Chemometric approach to find relationships between physiological elements and elements causing toxic effects in herb roots by ICP-MS

In this paper 13 elements, both physiological and causing toxic effects, were determined by inductively coupled plasma mass spectrometry in roots of 26 species of herbs used in Traditional Chinese Medicine. The herbs were purchased from online shop in two batches 1 year apart to verify the variability of elemental content in time. The multivariate statistical methods—multiple regression, canonical variates and interaction effect analysis—were applied to interpret the data and to show the relationships between elements and two batches of herb roots. The maximum permissible concentration of Cd (0.3 mg kg−1) was exceeded in 7 herb roots which makes 13% of all specimens. The multiple regression analysis revealed the significant relationships between elements: Mg with Sr; V with Pb, As and Ba; Mn with Pb; Fe with As and Ba; Co with Ni and Sr, Cu with Pb, Cd and As; Zn with Pb, Cd, As and Ba. The canonical variates analysis showed that the statistical inference should not be based solely on the type of herb or number of batch because of the underlying interaction effects between those two variables that may be a source of variability of the content of determined elements.

The multielement analysis of roots of herbs after digestion was performed in ICP-MS (Agilent 7700x, USA) equipped with octopole reaction system (ORS) working in no-gas and helium mode for the reduction of spectral interferences. Non-spectral and matrix interferences were minimized by diluting the samples and using an internal standard solution containing 10 µg L −1 Rh, introduced in parallel with the analyzed solution via T-piece. To ensure the accuracy and correctness of the analytical procedure three CRMs were used: Mixed Polish Herbs (INCT-MPH-2, Institute of Nuclear Chemistry and Technology, Warsaw, Poland), Trace Elements in Spinach Leaves (1515, NIST, USA), Apple Leaves (1570a, NIST, USA). The CRMs were digested in triplicates according to the procedure described in sample preparation chapter. Ten procedural blanks were obtained in the same procedure from the digestion reagents in the digestion vessel. The average value measured for procedural blanks was subtracted from the concentrations measured for digested roots. After calibration the measurement control was conducted through the analysis of standard solutions and CRMs after every ten herb samples.
The operating conditions of ICP-MS are presented in Table 2.
Statistical and chemometric evaluation. In order to evaluate the relationships between elements in herb roots the multivariate methods were used to analyze the results of the measurement. Canonical variate coordinates are directions in multivariate space that maximally separate the predefined groups of the analyzed dataset 33,34 . Canonical variate analysis (CVA) and modifications thereof are widely used in various scientific fields, not only to study the content of elements 35 , but also environmental 36 , agriculture and biological sciences 37 and economics 38 . Additionally, the CVA, similar to principal component analysis, also showed relations between selected experimental objects and the variables that describe them 39,40 . In this case, the experimental object was defined as the values of content of elements for the particular herb root in the individual batch. This method was  39 were calculated based on the results of element content in roots. The differences between herb roots may be either due to the fact the herb root factor causes some interaction with the batch factor or because there are some significant main effects of the former factor, or due to both 41 . The multiple regression method was used, and the parameters of equations which address separately the physiological elements (Mg, V, Mn, Fe, Co, Cu, Zn) were estimated in order to determine the strength and direction of impact of elements causing toxic effect on physiological elements in herb roots. The relationships between the content of physiological elements in herb roots and the analyzed elements causing toxic effects were determined using the multiple linear regression equation: where y -content od physiological element (in mg kg −1 ) in herb roots, β i -regression coefficient, e-random error.
Multiple linear regression and stepwise regression with backward elimination were performed using the statistical software Statistica 13 (Statsoft, Poland).

Results and discussion
The proper conclusions drawn from the experimental data require that the measurements were performed accurately. In order to verify that the developed analytical procedure is fit for intended purpose it was validated by analyzing series of calibration standards, blank samples and matrix-matching CRMs, and by estimating the parameters characterizing the analytical method: working range, linearity, limits of detection and quantification, precision and trueness. After validation of the analytical procedure the elements in samples of herb roots were determined and the obtained results were evaluated with the basic statistics, as well as multivariate statistical methods.

Method validation.
Obtained values with validation parameters are presented in Table 3. Calibration curves for elements determined by ICP-MS were in the range defined by the estimated limit of quantification and the standard with highest concentration. The lowest upper limits of working range equal 50 µg L −1 have V, Co, As, Cd and Pb. The standards with highest upper limits of working range equal 20,000 µg L −1 were prepared for Mg, 5000 µg L −1 for Fe and 2000 µg L −1 for Mn and Zn. Calibration curves for all elements were linear over the entire concentration range which resulted in high correlation coefficients > 0.9998 with exception of 0.9991 for Mg.
Instrumental detection limits (IDL) were calculated as three standard deviations (SD) of the calibration blank samples. The lowest values of IDL were 0.0025 and 0.0088 µg L −1 for Co and V, respectively, and the highest: 0.48 and 0.46 µg L −1 for Zn and Fe, respectively. Method detection limits (MDL) were calculated as three SD from the procedural blank samples. The lowest values of MDL were 0.010 and 0.018 µg L −1 for Co and Cd, respectively, and the highest: 8.1 and 0.91 µg L −1 for Mg and Fe, respectively. Instrumental quantification limits (IQL) were calculated as three times IDL, with the lowest values: 0.0076 and 0.027 µg L −1 for Co and V, respectively, and the highest: 1.4 µg L −1 for Fe and Zn. In all analyzed herbs the obtained concentrations are above the detection limit.
The precision was calculated as the relative standard deviation of three repetitions of calibration standards on three concentration levels and expressed in % as coefficient of variation (CV) with distinction to short-term precision measured in a single analytical run (repeatability) and long-term measured in the span of 3 days (intermediate precision). Repeatability values for all analyzed elements were in the range from 0.68% for Cu to  32 , NIST SRM 1570a (trace elements in spinach leaves) 28,30,32 , NCS DC73349 (trace elements in bush branches and leaves) 42 , Japan CRM NIES no. 7 (tea leaves) 31 . A review article on determination of elements in herbs and CRMs used for method validation was published by Pohl et al. 43 .

Determination of elements in herbs.
The elements were grouped depending on their effect on the human health: physiological elements (Mg, V, Fe, Mn, Co, Cu, Zn) and the second group contains elements that are causing toxic effects (Pb, Cd, As, Ni, Ba, Sr). Elements causing toxic effect have negative effect on health even in low concentrations which include Pb, Cd, As Ni, while Sr and Ba are consider not toxic in the concentrations usually find in food, but may induce toxic effects in higher doses. It is worth noting that certain elements may have detrimentally different toxicity depending on the speciation form, for example BaCl 2 and BaSO 4 . However, the speciation analysis is not in scope of this study and only the general content of elements is discussed.
The results of determination of 13 elements in 26 herbs each analyzed in two batches purchased 1 year apart ( n = 52 ) are presented in Table 4. The two batches of herbs are described as "A" and "B", referring to samples obtained in 2020 and 2021, respectively. The basic descriptive statistics of the measured contents of elements in herb roots are gathered in Table 5. The obtained results show considerable variation among elements and tested herbs. The largest values were measured for Mg with the maximum content of 6947 mg kg −1 in herb 19A (Puerariae). The lowest values were measured for Cd with the minimum content of 0.0040 mg kg −1 in herb 25B (Trichosanthis). Relatively, the narrowest range of content values for all herbs characterizes elements: Mg, Zn and Cu, and the broadest range have Pb and Cd. The average contents of elements causing toxic effects measured for all herbs are in the order: Cd < As < Pb < Ni < Sr < Ba.
The ranges of contents of determined elements, arranged by the minimum measured content and expressed in mg kg −1 , are as follows (number of herb in parenthesis): Cd from 0.004 (25B, Trichosanthis) to 0.798 (26B, Vladimiriae soulier), As from 0.016 (18A, Pseudostellariae) to 0.865 (7B, Codonopsis), Pb from 0.021 (14A, Paeoniae alba) to 9.062 (12A, Morindae officinalis), V from 0.047 (14A, Paeoniae alba) to 3.635 (7B, Codonopsis), Co Table 3. The validation parameters of the developed analytical procedure for determination of 13 elements in roots of herbs. The IDL and IQL refer to the solution and MDL refers to the dry mass of roots. The low and high concentration refers to the standard solution with concentration close to the low limit of working range and the highest concentration standard, respectively. The medium concentration is (in µg L −1 ) 1.0 for V, Co, As, Cd and Pb, 10.0 for Ni and Cu, 100 for Mn, Fe, Zn, Sr and Ba, and 1000 for Mg. The Student's p value refer to the probability of statistical significance between measured and certified value in CRM. *Element not stated in certificate; # Element with informative value.   The method of cultivation and exact geographical origin of herbs in this study is unknown, besides that they come from China. The differences between contents of elements in herbs in this study and the literature data may arise from differences in: method of cultivation, type of soil, pollution of air, water and soil, pH of soil and water, type and amounts of fertilizers, plant species, climatic condition, plant growth time and conditions, part of the plant, storage conditions and possibly more. Some regions of China are more exposed to discharge of wastewaters, solid and gas waste which are dumped to the environment as a result of mining and smelting industry, traffic, fertilization, soil amendment with sludge from wastewater treatment plants, inadequate irrigation practices which raise concerns regarding their impact on environment, plants cultivation and inhabiting people 3,45 . Gao et al. showed that soil from polluted areas and the herbs growing there are severely contaminated, in comparison to soils and herbs from unpolluted areas 3 .
According to WHO the maximum permissible amounts in dried plant material are 10 mg kg −1 of Pb and 0.3 mg kg −1 of Cd 46 . WHO did not propose limits for any other element, however, examples of national limits were proposed by several countries (for example Canada, China, Malaysia, Korea, Singapore, Thailand) and are within ranges: 2-5 mg kg −1 of As, 10-20 mg kg −1 of Pb, 0.3-1 mg kg −1 of Cd 47 . Considering the proposed limits, in most cases the maximum amounts of elements causing toxic effects is not exceeded in the tested roots of herbs. None of the investigated herb roots exceeded 10 mg kg −1 of Pb, and the maximum value of 9.1 mg kg −1 was measured for herb 12A (Morindae officinalis). In 7 herbs the content of Cd exceeded 0.3 mg kg −1 (13% of all tested samples) with the maximum value of 0.80 mg kg −1 for herb 26B (Vladimiriae soulier). The highest content of As-0.86 mg kg −1 -was obtained for the herb 7B (Codonopsis) without exceeding the national limits.
The proper manner of preparing the medicinal herbs is the infusion, and the risk of introducing high amounts of harmful ingredients, including elements, is usually low due to the low extraction efficiency. However, the extraction efficiency varies substantially in different herbs and also different parts of plant and can be boosted in acidic conditions, i.e. by addition of lemon juice 48 . For black tea infusions the extractions efficiency of As, Cd, Pb, Ni, Sr and Ba were estimated as < 36.8%, < 40.3%, < 58.6%, 0.1-100%, 0.6-27.9% and 0.4-7.6%, respectively 49 . Wang et al. found that the extraction efficiency for Chinese herbal medicines was in range 6-62% for Cd, 0-8% for Pb and 1-15% for As 50 . The recommended dose of some herbal medicines is up to 30 g per single use and the patient may have been prescribed the multiple doses a day. The prolonged ingestion of elements causing toxic effects, especially As, Cd and Pb, daily through weeks or months may contribute to systemic and chronic toxicity. Although, the maximum permissible amounts per mass unit of dried herb in this study is not exceeded in most specimens, the use of larger amounts of herbal material to prepare the infusion will proportionally increase the amounts of ingested harmful ingredients. The herbal medicines may also be consumed directly as a www.nature.com/scientificreports/ functionalized, nourishing food known as medicine food homologous (MFH) which may considerably increase the intake of hazardous elements 29 .
The roots of herbs in this study were purchased in two terms 1 year apart and the results of both batches were taken into statistical analysis. For some herbs the differences between two batches were small, for example for herbs 20, 21, 10, 9, and for some were relatively large, for example for herbs 7, 1, 22, 3. The evaluation of complex measurement results for numerous elements and samples require the advanced multivariate statistical methods which will be discussed in detail in next chapters.
Chemometric evaluation. The multielemental analysis of herbs provided a large dataset of 13 elements content in 52 samples. Such large datasets are difficult to interpret with basic statistics, so the multivariate methods were applied. The multivariate statistics allow to reduce the multidimensionality of data in order to find the most prominent relationships among the measured values for all tested herbs. The applied multivariate methods are: CVA based on interactions and interaction effects and multiple regression analysis.
Canonical variates analysis. The analysis of canonical variates was applied to show, simultaneously, the variability among all tested herb roots and between two herb roots of the same species, i.e. the same product purchased twice, 1 year apart. Some variability is expected, possibly due to the delivery of a completely different batch, but could also be the same batch that has been stored for a year in a storage facility. However, the content of the determined elements should not be expected to change after 1 year of storage of the herb roots, as it is a well-dried material and it is assumed that it was properly stored. The analyzed elements are not volatile and no losses during storage should occur. Drying is the standard way of preserving herbs which allow to keep the original characteristics and renders them available in any season 51 . Figure 1 shows what is the distribution of variability of concentrations of elements among two batches of herbs. The dimensionality reduction resulted in two coordinates which explain the 61.3% of total variability of determined elements in tested herb roots from two batches. The greater the variability, the longer the arrow. Dashed lines connect points that represent roots of herbs purchased in two terms. The length of this line indicates differences between two batches of the same herb. In terms of the content of elements, the herb roots differ substantially in the following years when the points for two batches are farther apart. The interaction effects are intended to show the effects of variables (type of herb and batch number) on the content of element and are presented in Fig. 2. The distance between the graph origin (the general mean) and the given point in graph (experimental object) is the Mahalanobis distance. The exact values of Mahalanobis distance for all points representing herbs from two batches presented in Figs. 1 and 2 are gathered in supplementary Table S1. The interaction effects for individual elements are presented in supplementary Fig. S1.
In some cases, the points corresponding to the two batches of roots of the same herb overlap or are very close together (see Fig. 1), for example 2, 4, 9, 10, 15, 25, so differences in element content between two batches are relatively small. The largest Mahalanobis distances between two points are for herbs 3, 7, 12, 18, 24. For those herbs the differences of content of element between two batches are relatively large. For most pairs of points, the direction of line connecting them is similar to the direction of arrows related to Mg and Mn, which indicates  Fig. 1). In terms of intergroup relationships, Mn and Mg have the inverse relationship, while Ba and Fe have positive. A small relationship of Fe with Mn and Zn is observable, but the increase in Fe and Ba content has no significant effect on Mg content. Herb 7 shows relatively large differences of element content between two batches but the interaction effect is not large. And conversely, in herb 23 (Scrophulariae) the differences between two batches are not large (see Fig. 1), but the interaction effects graph show the considerable differences (see Fig. 2). The CVA analysis allowed to find the herb roots which are outliers in terms of content of elements and also helped to reject those results from the dataset resulting in the model of the regression analysis with n = 41 . The limiting distance form the origin of the graph that allowed to reject 20% of herb roots is outlined by green circle in Fig. 1.
The results presented in the interaction effects graph suggest that the data interpretation should not be done based solely on the average of tested samples (herb roots) or on the average of the batches purchased 1 year apart. The analysis of data should not be restricted only to the specific herb or to the specific batch. When analyzing the data of elemental content in numerous herbs gathered in batches one has to take into account the variability associated with different herbs in analyzed samples, as well as different batches, simultaneously. There are differences of content of elements in tested herb roots, as well as in two batches, and the changes of contents of elements are not predictable and dependent from each other. When the interaction effect is significant then the values of variables do not change proportionally. The CVA shows that by including the additional batch of herbs the data inference show more profound relationships among determined elements in roots of herbs and the conclusions of the experiment may be more deep and wide.
The CVA described in this study seems to be of practical use with handling the results of analysis as a continuous monitoring process of a product in specified time intervals. This is especially valid when the changes in product composition or other characteristics has impact on the effect for consumer and may pose risk for health when it does not meet the specification.
Regression analysis. The regression analysis was performed in order to find relationships between the determined elements in herb roots. The elements were divided into groups so that each element essential for humans was compared individually with a group of elements causing toxic effect. Two approaches to regression analysis, divided into 2 batches of samples of different sizes ( n = 52 and n = 41 ), were taken and compared with each other resulting in four models of regression. Model n = 52 contains all measurement results of 26 herbs from  Depending on the type of regression all elements causing toxic effects will be included in model or some of them will be excluded. The stepwise regression with backward elimination is also a multiple regression but some of the elements causing toxic effect are eliminated from the model, depending on the desired empirical significance level p, which in this study is 0.05. The results of 4 models of the regression analysis are presented in Fig. 3. Each physiological element is compared with elements causing toxic effect with the resulting regression coefficient. The column outlined by a dashed line shows the relation between two elements that is significant. The stepwise regression eliminates variables that are the least influencing to the total variation, therefore, the missing column may occur for 2 models: stepwise regression n = 52 and n = 41 . The results of regression analysis are presented in supplementary  www.nature.com/scientificreports/ The calculated regression parameters for the four models show numerous similarities for some elements, which proves a significant relationship between them, regardless of the regression model. The multiple and stepwise regression show similar results but only in a specific group of samples: n = 52 or n = 41 . The direction of relationship is concordant in most cases, as is presented by four models of regression in Fig. 3. The only difference in significance is visible for pairs of elements: Co-Cd and Zn-As for n = 52 and Cu-Pb for n = 41 . Much pronounced differences are visible when comparing groups n = 52 with n = 41 where many differences are visible in graph. The most prominent are V-Pb, Mn-Pb, Co-Cd, Cu-Pb, Cu-Cd, Cu-As and all elements with Zn show more or less clear differences. Those elements have some non-standard or less predictable relationships, in comparison to other physiological elements. This suggests that the rejection of outliers, as is the case in group n = 41 , have much more notable impact on relationships between elements in studied samples than changing the type of regression from multiple to stepwise with backward elimination.
There are also several correlations which are significant for all 4 models of regression. Vanadium is correlated with arsenic ( 0.66 < β < 0.76 ) and with barium ( 0.16 < β < 0.25 ), Iron is correlated with arsenic ( 0.69 < β < 0.85 ), and cobalt is correlated with nickel ( 0.36 < β < 0.65 ) and negatively correlated with strontium ( −0.49 < β < 0.35 ). There were samples with much higher content of certain element compared to the rest of samples, even over an order of magnitude. Such high concentration values in a single or few samples may have large effect on existing relationships between elements. The regression analysis in the model n = 41 after removal of outliers showed relationships that did not exist or were not significant in the model n = 52 , e.g. Mg-Ni, V-Pb, Cu-Pb, and also Zn with all elements causing toxic effect. The elements causing toxic effects that have weak and the least significant relationships with physiological elements are Ni and Cd. Nickel is positively and significantly related only with Co, according to four regression models. The model n = 41 shows that Cd is in inverse relationship with Zn, while according to the model n = 52 , Cd has positive relationship with Cu. Arsenic has strong and significant relationship with V and Fe. Nkansah et al. estimated the Pearson correlation coefficients for elements in black and green tea from Ghana and the significant and positive correlation were found for Zn-Cd ( r = 0.625 ) and Zn-As ( r = 0.591 ), however, in this study the stepwise regression ( n = 41 ) for those pairs showed also significant relationships, but in the opposite direction 31 .

Conclusions
In this study the ICP-MS technique allowed to determine 13 elements, both physiological (Mg, V, Mn, Fe, Co, Cu, Zn) and causing toxic effects (Ni, As, Sr, Cd, Ba, Pb) in 26 roots of herbs originating from China which are used in TCM. The herbs were acquired from online shop in two batches 1 year apart to check the variation of elemental composition of the same product bought in time interval. The analytical procedure of sample preparation and analysis by ICP-MS was developed and validated. The use of ICP-MS with octopole reaction system and helium gas allowed to minimize the spectral interferences arising from matrix and argon gas. The application of matching-matrix CRMs allowed to estimate the parameters characterizing the performance of the analytical method and assured the traceability of measurement result. The multivariate statistical methods, such as the analysis of canonical variates and interaction effects, and multiple regression analysis allowed to find the relationships between the determined elements in roots of herbs, as well as between the two batches bought 1 year apparat. The obtained results showed the broad range of content of determined elements among all examined species of herbs.
The CVA allowed to show the magnitude of differences between two batches of herbs, taking into account each determined element and showing that for some herbs the differences were much more prominent than in others. The statistical inference should not be based solely on the type of herb or number of batch because of the underlying interaction effects between those two variables that may be a source of variability of the content of elements. The multiple regression analysis revealed the significant relationships between elements in all tested herb roots: Mg with Sr; V with Pb, As and Ba; Mn with Pb; Fe with As and Ba; Co with Ni and Sr, Cu with Pb, Cd and As; Zn with Pb, Cd, As and Ba.
The only element causing toxic effects that was found to excess the maximum permissible limit was Cd. The maximum permissible amount of Cd in dried plant material proposed by WHO is 0.3 mg kg −1 and this limit was exceeded in 7 herbs which makes 13% of all tested samples. No other element causing toxic effects was above the permissible limit, however, the prolonged consumption of roots of herbs and the infusions as the therapy in TCM may potentially be a cause of detrimental effects on health. Even if the maximum permissible limits are not reached, the recommended dose of certain herb roots can reach up to 30 g, and the multiple doses over a long time of therapy may constitute a considerable source of elements causing toxic effects like As, Cd, Pb or Ni that can develop chronic toxicity in organism.