CORRECTING ESTIMATES OF ROOT CHEMICAL COMPOSITION FOR SOIL CONTAMINATION

H. W. Hunt1,2, D. E. Reuss1 and E. T. Elliott1

1Natural Resource Ecology Laboratory, Colorado State University, Fort Collins, CO 80523, USA
2Department of Rangeland Ecosystem Science, Colorado State University, Fort Collins, CO, 80523, USA

ABSTRACT

The simplest method to correct estimates of root nutrient concentration for contamination by adhering soil is to express concentration on an "ash-free" basis-- total sample nutrient divided by the ash-free weight of the sample. Such ash-free corrections can seriously inflate estimates of root nutrient concentration. Equations exist to correct estimates of plant composition for soil contamination, but they require information on the composition of the contaminating soil. Such information is difficult to obtain because soil fractions adhering to roots after washing may differ from whole soil.

We developed a statistical method to indirectly estimate the composition of adhering soil by eliminating the correlation between estimates of plant composition and the degree of soil contamination. This method was tested against mixtures of grass shoots with soil, and then applied to roots from field-collected soil cores for two grass species. The estimated properties of the contaminating material differed from whole soil, and the differences were great enough to seriously affect root N estimates. The nature of the contaminating material may have to be determined for each new combination of soil and plant species. Our approach requires enough replicate samples for regression analysis, but does not rely on separate soil samples and should be applicable over a wide range of experimental conditions.

Key words: plant chemical composition, soil contamination, ash-free estimates, soil-corrected estimates

INTRODUCTION

Washed or otherwise cleaned samples of roots from plants growing in soil typically have high ash contents of up to 40-60% due to residual adhering soil. Although papers reporting chemical composition of roots often fail to detail the method of calculation used, a common approach is to report concentration on an ash-free basis (e.g., Hansson and Steen 1984, Pettersson et al. 1986). However, the ash-free calculation fails to account for the chemical content of the soil or the ash content of plant tissue, and thus can give misleading estimates. Blair (1988) and Misra (1994) discussed these problems and developed equations to estimate the extent of soil contamination from sample ash content, and to correct estimates of plant nutrient concentration for nutrients contributed by the soil. However, these authors assumed that the contaminating soil was identical to separate samples of whole soil. The properties of soil remaining attached to roots could differ from those of whole soil, because of fractionation during washing. The objectives of our research were to develop a method to indirectly estimate the composition of contaminating soil through an optimization procedure, to apply the estimator to washed root samples from established perennial grasses, and to test the validity of previous assumptions about the composition of contaminating soil.

ESTIMATING EQUATIONS

The fraction ash in a sample of plant tissue contaminated with soil (As) is a function of the ash fraction of the uncontaminated plant tissue (Ap), the ash fraction of the soil (Ao), and the fraction of the sample that is plant tissue (Fp): 

   As  =  Fp*Ap + (1 - Fp)*Ao .       (1)

If the ash contents of the soil and the uncontaminated plant tissue are known, then the fraction plant material in a sample can be estimated from sample ash by rearranging Eq. 1: 

              Ao - As
   Fp   =    _________ .              (2) 
              Ao - Ap

If Ws is the weight of the sample, the ash-free weight of plant tissue (Wpf) is: 

   Wpf   =   Ws*Fp*(1 - Ap) .         (3)

Equation 1 will apply for other components besides ash, for example lignin or nitrogen. Let Xs be the concentration (fraction) of substance X in the sample, Xp that in the plant tissue, and Xo that of the soil. Then we have: 

   Xs  =  Fp*Xp + (1 - Fp)*Xo .       (4)

If the concentration of the substance in soil is known, and the ash content of the sample is used to estimate Fp according to Eq.2, then Eq. 4 can be solved for concentration of the substance in the plant: 

          Xs  -  Xo*(1 - Fp)
   Xp  =  __________________  .       (5)
                  Fp

Concentration of the substance on an ash-free basis (Xpf) is: 

   Xpf  =  Xp/(1 - Ap) .              (6)

We refer to estimates derived using equations 1-6 as "soil-corrected". Although these equations have been in the literature since Blair (1988), most authors either fail to state their method of calculating plant composition, or else calculate an "ash-free" substance concentration (Xa) as the amount of the substance in the sample divided by the ash-free fraction of the sample: 

            Xs
   Xa  =  _______   .                 (7)
          1. - As

Equation 7 is equivalent to Eq. 6 if plant ash is zero, soil ash is 1.0, and the concentration of the substance in soil is zero. The ash-free estimate Xa will overestimate the actual concentration Xpf if the concentration of substance X in the soil on an ash free basis is greater than the concentration of X in the plant on an ash free basis: 

      Xo            Xp
   ________  >   ________    .        (8)
    1 - Ao        1 - Ap
and Xa will underestimate Xpf if the reverse is true. Thus, there may be a systematic bias in ash-free estimates.

The "X" in the terms of Eqs. 4-8 will be replaced below by an "L" when applying the equations to lignin and by an "N" when applying them to nitrogen.

MATERIALS AND METHODS

Ash-free and soil-corrected estimates of plant chemical composition were evaluated using two data sets. In the first, soil and clean grass shoots were mixed together in ratios of 0:10, 1:9, 2:8, 5:5, 8:2, and 10:0 by weight. Lignin concentrations were determined for several replicates of each mixture using the method of Goering and van Soest (1970), which entails digestion in an acid detergent solution to remove cell contents and hemicellulose, followed by 72% sulfuric acid to remove cellulose, and finally combustion to remove lignin. Ash content was taken as the residual material left after combustion. Soil gives a positive test for lignin, about 95% of soil organic matter (data not shown), because humic materials resist acid digestion. Part of the mineral component of the soil must have been dissolved by the acid treatments of the lignin determination, because the ash content of soil was only 82% by this method, compared to about 95% ash determined directly by combustion. This difference presents no problems for the calculations, since the ash contents of soil, plant material and mixtures were all determined using the same method.

Since the lignin content of the plant material was the same in all soil:plant mixtures, the estimate of plant lignin (Lpf, Eq.6), should be independent of sample ash (As). If soil ash (Ao), soil "lignin" (Lo), and shoot ash (Ap) were unknown, they could be estimated by finding the values that eliminate any relationship between sample ash and estimates of plant lignin. Thus, we estimated the values of parameters (Ao, Lo and Ap) necessary to minimize the fraction of variance accounted for (R2) in the regression of Lpf vs. As. Any of several standard optimization procedures (e.g., Newton-Raphson, Ralston 1965; TRUST, Barhen et al. 1997) could be used to accomplish this minimization. We used a simplex method (Nelder and Mead 1965).

The second data set consisted of root samples from two species of perennial grass (Pascopyrum smithii and Bouteloua gracilis) from native shortgrass prairie in NE Colorado. Samples consisted of entire soil plugs with associated vegetation collected by driving steel cylinders (25 cm diameter by 45 cm high) into the ground. Cylinders were either (1) removed and held in growth chambers for two growing seasons under twelve different combinations of atmospheric CO2, temperature and water regimes, (2) left in the ground in the field for two years, or (3) removed and immediately processed. Roots were separated from soil by immersion and soaking in water, gentle agitation by hand and by a water jet while submerged, and decanting and collection on a 4.75 mm mesh screen (large roots) followed by a 0.6 mm screen (small roots). Ash content of the samples was determined by combustion in a muffle furnace, and N concentration by micro-Kjeldahl digestion. Hunt et al. (1996) give a full description of experimental design, chemical methods, and procedures for washing and separating the large and small root fractions. The washed roots appeared to be clean , but ash contents up to almost 80% indicated that attached soil remained. The properties of the attached soil were estimated by minimizing the relationship (R2) between sample N (Npf) and sample ash (As), using the optimization method identified above. It is conceivable that the efficiency of soil removal varied as a function of root system properties such as total biomass or individual root size. Since both root biomass and soil-corrected root N concentration varied with treatment (Hunt et al. 1996), there could be real correlations between root N and sample ash across the whole data set. In other words, a significant regression of root N vs. sample ash could be real and not an artifact to be eliminated. We circumvented this problem by estimating soil properties necessary to minimize the average of the within treatment (across replicate) correlations between root N and sample ash. The number of replicates varied between 8 and 15 among the 14 treatments.

RESULTS

Soil-Shoot Mixtures

Figure 1 compares the ash-free and soil-corrected estimates of plant lignin as a function of sample ash. R2 for the regression of soil-corrected shoot lignin vs. sample ash (omitting pure soil samples) was less than 0.00001. The two methods gave similar results for sample ash up to 20%. For sample ash of 65% (80% soil), the ash-free method overestimated plant lignin by a factor of 2.4. For 100% soil, both methods returned nonsense values for plant lignin. The precision of both estimators declined appreciably as sample ash increased. Table 1 shows good agreement between directly measured and soil-corrected estimates of plant and soil properties.

Roots

The average R2 for regressions of soil-corrected estimates of large root N concentrations vs. sample ash was 0.14 in P. smithii samples and 0.30 for B. gracilis. For P. smithii, one of the 14 regressions was significant (P < .05), no more than expected by chance (14 X .05 = 0.7). In B. gracilis, three of 14 regressions were significant (two positive and one negative slope), indicating some systematic departure from the theory. In the corresponding regressions of ash-free estimates of N concentrations, 27 of 28 slopes in the two species were positive and 11 (all with positive slopes) were significant. Figure 2 shows that the ratio of the ash-free to soil-corrected root N estimates increased with sample ash in both species. Assuming the soil-corrected estimates are accurate, the ash-free method overestimated root N concentration by a factor of up to two or more. The overestimate was by less than 15% for ash contents up to about 0.40, which includes about 93% of the 256 samples included in the analysis. Soil ash determined by combustion was greater than that from the lignin assay, and a sample ash of 0.8 in Fig. 2 corresponds to a sample soil content of about 80%, rather than 100% as in Fig. 1 .

The carbon to nitrogen ratio (C/N) of the organic fraction of the soil was estimated by assuming that the organic fraction was 1.0 minus the ash fraction, that the organic fraction was half carbon, and that all the N was in the non-ash fraction. C/N estimates based on independent estimation of all three parameters (Ao, No and Ap) were unrealistically low. Sensitivity analysis of the equations indicated that estimates of root N varied with the value of No (the N content of the soil) but were almost independent of moderate variation in Ao (soil ash) and Ap (root ash). Therefore we reduced the number of parameters from three to two, by requiring that the C/N ratio of organic matter take more realistic values of 10 in P. smithii soil and 8 in B. gracilis soil, based on data for these soils ( Table 2). This assumption had little effect on root N estimates or regressions of root N vs. sample ash.

The indirectly estimated properties of the material contaminating the washed roots differed from the properties of whole soil (Table 2). To test whether this difference was important to the root N estimates, we applied the soil-correction equations using measured whole soil properties instead of optimized properties. Using whole soil properties led to higher N estimates in both species, and to stronger correlations between N estimate and sample ash in P. smithii (Table 2). The estimated N content of contaminating material was greater than that of whole soil in both species. On the assumption that the sand sized fraction of soil would be more easily washed from roots, we also estimated root N based on the composition of the sand-free (silt plus clay) soil fraction (Table 2). Properties of the sand-free fraction were estimated from known sand contents (56% in the P. smithii soil and 79% in B. gracilis) and from the N and organic matter content of the sand sized fraction of two prairie soils (Anderson et al. 1981). In P. smithii, the N content of the sand free fraction was similar to that estimated indirectly and gave very similar root N estimates. However, in B. gracilis, the N content of the sand-free fraction was appreciably greater than that estimated by optimization, and resulted in some negative N estimates.

We also obtained soil-corrected N estimates for small roots (data not shown), which were analyzed entirely separately from the large roots. As in the case of large roots, the indirectly estimated properties of contaminating material (.958 ash and .00212 N for P. smithii; .976 ash and .00147 N for B. gracilis) were intermediate between those of whole soil and the sand-free fraction.

Range of Bias in Ash-free Estimates

The degree of bias in ash-free estimates will vary in a complex manner with the extent of soil contamination and the difference in concentration between the plant and the contaminating material. To explore these relationships, we applied eqs. 2, 4 and 7 to estimate the bias for some commonly measured plant tissue properties (Table 3). The bias for most substances was positive, suggesting that equation 8 is generally true. Only nonstructural carbohydrates showed a negative bias, and this was significant only in heavily contaminated samples. For the macronutrients N, P and S, the bias in ash-free estimates increased as plant concentration decreased, and bias appeared to be acceptable for moderately clean samples with sample ash of 20%. However, heavily contaminated samples led to gross overestimates. Bias for the cations appeared to exceed that for macronutrients, and was important even in clean samples for Ca, which had appreciably greater concentration in contaminating material than in plant tissue.

The bias values in Table 3 must be considered provisional because (1) little is known about the composition of contaminating material, (2) different chemical methods are often applied to soil and plant samples, and it is uncertain how well data from soil analyses will apply to the contaminating material in plant samples, and (3) soil and plant composition each may vary widely.

DISCUSSION

The bias in ash-free estimates of plant tissue composition might vary greatly depending on the substance measured, chemical method for measurement, root system properties, soil properties, method of washing or cleaning samples, and degree of soil contamination. Thus it is difficult to make generalizations about the conditions under which ash-free estimates are acceptable. However, ash-free estimates may work well for measurements such as total nonstructural carbohydrates which have much greater concentrations in plant tissue than in soil, and for substances with similar concentrations (ash free basis) in the plant and the soil. Ash-free estimates of macronutrients and some cations may be strongly biased in moderately contaminated samples. Intensive washing to remove a high fraction of soil may reduce bias in ash-free estimates, but carries the risks of leaching losses (Misra 1994) and of fragmenting and loss of plant tissues, especially fine roots. On the other hand, insufficient washing may considerably decrease the precision of both ash-free and soil-corrected estimates.

The soil-corrected method can eliminate the systematic biases of ash-free estimates. The main disadvantages of the method are that it requires additional statistical analysis, and requires enough replicate samples for regression of plant composition against sample ash content. It will be difficult to test the assumption that the composition of the contaminating material is not correlated with plant tissue composition. Indeed, the number of significant regressions in the B. gracilis data indicates some departure from theory. A likely explanation for this departure is that texture and soil N varied significantly among replicates, because of spatial variation at the field site where the root samples were collected. Such soil variation would not be a factor in pot studies using homogenized soil. Of course, soil variation and correlations between plant and soil properties will affect ash-free as well as soil-corrected estimates. Thus, we expect that soil-corrected estimates will always be more accurate than ash-free estimates.

No more than about 0.15% of the original soil in the cores remained attached to roots after washing, which indicates the great potential for fractionation between washed and adhering soil. The properties of the contaminating material appeared to be intermediate between whole soil and the sand-free fraction for both large and small roots and in both soil-species combinations. Since the differences between properties of whole soil vs. the sand-free fraction had an appreciable effect on large root N estimates, the nature of contaminating material will have to be determined for each experimental situation. It would be useful to develop a direct method of determining the composition of contaminating material, or somehow to establish statistical relationships between the properties of whole soil and contaminating material.

In the absence of direct information on the composition of the contaminating material, its properties can be estimated indirectly by minimizing the relationship between N estimate and sample ash. This minimization should be carried out across replicates within treatments, to reduce the likelihood of confounding between root properties and washing efficiency. This approach should be widely applicable and has the advantage of not requiring independent information on soil properties. Software for calculating soil-corrected estimates is available.

ACKNOWLEDGMENTS

C. A. Monz contributed to sample processing. J. M. Blair and two anonymous reviewers gave constructive criticism. Research supported by National Science Foundation grants BSR-8818269, BSR87-06429 and DEB-9112571.

LITERATURE CITED