Traffic, Drugs, Mental Health, and Disinfectants: Changes in Sewage Sludge Chemical Signatures During a COVID-19 Community Lockdown

The COVID-19 pandemic and related
shutdowns have caused changes in everyday activities for many people, and signs
of those changes are present in the chemical signatures of sewage sludge
produced during the pandemic. We analyzed primary sewage sludge samples from a
wastewater treatment plant in New Haven, CT USA collected between March 19 and
June 30, 2020. This time period encompassed the first wave of the COVID-19
pandemic, the initial statewide stay at home order, and the first phase of reopening.
We used liquid chromatography coupled with high resolution mass spectrometry and
targeted and suspect screening strategies to identify contaminants in the
sludge. We and found evidence of increasing opioid, cocaine, and antidepressant
use, as well as upward trends in chemicals used in disinfectants and sunscreens
during the study period. Benzotriazole, an anti-corrosion chemical associated
with traffic pollution, decreased through the stay-at-home period, and
increased during reopening. Hydroxychloroquine, a drug that received
significant attention for its potential to treat COVID-19, had elevated concentrations
in the week following the implementation of the United States Emergency Use
Authorization. Our results directly relate to nationwide reports of increased
demand for fentanyl, antidepressants, and other medications, as well as reports
of increased drug overdose deaths during the pandemic. Though wastewater surveillance during the
pandemic has largely focused on measuring SARS-CoV-2 RNA concentrations,
chemical analysis can also show trends that are important for revealing the public
and environmental health effects of the pandemic.


37
The COVID-19 pandemic has dramatically increased the practice of wastewater-based 38 epidemiology, with scientists and public health practitioners worldwide monitoring levels of SARS-CoV-39 2 RNA in untreated wastewater (1). Measurements of SARS-CoV-2 in wastewater and sludge are associated 40 with daily case rates from testing and COVID-19 related hospitalizations, and can provide early information 41 about potential clusters and outbreaks of COVID-19 (2, 3). Historically, wastewater-based epidemiology 42 has focused primarily on chemical contaminants, which can provide information about the habits of the 43 population within the catchment area of a treatment plant. Chemical analysis of wastewater has been used 44 to track use of licit and illicit drugs and pharmaceuticals such as antidepressants, benzodiazepines, opioids 45 and asthma medications, as well as exposure to pesticides and plasticizers (4-6). Wastewater analysis can 46 be a highly efficient way to gather information about topics such as use of illegal drugs and psychoactive 47 medications, without identification of individual persons. Additionally, wastewater analysis has been used 48 to track antiviral and antibiotic use during influenza pandemics throughout the world (7-9). 49 The COVID-19 pandemic has affected many aspects of daily life beyond the direct effects of the 50 virus, and we hypothesized that these changes would be visible in the organic chemical signature of 51 wastewater. Our objectives were to characterize temporal variation of chemical contaminants in sewage 52 sludge during the COVID-19 outbreak and associated lockdown and to relate our findings to the health and 53 activities of local residents as well as broader global trends. We used both targeted and suspect screening 54 methods to cover a broad range of contaminants including common analytes such as pharmaceuticals and 55 illicit drugs (4), but also more unusual compounds for wastewater epidemiology studies such as 56 disinfectants, UV-filters, and pesticides. Previous studies evaluating chemical concentrations in wastewater 57 during the COVID-19 pandemic have focused on limited numbers of analytesprimarily licit and illicit 58 drugs (10-14). Additionally, to our knowledge, this study is the first to report trends in wastewater 59 concentrations for chemicals with direct significance to the COVID-19 pandemic including 60 hydroxychloroquine and disinfectants. Samples were taken at the East Shore Water Pollution Abatement 61 Figure 1. Timeline showing key pandemic related events and the timing of sample collection. We analyzed daily samples for four weeks during the initial increase in local COVID-19 cases. We analyzed weekly composite samples for a total of 15 weeks which covered the early stages of the pandemic and shut down as well as the initial stages of re-opening. All dates are within the year 2020.
TraceFinder to screen our data using an in-house database of approximately 1800 compounds. The database 104 contains exact MS1 and MS2 masses and retention times for many compounds that have previously been 105 measured in house or by collaborators with the same (or very similar) instrument methods used in this 106 project. The database also contains MS1 and MS2 masses that are provided in the Thermo Scientific 107 EFS_HRAM database in TraceFinder (without retention times). Compound identifications using the 108 screening method were based on exact mass matches for MS1 and MS2 masses, isotope pattern matching, 109 and retention time matching where available. Only the Full MS/AIF data was used in the TraceFinder 110 methods. The third method used Compound Discoverer version 3.1 software (Thermo Scientific), and 111 identified compounds based on the ddMS2 data for the 5-week composite samples and spectral matches 112 with the mzCloud database. The full MS data for the daily and weekly samples was then screened for the 113 identified compounds. Peak areas were used for semi-quantitative trend analysis for the compounds 114 identified with Compound Discoverer and TraceFinder screening methods. Each identification was 115 assigned a confidence level based on available evidence. In the main text, identifications based on analytical 116 standards are referred to as "confirmed" while confident screening results (from TraceFinder and 117 Compound Discoverer) are "probable" and screening results where more ambiguity remains are listed as 118 "tentative" (17). More information, including detailed, software specific confidence levels for each 119 identification, is available in sections S.1.4-7, and section S.2.2. 120 Trends over time for each identified compound in daily and weekly samples were determined using 121 two types of analysis: linear regression and multigroup analysis. Multigroup statistical tests used were 122 determined based on the normality and homoscedasticity of each dataset. Trends listed as "increase" in 123 Table 1 indicate a statistically significant positive linear regression (p ≤ 0.05) or a multigroup analysis 124 where there were statistically significant differences between groups (p ≤ 0.05) and an increase in average 125 compound levels in the sludge. Trends listed as "decrease" in Table 1 indicate a statistically significant 126 negative linear regression (p ≤ 0.05) or a multigroup analysis where there were statistically significant 127 differences between groups (p ≤ 0.05) and a decrease in average compound levels in the sludge. 128 Concentrations based on an external calibration curve were used for trend analysis where available (for a 129 portion of the "confirmed" compounds); peak area was used for all other trend analyses (for all other 130 compounds). Detailed statistical methods and results for trend determination are available in sections S.1.8 131 and S.2.2. Table 1 also includes the relative standard deviation (RSD) of each compound concentration or 132 peak area (from replicate unspiked samples, n ≥ 3) as an estimate of measurement error. 133 Ten additional standards were purchased and analyzed after data analysis took place in an effort to 134 improve annotation confidence for interesting results. We found that 9 of 10 compounds were correctly 135 identified (amitriptyline, citalopram, diphenhydramine, triclocarban, didecyldimethylammonium, 136 acetaminophen, benzotriazole, sertraline, and oxybenzone). Results for these compounds are reported as 137 "confirmed", but trend analysis is based on peak area due to lack of quantitative standards run alongside 138 the samples.   immediate or drastic as it would be for a drug with a shorter half-life. Our data indicates that the EUA and 159 the large amount of publicity generated around hydroxychloroquine had significant impact on the amount 160 used in the New Haven area, which includes two major hospitals. Hydroxychloroquine is normally used to 161 treat malaria, lupus and rheumatoid arthritis (20), which are unlikely to have changed during the pandemic. 162 Azithromycin concentrations decreased over the study period (weekly samples, Figure 2b). Azithromycin 163 is only sometimes used in combination with hydroxychloroquine (21) and is more frequently used to treat 164 bacterial respiratory infections which typically decline in the spring (22). Acetaminophen, which can be 165 used to treat COVID-19 symptoms such as fever and headache, had limited availability during the 166 pandemic, likely due to increased demand (23). Correspondingly, acetaminophen sludge concentrations 167 increased in our weekly sample analysis ( Table 1, Table S8). 168 Disinfectant use for cleaning both hands and surfaces has grown during the pandemic (24). Previous 169 studies have shown pandemic related increases in concentrations of quaternary ammonium disinfectants in 170 household dust (25), and higher risk of health effects due to increased exposure (26). Levels of two 171 quaternary ammonium disinfectant chemicals increased in sludge during the overall study period (weekly 172 samples, Figure 1d, Table S8). Triclocarban, an antibacterial compound used in consumer and medical 173 grade handwashes increased in concentration in our daily sampling period (Figure 1c). Triclocarban was 174 previously banned in medical grade hand washes (2017) and rubs and consumer hand washes (2016) for its 175 endocrine disruption potential and other negative health effects (27-29). However, the most recent ruling 176 against triclocarban (regarding consumer antiseptic rubs) took place in 2019, with an effective date of April 177 13, 2020 (30). Thus, it is likely that triclocarban products use had not yet been fully phased out during our 178 study period. Additionally, the pandemic is likely to have prompted increased use of soaps and hand 179 sanitizers that were previously stored. We identified an additional 3 disinfectant compounds for which 180 there were no trends detected during the study period ( Table 1). 181  (Table S8). All scatterplot error bars show the RSD for each compound, calculated from one set of replicate samples.

183
The ongoing epidemic of opioid abuse across the US has included the State of Connecticut (31). 184 Additionally, there are pandemic-related increases in legal use of opioids; in April of 2020, the U.S. Drug 185 Enforcement Agency authorized increased production quotas for fentanyl, morphine, hydromorphone, 186 codeine to meet COVID-19 treatment needs, as well as for methadone, to ensure addiction treatment 187 centers are adequately supplied (32). Sludge concentrations of fentanyl, methadone, and hydromorphone 188 increased during our study period (weekly samples, Figure 3a). Fentanyl and methadone are commonly 189 used both legally and illegally. Hydromorphone is itself a drug, but it is also a metabolite of morphine, 190 codeine, and other opioids, thus its increasing levels are an indication of overall increase in opioid 191 concentrations (33). Levorphanol, an opioid used for pain management and as a preoperative drug (34), 192 decreased in both daily and weekly sludge samples (Figure 3a, Table 1). This decrease is potentially due 193 to the reduction in elective procedures during the study period (35). We did not observe trends over time 194 for an additional four opioids ( Table 1). We note that our method was not capable of measuring heroin at 195 these low concentrations (section S.2.1). 196 Concentrations of cocaine and two of its metabolites (ecgonine methyl ester and benzoylecgonine) 197 also increased in the weekly samples (Figure 3b, Table S8). Anhydroecgonine, a metabolite specific for 198 crack cocaine (36), decreased in the weekly samples, suggesting the possibility of a shift in local cocaine 199 use patterns (Figure 3b). We saw no trends for methamphetamine, though the party drug TFMPP decreased 200 during the study period ( Table 1, Table S8). Cannabis related compounds did not show a consistent trend. 201 Interestingly THC-A, the non-psychoactive precursor to THC found in raw plant material increased, 202 whereas THC (transformed from THC-A by decarboxylation during heating above 105⁰C for example in 203 cooking or smoking) decreased across the study period ( Table 1, Table S8). 204 The pandemic has increased risk factors for the development of substance abuse disorders and 205 overdoses, such as isolation and economic distress. High COVID-19 related worry has been shown as a 206 predictor of beginning substance use during the pandemic (37), and increasing numbers of overdoses have 207 been reported nationwide (38). An increase in the amount of emergency responses necessary for opioid 208 overdoses has occurred in some locations (39). Locally, there were 36 fatal overdoses during the study 209 period in the towns/cities served by the East Shore Water Pollution Abatement Facility in New Haven (New 210 Haven, East Haven, Woodbridge, and Hamden) (40). Thirty-two of these overdoses involved opioids, 211 including 28 where fentanyl was detected. Cocaine was involved in 17 of the overdose deaths. Most cases 212 included multiple drugs (40). Additionally, the COVID-19 pandemic has caused many changes in 213 treatments for both pain and substance abuse disorders, which usually depend heavily on in-person 214 interactions and carefully controlled access to medications. New systems for opioid distribution and 215 telemedicine appointments have been developed but there is continued concern over their effectiveness 216   We also observed various trends for other pharmaceuticals identified in our analysis ( Table 1,  230   Table S8, Figures S3-S5). Some of these trends are likely related to pandemic-induced changes in 231 behaviour, while others are not. For example, tolycaine, a local anaesthetic used in dental injections (49), 232 decreased in the sludge samples, which corresponds to a decrease in dental appointments during the 233 shutdown (50). Pramocaine, a mild anaesthetic used in over-the-counter creams (51), had increasing levels 234 in sludge which is more likely due to seasonal changes in exposure to insect bites and poison ivy than to 235 pandemic related changes. Diphenhydramine, an allergy medication, also increased during the study period 236 (Table 1, Table S8). 237 Personal care product ingredients and other chemicals 238 We found that benzotriazole, a corrosion inhibitor frequently used on cars and a known contaminant 239 in road dust (52), had trends in sludge that corresponded to the shut down and phase one reopening that 240 occurred during our study period (Figure 5a). There was a decrease in the daily and weekly composite 241 sample concentrations at the beginning of the study period, and then an increase in weekly composite 242 sample levels starting in the weeks before Phase 1 reopening. We hypothesize that the benzotriazole trends 243 are due to changes in the amount of traffic. Doucette et al., found that traffic in Connecticut decreased 43% 244 during the stay-at-home order that began in the first week of our study period (53), and air pollutants related 245 to traffic decreased during stay-at-home orders in other locations (54, 55). With fewer cars on the road, less 246 benzotriazole washes off cars onto the road, and thus less is dissolved the in the runoff water that enters the 247 combined sewer system. Benzotriazole is also used on aircrafts as a de-icer and corrosion inhibitor (56). 248 There is one small airport in the study area that, like many other airports, experienced decreased traffic 249 during the stay-at-home order. Benzotriazole is also used in household dishwasher detergents, which is 250 likely a smaller source to combined sewer wastewater systems. 251 All the UV-filter compounds detected increased in the weekly composite samples (Figure 5b). 252 This trend is likely due to the increase in sunscreen use that corresponds to the seasonal change that occurs 253 in Connecticut between March and June. A slight decrease in oxybenzone levels was observed in the daily 254 samples and the first weekly samples which may be reflective of decreased cosmetic usage during the stay 255 at home order while there was still wintery weather. We suspect that the other trends we found in this 256 category were not affected by the pandemic or stay at home order ( Table 1, Table S8). 257 258 Broader relevance, limitations, and future directions 259 Though our results are specific to the New Haven, CT area, many of the trends that we found are 260 more broadly relevant. We observed increased concentrations for medications whose demand increased 261 during the pandemic (47) and increasing trends for illegal drugs that align with the increasing number of 262 overdoses nationwide (38). Wastewater monitoring can be a way to monitor drug usage during this time 263 when other monitoring strategies have been disrupted by the pandemic (57, 58). Moreover, if wastewater 264 trends can be associated with public heath monitoring data, wastewater-based information can play an 265 important role in providing real-time estimates or early warnings of a variety of infectious and non-266 infectious disease. We note that our results on drugs of abuse differ from those reported by wastewater 267 monitoring programs in Europe, where there has been an overall decrease in illicit drug use (10). 268 Specifically, a study in Austria found decreased use of cocaine, amphetamine, and MDMA during the initial 269 COVID-19 lockdown, which were partially compensated for by increased methamphetamine use (13). 270 They saw no changes in cannabis or methadone related compounds relative to other years (13). 271 Additionally, wastewater monitoring and drug use surveys in Australia have revealed record low levels of 272 fentanyl and oxycodone, but regional increases in cocaine, heroin, methamphetamine, and cannabis (11). 273 The differing trends may be related to differences in pandemic severity and local political responses, but 274 are also reflective of existing trends from before COVID-19; the opioid crisis that is prominent throughout 275 the US has not affected Australia nor Europe to the same extent (10, 11). Trends we observed for 276 pharmaceuticals are more similar to those reported by the Austrian study; though there is some variation 277 individual compound results, both studies show consistent levels of long term medications such as beta-278 blockers and anticonvulsants and lowered levels of short term medications such as analgesics and 279 pharmaceuticals (13). 280 In addition to human health related trends, our results also reveal trends in chemical releases that 281 may affect the environment. Though our samples did not undergo the complete wastewater treatment 282 process, many of the compounds we detected are not fully removed by standard treatment trains (59-61) 283 and are released with the effluent water or sewage sludge. We detected endocrine disrupting compounds 284 including triclocarban, oxybenzone, and sertraline that can have negative impacts on marine organisms and 285 cycle back to humans via consumption of local seafood (62, 63). Much concern has been expressed about 286 the potential ecological impacts of increased pharmaceutical loads in wastewater, particularly in developing 287 areas where wastewater treatment is limited and access to antibiotic and antiviral medications is not 288 controlled by prescriptions (64-67). Spread of resistance to antibiotic and antiviral medications is also a 289 potential concern (64, 67). 290 While our analytical method was designed to include a wide range of chemicals, the scope of any 291 analysis is inherently limited. We intentionally included both liquid and solid portions of primary sludge to 292 measure both hydrophilic and hydrophobic chemicals. However, this prohibited the exact quantification of 293 chemicals in either phase. We therefore are not able to use our data to back calculate per capita consumption 294 as has been done in other wastewater studies (4, 12, 13). Additionally, we designed our sample preparation 295 method for the relatively small volume of sample available from corresponding research on levels of SARS-296 CoV-2 RNA in primary sludge; we could not use solid phase extraction to preconcentrate the liquid portion 297 of our samples, as is common in wastewater studies (59, 60). This likely caused a decrease in the number 298 of liquid phase contaminants we detected. Additionally, our unique method makes our quantitative results 299 difficult to relate to other studies, though trends over time can still be compared. We note that our analytical 300 methods were highly effective, and our sample collection and preparation method was simple, fast, and did 301 not require specialized supplies. Sewage sludge is a well-mixed, concentrated source that does not require 302 complex sampling equipment. Though we collected data over a relatively long period of time in 2020, our 303 sampling campaign did not begin until the pandemic was underway; therefore, we cannot directly compare 304 our results to those from previous years. The data presented in this manuscript represents only a small 305 fraction of what was collected using our high-resolution mass spectrometry methods. We plan to conduct 306 further investigation of chemicals in the sludge that were not easily identifiable using our databases and 307 investigate chemical correlations with measured levels of SARS-CoV-2, as in Wang et al., 2020 who 308 reported statistical relationships between a variety of chemical features and virus RNA levels (12). 309 Overall, the first wave of the COVID-19 pandemic and the related shut down had a significant 310 influence on the chemical fingerprint of primary sludge in New Haven, CT. We found upwards trends in 311 hydroxychloroquine and disinfectant concentrations in sludge, reflecting increased use during the initial 312 wave of the COVID-19 pandemic. We also saw increases in drugs of abuse and antidepressants, and 313 seasonal changes for chemicals such as UV-filters that are used in sunscreens. Importantly, we found that 314 benzotriazole concentrations showed different trends during and after the local stay at home order, a key 315 indication that benzotriazole can be used as a marker for the influence of traffic on wastewater and sludge 316 in combined sewer systems. Overall, our findings relate strongly to trends in public and environmental 317 health worldwide and show specific trends that may not have been picked up in other types of analysis. 318 Sewage sludge surveillance is a promising strategy to monitor a variety human behavioural changes during 319 the pandemic that have public health consequences. 320 When samples were missing, an additional 75 µL was used from an adjacent day to maintain a total aqueous 34 volume of 525 µL per composite (Table S2). In a separate tube, 150 ± 5 mg solids from each day were 35 combined, with double solids added from an adjacent day for missing samples (Table S2)  Three extraction blanks and one spiked solvent sample were extracted alongside each batch of 47 samples. Additionally, we used extra primary sludge (collected in July 2020) to conduct recovery tests and 48 assess variability in our results. Six subsamples of the July sludge were extracted using the daily sample 49 method: 3 replicates spiked with 50 µL of 20 ng/mL standard mixture and 3 replicates unspiked. The spike 50 was added to a 1 mL sample of sludge prior to separation of liquids and solids. 51

Supporting Information
Samples were analyzed using an Ultimate 3000 liquid chromatograph coupled with a Q-Exactive 52 mass spectrometer (Thermo Scientific) and positive electrospray ionization. The LC solvent gradient 53 progressed linearly from 5% B to 95% B for the first 45 minutes, held at 95% B from 45 to 55 minutes, 54 then re-equilibrated at 5% B for an additional 7 minutes. Flow rate was 0.2 mL/min and injection volume 55 was 3 µL. The autosampler temperature was maintained at 10°C and the column oven at 40 °C. Positive 56 electrospray ionization source parameters are shown in Table S3. 57 The full MS-AIF method was run at a resolution of 140,000 with an AGC target of 3e6 and 58 maximum IT of 20 ms. The scan range for full MS was 100 to 1200 m/z. The scan range for AIF was 70 to 59 1050, with stepped NCE at 15, 30, and 45. This method was used for two sample runs on the instrument 60 for weekly and daily samples respectively. 61 For the full MS-ddMS2 methods, the full MS had a resolution of 70,000, AGC target of 3 x 10 6 , 62 maximum IT of 20 ms, and scan range of 100 to 1200 m/z. The ddMS2 scans had a resolution of 17,500, 63 AGC target 2 x 10 5 , maximum IT 100 ms, loop count 5, isolation window 1.0 m/z, and stepped NCE at 15, 64 30, and 45. For determining dd scans, the minimum AGC target was 10 3 , intensity threshold was 10 4 , 65 dynamic exclusion was on (10.0 s), and "if idle…" pick others (instrument scans other ions when not busy 66 with the contents of the inclusion list). Apex peak trigger was on. We used an estimated chromatographic 67 peak width of 15 s. These settings remained the same for all ddMS2 methods, though the inclusion list 68 changed with each run. We had a "targeted" ddMS2 method where the inclusion list contained all of the 69 targeted compounds (Table S1). Additionally, we performed 10 or 11 ddMS2 acquisitions for each 5-week 70 composite sample using the inclusion lists generated by iterative inclusion (described below), to ensure 71 detection of as many compounds as possible. The full MS-ddMS2 method was used in a separate instrument 72 run from the daily and weekly samples and was used only on the 5-week composite samples. 73

74
We used Intelligent Acquisition software to generate inclusion lists to ensure we collected ddMS2 75 spectra for as many of the features in our data as possible. This software is currently in beta testing and is 76 an expansion of existing iterative exclusion software (1, 2). We used the version received from the creators 77 as of July 22, 2020 (code revised July 15, 2020). Iterative inclusion uses Full MS scan data to identify 78 peaks, then generates a series of inclusion lists that can be used to collect ddMS2 data on all the identified 79 peaks. The Intelligent Acquisition method begins with file conversion using MSConvert (3). Settings are 80 shown in Table S4. 81 S7 We generated three batches of inclusion listsone for each of the three 5-week composite samples. 83 Sample files used to generate each batch included the Full MS/AIF data collected for the 5-week composite 84 sample as well as the Full MS/AIF data collected for the 5 weekly samples associated with each 5-week 85 composite (6 input sample files). Three extraction blank Full MS/AIF files were designated as blanks (same 86 files used for all three batches). Files were sorted and labeled as designated in the Intelligent Acquisition 87

Inclusion instructions. 88
Inclusion list generation was completed in R version 4.0.2 (4) using the script provided with 89 Intelligent Acquisition. Settings used for peak selection and list generation are shown in Table S5. We did 90 not use the exact mass library inclusion option. Eleven lists were generated for the first two 5-week 91 composite samples, and 10 for the third. We saved a version of our instrument method with each inclusion 92 list and ran the 5-week composite samples with each of their associated inclusion lists. 93 As a comparison, we also ran each 5-week composite sample using a ddMS2 method with an inclusion list 94 containing only the targeted compounds from the standard mixture. 95 Additionally, peaks were manually curated to ensure accurate integrations. Calibration curve levels 102 ranged from 0.1 ng/mL to 100 ng/mL, and were based on area, weighted 1/x, ignored the origin, and were 103 linear or quadratic, depending on the best fit. Not all compounds were above detection in the lowest 104 calibration standards (Table S1). We note that our sample preparation method was primarily designed to 105 capture as many contaminant compounds as possible and not to provide an exact quantitative relationship  After processing, we filtered the Compound Discoverer results to include only compounds annotated 247 in mzCloud with a match score greater than 75%, that also had ddMS2 spectra for the preferred ion. We 248 further curated the mzCloud results to include only anthropogenic compound unlikely to have a natural 249 source in sludge (e.g. amino acids, lipids, hormones, and compounds common in plants and bacteria were 250 removed). Peak areas for the remaining annotated compounds were exported and used for trend analysis. 251 We note that not all aspects of the CD analysis were used in the data reported in this manuscript (e.g. the 252 ChemSpider results), but we plan to continue investigating the results. 253 S13 S.1.7. Classification of Annotation Confidence 254 All compounds identified by targeted analysis or annotated using screening methods were 255 assigned a confidence level using a customized ranking system based on Schymanski et al., 2014 (Table  256 S7) (6). In the main text Table 1, all level TF-1, TF-1a, and CD-1 compounds are listed as "confirmed", 257 TF-2, TF-2a, CD-2, and CD-2a as "probable", and TF-3 and CD-3 as "tentative". 258 Initially identified at a lower CD confidence level then later confirmed with a standard (retention time, exact mass, and MS2 spectral match)

CD-2a
Probable Structure: mzCloud spectral match Exact mass match, ddMS2 spectra for preferred ion, top mzCloud match score, but other matches also found with score >75%

CD-3
Tentative Candidate: mzCloud spectral match Exact mass match, ddMS2 spectra for preferred ion, mzCloud match score >75%, multiple peaks annotated as the same compound  Figure S1. Briefly, targeted (sludge extract concentration in ng/mL) S14 and suspect screening (peak area) data were analyzed independently due to the different measurement units. 264 Compounds with more than 30% missing values were removed. We explored the temporal trends by 265 conducting linear regressions on both daily and weekly datasets. P-value and adjusted R 2 were used to 266 evaluate the significance of the trends over time. For all analyses, a level of α = 0.05 was used for the 267 determination of significance. 268 We also used multivariate methods to compare each week of daily sample data and each 5-week 269 segment of weekly data. All data was tested for normality using the Shapiro-Wilk test. To test for the 270 variance, Bartlett's test was applied to normal-distribution data; Levene's test was used for non-normal 271 data. Multi-group hypothesis tests were conducted to compare mean level differences among groups of 272 weekly and monthly samples. Seven daily data points were grouped for each week (4 groups); weekly data 273 was grouped in 5-week increments (3 groups). One-way ANOVA and Tukey's HSD, Welch's ANOVA We assessed extraction recovery based on triplicate extractions of spiked and unspiked sludge (six 284 extracts total). Due to the nature of our extraction method, percent recovery is difficult to assess, as 285 compounds partition differently between the solid and liquid phases of the sludge. As per our spiking 286 method (see above), we estimate 100% recovery to be 5 ng/mL but report the extraction results in terms of 287 measured concentration rather than percent in acknowledgement of potential variability ( Figure S2. showing that overall, our extraction method performed well for a wide range of compounds, with low levels 291 of matrix interference. Note that several standard compounds are not included in Figure S2. Lisinopril and 292 a-amantin were not detected in any spiked samples due to high instrument detection limits (Table S1) and 293 amoxicillin, heroin, and oxamyl were only detected in the solvent spiked samples (likely due to a 294 combination of instrument detection limits, matrix interference, and strong sorption to sludge solids.). There 295 was a significant background level of triphenylphosphate, which prohibited its analysis. 296

S18
The injection standard d4-Imidacloprid was added to all samples at 10 ng/mL. For the weekly 301 composite sample run, intensity of the injection standard stayed within ±15 % of the average for all samples 302 with no trend over time, retention time stayed within 2 %, and the average m/z delta was -1.63 ppm. For 303 the daily sample run, intensity of the injection standard stayed within ±25 % of the average for all samples 304 with no trend over time, retention time stayed within 0.3 %, and the average m/z delta was -2.25 ppm. For 305 the ddMS2 sample run, intensity of the injection standard stayed within ±16 % of the average for all samples 306 with no trend over time, retention time stayed within 5 %, and the average m/z delta was -2.83 ppm. 307 We ran QC injections containing our standard mixture every 10-15 samples during each instrument 308 run to monitor for changes in m/z, RT, and abundance. Overall, we found that most standard masses stayed 309 within 2 ppm and all within a range of 5 ppm across the runs. Retention times for most compounds had a 310 relative standard deviation of less than 3 %, with all less than 10%. Abundance levels typically had an RSD 311 less than 20% within each run, with a few outliersparticularly for compounds with higher detection limits. 312 We note that a 0.5 ng/mL standard was used as the QC sample for the weekly sample run, which was below 313 detection for multiple compounds. We used a 5 ng/mL standard for the subsequent ddMS2 run and daily 314 sample run, which produced better results. In the daily sample run, we saw a decrease for most compounds 315 between the standard curve at the beginning of the run and the subsequent QC injections. However, the 316 abundance was similar for all QC injections run between samples, so we assume the trends in our results 317 are valid, if not fully accurate from a quantitation standpoint. 318 Detection information for compounds identified in the sludge samples is provided in Table 1 Table S8 shows the full list of compound annotation results with confidence levels assigned as per Table S7. Adjusted R 2 and p values are 324 provided for all linear regressions for compounds in daily and weekly samples. Additionally, the multi group comparison test used for each analysis 325 is listed alongside the p-values determined. Boxplots of all the statistically significant results from the multi group comparisons are shown in Figure  326 S3 for daily samples, and Figures S4-5