A combined bioinformatics and LC-MS-based approach for the development and benchmarking of a comprehensive database of Lymnaea CNS proteins

Protein networks perform key functions in all living organisms. The study of such complex biological functions has led to the development of liquid chromatography–mass spectrometry (LC-MS)-based platforms that enable researchers to perform quantitative system-wide analysis of proteomes, including protein–protein interactions, post-translational modifications and spatial localization of proteins, even at the single cell level. Furthermore, several labelling and pre-fractionation techniques allow researchers to improve the detection of low abundance and possibly functionally relevant proteins.

The great pond snail, Lymnaea stagnalis, is used as a model organism in a wide range of biological research fields, such as the study of host–parasite interactions, ecotoxicology, evolution, developmental biology, learning and memory, ageing and age-related memory decline, genome editing, ‘-omics’ and human disease modelling (Benjamin et al., 2021; Benjamin and Kemenes, 2020; Fodor et al., 2020a; Rivi et al., 2020).

In our laboratory, we are interested in elucidating evolutionarily conserved molecular mechanisms involved in the formation of long-term memory after classical conditioning and memory impairment associated with ageing as well as amyloid-β-induced memory decline. In memory consolidation and underlying synaptic plasticity, we and others have already established important roles for a variety of key enzyme proteins, such as nitric oxide synthase (NOS) (Kemenes et al., 2002), mitogen-activated protein kinase (MAPK) (Ribeiro et al., 2005), protein kinase A (PKA) (Kemenes et al., 2006), Ca²⁺/calmodulin-dependent protein kinase II (CaMKII) (Naskar et al., 2014) and the transcription factors cAMP-response element binding protein 1 (CREB1) (Ribeiro et al., 2003; Sadamoto et al., 2004) and CCAAT enhancer binding protein (C/EBP) (Hatakeyama et al., 2006). In age-related memory decline in L. stagnalis, we have found several impaired signalling pathways, including pituitary adenylate cyclase activating polypeptide (PACAP) and insulin-like growth factor-1 (IGF-1); when these are restored by administering exogenous PACAP or IGF-1, the age-related learning deficiency is reversed (Pirger et al., 2014). We have already established L. stagnalis as a useful invertebrate model for amyloid-β-associated memory impairment. Specifically, we found that amyloid-β peptides 1–42 and 25–35 were able to inhibit long-term memory recall in L. stagnalis (Ford et al., 2017; Ford et al., 2015). Recently, the sequencing of the whole transcriptome of the central nervous system (CNS) of L. stagnalis identified several evolutionarily conserved sequences of genes involved in human ageing, and age-related and amyloid-β-induced memory loss, including gelsolin, presenilin, huntingin, Parkinson disease protein 7 (PARK-7/DJ-1) and amyloid precursor protein (Fodor et al., 2021; Fodor et al., 2020b). However, the lack of comprehensive proteomics information in L. stagnalis has hindered further progress with research aimed at understanding the roles of these proteins in the context of large-scale protein networks in the CNS.

Although quantitative reverse transcription–polymerase chain reaction (qRT-PCR) is a well-established method to quantify specific gene transcripts of interest, it does not allow the quantification and detection of the function of a specific protein in its protein network because of the possibility of post-translational modifications, e.g. phosphorylation. To enable us to provide these types of important information, in the present study we performed large-scale proteomics experiments using Nanoscale liquid chromatography–mass spectrometry (nanoLC-MS) to analyse protein expression and post-translational modification in the CNS of L. stagnalis.

One important prerequisite for a successful proteomic workflow and the accurate identification of proteins is the existence of a protein sequence database of the organism of interest that can be used for comparison of the peptide sequences acquired from tandem mass spectrometry (MS/MS) fragmentation spectra with protein sequences in the database of the same organism.

In the Universal Protein Knowledgebase (UniProtKB) (UniProt, 2019), the only available and useful protein database for L. stagnalis consists of 519 proteins, of which 48 have been reviewed but the remaining 471 have not (Uniprot.org – last modified on 3 August 2020). Because of this lack of specific protein sequence information, previous proteomics analysis of L. stagnalis was performed by utilizing the entire UniProtKB/Swiss-Prot database with protein sequences from all available organisms (Rosenegger et al., 2010; Silverman-Gavrila et al., 2011), or using a Metazoa-specific database (Giusti et al., 2013) to identify proteins.

To prepare a more comprehensive and representative database for CNS proteins of L. stagnalis, which will enable us to identify proteins accurately by liquid chromatography with tandem mass spectrometry (LC-MS/MS) analysis, we selected the transcriptome dataset of Sadamoto et al. (2012) from the available transcriptome datasets (Bouetard et al., 2012; Davison and Blaxter, 2005; Dong et al., 2021; Feng et al., 2009; Sadamoto et al., 2004; Sadamoto et al., 2012) published in the NCBI database because it contains the transcriptome of the whole L. stagnalis CNS including the buccal ‘learning’ ganglia and the central cerebral ring.

The transcripts from the Sadamoto et al. (2012) dataset (NCBI accession number PRJDB98) were filtered for coding regions and the remaining transcripts were then searched for homology against all UniProtKB molluscan entries. The resulting Lymnaea CNS protein database, LymCNS-PDB, was then utilized in a proof of principle experiment by performing LC-MS-based proteomics analysis with proteins isolated from the CNS of L. stagnalis. Furthermore, we prepared a database (DB) with all proteins from the LymCNS-PDB using matching amino acid sequences of the other molluscan species from UniProtKB. As the two databases were the same size, we were able to compare the number of identifications from a L. stagnalis-specific database with those in a non-specific molluscan database.

Specimens of Lymnaea stagnalis (Linnaeus 1758) were raised in the breeding facility of the University of Sussex, where they were kept in 20–22°C copper-free water under a 12 h light and dark cycle. They were fed on lettuce 3 times and a vegetable-based fish food twice a week.

Whole CNS samples from 90 snails aged 3 months and weighing approximately 1.5 g were prepared as follows. The shell of the snail was cut, and the body carefully removed and pinned to a Sylgard-coated dish containing Hepes-buffered saline, then dissected under a stereomicroscope (E-Zoom6, Edmund Optics, Barrington, NJ, USA). The CNS was accessed by an incision in the dorsal body region isolated from the buccal mass by the severing of all the peripheral nerves, and then immediately placed in Eppendorf tubes on dry ice. Three tubes, each containing 30 CNS, were stored at −80°C.

Frozen CNS samples were thawed for 30 min at room temperature (RT; 20–22°C) before adding lysis buffer (6 mol l⁻¹ urea, 2 mol l⁻¹ thiourea in 10 mmol l⁻¹ Hepes pH 8.0) and 30 ceramic beads (1.4 mm zirconium oxide beads) for homogenization in a Precellys 24 Homogeniser (Bertin Instruments) using two cycles for 30 s at 6800 rpm with a 60 s break in between, followed by a centrifugation for 1 h at 14,000 g to remove the debris. The supernatant was transferred to a fresh tube and the extracted proteins were reduced for 30 min in 10 mmol l⁻¹ dithiothreitol and alkylated for 30 min in 55 mmol l⁻¹ iodoacetamide (in the dark). Proteins were first pre-digested with LysC for at least 3 h at RT and after dilution with 3 volumes of 50 mmol l⁻¹ ammonium bicarbonate buffer, the main digestion was performed with trypsin overnight at RT. Peptide samples (in triplicate) were desalted with C18 Hypersep cartridges (Thermo Fisher) and eluates were concentrated in a SpeedVac concentrator (Savant) and prefractionated into 6 fractions using the immobilized pH gradient (IPG) strip-based peptide fractionation method as previously described (Eravci et al., 2014).

The 6 desalted peptide fractions were analysed in triplicate (18 samples in total) by a reversed-phase capillary nano liquid chromatography system (Ultimate 3000, Thermo Scientific) connected to a Q Exactive HF mass spectrometer (Thermo Scientific). Samples were injected and concentrated on a trap column (PepMap100 C18, 3 µm, 100 Å, 75 µm i.d.×2 cm, Thermo Scientific) equilibrated with 0.05% trifluoroacetic acid in water. After switching the trap column inline, LC separations were performed on a capillary column (Acclaim PepMap100 C18, 2 µm, 100 Å, 75 µm i.d.×25 cm, Thermo Scientific) at an eluent flow rate of 300 nl min⁻¹. Mobile phase A contained 0.1% formic acid in water, and mobile phase B contained 0.1% formic acid in 80% acetonitrile, 20% water. The column was pre-equilibrated with 5% mobile phase B and peptides were separated using a gradient of 5–44% mobile phase B within 100 min. Mass spectra were acquired in a data-dependent mode utilizing a single MS survey scan (m/z 350–1650) with a resolution of 60,000 in the Orbitrap, and MS/MS scans of the 15 most intense precursor ions with a resolution of 15,000. HCD fragmentation was performed for all ions with charge states of 2+ to 5+, normalized collision energy of 27 and an isolation window of 1.4 m/z. The dynamic exclusion time was set to 20 s. Automatic gain control (AGC) was set to 3×106 for MS scans using a maximum injection time of 20 ms. For MS2 scans, the AGC target was set to 1×105 with a maximum injection time of 25 ms.

MS and MS/MS raw data were analysed using the MaxQuant software package (version 1.6.12.0) with an implemented Andromeda peptide search engine (Tyanova et al., 2016). Data were searched against the FASTA formatted protein database of L. stagnalis described here.

To predict coding regions in the NCBI PRJDB98 transcriptome dataset, we analysed all available transcript assemblies using the Trinity TransDecoder (version 5.5.0) bioinformatics tool. Prior to this, we had downloaded all available Swiss-Prot and TrEMBL amino acid sequences for molluscan proteins from UniProtKB and characterized them by whether they were sequences for L. stagnalis, other euthyneura species or other molluscan species.

All sequences were compared with the transdecoded database from the NCBI PRJDB98 transcriptome dataset mentioned above, in order to find potential matches using the blast-style program Many-against-Many sequence searching (MMSeqs2) (Steinegger and Soding, 2017). In each case, proteins were matched by preference, first against L. stagnalis, then other euthyneura and, finally, against other molluscan proteins. Duplicate proteins as well as matches to proteins termed ‘uncharacterized’ or ‘hypothetical’ were removed from the final dataset.

For all matching entries, the header of the TrEMBL/Swiss-Prot protein was used to update the headings of the respective uncharacterized L. stagnalis amino acid sequences from the transdecoded PRJDB98 transcriptome dataset, indicating the original GenBank accession number, the UniProt ID and the protein name of the matching molluscan protein, and the P-value for matching accuracy provided by the MMSeqs2. The constructed database LymCNS-PDB is saved in a ‘FASTA’ format following the UniProtKB parsing rules for FASTA headers (see Fig. 1).

To compare our database with the non-specific database, we prepared a molluscan database with all proteins identified from our LymCNS-PDB database using the amino acid sequences of the matching molluscan species from UniProtKB. To allow optimal comparison, the two databases are the same size and contain the same number of proteins.

The NCBI PRJDB98 transcriptome dataset (Sadamoto et al., 2012), which contains 116,265 transcripts, was analysed with a Trinity TransDecoder (version 5.5.0) to predict coding regions in the present assemblies; 22,180 ‘transdecoded’ transcripts were then used for homology searches with MMSeqs2 against 211,200 protein entries from the UniProtKB database with the preference for Lymnaea stagnalis>Euthyneura>Mollusca. We were able to match 16,142 transcripts to 9628 proteins from the UniProtKB database (Fig. 2).

Three-quarters of all matching entries were from organisms with a large number of available protein sequences in UniProtKB: Mizuhopecten yessoensis (MIZYE; Yesso scallop with 22,614 protein entries matching to 4781 PRJDB98 transcripts of 3469 unique proteins; Crassostrea gigas (CRAGI; Pacific oyster) with 27,077 protein entries matching to 2672 PRJDB98 transcripts of 1997 unique proteins and Biomphalaria glabrata (BIOGL), a species of pulmonate freshwater snail, with 31,775 protein entries matching to 3183 PRJDB98 transcripts of 1729 unique proteins.

A rather low number of matching entries were from the following molluscan organisms: a mixture of several Eupulmonata (9EUPU), with the majority of protein entries (75%) from Arion vulgaris within this taxonomic clade of air-breathing snails, with 65,368 protein entries matching to 1072 PRJDB98 transcripts of 614 unique proteins; Elysia chlorotica (ELYCH; eastern emerald elysia) with 23,887 protein entries matching to 928 PRJDB98 transcripts of 525 unique proteins; Lymnaea stagnalis (LYMST; great pond snail) with 442 protein entries matching to 1245 PRJDB98 transcripts of 391 unique proteins; Aplysia californica (APLCA; California sea hare) with 443 protein entries matching to 755 PRJDB98 transcripts of 136 unique proteins; Pomacea canaliculata (POMCA; golden apple snail) with 21,514 protein entries matching to 144 PRJDB98 transcripts of 110 unique proteins; and the remaining 237,428 molluscan protein entries matching to 1029 PRJDB98 transcripts of 657 unique proteins (Fig. 2 and Table 1).

To demonstrate the suitability of our newly created protein database (LymCNS-PDB), we performed a MS-based proteomics experiment of the CNS from L. stagnalis, which resulted in the identification of 3810 unique proteins. The identified proteins were derived from the following molluscan organisms with significant matching homology to the corresponding amino acid sequences of the L. stagnalis transcripts from the PRJDB98 dataset: Mizuhopecten yessoensis (MIZYE) with 1478 unique proteins; Crassostrea gigas (CRAGI) with 775; Biomphalaria glabrata (BIOGL) with 590; Eupulmonata (9EUPU) with 275; Elysia chlorotica (ELYCH) with 177; Lymnaea stagnalis (LYMST) with 121; Aplysia californica (APLCA) with 82; Pomacea canaliculata (POMCA) with 32; and other molluscan organisms with 280 (Fig. 3 and Table 1). In contrast, when the same experimental data were analysed against the non-specific molluscan database, only 920 unique proteins were identified with 44 proteins from Mizuhopecten yessoensis, 121 proteins from Crassostrea gigas, 341 proteins from Biomphalaria glabrata, 104 proteins from Eupulmonata, 52 proteins from Elysia chlorotica, 121 proteins from Lymnaea stagnalis, 41 proteins from Aplysia californica, 3 proteins from Pomacea canaliculate and 34 proteins from other molluscan species.

To provide a functional categorization of all proteins included in our LymCNS-PDB, protein sequences were further annotated and classified based on EuKaryotic Orthologous Groups (KOG) categories by using RPSBLAST 2.2.15 on NCBI KOG 2/2/2011 database (http://weizhong-lab.ucsd.edu/webMGA/server/kog/). Fig. 4 shows the distribution of KOG annotations evaluated on the proteins in the LymCNS-PDB and Table S1 provides more detailed information on this distribution. The similarity of this KOG annotation pattern to the distribution in the comparison of KOG annotations of protein-coding transcripts expressed in the CNS of key vertebrate and invertebrate neuroscience model organisms presented by Dong et al. (2021) indicates that all those functional categories are also present in our LymCNS-PDB.

The pond snail L. stagnalis is an invertebrate model organism used highly successfully in both basic and translational neuroscience research aimed at understanding the neural and circuit mechanisms of a variety of behaviours because of its numerically simple and well-characterized CNS (Benjamin, 2008; Fodor et al., 2020a; Rivi et al., 2020). However, the molecular characterization of the different functions of the CNS has been limited by the lack of a comprehensive proteomics database. In this investigation, we set out to develop an extensive proteomics database based on the mRNA transcript assemblies from the NCBI Bioproject PRJDB98 obtained by de novo sequencing and transcriptome analysis (Sadamoto et al., 2012).

Although a recently published paper by Dong et al. (2021) presented a more comprehensive L. stagnalis transcriptomic database, it only included RNA transcripts from the ring without the buccal ganglia, and the identification of protein sequences was focused mainly on different ion channels (Dong et al., 2021). In our approach, we developed a more general proteomics database that includes proteins involved in several CNS functions such as learning, fundamental decision making and feeding-related motivational states, among others. As previous published studies have identified the buccal ganglia as the location of circuitry involved in the expression of both appetitive and aversive memories (Ito et al., 2012; Marra et al., 2010), encoding hunger states (Crossley et al., 2016; Staras et al., 2003) as well as fundamental decision making (Crossley et al., 2018), we developed a proteomics database created from a transcriptomic database that included RNA transcripts from the buccal ganglia as well as the ring ganglia. For this reason, we used the L. stagnalis transcriptomics database published by Sadamoto et al. (2012) as this database contains the transcriptome from the whole CNS, not just the ring ganglia (Sadamoto et al., 2012).

By matching all mRNA transcript assemblies from the NCBI Bioproject PRJDB98 to all available molluscan proteins on the UniProtKB database, we succeeded in creating the proteomics database LymCNS-PDB with 9628 proteins, containing the translated amino acid sequences of their respective mRNAs from the L. stagnalis CNS, as well as obtaining all the other information (e.g. protein name) from their matching molluscan counterparts from UniProtKB. Most of the matches to certain molluscan species were due to the large number of available protein sequences of this organism in UniProtKB. Species that have a smaller number of identified proteins have a greater match compared with those with a larger number of identified proteins. For example, Mizuhopecten yessoensis with 22,614, Crassostrea gigas with 27,077 and Biomphalaria glabrata with 31,775 protein entries in UniProtKB were matched to 3469, 1997 and 1729 PRJDB98 sequences, respectively. These matches represent 75% of all the matching proteins (Fig. 2), even though two of these three organisms, M. yessoensis and C. gigas, have the most distant phylogenetic relationship to L. stagnalis amongst all of the matching organisms (Fig. 5). In contrast, the group with a very high number of UniProtKB entries (65,368), Eupulmonata (with approximately 49,000 UniProtKB entries from Arion vulgaris) only matched to 614 PRJDB98 sequences, even though it has a much closer phylogenetic relationship to L. stagnalis (Fig. 5).

The number of protein identifications was increased by performing a pre-fractionation of the tryptic peptides from the CNS of L. stagnalis using IPG gels before analysis of the fractions by nanoLC-MS (Eravci et al., 2014).

Analysis of the MS/MS fragmentation spectra using the LymCNS-PDB led to the identification of 3810 unique proteins, representing almost 40% of the entire LymCNS-PDB database. To our knowledge, this is the highest number of protein identifications in a proteomics experiment using L. stagnalis.

The proportion of different organisms identified by MS analysis using the LymCNS-PDB (Fig. 3) shows the same distribution as that for the organisms identified from all molluscan entries in the NCBI Bioproject PRJDB98 assemblies (Fig. 2). The similarity of the two distributions indicates that the LymCNS-PDB can be successfully used for the identification of proteins with extracted samples from L. stagnalis CNS without any bias towards one of the matched molluscan organisms.

To compare our database with amino acid sequences specific for L. stagnalis with other studies that were using a non-specific database for the identification of proteins from this species (Giusti et al., 2013; Rosenegger et al., 2010; Silverman-Gavrila et al., 2011), we prepared a molluscan database with all proteins from our LymCNS-PDB, using the amino acid sequences of their matching counterparts from other molluscs instead of the L. stagnalis sequences derived from the NCBI PRJDB98 transcriptome dataset. In comparison to the 3810 unique proteins identified using the LymCNS-PDB the unspecific molluscan database led to the identification of only 920 unique proteins, which presumably contain orthologous sequences in evolutionarily conserved regions, homologues to the sequences of certain tryptic peptides from the L. stagnalis CNS.

This comparison clearly shows the benefits of using a proteomics database with specific amino acid sequences for the organism under investigation, as using the LymCNS-PDB for the identification of proteins from the L. stagnalis CNS resulted in a 4 times higher number of identified proteins compared with using the non-specific molluscan database.

We have generated the most extensive proteomics database to date that is specifically for proteins from the CNS of L. stagnalis, the LymCNS-PDB. We have successfully used this database, as a proof of principle, to identify proteins in the isolated L. stagnalis CNS. Recently, we also successfully used it to reveal quantitative protein expression differences between CNS preparations made from classically conditioned and control animals (Anagnostopoulou et al., 2021b), food-deprived and satiated animals (Eravci et al., 2021), and young and aged animals (Anagnostopoulou et al., 2021a).

The LymCNS-PDB database provides a valuable tool to open new avenues for future research on proteomics to identify and quantify a plethora of proteins, which are involved in the molecular mechanisms of different neurobiological functions in the CNS of L. stagnalis, including learning and memory formation, ageing, age-related memory impairment as well amyloid-β-induced memory decline, feeding patterns, defensive responses and neuro-hormonal behavioural circuits involved in reproduction.

The authors would like to thank Dr Paul Johnston from the Freie Universität Berlin for analysis of the NCBI PRJDB98 transcriptome dataset with Trinity TransDecoder and providing the transcoded dataset. For mass spectrometry, we acknowledge the assistance of the Core Facility BioSupraMol supported by the Deutsche Forschungsgemeinschaft (DFG).