Journal of Environmental Management 203 (2017) 383e390

Contents lists available at ScienceDirect

Journal of Environmental Management

journal homepage: www.elsevier.com/locate/jenvman

Research article

Multivariate soil fertility relationships for predicting the

environmental persistence of 2,4,6-trinitrotoluene (TNT) and 1,3,5-
trinitro-1,3,5-tricyclohexane (RDX) among taxonomically distinct soils
Chelsea K. Katseanes a, 1, Mark A. Chappell b, *, 1, Bryan G. Hopkins a, Brian D. Durham b,
Cynthia L. Price b, Beth E. Porter b, Lesley F. Miller b
a
Department of Plant and Wildlife Sciences, Brigham Young University, Provo, UT, USA
b
Environmental Laboratory, US Army Engineer Research & Development Center, Vicksburg, MS, USA

a r t i c l e i n f o a b s t r a c t

Article history: After nearly a century of use in numerous munition platforms, TNT and RDX contamination has turned up
Received 8 February 2017 largely in the environment due to ammunition manufacturing or as part of releases from low-order
Received in revised form detonations during training activities. Although the basic knowledge governing the environmental fate
2 August 2017
of TNT and RDX are known, accurate predictions of TNT and RDX persistence in soil remain elusive,
Accepted 4 August 2017
particularly given the universal heterogeneity of pedomorphic soil types. In this work, we proposed
overcoming this problem by considering the environmental persistence of these munition constituents
(MC) as multivariate mathematical functions over a variety of taxonomically distinct soil types, instead of
Keywords:
2,4,6-Trinitrotoluene
a single constant or parameter of a specific absolute value. To test this idea, we conducted experiments
1,3,5-Trinitro-1,3,5-tricyclohexane where the disappearance kinetics of TNT and RDX were measured over a >300 h period in taxonomically
Soil fertility distinct soils. Classical fertility-based soil measurements were log-transformed, statistically decomposed,
Partial least squares regression and correlated to TNT and RDX disappearance rates (k-TNT and k-RDX) using multivariate dimension-
Contaminant environmental persistence reduction and correlation techniques. From these efforts, we generated multivariate linear functions
for k parameters across different soil types based on a statistically reduced set of their chemical and
physical properties: Calculations showed that the soil properties exhibited strong covariance, with a
prominent latent structure emerging as the basis for relative comparisons of the samples in reduced
space. Loadings describing TNT degradation were largely driven by properties associated with alkaline/
calcareous soil characteristics, while the degradation of RDX was attributed to the soil organic matter
content e reflective of an important soil fertility characteristic. In spite of the differing responses to the
munitions, batch data suggested that the overall nutrient dynamics were consistent for each soil type, as
well as readily distinguishable from the other soil types used in this study. Thus, we hypothesized that
the latent structure arising from the strong covariance of full multivariate geochemical matrix describing
taxonomically distinguished “soil types” may provide the means for potentially predicting complex
phenomena in soils.
Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction thermal and physical shock. Yet, the widespread deployment of

Developing TNT (2,4,6-trinitrotoluene) and RDX (1,3,5-trinitro-
1,3,5-tricyclohexane) explosive compounds was an important 20th
century innovation to military and industrial technology, providing
marked improvements in explosive yield and stability against
* Corresponding author.
E-mail address: mark.a.chappell@usace.army.mil (M.A. Chappell).
1
Co-first authors.
these munition constituents (MC) resulted in substantial soil and
groundwater legacy contamination associated with manufacture
and military training, storage, and demolition activities. These
legacy contamination issues were previously documented in detail
over the last three decades (Binks et al., 1995; Jenkins et al., 2001;
Pennington et al., 2001; Hewitt et al., 2007; Clausen, 2011) across a
variety of military training and demolition ranges as well as
ammunition plants, with extensive efforts undertaken to control,
predict, and mitigate on-site contamination.

http://dx.doi.org/10.1016/j.jenvman.2017.08.005
0301-4797/Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
384 C.K. Katseanes et al. / Journal of Environmental Management 203 (2017) 383e390
Discovering and parameterizing the biogeochemical processes
controlling the fate and persistence of MC are viewed as key to
predicting their behavior in the soil environment. Typically, the
major processes controlling MC environmental fate are partitioned
in terms of the sorption, biodegradation rate, and transport pa-
rameters in soils. Previously, we demonstrated the limitations of
predicting TNT and RDX sorption purely from the perspective of
humified organic matter phases in soils and sediments (Chappell
et al., 2011a, and references therein) citing uncertainties of three
to four orders of magnitude for the distribution coefficient (KD)
describing TNT and RDX in soils (a similar range existing for KOC
values representing the KD values normalized by the soil total
organic carbon, TOC concentration). We argued that estimating MC
sorption through KOC values represented a univariate over-
simplification of soil surface-promoted “partitioning” behavior.
Multilinear regression modeling of sorption data emphasized the
importance of soil Ca and Fe content, the cation exchange capacity
(CEC), and percent clay composition and, for TNT, the insignificance
of TOC (Chappell, 2011; Chappell et al., 2011a; Bridges et al., 2017).
Katseanes et al. (2016) expanded these functions using partial least
squares regression (PLS) modeling on a sample set containing a
more extensive soil characterization matrix, finding that soil K, %
CaCO3, and the soil buffering capacity for exchangeable NHþ
(PBCNH4) - a soil Quantity-Intensity based parameter as statistically
significant.
While there exists a good knowledge of the fundamental
mechanisms driving the biodegradation of TNT and RDX, theoret-
ically rigorous representations of these mechanisms in the complex
multiphasic soil system remain challenging. TNT and RDX can
degrade via a number of different abiotic reductive mechanisms
(Price et al., 1995, 1997; Devlin et al., 1998; Emmrich, 1999, 2001;
Hofstetter et al., 1999; Hansen et al., 2003; Nefso et al., 2005;
O'Sullivan et al., 2011), yet the MC environmental persistence is
largely mediated by soil/sediment biodegradation via microbio-
logical activity (McCormick et al., 1976, 1981; Kaplan and Kaplan,
1982; Bradley and Chapelle, 1995; Comfort et al., 1995; Price
et al., 1995; Lewis et al., 1996, 1997; Riefler and Smets, 2000;
Spain et al., 2000; Esteve-Nunez et al., 2001; Walker et al., 2006;
Yost et al., 2007; Rosen and Lotufo, 2010; Chappell et al., 2011b;
Montgomery et al., 2011). All of these compounds degrade in
contact with the soil solid phase, but the degradation rate is
enhanced at higher pH and lower Eh values, characteristic of basic,
electron-rich environments. Accordingly, the commonly identified
degradation products represent basic amino or aminohydroxy de-
rivatives (Price et al., 2001; Thorn and Kennedy, 2002; Yost et al.,
2007) consistent with reduction reactions detailed in the scienti-
fic literature (Thorn et al., 2002; Qasim et al., 2007). Yet, with all this
mechanistic information, it remains difficult to confidently predict
MC persistence in soils. For example, the distribution of published
MC degradation parameters in soils demonstrated a high uncer-
tainty associated with the calculated degradation coefficients (k,
h1) in soil (Bridges et al., 2017). The values for kTNT and kRDX were
0.226 ± 0.924 (median ¼ 0.0087, min ¼ 0.0003, max ¼ 4.36) and
0.030 ± 0.064 (median ¼ 0.008, min ¼ 9  105, max ¼ 0.240) for
TNT and RDX, respectively. By comparison, RDX generally exhibits a
greater resistance against degradation in soil suspensions than TNT,
showing usually minimal degradation potential under aerobic soil
conditions, but requiring prolonged periods of anaerobic soil con-
ditions (Shen et al., 2000; Price et al., 2001; Larson et al., 2008)
system Eh.
The main purpose of this study was to investigate our hypoth-
esis that soil fertility data served as a predictor of the degradation
kinetics of TNT and RDX across varying range of taxonomically
distinct soils. In agriculture, Liebig's Law of the Minimum provides
a useful conceptual theory for connecting nutrient deficiencies to
limits in optimal crop yield, stating that maximum crop yield is not
determined necessarily by the concentration of all resources (such
as nutrients, water, etc.), but by those that are the most limiting.
Applying Lieberg's Law to the incidental MC biodegradation in soil,
one would expect the existence of nutrient threshold concentra-
tions that limit microbial activity in soils to also influence MC
persistence. Currently, we are not aware of any empirical functions
for k-TNT and k-RDX related to soil properties, although the influence
of major soil components such as water, C, N, and P on TNT and RDX
biodegradation rates have been reported previously. For example,
elevated MC degradation rates were reported with additions of
various C substrates (McCormick et al., 1984; Boopathy, 2002;
Fahrenfeld et al., 2013) that stimulated microbial activity. On the
other hand, studies also showed that addition of N sources alone
such as nitrate and ammonium reduced MC degradation rates
(Sherburne et al., 2005; Beller, 2002; Singh et al., 2008; Bernstein
et al., 2011). In these studies, easily degradable N sources pre-
cluded MC degradation, as there was less of a need for microor-
ganisms to break down the more recalcitrant MC molecules.
Combinations of multiple nutrient deficiencies are also possible.
For example, both C and N could be limiting factors that affect
microbial activity in the same soil. In this case, the N limitation
4 could promote microbial-mediated MC degradation due to a lack of
more easily degradable sources. Limiting C concentrations will
overall inhibit the growth of microbial populations and biotically
mediated transformations in soils. There also could also be further
complex interactions/limitations due to deficient micronutrient
concentrations. It is possible for deficiencies of Fe or Mn, for
example, to be the main limiting factor for biologic growth e a
focus of soil fertility evaluations of agricultural soils but rarely
considered when evaluating microbial driven degradation of
xenobiotics.
Here, we show ordination-based methods to be particularly
useful for comprehending different soil and environmental samples
characterized with extensive physical and chemical attribute
matrices, viewing soils as a formulated material composed of a
variety of constituents. In this work, we focus on rigorous and
explicit applications to draw out the latent structure among soil
fertility variables and develop multivariate models predicting TNT
and RDX degradation in different soil types. The development of
predictive multivariate models tuned to a local soil's taxonomic
designation would have direct benefit to military range managers
seeking to interpret the environmental risks of training activities on
sites.
2. Materials and methods
2.1. Soil selection and characterization
Non-contaminated soils were collected from various locations in
the United States representing different soil types and inherent
fertility characteristics: Catlin (Fine-silty, mixed, superactive, mesic,
Oxyaquic Argiudolls), Playas-Saltair (Fine-silty, mixed, mesic Typic
Salorthids), Ruston (Fine-loamy, siliceous, semiactive, thermic
Typic Paleudults), Skumpah (Mixed, mesic Typic, Torripsamments),
Smithdale (Fine-loamy, siliceous, subactive, thermic Typic Haplu-
dults), and Sunev soil (Fine-loamy, carbonatic, thermic Udic Cal-
ciustolls). All soils were collected and used in experiments within
one calendar year. Details of the processing, preparation, and
characterization of these soils are reported in Katseanes et al.
(2016).
2.2. Munition degradation experiments
Batch incubations were conducted using soil suspensions in
 C.K. Katseanes et al. / Journal of Environmental Management 203 (2017) 383e390
airtight reactors created from 2.8-L Fernbach flasks following the
procedure of Chappell et al. (2011b). Each flask was equipped with a
rubber stopper modified to include a septum, Eh, pH, temperature
probes, and a sampling tube (See Fig. S1). Soil was added in a 1:10
soil-to-water ratio using double-deionized water. The background
electrolyte was set by the equilibrium concentration of salts
released from the suspended soils. The Fernbach flasks were placed
in a water bath with an automatic cooling system to maintain
temperature. Sediments were kept in suspension with a magnetic
stirrer. Background levels of Eh, pH, temperature, and CO2 were
measured before spiking the systems with TNT or RDX. Reactors
were covered with aluminum foil prior to adding dissolved muni-
tion constituents to prevent incidental photodegradation of com-
pounds. Solutions were equilibrated for three days before adding a
spike solution of 1.3 mL of 40 mg TNT or RDX mL1 in acetone to
reach an overall initial concentration of approx. 20e25 mg TNT or
RDX L1 in the reactor. Suspension and gas samples were taken
with time after the addition of munition. The reactors were oper-
ated as atmospherically closed systems but were opened daily for
10 min to limit the development of anaerobic conditions. Collected
suspension samples were filtered through 0.45 mm filters to sepa-
rate solids from liquids. Dissolved TNT and its degradation products
2-amino-4,6-dinitrotoluene (2ADNT) and 4-amino-2,6-
dinitrotoluene (4ADNT) and RDX concentrations were measured
using by EPA Method 8330B on a HP1100 Series LC/MSD system
high-pressure liquid chromatography (HPLC) system (Hewlett
Packard Company, Palo Alto, CA, USA). Additional carbon dioxide,
dissolved organic carbon, and nutrient measurements made in the
batch systems are described in the Supporting Information section.
Collected suspensions were also centrifuged to collect solids, and
extracted using a 50:50 acetonitrile/water solution as described by
EPA Method 8330B.
2.3. Data modeling
Kinetic data for MC disappearance were modeled by least-
squares fit using a first-order rate law. The rate of disappearance
(kMC) was calculated from the following:
MCðtÞ ¼ MCasym þ MC0 ekMC t
where, MC (t) ¼ concentration MC at each time point,
MCasym ¼ asymptote of the modeled line, MC0 ¼ modeled original
MC concentration (at t ¼ 0), and k-MC ¼ the first order rate constant
describing MC (either TNT or RDX) disappearance. Additionally, the
slope (k-1) of the initial linear portion of the kinetic curve was
calculated by:
k1 ¼ MC0  kMC
For this work, the degradation rate data were not normalized by
MC0 (as is commonly done during the kinetic modeling) in order to
preserve the quantitative differences in the absolute values of the
rate constants among the different soils. Thus, kMC and k-1 were
defined in units of mg L1 h1. Thus, the persistence half lives (t1/2)
were calculated by:
t1=2 ¼ ð0:697=kMC Þ  MCorginal
where, MCoriginal is the actual initial concentration of MC at the
beginning of the experiments (t ¼ 0), thus distinguishing it from
the fitted parameter MC0 in Eq. (1).
Correlations between latent structure contained within soil
fertility characterization data and kMC were explored using multi-
variate Partial Least Squares (PLS) regression. Data transformations
385
(log10), PLS modeling calculations, and variable selection were
described previously (Katseanes et al., 2016), except we limited
transformations to those variables with high ratios between their
minimum and maximum values and were highly skewed (based on
test of normality) (Eriksson et al., 2013) The data were also exam-
ined using an orthogonal partial least squares (OPLS) algorithm
(Trygg and Wold, 2002), which verified that none of the Y-
explained variance in the PLS prediction models was attributed to
non-correlated (and orthogonal) structure in X.
3. Results
3.1. Soil fertility interpretation
A detailed chemical and physical assessment of the soils was
discussed previously (Katseanes et al., 2016). To summarize, the
Catlin and Sunev Mollisols were considered the most fertile soils
used in this study, (likely to support the greatest microbiological
activity) while the Ruston and Smithdale Ultisols were the most
acid and possessed the lowest macronutrient levels, and lowest CEC
values e properties typical of soils from the Southeastern U.S., and
were expected to support much lower microbiological activity
relative to the Mollisol soils. The Playas-Saltair and Skumpah soils
were categorized as sodic and saline-sodic soils, respectively, and
considered to be the least fertile, greatly limiting microbiological
activity due to their alkaline pH, high %CaCO3, and soluble salts, as
well as likelihood to promote abiotic alkaline hydrolysis of TNT
(Hansen et al., 2003, and references therein).
3.2. Kinetics of MC disappearance
Kinetic data for the disappearance of TNT and RDX in the
different soil suspensions (Fig. 1) were, for the most part, well-
described by the first-order kinetic model (Tables 1 and 2). The
initial fate of MC disappearance was rapid (note that MC concen-
tration at t ¼ 0 represents the calculated initial concentration) as
represented by k-TNT-1 and k-RDX-1. For TNT, k-TNT-1 values were
highest in the Bonneville Salt Flats soils (37e38 mg L1 h1) fol-
lowed by the fertile Mollisols (1e2 mg L1 h1) and then the poorly
(1) fertile Ultisols (0.4e0.6 mg L1 h1). For RDX, k-RDX-1 values were
highest for the fertile Mollisols soils (6e20 mg L1 h1), following
Bonneville Salt Flats soils (0.3e0.5 mg L1 h1), and again, the
poorly fertile Ultisols (0.007e0.012 mg L1 h1). In general, the kTNT
and kRDX values were less than ki-1 values (assumed related to the
previously reported sorption values), but generally followed the
same order in terms of soil types. For TNT, the highest k values were
attributed to the Bonneville Flats soils (2.3 mg L1 h1), followed by
the fertile Mollisol soils (0.102e0.114 mg L1 h1) and then the
(2) poorly fertile Ultisols (0.037e0.044 mg L1 h1). Note that this
trend is generally reflected in the calculated initial disappearance
rate, k-TNT-1. For RDX, the highest k values were observed for the
fertile Mollisol soils (1.4e2.0 mg L1 h1) followed by the Bonne-
ville soils (0.087e0.174 mg L1 h1) and the poorly fertile Ultisol
soils (0.012e0.014 mg L1 h1).
The implication of alkaline hydrolysis mechanisms in the
degradation of TNT seemed obvious in these experiments. We
(3) attempted to clarify the degree in which persistence was driven by
abiotic vs. biotic conditions by simultaneously measuring different
soluble solutes and CO2 (assuming soil microbiological respiration)
generated by the experiments. For TNT, we observed clear increases
in the concentration of 4ADNT with time for the Catlin, Smithdale,
and Sunev soils (Fig. S-1). Since alkaline hydrolysis results in the
same degradation products as biodegradation, 2ADNT was
measured in both Bonneville Flat soils.
386 C.K. Katseanes et al. / Journal of Environmental Management 203 (2017) 383e390

Fig. 1. (A) Measured TNT and (B) RDX concentrations in solution with time.

Table 1
Fitted kinetic parameters for TNT degradation based on Eq. (1).

Soil TNTasym, (mg L1) TNT0 (mg L1) k-TNT (mg L1 h1) Adjusted r2 kTNT-1 (mg L1 h1) t1/2 a(h)

Catlin <0.01 20.00 ± 0.14 0.114 ± 0.030 0.99965 2.290 121

Playas-Saltair 3.67 ± 0.12 16.32 ± 0.51 2.273 ± 0.880 0.99331 37.103 6
Ruston 9.81 ± 0.96 10.04 ± 1.02 0.044 ± 0.018 0.94583 0.443 315
Skumpah 3.27 ± 0.31 16.73 ± 0.33 2.314 ± 0.506 0.99793 38.721 6
Smithdale 4.40 ± 2.31 15.38 ± 2.34 0.037 ± 0.015 0.93384 0.567 375
Sunev 7.56 ± 2.47 11.75 ± 2.60 0.102 ± 0.054 0.84205 1.21 136
a
Calculated by t1/2 ¼ [Ln(2)/k-TNT]  original TNT.

Table 2
Fitted kinetic parameters for RDX degradation based on Eq. (1).

Soil RDXasym, (mg L1) RDX0 (mg L1) k-RDX (mg L1 h1) Adjusted r2 kRDX-1 (mg L1 h1) t1/2 a
(h)

Catlin 12.43 ± 0.36 10.59 ± 0.67 1.954 ± 0.469 0.965 20.692 8

Playas-Saltair 19.15 ± 0.23 6.31 ± 2.88 0.087 ± 0.067 0.704 0.549 202
Ruston 19.19 ± 0.09 0.87 ± 0.09 0.014 ± 0.005 0.940 0.012 996
Skumpah 19.73 ± 0.21 1.46 ± 0.81 0.174 ± 0.46 0.466 0.254 83
Smithdale 13.96 ± 0.11 0.57 ± 0.13 0.012 ± 0.008 0.694 0.007 830
Sunev 18.03 ± 0.21 4.97 ± 0.50 1.367 ± 1.696 0.914 6.794 12
a
Calculated by t1/2 ¼ [Ln(2)/k-RDX]  original RDX.

3.3. Nutrient and CO2 fluctuations during the kinetic experiments
Headspace measurements showed clear increases in CO2
(Fig. S.3) for the fertile Catlin and Sunev Mollisols during the in-
cubation experiments for both TNT and RDX. We attributed these
CO2 increases as an indicator of biomass activity occurring within
the soils e a potential basis for distinguishing between (incidental)
biotic and abiotic degradation. Furthermore, we observed fluctua-
tions in certain dissolved solutes as indicators of bioactivity, such as
decreases in NH4-N, NO3-N, NO2-N, TOC, PO4, and SO4 with time in
some of the soils (Figs. S.4e7), especially the more fertile Mollisols
e the exception being the very alkaline Playas-Saltair, where NH4-N
losses most likely resulted from the high pH driving ammonia
volatilization. Increasing concentrations of exchangeable cations,
such as Na, Ca, K, and Mg (Figs. S.8e9) suggested that the continual
mechanical stirring promoted mineral dispersion and dissolution.
Fluctuations in the dissolved Fe and Mn suggested that the system
redox (Eh) were generally poised by these heavy metal oxides
(Figs. S.10e11). Follow-up measurements to closely monitor pH, Eh,
and temperature (Fig. S.12) showed that the soil systems remained
generally aerobic over the duration of the incubations.
 C.K. Katseanes et al. / Journal of Environmental Management 203 (2017) 383e390 387

3.4. Multivariate modeling: MC disappearance kinetics and soil Table 3

fertility characteristics Fitting statistics for the PLS model predicting the persistence (k-TNT and k-RDX) for TNT
and RDX for the taxonomically distinct soils used in this study.

PLS modeling was used to correlate the most statistically Parameter k-TNT k-RDX
important soil fertility variables for predicting TNT and RDX Slope 0.9996 1.001
degradation rates in the batch experiments. Based on the shape of Offset (mg L1 h1) 0.010 6.034  105
the validation curve, we chose a two-factor PLS model (Fig. S.13), RMSEC [RMSEV] 0.3576 [0.5976] 0.3812 [0.4668]
R2 0.9417 0.8774
which explained 89% of the variance in the y-data. Fig. 2 shows that
PLS generated linear calibration functions (statistical diagnostics
presented in Table 3). The distribution of data points along the
linear trend line remained skewed due to the three-order of
magnitude range in kMC values in spite of mathematical pre-
processing (e.g., log transformations). For TNT, the Bonneville Salt
Flats Playas-Saltair and Skumpah soils represented the upper
portion of the regression curve (presumably due to abiotic hydro-
lysis), while the upper portion of regression curve for RDX was
driven by the fertile Mollisols. Thus, RMSE values (Table 3) were
sufficiently high so that only the kMC values could be predicted with
confidence on the Salt Flats soils for TNT and the Mollisol soils for
RDX. kMC values for soils at the bottom of the calibration curve
could not be statistically distinguished among one another,
reflecting an inherent limitation in the small dataset used for this
study.
In spite of the relatively high RMSE values (limiting the reso-
lution of the predictions), substantial latent structure was apparent
in the score and loading plots for the PLS model. For example, the
Salt Flats soils were negatively loaded in Factor 1 (Fig. 3A) while the

Fig. 3. Scores (A) and X-Y loading (B) plots showing the distribution of samples in
reduced space in Factors 1e2 (representing 89% of the explained variance in the data)
from PLS-based pedo-informatic models correlating the modeled persistence rate
constants, k-TNT and k-RDX for TNT and RDX to the chemical and physical properties of
the taxonomically distinct soils selected for this study.

non-saline/sodic soils were positively loaded in Factor-1. Note the

Fig. 2. Correlation curves for the 2-Factor PLS-based pedo-informatic functions pre-
dicting the measured and predicted modeled persistence rate constants, k-TNT and k-RDX
for (A) TNT and (B) RDX to the chemical and physical properties of the taxonomically
distinct soils selected for this study.
clustering of loadings for the alkaline soil characteristics (pH, high %
CaCO3, soluble salts, CaP, Cl, Na, S, K) and kTNT in the negative
reduced space of Factor 1 (Fig. 3B). On the other hand, the explained
variance in Factor 2 may be explained in terms of the soil nutrient
and other constituent concentrations. Thus, the soils with higher
concentrations of constituents were positively loaded in Factor 2,
and the soils low in constituents negatively loaded in Factor 2. In
particular, kRDX mapped with the constituents that correlated with
the Catlin and Sunev Mollisols. Here, the PLS modeling pointed to
the importance of the soil fertility and this connection to soil bio-
logical activity (implied in the incidental biodegradation of the TNT
and RDX compounds). These relationships were not be identified
based on constituent concentrations alone, as demonstrated pre-
viously using principal components analysis (PCA) where the
388 C.K. Katseanes et al. / Journal of Environmental Management 203 (2017) 383e390

alkaline soil properties clustered with the other fertility-based the overall model. Yet, even with this limitation, the regression
constituent concentrations (Katseanes et al., 2016). function suggested an inherent connection of soil microbiological
Using stability tests (Martens and Næs, 1989), we calculated activity to the potential decomposition of soil organic carbon.
functions for MC degradation rates for the different soil types based
on the statistically significant regression coefficients for the two-
3.5. Soil functionality profiles
factor validated model. For TNT, kTNT ¼ 0.071 NH4-N  0.035
NO3-N þ 0.0763 (log Na) þ 0.0700 B þ 0.0617 (log Cl) þ 0.0796 %
Whenever testing different soils, we commonly assume that
CaCO3, predicting TNT degradation kinetics in the different soils as
each soil exhibits behavior bound within certain external limits.
directly proportional to particular saline/sodic soil properties and
Thus, each soil type is expected to exhibit its own characteristic
inversely related to inorganic nitrogen species as described previ-
behavior, regardless of the actual MC transformation rates. To
ously. Thus, both abiotic and biotic degradation processes were
demonstrate this, we conducted PCA on the previously discussed
apparently captured in the prediction function. Interestingly, the
solution-phase data to determine if there was sufficient latent
regression equation emphasized the importance of soil B for pre-
structure to distinguish the soils. We calculated an overall 5-PC
dicting kTNT, yet follow-up studies conducted by us did not articu-
model explaining 85% of the variance in the data. The score plot
late this relationship (data not shown). Our previous PCA of the soil
(Fig. 4A) showed separation of the samples based on which MC was
properties (Katseanes et al., 2016) showed that soil B was highly
tested: soils from the RDX incubation experiments loaded posi-
correlated with soil CaCO3, but OPLS analysis (described in the
tively in PC 1 (37 %EV) while soils used in the TNT experiments
Materials and Methods section) did not detect orthogonal structure
negatively loaded in PC 1. The loading plot (Fig. 4B) showed that
in X contributing to the functions describing k-MC.
these differences were largely driven by the higher CO2 levels
For RDX, kRDX ¼ 0.102 %OM. Here, organic matter represented
evolved during the TNT incubation experiments relative to the
the only statistically significant variable for predicting kRDX, most
experiments with RDX. In PC 2 (27 %EV) we observed the similar
likely given the low number of samples (and thus low degrees of
separation of the soils with the saline/sodic soils positively loaded
freedom) and the bias introduced by the two saline/sodic soils on
in PC 2, the acid Ultisols negatively loaded in PC 2, and the fertile

Fig. 4. PCA results from a 5-PC model explaining 84.5% of the data variance in the solution-phase constituents measured during the TNT and RDX soil incubations experiments. (A)
Score plot comparing PC 1e2, (B) corresponding loading plot for the variables for PC 1e2; (C) score plot after rotating the matrix by 90 to give PC 2e3 (C) corresponding loading
plot for PC 2e3.
 C.K. Katseanes et al. / Journal of Environmental Management 203 (2017) 383e390
Mollisols clustered in the middle of the plot, as reported above. But,
it can be seen that the soil types readily clustered regardless
whether the incubation involved TNT or RDX when the matrix was
rotated to show PC 2e3 (Fig. 4C). Here, the more fertile Mollisols
were positively loaded in PC 3 (6.9 %EV) and the less fertile soils
negatively loaded in PC 3. Thus, the latent structure in the solution
phase data suggests that the inherent behavior of the different soil
types was preserved even if there were notable differences in the
overall soil biological activity driving MC degradation.
This study showed how the latent structure contained in clas-
sical soil fertility data was used to generate prediction functions for
the degradation of MC in soils. Without application of a design of
experiments (DOE) approach, the PLS functions cannot be claimed
as statistically “causal”. Yet, the fact that the latent structure in the
soil characterization data correlated with the complex processes
driving the chemical transformation rates of TNT and RDX is
notable, particularly given the low degrees of freedom of this small
dataset. Similar to our previous work, this supports our theory that
the latent structure contained in soil characterization data may be
key to overcoming the large uncertainty in environmental models
for predicting the complex biogeochemical processes in soils e this
uncertainty largely attributed to soil's nearly universal spatial
heterogeneity. Thus, exploring this latent structure in soil charac-
terization data opens up opportunities to impose PLS models onto
hierarchical frameworks, such as pedomorphological soil classifi-
cation systems e an approach we referred previously to as Pedo-
informatics (Chappell, 2016). Current research is underway to
explore multivariate models using larger soil datasets in order to
enhance the resolution (decrease the RMSE) in the PLS prediction
models. These datasets are currently being generated using soils
collected from different suborders (with the NRCS soil classification
system).
Acknowledgements
The use of trade, product, or firm names in this report is for
descriptive purposes only and does not imply endorsement by the
U.S. Government. The tests described and the resulting data pre-
sented herein, unless otherwise noted, were obtained from
research conducted under the Environmental Quality Technology
Program of the US Army Corps of Engineers by the US Army Engi-
neer Research and Development Center (ERDC). Permission was
granted by the Chief of Engineers to publish this information. The
findings of this report are not to be construed as an official
Department of the Army position unless so designated by other
authorized documents. The authors express gratitude to Dr. Eliz-
abeth Ferguson, Technical Director of the US Army ERDC Environ-
mental Quality Technology Program for support of this research.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.jenvman.2017.08.005.
References
Beller, H.R., 2002. Anaerobic biotransformation of RDX (hexahydro-1,3,5-trinitro-
1,3,5-triazine) by aquifer bacteria using hydrogen as the sole electron donor.
Water Res. 36, 2533e2540.
Bernstein, A., Adar, E., Nejidat, A., Ronen, Z., 2011. Isolation and characterization of
RDX-degrading Rhodococcus species from a contaminated aquifer. Biodegra-
dation 22, 997e1005.
Binks, P.R., Nicklin, S., Bruce, N.C., 1995. Degradation of hexahydro-1,3,5-trinitro-
1,3,5-triazine (RDX) by Stenotrophomonas maltophilia PB1. Appl. Environ.
Microbiol. 61, 1318e1322.
Boopathy, R., 2002. Effect of food-grade surfactant on bioremediation of explosives-
contaminated soil. J. Hazard. Mater. 92, 103e114.
389
Bradley, P.M., Chapelle, F.H., 1995. Factors affected microbial 2,4,6-trinitrotoluene
mineralization in contaminated soil. Environ. Sci. Technol. 29, 802e806.
Bridges, T.D., Lotufo, G.R., Chappell, M.A., Price, C.L., George, R.D., Glisch, E., 2017.
Review and Synthesis of Evidence Regarding Environmental Risks Posed by
Munitions Constituents (MC) in Aquatic Systems. U.S. Army Engineer Research
& Development Center, Vicksburg, MS.
Chappell, M.A., 2011. Solid-phase considerations for the environmental fate of TNT
and RDX in soil. In: Chappell, M.A., Price, C.L., George, R.D. (Eds.), Environmental
Chemistry of Explosives and Propellant Compounds in Soils and Marine Sys-
tems: Distributed Source Characterization and Remedial Technologies. Amer-
ican Chemical Society, pp. 1e25.
Chappell, M.A., 2016. In: Pedo-informatic Theory for Predicting Complex Processes
in Soils. USACE-ERDC, U.S.A.
Chappell, M.A., Price, C.L., Miller, L.F., 2011a. Solid-phase considerations for the
environmental fate of nitrobenzene and triazine munition constituents in soil.
Appl. Geochem. 26, S330eS333.
Chappell, M.A., Price, C.L., Porter, B.E., Pettway, B.A., George, R.D., 2011b. Differential
kinetics and temperature dependence of abiotic and biotic processes control-
ling the environmental fate of TNT in simulated marine systems. Mar. Pollut.
Bull. 62, 1736e1743.
Clausen, J.L., 2011. Energetic residue observations for operational ranges. In:
Chappell, M.A., Price, C.L., George, R.D. (Eds.), Environmental Chemistry of Ex-
plosives and Propellant Compounds in Soils and Marine Systems: Distributed
Sources Characterization and Remedial Technologies. American Chemical So-
ciety, Washington, DC, pp. 107e136.
Comfort, S.D., Shea, P.J., Hundal, L.S., Li, Z., Woodbury, B.L., Martin, J.L., Powers, W.L.,
1995. TNT transport and fate in contaminated soil. J. Environ. Qual. 24,
1174e1182.
Devlin, J.F., Klausen, J., Schwarzenbach, R.P., 1998. Kinetics of nitroaromatic reduc-
tion on granular iron in recirculating batch experiments. Environ. Sci. Technol.
32, 1941e1947.
Emmrich, M., 1999. Kinetics of the alkaline hydrolysis of 2,4,6-trinitrotoluene in
aqueous solution and highly contaminated soils. Environ. Sci. Technol. 33,
3802e3805.
Emmrich, M., 2001. Kinetics of the alkaline hydrolysis of important nitroaromatic
co-contaminants of 2,4,6-trinitrotoluene in highly contaminated soils. Environ.
Sci. Technol. 35, 874e877.
Eriksson, L., Byrne, T., Johansson, E., Trygg, J., Vikstro€m, C., 2013. Multi-and Mega-
variate Data Analysis: Basic Principles and Applications, third ed. Umetrics
Academy, Malmo € , Sweden.
Esteve-Nunez, A., Caballero, A., Ramos, J.L., 2001. Biological degradation of 2,4,6-
trinitrotoluene. Microbiol. Mol. Biol. Rev. 65, 335e352.
Fahrenfeld, N., Zoeckler, J., Widdowson, M.A., Pruden, A., 2013. Effect of bio-
stimulants on 2,4,6-trinitrotoluene (TNT) degradation and bacterial community
composition in contaminated aquifer sediment enrichments. Biodegrad 24,
179e190.
Hansen, L.D., Larson, S.L., Davis, J.L., Cullinane, J.M., Nestler, C.C., Felt, D.R., 2003.
Lime Treatment of 2,4,6-trinitrotoluene Contaminated Soils: Proof of Concept
Study. U.S. Army Corps of Engineers, Vicksburg, MS.
Hewitt, A.D., Jenkins, T.F., Walsh, M.E., Walsh, M.R., Bigl, S.R., Ramsey, C.A., 2007.
Protocols for Collection of Surface Soil Samples at Military Training and Testing
Ranges for the Characterization of Energetic Munition Constituents. U.S. Army
Engineer Research and Development Center, Hanover, NH.
Hofstetter, T.B., Heijman, C.G., Haderlein, S.B., Holliger, C., Schwarzenbach, R.P., 1999.
Complete reduction of TNT and other (poly)nitroaromatic compounds under
iron-reducing subsurface conditions. Environ. Sci. Technol. 33, 1479e1487.
Jenkins, T.F., Pennington, J.C., Ranney, T.A., Berry Jr., T.E., Miyares, P.H., Walsh, M.E.,
Hewitt, A.D., Perron, N.M., Parker, L.V., Hayes, C.A., Wahlgreen, E.G., 2001.
Characterization of Explosives Contamination at Military Firing Ranges. U.S.
Army Engineer Research & Development Center, Hanover, NH.
Kaplan, D.L., Kaplan, A.M., 1982. Thermophilic biotransformations of 2,4,6-
trinitrotoluene under simulated compositing conditions. Appl. Environ.
Microbiol. 44, 757e760.
Katseanes, C.K., Chappell, M.A., Hopkins, B.G., Durham, B.S., Price, C.L., Porter, B.E.,
Miller, L.F., 2016. Multivariate functions for predicting the sorption of 2,4,6-
trinitrotoluene (TNT) and 1,3,5-trinitro-1,3,5-tricyclohexane (RDX) among
taxonomically distinct soils. J. Environ. Manag. 182, 101e110.
Larson, S.L., Martin, W.A., Escalon, B.L., Thompson, M., 2008. Dissolution, sorption,
and kinetics involved in systems containing explosives, water, and soil. Environ.
Sci. Technol. 42, 786e792.
Lewis, T.A., Ederer, M.M., Crawford, R.L., Crawford, D.L., 1997. Microbial trans-
formation of 2,4,6-trinitrotoluene. J. Ind. Microbiol. Biotechnol. 18, 89e96.
Lewis, T.A., Goszczynski, S., Crawford, R.L., Korus, R.A., Admassu, W., 1996. Products
of anaerobic 2,4,6-trinitrotoluene (TNT) transformation by Clostridium bifer-
mentans. Appl. Environ. Microbiol. 62, 4669e4674.
Martens, H., Næs, T., 1989. Multivariate Calibration. John Wiley & Sons, Inc., New
York.
McCormick, N.G., Cornell, J.H., Kaplan, A.M., 1981. Biodegradation of hexahydro-
1,3,5-trinitro-1,3,5-triazine. Appl. Environ. Microbiol. 42, 817e823.
McCormick, N.G., Cornell, J.H., Kaplan, A.M., 1984. The Fate of Hexahydro-1,3,5-
trinitro-1,3,5-triazine (RDX) and Related Compounds in Anaerobic Denitrify-
ing Continuous Culture Systems Using Simulated Waste Water. Army Natck
Research and Development Center.
McCormick, N.G., Feeherry, F.E., Levinson, H.S., 1976. Microbial transformations of
2,4,6-trinitrotoluene and other nitroaromatic compounds. Bull. Environ.
390 C.K. Katseanes et al. / Journal of Environmental Management 203 (2017) 383e390
Contam. Toxicol. 28, 33e38.
Montgomery, M.T., Boyd, T.J., Smith, J.P., Walker, S.E., Osburn, C.L., 2011. 2,4,6-
Tritnitrotoluene mineralization and incorporation by natural bacterial assem-
blages in coastal ecosystems. In: Chappell, M.A., Price, C.L., George, R.D. (Eds.),
Environmental Chemistry of Explosives and Propellant Compounds in Soils and
Marine Systems: Distributed Source Characterization and Remedial Technolo-
gies. American Chemical Society, pp. 172e184.
Nefso, E.K., Burns, S.E., McGrath, C.J., 2005. Degradation kinetics of TNT in the
presence of six mineral surfaces and ferrous iron. J. Hazard. Mater. B123, 79e88.
O'Sullivan, D.W., Denzel, J.R., Luning-Prak, D.J., 2011. Photolysis of 2,4,6-
tritnitrotoluene in seawater: effect of salinity and nitrate concentrations. In:
Chappell, M.A., Price, C.L., George, R.D. (Eds.), Environmental Chemistry of Ex-
plosives and Propellant Compounds in Soils and Marine Systems: Distributed
Source Characterization and Remedial Technologies. American Chemical Soci-
ety, pp. 157e169.
Pennington, J.C., Jenkins, T.F., Brannon, J.M., Lynch, J.C., Berry Jr., T.E., Hayes, C.A.,
Miyares, P.H., Walsh, M.E., Hewitt, A.D., Perron, N.M., Delfino, J.J., 2001. Distri-
bution and Fate of Energetic on DoD Test and Training Ranges: Annual Report 1.
U.S. Army Engineer Research and Development Center, Vicksburg, MS.
Price, C.B., Brannon, J.M., Hayes, C.A., 1995. Transformation of 2,4,6-trinitrotoluene
under Controlled EH/pH Conditions. U.S. Army Engineer Waterways Experi-
ment Station, Vicksburg, MS.
Price, C.B., Brannon, J.M., Hayes, C.A., 1997. Effect of redox potential and pH on TNT
transformation in soil-water slurries. J. Environ. Eng. 123, 988e992.
Price, C.B., Brannon, J.M., Yost, S.L., Hayes, C.A., 2001. Relationship between redox
potential and pH on RDX transformation in soil-water slurries. J. Environ. Eng.
127, 26e31.
Qasim, M.M., Moore, B., Taylor, L., Honea, P., Gorb, L., Leszczynski, J., 2007. Structural
characteristics and ractivity relationships of nitroaromatic and nitramine ex-
plosives - a review of our computational chemistry and spectroscopic research.
Int. J. Mol. Sci. 8, 1234e1264.
Riefler, R.G., Smets, B.F., 2000. Enzymatic reduction of 2,4,6-trinitrotoluene and
related nitroarenes: kinetics linked to one-electron redox potential. Environ.
Sci. Technol. 34, 3900e3906.
Rosen, G., Lotufo, G.R., 2010. Fate and effects of composition B in multispecies
marine exposures. Environ. Toxicol. Chem. 29, 1330e1337.
Shen, C.F., Hawari, J., Ampleman, G., Thiboutot, S., Guiot, S.R., 2000. Enhanced
biodegradation and fate of hexahydro-1,3,5-trinitro-1,3,5-triazine (RDX) and
octahydro-1,3,5,7-tetranitro-1,3,5,7-tetrazocine (HMX) in anaerobic soil slurry
bioprocess. Bioremediat. J. 4, 27e39.
Sherburne, L.A., Shrout, J.D., Alvarez, P.J.J., 2005. Hexahydro-1,3,5-trinitro-1,3,5-
triazine (RDX) degradation by Acetobacterium paludosum. Biodegradation 16,
539e547.
Singh, R., Soni, P., Kumar, P., Purohit, S., Singh, A., 2008. Biodegradation of high
explosive production effluent containing RDX and HMX by denitrifying bacte-
ria. World J. Microbiol. Biotechnol. 25, 269e275.
Spain, J.C., Hugues, J.B., Knackmuss, H.-J. (Eds.), 2000. Biodegradation of Nitro-
aromatic Compounds and Explosives. Lewis Publishers, CRC Press LLC, Boca
Raton, FL.
Thorn, K.A., Kennedy, K.R., 2002. 15N NMR investigation of the covalent binding of
reduced TNT amines to soil humic acid, model compounds, and lignocellulose.
Environ. Sci. Technol. 36, 3787e3796.
Thorn, K.A., Pennington, J.C., Hayes, C.A., 2002. 15N NMR investigation of the
reduction and binding of TNT in an aerobic bench scale reactor simulating
windrow composting. Environ. Sci. Technol. 36, 3797e3805.
Trygg, J., Wold, S., 2002. Orthogonal projections to latent structures (O-PLS).
J. Chemom. 16, 119e128.
Walker, S.W., Oshburn, C.L., Boyd, T.J., Hamdan, L.J., Coffin, R.B., Montgomery, M.T.,
Smith, J.P., Li, Q.X., Hennessee, C., Monteil, F., Hawari, J., 2006. Mineralization of
2,4,6-trinitrotoluene (TNT) in Coastal Waters and Sediments. Naval Research
Laboratory, Washington DC.
Yost, S.L., Pennington, J.C., Brannon, J.M., Price, C.B., 2007. Environmental process
descriptors for TNT, TNT-related compounds and picric acid in marine sediment
slurries. Mar. Pollut. Bull. 54, 1262e1266.