Sampling was undertaken off the Californian coast in waters influenced by two opposing currents. The California Current flows equatorward from the North Pacific Current (∼ 50∘ N) to Baja California (∼ 15–25∘ N). Southerly along-shore winds drive upwelling of cold nutrient-rich waters in early spring and summer in central California, and weaker but more sustained upwelling further toward the south. The relatively warm, saline Davidson Current and California Undercurrent flow poleward over the continental shelf. During the southern California summer, the California Current moves farther offshore, and the Davidson Current predominates near shore (Checkley and Barth, 2009). For this study, samples were collected along the narrow central California shelf ∼ 1 km off Bodega Head, California (38.3∘ N, 123.0∘ W), and in the Southern California Bight off Santa Catalina Island (33.4∘ N, 118.4∘ W; Fig. 1). At both sites, local variation in foraminiferal abundances and species composition is well understood (Thunell and Sautter, 1992; Field, 2004, Davis et al., 2016), driven by periods of upwelling, relaxation or downwelling, and/or seasonal predominance of the Davidson Current. The Santa Catalina Island site is close to the San Pedro Ocean Time-series (SPOT; 33.55∘, 118.4∘ W) station, where the composition of the bacterial assemblage in the water column has been monitored for over a decade (Chow et al., 2013; Cram et al., 2015).

The Bodega Head samples were collected during November 2014, when there is a relaxation in the upwelling before the winter storms, and also in April 2015 (Table 1). Normally, the southerly along-shore winds start to drive renewed upwelling in early spring (Garcia-Reyes and Largier, 2012) but the samples for this study were collected prior to the delayed upwelling season of 2015, an unusual phenomenon possibly due to the strong El Niño–Southern Oscillation (e.g. Jacox et al., 2015). Bodega Head samples were obtained from vertically integrated 150 µm mesh-size net tows, deployed to a maximum depth of 160 m, or to 10 m above the seafloor at shallower sites. Tow material was transferred to ambient surface seawater and kept chilled during transit to shore at the Bodega Marine Laboratory, where live foraminifera were wet picked. G. bulloides were then identified morphologically to the morphospecies level, rinsed in 0.6 µm filtered surface seawater and preserved in RNALater^® (Ambion^™). This reagent conserves cell integrity, inhibits nucleases at ambient temperatures, and dissolves the calcite shell. The Santa Catalina Island samples were collected by scuba diving or net tows during July/August 2013 towards the end of the weaker upwelling season off southern California. Collected foraminifera were treated as at the Bodega Marine Laboratory and transferred to RNALater^® at the Wrigley Marine Science Center.

To remove the shell and shell-associated, external contaminants, each individual specimen was decalcified by exposure to RNALater^® (Ambion^™). The cell was then washed in filter-sterilised, salt-adjusted phosphate-buffered saline (PBS) or sterile artificial seawater, transferred to a new sterile 1.5 mL tube and washed a further three times before being transferred to DOC DNA extraction buffer (Sect. 2.4; Holzman and Pawlowski, 1996) for DNA analysis, or 4 % (w/v) paraformaldehyde in salt-adjusted PBS for microscopy. Preliminary experiments with the kleptoplastic foram Elphidium species confirmed the efficacy of this treatment. The majority of prokaryote shell contaminants (> 80 %) were removed in the washing steps following dissolution in RNALater^® (Bird et al., 2017).

DNA was extracted from individual foraminifer specimens using the DOC extraction method to identify the specific genotype (Holzman and Pawlowski, 1996). PCR amplification of the foraminiferal 18S rRNA gene was performed according to Seears et al. (2012). DNA sequencing was carried out using the BigDye^® Terminator v3.1 Cycle Sequencing Kit and an ABI 3730 DNA sequencer (both Applied Biosystems).

Foraminifer cells were stained with 4′,6-diamadino-2-phenylindole (DAPI), which forms a highly fluorescent DAPI–DNA complex that allows the visualisation of bacterial cells and eukaryotic cell nuclei under fluorescence microscopy. Individual decalcified and washed foraminifera were fixed in 4 % (w/v) paraformaldehyde in salt-adjusted PBS for 4 h at 4 ∘C. Fixed cells were transferred to a polylysine-coated microscope slide and dehydrated through an ethanol series of 70, 90 and 100 % ethanol. Cells were stained in 1 µg mL-1 DAPI (dilactate, Sigma-Aldrich) in PBS for 3 min and then rinsed with sterile deionised water. The stained preparations were mounted in AF1 mountant solution (Citifluor) and bacteria and eukaryotic nuclei visualised using a Zeiss Axio Imager Fluorescence microscope equipped with a DAPI filter set.

An unstained specimen of G. bulloides was also examined by fluorescence microscopy to observe the background levels of autofluorescence under the DAPI filter set to compare with the appearance of DAPI stained individuals. A TRITC filter set (excitation wavelength 540 nm, emission wavelength 580 nm) was used also on unstained individuals to investigate for the presence of autofluorescent, phycoerythrin-containing cyanobacterial cells.

DNA for 16S rRNA gene metabarcoding of the bacterial population within the foraminifera was extracted from decalcified and washed planktonic foraminiferal cells by the DOC extraction method (Holzman and Pawlowski, 1996). The DNA from the six samples – three independent G. bulloides isolates (BUL34, BUL36, BUL37) and a single non-spinose Neogloboquadrina dutertrei (DUT55) collected in July/August off Santa Catalina Island and two additional G. bulloides cells collected in November off Bodega Head (BUL22, BUL23; Table 1) – was amplified alongside three reagent controls containing (i) no DNA template (two replicates) and (ii) DOC buffer only. The V4 region of the 16S rRNA gene was chosen for amplification using the 515F forward primer and a barcoded 806R reverse primer series (Caporaso et al., 2012). This primer set is widely used by the Earth Microbiome Project (Gilbert et al., 2010), and therefore the amplification biases are known and well documented. For example, there is a bias against amplification of the SAR11 group of marine Alphaproteobacteria and a bias towards over-amplification of Gammaproteobacteria (Apprill et al., 2015; Walters et al., 2016; Parada et al., 2016). Each DNA sample and control was PCR-amplified with a unique barcode tag that enabled demultiplexing of the samples after being pooled for sequencing. The thermal cycling conditions are detailed by Caporaso et al. (2012). PCR reactions contained 1× Taq buffer plus additional MgCl2 (final concentration 2.5 mM), 0.2 mM of each dNTP, 0.25 µM of each primer, 1 µL of template DNA and 1.25 U of Taq DNA polymerase (Roche Applied Science), with the volume made up to 25 µL with PCR-grade water (Sigma). All PCR reactions were set up in a PCR6 vertical laminar airflow cabinet with UV sterilisation (Labcaire Systems, Bristol, UK) as described by Pagaling et al. (2014). Reaction tubes and PCR mixtures were treated for 15 min with 15 W UV light (wavelength = 254 nm) to destroy contaminating DNA, prior to addition of dNTPs, Taq polymerase and template DNA (Padua et al., 1999). The six functional and three control PCR reactions were run on a 1 % agarose gel and the products were purified with the Wizard^® SV Gel and PCR Clean-Up System (Promega). The purified amplicons were quantified using a Quant-iT PicoGreen ds-DNA assay kit (Life Technologies) prior to pooling at equimolar concentrations for DNA sequencing. The total number of quality-filtered sequencing reads including controls was 862 954. DNA sequencing was performed at Edinburgh Genomics using an Illumina MiSeq v3 to generate 250 base pair (bp) paired-end reads.

TEM was used to observe and document the structural relationships between the endobiotic bacteria and foraminiferal cells. Decalcified G. bulloides were fixed in 3 % glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.3, for 2 h followed by three 10 min washes in 0.1 M sodium cacodylate. Specimens were then post-fixed in 1 % osmium tetroxide in 0.1 M sodium cacodylate for 45 min, followed by a further three 10 min washes in 0.1 M sodium cacodylate buffer. Specimens were then dehydrated in 50, 70, 90 and 100 % ethanol (X3) for 15 min each, then in two 10 min changes in propylene oxide prior to being embedded in TAAB 812 resin. Sections (1 µm thick) were cut on a Leica Ultracut ultramicrotome, stained with toluidine blue, and then viewed under a light microscope to select suitable specimen areas for investigation. Ultrathin sections (60 nm thick) were cut from selected areas, stained in uranyl acetate and lead citrate and then viewed with a JEOL JEM-1400 Plus TEM.

Synechococcus cells were found in large numbers inside G. bulloides and were genetically characterised. A 422 bp fragment of the Synechococcus 16S rRNA gene was amplified from total DNA extracted via the DOC method from individual specimen BUL34 (Table 1). This provided a larger, more informative fragment for phylogenetic analysis compared with the 253 bp generated by 16S rRNA metabarcoding. Cyanobacterial specific primers were used (CYA359f 5′-GGGGAATCYTTCCGCAATGGG-3′ and CYA781R a and b 5′-GACTACWGGGGTATCTAATCCCWTT-3′; Nübel et al., 1997) and thermocycler conditions were as follows: 94 ∘C for 2 min followed by 30 cycles at 94 ∘C for 15 s, 55 ∘C for 15 s and 72 ∘C for 30 s, followed by a final extension at 72 ∘C for 5 min. PCR reactions were performed with MyTaq REDDY mix (Bioline) and 0.25 µM of each primer and 1 µL of template DNA, with the volume made up to 25 µL with PCR-grade water (Sigma). The PCR product obtained was cloned (TOPO^®-TA cloning kit, Invitrogen) and Sanger-sequenced.

Clone sequences were aligned with reference Synechococcus 16S rRNA gene sequences retrieved from the GenBank database (NCBI) using ClustalW software within the package MEGA6 (Tamura et al., 2013). Phylogenetic trees (maximum likelihood, neighbour joining, minimum evolution, UPGMA, maximum parsimony; Sect. 3.5) were generated using the default settings of MEGA6 with 500 bootstrap resamplings to determine the closest taxonomic affiliations (i.e. clade designation sensu Fuller et al., 2003) of the G. bulloides-associated Synechococcus. Informed by this analysis, further primers were designed that target other signature genes harboured by the Synechococcus clades identified, including that for rbcL. This phylogenetically informative gene encodes the large subunit of RuBisCO (ribulose-1,5-bisphosphate carboxylase/oxygenase), the primary CO2-fixing enzyme found in cyanobacteria. Primers (SynrbcL_For 5′-CGGCAACTTCTTCGATCAGG-3′; SynrbcL_Rev1 5′-ATGTCGCGGCTTTCTTTCTC-3′; SynrbcL_Rev2 5′-CCGGCTTCCATAAGGATGTC-3′) were designed with Primer 3 (http://primer3.ut.ee/) that target a 252 bp fragment of rbcL from the most closely related Synechococcus spp. (i.e. strains CC9902, CC9311 and WH8120; see below).

Purified DNA from 14 G. bulloides specimens (Table 1) generated products of the correct size on PCR amplification with the rbcL primers. The product obtained from isolate BUL34 was selected and TA-cloned and DNA-sequenced as described above. PCR reactions were performed in a Biometra Personal Thermocycler using MyTaq REDDY mix (Bioline) and 0.25 µM of each primer and 1 µL of template DNA, with the volume made up to 25 µL with PCR-grade water (Sigma). Thermocycling conditions were 94 ∘C for 2 min followed by 30 cycles at 94 ∘C for 15 s, 56 ∘C for 15 s and 72 ∘C for 30 s, followed by a final extension at 72 ∘C for 5 min.

Partial 18S rRNA (SSU) gene sequences amplified from BUL34 and DUT55 have been submitted to Genbank (NCBI, accession numbers KX816046 and KX816047 respectively). G. bulloides specimen BUL34 is type IId and N. dutertrei specimen DUT55 is type Ic and it is the first time a ∼ 1000 bp fragment has been amplified for this genotype. Both genotypes have been found routinely in the Southern California Bight (Darling et al., 2003).

Fluorescence microscopy examination of an unstained, fixed G. bulloides specimen (BUL21; Table 1; Supplement Fig. S1) under the DAPI filter set demonstrated high levels of diffuse autofluorescence across the entire cell. G. bulloides cells that were first stained with DAPI (n=6; Table 1) also showed background autofluorescence, but in addition some more highly fluorescent regions 3–10 µm in size were also observed (Fig. 2a). These stained structures are probably DNA from the foraminiferal nucleus and also DAPI–DNA complexes in organisms sequestered within food vacuoles. Of greater present significance, however, were the very many fluorescent, globular structures of approximately 1 µm diameter observed consistently within all of the G. bulloides cells analysed. Their small, regular size is consistent with the presence of intact intracellular (coccoid) bacteria residing within the foraminiferal cell (Fig. 2b). Further microscopic examination of G. bulloides using the TRITC filter set also revealed high autofluorescence across the cell but, in addition, brighter fluorescence from many of these approximately 1 µm diameter DNA-containing structures (Fig. 3). This observation is entirely consistent with the presence of the photosynthetic pigment, phycoerythrin (excitation maxima ∼ 495, 545 nm, emission maximum 565–580 nm), within many of these intracellular (cyano)bacteria (see below).

16S rRNA gene metabarcoding was carried out on five specimens of G. bulloides and a single specimen of the non-spinose species Neogloboquadrina dutertrei for comparison (Table 1). The raw dataset has been submitted to the Sequencing Read Archive (SRA, NCBI) (Bioproject accession number PRJNA341960; SRA accession numbers SRR4271458, SRR4271479, SRR4271493, SRR4271505, SRR4271506, SRR4271507, Bird et al., 2017). A total of 862 954 sequences were generated from the six samples and three controls after quality filtering and removal of short reads (< 245 bp) and chimeric sequences (Sect. 2.6.1). In closed-reference picking, after removal of pynast failures, all control sequences (288 985) contaminant OTUs (including 161 282 sequences as result of the single contaminating Bradyrhizobiaceae OTU across all six samples) and OTUs with an abundance of less than 10 sequences across all samples, a total of 214 087 sequences were clustered into OTUs and assigned taxonomy (Sect. 2.6.1 and 2.6.2). The numbers of sequences and OTUs generated in individual specimens for both closed-reference picking and de novo picking are listed in Supplement Table S1. The OTU profiles within a specimen were highly similar between de novo picking and closed-reference picking with normalisation by copy number (NBCN, Sect. 2.6.2). Therefore, we present results for closed-reference picking with NBCN and indicate when de novo picking OTUs are being represented. Rarefaction curves (Fig. S2) for OTU richness confirmed that sequencing depth was sufficient to capture the full bacterial assemblage diversity.

All replicates of G. bulloides contained a highly distinctive assemblage of OTUs. The OTU assemblage of an individual N. dutertrei (Table 1) sampled at the same time and from the same water mass is shown for direct comparison of the bacterial assemblages of two morphospecies of foraminifera (Fig. 4). Within the five BUL specimens investigated (Table 1), 37–87 % of all sequences belonged to five OTUs assigned to the unicellular, cyanobacterial genus Synechococcus (Fig. 4). This was by far the greatest abundance of a single genus of bacteria; no other genera, or indeed family or order, of bacteria were found across all five BUL specimens at relative abundances consistently more than 2 %. Therefore, the next highest relative abundance group must be described at the class level. Across four (BUL22, BUL23, BUL34, BUL36) of the five BUL specimens 15–31 % of sequences belonged to the class Alphaproteobacteria and were dispersed amongst 81 OTUs. The fifth and outlying specimen (BUL37, containing 87 % Synechococcus sequences) contained only 2.6 % Alphaproteobacteria across 37 OTUs. It contained marginally more phylum Actinobacteria (1.8 %) and class Betaproteobacteria (1.4 %) sequences, but the sequence abundances of these classes were more similar to the other BUL specimens. The high relative abundance of Synechococcus within this specimen might be due to a lack of feeding on other bacteria or algae immediately prior to sampling, indicative of a high turnover rate for prey bacteria and algal cells.

There were no chloroplast-affiliated OTUs in specimens BUL22 and BUL23. However, 6.4, 27 and 3 % of sequences in specimens BUL34, BUL36 and BUL37 respectively were allocated to chloroplast 16S rRNA gene OTUs from a variety of sources. These were two OTUs from a mixotrophic protist belonging to the diverse, protozoan phylum Cercozoa (Cavalier-Smith and Chao 2003); three OTUs from the phylum of Haptophyte algae, which includes coccolithophores; eight OTUs from the phylum Stramenopile, which includes both diatoms and chrysophyte algae; and, finally, 13 OTUs from the group Streptophyta (an unranked clade of plants that includes green algae).

In de novo picking, G. bulloides specimens BUL34, BUL36 and BUL37 contained varying percentages of sequences (25.4, 0.3 and 1.6 % and respectively) in three taxonomically unassigned OTUs, whereas BUL22 and BUL23 did not contain any unassigned sequences. Of all the unassigned sequences, 94 % belonged to a single OTU (e.g. 5878 sequences in sample BUL34) that was 99 % identical (100 % coverage of 253 bp) to an unidentified marine bacterial clone (accession number HQ673258) retrieved from the northeast subarctic Pacific Ocean (Allers et al., 2013). The nearest match (87–89 % similarity over 100 % coverage) to an identified phylum was to a large number of uncultured Verrucomicrobiae bacteria of the phylum Verrucomicrobia from a wide range of habitats, including marine environments. Whilst this was not a particularly close match, it is within the defined limit of > 85 % DNA identity that delineates a phylum (Hugenholtz et al., 1998; Rappé and Giovanonni, 2003), albeit based only on a 253 bp fragment. The variation in abundances of this OTU found in each G. bulloides individual analysed (25 % in BUL34; 0.3 % in BUL36 and 1.6 % in BUL37) might indicate this bacterium has a patchy distribution and is an opportunistic food source.

Transmission electron microscopy (TEM) imaging was carried out on four G. bulloides specimens (BUL32, BUL39, BUL69, BUL71; Table 1, Fig. 5) to observe whether any endobionts were present within the cell. No intracellular eukaryotic cells were observed, confirming a lack of algal symbionts. However, numerous intact coccoid cells containing carboxysomes (Fig. 5a) within the central cytoplasm surrounded by thylakoid membranes, characteristic of the cyanobacterium Synechococcus, were observed throughout the cytoplasm and also in vacuoles of all individual G. bulloides observed (Fig. 5b). Approximately 5 % of the observed Synechococcus cells were undergoing cell division (Fig. 5c).

To compare foraminiferal cellular Synechococcus concentrations with those of the water column, the concentration of Synechococcus cells per millilitre of foraminiferal cytoplasm was calculated by assuming a conservative average host cell diameter of 200 µm (Spero and Lea, 1996; Aldridge et al., 2012), a spheroid morphology (Geslin et al., 2011) and that the cytoplasm was equivalent to 75 % of the shell volume (Hannah et al., 1994). Based on averaged cell counts from the TEM images, the total number of Synechococcus cells within G. bulloides occupied less than 2 % of the foraminiferal cell volume but was equivalent to 3.8 × 109 Synechococcus cells mL-1. This is far higher than the well established range of concentrations of Synechococcus found throughout the global ocean that range from 1 × 102–1.5 × 106 Synechococcus cells per millilitre of seawater (Partensky et al., 1999; Paerl et al., 2011). In the Southern California Bight, Synechococcus cell counts are generally fewer than 1.5 × 105 cells mL-1 but can reach 6 × 105 cells mL-1 during the blooms that are generally observed in late spring to early summer (Tai and Palenik, 2009; Tai et al., 2011). The concentration of Synechococcus in the G. bulloides cell was therefore up to 4 orders of magnitude greater than peak bloom concentrations measured in the California Bight. This suggests that Synechococcus cells accumulate within the cytoplasm of G. bulloides type IId.

Five Synechococcus OTUs were assigned in 16S rRNA gene metabarcoding with closed-reference picking. However, more than 99 % of the BUL Synechococcus sequences were assigned to just one of these OTUs. The representative nucleotide sequence (253 bp) of this OTU is a 100 % match to the coastal clade IV Synechococcus sp. strain CC9902, originally isolated from the California Current. Two further OTUs were highly similar to this abundant OTU and were 99 % identical to Synechococcus sp. strain CC9902. The remaining two OTUs both had a nucleotide match of 99 % with Synechococcus sp. strain WH8020, a clade I strain also found typically in coastal waters. In order to confirm these clade assignments, phylogenetic analysis (Fig. S3) of a larger (422 bp) fragment of the Synechococcus 16S rRNA gene generated from BUL34 total DNA was performed. Ten clones (GenBank accession numbers KX815969–KX815978) clustered with the clade IV Synechococcus sp. strain CC9902 and two clones (GenBank accession numbers KX815979 and KX815980) clustered with clade I strains CC9311 (another California Current isolate) and WH8020, in agreement with the 16S rRNA gene metabarcoding data. The topologies of the phylogenetic trees produced were all in overall agreement with well-established analyses of Synechococcus 16S rRNA genes (Scanlan et al., 2009), confirming the phylogenetic resolution of the sequence data included in the present study.

In addition, a 252 bp fragment of the Synechococcus rbcL gene was cloned and 230 bp of this clone was DNA-sequenced (GenBank accession number KX816048) from BUL34 (Table 1). A GenBank BLAST search (NCBI) found 100 % nucleotide sequence identity with the RuBisCo large subunit coding region of Synechococcus sp. strain CC9902 and 92 % identity with Synechococcus sp. strain WH8020, confirming the presence of Synechococcus sp. strain CC9902, or a very closely related clade IV strain. The DNA of 13 further G. bulloides specimens (Table 1) also yielded products of ∼ 252 bp on amplification with the Synechococcus rbcL primers confirming the consistency of the association between G. bulloides type IId and Synechococcus strains in the California Current year round.

G. bulloides has consistently been reported to be barren of protist algal symbionts (Febvre-Chevalier, 1971; Gastrich, 1987; Hemleben et al., 1989; Spero and Lea, 1996). The current study supports this conclusion, since no intact algal cells were found in any of the G. bulloides cell sections examined using TEM. However, we do have strong evidence that G. bulloides type IId contains large numbers of the photoautotrophic picocyanobacterium Synechococcus. Intact Synechococcus cells accumulate within the host cytoplasm in abundances far greater than those found under bloom conditions in the California Bight or in any other foraminiferal species investigated. How this association occurs is unclear, but G. bulloides type IId is the only foraminiferal species currently known to associate with Synechococcus, an observation that implies a specific potentially mutualistic benefit, particularly since G. bulloides does not exploit protist algal symbionts as in other spinose species. Based on the observations discussed below, we propose that these picocyanobacteria are abundant, metabolically active endobionts living within the G. bulloides cell, rather than prey.

There are some obvious potential metabolic benefits to each organism in a G. bulloides–Synechococcus partnership. Firstly, the foraminifer might benefit from a supply of photosynthetically fixed carbon, as is the case with the foraminifera that harbour protist algal symbionts (Caron et al., 1995; Uhle et al., 1997, 1999). If this were the sole benefit, however, one would question why G. bulloides preferentially recruits Synechococcus for this purpose, rather than the more conventional algal symbionts found in other species. One possible explanation is that G. bulloides inhabits a wide range of depths that often extend below the photic zone and it is also common in unstable upwelling waters, where potential algal symbionts may not thrive. Synechococcus has been found alive in aphotic waters at depths of 600 m (Bernhard et al., 2000), and has been shown to assimilate carbon mixotrophically (Paoli et al., 2008). It could therefore augment phototrophy with carbon assimilated through (photo)heterotrophy, depending on the water column depth of the host. In addition, some Synechococcus strains within clade I and those so far characterised in clade IV, as found in G. bulloides, are chromatic adapters, able to modify their pigment composition and absorption properties depending on the underwater light field (Six et al., 2007). Such adaptability might make Synechococcus a more compatible symbiont for the G. bulloides lifestyle.

Alternatively, Synechococcus may have additional or quite separate functional roles in association with G. bulloides beside endobiotic photosynthetic activity within the photic zone. For example, approximately half of the nitrogen assimilated by the host cell in the Orbulina universa foraminifer–symbiont system is transferred via the algal symbionts, a contribution that increases further to ∼ 90–100 % in nitrate-depleted waters (Uhle et al., 1999). Synechococcus has a very high affinity for combined nitrogen (e.g. nitrate, nitrite and ammonium) and accumulates expanded stores of this element within its light-harvesting phycobilisomes under N-replete conditions (Wyman et al., 1985). Likewise, Synechococcus sequesters large stores of P within its cells as polyphosphate, even under low external concentrations (Martin et al., 2014). These nutrient reservoirs could be readily mobilised and exploited by the foraminiferal cell, particularly prior to gametogenesis, when planktonic foraminifera require extra elemental resources for DNA production (Hemleben et al., 1989). For Synechococcus, being housed within a foraminiferal cell could protect it from grazers and the multitude of cyanophages present in the water column (Suttle and Chan, 1994; Mühling et al., 2005). Synechococcus may also benefit presumably from a supply of host metabolic by-products or from specific nutrients as products of prey digestion.

TEM in combination with 16S rRNA gene metabarcoding enables identification of both bacteria and eukaryotic chloroplasts within the foraminiferal cell. This methodology does not amplify eukaryotic, nuclear-encoded (18S) rRNA genes and, as a result, does not provide any information about the non-chloroplast-bearing zooplankton prey of G. bulloides. Observations of large numbers of freshly collected specimens of G. bulloides confirm that they feed on small zooplankton prey as well as phytoplankton (Spero and Lea, 1996). Amongst the latter, a preference for some species of diatoms and chlorophytes over dinoflagellates or chrysophytes has been reported (Lee et al., 1966). Interestingly, however, two of the five G. bulloides specimens in this metabarcoding study (BUL22, BUL23) did not contain any chloroplast DNA, indicating that they had not fed on phytoplankton prior to sampling. However, these specimens were sampled in November (off Bodega Head) during the period of relaxation in upwelling, from vertically integrated net tows and may have been obtained from resident populations as deep as 150 m. In contrast, the three individuals in which chloroplast 16S rRNA sequences were present (BUL34, BUL36 and BUL37) were sampled from shallow water nets in July/August (off Santa Catalina Island) towards the end of the weak summer upwelling period. These differences in OTU composition therefore could be as a result of location, depth or seasonal differences in available diet. The three G. bulloides with chloroplast sequences (6.4, 27 and 3 %, respectively) were clearly feeding on a range of photosynthesising eukaryotes (Sect. 3.3). OTUs indicate these to be Cercozoa (mixotrophic protists), Streptophyta (which includes green algae), Haptophyta (which includes coccolithophores) and Stramenopiles (which includes both diatoms and chrysophyte algae).

Our data suggest that G. bulloides may also utilise bacteria as a significant food source. G. bulloides contained 33.7–62.5 % of non-Synechococcus bacterial sequences within the cell (BUL37 was an outlier with only 10 %, Fig. 4), corresponding to a diverse assemblage of 200 OTUs. We assume that these sequences are derived from prey species because no intact bacteria lacking the carboxysomes and thylakoid membranes found in Synechococcus were observed in TEM images of the G. bulloides cytoplasm or non-digestive vacuoles. The most abundant group of sequences recovered (15–31 %, outlier BUL37 contained 2.6 %) comprise 81 OTUs belonging to the class Alphaproteobacteria (including those OTUs of the family Bradyrhizobiaceae that were not excluded as contaminants), perhaps indicating a preferential selection of specific members within this class. The remaining 17–47.5 % (outlier BUL37, 7.5 %) of sequences were made up of a diverse collection representing other major phyla of bacteria (Sect. 3.3; Fig. 4), including the Acidobacteria, Actinobacteria, Bacteroidetes, Firmicutes and Planctomycetes and the classes Beta- and Gammaproteobacteria of the phylum Proteobacteria.

G. bulloides type IId is found throughout the year in the Southern California Bight, where it is exposed to cool upwelling periods of high productivity and also to warmer periods characterised by more stratified, less productive conditions (Darling et al., 2003; Darling and Wade, 2008). It has a relatively high growth rate, possibly reproducing within 2–3 weeks (Spero and Lea, 1996; Lombard et al., 2009). In combination with our data, this suggests that G. bulloides type IId is a generalist predator with an opportunistic feeding strategy, utilising bacterioplankton as well as phyto- and zooplankton, the proportions of which may be seasonal and depth-dependent. Such opportunism may enable G. bulloides to grow and reproduce rapidly within its diverse habitat. We propose two hypotheses for the life strategy of G. bulloides type IId to survive the challenges presented within the broad seasonal changes in the region. The first is that it is a mixotrophic feeder (Mitra et al., 2016) and that the Synechococcus endobionts are photosynthesising symbionts contributing fixed carbon to the foraminiferal host. In this scenario, they would be fulfilling a functional role similar to that of algal symbionts in other spinose species, and in particular could provide additional resources during the maximum growth phase of the shallow-dwelling juveniles. Alternatively, or concurrently, Synechococcus may be exploited by G. bulloides type IId for its nutrient assimilation and storage capacity and then digested as an extra energy, nitrogen and phosphate source for DNA replication at reproduction.

Since G. bulloides occurs in great abundance in cool, high latitudes and mid- to lower-latitude upwelling systems (Kleijne et al., 1989; Naidu and Malmgren, 1996), it is one of the most commonly used planktonic foraminifera for palaeoclimate reconstruction (Sautter and Thunell, 1991; Spero and Lea 1996). In order to reconstruct past changes in oceanic conditions using the shell geochemical data, it is important to obtain a thorough understanding of the relationship between foraminiferal ecology and the geochemistry of its shell. This relationship is based on the assumption that each foraminiferal morphospecies represents a genetically continuous species with a unique habitat preference. However, since G. bulloides inhabits such a wide range of different ecosystems, it is not surprising that several ecologically distinct genotypes have been recognised (Darling et al., 1999; Kucera and Darling, 2002; Darling and Wade, 2008; Seears et al., 2012; Morard et al., 2013). Indeed, recent species delineation studies support species status for several of the G. bulloides genotypes (André et al., 2014), including G. bulloides type IId. Such diversity could result in genotype-specific geochemical signatures across the morphospecies (Healy-Williams, 1985; Bijma et al., 1998; de Vargas, 2001; Kucera and Kennet, 2002; Sadekov et al., 2016). Both Kucera and Darling (2002) and Morard et al. (2013) have demonstrated that, based on ecological knowledge, integrating G. bulloides genotypes into assemblage-based SST reconstructions significantly improves resolution. This demonstrates the value in understanding the ecology of genotypes within a morphospecies and the necessity of establishing whether the association between G. bulloides type IId and Synechococcus is universal across this foraminiferal morphospecies complex.

The discovery of intracellular bacteria within a palaeoceanographically significant foraminiferal host may lead to a significant improvement in our current understanding of foraminiferal shell geochemistry. The carbon isotopic composition of planktonic foraminifera has the potential to help reconstruct changes in the chemocline of the surface ocean, providing insights into changes in ocean circulation (Spero et al., 2003). However, the interpretation of δ13C data is often complicated by poor understanding of the causes of offsets between shell δ13C and the δ13C of the dissolved inorganic carbon from which foraminifera build their shells. In particular, the δ13C of G. bulloides shells deviates from predicted values more than that of any other extant species (Deuser et al., 1981; Kahn and Williams, 1981; Curry and Matthews, 1981; Kroon and Darling, 1995; Spero and Lea, 1996; Bijma et al., 1999), implying consistent use of metabolic carbon during calcification by this morphospecies (Deuser et al., 1981).

Symbiont photosynthesis as well as symbiont and host respiration alters the chemical microenvironment surrounding the host shell, which in turn influences their shell geochemical signatures (Rink et al., 1998; Wolf-Gladrow et al., 1999; Eggins et al., 2004). In this G. bulloides/Synechococcus association, respiration of both endobiont and host would contribute 13C-depleted CO2 to the calcifying microenvironment (Spero and Lea, 1996), whilst Synechococcus photosynthesis would counteract this by preferentially removing 12CO2 and hence elevating 13C / 12C ratios in the remaining dissolved CO2, as occurs in protist algal symbiont-bearing planktonic foraminifera (Spero et al., 1997). The large offset towards 13C-depleted values measured in G. bulloides suggests that Synechococcus respiration dominates the shell geochemical signature, and implies that photosynthesis is not the primary role of Synechococcus in this association (Sect. 4.2). The presence of metabolically active Synechococcus in G. bulloides type IId therefore may account for the unusual shell δ13C determined via culture-based studies conducted at Santa Catalina Island. Since G. bulloides type IId is abundant here, it is unlikely that variation in genotype has contributed to uncertainties in these calibrations, but applying these culture-based calibrations to other regions in which different genotypes dominate may produce erroneous results (Darling et al., 2003). It is therefore of particular importance to determine whether the Synechococcus/G. bulloides association exists in other G. bulloides genotypes in order to generate and apply genotype-specific palaeoclimate calibrations.