RES EARCH

MEMBRANES Molecular design of polyester membranes

Harnessing steric hindrance to retard hydrol-
More resilient polyester membranes for ysis, we used a monomer, 3,5-dihydroxy-4-
methylbenzoic acid (DHMBA; Fig. 1A), to
high-performance reverse osmosis desalination fabricate a polyester RO membrane with com-
plete chlorine tolerance. Specifically, the mono-
Yujian Yao1, Pingxia Zhang2, Fei Sun1, Wen Zhang1, Meng Li1, Gang Sha3, Long Teng3, mer was synthesized from 3,5-dihydroxy benzoic
Xianze Wang4*, Mingxin Huo4, Ryan M. DuChanois5, Tianchi Cao5, Chanhee Boo6, acid (DHBA) by aromatic bromination, electro-
Xuan Zhang1*, Menachem Elimelech5* philic aromatic alkylation, and Raney-nickel
reduction, successively (fig. S1). The chemical
Thin-film composite reverse osmosis membranes have remained the gold standard technology structure of DHMBA was characterized and
for desalination and water purification for nearly half a century. Polyamide films offer excellent verified by nuclear magnetic resonance spec-
water permeability and salt rejection but also suffer from poor chlorine resistance, high fouling troscopy, Fourier transform infrared spectros-
propensity, and low boron rejection. We addressed these issues by molecularly designing a polyester copy, and mass spectrometry (figs. S2 to S10).
thin-film composite reverse osmosis membrane using co-solvent–assisted interfacial polymerization The methyl group between the two phenolic
to react 3,5-dihydroxy-4-methylbenzoic acid with trimesoyl chloride. This polyester membrane groups was expected to provide steric hindrance
exhibits substantial water permeability, high rejection for sodium chloride and boron, and complete to impede the hydrolysis of ester bonds in al-
resistance toward chlorine. The ultrasmooth, low-energy surface of the membrane also prevents kaline conditions. We confirmed this hypoth-
fouling and mineral scaling compared with polyamide membranes. These membranes could increasingly esis by constructing model ester compounds

Downloaded from https://www.science.org at Xiamen University on April 22, 2024

challenge polyamide membranes by further optimizing water-salt selectivity, offering a path to made from DHBA or DHMBA with TMC and
considerably reducing pretreatment steps in desalination. then computationally calculating the torsion
potential energy (TPE) for the compound to

W
be attacked by hydroxyl ions (fig. S11). For
ith their excellent separation efficiency, treatment steps (e.g., coagulation, antiscalant the DHBA-TMC compound, OH– can eas-
polyamide thin-film composite reverse addition, disinfection, and dechlorination) to ily approach the ester carbon for all config-
osmosis (TFC-RO) membranes have remove potential foulants and remaining oxi- urations of DHBA and TMC, as evidenced by
been the gold standard technology dizing species, thus preparing the feed water the narrow range of the TPE barrier (0.56 to
for desalinating seawater or reusing for the delicate RO membranes (6). Although 0.98 kcal mol−1). By contrast, the TPE barrier
wastewater for nearly half a century (1, 2). fouling- and chlorine-resistant membranes have varied considerably with the molecular con-
These membranes are fabricated by interfacial been recently reported in the literature to po- figuration of DHMBA-TMC, and the energy
polymerization (IP), in which a reaction occurs tentially avoid these pretreatment steps, they barrier reached as high as 9.92 kcal mol−1.
at the interface between an aqueous phase con- have demonstrated considerably inferior de- Therefore, we expected the DHMBA struc-
taining m-phenylene diamine (MPD) and an salination performance relative to polyamide ture, which contains an additional methyl
organic phase containing trimesoyl chloride membranes (7,8). One exception is a polyester group, to have higher stability toward OH–
(TMC) (3). The thermodynamic instability of selective layer that was formed on an inter- attack than DHBA.
the IP process, however, produces a thin poly- mediate poly(piperazine amide) layer that was Inserting a –CH3 group into the DHMBA
amide film with high surface roughness (4), supported by a poly(ether sulfone) (PES) sub- monomer further lowers the reactivity of the
which favors adhesion of organic foulants or strate (9). Although the polyester selective layer entire molecule, because the steric hindrance
inorganic scalants. Although oxidizing agents exhibited exceptional chlorine resistance and makes the phenolic groups less nucleophilic in
such as chlorine could prevent membrane bio- desalination capacity, the membrane began the presence of base (11). We used density func-
fouling, polyamide films deteriorate in the pres- to hydrolyze at pH 8.0, such that seawater tion theory (DFT) simulations to verify this
ence of active chlorine through N-chlorination (pH 8.0 to 8.5), alkaline brackish groundwater, claim by calculating the highest occupied
at the amide linkages or direct aromatic sub- and municipal wastewater effluent (typically molecular energy (EHOMO) and lowest un-
stitution at the benzene rings (5). pH 7.0 to 8.5) (10) would require other forms occupied molecular energy (ELUMO) of DHBA
To overcome these limitations, industrial of chemical-intensive pretreatment (more spe- and DHMBA. The lower band gap energy of
water treatment requires a series of costly pre- cifically, pH adjustment). Developing a robust DHBA (3.67 eV) than DHMBA (3.81 eV) may
membrane that eliminates all pretreatment suggest that the latter is more chemically
steps, either through resistance to fouling and inert (figs. S12 and S13 and table S1). We
1
Key Laboratory of New Membrane Materials, Ministry of mineral scaling or chemical inertness to clean- also found DHMBA to be less polymerizable
Industry and Information Technology, School of
Environmental and Biological Engineering, Nanjing University ing agents (e.g., active chlorine), could substan- than DHBA, as shown by their measured mo-
of Science and Technology, Nanjing 210094, China. 2Key tially reduce the cost and environmental impact lecular weights (figs. S14 and S15). It is for
Laboratory of Science and Technology on High-Tech Polymer of desalination. this reason that we used a co-solvent–assisted
Materials, Institute of Chemistry, Chinese Academy of
Sciences, Beijing 100190, China. 3School of Material Science
To address these challenges, we molecularly IP process between DHMBA and TMC di-
and Engineering, Herbert Gleiter Institute of Nanoscience, designed a robust polyester TFC-RO membrane rectly on a PES substrate, as illustrated in
Nanjing University of Science and Technology, Nanjing using co-solvent–assisted interfacial polymer- Fig. 1B, to create a poly(isophthalester) selec-
210094, China. 4Engineering Research Center of Low-Carbon
Treatment and Green Development of Polluted Water in
ization. We molecularly designed a polyester tive layer, unlike our previous layer-by-layer
Northeast China, Ministry of Education, Northeast Normal TFC-RO membrane that is chlorine resistant approach (9). The mixed solvent enhances the
University, Changchun 130117, China. 5Department of using co-solvent–assisted interfacial poly- diffusion of the DHMBA molecules from the
Chemical and Environmental Engineering, Yale University,
merization. The membrane exhibits marked aqueous phase to the organic phase (fig. S16)
New Haven, CT 06520, USA. 6Department of Civil and
Environmental Engineering, Korea Advanced Institute of water permeability and high rejection for to create a defect-free, water-salt selective
Science and Technology (KAIST), Daejeon 34141, Republic of NaCl and boron. In addition, it has complete polymer network, in contrast to membranes
Korea. resistance toward chlorine and exception- fabricated using other solvents (fig. S17). We
*Corresponding author. Email: [email protected] (X.Z.);
[email protected] (X.W); [email protected] al resistance toward fouling and mineral then used a 2% isophthaloyl dichloride (IPC)
(M.E.) scaling. solution to react with the residual –OH groups

Yao et al., Science 384, 333–338 (2024) 19 April 2024 1 of 6

RES EARCH | R E S E A R C H A R T I C L E

Downloaded from https://www.science.org at Xiamen University on April 22, 2024

Fig. 1. Design, fabrication procedure, and desalination performance of membranes were posttreated with 2% IPC to end-cap residual phenolic
polyester RO membranes. (A) Structural evolution of aqueous monomers for groups. Experiments were conducted using an effective membrane area of
fabricating RO membranes. EWG, electron-withdrawing group. (B) Schematic 15.0 cm2, cross-flow velocity of 40.0 cm s−1, constant applied pressure of
illustration of the co-solvent–assisted IP procedure between DHMBA and TMC on 15.5 bar, temperature of 25.0 ± 0.5°C, and pH of 6.9 ± 0.1. (H) Seawater
the PES substrate. (C and D) Surface FE-SEM image (C) and atomic force desalination performance of the PIP-DHBA-DHBA, DHMBA, and SW30 mem-
microscopy (AFM) image (D) of the DHMBA membrane. (E) Three-dimensional branes toward a 35,000 mg L−1 NaCl feed solution. Data for the PIP-DHBA-
reconstruction of a dataset from the DHMBA membrane for all atoms. DHBA membrane were adopted from (9). Experiments used an effective
(F) Concentration heatmap of the C element in the DHMBA membrane. membrane area of 15.0 cm2, cross-flow velocity of 40.0 cm s−1, constant
(G) Separation performance of the PIP-DHBA-DHBA, DHMBA, and BW30 applied pressure of 55 bar, temperature of 25.0 ± 0.5°C, and pH of
membranes toward a 2000 mg L−1 NaCl feed solution. Both polyester 7.7 ± 0.1.

in the polyester matrix to prevent undesired formation of a self-polymerized hyperbranched constructing a three-dimensional structure of
side reactions with chlorine during membrane polyester (fig. S18). In field emission scanning the top layer using atom probe tomography,
operation. electron microscopy (FE-SEM) images of the the C element was uniformly distributed over
polyester membrane, we observed a much the scanning area in the DHMBA membrane
Membrane characteristics and smoother membrane surface than that of the (Fig. 1, E and F), whereas more heterogeneity,
desalination performance SW30 membrane (Fig. 1C and fig. S19). The including a cavity likely induced by nano-
We characterized the physicochemical proper- average roughness of the membrane surface bubbles (12), was observed for the SW30
ties of the fabricated polyester TFC-RO membrane (Ra = 2.36 ± 0.23 nm), determined by atomic membrane (figs. S21 to S26). The discrepancy
(the DHMBA membrane) to ensure that a thin, force microscopy (Fig. 1D), further confirmed between these membranes may originate from
defect-free film was formed. The cross-linking the smooth nature of the polyester layer, the thermodynamics and kinetics in the IP
density was calculated from surface x-ray which exhibits more than one order of mag- fabrication process (13–15). Finally, the thick-
photoelectron spectroscopy (XPS) (table S2) nitude lower average roughness than the ness of the polyester film was identified with
to be greater than 92%, demonstrating the SW30 polyamide membrane (fig. S20). By re- laser ellipsometry (33.9 ± 0.83 nm; table S3),

Yao et al., Science 384, 333–338 (2024) 19 April 2024 2 of 6

RES EARCH | R E S E A R C H A R T I C L E

RO membrane (SW30, DuPont), which exhib-

ited 99.0 ± 0.6% salt rejection when desalinating
a 35,000 mg L−1 NaCl feed solution simulating
seawater. Under these conditions, the water flux
of the DHMBA membrane (89.3 ± 6.9 L m−2 h−1
at 55 bar) was considerably higher than that
of the SW30 membrane (59.6 ± 4.1 L m−2 h−1 at
55 bar) while maintaining a rejection of 95.0 ±
0.6% (Fig. 1H and fig. S28B).
Whereas the polyester RO membrane exhib-
ited lower salt rejection than SW30, it provides
higher rejection of boron, which is a seawater
constituent (4 to 6 mg L−1; table S4) that often
requires additional RO passes to fall under
regulatory limits (16). The DHMBA membrane
reduced a 5 mg L−1 boron solution to 0.32 to
0.36 mg L−1 in a single pass (93.2 ± 0.3%
rejection) (Fig. 1H), which falls below the
permissible concentration for irrigation water

Downloaded from https://www.science.org at Xiamen University on April 22, 2024

(0.4 mg L−1) and drinking water standards
(1.0–2.4 mg L−1) (17, 18). By contrast, the SW30
membrane exhibited a much lower boron
removal of 74.4 ± 0.8% under the same test
conditions (Fig. 1H). This high rejection was
maintained for a highly concentrated boron
solution, ruling out boron adsorption to the
membrane as a potential removal mechanism
(table S5). The higher boron rejection of the
DHMBA membrane can likely be attributed
to its chemistry and charge characteristics, be-
cause the estimated pore radii of the DHMBA
and SW30 membranes were comparable (fig.
S29). Specifically, the lower surface carboxyl
group density (figs. S30 and S31A) and higher
volumetric carboxyl group concentration within
the polyester layer (fig. S31B) may result in lower
partitioning and higher friction of boric acid
within the membrane, respectively, compared
Fig. 2. Desalination performance and morphology of the DHMBA and SW30 membranes after with the SW30 membrane.
chlorine exposure. (A to C) Water flux and salt rejection performance of the DHMBA and SW30 membranes
exposed to NaOCl aqueous solution at pH 0.0 (A), pH 7.0 (B), and pH 9.0 (C). The total free Cl exposure Membrane chlorine stability
(CTFC) was 8000 ppm hour–1 (50 ppm NaOCl solution for 160 hours) for the SW30 membrane and Subsequently, chlorine stability tests were per-
100,000 ppm hour–1 (50 ppm NaOCl solution for 2000 hours) for the DHMBA membrane. DHMBA and SW30 formed on the DHMBA and SW30 membranes
membranes were immersed in NaOCl solution under continuous stirring at 25°C, and the solution pH was (Fig. 2, A to C). Analogous to other reports
adjusted using HCl (1.0 M). The chlorine solution was replaced with a fresh solution every 24 hours. (19–21), the performance of the polyamide-
At certain times, membrane sheets were removed from the solution, rinsed thoroughly with deionized water, based SW30 membranes rapidly deteriorated
and tested for their desalination performance with the following experimental conditions: an effective across all pH conditions in the presence of
membrane area of 15.0 cm2, 2000 mg L−1 NaCl feed solution, cross-flow velocity of 40.0 cm s−1, constant active chlorine, as observed in FE-SEM images
pressure of 15.5 bar, and temperature of 25.0 ± 0.5°C. (D to F) FE-SEM images of the top surface of the (Fig. 2, D to F). Its fast degradation under
SW30 membrane after a total free Cl exposure (CTFC) of 8000 ppm hour–1 (50 ppm NaOCl solution for acidic conditions is attributed to HOCl species,
160 hours) at pH 0.0 (D), pH 7.0 (E), and pH 9.0 (F). (G to I) FE-SEM images of the top surface of the which may directly attack the benzene rings
DHMBA membrane after a total CTFC of 100,000 ppm hour–1 (50 ppm NaOCl solution for 2000 hours) at and deteriorate the polyamide matrix (5). By
pH 0.0 (G), pH 7.0 (H), and pH 9.0 (I). contrast, the DHMBA membrane exhibited
exceptional oxidative resistance (Fig. 2, G to I),
transmission electron microscopy (~30 to 35 nm; RO membrane (BW30, DuPont) when desalinat- because of the substituents at the 1, 3, 4, and
fig. S19), and time-of-flight secondary ion mass ing a 2000 mg L−1 NaCl feed solution (Fig. 1G 5 positions in the benzene ring, which hinder
spectrometry (30 to 50 nm; fig. S27). For the and fig. S28A). Specifically, the DHMBA direct chlorination reactions at the 2 and
latter, clear boundary regions could be distin- membrane had a water permeance of 3.36 ± 6 positions. Stable desalination performance
guished for the two membranes, which fur- 0.27 L m−2 h−1 bar−1 and an observed salt re- was observed for the DHMBA membrane when
ther confirms the layer depth of their active jection of 98.2 ± 0.3%, compared with 3.24 ± exposed to aqueous molecular Cl2, which forms
layers. 0.16 L m−2 h−1 bar−1 and 96.6 ± 0.6% for BW30. in acidic conditions (pH 0), suggesting excellent
The fabricated DHMBA membrane demon- Although the salt rejection of the DHMBA mem- inhibition of the aromatic substitution reaction.
strated superior performance compared with brane exceeded that of BW30, it fell slightly The DFT calculations support this mechanism
a state-of-the-art commercial brackish water short of a commercial state-of-the-art seawater as well; both negative Mulliken population

Yao et al., Science 384, 333–338 (2024) 19 April 2024 3 of 6

RES EARCH | R E S E A R C H A R T I C L E

Downloaded from https://www.science.org at Xiamen University on April 22, 2024

Fig. 3. Gypsum scaling and organic fouling of the DHMBA and SW30 and 55 bar, respectively, to ensure the same initial water flux (~50.6 L m−2 h−1). The
membranes. (A) Change of normalized water fluxes of the DHMBA and SW30 composition of simulated seawater is shown in table S4. (C and D) Change of
membranes with time with a model gypsum solution having a saturation index normalized fluxes of the DHMBA and SW30 membranes with time using simulated
(SI) of 0.51. The operating pressures for the DHMBA and SW30 membranes seawater solution (table S4) containing 20 mg L−1 sodium alginate (SA) (C) or
were set at 35 and 55 bar, respectively, to ensure the same initial water flux 20 mg L−1 humic acid (HA) (D) as a model organic foulant. In all organic fouling
(~110 L m−2 h−1). Experiments were conducted using an effective membrane area of experiments, the operating pressures for DHMBA and SW30 membranes were set at
15.0 cm2, a mixed solution containing 1000 mg L−1 (9.01 mM) CaCl2 and 1000 mg 46 and 55 bar, respectively, to ensure the same initial water flux (~41 L m−2 h−1).
L−1 (7.04 mM) Na2SO4, pH of 6.6 ± 0.1, cross-flow velocity of 0.28 m s−1, and a (E, F, I, and J) FE-SEM images and the corresponding energy-dispersive x-ray
temperature of 25.0 ± 0.5°C. (B) Change of normalized water fluxes of the DHMBA spectroscopy (EDS) mapping images of the top surface of both membranes after
and SW30 membranes with time with a simulated seawater solution having an SI of scaling experiments. (G, H, K, and L) FE-SEM images of the top surface of both
0.52. The operating pressures for DHMBA and SW30 membranes were set at 46 membranes after organic (SA or HA) fouling experiments.

values and low Fukui function values were formance decline with increasing the solution images showed a nearly pristine surface for
obtained at these two carbon sites (C1 and C3; pH above 9 (fig. S32), where the ester carbon the DHMBA membrane after the desalination
table S1), demonstrating their inability to site can be attacked by abundant hydroxyl ions performance tests, whereas crystals or aggre-
donate electrons in an electrophilic substi- (fig. S33). This high pH, however, is outside gates could be identified on the SW30 mem-
tution reaction. The XPS results further sup- the range for water and wastewater treat- brane (Fig. 3, E, F, I, and J, and table S7). The
port these findings, because extremely low Cl ment, as well as seawater and brackish water DHMBA membrane also performed well in
content (<0.1%; table S6) was identified for the desalination. treating saline solutions containing model
DHMBA membrane after the stability tests. organic foulants (sodium alginate and humic
The DHMBA membrane also maintained steady Membrane fouling resistance acid), showing negligible water flux decline
desalination performance over 2000 hours at We compared the polyester membrane and compared with the SW30 membrane (Fig. 3, C
pH ~7.0, where HOCl is the dominant chlo- SW30 for their resistance to inorganic (mineral) and D). Fouling was less visible on the DHMBA
rine species. The membrane even tolerated a and organic fouling. While the water flux of membrane in the FE-SEM images compared
total free chlorine exposure of 100,000 ppm the polyamide membrane rapidly declined in with the thick and compact cake layer formed
hour–1 at a pH of up to 9.0. This reveals the treating a model gypsum solution and sim- on the SW30 membrane surface (Fig. 3, G, H,
impact of steric hindrance from –CH3 in in- ulated seawater, almost no performance change K, and L).
hibiting the alkaline hydrolysis of the ester was observed for the polyester membrane over AFM adhesion force measurements verified
linkages. Nevertheless, we still observed per- a 24-hour operation (Fig. 3, A and B). FE-SEM the much lower fouling propensity of the

Yao et al., Science 384, 333–338 (2024) 19 April 2024 4 of 6

RES EARCH | R E S E A R C H A R T I C L E

real seawater and periodic addition of chlorine,

the DHMBA membrane exhibited stable oper-
ation, with a water flux decline of merely 2%
after 15 days (Fig. 4, D and E, and table S9).

Discussion
We have demonstrated polyester RO mem-
branes with exceptional water-salt selectivity,
boron rejection, and resistance to chlorine,
fouling, and mineral scaling. The co-solvent–
assisted IP process enhanced diffusion of the
aqueous reactants to the organic phase, thus
forming a defect-free selective layer with high
desalination performance. The membranes
have a much smoother surface than commercial
polyamide seawater RO membranes, which may
account for their lower propensity for foul-
ing and scaling. The wholly aromatic poly-
ester with a highly cross-linked architecture

Downloaded from https://www.science.org at Xiamen University on April 22, 2024

imparted the membrane with chlorine re-
sistance and hydrolytic stability across a pH
range of 0 to 9. Because our membrane fab-
rication process is similar to that for conven-
tional TFC membranes, it may be possible to
adapt existing industrial production lines to
scale up manufacturing.

REFE RENCES A ND N OT ES
1. J. R. Werber, C. O. Osuji, M. Elimelech, Nat. Rev. Mater. 1,
16018–16025 (2016).
2. Y. J. Lim, K. Goh, M. Kurihara, R. Wang, J. Membr. Sci. 629,
119292 (2021).
3. G. M. Geise, Science 371, 31–32 (2021).
4. Z. Wang et al., Prog. Polym. Sci. 122, 101450 (2021).
5. R. Verbeke, V. Gómez, I. Vankelecom, Prog. Polym. Sci. 72,
1–15 (2017).
6. M. Qasim, M. Badrelzaman, N. N. Darwish, N. A. Darwish,
Fig. 4. Fouling comparison of the DHMBA and SW30 membranes in seawater desalination. N. Hilal, Desalination 459, 59–104 (2019).
7. H. B. Park, B. D. Freeman, Z. B. Zhang, M. Sankir, J. E. McGrath,
(A) Adhesion force normalized by particle diameter (Fmax/Rp) distribution between a COOH-modified particle Angew. Chem. Int. Ed. 47, 6019–6024 (2008).
probe and the DHMBA or SW30 membranes. Measurements were conducted 30 times on four different 8. R. Verbeke et al., Environ. Sci. Technol. Lett. 8, 818–824
points of the membrane surface, corresponding to a total of 120 measurements for each membrane. (2021).
9. Y. Yao et al., Nat. Sustain. 4, 138–146 (2021).
(B and C) QCMD mass-time plots of the DHMBA and SW30 membranes toward a mixed solution containing 10. G. Tchobanoglous, H. D. Stensel, R. Tsuchihashi, F. L. Burton,
1000 mg L−1 (9.01 mM) CaCl2 and 1000 mg L−1 (7.04 mM) Na2SO4 (saturation index, SI, 0.51, pH 6.6 ± 0.1) Wastewater Engineering: Treatment and Resource Recovery
(B) and simulated seawater (SI of 0.52; table S4) (C). (D) Performance of the DHMBA and SW30 membranes (McGraw-Hill, ed. 5, 2013).
11. E. Pop, M. Huang, M. E. Brewster, N. Bodor, Int. J. Quantum
toward real seawater desalination for 15 days. Water flux was monitored over the entire period. NaOCl Chem. 44 (S19), 77–85 (1992).
was added to the seawater solution (see measured chemical composition in table S4) at a concentration of 12. X. Ma et al., Environ. Sci. Technol. 5, 123–130
5 mg L−1 and supplied every three days. Experiments were conducted using an effective membrane area (2018).
13. M. F. Jimenez-Solomon, Q. Song, K. E. Jelfs, M. Munoz-Ibanez,
of 15.0 cm2, cross-flow velocity of 0.46 m s−1, temperature of 25.0 ± 0.5°C, and constant pressure of 50 bar A. G. Livingston, Nat. Mater. 15, 760–767
(DHMBA membrane) or 55 bar (SW30 membrane) to ensure the same initial water flux of ~22 L m−2 h−1. (2016).
For the SW30 membrane, once the flux decreased by 40%, an in situ cleaning procedure was performed 14. S. J. Vanhook, M. F. Schatz, J. Swift, W. McCormick,
H. L. Swinney, J. Fluid Mech. 345, 45–78 (1997).
using NaOH (0.1 wt %) for 30 min, HCl (0.2 wt %) for 30 min, and pure water for another 30 min, 15. D. Ren, Y. T. Jin, T. Y. Liu, X. Wang, ACS Appl. Mater. Interfaces
successively. (E) FE-SEM images of the top surface of the DHMBA membrane after the seawater experiment 12, 7605–7616 (2020).
and the corresponding EDS mapping of various elements. 16. R. Bernstein, S. Belfer, V. Freger, Environ. Sci. Technol. 45,
3613–3620 (2011).
17. K. L. Tu, L. D. Nghiem, A. R. Chivas, Separ. Purif. Tech. 75,
87–101 (2010).
DHMBA membrane compared with the SW30 a much lower accumulation of mass on the 18. A. Tal, Science 313, 1081–1084 (2006).
19. A. Antony, R. Fudianto, S. Cox, G. Leslie, J. Membr. Sci. 347,
membrane (Fig. 4A). AFM results revealed a DHMBA membrane surface compared with
159–164 (2010).
high average adhesion force (–3.33 mN m−1) the SW30 membrane (Fig. 4, B and C). Clean- 20. K. Huang et al., Environ. Sci. Technol. 53, 8167–8176
and wide adhesion force distribution between ing with deionized water (Fig. 4B) restored (2019).
21. V. T. Do, C. Y. Tang, M. Reinhard, J. O. Leckie, Environ. Sci.
the SW30 membrane surface and a model the initial state of the DHMBA membrane,
Technol. 46, 852–859 (2012).
organic foulant (a COOH-rich particle probe). whereas accumulated mass on the SW30 mem-
By contrast, a low average adhesion force brane was mostly irreversible. The ultrasmooth, AC KNOWLED GME NTS
(–0.89 mN m−1) and narrow distribution (–2.6 to low-energy, and less-charged surface of the Funding: This work was supported by the National Natural Science
0 mN m−1) was observed for the DHMBA mem- DHMBA membrane likely contributed to its Foundation of China (grants 22178178, 52170027, and
U20A20322) and the Natural Science Foundation of Jiangsu
brane. For gypsum scaling, quartz crystal mi- antifouling and antiscaling properties (figs. Province (grant BK20220081). Author contributions: X.Z
crobalance with dissipation (QCMD) showed S30 and S34 and table S8). Even in a test with conceived the initial idea and experimental design. X.Z. and M.E.

Yao et al., Science 384, 333–338 (2024) 19 April 2024 5 of 6

RES EARCH | R E S E A R C H A R T I C L E

supervised the study and experiments. Y.Y. performed the Competing interests: X.Z. and Y.Y. are inventors on authorized SUPPLEMENTARY MATERIALS
membrane fabrication and desalination experiments. F.S., W.Z., patents (China National Intellectual Property Administration science.org/doi/10.1126/science.adk0632
M.L., M.H., and X.W. performed some of the characterizations numbers ZL 201911277839.9 and ZL 201911270642.2) that cover Materials and Methods
and analyzed the data. P.Z. performed molecular dynamics the fabrication of polyester RO membranes. Data and materials Supplementary Text
simulations and analyzed the data. G.S. and L.T. performed the availability: All data are available in the main text or the Figs. S1 to S34
APT experiments and analyzed the data. C.B. performed the supplementary materials. License information: Copyright © 2024 Tables S1 to S9
AFM experiments and analyzed the data. All authors analyzed the the authors, some rights reserved; exclusive licensee American References (22–50)
results and commented on the manuscript. Y.Y., X.Z., R.M.D., Association for the Advancement of Science. No claim to original
and M.E. wrote the original draft of the manuscript. Y.Y., X.Z., US government works. https://www.science.org/about/science- Submitted 31 July 2023; accepted 19 March 2024
T.C., X.W., R.M.D., and M.E. reviewed and edited the manuscript. licenses-journal-article-reuse 10.1126/science.adk0632

Downloaded from https://www.science.org at Xiamen University on April 22, 2024

Yao et al., Science 384, 333–338 (2024) 19 April 2024 6 of 6