Samples were collected on board the RV Pourquoi pas? from 10 May to 11 June 2017. Water from the SML and the underlying water (ULW; 20 cm below the SML) was collected at 12 stations from 2.9 to 19.8∘E and 35.5 to 42.0∘N (Fig. 1). SML samples were collected from a zodiac using a glass plate sampler (Cunliffe and Wurl, 2014; Harvey, 1966). The dimensions of the silicate glass plate (50×26 cm) resulted in an effective sampling surface area of 2600 cm2 considering both sides. To avoid contamination during sampling, the zodiac was positioned upwind and in front of the research vessel. The glass plate was immersed and withdrawn perpendicular to the sea surface. With a Teflon wiper, SML samples were collected in acid-cleaned and rinsed bottles (Cunliffe and Wurl, 2014). Approximately 1.5 L of SML sample was collected in the course of 1 h. Sampling times are listed in Table 1. All sampling equipment was acid-cleaned (10 % HCl), rinsed with Milli-Q and copiously rinsed with seawater from the respective depth once the sampling site was reached. The ULW samples were collected simultaneously with two acid-cleaned and Milli-Q-rinsed glass bottles by immersing the closed bottles and opening them at approximately 20 cm.

The abundance and area of TEP was measured microscopically (Engel, 2009). The sample volume (10–30 mL) was chosen according to the prevailing TEP concentrations. Samples were filtered onto 0.4 µm Nuclepore membranes (Whatman) and stained with 1 mL Alcian Blue solution (0.2 gL-1 w/v) for 3 s. Filters were mounted on Cytoclear® slides and stored at -20 ∘C until analysis. Two filters per sample with 30 images each were analyzed using a Zeiss Axio Scope.A1 (Zeiss) equipped with a Zeiss AxioCam MRc. The pictures with a resolution of 1388×1040 pixels were saved using AxioVision LE64 Rel. 4.8 (Zeiss). All particles larger than 0.2 µm2 were analyzed. ImageJ was subsequently used for image analysis (Schneider et al., 2012). A filter prepared with 10 mL Milli-Q water served as a blank.

Bacterial cell numbers were determined from 2 mL samples fixed with 100 µL glutaraldehyde (GDA, 1 % final concentration). Samples were stored at -20 ∘C and stained with SYBR Green I (Molecular Probes) to determine abundance using a Becton and Dickinson (BD Biosciences) FACSCalibur flow cytometer equipped with a 488 nm laser. Bacterial cells were detected by the unique signature in a plot of side scatter (SSC) vs. green fluorescence (FL1). Yellow-green latex beads (Polysciences, 0.5 µm) were used as internal standards.

Picophytoplankton and picophytoneuston cell numbers were determined from 2 mL samples fixed and stored as for bacterial abundances. Samples were filtered through a 50 µm filter and analyzed with a flow cytometer (similar to Sect. 2.3). Enumeration of cells was conducted using a high flow rate (approximately 39–41 µLmin-1). The forward- or right-angle light scatter (FALS or RALS) as well as the phycoerythrin- and Chl a-related fluorescent signal was used to distinguish the cells. Cell counts were analyzed using the CellQuest Pro software (BD Biosciences). The method used here (fixative addition + slow freezing) follows recommendations by Lepesteur et al. (1993).

Samples (20 mL) for total hydrolyzable carbohydrates (TCHOs) >1 kDa were filled into precombusted glass vials (8 h, 500 ∘C) and stored at -20 ∘C. In the home lab, TCHO analysis was carried out using high-performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD) on a Dionex ICS 3000 ion chromatography system (Engel and Händel, 2011). Prior to analysis, samples were desalinated by membrane dialysis (1 kDa MWCO, Spectra/Por) at 1 ∘C for 5 h and hydrolyzed for 20 h at 100 ∘C in HCl (0.8 M final concentration) with subsequent neutralization using acid evaporation (N2, for 5 h at 50 ∘C). Two replicates were analyzed for each sample.

Water samples for sequencing (400 mL each) were passed through a 100 µm pore size mesh in order to remove meta-zooplankton that could dominate the 18S sequences and were subsequently filtered onto a Durapore membrane (Millipore, 47 mm, 0.2 µm) and immediately stored at -80 ∘C. In order to improve cell accessibility for the DNA extraction, filters in cryogenic tubes were immersed in liquid nitrogen, and the filter was crushed with a pestle. DNA was extracted according to a modified protocol from Zhou et al. (1996) by Wietz et al. (2015). The protocol included bead-beating, phenol–chloroform–isoamyl alcohol purification, isopropanol precipitation and ethanol washing. An additional protein-removal step by salting was used to avoid protein contamination.

Library preparation and sequencing was conducted at the Integrated Microbiome Resource at Dalhousie University, Halifax, Canada, and is described in detail elsewhere (Comeau et al., 2017). Samples were PCR-amplified in two dilutions (1:1 and 1:10) using the 18S rRNA gene primers E572F and E1009R (Comeau et al., 2011). Prior to pooling, samples were cleaned up and normalized using the Invitrogen SequalPrep 96-well plate kit (Thermo Fisher Scientific). Sequencing was conducted according to Comeau et al. (2017) on an Illumina MiSeq using 300+300 bp paired-end V3 chemistry.

Sequences were processed using the DADA2 pipeline (Callahan et al., 2016), and sequences shorter than 400 bp, longer than 444 bp, with more than eight homopolymers or any ambiguous bases were discarded. Sequences were aligned with the 18S rRNA gene sequences of the SILVA 132 alignment database (Quast et al., 2013). Subsequently, sequences that aligned outside of most of the dataset and chimeras were removed. Sequences were classified using the SILVA 132 database (Quast et al., 2013) and deposited at the European Nucleotide Archive (ENA accession number PRJEB23731). Sequences were not subsampled, and sequence numbers per sample ranged from 1063 (S8 SML) to 43 027 (S5 SML). However, for principal component analysis (PCA), all samples were subsampled down to 1063 sequences.

Statistical analyses and maps were produced using R (R Core Team, 2014) and bathymetry from NOAA (National Oceanic and Atmospheric Administration). The enrichment factor (EF) was used to compare the concentration of parameter A in the SML ([A]SML) to the concentration in the ULW ([A]ULW) and was calculated as follows (Eq. 1; World Health Organization, 1995): 1EF=[A]SML[A]ULW. An EF > 1 indicates enrichment, EF < 1 indicates depletion and EF = 1 indicates no difference between the SML and the ULW. The significance of differences between the SML and ULW and between the basins of eukaryote sequences and biogeochemical parameters were tested using the Kruskal–Wallis test and PERMANOVA. Correlations were calculated using Spearman's rank correlation.

Wind speed, surface water salinity and temperature were obtained at 5 m depth from the RV Pourquoi pas? system. Radiation measurements were obtained with a LI-COR radiation sensor (Li-200SZ) at wavelengths of 400 to 1100 nm. All parameters were measured every 5 min during sampling on the zodiac and averaged over the sampling period for statistical analyses (Table 1).

The eukaryotic communities in the SML and the ULW were similar (ANOSIM, p=0.039, R=0.1002). However, differences were detected in the eukaryotic community composition (Fig. 2) of the basins sampled (western Mediterranean, Tyrrhenian Sea and Ionian Sea). ANOSIM showed that the differences in the eukaryotic community composition were slightly larger across basins than between SML and ULW (p=0.0025, R=0.2263). However, the overall diversity and evenness (based on Shannon and Pielou indices) were not significantly different between basins (Fig. S1 in the Supplement).

A total of 16 orders were found in relative abundances over 5 % of the total eukaryotic community in one or more of all 12 stations (Fig. 3). The communities in the SML and ULW at most stations were similar, with Dinophyceae and Syndiniales (Dinoflagellata) and an unidentified Eukaryote class dominating the eukaryotic community. Zooneuston were found in most of the SML samples but rarely (n=2) in the ULW samples. Zooneuston were comprised of Ploimida (Rotifera), Maxillopoda (Cyclopoida and Calanoida) and Scyphozoa (Semaeostomeae).

Myconeuston and mycoplankton were found in high relative abundances in three ULW samples and in the corresponding SML samples (stations 7, 8 and ION_2) in the Ionian Sea. In the ULW of station 7, fungi made up more than half (54 %) of the total number of retrieved sequences. The vast majority of fungal amplicon sequence variants (ASVs) (64 out of 69) belonged to Ascomycota and Mucoromycota, with the remaining five belonging to the Chytridiomycota (n=3), Basidiomycota and Neocallimastigomycota. All fungal ASVs that were recovered throughout the cruise and their relative abundance are shown in Fig. 4. While fungal ASVs made up a significant amount of sequences in the Ionian Sea (stations to the right of Fig. 4), they were barely detectable at the other stations (p=0.014 for differences in fungal ASVs between basins tested with PERMANOVA).

Bacterial numbers did not show any significant differences between layers (Fig. 5a). In the SML, bacterial abundances ranged from 2.0×105 to 1.0×106 cellsmL-1 with an average of 5.2×105±2.3×105 cellsmL-1. In the ULW, bacterial numbers were on average 4.6×105±1.5×105 cellsmL-1 (range of 2.2×105 to 6.9×105 cellsmL-1) (Fig. 5).

Picophytoneuston (0.2–20 µm size range) abundance was on average 3.3×103±1.9×103 cellsmL-1 in the SML, and picophytoplankton abundance was on average 2.3×103±1.7×103 cellsmL-1 in the ULW (range of 1.4×103 to 8.5×103 cellsmL-1 in the SML and 9.5×102 to 7.1×103 cellsmL-1 in the ULW). Overall, cell counts determined by flow cytometry were significantly higher in the SML than in the ULW (p=0.002, n=12; Fig. 5b).

TEP concentration averaged 1.4×107±9.7×106 particlesL-1 (ranging between 3.6×106 and 3.7×107 TEPL-1) in the SML. In the ULW, the average TEP concentration was 3.6×106±2.1×106 particlesL-1 (ranging between 6.8×105 and 7.5×106 TEPL-1) in the ULW. TEP area in the SML was on average 9.7×107±1.2×108 mm2L-1 (1.5×107 to 4.5×108 mm2L-1). TEP area was lower in the ULW, with an average of 2.3×107±1.1×107 (2.9×106 to 3.9×107 mm2L-1). Both TEP abundance and area were significantly enriched in the SML (Fig. 5; p=0.01 and p=0.007, respectively). While irradiation, water temperature and salinity did not correlate with TEP abundance or area, wind speed did have a significant negative correlation with TEP abundance in the SML (R2=-0.73) and TEP area in the SML (R2=-0.75) and the enrichment factor for TEP area (R2=-0.63).

TCHO concentrations were similar between the SML and ULW (Fig. 5e), with no significant differences between depths (778±294 nM (range 562 to 1684 nM) in the SML and 605±97 nM (range 525 to 885 nM in the ULW).

The eukaryotic community composition between the SML and the ULW differed only slightly, with higher horizontal heterogeneity and significant differences between the communities of the western, Tyrrhenian and Ionian basins. The Shannon diversity did not differ significantly between depths or basins; however, a slight decrease in species richness from west to east could be observed (Fig. S1), possibly related to the transition from oligotrophic to ultra-oligotrophic conditions given the more pronounced water exchange with the Atlantic in the western basin (Reddaway and Bigg, 1996).

No diatoms were present at high relative abundances in our samples. In seasonal studies in the Mediterranean Sea, diatom contribution can be significant during blooms in March and April, but later in the year, as the water column stratifies, their contribution decreases (Marty et al., 2002). Even though diatoms most likely were not dominant in the samples, the extremely low abundance (<1 %) of diatoms in the samples might also indicate a bias of the primers used or removal of larger cells and aggregates during the pre-filtration step. Another potential bias is the dominance of dinoflagellate genera (Fig. 3). Dinoflagellates have a large number of 18S rRNA gene copies in comparison to other phytoplankton groups, and therefore their abundance in 18S rRNA gene sequencing is often overestimated (Godhe et al., 2008; Guo et al., 2016).

Previous studies suggested various factors that potentially drive the phytoplankton community composition. In addition to buoyancy of cells, high levels of UV radiation could potentially cause damage by photoinhibition. While dinoflagellates, one of the dominating phytoplankton groups, can produce photoprotective compounds, including mycosporine-like amino acids (MAAs) (Carreto et al., 1990; Häder et al., 2007), they can still be inhibited by high UV radiation (Ekelund, 1991). In the present study, inhibition by UV radiation is not indicated in the data since phytoplankton was enriched in the SML despite high radiation values (e.g., stations 4 and 7; Table 1). At the same time, TEP were significantly enriched during this sampling campaign while the phytoplankton community did not show significant differences. Previous studies suggested that TEP can protect phytoplankton and bacteria from UV radiation (Elasri and Miller, 1999; Ortega-Retuerta et al., 2009). Process oriented studies would be needed to determine whether TEP production was higher in the SML due to UV protection of phytoneuston or whether TEP formation rates were higher in the SML due to wind and wave shear at the surface (Carlson, 1993; Cunliffe et al., 2013).

Fungi were prevalent in the Ionian Sea (more than half of the sequences retrieved at one station), whereas they were scarce in the western basin and the Tyrrhenian Sea (Figs. 3 and 4). Most of the fungal ASVs present in the Ionian Sea belonged to Ascomycota and Mucoromycota. The question arises as to what drives the higher fungal relative abundances in this region of the Mediterranean Sea. While fungi, like dinoflagellates and other eukaryotic groups, can have varying amounts of 18S rDNA gene copy numbers, the patchy distribution of fungi found in this study makes a consistent bias unlikely. Marine fungi can live a saprotrophic lifestyle, degrading and recycling high-molecular-weight organic matter (Chrismas and Cunliffe, 2020; Cunliffe et al., 2017) and potentially competing with functionally similar bacteria. Some marine fungi are also phytoplankton parasites, potentially altering phytoplankton community composition through selective parasitism (Amend et al., 2019; Grossart et al., 2019). At present, we have a very limited understanding of the diversity and functional role of fungi in the SML (myconeuston). One previous study of the coastal myconeuston in the western English Channel off Plymouth (UK) showed that the SML was dominated by both Ascomycota and Basidiomycota (Taylor and Cunliffe, 2014), while our results support the dominance of Ascomycota in the SML.

So far, not many studies have looked at fungi in LNLC regions. A global comparison of fungal distribution (Hassett et al., 2020) has found that fungal diversity determined by amplicon sequencing varies between different oceanic regions, with Exophiala, belonging to Ascomycota, dominating the Ligurian Sea and an unclassified Ascomycota being the most abundant taxon in other regions.

ASVs identified as Solanales (Nicotiana) had quite high relative abundances in the easternmost stations. Since Solanales are land plants, presence of their DNA could suggest a terrestrial influence on the Ionian Sea, linked to wet or dry deposition that occurred before and/or during our sampling period in this basin. This is also corroborated by air mass trajectory backtracking using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (Fig. S3), which showed that aerosols were most likely of continental origin (Fu et al., 2021). This is also confirmed by atmospheric measurements indicating that chemical composition of dry and wet depositions were influenced by Eastern European air masses (Desboeufs et al., 2021). Station FAST_2 in the western basin was highly influenced by dust input in the area (Guieu et al., 2020; Tovar-Sánchez et al., 2020). This coincided not only with a strong increase in TEP abundance in the SML, but also with a distinct increase in the relative abundance of unidentified dinoflagellates in the SML (Fig. 3). The details of the impact of dust input on the organic matter and microbial community composition in the SML and ULW are discussed elsewhere (Anja Engel et al., unpublished data). However, no fungi were found at station FAST_2 either in the SML or in the ULW (Fig. 4), showing that dust input does not necessarily deposit fungi to the surface ocean; this potentially also holds true for the Ionian Sea. In addition to atmospheric inputs, riverine inputs can also influence the Mediterranean Sea (Martin et al., 1989). However, the Ionian Sea itself does not experience vast riverine input, and riverine influence is less pronounced in the open sea, making riverine sources of Mycophyta unlikely. Ascomycota and Mucoromycota have been recovered from a variety of marine environments (Bovio, 2019; Grossart et al., 2019; Hassett et al., 2019), implying that they also might be thriving in the SML of the Mediterranean Sea rather than originating solely from terrestrial sources.

Overall, the most abundant fungal ASV in the Ionian Sea (ASV 8) was identified as belonging to genus Cladosporium, which has been found in marine environments before (Cunliffe et al., 2017). Another explanation for the high relative abundance of fungi in the Ionian Sea might be that they are more adapted to dealing with the low-nutrient conditions found in the more eastern basin of the Mediterranean Sea.

Bacterial and microalgal numbers determined by flow cytometry decreased significantly from west to east, with bacteria showing the largest decline (Tovar-Sánchez et al., 2020). While total microalgal abundances determined by flow cytometry were low in the SML and in the ULW, they were comparable to other studies on phytoplankton abundance in the SML of the Mediterranean Sea (Joux et al., 2006). The microalgal numbers between 5 and 200 m water depth (data not shown) were higher than at the air–sea interface. Even though overall bacterial numbers decrease, further molecular analyses would be needed to determine if the bacterial community is changing from west to east and if certain bacterial taxa can benefit from the ultra-oligotrophic conditions. In contrast to bacteria and phytoplankton, spatial trends in TCHO and TEP, as well as DOC in the SML and dissolved organic carbon (DOC) and particulate organic carbon (POC) in the ULW, did not show significant differences between the Ionian Sea and the other basins (Freney et al., 2020; Trueblood et al., 2020). TEPs are often enriched in the SML (Engel and Galgani, 2016; Jennings et al., 2017; Wurl et al., 2009; Wurl and Holmes, 2008), in particular over oligotrophic regions (Jennings et al., 2017; Zäncker et al., 2017). This is in good accordance with trends observed in the present study (Durrieu de Madron et al., 2011; Fogg, 1995; Wikner and Hagstrom, 1988). Wind speed correlated negatively with TEP abundance and area in the SML, showing that wind can negatively affect TEP concentrations at the air–sea interface as has been previously suggested (Sun et al., 2018).

Since exchange of water with the Atlantic is most pronounced in the western basin and an anti-estuarine circulation prevails in the Mediterranean Sea, nutrient limitation increases eastwards. TEP production has been shown to be independent of stoichiometric ratios in the surrounding water in a previous study (Corzo et al., 2000). Since light limitation rarely occurs in the SML and TEP can potentially protect phytoplankton from light damage (Elasri and Miller, 1999; Ortega-Retuerta et al., 2009), phytoplankton might still photosynthesize and excrete carbohydrates that assemble to TEP. This would not only explain the lack of difference in TEP abundances between basins, but also TCHO concentrations. TCHO can also be produced by cell lysis (due to nutrient depletion).

TCHO and TEP could also provide available substrate and microhabitats for marine fungi with reduced competition by bacteria in the Ionian Sea. Malassezia and Cladosporium have been shown to assimilate carbon derived from TEP-associated algal polysaccharides in the English Channel (Cunliffe et al., 2017), highlighting that Cladosporium and other fungi might be able to profit from the decreased bacterial competition in the Ionian Sea. Further, previous studies have shown that the eastern Mediterranean Sea contains higher concentrations of organic pollutants (Berrojalbiz et al., 2011a, b), while a Cladosporium strain has the capacity to degrade polycyclic aromatic hydrocarbons (Birolli et al., 2018), highlighting another potential substrate for the fungi detected in the Ionian Sea.