Pilot-Scale Treatment of Neutral Pharmaceuticals in Municipal Wastewater Using Reverse Osmosis and Ozonation

High-performance liquid chromatography (HPLC) –grade methanol was obtained from Fisher Scientific (Oakville, Ontario). LC-MS–grade ammonium acetate and formic acid were purchased from Sigma Aldrich. Mixtures of reference standards for acetaminophen, caffeine, carbamazepine, ciprofloxacin hydrochloride, erythromycin, fluoxetine hydrochloride, sulfamethoxazole, and trimethoprim were obtained from Restek (Bellefonte, Pennsylvania). Individual reference standards for venlafaxine, cotinine, o-desmethylvenlafaxine, norfluoxetine, and metformin were obtained from Sigma-Aldrich. All reference standards utilized were greater than 98% purity. Isotope standards (above 98% purity) for metformin-${\mathrm{d}}_6$ hydrochloride, acetaminophen-${\mathrm{d}}_4$, caffeine-${\mathrm{d}}_3$ (10 mg), and fluoxetine-${\mathrm{d}}_5$ hydrochloride were obtained from Cayman Chemical (Ann Arbor, Michigan) and (±)-cotinine-${\mathrm{d}}_3$, caffeine-${{}^{13}\mathrm{C}}_3$, and carbamazepine-${\mathrm{d}}_{10}$ were purchased from Sigma–Aldrich. Stocks and working standards were prepared using LC-MS–grade methanol and stored in amber vials at $-20°\mathrm{C}$.

Experiments used for this study were conducted at the Advancing Canadian Wastewater Assets Research WWTP in Calgary, Alberta. The research WWTP is a globally unique research facility that is imbedded in a functioning municipal WWTP, using effluent from the secondary clarifiers as the input water. The full test facility has the capacity to process roughly $\mathrm{500,000}\mathrm{L}/$days. Treatment processes are designed to be run as a part of the active treatment train in the test facility, making data collected there the most direct comparison to full application. Prior to entering the research WWTP, municipal wastewater is collected in Calgary and treated at the WWTP by the following treatment train: screening and grit removal; primary clarification; three-phase biological treatment consisting of preanoxic, anaerobic, and anoxic zones; and secondary clarification. The active WWTP serves a municipal population of 250,000 with a capacity of up to $100\mathrm{ML}/$days. After secondary clarification, water can be diverted to the research WWTP, characteristics of which are provided in Table 1. Collection systems for the WWTP are by separate sewers. Experiments are performed during dry conditions, so minimal influence of infiltration is expected in the collections system.

Prior to treatment in the research plant, the wastewater is microfiltered utilizing tubular membranes with a nominal pore size of 0.02 μm and a total membrane area of $752.4{\mathrm{m}}^2$. After ultrafiltration there is no further modification of the wastewater, with tests being conducted at typical concentrations of PhACs found in it. Wastewater entering ozone and RO processes has a turbidity of $<0.2\mathrm{NTU}$ and a 5-day BOD $<5\mathrm{mg}/\mathrm{L}$. The temperature in the test facility during RO and ozone testing is 11°C–25°C.

Ozone experiments were conducted in an ozone contact tank with a total volume of 1,830 L, with a contact time of 15–20 min. A schematic of the ozone contact tank can be seen in Figs. 1(a and b). The influent flow rate to the ozone contact tank was maintained at $1.7\mathrm{L}/\mathrm{s}$. The ozone generated was mixed with air at a proportion of 4:96 (ozone:air), and the flow rate was adjusted to achieve a target aqueous ozone concentration of $50\mathrm{mg}/\mathrm{L}$. Ozone was produced through a process of air compression, drying, and oxygen concentration followed by corona discharge. The ozone was dissolved in wastewater by eductor prior to entering the contact tank. A process flow diagram can be seen in Fig. 2. For analysis of the ozone-treated wastewater, samples were taken before and after the ozone reactor. The ozone contact tank can be seen in Fig. 3.

RO experiments were conducted using a two-stage process. The arrangement contained two units in the first bank and one in the second bank. Each unit contained a spiral-wound RO membrane with an element diameter of 20.1 cm and a length of 101.6 cm. The total active surface area of each RO membrane was $37.2{\mathrm{m}}^2$. A high-pressure pump was used to pressurize the RO influent into the first RO unit. The result of the RO arrangement was a configuration analogous to a single unit with a length of 203.2 cm. The concentrate through the first RO bank was used as influent for the second RO bank. Samples were taken from sampling ports located before the RO system, as well as in the permeate and concentrate lines exiting the RO system. The RO module ran at a permeate flow rate of $7.08{\mathrm{m}}^3/\text{h}$. Following treatment in the RO system, the water was returned to the mixed liquor line for the activated sludge process of the functioning WWTP. The reverse osmosis unit can be seen in Fig. 4.

Samples were collected in 500-mL amber-glass bottles lined with PTFE caps. The sample bottles were rinsed in methanol and milli-Q water and dried in a hot-air oven at 80°C prior to collection of samples to minimize interferences. The samples were collected with minimal head space following recommended practices for trace contaminant analysis (Miao et al. 2005; USEPA 2007). The influent wastewater effluent was stored in a large tank of $3.79{\mathrm{m}}^3$ capacity prior to ozone and RO treatment. Triplicate samples were collected by grab sampling during and before and after treatment over a period of 30 min. As grab samples, it is noted that results represented a deamination of function at the specific time of testing, with potential fluctuations that were not accounted measurements of inlet concentration variability performed during this study. One sample bottle was filled with milli-Q water on-site at approximately the same rate as the actual samples were filled; it was used as blank sample to identify the possibility of contamination during sample collection. The samples were stored in 4°C immediately after collection and extracted within 48 h to avoid possible degradation.

The sample-processing protocol used by Chen et al. (2006) for analysis of neutral PhACs was adopted (Chen et al. 2006). All sample bottles were spiked with surrogate standard caffeine-${{}^{13}\mathrm{C}}_3$ (50 μL of $10\mathrm{\mu g}/\mathrm{mL}$) and weighed. The samples were concentrated by solid-phase extraction using 6 mL 200-mg Supel-Select HLB tubes on a Supelco Visiprep 24-port vacuum manifold (Sigma-Aldrich). Prior to extraction, the cartridges were conditioned with 8 mL methanol, followed by 4 mL milli-Q water. The samples were extracted at a flow rate of approximately $4\mathrm{mL}/\min$. The empty bottles were then reweighed to obtain the exact mass of each extracted sample. The cartridges were rinsed with 4 mL water and dried under vacuum for 1 min. Subsequently, the extracted contaminants were eluted using three aliquots of 4 mL methanol at a flow rate of approximately $1\mathrm{mL}/\min$. The extracts were concentrated by evaporation on a MICROVAP (Organomation Associates, Berlin, Massachusetts) under a gentle stream of nitrogen at 40°C. The concentrate was then reconstituted in 2 mL methanol and stored at $-20°\mathrm{C}$ for instrumental analysis. Prior to LC-MS analysis, 300 μL concentrate was mixed with 700 μL HPLC eluent and spiked with internal standard mix.

Sample analyses were performed using an Agilent 1260 HPLC coupled with an Agilent 6460 (Agilent Technologies, Santa Clara, California) triple quadruple mass spectrometer (QQQ) equipped with a Jet Stream electrospray ionization source (Agilent Technologies). Chromatographic separation of target analytes was performed using a Raptor ARC-18 (Restek) column ($100\times 2.1\mathrm{mm}$, 2.7 μm) fitted with a Raptor EXP guard column cartridge. A binary, gradient eluent system comprising 10 mM ammonium acetate/formic acid (0.1%) (A) and methanol (B) with a flow rate of $0.3\mathrm{mL}/\min$ was used as follows: (1) hold at 85% A for 2 min; (2) linear ramp to 45:55 (A:B) over 8 min; (3) hold at 45:55 (A:B) for 4 min; (4) linear ramp to 100% B over 1 min and hold for 5 min; and (5) linear ramp to 85% A over 1 min and re-equilibrate for 6 min. The total run time of the program was 25 min and the injection volume of the sample was 10 μL. The column compartment was maintained at 30°C throughout the run time. The Agilent 6460 MS instrument conditions were maintained as follows: drying gas temperature 180°C; drying gas flow $7\mathrm{L}{\min }^{-1}$; nebulizer pressure 25 psi; sheath gas temperature (positive 300°C; negative 300°C); sheath gas flow $12\mathrm{L}{\min }^{-1}$; capillary (positive 2,500 V; negative 0 V); nozzle voltage (positive 1,500 V; negative 0 V) delta EMV 200 V. The analysis was performed using dynamic multiple-reaction monitoring with individual time windows for every compound transition. The dynamic multiple-reaction monitoring transitions and associated fragmented and collision voltages applied for each compound are summarized in Table 2.

A seven-point calibration curve was established by injection of known calibration standards ranging from 1 to $100\mathrm{ng}/\mathrm{mL}$ for all compounds except ciprofloxacin. For ciprofloxacin, a five-point calibration curve ranging from 5 to $100\mathrm{ng}/\mathrm{mL}$ was used. All calibration injections were run in triplicate. For the sake of accuracy, one injection of all calibration standards was conducted daily. Instrument detection limits (IDLs) and instrument quantitation limits (IQLs) were determined according to the lowest concentration that gave a signal-to-noise ratio of $\ge 3$ and $\ge 10$, respectively. Quality control was maintained by analyzing the calibration standard of $25\mathrm{ng}/\mathrm{mL}$ at every 10 samples. Carryover between samples was monitored by injecting unspiked HPLC eluent at every 25 samples. Variations in concentration measured in quality control samples was less than 5%, and the carryover between samples was less than the IQL. Caffeine ${\mathrm{C}}_{13}$ was used as a surrogate standard to establish the fraction recovered during solid-phase extraction, which was 95%–105% in all cases. Error determination was performed based on variation in feed concentration, as feed variation is the largest source of variability within the study. Multiple inlet samples were taken across multiple run days, with one standard deviation within intraday concentration variability used for error bars in all figures.

The concentrations of all PhACs examined here were measured entering the RO unit, the permeate, and the reject concentrate. The percentage passing into the permeate of the RO system is presented in Fig. 5. The high removal efficiencies of PhACs in the RO system indicate that it is a highly effective treatment method for all compounds considered in the study except ciprofloxacin. The reduction in concentration between the inlet and outlet was determined to be 92.6%, 99.0%, 99.6%, 97.8%, 99.0%, 99.6%, 99.9%, and 99.2% for metformin, cotinine, trimethoprim, caffeine, venlafaxine, carbamazepine, erythromycin, and fluoxetine, respectively. Sulfamethoxazole was reduced to below the detection limit. The high removal rates demonstrated in this study are in line with those in other large-scale studies that have investigated the removal of PhACs from secondary municipal effluent. A comparable study of ambient concentrations of PhACs in a microfiltration-RO pilot plant produced comparable removal results of 97% for trimethoprim, 97% for sulfamethoxazole, 99% for carbamazepine, and 98% for erythromycin (Rodriguez-Mozaz et al. 2015). Additionally, other pilot-scale studies on the removal of PhACs by RO have shown 91% for acetaminophen, 90% for carbamazepine, and 99.5% for caffeine (Al-Rifai et al. 2011). It is notable that, while the literature broadly shows that RO is a highly effective method for the removal of PhACs from municipal wastewater, select tests show some compounds that are not well removed. For example, a higher maximum value in the RO permeate compared with the inlet for acetaminophen was demonstrated by Rodriguez-Mozaz et al. (2015).

That removal rates are compound-specific in RO has been noted in many peer-reviewed studies, with multiple plausible explanations posited. Size exclusion is the most significant contributing mechanism, so molecules with a molecular diameter smaller than the nominal pore size of the RO membrane represent a possible explanation for some compounds passing through the membrane (Malaeb and Ayoub 2011; Radjenović et al. 2008). However, most PhACs are of a large enough size that they should be retained by the membrane (molecular weight of $129.167\mathrm{g}/\mathrm{mol}$ or higher). It is therefore expected that they will be retained by the membrane based on size exclusion.

Electrostatic repulsion is another mechanism that can lead to removal of PhACs by the RO membrane. Electrostatic forces due to charged particles interacting with the membrane can cause repulsion, leading to rejection of the compounds. It should be noted, however, that membrane fouling can lead to decreased removal rates for a compound (Lin 2017) if ionic properties are a significant mechanism leading to membrane rejection for that compound. Fouling of a membrane can cause individual molecules to bond to the membrane, leading to a reduction in ionic repulsion. A result of membrane fouling is that compounds for which ionic repulsion is a consequential mechanism may no longer be removed.

Another mechanism for PhACs to be poorly rejected by a membrane is sorption to the membrane, followed by diffusion through it and subsequent desorption on its permeate side. The impact of sorption has been most noted for hydrophobic PhACs, with octanol water partitioning coefficient ($K_{OW}$) the characteristic most commonly used to describe hydrophobicity. A study on the impact of hydrophobic sorption to RO membranes noted that it becomes common for PhACs to pass through the membrane into the permeate through a sorption-diffusion-desorption mechanism when $K_{OW}>2.0$. It should be noted that this effect is often overlooked by many studies as it can be very slow (Dolar et al. 2013). For PhACs that are often measured in wastewater in the $\mathrm{ng}/\mathrm{L}$ range, it takes significant time for the membrane to become saturated and subsequently diffuse and desorb the compound on the other side. For this reason, experimental studies that work with new membranes can easily overlook this effect as a significant mechanism for poor removal rates of hydrophobic compounds.

The RO process used in this study has been running consistently for more than six months, with good removal of hydrophobic compounds persisting after this time. This indicates that, although it has been demonstrated in previous studies that hydrophobic sorption can be a major contributing factor in RO membrane effectiveness reduction, this is not the case for the compounds tested in this study. Several factors play into the strong performance of the membrane for hydrophobic PhACs, such as the complex matrix of municipal secondary effluent leading to many compounds competing for a limited number of sorption sites, as well as the membrane being less prone to hydrophobic sorption compared with membranes used in the study by Dolar et al. (2013).

A similar process for diffusion through an RO membrane is ionic sorption. This mechanism is infrequently discussed in literature as ionic forces on the membrane often lead to the repulsion of free ions in solution. However, it is possible that a PhAC can sorb to a membrane due to ionic forces, travel through the membrane, and then be released on the permeate side.

In this study, the only PhAC that showed inadequate removal across the membrane was ciprofloxacin. The reasons behind poor removal of ciprofloxacin of only 33% are unclear, as it does not match any criteria that would fit any common modality of poor treatment in RO.

The removal of two common PhAC metabolites were also determined: o-desmethylvenlafaxine was removed at a rate of 99.5%, and norfluoxetine was removed to below detection limits. While it was expected that these metabolites would be removed at rates similar to those of their parent compounds, few experimental data verifying this result have been published. With results indicating that removal of these metabolites follows similar removal by RO compared with their parent compounds, it is reasonable to expect that common metabolites of other PhACs will be removed similarly to their parent compounds unless changes to the molecular structure have led to a significant decrease in molecular diameter or if the compound had a significant increase in polarity.

Data for PhAC concentration in the RO reject can also be seen in Fig. 6, which indicates that the concentration of tested PhACs was generally increased in the reject stream. While RO is a highly effective process for the removal of PhACs from municipal wastewater, the creation of a concentrated waste stream is a significant concern that mandates management if RO is to be successfully applied on a large scale. RO does nothing to break down pollutants; it simply concentrates them. As substantial volumes of concentrated waste would be created in a full-scale application, applying RO at a full-scale necessitates that it be further treated or in some way not discharged to a receiving body (Subramani and Jacangelo 2014).

For further treatment of waste streams, AOPs such as ozonation, discussed in this paper, or ${\mathrm{UV}/\mathrm{H}}_2{\mathrm{O}}_2$ show strong potential as processes to treat RO waste streams (Dialynas et al. 2008; Zhou et al. 2011). As most AOPs have been shown to degrade PhACs by pseudo-first-order kinetics, it is advantageous to apply them to concentrated waste streams rather than dilute solutions, as is the case for PhACs in municipal wastewater (Lian et al. 2017; Tokumura et al. 2016). AOPs are mechanistically dependent on the molecular collision of highly reactive radical oxidants, with oxidation of whatever organics the radicals contact being the first to be degraded. With municipal wastes having complex matrices that include multiple organic and inorganic constituents, some at fairly high concentrations, it becomes increasingly unlikely that PhACs will be removed from the solution when concentrations are in the $\mathrm{ng}/\mathrm{L}$ range. As concentration of target PhACs is significantly increased through RO, molecular collisions with these contaminants by the radicals produced by AOPs becomes more likely, making the process more effective. Notably, AOPs have been demonstrated to have low efficacy in dilute solutions because of the potential for recombination of radicals if molecular collisions do not readily occur with target contaminants. For these reasons, AOPs present a highly appealing process for the treatment of RO concentrate waste.

While RO can be clearly demonstrated as effective in the removal of PhACs from municipal wastewater, its use must be counterbalanced by a plan to manage the concentrate steams.

Degradation efficiencies for various neutral PhACs under ozonation are shown in Fig. 7. More than 99.9% degradation efficiency (below detection limit) was noted for sulfamethoxazole, carbamazepine, erythromycin, and o-desmethylvenlafaxine. Trimethoprim and venlafaxine showed slightly lower removal, though both were reduced by more than 95%. The remaining compounds degraded significantly less effectively, with degradation between 16% and 85%. They can be ordered from least to most degraded as metformin, cotinine, ciprofloxacin, acetaminophen, norfluoxetine, fluoxetine, caffeine. The variation in degradation efficiencies noted for each compound was influenced by the inherent functional groups that determine their susceptibility to degradation by ozone and hydroxyl radicals rather than their inherent concentration. The mechanism of ozone reactions with organic compounds in aqueous solutions corroborates this rationale. Degradation of organic contaminants by ozonation occurs via two predominant mechanisms: (1) direct attack of ${\mathrm{O}}_3$; and (2) reaction with hydroxyl radicals (•OH), formed as a result of ${\mathrm{O}}_3$ decay. Reactions summarizing the dissociation of ozone to form •OH in water can be seen in Eqs. (S1)–(S7) in Supplemental Materials.

Proper control of ozone dose for treatment is highly consequential, as the overproduction of radicals can lead to their self-elimination, causing decreases in process performance (Beltrán et al. 1999):

Direct attack of ${\mathrm{O}}_3$ is highly dependent on the specific functional groups present in the target organic compounds. These include dipolar cycloaddition of unsaturated bonds (Criegee mechanism) and electrophilic reactions with aromatic compounds with electron donor groups (OH, ${\mathrm{NH}}_2$, $─{\mathrm{OCH}}_3$), and amines and thiols in ortho and para positions showing high electron densities. In contrast to the highly selective attack of ${\mathrm{O}}_3$, •OH radicals are nonselective and react rapidly with most functional groups in organic compounds. This is also a drawback of oxidation by •OH radicals, as they react with most organic components in wastewater, leading to their degrading various organics in solution rather than the compounds of interest. For this reason, compounds susceptible to ${\mathrm{O}}_3$ are quickly degraded in a wastewater matrix compared with compounds that are susceptible to reaction only with •OH radicals.

Rapid reactions of sulfamethoxazole and carbamazepine with ozone have been reported in the literature. Both have shown faster transformation under ozonation in comparison with many other PhACs (Huber et al. 2003). Somathilake et al. (2018) showed that degradation of carbamazepine is not significantly altered in the wastewater matrix when treated with ozonation (Somathilake et al. 2018). Amine and C═C double bonds in both sulfamethoxazole and carbamazepine have been shown to be directly susceptible to oxidation by ozone. In a study comparing degradation of carbamazepine and venlafaxine using ozonation, faster degradation of carbamazepine was observed, which was corroborated by the present study (Lester et al. 2013). McDowell et al. (2005) reported caffeine as one of the faster-degrading compounds under ozonation due to the presence of C═C double bonds (Mcdowell et al. 2005). The lack of susceptible functional groups in other compounds hinders the fast electrophilic attack by ozone. In a study comparing degradation of acetaminophen with ${\mathrm{UV}/\mathrm{H}}_2{\mathrm{O}}_2$ and ozonation, faster degradation was observed with ${\mathrm{UV}/\mathrm{H}}_2{\mathrm{O}}_2$ by three orders of magnitude (Hamdi et al. 2014). This clearly indicates that acetaminophen is more prone to attack by •OH radicals than by ${\mathrm{O}}_3$.

Regarding contaminants that were recalcitrant to ${\mathrm{O}}_3$ degradation, poor results for the degradation of metformin was mirrored in bench-scale studies. In one study, $8\mathrm{mg}/\mathrm{L}$ with 30-min contact time was able to degrade only 60% (Quintão et al. 2016). Considering that the present study operated at a significantly higher concentration of ozone ($50\mathrm{mg}/\mathrm{L}$), it can be noted that this real application setting displayed significantly reduced degradation compared with that expected based on the controlled lab study. It is probable that other organics in solution are selectively degraded before metformin, causing the discrepancy. A study using multiple PhACs found poor degradation similar to that in this study for cotinine by ozone, while degradation of ciprofloxacin was noted to be higher than was found in this study; the reduced efficacy was likely due to the complex matrix of real municipal wastewater (Rosal et al. 2010).

In pilot-scale tests for the treatment of PhACs in drinking water, it was determined that carbamazepine, caffeine, and cotinine all showed very high removal rates by ozone treatment under an ozone dosage of only $2\mathrm{mg}/\mathrm{L}$ and a contact time of 20 min (Hua et al. 2006). While high removal of carbamazepine was also found in this study, a lower rate for caffeine and very little degradation for cotinine was noted. To allow for contrasting the effect of the water matrix on process effectiveness, it is consequential to compare data that have been published at a similar application scale using a different water matrix. The comparatively lower degradation in wastewater can be primarily attributed to dissolved organic matter in the wastewater matrix which acts as ozone and radical scavengers (Durán et al. 2011).

While ozone has been shown to be highly effective for the degradation of many different PhACs in wastewater, it is unlikely that conditions can be met that would oxidize all contaminants. If reduction of PhACs follows pseudo-first-order kinetics, as has often been suggested in the literature, impractical contact times or an extremely high ozone dose would be required for one log reduction in some recalcitrant contaminants (Dai et al. 2015; Huang and Liu 2015; Patel et al. 2019). As such, the selective degradability of many contaminants shown in this application-scale study presents a significant impediment to the robustness of this technology.

An additional challenge in ozone treatment for the removal of PhACs from wastewater is the potential for transformation by-products to be created from incomplete contaminant degradation. While the final objective of oxidation treatments is the compete mineralization of organic contaminants, this may not be the result. Particularly for ozone treatment, lab-scale literature provides an indication that much of the total mass of PhACs measured continues to exist in some altered form, as indicated by low reductions in total organic carbon after treatment (Hollman et al. 2018; Mehrabani-Zeinabad et al. 2015; Rosal et al. 2008; Somathilake et al. 2018). Analytical methods for quantification of PhACs identify a reduction in a specific compound, but they are not capable of determining if by-products have been produced, even if the only change to the original compound is a simple functional group substitution. As some by-products can have a toxicological profile similar to that of their parent compounds, it is important to understand the degree of mineralization or to test for specific environmental impacts of treated water. In application-scale testing of municipal wastewater, total organic carbon is not an effective measure of degree of mineralization, because of the very low concentration of contaminants combined with the high concentration of other organics in wastewater. It is therefore necessary that we rely on bench-scale experimental results for determining the total mineralization potential of ozone processes. Current literature in this area indicates that, for the contaminants discussed in this study, a high degree of mineralization can be achieved for ozone treatment of sulfamethoxazole while only a marginal reduction in total organic carbon or substantial detected by-products has been seen for carbamazepine, erythromycin, and trimethoprim (Beltrán et al. 2008; Kuang et al. 2013; Luiz et al. 2010; Somathilake et al. 2018). The PhACs norfluoxetine and o-desmethylvenlafaxine were considered in this study to provide an indication of how effective the process is for the removal of common metabolites. Both metabolites were significantly removed compared with their parent compounds. As these metabolites are commonly formed during biological treatment and are amenable to ozone oxidation, an indication is given that a combined process for enhanced biological treatment followed by ozone treatment may be a viable option to enhance PhAC treatment in existing facilities.