). Pu’er and Baoshan represent the two largest coffee-producing regions in China. These regions, situated at altitudes between 900 and 1750 m, are characterized by large diurnal temperature variations, volcanic red soils, and abundant rainfall—factors that are highly favorable for the growth of high-quality Arabica coffee [4–6].

Among the mineral elements (MEs) analyzed in this study—including K, Mg, Ca, Cr, Cu, Mn, Zn, Fe, Pb, Mo, and As, the elements K, Mg, Ca, Cu, Mn, Zn, Fe, and Mo are generally considered beneficial to plant growth, whereas As, Pb, and Cr are commonly classified as potentially toxic elements (PTEs) that may pose risks to plant health [1, 7].These MEs in soils may originate from either natural (lithogenic, non-anthropogenic) and anthropogenic sources, including agricultural, industrial, or atmospheric inputs [8–11]. Natural weathering is considered the dominant mechanism for elemental release in large-scale native soils [12].

However, the long-term accumulation of these elements in agricultural soils may increase their transfer into the human food chain [13], raising growing concerns about the origin, transformation, and fate of trace elements in soils [14, 15]. This is particularly relevant for cropland soils, where food safety and quality are directly impacted [16–18]. Principal Component Analysis (PCA) has been widely used in previous studies to identify the sources of soil-borne MEs [19]. In the context of coffee cultivation, understanding the sources of MEs in plantation soils may not only help improve soil management but also provide a chemical “fingerprint” for origin traceability of coffee beans [20, 21]. Furthermore, analyzing the mineral nutrient profiles of both soils and beans can offer a scientific basis for optimizing coffee production regions [22, 23]. Given that the relationship between soil and bean elemental concentrations is typically multivariate and exhibits strong linear correlation, multicollinearity may arise during regression modeling. Multivariate statistical approaches are particularly effective in handling such complex interdependencies [24, 25]. Among them, Partial Least Squares Regression (PLSR) stands out as a hybrid method that integrates the advantages of PCA, multiple linear regression (MLR), and correlation analysis [26]. PLSR not only reduces dimensionality but also improves the predictive power of regression models [27], making it highly suitable for this study’s goals—to trace the sources of soil MEs and explore their relationships with coffee bean composition.

Although numerous studies have been conducted on coffee cultivation in Yunnan, most have focused on flavor characteristics [28], quality evaluation [29], ecological adaptability [30], and industrial innovation [31]. However, there is still a paucity of research investigating the accumulation characteristics of mineral elements in both soils and coffee beans across different production regions. In particular, few studies have simultaneously examined the correlations between soil nutrients and elemental composition in beans, especially within the context of region-specific differentiation. This study addresses this gap by investigating 11 mineral elements in the soils and Arabica coffee beans of two key coffee-producing regions in China—Baoshan and Pu’er. The objectives are to (i) characterize the enrichment patterns of mineral elements in soils and beans, (ii) identify the geogenic and anthropogenic sources of these elements in soils, (iii) determine the key soil nutrient factors influencing mineral accumulation in beans, and (iv) apply partial least squares regression (PLSR) with variable importance in projection (VIP) analysis to identify the dominant soil predictors for bean mineral composition. The findings aim to provide a quantitative framework for understanding regional differences in elemental transfer from soil to bean and to support the development of a robust elemental fingerprinting system for coffee origin traceability in China.

Materials and methods

Sampling and analysis

In 2024, during the coffee cherry ripening season, sampling was conducted in two representative coffee-growing areas in Pu’er (Menglian and Simao) and two in Baoshan (Lujiangba and Longling), as shown in Fig. 1. Mature coffee cherries were hand-picked by trained personnel wearing gloves to avoid contamination. All debris, including soil particles, rainwater, leaves, and branches, was carefully removed in the field. The cleaned samples were sealed in polyethylene bags and stored for further analysis. A total of 23 coffee cherry samples were collected from Pu’er and 29 from Baoshan. Soil samples were collected from the root zones of the corresponding coffee plants from which the cherries were harvested. The sampling depth corresponded to the main rooting zone, approximately 5–15 cm beneath the surface. For each tree, five subsamples were taken within a 1-meter radius and thoroughly mixed to form a composite sample. Depending on field conditions, 1–2 soil samples were obtained per coffee plant, resulting in 31 soil samples from Pu’er and 54 from Baoshan. Each composite soil sample weighed approximately 1 kg and was stored in inert polyethylene bags to prevent contamination. Additionally, 4 bedrock samples were collected from each region—2 from Menglian and Simao in Pu’er (sandstone) and 2 from Lujiangba and Longling in Baoshan (mudstone). All bedrock samples were also sealed in polyethylene bags and transported to the laboratory for further geochemical analysis.

Fig. 1.

Sampling location map of the study area

All samples in this study were collected with permission from the coffee plantation management authorities. The samples from Baoshan were sourced from Biton Coffee Plantation in Lujiangba and Lameng Town Coffee Plantation in Longling County. The samples from Pu’er were obtained from Beigui Coffee Plantation in Simao and Daya Coffee Plantation in Menglian County. After collection, all samples were directly sent to the Central Laboratory of the Yunnan Provincial Geological and Mineral Exploration and Development Bureau for analysis.

During the coffee bean preparation stage, harvested coffee cherries were first sun-dried to remove moisture. Once fully dried, the outer husks were manually removed using gloves to obtain the green coffee beans. The beans were then ground into fine powder using a ceramic grinder. For each sample, 30 g of the powdered beans were weighed and stored in polyethylene plastic bags. The powdered samples were kept in desiccators containing calcium chloride to maintain constant dry weight prior to digestion. All soil samples were air-dried at room temperature, followed by removal of coarse gravel and plant roots. The dried soils were then passed through a 200-mesh sieve and stored in polyethylene bags for subsequent analysis. Rock samples were first oven-dried at 105 °C for 24 h, crushed with a jaw crusher, and ground to pass through a 200-mesh sieve. The powdered rock samples were also sealed in plastic bags for further elemental analysis. For digestion, 0.1 g of each homogenized coffee bean sample was accurately weighed into a 50 ml tall-form beaker, followed by the addition of 10 ml HNO₃ [32, 33]. Similarly, for soil and bedrock samples, 0.1 g of each homogenized powder was added into a 50 ml tall-form beaker, followed sequentially by 10 ml HNO₃, 10 ml HF, and 2 ml HClO₄. All samples were covered and pre-digested for 1 h, then subjected to microwave-assisted digestion at 120 °C for 5 min to achieve complete dissolution. After cooling, the digested solutions were diluted to volume and analyzed using inductively coupled plasma mass spectrometry (iCAP-RQ, Thermo Fisher Scientific, USA) to determine the concentrations of 11 MEs: Cr, Cu, Mn, Zn, Ca, K, Mg, Fe, Pb, Mo, and As.

For soil pH determination, 10 g of air-dried and sieved soil sample was placed in a 50 ml tall-form beaker. Then, 25 ml of ultrapure water was added. The container was sealed and shaken vigorously using a magnetic stirrer for 2 min, followed by a 30-minute standing period to form a stable suspension. A pH meter was calibrated using standard buffer solutions at 25 °C (first with pH 6.86, then pH 4.01). The pH electrode was immersed vertically into the suspension to approximately two-thirds of its depth, and the pH value was recorded under controlled room temperature (25 °C). For SOM measurement, 0.5 g of sieved soil sample was weighed into a 150 ml Erlenmeyer flask. Then, 0.1 g of powdered silver sulfate was added as a catalyst, followed by the precise addition of 10 ml of 0.4 mol/L potassium dichromate–sulfuric acid solution. The mixture was thoroughly shaken and placed on an electric sand bath at 160 °C for 5 min, ensuring that the total volume in the flask did not exceed 80 ml. After cooling, the residual dichromate was titrated with ferrous sulfate solution to determine the soil organic matter (SOM) content.

To ensure the accuracy and reliability of the analysis results, blank samples and duplicate samples were set throughout the entire experimental process for quality control. In addition, national standard soil reference materials (GSS 38 for K, Ca, and Mg; GSD-30 for Cr, Cu, Mn, Zn, Fe, Pb, and Mo) and a national standard coffee bean reference material (GSB-30a) were used during testing. Calibration curves were established using linear regression (R² ≥ 0.99). Recoveries of all measured elements were maintained within the range of 75.7% to 100%, specifically Cr 86.2%, Cu 89.5%, Mn 99.4%, Zn 75.7%, Fe 90.7%, Pb 98.6%, Mo 100%, As 80.0%, K 98.2%, Ca 97.8%, and Mg 99.7%. Although the recovery of Zn was relatively lower (75.7%), it is still within an acceptable range for trace element analysis in soils and plant matrices [20, 23]. The detection limits of each element were as follows: Cr, Cu, Mn 0.05 mg/kg; Zn 0.50 mg/kg; Fe 1.0 mg/kg; Pb 0.02 mg/kg; Mo 0.01 mg/kg; As 0.002 mg/kg; K, Ca, Mg 1.0 mg/kg (Table S1). In addition, to identify the rock types of bedrock in the two production regions, 4 bedrock samples were examined under a polarizing microscope (Leica DM4500P) to determine their mineral compositions.

Data analysis

Descriptive statistics and one-way Analysis of Variance (ANOVA) were performed using SPSS 23.0 software (IBM, USA), with a significance level set at p < 0.05. Descriptive results are presented as mean ± standard deviation (SD). Group differences were assessed using ANOVA, followed by post hoc multiple comparison tests: Tukey’s HSD test was applied when homogeneity of variance was assumed, while Tamhane’s T2 test was used otherwise. All boxplots were generated using R Studio (Posit, USA), and sampling maps were created using ArcMap 10.8 (Esri, USA). To compare the concentration levels of MEs in soils and coffee beans against threshold values, the Chinese Environmental Quality Standard for Agricultural Soils (GB 15618–2018) [34] and the National Food Safety Standard (GB 2762–2017) [35] were used as references.

Bioconcentration factor (BCF)

BCF is defined as the ratio of the concentration of a given mineral element in the edible part of the plant to its corresponding concentration in the soil sample [36]. This parameter reflects the transfer potential of elements from soil to plant tissues. It is calculated using the following equation, as proposed by Kachenko and Singh (2004) [37]:

1

where is the bioconcentration factor of the mineral element , denotes the concentration of element in the coffee bean (mg/kg), and denotes the corresponding concentration in the soil (mg/kg). A BCF value greater than 1 indicates that the coffee bean enriches the respective element, suggesting active accumulation from soil to plant tissue [38].

Enrichment factor (EF)

EF is commonly used to assess the source and relative enrichment of elements in soil. EF is typically calculated as the ratio between the concentration of a given element in the soil and its concentration in the local parent rock [39]. A geochemically stable element is generally selected as the reference element. In this study, titanium (Ti) was chosen [40] primarily because it is chemically stable in soils, minimally absorbed by plants, and largely unaffected by environmental or anthropogenic activities. Additionally, using Ti avoids the potential variability associated with commonly used reference elements such as Fe, which is included as target element in this study and whose concentration can fluctuate with soil pH and organic matter content. For soil samples, the Ti concentration of each individual sample was used in the EF calculation, while for the parent rock, only two representative samples were collected in each region. Therefore, weighted average concentrations of Ti in the bedrock were used as the reference values. The average Ti concentrations in Pu’er and Baoshan bedrock were 2865.8 mg/kg and 2615.5 mg/kg. Typically, when EF ≈ 1, the element is considered to originate mainly from natural geological processes such as bedrock weathering. When EF > 1 and significantly greater than 1, it may suggest external inputs or anthropogenic influences. Conversely, EF < 1 may indicate element depletion, migration, or adsorption during pedogenesis [39].

Multiple Linear Regression (MLR)

To investigate the effects of soil properties—namely pH and SOM—on the concentrations of 11 MEs (Cr, Cu, Mn, Zn, Ca, K, Mg, Fe, Pb, Mo, and As), MLR models were constructed separately for the Pu’er and Baoshan regions. In each model, soil pH and soil organic matter (SOM) were consistently treated as independent variables, while the concentration of each mineral element (Cr, Cu, Mn, Zn, Ca, K, Mg, Fe, Pb, Mo, and As) served as the dependent variable. All regression analyses were conducted in R Studio (Posit, USA) using the “lm()” function, which is a built-in function in R for performing linear regression. For each model, the coefficient of determination (R²), regression coefficients (β), and corresponding p-values were extracted.Prior to regression, the normality of the data was assessed using the Kolmogorov–Smirnov test [41]. To test for the independence of residuals, the Durbin–Watson (DW) statistic was calculated for each model [42]. For models with DW values significantly below 2 (especially DW < 1), the standard errors were corrected using the Newey–West method to ensure the robustness of the statistical inference regarding regression coefficients. This correction was implemented using the “sandwich::NeweyWest()” and “lmtest::coeftest()” functions, where “NeweyWest()” provides heteroskedasticity- and autocorrelation-consistent standard errors, and “coeftest()” is used to compute hypothesis tests based on the corrected standard errors [43].Variance inflation factors (VIFs) below 5 indicated acceptable levels of multicollinearity, and R² values greater than 0.10 were considered to have practical significance [44, 45].

Principle Component Analysis (PCA)

To investigate the sources of 11 MEs in soils from the two major coffee-producing regions of Baoshan and Pu’er, PCA was conducted using the “prcomp()” function from the “stats” package in R Studio (Posit, USA) [46]. PCA plots were generated using Origin Pro 2025 (OriginLab, USA). The analysis was applied separately to the concentration data of the 11 MEs in soils from both regions [47]. Prior to analysis, all data were standardized to remove the effects of different units and scales. The suitability of the dataset for PCA was assessed using the Kaiser–Meyer–Olkin (KMO) test, where KMO values greater than 0.5 indicate adequacy for PCA [48]. Typically, the first few components with cumulative variance greater than 70% are selected to interpret the dominant elemental source structures [49].

Partial Least Squares Regression (PLSR)

To explore the elemental compositional differences between soil and coffee bean samples in the two major coffee-producing regions of Baoshan and Pu’er, this study employed PLSR modeling using the “plsr()” function from the “pls” package in R Studio (Posit, USA) [50]. The predictor matrix (X) consisted of elemental concentrations in soils, while the response matrix (Y) comprised elemental concentrations in coffee beans. PLSR constructs latent variables (LVs) to extract the most relevant information from the original variables, effectively addressing multicollinearity among predictors. Each LV represents an optimal projection direction that maximizes the covariance between a linear combination of predictors and responses [24]. For variable selection, the Variable Importance in Projection (VIP) score was used to evaluate the contribution of each predictor to the overall model, calculated as a weighted sum of the absolute loadings across all LVs [50, 51]. Variables with VIP values greater than 1 were considered to have significant contributions to the model and strong explanatory power. In addition, the Variable Identification Value (VID) was introduced to reflect the importance of each variable in class discrimination. VID was calculated based on the difference in projection coefficients between sample categories, where higher VID values indicate greater discriminatory ability [52].

Results

Concentrations of mineral elements in soil and coffee bean samples

Descriptive statistics of ME concentrations in soil and coffee bean samples from Baoshan and Pu’er are summarized in Table S2. All tested soil samples exhibited pH values below 5.5. Importantly, the concentrations of PTEs—As, Cr, Pb—in all soil samples were below the thresholds set by the Chinese Environmental Quality Standard for Agricultural Soils (GB 15618–2018). Similarly, As, Cr, and Pb concentrations in all coffee bean samples were within the limits specified by the National Food Safety Standard (GB 2762–2017), indicating overall environmental safety and food security for the studied coffee-growing regions.

To facilitate visualization of the distribution and inter-regional differences in the concentrations of 11 MEs, all concentrations were log₁₀-transformed and presented as boxplots (Fig. 2). Comparisons were made between Pu’er and Baoshan regions for both soil and coffee bean samples. Statistical differences between groups were assessed using one-way ANOVA at a significant level of p < 0.05, with detailed results shown in Table S3. Statistically, no significant differences were observed in the concentrations of any of MEs between the two regions in either soil or coffee bean samples. However, based on mean values, As, Ca, and Pb concentrations in Pu’er soils were higher than those in Baoshan, whereas the concentrations of all other elements were lower. However, visual inspection of the boxplots revealed certain trends in mean values: As, Ca, and Pb concentrations were higher in Pu’er soils than in Baoshan soils, whereas the concentrations of all other elements were lower. In contrast, when comparing coffee bean samples between the two regions, the mean concentration of Cr was higher in Baoshan beans than in Pu’er beans, while all other MEs were higher in Pu’er beans, suggesting a generally stronger ME uptake capacity in coffee plants from Pu’er.

Fig. 2.

Boxplots of 11 ME concentrations (log₁₀) in soils and coffee beans from Pu’er and Baoshan. The boxes represent the interquartile range (IQR, 25th–75th percentile), the horizontal line inside each box indicates the median, and the whiskers extend to 1.5 × IQR. The dots represent outliers beyond 1.5 × IQR

Soil-to-Bean Element Transfer Coefficients (BCF)

To assess the bioaccumulation characteristics of soil-derived elements in coffee beans, the BCFs of 11 MEs were calculated for both Pu’er and Baoshan regions according to Eq. (1) (Table S4). The results showed that only K and Ca had BCF values greater than 1, indicating active enrichment. To compare inter-regional differences, all BCF data were log₁₀-transformed and presented as boxplots (Fig. 3), and group differences were evaluated using ANOVA (p < 0.05) (Table S5). The results revealed that most elements showed significant inter-regional variation. In particular, Pb, As, Mn, Zn, Fe, Cr, and Cu demonstrated statistically significant differences (p < 0.05), with the degree of variation ranked as follows: Pb > As > Mn > Zn > Fe > Cr > Cu. Conversely, K, Ca, and Mo did not exhibit significant regional differences.

Fig. 3.

Boxplot of BCF Values (Log₁₀) in Pu’er and Baoshan. The boxes represent the interquartile range (IQR, 25th–75th percentile), the horizontal line inside each box indicates the median, and the whiskers extend to 1.5 × IQR. The dots represent outliers beyond 1.5 × IQR

Boxplot results further showed that the BCFs of Pb, As, Mn, Zn, Fe, and Cu were significantly higher in coffee beans from Pu’er than those from Baoshan (p < 0.05), indicating a stronger accumulation capacity for these elements in Pu’er-grown coffee. In contrast, the BCF of Cr was higher in Baoshan samples than in Pu’er (p < 0.05) (Table S5), suggesting greater enrichment and bioavailability of Cr in Baoshan soils. These differences may be associated with variations in soil properties between the two regions.

Source apportionment of MEs in soil

Rock-to-soil Element Enrichment (EF)

To investigate the sources of 11MEs in soils from the Pu’er and Baoshan coffee-producing regions, it is essential to first distinguish between natural (lithogenic) and anthropogenic origins. In this study, the elemental concentrations in bedrock samples from both regions were statistically summarized (Table S6) and calculated enrichment factors (EFs). The resulting EF values are presented in Table S7 and visualized in Fig. 4. Group differences were evaluated using ANOVA (p < 0.05), with detailed statistical results shown in Table S8.

Fig. 4.

Boxplot of EF Values (Log₁₀) in Pu’er and Baoshan. The boxes represent the interquartile range (IQR, 25th–75th percentile), the horizontal line inside each box indicates the median, and the whiskers extend to 1.5 × IQR. The dots represent outliers beyond 1.5 × IQR

According to the statistical analysis, all elements except Mn, K, and As exhibited significant differences in EF values between the two regions. The degree of interregional variation, based on F-values, was ranked as follows: Pb > Mg > Mo > Fe > Cu > Cr > Ca > Zn. These results indicate that Pb and Mg demonstrate the most pronounced geochemical background differences between Pu’er and Baoshan, suggesting distinct geological or environmental controls on their distribution.

In terms of EF value distribution, all MEs in Pu’er—except for Zn and K—exhibited peak EF values greater than 1, indicating potential external enrichment. Specifically, the proportions of samples with EF > 1 for Cr, Cu, Mn, Ca, Mg, Fe, Pb, Mo, and As were 90.32%, 83.87%, 64.52%, 6.45%, 19.35%, 22.58%, 87.10%, 87.10%, and 83.87%, respectively. In Baoshan, peak EF values exceeded 1 for all elements except Cu, Ca, and Mg. The proportions of samples with EF > 1 were 98.15% for Cr, 25.93% for Mn, 51.85% for Zn, 1.85% for K, 72.22% for Fe, 14.81% for Pb, 90.74% for Mo, and 88.89% for As.

The majority of elements in both regions exhibited peak EF values exceeding 1, suggesting the presence of exogenous inputs. Conversely, elements with EF values consistently below 1 were likely derived primarily from natural bedrock weathering. Notably, Ca, Mg, and Fe in Pu’er, as well as Mn, K, and Pb in Baoshan, showed relatively low proportions of EF > 1, indicating a dominant lithogenic origin with minimal anthropogenic influence. In contrast, Cr, Cu, Mn, Pb, Mo, and As in Pu’er, and Cr, Zn, Fe, Mo, and As in Baoshan displayed high proportions of EF > 1, implying substantial anthropogenic contributions in addition to natural weathering processes.

Based on the boxplot analysis, the average enrichment levels of Cu, Mn, Ca, Mg, Pb, and As were higher in Pu’er compared to Baoshan, while Cr, Zn, K, Fe, and Mo showed greater enrichment in Baoshan. These results collectively indicate that the enrichment characteristics of soil mineral elements in the two coffee-producing regions are governed by distinct geological backgrounds and environmental evolution pathways, with both commonalities and differences. The commonalities include MEs such as Mg, Ca, and K, which were predominantly derived from bedrock weathering in both regions.

Influence of soil pH and SOM on Mineral Element Concentrations (MLR)

To further investigate the potential influence of soil properties on mineral element concentrations, MLR was conducted using the concentrations of 11 MEs as dependent variables and soil pH and SOM as independent variables. The results are summarized in Table S9.

In the Pu’er region, soil pH showed generally low explanatory power for the elemental concentrations, with R² values less than 0.5 for all MEs. However, Cu, Mn, Zn, Ca, Mg, Fe, and Pb showed statistically significant relationships (p < 0.05) with R² values greater than 0.1, suggesting meaningful regression results. Among them, except for Fe, all other elements exhibited positive regression coefficients (β), indicating that Cu, Mn, Zn, Ca, Mg, and Pb were positively correlated with pH, whereas Fe was negatively correlated. This implies that with increasing soil pH, the concentrations of Cu, Mn, Zn, Ca, Mg, and Pb tend to increase, while Fe concentrations decrease. SOM exhibited a stronger explanatory power for most elements. For example, the R² value for Fe reached 0.65 at Pu’er and 0.38 at Baoshan. In addition, Cr, Cu, Mn, Zn, Mg, Pb, and Mo also showed significant relationships (R² > 0.1, p < 0.05). Notably, Fe displayed a positive correlation with SOM, whereas the other MEs (Cr, Cu, Mn, Zn, Mg, Pb, and Mo) were negatively correlated. These results suggest that higher SOM content in soils corresponds to higher Fe concentrations, but lower concentrations of Cr, Cu, Mn, Zn, Mg, Pb, and Mo. These results indicate that the effect of SOM on metal bioavailability is element-specific. While higher SOM content generally enhances the bioavailability of some metals, in our study, Fe concentrations increased with SOM, whereas Cr, Cu, Mn, Zn, Mg, Pb, and Mo concentrations decreased. This inverse relationship for certain metals may be attributed to adsorption and complexation processes, in which organic matter can immobilize specific trace metals and reduce their bioavailable fractions in the soil. Thus, SOM may enhance the bioavailability of certain elements but decrease it for others, depending on their interactions with organic matter functional groups. In the Baoshan region, both pH and SOM exhibited weak explanatory power for elemental concentrations. Regarding pH, Cu had the highest R² value of 0.40, while Cr, Mn, and Mg also showed R² values greater than 0.1 (p < 0.05). Among them, Mg was positively correlated with pH, whereas Cr, Mn, and Fe showed negative correlations. This suggests that soils with relatively higher pH within the acidic range tend to have higher Mg concentrations but lower levels of Cr, Mn, and Fe. As for SOM, Cu again demonstrated the highest R² value (0.44). Additionally, Cr, Mn, Zn, Fe, Mo, and As all showed R² values greater than 0.1 (p < 0.05), and all were positively correlated with SOM, indicating that higher organic matter content tends to increase the concentrations of these elements in soil.

Identification of soil mineral element sources (PCA)

PCA was conducted to identify the underlying structure of ME concentrations in soil samples. Prior to PCA, the Kaiser–Meyer–Olkin (KMO) test was performed to assess sampling adequacy. The KMO values were 0.51 for Pu’er and 0.73 for Baoshan, both exceeding the threshold of 0.5, indicating that the data were suitable for PCA analysis. The PCA results revealed distinct differences in the loading patterns of elements between the two regions (Fig. 5). In the Pu’er region, three principal components (PCp1, PCp2, and PCp3) were extracted, collectively explaining 73.74% of the total variance. In contrast, four principal components (PCb1, PCb2, PCb3, and PCb4) were identified in the Baoshan region, accounting for a cumulative variance of 76.50%. The rotated component matrices for both regions are presented in Table S10.

Fig. 5.

Principal component analysis (PCA) results for the first three principal components related with 11 MEs in soil from Pu’er and Baoshan

In the Pu’er soil samples, PCp1 accounted for 42.69% of the total variance and was heavily loaded with Cu (0.944), Mn (0.820), Zn (0.642), Ca (0.618), Mg (0.784), Fe (− 0.609), and Pb (0.933). PCp2, explaining 18.34% of the variance, showed strong loadings for Cr (0.686) and Mo (0.787). PCp3 explained 12.71% of the variance and was mainly associated with K (0.613) and As (0.701). In the Baoshan soil samples, four principal components were extracted, cumulatively explaining 76.50% of the total variance. PCb1 accounted for the largest proportion (42.86%) and was heavily loaded with Cr (0.884), Cu (0.897), and Fe (0.816). PCb2 explained 13.17% of the variance and was mainly associated with Zn (0.747), Pb (0.897), Mo (0.698), and As (0.804). PCb3, which explained 10.87% of the variance, was dominated solely by Mn (0.881). PCb4 explained 9.60% of the total variance and was characterized by Ca (0.694) and Mg (0.628).

Key soil predictors for discriminating coffee beans from Pu’er and Baoshan (PLSR)

To further elucidate the relationships between soil and coffee bean MEs and their contribution to geographical origin discrimination, PLSR was employed. The model was constructed using the concentrations of 11 MEs in soils as the predictor matrix (X) and the corresponding 11 MEs in coffee beans as the response matrix (Y).Three latent variables (LVs) were extracted, each representing a linear combination of the original variables that captures the main patterns of variation and covariation between soil and coffee bean MEs, thus summarizing the majority of the variation in the datasets.

In Pu’er, the first latent variable (LV1) explained 31.56% of the variation in X and 21.72% in Y, indicating its dominant role in extracting elemental structure differences among samples. The second (LV2) and third (LV3) latent variables accounted for 38.98% and 6.03% of the variation in X, and 7.53% and 5.93% in Y, respectively, suggesting that they provided supplementary discriminative power to the model. In contrast, the Baoshan model showed relatively lower explanatory power. LV1 accounted for 17.03% of the variation in X and 13.45% in Y. LV2 and LV3 explained 18.11% and 11.12% of the X variation, and 8.96% and 7.08% of the Y variation, respectively. The lower model performance in Baoshan may be attributed to the higher degree of anthropogenic influence on soil MEs in this region, resulting in more complex and heterogeneous elemental patterns that reduce model robustness. Overall, the PLSR results indicate that soil–bean elemental transfer relationships are more stable and interpretable in Pu’er, where natural geochemical processes dominate, while in Baoshan, external inputs may obscure the soil–plant elemental signal and weaken origin traceability.

The three-dimensional scores plot (Fig. 6) revealed a clear separation between the Baoshan and Pu’er samples along the first latent variable (LV1), indicating that LV1 has a strong discriminative effect with respect to geographic origin. Although LV2 and LV3 did not fully separate the two groups, partial differentiation among individual samples was observed, suggesting the presence of potential covariation patterns among certain elemental combinations. In the LV1–LV2 plane, Pu’er samples were predominantly distributed in the left quadrant and exhibited a relatively compact clustering pattern, implying a more consistent elemental composition likely controlled by uniform geogenic factors. In contrast, Baoshan samples were widely dispersed towards the right along LV1, with some individuals showing substantial deviations along LV2. This distribution pattern implies greater heterogeneity in elemental composition, potentially influenced by multiple environmental drivers. When considered alongside PCA results, the observed dispersion of Baoshan samples may reflect the combined effects of mining activities and agricultural inputs, both of which can introduce external elemental sources and alter geochemical signatures in the soil–plant system.

Fig. 6.

3D Score Plot Based on PLSR for Pu’er and Baoshan

To identify the most influential soil MEs contributing to the prediction of corresponding elemental concentrations in coffee beans, the VIP (Variable Importance in Projection) scores were calculated for both regional PLSR models and visualized in the form of a heatmap (Fig. 7). As shown, Zn and Mo exhibited consistently high VIP scores across both regions, indicating their dominant roles not only in the predictive accuracy of the model but also in the soil-to-bean elemental transfer process. Notably, Cr and Cu displayed substantially higher VIP values in Pu’er than in Baoshan, whereas K and As were more influential in the Baoshan model. These differences suggest regional variability in plant uptake efficiency or elemental bioavailability, which may be related to site-specific environmental factors such as atmospheric deposition from coal ash combustion, as supported by the PCA findings. To further identify soil elements with strong discriminative power between the two geographic origins, the VID (Variable Identification Value) was introduced as a measure of classification importance. As illustrated, both Ca and Mg yielded VID scores exceeding 0.2 in both Pu’er and Baoshan, highlighting their relevance in differentiating soil geochemical characteristics between regions. In particular, Ca (0.65) and Cr (0.44) exhibited markedly higher VID scores in Baoshan, potentially reflecting the influence of local lithology or distinct agricultural practices. In contrast, Mn, Mg, and Pb showed greater VID values in Pu’er, suggesting their association with parent material composition or region-specific pedogenic processes, aligning with the source distinctions revealed by PCA.

Fig. 7.

Heatmap of VIP and VID Values for Pu’er and Baoshan

Integration of VIP and VID results revealed that Cr, Cu, Ca, and Mg simultaneously exhibited high importance in both prediction (VIP) and classification (VID), designating them as critical dual-purpose variables. These MEs likely reflect both geogenic background differences and differential bioavailability, suggesting high potential for stable cross-media transfer and regional discrimination. This dual-importance insight provides a theoretical basis for establishing multi-element fingerprinting systems for coffee origin traceability.

Discussion

This study systematically investigated the geochemical characteristics of 11 MEs in soils and their transfer patterns to Arabica coffee beans in two major coffee-producing regions in China, Pu’er and Baoshan. Firstly, the concentrations of MEs in soils and coffee beans and their BCF were assessed, revealing the uptake and translocation capacities of plants for different elements. Secondly, the natural (lithogenic) and anthropogenic sources of soil elements were elucidated through EF analysis and PCA. Thirdly, MLR analysis was applied to compare the distribution of key elements in soils between the two regions and their relationships with soil physicochemical properties. Finally, PLSR was used to quantify the influence of soil elements on coffee bean mineral composition and to identify key discriminant factors for potential use in origin traceability.

In terms of concentrations, K, Mg, and Ca in soils of both Pu’er and Baoshan exhibited notable variability and high dispersion, likely associated with plot-specific fertilization practices and differences in nutrient uptake by individual coffee plants, a phenomenon also commonly observed in other cropping systems [53–55]. Overall, all MEs were present at higher concentrations in soils than in coffee beans, with Fe, Mn, Cr, Zn, As, and Pb showing particularly large differences, indicating limited plant uptake or translocation capacity, consistent with previous studies [56]. Conversely, K, Ca, Mg, Cu, and Mo displayed relatively smaller concentration differences between soils and beans, suggesting higher bioavailability and efficient plant uptake [36, 56].

Regarding soil–bean transfer characteristics, BCF analysis showed that K and Ca had values greater than 1, reflecting active accumulation. This is likely because they are essential macronutrients, generally abundant and evenly distributed in soils, and their uptake is tightly regulated by plant physiological mechanisms, resulting in relatively stable BCFs across regions [57, 58]. In contrast, BCFs for other elements were below 1, suggesting limited uptake. Although absolute uptake may be constrained, minor differences in soil geochemical properties, bioavailable fractions, pH, organic matter content, and metal complexation can still produce significant interregional differences in BCF [58, 59]. Notably, the BCFs of Pb, As, Mn, Zn, Fe, and Cu in Pu’er coffee beans were significantly higher than those in Baoshan (p < 0.05), indicating a stronger accumulation potential for these elements in coffee grown in Pu’er. In contrast, the BCF of Cr was higher in Baoshan samples than in Pu’er (p < 0.05). This phenomenon can be attributed to local anthropogenic inputs in Baoshan soils, such as mining activities and atmospheric deposition, as revealed by PCA, EF, and MLR analyses. Specifically, PCA of Baoshan soils indicated that Cr is a dominant element in PCb1, with EF values suggesting primarily anthropogenic sources, likely related to mining, ore transport, and industrial emissions, while its mobility is still partially regulated by soil pH and SOM [60–62]. In comparison, Cr and Mo in Pu’er soils (PCp2) are also influenced by industrial sources, but the lower local industrial intensity results in comparatively lower Cr accumulation in coffee beans. Since the same coffee variety (Coffea arabica L.) was cultivated in both regions, the observed differences are largely attributable to soil geochemical background and environmental factors rather than varietal differences. For example, higher organic matter content in Pu’er soils (Fig. S1–S4) may enhance metal complexation and bioavailability [63, 64], and slight differences in soil pH between the regions may further affect metal solubility and plant uptake. Overall, these findings provide a mechanistic explanation for the observed BCF differences, highlighting the combined effects of natural geochemical background and anthropogenic activities on elemental accumulation in coffee beans.

From a bedrock–soil perspective, soil enrichment characteristics of MEs in Pu’er and Baoshan showed both commonalities and notable differences. In Pu’er, average enrichment levels of Cu, Mn, Ca, Mg, Pb, and As were higher than in Baoshan, whereas Cr, Zn, K, Fe, and Mo were more enriched in Baoshan. These results indicate that soil enrichment patterns in the two coffee-producing regions are controlled by distinct geological backgrounds and environmental evolution pathways, with both shared and unique features. Shared elements, including Mg, Ca, and K, were mainly derived from bedrock weathering in both regions, likely due to their high abundance in the crust and wide occurrence in sedimentary rocks, making their enrichment primarily controlled by parent rock mineral composition [65]. Based on the mineral compositions of the bedrocks in Pu’er and Baoshan, muscovite is commonly present in both, which likely serves as a primary source for these three elements (Figs. S1–S4).

Regarding the relationship between ME concentrations and soil physicochemical properties, MLR analysis showed that Fe was significantly negatively correlated with pH and Mg significantly positively correlated with pH in both regions, indicating that soil pH strongly influences their mobility and enrichment [66, 67]. Mn exhibited opposite correlations with pH in the two regions: positive in Pu’er and negative in Baoshan, likely due to its redox sensitivity. Although Fe is also redox-sensitive, it mainly exists in mineral-bound forms, limiting mobility, consistent with abundant Fe-rich clay fillings in the bedrocks of both regions (Figs. S1–S4). Pu’er soils are moist and reductive, favoring Mn dissolution, whereas Baoshan soils are at higher elevation with better drainage, tending to be more oxidative, thus reducing Mn solubility [4, 56]. Regarding SOM, Cr, Cu, Mn, Zn, and Mo were negatively correlated with SOM in Pu’er but positively correlated in Baoshan. This difference may stem from variations in SOM composition: Pu’er SOM is rich in humic substances that can bind metals into insoluble complexes or precipitates, reducing mobility [29]. In contrast, Baoshan soils may favor a “complexation-facilitated transport” mechanism, where metal elements form soluble organometallic complexes that migrate and accumulate with SOM [68], although the specific SOM composition requires further study for confirmation.

In terms of soil ME sources, PCA in Pu’er soils extracted three principal components. The first component (PCp1) loaded heavily on Cu, Mn, Zn, Ca, Mg, Fe, and Pb; combined with EF and MLR results, PCp1 primarily represents natural weathering. Although Pb EF values suggest possible anthropogenic input, its significant positive correlation with pH and negative correlation with SOM indicate that its mobility and accumulation are also regulated by soil physicochemical properties, confirming natural weathering as the main source of PCp1. This suggests that the baseline geochemical background for most common MEs in Pu’er soils is mainly controlled by parent rock composition. The second component (PCp2) was dominated by Cr and Mo, whose high EF values suggest predominant anthropogenic sources, likely related to industrial smelting and mining activities [69–71]. Although Pu’er is not a traditional industrial city, local polymetallic deposits are abundant, and reports of small-scale smelting exist [72, 73], supporting the interpretation of PCp2 as reflecting industrial activity. Moreover, PCp2 elements showed certain correlations with soil pH and SOM in MLR analysis, suggesting that soil physicochemical properties still regulate the mobility and bioavailability of these industrial-source elements. The third component (PCp3) was dominated by K and As; EF analysis indicates K is from natural sources, while As is anthropogenic. In MLR, both elements showed no significant correlation with pH or SOM. Their co-occurrence is typically associated with agricultural inputs, particularly fertilizer application [74]. Studies suggest that potassium fertilization can reduce plant As uptake [75], further supporting the interpretation of PCp3 as reflecting agricultural activity. This indicates that in Pu’er, fertilization significantly influences the accumulation of specific elements, while their association with soil baseline properties is weak.In Baoshan soils, PCA extracted four principal components. PCb1 was dominated by Cr, Cu, and Fe; EF analysis suggests Cr and Fe are anthropogenic and Cu is natural, but all three elements showed significant correlations with soil pH and SOM, indicating mobility regulated by soil properties. Considering local mining geography and historical emissions, PCb1 likely reflects the influence of atmospheric deposition or mine leachate. PCb2 was mainly associated with Zn, Pb, Mo, and As; EF values indicate clear anthropogenic sources, possibly related to mining [76], ore transport [77], and agricultural inputs [78–81]. Baoshan lies within the “Three Rivers” metallogenic belt, a mineral-rich region [82], supporting the interpretation of PCb2 as influenced by mining and agricultural activities. PCb3 (10.87% variance) was dominated by Mn (0.881); EF analysis suggests mixed natural and anthropogenic sources, and MLR indicates significant correlation with soil properties. Considering Mn’s redox sensitivity, this component may reflect the influence of soil environmental conditions—particularly moisture, redox status, and pH—on element accumulation. PCb4 (9.60% variance) was dominated by Ca (0.694) and Mg (0.628); EF analysis indicates both are of natural origin. These alkaline earth metals are widely distributed in the crust and typically released through bedrock weathering, suggesting PCb4 reflects geological or lithogenic background. Overall, in Baoshan, baseline geochemical conditions still dominate the distribution of some MEs (especially Ca and Mg), while anthropogenic activities more strongly influence Cr, Pb, Zn, Mo, and As.

PLSR analysis further revealed soil–coffee bean element transfer patterns and potential for origin discrimination. In Pu’er, LV1 clearly distinguished samples, indicating that soil–bean element transfer is primarily controlled by stable natural geochemical processes. In Baoshan, model explanatory power was relatively lower, with samples more dispersed along LV1 and LV2, reflecting greater heterogeneity of soil–bean element composition. Combined with PCA results, this dispersal pattern likely reflects the combined influence of mining and agricultural activities, which introduce exogenous elements and alter the geochemical characteristics of the soil–plant system [24, 83]. VIP and VID analyses identified Cr, Cu, Ca, and Mg as having high importance in both regions, indicating that these elements are influenced by geological background yet effectively accumulated in plants, making them reliable cross-medium discriminant markers [84–86]. Combined with PCA results, these elements’ distribution patterns reflect the integrated effects of geological, environmental, and anthropogenic factors in the soil–bean system, providing a theoretical basis for establishing a multi-element fingerprinting system for coffee origin traceability and informing regional coffee management and potential food safety risk assessment.

Despite these insights, this study has several limitations. Sample numbers were limited (Pu’er: 31 soils, 23 beans; Baoshan: 54 soils, 29 beans), and spatial coverage was constrained. The stability of key elements across different years or growing seasons was not evaluated. In addition, only bulk soil organic matter (SOM) content was measured, while its specific composition was not characterized, which may influence the understanding of element binding mechanisms and mobility. Although composite sampling, PCA-based dimensionality reduction, PLSR modeling, and VIP-based variable selection enhanced model robustness, future research should expand temporal (seasonal/annual) and spatial (additional regions or finer plot scales) sampling. Furthermore, incorporating detailed SOM compositional analysis and other soil physicochemical properties would allow for a more systematic assessment of the temporal stability and discriminative potential of key elements, and provide deeper insights into the mechanisms governing element transfer, thereby improving origin discrimination and mechanistic understanding of element mobility.

Conclusion

This study systematically compared the differences in MEs compositions in soils and coffee beans between two representative coffee-producing regions in Yunnan, China—Pu’er and Baoshan. The results showed that the concentrations of PTEs such as As, Cr, and Pb in both soil and coffee bean samples were below the national environmental quality standards for farmland soils and food safety limits, indicating no pollution risk. Elements such as K, Ca, Mg, Cu, and Mo showed relatively small differences in concentration between soils and beans, reflecting their high bioavailability. Although no statistically significant differences were observed for most elements, regional variations were evident in concentration levels, distribution patterns, and enrichment capacities. Specifically, the soils in Pu’er generally exhibited lower elemental concentrations than those in Baoshan, while the corresponding concentrations in coffee beans were higher, suggesting a higher uptake efficiency of MEs by coffee plants in Pu’er.

BCF analysis further confirmed this trend: the BCF values of Pb, As, Mn, Zn, Fe, and Cu in Pu’er samples were significantly higher than those in Baoshan, indicating a stronger accumulation potential for these elements in coffee beans from Pu’er. In contrast, Cr exhibited a higher BCF in Baoshan samples, suggesting greater accumulation of this element in coffee beans from that region. EF analysis revealed external inputs for most MEs in both regions, with Cr, Mo, and As predominantly influenced by anthropogenic activities, while Mg, Ca, and K were mainly derived from bedrock weathering. Pb and Mg exhibited the most pronounced differences in EF values between the two regions. MLR indicated that Fe was negatively correlated with soil pH, and Mg was positively correlated in both regions, highlighting the role of soil acidity in controlling their mobility. However, Mn, Cr, Cu, Zn, and Mo exhibited opposite trends in correlation with pH and soil SOM between Pu’er and Baoshan, suggesting significant differences in redox conditions and organic matter composition that affect elemental speciation and migration. PCA revealed that soils in Pu’er were mainly influenced by natural weathering, industrial smelting, and agricultural activities, while those in Baoshan were affected by atmospheric deposition, mining activities, agricultural inputs, and bedrock weathering, indicating a more complex geochemical background in Baoshan. The PLSR model extracted three LVs, with the cumulative explanatory power higher in Pu’er than in Baoshan. Notably, LV1 provided good separation between the two regions. Among all variables, Cr, Cu, Ca, and Mg were identified as key discriminant elements with high VIP and VID scores, indicating their dual importance in both mineral enrichment in coffee beans and geographic origin differentiation. These findings underscore the dual role of regional variations in bioavailability and geochemical characteristics and offer theoretical support for establishing a multi-element-based traceability framework for coffee origin authentication.

Authors’ contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Shuran Yang and Qianrui Huang. The first draft of the manuscript was written by Shuran Yang and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the Scientific and Technological Innovation Project entitled “Study on the Ecogeochemical Evaluation System in Plateau Specialty Coffee Producing Areas of Yunnan Province” and the Science and Technology Innovation Team for Highland Ecological Agriculture Geological Survey and Evaluation of Yunnan Education Department, China.

Data availability

The datasets used and analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.