Disruption of the Circulating Proteome in Neonates Receiving Extracorporeal Membrane Oxygenation Following Congenital Heart Disease Surgery: A Nested Case-Control Study

Abstract

This nested case-control study identified broad dysregulation of the circulating proteome in neonates receiving post-operative extracorporeal membrane oxygenation (ECMO) support after congenital heart disease surgery, including differential responses in those not surviving to hospital discharge. Tissue hypoxia and mitochondrial-associated proteins may represent novel candidate biomarkers for poor ECMO outcomes.

The use of extracorporeal membrane oxygenation (ECMO) in pediatric and neonatal populations has grown substantially in recent decades, with the total number of all-indication ECMO runs reaching >18,000 in 2022.¹ Despite increased use, mortality remains high across indications. Among the highest mortality are neonates cannulated to ECMO for cardiac failure (2019 through 2023 North American survival to hospital discharge 54%), most commonly children born with critical congenital heart disease (CHD).¹ These children also experience significant morbidity across multiple organ systems including: acute kidney injury (72%)², thrombosis (29–50%)³, hemorrhage (70%, including major events such as pulmonary or intracranial hemorrhage in 13%)³, and neurologic insults (17%).^4,5

While the multiorgan clinical sequelae of ECMO have been well-described, a gap exists in understanding how individual molecular signals drive systemic derangements and how those system derangements drive clinical findings. Proteomics may be used to identify dysregulation of protein signaling pathways, providing insight into mechanisms or predictors of injury in complex physiologic states. One pilot study measuring the circulating proteome in adult ECMO populations demonstrated dysregulation of cell death and markers of coagulation with some preliminary ability to differentiate survivors from non-survivors based on their proteomic signature on ECMO day 3.⁶ To date, no study has reported the proteomic response to ECMO following neonatal surgery for critical CHD or explored the dysregulated proteins associated with post-ECMO mortality. Contributing to this scientific gap is the challenge of approaching families of critically ill neonates for observational/mechanistic research studies under the highly stressful clinical circumstances that lead to the need for post-operative ECMO.

In this hypothesis generating pilot study, we sought to 1) measure the scope of protein dysregulation in neonates on post-operative ECMO compared with neonates undergoing CHD surgery who did not receive ECMO support, 2) determine global changes in the circulating proteome over the first 4 days of ECMO support, and 3) perform preliminary assessment of individual proteins responsible for driving variation between ECMO survivors vs. non-survivors. Furthermore, to overcome the logistical and ethical challenges associated with performing observational/mechanistic research in this critically ill patient population, we sought to leverage preoperative consent and prospective collection of residual clinical plasma samples for proteomic analysis using ultra-low sample volume proteomics techniques.

Methods

This pilot study was a nested case-control study within a larger, IRB approved cohort study of the circulating metabolome in infants after cardiopulmonary bypass (R01HL156936). Informed consent was obtained for all subjects preoperatively. Neonates who received ECMO following CHD surgery with cardiopulmonary bypass (ECMO cases) were compared with patients who did not (controls). Residual plasma samples (<20μl) were collected on days one through four of ECMO and on post-operative day seven in controls. Samples were stored at −80°C until single batch analysis using a >1,500 protein panel (OLINK Explore 1536, Uppsala, Sweden). Principal Component Analysis (PCA) was used to compare proteomic profiles, followed by Welch’s t-test to detect individual protein differences (FDR < 0.05). In one subject who experienced two ECMO runs, the second run was selected for t-testing to maintain statistical independence of the samples while maximizing the similarity in post-operative days between cases and controls. PCA loadings were used to identify proteins driving variation between the PCA profile of ECMO survivors vs. non-survivors (defined as death prior to hospital discharge). Protein network analysis with functional enrichment was performed in STRING and ShinyGO.

Results

Between 2021 and 2023, we enrolled and collected residual plasma samples from seven controls and three ECMO cases (four total ECMO runs). Basic demographic data are shown in Table I. Most subjects in both groups had complex CHD (STAT 4 or 5 eg, hypoplastic left heart syndrome, truncus arteriosus). Median cardiopulmonary bypass times (minutes: 179 [range 74, 306] vs. 163 [range 92, 227]; p=0.83) did not differ between cases and controls respectively. Median 24-hour post-operative vasoactive inotropic scores were lower in ECMO cases compared with controls (7 [range 7, 11] vs. 13 [range 10, 17]; p=0.03). All ECMO subjects achieved normalization of lactate and arteriovenous difference in oxygen content by ECMO day 2. For ECMO cases, ECMO day 1 residual plasma samples were collected immediately following cannulation on post-operative days 0 (ECMO Subject 1 [E1]), 0 (E2), 3 (E3 run 1), and 10 (E3 run 2). Both subjects cannulated on post-operative day 0 were cannulated in the cardiovascular operating room, one for inability to separate from bypass with severe biventricular dysfunction and one for high vasoactive/inotropic support requirements immediately following separation. ECMO day 4 samples were collected on post-operative days 3 (E1), 3 (E2), 6 (E3 run 1), and 13 (E3 run 2). All residual samples collected resulted in sufficient sample volume for full proteomic analysis.

Table I.

Demographic data

Subject	Diagnosis	Primary Procedure	Sex	Age at Surgery (days)	Weight (kg)	Gestational Age (weeks)
ECMO 1	HLHS	Norwood	F	6	3.2	39
ECMO 2	Truncus arteriosus	Truncus arteriosus repair	M	7	3.1	39
ECMO 3	TAPVR	TAPVR repair	M	6	3.4	39
Control 1	HLHS	Norwood	M	8	4.3	39
Control 2	TAPVR	TAPVR repair	F	13	3.4	38
Control 3	Hypoplastic aortic arch/VSD	Arch repair/VSD closure	M	14	3.0	36
Control 4	HLHS	Norwood	M	4	3.4	39
Control 5	HLHS	Norwood	M	3	3.4	39
Control 6	d-TGA	Arterial switch	F	4	3.8	39
Control 7	HLHS	Norwood	F	7	4.0	39

HLHS=hypoplastic left heart syndrome; TAPVR=total anomalous pulmonary venous return; VSD=ventricular septal defect; d-TGA=d-transposition of the great arteries

Circulating Proteome: ECMO Cases vs. Non-ECMO Controls

Complete proteomic data are provided in the Supplemental Data File. ECMO patients demonstrated marked early disruption of their circulating proteome, with PCA readily discriminating between cases (ECMO day 1) and controls based on their proteomic patterns (Figure 1a and b). Welch’s t-tests comparing controls to ECMO cases on day 1 (n=3) identified 222 proteins that differed significantly between the groups (Figure 2a). These included both up-regulated (n=57) and down-regulated (n=165) proteins compared with controls. The Log2 fold change ranged from −6 to 15, where positive values are higher in ECMO cases. The maximally up-regulated protein was CXCL8 (IL-8), a potent neutrophil-attracting chemokine responsible for response to tissue injury. Other notable up-regulated proteins included peroxidases responding to cell oxidative stress, inhibitors of tissue plasminogen activators, matrix metalloproteinases, and mediators of cell adhesion. Down-regulated proteins were enriched with proteins governing neuronal growth and pruning, including amyloid precursor-like protein 1, the most down-regulated protein within our network, and NCAN, a protein in which mutations are associated with developmental delay.

Figure 1: Principal Component Analysis of the Proteome in ECMO Cases vs. Controls.

The x- and y-axes represent principal components (PC) 1 and 2, vectors that explain the greatest percentage of the global proteomic variation. The percentage of variation explained by each PC is shown in parentheses. Individual circles on the graph represent the condensed proteome of a single subject at a single time point. A) The proteome readily distinguished between controls (red; n=7) and ECMO cases on ECMO day 1 (green; n=4) by PCA. Proteomic shift is also demonstrated between ECMO from day 1 (green) and day 4 (blue). ECMO subject 3 (E3) received two separate ECMO runs, each included here. Proteome shifts in individual cases include ECMO subject 1 (E1) who decannulated from ECMO on day 3 and trended towards the control profile. In contrast, the proteome remained distinct from controls in subjects E2 and E3, both of whom died after complicated ECMO courses. B) PCA analysis of all time points from ECMO Days 1–4 demonstrates a progressive shift in the proteomic profile of subject E1 (purple) towards controls beginning on ECMO day 2 (prior to decannulation) but no substantial change in the proteomic profile of subjects E2 (yellow) or E3 (green-ECMO run 1; pink-ECMO run 2) over the first 4 days of ECMO.

Figure 2: Network Analysis and Gene Ontology.

A) A protein interaction network of the 222 proteins circulating in the blood which differed between ECMO cases and controls (FDR <0.05). Identified cell signaling pathways fell across six key categories: including cell proliferation and turnover, immune system response, mediation of the nervous system/neural development, angiogenesis, apoptosis, and cardiac signaling/development. Domains are highlighted in shades of yellow to blue with corresponding color intensity indicating higher representation within the network. Node edges are colored by the Log2 Fold Change strength (red = up-regulated and blue = down-regulated). Protein node size indicates degree of node connectivity within the network, with larger nodes representing the most highly interconnected proteins. B) The Gene Ontology (GO) terms that are most highly represented within the 222 significantly altered proteins in ECMO subjects on ECMO day 1. Node color and ranking is reflective of the highest −log10(FDR). Node size is reflective of the number of genes in the network contributing to this ontologic term. Note, three ontologic terms (malaria, pertussis, and prostate cancer) were excluded, as they represented diseases known not to be present in our population.

Functional enrichment through Gene Ontology analysis of the protein network demonstrated similar themes (Figures 2a and b). The most altered cell signaling pathways fell into six key categories. Cell proliferation and turnover represented the most enriched network, with 136 (61%) of proteins involved, followed by immune system response and mediation of the nervous system/neural development, each enriched by ~1/3 of network proteins. Finally, the network was highly enriched in drivers of angiogenesis, apoptosis, and cardiac signaling/development, representing ~1/4 proteins within our set.

Circulating Proteome: ECMO Day 1 vs. ECMO Day 4

Within the ECMO group, E1 experienced an uncomplicated course (decannulated ECMO day 3 and survived to hospital discharge) while the other ECMO subjects experienced complicated courses (E2: died ECMO day 6 following progression to myocardial standstill on ECMO day 3; E3: 2 ECMO runs, died during the post-operative hospitalization from progressive capillary leak and multiorgan failure after second ECMO decannulation). PCA provided preliminary evidence for distinct proteomic responses between the ECMO survivors vs. non-survivors (Fig. 1a and b). Overall, the three ECMO runs in non-survivors (E2, E3 run 1, and E3 run 2) demonstrated relatively little change in the circulating proteome from days 1 to 4, while the one uncomplicated ECMO run (E1) showed daily changes in the circulating proteome that progressively moved closer to the proteomic signature of controls (Fig. 1a and b). Analysis of the PCA loading plots demonstrated that subject E1 separated from both controls and cases E2 and 3 primarily along PC1 (X-axis), driven by the highest circulating levels of cytokines IL-6, IL-8, and IL-10 in any subject (Figure 1a). By ECMO day 4, their cytokine levels dropped dramatically with a corresponding shift in the PCA plot towards the controls. In contrast, subjects E2 and 3 demonstrated a greater shift from controls along PC2 (Y-axis), which did not resolve by ECMO day 4 and was driven heavily by increased concentrations of 3 proteins: erythropoietin (EPO), fibroblast growth factor 23 (FGF23), and NDUFS6 (a subunit of mitochondrial complex 1). EPO and FGF23 are upregulated by tissue hypoxia (particularly kidney hypoxia) via hypoxia inducible factor (HIF) signaling⁷ while NDUFS6 is upregulated with ischemia⁸, suggesting decreased and persistently low tissue-level oxygen delivery in non-survivors despite achieving early normalization of lactate and arteriovenous difference in oxygen content on ECMO support. Interestingly, classic markers of myocardial injury (troponin I) and myocardial stretch/dysfunction (NT-proBNP) did not contribute to the variation between survivors and non-survivors.

In conclusion, we used prospectively obtained residual clinical plasma samples to measure proteomic changes associated with ECMO after neonatal CHD surgery. We report differences in the post-operative proteomic profiles of neonates with CHD on ECMO vs. those not receiving ECMO, driven by broad differences in proteins related to cell turnover, inflammation, neural development, angiogenesis, apoptosis, and cardiac signaling. Furthermore, we identify preliminary evidence suggestive of distinct proteomic profiles by clinical course, with normalization of the circulating proteomic profile in the ECMO survivor by ECMO day 4 and biomarker evidence of persistent activation of hypoxia signaling and mitochondrial pathways in non-survivors despite normalization of circulating lactate and arteriovenous difference in oxygen content. Our pilot study is limited by the small number of subjects and the variation in time to ECMO cannulation. The use of POD7 controls was designed to limit any residual effects of surgery/CPB on the circulating proteome, optimizing the comparison to the ECMO day 4 samples. However, response to surgery/CPB may confound the comparison between controls and ECMO day 1 samples, although the findings from ECMO subject 3 and the persistence of the proteomic dysregulation on ECMO day 4 in subjects 2 and 3 suggest proteomic dysregulation in ECMO patients independent of the response to CPB. Future studies validating proteomic changes in a larger neonatal CHD population and specifically targeting HIF and mitochondrial pathways over time with stratification of patients experiencing specific clinical sequelae may identify whether these dysregulated proteins could be used for diagnostic, predictive, or therapeutic purposes. In addition, as our study design did not consistently capture pre-cannulation samples, future studies should include assessment of the pre-ECMO proteome to evaluate for proteins systems associated with the need for subsequent ECMO support.

List of Non-Standard Abbreviations:

CHD

congenital heart disease

ECMO

extracorporeal membrane oxygenation

EPO

erythropoietin

FDR

false discovery rate

FGF23

fibroblast growth factor 23

HIF

hypoxia inducible factor

IQR

intra-quartile range

NDUFS6

NADH:ubiquinone oxidoreductase subunit S6/mitochondrial complex I

PCA

principal component analysis