Effects of background cations on the fouling of polyethersulfone membranes by natural organic matter: Experimental and molecular modeling study

Abstract

Adsorptive fouling of a polyethersulfone (PES) membrane by natural organic matter (NOM) in the presence of common metal cations was investigated with both experimental and computational molecular modeling techniques. NOM, calcium, magnesium and silicon were identified as important foulants through chemical analysis of two Midwestern surface waters. A model feed solution mimicking the lake waters and containing only NOM (Suwannee River NOM) and Ca2+ resulted in a fouling pattern similar to the surface waters; Mg2+ and Na+ caused much lower fouling at the same ionic strength as the calcium solution. Molecular modeling of the model solution allowed detailed probing of the fouling process. This work suggests that divalent ions (Ca2+ and Mg2+) may cause membrane fouling not by forming “ionic bridges” between the negatively charged functional groups on membrane surface and the negatively charged functional groups of NOM, but by promoting the aggregation of NOM molecules in solution. The carboxyl groups of NOM strongly associate with the divalent ions, while the sulfonyl groups in the polyethersulfone do not. Although Ca2+ and Mg2+ are both coordinated to the NOM carboxyl groups predominantly by outer-sphere-type complexation, Ca2+ associates with the carboxyl groups more strongly than Mg2+ due to the looser second hydration shell structure of Ca2+. The stronger Ca2+-NOM complexation is also manifested by the decreased mobility (diffusion coefficients) of the Ca2+ bound to the NOM.

Full text

Page 1

Available online at www.sciencedirect.com
Journal of Membrane Science 309 (2008) 128–140
Effects of background cations on the fouling of polyethersulfone
membranes by natural organic matter: Experimental and
molecular modeling study
Won-Young Ahna, Andrey G. Kalinichevb,∗, Mark M. Clarka
aDepartment of Civil and Environmental Engineering, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
bDepartment of Geology and NSF WaterCAMPWS, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
Received 9 August 2007; received in revised form 5 October 2007; accepted 10 October 2007
Available online 26 October 2007
Abstract
Adsorptive fouling of a polyethersulfone (PES) membrane by natural organic matter (NOM) in the presence of common metal cations was
investigated with both experimental and computational molecular modeling techniques. NOM, calcium, magnesium and silicon were identified as
important foulants through chemical analysis of two Midwestern surface waters. A model feed solution mimicking the lake waters and containing
only NOM (Suwannee River NOM) and Ca2+resulted in a fouling pattern similar to the surface waters; Mg2+and Na+caused much lower fouling
at the same ionic strength as the calcium solution. Molecular modeling of the model solution allowed detailed probing of the fouling process.
This work suggests that divalent ions (Ca2+and Mg2+) may cause membrane fouling not by forming “ionic bridges” between the negatively
charged functional groups on membrane surface and the negatively charged functional groups of NOM, but by promoting the aggregation of NOM
molecules in solution. The carboxyl groups of NOM strongly associate with the divalent ions, while the sulfonyl groups in the polyethersulfone do
not.AlthoughCa2+andMg2+arebothcoordinatedtotheNOMcarboxylgroupspredominantlybyouter-sphere-typecomplexation,Ca2+associates
withthecarboxylgroupsmorestronglythanMg2+duetotheloosersecondhydrationshellstructureofCa2+.ThestrongerCa2+-NOMcomplexation
is also manifested by the decreased mobility (diffusion coefficients) of the Ca2+bound to the NOM.
© 2007 Elsevier B.V. All rights reserved.
Keywords: Ultrafiltration; Membrane fouling; Natural organic matter (NOM); Calcium; Molecular modeling
1. Introduction
Prediction of ultrafiltration performance is not generally
possiblewithoutperformingpilot-scaletests[1,2]becausemem-
brane fouling is related to membrane material and solution
chemistry [1,3–9]. The inadequate understanding of membrane
foulinghasinmanycaseshinderedwideradoptionofmembrane
processes in large-scale drinking water treatment plants [10].
Several fouling mechanisms are described in the literature.
Hydrodynamic forces exerted on particles and formation of
cake layers have long been studied [5,11–13]. Membrane mod-
ules for large-scale applications are designed in cross-flow
configurations [5,14], and excessive cake formation can be
∗Correspondingauthor.Presentaddress:DepartmentofChemistry,Michigan
State University, East Lansing, MI 48824-1322, USA. Tel.: +1 517 355 9715.
E-mail address: kalinich@chemistry.msu.edu (A.G. Kalinichev).
controllediffiltrationisperformedundersub-critical-fluxcondi-
tions[15–17].Filterdepositsformedundertheseconditionstend
to be easily removed yielding sustainable water productivity.
An often more-problematic flux loss is caused by adsorption
of macromolecules on the membrane surface and in the mem-
brane pores; this type of fouling can be especially important
sincedrinkingwatersuppliescontainNOM[18–22].Hydropho-
bic membranes have long been considered more prone to the
adsorptive fouling than hydrophilic membranes [1,3,4]. Never-
theless, hydrophobic materials are still used because of their
physical and chemical stability, and several studies have exam-
ined surface modification of hydrophobic membranes through
additions of hydrophilic moieties [23–25].
Although NOM is considered as a major membrane foulant
in drinking water treatment, the exact fouling mechanism is
still debated. Researchers have attempted to determine if NOM
hydrophobicity is related to fouling. Amberlite XAD-8 and
XAD-4 resins [26,27] have been used to separate and isolate
0376-7388/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2007.10.023

Page 2

W.-Y. Ahn et al. / Journal of Membrane Science 309 (2008) 128–140
129
the hydrophobic and hydrophilic portions of NOM, and exam-
ine the fouling characteristics of the fractions separately. While
Nilson and DiGiano [28] claimed that the hydrophobic portion
of NOM from the Tar river in North Carolina caused almost all
fouling of a polysulfone NF membrane, Lee et al. [29] reported
that the neutral hydrophilic fraction of NOM from river waters
caused more fouling of various MF and UF membranes. Lee
et al. [29] argued that the hydrophilic fraction, composed of
polysaccharide-like colloidal matter, behaved very differently
from the humic portion of NOM that is often assumed to be the
primary foulant. The neutral hydrophilic portion of NOM was
also implicated in fouling by Carroll et al. [30].
Contrasting experimental results and theories such as these
indicate that the interactions between membrane surfaces and
membrane foulants are not well understood. A more detailed
picture of molecular interactions between membrane surfaces
and NOM are required in order to better understand the fouling
phenomenon.
Although the molecular structure of NOM is not very well
defined, various acidic functional groups are generally present
in NOM, and the concentration of carboxyl functional groups
determines the organic acid characteristics of the NOM [31].
Thus, NOM is usually classified by its solubility in water at
various pH values [32,33]. High ionic strength reduces the elec-
trostaticdoublelayerthicknessesofbothmembranesurfaceand
NOM [34,35], which is believed to cause more severe mem-
brane fouling [4,6,35]. In addition to the general ionic strength
effect, the complexation of divalent ions (i.e., Ca2+and Mg2+)
with acidic groups of NOM in natural waters has been consid-
ered to enhance fouling through NOM aggregation [4,6,36,37].
Ca2+appears to interact more strongly with negatively charged
carboxyl groups of NOM than Na+and Mg2+[38]. Binding
of metal ions to acidic groups of NOM has been extensively
studied [39–44] and quantitative models have been suggested
[42,45–48].AlthoughresearchershaveobservedthatCa2+inter-
acts stronger than Mg2+with acidic functional groups, there is
not yet an adequate explanation of why Ca2+causes greater
membrane fouling.
Computational molecular modeling techniques are success-
fully applied in the development of better molecular-level
understanding of various properties and processes in modern
chemical,biological,andmaterialssciences.Theycanbeusedto
predict the structures and chemical interactions of biomolecules
[49–53] and the behavior of polymer materials [54–56]. These
computational modeling techniques have also been used to
understand many membrane properties and processes on a fun-
damental molecular level [57–63]. In the present work, we
applied molecular dynamics (MDs) computer simulations to
model the processes related to adsorptive membrane fouling
phenomena.
Although there can be no unique chemical structural model
of NOM, several building blocks of humic substance struc-
tures have been proposed [64,65] and adopted in computational
molecularmodelingworks[66–69].SchultenandSchnitzer[64]
suggested a polymeric humic acid structure with an elemen-
tal composition of C342H388O124N12and a molecular weight
of 6651Da. Pyrolysis-GC/MS, pyrolysis-FIMS,13C NMR,
electron microscopy, and several other analytic results were
used to deduce this structure. Jansen et al. [65] proposed a
three-dimensionalstructure,theso-calledTemple-Northeastern-
Birmingham (TNB) humic acid building block. The TNB
structure had a chemical composition of C38H39O16N with a
molecularweightof765Da.Theauthorsscreenedoutunfeasible
chiral isomers that could not result from biological trans-
formations. This model was then successfully used in the
conformational modeling study [70].
The interaction between NOM and metal ions in aqueous
solutions has also been recently studied by molecular modeling
techniques [66–69]. Sutton et al. [66] evaluated the hydration of
the Schulten’s NOM model [64] and the interactions of Na+and
Ca2+ions with the NOM molecule. They reported that calcium
ions bind more strongly with NOM carboxylate groups than
sodium, and that calcium yields better NOM hydration than
sodium. It is thought that calcium better penetrates the NOM
structure and creates more space for water molecules [66]. Xu
etal.[67]combinedNMRspectroscopyandMDmodelingofan
NOM-CsClsolutiontoshowthatCs+ionsformedouter-sphere-
type complexes with the NOM, while Cl−ions did not interact
with the NOM molecule. As a result, the diffusional mobility
of the Cs+ions decreased significantly, especially at low CsCl
concentrations. The TNB model of NOM [65] was employed
in these simulations because of its compositional similarity
to the Suwannee River NOM (SRNOM) used in their experi-
ments. The molecular weight of the TNB model [65] also seems
consistent with recent studies suggesting that heterogeneous
molecules, which have a molecular weight of about 500Da,
aggregate in aqueous solutions to form various supramolecular
structures where hydrogen bonding between carboxyl groups
andvanderWaalsforcesbetweenaromaticandaliphaticcarbon
structures hold the structure together [71–73]. Multications can
increase apparent molecular weights of NOM through binding
[74,75]. Recently, Kalinichev and Kirkpatrick [68] performed
MD simulations to demonstrate that Ca2+forms predominantly
inner-sphere complexes with NOM, while the strong hydration
shell of Mg2+ions significantly decreases their ability to form
evenouter-spherecomplexeswithnegativelychargedfunctional
groups of NOM. Deprotonated carboxyl groups were the major
binding sites and the contribution of phenolic groups was neg-
ligible.
Inthisstudy,weinvestigatedpossiblemolecularmechanisms
ofmembranefoulingbyNOMinthepresenceofcommonmetal
cations. The main hypothesis at the outset of this work was that
membrane fouling could be caused by cation binding of the
partially negatively charged sulfonyl groups of the PES mem-
braneandthedeprotonatedcarboxylgroupsofNOMmolecules.
The degree of binding can vary depending on the electrostatic
charges and the sizes of the cations, because these parameters
affect the ionic hydration structure and energy [68]. In order to
quantitatively address the degree of binding interactions, mem-
brane fouling by NOM was evaluated by both experiments and
molecular modeling. Two Midwestern lake waters and Suwan-
neeRiverNOMwereusedfortheultrafiltrationexperimentsand
the TNB molecular model [65] was adopted as a representative
structure of NOM for the molecular dynamics simulations.

Page 3

130
W.-Y. Ahn et al. / Journal of Membrane Science 309 (2008) 128–140
2. Experimental
2.1. Methods
2.1.1. Preparation of water samples
Two different Midwestern lake waters were directly sam-
pled from the intake points of drinking water treatment plants.
Lake Decatur water was collected from the South Water Treat-
ment Facility at Decatur, Illinois. Lake Michigan water was
collected from the Membrane Drinking Water Treatment plant
in Kenosha, Wisconsin. The collected samples were stored in
a 4◦C cold room to prevent microbial activity. A two-step pre-
filtration, using a glass-fiber filter (AP40, Millipore, Billerica,
Massachusetts) and a 0.45-?m hydrophilic nylon filter (Magna,
GE Osmonics, Minnetonka, Minneapolis), was conducted to
remove suspended solids from the feed water.
Freeze-dried powdered Suwannee River NOM (SRNOM)
was purchased from the International Humic Substances Soci-
ety (Lot number: 1R101N, St. Paul, Minneapolis). Powdered
SRNOM was resolubilized in reagent-grade water to prepare
a stock solution, which was diluted to 6mg/L. The 6mg of
SRNOMwasequivalentto2.4mgofcarbon.SincetheSRNOM
solution contained fewer crude particles than the lake waters,
removal of suspended solids was achieved in a single-step using
a 0.45-?m hydrophilic nylon filter.
2.1.2. Ultrafiltration
A flat-sheet PES UF membrane with an average MWCO
of 20,000Da, was kindly supplied by the manufacturer, GE
Osmonics (Minnetonka, Minneapolis). A dead-end, stirred-cell
unit (Amicon Model 8050, Millipore, Billerica, Massachusetts)
was used to evaluate water flux through the UF membrane. Dur-
ing filtration, the solution was continuously stirred at a fixed
rotating speed of 275rpm. Filtration was performed under a
constant pressure of 105±2.07kPa, which was provided by a
nitrogen gas cylinder. The filtration process consisted of five
steps: wetting and cleaning the membrane, measuring clean
water flux, filtering the NOM water sample, cleaning the mem-
brane, and measuring clean water flux again. Forty milliliters
of deionized water was then introduced into the cell and stirred
without external pressure for 1min for the cleaning. This step
was repeated three times. Water flux was measured by weigh-
ing the permeate water over time with a digital balance (Model
PB3002-S, Mettler-Toledo, Columbus, Ohio). Data from the
digital balance were directly recorded using data-acquisition
software (Winwedge Standard, TAL Technologies, Philadel-
phia, Pennsylvania). Temperature was adjusted to 294±1K by
placement of a glass sample container in a continuously circu-
lating water bath before the experiment.
2.1.3. Analysis
NOM was quantified as dissolved organic carbon (DOC),
which was analyzed using a Phoenix 8000 TOC Analyzer
(Tekmar-Dohrmann, Cincinnati, Ohio).
Metal analyses of the samples were conducted using the
inductive coupled plasma (ICP) technique at the Illinois Water
Survey’s (ISWS) Analytical Chemistry and Technology Unit in
accordance with USEPA Method 200.7, Revision 4.4 (USEPA,
1994). The samples were preserved with 0.2% HNO3(Fisher
Optima High Purity Grade) immediately upon arrival in the lab-
oratory. Samples were digested with 1% HNO3and 0.5% HCl
Optima acids in metal-free polypropylene sample tubes using a
CPI ModBlockTM digestor. Digestates were analyzed for met-
als using a Thermo Jarrell Ash 61E ICP vacuum spectrometer
with a fixed channel polychromator.
Chemical elemental compositions of the clean membrane
and the fouled layers were analyzed by energy-dispersive
spectroscopy (EDS) using an environmental scanning electron
microscope(ESEM),PhilipsXL30ESEM-FEG(FEICompany,
Hillsboro,Oregon).Themembranesamplesweredriedinades-
iccator for 5 days before analysis. 4-nm thick Au–Pd layer was
sputter-coated in order to endow conductivity to the samples.
Chemical functional groups of membrane foulants collected
on the surface of membranes were analyzed using attenuated
total reflection Fourier transform infrared (ATR-FTIR) spec-
troscopy (Nexus 670 spectrometer, Thermo Nicolet, Madison,
Wisconsin). One liter of pre-filtered water sample was filtered
through a 20-kDa PES membrane without stirring with the pur-
poseofcollectinganintactlayeroffoulants.Themembranewas
dried in a desiccator over 5 days before ATR-FTIR analysis.
The membrane samples were cut into about 5 by 5 mm-sized
squares and were pressed against the surface of zinc selenide
(ZnSe) focusing element of an ATR accessory (Golden gate sin-
gle reflection diamond ATR series ML II, Specac, Woodstock,
Georgia). For each membrane sample, an averaged spectrum
was obtained from a series of 32 scans.
2.2. Results
2.2.1. Lake water filtration
Water samples from Lake Michigan and Lake Decatur were
also studied to simulate real drinking water supplies. Water flux
changes and rejection of organic and inorganic constituents in
the feed were monitored. The lake waters were filtered through
20-kDa PES membranes, which have been used in our lab
for many studies of adsorptive membrane fouling [21,76,77].
Organic and inorganic constituents of feed and permeate waters
weremeasured,andthemembranesurfaceandfouledlayerwere
analyzed. The analytical results were applied to formulate the
SRNOM feed solutions when the concentrations of NOM and
metal ions were set up.
Flux during filtration of Lake Michigan and Lake Decatur
waters decreased by 37% and 48%, respectively, compared to
initial flux values (Fig. 1). DOC rejection was low in both cases,
20%forLakeMichiganand13%forLakeDecatur(Table1).The
standardizedrinsingshowedlimitedfluxrecoveryataround60%
of the initial fluxes (Fig. 1). These results were typical of pre-
vious work with the same sources [76]. Twenty-nine inorganic
elements in the feed and permeate were analyzed with ICP-MS.
4.4% and 2.7% of calcium in Michigan Lake and Lake Decatur
waters were rejected by the membrane filtration. Magnesium
showednosignificantretention,whichwasincontradictionwith
thefollowingEDSanalysis(Table1).AdsorptionofCa2+onthe
polymer membrane surface has been previously observed [44].

Page 4

W.-Y. Ahn et al. / Journal of Membrane Science 309 (2008) 128–140
131
Fig. 1. Permeate fluxes of (a) two Midwestern lake waters and (b) SRNOM
solution with three metal ions.
The slight increase of sodium in the permeate water seemed to
occur when Ca2+replaced Na+that had been captured in the
polymer matrix via an ion-exchange type interaction. For Lake
Decatur water, 5.8mg/L of Si was detected in the feed and 13%
of Si was rejected by the membrane. Particles retained on the
glass-filter during the pre-filtration were analyzed with SEM,
which identified clay particles and silicon-rich unicellular algae
(diatoms). NOM and clay mineral debris are often considered
to cause membrane fouling [76].
Chemicalelementsandfunctionalgroupsofthefouledlayers
were analyzed with EDS and ATR-FTIR. As described in Sec-
tion 2.1, to overcome detection limit problems, fouling layers
were collected after filtering 1-L water samples through 20-kDa
PES membranes. EDS analysis detected calcium and silicon in
the fouling layers (Fig. 2). The detection of calcium and silicon
in the fouled layers, and the rejection of these elements by the
membrane, indicated that the two elements acted as membrane
foulants. Strong peaks of carbon, oxygen, and sulfur resulted
fromthePESmembraneitself.(Goldandpalladium,whichwere
used to coat the specimen, are also visible.) Silicon and calcium
peaks were detected in the fouling layer of Lake Decatur water
and a small calcium peak was detected in the Lake Michigan
foulant. Therefore EDS and solution analyses show that NOM,
silicon, calcium and magnesium were involved in the fouling
process.
ATR-FTIR analysis expanded the picture of the fouling lay-
ers. The filtration of lake waters commonly resulted in the
addition of carboxyl or amide groups to the clean membrane
(Fig. 3). Corresponding peaks were detected at wavenumbers of
1540 and 1650cm−1. Deprotonated carboxyl groups have two
modesofvibration.Peaksarisingfromasymmetricstretchingof
(C O) O appear between 1540 and 1650cm−1, while peaks
fromsymmetricstretchingappearbetween1200and1300cm−1
[78]. In addition, stretching of
and bending of C O H group between 1200 and 1300cm−1
suggest the presence of protonated carboxyl groups [78]. Due
to the similar structure of amide and deprotonated carboxyl
groups, vibration frequencies are similar as well. Primary and
secondary amide peaks are located at 1580 and 1650cm−1,
respectively [27,79]. Other common peaks were detected at
1130 and 1140cm−1for the filtration of lakes waters, which
was assigned to polysaccharide-like material [79]. The silicon
oxide bond (i.e., Si O) of clay minerals may also be apparent
at 1080cm−1[76,80].
According to the filtrations and analyses, NOM, calcium,
magnesium, and silicon seemed to be significant players in the
fouling process. We formulated SRNOM feed solutions so the
sum of individual metal ion concentrations yielded the same
ionic strength as the Lake Michigan water. In this work, we
excluded silicon in the model NOM filtration experiments and
the molecular modeling in order to focus on the metal-NOM
complexation.
C O group near 1700cm−1
2.2.2. SRNOM filtration with metal ions
Model metal-NOM solutions were prepared with SRNOM.
The SRNOM solutions containing three different metal ions
(Na+, Mg2+, or Ca2+) were filtered through 20-kDa PES
membranes in order to experimentally evaluate the degree of
Table 1
Organic and inorganic contents of Michigan Lake and Lake Decatur watersa
UnitSRNOM solution Lake MichiganLake Michigan (20kDa PES permeate)Lake Decatur Lake Decatur (20kDa PES permeate)
DOC
Fe
Ca
Mg
Na
Al
Si
S
mgC/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
2.43 (0.60 permeate)
<0.03
<0.18
<0.05
0.56
<0.13
<0.11
0.21
1.99
<0.03
36.50
12.35
7.08
<0.13
0.90
7.86
1.60
<0.03
34.89
11.87
7.20
<0.13
0.69
7.82
4.04
<0.03
48.02
18.22
4.33
0.03
5.85
5.60
3.52
<0.03
46.72
18.79
5.09
<0.13
5.07
5.40
aMetal analyses were conducted at the Illinois Water Survey’s (ISWS) Analytical Chemistry and Technology Unit. According to the CCV values made from the
same source as that used for calibration, 1.7–3.1% error was observed for a Ca standard measurement. Data presented in this table are averaged values of three
measurements.

Page 5

132
W.-Y. Ahn et al. / Journal of Membrane Science 309 (2008) 128–140
Fig. 2. EDS spectra of (a) clean membrane, (b) membrane fouled by pre-filtered Lake Michigan water, (c) membrane fouled by pre-filtered Lake Decatur water, and
(d) membrane fouled by pre-filtered SRNOM solution where x-axis corresponds to X-ray energy (keV) and y-axis corresponds to counts.
membrane fouling. The Ca2+-NOM solution showed the most
severemembranefoulingwiththenominalfluxdecreaseexhibit-
ingthefoulingpattern(Fig.1b)verysimilartotheonesobserved
for the lake waters samples (Fig. 1a). The Mg2+-NOM solution
resulted in a noticeably weaker fouling, while the Na+-NOM
solution did not cause any fouling at all (Fig. 1b).
The ionic strength of all metal-NOM solutions was set at
2.6mM using the chloride salt of each metal in order to set
each solution at the same electrostatic double layer thickness,
while metal-free SRNOM had an ionic strength of 0.025mM.
The Ca2+concentration was set at 35mg/L following the Ca2+
Fig. 3. ATR-FTIR spectra of 20kDa PES membranes filtered with lake waters
and SRNOM solution: ATR-FTIR spectrum of clean PES membrane was digi-
tally subtracted from the spectra of fouled layers.
concentration of the Lake Michigan water, while the concentra-
tionsofMg2+andNa+were21and61mg/L,respectively.Since
all three metal-containing solutions had the same ionic strength,
the electrostatic double layer thickness (1/κ) in these solutions
was assumed to be the same, calculated to be 59˚A according to
the Debye–H¨ uckel model:
κ2=e2?
iz2
ini0
εε0kT
,
(1)
wheree=1.60219×10−19Cistheelectroncharge;k=1.38066
×10−23J/K
ε0=8.854×10−14C2/cm/J
in vacuum; and εH2O= 80 is the relative permittivity of water
at T=298K.
Different fouling patterns at the same ionic strength contrast
with our original hypothesis, and clearly indicate that the foul-
ing is not merely an ionic strength effect; rather it is suggested
that ions having different charges bind chemical moieties of the
membrane surface and/or NOM via a specific complexation.
The divalent ions may form bridges between the membrane sur-
face and NOM or simply cause aggregation of NOM molecules
in solution. Although specific humic-metal complexations by
divalent ions has been already suggested in membrane literature
[4,6,35,38], the different membrane fouling tendencies by Ca2+
and Mg2+are not yet fully understood. Better knowledge of the
dependencyofcomplexationonthedifferentioniccharges,sizes
and ionic masses is necessary. The experimental investigation
is theBoltzmann
the dielectric
constant;
permittivityis