Available at www.sciencedirect.

com
journal homepage: www. elsevier.com/locate/watres
Bacterial community composition and structure of biolms
developing on nanoltration membranes applied to
wastewater treatment
Hanan Ivnitsky
a
, Ilan Katz
a
, Dror Minz
b
, Galit Volvovic
b
, Eyal Shimoni
c
, Elina Kesselman
c
,
Raphael Semiat
d
, Carlos G. Dosoretz
a,
a
Faculty of Civil & Environmental Engineering and Grand Water Research Institute, Technion, IIT Haifa, Israel
b
Institute of Soil, Water and Environmental Sciences, ARO, The Volcani Center, Bet-Dagan, Israel
c
Faculty of Biotechnology and Food Engineering, Technion, IIT Haifa, Israel
d
Faculty of Chemical Engineering and Grand Water Research Institute, Technion, IIT Haifa, Israel
a r t i c l e i n f o
Article history:
Received 1 February 2007
Received in revised form
8 May 2007
Accepted 11 May 2007
Available online 23 May 2007
Keywords:
Biofouling
Efuents desalination
Nanoltration
Biolm
DGGE
AFM
a b s t r a c t
The structure and microbial communities of biolms developing on cross-ow nanoltra-
tion (NF) membranes at different temperatures (20, 25 or 34 1C) and operation lengths
(8 h24days) were studied. Feedwater comprised tertiary quality wastewater efuent
or synthetic media mimicking efuents of intermediate quality. After each run, the
membranes were autopsied for bacterial enumeration, bacterial community composition
and microscopy visualization (SEM, CLSM and AFM/NSOM). Community composition
was analyzed by polymerase chain reactiondenaturing gradient gel electrophoresis
(PCRDGGE) coupled with sequence analysis of 16S rRNA gene fragments from dominant
bands.
Deposition of polysaccharides and initial bacterial colonization were observed within
8h, whereas developed biolms markedly affecting membrane permeability were evident
from days 23 onwards. Regardless of applied conditions, the heterotrophic plate counts in
the biolm were 34 10
6
CFU/cm
2
and the thickness of the biofouling layer was 2030 mm.
From a total of 22 sequences obtained from 14 independent experiments, most species
identied were Gram negative (19 of 22 sequences). Proteobacteria were found to be a
prevalent group in all cases (16 of 22 sequences) and among it, the b-subclass was the most
predominant (8 sequences), followed by the g-subclass (5 sequences). Pseudomonas/
Burkholderia, Ralstonia, Bacteroidetes and Sphingomonas were the dominant groups found in
most cases. Even though the microbial population might be important with respect to
biofouling patterns, membrane permeability decline seems to be more substantially
inuenced by the formation and accumulation of exopolymeric substances (EPS).
& 2007 Elsevier Ltd. All rights reserved.
1. Introduction
Membrane lifetime and permeate uxes in cross-ow pres-
sure-driven membrane separation processes are primarily
affected by the phenomena of concentrationpolarization
(i.e., accumulation of rejected species at the membrane
surface) and fouling (e.g., solute adhesion, microbial adhe-
sion) at the membrane surface (Gwon et al., 2003; Van der
Bruggen et al., 2003; Goosen et al., 2004). Biolm formation or
biofouling, i.e., microbial adhesion and growth, is one of the
ARTICLE IN PRESS
0043-1354/$ - see front matter & 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.watres.2007.05.021

Corresponding author. Tel.: +972 48294962; fax: +972 4 8228898.

E-mail address: [email protected] (C.G. Dosoretz).
WAT E R R E S E A R C H 41 ( 2007) 3924 3935
most complex fouling phenomena that causes severe perfor-
mance loss, modies membrane surface properties and
demands costly periodic cleaning or membrane replacement
(Ridgway et al., 1983; Ridgway and Flemming, 1996; Flemming
et al., 1997).
Cross-ow membrane separation systems offer a differen-
tial environment for biolm development. Indeed, the surface
for attachment is a selective porous substratum that allows
the transfer of certain molecules, whereas the concentration
polarization layer generates an environment of relatively high
nutrient and salt concentrations. Additionally, the permeate
ux generates a convective transport of both bacteria and
nutrients toward the membrane surface and acts as an
additional strong force assisting the cells to penetrate the
hydrodynamic layer (Christensen and Characklis, 1990; Flem-
ming, 2002; Goosen et al., 2004; Kim et al., 2006). Thus, in
biolms formed on membranes, motility seems to be
circumvented by the convective drag force perpendicular to
the membrane, which may affect the biolm population and
dynamics. Three out of four bacterial isolates retrieved from a
reverse osmosis (RO) membrane treating potable water
(Dermacoccus RO12, Microbacterium RO18 and Rhodopseudomo-
nas RO3) were reported to not exhibit swimming motility,
suggesting that they could be transported to the membrane
surface by other mechanisms such as convective permeate
ow (Pang et al., 2005).
Not all membranes undergo biofouling at the same rate,
due to differences in polymer composition, membrane sur-
face properties (e.g., hydrophobicity, roughness, pore size and
geometry, charge density, etc.) and chemical composition of
the feedwater (Knoell et al., 1999; Schafer et al., 2005). In most
cases, biofouling is caused by heterotrophic organisms that
convert dissolved organic material into biomass locally
(Ridgway and Flemming, 1996). The composition of fouling
biolms in some cases has been dominated by the endogen-
ous ora, which can differ profoundly with different fouling
sites and conditions, including responses to the effect of
biocides (Baker and Dudley, 1998; Kang et al., 2004).
Pseudomonas and Corynebacterium, and to a lower extent
Bacillus, Arthrobacter, Flavobacterium and Aeromonas, were the
predominant bacteria isolated from biolms formed on RO
membranes applied for drinking water treatment (Baker and
Dudley, 1998). Mycobacterium and Lactobacillus were also
reported as predominant isolated bacteria (Baker and Dudley,
1998; Knoell et al., 1999; Schafer et al., 2005). A study on RO
membranes applied for efuents desalination reported Acine-
tobacter, FlavobacteriumMoraxella and Pseudomonas alcaligenes
as the major groups isolated (Ridgway et al., 1983).
More recently, Chen et al. (2004) identied a-Proteobacteria
as the largest microbial fraction, followed by the g-Proteo-
bacteria, in microltration (MF) and RO membranes applied
for water purication, analyzed by 16S rRNA gene clone
library and uorescence in situ hybridization (FISH). Bradyrhi-
zobium, Rhodopseudomonas, Sphingomonas, and Bacillus/Clostri-
dium were the most predominant bacterial isolates in RO.
The microbial community structures between the MF and
RO were considerably different and only shared two common
a-Proteobacteria groups, Bradyrhizobium and Bosea, out of 17
isolates. Horsch et al. (2005) found g-Proteobacteria as the
predominant species at the earlier stages of biofouling,
whereas a- and b-Proteobacteria were the dominant species
of mature biolms formed on nanoltration (NF) and ultra-
ltration (UF) membranes applying feedwater from an
oligotrophic water reservoir, also applying FISH.
In a previous study, we reported on the biofouling layer
characteristics for NF of MF-preltered secondary efuents as
well as synthetic media inoculated with a Flavobacterium sp.
isolated from biofouled membranes (Ivnitsky et al., 2005).
Three dominant microbial populations sequenced upon
polymerase chain reactiondenaturing gradient gel electro-
phoresis (PCRDGGE) analysis of the 16S rRNA gene fragment
were identied as Pseudomonas, Ralstonia and Cytophaga.
The aim of this work was to investigate the potential for
biolm formation and bacterial communities composition on
the surface of NF membranes subjected to partial efuent
desalination. To trace the changes occurring in the biofouling
layer, community composition was assessed by PCRDGGE
and sequence analysis of 16S rRNA gene fragments from
DGGE bands, as a function of temperature, length of opera-
tion and feedwater source.
2. Materials and methods
2.1. Experimental setup and conditions
Biolm development was studied in a cross-ow apparatus
equipped with a MIC-RO 240 module (PCI) comprising one
NF-polyamide tubular membrane (AFC 30, PCI) with a
ltration area of 120cm
2
(30cm length by 1.25cm diameter)
and 200Da MWCO, as described previously (Ivnitsky et al.,
2005). All runs were conducted in semibatch mode with full
recirculation of both permeate and concentrate into the
feed tank, whose content was replaced daily. Pressure
was maintained at 5 bar and recirculation ow-rate at
132.5L/h (corresponding to a linear velocity of 0.3m/s
and Re of 5300) in all cases. The temperature was maintained
at 20, 25 or 34 1C, and the duration of the experiments
varied from 8h to 24 days, as indicated in Table 1. Feedwater
consisted of either tertiary quality wastewater efuents
or synthetic media mimicking efuents of intermediate
quality according to the stipulations effective in Israel.
Before each experiment, the system was intensively
cleaned with alkaline sodium hypochlorite and profusely
rinsed with tap water and a new membrane was installed.
Following each run, the membranes were dismounted and
evenly segmented into slices of approx. 1 cm length for the
different microbiological and microscopic analyses, as de-
scribed thereafter. A total of approx. 40 independent runs
were performed.
2.2. Feedwater
Tertiary quality efuents were generated in a continuous
membrane bioreactor (MBR) of 1.2m
3
capacity, 8 m
2
ltration
area and 0.4 mm pore size-at sheets MF membranes (Kubotta,
Japan), operating at the Technion campus. The MBR was fed
with primary efuent collected upon sedimentation of raw
wastewater from the Technion campus sewage collection
system. MBR efuents were free of suspended solids
ARTICLE IN PRESS
WATER RESEARCH 41 ( 2007) 3924 3935 3925
(o1NTU) and characteristic organic matter content was
(average7standard deviation, in mg/L): total organic carbon
(TOC) (12.774.0), BOD (o0.370.6) and COD (28.379.0). In-
dividual TOC values of each independent run are presented in
Table 1.
Synthetic efuents were free of suspended solids and
contained per liter of distilled water (in mg): NaCl, 300; CaCl
2
,
50; MgSO
4
, 50; NH
4
Cl, 1.4; H
2
NaPO
4
, 0.29 and an organic
substrate to meet a theoretical COD concentration of 20 mg/L
at a COD:N:P ratio of 100:5:1, which is known to be optimal for
bacterial growth. The sources of organic carbon supplemen-
ted were glucose or starch, as indicated in Table 1.
2.3. Chemical analysis
Electrical conductivity (EC) and pH were measured using a
Cyborscan (Eutech Instruments, Nijkerk, Netherlands). Tur-
bidity was measured with a 2100P turbidimeter (Hach Co.,
Loveland, CO). TOC was measured with an N/C 2100 multi-
analyzer (Analytik Jena AG, Jena, Germany).
2.4. Bacterial community analysis
2.4.1. DNA extraction and PCR
The tubular membrane slices were subsectioned into pieces
of approx. 1 cm
2
and subjected to DNA extraction by a
modied bead-beating method described previously (Oved
et al., 2001). The extracted DNA was stored at 201C until
further processed. PCR amplications for DGGE were per-
formed with the previously described bacterial primer pair
341F, with a GC clamp at its 5
0
end, and 907R (Muyzer et al.,
1998) using a T-gradient thermocycler (Biometra, Goettingen,
Germany) as described previously (Cytryn et al., 2006).
2.4.2. DGGE analysis
DGGE of PCR-amplied 16S rRNA gene fragments was
performed using a D-gene system (BioRad, Hercules, CA), as
described previously (Muyzer et al., 1998). Prevalent DGGE
bands were carefully excised on a Dark Reader Transillumi-
nator (Clare Chemical Research, Inc., Dolores, CO) and
puried as described previously (Ivnitsky et al., 2005; Cytryn
ARTICLE IN PRESS
Table 1 Details of membranes and recirculating uid samples for PCRDGGE analysis and bacterial plate counts
Sample
number
a
Sample
type
b
Organic carbon
amendment
Temperature
(1C)
Run
duration
(days)
c
TOC
o
d
(mg/l)
Permeability
decline (%)
HPC
e
(CFU/
cm
2
)
(CFU/
ml)
1 SyE[M] Glucose 34
16
3.3 60 1.4 10
7
3
*
SyE[F] Glucose 34 3.3 na 6.3 10
7
4 TeE[M] None 34
12
10.3 71 5.0 10
5
8 TeE[F] None 34 10.3 na 1.6 10
5
12
*
SyE[F] Starch 34 6 na 4.9 10
6
15 SyE[M] Starch 34 8 8.3 20 8.4 10
5
19 SyE[M] Glucose 34 8 3.6 25 3.4 10
6
21 SyE[M] Glucose 34 9.7 65 3.4 10
6
22 SyE[M] Glucose 34 7 9.7 65 1.7 10
6
23 SyE[M] Glucose 34 9.7 65 2.5 10
6
26 SyE[M] Glucose 34 4.7 22 1.0 10
6
27 SyE[M] Glucose 34 24 4.7 22 6.7 10
5
28 SyE[F] Glucose 34 4.7 na 9.5 10
6
30 TeE[M] None 34 11.7 35 1.0 10
7
34 TeE[M] None 34 4 11.7 35 4.2 10
5
35 TeE[M] None 34 11.7 35
37 TeE[M] None 34 3 11.1 12 3.0 10
5
43 TeE[M] None 20
2
20.6 36 1.3 10
7
46 TeE[F] None 20 20.6 na 1 10
6
47 TeE[M] None 20
7
16.7 52 7.4 10
6
49 TeE[M] None 20 16.7 52 7.1 10
6
52 TeE[M] None 20
0.4
12.4 0 4.2 10
6
55 TeE[F] None 20 12.4 na 1.2 10
6
62 TeE[M] None 25 1 9.6 8 2.8 10
6
64 TeE[M] None 25
7
9.2 31 2.2 10
6
65 TeE[M] None 25 9.2 31 3.3 10
5
na: not applicable.
a
Numbers indicate different experiments and correspond to the lane number in the DGGE gel besides
*
samples 3 and 12 (see Fig. 1).
b
TeE: tertiary efuents, SyE: synthetic efuents, M: membrane, F: uid. Samples were taken at the end of the run.
c
Each common block represent the same run.
d
TOC
o
: Total organic carbon concentration in the feedwater.
e
HPC: heterotrophic R2A-plate count of bacteria extracted from the membrane and recirculating uid samples.
WAT E R R E S E A R C H 41 ( 2007) 3924 3935 3926
et al., 2006). Puried bands were re-amplied and veried by a
second DGGE analysis.
2.4.3. Cloning of PCR products and sequencing
The pGEM-T Easy vector System (Promega, Madison, WI) was
used to clone PCR products amplied from excised DGGE
bands (Cytryn et al., 2006). Clones were screened by the
colony PCR amplication and veried by DGGE (as described
above). Plasmids from selected colonies were puried with
the Concert Rapid Miniprep Plasmid Purication System
(Gibco BRL Products, Rockville, MD) according to the protocol
provided by the manufacturer. Clones were sequenced using
the Applied Biosystems PRISM Dye Terminator Cycle Sequen-
cing Ready Reaction kit with Ampli Taq DNA polymerase
and the T7 primer as suggested by the manufacturer. The
sequencing products were analyzed with an Applied Biosys-
tems 377 DNA sequencer.
2.4.4. Phylogenetic analysis
Five hundred and thirty-base pair nucleotide sequences were
incorporated into a pre-aligned database of 16S rRNA
sequences, using the aligning tool supplied by the ARB
phylogenetic program package (Ludwig et al., 2004). Phyloge-
netic trees were generated with neighbor-joining and max-
imum likelihood methods provided with the ARB program
package using the Felsenstein correction method and apply-
ing a 50% cutoff lter to remove highly variable positions. The
topologies of the resulting trees were identical for both
methods.
2.4.5. Bacterial enumeration
Subsectioned membrane samples were shaken (30min,
150rpm) with glass beads (1 mm in diameter) in phosphate-
buffered saline (PBS) for biofouling layer extraction. The
suspension was subsequently serially diluted and plated.
Feedwater samples were serially diluted in PBS and plated.
Enumeration was performed by heterotrophic plate count
(HPC) on R2A media; attached bacteria were expressed in
CFU/cm
2
and suspended bacteria in CFU/ml. Plates were
incubated at 30 1C for 3 days and colonies were counted.
2.5. Microscopy
2.5.1. Scanning electron microscopy (SEM)
Membrane samples for SEM were subsectioned and xed with
glutardialdehyde (5% in 50mM sodium cacodilate). Upon
xation, samples were incubated in the dark in osmium
tetraoxide solution (1% in 50 mM sodium cacodiolite), fol-
lowed by a six-step dehydration with an ethanol/water
gradient (25100%) and nally three times immersion in pure
hexamethyldisilazane (HMDS), all at 4 1C. Following air-drying
in a fume hood until full HMDS evaporation, the dehydrated
specimens were further dried in a desiccator at 251C for 24h
and gold-coated in a sputtering device (Polaron SC515, Fisons
Instruments, UK). Digital images were obtained using SEM
(JSM-5400; Jeol, Japan) combined with the appropriate soft-
ware (Voyager II, Moran, Netherlands). All chemicals em-
ployed were of electron microscopy quality (Sigma Chemical,
St. Louis, MO).
2.5.2. Atomic force microscopy (AFM)
Membrane sample subsections (5 5mm) were mounted into
the liquid cell and scanned immediately. Scans were
performed in PBS in the Nanonics AFM/NSOM-100 system
(Nanonics Imaging Ltd., Jerusalem, Israel), using cantilevered
optic ber AFM-probe, with a 20nm diameter tip. Scanning
was performed at tapping mode, 12delay/ms, 50 scale lines,
150 frames, 4 substeps and 4 prescans, giving an overall scan
time of approx. 26 min.
2.5.3. Near-eld scanning optical microscopyAFM
(AFMNSOM)
Mounting and scanning was performed as described above.
Near-eld illumination of surfaces was performed using an
argon ion laser (Laser Physics Inc., West Jordan, UT). Laser
light was delivered to the surface through a cantilevered
AFMNSOM optical ber probe, coated with CrAl, possessing
an aperture diameter of 50100nm. Fluorescent staining was
performed with concanavalin A (ConA)FITC (Sigma) to target
the biolm polysaccharide fraction of exopolymeric sub-
stances (EPSs) as described previously (Johnsen et al., 2000).
The emitted uorescence signal (excitation wavelength of
488nm and an emission band lter of 515710 nm) was
collected by a photon counting module (Perkin Elmer
Optoelectronics, Vaudreuil, Canada), through the 50 objec-
tive lens of built-in Olympus-BXFM dual microscope (Olym-
pus Optical Ltd., Tokyo, Japan). Images were processed using
Quartz software (Cavendish Instruments Ltd., Shefeld, UK).
2.5.4. Confocal laser scanning microscopy (CLSM)
Confocal microscopy images of double-stained membrane
sample subsections were captured using a MRC 1024 CLSM
(Bio-Rad, Hempstead, UK) equipped with a Nikon Plan Apo
601.40 objective, detectors and lter sets for monitoring the
uorescent staining applied. Double staining was performed
with propidium iodide (PI) for nucleic acid staining (Sigma),
followed by ConAFITC for polysaccharide staining. Images
were processed using the BioRad confocal assistant software
(version 4.02). Biolm thickness was determined by multi-
plying the number of slices taken by the thickness of each
slice (23mm).
3. Results
3.1. Identication of the bacterial community and
characteristics of the biofouling layer
3.1.1. PCRDGGE bacterial community analysis and
sequencing
Membranes and recirculation uid from 14 independent
experiments, six runs with synthetic efuents and eight with
tertiary efuents (Table 1) were analyzed for the bacterial
community composition by PCRDGGE (Fig. 1). Most of the
dominant bands were successfully excised from the gels,
reamplied, cloned and sequenced. The sequences were
compared to published 16S rRNA sequences and a phyloge-
netic tree was constructed (Fig. 2). In parallel, the samples
were analyzed for HPC (see Fig. 3).
ARTICLE IN PRESS
WATER RESEARCH 41 ( 2007) 3924 3935 3927
From a total of 22 sequences obtained from 14 different
independent experiments conducted, most species identied
were Gram negative (19 of 22 sequences). Proteobacteria were
found to be a prevalent group in all cases (16 of 22 sequences)
and among it, the b-subdivision was the most predominant
(8 sequences), followed by the g-subdivision (5 sequences).
Pseudomonas/Burkholderia, Ralstonia, Bacteroidetes and Sphingo-
monas, were the dominant groups found in most cases.
General DGGE banding patterns and distribution of domi-
nant bands obtained from identical runs were more similar in
biolms developed in experiments involving synthetic rather
than tertiary efuents (see Fig. 1). This behavior may be
attributed to the presence of diverse sources of organic
materials and nutrients in wastewater efuents compared
with the relatively stable and simple composition of the
synthetic efuents. However, no dened effect of run dura-
tion or temperature on DGGE banding patterns can be
delineated.
The microbial populations identied by sequence homology
in the experiment involving synthetic efuents corresponded
phylogenetically to Ralstonia (bands 5, 8, 9), Bacteroidetes
(bands 1, 11), Sphingomonas (bands 2, 3), Dyella (band 4),
Microbacterium (band 10), uncultured Myxococcus and Cysto-
bacter (band 6) and Burkholderia (band 7). Microbial popula-
tions identied in runs with tertiary efuents corresponded
phylogenetically to Flavobacterium sp. (band 22), Pseudomonas
(bands 15, 18, 19), Legionella-like (band 12), Ralstonia sp. (band
13), Hydrogenophaga (band 14), Acidovorax (band 17), Bacillus
(band 16), Brevibacillus (band 20) and Delftia (band 21).
The duration of the experiments did not directly determine
the type of populations found on the membranes. The
microbial populations that prevailed in the recirculation uid
were not always found in the biolm. In some cases, however,
the bacterial populations found in the recirculation uid
became dominant also on membranes. These included
Bacteroidetes (see band 11 in lanes 1528, Fig. 1) and
Burkholderia (see band 7 in lanes 1928, Fig. 1) in synthetic
efuents, and Pseudomonas (see band 15 in lanes 4649, Fig. 1)
and Ralstonia (see band 13 in lanes 3046, Fig. 1) in tertiary
efuents. However, not all the dominant populations on the
membranes could be found in the recirculation uid during
sampling.
3.1.2. Bacterial plate counts
Membranes and recirculation uid were analyzed for hetero-
trophic plate counts (Table 1 and Fig. 3). Regardless of
temperature, run duration or feedwater type, the HPC in all
biolm samples rendered very similar numbers, in the order
of 10
6
CFU/cm
2
of membrane. For the synthetic efuents, the
average was 3.4 10
6
CFU/cm
2
, ranging from 6.7 10
5
to
1.4 10
7
CFU/cm
2
(median 2.1 10
6
). For the tertiary efuents,
the average was 4.4 10
6
CFU/cm
2
, ranging from 3.0 10
5
to
2.8 10
7
CFU/cm
2
(median 1.3 10
7
). HPCs in the recirculated
uid were in the range of 1.610
5
1.210
6
CFU/ml for the
tertiary efuents, and 4.910
6
6.310
7
CFU/ml for the syn-
thetic efuents. The higher counts in the recirculation uid of
the synthetic efuents were probably due to a higher
degradable organic matter, and therefore faster growth,
compared with tertiary efuents. Similar HPC values were
obtained in experiments performed at a pressure of 15 and
25 bar, or when culturing the samples on LB agar medium
(data not shown). Independent of pressure, time or tempera-
ture, the development of a biofouling layer to a convergent
countable magnitude was in the order of 10
6
CFU/cm
2
, which
seems to be the result of a steady-state nutrient supply to the
biolm and the shear forces developed on the membrane
surface. Still, a HPC increase of approx. one order of
magnitude was noticeable for tertiary efuents, with an
increase of the initial organic carbon level within the range
tested (920mgTOC/l) (see Table 1). No effect of the organic
carbon pool on HPC was found for synthetic efuents in the
range studied (310mgTOC/l). This agrees with the acknowl-
edged conception that the quality of the feedwater is a
decisive factor in biofouling development.
3.1.3. Estimation of the biofouling layer thickness by CLSM
CLSM was applied for the thickness estimation of mature
biolms formed on the membranes during NF of tertiary
efuents (double stained with PI and ConAFITC). The series
of CLSM micrographs shown in Fig. 4 correspond to the
subsequent optical sections of increasing thickness of a
typical biolm developed on the membranes after 7 days of
operation. Scanning into the depth of the biofouling layer
shows that the optical section at 20mm is near but above the
membrane surface, as evidenced by the pores seen in the
ARTICLE IN PRESS
Fig. 1 PCRDGGE ngerprints of 16S rRNA gene fragments amplied from DNA extracted from different membrane and
recirculated uid samples at the conclusion of each run. For lane numbers refer to Table 1.
WAT E R R E S E A R C H 41 ( 2007) 3924 3935 3928
ARTICLE IN PRESS
Fig. 2 Phylogenetic tree based on 530bp 16S rRNA gene sequences. The topology of the dendogram was generated using the
neighbor-joining method included in the ARB phylogenetic package. Similar branching was obtained using maximum
likelihood (see Section 2). Sequences were obtained from bands excised from the PCRDGGE gel shown in Fig. 1. The bar
indicates an estimated 10% sequence divergence.
WATER RESEARCH 41 ( 2007) 3924 3935 3929
background of the uorescent eld. Since the optical section
at 30 mm is already out of focus of the membrane surface, as
evidenced by the lack of uorescence, it can be concluded
that the thickness of the biolm is 2030mm. Similar biolm
thickness were observed under the different conditions
applied (data not shown), again suggesting the development
of a biofouling layer to a relatively convergent magnitude
under a broad range of conditions.
3.2. Study of biolm development on NF membranes
Characterizing the development of the biofouling layer on the
membrane surface as a function of the operating time during
nanoltration of tertiary efuents was performed by AFM, for
visualization of the early stages of the process, and by SEM,
for visualization of fully developed biolms. While SEM
provides imaging of previously xed and dried samples,
AFM allows investigation of the membrane surface in its
swollen state. Quantitative evaluation of the membrane
performance was assessed by permeability rejection proles.
3.2.1. AFM and NSOMthe earlier stages
The earlier stages of biolm development on the membranes,
i.e., rst 16 h of operation, were followed by AFM and NSOM
(Figs. 5 and 6). As a reference to the fouled membrane, a virgin
membrane, swollen in PBS, showed topography with uniform
surface features and relatively low height range (300nm)
(Fig. 5, left panel). Upon 8h of operation, sporadic features,
roughly 12 mm in length by 900nm height, which are typical
of microbial cells on surfaces, were evident (Fig. 5, middle
panel). These features seem to be singular microbial cells,
thus suggesting that initial colonization was done by singular
cells, as clearly shown in the 3D image (see Fig. 6, top, left
panel). They corresponded to the sporadic appearance
of microbial cells at this stage, imaged by SEM (data not
shown). It should be mentioned that microorganisms imaged
in their native environment (solution) by AFM have a swollen
ARTICLE IN PRESS
SyE(M) TeE(M) TeE(F)
1.010
5
1.010
6
1.010
7
1.010
5
1.010
6
1.010
7
C
e
l
l

c
o
u
n
t
s

(
C
F
U
/
c
m
2
)
C
e
l
l

c
o
u
n
t
s

(
C
F
U
/
m
l
)
SyE(F)
Fig. 3 Summary of heterotrophic plate counts on R2A of bacteria extracted from different membranes and recirculated uid
samples at the conclusion of each run. Bars indicate standard deviation. For details see Table 1.
0 m 10 m 5 m 8 m
20 m 15 m 30 m 12 m
Fig. 4 CLSM micrographs of a typical 7-day-old membrane for the estimation of biolm thickness. Panels from left to right
represent CLSM images taken from the outside layer of the biolm toward the membrane surface. Scanning depth is at the
top of each image. Red color is propidium iodide staining of bacterial nucleic acids and green color is ConAFITC staining of
EPS polysaccharides. Magnication 60, resolution 10821082lm.
WAT E R R E S E A R C H 41 ( 2007) 3924 3935 3930
extracellular matrix, hence revealing larger dimensions than
in xed and dried SEM samples (Pang et al., 2005). Surface
scanning of a membrane 16h after initial ltration exhibited a
larger number of individual microbial cell features as well as
microbial cluster sizes per similar area, indicating the
formation of microcolonies (Fig. 5, right panel). These
changes can also reect the possible changes in the commu-
nity composition and the recruitment of secondary micro-
organisms, facilitated by the convective motion toward the
membranes.
Simultaneous NSOM analysis of the same membranes was
performed to study the specic distribution of EPS, visualized
by uorescent staining of polysaccharides with ConAFITC
(Fig. 5, bottom panel). The development of an EPS layer on the
surface of the membrane with time is evident and although
polysaccharides are unevenly distributed over the surface,
they cover most of the space scanned. This is in line with the
CLSM images shown for a mature biolm, depicting enhanced
accumulation of polysaccharides, i.e., EPS, on the membrane
at earlier stages, similar to other literature reports (Flemming
et al., 1997; Park et al., 2005; Kim et al., 2006). A typical 3D
image of the membrane after 8h of ltration showing the
AFM topography (left) and NSOM image (right) of the
membrane is shown in Fig. 6, top panel. The prole of
uorescence emitted by ConAFITC bound to the membrane
as a function of ltration time, as calculated from NSOM
signals, is shown in Fig. 6, bottom panel, displaying that the
attachment of polysaccharides is already evident after 2h.
The earlier detection of dened microbial structures and EPS
on the surface of the membranes suggests a facilitated
colonization, probably due to the permeate drag component.
3.2.2. SEM and membrane performancethe mature biolm
The development of the biofouling layer with time during NF
of tertiary efuents was assisted by SEM of autopsied
membranes as well. Typical SEM micrographs gathered at
different operating times, starting at a virgin membrane, are
presented in Fig. 7. After less than 1 day, initial colonization
was evidenced by sporadic spots on the membrane area, in
line with the AFM data (data not shown). After 3 days, the
membrane was already covered by a mature, but still loose,
biolm, characterized by a predominance of rod-shaped cells
and an evident EPS matrix embedding the cells (Fig. 7 top,
center panel). After 7 days, a completely developed and tight
biolm composed of a diversied number of microorganisms
over multiple layers of cells, debris/rejected material and EPS
matrix, was observed (Fig. 7 top, right panel). After 12 days, a
very heterogeneous and compacted biofouling layer could be
seen in both top and side views (Fig. 7, bottom panel).
The qualitative SEM data correlated to a great extent with
membrane performance, quantitatively evaluated by the time
proles of permeability and rejection of both soluble organic
matter (as TOC) and inorganic ions (as EC), as presented in
Fig. 8. A relatively steady operation is seen during the rst 23
ARTICLE IN PRESS
Fig. 5 Representative AFM micrographs depicting polysaccharide attachment and biolm proliferation on the membrane
surface as a function of ltration time. Top: AFM topography scans; bottom: NSOM scans of the uorescence emitted by
ConAFITC bound to polysaccharides on the membrane. Left panels: virgin membrane (time 0); middle panels: 8h; right
panels: 16h.
WATER RESEARCH 41 ( 2007) 3924 3935 3931
ARTICLE IN PRESS
7 days
12 days
3 days Virgin membrane
Fig. 7 SEM micrographs of biolm developed on the membranes during nanoltration of tertiary efuents as a function of
operating time. The rightmost bottom image shows a side view of a specimen, denoting the membrane matrix. All other
images are top views.
0 2 8 16
0.0
0.1
0.2
0.3
0.4
0.5
Time (h)
F
l
u
o
r
e
s
c
e
n
c
e

(
i
n
p
u
t
/
V
)
Height /nm Height /nm
968.1
0
X
:
2
7
.
5

m
Y
:
2
7
.
5

m
968.1
0
X
:
2
7
.
5

m
Y
:
2
7
.
5

m
Fig. 6 AFM analysis. Top: typical 3D micrographs of AFM-topography (left) and NSOM image (right) of a membrane after 8h
ltration (3D images correspond to the respective middle panels shown in Fig. 5). Bottom: prole of uorescence emitted by
ConAFITC bound to the membrane as a function of ltration time, depicting polysaccharides buildup. Bars represent
standard deviation.
WAT E R R E S E A R C H 41 ( 2007) 3924 3935 3932
days, corresponding to the stage of biolm not yet affecting
membrane performance (Fig. 8, left panel). Then, a gradual
decrease in the permeate ux, with a subsequent reduction of
permeability of approx. 60% after 7 days and more than 90%
after 12 days of operation, was evident. This decrease
corresponded to the formation of a fully developed and tight
biofouling/deposition layer visualized by SEM.
In contrast with the marked decrease in permeability, only
a slight deterioration of the membrane selectivity was
observed during the 12-day period (Fig. 8, right panel), in
which the rejection of the inorganic ions decreased by approx.
10% (from 68% to 61%) and that of the organic matter by
17% (from 72% to 60%). These results exemplify the great
capability of membrane separation for improving the quality
of wastewater efuents, in which an almost constant product
quality can be achieved regardless of the production yield.
Similar time proles were obtained in almost all the 40
independent runs performed, and moreover, in some cases
the rejection was even improved upon fouling (data not
shown).
The ux decline behavior is similar to other reports (Gwon
et al., 2003; Park et al., 2005; Schafer et al., 2005). However, ux
(and permeability) in the present work was allowed to decline
more drastically in a short period of time with the purpose of
studying the evolvement of fully developed biolms under
dened hydrodynamic and feedwater conditions. Rejection
data are in full agreement with previous reports on NF of
wastewater efuents, in particular for organic matter rejec-
tion, which is in the order of 6580% TOC (Van der Bruggen
et al., 2003).
Analysis of the permeability data for the entire set of
conditions in both types of feedwaters presents an average
decline of 2% per mg TOC/l with an increase in feedwater TOC
concentration (see Table 1). Regarding the inuence of
run duration on permeability, a minor effect was seen on
synthetic efuent (0.5% decline per day) compared with a
more pronounced effect on tertiary efuent (5% decline per
day). These results are in general agreement with the HPC,
AFM and SEM data and the apparently greater population
diversity seen in tertiary efuent samples, and reconrm that
feedwater quality is perhaps the most determinant factor in
membrane biofouling.
4. Discussion
The structure and microbial communities of biolms devel-
oping on cross-ow NF membranes for efuent desalination
at different temperatures, lengths of operation and feedwater
composition were studied by microscopic and PCRDGGE
analyses. Deposition of polysaccharides and the beginning of
bacterial colonization were observed after 8 h, whereas
developed biolms affecting membrane permeability were
evident from days 23 onwards. HPC in the biolm was
3410
6
CFU/cm
2
and the thickness of biofouling layer was
2030mm, regardless of applied conditions. Most species
identied corresponded to Proteobacteria and only a small
fraction corresponded to Gram-positive and other Gram-
negative groups. b- and g-Proteobacteria were found to be
the most prevalent classes in all cases and Pseudomonas/
Burkholderia, Ralstonia, Bacteroidetes and Sphingomonas were the
dominant groups found in most cases.
With some minor differences, our results are in general
agreement with previous reports, regardless of the separation
technology applied, NF and RO, or the type of feedwater,
natural water and treated efuents (Ridgway et al., 1983;
Baker and Dudley, 1998; Chen et al., 2004; Horsch et al., 2005)
as they are in strict oligotrophic environments (Sarro et al.,
2005) or drinking water-network biolms (Schmeisser et al.,
2003). Horsch et al. (2005) attributed to g-Proteobacteria a
distinctive role in initial membrane colonization since they
were more predominant than b- and a-Proteobacteria in
primary biolms as compared with mature ones. According
to Chen et al. (2004), the prevalence of a-proteobacterial
populations in membrane systems is due to their tendency to
proliferate in oligotrophic ecosystems. The apparent differ-
ences in the microbial species observed between the different
studies could have been due to the differences in process
conguration and operating conditions (e.g., the pre-existing
microorganisms in the feedwater, hydrodynamics and trans-
membrane pressure) but also due to differences in the
analytical protocols, sampling procedures and extraction
methods applied.
In spite of a high diversity of microbial species found
on water systems biolms, generic patterns of bacterial
ARTICLE IN PRESS
0 3 6 9 12
0
1
2
3
4
Time (d)
P
e
r
m
e
a
b
i
l
i
t
y

(
L
/
m
2
.
h
.
b
a
r
)
0 3 6 9 12
0
25
50
75
100
EC
TOC
Time (d)
R
e
j
e
c
t
i
o
n

(
%
)
Fig. 8 Typical effect of biofouling on membrane performance during nanoltration of tertiary efuents. Left: permeability
prole; right: rejection proles of dissolved organic matter (as TOC) and inorganic ions (as EC).
WATER RESEARCH 41 ( 2007) 3924 3935 3933
colonization on membranes can be established, in which
Proteobacteria are the predominant species, with a minor
contribution of Gram positives and other Gram negatives.
This prole seems to be related to a broad ability of surface
colonization, adaptability to changing nutrients and capabil-
ity to proliferate under oligotrophic conditions of these
organisms. The dominant organisms found in our work and
in other reports show a wide metabolic diversity and tness
to the environment examined. Ralstonia and Sphingomonas are
known organisms present in most oligotrophic environments
(Kulakov et al., 2002; Chicote et al., 2004). Some strains of
Ralstonia seem to be involved in biomineralization and
chemisorption processes of metals (Sarro et al., 2005).
Burkholderia and Pseudomonas, which are ubiquitous bacteria
in soil and wastewater treatment plants, are related to the
degradation of a broad range of synthetic and natural organic
compounds, whereas Bacillus species display impressive
physiological diversity (Sarro et al., 2005). Nonetheless, the
specic role of each type of bacteria on the physiology of the
biolm as well as on membrane biofouling still remains
unclear.
The development of the biofouling layer to a convergent
countable magnitude in the order of 10
6
CFU/cm
2
and
converging thickness value of 2030mm regardless of the
conditions applied seems to be the result of a balance
between available nutrients, shear forces and pressure drop
across the membrane surface. Plate count values ranging
from 2 10
5
to 6 10
7
CFU/cm
2
were reported for biofouled RO
(Ridgway et al., 1983; Flemming et al., 1997; Baker and Dudley,
1998; Hu et al., 2005) or NF (Vrouwenvelder et al., 1998;
Ivnitsky et al., 2005; Jacquemet et al., 2005) membranes
applied for the treatment of brackish, sea or wastewater
efuents of differing qualities. Regardless of hydrodynamic
conditions, pressure drops across the membrane or for the
purpose of water or efuent purication, similar biolm
thicknesses on membranes, in the order of 1530mm, have
been reported (Ridgway et al., 1983; Baker and Dudley, 1998;
Chen et al., 2006).
The earlier detection of dened microbial structures and
EPS on the surface of the membranes by NSOM/AFM analysis
here suggests a facilitated colonization, probably due to the
permeate drag component. In the process of biolm develop-
ment, early stages of surface colonization is known to
proceed by motility-assisted locomotion and the resultant
architecture has often been associated with substrate avail-
ability (Christensen and Characklis, 1990; Stoodley et al.,
2002). The possibility of facilitated transport of bacteria and
nutrients in membrane devices may impart a different
pattern of initial colonization, and perhaps structure, in
biolms, developed on membranes (Pang et al., 2005; Eshed
et al., 2007). NSOM/AFM analysis further suggests that since
the membrane surface is rapidly changed once the biofouling
layer starts developing, cell-to-cell and cell-to-EPS interac-
tions may be more crucial to the development of the mature
biolm than their interaction with membrane itself. This
explains the convergent nature of the data found here and in
other reports.
Taken all together, our ndings strongly suggest that
even though the diversity of the microbial population and
the type and concentration of the nutrient pool are important
to the patterns of biofouling, once biofouling is initiated the
most dramatic effect on membrane permeability decline
might be due to the formation and accumulation of EPS.
The direct link between bacterial communities and mem-
brane fouling, as well as the implication of microbial diversity
on long-term application of membranes, is still unclear, and
further data should be gathered. Understanding of the
microbial community structure would provide useful insights
into the rational application of cleaning and disinfecting
protocols as well as to the development of tools for early
biofouling detection, in line with the series of concerted
actions needed to be implemented in order to hinder
biofouling (Flemming, 2002). Whether or not pathogenic
bacteria can develop in biolms on membranes also needs
to be assessed since this may be a limitation for safe
discharge of brines and sustainability of water reuse.
Acknowledgments
This work was funded by the Chief Scientist of the Ministry of
Agriculture, Tashtiot Program of the Ministry of Science &
Technology, the Grand Water Research Institute at the
Technion, and the Israel Science Foundation. We are grateful
to Y. David, J. Holzman, C. Rozenfeld and L. Kautskey for their
technical assistance.
R E F E R E N C E S
Baker, J.S., Dudley, L.Y., 1998. Biofouling in membrane systema
review. Desalination 118, 8190.
Chen, C.L., Liu, W.T., Chong, M.L., Wong, M.T., Ong, S.L., Seah, H.,
Ng, W.J., 2004. Community structure of microbial biolms
associated with membrane-based water purication processes
as revealed using a polyphasic approach. Appl. Microbiol.
Biotechnol. 63 (4), 466473.
Chen, M.Y., Lee, D.J., Yang, Z., Peng, F., Lai, J.Y., 2006. Fluorescent
staining for study of extracellular polymeric substances in
membrane biofouling layers. Environ. Sci. Technol. 40,
66426646.
Chicote, E., Moreno, D.A., Garca, A.M., Sarro , M.I., Lorenzo, P.I.,
Montero, F., 2004. Biofouling of the walls of a spent nuclear
fuel pool with radioactive ultrapure water. Biofouling 20,
3542.
Christensen, B.E., Characklis, W.G., 1990. Physical and chemical
properties of biolm. In: Characklis, W.G., Marshall, K.C. (Eds.),
Biolms. Wiley, New York, NY, pp. 93130.
Cytryn, E., Minz, D., Gieseke, A., van Rijn, J., 2006. Transient
development of lamentous Thiothrix species in a marine
sulde oxidizing, denitrifying uidized bed reactor. FEMS
Microb. Let. 256, 2229.
Eshed, L., Yaron, S., Dosoretz, C.G., 2007. The involvement of
permeate ux on biolm development on cross-ow mem-
brane ltration. In: Biolms 2007, Fourth ASM Conference on
Biolms, Book of Abstracts, Quebec City, Canada, p. 143.
Flemming, H.C., 2002. Biofouling in water systemscases, causes
and countermeasures. Appl. Microbiol. Biotechnol. 59 (6),
629640.
Flemming, H.C., Scaule, G., Griebe, T., Schmitt, J., Tamachkiarowa, A.,
1997. Biofoulingthe Achilles heel of membrane processes.
Desalination 113, 215225.
ARTICLE IN PRESS
WAT E R R E S E A R C H 41 ( 2007) 3924 3935 3934
Goosen, M.F.A., Sablani, S.S., Al-Hinai, H., Al-Obeidani, S., Al-Belushi,
R., Jackson, D., 2004. Fouling of reverse osmosis and ultraltration
membranes: a critical review. Sep. Sci. Technol. 39, 22612297.
Gwon, E., Yu, M., Oh, H., Ylee, Y., 2003. Fouling characteristics of
NF and RO operated for removal of dissolved matter from
groundwater. Water Res. 37 (12), 29892997.
Horsch, P., Goreno, A., Fuder, C., Deleage, A., Frimmel, F.H., 2005.
Biofouling of ultra- and nanoltration membranes for drinking
water treatment characterized by uorescence in situ hybri-
dization (FISH). Desalination 172, 4152.
Hu, J.Y., Song, L.F., Ong, S.L., Phua, E.T., Ng, W.J., 2005. Bioltration
pretreatment for reverse osmosis (RO) membrane in a water
reclamation system. Chemosphere 59 (1), 127133.
Ivnitsky, H., Katz, I., Minz, D., Shimoni, E., Chen, Y., Tarchitzky, J.,
Semiat, R., Dosoretz, C.G., 2005. Characterization of mem-
brane biofouling in nanoltration processes of wastewater
treatment. Desalination 185, 255268.
Jacquemet, V., Gaval, G., Rosenberger, S., Lesjean, B., Schrotter, J.C.,
2005. Towards a better characterisation and understanding of
membrane fouling in water treatment. Desalination 178 (13),
1320.
Johnsen, A.R., Hausner, M., Schnell, A., Wuertz, S., 2000. Evalua-
tion of uorescently labeled lectins for noninvasive localiza-
tion of extracellular polymeric substances in Sphingomonas
biolms. Appl. Environ. Microbiol. 66 (8), 34873491.
Kang, S.-T., Subramani, A., Hoek, E.M.V., Deshusses, M.A.,
Matsumoto, M.R., 2004. Direct observation of biofouling in
cross-ow microltration: mechanisms of deposition and
release. J. Membr. Sci. 244 (12), 151165.
Kim, A.S., Chen, H., Yuan, R., 2006. EPS biofouling in membrane
ltration: an analytic modeling study. J. Colloid Interface Sci.
303, 243249.
Knoell, T., Safarik, J., Cormack, T., Riley, R., Lin, S.W., Ridgway, H.,
1999. Biofouling potentials of microporous polysulfone mem-
branes containing a sulfonated polyetherethersulfone/poly-
ethersulfone block copolymer: correlation of membrane
surface properties with bacterial attachment. J. Membr. Sci.
157 (1), 117138.
Kulakov, L.A., McAlister, M.B., Ogden, K.L., Larkin, M.J., OHanlon, J.F.,
2002. Analysis of bacteria contaminating ultrapure water in
industrial systems. Appl. Environ. Microbiol. 68, 15481555.
Ludwig, W., Strunk, O., Westram, R., Richter, L., Meier, H., Yadhuku-
mar, Buchner, A., Lai, T., Steppi, S., Jobb, G., Forster, W., Brettske, I.,
Gerber, S., Ginhart, A.W., Gross, O., Grumann, S., Hermann, S.,
Jost, R., Konig, A., Liss, T., Lussmann, R., May, M., Nonhoff, B.,
Reichel, B., Strehlow, R., Stamatakis, A., Stuckmann, N., Vilbig, A.,
Lenke, M., Ludwig, T., Bode, A., Schleifer, K.H., 2004. ARB: a
software environment for sequence data. Nucl. Acids. Res. 32,
13631371.
Muyzer, G., Brinkhoff, T., Nubel, U., Santegoeds, C.M., Schafer, H.,
Wawer, C., 1998. Denaturing gradient gel electrophoresis (DGGE)
in microbial ecology. In: Akkermans, A.D.L., van Elsas, J.D.,
Bruijn, F.J. (Eds.), Molecular Microbial Ecology Manual. Kluwer
Academic Publishers, Dortrecht, pp. 127.
Oved, T., Shaviv, A., Goldrath, T., Mandelbaum, R.T., Minz, D.,
2001. Inuence of efuent irrigation on community composi-
tion and function of ammonia-oxidizing bacteria in soil. Appl.
Environ. Microbiol. 67, 34263433.
Pang, M.C., Hong, H., Guo, H., Liu, W.-T., 2005. Biolm forma-
tion characteristics of bacterial isolates retrieved from a
reverse osmosis membrane. Environ. Sci. Technol. 39,
75417550.
Park, N., Kwon, B., Kim, I.S., Cho, J., 2005. Biofouling potential of
various NF membranes with respect to bacteria and their
soluble microbial products (SMP): characterizations, ux
decline, and transport parameters. J. Membr. Sci. 258 (12),
4354.
Ridgway, F.R., Flemming, H.C., 1996. Membrane biofouling. In:
Mallevialle, P.E.O., Wiesner, M.R. (Eds.), Water Treatment
Membrane Processes. McGraw-Hill, New York, NY, pp. 6.16.62.
Ridgway, H.F., Kelly, A., Justice, C., Olson, B.H., 1983. Microbial
fouling of reverse-osmosis membranes used in advanced
wastewater treatment technology: chemical, bacteriological,
and ultrastructural analyses. Appl. Environ. Microbiol. 45,
10661084.
Sarro, M.I., Garcia, A.M., Moreno, D.A., 2005. Biolm formation in
spent nuclear fuel pools and bioremediation of radioactive
water. Int. Microbiol. 8, 223230.
Schafer, A., Andritsos, N., Karabelas, A.J., Hoek, E.M.V., Schneider, R.,
Nystrom, M., 2005. Fouling in nanoltration. In: Schaefer, A.,
Fane, A.G., Waite, T.D. (Eds.), Nanoltration: Principles and
Applications. Elsevier, Oxford, UK, pp. 169240.
Schmeisser, C., Sto ckigt, C., Raasch, C., Wingender, J., Timmis, K.N.,
Wenderoth, D.F., Flemming, H.-C., Liesegang, H., Schmitz, R.A.,
Jaeger, K.-E., Streit, W.R., 2003. Metagenome survey of biolms
in drinking-water networks. Appl. Environ. Microbiol. 69 (12),
72987309.
Stoodley, P., Sauer, K., Davies, D.G., Costerton, J.W., 2002. Biolms
as complex differentiated communities. Annu. Rev. Microbiol.
56, 187209.
Van der Bruggen, B., Vandecasteele, C., Van Gestel, T., Doyen, W.,
Leysen, R., 2003. A review of pressure-driven membrane
processes in wastewater treatment and drinking water
production. Environ. Prog. 22 (1), 4656.
Vrouwenvelder, H.S., van Paassen, J.A.M., Folmer, H.C., Hofman,
J.A.M.H., Nederlof, M.M., van der Kooij, D., 1998. Biofouling of
membranes for drinking water production. Desalination 118
(13), 157166.
ARTICLE IN PRESS
WATER RESEARCH 41 ( 2007) 3924 3935 3935