Seawater was collected during six seasonal cruises in the Levantine Basin, southeast Mediterranean Sea, on-board the R/V Bat-Galim between 2021–2023. Three cruises were held during the stratified period and three during the winter mixing. The mixed layer depth was calculated using a temperature difference of Δ0.3 °C (Mena et al., 2019). Two “deep” stations were sampled in each cruise; one located at the edge of the continental shelf (H05 33.00° Lat, 34.50° Lon, bottom depth ∼ 1500 m, 50 km from the coast) and the other at the edge of Israel's exclusive economic zone (H06 33.15° Lat, 34.16° Lon, bottom depth 1750 m, 90 km from shoreline). Seawater was sampled at discrete depths throughout the water column, from the surface (∼ 0.5 m) to the bottom (1500–1750 m) using Niskin bottles. Sampling depths were chosen in real-time based on measurements of conductivity, temperature, depth (CTD) (Seabird 19 Plus), chlorophyll fluorescence (Turner designs, Cyclops-7) and PAR (Sea Bird). The raw hydrological data can be freely downloaded from https://isramar.ocean.org.il/isramar2009/ (last access: 25 February 2025). Measurements included DIC (NaH¹⁴CO₃) uptake under ambient light (hereafter light primary productivity, LPP) or under full dark conditions (DCF), bacterial productivity (BP) and nutrient quantification.

Samples for nitrite (NO2-) and ammonium (NH4+) concentrations were collected only in the 2023 cruises. The samples were pre-filtered (0.45 µm), placed in acid-washed plastic vials, and were kept frozen at -20 °C until analysis. Nutrients were measured with a Seal Analytical AA-3 system. The limits of detection for NO2- and NH4+ were 0.06 and 0.09 µM, respectively.

Seawater was collected in triplicates into transparent (for LPP measurements) or dark (for DCF) Nalgene bottles (45–250 mL) and spiked with NaH¹⁴CO₃ (Perkin Elmer, specific activity 56 mCi mmol⁻¹) following Nielsen (1952). The bottles were maintained in on-deck incubators covered with a gradient of neutral mesh simulating the irradiance intensity (no change in spectrum) at 100 %, 50 %, 10 %, 1 %, and 0.1 % of surface light intensities or under complete dark conditions (Belkin et al., 2022; Reich et al., 2024). Incubators were kept at constant ambient surface temperatures (∼ 19–20 °C in winter and ∼ 28–29 °C in the summer cruises). We acknowledge that temperature differences between surface and deeper depths may alter the LPP or DCF rates measurements, especially during the summer when the water column is stratified. While in situ measurements may offer more precise rate estimates, they are generally impractical during research cruises that involve sampling at multiple locations and times throughout the day and night. Nevertheless, preliminary comparisons between the incubation setup used here versus in situ incubations using a mooring line showed negligible differences in primary productivity, falling within the expected range of measurement variability (see also Reich et al., 2022). All the incubation bottles were spiked at sunrise and terminated after 24 h (Reich et al., 2022; Robinson et al., 2009) by filtering. the particulate matter onto GF/F filters using low vacuum pressure (<50 mmHg). Next, the excess ¹⁴C-bicarbonate was removed by fuming with 50 µL of 37 % hydrochloric acid overnight. Finally, 5 mL scintillation cocktail (ULTIMA-GOLD) was added, and the disintegrations per minute (DPM) from the particulate matter concentrated on the filters were counted using a TRI-CARB 4810 TR (Packard) liquid scintillation counter. Blank seawater spiked with NaH¹⁴CO₃ was filtered immediately without incubation and the reads were subtracted from the sample's DPM. The blank DPM reads were usually negligible (<5 % of the sample's DPM). Aliquots (50 µL) from random spiked samples were placed onto new GFF filters, added with 50 µL ethanolamine and scintillation liquid, and counted immediately without incubation to account for the “added activity” of the radiolabeled working solution used. LPP was calculated as the difference between the DPM retrieved from the samples incubated under ambient light (total primary production) and the “dark” bottles. Dark or light dissolved inorganic carbon fixation was calculated based on the Bermuda Atlantic Time-series Study (BATS) protocol (https://bios.asu.edu/bats/bats-data, last access: 25 February 2025). More details can be found in Reich et al. (2024).

Triplicate samples per depth (1.7 mL) were incubated in the dark with 10 nmol L^{−1 3}H-leucine L⁻¹ (Perkin Elmer, specific activity 123 Ci mmol⁻¹) for 4–6 h under ambient temperature (Simon et al., 1990). The incubations were terminated with 100 µL of trichloroacetic acid (100 %), processed as described by Smith and Azam (1992), and counted using a TRI-CARB 4810 TR (Packard) liquid scintillation counter. Killed control samples containing ³H-leucine L⁻¹ and trichloroacetic acid (without incubation) were also measured and these control sample's DPMs were subtracted from the sample's reads. A conversion factor of 3 kg C mol⁻¹ mol⁻¹ leucine incorporated was used, assuming an isotopic dilution of 2.0 (Simon et al., 1990).

DNA was extracted from water samples with the PowerWater kit (Qiagen, USA), using the FastPrep-24™ Classic (MP Biomedicals, USA) bead-beating to disrupt the cells (2 cycles at 5.5 m s⁻¹, with a 5 min interval). The V4 region (∼ 300 bp) of the 16S rRNA gene was amplified from the DNA (∼ 50 ng) using the 515Fc/806Rc primers amended with relevant tags (Apprill et al., 2015; Parada et al., 2016). PCR conditions were as follows: initial denaturation at 94 °C for 45 s, 30 cycles of denaturation (94 °C for 15 s), annealing (15 cycles at 50 °C and 15 cycles at 60 °C for 20 s) and extension (72 °C for 30 s). Two annealing temperatures were used to account for the melting temperature of both forward (58.5–65.5 °C), and reverse (46.9–54.5 °C), primers.

Demultiplexed paired-end reads were processed in QIIME2 V2022.2 environment (Bolyen et al., 2019). Reads were truncated based on quality plots, checked for chimeras, merged and grouped into amplicon sequence variants (ASVs) with DADA2 (Callahan et al., 2016), as implemented in QIIME2. The amplicons were classified with Scikit-Learn classifier that was trained on Silva database v138 (16S rRNA; Glöckner et al., 2017). Mitochondrial and chloroplast sequences were removed from the 16S rRNA amplicon dataset. Downstream analyses were performed in R v4.1.1 (R Core Team, 2020), using packages Phyloseq (McMurdie and Holmes, 2013) and Ampvis2 (Andersen et al., 2018). Indicator species analyses were performed using Indic species package v1.7.9 (De Cáceres et al., 2009). Amplicon reads were deposited to the NCBI SRA archive under project number PRJNA1215023.

BR was calculated based on the following equation and assuming an average open-ocean bacterial growth efficiency (BGE) of 20 % (Herndl and Reinthaler, 2013) similar to previous direct measurements from the Mediterranean Sea ranging from 0.21–0.29 (Zweifel et al., 1993). 1BGE=BPBP+BR BCD was then calculated as the sum of BP and BR (Gasol et al., 1998). Zooplankton respiration (ZR) and excretion were compiled from Belkin et al. (2022).

As expected, LPP was restricted to the photic layer with highest rates usually measured at the surface (∼ 0.5 m) that gradually decreased to reach minimum rates at the bottom of the photic layer (∼ 180 m) (Fig. 1a). Relatively low LPP values were measured during the stratified summer (∼ 0.1–0.8 µg C L⁻¹ d⁻¹), whereas higher rates were measured during the winter mixing period (∼ 0.2–7.4 µg C L⁻¹ d⁻¹) (Fig. 1a). This resulted in ∼ 10-fold higher integrated rates measured during the mixed period compared to those measured during the stratified period (Table 1), in accordance with studies from the area (Psarra et al., 2005; Reich et al., 2022; Sisma-Ventura et al., 2022b). In contrast with LPP, DCF was not restricted to the photic layer and ranged from 0 to ∼ 0.4 µg C L⁻¹ d⁻¹ throughout the water column (Figs. 1b, 2a), without significant differences in the absolute rates between the photic and aphotic zones (t-test, p>0.05, Fig. 2b). The observed decrease in DCF rates with depth (Fig. 2a) during the summer cruises may be partly attributed to a decline in the abundance of chemoautotrophs with depth. For example, Agogué et al,. (2008) reported a decline in archaeal amoA gene copy numbers with depth in the eastern North Atlantic. Normalizing DCF rates to chemoautotrophic microbial cell abundance (or gene copy) could reveal a different vertical pattern. Another possible explanation for the decline in DCF rates with depth may be related to the weakening flux of sinking organic matter with depth that limits the substrates that fuel DCF (discussion below). The integrated photic DCF was typically lower than the rates reported in the central and western Mediterranean Sea (La Cono et al., 2018). The aphotic DCF rates were ∼ 3.5 fold higher during the mixed than during the stratified period (Table 1, Fig. 2a).

The higher aphotic DCF in the mixed versus the stratified periods may be related to more bioavailable carbon that is transported from the photic layer as marine snow and supplies organic carbon to heterotrophic activity in the winter (coinciding higher LPP). However, given the oligotrophic nature of the southeast Mediterranean Sea (Berman-Frank and Rahav, 2012; Reich et al., 2022), most of the organic carbon (both particulate and dissolved originating from LPP) is recycled within the photic layer. Only a small fraction fluxes down to the aphotic zone and has been recorded in sediment traps (Alkalay et al., 2024).

Another possible mechanism that may, potentially, explain higher DCF rates during the winter versus the summer is anaplerosis. The extent of anaplerotic reactions is primarily driven by the availability of labile organic carbon to heterotrophs (Dijkhuizen and Harder, 1984). Therefore, assuming anaplerosis drives DCF, we expect it will be positively coupled to BP.

Yet, our results do not support the likelihood of significant anaplerosis reactions. This is predominantly evident from the spatiotemporal distribution of aphotic BP (Fig. 1c) differing considerably from that of the DCF (Fig. 1b) and does not correlate with it (Fig. 3a). In fact, BP seems to be coupled with LPP at the photic layer reaching ∼ 0.4 µg C L⁻¹ d⁻¹ (not shown). Excluding some sporadic measurements, aphotic BP rates were usually of similar magnitude and typically <0.1 µg C L⁻¹ d⁻¹ (Fig. 1c). Moreover, the highest integrated aphotic BP was measured during the summers of 2021 and 2022 and not during the winter cruises when generally higher DCF was recorded (Table 1, Fig. 2a).

Despite the lack of a clear positive relationship between DCF and BP, DCF may contribute to bacterial carbon demand (BCD) in the aphotic zone. Thus, we use a literature standard, a bacterial growth efficiency of 0.20 (Gasol et al., 1998) to calculate BR and BCD (see the “material and methods” section for more details). This calculation yielded bacterial respiration (BR) ranging from 29–494 mg C m⁻² d⁻¹ (average 209 ± 172 mg C m⁻² d⁻¹), and the concurrent BCD ranges from 36 to 648 mg C m⁻² d⁻¹ (average 262 ± 121 mg C m⁻² d⁻¹). Under these circumstances, exudation of DOC from primary productivity at the photic layer estimated as 20 % of the rates (Teira et al., 2003) equals to ∼ 1–130 mg C m⁻² d⁻¹.

This new DOC, that originated from the photic zone therefore cannot support the aphotic BCD in our system in all of our observations. However, if we consider the contribution of DOC produced by aphotic DCF, part of the missing carbon may be accounted for. Thus, when considering aphotic DCF in addition to the sequestered DOC from the photic layer, the “abnormally high” aphotic BCD could be explained in full (≥100 %) in ∼ 35 % of the observations (Fig. 1d). In the other 65 % of the observations the missing carbon sources needed to support the aphotic BCD remain an enigma. We note that these calculations are based on global averages and assumptions and therefore may be subject to some uncertainties. For example, BGE can vary between seasons and sites (del Giorgio and Cole, 1998). In the Mediterranean Sea, long-term measurements of BGE ranged from 0.21 (similar to our calculations and the global average used by Herndl and Reinthaler, 2013) to 0.29 (Zweifel et al., 1993). If the 0.29 value is used, the contribution of DCF to the aphotic BCD increases to ∼ 45 % of the observations rather than ∼ 35 % when using BGE of 0.2. Similarly, if we apply an exported DOC estimate of ∼ 4 % from the photic zone, as reported for the Ionian Sea/western Mediterranean (Moutin and Raimbault, 2002), the relative contribution of DCF to aphotic BCD would be even higher than in our current calculations, which assume ∼ 20 % DOC export (Teira et al., 2003). These uncertainties warrant future investigation. Yet, even when using conservative estimates for BGE and DOC export as done here, the contribution of DCF to aphotic BCD remains substantial.

Evidence suggests that dissolved methane may be more abundant in oxygenated environments than previously thought (Grossart et al., 2011). Methane can potentially be one of the missing energy sources for marine microbes and support high BCD (Brankovits et al., 2017) as observed at the aphotic southeast Mediterranean Sea (Fig. 1d). In agreement, methanotrophs were found in aphotic cold seeps at the southeast Mediterranean Sea (Sisma-Ventura et al., 2022a), as well as across the aphotic water column in our samples (see discussion below).

Interannual variability in DCF, but not in LPP or BP, was observed with higher rates recorded in March 2021–March 2022 and lower rates observed in August 2022–August 2023 (Fig. 1b). Inorganic nutrients such as PO43+ or NO3-+NO2- are unlikely to explain this variability as their ambient levels were similar between periods (https://isramar.ocean.org.il/isramar2009/, last access: 25 February 2025). Alternatively, we surmise that differences in the bioavailability and concentration of sinking organic particles, possibly attributed by the BiOS (Bimodal Oscillating System) oscillation circulation of deep water between the Adriatic and Ionian seas, could potentially explain the higher aphotic chemoautotrophic activity in March 2021–March 2022 versus August 2022–August 2023. This mechanism is known to influence the bioavailability of organic nutrients in the deep Mediterranean Sea by modulating deep-water circulation and ventilation patterns (Civitarese et al., 2010). These shifts affect the transport and residence time of organic matter (Civitarese et al., 2023), thereby potentially altering availability of organic nutrients to aphotic microbial populations, including to chemoautotrophs. Supporting this hypothesis are recent studies from the northern Red Sea and South China Sea showing that DCF is limited by labile organic nutrients such as phosphonates and even carbon-rich molecules (Reich et al., 2024; Zhou et al., 2017). Aphotic free-living chemoautotrophs are likely to encounter an increasingly refractory pool of dissolved organic matter for metabolism that may result in lower DCF rates, as shown in exported material through the water column (Santinelli et al., 2013). Particle-attached chemoautotrophs may have access to higher concentrations of organic substrates. Therefore we surmise these microbes would preferentially have a particle-attached lifestyle in the deep ocean. The patchy nature of particulate matter sinking and lateral transport during wintertime (Alkalay et al., 2024) and aggregate concentrations (Bar-Zeev et al., 2012) in the deep southeast Mediterranean Sea could also potentially explain the interannual variability in DCF between periods. Understanding how chemoautotrophs transform labile dissolved organic matter into refractory dissolved organic matter, which is an essential process in the “microbial carbon pump”, is crucial as it influences the efficiency of the biological pump (Jiao et al., 2010).

Oxidizing reduced inorganic compounds as electron donors (e.g., NO2- or NH4+) may provide chemoautotrophic prokaryotes sufficient energy to fix DIC (Hügler and Sievert, 2011). We therefore measured the vertical distribution of NO2- and NH4+ (only in the March and August 2023 cruises) and examined if these chemical species are coupled or uncoupled with DCF at the aphotic zone. Our results show a negative-linear relationship between DCF and NO2- (Fig. 3b) and NH4+ (Fig. 3c), suggesting nitrification. This is because chemoautotrophs consume NO2- and NH4+ during nitrification to yield energy to fix DIC in the aphotic zone (REF), thereby reducing nutrient standing stocks in the water. We surmise that this “depletion” may, theoretically, explain the observed negative correlation between nutrient levels and chemoautotrophic activity. In agreement, both ammonia and nitrite oxidizers were found in the aphotic zone of all cruises (DNA level, discussion below), further highlighting their potential role as contributors to DCF in the southeast Mediterranean Sea. Nitrification measurements along with metagenomic tools, DCF (and BP) in aphotic water should be included in future dedicated studies to better refute or reinforce that oxidation of NO2- or NH4+ may provide chemoautotrophic prokaryotes the energy to fix DIC.

Analyses of 16S rRNA gene amplicons suggest that diverse bacteria and archaea may drive DCF in the aphotic southeast Mediterranean Sea (Fig. 4a). Microbes found in our collected genetic material primarily include the order Nitrosopumilales ammonia-oxidizing archaea, which become dominant below DCM (up to ∼ 30 % read abundance near the bottom), corresponding to previous estimates based on in-situ fluorescent hybridization (De Corte et al., 2009). Nitrite-oxidizing Nitrospirales comprised ∼ 1 % read abundance at depths below 300 m. Among these lineages, analyses of indicator species identified seasonal variation in abundance of the orders Nitrosopumilales and Nitrososphaeria that were more prominent in the stratified period than in wintertime (p-value < 0.05) when the water column is mixed, while the deep-sea community in general exhibited only mild seasonal changes (Fig. 4b). These ammonia oxidizers may thus drive ammonia depletion during summertime at the southeast Mediterranean Sea.

Additionally, we identified consistent occurrence of UBA10353 (Arenicellales, including the UBA868 group) and SAR324 gammaproteobacterial groups at depths below the deep chlorophyll maxima (Fig. 4a). Members from these ubiquitous clades are mixotrophs that can fix inorganic carbon conserving energy from sulfur oxidation (Baltar et al., 2023; Jaffe et al., 2024; Swan et al., 2011). We did not find the SUP05/Arctic96BD-19 gammaproteobacterial sulfur oxidizers that occur in productive dark oxygenated waters (Swan et al., 2011). We note that Methylococcales methane oxidizers were typical in the near-bottom water layer, as found previously in the southeast Mediterranean Sea basin (Sisma-Ventura et al., 2022a; Techtmann et al., 2015), suggesting methylotrophy as a potential mechanism of 1-carbon molecule acquisition in the dark southeast Mediterranean Sea. While this DNA-based community analysis provides insight into the potential contributors to DCF, it does not reflect a direct link as we cannot determine which of the identified microbes are responsible for the measured DCF rates. We stress that future studies should examine the link between DCF and specific microbial groups such as archaea by targeting RNA-level expression and functional genes (e.g., amoA and amoB), as demonstrated by Agogué et al. (2008).