Giardia lamblia infection risk modeling in Mexico city's flood water

Floods are common events in urban areas and represent a concern due to their impacts on human health (Holcer et al. 2015). Biophysical and sociopolitical factors, including extreme precipitation events, urban growth, prevailing decision-making processes, lack of infrastructure, or insufficiency of sewerage and drainage capacity, contribute to urban flood frequency and magnitude (Eakin et al. 2017). Urban flood water may be contaminated with fecal material due to its combination with wastewater from sewer systems (Holcer et al. 2015). Inappropriate handling of human and animal waste in rural and urban areas causes the occurrence of feces in dust; thus, domestic animal and bird excreta on ground surfaces have the potential for mixing with rainwater (Rosas et al. 2019). Thereby, flood water can acquire pathogenic microorganisms such as bacteria, viruses, helminths, and protozoa (Sterk et al. 2008; ten Veldhuis et al. 2010). Although there is no direct evidence that pathogens can be transmitted from flood water to humans, several studies support the hypothesis that floods can be a microorganism transmission pathway (Cortés-Ortiz et al. 2002; ten Veldhuis et al. 2010). This suggests that urban flood water represents a relevant source of pathogen spread in human population and a risk that should be assessed (Cortés-Ortiz et al. 2002; Sterk et al. 2008).

Since the XVII century, city water management and sanitation have focused on decision-making cycles promoting hard infrastructure solutions (Tellman et al. 2018); however, these have been overwhelmed by urban sprawl. Mexico City was built in a closed basin shaped by a lacustrine system (Ezcurra et al. 2006) that, in addition to the climatic conditions, has led to recurrent flooding events (Chaussard et al. 2014; Matos-Moctezuma 2018). Studies on Mexico City have explored the relationships between the occurrence of gastrointestinal diseases, climatic, hydrological, infrastructural, and socioeconomic factors (Baeza et al. 2018). Cortés-Ortiz et al. (2002) found a relationship between flooding and gastrointestinal diseases caused by enterohemorrhagic and enteropathogenic Escherichia coli, two of the most important pathotypes (Lanata & Mendoza 2002). Baeza et al. (2018) showed an association between disease incidence and urban floods in Mexico City's lowlands and household infrastructure deficiency in the highlands. Despite these studies, there is, to our knowledge, no information on the risk of infection from exposure to pathogens related to flood water in this densely populated city.

In this study, we carried out a Quantitative Microbial Risk Assessment (QMRA) to evaluate the risk of infection from exposure to Giardia lamblia in urban floodwater. QMRA is a mathematical modeling framework used to estimate the risk of infection when a population is exposed to microorganisms from the environment (Haas et al. 2014). Regarding these health issues, we attempt to answer the following research questions: (i) Is there evidence of fecal contamination in Mexico City's floodwater, (ii) What is the risk of infection from exposure to G. lamblia in flood water?, and (iii) Is there a relationship between the risk of G. lamblia infection with urban flood frequency and volume? This study represents one of the few empirically based quantitative risk assessments performed in Mexico City. The results provide a novel source of information with respect to a new potential pathogenic source. This original assessment framework could be used to support adaptive urban flood and health management plans related to the spread of potential pathogens in urban contexts, especially in developing countries.

Mexico City lies within the Mexico City Metropolitan Area (MCMA), one of the largest megacities worldwide with nearly 22 million inhabitants (INEGI 2020). Mexico City is located at 19° 35′ 34″ N and 99° 21′ 54″ W at 2,240 meters above sea level (m.a.s.l.), and constitutes an endorheic system surrounded by mountains (INEGI 2017). The rainy season ranges from the months of May to October with a mean annual precipitation from around 600 mm in the north up to 1,200 mm in the south (Romero-Lankao 2010). The water infrastructure in Mexico City consists of a surface and groundwater mixed-distribution system, as well as an intricate wastewater collection system. The water distribution system has 11,000 kilometers of distribution lines and 243 storage tanks with a capacity of nearly 1.5 million cubic meters. The wastewater collection system combines wastewater and stormwater collection in a single network. Eventually, all of the sewers conduct the wastewater through four artificial exits, located at the northern end of the basin (National Research Council 1995).

We collected 25 flood water samples in 23 different census blocks (Basic Geostatistical Areas) from August 2015 to October 2016. Water samples were taken from puddles located in the street. Sampling was conducted in areas presenting the highest number of flood reports (2007–2014 period), according to Sistema de Aguas de la Ciudad de México (SACMEX, local water authority). Sampling was based on daily precipitation reports obtained through the local news. The time between the precipitation report and the sampling was less than 24 h; therefore, all samples were taken after the occurrence of the precipitation events. As reference, during June 2015, we collected influent samples from two wastewater treatment plants (WWTP) located in Mexico City: WWTP Cerro de la Estrella and San Luis Tlaxialtemalco, with one sample from each WWTP. All samples were collected in triplicate in sterile 250-mL polypropylene bottles and transported to the laboratory for analysis, the temperature was maintained at 4 °C until their processing. Descriptive data of accumulated annual precipitation between 2000 and 2017 in the sampling sites are presented in Table A1.

Fecal indicator bacteria were analyzed within the first 24 h after sample collection. Fecal coliforms, fecal enterococci, and E. coli densities were analyzed by standard membrane filtration methods. Counts are reported as colony forming units per 100 mL (CFU/100 mL; APHA 2017). To complement the fecal contamination analysis, we evaluated the presence of E. coli pathotypes in the flood water samples. To determine E. coli pathogenicity, five colonies per water sample were isolated and a multiplex PCR was performed to confirm the presence of four pathotypes: enterotoxigenic; enteropathogenic; shiga-toxin-producing, and enteroaggregative E. coli (ETEC, EPEC, STEC, and EAEC, respectively). A 25-mL reaction mixture contained 10 mL of DNA template, 2.5 mL buffer 10X, 1.5 mL of MgCl₂ 25 Μm, 0.625 mL primer forward of each pathotype, 0.625 mL primer reverse of each pathotype, all at 1-Μm concentrations, 0.5 mL DNTP 10 Μm, 0.25 mL Taq polymerase of 2,000 U, and 5.25 mL water. PCR amplification conditions were initial activation step 95 °C 3 min, 30 cycles denaturation 95 °C 30 s, annealing 53 °C 30 s, extension 72 °C for 30 s, and final elongation 72 °C for 5 min. Target genes and primer sequences are in Table 1.

Fifty mL of each flood water sample were centrifuged at 3,500 rpm for 15 min, supernatant was discarded, and the pellet was resuspended with 1 mL PBS. Samples were stored at −20 °C until their analysis. This procedure was performed in triplicate for each sampling site. The quantification of G. lamblia cysts was determined by calculating the mean across triplicate measurements per sampling site. Cyst quantification was performed by the indirect immunofluorescence method in liquid phase with monoclonal antibodies (mouse monoclonal IgG3 Anti-Giardia lamblia, BDI27; Santa Cruz Biotechnology, CA, USA) (Rangel-Martínez et al. 2015). Additionally, a positive control was used to evaluate both the integrity of the reagents and the G. lamblia cyst walls. G. lamblia cysts measuring 8–12 μm were quantified in an Axiostar Plus Fluorescence Microscope (Carl Zeiss, Göttingen, Germany), with counts reported as cysts per liter (cysts/L). The G. lamblia concentration was calculated as follows: number of cysts*1,000/50 mL. The quantification analysis ensures a 60% recovery rate of G. lamblia cysts (Tapia-Palacios 2012).

We applied a QMRA to quantify infection risk probabilities from exposure to G. lamblia in Mexico City urban flood water. The QMRA method entails four steps:

• (1)

  Hazard identification: G. lamblia is one of the most prevalent pathogens at the global level, mainly in countries with emerging economies, in which prevalence has been assessed in about 55.3% (Smith & Paget 2007).

• (2)
  Exposure assessment: Exposure to G. lamblia takes place during a flood event. Two different infection mechanisms were modeled: the case of pedestrians splashed by passing traffic (hereafter ‘adults’), and the case of children playing in the water (hereafter ‘children’). The degree of G. lamblia exposure described by the ingested dose was calculated according to Equation (1) (Haas et al. 2014):

  

  (1)

  where: μ = dose [number of G. lamblia], c = pathogen concentration in flood water samples [cyst/mL for G. lamblia], and v = intake volume [mL].

• Pathogen concentration was based on the analysis described in the previous section. Ingested volumes were modeled using a log normal distribution with mean = 10 mL and σ = 10 (adults), and mean = 30 mL and σ = 30 (children) according to previous reports (Sterk et al. 2008).

• (3)
  Dose-response modeling: This step involved estimating the probabilities of risk infection risk for adults and children utilizing an exponential dose-response model for Giardia spp. (Equation (2); Rendtorff 1954):

  

  (2)

  where: Pinf = single exposure risk of infection by a pathogen, = pathogen dose, obtained from sampling and intake volume, in the G. lamblia case, the organism-specific constant r = 0.019.

• To incorporate the uncertainty associated with some of the parameters, we performed 20,000 Monte Carlo simulations for each sampling site (n = 25). This method is an iterative technique employed to include the stochastic effect during the modeling process and is applied when the parameters that describe the distribution of one or more model variables are not known or assumed (Poulter 1998). For each Monte Carlo run, we solved Equations (1) and (2) by using the following: (1) the pathogen concentration calculated in each sampling site; (2) an intake volume value stochastically sampled from the theoretical log-normal distributions using mean and σ reported previously for adults and children (Sterk et al. 2008), and (3) the specific constant (r) associated with Equation (2). We then calculated the geometric mean, standard deviation, and 95% confidence intervals in terms of each Monte Carlo simulation set. All simulations were performed with R software version 3.5.2 (R Development Core Team 2019).

• (4)

  Risk characterization: The risk infection values were weighted based on U.S. Environmental Protection Agency (USEPA 2018) recommendations.

A historical flooding database (2015–2017) was provided by SACMEX. The flooding database included date and location of the flood, hereafter ‘neighborhood’ of the event, the in situ flood attributes, such as water width (m), length (m), and depth (m), plus the observed cause of the flood; records with incomplete information were removed from the dataset. Given the lack of a specific metric that allowed us to distinguish between puddles and floods, we thus considered them indistinctly. We calculated flood volume (m³) by multiplying width, length, and depth of each recorded flood event. An aggregated frequency and volume value at the neighborhood level was calculated by summing up all flood frequencies occurring during the recorded years and by averaging the volume values in a given neighborhood. In the dataset, frequency and volume records were weakly associated (r = 0.12; p <0.001), meaning that large floods are not typically the most frequent ones. Because flood water sampling was performed at the census block level, we needed to guarantee spatial resolution consistency with the flooding database; therefore, we downscaled aggregated frequency and volume values from neighborhood to census block level with Baeza et al. (2018) methodology.

We assigned different weights to each influencing factor to reflect the heterogeneity associated with the ‘importance’ evaluation of each factor in determining the risk of infection from exposure to G. lamblia (Malczewski 2011). Severity index 90–10 with weights wf = 0.9 and wv = 0.10 implies that frequency factor is nine times more important than volume. This indicates that both the risk of infection from exposure to G. lamblia and the presence of E. coli pathotypes are indifferent to flood volume changes, but not to changes in flood frequency. Severity index 50–50 with weights wf = 0.5 and wv = 0.5 possesses equal importance in both factors, while severity index 10–90 with weights wf = 0.10 and wv = 0.9 implies that flood volume is nine times more important than frequency.

Linear regressions were fit to the data and ANOVA analyses were performed to test for an effect of each severity index on the infection risk of G. lamblia in adults and children. Associations among variables were determined using Pearson product moment correlation coefficients (R package ‘stats’). The presence of E. coli pathotypes was treated as a binary response variable (1 or 0); thus, we performed simple logistic regressions (R package ‘glm’) and associated Chi-square tests to evaluate the effect of each severity index on the response variable in our flood water samples. Additionally, we computed the point-biserial correlations between presence of E. coli pathotypes and flood severity indices (R package ‘ltm’).

We mapped the spatial distribution of risk of G. lamblia infection and flood water severity indices to identify city hot spots. Additionally, we used the Urban Marginalization Index (UMI) from Consejo Nacional de Población (CONAPO, Mexican government instance), to include the socioeconomic dimension of flood severity impacts and associated health conditions. The UMI was built employing variables such as access to education and health care, availability of first-order goods, and enjoyment of adequate housing rights.

We utilized QGIS version 2.18.18 to categorize infection risk probabilities based on a scale including low, moderate, and high levels. Due to the lack of a limit for G. lamblia cysts in drinking water that could be used to compare the risk, we set the lowest threshold according to permissible risk values of 0.0001 in a yearly exposure of pathogens, based on U.S. Environmental Protection Agency (USEPA 2018) recommendations. It is noteworthy that the calculated risk values are assumed as one exposure event according to the concentration of pathogens quantified in the laboratory; thus, the annual risk must be greater than or equal to the risk reported here. Moderate- and high-risk categories were built utilizing the natural breaks method (Chou 2013). The moderate category comprised infection risk probabilities ≥0.0001 and <0.2, whereas the high category comprised probabilities ≥0.2.

We found fecal coliform and fecal enterococcus in 96 and 92% of floodwater, respectively. E. coli was found in 76% of the samples, detecting at least one pathogenic E. coli (ETEC, EPEC, or EAEC) in 20% of samples; STEC was not detected. G. lamblia was found in 56% of the samples. Bacterial and G. lamblia count magnitudes are presented in Table 2.

QMRA modeling results are shown in the Appendix (Table A2). The analysis showed a greater average risk of infection from exposure to G. lamblia in children than in adults (Figure 1). However, there was not a significant difference between both groups (t = 1.36; p= 0.35). ANOVA results revealed significant relationships between the risk of G. lamblia infection in adults and severity indices 90–10 and 50–50 (Table 3, F = 11.4; p= 0.002, and F = 9.07; p= 0.006; respectively). We also found significant relationships between the risk of G. lamblia infection in children and severity indices 90–10 and 50–50 (Table 3, F = 9.2; p= 0.005 and F = 7.7; p= 0.01; respectively).

Moderate positive significant correlations between the risk of G. lamblia infection risk in both adults and children and flood severity indices 90–10 and 50–50 were revealed, indicating that greater infection risk probabilities are, in part, associated with greater flood severity values (Figure 2(a), 2(b), 2(c) and 2(d); Table 3). The presence of E. coli pathotypes in flood water samples was not found to vary among severity indices. Low values of the biserial correlation are supportive of this positive but weak association (Table A3).

Twenty-eight percent of urban flood water samples were low risk from exposure to G. lamblia for both adults and children, 52% were moderate risk, and 4% were high risk (Figure 3(a)). Approximately, 73–76% of the total Mexico City area was classified as low flood severity for severity indices 90–10 and 50–50, respectively, 20–22% as moderate flood severity, and 3–4% as high flood severity (Table A4). Areas of moderate-to-high risk of G. lamblia infection in both adults and children were located in the city's periphery, particularly in southern and southeastern areas, where moderate-to-high-severity flood events occur (Figure 3(b)).

These areas of greater risk of infection lie within three administrative boroughs located in the city's periphery: Iztapalapa, Xochimilco, and Tláhuac. These three boroughs were also associated with moderate-to-high marginalization levels (Figure 3(b)), with Iztapalapa the most densely populated borough in Mexico City. Nearly 20% of the flood water samples exhibited the presence of at least one E. coli pathotype, and 80% of these samples were found in the city's periphery.

Our study reveals that urban flood water and WWTP samples have similar maximal fecal indicator density values. This result coincides with previous studies in Mexico, in which bacteria in wastewater has been reported in orders of 6–9 logs (Jiménez 2005). Fecal coliforms, fecal enterococci, and E. coli densities exceeded the permissible limits suggested by Mexican Federal Regulation NOM-003-SEMARNAT-1997 (DOF 1998) and USEPA recommendations for recreational water (USEPA 2018). We acknowledge that recreational water limits are not the ideal reference; nonetheless, these limits offer a way to compare the magnitude of the measured densities.

Our study reveals that urban flood water and WWTP present high pathogen densities. MCMA wastewater flow is estimated to reach 43 m³/s during the dry season, with approximately 340-m³/s peaks during the rainy season. Of this amount, only 6.5 m³/s are treated and reused through 91 wastewater treatment plants of different capacities, 69 of these located in Mexico City, the majority of these operating at a >50% capacity (Mazari-Hiriart et al. 2001; Burns 2009). This means that only about 15% of wastewater is treated, revealing a very inefficient potential reuse of safe water for other purposes, including irrigation. This entertains two important socioenvironmental implications: Efficiency of treatment processes need to be improved in order to reduce the health risk of contamination during flood events, and lack of treatment infrastructure does not allow the reuse of safe water in a city undergoing water scarcity.

QMRA modeling results suggest that there was a greater risk of infection from exposure to G. lamblia in children than in adults. This pattern can be explained partially by the assumed differential accidentally consumed volumes established in the model, which were higher for children. Exposure in children can potentially increase due to their activities at play; however, adults may be more exposed due to routines related with their work activities. Guerra et al. (2018) estimated that nearly 70% of Mexico City's working-age population commute to work by public transportation (e.g., buses, taxis, employee shuttles), on foot, or by bicycle. According to 2015 Mexican Intercensal data, residents of the city's outskirts spend longer average travel times commuting to the city's central areas; therefore, they could be at risk for higher exposure from pathogens during flood events (Mazari-Hiriart et al. 2018).

Three flood severity indices were built to include the uncertainty associated with the relative effect of flood water frequency and volume on risk infection from exposure to G. lamblia. Positive significant associations between severity indices 90–10 and 50–50 and risk of G. lamblia infection occurred in both adults and children. This suggests that flood water frequency is more strongly related with greater risks of infection from exposure to G. lamblia. The latter statement underscores that flood frequency, rather than flood volume, is a more likely mechanism leading to a greater risk of G. lamblia infection in Mexico City. As a result, recurrent floods could be sufficient for maintaining the population exposed to the pathogen during the majority of precipitation events occurring during the rainy season (Ochoa-Rodriguez et al. 2015). The adverse health effects of the risk of G. lamblia infection could be intensified in 22% of Mexico City's total area where moderate-to-high-severity floods are experienced (Table A4; Figure 3(a)). Our results are consistent with other studies conducted in developed urban contexts worldwide (Okaka & Odhiambo 2018). For instance, ten Veldhuis et al. (2010) reported that frequent floods are conducive to greater risks of infection from exposure to G. lamblia through direct contact with flood water. However, QMRA studies are scarce and are often performed in temperate, but not tropical, zones; therefore, it is risky to compare our results with those.

Distribution patterns of the risk of G. lamblia infection within Mexico City suggest that boroughs located in the city's periphery, such as Iztapalapa, Xochimilco, and Tláhuac (Figure 3(a) and 3(b)) are at a greater risk for infection than the boroughs in the city's center. These peripheral areas are usually characterized by high economic informality, fast growth, and water and sanitation infrastructure deficits (Baeza et al. 2018). As urban sprawl continues to develop, larger areas of impervious surfaces are created, thus resulting in greater runoff and flood risks (Ochoa-Rodriguez et al. 2015). In addition, extreme precipitation during the rainy season is already pushing the capacity of the sewage system to its limit, diminishing the response for adequate wastewater management, thus increasing urban vulnerability to flood related health risks (Ochoa-Rodriguez et al. 2015; Baeza et al. 2018).

Health risks associated with urban floods are a consequence of the interaction among biophysical (e.g., geology, climate, hydrology) factors, combined with the sociopolitical infrastructure (Baeza et al. 2018). Deficiencies in WWTP operation, marginalization levels, and existing inequities to access to health services, as well as to decision-making processes (Eakin et al. 2017) highlight the need for better understanding of the underlying mechanisms shaping health vulnerability in a city as complex as Mexico City. The results of our study build on previous conceptual views of the situation in this megacity. For example, Baeza et al. (2018) used gastrointestinal health reports to test for spatial correlations among biophysical, socioeconomic, and infrastructure variables. These authors’ results indicate the existence of dynamic interactions between water management at the regional scale and control of waterborne diseases in public health planning.

Mexico City currently faces myriad challenges that could potentially hinder the capacity of both water management and public health systems to suppress the outbreak of gastrointestinal diseases following flooding events. As a result, multiple regional and local actions should be undertaken to alleviate these risks and promote conditions to transit into a healthy city in the long term, as promoted worldwide (Baeza et al. 2018). For instance, recommendations to shape cities for health include integration of sustainable urban planning and decision-making processes (Rydin et al. 2012), building capacity for sustainability to reduce health inequality through the development of partnerships and skills (Gugglberger et al. 2016), facilitating effective actions for urban health through coordination among different social sectors–academia, citizens, and stakeholders–and, finally, implementing technological, sampling, and analytical tools designed for long-term public health data collection, analysis, and integration to effectively inform public policies (Lin et al. 2018).

There are some limitations regarding the implementation of the sampling design and other methodological issues which could improve our infection risk estimates. For instance, the reduced number of samples (one sample per site) represents a potential source of uncertainty since microorganisms behave differentially according to the distribution of key biophysical factors. Despite our sampling efforts, the lack of material and human resources, i.e. infrastructure and transport, number of technicians, and the restricted access to flood areas, together with the spatial and temporal variability of rainfall in Mexico City, imposed great challenges in terms of sampling simultaneously in all sites or at the exact time when the flood occurred. We acknowledge that the 24-hour lag time between a rainfall event and the deployment of our sampling campaign may result in an underestimation of the total load of microorganisms found in the flood samples. Similarly, greater percentages of G. lamblia recovery rates could improve cyst quantification and consequently infection risk estimates. However, it is worth noting that a 60% recovery rate is appropriate for environmental samples (Tapia-Palacios 2012). Given these limitations, the risk of infection from exposure to G. lamblia in urban flood water could be greater than the values reported in this study.

These limitations, however, could be partially solved through access to real-time precipitation data, when available, and by leveraging citizen science. The latter could provide an opportunity for technical support to improve the sampling and real-time data collection, while keeping citizens and stakeholders engaged in participatory decision-making processes. In particular, integrating the citizenry's knowledge, needs, and preferences, are of paramount importance when designing and implementing health studies aimed at fostering sustainability in urban contexts (Hadorn et al. 2006).

The three severity indices utilized in this study were built using a weighted linear combination technique that assigns different weights to influencing factors, flood frequency, and volume according to their relevance in determining the risk of infection. The factors’ weights emphasize judgments on the relative importance attached to the information related to each influencing factor (Chou 2013). However, we acknowledge that more than three possible sets of weight combinations are plausible. Therefore, we recommend the use of a multi-criteria decision analysis (MCDA) framework conducted with different stakeholders as an alternative approach to elicit the weights of factors in a systematic and robust manner.

QMRA outcomes are sensitive to assumptions, given the sampled ranges of the system's parameters, such as ingested volume and the probability of pathogen survival. To deal with this uncertainty, we stochastically sampled from literature-based reported distributions of some of the system's parameters. However, further research should focus on parameterizing the QMRA model from local empirical data so that specific conditions could be explicitly simulated. Moreover, simulating risk employing multiple exposure events based on mobility data, if available, could also improve model estimates.

Urban flood water in Mexico City represents a pathogenic transmission source due to the presence of high densities of fecal indicators and pathogens. QMRA results indicate that the risk of G. lamblia infection surpasses recommended international limits in more than one-half of the samples. Our analysis suggest that greater flood frequency could be conducive to a higher risk of G. lamblia infection, especially in the city's periphery, where marginalization is usually high. Consequently, coordination between the water management and public health systems, along with robust and long-term decision making, must be a priority for sustainable urban planning in developing countries such as Mexico. Exposure to the pathogens present in urban flood waters, as well as the associated risk, should be considered for the population's health and well-being in developing countries. The latter is especially needed, given the expected increases in future flooding occurrences due to climate change.

We acknowledge the support of Dr Hallie Eakin of the School of Sustainability, Arizona State University (ASU), and Dr Luis A. Bojórquez-Tapia from the Laboratorio Nacional de Ciencias de la Sostenibilidad, Instituto de Ecología, UNAM, during the development of this project. We would also like to thank Fidel Serrano-Candela, Luis Villareal-Ávila, and Blanca Hernández-Bautista for their support at different stages of the research. The data provided by Sistema de Aguas de la Ciudad de México (SACMEX), by Consejo Nacional de Población (CONAPO), and by Secretaría de Salud (Ministry of Health) was fundamental for model implementation.

This work was supported partially for field and laboratory work by the [National Science Foundation] under grant [CNH Grant 1414052] for project The Dynamics of Multi-Scalar Adaptation in the Megalopolis: Autonomous Action, Institutional Change and Social-Hydrological Risk (MEGADAPT) and by the [Inter-American Institute for Global Change Research] under Grant [CRN3108].

The authors declare that they have no conflict of interest.

All relevant data are included in the paper or its Supplementary Information.