Proteome-Based Clustering Approaches Reveal Phylogenetic Insights into Amphistegina

0. INTRODUCTION

Foraminifera are a highly diverse group of unicellular protists that have a long evolutionary history. They represent key bioindicators in paleontological and environmental studies, but may be hard to differentiate and identify correctly due to their small size and high phenotypic plasticity (Narayan et al., 2022; Macher et al., 2021; Prazeres et al., 2020a; Bhatt and Trivedi, 2018; Hallock et al., 2003). Reconstructions of their phylogenetic and evolutionary schemes are of great importance for paleogeographic applications. Yet, as for decades these were solely based on diagnostic characters of their morphology, modern and fossil foraminiferal taxa have an extensive history of taxonomic revisions (Pawlowski and Holzmann, 2008; Haynes, 1992; Loeblich and Tappan, 1988; Loeblich et al., 1957). The application of molecular tools to identify recent species and resolve their phylogenetic relationship has therefore in recent years become increasingly prominent and often revealed a high genetic diversity at the cryptic level (Bhatt and Trivedi, 2018; Pawlowski et al., 2014; Pawlowski and Lecroq, 2010; Pochon et al., 2006; Flakowski et al., 2005; Pawlowski and Holzmann, 2002). However, foraminifera have fast evolving ribosomal genes (Pawlowski and Holzmann, 2002) and a wide occurrence of intragenomic variability (Weber and Pawlowski, 2014). Genetic markers do not always yield satisfying results, and recent studies tend to combine molecular with morphological approaches (Macher et al., 2021; Morard et al., 2019; Pawlowski et al., 2013). Yet, the phylogenetic relationships within one of the most prominent and abundant foraminiferal genera, Amphistegina, have not been sufficiently revisited using modern methodologies, so uncertainties still exist (Renema, 2018; Langer and Hottinger, 2000).

Amphistegina is the only extant genus of the family Amphisteginidae and belongs to the group of large benthic foraminifera (LBF), which live in symbiosis with photosynthesizing microalgae (Prazeres et al., 2021; Boudagher-Fadel, 2018; Renema, 2018; Lee, 2006; Leutenegger, 1977). It has a circum-global distribution and can be found in all (sub-)tropical oceans, and it also has one of the widest latitudinal extensions among the LBF, often dominating assemblages of shallow marine ecosystems from one to more than 100 m depth (Renema, 2019; Weinmann et al., 2013; Hallock, 1985; Hansen and Buchard, 1977). Moreover, Amphistegina spp. are currently invading into new higher latitudinal habitats such as the Mediterranean Sea (Mouanga and Langer, 2014; Langer et al., 2013; Triantaphyllou et al., 2009) and are predicted to further extend their distribution range with climate change (Guastella et al., 2019; Weinmann et al., 2013). Amphisteginids can locally occur in very high densities and thereby contribute significantly to coastal carbonate sediment production around several atolls and reef islands, especially in the western Pacific (Doo et al., 2012; Renema, 2003; Langer et al., 1997). Amphistegina species are also relatively easy to culture and have been found to bleach in response to environmental changes associated to climate change (Schmidt et al., 2011; Hallock, 1999; Hallock et al., 1993, 1986), which has on the one hand led to their increased application as bioindicators for coral reef condition and on the other hand to intense laboratory experiments using them for e.g. physiological or calcification studies (Stuhr et al., 2017; Prazeres et al., 2015; Bentov et al., 2009; Talge and Hallock, 1995; Hallock et al., 1986; ter Kuile and Erez, 1984; Hallock and Hansen, 1979; Zmiri et al., 1974). Many modern approaches are first tested and applied on these foraminifera, especially A. lobifera (Titelboim et al., 2021; Ross and Hallock, 2018; Stuhr et al., 2018a; Glas et al., 2012; Prazeres et al., 2011). Thus, they can be regarded as a model organism for modern diatom-bearing perforate hyaline LBF.

The genus Amphistegina was first described by d'Orbigny (d'Orbigny, 1826) and then Ehrenberg (Ehrenberg, 1838). Current studies suggest that at least 11 extinct relatives (Larsen, 1978) have been widely distributed since the Eocene and throughout the Miocene (Todd, 1976), and it contains a minimum of six to seven commonly described recent morphospecies (Förderer and Langer, 2019; Renema, 2018; Debenay, 2013; Murray, 2006; Langer and Hottinger, 2000; Hottinger et al., 1993; Larsen, 1978). These are generally grouped in two morpho-groups, with lineage I (Larsen, 1978), or the A. radiata- group (Renema, 2018) containing the more symmetrical forms A. radiata and A. papillosa, as well as, according to Larsen (1978), also A. gibbosa (shown in Fig. 3) and A. bicirculata. Lineage II or the A. lessonii-group has widely spaced curved septa and is therefore more asymmertrical in axial view (Renema, 2018). It contains A. lessonii, A. lobifera (both shown in Fig. 3) and, according to e.g. Renema (2018), also A. bicirculata and A. quoi, of which the latter is rarely recognized by others. A. gibbosa is not described by Renema (2018), however, briefly mentioned as "western Atlantic sister species of A. lobifera and A. lessonii". Hohenegger (2014) distinguishes between the A. lessonii kinship and the A. radiata kinship like Renema (2018), though he does not mention the existence of A. gibbosa. Besides these, also A. madagascariensis is occasionally described (Förderer and Langer, 2019; Murray, 2006; Zmiri et al., 1974), but assumed to be a variety of A. lessonii (Förderer and Langer, 2019).

Most of these extant species are widely distributed throughout the Indo-Pacific and show a depth-related niche partitioning that is related to shape trends (Hohenegger, 2004; Hallock and Hansen, 1979), but partly overlaps, resulting in locally mixed assemblages (more details on species descriptions and distributions can be found in Förderer and Langer, 2019; Renema, 2018; Debenay, 2013; Murray, 2006; Loeblich and Tappan, 1994, 1988; Hottinger et al., 1993; Larsen, 1978; Hansen and Buchard, 1977). A. gibbosa, in contrast, is generally assumed to be the only Amphistegina species present in the Atlantic, specifically the Caribbean region. This is however controversial (Hallock, 1999) and some authors identified Amphistegina found along the coast of Brazil as A. lessonii instead (Alves and Júnior, 2020; Prazeres et al., 2012, 2011). In light of their ubiquitous appearance in shallow warm-water carbonate systems it is even more surprising that species boundaries and numbers remain widely uncertain (Renema, 2018) and that therefore a reassessment of their phylogenetic and evolutionary relationships is long overdue.

Molecular studies on Amphistegina have so far only focused on genetic diversity within single species. Schmidt et al. (2016) showed a distinction between A. lobifera from Israel (Mediterranean and Red Sea) and Australia (Great Barrier Reef). This was confirmed by a detailed study of Prazeres et al. (2020b), who differentiated between six genotypes of A. lobifera that could indeed represent at least four separate species and have distinct distribution ranges structured along a longitudinal gradient. Subsequently, Ramadhani et al. (2022, under review) determined three distinct genetic types of A. lessonii across its global range, with several of these co-occurring in the centre of its distribution. This further suggests a high cryptic diversity within the genus and demands for broader investigations of these prolific foraminifera using modern approaches.

Besides RNA and DNA sequencing-based molecular approaches, different mass spectrometry (MS)-based proteomics approaches are increasingly used to give insights into the relationship between samples and organisms (Aebersold and Mann, 2003). The most common method is to first generate short peptides and determine their mass-to-charge (m/z) ratios by MS. These measured mass spectra are then compared to a given database to find the matching peptide sequences and in turn identify the corresponding proteins. However, this method fails to recognize peptides that do not have existing sequence data in the given database, or for organisms that do not have a suitable reference database at all (MacCoss, 2005). This has limited the application of proteomics methods largely to the analysis of model organisms. However, in recent years instrumental and methodological advances enabled the adaption of these methods to a wider range of samples (Heck and Neely, 2020). In cases where reference databases are unavailable, de novo peptide sequencing, i.e., direct inference of peptide sequences from tandem mass spectra, is often the only option. Because this may have some limitations, a novel synergistic approach optimizes de novo peptide sequencing by combining the outcomes of several commonly available algorithms (Blank-Landeshammer et al., 2017). Another MS-based approach is to cluster similar peptide mass spectra by similarity, using different algorithms (Rieder et al., 2017b). These commonly used clustering approaches can be further improved by the DISMS2 filter, which allows to also consider precursor mass, precursor charge or retention time. By comparing thousands of peptide spectra, without de novo annotation or reference database, the similarities among sample proteomes are calculated and clustered, which may indicate their phylogenetic proximity (Rieder et al., 2017a).

To better understand the relationship of recent Amphistegina and at the same time test the different proteomics-based approaches on species with no appropriate reference database, we applied and compared the results of de novo peptide sequencing as well as clustering with DISMS2 to three recent Amphistegina species. These were collected from different sites, partly repeatedly over several years, to test for the influence of spatial and temporal variations. LBF are holobionts, i.e., a combination of the foraminiferal host organism, their photosymbiotic algae and microbiome. As these associated organisms were included in the proteomics workflow, they are also contained in the resulting MS data. We therefore applied a DNA sequencing approach to analyse the dominant photosymbiont community, and their potential influence on the holobiont similarity clusters.

1. MATERIALS AND METHODS

1.1 Sample Collection

Samples were collected by snorkelling or SCUBA diving during different field campaigns (Stuhr et al., 2021, 2018b) following standard protocols in all locations (Red Sea (RS), Eilat, Israel; Western Indian Ocean (WIO), Zanzibar, Tanzania; Western Atlantic (WA), Florida Keys, USA), and either directly frozen at -80 ℃ or cultured at ZMT for several months (Table 1), in which case LBF were kept in sterilized seawater for three days prior to freezing in order to limit contamination by the microbiome within the culture dishes. They were then either extracted for DNA analysis in Bremen or transported on dry ice to ISAS, Dortmund, for proteomic analyses.

Table  1.  Site, depth, and year of collection of foraminifera populations analysed in this study

Species	Site	Year	Depth (m)	Collection	Abbreviation
Marginopora vertebralis	Zanzibar, WIO	2014	3	Culture	M. vertebralis WIO
Amphistegina lessonii	Eilat, RS	2016	3	Culture	A. lessonii RS16
Amphistegina lessonii	Eilat, RS	2016	3	Field	A. lessonii RS*16
Amphistegina lessonii	Zanzibar, WIO	2015	3	Culture	A. lessonii WIO15
Amphistegina lessonii	Zanzibar, WIO	2014	10	Culture	A. lessonii WIO14
Amphistegina lobifera	Eilat, RS	2016	3	Culture	A. lobifera RS16
Amphistegina lobifera	Eilat, RS	2016	3	Field	A. lobifera RS*16
Amphistegina lobifera	Zanzibar, WIO	2015	3	Culture	A. lobifera WIO15
Amphistegina gibbosa	Florida, WA	2015	18	Culture	A. gibbosa WA15
Amphistegina gibbosa	Florida, WA	2014	18	Culture	A. gibbosa WA15
Foraminifera frozen immediately after collection (field) are marked with an asterisk *.

 | Show Table

DownLoad: CSV

1.2 Proteomic Analyses

1.2.1   Sample preparation

Eight specimens per replicate were pooled and processed as described previously (Stuhr et al., 2018a; Rieder et al., 2017a). After protein digestion and quality-control, amino acid analysis was performed on all samples. Thus, sample amounts were corrected to compensate for systematic errors and to guarantee equal starting material for the subsequent LC-MS analysis.

1.2.2   LC-MS/MS analysis

Using an Ultimate 3000 nano-RSLC system coupled to an Orbitrap Fusion Lumos mass spectrometer (both Thermo Scientific), 330 ng per sample were analysed. A 100 µm × 2 cm C18 trapping column was used for preconcentration of peptides for 10 min using 0.1% TFA (v/v) with a flow rate of 20 µL/min followed by separation on a 75 µm × 50 cm C18 main column (both PepMap RSLC, Thermo Scientific) with a 90 min LC gradient ranging from 3%–35% of buffer B: 84% acetonitrile, 0.1% formic acid at a flow rate of 250 nL/min. The Orbitrap Fusion Lumos MS was operated in data-dependent acquisition mode and MS survey scans were acquired from m/z 300 to 1 500 at a resolution of 120 000 using the polysiloxane ion at m/z 445.12002 as lock mass. Isolation of precursors was performed by the quadrupole with a window of 0.4 m/z. Acquisition was performed in top speed mode, selecting the most intense signals within a cycle time of 3 s between survey scans and subjecting them to higher energy collisional dissociation with a normalized collision energy of 32% at a resolution of 15 000, taking into account a dynamic exclusion of 15 s. Automated gain control target values were set to 2 × 10⁵ for MS and 5 × 10⁴ MS/MS. Maximum injection times were 120 and 250 ms, respectively. Precursor ions with charge states of +1, > +7 or unassigned were excluded from MS/MS analysis and monoisoptic peak determination was set to 'peptide'.

1.2.3   Proteome-wide distance measure DISMS2

DISMS2 is a flexible algorithm that computes proteome-wide distances between MS/MS runs consisting of thousands of unidentified spectra, without additional information (Rieder et al., 2017a). Here, the ProteoWizard tool MSConvertGUI (Chambers et al., 2012) was used to convert Thermo RAW files into mzXML files. The open data format mzXML can be read with the R package readMzXmlData (Gibb, 2015). The washing step of the liquid chromatography was excluded. Thus, only spectra of the first hundred minutes are part of the analysis. First, distances for all these spectra were computed. Second, in each run 2 000 spectra with highest total ion signal were extracted. The preselection of spectra has already been used before (Palmblad and Deelder, 2012).

We used the same full factorial design as in the original paper (Rieder et al., 2017a) to find optimal parameter settings for all spectra. Out of in total 81 factor combinations the combination with the highest partial R squared (0.618 3) was chosen. Partial R squared is a goodness of fit measure that was calculated based on adonis, an ANOVA like approach for distance matrices. Optimized parameters correspond to the values from the original paper. In the pre-processing step no selection of highest peaks (topn = ∞) is made and binning is performed with binsize bin = 0.2. In the step that includes checking of constraints the size of retention time window is ret = 3 000 and the accepted precursor mass shift limit is prec = 10 ppm. The cosine distance (dist = d_cos) is chosen as distance measure between spectra and the cutoff for distance of spectra is cdis = 0.3.

1.2.4   De novo peptide sequencing data analysis

Conversion of generated.raw files and subsequent data analysis by de novo peptide sequencing was performed as described by Blank-Landeshammer et al. (2017). Only full de novo sequences of agreeing spectrum annotations by PEAKS, pNovo+ and Novor—fulfilling an expected false discovery rate (FDR) of 5%—were used for further analysis.

For distance calculation, first the similarity of peptide identifications was determined by calculating the Sørensen-Dice coefficient (Sørensen, 1948; Dice, 1945) for all sample pairs, where |X| and |Y| represent the number of unique peptides in the samples.

For comparison of populations, all identified peptides within the replicates were pooled and compared as described above.

All statistical analyses were performed and figures were produced with the statistical programming software R (R Core Team, 2016), Version 4.1.0. Package readMzXmlData (Palmblad and Deelder, 2012) was used for reading MS data, and packages ape (Paradis et al., 2004) and phytools (Revell, 2012) were used for plotting dendrograms. In each case, dendrograms were generated by using complete linkage hierarchical clustering and triplicates of M. vertebralis WIO14 as an outgroup. The dendrograms of two different approaches were visualized facing each other for direct comparison.

1.3 Genetic Analyses

1.3.1   Molecular analysis of photosymbionts

Symbiont DNA was extracted and prepared for 18S sequencing as previously described in Stuhr et al. (2018b). Briefly, 36 foraminifera (n = 4 per population) were crushed with a micropestle and digested in a CTAP/protein kinase K lysis buffer, and subsequently extracted in phenol-chloroform (Green and Sambrook, 2012). After ethanol washing, a 443-bp fragment of the 3´end of the photosymbionts' 18S rDNA was amplified using primers SymFS1 and 1528R (Schmidt et al., 2015) and Taq polymerase using previously described cycling conditions (Stuhr et al., 2018b). PCR products were SAP digested and Sanger sequenced (Starseq, Mainz) without cloning. The sequences were trimmed for primers and with a quality threshold of 0.01 (1% error probability) in Geneious Prime 2021.1.1. The resulting contigs were deposited at ENA (project accession number PRJEB49701) and aligned after adding sequences with close BLAST hits or from diatoms previously described or sequenced from LBFs extracts (Prazeres et al., 2021; Stuhr et al., 2021, 2018b; Barnes, 2016; Lee, 2006). Alter (Glez-Peña et al., 2010) was used to condense haplotypes with identical sequences. MEGA X (Kumar et al., 2018) was used to determine the best fitting substitution model and finally the taxonomic assignment was done by calculating maximum- likelihood phylogenetic trees in RAxML (Kozlov et al., 2019).

2. RESULTS

2.1 Proteome-Based Clustering

The comparison of dendrograms resulting from proteome-wide distance measures DISMS2 on either all or only the top 2 000 spectra (Fig. 1), illustrating distances between MS/MS runs of different LBF populations and their technical replicates, showed relatively similar overall results, grouping all samples according to species. However, if all spectra were included in the analysis (left side), the uncultured A. lessonii RS16 samples and one A. lessonii WIO15 sample grouped together with all A. lobifera, instead of with the remaining A. lessonii. In contrast, the A. lessonii WIO14 samples formed a separate cluster when using only the top 2 000 spectra for DISMS2 (right side), yet these did not merge with samples of other species. Samples also cluster more according to their collection site and year using this approach, e.g. with all A. lobifera triplicates grouping by their population and then by location. A. gibbosa formed a separate cluster in both approaches and hence clustered dominantly by population and then collection year. The average distance between technical replicates of the populations was between 40% and 60% when all spectra were considered and between 35% and 50% for the top 2 000 spectra. Nodes connecting all samples of each species were at ~80% and around 50%–70% distances, respectively.

Figure  1.  Proteome-wide distance measure clusters computed via DISMS2 with optimized parameters based on either pairwise distances of all spectra (left side) or only the top 2 000 spectra (right side) produced by 30 MS/MS runs of triplicates of different population of large benthic foraminifera, using complete linkage hierarchical clustering.

DownLoad: Full-Size Img PowerPoint

Since DISMS2 with the top 2 000 spectra showed a very coherent result regarding grouping of triplicates and populations, this was considered as the more reliable approach and further compared to the output of a de novo annotation approach. Complete linkage hierarchical clustering of de novo peptide sequences resulted in widely similar grouping (Fig. 2). One sample of A. lessonii WIO15 grouped more closely with A. lobifera than the other A. lessonii samples, which however clustered closer together without building a separate group for the population sampled in 2014. Moreover, within A. lobifera the samples of the cultured RS16 and WIO15 populations grouped first together, and then the uncultured RS16 samples. Apart from these minor differences, populations and triplicates within all LBF, including A. gibbosa and M. vertebralis, showed widely analogous clusters. In both approaches, A. lobifera and A. lessonii are clearly more similar to each other than they are to A. gibbosa. The average distance between technical replicates was between 35% and 50% for de novo peptide sequences. Nodes connecting all samples of each species were around 50%–60% distances.

Figure  2.  Proteome-wide distance measure clusters computed via de novo peptide sequences (left side) or DISMS2 on top 2 000 spectra with optimized parameters (right side) produced by 30 MS/MS runs of triplicates of different population of large benthic foraminifera, using complete linkage hierarchical clustering.

DownLoad: Full-Size Img PowerPoint

Because only one sample did not cluster with the remainder of its population and species, we regarded the de novo approach as providing the most representative results. When triplicates of each population were pooled, the resulting hierarchical cluster (Fig. 3) grouped robustly by LBF species, with A. lessonii and A. lobifera being most similar to each other, followed by A. gibbosa. Within species, no dominant grouping by collection site of year could be identified.

Figure  3.  Proteome-wide distance measure clusters computed via de novo peptide sequences produced by 30 MS/MS runs of pooled triplicates of different population of large benthic foraminifera, using complete linkage hierarchical clustering. Images on the right side show from top to bottom representative specimens of Amphistegina lessonii, A. lobifera and A. gibbosa. Marginopora vertebralis is not displayed.

DownLoad: Full-Size Img PowerPoint

2.2 Photosymbiont Barcoding

To understand whether the similarity patterns of the tested LBF reconstructed by proteomics-based approaches may reflect their different photosymbiont communities, we applied a widely used DNA barcoding approach and calculated phylogenetic a tree of the resulting ~450 bp 18S rDNA sequences. This molecular fingerprinting of all samples (n = 24, Fig. 4) revealed that all A. gibbosa (n = 5) contained the same diatom symbiont sequence, while A. lessonii (n = 11) yielded seven haplotypes and A. lobifera (n = 8) yielded five haplotypes. All these sequences fall within the order Fragilariales that cluster closely together with sequences obtained from formerly sequenced A. lobifera from Israel, Indonesia, Australia, the Maldives and Papua New Guinea (Prazeres et al., 2021; Stuhr et al., 2021), A. lessonii from Zanzibar (WIO), Palau, Papua New Guinea, Australia and Eilat (RS)(Ramadhani et al., 2022, under review; Prazeres et al., 2021; Stuhr et al., 2018b) and A. gibbosa from Florida (WA) (Stuhr et al., 2018b). In contrast, A. radiata from Papua New Guinea (Prazeres et al., 2021) host endosymbionts from Triceratiales instead, another order of diatoms. Although the obtained sequences (443 bp) are too short for a taxonomic assignment to the species level, our results indicate some variability in the diatom assemblages, especially within A. lessonii. The resulting dendrogram also showed a separation of A. lobifera endosymbiont sequences from Greece and A. gibbosa photosymbionts from the other samples, while A. lessonii and A. lobifera photosymbionts grouped weakly by collection sites. The minor division of greek A. lobifera from the remaining samples could be due to a change in their main associated pennate diatoms during the spreading of the species on its northwestward expansion in the Mediterranean (Guastella et al., 2019; Langer et al., 2013; Triantaphyllou et al., 2009) as an adjustment to varying environmental conditions or native diatom communities. While Schmidt et al. (2016) found no systematic difference in the photosymbiont community of Red Sea and Mediterranean populations, the diatom population in other areas, potentially more marginal in terms of A. lobifera's temperature range, are likely to bear a different local diatom community with which the foraminifera can associate. Moreover, it is not clear whether all A. lobifera in the Mediterranean are really Lessepsian invaders or might have actually been present in lower densities already before the opening of the Suez canal (Langer, 2008), in which case they could have adjusted their photosymbiont community to the different environmental conditions over much longer time scales.

Figure  4.  Phylogenetic assignment of diatom-symbiont sequences from Amphistegina lobifera and A. lessonii collected in Zanzibar, Tanzania, and Eilat, Israel, or A. gibbosa from the Florida Keys, USA. These new sequences are highlighted in bold and underlined. Bootstrap support above 50% is given at the respective nodes.

DownLoad: Full-Size Img PowerPoint

3. DISCUSSION

Large benthic foraminifera are among the most biodiverse and ecologically intriguing calcium-carbonate producers worldwide (Narayan et al., 2022; Förderer and Langer, 2019; Förderer et al., 2018; Doo et al., 2012; Langer et al., 1997). Acting as bioindicators, they let us characterize recent and reconstruct paleoenvironments (Prazeres et al., 2020a; Emrich et al., 2017; Pisapia et al., 2017; Narayan et al., 2015; Carnahan et al., 2009; Hallock et al., 2003; Hallock and Glenn, 1986), based on our knowledge of their physiology, geochemistry, biogeography, ecology and evolutionary biology. Furthermore, their evolutionary history and proliferation through geological time inspires research and ideas on the various modes of adaptation, extinction, and radiation during periods of environmental change (Prazeres and Renema, 2019; Hohenegger, 2014; Pomar and Hallock, 2008). Although with the development of modern methods, we are now able to reveal cryptic species and their underlying phylogenetic relationships, the actual diversity and natural history of some of the most prominent LBF remain undiscovered (Stulpinaite et al., 2020; Bhatt and Trivedi, 2018; Renema, 2018; Pawlowski and Holzmann, 2002). By combining new MS-based techniques with advanced de novo peptide sequencing or similarity clustering algorithms we here present a pilot study, delineating three Amphistegina species commonly found in association with coral reefs.

The results of both our proteomics-based approaches showed a separation of the three species. However, as DISMS2 applied to all spectra inadequately grouped the uncultured A. lessonii RS16 samples and one A. lessonii WIO15 sample with A. lobifera, this approach gives the least reliable results. This is further indicated by the highest distances between population replicates as well as nodes connecting all samples of the same species (excluding described faulty grouped samples).

Although these distances were lower when using the top 2 000 spectra only, the A. lessonii WIO14 formed a group separated from the other A. lessonii and also the A. lobifera samples. Hence, the de novo peptide sequences-based dendrogram represented the most trustworthy outcome, with all but one sample (A. lessonii WIO15.2) clustering correctly by species, and principally also the technical triplicates by population, with comparably low distances. This might be because the de novo peptide sequencing algorithms act as efficient filters to separate true (i.e., biologically relevant) spectra from background spectra and those stemming from laboratory contaminations compared to unsupervised clustering. We therefore demonstrate that the approach of Blank-Landeshammer et al. (2017) is capable of comprehensively delineating LBF taxa without a reference database. Yet, as multiple algorithms can be easily applied to the same MS/MS dataset, a combination with DISMS2 (Rieder et al., 2017a) applied to top spectra may serve to further confirm the outcome and boost reliability.

The resultant dendrograms provide first hints into the phylogeny of the three studied Amphistegina species. Clearly, the Indopacific species A. lobifera and A. lessonii are more closely related to each other than they are to their counterpart A. gibbosa in the Atlantic (Figs. 1 to 3). This confirms that the morphologically relatively similar and therefore sometimes confused species A. lessonii and A. gibbosa are indeed separate species, and agrees with their phylogenetic separation suggested by Larsen (1978) into relatively widely separated lineages. This pattern is reflected in the photosymbiotic diatoms harbored by these populations (Fig. 4). Hence, the differences detected for the LBF holobionts may be influenced and partly based on the differences of the photosymbiont communities. Overall, as previously recognized, A. lessonii and A. lobifera contain a more diverse community of Fragilariales (Ramadhani et al., 2022, under review; Prazeres et al., 2021; Stuhr et al., 2021, 2018b; Barnes, 2016; Schmidt et al., 2016), while A. gibbosa contains only one strain of a Fragilariales species (Stuhr et al., 2018b; Barnes, 2016), fairly distinct from those harboured by the aforementioned species. Interestingly, also A. radiata appears to have a higher specificity in its photosymbiont assemblages, which is a member of Triceratiales instead of Fragilariales (Prazeres et al., 2021; Barnes, 2016). Whiles this may be an adaptation to their usually more light-reduced habitat (Prazeres et al., 2021), the variation in symbiont flexibility could likewise reflect their phylogenetic distinction into the two lineages suggested by Larsen (1978), with more restricted associations in lineage I (A. radiata-A. gibbosa) and highly flexible associations in lineage II (A. lessonii-A. lobifera). In this regard, it would be highly interesting to examine the photosymbiont communities of A. papillosa, which is generally assumed to belong to lineage I or the A. radiata group/kinship, to see whether it also shows a relatively high symbiont specificity.

Despite a potential effect of the global symbiont community on the proteome-based similarity clusters, the well-defined separation of morpho-species signifies the LBF host as the major driver. If the symbionts had an overriding impact, A. lobifera and A. lessonii as well as samples with morpho-species should cluster more mixed, as specimens from different collection sites or years showed similar symbiont sequences. In contrast, recent environmental conditions appear to have a detectable influence. For example, A. lobifera and A. lessonii from the Red Sea that were cultured prior to sampling (RS16) are more similar to their likewise cultured counterparts from the western Indian Ocean (WIO15) than they are to subsamples of the same Red Sea population sampled immediately without culturing (RS16*). Since the proteome is largely influenced by environmental factors and may thereby influence evolutionary processes itself (Baer and Millar, 2016) it appears reasonable that populations cultivated under alike conditions will have more similar proteomes. Conversely, A. lessonii WIO14 was in all dendrograms slightly disjointed from the remaining A. lessonii populations. Notably, it is the only A. lessonii population that was sampled in at least 10 m depth, while all others originated from 2–3 m depth. Whether this distinction in A. lessonii from Zanzibar may thus represent a genotypic differentiation related to their original habitat, or if this population was somehow differently affected by culturing, will have to be clarified in future studies. Conditions such as the microbiome or phenological phases can potentially vary between cultures, even under similar light, temperature, and water geochemistry, especially if these have been developing over long time.

4. CONCLUSIONS

Our proteome-based clustering approaches along with phylogenetic analyses of their photosymbiont communities indicate that especially de novo peptide sequencing provides a reliable tool to reconstruct the evolutionary relationship among foraminiferal holobionts. Although the wide separation of A. gibbosa from A. lessonii and A. lobifera along with their distinctive type of Fragilariales symbionts may indicate a partial influence of the photosymbiont communities, the well-defined species separation in between A. lobifera and A. lessonii contradicts their highly widely overlapping symbiont communities. This difference between proteome-based clusters and the phylogenetic dendrogram of the photosymbionts therefore signifies that the symbiont community cannot be the major clustering driver. This further confirms that proteomics approaches allow us to draw phylogenetic information, which here reproduces the separation in two lineages as suggested by Larsen (1978). This study provides a basis for future studies that should include the remaining Amphistegina morpho-species and include populations from their wide spatial distribution, across depth and their entire geographic range, but should moreover also be extended to other genera of foraminifera. Shedding light onto the true (cryptic) diversity and unravelling the evolutionary history of the prominent genus Amphistegina will help ecologists and palaeontologists alike towards understanding paradigms in the face of changing environmental conditions, past and future adaptions, or potential radiation events.