Nonionic Surfactants – A Multivariate Study – by

Nonionic Surfactants – A Multivariate Study – by
Nonionic Surfactants
– A Multivariate Study –
Lise-Lott Uppgård
Akademisk avhandling
Som med tillstånd av rektorsämbetet vid Umeå universitet för erhållande av Filosofie
Doktorsexamen vid Teknisk-Naturvetenskapliga fakulteten, framlägges till offentlig
granskning vid Kemiska institutionen, Umeå universitet, sal KB3B1, KBC, lördagen den
16 november 2002 kl.10.30.
Copyright © 2002 by Lise-Lott Uppgård
2nd Edition
ISBN: 91-7305-330-9
Printed in Sweden by VMC-KBC
Umeå University, Umeå 2002
Nonionic Surfactants – A Multivariate Study
Lise-Lott Uppgård, Organic Chemistry, Department of
Chemistry, Umeå University, SE-901 87 Umeå, Sweden
In this thesis technical nonionic surfactants are studied
using multivariate techniques. The surfactants studied were
alkyl ethoxylates (AEOs) and alkyl polyglucosides (APGs).
The aquatic toxicity of the surfactants towards two
organisms, a shrimp and a rotifer, was examined. The
specified effect was lethality, LC50, as indicated by
immobilisation. In a comparative study, the LC50 values
obtained were used to develop two different types of
model. In the log P model the toxicity was correlated to log
P alone, while in the multivariate model several
physicochemical variables, including log P, were
correlated to the toxicity. The multivariate model gave
smaller prediction errors than the log P model.
Further, the change in reactivity when a surfactant mixture
was added to dissolving pulp under alkaline conditions was
studied, using the amount of residual cellulose as a
measure of the reactivity. Ten AEO/APG mixtures were
tested, and the mixture with greatest potential was studied
in more detail. An optimum in the amount of added
surfactant was found that seems to coincide, according to
surface tension measurements, with the CMC.
AEO, APG, aquatic toxicity, cellulose reactivity, Fock,
MLR, multivariate data analysis, nonionic surfactant, PCA,
physicochemical properties, PLS, QSAR, surface activity,
ISBN: 91-7305-330-9
LIST OF PAPERS .........................................................................................................7
LIST OF ABBREVIATIONS .......................................................................................8
INTRODUCTION .........................................................................................................9
NONIONIC SURFACTANTS ....................................................................................11
SURFACTANT SYNTHESIS ...........................................................................................12
AMPHIPHILIC SELF-ASSEMBLY ..................................................................................13
SURFACTANT CHARACTERIZATION ............................................................................14
MULTIVARIATE METHODS ..................................................................................17
MULTIPLE LINEAR REGRESSION.................................................................................18
PRINCIPAL COMPONENT ANALYSIS ............................................................................18
PARTIAL LEAST SQUARES REGRESSION .....................................................................19
MODELLING NONLINEARITIES ....................................................................................20
MODEL VALIDATION AND PREDICTIONS ....................................................................20
MODELLING TOXICITY .........................................................................................23
LOG P FOR PREDICTING THE TOXICITY ......................................................................25
SHOULD LOG P OR SURFACTANT PARAMETERS BE MODELLED? ...............................28
REACTIVITY STUDIES ............................................................................................37
CELLULOSE ................................................................................................................37
VISCOSE PROCESS .....................................................................................................38
REACTIVITY MEASURE ..............................................................................................39
REACTIVITY INCREASE ..............................................................................................40
OUR WORK ................................................................................................................40
REPLICATE STUDY .....................................................................................................41
ADDITION OF SURFACTANT ........................................................................................41
SURFACTANT ACTION ................................................................................................43
CONCLUDING REMARKS ......................................................................................45
REFERENCES ............................................................................................................49
List of Papers
based on the four papers listed below. In the text, the
papers will be referred to by their Roman numbers.
Multivariate Quantitative Structure-Activity Relationships for
the Aquatic Toxicity of Technical Nonionic Surfactants
Lise-Lott Uppgård, Åsa Lindgren, Michael Sjöström and
Svante Wold
Journal of Surfactants and Detergents 3(1): 33-41 2000.
Multivariate Quantitative Structure-Activity Relationships for
the Aquatic Toxicity of Alkyl Polyglucosides
Lise-Lott Uppgård, Michael Sjöström and Svante Wold
Tenside/Surfactants/Detergents 37(2): 131-138 2000.
Improving Cellulose Reactivity by Addition of Nonionic
Lise-Lott Uppgård, Åsa Lindgren and Michael Sjöström
In manuscript
A Multivariate Physicochemical Characterisation of Technical
Lise-Lott Uppgård, Anders Berglund, Ingegärd Johansson,
Michael Sjöström, and Christine Strandberg
In manuscript
List of Abbreviations
Alkyl Ethoxylate
Alkyl Polyglucoside
Critical Micelle Concentration
Ecological Structure Activity Relationships
(U.S.) Environmental Protection Agency
Multiple Linear Regression
Nonlinear Iterative PArtial Least Squares
Ordinary Least Squares
Principal Component
Principal Component Analysis
Partial Least Squares projections to latent structures
Quantitative Structure Activity Relationships
Root Mean Squared Error of Prediction
Structure Activity Relationships
one frequently comes into contact with surface
active agents, or so-called surfactants. In food products, naturally
occurring surfactants act as emulsifiers, for example casein in milk
and other dairy products. In our households, synthetic surfactants are
active ingredients in detergents, cosmetics, and pharmaceuticals. In
industry, surfactants are used in pulp and paper manufacturing, in oil
recovery, in flotation where different minerals are separated from each
other and a host of other applications.
Over the years, both the type and uses of surfactants have changed.
The oldest record of surfactant production has been found in clay
cylinders originating from ancient Babylon, dating back to about 2800
B.C. A soap-like material was found in the cylinders and the inscriptions
say that fats were boiled with ashes, a known soap-making method, but
the purpose of the soap was not revealed.
In the 19th century progress in the chemical industry made large-scale
commercial production of surfactants possible, although it was still based
on fats and ashes. Soon, the surfactants were used for cleaning clothes,
and their increased use led to the development of more efficient surface
active agents [1].
In the 1950’s, the synthetic textiles became increasingly common and
types of surfactants other than simple soaps were needed. The first
surfactants were anionic with a branched hydrocarbon chain, which made
them difficult for microorganisms to biodegrade. Subsequently, foam
accumulated on many lakes and rivers since the surfactants were not
degraded. Therefore, straight-chained surfactants that are more easily
degraded were introduced and the foaming problem was reduced [2].
Further problems were caused by the presence of polyphosphate ions
in detergent blends, which help form ion complexes in order to soften the
Nonionic Surfactants
molecules consisting of two
different parts: one hydrophilic (“water loving”) and the other
hydrophobic, (“water rejecting”). This combination makes the
surfactant ambivalent; the hydrophilic head group is attracted to polar
environments, for example water, while the hydrophobic tail is attracted
to nonpolar environments, for example oil. Consequently, the surfactants
can dissolve either in water or oil and have the capacity to solubilise water
in oil and vice versa, thus creating homogeneous systems. However,
surfactants tend to self-assemble or aggregate when dissolved in either
polar or nonpolar environments. The so-called hydrophobic effect is the
driving force for this aggregation in water, caused by the hydrophobic tail
avoiding contact with the solvent i.e. the water. A single surfactant
molecule is called a monomer. When the monomers self-assemble into
aggregates there is a gain in entropy from the release of structured water
molecules surrounding the monomer. This is referred to as hydrophobic
One force opposing self-assembly in such circumstances is the head
group repulsion. This repulsion originates from the hydrophobic force
minimizing the unfavourable water/hydrocarbon contact area, and thus
bringing the head groups into close proximity with each other at the
aggregate/water interface. The strength of the head group repulsion
depends on the nature of the surfactant. With nonionic surfactants the
head group repulsion is mainly due to steric hindrance and with ionic
surfactants the repulsion is of an electrostatic character. The electrostatic
repulsion between head groups can be reduced by the addition of an
electrolyte, i.e. a salt, which can screen the electrostatic repulsion [4].
The hydrophobic part of the surfactant consists of a hydrocarbon
chain. This hydrocarbon chain can be obtained from petrochemical
sources or from naturally occurring fats and oils, such as coconut oil or
palm kernel oil, as mentioned earlier. Classification of surfactants is
usually based on the hydrophilic part. There are two main groups of
surfactants: ionic and nonionic. Ionic surfactants can be further divided
into three subgroups: anionic, cationic and zwitterionic. Anionic
surfactants are negatively charged and their hydrophilic part very often
consists of a sulphate or phosphate group. In contrast, the cationic
surfactants are positively charged, frequently with a nitrogen-based
hydrophilic part. The zwitterionic surfactants are neutral overall but
contain functional groups with both a negative and a positive charge.
As mentioned in the Introduction, the predominantly used nonionic
surfactants were originally AEOs, but since about fifteen years ago new
kinds of nonionic surfactants, APGs, have been receiving increasing
attention. These surfactants are interesting since they are manufactured
from renewable raw materials and are considered to be environmentally
friendly [5].
Surfactant Synthesis
The synthesis of AEOs and APGs involves polymerization processes.
In the synthesis of AEOs fatty alcohol is reacted with ethylene oxide using
a base catalyst and water elimination. APGs are synthesized by direct
glucosidation between an excess of alcohol and glucose using an acid
catalyst and water elimination. In both cases, the final surfactant product
is of technical grade i.e. a complex mixture of surfactants with quite a
broad range of molecular weights. To further complicate matters the
hydrophobic part of the surfactant product can originate from a mixture of
fatty alcohols.
Figure 2. An APG, where DP = 1 and R = any alkyl chain. The anomeric property on
carbon 1 is illustrated with a wavy bond.
In the case of APG the technical product contains both a and b
anomers of glucose. The anomeric property exists because there are two
possible modes of attaching the hydrocarbon chain to the glucose ring, i.e.
by a or b linkages. When the hydroxyl group on the glucose ring is axially
oriented at carbon 1 (see Figure 2) an a linkage will occur, and a b
linkage is formed when the hydroxyl group is equatorially oriented.
The linkage between the glucose units is usually referred to as a
glucoside linkage or glucoside bond. In nature, the formation and
breakdown of these glucoside bonds are enzymatically controlled by
glucosidases [6]. This property makes these surfactants potentially more
easily biodegradable i.e. more environmentally friendly than other
nonionic surfactants, for example AEOs.
Amphiphilic Self-Assembly
At low surfactant concentrations in water, surfactants occur as
monomers in the solution as well as at the interface. The distribution of
monomers between the solution and the interface is in equilibrium. As the
surfactant concentration is increased the monomers start to interact, selfassemble, and aggregation starts when the surfactant concentration
exceeds a certain value, the critical micelle concentration or CMC.
Surface tension measurements can be used to determine the CMC value
since the surfactant monomers disrupt the hydrogen bonding between
water molecules at the surface and thus lower the surface tension. At the
CMC the water/air interface is saturated with surfactant monomers and
the reduction of the surface tension strongly diminishes. The magnitude of
the CMC value is specific for each surfactant.
Amphiphilic molecules can associate into a variety of structures in
aqueous solutions. These structures can transform from one to another
when solution conditions are changed, for example, the electrolyte
concentration, temperature, pressure or pH. Possible structures are
restricted by the forces acting to keep amphiphiles’ polar and nonpolar
parts in favourable environments.
Israelachvili [7] has established an equation related to the geometry or
packing properties of the surfactant. The dimensionless packing parameter
or shape factor, Ns = v/la0, relates amphiphilic molecular structure to the
preferred curvature properties of the aggregates, where v and l represent
the volume and length of the hydrocarbon chains, respectively, and a0 is
the area per head group. Small values of Ns imply highly curved
aggregates, micelles, but when Ns is close to unity, planar bilayers
(membranes), usually form.
Ns < 1/3
Ns = 1/3 - 1/2
Ns = 1/2 - 1
Figure 3. Overview of different amphiphilic aggregate shapes in water.
Surfactant Characterization
Characterisation can be seen as a progression from a visual description
to a numerical representation of molecules, based on various molecular
properties, such as log P, molecular weight and branching patterns. All
these variables depend on the structure of the molecule. Some of them are
easy to calculate, for example molecular weight, which is atom dependent,
while others, such as log P, depend on atom types, connectivity and
Ideally, such characterization should be as broad as possible i.e. using
many different descriptors that are considered relevant and interpretable
for the specific problem. In general, the more complex an observed
system is, the more unlikely it is that a single descriptor variable will
provide sufficient relevant information. Hence, the structural and
physicochemical description is preferably multivariate, to a degree that
depends on the nature of the problem or phenomena being considered. If
there is no prior information or knowledge about the importance of the
descriptor variables, it is difficult to predict which variables will be
useful. However, it is usually recommended that variables describing the
hydrophobicity, size and electronic properties should be included [8].
Structural and physicochemical properties can be categorised in many
ways. One way of doing this is to divide them into substituent and
molecular classes, i.e. descriptors related to a certain part or moiety of a
molecule, and those related to the whole molecule, respectively. In
addition, some properties can be calculated while others are empirically
Commercial surfactants are mixtures, usually of isomers and
homologues, rather than pure compounds with a single general structure.
It is therefore important to measure surfactant properties. Numerous
publications concerning surfactants are based on pure single surfactant
compounds relating one descriptor variable, often log P, to a property of
interest, toxicity for instance. However, few studies on surfactants of
technical grade have been published. Å. Lindgren [9] multivariately
characterized the AEOs considered in Paper I, and I. Johansson [10]
initiated the characterisation of the APGs that were further studied in
Papers II and IV. The physicochemical characterisation of AEOs and
APGs presented in Papers I and IV, respectively, is mainly based on
measured properties.
The surfactants assessed in Papers II and IV were characterized
according to 19 variables. Here, the descriptor variables are roughly
divided into five subgroups; describing the hydrophilic part of the
surfactant, the hydrophobic part, the whole surfactant molecule, surface
activity parameters and the composition of the surfactant product, see
Table 1.
Hydrophilic Part
Hydrophobic Part
Surface Activity
Surfactant Product
Abbreviation Description
Degree of Polymerization
Number of carbon atoms
Number of carbon atoms in the
longest chain
Ratio of redC to C
Molecular Weight
Ratio of DP to C
Ratio of DP to redC
Hydrophilic-Lipophilic Balance
Critical Micelle Concentration
Contact Angle
Surface Tension
Wetting rate
Foam height
Remaining amount of nonAlOH
glucosidated alcohol
Monoglucoside content in the
technical surfactant product
Table 1. The descriptor variables used to characterize the APGs, Paper IV.
Using molecular modelling to generate descriptor variables has the
advantage that synthesis of the surfactant product is unnecessary.
However, one problem that arises when using calculations to obtain
surfactant variables is how to treat the surfactant mixture. Basically, there
are two ways to handle this problem. The calculations can be based on the
“average structure” or, if the composition of the surfactant mixture is
quantitatively known, the variable can be calculated for each component,
multiplied by respective mole fraction and summed. In this way a
weighted average variable is obtained. Another problem is that the
variables generated by molecular modelling may be difficult to interpret.
Some variables, for instance log P, describing the hydrophobicity, are
relatively easy to correlate to structural features of the molecule, but this
is more difficult for certain other variables, such as connectivity indices.
Alternatively, measuring surfactant properties of a technical product
focuses on the surfactant mixture and gives immediate answers about how
the mixture behaves. However, a drawback is that the properties are
unique for the surfactant mixture they were derived from.
Multivariate Methods
about the complex world we live in, models of
differing complexity have been developed in order to explain and
interpret observed phenomena. A model is an approximation of an
observed phenomenon and is often expressed as a function of variables
that have been measured or estimated. An empirical function like this,
also called a soft model, is a mathematical expression of varying
complexity. If covering a small interval, it may be derived from a
relatively straightforward mathematical procedure, such as a Taylor
expansion. The true function describing the whole system might be much
more complex than the empirical model. The limitation with empirical
models is that they are only valid within the restricted region studied i.e.
the models are local. The reason for using empirical models is that the
reality is often far too complex to be described by current physical
theories and laws, i.e. so-called hard models.
The interpretation of data is an important part of all research. Usually,
the data are collected by observations, tabulated, compared and
interpreted. Data comparisons can be performed in many ways, for
example, by comparing the columns in a data table or by plotting the data.
These approaches may be insufficient since most investigated systems are
complex, and studying one variable at a time may result in meaningless or
incorrect conclusions. However, multivariate analysis provides a powerful
tool for interpreting large amounts of data.
In multivariate analysis all the collected data are simultaneously
accounted for, resulting in a model describing the overall system. Hence,
it is possible to get an overview of the data and to identify important,
hidden information. A short presentation of the methods used in this
thesis is presented below. For a more comprehensive overview the reader
is referred to the cited references.
Multiple Linear Regression
Multiple Linear Regression (MLR), also known as Ordinary Least
Squares (OLS), is a regression method that can be used for establishing a
relationship between variables, X, and a response, y. MLR requires the
variables to be independent of each other, that there are fewer variables
than observations, and that the chosen variables are relevant to the
investigation. Further, the data in the descriptor matrix, X, is assumed to
be exact and free of noise. A general quadratic MLR model with two
independent variables can be expressed as follows:
yobs = Xb + e = b0 + b1x1 + b2x2 + b12x1x2 + b11x12 + b22x22 + e
where b0 is a constant, the regression coefficients b1 and b2 are the linear
terms of the main effects, x1 and x2, and b12 is the cross term of the
interaction between x1 and x2. The coefficients b11 and b22 i.e. the
quadratic terms of x1 and x2, will indicate whether there is a
maximum/minimum present. The last term, the residual, e, is the
difference between the experimental data and the estimated model and
contains errors of measurement as well as model errors. The least squares
solution for calculating the coefficients, b, is given by
b = (X´X)-1Xy
Due to the requirement for independent variables in MLR modelling,
it is essential to use experimental design to get reliable models [11].
Principal Component Analysis
Principal Component Analysis (PCA) is used on multivariate data to
reveal patterns and groupings among the objects. A data table with N
objects and K variables can be viewed as N points in a K dimensional
space. This K dimensional space can be reduced to an A dimensional
subspace, a hyperplane, based on principal components, PCs. These are
orthogonal to each other and linear combinations of all the original
descriptor variables, X. Mathematically, PCA corresponds to a
decomposition of the multivariate data matrix X into the product of two
smaller matrices, T and P´, where T is an NxA score matrix which gives
each object its positional co-ordinates in the subspace, i.e. on the
hyperplane, P is a KxA loading matrix and gives the direction of the
hyperplane, and A is the number of PCs. The angle between a variable and
a PC can be interpreted as the arccos of the loading for that particular
variable, provided that åp´p=1. The smaller the angle between a variable
and a PC, the larger loading and the more the variable contributes to the
corresponding score vector. The variance not explained by the A PCs is
found in the NxK residual matrix, E.
X = TP´+ E
One algorithm used for finding the PCs is NIPALS (Nonlinear
Iterative Partial Least Squares). Here, the components are extracted one at
a time, the first PC is the one that explains the most variation in X, the
second PC explains the second most variation, and so on. More detailed
descriptions of PCA are given in [12] and [13].
Partial Least Squares Regression
Partial Least Squares (PLS) regression is a multivariate regression
method that relates a descriptor matrix, X, with a response matrix, Y.
Essentially, PLS maximises the covariance between the X-scores T and
the Y-scores U. The contribution of each x-variable to the explanation of
Y is calculated for each model dimension resulting in a weight matrix, W.
This weight matrix, W, contains the structure in X that maximises the
covariance between T and U in each dimension. The corresponding
matrix for the response matrix is denoted C. The matrix of X-loadings, P,
is calculated for each dimension in order to perform the appropriate
decomposition of X. The decomposition of X and Y can be described as:
X = TP´+ E
Y = TC´+ F
As opposed to MLR which requires, for example, independent, errorfree x-variables, the projection method PLS can handle collinear data that
has many more variables than objects, many response variables, and also,
to some extent, missing values. PLS extracts the latent structure in the
data and the estimate of the latent structure will be more precise the larger
the number of collinear variables included in the PLS modelling, just as
the accuracy of a mean value can be increased by performing repeated
A more comprehensive presentation of PLS is given in [14].
Modelling Nonlinearities
In cases where data deviate from linearity, linear modelling methods
such as MLR and PLS might be inadequate. A number of different
approaches to nonlinear PLS have been published, including neural
network [15], spline PLS analyses [16] and others. Another way to solve
nonlinear problems, but still use linear methods, is to make the
relationship between X and Y linear. This can be done in several ways,
e.g. by variable transformation, addition of squares and cross terms [17]
or by GIFI transformation, which will be described below.
In Paper IV, surfactant foaming was modelled as a function of the
physicochemical characterisation. A nonlinear relationship was expected
since surfactant foaming is known to have an optimum, so the GIFI-PLS
approach was used in the final PLS modelling.
The method of GIFI-PLS [18, 19] can be useful if nonlinearities are
present in the data. GIFI-PLS is flexible and assumes no particular form
of nonlinearity, even discontinuous relations between X and Y may be
approximated. The original GIFI-PLS approach, GIFI 1, divides the x
variable into a number of bins. Each bin is then represented by 1/0
dummies: 1 if the object has a value within the range of the bin and 0
otherwise. Each bin gives a new variable, resulting in an expansion of the
X matrix. This new X matrix is then used in linear PLS modelling and the
pattern among the regression coefficients reveals the type of nonlinearity
present. The limitation with this type of coding is the loss of resolution. If
many bins are used there is an increased risk of modelling noise, and if
too few bins are used objects with large differences in the original
variable are placed in the same bin.
Other approaches can be used to set up the GIFI coding and for more
comprehensive presentations of GIFI-PLS the accounts in [18] and [19]
are recommended. More detailed descriptions of other nonlinear methods
are given in [15-17].
Model Validation and Predictions
Models should be validated to ensure their relevance to the problem
under consideration. Several methods for model validation has been
developed, some of which are presented below. Two common parameters
of model quality, R2 and Q2, provide estimates of the ability of a model to
describe the data from which it is derived. They are expressed in terms of
explained variance and predictive quality, respectively and R2 indicates
how much of the variance in the observed data is explained by the model,
while Q2 indicates the accuracy of the model’s predictions. The explained
variance, R2, and predictive ability, Q2, are calculated as follows
R2 = 1-
å (Y
å (Y
- Ycalculated ) 2
- Ymean ) 2
Q2 = 1 -
å (Y
å (Y
- Ypredicted ) 2
- Ymean ) 2
The internal validation method, cross validation [20], generates Q2
values and is used during model development. Cross validation is based
solely on a modelling set, and during model development some of the data
are kept out, and are then predicted by the model and compared with the
observed values. This procedure is repeated until all data point has been
kept out once and only once.
Permutation tests [21] provide a way to check chance correlations
between the X and Y matrices. The order of Y is randomly permuted a
number of times, and separate models are fitted to all permuted Y’s. If, in
each case, the values of the fit (R2) and the prediction quality (Q2) are
substantially lower than corresponding values for the original model, the
model is considered valid since it predicts better than chance. This
method also indicates if a too complex model has been generated i.e. if
the data have been overfitted.
However, the predictive ability of a model is best verified by external
validation [21], e.g. by biological testing of some additional compounds
in the same way as the modelling set compounds. The experimental data
collected for the validation compounds are then compared with the values
predicted by the model. The compounds in the validation set should be
within the physicochemical domain of the modelling set. However,
limitations in the number of experiments, test animals, time or money
may make it difficult to perform external validation.
Modelling Toxicity
as mentioned earlier, used in personal
hygiene products, laundry and dishwashing detergents, which are
discharged into domestic wastewater after use and thus enter the
aquatic environment. Because of the increasing use of surfactants, the
identification of compounds with low toxicity and good surface activity
properties is of great interest. Development of Quantitative StructureActivity Relationship (QSAR) models provides a possible tool for this
search. QSARs relate physicochemical data, such as molecular weight and
solubility, to biological responses. The biological data describe properties
such as toxicity, pharmacological effects and carcinogenicity. Using this
technique only a relatively small number of experiments will be needed to
construct a model, which will give indications about compounds that are
expected to be harmful without any further biological testing.
The six-step strategy for the construction of a valid QSAR model is
based on statistical experimental design and multivariate modelling of the
relationships between chemical descriptors and biological responses [22].
The steps are: formulation of classes; characterization of the selected
class; selection of compounds to test; biological testing; model
development; model validation and prediction of untested compounds.
Each step is described in more detail below.
1. Classes
Classes should be composed of similar compounds since the
mechanism of biological action usually differs between different classes
of compounds. Chemicals can generally be divided into four major classes
according to their mode of toxic action. Class I compounds are relatively
unreactive chemicals with a non-specific mode of action, also known as
narcosis. Narcosis corresponds to the minimal level of toxicity and is also
referred to as baseline toxicity. It is often assumed that neither a specific
chemical nor a unique receptor is involved in narcosis. Narcosis due to
environmental pollutants in aquatic organisms is, according to van Wezel
[23], defined as a non-specific reversible disturbance of the functioning of
the membrane, caused by accumulation of the pollutant(s) in hydrophobic
phases within the organism. The disturbance of membrane function
results in decreased activity and diminished ability to react to stimuli, and
can ultimately lead to death.
More polar chemicals are defined as class II, reactive chemicals as
class III and chemicals with a specific mode of action as class IV
compounds. Compounds of classes II-IV cause mortality at much lower
concentrations than corresponding baseline toxicity compounds [24].
However, this is a rough classification and to refine it, subclasses are
required. For instance, surfactants are generally considered to be class I
compounds, but they can be further divided into anionic, cationic and
nonionic sub-classes. The nonionic surfactants can be further divided into
structural classes, such as AEOs and APGs.
2. Characterization
The characterisation, as mentioned in Chapter 4, should ideally be as
broad as possible i.e. many different descriptors, substituent and
molecular, measured and calculated should be used. The characterisation
depends on the considered compounds and also on the aim with the
system under investigation. Thus, it is important to consider the relevance
and interpretability of the descriptors used. For instance, if the aim is to
propose new environmentally friendly compounds it is important to
include interpretable structural descriptors in the characterisation.
3. Compound Selection
By analysing the data table with PCA, the original number of variables
can be reduced into a few information-rich PCs, see Chapter 5. These
components can be used as variables in multivariate design. In Paper I,
three principal components were used together with D-optimal design to
select a well-balanced modelling set. In Paper II, a 3D PCA score plot
formed the basis for the modelling set selection. It is important that the set
of compounds used as a modelling set should be well-distributed in the
variable space, i.e. the X-space, and contain representative compounds,
otherwise the predictability of the QSAR model will be lost.
4. Biological Testing
Some of the bioassays used today for testing are costly because they
require continuous culturing and maintenance of livestock to ensure there
are enough healthy test animals for toxicity tests. Differences in animal
livestock cultures are sources of variation both within and between
laboratories. However, the year-round commercially available fairy
shrimp, Thamnocephalus platyurus [25], and rotifer, Brachionus
calyciflorus [26], are well-suited as test organisms since they can be
stored for months and hatched upon demand by simple manipulations.
Further advantages are the reproducibility, ease and rapidity of testing
they allow. To get comparable results it is important that the biological
testing of all the selected compounds should be done under as similar
conditions as possible.
Since a compound may induce many different responses it can be
advantageous to include several biological effects in the QSAR model.
For the toxicity tests on shrimp and rotifer described in Papers I and II,
the specified effect was lethality (LC50) or, more precisely,
immobilisation. If the shrimp or rotifer was immobile and did not react
when touched with a glass rod it was considered dead.
5. Model Development
Regression methods relate the physicochemical variables, X, to the
biological responses, Y, see Chapter 5. It may be necessary to transform
some of the descriptor variables or remove compounds that have deviant
chemical and/or biological properties. In cases with many variables, a
multivariate regression method is needed. There might also be missing
values and noise in the data set. PLS has proven to be a well-suited
multivariate method in such cases [14].
6. Validation and Predictions
Models should be validated to ensure their accuracy, especially if the
purpose is to use the derived model for prediction of untested samples.
There are a number of ways to test a model, some of which are presented
in Chapter 5. When the model is considered to provide a good description
of the studied phenomena it can be used for predicting the response of
compounds that have not been previously examined.
Log P for Predicting the Toxicity
In traditional QSAR one single variable is often used to describe the
responses of interest. Over the years the most popular variable has been
Leo and Hansch’s octanol/water partition coefficient, log P, which
describes the hydrophobicity [27]. The concept underlying this method is
that the various fragments of a molecule contribute additively to its log P
value. The application of log P in QSAR is based on the assumption that
log P can be accurately measured or calculated, and that this variable is
sufficient to describe the toxicity of compounds and their uptake into
Surfactants are known to interact with biological membranes [28, 29].
At low surfactant concentrations the monomer adsorbs onto and absorbs
into the membrane. According to Muller [30], this increases both
membrane permeability and transmembrane solute transport. At
concentrations higher than the CMC the lamellar structure of the
membrane begins to disintegrate, and the membrane-bound proteins
become solubilized.
Könemann [31] correlated the aquatic toxicity of 2-3 month old guppy
fish, Poecilia reticulata, to log P for various pollutants from different
chemical classes. Several of the tested substances have an anaesthetic,
narcotic, potency. Quick recovery of the exposed fish was observed on
transfer into clean water, indicating the compounds had an anaestheticlike effect. The resulting QSAR model, shown below, is interpreted as
reflecting narcosis or baseline toxicity.
-logLC50 (M) = 0.871*log P + 1.13
LC50 is in mol per litre and log P is the octanol/water partitioning
coefficient. LC50 is defined as the concentration causing a specified effect,
for example immobilisation or mortality, in 50% of the test organisms.
Veith et al. correlated the aquatic toxicity of fathead minnow,
Pimephales promelas, to log P for a variety of polar compounds such as
amides, primary amines, substituted phenols, and pyridines [32]. The
resulting QSAR model is believed to reflect polar narcosis:
-logLC50 (M) = 0.65*log P + 2.29
here, again, LC50 is in mol per litre and log P is the octanol/water
partitioning coefficient.
The acute, short-term, toxicity of nonionic surfactants is here
modelled by Könemann’s general narcosis QSAR (♦) and Veith’s polar
narcosis QSAR (●). Figure 4 presents -log P versus the toxicity, i.e. log
LC50 values, for 36 AEOs. The toxicity values were obtained by
measuring the LC50 values of shrimp and rotifer, and by calculations using
the equations derived by Könemann and Veith. As shown in Figure 4 the
surfactants are better modelled by the polar narcosis QSAR than the
general narcosis QSAR. Deviation of predictions of general narcosis
toxicity levels for compounds from measured LC50 values indicates that
toxicity is probably caused by another mechanism, or that the log P value
alone is insufficient to explain toxicity. Significant differences can be
found when studying partitioning behaviour of polar and nonpolar
compounds in several kinds of organic solvents. Therefore, the
interpretation of toxicity data can depend strongly on the choice of
partition coefficient. Vaes et al. showed that for nonpolar and polar
narcotics the difference in toxicity QSARs disappeared when using log
Klipid/water as the hydrophobicity descriptor instead of log P [33]. Klipid/water
is the DMPC/water partitioning coefficient where DMPC is L-adimyristoyl phosphatidylcholine.
log LC50
General Narcosis QSAR
Polar Narcosis QSAR
Measured Toxicity, Shrimp
Measured Toxicity, Rotifer
-log P (PACO)
Figure 4. Shrimp and rotifer log LC50 values versus –log P values for 36 AEOs. The
LC50, values were obtained by measuring the toxicity towards shrimp and rotifer and by
calculations using Könemann’s general narcosis and Veith’s polar narcosis QSAR.
Should Log P or Surfactant Parameters be Modelled?
There has been a controversy in the literature about the use of log P
and other physicochemical descriptors to describe the aquatic toxicity of
surfactants. Roberts has been arguing for the use of log P [34] whereas
Morrall and Rosen have been arguing in favour of other descriptors [35,
Morrall and Rosen state that the use of log P is based on assumptions
valid for non-surface-active compounds (non-surfactants), but not
necessarily for surfactants. Roberts on the other hand states that nonsurfactants and surfactants form a continuum, with no clear boundary
between them. Consider the compound RO(CH2CH2O)nH, for example
(see Figure 5). It is a surfactant when R is n-dodecyl (C12) and n has any
positive value, but a non-surfactant when R is ethyl (C2) and n has any
positive value. But, if R = hexyl (C6) and n = any positive value, or R =
dodecyl and n = 0 is it then a surfactant or a non-surfactant?
R = C 12
R = C 12
Figure 5. Schematic figure illustrating the compound R-O-(CH2-CH2-O)n-H when the
alkyl chain (R) and number of ethoxylates (n) are varied.
Morrall and Rosen state that when using log P to describe toxicity it is
assumed that solute molecules do not interact with each other, but
surfactants exhibit concentration-dependent interactions with other
surfactants in both water and octanol. A second assumption is that the
presence of octanol in water and water in octanol does not affect the
results. However, surfactants interact significantly with both octanol and
water, and these interactions can increase the respective saturation
concentrations in either one or both of the phases. At equilibrium, log P is
the ratio of the concentrations of the compound in true solution in each of
the two solvent phases, according to Roberts. Although the presence of
surfactants affects the solubility of octanol in water and vice versa, the
same phenomenon is encountered with non-surfactant solutes. This can in
principle be accounted for by measuring log P at various concentrations
and extrapolating to infinite dilution.
Further, Roberts states that log P values should be straightforward to
calculate by the method of Leo and Hansch, since surfactants tend to be of
a relatively simple structure. On the other hand, according to Morrall and
Rosen the correlations used for calculating log P are related solely to the
hydrophobic part of the surfactant, although it is well known that
surfactant behaviour also depends on the polar head group. Despite this,
the correlations used by Roberts and others for calculating log P values
have shown that within a homologue series these values can be sufficient
to describe certain trends in surfactant toxicity. However, Morall and
Rosen argue that the empirical correlations used do not hold as well
across surfactant types, without special rules for adjusting the contribution
of the head group to the log P. Therefore, they claim that an alternative set
of physicochemical variables is needed.
The idea underlying the work by Morrall and Rosen is that surfactant
toxicity is best predicted from variables that describe the surface active
properties of the surfactants rather than by log P. Their selection of
variables was based on the hypothesis that in order to produce toxic
effects surfactants must not only adsorb onto, but also penetrate into the
organism. They chose the surfactant concentration in aqueous solution
required to reduce the surface tension of the solution by 20mN/m (pC20)
as a measure of adsorption efficiency. Then, as a measure of the ability of
the surfactant to penetrate into the organism the surfactant’s minimal
hydrated cross-sectional area (Amin) was used. Although the external cell
membrane is generally considered the site of action for surfactants, a good
correlation was found between toxicity and the pC20/Amin variable. A
plausible explanation, given by Roberts, for the observed correlation is
that pC20 and Amin in combination may actually model log P.
The work of Morrall and Rosen is based on surface tension
measurements of pure, single surfactant compounds, but Roberts argues
that commercial surfactants are usually mixtures of isomers and
homologues rather than pure single compounds, and for this reason a
variable that can be calculated for each component in the mixture is more
valuable in practice than one that has to be measured experimentally for
each component.
In this thesis two papers are discussed, Papers I and II, where
physicochemical variables of technical nonionic surfactants, i.e. isomer
and homologue surfactant mixtures, are correlated with their aquatic
toxicity. In both cases, the physicochemical characterisations are based on
both measured and calculated variables. The characterisation of the
AEOs, in Paper I, includes estimates of log P, calculated from the
average structure. A PCA on the physicochemical characterisation
resulted in three PC’s describing 79% of the variation. The PCA score
plot in Figure 6 shows the 18 of the total number of 36 surfactants
selected as the modelling set, and the remaining 18 surfactants comprising
the test set. This follows the way that the surfactants were selected in
Paper I.
Figure 6. PCA score plot of the first three PCs from the AEO study. The 18 surfactants
represented by black spheres were used for model development and the remaining 18
were used for validation.
For a comparative study, two different types of model were developed,
one multivariate and one based on log P alone. In the first case the
physicochemical variables are correlated with the toxicity and in the
second case only one variable, log P, is correlated with the toxicity. PLS
regression was used to develop the model in the multivariate case where
the toxicity towards the shrimp and the rotifer was modelled
simultaneously. In the log P case, linear regression was used and two
models were developed, one for each organism. The results, presented in
Figure 7a-d, clearly show that the multivariate model describes the data
better and gives smaller prediction errors.
Calculating the RMSEP (root mean squared error of prediction) values
for the multivariate and log P models confirmed these observations, see
Table 2. Use of the multivariate approach, when analysing these data,
decreases the prediction error for the shrimp and rotifer data by 28% and
47%, respectively.
Answering the question posed in the above heading, i.e. whether it is
best to use log P or surfactant parameters, I believe (based on the results
presented above) that using one single variable for toxicity modelling is
inadequate, while using a combination of both measured and calculated
variables gives satisfactory results, therefore surfactant properties are
Log P
Table 2. RMSEP values for the multivariate and the Log P model, respectively.
Multivariate Model
Log P Model
S hrimp
Multivariate Model
Log P Model
Me asured
Figure 7. Measured LC50 values versus values calculated for the modelling set
surfactants (filled symbols) and predicted for the validation set surfactants (open
symbols). Both multivariate and log P data are shown.
The U.S. Environmental Protection Agency (EPA) has been using
structure-activity relationships (SARs) since 1981 to predict the aquatic
toxicity of new industrial chemicals in the absence of test data. The
computer program ECOSAR (Ecological Structure Activity
Relationships) [37], is used to estimate the toxicity of chemicals used in
industry and discharged into water. The SARs presented in the ECOSAR
program are based on their similarity of structure to chemicals for which
aquatic toxicity has been previously measured. The acute toxicity of
chemicals to fish, invertebrates, and algae has been considered during the
development of the SARs, although for some chemical classes SARs are
available for other effects and organisms. Most SAR calculations are
based on log P, but the surfactant SAR calculations for AEOs are based
on the average length of their carbon chains and number of ethoxylate
The ECOSAR program was used in order to calculate the toxicity of
the 36 AEOs towards Daphnia/Fish, and the results are presented in
Figure 8a-d. As shown in Figure 8a the calculated responses of the
Daphnia/Fish are similar, parallel, to the measured responses of the
shrimp and rotifer. To assess the similarity of the results generated by the
ECOSAR program and our analysis (assuming that different organisms
have similar responses to a compound) these toxicity data are plotted
versus each other.
In Figure 8b the toxicity of the AEOs towards the shrimp is plotted
versus their toxicity towards the rotifer. Linear regression shows there is a
good correlation (R2=0.88) with the slope close to unity (1.02) confirming
the similarity in response. In Figure 8c and 8d the ECOSAR-calculated
toxicity values for the Daphnia/Fish are plotted versus the shrimp and
rotifer values, respectively. Linear regression gives good correlations
(R2=0.68 and 0.83), with slopes of 0.72 and 0.86, respectively. The
ECOSAR program underestimates to some extent the LC50 values, i.e. the
surfactants are estimated to be more toxic than they really are. The
deviations from zero of the intercepts (0.2 for the measured case,
compared to 0.6 and 0.8 for the Daphnia/Fish versus shrimp and rotifer,
respectively), also indicate that the ECOSAR program tends to
underestimate the LC50 values.
y = 1.02x + 0.2
R = 0.88
Measured Toxicity
ECOSAR: Daphnia/Fish Toxicity
y = 0.86x + 0.8
y = 0.72x + 0.6
R = 0.83
R = 0.68
ECOSAR: Daphnia/Fish
ECOSAR: Daphnia/Fish
Figure 8. a) Measured toxicity of the 36 AEOs towards the shrimp (triangles), and rotifer
(squares) versus their toxicity towards Daphnia/Fish as calculated by the ECOSAR
program. b) Measured toxicity towards shrimp versus measured toxicity towards rotifer.
c) Calculated toxicity towards Daphnia/Fish versus measured toxicity towards shrimp. d)
Calculated toxicity of Daphnia/Fish versus measured toxicity towards rotifer.
According to the ECOSAR technical reference manual [38] the
ECOSAR program uses local models for calculating toxicity i.e. different
SAR equations are used when the number of carbon atoms in the
hydrophobic part of the surfactants differs. In fact, there is one equation
for each carbon number ranging from 8 to 18. Ideally, however, the SAR
used for calculating the toxicity should cover the whole range of carbon
numbers. Further, the SAR models have been developed from the
responses of a variety of different species. For instance, the development
of the SAR equation for AEOs with 17 carbon atoms is based on fathead
minnow, rainbow trout, golden orfe and harlequin fish, i.e. four different
fish types.
Despite the rough methods used to develop the SAR models, the
ECOSAR program seems to work well if the aim is to get an overview of
the toxicity. However, compared to the results presented in this thesis,
ECOSAR appears to generally overestimate toxicity, see Figure 8c and
8d. Whether this overestimation of toxicity depends on incorrectness in
the program or if Fish/Daphnia is more sensitive towards surfactants than
shrimp and rotifer is difficult to determine.
Reactivity Studies
in many different applications. As
mentioned in the Introduction surfactants are used, inter alia, in
pulp and paper manufacture, household cleaning, cosmetics, and oil
recovery. The presence of surfactants is also believed to increase the
reactivity of dissolving pulp in the viscose process.
We encounter carbohydrates at almost every turn of our daily lives.
They occur in every living organism and are essential to life.
Carbohydrates give structure to plants, flowers, vegetables, and trees, and
serve as chemical energy-storage systems. The sugar and starch in food,
the wood of our houses, the paper on which this thesis is printed, and the
cotton of our clothes are made of carbohydrates. The name carbohydrate
originates from the first obtained pure simple carbohydrate, glucose.
Glucose, which has the molecular formula C6H12O6, was originally
thought to be a “hydrate of carbon”, C6(H2O)6. Even after that view was
abandoned the name remained, and today carbohydrate refers loosely to a
broad class of polyhydroxylated compounds. The carbohydrates may be
classified into three major groups. Monosaccharides, such as glucose,
mannose, galactose, and xylose, are simple sugars. Oligosaccharides are
sugars with less than ten monosaccharides joined together by glycosidic
linkages; examples of disaccharides are sucrose, fructose and lactose.
Polysaccharides, such as starch, glycogen and cellulose, are large sugars
with numerous monosaccharides joined together by glycosidic linkages
Cellulose is a polysaccharide made from glucose linked in b,1:4
fashion in very long unbranched chains. The b-glycosidic linkages make
the cellulose basically linear. The linear arrangement of the glucose units
in cellulose presents a uniform distribution of hydroxyl groups on the
outside of each chain. When two or more cellulose chains make contact,
the hydroxyl groups are ideally situated to form hydrogen bonds between
the chains, “zipping” them together. Zipping many cellulose chains
together results in a highly insoluble, rigid, and fibrous polymer. Some
parts of this polymer are highly ordered, forming crystalline regions,
while other parts are less ordered or completely unordered, forming
amorphous regions. The amorphous regions are fully accessible to water
and reagents, but the chain packing in the crystalline regions is very
dense, making them inaccessible to reagents and relatively unreactive. At
the super molecular level cellulose occurs in several polymorph forms
designated cellulose I, II, III and IV. The natural product, resulting from
biosynthesis, is referred to as cellulose I. Cellulose II is obtained by
treating cellulose I with sodium hydroxide, NaOH [40].
The definition of viscose given by the Bureau International pour la
Standardisation des Fibres Artificielles (BISFA) is simply “a cellulose
fibre obtained by the viscose process” [41]. After the pulp and paper
industry, the viscose industry is one of the largest processors of cellulose.
The viscose industry converts dissolving pulp, which is highly refined
pulp containing 95-98 % cellulose, into filaments, fibres, cords, casing
and film. These materials are used for many purposes, for example in
clothing, fabrics, tyres, packing material, diapers, tea bags, and non-edible
sausage casings [42].
Viscose Process
The conventional viscose process uses alkali and CS2 to break down
the inaccessible crystalline regions and convert cellulose into a liquid
called viscose. The process involves treatment of the dissolving pulp with
alkali, producing alkali cellulose, an aging step in which oxygen reacts
with the alkali cellulose chains to reduce the DP or average number of
glucose units in the cellulose polymer chain, and reaction with CS2 to
make cellulose xanthate.
Dissolving pulp
Cellulose-OAlkali cellulose
Cellulose xanthate
Regenerated cellulose
The charged, bulky, xanthate groups disrupt the forces holding the chains
together in the crystalline lattice formation, thereby promoting
solubilisation in alkali. Dissolution of cellulose xanthate in alkali results
in liquid viscose, which is taken through a spinning step in an acid bath,
thereby neutralizing the alkali in the viscose and reversing the xanthation
reaction to regenerate solid cellulose and CS2. During the spinning
process H2S is produced as a side-product. Depending upon the type of
spinning dye used, the cellulose is regenerated as fibre, cord, casing or
film [43].
The physical properties of cellulose fibres, as well as their chemical
behaviour and reactivity, are strongly influenced by the arrangement of
the cellulose molecules with respect to each other and the fibre axis [44].
Mercerization of cellulose irreversibly changes the native cellulose I
lattice structure to a more stable, lower energy structure, cellulose II. In
this process, alkali penetrates the cellulose fibre, swells it, and in some
fashion causes a change from a parallel-chain crystal structure, cellulose I,
to an antiparallel structure, cellulose II [40].
The formation of alkali cellulose is generally facilitated by a low
degree of crystallinity in the cellulose, and begins in the amorphous
regions. The directions of the chains in the cellulose fibre is random along
the fibre axis, i.e. approximately equal numbers of ‘up’ and ‘down’ chains
occur in the amorphous regions. NaOH interacts with chains running in
both directions in the amorphous regions and continues into small
crystallite zones, also running in both directions. It is within these
interfacial regions where NaOH can be expected to enter with relative
ease and the initial conversion to alkali cellulose occurs [45, 46].
Reactivity Measure
The viscose filter value is the most commonly used measure of the
reactivity of dissolving pulp. However, preparing a test viscose and
measuring its filterability is quite tedious and requires special equipment
A simpler method, presented by Fock (1959), to measure the reactivity
of dissolving pulp is to measure the content of non-reacted cellulose, i.e.
residual cellulose. This method is supposed to simulate the viscose
process at a laboratory scale. The first step in the Fock method is addition
of NaOH(aq) and CS2 to finely cut dissolving pulp i.e. cellulose. The
alkali solution deprotonates the hydroxyl groups in cellulose, forming
alkali cellulose and water. The alkali cellulose then reacts with CS2 and
cellulose xanthate is formed. The cellulose xanthate is neutralised with
sulphuric acid and cellulose is regenerated. Elg Christoffersson et al. have
shown that the viscose filter value and the residual cellulose estimates
correlate well [48].
Reactivity Increase
Increasing the reactivity of dissolving pulp can reduce the amount of
CS2 and NaOH used in the viscose process, thus reducing chemical costs.
From an environmental point of view increasing reactivity is also
beneficial since it lowers the amount of CS2 and by-products, such as
H2S, that is produced and released into the environment. The reactivity of
cellulose can be increased by various treatments, including (i) high
temperature mercerization, (ii) electron treatment, or (iii) addition of
(i) In high temperature mercerization (HTM), the cellulose is added to
a NaOH solution preheated to 105oC. This splits the fibrils into smaller
components (microfibrills), decreases the abundance of the crystalline
phase and the sizes of the crystallites, and removes the natural impurities
of cellulose, thus increasing reactivity [49].
(ii) Electron treatment disrupts the crystalline structure of cellulose
and forms various radiolytic products. These radiolytic products differ in
structure from their neighbours and cannot participate effectively in
hydrogen bonding. This results in localized defects that weaken the lattice
structures and enhance penetration of reagents into these regions, thus
increasing reactivity [42].
(iii) Surfactants can be added at various stages in the viscose process.
In mercerization the surfactant increases the wetting of the alkali solution,
in xanthogenation surfactants influence the dissolution of CS2, and in the
spinning bath surfactants decrease the adhesion between the viscose and
the spinneret [50].
Our Work
The work presented in Paper III examines whether the reactivity of
dissolving pulp could be improved by using surfactants that are soluble in
alkaline conditions. Resulting changes in reactivity were measured by
Fock’s method. The standard deviation in Fock’s method is known from
other studies [48] to vary by ±3 response units. Also, biological materials,
such as dissolving pulp, are known to vary between different samples.
Thus, to get valid, representative response values, series of experiments
should be repeated several times, and after a number of replicates the
responses should be averaged. Therefore, to find the minimum number of
replicates needed in this study, a series of experiments was performed.
Replicate Study
Four sheets of dissolving pulp from the same batch were used, and
five fixed positions on each sheet were investigated five times, resulting
in 100 experiments in total, Figure 9. In every experimental run, each
sheet was represented by at least one sample. The selection was random
i.e. all four corners and the middle samples were randomly selected five
times from each sheet. The overall average of the response was calculated
and compared with the average after n runs, where n=1, 2, 3,…100. The
confidence limit was set to ±3 response units, and the data were averaged
after five replicates to obtain a representative response.
Figure 9. Schematic illustration of how the samples in the replicate study were selected
from the dissolving pulp sheets. In every experimental run, each sheet was represented by
at least one sample.
Addition of Surfactant
At first the APGs were tested alone with no success, and later in a
successful combination with AEOs. In alkaline solutions APGs become
anionic, i.e. the hydroxyl groups are deprotonated by NaOH, and they lose
their surface activity as they become more and more soluble at higher
NaOH concentrations. On the other hand, AEOs are insoluble and retain
their surface activity when brought into alkaline solutions by the APGs.
Two AEOs were permuted with five APGs, resulting in ten surfactant
mixtures and the combination that showed the greatest potential was
investigated further with two types of designs, a full factorial design and a
central composite circumscribed (CCC) design [11]. The full factorial
design showed that the NaOH concentration was the most influential
variable. Increased NaOH concentration in the reaction solution probably
increased reactivity by increasing penetration of the NaOH into the
cellulose fibre i.e. it mainly affected the crystalline regions.
In the CCC design the NaOH concentration was kept constant and the
influence of time and the amount of added surfactant were studied in more
detail. The resulting model reveals that there is an optimum, either a
single point or plateau, for the amount of added surfactant. The
experiment where surfactant amount was screened indicated that this
optimum was a plateau, but use of the derived CCC model to calculate
improvements in reactivity indicated that it was a single point optimum,
since the reactivity slightly decreased when more than 0.07g/ml of
surfactant was added. The range of surfactant added was set in the region
where the increase in reactivity is most significant, and a too narrow range
was tested to establish clearly whether the optimum is a single point or a
plateau. Therefore, further experiments are needed with a wider range of
surfactant levels to address this issue. In detergency studies performed by
Å. Lindgren the maximum for the detergency effect proved difficult to
locate, because it appeared to ‘move’ [51]. The reason for this ‘moving
maximum’ was the setting of the variables in the three experimental
designs. In the first design the range of surfactant concentrations tested
was too narrow, in the second the range was set on the plateau, i.e. no
maximum at all was found, and in the third a maximum was found with
different variable settings from the first design. Finally, all three designs
were combined and the results revealed that the detergency effect reaches
a plateau, at which further large increases in surfactant concentration have
little influence on the detergency effect.
Surface tension measurements showed an interesting feature, since
plots of the surfactant concentration versus surface tension revealed two
breakpoints on the curve, see Figure 10. The first is thought to be the
CMC, while the second is not clearly established but indicate that the
surfactant leaves the solution and accumulates at the solution surface. The
first breakpoint coincides with ‘the optimum surfactant concentration’ i.e.
0.07g/ml. At the CMC the amount of free monomers in the solution is
constant, and adding more surfactant to the solution increases the number
and/or size of aggregates [52]. If this breakpoint is the CMC, adding more
surfactants to the solution will not increase reactivity further since
additional surfactant monomers will only cause an increase in the number
or size of aggregates, which do not enhance reactivity.
Breakpoint 1
Surface Tension (mN/m)
Breakpoint 2
Surfactant Concentration (g/ml)
Figure 10. Surfactant concentration versus measured surface tension showing two
interesting breakpoints, at approximately 0.065 g/ml and 0.6 g/ml, respectively. The first
breakpoint is thought to reflect the CMC and at the second the surfactant is believed to
leave the solution and accumulate at the solution surface.
Surfactant Action
Serebryakova et al. have studied the behaviour of surfactants at
different stages of the viscose process. They state that addition of
surfactants in the mercerization of cellulose improves the properties of
alkaline cellulose by promoting wetting by the alkali solution, swelling of
the fibres, removal of hemicellulose, and prevention of lump formation.
During xanthogenation the surfactants enhance the dissolution of CS2 in
NaOH due to its emulsification and thus alter the kinetics of penetration
of CS2 into the alkaline cellulose. By adding surfactants, the duration of
xanthogenation can be reduced by 30-40% without changing the quality of
the viscose. The presence of surfactants in the viscose and spinning bath
decreases their surface tension so that the adhesion between viscose and
spinneret decreases and the spinning stability increases. The authors also
give recommendations of surfactants to use in different stages of the
viscose process [50].
In the study described in Paper III the surfactant mixture was added to
the reaction flask at the same time as the NaOH and CS2, so the specific
action of the surfactant mixture used is uncertain. However, there are at
least two ways it could act: as a wetting agent, i.e. increasing the wetting
of the cellulose fibre, and/or as a solubiliser of CS2. The contact angle, i.e.
the wetting ability, in 40% alkaline solution of three APGs, with and
without an AEO, has been investigated by Johansson et al. [53]. The
contact angle of the AEO/APG mixture was smaller than that of the APGs
alone i.e. the mixture was more efficient as a wetting agent. The wetting
ability was also studied in 30% alkaline solution by measuring the
shrinkage rate of a yarn. Here, they found similar shrinkage rates for
solutions containing the AEO/APG mixture as for a fully formulated
commercial product. An increase in fibre wetting would increase the
penetration ability of the alkali into the cellulose fibre and thus improve
The low solubility of CS2 in water, 0.294% at 20°C, might be
increased by addition of surfactants, and micelle formation. The micelles
thus formed would contain CS2 in the interior, so the CS2 concentration
close to the fibre surface would be increased. This would mean that the
CS2 reaches the cellulose surface much more readily and quickly, leading
to an increase in the cellulose reactivity if the CS2 transport to the fibre
surface is the rate-limiting step.
A disadvantage of using surfactants is the foam production they may
induce in the viscose and spinning bath, therefore surfactants with high
foaming power and high foam stability should be avoided. By using
multivariate characterisation and analysis, surfactants with low foaming
properties can be found (Paper IV).
The work presented in Paper III shows that the cellulose reactivity
could be improved by adding surfactant mixtures that are soluble in
alkaline conditions. The surfactant mixture studied gave a reactivity
increase of 20%units i.e. showing that such mixtures have great potential
as additives in alkaline processes where cellulose is modified.
Concluding Remarks
Nonionic Surfactants – A Multivariate Study,
implies, this thesis discusses nonionic surfactants from a
multivariate point of view.
Since surfactants of technical grade are mixtures of isomers and
homologues their characterization is not as straightforward as for single
component compounds. Measuring surfactant properties such as surface
tensions, contact angles, etc., is relatively straightforward, but trying to
use molecular modelling is more difficult. One problem that almost
immediately arises is how to treat the surfactant mixture: as an average
molecule or as a weighted average molecule. By treating the mixture as an
average molecule only one calculation is needed, whereas treating it as a
weighted average molecule requires several calculations, one for each
isomer and homologue in the mixture, with weighting factors to account
for the distribution of these isomers and homologues.
In Paper I, the characterization of AEOs was extended with calculated
log P values derived from the average structure. In this thesis a
comparative study between a multivariate model and a log P model was
included. The log P model was based solely on log P values, whereas the
multivariate model included log P and other physicochemical variables.
The results show that the multivariate model gave smaller prediction
errors than the log P model. The multivariate model also gave additional
information about features of surfactant structure related to toxicity, for
instance that the branching of the hydrophobic part was important, see
Paper I.
Surfactants are useful in many different applications. Surfactants can,
for instance, increase the reactivity of cellulose in the viscose process. A
study of the changes in reactivity when a mixture of AEOs/APGs was
added to the reaction solution is presented in Paper III. The results
showed that mixtures of AEOs and APGs have great potential as additives
in alkaline processes where cellulose is modified.
To extend the work presented in Paper III, a multivariate design
should be used for selecting other AEOs to combine with APGs.
Combining the results from a multivariate design together with data from
the toxicity tests described in Papers I and II, might lead to an
environmentally friendly AEO/APG mixture that further increases the
cellulose reactivity. However, I think that it is important to perform
biodegradation studies on AEOs, APGs or on AEO/APG mixtures before
drawing any final conclusions about their toxicity. A surfactant that is
initially very toxic and easily biodegraded might cause less harm to the
environment than a less toxic surfactant that is very slowly degraded.
end is near… och cirkeln är sluten. Redan som
liten parvel undersökte jag ytaktiva ämnen, företrädesvis då tvål.
Mina metoder var inte särskilt sofistikerade, jag provsmakade
helt enkelt en härlig grön tvål. Min mamma är lättad över att jag idag har
funnit andra sätt att karaktärisera ytaktiva ämnen på eftersom det blir en
hel del lödder om man dricker mjölk efter man ätit tvål...
Ett stort tack till alla er som gjort denna avhandling möjlig och som
bidragit till att göra livet glatt och enkelt. Ni är många, och nedan är några
jag vill nämna lite extra.
Jag börjar med mina handledare Michael Sjöström och Svante Wold
som gav mig förtroendet att göra detta.
Åsa Lindgren, min mentor, för din aldrig sinande entusiasm. Tänk, att
du alltid hittar någonting positivt hur nattsvart det än ser ut! Tack LillStumpan!
Ingegärd Johansson, Christine Strandberg och Bosse Nilsson vid Akzo
Nobel Surface Chemistry i Stenungsund för ovärdelig teoretisk och
praktisk hjälp med bla. APG karaktäriseringen.
Asbjörn Bråtekas vid Borregaard ChemCell och Bertil Lindquist vid
Domsjö Paper för att ha försett mig med dissolving massa i obegränsade
Till alla er på Organisk Kemi som fixat med papper, Carina Sandberg
och Ann-Helén Waara, mixtrat med mina datorer, Bert Larsson, eller på
annant sätt hjälpt mig eller förgyllt fikarasterna med glada skratt.
Paul Geladi för alla nyttiga artiklar du letat fram och för all choklad du
släpat hem från Belgien.
Min rumskamrat, Jon för att du med ditt lugn och din hjälpande hand
sett till att datorn fortfarande står kvar på skrivbordet. Otroligt!
Kristina för att du lurade in mig i dissolving massans mystiska värld.
Till alla innebandylag jag haft äran att spela med; Bunsen Burners,
Bunsen Flames, Policestation mfl., och alla beachvolleyspelare som frusit
om fötterna tillsammans med mig på IKSU; PelleP, FreddaFred, Martin,
Göran, Malin K, David och Pära.
Det största tacket till alla mina vänner och min familj för att ni alltid
finnas till hands, Ni är guld värda!
Och sist men inte minst, vill jag tacka mannen och pojken i mitt liv,
Anders för att du fixar och donar och gör så att jag har det bra och Alvin
för det charmtroll du är.
Slutligen vill jag bara säga...
‘I did it my way…’
P. Anka, J. Revaux, C. Francois & G. Thibault
Rosen, M., Surfactants and Interfacial Phenomena. 2nd ed. 1989, New York, USA:
John Wiley & Sons Inc.
Hill, K., W. von Rybinski, G. Stroll, ed. Alkyl Polyglycosides - Technology,
Properties and Application. 1997, VCH: Weinheim, Germany.
Stryer, L., Biochemistry. 1981, New York, USA: W.H. Freeman and Company.
Israelachvili, J.N., Intermolecular and Surface Forces. 2nd ed. 1992, London, UK:
Academic Press.
Hellberg, S., M. Sjöström, B. Skagerberg, C. Wikström and S. Wold, Peptide
Quantitative Structure-Activity-Relationships, a Multivariate Approach. Journal of
Medicinal Chemistry, 1987. 30(7): p. 1126-1135.
Lindgren, Å. and M. Sjöström, Multivariate Physicochemical Characterisation of
Some Technical Nonionic Surfactants,. Chemometrics and Intelligent Laboratory
Systems, 1994. 23: p. 179-189.
Johansson, I., C. Strandberg, B. Karlsson and B. Gustavsson, Structure-property
relationships for some alkyl glucosides. in 4th World Surfactant Congress. 1994.
Barcelona, Spain.
Box, G.E.P., W.G. Hunter, and J.S. Hunter, Statistics for Experimenters, An
Introduction to Design, Data Analysis, and Model Building. 1978, New York, USA:
John Wiley & Sons, Inc.
Jackson, J.E., A Users Guide to Principal Components. 1991, New York: Wiley.
Joliffe, I.T., Principal Component Analysis. 1986, New York: Springer.
Wold, S., M. Sjostrom, and L. Eriksson, PLS-regression: a basic tool of
chemometrics. Chemometrics and Intelligent Laboratory Systems, 2001. 58(2): p.
Holcomb, T.R. and M. Morari, Pls Neural Networks. Computers & Chemical
Engineering, 1992. 16(4): p. 393-411.
Wold, S., Nonlinear Partial Least-Squares Modeling .2. Spline Inner Relation.
Chemometrics and Intelligent Laboratory Systems, 1992. 14(1-3): p. 71-84.
Berglund, A. and S. Wold, INLR, implicit non-linear latent variable regression.
Journal of Chemometrics, 1997. 11(2): p. 141-156.
Eriksson, L., E. Johansson, F. Lindgren and S. Wold, GIFI-PLS: Modeling of NonLinearities and Discontinuities in QSAR. Quantative Structure Activity
Relationships, 2000. 19: p. 345-355.
Berglund, A., N. Kettaneh, L. Uppgård, S. Wold, N. Bendwell and D.R. Cameron,
The GIFI approach to non-linear PLS modeling. Journal of Chemometrics, 2001.
15: p. 321-336.
20. Wold, S., Cross-Validatory Estimation of the Number of Components in Factor and
Principal Components Models. Technometrics, 1978. 20(4): p. 397-405.
21. Eriksson, L., E. Johansson, and S. Wold. QSAR Model Validation. in Quantitative
Structure-Activity Relationships in Environmental Science - VII. QSAR 1996. 1997.
Elsinore, Denmark: SETAC, Pensacola, Florida, USA.
22. Jonsson, J., L. Eriksson, M. Sjöström and S. Wold, A Strategy for Ranking
Environmentally Occuring Chemicals. Chemometrics and Intelligent Laboratory
System, 1989. 5: p. 169-186.
23. van Wezel, A.P. and A. Opperhuizen, Narcosis due to Environmental Pollutants in
Aquatic Organisms: Residue-Based Toxicity, Mechanisms, and Membrane Burdens.
Critical Review in Toxicology, 1995. 25: p. 255-279.
24. Verhaar, H.J.M., C.J. van Leeuwen and J.L.M. Hermens, Classifying Environmental
Pollutants. 1. Structure-Activity Relationships for Prediction of Aquatic Toxicity.
Chemosphere, 1992. 25(4): p. 471-491.
25. Centeno, M.D.F., G. Persoone, and M.P. Goyvaerts, Cyst-based toxicity tests. IX.
The potential of Thamnocephalus platyurus as test species in comparison with
Streptocephalus proboscideus. Environmental Toxicology and Water Quality, 1995.
10: p. 275.
26. Snell, T.W., B.D. Moffat. C.R. Janssen and G. Persoone, Acute Toxicity Tests
Using Rotifers: IV. Effects of Cyst Age, Temperature, and Salinity on the Sensitivity
of Brachionus calyciflorus. Ecotoxicology and Environmental Safety, 1991. 21: p.
27. Hansch, C., and A.J. Leo, Substituent Constants for Correlation Analysis in
Chemistry and Biology. 1979, New York, USA: John Wiley & Sons, Inc.
28. Abel, P.D., Toxicity of Synthetic Detergents to Fish and Aquatic Invertebrates.
Journal of Fish Biology, 1974. 6: p. 279-298.
29. Granmo, Å. and S. Kollberg, Uptake Pathways and Elimination of a Non-Ionic
Surfactant in Cod (Gadus morrhua L.). Water Research, 1976. 10: p. 189-194.
30. Muller, M.T., A.J.B. Zehnder and B.I. Escher, Liposome-Water and Octanol-Water
Partitioning of Alcohol Ethoxylates. Environmental Toxicology and Chemistry,
1999. 18(10): p. 2191-2198.
31. Könemann, H., Quantitative structure-activity relationships in fish toxicity studies:
part 1. Relationships for 50 industrial pollutants. Toxicology, 1981. 19: p. 209-221.
32. Veith, G., and S.J. Broderius, Structure-toxicity relationships for industrial
chemicals causing type (II) narcosis syndrome, in QSAR in Environmental
Toxicology - II, K.L.E. Kaiser, Editor. 1987, Reidel Publishing Company.
33. Vaes, W.H.J., E. Urrestarazu Ramos, H.J.M. Verhaar and J.L.M. Hermens., Acute
Toxicity of Nonpolar Versus Polar Narcosis: Is There a Difference?,. Environmental
Toxicology and Chemistry, 1998. 17: p. 1380-1384.
34. Roberts, D.W., Aquatic Toxicity-Are Surfacant Properties Relevant. Journal of
Surfactants and Detergents, 2000. 3(3): p. 309-315.
35. Morrall, S.W., M.J. Rosen, Y.-P. Zhu, D.J. Versteeg, S.D. Dyer. Physicochemical
Descriptors for Development of Aquatic Toxicity QSARs for Surfactants. in QSAR
96. 1996. Elsinore, Denmark: SETAC.
36. Rosen, M.J., L. Fei, Y-P. Zhu and S.W. Morrall, The Relationship of the
Environmental Effect of Surfactants to Their Interfacial Properites. Journal of
Surfactants and Detergents, 1999. 2(3): p. 343-347.
39. Solomons, G.a.C.F., Organic Chemistry. 7th ed. 2000, New York: John Wiley &
Sons, Inc.
40. Sjöström, E., Wood Chemistry - Fundamentals and Applications. 1981, Orlando,
Florida, USA: Academic Press, Inc.
42. Stepanik, T.M., et al., Electron-Processing Technology: A Promising Application
for the Viscose Industry. Radiation Physics and Chemistry, 1998. 52: p. 505-509.
43. Treiber, E.E., Formation of fibers from cellulose solutions. Cellulose Chemistry and
Its Applications, ed. T.P.N.a.S.H. Zeronian. 1985, Chichester, UK: Ellis Horwood
Ltd. 455-479.
44. Majdanac, L.D., D. Poleti, and M.J. Teodorovic, Determination of the Crystallinity
of Cellulose Samples by X-ray diffraction. Acta Polmerica, 1991. 42(8): p. 351-357.
45. Nishimura, H. and A. Sarko, Mercerization of Cellulose. III. Changes in Crystallite
Sizes. Journal of Applied Polymer Science, 1987. 33: p. 855-866.
46. Nishimura, H. and A. Sarko, Mercerization of Cellulose. IV. Mechanism of
Mercerization and Crystallite Sizes. Journal of Applied Polymer Science, 1987. 33:
p. 867-874.
47. Internal Method, S.R., Bestämning av filtrerbarhet för viskos genom
konstanttryckfiltrering. Svenska Rayon AB, SE-660 50 Vålberg, Sweden.
48. Elg Christoffersson, K., M. Sjöström, U. Edlund, Åsa Lindgren and M. Dolk,
Reactivity of Dissolving Pulp: Characterisation Using Chemical Properties, MNR
Spectroscopy and Multivariate Data Analysis. Cellulose.
49. Pavlov, P., V. Makaztchieva, and E. Lozanov, High Reactivity of Cellulose After
High Temperature Mercerization. Cellulose Chemistry and Technology, 1992. 26:
p. 151-160.
50. Serebryakova, Z.G. and L.G. Tokareva, Surfactants and Modifiers in Production of
Viscose Fibres (Review). Fibre Chemistry, 1996. 28(2): p. 91-94.
51. Lindgren, Å., Multivariate Structure-Property Relationships of Non-Ionic
Surfactants, in Department of Organic Chemistry. 1995, Umeå University: Umeå.
52. Shaw, D.J., Introduction to Colloid and Surface Chemistry. 1992, Trowbridge,
Wiltshire: Redwood Books.
53. Johansson, I., C. Strandberg, B. Karlsson, G. Karlsson and K. Bergström, Use of
mixtures of alkyl alkoxylates and alkyl glucosides in strong electrolytes and highly
alkaline systems. Industrial Applications of Surfactants, ed. D. Karsa. Vol. 4. 1999,
Camebridge: Royal Chemical Society.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF