Haptoglobin phenotypes and structural variants associate with post-exertional malaise and cognitive dysfunction in myalgic encephalomyelitis

Abstract

Background

Myalgic encephalomyelitis (ME) is a chronic, multisystem illness characterized by post-exertional malaise (PEM) and cognitive dysfunction, yet the molecular mechanisms driving these hallmark symptoms remain unclear. This study investigated haptoglobin (Hp) as a potential biomarker of PEM severity and cognitive impairment in ME, with a focus on Hp phenotypes and structural proteoforms.

Methods

A longitudinal case–control study was conducted in 140 ME patients and 44 matched sedentary healthy controls. In the discovery phase, global plasma proteomic profiling was performed in 61 ME patients and 20 controls before and after a standardized, non-invasive stress protocol in order to induce PEM. Associations between Hp levels, phenotype, and cognitive performance were assessed. In the validation phase, plasma Hp concentrations and proteoform composition were analyzed in an independent cohort of 89 ME patients and 24 controls using high-performance liquid chromatography (HPLC).

Results

ME patients demonstrated a significant reduction in Hp levels following post-exertional stress. Lower baseline Hp concentrations were associated with impaired cognitive performance. Hp phenotypes were differentially associated with symptom burden, with the Hp2-1 phenotype enriched in ME and linked to greater PEM severity and cognitive deficits compared to Hp1-1 and Hp2-2. HPLC analysis revealed altered Hp proteoform profiles in the Hp2-1 subgroup, including increased high-mass tetrameric and pentameric forms and shorter retention times indicative of structural changes. In contrast, the Hp1-1 phenotype was associated with milder symptoms and greater cognitive resilience.

Conclusions

These findings suggest that Hp phenotype and proteoform structure modulate the physiological response to post-exertion in ME, offering insight into the molecular basis of PEM and its clinical heterogeneity. Hp may serve as a translational biomarker for patient stratification and a potential therapeutic target to mitigate oxidative stress and cognitive dysfunction in ME.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-025-07006-z.

Background

Myalgic encephalomyelitis (ME) is a chronic, disabling disease characterized by profound fatigue, unrefreshing sleep, and cognitive impairment. The hallmark symptom, post-exertional malaise (PEM), involves exacerbation of symptoms following minimal exertion (physical or mental) and lacks a clear mechanistic explanation [1, 2]. Diagnostic challenges are compounded by overlapping symptoms with conditions such as fibromyalgia [3].

Emerging evidence indicates that vascular and hematological abnormalities, particularly impaired red blood cell deformability observed in ME, may contribute to post-exertional malaise (PEM), potentially through mechanisms involving hemolysis-induced oxidative stress [4–6]. Haptoglobin (Hp), a key hemoglobin-binding protein, plays a critical role in mitigating oxidative damage under hemolytic condition [7]. In addition to its direct interaction with hemoglobin (Hb) through the CD163 receptor–mediated clearance pathway, Hp also functionally collaborates with the LRP-1–hemopexin axis in heme detoxification [8, 9]. While the Hp–CD163 pathway primarily mediates Hb clearance, Low-density lipoprotein receptor-related protein 1 (LRP-1) plays a key role in the uptake of heme bound to hemopexin. Hp exists in two allelic forms—Hp1 and Hp2—producing three isoforms (Hp1-1, Hp2-1, and Hp2-2). Structurally, Hp 1-1 forms dimers consisting of two α1β units, connected by disulfide bonds. Hp 2-2 exists as cyclic α2β oligomers, while Hp 2-1 forms a mixture of linear oligomers of α1β and α2β units. Notably, Hp genetic polymorphism significantly influences its quaternary structure, and different Hp genotypes exhibit varying affinities for binding hemoglobin (Hb) [10, 11]. The Hp2 allele, associated with lower antioxidant capacity, has been linked to worse clinical outcomes in other oxidative stress-related conditions [12, 13].

Despite significant advances in ME research, few studies have addressed the role of Hp in PEM or its impact on cognitive dysfunction. This study investigates the relationship between Hp levels, phenotypes/genotypes, PEM severity, and cognitive impairment in ME patients using a novel non-invasive stress protocol.

Methods

Experimental design and study population

This longitudinal case–control study involved a discovery phase conducted before the COVID-19 pandemic and a validation phase carried out thereafter. The overall experimental design is summarized in Fig. 1. The demographic and clinical data of the participants at the discovery and validation phases are described in Table 1 and Supplementary Table S1, respectively. This study was approved by the Institutional Review Board of CHU Sainte-Justine (protocol # 4047) and was conducted in accordance with the guidelines of human ethics regulations. Signed informed consent was obtained from all participants.

Fig. 1.

Graphical summary of the experimental design of the study. ME myalgic encephalomyelitis, HC healthy controls

Table 1.

Clinical and demographic characteristics of study participants—discovery phase

	ME (N = 61)	HC (N = 20)
Proteomic cohort
Age (years)	51.7 ± 1.5	47.9 ± 2.5
Sex (male/female)	20/41	11/9
BMI	25.0 ± 0.6	25.0 ± 0.9
Ilness duration	13.6 ± 1.7	N/A
36-Item Short Form Health Survey (SF36)
Physical score	31.0 ± 1.6***	92.0 ± 1.2
Mental score	47.0 ± 3***	89.0
Multidimensional Fatigue Inventory-20 (MFI) Scores
General fatigue	18.0 ± 0.3***	6.0 ± 0.5
Physical fatigue	18.0 ± 0.3***	5.0 ± 0.4
Reduced activity	16.0 ± 0.4***	6.0 ± 0.5
Reduced motivation	10.0 ± 0.4***	6.0 ± 0.4
Mental fatigue	15.0 ± 0.5***	7.0 ± 0.5
DePaul Symptom Questionnaire (DSQ) Scores
Neuroendocrine, Autonomic and Immune Dysfunction score	40.0 ± 2.2***	6.0 ± 1.0
Cognitive Dysfunction score	56.0 ± 2.5***	11.0 ± 2.0
Post-exertional malaise score (PEM)	69.0 ± 2.4***	7.0 ± 1.0
Sleep Disturbance score	49.0 ± 2.2***	14.0 ± 2.0

Values for the different SF-36, MFI-20 and DSQ categories are described as scores. All data are represented as mean ± standard error of the mean. A two-tailed Student's t-test was used for the comparison between ME/CFS patients and healthy controls. The questionnaire scores were significantly different between ME vs. HC. ***p-value < 0.001

ME symptoms evaluation and participant health status

All participants underwent a comprehensive health status assessment using three validated self-reported questionnaires before undergoing the stress test. The Short Form 36-Item Health Survey (SF-36) was used to assess general mental and physical health [14], while the Multidimensional Fatigue Inventory-20 (MFI-20) evaluated fatigue across five categories: General Fatigue, Physical Fatigue, Reduced Activity, Reduced Motivation, and Mental Fatigue [15]. The sum of the MFI-20 categories provided a fatigue severity score (score range of 0–100), allowing patients to be stratified into two groups: mild-moderate (score of 51–85), and severe (score of 86–100). The DePaul Symptom Questionnaire (DSQ) captured the core symptoms of ME across four domains: Neuroendocrine, Autonomic, and Immune Dysfunction; Cognitive Dysfunction; Post-Exertional Malaise (PEM); and Sleep Disturbances [16].

Post-exertional stress challenge

PEM, the hallmark symptom of ME, serves as a key criterion for distinguishing ME from other overlapping conditions. The application of a standardized provocation maneuver specifically designed to induce PEM offers a unique opportunity to uncover the molecular mechanisms driving PEM and its relationship to symptom severity. By focusing on PEM as the central feature, this approach minimizes the confounding effects of factors such as polypharmacy, illness duration, and sex differences commonly associated with ME. This ensures a more reliable and controlled investigation into the underlying biological processes of the disease. To safely induce PEM in ME patients, including those who are severely ill and housebound, we employ a post-exertional stress challenge using a therapeutic massager device (ABR therapeutic massager device, Panacis Medical Ltd., Ottawa, ON, Canada). This innovative approach, previously described in our earlier studies, utilizes an inflatable cuff applied to the participant’s arm, delivering pulsatile compressions with variable amplitude (0–4 psi at 0.006 Hz) [17]. Unlike traditional PEM-inducing methods such as the cardiopulmonary exercise test (CPET), the ABR method is non-invasive, requires no active effort from participants, and has been demonstrated to facilitate the safe inclusion of very sick ME patients (housebound individuals). PEM severity was further evaluated using a modified version of the DePaul Post-Exertional Malaise Questionnaire (DPEMQ), which was administered by a clinical research nurse five days after the stress test [18]. This version focused on the frequency and severity of symptoms during the immediate post-exercise period rather than over the previous six months. Thirteen items contributed to the total DPEMQ score, with higher scores indicating greater PEM severity. This systematic approach provided a comprehensive assessment of symptom burden and its relationship to study findings. DPEMQ was also used to evaluate PEM symptoms in patients of different Hp phenotypes.

Longitudinal cognitive assessment

Participants underwent cognitive assessments at two time points including at baseline (T0), and immediately after PEM induction using a standardized provocation maneuver (post-exertional stress-test) of 90 min (T90), using the BrainCheck test (Braincheck, Austin, TX, USA), a digital platform designed to evaluate and monitor cognitive function over time. A standard BrainCheck evaluation includes assessments of memory (immediate and delayed recognition) as well as cognitive processing and executive function. These assessments include the Digit-Symbol Substitution task, Trail-Making Test A and B (which evaluate mental flexibility and attention), and the Stroop Interference task. Upon completion, the platform provides standard scores and percentile ranks for each individual test. To mitigate potential learning effects associated with repeated cognitive testing, participants completed a practice session prior to the baseline assessment. This session featured shortened versions of the cognitive tasks, allowing participants to familiarize themselves with the format, timing, and question types of the BrainCheck platform, while also reducing test-related anxiety and minimizing potential confounding from repeated exposure.

Blood sample collection and plasma sample preparation

Peripheral blood samples were taken from all the participants and collected at two time points in EDTA-coated tubes, one at baseline (T0) and the other one 90 min after the application of the post-exertional stress test (T90) for a total of 162 blood samples. Plasma samples were obtained after centrifugation at 216 × g for 10 min and stored at − 80 °C until use.

Chemicals

Trifluoroacetic acid (TFA, ≥ 99.0%) and ammonium bicarbonate (AmBic) were obtained from Sigma Aldrich (St Louis, MO, USA). Iodoacetamide (IAA, 98%) was purchased from Acros Organics (Geel, Belgium). Dithiothreitol (DTT) was supplied by PanReac AppliChem (Chicago, IL, USA). Sequencing Grade Modified Trypsin was purchased from Promega (Madison, WI, USA). Acetonitrile (ACN) was supplied by Merck (Darmstadt, Germany). Pierce™ HeLa Protein Digest Standard, formic acid (FA, 99 + %), and urea, as well as Pierce™ Water (LC–MS Grade), 0.1% Formic Acid in 80% Acetonitrile (Optima® LC/MS) and Flush solution MB124 for the LC–MS separation (45% 2-propanol, 45% acetonitrile, 10% acetone) were obtained from Thermo Fisher Scientific (Waltham, USA). Buffer A (Equilibrate/Load/Wash) and Buffer B (Elution) solutions used for depletion of high abundant proteins were purchased from Agilent Technologies (Cedar Creek, TX, USA). Ultrapure water was prepared by Milli-Q water purification system (Millipore, Bedford, MA USA).

Plasma sample preparation for proteomic analysis

Plasma samples were chosen for preparation and further analysis in a random order. All the steps were carried out according to the Standard Operating Procedures of the MS Based Proteomics Facility at Uppsala University. The full list and the description of the samples can be found in Supplementary Table S7.

Depletion of highly abundant plasma proteins

In order to enrich the plasma samples with low abundant proteins for the subsequent LC–MS/MS analysis, fourteen of the most highly abundant proteins (human albumin, IgG, α1-antitrypsin, IgA, transferrin, haptoglobin, fibrinogen, α2-macroglobulin, α1-acid glycoprotein, apolipoprotein A-I, apolipoprotein A-II, IgM, transthyretin, and complement C3) were depleted using a Multi Affinity Removal Spin (MARS) Cartridge Human-14, 0.45 mL, (Agilent Technologies) according to the manufacturer’s instructions. Prior to the depletion, all the plasma samples were filtered to remove any particulates and to prevent possible blocking of the MARS cartridge as follows. Aliquots of 10 μL of plasma samples diluted with 200 μL of buffer A were loaded onto 0.22 μm Cellulose Acetate Spin Filters (Agilent Technologies) and centrifuged at 16 000 × g for 1 min in the accuSpin Micro-17 microcentrifuge (Thermo Fisher Scientific). After the filtration, a 190 μL aliquot of each sample was loaded on top of the MARS cartridge placed in a 2 mL Eppendorf tube. The cartridge was centrifuged at 100 × g for 1 min, followed by collection of the flow-through fraction. Two successive wash steps were carried out to ensure the maximum yield. A volume of 400 μL of buffer A was placed onto the cartridge, followed by centrifugation at 100 × g for 2 min after letting it sit for 5 min at room temperature. Flow-through fraction was collected, the MARS cartridge was placed into a new 2 mL Eppendorf tube, and 400 μL of buffer A were loaded to the cartridge, which was then centrifuged again at 100 × g for 2 min. Finally, the flow-through fractions were combined and dried using a vacuum concentrator (Eppendorf Concentrator PLUS, Hamburg, Germany) at 45 °C until they were completely dry. The dried pellets were stored at – 80 °C until the further sample preparation steps took place. After every depletion the MARS cartridge was washed with 4 mL of buffer B to elute the bound proteins and flushed with 4 mL of buffer A to re-equilibrate the resin.

Tryptic digestion of plasma proteins

The proteins were reduced, alkylated and digested in solution according to the following procedure. The dried pellets were dissolved in 50 μL of digestion buffer (6 M urea, 50 mM AmBic), sonicated using the ultrasonic unit Elmasonic S 10 (H) (Elma Schmidbauer GmbH, Singen, Germany) for 15 min, followed by addition of 70 μL of 50 mM AmBic to each sample. The samples were then shaken using the Heidolph Multi Reax shaker (Heidolph, Schwabach, Germany) for 50 min until the pellets were fully dissolved. At the next step, 10 μL of 45 mM DTT solution were added to every sample before the incubation at 50 °C for 15 min using the Eppendorf Thermomixer Comfort (2 mL block) (Eppendorf, Hamburg, Germany) to reduce the disulfide bridges. The mixtures were cooled to room temperature, and 10 μL of 100 mM IAA solution were added to each of them, followed by incubating the samples for 20 min in the darkness at room temperature in order to carbamidomethylate the cysteines. Finally, 20 μL of 0.1 μg/μL trypsin in 50 mM AmBic solution were added to every sample, followed by an overnight incubation at 37 °C.

Spin column sample clean up

Prior to the LC–MS/MS analysis the samples were purified and desalted using Pierce™ C-18 Spin Columns (Thermo Fisher Scientific). The columns were activated with 2 × 200 μL of 50% ACN and equilibrated with 2 × 200 μL of 0.5% TFA. A volume of 15 μL of 20% TFA solution was added to the samples before loading them onto the columns. In order to ensure the complete binding, three repeated loading cycles were used, followed by four washing steps with 200 μL of 0.5% TFA. Finally, the peptides were eluted with 2 × 50 μL of 70% ACN, the samples were dried at 30 °C in the vacuum concentrator and then stored at − 80 °C until they were analyzed. Centrifugation at 1500 × g for 1 min was used for all the steps of the purification procedure.

Nano LC–MS/MS analysis

The dried samples were reconstituted in 80 μL of 0.1% FA and analyzed using a Q Exactive™ Plus Orbitrap™ mass spectrometer (Thermo Scientific, Bremen, Germany) equipped with a nano-electrospray ion source. The peptides were separated by the reversed phase liquid chromatography using an EASY-nLC 1000 system (Thermo Scientific, Bremen, Germany) with a set-up consisting of a pre-column and an analytical column. The pre-column was an Acclaim™ PepMap™ 100 C18 column with the length of 20 mm, internal diameter (ID) of 75 μm and 3 μm particle size (PS) (Thermo Scientific, Lithuania), while the analytical column was an EASY-Spray™ PepMap™ Neo C18 column with the length of 150 mm, ID of 75 μm and 2 μm PS (Thermo Scientific, Lithuania). The injection volumes were 5 μL, and the separations were performed at a flow rate of 250 nL/min with mobile phases A (0.1% aqueous FA) and B (0.1% FA in 80% ACN). A 90 min gradient varying from 5 to 95% B was used for chromatographic separation. Every sample was run as a single replicate. The system was flushed with a Flush solution MB124 after every two samples using a 16 min gradient varying from 3 to 95% B. Analysis of 0.25 μg/μL HeLa Protein Standard using a 60 min gradient varying from 5 to 100% B, performed every 7 days after the mass spectrometer m/z axis calibration procedure, was used as the system quality control. The mass spectrometer was operated in a positive ion mode, acquiring survey mass spectra with the resolving power of 140,000, m/z scan range from 400 to 1700, using an automatic gain control (AGC) target of 3 × 10⁶ in a data-dependent acquisition mode. Thus, ten most intense ions were chosen for the further higher-energy collisional dissociation (HCD) fragmentation, and the MS/MS spectra were acquired with an AGC target of 1.6 × 10³ at the resolution of 17 500, m/z scan range from 200 to 2000.

Mass spectrometry data analysis

The acquired data (RAW-files) were processed in MaxQuant 2.0.3.0. The database searches were performed using the implemented Andromeda search engine and a FASTA database containing proteins from Homo sapiens proteome extracted from UniProt database (Release February 2022). The same search results were used both for the qualitative and quantitative analysis. The search parameters included trypsin for enzyme specificity; Carbamidomethyl (C) was set as the fixed modification, while Oxidation (M), Acetyl (Protein N-term), and Deamidation (NQ) were set as variable modifications. The search criteria for protein identification were set to at least one unique and two matching peptides. Label free quantification was applied for comparative proteomics. In order to make such comparison possible, all the sample preparation steps, as well as the further nano LC–MS/MS analysis, were carried out in as similar conditions as possible, including the use of the same cartridge for depleting all the samples and analyzing them using the same chromatographic column.

Non-invasive cerebral hemoglobin monitoring using MASIMO NIRS

Cerebral total hemoglobin (CHb) was continuously monitored non-invasively using the MASIMO Root® patient monitoring and connectivity platform, equipped with the O3™ Regional Oximetry Module (Masimo Corporation, Irvine, CA, USA). Each participant was fitted with two disposable O3™ adult sensors placed bilaterally on the forehead, covering the fronto-temporal region to optimize signal quality and minimize interference from ambient light. The MASIMO O3™ module utilizes Near-Infrared Spectroscopy (NIRS) technology to assess localized tissue hemoglobin saturation (rSO₂) and total hemoglobin concentration (CHb). The sensor emits multiple wavelengths of near-infrared light into the tissue, and detectors measure the reflected and absorbed light. By analyzing the absorption spectra, the system estimates the relative concentrations of oxygenated hemoglobin (HbO₂) and deoxygenated hemoglobin (HHb), which are then combined to derive CHb. This metric reflects the total blood volume within the monitored cerebral region. CHb data were acquired continuously at a sampling rate of one data point every 2 s throughout the experimental protocol, which included a baseline resting period, an exertional stress test, and a post-exertional malaise (PEM) induction phase. Sensor placement was standardized across participants to ensure consistency and reliability of the measurements. All data were stored on the MASIMO Root device and later exported for offline analysis. For quantitative analysis, mean CHb values were computed for predefined time windows. Baseline CHb was calculated as the average over a 5-min resting period preceding the 90-min stress test. Post-PEM CHb was determined by averaging values over a 5-min window immediately following the cessation of the PEM induction period.

Hp phenotyping

Hp phenotyping in the two cohorts was performed using two separate methods including Western blotting and HPLC.

Western blotting

Hp phenotyping was conducted using polyacrylamide gel electrophoresis followed by immunoblotting to identify α1 and α2 chains with different sizes, according to previously reported protocols [19]. Different Hp types were identified through Western blotting analysis, using plasma samples diluted in phosphate-buffered saline (PBS) at a ratio of 1:75. Next, diluted plasma samples were mixed with an equal volume of 2 × SDS sample buffer (Bio-Rad, Hercules, CA, USA) and boiled at 95 °C for 8 min. A 10 μl aliquot of the prepared sample was then loaded on 15% of polyacrylamide gel and electrophoresed for 150 min at 100 V (Bio-Rad). After transferring to the PVDF membrane, they were blocked with 5% of BSA in PBS with 0.02% tween for 1 h. The membranes were incubated overnight with monoclonal mouse antihuman haptoglobin Antibody (Santa Cruz Biotechnology, Dallas, TX, USA) selectively recognizing the alpha subunits of Hp diluted in a 1:1000 ratio in 5% BSA blocking buffer at 4 °C. After washing two times with PBST, the membranes were incubated with horseradish peroxidase (HRP)-conjugated goat anti-mouse IgG (Abcam, Cambridge, United Kingdom) in a 1:5000 ratio for 1 h at room temperature. The membranes were treated with western ECL substrate (Bio-Rad), and chemiluminescence was detected using the Imagelab chemidoc MP imaging system (Bio-Rad), following the final wash.

HPLC analysis and sample preparation

A volume of 20 µL of thawed patient plasma was combined with 20 µL of hemoglobin solution and vortexed. Then, 600 µL of mobile phase used for HPLC analysis was added to the mixture and vortexed. The obtained sample mixture was transferred into the Whatman Mini-UniPrep PVDF 0.45µm filtering vial, and 20 µL was injected into HPLC using the full loop mode.

Hemoglobin solution purification and preparation

One milliliter of fresh whole blood was mixed with Na₂EDTA at a final concentration of 2.2 mg/mL and centrifuged in a 15 mL conical polypropylene (PP) tube at 5000 × g for 3 min to separate erythrocytes. The plasma supernatant was discarded, and the erythrocyte pellet was reconstituted in 10 mL of TRIS buffer (0.15 M, pH 8), vortexed, and centrifuged again at 5000 × g for 3 min. The supernatant was discarded, and this washing step was repeated a total of ten times to thoroughly remove plasma components. The washed erythrocyte pellet was then reconstituted in 10 mL of 0.9% saline, vortexed, and centrifuged at 5000 × g for 3 min. After discarding the supernatant, the final erythrocyte pellet was reconstituted in 2 mL of distilled water and subjected to a freeze–thaw cycle at − 40 °C to induce complete erythrocyte lysis. The lysate was then centrifuged at 18,000 × g for 15 min to remove cell membranes. The resulting supernatant was carefully collected into separate tubes and stored frozen for further analysis.

HPLC measurements

Plasma samples prepared as mentioned above were analyzed using a Beckman System Gold HPLC (508 A/S, 166 UV–Vis detector and 126 solvent module). The sample tray was maintained at + 5 °C, and the chromatographic separation was performed with isocratic elution at a flow rate of 0.3 mL/min using a Biozen 1.8 µm SEC-3 column (300 × 4.6 mm) (Phenomenex, Inc., Torrance, CA, USA) operated at 30 °C. The UV–Vis detector was set at 410 nm. The mobile phase was prepared from 9 g ammonium acetate, 18 g potassium chloride, and 0.2 g sodium azide dissolved in 1 L of Milli-Q water (18.2 mΩ) adjusted to pH 6.9 with acetic acid. All reagents were obtained from Sigma-Aldrich. For hemoglobin samples, a Tris buffer (0.15 M, pH 8) was used, while plasma samples were diluted in the same buffer as the mobile phase. Saline (0.9% NaCl) and Milli-Q water were also used as needed.

Measurement of soluble LRP-1 (sLRP-1)

Plasma sLRP-1 levels were quantified using a sandwich enzyme-linked immunosorbent assay (ELISA) kit (Cat. Abx152232, Abbexa Ltd, Cambridge, UK), following the manufacturer’s protocol. All samples were analyzed in duplicate to ensure accuracy, and mean values were used for analysis.

Statistical analysis

In proteomics analysis, proteins labeled as contaminants and those matching the reversed database in the search results were removed from further calculations. For the quantitative analysis, the following comparisons were performed: ME samples vs. HC samples; ME samples after the stress test vs. HC samples after the stress test; ME samples versus (vs.) ME samples after the stress test (paired comparison); HC samples vs. HC samples after the stress test (paired comparison). Differential protein expression between two groups was assessed using Student’s two-tailed t-test applied to observed (non-missing) values, and nominal p-values were calculated. Imputation methods were avoided to minimize bias and improve the robustness of the analysis. To account for multiple hypothesis testing, p-values were adjusted using the q-value method [20] via the R qvalue package [21] which estimates the minimum false discovery rate (FDR) at which a test may be considered significant. This approach offers a balance between identifying true positives and controlling for false discoveries, making it particularly suitable for high-dimensional proteomics data with sparse signals. The questionnaire score results and other clinical characteristics were presented as mean ± SEM. The statistical differences between two groups were assessed using the Student’s t-test or the paired t-test for normally distributed data, and the Mann–Whitney test for non-normally distributed data. For comparisons involving three groups, one-way ANOVA followed by Tukey’s post hoc test or Holm-Šídák's correction was used for normally distributed data, while the Kruskal–Wallis test followed by Dunn’s multiple comparison test was applied to non-normally distributed data. p-values smaller than 0.05 were considered significant. Effect sizes were calculated to quantify the magnitude of observed group differences, using methods appropriate to the data’s distribution and structure. Cohen’s d was used for parametric comparisons between two groups, with thresholds of 0.2, 0.5, and 0.8 indicating small, medium, and large effects. Cohen’s f was applied, with values of 0.10, 0.25, and 0.40 corresponding to small, medium, and large effects. In non-parametric two-group comparisons, a rank-based effect size ranging from 0 to 1 was used, where higher values indicate greater group separation between groups. In non-parametric comparisons with three groups, a variance-based effect size ranging from 0 to 1 was calculated, interpreted similarly to eta-squared: 0.01 (small), 0.06 (medium), and 0.14 (large). Statistical analysis was performed using GraphPad Prism (version 8, GraphPad Software, Inc., San Diego, CA, USA).

Results

Study population characteristics

The discovery plasma proteomic cohort consisted of 81 participants, including 61 patients with ME and 20 sedentary healthy controls (HC) matched for age and sex (Table 1). Healthy controls were carefully selected to exclude individuals with any familial history of ME or related conditions, such as fibromyalgia (FM) or multiple sclerosis (MS). Patients classified as FM based on our previously described panel of miRNAs, were excluded from the study [12]. The validation cohort included 89 patients diagnosed with ME and 24 matched sedentary healthy controls (HC) (Supplementary Table S1). All patients with ME who were selected met the Canadian Consensus Criteria (CCC) and all participants were Caucasian (French-Canadian) individuals of European ancestry. In the discovery cohort, the ME group included 41 females and 20 males, while the HC group consisted of 9 females and 11 males. The validation phase comprised of 78 females and 11 males in the ME group and 16 females and 8 males in the HC group. There were no significant differences in average age or Body Mass Index (BMI) between both groups. Illness duration did not differ significantly between ME patients across the two cohorts (p > 0.5). All the questionnaire scores were significantly different between ME and HC groups. (p-value < 0.001).

Global plasma proteome profiles in ME patients and healthy controls

In the discovery phase, label-free quantitative proteomics was performed on 162 plasma samples collected from 61 individuals diagnosed with ME and 20 matched sedentary healthy controls. Samples were obtained at baseline (T0) and 90 min after a standardized passive stress test (T90) designed to induce PEM. Mass spectrometry analysis using MaxQuant identified a total of 846 proteins across all samples. The number of quantified proteins per sample ranged from 253 to 514, with an average of 359. Chromatographic separation provided a well-distributed peptide profile across the LC–MS gradient (Supplementary Figure S1), and detailed protein counts per sample are available in Supplementary Table S7.

Subtle group-level differences in the plasma proteome at baseline

To identify differences between ME and HC groups at baseline, two comparative strategies were applied. In a conservative approach that included only proteins consistently quantified across all 81 samples (61 ME and 20 HC), 101 proteins met the criterion of no missing values. Among these, no differentially expressed proteins met the threshold of multiple test adjusted significance (q-value < 0.05). Seven proteins showed nominal significance (p < 0.05), with two upregulated and five downregulated in the ME group. However, none of these seven proteins exceeded ± 1.5 fold-change (FC) (Supplementary Table S2). A more inclusive analysis, allowing up to 20% missing values, considered proteins quantified in at least 49 of 61 ME samples and 16 of 20 HC samples. This relaxed filtering increased the number of analyzable proteins to 173. Within this broader dataset, 11 proteins exhibited nominal significant differences between ME and HC (p < 0.05), including two upregulated and nine downregulated proteins; none passed multiple test correction (q-value < 0.05). Only one protein, the IgGFc-binding protein, showed a fold-change (FC) exceeding the ± 1.5 threshold (FC = 0.55). No protein was found to be exclusively present in all ME samples and entirely absent in all HCs, or vice versa, emphasizing the subtlety of proteomic distinctions between groups at rest.

Stress-induced proteomic responses reveal disease-specific patterns

Plasma samples collected after the stress test (T90) were analyzed to explore group-level differences post-exertion. As with the baseline comparisons, no protein was found to be uniquely present or absent in all individuals of either group. In the strictest comparison, limited to proteins quantified in all 81 post-stress samples, 107 proteins were included. Among these, none passed multiple test correction (q-value < 0.05). Three proteins showed nominal significant differences between ME and HC (p < 0.05), but none exceeded the ± 1.5 FC threshold. When the analysis was expanded to include proteins with up to 20% missing values, 176 proteins were analyzed, with five showing nominal significant differences between groups. Again, FC remained below ± 1.5 (Supplementary Tables S4 and S5; Supplementary Figure S3).

Within-group comparisons revealed more dynamic changes. In ME patients, a paired analysis of baseline (T0) and post-stress (T90) samples identified 97 proteins consistently quantified across both timepoints. Of these, 20 proteins showed nominal significant changes (p < 0.05), including 15 that were upregulated and 5 downregulated following stress (Table 2; Fig. 2). Using thresholds of FC ≥ 1.5 and q-value < 0.05, two proteins, ApoA1 and Hp, were identified as significantly altered (Supplementary Table S6). In the HC group, 118 proteins were consistently quantified across both timepoints, and 17 proteins were significantly altered post-stress (10 downregulated, 7 upregulated; Table 3; Fig. 3). Although none of these proteins reached a 1.5-fold change, some consistent trends were observed. Heat shock cognate 71 kDa protein (HSP7C) and vinculin (VINC) were upregulated in both groups, likely reflecting non-specific physiological responses to the stress protocol. Notably, several proteins demonstrated divergent responses between ME and HC groups. Fibulin-1 (FBLN1) was downregulated in ME patients but upregulated in healthy controls, whereas alpha-2-HS-glycoprotein (FETUA) was upregulated in ME and downregulated in HC. These inverse responses, although modest in magnitude, suggest possible ME-specific regulatory disruptions in response to exertion. A complete listing of proteins with opposite post-stress trajectories between groups is provided in Supplementary Tables S7 and S8.

Table 2.

Proteins with significant changes in ME: Baseline vs post-stress test values

Protein name	UniProt entry name	Ratio	p-value
Up-regulated
Haptoglobin	HPT_HUMAN	1.84	0.007
Apolipoprotein A-I	APOA1_HUMAN	1.50	0.002
Alpha-2-HS-glycoprotein	FETUA_HUMAN	1.22	0.04
Fibrinogen gamma chain	FIBG_HUMAN	1.20	0.01
Fibrinogen beta chain	FIBB_HUMAN	1.12	0.04
Down-regulated
Lumican	LUM_HUMAN	0.97	0.05
N-acetylmuramoyl-L-alanine amidase	PGRP2_HUMAN	0.96	0.04
Thrombospondin-1	TSP1_HUMAN	0.96	0.01
Fibulin-1	FBLN1_HUMAN	0.96	0.03
Attractin	ATRN_HUMAN	0.96	0.03
Apolipoprotein E	APOE_HUMAN	0.95	0.01
HCG40889, isoform CRA_b	A0A024R962_HUMAN	0.95	0.01
Heat shock cognate 71 kDa protein	HSP7C_HUMAN	0.94	0.003
Moesin	MOES_HUMAN	0.94	0.003
Fructose-bisphosphate aldolase	V9HWN7_HUMAN	0.94	0.001
Peptidyl-prolyl cis–trans isomerase A	PPIA_HUMAN	0.93	0.01
Actin, cytoplasmic 2	ACTG_HUMAN	0.93	0.0004
Vinculin	VINC_HUMAN	0.88	0.0001
Talin-1	TLN1_HUMAN	0.86	0.000001
Alpha-actinin-1	ACTN1_HUMAN	0.85	0.00003

Fig. 2.

Volcano plot displaying the difference in the protein levels between the ME (after the stress test) and ME groups. Log₂ protein ratios are plotted against the negative log₁₀ p-values. Increased protein levels are represented on the positive x-axis, while the decreased levels are shown on the negative x-axis

Table 3.

Proteins with significant changes in HC: Baseline vs post-stress test values

Protein name	UniProt entry name	Ratio	p-value
Up-regulated
Pigment epithelium-derived factor	PEDF_HUMAN	1.87	0.05
Complement C1q subcomponent subunit A	C1QA_HUMAN	1.55	0.02
Complement C1q subcomponent subunit B	C1QB_HUMAN	1.38	0.02
Hemopexin	HEMO_HUMAN	1.30	0.001
Fibulin-1	FBLN1_HUMAN	1.19	0.01
Carboxypeptidase N catalytic chain	CBPN_HUMAN	1.18	0.02
Inter-alpha-trypsin inhibitor heavy chain H4	ITIH4_HUMAN	1.15	0.03
Down-regulated
Alpha-2-HS-glycoprotein	FETUA_HUMAN	0.92	0.05
Vinculin	VINC_HUMAN	0.91	0.03
Filamin-A	FLNA_HUMAN	0.88	0.04
Moesin	MOES_HUMAN	0.85	0.02
Heat shock cognate 71 kDa protein	HSP7C_HUMAN	0.84	0.01
Talin-1	TLN1_HUMAN	0.84	0.001
Tropomyosin 1 (Alpha), isoform CRA_m	H7BYY1_HUMAN	0.80	0.01
Zyxin	ZYX_HUMAN	0.80	0.004
Tropomyosin 3 isoform 1	A0A0S2Z4G4_HUMAN	0.79	0.005
Tropomyosin alpha-4 chain	TPM4_HUMAN	0.75	0.001

Fig. 3.

Volcano plot displaying the difference in the protein levels between the HC (after the stress test) and HC groups. Log₂ protein ratios are plotted against the negative log₁₀ p-values. Increased protein levels are represented on the positive x-axis, while the decreased levels are shown on the negative x-axis

Haptoglobin as a candidate marker of PEM

Among the proteins altered in response to stress, Hp emerged as the most significantly dysregulated in ME patients (Table 2, Fig. 2 and Supplementary Figure S6). Proteomic analysis revealed that Hp levels were significantly higher at baseline compared to 90 min post-stress in the ME group (1.84-fold change, p = 0.007). This stress-induced decrease in Hp level was specific to ME patients and was not observed in healthy controls, in whom Hp levels remained undetectable at both time points. This distinct post-stress pattern positions Hp as a potential disease-specific marker of PEM. The pronounced post-exertional reduction in Hp, a key scavenger of free hemoglobin, in ME patients suggests a diminished capacity to buffer hemolysis-associated oxidative and inflammatory stress. Supporting this, the non-invasive cerebral monitoring using MASIMO's NIRS revealed a significant post-PEM increase in cerebral total hemoglobin in ME patients, a pattern absent in healthy controls (Supplementary Figure S4). This rise in cerebral hemoglobin may reflect localized vascular dysregulation and further supports a PEM-induced hemolytic response. This molecular response could represent a key feature of the pathophysiology underlying PEM and broader symptom exacerbation in ME.

Replication of Hp reduction using a high-sensitivity quantification method

The identification of Hp as a top candidate biomarker warrants careful consideration in light of the technical protocol employed. The plasma samples used in the global proteomic assay were processed using a MAR-14 immunodepletion column, which targets and removes high-abundance plasma proteins, including acute-phase reactants such as Hp. In theory, this step should have substantially reduced or eliminated detectable Hp levels across all samples. The fact that Hp remained quantifiable among the ME group, and showed consistent, biologically plausible differences post-stress, raises important questions about the molecular form and binding context of the Hp detected. To validate and extend the discovery phase results obtained using MAR-14-depleted plasma proteomics, a high-sensitivity method for direct Hp quantification in total plasma (without prior depletion) was applied to the same cohort. This orthogonal approach confirmed a significant post-stress reduction in Hp levels in ME patients, with the medium effect size (Fig. 4). In contrast, no significant change in Hp levels was found in HC group, confirming the disease specificity of the stress-induced Hp depletion (Fig. 5). This replication strengthens the evidence that Hp is a stress-responsive protein in ME and may be a biomarker of PEM. The ability to detect and quantify Hp despite MAR-14 depletion in the discovery phase suggests that certain Hp isoforms or complexes may have been partially retained in the proteomics workflow. Possible explanations include incomplete depletion due to structural modifications of Hp or altered binding affinities influenced by previously reported genotype-dependent modifications [22]. These findings justified further investigation into phenotype-specific responses and structural properties of Hp in ME.

Fig. 4.

Symptom severity based on Hp phenotypes. A The correlation between baseline (T0) Hp levels and cognitive score according to DSQ. There was a negative correlation between Hp level at baseline and cognitive dysfunction patients. B The baseline plasma Hp levels in the ME group based on the MFI-20 severity score, which categorizes patients into two groups: mild/moderate and severe. Patients with more severe symptoms exhibited lower plasma Hp levels compared to those with mild or moderate symptoms. All data are represented as mean ± standard error of the mean Pearson correlation was used to assess the correlation with an R of − 0.4 (95% CI − 0.6 to − 0.12) and a p-value of 0.006. Student's t-test was used when comparing two groups. Results were considered significant at *P value < 0.05 and **p-value < 0.01

Fig. 5.

Validation of proteomics findings using HPLC. A Hp concentrations at T0 and T90 in ME patients, showing a significant post-stress decrease consistent with proteomic analysis. B Hp concentrations at T0 and T90 in HC, showing no significant change following the stress challenge suggesting a disease-specific PEM-induced hemolytic response in ME. Paired t-tests were performed on log₂-transformed data to assess within-group changes in Hp levels. Results were considered significant at **p-value < 0.01

Hp phenotypes and baseline associations with symptom severity

To explore the clinical significance of Hp dynamics, phenotype distribution and baseline plasma Hp levels were analyzed in relation to symptom severity. In the discovery cohort, baseline plasma Hp levels were lower in the most severely affected subgroup as classified by MFI-20 scores with the moderate to strong effect size (r = 0.56). Moreover, lower plasma Hp levels at baseline were associated with greater cognitive dysfunction as measured by the DSQ (Fig. 4A, B), suggesting a link between Hp depletion and symptom burden. Hp phenotype analysis using Western blotting and size-exclusion HPLC revealed differences in distribution between ME patients and HC group (Supplementary Figures S5-S6). In the ME group, the phenotype distribution was 50% Hp2-2, 40% Hp2-1, and 10% Hp1-1, while the HC group exhibited 31% Hp2-2, 47% Hp2-1, and 21% Hp1-1. The Hp1-1 phenotype was approximately twice as common in HC compared to the ME group, consistent with known population distributions [8]. The underrepresentation of individuals with the potentially protective Hp1-1 phenotype in ME raises important questions about genotype-linked susceptibility.

Phenotype-specific Hp dynamics and association with PEM severity

In the validation cohort (n = 89 ME patients), Hp phenotype distributions closely matched those observed in the discovery phase: 51.6% Hp2-2, 40% Hp2-1, and 8.4% Hp1-1 in ME patients versus 21% Hp2-2, 58% Hp2-1, and 21% Hp1-1 in HC group. For the PEM analysis using the DPEMQ score, participant distribution across Hp phenotypes was as follows: Hp1-1 (n = 8), Hp2-1 (n = 47), and Hp2-2 (n = 34). Analysis of DPEMQ scores indicated that ME patients with Hp2-1 or Hp2-2 phenotypes experienced more severe PEM symptoms compared to those with the Hp1-1 phenotype with the mild-moderate effect size (Fig. 6A). Due to the small numbers of individuals in the Hp1-1 group, a non-parametric test for DPEMQ analysis was used, despite the values passing normality testing. Using the non-parametric approach, the difference between the groups did not reach statistical significance (p = 0.09), highlighting a known limitation of non-parametric tests in small, unbalanced groups. Nevertheless, the consistent trend observed across both parametric and non-parametric analyses supports a biologically plausible interpretation: the Hp1 allele may confer some degree of protection against PEM severity. This hypothesis is further supported by the relative underrepresentation of individuals with the Hp1-1 phenotype in the cohort. Further item-level analysis of the 13 DPEMQ questions showed a significant difference among the three Hp phenotype groups for Question 10, which assesses the symptom of “dizziness”. This difference remained statistically significant even when using non-parametric analysis. Patients with the Hp1-1 phenotype reported lower dizziness severity compared to Hp2 carriers, with the Hp2-1 group exhibiting the highest symptom burden (Fig. 6B). These results suggest a protective effect of the Hp1 allele and a potential genetic vulnerability linked to the Hp2 allele under stress conditions. The same cohort was included in the cognitive assessment, except for one Hp2-2 patient who did not complete the BrainCheck test. The same pattern was further reinforced by cognitive performance data; although baseline scores were similar across phenotypes, individuals with the Hp1-1 phenotype demonstrated better performance at T90 across multiple cognitive domains, including the Stroop task, immediate memory recognition, and total BrainCheck scores (Fig. 6C–E). The calculated effect size for cognitive performance indicated a mild to moderate effect (r = 0.23–0.27), with immediate memory recognition demonstrating the highest effect size among the domains assessed. The poorest performance was observed in the Hp2-1 group, highlighting its potential link to PEM-associated cognitive decline.

Fig. 6.

PEM severity and cognitive function in patients with ME according to different Hp profiles. A DPEMQ analysis revealed a significant difference between ME patients with three different Hp phenotypes. ME patients with either the Hp2-1 or Hp2-2 phenotype exhibited a significantly higher total DPEMQ score, indicating greater PEM severity after the stress test, compared to those with the Hp1-1 phenotype. B Patients with the Hp1-1 phenotype reported lower dizziness severity compared to Hp2 carriers, with the Hp2-1 group exhibiting the highest severity. C–E BrainCheck analysis at T90 revealed significant differences across multiple parameters, including combined BrainCheck score, Stroop and immediate recognition, among ME patients with different Hp phenotypes. In all these cognitive elements, patients with the Hp1-1 phenotype demonstrated better performance compared to those with the Hp2-1 and Hp2-2 phenotypes. One-way ANOVA was performed to assess significant differences among the normally distributed groups followed by Tukey post hoc test for multiple comparisons. Kruskal–Wallis test followed by Dunn’s test for multiple comparisons was used for non-normally distributed groups. Results were considered significant at *p-value < 0.05 and **p-value < 0.01

Structural features of the Hp2-1 phenotype in relation to PEM and cognitive dysfunction

Detailed structural analysis of Hp phenotypes using HPLC revealed characteristic oligomeric profiles for each genotype. The Hp1-1 phenotype presented a single major peak corresponding to the dimer, while the Hp2-1 phenotype displayed a more complex chromatographic profile with five distinct peaks: dimer, trimer, tetramer, pentamer, and hexamer. The Hp2-2 phenotype exhibited two overlapping peaks representing tetramer and pentamer forms. These findings were consistent with previous mass spectrometry-based characterizations by Tamara et al. [22], whose nomenclature was adopted in this study. In ME patients with the Hp2-1 phenotype, the relative abundance of Hp tetramer and pentamer forms was significantly increased compared to healthy controls with the same phenotype. Additionally, shorter retention times were observed in ME patients, indicating a shift in molecular structure or mass, with the medium to large effect (r = 0.7). Notably, the abundance of these higher-order oligomers correlated with PEM symptom severity and cognitive dysfunction (Fig. 7A–H). Previous work by Tamara et al. demonstrated that higher-mass Hp forms, such as those present in Hp2-1 oligomers, exhibit reduced glycosylation, a modification associated with diminished functional capacity, particularly in binding and clearing free hemoglobin. While our data do not directly assess glycosylation, the detection of Hp proteoforms with higher retention times and reduced function is consistent with these findings and suggests that genotype-dependent structural characteristics, potentially including lower glycosylation in large oligomers, may contribute to impaired hemoglobin clearance following exertion in ME. Together, these findings point to a distinct biochemical and structural profile of Hp in ME patients, particularly in those with the Hp2-1 phenotype, that may contribute to vulnerability to PEM and associated cognitive decline. These phenotype- and structure-specific alterations warrant further investigation as potential biomarkers and mechanistic contributors to ME pathophysiology.

Fig. 7.

Analysis of haptoglobin oligomer distribution, retention times and symptom severity in Hp2-1 Individuals. A–D The proportions of Hp tetramer and pentamer were higher in ME patients compared to healthy controls (HC), which was associated with shorter retention times. E–H The percentage of Hp tetramer was positively correlated with the severity of PEM, as measured by the DSQ, and negatively correlated with cognitive function, as assessed by the combined BrainCheck scores. In addition, higher levels Hp pentamer were linked to greater cognitive dysfunction severity, as indicated by DSQ scores, and to reduced mental flexibility. All data are represented as mean ± standard error of the mean Pearson correlation was used to assess the correlations. Student's t-test was used for comparing two groups. Results were considered significant at *p- value < 0.05 and ****p-value < 0.0001

Association between soluble LRP-1 levels and symptom severity

To investigate the potential synergistic role of LRP-1 in heme detoxification, we measured soluble plasma LRP-1 (sLRP-1) levels. LRP1 is a membrane-bound receptor that can undergo proteolytic shedding from the cell surface, generating a soluble form. This shedding process is known to be induced under chronic inflammatory conditions, such as those observed in ME, and has been previously linked to neuroinflammation [23]. Our results revealed significantly elevated plasma sLRP-1 levels in the ME group compared to healthy controls (Supplementary Figure S8A). Importantly, higher sLRP-1 levels were significantly correlated with poorer sleep scores in ME patients, as measured by the DSQ (Supplementary Figure S8B). Furthermore, a comparison of sLRP-1 levels across different Hp phenotypes revealed that individuals with the Hp2-1 phenotype exhibited the highest levels. This group also demonstrated more severe PEM symptoms, suggesting a possible phenotype-dependent dysregulation of the oxidative stress management system (Supplementary Figure S8C).

Discussion

This study provides the first evidence that Hp plays a significant genotype-dependent role in the pathophysiology of ME, particularly in relation to PEM and cognitive dysfunction. Using a two-cohort design combining global plasma proteomics, Hp phenotype determination, and post-stress quantification, we identified a reproducible and disease-specific reduction in plasma Hp levels following a standardized exertional challenge in ME patients, but not in sedentary healthy controls. This acute depletion was most pronounced in individuals with the Hp2-1 phenotype and correlated with both PEM severity and objective declines in cognitive performance. These findings support the notion that PEM may involve a hemolysis-like stress response, and that individual vulnerability to such a response is shaped by Hp genotype. Elevated plasma sLRP-1 levels in ME patients further support this observation, which might reflect increased receptor shedding driven by chronic inflammation and neuroinflammatory processes. The significant association between higher sLRP-1 levels and poorer sleep quality further supports a link between LRP-1 dysregulation and symptom burden in ME. Additionally, the notably higher sLRP-1 levels in individuals with the Hp2-1 phenotype, who also exhibited more severe PEM symptoms, suggest a Hp phenotype-dependent alteration in oxidative stress and heme detoxification pathways.

Importantly, this study is the first to identify a genetic predisposition to PEM severity in ME linked to Hp genotype: the Hp1 allele appears protective, while the Hp2 allele, particularly in its heterozygous form (Hp2-1), is associated with symptom exacerbation. This is a novel finding not only within the field of ME, where genetic contributions to PEM have remained largely speculative, but also in the broader context of Hp biology. Structural and oligomeric differences between Hp phenotypes have been previously described, but their impact on stress-induced symptoms and cognitive decline in chronic inflammatory conditions has not been demonstrated. Our study addresses this gap by correlating the presence and abundance of higher-order Hp oligomers with symptom severity and functional impairment. Notably, the altered chromatographic profiles observed in ME patients with the Hp2-1 phenotype, characterized by increased tetramer and pentamer forms and reduced retention times, suggest underlying structural modifications that may impair Hp’s protective functions in oxidative stress and hemoglobin clearance.

A key strength of this study lies in the detection of Hp in MAR-14-depleted samples, a step typically used in proteomic workflows to remove high-abundance acute-phase proteins such as Hp. The ability to detect biologically meaningful differences in Hp levels post-stress, despite the use of this depletion protocol, underscores the robustness of the signal and suggests that specific Hp isoforms or complexes, possibly modified or bound in ways that affect depletion efficiency, are retained in ME plasma. This finding is particularly relevant for the proteomics field, where acute-phase proteins are often deliberately excluded due to concerns about dynamic range and technical variability. By demonstrating that disease-relevant information can still be extracted from partially retained forms of these proteins, this study challenges the assumption that their removal always enhances data quality. Moreover, it underscores the need to revisit proteomic exclusion criteria in diseases such as ME, where stress-responsive proteins may carry pathophysiological importance.

The use of orthogonal validation approach further strengthens our conclusions. High-sensitivity quantification of total Hp in non-depleted plasma using an HPLC method replicated the stress-induced reduction seen in the proteomic discovery cohort and highlighted phenotype-specific vulnerabilities, particularly in Hp2-1 individuals. Associations with cognitive outcomes add an additional layer of clinical relevance, indicating that structural features of Hp not only track with biochemical changes but may also translate to functional impairments. Although participants completed a brief practice session to reduce the potential learning effects from repeated cognitive testing, this remains a possible source of bias in the post-exertion results and should be considered when interpreting the findings. Although the precise mechanisms remain to be clarified, the observed correlation between Hp oligomer distribution and PEM severity supports the hypothesis that altered Hp architecture, possibly driven by genotype-dependent glycosylation or other post-translational modifications, affects its efficacy in Hb scavenging and oxidative stress buffering under exertional stress.

In the broader context of ME research, this study contributes a multidimensional characterization of a potential molecular driver of PEM. Previous studies have focused on immune dysregulation, energy metabolism, and oxidative stress. However, few have linked these features to stress-responsive plasma proteins with genotype-specific functions. Our findings suggest that hemolysis-like processes, as reflected by stress-induced Hp depletion, may underlie symptom exacerbation in ME and could be modulated by inherited factors. This expands the conceptual framework of PEM beyond generalized inflammation or redox imbalance and introduces a targeted candidate biomarker that integrates environmental triggers with genetic susceptibility.

While this study was primarily exploratory in design, the reproducible and statistically significant reduction in Hp observed post-exertion, despite standard depletion protocols, highlights its robustness and pathophysiological relevance. Although most differentially expressed proteins did not exceed conventional FC thresholds, this likely reflects the nuanced and tightly regulated nature of circulating factor responses in ME, even under exertional stress. Importantly, our cohort was composed primarily of severely affected, housebound individuals, patients who are rarely included in clinical studies due to the demands of protocols such as cardiopulmonary exercise testing (CPET), which typically excludes this population. The standardized 90-min provocation maneuver used here was carefully tailored to this severity level and, based on extensive clinical experience, is sufficient to trigger PEM in this group. While PEM often evolves over 24–48 h in less severely affected individuals, severely ill patients frequently experience a more immediate cascade of molecular and symptomatic changes. Our ability to detect early Hp alterations within this timeframe supports this notion, though we acknowledge that additional time points would help capture the broader temporal dynamics of PEM and its systemic consequences.

The consistent Hp-specific findings across cohorts and analytical platforms reinforce the strength of this signal. Although mechanistic analyses of Hp structural changes were beyond the scope of the current study, phenotype-specific differences in Hp oligomer profiles provide a compelling rationale for such investigations. The relatively low prevalence of the Hp1-1 genotype in ME did limit statistical power for some direct comparisons, but consistent directional trends support its potential protective role and justify further study in larger or genetically stratified cohorts. Crucially, the identification of Hp as both a functional and structural correlate of PEM severity opens promising therapeutic avenues. Emerging strategies, such as intravenous infusion of purified recombinant Hp1-1, may offer targeted protection for ME patients with Hp2 allele associated vulnerabilities [7, 24, 25]. Our findings strengthen the rationale for these approaches and suggest that modulating Hp levels or proteoforms could help prevent or mitigate post-exertional crashes and associated cognitive impairment, commonly referred to as brain fog. While further validation and mechanistic work are needed, this study lays the groundwork for phenotype-informed interventions aimed at restoring redox homeostasis and enhancing hemoglobin clearance capacity in genetically susceptible individuals.

Conclusions

This study identifies Hp as a novel, genetically influenced marker and potential mediator of PEM in ME. It highlights the relevance of stress-responsive proteins traditionally excluded from proteomic analyses, establishes a biochemical-genetic-functional axis of vulnerability in ME, and opens the door to phenotype-targeted therapeutic strategies. These findings advance the field by integrating proteomics, genetics, and functional outcomes to uncover a plausible mechanism linking exertional stress to symptom exacerbation in ME, and may also inform broader research into Hp biology in chronic inflammatory and neurocognitive conditions.

Abbreviations

ACN

Acetonitrile

AGC

Automatic gain control

AmBic

Ammonium bicarbonate

ANOVA

Analysis of variance

ApoA1

Apolipoprotein A1

BMI

Body mass index

BSA

Bovine serum albumin

C

Carbamidomethyl (C)

CCC

Canadian Consensus Criteria

CD163

Cluster of differentiation 163

CHb

Cerebral total hemoglobin

CHU Sainte-Justine

Centre Hospitalier Universitaire Sainte-Justine

COVID-19

Corona virus disease-19

CPET

Cardiopulmonary exercise test

DPEMQ

DePaul Post-Exertional Malaise Questionnaire

DSQ

DePaul Symptom Questionnaire

DTT

Dithiothreitol

ECL

Electrochemiluminescence

ELISA

Enzyme linked immunosorbent assay

EDTA

Ethylenediaminetetraacetic acid

FA

Formic acid

FASTA

Text-based format for nucleotide and protein sequences

FBLN1

Fibulin-1

FC

Fold-change

FDR

False discovery rate

FETUA

Alpha-2-HS-glycoprotein

g

Gravitational force

Hb

Hemoglobin

HbO₂

Oxygenated hemoglobin

HC

Healthy control

HCD

Higher-energy collisional dissociation

HeLa

Immortalized cell line

HHb

Deoxygenated hemoglobin

Hp

Haptoglobin

HPLC

High-performance liquid chromatography

HRP

Horseradish peroxidase

HSP7C

Heat shock cognate

Hz

Hertz

IAA

Iodoacetamide

ID

Internal diameter

IgA

Immunoglobulin A

IgG

Immunoglobulin G

IgM

Immunoglobulin M

LC-MS

Liquid chromatography-mass spectrometry

LRP-1

Low-density lipoprotein receptor-related protein 1

M

Molarity

M

Oxidation

MAR-14

Multi affinity removal column, human-14

MARS

Multi Affinity Removal Spin

MB124

Flush solution for the LC–MS separation

ME

Myalgic encephalomyelitis

MFI-20

Multidimensional Fatigue Inventory Questionnaire

mL

Mililiter

mM

Milimolar

MS/MS

Tandem mass spectrometry

m/z

Mass-to-charge ratio

Na₂EDTA

Disodium ethylenediaminetetraacetate

NaCl

Sodium chloride

NIRS

Near-Infrared Spectroscopy

NQ

Deamidation

PBS

Phosphate-buffered saline

PEM

Post-exertional malaise

PS

Particle size

p-value

Probability value

PVDF

Polyvinylidene fluoride

q-value

False discovery rate or FDR

rSO₂

Regional oxygen saturation

SDS

Sodium dodecyl sulfate

SEM

Standard error of the mean

SF-36

36-Item Short-Form Health Survey

sLRP1

Soluble Low-density lipoprotein receptor-related protein 1

T0

Baseline, pre-stress-test

T90

90 Minutes post-stress-test

TFA

Trifluoroacetic acid

TRIS

Tris(hydroxymethyl)aminomethane

UV-Vis

Ultraviolet-visible spectroscopy

VINC

Vinculin

μL

Microliter

Author contributions

AMoezzi contributed to the design of the study, performed all the experiments characterizing plasma Hp phenotyping, conducted the Hp phenotypes and questionnaire data analysis and wrote the manuscript. AU performed the mass spectrometry experiments (global plasma proteomics), the analysis of proteomics data and wrote the manuscript in collaboration with AW. PL participated in the proteomic data analysis. BRA and CL participated in the Hp data analysis. WE coordinated the study and sample delivery across various OMF collaborative ME/CFS research centers and contributed to classifying and analyzing severity scores, ensuring cohesive collaboration and accurate data interpretation. IC contributed to different bioinformatic analyses. AF contributed to the overall coordination, ethics and multiple revisions of the manuscript. CG participated in the conception of the study design and more specifically to the provocation maneuver inducing PEM. ON set up the HPLC method for Hp quantification and phenotyping. AMoreau, JB and WX contributed to the design of this study, participated in different analyses, wrote the manuscript and secured the funding for this study. All co-authors reviewed and approved the final version of the manuscript.

Funding

This work was supported by a grant (DOMINO-ME project to Pr. Alain Moreau and Pr. Jonas Bergquist) funded by Open Medicine Foundation (USA), and a generous anonymous donor through Open Medicine Foundation Canada (Pr. Alain Moreau). This work was also supported by the Mass Spectrometry Based Proteomics Facility at Uppsala University, the Estonian Research Council (grant PRG690I to Pr. Jonas Berquist), the Excellence in Analytical Chemistry (EACH) program (to Pr. Jonas Bergquist) and a Computational Science grant (to Dr. Wenzhong Xiao) funded by Open Medicine Foundation (USA).

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding authors upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board of CHU Sainte-Justine (protocol #4047). Informed consent was obtained from all participants involved in the study.

Consent for publication

Not applicable.

Competing interests

Pr. Alain Moreau is the Director of the Interdisciplinary Canadian Collaborative Myalgic Encephalomyelitis (ICanCME) Research Network, a national research network funded by The Canadian Institutes of Health Research (grants MNC- 166242 and MNC – 196095). Pr. Jonas Bergquist in Chief Medical Officer of Open Medicine Foundation (USA). Pr. Alain Moreau, Pr. Jonas Bergquist and Dr. Wenzhong Xiao are Members of the Scientific Advisory Board of the Open Medicine Foundation (USA).