Predictability of Vibrio cholerae in Chesapeake Bay

ABSTRACT Vibrio cholerae is autochthonous to natural waters and can pose a health risk when it is consumed via untreated water or contaminated shellfish. The correlation between the occurrence of V. cholerae in Chesapeake Bay and environmental factors was investigated over a 3-year period. Water and plankton samples were collected monthly from five shore sampling sites in northern Chesapeake Bay (January 1998 to February 2000) and from research cruise stations on a north-south transect (summers of 1999 and 2000). Enrichment was used to detect culturable V. cholerae, and 21.1% (n = 427) of the samples were positive. As determined by serology tests, the isolates, did not belong to serogroup O1 or O139 associated with cholera epidemics. A direct fluorescent-antibody assay was used to detect V. cholerae O1, and 23.8% (n = 412) of the samples were positive. V. cholerae was more frequently detected during the warmer months and in northern Chesapeake Bay, where the salinity is lower. Statistical models successfully predicted the presence of V. cholerae as a function of water temperature and salinity. Temperatures above 19°C and salinities between 2 and 14 ppt yielded at least a fourfold increase in the number of detectable V. cholerae. The results suggest that salinity variation in Chesapeake Bay or other parameters associated with Susquehanna River inflow contribute to the variability in the occurrence of V. cholerae and that salinity is a useful indicator. Under scenarios of global climate change, increased climate variability, accompanied by higher stream flow rates and warmer temperatures, could favor conditions that increase the occurrence of V. cholerae in Chesapeake Bay.

Vibrio cholerae is both the causative agent of cholera and a natural inhabitant of the aquatic environment. Nearly 200 V. cholerae serogroups have been identified to date (70), but only two serogroups, serogroups O1 and O139, are associated with epidemic cholera. V. cholerae was first isolated from the Chesapeake Bay in the 1970s and was suggested to be an autochthonous member of the aquatic environment (17). Further studies demonstrated clearly that V. cholerae is, in fact, autochthonous to the Chesapeake Bay and to the aquatic environment in general (15,18,40). V. cholerae has since been detected in natural waters worldwide, including areas where clinical cases of cholera did not exist (32,38,43,69). These studies showed that the majority of environmental isolates of V. cholerae are members of non-O1, non-O139 serogroups. However, various non-O1, non-O139 V. cholerae strains have repeatedly been isolated from patients with diarrhea (20,59) and have shown a capacity to provoke localized diarrheal outbreaks (2,19,53,56).
Colwell (15) proposed that the natural aquatic environment serves as the reservoir for V. cholerae and that it may play a critical role in pandemics of cholera. Horizontal gene transfer, which has been demonstrated in V. cholerae by Waldor and Mekalanos (66), has been proposed as a mechanism for the emergence of new pathogenic strains (4), and recent studies have confirmed such mechanisms (23). Environmental isolates of V. cholerae have been shown to harbor genotypes associated with virulence (10,54), thus supporting the hypothesis that the natural population of V. cholerae, including V. cholerae non-O1, non-O139 isolates, can serve as a precursor for new pathogenic or epidemic strains. V. cholerae can pose a public health risk ranging from limited outbreaks to epidemics when it is ingested via untreated water or contaminated shellfish (49,51). Because of this inherent risk, it is relevant to understand the mechanisms that affect the natural population of V. cholerae in the environment. Studies in coastal and estuarine regions of different parts of the world have shown that temperature and salinity play roles in the occurrence of V. cholerae in the environment (3,39,65). The objective of the study reported here was to model the occurrence of V. cholerae in the Chesapeake Bay as a function of environmental parameters. In particular, the occurrence of V. cholerae was investigated with respect to salinity, temperature, association with plankton, the Susquehanna River freshwater influx, and seasonality in order to understand how changing climate conditions might affect the ecology of V. cholerae in the environment.

MATERIALS AND METHODS
Shore sampling sites. Sampling was done from January 1998 to February 2000 at coastal sites throughout the upper Chesapeake Bay (Fig. 1). Samples were collected from surface water, about 1 m from the surface. Station F (Susquehanna Flats) is located at the northern tip of the bay at the mouth of the Susquehanna River, where the water salinity is Ͻ0.5 ppt. Other sampling sites are located in mesohaline water with salinities ranging from 2 to 15 ppt. Station H, on the eastern shore, is at the Horn Point Laboratory on the Choptank River. Station K is on Kent Island at an open water point of the main stem of Chesapeake Bay. On the western shore, station S is located at the Smithsonian Environmental Research Center on the Rhode River. In January 1999, a site in the inner harbor of Baltimore (station B) was added to the study. Plankton samples could not be collected in January 2000 except in Baltimore harbor, because the sampling sites in Chesapeake Bay were frozen.
Sampling was accomplished, in most cases, within a 2-day period and was done twice monthly during the summer (June, July, August) and monthly the rest of the year. During the entire study, a total of 32 sampling trips were accomplished over 26 months. The seasons were defined as follows: winter, December to February; spring, March to May; summer, June to August; and autumn, September to November.
Cruise sampling. Additional sampling was carried out aboard the research vessel Cape Henlopen in August and September 1999 and in June and August 2000. Nine stations on a north-south transect from the upper bay to the mouth of the bay (Fig. 1) were sampled on each cruise within 2 days. Temperature and salinity were recorded by using on-board instruments. Plankton were collected by using a plankton net on a 10-m vertical tow. In 1999 three fractions, water, P20, and P64 (as described below), were collected, and in 2000 two fractions, water and P64, were collected. The enrichment procedure (see below) was used to detect V. cholerae. Sample collection. One surface water fraction (fraction W) and two plankton fractions were collected in sterile plastic containers at each of the shore sampling sites. To collect plankton samples, simple 20-and 64-m-mesh, 30-cm-diameter, three-point bridle plankton nets fitted with a polyvinyl chloride collecting bucket (Aquatic Research Instrument, Lehmi, Idaho) were used. The small-plankton fraction (P20) was comprised of phytoplankton and small microzooplankton that passed through the 64-m-mesh net and were captured by the 20-m-mesh net. The large-plankton fraction (P64) included microzooplankton captured by the 64-m-mesh net. Plankton were collected either by towing or by pumping water for 5 min (ca. 500 liters) through the nets. The two methods were used in parallel in the field for 1 month. When the total numbers of bacteria were compared by the method of Bland and Altman (5), the two methods were found to be not significantly different (P Ͼ 0.55).

FIG
Sample processing. In the field, 20 ml of each water sample and 10 ml of each concentrated plankton sample were fixed with 2% (final concentration) formaldehyde (Fisher Scientific, Pittsburgh, Pa.) to obtain total bacterial direct counts and for plankton characterization, respectively. Samples for other assays (see below) were transported at the ambient temperature to the laboratory and processed within 6 h of collection.
Environmental parameters. Surface water temperature and salinity were measured at the sites by using a field thermometer and a diffractometer ( Total bacterial direct counting. Ten-milliliter portions of fixed water samples were incubated in the dark with 10 M (final concentration) DAPI (4Ј,6-diamino-2-phenylindole) (Sigma, St. Louis, Mo.) at room temperature for 2 h. After staining, 1 to 3 ml of each sample was filtered through a 25-mm-diameter, 0.2-m-pore-size polycarbonate black membrane (K02BP02500; Osmonics, Inc., Minnetonka, Minn.) in triplicate. Filters were mounted on microscope slides by using low-fluorescence immersion oil (type A or FF) and were examined by using a ϫ100 oil immersion lens on a fluorescent microscope fitted with a 420-nm UV filter. A minimum of 400 cells or 10 microscope fields were counted, as recommended by Kepner and Pratt (44). Cell concentration was computed as follows: number of cells per milliliter ϭ (average number of cells per square) ϫ (number squares per filter) ϫ (dilution factor)/sample volume.
Plankton identification. Zooplankton analysis was done for formalin-fixed P64 fractions collected from June 1998 to February 2000 by Nuruddin Mahmood at the Institute of Marine Biology, University of Chittagong, Chittagong, Bangladesh. Organisms were identified to the level of order for adult crustaceans and to the level of subclass for nauplii. For statistical analysis the zooplankton data were log transformed (log[x ϩ 1]) and grouped in the following categories: copepod nauplii, copepodites, adult calanoid copepods, adult cyclopoid copepods, adult harpacticoid copepods, rotifers, barnacle nauplii, and others (cladocerans, polychaetes, ostracods, oligochaetes, amphipods, nematodes, insect larvae, and cumaceans), which when combined represented less than 2.5% of the total number. Enrichment, culture, and molecular probes. V. cholerae was isolated by using the alkaline peptone water enrichment procedure described elsewhere (12). Briefly, 250-ml water samples or 25-ml plankton samples were filtered with a 0.22-m-pore-size Supor-200 membrane (Gelman Laboratories, Ann Arbor, Mich.) and incubated for 6 to 18 h at 30°C in 100 ml of alkaline peptone water (1% [wt/vol] peptone, 1% [wt/vol] sodium chloride; pH 8.6). Two loopfuls from each broth culture were streaked onto thiosulfate citrate bile salt sucrose agar (Oxoid Ltd., Basingstoke, England) in triplicate. After overnight incubation at 37°C, presumptive V. cholerae colonies were subcultured on Luria-Bertani agar (Difco, Detroit, Mich.) containing 2% (wt/vol) NaCl, a nonselective medium.
Isolates were screened by using a rapid biochemical test (12) that selected for colonies which were negative for both arginine dihydrolase activity and esculin hydrolysis. Presumptive V. cholerae isolates were confirmed by PCR by using V. cholerae-specific primers based on the 16S-23S rRNA intergenic spacer regions (13) ( Table 1). Alternatively, presumptive colonies were dot blotted onto Luria-Bertani agar, transferred to filter paper (68), and hybridized with a 32 P-labeled V. cholerae probe (Table 1). Positive colonies were identified as dark spots by autoradiography. V. cholerae O1 and Vibrio mimicus were grown on the same petri dish and processed along with presumptive colonies, serving as positive and negative controls, respectively. Enrichment data were collected from January to August 1998 and from March 1999 to February 2000.
Serology and identification of the O1 and O139 serogroups. V. cholerae isolates were examined to determine if they were members of the O1 and O139 serogroups by slide agglutination by using a polyclonal antibody specific for O1 surface antigen and a monoclonal antibody specific for O139 surface antigen (Centre for Health and Population Research, Dhaka, Bangladesh), respectively. A saline solution was used as a control to identify self-agglutinating isolates that could not be tested by this method. Self-agglutinating strains were tested by PCR by using O1-and O139-specific primers (Table 1). DFA O1. V. cholerae O1 was detected by using a direct fluorescent-antibody assay and direct viable counting for V. cholerae serogroup O1 (DFA O1) as described elsewhere (9). Briefly, water and plankton samples were incubated overnight at 30°C with 0.002% nalidixic acid and 0.025% yeast extract with shaking (45). Samples were homogenized with a glass tissue grinder, fixed with 2% formaldehyde, and stored at room temperature until they were processed. Samples were processed according to the DFA O1 kit's instructions; 10-l portions of samples were air dried on a microscope slide, fixed with ethanol, and labeled with a V. cholerae DFA reagent specific for V. cholerae O1 (Cholera DFA; New Horizons Diagnostics Corporation, Columbia, Md.). Slides were examined at a magnification of ϫ1,000 by using an epifluorescence microscope (Olympus model AH-2) and an HBO 200W/2 mercury lamp (OSRAM, Munich, Germany).
Chlorophyll a determination. Water samples (25 ml, in triplicate) were filtered with 47-mm-diameter glass microfiber filters (GF/F; Whatman International Ltd., Maidstone, England). The filters were wrapped in aluminum foil and frozen immediately at Ϫ80°C. For extraction of the chlorophyll a pigment, the frozen filters were placed in 10 ml of methanol at room temperature for at least 6 h in the dark. The supernatants were transferred to 15-ml centrifuge tubes and centrifuged at 1,200 ϫ g for 15 min. Fluorescence measurements were obtained with a luminescence spectrophotometer (Perkin-Elmer LS-5B) with the excitation and emission wavelengths set at 663 and 430 nm, respectively. By employing methods described by Holm-Hansen and Rieman (31), fluorescence was measured before and after acidification with 0.1 ml of 0.1 N HCl. Pure chlorophyll a extracted from algae (C-6144; Sigma) was used to establish a standard curve. Concentrations of the standard were determined by spectrophotometry (63) by using the specific absorption coefficient for methanol (␣ ϭ 79.95 liters g Ϫ1 cm Ϫ1 ) (37,52). Corrected chlorophyll a concentrations were computed as described elsewhere (1).  VOL. 69,2003 PREDICTABILITY OF V. CHOLERAE Statistical analyses. Because the methods used for detection of V. cholerae were not quantitative, the statistical analysis was done by using the presence or absence of V. cholerae in each sample. Logistic regression (PROC LOGISTIC from the statistical software package SAS, version 8.2; SAS Institute Inc., Cary, N.C.) was used with the presence or absence of V. cholerae as the response of interest and provided an odds ratio value. A stepwise selection process was used, in which descriptive variables were entered in turn into the model. Variables remained in the model only if they were significant at each step, when a new variable was added. Two analyses were done; one corresponded to the presence or absence of V. cholerae for the water fraction, and the other corresponded to the presence or absence of V. cholerae for the combined plankton fraction. In the latter case, the combined plankton sample was considered positive when either plankton fraction (P20 or P64) was positive. A response surface, as described by Carter et al. (7), was used for the analysis with temperature and salinity. The cross-product term between salinity and temperature indicates an interaction, and the cross-product with the same variable indicates a quadratic effect. Cochran's Q statistic for proportions (25) was used to test for significance between proportions in the three sampling fractions (water and two plankton fractions). Binary classification trees were computed by using S-Plus, version 6.0 (Insightful Corp., Seattle, Wash.), as a nonparametric alternative to logistic regression (21). For both the logistics regression and the binary classification tree, model agreement was evaluated by determining whether the model correctly predicted the presence or absence of V. cholerae for each sample used in the analysis. Throughout the analysis, a difference was considered significant if the P value was Ͻ0.05.

Detection of V. cholerae.
The enrichment method used in this study detects culturable V. cholerae, and the DFA O1 method was used to detect both culturable and nonculturable cells of the O1 serogroup. The two methods were in agreement 76% of the time; i.e., samples yielded the same results (both positive or both negative) with both methods. When the two methods did not agree, in 60% of the cases samples that tested positive by DFA O1 were negative as determined by enrichment. Overall, 21.1% of the samples (n ϭ 427) were positive for V. cholerae as determined by enrichment, including 22.6% of the shore samples (n ϭ 337) and 15.6% of the cruise samples (n ϭ 90); 23.8% of the shore samples (n ϭ 412) were positive for V. cholerae as determined by DFA O1 ( Table 2). The proportion of positive samples at the freshwater site (sampling site F) was 11.0% for both the enrichment (n ϭ 73) and DFA O1 (n ϭ 91) analyses. DFA O1-positive samples and culturable V. cholerae isolates were obtained at each of the five shore sampling sites.
Combined shore and cruise samples yielded 361 non-O1, non-O139 V. cholerae isolates obtained by enrichment. All isolates were tested by PCR with primers for V. cholerae (i.e., primers for identifying V. cholerae regardless of the serogroup). All isolates were tested by serology to determine whether they were O1 and O139 isolates. However, the self-agglutinating strains (which could not be analyzed by serology) were tested by PCR to determine whether they were O1 and O139 isolates (see Materials and Methods).
The data obtained for shore samples by both the enrichment and DFA O1 methods showed that there was an increased frequency of positive samples during the warmer months (Fig.  2). The enrichment and DFA O1 data showed similar patterns in 1998, whereas the seasonal pattern was not as marked for enrichment data in 1999. A smaller proportion of the samples was positive in 1999 than in 1998, and the reduction was greater when the enrichment method was used than when the DFA O1 method was used (Fig. 2). The enrichment method is not quantitative. However, given a consistent sampling effort, there was a striking difference between the number of isolates collected in 1998 and the number of isolates collected in 1999 for the brackish near-shore samples. At sampling sites K, S, and H, the yield of V. cholerae isolates (i.e., the number of isolates) in 1999 was only approximately 5% of that in 1998 (17 and 289 isolates, respectively). As determined by enrichment, the proportion of positive samples decreased from 97.8% in summer 1998 to 11.12% in summer 1999, and as determined by DFA O1 the proportion of positive samples decreased from 75.0 to 20.4% (Table 2). In 1999, sampling site B yielded the largest number of isolates and the highest proportion of positive samples compared to the other shore sites, as determined by both enrichment (17.9% versus 2.6 to 12.8%) and DFA O1 (45.5% versus 8.9 to 35.6%). Sampling site F yielded approximately one-tenth the number of isolates that the other shore sites yielded in 1998. However, the number of isolates at this site was nearly constant in 1998 and 1999, and the proportion of positive samples, unlike the proportion of positive samples at the other shore sites, increased in summer 1999 compared to summer 1998, as determined by both detection methods ( Table   TABLE 2  2). Furthermore, for site F the typical seasonal pattern observed at the other shore sites was not clearly visible (Fig. 2). For these reasons and because the logistic regression model for freshwater appeared to differ substantially from that for the brackish sites (data not shown), the samples from the freshwater sampling sites were analyzed separately from those from the brackish sites. For cruise samples, culturable V. cholerae isolates were obtained from seven of the nine sampling sites along the northsouth transect of Chesapeake Bay (Fig. 3A). In 1999 the percentage of positive samples was significantly lower than the percentage of positive samples in 2000 (9.3 versus 23.7%) ( Table 2), and one-half as many isolates were obtained (8 versus 16 strains). The results obtained with the cruise samples also suggest that both in terms of the percentage of positive samples and in terms of the number of isolates, V. cholerae occurs more frequently in the northern part of the bay (sites 908 through 834), where the salinity is lower (Fig. 3B). No such site difference was detectable among the brackish shore sites located in a smaller geographical area, where the salinity and temperature conditions did not vary as much at a given time.
The proportion of positive samples was higher for water samples than for plankton samples, but it was significantly higher only for enrichment samples collected at the brackish sites, at both the shore and the cruise (Fig. 4). In the case of enrichment, the two plankton fractions (P20 and P64) did not differ significantly. In the case of DFA O1, the two plankton fractions differed for the freshwater samples because V. cholerae was not detected in the P64 fraction.
Environmental parameters. The ranges of the environmental parameters measured in this study are summarized by site in Table 3. The salinity at the brackish sites was inversely correlated with the Susquehanna River flow, with a 3-week lag (P Ͻ 0.001; R ϭ Ϫ0.56). Table 4 shows that the measured salinities at brackish sites reflected the general flow pattern of the Susquehanna River compared with the long-term average. Salinity at the brackish sites, however, was not correlated with a lag in precipitation (1 to 21 days) or with 3-or 7-day rain accumulation. Temperature and to a lesser extent salinity displayed annual seasonal patterns (Fig. 2). For all sites, the total bacterial counts exhibited a seasonal pattern that followed the temperature variation. The log of the total bacterial count  tion exhibited great variability without a seasonal pattern. Similarly, there was no clear pattern for pH, seasonal or otherwise, and sites did not vary together. Table 5 shows, for the most abundant taxa, the percentages of samples in which a taxon was present or had a relative abundance in the sample that was greater than 20%. For the four brackish water, shore sites (sites B, K, S, and H), copepod nauplii were the most abundant zooplankton in 70.2% of all the samples (n ϭ 94). Furthermore, the number of copepod nauplii exceeded 20% of the number of zooplankton in 83.0% of the samples. Adult calanoid copepods, copepodites, and barnacle nauplii were the next most common groups ( Table 5). The freshwater site (site F) differed from the other shore sites in that rotifers were considerably more common (73.1 versus 33.0% of the samples) and accounted for a higher proportion of the organisms (50.0% of the samples had a relative abundance greater than 20%, compared to 10.6% of the samples from brackish sites), whereas barnacle nauplii were less common. While adult calanoid copepods were observed more frequently than adult cyclopoid copepods at brackish sites, the trend was reversed for the freshwater site (Table 5). Table 6 shows the relative distribution of V. cholerae-positive samples as a function of temperature and salinity. For shore samples, a higher temperature led to an increase in V. cholerae occurrence, except when the salinity was more than 14 ppt. Use of the odds ratio as a measure of association indicated that occurrence of V. cholerae was nearly 20 times more likely at a temperature of 19°C or higher as determined by enrichment culture and nearly 6 times more likely as determined by DFA O1. The salinity range from 2 to 8 ppt yielded the highest percentage of positive samples at temperatures of more than 19°C for both enrichment and DFA O1 analyses. At temperatures below 19°C, V. cholerae was detected more often by DFA O1 (10.9%) than by enrichment culturing (5.4%). At temperatures below 19°C, V. cholerae was detected more often by enrichment culture at salinities between 2 and 8 ppt, whereas it was detected more often by DFA O1 at salinities between 8 and 14 ppt. At temperatures of 19°C or more and salinities between 2 and 14 ppt, detection of V. cholerae was 5.2 times more likely by enrichment culture and 4.1 times more likely by DFA O1 than when the combination of these conditions was otherwise. In freshwater, temperatures higher than 19°C favored V. cholerae detection by both methods, but to a lesser extent than for the brackish water samples. Data from the summer cruises showed that elevated temperature alone was not sufficient to yield a high percentage of positive samples, since a temperature above 21°C yielded 15.6% positive samples (n ϭ 90), compared to 37.4% positive samples at the shore sites (n ϭ 174), and the percentage of positive samples dropped to 9.5% when the salinity was greater than 14 ppt. Even though V. cholerae was detected less frequently at a salinity greater than 14 ppt, it was not absent from high-salinity water, since V. cholerae was isolated from a site near the mouth of Chesapeake Bay where the salinity was 21.1 ppt.
Modeling. Statistical analysis showed that the models performed poorly when the freshwater site was included with the other sites. Consequently, the analysis was done separately for freshwater and brackish water sites. The selected models shown in Table 7 include only data from the brackish water sites, shore sites, and cruise sites. The water fractions and the combined plankton fractions were analyzed separately. For the freshwater sampling site alone, no significant explanatory variables were found. Numbers of adult copepods and copepodites appeared in the stepwise selection model but not at a significant level. For the brackish water sites, the logistic regression showed that temperature and salinity were at the core of a basic model for both the enrichment culture and DFA O1 data. Even when other parameters were allowed to enter the model  in a stepwise algorithm, they provided only minor improvement over the basic temperature-salinity model. Sometimes the variable total bacterial count, which is strongly correlated with temperature, was substituted for the variable temperature by the stepwise logistic regression. For example, in model 6 (Table 7), the agreement between the observed and predicted responses improved only from 75.5 to 77.8% when the additional variables chlorophyll a, rotifer, and barnacle nauplii were entered into the model. Also in this modified model, the variable total bacterial count was substituted for temperature. Similarly in model 1, the agreement between the observed and predicted values increased only from 82.2 to 83.1% when the variable copepod nauplii was added to the model. Analysis with the binary classification tree gave patterns similar to that obtained by the logistic regression analysis with stepwise selection. This method pointed toward a lower threshold for temperature (around 17°C) and toward an upper boundary for salinity (between 8 and 10 ppt). For enrichment cultures, the percentage of agreement between the observed and predicted values was slightly better with the binary tree ( Table 7). For DFA O1, the binary classification tree gave greater improvement over logistic regression. For both modeling methods, the results showed that the combined temperature and salinity conditions predict the presence of V. cholerae with an accuracy between 75.5 and 88.5%.

DISCUSSION
Enrichment culturing versus DFA O1. The fact that DFA O1 analysis yielded a higher frequency of positive samples than the enrichment culture method suggests that under environmental conditions suboptimal for V. cholerae growth, more cells may be nonculturable and, therefore, may not be detected by enrichment culturing (16). This observation is in agreement with previous studies that showed that V. cholerae exists in a viable but nonculturable (VBNC) state when environmental conditions such as temperature, salinity, osmotic pressure, pH, and nutrient concentration become less favorable for growth  and replication (33,34). Nonetheless, the pattern of detection observed here by enrichment culture corroborated the pattern obtained in a study done in the Choptank River (Chesapeake Bay, sample site H) with a non-culture-based detection method (29,30). Using fluorescent-oligonucleotide direct counting, Heidelberg et al. (30) showed that the abundance of the V. cholerae-V. mimicus subgroup was greater in the summer and that the relative proportion of this subgroup within the Vibrio-Photobacterium group increased with higher temperatures. In this study, V. cholerae isolates were V. cholerae non-O1 and non-O139. In contrast, detection by direct fluorescence by using the O1 antibody (DFA O1) gave positive results for the V. cholerae O1 population. The fact that the two methods did not always agree can be explained by the fact that the fluorescent-antibody assay principally assesses the nonculturable subpopulation of V. cholerae O1, whereas the enrichment procedure detects only culturable cells. This suggests that V. cholerae O1 in Chesapeake Bay is mostly VBNC. The assay, by itself, does not necessarily measure the public health risk associated with V. cholerae O1, since not all O1 strains may be pathogenic (47) and the genetic characteristics of nonculturable V. cholerae in Chesapeake Bay have yet to be determined. Nevertheless, V. cholerae O1 has previously been isolated from Chesapeake Bay (18), and VBNC strains have been shown to conserve their antigenic determinants (9), suggesting that a subpopulation of potentially pathogenic V. cholerae, either VBNC V. cholerae or V. cholerae that is difficult to culture, is present in Chesapeake Bay. Furthermore, because of the possibility of horizontal gene transfer and the presence in the environment of V. cholerae variants that potentially carry toxigenic genetic elements (24), an increase in the V. cholerae population or VBNC V. cholerae O1 in the environment may lead to an increase in public health risk.
Roles of zooplankton, temperature, and salinity. While all of the brackish sites had similar patterns of dominant zooplankton groups, the freshwater site differed from the brackish sites. This may be one reason why the detection pattern of V. cholerae was very different at the freshwater site (detection was infrequent and erratic). In this study, V. cholerae was detected more often in water samples than in plankton samples, and no clear association could be established between detection of V. cholerae and the presence of a specific group of zooplankton. These observations are similar to those reported in 1989 by Venkateswaran et al. (65) for non-O1 V. cholerae in the coastal waters of Japan.
In this study, V. cholerae showed a seasonal pattern of occurrence, which was correlated with higher temperatures. The data indicate that there is a temperature threshold between 17 and 19°C, and the frequency of occurrence of V. cholerae is significantly greater at temperatures above 19°C (P Ͻ 0.001). Furthermore, the results indicate that there is an optimal salinity range for V. cholerae detection (namely, between 2 and 14 ppt, especially at salinities below 8 ppt). These observations are in agreement with previous studies done in Chesapeake Bay, where V. cholerae was isolated only at salinities ranging from 4 to 17 ppt and only at temperatures above 10°C (40) and was more frequently detected when temperatures were greater than 17°C (14). In recent studies carried out in Southern California (39), Jiang found that high concentrations of V. cholerae were more frequently detected at salinities below 10 ppt but above 0 ppt. Laboratory microcosm studies also suggested that optimum growth of V. cholerae occurs at a range of salinities and that below or above this range growth decreases. For example, Cavari and Colwell (8) showed that there was rapid V. cholerae growth in microcosms at salinities of 5 and 10 ppt compared to the growth at a salinity of 0 or 30 ppt. Singleton et al. (60) documented that there was an absolute requirement for NaCl in the 10 strains of V. cholerae which they tested and showed that growth was optimum in microcosms at salinities of 15 and 25 ppt. Huq et al. (36) found little difference in the growth of V. cholerae in microcosms at salinities of 5, 10, and 15 ppt, except that at a salinity of 5 ppt V. cholerae culturability was reduced in the absence of copepods. Results of previous studies describing conditions that favor detection of V. cholerae are summarized in Table 8.
These results offer insight into cholera epidemiology both in developed countries and in developing countries, such as Bangladesh, where cholera is endemic. The seasonality of cholera in regions where cholera is endemic suggests that environmental conditions are linked to cholera epidemics (57). Recent studies (6,48) further support the role of environmental factors in endemic cholera, indicating that the ecology of V. cholerae in natural waters is influenced by environmental conditions, such as sea surface temperature. Along these lines, a study of diarrheal diseases, including cholera, in Peru indicated that there was a link with temperature and that there was a 19°C threshold, above which the number of cases increased (11).

Role of the Susquehanna River.
In the upper Chesapeake Bay, defined as the section north of the mouth of the Potomac River, which includes the shore sampling sites used in this study, the Susquehanna River provides nearly 90% of the overall freshwater influx (58). In this area, salinity is coupled to the freshwater flow from the Susquehanna River (28,58). Salinity averaged by season at the brackish sites was inversely related to the general flow trend of the Susquehanna River (Table 4). The first half of 1998 was marked by higher-than-usual river flow, and consequently, the average salinities at the near-shore sites in the winter, spring, and summer of 1998 were significantly lower (P Ͻ 0.005) than the values during the corresponding times of year in 1999, a low-flow year.
In contrast, the period from August to December 1998 was marked by very low flow, and the average autumn salinity turned out to be higher in 1998 than in 1999, but the values were not statistically different. At the cruise sites, the average salinity along the north-south transect was significantly higher (P Ͻ 0.02) in 1999, a year of low flow, than in 2000, a year of average flow. These observations illustrate (i) that salinity varies seasonally (for example, a spring freshet [i.e., lower salinity] was apparent in both 1998 and 1999) and (ii) that there is annual variability due to the magnitude of the Susquehanna River inflow, as described by Schubel and Pritchard (58).
Due to variation in the flux of freshwater and, along with it, nutrients, there is significant year-to-year variability in salinity, turbidity, and dissolved organic and inorganic matter throughout the bay, especially in the northern part. For example, the timing, location, duration, and amplitude of the chlorophyll peak, which depends on nutrient input, can vary up to 80% in certain parts of the bay from one year to the next (28). The average salinities in spring and summer were significantly lower in 1998 than in 1999, while the temperature patterns were similar. The periods of lower salinity, given similar temperature conditions, resulted in increased V. cholerae occurrence and much higher numbers of isolates. In contrast, little year-to-year variability was observed at the freshwater sampling site. Interestingly, V. cholerae was also detected much less frequently at that site, suggesting that the dynamics that regulate V. cholerae are very different in freshwater. The cruise samples collected in the summers of 1999 and 2000 also confirmed that 1999 was a year in which low V. cholerae yields were obtained. In this analysis, salinity may be the effective parameter per se, as laboratory studies have determined, but it may also be a proxy for other factors, such as turbidity or nutrient load, that were not measured in this study. Singleton et al. (61) studied the combined effects of nutrients and salinity on V. cholerae growth and showed that when the nutrient concentration was elevated, the V. cholerae concentration increased and the range of salinity tolerated was larger. Thus, an increase in the riverine influx, laden with nutrients, may favor the growth of V. cholerae by increasing the growth potential and by making the bacterium able to thrive over a wider salinity range. Whatever the mechanisms, these results suggest that year-to-year salinity variability or other factors associated with freshwater influx may generate large variations in the V. cholerae populations. For example, lower salinities in the upper bay due to increased Susquehanna River flow, a possible outcome of cli-mate change scenarios that include a doubling of atmospheric carbon dioxide (27), may favor increased occurrence of V. cholerae in Chesapeake Bay. Conversely, drought conditions, such as those experienced in recent years in the Chesapeake Bay region, may produce conditions less conducive to V. cholerae proliferation in the estuary. In the past century, precipitation has been changing in the United States, and there have been increases in extreme precipitation events (41,42). These changing rainfall patterns, most evident in the mid-Atlantic region, are consistent with expectations of a more vigorous hydrological cycle expected to accompany global warming (26,50,64). Based on the salinity findings of this study, expected changes in stream flows could modulate V. cholerae populations in Chesapeake Bay. Furthermore, the effects of sea level rise (a potential consequence of global warming) on the salinity regimen of an estuary such as Chesapeake Bay may also be important for long-term assessment.
Predictability. The statistical models used in this study described the occurrence of V. cholerae in Chesapeake Bay mostly as a function of temperature and salinity. This knowledge is valuable in itself because it provides an understanding of the large-scale processes that dominate the ecology of V. cholerae in Chesapeake Bay. Inclusion of the variables for plankton resulted in only marginal improvement to the models. Since the distribution of plankton in Chesapeake Bay is known to be patchy (28), an adequate model may require much higher temporal resolution and spatial resolution (on the order of days and kilometers, respectively) at the data acquisition step to properly describe the role of plankton. Thus, the association between plankton and V. cholerae, which has been well documented (22,35,62), may come into play at a small scale within large-scale processes.
Both the logistic regression and the binary classification tree proved to be useful modeling methods for describing V. cholerae occurrence and pointed out the most relevant parameters. As discussed by De'ath and Fabricius (21), binary classification trees are well suited for describing complex ecological data. In this study, binary classification trees dealt effectively with the threshold situation and outperformed logistic regression models. Such modeling techniques could be useful for describing V. cholerae occurrence in other estuarine or coastal regions. The analysis showed that freshwater and brackish sites should be analyzed separately. Even though this study did not yield significant results for the freshwater site, models could be applied to freshwater or nearly freshwater conditions if several freshwater sampling sites are included in order to provide enough power for analysis. To become a more accurate predictive tool, the models need to be developed over a period of time that is longer than 2 years, so that year-to-year variability can effectively be taken into account.
In conclusion, the results provided by this study showed that there is clear seasonality in the occurrence of V. cholerae in the temperate estuarine environment of Chesapeake Bay, as detected both by enrichment culturing and direct fluorescentantibody detection for V. cholerae O1. Even though V. cholerae was found to occur at a wide range of salinities (from freshwater to a salinity of 21 ppt), the salinity most favorable for V. cholerae was between 2 and 14 ppt, and the optimal salinity appeared to be between 2 and 8 to 10 ppt. In brackish water, temperature and salinity are the two major parameters deter-VOL. 69,2003 PREDICTABILITY OF V. CHOLERAE 2783 mining the frequency of occurrence and geographical distribution of V. cholerae. However, in freshwater or at a high salinity (conditions that may be suboptimal for growth of V. cholerae), another factor, such as zooplankton, may be the dominant variable that explains the occurrence of V. cholerae. In the upper Chesapeake Bay, where V. cholerae was most frequently detected, salinity was linked to freshwater input from the Susquehanna River. Climate and weather conditions greatly contribute to river discharge, and there is great year-to-year variability. In this study, lower-than-normal salinity resulted in high V. cholerae occurrence in the spring and summer, suggesting that climate patterns have the potential to shift the estuarine ecosystem toward circumstances for which very favorable conditions for V. cholerae can occur. Since the salinity of Chesapeake Bay can affect the occurrence of V. cholerae, the predictability of V. cholerae in Chesapeake Bay depends on large-scale phenomena that regulate parameters such as temperature and salinity. Statistical models provided a useful tool for describing such phenomena and can be used in other estuarine regions of the world, like the Bay of Bengal, where cholera is endemic.