Comparative analysis shows that bacterivory, not viral lysis, controls the abundance of heterotrophic prokaryotic plankton

Abstract

Empirical models derived from literature data were used to compare the factors controlling prokaryotic abundance (PN) and prokaryotic heterotrophic production (PHP) in solar salterns. These empirical relationships were generated as multiple linear regressions with PN or PHP as dependent variables, while the independent variables were chosen to reflect the likely sources of organic matter, inorganic nutrients and temperature. These variables were then measured in solar salterns and the predictions made by the general relationships were compared to actual saltern values of PN and PHP. Saltern ponds of salinity higher than 100‰ departed significantly from the general relationships, while the ponds of salinity lower than 100‰ fitted well within the range of values predicted by the general models. The most likely explanation for the discrepancy of the former was the absence of bacterivory. This hypothesis was tested with data from other very different aquatic systems: karstic lakes with anaerobic hypolimnia and two marine areas in the Mediterranean and the Southern Ocean. The anoxic regions of karstic lakes departed significantly from the predictions of the general model, while the oxic layers conformed to the predictions. As in the case of salterns, this difference could be explained by the presence of significant predation in the oxic, but not in the anoxic, layers of these lakes. Finally, two marine areas with similar predation pressure on prokaryotes but very different impacts of viral lysis were tested. In all cases, PN values conformed to the predictions, suggesting that lysis due to viruses is not the main factor controlling PN in aquatic systems, which is more likely to be determined by the balance between bacterivory and resource supply. The present work also demonstrates the usefulness of empirical comparative analyses to generate predictions and to draw inferences on the functioning of microbial communities.

Multiple regression analysis, Solar saltern, Bacterivory, Viral lysis, Bacterial abundance

1 Introduction

The current paradigm of planktonic food webs recognizes the fundamental role played by heterotrophic prokaryotes [1]. However, the mechanisms determining the actual values of abundance and activity of prokaryotes remain controversial and may be different in different ecosystems [2–5]. Accumulation of data from a wide range of aquatic ecosystems during the past two decades has allowed the comparative study of their microbial food webs [6–11]. One approach to this comparative study is the establishment of empirical relationships between prokaryotic plankton parameters and other environmental variables judged to be relevant. Thus, a relationship between bacterial abundance and chlorophyll a[8] is believed to reflect the strength of the link between bacteria and autotrophic phytoplankton across a wide range of aquatic systems. Likewise, a relationship between bacterial heterotrophic production and primary production has been found to be significant across systems [6]. These empirical relationships, usually in the form of linear regressions between exponentially transformed variables, define a region of probable values within the universe of all possible values [9].

One of the main weaknesses of this approach is that the addition of data from new environments may change the relationships. In fact, most data have been gathered in the ecosystems most accessible to well funded research institutions (see for example Fig. 1 in [10]). The whole range of existing systems with different values for the microbial parameters has not been explored. Outliers are usually eliminated from the analysis, for example, Lake Elmenteita was excluded from the regression by Bird and Kalff [8]. In fact, the Lake Elmenteita data on bacterial abundance [12] provided the first clue that hypersaline lakes might be different from other more ‘common’ systems. This impression was confirmed by the studies of Finlay et al. [13] and recently by Zinabu and Taylor [14], who found a regression line between bacterial abundance and chlorophyll a with a slope significantly different from those in the literature. To establish the limits of the general models and to check their underlying assumptions the outliers may be actually very helpful. We decided to use hypersaline ecosystems in a way analogous to that in which mutants are used in physiology: by studying systems in which a given function is impaired or exacerbated, the mechanisms regulating the ‘common’ functioning of the system can be better appreciated. In this work we used the ponds in solar salterns as such outlier ecosystems.

1

Conceptual model for the analysis of the factors determining the abundance and production of prokaryotic plankton. The variables that were included in the regression analysis are shown in boldface. See text for explanation.

The analysis of salterns in a previous paper [15] suggested that predation was the most important cause for the differences in prokaryotic plankton abundance in salterns versus other systems. Here we use karstic lakes to confirm this suggestion in a very different kind of environment, where water layers with and without predation were present. Finally, we use two different marine areas with a low and high impact of viral lysis to see whether this factor could also be important in regulating prokaryotic abundance (PN) in aquatic systems.

Throughout the paper we use the terms prokaryotic plankton and prokaryotic heterotrophic production (PHP) instead of bacterioplankton and bacterial heterotrophic production. This is to recognize the presence and abundance of Archaea in all the sytems analyzed. Archaea are known to make up to 20% of the prokaryotic count in the ocean [16], between 80 and 90% in the high salinity ponds in the salterns (Antón and Roselló-Mora, personal communication) and they have been shown to be present in the anaerobic layers of karstic lakes [17]. However, we retain the term ‘bacterivory’ for convenience, since any other alternative would be unnecessarily cumbersome and confusing.

2 Materials and methods

2.1 Environments studied

The environments studied are summarized in Table 1. The data from salterns are those published in Pedrós-Alió et al. [15] and Oren [18,19]. These salterns are formed by a series of connected ponds. As water evaporates and salinity increases, water is pumped or gravity-fed to the next pond, such that the salinity in each particular pond is kept, within narrow limits, essentially constant. The ponds provide a range of salinities from that of seawater to sodium chloride precipitation. Each pond can thus be considered at equilibrium and the biota in any given pond as a well-adapted and established community for that particular salinity. We have previously shown that viral lysis has a relatively low impact throughout the gradient [20]. Predation on prokaryotes, on the other hand, has a very strong impact at lower salinities, but completely disappears above 250‰ salinity [15].

1

Some characteristics of the aquatic systems included in the analysis

System	Coordinates		Depth (m)	Temperature (°C)	Prokaryotic abundance (cells ml−1)	Bacterivory	Viral lysis	Comments	Refs.
	latitude	longitude
Salterns				30	106–108			high salinity
Bras del Port	38°12′N	0°36′W	≤1			high to absent	not determined		[15]
La Trinitat	40°35′N	0°41′E	≤1			high to absent	lowa		[15]
Eilat	29°N	35°E	≤1			not determined	not determined		[18,19]
Karstic lakes				7–20	106–107		not determined	sulfide in hypolimnion	[37]
Lake Cisó	42°8′N	2°45′E	8			high to lowb	not determined	anaerobic holomictic	unpub.c
Lake Vilar	42°8′N	2°45′E	9				not determined	biogenic meromictic	unpub.c
La Cruz	39°59′N	1°52′W	24				not determined	crenogenic meromictic	unpub.c
El Tobar	40°33′N	2°3′W	19.5				not determined	biogenic meromictic	unpub.c
Arcas-2	39°59′N	2°8′W	14.5				not determined	aerobic holomictic	unpub.c
El Tejo	39°59′N	1°52′W	11				not determined	aerobic holomictic	unpub.c
Open marine areas
Mediterranean Sea	40–42°N	2–6°E	≤2000	14–20	104–105	moderated	below detectione	oligotrophic	[27]
Southern Ocean	62–65°S	59–66°W	≤4000	−1.8/+4	104–105	moderatef	very highg	cold	[38]

^a [20].

^b [23,,25].

^cCalderón-Paz, J.I. (1997) Ph.D. thesis, University of Barcelona.

^dMarrasé, C. and Vaqué, D., in preparation.

^e [28].

^f [29].

^g [30].

The karstic lakes analyzed are located in two Spanish regions: Banyoles (Girona) and Cuenca. Information about these systems can be found in Miracle et al. [21] and Pedrós-Alió and Guerrero [22]. They are characterized by a very sharp stratification that separates an oxic epilimnion from an anoxic hypolimnion. While predation on prokaryotes is relatively high in the oxic layers [23,24], it is much reduced in the anoxic layers [24,25].

The Mediterranean waters studied show strong vertical stratification in the summer. The chlorophyll vertical profile presents a deep maximum slightly above the thermocline [26]. We studied a transect offshore from Barcelona in June 1995 (boreal summer) that covered three different zones [27]: the coastal area on the Continental Shelf (depth <200 m), the Shelf break frontal area (depth 500–1500 m) and the deep open sea area (depth ∼2000 m). Predation in this area was moderate (Marrasé and Vaqué, in preparation), while viral activity was below detection limits [28].

The area studied in the Southern Ocean included stations within the Gerlache and Bransfield Straits as well as in the Bellinghausen Sea, south of Drake Passage. The study was carried out during December 1995 (austral summer) and each zone had different stratification characteristics as well as chlorophyll a concentrations. Bacterivory was uniformly moderate [29], while viral impact was rather high [30].

The conceptual model used is shown in Fig. 1. Prokaryotic plankton abundance is the result of a balance between loss factors (mostly viral lysis, bacterivory, advection and sedimentation) and PHP. PHP, in turn, is a consequence of the available carbon and physicochemical parameters such as temperature and inorganic nutrients that probably impose an upper limit to production. Finally, the available organic carbon depends on inputs from phytoplankton (represented by chlorophyll in the model), allochthonous carbon inputs and other sources, such as sloppy feeding or excretion by herbivores. We screened the literature for papers with values for as many variables as possible to build quantitative relationships able to predict prokaryotic plankton abundance and production from the values of the other variables. We built multiple linear regressions with the choice of dependent and independent variables based on the model in Fig. 1. The predictions could then be compared to actual values in particular aquatic systems in which one of the factors was missing (for example predation in the high salinity ponds). We expected this analysis to reveal the relative importance of different factors in determining the actual values of prokaryotic plankton abundance and production.

2.2 Construction of the general relationships

A database of microbial parameters in different aquatic systems was gathered from the literature (Calderón-Paz, 1997, Ph.D. thesis, University of Barcelona). This database was created by combining those of White et al. [7] and Vaqué et al. [31], eliminating those studies in which some of the essential variables (PN and PHP, temperature and chlorophyll a) were missing and those where PHP was measured with techniques other than leucine or thymidine incorporation. In total, the database included 705 data points from 53 different studies. To compare values from systems with very different depths (from deep sea to shallow salterns), values for the respective photic zones were integrated and divided by the depth of integration, thus giving a weighted average of the photic zone. The values in this database were then used to derive general relationships between prokaryotic biomass and production and variables such as temperature and chlorophyll concentration. The purpose was to obtain empirical equations from the data gathered in many different systems that would reflect the relationships shown in Fig. 1. Except for temperature, all the other variables were logarithmically transformed to satisfy the criteria of normality and homogeneity of variances required by regression analysis. Simple and multiple linear regressions were carried out by the least squares method (model I), using backward elimination for the multiple regressions [32].

3 Results

3.1 General models

3.1.1 Simple regressions

Given that PN and PHP are commonly found to covary with chlorophyll a concentration (Chla) [6,8] and temperature [7], we first inspected the univariant relationships between these variables (Table 2). As expected, all these relationships were significant as were those of PN and PHP for both freshwater and marine systems. In marine data, PN was better related to Chla (r²=0.61) than to temperature (r²=0.11) while PHP was better correlated to temperature than to Chla (Table 2). In freshwater, however, both PN and PHP were better correlated to temperature than to Chla. PN and PHP covaried with an r² of 0.5–0.6.

2

Simple linear regressions (Y=a+bX) for the prediction of PN and PHP

Y	X	n	a±95% C.I.	b±95% C.I.	r2
Marine systems
PHP	Temperature	244	−0.10±0.13	0.07±0.01	0.52
PHP	Chla	94	0.93±0.09	0.34±0.13	0.22
PN	Temperature	244	−0.20±0.11	0.02±0.01	0.11
PN	Chla	90	0.16±0.03	0.24±0.04	0.61
PN	PHP	244	−0.32±0.05	0.45±0.05	0.55
Freshwater systems
PHP	Temperature	242	0.60±0.14	0.05±0.01	0.33
PHP	Chla	143	1.10±0.15	0.38±0.16	0.13
PN	Temperature	217	0.24±0.08	0.02±0.01	0.21
PN	Chla	159	0.18±0.08	0.36±0.09	0.27
PN	PHP	201	0.12±0.07	0.34±0.05	0.50

All variables except temperature have been logarithmically transformed. Units for each variable are: PN=×10⁹ cells l⁻¹; PHP=μg C l⁻¹ day⁻¹; Chla=μg l⁻¹; temperature=°C. 95% C.I.: 95% confidence interval; n=number of cases used in the regression. All regressions were highly significant (P<0.001).

3.1.2 Multiple regressions

Multiple regressions for marine and freshwater systems turned out to be significantly different (with an analysis of covariance). Following Billen et al. [4], Ducklow [5] and White et al. [7], we expected PHP to depend on the sources of organic matter (with Chla as a surrogate variable) and temperature. PN was expected to depend on PHP as possibly modulated by temperature. Two empirical relationships were derived to predict bacterial abundance and production from other variables in marine systems (Table 3):

3

Multiple linear regressions (Y=a+bx₁X₁+…+bx_nX_n) for predictions of PN and PHP in marine and freshwater systems

	n	a	bTemp	bChla	bPHP	β1	β2	β3	Adjusted r2
Predictions of PHP
Freshwater	133	0.23±0.09	0.05±0.01	0.43±0.06	–	0.66	0.39	–	0.59
Marine	94	–	0.07±0.00	0.40±0.04	–	0.64	0.27	–	0.82
Predictions of PN
Freshwater	140	–	0.01±0.00	0.10±0.03	0.32±0.04	0.15	0.15	0.64	0.70
Marine	249	−0.18±0.04	−0.02±0.00	–	0.58±0.03	−0.32	–	0.99	0.65

β₁, β₂ and β₃ are the standardized partial coefficients for variables Temp, Chla and PHP, respectively. All variables except temperature have been logarithmically transformed. Units for each variable are: PN=×10⁹ cells l⁻¹; PHP=μg C l⁻¹ day⁻¹; Chla=μg l⁻¹; temperature=°C. All regressions were highly significant (P<0.001).

Log₁₀ PHP=0.07Temp+0.40 Log₁₀ Chla (r²=0.82, P<0.001, n=94)

Log₁₀ PN=−0.18+0.58 Log₁₀ PHP−0.02Temp (r²=0.65, P<0.001, n=249)

where PN is the prokaryotic number in 10⁹ cells per liter, PHP is the prokaryotic heterotrophic production in μg C l⁻¹ day⁻¹, Temp is temperature in °C and Chla is chlorophyll a concentration in μg l⁻¹. The data obtained in our own studies of salterns and marine regions were then compared to these regressions.

The regressions for the freshwater systems were (Table 3):

Log₁₀ PHP=0.23+0.05Temp+0.43 Log₁₀ Chla (r²=0.59, P<0.001, n=133)

Log₁₀ PN=0.04 Log₁₀ PHP+0.01Temp+0.1 Log₁₀ Chla (r²=0.70, P<0.001, n=140)

Note that for freshwater systems, PN was also a function of Chla.

3.2 Comparison of data from the salterns to the general relationships

To simplify comparisons of microbiological parameters with other marine and freshwater systems, we assigned individual ponds in the salterns to one of four groups according to their salinities. These groups correspond to the four biological domains established by Ortí-Cabo et al. [33] and Rodríguez-Valera [34]: salinity less than 100‰, salinity between 100 and 200‰, salinity between 200 and 300‰ and salinity greater than 300‰. The means and standard errors for each group were calculated for every variable measured [15]. These values could then be compared to the general relationships derived from the literature data.

Fig. 2 shows the general relationship between PN and Chla (Fig. 2A) and between PHP and Chla (Fig. 2B). In addition to the regression line obtained with values from the literature, the 95% confidence limits for the prediction of the dependent variable are also shown as discontinuous lines. The length of the lines indicates the range of values of each variable that were included in the calculations that generated the regression (the data available in the data set). Values of Chla for the four saltern groups fell within the range of values in the database. Both PHP and PN values, however, were clearly above the highest value in the database (Fig. 2A,B). Thus, the extreme physico-chemical conditions of the salterns seem to result in very high values of PHP and extremely high values of PN with moderate values of Chla, in comparison to the ‘common’ systems included in the database.

2

Linear regressions between PN (in 10⁹ cells l⁻¹, A) and PHP (in μg C l⁻¹ day⁻¹, B) with respect to Chla (in μg l⁻¹). The continuous line is the regression derived from literature data. The discontinuous lines indicate the 95% confidence limits. The lines are drawn only across the range of values existing in the database used to derive the regressions. The ponds in the solar salterns have been grouped according to their salinity (see text). Mean values and standard errors within each group of ponds are shown.

Fig. 3A shows the relationship between PHP and temperature. Values of PHP fell within the 95% confidence interval of the predictive equation. This relationship, therefore, was not different in salterns and in more common systems. Likewise, the relationship between PN and PHP (Fig. 3B) correctly predicted the values found in the salterns, even though the PN values were much higher than any of those in the database.

3

A: Linear regression between PHP (in μg C l⁻¹ day⁻¹) and temperature (in °C). B: Linear regression between PN (in 10⁹ cells l⁻¹) and PHP (in μg C l⁻¹ day⁻¹). The continuous lines are the regressions derived from literature data. The discontinuous lines indicate the 95% confidence limits. The lines are drawn only across the range of values existing in the database used to derive the regressions. The ponds in the solar salterns have been grouped according to their salinity (see text). Mean values and standard errors within each group of ponds are shown.

Values of PN and PHP could also be compared to the predictions from the multiple linear regressions derived from the literature data. Since showing graphs similar to those for simple linear regressions (e.g. those in Figs. 2 and 3) would require complex three dimensional graphs, we have plotted the residual of each value (that is, the difference between the actual value and its prediction) against its prediction (Figs. 4 and 5). The 95% confidence limits for the predictions of Y are shown as discontinuous lines. A residual equal to zero (point on the horizontal continuous line) means that the prediction from the multiple regression model is perfect. This implies that the relationships between the variables used in the regression are similar in the tested and in the ‘common’ systems. As residuals move away from the zero line, the predictions become less accurate. Finally, if the residuals are beyond the 95% confidence lines, the prediction is significantly different from the actual values and, thus, it can be concluded that the relationships among the variables involved in the regression model are fundamentally different in the test and in the ‘common’ systems.

4

Comparison of values of PN (in 10⁹ cells l⁻¹, A) and PHP (in μg C l⁻¹ day⁻¹, B) from the salterns with their predictions from the general relationships (multiple linear regressions) derived from literature data. Symbols as in Figs. 2 and 3. In C, the same plot as in A is shown for values from the study of the Eilat salterns in Israel (data from [19,20]).

5

A: Comparison of values of PN (in 10⁹ cells l⁻¹) from different karstic lakes with their predictions from the general relationships (multiple linear regressions) derived from literature data. Numbers indicate each lake: 1, Lake Cisó; 2, Lake Vilar; 3, Lake La Cruz; 4, Lake Tobar and 5, Lake Lagunillo del Tejo. B: Comparison of values of PN (in 10⁹ cells l⁻¹) from two marine areas: the northwestern Mediterranean Sea during June 1995 (filled symbols) and the Southern Ocean during December 1995 (empty triangles) with their predictions from the general relationship.

In Fig. 4A the predictions of PN are shown. First, all values were beyond the range of values used to derive the general relationship (they were all to the right of the discontinuous lines). This reflects that PN in the salterns was much larger than in other marine systems. Second, the residuals for the lowest salinity ponds were within the 95% confidence interval and they were relatively small. This indicates that the ponds with salinities lower than 100‰ showed similar relationships among PN on the one hand, and temperature and PHP on the other. Finally, the residuals for the three groups of ponds with salinities higher than 100‰ were significantly beyond the upper 95% confidence limit. Thus, the actual values of PN were significantly larger than the values predicted by the relationship. This implies that the relationships among the variables involved in the regression (PN as dependent variable and PHP and temperature as independent variables) were different in salterns from those in most marine systems.

Fig. 4B shows the equivalent graph for the prediction of PHP. Again, the values of PHP found in the salterns were larger than the largest values in the database. PHP values showed a tendency to be higher than their predictions (all were above the zero line). The residuals, however, were within the 95% confidence interval and, therefore, the predictions were very close to the actual values in all the saltern ponds. Thus, despite the difference in absolute values, the relationships among the variables involved in the regression (PHP as dependent variable and temperature and Chla as independent variables) were not significantly different from those in most marine systems. Since the simple linear regression between PHP and chlorophyll a significantly underestimated PHP in the salterns (Fig. 2B), it must be concluded that temperature was the factor determining these high PHP values in comparison to ‘common’ systems. This conclusion is in accordance with the relationship between PHP and temperature shown in Fig. 3A.

3.3 Comparison of data from the karstic lakes and marine areas to the general relationships

Fig. 5 shows the same kind of graph as Fig. 4 for the karstic lakes and the two marine areas studied. In the case of the karstic lakes the regression line used was that built with data from freshwater systems only (Table 3). Epilimnetic samples showed residuals that were not significantly different from zero (Fig. 5A). Hypolimnetic samples, on the contrary, had residuals above the 95% confidence limit. Metalimnetic samples were sometimes above and sometimes below the 95% confidence limit, reflecting the intermediate characteristics of these layers between the oxic and anoxic layers of the lakes. The holomixis sample for Lake Cisó was above the limit. This is in accordance with the fact that the whole lake is anoxic during holomixis [22]. The values of abundance for the two marine areas (northwestern Mediterranean and Southern Ocean) were at the lower range of the values in the database (Fig. 5B). However, they were all within the 95% confidence interval of their predictions.

4 Discussion

One way to compare the microbial food webs of different systems is through the use of empirical models [4–11]. These models combine the data from a large number of different studies and look for trends in the relationships between variables. For example, there is a significant relationship between bacterial number and Chla when data from many different systems are combined and a regression is performed. This significant regression reflects a relationship between the two variables and, thus, a characteristic of microbial communities. Although we assume that the relationship should be ‘causal’, the models do not require this to be the case. If data from a particular system, such as the solar salterns, are compared to this general relationship, they may either conform to the regression or show statistically significant differences. In the former case, one may reason that the relationships between the examined variables are the same in the particular system and in the systems used to derive the regression. In the latter case, however, the relationship should be different.

In the case of PN (Fig. 4A) the group of ponds with salinities lower than 100‰ showed values very close to the predictions provided by temperature and PHP. Thus, we conclude that the mechanisms regulating PN in these ponds are the same as those in most marine systems. All the other ponds, however, showed values significantly higher than predicted ones, indicating the existence of differences in the regulation of PN. The PN in any given system is the result of a balance between growth and losses. In most marine systems, and in the lower salinity group of ponds, the final balance is the same. However, in the ponds with salinities higher than 100‰ the balance is displaced towards higher biomass values. This could be the result of either faster growth or lower loss factors. Since growth rates are actually slower in the ponds with higher salinities [15], the reason for this higher abundance must be the reduced losses. We have shown that viral impact is low throughout the salinity gradient [20]. Therefore, viral lysis cannot be the reason for these changes as salinity increases. Bacterivory, on the other hand, is much reduced or non-existent in the ponds with higher salinities [15] and, therefore, it seems to be the main factor responsible for the differences between salterns and other systems.

Even though values of PHP were always higher than their predictions (Fig. 4B), these differences were never significant: sometimes the residuals were very close to the upper 95% confidence limit but always below it. Therefore, we must conclude that the main factors regulating PHP are temperature and resource supply (as represented by Chla), just as in most other environments. Thus, salinity per se does not play a role in limiting PHP. The organisms present at each salinity, therefore, must be well adapted to the ambient salinity.

To see whether these results were also true for salterns in other parts of the world, we screened the literature for other studies of solar salterns. Only the work of Oren [18,19] in the Eilat salterns (Israel) included the necessary variables. Unfortunately, there was no data for Chla and, thus, the comparison could only be carried out with the prediction of PN, but not with that of PHP. The ponds in Oren's study were assigned to the same four groups of salinity and the multiple linear regression was used to obtain predictions of prokaryotic number. As can be seen in Fig. 4C the results were exactly as those found in the two Spanish salterns: the ponds with salinities lower than 100‰ had very low residuals and all the other ponds were above the upper 95% confidence limit. Thus, the pattern found is fairly robust, since it has been found in three different solar salterns.

To confirm the importance of bacterivory in determining the PN in aquatic systems, it had to be shown that other environments without bacterivory also had abundances larger than the predictions. Small karstic lakes in northeastern Spain are ideal systems to test this point. These lakes tend to be sharply stratified [21]. While the epilimnion is oxic and has a microbial food web similar to that of most other lakes, the hypolimnion is generally anoxic with large concentrations of sulfide. This alters many characteristics of the microbial community. Bacteria tend to be more abundant and greater in size [25,35] and the abundance of bacterivores is clearly reduced [25]. As a consequence, the impact of bacterivory on the prokaryotic assemblage is low [25]. If our contention that bacterivory is the most important factor determining PN were true, the predictions from the general relationships should fall within the 95% confidence interval for the oxic layers and outside this interval for the anoxic layers of these lakes. This is exactly what happened (Fig. 5A): all the epilimnetic samples were within the 95% confidence interval and all the hypolimnetic samples were beyond the upper 95% confidence interval.

It could be argued that some other difference between epilimnetic and hypolimnetic samples (such as light regime for example) was the cause of these different results. However, when Lake Cisó mixes in the winter it becomes completely anoxic [22] and the microbial community for the whole water column is very similar to that of the anoxic hypolimnion during stratification [24]. In accordance with our hypothesis the holomixis samples from Lake Cisó were also beyond the upper 95% confidence limit (Fig. 5A).

As a final test of our hypothesis, we checked the possible influence of different degrees of viral lysis on the abundance of prokaryotic plankton. Part of the behavior of our data could have been the result of concerted changes in the importance of viral lysis. This loss factor has been shown to account for up to 50% of total bacterioplankton losses in a coastal environment [36]. However, we have found viral lysis to be very low in the salterns [20]. Thus, we decided to test systems with high and low impact of viral lysis but similar values of bacterivory. The northwestern Mediterranean and some areas of the Southern Ocean were shown to have moderate but similar levels of bacterivory (Marrasé and Vaqué, in preparation, [29]). The former system was sampled during a cruise in June 1995 and viral lysis was found to be undetectable [28]. The water in the Gerlache and Bransfield Straits was sampled during a cruise in December 1995 and viral lysis was found to be more important than bacterivory in all stations where both values could be compared directly. In fact, viral lysis accounted for 50–100% of PHP in different stations [30]. If viral lysis were an important factor determining PN, we should see differences between these two systems in their respective residual plots. As can be seen in Fig. 5B, all the samples were within the 95% confidence interval and, therefore, viral lysis was not an important factor in the determination of the balance between prokaryotic growth and losses that ultimately determines the PN found in any given aquatic system.

In conclusion, the method of comparing results from a given system to an empirically derived general relationship seems to be very effective in revealing whether the functioning of the microbial food web is similar to that of other systems. The analysis of ‘peculiar’ systems such as the solar salterns clearly identified bacterivory as the factor responsible for maintaining the PN values usually found in ‘common’ systems such as the coastal oceans. Conversely, this analysis showed that PHP was regulated by the same factors (temperature and resource supply) as in most ‘common’ aquatic environments. The analysis of karstic lakes and two different marine zones confirmed that bacterivory and not viral lysis is the main factor determining the actual abundance of prokaryotic plankton in a wide range of aquatic systems.

Acknowledgements

This study was supported by DGICYT Grant PB95-0222-C02-01. We thank Mr. Juan Duch from INFOSA SA and Mr. Miguel Cuervo-Arango for permission to work at La Trinitat and Bras del Port salterns, respectively. We thank Paul A. White for access to the data of [7]. J.I.C.-P. was a recipient of a scholarship from the Generalitat de Catalunya.