View Article Online / Journal Homepage / Table of Contents for this issue

PAPER www.rsc.org/analyst | Analyst

Non-linear multivariate curve resolution analysis of voltammetric pH

titrations†
Jose Manuel Dıaz Cruz,* Josep Sanchıs, Elena Chekmeneva, Cristina Ari~no and Miquel Esteban
Received 12th February 2010, Accepted 23rd April 2010
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM.

First published as an Advance Article on the web 28th May 2010

DOI: 10.1039/c0an00088d

A new chemometric approach is put forward, dealing with the non-linear behaviour observed in the
multivariate curve resolution (MCR) analysis of certain overlapping voltammetric signals obtained in
titrations of metal complexes where pH is progressively changed. In such cases, non-reversible
reduction signals move along the potential axis as a consequence of the involvement of H+-ions in the
electrochemical process and cause a dramatic loss of linearity, which hinders accurate MCR analysis.
The method proposed is based on the least-squares fitting of peak potential vs. pH datasets to
parametric linear and sigmoid functions through the decomposition of the data matrix into both
a concentration profile matrix and a unit signal matrix, in a similar way as in the alternating least-
squares algorithm of MCR (ALS). Such calculations are carried out through several home-made
Matlab programs which are freely available as Supplementary Material of the present work†. The fitted
parameters, along with the evolution of resolved concentrations and potential shifts with pH, provide
valuable information on the complexation/reduction processes. The method is tested first on the
relatively simple Cd(II)–NTA system and then applied to the study of the binding of Cd(II)-ions by
glutathione (g-Glu-Cys-Gly, GSH) and the phytochelatin PC2 ((g-Glu-Cys)2-Gly).

1. Introduction designed for spectroscopic data,5 once it was adapted to

Voltammetric techniques provide valuable information on the
binding of heavy metals by biomolecules. However, the presence
of many overlapping signals makes the interpretation of this
information difficult and requires the use of chemometric
methods.1–3 Unfortunately, most of these methods are based on
the assumption of the linear contribution of all species to the
overall signal, which is not always true in electrochemical
measurements.
Among the possible sources of non-linearity, the displacement
of the signals along the potential axis deserves special attention,
since this is a common situation in the presence of fast inter-
conversion equilibria or changes in electrochemical reversibility.
For this kind of non-linear system, intrinsically non-linear che-
mometric methods like artificial neural networks (ANN) can be
used or, instead, linear methods like partial least squares (PLS)
can be adapted with a high number of latent variables to model
non-linearity. However, since these approaches require an
extensive and time-consuming calibration, in a recent paper we
proposed a simpler strategy for the analysis of non-linear elec-
trochemical data through the shiftfit program.4 This is based on
a least-squares correction of the potential shift of moving signals
and the further analysis of corrected data by means of linear
methods such as multivariate curve resolution by alternating
least squares (MCR-ALS). Although MCR-ALS was originally
Departament de Quımica Analıtica, Universitat de Barcelona, Martı i
Franques, 1-11, E–08028 Barcelona, Spain. E-mail: josemanuel.diaz@ub.
edu; Fax: (+34) 93 402 12 33; Tel: (+34) 93 403 91 16
† Electronic supplementary information (ESI) available: Additional
figures and pHfit programs including tutorial and sample data. See
DOI: 10.1039/c0an00088d
voltammetry,6 it produced a series of excellent results in the
analysis of many metal–biomolecule systems.3,7
In a first approach, the correction method was applied to the
study of electrochemically labile complexes, where the uncon-
strained fitting of potential shifts produced values that were
further adjusted by means of the DeFord–Hume method to
calculate the corresponding stability constants.4 Another inter-
esting application was the correction of the non-linearity intro-
duced by the movement of anodic signals in the study of metal
binding by thiol-rich peptides.8 In both cases, the unconstrained
character of the fitting was not a problem, as long as signals were
moving inside well-separated potential ranges. In contrast,
further experiments in the presence of many overlapping and
moving peaks highlighted important weaknesses in the correction
methodology, which easily ‘confuses’ signals as they get closer.
A typical example of this is the voltammetric pH titration of
a heavy metal ion / thiol-peptide system, i.e. the addition of
a strong base (or acid) solution to a mixture of both metal and
ligand and the registration of voltammograms at different pH
values. This usually results in a great deal of overlapping and
moving signals with markedly non-linear behaviour (an asymp-
totic decrease of singular values is seen when principal compo-
nent analysis, PCA, is applied), which render the correction
procedure useless. Voltammetric pH titrations contain informa-
tion about the stoichiometry of the complexes, especially when
H+-ions are involved in their electrochemical reduction. In this
case, pH variation produces important shifts (DE) in the peak
potential of the signals.9,10 From the slope of the DE vs. pH plot,
a semi-empirical equation9,11 is used to estimate the ratio between
the electrons and H+-ions involved in the reduction.9,12–14 In the
presence of overlapping signals, deconvolution is required for

This journal is ª The Royal Society of Chemistry 2010 Analyst, 2010, 135, 1653–1662 | 1653

accurate determination of peak potentials. As the high non-
linearity of the data prevents the use of linear methods for the
overall analysis of the full dataset, individual deconvolution has
to be carried out for each voltammogram,15–18 which implies
a risk of inconsistent results if the deconvolution criteria are not
uniform throughout the full dataset.
The study reported here sought to upgrade this methodology
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM.
by attaining a simultaneous and consistent analysis of the full set
of voltammograms and also by dealing with non-linear DE vs.
pH relationships. This was done through a new program, pHfit,
which essentially uses the same strategy as shiftfit,4 but forces the
fitted DE values to follow a fixed relationship with pH, according
to linear and sigmoid functions whose coefficients are optimized.
In addition, the physical meaning of such coefficients is discussed
in order to get valuable information about the complexation/
reduction processes under study.
2. Theory
Let us consider a metal ion Mn+ that is reversibly reduced at
a mercury electrode to M(Hg) and also reacts with a ligand HmL
to form the electrochemically inert and electroactive complex
ML(nm)+ in solution. Electrochemically inert means that the
association-dissociation kinetics of the complex are so slow that
it practically does not dissociate during the time scale of the
electrochemical measurement. If the protonation equilibria of
the ligand are involved in the electrochemical reaction, the
reduction of Mn+-ions can take place according to different
simultaneous processes:9,19,20
i) Mn+ + ne % M(Hg)
ii) MLHmn+ + ne % M(Hg) + HmL
iii) ML(nm)+ + mH+ + ne % M(Hg) + HmL
iv) ML(nm)+ + ne % M(Hg) + Lm
Usually, reaction i) takes place at potentials quite less negative
than those of processes ii)–iv), thus producing a separate signal
for the reduction of the free metal ion, which, for the totally inert
case, remains at constant potential. In contrast, processes ii)–iv)
are simultaneous and compete with each other to produce
a signal which moves along the potential axis, depending on their
relative predominance. Despite the inherent complexity of the
problem, many authors assume a semiempirical relationship
between the peak potential of the signal (Ep) and pH:
m
Ep ¼ k  0:058 pH (1)
n peakmaker program can be useful.4 It allows one to fit Gaussian
with k being a constant related to the formal reduction potential
of the metal ion. This means that the slope of an Ep vs. pH linear
plot leads to the obtaining of the ratio between the protons and
electrons exchanged in the electrochemical reduction of the
complex, which is useful in the formulation of complexation
models.
Unfortunately, experimental systems do not behave in such
a simple way. Indeed, the situation is quite dependent on the
distribution of the differently protonated species of the free
1654 | Analyst, 2010, 135, 1653–1662
View Article Online
ligand. For instance, process iii) becomes negligible at too high
pH values (the concentration of H+ is too small to protonate
either the ligand or the complex) or too acidic pH values (the
ligand is already protonated in the complexed form or
complexation is negligible), whereas processes ii) and iv) are
unaffected by pH variation. These considerations are confirmed
by the sigmoid evolution of many experimental DE vs. pH plots.
To model this behaviour, we chose a parametric sigmoid equa-
tion in the form:
a
DE ¼ þ d (2)
1 þ eb ðpH  cÞ
with a, b, c, d being adjustable parameters. In the vicinity of the
inflexion point, the sigmoid can be approached by a straight line,
whose slope is given by:
ab
max: slope ¼  (3)
4
Then, shiftfit program4 was modified to include five different
types of peaks simultaneously, depending on their evolution with
pH (pHfit program):
- Type 0: The peak does not move; DE is always zero.
- Type 1: The peak moves in a non-prefixed way, so that DE
values are adjusted individually for each pH value and are not
forced to follow any parametric equation.
- Type 2: The peak moves according to a linear DE vs. pH
relationship, so that a straight line is fitted inside a particular
range of pH values.
- Type 3: The peak moves according to the sigmoid evolution
predicted by eqn (2): a, b, c and d parameters are fitted inside
a range of pH values.
- Type 4: The peak undergoes a sigmoid evolution, but the
reference signal is placed at either the highest or the lowest DE
value (d equals zero).
In the case of signal types 2, 3 and 4, the slope obtained in the
fitting can be used to apply eqn (1) and calculate electron-to-
proton ratios, whereas for types 3 and 4 an additional parameter,
inflexion point c, can be used to relate it with the pKa values of
the functional groups involved in metal binding.
Fig. 1 summarises how the pHfit algorithm works. In the
proposed example, the three signals, A, B and C, are of types 3
(sigmoid), 0 (immobile) and 2 (linear), respectively. The experi-
mental matrix Iexp contains all measured currents and has as
many rows as voltammograms (i.e. considered pH values) and as
many columns as scanned potentials. From the visual inspection
of Iexp, pure signals of reference are established and arranged into
V0 matrix. This can be arbitrarily chosen and is usually done at
pH values where the signal is well defined and does not suffer
strong overlapping with other signals. For this purpose, the
peaks with a mouse at both sides of every peak maximum by
visual inspection of the superimposed experimental data matrix.
Besides the pure voltammograms of the reference, the behav-
iour of each signal is required, and this is done in the first row of
the isshift matrix (numbers from 0 to 4). The second and third
rows define the existence range of every component (row
numbers where the signal appears and disappears) and the range
to be used for fitting parametric equations. These ranges are
defined initially by inspection of the experimental data matrix
This journal is ª The Royal Society of Chemistry 2010
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM. View Article Online

Fig. 1 Flowchart of the proposed methodology for non-linear multivariate curve resolution analysis of a voltammetric pH titration dataset by means of
pHfit and pHcalc programs. The proposed example includes three kinds of signals depending on their evolution with pH: sigmoid (A), immobile (B) and
linear (C).

This journal is ª The Royal Society of Chemistry 2010 Analyst, 2010, 135, 1653–1662 | 1655

and can be changed prior to each new run of pHfit. As for the
potential ranges where the maximum of every peak is expected,
they are selected inside the pHfit program by clicking with
a mouse on a plot of the data matrix. A proper selection of this
range is especially important to avoid ambiguities with respect to
neighbouring signals.
The next step is the initial estimation of potential shifts (DE)
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM.
for all moving signals in all voltammograms. This is done by
searching each voltammogram for relative maxima inside the
regions determined by the pH and potential existence ranges.
Then, DE is computed as the difference between the potential of
each maximum and the peak potential of the corresponding pure
signal in V0. When potential shifts are expected to follow a linear
or sigmoid relationship with pH, such initial estimations of DE
are used to compute initial values of the parameters a, b, c and
d (sigmoid) or a0 and b0 (straight line) to start an iterative least-
squares optimization based on the lsqcurvefit tool of the Matlab
program.4 This is done by the two home-made programs, pHfit
and pHcalc. The former computes at each pH value the pure
voltammogram of every component shifted from its reference
position by the DE value predicted by the sigmoid or linear
function with the initial set of parameters. Then, the voltam-
mogram measured at this pHi (Iexp,i) is decomposed as a product
of a vector Ci, containing the concentrations of all components,
with the matrix VshiftT containing all shifted pure voltammo-
grams (‘T’ being transposed matrix) plus an error. The product
Ci VshiftT is the reproduced voltammogram at this pHi (Irep,i).
When all reproduced voltammograms have been evaluated, they
are column-wise integrated into the reproduced matrix Irep. In
a similar way, the concentration values computed at every pH are
integrated into the C matrix. The reproduced and experimental
matrices are then compared by means of the percentage of lack of
fit (lof), defined as usual.5–7
Then, the pHfit program changes the initial values of the
sigmoid and/or linear parameters, uses pHcalc to compute the
experimental and reproduced matrix and will keep the iterative
process running until a satisfactory lack of fit is achieved. From
the optimal values of a–d and/or a0 –b0 , useful parameters like the
m/n ratio or the pH of the inflexion point can be estimated.
Additionally, a matrix corrected for non-linearity, Icor, can be
obtained as:
T matrices by means of home-made programs implemented in
Icor ¼ C V0 (4)
which can be further analysed with linear chemometric tech-
niques like MCR-ALS or PLS.
It must be pointed out that immobile components are
described by a pure signal, which is not changed throughout the
entire optimization. As for type-1 components (with uncon-
strained potential shifts), they require two iterative steps: first,
the parameters of sigmoid- and linear-behaving components are
optimised by use of fixed values for the unconstrained potential
shifts (those obtained in the initial estimation); and, second, these
unconstrained DE values are optimised by use of the optimal
potential shifts achieved in the first step for the other
components.
The standard deviations associated with the main optimised
parameters were computed from the Jacobian matrix, following
the usual methodology in non-linear least squares fitting.21,22
1656 | Analyst, 2010, 135, 1653–1662
View Article Online
3. Experimental section
3.1. Chemicals and instruments
All reagents were Merck, Sigma-Aldrich or Panreac analytical
grade. Stock solutions of Cd2+ 102 mol L1 were prepared from
Cd(NO3)2$4H2O and standardized complexometrically. 1 mol L1
and 0.1 mol L1 HNO3 and KOH solutions were prepared from
Titrisol (Merck). Phytochelatin (PC2, (g-Glu-Cys)2-Gly) as tri-
fluoroacetate salt was provided by DiverDrugs S.L. (Barcelona,
Spain), with purity of 91.8% (HPLC-MS). Ultrapure water
(Milli-Q Plus 185, Millipore) was employed in all experiments,
which were carried out at room temperature (20  1  C). All
solutions were deoxygenated with purified nitrogen (Linde N50)
and subsequently stored tightly closed.
Measurements by differential pulse polarography (DPP) were
carried out with 757 VA Computrace (Metrohm) attached to
a personal computer with data acquisition software (Metrohm).
The working, reference and auxiliary electrodes were a static
mercury drop electrode (SMDE), Ag/AgCl/KCl (3 mol L1) and
a glassy carbon electrode, respectively. Unless otherwise indi-
cated, instrument parameters were as follows: voltage step of
0.006 V, pulse amplitude of 0.05 V, pulse time of 0.04 s and drop
time of 1 s. Glass cells (Metrohm) were used in all the
measurements. pH values during the experiments were measured
by means of an Orion SA 720 pH meter.
3.2. Voltammetric titrations
Measurements were performed at different pH values for fixed
metal-to-ligand ratios. At the beginning of every experimental
run, 25 mL of a solution containing Cd2+and the ligand in the
right proportion and also 0.1 mol L1 of KNO3 were poured into
the cell and deoxygenated with purified nitrogen for 20 min, and
a DPP scan was recorded. Then, HNO3 and KOH solutions were
added successively to change steadily the pH of the cell solution
inside the range 2 to 11. After every addition, purified nitrogen
was passed through the solution for 1 min, pH was measured by
means of the glass electrode and a new DPP curve was recorded.
3.3. Data treatment
DPP curves were baseline-corrected and converted into data
Matlab.23 Some details about the peakmaker program and the
algorithm of unconstrained potential shift correction can be
found in ref. 4–8. As for the constrained correction, the processes
depicted in Fig. 1 are carried out by means of a set of home-made
programs implemented in Matlab23 and available as supple-
mentary material† of the present paper.
4. Results and discussion
4.1. Preliminary tests with the Cd(II)–NTA system
Prior to considering overlapping moving signals, the relatively
simple Cd(II)–NTA system was chosen as a preliminary test for
pHfit. Fig. 2a shows the structure of the NTA molecule.
Previous studies19,20,24,25 demonstrated that Cd(II) and NTA
form quite a stable 1 : 1 ML complex and that the 1 : 2 ML2
form is produced only in the presence of a large excess of
This journal is ª The Royal Society of Chemistry 2010
 View Article Online

increasing pH values, the signal of the free metal decreases (due

to increasing complexation) and moves to more negative
potentials (due to increasing complex dissociation). As for the
signal of the complex, it increases with pH (due to increasing
complexation) and moves to more negative potentials (due to
the pH-dependence of electrochemical kinetics). Although both
signals progressively move with pH, they do not overlap with
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM.

each other, so that the evolution of the respective peak

Fig. 2 Structures of the molecules NTA (a), GSH (b) and PC2 (c).
ligand. The ML complex is electroactive and its association-
dissociation kinetics are intermediate between the (electro-
chemically) labile and inert cases with a strong dependence on
pH. Low pH values decrease its inert character by increasing
ML dissociation, whereas higher pH values get the system
closer to totally inert behaviour. As a result, two separate
signals are measured by DPP in mixtures of Cd(II) and NTA:
a first one, at more positive potentials, corresponding to the
reduction of free Cd(II)-ions coming from the solution and
from complex dissociation; and a second one, at more negative
potentials generated by the reduction of the complex.
Fig. 3a shows the evolution of both signals during the DPP
pH titration of a Cd(II)–NTA mixture at 2 : 1 ratio. At
potentials is easily monitored by classic, univariate procedures
of peak evaluation. The results of this conventional strategy are
depicted as points in Fig. 3d. Despite that the absence of signal
overlapping makes it unnecessary to use multivariate tech-
niques to determine the evolution of peak potentials, this gives
us a chance to test the reliability of the proposed pHfit
algorithm.
When pHfit was applied to the matrix in Fig. 3a, it produced
consistent results with a (quite low) lack of fit of 3.57%. Fig. 3d
compares the sigmoids adjusted by pHfit with the regression
lines obtained from the conventionally measured points.
Finally, Fig. 3c shows the pH evolution of the corresponding
concentration profiles as computed by pHfit. The comparison
of Fig. 3c and 3d shows a clear sigmoid behaviour of the
potential of the free metal signal, with maximum variation
coinciding with the region of faster decrease in free metal
concentration. The signal of the complex also appears to follow
a sigmoid relationship, but in this case the upper part of the

Fig. 3 Experimental data matrix (a) measured for the Cd(II)–NTA system in a DPP pH titration at a metal-to-ligand ratio 2 : 1(total Cd concentration
2  104 mol L1). The application of pHfit program produced the potential shifts (d) and the concentration profiles (c) shown in the graph. In (b)
a comparison is made between the normalised singular values of the experimental (C) and the corrected matrix (B). In (d), points denote the potential
shifts of the free metal signal (O) and the signal of the complex (B) as compared to reference peaks and measured in the traditional, non-multivariate
way; solid lines show the fitted sigmoids according to pHfit program and dashed lines denote regression lines obtained from a fraction of the
conventionally-measured points.

This journal is ª The Royal Society of Chemistry 2010 Analyst, 2010, 135, 1653–1662 | 1657

Table 1 Parameters fitted to the potential shifts obtained in DPP pH titrations of the Cd(II)–NTA system according to the linear and the sigmoid (pHfit)
approaches. For all three metal-to-ligand ratios, the KNO3 concentration was 0.1 mol L1, whereas the total concentrations of Cd(II)/NTA were 0.2/0.1,
0.1/0.1 and 0.1/0.2 mmol L1, respectively. Standard deviations from respective fittings are given in parentheses. Linear fittings were carried out on sets of
7–8 points and produced r values between 0.993 and 0.998
Slope M signal
M : L ratio Lof (%) sigmoid fit Linear fit Sigmoid fit
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM.
2:1 3.57 0.014 (0.001) 0.016 (0.001)
1:1 3.23 0.013 (0.001) 0.016 (0.001)
1:2 4.12 0.015 (0.001) 0.018 (0.001)
sigmoid is poorly defined because it corresponds to the region
where the concentration of the complex is negligible. Fig. 3d
also shows that the sigmoid is a better description of DE
evolution than the linear approach, since it works accurately
throughout the entire experimental range and coincides very
well with the line near the inflexion point. Fig. 3b compares the
normalised singular values obtained in the decomposition of
both the experimental and the corrected data matrices. In the
first case, the system is far from linear, since at least three
components are required to describe the evolution of two
processes (i.e., the reductions of the free metal and the
complex). If two components are selected, only 88% of data
variation is explained. In contrast, the corrected data matrix
appears to be perfectly linear, since two components explain
more than 99% of data variation.
Table 1 summarises the numeric results obtained in the
application of the pHfit program to the DPP data and compares
them with the results of the linear fit applied to conventionally
measured potentials. The inspection of the table suggests quite
a low lack of fit in the application of the pHfit program and good
agreement between the slopes computed by linear and sigmoid
methodologies.
In conclusion, this preliminary study demonstrates that pHfit
can satisfactorily extract the potential shifts of non-overlapping
peaks from the experimental matrix and convert it into a cor-
rected linear matrix. The next step will be to verify whether the
program still works properly in the case of several overlapping
and moving peaks.
4.2. Application to the Cd(II)–GSH system
Although the binding of Cd(II) by GSH has been extensively
studied, the complexity of the system still generates discrep-
ancies about some of the postulated species.26 Voltammetric
studies make clear that in most cases an inert ML2 complex is
formed, which is reduced at potentials more negative than
those required for the reduction of the free metal ion.15,27
Under certain conditions, a second complex with stoichiometry
1 : 1 can be observed with characteristics intermediate between
labile and inert behaviour. This produces a signal strongly
overlapping the free metal signal.15,27 At fixed pH, MCR-ALS
allowed an accurate resolution of the overlapping signals.
However, in the experiments in which pH was progressively
changed, most signals were moving towards negative potentials,
which caused a dramatic decrease in linearity (shown by a large
number of components in PCA) that hindered the application
of MCR-ALS. Then, the evolution of peak potentials as
1658 | Analyst, 2010, 135, 1653–1662
View Article Online
Slope ML signal Inflexion point pH
Linear fit Sigmoid fit M signal ML signal
0.059 (0.002) 0.055 (0.001) 4.28 (0.01) 4.66 (0.06)
0.061 (0.002) 0.057 (0.001) 4.26 (0.01) 4.61 (0.05)
0.064 (0.002) 0.064 (0.001) 4.46 (0.01) 4.63 (0.02)
a function of pH had to be monitored by deconvolution of
individual voltammograms, a relatively imprecise and subjec-
tive method.15
In the present work, the Cd(II)–GSH system was studied
under conditions similar to the reference ones.15 In the exper-
imental matrix obtained at 1 : 1 ratio (Fig. 4a), three peaks are
clearly distinguished: i) the anodic signal of the free ligand L,
initially at ca. 0.35 V, ii) a peak at ca. 0.65 V attributed to
the ‘‘labile metal’’ (metal free plus weakly bound to 1 : 1
complex), and iii) a peak around 0.8 V corresponding to the
ML2 complex. It should be noted that, in contrast with the
experiments at fixed pH,15 the smooth evolution of signal ii)
does not allow one to differentiate between free and weakly
bound metal.
The application of pHfit allows a satisfactory resolution of the
system with a lack of fit of 5.43%, much lower than the value
13.04% of the reproduced PCA matrix using three components.
Fig. 4b and 4c summarise the evolution as a function of pH of the
optimised potential shifts and concentrations for all three
components. Additionally, Fig. 4d shows the distribution of the
species of the free ligand as computed from the acidity constants
of GSH found in the literature28 (pK1 ¼ 2.09, pK2 ¼ 3.48, pK3 ¼
8.67, pK4 ¼ 9.54).
Given the structure of the molecule (Fig. 2b), the first two
dissociations of H4L correspond to carboxylic groups and the
third and fourth dissociations are a mixed contribution of
ammonium and thiol groups. Fig. 4b and 4c show that, at pH
values lower than 5, there is no significant complexation. The
only remarkable fact in this pH region is the large and almost
linear potential shift of the anodic signal related to the presence
of free ligand.
At pH values higher than 5, both a decrease in the labile
fraction of the metal and an increase in the fraction of the inert
ML2 complex (Fig. 4c) suggest a progressive binding of Cd(II) to
the thiol group of GSH, probably with some contribution of the
amino group. Although the formation of the ML2 complex
clearly starts at a pH near 5, Fig. 4b shows that the corresponding
peak potential remains practically unchanged until pH values are
close to 6 and then starts an important shift towards negative
potentials until pH values are close to 9 by following a sigmoid
pattern. The signal of labile metal also suffers a sigmoid potential
shift in this range, but with much lower amplitude.
Fig. 4d helps explain these facts. At pH 6, carboxylic groups
of GSH are deprotonated, whereas thiol and amino groups are
protonated. At pH higher than 6, a progressive deprotonation
of thiol/amino groups starts, which finishes at a pH slightly
higher than 9. This means that inside this pH range (6 to 9),
This journal is ª The Royal Society of Chemistry 2010
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM. View Article Online

Fig. 4 Experimental data matrix (a) measured for the Cd(II)–GSH system in a DPP pH titration at a metal-to-ligand ratio 1 : 1(total Cd concentration
1  105 mol L1). The application of pHfit program yielded the potential shifts (b) and the concentration profiles (c) shown in the graph. In (d), the
distribution of free GSH species as a function of pH is shown (acidity constants taken from Ref. 28).

protonation of the ligand can play a role in the reduction of the 4.3. Application to the Cd(II)–PC2 system
complex. Ideally, if ML2 complex was formed exclusively by
After some experience was gained in the application of pHfit to
two Cd(II)–thiol bonds and all thiol groups in the free ligand
the systems above, an exhaustive pH study was carried out on the
were protonated, the proton-to-electron ratio would be 1 and
especially intricate Cd(II)–PC2 system. Unlike GSH, PC2 has two
the maximum slope of the sigmoid would be 0.058. However,
thiol groups (Fig. 2c), which greatly increases the possibilities of
Fig. 4d suggests that a fraction of the thiol groups of the free
metal coordination and produces a large number of overlapping
ligand is deprotonated as pH moves from 6 to 9, which should
signals in voltammetric measurements. The application of MCR-
cause a progressive decrease in the proton-to-electron ratio.
ALS was especially successful in the resolution of DP voltam-
Such a decrease is difficult to quantify, since the acidity
mograms recorded during titrations of Cd(II) with PC2, and vice
constants K3 and K4 are average parameters that do not
versa, at constant pH.29 However, the strong non-linearity of the
correspond exactly to the microscopic dissociation of thiol and
data obtained in voltammetric pH titrations prevented the
amino groups, respectively. Moreover, a hypothetical contri-
application of MCR-ALS to get complementary information. To
bution of amino groups to the binding could also modify the
obtain such information, in the present study DP voltammetric
ratio. The experimental value for the slope was 0.037 (Table 2),
pH titrations were performed at several Cd-to-PC2 ratios and
which means a proton-to-electron ratio of 0.64, somewhat lower
pHfit was used to analyze the resulting data matrices, which are
than 1, as expected. Table 2 shows the results of additional
shown in Figure I of Supplementary Material.†
experiments carried out at metal-to-ligand ratios different from
A look at the data matrices suggested that, despite the large
1. Although the slope of the ML2 signal is in all cases lower than
number of signals present, most of them are common to many
0.058, it considerably increases at increasing metal proportion,
experiments. Indeed, 6 kinds of signals could be identified, which
which could be related to a different contribution of amino and
were numbered according to their order of appearance along the
carboxylic groups to the binding.
potential axis:

Table 2 Parameters fitted to the potential shifts obtained in DPP pH titrations of the Cd(II)–GSH system according to pHfit approach. For all three
metal-to-ligand ratios, the KNO3 concentration was 0.1 mol L1, whereas the total concentrations of Cd(II)/GSH were 10/5, 10/10 and 10/20 mmol L1,
respectively. Standard deviations from respective fittings are given in parentheses

Mlabile ML2

M : L ratio Lof (%) Slope Inflexion point Slope Inflexion point

2:1 3.54 0.024 (0.001) 8.60 (0.02) 0.049 (0.001) 7.96 (0.04)
1:1 5.43 0.006 (0.001) 7.42 (0.06) 0.037 (0.001) 7.31 (0.02)
1:2 7.42 0.022 (0.001) 7.23 (0.04) 0.027 (0.001) 8.02 (0.02)

This journal is ª The Royal Society of Chemistry 2010 Analyst, 2010, 135, 1653–1662 | 1659
 View Article Online

1. At ca. 0.45 V: immobile anodic signal of PC2 (type 0).
2. At ca. 0.60 V: anodic signal of PC2, fitted as type 1 or 2
signal.
3. At ca. 0.68 V: immobile signal of the free Cd(II) (type 0).
4. At ca. 0.75 V: signal of weakly bound Cd(II), fitted as
type 2 or 3 signal.
5. At ca. 0.80 V: anodic signal of PC2, fitted as type 3
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM.
As an example, Fig. 5 summarises the results obtained in the
application of pHfit to the data obtained at 1 : 4 ratio. The
experimental matrix (a) shows strong overlapping of the 6 signals
mentioned above. The use of the peakmaker program allowed us
to define the pure signals of reference shown in Fig. 5e. The
application of pHfit generates, with a lack of fit of 16.7%, the
reproduced matrix shown in Fig. 5b and the corrected matrix of

signal. Fig. 5c. The comparison of the singular values of the experi-
6. At ca. 0.90 V: signal of strongly bound Cd(II), fitted as mental and corrected matrices (Fig. 5d) suggests an important
type 3 or 4 signal. increase in data linearity after the application of the pHfit

Fig. 5 Experimental data matrix (a) measured for the Cd(II)–PC2 system in a DPP pH titration at a metal-to-ligand ratio 1 : 4 (total Cd concentration
5  106 mol L1). The application of pHfit program by using the reference pure signals in (e) yielded the reproduced (b) and corrected (c) matrices, the
potential shifts (g) and the concentration profiles (f) shown in the graph. The singular values of the experimental (C) and corrected (B) matrices are
compared in (d). The distribution of free PC2 species as a function of pH is summarised in (h), as computed from the acidity constants taken from
Ref. 30. All components are denoted by numbers, whose meaning is explained in the text. In (g), black circles denote the initial values of DE computed
from the maxima of experimental voltammograms inside the potential estimation ranges.

1660 | Analyst, 2010, 135, 1653–1662 This journal is ª The Royal Society of Chemistry 2010
 View Article Online

Table 3 Parameters fitted to the potential shifts obtained in DPP pH titrations of the Cd(II)–PC2 system according to pHfit approach. For all metal-to-
ligand ratios, the KNO3 concentration was 0.1 mol L1, whereas the total concentrations of Cd(II)/PC2 were 0/20, 20/10, 10/10, 10/20 and 5/20 mmol L1,
respectively. Standard deviations from respective fittings are given in parentheses

M : L ratio

0:1 2:1 1:1 1:2 1:4

Lack of fit (%) 12.6 11.8 11.4 15.4 16.7

Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM.

Slope Signal 1 — — — — —
2 0.070 (0.001) — — 0.051 (0.003) 0.048 (0.001)
3 — — — — —
4 — 0.063 (0.001) 0.082 (0.001) 0.057 (0.001) 0.050 (0.001)
5 0.053 (0.001) — 0.034 (0.001) 0.046 (0.001) 0.070 (0.001)
6 — 0.039 (0.001) 0.036 (0.001) 0.028 (0.001) 0.041 (0.001)
Inflexion point 1 — — — — —
2 4.44 (0.02) — — — —
3 — — — — —
4 — — 5.13 (0.01) 4.93 (0.01) —
5 8.60 (0.01) — 9.54 (0.05) 8.65 (0.06) 8.33 (0.02)
6 — 8.13 (0.03) 8.02 (0.02) 8.16 (0.02) 7.97 (0.02)

program. Fig. 5f shows the concentration profiles obtained and

Fig. 5g gives the curves fitted to explain the evolution of peak
potentials as a function of pH. The optimised parameters
defining such curves for this and the other experiments are
collected in Table 3.
A good complement to understanding the trends in these
graphs is the distribution diagram of the species of the free ligand
as a function of pH (Fig. 5h), which has been computed by using
the acidity constants given in Ref.30 (pK1 ¼ 2.39, pK2 ¼ 3.18,
pK3 ¼ 4.01, pK4 ¼ 8.75, pK5 ¼ 9.03, pK6 ¼ 10.04). Such
constants are related to the presence of three carboxylates that
become totally deprotonated at pH 6 and two thiols and one
amino group that are progressively deprotonated as pH increases
from 6 to 10.
As Fig. 5f shows, the anodic signal 1 is quite important at
acidic pH and slowly decreases as pH increases and complexes
start to form, which suggests that it could be related to the free
PC2. The sharp disappearance of component 3 at pH close to 4.0
confirms the strong complexing power of PC2: even with all thiol
and amino groups protonated, Cd(II)-ions are weakly bound to
PC2. This is confirmed by the appearance of component 4
(attributed to weakly complexed cadmium), which predominates
until pH 6.0 and then seems to suddenly transform into
component 6, assigned to strongly bound Cd(II)-ions. This
suggests that, just when the first thiol groups start to slightly
deprotonate (pH close to 6), all weakly bound Cd(II)-ions start to
form stronger bonds with them, so that at pH not much higher
than 6 all these ions are strongly bound. As for components 2 and
5, they are anodic signals of PC2 already detected in the absence
of metal. However, their evolution with pH is strongly affected
by the presence of Cd(II), as the fitted parameters in Table 3
show. Additionally, the trends of these components in both
Fig. 5f and 5g are analogous to the evolution of the two kinds of
bound metal: signal 2 seems to be related to weakly bound
cadmium (signal 4), whereas signal 5 seems to be the counterpart
of strongly bound Cd(II)-ions (signal 6).
The comparison of Fig. 5g and 5h suggests an important
consideration: signal 4 of weakly complexed Cd(II) changes its
peak potential linearly in its entire existence range and the same
happens to the anodic signal 2 related to it. This happens in a pH
Fig. 6 Comparison of the concentration profiles as a function of pH
obtained by pHfit program from DPV measurements on the Cd(II)–PC2
system at different metal-to-ligand ratios: 1) 2 : 1 (B), 2) 1 : 1 (O), 3)
1 : 2 (,) and 4) 1 : 4 (>). Dashed lines denote weakly bound Cd(II)
whereas solid lines stand for strongly bound Cd(II). Every concentration
profile has been normalised by the maximum value of the concentration
of strongly bound Cd(II).
range in which all thiol groups are protonated, which is consistent
with the formation of Cd–S bonds. Additionally, Table 3 shows
that the slope of the DE vs. pH straight line is 0.050, quite close to
the value 0.058 that would produce a proton-to-electron ratio of 1.
This suggests the formation of 2 bonds with protonated groups,
very probably two thiols with a possible contribution of carbox-
ylates. In the case of signal 6 of strongly bound Cd(II), the
evolution of the peak potential with pH is clearly sigmoid and the
same happens to the related signal 5. This is not surprising if we
take into account that, in the pH range in which these components
predominate, there is progressive deprotonation of all thiol and
amino groups that are related to binding. The maximum slope of
the sigmoid (Table 3) has a value of 0.041, clearly lower than 0.058.
When the concentration profiles obtained at different metal-
to-ligand ratios are compared for both kinds of bound cadmium
(Fig. 6), some general patterns can be deduced. As the ligand
proportion increases (signals 1 to 4), the weak binding of Cd(II)
starts at lower pH values and begins to transform into stronger
binding at slightly lower pH values, too. This is not surprising if
we assume that increased excess of the ligand ensures more

This journal is ª The Royal Society of Chemistry 2010 Analyst, 2010, 135, 1653–1662 | 1661

deprotonated groups even at lower pH, which favours the
formation of weak and (especially) strong bonds. As for the DE
plot (See Supplementary Material†), quite similar behaviour is
found at the different ratios, which suggests that the nature of
both strong and weak cadmium binding is common to different
metal-to-ligand ratios (i.e., it could involve the same functional
groups) and that such a ratio would just determine the propor-
Published on 28 May 2010. Downloaded by Universidad Autonoma de Coahuila on 3/26/2019 8:11:23 PM.
tion between the two kinds of binding.
Finally, the comparison of slopes and inflexion points (Table
3) shows a certain inconsistency with the narrow standard
deviations of the fitting. This suggests that the real uncertainty of
these parameters is mostly given by the reproducibility of the
experiments and not by the accuracy of the fitting of a single
dataset. Indeed, some replicates of the dataset given (not shown)
suggested accuracy of ca. 0.01 for the slope and ca. 0.5 for the
inflexion point. If we take into account these limitations, Table 3
suggests that weak binding (signal 4) generates proton-to-elec-
tron ratios slightly higher than 1 and also suggests that, as well as
thiol groups, it could involve carboxylates (inflexion points close
to 5, a typical value for carboxylic pKa). Finally, Table 3 shows
slopes clearly lower than 0.058 in the case of strong binding.
Nevertheless, this does not mean proton-to-electron ratios lower
than 1, since, in the pH regions where such binding happens,
most of the groups involved are deprotonated.
5. Conclusions
The facts described above demonstrate the applicability of the
proposed methodology for an accurate analysis of voltammetric
pH titration data, including many overlapping and moving
signals. This is especially promising in voltammetric studies of
heavy metal binding by biomolecules, since several parameters
containing valuable information about the binding process can be
fitted. These include the maximum slope of the DE vs. pH curves
(which is related to the proton-to-electron ratios of electro-
chemical reactions) and the inflexion point (which provides qual-
itative information about the acidity of the functional groups
acting as ligands). Moreover, the optimised concentration profiles
provide additional information on the nature of the binding. In less
favourable situations, the method can be used simply to linearise
voltammetric datasets for further analysis with more conventional
techniques like MCR-ALS or PLS. In this case, however, the risk
of producing ‘chemometrical artifacts’ should be minimised by
a proper theoretical justification of the linearity correction.
For the systems considered, although most DE vs. pH slopes
can be determined with reasonable precision, these values could
not be used for accurate determination of the corresponding
proton-to-electron ratios. Unfortunately, in most cases, different
protonated forms of the ligand coexist in the pH regions where
complexation takes place. Then, fitting just produces an average
parameter among different proton-to-electron ratios. All this
means that one has to be careful when proposing complexation
mechanisms and stoichiometries from DE vs. pH slopes, espe-
cially if the protonation constants of the ligand are unknown.
Finally, in more general terms, the methodology proposed can
be seen as an alternative to the MCR-ALS strategy of ‘shape
1662 | Analyst, 2010, 135, 1653–1662
View Article Online
constraints’, which has provided successful results in the analysis
of many kinds of data, so that further adaptations of the
proposed method to non-electrochemical signals progressively
moving along the x-axis can be confidently expected.
Acknowledgements
The authors acknowledge financial support from the Spanish
Ministerio de Educaci
on y Ciencia (Project CTQ2009-09471) and
also thank Miquel Grabulosa for some preliminary experiments.
J. Sanchıs thanks the University of Barcelona for a grant of
collaboration with research groups.
References
1 S. D. Brown and R. S. Bear Jr., Crit. Rev. Anal. Chem., 1993, 24, 99.
2 E. Richards, C. Bessant and S. Saini, Electroanalysis, 2002, 14, 1533.
3 M. Esteban, C. Ari~ no and J. M. Dıaz-Cruz, TrAC, Trends Anal.
Chem., 2006, 25, 86.
4 A. Alberich, J. M. Dıaz-Cruz, C. Ari~ no and M. Esteban, Analyst,
2008, 133, 112.
5 R. Tauler, A. Smilde and B. R. Kowalski, J. Chemom., 1995, 9, 31.
6 J. M. Dıaz-Cruz, R. Tauler, B. S. Grabaric, M. Esteban and
E. Casassas, J. Electroanal. Chem., 1995, 393, 7.
7 M. Esteban, C. Ari~ no, J. M. Dıaz-Cruz, M. S. Dıaz-Cruz and
R. Tauler, TrAC, Trends Anal. Chem., 2000, 19, 49.
8 A. Alberich, J. M. Dıaz-Cruz, C. Ari~ no and M. Esteban, Analyst,
2008, 133, 470.
9 S. J. Slattery, J. K. Blaho, J. Lehnes and K. A. Goldsby, Coord. Chem.
Rev., 1998, 174, 391.
10 W. Stumm, J. J. Morgan, Aquatic Chemistry, Wiley, New York, 2nd
Ed., 1981, Chapter 7.
11 J. Heyrovsky, J. Kuta, Principles of Polarography, Academic Press,
New York, 1966, p. 161.
12 G. Hilt, T. Jarbawi, W. R. Heineman and E. Steckhan, Chem.–Eur. J.,
1997, 3, 79.
13 H. Alemu and B. S. Chandravanshi, Electroanalysis, 1998, 10, 116.
14 I. Cukrowski, J. Electroanal. Chem., 1999, 460, 197.
15 M. S. Dıaz-Cruz, J. Mendieta, R. Tauler and M. Esteban, J. Inorg.
Biochem., 1997, 66, 29.
16 J. Mendieta, M. S. Dıaz-Cruz, A. Monjonell, R. Tauler and
M. Esteban, Anal. Chim. Acta, 1999, 390, 15.
17 B. H. Cruz-Vasquez, J. M. Dıaz-Cruz, C. Ari~ no, M. Esteban and
R. Tauler, Analyst, 2002, 127, 401.
18 E. Chekmeneva, J. M. Dıaz-Cruz, C. Ari~ no and M. Esteban, Anal.
Biochem., 2006, 348, 252.
19 J. Koryta, Collect. Czech. Chem. Commun., 1959, 24, 2903.
20 J. Koryta, Collect. Czech. Chem. Commun., 1959, 24, 3057.
21 P. Gans, Data Fitting in the Chemical Sciences, Wiley, New York,
1992, Chapter 3, pp. 45–47.
22 J. M. Dıaz-Cruz, J. Agull o, M. S. Dıaz-Cruz, C. Ari~no, M. Esteban
and R. Tauler, Analyst, 2001, 126, 371.
23 The MathWorks (2006) MATLAB version 7.3.0.267. The Math-
Works, Natick.
24 G. Anderegg, Pure Appl. Chem., 1982, 54, 2693.
25 M. Torres, J. M. Dıaz-Cruz, C. Ari~ no, B. S. Grabaric, R. Tauler and
M. Esteban, Anal. Chim. Acta, 1998, 371, 23.
26 D. Dolphin, O. Avramovic, R. Poulson ed., Glutathione. Chemical,
Biochemical and Medical Aspects, Part A. John Wiley, New York,
1989.
27 L. Wagner-Roos, H. Zahn, J. M. Sequaris and P. Valenta, Toxicol.
Environ. Chem., 1989, 22, 77.
28 L. Pillai, R. D. Boss and M. S. Greenberg, J. Solution Chem., 1979, 8,
635.
 akova, J. Velek, C. Ari~
29 B. H. Cruz, J. M. Dıaz-Cruz, I. Sest no and
M. Esteban, J. Electroanal. Chem., 2002, 520, 111.
30 J. Johanning and H. Strasdeit, Angew. Chem., Int. Ed., 1998, 37, 2464.
This journal is ª The Royal Society of Chemistry 2010