Platynereis spp. were collected via snorkelling or scuba from either inside (40°43′53″N, 13°57′47″E) or outside (40°43'33.33N, 13°57'36.38E and 40°44′48″N, 13°56′39″E) the carbon dioxide (CO2) vent on the island of Ischia (Italy) and used in a reciprocal transplant experiment. The effect of exposure to high or low partial pressures of CO2 (pCO2) conditions on the metabolome (metabolome, and lipidome) of worms from different pCO2 regimes was investigated to understand the effect of exposure to different pCO2 conditions on the cellular physiological response. This experiment was conducted between 04/09/2013 and 16/09/2013. The experiment was staggered during this time so all worms could be processed. After five days exposure to either low or high CO2 conditions worms were snap frozen in liquid nitrogen and shipped to the University of Birmingham for metabolomic analysis which was finalised on 21/01/2016. Metabolomic profiles of worms were characterised using a mass spectrometry approach. A standard mass spectrometry based metabolomics workflow was used to analyse both the polar and lipid extracts from the samples (Kirwan et al. 2014). Raw mass spectral data were processed using the SIM-stitching algorithm, using an in-house Matlab script. The data matrices were normalized using the PQN algorithm. Missing values were imputed using the KNN algorithm. The resulting data matrix was analysed using univariate statistics, described below. The same matrix was transformed using the generalised logarithm to stabilise the technical variance across the measured peaks prior to analysis using multivariate statistics. Signals were putatively annotated with empirical formulae calculated by the MIPack software (Weber et al. 2010), searching the KEGG (Kanehisa et al. 2012) and LipidMaps (Fahy et al. 2007) databases, and confirmed by performing calculations based on the original spectra in Xcalibur 2.0.7 (Thermo Fisher Scientific).

Processed metabolomics data:For both polar and lipid extracts, first processed metabolomics data were analysed using Perseus (Tyanova et al., 2016) and MetaboAnalyst 4.0 (Chong et al., 2019). Principal components analysis (PCA) was used to unravel the data structure and any group separations. A sample cluster analysis was performed using Euclidean distance and Ward's clustering algorithm. At this point three outlying samples were identified and removed from further analysis in the polar extracts dataset. Volcano plots were used in Perseus to identify differentially abundant metabolites between individuals transplanted to the same environment (SE, which includes CC and AA) versus those transplanted to a different environment (DE, which includes CA and AC) and (ii) individuals from the control site of origin (CC and CA) versus individuals from the acidified site of origin (AA and AC). Volcano plots were based on t-tests with 250 randomizations, FDR 0.05 and s 0.1. Next, a heatmap and hierarchical cluster analysis was carried out after Z-scoring the data matrix. Spearman correlation was used as distance measure and complete linkage as clustering method. Polar extracts:To identify enriched compound types and relevant biological pathways within the significantly different m/z peaks, we also applied the mummichog and GSEA algorithms to the data, using the module 'MS peaks to pathways' in MetaboAnalyst 4.0 to predict network activity from untargeted metabolome data (see Li et al., 2013). Data was input as a table with m/z peaks, p-values and fold-changes. The following parameters were used in the analysis: molecular weight tolerance 5 ppm, negative mode, p-value <0.01 for peak significance in mummichog, database 'non-lipids – sub chemical class', which contains 778 main non-lipid chemical class metabolite sets from RefMet. The same analysis was then run using the KEGG database for the model species Caenorhabditis elegans Maupas 1900, for pathway identification. Significantly enriched compound classes and pathways were based on p < 0.05. Lipid extracts: To identify relevant biological pathways within the significantly different m/z peaks, the ions with a putative annotation were categorized into a main lipid class, and the biological function of each lipid class was retrieved from the Encyclopedia of Lipidomics (Wenk, 2019) and HMDB Metabocards. As lipid annotation is still a major bottleneck in untargeted lipidomics, we also applied the mummichog and GSEA algorithms to the data, using the module 'MS peaks to pathways' in Metaboanalyst 4.0 to predict network activity from untargeted lipidome data, bypassing the need to identify lipids (see Li et al., 2013). Data was input as a table with m/z peaks, p-values and fold-changes. The following parameters were used in the analysis: molecular weight tolerance 5 ppm, negative mode, p-value <0.01 for peak significance in mummichog, database 'lipids – main chemical class', which contains 77 main lipid chemical class metabolite sets from RefMet. Significantly enriched lipid classes were based on p<0.05.Abbreviations: SIM-stitching algorithm: collection of multiple adjacent selected ion monitoring (SIM) windows that are ‘stitched’ together computationally PQN algorithm: Probabilistic Quotient Normalization algorithm KNN algorithm: K nearest neighbours algorithm KEGG: Kyoto Encyclopedia of Genes and Genomes FDR: false discovery rate GSEA algorithm: Gene set enrichment analysis algorithm* HMDB: Human Metabolome Database