Seawater samples were collected in Otsuchi Bay, located on the Pacific coast of northeastern Japan (39°20′36″N, 141°56′47″E; water depth: 46 m) on January 24, March 11, May 30, and July 24 of 2013. Dynamics of bacteria and viruses in Otsuchi Bay have been described elsewhere (Nagata et al., In press). Samples were collected at a depth of 5 m using a 12-L Niskin bottle (Sea-Bird Scientific, United States), gravity filtered through a 20-μm mesh-size nylon, and then used for subsequent analysis. Prescreening was used to remove larger particles, including occasionally abundant diatom cells (Tachibana et al., 2017) that can clog filters. Bacteria associated with particles larger than 20 μm were not considered in this study. All sampling tanks, cylinders, and incubation bottles used for the experiments were thoroughly cleaned through sonicating in alkaline detergent (5% Extran, Merck Ltd., Germany) and then soaked overnight. Then, the containers were rinsed with hot water to remove the detergent, and then rinsed with Milli-Q water. After that, the containers were soaked in 1 N HCl overnight and thoroughly rinsed with Milli-Q water. Before each incubation experiment, the incubation bottles and cylinders were rinsed with virus-free seawater (see below) three times to wet their inner surfaces.

Viral production was measured according to the method of Winget et al. (2005). Approximately 1 l of seawater sample was concentrated using a 0.2-μm pore-size polyethersulfone tangential flow filtration filter (Vivaflow 200, Sartorius Stedim Lab Ltd., Germany). After the concentrate reached about 100–200 ml in volume, virus-free seawater (filtrate from 100-kDa polyethersulfone filter tangential flow filtration; Vivaflow 200, Sartorius Stedim Lab Ltd., Germany) was added to the original volume. This washing procedure was repeated one or two additional times. In the final reduction step, the sample volume was reduced to approximately 100 ml. The filter was then slowly back-flushed to recover the remaining concentrate in the filtration cartridge and tubes. Seawater was pumped through the tangential filtration system using a peristatic pump (Masterflex L/S variable speed modular drive 7553-80, Cole-Parmer, United States), with a gentle backpressure of approximately 20 kPa (2.9 psi). This backpressure is less than half of that used by Winget et al. (2005) with the recommended flow conditions (8–10 psi). The concentrate was reconstituted with virus-free seawater to its original volume, and then incubated in triplicate 125-ml polycarbonate bottles (Nalgene, Thermo Fisher Scientific, United States). Samples were incubated at in situ temperature in the dark. Subsamples of 2 ml were collected every 3 h for the first 12 h and then at 24 h. Subsamples were placed in cryovials (Nalgene, Thermo Fisher Scientific, United States) and fixed with 0.02-μm (Anotop™ 25, Whatman, United Kingdom) prefiltered glutaraldehyde (20% solution for electron microscopy, FUJIFILM Wako Pure Chemical Co. Japan), at a final concentration of 1%, for 15 min in the dark. The samples were flash frozen in liquid nitrogen and stored at −80°C until flow cytometric analysis (see below). The viral production rate was determined as the slope of the linear regression of viral abundance against time, with a correction for bacterial loss due to sample processing (Winget et al., 2005).

In theory, the virus reduction assay is applicable to <0.2-μm communities, using finer (e.g., 0.1-μm) pore-size filters. However, the filtration process would result in further loss of bacterial cells and, thus, may introduce analytical challenges in detecting viral production. Furthermore, a significant fraction of bacteria may pass through 0.1-μm pore-size filters (Fukuda et al., 1998). For these reasons, we did not perform the virus reduction assay using <0.2-μm communities. Instead, we examined the net increase of viral abundance in <0.2-μm communities in samples that were or were not diluted using virus-free seawater. With dilution, both bacterial and viral abundances are reduced by the same proportion. Because the encounter rate of cells and viruses is the product of bacterial abundance, viral abundance, and the encounter kernel (Brownian collisions in the case of <0.2-μm communities; Murray and Jackson, 1992), a dilution factor of ×0.5, for example, means that the encounter rate between cells and viruses is reduced to 0.25 (=0.5²) times its original value. Thus, viral re-infection during incubation is strongly inhibited by dilution. The increases of viral abundance in both undiluted and diluted samples indicate that viral production occurs due to previously infected cells. If viruses increase only in undiluted samples and not in diluted samples, viral production is likely to be attributable to re-infection. A disadvantage of the dilution approach is that, in contrast to the virus reduction assay, both bacterial and viral abundance decrease by the same proportion, complicating detection of viral production at high dilution rates. Therefore, we prepared a dilution series (except in July) and assumed that if viral abundance increased in at least one of the dilutions, viral production was associated with previously infected cells. Note that our approach differs from the dilution method used for determining the grazing pressure and virus-induced mortality of phytoplankton (Worden and Binder, 2003), in which a response of the apparent cell growth rate to the extent of dilution is analyzed to estimate cell mortality. In our case, we simply followed the idea of the virus reduction assay using a dilution series.

Dilution series were prepared by diluting the <0.2-μm filtrate with virus-free sea water (100-KDa filtrate, see description above) to 20, 40, and 70% of the initial concentration, or left undiluted (100%), hereafter referred to as the 0.2, 0.4, 0.7, and 1 dilutions (the values indicate the fraction of original seawater), respectively (in July, we prepared only the 0.2 dilution). Triplicate of 100-ml samples of the diluents were placed into125-ml Nalgene polycarbonate bottles and incubated at in situ temperature in the dark. Samples for flow cytometric analysis were collected at a 3-h interval over 12 h (referred to as T0, T3, T6, T9, and T12) and at 24 h after the beginning of the incubation (T24). Subsamples for flow cytometry were collected in cryovials (Nalgene, Thermo Fisher Scientific, United States) and fixed with 0.02-μm (Anotop™ 25, Whatman, United Kingdom) prefiltered glutaraldehyde (20% solution for electron microscopy, FUJIFILM Wako Pure Chemical Co., Japan), at a final concentration of 1% for 15 min in the dark. The samples were flash frozen in liquid nitrogen and stored at −80°C until further flow cytometric analysis (see below).

The time courses of viral abundance were discontinuous (see section “Results”). The difference between the maximum and minimum viral abundance (ΔVA) was calculated over the time interval during which viral abundance increased significantly. The frequency of infected cells (FIC) was estimated by dividing ΔVA by (BS × BA) (Winter et al., 2004), where BS is burst size and BA is bacterial abundance at T0. We assumed a BS of 24, the average obtained in marine environments (Parada et al., 2006, 2008). The sensitivity of FIC estimates to the assumed BS was examined using a range of BS values (20–40) (Winter et al., 2004).

Viral production in the <0.2-μm filtrate (VP_<0.2) was estimated using a modification of the “piece-wise” method (Winter et al., 2004; Kostner et al., 2017). This method was proposed for determining viral production through the virus reduction assay in cases where viral abundance shows a discontinuous (pulsed) dynamic rather than a linear increase. Viral production is estimated by dividing the ΔVA or the sum of ΔVA by the time elapsed from the beginning of the incubation to observation of the last peak (Winter et al., 2004). In our modified method, we divided ΔVA by 1 (day), assuming a diel cycle of viral production (Winter et al., 2004; Aylward et al., 2017). Where applicable, VP_<0.2 was derived through correction with the dilution factor. Our estimates of VP_<0.2 are conservative because they are based on the net increase in viral abundance with no correction for viral decay. For one case where re-infection was suspected (T0–T3 in March, see section “Results”), the corresponding ΔVA value was excluded from the calculation of FIC and VP_<0.2.

Viruses and bacteria were counted using flow cytometry (Brussaard, 2004; Yang et al., 2010). Prior to analysis, samples were slowly thawed on ice. To count viruses, samples were diluted 100-fold with Tris-ethylenediaminetetraacetic acid (EDTA) buffer (hereafter, TE; 10 mM Tris-hydrogen chloride, 1 mM EDTA, pH 8.0; Nippon Gene Co., Japan; prefiltered through 0.02-μm pore-size inorganic membrane filters, AnotopTM 25; Whatman, United Kingdom). Then, the samples were stained with SYBR Green I (Molecular Probes, United States) at a final concentration of 5 × 10^–5 at 80°C for 10 min at room temperature in the dark. To obtain bacterial counts, samples were diluted 10-fold with TE buffer, stained with ×10^–4 SYBR Green I, and incubated for 15–30 min at room temperature in the dark. After the addition of 1-μm yellow-green beads (Molecular Probes, United States), samples were injected into the flow cytometer (FACSCalibur; Becton Dickinson, United States). For both viral and bacterial samples, three blanks (treated in the same manner but without the addition of seawater samples) were prepared alongside the samples. The optimal voltage range of the green fluorescence (FL1) photomultiplier was determined by checking the electronic noise level of 0.02-μm filtered TE buffer. Using this voltage setting, three blanks were run to identify a noise window for separation of viral or bacterial counts from noise. The contribution of noise to the total count was always <5% for viruses and <0.1% for bacteria. High and low nucleic acid content bacteria (HNA and LNA bacteria, respectively) were discriminated on the cytogram of side scatter versus FL1. Consistent with previous observations in diverse aquatic environments (Li et al., 1995; Bouvier et al., 2007), the HNA and LNA bacteria were clearly distinguishable on visual inspection of the cytogram. Therefore, manual gating was used to discriminate these subgroups (Bouvier et al., 2007). The flow cytometric data were analyzed using the CellQuest software (Becton Dickinson, United States).

SigmaPlot v.14.4 (Systat Software, United States) was used for statistical analysis. Whenever the viral abundance increased between two sampling times, the difference in viral abundance between each time interval was assessed using Student’s t-test (p-values < 0.05 were considered significant), except for between T0 and T3. Because we took only a single measurement at T0, the significance of the difference in viral abundance between T0 and T3 could not be tested using Student’s t-test, and instead we used the 95% confidence interval of the T3 value. If the T0 value does not overlap with the confidence interval of the T3 value, we considered the T0 value significantly different from the T3 value. For calculation of error propagation, the coefficient of variability (CV, standard deviation/mean × 100) of the T0 value was assumed to equal the CV of the T3 value. The CV values for viral abundance estimates for triplicate incubations were generally <5%.

Experiments were conducted in winter (January and March) and the stratified period (May and July) (Supplementary Table 1). Table 1 summarizes the bacterial and viral abundance parameters in the filtrate of the tangential flow filtration. This table is an overview, rather than an assessment, of the differences in parameters among months (because the tangential flow filtration, which uses expensive filter cartridges, was conducted only once for each month, the uncertainty of each parameter associated with the operational error was unknown). The fraction of LNA bacteria that passed through the 0.2-μm filter was 17–35%, with a corresponding value for HNA bacteria of 0.08–5%. The contribution of LNA bacteria to the total bacterial communities in the 0.2-μm filtrate was 78–99% (Table 1). The fraction of viruses that passed through the 0.2-μm filter was 68–89% (Table 1).

Viral abundance increased linearly during the initial 6 or 12 h of incubation (r² = 0.50–0.92; p < 0.05, Supplementary Figure 1). Viral production estimated using the virus reduction assay (VP_red; 10⁹ viruses l^–1 d^–1) was lower in winter (3.41 ± 0.23 in January and 1.05 ± 0.34 in March) than during the stratified period (14.2 ± 0.55 in May and 8.36 ± 2.13 in July).

Viral dynamics in the <0.2-μm filtrates were discontinuous. For example, in January, viral abundance increased significantly between T0 and T3 (Figure 1). After T3, viral abundance remained constant until T9 and then declined. Similarly, in other months, viral abundance increased during only a certain time interval (Supplementary Figure 2). These increases in viral abundance were considered to reflect viral production from previously infected cells, as viral abundance increased in diluted samples. An exception was found for the time interval of T0–T3 in March, when viral abundance increased only in the undiluted treatment. The ΔVA for this time interval was not used for the calculation of FIC and VP_<0.2 (see section “Materials and Methods”). LNA bacteria dominated the community (%LNA > 80%) throughout the incubation period except for the later portion of incubation in May and July, when %LNA decreased after T12 (Figure 1 and Supplementary Figure 2). We found no substantial decrease in %LNA during the time intervals when viral abundance increased.

The FIC was estimated using the data from the different dilutions in January, March, and May. The FIC estimates were consistent among different dilutions, except in March when the FIC for the 1 dilution was approximately half of those for the 0.4 and 0.7 dilutions (Table 2). The FIC was also sensitive to the BS assumption. Using the range of BS (20–40) reported by Winter et al. (2004), FIC estimates differed twofold depending on the BS assumption. For each month, the average FIC using the default BS setting (24) ranged from 42 to 93% in January, March, and May, and was 7% in July (Table 2). The average VP_<0.2 (10⁹ viruses l^–1 d^–1) for each month varied in the range from 0.76 to 3.99 (Table 3). From VP_<0.2 and VP_red, we estimated the total viral production, VP_tot, using the following equation: VP_tot = (1−f) × VP_red + VP_<0.2, where f is the fraction of bacteria that passed through the 0.2-μm pore-size filter. VP_red is multiplied by (1−f) because the calculation of VP_red corrects for any loss of bacteria during the tangential flow filtration using the bacterial recovery rate (this correction implicitly includes the loss of bacteria due to the passage of bacteria through 0.2-μm pore size filters). Then we examined the percentage of VP_<0.2 relative to VP_tot and found that VP_<0.2 accounted for a large proportion (36–43% on average) of VP_tot in January, March, and July, although the percentage of VP_<0.2 in May was lower (11%) (Table 3). Comparison of the VP_red and VP_tot revealed that VP_tot was 1.3–1.7 times higher than VP_red in January, March, and July, whereas VP_tot/VP_red was near 1 in May (Table 3).