Enhanced viral production and virus-mediated mortality of bacterioplankton in a natural iron-fertilized bloom event above the Kerguelen Plateau

Above the Kerguelen Plateau in the Southern Ocean natural iron fertilization sustains a large phytoplankton bloom over 3 months during austral summer. During the KEOPS1 project (KErguelen Ocean and Plateau compared Study1) we sampled this phytoplankton bloom during its declining phase along with the surrounding highnutrient–low-chlorophyll (HNLC) waters to study the effect of natural iron fertilization on the role of viruses in the microbial food web. Bacterial and viral abundances were 1.7 and 2.1 times, respectively, higher within the bloom than in HNLC waters. Viral production and virus-mediated mortality of bacterioplankton were 4.1 and 4.9 times, respectively, higher in the bloom, while the fraction of infected cells (FIC) and the fraction of lysogenic cells (FLC) showed no significant differences between environments. The present study suggests viruses to be more important for bacterial mortality within the bloom and dominate over grazing of heterotrophic nanoflagellates (HNFs) during the late bloom phase. As a consequence, at least at a late bloom stage, viral lysis shunts part of the photosynthetically fixed carbon in iron-fertilized regions into the dissolved organic matter (DOM) pool with potentially less particulate organic carbon transferred to larger members of the food web or exported.


Introduction
A quarter of a century ago the importance of viruses as the most abundant biological entity in the oceans (Bergh et al., 1989) and their role in the material and energy cycles were recognized (Proctor and Fuhrman, 1990;Suttle et al., 1990). Shortly afterwards, Smith et al. (1992) conducted the first study on viral distribution and their relationship to bacteria in the Southern Ocean. Since then, studies on viral abundance and production or infectivity in the cold, high-latitude marine environments remained limited or have only recently been accumulating (Bird et al., 1993;Brussaard et al., 2008b;Evans and Brussaard, 2012;Evans et al., 2009;Guixa-Boixereu et al., 2002;Higgins et al., 2009;Manganelli et al., 2009;Marchant et al., 2000;Suttle, 2008, 2013;Smith et al., 1992;Steward et al., 1996;Strzepek et al., 2005;Weinbauer et al., 2009). These observations demonstrate that viruses are ecologically as important in these cold environments as in the world's other oceans.
Viral lysis of cells converts particulate organic matter into dissolved and colloidal organic matter, reduces the carbon flow to higher trophic levels and increases the residence time of carbon and mineral nutrients in the euphotic zone (Fuhrman, 1999). By this process, called the "viral shunt" (Wilhelm and Suttle, 1999), heterotrophic bacteria are Published by Copernicus Publications on behalf of the European Geosciences Union.
supplied with substrate, which finally increases respiration (Bonilla-Findji et al., 2008;Middelboe and Lyck, 2002). This could reduce the efficiency of the biological carbon pump, i.e. the process which transforms inorganic to organic carbon, part of which is then transferred to the deep ocean (Suttle, 2007). The relative significance of viral lysis and protistan grazing can strongly vary on temporal and spatial scales (Boras et al., 2009;Fuhrman and Noble, 1995). This has also been shown for cold marine environments (Boras et al., 2010;Guixa-Boixereu et al., 2002;Steward et al., 1996;Wells and Deming, 2006).
In about one-third of the World Ocean -including the subarctic northeast Pacific, the equatorial Pacific and the Southern Ocean, phytoplankton growth is limited by available iron, resulting in excess dissolved inorganic phosphorus and nitrogen (Martin and Fitzwater, 1988). In these highnutrient-low-chlorophyll (HNLC) regions, bacterioplankton are thought to be the key player of the "microbial ferrous wheel" (Kirchman, 1996), i.e. the uptake and remineralization of iron. Bacterioplankton contains more than twice the iron per carbon unit than eukaryotic phytoplankton, and they can thereby store up to 50 % of the biogenic iron in the HNLC ocean (Tortell et al., 1996). Viral activity has a potential impact on nutrient regeneration. Typically, nutrients released as a result of viral lysis are thought to be organically complexed, which may facilitate their use by marine plankton (Poorvin et al., 2004;Rue and Bruland, 1997). Iron released by viral lysis can account for more than 10 % of ambient Fe concentrations (Gobler et al., 1997) and thus potentially relieve its limitation in depleted environments. Furthermore, marine viruses may serve as nuclei for iron adsorption and precipitation, and they thus represent a significant reservoir of iron in seawater (Daughney et al., 2004). Despite their key role, viruses are hardly included in iron enrichment studies. These experiments were originally stimulated by the "iron hypothesis" (Martin, 1990), which assigns iron a paramount role in controlling ocean productivity and consequently atmospheric carbon dioxide concentrations. Only 2 out of 13 iron fertilization experiments so far performed (Secretariat of the Convention on Biological Diversity, 2009) report on viral abundance and activity (Higgins et al., 2009;Weinbauer et al., 2009). Both studies -from the subarctic and Southern Ocean, respectively -found that viral production was significantly enhanced after iron fertilization.
Above the Kerguelen Plateau in the Southern Ocean, the largest HNLC ocean, a large phytoplankton bloom occurs annually during austral summer. The continuous supply of Fe and major nutrients from below has been shown to sustain this massive bloom (Blain et al., 2007). The region off Kerguelen provides the opportunity to study natural iron fertilization in the Southern Ocean and to compare it to blooms induced by mesoscale Fe additions. Within the KEOPS1 project (KErguelen Ocean andPlateau compared Study1, 2005-2007), we sampled the phytoplankton bloom above the Kerguelen Plateau during its late successional stage ( ∼ 3rd month) along with the surrounding HNLC waters. The aim of the present study was to assess the role of viruses within the microbial food web affected by natural Fe fertilization and to elucidate the possible implications for the final destiny of organic carbon. For this purpose, we measured viral production, the fraction of infected cells (FIC), lysogeny and estimated bacterial mortality through viral lysis in the bloom and surrounding HNLC waters.

Description of the study site
Sampling was performed in the Indian sector of the Southern Ocean above the Kerguelen Plateau (49-53 • S, 72-78 • E) in austral summer (18 January-13 February 2005) onboard the R/V Marion Dufresne in the framework of the project KEOPS (Blain et al., 2008). We sampled a large phytoplankton bloom dominated by diatoms from its peak to its decline (Mosseri et al., 2008) (Fig. 1). Satellite images dated the onset of this bloom more than 2 months before its first visit (Blain et al., 2007). Hydrographic conditions are described in detail in Park et al. (2008). Dissolved Fe concentrations in the surface mixed layer were low and similar on and off the plateau (0.09±0.03 nM) but increased with depth above the plateau, reaching a mean maximum of 0.35 nM at 500 m. This strong vertical gradient in combination with physical features such as internal waves and tidal activity  Uitz et al. (2009) sustained the phytoplankton bloom above the plateau (Blain et al., 2007).

Sampling strategy
Water was collected using General Oceanics 12 L Niskin bottles mounted on a rosette with a Sea Bird SBE19 plus CTD sensor for salinity, temperature and oxygen from two to three depths (within and below the surface mixed layer) at the following stations to cover the centre and borders of each of three transects (A, B, C): A3, A11, B1, B5, B11, C3 and C11 (Fig. 1, Table 1). The stations A3 and C11 were considered as the most contrasting stations and sampled repeatedly. The first sampling of station A3 (A3-1) was done during the peak of the bloom, and about 2 weeks later station A3 was re-sampled at a fourth visit (A3-4) during the decline of the bloom. Station B5 was situated within a new phytoplankton bloom above the Kerguelen Plateau . Station A11 was located in iron-fertilized waters. The annually occurring spring bloom developed prior to our visit, explaining the low concentrations of Chl a at this site (Table 1). Station C11 was in HNLC waters off the Kerguelen Plateau and was sampled twice (Table 1). Stations B11 and C3 were in different environments with relatively low Chl a contents and will also be considered as representative of HNLC conditions in this study.

Enumeration of viruses and prokaryotes
Subsamples (2 mL) were fixed with glutaraldehyde (0.5 % final concentration), incubated for 15-30 min at 4 • C, subsequently frozen in liquid nitrogen and stored at −80 • C. Within a few days samples were thawed and viral particles and bacteria were stained with SYBR Green I (molecular probes) and quantified using a FACScalibur (Becton and Dickinson) flow cytometer after dilution with TE buffer (10 mM Tris, 1 mM EDTA, ph = 8). For viruses an optimized protocol by Brussaard (2004) was followed. Viruses and prokaryotes were determined in plots of 90 • light scatter (SSC) and green DNA fluorescence. Differences in the green fluorescence and side scatter signature in the cytometric plot allowed to separate prokaryotes with low nucleic acid content (LNA) from prokaryotes with high nucleic acid content (HNA) as previously described by Gasol et al. (1999). Similarly, different size classes of viruses were distinguished on the basis of green fluorescence. Abundances were calculated by using the flow rate measurements. Flow-cytometric assessment of viral abundance may encompass particles other than viruses such as bacterial vesicles (Biller et al., 2014). However, since bacterial and viral parameters were related www.biogeosciences.net/11/6841/2014/ Biogeosciences, 11, 6841-6853, 2014 significantly (Table 4), a potential overestimation of viral abundances probably did not bias the conclusions of the study.
To convert bacterial abundance (BA) to biomass, we used a conversion factor of 12.4 fg C cell −1 for oceanic prokaryotes (Fukuda et al., 1998).

Bacterial production
The incorporation of 3 H leucine into protein  was used to estimate the production of heterotrophic bacteria (BP). At each depth, 1.5 mL duplicate samples and a trichloroacetic acid (TCA)-killed control were incubated with a mixture of L−[4, 5-3H] leucine (Amersham, 160 Ci mmol −1 ) and nonradioactive leucine added at final concentrations of 7 and 13 nM for the upper 100 m, and 13 and 7 nM for the 100-200 m depth layer. Samples were incubated in the dark at the ambient temperature of the depth where samples were collected. The incubation time (2-3 h) was tested to satisfy linear incorporation with time. We checked by concentration kinetics (2.5, 5, 10, 20 and 40 nM) at three stations inside and outside the bloom at 5 and 175 m depths that there was no isotopic dilution. The theoretical conversion factor of 1.55 kg of C mol −1 was used to convert leucine incorporation rates to prokaryotic carbon production (Kirchman, 1993).

Viral production, the fraction of infected cells and the fraction of lysogenic cells
Lytic viral production (VP 1 ), the FIC, induced viral production (VP i ) and the fraction of lysogenic cells (FLC) were estimated using the virus reduction approach (VRA; Weinbauer et al., 2010;Wilhelm et al., 2002). The rationale behind VRA is to reduce viral abundance in order to stop new viral infection. Thus, the viruses produced originate from already-infected cells. Briefly, bacteria from 200 mL raw seawater were concentrated using a tangential flow system with a peristaltic pump (Watson-Marlow 323) equipped with a 0.2 µm cartridge (VIVAFLOW 50). To obtain virus-free seawater, the 0.2 µm pore-size ultrafiltrate was passed through a 100kDalton cartridge (VIVAFLOW 50). The bacterial concentrates were brought up to the original volume with virusfree seawater and incubated in duplicate 50 mL Falcon tubes in the dark at ±2 • C in situ temperature for 24 h. Lysogeny was estimated by adding mitomycin C (SigmaChemical Co., Cat. No. M0503, final concentration 1 µg mL −1 ) to duplicate 50 mL Falcon tubes in order to induce the lytic cycle in lysogens; untreated duplicate samples served as controls (Paul and Weinbauer, 2010). Subsamples (2 mL) for viral and bacterial abundance from each incubation were taken immediately (t 0 samples) and every 3-4 h, fixed with glutaraldehyde (0.5 % final concentration), incubated for 15-30 minutes at 4 • C, subsequently frozen in liquid nitrogen and stored at −80 • C until enumeration using a flow cytometer as described above. VP l was calculated as where V 1 and V 2 are viral abundances and t 1 and t 2 the elapsed time. Dividing the number of produced phages by an estimated burst size (BS, i.e. the number of phages released during the lysis of a single host) yields the number of lysed cells and thus gives an estimation of FIC (Weinbauer et al., 2002). FIC was calculated as where BA is the bacterial abundance at t 0 . The difference in phage production between the lysogeny treatment and the control is VP i , calculated as where V MC and V C are the maximum difference in viral abundance at corresponding time points in control and mitomycin C treatments, respectively. Dividing the number of induced phages by BS and the bacterial abundance at t 0 (BA) gives an estimate of the FLC: Calculations were performed for each replicate separately.

Contact rates
The rates of contact (R, number mL −1 d −1 ) between viruses and bacteria were calculated by using the following equations (Murray and Jackson, 1992).
where Sh is the Sherwood number (1.06 for a bacterial community with 10 % motile cells; Wilhelm et al., 1998), d is the diameter of the target; VA and BA are the abundances of viruses and bacteria, respectively; and D v is the diffusivity of viruses.
where k is the Boltzmann constant (1.38 × 10 −23 J K −1 ), T is the in situ temperature (∼ 275 K), µ is the viscosity of water (Pascal s −1 ) and d v is the diameter of the viral capsid (∼ 60 nm). The contact rates were divided by in situ bacterial abundance to estimate the number of contacts per cell on a daily basis.

Statistics
Normal distribution of data was checked using the Shapiro-Wilk W test. Differences between different trophic situations were analysed by the Kruskal-Wallis test for nonparametric data and by one-way ANOVA (analysis of variance) for normally distributed data. Spearman rank correlation for non-normally distributed data was applied. Significance was considered for P < 0.05.

Bacterial and viral abundances
From surface water down to 200 m, BA was on average 1.7 fold higher within the Fe-fertilized (3.9 × 10 5 mL −1 ) than in HNLC waters (2.4 × 10 5 mL −1 , Kruskal-Wallis test, P < 0.0001, Table 2; Fig. 2). Similarly, viral abundance (VA) averaged 9.9 × 10 6 mL −1 at the Fe-fertilized stations and was twice as high as in the HNLC environments (4.7 × 10 6 particles mL −1 , Kruskal-Wallis test, P < 0.05, Table 2). VA ranged from 3.1 to 14.2 × 10 6 mL −1 , with the highest values found at the main bloom station A3 and the lowest value detected in the deep layer of the HNLC station B11. Viruses were homogeneously distributed with depth at the HNLC stations. The virus-to-bacteria ratio (VBR) ranged from 11 to 34 and averaged 21 without significant differences between stations or trophic situations.

Contact rates
Contact rates were significantly higher at the Fe-fertilized stations than in HNLC waters (Kruskal-Wallis test, P < 0.05, Table 2). At the Fe-fertilized stations, on average 29.4 ± 11.1 viruses contacted a bacterial cell per day, while in the HNLC waters contact rates were 14.2 ± 4.4 viruses cell −1 d −1 , with the highest values at the bloom station A3 and the lowest at the HNLC station B11 in accordance to the highest and lowest viral abundances, respectively (see Fig. 2).

Bacterial production, viral production, fraction of infected cells and lysogeny
Bacterial production ranged from 0.1 to 0.7 µgC L −1 d −1 at the HNLC stations and from 0.1 to 2.5 µgC L −1 d −1 at the Fefertilized stations ( Table 2). The highest values were found throughout the depth profile of the main bloom station A3-1 and the lowest values were measured between 150 and 200 m at the HNLC stations. Despite the wide range of values, BP was on average 4 times higher at the Fe-fertilized stations than at the HNLC stations (Kruskal-Wallis test, P < 0.0001, Table 2). Initial virus abundance in the VRA was 45 ± 25 % (11-88 %) of in situ abundance. The recovery efficiency for bacteria in the VRA was on average 26 ± 18 % (5-83 %).
Although FIC values at the Fe-fertilized stations almost doubled those in HNLC waters, this difference between environments was not significant (Kruskal-Wallis test, Table 2). Average values for duplicate assays ranged from 4 to 47 % (average: 22 %) in fertilized waters and from 3 to 23 % (average: 12 %) in HNLC waters. Lysogenic infection of bacterioplankton could be detected only in 7 out of 15 lysogenic phage induction essays and ranged from 1 to 31 % in fertilized waters and from 1 to 4 % in the HNLC environment.

Relation between the different parameters
Spearman rank correlation coefficients ρ for chlorophyll a, viral and bacterial parameters from HNLC and bloom stations are shown in Table 4. BA and BP correlated positively throughout trophic situations, but only in HNLC waters did BA and BP increase with Chl a. In the fertilized waters VA correlated positively with BP, while in HNLC waters VP increased with BA. Only in these waters did VP l correlate sig-nificantly and positively with the fraction of infected cells (Table 4).

Discussion
Viruses were the dominant mortality factor of bacteria during the late stage of a phytoplankton bloom induced by natural iron fertilization in the Southern Ocean (second visit to A3) but accounted for a small part of bacterial mortality within a new bloom (station B5, Table 3). Additionally, observations from the early bloom phase showed that heterotrophic nanoflagellates (HNFs) dominated the loss of BP, and viruses accounted for only 10 % of bacterial mortality (Christaki et al., 2014). These seasonal dynamics point to a switch from an efficient functioning of the microbial food web during the onset of the phytoplankton bloom to a microbial food web where organic carbon is mainly processed by the viral shunt. The increase in virus-mediated release of dissolved organic carbon over time has important consequences for the fate of part of the photosynthetically fixed carbon and reduces its transfer to higher trophic levels and export.

Comparison of viral data within high-latitude marine environments
Viral production rates in the present study match well the data obtained from the Australian sector of the Southern Ocean (Evans et al., 2009) and are within the range of VP rates from an iron-induced bloom in the subarctic Pacific (Higgins et al., 2009). However, our VP rates are high when compared to data from an artificial iron-fertilization experiment in the Southern Ocean  or those from other high-latitude marine environments, i.e. the Arctic Sea (Steward et al., 1996;Boras et al., 2010) (Table 5). Differences between studies could be due to spatiotemporal variations of VP; however, it is also conceivable that differences between methods Winget et al., 2005) have contributed to the variability of reported VP data.
In the present study, the burst size averaged 128 viruses per bacterial cell throughout the experiments. This value is high compared to two studies from the Southern Ocean where measured BS was about 40 viruses per bacterial cell (Strzepek et al., 2005;Weinbauer et al., 2009) and to a study in early spring above and off the Kerguelen Plateau where BS evaluated with TEM observations varied from 6 to 88 viruses per bacterial cell (mean ± SD, 22 ± 15; Christaki et al., 2014). These different BS could be inherent to the study regions or due to the used method, i.e. estimating BS by an increase in VA and a decrease of BA in the VRA (Wells and Deming, 2006), which can result in increases of BP and thus potentially increase VP Weinbauer et al., 2009;Winget et al., 2005). However, Steward et al. (1996) found BS as high as 270 for areas of high productivity in www.biogeosciences.net/11/6841/2014/ Biogeosciences, 11, 6841-6853, 2014 the Chukchi Sea, and studies from the North Sea have reported 100 phages produced per lysed bacterium (Bratbak et al., 1992).

Viruses in HNLC waters versus a phytoplankton bloom induced by natural iron fertilization
Viral distribution during the late stage of the phytoplankton bloom above the Kerguelen Plateau as well as its relation to the bacterial hosts (e.g. VBR) and phytoplankton biomass is extensively reported, discussed and compared to existing data from similar regions in Brussard et al. (2008b). During the late bloom stage, average viral abundance at the bloom stations was twice as high as in HNLC waters (Brussaard et al., 2008b), while during the early bloom viral abundance remained unaffected (Christaki et al., 2014). Data from mesoscale Fe fertilization experiments showed that viral abundance inside the fertilized patch was higher  or not substantially different from outside (Higgins et al., 2009). The authors of the latter study explained the lack of differences between inside and outside the fertilized patch with the time-lag of the microbial response to the induced bloom, since viral abundance and production were only increasing at the end of their observations (day 12 after iron fertilization). This observation is in line with the increase in viral abundance and activity on a seasonal scale in the Kerguelen bloom (Christaki et al., 2014). The present study observed a mature bloom and could thus track a period with a more pronounced microbial response. The 4-times-higher viral production at the naturally Fe-fertilized study sites compares well to the 3-fold increase in phage production after an induced bloom through iron addition . Interestingly, Christaki et al. (2014) reported higher VP rates already at the early bloom stages. Thus, there is a trend of higher viral production in the iron-fertilized bloom compared to the surrounding HNLC waters consistent with existing data on iron fertilization . Complementary, within the bloom, HNFs did not seem to control enhanced bacterial production rates, while in HNLC waters HNFs consumed 95 % of bacterial production . These studies suggest that there is a switch towards viral lysis dominating in the bloom situations. More generally, this is in accordance with previous studies across environments which showed viral influence to be more important in more eutrophic waters (Weinbauer et al., 1993;Steward et al., 1996), particulary in the cold environments such as the Arctic (Steward et al., 1996) or the Southern Ocean, where Guixa-Boixereu et al. (2002) found that viruses were responsible for the entire bacterial mortality. The high virally induced mortality in the bloom could also be a reason for low biomass accumulation, despite the high BP. We calculated carbon release rates through viral lysis in two ways: first, based on VP, and, second, based on VMM related to FIC by a model of Binder (1999) (Table 6). Independent of the absolute values, which were 1 order of magnitude higher in the former than in the latter way (Table 6), C release through viral lysis was 5-8 times higher in the Fe-fertilized than in the surrounding HNLC waters.
The percentages of lysogens (i.e. bacteria containing temperate viruses) were more variable in the fertilized (0-31 %) than in the HNLC waters (0-4 %) but not significantly different between environments. Consistent with our study, Weinbauer et al. (2009) did not find differences inside and outside the iron-enriched patch during a fertilization experiment in the Southern Ocean. The proportion of the lysogenized bacterial population can vary extensively, for example, from 1.5 to 11.4 % in the Gulf of Mexico (Weinbauer and Suttle, 1996), from 4 to 38 % in the Canadian Arctic Shelf (Payet and Suttle, 2013) and from 0 to 100 % in Tampa Bay, Florida (Williamson et al., 2002). According to conceptual models, lysogeny should occur preferentially in environments where the contact rate between infective phages and hosts is too low to sustain the lytic lifestyle . Empirically, this has been proven by Weinbauer et al. (2003), who studied the frequency of lysogenic cells in contrasting Biogeosciences, 11, 6841-6853, 2014 www.biogeosciences.net/11/6841/2014/ Table 6. C and Fe release rates (L −1 d −1 ) through viral lysis calculated from VP (12.4 fg C cell −1 , Fukuda et al., 1998) and from FIC following the model by Binder (1999) using bacterial iron quota of 7.5 µMol Fe mol C −1 (Tortell et al., 1996). Averages are given in parenthesis.
Release based on VP Release based on FIC Fertilized stations 0.60-7.97 (4.18 ± 3.15) 0.08-1.06 (0.56 ± 0.42) 0.03-1.58 (0.42 ± 0.49) 0.003-0.21 (0.06 ± 0.07) HNLC stations 0.28-1.60 (0.86 ± 0.43) 0.04-0.21 (0.11 ± 0.06) 0.004-0.12 (0.05 ± 0.05) 0.001-0.02 (0.01 ± 0.01) Ratio 4.9 * 4.9 * 7.9 * 7.9 * * Values are significantly higher in the Fe-fertilized than in the HNLC stations (Kruskal-Wallis, P < 0.05).  marineenvironments and found the highest incidence of lysogens in deep waters where the host abundance is typically low, and by Payet and Suttle (2013) during a seasonal cycle study in the Arctic where lysogenic infection prevailed in periods of low system productivity. Apparently, this was not the case in the present study as the fraction of lysogenic cells was not different between trophic situations although bacterial and virus abundances and contact rates were significantly lower at the HNLC than at bloom stations (Table 2). It was suggested that enhanced growth causes temperate viruses to enter the lytic cycle (Wilson and Mann, 1997). Both filtration and incubation could have stimulated bacterial production in the virus reduction approach ) and consequently induced prophages in the mitomycin C treatment controls. Additionally, it has to be stressed that mitomycin C used as an inducing agent of lysogens in the natural bacterial communities may not induce all prophages and be toxic to some bacteria (Paul, 2008;Paul and Weinbauer, 2010). Thus, the apparent low incidence of lysogenic infection, particularly in HNLC waters, might be an artefact. However, it could also be that the study period was not long enough to induce potential changes of lysogenic infection. In addition, our study provides no evidence that lysogens were induced by relieving iron addition.

Role of viruses for sustaining phytoplankton productivity by Fe supply
Bacteria store about 50 % of the biogenic iron in HNLC areas (Tortell et al., 1996) and the mode of bacterial mortality will affect the way of Fe regeneration and bioavailability (Kirchman, 1996;Mioni et al., 2005;Strzepek et al., 2005). While viral lysis liberates organically complexed iron, which may be assimilated rapidly, grazing mainly sets free inorganic Fe (Gobler et al., 1997;Poorvin et al., 2004). Assimilation studies with a model heterotrophic bacterium demonstrated that Fe in the virus-mediated cell lysates was more bioavailable than the siderophores produced by the same cells, supporting the importance of virus-mediated Fe regeneration in marine surface waters (Poorvin et al., 2011). We calculated Fe release rates in two ways: first, based on VP, and, second, based on VMM related to FIC by a model of Binder (1999). The former resulted in average iron regeneration rates due to viral lysis of bacteria of 4.18 and 0.86 pMol Fe d −1 in fertilized and HNLC waters, respectively, while the latter resulted in more realistic values ranging from 0.03 to 1.58 pM d −1 (average: 0.42 ± 0.49 pM d −1 ) in iron-fertilized waters and from 0.004 to 0.12 pM d −1 (average: 0.05 ± 0.05 pM d −1 ) in HNLC waters (Table 6). These values are similar to those found in the Southern Ocean (Evans and Brussaard, 2012) and an iron-induced bloom (ibid., Weinbauer et al., 2009) but low compared to other studies. Poorvin et al. (2004) reported Fe regeneration rates of 19.2-75.5 pM d −1 in HNLC waters off Peru, and Strzepek et al. (2005) found a high range over 2 orders of magnitude of 0.4-28 pM d −1 in HNLC waters southeast of New Zealand. Fe regeneration rates are calculated from virally induced bacterial loss, which is inversely related to burst size. When taking into account that the calculated burst size in the present study was 5 times higher than the assumed BS in the study of Poorvin et al. (2004), the values in the present study compare well to data on Fe regeneration through viral activity from artificial fertilization experiments and other environments. Significantly more iron was released by viral lysis within the naturally Fe-fertilized bloom than at the HNLC stations (P < 0.05, Kruskal-Wallis, Table 6). The concentration of dissolved iron in the surface mixed layer on and off  (Blain et al., 2007), and the estimated biogenic iron pool at the main bloom station equaled 80 ± 9 pM (Sarthou et al., 2008). Taking into account the total Fe demand of the producers within the bloom of 6.04 ± 0.62 pM d −1 (Sarthou et al., 2008), the remobilization of iron through viral lysis above the Kerguelen Plateau following the model by Binder (1999) accounts for up to 26 % of the demand of the producers, and this appears to be a non-negligible iron source for sustaining plankton productivity.

Implications for carbon cycling and sequestration
Bacterial biomass and production were increased respectively from 287 to 797 mg C m −2 and from 23.5 to 304 mg C m −2 d −1 between the HNLC (C11) and the ironfertilized (A3) areas . Bacterial abundance and production are often correlated with viral abundance and production. Thus, elevated bacterial activity in the (natural or induced) bloom could explain the enhanced viral abundance and production found in previous in situ Fe enrichment studies (Arrieta et al., 2000;Higgins et al., 2009;Weinbauer et al., 2009). The finding of higher viral lysis rates of bacteria in the sites of natural Fe fertilization, where HNF grazing could only explain a small fraction of bacterial mortality , has important implications for the carbon cycling. Due to enhanced viral lysis, less carbon will be transferred to larger members of the food web but becomes again part of the DOM pool (Middelboe et al., 1996). This viral shunt should result in elevated bacterial production and respiration; thus more CO 2 would be produced and less carbon sequestrated. Experimental studies indicate that most of the lysis products belong to the labile fraction of DOM and are consequently rapidly degraded (Weinbauer et al., 2011). By the transformation of bacterial biomass into DOM, viruses have the effect of retaining carbon and nutrients in the photic zone (Suttle, 2007). Thus, viral lysis of bacteria could shortcircuit the biological pump (Brussaard et al., 2008a). However, there are other possible scenarios. For example, microbial activity converts part of the organic matter into recalcitrant DOM (RDOM) that is resistant to microbial utilization and can persist in the interior of oceans for up to thousands of years. The detailed role of viral lysis in this new concept of the microbial carbon pump (MCP) (Jiao et al., 2010) is still poorly known. However, a compilation of data suggests that viral lysis increases the DOM pool and the ratio of recalcitrant vs. labile organic matter (Weinbauer et al., 2011). Thus, enhanced viral lysis of bacteria due to Fe fertilization could result in an enhanced carbon sequestration not related to the biological pump.
Rates of bacterial production ([ 3 H] leucine incorporation) and respiration (< 0.8 µm size-fraction) were 5-6 times higher in the bloom at station A3 than those in surrounding HNLC waters, indicating that heterotrophic bacteria within the bloom processed a significant portion of primary production, with most of it being rapidly respired , fuelling the CO 2 pool. This scenario is coherent with the finding of small particulate organic carbon export fluxes to depth necessary for long-term sequestration (de Baar et al., 2005;Street and Paytan, 2005), despite the role of iron in regulating primary productivity. However, most in situ mesoscale iron enrichment experiments so far performed in the HNLC regions did not last long enough to follow the termination of the bloom (Buesseler and Boyd, 2003;Smetacek et al., 2012). In the present study, we sampled a bloom in its late successional stage and could thereby track the fate of fixed carbon by an iron-fertilized phytoplankton bloom. Figure 4 shows a simple sketch to highlight the importance of each compartment of the microbial food web in the transfer of organic material in an Fe-fertilized bloom compared to HNLC waters. Sequestration of material in viruses, bacteria and dissolved matter may lead to stronger retention of nutrients in the euphotic zone in systems with high viral lysis rates of bacteria, because more material remains in these small, non-sinking forms. This could be of major importance for large-scale iron fertilization of ocean regions as a means of enhancing the ability of the ocean to store anthropogenic CO 2 and mitigate 21st-century climate change.

Conclusions
Enhanced bacterial production following the iron-fertilized phytoplankton bloom induced a switch from grazing to viral lysis as major mechanisms causing bacterial mortality. This could change the carbon flow through the microbial food web. We suggest that enhanced viral lysis of bacteria short-circuits the biological pump but potentially primes the microbial carbon pump.