Proteomic changes associated with racial background and sepsis survival outcomes

Abstract

Intra-abdominal infection is a common cause of sepsis, and intra-abdominal sepsis leads to ~156 000 U.S. deaths annually. African American/Black adults have higher incidence and mortality rates from sepsis compared to Non-Hispanic White adults. A limited number of studies have traced survival outcomes to molecular changes; however, these studies primarily only included Non-Hispanic White adults. Our goal is to better understand molecular changes that may contribute to differences in sepsis survival in African American/Black and Non-Hispanic White adults with primary intra-abdominal infection. We employed discovery-based plasma proteomics of patient samples from the Protocolized Care for Early Septic Shock (ProCESS) cohort (N = 107). We identified 49 proteins involved in the acute phase response and complement system whose expression levels are associated with both survival outcome and racial background. Additionally, 82 proteins differentially-expressed in survivors were specific to African American/Black or Non-Hispanic White patients, suggesting molecular-level heterogeneity in sepsis patients in key inflammatory pathways. A smaller, robust set of 19 proteins were in common in African American/Black and Non-Hispanic White survivors and may represent potential universal molecular changes in sepsis. Overall, this study identifies molecular factors that may contribute to differences in survival outcomes in African American/Black patients that are not fully explained by socioeconomic or other non-biological factors.

Introduction

In the United States, more than 1.5 million Americans are diagnosed with sepsis annually, and ~16% of these patients do not survive within twelve days post-hospital admission.¹ Sepsis is life-threating organ dysfunction caused by a dysregulated host response to infection.² Intra-abdominal infections, a leading source of sepsis,^3–7 are highly heterogeneous, with a wide range of disease severity, pathogenic sources, and infection subtypes.⁸ Sepsis severity is associated with higher mortality from intra-abdominal infections, potentially due in part to a high number of pathogenic organisms that result in a pro-inflammatory state.^3,6–9

Racial and ethnic disparities exist in both the incidence and mortality rates of sepsis patients from a variety of backgrounds.^10–15 For instance, African American/Black patients have twice the incidence^10,16 of sepsis than Non-Hispanic White patients and have higher hospitalization¹⁷ and mortality^15,18–20 rates. Higher incidence in this population is observed across multiple primary causes of infection.^21–24 Factors such as socioeconomic status, chronic disease burden, healthcare access, quality of patient care, and aging contribute to disparities in sepsis incidence and mortality, but data suggest there are residual differences unexplained by these factors.^{12,15,16,18,19,25–27} Interestingly, differences in the host innate response that are collinear with racial and ethnic backgrounds have been observed.^28–36 For instance, zinc finger, metal binding, and iron homeostasis genes differed in Black and White community acquired pneumonia (CAP) patients.²⁸ Several differences in inflammatory and coagulation factors have also been reported between individuals of African and European/Caucasian ancestry.^29–34 Additionally, apolipoprotein L1 alleles are common in African American/Black adults and are associated with a higher risk of developing sepsis.^35,36 Therefore, molecular studies which investigate sepsis patient outcomes must be done using diverse cohorts to further understand the biological response to infection.

Previous molecular studies investigating sepsis patient outcomes have identified proteins and metabolites involved in immune response, coagulation, metabolism, and cellular assembly and movement pathways that consistently differ between sepsis survivors and non-survivors.^37–46 However, most of these studies have been conducted with small cohorts (N < 40) that are not racially or ethnically diverse and that focus on sepsis secondary to CAP. Whether or not there are any molecular changes associated with sepsis survival that can partially explain disparities in incidence or mortality in African American/Black patients has not been explored.

Herein, we employed quantitative proteomics to examine a patient cohort that included both Non-Hispanic White and African American/Black patients to identify proteins associated with differences in survival from sepsis due to intra-abdominal infection. Plasma samples were obtained from patients enrolled in the Protocolized Care for Early Septic Shock (ProCESS) cohort,⁴⁷ which was a randomized trial comparing alternative six-hour fluid resuscitation strategies in septic shock patients across 31 academic U.S. hospitals.⁴⁷ Effectiveness of these strategies was measured by the rate of death 60 days and 1 year post-emergency department (ED) admission, with treatment strategies showing equal effectiveness.⁴⁷ We chose the ProCESS cohort because it is diverse with regards to geographical location, patients’ racial and ethnic backgrounds, and primary cause of sepsis. This study focuses on the subset of patients that developed sepsis from primary intra-abdominal infections (N = 107 total) to minimize effects due to heterogeneity in infection source. This is the first study to evaluate if molecular changes can partially explain disparities in sepsis survival outcomes for African American/Black patients.

Materials and methods

Ethics statement, study design, patients, & plasma collection

The Institutional Review Boards at the University of Pittsburgh and all 31 hospitals approved the parent trial, and this subanalysis was approved by Vanderbilt University. We conducted a study using patients enrolled in ProCESS, a randomized, multicenter trial comparing alternative six-hour fluid resuscitation strategies for septic shock.⁴⁷ From March 2008 to May 2013, 1341 septic shock patients were enrolled across 31 U.S. EDs. Details and eligibility criteria of ProCESS have been described previously.^47,48

A total of 107 ProCESS patients were selected for this study based on self-identified race (Non-Hispanic White or African American/Black), primary infection source (intra-abdominal infection), and availability of plasma. Site and etiology of infection were based on criteria previously described.⁴⁸ Plasma samples were collected at ED admission when patients were enrolled in the ProCESS cohort and prior to treatment (“time zero”). Survivorship was defined as a person’s status at 90 days post-ED admission. There was no reported impact of resuscitation strategy on patient survival outcomes.⁴⁷ Therefore, treatment group was not included as a covariate in this study.

Plasma samples were analyzed using a discovery-based, quantitative proteomics workflow with automation of sample preparation steps in a 96-well plate format. Plasma was depleted of abundant proteins, and TMTpro-labeled peptides were subjected to high pH reverse-phase (RP) fractionation. Fractionated peptides were further separated and detected with liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis (Fig. 1). Details are described below.

Fig. 1.

Plasma proteomics workflow. Samples from each group were randomized into eight batches such that each batch contained one quality control pool and at least one sample from each of the four sample groups (Non-Hispanic White survivors, Non-Hispanic White non-survivors, African American/Black survivors, and African American/Black non-survivors). The BCA assay, digestion, desalting, and TMTpro labeling steps were performed using a Biomek i7 liquid handling workstation, allowing for six batches to be prepared simultaneously. After TMTpro pooling, batches were fractionated and analyzed via LC-MS/MS, and raw data was processed with Proteome Discoverer 2.3. Figure was created with BioRender.com. Abbreviations: MARS-14 = Multiple Affinity Removal System Top 14; BCA = bicinchoninic acid; TMTpro = Tandem Mass Tags pro (16-plex); RPLC = reverse phase liquid chromatography; LC-MS/MS = liquid chromatography-tandem mass spectrometry.

Immunoaffinity depletion of high-abundant proteins

The multiple affinity removal system Hu-14 (MARS-14) column (Agilent; Santa Clara, CA, USA) was used to deplete albumin; immunoglobulins (Ig) G, A, and M; transferrin; haptoglobin; antitrypsin; fibrinogen; α2-macroglobulin; α1-acid glycoprotein; apolipoproteins AI and AII; complement C3; and transthyretin. Two 40 μL aliquots of crude plasma were each injected onto a MARS-14 column following the manufacturer’s protocol. Unbound fractions were concentrated using a 10k molecular weight cutoff concentrator (Millipore Sigma; St. Louis, MO, USA) at 14 000 × g at 4 °C for 12 min. The two concentrated aliquots were combined into one sample and frozen at −80 °C until further analysis. Protein concentrations were measured using a bicinchoninic acid (BCA) protein assay (Thermo Fisher Scientific; Waltham, MA, USA). A quality control (QC) sample was made by pooling equimolar amounts of each sample (N = 107) at the protein level.

Protein digestion & desalting

Digestion, desalting, and labeling steps were performed on a Biomek i7 liquid handling workstation (Beckman Coulter; Atlanta, GA, USA). Protein (60 μg) was denatured in 100 mM ammonium bicarbonate buffer, reduced with 200 mM dithiothreitol (Thermo Fisher Scientific; Waltham, MA, USA) for 55 min at 55 °C and alkylated with 200 mM iodoacetamide (Thermo Fisher Scientific; Waltham, MA, USA) for 30 min at room temperature. Each sample was incubated with trypsin/Lys-C (Promega; Madison, WI, USA) at 50 : 1 sample : enzyme ratio for 14 h at 37 °C and then acidified with 5% formic acid (FA). Digested samples were desalted on a TARGA C18 filter plate (The Nest Group; Ipswich, MA, USA) using a positive pressure apparatus. Samples were dried using centrifugal evaporation and frozen at −80 °C.

Tandem mass tag (TMT) labeling & batch design

Samples were randomly assigned to 8 TMTpro batches, with 15 patient samples plus a QC sample in each batch (see Table S1 for batch assignments, ESI†). Batches were balanced across numbers of patient survivors and non-survivors and patients’ racial/ethnic backgrounds and ages.

Each sample was reconstituted in 100 mM triethyl ammonium bicarbonate (Millipore Sigma; St. Louis, MO, USA), and 25 μg was labeled with a TMTpro reagent following the manufacturer’s protocol (Thermo Fisher Scientific; Waltham, MA, USA) using the Biomek i7 workstation. Each TMTpro reagent was solubilized in anhydrous acetonitrile (ACN) and added to a sample in a 1 : 8 sample : tag ratio. After 1 h of incubation, the reaction was quenched with 5% hydroxylamine (Millipore Sigma; St. Louis, MO, USA). Samples were pooled and the ACN dried via centrifugal evaporation. Pooled samples were then acidified with 5% FA, desalted as previously described, and frozen at −80 °C.

Basic RPLC offline fractionation

Pooled samples were reconstituted in loading buffer (1.0 mM ammonium formate, 2% ACN, pH 10) and loaded (295 μg) onto a Zorbax 300-Extend-C18 column (Agilent; Santa Clara, CA, USA). The gradient was as follows: 0–7 min, 0% B; 7–13 min, 0–16% B; 13–73 min, 16–40% B; 73–77 min, 40–44% B; 77–82 min, 44–60% B; 82–96 min, 60% B. Mobile phase A was 4.5 mM ammonium formate, 2% ACN (pH 10), and mobile phase B was 4.5 mM ammonium formate, 90% ACN (pH 10).⁴⁹ One fraction was collected per minute, for a total of 96 fractions. These fractions were pooled into a final set of 24 concatenated fractions, which were dried by centrifugal evaporation. Fractions were reconstituted in LC-MS grade water with 0.1% FA and frozen at −80 °C.

LC-MS/MS analysis

Each fraction (1 μg) was injected onto a Dionex UltiMate 3000 LC system coupled to a Q-Exactive HF mass spectrometer (both Thermo Fisher Scientific; Waltham, MA, USA) operated in positive mode. Peptides were loaded onto a commercial C18 trap column (75 μm i.d. × 2 cm, 100 Å, 3 mm; Thermo Fisher Scientific, Waltham, MA, USA) prior to separation on an in house C18 packed column (100 μm i.d. × 30 cm, 100 Å, 2.5 μm) with a pulled tip over the following 185 min gradient: 0–14 min, 2% B; 14–17 min, 2–16% B; 17–145 min, 16–25% B; 145–150 min, 25–85% B; 150–158 min, 85% B; 158–160 min, 85–4% B; 160–185 min, 4% B. Mobile phase A was 0.1% FA and mobile phase B was 0.1% FA in ACN. Full MS spectra were collected in the Orbitrap (350–1500 m/z, 120 000 resolution, automated gain control [AGC] 3 × 10⁶, maximum injection time 50 ms). The instrument was operated in data-dependent acquisition (DDA) mode to acquire the top 15 MS/MS spectra in the Orbitrap using higher-energy collisional dissociation (normalized collision energy 28%, isolation width 0.7 m/z, AGC 1 × 10⁵, 45 000 resolution) and dynamic exclusion of 30 seconds. Fractions were analyzed in duplicate injections.

Data analysis

RAW files were analyzed using Proteome Discoverer v2.3 software (Thermo Fisher Scientific; Waltham, MA, USA). All technical replicates and fractions from each batch were combined into one result file and searched against the UniProt human reviewed protein database (01/14/21, 42 294 sequences) using SEQUEST-HT. The following modifications were included: fixed modification of cysteine carbamidomethylation and variable modifications of methionine oxidation, TMTpro on lysine residues and peptide N-termini, acetylation and methylation on protein N-termini, and methionine loss with acetylation. A maximum of two trypsin miscleavages were allowed. TMTpro was set as the quantification method in Proteome Discoverer, and reporter ion quantification was based on intensity with a reporter ion signal-to-noise threshold of 10. The selected protein groups will be referred to as proteins in the rest of this analysis. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (https://proteomecentral.proteomexchange.org/) via the PRIDE⁵⁰ partner repository with the dataset identifier PXD030476.

Peptides were filtered to only include those identified with high confidence (<1% false discovery rate [FDR]) and their corresponding proteins. This protein list was further filtered to those with at least two peptide spectral matches (PSMs). TMT reporter ion intensities of quantified proteins were normalized using a modified two-step process involving within-batch and across-batch normalization.^51,52 Within-batch normalization was based on the intensity of the pooled sample, and across-batch normalization was based on the geometric mean of the pool’s reporter ion intensities. These normalized data were used for further statistical analysis.

We used linear regression methods to identify protein factors associated with survival and race. In the regression model, the protein intensity was the dependent variable. For each protein factor, two regression models were fitted: one with only 90 day survival (yes or no) as the independent variable, and one with survival, race, and the interaction between them as independent variables. Because the time point of sample collection was prior to randomization into treatment group and because treatments showed equal effectiveness,⁴⁷ treatment group was not included as a covariate in this study. All models were also adjusted for age, and robust standard errors were calculated to account for the batch clusters. We excluded N = 901 proteins with reporter ion values missing in QC channels and/or that were detected in 20 samples or less (<19% of all 107 samples) due to small sample size. Mean group difference with 95% confidence interval, as well as p values, were reported for each included protein and each comparison. p values were adjusted using FDR to account for multiple comparisons. Adjusted p values <0.05 were considered statistically significant. Analyses were performed using R version 4.1.1 and the rms package.⁵³

Fold change values were calculated for each protein by dividing the mean of TMT levels of survivors’ by the mean TMT levels of non-survivors’. Fold change cutoffs were established using a power analysis strategy previously described.⁵⁴ Briefly, at a power of 0.8 and significance level of 0.05, this strategy accounts for technical variation (i.e., variation in TMT reporter ion abundances for each protein across technical replicates), biological variation (i.e., variation in TMT reporter ion abundances for each protein across biological replicates, or patients within the same sample group), and the number of technical and biological replicates. Specific criteria for this study included: (1) N = 2 technical replicates, and (2) the smallest possible number of biological replicates in a patient subgroup. Resulting fold change cutoffs were ≥1.48 or ≤0.67 for comparisons of survivors vs. non-survivors; ≥1.62 or ≤0.62 for comparisons of Non-Hispanic White survivors vs. non-survivors; and ≥2.52 or ≤0.40 for comparisons of African American/Black survivors vs. non-survivors. These cutoffs were overly strict for this analysis, as we used the cutoff corresponding to the protein for which the smallest number of patients (biological replicates) in that sample group had reporter ion values.

Ingenuity Pathway Analysis (IPA) was used to identify significant biological pathways and functions (Benjamini–Hochberg [B–H] corrected p < 0.05). Search Tool for Retrieval of Interacting Genes/Proteins (STRING)⁵⁵ was also used to identify clusters of differentially-expressed proteins and significantly enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) or reactome pathways at medium confidence (minimum interaction score = 0.400).

Western blot analysis

To measure plasminogen (PLG) and C-reactive protein (CRP), which are implicated in intra-abdominal infections and sepsis,^56–59 Western blot analysis was performed. Ten samples were randomly selected from each of the four sample groups, except for African American/Black non-survivors (N = 4). Proteins were fractionated by SDS-PAGE (120 V loading, 160 V for ~70 min) and transferred to a nitrocellulose membrane using a wet transfer at 100 V for 70 min. After incubation with 5% non-fat milk in Tris-buffered saline with Tween-20 (TBST; 50 mM Tris, 150 mM NaCl, pH 7.5, 0.1% Tween-20) for 60 min, the membrane was washed 4 × 4 min with TBST and incubated overnight with antibodies against CRP (Abcam ab227507, 1 : 10 000, Cambridge, UK) or PLG (Abcam ab180046, 1 : 1000, Cambridge, UK) at 4 °C. The membrane was washed 4 × 4 min with TBST and incubated with a 1 : 10 000 dilution of fluorescent-labeled anti-rabbit (IRDye 800CW; Li-Cor Biosciences, Lincoln, NE, USA) secondary antibody for 30–60 min. Blots were washed 4 × 4 min with TBST prior to imaging using a ChemiDoc MP imaging system (Bio-Rad; Hercules, CA, USA). ImageLab v6.1 software (Bio-Rad; Hercules, CA, USA) was used for protein band quantification.

Results

Patient characteristics

Overall, of the total 107 patients in this study, 62 (58%) survived and 45 (42%) did not survive 90 days post-ED admission. Demographic characteristics were collected upon enrollment in ProCESS and randomization into a treatment group (Table 1). Survivors were significantly younger than non-survivors (p = 0.003) by eight years, and notably, the number of surviving African American/Black patients was lower than the number of surviving Non-Hispanic White patients. There were no significant differences in frequencies of kidney and liver comorbidities between survivors and non-survivors.

Table 1.

Demographics of the patients selected for this study

Characteristic	Survivorsc	Non-survivorsc	Statistic
Non-Hispanic White/African American/Black (N)	41/21	38/7
Agea (years)	57 ± 15	65 ± 14	0.003
Renal failureb (N, %)	13 (21.0%)	13 (28.9%)	0.889
History of chronic dialysisb (N, %)	7 (11.3%)	4 (8.9%)	0.163
History of cirrhosisb (N, %)	7 (11.3%)	10 (22.2%)	2.332
History of hepatic failure, encephalopathy, or comab (N, %)	2 (3.2%)	4 (8.9%)	2.510
History of chronic hepatitis or other liver diseaseb (N, %)	8 (12.9%)	12 (26.7%)	3.250

^aAverage ± 1 standard deviation; p-value statistic was calculated using a two-tailed t-test.

^bNumber of patients who responded “yes” and percentage of total within survival outcome group; χ² statistics were calculated using contingency table analysis.

^cNumber of plasma samples. Bold denotes significant differences between survivors and non-survivors (p < 0.05 or χ² > 3.841).

Differentially-expressed proteins associated with survival

In total, 1639 proteins (82 320 peptides, 820 209 PSMs) from an initial 2540 proteins were analyzed with linear regressions (survival only model; Fig. 2 and Fig. S1, ESI†). After FDR adjustment (adjusted p < 0.05), 180 proteins (11.0% of the 1639 proteins in regression) had significant fold changes between survivors vs. non-survivors (Fig. 2 and Table S2, ESI†). The most statistically significant pathways associated with survivors were: hepatic fibrosis/hepatic stellate cell (HSC) activation (HF/HSCA), tumor microenvironment, wound healing signaling, and glycoprotein VI signaling (Fig. S2, ESI†). Other pathways are broadly related to inflammation and metabolism (Fig. S2, ESI†) and are consistent with prior reports of changes in sepsis survivors.^37–46 Furthermore, PLG and CRP, proteins that are commonly implicated in sepsis,^56,57 were verified as positive controls (raw images in Fig. S3 and processed images in Fig. S4, ESI†). PLG levels were significantly higher in survivors compared to non-survivors (Fig. S4B, ^ESI,^† p = 0.038), and CRP levels were not significantly different between survivors and non-survivors (Fig. S4D, ESI†). Similar results were obtained with race-stratified data (Fig. S4C and E, ESI†).

Fig. 2.

Data analysis workflow. TMT reporter ion intensities for proteins were normalized and analyzed using linear regression models with and without race stratification (see experimental methods). Proteins were first filtered to those with p < 0.05, then to those with an FDR-adjusted p < 0.05 and significant fold change values (see experimental methods). Figure was created with BioRender.com.

Protein levels between survivors and non-survivors in Non-Hispanic White and African American/Black patients were then independently compared using linear regression analyses of (1) survival outcome from race-stratified data and (2) the interaction of survival outcome and race.

Differentially-expressed proteins associated with survival outcome within racial groups

In the first comparison, after FDR adjustment (adjusted p < 0.05), there were 55 (3.4% of the 1639 proteins in regression) and 65 (4.0%) differentially-expressed proteins when comparing survivors and non-survivors in either Non-Hispanic White or African American/Black patients, respectively (Table S3, ^ESI,^† and Fig. 3A and B). A majority of these differentially-expressed proteins were lower in survivors (i.e., 51 proteins [3.1%] in Non-Hispanic White survivors and 61 proteins [3.7%] in African American/Black survivors). Only four differentially-expressed proteins were higher in survivors of both Non-Hispanic White or African/American Black backgrounds (Fig. 3A–C).

Fig. 3.

Analysis of data stratified by race. Volcano plot of protein log₂ of averaged TMT reporter ion abundances’ ratio of survivors to non-survivors as a function of p-value (N = 1639 proteins in the regression model) in (A) Non-Hispanic White and (B) African American/Black patients. Light grey data points indicate proteins with insignificant p-values (p ≥ 0.05) and/or insignificant fold change values (1.67 > fold change > 0.62 in Non-Hispanic White patients and 2.52 > fold change > 0.40 in African American/Black patients). (A) Pink data points indicate proteins with both significant p values (p < 0.05) and fold change values (≥1.67 or ≤0.62) in Non-Hispanic White patients (N = 55 proteins). (B) Blue data points indicate proteins with both significant p values (p < 0.05) and fold change values (≥2.52 or ≤0.40) in African American/Black patients (N = 65 proteins). Dashed lines indicate p-value and fold change cutoffs. Example proteins are highlighted. (C) Venn diagram of the overlap of differentially-expressed proteins (pink and blue proteins from (A) and (B)) between survivors and non-survivors in Non-Hispanic White and African American/Black patients. (D) IPA canonical pathway analysis of the proteins in (C) (i.e., N = 55 differentially-expressed proteins in Non-Hispanic White patients and N = 65 differentially-expressed proteins in African American/Black patients). Canonical pathways were identified from the IPA library using Fisher’s exact test adjusted for multiple hypothesis testing using the Benjamini–Hochberg (B–H) correction. * indicates pathways significant in both racial groups, # indicates pathways significant in Non-Hispanic White patients only, and + indicates pathways significant in African American/Black patients only. Pink bars refer to Non-Hispanic White patients, and blue bars refer to African American/Black patients. Abbreviations: IL1R2 = interleukin-1 receptor type 2; VCAM1 = vascular cell adhesion molecule 1; CCN2 = cellular communication network family member 2; IL1RL1 = interleukin-1 receptor like 1; BCHE = cholinesterase; IGFALS = insulin-like growth factor-binding protein complex acid labile subunit; IGFBP3 = insulin-like growth factor-binding protein 3; GPLD1 = phosphatidylinositol-glycan-specific phospholipase D; HRG = histidine rich glycoprotein; IGF2 = isoform 2 of insulin-like growth factor II; C4BPA = C4b-binding protein α chain; B–H = Benjamini–Hochberg; IGF-1 = insulin-like growth factor-1; Th1 = T-helper 1; Th2 = T-helper 2.

Nineteen (1.2%) proteins in survivors were differentially-expressed in both African American/Black and Non-Hispanic White patients (Fig. 3C). The directionality of change in survivors is mostly lower in expression and is also similar in both racial groups. It is likely that these 19 proteins represent proteins universally related with sepsis patient survival outcomes regardless of racial and ethnic background. Eight of these 19 proteins are in STRING networks in both racial groups, although they are connected differently between racial groups (Fig. S5A and B, ESI†). These eight proteins are primarily involved in antigen processing and presentation, phagocytosis, and immunoregulatory interactions between lymphoid and non-lymphoid cells. Interestingly, there are also differentially-expressed proteins in survivors that are specific to African American/Black (N = 46 proteins [2.8%]) or Non-Hispanic White (N = 36 proteins [2.2%]) patients (Fig. 3C). In either racial group, we consistently observe that most differentially-expressed proteins are lower in survivors.

HF/HSCA is the most significant pathway associated with proteins in race-stratified comparisons (Fig. 3D), and this pathway is consistent with or without race stratification (Fig. S2, ^ESI,^† and Fig. 3D). The most differentially-expressed proteins in this pathway in either racial group are highlighted in Fig. 3A and B: cellular communication network family member 2 (CCN2), interleukin (IL)-1 receptor type 2, vascular cell adhesion molecule 1 (VCAM1), and IL-1 receptor-like 1. Insulin-like growth factor-1 (IGF-1) signaling and the tumor microenvironment pathways were significant pathways represented in Non-Hispanic White patients. T-helper (Th) 1 and 2 activation, wound healing signaling, Th2 pathway, and the complement system were significant pathways represented in African American/Black patients. Similar biological processes and network clusters were identified with STRING.

Proteins with significant race–survival interaction

In the second comparison, we performed a linear regression with a race–survival interaction term, which resulted in 49 proteins (3.0% of the 1639 proteins in regression) displaying a statistically significant race–survival interaction (adjusted p < 0.05, Table S4, ESI†). In a heat map analysis, two clusters of proteins were observed: a small cluster that is higher in survivors and a larger cluster that is lower in survivors (Fig. 4A). Patients generally clustered based on survival outcome. However, some proteins clustered by racial background of the patient, suggesting heterogeneity at the molecular level with this patient cohort.

Fig. 4.

(A) Heatmap with clustering of N = 49 proteins with significant race–survival interactions (FDR-adjusted p < 0.05). ClustVis⁶⁰ was used to create a heatmap from the log₁₀ of TMT reporter ion intensities for these proteins across N = 107 patients. The columns correspond to individuals, while the rows correspond to proteins. Rows are centered; unit variance scaling is applied to rows. Imputation is used for missing value estimation. Both rows and columns are clustered using correlation distance and average linkage. (B–E) Box plots of log₁₀ TMT intensities of example proteins with significant race–survival interactions (FDR-adjusted p < 0.05): (B) C4b-binding protein α chain (C4BPA); (C) C4b-binding protein β chain (C4BPB); (D) serum paraoxonase/arylesterase 1 (PON1); (E) isoform 2 of human leukocyte antigen (HLA) class I histocompatibility antigen, A α chain (HLA-A). The horizontal line in the box represents the median and the X represents the mean. * indicates p < 0.05 and ** indicates p < 0.01, as determined by 2-way ANOVAs with replication and Tukey–Kramer post hoc tests. For both the heatmap and box plots, green represents Non-Hispanic White non-survivors, purple represents Non-Hispanic White survivors, red represents African American/Black non-survivors, and blue represents African American/Black survivors.

Half of the proteins with significant race–survival interaction (described above) are also differentially-expressed in the first race-stratified comparison. Example proteins with statistically significant race–survival interaction are shown in Fig. 4B–E. C4b-binding protein α (C4BPA) and β chains (C4BPB; Fig. 4B and C, respectively) are examples of proteins whose association with survival are specific to one of the racial groups. We note, however, that both adjusted p and fold change values are only significant in C4BPA in African American/Black patients. Serum paraoxonase/arylesterase 1 (PON1; Fig. 4D), appears higher in survivors of both racial groups, which is consistent with other studies.^61–65 Isoform 2 of human leukocyte antigen (HLA) class I histocompatibility antigen, A α chain (HLA-A) is decreased in survivors of both racial groups (Fig. 4E).

These 49 proteins are involved in acute phase response (APR) signaling and the complement system. More specifically, natural killer (NK) cell mediated cytotoxicity was an enriched KEGG pathway in STRING analysis. STRING also identified a local network cluster of proteins with functions related to the complement and coagulation cascades and lipoprotein particles. Type I interferon binding and receptor activity was also enriched (Fig. S6, ESI†).

Discussion

Alterations in inflammation and coagulation biomarkers (IL-6, tumor necrosis factor, IL-10, thrombin–antithrombin complex, and D-dimer) between survivors and non-survivors were previously identified in the ProCESS cohort using enzyme-linked immunoassays (ELISAs) and latex agglutination assays.⁶⁶ However, whether other molecular changes or pathways are associated with sepsis survival in a diverse cohort have not been investigated. In this study, we used a semi-automated plasma proteomics workflow to analyze plasma samples (N = 107) from African American/Black and Non-Hispanic White sepsis survivors and non-survivors with primary intra-abdominal infection from the ProCESS cohort.⁴⁷

HF/HSCA was the most significant pathway represented between survivors and non-survivors. HF/HSCA, its proteins, and related inflammatory molecules and pathways have previously been reported as altered in sepsis survivors compared to healthy controls³⁸ and in sepsis and inflammatory studies.^42,67–76 HF/HSCA changes may be an early sign of liver dysfunction, which can affect patient outcomes.^77,78 Other pathways that are significantly represented in survivors in this study are those related to lipid metabolism, such as liver X receptor/retinoid X receptor activation and atherosclerosis signaling, which are commonly implicated in proteomic studies of sepsis secondary to CAP or hospital-acquired pneumonia.^{37,38,42,54,61–65} These previous studies that have observed similar findings have included proteomics, transcriptomics, genomics, and metabolomics strategies.

We note that ~25% of the differentially-expressed proteins we observed were <30 kDa, and kidney function has been shown to impact the plasma levels of these low molecular weight proteins.⁷⁹ However, because we did not observe significant differences in the frequencies of renal failure or history of dialysis between survivors and non-survivors, kidney function is likely not a driver of observed proteomic differences based on survival outcome.

Molecular pathways associated with survival outcomes in African American/Black patients

Sepsis mortality rates from intra-abdominal and other causes of primary infection are disproportionately high in underrepresented populations, such as African American/Black adults, and are not completely explained by differences in socioeconomic factors.^{12–16,18,19,25–27} To understand potential molecular aspects of disparities in sepsis mortality rates amongst African American/Black adults, we re-analyzed our proteomics data by stratifying patients based on self-identified race and conducting linear regressions with an interaction term of survival outcome and race.

In these analyses, HF/HSCA was still the most significant pathway represented in both racial groups. A majority of differentially-expressed proteins in the HF/HSCA pathway were decreased in survivors of African American/Black and Non-Hispanic White racial groups. In this pathway, CCN2 and VCAM1 changed consistently in survivors from both racial groups and should be further explored as potential prognostic biomarkers associated with patient survival. Decreased levels of these proteins in survivors suggest less pro-inflammatory activity, specifically less angiogenesis, proliferation of chondrocytes, leukocyte migration, and collagen production.^67–70

Notably, VCAM1 levels were increased in phenotypes associated with higher inflammation and mortality in the ProCESS cohort.⁸⁰ VCAM1 was decreased in survivors in a cohort inclusive of African American/Black patients⁷⁰ and in Taiwanese patients^81,82 via Luminex immunoassays or ELISA and is overall a predictor of sepsis mortality.^83,84 RNA-sequencing analysis (RNA-seq) and Western blotting have indicated that VCAM1 levels directly correlate with inflammation in control mice and mice with systemic inflammatory response syndrome.⁸⁵ Additionally, similar trends have been demonstrated with CCN2 using mRNA expression in sham and cecal ligature and puncture rats,⁶⁸ and with Western blotting in the peritoneal cavity of peritoneal dialysis patients.⁶⁹ These previous studies, both at the proteomic and transcriptomic level, support our findings that VCAM1 and CCN2 could potentially be prognostic biomarkers for patients from a variety of racial/ethnic backgrounds.

However, several pathways related to the host immune response were significantly represented in African American/Black survivors, specifically the complement system and pathways related to T cell activation and proliferation. On the other hand, IGF-1 and tumor microenvironment pathways were significantly represented in Non-Hispanic White survivors. Although these pathways have not been discussed previously in African American/Black sepsis patients, RNA-seq has implicated genes related to Th2 induction and complement activation in African American patients with atopic dermatitis.⁸⁶ ELISAs have shown that Th2-related cytokines lead to disease flares in African American systemic lupus erythematosus (SLE) patients.⁸⁷ Single nucleotide polymorphisms (SNPs) in genes important for Th1 and Th2 immunity and T regulatory cell-mediated immunosuppression have been identified in African American women that greatly increase breast cancer risk.⁸⁸ Our findings extend the implication of T cell and complement pathways in various immune-related diseases to sepsis in African American/Black patients. Our findings may point to a specific difference in the adaptive immune response between African American/Black and Non-Hispanic White patients that may partially explain differences in patient response and ultimately sepsis survival outcomes.

We also identified 49 proteins with significant race–survival interactions, simply implying that race and survival outcome together influence protein levels. These 49 proteins are involved in the APR and complement system, the latter of which is also significantly represented in African American/Black patients. Some of these proteins, such as PON1 and PLG, have similar changes in protein level in survivors from both racial groups. On the other hand, C4BPA and C4BPB are examples of proteins which appear to change differently in survivors based on racial background, further implicating the complement system (specifically complement regulation) as a point of potential difference between racial groups. Additionally, interferon α/β receptors 1 and 2, HLA-A, and leukocyte immunoglobulin-like receptor subfamily B member 4 have significant race–survival interaction terms. This finding suggests that type I interferon binding and receptor activity and NK-cell mediated cytotoxicity are also specific aspects of the innate immune response that may be different between racial groups. These findings may also implicate downstream pathways, such as the Janus kinase-signal transducer and activator of transcription (JAK-STAT),⁸⁹ as areas of the immune response that differ between racial groups and could be explored in future studies.

Gene expression profiling of neonatal sepsis patients has revealed that in early sepsis, several genes related to interferon 1 signaling are decreased in survivors and that increased interferon 1 signaling is associated with mortality.⁹⁰ Additionally, plasma levels of type I interferons were increased in African American patients with SLE compared to White patients with SLE in a luciferase assay.⁹¹ Conversely, genotyping of SLE patients revealed a SNP in the MAVS gene, which can induce type I interferon expression, that was found almost exclusively in African American patients and that correlated to low type I interferon and pro-inflammatory molecule production.⁹² Correspondingly, single-cell RNA seq showed an association between higher degrees of European ancestry and increased type I interferon activity in early infection of influenza.⁹³ In regard to the HLA system, a PCR-based study revealed that 11 alleles of HLA-A were frequently observed in African Americans, but were observed with low frequencies or not observed at all in other racial/ethnic groups in the U.S.⁹⁴ Different polymorphisms in HLA-A based on race/ethnicity can affect susceptibility to different immune-related diseases, such as multiple sclerosis and S. aureus infections.^95–97 Although these previous studies have either been in neonatal sepsis patients or other immune-related diseases, they provide genomic and transcriptomic support for variations in the type I interferon signaling pathway and HLA system based on race/ethnicity and the subsequent impacts on adult sepsis survival that we observed.

While highlighting similarities between racial groups, our study also suggests immune response differences between African American/Black and Non-Hispanic White survivors. Race-specific and common proteins across both groups highlight proteome heterogeneity in sepsis patients and the need for larger studies with patients from diverse racial/ethnic backgrounds. Diversity in patient cohorts will help to better explain residual differences in patient outcomes.

Langley et al. also have studied sepsis survivorship in a diverse cohort using an integrated plasma proteomics and metabolomics approach.⁴² Their cohort included a majority of Black CAP patients, but also included White patients.⁴² Fifty-six proteins were differentially-expressed between survivors and non-survivors at ED admission in their study and were primarily involved in the complement and coagulation systems and fatty acid transport.⁴² Their findings in sepsis patients from CAP were generally consistent with our study in sepsis patients from intra-abdominal infection. Proteins such as fatty acid binding protein 4 and Fc fragment of IgG binding protein are decreased in survivors from both studies (Tables S2 and S3, ESI†).⁴²

The limited number of proteomic studies with diverse sepsis patient groups, along with consistency of studies to date (i.e., racial background, primary infection source),^{37–40,42,43,98} requires further exploration of the role of HF/HSCA, APR, coagulation, complement, and lipid metabolism in sepsis. Future studies should include a larger number of African American/Black patients and explore changes in other primary infection sources such as urinary tract infection. Additionally, targeted proteomic studies could be used to characterize changes in the aforementioned pathways in African American/Black patients in comparison to Non-Hispanic White patients.

Future studies could also extend to other ’omics to further explore differences in sepsis survival outcomes, particularly in the immune system. Previously in a transcriptomic study comparing gene activation in Black and White CAP patients that did not explore sepsis risk or outcome from CAP, zinc finger, metal binding, and homeostasis genes were differentially activated between the racial groups.²⁸ We did not identify proteins with similar functions as differentially-expressed in the present study, possibly due to a difference in primary infection source and/or differences in cohort design related to disease severity. Another transcriptomic study found 28 inflammation/immune system genes that were differentially-expressed between survivors and non-survivors of sepsis.⁹⁹ The majority of significant proteins in our study also have inflammatory/immune-related functions, although they do not correspond directly to the genes identified previously. This could be because the previous study collected samples between days 2–4 post-ED admission and only studied Caucasian patients.⁹⁹ Additionally, mRNA levels do not correlate well with protein expression during stress responses like sepsis.¹⁰⁰ Sepsis-related inflammatory changes are also highly variable on a transcriptional level.¹⁰¹ Previous genomic studies have associated apolipoprotein L1 alleles that are common in African American/Black adults with a higher sepsis risk and disease severity,^35,36 but not mortality.³⁵

Study strengths and limitations

A major strength of this study is the inclusion of plasma samples from both Non-Hispanic White and African American/Black patients. Biological pathways implicated in this work are consistent with prior studies, and key proteins associated with survival are also consistent across racial background. The primary cause of infection, intra-abdominal, is well established, and the ProCESS cohort is geographically diverse, which are additional study strengths.

Our group has previously demonstrated that age affects the plasma proteome of 50–65 and 70–85 year old sepsis patients, with differences noted between these age groups in APR signaling, coagulation, lipid metabolism, and reactive oxygen species generation pathways.⁵⁴ Therefore, age was included as a covariate in regression models. All ProCESS hospitals were academic, which eliminates hospital type as a contributing factor to differences observed in this study, as it has been shown to impact sepsis disparities.^27,102 We also only analyzed samples from patients with primary intra-abdominal infection to eliminate variability in infection source.

We utilized a randomized, blinded study design and ensured samples from each study group were included in every TMT batch in order to minimize error and bias in sample preparation. We opted to use MS/MS quantification to increase identifications, which is reasonable for an exploratory study. MS/MS methods produce higher identifications and fewer missing values than MS³ methods with similar levels of quantitative precision.^103–105

Limitations of this study begin with our sample size, which was a total of N = 107. While this is larger than most similar studies examining proteomic differences between sepsis survivors and non-survivors,^{37–41,43,46,98} we note that this sample size could be improved. Sufficient sample size is critical due to the heterogeneity within and across racial and ethnic groups,^26,42 and more representation in patient cohorts is necessary.^{37–40,43,98}

Prior to our regression analysis, 901 proteins (35.5% of identifiable proteins) were filtered out due to missing TMT values. This loss may be due to batch effects caused by the stochastic nature of DDA.¹⁰⁶ Data-independent acquisition strategies would potentially improve the dynamic range of plasma proteins measurable.^107,108

Another limitation is the study design which did not control for co-variate factors (e.g., sex,^{13,15,17,20,21} geographical region,^{13,15,20,25,109,110} income,^11,16 education,^11,111,112 health insurance,^26,113,114 and the presence of comorbidities [such as cardiovascular disease,^13,115 diabetes,^{12,13,16,21,27,113} and HIV/AIDS^14,21,27,113]) that may impact patients’ plasma proteomes and contribute to sepsis disparities.

Conclusions

Plasma proteomics analysis was conducted on a racial and ethnically diverse cohort of samples from sepsis survivors and non-survivors with primary intra-abdominal infection. Overall, this study highlights liver cell activation as a key pathway that can differentiate sepsis survivors from non-survivors at the point of ED admission while also emphasizing the role of inflammation and lipid metabolism in sepsis survival. Liver cell activation was significantly represented in patients regardless of racial/ethnic background. We also identified several differentially-expressed proteins that were specific to Non-Hispanic White or African American/Black survivors. Proteins with significant race–survival interactions were related to the acute phase response and complement system, which are key inflammatory pathways related to sepsis. Additionally, several pathways related to the host immune response were significantly represented in African American/Black patients. These findings may help explain the residual differences in patient outcome unexplained by socioeconomic factors and highlight the need for future studies in diverse patient cohorts.

Data availability

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (https://proteomecentral.proteomexchange.org/) via the PRIDE partner repository with the dataset identifier PXD030476.