Estimating Potential Environmental Loadings of ... - ACS Publications

Environ. Sci. Technol. 2004, 38, 3370-3380

Estimating Potential Environmental Loadings of Cryptosporidium spp. and Campylobacter spp. from Livestock in the Grand River Watershed, Ontario, Canada SARAH M. DORNER,* PETER M. HUCK, AND ROBIN M. SLAWSON† NSERC Chair in Water Treatment, Department of Civil Engineering, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada N2L 3G1

Exposure to waterborne pathogens in recreational or drinking water is a serious public health concern. Thus, it is important to determine the sources of pathogens in a watershed and to quantify their environmental loadings. The natural variability of potentially pathogenic microorganisms in the environment from anthropogenic, natural, and livestock sources is large and has been difficult to quantify. A first step in characterizing the risk of nonpoint source contamination from pathogens of livestock origin is to determine the potential environmental loading based on animal prevalence and fecal shedding intensity. This study developed a probabilistic model for estimating the production of Cryptosporidium spp. and Campylobacter spp. from livestock sources within a watershed. Probability density functions representing daily pathogen production rates from livestock were simulated for the Grand River Watershed in southwestern Ontario. The prevalence of pathogenic microorganisms in animals was modeled as a mixture of β-distributions with parameters drawn from published studies. Similarly, Γ-distributions were generated to describe animal pathogen shedding intensity. Results demonstrate that although cattle are responsible for the largest amount of manure produced, other domesticated farm animals contribute large numbers of the two pathogenic microorganisms studied. Daily pathogen production rates are highly sensitive to the parameters of the Γ-distributions, illustrating the need for reliable data on animal shedding intensity. The methodology may be used for identifying source terms for pathogen fate and transport modeling and for defining and targeting regions that are most vulnerable to water contamination from pathogenic sources.

Introduction To ensure that drinking water supplies are adequately protected and treated, it is important to understand the risk of exposure to potentially pathogenic waterborne microorganisms because they represent acute threats to public health. * Corresponding author e-mail: [email protected]; phone: (519)888-4567, ext 3828; fax: (519)746-7499. † Present address: Department of Biology, Wilfrid Laurier University, 75 University Avenue West, Waterloo, Ontario, Canada N2L 3C5. 3370

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 12, 2004

The highly transient and variable nature of pathogenic microorganisms in the environment complicates efforts to estimate or model loadings of these contaminants in water supplies. The importance of quantifying the environmental loadings of pathogens has been emphasized by recent work by Atwill et al. (1), who estimated daily loading of Cryptosporidium parvum by adult beef cattle in California. Waterborne disease outbreaks have frequently been associated with intense hydrologic events that carry heavy pathogenic loads (2), which suggests that nonpoint sources are important origins of fecal contamination. It is essential that drinking water treatment plants be capable of handling peak concentrations of pathogens. Unlike other forms of nonpoint source contaminants such as sediment and nutrients, pathogenic microorganisms are much more transient in nature and may be clustered in space and in time (e.g., ref 3). It is not possible to directly measure source terms for pathogen fate and transport modeling. The true risk from pathogens is difficult to determine as the data to capture their natural variability are frequently not available. There exists a high level of uncertainty in methods used for enumeration of pathogens such as Cryptosporidium. Furthermore, not all methods for detection and enumeration can differentiate between organisms that are nonviable, viable, or infective. Recent studies have demonstrated that indicators of fecal contamination are not well correlated to pathogenic microorganisms (e.g., ref 4), thus it is important to understand the origins, fate, and transport of the pathogens themselves to comprehend the true risk they pose to drinking water treatment utilities as well as to recreational users of water resources. A number of microorganisms have emerged as being of potential concern to drinking water utilities, and these include viruses such as enterovirus, calicivirus, hepatitis, and Norwalk virus; protozoa such as Cyclospora cayetanensis, humanpathogenic microsporidia, and Toxoplasma gondii (5); and bacteria such as the Mycobacterium avium complex, Helicobacter pylori, pathogenic Escherichia coli, and Campylobacter jejuni (6). Cryptosporidium spp. are protozoan pathogens that have been a major driver for drinking water treatment design and operation in recent years. This is because of their ability to survive in the environment (7) and resistance to conventional (chlorine) disinfection (8). A number of outbreaks have been caused by consumption of contaminated or inadequately treated water (8). Campylobacter spp. are bacterial pathogens that have been reported worldwide as being among the most common causes of diarrheal illnesses (9). Nonpoint sources of pathogenic contamination are more difficult to quantify than point sources or other forms of contaminants such as sediments or nutrients. Pathogens are usually present sporadically, depending on the disease status of the human and animal populations in a given region, thus predicting their occurrence poses a greater challenge than contaminants with less variable sources, such as sediments. This paper presents a method to quantify nonpoint sources of pathogenic production in a watershed illustrated with two examplessCryptosporidium spp. and Campylobacter spp. Although these organisms have been identified as causing waterborne disease, their presence in farm animals may also pose a health risk to farm workers and rural residents. In an analysis of endemic cryptosporidiosis cases in Ontario in 1996-1997, surface water was the most commonly reported probable source of infection (26% of 157 cases), followed by livestock (21% of 157 cases) (10). 10.1021/es035208+ CCC: $27.50

 2004 American Chemical Society Published on Web 05/07/2004

FIGURE 2. Major livestock distribution within the entire Grand River Watershed.

FIGURE 1. Location of the Grand River Watershed in southwestern Ontario. Animal Prevalence. Cryptosporidium has been demonstrated to be present in all types of farm animals including cattle (e.g., ref 11), sheep (e.g., ref 12), pigs (e.g., ref 13), and poultry (e.g., ref 14). Domesticated farm animals may carry multiple species of Cryptosporidium (e.g., ref 15), and immunocompromised individuals may be susceptible to a wide range of species and genotypes (16). Until recently, only C. parvum was believed to be of public health significance for immunocompetent individuals (8), but reports have demonstrated that C. meleagridis, originally associated with birds, has infected human immunocompetent hosts (17). Campylobacter spp. have been reported in cattle (e.g., ref 18), sheep (e.g., ref 19), pigs (e.g., ref 20), and poultry (e.g., ref 21).

blended Grand River water with groundwater for Kitchener, Waterloo, and several smaller communities. The Brantford and Oshwegan treatment plants obtain their entire water supply from the Grand River. There are 26 wastewater treatment plants servicing approximately 680 000 people that discharge into the Grand River and its tributaries. All sewage receives secondary treatment, with 19% treated to the tertiary level. However, a study examining pathogen levels in several areas within the Grand River Watershed found no significant increases downstream of wastewater treatment plants (23). Wastewater may be a greater source of pathogenic contamination during spills or sewage bypasses. Probabilistic Model of Daily Pathogen Production. Loadings of pathogens from nonpoint sources were estimated from livestock numbers from the 2001 Canadian Census of Agriculture (22) using a probabilistic approach with data from the literature regarding animal pathogen prevalence and shedding intensity. β- and Γ-distributions were used to model the probability distributions of pathogen prevalence and shedding intensity, respectively. Estimation of Pathogen Prevalence in Animals. β-distributions for modeling animal prevalence were selected because the random variable is restricted to a finite interval (i.e., from 0 to N), and the distribution can take on a variety of shapes allowing for a high level of flexibility in fitting the distribution to data. A similar model was used for describing the prevalence of Cryptosporidium and Giardia in animals by Medema et al. (24). The probability that an animal is positive (Pap) is described as

Theoretical Basis and Calculations Study Area. The case study region is the Grand River Watershed located in southwestern Ontario, Canada (Figure 1). The eastern boundary of the watershed is located approximately 80 km west of Toronto. With an area of close to 7000 km2, the Grand River Watershed is the largest in southern Ontario. The total population living in the watershed is greater than 800 000 and is expected to grow by 37% in the next 20 yr. The watershed is intensively farmed with 80% of the land being used for agriculture, including pasture. The central portion of the watershed is heavily urbanized with approximately 500 000 people living in five citiessBrantford, Cambridge, Guelph, Kitchener, and Waterloo. Forest covers 14% of the land but reaches 30% in the eastern portion of the watershed. Figure 2 presents the major livestock distribution for the entire watershed, estimated from the 2001 Canadian Census of Agriculture (22). Water supply in the watershed is from both groundwater and surface water. In the northern region of the watershed, groundwater is used exclusively. In the central and southern regions, three water treatment plants draw water from the Grand River. The Mannheim water treatment plant supplies

Pap ) β(Rb, βb)

(1)

Rb ) r + 1

(2)

βb ) s - r + 1

(3)

where

For each census category (e.g., calves 1 yr and under), r is the number of positive animals and s is the number of animals sampled. A positive animal is an animal from which a fecal sample has been obtained, and it has been demonstrated that the pathogen is present in the fecal sample in numbers at or above the detection limit for the microbiological method used for testing. In the present study, the parameters r and s were drawn from published studies examining pathogen prevalence in animals. The individual β-distributions in the mixture were not weighted, although it is possible to do so. The probability of y positive animals in the ith census category, P(yi), was modeled as a mixture of generalized β-distributions: VOL. 38, NO. 12, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3371

TABLE 1. Cryptosporidium Data for β-Distributions

TABLE 2. Campylobacter Data for β-Distributions

animal type

ref

% positive

n sampled

rbi,k

βbi,k

calves, under 1 yr

15 29 30 31 32 33 11 34 35 36 37 33 30 15 37 33 38 31 35 33 31 35 39 30 39 36 39 13 30 40 30 12 30 41 36 12 30 42 14 43 44 45 43 43 45

24.6 25.0 16.0 1.5 47.9 29.6 5.0 17.9 3.9 11.2 62.4 17.8 9.3 40.2 62.4 17.8 71.8 1.5 2.7 7.7 1.7 2.2 0 100 29.6 19.6 34.3 92 5.1 11 27.3 7.8 22.5 37.8 9.5 59 17.1 33.8 24.3 27.3 15.1 33.3 10 5.9 36.8

175 512 25 2270 844 199 20 1628 228 321 553 225 54 117 553 225 131 998 75 130 810 224 210 4 98 275 312 595 78 14 22 205 40 217 190 583 35 160 70 33 225 12 30 17 68

44 129 5 35 405 60 2 293 10 37 346 41 6 48 346 41 95 16 3 11 15 6 1 5 30 55 108 548 5 3 7 17 10 83 19 345 7 55 18 10 35 5 4 2 26

133 385 22 2237 441 141 20 1337 220 286 209 186 50 71 209 186 38 984 74 121 797 220 211 1 70 222 206 49 75 13 17 190 32 136 173 240 30 107 54 25 192 9 28 17 44

adult beef cows

adult dairy cows

heifers sows or gilts nursing or weaner pigs grower or finishing pigs ewes lambs

horses turkeys poultry, broilers

poultry, breeders poultry, layers

m

P(yi) )

∑w

ikβ(yi,

Θi,k)

(4)

k)1

where Θi,k is the set of parameters (Rbi,k and βbi,k) for the β-distribution obtained from a given prevalence study in a given animal census category, m is the number of β-distributions in the mixture, and wi,k is a weighting factor equal to 1/m. Changes in the reported values of animal pathogen prevalence over all types of animals expected to be present in the watershed were investigated to determine if microbiological detection method improvements over time resulted in fewer or greater numbers of positive samples. Reported animal prevalence values for Cryptosporidium or Campylobacter did not demonstrate any meaningful long-term temporal trends. Therefore, it was determined that weighting the b-distributions according to changes in methods of detection or more heavily toward more recent studies would not improve the simulation results. Pathogen prevalence may differ according to geographical regions as management practices and climate may have an influence on infection levels in animal populations. However, in most watersheds insufficient data are available regarding pathogen prevalence 3372

9


animal type

% ref positive

n sampled

rbi,k

βbi,k

poultry, broilers 46 27 12 233 3 305 8 930 47 35.4 929 330 601 48 100 60 61 1 21 32.1 1 600 515 1 087 49 42.5 89 110 37 873 51 240 27 3.1 19 700 000 606 001 19 094 001 50 94.7 360 342 20 51 100 254 255 1 poultry, layers 52 66.1 280 186 96 53 42.9 105 46 61 turkeys 54 66.8 20 14 8 sows and gilts 47 45.9 316 146 172 20 79.7 59 48 13 growers or 13 91.9 595 548 49 finishing pigs 55 100 24 25 1 56 98.1 160 158 4 nursing or 20 63.6 93 60 35 weaner pigs 36 79.3 294 234 62 lambs 57 91.7 360 331 31 58 87.3 840 734 108 36 18.3 93 18 77 sheep 19 55.3 750 416 336 calves 59 66.7 42 29 15 36 19.7 304 61 245 60 31.1 212 67 147 beef cattle 18 35.7 196 71 127 61 89.4 360 323 39 47 46.8 94 45 51 dairy cattle 59 59.6 94 57 39 62 37.7 2 085 787 1 300 63 22.3 273 62 213 60 9.2 120 12 110

in local animal populations, and is it difficult to ascertain which studies of animal populations are most similar to animals in another given geographical region. Only studies that randomly selected farms and animals for prevalence testing were included in the development of the model; therefore, data on pathogen prevalence from traceback studies (where the herd was identified as the source of an outbreak) were not included. Studies of herds or flocks with known histories of the presence of pathogens were also not included. Efforts were made to ensure that selected data from the literature had not already been included in another selected published study in order to be certain that all studies were independent of one another. Over time, the cumulative prevalence of pathogens may be close to 100% for some animals [e.g., pigs (25) and calves (26)]. However, to obtain an estimate of the prevalence as a snapshot in time, it is necessary to select random samples independently. A separate β-distribution was generated from each study and for each census category (i.e., type and age of animal). A mixture of the β-distributions was created to model the overall pathogen prevalence for each census category. Tables 1 and 2 provide the data from which the β-distribution parameters were calculated for Cryptosporidium and Campylobacter, respectively, in addition to the calculated parameters Rbi,k and βbi,k. In many cases, there is a wide degree of variability in the parameters of the β-distributions for a given animal category as anywhere from 0 to 100% of sampled animals may be potentially positive for the pathogens. Reported prevalence of fecal shedding of Cryptosporidium and Campylobacter among livestock varies widely partly because investigators use methods of detection with differing specificity and sensitivity, but a large portion of the variation results from sampling different livestock populations under varying management practices (1).

FIGURE 3. β-distribution mixture for Cryptosporidium prevalence in calves (1 yr and under). P(x) is the probability of detecting x percent of animals positive for Cryptosporidium. The composite distribution is the mixture resulting from individual distributions, according to eq 4. Studies with a large sample size will produce β-distributions with a small variance and a large probability mass on the measured percentage of positive animals. In the mixture of β-distributions, all individual β-distributions are considered equally likely, and random samples are uniformly drawn from them. If a given study has a very large number of animals sampled [e.g., Campylobacter in 19 700 000 broiler chickens reported by Perko-Ma¨kela¨ et al. (27)], the individual β-distribution generated from that study will have low variance. Since each study was given equal weighting in the overall mixture, the individual β-distribution of a study with a large number of animals tested will not overwhelm the mixture. Studies were not rejected on the basis of results that were different than the majority of other studies (i.e., outliers) as they were also viewed as potentially representative of the true β-distribution that is unknown. When several studies are in agreement with one another, the mixture of β-distributions has a larger probability mass in the shared region of agreement. Mixtures of β-distributions have been used to model the prevalence of Campylobacter in chicken flocks (28). Figures 3 and 4 demonstrate the mixtures of β-distributions from a particular census categoryscalves 1 yr and under. Reported prevalence of Campylobacter in calves is greater than that of Cryptosporidium. The presence of Cryptosporidium in calves has been more widely examined than Campylobacter, thus the mixture of β-distributions for Cryptosporidium is based on a greater number of studies and animals. The β-distribution mixture for Campylobacter has little probability mass on zero and a large probability mass greater than 50%, indicating that Campylobacter has been commonly detected in calves. Estimation of Animal Shedding Intensity. To estimate the production of organisms per unit of fresh weight manure, frequency distributions were derived from data in the literature. The frequency distributions were derived from test-positive animals only, meaning that zeros were not included in the distribution. For the present research, for studies that reported samples that were positive, but below

the sensitivity of a given enumeration method, numbers were randomly generated between 100 and the reported sensitivity limit of the method. The probability densities for shedding intensities (in log number/g) for infected animals were represented by Γ-distributions. Γ-distributions were selected as they can take on a variety of shapes and could be fit to published shedding intensity data. Medema et al. (24) modeled concentrations of Cryptosporidium oocysts in manure using Γ-distributions. The probability distribution representing animal shedding intensity in log number of pathogens per gram of fresh weight manure, P(zi), was estimated as

P(zi) ) Γ(Rgi, βgi)

(5)

where Rgi and βgi are parameters for the Γ-distributions that model fecal shedding intensity, estimated from published studies for a given census category. A separate Γ-distribution was used to model the pathogen shedding intensity for each of the census categories for animal type. The natural variability of Cryptosporidium and Campylobacter shedding intensity reported in infected animals is very large (ranging approximately from 100 to 107), thus parameters for the Γ-distribution were found only to fit in the log domain. The parameters for the Γ-distribution, Rgi and βgi, were estimated using the method of moments (e.g., ref 64) as

Rgi )

Χ h2 n-1 2 s n

(6)

n-1 2 s n βgi ) Χ h

(7)

where s2 is the sample variance, Χ h is the sample mean, and n is the number of sample points. VOL. 38, NO. 12, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3373

FIGURE 4. β-distribution mixture for Campylobacter prevalence in calves (1 yr and under). P(x) is the probability of detecting x percent of animals positive for Campylobacter. The composite distribution is the mixture resulting from individual distributions, according to eq 4.

TABLE 3. Γ-Distribution Parameters for Cryptosporidium Shedding Intensity animal type

ref

rgi

βgi

lambs sheep cattle (dairy and beef) calves pigs, nursing or weaners poultry, broilers

40 40 66, 11 66, 26 25 67, 68

37.77 12.47 14.01 3.708 12.09 29.31

0.06223 0.1083 0.1370 0.6247 0.1585 0.1560

TABLE 4. Γ-Distribution Parameters for Campylobacter Shedding Intensity animal type

ref

rgi

βgi

poultry, broilers poultry, turkeys lambs sheep, ewes calves beef cattle dairy cattle nursing or weaner pigs sows and gilts

50, 69 54 57 19 70, 19 70 19 71 72, 71

27.78 5.621 15.75 3.263 11.45 3.987 5.181 4.419 4.207

0.2558 0.8433 0.2434 0.9315 0.4140 0.6835 0.8509 0.6319 0.8859

Tables 3 and 4 provide the sources and parameter estimates for the Γ-distributions representing shedding intensity. The frequency distributions were generated for each census class over all estimates for each census category. Shedding intensity values for some census categories are not available as no studies have been published. In cases where data are not available, Γ-parameters from the most similar census category were used. For Cryptosporidium shedding in lambs, Bukhari and Smith (65) published results from an experimental infection study; however, their results are several orders of magnitude higher than for other animal categories and were therefore not used in this study. Instead, it was assumed that lambs would shed Cryptosporidium in numbers approximately an order of magnitude greater than 3374

9


FIGURE 5. Cumulative Γ-distributions modeling Cryptosporidium shedding intensity in domesticated farm animals. adult sheep, based on observed differences between lambs and adult sheep with respect to Campylobacter (57, 19). The Γ-distribution parameters for Cryptosporidium in lambs were derived from the adult sheep study. In general, fewer studies are available that provide data on animal pathogen shedding intensity than are available on pathogen prevalence. Differences between the estimated Rgi and βgi parameters for the various animal categories are the result of younger animals shedding larger numbers of pathogens than older animals as well as differences between animal types. For both Cryptosporidium and Campylobacter, higher numbers of pathogens per gram of fresh weight manure have been reported in poultry as compared to other animal types. Figures 5 and 6 demonstrate the cumulative probability density functions of shedding intensity by livestock for Cryptosporidium and Campylobacter. For both microorganisms, poultry have the greatest reported shedding intensity per gram of fresh weight manure. Younger animals have been shown to shed pathogenic organisms in greater numbers (e.g., ref 61), which is captured in the parameters of the Γ-distributions and demonstrated in Figures 5 and 6.

comparable to rates in the literature, such as the Midwest Plan Service’s (74) section of manure characteristics. The conditional probability of the total daily numbers of pathogens produced from all livestock sources given the daily numbers of pathogen produced by animals in given census categories was modeled as N

P(XTot|x1, x2, ..., xN) ) δ(XTot -

∑x )

(9)

i

i)1

The probability of total daily pathogen production rate from all livestock sources in a given region, XTot [number/day], was modeled as

P(XTot) ) FIGURE 6. Cumulative Γ-distributions modeling Campylobacter shedding intensity in domesticated farm animals.

∫ ∫ ...∫ ∑∑...∑∫ ∫ ...∫ ∫ ∫ ...∫ x1

x2

xN

y1

y2

yN

z1

z2

zN

z1

z2

zYmax

×

N Ymax

∏∏P(x |y , z

P(XTot|x1, x2, ..., xN)

TABLE 5. Manure Production as Excreted by Farm Animals animal census category

manure production (kg/day)

bulls, 1 yr and over dairy cows beef cows heifers for beef herd replacement heifers for dairy herd replacement heifers for slaughter or feeding steers, 1 yr and over calves, under 1 yr boars sows and gilts for breeding nursing and weaner pigs grower and finishing pigs rams ewes lambs broilers, roasters and cornish pullets under 19 weeks, intended for laying laying hens 19 weeks and over laying hens in hatchery supply flocks turkeys other poultry horses

23 50 23 21 20 20 20 5 11 15 2 5 4 4 1 0.04 0.04 0.11 0.11 0.18 0.11 28

Estimation of Daily Pathogen Production. The conditional probability of the number of pathogens produced from animals in an individual census category, given the number of animals positive for the pathogen and their shedding intensities, was estimated as yi

P(xi|yi, zi,j) ) δ(xi -

∑10

z

i,jMi)

i

i

i,j)P(yi)P(zi,j)

×

i)1 j)1

(8)

j)1

where xi is the total number of pathogens produced from animals in a given animal census category, yi is the number of animals positive for Cryptosporidium or Campylobacter in a given animal census category, zi,j is the log number of Cryptosporidium or Campylobacter per gram of fresh weight manure for an individual animal in a census category, and Mi is the manure production in grams per animal per day in a given animal census category. Manure production rates were estimated “as excreted”. Table 5 presents the manure production values used for this study. Values have been obtained from the Ontario Ministry of Agriculture and Food’s Nutrient Management Software (73) and modified from rates on a volumetric basis to a mass basis based on an assumed fresh manure density of approximately 1 kg/L (73). The manure production rates are

dx1, ..., dxN, dy1, ..., dyN, dz1,1, ..., dzN,1, ..., dzN,Ymax (10) where x1, x2, ..., xN are the total daily pathogen production rates from individual animal census categories (i.e., calves 1 yr and under), number/day; y1, y2, ..., yN are the number of animals positive for the pathogen in the given region; z1,1, z2,1, ..., zN,1, ..., zN,Ymax are the shedding intensities of individual animals; N is the number of animal census categories; and Ymax is the number of positive animals within an animal census category. Due to the size and complexity of the model relating the total daily pathogen production (XTot) to its parameters, the probability distribution on XTot was estimated using Monte Carlo methods. Figure 7 illustrates the general steps of the method used for the Monte Carlo simulations. The method builds upon the approach by Dorner et al. (75). Each iteration of the simulation provided a single estimate for XTot in the probability distribution. Given the number of animals in the region, the number of animals positive for Cryptosporidium or Campylobacter was obtained from random sampling of a mixture of generalized β-distributions. Then for each positive animal, the number of pathogens shed by an individual animal was taken from random sampling of a Γ-distribution. Using daily manure production rates, the daily pathogen production was calculated by summing the number of pathogens shed by all positive animals. As different animal types and age classes may show varying pathogen prevalence or shedding intensities, the simulations were performed according to the animal categories as defined in the 2001 Canadian Census of Agriculture (22). The total daily pathogen production rate was obtained by summing the daily pathogen production from all individual census categories. The solution is computationally intensive and time-consuming as more than 1 000 000 animals may reside in a given region, which translates into a large number of positive animals for which shedding intensities were simulated. The simulations were performed using Matlab, and random numbers for the distributions were obtained using Matlab’s functions from the statistics toolbox. The number of iterations necessary to integrate the probability distributions based on the stabilization of the mean and standard deviation of the log-transformed data was found to be approximately 3000. The simulations were performed for all consolidated census subdivisions, which have the finest resolution available of data for livestock.

Results and Discussion Modeled estimates of daily mean pathogen production are presented in Figures 8 and 9. The error bars representing the VOL. 38, NO. 12, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3375

FIGURE 7. Flowchart of the method to determine daily pathogen production in a given region.

FIGURE 8. Modeled daily mean production of Cryptosporidium in townships of the Grand River Watershed. Error bars represent standard deviation. standard deviation in Figures 8 and 9 capture the natural variability associated with the estimates. The transient nature of pathogen presence in livestock and the high variability of shedding intensities among positive farm animals are the primary sources of the variability captured by the model. Additional uncertainties not considered include the uncertainty relating to livestock numbers and the variability of manure production per individual animal. Differences of several orders of magnitude between estimate daily rates of pathogen production exist between regions. In general, townships with high daily rates of Cryptosporidium production also have high daily rates of Campylobacter production because the production rates for both pathogens are based upon the number of animals and the estimated manure production (see Figure 10) within a region. However, differences in the relative ratios of animal types can affect the relative distribution of the pathogens across different regions. This highlights the need for assessing regions, animal classes, and pathogen types individually and that there are important spatial variations of pathogen production within a watershed. 3376

9


Figures 11 and 12 show the locations of the regions classified by daily pathogen production rates per hectare. The Six Nations of the Grand River Territory is a region of the watershed with no estimates of pathogen production, as data on livestock are unavailable. The zones with the highest estimated category of daily Cryptosporidium and Campylobacter production are located in the clay till portion of the watershed (76), where most of the watershed’s runoff is generated and where a high proportion (>50%) of the agricultural land is tile drained. These regions with high runoff potential and large daily pathogen production are potentially the source of the greatest amount of pathogenic contamination in surface waters. The tributaries to the Grand River that flow through these areas are the Conestogo River and Canagagigue Creek, which have been shown to have increased levels of indicator organisms such as fecal coliforms and E. coli in their waters as compared to other sampled regions of the watershed (23). The lowest estimates of daily pathogen production occurred in the heavily urbanized regions in the central portion of the Grand River Watershed. The eastern regions

FIGURE 9. Modeled daily mean production of Campylobacter in townships of the Grand River Watershed. Error bars represent standard deviation.

FIGURE 11. Modeled daily mean Cryptosporidium production in the Grand River Watershed. FIGURE 10. Estimated daily manure production in the Grand River Watershed. of the watershed, which are less intensively farmed, also demonstrated lower estimates of pathogen production. Urban regions have lower estimates of pathogen production from livestock; however, urban streams may still carry heavy pathogenic loads as a result of urban stormwater as well as discharges from wastewater treatment plants. Ducks and geese frequently reside in urban regions and may be an important source of contamination in urban subwatersheds, as Campylobacter and Cryptosporidium species have been detected in migrating birds (e.g., refs 77 and 78). Campylobacter spp. have been detected in stream samples obtained from several regions of the Grand River Watershed using quantitative PCR in mean concentrations ranging from 69 to 324 cells/100 mL with 60-85% of all samples (n ) 192) positive for the organism (23). A reported detection frequency

of Cryptosporidium in the Grand River at the intake of a drinking water treatment plant was 33.7% of samples analyzed (n ) 100), and the maximum concentration reported was 1 oocyst/L (79). In the Grand River Watershed, reported water concentrations of Campylobacter are greater than reported concentrations of Cryptosporidium by several orders of magnitude. The results of the present study also suggest that Campylobacter production may be greater than Cryptosporidium production by several orders of magnitude. It is not possible to validate the probabilistic model through direct comparison with measured stream values because processes such as die-off and transport have not been considered, nor are actual pathogen loading data available. Estimates of greater production of Campylobacter as compared to Cryptosporidium are expected and reasonable based on reported prevalence and shedding intensity data, and are in accordance with available environmental observations of stream samples, which show a higher frequency of detection VOL. 38, NO. 12, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3377

TABLE 6. Modeled Daily Cryptosporidium and Campylobacter Production in Woolwich Township animal category (n animals) cattle (36 338) calves (7175) pigs (44 011) poultry (1 259 844) sheep (795) horses (1214)

FIGURE 12. Modeled daily mean Campylobacter production in the Grand River Watershed. and higher concentrations of Campylobacter as compared to Cryptosporidium. However, it is important to consider that variations in survival times and differences between microbiological methods used for Campylobacter and Cryptosporidium may impact the estimate of the numbers of microorganisms in a sample. Furthermore, it has not been common practice to include estimates of viability or recovery efficiency of the microbiological methods used for enumeration in the reporting of results, which adds uncertainty to modeling and monitoring results. Differences between estimated numbers of pathogenic microorganisms produced and concentrations measured in streams suggest that processes occurring on land of physical removal such as natural filtration, management practices such as manure storage, and inactivation from environmental stresses are important and account for the removal of microorganisms by several orders of magnitude. However, the results also demonstrate that the potential exists for large numbers of organisms to be present on any given day and that livestock can be an important source of pathogenic microorganisms such as Cryptosporidium and Campylobacter. Efforts are currently under way to model the physical/ chemical/biological processes for pathogenic fate and transport. Variations of pathogen prevalence in livestock has on occasion been associated with season (e.g., refs 57, 59, and 61); however, larger differences have been associated with animal age, with younger animals showing a higher prevalence and shedding of pathogens (30, 31). As such, management practices such as calving become important with respect to peaks in numbers of pathogens emitted to the environment. Calves have been shown to have the greatest prevalence and shedding intensity of Cryptosporidium at approximately 15 days of age (29, 66). Although calves produce less manure than adult cows, the greater shedding intensity of calves can translate to greater production of pathogens relative to the total number of animals. In Ontario, calving occurs year round; however, for beef calf production, the majority of calves are born in the spring. As survival of these organisms is greater at lower temperatures (80, 81) and frequently more precipitation occurs in the spring, the spring season is potentially a critical time for drinking water utilities. The prevalence of pathogens in herds and flocks is frequently clustered in space and in time, but clustering was 3378

9


Cryptosporidium Campylobacter mean log count ((SD) mean log count ((SD) 12.24 (1.05) 12.16 (1.16) 10.24 (0.364) 13.02 (0.122) 7.680 (0.245) 9.077 (0.182)

19.31 (1.39) 15.27 (0.738) 20.09 (1.24) 21.01 (1.20) 13.60 (1.33) 16.76 (1.65)

not considered in this study because individual farm-level data is not available. Furthermore, the model assumes independence between animal census categories. Correlations are likely to exist between census categories at the farmlevel as a result of clustering. Despite some studies examining vertical transmission [e.g., in cow-calf herds (66) or in chicken flocks (46)], data to quantify correlations between categories are generally not available. It is not possible to compare model results with environmental samples, as it is not possible to directly measure the daily production of pathogenic microorganisms. However, an understanding of the reliability of the model to estimate pathogen numbers is necessary. As a result of the high degree of variability in pathogen shedding intensity, the model is highly sensitive to the parameters of the Γ-distribution. This illustrates the need for reliable data on pathogen shedding by livestock. For some categories of animals, data on pathogen shedding intensity was not available (for example, boars). Variability in measured numbers of pathogenic organisms per fresh weight manure may result from differences in methods used for detection and enumeration but may also result from whether the animals were experimentally or naturally infected. Atwill et al. (1) estimated the environmental loading of C. parvum from adult beef cattle and found that insensitive methods of detection could result in an upward bias of arithmetic mean fecal shedding estimates if low fecal shedding numbers were not included in the calculation of the mean. As the model in the present study is also sensitive to shedding intensity, an upward bias is possible if the fecal shedding intensity from animals shedding low numbers was not included in the frequency distribution. Therefore, efforts were made to include low shedding intensity numbers in the frequency distributions by simulating shedding intensities from animals that were positive but below the sensitivity of an enumeration method. Overall, the estimate of daily environmental loading is dominated by animals shedding large numbers of pathogens rather than low numbers. Table 6 shows the daily pathogen production by animal category as estimated by the probabilistic model, for the region with the greatest level of Campylobacter and Cryptosporidium produced, Woolwich Township. From this comparison, it appears that although cattle are the greatest generators of manure in the watershed and may generate a larger number of pathogens per individual animal, other livestock animals such as poultry that do not generate as much manure (but shed large numbers of organisms relative to the manure that they produce) may also be important for the production of overall levels of pathogenic microorganisms. The relative proportion of pathogens produced by various animal types will remain the same over time as the model incorporates the effects of manure production and the number of animals in its estimates. Temporal variations may exist; however, these would be related to other factors such as farm-level management practices (e.g., calving periods) that were not considered in the model or changes in the relative numbers of animals in the watershed.

From a recreational standpoint, the results from the present study suggest that Campylobacter may be a greater cause for concern than Cryptosporidium as they are produced in much greater numbers, based on current methods of detection and enumeration. However, Cryptosporidium oocysts may have a greater ability to survive in the environment than Campylobacter (80, 81) and could potentially be transported greater distances from their sources. The lack of correlation of Campylobacter and Cryptosporidium to indicator organisms such as E. coli and fecal coliforms in the Grand River Watershed and other regions (23, 82) demonstrates the inadequacies of the use of indicator organisms for assessing water quality, and greater emphasis should be on understanding sources and the natural processes that will affect levels of the pathogenic microorganisms in the stream. The method presented in this study is a step toward a better understanding pathogen sources in the environment in order to advance the development of pathogen environmental fate and transport models.

Acknowledgments This study is funded by the Canadian Water Network and the Natural Sciences and Engineering Research Council of Canada (NSERC). The authors acknowledge the Grand River Conservation Authority, the City of Brantford, and the Regional Municipality of Waterloo for providing data. We thank Bill Anderson from the NSERC Chair in Water Treatment for technical assistance. We also thank Chris Pal from the School of Computer Science at the University of Waterloo and John Winn from Microsoft Research Cambridge for discussions on probabilistic modeling.

Literature Cited (1) Atwill, E. R.; Hoar, B.; das Graç as Cabral Pereira, T. K. W.; Rulofson, F.; Nader, G. Appl. Environ. Microbiol. 2003, 69, 46044610. (2) Hrudey, S. E.; Huck, P. M.; Payment, P.; Gillham, R. W.; Hrudey, E. J. J. Environ. Eng. Sci. 2002, 1, 397-407. (3) Berndtson, E.; Emanuelson, U.; Engvall, A.; Danielsson-Tham, M. L. Prev. Vet. Med. 1996, 26, 167-185. (4) Johnson, J. Y. M.; Thomas, J. E.; Graham, T. A.; Townshend, I.; Byrne, J.; Selinger, L. B.; Gannon, V. P. J. Can. J. Microbiol. 2003, 49, 326-335. (5) AWWA Research Division Microbial Contaminants Research Committee. J. Am. Water Works Assoc. 1999, 91 (9), 110-121. (6) AWWA Research Division Microbial Contaminants Research Committee. J. Am. Water Works Assoc. 1999, 91 (9), 101-109. (7) Walker, M. J.; Montemagno, C. D.; Jenkins, M. B. Water Resour. Res. 1998, 34, 3383-3392. (8) Fayer, R.; Speer, C. A.; Dubey, J. P. In Cryptosporidium and Cryptosporidiosis; Fayer, R., Ed.; CRC Press: Boca Raton, FL, 1997; pp 1-42. (9) Blaser, M. J. J. Infect. Dis. 1997, 176, S103-105. (10) Majowicz, S. E.; Michel, P.; Aramini, J. J.; McEwen, S. A.; Wilson, J. B. Can. J. Public Health 2001, 92, 62-66. (11) Ralston, B. J.; Allister, T. A.; Olson, M. E. Vet. Parasitol. 2003, 114, 113-122. (12) Causape´, A. C.; Quı´lez, J.; Sańchez-Acedo, C.; del Cacho, E.; Lo´pez-Bernad, F. Vet. Parasitol. 2002, 104, 287-298. (13) Harvey, R. B.; Young, C. R.; Ziprin, R. L.; Hume, M. E.; Genovese, K. J.; Anderson, R. C.; Drolesky, R. E.; Stanker, L. H.; Nisbet, D. J. J. Am. Vet. Assoc. 1999, 215, 1601-1604. (14) Papadopoulou, C.; Xylouri, E.; Zisides, N. Avian Dis. 1988, 32, 842-843. (15) Enemark, H. L.; Ahrens, P.; Lowery, C. J.; Thamsborg, S. M.; Enemark, J. M. D.; Bille-Hansen, V.; Lind, P. Vet. Parasitol. 2002, 107, 37-49. (16) Morgan, U.; Weber, R.; Xiao, L.; Sulaiman, I.; Andrew Thompson, R. C.; Ndiritu, W.; Lal, A.; Moore, A.; De plazes, P. J. Clin. Microbiol. 2000, 38, 1180-1183. (17) Pedraza-Dıáz, S.; Amar, C.; McLauchlin, J. FEMS Microbiol. Lett. 2000, 189, 189-194. (18) Beach, J. C.; Murano, E. A.; Acuff, G. R. J. Food Prot. 2002, 65, 1687-1693. (19) Jones, K.; Howard, S.; Wallace, J. S. J. Appl. Microbiol. 1999, 86, 531-536.

(20) Young, C. R.; Harvey, R.; Anderson, R.; Nisbet, D.; Stanker, L. H. Res. Vet. Sci. 2000, 68, 75-78. (21) Heuer, O. E.; Pedersen, K.; Andersen, J. S.; Madsen, M. Lett. Appl. Microbiol. 2001, 33, 269-274. (22) Statistics Canada. 2001 Census of Agriculture Farm Data; 2002; CD ROM. (23) Anderson, W. B.; Huck, P. M.; Gaulin, T.; Dorner, S. M.; Slawson, R. M. Proceedings of the 2003 American Water Works Association Water Quality Technology Conference. (24) Medema, G. J.; Ketelaars, H. A. M.; Hoogenboezem, W. Cryptosporidium and Giardia: occurrence in sewage, manure, and surface water; Association of River WaterworkssRIWA: Amsterdam. 2001; 171 pp. (25) Guselle, N. J.; Appelbee, A. J.; Olson, M. E. Vet. Parasitol. 2003, 113, 7-18. (26) Xiao, L.; Herd, R. P. Vet. Parasitol. 1994, 55, 257-262. (27) Perko-Ma¨kela¨, P.; Hakkinen, M.; Honkanen-Buzalski, T.; Ha¨nninen, M.-L. Epidemiol. Infect. 2002, 129, 187-192. (28) Hartnett, E.; Kelly, L.; Newell, D.; Wooldridge, M.; Gettinby, G. Epidemiol. Infect. 2001, 127, 195-206. (29) Maldonago-Carmago, S.; Atwill, E. R.; Saltijeral-Oaxaca, J. A.; Herrera-Alonso, L. C. Prev. Vet. Med. 1998, 36, 95-107. (30) Olson, M. E.; Thorlakson, C. L.; Deselliers, L.; Morck, D. W.; McAllister, T. A. Vet. Parasitol. 1997, 68, 375-381. (31) Wade, S. E.; Mohammed, H. O.; Schaaf, S. L. Vet. Parasitol. 2000, 93, 1-11. (32) Castro-Hermida, J. A.; Gonza´lez-Losada, Y. A.; Ares-Maza´s, E. Vet. Parasitol. 2002, 106, 1-10. (33) Quı´lez, J.; Sańchez-Acedo, C.; del Cancho, E.; Clavel, A.; Causape´, A. C. Vet. Parasitol. 1996, 66, 139-146. (34) Lefay, D.; Naciri, M.; Poirier, P.; Chermette, R. Vet. Parasitol. 2000, 89, 1-9. (35) Sakai, H.; Tsushima, Y.; Nagasawa, H.; Ducusin, R. J. T.; Tanabe, S. Uzuka, Y. Sarashina, T. J. Vet. Med. Sci. 2003, 65, 125-127. (36) Adesiyun, A. A.; Kaminjolo, J. S. Prev. Vet. Med. 1994, 19, 151165. (37) Scott, C. A.; Smith, H. V.; Mtambo, M. M. A.; Gibbs, H. A. Vet. Parasitol. 1995, 57, 277-288. (38) Lorenzo Lorenzo, M. J.; Ares Maza´s, E. A.; Villacorta Martıńez de Maturana, I. Vet. Parasitol. 1993, 47, 9-15. (39) Quı´lez, J.; Sańchez-Acedo, C.; Clavel, A.; del Cacho, E.; Lo´pezBernad, F. Vet. Parasitol. 1996, 67, 83-88. (40) Ortega-Mora, L. M.; Requejo-Fernańdez, J. A.; Pilar-Izquierdo, M.; Pereira-Bueno, J. Int. J. Parasitol. 1999, 29, 1261-1268. (41) Mun ˜ oz, M.; AÄ lvarez, M.; Lanza, I.; Ca´rmenes, P. Epidemiol. Infect. 1996, 117, 203-211. (42) Woodmansee, D. B.; Pavla´sek, I.; Pohlenz, J. F. L.; Moon, H. W. J. Parasitol. 1988, 74, 898-900. (43) Ley, D. H.; Levy, M. G.; Hunter, L.; Corbett, W.; Barnes, H. J. Avian Dis. 1988, 32, 108-113. (44) Kichou, F.; Saghir, F.; El Hamidi, M. Avian Pathol. 1996, 25, 103-111. (45) Itakura, C.; Goryo, M.; Umemura, T. Avian Pathol. 1984, 13, 487-499. (46) Pearson, A. D.; Greenwood, M. H.; Feltham, R. K. A.; Healing, T. D.; Donaldson, J.; Jones, D. M.; Colwell, R. R. Appl. Environ. Microbiol. 1996, 62, 4614-4620. (47) Nielsen, E. M.; Engberg, J.; Madsen, M. FEMS Immunol. Med. Microbiol. 1997, 19, 47-56. (48) Thomas, L. M.; Long, K. A.; Good, R. T.; Panaccio, M.; Widders, P. R. Appl. Environ. Microbiol. 1997, 63, 1874-1877. (49) Wedderkopp, A.; Gradel, K. O.; Jørgensen, J. C.; Madsen, M. Int. J. Food Microbiol. 2001, 68, 53-59. (50) Wallace, J. S.; Stanley, K. N.; Jones, K. In Campylobacters, Helicobacters, and Related Organisms; Newell, et al., Eds.; Plenum Press: New York, 1996; pp 323-328. (51) Gomes, C.; Santos, A.; Soqui, M.; Cabrita, J. In Campylobacters, Helicobacters, and Related Organisms; Newell, et al., Eds.; Plenum Press: New York, 1996; pp 301-306. (52) Petersen, L.; Nielsen, E. M.; On, S. L. W. Vet. Microbiol. 2001, 82, 141-154. (53) Kazwala, R. R.; Collins, J. D.; Hannan, J.; Crinion, R. A. P.; O’Mahony, H. Vet. Rec. 1990, 126, 305-306. (54) Wallace, J. S.; Stanley, K. N.; Jones, K. J. Appl. Microbiol. 1998, 85, 224-230. (55) Nesbakken, T.; Eckner, K.; Høidal, H. K.; Røtterud, O.-J. Int. J. Food Microbiol. 2003, 80, 231-240. (56) Weijtens, M. J. B. M.; van de Plas, J.; Urlings, B. A. P.; Bijker, P. G. H. In Campylobacters, Helicobacters, and Related Organisms; Newell, et al., Eds.; Plenum Press: New York, 1996; pp 363367. VOL. 38, NO. 12, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

3379

(57) Stanley, K. N.; Wallace, J. S.; Currie, J. E.; Diggle, P. J.; Jones, K. J. Appl. Microbiol. 1998, 84, 1111-1116. (58) Wallace, J. S.; Stanley, K. N.; Jones, K. In Campylobacters, Helicobacters, and Related Organisms; Newell, et al., Eds.; Plenum Press: New York, 1996; pp 359-362. (59) Atabay, H. I.; Corry, J. E. L. J. Appl. Microbiol. 1998, 84, 733740. (60) Nielsen, E. M. Lett. Appl. Microbiol. 2002, 35, 85-89. (61) Stanley, K. N.; Wallace, J. S.; Currie, J. E.; Diggle, P. J.; Jones, K. J. Appl. Microbiol. 1998, 85, 472-480. (62) Wesley, I. V.; Wells, S. J.; Harmon, K. M.; Green, A.; ShroederTucker, L.; Glover, M.; Siddique, I. Appl. Environ. Microbiol. 2000, 66, 1994-2000. (63) Meanger, J. D.; Marshall, R. B. N. Z. Vet. J. 1989, 37, 18-20. (64) Dougherty, E. R. Probability and Statistics for the Engineering, Computing, and Physical Sciences; Prentice Hall: Upper Saddle River, NJ, 1990; 800 pp. (65) Bukhari, Z.; Smith, H. V. Epidemiol. Infect. 1997, 119, 105-108. (66) Faubert, G. M.; Litvinsky, Y. J. Parasitol. 2000, 86, 495-500. (67) Tumova´, E.; Skrivan, M.; Marounek, M.; Pavlasek, I.; Ledvinka, Z. Avian Dis. 2002, 46, 203-207. (68) Taylor, M. A.; Catchpole, J.; Norton, C. C.; Green, J. A. Vet. Parasitol. 1994, 53, 7-14. (69) Stern, N. J.; Clavero, M. R. S.; Bailey, J. S.; Cox, N. A.; Robach, M. C. Poult. Sci. 1995, 74, 937-941. (70) Stanley, K. N.; Wallace, J. S.; Jones, K. In Campylobacters, Helicobacters, and Related Organisms; Newell, et al., Eds.; Plenum Press: New York, 1996; pp 163-167. (71) Weijtens, M. J. B. M.; Reinders, R. D.; Urlings, H. A. P.; Van der Plas, J. J. Appl. Microbiol. 1999, 86, 63-70. (72) Weijtens, M. J. B. M.; van der Plas, J.; Bijker, P. G. H.; Urlings, H. A. P.; Koster, D.; van Logtestijn, J. G.; Huis in’t Veld, J. H. J. J. Appl. Microbiol. 1997, 83, 693-698.

3380

9


(73) Ontario Ministry of Agriculture and Food. NMAN Software Instruction Manual; 2001. (74) Midwest Plan Service. Manure Characteristics MWPS-18; Manure Management Systems Series; 2000; 24 pp. (75) Dorner, S. M.; Huck, P. M.; Slawson, R. M. Proceedings of the American Water Resources Association’s Annual Congress: Watershed Management for Water Supply Systems, New York, 2003. (76) Grand River Conservation Authority. Grand River Regional Groundwater StudysTechnical Report; 2001; 312 pp. (77) Waldenstro¨m, J.; Broman, T.; Carlsson, I.; Hasselquist, D.; Acterberg, R. P.; Wagenaar, J. A.; Olsen, B. Appl. Environ. Microbiol. 2002, 68, 5911-5917. (78) Kuhn, R. C.; Rock, C. M.; Osima, K. H. Appl. Environ. Microbiol. 2002, 68 (1), 161-165. (79) Sobrinho, J. A. H.; Rosen, J. S.; Lechevallier, M. W.; Frey, M. M.; Clancy, J. L. Proceedings of the American Water Works Association Water Quality Technology Conference; 2001; Paper M8-3. (80) Chynoweth, R. W.; Hudson, J. A.; Thom, K. Lett. Appl. Microbiol. 1998, 27, 341-344. (81) Sattar, S. A.; Chauret, C.; Springthorpe, V. S.; Battigelli, D.; Abbaszadegan, M.; LeChevallier, M. Giardia Cyst and Cryptosporidium Oocyst Survival in Watersheds and Factors Affecting Inactivation; AWWA Research Foundation: 1999; 122 pp. (82) Chauret, C.; Armstrong, N.; Fisher, J.; Sharma, R.; Springthorpe, S.; Sattar, S. J. Am. Water Works Assoc. 1995, 87, 76-84.

Received for review October 29, 2003. Revised manuscript received March 22, 2004. Accepted April 6, 2004. ES035208+

Estimating Potential Environmental Loadings of ... - ACS Publications

Recommend Documents