Pharmaceuticals’ removal by constructed wetlands: a critical evaluation and meta-analysis on performance, risk reduction, and role of physicochemical properties on removal mechanisms

This paper presents a comprehensive and critical analysis of the removal of pharmaceuticals (PhCs), the governing physicochemical properties, and removal mechanisms in constructed wetlands (CWs). The average removal ef ﬁ ciency of the most widely studied 34 PhCs ranges from 21% to 93%, with the exception of one PhC that exhibited negative removal. Moreover, CWs are effective in signi ﬁ cantly reducing the environmental risk caused by many PhCs. Based on risk assessment, 12 PhCs were classi ﬁ ed under high risk category (oxytetracycline > o ﬂ oxacin > sulfamethoxazole > erythromycin > sulfadiazine > gem ﬁ brozil > ibuprofen > acetaminophen > salicylic acid > sulfamethazine > naproxen > clarithromycin), which could be considered for regular monitoring, water quality standard formulation and control purposes. Biodegradation (aerobic and anaerobic) is responsible for the removal of the majority of PhCs, often in conjunction with other mechanisms (e.g., adsorption/sorption, plant uptake, and photodegradation). The physicochemical properties of molecules play a pivotal role in the elimination processes, and could serve as important predictors of removal. The correlation and multiple linear regression analysis suggest that organic carbon sorption coef ﬁ cient (Log Koc), octanol-water distribution coef ﬁ cient (Log Dow), and molecular weight form a good predictive linear regression model for the removal ef ﬁ ciency of PhCs (R 2 ¼ 0.65, P -value < 0.05).

. Therefore, PhC pollution can be seen as a worldwide concern for almost every countryno matter how much of the total wastewater produced is treated before discharge into the environment.
Although PhCs are found to be in relatively small concentrations (e.g., ng L À1 to μg L À1 ) in water resources, their presence (as individual compounds, transformation products (TPs), and multitude of compounds) could pose risk for aquatic and terrestrial life. The continuous discharge of PhCs through various sources including WWTPs could make these 'pseudo-persistent' organic chemicals a potential source of risk, especially when present in large concentrations, and the combination of a wide range of compounds that may act synergistically (e.g., Gorito et al. ). Furthermore, in WWTPs, during biological treatment, the development of antibiotic resistant bacteria (ARB) and/ or antibiotic resistance genes (ARGs) due to the subtherapeutic concentrations of antibiotics is of major concern. Antibiotic resistance is the ability of bacteria and other microorganisms to resist the effects of an antibiotic to which they were once sensitive (Berglund et  This study aims to fill some of the above-mentioned knowledge gaps by building on the previous reviews and a large number of published case studies, including recently published sources (e.g., after 2013, as most of the previous reviews were published before 2014). Therefore, the main objectives of this study are: (1) to conduct a comprehensive assessment of a large number of PhCs in wastewater and their removal by four types of CWs; (2) to critically evaluate and summarize the available evidence on major PhCs' removal mechanisms in CWs; (3) to examine the impact of physicochemical properties of PhCs on their removal mechanisms; and (4) to assess the environmental risk posed by a large number of PhCs, and contribution of CWs in risk reduction.

METHODOLOGY
The research papers, review papers, and books were searched from various sources, such as Scopus, Google Scholar, and individual journal websites, related to the performance of different types of CWs for the removal of different categories of PhCs. The snowball sampling method yielded over 100 journal articles, which were further screened and used for the purpose of this research.
The screening was carried out to check the quality of pub-  A detailed analysis of the reported PhCs was conducted from the studied literature including the designed database, which focused on therapeutic classes, types of PhCs, and impact of their physicochemical properties and contribution of removal mechanisms in CWs. The mechanisms were identified for the selected PhCs as presented in the published case studies. The majority of the studies only attributed removal to certain mechanisms (e.g., biodegradation, adsorption/sorption, plant uptake, and photodegradation). The relative contribution of mechanisms in removal was only quantified in a few experimental studies. Therefore, the analysis on removal mechanisms was based on a critical oversight from both qualitative and quantitative information. The information on the physicochemical properties of PhCs was gathered from various sources (e.g., journal papers, reports, and websites) for molecular formula/structure/weight, water solubility, dissociation constant (pKa), organic carbon sorption coefficient (Log Koc), octanol-water partition coefficient (Log Kow), and distribution coefficient (Log Dow). The available evidence on the role of these properties in the removal of PhCs in CWs was comprehensively and critically analyzed. The linkages between physicochemical properties and removal mechanisms were delineated from this analysis. Moreover, a statistical analysis (Pearson correlation and multiple linear regression) was conducted for removal efficiency and physicochemical properties. The causality of the observed relationship was established through a synthesis of available knowledge and the authors' own insights.
Additionally, risk assessment was carried out by estimating RQ. Following the recommendations by Hernando

RESULTS AND DISCUSSION
Removal of PhCs and TPs by CWs      This prohibits generalization of individual case study results for these PhCs. Therefore, a comprehensive assessment is not possible for every compound. However, we identified 34 PhCs that were studied by several authors for a comprehensive assessment and critical review of the available knowledge. Therefore, the following sections present results and discussion on these selected PhCs.

Removal of widely studied PhCs by CWs
The following insights can be drawn from the analysis presented in Table 2 tive removal efficiency ranging from À283% to À4,067%. μg L À1 ), Daphnia (7.8 μg L À1 ), algae (0.02-4.3 μg L À1 ), and invertebrates (15 μg L À1 ). In this case, the lowest value of 0.02 μg L À1 was used as the PNEC for our assessment.
Then, RQ was calculated using the lowest PNEC value and the MEC of influent and effluent of PhCs. These calculations were performed for the selected PhCs based on all the available data points. The mean RQ were estimated from this analysis and discussed in detail in this section. Since mean could be biased towards high values, median and various other percentiles were also estimated. The RQ was also estimated based on extremes (minimum and maximum values). The resulting statistics are given in Supplementary Materials 4: Table S7. The mean RQ estimates are given by Figure 2 and Our results also reveal that, in general, the estimated Obviously, RQs' assessment based on C. dubia will be much higher than estimated using D. magna in this case.   Risk rank is based on our results (*); Risk is categorized into four levels: high risk (RQ > 1.0), medium risk (0.1 RQ 1.0), low risk (0.01 RQ 0.1), and no risk (RQ < 0.01).
categorization based on the 25th percentile (P25) shows only a few PhCs under the high risk category, and may result in severe underestimation of risk. However, the risk categorization based on the 75th percentile (P75) is more stringent than the mean, and could also be used as a risk classification threshold, then a more stringent approach can be adopted.
The extreme value analysis was not recommended for the risk classification because of significant underestimation in the case of using minimum RQ value or significant overestimation in the case of maximum RQ value. Finally, while aiming at achieving the best ecosystem protection can be recommended in theory, several trade-offs may apply in practice, which require sound scientific evidence to make an informed decision on the target levels of PhCs in a specific aquatic environment.

Role of physicochemical properties of PhCs and removal mechanisms in CWs
Based on the available evidence, synthesis on the role of physicochemical properties and removal mechanisms for 34 PhCs was conducted. For a few PhCs, experimental studies were available to quantify the relative contribution of various mechanisms. This work is summarized in Figure 3 (hydroponic microcosms and media adsorption experiments), Figure 4 (CWs), and Supplementary Materials 5: Tables S8 and S9. Correlation and multiple linear regression analyses were performed to examine the linkages between removal efficiency and physicochemical properties, which are discussed in this section (details are given in Table 4 and Supplementary Materials 6: Tables S10-S14). The summary of removal mechanisms reported in the literature for the PhCs is presented in  The studies examined the contribution of one or more removal mechanisms. When the sum of the reported contribution by different mechanisms exceeded 100%, we standardized the contribution from each mechanism to 100% by adding removal of all the studied mechanisms and dividing it by the total. For example, in the case of ibuprofen, the contribution by biodegradation, adsorption, and plant uptake was 30%, 53%, and 24.6%, respectively. The total removal is 107.6%. However, out of 100%, the contribution of biodegradation, adsorption, and plant uptake was 27.9%, 49.3%, and 22.9%, respectively. Adsorption is the adhesion of dissolved solid molecules (adsorbate) to a surface of the substrate (adsorbent). It is a surface phenomenon.
Absorption is a process in which a fluid (absorbate) permeates a solid (                  It is widely considered that organic compounds with moderate hydrophobicity (1.0 < Log Kow < 3.5; Log Dow < 2.5) and low molecular weight (MW < 500 g mol À1 ) have adequate properties to move through cell membranes, Biodegradation is the major removal mechanism in most of the studied PhCs (19 out of 34 selected PhCs) ( Therefore, experimental studies are essential to establish biodegradability of every PhC. Moreover, biodegradation as a dominant process does not guarantee higher removal possibilities in all the cases (e.g., bezafibrate, gemfibrozil, mirtazapin, sotalol, and trimethroprim) (Tables 2 and 5).
This reveals the complexity of the biodegradation process itself (how much a compound is biodegradable) but also the role of other processes in CWs and physicochemical properties of PhCs.
Furthermore, for the compounds which are most hydrophilic (Log Kow < 1.0), the most water soluble (WS > 1,000 mg L À1 ), and have the lowest molecular weight (MW < 100 g mol À1 ), adsorption cannot be con- When adsorption/sorption is a dominant removal mechanism, the removal efficiencies are either moderate (e.g., atenolol and codeine) or low (e.g., fexofenadine, ranitidine, and carbamazepine) even in the CWs that can provide good media for adsorption/sorption (Tables 2 and 5). It has been observed that adsorption/sorption potential of a CW may decrease due to creation of biofilms around the filter media that may prohibit access to adsorption/sorption surfaces (Dordio et al. a, ).  (Table 4), which can be further supported by the negative correlation (although non-significant) of Log Koc and water solubility ( Table 4). The Log Koc is a good indicator of the sorption potential and mobility of the PhCs, which can influence their fate in CWs. We found a significant negative correlation of Log Koc with removal efficiency (r: À0.697) (Table 4), which could explain about 49% of the variance in the available data (R 2 : 0.49) (Table S10). This indicates the possibility of using Log Koc as a screening parameter, even though its proportionate contribution in overall removal may be limited. Similarly, Log Dow shows a significant negative correlation with removal efficiency (r: À0.420) (Table 4), although this could only explain about 18% variance in the available data (R 2 : 0.18) (Table S11). A multiple linear regression of removal efficiency with these two physicochemical properties was found as a significant predictor (R 2 : 0.60; P-value <0.05) (Table S13).
In particular, the PhCs that have chlorine in their molecular structure (e.g., bezafibrate, clofibric acid, diclofenac, and furosemide) are considered recalcitrant to biodegradation. Most of these PhCs are very difficult to remove by CWs, and thus show low removal efficiencies (<50%) with the exception of furosemide (72%) (Tables 2 and 6). As mentioned above, molecular structure/weight plays an important role in the removal processes, although molecular weight did not show a statistically significant correlation with removal efficiency (r: À0.384 and R 2 : 0.15) (Tables 4   and S12). Nevertheless, when molecular weight was included in the multiple linear regression with Log Koc and Log Dow, the result was the best possible model for predicting removal efficiency (R 2 : 0.65; P-value <0.05) (Table S14), compared to when other physicochemical properties (water solubility, Log Kow, and pKa) were included in regression with Log Koc and Log Dow. A statistically significant model was also possible with four physicochemical properties but was not able to explain more variance compared with the best model. Therefore, we preferred a model with three variables, which was derived using the data of all types of examined CWs. The novel relationship developed in this study is given below and details on statistics are presented in Table S14.
RE ¼ 96:16 À 14:18 Log Koc À 3:72 Log Dow À 0:03 MW (1) where: RE is removal efficiency in % and MW is molecular weight in g mol À1 ; Log Koc and Log Dow are already defined in this paper.
The other processes were found to be dominant only in very few PhCs such as photodegradation in the case of diclofenac, clarithromycin, and ketoprofen, and hydrolysis of furosemide. These processes also demonstrate low to moderate removal efficiencies of these PhCs (Tables 2   and 5).
The above insights clearly indicate the removal mechan- Analgesic/anti-inflammatory drugs

Diclofenac
The removal efficiency of diclofenac was moderate in HCW (56 ± 32%) and VFCW (50 ± 17%), and comparatively lower in FWSCW (42 ± 24%) and HFCW (39 ± 24%). It is suggested that the presence of chlorine in its structure ( ( Figure 3 and Table S8). This was confirmed by Zhang et al. (c), who calculated the bioaccumulation factor (BAF) and reported that its BAF in the shoots was less than half (0.17-0.51) compared with BAF in the roots (0.40-1.36). This can be attributed to both its high hydrophobicity and relatively low water solubility (4.52 mg L À1 at 25 C) ( Table 6). It has been suggested that organic com-  (Table S2), and planted and unplanted HCW (38% and 25%, respectively) (Hijosa-Valsero et al. a) (Table S4) might be due to indirect positive effects of plants' presence such as degradation by enzymatic exudates as well as an increase in the amount of oxygen released by the plant roots in the rhizosphere which can support high microbial activity (biodegradation). However, in hydroponic microcosms it has been revealed that the contribution of biodegradation to its removal was low (3.0%) (Zhang et al. b) (Figure 3 and Table S8). Its high removal by photodegradation was achieved in hydroponic microcosms (79 ± 2%) (Zhang et al. c, b) (Figure 3 and Table S8) and it was confirmed in the

Ibuprofen
The removal efficiency of ibuprofen was much higher in VFCW (79 ± 24%) but moderate in HCW (62 ± 29%), FWSCW (57 ± 28%), and HFCW (53 ± 27%). Its removal by plant uptake is expected to be low since it is hydrophobic  (Table 6). This can be explained by its low plant uptake (0.4-5.0%) in the planted CWs ( Figure 4 and Table S9) (Zhang et al. a, b). Therefore, the removal efficiency differences between the planted and unplanted hydroponic microcosms (78% and 30%, respectively) (Matamoros et al. ) (Figure 3 and  (Figure 3 and Table S8). Nevertheless, its detection in the plant leaves (Typha angustifolia) with an even distribution in the lamina and sheath tissues reveals the phytoextraction process for its removal as well. It had been taken up by the plant roots, which is translocated from the roots to the leaves and accumulated in leaf tissues. For instance, its root uptake was partially transformed to carboxyibuprofen, 2-hydroxyibuprofen, and 1-hydroxyibuprofen in the sheath (1,375, 236, and 302 μg kg À1 , respectively) and in the lamina (1,051, 694, and 179 μg kg À1 , respectively).
The accumulation of its metabolites in the plant leaves indicates its phytotransformation in the plant tissues (Li et al. b). Although it has Log Kow < 4 and Log Dow < 2.5, its high sorption coefficient (Log Koc ¼ 2.60) ( Table 6) Table S9).
Since it is anionic under neutral pH condition, the repulsion with the negatively charged biomembranes might restrict its removal by plant uptake. However, Hijosa-Valsero et al.
(b) reported that it was not removed in unplanted HCW but removed in planted HCW (29-97%) (Table S4). This removal by the planted system could be due to degradation by enzymatic exudates as well as biodegradation.

Acetaminophen
The removal efficiency of acetaminophen was higher in   Table S9). Furthermore, Koottatep et al. () suggested that by-product transformation contributed up to 15% of its total removal (Table S9).

Oxytetracycline
The removal efficiency of oxytetracycline was higher and almost the same in FWSCW (97%) and VFCW (96 ± 4%  (Figure 4 and  (Figure 4 and Table S9). However, the enhancement in removal in the presence of plants might be due to the improvement in biodegradation (Dordio & Carvalho ).

Sulfadiazine
The removal efficiency of sulfadiazine was moderate in FWSCW (61 ± 35%) and VFCW (52 ± 22%) but low in HFCW (46 ± 30%). Adsorption to the substrate cannot be considered its main removal mechanism in CWs due to its high water solubility (28.14 g L À1 at 25 C) and very low hydrophobicity (Log Kow ¼ À0.09), although its molecular weight is moderate (250.28 g mol À1 ) ( Table 6). This can be seen by the non-significant difference in the removal efficiency between HFCW and FWSCW in winter (19 ± 2% and 19 ± 5%, respectively) which represented full substrate system and half substrate system, respectively (Dan et al. (Tables S1 and S2). Similarly, although its sorption capacity is moderate (Log Koc ¼ 1.87), due to its neutral or anionic nature under neutral pH conditions (  (Tables S1 and S2), since substrates provide a surface area suitable for the growth of microorganisms and the formation of biofilm (Dan et al. ).

Sulfamethazine
The removal efficiency of sulfamethazine was moderate in HCW (74%) and FWSCW (53 ± 48%) but lower in HFCW (45 ± 27%) and VFCW (28 ± 27%). The high water solubility (11.24 g L À1 at 25 C) and high hydrophilicity (Log Kow ¼ 0.89; Log Dow ¼ 0.79) do not favor its adsorption to the substrate in CWs, although its molecular weight is moderate (278.33 g mol À1 ) (Table 6). Similarly, although its sorption capacity is moderate (Log Koc ¼ 2.28), due to its neutral or anionic nature (Table 6), its binding to biomass (sorption) is also likely to be negligible (Dan et al. ). This is evident by its removal in HFCW using different substrate materials (oyster shell, zeolite, medical stone, and ceramic), and it was below the limit of detection in substrate media (Chen et al. a) (Figure 4 and  (Table S1).
Additionally, due to its neutral or anionic form (  (Table S4), which indicates that in planted CWs direct uptake by the plants is minimal due to its low Log Kow, but the plants also support biodegradation (Choi et al. ; Liang et al. ). This is evident by the major contribution of biodegradation pathways (68%) to its total removal of 71% in hydroponic systems (Choi et al. (Figure 3 and Table S8). In unplanted CWs, the removal may be because the substrates provide a surface area suitable for the growth of microorganisms and the formation of biofilm for biodegradation (Dan et al. ; Choi et al. ). This is obvious by its higher removal in biotic systems (73%) compared with abiotic systems (67%) during a soil adsorption experiment under biotic and abiotic conditions (Choi et al. ). Furthermore, Choi et al. () investigated the possibility of photodegradation and observed 23% of its removal by this process in a photodegradation experiment (Figure 3 and Table S8). Thus, a slightly higher increase in the removal efficiency by unplanted HCW (FWS on top of HFF) compared with planted HCW indicates that photodegradation might contribute to its removal (Hijosa-Valsero et al. a).

Gemfibrozil
The removal efficiency of gemfibrozil was higher in HCW (95%), moderate in HFCW (58 ± 23%) but low in VFCW (45 ± 9%) and FWSCW (12 ± 2%). Its high hydrophobicity (Log Kow ¼ 4.77), low water solubility (4.964 mg L À1 at 25 C), moderate molecular weight (250.33 g mol À1 ), high sorption coefficient (Log Koc ¼ 2.636), and anionic form under neutral conditions (pH ¼ 7) (Table 6) favors its adsorption onto soil particles following complexation with metal ions. However, its adsorption to substrate as one of its removal pathways has not been reported by any of the reviewed studies which investigated its removal (Tables S1-S4). Furthermore, its low to moderate removal in most of the cases indicates its lower retention due to the development of negatively charged biofilms on substrate surfaces over time (Dordio et al. a, ), which might obstruct its binding to biomass. Similarly, due to its anionic nature,

SUPPLEMENTARY MATERIAL
The Supplementary Material for this paper is available online at https://dx.doi.org/10.2166/wh.2020.213.