Introduction

Nanofiltration (NF) membranes have emerged as versatile tools in water treatment and other industries, allowing selective separation of solutes based on size exclusion and other molecular-level effects including Donnan exclusion, dielectric effects, and van der Waals forces1,2,3,4,5,6,7,8,9,10,11,12,13,14. Among these effects, it is commonly accepted that Donnan (charge) exclusion, arising from the charge of amine and carboxyl fixed groups on the surface of the polyamide active layer, plays a critical role in the separation capabilities of NF membranes15,16,17,18,19,20,21,22. However, questions remain regarding the effects of solution ions on the membrane charge and the consequent influence on membrane performance23,24,25,26; such effects can be particularly important considering recent efforts to achieve separation between similar ions27,28,29,30,31,32,33. Therefore, a molecular-level examination of the interactions between solution ions and the membrane is imperative to gain insights into membrane behavior, ultimately aiding the development of more efficient and customized NF membranes.

Measuring the zeta potential (ZP) of the membrane surface using the streaming potential method has been widely adopted to evaluate the extent and sign of membrane surface charge34,35. Additionally, trends of salt rejection versus pH are frequently used to explain the protonation behavior of the polyamide layer15,36,37,38,39,40. However, discrepancies in the interpretations of these measurements have been reported in the literature34,35,38. For instance, the minimum point of salt rejection curves is often ascribed as the isoelectric point of the membrane41; however, the ZP of the membrane at this supposed isoelectric point is negative for many commercial NF membranes. In another example, the literature is inconsistent in describing the effect of salt concentration on the ZP of the membrane; that is, some studies reported an increase in the negative charge of the membrane at higher salt concentrations42, while other studies reported the opposite38,43. A primary goal of our research is filling these knowledge gaps.

Such discrepancies and inconsistencies in ZP measurements and interpretations suggest that NF membrane charge at a given pH is not an intrinsic (fixed) property, as it is influenced, inter alia, by the solution composition35,44. More specifically, membrane charge is not solely determined by the density and the pKa values of the fixed groups, as it also depends on the specific ionic composition in the solution45. In this regard, different ions (including those having similar charge, valency, and hydrated radius) may differ in their stickiness46, i.e., their ability to adsorb to the membrane surface, by substituting their hydration shell with surface groups47, and affect its charge. Therefore, systematically exploring the effect of solution composition on the charge and performance of NF membranes can help address the aforementioned knowledge gaps and elucidate how solution composition impacts Donnan exclusion under varying conditions.

The current study systematically investigates the ZP and ion permeability of a polyamide NF membrane under a wide range of pH values, ionic compositions, and ionic strength to explore how solution ions affect the charge and performance of the membrane. Our results imply that the membrane-solution interactions are ion-specific, where “stickier” weakly hydrated ions (e.g., Cs+) can more easily shed their hydration shell to attach to the membrane backbone and affect its charge, compared to less “sticky” strongly hydrated ions (e.g., Li+). Additionally, we show that chlorides, which have greater polarizability than the investigated cations, can adsorb to uncharged membrane segments, enhancing the Donnan exclusion of solution anions. Overall, our work clarifies existing discrepancies in ZP studies and demonstrates the contribution of adsorbed solution ions to the charge and performance of NF membranes by correlating between ZP, the ionization behavior of the fixed groups, and ion rejection and permeability.

Results and discussion

Verification of the pK a value of carboxyl groups on the polyamide surface

Membrane charge is directly impacted by the extent of protonation of its fixed groups, i.e., carboxylic and amine groups, a byproduct of incomplete polymerization and crosslinking of thin-film composite membranes such as the NF270 membrane48. Therefore, knowing the equilibrium constants (Ka) and their corresponding pKa values for the protonation reactions of these groups (Eqs. (1) and (2)) is imperative for understanding membrane surface charge. However, the pKa values of fixed carboxyl and amine groups bonded to polyamide membranes may differ from thermodynamic values of the acids with similar structures (i.e., ~4.249 and ~7.850,51 for benzoic acid and N-benzoylpiperazine, respectively), depending on the polymer chemistry. Relevant data on these acids and the polyamide structure is given in Supplementary Table 1. Our first set of experiments was therefore aimed at evaluating the pKa values of the fixed groups bonded to the NF270 membrane and suggesting explanations for discrepancies between the membrane zeta potential (ζ) values and expected charge (Fig. 1).

$${\rm{R}}-{\rm{COOH}}\leftrightarrow {{\rm{R}}-{\rm{COO}}}^{-}{+{\rm{H}}}^{+}$$
(1)
$${{\rm{R}}-{\rm{NH}}}_{2}^{+}\leftrightarrow {{\rm{R}}-{\rm{NH}}+{\rm{H}}}^{+}$$
(2)
Fig. 1: Zeta potential and performance of the NF270 membrane as a function of solution pH.
figure 1

a Average (n = 4) ζ of the NF270 membrane as a function of pH using 1 mM NaCl solution at 23 ± 0.8 °C (Error bars for ZP measurements that are below the order of graph point size are not shown). b Slope between adjacent points depicted in panel a, i.e., Δζ / ΔpH (pink closed circles), and NaCl rejection (blue open circles) as a function of pH. c Schematics of membrane charge alteration and speciation of fixed groups with pH. Weak acids completely dissociate at pH greater (app. 2 pH units) than their pKa values: carboxylic (green ⊖) and amine (blue ⊕) at pH > ~6 and ~10, respectively. Chloride ions are attached to neutral segments on the membrane surface, regardless of pH. Error bars in a and b represent standard deviation.

Full size image

A straightforward approach for determining the pKa of the membrane’s fixed groups is to analyze the membrane ζ vs. pH curve (Fig. 1a) while considering the general shape of concentration vs. pH curves of weak acid systems (Supplementary Fig. 1). Assuming that the pKa of the carboxylic groups on the NF270 membrane is 4.5 (i.e., close to the reported pKa of benzoic acid), a good correlation is observed between the trends of the expected concentration of the charged species (i.e., R-COO−) and the ZP. Specifically, at pH <6.5, similar steep inclines in R-COO− concentration (Supplementary Fig. 1a) and membrane negative charge (Fig. 1a) are observed (with the steepest incline taking place at pH 4.5), followed by steady values of both parameters at pH >6.5. This parallel behavior suggests that the pKa of carboxylic groups on the NF270 membrane surface is around 4.5. This notion is corroborated by comparing the change in ζ with pH (Fig. 1b, closed pink circles) to the general shape of buffer capacity curve of a weak acid (Supplementary Fig. 1c), as both demonstrate a negligible change at pH far above the supposed pKa (i.e., at pH >6.5), while the change is maximal (in absolute values) at pH = pKa. We attribute the slight decrease in the ζ values at pH >8.5 (Fig. 1a) to dissociation of amine groups (Eq. 2), suggesting also that the density of R-NH2+ on the surface of the NF270 membrane is considerably lower compared to R-COO− groups. This conclusion is reasonable since amine groups only form on the edge of the polyamide chains, while the formation of carboxyl groups can take place inside the polyamide chain51, which is substantiated by studies quantifying functional groups on similar membranes48,52,53,54. The exact value of the pKa of membrane amine groups is hard to locate since their density is low and there is technical difficulty in obtaining results at pH >10. Nevertheless, based on our findings (i.e., slight decrease in ζ values at pH >8.5 and a decrease in rejection at pH >10), we estimate that the pKa of amine groups is between pH 8.5 and 9.0.

Correlation between salt rejection and zeta potential curves

Membrane charge impacts the membrane performance via Donnan (charge) exclusion, as often demonstrated by the salt rejection vs. pH curves, supposedly demonstrating good match between the pH at which the minimum rejection is achieved and the pH where minimum fixed groups on the membrane surface are charged15,40,55,56. However, according to previous studies (e.g., Tu et al.57) and the above explanation, the minimum number of charged groups on the surface of the NF270 membrane is achieved at pH <3.0 (Fig. 1c), whereas the minimum rejection is at pH = pKa = 4.5 (Fig. 1b, blue circles). Therefore, our results along with previously accumulated observations call for an alternative explanation for the relation between rejection vs. pH curves and membrane effective charge. Figure 1b shows an increase in rejection along the pH at 4.5 ≤ pH ≤ 9.0, which is explained by the increase in the number of deprotonated carboxyl groups at higher pH. Note that since the amine groups remain charged at pH smaller than their pKa, the increase in the number of deprotonated carboxyl groups at pH < 9 corresponds to an increase in the overall number of charged fixed groups. However, at pH > 9 most of the amine groups are deprotonated (i.e., neutral); therefore, the membrane fixed charge becomes more negative, while the overall number of charged groups is decreased. Naturally, a decrease in the overall number of charged groups leads to a decrease in Donnan exclusion58, which is demonstrated by a decrease in rejection of NaCl at pH > 9.0 (Fig. 1b). Notably, the decrease in rejection displayed from pH ~ 3.0 to pH 4.5 contradicts the increase in deprotonated carboxyl groups at this pH range, indicating that Donnan exclusion is not the only factor controlling the rejection at this pH range. Possibly, at pH < pKa of the carboxyl groups, where membrane charge is minor, swelling of the polymer can outcompete the effect of charge exclusion. Thus, at pH < pKa of the carboxyl groups, the rejection decreases with increasing pH due to increased swelling at higher pH, as previously shown for this pH range54,59.

Irreversible and influential adsorption of chlorides

We discussed above the considerably higher density of carboxyl groups on the membrane surface compared to amine groups. Based on this idea, even under a conservative assumption of 10:1 ratio of membrane carboxyl to amine groups42,60, at 1 pH unit below the pKa of the carboxyl groups (i.e., pH ~3.5, in which [R – COO−] : [R – NH2+] ≈ 1, Supplementary Fig. 1), ζ is presumed to be close to zero or slightly positive. Nevertheless, multiple measurements conducted in this study and previous ones51,61,62 showed a clearly negative ζ at pH 3.5 (Fig. 1a). It was shown before that polarizable ions experience free energy minimum at the water-organic phase interface63 and also that anions can go nearer to nonpolar or hydrophobic surfaces because they are more weakly hydrated than cations. Therefore, a likely explanation for the seemingly low membrane ζ values is Cl− adsorption to neutral segments of the membrane surface42 due to its enhanced “stickiness” properties (i.e., weakly hydrated and highly polarizable). Notably, earlier studies demonstrated that ions (predominantly anions) bind to the hydrophobic groups of the membranes64,65,66,67; our experiments conducted to further investigate this finding suggest that chloride ions better adsorb to the hydrophobic segments of the membrane, as elaborated below.

Ion concentrations in the membrane were probed using X-ray photoelectron spectroscopy (XPS) measurements for membranes that were soaked in NaCl solution (and then thoroughly rinsed with DI water), which demonstrated the adsorption of the salt ions to the membranes (Fig. 2a, b). More specifically, an increase in the intensity of both sodium and chlorine on the membrane surface was observed after soaking the polyamide membrane in NaCl solution as compared to soaking in DI water. Atomic percent calculations based on XPS survey scans showed that the chlorine atomic fraction was 5.3%, 9.3%, and 16% for samples soaked in DI water, 1 mM NaCl, and 10 mM NaCl, respectively. These data further indicate that ion sorption may be occurring in the membrane.

Fig. 2: Irreversible and influential adsorption of chlorides.
figure 2

a, b X-ray photoelectron spectroscopy core scans of sodium 1s and chlorine 2p for membranes soaked in 0 mM, 1 mM, and 10 mM sodium chloride. Samples were soaked in their respective solutions for 24 hr at 40 °C and thoroughly rinsed in DI water. c Three consecutive experiments examining NaCl permeability as a function of temperature (i.e., in each cycle, the temperature was elevated from 20 °C to 40 °C) to explore chloride irreversible adsorption. The same membrane coupons were used throughout the three experiments. Experimental conditions: 1 mM NaCl, applied pressure of 18 bar, and crossflow velocity of 2.13 m s−1. Error bars in c represent standard deviation.

Full size image

To further examine the hypothesis of chloride adsorption and its potential effect on Donnan exclusion, an additional set of experiments was conducted using the assumption that Cl− adsorption onto the membrane surface is promoted at elevated temperatures68. More specifically, at constant salt concentration the temperature was elevated sequentially from 20 °C to 40 °C three times to test the effect of chloride adsorption on water and salt permeabilities (Fig. 2c). Our results show the irreversible effect of membrane exposure to high temperature (40 °C) at the presence of merely 1 mM NaCl. That is, increasing the temperature in the first cycle results in a lower salt permeability in the following cycles for a given temperature. We attribute this decrease in salt permeability to irreversible chloride adsorption occurring at higher temperatures, contributing to the negative charge of the membrane and therefore to enhanced Donnan exclusion in the following cycles. Notably, salt permeability at the highest temperature tested remained nearly constant over the three cycles since within the tested temperature range chloride adsorption increases with temperature during the first cycle. On the other hand, water permeability did not show a statistically significant pattern over the cycles (Supplementary Fig. 3), supporting the idea that increased charge exclusion, resulting from chloride adsorption, affected the salt permeability, as opposed to steric effects alone that affected the water permeability.

Zeta potential behavior with varying ionic strength and composition

Previous studies demonstrated contradicting effect of ionic strength (I) on membrane ZP, showing either a decline in the absolute value of ζ with increasing I11 or the opposite38,69. The former observations (i.e., inverse relationships) are explained by charge screening, while the latter imply that adsorption of solution ions (e.g., chlorides) to the membrane surface may occur. To elucidate such phenomena, the effect of I (or salt concentration) on ζ was examined at different pH and using different types of cations (Fig. 3a–c).

Fig. 3: Effects of specific conditions on the zeta potential of the NF270 membrane.
figure 3

The membrane was in contact with single-salt solutions of CsCl (blue squares) or LiCl (green triangles) at various concentrations and pH values. a–c Average (n = 4) ζ obtained for solutions of different concentrations of CsCl and LiCl at 21 ± 2 °C and pH 2.62 ± 0.15, pH 4.51 ± 0.02, and pH 8.9 ± 0.37, respectively. d The stickiness tendency of Cs+ (compared to Li+) is noticeable only for 1 mM solutions (glowing symbols) at pH > 5.5. At high ionic strength, the |ζ| is lower as the electric double layer is compressed.

Full size image

Our results show that for ionic strength higher than 1 mM (regardless of the tested cation or the pH), higher salt concentration resulted in better charge screening and thus reduced absolute value of ZP (|ζ|, Fig. 3a–c). However, further examination of the effect of solution ions on ZP at low I conditions (<1 mM) demonstrates a different trend where increasing I leads to a decreased ζ, suggesting that adsorption of Cl− ions to uncharged segments of the membrane is a prominent factor for highly diluted solutions. More specifically, under the conditions of negatively charged fixed groups (i.e., pH ≥ 4.5, Fig. 3b, c) and very low salt concentrations (i.e., I < 1 mM), chlorides stick to the membrane surface to higher extent than the cations42, presumably due to the higher polarizability (and therefore “stickiness”) of chlorides compared to the cations (Table 1). As a result, elevated membrane negative charge (i.e., lower ζ) is attained by elevating the concentration of Cl− in the solution. At higher salinities (I > 1 mM), however, it is likely that chloride adsorption reaches its maximum and charge screening by the solution cations becomes prominent, leading to a decrease in the absolute value of ζ. Notably, at concentration of 50 mM, charge screening is considerable regardless of the cation, resulting in identical ζ for both cation solutions.

Table 1 Main properties of the ions examined in this work
Full size table

To support the effect of Cl− adsorption on membrane charge at low salinities, the effect of I on ζ was examined at pH 2.6 (Fig. 3a), where membrane fixed groups are slightly positive and therefore both phenomena (i.e., charge screening and chloride adsorption) are expected to impact ζ in the same manner. Figure 3a shows a consistent trend of decrease in ζ with the increase of I, corroborating the hypothesis of Cl− adsorption being the prominent factor effecting ZP at low salinities.

It was recently shown that even ions possessing similar hydrated size and charge such as Li+ and Cs+ show different permeabilities through NF membranes37,70,71,72. Such differences in the transmembrane permeation of similar ions stem from delicate and more complex molecular-level interactions beyond simple size and charge exclusion71,73,74. Similarly, the differences in “stickiness” of Li+ and Cs+ can explain the different ζ values attained with CsCl and LiCl solutions (Fig. 3d). Noticeably, at high pH (>6) the CsCl solution has a less negative ζ than LiCl, which can be attributed to Cs+ ‘s softer hydration shell (or lower hydration enthalpy, i.e., |ΔHhyd (Cs+)| < |ΔHhyd (Li+)|, Table 1), which facilitates its increased adsorption to the surface compared to Li+. In particular, our results demonstrate that the differences in the ZP values achieved with similar ions are most noticeable for moderate concentrations (between 0.5 and 1 mM). At high concentrations (50 mM), similar ZP values were measured for LiCl and CsCl presumably due to the screening effect at higher concentrations that outcompetes ion-specific effects. We note that the differences in the stickiness of the ions have only a slight impact on the membrane ζ values. Thus, only considerable differences in stickiness (such as between Li+ and Cs+) are noticeable, while similar cations (e.g., K+ and Cs+ or Li+ and Na+) show insignificant differences in ζ values (Supplementary Fig. 4).

Correlation between membrane zeta potential and ion permeability

It was shown that ZP is not an indicator of the fixed membrane charge alone, but also represents contribution from the solution ions. In other words, ZP represents the effective charge of the membrane, which is the charge experienced by the solution at the vicinity of the membrane resulting from the combination of the charge of the membrane fixed groups and solution ions. Therefore, it is reasonable that ZP is a good indicator for Donnan exclusion, which is also a result of the effective charge rather than the fixed charge of the membrane alone. To better understand to what extent ZP is a predictor for Donnan exclusion, the correlations between ZP and ion permeability were examined for chloride salts of Li+ and Cs+ representing ions of low and high stickiness (Table 1), respectively (Fig. 4a). In addition, I and pH were altered to examine salt permeability under various known ζ values (inset of Fig. 4a).

Fig. 4: Correlation between membrane zeta potential and ion permeability.
figure 4

a LiCl (green) and CsCl (blue) permeabilities at different conditions and the respective ζ (inset). Experimental conditions during the permeability tests: 5 bar (72.5 psi), 24.8 ± 0.1 °C, and crossflow velocity of 2.13 m s−1. Ratios between LiCl and CsCl measurements of ζ and Ps, rζ and rP, respectively, are indicated next to the bars of each condition. Donnan effect on ion permeation at b pH 4.0 and low ionic strength; c pH 7.5 and high ionic strength; and d pH 7.5 and low ionic strength. Membrane charge is indicated by black ⊕/⊖ circles.

Full size image

Our results show a clear inverse correlation between the parameters; that is, salt permeability is reduced with the increase in |ζ|, which implies that ZP is a reasonable indicator for Donnan exclusion. For instance, at pH 4.0 and I = 1 mM, the |ζ| is relatively low due to weak membrane charge with the accompanying reduced Donnan effects, which results in relatively high ion permeability (Fig. 4b). At pH 7.5 and high ionic strength, the |ζ| is greater and consequently ion permeability is reduced. The main reason for the lower permeability under these conditions is schematically explained in Fig. 4c, showing higher membrane charge, which is partly neutralized by the ions. At pH 7.5 and low ionic strength (Fig. 4d), there is limited charge screening, therefore many membrane charges generate strong Donnan effects resulting in very low permeability. The ZP results may also imply subtle differences in Donnan exclusion experienced by Li+ and Cs+, as reflected by the good correlation between the ratios of LiCl and CsCl ζ (\({r}_{\zeta }={\zeta }_{{\rm{LiCl}}}/{\zeta }_{{\rm{CsCl}}}\)) and Ps (\({r}_{P}={P}_{{\rm{s}}}({LiCl})/{P}_{{\rm{s}}}({CsCl})\)). Notably, the largest rζ was obtained at pH 7.5 and 1 mM, where the negative membrane charge is high (Fig. 1a, c) and charge screening is minor (Fig. 4d), leading to noticeable differences in Cs+ and Li+ permeabilities (rP = 0.74). It is stressed that we tested the match between ZP and Donnan exclusion only for pH ≤ 7.5, where the fixed membrane charge is solely influenced by protonation of carboxyl groups (Eq. (1)), while the amine groups are positively charged throughout the pH range. Thus, in the tested conditions, ZP is directly related to the overall number of charged fixed groups. On the other hand, at pH > 7.5, the increase in |ζ| with pH is the outcome of the protonation of charged amine groups (i.e., Eq. (2)), resulting in a decrease in overall number of charged groups, but an increase in the overall negative charge. Lastly, we considered the potential effect of the carbonate system and water ions on the pH near the membrane wall. Based on the pKa values, the concentrations of the various species (Supplementary Table 3), and the expected permeability of the species, we computed the potential effect on the pH (e.g., Supplementary Fig. 2). These calculations showed the minor effect of the carbonate system and water ions on our results.

Discussion

Understanding the effects of solution ions on Donnan exclusion and transport through NF membranes can aid growing efforts to improve the solute-specific selectivity of NF membranes. Motivated by knowledge gaps in previous research using ZP to study membrane surface charge, we systematically evaluated membrane ZP over a range of pH, salinity, and ionic composition and corelated the results to membrane permeability tested under similar conditions. Initially, we determined the pKa of carboxyl and amine groups to be ~4.5 and ~8.5, respectively. We then proceeded to explain how increased swelling accounts for the discrepancy between the isoelectric point assumed at the minimum point of salt rejection curves and the isoelectric point measured by ZP analysis. We also explain how chloride adsorption accounts for membranes’ charge remaining negative despite the pKa of its functional groups indicating it should be positive. We then systematically evaluated membrane ZP over a range of pH, salinity, and ionic composition to reconcile conflicting reports of the effect of solution concentration on ZP. More specifically, chloride adsorption can explain the increase in ZP absolute value with increasing solution concentration until the point of saturation of bonding sites on the polymer backbone, after which condensing of the electric double layer accounts for decreasing the absolute value of ZP. We conclude with demonstrating that Cs+ ions, with lower hydration enthalpy (and therefore higher stickiness), adsorb more easily to the membrane surface than Li+ ions and showing how ZP is a good indicator of Donnan exclusion, which is evidenced in permeability experiments.

Methods

Membranes and chemicals

Analytical grade chloride-based salts of alkali metals including lithium (LiCl), sodium (NaCl), potassium (KCl), and cesium (CsCl) were purchased from Acros Organics, BioLab, Merck, and Fisher Chemicals, respectively. The ions constituting these salts systematically differ in several properties that can affect the ability of the ions to adsorb to the membrane surface and affect its charge (Table 1). Single-salt solutions for ZP measurements and for NF filtration tests were prepared by dissolving the powdered salt in ultrapure water (Merck, <0.055 mS cm−1) and deionized water (DI) (Zailon, <0.8 mS cm−1), respectively. Hydrochloric acid (HCl) and sodium hydroxide (NaOH) at concentrations of 0.05 M were used to adjust the solution pH. Flat sheets of a thin film composite NF membrane (NF270, Dow Filmtec) were used for all experiments. This membrane is considered a “loose” NF membrane75, where charge effects possess a critical role in the separation54,57,76. All membrane coupons were pretreated with a 25% v/v isopropyl alcohol solution for 30 min and then rinsed with deionized water for 30 min, three times. Each membrane coupon was used for a single experiment of ZP or filtration (i.e., each experiment was performed with different coupons except for the experiment described in Fig. 2c).

Zeta potential measurements

Membrane surface charge was evaluated by measuring streaming potential using the SurPASS3 electro kinetic analyzer for solid surface analysis (Anton Paar GmbH, Austria). The streaming potential is converted to zeta potential using the Helmholtz–Smoluchowsi equation:

$$\zeta =\frac{{{dU}}_{{\rm{str}}}}{{dp}}\frac{\eta }{{\varepsilon }_{{\rm{r}}}\cdot {\varepsilon }_{0}}\frac{L}{A\cdot R}$$
(3)

where ζ is the zeta potential, Ustr is the streaming potential, p is the applied hydraulic pressure, η is the liquid viscosity, εr is the liquid permittivity, ε0 is the permittivity in vacuum, R is the electrical resistance across the medium, and L and A are the length and cross-sectional area of the channel, respectively.

Four chloride-based single salt solutions (LiCl, NaCl, KCl, and CsCl) at concentrations of 0.1, 0.2, 0.5, 1, 2, 10, 50, and 75 mM were used during the ZP measurements. For each solution, stabilization of the measurements was first obtained at the initial pH (i.e., around pH 6.0). Next, the pH was elevated to pH 9.0 by injecting NaOH and gradually reduced towards pH 2.5, while measuring the ZP at intervals of around 0.5 pH unit. All experiments were performed at room temperature (23 ± 3 °C) with a continuous degassing of carbon dioxide from the solution by bubbling N2 gas into the beaker from which the solution is pumped to the measuring cell. The cell gap was adjusted to 100 ± 10 um and the following criteria were verified for ensuring reliable results: correlation coefficient (linearity of the streaming potential vs pressure ramp) >0.989 [-], absolute value of Asymmetry <10 mV, and static resistance <104 kOhm.

X-ray photoelectron spectroscopy measurements

X-ray photoelectron spectra were acquired with a Kratos Supra X-ray photo electron spectrometer. Dow NF270 membrane samples were soaked for 24 h in DI water or 1 mM NaCl solution at 40 °C, rinsed with DI water, and then soaked in DI water at room temperature for at least 6 h. Prior to testing, samples were left to dry in ambient conditions overnight. Survey spectra were used to calculate the element atomic percent on the surface and high-resolution spectra provided information on elemental chemical state. The Al K alpha X-ray source was operated at 1486.69 kV and 15.00 mA current emission. Survey spectra were acquired from 1200 eV to 0 eV at a resolution of 160, while high resolution scan ranges were based off the element being analyzed and a resolution of 20.

Nanofiltration tests

The experimental setup used for the permeability and rejection tests is operated in a crossflow mode and consists of a 10 L feed tank, high pressure pump, and three custom-made stainless steel cells accommodating the 15.3 cm2 flat-sheet membrane coupons. The reject and permeate streams of the three cells were continuously drawn back to the feed tank. The applied pressure, crossflow velocity, feed pH, and temperature were continuously monitored and maintained constant unless otherwise mentioned. In each experiment, following the alteration of the feed property, i.e., temperature, salt concentration, or pH, the system was allowed to stabilize for 30 minutes before permeate samples were taken. Prior to each experiment, the membrane coupons were compacted overnight with DI water by applying a pressure higher than the operation pressure by at least 2 bar.

Single-salt solutions were used for each experiment, so that salt concentrations in the feed and permeate streams (CF and CP, respectively, mol L−1) were deduced from electrical conductivity measurements (Eutech Instruments, CON2700). In addition, the weight of the permeate stream collected over a given time was used for calculating JW (L m−2 h−1), the water flux through the membrane, from which Js (=JW ∙ Cp), the salt flux, was also calculated. Concentration polarization was taken into consideration using the film theory77,78:

$${C}_{{\rm{m}}}=\left({C}_{{\rm{f}}}-{C}_{{\rm{p}}}\right)\exp \left(\frac{{J}_{{\rm{W}}}}{k}\right)+{C}_{{\rm{p}}}$$
(4)

where, Cm, (mol L−1) is the salt concentration on the membrane wall of the feed side; k is the mass transfer coefficient (m s−1) computed based on Sherwood correlation for turbulent flow in rectangular channel79 and parameters in Supplementary Table 2. Water permeability, PW, and salt permeability, Ps (m s−1), were calculated for each membrane cell using the solution-diffusion model:

$${P}_{{\rm{W}}}=\frac{{J}_{{\rm{W}}}}{\triangle P-\triangle \pi }$$
(5)
$${P}_{{\rm{S}}}=\frac{{J}_{{\rm{s}}}}{{C}_{{\rm{m}}}-{C}_{{\rm{p}}}}$$
(6)

where ΔP and ΔΠ are the pressure gradient and osmotic pressure gradient across the membrane, respectively.