Donnan dialysis for phosphate recovery from diverted urine

There is a critical need to shift from existing linear phosphorous management practices to a more sustainable circular P economy. Closing the nutrient loop can reduce our reliance on phosphate mining, which has well-documented environmental impacts, while simultaneously alleviating P pollution of aquatic environments from wastewater discharges that are not completely treated. The high orthophosphate, H x PO ( 3 (cid:0) x )(cid:0) 4 , content in source-separated urine offers propitious opportunities for P recovery. This study examines the use of Donnan dialysis (DD), an ion-exchange membrane-based process, for the recovery of orthophosphates from fresh and hydrolyzed urine matrixes. H 2 PO (cid:0) 4 transport against an orthophosphate concentration gradient was demonstrated and orthophosphate recovery yields up to 93% were achieved. By adopting higher feed to receiver volume ratios, DD enriched orthophosphate in the product stream as high as ≈ 2.5 × the initial urine feed concentration. However, flux, selectivity, and yield of orthophosphate recovery were detrimentally impacted by the presence of SO 42 (cid:0) and Cl (cid:0) in fresh urine, and the large amount of HCO 3 (cid:0) rendered hydrolyzed urine practically unsuitable for P recovery using DD. The detrimental effects of sulfate ions can be mitigated by utilizing a monovalent ion permselective membrane, improving selectivity for H 2 PO (cid:0) 4 transport over SO 42 (cid:0) by 3.1 × relative to DD with a conventional membrane; but the enhancement was at the expense of reduced orthophosphate flux. Critically, widely available and low-cost/waste resources with sufficiently high Cl (cid:0) content, such as seawater and waste water softening regenerant rinse, can be employed to improve the economic viability of orthophosphate recovery. This study shows the promising potential of DD for P recovery and enrichment from source-separated urine.


Introduction
Global food security is indispensably dependent on the sufficient supply of bioavailable phosphate for fertilizers. The growing global population is projected to drive an ≈ 50− 85% increase in phosphate fertilizer demand by 2050 (Mogollóna et al., 2018). At the same time, the prevailing practice for phosphorus (P) fertilizer production, phosphate rock mining, relies on finite and diminishing deposits (reserves are predicted to last only 50− 100 years, with production projected to decline after 2033) (Elser and Bennett, 2011;Smil, 2000). Furthermore, phosphate rock mining and beneficiation require significant energy inputs of 0.80− 1.66 kWh/kg-P (2002; Reta et al., 2018). In addition to the substantial energy required for industrial P production, downstream anthropogenic waste streams require further chemical and energy inputs for phosphate management (Bleiwas, 2011;Maurer et al., 2003;Schaubroeck et al., 2015). On average, a human excretes 820− 1200 mg-P daily, with 67% in urine and the remainder in feces (H. Jönsson, 2004;Karak and Bhattacharyya, 2011;Mihelcic et al., 2011). But due to the high costs, wastewater treatment plants (WWTPs) in the U.S. are not commonly equipped with advanced treatments dedicated to phosphorous removal (Larsen et al., 2009). Without adequate elimination at WWTPs, the nutrient is discharged into aquatic ecosystems, which results in eutrophication, harmful algal blooms, and hypoxic dead zones (Anderson et al., 2008;Conley et al., 2009;Diaz and Rosenberg, 2008;Michalak et al., 2013). These ecotoxic environments harm aquatic organisms and can pose public health threats from algal and cyanobacterial toxins in recreational waterbodies and drinking water supplies (Brooks et al., 2016;Hitzfeld et al., 2000). For these reasons, the biogeochemical flow of phosphate is flagged as exceeding the safe operating space for humanity and poses high risks under the planetary boundaries framework (Steffen et al., 2015). The current approach for P management has clear shortcomings and a new paradigm that is more sustainable is urgently needed (Guest et al., 2009;Li et al., 2015;Verstraete et al., 2009).
There has been considerable efforts to separate phosphorus from waste streams at WWTPs, (Blackall et al., 2002;Larsen et al., 2007;Oehmen et al., 2007;Yeoman et al., 1988) but the focus of most methods is P removal without capture. Such approaches can mitigate the downstream environmental problems, but still operate as an inefficient take-make-dispose linear economy model. Instead, orthophosphate, , can be simultaneously removed and recovered from anthropogenic wastewaters within a circular economy model, specifically, phosphorous is captured from wastewater and recycled back into the food chain (Commoner, 1971;McCartney et al., 2020;Stahel, 2016;W. McDonough, 2002;Webster, 2015). Closing the nutrient loop can ease the demand for phosphate mining to strengthen food security and alleviate P pollution of aquatic environments, thereby represents a transformative evolution to a more sustainable approach for phosphorous management.
The theoretical minimum energy for P recovery, governed by thermodynamic principles, is substantially lower for the feed of sourceseparated urine (i.e., the liquid is diverted away from black water and isolated) compared with other wastewaters (McCartney et al., 2021). For instance, recovery of orthophosphate from urine is ≈13− 34% less energy-intensive than treated wastewater effluent (McCartney et al., 2021). This is because urine is rich in P (total orthophosphate, TOP = 19− 48 × 10 − 3 mol/L), (Fittschen and Hahn, 1998;Larsen et al., 2013;Simha and Ganesapillai, 2017;Udert et al., 2003a) whereas treated wastewater effluent is over 1− 2 orders of magnitude more dilute. There has been considerable efforts to study the recovery of orthophosphates from urine and significant strides have been made, with the primary approach being mineral precipitation of slow-release fertilizers, such as struvite and magnesium potassium phosphate Maurer et al., 2006;Rahman et al., 2014;Randall and Naidoo, 2018;Ronteltap et al., 2007b;Udert et al., 2003aUdert et al., , 2003bWang et al., 2005). However, the technique has potential risks for contamination by pharmaceuticals, endocrine disrupting compounds, and pathogens (Boer et al., 2018;Kemacheevakul et al., 2015;Lahr et al., 2016;Mullen et al., 2017;Ronteltap et al., 2007a;Tang et al., 2019). Alternatively, sorbents, e.g., metal (oxy)hydroxide, (Karageorgiou et al., 2007;Lu et al., 2013;Tanada et al., 2003;Xie et al., 2014;Zhang et al., 2009) can be used to separate orthophosphate from other constituents in urine. But sorbent regeneration is chemically costly and the technique is further disadvantaged by the unavoidable generation of waste brines that cannot be easily disposed (Maul et al., 2014).
Ion exchange membrane (IEM) processes offer viable options for orthophosphate separation from urine without being encumbered by the above limitations. Donnan dialysis (DD) using IEMs has demonstrated promising potential for orthophosphate separation from other wastewaters besides urine, achieving high recovery yields (up to 98.4%) (Chen et al., 2014;Shashvatt et al., 2021;Trifi et al., 2009). However, the previous studies examined orthophosphate recovery from solutions containing only orthophosphate anions, i.e., no other co-ions are included in the water chemistry (Chen et al., 2014;Shashvatt et al., 2021;Trifi et al., 2009). Actual waste streams, including urine, are complex water matrices with many anions, cations, and uncharged species. In particular, the ionic composition of urine is uniquely different from, say, raw sewage or treated secondary effluent, with significantly greater amounts of SO 4 2− , Cl − , and HCO 3 − . Based on the general understanding of Donnan equilibrium theory, the presence of other anions in addition to orthophosphate is expected to have non-negligible and potentially detrimental effects on the process. In a past DD study, the presence of co-ions was shown to have deleterious impacts on arsenate removal from groundwater (Zhao et al., 2010), underscoring the need to better understand these effects to enable P recovery from urine using DD.
In this study, the performance of Donnan dialysis to drive the separation and recovery of orthophosphate from diverted human urine is assessed. First, the working principles of anion exchange membranes and Donnan dialysis are introduced. The exchange of H 2 PO 4 − and Cl − across an anion-exchange membrane is then examined with DD experiments driven by the ion concentration gradients between an orthophosphate feed solution and a receiver solution of high chloride content. Next, the impact of receiver solution chloride concentration on P recovery is analyzed using simulated streams of brackish water, seawater, and desalination brine. The capability of DD to enrich orthophosphate in the receiver solution above the initial urine feed concentration is evaluated by adopting higher feed to receiver volume ratios. Then, the influence of other anions in urine on orthophosphate flux and recovery yield is studied, specifically, SO 4 2− and Cl − in fresh urine, and SO 4 2− , Cl − , and HCO 3 − in hydrolyzed urine. Enhancements in selectivity for orthophosphate transport over other anions using a monovalent ion permselective membrane are assessed and the potential of DD to harvest fertilizer products of aqueous orthophosphate solutions is investigated. Finally, implications of Donnan dialysis for P separation and recovery from urine are discussed.

Ion exchange in Donnan dialysis
In Donnan dialysis, DD, flux of driver ions down a concentration gradient, across an ion-exchange membrane (IEM), sets up an electrochemical potential to drive the transport of target ions in the opposite direction (Sarkar et al., 2010;Strathmann, 1995;Tanaka, 2015). IEMs are polymeric thin-films with a high density of charged functional groups to allow the selective permeation of oppositely-charged counterions, while retaining like-charged co-ions (cation and anion exchange membranes, CEMs and AEMs, are selective for cation and anion transport, respectively) (Strathmann, 2004(Strathmann, , 2010. The IEM separates the feed solution (FS) and receiver solution (RS). Driver counterions permeate from the higher concentration RS to the lower concentration FS. Because co-ions are rejected by the charge-selective membrane, target counterions in the FS migrate across the membrane in the opposite direction, into the RS, to preserve electroneutrality (Asante-Sackey et al., 2021;Chen et al., 2014;Hichour et al., 2000;Sarkar et al., 2010;Shashvatt et al., 2021;Velizarov, 2013;Yan et al., 2018), i.e., the electrochemical potential gradient across the IEM drives the exchange of counterions between the FS and RS in DD, with co-ions retained in the initial solutions (Sarkar et al., 2010). Note that the charge fluxes of driver and target ions must be equal to maintain electroneutrality. Importantly, an adequately large driver ion concentration difference across the membrane can enable migration of target ions against a concentration gradient, to achieve "uphill" transport (Tanaka, 2015).
In DD recovery of orthophosphates, H x PO target ions from the FS with driver anions in the RS (Chen et al., 2014;Shashvatt et al., 2021;Trifi et al., 2009). Chloride is a suitable driver ion because Cl − is present in high concentrations in low-cost or waste streams that are widely available, such as brackish water, seawater, desalination brine, and waste effluent from water softening regeneration. Critically, Cl − concentrations in these streams are significantly higher than in urine (>200 × 10 − 3 mol/L, compared to ≈100 × 10 − 3 mol/L), to establish a sufficiently large electrochemical potential gradient across the AEM for uphill transport of H x PO (3− x)− 4 . Therefore, this study will investigate DD recovery of orthophosphates using chloride as the RS driver counterion.

Orthophosphate recovery at Donnan equilibrium
Target and driver counterions in the FS and RS exchange until Donnan equilibrium is reached, i.e., electrochemical potential gradient across the membrane = 0. For the system with H x PO and Cl − as target and driver ions, respectively, concentrations in the FS and RS at Donnan equilibrium are governed by Eq (1) (derivation is detailed in eqs S1-4 of the Supplementary Material): where subscripts FS and RS denote feed and receiver solutions, respectively, and f signifies final equilibrium. Note that Eq. (1) assumes perfect cation rejection and negligible water transport across the AEM.
To determine final target and driver ion concentrations at Donnan equilibrium using known initial concentrations, the principles of electroneutrality and material balances can be applied and are represented by eqs S5-8 of the Supplementary Material. For equivalent FS and RS volumes (i.e., as the sole initial anion in the FS, and Cl − as the sole initial anion in the RS, Eq. (2) describes the relationship between initial and final concentrations at Donnan equilibrium: ] RS,f can be determined using initial concentrations

Materials and chemicals
Commercial anion exchange membranes of Selemion AMV and Selemion ASVN, a monovalent ion-selective AEM, were procured from Asahi Glass Co. (Japan). Four DD cells with different chamber volumes were fabricated using Mars Pro-MSLA 3D Printer acquired from Elegoo (China). The solutions in each chamber were constantly stirred using magnetic stirring bars. The feed and receiver chamber volumes, respectively, are: 20 and 20 mL; 40 and 20 mL; 80 and 20 mL; and 40 and 40 mL, for feed to receiver solution volume ratios of 1:1, 2:1, 4:1, and 2:2, respectively. The effective membrane areas in all cells are 9.0 cm 2 . Na 2 HPO 4 ⋅7H 2 O, Na 3 PO 4 ⋅12H 2 O, NaCl, MgCl 2 ⋅6H 2 O, MgSO 4 ⋅7H 2 O, and NH 4 Cl salts were obtained from Alfa Aesar (Ward Hill, MA). KCl salt, 85% H 3 PO 4 solution, and 2.5 M H 2 SO 4 were acquired from Lab Chem (Zelienople, PA), Fisher Scientific (Waltham, MA), and Titripur (St. Louis, MO), respectively. All chemicals are ACS grade and were used as received. All solutions were prepared using deionized (DI) water from a Milli-Q ultrapure water purification system (Millipore Co., Burlington, MA).

Characterization of orthophosphate recovery at Donnan equilibrium
Orthophosphate, sulfate, and bicarbonate transport from FS to RS and chloride transport from RS to FS were evaluated in different DD operations. Three FS were utilized in orthophosphate recovery experiments: i. feed solution with the total orthophosphate, TOP, concentration and pH of fresh urine, (Fittschen and Hahn, 1998;Larsen et al., 2013;Simha and Ganesapillai, 2017;Udert et al., 2003a) but without other anions (16.5 × 10 − 3 mol/L Na 2 HPO 4 ⋅7H 2 O and 13.5 × 10 − 3 mol/L H 3 PO 4 ; pH = 6.0), ii. simulated fresh urine with the TOP, sulfate, and chloride concentrations, and pH of actual fresh urine (Fittschen and Hahn, 1998 , transport in the opposite direction, i.e., from FS to RS. The charge fluxes of driver and target ions must be equal to maintain electroneutrality, i.e., [ (2) S.N. McCartney et al. et al., 2003a) (24 × 10 − 3 mol/L Na 2 HPO 4 ⋅7H 2 O, 6 × 10 − 3 mol/L Na 3 PO 4 ⋅12H 2 O, 100 × 10 − 3 mol/L NaCl, and 16 × 10 − 3 mol/L H 2 SO 4 ), and iii. simulated hydrolyzed urine with the TOP, sulfate, chloride, and bicarbonate concentrations, and pH of actual hydrolyzed urine (Fittschen and Hahn, 1998;Udert et al., 2003a) (30 × 10 − 3 mol/L Na 2 H-PO 4 ⋅7H 2 O, 100 × 10 − 3 mol/L NaCl, 144 × 10 − 3 mol/L Na 2 CO 3 , and 106 × 10 − 3 mol/L NaHCO 3 ), as presented in Table 1. All solutions simulated the TOP concentration of undiluted urine (i.e., not mixed with flush water). Note that hydrolyzed urine is formed when urea, CO(NH 2 ) 2 , in urine hydrolyses to form ammoniacal nitrogen and bicarbonate, which in turn increases the pH from ≈6.0 to ≈9.2 (Udert et al., 2003a(Udert et al., , 2003b. At those pHs, the predominant orthophosphate species in simulated fresh and hydrolyzed urine solutions are H 2 PO 4 − and HPO 4 2− , respectively. Three receiver solutions with NaCl concentrations of 200 × 10 − 3 mol/L, 600 × 10 − 3 mol/L, and 1000 × 10 − 3 mol/L were utilized to simulate brackish water, seawater, and seawater desalination brine, respectively. DD cells with feed chamber volumes of 20, 40, and 80 mL were employed for orthophosphate recovery experiments with different feed to receiver solution volume ratios. In summary, FS anion composition and RS Cl − concentration were the parameters assessed in different FS to RS volume ratios. Unless stated otherwise, AMV membranes were used in the experiments. Pressure and temperature were at ambient conditions in all experiments. Total orthophosphate, sulfate, and bicarbonate concentrations in the FS and RS, and chloride concentration in the FS were sampled and analyzed using ion chromatography. The measurement intervals were 12, 24, or 48 h for experiments with FS to RS volume ratios of 1, 2, and 4, respectively, with longer time allotted for experiments with higher ratios to approach equilibrium. The experiments were considered to have effectively equilibrated when the moles of exchanged orthophosphate ions remained consistent over three consecutive measurements (< 5% change); experimental duration is < ≈96 h for the longest run (FS to RS volume ratio of 4). Note that the sampling times are relatively long because the solution volumes are large compared to the effective membrane area. Additionally, the equilibration is asymptotic and, hence, the ions exchange rates slow significantly as the process approaches equilibrium. Water transport was assessed by measuring feed and receiver solution volumes at the end of each experiment. Recovery yield, Y, is defined as the moles of TOP in the final RS normalized by the moles in the initial FS.

Evaluation of anion transport kinetics
Anion fluxes from FS to RS were examined in DD kinetic experiments, i.e., final Donnan equilibriums were not reached in the tests. Fluxes were determined from the rate of change of anion concentrations in the RS over 2.0 and 8.0 h for the AMV and ASVN membranes, respectively. The sampling durations were selected for sufficient ions to permeate across the small membrane areas of the benchscale setup such that the solution concentrations were above detection limits of the IC (the ASVN experiments required more time because anion fluxes are relatively lower). Water transport across the AEMs during the kinetic experiments was not observed and deemed to be practically negligible. The change in moles of anion in the RS over time normalized by the membrane area yields flux, J i . Flux selectivity, J i /J T , is defined as J i normalized by the sum of all anion fluxes from feed to receiver solution.
Anion fluxes and flux selectivities were assessed for the AMV membrane using 600 × 10 − 3 mol/L NaCl RS with the three FS of TOP-only solution, simulated fresh urine, and simulated hydrolyzed urine. J i and J i /J T were determined for the ASVN membrane using simulated fresh and hydrolyzed urine as FS and RS of 600 × 10 − 3 mol/L NaCl. To demonstrate the potential of DD to recover orthophosphate and yield aqueous products, experiments were conducted using the ASVN membrane, simulated fresh urine as FS, and four different RS. The RS investigated are: i. 600 × 10 − 3 mol/L NaCl, ii. 600 × 10 − 3 mol/L KCl to simulate aqueous potash fertilizer, iii. waste water softening regenerant rinse, WWSRR (547 × 10 − 3 mol/L KCl and 48 × 10 − 3 mol/L MgCl 2 ⋅6H 2 O), iv. and simulated diluted bittern, DB (252 × 10 − 3 mol/L MgCl 2 ⋅6H 2 O, 77 × 10 − 3 mol/L KCl, 25 × 10 − 3 mol/L MgSO 4 ⋅7H 2 O, 26 × 10 − 3 mol/L NaCl, and 8 × 10 − 3 mol/L NH 4 Cl). Further information on the composition of WWSRR and DB can be found in the Supplementary Material. Table S5 of the Supplementary Material summarizes the conditions for experiments presented in the following Figures.

Driver ion concentration gradient drives transport of orthophosphate across the anion exchange membrane
Donnan dialysis experiments were conducted using an initial FS of 30 × 10 − 3 mol/L H 2 PO 4 − and RS of 600  at pH = 6.0 (16.5 × 10 − 3 mol/L Na 2 HPO 4 ⋅7H 2 O and 13.5 × 10 − 3 mol/L H 3 PO 4 ) and 600 × 10 − 3 mol/L NaCl, respectively. Both feed and receiver solution volumes are 20 mL and effective AMV membrane area is 9.0 cm 2 . (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) ] RS,f at Donnan equilibrium calculated using Eq (2) (patterned and empty blue columns, respectively). Labels above these columns denote percentage ] RS,f relative to the prediction at Donnan equilibrium.
Theoretical orthophosphates captured in the receiver solution at Donnan equilibrium increase with greater feed concentrations of driver ] RS,f by 3.7%, lower than the theoretical enhancement of 11.3%.
[H x PO ] RS,f for the three different [Cl − ] RS,0 are lower than theoretical final concentrations due to dilution by simultaneous water transport from FS to RS and imperfect cation exclusion by the AEM. The high concentration of NaCl in the RS generates an osmotic pressure gradient that drives water permeation from FS to RS, i.e., osmosis. Additionally, electro-osmosis, where water molecules are dragged along with the permeating charged ions, contributes to water transport (J. Veerman et al., 2009;Spiegler, 1958;Yip and Elimelech, 2012).
flux from FS to RS is opposite in direction to water transport by Cl − flux (and Na + flux, explained next) from RS to FS. Net electro-osmosis is <10% of the measured water permeation; therefore, the contribution from the phenomenon is relatively minor and osmosis is the primary driver of water transport. The RS volume increased by 5.3-21% at the end of the DD experiments performed here. Water flux dilutes the RS and consequently lowers the TOP concentration. Note that, in contrast, minimal water transport was measured for the experiment presented in Fig. 2 because the experiment duration was comparatively shorter (6.0 h instead of 72 h).
Additionally, co-ions are not completely rejected by ion-exchange membranes (Beck and Ernst, 2015;Pessoa-Lopes et al., 2016). The Na + concentration gradient set up by the high [NaCl] in the RS results in Na + permeation from RS to FS, i.e., co-ion (cation) leakage. To maintain electroneutrality, every Na + ion that leaks across the AEM compels one Cl − ion to permeate in the same direction; therefore H 2 PO 4 − and Cl − are transported below equimolar ratio and [ ] RS,f . Na + transport was not specifically characterized in the experiments, but was estimated to be in the approximate region of 5% of anion transport, based on the membrane permselectivity ] RS,f experimentally captured in the RS. B) Experimental and predicted orthophosphate recovery yields, Y, (patterned and empty green columns, respectively) in DD with the different ] RS,f and accounting for osmotic water flux the FS to the RS. Labels above the columns signify the percent decrease in H x PO (3− x)− 4 experimentally captured in the RS relative to theoretical prediction (based on Eq. (2)). Initial feed solution contains 30 × 10 − 3 mol/L H 2 PO − 4 at pH = 6.0 (i.e., same composition as the FS of Fig. 2). Receiver solutions with initial concentrations of 200, 600, and 1000 × 10 − 3 mol/L NaCl to simulate brackish water, seawater, and seawater desalination brine, respectively, were investigated. In all operations V FS /V RS is 1:1. All experiments utilized AMV membranes. Data points and error bars are means and standard deviations, respectively, of duplicate experiments. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) of ≈0.95 reported in the manufacturer's specifications. Both water and co-ion leakage are not accounted for in Eq. (2) and cause the experi- ] RS,f to deviate below the predicted Donnan equilibrium. Note that transport of Cl − from RS to FS and H 2 PO 4 − from FS to RS were nearly identical during the 6 h of DD operation presented in Fig. 2; however, given the relatively longer duration of the phosphate recovery experiments presented in Fig. 3A (72 h), cumulative co-ion transport is more pronounced. Additionally, the ballpark estimate of cation leakage is consistent with the deviations between experimental and predicted orthophosphate concentrations. ] RS,f of [Cl − ] RS,0 = 1000 mol/L being marginally smaller than [Cl − ] RS,0 = 600 mol/L is likely attributed to inevitable experimental uncertainties inherent to the measurement techniques). The greater deficits are due to both water permeation from FS to RS (quantified by measuring changes in the solution volumes) and cation leakage from RS to FS being more prominent in operations with higher [Cl − ] RS,0 . Osmotic pressure is essentially proportional to [NaCl] and are 9.91, 29.7, and 49.6 bar for 200, 600, and 1000 × 10 − 3 mol/L NaCl, respectively. As a result of the greater osmotic pressure difference between the RS and FS, increased water permeation is observed when [Cl − ] RS,0 is higher; net water fluxes are 1.93 × 10 − 2 , 6.37 × 10 − 2 , and 7.10 × 10 − 2 L/m 2 h for initial RS of 200, 600, and 1000 × 10 − 3 mol/L NaCl, respectively. Additionally, Na + leakage is more exacerbated with conditions of higher initial RS [NaCl] due to increased Na + concentration gradients. (Beck and Ernst, 2015;Pessoa-Lopes et al., 2016) Consequently, the undesired co-permeation of Cl − is elevated. In principle, [Cl − ] FS,f (gray columns in Fig. 3A) should be equivalent to ] RS,f are larger at higher [Cl − ] RS,0 as co-ion leakage and osmosis are more exacerbated. Although increasing [Cl − ] RS,0 resulted in higher TOP concentrations recovered in the receiver solution, the greater deviations between experimental and predicted DD performance also indicate that the inefficiencies of water and co-ion leakage are more pronounced.
Recovery yield, Y, is defined as the percentage of orthophosphate moles from the initial FS that are recovered in the RS and is presented in Fig. 3B, with patterned and empty green columns denoting experimental and predicted Y, respectively. Experimental Y is calculated using the product of TOP concentration and solution volume at the end of the test ] RS,f V RS,f ), i.e., the effects of osmotic and electro-osmotic water transport are accounted for. Labels above the columns are the differences between experimental and predicted Y. The experimental orthophosphate recovery yields are very close to Ys predicted at Donnan equilibrium using Eq. (2) (within 5%). The minor shortfalls in recovery yield are attributed to Na + leakage and Cl − co-permeation. Despite the imperfect permselectivity of the AEM, experimental Ys are high and exceed 90% for [Cl − ] RS,0 of 600 × 10 − 3 and 1000 × 10 − 3 mol/L. Even the lowest [Cl − ] RS,0 of 200 × 10 − 3 mol/L produced Y of 83%. The high yields achieved signify that DD can be a promising technique for orthophosphate recovery. Increasing [Cl − ] RS,0 improves Y, but with diminishing returns. For example, raising [Cl − ] RS,0 from 200 × 10 − 3 to 600 × 10 − 3 mol/L increases experimental Y by 9.5%. However, further increasing [Cl − ] RS,0 to 1000 × 10 − 3 mol/L only marginally improved the orthophosphate recovery yield by 1.4% (this trend is also predicted by the Donnan equilibrium theory of Eq. (2)). Subsequent phosphate recovery experiments will, thus, utilize [Cl − ] RS,0 = 600 × 10 − 3 mol/L.

Donnan dialysis can enrich orthophosphate in the receiver solution
It is advantageous to produce concentrated orthophosphate solutions to be utilized in downstream applications as fertilizer. Orthophosphate   ] RS,f V RS,f and, therefore, accounts for water transport from the FS to the RS due to osomosis and electro-osmosis. Predicted ] RS,f and recovery yields at Donnan equillibrium, calculated using eqs S14-17 in the Supplementary Material, are depicted as blue and green dashed lines, respectively. The initial feed and receiver solutions are the same as the FS and RS of Fig. 2. For comparison, the initial total orthophosphate concentration in the simulated urine feeds is depicted as a dotted blue line.

Enrichment factors, defined as [H x PO
] RS,f V RS,f and, therefore, accounts for osmotic water flux from the FS to the RS. All experiments utilized AMV membranes. Data points and error bars are means and standard deviations, respectively, of duplicate experiments. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) detrimentally elevated with increased V FS /V RS and are − 15.7%, − 22.3%, and − 24.8% for V FS /V RS of 1, 2, and 4, respectively. However, similar to the results presented in Fig. 3, experimental recovery yields (green square symbols in Fig. 4) are comparable to theoretical Y at Donnan equilibrium (<3% difference between green square symbols and green dashed lines), as water permeation is accounted for in the determination of moles of TOP recovered. This minor difference is attributed to Na + leakage and Cl − co-permeation, i.e., imperfect AEM permselectivity, (Beck and Ernst, 2015;Pessoa-Lopes et al., 2016) which results in reduced driving force available for H x PO 4 (3− x)− transport.
Varying V FS /V RS produces a tradeoff between enrichment and recovery yield, as predicted by the Donnan equilibrium theory (eq S14-17): increasing V FS /V RS enhances the enrichment factor, but the recovery yield is slightly compromised (dashed blue line and dashed green line in Fig. 4 exhibit positive and negative slopes, respectively). This is because with a relatively smaller receiver solution volume, the initial ratio of moles of Cl − in the RS to moles of H x PO 4 (3− x)− in the FS (i.

Competing anions in the urine feed solutions lower orthophosphate transport and recovery
Results presented thus far utilized a feed solution with TOP concentration of fresh urine, i.e., orthophosphate as the sole anion. However, actual fresh urine contains orthophosphate, sulfate, and chloride anions, and hydrolyzed urine additionally has significant bicarbonate concentration (Fittschen and Hahn, 1998;Larsen et al., 2013;Simha and Ganesapillai, 2017;Udert et al., 2003a). To examine the effects of other anions on orthophosphate recovery, DD experiments were conducted using simulated fresh urine and hydrolyzed urine (anion compositions are summarized in Table 1), for comparison with the TOP-only solution. In all operations, the RS has 600 × 10 − 3 mol/L of Cl − . Fig. 5A   , whereas simulated hydrolyzed urine additionally has 250 × 10 − 3 mol/L HCO 3 − . Note that the transport of water from feed solution to receiver solution is accounted for in the experimental orthophosphate recovery yields. Predicted final RS concentrations and predicted orthophosphate recovery yields at Donnan equilibrium, calculated using eqs S14-17 in the Supplementary Material, are depicted as empty columns. The labels above the columns indicate the change relative to the orthophosphate-only FS. All experiments were operated with V FS /V RS equal to 2 and 600 × 10 − 3 mol/L NaCl as the receiver solution. All experiments utilized AMV membranes. Data points and error bars are means and standard deviations, respectively, of duplicate experiments. ] RS,f is significantly lower than in DD with TOP-only FS ( Figure S1 of the Supplementary Material).
Consistent with the Donnan equilibrium predictions, experimental ] RS,f in DD with fresh and hydrolyzed urine are 29.1% and 39.9% lower, respectively, relative to TOP-only feed solution (Fig. 5A) ] RS,f in DD with hydrolyzed urine resulted in considerably larger deviations from Donnan equilibrium values (− 49.4%) than with TOP-only feed solution and fresh urine (− 22.3% and − 21.0%, respectively). One possible explanation for these observations is that the hydrolyzed urine experiments were terminated prior to actual equilibrium. The experimental protocol used   Figure 5B presents experimental and predicted orthophosphate recovery yields, Y (patterned and empty blue columns, respectively); note that Y accounts for water transport (as discussed earlier). Labels above the arrows designate the percent reduction in experimental Y relative to orthophosphate-only FS. Recovery yield trends are generally similar to ] RS,f trends. Specifically, Y are 24.5% and 68.4% lower with fresh and hydrolyzed urine, respectively, relative to DD with TOP-only FS, which qualitatively agrees with theoretical predictions of 30.3% and 40.3% reduction. Experimental DD with TOP-only feed solutions and fresh urine deviate slightly from predicted Y at Donnan equilibrium (<3%). As previously discussed, Na + leakage from RS to FS results in less ion transport from FS to RS, which explains the experimental Y being slightly lower than the predicted value for DD with TOP-only FS. For fresh urine, co-ion leakage is lower because the Na + present in the FS reduces the Na + concentration gradient, which drives co-ion permeation. Therefore, the experimental and predicted Ys are comparable for DD with fresh urine (minor difference is due to unavoidable random fluctuations in experimental measurements). In contrast, Y is significantly lower than the theoretical recovery yield at Donnan equilibrium for hydrolyzed urine (50.0% lower). As conjectured in the preceding paragraph, the substantially greater discrepancy is possibly explained by the premature termination of the DD experiment before the ion concentrations were fully equilibrated.

Orthophosphate fluxes are diminished by the competing anions
Parallel DD kinetics experiments were conducted with the same initial feed solutions presented in Fig. 5, i.e., TOP-only FS, fresh urine, and hydrolyzed urine. To elucidate the impacts of individual ionic species in the urine matrix on ion fluxes, additional DD kinetic experiments were carried out with feed solutions of H x PO 4 (3− x)− + SO 4 2− and H x PO 4 (3− x)− + Cl − . In all operations, the RS has 600 × 10 − 3 mol/L of Cl − and V FS /V RS = 1. Fig. 6 shows molar ion fluxes, J i , of H x PO 4 (3− x)− , SO 4 2− , and HCO 3 − from FS to RS (subscript i is P, S, and C, respectively) as patterned blue, orange, and red columns, respectively; labels above blue columns indicate reduction in J P relative to DD with the orthophosphate-only FS.

As observed with experimental [H x PO
] RS,f and Y ( Fig. 5A and B), J P is reduced in DD with simulated fresh or hydrolyzed urine relative to orthophosphate-only FS (90.1% and 97.3% lower, respectively).  Table S6 and Fig. S2).
Instead, the observed differences in J P could possibly be explained by different affinities of the various ions to sorb into the anion exchange membrane. Specifically, if other anions in the FS, such as SO 4 2− and Cl − , more preferentially partition into the membrane than H x PO 4 (3− x)− , the relative concentration of orthophosphate ions within the IEM will be disproportionately lower compared to the feed solution (Malewitz et al., 2007). As flux is directly proportional to the ion concentration in the IEM, J P will consequently be depressed. Additionally, the higher charge of multivalent ions favors sorption into the charged membrane more than monovalent ions (Galizia et al., 2017 (details on the mechanisms of valence-selectivity can be found in literature) (Fan et al., 2022;Lu et al., 2011;Saracco, 1997;Saracco and Zanetti, 1994). Because the predominant orthophosphate species in fresh urine is H 2 PO 4 − , DD with the ASVN MIPM is expected to improve selectivity for H x PO 4 (3− x)− over SO 4 2− relative to the conventional AMV.
DD kinetic experiments were conducted with the AMV and ASVN membranes using simulated fresh urines as FS, 600 × 10 − 3 mol/L Cl − for RS, and V FS /V RS = 1. Fig. 7A presents the ion fluxes from FS to RS, J i , and Fig. 7B shows ion flux selectivity, defined as the molar ion flux normalized by the sum of two fluxes, J i /J T . H 2 PO 4 − and SO 4 2− fluxes are denoted by blue and orange columns, respectively, whereas patterned and solid columns signify AMV and ASVN, respectively. Similar to the results presented in Fig. 6, Fig. 7A shows J P < J S in DD with the AMV membrane. Flux selectivity J P /J T is 0.21 (Fig. 7B) .8% lower for J P and J S , respectively). The monovalent ion selective coating on the ASVN membrane is composed of highly crosslinked resin, which creates additional steric hindrance for permeating ions (Fan et al., 2022;Lu et al., 2011;Saracco, 1997;Saracco and Zanetti, 1994). Consequently, the overall ASVN membrane has greater resistance for ion transport compared to the AMV membrane (this is reflected in the manufacturer's specifications on resistance with various salt solutions, summarized in Table S7). Thus, there exists a tradeoff between ion selectivity and permeability: using MIPMs instead of conventional AEMs yields improved monovalent ion selectivity but reduced ion fluxes. In applications of DD for orthophosphate recovery, MIPMs can improve TOP yield, but at the expense of slower kinetics; a lower J P would necessitate larger membrane areas for the same productivity of orthophosphate recovery.

Orthophosphate can be captured as a fertilizer solution using Donnan dialysis
The previously presented results utilized receiver solutions with high Cl − concentrations to demonstrate the potential of Donnan dialysis to separate orthophosphate from urine, which can have contaminants, e.g., pharmaceuticals, endocrine disrupting compounds, and opportunistic pathogens, that prohibit the direct application of urine for fertilization. Orthophosphate was captured in receiver streams of NaCl solutions for the experiments. However, Na + is not well tolerated by most plant species at concentrations > 4.3-13 × 10 − 3 mol/L, (Genc et al., 2007;Kronzucker et al., 2013) i.e., the eventual DD receiver solutions contain too much sodium. One option to exclude Na + is to precipitate phosphate fertilizers as mineral solids, such as struvite, NH 4 MgPO 4 ⋅6H 2 O, and potassium magnesium phosphate, KMgPO 4 ⋅6H 2 O, from the TOP-enriched RS by adding Mg 2+ and NH 4 + or K + . As highlighted in the Introduction section, some studies reported the presence of , SO 4 2− , and HCO 3 − anion fluxes from FS to RS, J i , in kinetic experiments with different urine matrices as initial feed solutions. The anions of the five FS are: i) orthophosphate only, ii) orthophosphate and sulfate, iii) orthophosphate and chloride, iv) orthophosphate, sulfate, and chloride (fresh urine), and v) orthophosphate, sulfate, bicarbonate, and chloride (hydrolyzed urine). Labels above the columns indicate the change in flux relative to the orthophosphate-only feed solutions.
[H x PO contaminants in phosphate minerals directly precipitated from urine (Boer et al., 2018;Kemacheevakul et al., 2015;Lahr et al., 2016;Mullen et al., 2017;Ronteltap et al., 2007a;Tang et al., 2019). The anion exchange membrane in DD serves as a barrier, retaining the contaminants of concern in the urine feed (Arola et al., 2019;Banasiaka et al., 2011;Ma et al., 2021aMa et al., , 2021bPronk et al., 2006;Vanoppen et al., 2015). Therefore, utilizing Donnan dialysis to separate orthophosphate from the urine matrix prior to mineral precipitation can sidestep the issue of possible pollutants contaminating the fertilizer products. Future studies need to be conducted to better understand the rejection of pharmaceuticals and other pollutants by AEMs. An alternative method to circumvent both the sodium and product contamination issues is to recover the orthophosphates in DD receiver solutions with little or no Na + . In this approach, initial receiver solutions that contain sufficiently high concentrations of Cl − but with cations other than Na + are used. Additionally, the cations can be K + and NH 4 + to further valorize nutrient content of the fertilizer product. For instance, aqueous solutions of potash fertilizer, KCl (aq) , can be used as the RS (simulated as 600 × 10 − 3 mol/L KCl here), to draw additional value through the exchange of orthophosphate and Cl − . Another RS option is waste water softening regenerant rinse (WWSRR), which contains a high concentration of chloride anions and potassium and magnesium as the cations (approximately 547 × 10 − 3 mol/L K + , 48 × 10 − 3 mol/L Mg 2+ , and 643 × 10 − 3 mol/L Cl − ). Diluted bittern (DB), a concentrated byproduct of table salt production from seawater, is yet another alternative. DB contains approximately 77 × 10 − 3 mol/L K + , 252 × 10 − 3 mol/L Mg 2+ , 25 × 10 − 3 mol/L SO 4 2− , 8 × 10 − 3 mol/L NH 4 + , 615 × 10 − 3 mol/L Cl − , and 26 × 10 − 3 mol/L Na + . Note that the concentration of Na + in diluted bittern is sufficiently low and can be tolerated by some plant species, such as beets (Kronzucker et al., 2013;Lawlor and Milford, 1973). DD experiments were conducted using simulated aqueous potash solutions (APS), WWSRR, and DB, with the ASVN membrane, simulated fresh urine as FS, and V FS /V RS of 1. Fig. 8 presents fluxes of H x PO 4 (3− x)− and SO 4 2− from FS to RS.
In general, J P (and J S ) of the different simulated receiver solutions are comparable with the 600 × 10 − 3 mol/L NaCl RS. Similar J P are expected because all operations utilized the same FS and have fairly similar [Cl − ] RS,0 (< 7% difference). Importantly, DD with the monovalent ion selective membrane (ASVN) consistently achieved selectivity for H x PO 4 (3− x)− flux over SO 4 2− for all the simulated receiver solutions (J P /J T = 0.53-0.61). This highlights the flexibility of DD to utilize different Cl − -rich streams, including waste and low-cost sources, as the receiver solution for orthophosphate recovery from fresh urine. Additionally, the RS can be rationally selected to promote mineral precipitation, to enable recovery of solid orthophosphates. For instance, precipitated phosphate salts of calcium and magnesium can be further isolated from the liquid stream when Ca 2+ -and Mg 2+ -rich WWSRR is used as the RS.

Implications
The current P management practices are unsustainable. Phosphate is  columns denote the selectivity for orthophosphate, J P /J T . Horizontal line is orthophosphate flux in DD with 600 × 10 − 3 mol/L NaCl. All experiments utilized ASVN membranes. Data points and error bars are means and standard deviations, respectively, of duplicate experiments. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) mined and processed from diminishing reserves at immense energy costs. P-rich excretions are heavily diluted in the traditional wastewater system, which contributes to the high energy and chemical expenditures in efforts to remove the phosphates from wastewaters. When not adequately removed, P is released into the environment, causing ecological harm and public health concerns. Further compounding to these issues are environmental problems associated with phosphate fertilizer manufacturing, in particular, the radioactive byproducts that are generated. Improper storage and management of the radioactive wastes pose leakage risks (Nelson et al., 2021;Sandhu et al., 2018). These hazards were starkly exposed in March 2021 when the Piney Point, Florida, production plant leaked radioactive phosphogypsum into Tampa Bay because of an engineering failure in the aged infrastructure. Residents of the surrounding area were evacuated due to risks of contact with the contaminated waters (Nelson et al., 2021). The inefficient linear economy approach and this unfortunate incident underscore the need for a paradigm shift to a circular P economy with sustainable phosphate capture and reuse.
The high concentrations of orthophosphate anion in human urine offer propitious opportunities for recovery. Donnan dialysis can utilize driver ions to exchange for H x PO 4 (3− x)− in the urine feed, capturing P in the receiver solution for application as fertilizer. This study demonstrates that Donnan dialysis can i) recover orthophosphate from urine, ii) enrich orthophosphate in the receiver solution, iii) selectively capture orthophosphate over other anions by utilizing fresh urine and monovalent ion permselective membranes, and iv) leverage on widely available and low-cost/waste resources to drive orthophosphate recovery. Importantly, using receiver streams with adequately high driver ion concentrations in DD (Cl − in this study) can enable orthophosphate transport against a concentration gradient and attain "uphill transport" to reach practically feasible recovery yields (> 80% demonstrated in this investigation). By employing a smaller receiver solution volume relative to the feed, DD can achieve enrichment of orthophosphate, i.e., P concentration in the product is higher than in fresh urine. A fertilizer product with high orthophosphate concentrations is of greater economic value and additionally facilitates transportation. The analysis also reveals the rationale for using fresh, rather than hydrolyzed urine, for P recovery. Specifically, the high bicarbonate content of the latter source is detrimental to orthophosphate flux and selectivity. This indicates that DD orthophosphate recovery should be performed immediately after urine diversion and collection, prior to urea hydrolysis. Alternatively, dosing with inhibitory compounds, electrochemical treatment, or acid/ base addition can inactivate the urease enzyme, (Hellström et al., 1999;Ikematsu et al., 2007;Lv et al., 2020;Randall et al., 2016;Saettaab et al., 2020;Svane et al., 2020;Udert et al., 2003aUdert et al., , 2003b thus suppressing bicarbonate formation. Besides bicarbonate, sulfate can compete with orthophosphate anions to exchange with driver chloride ions, hence reducing recovery efficiency. The study highlights the applicability of using monovalent ion permselective membrane to drive more selective transport of orthophosphate over sulfate. However, the improvements in selectivity are at the cost of decreased kinetics. Therefore, the overall DD process would need to simultaneously consider TOP recovery yields and membrane requirements, i.e., tradeoffs between with economic benefits with capital and operating costs. Thoughtful selection of the receiver stream offers flexibility in tailoring the water chemistry and nutrient profile of the fertilizer product. Critically, economic viability of P recovery with DD can be enhanced by utilizing waste/low-cost streams as the receiver solution. For example, waste water softening regenerant rinse generated in residential buildings can be repurposed to supply the high chloride concentration needed in the receiver solution, to drive DD recovery of orthophosphate from diverted urine from the same premises. Future investigations will be needed to understand the potential impacts of other compounds in urine, such as pharmaceuticals, pathogens, and other contaminants, on DD performance and phosphate recovery (including membrane fouling).
The insights from this study are broadly applicable to other DD processes for resource recovery or contaminant removal, particularly for streams with multiple anions and/or cations. Some examples are NH 4 + recovery from wastewater, metal ion recovery from electroplating rinse, and removal of NO 3 − and H x AsO 4 (3− x)− from drinking water. The approach for determining ion concentrations in feed and receiver solutions at Donnan equilibrium presented here (eqs S9-13) can be utilized to project target ion recovery potential or contaminant removal efficiency from mixed electrolyte solutions. Furthermore, the systematic analysis of factors influencing fluxes of different ions underscores the role of competitive ion sorption on transport kinetics and can elucidate ion transport behavior in solutions with complex compositions. In applications where multivalent species are present together with the monovalent target ion, as in NH 4 + recovery from wastewater or NO 3 − removal from drinking water, MIPMs may be useful to improve selectivity for the target species with an acceptable sacrifice in permeation flux.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data Availability
Data will be made available on request.