Light regulation of coccolithophore host-virus interactions

EhV replication was controlled by a trade-off between host nucleotide recycling and de novo synthesis, and that photoperiod and photon ﬂux could toggle this switch. (cid:1) Laboratory results supported ﬁeld observations that light was the most robust driver of EhV replication within E. huxleyi populations collected across a 2000 nautical mile transect in the North Atlantic. (cid:1) Collectively, these ﬁndings demonstrate that light can drive host – virus interactions through a mechanistic interplay between host metabolic processes, which serve to structure infection and phytoplankton mortality in the upper ocean.


Introduction
Virus infection is a primary mechanism of high lysis rates of phytoplankton populations (van Boekel et al., 1992;Bratbak et al., 1993;Brussaard et al., 1995;Valiela, 1995;Agust ı et al., 1998). It is estimated that algal viruses turn over more than a quarter of the photosynthetically fixed carbon, fueling microbial foodwebs and short-circuiting carbon export to higher trophic levels and the deep sea (Fuhrman, 1999;Suttle, 2007) by releasing dissolved organic matter into the surrounding water. At the same time, viral-induced transparent exopolymer particle production (Joassin et al., 2011;Vardi et al., 2012) suggests infection may stimulate vertical sinking flux and enhance biological pump efficiency. Despite the impact on phytoplankton communities, viralinduced mortality is not routinely accounted for in models of ecosystem processes and carbon export, due to inadequate mechanistic and quantitative understanding of the environmental factors that regulate host-virus interactions.
Emiliania huxleyi and its associated Coccolithovirus (EhV) is a highly studied marine eukaryotic algal host-virus model system due to its ecological relevance and the availability of genetically diverse hosts and EhV strains in culture (Schroeder et al., 2002;Bidle & Vardi, 2011). Massive spring blooms of E. huxleyi in the North Atlantic (Townsend et al., 1994;Tyrell & Merico, 2004) are routinely terminated by the giant, lytic, double-stranded DNA containing EhVs (Bratbak et al., 1993;Schroeder et al., 2002;Lehahn et al., 2014;Laber et al., 2018;Sheyn et al., 2018), and studies have revealed that host-virus interactions are critically controlled through a lipid-based chemical arms race and subcellular regulation of autophagy and programmed cell death pathways (Bidle et al., 2007;Vardi et al., 2012;Rose et al., 2014;Rosenwasser et al., 2014;Schatz et al., 2014;Sheyn et al., 2016). These studies have collectively set the stage for detailed exploration into the environmental conditions that regulate key cellular pathways during infection.
As one of the most fundamental and readily measured resources in the ocean, light critically regulates E. huxleyi distribution, growth, and productivity, and in turn viral genome replication and virion production given the need for host resources (e.g. nucleotides, lipids, proteins, and energy). Decreased photosynthetic efficiency (Evans et al., 2006;Bidle et al., 2007) and increased nonphotochemical quenching (Llewellyn et al., 2007) during EhV infection suggest an uncoupling of photosynthetic electron flow and impaired photophysiology (Evans et al., 2006), findings that are supported by transcriptomic analysis of photosynthetic-related genes during infection Gilg et al., 2016). However, the

Viral infection
Exponentially growing E. huxleyi cultures were infected with EhV (virus-to-host ratio: 5), c. 1-2 h into the light phase, unless otherwise stated. In comparative experiments (e.g. LD vs continuous light or LD vs dark), infections were performed concurrently using the same stock of freshly propagated viral lysate. For dark experiments, EhV was added at the onset of the light period to allow adsorption and internalization. At the end of the light phase, a subset of the cultures was wrapped in foil and kept dark for the remainder of the experiment. Control cultures continued on the normal LD cycle. For 3-(3,4-dichlorophenyl)-1,1dimethylurea (DCMU) experiments, 5 lM was added to control and infected cultures 24 h postinfection (hpi). Burst size was calculated by dividing the total number of viruses produced by the number of host cells lost (i.e. maximum host abundance minus final host abundance). Cell biomass was collected for enzyme activity measurements and immunoblots by filtration onto 47 mm diameter, 1.2 lm pore-size polycarbonate filters, snap frozen in liquid nitrogen, and stored at À80°C.

Photosynthetic measurements
Electron transfer rates and the maximum photosynthetic efficiency of photosystem II (PSII), expressed as F v /F m , were measured using a custom-built fast fluorescence induction and relaxation system (Gorbunov & Falkowski, 2005). The minimum (F o ) and maximum (F m ) fluorescence yields, F v /F m ((F m ÀF o )/F m for dark-adapted cells and DF 0 =F 0 m under ambient light), and r PSII (the functional absorption cross-section of PSII) were calculated from the analysis of fluorescence induction at the microsecond time scale. The photosynthesis vs irradiance curves were reconstructed from measurements of electron transport rates (P ¼ E Â r PSII Â DF 0 =F 0 m ) as a function of ambient irradiance E (lmol m À2 s À1 ).

Viral adsorption
Free, extracellular virus abundance was used as a proxy for adsorption, where the disappearance of viruses from the medium is as an indication of viral adsorption onto host cells. Nonspecific adsorption was assessed by adding viruses to cell-free media. A drop in free virus was detected within 5 min, but returned to initial concentrations after 15 min, suggestive of nonspecific adsorption, presumably to the wall of the flask, and subsequent release of viruses. Therefore, measurements earlier than 30 min were not used.

Quantitative immunoblot analysis
Protein extraction and quantitative immunoblots were performed as described (Brown et al., 2008;Thamatrakoln et al., 2013). Immunoblots were probed with primary antibodies (Agrisera, Vännäs, Sweden) against PsbD (a proxy for PSII) and RbcL (a proxy for RuBisCO) at a dilution of 1 : 40 000, followed by a horseradish-peroxidase-conjugated secondary antibody at 1 : 10 000 dilution. Chemiluminescence detection was performed using ECL Select (GE Healthcare, Chicago, IL, USA) and imaged using a Chemidoc XRS+ CCD imager (Bio-Rad, Hercules, CA, USA). Adjusted volume values were obtained, and standard curves of serially diluted recombinant proteins were used to estimate the amount of protein as described (Brown et al., 2008;Thamatrakoln et al., 2013).

Nonlinear model of host-virus nucleotide dynamics
A mathematical model was developed to characterize and disentangle the effect of host growth, host nucleotide recycling and de novo nucleotide synthesis on host-virus population dynamics. By extension to Wikner et al. (1993), the model resolves time variation in the concentration of nucleotides (nucleotides ml À1 ) within uninfected hosts (S), infected hosts (I ), internal viruses produced by host nucleotide recycling (P ), internal viruses produced by de novo nucleotide synthesis (D), and free (extracellular) viruses (V ). Direct measures of the individual nucleotide pools were not made given methodological limitations. Rather, the model was constrained with total cellular nucleotides (S + I + P) and free viral nucleotides (V ), which were converted from host and virus abundance respectively and known genome size (Supporting Information Table S1). We hypothesized that host growth, nucleotide recycling and de novo synthesis each have a distinct and quantifiable influence on the dynamics of total cellular and free viral nucleotides and that these distinct signatures would lead to constraint on the light dependence of nucleotide synthesis rates. Parameter definitions and units are in Table S1.
In Eqn 1, hosts synthesize nucleotides at rate l h . The linear interaction term /SV, controls the total flux of nucleotides from susceptible hosts and free, extracellular viruses to newly synthesized, intracellular viruses. The parameter v x represents the portion of this flux that originates from the virus, and 1 À v x represents the portion that originates from the host, with where V ind and S ind are the nucleotide contents of an individual virus and host respectively (Table S1). In Eqn 2, infected hosts synthesize nucleotides at rate l h , and nucleotides are either converted to viral progeny at rate v p l v or lysed at rate v p d. The parameter v p is included to mimic a delay in internal viral production and host lysis due to internal assembly of viral progeny. This delay parameterizes the latent period (i.e. the time during which the virus is replicating prior to host lysis). Unlike previous models that mimic the latent period with a time lag (e.g. Wang, 2006), v p forces a delay due explicitly to the internal depletion of host resources by the virus. Specifically, v p forces the rate of lysis to increase as host resources are converted to intracellular viruses: In Eqn 7, as host nucleotides I are converted to internal viral nucleotides P + D, v p increases from very low values and then saturates close to one, which in turn causes the rate of lysis v p d in Eqn 2 to increase, and then saturate at a fixed rate.
Nucleotides required for viral genome replication can be obtained in two ways, either through recycling of existing host nucleotides or de novo nucleotide synthesis. Recycling, by definition, leads to a depletion of host nucleotides, whereas de novo synthesis does not. Eqn 3 accounts for the accumulation of internal viruses by nucleotide recycling at rate v p l v , and the loss of internal viruses due to lysis at rate v p d. Eqn 4 accounts for the production of internal viruses by de novo synthesis at rate v p l 0 v , as well as accumulation of internal viruses due to adsorption, and losses due to host lysis at rate v p d. Having the rate of nucleotide recycling and de novo nucleotide synthesis dependent on v p accounts for the time it takes for a virus to co-opt the host biosynthetic machinery. Note that the sum of Eqns 3 and 4 represents a single mass balance for the total number of internal nucleotides, P + D. We separate P and D because it is only appropriate to count nucleotides generated via recycling P, with host nucleotides inferred with measured cell abundance and known nucleotide content (i.e. genome size; Table S1). Nucleotides generated via de novo synthesis are assumed to be 'invisible' until host lysis.
Eqn 5 assumes the free virus nucleotide concentration V is a balance between production of free viruses by host lysis at rate v p d, and uptake of viruses due to adsorption to host cells at rate v x /S. For all experiments, model solutions were obtained by numerically integrating Eqns 1-5 using initial conditions that matched the experimental observations.

Parameter constraint with the Metropolis algorithm
Our goal was to quantify the effect of host growth, nucleotide recycling and de novo nucleotide synthesis on total cellular (S + I + P) and free viral (V) nucleotide dynamics. We used the Metropolis algorithm (Metropolis et al., 1953) to test formally whether rates of nucleotide synthesis could be uniquely constrained with knowledge of total cellular and free-virus dynamics. We leveraged this constraint to explore the light dependence of viral replication. The Metropolis algorithm is a Monte Carlo type simulation that quantifies parameter uncertainty using a random walk to explore parameter combinations that explain observed dynamics. When the observations are unable to constrain model parameters, the random walk does not converge, and the algorithm predicts ever wider parameter distributions. When the observations are sufficient to constrain model parameters, the algorithm converges on unique parameter distributions.
The five model parameters controlling infection dynamics are virus adsorption /, host growth rate l h , host nucleotide recycling rate l v , de novo nucleotide synthesis rate l 0 v , and lysis rate d. The adsorption parameter was fitted using data from adsorption experiments. The remaining parameters h ¼ fl h ; l v ; l 0 v ; dg were fitted using population-level observations, labeled here D(t i ), of hostvirus dynamics. The Metropolis algorithm starts with an initial parameter set h current manually determined so that the model solutions are reasonably close to the observations. A new parameter set, h proposed , is then drawn from appropriate distributions. We used lognormal distributions with means matching the old distribution, and standard deviations chosen manually to ensure average acceptance ratios within the range 23-44% to promote efficient convergence (Gelman et al., 1996). The new parameters were chosen from lognormal distributions to prevent negative values. Solutions y(t i |h proposed ) to the model with the new parameter set are then calculated. For both the old and the new models, the sumof-squared difference between the model and the observations is divided by the measurement uncertainty of the observations, assumed here to be the average variance of each triplicate r 2 .
A likelihood function is defined in the following way: A likelihood ratio is then calculated by taking the ratio of the new to the old likelihood function: If the likelihood ratio is larger than a random number r between 0 and 1, the new parameter set becomes h current , and the process is repeated. If the likelihood ratio is less than r, the proposed set of parameters is disregarded and the process repeated with the old parameter set. Whenever a new parameter set is adopted, those parameters are recorded. Over time, enough parameters are stored to estimate parameter distributions that minimize error between the model and the observations. The distributions represent the full range of parameters that give suitable fits with respect to uncertainty in the observations. By repeating the fitting exercise for both the light and the dark treatments, estimates of parameter distributions for l h , l v , l 0 v , and d were generated. All simulations were repeated numerous times from different initial parameter guesses to ensure consistent convergence. Parameters were assumed to be constrained by the data if the variance in the posterior distribution did not increase with respect to the number of iterations of the optimization algorithm. In some cases, this led to poor constraint; for example, in the continuous light treatments. By looking for differences in the fitted parameter distributions between different light treatments, we were able to discern whether key processes, namely de novo synthesis or recycling of host nucleotides, were dependent on light intensity.

Field sampling and statistical analysis
Emiliania huxleyi populations (and associated EhVs) were collected during the North Atlantic Virus Infection of Coccolithophores Expedition (http://www.bco-dmo.org/project/2136) using Niskin bottles on a 24-position rosette equipped with an SBE conductivity-temperature-depth profiler (Sea-Bird Scientific, Bellevue, WA, USA). Cell biomass (and associated viruses) was collected from c. 3-5 l of 200 lm mesh-filtered seawater onto large (142 mm diameter), 0.8 lm pore-size polycarbonate filters (Millipore) to minimize clogging. Filters were immediately submerged in 10 ml extraction buffer (100 mM Tris-HCl pH 8, 250 mM EDTA pH 8, 100 mM sodium chloride, and 1% sodium dodecyl sulfate) and homogenized at maximum speed on a Vortex Genie (Mo Bio, Carlsbad, CA, USA) for 10 min in the presence of 2 ml molecular-grade zirconium beads (equal amount of 100 lm and 400 lm diameter beads; OPS Diagnostics, Lebanon, NJ, USA). Filters were subjected to three freeze-thaw cycles via submerging in liquid nitrogen and thawing at 50°C, homogenized for 5 min after each cycle and stored at À80°C. Upon processing, filters were thawed and 5 ml was incubated for 1 h at 50°C with 100 lg Proteinase K, followed by phenol : chloroform : isoamyl alcohol, 25 : 24 : 1 v/v/v extraction. Nucleic acids were precipitated with two volumes of 100% ethanol and 0.2 M sodium chloride. Following centrifugation (20 min, 10 000 g), the DNA

Research
New Phytologist pellet was washed with 20 ml of 70% ethanol, centrifuged again (10 min, 10 000 g) and then resuspended in 19 TE buffer (pH 8). Trace impurities that could inhibit enzymatic amplification reactions were removed using the PowerClean ® DNA Clean-Up Kit (Mo Bio). Quantitative PCR using SYBR ® Green was used to quantify E. huxleyi (via cytochrome oxidase I) and EhV (via major capsid protein) (Coolen, 2011). Tenfold serially diluted genomic DNA (extracted from flow cytometry quantified E. huxleyi (RCC1216) and EhV (EhV86) was used to calibrate the cytochrome c oxidase I-and major capsid protein-specific quantitative PCR. Data were analyzed by linear regression in R using the lm() function and model fit verified with the plot(model) function. All data were log 10 -transformed for consistency.

EhV infection, the PPP, and photosynthesis
Genes associated with de novo nucleotide synthesis, namely those involved in the PPP, were found to be upregulated during infection of E. huxleyi . To determine whether these observed increases in transcript abundance translated into higher rates of nucleotide synthesis, we measured the enzymatic activity of key enzymes of the PPP during the early stages (< 24 hpi) of EhV infection. EhV infection arrested host growth within 24 hpi, concomitant with an increase in the production of extracellular viruses (Fig. S1). PPP activity in uninfected controls remained steady at c. 9 nM NADPH min À1 lg protein À1 (Table 1). In contrast, PPP activity increased in infected cultures by c. 1.6-fold between 4 and 14 hpi, reaching maximum levels that were > 2.5-fold higher than uninfected controls by 24 hpi. Given that the PPP shares enzymes with the Calvin cycle, upregulation of the former should result in the downregulation of the latter. Measured protein expression levels of RuBisCO (Fig. 1a), the rate-limiting enzyme of the Calvin cycle, showed that the induced PPP activity was coincident with a 50% decrease in the biochemical potential of the Calvin cycle. In contrast, PsbD (a proxy for PSII) protein expression was similar between control and infected cells (Fig. 1a), suggesting photosynthetic electron transport was not inhibited in infected E. huxleyi.
No significant difference was observed in the electron transport rate between control and infected cells 5 hpi (data not shown) or 24 hpi (Fig. 1b). We further tested the role of PSII in viral production by inhibiting PSII activity with DCMU (Fig. 1c). There was a c. 15% drop in cell abundance in infected cultures that was not observed in infected cultures without DCMU (Fig. 1c, inset). Associated viral abundance was reduced by c. 50% in infected cultures treated with DCMU (Fig. 1d), which was not due to lower host growth, as the decrease in host abundance in the presence of DCMU was less (c. 30%) than the accompanying decrease in viral production. The infection was carried out for an additional 48 h to assess whether viral replication in DCMU-treated cultures was delayed, but cultures failed to produce the equivalent number of viruses as untreated, infected cultures, even at 96 hpi.

Light critically regulates infection
Light heavily impacted virus adsorption to E. huxleyi cells. There was little adsorption when EhVs were added to E. huxleyi at the onset of the dark period (see the Materials and Methods section). Only upon transition to the light 10 h later was there notable adsorption, observed as a 78% drop in free EhV abundance within 2 h (Fig. 2a). By contrast, when EhV was added at the onset of the light period, a near immediate drop of 44% in free EhV abundance was observed (Fig. S2a), consistent with previous studies (Mackinder et al., 2009).
EhV production, defined as the number of EhVs measured upon lysis, was also light dependent. Comparative infection dynamics in the dark were performed after EhV was allowed time to adsorb to host cells in the light. Although uninfected E. huxleyi cultures did not grow in the dark (Fig. 2b), cell lysis and declines in host abundance were observed for both infected cultures regardless of light or dark treatment (Fig. 2b). However, the 71% lower virus production in the infected, dark treatment ( Fig. S2b) resulted in a 3.5-fold lower burst size (i.e. the number of viruses produced per E. huxleyi cell) compared with infected cultures grown on an LD cycle (Fig. 2c).
Under continuous light, E. huxleyi growth and F v /F m were similar to cultures grown on a LD cycle (Figs 2d, S2c), indicating cells were not physiologically stressed, consistent with previous observations (Nielsen, 1997). However, infected cultures under continuous light produced c. 50% more viruses (Fig. S2d), resulting in a 1.5-fold higher burst size (Fig. 2e). Infected cultures in continuous light also had nearly twofold higher PPP activity than LD infected cultures (average over the entire experiment was 4.7 AE 1.4 nM À1 and 2.5 AE 0.6 nM NADPH min À1 lg protein À1 respectively), with significant increases observed as early as 6 hpi under continuous light compared with 20 hpi under LD conditions (Fig. 2f,g).
We also tested how photon flux impacted host-virus interactions. The maximum steady-state growth rate l max at irradiance  (Fig. 3a). At all irradiance levels except 2000 lmol m À2 s À1 , F v /F m was similar, suggesting growth at an irradiance of 25 lmol m À2 s À1 was limited by light rather than impaired photosynthetic efficiency (Fig. 3a). The lower F v /F m at an irradiance of 2000 lmol m À2 s À1 suggests the decreased growth may have been due to impaired photophysiology. When infected at an irradiance of 25, 150 or 300 lmol m À2 s À1 , host lysis occurred c. 4 d postinfection (dpi; Fig. 3b). By contrast, lysis occurred within 2-3 dpi at an irradiance > 500 lmol m À2 s À1 (Fig. 3b). EhV production was maximal at intermediate irradiance (150 and 300 lmol m À2 s À1 , Fig. 3c) and was approx. five-fold lower at irradiances of 25 and 500 lmol m À2 s À1 , and 16-fold lower at irradiances of 1000 and 2000 lmol m À2 s À1 (Fig. 3c). When expressed as a burst size, the intermediate irradiance levels produced the highest number of viruses per host (Fig. 3d). This relationship was robust, being observed across a large number (n = 5-25) of discrete samples (Fig. S3).
Photoperiod and irradiance serve as a switch between de novo and recycled nucleotide synthesis Given the increased PPP activity concomitant with decreased RuBisCO expression, we hypothesized enhanced viral production in the light was due to a concerted strategy to increase nucleotide synthesis and viral genome replication, while maintaining host integrity. To test this, we developed a mathematical model to quantify the dependence of free virus population dynamics on the rate of de novo nucleotide synthesis (Eqns 1-5; Fig. S4), accounting for the potentially interacting and compensating effects of host growth and recycling of host material. Our premise was that host growth l h , nucleotide recycling l v and de novo nucleotide synthesis l 0 v each have distinct and constrainable effects on population dynamics. We leveraged the dependence of host-virus population dynamics on these internal processes to infer their light dependence.
In Fig. 4 we show the contrasting effect within the model of nucleotide recycling l v and de novo nucleotide synthesis l 0 v on population dynamics. Increasing the rate of nucleotide recycling leads to an earlier decline of the host population (Fig. 4a)

Research
New Phytologist lower viral production due to the premature lysis of the host (Fig. 4b). Increased de novo synthesis also leads to early host decline (Fig. 4c) but results in enhanced viral production (Fig. 4d). The early decline of the host with increased de novo nucleotide synthesis is a consequence of our assumption that the rate of lysis is influenced positively by the accumulation of  internal viral nucleotides (Eqns 3-5, 7). However, the concomitant increase in the rate of de novo nucleotide synthesis allows increased viral production. The contrasting dynamics in Fig. 4(a,  b) and Fig. 4(c,d) respectively provide a distinct signature that allows the model to make predictions about nucleotide recycling l v and de novo synthesis l 0 v , based on host and virus population dynamics.
To assess the light dependence of these rates, we first assessed whether the model with these assumptions was consistent with the observed host-virus population dynamics. We then asked, using a formal fitting procedure with explicit quantification of parameter uncertainty (Metropolis et al., 1953), whether there were clear differences in rates of recycling and de novo nucleotide synthesis between light treatments.
The model successfully captured host-virus nucleotide dynamics in cultures infected in the dark, on an LD cycle, and under continuous light (Fig. S5). For the dark and LD treatments, the Metropolis algorithm converged on unique parameter distributions (Fig. 5), demonstrating that the empirical measurements were sufficient to constrain model parameters in these conditions. The continuous light treatment had too few measurements to constrain parameter values (Fig. S5e,f).
The model output suggests that host nucleotide recycling and de novo nucleotide synthesis were both strongly dependent on light, and the contribution of nucleotides to viral genome replication by each pathway could be toggled by external light conditions. In dark treatments, the model predicted rapid nucleotide recycling, accompanied by relatively slow de novo synthesis (Fig. 5a,b). This prediction is due to rapid demise of host populations in the dark and low viral production (Figs 2b, 4a,b). Conversely, LD treatments showed slower recycling and more rapid de novo synthesis (Fig 5a,b), arising from delayed host lysis and a large burst of freevirions toward the end of the infection (Figs 2, 4c,d). Data are mean AE SE of duplicate cultures, except for data from light irradiance levels 500 and 1000 lmol m À2 s À1 , which are from triplicate cultures. Data shown are from a single experiment but are representative of two to four independent experiments. Note that gradient shading of bars, lines, and symbols reflects the relative light levels (from dark to light).
New Phytologist ( Fig. 4 Sensitivity analysis of infection dynamics to the rate of host nucleotide recycling l v and de novo synthesis l 0 v during the time-course of infection. When a higher rate of nucleotide recycling is assumed, (a) host abundance declines more rapidly and (b) free, extracellular virus abundance decreases. When a high rate of de novo synthesis is assumed, (c) host abundance declines more rapidly due to faster internal progression of infection and (d) viral abundance increases. The distinct signature of these parameters on host-virus population dynamics allows their light dependence to be quantified. Contrasting parameter predictions during LD vs dark treatments were insensitive to a wide range of virus adsorption rates /, suggesting predictions of de novo synthesis and recycling rates are robust to any uncertainty in our predictions of / (Fig. S6).
Repeating the fitting procedure for experiments at different light intensities revealed that the rate of nucleotide recycling and de novo synthesis were also strongly dependent on irradiance level. Nucleotide recycling was lowest at intermediate irradiance (150-500 lmol m À2 s À1 ) and highest at low and high irradiances (25, 1000-2000 lmol m À2 s À1 ; Fig. 5c). De novo synthesis was opposite, being the highest at intermediate irradiance and lowest at low and high irradiance (Fig. 5d). In some cases, predictions of recycling and de novo synthesis rates were poorly constrained by the experimental data (Fig. 5c,d), but the clear difference across light intensities in spite of this large error supports the hypothesis that key infection parameters are light dependent. Taken together, these model predictions demonstrate that slow recycling and rapid de novo synthesis combine to impact EhV production and that light can lead to a systematic shift between nucleotide recycling and de novo synthesis.

Light structures infection of natural populations
The termination of dense coastal and open ocean North Atlantic blooms of E. huxleyi, which can span 10 5 km 2 (Holligan et al., 1993;Tyrell & Merico, 2004) and last over a period of days to weeks, is commonly attributed to EhV infection (Bratbak et al., 1993;Vardi et al., 2012;Lehahn et al., 2014). We used the 2012 North Atlantic Virus Infection of Coccolithophores Expedition as a platform to extend our laboratory-based mechanistic findings and test the hypothesis that irradiance structures the infection of natural populations of E. huxleyi. We previously demonstrated these populations were at various stages of infection (Laber et al., 2018) using a suite of structurally distinct glycosphingo-and betaine-like lipids that represent functional biomarkers of the E. huxleyi-EhV infection process (Vardi et al., 2009(Vardi et al., , 2012Fulton et al., 2014;Hunter et al., 2015). Using targeted, quantitative PCR we quantified the intracellular abundance of E. huxleyiassociated EhVs (expressed as a virus-to-microbe ratio, VMR) at three to five depths (97 discrete samples) in the euphotic zone at 21 stations across the c. 2000 nautical mile cruise track (Figs 6, S7). At 18 of the 21 stations, there was a subsurface VMR maximum that corresponded to c. 1-10% of the surface irradiance I 0 . The relationship between VMR and percentage surface I 0 (Fig. 6b) at these stations was similar to that between burst size and irradiance level seen in laboratory cultures (Fig. 3d), in that there was an optimal percentage surface I 0 that supported the highest VMR; above and below that optimum, VMR decreased. Taking the entire dataset together, there was a significant positive relationship between VMR and depth (P = 6.19 9 10 À4 , slope 0.02, R 2 = 0.12; linear regression with log-transformed VMR and depth), despite the inherent variability in pooling across stations. These observations corroborate active, subsurface infection of these populations (Laber et al.,  Sheyn et al., 2018), with the degree of infection correlating with depth. We then analyzed physical (light, temperature, salinity, mixed layer depth) and chemical (phosphate, nitrate and nitrite, ammonium, dissolved oxygen) aspects of this depth distribution to determine which parameters were the most robust predictors of the VMR-depth relationship. Light was the only significant (P = 3.14 9 10 À4 ) and robust (R 2 = 0.13) predictor of VMR (Table 2; Fig. S8) across depth, whereas the other parameters were either not significant (P > 0.05) or only weakly (R 2 ranged from 0.04 to 0.06, often prone to the impact of outliers) related to VMR.

Discussion
Elucidating the cellular processes that fundamentally regulate EhV infection is essential for developing a mechanistic framework for interpreting and modeling algal host-virus interactions. Given their inability to replicate without a host, it follows that obtaining nucleotides, either through host genome recycling or de novo synthesis, represents the first and foremost challenge to viral production. As the primary mechanism for de novo nucleotide synthesis, an increased PPP activity during infection has been documented in a wide range of systems, including animals, plants, and cyanobacteria (Bissell et al., 1973;Sindel a r et al., 1999;Lachaise et al., 2001;Thompson et al., 2011). Widespread enrichment of genes associated with the PPP has also been observed in marine metaviromes (Enav et al., 2014).
Here, we show that EhV infection increases host PPP activity, consistent with increased PPP-associated gene expression . We suggest this induced activity allows EhV to meet the requirement for genome replication that cannot be supported by host nucleotide recycling alone. With a genome size of c. 400 000 bp (Nissimov et al., 2012) and burst sizes ranging from 500 to 1000 virions per host (this study; Castberg et al., 2002), EhV201 would conservatively require 2 9 10 8 nucleotides for replication. Even if all 1.42 9 10 8 nucleotides of the E. huxleyi genome (Read et al., 2013) were recycled, EhV would still require an additional 6 9 10 7 nucleotides to achieve a mature burst size. This disparity becomes even more pronounced when the discrepancy between the guanine-cytosine content of E. huxleyi (66%) and EhV (40%) is taken into account. Although the plastid genome could also provide a source of recycled nucleotides, our observations of high PsbD expression (a gene encoded by the plastid genome) in infected cultures argues against plastid genome recycling.
The importance of de novo nucleotide synthesis is further supported by reduced RuBisCO protein levels and maintenance of electron transport rates and PSII activity. By diverting photosynthetic energy toward the PPP, viruses may enhance nucleotide synthesis while preserving energy that would otherwise be spent fixing unnecessary carbon. While functionally equivalent to similar observations in cyanophage (Lindell et al., 2004;Thompson et al., 2011), the mechanism by which EhV maintains host photosynthesis while decreasing RuBisCO is unclear.
Light-dependent adsorption and entry of EhV to host cells is conceptually similar to cyanophage (Cs eke & Farkas, 1979;Kao et al., 2005;Jia et al., 2010), and we hypothesize this dependence is related to the role lipid rafts play in facilitating EhV entry (Rose et al., 2014). Lipid rafts, membrane microdomains that sense extracellular stimuli and activate various signaling cascades through protein-protein interactions, have been implicated as entry sites of pathogens and viruses in other systems (Neilan et al., 1999;Chazal & Gerlier, 2003;Hajishengallis & Lambris, 2011). Proteomic analysis of purified lipid rafts from infected E. huxleyi cells suggests EhVs utilize a similar entry mechanism (Rose et al., 2014). Intriguingly, E. huxleyi lipids rafts were also found enriched in light-harvesting and Chla/b binding proteins (Rose et al., 2014). Although a role for lipid rafts in light capture is unknown, we posit that light regulates lipid raft production and thereby functionally limits entry of EhV during periods of darkness.
Viral replication and production were also impacted by light, with fewer viruses produced when infected cultures were maintained in the dark. As obligate photoautotrophs, prolonged darkness represents a significant physiological stress, and the lack of viral production may simply be due to host senescence. However, if cell mortality was significant, we would expect a dramatic decrease in host abundance. Rather, we observed no net change in cell abundance, suggesting either equal growth and mortality or that cells were neither dividing nor dying. By contrast, when host cultures were maintained under continuous light, viral production was enhanced, further supporting the hypothesis that light is required for maximum viral production. In addition, elevated basal PPP activity in continuous light cultures compared with LD cultures suggests increased EhV production is supported by the availability of additional nucleotides through enhanced de novo synthesis. Despite evidence that algal viruses have the potential to be a primary control on natural communities (Bratbak et al., 1993;Schroeder et al., 2003;Vardi et al., 2012;Laber et al., 2018;Sheyn et al., 2018), their impacts are not quantitatively assessed in ecosystem models resolving coupled interactions between phytoplankton, grazers, and bacteria. This is in part due to the lack of empirical data on key parameters, such as encounter and adsorption with hosts (/), replication rates (e.g. l v , l 0 v ) and host lysis (d) (Record et al., 2016). Our empirically based mathematical model of EhV production hypothesizes that viral production is influenced by the rate of nucleotide recycling and de novo synthesis. In the dark, when host metabolism is suppressed, viral production is diminished due to a combination of rapid nucleotide recycling and diminished de novo synthesis. When hosts grow rapidly at intermediate irradiance, reduced host nucleotide recycling can support enhanced viral production because host integrity and growth are preserved, leading to a larger pool of available host nucleotides for viral genome replication. At high light, viral production is diminished due to a combination of rapid nucleotide recycling and diminished rates of de novo synthesis. Together, these insights form a base with which to explore ecosystem effects of host-virus metabolism in contrasting aquatic light environments.
Our laboratory-based data suggest EhV adsorption, and subsequent conversion and/or synthesis of nucleotides, is enhanced in the light and that maximum viral replication occurred at mid-range irradiance levels. At high light, viral production may be inhibited by high reactive oxygen species production (Sheyn et al., 2016), whereas at low light there may simply not be enough de novo nucleotide synthesis to support high viral production. Alternatively, the shorter latent period (i.e. the time until host lysis) at the higher irradiance levels (> 500 lmol m À2 s À1 ) may itself represent an advantageous strategy whereby the virus induces host lysis to ensure virion release. This has been observed in bacteriophage, whereby the optimal lysis time is dependent on host abundance and physiology (Wang, 2006). We also cannot rule out a direct effect of light on the viruses. Although ultraviolet radiation is the major cause of viral decay in the ocean (Suttle & Chen, 1992;Noble & Fuhrman, 1997), photosynthetically active radiation has also been documented to facilitate decay (Traving et al., 2014;Wei et al., 2018). Exploring the effect of photosynthetically active radiation on viral abundance, infectivity, and adsorption needs to be further investigated to better understand the mechanism by which light impacts the latent period and viral production. Regardless, our data suggest virus activity in the upper ocean may be vertically structured by light. Within stations, the highest intracellular EhV abundance measured in natural populations occurred at depths corresponding to c. 1-10% of the surface irradiance I 0 , near the base of the euphotic zone, with lower abundances above and below this level. This was not due to the number of available hosts, given that VMR is normalized to host abundance; and importantly, these data are consistent with our laboratory findings that EhV replication is maximal between irradiances of 150 and 300 lmol m À2 s À1 , levels that would correspond to c. 7-15% surface I 0 on a sunny day (c. 2000 lmol m À2 s À1 ). When the entire dataset is pooled (97 discrete samples across 2000 nautical miles) and tested against a variety of environmental parameters, light was the only one found to have a significant relationship with VMR, with increasing light leading to lower VMR. The somewhat low, but significant, R 2 value (0.13) may be partially attributed to sampled populations being at different stages of infection (Laber et al., 2018), but nonetheless suggests a relationship between light and infection. Although we lose the resolution to disentangle the relationship seen at individual stations when we pool all of the samples, taken together with our laboratory data, these findings demonstrate that light can serve as a driving mechanism by which EhV infection can be vertically structured in natural populations.
Our laboratory-, field-, and modeling-based work form a mechanistic and empirical foundation on which to constrain key parameters regulating algal host-virus interactions. Our finding that light regulates critical aspects of EhV replication and production opens the door for mechanistic models of host-virus impacts on ecosystem function across a gradient of light environments. This, in turn, provides a novel and rigorous means to disentangle competing controls on phytoplankton populations, and the impact of viral infection on carbon cycling and nutrient dynamics.
quantitative PCR data from the North Atlantic field samples, FN performed the flow cytometry, BK analyzed the field data and contributed statistical analysis. KT, DT and KDB wrote the paper with contributions from BK, MJLC and MJF.

Supporting Information
Additional Supporting Information may be found online in the Supporting Information section at the end of the article: