Serum Proteomic Profiling Implicates a Dysregulated Neurohormonal–Inflammatory Axis in Post‐Fontan Sinus Tachycardia

Abstract

Background

Postoperative sinus tachycardia is a poorly understood complication following the Fontan procedure. The molecular signaling cascades triggering acute tachycardia remain uncharacterized, limiting therapeutic innovation. Here, we present a retrospective study leveraging serum proteomics and machine learning to identify the molecular drivers of postoperative Fontan sinus tachycardia.

Methods

We integrated a clinically relevant ovine Fontan model with continuous telemetric heart rate monitoring and human patient data. Serum proteomics coupled with least absolute shrinkage and selection operator and Boruta machine learning algorithms were used to identify protein panels predictive of postoperative sinus tachycardia. Cross‐species validation was performed by comparing proteomic signatures from sheep and pediatric patients undergoing Glenn or Fontan surgery.

Results

Ovine Fontan animals demonstrated significant heart rate elevation beginning on postoperative day 1, peaking at postoperative day 3 (159.4±11.7 bpm versus preoperative, 105.3±10.5 bpm; P=0.0002), before trending toward baseline by postoperative day 10. This pattern was mirrored in human patients with a more modest magnitude. Surgical controls did not exhibit tachycardia. The principal component most correlated with heart rate (principal component 1: r=0.78, P=2.2×10⁻⁴) was enriched for inflammatory and neural pathways. The Boruta algorithm identified an 11‐protein panel with strong predictive power (area under the receiver operating characteristic curve, 0.963). Cross‐species comparison demonstrated that angiotensinogen, angiotensin‐converting enzyme, and pentraxin 3 were similarly dysregulated in both species postoperatively.

Conclusions

This study provides molecular evidence implicating a dysregulated neurohormonal–inflammatory axis in acute postoperative Fontan sinus tachycardia and establishes a foundation for developing targeted diagnostics and therapeutics for this complication.

Nonstandard Abbreviations and Acronyms

DIA

data‐independent acquisition

LASSO

least absolute shrinkage and selection operator

PC

principal component

PCGSE

principal component gene set enrichment

POD

postoperative day

RAAS

renin–angiotensin–aldosterone system

Research Perspective.

What Is New?

• This study provides the first direct molecular evidence identifying a dysregulated neurohormonal and inflammatory axis involving angiotensinogen, angiotensin‐converting enzyme, and pentraxin 3 as a primary driver of acute postoperative sinus tachycardia following the Fontan procedure.

• Proteomic profiling combined with machine learning reveals a highly predictive biomarker panel and suggests the activation of a noncanonical renin–angiotensin–aldosterone system pathway potentially mediated by chymase.

What Question Should Be Addressed Next?

• Future investigations must perform targeted tissue analysis to validate local chymase expression and evaluate whether novel therapeutics directed against these specific upstream neurohormonal or inflammatory targets can successfully mitigate sinus tachycardia in patients with single‐ventricledisease.

The Fontan procedure represents the final surgical palliation for patients with single‐ventricle diseases, a group of congenital heart defects that were once uniformly fatal in early childhood. ¹ , ² This staged surgical approach transformed patient outcomes, creating an expanding global population of survivors who now reach adolescence and adulthood. Nevertheless, the Fontan operation, while lifesaving, is palliative. ³ , ⁴ The Fontan surgery establishes a fragile and abnormal circulation that results in systemic venous hypertension and chronic ventricular preload deficiency. ⁵ , ⁶ , ⁷ This unique hemodynamic state of elevated central venous pressure and relatively low cardiac output is the primary driver of a progressive, multisystem disease burden that defines the long‐term clinical course for these patients. The consequences are ubiquitous and severe, including, but not limited to, congestive hepatopathy, protein‐losing enteropathy, plastic bronchitis, and renal dysfunction. ⁵ , ⁸ , ⁹ , ¹⁰

Acute postoperative sinus tachycardia represents a pervasive and poorly understood complication in the immediate period following the Fontan procedure. The underlying causes are often attributed to a combination of factors, such as surgical scarring, the type of Fontan operation, and atrial dysfunction as well as the chronic effects of cyanosis and associated end‐organ dysfunction. ³ , ¹¹ , ¹² , ¹³ However, the specific molecular signaling cascades that trigger a tachycardic event, particularly in the vulnerable immediate postoperative period, remain largely uncharacterized. This knowledge gap limits therapeutic innovation and intervention, leaving clinicians with reactive management strategies that often fail to prevent recurrence or address the underlying causes.

Serum proteomics, the large‐scale study of circulating proteins, offers a powerful tool to investigate the molecular basis of complex cardiovascular diseases. ¹⁴ By directly measuring the proteins that carry out cellular functions, proteomics can uncover disease mechanisms and identify novel biomarkers that other technologies might miss. ¹⁵ , ¹⁶ Analyzing the serum proteome provides a unique window into the body's systemic response, capturing circulating proteins from multiple organ systems that reflect an individual's physiological state. While previous proteomic studies in patients with Fontan circulation identified broad signatures of inflammation, oxidative stress, and metabolic dysregulation, none pinpointed specific molecular pathways directly linked to the onset of tachycardia. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹

In this study, we sought to identify the molecular drivers of acute postoperative sinus tachycardia in the Fontan circulation by integrating a clinically relevant large animal model and human patient data with advanced proteomic and bioinformatic analyses. ²² We leveraged a Fontan surgical ovine model that recapitulates the acute postoperative course and allows for the correlation of high‐throughput serum proteomics with continuous telemetric monitoring of heart rate (HR). Using this translational framework, we applied least absolute shrinkage and selection operator (LASSO) and Boruta machine learning algorithms to identify protein panels that are highly predictive of postoperative tachycardia in our ovine model. To establish clinical relevance, we confirmed the trend of increased post‐Fontan HR and performed a comparative proteomic analysis of serum from human pediatric patients undergoing either the Fontan surgery or the immediately preceding procedure, the Glenn procedure, which connects the superior vena cava to the main pulmonary artery. This cross‐species validation confirmed that 3 key proteins from the predictive panel—angiotensinogen, angiotensin‐converting enzyme (ACE), and the inflammatory marker pentraxin 3—were similarly dysregulated in human patients postoperatively. This work offers novel mechanistic insight into post‐Fontan tachycardias and identifies a panel of candidate biomarkers that may enable the development of future diagnostics and targeted therapeutics to predict and mitigate this condition.

Methods

Data Availability

The data that support the findings of this study are available from the corresponding authors upon reasonable request. The mass spectrometry data for the human serum samples used in this study have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository under the data set identifier PXD068079. The ovine serum mass spectrometry data have been deposited under the separate PRIDE data set identifier PXD077206. All custom R scripts used for data normalization, differential enrichment, principal component (PC) analysis, and machine learning feature selection are publicly available in our GitHub repository located at https://github.com/ftakaes1/Ovine‐Fontan‐Model‐Serum‐Proteomic‐Profiling.git

Ethical Statement

All animal studies were performed according to Animal Research: Reporting of In Vivo Experiments Guidelines and under the approval and guidance of the nationwide children's hospital (NCH) Animal Welfare and Resource Committee (Approval AR13‐00079 and AR24‐00179). Sheep were housed in facilities that are US Department of Agriculture licensed and American Association for Accreditation of Laboratory Animal Care accredited. Housing space for sheep was in accordance with the 2011 Guide for the Care and Use of Laboratory Animals (including heating, ventilation, and air conditioning parameters; lighting; and airflow). Human HR measurements were collected and analyzed under institutional review board approval (STUDY00000005 and STUDY00000887). Human blood samples were collected under institutional review board approval (STUDY00005500) before surgery and before discharge from patients undergoing Glenn or Fontan surgery. Informed consent was received from parents before study enrollment. This study was performed as a retrospective analysis of collected data and samples.

Telemetry Device Implantation

Fontan (n=7) and surgical control (n=4) Dorset sheep (Ovis aries) were anesthetized and received implanted continuous monitoring telemetry devices (DSI PhysioTel Digital L21‐F1 Large Animal Implantable Telemetry Unit). Sheep subject characteristics are detailed in Table S1. The procedure was performed as described previously. ²³ Briefly, the device itself was placed in a subcutaneous pocket either in the right lower abdomen adjacent to the inguinal crease or at the base of the left neck between the scapula and cervical spine, while the bipotential ECG leads were placed subcutaneously over the sternum. Pressure transducers were cannulated either in the femoral artery or carotid artery to obtain vital information. After the implantation, Fontan animals were followed via continuous telemetry monitoring for 10 days at a data collection rate of 6 collections per minute (once every 10 seconds). Surgical control animals were followed for 5 days after the operation in the same manner. Continuous data collected from the telemetry units appeared in real‐time in the linked Ponemah software, which was used to inform intensive care unit (ICU) medical decision making. Collected data were saved and exported to Microsoft Excel for further analysis. Due to software data collection settings, continuous ECG tracings were available for intensive care monitoring postoperatively but were not saved in a format suitable for retrospective analysis.

Establishing the Fontan Connection

The Fontan cohort of ovine subjects underwent an additional surgical operation in a separate anesthetic event at least a month after their telemetry device implantation. This operation established a Fontan connection by detaching the inferior vena cava from the right atrium and reconnecting it to the main pulmonary artery via an end‐to‐end anastomosis facilitated by a polytetrafluorethylene conduit, as described previously. ²⁴ Following the operation, sheep were immediately extubated and closely monitored 24/7 for ≈2 weeks in an ICU. The criteria for initiating esmolol treatment were having a post‐Fontan sinus tachycardia with an HR ≥160 bpm and clinical symptoms of poor nutritional intake or lethargy.

Postoperative Blood Sample Collection

Following the operation, sheep were immediately extubated and closely monitored 24/7 for ≈2 weeks in an ICU. For both the Fontan cohort and surgical control cohort, blood samples were collected daily from the date of surgery to the end of each animal's patient‐specific ICU monitoring end point. Sheep‐specific duration of ICU care is noted in Table S1. Collected blood samples were used for ICU monitoring and for serum proteomic analysis. The duration of this sample collection for proteomics extended to a maximum of postoperative day (POD) 6. Samples were clotted using a serum blood collection tube (BD Vacutainer, Catalog No. 02‐683‐94) for 15 minutes at room temperature. Samples were then spun at 1500g for 15 minutes at 4 °C followed by immediate serum collection from the supernatant. Serum samples were stored at −80 °C for long‐term use.

Human blood samples were collected under institutional review board approval before discharge from patients undergoing either Glenn (n=5) or Fontan (n=5) surgery. Samples were clotted using a serum blood collection tube (BD Vacutainer, Catalog No. 02‐683‐94) for 15 minutes at room temperature. Samples were then spun at 1500g for 15 minutes at 4 °C followed by immediate serum collection from the supernatant. Serum samples were stored at −80 °C for long‐term use. Human patient characteristics for the proteomic analysis are detailed in Tables S2 and S3.

Acute Post‐Fontan Data Set Analysis and Statistics

Data analysis of telemetry measurements included preoperative measurements through POD 10 for the Fontan animals, and the morning after the telemetry implantation operation (POD 1) through POD 5 for the control cohort. Data were separated by day or night according to the 12‐hour light/dark cycle of the vivarium, with the hours from 6:00 am to 5:59 pm being day to 6:00 pm to 5:59 am being night. All recorded measurements within this time frame were averaged to generate 1 value for each POD for each assessed parameter. Heart rate values recorded during periods of esmolol infusion were included in the overall time trend analyses to accurately reflect the continuous retrospective data. As there were no statistically significant differences between day and night measurements on any POD (Figure S1A through S1C) and all sample collection was performed during the day, data assessed for the purpose of these study comparisons were the measurements and blood samples collected during the day.

To assess the translatability of our ovine large animal model, we also analyzed HR data of human patients with Fontan circulation for evidence of the same trends (n=26). HR data of human patients with Fontan circulation were pulled from patient charts and separated into each POD, defined as each calendar day after the operation, with the date of surgery being POD 0, out to POD 7, as it was the most common date of discharge from the hospital. These data were obtained from a retrospective chart review of the electronic medical record, rather than continuous telemetry recording with measurements taken at consistent intervals. Therefore, to avoid overrepresenting certain periods within an hour, only the first measurement made for each hour was included in the POD average calculation.

Statistical analysis was performed using Prism 10 (GraphPad Software, La Jolla, CA). For unpaired between‐group comparisons, we used linear regression with treatment group as the predictor, providing a unified framework for estimating effect sizes and CIs across all comparisons. For longitudinal physiological measurements (HR, systolic blood pressure, diastolic blood pressure) involving repeated observations within individual animals, we used a mixed‐effects model fitted via restricted maximum likelihood in GraphPad Prism, with individual animals treated as a random effect to account for the paired nature of the data. Dunnett's post hoc test was used to compare postoperative time points against the preoperative baseline. Effect size was determined through the β coefficient and 95% CI of each linear regression comparison, as well as through the calculation of Cohen's d and Hedges' g (accounting for small sample size bias) values. All HR statistics can be found in Tables S4 through S6.

Protein Digestion and Peptide Cleanup

For sheep serum, samples were diluted 1:10 with 8 mol/L urea and subjected to probe sonication. Protein concentration was determined using a bicinchoninic acid assay, and 100 μg of protein was reduced with 5 mmol/L dithiothreitol for 30 minutes at room temperature, followed by alkylation with 10 mmol/L iodoacetamide in the dark for 30 minutes at room temperature. Samples were digested overnight with 4 μg of Lysyl endopeptidase (Wako), diluted to a 1 mol/L urea concentration, and further digested overnight with 4 μg of trypsin (Thermo Fisher Scientific) at room temperature. The resulting peptide solutions were acidified to a final concentration of 1% formic acid and desalted using 10 mg HLB solid‐phase extraction columns. The HLB columns were conditioned with 500 μL of methanol and equilibrated twice with 500 μL of 1% formic acid. Samples were loaded onto the columns and washed twice with 500 μL of 1% formic acid. Peptides were eluted using 1 mL of 50% acetonitrile. A 10% aliquot was removed from each sample, and all eluates were subsequently dried to completeness using a SpeedVac vacuum concentrator (Speedvac, Labconco).

For human serum, samples (5 μL) were diluted 1:10 with 8 mol/L urea. Protein reduction was performed using 5 mmol/L dithiothreitol for 30 minutes at room temperature, followed by alkylation with 10 mmol/L iodoacetamide in the dark for 30 minutes at room temperature. Samples were then digested with 4 μg of Lysyl endopeptidase (Wako) overnight. The reaction mixture was then diluted to a final urea concentration of 1 mol/L and subjected to further digestion with 4 μg of trypsin (Thermo Fisher Scientific) overnight at room temperature. Following digestion, the peptide solutions were acidified to a final concentration of 1% formic acid and desalted using 10 mg HLB columns. The columns were conditioned with 500 μL of methanol and equilibrated twice with 500 μL of 1% formic acid. Samples were loaded onto the columns and washed twice with 500 μL of 1% formic acid. Peptides were eluted with 1 mL of 50% acetonitrile and dried to completeness using a centrifugal vacuum concentrator (SpeedVac, Labconco).

Liquid Chromatography and Mass Spectrometry

For sheep serum, each sample was resuspended in 200 μL of loading buffer containing 0.1% formic acid, and a 20 μL aliquot was loaded onto Evotips for analysis by liquid chromatography–tandem mass spectrometry. Peptide eluents were separated using an Evosep One system (Evosep) coupled to an IonOpticks column (15 cm x 75 μm internal diameter, packed with 1.7 μm resin). Buffer A consisted of water with 0.1% (vol/vol) formic acid, and buffer B consisted of acetonitrile with 0.1% (vol/vol) formic acid. Peptide elution was performed using the preset 40SPD Whisper Zoom method. Peptides were analyzed on an Orbitrap Astral mass spectrometer (Thermo Fisher Scientific) equipped with a FAIMS Pro high‐field asymmetric waveform ion mobility spectrometry interface (Thermo Fisher Scientific). One compensation voltage of −35 V was applied for FAIMS separation. Each cycle consisted of a full MS1 scan across an m/z range of 380 to 980 at a resolution of 240 000, using 500% AGC, and a 3‐millisecond injection time. Data‐independent acquisition (DIA) scans using higher‐energy collisional dissociation were collected with 200 isolation windows of 3 m/z across the precursor range of 380 to 980 m/z using a narrow‐window DIA acquisition strategy, with a cycle time of 0.6 seconds and a 3‐millisecond injection time. The collision energy was set to 27%, and the tandem mass spectrometry scan range was configured from 150 to 2000 m/z.

For human serum, each sample was resuspended in 250 μL of loading buffer containing 0.1% formic acid, and a 1 μL aliquot was analyzed by liquid chromatography coupled to tandem mass spectrometry. Peptide eluents were separated on an Evosep column (8 cm × 150 μm internal diameter packed with 1.5 μm resin) using a Vanquish Neo system (Thermo Fisher Scientific). Buffer A consisted of water containing 0.1% (v/v) formic acid, and buffer B consisted of 80% acetonitrile in water with 0.1% (v/v) formic acid. Elution was carried out over a 16‐minute gradient increasing from 1% to 50% solvent B. Peptides were analyzed using an Orbitrap Astral mass spectrometer equipped with a high‐field asymmetric waveform ion mobility spectrometry (FAIMS Pro) ion mobility source (Thermo Fisher Scientific). A single compensation voltage of −35 V was applied. Each acquisition cycle began with a full MS1 scan covering a mass range of 380–980 m/z at a resolution of 240 000, 500% AGC and 3‐millisecond injection time. DIA was performed using higher‐energy collision‐induced dissociation with 300 isolation windows of 2 m/z covering the full precursor range (380 to 980 m/z) using a narrow‐window DIA acquisition strategy. The total cycle time was set to 0.6 seconds, with a maximum injection time of 2.5 milliseconds. Collision energy was set to 27%, and the scan range was set to 150 to 2000 m/z.

Database Search

Spectronaut (version 19.9.250512.62635) was used to search all raw files using a library‐free directDIA approach against a human UniProt database (downloaded February 2019) and a sheep UniProt database (downloaded July 2024). The false discovery rate was restricted to 0.01 at all levels, including peptide‐spectrum match, peptide, and protein. Protein quantification was derived using the MaxLFQ algorithm, a widely adopted standard quantification method for label‐free proteomics that estimates protein‐level intensities by solving a least‐squares optimization of pairwise peptide intensity ratios across samples. ²⁵ All remaining parameters were maintained at standard Spectronaut defaults.

Proteomic Data Set Preprocessing

Raw protein intensity values from the proteomic analysis were first subjected to quality control. To determine the optimal normalization strategy for our quantitative label‐free proteomics data, we evaluated several standard methods using the NormalyzerDE package. ²⁶ Variance stabilizing normalization was selected because it outperformed the alternatives in minimizing technical variance while preserving biological signal. Specifically, variance stabilizing normalization achieved the highest correlation among replicates (R ²=0.9) compared with other methods. Furthermore, relative log expression plots confirmed that variance stabilizing normalization centered the sample medians at 0 and minimized the variance spread across all biological replicates (Figure S2). Our selection was further supported by systematic evaluations in the literature showing that variance stabilizing normalization consistently provides superior variance reduction and downstream differential expression performance in label‐free proteomics. ²⁶ , ²⁷ Following normalization, our data set also revealed that missing values were predominantly skewed toward proteins with low intensities, suggesting a missing not at random data structure (Figure S3A). To address this, missing values were imputed using a quantile regression imputation of left‐censored data method via the impute.QRILC function from the imputeLCMD R package (R Foundation for Statistical Computing, Vienna, Austria). ²⁸ For subsequent analyses, the human data set was stratified into separate Fontan and Glenn surgical cohorts.

Proteomic Data Analysis

Differential enrichment analysis was performed to identify proteins that were significantly altered between preoperative and postoperative states. For the ovine data set, the NormalyzerDE R package was used to compare the postoperative and preoperative groups. ²⁶ For the human data set, differential enrichment results were obtained from a direct comparison between preoperative and postoperative samples, performed separately for the Fontan and Glenn surgical cohorts. For both species, proteins were considered significantly enriched if they met a threshold of an absolute log₂ fold change ≥0.48 and an adjusted P value ≤0.05. Adjusted P values were calculated using the Benjamini–Hochberg method.

Unsupervised analyses were conducted to visualize the overall proteomic profiles. For the human data, these analyses were performed on the combined patient cohort as well as on the stratified Fontan and Glenn groups. PC analysis was performed using the prcomp function in R. Before decomposition, the protein intensity matrix was column‐centered (mean‐subtracted) and scaled to unit variance using the center and scale arguments in the prcomp function. Hierarchical clustering was performed on the top 100 most variable proteins, selected by ranking all quantified proteins by their sample‐wise variance and retaining those with the highest variance. This threshold was chosen to focus visualization on proteins with sufficient biological variation across samples while excluding low‐variance features that contribute primarily to noise. Clustering was performed using the pheatmap R package. ²⁹ This analysis used the “ward.D2” clustering method with Manhattan distance and row‐wise scaling.

Pathway enrichment analysis was performed on the lists of significantly upregulated proteins from the pre‐ and postoperative groups using Metascape. ³⁰ For visualization, the top 25 pathways ranked by enrichment score were selected.

Spearman Correlation and PC Gene Set Enrichment

Principal components of the quantile regression imputation of left‐censored data‐normalized ovine proteome were correlated with measured HRs using Spearman's rank correlation through the cor.test function in R. Biological pathways represented by the significantly correlated principal components were subsequently identified using principal component gene set enrichment (PCGSE) analysis through the pcgse package in R. ³¹ The analysis was run against the Reactome gene set database (c2.cp.reactome.v2024.1.Hs.symbols.gmt). P values were corrected using the Benjamini–Hochberg method, and pathways with an adjusted P value ≤0.05 were considered significantly enriched.

LASSO‐ and Boruta‐Based Feature Selection

To identify the most predictive proteins for tachycardia, 2 feature selection methods were used: least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm. LASSO regression was selected for its ability to produce sparse models through L1 penalization, which is well suited to high‐dimensional proteomic data sets where the number of features exceed the number of samples. ³² , ³³ The Boruta algorithm was chosen because it evaluates all features against randomized shadow variables and retains those consistently more important than chance, without penalizing correlated features. ³⁴ , ³⁵ The ovine proteomic data set was first partitioned into training (50%, 12 replicates) and testing (50%, 12 replicates) sets for model development and evaluation. The LASSO model was implemented using the glmnet R package. Due to the highly constrained sample size, the optimal regularization parameter (λ) was determined using a leave‐one‐out cross‐validation approach rather than K‐fold cross‐validation to maximize the training data available in each iteration and minimize bias. ³⁶ , ³⁷ Leave‐one‐out cross‐validation was performed using the caret R package. ³⁸ Proteins with non‐0 coefficients in the final model were selected as potentially predictive features.

As an orthogonal approach, we used the Boruta algorithm, a random forest–based wrapper. To account for the inherent randomness of the algorithm and ensure a robust feature set, a stability selection protocol was implemented. ³⁹ The Boruta algorithm was executed 100 independent times on the training data. Only proteins that were confirmed as ‘important’ in at least 90% of these iterations were retained for the final feature set, ensuring that the selected signature was stable across multiple resamples. For both the LASSO‐ and Boruta‐derived protein lists, the optimal predictive panel was determined by iteratively testing all unique combinations of the selected features. A logistic regression model, using the glm function in R, was built for each combination on the training data, and its predictive performance was evaluated on the test data by calculating the area under the receiver operating characteristic curve (AUC) using the pROC R package. ⁴⁰

Enzyme‐Linked Immunosorbent Assay

Ovine serum concentrations of ACE (Catalog No. MBS737781), angiotensinogen (Catalog No. MBS286846), and pentraxin 3 (Catalog No. MBS1603320) were quantified using commercial sandwich ELISA kits from MyBioSource (San Diego, CA). Frozen serum samples were thawed on ice and centrifuged at 1500g for 15 minutes to pellet debris. Samples were then diluted 1:2 in PBS to ensure optical density readings fell within the parameters of the standard curve.

All assays were conducted following the manufacturer's protocols for each protein assay. In summary, standards and diluted samples were pipetted into the provided microplate wells precoated with target‐specific antibodies. After the required incubation and wash steps, horseradish peroxidase conjugates and substrate solutions were added to induce a colorimetric reaction. The reaction was terminated with the addition of a stop solution. Absorbance was immediately measured at 450 nm using a BioTek Synergy 2 microplate reader (BioTek Instruments, Inc., Winooski, VT). Sample concentrations were determined via standard curve interpolation and adjusted for the initial dilution factor.

Results

Ovine Post‐Fontan Sinus Tachycardia Peaks at POD 3 and Begins Returning to Baseline

Several parameters were assessed from the implanted telemetry devices, including HR, systolic blood pressure, and diastolic blood pressure. Only HR significantly differed between preoperation measurements and postoperation, while the other parameters did not change significantly (Figure S1D and S1E). HR was found to increase starting immediately after the Fontan operation, with statistically significant differences beginning on POD 1 (P=0.023) and continuing to rise on POD 2 (P=0.0093). The postoperative sinus tachycardia peaked on POD 3 compared with preoperative HRs (preoperative, 105.32±10.51 bpm versus postoperative, 159.40±11.74 bpm; P<0.0002). Though still maintaining statistical significance, at POD 5, the trend of increasing tachycardia reversed and the difference between preoperative HR and postoperative HR began to decrease and trended toward a return to baseline (P=0.0150). This pattern continued as POD 10 measurements continued decreasing in statistical difference compared with preoperative HR measurements (P=0.2032; Figure 1A, Table S4).

Figure 1. Postoperative HR analysis of ovine and human subjects.

A, Left, Average daily heart rate for each POD following the Fontan operation in ovine subjects (n=7). Statistically significant increases in average HRs were observed beginning from POD 1 (*P=0.0234), peaked at POD 3 (***P<0.0002), and began to resolve by POD 5 (*P=0.0150). A, Right, Average measured heart rate for each POD following the Fontan operation in human patients (n=26). Patients experienced a statistically significant increase in HR beginning at POD 2 following the operation (***P=0.0007), which peaked at POD 3 (****P<0.0001), and began to resolve by POD 5 (**P=0.0074). B, Average daily HR following either the Fontan operation (n=7, red line) or a surgical control operation (n=4, blue line) in ovine subjects. Animals receiving the Fontan operation had statistically significant HR increases compared with surgical control animals beginning at POD 2 (*P=0.029), which peaked at POD 3 (***P=0.0005). HR indicates heart rate; and POD, postoperative day.

Due to the retrospective nature of this data analysis and the intensive care treatment after the operation, animals reaching an HR of ≥160 bpm were administered esmolol. Notably, the ovine subjects were generally symptomatic from the tachycardia. They demonstrated decreased food and water intake as well as reduced physical activity. These clinical signs improved following HR control with esmolol. Of the assessed Fontan animals, 6 of 7 animals required esmolol intervention during their intensive care, with the average number of days on the medication being 3.86±2.48 days. Four of 6 of these animals required esmolol treatment beginning on POD 3. One animal required esmolol beginning on POD 4 and another beginning on POD 6. The longest duration any animal required esmolol intervention during their intensive care treatment was 7 days, and the shortest treatment course was 3 days (Table S1).

Pediatric Patients With Fontan Circulation Demonstrate a Trend of Increased Postoperative HR

The dramatic HR increases observed in the ovine Fontan model were compared with trends in human patient data. While no statistically significant differences were observed on POD 1, the pediatric patients experienced a modest HR increase beginning POD 2 (P=0.0007), which peaked on POD 3 (P<0.0001), before trending toward resolution on POD 5 (P=0.0074) and POD 7 (P=0.3492) (Figure 1A, Table S5). The magnitude of observed change was lower compared with the ovine model, with the difference at POD 3 being an average change of 16.7 bpm in humans, compared with 54.08 bpm in sheep. Additionally, no human patients required HR modulation with esmolol.

Ovine Surgical Controls Did Not Experience Postoperative Sinus Tachycardia

To determine if this postoperative sinus tachycardia was specific to the Fontan operation or generally attributable to surgical insult, the HR patterns were compared with a separate cohort of sheep receiving a surgical operation for the implantation of a telemetry monitoring device. This operation consisted of an anesthetic event, incision, device implantation, and wound closure. We compared the HRs of the Fontan cohort and surgical control cohort on POD 1, 2, 3, and 5, as these corresponded to the rise, peak, and resolution of tachycardia in our Fontan operation animals, respectively.

At POD 1, the HR of the Fontan animals was not significantly different than the surgical control cohort (P=0.0579). On POD 3, in contrast with the tachycardia seen in the Fontan cohort, the HR of the surgical control cohort remained relatively constant with preoperative measurements, representing a statistically significant difference between cohorts (P=0.0005). By POD 5, there was no longer a statistically significant difference in their observed HRs as the Fontan cohort HRs begin to return to baseline (P=0.0942) (Figure 1B, Table S6).

Our analysis showed that our observed postoperative sinus tachycardia was not due to surgical insult alone, as the surgical control cohort animals maintained relatively constant HRs and significantly lower HRs than the Fontan animals throughout their recovery. The surgical control cohort's HR measurements did not show the same trend of increasing and resolving tachycardia observed in the Fontan cohort.

As none of the assessed surgical control animals experienced postoperative sinus tachycardia, none required esmolol treatment. This suggests that the postoperative sinus tachycardia observed following the Fontan connection is not driven by anesthetic complications or postoperative pain but rather has a separate pathogenesis.

Sheep Serum Proteome Is Altered by the Fontan Operation

To investigate potential molecular mechanisms driving the postoperative sinus tachycardia observed in our sheep, we performed a comparative proteomic analysis of serum collected before and after the Fontan operation. Hierarchical clustering of the top 100 most variable proteins in our data set revealed a clear separation between preoperative and postoperative serum profiles (Figure 2A). PC analysis reinforced that finding with distinct clustering between both sample groups (Figure 2B). A scree plot showed PCs 1 to 3 representing >50% of the data set's variance (Figure S3B). Of the 24 PCs, PC1 was the primary driver for clustering as comparisons between other PCs in the absence of PC1 showed significant overlap (Figure S3C). Differential enrichment analysis between preoperative and postoperative groups revealed 104 proteins upregulated in the preoperative group and 125 proteins being upregulated in the postoperative group (Figure 2C).

Figure 2. Unsupervised and pathway analysis of the sheep serum proteome before and after Fontan procedure.

A, Hierarchical clustering heatmap of the top 100 most variable proteins in our data set. Clustering for this analysis was done using the “ward.D2” method with Manhattan distance calculations. B, PC analysis graph of PC1 vs PC2 of the serum proteomic data set. Shaded areas for the groups were manually added to highlight each group's respective area. Preoperative (teal, n=6 with no technical replicates) samples showed clear separation from the postoperative (red, n=6 at different time points) group. C, Volcano plot of the top differentially enriched proteins in a preoperative vs postoperative group comparison. A total of 103 and 125 proteins were found to be upregulated in the preoperative and postoperative groups, respectively. D, Reactome gene sets pathway analysis of upregulated proteins found in the preoperative samples. E, Reactome gene sets pathway analysis of upregulated proteins found in postoperative samples. HDL indicates high‐density lipoprotein; HLH, helix‐loop‐helix; L1CAM, L1 cell adhesion molecule; NOTCH, neurogenic locus notch homolog protein 2; and PC, principal component.

We next sought to understand the functional implications of these changes by performing pathway analysis to identify biological processes enriched in each group. Preoperative samples showed an upregulation for pathways predominantly involved in high‐density lipoprotein remodeling, the complement system, and hemostasis (Figure 2D). The postoperative group was significantly enriched with proteins related to robust inflammatory and protein turnover processes, including neutrophil degranulation and ATP‐independent proteasomal degradation, but more interestingly, our results also identified an enrichment of the metabolism of angiotensinogen to angiotensin pathway in the postoperative state (Figure 2E).

Inflammatory and Neural Pathways Correlate With Postoperative Fontan Sinus Tachycardia

We next investigated which pathways were most directly associated with the postoperative sinus tachycardia observed in the animals. To achieve this, we correlated the PCs of the serum proteome with HRs measured at both pre‐ and postoperative time points. Of all the PCs, only PC1 (Spearman correlation=0.78, adjusted P value=2.2 × 10⁻⁴) showed a statistically significant correlation with the HR measurements (Figure 3A). All other PCs had low correlation scores and failed to meet the statistical significance threshold (Figure S4). A complete list of PC1 loadings can be found on Table S7. To define the biological functions represented by this PC, we performed PCGSE analysis. This analysis revealed PC1 to be significantly upregulated with pathways indicative of a strong inflammatory response, such as innate and adaptive immunity and cytokine signaling, alongside pathways involved in cellular responses to stimuli and nervous system development (Figure 3B). The full test statistics for the PCGSE can be found in Table S8. Collectively, these findings suggest that the postoperative sinus tachycardia observed in our sheep Fontan model is driven by a systemic inflammatory state.

Figure 3. PC gene set enrichment analysis of the sheep serum proteome.

A, Spearman correlation between the first PC and sheep HR measurements. The red shaded area around the best fit line represents the 95% CI. Reported P values were corrected using the Benjamini–Hochberg method and adjusted P values <0.05 were considered significant. B, PC gene set enrichment analysis of the PC1 protein loadings. Pathways were ranked based on their overall gene count in the principal component loadings. HR, heart rate; MAPK, mitogen‐activated protein kinase; MHC, major histocompatibility complex; and PC, principal component.

Identification of Key Protein Drivers of Postoperative Sinus Tachycardia

We then sought to pinpoint the specific proteins found in our pathway analysis that may be mediating the tachycardia. To narrow down the most influential proteins from our data set, we used 2 distinct feature selection methods: LASSO and Boruta (Figure 4A). For the LASSO model, we found the optimal regularization parameter to be 0.58 through a leave‐one‐out cross‐validation method (Figure S5A and S5B). Our final regression with this parameter yielded a final set of 22 proteins as the most predictive features associated with HR. To identify the optimal combination of proteins that could be acting as tachycardia drivers, we iteratively tested all 22 unique protein combinations. Our evaluation identified top‐performing panels of varying sizes, notably a unique 20‐protein combination (AUC, 0.481), a 16‐protein combination (AUC, 0.500), and an 11‐protein combination (AUC, 0.704) that showed the highest predictive potential (Figure 4B). As an orthogonal approach, we used the Boruta algorithm to determine if an alternative feature selection method could identify a protein signature with greater predictive power. Surprisingly, the Boruta model identified 12 proteins with strong discriminatory power. By analyzing each protein's intensity, we found that all 12 proteins showed a statistically significant difference in quantity between the preoperative and postoperative groups (Figure 4C). Eleven of the 12 showed high mean variable of importance (Figure 4D). After using a similar iterative combination approach, the 11‐protein combination of ADAMTS13, angiotensinogen, apolipoprotein A1, apolipoprotein M, LOC101117146, nidogen‐1, proteasome α4, pentraxin 3, surfactant protein B, W5PRZ7, and YWHAE showed the highest predictive power (AUC, 0.963; Figure 4E). Finally, the predictive accuracy peaked with this 11‐protein panel, as no other combination size achieved a comparable AUC value. These results demonstrate that the Boruta feature selection method identified a superior predictive signature, pinpointing an 11‐protein panel (ADAMTS13, angiotensinogen, apolipoprotein A1, apolipoprotein M, LOC101117146, nidogen‐1, proteasome α4, pentraxin 3, surfactant protein B, W5PRZ7, YWHAE) as potential key drivers of post‐Fontan sinus tachycardia in our animal model.

Figure 4. Identification of a candidate protein signature driving post‐Fontan sinus tachycardia.

A, Graphical representation of our feature selection workflow. To identify the most robust protein signature, 2 orthogonal methods were used. First, the LASSO method with leave‐one‐out cross‐validation was used. Second, the Boruta algorithm was applied using a stability selection approach, where the algorithm was run 100 times to identify features consistently ranked as important. B, ROC curves for LASSO protein candidates. The ROC curve plots the true positive rate (sensitivity) against the false positive rate (1‐specificity). The diagonal line represents a random classifier with no predictive power (AUC, 0.5). To identify the optimal predictive signature, all unique combinations of the 22 candidate proteins were tested. The curves shown highlight the top‐performing combinations, with the shaded area representing the AUC. C, Distribution of mean importance scores for the 12 proteins identified by Boruta. The plot illustrates the stability of feature importance across 100 iterative runs of the algorithm. Each point represents the importance value calculated for a given protein from a single run. D, Violin plot illustrating the intensity distribution of the 12 Boruta candidate proteins. The plot compares the distribution of protein intensity values between the preoperative (teal) and postoperative (red) groups for each of the 12 candidates. E, ROC curves for the Boruta candidate proteins. The curves plot the true‐positive rate vs the false‐positive rate, where the diagonal line represents random chance. The shaded area indicates the area under the curve (AUC). AUC indicates area under the receiver operating characteristic curve; LASSO, least absolute shrinkage and selection operator; and ROC, receiver operating characteristic.

Translational Analysis of the Human Fontan Proteome Validates Angiotensinogen, ACE, and Pentraxin 3 as Cross‐Species Markers of Post‐Fontan Sinus Tachycardia

To determine the clinical relevance of our findings in the ovine model, we performed a comparative proteomic analysis on serum collected from human patients undergoing either a Glenn or Fontan operation, using a similar pre‐ and postoperative study design. Hierarchical clustering of the top 100 most variable proteins revealed unclear separation between both groups (Figure 5A). A clear separation between pre‐ and postoperative samples emerged only after stratifying the cohort by surgical stage (Glenn versus Fontan). This distinct clustering was exclusively observed in the Fontan patient group, while the Glenn patient samples showed no clear separation (Figure 5B and 5C). Further comparisons with different PCs showed similar results with PC1 driving much of the clustering between the pre‐ and postoperative groups in the Fontan group (Figure S6). Differential enrichment analysis also revealed no upregulated proteins in the post‐Glenn cohort (Figure 5D). On the other hand, 18 upregulated proteins were found in the post‐Fontan cohort (Figure 5E). To ensure the lack of differentially expressed proteins in the postoperative Glenn cohort was not due to elevated noise from a small sample size, we evaluated the intragroup variance. We calculated the coefficient of variation for all quantified proteins using linear intensity values. The median coefficient of variation for the postoperative Glenn cohort (32.3%) was equivalent to the postoperative Fontan cohort (34.0%). A Wilcoxon rank‐sum test confirmed no significant elevation in variance within the Glenn cohort (P=0.30, Figure S6). We compared the 18 upregulated proteins from the post‐Fontan human cohort with the upregulated proteins from our postoperative sheep. This cross‐species analysis revealed 6 conserved proteins: leucine‐rich α₂‐glycoprotein, angiotensinogen, cathepsin D, glucose‐6‐phosphate isomerase, peptidoglycan recognition protein 1, and pentraxin 3 (Figure 5F). Notably, angiotensinogen and pentraxin 3 were also key features identified by our machine learning model as highly predictive of sinus tachycardia. Although ACE narrowly missed the statistical significance threshold for upregulation in the human cohort (Log₂fold change=−0.78, adjusted P‐value=0.18), its critical association with angiotensinogen prompted us to investigate it further. Dot plots of those 3 proteins revealed that angiotensinogen, pentraxin 3, and ACE all demonstrated a similar pattern of decreased postoperative expression in both human and sheep Fontan groups (Figure S7). Therefore, we assessed the magnitude of the physiological response via targeted effect size calculations. An evaluation of our cross‐species markers revealed robust effect sizes (Cohen's d) for angiotensinogen (3.68), ACE (−1.61), and pentraxin 3 (1.35) in the Fontan cohort (Table S9). In contrast, these effect sizes were substantially lower or negligible in the Glenn cohort (angiotensinogen, 0.40, ACE, −0.02, pentraxin 3, 0.92; Table S9). To validate these untargeted proteomic findings, we performed an ELISA quantifying pentraxin 3, angiotensinogen, and ACE in banked ovine serum preoperatively and postoperatively. Due to limited remaining sample volumes, this orthogonal validation could not be performed on the human cohort. Notably, however, the ovine ELISA results demonstrated protein expression trends consistent with our quantitative proteomic data with elevated pentraxin 3 and angiotensinogen presence in postoperative samples and increased ACE enrichment in preoperative samples (Figure 5G). Together, these findings suggest that the neurohormonal and inflammatory proteins angiotensinogen, pentraxin 3, and ACE are strong candidates for key drivers of post‐Fontan sinus tachycardia.

Figure 5. Proteomic profiling of human serum samples before and after Fontan operation.

A, Hierarchical clustering of the top 100 most variable proteins in pediatric Glenn and Fontan human serum samples. The heatmap includes samples from both patients undergoing the Glenn procedure (yellow) and patients undergoing the Fontan procedure (dark purple), annotated by preoperative (teal) and postoperative (red) time points. B, PC analysis of the Glenn patient cohort. The plot of PC1 vs PC2 shows no clear separation between preoperative (teal) and postoperative (red) samples. The shaded area was added manually to highlight the overlapping sample distribution. C, PC analysis of the Fontan patient cohort. The plot of PC1 vs PC2 shows a distinct separation between preoperative (teal) and postoperative (red) samples. The shaded areas were added manually to emphasize the clear sample clustering. D, Volcano plot of differential protein expression in the Glenn cohort (n=5). The plot compares postoperative vs preoperative samples. Proteins with an absolute log₂ fold change ≥0.48 and an adjusted P value ≤0.05 (indicated by dashed lines) were considered statistically significant. E, Volcano plot of differential protein expression in the Fontan cohort (n=5). The plot compares postoperative vs preoperative samples. Proteins with an absolute log₂ fold change ≥0.48 and an adjusted P value ≤0.05 (indicated by dashed lines) were considered statistically significant. F, Venn diagram of upregulated post‐Fontan proteins. The diagram illustrates the comparison between upregulated proteins in the ovine (blue) and human (purple) cohorts, revealing a 6‐protein overlap (pink) between the 2 groups. G, ELISA validation of key dysregulated proteins in the Fontan ovine model. Protein concentrations of ACE, angiotensinogen, and pentraxin‐3 were quantified in preoperative and postoperative sheep serum samples. Bar graphs represent the mean concentration with individual samples plotted. Error bars indicate SEM. Statistical significance between preoperative and postoperative groups was determined using a 2‐way Student's t test (*P<0.05). ACE indicates angiotensin‐converting enzyme; AGT, angiotensinogen; PC, principal component; and PTX3, pentraxin‐3.

Discussion

To our knowledge, the present study provides the first direct molecular evidence implicating a dysregulated neurohormonal–inflammatory axis, specifically centered on the renin–angiotensin–aldosterone system (RAAS), as a principal driver of acute postoperative sinus tachycardia in the Fontan circulation. We first began our investigation through a robust characterization of an ovine Fontan model, which demonstrated high fidelity in replicating the acute postoperative clinical course. The temporal pattern of tachycardia observed in the sheep, with a significant increase beginning on POD 1 and peaking on day 3 before trending toward resolution, closely mirrors the timeline observed in pediatric patients recovering from Fontan surgery. This physiological concordance provides confidence that the molecular events captured in the sheep during this period are relevant to humans. The inclusion of a surgical control group further strengthened our findings that the observed sinus tachycardia is not a general response to anesthesia or postoperative pain, but rather a specific pathophysiological response to the unique and abrupt hemodynamic alterations imposed by the Fontan circulation.

The systemic changes induced by the Fontan connection were mirrored by a shift in the serum proteome. Unsupervised analyses, including both principal component analysis and hierarchical clustering, revealed an unambiguous separation between preoperative and postoperative serum profiles in the ovine model. Interestingly, this observation mirrors our previous findings on the post‐Fontan serum extracellular vesicle transcriptome. ⁴¹ This suggests that the transition to a Fontan circulation, specifically, imposes a systemic insult significant enough to alter the pre‐ and postoperative states. An extracellular vesicle deconvolution analysis from our prior work implicated the liver, brain, and other tissue types as primary contributors to the systemic response to the Fontan procedure. ⁴¹ Therefore, further experimental work is necessary to deconvolute the complex serum proteome and definitively identify the key tissues driving the proteomic alterations we observe following the Fontan connection.

Having established that the Fontan procedure induces a significant systemic shift in the serum proteome, we sought to elucidate the functional consequences of these alterations. Pathway enrichment analysis of the postoperative proteins revealed a robust inflammatory signature, including the significant enrichment of pathways such as neutrophil degranulation and ATP‐independent proteasomal degradation. Intriguingly, this analysis also identified the metabolism of angiotensinogen to angiotensin as a key upregulated pathway, providing the first objective evidence implicating the RAAS system in our animal model. To directly connect these proteomic shifts to the observed tachycardia, we used a PCGSE and Spearman's rank correlation analysis. Spearman's rank correlation was chosen for this analysis because it does not assume linearity and is robust to outliers, properties that are particularly important when evaluating monotonic associations in a small cohort where individual extreme values could disproportionately influence parametric correlation estimates. When combined with PCGSE, we sought to identify the biological processes that are overrepresented within the principal components, thereby linking the principal component loadings directly to their biological function. These tools provided a powerful second line of evidence, demonstrating that the principal component most significantly correlated with HR (PC1) was enriched for pathways related to innate and adaptive immunity, interleukin signaling, and nervous system development. The subsequent identification of angiotensinogen, ACE, and pentraxin 3 as key predictive proteins by machine learning serves as a specific, protein‐level validation of this broader pathway‐level discovery. The inclusion of pentraxin 3 in this predictive signature is particularly telling. Pentraxin 3 can be produced rapidly by cardiac and endothelial cells in direct response to surgical trauma and inflammation, and its elevation is a hallmark of myocardial stress and local inflammation, which is known to create a proarrhythmic environment by promoting electrical instability. ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ Its strong predictive power in our models, therefore, likely represents the initial inflammatory trigger for cardiac irritability. ⁴⁶

The specific proteomic signature of the RAAS components in our analysis presents another interesting point of discussion. A previous report highlighted the activation of the RAAS pathway following a Fontan procedure. ⁴⁷ However, our results show a unique regulatory pattern that warrants further scrutiny. While angiotensinogen, the precursor substrate, was significantly upregulated, we observed a downregulation of ACE in the postoperative group. This pattern deviates from a classic RAAS cascade and points toward an alternative, ACE‐independent pathway for angiotensin II generation, likely mediated by the serine protease chymase. In states of profound cardiac stress, previous reports show that the heart can shift its primary mechanism of angiotensin II production to a local, tissue‐based chymase‐dependent pathway. ⁴⁸ , ⁴⁹ Furthermore, cardiac chymase is not only more efficient at generating angiotensin II than ACE, but its activity is not inhibited by ACE inhibitors, potentially explaining the persistence of increased heart rate in some patients despite standard therapy. ⁵⁰ , ⁵¹ However, it is important to note that direct measurements of angiotensin II, renin, aldosterone, and chymase activity were not performed in this study, and this proposed pathway remains a hypothesis derived from the observed proteomic associations rather than direct biochemical evidence.

However, our finding of a normotensive state amidst robust RAAS activation represents a significant departure from standard hemodynamics. Because the Fontan circuit lacks a subpulmonary pump, it is characterized by chronic preload deficiency and a fundamentally low cardiac output state. ⁵² , ⁵³ Under these physiological constraints, the preservation of mean arterial pressure becomes more dependent on compensatory increases in systemic vascular resistance. ⁵⁴ The upregulation of angiotensinogen identified in our proteomic screen likely provides the substrate necessary to fuel this homeostasis. ⁵⁵ In this hemodynamic context, normotension may serve as evidence of an active neurohormonal response needed to prevent circulatory collapse. Although direct measurement of local angiotensin II levels was not performed, this may offer a potential explanation for the divergence between our proteomic signatures and the observed clinical phenotype. Mechanistically, the dissociation between systemic pressure and HR can be driven by the enzymatic compartmentalization of the previously mentioned chymase pathway in conjunction with inflammatory modulation by pentraxin 3. Although our data demonstrate systemic ACE downregulation, the high levels of angiotensinogen are consistent with a possible shift toward local angiotensin II production within the myocardium, though this interpretation requires direct tissue‐level validation. Previous studies suggest that chymase activity is strictly localized to the interstitium due to its rapid neutralization by plasma serpins. ⁵⁶ This biological containment may allow for a local neurohormonal surge that drives sinus tachycardia while the systemic vasculature remains shielded from hypertension. The identification of pentraxin 3 in our panel adds a layer of inflammatory modulation to this model. This mechanism may reinforce the high‐resistance state required to sustain blood pressure in the postoperative Fontan heart.

Taking the above together, we propose a hypothesis where the acute surgical inflammation, marked by pentraxin 3, acts as the initial trigger for cardiac irritability, which is then amplified and sustained by the noncanonical RAAS pathway (Figure 6). The directional relationship between systemic inflammation and sinus tachycardia is complex and likely represents a bidirectional feed‐forward loop. Systemic inflammation is a well‐established catalyst for arrhythmogenesis. Elevated proinflammatory cytokines, such as tumor necrosis factor‐α, interleukin‐1β, and interleukin‐6, directly alter cardiac electrophysiology by disrupting intracellular calcium handling and modifying ion channel function. ⁵⁷ , ⁵⁸ , ⁵⁹ Conversely, tachycardia itself acts as a mechanical and metabolic stressor that can worsen existing inflammation. Previous clinical population studies have shown that elevated resting HRs independently correlate with increased systemic inflammatory markers like high‐sensitivity C‐reactive protein and interleukin‐6. ⁶⁰ , ⁶¹ While current literature favors acute surgical inflammation as the primary initiating driver of early postoperative arrhythmias, the ensuing tachycardia likely exacerbates local myocardial stress, perpetuating a cycle of sustained inflammation and autonomic dysfunction. Future studies should directly investigate the mechanistic interdependence between RAAS and pentraxin 3 expression to fully deconvolute this axis.

Figure 6. Proposed hypothesis for a neurohormonal–inflammatory axis driving acute postoperative Fontan sinus tachycardia.

This schematic illustrates a working model derived from the proteomic associations identified in this study. The abrupt transition to Fontan hemodynamics may trigger a systemic inflammatory response characterized by the upregulation of pentraxin‐3 and the recruitment of inflammatory cells. While systemic ACE expression is downregulated, the increase in circulating angiotensinogen can provide a continuous substrate for angiotensin II production. This process can shift from a systemic ACE‐dependent cascade to a localized, myocardial chymase‐dependent pathway. The resulting compartmentalized surge in angiotensin II within the myocardial interstitium could promote cardiac irritability and sinus tachycardia while maintaining necessary systemic vascular resistance through vasoconstriction and aldosterone release. ACE indicates angiotensin‐converting enzyme; AGT, angiotensinogen; and PTX3, pentraxin‐3.

Finally, this study has several limitations that should be acknowledged. In human patients, Fontan palliation is typically performed 2 years following the Glenn operation. In the ovine surgical model, however, the Fontan physiology is established in a single stage where the Glenn and inferior vena cava‐to‐pulmonary artery extracardiac conduit are performed simultaneously. Most of the blood flow returning to the heart is derived from the inferior vena cava. This abrupt hemodynamic shift may be responsible for the more dramatic changes observed in the ovine HR data. This is further supported by our proteomic data showing that significant differences in human patients were observed only after establishing the Fontan circulation and did not appear following the Glenn circulation. Furthermore, due to the retrospective nature of this analysis, representative ECG tracings corroborating our findings from these study animals were not available.

Slight differences exist within the postoperative management that may influence our HR and proteomics results. All medications taken by sheep and human subjects used for HR collections can be seen in Tables S10 and S11. A notable limitation of our retrospective design is the potential attenuation of the RAAS pathway by concurrent medications. Specifically, all patients received furosemide and spironolactone by POD 2 following the Fontan procedure, and 15 of 26 patients continued preoperative prescriptions for either enalapril or losartan. Both our ovine subjects and human patients received spironolactone and furosemide diuretic therapy postoperatively. Management strategies varied depending on patient needs, duration of treatment, and the use of other diuretic medications such as chlorothiazide in humans. The average postoperative HR in both human and animal subjects may have also been influenced by the as needed administration of opiate narcotics in the early postoperative period. However, all subjects assessed did not require narcotic analgesia by POD 3. Esmolol treatment was not required in human patients because they remained asymptomatic, unlike the ovine subjects, which exhibited symptoms from the postoperative sinus tachycardia. Within the animal cohort, the administration of esmolol to treat severe sinus tachycardia was needed in most but not all cases, introducing a potential confounding variable. Additionally, the inclusion of HR data during esmolol infusion limits our ability to assess the true maximum post‐Fontan HR. The primary goal of the intensive care management was to ensure the long‐term survival of the ovine Fontan model, necessitating these interventions. Regarding the proteomic profiling, while a small subset of samples from POD 5 were collected during esmolol treatment, the majority of the analyzed samples were entirely free of this intervention. Because the tachycardia response remained robust and the untreated samples formed the bulk of the data set, the identified proteomic signature should reflect the dominant biological response.

Another limitation involves the analytical constraints of predictive modeling. While the machine learning models yielded a highly predictive protein signature, they were developed on a restricted animal cohort. Training these algorithms on a limited number of replicates restricts their capacity to fully capture complex biological heterogeneity. These findings therefore require independent validation in larger cohorts to verify that the selected feature set generalizes effectively across broader populations. Additionally, AUC estimates derived from small data sets are inherently subject to higher variability under resampling and may be optimistically biased, as limited sample sizes can reduce the stability of performance metrics. Accordingly, the reported AUC values should be interpreted as preliminary estimates of discriminative capacity that require confirmation in an adequately powered external cohort.

Finally, the reliance on systemic serum proteomic signatures also restrains our findings. Circulating proteomic profiles provide valuable insights into systemic physiological shifts, but they do not offer direct evidence of localized atrial tissue remodeling. Despite this constraint, the identification of a highly predictive systemic biomarker signature represents a significant strength of this study. Future investigations must incorporate direct tissue analysis to effectively bridge the gap between these systemic proteomic alterations and local cardiac remodeling. Specifically, targeted quantification of local chymase expression alongside assessments of profibrotic and inflammatory cellular infiltration directly within the atrial tissue will be critical next steps in validating these proposed mechanisms. The short postoperative follow‐up period also limits our ability to assess the long‐term molecular signature after the Fontan procedure.

Ultimately, our findings pinpoint angiotensinogen, ACE, and pentraxin 3 signatures as promising, functional readouts of the neurohormonal and inflammatory effectors driving post‐Fontan sinus tachycardia. Future therapeutic strategies should investigate whether targeting these upstream molecular drivers mitigates the arrhythmic burden in this vulnerable patient population. Specifically, therapies targeting chymase‐dependent angiotensin II generation or the inflammatory amplification induced by pentraxin 3 may offer novel avenues for clinical management.

Sources of Funding

Funding for this study was supported through the Additional Ventures Cures Collaborative grant, the American Heart Association Predoctoral Fellowship (25PRE1375268) to F.T., and the Additional Ventures Collaborative Sciences Award (24AVCSASV1276047) to M.E.D. and C.K.B.

Disclosures

None.

Supporting information

Tables S1–S11

Figures S1–S7

ARRIVE Checklist