High-throughput metabolomics is a powerful tool for systematic metabolite profiling of complex biological systems to understand disease mechanisms and to identify novel clinical biomarkers for diagnosis, prognosis and treatment response. For example, in recent years, this technique has been successfully applied for developing biomarkers and gaining deeper knowledge of common and devastating diseases like cancer [1
], type 2 diabetes [3
] or cardiovascular diseases [6
]. One of the most common types of sample matrix used in these research areas is blood-based material, such as serum or plasma, because of its minimally invasive accessibility and the extensive coverage of the human metabolome. However, since metabolites are sensitive to pathological alterations and improper sample handling, accurate quality assurance as well as quality control are mandatory to obtain reliable results and to ensure reproducibility [9
]. The main source of laboratory uncertainty is pre-analytical variability. Failures in identifying confounders could lead to serious misinterpretations and erroneous clinical decisions [19
]. Inability to complete metabolic biomarker studies due to pre-analytical confounding factors have been reported [20
], and similar challenges have been observed in transcriptomics, peptidomics and proteomics research [21
]. The most relevant technical issues in the pre-analytical phase are sample collection, processing, transport, and storage [16
]. In clinical research, and particularly in -omics approaches, sample quality must be guaranteed by standard operating procedures (SOPs) to eliminate pre-analytical bias caused by inappropriate sample handling or inadequate storage conditions. However, in multicenter studies it is difficult to ensure that each institution strictly adheres to the sample preparation procedures as defined by SOPs. While specific SOPs can be determined for sample collection, processing and transport, changes in metabolite concentrations during storage are challenging to control and cannot be completely avoided. Nevertheless, the experimental design of long-term retrospective and prospective epidemiological studies often requires the use of frozen samples that were collected months, years or even decades ago, and were thus subject to different retention periods prior to analysis [25
]. Therefore, knowledge of metabolite stability during long-term storage is of paramount importance to allow unbiased comparisons of samples collected at various time points and stored for different periods of time.
Cryoconservation of biospecimens in liquid nitrogen is considered the preferred method of sample preservation due to excellent sample stability [26
]. As an example, ascorbic acid concentrations were reported to be significantly stable over a storage period of eleven years in liquid nitrogen tanks [27
]. However, liquid nitrogen poses handling hazards and might be too expensive in biobanking environments [28
], Therefore, sample storage in freezers at a temperature of at least −80 °C was recommended to maintain long-term integrity of biomarkers [29
]. Criteria for selecting the optimum storage temperature were reviewed by Hubel et al. [30
]. Most of the existing studies on metabolite stability during sample storage focus on short-term stability or effects of repeated freeze-thaw cycles [12
], while the impact of long-term storage on the metabolic fingerprint is not yet fully understood. Indeed, Hustad et al. investigated the influence of storage time on biomarkers related to vitamin B metabolism in serum and plasma samples stored for up to 29 years [34
], Yang et al. examined two plasma sample cohorts at two different time points within a five-year-framework [11
], and Abuja et al. simulated the effect of storage time by repeated temperature changes [26
], Recently, Haid et al. found that the levels of amino acids, acylcarnitines, glycerophospholipids, sphingomyelins and the sum of hexoses in plasma samples are altered after five years of storage [35
In the present work, we investigated the impact of storage time on the human ethylenediaminetetraacetic acid (EDTA) plasma metabolome in samples that were stored at −80 °C for up to 16 years prior to analysis. Besides providing new basic insights into the stability of metabolites during long-term storage in the freezer, these results and the derived knowledge of differing sensitivity of metabolites to storage effects will be valuable in the development of novel biomarkers.
In this work, we investigated the stability of 231 human plasma metabolites from ten different ontology classes during long-term storage at −80 °C over a period of up to 16 years. Due to the longitudinal study design with repeated sampling of the same individuals at 70, 75 and 80 years of age it was not possible to distinguish subject age-related from storage-time-dependent effects when comparing storage times across different subject age groups. Therefore, within each subject age group in this longitudinal study we investigated the influence of longer storage against shorter one. Hence the difference between the oldest and youngest samples in any one subject age group was about four to seven years. We found that 98% of the analyzed metabolites remained stable in the first seven years of storage, but upon prolonged storage of up to 16 years, time differences of few years resulted in a statistically significant change in concentration in up to 26% of the analyzed metabolites. All metabolite classes were affected to a certain degree, with complex lipids, fatty acids, energy metabolism molecules, and amino acids being the most affected by long-term storage. Therefore, these data indicate that the human plasma metabolome is adequately stable to long-term storage effects at −80 °C for up to seven years but is sensitive to even few years of additional storage if stored for longer time periods at −80 °C. Liquid nitrogen may be an alternative for highly unstable metabolites that require storage at lower temperatures but should be tested in further studies.
It is noteworthy, however, that many affected metabolites displayed significant changes only in the age group of 75 years, i.e., samples were stored for seven up to nine years and nine up to eleven years. Those significant effects vanished upon even longer storage of 14 up to 16 years versus eleven up to 14 years in the age group of 70 years.
Since the reasons for the observed effects may be related to oxidation reactions, acid-base-driven hydrolysis and enzymatic activities it is possible that these reactions reach a substrate-product-equilibrium upon prolonged storage of eleven up to 16 years. Previous studies have reported that hydrolysis cannot be stopped by an increase in viscosity, because the movement of H+ and OH− ions is possible even in solid ice [36
]. In addition, freezing of aqueous solutions results in an increase in concentration of reactants, catalysts, electrolytes, and solvents in the remaining liquids, accompanied by changes in pH, solubility, viscosity, ionic strength, and thermodynamic properties [38
However, the majority of samples from these storage groups were classified as poor quality according to the MxP Sample Processing Control meaning that irrespective of the decline of significant metabolite changes as compared to shorter storage times these samples should not be used in metabolomic analyses.
In a recent study, Haid et al. showed that amino acids are sensitive to long-term storage at −80 °C [35
]. We confirmed this aspect for sample storage and further found that asparagine, cysteine and cystine, which are not included in their work, were all significantly reduced in concentration during storage. In contrast to the amino acids mentioned above, aspartate was significantly increased upon longer storage. In general, these effects can be explained by the fact that peptide bonds and amino acid side chains are susceptible to non-enzymatic hydrolysis. Particularly, asparagine can be converted to its dicarboxylic acid counterpart aspartate by deamination [39
]. Cysteine instability can be explained by rapid oxidation to cystine. However, since cystine levels were also reduced over the storage period, the reduction of both cysteine and cystine could be a result of oxidative conversion to unidentified derivates, as described previously [33
]. Moreover, non-enzymatic oxidation may explain the observed reduction of pyruvate concomitant with an increase in lactate (see Table S1
) if samples were stored for up to eleven years versus those stored for up to nine years. Likewise, the observed increase in glycerate may derive from oxidation of glyceraldehyde, an intermediate of sugar metabolism. However, changes in pyruvate and lactate have also been noted upon prolonged blood or plasma processing and attributed to erythrocyte-derived enzymatic actions [14
In addition to amino acids Haid et al. analyzed the alteration of LPCs and PCs after storage for five years. However, contrary to our observations LPCs were found reduced or unchanged in this publication with the PCs showing a more heterogenous behavior [35
]. We could confirm that upon storage for up to seven years LPCs remained largely unchanged. However, after longer storage even four or five years of additional storage led to an increase in most of the measured LPCs. That is in good agreement with Kamlage et al., who found increased LPC concentrations upon prolonged serum processing at room temperature [18
]. In contrast, the changes observed in PCs upon longer storage were inconsistent. Whereas, PCs with fatty acids of less than 20 carbon atoms and two double bonds at most remained mostly unchanged, PCs containing polyunsaturated fatty acids were reduced upon longer storage (see Table S1
). Interestingly, we measured two arachidonic acid containing PCs. One of those is shown in Figure 2
E. The other one, PC (C16:0, C20:4) (data not shown), displayed a similar behavior but failed to reach FDR < 0.05. The increase of LPCs and the concomitant decrease in polyunsaturated fatty acid-containing PCs may likely derive from phospholipase activity. Several isozymes of phospholipase A2 are calcium-independent or require minimal calcium amounts [41
] and hence may be active in EDTA plasma even at low temperatures. Interestingly, several phospholipase A2 isozymes display great specificity for arachidonic acid at the sn-2 position [41
] and may explain the reduction in arachidonic acid containing phosphatidylcholines. Furthermore, some phospholipase A2 enzymes are predominantly active on oxidized phosphatidylcholines that may derive from oxidation processes upon long-term storage [42
]. It remains to be analyzed, however, if those polyunsaturated fatty acids are indeed released and if there are downstream reactions that may conceal their potentially increased free fatty acid concentrations. Because of their numerous double bonds, they are vulnerable to chemical reactions such as oxidation processes. The observed reduction of polyunsaturated fatty acid containing PCs is of great importance since arachidonic acid and its downstream products are important mediators of inflammatory processes [44
] and hence studies aimed at detecting biomarkers in inflammatory diseases should always account for storage time as a potential confounder for their results.
Another interesting metabolite group that has been significantly impacted by long-term storage is lipid hydroperoxides, which have been increased over time by lipid oxidation and auto-oxidation processes. This so-called lipid peroxidation is a free radical-generating process that leads to the oxidative modification of lipids. Although lipid hydroperoxides have been described as biomarkers for the assessment the oxidative stress status and associated diseases [45
], their susceptibility to long-term storage should be considered in the development of biomarkers containing these metabolites.
A limitation of our study is that lifestyle changes, diseases, etc. as a result of ageing of the individuals may also play a role on the changes of metabolite concentrations, i.e., effects of long-term storage on the plasma metabolome overlap with effects of biological ageing of the subjects. To overcome this problem metabolite concentrations were compared within subject age groups. However, that resulted in much smaller time differences between longer-stored and shorter-stored samples of seven, four and five years in the subject age groups of 70 years, 75 years, and 80 years, respectively. Furthermore, we focused our observations on EDTA plasma samples. Hence, it is possible that other blood-derived biospecimens such as serum or heparin plasma may display different changes or sensitivities in long-term storage.
In conclusion, access to high-quality samples that are collected and handled in standardized ways to minimize or even exclude confounding factors is key to the “bench to bedside” goal of translational research. We now have provided evidence that long-term storage of samples has a major impact on the stability of metabolites in human plasma, which in turn influences data analysis in metabolomics studies. In this context, we could demonstrate that metabolite profiling is well-suited for identifying low-quality samples prior to data analysis. However, since nearly all tested metabolites were stable for up to seven years at −80 °C, biomarker studies based on frozen samples should be performed as soon as possible after sampling.