Microbial food web dynamics during spring phytoplankton blooms in the naturally iron-fertilized Kerguelen area (Southern Ocean)

Microbial food web dynamics were determined during the onset of several spring phytoplankton blooms induced by natural iron fertilization o ﬀ Kerguelen Island in the Southern Ocean (KEOPS2). The abundances of heterotrophic bacteria and nanoﬂagellates, bacterial heterotrophic production, bacterial respiration, and bacterial 5 growth e ﬃ ciency, were consistently higher in surface waters of the iron-fertilized sites than at the reference site in HNLC (high nutrient low chlorophyll) waters. The abundance of viral like particles remained unchanged, but viral production increased by a factor of 6 in iron-fertilized waters. Bacterial heterotrophic production was signiﬁcantly related to heterotrophic nanoﬂagellate abundance and viral production across all sites, 10 with bacterial production explaining about 70 and 85 %, respectively, of the variance of each in the mixed layer (ML). Estimated rates of grazing and viral lysis, however, indicated that heterotrophic nanoﬂagellates accounted for a substantially higher loss of bacterial production (50 %) than viruses (11 %). Combining these results with rates of primary production and export determined for the study area, a budget for the ﬂow 15 of carbon through the microbial food web and higher levels during the early (KEOPS2) and the late phase (KEOPS1) of the Kerguelen bloom is provided. incubated leucine leucine 13 and for 9 of 3 H leucine situ the deep mixed the represent a major source of CO at the early bloom This picture di ﬀ ers from that obtained during the late bloom phase, where BGE was but the viral shunt prevented an e ﬃ cient transfer of the BP to higher trophic levels. This study highlights the variability of the bacterial contribution to and export according to the trophic dynamics at di ﬀ erent underlying varying mechanisms channelling To improve our understanding of further investigation into time series over determine timing factors that induce


Introduction
The Southern Ocean has a unique geography with major implications for the global ocean circulation and climate system. It is also the largest HNLC (high nutrient low 20 chlorophyll) ocean where iron limits phytoplankton primary production, resulting in a large stock of unused major inorganic nutrients (Martin and Fitzwater, 1990). Natural and artificial iron fertilization studies have suggested regional variability in the ecosystem response to iron addition and in the carbon export (Blain et al., 2007;Pollard et al., 2009;Boyd et al., 2007). Although the underlying mechanisms are not fully un-of Kerguelen Island. The sampling strategy covered spatially diverse fertilized regions at early bloom stages (October-November 2011).
The main objective of the present study was to provide for the first time insight into seasonal dynamics of the microbial food web functioning in the context of natural Fe fertilization of the Southern Ocean. The major biogeochemical and biological parameters 10 reported in this study are full depth profiles of microbial stocks (viruses, heterotrophic bacteria, and heterotrophic nanoflagellates), bacterial production, potential bacterial grazing, and viral lysis, while respiration was measured at selected depths in surface waters.

Sample collection
The present study was carried out during the KEOPS2 cruise from 15 October to 20 November 2011. Water samples for this study were collected at 10 stations along a North-South transect (TNS; stations 1-10) and at 7 stations along an East-West transect (TEW; Stations 1-8) (Fig. 1 of 2 surface drifters, the other 5 main stations were sampled in a quasi-Lagrangian manner within a complex meander south of the Polar Front (E1, E2, E3, E4E, and E5). Station A3 was visited twice (A3-1 and A3-2) during the onset of the bloom (KEOPS2), which complemented previous investigations following the decline of the phytoplankton bloom (KEOPS1). The reference site (Station R) in High Nutrient Low Chlorophyll 5 waters (HNLC) was situated west of the plateau (Table 2). All water samples were collected with 12 L Niskin bottles mounted on a rosette equipped with a CTDO Seabird SBE911-plus. Sampling for microbial parameters presented here was performed at 11-14 depths at each station. 10 grazing

Abundances of microbial components and heterotrophic nanoflagellate
The abundance of virus-like particles (VLP) of heterotrophic bacteria ((HB) − sensus stricto heterotrophic Bacteria + Archaea) and HNF (heterotrophic nanoflagellates) were determined by flow cytometry. Subsamples (2.5 mL for VLP and HB, and 4.5 mL for HNF) were fixed with a Transmission Electron Microscope (TEM) grade glu-15 taraldehyde (final concentration 1 %) for VLP and HNF; and with formaldehyde (2 % final concentration) for HB. VLP and HB, and HNF were refrigerated for 10-20 min and 2 h, respectively, then frozen in liquid nitrogen and stored at −80 • C until analysis. Counts of VLP and HB were made using a FACSCalibur flow cytometer (BD-Biosciences) equipped with an air-cooled laser, providing 15 mW at 488 nm with the 20 standard filter set-up. VLP and HB were stained with SYBRGreen I, as described in detail in Marie et al. (1999) and Brussaard (2004). Populations of VLP and HB differing in fluorescence intensity were distinguished on plots of side scatter vs. green fluorescence (530 nm wavelength, fluorescence channel 1 of the instrument). Flow cytometry list modes were analyzed using CellQuest Prosoftware (BD Bioscences, version 4.0). Introduction To establish the size of cells of the three cytometric populations identified during this study (HNF1, HNF2, and HNF3, Fig. A1), cells from each population were sorted with a FACSAria cell sorter (BD-Biosciences). The sorted cells (1000-3000 cells per sample) were collected on Nuclepore filters (0.2 µm pore size, 25 mm diameter), and examined using a Zeiss AX10 microscope at 1000×. The mean biovolume of each cyto-5 metric population was calculated based on the linear dimensions of the cells, applying a prolate spheroid equation. Clearance rates (nL HNF −1 h −1 ) were estimated based on 10 5 biovolume −1 h −1 (Fenchel, 1982;Christaki et al., 2001). The potential of HNF grazing accounting for the relative loss of bacterial heterotrophic production (BP) was then calculated as: 10 % BP loss = bacterial cells "cleared" L 1 h −1 × 100/bacterial cells produced L −1 h −1 (1)

Conclusions References
The number of bacterial cells produced was calculated from BP, as determined by leucine incorporation (see below), using a conversion factor of 12.4 fg C cell −1 (Fukuda et al., 1998).

Phage-infected bacteria and burst size
For observations on the Transmission Electron Microscope (TEM), the 4.5 mL subsamples collected at different depths were pooled before ultracentrifugation for the following layers: the mixed layer ( phages. Burst size in each of the 30 samples was defined as the average number of viral particles in all visibly infected cells. This is likely the minimum burst size, as more viral particles may accumulate within an infected cell before it lyses. To estimate viral induced bacterial mortality as a % of the bacterial production (VIBM), the frequency of infected cells (FIC with data given as percentages) was first calculated from the 5 frequency of FVIC according to Weinbauer et al. (2002): The proportion of the total bacterial mortality that was due to virally induced lysis was calculated according to Binder (1999): Viral production (VP) was estimated according to Weinbauer et al. (2003): The BS used in the equation was the BS of each of the 30 samples observed with the TEM.

Bacterial production, respiration, and bacterial growth efficiency
The incorporation of 3 H leucine was used to estimate BP. Leucine concentrations and incubation times were tested on board and adjusted for different depths in order to ob-20 tain a sufficient radioactivity signal and to maintain linear uptake during the incubation. At each depth, 20 mL triplicate samples and a trichloroacetic acid (TCA)-killed control were incubated with a mixture of L-[4,5-3 H] leucine (Amersham, 144 Ci mmol −1 ) with Bacterial respiration was determined at all main stations, except R and E2, at one to three depths within the ML (Table 1) as described in Obernosterer et al. (2008). At Station R, bacterial respiration rates were estimated from both dark community respiration (unfiltered seawater) and the fraction of dark community respiration accounted for by bacterial respiration as determined for the iron-fertilized sites (see Sect. 3.3).
Briefly, rates of respiration were determined from dissolved oxygen consumption in 24 h dark incubations of 0.8 µm-filtered samples. Dissolved oxygen was determined by Winkler titration using a PC-based system with a photometric endpoint detector (Lefèvre et al., 2008). Bacterial growth efficiency was determined from in situ bacterial production and respiration rates as: We used a respiratory quotient of 1, as determined for the Kerguelen study region during KEOPS1 (Lefèvre et al., 2008)

Study sites
The hydrographic conditions during KEOPS 2 are reported in detail in . The "historical" A3 station situated above the Kerguelen plateau (Blain et al., 2007(Blain et al., , 2008 was characterized by a deep ML (Z ML , depth of the mixed layer from 150 m to 5 170 m) (Fig. 2). Here mean concentrations of Chl a in the ML increased from 0.6 µg L −1 to 2.0 µg L −1 between the first and second visit three weeks later (

Distribution of microbial community components
VLP, HB, and HNF abundances in the upper 200 m were of the order of 10 9 , 10 8 , and 10 5 particles cells −1 L −1 , respectively (Fig. 3, Table 3). HB abundance, the %HNA (high nucleic acid) containing bacterial cells, and HNF abundances were significantly higher in the upper 200 m of the fertilized stations compared to the HNLC-site R 5 (Mann-Whitney, p < 0.05); this difference disappeared below 200 m ( Table 3). The abundance of VLP was not significantly different between sites at any of the depth layers considered ( Table 3). The variation between HNLC and Fe-fertilized sites was most pronounced for HNF (range ∼ 200-900×10 3 L −1 ), if compared to VLP (range 1.4-1.7 × 10 9 L −1 ), and HB (range 2.7-4.7 × 10 8 L −1 ). HNF were distinguished in the three 10 cytometrically identified subpopulations HNF1, HNF2, and HNF3 based on their cytometric signatures (Fig. A1). The individual cell biovolumes determined after cell sorting of each of these populations were 3.9 ± 1.6 µm 3 for HNF1, 36.9 ± 9.3 µm 3 for HNF2, and 62.2 ± 41.1 µm 3 for HNF3 µm 3 (all sites and depths pooled). The calculated clearance rates were 0.4 ± 0.2, 3.7 ± 0.4 and 6.2 ± 1.3 nL HNF −1 h −1 for HNF1, HNF2, and HNF3, respectively. Size-specific clearance rates differed only slightly among bloom sites, and no significant differences between the Fe-fertilized sites and HNLC waters were detected (Mann-Whitney tests, p > 0.05). The relative abundances of these cytometrically identified subpopulations were similar among sites and throughout the water column accounting for roughly 46 % for HNF1, 52 % for HNF2, and 2 % for HNF3 of the 20 total HNF abundance (Table 3).

Bacterial production, growth rates, respiration, and growth efficiency
BP and bacterial growth rates were 5.2 nmol C L −1 d −1 and 0.018 d −1 in the ML at station R, and they were overall higher in the Fe-fertilized region. BP ranged between 9.9 and 133.8 nmol C L −1 d −1 and bacterial growth rates varied between 0.025 d −1 and 25 0.210 d −1 (mean ML at Stations A3-1 and F, respectively) ( Fig. 4 and Table 4). At station A3, BP and bacterial growth rates increased 4 to 5 fold between the first between 4.6 % and 25 % (mean 16.5 ± 7.0 %) in Fe-fertilized stations, and 43 % at station R of the 0-1000 m integrated bacterial production (data not shown). Bacterial respiration rates varied by 8-fold among the Fe-fertilized sites, with lowest and highest rates at Station E1 (mean ML 0.23 ± 0.06 µmol C L −1 d −1 , n = 3) and E5 (1.73 µmol C L −1 d −1 , n = 1), respectively (Table 5). At the Fe-fertilized sites bacterial respiration accounted 10 on average for 59±20 % of dark community respiration (Cavagna et al., 2014). We estimated bacterial respiration at Station R to vary between 0.25±12 µmol C L −1 d −1 (n = 4) (dark community respiration in unfiltered seawater) and 0.14±0.07 µmol C L −1 d −1 (59 % of dark community respiration) ( Table 5). Due to the lower contribution of phytoplankton and HNF to overall microplankton biomass in HNLC compared to Fe-fertilized waters, 15 the contribution of bacterial to dark community respiration is likely to be higher at Station R than the mean value determined for the Fe-fertilized sites. We therefore use the respiration rate in unfiltered seawater as an upper estimate for bacterial respiration throughout the manuscript ( Table 5). The cell-specific respiration revealed a similar pattern with lowest rates at stations E1 (mean ML 0.54 ± 0.13 fmol O 2 cell −1 d −1 , n = 3), 20 and highest rates at station E5 (3.76 fmol O 2 cell −1 d −1 , n = 1) ( Table 5). BGE ranged between 3 % and 17 % in the ML of the fertilized sites (mean 9 ± 7 %, n = 14). One exceptionally high value of 28 % at the base of the ML at Station FL was registered. At Station R, BGE was 3 ± 1 % (n = 4) based on respiration rates in unfiltered seawater. Assuming BR accounts for 59 % of dark community respiration, as described above, 25 would increase the BGE at Station R to 4 ± 2 %. BP and bacterial growth rates were 5.2 nmol C L −1 d −1 and 0.018 d −1 in the ML at station R, and they were overall higher in the Fe-fertilized region, varying up to 13.5 and 8-fold, respectively, among fertilized sites ( Fig. 4 and bacterial growth rates increased 4 to 5 fold between the first (9.9 nmol C L −1 d −1 and 0.025 d −1 , mean, ML) and the second visit (39.6 nmol C L −1 d −1 and 0.122 d −1 ). Highest BP and bacterial growth rates were observed at station FL (133.8 nmol C L −1 d −1 and 0.210 d −1 ). Within the stationary meander, BP and bacterial growth rates increased from 30 nmol C L −1 d −1 and 0.068 d −1 at Station E1, to 54.7 nmol C L −1 d  (Table 5). At Station R respiration rates were 0.25 ± 12 µmol C L −1 d −1 (n = 4) ( Table 5). At the Fe-fertilized sites bacterial respiration accounted on average for 59 ± 20 % of dark community respiration (Cavagna et al., 2014). The cell-specific respiration revealed a similar pattern with lowest rates at stations E1 (mean ML 0.54 ± 0.13 fmol O 2 cell −1 d −1 , n = 3), and highest rates at station E5 (3.76 fmol O 2 cell −1 d −1 , n = 1) ( Table 5). BGE ranged between 3 % and 17 % in the ML of the fertilized sites (mean 9±7 %, n = 14). One exceptionally high value of 28 % at the base of the ML at Station FL was registered. At Station R, BGE was 3 ± 1 % (n = 4) based on respiration rates in unfiltered seawater. Assuming BR accounts for 59 % of 20 dark community respiration, as determined for the fertilized sites, would increase the BGE at Station R to 4 ± 2 %.

Bacterial losses and viral production
The potential grazing capacity estimated by the clearance rates of HNF revealed losses of BP of 50 %, 70 %, and 85 % for the ML, the Z ML -200 m, and the > 200 m layers, respectively ( Fig. 5a). In surface waters (ML) the loss of BP due to grazing varied between ∼ 30 % and ∼ 60 % at the fertilized stations, with the exception of Station A3-1 where this value was 80 %. The BP loss due to grazing at station R accounted for about 70 % of the bacterial production ( Fig. 5a). BP loss due to viral lysis was comparatively low, and varied from undetectable to 24 % of BP in the ML (Fig. 5b). The viral induced loss of BP were 11 %, 2 %, and 2 % for the ML, Z ML -200 m, and > 200 m layers, respectively ( Fig. 5b). A higher percent of viral mortality was encountered in the ML at 5 stations E4E, E2, E1, and R. Overall there was no difference in the percent loss of BP induced by grazing and viral lysis between the Fe-fertilized sites and the HNLC site R ( Fig. 5a and b). The sum of bacterial mortality due to HNF grazing and viral lysis varied from 47 % at station E1, to 100 % at stations E3 and E5. Together, grazing and viral lysis accounted for an average of 83 % bacterial mortality at all stations ( Fig. 5a and b). The 10 empirically estimated burst size of bacteria (BS) was 22 ± 15 virus cell −1 (mean ± SD, n = 30), and varied from 6 (A2, in the ML) to 88 (FL, in the Z ML -200 m) without any specific pattern related to Fe-fertilization or depth. Viral production (VP) calculated based on the BS and BP (Eq. 4) was 2 ± 1 and 0.34 ± 0.07 × 10 8 viruses L −1 d −1 in the ML of the fertilized stations, and at station R, respectively.

15
Combining these results revealed several significant relations (Table 6). In the ML, HB and BP were significantly related to Chl a. BP and the abundance of HNF were significantly related in the three layers considered, with BP explaining up to 79 % of the HNF variability (r 2 = 0.79). While BP and VLP -except in the Z ML − 200 m layershowed little relation, BP and VP showed highly significant relations in the first 200 m, 20 and a weaker, but still significant relation in the deep layer. It is worth noting that the relations with VP were insignificant when an overall mean BS of 22 virus cell −1 was applied, but became highly significant when the specific BS for each sample was applied. VP and HB abundance were significantly related in the upper 200 m, however the determination coefficient was relatively low (r 2 = 0.5). The relation between BP and BR 25 was significant with BP explaining ∼ 50 % of the BR variance (r 2 = 0.51). The relation between VP and BR was weaker, but still significant (r 2 = 0.32). Notably, there was no detectable relation between VP and VLP at any depth. The stocks of HB were sig- To date, 6 artificial mesoscale, and 2 detailed natural iron fertilization experiments have been conducted in the Southern Ocean and microbial food web dynamics have been considered in part in these studies ( Table 7). The most commonly determined microbial parameters, such as HB abundances and BP, are generally enhanced upon iron fertilization. However, remarkable variability in the extent of the response exists among 10 different regions of the Southern Ocean (Table 7). During KEOPS2, a patchwork of blooms induced by large-scale natural iron fertilization above the plateau and in the oceanic region off Kerguelen Island was investigated . The results obtained from KEOPS2 add to previous studies by providing an extensive description of microbial food web dynamics during the onset of spring phytoplankton blooms under 15 varying hydrographic and biogeochemical conditions. Natural iron fertilization of the Southern Ocean induced rapid responses of members and fluxes of the microbial food web. The intensity of the response and its variability among sites were most pronounced in the 200 m surface layer, and this likely reflects the hydrographic and biogeochemical characteristics at the given sites. The depth of 20 the mixed layer (Z ML ) varied considerably, extending down to 170 m above the plateau at Station A3, and to 40 m north of the Polar Front at Station FL. To note that although surface Chl a concentration, which is a proxy for phytoplankton biomass, varied by a factor of 6 in the ML among sites, the integrated Chl a at A3-2 and FL stations were similar (Lasbleiz et al., 2014). Noticeable differences in surface temperature, a key fac- results showed the BP response to be strong and varied, accounting for increases of mean values in the ML of up to 13-fold between bloom stations (e.g. A3-1 compared to FL, Table 4), and up to 26-fold between bloom stations and HNLC station R (e.g. FL compared to R, Table 4). This enhancement in BP, together with that observed during CROZEX (9-fold; Zubkov et al., 2007) and KEOPS1 (6-fold;Christaki et al., 2008), was 5 much higher than the enhancement reported from artificial fertilization experiments (roughly 2-fold; Table 7). During KEOPS2, BR was up to 8-fold enhanced by iron fertilization and varied by a factor of 8 among sites among bloom stations. During the late bloom phase, above the plateau, bacterial respiration was about 3-fold higher in the bloom than in HNCL waters (Obernosterer et al., 2008), but to the best of our knowledge, no BR rates are available for comparison from other fertilization studies in the Southern Ocean.
In contrast to bacterial metabolism, the abundance of HB increased overall to a lesser extent (roughly 2-fold), and thus similarly to those reported previously in natural and artificial iron fertilization experiments (Table 7). Besides bulk abundance, the 15 %HNA bacteria were also significantly higher in the ML of fertilized stations (59 %) relative to the HNLC reference site R (47 %) (Table 3). Oliver et al. (2004) reported a relatively minor increase of the fraction of HNA cells (up to 45 %) within the ironfertilized patch during SOFeX. By contrast, the % HNA bacteria accounted for up to 80 % of total bacterial abundance at the late stage of the Kerguelen bloom at Station 20 A3 (Obernosterer et al., 2008). The smaller increase in HB abundance indicates efficient top-down control of these members of the microbial food web that are markedly stimulated by natural iron fertilization.
Although HNF showed significantly higher abundances in the 200 m surface layer in the fertilized stations compared to station R, the HNF size distribution, resulting from 25 sorted cell observations, was remarkably stable, with the two smaller sized populations contributing > 45 % each, at all stations and depths ( Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ever, this was not the case in the present study, where the total number of VLPs was similar in fertilized sites and HNLC waters. On average ∼ 80 % of the VLPs during KEOPS2 showed low green fluorescence, and were most probably bacteriophages. The high green fluorescence VLPs represented ∼ 20 % of total VLPs in the study area, and according to Brussaard et al. (2008); this group may represent larger algal viruses.

5
Although viral stocks did not differ across stations in this study, VP was about 6 times higher in the ML of the fertilized stations compared to HNLC waters. This result is in line with observations from the late bloom phase (KEOPS1, Malits et al., 2014) and artificial iron fertilization experiments (Higgins et al., 2009;Weinbauer et al., 2009). Contrary to previous observations (Weinbauer et al., 2009), the trend of higher bacterial and viral production in iron-fertilized waters was not accompanied by a higher virus induced loss of BP. Lysogeny was not considered in the present study. The proportion of the lysogenized bacterial population can vary extensively from undetectable to 100 % (Weinbauer and Suttle, 1996;Williamson et al., 2002). Across a system study, lysogeny was highest in deep sea waters (Weinbauer et al., 2003) where the contact rate between infective 15 phages and hosts is too low to sustain the lytic life style (Paul et al., 2002). The only study on lysogeny during artificial Fe-fertilization experiments in the Southern Ocean did not observe differences inside and outside the patch (Weinbauer et al., 2009). During KEOPS1 the fraction of lysogenic cells was 8 % and 6 % of the total bacterial cells infected by viruses in the bloom, and in HNLC waters, respectively (Malits et al., 2014).

20
Based on these observations (Weinbauer, 2009;Malits et al., 2014), and this paper's independently obtained results (such as the low viral abundance and the low frequency of infected cells) supports the idea that the loss of BP due to lysis was low at the onset of the phytoplankton bloom. Notably, viruses and HNF revealed opposite vertical trends, with HNF grazing increasing and viral lysis decreasing with depth.

Microbial food web dynamics in response to iron fertilization
The extent of change of the microbial parameters considered in the present study at different fertilized sites across variable hydrographic and biological regimes appears to 7001 Introduction viruses to rise following blooms of their specific bacterial hosts. Indeed, the composition of the bacterial community changed markedly during the bloom above the plateau (West et al., 2008;Obernosterer et al., 2011). Taken together, these results suggest that the mechanism of natural fertilization through the continuous supply of iron strongly affects bacterial heterotrophic metabolism, and as a consequence HNF grazing, with 20 important consequences for the cycling of carbon through the microbial food web.

Implications of microbial food web dynamics for carbon cycling in the bloom above the Kerguelen plateau
The investigation of the spring phytoplankton bloom located in the southeastern part above the Kerguelen plateau during two distinct phases has provided for the first time 25 insight into seasonal dynamics of the microbial food web functioning in the context of natural Fe fertilization of the Southern Ocean. Combining our results with rates of primary production, mesozooplankton activity, and export determined during the KEOPS 7002 Introduction project allowed to propose a budget for the flow of carbon through microbial and higher trophic levels in the southeastern bloom (A3 station, Fig. 6). The main purpose of the carbon budget presented here (Fig. 6) is to place the microbial loop in the context of the food web and to compare the fluxes to the potential export and/or accumulation of phytoplankton biomass during the early and late phases of the bloom.

5
Gross community production (GCP) integrated over the ML was about 7fold higher during the early (620 mmol C m −2 d −1 ) than the late bloom phase (95 mmol C m The ratios between the BCD and GCP were in the same range at the different phytoplankton bloom sites (BCD : GCP ∼ 0.1-0.3), and overall lower than those observed during the late bloom phase (BCD : GCP ∼ 0.3-0.7) (Table A1). These latter estimates are similar to those reported for spring phytoplankton blooms in the polar frontal zone, 20 the marginal ice zone, and the southern Antarctic circumpolar Current (0.3 to 0.6 Lochte et al., 1997). To estimate the fraction of primary production that is processed by heterotrophic bacteria most studies report on the ratio BP : PP. For KEOPS, this ratio was on average 0.04±0.02 for the early phytoplankton blooms compared to an average of 0.10 ± 0.03 for the late bloom phase at Station A3. At the HNLC site R the BP : PP ratio was 0.05. Except for the late bloom stage at A3, the ratios reported here are in the same range as those reported for the Ross Sea (0.04, Ducklow, 2000) and the Arctic Ocean, when primary production was greater than 10 mmol C m Even though the absolute values of BP transferred to HNF and to mesozooplankton were the same during the early and the late bloom (3 and 1.5 mmol C m −2 d −1 , respectively, Fig. 6) the importance of grazing induced consumption of BP was more important during the early than the late phase of the bloom (50 % and 36 %, respectively).
Interestingly, viral activity showed the opposite pattern, since viral induced mortality 5 changed significantly over time, becoming the dominant top-down control of BP during the late bloom phase (Fig. 6, Malits et al., 2014). This suggests that a larger proportion of BP was channeled to HNF and to higher trophic levels during the early bloom, while most of BP returns to DOM through the viral shunt during the late bloom (Wilhelm and Suttle, 1999). These contrasting scenarios have important implications for organic 10 matter and nutrient cycling. During the early stage of the bloom, the different loss terms of GCP were bacterial: (i) (105 mmol C m −2 d −1 ) other microplankton (24.5 + 1.5 mmol C m −2 d −1 ) and mesozooplankton (20 mmol C m −2 d −1 ) respiration, (ii) 0.2 mmol C m −2 d −1 to dissolved pool through bacteria lysis and, (iii) 38 and 1.5 mmol C m −2 d −1 , respectively, were trans-15 ferred directly or through the microbial food web to mesozooplankton for biomass accumulation or export through faecal pellet production. Substracting the sum of the above terms from GCP yields 429.3 mmol C m −2 that remained available in the water column for phytoplankton biomass accumulation and/or export (Fig. 6). Comparison to the POC flux (4 mmol C m  (Fig. 6).
During the late stage of the bloom, organic carbon lost through micro-and mesozoo-25 plankton respiration amounted to 51 mmol C m −2 d −1 , and 17 and 1.5 mmol C m −2 d −1 , respectively, were transferred directly or indirectly to mesozooplankton. Thus, 19.5 mmol C m −2 of GCP were not processed in the water column (Fig. 6) our carbon budget indicates a negligible potential for phytoplankton biomass accumulation during the late bloom phase (Fig. 6). Production was measured at one time point during the onset of the bloom, and the error associated to the integrated flux is 10 %. Daily fluxes are more likely variable due to the influence of environmental factors, such as light intensity, which could in part 5 account for the observed discrepancy. Mesozooplankton ingestion during the early bloom phase are probably overestimated, because rates were extrapolated from the late bloom phase where mesozoplankton abundances were substantially higher (Carlotti et al., 2008;Jouandet et al., 2014).