Introduction

Ecosystem overgrowth by microbes, a process known as microbialization, is a major threat to coral reefs worldwide^1–2. Microbialized reefs are subject to a positive feedback loop in which proliferating algae generate excess dissolved organic carbon (i.e., sugar) that fuels the growth of increasingly pathogenic and copiotrophic microbes that thrive in nutrient-rich environments^1,3,4. This process exacerbates coral diseases and clears space for further algal growth^5–7.

Bacterial metabolism on degraded coral reefs is directed towards high-throughput rather than efficient carbohydrate utilization^1,8. For instance, while bacterial metabolism on healthy reefs is dominated by genes for the efficient but multi-step Embden-Meyerhof-Parnas (EMP) pathway, degraded reefs exhibit greater representation of genes for the inefficient but short Entner-Doudoroff (ED) glycolysis pathway ^1,9–11. This shift on degraded reefs enable bacteria to readily exploit ecosystem organic carbon enrichment to build up cellular stockpiles of central metabolites like glyceraldehyde-3-phopshate (G3P) and pyruvate (‘anaplerotic’ reactions) as well as use those central metabolite pools to produce cellular energy such as ATP, reductants like NADH, and amino and nucleic acid biosynthetic precursors required for bacterial over-proliferation (‘cataplerotic’ reactions)^1,6,10.

Viral predation could, in principle, preclude microbialization by killing proliferating microbes, as lytic infection and lysis rates are predicted by established predator-prey models to rise with host densities and growth rates^{12– 15}. However, viral communities switch from lytic to temperate infection during eutrophication and microbialization¹³. This transition towards non-lethal, temperate infection can facilitate further increases in microbial biomass due to suppressed lysis^13,16. Further, temperate viral communities encode higher frequencies of pathogenicity genes that could protect bacteria from consumption by eukaryotic grazers^13,17–19. These genes could facilitate microbialization by limiting protist predation, the major non-viral predation guild in the ocean^20,21. Altogether, rather than controlling bacterial populations, viruses may form a second ‘Temperate Ratchet’ feedback loop that traps ecosystems in persistent microbialization through suppressed lysis rates and protist grazing on pathogenic copiotrophs^13,22.

In addition to these population-level effects, viruses may also exert gene-level effects on microbialization by reshaping bacterial metabolism. Viruses shuttle a broad array of metabolic genes between hosts via horizontal gene transfer, thereby altering bacterial metabolism and ecology^23–26. These include central carbon metabolism genes implicated in the metabolic switching of bacterial communities on microbialized reefs, such as the pentose phosphate pathway (PPP) and the ED and EMP glycolysis pathways^1,27. In lytic viruses, these pathways support virion synthesis by converting general cellular metabolic pools like G3P and pyruvate to specialized metabolic precursors (i.e., they are cataplerotic reactions). For example, the PPP provides nucleic acid and amino acid precursors, whereas EMP generates cellular energy like ATP and NADH^26,28,29.

However, these genes can also be co-opted by bacteria during infection. As a result, while viral metabolic genes can combat microbialization by promoting the production of lytic viruses that would suppress bacterial abundance, they are also capable of exacerbating reef decline by enhancing bacterial proliferation. Together, these processes introduce a novel viral-mediated metabolic mechanism of reef decline or resilience beyond lytic-to-lysogenic switching.

However, the role of viral-encoded metabolic genes in coral reef microbialization remains unexplored. To address this gap, we leveraged 19 metagenomes of virus-like particles (VLPs; viromes) used in several prior publications^10,13,30, from Pacific coral reefs spanning a gradient from healthy to degraded states and from lytic to temperate infection dominance, to characterize changes in the frequencies of viral metabolic genes across a viral lifestyle spectrum. We found that temperate viral communities were enriched in anaplerotic central carbon metabolism genes, particularly those associated with copiotrophic growth like Entner-Doudoroff glycolysis and the generation of central metabolite pools that can be used for diverse cellular processes³¹. Lytic communities had higher levels of cataplerotic non-oxidative pentose phosphate pathway genes and oxidoreductases that generate specialized metabolic precursors, cellular energy and redox potential. Incorporating our findings into a conceptual model revealed that lytic viral metabolism can enhance viral production and microbial surveillance on healthy reefs (“viralization”)³², while temperate viral metabolism may accelerate microbialization in degraded systems. These findings implicate viral metabolism in feedback loops that either enhance reef resilience or exacerbate decline. Because viral lifestyle switching is a known function of cellular physiological state, with elevated nutrient availability driving non-lethal temperate infection, while nutrient limitation promotes lytic infection, our work suggests that metabolic management, such as limiting nutrient inputs, could promote rapid recovery of degraded reefs.

Methods

Metagenome sampling and processing

Metagenomes were sampled from virus-like particles (VLPs) from coral reefs across the central Pacific Ocean (n = 19; see Supplementary Table 1 for sample details) and samples processed as described previously^13,33. For each viral metagenome, approximately 60-100 L of seawater were pumped from the reef surface into acid-washed, triple-rinsed low-density polyethylene bags and concentrated to < 500 mL using a 100 kDa tangential flow filter. Concentrates were then 0.45 µm-filtered to remove bacteria and 0.5 % final concentration chloroform was added to destroy any residual cells and extracellular vesicles³⁴. Samples were stored at 4 °C until subsequent processing. The presence of concentrated VLPs was confirmed by epifluorescence microscopy of Sybr Gold stained viruses on anodisc filters³³.

VLPs were purified from concentrates using a cesium chloride step gradient approach^13,33,35. The absence of cells in step gradient outputs was confirmed by microscopy. Purified samples were then DNAse I-treated at 37 °C for 2 hours and denatured at 65 °C for 15 minutes to remove extra-virion free DNA (peDNA; protected environmental DNA)³⁶. Viral DNA was extracted using the formamide/chloroform isoamyl alcohol technique^13,33,35. The absence of contaminating host DNA in all samples was confirmed by the lack of PCR amplification of the universal 16S rDNA gene using primers 27F (AGA GTT TGA TCC TGG CTC AG) and 1492R (TAC GGY TAC CTT GTT ACG ACT T)^13,33,35. Note that our analysis is based on correlations across many samples, minimizing the impact of any contamination on our findings. VLP DNA was amplified using the Linker Amplified Sequencing Library approach¹⁸ prior to sequencing on an Illumina MiSeq platform.

Bioinformatic processing

Low-quality reads < 100 bp in length and with mean PHRED quality scores of < 25 were excluded using Prinseq³⁷. Samples were then dereplicated with TagCleaner³⁸, human contaminants were removed using DeconSeq³⁹, and sequences were posted to MG-RAST (www.mg-rast.org; see Data Availability section). Accession numbers and summary statistics for the viral metagenomes used in this paper are shown in Supplementary Table 1.

Annotating metagenomes

Viral metagenomes were annotated by DIAMOND alignment against the SEED subsystems database on MG-RAST v4 using the default MG-RAST parameters as a very commonly used balance of accuracy and specificity⁴⁰: e-value = −5 (annotated in MG-RAST as e-value = 5), 60 % identity, > 15 base length, a minimum abundance of 1, and all analyses conducted using representative hits (February 2020)^13,41. Level 1 Gene categories with average frequencies below 0.5 % were omitted. Temperateness of each viral metagenome was assessed as the frequency of genes annotated in the Level 3 SEED category of “Phage_integration_and_excision” (i.e., how common genes involved in integration and excision were in each metagenome)¹³. The frequency of genes in high-order classification (e.g., SEED level 1 with categories shown in Figure 1) was calculated by summing the frequency of the genes making up each category. All analyses were normalized to account for different sequencing depth between samples. This was done by calculating gene frequency as the number of hits to a given gene divided by all known genes in that sample (“percent known”)¹⁴, making all analyses comparable and unaffected by the effects of varied sequencing depth between samples.

Figure 1:

Pathway annotation

Pathway visualizations were guided by KEGG maps (www.genome.jp/), pathway-specific Wikipedia pages (www.wikipedia.org/), and enzyme Wikigenes entries (www.wikigenes.org/). Pathway maps were drawn using EC Number names of each enzyme (Table 1).

Table 1:

Anaplerotic and cataplerotic gene categorization

Genes were categorized based on whether they (i) generated central metabolite pools like glyeraldehyde-3-phopshate (G3P) and pyruvate (‘anaplerotic’ or ‘filling up’ reactions) that are flexibly used in multiple pathways by cells, or (ii) used these central metabolite pools to generate cellular energy or metabolic precursor products that are used for specific purposes (i.e., the products are ‘used’ metabolically; known together as ‘cataplerotic’ or ‘drawing down’ reactions), or (iii) were neither anaplerotic nor cataplerotic (‘neither’). Note that although anaplerotic and cataplerotic reactions often refer to steps in the TCA, we are using the term more generally here³¹.

Statistical analyses

The correlation between gene frequency, either at a Level 1 SEED category or individual gene level, and viral community temperateness were assessed using Spearman’s Rho (ρ_s), a non-parametric rank-based measure of how monotonic the relationship between x- and y-values is (i.e., it is a non-parametric correlation coefficient). For supplementary diversity and photosynthesis gene analyses we estimated ρ_s using the R cor.test() function with exact = FALSE to avoid ties. For our main analyses shown in Table 1 and in Figures 1, 2, 3, and 4, we estimated ρ_s values using the stat.spearmanr() function in the Python scipy.stats package with bootstrapping to ensure robust median ρ_s assessment over 1000 iterations of bootstrapping with replacement (see Supplementary Figure 2 for resulting distributions for central carbon metabolism genes). Confidence intervals of the median were assessed as the values corresponding to the 2.5th and 97.5th percentiles of the median values obtained from bootstrapping (95 % CI). For clarity, 95 % confidence intervals are listed in square brackets, as ρ_s [95 % CI minimum, 95 % CI maximum].

Figure 2:

Figure 3:

Bootstrapping code can be found at https://github.com/hopefulmonstersucla/viral-metabolism/tree/main/code. Note that correlations are significant at p < 0.05 when 95% bootstrap confidence intervals exclude zero, equivalent to classical null hypothesis testing.

Diversity metrics were assessed using the R vegan package, with the diversity() command used for Shannon diversity (natural log-transformed; H’). Median ρ_s at the pathway level were calculated as the median value of all of the ρ_s values for each enzyme in the pathway (i.e., they are a simple overall median).

Qualitative model

We created a conceptual model to demonstrate qualitative trends in bacterial, viral, coral, and algae abundance stemming from the changes in viral metabolism found here (Figure 5). Note that this model is not intended to be quantitative and was designed aid our understanding of trends in organismal abundance in response to the incorporation of viral metabolic capacity. Parameter values were implemented as high or low relative values to reflect known processes on healthy and degraded reefs (Supplementary Table 2 for parameter values). For example, high rates of disease^7,42 and bacterial metabolism¹ were used in degraded coral reef simulations, while high lysis rates were used in healthy systems^13,16.

Bacterial, viral, coral, and algae populations were initialized at 0.5 arbitrary units. The model was implemented using an iterative loop over six iterations (i.e., time in arbitrary units). This number of iterations was chosen as a balance of having enough iterations to see qualitative population trends, but without the need to make a more complicated model to generate long-term dynamics. At the beginning of each time step, new abundance values for each species (B_new, C_new, V_new) except algae were calculated based on the previous values (A_new). These updated values were applied at the end of each time step. Algae abundance was calculated last, occupying any remaining share of the benthos not occupied by coral. Once a population reached or fell below zero, it was excluded from further dynamics and not allowed to re-enter the ecosystem.

Code can be found at https://github.com/hopefulmonstersucla/viral-metabolism/tree/main/code.

For the ‘no viruses’ model shown in Figures 5a and 5d, we used the following equations, where terms with changed parameter values between Figures 5a and 5d are in bold (see Supplementary Table 2):

In Equation 1, bacteria (B) proliferate due to carbon provided by algae, with carbon supply itself a product of algal population (A) as well as the rate of delivery (a). To make the model as conservative as possible, we did not vary the rate of carbon delivered to bacteria by the algae, although it likely is higher on degraded reefs and enhances bacterial grow more than shown by the model⁹. Bacterial proliferation is also enhanced by metabolic changes (m) reinforced by bacterial population size on degraded reefs¹. Note that, in the absence of viruses the bacteria have no loss processes, including natural mortality, even though their proliferation may be constrained by a lack of algae to support growth.

In Equation 2, coral populations (C) proliferate as a product of coral population size and growth rate (c). Coral growth is limited by algal population size and allelochemical inhibition (i)⁴³ as well as by bacterial pathogenicity causing coral disease (d)^7,44. To make the model as conservative as possible, we did not alter the direct inhibitory effect of algae on coral across healthy and degraded reefs, although it is enhanced on unhealthy reefs^43,44.

In Equation 3, algal population size (A) is solved in the last step of each iteration as the remainder of the benthos not occupied by coral (1 – C) to account for the fact that the benthos is a finite space.

Note that Equations 2 and 3 are used for all iterations of the model with or without viral lysis and metabolism.

For the ‘with lytic viruses’ model shown in Figures 5b and 5e we used the following equations, with bold terms reflecting varied parameter values and with new terms compared to Equations 1 and Equation 2 underlined:

Building on Equation 1, lytic viruses are introduced in Equation 4 as a mortality term for bacteria as an outcome of viral population size (V) and lysis rate (k).

A viral population (V) is added to the model in Equation 5, where viruses proliferate at burst size (k) from lysis if bacteria. Note that k changes between healthy and degraded reef scenarios, consistent with Piggyback-the-Winner dynamics suppressing lysis on degraded reefs¹³.

For the ‘with viral lysis and metabolism’ model shown in Figures 5c and 5f we used the following equations, with bold terms reflecting varied parameter values and with new terms compared to Equations 4 and 5 underlined:

This model builds on Equation 4 by adding viral metabolic function (t) that enhances bacterial proliferation in degraded reef scenarios associated with temperate Piggyback-the-Winner dynamics in Equation 6¹³. Equation 7 builds on Equation 5 by adding lytic functions that enhance virion production in healthy reefs (l).

Results

Hallmarks of viral temperateness

To assess how viral metabolism across the lytic-temperate spectrum could affect coral reefs, we examined how the fraction of sequenced reads that mapped to known genes in broad metabolic categories (SEED Database Level 1 annotation) changed with community temperateness. Building on prior work¹³, we used the frequency of integrase and excisionase genes as a proxy for community temperateness (i.e., hallmarks of temperateness) in viral communities across the Central Pacific Ocean (Supplementary Table 1). As a measure of how consistently or monotonously gene frequency and sample temperateness covary, Spearman Rho (ρ_s) values reflect the strength and sign of their correlation. In this analysis, temperate-associated genes have positive correlations (ρ_s > 0) while lytic-associated genes have negative correlations (ρ_s < 0. For clarity, 95 % confidence intervals are listed in square brackets as ρ_s [95 % CI minimum, 95 % CI maximum], and correlations where the confidence intervals exclude zero are significant at the p < 0.05 level.

Gene names and corresponding pathways, EC numbers, median ρ_s and 95 % CIs, gene category (anaplerotic vs cataplerotic), and reports of the gene in viruses are shown in Table 1. Figure 1 and Figure 2 display the sample-level patterns of temperateness vs gene frequency in all Level 1 SEED categories and in central carbon metabolism functions, respectively. The correlations shown in Figure 2 are represented in a metabolic pathway map in Figure 3 and analyzed according to Table 1 categories in Figure 4. Finally, the implications of our central findings are incorporated into a schematic ecosystem model shown in Figure 5.

Figure 4:

Figure 5:

Temperate viral communities are enriched in metabolic genes

The frequency of genes in the Phages, Prophages, Transposable Elements, and Plasmids category declined with increasing temperateness of viral communities (Figure 1; ρ_s = −0.81 [−0.95, −0.58]). In contrast, the frequency and diversity of ‘bacterial’ metabolic genes that encode known cellular functions increased in temperate viral communities (Supplementary Figure 1; ρ_s = 0.63 and 0.81, respectively; c.f., genetic diversity declines with temperateness when unknown genes are included¹³). Altogether, this analysis indicates that temperate viral communities are more metabolically similar to bacteria than are lytic communities.

Functional homologues as viral lifestyle markers

Analogous to taxonomic markers in cells, where the prevalence of class I and class II homologues of fructose bisphosphate aldolase EC4.1.2.13 genes relates to bacterial taxonomy⁴⁵, the divergent correlations in these genes across a viral temperateness spectrum may reflect viral life strategies (EC 4.1.2.13 I ρ_s = −0.65 [−0.86, 0.33] while EC4.1.2.13 II ρ_s = 0.56 [−0.18, 0.83]; see Table 1 for summary statistics).

Photosynthesis genes in lytic communities

The decline in ‘Phages, Prophages, Transposable Elements, and Plasmid’ genes from ∼ 50 % of the known genes in lytic viromes to ∼ 5 % in temperate viromes caused the gene frequency in almost all other categories to increase (as total gene frequencies sum to 100 %; 17 of 22 categories had ρ_s > 0.5; Figure 1). A notable exception is Photosynthesis genes: these had higher frequencies in lytic communities vs. temperate communities in the Level 1 analysis (ρ_s = −0.78 [−0.92, −0.53]; Figure 1). Photosynthesis genes, including rhodopsins, were also lytic-associated at the individual-gene level, where 16 out of 20 photosynthesis genes had negative ρ_s values (Supplementary Table 3). Finally, photosynthesis-related electron transport (Pet) genes that couple Photosystems II and I are absent from the viromes (Supplementary Table 3).

Carbohydrate metabolism genes

Given the high prevalence of carbohydrate genes in the viromes, their elevated frequency in temperate communities (Figure 1; ρ_s > 0), and their ecological importance, we next focused on genes comprising central carbon metabolism^25,26 Embden-Meyerhof-Parnas (EMP) and Entner-Doudoroff (ED) glycolysis pathways, the pentose phosphate pathway (PPP) and the Tricarboxylic Acid Cycle (TCA; Figure 2). Note that for this and subsequent analyses we retained only sequences with hits to SEED database entries with high-confidence Enzyme Commission (EC) annotations⁴⁶.

Entner-Doudoroff genes are enriched in temperate viromes

Few patterns were readily apparent from the ρ_s values for genes in each pathway (Figure 2; Supplementary Figure 2). When viewed as a pathway map, though, the ED pathway was more clearly enriched in temperate viral communities (i.e., the red arrows running through the center of Figure 3). This map identified the ED pathway as the only pathway that changed consistently across the lytic-temperate spectrum (Figure 4a), and was exclusively over-represented in temperate viral communities (i.e., ρ_s > 0; EC 1.1.1.49 ρ_s = 0.26 [−0.24, 0.68], EC3.1.1.31 ρ_s = 0.43 [0.03, 0.77], EC4.1.2.14 ρ_s = 0.67 [0.36, 0.90], and EC4.2.1.12 ρ_s = 0.71 [0.47, 0.89]). Altogether, this gave the ED pathway an overall median ρ_s = 0.55 in contrast to overall median ρ_s = 0.03 for EMP glycolysis, ρ_s = 0.175 for gluconeogenesis (GNG), ρ_s = 0.03 for the PPP, and ρ_s = −0.10 for the TCA (see Figure 4a).

Anaplerotic temperate metabolism

Given the role of ED glycolysis in generating central metabolite pools for use in diverse pathways (i.e., ED is an anaplerotic pathway), we investigated whether temperate viral communities were enriched in anaplerotic genes more generally (see Table 1 for a list of anaplerotic and cataplerotic genes)³¹. Temperate viral communities were, indeed, enriched in anaplerotic genes, with an overall median ρ_s = 0.44 for anaplerotic reactions and only one gene even weakly lytic associated (6-phosphogluconate dehydrogenase EC 1.1.1.44 from the PPP; ρ_s = −0.06 [−0.54, 0.45]; Figure 4b).

Amino acid precursor genes are enriched in lytic viromes

In contrast to the ED-enrichment of temperate communities, the pathway map does not show many areas of contiguous lytic-associated genes (Figure 3). However, the enrichment of non-oxidative PPP genes in lytic communities, especially transaldolase EC 2.2.1.2 (ρ_s = −0.49 [−0.87, 0.02]). This enzyme consumes the central metabolite pool of glutaraldehyde-3-phosphate (G3P) to produce fructose 6-phosphate and erythrose 4-phosphate. Given that other fructose-6-phosphate producing genes are not similarly enriched in lytic communities (e.g., EC 2.2.1.1 in the PPP ρ_s = −0.08 [−0.52, 0.35], EC5.3.1.9 in EMP ρ_s = 0.45 [0.12, 0.73], and EC3.1.3.11 in gluconeogenesis ρ_s = 0.24 [−0.20, 0.66]), this suggests that the primary outcome of the non-oxidative PPP in lytic viral communities is the cataplerotic consumption of G3P for the generation of erythrose-4-phosphate, a precursor for amino acid biosynthesis⁴⁷.

Redox genes are over-represented in lytic communities

Given this cataplerotic focus in lytic communities, we investigated whether lytic communities were more strongly associated with individual cataplerotic reactions regardless of pathway rather than with whole pathways. Focusing on the distribution of anabolic redox steps across all the central carbon metabolism pathways, we found that NADPH-generating reactions were not consistently associated with either lytic or temperate communities (EC1.1.1.44 ρ_s = −0.06, EC1.1.1.42 ρ_s = −0.19, and EC1.1.1.49 ρ_s = 0.26, giving a median ρ_s = −0.06). In contrast, NADH-producing oxidoreductase genes were universally over-represented in lytic communities (i.e., all ρ_s < 0; median ρ_s = −0.13 across all NADH-producing reactions from EC1.1.1.37 ρ_s = −0.19 [−0.61, 0.29], EC1.1.1.41 ρ_s = −0.13 [−0.54, 0.34], EC1.2.1.12 ρ_s = −0.20 [−0.60, 0.26], EC1.2.4.1 ρ_s = −0.12 [−0.60, 0.36], EC1.2.4.2 ρ_s = −0.10 [−0.59, 0.42]; Table 1; Figures 2, 3, and 4). This included all redox genes in the TCA being lytic associated (i.e., ρ_s < 0, with an overall median ρ_s = −0.13) in contrast to the other, non-oxidoreductase genes which had a median ρ_s = 0.15, highlighting the specific overrepresentation of redox genes in lytic communities (Figure 4a).

Cataplerotic lytic metabolism

Given the association between lytic communities and cataplerotic reactions observed in the PPP and the TCA, we then investigated whether lytic communities are broadly enriched in cataplerotic genes. We observed that the overall median ρ_s of cataplerotic reactions was −0.125 (i.e., these genes were enriched in lytic communities) compared to ρ_s = 0.44 and ρ_s = 0.11 for anaplerotic reactions and reactions that are neither cataplerotic nor anaplerotic, respectively. This result highlights the association between lytic viral communities and cataplerotic metabolism.

Implications for Healthy Coral Reefs

To qualitatively describe the ecosystem effects of viral metabolism, we constructed a schematic model of algal-bacterial-viral-coral interactions, with all population sizes measured in illustrative arbitrary units (Figure 5). Parameterizing the model to healthy reef values (Figure 5a; note that parameterization of the healthy reef scenario largely insulates coral from the effects of bacterial overgrowth due to the small d value used; see Supplementary Table 2 for parameters and values)^1,7,8,13,42, the system became coral-dominated by t = 5, with coral-bacterial coexistence (coral and bacteria had increased from 0.5 a.u. initial values to 1.0 a.u. and 0.89 a.u., respectively, by t = 6). While the addition of viral lysis (Figure 5b) did not markedly change coral or algal dynamics, it led to the extinction of bacteria by t = 6 and to an ecosystem dominated by coral and viruses (coral and viruses had increased from 0.5 a.u. to 1.0 a.u. and 0.97 a.u., respectively, by the end of the simulation). Finally, adding in viral metabolism led to an earlier extinction of bacteria (by t = 5) and enhanced viral population size (viruses were 2.29 a.u. by t = 6), although coral and algal population dynamics were unaffected (Figure 5c). Altogether, the model recaptured the coral-dominated status of healthy coral reefs and showed that viral metabolism likely exerts a profound negative effect on bacterial populations and a strongly positive effect on viral populations on healthy reefs.

Implications for Degraded Coral Reefs

In contrast to healthy systems, our model suggests that degraded coral reefs would be overgrown with algae rather than coral (Figure 5). Note that the parameterization of the degraded reef scenarios leads to enhanced deleterious bacterial effects on coral health. Coral went extinct by t = 5 in the degraded scenarios without viruses and with viral lysis only (Figure 5d and e), as algal overgrew the benthos and bacterial proliferated to levels much higher than in any of the healthy scenarios (B = 2.75 a.u. and 2.64 a.u. in the no virus and lysis-only models, respectively, by t = 6). In contrast to bacteria, viral populations did not proliferate markedly in the viral-lysis model, increasing from 0.5 a.u. to only 0.64 a.u. by t = 6 (Figure 5e). Adding in viral metabolism that enhanced bacterial proliferation (t) allowed bacteria to proliferate even further (reaching 3.80 a.u.), supporting a modest rise in viruses (to 0.75 a.u.), and driving the coral to extinction sooner (by t = 4; Figure 5f). Altogether, our qualitative model suggests that viral metabolism supporting bacterial proliferation can enhance microbial overgrowth of reefs and promote shifts towards algal dominance of the benthos.

Discussion

The role of microbial overgrowth in coral reef decline is well established^1–2,7. It is further intensified by shifts in bacterial metabolism and viral lifestyles that promote bacterial proliferation^13,16. Here, we show that this process is further exacerbated by viral-encoded metabolic genes, whereas healthy ecosystems may be protected by viral lysis and metabolic genes that enhance virion production. Because temperate viral infection is tightly linked to host physiology and environmental conditions such as nutrient availability^13,48,49, targeting bacterial metabolic states through manipulation of nutrients could enable a shift in viral effects from degradative to protective, presenting a promising new conservation strategy.

Viruses become increasingly temperate on degraded reefs, following Piggyback-the-Winner (PtW) dynamics^13,22, in which abundant, rapidly proliferating bacterial hosts select for non-lethal, lysogenic infection. This switch in viral lifestyle suspends the top-down control exerted by lytic viruses, allowing bacterial populations to expand unchecked by lethal infection^13,49. Temperate infection not reduces lysis but also promotes horizontal gene transfer of pathogenicity genes⁵⁰, which can help bacteria evade predation by eukaryotic grazers and potentially compromise coral health^13,22. This forms a positive feedback loop wherein bacterial growth triggers viral lifestyle switching, which in turn facilitates further bacterial growth, establishing a ‘Temperate Ratchet’²² that locks ecosystems into ever-exacerbating bacterial overgrowth and coral disease¹³. Worse, here we show that the temperate viruses on degraded reefs also harbor metabolic genes, such as those for Entner-Doudoroff glycolysis, that can directly support bacterial proliferation and overgrowth of reefs¹, thereby enhancing the Temperate Ratchet²².

Healthy reefs, in contrast, appear to be guarded by lytic viruses. Lytic infection rates are driven by virus-host predator-prey encounter frequencies that are the outcome of host and viral abundances^12–14. As a result, as bacterial populations grow, so should lytic infections and lysis, precluding bacterial overgrowth^13,16. Further, the viral communities on healthy reefs do not harbor the high frequency of pathogenicity genes seen on degraded reefs, meaning that grazing and coral health are not being hindered by viral-encoded genes, which also helps to keep bacterial densities low⁷. Here we show that lytic viral communities on healthy reefs are enriched in metabolic genes associated with efficient energy production and biosynthesis, such as those for the Embden-Meyerhof-Parnas glycolysis pathway^1,9 and oxidoreductases involved in ATP generation and NADH reduction⁵¹. These metabolic features support robust virion production that can support reef ‘viralization’ through high viral abundances and high rates of virus-host encounters, infection, and lysis. In this way, lytic viral metabolism and reef viralization acts as a counterbalance to the Temperate Ratchet, helping to stabilize healthy reef ecosystems³².

Viral metabolism associated with the lytic-to-temperate shift likely exacerbates coral reef decline, as shown here. Building on ecosystem-scale processes like viral lifestyle switching^13,16, the metabolic capacity of temperate and lytic viruses can further drive reefs into two stable states: healthy reefs protected by metabolically enhanced lytic control of bacteria, and degraded reefs ratcheted into persistent microbialization under temperate infection and metabolism. Ultimately, degraded coral reefs experience a metabolic disorder, where excess nutrients trigger ecosystem-scale dysregulation through microbial and viral lifestyle and metabolic switching^1–2,6. However, the physiological basis of this switch also suggests that conservation measures targeting nutrient supply could undercut copiotrophic bacterial and viral metabolism^3,9,10. Our findings suggest that doing so could lead to a ‘re-viralization’ of the reef system, harnessing viral metabolic and lifestyle switching to enhance reef restoration and resilience.

Data Availability

Viral metagenomes can be found on MG-RAST under the accession numbers 4683670.3, 4683674.3, 4683677.3, 4683680.3, 4683683.3, 4683684.3, 4683686.3, 4683703.3, 4683704.3, 4683712.3, 4683718.3, 4683720.3, 4683731.3, 4683733.3, 4683745.3, 4683746.3, and 4683747.3 (see Supplementary Table 1 for library data summaries; www.mg-rast.org). Fasta files for each sample, statistical and modeling code, and annotations and simulations data can be found at https://github.com/hopefulmonstersucla/viral-metabolism.

Supplementary Tables and Figures

Supplementary Table 1:

Supplementary Table 2:

Supplementary Table 3:

Supplementary Figure 1:

Supplementary Figure 2:

Acknowledgements

This work was supported by a BIG Summer stipend from Simons Foundation (awarded to Alexander Hoffmann for CU and NY) as well as by NSF Award OCE-2201645 to BK that provided Hopeful Monsters Summer stipends to JK, MK, GD, NF, EG, AK, and IT. LWK was supported by NSF award OCE-2118617. RAE was supported by an award from the NIH NIDDK RC2DK116713, an award from the Australian Research Council DP220102915, and an award from the Gordon and Betty Moore Foundation number 9871 (Perpetual Viral Origins). JG was supported by NSF Biology Integration Institute award 2119968. ML was supported by an NSF Postdoctoral Research Fellowship in Biology award 2209377. The authors would like to thank the captain, crew, and staff of the NOAAS Hi’ialakai (R 334) and National Oceanic and Atmospheric Administration (NOAA) shore support, as well as the members of the Pacific Reef Assessment and Monitoring Program of NOAA’s National Coral Reef Monitoring Program.

Additional information

Funding

US National Science Foundation (2201645)

• Ben Knowles