Patterns of resource limitation of bacteria along a trophic gradient in Mediterranean inland waters

The nature of the resource that limits heterotrophic bacteria, i.e. mineral nutrients or carbon (C), has consequences for biogeochemical cycles in aquatic ecosystems. Our aim was to identify the resource [C or phosphorus (P)] that mainly limits bacteria in a set of 31 Mediterranean inland water ecosystems spanning a wide trophic range. We followed an intersystem observational approach with three complementary perspectives, comparing the bacterial demand with the resource supply in terms of both the quantity (demand : supply ratio for C and P) and quality (C : P ratio of demand and supply), and assessing the relative strength of each resource in controlling bacterial production. The trophic gradient revealed a shift in the main limiting resource for bacteria, from C at the oligotrophic end (typically high-mountain, low-productivity lakes) to mainly P at the eutrophic end (typically nonmountain, high-productivity lakes).


Introduction
Heterotrophic prokaryotes (hereafter 'bacteria') are the most abundant and important organisms involved in biogeochemical fluxes on local and global scales. They remineralize ecosystem biomass via general and specialized biochemical pathways and act as a sink (immobilization) and a source (secondary production of biomass) of carbon (C) and nutrients for higher trophic levels (Cho & Azam, 1988;Del Giorgio & Cole, 1998;. The dependence of heterotrophic bacterial growth on organic C and mineral nutrients is a well-established ecological paradigm, with an inadequate supply of either resource driving bacteria toward resource limitation, a frequent situation in the natural environment (Sterner & Elser, 2002).
Bacteria are the main consumers of organic matter in ecosystems ; therefore, the nature of the resource that limits bacterial growth (i.e. C or mineral nutrients) impacts biogeochemical C cycling. This is because limiting elements tend to be utilized for growth and transferred in food chains with high efficiency, while nonlimiting elements (i.e. those present in excess) must be disposed of and may be recycled (Hessen et al., 2004;Hessen & Anderson, 2008). Thus, when bacteria are C limited, labile C is consumed for bacterial growth, while 'semi-labile' (slowly degradable) C accumulates in the surrounding waters (Carlson et al., 2002). However, when bacteria are limited by mineral nutrients, mainly by phosphorus (P) due to the usually low bacterial C : P and N : P ratios (e.g. Vrede et al., 2002), the labile fraction of organic C also tends to accumulate (Søndergaard & Middelboe, 1995;Cotner et al., 1997;Vadstein et al., 2003). This scenario connects with the 'malfunctioning microbial loop' hypothesis (Thingstad et al., 1997) that bacterial C consumption can be restricted due to food web interactions (e.g. algae-bacteria competition for mineral nutrients, grazing control on bacteria), favoring the accumulation of degradable C, which becomes subject to chemical transformation and transport.
Experimental findings indicating that the supply of inorganic nutrients stimulates C utilization (Zweifel et al., 1993), whereas only labile C addition promotes the growth of Climited bacteria and C utilization (Carlson et al., 2002), are compatible with this scenario. Therefore, ecosystems might be expected to have a higher capability to temporarily accumulate dissolved organic C (DOC) when their bacteria are limited by nutrients than when they are C limited (Vadstein et al., 2003;Vrede, 2005).
However, the consumption of organic C and its fate upon assimilation by bacteria is largely governed by the bacterial growth efficiency (BGE), which is the share of the total bacterial assimilation of organic C used for growth (BGE = BP/[BP1BR], where BP is bacterial production and BR bacterial respiration). Although there is a wide variation in BGE within and among aquatic systems, a consistent increase in BGE along gradients of higher productivity or substrate quality (i.e. lower C : nutrient ratio) has been found (Del Giorgio & Cole, 1998;Biddanda et al., 2001;Jansson et al., 2006). Hence, bacterial growth appears to be energetically more costly in oligotrophic ecosystems or with low substrate quality (e.g. high C : nutrient ratio). This is likely due to the need to keep the metabolic machinery active to face poor and fluctuating nutrient environments, where the ability to respond to sudden (pulsed) increases in nutrient levels is an essential property for survival (Del Giorgio & Cole, 1998). As a consequence, at least for ecosystems with a low input of allocthonous C, a larger share of the assimilated C is dismissed by bacterial respiration in oligotrophic waters (low BGE) than in eutrophic waters (high BGE), and greater amounts of primary production (PP) should pass through the bacterioplankton in oligotrophic than in eutrophic waters whether BP constitutes roughly a similar fraction of PP along a trophic gradient (Biddanda et al., 2001). Therefore, the capability of a lake to temporarily accumulate C depends on the BGE and, in turn, on the supply of organic C and mineral nutrients to support the respective bacterial demand.
Bacteria have been reported to be limited by C and P, alone or in combination, over a wide range of ecosystems of varying trophic status (Supporting Information, Table S1). Although published data do not indicate that the trophic status of lakes determines which element will limit bacterial growth (Vrede, 2005), P limitation appears to be more frequent than C limitation in oligotrophic ecosystems, whereas limitation by both C and P appears to be frequent in eutrophic ecosystems. Accordingly, Jansson et al. (2006), based on the opposing trend shown by the BGE and bacterial nutrient use efficiency along an experimental gradient of DOC/P supply in a moderately humic lake, proposed that bacterial P limitation would be more frequent in oligotrophic lakes, C limitation in eutrophic lakes, and that dual C and P limitation would occur in a wide variety of eutrophic-oligotrophic lakes. However, this pattern may not be true for ecosystems from ecoregions as Mediterranean, where inland waters are expected to have a predominantly autotrophic nature. The reason is the limited watershed export as a consequence of Mediterranean ecoregional traits, i.e. scarceness of regional vegetation, wetlands, developed soil, and rainfall (Ortega-Retuerta et al., 2007), which can restrict the input of terrestrial C (Xenopoulos et al., 2003). Taking into consideration that bacteria in autotrophic ecosystems depend on algal C while bacteria in heterotrophic ecosystems are, to a large extent, uncoupled from this dependence, we hypothesized that, in autotrophic Mediterranean inland waters, bacteria will be limited mainly by organic C in oligotrophic ecosystems, where the bacterial demand for C (intensified by low BGE) cannot be satisfied at low PP (as the main supply of organic C); by contrast, bacteria will be limited mainly by P in eutrophic ecosystems due to competition with phytoplankton for this resource, as the high PP provides sufficient amount of organic C to meet the bacterial demand for C; finally, bacteria will be colimited by C and P in ecosystems with an intermediate trophic status. Therefore, our objective was to identify which resource (C or P) limits bacteria in a set of Mediterranean inland aquatic ecosystems by comparing the supply of available C and mineral nutrients with the respective bacterial demand along a trophic gradient. Our approach is intended to provide an initial insight into the patterns of resource limitation of bacteria in ecosystems that have been underrepresented in studies on this issue (Table S1), which might differ from the patterns expected for more heterotrophic ecosystems.

Materials and methods
The study was performed in a set of 31 Mediterranean inland water ecosystems located in the southern half of Spain ( Fig. 1). All ecosystems have a small watershed covered by scarce vegetation (Table 1). The physical, chemical, and biological variables were simultaneously measured for each ecosystem during midsummer in 1999, 2001, and 2005. The mixing depth was calculated as the maximum depth at which the water temperature varies 1 1C relative to the temperature at the surface layer. Temperature profiles were measured using a YSI meter. Water used for measuring chemical (e.g. total and dissolved P [TP and TDP, respectively]) and the biological variables (e.g. seston C : P ratio, PP and BP) consisted of an integrated sample representative of the mixed layer water column. Samples were taken using a sampling vertical tube, and, except for the aliquots used for TP measurements, sieved by a 70 mm net to remove metazooplankton. Water for TDP determinations was filtered previously through disposable 0.45-mm pore size filters (Sartorious). TP and TDP were measured using the acid molybdate technique after digestion with a mixture of potassium persulfate, boric acid, and sodium hydroxide at 120 1C for 30 min (APHA, 1992). For seston C : P ratio measurements, aliquots were filtered through precombusted (1 h at 550 1C) glass fiber filters (1.0 mm pore size; GF/B Whatman) at a low pressure (o 100 mm Hg). The filters were then immediately analyzed for P or were dried (24 h at 60 1C) and kept desiccated until C analysis. Particulate C was determined using a Perkin-Elmer model 2400 (Perkin-Elmer Corporation) elemental analyzer, and particulate P was determined using the acid molybdate technique (APHA, 1992). The C : P ratios were calculated on a molar basis.
The functional variables were measured after short-term incubations, thereby minimizing the methodological shortcomings associated with long-term incubations in some experimental procedures, for example alteration of feedback loops within microbial assemblages or changes in the bacterial community composition (Caron, 2001;Massana et al., 2001). Thus, the total PP, particulate PP (pPP) and excretion of organic C (EOC) by algae were measured as in Carrillo et al. (2002Carrillo et al. ( , 2008. Briefly, sets of four 50-mL quartz flasks (three clear and one dark), filled with the lake water sample and added with 0.37 MBq of NaH 14 CO 3 (specific activity: 310.8 MBq mmol À1 ) per flask, were incubated under in situ conditions under 75-80% of incident PAR irradiance for 4 h symmetrically distributed around noon. PP was measured as the total organic C produced by acidifying a 4-mL subsample in a 20-mL scintillation vial with 100 mL of 1 N HCl and allowing the vial to stand open in a hood for 24 h (no bubbling). pPP (4 1 mm) was segregated from the EOC by algae (o 1 mm; EOC) by filtration through 1 mm pore-size filters of 25 mm diameter (Nuclepore Whatman). A low pressure ( o 100 mm Hg) was applied to minimize cell breakage. The volume filtrated onto each filter was previously adjusted to the algal concentration to avoid clogging of the filter. The total CO 2 in the lake water was calculated from the alkalinity and pH measurements (APHA, 1992). All inorganic 14 C was removed by acidifying overnight. The vials were measured in a scintillation counter equipped with autocalibration (Beckman LS 6000 TA). In all the calculations, data were corrected by dark values. BP was measured following the procedure of Smith & Azam (1992). Briefly, sets of five (three1two blanks) sterile microcentrifuge tubes, filled with 1.5 mL of the lake water sample and added with L-[4,5-3 H]leucine (specific activity: 5.2-5.8 TBq mmol À1 , final concentration 4 20 nM) for lakes nos. 1, 2, 14, and 17, or with [methyl-3 H] thymidine (specific activity: 2.6-3.2 TBq mmol À1 , final concentrations 12-20 nM) for the remaining lakes, were incubated at in situ temperature in the dark for 1 h around noon. After extraction with 5% (final concentration) cold trichloroacetic acid (TCA), the tubes were centrifuged at 16 000 g, rinsed with 5% TCA, and measured in a scintillation counter equipped with autocalibration (Beckman LS 6000 TA). In all the calculations, data were corrected by blank values. Conservative factors were applied to convert incorporated tracers to BP expressed in C terms, i.e. 1.55 kg C mol À1 of leucine (Simon & Azam, 1989;Hoppe et al., 2002), 1 Â 10 18 cells mol À1 of thymidine (Bell, 1993), and 2 Â 10 À14 g C per cell (Lee & Fuhrman, 1987).
Bacterial requirements for C (BRC) were estimated as BP/ BGE, where BGE is calculated from the equation BGE = (0.03710.65BP)/(1.81BP) reported by Del Giorgio & Cole (1998), which provides more realistic estimates compared with the application of a constant BGE value. Bacterial respiration (BR) was estimated from BGE and BP values. Bacterial requirements for P (BRP) were estimated  Thomas et al., 1996). See Table 1 for their identification and characterization.  from the BP (on a molar basis) and a literature-based average bacterial molar C : P ratio of 40 for actively growing bacteria (Chrzanowski et al., 1996;Vrede et al., 2002), which implies a conservative estimate of the highest bacterial demands for P. Algal requirements for P (ARP) were estimated from pPP (on a molar basis) and the seston C : P measured in each ecosystem, as seston was chiefly composed of algae. Despite their best justification, the use of fixed coefficients to estimate these variables may disregard potential variations of the coefficients between locations. Therefore, the absolute values of these variables should be considered with caution. However, the potential error introduced by the estimations would not invalidate the main outcome of the present study, based on the agreement between the trends found from regression slopes with mean values (see Discussion).
Variations along the trophic gradient of the balances provided by the BR : PP and PP : BP ratios were studied by standardizing the variables to daily rates, i.e. extrapolating PP data from the 4-h incubation to the day length (14.5 h) and extrapolating the BR and BP data from 1-h incubations to a full 24-h day, thereby assuming these ratios to be constant throughout the day. Because of uncertainty about the night-time EOC values, the resources supplied and required by bacteria and algae were standardized for 14.5 h of day time (EOCd, BRCd, BRPd, ARPd). The potential error introduced by these standardizations is expected to be small for the purpose of this study, because the sum of the PP values measured for discrete periods during the day time was similar to the parallel PP values measured for an entire day time period in an Iberian lake included in the present study (Medina-Sánchez, 2002). Moreover, the entire water volume was considered rather than single depths, and the diel variation in BP and PP can be considered small in comparison with variations across systems (see Karlsson et al., 2002 and references therein).
C supply to bacteria was measured as EOC, based on (1) the higher preference of bacteria for C of autochthonous rather than of terrestrial origin, even in net heterotrophic lakes in which bacterial growth is considerably subsidized by terrestrial C (Kritzberg et al., 2005(Kritzberg et al., , 2006, and (2) experimental findings of bacterial dependence on organic C freshly released by algae in an Iberian lake included in the present study . The importance of measuring algal exudation as a C source in ecosystems has been underlined elsewhere (e.g. Karl et al., 1998) and is being progressively incorporated into studies on bacterial activity in ecosystems (e.g. Medina-Sánchez et al., 2004Alonso-Sáez et al., 2008). P supply was measured from the TDP pool because it comprises both the inorganic and the organic dissolved fractions of P and can be considered the variable that best quantifies the bulk of P available to microorganisms by direct uptake and/or by the activity of attached or dissolved phosphatases (Currie et al., 1986). As TDP is competitively consumed by both phyto-and bacterioplankton, the share of the P supply to bacteria (TDPb) was estimated as the TDP pool multiplied by the fraction of BRP in relation to osmotrophs' (algae plus bacteria) requirements for P, i.e. TDPb = TDP Â BRPd/(BRPd1ARPd).
In terms of quantity, the ratio between the amount of each resource required and supplied per day was examined along the trophic gradient. Thus, the BRCd/EOCd ratio was considered as a conservative proxy of the degree of C limitation of bacteria, because values 4 1 mean that the amount of organic C freshly released by algae is insufficient to support the bacterial demand for C (assuming a potentially 100% assimilation of this C). The BRPd/TDPb ratio was considered as a proxy of the degree of P limitation of bacteria, because values 4 1 mean that the share of the P supply to bacteria is insufficient to support the bacterial demand for P per day in the absence of an external P input (they also mean that the TDP turnover rate is 4 1 day À1 , which implies that TDP is exhausted after a day time in the absence of an external P input).
In terms of quality of the resource, the variation of the C : P ratio of resource supply and bacterial demand along the trophic gradient was assessed by plotting EOCd : TDPb and BRCd : BRPd ratios against TP and examining the relative position of the regression lines.

Statistical analysis
A homogeneity-of-slopes model was use to test differences in the slopes of the regression lines of C : P ratios of resource supply and bacterial demand against TP (interaction with covariate). Stepwise multiple-regression analyses were carried out to assess the relative strength of each resource (EOC, TDP) in controlling BP for oligomesotrophic or eutrophic ecosystems (i.e. with TP below or above 35 mg P L À1 , respectively; Thomas et al., 1996). Linearity and orthogonality among independent variables were verified by a previous correlation analysis and controlled by specifying 0.4 as the minimum acceptable tolerance (Stat-Soft Inc, 2005). The F values entering the multiple-regression model were established on the basis of the number of independent variables and cases. The normal distribution of residues for all regressions was checked using Shapiro-Wilks' W-tests. Student's t-tests were used to compare the degree of bacterial limitation by C or P (BRCd/EOCd or BRPd/TDP, respectively) between oligomesotrophic and eutrophic ecosystems. The normal distribution for withingroup data was checked using Shapiro-Wilks' W-test. Homoskedasticity was verified by means of Levene's and Cochran-Hartley-Bartlett's tests. For all statistical analyses, data were log-transformed to favor a normal distribution and homoskedasticity and to minimize the influence of outliers. STATISTICA 7.1 software (StatSoft Inc, 2005) was used for the analyses.

Results
A wide trophic gradient was shown by this set of lakes (TP: 3.5-454 mg P L À1 ). The daily BR : PP ratio was negatively related to the TP gradient (Fig. 2a), and the slope and 95% confidence intervals of this regression reached the threshold of log(BR : PP) = 0 (BR equals PP) at the very low end of the x-scale, i.e. under the most oligotrophic conditions, indicating the autotrophic nature of most of the ecosystems under study. Interestingly, the daily PP : BP ratio yielded values ranging from c. 3 to 600 (average of 82.1) and was not related to the TP trophic gradient (Fig. 2b), suggesting the greater contribution of algae than bacteria in energy mobilization (i.e. the initial entry of energy into food web, from exogenous energy and C sources; Jansson et al., 2007) in these Mediterranean ecosystems.
Given these underlying traits of the ecosystems studied, a change in the resource that potentially limits bacteria (mainly from C in oligotrophic lakes to mainly P in the eutrophic lakes) can be inferred by combining (1) the patterns followed by the supply and bacterial demand for C and P along the gradient (resource-quantity perspective), (2) the patterns followed by the resource ratio (i.e. C : P) of supply and bacterial demand along the gradient (resourcequality perspective), and (3) the results of stepwise multiple regression analyses with direct measurements, i.e. BP vs. supply of C (EOC) and of P (TDP), when ecosystems were separated into two groups according to their trophic status (Thomas et al., 1996).

Resource-quantity perspective
The regression of bacterial demand : supply ratios vs. the TP gradient showed opposite patterns for C and P. Thus, the BRCd : EOCd ratio was negatively related to the TP gradient (Fig. 3a), and the slope of this regression (with 95% confidence intervals) crossed the threshold at which bacterial demand equals supply, i.e. log(BRCd : EOCd) = 0 at the lower end of the x-scale (i.e. toward the most oligotrophic conditions). This suggests that bacteria tended to require more organic C (BRCd) than was provided by the total supply of freshly released organic C by algae (EOCd) under more oligotrophic conditions. By contrast, an excess of C supply over bacterial C demand tended to increase under more eutrophic conditions, with significantly lower BRCd : EOCd ratios in the eutrophic (TP 4 35 mg P L À1 ) than in the oligomesotrophic (TP o 35 mg P L À1 ) ecosystems (Fig. 4a). In relation to P, the slope and 95% confidence intervals of the BRPd : TDP ratio vs. the TP regression crossed the threshold of log(BRPd : TDP) = 0 at the middle of the x-scale, indicating that the P supply tended, in the absence of an external P input, to be depleted by bacterial demand (BRPd) toward eutrophic conditions (Fig. 3a). By contrast, an excess of P supply over bacterial P demand tended to increase under more oligotrophic conditions, with significantly lower BRPd : TDP ratios in the oligomesotrophic (TP o 35 mg P L À1 ) than in the eutrophic (TP 4 35 mg P L À1 ) ecosystems (Fig. 4b).

Resource-quality perspective
The BRCd : BRPd ratio (demand C : P) and the EOCd : TDPb ratio (supply C : P) were significantly related to the TP gradient in opposed directions (Fig. 3b). Thus, the slope of BRCd : BRPd vs. TP regression was negative and significantly different from the slope of EOCd : TDPb vs. TP regression, which was positive (slope model, F 1, 58 = 29.41, P o 0.0001). Because of the contrasting behavior of these regressions, the C : P ratio of bacterial demand (BRCd : BRPd) tended to be increasingly higher than the C : P ratio of resource-supply (EOCd : TDPb) toward the oligotrophic end of the gradient. The two slopes and 95% confidence intervals crossed at the middle of the trophic gradient, and the C : P ratio of bacterial demand (BRCd : BRPd) tended to be increasingly lower than the C : P ratio of resource-supply (EOCd : TDPb) toward the eutrophic end of the gradient.
Stepwise multiple regressions with direct measurements C (EOC) was the only resource that explained the BP variance in the oligomesotrophic ecosystems (TP o 35 mg P L À1 ). In the eutrophic ecosystems (TP 4 35 mg P L À1 ), both C (EOC) and P (PDT) explained the BP variance, although PDT explained most (55%) of this variance, whereas EOC contributed only with an additional 15% (Table 2).

Discussion
The main finding of the present study was that the trophic gradient of a set of Mediterranean Iberian inland waters revealed a shift in the resource limitation of bacteria, from limitation mainly by C in the oligotrophic ecosystems to limitation mainly by P in the eutrophic ecosystems. This outcome was obtained by adopting an intersystem observational approach widely used in the literature (e.g. Del Giorgio & Cole, 1998;Biddanda et al., 2001;Karlsson et al., 2002;Kritzberg et al., 2005) and, despite the potential error introduced in the estimation of variables related to bacterial demands (see Materials and methods), is supported by the agreement between the trends found from three complementary perspectives, i.e. comparing the resource supply with the bacterial demand in terms of both the quantity and the quality, and assessing the relative strength of each resource in controlling BP by means of stepwise multiple-  regression analyses. This outcome supports our hypothesized patterns of bacterial resource limitation for autotrophic ecosystems along a trophic gradient.
The patterns of resource limitation of BP reported here may stem from the autotrophic nature of most study ecosystems, their geographic regional pattern, and the balance between commensalistic (for algal C) and competitive (for P) algae-bacteria relationship. The autotrophic nature of most Mediterranean ecosystems studied here is supported by their high PP : BP ratios (3-600, averaging 82.1) and their BR : PP ratios below 1 throughout the gradient. The PP : BP values were even higher than those reported by Karlsson et al. (2002), and were not related to the trophic gradient, suggesting that energy mobilization was invariably dominated by autotrophic processes along the trophic gradient, in contrast to other ecosystems such as temperate forest lakes, where allocthonous dissolved organic matter (DOM) is a major source of energy mobilization via bacteria (Karlsson et al., 2002). The predominantly autotrophic nature of ecosystems studied here intensified toward the eutrophic end of the gradient, because the estimated bacterial respiration tended to be progressively lower than the PP, in agreement with the general trends reported elsewhere (Del Giorgio et al., 1997;del Giorgio & Cole, 1998;Biddanda et al., 2001). This general autotrophic nature may be conditioned by some traits linked to Mediterranean region (e.g. scarceness of terrestrial vegetation, wetlands, developed soil and rainfall, and intense solar radiation) that limit watershed export, restrict terrestrial inputs of DOM to short rainy periods, promote long water residence times, and subject DOM to intense photoalteration (Ortega-Retuerta et al., 2007 and references therein). Therefore, this environment may favor autotrophs rather than bacteria, as indicated by the generally high PP : BP ratios found, and the fact that a considerable share of the DOC in these ecosystems is of autochthonous origin, which is partially supported by the low chromophoric content of their organic matter (Ortega-Retuerta et al., 2007). Interestingly, most of the oligotrophic ecosystems are clustered in a mountainous region within or near the Sierra Nevada range and have a low DOC content (Reche et al., 2005), probably because of the low PP and the small unforested siliceous bedrock watershed. The high transparency of the water and the high fluxes of UV radiation may decrease DOC bioavailability to bacteria even further by photopolymerization (Obernosterer et al., 2001;Tranvik & Bertilsson, 2001) or by indirect effects through the interaction with reactive oxygen species (Ortega-Retuerta et al., 2007). Because bacteria are superior competitors than algae for P under Plimiting conditions (Cotner & Wetzel, 1992, Thingstad et al., 1993, bacteria may become more restricted by the availability of C than of P in these oligotrophic waters (where the BGE tend to be low) and would heavily depend on freshly released algal C, as reported previously in a Sierra Nevada lake Medina-Sánchez et al., 2002). By contrast, most of the eutrophic ecosystems in the present series are on low lands, surrounded by watersheds devoted to agriculture or pasture, and show higher PP and EOCd values ( Table 1). The latter may sustain the higher DOC content found in these inland waters (from 4 3 to 4 45 mg C L À1 ) than in the Iberian high-mountain lakes (from o 0.5 to o 1.5 mg C L À1 ), leading to a positive correlation between the trophic gradient (e.g. TP) and the DOC concentration (Fig. 5, see data in Pulido-Villena & Reche, 2003;Reche et al., 2005;Ortega-Retuerta et al., 2007). Because of the increased bacterial demand for P to support an enhanced BP and BGE, and the superiority of algae over bacteria as competitors for P under P-rich conditions (Cotner & Wetzel, 1992, Thingstad et al., 1993, consistent with a higher increase of autotrophic than heterotrophic biomass after experimental nutrient enrichment (e.g. Duarte et al., 2000), bacteria may become more restricted by P than by C availability in these eutrophic ecosystems.
Although the patterns of resource limitation found in these Mediterranean lakes may first appear to be somewhat counterintuitive (e.g. P as the main limiting resource of bacteria in P-rich ecosystems), Liebig's law of minimum in relation to the needs of bacteria is the basis of the explanation of the patterns discussed above. Thus, for bacteria, the limiting resource, i.e. the 'essential material available in amounts most closely approaching the critical minimum needed will tend to be the limiting one' (Odum, 1959), will shift from C to P along the trophic gradient according to the explanation discussed above. However, the straightforward applicability of Liebig's law at the community level was questioned recently by Danger et al. (2008). Based on a model derived from Tilman (1982) and an experimental test of its predictions, they demonstrated that bacterial communities were colimited over wider ranges of supplied nutrient ratios in comparison with a single bacterial species. Nevertheless, they also showed that, in accordance with the predictions of the model, the bacterial community becomes single-resource limited (e.g. by C) at end ranges of the supplied nutrient ratio gradient. Thus, in the present study, the supply of C and P in relation to the bacterial demand generates a scenario, in terms of the resource quantity and quality, that is compatible with bacterial community monolimitation by C toward the oligotrophic end of the TP gradient, and with biochemical and community colimitation (sensu Arrigo, 2005) by C and mainly P toward the eutrophic end of the gradient. Hence, our observations would be in line with the view that the transition from one single limiting nutrient to another may be smoother than originally suggested by Liebig, and may pass through a transition phase in which the organism or the community is colimited (Schade et al., 2005;Danger et al., 2008). This perspective adds to the view of the two possible steady states of the microbial food web (sensu Thingstad et al., 1997Thingstad et al., , 2008. This theoretical approach would be in agreement with our findings for the middle of the trophic gradient that may correspond to a transition phase before a strict P limitation under extreme eutrophic conditions. In summary, the patterns of resource limitation of BP reported here stem from the predominant autotrophic nature of the study lakes, which conditions the balance between bacterial commensalism and competition with algae. In contrast to the scenario of lakes dominated by high inputs of terrestrial organic C, the fact that autochthonous C (algal exudates) controlled BP in the study ecosystems precludes that bacteria outcompete algae, as indicated by the generally high PP : BP ratios found. The shift between C and P as the main limiting resource for bacteria along the trophic gradient is the result of a balance between bacterial demand and resource supply, modulated by interaction with the algae, and governed by Liebig's law of minimum. Therefore, these findings may be relevant to lakes receiving low inputs of allocthonous DOC (e.g. many lakes placed outside the boreal region).

Ecological implications
Based on our results and current knowledge on the potentiality of aquatic ecosystems to accumulate C according to the resource limitation of their bacteria (Thingstad et al., 1997(Thingstad et al., , 2008Carlson et al., 2002;Cole et al., 2007), a Mediterranean eutrophic ecosystem would tend to accumulate more C compared with an oligotrophic ecosystem of similar size, as long as patterns of resource limitation of bacteria persist over the long term. This proposition agrees with the recent recognition that inland freshwater ecosystems play a greater role than previously thought in regional and global C balance, especially small aquatic ecosystems linked to agricultural and intensive land use, in which C storage in sediments tends to be higher (Cole et al., 2007;Tranvik et al., 2009;Williamson et al., 2009, and references therein). However, the role of bacteria in C accumulation in these Mediterranean ecosystems may be less than first thought, because their autotrophic nature and the high PP : BP ratio found suggest a rather modest contribution of bacteria to the biogeochemistry in these ecosystems.