Raw water quality
In total 68 (23 %) of the 291 targeted compounds were detected above the MDL in at least one raw water sample, and the number of detected compounds per sampling site ranged from 49 to 64. The detected micropollutants represented a wide variety of compounds including pesticides (n=27), pharmaceuticals (n=15), PFASs (n=10), organophosphorus flame retardants (n=7), food additives (n=2), industrial chemicals (n=2), surfactants (n=3), and a fatty acid (n=1), stimulant (n=1). The detected compounds and their concentrations at each sampling point are given in Table A4 in the Supporting Information.
The number of detected micropollutants on a category basis were rather similar between the five DWTPs (Figure 2; raw water). This indicates a rather even distribution of emission sources in the Tai Hu Lake and the Yangtze River Delta area, with the exception of pesticides, which showed a higher number of detected compounds at the river network site. Overall, pesticides was the most detected pollutant category with on average 35% ± 2% of all compounds detected (including all samples and sampling sites; n=18), followed by pharmaceuticals (19% ± 2%), PFASs (17% ± 2%), ‘Others’ (16% ± 2%: food additives, industrial chemicals, surfactants, stimulant and fatty acid), and flame retardants (13% ± 1%).
The total concentrations of the micropollutants in the raw water (Figure 3) showed that the wetland river network was the most polluted site with respect to organic micropollutants (4 000 ng L-1), followed by the Tai Hu Lake (1 600 ng L-1 for Su Tai and Wu Tai, respectively; using the same source water). As expected, Yangtze, with its raw water already treated through serval steps at another DWTP (Figure 1b), showed the lowest total level (1 000 ng L-1).
Overall, the most prevalent micropollutant was sucralose (food additive), which contributed to about one quarter of the total concentration in the water from Su Tai and Wu Tai (Figure 3), and even more in the water from the Yangtze River (48%; Figure 3). In contrast, sucralose made up only 5% of the total concentration in the river network. Sucralose is frequently detected in rivers and lakes all over the world, with a reported concentration range of 0.08-1.0 μg L-1 in Europe , 18-175 ng L-1 in River Rhine , and 200-400 ng L-1 in northern China .
There were also other clear differences in the composition of the three source waters: Yangtze, river network and Tai Hu (Su Tai/Wu Tai). In the river network, the fraction made up by pesticides was 71% of the cumulative concentration, which was significantly higher than at the other three sites (8-23%; Fig. 3). The cumulative concentration was 2 800 ng L-1 in the river network, which is almost 10 times higher than the average concentrations at Tai Hu (340 ± 100 ng L-1) and 30 times higher than the Yangtze River raw water (97 ng L-1). The reason for the high dominance of pesticides in the river network can be attributed to the land use in the river catchment, which has a high proportion of agricultural areas. Moreover, although both Su Tai and Wu Tai are using Lake Tai Hu raw water, the Su Tai samples showed higher levels (370 ± 47 ng L-1; 23% of total) than that at Wu Tai (130 ng L-1; 8%). The pesticides with the highest concentrations at river network were carbendazim, imidacloprid, bentazon and azoxystrobin, which were detected at levels in the range 420-670 ng L-1. Overall (all sites), other common pesticides included bentazon (85 ± 47 ng L-1), carbendazim (60 ± 40 ng L-1), N, N-diethyl-m-toluamide (DEET) (38 ± 32 ng L-1) and atrazine (32 ± 11 ng L-1), which showed high detection frequency in source waters. These concentrations are corresponding or higher to those reported from Europe, where e.g., DEET averaged 4.0 ng L-1 and atrazine averaged 2.2 ng L-1 in rivers in southeast Spain , and reported concentration ranges in River Rhine were 1-16 ng L-1 for bentazone, 7-105 ng L-1 carbendazim, 6-120 ng L-1 for DEET, and 1-6 ng L-1 for atrazine .
The number of detected PFASs differed between the sampling sites, with the lowest number at the Yangtze River site (n=7; 14% of total number of detected compounds), whereas for other raw waters, the number of detected PFASs ranged from 9 to 10 (18%-18%), indicating upstream point sources of PFASs. The average concentration of ∑PFASs in the Tai Hu raw water (168 ± 22 ng L-1) was significantly higher (two-sided Student’s t-test) than in the Yangtze River water (20 ng L-1), and the water from the wetland river network (56 ng L-1) (p=0.0003). Perfluorohexane sulfonic acid (PFHxS) was dominant in the Tai Hu water (60% of the ∑PFASs), whereas perfluorooctanoic acid (PFOA) was dominant in the water from wetland river network (63%). On the other hand, PFHxS, PFOA and perfluorobutanoic acid (PFBA) were equally dominant in the Yangtze River water (22%, 23 % and 27%, respectively). The PFAS concentrations were generally comparable to other river water concentrations in China  or in the River Rhine watershed in Europe , but lower than other parts of the world such as Sweden , or India .
Organophosphorus flame retardants was the major group of compounds in the Yangtze and Wu Tai water, with cumulative concentrations of 290 ng L-1 and 405 ng L-1 (28% and 26%), respectively. These levels are similar to Su Tai levels (266 ± 48 ng L-1, but there the fraction of flame retardants was lower (16%). Previous research has also reported frequent detection of organophosphorus flame retardants, e.g., in Yangtze River water  and river water in Sweden . Since the use of brominated flame retardants is decreasing because of environmental and human health risks, organophosphorus flame retardants has become widely produced and used since the 1970s . Phosphorus flame retardants have been detected in abiotic compartments as well as in living organisms including humans [29, 30], indicating potential risk for environmental and human hazards.
The concentrations of the compound category ‘Others’, such as industrial chemicals (Table A4 in Supporting Information), were in the range 250-400 ng L-1. The most dominating compound was butyl dihydrogen phosphate, a surface-active agent used as e.g., additive in lubricants and paints, with an overall (all sites) average concentration of 140 ± 48 ng L-1.
An overall observation was that there was a composition profile change from north to south, i.e. going from Yangtze (N4), towards Wu Tai (N3), Su Tai (N1+N2) and river network (N5), with an increasing proportion of pesticides and a decreasing proportion of flame retardants in the samples. This tendency shows that local emission sources and land use play a large role for the overall pollution. It also suggests that the Yangtze River in the north and the river network in the south may influence different parts of Lake Tai Hu.
Drinking water quality
The number of micropollutants detected above MDL decreased with on average 22% through the drinking water treatment process (Figure 2 a). In total, 51 compounds were detected in at least one sample including pesticides (n=17), pharmaceuticals (n=9), PFASs (n=10), organophosphorus flame retardants (n=7), food additives (n=2), industrial chemicals (n=2), surfactants (n=3), and a fatty acid (n=1). Similar to the raw water, the pesticide group was the most prevalent pollution category with on average 13 compounds per sample, followed by PFASs (n=10), flame retardants (n=7), pharmaceuticals (n=6), and food additives (n=1).
The total concentrations of the micropollutants mostly decreased but also increased through the drinking water treatment process (Figure 3, bar charts). The overall (n=9) average total concentration was 730 ± 160 ng L-1, and the concentration range was 720-2 700 ng L-1. As in the raw water, sucralose dominated the concentration in every sample with a top concentration of 1 900 ng L-1 (N1(s); 71% of total) and an overall average concentration of 630 ± 520 ng L-1, which corresponds to a proportion of 30% or more in most of the samples. The average concentration is far higher than levels found in drinking water from northern China (110-160 ng L-1) , and also higher than waste water treatment plant (WWTP) influents in some American cities (78-120 ng L-1) .
The cumulative concentrations of other detected micropollutants (without sucralose) were in the range of 540-1 100 ng L-1 including all sites. The lowest value was 550 ng L-1 (river network), which is approximately 10 times higher than the micropollutant concentration in drinking water from northern Europe (Sweden) , where a similar set of compounds were targeted (134 compounds and an overlap of 43% with the current study). Because of the relatively high removal rates of pesticides and pharmaceuticals, the PFASs, flame retardants and the category ‘Others’ were the dominant micropollutants in the drinking water (Figure 3).
The concentration of ∑PFASs ranged from 110 to 200 ng L-1 at N1-N3 (Tai Hu Basin water; Su Tai/Wu Tai), and was 27 ng L-1 in the sample from N4 (Yangtze River water). The PFHxS was the dominant PFAS in Tai Hu (60-120 ng L-1, N1-N3), while PFBA was the highest PFAS in Yangtze River (11 ng L-1) and PFOA was the highest in the river network water (54 ng L-1). The PFAS concentrations in drinking water were generally higher than those in drinking water from USA, Brazil, Spain, and France [33-35].
Flame retardants was the main micropollutant category in drinking water from the Yangtze River (370 ng L-1; 42 % of total) and generally contributed to more than 20% of the total concentration of all compounds in the drinking water at all sites, except for N1(s). Overall (all sites), the compounds with the highest concentrations were TEP (average 120 ng L-1) and TCEP (average 96 ng L-1). These two organophosphorus flame retardants are at almost the same concentration as levels reported from the Netherlands (48-100 ng L-1) . Another compound with high concentration was butyl dihydrogen phosphate (average 130 ng L-1), an industrial chemical, rarely (if ever) reported as present in drinking water before. The occurrence of micropollutants in drinking water is dependent on the quality of water source (Figure 3) as well as on the treatment efficiency at the DWTP. In the next section, the impact of the treatment process is examined.
The removal efficiencies for all compounds at all DWTPs are given in Table A5 in Supporting Information and are illustrated in Figure 4. The overall average removal efficiency of all compounds averaged 24 ± 149 % (N1-N5, n=9, including the two extra samples and the duplicates at N1+N2). The removal of pesticides averaged 65 ± 50 % and the pharmaceuticals 74 ± 49 %, while the removal of other compound categories were significantly lower (p<10-8, two-sided Student’s t-test), with average efficiencies of 3 ± 99 % for ‘Others’ (food additives, surfactant, industrial chemicals, fatty acid), -1 ± 13% for flame retardants, and -140 ± 329% for PFASs. Thus, the flame retardants showed negative treatment efficiencies, which means they boosted through the treatment process. Similarly, the removal of PFASs was often inefficient or even negative at the investigated DWTPs. Previous studies have shown that conventional treatment techniques are inefficient in removing PFASs . Negative removal efficiency may be explained by time-dependent concentrations and the lag time between sampling raw and finished water, degradation of PFAS precursors to persistent PFASs  and desorption/breakthrough of PFASs from granular activated carbon (GAC) or other filter types within the DWTP . GAC and anion exchange filters are currently used for removal of PFASs in DWTPs; however, development of new treatment options are needed, in particular for the shorter chain PFASs [38-40].
There were no significant differences in the removal efficiencies of all detected compounds among the five DWTP (p>0.06, two-sided Student’s t-test), although the DWTPs used different treatment strategies (Figures 1b and 4a). It should be noted that three of the DWTPs use ozonation, i.e. advanced treatment technology; yet removal efficiencies are moderate to low. This demonstrates the major challenge in removing organic micropollutants through modern drinking water treatment technologies.
The removal efficiencies among different treatment strategies are shown in Figure 4b. N1, N2 and N5 have ozone treatment and biological activated carbon filter (BACF) in contrast to N3 and N4. Additionally, N5 has an extra biological pretreatment step as compared to N1-N4. The removal efficiencies of the pesticides and the pharmaceuticals at N3+N4 were significantly lower than at N1+N2 and N5 (p<0.05, two-sided Student’s t-test), which shows that the conventional treatment (sedimentation, filtration and disinfection) cannot remove pesticides and pharmaceuticals in water effectively. On the other hand, N1+N2 and N5 with their more advanced treatment process (including ozonation) were successful with these two substance groups. This is also illustrated for N5 in Figure 3, with a 5-fold drop in total concentration from raw to drinking water (from 3990 to 720 ng L-1, going from 76% ‘pesticides + pharmaceuticals’ of total to 17%). The removal efficiency of flame retardants in N5 was also rather efficient, with significantly higher efficiency than N1+N2 and N3+N4 (p<0.02, two-sided Student’s t-test). This suggests that the biological treatment has the ability to remove flame retardants in water.
Potential human health risks of sucralose
The artificial sweetener sucralose was the dominant micropollutant in both raw and drinking water and this may cause human health concern. Sucralose is widely used in the food industry since the 1970s, because of its sweet taste and stability. Sucralose has been shown to be a persistent, ubiquitously occurring environmental pollutant [20, 41] and is commonly detected in wastewater, surface water, groundwater and drinking water [20, 22, 42]. It has been tested for toxicity, and there are some reports on that exposure may led to negative effects e.g., on locomotion and physiological behavior of crustaceans . However, there is a lack of studies focusing on chronical and low-dose effects [44, 45]. Rahn et al.  found that sucralose can degrade and produce chloropropanols (3-monochloropropanediol and 1,2- and 1,3-dichloropropanols) under thermal decomposition. These metabolites are known as a potentially toxic class of compounds that may cause cancer . Another study indicated that ingestion of sucralose may affect the glucose metabolism of obese people who rarely use non-nutritive sweeteners . More studies on the toxicity of sucralose are needed, and it is also important to consider the potential toxicity of its metabolites, which e.g., are formed during cooking under high temperature , a common practice in China.