We conducted six oceanographic surveys: two during autumn (October and November 2016), two during winter (February 2016 and January 2017), one in summer (August 2017), and one in spring (March 2018) on board the R/V Thuwal and R/V Al Azizi. Sampling was conducted following a latitudinal transect along the Red Sea within a region limited by coordinates 17.25 to 27.82∘ N and 34.83 to 41.39∘ E (Fig. 1). At each station, vertical profiles of temperature and salinity were obtained with a Sea-Bird SBE 911plus CTD profiler (Sea-Bird Electronics, Bellevue, WA, USA), equipped with additional sensors to measure the attenuation of photosynthetically active radiation (PAR) (Li-cor biospherical PAR sensor), in vivo fluorescence (WetLabs ECO-FL fluorometer), and dissolved oxygen concentration (Sea-Bird SBE 43 dissolved oxygen sensor). Water samples for chemical and biological measurements were collected between 07:00 and 09:00 local time, using a rosette sampler equipped with 12 Teflon Niskin bottles (12 L) that were provided with silicone O-rings and seals.

Water samples for nutrient analyses were collected in 50 mL polyethylene bottles and kept frozen (-20 ∘C) until further processing. Inorganic nutrient concentration was determined with a SEAL AA3 segmented flow analyser (SEAL Analytical Inc., WI, USA) using standard methods (Hansen and Koroleff, 1999). The detection limits were 0.05 µM for nitrate, 0.01 µM for nitrite, 0.01 µM for phosphate, and 0.08 µM for silicate. For the chlorophyll a analysis, 200 mL samples were taken at 10 discrete depths (between 5 and 200 m) and filtered through Whatman GF/F filters. The filters were kept frozen (-20 ∘C) until further analysis. Pigments were extracted for 24 h using 90 % acetone and left overnight in the dark at 4 ∘C. The Chl a concentration was estimated with the non-acidification technique using a Trilogy fluorometer equipped with a CHL-NA module (Turner Designs, San Jose, USA), previously calibrated with pure Chl a.

Plankton metabolic rates were determined in vitro by measuring the changes in dissolved oxygen concentration after 24 h light–dark bottle (Winkler) incubations (Carpenter, 1965). This methodology, commonly used to determine plankton metabolic rates (Williams et al., 1979; Duarte and Agustí, 1998; Bender et al., 1999; Robinson and Williams, 1999; Ducklow et al., 2000; Serret et al., 2001, 2009; Robinson et al., 2002; García-Martín et al., 2017), allows the diel cycle of oxygen and carbon fluxes derived from photosynthetic mechanisms (light-dependent reactions) and also those linked to the acquisition of energy by both autotrophic and heterotrophic microorganisms (light- and dark-dependent reactions)(Robinson and Williams, 2005; Williams and del Giorgio, 2005) to be accounted for .

Water samples were collected at three different optical depths (ζ) through the water column. One at the surface (100  %–80  % of incident PAR), another towards the bottom of the photic layer (8  %–1  % of incident PAR), and one intermediate sample at a depth of the chlorophyll maximum (Chl a max). In case the Chl a max was sampled at the surface or bottom layers, the intermediate sample was taken between 1.5 and 2.3 ζ (i.e. 22  %–10  % of incident PAR). Seawater was collected directly from the Niskin bottles to fill a total of 21 (100 mL) Winkler bottles. The bottles were carefully filled using silicone tubing, and the water was allowed to overflow during the filling, taking special care to avoid the formation of air bubbles. Surface samples were collected in 100 mL quartz bottles. From each depth, seven of the bottles were immediately fixed with manganese sulfate (MnSO4) and potassium hydroxide / potassium iodide solution (KI / KOH) to determine the initial oxygen concentration while the other 14 bottles, 7 light and 7 black, were incubated on deck in surface water flow-through tanks. Due to the difference in temperature between the surface and deep waters, particularly during the summer and autumn surveys, we decided to include in our analyses only those samples collected above the thermocline. Changes in temperature and PAR in the incubation tanks were recorded with HOBO Pendant^® data loggers (Onset, Massachusetts, USA).

Before the incubation, the bottles were covered with neutral mesh to reduce the incident PAR according to the sampled depth. At the end of the incubation period, light and dark bottles from each depth were fixed to determine final O2 concentrations. The oxygen concentration was measured by automated high-precision Winkler titration with a potentiometric end-point detection (Oudot et al., 1988) using a Mettler Toledo T50 Titration Excellence auto-titrator attached to an InMotion autosampler. NCP was calculated as the difference in the oxygen concentration between the light bottles after the 24 h incubation period ([O2]L-24h) and the oxygen concentration measured before the incubation ([O2]Tzero) (i.e. NCP = ([O2] L-24h - [O2]Tzero). CR rates (mmol O2 m-3 d-1) were calculated as the difference of the oxygen concentration after the 24 h incubation period in the dark bottles ([O2]D-24h) and the initial oxygen concentration ([O2]Tzero) (i.e. CR = [O2]Tzero - ([O2]D-24h). GPP (mmol O2 m-3 d-1) was calculated as the sum of NCP and CR.

Due to the consistent relationship existing between plankton metabolism and temperature across diverse marine regions (Regaudie-de-Gioux and Duarte, 2012; García-Corral et al., 2014), we examined how plankton metabolic rates covariate with temperature in the Red Sea, a system whose temperature range is higher than previously encountered in marine planktonic metabolism research. We determined the relationship between metabolic rates and temperature by fitting an ordinary least squares linear regression equation to the relationship between the natural logarithm of the Chl a-specific metabolic rates (B0) and the inverse of the absolute temperature × k (i.e. 1/kT), where k is the Boltzmann's constant (8.617734×10-5 eV K-1) (Gillooly et al., 2001; Brown et al., 2004): 1LnB0=-Ea(1/kT)+C.

In these so-called Arrhenius plots, the slope of this relationship represents the average activation energy (Ea = negative slope), characterising the extent of thermal dependence of metabolic processes.

Statistical analyses and figures were done using the statistical and machine learning toolbox in MATLAB version R2018b (Mathworks Inc, Natick, MA, USA) and with the R statistical computing package using RStudio 1.1419. Pearson correlation tests were used (corrplot function in R) to determine the relationship between environmental variables (temperature, nitrate + nitrite (NOx), phosphate, and silicate concentration) and their latitudinal distribution, and to determine the relationship between volumetric measurements of GPP, CR, NCP, and environmental variables (temperature, NOx concentration, Chl a, and latitude). We used ordinary least squares (OLS) simple regression models (fitlm function in MATLAB) to describe the potential relationships between different planktonic metabolic rates and between metabolic rates and environmental variables, and to predict the response of the Chl a-normalised GPP (and CR) to temperature (Arrhenius plots described in Sect. 2.3). To test if the activation energies (obtained from the Arrhenius plots) were significantly different, we performed an analysis of covariance (ANCOVA) by using the aoctool in MATLAB. The variability of planktonic metabolic rates between cruises was statistically analysed using non-parametric Kruskal–Wallis tests. Mean values and their standard error (SE) are reported throughout the text.

Hydrographic (temperature and salinity) and chemical variables (nutrient concentrations) depicted a marked latitudinal gradient typical of the Red Sea. At the southern-most area, sea surface temperature (SST) fluctuated between 28 ∘C (winter–spring) and 32 ∘C (summer), while at the far-northern sampling site SST ranged between 23 ∘C (winter) and 27–28 ∘C (summer–autumn) (Fig. 2). Overall, all macronutrients observed a significant inverse correlation with latitude (Pearson correlation coefficients r<-0.4, p<0.05) (Fig. 3). Nitrite + nitrate (NOx) decreased from 6.1±0.9 µM in the southern region to 2.9±0.3 µM towards the northern Red Sea, while on average, phosphate concentration ranged from 0.5±0.01 µM in the south of the Red Sea to 0.1±0.01 µM towards the northern stations (data not shown). Phytoplankton biomass (measured as Chl a concentration) also decreased significantly towards the north of the Red Sea (Pearson's correlation, r = -0.4, p<0.001, and n = 77) (Table 1). We found the highest autotrophic biomass during the autumn and winter cruises. During this period, surface Chl a ranged from 0.6 to 0.8 mg m-3 in the southern region and ranged from 0.2 to 0.3 mg m-3 in the north (Fig. 2). In general, our results confirm that all variables correlated significantly with latitude, highlighting the prevalence of the south–north gradient in temperature, salinity, nutrient availability, and chlorophyll a concentration across the Red Sea.

Analogous to the environmental variability, planktonic metabolism followed the same significant north–south decreasing pattern with latitude (Fig. 4). The inverse correlation of GPP rates with latitude was highly significant (Pearson correlation coefficient r = -0.6, p<0.001, n = 77) (Table 1), as found for autotrophic biomass, thus explaining the strong correlation observed between GPP and Chl a concentration (Pearson correlation coefficient r = 0.7, n = 77) (Table 1). GPP rates decreased on average by 79 %, from 4.1±0.5 mmol O2 m-3 d-1 (≈49.2 mgC m-3 d-1, assuming the photosynthetic quotient (PQ) = 1) at the southernmost station of the Red Sea to 0.9±0.1 (≈ 10 mgC m-3 d-1; PQ = 1) at the northern site, while CR decreased on average by 73  %, from 3±0.4 mmol O2 m-3 d-1 (≈36 mgC m-3 d-1, assuming the respiratory quotient (RQ) = 1) in the south to 0.8±0.1 in the north (≈9.6 mgC m-3 d-1; RQ = 1) (Fig. 4). We did not find any significant correlation between NOx availability and GPP (Pearson correlation coefficient, p>0.05, n = 56), CR (Pearson correlation coefficient, r = 0. 2, p>0.05), or NCP rates (Pearson correlation coefficient, r = -0.2, p>0.05, n = 56) (Table 1); however, all metabolic rates were positively correlated with temperature (Table 1).

The highest GPP and CR rates measured along the Red Sea came from data collected during the autumn and winter cruises, when GPP and CR rates reached values above 6 and 4 mmol O2 m-3 d-1, respectively (Fig. 5), and when the mean values were the highest (GPPautumn-winter = 2.9±0.3–2.3±0.3 mmol O2 m-3 d-1; CRautumn-winter= 2.5±0.3–2±0.2) (Fig. 5). However, despite the overall variability between autumn–winter and spring–summer seasons, when all data are considered together, planktonic GPP and CR rates were not significantly different between seasons (Kruskal–Wallis H test: χ2=6.83, p = 0.08; χ2=4.14, p = 0.25, respectively). Furthermore, the balance between planktonic autotrophic production (GPP) and respiratory losses (due to the heterotrophic and autotrophic metabolism, CR) (i.e. NCP rates) revealed that NCP rates also decreased towards the northern region (by 94 %). From 1.1±0.3 mmol O2 m-3 d-1 at the southern stations to 0.1±0.1 mmol O2 m-3 d-1 above 26∘ N (Fig. 4). The average NCP measured during our cruises was 0.3±0.1 mmol O2 m-3 d-1 (Fig. 5), which indicates an overall prevalence of autotrophic communities (Fig. 5). However, a closer look at our data revealed the mean NCP rate in spring was -0.3±0.2 mmol O2 m-3 d-1 (Fig. 5), while during summer, NCP rates in the northern region ranged from -0.6 to -0.1 mmol O2 m-3 d-1, which provides evidence that planktonic metabolism was governed by heterotrophic communities during both spring and summer in the northern region.

When we evaluated the relationship of GPP with CR and NCP, the analysis showed that both CR and NCP increased significantly with GPP (R2=0.62 and 0.49, respectively; p<0.001) (Fig. 6). From the functional relationships between GPP with CR and NCP, we calculated the threshold of GPP for metabolic equilibrium for the region. By solving for GPP = CR and for NCP = 0 (from the relationship describing NCP as a function of GPP), and by using the slope and intercept shown in Fig. 6a and b, we determined that the GPP threshold that separates autotrophic from heterotrophic planktonic communities in the Red Sea is 1.7 mmol O2 m-3 d-1 (range 1.2–1.9 mmol O2 m-3 d-1).

Due to the pervasive influence of temperature in regulating metabolic rates, we further explored the temperature dependence of GPP and CR by analysing the relationship between chlorophyll a-specific metabolic rates and temperature. Our analysis revealed that both GPP and CR tended to increase with temperature albeit with different activation energies, i.e. Ea was significantly higher for GPP (1.20±0.17 eV) than for CR rates (0.73±0.17 eV) (ANCOVA, F = 3.96, p = 0.04) (Fig. 7). We also tested whether the temperature-dependence response was consistent between cruises (Fig. 8). Our results indicated a relatively higher activation energy for GPP during the summer cruise (2.31±0.75 eV) and for CR in spring (2.60±0.85 eV). However, the observed differences in the activation energies for GPP were not significantly different between seasons (ANCOVA, F = 0.38, p = 0.8).

Our results demonstrate that planktonic metabolic rates are markedly different between the southern and northern regimes of the Red Sea, with a northward increase in the overall mean GPP and CR by factors of 5 and 4, respectively (i.e. an absolute increase in GPP rates of 3.2 mmol O2 m-3 d-1≈38.4 mgC m-3 d-1, while absolute CR rates increased by 2.2 mmol O2 m-3 d-1≈26.4 mgC m-3 d-1). Although, sensu stricto, the overall balance between autotrophic metabolism and planktonic community respiration (i.e. NCP) indicated a prevalence of autotrophic communities during our samplings along the Red Sea. Heterotrophic communities prevailed during the spring, and in the northern stations during the summer, which highlights the shift in the trophic conditions in the basin. Consistent with these findings, our data revealed that the GPP threshold that separated autotrophic from heterotrophic communities in the Red Sea (1.7 mmol O2 m-3 d-1) is similar to that reported across oceanic communities elsewhere (Duarte and Agustí, 1998; Duarte and Regaudie-de-Gioux, 2009), agreeing with the oligotrophic characteristics that govern the basin at certain periods or locations. The latitudinal differences depicted in our results mirror the increasing north–south pattern in Chl a concentration and photosynthetic carbon fixation rates previously reported for the Red Sea (Acker et al., 2008; Raitsos et al., 2013; Qurban et al., 2014; Kheireddine et al., 2017), which are supported by the presence of different planktonic communities (Al-aidaroos et al., 2016; Pearman et al., 2016; Robitzch et al., 2016; Kheireddine et al., 2017; Kottuparambil and Agusti, 2018).

The lower productivity of the northern section of the Red Sea explains the dominance of heterotrophic communities therein. Still, sustaining heterotrophy in oligotrophic regions requires an allochthonous source of organic matter (Duarte et al., 2011, 2013). The arid nature of the northern Red Sea, with the watershed consisting mostly of deserts, leads to the absence of rivers and significant organic carbon inputs to the sea. Dust inputs are important, however, and whereas they have shown no effect on primary production (Torfstein and Kienast, 2018), they are a source of organic carbon (Jurado et al., 2009) that can partially supply the organic matter required to sustain heterotrophic communities. Moreover, the Red Sea supports highly productive coral reefs, mangrove forests, seagrass meadows, and algal communities in its extensive shallow coastal areas (Rasul et al., 2015; Almahasheer et al., 2016), which may export significant organic carbon to the pelagic compartment, thereby helping to sustain heterotrophic plankton communities in the northern Red Sea.

Temperature is a master variable that regulates many components of ocean dynamics, such as vertical stratification and most aspects of organismal biology, from setting boundaries in the distribution of organisms (Clarke, 1996) to controlling biochemical reactions that constrain the energy for metabolic processes (Gillooly et al., 2001). Hence, temperature is likely a significant driver of metabolic processes in the Red Sea, one of the warmest tropical marine ecosystems (Raitsos et al., 2011; Chaidez et al., 2017). Indeed, our results showed a positive response of planktonic metabolism to temperature. Moreover, the functional relationships between metabolic rates with temperature suggested that both GPP and CR were positively enhanced with increasing temperature but at a different pace.

The metabolic theory of ecology (MTE) relates the metabolic rate of an organism with its mass and temperature. This theory hypothesises that individual metabolic rates relate to temperature with a relatively constant activation energy (Ea ∼0.63 eV) for a wide range of taxa, from unicellular organisms to plants and animals (Gillooly et al., 2001; Brown et al., 2004). For aerobic respiration, Ea values vary between 0.41 and 0.74 eV at temperatures between 0 and 40 ∘C (Gillooly et al., 2005), while for photosynthetic processes the predicted Ea is lower, ∼0.32 eV (Allen et al., 2005). From a thorough compilation of data obtained for a wide range of marine systems (from polar to subtropical and tropical oceanic regions), Regaudie-de-Gioux and Duarte (2012) found that overall, the activation energies for photosynthetic production (GPP) varied between 0.29 and 0.32 eV, and for respiratory processes (CR) between 0.65 and 0.66 eV.

The Ea for GPP (1.20±0.17 eV) obtained for the Red Sea was higher than the overall value predicted by the MTE, while the Ea values for CR were below those for GPP (0.72±0.17 eV) unlike those observed elsewhere in open oceanic waters (Regaudie-de-Gioux and Duarte, 2011; García-Corral et al., 2017). Furthermore, these Ea values imply that GPP rates increased faster (5.1-fold) than CR rates (2.7-fold), in the Red Sea's thermal range (22–32.5 ∘C). These findings differ with the expected double increase in heterotrophic respiration (regarding photosynthetic processes) with temperature (Harris et al., 2006) but are closer to the results obtained by García-Corral et al. (2017), who recently reported Ea values for GPP of 0.86 , 1.48, and 1.07 eV for the Atlantic, Indian, and Pacific oceans, respectively, while the Ea values for CR found in the Atlantic, Indian, and the Pacific oceans were 0.77, 0.57, and 0.82 eV, respectively.

The apparent contradiction between our findings and the general patterns predicted by the MTE is, however, not surprising. In their model, Allen et al. (2005) predict the activation energy of photosynthesis per chloroplast (for temperatures between 0 and 30 ∘C) using the temperature-dependence parameters obtained by Bernacchi et al. (2001) for RuBisCO carboxylation rates in one species (tobacco leaves). Although the temperature range selected by Allen et al. (2005) comprises the optimum temperatures of growth rates for a wide range of functional groups of marine primary producers (Chen, 2015; Thomas et al., 2016), the temperature observed in the Red Sea exceeded this range. Due to the fast generation times of microbes (Collins, 2010), we can expect that photosynthetic planktonic communities are acclimated or even locally adapted to the thermal conditions they experience. So by favouring certain photosynthetic or thermal traits, they can enhance their metabolism and growth to the ambient temperature, up to their thermal optimum (Galmés et al., 2015; Thomas et al., 2016). Therefore, it is likely that the acclimation or local adaptation (in the long term) of photosynthetic traits in Red Sea plankton optimises the metabolic response at the high temperatures reached, resulting in a steeper response to temperature than predicted by the MTE. Moreover, as the trait responses to temperature vary among phylogenetic groups (Galmés et al., 2015, 2016; Thomas et al., 2016), we anticipated a certain degree of discrepancy if we characterise the photosynthetic response (GPP) of planktonic communities by considering only one trait (i.e. RuBisCO carboxylation) of one species.

However, we must bear in mind that the metabolic response of individuals is not only temperature dependent, and that resource supply also plays an essential role (Brown et al., 2004; Allen and Gillooly, 2009). Our results provide evidence that the increased response of planktonic metabolism towards warmer temperatures was mostly confined to the southern half of the Red Sea, which receives the direct inflow of the enriched Intermediate Water coming from the Gulf of Aden during the winter monsoon (Raitsos et al., 2015; Wafar et al., 2016). Recent findings have demonstrated that mass-specific carbon fixation rates of phytoplankton communities can be enhanced with increasing temperature when nutrients are not limiting their growth (Marañón et al., 2014, 2018). Therefore, it is likely that the intertwined effect of both the warmer temperatures and the higher nutrient availability towards the south of the Red Sea are key drivers regulating the metabolic response of planktonic communities. Thus, unlike the global ocean, where nutrient concentration is inversely correlated with temperature (e.g. Agawin et al., 2000), in the Red Sea nutrient concentration and temperature are positively correlated. This anomaly may explain the steep increase in Ea for GPP, as primary producers in the warmer region are being supported by the inflow of nutrient-enriched waters from the Indian Ocean.

The elevated Ea for GPP compared to CR in Red Sea plankton is also an anomaly, likely associated with the lack of allochthonous nutrient supply due to the absence of rivers and vegetation in the arid watershed of the Red Sea. The warm oligotrophic ocean is characterised by plankton communities that are in metabolic balance or net metabolically imbalanced (Duarte and Agusti, 2008; Duarte et al., 2013). In contrast, the warm southern Red Sea tends to support autotrophic metabolism, sustained by the input of nutrient-enriched waters while low allochthonous carbon inputs may constrain CR. As a result, NCP tends to increase, rather than decrease with increasing temperature (Regaudie-de-Gioux and Duarte, 2011; García-Corral et al., 2017). These patterns in plankton metabolism in the oligotrophic and warm Red Sea deviate from those characterising the subtropical and tropical gyres of the open ocean, but it provides an opportunity to explore the mechanistic basis for patterns in plankton metabolism with temperature, which would otherwise remain obscured by the underlying prevalent negative relationship with nutrient concentrations.