In order to study the influence of both predation pressure and the autotrophic activity on bacterial community of Lakes Annecy and Bourget, we carried out a fractioning approach and performed incubation in either darkness or ambient light. The originality and strength of this study comes from the fact that such experiments were conducted (i) in two ecosystems with either oligotrophic or mesotrophic status and (ii) at two distinct periods of the year (i.e. early-spring and summer) where microbial planktonic dynamics and composition are likely to display clear differences [8, 24, 25]. Although the use of microcosms may introduce some bias into the development of microbial communities compared with those occurring naturally in the field (due to confinement and handling effects), these experimental tools are still very useful for investigating how processes such as mortality factors induce temporal variation in bacterial dynamics, structure and activity . Incubation time (4 days) coupled with the volume of microcosms (2.5 L) considered in this study have previously been used successfully in other experimental studies [18, 22]. We assumed that our design was thus realistic enough compared to the generation time of microorganisms and aimed to obtain significant changes in bacterial and viral activity . A comparison of virus and flagellate abundances at the onset of the experiments with in situ conditions and among treatments with different viral and flagellate effects was successful. However, the experimental protocol resulted in a reduction of HNF at the start of the experiment and we thus might have underestimated their influence. Clear effects of HNF were observed at the end of the experiment, when flagellate abundance was about twice as high as in situ (Tables 1 and 2).
Grazing effect on viral activity
According to the model of Miki and Yamamura , grazers should reduce the role of the viral loop. As viral production was higher in VF and VFA treatments than in V in most cases, particularly after 48 h incubation, our results instead suggested a stimulation of viral activity in the presence of grazers. Such a stimulation of viral production by the presence of small eukaryotes (grazers) was observed in all experiments for the two lakes. These results corroborate the findings of Jacquet et al.  who observed a clear and positive relationship between flagellate concentration and VIBM (virus-induced bacterial mortality) in Lake Bourget (r = 0.99, P < 0.05) at three different periods of the year (winter, spring and summer), suggesting a synergistic cooperation between grazer and virus activity. Our new results extend the occurrence of this process at other periods of the year and in the oligotrophic Lake Annecy. Similar beneficial effects of protozoan grazing on viruses have been reported in various lacustrine systems with different trophic statuses [21, 23, 26]. This means that the trophic status cannot be 'used' as an environmental factor to change the balance between positive and negative effects of flagellates on viruses , and it is likely that there are probably different processes involved in enhancing viral activities in response to grazing activity .
To the best of our knowledge, Šimek et al.  were first to suggest that protozoan grazing may influence and increase viral lysis. From that time, other studies reported such a synergistic effect in contrast to freshwater systems [21, 26, 27]. Nevertheless, an antagonistic interaction between these two compartments was also noted elsewhere [30, 31]. Mechanisms by which HNF affect viral activity are still unclear and many hypotheses have been proposed to explain such a cooperative interaction (reviewed by Miki and Jacquet ). In brief, grazing activity could stimulate bacterial growth rates, by releasing organic and inorganic nutrients. Higher bacterial growth rates might be associated with enhanced receptor formation on cell surface which may result in a greater chance of phage attachment and in fine higher infection frequencies. Thus, grazer stimulation of viral proliferation could occur through cascading effects from grazer-mediated resource enrichment . We observed, in this study, a strong stimulation of bacterial production in treatments with grazers which may corroborate this assumption in both lakes. A link between infection and host production has been reported previously (summarized in Weinbauer ) and, recently, experimental studies showed that viruses may preferentially lyse active cells [18, 32]. Our results showed that autotrophic activity contributed to this stimulation, mainly in the early summer experiment (for both lakes), while heterotrophic flagellates were always involved in this positive feedback. A shift in the bacterial community structure could also contribute to the synergistic interaction observed in this study. According to Weinbauer et al. , grazing can favour species which are capable of rapid growth and resistant to grazing, which may stimulate viral infection. Our results also seemed to support this hypothesis since both high bacterial production and specific bands were only observed in treatments VF and VFA.
Stimulation of viral production was much more variable between lakes than between seasons and it was clearly higher in Lake Bourget. This suggests that environmental conditions encountered in the mesotrophic system might promote higher viral activity compared to more oligotrophic conditions. This hypothesis agrees with Lymer et al.  or Pradeep and Sime-Ngando  who observed, during a microcosm experiment, an enhancement of both viral abundance and FIC (frequency of infected cells) in P-enriched samples as a result of nutrient stimulation of bacterial growth, which in turn enhanced viral activity. However, it is noteworthy here that although phosphorus concentration was 2-fold higher in Lake Bourget than in Lake Annecy (Table 1), no significant difference was recorded in bacterial production between the two lakes (t test, P > 0.005). Some studies have suggested that nutrient availability may have an important influence on viral life strategies (e.g. [35, 36]). As lysogenic infection is considered the most favourable method of bacterial infection in water characterized by low bacterial abundance and primary production, this may also explain the relatively weak stimulation of viral production observed in Lake Annecy compared to Lake Bourget .
In Lake Annecy, and in contrast to viral production, the effects of flagellate presence on viral abundance seemed to be highly variable between the two periods (LA1 vs. LA2). This variation revealed viral abundance stimulation in early-spring (LA1) and repression in summer (LA2), for both treatments (VFA and VF). This result could suggest a direct grazing of flagellates on viruses during summer. Virivory by flagellates has been previously reported [37, 38] and according to Domaizon et al. , all flagellates do not act similarly because of large differences between taxon-specifc ingestion rates. During our study, heterotrophic flagellates were mainly represented by Oikomonas (45 and 48% during LA1 and LA2, respectively). Also, the grazing impact of flagellates on viruses has always been reported to be relatively low, resulting in < 4% loss [37, 38]. Hence, direct grazing of flagellates on viruses was unlikely to explain the repression of viral abundance in LA2. Other factors should be invoked  and would need further investigation.
Effect of both flagellates and viruses on bacterial activity
Higher bacterial production in both VF and VFA treatments than V suggested that grazers and viruses acted additively to sustain (directly or indirectly) bacterial activity in Lake Annecy and Lake Bourget. Such bacterial growth and activity stimulation in the presence of grazers and viruses has been previously reported in the oligotrophic Sep reservoir, Massif Central, France [23, 26]. However, the control and reduction of bacterial production by the two mortality agents have been observed in other aquatic systems [18, 21, 22]. Such variability in possible responses could be due to the initial bacterial community composition and environmental conditions.
The increase in bacterial production with the presence of both predators (flagellates and viruses) could be explained by the fact that grazing activity and viral lysis are likely to release inorganic and organic nutrients which may stimulate bacterial activity. Obviously, the absence of direct measurements of grazing rates of flagellates on heterotrophic bacteria communities, for instance using fluorescently labelled bacteria (FLB) , prevented us from drawing firm conclusions about the grazing pressure of HNF on bacteria and our results should be considered in light of that. However, it has been suggested that a minimal proportion of 1,000 heterotrophic bacteria for one heterotrophic flagellate is characteristic of microbial food webs in which flagellates preferentially consume bacteria [39, 41, 42]. The value for this ratio was higher than 1,000 in each treatment (VFA vs. VF) and for each experiment (early spring vs. summer). Indeed it varied between 1,632 and 3,866 bacteria per flagellate in Lake Annecy (mean value: 2,795), and between 2,619 and 8,857 in Lake Bourget (mean value: 5899), suggesting that heterotrophic bacteria were abundant enough to support the development of the heterotrophic flagellates that were present.
Seasonal variability in the stimulation of bacterial production seemed to be more important than the trophic status variability, with highest mean values recorded in summer (+33.5% and +37.5% in Lakes Bourget and Annecy, respectively), a period which corresponds to low total phosphorus conditions and high temperature in surface waters (Table 1). Thus, the input of nutrient resources by viral and grazing activities, under such summer conditions, is likely to stimulate the bacterial community much more than under the cold early-spring conditions (temperature = 6-7°C). Moreover, Thomas et al.  observed that the abundance of HDNA (high nucleic acid containing bacteria) is lower in spring than in summer in Lake Bourget (less than 40% of the total bacterial abundance), and this group is considered to be more active in comparison to LDNA (low nucleic acid bacteria) [43, 44]. This could also explain the low stimulation of bacterial production in early spring compared to that in summer.
For most experiments (LA1, LB1 and LB2), the stimulation of bacterial production, at the end of experiments, was much higher in VFA than in the VF treatment (Figure 4) which could be attributed to an increase in substrate availability and regenerated nutrients, resulting from grazing pressure of flagellates on both heterotrophic bacteria and autotrophic communities, in treatment VFA [45, 46]. Nevertheless, the opposite situation was observed in Lake Annecy during the summer period (LA2 experiment) suggesting weaker competition by the bacterial community for nutrient resources in the presence of autotrophs, at this period. As a possible explanation, the abundance of autotrophs (represented mainly by picocyanobacteria and PNF) was indeed 2- to 4-fold higher in summer than in early spring while bacterial abundance was 2-fold lower (Table 1).
Impact of HNF on bacterial community structure
We are aware that the DGGE fingerprinting method presents some bias and only reflects the microorganism populations that are present at relatively high concentrations. For example, while Muyzer et al.  claimed that the reported sensitivity of DGGE is 1% of the template DNA, Casamayor et al.  reported that the number of bands is related to the number of populations that account for more than 0.3-0.4% of the total cell counts. In addition, some other bias such as insufficient or preferential disruption of cells during the DNA extraction step, amplification bias (chimera and heteroduplex formation) and band co-migration in the DGGE gel can occur and consequently over- or underestimate the number of bands. However, such limitations are not specific to DGGE and may also be found in other molecular fingerprinting techniques . Therefore, it must be kept in mind that only major changes in the bacterial community composition could be monitored using DGGE. That is exactly what we observed in this study as all sequenced bands belonged to Actinobacteria and Proteobacteria, known to be the most dominant phyla in lakes [50, 51]. Thus our results have to be interpreted with caution because the structure of some "non-dominant" phyla, non-detectable with the DGGE technique, could have changed according to the treatments performed in this study.
We found that some bands were specific to each treatment suggesting that some bacterial phylotypes were able to develop and thwart the predation pressure. Such specificity has already been reported in other experimental studies [18, 21, 22]. Phylotypes, observed in both VFA and VF treatments, were likely to be resistant to both grazing and infection [21, 22]. Nevertheless, the presence of phylotypes only in VF (not in VFA) might indicate sensitivity to the autotrophic activity as a result of a weak ability to compete for resources. Phylotypes only present when viruses were the exclusive mortality agents would probably not be able to deal with the combined pressure of grazing and viral lysis  or were strongly susceptible to grazing as already suggested by Zhang et al. . Finally, the appearance of bands in both VF and VFA treatments could be due to phylotypes benefiting from the presence of predators, e.g., via the production of DOM or by the removal of competitors.
A few previous experimental studies have shown that viruses can influence the presence or absence of specific bacterial phylotypes but, typically, result in a reduction of the number of detected phylotypes [53, 54]. In contrast, we observed during the summer period an increase in the apparent richness when viruses were the exclusive mortality agents (i.e. the number of detectable bands) giving support to the "killing the winner hypothesis". The stimulation of bacterial diversity in the presence of viruses was also reported in other lacustrine systems by Weinbauer et al.  and other experimental studies performed in coastal marine systems observed the same trend [18, 22]. However, the relative stability of the apparent richness during early spring experiments, in treatment V, highlighted the seasonal variability of virus effects on bacterial diversity. This high variable impact of viruses upon bacterial community structure, already reported by Hewson and Fuhrman , could suggest the influence of stochastic processes.
Since no decrease in the number of bands was observed in either treatment VF or VFA, our result could not support the hypothesis of Miki and Yamamura  according to whom grazing on infected cells "Kills the killer of the winner" and thus reduces bacterial species richness. In some cases, the combined effect of viruses and flagellates on bacterial fingerprint diversity was more consistent than the effect of viruses alone, suggesting that both predators acted additively to sustain apparent richness. According to Zhang et al.  the 'killing the winner' hypothesis is mediated by both predators and not just by one type of predator (viruses). Thus, all predators (viruses and flagellates) could act additively in controlling the winners of the competition for resources and caused an increase in detectable phylotypes. In addition, stimulation of bacterial production and related viral lysis also suggested input of nutrients and substrates from grazing and lysis activities which may decrease the competition pressure within bacterial community, thereby increasing the competitiveness of the minor phylotypes .
The effect of both predators on the bacterial diversity was not apparent in all experiments, suggesting more variability and complexity in the interactions between bacterial diversity, viruses and grazers than hitherto assumed. Diverse patterns between predators and bacterial diversity were reported in other studies [18, 19, 55]. Such variability could be explained by the change in the balance between bacterial production and protistan grazing  or to chaotic behaviour due to competition among predators for the same prey . Overall, previous work performed in both Lakes Annecy and Bourget, indicated that the strong complexity of the combined physico-chemical and biological parameters (with a larger effect of abiotic factors) is mainly responsible for the evolution of the bacterial community structure .