This systematic review was conducted in accordance with a previously developed protocol. Both the protocol and the systematic review were registered in PROSPERO (No. 1181037). The review was conducted in accordance with the PRISMA 2020 guidelines (14).

The inclusion and exclusion criteria for studies were determined using the PICOS framework:

Population (P).

Inclusion Criteria: Studies involving human participants who received any type of COVID-19 vaccine. Participants must not have an active SARS-CoV-2 infection at the time of sampling. Only studies analysing venous blood samples (serum, plasma, or whole blood) collected after vaccination will be included.

Exclusion Criteria: Studies using non-human models (animal, cell lines, in vitro, in silico), studies of infection without vaccination, studies based on biological fluids other than blood. Studies involving participants with chronic diseases will be excluded unless a healthy, vaccinated control group is analysed separately.

Intervention (I).

Intervention is the administration of any COVID-19 vaccine. This includes, but is not limited to, vaccines: mRNA-based (e.g., BNT162b2, mRNA-1273); vector-based (e.g., ChAdOx1 nCoV-19, Ad26.COV2.S); inactivated (e.g., CoronaVac, BBIBP-CorV); subunit protein (e.g., NVX-CoV2373).

Comparison (C).

Comparisons may include: pre-vaccination samples from the same individuals, unvaccinated control groups, different types of vaccines, and different time points after vaccination.

Outcomes (O).

Primary Results: Changes in metabolomic profiles (metabolites, lipidome, metabolic pathways) in venous blood after vaccination.

Secondary Outcomes: Data evaluating the immune response to vaccination. Identification of specific metabolites or pathways associated with immune response or adverse effects. Differences between vaccine types or sampling time points.

Analytical Techniques: Interest is in studies using LC-MS, GC-MS, CE-MS, NMR, and other analytical methods to characterize the metabolome.

Study Type (S).

Inclusion Criteria: Original studies, including clinical trials, cohort studies, cross-sectional studies, and experimental studies that report metabolomic data.

Exclusion Criteria: Reviews, meta-analyses, protocols, letters, conference abstracts, and preprints. Only peer-reviewed articles in English were included.

The following databases were searched: PubMed/MEDLINE, Embase, Scopus, and Web of Science. The studies were retrieved using a Boolean sequence: (metabolome OR metabolomics OR “metabolic profile” OR “metabolic profiling” OR lipidome OR lipidomics) AND (“COVID-19” OR “SARS-CoV-2”) AND (vaccine OR vaccination OR vaccinated OR immunization).

The selection of studies was carried out by two independent reviewers. In order to optimize the abstract screening process in this review, automation based on a large language model (LLM) was implemented. This model, operationally integrated with the data management platform (Catchii) (15), acts as an inference engine. This model translates the protocol criteria (PICOS) into formal classification instructions, allowing for prioritized selection of material. Each record obtained using LLM was verified by a researcher and served only as support in the screening process.

Stage 1: Both reviewers will independently conduct a preliminary review of titles and abstracts for compliance with inclusion criteria.

Stage 2: Studies deemed potentially eligible will proceed to full-text assessment.

Disagreements in decisions were resolved by reaching a consensus.

The following information was extracted from the full texts of the included articles and supplementary materials: Study characteristics: author, year, country, study design; Population: sample size, demographics, vaccine type; Samples: sample matrix (serum, plasma, whole blood), time point of collection. Analytical method; Main metabolomic results: identified metabolites, altered pathways, clinical data, and additional data that may influence the interpretation of results, such as limitations, additional research methods. The visualization was created using the robvis tool (16).

The risk of bias will be independently assessed by two reviewers. The ROBINS-I V2 (Risk Of Bias In Non-randomized Studies - of Interventions, Version 2) (17) will be used to assess non-randomized studies, while the Cochrane RoB 2 (Risk of Bias tool for randomized trials, Version 2) will be used to assess randomized controlled trials (RCTs) (18).

The results were summarized narratively. Studies will be grouped by vaccine type, time point, and analytical technique used to synthesize consistent metabolic signatures.

Ten studies were included in the synthesis (11–13, 19–25) meeting the inclusion criteria, focusing on serum/plasma metabolomic profiling in response to COVID-19 vaccination. The study populations (total n > 1,200) mainly included healthy adults of various ages vaccinated with mRNA-based preparations (Pfizer/Comirnaty) and inactivated vaccines (Sinopharm, CoronaVac, BBIBP-CorV). Detailed results of individual stages of the research selection process are presented in Figure 1.

The studies were predominantly longitudinal in design, with sampling time points covering the period from baseline (pre-vaccination) to the short- and medium-term phases, reaching a maximum of 480 days after the first dose (21). The most commonly used analytical techniques were liquid chromatography-mass spectrometry (LC-MS) and nuclear magnetic resonance (NMR) spectroscopy.

The characteristics of the included studies are presented in Table 1. A summary of metabolomics data and key conclusions is presented in Table 2. An analysis of integrated key metabolomics data is summarised in Table 3.

Detailed metabolomic data at the study level and individual study conclusions are presented in Table 2. To account for the heterogeneity of results, Table 3 presents a synthesis of results from different studies, highlighting consistent metabolic signatures that persist across different vaccine platforms and analytical methods.

A key recurring inaccuracy in non-randomized studies was selection bias, where comparison groups were defined post-hoc based on the antibody levels achieved. This approach creates a fundamental bias in estimating the relationship between the metabolic biomarker and the immune response. A second significant weakness was insufficient control for confounding factors such as age, sex, or BMI. In studies where groups were unevenly distributed in terms of age and gender, the lack of statistical adjustment raised doubts about the reliability of the estimated differences in metabolism. In addition, metabolic analyses often contained inaccuracies related to missing data or non-standard data handling. Some key analyses were based on data with significant sample loss (even >70%) or used simplified imputation methods (e.g., 1/5 of the minimum value) for LC-MS data, which reduced the reliability of the results. There was also a risk of selective reporting of results, with researchers choosing the most striking metabolites (e.g., VLDL-TG, Pregnenolone) from hundreds of measured variables for discussion without relying on a pre-specified analysis plan. In contrast to the above, studies using a self-controlled (pre-post) design on homogeneous cohorts (12, 13, 22) demonstrated a stronger methodology in terms of controlling for baseline confounders. In contrast, the only randomized controlled trial analysed (24) did not show any significant inaccuracies, confirming its high methodological quality. The assessment of the risk of bias in specific publications is summarised in Figure 2.

This systematic review, conducted in accordance with the PRISMA 2020 guidelines (14), synthesizes data from 10 studies involving more than 1,200 healthy participants, analysing metabolomic changes in venous blood after vaccination against COVID-19. The studies focused mainly on mRNA (Pfizer/Comirnaty) and inactivated (CoronaVac, BBIBP-CorV, Sinopharm) vaccines, excluding vector and subunit platforms due to a lack of eligible publications. Key findings indicate consistent, albeit heterogeneous, modifications in amino acid, lipid, and energy metabolism associated with immune system activation. These results complement earlier reports in vaccinology (9, 10), highlighting the role of the metabolome as an early indicator of immune response.

Despite differences in sensitivity between platforms (higher for LC-MS in the case of lipids, better repeatability for NMR in the case of major metabolites), the kynurenine pathway and changes in glutamine levels show high consistency in both techniques. These changes are most pronounced in the acute phase (1–2 days after vaccination), while lipoprotein profiles (detected mainly by NMR) evolve over a longer period (up to 21 days), suggesting that the early metabolic response is dominated by amino acid metabolism and the late response by lipid remodelling (26).

The complementary nature of NMR and LC-MS platforms is crucial for a comprehensive metabolomic landscape. While LC-MS offers higher sensitivity for low-abundance metabolites, 1H-NMR provides superior reproducibility and minimal sample preparation, as previously highlighted in the context of neuroimmunological disorders.

The results of this systematic review correspond to the broader context of metabolomic research on post-vaccination responses, both in relation to other vaccines and to the pathophysiology of SARS-CoV-2 infection. At the same time, they reveal specific features of COVID-19 vaccination resulting from unique technological platforms and the pandemic context.

Despite identifying consistent metabolic patterns, the study results are subject to limitations typical of metabolomics, such as selection bias (in the case of post-hoc definition of HR/LR groups) and insufficient control of confounding factors (age, gender, BMI).

A significant limitation of this review is the exclusion of vector-based (e.g., AstraZeneca) and protein subunit (e.g., Novavax) vaccines due to a lack of eligible metabolomics studies at the time of the search. Consequently, the identified signatures particularly the ‘Warburg-like’ switch seen in inactivated vaccines may not be fully generalizable to all technology platforms. While amino acid mobilization appears to be a conserved feature of general immune activation, it remains unclear whether the specific lipidomic shifts are platform-dependent or universal. Future research should fill this gap to determine a truly ‘universal’ metabolic code of vaccination.

Further prospective studies are needed combining metabolomics with proteomics and transcriptomics (which is already partially happening) to create a more complete picture of the molecular mechanisms of the immune response to vaccination. Prospective validation of biomarkers such as the Kyn/Trp ratio or baseline ceramide/amino acid levels for practical clinical use in identifying individuals at risk of poor response (Low Responders). Confirmation that metabolic changes induced by vaccination are transient and do not lead to permanent metabolic disorders.