Impact of service line replacement on lead, cadmium, and other drinking water quality parameters in Flint, Michigan

This study elucidates the short- and long-term impacts of lead service line replacement in Flint homes following a corrosion event.


S3
used for MDL and LOQ determination was greater than the MDL but smaller than 10 times the MDL, as is required for the dilution level to be appropriate for MDL and LOQ calculations.
Sample processing, DNA extraction, and qPCR. Within 4 h of sample collection, each water sample was filtered through a 0.2 µm sterile membrane polycarbonate filter (EMD Millipore, Billerica, MA) on top of a 0.45 µm sterile cellulose ester backing membrane (Thermo Fisher Scientific). The filter was then transferred to a 2 ml nuclease free screw-cap tube and stored at -20°C. DNA was extracted from filters with a modified Maxwell® 16 LEV Blood DNA Kit (Promega, Madison, WI). 3 Briefly, each filter was dissolved in 500 µL of a 49:1 chloroformisoamyl alcohol mixture (Sigma-Aldrich, St. Louis, MO). Three rounds of physical and chemical lysis were achieved by bead beating with 0.5 g of zirconium beads (BioSpec Products, Bartlesville, OK) and Maxwell lysis buffer. Extracted DNA was dissolved in 50 µl of molecular grade water and stored at -20°C until qPCR was performed. qPCR was conducted using a RealPlex 2 Mastercycler System (Eppendorf, Hauppauge, NY) to detect total bacteria, Legionella pneumophila, and Mycobacterium and Legionella genera. Targeted genes, primer sequences, amplicon sizes, limit of detection (LOD), LOQ, and standard sources for all qPCR assays are shown in SI Table S3. Each 10 µL reaction contained 1 µL DNA template, 1X Fast Evagreen qPCR Master Mix (Biotium, Fremont, CA), and 0.625 mg/ml bovine serum albumin (Life Technologies, Inc., Waltham, MA). For the quantification of total bacteria, Mycobacterium spp., and L. pneumophila, primers were added to achieve a final concentration of 500 nM, whereas a 400 nM final concentration was used in the Legionella genus assay. All assays had initial denaturation at 95°C for 5 min, except the total bacteria assay (95°C for 2 min). Cycling condition times, temperatures and cycle number varied (SI Table S3). All assays were followed by a melt curve analysis from 55 to 95°C for 20 min after post-cycling denaturation at 95°C for 5 min and S4 annealing at 55°C for 15 s. qPCR reactions were performed in triplicate and results were averaged for all samples and standards. No-template controls were carried out in duplicate for each qPCR run. Samples were quantified from a standard curve consisting of 10-fold serially diluted qPCR standards (10 1 -10 6 ) gene copy/µL. qPCR standards consisted of purified PCR products prepared from either pure culture extracts or extracted environmental sample DNA (SI Table S3) and quantified using the Qubit double-stranded DNA high sensitivity kit (Invitrogen, Waltham, MA).
A Legionella pneumophila strain Lp02 DNA extract was provided by Dr. Michele Swanson's laboratory as the L. pneumophila qPCR assay standard source.
Linear mixed-effects modeling. All linear mixed effects models were comprised of fixed effects (described below) and one random effect (i.e., home number). Prior to model generation, imputation of left-censored data was conducted on data points below the LOD by replacement with one-half the LOD, and censored data between the LOD and LOQ were replaced with the average of those two values. 4 Lognormal transformations of all metals, chlorine residuals, and total bacteria concentrations were performed prior to model selection to normalize data. Collinearity of explanatory variables was assessed using a correlation matrix (SI Table S4). Any variables in the matrix with a significant Kendall correlation were not included together in a model. Manual model selection was conducted using the log-likelihood of each model. Specifically, nested models were used to determine which parameters (i.e., pH, temperature, free chlorine residual, dissolved phosphorus level, percentage of a certain pipe material, and private service line (SL) type) increased the log-likelihood of the model significantly, and those parameters were left in the final model. All mixed-effect models contained only data of the same type of water sample (i.e., distribution system, premise plumbing, or hot water). Visit number (i.e., pre-SL replacement, two weeks and five weeks post-SL replacement) and sampling season were included as categorical S5 variables in all models regardless of log-likelihood values, unless otherwise specified. The lmer() function from the lme4 package was used to generate all models and determine the p-values associated with each explanatory variable. 5,6 Results and Discussion Temporal trends from Spring to Fall 2016 in total bacteria and select opportunistic pathogen abundance in water samples. Total bacteria concentrations ranged from 3.68×10 2 to 1.56×10 9 gene copies/L and Mycobacterium spp. and Legionella spp. concentrations were as high as 1.08×10 7 and 6.18 ×10 6 gene copies/L, respectively (SI Figure S8). Rhoads et al. were detected in 69% of samples in the spring and 53% of samples in the fall. While bacterial levels tend to increase with increasing temperature, our results signify that water quality changes other than temperature changes from Spring 2016 to Fall 2016 had a more profound impact on bacterial levels. Particularly, free chlorine levels inversely correlated with total bacterial concentrations in distribution system samples (SI Table S8), suggesting the addition of extra chlorine to the water in Flint starting in the summer of 2016 10 resulted in successful reduction of bacterial levels.
Total bacterial abundance in water samples before and after SL replacement. Total bacterial levels in each type of sample collected before, two weeks after, and five weeks after SL replacement did not change significantly (Wilcoxon Signed Rank Test, all p-value > 0.05). In addition, the total bacterial concentrations in distribution system samples were significantly lower than those in premise plumbing and hot water samples (Wilcoxon Signed Rank Test, all p-values < 0.05), consistent with previous studies. 11,12 Primary reasons for this trend include the ability of bacteria to grow in stagnant water that typically undergoes disinfectant residual decay and biofilm sloughing during flow changes. 13 Our modeling results for premise plumbing samples support these explanations, as chlorine residual was significantly inversely correlated with total bacterial levels (SI Table S9). Linear mixed-effects models established for total bacteria in premise plumbing samples indicate the positive correlation of increased total metal concentrations (e.g., total copper, total lead, total iron) with increased total bacterial abundance. For instance, a model developed with total lead levels shows a 0.6-log increase in total bacterial levels corresponding to a one-log increase in lead concentrations (SI Table S9). These results indicate the release of S7 particulate metals might be accompanied with biofilm sloughing from piping in premise plumbing samples.
S8 Figure S1. a) Free chlorine and b) orthophosphates levels in Detroit Water and Sewer Department water entering the Flint treatment plant, in water at a Flint tap, and applied to water prior to entering the Flint distribution system, and at a Flint tap throughout 2016. Data were obtained from 2016 City of Flint water treatment plant monthly operating reports.   Figure S5. Profiles of the calculated percentages from each plumbing source (premise plumbing, private SL, and public SL) contributing to the (a) lead and (b) cadmium levels in water. The lead and cadmium concentrations represent the pre-LSL replacement conditions. Gray shading represents the baseline lead/cadmium levels, which indicate the contribution of metal concentrations from water in the distribution system being used at the tap. Red shading is considered to be the additional lead/cadmium contribution from each portion of pipe over the baseline metal level. The data points (red symbols) in each line were calculated by averaging the results of replicate pre-LSL replacement sampling events.   a Initial water sample data taken at these homes are not included in this study because no post-replacement samples were taken.  In addition, any samples in which one or more triplicates did not return a positive cycle value were considered below detection. bp -base pairs S20 Table S4. Kendall's Tau correlation coefficient (above) and associated p-values a (below) obtained using Kendall correlation analysis between all water quality parameters with all sample types. a Cadmium masses were determined by integrating Home 11 total cadmium concentrations of the sequential sample profile obtained over the different portions of pipe in the home. Rectangular integration was used (average concentration of each sample was assumed to be the average concentration of that 1 L sample). b Volumes of water in the premise plumbing, private SL, and public SL segments were determined through a home survey of pipe length, material, and diameter. c Includes the total integrated cadmium concentration of 12 1 L samples. Table S8. Kendall's Tau correlation coefficient (above) and associated p-values a (below) obtained using Kendall correlation analysis between all water quality parameters with distribution system sample type. a The slope (β), confidence interval (CI), and p-value of any fixed effects with a significant impact on the response variable are indicated in bold.