Urinary arsenic profiles reveal exposures to inorganic arsenic from private drinking water supplies in Cornwall, UK

Private water supplies (PWS) in Cornwall, South West England exceeded the current WHO guidance value and UK prescribed concentration or value (PCV) for arsenic of 10 μg/L in 5% of properties surveyed (n = 497). In this follow-up study, the first of its kind in the UK, volunteers (n = 207) from 127 households who used their PWS for drinking, provided urine and drinking water samples for total As determination by inductively coupled plasma mass spectrometry (ICP-MS) and urinary As speciation by high performance liquid chromatography ICP-MS (HPLC-ICP-MS). Arsenic concentrations exceeding 10 μg/L were found in the PWS of 10% of the volunteers. Unadjusted total urinary As concentrations were poorly correlated (Spearman’s ρ = 0.36 (P < 0.001)) with PWS As largely due to the use of spot urine samples and the dominance of arsenobetaine (AB) from seafood sources. However, the osmolality adjusted sum, U-AsIMM, of urinary inorganic As species, arsenite (AsIII) and arsenate (AsV), and their metabolites, methylarsonate (MA) and dimethylarsinate (DMA), was found to strongly correlate (Spearman’s ρ: 0.62 (P < 0.001)) with PWS As, indicating private water supplies as the dominant source of inorganic As exposure in the study population of PWS users.

a considerable number of people in the region may be subject to elevated levels of As in their drinking water, an exposure route not comprehensively investigated in Cornwall, nor indeed the UK as a whole, to date.
The identification of elevated concentrations of As in drinking water alone can help provide an indication of the population at risk. However, the use of exposure biomonitoring, the analysis of biological material for the presence of chemicals and their metabolites, allows for a more direct quantification of internal exposure 24 underpinning environmental chemical attributable health risks. A common approach to As biomonitoring is the analysis of urine samples for inorganic arsenite (As III ), arsenate (As V ) and methylated metabolites methylarsonate (MA) and dimethylarsinate (DMA) that are excreted in the urine following metabolism in the liver 25 . It is accepted that post intake, inorganic As V is reduced to As III followed by methylation to MA and DMA 26 . The process was formerly considered to be a detoxification pathway, but findings of genotoxic intermediate trivalent forms of MA and DMA suggest otherwise 27 . The exact mechanisms of As biomethylation are subject to ongoing investigation 28 . For the purpose of exposure assessment, the methylation of inorganic to organic As species justifies the quantification of MA and DMA, whilst acknowledging that direct intake of both of these species from dietary sources has been reported 29 . The majority of As is excreted within 4 days of dosage 30 , making urinary As a useful measure of recent exposure, and has been used, for example, to demonstrate rice as a significant dietary exposure pathway in rice consumers in the UK 31 , USA 32 and West Bengal 33 . Several studies have used urinary As to model the risk of health end-points and toxicological responses resulting from exposure to inorganic As. These include type 2 diabetes 34 , a mortality follow-up of a population with baseline urine measurements which found a significant association with lung cancer 35 and increased genotoxicity measured by micronuclei frequency in urothelial cells 36 . Urinary As biomonitoring, albeit on a small number of volunteers and in relation to soil and dust exposure, has been carried out in Cornwall on two previous occasions 37, 38 and elevated concentrations were observed relative to control areas with low environmental As.
A number of considerations need to be taken into account when urinary As is used as a biomarker of exposure. Firstly, total urinary As results can be influenced by high concentrations of arsenobetaine (AB), an organo-arsenical found in seafood, widely thought to be non-toxic 39 and readily excreted unaltered following dietary intake. This makes it necessary to perform speciation analysis on urine samples to quantify the individual As species and exclude the contribution from AB which does not reflect exposure to more hazardous environmental inorganic As. Secondly, the variation in hydration status among volunteers means that both first morning void (FMV) and spot urine samples differ markedly in their dilution, both giving imperfect estimates of 24 hr excretion 40 . Therefore, in order to be used as a robust indicator of exposure, urinary As concentrations require adjustment for dilution to eliminate variation from fluid balance. Creatinine and specific gravity (SG) adjustment are widely used, but both methods are susceptible to interferences. Variation in urinary creatinine has been demonstrated between demographic groups 41 and in response to variations in muscle mass 42 and malnutrition 43 , while possibly a more relevant deterrent of applying this adjustment factor is its observed relationship with As methylation efficiency 44 . Alternatively, because SG is routinely measured by refractometry, the presence of urinary solutes such as protein (proteinuria), glucose (glucosuria) and ketones (ketonuria) alters the refractive index of the liquid irrespective of its dilution, thus giving inaccurate dilution estimates 45 . One alternative adjustment factor, seldom used in biomonitoring studies, is urinary osmolality. Previously overlooked due to the lack of widespread availability and relative cost of the instrumentation required 46 , osmolality is regarded as the 'gold standard' and definitive measure of urinary concentration in the clinical and veterinary sciences community 47 . In the case of cryoscopic osmometry, freezing point depression is measured. Freezing point is a colligative property reflective of solute content, expressed here by osmolality (osmoles of solute per mass unit of solvent) and is not vulnerable to the same interferences as SG measurement by refractometry. Given the absence of 24 hr or timed excretion data in the present study, osmolality adjustment was preferred over the two alternative options.
This study aimed to: (1) assess human exposure to inorganic environmental As in a population of PWS users in Cornwall using non-invasive urinary As exposure biomonitoring, (2) assess to what extent the biomarker response can be attributed to PWS drinking water as an exposure route and (3) observe the effect of osmolality adjustment to better define the relationship between urinary As and PWS drinking water As.

Results
Study group demographics. The extent of the study area and spatial distribution of households is shown in Fig. 1. A total of 215 volunteers from 129 households participated in the study. Of these volunteers, 207 from 127 households consisting of 108 males (52%) and 99 females (48%), reported using their water for drinking and provided both a drinking water and urine sample. Henceforth, unless otherwise stated, this sub-group will be the focus of the present article. The mean volunteer age was 62 years old (range: 18-90). The age and gender distribution is shown in Fig. 2. The study group was classified as a 99% rural population (see supplementary information).
PWS drinking water and urine samples. Summary statistics for total As in drinking water samples and total and speciated As in urine samples (unadjusted and osmolality adjusted) are displayed in Table 1 and plotted in Fig. 3. Geometric means (GM) were calculated in addition to arithmetic means (AM) as the data were positively skewed. Of the 127 households, 126 (99%) had detectable (> 0.02 μ g/L) As in their drinking water, 62 (49%) had ≥1 μ g/L and 15 (12%) exceeded the current WHO guidance value 48 and UK PCV 23 of 10 μ g/L. This corresponds to 21 of the 207 (10%) volunteers being exposed to drinking water As concentrations above 10 μ g/L. The maximum PWS drinking water arsenic concentration was 233 μ g/L.
All volunteers had detectable (> 0.2 μ g/L) concentrations of unadjusted urinary total As; with a maximum observed concentration of 426 μ g/L. Speciation data yielded a 98% mean recovery of total As and precision, expressed as relative standard deviation (RSD), was 9%. Despite requesting volunteers to refrain from eating seafood for the 4 days prior to sample collection, a large contribution of total As was from organic AB. Arsenobetaine was detected (LOD 1.3 μ g/L) in 152 (73%) samples whilst the mean contribution of AB to total urinary As was Scientific RepoRts | 6:25656 | DOI: 10.1038/srep25656 49%; (range: 0-98%). Findings of inorganic As III and As V were lower, with 56 (27%) and 10 (5%) of samples having detectable concentrations (> 0.8 μ g/L; > 1.5 μ g/L) respectively. The sum of As III and As V ranged from < LOD (0.8 μ g/L and 1.5 μ g/L respectively) to 19.2 μ g/L. All samples had detectable concentrations of DMA and 107 (52%) had detectable concentrations of MA. Dimethylarsinate was the dominant arsenic species with the exception of AB. The sum of inorganic As (As III and As V ) and its organic methylated metabolites (MA and DMA), referred to here as U-As IMM , ranged from 0.9 to 124 μ g/L with an arithmetic mean (AM) of 9.0 μ g/L and a GM of 5.8 μ g/L.
Urinary osmolality ranged from 181-1161 mOsm/kg, reflecting a large variation in urinary dilution amongst volunteers. Post osmolality adjustment, AM urinary total As moderately decreased from 36.8 to 36.1 μ g/L and the GM slightly increased from 15.8 to 17.1 (range: 2.2-404 μ g/L). The osmolality adjusted U-As IMM AM and GM also decreased to 8.6 and increased to 6.3, respectively.
Additionally, 30 (14%) urine samples were collected as spot samples at the time of visit as opposed to first morning voids (FMV). To address this, a Welch's independent two-group t-test was used to assess the difference between the two collection methods. For unadjusted and osmolality adjusted U-As IMM and urinary osmolality no significant difference was observed (P = 0.20, P = 0.30 and p = 0.43, respectively).  Correlation analysis. Scatterplots showing urinary As vs PWS drinking water total As, both before and after AB and dilution adjustment, are shown in Fig. 4. Figure 4a shows that total As in drinking water was not a good predictor of urinary total As, with a large variation in urinary total As even for volunteers with low PWS drinking water As concentrations. However, when corrected for AB ( Fig. 4b) a more positive correlation was observed. Correcting for urinary dilution using osmolality measurements further improved the correlation between urinary As (U-As IMM ) and PWS drinking water As (Fig. 4c). To test the strength of these correlations, Spearman's rank correlation coefficient was used as both variables were non-normally distributed (Shapiro-Wilk test: P < 0.001 for drinking water total As, and both unadjusted and osmolality adjusted urinary total and U-As IMM ) and the results from this analysis are shown in Table 2. Following adjustment for AB, a stronger correlation was observed between drinking water and urine samples (Spearman's ρ = 0.36 (P < 0.001) and 0.58 (P < 0.001) pre and post AB exclusion respectively). This correlation strengthened slightly (Spearman's ρ = 0.62, P < 0.001) following osmolality adjustment. The correlation between creatinine adjusted U-As IMM (μ g/g Cre) and drinking water As is also shown in comparison to unadjusted and osmolality adjusted results (Fig. 5) and is weaker (Spearman's ρ = 0.53, P < 0.001) than both. In addition, correlations were calculated on subsets of different drinking water As concentrations and were found to weaken with decreasing concentration. For drinking water As versus osmolality corrected U-As IMM , Spearman's ρ was 0.81 (P < 0.001) when drinking water As was >10 μ g/L compared to 0.21 (P = 0.031) when < 1μ g/L. This is shown in Fig. 6.
Finally, 74 households consisted of >1 volunteer, all of whom were included in correlation analyses. Volunteers (observations) sharing a household (sampling unit) were therefore not independent. Correlations between U-As IMM concentrations of volunteers from the same household (n = 74) were calculated as ρ = 0.59 (P < 0.001) and ρ = 0.66 (P < 0.001) for unadjusted and osmolality adjusted concentrations, respectively. This had the potential to influence the strength of correlations and, therefore, correlations were re-calculated by randomly selecting one volunteer per household for inclusion. These results are presented in Table 3 and, although some correlations (particularly those calculated for lower drinking water As concentration groups) were numerically different, the overall pattern remained the same. Furthermore, the correlations re-calculated on osmolality adjusted U-As IMM concentrations agreed strongly across drinking water concentrations groups with those originally calculated with the inclusion of all volunteers.

Discussion
The present study shows that exposure to inorganic As in drinking water, although not widespread, is occurring within the Cornwall study population with 10% of the present study group exposed to > 10 μ g As/L in drinking water. Although not a true representation of the actual proportion of population exposure, this study builds on the findings 22 of its precursor survey by confirming human exposure from PWS that exceeded the PCV, with high As concentrations in drinking water reflected by dilution and AB adjusted U-As IMM .
The maximum U-As IMM concentration measured in the present study (124 μ g/L) was comparable with values found in West Bengal 49 , one of the world's worst affected regions, some of the highest recorded elsewhere in Europe 50 , and was higher than any found previously in Cornwall 37,51 . In 1998 Kavanagh and co-workers 51 reported a range of 2.7-58.9 μ g As/g creatinine (U-As IMM ) in urine collected from residents (8 boys aged 3-8; 9 adults aged 30-43) of Gunnislake, Cornwall, although the drinking water supply status of the volunteers was not reported. This demonstrates that the larger sample population in this study revealed further exposure incidences in the region and a previously uninvestigated exposure route, both in Cornwall and the UK to date.
Correlation analysis of exposure and response variables showed that the strength of the correlation between drinking water and U-As IMM reduced with decreasing levels of exposure to total As in drinking water. Variation among U-As IMM results in volunteers with < 1 μ g/L in drinking water was evident, with some urinary U-As IMM  results still higher than 10 μ g/L. As mentioned, Cornwall is an area of high environmental As and these observations suggest that in low drinking water As concentration scenarios, confounding exposure variables such as direct soil ingestion from home grown produce consumption, dust ingestion/inhalation or contact with high As bearing mine wastes could be more prominent. The importance of these exposure routes will be the focus of further research incorporating the analyses of garden/vegetable patch soils and household dust.  (c) Osmolality adjusted U-As IMM . Linear regression lines are for reference only. A poor relationship between drinking water total As and unadjusted urinary total As is evident (a) due to seafood intake and the large contribution of AB on urinary total As results. This is illustrated by the red dashed line showing high urinary total As results at low drinking water As exposure.  Table 2. Correlation analysis of exposure and outcome variables for all volunteers. A strong correlation (bold font) is only observed for U-As IMM (osmolality adjusted) for drinking water As > 10 μ g/L. All household volunteers were included in analyses.
Additionally, with the exception of AB, DMA was the dominant species measured in urine samples. This is not unexpected, as DMA is the major endpoint of As metabolism in mammals, typically accounting for 60-80% of stable urinary As species excluding AB 52 . This outcome requires further consideration given the low drinking   Fig. 3c plotted on log scale axes to show contrasting exposure-response relationships of participants exposed to different concentrations of As in drinking water. Spearman's correlation coefficients (ρ ) are displayed for the different drinking water As ranges.

Table 3. Correlation analysis of exposure and outcome variables for single volunteers per household.
A strong correlation (bold font) is only observed for U-As IMM (osmolality adjusted) for drinking water As >10 μ g/L. One volunteer per household was chosen at random for inclusion in analyses.
water concentrations of the majority of individuals. This is in agreement with the study of Leese et al. 29 who reported high concentrations of AB in urine samples from an unexposed population 29 , in which DMA was also the dominant species after AB. Given the unexposed status of their study population, Leese et al. 29 conclude that dietary sources are responsible for the presence of DMA as well as AB. In addition, they advise that organic methylated species in urine samples do not necessarily indicate exposure to inorganic As. In the case of individuals not exposed to As in their drinking water, future efforts should be made to model the proportion of DMA likely to derive from direct dietary intake versus that excreted as a product of the metabolism of inorganic species.
No robust reference value for U-As IMM applicable to a UK population currently exists and existing values applicable elsewhere are discussed. Commonly cited is the Agency for Toxic Disease Registry (ATSDR) 100 μ g/L total urinary As 53 . This was not selected for comparison in the present study due to the large contribution of AB to urinary total As and unless seafood consumption can be categorically ruled out then this value is not recommended. Of 207 urine samples, 12 (6%) exceeded 21.5 μ g/L of unadjusted U-As IMM , the approximate creatinine adjusted concentration found in a recent study 35 to correspond to a lung cancer hazard ratio (HR) of 2.0 which is equivalent to double the risk of developing the disease. This value is more appropriate for comparison as it is not affected by AB, however it is noted that because it refers to creatinine adjusted urinary As results from a sample of almost 4000 American Indians, it is not directly applicable to the group studied here. An arguably more appropriate value is the occupational biological effect index (BEI) provided by the American Conference of Government Industrial Hygienists 54 (ACGIH) (35 μ g/L of unadjusted U-As IMM ), of which 8 (4%) samples exceeded. Whilst acknowledging that this was derived for use with occupational exposure, the BEI was chosen as the comparison value provided to volunteers on feeding back their individual urinary As results (unadjusted U-As IMM ).
In order to assess the magnitude of exposure it is important to consider how the sample in the present study relates to the underlying population of PWS users in Cornwall and elsewhere in the UK. As demonstrated in Fig. 2a, the sample of volunteers obtained in the present study was biased and is unlikely to reflect the true proportion of exposure in the underlying population. Furthermore, high-As bearing PWS were over-sampled to ensure that a range of exposure scenarios were captured to model the biomarker response. Therefore, the proportion of drinking water As PCV exceedances in the present study is higher than that observed in the wider population of PWS users (12% in the present biomonitoring study versus 5% in the 2011-2013 PWS survey).
The relationship between the current sample and the underlying population is a matter for further investigation.
Urinary As biomonitoring is useful in assessing recent exposure 25 , and therefore results offer a snapshot of a relatively narrow exposure window, especially given that FMV/spot samples were taken as opposed to 24 hr collections, making it impossible to assess day-to-day variation in individual excretion patterns. Chronic exposure to As cannot be fully assessed by exposure incidence alone, an assessment of longevity is also needed. The analysis of alternative biomarkers such as hair and toenails is ongoing and may provide evidence of longer term exposure, as will analysis of the temporal stability of As in drinking water samples.
In conclusion, it has been demonstrated that, following the necessary adjustments of urinary As concentrations for AB intake and urinary fluid balance, a strong positive correlation was observed between As concentrations in PWS drinking water and urinary As excretion-indicative of ongoing human exposure to inorganic As in PWS drinking water in Cornwall. Given the comparisons to existing guidance values for other populations, the results of the present study are a cause for concern, albeit for a minority of cases. Efforts should be made to raise wider public awareness of the potential hazards associated with PWS usage and, where analytes exceed the PCV, recommendations for treatment should be made given that it has been demonstrated 55 that installation of appropriate treatment systems is effective in reducing exposure to As and other elements. This work has raised points for further investigation which should include: whether chronic/long-term exposure is evident; the importance of additional exposure routes; further refinement of As biomonitoring techniques to account for dietary sources of organic As species in addition to AB; identification of specific population groups at risk. Such groups may be dictated geographically or as a result of individual susceptibility or behavioural risk factors. Particular 'hotspots' of high exposure require identification using spatial/geostatistical methods and ongoing questionnaire analysis. Finally, the health implications of PWS usage in the UK warrant more investigation by detailed analysis of supply distribution, consumption patterns, geochemical risk modelling in conjunction with health surveillance datasets.

Methods
Ethical approval and consent. In accordance with approved guidelines, written informed consent was obtained from all volunteers and only those who were able to provide such were included in the study. In addition, all methods were followed in accordance with approved guidelines. Ethical approval for the study was provided Sampling strategy and recruitment methods. The sampling frame consisted of volunteers previously involved in the 2011-2013 PWS survey carried out by the BGS on behalf of the former Health Protection Agency (HPA), now part of Public Health England (PHE). Households with a PWS at which volunteers resided formed the sampling units. Observational units consisted of those individual volunteers who met the following inclusion criteria: ≥ 18 years of age; did not suffer from a health condition that could prevent them from participating in the study; had not been identified from the previous phase as unwilling/unable to participate further; provided informed consent. Prospective volunteers were contacted via an information/invitation letter prior to receiving a telephone call. All of those with > 1 As μ g/L being found in their drinking water in the previous survey were contacted to include as many as possible in the study. Numbers were then made up with households in the < 1 As μ g/L category. This approach was designed to maximise the range of observed exposures in the study group. Sample collection and pre-treatment. Household visits were made to volunteers by sampling teams.
Urine and point of use drinking water samples were collected and an exposure assessment questionnaire administered to volunteers using Microsoft Access 2007 on a laptop/tablet device to ascertain whether volunteers were using their PWS for drinking.
Drinking water samples were collected by running the tap most frequently used for drinking for a minimum of 3 minutes to purge any standing water from the pipes before collecting the water in pre-rinsed (with the water being sampled) LDPE containers (Nalgene, USA). Samples were stored in a cool box during transit. Samples were acidified with 1% v/v HNO 3 on return to the field laboratory, and then with an additional 0.5% v/v of HCl on return to the Inorganic Geochemistry Facility at the British Geological Survey.
For urine collection, volunteers were asked to refrain from eating seafood for a minimum of 4 days prior to providing a sample. HDPE containers (60 mL) (Nalgene, USA) were mailed in advance to volunteers who were asked to provide a FMV, mid-stream urine sample on the day of their visit and store it in the refrigerator until collection by the sampling team. Where instructions were not followed (n = 30), a spot urine sample was collected at the time of the visit where possible. Samples were stored in a cool box during transit and, on return to the field laboratory, filtered through 0.45 μ m Acrodisc ® syringe filters (PALL Life Sciences, USA) into 30 mL HDPE containers (Nalgene, USA) and then frozen at −30 °C until analysis.
Arsenic calibration standards were prepared from an in-house multi-element stock in which the As contribution was from a 1000 mg/L PrimAg ® grade mono-elemental stock solution (Romil, UK). Arsenic QC standards (5 μ g/L) were prepared from a multi-element stock solution of various concentrations with As at 20 mg/L (Ultra Scientific, USA). A Tellurium (Te) ICP-MS internal standard was prepared from a PlasmaCAL 10,000 mg/L stock solution (SCP Science, Canada). The following standards were used for the calibration of individual As species as follows: As III : 1000 As mg/L stock solution of arsenic trioxide (As 2 O 3 ) (Inorganic Ventures, USA); As v : 1000 As mg/L stock solution of arsenic (V) oxide hydrate (As 2 O 5 ·xH 2 O) (Inorganic Ventures, USA); MA: 50 As mg/L in-house stock solution of monomethylarsonic acid ((CH 3 AsO(OH) 2 ) prepared from solid (Sigma-Aldrich, USA); DMA: 50 As mg/L in-house stock solution of dimethylarsinic acid ((CH 3 ) 2 AsO(OH)) prepared from solid (Greyhound Chromatography, UK); AB: 1031 As mg/L BCR-626 standard solution of arsenobetaine ((CH 3 ) 3 As + CH 2 COO − ) (LGC, UK).

Total arsenic determination by ICP-MS.
Urine samples were thawed at room temperature and refrigerated at 4 °C prior to analysis. Due to the high matrix of urine, samples (1 mL) were diluted x10 with 1% v/v HNO 3 and 0.5% v/v HCl to reduce the effects of high concentrations of sodium (Na) on signal stability. Acidified PWS drinking water samples were refrigerated at 4 °C prior to analysis and analysed neat. Total As concentrations in both water and urine samples were determined using inductively coupled plasma mass spectrometry (ICP-MS). An Agilent 7500 Series ICP-MS instrument (Agilent Technologies, USA) was used under the operating conditions described by Watts et al. 56 . The instrument was fitted with a MicroMist low-flow nebulizer (Glass Expansion, Australia) and sample introduction was accelerated using an ASXpress rapid sample introduction system (Teledyne CETAC Technologies, USA). A three-point calibration was used with As concentrations at 1, 10 and 100 μ g/L. Arsenic was detected in helium (He) collision cell mode to reduce potential mass 75 polyatomic interferences such as argon chloride ( 40 Ar 35 Cl + ). A Te internal standard was introduced simultaneously via a T-piece and the Te signal response used to fit urinary As data. The limits of detection (LOD) were calculated as 3σ of analytical run blanks and were 0.02 and 0.2 As μ g/L for drinking water and urine samples respectively.

Arsenic speciation by HPLC-ICP-MS.
Urine samples (150 μ L) were diluted × 10 with deionised water and As speciation was measured using high performance liquid chromatography coupled to ICP-MS (HPLC-ICP-MS) using the method described by Button et al. 57 . In summary, a GP50 gradient pump and an AS auto-sampler (Dionex, USA) were coupled to the ICP-MS instrument with PEEK tubing. Chromatography was performed with a PRP-X100 anion exchange column and a PRP-X100 guard column (Hamilton, USA) using gradient elution with the mobile phase (pH 8.65, 1 mL/min) alternating between 4 and 60 mM NH 4 NO 3 . A 3-point calibration was used with 1, 10 and 50 As μ g/L solutions of As III and a mixed solution of 1, 10 and 50 As μ g/L As V , MA, DMA and AB. Figure 7 shows a standard chromatogram obtained for calibration solutions. The LODs for this method (3σ of blank values) are reported by Watts et al. 58 : 0.8; 1.5; 0.7; 0.3; 1.3 As μ g/L for As III , As V , MA, DMA and AB respectively. It is noted that this method cannot distinguish the trivalent and pentavalent forms of both MA and DMA which vary in genotoxicity 59 .
Statistical analysis. Statistical analysis (including the production of exploratory plots) was performed using R version 3.0.0 (base package) 62 . Welch's independent two sample t-test was used to assess the difference between results of spot and FMV urine collections. A Shapiro-Wilk test was used to determine the normality of exposure and outcome variables before and after applying log transformation. Correlation tests were performed using Spearman's rank correlation coefficient accompanied by a significance test to exclude the possibility of the observed correlations resulting from random sampling. Descriptive statistics, with the exception of the geometric mean, were obtained using the 'psych' package 63 . In the case of speciation data, where manual peak integration resulted in samples with zero or negative values for particular species (As III and As V ), left censoring was required to enable data for log transformation and the calculation of geometric means. Values < LOD were therefore replaced with that of half the appropriate LOD.
Mapping. All maps displayed as figures in this manuscript were compiled using ESRI ArcGIS Desktop version Dissemination of results to households. A letter containing individual result data was fed back to households. Where a PCV exceedance was highlighted, specific advice was provided to participants on any potential health risks and suggested corrective actions were given. All participants were provided with appropriate contact details for any follow-up enquiries. The letter and guidance were developed by PHE along with BGS and the Local Authority. The letter was sent from the Local Authority, as the regulator for PWS in England. Figure 7. Arsenic speciation standard chromatogram. Chromatograms obtained for standard calibration solutions at 1, 10 and 50 μ g/ L. Calibration of arsenate (As V ), methylarsonate (MA), dimethylarsinate (DMA) and arsenobetaine (AB) was performed with mixed solutions of arsenic (V) oxide hydrate (As 2 O 5 ·xH 2 O), monomethylarsonic acid ((CH 3 AsO(OH) 2 ), dimethylarsinic acid ((CH 3 ) 2 AsO(OH)) and arsenobetaine ((CH 3 ) 3 As + CH 2 COO-) respectively. Calibration of arsenite (As III ) has been plotted simultaneously and was achieved with separate solutions of arsenic trioxide (As 2 O 3 ).