Interaction between phytoplankton and heterotrophic bacteria in Arctic fjords during the glacial melting season as revealed by eDNA metabarcoding

Abstract The hydrographic variability in the fjords of Svalbard significantly influences water mass properties, causing distinct patterns of microbial diversity and community composition between surface and subsurface layers. However, surveys on the phytoplankton-associated bacterial communities, pivotal to ecosystem functioning in Arctic fjords, are limited. This study investigated the interactions between phytoplankton and heterotrophic bacterial communities in Svalbard fjord waters through comprehensive eDNA metabarcoding with 16S and 18S rRNA genes. The 16S rRNA sequencing results revealed a homogenous community composition including a few dominant heterotrophic bacteria across fjord waters, whereas 18S rRNA results suggested a spatially diverse eukaryotic plankton distribution. The relative abundances of heterotrophic bacteria showed a depth-wise distribution. By contrast, the dominant phytoplankton populations exhibited variable distributions in surface waters. In the network model, the linkage of phytoplankton (Prasinophytae and Dinophyceae) to heterotrophic bacteria, particularly Actinobacteria, suggested the direct or indirect influence of bacterial contributions on the fate of phytoplankton-derived organic matter. Our prediction of the metabolic pathways for bacterial activity related to phytoplankton-derived organic matter suggested competitive advantages and symbiotic relationships between phytoplankton and heterotrophic bacteria. Our findings provide valuable insights into the response of phytoplankton-bacterial interactions to environmental changes in Arctic fjords.


Introduction
The Arctic glacial fjord in Svalbard is a unique and sensitive ecosystem experiencing r a pid envir onmental c hanges owing to global warming (Svendsen et al. 2002 ).The distinctive characteristics of Arctic glacial fjords are expected to significantly impact the diversity of the Arctic marine ecosystem, encompassing various marine organisms from microscopic (eukaryotic phyto-and zooplankton and heter otr ophic bacteria) to macr oscopic (fish and benthic inv ertebr ates) species.These or ganisms and with their inter actions, ar e crucial in controlling the productivity and biodiversity of the Arctic fjord ecosystems (Hop et al. 2002 ).P articularl y, the microbial food web in fjord ecosystems is primarily influenced by phytoplankton that supports primary production, and bacteria that tr ansform phytoplankton-deriv ed or ganic matter into inorganic nutrients (Rokkan Iversen and Seuthe 2011 , Buchan et al. 2014 ).
The interactions between phytoplankton and bacteria, which ar e closel y associated in the microbial food web, have been postulated for decades (Seymour et al. 2017 ).Phytoplankton, such as diatoms (eukaryotes), dinoflagellates (eukaryotes) and c y anobacteria (prokary otes), are responsible for almost 50% of global photosynthesis in aquatic environments (Falkowski 1994, Field et al. 1998 ).Furthermor e, these phototr ophic organisms are consequently linked to biogeochemical cycles mediated by heter otr ophic bacterial metabolism (Seymour et al. 2017 ).Emer ging e vidence indicates that the inter action of phytoplankton and bacteria at the base of the food web controls carbon and nutrient cycling in oceans (Seymour et al. 2017 ).Ho w e v er, its ecological interface in Arctic fjords remains uncertain.
Phytoplankton communities in the Arctic Svalbard have been extensiv el y surv eyed under micr oscopic observ ation (Piwosz et al. 2009, Hodal et al. 2012, Marquardt et al. 2016, Caroppo et al. 2017, Hegseth et al. 2019, Payne and Roesler 2019, Bae et al. 2022, Kohlbach et al. 2023 ).Recent observations suggested that summer fjor d w aters may not be conducive to the survival of eukaryotic phytoplankton, such as diatoms or dinoflagellates, because of nutrient limitations and salinity changes resulting from glacial melting (Bae et al. 2022 ).Unlike other oceans, underwater light conditions are determined by the thickness of sea ice and its snow cover in the Arctic Ocean; these conditions control the formation of early phytoplankton assemblages.Additionally, the post-phytoplankton bloom in summer Arctic fjords is limited because of the reduced light transparency resulting from increased sediment-laden waters after glacial melt disc har ge (Wiktor 1999 ).Similarly, bacterial surveys in Arctic fjords have revealed that Arctic hydr ogr a phy affects the bacterial community composition in the Svalbard fjords (Zeng et al. 2009, Piquet et al. 2016, Jain and Krishnan 2017, Han et al. 2021, Han et al. 2022b ); this implies a role for bacterial degradation of organic matter produced and released by phytoplankton (Han et al. 2021(Han et al. , 2022 ) ). Ho w e v er, phytoplankton-associated bacterial communities have been studied less fr equentl y in Ar ctic fjor ds than in other en vironments .This information can explain how inorganic nutrients ar e r esupplied to primary producers through organic matter degradation in Ar ctic fjor ds.
A phytoplankton cell conserves its immediate microenvironment (phycosphere).It is a unique and dynamic niche where the cell continuously releases organic molecules into the surrounding water, including the bacterial habitat (Seymour et al. 2017 ).Phytoplankton-deriv ed or ganic matter, including pol yphenols , alkaloids , terpenes , pol ysacc harides , fatty acids , sterols , lactones , proteins and peptides , significantly contributes to the resupply of nutrients to primary producers through bacterial degradation (Bhowmick et al. 2020 ).T hus , understanding the fate of phytoplankton-deriv ed or ganic matter in Ar ctic glacial fjor ds is essential to gain insights into the ecological and biogeochemical processes in this unique region and to predict how bacterial responses to future environmental changes in the Arctic ecosystem will impact these processes.
Although the investigation of the phytoplankton-bacterial interface has been challenging because of its complex and dynamic natur e, r ecent adv ances in molecular tec hniques hav e pr ovided new tools to investigate the phytoplankton and bacterial communities, including the use of environmental DN A (eDN A) metabarcoding.eDN A metabar coding has emerged as a po w erful tool for biodiversity assessment in various environments (Yoccoz 2012, Deiner et al. 2017 ).This technique involves amplifying and sequencing a specific DNA marker from environmental samples to identify the organisms in a community.Compared with other conventional methods, eDN A metabar coding is a more efficient and cost-effectiv e a ppr oac h for assessing the composition and div ersity of biological communities (Pompanon et al. 2012, De Barba et al. 2014 ).Recent studies have emplo y ed eDN A metabar coding to investigate the bacterial and phytoplankton communities in marine environments using 16S (Agogué et al. 2011, Ghiglione et al. 2012, Han et al. 2014, 2015, Techtmann et al. 2015, Hernando-Morales et al. 2017 ) or 18S (Logares et al. 2012, De Vargas et al. 2015, De Luca et al. 2021, Šupraha et al. 2022, Cui et al. 2023 ) rRNA genes, r espectiv el y.Mor eov er, compr ehensiv e 16S and 18S rRNA comm unity anal yses hav e v alidated the inter actions between phytoplankton and bacterial communities on the southern coast of the Korean Peninsula (Han et al. 2022a ) and in the coastal ocean of southern California (Needham et al. 2018 ).Howe v er, no study has investigated the phytoplankton-bacteria interface using the eDN A metabar coding a ppr oac h in Arctic fjords.To bridge this gap, in this study, we used eDNA metabarcoding to investigate the bacterial communities associated with phytoplankton in the Svalbard Arctic fjord.A diverse and distinct community was r e v ealed in r esponse to dynamic environmental conditions during the summer melting season in four different Arctic fjords (van Mijenfjorden, van Keulenfjorden, Hornsund and Storfjorden).

Description of sample prepar a tion and datasets
The survey of Arctic Svalbard fjords (v an Mijenfjorden, v an K eulenfjorden, Hornsund and Storfjorden) was conducted using R/V Helmer Hanssen in August 2019 (Fig. 1 A).Seawater samples ( n = 21, each 2 L of seawater) and hydr ogr a phic data (fluorescence, oxygen, salinity and temper atur e) wer e collected fr om the fjords at discrete depths using a shipboard conductivity-temper atur edepth rosette system (Seabird SBE 911plus; Sea-Bird Electronics, Belle vue, USA).Sampled depths ar e pr ovided in the supplementary Excel data and indicated in the sample IDs.Each 1 L of sampled seawater was immediately filtered in the field using a 0.2-μm hydr ophilic PVDF membr ane (Merc k, Darmstadt, Germany) and the filtered membranes were then stored in a deep freezer ( −80 • C) until eDN A extraction. eDN A w as extr acted fr om two membr ane filters at each sampled depth using a DNeasy Po w erWater Kit (QIAGEN, Hilden, Germany) following the manufacturer's instructions and its quantity was estimated using a Qubit 4.0 Fluorometer (In vitrogen, C A, USA).Next, 50 ml of seawater were additionall y filter ed (0.45-μm HDPE syringe filter) and frozen ( −20 .The e volv ed dimethyl sulfide was then quantified using gas c hr omatogr a phy with an attac hed flame photometric detector (Agilent 6890 N; Agilent Technologies, Wilmington, USA) (Park et al. 2014 ).This study compiled the datasets into 21 samples of 16S and 18S rRNA gene sequences and their metadata.18S rRNA gene sequences were constructed from this study, and 16S rRNA sequences were selectively retrieved from the National Center for Biotechnology Information (NCBI) under accession number PR-JNA669167.The metadata for these sequences constituted the hydr ogr a phic measur ements, nutrients and dissolv ed or ganic carbon (DOC), r etrie v ed fr om the pr e vious study (Han et al. 2022b ).

eDNA metabarcoding and ribosomal gene quantification
A metabarcoding-based a ppr oac h was a pplied to sequence the rRNA hyperv ariable r egion of eDNAs using an Illumina MiSeq platform (Macr ogen, Seoul, South Kor ea).The eDN A metabar coding with two-step PCRs amplified the 16S and 18S rRNA genes to analyze bacterial and eukaryotic communities.Information on used primer sequences and detailed conditions of the two-step PCRs were previously described (Han et al. 2022a ).Briefly, the first amplicon PCRs were carried out in triplicate using a KAPA HiFi Hotstart ReadyMix PCR kit (KAPA BioSystems, MA, USA) with sequencing primers for 16S (V3-V4) and 18S (V8-V9) rRNA genes.The PCR amplicons were purified using a Qiaquick PCR purification kit (Qiagen, CA, USA), and then the second index PCRs proceeded using the PCR amplicons according to Illumina's instructions.Concentrations of the index PCRs were measured by the Qubit 4.0 Fluorometer (In vitrogen, C A, USA) after amplicon purification.The purified amplicons were all mixed in equimolar amounts to construct the Illumina MiSeq library and subjected to sequencing.The obtained metabarcoding sequences were submitted to the Sequence Read Arc hiv e of the NCBI under the accession number PRJNA669202 (eukaryotic 18S rRNA gene).

Sequence data processing and analysis
Sequencing data analysis was performed using the Mothur softwar e pac ka ge (v.1.40.5)(Sc hloss et al. 2009 ) based on the MiSeq standard operating procedure (SOP) (Kozich et al. 2013 ).Briefly, 5 699 514 sequences for the 16S rRNA and 2 523 682 sequences for the 18S rRN A w er e obtained.The pr ocess of sequence quality filtration commenced by correcting amplification and sequencing errors, follo w ed b y singleton r emov al and r andom subsam-pling of sequences using the given commands in Mothur SOP.Before the random sampling, non-bacterial sequences like chloroplasts, mitochondria and unknown wer e r emov ed using the "remo ve .lineage"command after taxonomic classification using the "classify.seqs"command.The sequence count number for each sample was normalized with 10 000 sequences of both 16S and 18S rRNA genes in the random subsampling.These filtered sequences wer e cluster ed into the amplicon sequence v ariants (ASVs) to inv estigate micr obial alpha and beta div ersity and comm unity composition.In the Mothur Miseq SOP, abundance-unweighted species richness indices (Chao1 and ACE) were calculated using ASVs for the alpha diversity, and a phylip-formatted distance matrix using thetayc dissimilarity was generated for the beta diversity.Non-metric multidimensional scaling was applied to visualize beta diversity, and its statistical separation was supported by analysis of molecular variance .T he composition of the micr obial comm unity was determined a gainst the Silv a.seed_v132 database.
Statistical analysis was performed using R software (v.4.2.0) ( https://www.R-pr oject.org ) with v arious R pac ka ges.Principal component analysis (PCA) and simple linear r egr ession anal ysis were carried out using the function "prcomp" and "stats" of the pac ka ge "stats" (R cor e team 2017 ).Multi-r esponse perm utation pr ocedur e (MRPP) and indicator species analysis were carried out using the "mrpp" function of the vegan package (Oksanen and Blanchet 2017 ) and the "indval" function of the labdsv package (Roberts and Roberts 2016 ), r espectiv el y.Micr obial network inter actions with envir onmental factors wer e estimated using the RCy3 pac ka ge (Gustavsen et al. 2019 ) connected to the Cytosca pe software (v.3.10.0)( https://cytosca pe.or g ).The network anal ysis proceeded with sequences assigned to ASVs or class-level taxonom y.Ad ditional data curation was performed to minimize the potential inaccuracies caused by a significant number of zero values in the ASVs dataset.In the initial stage, 17 861 bacterial ASVs and 4376 eukaryotic ASVs, all with a sequence count exceeding 10, wer e filter ed fr om the dataset.Subsequentl y, major ASVs wer e selected based on each ASV representing more than 0.1% of the total dataset.Using the curated ASV dataset, the network model was similarly constructed following the pr ocedur es used for the class-le v el taxonomy dataset.Spearman's correlation matrix was calculated by the r elativ e abundance of bacterial and eukaryotic populations (class le v el) and v alues of envir onmental factors.Cooccurrence patterns between microbial populations or between population and environmental factors were determined by significant coefficient values more than 0.5 ( P < 0.05) in Spearman's correlation and were visualized in network interactions .T he functional potential of bacterial communities was estimated using the software PICRUSt2 (Douglas et al. 2020 ).Predictive functional abundance tables were classified using the Kyoto Encyclopedia of Genes and Genomes (KEGG).The abundance of KEGG data was display ed b y a heatmap visualization using the "HeatmapAnnotation" function of the ComplexHeatma p pac ka ge (Gu et al. 2016 ).

Water mass properties of Svalbard fjords
The water samples collected from distinct depths in the Svalbard fjords (Fig. 1 A and Table 1 : van Mijenfjorden (n = 7), van Keulenfjorden (n = 4), K1-K2, Hornsund (n = 3) and Storfjorden (n = 7)) w ere explained b y their w ater mass properties in the PCA patterns (Fig. 1 B).The proportion of variance in the PCA ordination was r epr esented b y tw o axes: PC1 (47.66%) and PC2 (22.60%).Among the water mass pr operties, temper atur e and salinity primarily defined the PC1 axis, whereas silicates mostly explained the PC2 axis.Water mass properties on the PCA r e v ealed that surface ( ≤10-m depth) w aters w er e separ ated fr om subsurface ( > 10 m) waters on the PC1 axis, and the significance of their separation w as supported b y MRPP ( P < 0.05).The separation between the surface and subsurface w aters w as visualized using temper atur e and salinity dia gr ams ( Fig. S1 ).Mor eov er, the PC1 axis suggested an increase in salinity and a decrease in temperature, DIN and c hlor ophyll fluor escence during the transition of water masses from the surface to subsurface la yers .Indeed, DIN and chlorophyll fluor escence wer e higher in the surface waters than in the subsurface waters, and their variations revealed significantly stronger correlations with depth ( Fig. S2 ).Contr astingl y, the PC2 axis suggested spatially distinguishable waters with variations in silicate and DOC in van Mijenfjorden and Storfjorden.For example, the silicate concentration in van Mijenfjorden was higher than that in Storfjor den; ho w e v er, the DOC v alues sho w ed the opposite pattern ( Fig. S3 ).eDNA concentrations varied among fjord waters but sho w ed no depth-wise distribution ( Fig. S4 ).Despite such eDNA variation, the copy numbers of 16S and 18S rRNA genes showed a r elativ el y conserv ed distribution among the fjor d w aters ( Fig. S4 ).DMSP concentr ations wer e estimated using Pearson's correlation, in conjunction with simple linear r egr ession anal ysis, to anal yze their relationship with changes in depth ( Fig. S5 ).The concentration of DMSP demonstrated a significant negative correlation with depth (r: −0.73, R 2 : 0.48, P < 0.05), manifesting higher values in surface waters (11.51 ± 9.97 nM) than in subsurface waters (1.31 ± 0.47 nM).

Di v ersity and community composition of 16S and 18S rRNA gene sequences
Each of the 210 000 sequences of the 16S and 18S rRN A genes w as clustered at 17 861 and 4378 ASVs , respectively.T hese ASVs were analyzed for alpha and beta diversity.The alpha diversity indices (Chao1 and ACE) of both 16S and 18S rRNA sequences were not significantl y corr elated with depth c hange ( ρ < 0.5, P > 0.01) but sho w ed mor e v ariability in 18S rRNA than in 16S rRNA ( Fig. S6 ).The less variable Chao1 and ACE indices of 16S rRNA suggested spatially similar species richness of bacterial communities among the fjor d w aters .T he 18S rRNA indices wer e r elativ el y lower in van Mijenfjorden and Storfjorden than in the other fjords, indicating a higher species richness of eukaryotes in van Keulenfjorden and Hornsund.Further, the beta diversity in 16S and 18S rRNA on non-metric multidimensional scaling revealed separate clusters between the surface and subsurface waters (Fig. 2 ), and this separ ation was significantl y supported by anal ysis of molecular variance ( P < 0.01; Table S1 ).
Such distributions of bacterial and eukaryotic communities wer e mor e distinctl y visualized at the taxonomic class le v el.Heter otr ophic bacteria, suc h as Alpha pr oteobacteria, Gamma pr oteobacteria and Bacter oidia, wer e dominant among the samples, and their r elativ e abundance sho w ed a depth-wise distribution.Still, the r elativ e abundance of photosynthetic Oxyphotobacteria was limited to the surface waters (Fig. 4 A).Other photosynthetic populations in the eukaryotic communities revealed spatially distributed patterns in the fjord waters (Fig. 4 B).For example, Dinophyceae was the most predominant photosynthetic player in both surface and subsurface waters from van Keulenfjorden and Hornsund, and in subsurface waters from van Mijenfjorden and Storfjor den.Prasinophytae w as dominant in the surface waters of van Mijenfjor den, whereas Phaeophyceae w as pr edominantl y found in the surface waters of Storfjorden.Contr astingl y, Diatomea was not dominant compared with other phytoplankton, but was more e v enl y distributed in the fjord waters.In the case of zooplankton, Intr amacr onucleata was abundant in van Keulenfjorden and Hornsund, and its spatial variation was like that of Dinophyceae.Anthoc y anins w er e r elativ el y dominant in Storfjorden, but other zooplankton (Arthropoda_unclassified or Ostracoda) occasionally occurred in some subsurface waters.At the genus level of taxonomy, the structure of both bacterial and eukaryotic communities was complex, with a considerable number of taxa remaining taxonomically unclassified ( Fig. S8 ).Despite this, the distribution of the most predominant taxa at the class le v el was reflected at the genus level.For example, the distributions of two cold-ada pted gener a, SAR11 clade Ia (Kr aemer et al. 2020 ) and Psychrobacter (Rodrigues et al. 2009 ), r epr esented the pr e v alence of Alpha pr oteobacteria and Gamma pr oteobacteria, r espectiv el y, within the bacterial community.In particular, the r elativ e   S2 ).Although none of the significant r epr esentativ es for fjord-specific populations (indicator value < 0.6) were determined by indicator species anal ysis, fiv e r epr esentativ e populations for surface or subsurface water had significant indicator values ( > 0.7, P < 0.01) (Table 2 ).In bacterial populations, the prokaryotic phytoplankton, Oxyphotobacteria, specificall y occurr ed in the fjord surface waters, but heter otr ophic bacteria, suc h as Acidimicr obiia and Verrucomicr obiae, wer e r epr esentativ e of the subsurface waters.Like bacterial populations, Cryptomonadales_cl (eukaryotic phytoplankton) and Per onospor omycetes (heter otr ophic eukaryotes) r epr esented the surface and subsurface waters, r espectiv el y.These micr obial population dynamics in the Svalbar d fjor ds implied the potential role of photosynthetic (surface) and heter otr ophic (subsurface) players in the planktonic food web.

Netw ork anal ysis of the microbial popula tions
The network anal ysis pr ovided insights into the dynamics of microbial populations by analyzing major ASVs (each representing > 0.1% of the total dataset) through Spearman's correlation at a significant le v el ( ρ > 0.50, P < 0.05).Evidently, nodes determined by ASVs of heterobacteria surrounded the ASVs of autotrophic bacteria and other eukaryotes, including phytoplankton, zooplankton, fungi and unclassified eukaryote.To extract specific interactions among these populations, microbial population dynamics at the class le v el with water mass properties were visualized in the network anal ysis (Fig. 5 ), wher ein micr obial co-occurr ences and their interactions with the water mass properties were evaluated

Prediction of bacterial metabolisms
Bacterial metabolism predictions were conducted using PICRUSt2, based on ASVs derived from 16S rRNA sequences.In total, 17 861 ASVs were identified from 21 000 sequences.Rare ASVs with < 10 sequences were excluded to optimize the use of limited computing r esources, r esulting in 1874 ASVs, corr esponding to 85.13% of the total sequence dataset.Following the exclusion of r ar e ASVs, the metabolic pr ediction identified pr edominant pathways related to the metabolism of carbohydrates, amino acids , cofactors , vitamins and nucleotides within the fjord waters ( Fig. S10 ).We investigated the pathways associated with energy metabolism, xenobiotic biodegradation, terpenoid and polyketide metabolism, and secondary metabolite biosynthesis (Figs 6 and S11 ).Relativ el y featur ed pathways in eac h metabolism wer e found (energy metabolism: sulfur metabolism (36.31%) and nitrogen metabolism (23.06%); xenobiotics biodegradation: benzoate degradation (37.55%) and chloroalkane and chloroalkene degradation (25.35%); metabolism of terpenoids and polyketides: terpenoid backbone biosynthesis (74.79%); biosynthesis of other secondary metabolites: penicillin and cephalosporin biosynthesis (52.27%) and phen ylpr opanoid biosynthesis (22.44%)).

Discussion
The regional hydrographic variability of Svalbard is primarily influenced by two main currents (Arctic and Atlantic waters), and their hydr ogr a phic balance with glacial meltwater controls the water mass properties and subsequent microbial responses in the Svalbard fjords (Rokkan Iversen and Seuthe 2011 , Han et al. 2021Han et al. , 2022 ) ).This study categorized the fjord waters into surface and subsurface lay ers accor ding to w ater mass properties .T he microbial diversity and community composition in these two layers F igure 5. Microbial netw ork model with w ater mass properties.In the netw ork model, eac h node indicates micr obial populations (pale pink circle: zooplankton; dark blue circle: cyanobacteria; purple circle: fungi; green circle: phytoplankton; orange circle: heter otr ophic bacteria) or water mass pr operties (gr a y diamond).T he size of micr obial nodes and edge between nodes indicate their r elativ e abundance and corr elation, r espectiv el y.Thickness of edge represents a correlation coefficient (the thicker the line, the stronger correlation), and the solid lines and dotted lines represent positive and negative correlations, respectively.The isolated nodes with no edge indicate there was no interaction at a significant level (correlation coefficient < 0.5, P > 0.05).
wer e distinct.Unlike pr e vious studies on eDN A metabar coding in Ar ctic fjor ds, w e investigated the interactions between phytoplankton and bacterial communities using compr ehensiv e 16S and 18S rRNA gene metabarcoding.The 16S rRNA metabarcoding results suggested a homogenous community composition comprising a few dominant heterotrophic bacteria in the summer fjor d w aters.Ho w e v er, we found that the distribution of SAR11 clade Ia, specifically within the Alphaproteobacteria, reveals the subtle yet significant sample-to-sample variation within what initiall y a ppear ed as a uniform bacterioplankton comm unity.This finding suggests the complex interactions between microbial communities and their en vironments , indicating that taxa with widespread distribution, like Alphaproteobacteria, display considerable ecological variation.This variation highlights the ada ptiv e ca pacity of these or ganisms acr oss differ ent envir onmental nic hes, r eflecting their substantial r ole in ecosystem dynamics .Con v ersel y, the div ersity indices deriv ed fr om the 18S rRNA data suggested a spatiall y div erse composition for eukaryotic communities.Notably, phytoplankton populations within eukaryotic communities exhibited spatial variations in surface waters, whereas c y anobacteria, the photosynthetic population in bacterial comm unities, wer e dominant in surface waters.Although the r elativ e abundance of heter otr ophic bacteria v aried fr om surface to subsurface waters, their widespr ead pr esence thr oughout the fjord waters indicates a robust heterotrophic component capable of utilizing a broad range of organic substrates, often originating from phytoplankton.T his interpla y between phytoplankton and heter otr ophic bacteria suggests an ecological interface within the Ar ctic fjor ds .For example , the fate of phytoplankton-derived organic matter is linked to the biogeoc hemical pr ocesses mediated by heter otr ophic bacterial metabolism during the oceanic nutrient cycle (Seymour et al. 2017 ).
DMSP, synthesized by marine plankton, is ubiquitous and is the most important source of biogenic sulfur and carbon in the ocean (Zindler et al. 2013 ).This phytoplankton-deriv ed or ganic matter can be converted to substantial nutrients by heterotrophic bacterial activity.Indeed, the measured DMSP in the fjord waters suggests the existence of DMSP-rich phytoplankton in the summer surface waters of the Svalbard fjords .Moreo ver, we found that Dinophyceae is a predominant eukaryotic plankton in fjord waters, and such dominant phytoplankton have been considered one of the main producers of DMSP pools in the ocean (Keller et al. 1989, Steinke et al. 2002, Zindler et al. 2013 ).
We summarized a compar ativ e description of DMSP in a microbial network model to understand the fate of phytoplanktonderiv ed or ganic matter in fjor d w ater.Micr obial co-occurr ence in the network model suggested biological inter actions, suc h as competition, mutualism and predation, among the microbial communities in the Svalbard fjords.Microbial connections in the network r epr esented the tr ophic r ole of the planktonic food web.For example, the link between Diatomea (phytoplankton) and Arthropoda (zooplankton) indicated their general feeding habits within plankton comm unities.Notabl y, a similar distribution of Dinophyceae (phytoplankton) and Intr amacr onucleata (zooplankton) was observed in the network model, with a strong linkage.Given that Intr amacr onucleata is a subphylum of ciliates that feed on small ph ytoplankton, such as Dinoph yceae, the dominant Dinoph yceae in the summer fjord waters may support the distribution of phytoplankton feeders, such as Intramacronucleata.Moreover, organic detritus pr oducts, suc h as DMSP, fr om plankton feeding habits, ma y pro vide nutrients for heter otr ophic bacterial gr owth in the fjor d w aters .T his assumption is supported by the qPCR measurement of bacterial DMSP degradation genes in summer Svalbard waters (Han et al. 2021 ).
We assumed that heter otr ophic bacteria influenced DMSP metabolism in fjord waters.First, the homogeneous distribution of heter otr ophic bacteria may fr equentl y include DMSP in the fjor d w aters.Although the most dominant bacteria, such as Alpha pr oteobacteria, Gamma pr oteobacteria and Bacter oidia, wer e not linked to phytoplankton in the network model, these predominant heter otr ophic bacteria may consume dissolv ed DMSP in the fjord waters after phytoplankton release .Nonetheless , less dominant bacteria, such as Verrucomicrobiae and Actinobacteria, wer e dir ectl y connected to ph ytoplankton (Prasinoph ytae or Dinophyceae) in the network model.As measured in this study, the DMSP concentration in natural seawater generally falls within the tens of nanomolar range (Kettle et al. 1999, Barak-Gavish et al. 2018 ).Ho w e v er, phytoplankton-deriv ed or ganic matter is found at higher concentrations in the immediate surroundings of phyto-plankton cells (phycospher e), wher e the organic matter released by the cell enhances the local environment of the surrounding water (Stocker 2012, Seymour et al. 2017 ).Mutualistic infochemical exchange within the phycosphere is an attractive strategy for recycling phytoplankton detritus, including DMSP, using heter otr ophic bacteria.Indeed, marine Actinobacteria, connected to both Prasinophytae and Dinophyceae in the network, have been recognized for their significant involvement in the decomposition of cellulose, chitin, agar, laminarin, alginates and various hydr ocarbons (Maniv asa gan et al. 2014 ), and incr easing e vidence suggests that Actinobacteria are involved in DMSP metabolism (Yoch et al. 2001, Mizuno et al. 2015, Liu et al. 2018 ).Ther efor e, understanding the interactions between phytoplankton and heter otr ophic bacteria is essential for predicting the responses of ecological and biogeochemical processes in Ar ctic fjor ds and other marine en vironments .
Increasing information on phytoplankton-bacterial interactions has r e v ealed a pattern of mutual exchange of substrates, such as essential vitamins and nutrients (Croft et al. 2005, Amin et al. 2009, Wang et al. 2014 ), and infoc hemicals, whic h conv ey information (Pohnert et al. 2007, Sey edsay amdost et al. 2011, Amin et al. 2015, Barak-Gavish et al. 2018 ), within the phycosphere (Segev et al. 2016, Landa et al. 2017 ).We predicted metabolic pathways for potential bacterial activity in phytoplankton-derived organic matter, including DMSP .Additionally , we assumed that the interactions between phytoplankton and bacteria could be estimated using the following metabolic pathways: sulfur metabolism (energy metabolism), benzoate degradation (xenobiotic biodegradation), ter penoid bac kbone biosynthesis (metabolism of terpenoids and polyketides), and penicillin and cephalosporin biosynthesis (biosynthesis of other secondary metabolites).
Hypothetical pathways for sulfur metabolism involved in bacterial DMSP and benzoate degr adation wer e suggested in fjord waters.DMSP serves as a chemoattractant and facilitates bacterial interactions with ph ytoplankton.Ph ytoplankton-derived aromatic benzoate compounds promote bacterial growth, and bacterial interactions with phytoplankton are influenced by their capacity to utilize benzoate (Barak-Gavish et al. 2023 ).For example, the bacterial detection of DMSP, along with more specific compounds, such as benzoate, contributes to the recognition of phytoplankton hosts within the phycosphere by bacteria.Similarly, the biosynthesis of terpenoids (or terpenes) and β-lactam antibiotics (penicillin and cephalosporin) ma y pro vide a competitive advanta ge for phycospher e de v elopment.For example, ter penoids play div erse r oles in mediating anta gonistic and beneficial inter actions among or ganisms a gainst pr edators , pathogens and competitors , and convey messages to mutualists, signaling the presence of nutrients and potential threats (Gershenzon and Dudar e v a 2007 ).Although terpenoids are mainly produced by plants and fungi, some bacterial taxa in the phycosphere also synthesize terpenoid compounds (Lu et al. 2023 ).Mor eov er, synthesizing antibacterial compounds r epr esents a mec hanism thr ough whic h a specific bacterial species can outcompete another, leading to enhanced and successful colonization linked with phytoplankton (Zhu et al. 2021 ).
Although our assumptions derived from this study provide meaningful information, certain limitations need careful consideration.First, this study focused on the post-blooming of phytoplankton in the summer season owing to the limitation of field observation.Ho w ever, microbial interactions within summer may not r epr esent the entir e nutrient cycle of the Arctic fjords.Considering the microbial response to the pre-blooming in the spring season could provide a comprehensive understanding of the interaction between phytoplankton and bacterial communities in the fjords.Second, the influence of heter otr ophic bacteria on DMSP metabolism is derived from the predicted metabolic functions.The prediction should be validated through the direct measurement of DMSP metabolism.Addressing these limitations could contribute to a robust and comprehensive understanding of the ecological and biogeochemical processes in the Arctic fjords.In addition, using rRNA gene regions for metabarcoding can lead to an ov er estimation of microbial community abundances, attributed to variations in rRN A gene cop y numbers across different species.Species within microbial communities exhibit significant variation in the number of rRNA gene copies in their genomes.Consequently, taxa with higher gene copy numbers might be perceived as more abundant than they truly are in metabarcoding studies, as each gene copy is equall y likel y to be sequenced.This variation necessitates a cautious interpretation of metabarcoding data, acknowledging its semi-quantitative nature.Despite these challenges, metabarcoding is a powerful tool for exploring biodiversity and understanding ecological and biological processes , pro vided its limitations are recognized and appropriately addressed.
In conclusion, eDN A metabar coding is a pr omising tec hnique for studying phycosphere community composition in the Arctic fjor d of Svalbar d.Although r ecent micr oscopic observ ations hav e r e v ealed the biogeogr a ph y of ph ytoplankton and their contribution to the Arctic fjord food web, compr ehensiv e studies validating the interactions between phytoplankton and bacterial communities in Arctic fjords are currently limited.Our results provide valuable insights into the responses of phycospheres to environmental change in this sensitive ecosystem.In particular, the metabolic pathw ays predicted b y PICRUSt2 suggested a potential interaction between phytoplankton and heter otr ophic bacteria.Our predicted metabolic processes may provide clues for seeking direct evidence of the role of the phycosphere in Arctic fjords using further deep sequencing, such as shotgun metagenome sequencing.

Ac kno wledgements
We would like to thank the shipboard scientific party, captain and cr e w of the R/V Helmer Hanssen for their help during the 2019 cruise.

Figure 1 .
Figure 1.Description of r esearc h ar ea and samples in the Sv albar d. (A) Sampled locations and (B) w ater mass properties of sampled w aters visualized by principal component analysis (PC A).T he geographic locations of the sampled area in Svalbard were visualized using the ggmap package (Kahle et al. 2019 ) in R. In PCA, samples enclosed with red colored dashed lines indicate surface waters (shallower than 10-m depth).
(EUK345f: 5 -AA GGAA GGCA GCA GGCG-3 ; EUK499r: 5 -C ACC AGACTTGCCCTCY AA T-3 ) sets .T he quantification standard for the threshold cycle (Ct) calibration of a target gene constitutes a dilution series of a known amount of genomic DN A (gDN A) for Esc heric hia coli ATCC 11775 and Yarrowia lipolytica ATCC 20306.qPCRs for the target genes in samples and standards were performed with the following conditions: initial denaturation at 94 • C for 3 min, follo w ed b y 40 cycles of denaturation at 94 • C for 20 s, specific annealing of each target gene at 60 • C for 30 s and elongation at 72 • C for 30 s.At the end of each run, a dissociation melt curve of the PCR product was determined to verify amplicon specificity.The copy number of 16S and 18S rRNA genes was calculated using the following formula with putative gDNA of E. coli (5 038 133 bp genome and se v en copies of 16S rRNA gene) (Meier-Kolthoff et al. 2014 ) and Y. lipolytica (20 262 281 bp genome and 182 copies of 18S rRNA gene) (Tejerizo et al. 2017 ), r espectiv el y. rRNA genes per nanogram gDNA = [6.022× 10 23 (Avogadro's constant) / (base pairs of genome × 660 (av er a ge mass of 1 bp of dsDNA) × 10 9 (conversion factor)] × number of rRNA genes per genome).

Figure 3 .
Figure 3. Taxonomic classification of (A) 16S and (B) 18S rRNA data at phylum le v el.

Figure 4 .
Figure 4. Composition of (A) bacterial and (B) eukaryotic communities at class level.

Figure 6 .
Figure 6.Prediction of metabolic pathways for bacterial activity in energy metabolism, xenobiotic biodegradation, metabolism of terpenoids and polyketides, and biosynthesis of secondary metabolites.

Table 1 .
Sample description.Detailed information on associated metadata for sequenced samples is provided in the supplementary data.

Table 2 .
Indicator species analysis for 16S and 18S rRNA data in surface or subsurface water groups .T he highest indicator values with significance ( P < 0.01) are highlighted in bold and underlined.