NMR-Based Metabolomic Analysis of Serum and Interstitial Fluid from Patients with Breast Cancer-Related Lymphedema

1. Introduction

Despite the advanced diagnostic capabilities of modern medicine, certain diseases and dysfunctions in the human body remain challenging to diagnose and treat early on. Often, this is due to a limited understanding of the underlying mechanisms driving the formation and progression of the pathology, leading to restricted therapeutic options. Breast cancer (BC) ranks among the most prevalent cancers globally, while recent advancements in diagnosis and treatment have improved survival rates; however, the survival is associated with an increased risk of complications. One of the most severe complications is breast cancer-related lymphedema (BCRL) [1], which significantly diminishes quality of life for affected individuals [2]. The development of BCRL before the appearance of visible symptoms can last from several months to up to 10 years post-surgery, affecting approximately 20-49% of breast cancer survivors [3].

Lymphedema or lymphoedema is a condition characterized by regional lymphatic dysfunction due to impaired lymphatic fluid flow or its complete absence. The disease progression is accompanied by chronic inflammation and results in gradual accumulation of protein-rich fluid in the interstitial space. This results in limb swelling, as well as fibrotic changes in the skin and subcutaneous tissues due to compromised lymphatic system function [4].

Current secondary lymphedema treatment options, including both conservative and surgical approaches, primarily address symptoms, slow disease progression, and mitigate complications. However, these treatments often fail to substantially improve the quality of life for patients [5]. Therefore, there is an urgent need for novel and effective early diagnostic and therapeutic strategies for BCRL.

A most promising approach to tackling these challenges involves the use of high-throughput omics technologies, particularly metabolomics [6]. Various omics investigate different molecular levels of tissue functioning organization, encompassing the genome, transcriptome, proteome, metabolome, lipidome, etc. While the genome represents the most stable molecular level, the metabolome stands out as the most dynamic and responsive to changes caused by numerous internal and external factors. Moreover, metabolomic profile exerts a profound influence on a tissue physiology. The concentrations of metabolites in tissues or fluids are typically regulated within narrow ranges by their roles in metabolic processes. However, the onset of pathological conditions, such as malignant tumors and postoperative complications, can result in significant alterations in both the metabolic cycles of affected tissues and the blood metabolome [7].

The correct collection of biosamples for metabolomic purposes is a crucial step, performed even before the initial treatment of samples, such as metabolic activity quenching or extraction of metabolites. While blood sampling is a routine clinical procedure, sampling of interstitial fluid (ISF) presents a challenge. Recently, we developed and tested a novel ISF sampling method [8], which was employed in this study and allowed for the successful acquisition of this valuable material from BCRL patients.

In the current study, we employed a quantitative metabolomic approach based on the nuclear magnetic resonance (NMR) to investigate changes in the human lymphatic system's function associated with BCRL progression. The objectives were to identify metabolomic biomarkers of lymphedema in the ISF and blood serum of patients who had undergone BC treatment and to elucidate the molecular mechanisms underlying lymphedema onset and progression. This research aims to deepen insights into lymphatic system function and offers potential for the development of promising and minimally invasive methods for early BCRL diagnosis and personalized treatment.

2. Materials and Methods

2.1. Materials

Methanol and chloroform (HPLC grade) from J.T. Backer (Radnor, USA), 99.9% D₂O from Astrachim (St. Petersburg, Russia), sodium 4,4-dimethyl-4-silapentane-1-sulfonate (DSS) from Cambridge Isotope Laboratories (Tewkesbury, USA) were used as received. To prepare 50 mM deuterated phosphate buffer, monosodium phosphate dihydrate and disodium phosphate from Applichem (Darmstadt, Germany) were dissolved in D₂O and vacuum dried overnight to replace hydrogen atoms with deuterium. The procedure was repeated twice. H₂O was deionized using an Aqualab-AL-1 Plus water system (Aqualab, Moscow, Russia) to the quality of 18.2 MΩ*cm.

2.2. Sample Collection

The investigations were conducted in accordance with the principles outlined in the Declaration of Helsinki 2008 (https://www.wma.net/what-we-do/medical-ethics/declaration-of-helsinki/doh-oct2008/, accessed on 01.08.2024), which delineate ethical principles for medical research involving human subjects. Ethical approval was obtained from the International Tomography Center SB RAS (#ECITC-2020-12 from 16.12.2020) and the Research Institute of Clinical and Experimental Lymphology – Branch of the Institute of Cytology and Genetics SB RAS (#159 from 20.07.2020). Written informed consent was obtained from all patients after a thorough explanation of the study's nature and potential consequences. No special permission from national or local authorities was required.

Blood (n=22) and interstitial fluid (n=9) samples were collected from patients who had previously undergone breast cancer treatment and who subsequently developed and were diagnosed with BCRL (n=22, age 69.6 ± 3.4 years). All patients were in remission and experienced no recurrence of BC. Eight patients had a prior history of erysipelas (Table S1). More than a year had passed since the last episode of BC relapse.

Control blood samples (n=12, age 63.7 ± 6.5 years) were collected from age-matched patients without diagnosed breast cancer and without lymphedema (Table S1). Peripheral blood was obtained from the ulnar vein in the morning after overnight fasting under aseptic conditions into a vacutainer containing a coagulation activator. Within 10-15 minutes of collection, the blood samples were centrifuged (3000×g, 10 min), and the plasma obtained was transferred into separate Eppendorf vials.

Interstitial fluid from the affected arm from the patients with secondary lymphedema was collected by the method described earlier [8]. Briefly, an ultrasound examination of the affected limb was performed using a 5 MHz and 10 MHz linear array probe. Areas of interstitial fluid accumulation were identified and marked. At these areas, the skin was punctured with a 27G needle and the interstitial fluid that appeared at the puncture site was collected using a 22G plastic catheter and transferred into clean Eppendorf vials. The vials with biofluid samples were immediately frozen and stored at -70°C until analysis.

2.3. Sample Preparation

For the preparation of extracts from blood serum for metabolomic measurements, we adhered to a standard protocol described in detail previously [9,10,11]. Briefly, after thawing human blood plasma, the clot was removed from the vial, yielding blood serum. Subsequently, 300 μL of H₂O, 600 μL of cold (-20 °C) methanol and 600 μL of cold (-20 °C) chloroform were added to 300 µL of serum. The ISF volume varied from 50 to 200 μL, and the volume of added water was adjusted to match the ISF volume in each sample. Twice the volume of cold (-20 °C) methanol and cold (-20 °C) chloroform was added to each ISF sample. The resulting mixtures were stirred on a shaker at +4 °C for 15 minutes, followed by incubation at -20 °C for 30 minutes. Subsequently, the mixtures were centrifuged at 16,100×g for 30 minutes at +4 °C. Centrifugation resulted in the separation of a mixture into two immiscible phases. The upper water-methanol phase was collected and vacuum dried overnight.

2.4. NMR Measurements

The dried extracts were dissolved in 600 µL of 50 mM of deuterated phosphate buffer (pH 7.2) containing 20 µM of sodium 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS) as an internal standard. The ¹H NMR measurements were conducted at the Center of Collective Use "Mass Spectrometric Investigations" SB RAS, using an AVANCE III HD 700 MHz NMR spectrometer (Bruker BioSpin, Rheinstetten, Germany), equipped with 16.4T magnet and corresponding proton Larmour frequency of 700 MHz. NMR spectra for each sample were acquired in a standard 5 mm glass NMR tube using a 5 mm TXI ATMA NMR probe. We used 70-degree detection pulse (pulse sequence name zgpr) and summed 64 free induction decay (FID) signals while maintaining the sample temperature at 25°C. For two ISF samples with volumes below 100 µL, we collected 128 FIDs to improve the signal-to-noise ratio. Prior to acquisition, low-power radiation was applied at the water resonance frequency to presaturate the water signal. To allow for the relaxation of all spins, a repetition time of 20 seconds was used between scans.

The resulted dataset is available at the NIH Common Fund's National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench (https://www.metabolomicsworkbench.org, accessed on 03.10.2024), where it has been assigned Project ID ST003506 (http://dx.doi.org/10.21228/M8FR6S, accessed on 22.10.2024). This dataset is also available in AMDB repository by the following link: https://amdb.online/amdb/experiments/291/ (accessed on 03.10.2024). The dataset contains NMR raw data, sample descriptions, metabolite concentrations, and experimental protocols.

2.5. Identification and Quantification of Metabolites in NMR Data

Metabolite identification was conducted by analyzing their NMR spectra, which were sourced from literature, databases (HMDB, METLIN, BMRB, and SpectraBase), an in-house NMR library [9,10,12], and AMDB [13]. Baseline processing, identification, and integration of spectral NMR peaks (quantification) were performed using MestReNova v12.0 (Mestrelab Research, A Coruna, Spain). In cases where NMR signal assignment was not straightforward, the identification of metabolites was verified by spiking extracts with commercially available standard compounds. Despite these efforts, several signals in the NMR spectra remained unassigned. Metabolite concentrations in the samples were determined in µM by integrating NMR signals relative to the internal standard DSS, and then normalized to the biofluid volume. On average, 60-80 compounds were identified in the samples. However, the quantification of some compounds was unreliable due to weak signals or signal overlap. The final table contains only 48 reliably identified and quantified metabolites. The raw data may contain additional information on yet unidentified metabolites.

2.6. Data Analysis

Univariate and multivariate statistical analyses, biomarker validation, and pathway analysis of an ¹H NMR-based metabolomics dataset were performed on the MetaboAnalyst 6.0 web platform (www.metaboanalyst.ca, accessed on 01.08.2024) with autoscaled data. A data overview was performed with principal component analysis (PCA), and it was plotted for three sample groups. Pairwise comparisons were then used to identify significant differences between two corresponding groups. Univariate analysis for detecting differences between two groups was based on a volcano plot, combining fold change (FC) values with a threshold of FC > 1.5 and p-values from a statistical significance test (Mann-Whitney U-test) with false discovery rate (FDR) correction, and a threshold of p < 0.05. Differentiating metabolites in multivariate statistics were identified by orthogonal partial least squares-discriminant analysis (oPLS-DA) using the variable importance in projection (VIP) score, with a VIP-score threshold ≥ 1. Potential biomarkers were examined and validated with the area under the curve (AUC) measure from receiver operating characteristic (ROC) method, employing Monte Carlo cross-validation (MCCV), which is used to evaluate the accuracy and precision of biomarkers. Multivariate ROC curve exploratory analysis was performed using a linear support vector machine classifier and feature ranking methods. Quantitative Metabolite Set Enrichment Analysis (QMSEA) was performed on 80 metabolite sets from the KEGG human metabolic pathways database (data from December 2023) using the globaltest package to test for associations between metabolite sets and outcomes.

3. Results

3.1. Metabolite Concentration in Samples

Using high-field (high-resolution) ¹H NMR spectroscopy, we analyzed blood serum and interstitial fluid specimens collected from patients with BCRL and control subjects. In total, we identified and quantified 48 most abundant metabolites across the samples. Table 1 provides the average concentrations (in µM) and standard deviations for these metabolites in each group, while Table S2 lists the concentrations for each individual sample. A representative ¹H NMR spectrum for interstitial fluid, with corresponding metabolite identification are presented in Figure 1.

We applied principal component analysis (PCA) to overview the quantitative metabolomic data obtained for serum and ISF samples from BCRL patients and controls in order to establish major similarities and differences between the samples. The results are presented in Figure 2.

PCA shows clear separation between the BCRL and control samples, while serum and ISF samples from the BCRL patients are not separated.

3.2. Comparison of Serum and ISF Metabolomic Composition

To analyze minor differences between serum and ISF, we performed multivariate oPLS-DA and volcano plot analyses. The results are presented in Figure 3.

Statistically significant differences in metabolite concentrations between serum and ISF of BCRL patients were observed in the volcano plot (Figure 3C) for: inosine, ketoleucine, tryptophan, lactic acid, 3-methyl-2-oxovaleric acid, 2-ketoisovaleric acid, phenylalanine, 2-hydroxy-3-methylbutyric acid, 2-hydroxybutyric acid, acetic acid, and dimethyl sulfone. All these metabolites, except dimethyl sulfone, were also highlighted by multivariate oPLS-DA as having the highest VIP-scores (Figure 3B). The oPLS-DA plot (Figure 3A) demonstrates a clear separation of both groups and corresponding samples along the T-score axis, which represents between-group variation. Additional metabolites identified by oPLS-DA as differentiating between groups included creatinine, serine, formic acid, proline, and citric acid. Notably, nearly all metabolite concentrations were higher in serum than in ISF.

Additionally, we made a pairwise comparison of metabolite concentrations in serum and ISF (Figure S1). For each patient, the ratio of metabolite concentrations in serum to ISF was calculated, and then ratios were averaged across patients. The highest mean serum/ISF ratio (3.6) was found for ascorbic acid, followed by 2-hydroxy-3-methylbutyric acid (2.8), 3-hydroxybutyric acid (2.2), ketoleucine (2.0), and tryptophan (2.0). The lowest ratio was found for 3-hydroxyisobutyric acid (0.45). Inosine was below the limit of detection (LOD) in ISF, while pyroglutamic acid was not detected in serum, but showed high variability of concentrations in ISF. Most other metabolites displayed ratios between 0.97 and 1.23 (Figure S1). The overall mean ratio across all metabolites was 1.27 ± 0.53, and a median value was 1.08.

3.3. Comparison of Serum Metabolomic Composition from BCLR Patients and Controls

To achieve the primary objective of the current study – identifying differences between serum samples from BCRL patients and controls – we performed multivariate oPLS-DA and volcano plot analyses. The results of these analyses are presented in Figure 4.

The most pronounced differences between serum samples from BCRL patients and controls, as shown in the volcano plot (Figure 4C), were found for pyruvic acid, 2-ketoisovaleric acid, ketoleucine, 3-methyl-2-oxovaleric acid, tryptophan, proline, creatine, isobutyric acid, propylene glycol, ascorbic acid, and formic acid. Among these, all except propylene glycol and ascorbic acid are also highlighted by multivariate oPLS-DA as metabolites having highest VIP-scores (Figure 4B). In the oPLS-DA scores plot (Figure 4A), both groups and corresponding samples are well separated along the T-score axis, representing between-group variation. Additional metabolites identified by oPLS-DA as differentiating between groups included isoleucine, leucine, threonine, creatinine, glycerol, and valine. Concentrations of all these compounds are increased in the serum of BCRL patients.

3.4. Biomarker Analysis and Analysis of Affected Metabolic Pathways in BCRL

An initial analysis of potential biomarkers indicating the difference between serum samples from BCRL patients and the control group was performed using univariate ROC curve analysis for individual metabolites. The results (Table 2) reveal that four metabolites achieve an AUC metric of 1, while seven compounds exhibit AUC values greater than 0.8. A full table of the univariate biomarker analysis is provided in Table S3. These 11 compounds match the differential compounds indicated in prior analyses by the oPLS-DA and the volcano plot.

Therefore, these compounds can be considered promising candidates for the role of lymphedema biomarkers. To verify this, we performed a multivariate biomarker analysis using a combination of the top five metabolites identified though univariate ROC analysis and ranked by AUC: 2-ketoisovaleric acid, ketoleucine, pyruvic acid, 3-methyl-2-oxovaleric acid, and tryptophan. The results of the ROC-AUC analysis, presented in Figure 5 demonstrate excellent performance, with an AUC of 1 and zero misclassified samples. The combination of all 11 metabolites also shows an excellent AUC measure (Figure S2).

To determine the metabolic pathways altered in BCRL, we performed quantitative metabolite set enrichment analysis (QMSEA), with the results presented in Figure 6. Table 3 provides the statistical characteristics of the most affected metabolic pathways, indicating the significance and reliability of observed changes, as well as the metabolites from our dataset associated with these metabolic pathways. The complete list of analyzed pathways and corresponding QMSEA characteristics is available in Table S4.

To visualize the variability in metabolite concentrations in the groups, we prepared boxplots for differentiating metabolites found in this study (Figure 7): 2-ketoisovaleric acid, ketoleucine, pyruvic acid, 3-methyl-2-oxovaleric acid, tryptophan, isoleucine, proline, creatine, isobutyric acid.

4. Discussion

Analysis of quantitative data obtained by NMR method on concentrations of the most abundant metabolites indicates that serum and ISF in BCRL patients have a rather similar metabolomic composition: concentrations of the majority of metabolites (Table 1) are comparable, although nearly all compounds exhibit slightly higher concentrations in serum compared to ISF (Figure 3C). PCA analysis (Figure 2) supports this observation showing no clear separation between the serum and ISF sample groups. The most pronounced statistically significant differences between serum and ISF revealed by a combination of univariate and supervised multivariate statistics were (oPLS-DA) observed for inosine, ketoleucine, tryptophan, lactic acid, 3-methyl-2-oxovaleric acid, 2-ketoisovaleric acid, phenylalanine, 2-hydroxy-3-methylbutyric acid, 2-hydroxybutyric acid, acetic acid, creatinine, serine, formic acid, proline, and citric acid. The metabolomic composition of ISF may differ from that of serum due to the disruption of ISF outflow into the lymphatic vessels in secondary lymphedema, accompanied by altered biochemical processes in the extracellular matrix and chronic inflammation. Assuming that under normal conditions the metabolomic compositions of serum and ISF are in dynamic equilibrium [14,15], it can be concluded that the increased generation of these metabolites in the blood or their elevated consumption in ISF leads to a partial disruption of this equilibrium. The implications of the increased metabolite generation are discussed below.

The most pronounced differences in the serum of BCRL and control patients, identified through a combination of univariate and multivariate statistics (Figure 4), were observed for pyruvic acid, 2-ketoisovaleric acid, ketoleucine, 3-methyl-2-oxovaleric acid, tryptophan, proline, creatine, isobutyric acid, propylene glycol, ascorbic acid, formic acid, isoleucine, leucine, threonine, creatinine, glycerol, and valine; all concentrations are increased in BCRL patients. Biomarker analysis using the ROC-AUC method further supports the differences in 11 of these metabolites, demonstrating high diagnostic significance (AUC > 0.8, Table 2).

It should be noted that the magnitude of the increase in the differential metabolite concentrations in BCRL serum, compared to control serum, was noticeably greater than the magnitude of differences observed between BCRL serum and BCRL ISF, as well as the differences between control serum and BCRL ISF (Figure 7, Figure S3). This suggests that the observed metabolomic changes primarily occur in the blood rather than in ISF. Subsequently, these alterations in the blood spread to ISF [14], resulting in less pronounced changes in metabolite concentrations in ISF. Unfortunately, direct comparison of ISF from BCRL and control groups was not possible, due to study limitations related to the current difficulties of obtaining ISF from control patients and from unaffected arms of BCRL patients. Nevertheless, the proposed interpretation of the observed alterations in metabolite concentrations among the sample groups appears to be the most plausible. Therefore, blood serum emerges as a promising fluid for the development of early lymphedema diagnostics additionally to ISF. Moreover, the ease of collecting blood samples compared to ISF could expedite further research on BCRL.

The pathogenesis of lymphedema involves chronic inflammation, increased adipose tissue, interstitial fluid accumulation, and subsequent tissue fibrosis leading to the development of the disease. Based on the results of the enrichment analysis, we assessed the significance of biochemical pathways in the development of lymphedema. It should be noted that our study had another limitation due to the lack of data from patients who had undergone BC treatment but did not develop BCRL. However, a few recent studies have reported changes in the blood plasma metabolome following surgical and chemotherapy treatment of BC [16,17]. Pathways altered after 1 month include purine metabolism, galactose metabolism and several others [16]. One year after BC treatment changes in lysine degradation, branched-chain amino acids (BCAA) metabolism, linoeic acid metabolism, and tyrosine metabolism have been observed [17]. In the current study, most patients undergone chemotherapy and surgery more than one year prior to sample collection (Table S1), and all patients showed no signs of BC recurrence.

Most of the enriched metabolic sets listed in the Table 3 characterize the activation of energy production pathways. The processes of BCRL development are accompanied by activation of the defense mechanisms required by the body for an effective response to the disease. Usually, energy deficiency manifests itself in an increase in the content of intermediates of the tricarboxylic acid cycle, protein metabolism and fatty acid oxidation. In our case, the processes related to the TCA cycle, the degradation of valine, leucine and isoleucine (BCAA), the metabolism of other amino acids, pyruvate, and lipoic acid are noted. These processes indicate the activity of glycolysis or gluconeogenesis in the body (which is also in the QMSEA list, Table 3), and they are involved in replenishing the energy needs of the body to fight the disease. It is worth noting the activation of the synthesis of pantothenate and coenzyme A, which are important substrates of the metabolism of carbohydrates, proteins and fatty acids, complements the hypothesis of activation of energy production processes [18].

In addition to the energy role, metabolic processes involving glutamine may reflect increased activity of immune cells, since glutamine is used as an energy source by immune cells – lymphocytes, neutrophils and macrophages [19]. Branched-chain amino acids have been shown to play a signaling role in regulatory T-cells, which are a subset of immune cells that, among other activities, perform immunosuppressive functions [20].

The metabolism of methylating agents – serine, glycine, betaine, and methionine – affects key biochemical processes. It is particularly interesting that hypermethylation, according to previous studies, is associated with shifting immune responses toward a T-helper 2 (Th2) proinflammatory response, and also contributes to an increase in cytokine production, thereby generally intensifying inflammation [21,22]. It was described earlier [4] that the inflammatory response in secondary lymphedema typically involves upregulated expression of various cytokines and chemokines, as well as an increase in the accumulation and activation of immune cells. In a mouse model, an increased number of CD4+ T cells and increased Th2 differentiation were observed [23]. Th2 cells play a significant role in the pathogenesis of lymphedema, as inhibition of their differentiation has been shown to reduce fibrosis and improve lymphatic function [23,24]. Thus, information on altered in the metabolism of methylating agents may be used for the development of subsequent studies on BCRL.

The data obtained demonstrate the potential for further search and verification of biomarkers though targeted monitoring of metabolites in these pathways on a larger cohort, enabling a more reliable assessment of deviations in the biochemical processes involving these metabolites.

5. Conclusions

Based on the results of the statistical analysis of ¹H NMR quantitative data obtained from serum and ISF samples from BCRL and control patient groups, we made the following observations. Serum and ISF display a rather similar metabolomic composition, though concentrations of nearly all compounds are slightly higher in serum compared to ISF. Significant differences in metabolite concentrations between the serum of BCRL and control patients were identified using multiple statistical methods. Differential metabolites that may serve as potential biomarkers for lymphedema are 2-ketoisovaleric acid, 3-methyl-2-oxovaleric acid, ketoleucine, pyruvic acid, tryptophan, proline, creatine, isoleucine, isobutyric acid, leucine, and creatinine. The most significantly altered metabolic pathways, as evaluated by enrichment analysis, include “Arginine and proline metabolism”, pathways of branched-chain amino acids, “Pantothenate and CoA biosynthesis”, “Tryptophan metabolism”, “Glycolysis/Gluconeogenesis”, “Glycine, serine and threonine metabolism”, “Alanine, aspartate and glutamate metabolism”, “Glyoxylate and dicarboxylate metabolism”, all of which are associated with the activation of energy production and inflammation.