Large, prospective biorepositories are a prerequisite in order to explore the molecular basis and biomarkers of favourable and unfavourable transplant outcomes. In the first large-scale transplant biobank, the UK QUOD initiative, donor samples and data are handled and stored according to a strict and uniform protocol and thereby facilitate donor-based, outcome-focused research. Unfortunately, since a similar initiative is not yet available for recipient samples, critical questions regarding a possible role of (molecular) recipient factors in transplant outcomes can currently not be adequately investigated nor addressed [6, 7]. Establishment of an adequately sized biobank that incorporates sufficient cases to address short term complications and has adequate long-term follow up information will be extremely time-consuming, thereby interfering with a timely evaluation of the role of recipient factors.

A possible alternative and rich source of recipient samples are the sera that have been stored for immunological matching and surveillance. However, since these sera have been collected by different centres using local clinical protocols, there probably is a large variability in sample handling and storage practices, whilst the material may also have been exposed to an unknown number of freeze–thaw cycles; factors that may all impact serum/plasma quality for future proteomic and metabolomic analysis [10,11,12]. In the light of the huge potential of these clinical sera, and the fact that a prospective collection will take years to gain any relevant numbers, we decided to investigate to what extent these samples might still provide relevant biological information to build biological pathways from. In this feasibility study, we applied a state-of-the-art shotgun and data-dependent acquisition metabolomics and proteomics approach in order to estimate the informative value of long-term stored, immunological surveillance lab-derived recipient serum samples, taking along uniformly biobanked QUOD donor plasma samples in parallel analyses.

A potential interference in this study is that it relies on two different sample groups. In an ideal world, one would have used uniform biological samples (serum or plasma) from a homogenous population (recipients or donors), and in parallel test fresh and long-term stored (> 20 years) samples. However, this is not feasible, and unfortunately, even a potential use of artificially aged samples as a reference, i.e., through exposing fresh samples to multiple freeze–thaw cycles, cannot reflect actual storage [10]. Nevertheless, the impact of non-uniform biological samples appears less than commonly thought, since serum and plasma proteomes are reportedly very similar, apart from fibrinogen levels [20]. To minimize the impact of biological differences, the pathway enrichments were performed against a reference database rather than one-to-one (i.e., plasma vs serum).

The advantage of the applied shotgun method is that it yields a broad overview of all proteins and metabolites present and is therefore an appropriate strategy for an explorative approach. However, it does not provide the optimal analytical circumstances for every specific compound, in contrast to a targeted approach. This is reflected e.g., by the incomplete identification of the amino acids class of components in both plasma and serum (Additional File 4 and 5). Also, for proteins, incomplete patterns were observed: i.e., complement protein C3 has nine protein fragments [17], while shotgun proteomics only identified three fragments in serum and four in plasma (Fig. 3). Nevertheless, the shotgun approach enables the generation of global metabolomics and proteomics data as required for pathway analysis, which is essential to answer biological questions.

In this feasibility study, LC–MS metabolomics was successfully applied in both groups: 891 and 803 compounds were identified in the recipient serum and donor plasma samples, respectively, of which 361 and 425 were endogenous metabolites. The majority (75%) of the compounds were identified with high confidence, and metabolites from similar metabolite classes were found in both groups (Fig. 1C), as expected for human blood samples. The median m/z ratio was higher for serum samples (313 m/z) compared to the plasma samples (230 m/z). Although a storage effect cannot be excluded, this difference more likely results from the clinical difference between the groups: whilst the plasma samples are derived from donors with adequate renal function, the serum samples all originate from patients suffering from end-stage renal disease, a condition known to result in accumulation of metabolites [21, 22], which could have contributed to the higher m/z ratio in the serum samples.

Although the measured protein concentrations were within the theoretical serum/plasma protein concentration range of 60–80 mg/mL [23], the concentration was higher in serum (69 ± 9 mg/mL) compared to plasma (61 ± 11 mg/mL), whilst the opposite was expected [24]. The most likely explanation is the prolonged storage of the serum samples, during which water will have evaporated, which has resulted in concentration of the sample. Whereas this affects absolute metabolite and protein abundance, and thus affects specific biomarker analysis, it does not impact the relative concentrations, and will therefore not interfere with general pathway analysis based on the presence of proteins rather than absolute abundances.

Using the power of shotgun LC–MS proteomics, it was possible to quantify not only the master proteins but also protein fragments in plasma and serum. In recipient serum, 42.5% of all proteins were fragments, whilst in the donor plasma group 26.5% of the proteins were fragments. This was also reflected in the size distribution with an overall lower MW of serum sample proteins compared to plasma. Although it cannot be ruled out that this increased fragmentation in kidney recipient sera is of biological origin [25], it is also an expected consequence of the non-uniform sampling and extended storage, which could e.g. lead to deamidation. In line with this high abundance of protein fragments in the serum samples of our study, a high level of deamidation (70%) was observed. Deamidation is a non-enzymatic, chemical post-translational modification that can occur in peptides and proteins during their lifespan (in vivo and in stored samples), which can have significant implications for the stability and integrity of these molecules in biological samples as it can make proteins more susceptible to degradation by proteolytic enzymes [26]. Whilst the observed fragmentation of proteins will interfere with analyses based on antibody-based assays or aptamer platforms, it has a limited effect on LC–MS-based pathway analysis; a fragment will still be identified as a fragment originating from its parent protein. The fragment can thus be taken along in the analysis as if it was an intact protein, if required.

The analytical strategy adequately generated metabolic and proteomic profiles from the recipient serum samples. To test to what extent the acquired data could be mapped along theoretical pathways, an integrated pathway analysis was performed in which enriched pathways in the recipient serum and donor plasma were mapped against the KEGG reference database. Integration of the qualitative proteomics and metabolomics data of both recipient serum and donor plasma samples resulted in the significant enrichment of a total of 79 pathways. Although the complement and coagulation cascades pathway had the highest impact and lowest p-value, interpretation of this pathway is potentially interfered as coagulation cascade activation in serum can trigger the complement cascade [27]. Consequently, this pathway is suboptimal for exploration of the potential impact of storage artifacts. Therefore, it was decided to focus on ECM-receptor interaction pathway as a universally enriched pathway for integrated pathway analysis. In-depth STRING analysis of this pathway showed major overlap between donor plasma and recipient serum samples (Fig. 4). Interestingly, with only three nodes missing from the serum analysis, the overall overview of the ECM receptor interaction pathway remained intact. This illustrates that older serum samples from non-uniform, clinical biorepositories can be applied for general qualitative proteomic and metabolomic integrated pathway analysis, yielding a resolution similar to that of strictly biobanked samples.

Besides the difference in biofluids, another limitation that especially affects metabolite abundances is the difference in nutritional state: the donor plasma samples are obtained from individuals who are in the process of dying and thus in a fasting, glucose-infusion state, whilst the serum samples originate from non-fasting recipients using a variety in diets. Moreover, donor type (brain death vs circulatory death) may also impact metabolism [28]. Again, we have therefore decided to not compare the two groups one-to-one.