Articles

https://doi.org/10.1038/s41565-020-00796-x

Biomimetic artificial water channel membranes

for enhanced desalination
Maria Di Vincenzo 1, Alberto Tiraferri 2, Valentina-Elena Musteata3, Stefan Chisca3,
Rachid Sougrat4, Li-Bo Huang1,5, Suzana P. Nunes 3 and Mihail Barboiu 1,5 ✉

Inspired by biological proteins, artificial water channels (AWCs) can be used to overcome the performances of traditional desali-
nation membranes. Their rational incorporation in composite polyamide provides an example of biomimetic membranes applied
under representative reverse osmosis desalination conditions with an intrinsically high water-to-salt permeability ratio. The
hybrid polyamide presents larger voids and seamlessly incorporates I–quartet AWCs for highly selective transport of water.
These biomimetic membranes can be easily scaled for industrial standards (>m2), provide 99.5% rejection of NaCl or 91.4%
rejection of boron, with a water flux of 75 l m−2 h−1 at 65 bar and 35,000 ppm NaCl feed solution, representative of seawater
desalination. This flux is more than 75% higher than that observed with current state-of-the-art membranes with equivalent
solute rejection, translating into an equivalent reduction of the membrane area for the same water output and a roughly 12%
reduction of the required energy for desalination.

W
ater scarcity is one of the important challenges of our the creative strategies for addressing such scale-up challenges is to
time1,2. Membrane-based technologies have a tremen- combine the PA material, known for its scalability via the integra-
dous role in addressing this challenge by allowing effi- tion within a typical roll-to-roll processing system, with the highly
cient water purification and desalination3. Desalination membranes permeable and selective AWCs. The key challenge in the construc-
allowing the transport of water while rejecting solutes are used in tion of such a hybrid material is the required adaptive gentle inter-
seawater reverse osmosis and brackish water reverse osmosis. action between PA and water channels, preventing the formation
Thin-film composite (TFC) polyamide (PA) membranes pre- of defects.
pared via interfacial polymerization (IP), have served as standard Herein, we report the incorporation of AWCs within scalable
for seawater reverse osmosis and brackish water reverse osmosis biomimetic membranes that outperform the classical reverse osmo-
over the last 40 years4. The control of the film morphology5–7 or the sis membranes. The new strategy leads to a greater fundamental
incorporation of nanofillers8–10 have both been shown to increase understanding of how incorporation of AWCs, starting from their
water permeability while generally leading to lower selectivity, as colloidal self-assembled superstructures, can be optimized at the
the resulting defects in the active layer affect the solute retention. nanoscale to facilitate the selective transport of water mainly occur-
To overcome the selectivity-permeance trade-off while keeping a ring through channels. One novelty here is related to the identifi-
water permeance of at least 2–4 l m−2 h−1 bar−1 (ref. 11), completely cation of unexpected sponge-like superstructures of the hybrid
new approaches would be required. PA–AWC materials. Moreover, all other selective channels or mate-
Biological aquaporins12,13 exhibiting high water permeability and rials previously reported in the literature23–27 remain applicable in
total ionic rejection, have been embedded in artificial bio-assisted principle, while the biomimetic membranes reported here have
membranes. Nonetheless, there are many challenges to over- been tested and applied under seawater desalination conditions
come with the goal to preserve their activity at high-salinity and with representative pressure and crossflow conditions of full-scale
high-pressure conditions, or to obtain high-density aquaporin-based systems. The relatively straightforward quantification of biomi-
membranes14. Parallel to these investigations, straightforward metic imidazole-quartet, I–quartet AWCs, relevant to subsequently
synthetic biomimetic artificial water channels (AWCs) have been construct reverse osmosis membranes on a larger scale based on
proposed15,16. The development of AWCs17–21 has increased the pros- such probes, is a particularly important strategy.
pect of using these materials in desalination technologies, since
their structure can be tuned to control the water translocation in Kinetic formation of AWC–PA membranes prepared via IP
a biomimetic fashion22 with near-perfect selectivity23. Flat-sheet Our fabrication strategy of AWC–PA layers is adding one
films (>cm2) embedding graphene, carbon nanotubes24–26 or pil- preliminary step to the conventional IP4 method (Fig. 1a):
lar[5]arenes27 have led to large improvements in water permeance Hexylureido-ethyl-imidazole (HC6) was dissolved in a ethanolic
(20–65 l m−2 h−1 bar−1), but their salt rejection is typically limited aqueous solution, followed by sonication to obtain a colloidal AWC
with little interest for seawater desalination. These results are related aggregate that was dispersed onto the active surface of a commer-
to important scale-up challenges in translating molecular transport cial polysulfone (PSf) support before starting the traditional IP
to performant metre-scale membranes. We postulate that one of procedure. During the IP, the AWCs can produce highly dispersed

Institut Européen des Membranes, Adaptive Supramolecular, Nanosystems Group, University of Montpellier, ENSCM, CNRS, Montpellier, France.
1

2
Department of Environment, Land and Infrastructure Engineering, Politecnico di Torino, Turin, Italy. 3King Abdullah University of Science and Technology,
Biological and Environmental Science and Engineering Division, Advanced Membranes and Porous Materials Center, Thuwal, Saudi Arabia. 4King Abdullah
University of Science and Technology, Core Laboratories, Thuwal, Saudi Arabia. 5Lehn Institute of Functional Materials, School of Chemistry, Sun Yat-Sen
University, Guangzhou, China. ✉e-mail: [email protected]

190 Nature Nanotechnology | VOL 16 | February 2021 | 190–196 | www.nature.com/naturenanotechnology

Nature Nanotechnology Articles
a HC6 H MPD
N AWC colloidal Crystallization Cl O Interfacial
C
O H aggregates AWC–MPD TMC polymerization
C6H13 N
N N H 2N NH2
H H Cl O
C C PA
O Cl
Impregnation 1 Impregnation 2 Impregnation 3
AWC aqueous phase MPD hexane TMC

NH

NH
O C

HN
HN

NH
NH

O
NH
O

C
O

C
NH
C CH2 O

NH
C

C
NH
HN NH HN

NH
NH NH
O C

CH
HN

2
CH
HN
O

2
NH
C CH2 O C NH

O
HN NH HN
NH

C
NH
O C

NH
NH NH

C
O C

Colloidal AWC–MPD

NH
O
C CH2 O C O
HN NH HN C CH2

aggregates
NH O C
HN NH HN
NH

NH
O C

AWC–PA O
sponge crystals C
NH
CH2
HN
O C
HN
NH

Nanoscale self-assembly
AWC–PA

b d e

c f g

Fig. 1 | Membrane preparation and SEM characterization. a, Synthetic procedure to obtain the biomimetic TFC–HC6 membrane and hypothetical
mechanism: impregnation 1 of the PSf ultrafiltration support with an aqueous/ethanolic solution of HC6, followed by the impregnation 2 with an aqueous
solution of MPD monomer and then the impregnation 3 with a hexane solution of TMC monomer to begin to the IP and obtain the composite PA TFC–HC6
membranes incorporating AWCs. b,c, Representative SEM micrographs of cross-sectional (b) and top view (c) of the uppermost layer of TFC–HC6 films.
Magnification, S-4,800 ×100.0 k. Scale bars, 300 nm. d–g, SEM micrographs of cross-sectional (d,f) and of the surface (e,g) showing compact structures of
control TFC membranes (d,e) and the formation of large voids at the PSf–PA interface of TFC–HC6 membranes (f,g). The elemental analysis was performed
for each membrane by using EDX. Magnification, S-4,800 ×50.0 k in d,f; S-4,800 ×25.0 k in e,g. Scale bars, 600 nm in d,f; 1,200 nm in e,g.

crystalline phases that are probably cross-linked with growing PA 0.1 w/w% solution of trimesoyl chloride (TMC) monomer during
oligomers into hybrid nanoparticles of 30–40 nm in diameter and the ensuing IP; the membranes thus formed are referred to as ‘opti-
integrated into the PA after a controlled aggregation/crystallization mized TFC–HC6 membranes’. The reference TFC membranes were
process. The morphology of the layer depends on the density of the prepared with the same procedure.
AWC nanoparticles interacting with the surrounding PA matrix. To gain insight into the mechanism of formation of colloi-
To improve the homogeneity of the AWCs distribution, different dal aggregates, we monitored different samples using dynamic
IP protocols were investigated. In the best procedure, the AWC light scattering (DLS). When 0.2 ml of water was added to a 1-ml
solution was left to sit on the PSf for 60 s and a 2 w/w% solution of ethanolic solution of 0.5% HC6, a reversible formation of aggre-
m-phenylenediamine (MPD) monomer was made to react with a gates (Dh = 600–800 nm) was observed, which was followed by a

Nature Nanotechnology | VOL 16 | February 2021 | 190–196 | www.nature.com/naturenanotechnology 191

Articles Nature Nanotechnology

a b
PA ridge-and-valley
structure

PSf suport

e f

Fig. 2 | Morphological TEM characterization of the membranes. a,b, TEM cross-sectional images of the pristine PA on porous PSf support (a) and detail
of the top layer (b) of the TFC–HC6 membrane with high magnification (membranes stained with OsO4). Scale bars, 200 nm. c, 3D reconstruction TEM
tomography of TFC–HC6 membrane in different tilting angles. d, Instantaneous screen shots of TFC–HC6 membrane tomography at a fixed angle: the
PA layer formed on the surface of the PSf porous support (not visible) is marked in green. The less dense grey brighter spots, which could correspond to
the AWCs–PA nanoparticles are marked in purple and represent an occupied relative volume of 20%. Scale bar, 200 nm. e,f, Schematic representations
of the selective transport of water (blue spheres) versus ions (green and red spheres) across compact pristine TFC (e) and AWC-embedded TFC–HC6
membranes with a larger surface area (f).

192 Nature Nanotechnology | VOL 16 | February 2021 | 190–196 | www.nature.com/naturenanotechnology

Nature Nanotechnology Articles
a b c
100 20 100.0
Feed: seawater 99.5% Commercial Optimized TFC
Commercial
90 ΔP = 65 bar seawater membrane
seawater
membrane with HC6
80 99.9 membrane

Inverse solute permeability

99.5%
Measured flux (l m–2 h–1)

coefficient, 1/B (l–1 m2 h)

15

Real rejection, Rr (%)

70 Optimized TFC
Obs. rej.
99.7%
membrane
60 99.2% with HC6 99.8
50 10
TFC membrane
40 99.7 with HC6
Control TFC
30

TFC with HC6

Control TFC
5
Commercial

TFC membrane

Optimized
20 99.6 Control TFC
with HC6
10
0 0 99.5
1.0 1.5 2.0 2.5 3.0 1.0 1.5 2.0 2.5 3.0
Water permeance (l m–2 h–1 bar–1) Water permeance (l m–2 h–1 bar–1)

d e f
100 10 100.0
Obs. rej. Feed: brackish water Optimized TFC
90 95.0% ΔP = 18 bar Optimized TFC membrane
membrane 99.8
with HC6
80 8
Inverse solute permeability

with HC6
Measured flux (l m–2 h–1)

coefficient, 1/B (l–1 m2 h)

97.2%
99.6

Real rejection, Rr (%)

70 99.3%
Commercial
60 6 99.4 brackish water
98.5% TFC membrane TFC membrane membrane
Control TFC

50 Control TFC with HC6

99.2 with HC6
40 4 Commercial Control TFC
98.5%
brackish water 99.0
30 Commercial
98.6% Commercial membranes
98.2% Commercial
Commercial

20 2 seawater 98.8 seawater

Optimized

brackish water
seawater

TFC with

membrane membrane
10 membrane
HC6

98.6
0 0
Commercial 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9
brackish water
Water permeance (l m–2 h–1 bar–1) Water permeance (l m–2 h–1 bar–1)

Fig. 3 | Performance of the membranes in the desalination of seawater and brackish water. a–c, Seawater. d–f, Brackish water. a,d, Experimental water
flux and observed solute rejection of control TFC membranes, commercial membranes and TFC–HC6 membranes incorporating AWCs at 0% recovery.
The filtration conditions were: 65 bar applied pressure with 35,000 ppm NaCl feed solution at pH 8 for seawater desalination; 18 bar applied pressure
with 5,800 ppm of NaCl feed solution at pH 7 for brackish water desalination. b,e, Selectivity–productivity trade-off graphs plotting the inverse solute
permeability coefficient of the membranes as a function of their respective water permeance; the solid line depicts the proposed TFC upper bound
relationship. c,f, Calculated real rejection of the membranes as a function of respective water permeance. The feed solution temperature in all the tests was
27 ± 1 °C. Obs. rej., observed rejection.

redispersion via ultra-sonication for 15 min. (Supplementary of the two minor (<10%) polymorphs (2θ = 3.62° or 4.09° (100))
Fig. 2a). Small angle X-ray scattering (SAXS) profiles provided (Supplementary Fig. 6).
information and confirmed that, no structures smaller than 150 nm The match between the distances of the ordered rows observed
were detected for diluted HC6 (Supplementary Fig. 3). For more in the transmission electron microscopy (TEM) micrograph and
concentrated solutions of 1.0–2.0% HC6, smaller aggregates the periodicity of lamellar sheets in X-ray single crystal struc-
(Dh = 70–200 nm) were observed after adding 2–6 ml of water to ture indicates that the nanoparticles have a crystalline layered
1 ml of HC6 ethanolic solution, without any visible precipitation arrangement of I-quartets. We propose that this phenomenon
(Supplementary Fig. 2b,c). Smaller aggregates (Dh = 70 nm) were occurs in the first step of membrane synthesis when the solution
obtained when a 2 w/w% aqueous solution of MPD was used instead is poured on the surface of the PSf support. Subsequently, the IP
of pure water (Dh = 200 nm), reminiscent of H bonding between procedure entails the contact of the support with an aqueous solu-
colloidal aggregates of AWC and MPD (Supplementary Table 1). tion of MPD and aggregated phases of AWC–MPD are formed.
Additional SAXS experiments performed on samples of 1% HC6 in Specifically, a heterogeneous nucleation is promoted by H bond-
5:1 ethanol:water, on addition of water to shift the ratio to 1:1 indi- ing between MPD and colloidal AWC, as shown by DLS analysis.
cate the presence of mass-fractal structures with a roughly 15-nm We did not observe a massive crystallization process, indicating
radius of gyration and Porod slope around three (Supplementary that this colloidal solution does not form an solid interlayer as
Table 2 and Fig. 4). These conditions simulate those corresponding previously observed for other solid carbon nanotubes, graphene
to the best results for the nanoaggregates formation and IP mem- oxide or metal–organic frameworks fillers28. Then, the membrane
brane preparation in this study. is immersed in the TMC hexane solution where the IP between
To better explain the hybrid particles formation and their inte- MPD and TMC occurs roughly 10 s at the organic phase side of the
gration into PA, it is important to consider their nano-crystallization interface. In this step, the diffusion of the preformed AWC or MPD
from water and ethanol solutions. The X-ray powder diffraction colloids to the organic phase is disfavoured compared with the dif-
patterns of bulk crystalline powders obtained from colloidal sus- fusion of MPD molecules, but they may interact with growing PA
pensions are indicative of lamellar phases, with the main diffraction segments via H bonding. This process results in the formation of
peaks at 2θ = 3.46° (100), 6.95° (200) and 8.04° (300), corresponding cross-linked hybrid AWC–PA nanoparticles. They are embedded
to interplanar distances of 25.4, 12.5 and 8.4 Å, respectively20,22, and into the incipient PA layer formed on top and surrounding the

Nature Nanotechnology | VOL 16 | February 2021 | 190–196 | www.nature.com/naturenanotechnology 193

Articles Nature Nanotechnology

a Cumulative permeate volume (l) c Recovery (%) e

0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0 5 10 15 20 25 30 35 40 45 50
80 100.0 80 3.6
Optimized TFC
membrane with HC6

Water permeance (l m–2 h–1 bar–1)

70 99.5 70
TFC membrane
3.2 with HC6
Concentration
Measured flux (l m–2 h–1)

60 99.0 60

Observed rejection (%)

in the retentate (g l–1)
50 98.5 50 2.8

Value
40 98.0 40
Permeate flux 2.4
30 97.5 30
(l m–2 h–1)
20 97.0 20 Retentate
2.0
–2 –1 –1 flow rate (m3 per h)
10 A = 2.45 l m h bar 96.5 10
(measured with ΔP = 65 bar)
0 96.0 0 1.6
0 5 10 15 20 25 30 35 40 45 50 20 30 40 50 60 70

0
8

263
6
74
37

11

14

22
18
Applied pressure (ΔP)

b Recovery (%) d Membrane area (m2) f

0 5 10 15 20 25 30 35 40 45 50 0 40 80 120 160 200 240
80 1.8 80 100.0
Option 1: 6 elements, ΔP = 66.8 bar
NaCl concentration of the 2 3 Feed: seawater
area = 222 m specific energy = 4.36 kWh per m 99.9
Retentate NaCl concentration (g l–1)

Permeate NaCl concentration (g l–1)

70 total collected 1.6 70 ΔP = 65 bar

Option 2:
permeate = 0.29 g l–1 7 elements, ΔP = 59.1 bar 99.8 Flux =
1.4
Permeate flux (l m–2 h–1)

60 60 2 65.1

Observed rejection (%)

area = 260 m −2 −1
77.9
specific energy = 4.12 kWh per m3 99.7 75.8 l m h l m−2 h−1
1.2 68.2 −2 −1
lm h lm h
−2 −1
50 50 Commercial membrane
99.6 77.3
1.0 7 elements, ΔP = 66.8 bar l m−2 h−1
2
area = 260 m
40 40 99.5

Long-term operation
specific energy = 69.8
0.8

Surfacant pH 11.5
4.66 kWh per m3

Citric acid pH 2.5

30 30 99.4 l m−2 h−1
0.6
99.3

6 bar 15 min
As prepared
20 20

Backwash
0.4 99.2

EtOH
10 0.2 10 99.1
0 0 0 99.0
0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0 5 10 15 20 25 30 35 40 45 50
Cumulative permeate volume (l) Recovery (%)

Fig. 4 | Application of the membranes in typical seawater desalination operation. a,b, Experimental water flux, observed solute rejection, feed and
permeate stream concentrations as a function of recovery and cumulative permeate volume; here, the filtration conditions were: 65 bar applied pressure
with 35,000 ppm NaCl feed solution at pH 8 and a final recovery of 50%. A is the intrinsic water permeability. c,d, Validation of the vessel-scale model for
SW30XHR commercial membranes (c) and outcome of the application of the model for the commercial membranes (d) and for the TFC–HC6 membranes
incorporating AWCs; here, the profile of permeate flux within the vessel is plotted as a function of (solid lines) space, that is cumulative membrane area,
or (dashed lines) recovery rate. e, Experimental water permeance in tests comprising a stepwise increase followed by a stepwise decrease of applied
pressure. f, Experimental performance of membranes subjected to mechanical or chemical stress; here, the filtration conditions were: 65 bar applied
pressure with 35,000 ppm NaCl feed solution at pH 8 and recovery of 0%.

nanoparticles. The new type of hybrid material is therefore gen- a diameter of roughly 20–40 nm and are homogeneously enclosed
erated via a nucleation or IP mechanism, which is supported by (30–40 spots per 100 µm2) within the PA (Fig. 2b). The PA layer was
experimental results reported here. preferentially stained with OsO4 or RuO4 to improve the contrast of
Scanning electronic microscopy (SEM) showed the presence of the PA regions distributed around the brighter spots, which appears
the sponge-like nanoparticles incorporated within the PA layer of darker or more grey in the TEM images of TFC (Fig. 2a) and TFC–
TFC–HC6 membranes (Fig. 1b,c), while they were not observed for HC6 membranes (Fig. 2b). These areas are also clearly visible in the
pristine membrane TFC. Close inspection of the SEM cross-sections RuO4 stained two-dimensional images (Supplementary Fig. 7) and
and membrane surfaces (Fig. 1e,g) reveals a typical ridge-and-valley tomography (Supplementary Fig. 8) TEM images.
surface morphology28–31. However, the features observed with the The presence of AWCs is corroborated by results obtained with
incorporation of AWC are clearly larger than those of the pristine attenuated total reflectance–Fourier transform infrared spectrom-
TFC membrane (Fig. 2d,e). This is also confirmed by atomic force etry (Supplementary Fig. 10) and energy dispersive X-ray (EDX)
microscopy results (Supplementary Fig. 5). The TEM confirmed spectrometry (Fig. 1e,g and Supplementary Table 3): (1) in addition
that, compared to reference pristine TFC layers (Fig. 2a), which to the vibrational bands of the PA of the TFC membrane, Fourier
have an homogeneous thickness of roughly 250 nm, the TFC–HC6 transform infrared spectra of TFC–HC6 contain the –CH2–,as
layers (Fig. 2b) have alternating thinner (around 250 nm) and and –CH2–,sym stretches in the range 2,750–2,957 cm−1, as well as
thicker (around 500 nm) layered regions, a morphology that might urea-NHC=O and imidazole stretches at 1,612 and 1,579 cm−1,
also contribute to a higher surface area. It is also clear that we have indicating the incorporation of the HC6 within PA; (2) the sharp
a complex internal structure when the AWCs are added, with differ- band at 3,300 cm−1 attributed to the O-H stretching vibration of
entiated domains in the TFC–HC6 layers. The AWC–PA nanopar- strongly H-bonded water within I-quartets became much broader,
ticles, corresponding to the internal brighter spots with a lower reminiscent of the presence of more mobile water within the
electron density observed by TEM (Supplementary Fig. 7b–d), have hybrid TFC–HC6 membrane and (3) the elemental composition

194 Nature Nanotechnology | VOL 16 | February 2021 | 190–196 | www.nature.com/naturenanotechnology

Nature Nanotechnology Articles
determined by EDX varied when AWCs were incorporated in the recipe for substitute ocean water of the ASTM D1141-98 (2013) as
PA, exhibiting a trend with up to a 10–12% higher fraction of per- feed solution. A water flux of 66.8 ± 3.8 LMH was obtained, with
centage of carbon for TFC–HC6 layers and a concurrent reduction an observed conductivity rejection of 99.4 ± 0.1% and an observed
in percentage of oxygen. boron rejection of 91.2 ± 1.5%. Details of the experimental condi-
The complexity of the structure can be clearly seen in the tions and the ionic composition of the feed and permeate samples,
three-dimensional (3D) TEM tomography videos (Supplementary as well as the values of observed rejection of individual ions are pre-
Fig. 8) or as screen shots of the 100-nm slices while tilting them at sented in Supplementary Table 8.
different angles (Fig. 2c), showing that larger voids and bright spots The TFC–HC6 membranes incorporating AWCs were further
can be observed inside the TFC–HC6 when compared with the assessed when applied in high-recovery filtration, as in commercial
pristine TFC micrographs (Supplementary Figs. 7a and 8a), where desalination processes. Experiments were conducted by achieving
very few or no bright spots are seen. The PSf support has much less 50% recovery of freshwater from a synthetic feed approximating
contrast, appearing as light grey areas in the normal TEM (below the salinity and the pH of typical seawater. The data in Fig. 4a,b,
the yellow line in Fig. 2a), while the PA layer formed on the surface show the trend of permeate flux as a function of recovery: the
of the PSf porous support is clearly darker. In the tomography in flux profile decreases steadily from the initial value of 75 l m−2 h−1
Fig. 2d, the position of the interface between the PSf support and due to the reduction in driving force as the feed solution became
the formed PA layer (green in Fig. 2d) is represented by a dashed increasingly concentrated. The average water flux of the test was
line. The less dense white grey spots that are proposed to be mostly 34.4 l m−2 h−1 and the concentration of the product water was
composed by AWC–PA nanoparticles are marked purple in Fig. 2d. 290 mg l−1, a value related to an excellent palatability based on
The relative volume occupied by the bright (purple in Fig. 2d) spots the World Health Organization Guidelines for Drinking-Water
was estimated to be 20% by conducting the segmentation and calcu- Quality. This experiment approaches the flux profile and the solute
lating the total voxel of the purple zones relative to the total volume concentrations that would be obtained inside a pressure vessel of
occupied by the complete PA–AWC layer. a full-scale system for a single pass sweater reverse osmosis pro-
cess, and it confirms the promising performances of the TFC–HC6
Desalination performances and stability under reverse membranes under representative full-scale conditions. We surmise
osmosis conditions that the enhanced fluid dynamics of real membrane elements would
The free-defect distribution of sponge-like PA–AWC regions is further reduce the effect of concentration polarization when com-
essential to increase their performance, since they determine the pared with laboratory-scale experiments, thus realizing even larger
pathways for the water flow across the active film (Fig. 2e,f). Here, water fluxes and a higher permeate quality. Therefore, a vessel-scale
biomimetic membranes were tested under high-pressure crossflow model was implemented to compute the membrane area and the
conditions, representative of seawater desalination operations in energy required to extract high-quality water from a feed of sea-
real plants. Specifically, optimized TFC–HC6 layers comprising water by applying current commercial reverse osmosis membranes
AWCs provided a consistent increase in water flux of roughly 75 and the bioinspired membranes proposed in this work.
and 150% in the desalination of mimicked seawater (Fig. 3a) and As shown in Fig. 4c, the vessel-scale model, represented by
brackish water (Fig. 3d), respectively, compared to laboratory-made solid lines, was first validated with high-rejection SW30XHR-400i
control TFC membranes. DuPont elements against the results provided by the software Wave
This notable productivity was achieved while maintaining or, in (DuPont) and plotted as black dots. The model accurately described
many instances, while improving the observed NaCl solute rejec- the profile of flux and concentration within the system. Once vali-
tion to roughly 99.5%, relative to traditional TFC membranes. Our dated, the model was thus applied in two optional implementations
membranes also displayed a better combination of productivity of our bioinspired membranes (Fig. 4d). The modelling outcome
and selectivity compared to marketed state-of-the-art membranes, suggests that, if the same pressure of 66.8 bar is applied in the ves-
which were tested in the laboratory under analogous conditions. sel, a 50% recovery in desalination may be achieved with six instead
Please note that an increase in the solute permeability coefficient of seven elements in series, which are needed instead when apply-
(B) is commonly expected when the water permeance (A) increases, ing current commercial seawater desalination membranes. In other
as B represents the flux of solutes dissolved in the permeating water: words, the larger permeance of the membrane TFC–HC6 incor-
thus, as water flux increases, the solute flux is also higher at equiva- porating AWCs would allow the use of a smaller membrane area,
lent values of solute rejection. translating to near proportional savings in capital and membrane
However, in this study, higher water permeance was obtained replacement costs, also with a reduction in the specific energy from
with little increase, or even with a decrease of solute permeability 4.66 to 4.36 kWh for each m3 of product freshwater. A possibly more
coefficient. This phenomenon is explained with the concurrent advantageous configuration would involve the same number of
increase of both water flux and solute rejection (Fig. 3b,c,e,f). A elements with respect to current commercial membranes, that is,
conventional and robust method to compare different membranes identical membrane area and capital costs, but a lower applied pres-
is to plot their performance in a permeability–selectivity trade-off sure of 59.1 bar. According to the model, this arrangement would
chart (Fig. 3b,e). While the data associated with laboratory-made provide a more uniform permeate flux within the vessel and a note-
control membranes are placed below the upper bound, mostly worthy 12% reduction of the required energy to produce the same
due to low water permeance, the membranes incorporating AWCs amount of freshwater. It is important to highlight that the applica-
defeat the trade-off. This result is especially true for optimized tion of membrane elements is subject to several constraints in terms
TFC–HC6 membranes, whose permeability–selectivity perfor- of minimum and maximum crossflow rate and maximum permeate
mance places them well beyond the bound line. The improvements flux, as well as maximum recovery rate, which can only be evaluated
accomplished with the incorporation of AWCs are especially evi- at the time of full-scale system design.
dent when the real rejection provided in desalination experiments A final set of experiments were conducted to understand the
is plotted against the water permeance (Fig. 3c,f). Real rejection is resilience of the layers to mechanical and chemical stress. The mate-
evaluated taking into account concentration polarization phenom- rials displayed a partially plastic behaviour, as membranes could
ena that occurred during filtrations and is a better parameter than not entirely recover their original permeance once subjected to high
observed rejection for understanding the intrinsic selectivity of the hydraulic pressures (Fig. 4e). Nevertheless, the new membranes
membranes. To assess the performance of the membranes with real were notably robust and maintained their selectivity after being
seawater, low-recovery experiments were also performed using the subject to mechanical solicitations, including changes in pressure,

Nature Nanotechnology | VOL 16 | February 2021 | 190–196 | www.nature.com/naturenanotechnology 195

Articles Nature Nanotechnology

to backwash and in long-term filtration experiments. The transport 8. Jeong, B. H. et al. Interfacial polymerization of thin film nanocomposites: a
performance of the bioinspired membranes was also unaffected by new concept for reverse osmosis membranes. J. Memb. Sci. 294, 1–7 (2007).
9. Duan, J. et al. High-performance polyamide thin-film-nanocomposite reverse
exposure to relatively harsh environments representative of real osmosis membranes containing hydrophobic zeolitic imidazolate
systems, where membranes would undergo cycles of physical and framework-8. J. Memb. Sci. 476, 303–310 (2015).
chemical cleaning, often with the use of acidic, basic and/or amphi- 10. Ratto, T. V., Holt, J. K. & Szmodis, A. W. Asymmetric nanotube containing
philic compounds. The results in Fig. 4f indicate that similar or membranes. US patent 7,993,524 (2011).
11. Werber, J. R., Deshmukh, A. & Elimelech, M. The critical need for increased
higher NaCl rejections to pristine membranes were observed after
selectivity, not increased water permeability, for desalination membranes.
these were exposed to solutions of different chemistry. Perhaps the Environ. Sci. Techol. Lett. 3, 112–120 (2016).
most notable outcome is related to the ability of the materials to 12. Agre, P. Aquaporin water channels (Nobel lecture). Angew. Chem. Int. Ed. 43,
maintain high performance following immersion in pure ethanol, 4278–4290 (2004).
which was used during HC6 synthesis and to solubilize this com- 13. Eriksson, U. K. et al. Subangstrom resolution X-ray structure details
aquaporin–water interactions. Science 340, 1346–1349 (2013).
pound during membrane fabrication. This result indicates that the 14. Tang, C. Y., Zhao, Y., Wang, R., Hélix-Nielsen, C. & Fane, A. G. Desalination
AWCs were effectively incorporated within the PA matrix and no by biomimetic aquaporin membranes: review of status and prospects.
detrimental effects of leaching or resolubilization were detected, as Desalination 308, 34–40 (2013).
the membranes showed no loss in performances (Fig. 4f). 15. Barboiu, M. Artificial water channels. Angew. Chem. Int. Ed. 51,
11674–11676 (2012).
16. Barboiu, M. & Gilles, A. From natural to bioassisted and biomimetic artificial
Conclusions water channel systems. Acc. Chem. Res. 46, 2814–2823 (2013).
This rationalization corroborates the results discussed above, indi- 17. Leduc, Y. et al. Imidazole-quartet water and proton dipolar channels.
cating that the embedded self-assembled AWCs did not generate Angew. Chem. Int. Ed. 50, 11366–11372 (2011).
defects of the uppermost active layer. Indeed, the highly selective 18. Hu, X. B., Chen, Z., Tang, G., Hou, J. L. & Li, Z. T. Single-molecular artificial
AWC are intrinsically promoting the preferential selective pas- transmembrane water channels. J. Am. Chem. Soc. 134, 8384–8387 (2012).
19. Si, W. et al. Selective artificial transmembrane channels for protons by
sage of water with a substantially better perm-selective behaviour formation of water wires. Angew. Chem. Int. Ed. 50, 12564–12568 (2011).
through a selective AWC-embedded layer with larger surface area. 20. Licsandru, E. et al. Salt-excluding artificial water channels exhibiting
It also indicates an excellent structural compatibility of AWCs with enhanced dipolar water and proton translocation. J. Am. Chem. Soc. 138,
the surrounding PA matrix, a necessary requisite to construct a 5403–5409 (2016).
seamless active layer without defects. 21. Tunuguntla, R. H. et al. Enhanced water permeability and tunable ion
selectivity in subnanometer carbon nanotube porins. Science 357,
792–796 (2017).
Online content 22. Kocsis, I. et al. Oriented chiral water wires in artificial transmembrane
Any methods, additional references, Nature Research report- channels. Sci. Adv. 4, eaao5603 (2018).
ing summaries, source data, extended data, supplementary infor- 23. Werber, J. R. & Elimelech, M. Permselectivity limits of biomimetic
mation, acknowledgements, peer review information; details of desalination membranes. Sci. Adv. 4, eaar8266 (2018).
24. McGinnis, R. L. et al. Large-scale polymeric carbon nanotube membranes
author contributions and competing interests; and statements of with sub-1.27-nm pores. Sci. Adv. 4, e1700938 (2018).
data and code availability are available at https://doi.org/10.1038/ 25. Mi, B. Graphene oxide membranes for ionic and molecular sieving. Science
s41565-020-00796-x. 343, 740–742 (2014).
26. Yang, Y. et al. Large-area graphene-nanomesh/carbon-nanotube hybrid
Received: 27 May 2020; Accepted: 13 October 2020; membranes for ionic and molecular nanofiltration. Science 364,
1057–1062 (2019).
Published online: 9 November 2020
27. Shen, Y. X. et al. Achieving high permeability and enhanced selectivity for
Angstrom-scale separations using artificial water channel membranes.
References Nat. Commun. 9, 2294 (2018).
1. Mekonnen, M. M. & Hoekstra, A. Y. Four billion people facing severe water 28. Lin, L., Lopez, R., Ramon, G. Z. & Coronell, O. Investigating the void
scarcity. Sci. Adv. 2, e1500323 (2016). structure of the polyamide active layers of thin-film composite membranes.
2. Eliasson, J. The rising pressure of global water shortages. Nature 517, J. Memb. Sci. 497, 365–376 (2016).
6–7 (2015). 29. Pacheco, F., Sougrat, R., Reinhard, M., Leckie, J. O. & Pinnau, I. 3D
3. Shannon, M. A. et al. Science and technology for water purification in the visualization of the internal nanostructure of polyamide thin films in RO
coming decades. Nature 452, 301–310 (2008). membranes. J. Memb. Sci. 501, 33–44 (2016).
4. Cadotte, J. E., Petersen, R. J., Larson, R. E. & Erickson, E. E. A new thin-film 30. Wong, M. C. Y., Lin, L., Coronell, O., Hoek, E. M. V. & Ramon, G. Z. Impact
composite seawater reverse osmosis membrane. Desalination 32, 25–31 (1980). of liquid-filled voids within the active layer on transport through thin-film
5. Karan, S., Jiang, Z. & Livingston, A. G. Sub-10 nm polyamide nanofilms composite membranes. J. Memb. Sci. 500, 124–135 (2016).
with ultrafast solvent transport for molecular separation. Science 348, 31. Li, Y. et al. Probing flow activity in polyamide layer of reverse osmosis
1347–1351 (2015). membrane with nanoparticle tracers. J. Memb. Sci. 534, 9–17 (2017).
6. Chowdhury, M. R., Steffes, J., Huey, B. D. & McCutcheon, J. R. 3D printed
polyamide membranes for desalination. Science 361, 682–686 (2018). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
7. Tan, Z., Chen, S., Peng, X., Zhang, L. & Gao, C. Polyamide membranes with published maps and institutional affiliations.
nanoscale Turing structures for water purification. Science 360, 518–521 (2018). © The Author(s), under exclusive licence to Springer Nature Limited 2020

196 Nature Nanotechnology | VOL 16 | February 2021 | 190–196 | www.nature.com/naturenanotechnology

Nature Nanotechnology Articles
Data availability with input from all authors. All authors discussed the results and commented on the
The data that support the findings of this study are available from the corresponding manuscript
author on reasonable request.
Competing interests
Acknowledgements M.B., M.D.V. and A.T. are inventors on a provisional patent application submitted in
This work was supported by Agence Nationale de la Recherche grant nos. ANR- 2019 by the SATT AxLR on behalf of the Centre National de la Recherche Scientifique
18-CE06-0004-02, WATERCHANNELS and ERANETMED2-72-357 IDEA. The authors (CNRS) for the design, synthesis and performances of the membrane materials in
thank D. Cot (University of Montpellier) for SEM experiments, F. Ricceri (Politecnico di this study.
Torino) for help with filtration experiments and M. Deleanu (University of Montpellier)
for help with the organic synthesis of HC6 and characterization.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
Author contributions s41565-020-00796-x.
M.B. conceived the project and designed the experiments. M.D.V. fabricated the Correspondence and requests for materials should be addressed to M.B.
membranes, performed IR, EDX characterization and SEM analysis. L.-B.H. performed
X-ray powder diffraction analysis. M.D.V. and A.T. designed and performed the filtration Peer review information Nature Nanotechnology thanks Isabel Escobar, Jun-li Hou and
experiments. S.P.N. designed and V.-E.M., S.C. and R.S., performed DLS, SAXS and Claus Helix Nielsen for their contribution to the peer review of this work.
TEM experiments and conducted the experimental analysis. M.B. wrote the manuscript Reprints and permissions information is available at www.nature.com/reprints.

Nature Nanotechnology | www.nature.com/naturenanotechnology