7-day longitudinal proteomics of critically ill patients: a pilot study

Abstract

An adult’s health, indicated by measurable parameters, is stable over time. With the exception of circadian rhythms, variability in these parameters typically does not exceed 20%. In this pilot study, we looked into the stability of proteome in intensive care unit (ICU) patients. This was a single-center, prospective, observational pilot study of blood plasma from adult ICU patients with statistically heterogeneous patterns of clinically observed parameters. Eight week-long batches from seven patients (one patient participated twice) were analyzed by means of bottom-up proteomics. The data were analyzed with MaxQuant software against reference proteome. The obtained intensities were further processed with in-house R and Python scripts. In total, 218 proteins were identified; however, only 68 proteins appeared in all samples from all patients. Most proteins remained stable within observation (within-patient variance was less than 30%). The random-effects model also confirmed high impact of within-patient variance on the protein levels. The effects of time on the protein level variances did not exceed 5%. Z-score-based hierarchical clustering analysis revealed that the daily data of each patient were clustered together indicating that the plasma proteome of ICU patients both bears individual traits and remains stable during short-term progression of the patients’ condition. Therefore, in this pilot group of patients, the analysis over seven consecutive days fails to reveal proteome dynamics.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-55106-7.

Introduction

Critically ill patients residing in intensive care units (ICU) require mechanical ventilation and have high risks of organ failure and systemic inflammation; thus, ICU therapy is generally based on supportive care. The non-specific markers of systemic inflammation are the level of C-reactive protein (CRP) and various leukocyte ratios^1,2. More specific markers include interleukin levels, as well as presepsin (CD14)³ and procalcitonin (PCT)⁴. Multiple organ failure is most commonly assessed based on the Sequential Organ Failure Assessment (SOFA) score, which includes the patient’s oxygenation levels, blood pressure, as well as CRP and biochemical markers of organ insufficiency⁵. Usually, the temporal dynamics of biomarkers are considered either in connection with a chronic disease or in connection with a long treatment process to control its results, in particular, to evaluate the effectiveness of cancer treatment, resolution of multiple organ failure and to predict patient survival⁶. The disadvantage of dynamic monitoring of biomarkers is that it does not always provide an opportunity to determine the individual trajectory in a prolonged course of the disease by providing general information at the population level⁷.

For ICU patients, the development of multiple organ dysfunction syndrome (MODS) is of utmost importance, and the difficulty in finding an effective therapy to prevent or treat MODS relates to the overlapping nature of the multiple systems priming each other for an exaggerated physiologic response^8–10. Previously, Xu et al. demonstrated that the development of organ failure could be predicted based on the dynamics of the patient’s markers¹¹. Altogether, organ failure could be considered as a severe destabilization of the organism. Progressive development of organ failure could be predicted based on the dynamics of the patient’s markers and 72‑h SOFA score trajectories and improvement of the patient’s condition is also related to the biomarker levels. In intensive care, we historically consider that the mechanisms of response to shock and infection are similar in all people. But nowadays we have some arguments that individual genomics and proteomics may play a considerable role in the response to critical conditions. Previously, Philipp Geyer et al.¹² have shown that the plasma proteome of healthy individuals is stable by comparing intra- and inter-group variability using single‑pass shotgun proteomic analysis. To determine inter-individual variability, plasma samples were collected from five female and five male donors in triplicate by finger prick. It has been shown that the plasma proteome as a whole has much greater inter-individual than intra-individual variability (19% and 55% of proteins within a CV of 20%, respectively).

Proteomics allows simultaneous analysis of relative abundances of hundreds of proteins in a 10‑µl sample, and thus it has high potential for ICU patient diagnostics. Indeed, more than a dozen proteomic studies of ICU patients appeared in the last decade (rev. in¹³). In the recent study of Palmowski et al., proteomic analysis revealed significant differences in several dozens of plasma proteins between sepsis survivors and non‑survivors¹⁴. In another study by Van Nynatten et al., targeted Olink‑based blood plasma proteomics allowed accurate distinguishing between sepsis patients and healthy control subjects¹⁵. In a recent study of Mitsuyama et al., longitudinal proteomic analysis of ARDS patients revealed that coagulation and complement cascades were more commonly activated in blood plasma in the acute phase of ARDS than in the subacute phase¹⁶. Quite recently, Bracht et al. performed LC‑MS analysis of blood plasma for 333 septic patients on days 1 and 4 of the disease¹⁷. Four clusters of proteins related to the patients’ prognosis were distinguished.

In the present pilot study, we hypothesized that, in contrast to healthy people, there could be a sub‑proteome of plasma proteins that destabilizes in ICU patients as their condition worsens and stabilizes as their condition improves. We monitored changes in the large‑scale proteome structure over seven days of patients’ stay in intensive care. During the observation period, the patients’ condition changed according to three criteria (physician assessment, CRP, and mSOFA).

Results

Patient characteristics

This pilot study included 7 ICU patients (Table 1, SI‑1, Fig. S1). Patient 1 participated twice in different periods of the disease (Table 1, Table S1: period of improvement – 1a, and period of deterioration – 1b), as we were interested in identifying patient‑specific patterns. For the same reason, the selected group was highly heterogeneous in clinical parameters (Table S2, 2‑way ANOVA p < 0.001 for all parameters). The 7 patients selected for the study were 4 men (age 69.5 ± 3.9 years, height 171.0 ± 2.1 cm, body weight 76.8 ± 6.4 kg) and 3 women (age 60.0 ± 7.6 years, height 160.7 ± 5.2 cm, body weight 68.8 ± 10.7 kg). The ICU subjects were monitored with the recording of basic vital signs and the calculation of dysfunction scores (mSOFA), on mechanical ventilation and continuous infusion therapy (Table S1). Antimicrobial therapy was administered to all patients due to hospital‑acquired pneumonia (mostly K. pneumoniae, Table S1). Blood component transfusion was performed in 62.5% of cases (P1, P4, P5, P7).

Table 1.

Brief patients’ characteristics.

ID/Group	Main clinical problems/Events	Associated pathology (CCI)#
Patient P1 (Male, Age 75, BMI 24.22)	Recurrent CAP, pulmonary sepsis. Multiple myeloma. Day 1: Systemic mycosis (Candida glabrata). Day 14: Clostridial colitis. 1a: Days 1–7; 1b: Days 16–23. Day 78: Biological death.	AH. COPD. CVD. CKD. Malignancy. (10)
Patient P2 (Male, Age 73, BMI 31.83)	Day − 14: Recurrent ischemic stroke in right MCA. HAP, CRBSI, Clostridial colitis, sepsis, septic shock. Day 8: Biological death	AH. CVD. T2DM. Obesity. (7)
Patient P3 (Female, Age 61, BMI 35.16)	Giant olfactory meningioma. Post-surgical cerebral edema with axial brain dislocation. Day − 25: Microsurgical tumor resection. HAP, sepsis. Day 17: Discharged	AH. CVD. T2DM. Obesity. (5)
Patient P4 (Male, Age 72, BMI 22.59)	Pancreatic head cancer (cT3N2M0). Day − 8: Pancreatoduodenectomy. HAP, SIRS. Day 7: Discharged	COPD, CVD. Malignancy. (6)
Patient P5 (Female, Age 46, BMI 19.38)	Day − 20: Sequelae of severe associated trauma and TBI, TSAH of the left temporal region. Cerebral edema and dislocation. HAP, SIRS. Day 31: Discharged	- (0)
Patient P6 (Male, Age 58, BMI 26.17)	Cerebral meningioma. Post-surgical cerebral edema with dislocation. Day − 7: Microsurgical tumor resection. Day − 2: DC and hematoma evacuation. Secondary CNS infection. Day 7: Discharged	AH. CVD. (4)
Patient P7 (Female, Age 72, BMI 26.19)	Day − 5: NSTE-ACS, hypertensive crisis ESKD as an outcome of systemic ANCA-associated vasculitis with kidney and lung involvement. HAP, sepsis. Day 19: Biological death	AH. COPD. CVD. CKD. MHD. Rheumatic disease. (8)

# Abbreviations: AH – arterial hypertension; BMI – body mass index; CAP - community acquired pneumonia; CCI - Charlson Comorbidity Index; CKD – chronic kidney disease; CNS – central nervous system; COPD – chronic obstructive pulmonary disease; CRBSI – catheter-related bloodstream infection; CVD – cardiovascular diseases; DC – decompressive craniectomy; ESKD – end-stage kidney disease; HAP – hospital acquired pneumonia; MCA – middle cerebral artery; MHD – maintenance hemodialysis; NSTE-ACS – non-ST-segment elevation acute coronary syndrome; T2DM – type 2 diabetes mellitus; TBI – traumatic brain injury; TSAH – traumatic subarachnoid hemorrhage.

According to the dynamics of clinical parameters (Table S1, Fig. S1), especially mSOFA score (Fig. 1b) and CRP level (Fig. 1a), the samples could be divided either by positive or negative trend (by CRP, Table S2), or by the presence of severe multiple organ dysfunction (mSOFA ≥ 5). In order to reveal proteomic patterns associated with a more severe patient’s condition, we looked into both approaches for the patients’ distinction. CRP and physician opinion significantly changed between the early (1‑2‑3) and late (5‑6‑7) days of analysis (Table S2).

Fig. 1.

Patterns of clinical parameters and comparison between identified protein datasets. (a, b) Dynamics of the C-reactive protein (CRP) (a) and mSOFA score (b) values vs. day of analysis (see Table 1), other clinical parameters are given in Figure S1 and Table S1. 2-way analysis of variance statistical significance is indicated: “ns” corresponds to p > 0.05; *** corresponds to p < 0.001. (c, d) Venn diagrams of identified proteins. c. Comparison between proteins identified at least once in the positive trend group (by CRP) vs. negative trend group (by CRP). d. Comparison between proteins identified at least once in all samples taken at points with mSOFA ≥ 5 vs. samples taken at points with mSOFA < 5.

The elevated CRP levels were not consistently associated with higher mSOFA scores at the time of patients’ study inclusion. This discrepancy could be explained by two main factors. First, the 7‑day observation period was held at different stages of each patient’s overall ICU stay, and therefore the measured parameters could reflect different phases of the disease course. Second, SOFA and CRP do not necessarily follow a linear relationship and thus may exhibit divergent trajectories^18–20. Although both parameters reflect clinical status and its dynamics, CRP primarily indicates the intensity of the infectious or inflammatory process, whereas SOFA assesses the severity of multi‑organ dysfunction.

Analysis of identified plasma proteins

In total, 218 proteins were reliably identified at least once; 203 out of them were identified in the samples related to positive CRP trend, 211 in those related to negative CRP trend (Fig. 1c), 206 in those related to high mSOFA, and 207 related to low mSOFA (Fig. 1d). Most of the proteins found in the more severe groups (red in Fig. 1c, d) were various immunoglobulins (IGLV2‑18, IGLV1‑51, IGLL1, IGKV1‑27, IGHV3‑49, IGHV3‑38, IGHV3‑33, IGHV1‑3, IGHV1‑24, IGLV3‑10), most of which are known to be highly patient‑specific²¹, and could not be mapped to any functional group by the String DB²², suggesting that only total immunoglobulin load could be considered as a marker¹⁷. All proteins indicated in Fig. 1c and d were excluded from further analysis.

When looking only for those proteins which were identified in all days in all patients, only 68 proteins remained (Fig. S2a). The appearance of 22 proteins seemed to be associated with CRP level trends (Fig. S2a); some appeared only when the CRP level was going up, while some when it was going down. These 22 proteins could be mapped to two main clusters, namely, complement activation and LDL particles (Fig. S2b). Of interest, 11 of the 22 proteins, including apolipoproteins APOC1, APOC2, APOD and APOL1, and SERPIN proteins, were previously identified as downregulated upon significant weight loss in obese persons²³, indicating the significance of body weight in patient survival. Furthermore, 11 other proteins (apolipoproteins, SERPIN proteins, and complement components) were previously identified by Gómez et al.²⁴ as serum markers of non‑invasive muscular bladder cancer. Therefore we conclude that the named proteins’ appearance in blood plasma indicates some kind of general stress response.

Analysis of protein abundances and their variations

For label‑free plasma proteomic analysis, the question of stability of the plasma proteome, i.e., whether the concentrations and composition of the proteome change with time, is of utmost importance². Most ICU patients receive frequent blood product transfusions and infusions of therapy (Table S1), thus potentially destabilizing the plasma proteome. In order to investigate this point, we analyzed the consistency and variation in iBAQs of different proteins (Fig. 2, S2). Analysis of protein representation between the samples related to positive vs. negative CRP trend (Fig. 2a), or more or less severe condition (by mSOFA, Fig. 2b) showed that only complement factor D (CFD) and immunoglobulins were significantly represented in the more severe group (red in Fig. 2a, b). A statistically significant (p‑value < 0.05) 1.5‑fold increase in the samples related to a less severe condition (green in Fig. 2a, b) was observed for various complement components and serine protease inhibitors. Analysis of protein representation between the early and late days of the positive trend (by CRP) (Fig. S2c) and negative trend (Fig. S2d) groups separately revealed that acute phase proteins, including CRP and SAA1, complement components and serine protease inhibitors could be associated with a poorer patient’s condition, while various immunoglobulins and apolipoproteins C3 and C4 were associated with a better patient’s condition (Fig. S2). The same proteins were previously found in several studies looking at COVID‑19 diagnostics and severity^25,26.

Fig. 2.

Analysis of protein abundances and their variations. (a, b). Volcano plots for protein abundance comparison between the positive trend group (by CRP) vs. negative trend group (by CRP) (a), or samples taken at points with mSOFA ≥ 5 vs. samples taken at points with mSOFA < 5. (c, d). Between-patient vs. within-patient variances calculated for the given random-effects model. (c) Color code for the Akaike Information Criterion (AIC) for model performance (red – best, blue – worst). (d) Color code for the R² statistics. (e) Normalized iBAQ values for the named proteins.

The main objective of our study was to assess the proteome stability in the ICU conditions. We assumed that the observed variability is related to a change in the patient’s condition, expressed by clinical parameters²⁷. Therefore, we used a random effects model²⁸ to estimate the variability of protein representations when attempting linear regression in two variants: regression of iBAQs vs. mSOFA and physician opinion (Fig. 2c, Table S3), and regression of iBAQs vs. potassium and platelet levels (Fig. S3, Table S4).

Both models revealed that both between‑patient and within‑patient variations were quite high (around 30‑50%), with the between‑patient variation higher than the within‑patient one for most proteins. A few proteins demonstrated much higher within‑patient variation in the random‑effects model (HPR, ITIH3, C1S, and others, Fig. 2c, Fig. S3), which surprisingly corresponded to lower values of the Akaike Information Criterion (AIC); therefore, these proteins could be indicative of the patient’s condition if their variation were taken into account. Altogether, we conclude that the stability of the proteome obtained by panoramic mass spectrometry is preserved even in the ICU conditions, with some proteins being highly patient‑specific.

In order to explicitly determine the effects of time in the described longitudinal data, we trained the random effects model to estimate the capability of the protein level description by each clinical parameter alone, or together with time (Fig. 2d, Table S5). The R² statistics were used to evaluate the success of the description. However, for most of the proteins under consideration, no clinical parameter allowed us to reliably describe the level either independently or together with the time parameter (Table S5). The contribution of the time parameter to the description of protein levels did not exceed 5%, and in most cases was less than 1%, which indicates the absence of independent dynamics of protein levels, illustrated in Fig. 2e. Proteins reliably described by the mSOFA value (Fig. S2d, Table S5) were related to complement and coagulation cascades, and were also previously found as markers for COVID‑19²⁶ and for sepsis severity²⁹.

Associations between protein abundances and clinical data

Despite the absence of obvious dynamics of the plasma proteome of ICU patients, the performed analysis indicates the presence of correlations between protein levels and clinically measured parameters. In order to find these associations, we performed canonical correlation analysis (CCA³⁰, . In Figure S4, CA matrices for all data (Fig. S4a), for the positive trend (by CRP) group (Fig. S4b), and for the negative trend group (Fig. S4c) are given. CCA for the 68 proteins present in all samples is given in Fig. 3a. The relations with clinical data allow clustering of related clinical parameters; thus we can select the groups of proteins most indicative of a patient’s condition (Fig. 3). Cluster A seems to indicate kidney failure and inflammation; it contains diverse proteins, including complement and coagulation cascades, proteolysis and transport proteins (Fig. 4b). The same proteins were found previously in many studies, and altogether indicate an acute phase response of the patient^31,32. Clusters B and C contained various blood microparticle proteins. Cluster D contains proteins of the lipoproteins (Fig. 4b), also found by Zhang et al. as associated with HDL maturation³³.

Fig. 3.

CCA analysis of iBAQs associations with clinical parameters. Spearman correlation coefficient, only data with > 90% points present was included in the analysis. a. Correlation matrix with hierarchical clustering. b. Z-scores for normalized iBAQs changes for the protein clusters named in panel a.

Fig. 4.

Hierarchical clustering of proteins and patients represented by their normalized iBAQs Z-scores. a. Clustering of patients data points (“Days”) based on protein abundances from Clusters A and D (Fig. 3). b. String DB analysis of corresponding genes, String DB edges color code (string-db.org) and String DB biological pathway mapping is given in the legends.

CCA performed for positive or negative trend (by CRP) groups separately (Fig. S5, S6) gave similar clusters. For the positive trend (by CRP) group, the largest cluster contained microparticle proteins, which highly correlated with blood cell counts and anticorrelated with kidney failure indicators (Fig. S5a). Network analysis³⁰ of the whole correlation matrix (Fig. S4b), as well as String DB analysis, confirmed high connectivity within this cluster of proteins (Fig. S5b, c). For the negative (by CRP) trend group, the clusters of HDL remodeling, complement and coagulation cascades were once again revealed (Fig. S6). Of interest, LPS‑binding CD14 and LBP proteins were selected for the negative dynamics group as correlating with clinical CRP level and the patient’s temperature, which reflects that the patients from the negative dynamics group may develop gram‑negative sepsis (Table 1)³.

Thus we conclude that the selected proteins could be used for the patient profiling. In Fig. 3b, mean dynamics of the protein clusters for each patient is given. Of interest, both patients with chronic kidney disease (CKD, Patients 1 and 7) demonstrated highest expression of the proteins from Cluster D, related to blood lipoproteins, and lowest expression of proteins from cluster A, related to blood coagulation and fibrinolysis. These clusters contain various blood microparticle proteins, probably related to the total lipid content in blood plasma, thus leading to an increased blood viscosity and ability for coagulation³⁴. While patients with CKD are known to have higher risks of thrombosis and elevated fibrinogen levels³⁵, in ICU these patients have higher risks of development of disseminated intravascular coagulation (DIC), which is usually accompanied by lower fibrinogen levels, and blood pressure⁹, and could be distinguished by SOFA and APACHE II scores.

Patient- and group-specific components of blood plasma proteome

In order to look for the patient‑specific patterns in the blood plasma proteome, we performed clustering analysis taking only proteins from Clusters A and D as features (Fig. 4a). Alternatively, we performed hierarchical clustering of the iBAQs’ all‑samples Z‑scores (Fig. S7), or within‑patient Z‑scores (Fig. S8), or by the all‑samples relative iBAQs (fold changes compared to total average) (Fig. S9). When calculating the representation of a protein in relation to the entire sample, both Z‑scores and folds led to the clustering of all or almost all days belonging to the same patient together. In this case, all days of Patient 1, represented by 14 points, were clustered together. Patients 2 and 5 were clustered together, but with a clear division between them (Fig. 4a). Both patients have negative CRP trend (Fig. 1a), but while Patient 2 has severe organ dysfunction (mSOFA > 5), Patient 5 has low mSOFA values (Fig. 1b). Most time points of Patients 3, 4, and 7 were also distinctly clustered by the selected proteins (Fig. 4a), while clustering based on all proteins did not allow their clear discrimination (Fig. S7, S9). The calculation of Z‑scores within each patient did not allow distinguishing between patients or trends, suggesting that within‑patient iBAQ dynamics is not informative (Fig. S8).

The distinction between the patients is most noticeable for the cluster A proteins, which contained coagulation, proteolysis and immune response proteins (Fig. 4b). Of interest, 15 of the 22 proteins, mainly from the complement and coagulation cascades, were also found by Mukherjee et al. as downregulated in patients with rheumatic mitral stenosis and considered to be associated with chronic inflammation³⁶. Indeed, in our study, these proteins, especially complement components C3, C5 and C4B, were up‑regulated in the patients with less severe condition (by CRP and mSOFA, patients 3, 4, and 6), and down‑regulated in the patients with more severe condition.

Discussion

Here we addressed the question of proteome dynamics and stability in the critical care unit situation. The seven patients included in the study were balanced in terms of age and sex, while their condition and comorbidities varied (Table 1). It should be noted that all patients were on mechanical ventilation and continuous infusion therapy, but the blood sample for the analysis was taken at 5 a.m. each day before any procedures; therefore, the plasma protein content should be reflective of the patient’s condition.

More than half of the identified proteins were either group‑specific, or patient‑specific, or even day‑specific (Fig. 1). The immunoglobulins were the most varied, which agrees well with previous findings²¹, however, together with complement components, they were associated with a more severe patient condition, determined by CRP and mSOFA (Fig. 1). These observation could reflect start of systemic inflammation in these patients⁵, or response to in-hospital infection, as they could be mapped to the S. aureus infection by the KEGG mapper³⁷, and these results agree well with plasma proteomics for septic patients^17,38.

Components of blood lipoproteins were found as indicative of the patients’ condition (Fig. 4b). Among these, apolipoproteins APOC1, APOC2, APOD and APOL1 were also previously identified as downregulated upon significant weight loss in obese persons²³, indicating the significance of body weight in patients survival, while others were identified by Gómez et al.²⁴ as serum markers of non‑invasive muscular bladder cancer. Of interest, abundances of APOA4, APOC3 and APOE, components of low‑density lipoproteins and chylomicrons, correlated with multiple organ failure scores and kidney failure markers, and were most presented in all days of Patient 7 having kidney failure secondary to inflammation. On the other hand, high‑density lipoproteins involved in cholesterol reverse transport, APOA1 and APOA2, as well as serum albumin and clusterin, correlated with blood cell counts, and were up‑regulated for patients with less severe state, especially patients 4 and 6, who were discharged by the end of the analysis (Table 1). Differential expression of apolipoproteins was previously found in multiple studies related to systemic inflammation conditions such as COVID‑19 diagnostics and severity^25,26, sepsis severity²⁹, cerebral injury³⁹, and others^31,32,40. The appearance of blood microparticle protein as both group-specific and patient-specific could reflect two features of the critical condition: either a total increase in the content and conditions of the extracellular vesicles^33,41, probably related to the procoagulant blood status³⁴ in DIC; or the relationship between the patient’s survival and lipid metabolism proposed earlier^42,43.

As Philipp Geyer et al.²³ have shown, inter‑individual variability in proteome profiles is not very high. In our case, inter‑individual variability was also lower than the differences between individuals, indicating overall stability of the patient’s proteome even in critical conditions. However, since here we obtained a negative result—namely, that panoramic proteomics data are unlikely to be used for diagnostic purposes in the intensive care unit — the small sample size may have led to a type II error, and probably it needs verification on a larger cohort. Heterogeneous experimental designs and limited numbers of study participants are common limitations in critical illness research. The problem of biomarker discovery remains challenging, even in fairly homogeneous patient groups⁴⁴. Abundance changes in the immune system proteins (complement and immunoglobulins) that we found are also consistent with the results of a systematic review of critical conditions studies⁴⁵.

This study is a pilot, and has a number of important limitations. They include enrollment of diverse patients with different diagnoses and comorbidities, a small cohort size, limited observation period and focus on the identification of the proteomic markers. This prevented us from identifying uniform patterns across all patients, and allowed us to only observe individual differences between them. Laboratory data of some patients were not complete (due to limitations of treatment protocols in the hospital), limiting us in performing correlation analysis. Within the selected 7‑day period we did not observe any sudden deteriorations or sudden recovery; every patient’s clinical course had a slow‑moving trend towards worsening or improvement. Maybe we need more information about proteome fluctuations in acute incidents. Follow-up studies could build upon this by employing larger cohorts, and extended observation periods. Unfortunately, the small differences between the protein levels, although similar to those observed in other proteomics studies^12,14, prevent us from hope to validate the results by means of ELISA. Other more sensitive proteomic methods, such as the proximal extension assay¹⁵, could be used in the further studies to better quantitate the markers, which were identified as promising by the present study.

The authors took into account the small patient sample size when analyzing the data. A key limitation of reducing the sample size is the study’s limited ability to detect large or qualitative differences between patients or measured parameters. Overall, our findings are not quantitative enough for clinical analysis of changes in the patient’s condition, but they do allow us to characterize the observed trends.

Complex instrumental studies, such as functional magnetic resonance imaging (fMRI), tend to require modest sample sizes and more ambiguous group selection criteria, often characteristic of behavioural studies^46,47. In our pilot study, we are working with a modest sample, characterized by symptoms of multiple organ failure (critical condition).

Materials and methods

Study design and participants

This pilot, single‑center, prospective, observational clinical study was conducted with the patients of the NMRC “Treatment and Rehabilitation Center” between June 2023 and May 2024. The study group inclusion criteria were met by the patients aged 25 to 80 years who were admitted to the Intensive Care Unit (ICU) with any life‑threatening condition, either following major surgical interventions, or with complications of pre‑existing conditions, or with acute organ failure. Exclusion criteria encompassed a wide range of conditions that could significantly affect the course and outcome of the primary disease. Patients with consequences of poisoning of any etiology, burn disease, or radiation sickness, as well as those with acute toxic‑metabolic disturbances (alcoholic or narcotic coma, anaphylactic shock), were excluded. Patients with genetic disorders, as well as pregnant women, were not eligible for inclusion. Patients indicative of an extreme severity of condition—for example, after cardiopulmonary resuscitation—were not enrolled. Other patients who stayed in the ICU for less than 7 days or who declined to participate in the study (or for whom relatives’ consent was not obtained) were also excluded.

All clinical and biological parameters (Table 1, Table S1, SI‑I), such as demographic characteristics, ventilation parameters, length of hospital stay, and comorbidities, were collected from electronic medical records and discharge summaries. Six out of 7 patients had significant comorbidities, as indicated by the comorbidity index (CCI, Table 1), the most frequent being cardiovascular disease. Three patients underwent major surgical interventions. Treatment outcomes varied: three patients were transferred in stable condition to other departments or discharged, one was discharged for outpatient follow‑up, and three cases resulted in biological death, underscoring the overall high severity of the patients’ conditions in this cohort. Disease severity was assessed using the Sequential Organ Failure Assessment (SOFA) score (range 0–24)³¹. A modified SOFA score was also calculated for each patient^32,48,49. The data for the 7 patients are summarized in Table 1 and S1, and Figure S1. All patients’ condition and dynamics were significantly different according to 2-way analysis of variance (Table S2).

Justification of the sample size. Access to patients in the ICU is limited, and the required duration of stay in the ICU (at least 7 days) required careful selection of study participants. Moreover, we set ourselves the task of conducting a comprehensive proteomic analysis without targeted search for specific proteins, and despite the low number of participants we could draw a conclusion about the stability of the proteome in ICU patients. Thus the null hypothesis was that the plasma proteome was stable despite significant changes in the patients’ condition. To test this hypothesis, assuming a 15% measurement error in the proteomics data, to obtain a 95% confidence level, the required sample size is 44. The seven patients provided 56 multivariate observation points; these were divided into four groups of 14 points each, corresponding to deteriorating vs. improving patient condition, early vs. late days, and those were significantly different by some clinical parameters (Table S2).

This study involving humans was conducted in accordance with the Declaration of Helsinki and was approved by the institutional review board of National Medical Research Center “Treatment and Rehabilitation Center” of the Ministry of Health of the Russian Federation (approval number: EC/052 from 04/12/2023). Written informed consent (blank form at the Supporting Information SI-2) was obtained from all participants or their legal representatives prior to enrollment.

Sample collection and batch selection

Patient blood samples were collected daily during their stay in the ICU. Blood samples were collected in tubes containing anticoagulant (EDTA), and immediately centrifuged at 800 g for 15 min to separate plasma, then at 3000 g for 20 min, at + 4 °C. The separated plasma was stored at − 80 °C until measurement. From all collected data, eight week-long batches of samples were used for proteomic analysis (Table S1).

Sample preparation

In total, 56 individual blood plasma samples obtained during 7 sequential days for 7 ICU patients (Patient 1 participated twice) and 32 bovine serum albumin samples (protein concentration was 70 g/l) were processed in a 96-well plate with Tecan Freedom EVO 150/8 liquid handling station in a single batch. Plasma samples were assigned random positions in a plate, 10 µl of each sample was transferred into well plate and processed. In the first step, the samples were peptized with urea (6 M), and disulfide bonds were reduced with dithiothreitol (13 mM) during 30-minute incubation at 37 °C, in the second step free thiol groups of cysteines were alkylated with freshly prepared iodoacetamide (40 mM) during 30-minute incubation at room temperature in the dark. Both reactions were performed in the presence of 300 mM Tris adjusted to pH 8.0 with hydrochloric acid. After that samples were diluted with the same buffer solution to drop urea concentration down to 0.55 M. 14 mg of bovine trypsin in 1 M calcium chloride prepared immediately before addition by in-house procedure to be published later was added to samples so that estimated weight ratio of protein to protease was 50 to 1 and calcium chloride concentration was 3.5 mM. After 1-hour incubation at 37 °C, the same amount of bovine β-trypsin (produced in-house from crystallized trypsin, Samson Med, Russia) solution was added, so that estimated protein to protease ratio became 25 to 1, and calcium chloride concentration was 7 mM. The samples were incubated overnight for at least 18 h. After incubation, the samples were diluted with trifluoracetic acid (TFA) and acetonitrile (ACN) solution, so that the final urea concentration was 150 mM, TFA was 1% volumetric, and acetonitrile was 5%. Internal standard mixture was added to all samples and calibration standard mixture was added to the samples of bovine sample albumin tryptic digest. The final sample volume was 1400 µl.

Peptides were extracted by solid-phase extraction using 96-well plate with C18 for desalting and Tecan Resolvex A200 positive pressure processor. All solvents used as a mobile phase were modified with 0.2% TFA and volumes of 600 µl were used for each step. The plate was conditioned with three volumes 5% ACN. The sample digests were loaded, and wells were washed three times with 5% ACN, and the bound peptides were eluted with a volume of 30%, 50% and 80% acetonitrile. The eluates were combined and dried using a speed vacuum concentrator. Plasma samples and calibrants were then resolubilized in 35 µl of 5% ACN with 0.1% TFA.

Panoramic LC-MS analysis

Samples were analyzed with 5 µl injection, thus equivalent of 1.42 ul of plasma was injected. LC-MS analysis was performed using an Orbitrap Exploris 480 mass spectrometer coupled with an Ultimate 3000 RSLCnano chromatographic system via a Nanospray Flex ion source (Thermo, USA).

Chromatographic separation of peptides was performed by reversed-phase chromatography using a C18 PepMap 100 trap column (Thermo, USA) and Peaky-C18 capillary (ReprosilPur C18 AQ, Dr. Maisch, Germany, 50 cm x 75 μm, 1.9 μm, Molecta, Russia) column at 60 °C. Loading solvent was modified with 0.1% of TFA and mobile phase solvents were modified with 0.1% formic acid. Each sample was injected onto trap column in 5% ACN at a flow rate of 10 µl/min for 12 min. Peptides were eluted with a flow rate of 0.25 µl/min with a gradient of solvent B (80% ACN) in solvent A (5% ACN). Gradient started with 5% of solvent B, increasing to 10% over 3 min, then to 55% over 55 min and to 65% over 3 min.

The source voltage was set to 2700 V, and capillary temperature to 275 °C. Mass spectrometer was operated in the data-dependent analysis in positive mode. In each cycle time was set to 1 s with one survey scan. Survey scan was recorded with resolution at 200 m/z set to 60,000 and mass-to-charge range from 200 to 1500 m/z. Fragment ion spectra were recorded with resolution of 15,000. Collision energy parameter was set to 30 and isolation window to 1.4 m/z. Funnel RF Level was set to 50, automatic gain control (AGC) parameter was set to “Standard” and injection time limit to “Auto” for both scan types.

Data processing

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository with the data set identifier PXD070474 (https://www.ebi.ac.uk/pride/archive/projects/PXD070474).

Targeted LC-MS data was loaded into Skyline software and exported as a table with peak areas for each transition in each sample. Exponential calibration curves (linear in log-log scale) were calculated for each peptide based on TIC area ratio of unlabelled to labelled peptides in BSA-based calibration samples and estimated concentrations were computed for each peptide in each sample by in-house with in-house R script.

Proteins were relatively quantified using MaxQuant software (2.3.1.0). Data was searched against sequences of reference human proteome (UniProt UP000005640) and common contaminants. Search settings were as follows: variable modifications included N-term acetylation, deamidation of asparagine and glutamine and oxidation of methionine residues, carbamidomethylation of cysteines was set as fixed modification, maximum number of missed cleavages was 2, minimum peptide length was 5.

Statistical analysis

To compare the proteins of the positive and negative groups, volcano plot analysis was used to visualize significant changes in the expression list. Significance was defined as a false discovery rate < 0.05 with log2 fold change >|0.5|. Differences among the clinical parameters were evaluated by 2-way ANOVA (α = 0.05), p-values < 0.05 were considered to indicate statistical significance (Table S2). Pooled OLS regression model, Fixed-effects and Random-effects models implemented in Python statmodels⁵⁰ were trained to explain protein normalized iBAQ levels either with mSOFA and Opinion (Normalized iBAQ_{p, t} ~ β1•mSOFA_{p, t} + β1•Opinion_{p, t} + µ_p + є_{p, t}, p – patient, t - time), or with Potassium and Platelet levels (Normalized iBAQ_{p, t} ~ β1•Potassium_{p, t} + β1•Platelet_{p, t} + µ_p + є_{p, t}), or with other clinical parameters alone or with Time (in days). For each protein σ_µ/< iBAQ > was considered as between-patient variance, and σ_є/< iBAQ > as within-patient variance. The association of proteins with other clinical data was assessed using hierarchical clustering based on Z-scores, implemented in Python Seaborn clustermap⁵¹. STRING DB enrichment analysis²² was performed for the selected protein groups.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: ANS IOB EYB SVT KSGMethodology: ANS IOB LAD IVK AVG KSGInvestigation: ANS IOB MMG NAK AVK OVK TAL ESVisualization: ANS MMGFunding acquisition: IOB LAD SVT KSGProject administration: SVT KSGSupervision: LAD KSGWriting – original draft: ANSWriting – review & editing: all authors.

Funding

The study was supported by the SBM Institute grant 122030900062-5.

Data availability

All data supporting the findings of this study are available within the paper and its Supplementary Information. Raw proteomic data are available via ProteomeXchange with identifier PXD070474 (https://www.ebi.ac.uk/pride/archive/projects/PXD070474).

Declarations

Competing interests

The authors declare no competing interests.