1. Ecology
Download icon

Biophysical and physiological processes causing oxygen loss from coral reefs

  1. Cynthia B Silveira  Is a corresponding author
  2. Antoni Luque
  3. Ty NF Roach
  4. Helena Villela
  5. Adam Barno
  6. Kevin Green
  7. Brandon Reyes
  8. Esther Rubio-Portillo
  9. Tram Le
  10. Spencer Mead
  11. Mark Hatay
  12. Mark JA Vermeij
  13. Yuichiro Takeshita
  14. Andreas Haas
  15. Barbara Bailey
  16. Forest Rohwer
  1. San Diego State University, United States
  2. University of Hawaii at Mānoa, United States
  3. Rio de Janeiro Federal University, Brazil
  4. University of Alicante, Spain
  5. CARMABI Foundation, Curaçao
  6. University of Amsterdam, Netherlands
  7. Monterey Bay Aquarium Research Institute, United States
  8. Utrecht University, Netherlands
Research Article
  • Cited 0
  • Views 253
  • Annotations
Cite this article as: eLife 2019;8:e49114 doi: 10.7554/eLife.49114

Abstract

The microbialization of coral reefs predicts that microbial oxygen consumption will cause reef deoxygenation. Here we tested this hypothesis by analyzing reef microbial and primary producer oxygen metabolisms. Metagenomic data and in vitro incubations of bacteria with primary producer exudates showed that fleshy algae stimulate incomplete carbon oxidation metabolisms in heterotrophic bacteria. These metabolisms lead to increased cell sizes and abundances, resulting in bacteria consuming 10 times more oxygen than in coral incubations. Experiments probing the dissolved and gaseous oxygen with primary producers and bacteria together indicated the loss of oxygen through ebullition caused by heterogenous nucleation on algae surfaces. A model incorporating experimental production and loss rates predicted that microbes and ebullition can cause the loss of up to 67% of gross benthic oxygen production. This study indicates that microbial respiration and ebullition are increasingly relevant to reef deoxygenation as reefs become dominated by fleshy algae.

eLife digest

Rising water temperatures, pollution and other factors are increasingly threatening corals and the entire reef ecosystems they build. The potential for corals to resist and recover from the stress these factors cause ultimately depends on their ability to compete against fast-growing fleshy algae that can rapidly take over the reefs.

Living on the fleshy algae, the coral and in the surrounding water are communities of bacteria and other microbes that help maintain the health of the coral reef. Both corals and algae modify the chemical and physical environment of the reef to alter the composition of the microbial communities for their own benefit. Algae, for instance, release large amounts of sugars and other molecules of organic carbon into the water. These carbon molecules are then taken up by the bacteria, along with oxygen, to produce chemical energy via a process called respiration. This could cause the levels of oxygen in the water to decrease, potentially damaging the corals and creating more open space for the algae.

Previous studies have revealed how communities of microbes on coral reefs use organic carbon, but it remains unclear how they affect the levels of oxygen in the reefs. To address this question, Silveira et al. used an approach called metagenomics to analyze the bacteria in samples of water from 87 reefs across the Pacific and the Caribbean, and also performed experiments with reef bacteria grown in the laboratory.

The experiments showed that bacteria growing in the presence of fleshy algae became larger and more abundant than bacteria growing near corals, resulting in the water containing lower levels of oxygen. Furthermore, the fleshy algae produced bubbles of oxygen that were released from the water. Silveira et al. developed a mathematical model that predicted that these bubbles, combined with the respiration of bacteria that live near algae, caused the loss of 67% of the oxygen in the water surrounding the reef.

These findings represent a fundamental step towards understanding how changes in the levels of oxygen in water affect the ability of coral reefs to resist and recover from stress.

Introduction

Anthropogenic pressures are shifting the composition of reef primary producer communities from calcifying to non-calcifying organisms globally (Cinner et al., 2016; Smith et al., 2016). These changes occur as fleshy algae gain competitive advantage over corals through their interaction with heterotrophic microbes (Barott and Rohwer, 2012; Jorissen et al., 2016). High Dissolved Organic Carbon (DOC) release rates by fleshy algae stimulate overgrowth of heterotrophic bacteria (Haas et al., 2011; Kelly et al., 2014; Nelson et al., 2013). The exacerbated heterotrophic growth creates hypoxic zones at the coral-algae interface that kill corals (Haas et al., 2013a; Haas et al., 2014b; Roach et al., 2017; Smith et al., 2006). At the reef-scale, algae dominance stimulates heterotrophic bacteria in the water above the reef and leads to an increase in microbial biomass and energetic demands, described as reef microbialization (Haas et al., 2016; McDole et al., 2012; Silveira et al., 2015; Silveira et al., 2017). Overgrowth of heterotrophic bacteria creates a feedback loop of coral death, opening space for more algae overgrowth and microbial biomass accumulation (Dinsdale and Rohwer, 2011; Rohwer et al., 2002). While DOC-microbe relationships have been extensively described both experimentally and in situ, the oxygen fluxes within the microbialization feedback loop are not fully understood.

Differences in microbial responses to coral and algae dominance stem from the physiology of these primary producers. Calcifying organisms, including scleractinian corals and crustose coralline algae (CCA), invest 50% to 80% of their daily fixed carbon in respiration to sustain the energetic costs of calcification (Hatcher, 1988; Houlbrèque and Ferrier-Pagès, 2009; Tremblay et al., 2012). Fleshy macroalgae allocate 10% to 30% to respiration, and release up to 60% of their primary production as dissolved organic carbon (DOC) in the water (Cheshire et al., 1996; Crossland, 1987; Jokiel and Morrissey, 1986; Peninsula et al., 2007). Fleshy algae also allocate a higher proportion of their daily synthesized carbon to biomass compared to corals, sustaining high herbivory pressure (Duarte and Cebrián, 1996; Falkowski et al., 1984; Tremblay et al., 2016). Together, these processes are predicted to increase DOC-to-O2 ratios in coral exudates compared to algae, but experimental data have shown the opposite to be true: the ratio between DOC and O2 released by fleshy algae in bottle incubations is higher compared to corals (Haas et al., 2013b; Haas et al., 2011).

Heterotrophic microbes growing on algal exudates produce fewer cells per unit of carbon consumed compared to growth on coral exudates (Haas et al., 2011; Nelson et al., 2013; Silveira et al., 2015). Microbes growing on algae-dominated reefs shift their metabolism from classic glycolysis (Embden–Meyerhof–Parnas pathway, EMP) towards Pentose Phosphate (PP) and Entner-Doudoroff (ED) pathways (Haas et al., 2016). All three pathways consume glucose in a series of redox reactions that produce pyruvate, however they transfer electrons to distinct reduced coenzymes that act as intermediary electron transfer molecules (Klingner et al., 2015; Spaans et al., 2015). These reduced coenzymes have different fates in the cell, and as a result, the classic glycolic pathway generates more ATP and consumes more oxygen, while ED and PP leave a higher fraction of electrons accumulated as biomass (Fuhrer and Sauer, 2009). These differences predict that microbes found in microbialized reefs have different oxygen consumption rates when compared to healthy reefs (Flamholz et al., 2013; Stettner and Segrè, 2013).

Fleshy turf and macroalgae release up to three times more O2 in the surrounding seawater than calcifying organisms (Haas et al., 2011; Naumann et al., 2010; Nelson et al., 2013; Silveira et al., 2015). Yet, algae-dominated systems have consistently lower O2 standing stocks (Calhoun et al., 2017; Haas et al., 2013b; Martinez et al., 2012). Fleshy algae can produce O2 bubbles through heterogeneous nucleation resulting from O2 super-saturation at the algae’s surface (Kraines et al., 1996). The gaseous O2 in bubbles is not detected by dissolved O2 instruments, and may cause underestimation of autotrophy in oxygen-based primary productivity methods (Atkinson and Grigg, 1984; Kraines et al., 1996). While several studies recognize bubble formation on algal surfaces in benthic ecosystems, the fraction of photosynthetic oxygen lost as bubbles from coral reef primary producers has never been quantified (Freeman et al., 2018; Odum and Odum, 1955).

Here we test the hypothesis that O2 loss in coral reefs is a result of the high microbial oxygen consumption and the biophysical loss through ebullition from algae surfaces. The analysis of organic carbon consumption pathways encoded in bacterial metagenomes from 87 reefs in the Atlantic and Pacific combined with experimental quantification of cell-specific DOC and oxygen consumption showed that fleshy macroalgae and microbes remove larger amounts of photosynthetic O2 by ebullition and respiration compared to corals. As a result, the incubation bottles have similar to lower oxygen concentrations, providing a mechanistic explanation for the lower dissolved O2 observed on reef ecosystems shifted to algal dominance.

Results

Abundance of carbon metabolism genes across algal cover gradient

Different carbon consumption pathways employed by bacteria are associated with varying levels of oxygen consumption rates (Russell and Cook, 1995). To investigate the carbon metabolic pathways selected among reef microbes, 87 metagenomes from reefs across a gradient in algal cover were analyzed for the relative abundance of genes encoding rate-limiting enzymes of central carbon metabolism (Figure 1—source data 1). On coral reefs, metagenomic data reflect the strain-level genomic adaptation that occurs within hours, the timescale of residence time depending on the tides and current regime (Nelson et al., 2011). This selection occurs as offshore microbial communities are transported onto the reef environment and water masses get enriched with benthic exudates changing their biogeochemistry (Nelson et al., 2011; Kelly et al., 2014). In our dataset, microbial biomass in the reef boundary layer increased with fleshy algae cover (linear regression, p=6.25 x 10-8, R2 = 0.29).

A dimension-reduction random forest regression analysis was applied to genes encoding rate-limiting enzymes of central carbon metabolism, stress response and control genes, for a total of 23 variables (Supplementary file 1), using fleshy algae cover as predicted variable. Random forest is a non-parametric method that does not rely on normality, and along with other machine-learning approaches, is the method of choice for the identification of a robust subset of variables within complex meta-omics data (Li, 2015; Liu et al., 2019; Thompson et al., 2019). The relative abundance of these genes could explain 14.3% of the variance in fleshy algae cover. The genes with highest and significant explanatory power in the permutation test were phosphoenolpyruvate carboxylase, aspartate aminotransferase, oxoglutarate dehydrogenase, glucose 6-phosphate dehydrogenase and KDPG aldolase (Figure 1 displays the enzyme genes with highest explanatory power according to their increasing mean squared error: 53.54%, 22.03%, 23.99%, 26.08% and 17.89%, and permutation test p-values = 0.009, 0.01, 0.03, 0.01 and 0.03, respectively; Figure 1—source data 1).

Figure 1 with 3 supplements see all
Increase in genes encoding anabolic pathways and decrease in genes encoding carbon oxidation reactions with increasing fleshy algae cover in situ.

Relative abundances of genes encoding rate-limiting enzymes in reef metagenomes are plotted in relation to fleshy algae cover (sum of fleshy turf and macroalgae). Individual metagenomes are listed as rows and sorted by the fleshy algae cover, ranging from 15% to 94.1%. Enzyme gene abundance was scaled by column to allow between-enzyme comparisons as indicated by the z-score, where blue indicates low relative abundance and orange indicates high relative abundance. Only enzyme genes significantly predicting fleshy algae cover in the random forest analysis are shown. A complete list of enzymes and abundances is provided as Figure 1—source data 1.

Phosphoenolpyruvate carboxylase and aspartate aminotransferase are involved in anaplerotic routes, reactions that replenish the Krebs cycle when its intermediaries are diverted for biosynthesis. Glucose 6-phosphate dehydrogenase shunts glucose to both Entner-Doudoroff and Pentose Phosphate pathways, and KDPG aldolase is unique to the Entner-Doudoroff pathway. These four enzyme genes had a positive relationship with fleshy algae cover (identified by random forest by their increasing mean squared error, followed by permutation test with p-values p=0.009, 0.01, 0.01 and 0.03, respectively), and participate in pathways of incomplete carbon oxidation, that is, lower oxygen consumption. Oxoglutarate dehydrogenase relative abundance had an inverse relationship with fleshy algae cover (random forest increasing mean squared error, p=0.03). This enzyme catalyzes an oxidative decarboxylation step in the Krebs cycle, a pathway that produces most of the NADH that fuels oxygen consumption by oxidative phosphorylation. These results suggest that the bacterial community is being increasingly selected for pathways related to overflow metabolism and incomplete carbon oxidation at high algal cover, which would predict a decoupling between organic carbon and oxygen consumption rates by microbes.

The gene rpoB (RNA polymerase) was used as a control and did not vary in relative abundance across the algal cover gradient (Figure 1—source data 1). Genes involved in oxidative stress did not show significant relationships either. To test if the changes in enzyme gene relative abundances could be explained by taxonomic shifts of bacterial genera that consistently display larger genome sizes and duplications in the genes described above, the relationship between fleshy algae cover and taxonomic profiles at the genus and species levels was tested using random forest. The relative abundances of each taxon were calculated accounting for genome size (Cobián Güemes et al., 2016). Seven species significantly changed in abundance across the algae cover gradient (Figure 1—source data 1, Prochlorococcus marinus, Fibrisoma limi, Prolixibacter bellariivorans, Pelagibacter ubique, Ruegeria mobilis, Phaeobacter italicus, and Psychrobacter pacificensis). At the genus level, five taxa (Prochlorococcus, Thermotoga, Methanobacterium, Leuconostoc, and Rhodopirellula) changed their abundance with increasing fleshy algae cover. We searched if the taxa varying in abundance consistently display copy number variations in the genes that significantly vary with algae cover. The only taxon that could potentially explain the abundance trends in enzyme genes was Prochlorococcus, which lacks an oxoglutarate dehydrogenase gene but encodes all the other four enzymes. However, there was no significant relationship between the abundances of Prochlorococcus and oxogluratae dehydrogenase (p=0.26 in the linear regression of oxoglutarate dehydrogenase and log10-transformed Prochlorococcus abundance). Figure 1—source data 1 shows the relative abundances of the 20 most abundant species that together made up 97.8 ± 3.0% of the identified reads at species level in the metagenomes.

Bacterial DOC and O2 consumption

To test the hypothesis of a decoupling between organic carbon and oxygen consumption by heterotrophic bacteria predicted by the metagenomic analysis, cell-specific carbon and O2 consumption rates were obtained by incubating reef heterotrophic bacteria with coral and algae exudates (Figure 2—source data 1; Experiment one in Materials and methods). The amount of DOC consumed per cell was higher, that is, low cell yields per carbon consumed, in fleshy macroalgae treatments compared to coral and control treatments (Figure 2A, Kruskal-Wallis χ2(2)=13.7, p=0.001; Wilcoxon, p=0.005 and 0.008 for pairwise comparisons of algae vs controls and corals, respectively). However, the amount of O2 consumed per cell was not different between treatments (Figure 2B, Kruskal-Wallis χ2(2)=0.5, p=0.77).

Decoupling between microbial DOC and O2 consumption in fleshy macroalgae exudates.

Cell-specific carbon and O2 consumption data from Experiment 4: dark incubations of microbial communities in primary producer exudates. (A) Cell-specific DOC consumption. (B) Cell-specific O2 consumption. Primary producer treatments had a significant effect on specific DOC consumption only (Kruskal-Wallis p<0.05) and letters above boxes indicate p<0.05 for Wilcoxon pairwise tests with FDR correction. Orange indicates calcifying and green indicates fleshy organisms. Boxes depict the median and the first and third quartiles. Whiskers extend to 1.5 times the size of the interquartile ranges. Outliers are represented as dots above or below whiskers.

Bacterial abundances and cell volumes

Incomplete carbon oxidation and low oxygen consumption relative to organic carbon are hallmarks of increases in overflow metabolism and organic carbon accumulation as biomass. To test whether low O2-to-DOC consumption ratios in algae exudates are causing biomass accumulation, heterotrophic bacteria were incubated in coral, CCA and fleshy turf and macroalgae exudates and their biomass was quantified accounting for changes in both abundance and cell volume. The change in bacterial abundance in all primary producer bottles was higher compared to controls (Figure 3A, Kruskal-Wallis χ2(4)=20.6, p=0.0003, Wilcoxon p=0.04, 0.02. 0.02. and 0.02 for pairwise comparisons of coral, CCA, turf and macroalgae against control, Figure 3—source data 1). The change in abundance was higher in fleshy algae treatments compared to calcifying treatments (Kruskal-Wallis χ2(2)=19.1, p=6.9×10−5, Wilcoxon p=0.001 for pairwise comparison of fleshy vs calcifying). Microbial cell volume changed differently in treatments during incubations (Figure 3B, Kruskal-Wallis χ2(5)=1235.2, p<2×10−16, Figure 3—source data 2). Bacteria growing in control bottles showed no change in volume (Wilcoxon, p=0.56 for pairwise comparison of control vs T0h). Cell volume decreased when bacteria grew on coral and CCA exudates, but increased when growing on turf and macroalgae exudates, and showed no change in controls (Wilcoxon p<2×10−16 for all pairwise comparisons vs T0h and vs control).

Figure 3 with 1 supplement see all
Differences in microbial biomass accumulation in primary producer exudates.

(A) Changes in microbial abundance normalized by primary producer surface area. (B) Cell volume distributions. (C) Changes in total microbial biomass, accounting for abundance and cell volume, normalized by primary producer surface area. Primary producer treatments had a significant effect on all three microbial variables (Kruskal-Wallis p<0.05), and letters above boxes indicate p<0.05 for Wilcoxon pairwise tests with FDR correction. Orange indicates calcifying and green indicates fleshy organisms. Boxes depict the median and the first and third quartiles. Whiskers extend to 1.5 times the size of the interquartile ranges. Outliers are represented as dots above or below whiskers.

When taking abundance and cell volume into account, the change in total bacterial biomass was significantly different among treatments (Figure 3C, Kruskal-Wallis χ2(4)=21.6, p=0.0002, Figure 3—source data 1). Coral exudates decreased total bacterial biomass, while turf and macroalgae increased total bacterial biomass, and CCA lead to no change (Wilcoxon, p<0.05 for pairwise comparisons of coral, turf and macroalgae vs control). The change in biomass in the two fleshy algae together was greater than in the calcifying organisms together (Kruskal-Wallis χ2(2)=19.1, p=7.1×10−5; Wilcoxon, p=0.0005 for pairwise comparison of fleshy vs calcifying). This same pattern was observed in two independent experiments, one performed in the island of Curaçao, in the Caribbean (Figure 3) and one performed at the Hawaiian Institute of Marine Biology (HIMB, Figure 3—figure supplement 1), with distinct sets of primary producers, suggesting that this result may be broadly applicable to other reefs (Figure 3—source datas 1 and 2).

Oxygen production, consumption, and ebullition

While bacteria became larger and more abundant when growing on algae exudates, their oxygen consumption was lower than that predicted by their DOC consumption: the theoretical value for full carbon oxidation through aerobic metabolism is 1:1, changing due overflow metabolism, futile cycles, uncoupling, and other processes reviewed in Russell and Cook (1995). To resolve the oxygen budget, heterotrophic bacteria were incubated together with the primary producers in POP (Peripheral Oxygen Production) incubation chambers that allow to quantify O2 in both dissolved and gas fractions. Net dissolved O2 production was significantly different among primary producer treatments (Figure 4A, Kruskal-Wallis χ2(4)=19.5, p=0.0006, Figure 4—source data 1). Primary producers showed higher net dissolved O2 production compared to controls (Wilcoxon with FDR-corrected pairwise comparisons p=0.004, 0.01, 0.01, 0.004 for comparisons between controls vs coral, CCA, turf and macroalgae, respectively). Net dissolved O2 production was lower in macroalgae incubations, but not significantly different between fleshy and calcifying organisms (5.99 ± 2.29 and 6.33 ± 2.20 μmol cm−2 day−1 for fleshy and calcifying organisms, respectively, mean ± SE; Wilcoxon p=0.68). There was a significant difference between gaseous O2 production observed in most primary producer bottles, while no gas was observed in control bottles (Figure 4B, Kruskal-Wallis χ2(4)=30.2, p=4.361e-06, Wilcoxon p=0.03, 0.002, 0.002, and 0.002 for pairwise comparisons between controls vs corals, CCA, turf and macroalgae, respectively). Fleshy organisms produced significantly more gaseous O2 than calcifying organisms (0.42 ± 0.35 and 2.33 ± 1.35 μmol cm−2 day−1 for calcifying and fleshy, respectively, mean ± SE; Wilcoxon p=0.0001 for comparisons between fleshy vs controls and vs coral). Fleshy macroalgae had the highest fraction of net photosynthetic O2 (sum of dissolved and gaseous O2) released in the form of gas (37.33 ± 8.34%), followed by turf algae (13.78 ± 1.33%), CCA (10.19 ± 2.88%), and corals (5.00 ± 5.55%) (Figure 4C). The difference in the fraction of O2 accumulated as gas was significant among treatments (Kruskal-Wallis χ2(4)=29.2, p=7.067e-06), being higher in fleshy organisms (Kruskal-Wallis χ2(2)=26.3, p=1.904e-06; Wilcoxon p<9.5e-05 for fleshy vs calcifying).

Photosynthetic O2 loss as gas from fleshy macroalgae.

(A) Dissolved O2 production rates normalized by organism surface area. (B) Gaseous O2 production rates normalized by organism surface area. (C) Fraction of total photosynthetic O2 production in the form of gas. Solid squares correspond to Experiment 1 and triangles to Experiment 2. In A, the gray-filled triangles indicate replicates where dissolved oxygen was above the accuracy range of the probe used, and lower bound values (within probe accuracy range) were utilized. In C, the gray-filled triangles indicate the estimated values associated with lower bound values from A. Primary producer had a significant effect on all three variables (Kruskal-Wallis p<0.05), and letters above boxes indicate p<0.05 for Wilcoxon pairwise tests with FDR correction. Orange indicates calcifying and green indicates fleshy organisms. Boxes depict the median and the first and third quartiles. Whiskers extend to 1.5 times the size of the interquartile ranges. Outliers are represented as dots above or below whiskers.

In the POP experiments, the oxygen in the gas phase was close to equilibrium with the dissolved oxygen, and in some of the fleshy algae and CCA, the dissolved oxygen concentration was above accuracy limits of the probe. During the course of the experiment, we observed that a headspace was formed by bubbles that were formed on the algae surfaces and rose to the top of the incubation bottles. To quantify ebullition rates and to test whether this effect would be observed in an open system, bubble production was quantified in open-tank experiments using image analysis (Figure 5A—source data 1). The green fleshy algae Chaetomorpha had the highest bubble production rates (10.3 ± 0.45 bubbles per min per dm2, mean ± SE; Video 1), followed by the fleshy red algae Gracilaria sp. (1.29 ± 1.21 bubbles per min per dm2). Bubble production by the two coral species analyzed was close to zero (0.1 ± 0.14 and 0.006 ± 0.02 bubbles per min per dm2 for Favia sp. and Montipora sp., respectively). Ebullition rate was significantly higher in fleshy compared to calcifying organisms (Kruskal-Wallis χ2(3)=349.46, p =<2.2e-16; Wilcoxon p<2e-16 for fleshy vs calcifying). Bubbles were observed forming on the algal surfaces, detaching and rising to the atmosphere starting as early as 20 min into the experiment. The ebullition rate in the open-tank experiment was within the range of ebullition rates estimated to be necessary to explain the volume and oxygen concentrations observed in the POP experiments (estimates indicated by the gray bar in Figure 5A). The dissolved oxygen concentration also increased through the course of the incubations (Figure 5B), however, there was no relationship between the rate of bubble production and the dissolved oxygen concentration (linear regression p=0.16, R2 = 0.04) or the change in oxygen concentration over time (linear regression p=0.27, R2 = 0.03).

Figure 5 with 1 supplement see all
Ebullition by fleshy algae.

(A) Rate of bubble production normalized by organism surface area in open-tank experiments. The green symbols indicate fleshy algae and the orange ones indicate scleractinian corals. The dotted line indicates the mean and the gray bar indicates the range of ebullition rates estimated to explain the amount of gaseous oxygen observed in POP experiment 2 (Figure 4). (B) Dissolved oxygen concentrations over time in the same open-tank experiments depicted in 5A. (C) Representative image of bubbles formed on the surface of the green algae Chaetomorpha sp. incubated in an open tank at 300 PAR. (D) Predicted loss of oxygen through microbial respiration and ebullition as a percentage of the gross benthic oxygen production in a model reef with varying algae and coral cover (x axis). Oxygen loss was estimated by a weighted linear model incorporating per-cell respiration rates shown in Figure 2, microbial abundances sustained by primary producers shown in Figure 3, and ebullition rates shown in Figure 5A.

Video 1
Representative video of open-tank experiment with Chaetomorpha sp. at 150 umol quanta m−2 s−1.

This specific video fragment was recorded 20 min after the beginning of the experiment.

The mean bubble diameter produced by the algae Gracialaria sp. was larger than that produced by the algae Chaetomorpha sp. (Figure 5C and Figure 5—figure supplement 1; 0.79 ± 27 mm in diameter and 0.64 ± 0.23 mm for Gracialaria sp. and Chaetomorpha sp., respectively, however these differences were not statistically significant). The incubation of these same algae species at higher PAR values (300 to 650 µmol quanta m−2 s−1, normally observed on a reef in a sunny day) yielded different bubble production rates: at 300 µmol quanta m−2 s−1, the algae Chaetomorpha sp. had the highest ebullition rates (Figure 5C and Video 2, 16.2 ± 1.9 bubbles per min per dm2). The rate decreased to 0.9 ± 0.04 at 650 µmol quanta m−2 s−1. The algae Gracilaria sp. also produced fewer bubbles per minute at 650 µmol quanta m−2 s−1 (0.9 ± 0.08).

Video 2
Representative video of open-tank experiment with Chaetomorpha sp. at 300 µmol quanta m−2 s−1.

This specific video fragment was recorded 7 hr after the beginning of the experiment.

Prediction of oxygen loss by microbial respiration and ebullition

To predict the proportion of the benthic gross oxygen production that is lost through microbial respiration and ebullition, a weighted linear model combined the cell-specific respiration rates from Figure 2, the microbial abundances sustained by each primary producer from POP experiments, and the rates of oxygen ebullition from Figure 5. Each of these rates was applied to an idealized reef of 1 m2 benthic surface covered by fleshy green algae and/or scleractinian coral, and the microbial community within 1 m3 of the water column above. Both the microbial respiration and ebullition rates increased with increasing algal cover in this idealized model (Figure 5D—source data 2). At 0% algal cover, 10.4% of the benthic oxygen production is consumed by microbes, compared to 35.9% at 50% algal cover, and 47.0% at 100% algal cover. The amount of oxygen loss by ebullition is predicted to increase from 1.0% at 0% algal cover to 14.0% at 50% algal cover and 19.7% at 100% algae cover. Combined, microbial respiration and ebullition are predicted to consume 50% of the gross oxygen production at 50% algal cover, and 66.7% of the oxygen production at 100% algal cover.

Discussion

Due to high per cell activity and relative size, a small change in bacterial biomass may represent large shifts in ecosystems energy allocation, biogeochemistry landscape, and trophic relationships (DeLong et al., 2010; McDole et al., 2012). In coral reefs, the global increase in microbial biomass was hypothesized to fuel positive feedbacks of coral mortality through the creation of hypoxic zones by microbial respiration (Altieri et al., 2017; Haas et al., 2016; Smith et al., 2006). The work presented here shows that a physiological transition to overflow metabolisms is the underlying mechanism of bacterial biomass accumulation observed in coral reefs. The loss of oxygen through ebullition facilitates this type of metabolism and adds to the effect of high microbial respiration that depletes oxygen in otherwise oxygen-rich, algae-dominated systems.

Bacterial overflow metabolism in algae-dominated reefs

The metagenomic results in Figure 1 show an increase in genes for pathways that shunt carbon to biosynthesis and incomplete oxidation, that is, low O2 consumption (Russell and Cook, 1995; Haas et al., 2016). The abundance of genes encoding enzymes involved in the Pentose Phosphate (PP) and Entner-Doudoroff (ED) pathways, increased with fleshy algae cover (Figure 1). These pathways are alternatives to the canonical glycolytic route for carbohydrate consumption and produce more NADPH than NADH as intermediate electron acceptors. These two reduced coenzymes have distinct cell fates, with NADH preferentially donating electrons to oxidative phosphorylation that consumes oxygen during ATP production (Pollak et al., 2007; Spaans et al., 2015). Thus, cells utilizing more NADPH pathways consume less O2 relative to carbon compared to cells utilizing more NADH, and store a larger fraction of the consumed carbon into biomass.

The tricarboxylic acid cycle (TCA or Krebs cycle), is a canonically oxidative path for complete decarboxylation of pyruvate. But the TCA cycle can work as a biosynthetic pathway by providing intermediates to amino acids, nitrogenous bases and fatty acids syntheses (Cronan and Laporte, 2013; Sauer and Eikmanns, 2005). TCA cycle stoichiometry is maintained through the resupply of these intermediates by anaplerotic reactions that use pyruvate (or phosphoenolpyruvate) (Owen et al., 2002). In the reefs analyzed here, the abundance of genes encoding anaplerotic enzymes increased, and genes encoding an oxidative decarboxylation step in the Krebs cycle decreased with increasing algal cover. These results indicate a shift towards the use of the Krebs cycle as an anabolic route (Bott, 2007; Sauer and Eikmanns, 2005). Anaplerotic activity increases in rapidly growing bacteria with high rates of amino-acid synthesis (Bott, 2007). These routes shunt organic carbon into biomass accumulation, as opposed to oxidation with electron transfer to O2 (Russell and Cook, 1995).

This overflow metabolism is analogous to the Warburg effect, where cells undergoing fast metabolism are limited by enzymes' catalytic rates, as opposed to substrate concentrations (Vander Heiden et al., 2009). This type of metabolism is classically described in cancer cells and fast-growing yeast, and consumes large amounts of organic carbon without complete oxidation to CO2, even in the presence of O2. In bacteria, the Warburg effect is a physiological response to changing proteomic demands that optimizes energy biogenesis and biomass synthesis under high energy supply (Basan et al., 2015). This biochemical transition occurs because the proteome cost of energy biogenesis by respiration exceeds that of fermentation. A switch to overflow metabolism would benefit heterotrophic bacteria living in a biogeochemical environment of high total DOC but low relative O2 that is observed in algae exudates.

The overflow metabolism predicted by metagenomic analysis was observed in experimental incubations of microbes. When growing on algal exudates, microbes increased cell volume and DOC consumption per cell (Figures 2A and 3). Yet, these cells displayed the same cell-specific O2 consumption compared to small cells growing on coral exudates (Figure 2B). These results indicated that microbes growing on coral exudates fully oxidize the organic carbon consumed, channeling metabolic energy to maintenance costs through NADH- and ATP-producing pathways (Russell and Cook, 1995). On algal exudates, bacteria had higher DOC consumption, with a greater fraction of the organic carbon being only partially oxidized and shunted to NADPH and biosynthesis. The deviation of the excess carbon to these pathways caused less O2 consumption (relative to carbon) and increased community biomass (Figures 2B and 3).

We considered and rejected alternative hypotheses for DOC and O2 consumption patterns observed in our experiments, such as broken TCA cycles and reactive oxygen species (ROS) detoxification (Mailloux et al., 2007; Steinhauser et al., 2012). There was no evidence for an increase in genes encoding these pathways in our dataset (Figure 1—source data 1). Likewise, the metagenomic transitions observed in situ could not be explained by the rise or disappearance of specific taxonomic groups that changed in abundance with increasing algae cover (Dinsdale et al., 2008; Kelly et al., 2014). The disconnection between functional and taxonomic profiles is common in microbial communities and has been previously observed in coral reefs (Kelly et al., 2014). This discordance is due to strain-level variability in functional genes a result of genomic adaptation to multiple environment selective pressures (Klingner et al., 2015; Martiny et al., 2006; Martiny et al., 2009).

Overflow metabolism predicts a shift in ecosystem biomass allocation

The cell volume and abundance changes observed in our experiments helps to explain the higher microbial biomass observed in algae-dominated reefs across the Pacific, Caribbean and Indian Oceans (Haas et al., 2016). In the Pacific, an increase in total bacterial biomass caused the total bacterial energetic demands to surpass that of macrobes in degraded reefs (McDole et al., 2012; McDole Somera et al., 2016). The increase in overflow metabolism observed in our dataset provides a mechanism for the increase in biomass and energetic demand observed in degraded reefs, affecting reef-scale trophic interactions (De'ath and Fabricius, 2010; Haas et al., 2016; Russell and Cook, 1995; Silveira et al., 2015; Wilson et al., 2003).

The changes in biomass observed in situ cannot be attributed to changes in abundance of a single clade. The only clade with an increase relative abundance that could potentially explain the change in cell sizes was Prochlorococcus, which has cell volumes of ~0.1 µm3 (Kirchman, 2016). However, this is the same average cell size observed in the offshore water used as inoculum in our incubation experiments, and the consumption experiments were run in the dark, selecting for heterotrophic microbes (Figure 3B). The other groups with significant relationships with fleshy algae cover contributed to only 0.9% to 1.4% of the community (Figure 1—source data 1), and their average cell sizes are not consistent with the observed increase in cell size in algae: Thermotoga has large cells, but their abundance decreased with algae, Leuconostoc, Rhodopirellula, and Methanobacterium also decreased in abundance with increasing algae. Therefore, none of these clades can explain the change in size observed in this study.

Ebullition causes O2 loss from coral reefs

Photosynthesis and respiration have a theoretical 1:1 molar ratio of organic carbon and O2 produced and consumed (Williams et al., 1979). In holobionts with different proportions of heterotrophic and autotrophic components, such as corals and algae, respiration can consume different fractions of the oxygen and organic carbon produced in photosynthesis (Tremblay et al., 2016). The cost of calcification in corals and calcifying algae can also increase the rate of oxygen consumption due to its energetic costs (Muscatine et al., 1981). As a result, the DOC:O2 ratios in exudates from calcifying holobionts are predicted to be higher than in fleshy algal exudates. This pattern contradicts experimental data showing higher DOC:O2 ratios in microbe-free algae incubations (Haas et al., 2011). Our results suggest that O2 ebullition is the likely cause of this observation. The organic carbon released by algae can be fully solubilized, while a fraction of the O2 nucleates and escapes, leaving behind a high DOC:dissolved O2 ratio. The lack of a relationship between dissolved oxygen concentrations and ebullition rates suggests that the bubbles are formed by heterogenous nucleation on the algae surface (Freeman et al., 2018). Differential carbon allocation into biomass can also alter these ratios, and the quantitative analysis of carbon incorporation using isotope probing in combination with dissolved and gaseous O2 dynamics is the next step to resolve these budgets.

O2 ebullition and bubble injection by hydrodynamics are recognized as a source of error when estimating production in shallow water ecosystems, yet the extent to which ebullition affects these estimates is rarely quantified due to methodological challenges (Cheng et al., 2014; Kraines et al., 1996). O2 ebullition observed in this study corresponded to 5–37% of net community production. Previous studies quantified that ebullition accounted for the loss of up to 21% and 37% of the O2 production in a salt marsh and a lake, respectively (Howard et al., 2018; Koschorreck et al., 2017). If in situ bubble nucleation and rise rates are comparable to those in incubations, our results imply that coral reef gross primary production has been significantly underestimated, especially in algae-dominated states (Howard et al., 2018). Previous studies reported bubbles on the surface of turf algae, on sediments, and inside coral skeletons colonized by endolithic algae (Bellamy and Risk, 1982; Clavier et al., 2008; Freeman et al., 2018; Odum and Odum, 1955). The heterogeneous distribution of bubble nucleation over primary producers entails that benthic community composition determines the magnitude of O2 ebullition on reef-level O2 dynamics. Our model based on the combined results from microbial and primary producer incubations predicts that microbial respiration and ebullition together contribute to the loss of 11.5% to 66.7% of the gross primary production as reef algae cover increases from 0% to 100%. A combination of future in situ incubations and gas exchange studies is required to test these relationships.

Conclusion

Our study suggests that O2 depletion observed in algae-dominated reefs is a result of high bacterial densities and oxygen ebullition. Ebullition causes a decoupling between photosynthetic fixed carbon and O2, fundamentally changing the biogeochemical environment: O2 loss as bubbles causes an increase in dissolved DOC:O2 ratios, stimulating overflow metabolism in the microbial community (Figure 6). These biophysical and physiological changes cause a depletion in oxygen standing stocks that may negatively affect corals and other reef animals.

Conceptual model of oxygen loss mechanisms from coral reefs.

In healthy reefs dominated by calcifying organisms, benthic exudates have low DOC concentrations and low DOC:O2 ratios. The high ratio of electron acceptors (O2) to donors (DOC) stimulates oxidative metabolism in the heterotrophic bacterial community, which sustains low biomass accumulation. In algae-dominated reefs, the loss of oxygen by ebullition leaves behind exudates that are enriched in electron donors (DOC) relative to acceptors (O2). Combined with the higher DOC concentrations, this biogeochemical environment stimulates overflow metabolism in the heterotrophic bacterial community, with incomplete carbon oxidation and accumulation of biomass.

Materials and methods

Central carbon metabolism pathways

Request a detailed protocol

Microbial metagenomes were sequenced from water samples collected on coral reefs in the Pacific and in the Caribbean. The Pacific samples were collected during NOAA RAMP cruises from 2012 to 2014 to the Hawaiian Islands, Line Islands, American Samoa, and Phoenix Islands. Caribbean samples were collected around the island of Curaçao during the Waitt Institute Blue Halo expedition in 2015. Geographic coordinates for each sampling site are provided in source data 1 for Figure 1, along with benthic cover data. At each sampling site, SCUBA divers collected water from within 30 cm of the reef surface using Hatay-Niskin bottles (Haas et al., 2014a). One metagenome was generated from each sampling site, and 1 to 5 sites were sampled around each island in the Pacific, and 22 samples from around the island of Curaçao, in the Caribbean. Samples were brought to the ship, filtered through a 0.22 μm cylindrical filter within 4 hr, and stored at −20°C until laboratory processing. DNA was extracted from the filters using Nucleospin Tissue Extraction kits (Macherey Nagel, Germany) and sequenced on an Illumina HiSeq platform (Illumina, USA). Pacific microbial metagenomes described here were previously analyzed for the presence of CRISPR elements, competence genes and Shannon diversity in Knowles et al. (2016). All fastq files from Pacific and Curaçao metagenomes were quality filtered in BBTools with quality score >20, duplicate removal, minimum length of 60 bases, and entropy 0.7 (Bushnell, 2014). The 87 metagenomes are available on the NCBI Short Read Archive (PRJNA494971). The number of samples analyzed here surpassed that estimated by power analysis based on the work of Haas et al. (2016) (estimated n = 67 for α = 0.05 and β = 0.2 based on the relationship between microbial abundance and algae cover shown in Figure 2B of Haas et al., 2016). Relative abundances of genes encoding rate-limiting enzymes in central carbon metabolism pathways were utilized as proxy for the representation of that pathway in the community. The list of pathways and enzymes analyzed here is provided in Supplementary Table 1. Enzyme-specific databases were built using amino acid sequences from the NCBI bacterial RefSeq. BLASTx searches were performed using metagenomic reads against each enzyme gene database, using a minimum alignment length of 40, minimum identity of 60% and e-value <10−5. Reads mapping to the database were normalized by total high-quality reads resulting in the percent abundance of each enzyme gene. Relative gene abundances are provided as source data one for Figure 1.

The relationship between fleshy algae cover and enzyme gene relative abundances was analyzed by the statistical learning method of supervised random forests regression (Breiman, 2001). In the random forest procedure, each tree in the ensemble is grown using a different bootstrap sample of the original data. For the final prediction, the mean prediction regression from all the trees is used. Since each tree is grown with a bootstrap sample of the data, there is approximately one-third of the data that is ‘out of sample’ for each tree. This out of sample data acts as an internal validation dataset and can be used to determine variable importance for each predictor variable. This is done for each tree by randomly permuting each variable in the out of sample data and recording the prediction. For the ensemble of trees, the permuted predictions are compared with the unpermuted predictions. The magnitude of the mean decrease in accuracy indicates the importance of that variable. Significance of the importance measures is estimated using a permutation test. The random forest easily handles different types of variables and does not require the data to be rescaled or transformed. The random forest consists of an ensemble of decision trees for regression. Predictions, prediction error, and variable importance are estimated from the ensemble of trees. The random forest was grown using 1000 trees and four tree splitting variables. The mean squared error diagnostic plot of the random forest indicated it settled down and therefore enough trees have been used (Figure 1—figure supplement 1). The significance of the random forests variable importance was determined by a permutation method, which has an implicit null hypothesis test, and was implemented using the R package rfPermute. The permutation test p-values were obtained using 100 permutation replicates. Variable importance was selected based on the increasing mean squared error, a measure of the importance of a given feature for the correct prediction of a response variable, with alpha cutoff of 0.05. No outliers were removed from any analyses.

The changes in bacterial community composition across the gradient in algae cover were compared to changes in relative abundances of metabolic genes to access the relationship between taxonomy and function. This analysis can indicate whether specific taxa with differences in gene copy number and genome sizes would drive the functional variation. First, the abundance of bacterial genera in metagenomes was generated through k-mer profiles using FOCUS2 (Silva et al., 2016). FOCUS2 computes the optimal set of organism abundances using non-negative least squares (NNLS) to match the metagenome k-mer composition to organisms in a reference database (Silva et al., 2014). For annotation at species level, sequences were mapped against a database of bacterial genomes from RefSeq using Smalt at 96% sequence identity (Ponstingl and Ning, 2010). The abundance of each organism was computed using the FRAP normalization (Fragment Recruitment, Assembly, Purification), which calculates the fractional abundance of each bacteria in the metagenome accounting for the probability of a read mapping to a genome given the length of both the read and the genomes in the database (Cobián Güemes et al., 2016). Taxa with median relative abundance across samples equal to zero were removed from the analysis. The relative abundances of genera and species were separately analyzed through permutation regression random forests with percent algae cover as response variable in both cases. The random forest was grown using 1000 trees and four tree splitting variables. The mean squared error diagnostic plot of the random forest indicated it settled down and therefore enough trees have been used (Figure 1—figure supplement 1). Variable importance was selected based on the increasing mean squared error, a measure of the importance of a given feature for the correct prediction of a response variable, with alpha cutoff of 0.05. The complete genomes of the taxa varying in abundance with algae cover deposited in the NCBI RefSeq were manually checked for copy number variation in the genes identified by the functional analysis according the annotations associated with each genome deposited.

Bacterial DOC and O2 demands

Request a detailed protocol

This dark incubation experiment was performed in Mo’orea, French Polynesia. Briefly, exudates from corals and algae released during a light incubation period were 0.2 μm-filtered and inoculated with freshly collected unfiltered forereef seawater to add a compositionally-representative ambient microbial community to the exudate samples. The forereef water column sample contains a microbial community that is most representative of the offshore community that approaches the reef environment in the incoming current (Haas et al., 2011; Nelson et al., 2013). Inoculation in filtered seawater was utilized as control. All bottles were kept in the dark for 48 hr. DOC, dissolved O2 and microbial abundances were measured at the beginning and at the end of each dark incubation as described above. DOC samples were filtered through pre-combusted GF/F filters (Whatman, 0.7 μm nominal particle retention) and transferred to acid-washed HDPE vials. Samples were kept frozen until analysis according to Carlson et al. (2010). DOC and O2 consumption rates normalized by the bacterial cell yield at the end of each experiment resulted in cell-specific carbon and O2 demands. The fleshy organisms used in this experiment were turf algae and the macroalgae Turbinaria ornata and Amansia rhodantha, and the calcifying organisms were CCA and the coral Porites lobata. Five replicate incubations of each organism were performed (Power analysis estimate n = 4 for α = 0.05 and β = 0.2 based on the incubations of Haas et al., 2011 and Haas et al., 2013b).

Bacterial cell sizes and biomass

Request a detailed protocol

Changes in bacterial abundance, cell size and total biomass in response to primary producers derived from two independent experiments, described below:

Experiment 1: Microbial communities from the reef off the CARMABI Research Station, in the island of Curacao were incubated with four different primary producers: the scleractinian coral Orbicella faveolata, CCA, the fleshy macroalgae Chaetomorpha sp., and turf algae. O. faveolata colonies were collected at 12 m depth from the Water Factory site (12°10’91’ N, 68°95’49’ W) and cut into ~10 cm2 fragments. Coral fragments were kept for two weeks at the CARMABI Research Station flow-through tank system. The tank was subject to natural diel light cycles with light intensities comparable to 10 m depth, as measured using HOBO Pendant UA-002–64. CCA, turf, and macroalgae were collected off CARMABI immediately prior to the experiment. Five replicate incubations for each organism were conducted, and five control incubations had no primary producer (Power analysis estimate n = 4 for α = 0.05 and β = 0.2 based on the incubations of Haas et al., 2011 and Haas et al., 2013a). Surface area of organisms is provided in Figure 3—source data 1. Incubations lasted for 4 days at 24°C with natural diel light cycles. Bacterial communities at the start and end of the incubation were analyzed by fluorescence microscopy according to McDole et al. (2012). Briefly, cell volume was calculated by considering all cells to be cylinders with hemispherical caps and individual microbial cell volumes were converted to mass in wet and dry weight using previously established size-dependent relationships for marine microbial communities (Simon and Azam, 1989). Differences in cell abundance, cell volume, and total microbial biomass were calculated by the difference between final and initial values, and were normalized by the area of the primary producer in the incubation.

Experiment 2: Five specimens of the coral Favia sp. were obtained from a long-term holding tank at the Hawaiian Institute of Marine Biology (HIMB) and placed in independent 5 L polycarbonate containers filled with 0.2 μm-filtered seawater. Five specimens of the fleshy macroalgae Gracilaria sp. were collected off HIMB and placed in independent 5 L containers. Additional control containers were filled with filtered seawater. Primary producers were incubated in natural light conditions for 8 hr to release exudates. At the end of the incubation, 2 L of seawater containing exudates were 0.2 μm-filtered and inoculated with 1 L of unfiltered offshore seawater containing water column reef microbial communities. All bottles were incubated for 24 hr in the dark. For microbial abundance and biomass determination, 1 mL samples were collected and analyzed as described above. Differences in cell abundance, cell volume and total heterotrophic microbial biomass were calculated by the difference between final and initial values.

O2 release by benthic primary producers

Request a detailed protocol

The rates of dissolved and gaseous O2 production by benthic primary producers were measured in two tank experiments (POP Experiments 1 and 2). In both experiments, organisms were incubated in custom-made chambers named peripheral oxygen production (POP) bottles. POP bottles are bell-shaped Polyethylene Terephthalate (PET) bottles with a removable base and two sampling ports, one for dissolved analyte sampling, and one at the top for gas sampling. Primary producers were placed at the bottom and bubbles released from their surfaces during incubation accumulated at the top. Primary producers were placed on the base of the POP bottles, inside a larger tank filled with reef water. The bell-shaped container was then placed over the base, and the bottle was sealed. At the end of the incubation, the gas accumulated at the top of the bottles was collected in a syringe and the volume of gas was recorded. The gas was transferred to a wide-mouth container and O2 partial pressure was measured using a polarographic probe (Extech 407510) immediately upon collection. The POP chambers and containers used for measurements were located underwater inside a larger holding tank, so no gas was lost during the quantification procedures. Dissolved O2 was determined using a Hatch-Lange HQ40 DO probe.

POP Experiment 1: The scleractinian coral Montipora sp. and the fleshy macroalgae Chaetomorpha sp. were collected from coral and macroalgae long-term holding tanks maintained at SDSU. Specimens were collected from the tanks immediately prior to the experiment and placed inside POP bottles. Three specimens of each organism were individually incubated for 2 days with artificial seawater from the coral tank, along with three control bottles under cycles of 12 hr light (150 μmol quanta m−2 s−1) and 12 hr dark at 27°C.

POP Experiment 2: Four benthic primary producers were analyzed: the scleractinian coral Orbicella faveolata, CCA, the fleshy macroalgae Chaetomorpha sp., and turf algae. O. faveolata colonies were collected at 12 m depth from the Water Factory site in Curaçao (12°10’91’ N, 68°95’49’ W) and cut into ~10 cm2 fragments. Coral fragments were kept for two weeks at the CARMABI Research Station flow-through tank system. The tank was subject to natural diel light cycles with light intensities comparable to 10 m depth, as measured using HOBO Pendant UA-002–64. CCA, turf, and macroalgae were collected off CARMABI immediately prior to the experiment. Five individual incubations for each organism were conducted. Surface area of organisms is provided in Figure 3—source data 1. Incubations lasted for 4 days at 24°C with natural diel light cycles (PAR varied between 150 and 500 μmol quanta m−2 s−1). The dissolved O2 in macroalgae and CCA bottles were above the accuracy range of the oxygen probe used. In these cases, we used the highest value within the accuracy range of the probe for all following calculations. Therefore, the results shown are lower bounds of the actual O2 concentrations. Figure 3A indicates the bottles where supersaturation was observed with gray-filled symbols, and the estimated fraction of oxygen lost by ebullition in Figure 3C that were calculated using these lower bound values.

Open-tank ebullition experiments

Request a detailed protocol

Two species of algae (Chaetomorpha sp. and Gracilaria sp.) and two species of coral (Montipora sp. and Pocillopora sp.) from the SDSU long-term holding tanks were transferred to incubation tanks immediately prior to each experiment. The surface area of each organism, light intensity (150 to 630 µmol quanta m−2 s−1) and incubation times are provided in Figure 5—source data 1. The experiment started in the morning after a 12 hr period of darkness. Each organism was transferred to a 15 cm x 20 cm x 30 cm glass tank containing seawater from the holding tank. The tank was open on the top to allow free gas exchange with atmospheric air. A GoPro camera was positioned inside the tank facing the organism, and videos were recorded at 4K 30 fps (Videos 1 and 2 are provided as representative videos of the experiments with Chaetomorpha sp. at 150 and 300 PAR, respectively). Additional pictures were taken with a Canon RebelT3 with 18–55 mm lens to obtain bubble sizes. Dissolved O2 was measured throughout the experiments using a Hatch-Lange HQ40 DO probe. The number of bubbles produced per minute was determined by manual counts of bubbles rising from the algae in the video. Bubble sizes were determined from a total of 160 bubbles (at least 40 bubbles from each experiment) from the still images in Adobe Illustrator.

Estimates of oxygen loss by ebullition

Request a detailed protocol

The internal pressure, pi, of a bubble was estimated using the Laplace’s law using a standard surface tension model, pi=pe+2σw/Rb (Bowers et al., 1995). Here, pe was the external pressure (assumed at atmospheric pressure, pe=1 atm), σw the surface tension of water (71.99 ± 0.05 mN/m at 25°C, Pallas and Harrison, 1990), and Rb the radius of the bubble. The moles of oxygen in a bubble, nb, were estimated using the ideal gas law, nb=(piVb)/(RT), which is a good approximation for oxygen at normal conditions (Christensen et al., 1992). Here, R was the universal gas constant (8.314 J K–1 mol–1), T the temperature (25°C or 298.15 K), and Vb was the volume of the bubble, which was approximated as spherical, Vb=(4π/3)Rb3. The rate of bubble production necessary to generate the moles of oxygen in the gas phase of the POP experiment, nPOP, was obtained by assuming a constant bubble production: rb=Nb/tlight. Here Nb=nPOP/nb was the number of bubbles needed, and tlight was the total time of light exposure in the POP experiments (2 to 4 days of 12 hours of light per day). The rate was normalized by the covered area (Ap) of the primary producer, ρb=rb/Ap.

The estimated ebullition rate was obtained for macroalgae in the POP experiment using the surface values (Ap) and moles (nb) values reported in Figure 4—source data 1. The bubble diameter utilized was derived from the open tank ebullition experiment. The mean, first quartile (25%) and third quartile (75%) of bubble sizes were used to obtain the range of ebullition rates necessary to recover the gas in headspace of the POP experiments, indicated by the shaded area in Figure 5A.

Statistical tests

Request a detailed protocol

All statistical analyses were performed using the software R. The response variables from the incubation experiments were tested for normality using the Shapiro–Wilk test. Due to lack of normality (Shapiro–Wilk, p<0.05), the non-parametric Kruskal-Wallis test was used to test if there were differences in treatments followed by the post-hoc Wilcoxon test with the False Discovery Rate (FDR) multiple-comparison correction with a significance cutoff of p<0.05. Because gaseous O2 production and fraction of total O2 in the form of gas were not significantly different between POP Experiments 1 and 2 (Kruskal-Wallis, p>0.05), the results of both experiments were combined in subsequent tests. Statistical tests comparing calcification functional groups were performed by combining corals with CCA as calcifying and turf algae with fleshy macroalgae as fleshy organisms, based solely on the presence of absence of calcification. No outliers were removed.

Uncertainty quantification

Request a detailed protocol

All measurements from incubations had an associated error originating from probe accuracy, organism surface area measurement software, volume in incubation chambers, incubation time and microscope resolution. The uncertainty quantification of rates calculated using combination of measurements was performed based on error propagation (Taylor, 1997). For error generated from two or more variables, the derivative of the source errors were applied (Rice, 2006). All uncertainty propagated from the device measurements were at least five times smaller than the statistical standard deviations from the experiments, except for control incubations where the statistical variation is in the same order as the uncertainty, as expected (Figures 2, 3 and 4 source datasets). Therefore, uncertainty was not incorporated in statistical comparisons between treatments.

Predictive model

Request a detailed protocol

The relative contribution of microbial respiration and ebullition to the removal of oxygen in an idealized reef was performed using a weighted linear model. The per-cell respiration rates from the dark experiment (Figure 2—source data 1) were applied to the bacterial abundances sustained by one decimeter square (dm2) of primary producer observed in POP experiment 2 (Figure 3—source data 1). These rates were scaled up to a 1 m3 reef by normalizing the rate by the percent cover of coral and fleshy algae on the benthos (Figure 5—source data 2). The model focused on these two primary producer groups due to the availability of data on their ebullition rates. The derivation of the model is provided below.

The production of O2 in the gas phase per m2 was calculated from the POP experiments for corals, ρcoralgas, and fleshy algae, ρcoralgas. The net production of dissolved O2 per area (m2) was also obtained from the POP experiments for coral, ρcoralnet, and fleshy algae, ρalganet. The microbial respiration rate of dissolved O2 per cell was obtained from the dark incubation experiment (Figure 2). The microbial concentration from the POP experiment at the end point was assumed as the reference value to obtain the rate of microbial respiration per volume (m3) in the POP experiment. This was obtained for coral, Ωcoralmicrob, and fleshy algae, Ωalgamicrob, incubations. To estimate the gross production rate of dissolved oxygen in a reef, a boundary layer of a meter was considered, and above this layer the influence of the primary producers on the oxygen content was neglected (Barott and Rohwer, 2012; Shashar et al., 1996). The gross production of dissolved O2 per meter square of primary producers was, thus, estimated from the coral incubations, Ocoraldiss=ρcoralnet×1 m2+ Ωcoralmicrob×1 m3, and the fleshy algae incubations, Oalgadiss=ρalganet×1 m2+ Ωalgamicrob×1 m3. The gross production rate of O2 was obtained by combining the dissolved oxygen and gas oxygen obtained from the coral, Ocoralgross=Ocoraldiss+ ρcoralgas×1 m2, and algae, Oalgagross=Oalgadiss+ ρalgagas×1 m2, experiments. The benthos of an ideal reef was assumed to be covered by a combination of coral and algae, C+A=1, where C was the fraction of coral coverage and A the fraction of algal coverage. The gross production rate of O2 in a reef was estimated by the linear weighted model Oreefgross= COcoralgross+AOalgaegross. The production rate of O2 gas, Oreefgas, and microbial consumption rate of O2, Oreefmicrob, in the reef were calculated analogously. The coral coverage fraction was expressed in terms of algal coverage fraction, CA=1-A, to investigate the impact of algal coverage in the oxygen budget of the reef. The fraction of oxygen loss in ebullition was defined as E=Oreefgas/Oreefgross. In terms of the algal coverage, the ebullition fraction was given by the shifted Hill function of order one

EA=ΔOgasΔOgrossA+OcoralgasΔOgross OcoralgrossΔOgross+A

Here, Δ Ogross=Oalgagross-Ocoralgross and Δ Ogas=Oalgagas-Ocoralgas. A similar expression was obtained for the fraction of oxygen rate consumed by microbes as a function of the fraction of algal coverage, M(A), where the gas terms above become microbial terms.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
    Tricarboxylic acid cycle and glyoxylate bypass
    1. JE Cronan
    2. D Laporte
    (2013)
    EcoSal Plus, 1, 10.1128/ecosalplus.3.5.2.
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
    Fish or Germs? Microbial Dynamics Associated with Changing Trophic Structures on Coral Reefs
    1. EA Dinsdale
    2. F Rohwer
    (2011)
    In: Z Dubinsky, N Stambler, editors. Coral Reefs: An Ecosystem in Transition. Dordrecht: Springer. pp. 231–240.
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
    SMALT-a new mapper for DNA sequencing reads
    1. H Ponstingl
    2. Z Ning
    (2010)
    Intelligent Systems for Molecular Biology 2010 meeting.
  64. 64
    Mathematical Statistics and Data Analysis
    1. JA Rice
    (2006)
    Cengage.
  65. 65
  66. 66
  67. 67
    Energetics of bacterial growth: balance of anabolic and catabolic reactions
    1. JB Russell
    2. GM Cook
    (1995)
    Microbiological Reviews 59:48–62.
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
    Introduction to Error Analysis the Study of Uncertainties in Physical Measurements
    1. J Taylor
    (1997)
    University Science Books.
  81. 81
  82. 82
  83. 83
  84. 84
  85. 85
    Agreement between the C14 and oxygen methods of measuring phytoplankton production-reassessment of the photosynthetic quotient
    1. PJB Williams
    2. RCT Raine
    3. JR Bryan
    (1979)
    Oceanologica Acta 2:411–416.
  86. 86
    Detritus in epilithic algal matrix and its use by coral reef fishes
    1. SK Wilson
    2. DR Bellwood
    3. JH Choat
    4. MJ Furnas
    (2003)
    Oceanography and Marine Biology 41:279–309.

Decision letter

  1. Evan Howard
    Reviewing Editor
  2. Ian T Baldwin
    Senior Editor; Max Planck Institute for Chemical Ecology, Germany

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

Overfishing, eutrophication, and climate change are shifting the composition of many coral reefs towards more algae and less coral. Algal have greater net oxygen production than corals, yet paradoxically this shift in producers decreases ecosystem oxygen availability while stimulating bacterial growth, a process known as microbialization. Silveira and coauthors combine metagenomic field data and laboratory incubations of algae, corals, and reef bacteria to tell a novel and compelling story: 20% of the oxygen generated by reef algae could escape to the atmosphere in bubbles, leaving behind excess dissolved organic carbon. This imbalance stimulates metabolic pathways in heterotrophic bacteria that tilt the cellular scales towards greater organic carbon uptake and biomass accumulation, and thus greater ecosystem respiration. Taken together, these processes suggest a mechanistic underpinning for the observed ecosystem changes accompanying microbialization. The authors' findings reinforce the growing body of oceanographic and limnologic work indicating that photosynthetically stimulated bubbling can lead to underestimation of oxygen-based aquatic ecosystem metabolism fluxes. This work should further prompt in situ investigations of how bubbles and alternative metabolic pathways can lead to non-canonical relationships between carbon and oxygen fluxes and stimulate ecosystem shifts-these may be important factors driving the response of shallow, productive aquatic environments to changing climate.

Decision letter after peer review:

Thank you for submitting your work entitled "Biophysical and physiological processes causing oxygen loss from coral reefs" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Senior Editor in consultation with an outside expert. The reviewers have opted to remain anonymous.

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife.

The work is clearly very interesting and we all were convinced that the main thesis could well be right, but that at this stage, the data does not fully support your conclusions. In addition I wanted to note three points from my own reading of the work and the reviews, that might help you prepare for the next steps for this work:

1) The arguments that the microbial community uses less energy-efficient carbon consumption pathways is based on metagenomic analyses. However, metagenomic analyses alone are not sufficient for identifying what a microbial community is really doing, they only reveal the metabolic capabilities of the community. Transcriptomic or proteomic analyses would be essential to show that there was a shift of the community in their use of carbon consumption pathways.

2) It was not clear which advance the metagenomic analyses of this paper provides as compared to the advances presented in the earlier paper by Haas et al., 2016,using a smaller dataset of 60 coral reef sites (vs. 87 here). Was the same data set used as in Haas et al. albeit expanded by 27 additional sites?

3) As noted by the first reviewer, the statistical analyses require much additional work.

Reviewer #1:

I think the datasets presented in this manuscript are promising and could ultimately support the stated conclusions. However, this manuscript has multiple issues that I do not believe can be adequately addressed within the scope of this submission (i.e. without a further round of peer review).

In summary, the major issues I identify include:

1) Shortcomings with the metagenomics approach and related statistics: The authors have set themselves a harder task by using genomic rather than transcriptomic data, and haven't convincingly shown that the genomic data can answer their question about activity of distinct metabolic pathways. The choice of a classification technique in place of standard statistical tools that work well with such data is not explained by the authors, and the details provided are insufficient to show that the result is statistically robust.

2) Methodological and analysis errors with the oxygen ebullition experiments: The oxygen data does not appear to have undergone QA/QC, leading to spurious results based on readings entirely beyond the accurate range of the sensor, and in many cases beyond the sensor detection limits. There is no uncertainty analysis of resulting rates. The values presented are extremely supersaturated with respect to dissolved oxygen and characteristic of equilibrium conditions in the gas phase, meaning the results are unlikely to be environmentally relevant and may not be related to ebullition.

3) Ratios of DOC and oxygen uptake in incubations which are not consistent with the expectations laid out in the Discussion: The DOC and oxygen uptake rates are similar to the expected 1:1 ratio for fleshy algae exudates, while it is the control and coral experiments that have unexpected ratios. This implies that the expectation value presented is incorrect (because the rate of DIC increase is not equal to the rate of DOC uptake), that an unexplained process is occurring in the coral exudate and control samples, or that the uncertainties are substantial (there was again no uncertainty analysis of the rates in this experiment).

4) Lack of discussion as to what composes the bacterial community in incubation experiments and how that informs or caveats interpretation of the experimental results: Without discussion of what is present in the experiments it is hard to interpret those results. Is community composition changing? Are certain clades dominating the observed responses? Does the implicit choice of community structure in the incubations bias the results towards certain outcomes (e.g. a community that normally grows near algae has a stronger response to algal exudates than to a carbon source it is not adapted for?).

A longer explanation of these concerns (with specific details and some suggestions for ways forward) follows, with the idea that it may be useful to the authors should they choose to resubmit.

1) Metagenomics a) Metagenomics seems like an indirect tool to answer the question the authors have posed regarding gene expression as a function of algal cover. In the absence of transcriptomic data that identifies active gene expression, some logical leaps are required to connect gene abundance to active metabolic pathways in the microbial community, and the current paper does not provide sufficient detail to outline these or convince a reader that they are justified.

b) The statistical analysis of the metagenomic data is not adequately described for me to evaluate how robust it is. In particular there is no indication that the genetic data was normalized to bacterial abundance or some other sensible scheme to avoid biases due to very different sampled communities (dividing by the number of good reads is insufficient). It is unclear what if any null hypothesis was tested. 14% covariance between differences in the gene community and algal cover is a weak signal given that the five genes were explicitly chosen with the expectation of showing a signal, and at the chosen alpha value (0.05) 5% of results are false positives by definition. Using a random forest approach, the authors might evaluate the null hypothesis based on the distribution of variance explained over ~1000 random subsets of five genes, and then evaluate the significance of the 14% value relative to that null distribution (which in my experience is unlikely to have both mean and standard deviation equal to 0% as appears to be implicit in the current analysis).

c) I question why a random forest classification and regression approach was used when something like an ANOVA should work well and is commonly used with -omics data (with clearly testable hypotheses, or a PCA if not). Since a random forest approach was used, more details are required about the number of trees and tree size, the division of the data in training, validation, and test subsets and any resulting tuned hyperparameters, and finally whether stated errors refer to out of bag (classification) or misfit (regression) errors. My personal experience is that the default statistical software settings for random forests can lead to non-robust results, thus care must be taken in the selection (and description) of these characteristics.

d) Since the metagenomes are already assembled the authors should know which species/strain level bacteria are present. Thus I would think a more convincing assessment of the effect of copy number variability would be to look at orthologs and paralogs in at the species/strain level and then evaluate if certain extremes of algal cover (very low, very high) are associated with particular species with many gene replicates or deletions that drive the overall trend. Doing this at the genera level as is the current choice could mute any problematic signal, unless the genera is monospecific. The manuscript states, however, that there is in fact high strain level variation, meaning that this choice is potentially masking a problem. At a minimum, a good reference should justify the choice. And this analysis still does not address the potential complication of variability in gene expression between sites. Finally, from Supplementary file 1 it appears that autotrophic Prochlorococcus has a stronger relationship with algal cover than the heterotrophic bacteria that are the focus of this work. It may be necessary to confirm Prochlorococcus (and Synechococcus, etc.) are not obfuscating your signal.

2) Oxygen ebullition experiments a) Based on the Figure 4—source data 1, it appears that there was a failure to do QA/QC on the oxygen concentrations in the POP experiments. The dissolved oxygen concentration at the end of all of the non-control experiments are outside the stated accurate range for the instrument used, 0-20 mg/L or 0-200% saturation, whichever is lower (https://www.hach.com/hq40d-portable-multi-meter-ph-conductivity-salinity-tds-dissolved-oxygen-do-orp-and-ise-for-water/product-details?id=7640501639). Since this is an optode based optical system, the concentration based range is more notion than reality and the saturation state should always be the limiting bound at high oxygen conditions. Optode based dissolved oxygen sensors have nonlinear responses to concentration, so the deviation from the true value may vary in a way that is not readily predictable (even within the 0-200% range the accuracy varies substantially). Not only are all the experiments beyond the accurate range, many of them (including all the CCA experiments) are beyond the instrument bounds (the internal QA filter is apparently set to 30 mg/L given the constant endpoint values of ~935 μm in many experiments).

b) Hypothetically the above experimental data from below the upper instrument detection bound could be salvaged, if a full range calibration to Winkler titrations or another sensor were to be run on the same sensor cap originally used, at each of the experimental temperatures (the oxygen sensitivity of the optode foil is temperature dependent). See Bushinsky and Emerson, 2013 for an example of what this might look like (https://doi.org/10.1016/j.marchem.2013.05.001). But the sensor characteristics have a substantial "grow-in" if they are new, and change if they are dry for an extended period, so this would have the best results if the sensor cap had already seen several months of use prior to the experiments and has been kept continuously moist since. But without the beyond detection samples there is insufficient experimental power to determine significance…

c) I'm concerned with how the sampling of the headspace gas is described. The manuscript says the gas was transferred from a syringe to a wide mouthed container that the second probe was placed into. Gas diffuses quickly, so I don't see how this could lead to accurate measurements-but I may be misunderstanding, and this refers to the dissolved sample? At any rate this description needs to be clarified.

d) Setting aside the accuracy issues above, all of the headspace values are approximately in equilibrium with the dissolved oxygen you measured. Henry's law solubility coefficients predict concentrations of 0.36-0.94 mg/mL and 1.3-20.6 mg of oxygen in the gas phase above the water samples-close to the 0.4-0.85 mg/mL and 1.1-17 mg O2 measured (assuming 1 atm pressure, and neglecting any lag between production and diffusion). Thus I think that the presence of a headspace isn't proof of the importance of ebullition in a natural environment-it is proof that a highly supersaturated solution will push gas into a headspace (like CO2 is in a non-experimental pop bottle). I do suspect bubble nucleation was essential, since the controls were also supersaturated and formed no headspace, but they were a lot less supersaturated so this would be more convincing with evidence of rising bubbles in the chambers-otherwise thermodynamics alone is sufficient without any biophysical interactions. At any rate what the above suggests to me is these chambers are unlikely to reflect environmental outcomes regarding the amount of oxygen released in bubbles. In this light it is very hard to support the statement that in algal dominated reefs 1/3 of the oxygen escapes as bubbles.

e) There is no uncertainty analysis of these rates. Given the relatively coarse measurements of bubble volume (to nearest 0.5 mL), short durations (2-4 hours), and uncertainties in saturation state (even when the instrument is within its specified range), both the gas production and dissolved oxygen change must have significant uncertainties, but these are not quantified. I think they should be, and relative uncertainty of each experiment shown so a reader can gauge how significant differences between the box and whisker categories actually are. Also, 2 pts don't make for a convincing rate.

3) DOC and oxygen uptake experiments

The DOC and oxygen uptake ratios in these experiments are far from 1:1, but not in the manner described to support arguments about less complete oxidation of organic matter. To use respiratory quotients with DOC it must be assumed that DOC uptake is equivalent to CO2 (DIC) release. The arguments put forward in this manuscript about the incomplete oxidation of organic matter would argue against this for the macroalgae, but I'll set that aside for the moment. At any rate, the -DOC/-O2 ratio for the fleshy algae exudate is ~0.9, right in line with theoretical and measured values of +DIC/-O2 for bacterioplankton (Robinson 2019, https://doi.org/10.3389/fmars.2018.00533). It is instead the controls and corals that seem to have unexpected values on the order of 0.3. So I see three options: 1) the -DOC/-O2 rate ratio shouldn't actually be ~1:1 (or rather 0.9), and thus the discussion of respiratory quotients is a red herring. 2) the corals and controls are doing something unusual that requires an explanation. 3) There are major biases in the rates. I think the error analysis I recommended for the O2 production fluxes should be done here as well and could inform this question. So overall I agree that the incubations show more oxygen consumption per unit of DOC consumed with the control and coral exudates than with the algal exudates, but since none of these categories meet the stated expectations this needs to be explored in more detail to confirm that the result is real and robust.

4) Bacterial community structure

The bacterial community structure is not discussed. How it diverges across sampled environments is challenging to ascertain in Figure 1. I don't find any information on what the bacterial community looked like within the various incubation experiments. Thus I can't evaluate whether the incubation results are a function of, e.g., incubating a community adapted to one particular environment in a way that would bias the experimental outcomes. While there is a mention about changing community structure with regards to the incubations (Discussion), there is not data in the supplement regarding the community in these experiments, nor would I expect changing genetic composition over the course of experiments lasting hours to days (again, a question for transcriptomics?). This makes it hard to interpret what is happening in the experiments where change in biomass is evaluated. Are changes in biomass due primarily to a single clade? Are certain larger types of bacteria being selected for, or are particular bacteria growing larger while others do not? Is this an expected result based on where those bacteria normally are environmentally successful?

Reviewer #2:

This study provides a theoretical and experimental based evidence approach to explain the observations for O2 depletion in algal dominated coral reefs. The processes of ebullition and microbial metabolism are postulated to drive this lower observed oxygen level in algal dominated reefs. Overall the manuscript is very interesting and details a nice combination of environmental genomics based analysis linked with controlled experimental trials. The postulated physical and biological reasons for reduced O2 levels are supported by the experimental data, though the conclusions are still a little speculative. However this approach represents a great approach to hypothesis driven science; and while not able to definitely confirm these process (i.e. ebullition and microbial metabolism) as driving O2 depletion, it sets up the concepts to be further tested in environmental settings. Hence the manuscript is worthy of publication with a few comments detailed below highlighting potential areas for improvement.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for choosing to send your work entitled "Biophysical and physiological processes causing oxygen loss from coral reefs" for consideration at eLife.

We thank you for sharing your rebuttal letter. We would welcome a revision of your work for a new cycle of review and note that the revision should address the following concerns:

The text is missing or unclear on the rationale for a number of the statistical and analytical choices, and has weaknesses in some of the lines of evidence, including unresolved problems with the oxygen production experiments. Thus the current manuscript, in the opinion of two reviewers, presents the interpretations with a certainty and forcefulness which we did not think is warranted by the present constellation of suggestive (but not definitive) evidence.

Thank you for clarifying the oxygen units. However, all of the CCA and most of the fleshy macroalgae experiments were beyond the detection bounds of your instrument (20 mg/L), as evidenced by the constant endpoint concentrations (937.5 umol/1.5L=625µM=20 mg/L) across several nominally independent experiments. The headspace concentrations also still appear to reflect equilibrium conditions with the underlying water when the solution is strongly supersaturated (as it is in most of the non-control experiments). How relevant are these experiments for explaining oxygen loss under realistic environmental conditions (less supersaturated and open to the atmosphere)?

There are statistical techniques for dealing with non-normally distributed data (or more accurately, non-conditionally normal error distributions-A random forest regression may well be a good choice for this analysis, but a rationale should be explained, and the particulars of how it is used clearly stated.

Whether 5 or 16 genes, the correlations to environmental variables need more context. Our concern remains about the effect size versus a null hypothesis of some other random set of genes from your already sequenced genomes, which are far larger than 16 genes. The details required to assess what differences are "real" are currently insufficient. e.g. was the data normalized to bacterial abundance? A single 100mb genome will have far more reads than a 1mb genome, even if there are 10 times more of the smaller-genome organism. You may well have made considered choices for these and other potential issues in your analysis, but you tell the reader none of this, even in the supplement.

Regarding your choice of approach for the metagenomic data. Please provide context and defend the choice (and aspects of the analysis of this data) in the text. It would be a good choice to highlight the rationale of this approach in as succinct a manner in the text as you do in your rebuttal letter.

We look forward to your revision.

https://doi.org/10.7554/eLife.49114.sa1

Author response

[Editors’ note: the author responses to the first round of peer review follow.]

The work is clearly very interesting and we all were convinced that the main thesis could well be right, but that at this stage, the data does not fully support your conclusions. In addition I wanted to note three points from my own reading of the work and the reviews, that might help you prepare for the next steps for this work:

1) The arguments that the microbial community uses less energy-efficient carbon consumption pathways is based on metagenomic analyses. However, metagenomic analyses alone are not sufficient for identifying what a microbial community is really doing, they only reveal the metabolic capabilities of the community. Transcriptomic or proteomic analyses would be essential to show that there was a shift of the community in their use of carbon consumption pathways.

2) It was not clear which advance the metagenomic analyses of this paper provides as compared to the advances presented in the earlier paper by Haas et al., 2016, using a smaller dataset of 60 coral reef sites (vs. 87 here). Was the same data set used as in Haas et al. albeit expanded by 27 additional sites?

3) As noted by the first reviewer, the statistical analyses require much additional work.

Reviewer #1:

I think the datasets presented in this manuscript are promising and could ultimately support the stated conclusions. However, this manuscript has multiple issues that I do not believe can be adequately addressed within the scope of this submission (i.e. without a further round of peer review).

In summary, the major issues I identify include:

1) Shortcomings with the metagenomics approach and related statistics: The authors have set themselves a harder task by using genomic rather than transcriptomic data, and haven't convincingly shown that the genomic data can answer their question about activity of distinct metabolic pathways.

Previous studies have shown the fast genomic adaptation of offshore bacterial communities to the reef environment – hours to days, the timescale of residence time – and the validity of using metagenomics to describe functional adaptation of bacteria (Nelson et al., 2013, Wegley-Kelly et al., 2015). This genomic adaptation occurs at the strain level, with rapid selection of strain variants with highest fitness under different selective pressures (Martiny, 2006, 2008). Therefore, the metagenomic analysis is a strong tool to indicate which carbon metabolic pathways are selected in the reef environments. This analysis provided a hypothesis that was experimentally validated in our in vitro experiments, i.e., direct quantification of bacterial activity. Transcriptomics is much more variable with and, just like metagenomics, only demonstrates potential and not activity. This argument is described in the first paragraph of the subsection “Abundance of carbon metabolism genes across algal cover gradient”.

The choice of a classification technique in place of standard statistical tools that work well with such data is not explained by the authors, and the details provided are insufficient to show that the result is statistically robust.

In our study, we used a regression random forest analysis (not classification – subsection “Abundance of carbon metabolism genes across algal cover gradient”, second paragraph) with algae cover as response variable, and a permutation test with corresponding p-values. Deciphering the relationships between microbial community composition, function, and environmental variables is a statistically challenging task due to the multi-dimensionality of these datasets. This challenge can be addressed by applying machine learning tools, such as random forests (Thompson et al., 2019, Chang et al., 2017, Edoardo et al., 2016, Hunter et al., 2016). The use of more standard multiple linear regression, inference, and model selection requires normality assumptions, which are not needed for a random forest. Since a random forest is a tree-based method it has the advantage that it can be implemented with a large number of predictor variables, and the variable importance ranks obtained from random forest may be used as a variable selection tool. We have included a paragraph in the Results and expanded the Materials and methods describing the random forest methodology:

“The relationship between fleshy algae cover and enzyme gene relative abundances was analyzed by the statistical learning method of supervised regression random forests (Breiman, 2001). […] Variable importance was selected based on the increasing mean squared error, a measure of the importance of a given feature for the correct prediction of a response variable, with α cutoff of 0.05. No outliers were removed from any analyses.”

2) Methodological and analysis errors with the oxygen ebullition experiments: The oxygen data does not appear to have undergone QA/QC, leading to spurious results based on readings entirely beyond the accurate range of the sensor, and in many cases beyond the sensor detection limits. There is no uncertainty analysis of resulting rates. The values presented are extremely supersaturated with respect to dissolved oxygen and characteristic of equilibrium conditions in the gas phase, meaning the results are unlikely to be environmentally relevant and may not be related to ebullition.

The reviewer mistook the values presented in Figure 4—source data 1: these values are not concentrations (µM), but actual number of mols, accounting for the volume of the incubation chamber, 1.5 L. We used calibrated probed, and performed an uncertainty quantification now included in the Source datasets provided with the manuscript. In the cases where dissolved O2 was above the 20 mg/L upper limit, we conservatively used the value 20 mg/L. To test whether ebullition is environmentally relevant we performed additional open-tank experiments (Figure 5—source data 1).

3) Ratios of DOC and oxygen uptake in incubations which are not consistent with the expectations laid out in the Discussion: The DOC and oxygen uptake rates are similar to the expected 1:1 ratio for fleshy algae exudates, while it is the control and coral experiments that have unexpected ratios. This implies that the expectation value presented is incorrect (because the rate of DIC increase is not equal to the rate of DOC uptake), that an unexplained process is occurring in the coral exudate and control samples, or that the uncertainties are substantial (there was again no uncertainty analysis of the rates in this experiment).

The reviewer states that “to use respiratory quotients with DOC it must be assumed that DOC uptake is equivalent to CO2 (DIC) release”. However, a fraction of the DOC consumed by a microbial cell is not oxidized to DIC, but instead allocated into biomass. This fraction greatly varies with cell physiology as reviewed in Russel and Cook, 1995. Therefore, we disagree with the reviewer’s statement that DOC:O2 ratios are equivalent to respiratory quotients.

4) Lack of discussion as to what composes the bacterial community in incubation experiments and how that informs or caveats interpretation of the experimental results: Without discussion of what is present in the experiments it is hard to interpret those results. Is community composition changing? Are certain clades dominating the observed responses? Does the implicit choice of community structure in the incubations bias the results towards certain outcomes (e.g. a community that normally grows near algae has a stronger response to algal exudates than to a carbon source it is not adapted for?).

The bacterial community utilized in all experimental conditions was the same: forereef communities, which represent offshore microbes that have not been exposed to the reef flat biogeochemistry and, therefore, were not yet selected or subject to implicit choice of community structure (subsection “Bacterial DOC and oxygen demands”). The bacterial community composition could not explain the trends in gene abundance (Results: subsection “Abundance of carbon metabolism genes across algal cover gradient”; Discussion: subsections “Bacterial overflow metabolism in algae-dominated reefs” and “Overflow metabolism predicts a shift in ecosystem biomass allocation”). None of the taxa that increased with algae cover consistently encode multiple copies of the genes studied, or lack these genes.

A longer explanation of these concerns (with specific details and some suggestions for ways forward) follows, with the idea that it may be useful to the authors should they choose to resubmit.

1) Metagenomics a) Metagenomics seems like an indirect tool to answer the question the authors have posed regarding gene expression as a function of algal cover. In the absence of transcriptomic data that identifies active gene expression, some logical leaps are required to connect gene abundance to active metabolic pathways in the microbial community, and the current paper does not provide sufficient detail to outline these or convince a reader that they are justified.

As discussed above, previous studies have shown the fast genomic adaptation of bacterial communities to environmental change – residence time – and the validity of using metagenomics to describe functional adaptation of bacteria (Nelson et al., 2013, Kelly et al., 2014). Additionally, the metagenomic analysis was used here as a hypothesis-generating tool, which was then tested using controlled experiments of bacterial physiology.

b) The statistical analysis of the metagenomic data is not adequately described for me to evaluate how robust it is. In particular there is no indication that the genetic data was normalized to bacterial abundance or some other sensible scheme to avoid biases due to very different sampled communities (dividing by the number of good reads is insufficient). It is unclear what if any null hypothesis was tested. 14% covariance between differences in the gene community and algal cover is a weak signal given that the five genes were explicitly chosen with the expectation of showing a signal, and at the chosen alpha value (0.05) 5% of results are false positives by definition. Using a random forest approach, the authors might evaluate the null hypothesis based on the distribution of variance explained over ~1000 random subsets of five genes, and then evaluate the significance of the 14% value relative to that null distribution (which in my experience is unlikely to have both mean and standard deviation equal to 0% as appears to be implicit in the current analysis).

The data actually consists of 87 independent metagenomes (subsection “Abundance of carbon metabolism genes across algal cover gradient”, first paragraph) and 23 genes of rate-limiting steps in central carbon metabolism and control genes (see the second paragraph of the aforementioned subsection and Supplementary file 1). Fleshy algae cover was the response variable of the random forest regression analysis. We have included more details on the number of trees grown in the forest, the examination of the MSE, and the permutation test, which incorporates a null hypothesis test (see the aforementioned subsection and subsection “Central carbon metabolism pathways”, second paragraph, diagnostic plots are shown in Figure 1—figure supplements 2 and 3). The abundance of bacterial taxa and the potential contributions of taxa with different genomes sizes was analyzing using FRAP (Cobian-Guemes et al., 2016) (–subsection “Central carbon metabolism pathways”, last paragraph).

c) I question why a random forest classification and regression approach was used when something like an ANOVA should work well and is commonly used with -omics data (with clearly testable hypotheses, or a PCA if not). Since a random forest approach was used, more details are required about the number of trees and tree size, the division of the data in training, validation, and test subsets and any resulting tuned hyperparameters, and finally whether stated errors refer to out of bag (classification) or misfit (regression) errors. My personal experience is that the default statistical software settings for random forests can lead to non-robust results, thus care must be taken in the selection (and description) of these characteristics.

The metagenomes were analyzed using a random forest regression with permutation test. The choice of a non-parametric, machine-learning approach was due to its higher performance compared to traditional multivariate statistics in the analysis of high-dimensional data (Thompson et al., 2019; Li, 2015; Liu et al., 2019). Random forests are less sensitive to training set size, and are as good at predicting feature importance as neural networks and traditional indicator species approaches. In our study, the random forest was grown using 1000 trees and 4 tree splitting variables. An examination of the mean squared error plot of the random forest indicates that enough trees have been used (Figures 1—figure supplements 2 and 3). The permutation test p-values are obtained using 100 permutation replicates (subsection “in Abundance of carbon metabolism genes across algae cover gradient”, second paragraph; subsection “Central carbon metabolism pathways”, second paragraph). ANOVA and PCA are not appropriate tests in our case, as we are not comparing groups or performing clustering.

d) Since the metagenomes are already assembled the authors should know which species/strain level bacteria are present. Thus I would think a more convincing assessment of the effect of copy number variability would be to look at orthologs and paralogs in at the species/strain level and then evaluate if certain extremes of algal cover (very low, very high) are associated with particular species with many gene replicates or deletions that drive the overall trend. Doing this at the genera level as is the current choice could mute any problematic signal, unless the genera is monospecific. The manuscript states, however, that there is in fact high strain level variation, meaning that this choice is potentially masking a problem. At a minimum, a good reference should justify the choice. And this analysis still does not address the potential complication of variability in gene expression between sites. Finally, from Supplementary file 1 it appears that autotrophic Prochlorococcus has a stronger relationship with algal cover than the heterotrophic bacteria that are the focus of this work. It may be necessary to confirm Prochlorococcus (and Synechococcus, etc.) are not obfuscating your signal.

The taxonomic profiles of the community are reported in Figure 1—source data 1 and discussed (subsection “Abundance of carbon metabolism genes across algal cover gradient”, last paragraph; subsection “Bacterial overflow metabolism in algae-dominated reef”, last paragraph and subsection “Overflow metabolism predicts a shift in ecosystem biomass allocation”, last paragraph). We also included it as Figure 1—source data 1 in the current version. None of these taxa varying with algae cover (including Prochlorococcus) display consistent trends in the absence, presence, or copy number variation in the genes with highest importance in the functional analysis. The predictions from the functional metagenomic data from in situ microbial communities were consistent with physiological analysis in vitro, despite the high dimensionality of the data and complexity of environmental influences on microbes

2) Oxygen ebullition experiments a) Based on the supporting data Figure 4—source data 1, it appears that there was a failure to do QA/QC on the oxygen concentrations in the POP experiments. The dissolved oxygen concentration at the end of all of the non-control experiments are outside the stated accurate range for the instrument used, 0-20 mg/L or 0-200% saturation, whichever is lower (https://www.hach.com/hq40d-portable-multi-meter-ph-conductivity-salinity-tds-dissolved-oxygen-do-orp-and-ise-for-water/product-details?id=7640501639). Since this is an optode based optical system, the concentration based range is more notion than reality and the saturation state should always be the limiting bound at high oxygen conditions. Optode based dissolved oxygen sensors have nonlinear responses to concentration, so the deviation from the true value may vary in a way that is not readily predictable (even within the 0-200% range the accuracy varies substantially). Not only are all the experiments beyond the accurate range, many of them (including all the CCA experiments) are beyond the instrument bounds (the internal QA filter is apparently set to 30 mg/L given the constant endpoint values of ~935 μm in many experiments).

As described above, the reviewer mistook the values presented in Figure 4—source data 1. These values are not concentrations (µM), but number of mols in the chamber, accounting for the volume of the incubation chamber, 1.5 L. We used calibrated probed, and performed an uncertainty quantification now included in the Source datasets provided with the manuscript. In the cases where dissolved O2 was above the 20 mg/L upper limit, we conservatively used the value 20 mg/L. To test whether ebullition is environmentally relevant we performed additional open-tank experiments (Figure 5—source data 1).

b) Hypothetically the above experimental data from below the upper instrument detection bound could be salvaged, if a full range calibration to Winkler titrations or another sensor were to be run on the same sensor cap originally used, at each of the experimental temperatures (the oxygen sensitivity of the optode foil is temperature dependent). See Bushinsky and Emerson, 2013 for an example of what this might look like (https://doi.org/10.1016/j.marchem.2013.05.001). But the sensor characteristics have a substantial "grow-in" if they are new, and change if they are dry for an extended period, so this would have the best results if the sensor cap had already seen several months of use prior to the experiments and has been kept continuously moist since. But without the beyond detection samples there is insufficient experimental power to determine significance…

The misinterpretation of concentration versus absolute values in Figure 4—source data 1 were now clarified. The uncertainty propagation calculates are provided in source data for Figures 2, 3 and 4, and described in the Materials and methods (–subsection “Uncertainty quantification”).

c) I'm concerned with how the sampling of the headspace gas is described. The manuscript says the gas was transferred from a syringe to a wide mouthed container that the second probe was placed into. Gas diffuses quickly, so I don't see how this could lead to accurate measurements-but I may be misunderstanding, and this refers to the dissolved sample? At any rate this description needs to be clarified.

Air-tight containers and valves were used, and the oxygen concentration in the sample was measured immediately upon collection by transferring the gas volume to the wide-mouth container located next to the incubation chamber. All of the containers and incubation chambers were underwater inside a larger holding tank, and no gas was lost during quantification procedures (subsection “O2 release by benthic primary producers”, first paragraph).

d) Setting aside the accuracy issues above, all of the headspace values are approximately in equilibrium with the dissolved oxygen you measured. Henry's law solubility coefficients predict concentrations of 0.36-0.94 mg/mL and 1.3-20.6 mg of oxygen in the gas phase above the water samples-close to the 0.4-0.85 mg/mL and 1.1-17 mg O2 measured (assuming 1 atm pressure, and neglecting any lag between production and diffusion). Thus I think that the presence of a headspace isn't proof of the importance of ebullition in a natural environment-it is proof that a highly supersaturated solution will push gas into a headspace (like CO2 is in a non-experimental pop bottle). I do suspect bubble nucleation was essential, since the controls were also supersaturated and formed no headspace, but they were a lot less supersaturated so this would be more convincing with evidence of rising bubbles in the chambers-otherwise thermodynamics alone is sufficient without any biophysical interactions. At any rate what the above suggests to me is these chambers are unlikely to reflect environmental outcomes regarding the amount of oxygen released in bubbles. In this light it is very hard to support the statement that in algal dominated reefs 1/3 of the oxygen escapes as bubbles.

We now provide data showing the bubbles being formed on the algae and rising to the surface in an open system, where the oxygen concentration in the water can equilibrate with the atmosphere (Figure 5—source data 1, Videos 1 and 2). No relationship was observed between dissolved oxygen concentrations in the water and ebullition rates, suggesting heterogenous nucleation due to supersaturation on the algae surface microenvironment (Figure 5B, Videos 1 and 2 and subsection “Oxygen production, consumption, and ebullition”, second paragraph).

e) There is no uncertainty analysis of these rates. Given the relatively coarse measurements of bubble volume (to nearest 0.5 mL), short durations (2-4 hours), and uncertainties in saturation state (even when the instrument is within its specified range), both the gas production and dissolved oxygen change must have significant uncertainties, but these are not quantified. I think they should be, and relative uncertainty of each experiment shown so a reader can gauge how significant differences between the box and whisker categories actually are. Also, 2 pts don't make for a convincing rate.

We performed an uncertainty quantification of all incubation data, which is provided in the source data files for Figures 2, 3 and 4. All error propagated from instrument uncertainty was at least five times smaller than statistical standard deviations, and was not considered in the statistical tests (–subsection “Uncertainty quantification”).

3) DOC and oxygen uptake experiments

The DOC and oxygen uptake ratios in these experiments are far from 1:1, but not in the manner described to support arguments about less complete oxidation of organic matter. To use respiratory quotients with DOC it must be assumed that DOC uptake is equivalent to CO2 (DIC) release. The arguments put forward in this manuscript about the incomplete oxidation of organic matter would argue against this for the macroalgae, but I'll set that aside for the moment. At any rate, the -DOC/-O2 ratio for the fleshy algae exudate is ~0.9, right in line with theoretical and measured values of +DIC/-O2 for bacterioplankton (Robinson 2019, https://doi.org/10.3389/fmars.2018.00533). It is instead the controls and corals that seem to have unexpected values on the order of 0.3. So I see three options: 1) the -DOC/-O2 rate ratio shouldn't actually be ~1:1 (or rather 0.9), and thus the discussion of respiratory quotients is a red herring. 2) the corals and controls are doing something unusual that requires an explanation. 3) There are major biases in the rates. I think the error analysis I recommended for the O2 production fluxes should be done here as well and could inform this question. So overall I agree that the incubations show more oxygen consumption per unit of DOC consumed with the control and coral exudates than with the algal exudates, but since none of these categories meet the stated expectations this needs to be explored in more detail to confirm that the result is real and robust.

As described above, we disagree with the reviewer statement that “to use respiratory quotients with DOC it must be assumed that DOC uptake is equivalent to CO2 (DIC) release”, because a fraction of the DOC consumed by a microbial cell is not oxidized to DIC, but instead allocated into biomass (reviewed in Russel and Cook, 1995). When respiratory quotients were mentioned in the previous version, they referred to primary producer metabolism, not bacterial consumption. We rephrased this section to avoid any confusion (subsection “Ebullition causes O2 loss from coral reef”).

The interpretation that an increasing part of the organic carbon consumed by bacteria is not fully oxidized is supported by the decoupling between organic carbon and oxygen consumption (Figure 2), and the higher biomass accumulated per cell (Figure 3).

4) Bacterial community structure

The bacterial community structure is not discussed. How it diverges across sampled environments is challenging to ascertain in Figure 1. I don't find any information on what the bacterial community looked like within the various incubation experiments.

The bacterial community composition results are discussed in subsection “Bacterial overflow metabolism in algae-dominated reef” and subsection “Overflow metabolism predicts a shift in ecosystem biomass allocation”. We also include the data as source data for Figure 1 in this version of the manuscript.

Thus I can't evaluate whether the incubation results are a function of, e.g., incubating a community adapted to one particular environment in a way that would bias the experimental outcomes. While there is a mention about changing community structure with regards to the incubations (Discussion), there is not data in the supplement regarding the community in these experiments, nor would I expect changing genetic composition over the course of experiments lasting hours to days (again, a question for transcriptomics?). This makes it hard to interpret what is happening in the experiments where change in biomass is evaluated. Are changes in biomass due primarily to a single clade? Are certain larger types of bacteria being selected for, or are particular bacteria growing larger while others do not? Is this an expected result based on where those bacteria normally are environmentally successful?

The bacterial inoculum for the experiments was a forereef water column community, which represents the offshore community not yet adapted to a reef environment (subsection “Bacterial DOC and oxygen demands”). These communities reach the reef with incoming currents, and are exposed to reef carbon sources and biogeochemical environment. They are selected in the time frame of water residence times as the offshore water flows in and out of reef lagoons (hours), the same time frame of incubation experiments performed here (Nelson et al., 2011, 2013). In the experiments, bacterial communities are subjected to similar carbon sources and oxygen environment that the offshore communities encounter as they reach coral-dominated or algae-dominated reefs. Therefore, the selection pressures are similar and should lead to the similar physiological responses and genomic adaptation.

Specifically, the changes in cell size observed in situ cannot be attributed to changes in abundance of a single clade. The only clade showing high relative abundance that changed with algae cover was Prochlorococcus, which cell volumes comparable to that observed in offshore water, ~ 0.1µm3 (Kirchman, 2016), which is the average cell volume observed in the inoculum community of our incubation experiments. The other groups with significant relationships with fleshy algae cover contributed to only 0.9 to 1.4% of the community, and their average cell size is not consistent with the observed increase in cell size: Thermotoga has large cells, but their abundance decreased with algae. Leuconostoc, Rhodopirellula, and Methanobacterium also decrease in abundance with increasing algae. Therefore, none of these clades can explain the change in size observed in this study. We have included this detailed description in the Discussion (subsection “Overflow metabolism predicts a shift in ecosystem biomass allocation”).

Reviewer #2:

This study provides a theoretical and experimental based evidence approach to explain the observations for O2 depletion in algal dominated coral reefs. The processes of ebullition and microbial metabolism are postulated to drive this lower observed oxygen level in algal dominated reefs. Overall the manuscript is very interesting and details a nice combination of environmental genomics based analysis linked with controlled experimental trials. The postulated physical and biological reasons for reduced O2 levels are supported by the experimental data, though the conclusions are still a little speculative. However this approach represents a great approach to hypothesis driven science; and while not able to definitely confirm these process (i.e. ebullition and microbial metabolism) as driving O2 depletion, it sets up the concepts to be further tested in environmental settings. Hence the manuscript is worthy of publication with a few comments detailed below highlighting potential areas for improvement.

We thank the reviewer for the comments, which we fully took into consideration to write the revised version, including rephrasing the conclusions to avoid inappropriate extrapolations of the in vitro data to in situ dynamics.

[Editors’ note: the author responses to the re-review follow.]

Thank you for choosing to send your work entitled "Biophysical and physiological processes causing oxygen loss from coral reefs" for consideration at eLife.

We thank you for sharing your rebuttal letter. We would welcome a revision of your work for a new cycle of review and note that the revision should address the following concerns:

The text is missing or unclear on the rationale for a number of the statistical and analytical choices, and has weaknesses in some of the lines of evidence, including unresolved problems with the oxygen production experiments. Thus the current manuscript, in the opinion of two reviewers, presents the interpretations with a certainty and forcefulness which we did not think is warranted by the present constellation of suggestive (but not definitive) evidence. For an independent opinion, you may consider consulting with Dr. Clara Fuchsman (Horn Point Laboratory, cfuchsman@umces.edu).

We thank the editor and reviewers for the comments. We expanded the Materials and methods to describe the random forests in more detail (subsection “Central carbon metabolism pathways”, second paragraph), and also provided rationale for choosing the random forests (subsection “Abundance of carbon metabolism genes across algal cover gradient”, second paragraph) and other statistical tests performed (–subsection “Statistical tests”).

Regarding the evidences for the oxygen measurements, two important changes were made:

i) Additional experiments were performed to quantify the coral and algae ebullition rates in an open system (Figure 5). These experiments show that ebullition occurs due to bubbles forming on the algae surface. These bubbles detach from the algae and rise though the water column (Videos 1 and 2). There was no relationship between the rates of ebullition and dissolved oxygen concentration in the water column (Figure 5), suggesting that ebullition does not occur due to supersaturation in the water column, but through of heterogenous nucleation on the algae surface. The rates observed in the new experiments are within the range of rates estimated to be necessary to explain the gas produced in the previous POP experiments (Figure 4). A description of how these predicted rates were estimated is provided in the Materials and methods (subsection “Estimates of oxygen loss by ebullition”), and the Results section was updated pointing out that supersaturation was observed in some POP chambers and this issue was addressed in the open-tank experiment (subsection “Oxygen production, consumption, and ebullition”, second paragraph).

ii) Our main conclusion is that microbial respiration and ebullition contribute to the loss of a significant fraction of the oxygen produced by reef primary producers. This conclusion is supported by in vitro experiments in this study. In the current version, we included a weighted linear model (subsection “Predictive model”, Figure 5D) that incorporates the respiration rates from the dark incubation experiments (Figure 2), microbial abundances sustained by primary producers (Figure 3), and oxygen production rates (Figure 4) into an idealized reef with varying coral and algae cover. The model predicted the proportion of the gross oxygen production consumed by microbial respiration and ebullition across the algae cover gradient. This prediction sets a baseline for future in situ studies.

Thank you for clarifying the oxygen units. However, all of the CCA and most of the fleshy macroalgae experiments were beyond the detection bounds of your instrument (20 mg/L), as evidenced by the constant endpoint concentrations (937.5 umol/1.5L=625µM=20 mg/L) across several nominally independent experiments. The headspace concentrations also still appear to reflect equilibrium conditions with the underlying water when the solution is strongly supersaturated (as it is in most of the non-control experiments). How relevant are these experiments for explaining oxygen loss under realistic environmental conditions (less supersaturated and open to the atmosphere)?

We described this issue in the current version (subsection “Oxygen production, consumption, and ebullition”) and performed new open-tank experiments that are closer to realistic environmental conditions due to the equilibration with atmospheric oxygen. These experiments showed significantly high rates of ebullition in algae compared to corals (Figure 5—source data 1, Videos 1 and 2). The ebullition rates were not correlated with the dissolved oxygen concentration in the water, indicating that ebullition occurs due to heterogenous nucleation on algae surfaces. Ebullition has been previously detected in the field on coral reefs (Odum and Odum, 1955, Bellamy and Risk, 1982; Clavier et al., 2008 Freeman, 2018), although its contribution to the oxygen budget has never been quantified. Recently, ebullition was demonstrated to account for the loss of up to 21 and 37% of the O2 production in a salt marsh and a lake, respectively (Howard et al., 2018; Koschorreck et al., 2017). This phenomenon may be increasingly important due to the current algae phase-shifts occurring on coral reefs (Figure 5D).

There are statistical techniques for dealing with non-normally distributed data (or more accurately, non-conditionally normal error distributions-A random forest regression may well be a good choice for this analysis, but a rationale should be explained, and the particulars of how it is used clearly stated.

We agree and included a justification for the use of random forest (subsection “Abundance of carbon metabolism genes across algal cover gradient”, second paragraph) and a description of the test and parameters used (subsection “Central carbon metabolism pathways”, second paragraph).

Whether 5 or 16 genes, the correlations to environmental variables need more context. Our concern remains about the effect size versus a null hypothesis of some other random set of genes from your already sequenced genomes, which are far larger than 16 genes. The details required to assess what differences are "real" are currently insufficient. e.g. was the data normalized to bacterial abundance? A single 100mb genome will have far more reads than a 1mb genome, even if there are 10 times more of the smaller-genome organism. You may well have made considered choices for these and other potential issues in your analysis, but you tell the reader none of this, even in the supplement.

We approached the issue of finding genes that are truly correlated with algae cover in two ways:

i) We performed a permutation version of the random forest test, which has an implicit null hypothesis test. The five genes reported were the only ones with a p-value of less than 0.05 in this test (–subsection “Abundance of carbon metabolism genes across algal cover gradient”, second paragraph and legend of Figure 1). The data is not normalized by bacterial abundance because the goal is to test the relative enrichment or depletion of genes within a community. Normalizing by bacterial abundance would skew this analysis because an increase in cell abundance would drive the increase in gene abundances even with no relative enrichment for specific functions. We also used control genes, such as rpoB, a RNA polymerase gene that is housekeeping and single-copy, not expected to change in relative abundance across the environmental gradients. Oxidative stress genes did not change in abundance either.

ii) To test the effect of differences in genome sizes and the likelihood of organisms with larger genomes contributing disproportionately to the observed trends, we analyzed the effect of taxonomic composition on the functional changes. The current version includes analyses at genus and species level. The abundance of each taxon was calculated using the FRAP method, which accounts for the probability of a given read mapping to a genome accounting for the genome size (Cobian-Guemes et al., 2016) (–subsection “Central carbon metabolism pathways”, last paragraph). We then searched if the taxa varying in abundance consistently display copy number variations in the genes that significantly vary with algae cover. None of the taxonomic transitions were able to explain the functional trends observed (–subsection “Abundance of carbon metabolism genes across algal cover gradient”, last paragraph). More details on the influence of Prochlorococcus abundance are below in the response to reviewers and in the text (subsection “Abundance of carbon metabolism genes across algal cover gradient”, last paragraph; subsection “Bacterial overflow metabolism in algae-dominated reef”, last paragraph, –subsection “Overflow metabolism predicts a shift in ecosystem biomass allocation”, last paragraph).

Regarding your choice of approach for the metagenomic data. Please provide context and defend the choice (and aspects of the analysis of this data) in the text. It would be a good choice to highlight the rationale of this approach in as succinct a manner in the text as you do in your rebuttal letter.

We included the rationale for this analysis at the beginning of the Results section, immediately before we present the metagenomic results: “Different carbon consumption pathways employed by bacteria are associated with varying levels of oxygen consumption rates (Russell and Cook, 1995). […] On coral reefs, metagenomic data reflects the strain-level genomic adaptation that occurs within the timescale of residence time as offshore microbial communities are transported onto the reef environment, where water masses get enriched with benthic exudates changing their biogeochemistry (Nelson et al., 2011, Kelly et al., 2015).”

https://doi.org/10.7554/eLife.49114.sa2

Article and author information

Author details

  1. Cynthia B Silveira

    1. Department of Biology, San Diego State University, San Diego, United States
    2. Viral Information Institute, San Diego State University, San Diego, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Supervision, Validation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    For correspondence
    cynthiabsilveira@gmail.com
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9533-5588
  2. Antoni Luque

    1. Viral Information Institute, San Diego State University, San Diego, United States
    2. Computational Science Research Center, San Diego State University, San Diego, United States
    3. Department of Mathematics and Statistics, San Diego State University, San Diego, United States
    Contribution
    Formal analysis, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  3. Ty NF Roach

    Hawaii Institute of Marine Biology, University of Hawaii at Mānoa, Kāneohe, United States
    Contribution
    Conceptualization, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  4. Helena Villela

    Department of Microbiology, Rio de Janeiro Federal University, Rio de Janeiro, Brazil
    Contribution
    Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  5. Adam Barno

    Department of Microbiology, Rio de Janeiro Federal University, Rio de Janeiro, Brazil
    Contribution
    Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  6. Kevin Green

    Department of Biology, San Diego State University, San Diego, United States
    Contribution
    Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  7. Brandon Reyes

    Department of Biology, San Diego State University, San Diego, United States
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
  8. Esther Rubio-Portillo

    Department of Physiology, Genetics and Microbiology, University of Alicante, Alicante, Spain
    Contribution
    Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  9. Tram Le

    Department of Biology, San Diego State University, San Diego, United States
    Contribution
    Investigation
    Competing interests
    No competing interests declared
  10. Spencer Mead

    Department of Biology, San Diego State University, San Diego, United States
    Contribution
    Investigation
    Competing interests
    No competing interests declared
  11. Mark Hatay

    1. Department of Biology, San Diego State University, San Diego, United States
    2. Viral Information Institute, San Diego State University, San Diego, United States
    Contribution
    Investigation, Methodology
    Competing interests
    No competing interests declared
  12. Mark JA Vermeij

    1. CARMABI Foundation, Willemstad, Curaçao
    2. Department of Freshwater and Marine Ecology, Institute for Biodiversity andEcosystem Dynamics, University of Amsterdam, Amsterdam, Netherlands
    Contribution
    Resources, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  13. Yuichiro Takeshita

    Monterey Bay Aquarium Research Institute, Moss Landing, United States
    Contribution
    Conceptualization, Writing—review and editing
    Competing interests
    No competing interests declared
  14. Andreas Haas

    NIOZ Royal Netherlands Institute for Sea Research, Utrecht University, Texel, Netherlands
    Contribution
    Conceptualization, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  15. Barbara Bailey

    Department of Mathematics and Statistics, San Diego State University, San Diego, United States
    Contribution
    Data curation, Formal analysis, Writing—review and editing
    Competing interests
    No competing interests declared
  16. Forest Rohwer

    1. Department of Biology, San Diego State University, San Diego, United States
    2. Viral Information Institute, San Diego State University, San Diego, United States
    Contribution
    Conceptualization, Resources, Supervision, Investigation, Methodology, Writing—review and editing
    Competing interests
    No competing interests declared

Funding

National Science Foundation (G00009988)

  • Ty N F Roach

Gordon and Betty Moore Foundation (3781)

  • Forest Rohwer

Conselho Nacional de Desenvolvimento Científico e Tecnológico (234702)

  • Cynthia B Silveira

Spruance Foundation

  • Cynthia B Silveira

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank the crew and captains of the NOAA ship Hi’ialakai and the Waitt Institute ship Plan B that contributed to sampling. We thank the CARMABI and HIMB for the support conducting field experiments. We thank Robert Edwards at SDSU for access to computer servers funded by the NSF (CNS-1305112 to Robert A Edwards). This work was funded by the Gordon and Betty Moore Foundation (grant 3781 to FR) and Spruance Foundation. CBS was funded by CNPq (234702) and Spruance Foundation. TNFR was supported by the NSF (G00009988).

Senior Editor

  1. Ian T Baldwin, Max Planck Institute for Chemical Ecology, Germany

Reviewing Editor

  1. Evan Howard

Publication history

  1. Received: June 6, 2019
  2. Accepted: October 27, 2019
  3. Version of Record published: December 3, 2019 (version 1)

Copyright

© 2019, Silveira et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 253
    Page views
  • 20
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)