Meta-Regression Analysis of Commensal and Pathogenic Escherichia

May 19, 2014 - ... Connected to Waterways in the Ruamahanga Catchment, New Zealand ... Environmental Science & Technology 2016 50 (4), 1897-1905...
0 downloads 0 Views 2MB Size
Article pubs.acs.org/est

Meta-Regression Analysis of Commensal and Pathogenic Escherichia coli Survival in Soil and Water Eelco Franz,*,† Jack Schijven,‡,∥ Ana Maria de Roda Husman,†,§ and Hetty Blaak† †

National Institute for Public Health and the Environment (RIVM), Centre Infectious Disease Control, Bilthoven, The Netherlands National Institute for Public Health and the Environment (RIVM), Expert Centre for Methodology and Information Services, Bilthoven, The Netherlands § Utrecht University, Faculty of Veterinary Medicine, Institute for Risk Assessment Sciences, Utrecht, The Netherlands ∥ Utrecht University, Faculty of Geosciences, Department of Earth Sciences, Utrecht, The Netherlands ‡

ABSTRACT: The extent to which pathogenic and commensal E. coli (respectively PEC and CEC) can survive, and which factors predominantly determine the rate of decline, are crucial issues from a public health point of view. The goal of this study was to provide a quantitative summary of the variability in E. coli survival in soil and water over a broad range of individual studies and to identify the most important sources of variability. To that end, a meta-regression analysis on available literature data was conducted. The considerable variation in reported decline rates indicated that the persistence of E. coli is not easily predictable. The meta-analysis demonstrated that for soil and water, the type of experiment (laboratory or field), the matrix subtype (type of water and soil), and temperature were the main factors included in the regression analysis. A higher average decline rate in soil of PEC compared with CEC was observed. The regression models explained at best 57% of the variation in decline rate in soil and 41% of the variation in decline rate in water. This indicates that additional factors, not included in the current meta-regression analysis, are of importance but rarely reported. More complete reporting of experimental conditions may allow future inference on the global effects of these variables on the decline rate of E. coli.



INTRODUCTION Escherichia coli occurs in diverse forms in nature, ranging from commensal strains (i.e., those participating in a symbiotic relationship with their host in which E. coli derives some benefit while the host is unaffected) to those pathogenic on human or animal hosts. Both commensal and pathogenic Escherichia coli can survive for prolonged periods in soil, manure, water, and plants.1−5 This can be considered a public health risk factor for the environmental transmission of pathogenic types as well as for opportunistic commensals carrying antimicrobial resistance factors. Human exposure to clinically relevant E. coli by environmental transmission pathways may include drinking water consumption, ingestion of recreational water, ingestion of manure and soil particles, and the consumption of fresh vegetables.3 Fresh vegetables (especially fresh-cut leafy greens) are increasingly being recognized as important vehicles for the transmission of human pathogens that were traditionally associated with foods of animal origin.6,7 Recently, concern has been raised with respect to the presence of Gram negative bacteria carrying determinants for antimicrobial like extendedspectrum β-lactamases (ESBL) on fresh vegetables.8,9 Although vegetables can become contaminated with pathogens by various means, contamination during the primary production phase from (manure-amended) soil and irrigation water are considered the most important.3 © 2014 American Chemical Society

The extent to which E. coli can survive, and which biotic and abiotic factors are of importance for its survival are crucial issues from a public health point of view. Quantitative microbial risk assessment (QMRA) is an established methodology in food and water safety to quantify the magnitude of exposure and risk of human infection associated with particular pathogens and transmission pathways.10 Quantifying and incorporation of variability is key to probabilistic risk assessment models.11 Exposure assessment in QMRA involves quantification of pathogen dynamics (i.e., growth and/or survival) along the transmission chain of concern. However, it is important to realize that the behavior of bacteria is highly variable due to the heterogeneity in environmental conditions encountered and variation between strains with respect to the inherent strain capacity to withstand those stress conditions.12,13 Both abiotic (temperature, pH, soil moisture, soil type) and biotic (composition and diversity of the microbial community) factors affect the survival capabilities of bacteria introduced into the soil habitat.14 However, most studies considered the effects of soil characteristics independently. Because the extent to Received: Revised: Accepted: Published: 6763

April 10, 2014 May 19, 2014 May 19, 2014 May 19, 2014 dx.doi.org/10.1021/es501677c | Environ. Sci. Technol. 2014, 48, 6763−6771

Environmental Science & Technology

Article

Figure 1. Overview of keywords literature search. N is the number of publications.

scientific database Scopus (Figure 1). The search was not restricted to a time period but was restricted to studies published in the English language. Abstract-based relevance screening was first applied to identify primary research in English investigating the decline-rate of E. coli in soil or water. Primary research was defined as original research during which authors generated and reported their own data. The primary focus was on studies addressing the survival of commensal and pathogenic E. coli (further referred to as CEC and PEC, respectively) In the context of this study, survival is related to the biological/physiological process of persistence in a given matrix under influence of stresses associated with that matrix. Consequently, variation in survival among strains is related to variation in resistance to the encountered stresses. Therefore, studies specifically investigating the physical removement and transportation of cells (i.e., runoff and percolation from surface into groundwater) were excluded. Only studies in soil and water with agricultural relevance were selected, meaning that studies in which E. coli was exposed to physical, chemical and biological treatments specifically designed to enhance the reduction of E. coli populations were excluded. This exclusion also held for studies using partly or entirely sterilized soil or water. Finally, decline-rate data from studies in beach-sand,

which these factors affect survival most likely depends on interactions between the various environmental factors, the overall set of abiotic and biotic soil characteristics should be taken into account. Many of these factors are a function of the experimental setup: field versus laboratory setting, inoculation of laboratory grown strains or natural contamination, the presence or absence of a fecal inoculum carrier, presence or absence of vegetation, etc. The primary goal of this study was to provide a quantitative summary of the variability in E. coli survival in soil and water over a broad range of individual studies. A specific goal was to compare commensal and pathogenic E. coli. Finally, we aimed at identifying the most important sources of variability. A systematic search of available literature was conducted, data were extracted, and multiple regression analyses were performed. The results were interpreted in the context of implications for QMRA and the understanding of experimental and ecological factors of importance for environmental survival of commensal and pathogenic E. coli.



MATERIALS AND METHODS Literature Search and Selection Criteria. A systematic literature search was conducted on August 14th 2012 using the 6764

dx.doi.org/10.1021/es501677c | Environ. Sci. Technol. 2014, 48, 6763−6771

Environmental Science & Technology

Article

Table 1. Considered Factors Affecting E. coli Die-Off Rate in Water and sSoil multivariate regression analysis factor description

factor

matrix coli type experimental location soil and water type inoculation faecal carrier vegetation temperature pH

mat coli exp

soil or water commensal or pathogen lab or field

model 1 X X

model 2 X X

model 3 X X

model 4 X X

submat

soil: clay, loam, sand, peat, volcanic, unspecified Water: fresh water, wastewater, groundwater, drinking water yes or no (inoculating the matrix with laboratory grown culture) yes or no (introduction of strains in the matrix with faecal material as carrier yes or no −20 °C − 37 °C 4−9

X

X

X

X

X X X

X X X X

X X

X X

inoc fcar veg temp ph

categorical or numerical values

−log10 Ct /C0 t

water

X X

which factors affect the value of the decline-rate coefficient and to what extent. Table 1 lists the data sets that were analyzed for which factors. Multivariate regression analysis was conducted by means of linear model fitting using the statistical package R (version 2.14.0). For regression analysis μ was log-transformed, which was justified by the visual inspection of Q-Q plots (no obvious curvature) and the plots of residuals (no obvious trends) (not shown). The effects of the factors and all two-way interactions were considered. By means of stepwise model selection by Aikaike’s Information Criterion (AIC) with k, the multiple of the number of degrees of freedom used for the penalty in AIC set to 3.84, best regression models were selected. A k value of 3.84 corresponds to the Chi-square with 95% confidence and one degree of freedom. The derivation of AIC involves the notion of loss of information that results from replacing the true parametric values for a model by their maximum likelihood estimates (MLE’s) from a sample. The best model selected (i.e., the model with the smallest expected loss of information when MLE’s replace true parametric values in the model) in R was implemented in Mathematica 9.0.1.0 (Wolfram Inc., Champaign Illinois) into a function for abstracting the linear equations for combinations of values of the categorical and numerical factors. Model predictions of E. coli decline-rate included 95% confidence intervals for the mean decline-rate as well as 95% prediction intervals.

sediments, forest soil, seawater, brackish water, or saline buffers were excluded. A second screening of the studies was performed by addressing the whole document. In this phase, studies were primarily assessed for the quality of the experiment and data. Studies were excluded that only reported qualitative (presence/absence) data or only (q)PCR data. Tailing of survival curves based on qualitative enrichment data was excluded and the last quantitative point above the detection limit was taken as the last measurement. Data Extraction. Although the decline in E. coli numbers over time may proceed at a rate that is significantly different from a first order decline of concentration in time (e.g., refs 12 and 15), reduction in E. coli concentration on a log10 (per g soil or ml water) scale was approached by linear approximation based on the following arguments. First, the majority of studies showed no obvious nonlinear decline over time (i,e, no shoulder-forming and/or tailing). Second, detailed information about the rate of E. coli decay is often lacking and would necessitate request of raw data from the authors. This was considered not feasible. For the purpose of collecting a sufficient number of data for the meta-analysis of E. coli declinerate, reduction in E. coli concentration on a log10 scale was assumed to proceed at a constant rate in time: μ=

soil

(1)



where Ct is the concentration of E. coli as a function of time, [numbers per volume of weight], C0 is the initial concentration, t is the time, [day], and μ is the decline-rate coefficient on a log10 scale, [day−1]. In the rest of the manuscript μ is equal to the decline rate in log number per day (t = 1). Values of μ from literature were used, when explicitly given, or derived from reported D90 values, that is, the time to one log10 reduction, where μ is the reciprocal. In other cases, μ was estimated from plotted data in graphs or reported data in tables by reading the logarithmic reduction concentration over the duration of the experiment and dividing it by the latter. In addition to the decline rate, bacterial and environmental factors were extracted (Table 1). Other bacterial characteristics (strain identity, O and H antigen type, strain origin, antibiotic resistance profile, virulence profile) and environmental factors (light/dark, chemical composition, water content/activity, redox potential) of interest were recorded but excluded from further analysis since these were reported by only a small fraction of studies (not shown). Data Analysis. Best Model Selection. Multivariate regression analyses were conducted in order to determine

RESULTS Data Set Characteristics. The primary Scopus search provided 119 hits for soil and 224 hits for water (Figure 1). After screening and applying the eligibility criteria, 169 waterrelated studies and 65 soil studies were excluded from further analysis. Most water studies were removed because of the specific application of treatments to enhance bacterial declinerate (n = 122) or because these were focused on monitoring the microbial load of water runoff from soil (n = 23). Most soil studies were excluded because of (i) the specific focus on runoff, percolation and leaching (n = 25), (ii) the matrix of interest was not of agricultural relevance (shoreline/beach sand, sediment, forest soil, alpine grassland) (n = 12), (iii) the application of specific treatments (sterilizing) (n = 4), (iv) specific focus on rhizosphere soil (n = 2). After whole manuscript screening another 24 water studies and 12 soil studies were excluded because of overall unsatisfactory data (for example no extractable quantitative data, questionable methods), resulting in a final yield of 54 studies for soil and 55 studies for water. 6765

dx.doi.org/10.1021/es501677c | Environ. Sci. Technol. 2014, 48, 6763−6771

Environmental Science & Technology

Article

Figure 2. Pie charts of (a) all 551 data points on E. coli die-off extracted from the selected literature, and (b) the 356 data for which temperature and pH were available. S = soil; W = water; C = commensal; P = pathogen; L = lab; F = field; inoculation = y/n.

Figure 3. Linear relations between the decline rate μ in water (top) and soil (bottom) and temperature in °C (left) and pH (right). Circles are observations; the solid line is the fitted linear regression (see Results section for equations); the dotted line is the 95% confidence interval; the shaded area is the 95% prediction interval.

the nomenclature commensal and pathogenic E. coli (CEC and PEC). There was a highly skewed distribution of serotypes used for STEC decline-rate studies in soil and water: 85.3% and 80.0% of the studies in soil and water were conducted with serotype O157. Consequently, the pathotype and serotype of pathogenic E. coli were not included as factors in further statistical inference. Isolated Effects of Matrix, E. coli Type, Temperature, and pH. Considerable variation in reported decline rates in soil was observed, ranging from 0.002 log CFU day−1 to 0.750 log CFU day−1. In water, decline rates ranged from 0.007 log CFU day−1 to 7.27 log CFU day−1 for. Both CEC and PEC showed a significant slower average decline rate in soil than in water (P < 0.0001). In soil, the average decline rate of CEC (average 0.052 log CFU day−1) was significantly lower than for PEC (0.071 log CFU day−1) (P = P < 0.0001). This was not observed in water (average CEC 0.189 log CFU day−1; average PEC 0.151 log CFU day−1) (P = 0.19). Decline rates larger than 1 log cfu day−1 were only observed in water. Log-transformed decline rates were normally distributed with the following parameters

Figure 2 summarizes the obtained data in terms of the number of individual decline rate values from the selected studies. In total the data set contained 551 data points, with the number on E. coli decline-rate in soil being twice the number E. coli decline-rate in water. About a quarter (26%) of the data concerned decline-rate in soil of CEC, about 40% decline-rate in soil of PEC. Studies on decline-rate commensal and pathogenic E. coli in water each provided about one-sixth of the data. The data set contained 356 data points for which temperature and pH were reported (Figure 2b). Surprisingly, temperature and pH were relatively less often reported in the studies with CEC in water as compared to CEC in soil and PEC in soil or water. But otherwise, the sizes of the various groups of data are relatively similar. For the meta-analysis, μ was log-transformed. With respect to PEC, the selected studies only included Shigatoxin-producing E. coli (STEC). Attenuated strains (i.e., removed or truncated stx-genes) were also considered STEC although these are more accurately described as atypical enteropathogenic E. coli (a-EPEC). Therefore, we maintained 6766

dx.doi.org/10.1021/es501677c | Environ. Sci. Technol. 2014, 48, 6763−6771

Environmental Science & Technology

Article

Figure 4. Linear relations between the decline rate μ soil for commensal (left) and pathogenic E. coli (right) and temperature in °C (left) and pH (right). Circles are observations; the solid line is the fitted linear regression (see Results section for equations); the dotted line is the 95% confidence interval; the shaded area is the 95% prediction interval.

Table 2. Decline Rates (μ in Log10 CFU day−1) in Soil Excluding Factors Temp and pH: Observed (Data) and Predicted by the Multiple Regression Model (Model)a factors

a

data

model

exp

submat

inoc

fcar

veg

N

mean

min

max

mean

95%-CI

95%-PI

lab lab lab lab lab lab lab lab lab lab lab lab lab Lab lab field field field field field field field field field field field field field

clay clay clay loam loam loam loam loam peat peat sand sand sand sand volcanic clay clay loam loam loam loam loam loam sand sand sand sand volcanic

yes yes yes yes yes yes yes no yes yes yes yes yes no yes yes no yes yes yes yes no no yes yes yes no no

yes no no yes yes no no yes no no yes yes no yes no yes yes yes yes no no yes yes yes no no yes yes

no yes no yes no yes no no yes no yes no no no no no no yes no yes no yes no no yes no yes no

1 14 8 17 63 14 41 11 4 4 2 38 19 1 2 2 4 21 35 1 1 12 6 6 4 4 7 2

0.110 0.110 0.060 0.100 0.063 0.073 0.050 0.062 0.080 0.077 0.036 0.048 0.079 0.034 0.120 0.065 0.046 0.059 0.098 0.030 0.035 0.019 0.035 0.160 0.170 0.066 0.034 0.047

0.110 0.093 0.042 0.020 0.014 0.060 0.008 0.007 0.051 0.051 0.020 0.019 0.031 0.034 0.110 0.046 0.003 0.020 0.038 0.030 0.035 0.002 0.010 0.095 0.170 0.11 0.043 0.044

0.110 0.140 0.180 0.190 0.210 0.090 0.250 0.150 0.110 0.110 0.063 0.110 0.210 0.034 0.140 0.092 0.360 0.380 0.750 0.030 0.035 0.069 0.063 0.290 0.170 0.11 0.043 0.049

0.110 0.100 0.070 0.095 0.064 0.074 0.050 0.061 0.096 0.065 0.069 0.046 0.079 0.044 0.120 0.100 0.037 0.062 0.092 0.025 0.037 0.022 0.032 0.150 0.087 0.130 0.035 0.047

0.029−0.400 0.078−0.140 0.051−0.096 0.076−0.120 0.055−0.075 0.058−0.095 0.042−0.060 0.041−0.088 0.060−0.150 0.040−0.100 0.052−0.093 0.038−0.057 0.059−0.100 0.028−0.068 0.049−0.310 0.059−0.180 0.021−0.063 0.049−0.079 0.076−0.110 0.014−0.045 0.020−0.068 0.016−0.029 0.023−0.045 0.100−0.220 0.056−0.140 0.083−0.200 0.024−0.051 0.019−0.120

0.017−0.680 0.027−0.400 0.018−0.270 0.025−0.360 0.017−0.240 0.02−0.280 0.013−0.190 0.016−0.240 0.024−0.390 0.016−0.260 0.018−0.260 0.012−0.170 0.021−0.300 0.011−0.170 0.025−0.610 0.025−0.430 0.009−0.150 0.0016−0.230 0.025−0.340 0.006−0.100 0.009−0.160 0.006−0.082 0.008−0.120 0.038−0.580 0.022−0.350 0.033−0.52 0.009−0.140 0.009−0.230

CI = confidence interval; PI = prediction interval. Model statistics: df = 343, P =