Ac
ce
pte
dP
re
pr
i nt
Plants
PREDICTION OF MONO-, BI- , AND TRI-VALENT METAL CATION RELATIVE
TOXICITY TO THE SEAWEED GRACILARIA DOMINGENSIS (GRACILARIALES,
RHODOPHYTA) IN SYNTHETIC SEAWATER
LUIZ FERNANDO MENDES, LEONARDO ZAMBOTTI-VILLELA, NAIR SUMIE YOKOYA, ERICK LEITE
BASTOS, CASSIUS VINICIUS STEVANI, and PIO COLEPICOLO
Environ Toxicol Chem., Accepted Article • DOI: 10.1002/etc.2340
Accepted Article "Accepted Articles" are peer-reviewed, accepted manuscripts that have not been edited, formatted, or in any way altered by the authors since acceptance. They are citable by the Digital Object Identifier (DOI). After the manuscript is edited and formatted, it will be removed from the “Accepted Articles” Web site and published as an Early View article. Note that editing may introduce changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. SETAC cannot be held responsible for errors or consequences arising from the use of information contained in these manuscripts.
Ac
ce
pte
dP
re
pr
i nt
Plants Environmental Toxicology and Chemistry DOI 10.1002/etc.2340
PREDICTION OF MONO-, BI- , AND TRI-VALENT METAL CATION RELATIVE
TOXICITY TO THE SEAWEED GRACILARIA DOMINGENSIS (GRACILARIALES,
RHODOPHYTA) IN SYNTHETIC SEAWATER
LUIZ FERNANDO MENDES,†* LEONARDO ZAMBOTTI-VILLELA,† NAIR SUMIE YOKOYA,‡ ERICK
LEITE BASTOS,§ CASSIUS VINICIUS STEVANI,§ and PIO COLEPICOLO†
† Instituto de Química, Universidade de São Paulo, Departamento de Bioquímica, 26077, 05599-970
São Paulo, SP, Brazil
‡ Instituto de Botânica, Núcleo de Pesquisa em Ficologia, São Paulo, SP, Brazil
§ Instituto de Química, Universidade de São Paulo, Departamento de Química Fundamental, São
Paulo, SP, Brazil
Running title: Predicting metal toxicity to seaweed by QICAR
* Address correspondence to [email protected].
Additional Supporting Information may be found in the online version of this article.
© 2013 SETAC
Submitted 22 April 2013; Returned for Revisions 17 June 2013; Accepted 24 July 2013
Ac
ce
pte
dP
re
pr
i nt
Abstract: This study reports a 48-h aquatic metal-toxicity assay based on daily growth rates of the red
seaweed Gracilaria domingensis (Gracilariales, Rhodophyta) in synthetic seawater. The median
inhibitory concentration (IC50) for each metal cation was experimentally determined, and the ratios of
free ions (aqueous complex) were calculated by software minimization of the total equilibrium activity
(MINTEQA2) to determine the free median inhibitory concentration (IC50F). A model for predicting
the toxicity of fourteen metal cations was developed using the Generic Function Approximation
algorithm (GFA) with logIC50F values as the dependent variables and the following properties as
independent variables: ionic radius (r), atomic number (AN), electronegativity (Xm), covalent index
(Xm2r), first hydrolysis constant (|logKOH|), softness index (σp), ion charge (Z), ionization potential
(ΔIP), electrochemical potential (ΔEo), atomic number divided by ionization potential (AN/ΔIP), and
the cation polarizing power for Z2/r and Z/AR. The three-term independent variables were predicted as
the best-fit model (logIC50F: - 23.64 + 5.59 Z/AR + 0.99 |logKOH| + 37.05 σp , adj-R2: 0.88, pred-R2:
0.68, Friedman lack-of-fit score: 1.6). This mathematical expression can be used to predict metal-
biomolecule interactions, as well as the toxicity of mono-, bi- and trivalent metal cations, which have
not been experimentally tested in seaweed to date. Quantitative Ion-Character Relationships
(QICARs) allowed us to infer that the mechanism of toxicity might involve an interaction between
metals and functional groups of biological species containing sulfur or oxygen.
Keywords: Daily growth rates, GFA, MINTEQA2, Free ions, QICAR models
Ac
ce
pte
dP
re
pr
i nt
INTRODUCTION
Linear free-energy relationships have been successfully applied for decades to predict the
toxicity of metal cations in several organisms [1-9]. Nevertheless, the potential use of the Quantitative
Ion-Character Relationship (QICAR) to investigate metal toxicity to living organisms has been
underexplored in comparison to the Quantitative Structure-Activity Relationship (QSAR), which is
commonly used for organic toxicants [7,10]. As in the case of QSAR studies with organic compounds,
the physico-chemical properties of metallic species can be used to predict the metal-biomolecule
interactions of biological systems (e.g., tissues, cells, and organelles), as well as to rationalize observed
toxic effects [3,8,9].
The physico-chemical properties of metallic species are caused by their kinetic reactivity/lability,
the affinity and thermodynamic stability of metal-ligand bonds (metal-donor atom affinities), solubility
(metal ion speciation), stereochemistry, redox potential, periodic tendencies in charge/radius,
ionization and polarizability, and Brønsted acid-base equilibria [11]. These physico-chemical
properties can be translated as electronegativity (Xm), or the ability of an atom or functional group to
attract electrons or electron density towards itself in a covalent bond [12,13]; the Pauling ionic radius
(r, Å), the size of an ion in a crystal lattice; the atomic radius (AR, Å), the distance from the nucleus
to the boundary of the surrounding cloud of electrons; the ion charge (Z), the charge on an ion; the
atomic number (AN), the number of protons in the nucleus [14]; the softness index (σp), the
coordinate bond energy of metal fluoride/metal iodide/coordinate bond energy of the metal fluoride or
the tendency for the other electron shell to deform (i.e., polarizability) [7,15,16]; the ionization
potential (ΔIP, eV), the difference in ionization potentials between the ion oxidation number OX and
OX–1 or the energy required to detach an electron in its lowest energy state from a gaseous ground state
atom or molecule; the electrochemical potential (ΔEo, V), the absolute difference in electrochemical
Ac
ce
pte
dP
re
pr
i nt
potential between the ion and its first stable reduced state [17]; and the log of the first hydrolysis
constant (|logKOH|, KOH for Mn+ + H2O MOHn-1 + H+), which is the tendency of a metal cation to
covalently bind to the functional groups of important biomolecules normally containing oxygen-donor
atoms [17,18]. Two parameters can be combined to generate a new one such as the ionic index, or the
ratio of cation Z charge to its ionic radius (Z2/r), which reflects the ionic bond stability of metal-ligand
electrostatic interactions [19,20]. Other parameters of cation polarizing power can be obtained for Z/r,
Z/r2, Z/AR, and Z/AR2 [20]. The electronegativity (Xm) and Pauling ionic radius (r) are composed of
two fundamental ionic characteristics. The result of the last combination, or covalent index (Xm2r),
reflects the cation-binding tendency of soft ligands as sulfur-containing biomolecules [19,21].
Despite the considerable ecological and economic importance of marine seaweeds from the
Gracilaria genus to the pharmaceutical industry and as raw materials for agar production [22-25],
metal toxicity studies using the QICAR approach are scarce. For the present study, we modeled the
effects of some metal cation physico-chemical parameters on their toxicity to a representative species
of the red seaweed Gracilaria domingensis (Kütz.) Sonder ex Dickie in synthetic seawater medium. A
data set of fourteen previously tested metal cations was used to predict the toxicity and cationic-
biomolecule interactions in G. domingensis [24]. The best model was used to estimate the free median
inhibitory concentration (logIC50F) values of metal ions that were not experimentally tested [i.e.,
Ag(I), Cs(I), Ba(II), Hg(II), Fe(II), Fe(III), Cr(III), and Al(III)]. Moreover, based on the results
obtained, we proposed some hypotheses to explain the toxicity of metal cations to G. domingensis at
the molecular level.
Toxicological studies are time-consuming and labor-intensive, demanding trained personal to
maintain the organisms and to perform the experiments. Hence, it is very useful to generate linear free
energy relationships such the ones presented here that can reasonably predict the toxicity of different
classes of substances to a specific target organism. It is also noteworthy to mention that the synthetic
Ac
ce
pte
dP
re
pr
i nt
medium assures complete control over the culture and assay conditions, which is crucial to the
determination of toxicological parameters and experimental replicability [24].
MATERIALS AND METHODS
Determination of daily growth rates (DGR) and free median inhibitory concentration (IC50F)
All free median inhibitory concentration values (IC50F) used in this work were obtained from previous
measurements conducted by our group [25]. In summary, the toxicity of fourteen metal cations to
seaweed was obtained using a 48-h assay based on daily growth rates (DGR, expressed in μg d-1)
[24,25]. The assay consists in add fresh apical segments to metal-free (control) and metal-containing
synthetic culture medium at specific pH and salinity. The DGR values were determined by using the
expression: Δm(mfinal - minitial)/Δt(tfinal - tinitial), where m is the mass and ∆t is 48 h. The average DGR
values of triplicates were used to calculate the IC50 values (fitting growth sigmoidal curves with a
dose-response function). G. domingensis cultures were cultivated and maintained at optimal conditions
as determined previously by a multivariate factorial analysis [25]. The concentration of metal cations
was determined with an ICP-AES according to the standard methods of the U.S. Environmental
Protection Agency (U.S. EPA, method 6010C). Metal distribution (in percentage) to calculate IC50F
values was calculated by computational chemical equilibrium software minimization of the total
equilibrium activity (MINTEQA2, U.S. Environmental Protection Agency), version 3.0, [8,24,26]
under specific experimental conditions.
Model development
The generic function approximation (GFA) technique was performed to generate different Quantitative
Ion-Character Relationship (QICAR) models from the following descriptors (Table S1): Z: ion charge,
AN: atomic number, r: ionic radius in Å, AR: atomic radius in Å, ΔIP: ionization potential in eV, ΔEo:
electrochemical potential in V, Xm: electronegativity, |logKOH|: log of the first hydrolysis constant, σp:
Ac
ce
pte
dP
re
pr
i nt
softness index, Xm2r: covalent index, AN/ΔIP: atomic number divided by ionization potential, and the
cation polarizing power for Z2/r and Z/AR [3-8,18].
The use of the genetic function approximation (GAF) algorithm in QSAR and analogous studies
was introduced by Rogers and Hopfinger [27]. This method for generating statistical models of data
uses the process of evolution and offers many advantages over conventional linear regression analysis
[28]. Moreover, the GFA algorithm creates several models by evolving random initial models, which is
very convenient to analyze data considering a large number of independent variables.
The GFA algorithm (100 random models and 5,000 interactions to evolution) was used to search
the solution space and to identify the most favorable descriptor subsets with which to build QICAR
models. The Friedman lack-of-fit (L.O.F.) score was used to estimate the fitness of each model among
a given number of equations that fit the training set data. Given the small number of descriptors
(fourteen metal ions), the initial model population was set to 100 and the equations consisted of two to
four terms, including a constant. Equations were constructed by considering linear, quadratic and cubic
parameters. The mutation probability, which consists of the addition of a new term, was set to 0.1,
while the smoothing parameter (d) of the L.O.F. expression was set to 0.5. The predictive models were
also evaluated by using a coefficient of determination (R2), adjusted R2 (adj-R2), predicted R2 (pred-
R2), residual error (RMS), and p-value for significance of regression (S.O.R.).
RESULTS AND DISCUSSION
The effects of mono-, bi- and trivalent metals on the seaweed Gracilaria domingensis as
evaluated by DGR (daily growth rate) was previously reported by our group [24]. Table 1 displays the
toxicity of mono-, di-, and trivalent metal cations as total IC50 values as determined from DGR curves
vs. metal concentrations [24]. The IC50 data were converted to free ions (IC50F and logIC50F values)
by MINTEQA2 (computational chemical equilibrium software minimization of total equilibrium
activity) as shown in Table 1. On the basis of the ICF50, Cd(II), Cu(II), Pb(II), Zn(II), Ni(II), Co(II),
Ac
ce
pte
dP
re
pr
i nt
and La(III) were more toxic as measured by a 48-h aquatic metal-toxicity assay to track the growth of
seaweed Gracilaria domingensis in synthetic seawater. As expected, such metals as Li(I), K(I), Na(I),
Mn(II), Sr(II), Ca(II), and Mg(II) presented much lower toxicity.
A set of fourteen common free metal ion toxicity descriptions (which were plotted as the
logIC50F data vs. linear free-energy parameters) was used for the generation of Quantitative Ion-
Character Relationship (QICAR) models. The first ten best-scored models share very similar
information in terms of statistical parameters and types of descriptors (Table 2). These metals can be
classified into three- or four-term equations for independent variables and they contain common
descriptors, such as the cation polarizing power (Z/AR), softness index (σp), and the absolute value of
the log of the first hydrolysis constant |logKOH|. All listed predictive equations in Table 2 were
determined to be statistically significant at α = 0.05. Although some three-term models (e.g., 2 and 5)
had adj-R2 values (0.88 and 0.87, respectively) that were slightly better with respect to the remaining
models, the models were ultimately classified according to their Friedman lack-of-fit (L.O.F.) values.
The L.O.F. values increased from 1.6 (equation 1) to 2.6 (equation 10), and the RMS went from 0.8 to
1.5 when fitting data for all metal cations on the QICAR models. Additionally, by using a minimum
L.O.F., the best fitting mathematical approach was obtained for a three-term independent variable
(equation 1: logIC50F: -23.64 + 5.59 Z/AR + 0.99 |logKOH| + 37.05 σp). In several studies the covalent
index is an important parameter to predict metal-ligand binding tendencies and thus their toxicity to the
target organism [7-9,29]. Surprinsingly, the correlation determined with the parameter Xm2r was very
weak (R2 = - 0.40).
Equation 1 was applied and a strong correlation (R = 0.94, R2 = 0.88) was found between the
experimental toxicity of the training set [as log(IC50F) values], and a response predicted toxicity
(Figure 1) as determined by GFA (Table 2) was observed (Supporting Information, Table S2). The
individual correlation between physico-chemical parameters displayed in Table S1 and logIC50F
Ac
ce
pte
dP
re
pr
i nt
values in Table 1 showed that the smaller the σp and |logKOH| of a metal ion (Figure S1), the smaller its
logIC50F values in the synthetic seawater medium. In contrast, metal cation logIC50F values increased
proportionally with Z/AR (Figure S1).
Based in the soft index model, metal cations can be separated into three classes according to ion
softness, namely, hard, soft and borderline (Table 1) [30]. The results of the present study indicate that
soft ions (i.e., Cd), which form complexes with sulfhydryl groups, and borderline ions (i.e., Cu, Pb, Zn,
Co, and Ni), which form more stable complexes on S-donor atoms compared to O-donor atoms, were
more toxic to seaweed than hard ions (i.e., Ca, Li, Sr, Mg, Na, and K), which form more stable
complexes on O-donor atoms compared to S-donor atoms [8,31]. The exception is La(III) ions (hard),
which are more toxic than Mn(II) (borderline) ions. The observed toxicity pattern can also be
rationalized as follows: i) the higher the hardness of the metal cation, the lower its polarizability, and
the higher its electronegativity, which yields a feature of ionic binding; ii) the higher the softness of the
metal cation, the higher its polarization, the lower its electronegativity and the more covalent the bond,
which can be associated with soft-based donor atom interactions in biological systems [8,31].
Moreover, the softness parameter can be translated as the metal cation's ability to donate its valence
electrons [4]. Hence, the metal cation toxicity of mainly soft and borderline ions is directly
proportional to its potential for a change of oxidation state.
A combination of Z/AR and σp descriptors or variables would be related to the static effective
polarizability of the ion [32,33]. On the other hand, the correlation between σp and |logKOH| allowed us
to speculate that some metal ions might covalently bind to intermediate ligands bearing functional
groups containing S- and O-donor atoms (such as -SH, S2-, RSH, RS, R2S, -OH, -COOH, -PO42-, -O-,
CO, ROH, H2O, NO-, ROSO-, and others groups) [6,8], which are of vital importance to the proper
protein function during the algal life cycle. It was reported that the presence of Cd(II) and Pb(II) ions
decreases superoxide dismutase levels and induced the increase in glutathione and phytochelatin levels
Ac
ce
pte
dP
re
pr
i nt
as a consequence of covalent binding to S-donor atoms [34,35], which can compromise the defense
mechanism against oxyradicals and finally lead to cellular damage [36]. On the other hand, hard ions
(La, Ca, Li, Sr, Mg, Na, and K) have low or very low toxicity, which most likely reflects a less specific
toxic action, which might involve membrane depolarization, osmosis or electrolyte redox imbalance.
Toxicological assays are time-consuming, labor intensive and expensive. Hence, it is very
important to find other methods to estimate toxicity, which are based on fewer toxicity assays. With
this goal, it is possible to predict the toxic effects of nontested metal cations. The training set used to
develop these prediction models contains mono-, di-, and trivalent cations, so it was used to predict the
metal cation toxicity of Ag(I), Cs(I), Ba(II), Hg(II), Fe(II), Fe(III), Cr(III), and Al(III) (Figure 2, Table
S3). The following toxicity order was predicted within the test set for metal ions that were not
experimentally tested (logIC50F): -10.3 Hg(II) >> -5.3 Ag(I) > -4.2 Fe(III) >> -2.3 Cr(III) >> 1.6
Al(III) >> -0.5 Fe(II) > 1.3 Cs(I) > 1.6 Ba(II).
The Person’s softness index (σp) has been successfully applied to predict covalent interactions
between metals and sulfur-donor ligands in biological systems for several organisms, such as mice (R2
= 0.66) and fruit flies (R2 = 0.62) [37], bacteria (R2 = 0.73)] [17], crustaceans (R2 = 0.89), rainbow
trout (R2 = 0.82), and fathead minnow (R2 = 0,84)] [9,10]. The logarithm of the first hydrolysis
constant, or |logKOH|, has been used to explain metal cation stability, toxicity mode of action and
pathways involving O-donor ligands to Vibrio fischeri in the Microtox assay (R2 = 0.93) [3], nematode
(R2 = 0.89) [4], and sunflower (R2 = 0.77).
The results for G. domingensis indicate the importance of electrostatic interaction strength on
cation-ligand complexation and covalent interactions on metal-ligand complexation of sulfur- and
oxygen-donor atoms (soft and borderline ions). In conclusion, the toxicity observed for 22 different
mono-, di-, and trivalent metal cations to this macroalga could be predicted by a function of three
independent physico-chemical variables, namely, Z/AR, σp and |logKOH|.
Ac
ce
pte
dP
re
pr
i nt
SUPPLEMENTAL DATA
Tables S1–S3.
Figure S1. (306 KB DOC).
Acknowledgment—The authors are grateful for the financial and technical support for this study from
the following institutions and people: Ministério de Ciência e Tecnologia, CNPq, NAP-Biodiversidade
Marinha, INCT-Redoxoma, Fundação de Amparo à Pesquisa do Estado de São Paulo-FAPESP:
09/54718-4 (L.F.M), and NAP-PhotoTech (the USP Research Consortium for Photochemical
Technology) from C. V. Stevani.
Ac
ce
pte
dP
re
pr
i nt
REFERENCES
1. Shaw WHR. 1961. Cation toxicity and the stability of transitionmetal complexes. Nature
192:754-755.
2. Kaiser KL. 1980. Correlation and prediction of metal toxicity to aquatic biota. Can J Fish Aquat
Sci 37:211-218.
3. Newman MC, McCloskey JT. 1996. Predicting relative toxicity and interactions of divalent
metal ions: Microtox bioluminescence assay. Environ Toxicol Chem 15:275-281.
4. Tatara CP, Newman MC, McCloskey JT, Williams PL. 1997. Predicting relative metal toxicity
with ion characteristics: Caenorhabditis elegans LC50. Aquat Toxicol 39:279-290.
5. Newman MC, McCloskey JT, Tatara CR. 1998. Using metal-ligand binding characteristics to
predict metal toxicity: quantitative ion character-activity relationships (QICARs). Environ Health
Perspect 106:1419-1425.
6 Wolterbeek HT, Verburg TG. 2001. Predicting metal toxicity revisited: general properties vs.
specific effects. Sci Total Environ 279:87-115.
7. Ownby DR, Newman MC. 2003. Advances in quantitative ion character-activity relationships
(QICARs): using metal-ligand binding characteristics to predict metal toxicity. QSAR Comb Sci
22:241-246.
8. Mendes LF, Bastos EL, Stevani CV. 2010. Prediction of metal cation toxicity to the
bioluminescent fungus Gerronema viridilucens. Environ Toxicol Chem 29:2177-2181.
9. Zhou D-M, Li L-Z, Peijnenburg WGM, Ownby DR, Hendriks AJ, Wang W, Li D-D. 2011. A
QICAR approach for quantifying binding constants for metal-ligand complexes. Ecotoxicol Environ
74:1036-1042.
Ac
ce
pte
dP
re
pr
i nt
10. Walker JD, Enache M, Dearden JC. 2003. Quantitative cationic-activity relationships for
predicting toxicity of metals. Environ Toxicol Chem 22:1916-1935.
11. Hoeschele JD, Turner JE, England MW. 1991. Inorganic concepts relevant to metal binding
activity, and toxicity in a biological system. Sci Total Environ 109/110:477-492.
12. Pauling L. 1932. Nature of the chemical bond. IV. The energy of single bonds and the relative
electronegativity of atoms. J Am Chem Soc 54:3570-3582.
13. Allred AL. 1961. Electronegativity values from thermochemical data. J Inorg Nucl Chem
17:215-221.
14. Brown TL, Lemay HEJr, Bursten BE, Murphy CJ, Woodward PM. 2008. Chemistry: The central
science, 11th ed. Prentice Hall, Englewood Cliffs, NJ, USA.
15. Jones MM, Vaughn WK. 1978. HSAB theory and acute metal ion toxicity and detoxification
processes. J Inorg Nucl Chem 40:2081-2088.
16. Williams MW, Turner JE. 1981. Comments on softness parameters and metal ion toxicity. J
Inorg Nucl Chem 43:1689-1691.
17. McCloskey JT, Newman MC, Clark SB. 1996. Predicting the relative toxicity of metal ions using
ion characteristics: Microtox® bioluminescence assay. Environ Toxicol Chem 15:1730-1737.
18. Charles FB, Mesmer JRE. 1986. Hydrolysis of Cations, 2nd ed. Wiley-InterScience, London,
United Kingdom.
19. Nieboer E, Richardson DHS. 1980. The replacement of the nondescript term 'heavy metals' by a
biologically and chemically significant classification of metal ions. Environ Pollut Ser B1:3-26.
20. Zamil SS, Ahmad S, Choi MH, Park JY, Yoon SC. 2009. Correlating metal ionic characteristics
with biosorption capacity of Staphylococcus saprophyticus BMSZ711 using QICAR model. Bioresour
Technol 100:1895-1902.
Ac
ce
pte
dP
re
pr
i nt
21. Tatara CP, Newman MC, McCloskey JT, Williams PL. 1998. Use of ion characteristics to
predict relative toxicity of mono-, di- and trivalent metal ions: Caenorhabditis elegans LC50. Aquat
Toxicol 42:255-269.
22. Eklund BT, Kautsky L. 2003. Review on toxicity testing with marine macroalgae and the need
for method standardization-exemplified with copper and phenol. Mar Pollut Bull 46:171-181
23. Torres MA, Barros MP, Campos SCG, Pinto E, Rajamani S, Sayre R, Colepicolo P. 2008.
Biochemical biomarkers in algae and marine pollution: a review. Ecotoxicol Environ Saf 71:1-15.
24. Mendes LF, Zambotti-Villela L, Marinho-Soriano E, Colepicolo P, Stevani CV, Yokoya NS.
2013. Metal cation toxicity in the alga Gracilaria domingensis as evaluated by the daily growth rates
in synthetic seawater. J Appl Phycol DOI: 10.1007/s10811-013-0036-1
25. Mendes LF, Vale LAS, Martins AP, Yokoya NS, Soriano EM, Colepicolo P. 2012. Influence of
temperature, light and nutrients on the growth rates of the macroalga Gracilaria domingensis in
synthetic Seawater using experimental design. J Appl Phycol 4:1419-1426.
26. Mendes LF, Stevani CV. 2010. Evaluation of metal toxicity by a modified method based on the
fungus Gerronema viridilucens bioluminescence in agar medium. Environ Toxicol Chem 29:320-326.
27. Rogers D, Hopfinger AJ. 1994. Application of genetic function approximation to quantitative
structure-activity relationships and quantitative structure-property relationships. J Chem Inform Comp
Sci 34:854-866.
28. Rogers D. 1999. Genetic function approximation: evolutionary construction of novel,
interpretable, nonlinear models of experimental data. In Rational drug design. The IMA volumes in
mathematics and its applications, vol. 108, pp 163-189.
29. Brady JM, Tobin JM. 1995. Binding of hard and soft metal ions to Rhizopus arrhizus biomass.
Enzyme Microb Technol 17:791-796.
30. Pearson RG. 1963. Hard and soft acids and bases. J Am Chem Soc 85:3533-3539.
Ac
ce
pte
dP
re
pr
i nt
31. Pearson RG, Songstad J. 1967. Application of the principle of hard and soft acids and bases to
organic chemistry. J Am Chem Soc 89:1827-1835.
32 Gasteiger J, Hutchings MG. 1984. Quantification of effective polarisability. Applications to
studies of X-Ray photoelectron spectroscopy and alkylamine protonation. J Chem Soc Perkin Trans
2:559-564.
33. Barry W, Ninham BW, Yaminsky V. 1997. Ion binding and ion specificity: the hofmeister effect
and onsager and lifshitz theories. Langmuir 13:2097-2108.
34. Le Faucheur S, Behra R, Sigg L. 2005. Phytochelatin induction, cadmium accumulation and
algal sensitivity to free cadmium ions in Scenedesmus vacuolatus. Environ Toxicol Chem 24:1731-
1737.
35. Scheidegger C, Behra R, Sigg L. 2010. Phytochelatin formation kinetics and toxic effects in the
freshwater alga C. reinhardtii upon short- and long-term exposure to lead (II). Aquat Toxicol 101:423-
429.
36. Pinto E, Sigaud-Kutner TCS, Leitão MAS, Okamoto OK, Morse D, Colepicolo P. 2003. Heavy
metal-induced oxidative stress in algae. J Phycol 39:1008-1018.
37. Williams MW, Hoeschele HR, Turner JE, Jacobson K, Christie NT, Panton CL, Smith LH,
Witschi HR, Lee EH 1982. Chemical softness and acute metal toxicity in mice and Drosophila.
Toxicol Appl Pharmacol 63:461-469.
Ac
ce
pte
dP
re
pr
i nt
Captions
Figure 1. Predicted vs. observed log(IC50F) for the metal cation speciation training set: logIC50F: -
23.64 + 5.59 Z/AR + 0.99 |logKOH| + 37.05 σp.
Figure 2. Predicted log(IC50F) for the test set.
Ac
ce
pte
dP
re
pr
i nt
Tables Table 1. Total and free ion median inhibitory concentrations (IC50 and IC50F, respectively) were determined for the seaweed Gracilaria domingensis in synthetic seawatera.
IC50 IC50 % of free iona IC50F x % of free ion Metals Classification
(mg L-1) b (mM) b (%) b (mM) b log (IC50F)
Cd(II) Soft ions 3.60 ± 0.02 0.030 ± 0.0002 0.18 54 x 10-6 ± 36 x 10-8 -4.300 ± 0.003 Cu(II) Borderline ions 54.9 ± 0.3 0.90 ± 0.05 0.05 50 x 10-5 ± 3 x 10-6 -3.300 ± 0.003 Pb(II) Borderline ions 120 ± 3 0.60 ± 0.01 1.4 0.0080 ± 0.0002 -2.10 ± 0.01 Zn(II) Borderline ions 41.5 ± 0.7 0.60 ± 0.01 6.9 0.040 ± 0.001 -1.40 ± 0.01 Ni(II) Borderline ions 123 ± 1 2.10 ± 0.02 34.7 0.70 ± 0.01 -0.20 ± 0.01 Co(II) Borderline ions 164 ± 5 2.8 ± 0.1 57.7 1.60 ± 0.06 0.20 ± 0.02 La(III) Hard ions 811 ± 3 5.80 ± 0.02 31.9 1.90 ± 0.01 0.300 ± 0.001 Mn(II) Borderline ions 694 ± 1 12.60 ± 0.01 65.4 8.20 ± 0.01 0.9000 ± 0.0004 Ca(II) Hard ions 759 ± 2 18.90 ± 0.05 60.7 11.50 ± 0.03 1.100 ± 0.001 Li(I) Hard ions 96 ± 1 13.8 ± 0.1 84.2 11.6 ± 0.1 1.10 ± 0.04 Sr(II) Hard ions 4,153 ± 102 48 ± 1 68.8 32.6 ± 0.8 1.50 ± 0.01
Mg((II) Hard ions 2,863 ± 7 118.0 ± 0.3 49.4 58.3 ± 0.2 1.800 ± 0.001 K(I) Hard ions 5,925 ± 283 151 ± 7 85.3 130 ± 6 2.10 ± 0.02
Na(I) Hard ions 3,982 ± 163 173 ± 7 85.5 148 ± 6 2.20 ± 0.02 a In synthetic seawater at pH = 7.5 ± 0.3 over 48 h [24]. b Quantification of the IC50 values were accomplished for all individual metal cations using inductively coupled plasma atomic emission spectroscopy (ICP-AES, Spectro Genesis). b The ratio of free metal cations in solution at IC50 values were estimated by the software MINTEQA2 (Minimization of Total Equilibrium Activity, U.S. Environmental Protection Agency, Washington, DC, version 3.0).
Ac
ce
pte
dP
re
pr
i nt
Table 2. Summary of statistics for the top 10 QICAR models (free metal cations) made by the GFA learner and their regression statistics.
log(IC50F) = R2 adj-R2 pred-R2 RMS L.O.F. p-value 1 -23.64 + 5.59 Z/AR + 0.99 |logKOH| + 37.05 σp 0.88 0.84 0.68 0.8 1.6 7.7 x 10-5 2 -13.94 • 0.03 AN + 169.19 σp • 431.18 σp σp 0.91 0.88 0.83 0.7 1.9 1.8 x 10-5 3 -14.77 + 159.24 σp • 380.35 σp σp 0.83 0.79 0.76 0.9 1.9 6.8 x 10-5 4 -5.16 + 32.86 σp 0.63 0.60 0.44 1.3 2.0 0.0007 5 -18.09 + 0.41 |logKOH| + 161.02 σp • 431.95 σp σp 0.90 0.87 0.80 0.7 2.0 2.6 x 10-5 6 -2.04 + 1.29 ΔEo 0.61 0.57 0.46 1.3 2.1 0.001 7 -20.50 + 6.61 Z/AR + 34.90 σp + 0.05 |logKOH| |logKOH| 0.89 0.85 0.71 0.8 2.3 4.5 x 10-5 8 -14.81 • 0.19 AN/ΔIP + 172.73 σp • 428.83 σp σp 0.88 0.85 0.75 0.8 2.4 5.7 x 10-5 9 -7.05 + 0.64 |logKOH| 0.55 0.51 0.43 1.4 2.5 0.003 10 4.41 • 3.09 Xm 0.53 0.49 0.39 1.5 2.6 0.003
R2 is the coefficient of determination; Adj-R2 is R2 adjusted for the number of terms in the model; pred-R2 is the prediction (PRESS) R2, equivalent to q2 from a leave-1-out cross-validation; RMS is residual error; L.O.F. is the Friedman lack-of-fit score; S.O.R. p-value is the p-value for significance of regression. Note: a multiplicity correction has not been applied to the p-values, meaning that the values are optimistic.