PREDICTION of copper and zinc concentrations in plants based on soil characteristics

Jennifer K. Saxe, Christopher A. Impellitteri and Herbert E. Allen^*

Department of Civil and Environmental Engineering

University of Delaware

Newark, Delaware 19716

^*Corresponding author email: allen@ce.udel.edu

Abstract

The establishment of appropriate soil quality standards for heavy metal species including copper and zinc requires quantification of the aspects of soil chemistry that mitigate or exacerbate their potential availability and subsequent toxicity to terrestrial organisms. Many soil characteristics have been implicated as altering metal availability to plants in soil from that predicted by the total concentration in soil. Lettuce, barley, and mustard were grown in international and domestic field soils with widely varied physical and chemical characteristics, and the plant metal content was determined. An artificial neural network model was used to determine soil characteristics most influential in predicting copper and zinc concentrations in these plant species. Metal levels in soil extracts were more valuable for predicting plant copper and zinc concentrations than were the total soil concentrations. The inclusion of soil organic matter content, particle size distribution, and macronutrient concentrations further refine and improve the model’s predictive ability.

Introduction

Many soil environment characteristics have been implicated in the literature as potentially altering trace metal bioavailability. Some of these include partitioning within soil through cation exchange, specific adsorption, precipitation, and complexation, and solid-solution partitioning factors including pH, redox potential, soil texture, clay content, organic matter content, iron and manganese oxides, and other cations and anions present, and buffer capacity (Rieuwerts et al. 1998). Additionally, plant roots have been shown to change soil chemistry non-negligibly in the rhizosphere through production of chemical root exudates which may affect bioavailability of zinc and copper (Marschner 1991, Marschner 1993).

Using a combination of chemical soil extraction techniques and mathematical modifications accounting for the most important soil characteristics affecting bioavailability of copper and zinc, a predictive model for plant concentrations has been developed. An artificial neural network was employed to iteratively determine the most influential soil parameters for inclusion in the model.

Methods

Seeds of lettuce (Lactuca sativa L.), barley (Hordeum vulgare L.) and Indian mustard (Brassica juncea) were sown in 36 air-dried, sieved field soils of international origin. Plants were grown indoors for 31 days after seedlings emerged from soil at 15 °C under a 14/10 photoperiod before being harvested, oven-dried at 70 °C, microwave-digested in concentrated HNO₃, and analyzed for metals using ICP spectrometry.

Soils were extracted for total metals using EPA Method 3051A. Other chemical extractions were performed at a 10:1 solution:soil ratio and mechanically shaken for 16 hours before being centrifuged, membrane-filtered (0.45 mm), acidified if necessary, and analyzed for metals using ICP spectrometry. Dissolved organic carbon in water extracts was measured using a TOC analyzer. Additional soil characteristics were determined by the University of Delaware Soils Testing Laboratory (Sims and Heckendorn, 1991).

Results and Discussion

Plant tissue concentrations of copper and zinc were plotted against the total copper and zinc in corresponding soils. Total metal was a poor predictor of plant tissue concentrations (Figure 1). Copper was generally maintained in plant tissue in a narrow concentration range (2-30 mg/kg) despite the fact that total copper in the soils spanned three orders of magnitude (1-1095 mg/kg). There is also strong evidence that zinc is well regulated in plant tissue: soil totals ranged from 4 to 3704 mg/kg total zinc while plant tissue ranged only from 10 to 754 mg/kg when grown in these soils. Even mustard, a hyperaccumulator of metals whose tissue concentrations would be expected to correlate most closely with soil totals, was poorly

Figure 1: Copper (A, B) and zinc (C, D) concentration in plant tissue plotted against total copper and zinc concentration in soil. Regression lines are shown. (A, C) Data shown for the entire range over which plants grew. (B, D) Low range data from A and C shown at higher resolution.

approximated by total soil concentrations, although significantly better than for lettuce and barley.

Soil extractions using several extractants commonly chosen as surrogates for bioavailable metal in soil were performed. Extractants included water, HCl added as needed to maintain slurry pH at 4.0, 0.01 M HCl, 0.01 M CaCl₂, Mehlich-1 reagent, and HNO₃ plus H₂O₂ (EPA digestion method 3051A). The copper and zinc concentrations removed from soil using these extractions were compared to the concentrations of these elements in dried plant tissue in order to determine which extraction is most useful in predicting plant concentrations. Correlation coefficients (r) were computed for each species individually and for all species combined. Copper in plants was most closely correlated to the copper concentration in a 0.01 M HCl extract overall (n = 81, r = 0.48), for barley (n = 32, r = 0.50), and for lettuce (n = 29, r = 0.71). Mustard, a metal hyperaccumulator, was most closely approximated by total copper in soil however (n = 20, r = 0.79), and the HCl extract was most poorly correlated with plant copper concentrations (r = 0.63). Correlation coefficients between zinc in plants and in extracts was consistently greatest for the 0.01 M CaCl₂ extraction. This extraction was the best predictor of plant zinc concentrations overall (r = 0.62), and for barley, lettuce, and mustard individually (r = 0.72, 0.70, and 0.83, respectively). Since a 0.01 M HCl extraction and a 0.01 M CaCl₂ extraction were the best predictors of plant concentrations of copper and zinc, the results of these

Table 1: Range of soil parameters for soils in which plants were successfully grown. All variables listed were used in the neural network/parameter-screening model with the exception of the total copper and zinc concentrations. These were replaced by the concentration of copper or zinc in the soil extract which was determined to correlate most strongly to plant metal concentrations: 0.01 M HCl for copper and 0.01 M CaCl₂ for zinc.

Parameter	Maximum	Minimum	Unit
Total Zinc (EPA 3051A Digestion)	3704	4.34	mg/kg
Total Copper (EPA 3051A Digestion)	1095	1.47	mg/kg
Dissolved Organic Carbon in Water Extract	637	38.8	mg/kg
Acid Neutralization Capacity (to pH 4.0)	0.2559	0.0003	moles/L
pH in Water Slurry	8.0	4.1	pH
pH in Buffered Slurry (pH = 8.0)	8.0	7.0	pH
Percent Organic Matter by Walkley-Black Method	10.3	0.2	%
Phosphorus in Mehlich-1 Extract	591.8	1.7	mg/kg
Potassium in Mehlich-1 Extract	539	5.7	mg/kg
Calcium in Mehlich-1 Extract	5561	11	mg/kg
Magnesium in Mehlich-1 Extract	560	2.3	mg/kg
Manganese in Mehlich-1 Extract	311	0.1	mg/kg
Zinc in Mehlich-1 Extract	865	0.1	mg/kg
Copper in Mehlich-1 Extract	178	0.1	mg/kg
Iron in Mehlich-1 Extract	134	0.3	mg/kg
Percent Carbon by Combustion Method	7.714	0.114	%
Percent Sulfur by Combustion Method	0.183	0.003	%
Percent Nitrogen by Combustion Method	0.706	< 0.003	%
Ammonia-Nitrogen	240	6.2	mg/kg
Nitrate-Nitrogen	142	1.1	mg/kg
Percent Sand	92	14	%
Percent Silt	58	2	%
Percent Clay	52	6	%

extractions were used to provide information about the soil-derived in the neural network screening model.

The MATLAB neural network toolbox was used to build a backpropagation single-layer linear neural network using soil characteristics (scaled from zero to one, actual ranges for data shown in Table 1) as input variables and plant tissue concentrations as output variables. Model predictions are shown in Figure 2. Weights, or linear coefficients for soil parameters, after training the model consistently rank pH in buffered solution, macronutrient contents (N,P,K,S) soil texture, and organic matter as being most important in predictions. Their importance is consistent with mechanistic processes discussed in the reviews in the Introduction.

Figure 2: Predicted versus measured plant tissue concentrations for: A. copper and B. zinc. Predictions were made from a multiple-linear neural network derived model considering all species simultaneously and based on the soil parameters listed in Table 1. A. copper extracted by a 0.01 M HCl solution and B. zinc extracted by a 0.01 M CaCl₂ solution were input in lieu of total soil concentrations. A corner-to-corner 45° line would represent an unbiased prediction. Mustard concentrations are slightly underpredicted and barley are slightly overpredicted for both metals. Lettuce concentrations of zinc are underpredicted.

References

Marschner H (1991), In: Plant Roots: the Hidden Half. (Y Waisel et al., Editors), New York, Marcel Dekker, pp. 503-528.

Marschner H (1993), In: Zinc in Soils and Plants. (AD Robson, Editor), Dordrecht, the Netherlands, Kluwer Academic Publishers, pp. 59-77.

Rieuwerts JS, Thornton I, Farago ME, Ashmore MR (1998), Chem. Spec. Bioavail. 10(2): 61-75.

Sims JT, Heckendorn SE (1991), Methods of Soil Analysis, University of Delaware Soil Testing Laboratory.