RNAlater facilitates remote sampling of aquaculture Atlantic salmon liver for proteomic analysis

Abstract

During sample collection and transport, high‐quality nucleic acids, proteins and metabolites are preserved by snap‐freezing (SF) samples in liquid nitrogen or dry ice. In remote aquaculture facilities, SF materials are not readily available and can be hazardous during sampling and transport. RNAlater is widely used in aquaculture for transcriptome studies, but its effect on proteome stability needs further investigation. Here, we used proteomics to demonstrate that RNAlater (RL) preserved liver samples similarly to SF samples. Additionally, 69 proteins (∼2.6%) comprising ribosomal proteins, transcription cofactors, translational factors, proteases and transmembrane proteins were significantly less abundant in the RL proteome. These proteins are expected to be denatured by RL and are cellular components of ribosomes and nucleosomes and involved in proteolysis, transcription and translation. We also demonstrated that RL did not influence abundance of important proteins originally identified in liver samples from salmon reared under heat stress, hypoxia or specific feeding regimes. These findings suggest that preserving samples in RL is suitable for proteomics and offer a safer and effective preservation technology for use in remote aquaculture locations.

INTRODUCTION

For ‘-omics’ studies, immersing tissue samples in liquid nitrogen (snap-freezing [SF]) combined with storage at −80°C is preferred as it preserves nucleic acids (DNA and RNA) and stabilises proteins close to their native state (Abbaraju et al., 2011; Bae et al., 2019; Bennike et al., 2016; Hickl et al., 2019; Kruse et al., 2017). However, access to liquid nitrogen or dry ice is not always practical in field experiments or remote areas. Aquaculture farms are often located in remote areas where access is by unpaved roads or boat, when farming occurs in the ocean. Collection and transport of aquaculture samples for ‘-omics’ purposes often require a high degree of logistics focused on sourcing and handling either liquid nitrogen or dry ice to reduce the potential for accidents or injury. Express courier services and shipment of frozen samples in dry ice incur higher costs and are associated with more logistical difficulties and shipment regulations.

RNAlater (RL) is used to precipitate RNases while preserving cellular RNA integrity (Allewell & Sama, 1974; Michaud & Foran, 2011; van Eijsden et al., 2013). In aquaculture, RL is widely used in transcriptomic studies to uncover candidate genes relevant for reproduction, growth and health (Chandhini & Rejish Kumar, 2019). Although proteomics has been increasingly used in aquaculture research (Nissa et al., 2021), the suitability of RL as a preserving agent of aquaculture samples intended for proteomics analysis needs further exploration.

Here, high-resolution mass spectrometry (MS) was used to compare the proteomes of Atlantic salmon liver samples stored in liquid nitrogen or RL to determine whether RL is a reliable and effective alternative to SF to mitigate the hazards of travelling and sampling with liquid nitrogen or dry ice.

MATERIALS AND METHODS

Atlantic salmons (Salmo salar) were reared in 300 L tanks in a recirculating freshwater system (16 ± 0.5°C) at Bribie Island Research Centre (Queensland, Australia). Fish were hand-fed to satiation twice daily under a 12L:12D photoperiod water. Ten randomly selected salmons (50.3 ± 1.7 g) were euthanized using anaesthetic Aqui-S (20 mg L⁻¹) (New Zealand Ltd). Liver tissue (1 cm³) from five fish was stored by SF in liquid nitrogen (SF) or RL (Sigma-Aldrich). SF samples were stored at −80°C immediately. To replicate field conditions, RL samples were kept at room temperature for 24 h before −80°C storage. SF_FD samples (n = 5) were created by freeze-drying some SF samples for 48 h. Procedures were approved by CSIRO Animal Ethics Committee (CQAEC permit 2018-44).

Protein preparation for sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS–PAGE) and mass spectrometry (MS) analysis

Proteins were extracted in 500 μl of buffer (100 mM dithiothreitol; 4% sodium dodecyl sulphate; 100 mM Tris-HCl pH 7.6) from 100 mg of disrupted liver tissue (30 s at 5000 rpm, Precellys 24, Bertin Technologies). Clear supernatants were collected after centrifugation (14,000 × g for 30 min at 4°C), and total protein content was estimated by Bradford reagent (Sigma-Aldrich) using bovine serum albumin as standard. Proteins (10 μg) were resolved by molecular weight (protein preparation for sodium dodecyl sulphate polyacrylamide gel electrophoresis [SDS–PAGE] gel, 4%–12%, Bis-Tris; Life Technologies, Carlsbad, CA) and visualised with Coomassie blue (PageBlue, Thermo Scientific).

Peptides were obtained by trypsin digestion of 200 μg of total proteins as described in Wiśniewski et al. (2009) where 10 kDa filters were used to reduce, alkylate and digest proteins. Tryptic peptides were recovered by centrifugation, vacuum dried and resuspended to ∼4 μg μl⁻¹ in 50 μl of 0.1% formic acid.

Peptides (2 μl) were chromatographically resolved using an Ekspert 415 NanoLC (Eksigent, Dublin, CA, USA) coupled to a TripleTOF 6600 (SCIEX, Redwood City, CA, USA) and directed onto a trap column (SGE ProtoCol, polar C18, 3 μm, 120 Å, 10 mm × 0.3 mm) for desalting for 5 min at a flow rate of 10 μl min⁻¹ 0.1% FA. Peptides were separated in a ChromXP C18 column (3 μm, 120 Å, 150 mm × 0.3 mm) at a flow rate of 5 μl min⁻¹ with buffer A (0.1% FA and 5% DMSO) and buffer B (0.1% FA and 5% DMSO in acetonitrile). A linear gradient from 3% to 25% solvent B over 38 min was employed followed by 25%−32% B over 5 min, followed by 32%–80% B over 2 min and a 3 min hold at 80% B, transitioning to 3% B over 1, and 9 min of re-equilibration. HPLC eluent was directly to the DuoSpray source of the TripleTOF 6600 MS. The mass spectrometer was set to 5500 V for ion spray voltage, 25 psi for curtain gas and ion gas sources 1 and 2 were set to 15 and 15 psi. The heated interface was 150°C. Data acquisition was performed in information-dependent acquisition mode comprising a high-resolution time-of-flight-MS survey scan followed by 30 MS/MS, each with a 40-ms accumulation time. First stage MS analysis was performed within the mass range m/z 400−1250 with a 0.25-s accumulation time in positive mode. Product ion spectra were acquired for precursor ions over 200 counts s⁻¹ with charge state 2−5. Spectra were acquired over mass range m/z 100−1500 using the manufacturer rolling collision energy based on the size and charge of the precursor ion. Dynamic ion exclusion was set to exclude precursor ions after one occurrence with an 8 s interval and a mass tolerance of 50 ppm. Isotopes within 6 Da of the precursor mass were excluded.

Protein identification and elimination of redundancy

Proteins were identified in ProteinPilot software version 5.01 using the Paragon algorithm (Shilov et al., 2007; Tang et al., 2008) version 5.0.1.0. 4874. Search parameters were alkylation with iodoacetamide and trypsin with no restrictions placed on taxonomy. A S. salar protein database (UniProtKB, 86,853/20200204), including common laboratory contaminants entries, was used. Modifications were set to ‘generic workup’ and ‘biological’ modification as indicated by software consisting of all biological modifications listed in Unimod: acetylation, methylation and phosphorylation. The generic workup comprised 59 potential chemical modifications that may occur as a result of sample handling, for example, oxidation, dehydration and deamidation. The criteria for positive protein identification were proteins with ≥95% confidence. Only proteins with at least two peptides were filtered from the ProteinPilot protein summary report (1% global false discovery rate) and reported as identifications (Table S1).

Protein identification efficiency was derived from mean and standard deviation in GraphPad, Prism 7.05. Redundant proteins were eliminated using combined searches of all spectral datasets obtained for each preserving method. Protein sequences identified in combined searches were retrieved from the UniProtKB S. salar database using seqinr (Charif & Lobry, 2007) in R software (R Development Core Team, 2017). Redundant proteins within cohorts were eliminated in CD-HIT (Huang et al., 2010) using a 90% sequence identity threshold and keeping the longest sequence as representative protein.

Venn diagram and over-representation gene ontology

Shared and unique proteins in each method were visualised by Venn diagram (). Over-representation gene ontology (GO) analysis of unique proteins was used to determine bias by liquid nitrogen or RL. Where available, GO categories were obtained from UniProt . GO analysis was carried out in using default settings.

Protein abundance by principal component analysis (PCA), partial least-squares discriminant analysis (PLS-DA), GO enrichment analysis, panther protein classes and fold-change abundance

Peptide intensity values were extracted from the distinct peptide summary generated by ProteinPilot. Intensities for peptides derived from the same protein were summed to represent the abundance of that protein. This process was carried out for each replicate of the preserving methods. Missing values are often obtained in label-free bottom-up proteomics, and strategies for imputation of missing data are available and frequently employed in biomedical research (Bennike et al., 2016; Liu & Dongre, 2020). Here, proteins with missing values across any given biological replicate were eliminated, and comparisons were carried out using subsets of data frames comprising 698 proteins (SF vs. RL) and 990 (SF vs. SF_FD) (Table S2). Protein abundance variance was visualised by unsupervised principal component analysis (PCA) and by supervised partial least-squares discriminant analysis (PLS-DA) in SIMCA 16.0.2 (Sartorius Stedim Data Analytics AB). GO IDs of proteins with significant abundance (twofold-change; p < 0.05) detected by PLS-DA were retrieved from UniProt where available. The first GO ID and significant abundance p-value of each protein were used as cluster representative and subjected to GO enrichment analysis in Revigo (Supek et al., 2011) using default settings. Protein classes were obtained in Panther () using gene names obtained from UniProt and a human reference background. Protein abundance analysed by fold-change and p-value was determined using R software (R Development Core Team, 2017) adapting scripts from .

RESULTS AND DISCUSSION

Protein recovery was not significant between treatments (Figure 1a). Initially, the standard protein extraction protocol produced a 116.8% coefficient of variance (CV) for RL replicates. Improved CV (15.8%) was achieved after RL stored samples were first cooled down on ice (2 min) and disrupted without extraction buffer followed by further cooling down on ice (3 min) before homogenisation in extraction buffer. SF stored samples produced the best CV (7%). Different strategies to extract proteins from RNA-later stored tissues have included use of different buffers (Alyethodi et al., 2020), sonication (Bae et al., 2019; Bennike et al., 2016; van Eijsden et al., 2013), phenol and methanolic ammonium acetate precipitation (Kruse et al., 2017), and mechanical disruption in phenol–chloroform (Wang et al., 2018), shear force (Zhu et al., 2019), with most approaches producing favourable results for protein quantity and quality (van Eijsden et al., 2013). SF and RL SDS–PAGE comparison revealed similar protein separation patterns (Figure 1b). Higher gel band intensity in RL samples was attributed to interactions between Coomassie blue and nucleic acids (Wenrich & Trumbo, 2012) caused by higher RNA traces in RL. Poor staining of high molecular weight protein was observed in SF3 and RL5 (Figure 1b); however, protein identification by MS was not affected. Similar findings were reported in gills of gulf killifish Fundulus grandis preserved by SF where protein degradation presumably caused poor staining (Abbaraju et al., 2011).

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

Proteolysis is essential in bottom-up proteomics as it generates the peptides used for protein identification (Burkhart et al., 2012). Tryptic peptides bear lysine or arginine at their C-termini. Trypsin conformity was calculated following MS analysis to determine influence of RL over trypsin digestion efficiency. The combined trypsin efficiency by group was SF (98.7%), SF_FD (98.4%) and RL (98.5%), consistent with published data in SF (95%) and RL (94.9%) (Bennike et al., 2016).

Similar sized proteomes were obtained for SF (2510, CV 4.3%), SF_FD (2523, CV 5%) and RL (2644, CV 9.1%) (Figure 1d). There were 1897 common proteins between SF and SF_FD with 613 and 626 unique to each (Figure 2a). Between SF and RL, 2117 were common, whereas 393 and 527 were unique to each (Figure 2b). Other studies using high throughput proteomics reported similar sized proteomes in RL preserved samples, 2646 (Kruse et al., 2017) and 3840 (Bennike et al., 2016). A different study used SDS–PAGE and proteomics to assess preservation of gulf killifish F. grandis heart, brain, skeletal muscle, gill and liver and concluded that liver and heart samples were preserved better by SF compared to RL (Abbaraju et al., 2011).

GO analysis of uniquely identified proteins did not reveal significant differences between SF and SF_FD (Table S3). Comparing SF versus RL showed enrichment in two membrane categories and one metabolic process (Table S4) which was potentially caused by RL diffusion in tissue, as reported previously (Kruse et al., 2017). Negligible effects of RL preserved samples on GO term enrichment have been reported elsewhere (Bennike et al., 2016).

Sample clustering was tighter in SF compared to either SF_FD or RL samples (Figure 2c,d). Contrastingly, a different study in peripheral blood mononuclear cells reported tighter clustering patterns in cells stored in RL compared to fresh cells (Alyethodi et al., 2020). Further PLS-DA (Figure 3a) analysis identified 69 proteins significantly (>2 fold-change; p < 0.05) less abundant in RL (Table S5). Comparatively, a study in Arabidopsis thaliana found ∼200 proteins differentially abundant between samples preserved in RL or liquid nitrogen (Kruse et al., 2017).

[IMAGE OMITTED. SEE PDF]

The effect of RL on differentially abundant proteins was visualised by heat map (Figure 3b). GO enrichment analysis using significant p-values derived from PLS-DA was carried out and identified 32 molecular functions affected in RL samples. Some functions included transcription and translation processes (Figure 3c) that were expected to be ‘halted’ in RL samples by the action of RL to preserve nucleic acids and proteins (Kruse et al., 2017). Additionally, six cellular components and eight biological process categories involved in processing and transport of essential molecules (nucleosome, ribosome and protein transport) were significantly altered in RL (Figure 4a). The A. thaliana study also reported perturbations in GO categories associated with transport and RNA processing between RL and SF samples (Kruse et al., 2017).

[IMAGE OMITTED. SEE PDF]

To better understand the nature of the 69 proteins affected by RL, proteins were classified in Panther using available UniProt gene names and human reference background. Using zebrafish as background produced fewer protein classes compared to the better annotated human background (data not shown). Results indicated that RL affected proteins involved in transcription, translation and proteolysis where ribosomal and serine protease proteins were highly affected by RL (Figure 4b). Other affected protein classes that could be part of transcription machinery included methyltransferase, decarboxylase, ligase, nucleotide phosphate and chromatin-binding. These results are consistent with the intended purpose of RL that seeks to stabilise DNA, RNA and proteins by denaturing DNases, RNases and proteases. RL effect over transcriptional machinery has also been reported elsewhere (Kruse et al., 2017). Similar protein abundance in SF and SF_FD samples prevented discrimination by PLS-DA as the model was unable to fit any component.

No significant fold-change in protein abundance was identified between SF and SF_FD (Figure 4c), whereas SF and RL comparison identified seven proteins with significantly less abundance in RL (Figure 4d) and one less abundant in SF. The identity and function of these eight proteins suggested the influence of permeating RL over membrane integrity (Kruse et al., 2017) and ribosomal processes (Table 1). Variable effects of RL on proteome abundance have been reported. For example, no significant effects of RL have been observed on the proteomes of microbes (Saito et al., 2011), human colon, (Bennike et al., 2016), human pancreas and phosphoproteome (Bae et al., 2019); however, RL significantly impacted the proteomes of human stools (Hickl et al., 2019) and A. thaliana (Kruse et al., 2017).

TABLE 1 Significantly different proteins identified in salmon liver stored by snap-freezing (SF) and RNAlater (RL)

Accession	Protein	Function	log_fca	log_pval
A0A1S3LCD0	Heterogeneous nuclear ribonucleoprotein A1-like isoform ×2	RNA binding	−2.6339	1.6498538
A0A1S3LSW2	tricarboxylate transport protein, mitochondrial	Transmembrane transport	−2.3059	2.3583165
A0A1S3Q7G0	Ribosome-binding protein 1-like isoform ×2	Protein transport	−2.5032	1.5959901
A0A1S3QKS3	Maleylacetoacetate isomerase isoform ×2	Catalytic protein: l-phenylalanine degradation	−2.0577	2.8550735
B5X4K4	l-Lactate dehydrogenase	Catalytic protein: pyruvate fermentation to lactate	−2.5382	2.3142437
B5XAB9	40S ribosomal protein S29	Translation	−2.9439	2.386765
C0HAF1	Regulator of chromosome condensation	–	−2.4486	1.905497
A0A1S3T350	2-Oxoisovalerate dehydrogenase subunit alpha	Catalytic protein: conversion of alpha-keto acids to acyl-CoA and CO2	2.20468	1.4842538

Consistent with the purpose of RL, our study has shown that RL impacts proteins involved in transcription, translation, transport and proteolysis. Here, we also demonstrate no effect of RL on the abundance of proteins originally identified as stress markers in liver samples from salmon reared at elevated temperatures and moderate hypoxia (Beemelmanns et al., 2021; Nuez-Ortín et al., 2018) or with differential abundance under specific feeding regimes (Esmaeili et al., 2021; Sissener et al., 2010). Preserving liver in RL did not influence abundance of catalase, superoxide-dismutase, caspase, peroxiredoxin, heat shock proteins, glutathione-s-transferase and complement C1q-like protein 2 (Table S6). Calreticulin and thymidine phosphorylase were initially identified as up- and down-regulated in liver of Atlantic salmon fed a genetically modified soy strain (Sissener et al., 2010). Acetoacetyl-CoA synthetase, fatty acid synthase and long-chain-fatty-acid-CoA ligase 4-like were identified in liver of feed efficient Chinook salmon (Oncorhynchus tshawytscha), whereas alcohol dehydrogenase 1 was identified in feed inefficient Chinook salmon (Esmaeili et al., 2021). Here, neither protein was significant in RL samples (Table S6). This demonstrates that neither liquid nitrogen, freeze-drying or RL significantly influenced fold-change abundance and that an external stimulus (e.g. high temperature, hypoxia and diet) is required to trigger significant differences in the level of these proteins, as previously reported.

CONCLUSION

Our study demonstrates that preserving salmon liver in RL does not significantly influence proteome size and quality. We also found that 69 proteins (∼2.6% of the total proteome) involved in transcription, translation, transport and constituents of ribosome were significantly less abundant in RL samples due to denaturation caused by the protective mechanisms of RL to preserve nucleic acids and proteins. RL did not significantly influence the abundance of important proteins previously identified in the liver of salmon reared under stress or different diets. These findings suggest that RL can be used for proteomic samples and aliquoted into tubes prior to the collection of field samples offering a safer and effective preservation technology to use at remote aquaculture locations.

AUTHOR CONTRIBUTIONS

Anca G. Rusu: investigation; methodology; writing – original draft. James A. Broadbent: data curation; formal analysis; methodology; writing – review and editing. Nicholas M. Wade: formal analysis; writing – review and editing. Cedric J. Simon: formal analysis, writing – review and editing. Artur N. Rombenso: formal analysis, funding acquisition, writing – review and editing. Simone A. Osborne: formal analysis, writing – original draft, writing – review and editing. Omar Mendoza?Porras: conceptualization; data curation; formal analysis; methodology, supervision; writing – original draft; writing – review and editing.

ACKNOWLEDGEMENTS

This study was funded by CSIRO Agriculture and Food. The authors would like to thank David Blyth and Ha Truong for maintenance of salmon tanks and sampling. We also thank Dr Richard Taylor and Dr Sophia Escobar-Correas for their internal CSIRO review.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT

Mass spectrometry data associated to this manuscript is available at CSIRO repository at and available on request.

ETHICS STATEMENT

Procedures were approved by CSIRO Animal Ethics Committee (CQAEC permit #2018-44).

PEER REVIEW

The peer review history for this article is available at: .

References

Abbaraju, N.V., Cai, Y. & Rees, B.B. (2011) Protein recovery and identification from the gulf killifish, Fundulus grandis: comparing snap‐frozen and RNAlater^® preserved tissues. Proteomics, 11(21), 4257–4261. [DOI: https://dx.doi.org/10.1002/pmic.201100328]