Preoperative Lymphocyte Signature Predicts Pancreatic Fistula After Pancreatoduodenectomy

Preoperative Lymphocyte Signature Predicts Pancreatic Fistula After Pancreatoduodenectomy | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Preoperative Lymphocyte Signature Predicts Pancreatic Fistula After Pancreatoduodenectomy Jonathan Garnier, Gregoire Bellan, Anais Palen, Xavier Durand, and 10 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5219663/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 11 Feb, 2026 Read the published version in Communications Medicine → Version 1 posted You are reading this latest preprint version Abstract Postoperative pancreatic fistula (POPF) is the major driver of postoperative morbidity after pancreatoduodenectomy (PD). However, current preoperative prediction models lack precision. This study aimed to determine the ability of a high dimensional analysis from the patient’s peripheral immune system before PD using mass cytometry and sparse machine learning (ML), to predict POPF. Twenty-two patients in the prospective IMMUNOPANC trial (NCT03978702) underwent PD. Blood samples collected preoperatively were analyzed by combining single-cell mass cytometry and a new sparse ML pipeline, Stabl, to identify the most relevant POPF-predictive features. The logistic regression model output was evaluated using a five-fold cross-validation procedure. Eight (36%) patients experienced POPF (grade B, n = 7; grade C, n = 1). The multivariable predictive model comprised 11 features—six natural killer, three CD8 + T, and two CD4 + T lymphocyte cell clusters—revealing a preoperative POPF lymphocyte signature (Pancreatic Fistula Lymphocyte Signature, PFLS). The Stabl algorithm identified a predictive model classifying POPF patients with high performance (area under the receiver operating characteristic curve = 0.81, P = 2.04e-02). In summary, preoperative circulating immune-cell composition can predict POPF in patients undergoing pancreatoduodenectomy. Clinical application of the PFLS could potentially help identify high-risk populations and mitigate POPF risk. Biological sciences/Immunology/Translational immunology Health sciences/Medical research/Outcomes research Figures Figure 1 Figure 2 Figure 3 Figure 4 INTRODUCTION Postoperative pancreatic fistula (POPF) following pancreatectomy is a major healthcare problem. 1,2 Multiple factors are associated with POPF, including clinical characteristics (age, 3 sex, 4 high body mass index (BMI), 5 and sarcopenia 6 ), pancreas-related risk factors (pancreatic texture 3 and main duct size 7 ), and intraoperative blood loss. 8 National studies 9,10 have emphasized the need for pancreatic surgery to be performed in an expert center to reduce postoperative mortality and failure-to-rescue 11 rates owing to POPF. However, despite being performed at high-volume expert centers, clinically relevant POPF continues to occur, resulting in a significant healthcare burden. Robust tools to preoperatively predict POPF are lacking. Several efforts have focused on the intraoperative prediction of POPF, including developing a Fistula Risk Score (FRS), 8 and an updated alternative FRS (ua-FRS) 12 for pancreatoduodenectomy (PD). Conversely, some factors are influenced by subjectivity (especially parenchyma texture), and mitigation strategies could modify the estimated risk. 13 Objective postoperative biochemical markers, such as drain fluid amylase (DFA), 14 immune factors including C-reactive protein, 15 or immune status as the neutrophil-to-lymphocyte ratio (NLR), 16 are associated with POPF and the patient clinical course. The investigation of these intrinsic pathophysiological effects is limited, 17 and given the interactive and interconnected nature of the surgical immune response, 18 it may not be captured by a single straightforward risk score. Thus, a comprehensive understanding of contributing mechanisms, more accurate predictive tools, novel risk stratification, preventive strategies, therapeutic targets, and disruptive technology are needed. Application of high-dimensional omics technologies, such as mass cytometry by time-of-flight (CyTOF), has enabled the identification of patient-specific immune signatures that predict various surgical outcomes with high accuracy, including surgical-site infections after bowel resection, 19 protracted functional recovery, 20 and postoperative cognitive decline after primary hip arthroplasty. 21 This ability to evaluate a patient’s specific immune states before surgery has tremendous translational potential in assessing preoperative interventions to alter the patient’s immune state and surgical recovery. Therefore, better characterization of the immune system in patients with and without POPF is crucial for identifying predictive biomarkers and new therapeutic targets. This study aimed to determine the efficacy of utilizing high-dimensional analysis of the patient’s preoperative peripheral immune system, combining mass cytometry and sparse machine learning (ML), in predicting POPF after PD. RESULTS Study participant characteristics Twenty-two patients underwent PD; among them, 4 (18%) had pancreatic ductal adenocarcinoma (PDAC). Eight (36%) patients had POPF (grade B, n = 7 and grade C, n = 1), with a median delay of 3 days (range, 3–9 days). Patient characteristics are presented in Table 1 . Patients with POPF had significantly higher ua-FRS (33.5% vs. 17%, respectively; P = 0.014) and softer pancreatic parenchyma ( P = 0.018). Comparative pre-, intra-, and post-operative characteristics of patients with and without POPF are detailed in Supplementary Table 2. The overall 90-day mortality rate was 4.5%. Table 1 Study participant characteristics and postoperative outcomes of pancreatectomy Participants, No (%) Sex Male 11 (40) Female 11 (50) Age, mean (SD), years 67.7 ± 12 Report of surgery owing to COVID-19 0 Pathology No tumor 1 (4.5) Neoplasia 21 (95.5) Neoplasia including PDAC 4 (19) Soft pancreatic parenchyma texture 14 (64) Main pancreatic duct diameter 3.00 [2.00; 7.00] ua-FRS score * , median (IQR 25–75) 26.5 [16;34] Operative duration, mean (SD), min 443 ± 80.7 Outcomes Clavien ≥ 3 (%) 4 (18) Hemorrhage 3 (13) POPF 8 (36) Time to diagnosis of POPF £ , median (IQR 25–75), days 3 [3;4] Grade B £ 7 (0.875) Grade C £ 1 (0.125) Length of hospitalization, median (IQR 25–75) 18.0 [16.0; 23.5] 90-day mortality 1 (4.5) COVID-19, coronavirus disease; FRS, Fistula Risk Score; IQR, interquartile range; PDAC, pancreatic ductal adenocarcinoma; POPF, postoperative pancreatic fistula; SD, standard deviation; £, calculated on the POPF population Predictive model accurately identifies patients at risk for POPF based on their preoperative immune state Using the h-SNE approach in the whole IMMUNOPANC cohort (n = 39), we previously identified three mass cytometry data layers, comprising cell frequencies and maturation states, of six subpopulations of NK lymphocytes (16 clusters; Supplementary Fig. 2) and four subpopulations of both CD8 + T lymphocytes (18 clusters; Supplementary Fig. 3) and CD4 + T lymphocytes (21 clusters; Supplementary Fig. 4). Overall, 1,965 immune features were quantified, corresponding to either the frequency or the signaling activity of 36 receptors within 55 adaptive immune-cell subsets. We investigated the immune profiling of patient samples collected preoperatively to predict the subsequent development of POPF. Figure 2 describes the preoperative immune cartography in patients who underwent PD (n = 22). We employed Stabl, 22 a sparse ML method that combines multivariable predictive modeling with a data-driven feature selection process, using LR, RR, or RFC. The multivariable predictive model comprised eleven specific lymphocyte populations (clusters), as follows: six, three, and two in each lymphocyte family (NK, CD8 + T cells, and CD4 + T cells, respectively, Fig. 3 A), highlighting preoperative features predisposing patients to POPF after PD. The analysis yielded a model accurately classifying patients with and without subsequent POPF (Stabl + LR: area under the receiver operating characteristic curve [AUROC] = 0.81 [0.59, 0.98], P = 2.04e-02, Fig. 3 B, C). Figure S5 represents the comparison of the different prediction models on PreopD, with no improvement with more complex models such as RR (Stabl + RR: AUROC = 0.77 [0.51, 0.95], P = 4.5e-02) or RFC (Stabl + RF: AUROC = 0.73 [0.47, 0.93], P = 8.9e-02). Given our objective of achieving the highest possible accuracy in predicting the occurrence of POPF, the identification of true positive cases was of utmost importance. For example, when the recall was set at 90%, the specificity, PPV, and NPV of the PFLS were 70%, 64%, and 90%, respectively. Preoperative NK, CD4+, and CD8 + T cell attributes associated with POPF We identified the POPF immune fingerprint (pancreatic fistula lymphocyte signature [PFLS]) by comparing patients preoperatively with and without subsequent POPF in a very timely and concise period and in an expert center to reduce every other bias. Key differences in cell counts between patients with and without subsequent POPF were observed as an increase in early and late-stage NK cells, phenotypically defined as NKG2A + CD158ah- (clusters #6 and #13) and CD158ah+ (cluster #3), respectively, as well as NK CD56 bright NKG2A + cells (cluster #7). Furthermore, it is associated with a decrease in memory-like NKG2C + CD57 + NK cells (clusters #2 and 14). Regarding CD8 + T cells, effector memory (T EM ) CD8 + T cells expressing butyrophilin 3 and CD57, a marker defining a population of T cells that contributes to long-term immunological memory, 23 were enriched in patients with POPF (cluster #5). We also observed enrichment in a cluster of T EMRA CD8 + CD56 + CD16 + T cells (cluster #11), a phenotypic hallmark of T cells with NK cell-like functional properties. 24 These cells showed enhanced TCR-independent/CD16-dependent degranulation potential and expressed both CD57 and the exhaustion marker TIGIT. Notably, another cluster of T EMRA TIGIT + CD56- was decreased in patients with POPF (cluster #7). Regarding CD4 + T cells, exhausted regulatory T cells (T reg ) expressing CD16, CD56, CD127, CCR7, and TIGIT (cluster #18) were higher in patients with POPF, a population with reduced immunosuppressive potential recently identified in humans by Freuchet et al. 25 Stabl also identified a cluster of T reg with phenotypic characteristics of highly immunosuppressive potential, expressing the marker of residency CD103, the immune checkpoints ICOS and cytotoxic T-lymphocyte-associated protein-4 (CTLA-4), and the ectonucleotidase CD39 (cluster #17). To better understand the biological implications of the model, Stabl-selected features were organized on a chord diagram (Fig. 3 A). The most interconnected T lymphocyte features were CD56 + CD16 + TIGIT + T lymphocytes (clusters #11 for CD8 + T EMRA and #18 for CD4 + exT reg , \(\:\rho\:\) =0.72) and memory-like NK NKG2C+ (clusters #2 and #14, \(\:\rho\:\) =0.61). Another correlation was found between CLTA4 + CD4 + T reg (cluster #17) and both memory-like NK (clusters \(\:\#2,\:\rho\:\) =0.57 and #14, \(\:\rho\:\) =0.51). Predictive performance of the FRS in combination with mass cytometry lymphocyte signature A multivariable confounder analysis, including demographic and clinical factors known to influence POPF development, was performed but was underpowered (Supplementary Table 3). Comparing different pre-, intra-, and postoperative scores (Supplementary Table 4), the PFLS AUROC was 0.81, comparable to that of the ua-FRS. DFA and NLR on POD1 and POD3, respectively, showed similar or inferior AUROC values (0.86 and 0.72, respectively). Combining preoperative PFLS and intraoperative ua-FRS improved the prediction significantly (AUROC = 0.86, P = 7.35e-03). Figure 4 shows the variation in AUROC and prediction from PFLS and ua-FRS to the FRS-PFLS combination. DISCUSSION POPF is the most prevalent, as well as life-threatening, complication, affecting 10–30% of patients undergoing PD. 8,26,27 Furthermore, in patients with pancreatic tumors, the occurrence of POPF could lead to a complete failure of the oncologic strategy by delaying or annihilating the delivery of the indicated adjuvant chemotherapy. 28 Once POPF occurs, it is associated with a significant reduction in long-term survival. 29 However, the effective treatment of POPF remains a significant challenge in pancreatic surgery. POPF results in the loss of mechanical anastomotic integrity; therefore, screening the immune parameters involved in tissue repair 18,30 may improve the current prediction system based on clinical and subjective parameters. Precise prediction models for POPF development can provide proactive strategies for management and early intervention, as well as facilitate transparency in patient information. This study is an analysis of the circulating lymphocyte populations in patients before major pancreatic surgery (PD), aiming to identify immunological factors associated with the occurrence of subsequent POPF. The PFLS is the first sparse multivariable immune signature integrating high-dimensional mass cytometry data preoperatively in a novel ML pipeline to accurately identify patients at risk for developing subsequent POPF after PD. The predictive model comprised 11 single-cell features, was internally validated, and showed the benefit of combining with the currently available ua-FRS model (AUROC of 0.86). Postoperative immune suppression has been known for some time 31 ; however, studies showing the correlation between lymphocytes and complications after pancreatic surgery are scarce, 32 with small sample sizes from single institutions and without external validation. To advance this field, we propose an assessment of a patient’s immune profile in predicting POPF, aiming to improve the FRS prediction. This pilot study was motivated by a unique situation combining the following: 1) on-site mass cytometry expertise, enabling high-dimensional analysis of the phenotypic characteristics of immune cells; 2) pancreatic surgery expert centers rendering the best postoperative care possible; and 3) a new powerful ML pipeline that overcomes traditional statistical limitations by combining advanced multivariable analysis with Stabl and cross-validation assessment. To identify pertinent components of the intrinsic biology that may differentiate patients with POPF from those with complications, a variety of dimensions were evaluated, including clinical factors, immune-cell phenotype, and maturation profiles. The single-cell resolution afforded by mass cytometry provided new insights into cell-type-specific subtypes, which may contribute to the pathogenesis of POPF. The following three major immune dysfunctions characterize patients at risk for POPF: 1) a major increase in the NKG2A + NK population (four on six selected clusters); 2) a decrease in memory-like (NKG2C + CD57+) NK populations, which could be associated with NK hypomaturation 33 and a low response to infection; 3) an immune-depressive state with an increase in highly intercorrelated TIGIT + populations, CD8 + CD56 + CD16 + T EMRA cells, and CD4 + exT reg , which may be immunosuppressive and associated with an increased risk of subsequent nosocomial infection 34 and improper healing. Furthermore, the increase in the T reg CTLA4 + cluster in the non-POPF group could be owing to the increased proportion of PDAC in this population as a surrogate marker indicative of a more assertive underlying disease (less prevalent POPF in this subgroup because of the nature of the pancreatic parenchyma). A major mechanism for promoting the immunosuppressive state and progression in cancer involves the upregulation of pivotal immune inhibitory checkpoint pathways in T lymphocytes, such as CTLA-4/B7 or PD-1/PD-L1 in many neoplasias, 35 or more recently, TIGIT/PVR 36 in the PDAC tumoral microenvironment. Recently, the NKG2A axis has emerged as a novel immune checkpoint suppressing the cytolytic function of NK lymphocytes, 37 and is currently being investigated in clinical trials, including therapy-resistant non-small cell lung cancer, among others. Here, we revealed, for the first time, that similar circulating peripheral lymphocyte populations preoperatively are associated with severe postoperative complications. This could partly explain the dreadful situation of early recurrence after a surgical complication and the role of host immunity in such situations. 38 There are several implications of a better POPF stratification of high-risk patients on modifications of the clinical course before or during the surgery. Coupling POPF prediction with a structured prehabilitation program, which focuses on enhancing patients’ physical and immune resilience before surgery, may further reduce complications. These findings could also pave the way for immunotherapy, alone or combined with anti-CTLA4, anti-TIGIT, or anti-NKG2A antibodies in the perioperative context, to restore antitumor immunity 39 and prevent postoperative complications, similar to PD-1/PD-L1 pathway inhibition in sepsis. 40 Intraoperatively, there are fewer options but an extensive mitigation strategy could be adopted based on the patient’s risk profile. In very high-risk cases, ultimately, total pancreatectomy combined with intra-portal islet transplantation for cancer 41 (ongoing French study, TPIAT-01 NCT05116072) might be considered as a more definitive approach. A large validation cohort from India 42 revealed the ua-FRS AUROC to be 0.70, highlighting the need to optimize the POPF prediction model. On the one hand, to mitigate the issue of multicollinearity and prevent overfitting, the least absolute shrinkage and selection operator regression 43 (AUROC 0.85) or nonparametric models 44,45 , including random forest, neural networks, and extreme gradient boosting (AUROC 0.69–0.79), have been introduced. However, although the discriminatory power of the model may improve with the inclusion of postoperative indicators, this is detrimental to its clinical applicability. First, the excessive number of variables considered diminishes practicality; second, using postoperative indicators linked to POPF diagnosis (such as DFA) or occurring later in the postoperative pathway (such as delayed gastric emptying, site infection, or pathology) weakens its usability. On the other hand, predicting pancreatic characteristics preoperatively using a radiomics-based FRS is a recent promising challenge benefiting from ML integration. 46 Strengths and limitations To our knowledge, this is the first study combining high-dimensional mass cytometry analysis and a new sparse ML algorithm in pancreatic surgery. However, there are some limitations. First, one could argue that the POPF rate is relatively high (36%) for an expert center. It should be highlighted that this study focuses on a high-risk cohort without neoadjuvant treatment, characterized by a low PDAC rate (four patients, 19%), with mainly soft pancreatic parenchyma (64%) and high ua-FRS (median 26.5%). Second, although mass cytometry enables simultaneous detection of up to 50 parameters at the single-cell level, our study is hypothesis-driven; using fully unsupervised methods, such as single-cell RNA sequencing, to explore the whole immunome may enable the identification of additional critical parameters. Third, future experiments should integrate functional studies to validate the biological relevance of changes in phenotypes and their relationship to clinical outcomes. Fourth, the results should be converted to more widespread and faster technologies, such as flow cytometry, for bedside use. Recently, a backgating algorithm, Hypergate, was developed to minimize the number of markers for the detection of a population of interest. 47 A combination of both Hypergate and Stabl will enable the restriction of these investigations to a minimal panel of parameters likely to predict POPF, thereby facilitating the routine transfer of our prediction system. Finally, although mitigated by cross-validation, statistical-power assessment, and significant results, a small sample size can affect the statistical reproducibility and representativeness of the cohort; thereby, the findings require external validation. A prospective multicenter validation trial with the start-up SurgeCare is ongoing. Conclusion Analyzing the patient’s immunome before PD was found to have robust predictive potential to identify those at risk of developing POPF. Similarly, to reframe critical illness, 48 de-emphasizing POPF syndromes and focusing on the underlying biological changes will lead to better understanding, treatment, and prevention for patients and cost reduction for society. METHODS Ethical statements, study design, and participants This was a post-hoc analysis of the prospective IMMUNOPANC trial (NCT03978702). Briefly, patients undergoing pancreatectomy at the Paoli-Calmettes Institute were enrolled between February 23, 2021, and July 27, 2021, after approval by the Institutional Review Board (IMMUNOPANC IPC 2018-051) and the French ethics committee (CPP SUD-EST 6). Patients provided written informed consent before study participation. This study adhered to tenets of the Declaration of Helsinki, complied with the 2015 Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis Or Diagnosis statement, 49 and followed the European Directive 2001/20/CE and the General Data Protection Regulation 2016/679. Inclusion and exclusion criteria are reported in the Supplementary Methods. Figure 1 shows the study flowchart from patient blood sampling to computer analysis after batch normalization. Outcomes The primary clinical endpoint was the occurrence of POPF after PD, defined as a clinically relevant pancreatic fistula according to the International Study Group in Pancreatic Surgery (ISGPS) 2016 classification, 26 and graded as B or C. Data collection Variables collected are detailed in the Supplementary Methods. Patients were categorized into the following two groups: a control group (non-POPF) and a POPF group. All patients with PD had an estimated POPF risk based on the ua-FRS, 12 calculated prospectively and intraoperatively after pancreatic transection. Laboratory assessments used for POPF prediction were DFA and NLR on postoperative day 1 (POD1) and POD3, respectively, and were chosen based on externally validated scores. Study workflow, batch effect correction, and hierarchical-stochastic neighbor embedding (h-SNE) Peripheral blood mononuclear cells (PBMCs) were thawed and processed as reported in the Supplementary Table 1 (mass cytometry panel) and Fig. 1 (gating strategy to identify immune cells of interest in mass cytometry). Data acquisition was performed with Helios™ (Standard Biotools) at a rate of 300–400 events per second, and a total of 0.8–1.3 x 10 6 events were collected from each patient time point. Raw .fcs files were manually pretreated using FlowJo v10.8.1. Live cells were manually gated using the gating strategy (see Supplementary Fig. 1). The generated raw FCS files were preprocessed before batch correction. Calibration beads were removed, and DNA-positive cells were identified ( 191 Ir and 193 Ir). Dead cells were excluded based on cisplatin positivity, and the gating strategy was applied to manually gate the live cells. Doublets were excluded using Gaussian parameters—event length and residual. Batch normalization was performed using R-based CytoBatchNorm (R script CyTOF Batch Adjust) because the samples were acquired in in 18 batches over 1 year. Channel-specific batch-to-batch variation was evaluated using anchor samples, and adjustment factors were transferred to patient samples. For h-SNE analyses, consensus files were generated for each group of patients with a fixed number of cells to obtain a representative and balanced view of all patient groups. Data were arcsinh-transformed with a cofactor of 5. Patient clusters were defined using h-SNE analysis with default settings (30 perplexity and 1,000 iterations). The non-supervised dimensionality reduction algorithm of h-SNE implemented in Cytosplore (V2.2.1), 50 yielding three data layers comprising cell frequencies and maturation states of natural killer (NK) lymphocytes (16 clusters), CD8 + T lymphocytes (18 clusters), and CD4 + T lymphocytes (21 clusters) before and after pancreatectomy (n = 39; Supplementary Figs. 2, 3, and 4 present the annotation of NK cell clusters, CD8 + T cell clusters, and CD4 + T cell clusters, respectively). Clusters defined by h-SNE were used for further supervised analysis in the specific PD population (n = 22). Blood samples collected on the preoperative day were used to calculate the absolute cell count for each immune cluster. Samples from each patient were stained and processed simultaneously on the CyTOF to minimize variations between measurements. Power analysis A power analysis was performed to determine the required sample size based on the following parameters: an estimated AUROC of more than 0.90, indicating a high level of predictivity; a proportion of 0.40, representing the expected prevalence of POPF within our cohort; a confidence interval width of 0.25; and a confidence level of 90%. The resultant sample size required for our study was determined to be 21 patients. Statistical analysis Univariable analysis methods are described in the Supplementary Methods. A multivariable analysis was performed using the Stabl algorithm (Biomics, version 2, Paris, France), 22 a supervised ML framework that enables sparse, reliable, and predictive feature selection in high-dimensionality datasets. Preprocessing is described in the Supplementary Methods. To establish a predictive model forecasting the occurrence of postoperative POPF, the following three ML predictive classifiers were fitted: LR, ridge regression (RR), and random forest classifier (RFC). To assess our model’s predictive performance, we used a stratified 5-fold cross-validation approach (20% data are tested at each fold) to ensure that repartition of the observed outcome was preserved in each fold (implemented with the Python package scikit-learn v.1.2). The model performance was evaluated using precision, specificity, sensitivity/recall, positive predictive value (PPV), negative predictive value (NPV), AUROC, area under the precision-recall curve, and significance tested using an unpaired Mann–Whitney test. Metrics were performed on median predictions of the model over cross-validation. A post-hoc confounder analysis was performed using sex, BMI, and pancreatic texture with LR. Data visualization A chord diagram representation was used to visualize inter-omic correlations based on the 22 PreopD samples using the Python Holoviews v1.15 library. We used the Spearman correlation coefficient ( \(\:\rho\:\) ) to characterize the strength of the correlation. Declarations Acknowledgments This work was supported by the Fondation ARC (grant ARC#2022-00154 for ASC) and the Groupement d’intérêt scientifique -Infrastructures pour la Biologie, la Santé et l’Agronomie (GIS IBiSA). The team “Immunity and Cancer” was labeled “Equipe Fondation pour la Recherche Médicale (FRM) #2018-00198” (for D.O.). Author Contributions Conceptualization: Jonathan Garnier, Olivier Turrini, Anne-Sophie Chrétien, and Daniel Olive; Data curation: Jonathan Garnier, Olivier Turrini, Marie Sarah Rouvière, and Amira Ben-Amara; Formal analysis: Jonathan Garnier, Anne-Sophie Chrétien, Marie Sarah Rouvière, Amira Ben-Amara, Gregoire Bellan, and Xavier Durand; Investigation: Jonathan Garnier, Olivier Turrini, and Anne-Sophie Chrétien; Methodology: Jonathan Garnier, Olivier Turrini, Anne-Sophie Chrétien, Daniel Olive, Gregoire Bellan, and Xavier Durand; Project administration: Jonathan Garnier, Olivier Turrini, and Caroline Gouarné; Resources: Jonathan Garnier, Olivier Turrini, Anaïs Palen, Jacques Ewald, Caroline Gouarné, Anne-Sophie Chrétien, Daniel Olive, Frank Verdonk, and Brice Gaudilliere; Software: Jonathan Garnier, Anne-Sophie Chrétien, Benjamin Choisy, Gregoire Bellan, and Xavier Durand; Writing – original draft: Jonathan Garnier, Anne-Sophie Chrétien, Gregoire Bellan, and Xavier Durand; Writing – review & editing: all authors. Competing Interests Statement: The authors declare no conflicts of interest. Data Availability Statement : The datasets generated and/or analyzed during the current study are not publicly available because of patient privacy concerns but are available from the corresponding author upon reasonable request. References Ma, L. W., et al . The cost of postoperative pancreatic fistula versus the cost of pasireotide: results from a prospective randomized trial. Ann. Surg. 265 , 11–16 (2017). Williamsson, C., Ansari, D., Andersson, R. & Tingstedt, B. Postoperative pancreatic fistula-impact on outcome, hospital cost and effects of centralization. HPB (Oxford) 19 , 436–442 (2017). Wellner, U. F., et al . A simple scoring system based on clinical factors related to pancreatic texture predicts postoperative pancreatic fistula preoperatively. HPB (Oxford) 12 , 696–702 (2010). Yamamoto, Y. et al. A preoperative predictive scoring system for postoperative pancreatic fistula after pancreaticoduodenectomy. World J. Surg. 35 , 2747–2755 (2011). Roberts, K. J., et al . Scoring system to predict pancreatic fistula after pancreaticoduodenectomy: a UK multicenter study. Ann. Surg. 261 , 1191–1197 (2015). Nishida, Y., et al . Preoperative sarcopenia strongly influences the risk of postoperative pancreatic fistula formation after pancreaticoduodenectomy. J. Gastrointest. Surg. 20 , 1586–1594 (2016). Schuh, F., et al. A Simple Classification of Pancreatic Duct Size and Texture Predicts postoperative Pancreatic Fistula: A classification of the International Study Group of Pancreatic Surgery (ISGPS). Ann. Surg. 277 , e597–e608 (2023). Callery, M. P., Pratt, W. B., Kent, T. S., Chaikof, E. L. & Vollmer, C. M. A prospectively validated clinical risk score accurately predicts pancreatic fistula after pancreatoduodenectomy. J. Am. Coll. Surg. 216 , 1–14 (2013). Nimptsch, U., Krautz, C., Weber, G. F., Mansky, T. & Grützmann, R. Nationwide in-hospital mortality following pancreatic surgery in Germany is higher than anticipated. Ann. Surg. 264 , 1082–1090 (2016). Farges, O. et al. The theory and practice of pancreatic surgery in France. Ann. Surg. 266 , 797–804 (2017). El Amrani, M., et al . Referring patients to expert centers after pancreatectomy is too late to improve outcome. Inter-hospital transfer analysis in nationwide study of 19,938 patients. Ann. Surg. 272 , 723–730 (2020). Mungroop, T. H., et al . Updated alternative fistula risk score (ua-FRS) to include minimally invasive pancreatoduodenectomy: pan-European validation. Ann. Surg. 273 , 334–340 (2021). Casciani, F., Bassi, C. & Vollmer, C. M. Decision points in pancreatoduodenectomy: insights from the contemporary experts on prevention, mitigation, and management of postoperative pancreatic fistula. Surgery 170 , 889–909 (2021). Ven Fong, Z. V., et al . Early drain removal—the middle ground between the Drain versus No Drain debate in patients undergoing pancreaticoduodenectomy: A prospective validation study. Ann. Surg. 262 , 378–383 (2015). Van Hilst, J. et al . The inflammatory response after laparoscopic and open pancreatoduodenectomy and the association with complications in a multicenter randomized controlled trial. HPB (Oxford) 21 , 1453–1461 (2019). Garnier, J., et al . Establishment and external validation of neutrophil-to-lymphocyte ratio in excluding postoperative pancreatic fistula after pancreatoduodenectomy. BJS Open 7 , zrac124 (2023). Søreide, K., Healey, A. J., Mole, D. J. & Parks, R. W. Pre-, peri- and post-operative factors for the development of pancreatic fistula after pancreatic surgery. HPB (Oxford) 21 , 1621–1631 (2019). Verdonk, F., et al . Measuring the human immune response to surgery: multiomics for the prediction of postoperative outcomes. Curr. Opin. Crit. Care 27 , 717–725 (2021). Rumer, K. K., et al . Integrated single-cell and plasma proteomic modeling to predict surgical site complications: A prospective cohort study. Ann. Surg. 275 , 582–590 (2022). Fragiadakis, G. K., et al . Patient-specific immune states before surgery are strong correlates of surgical recovery. Anesthesiology 123 , 1241–1255 (2015). Verdonk, F., et al . An immune signature of postoperative cognitive decline in elderly patients. bioRxiv (2024). Hédou, J., et al . Discovery of sparse, reliable omic biomarkers with Stabl. Nat. Biotechnol. (2024). Ahmed, R., et al . CD57 + memory T cells proliferate in vivo. Cell Rep. 33 , 108501 (2020). Björkström, N. K., et al . Elevated numbers of FcγRIIIA+ (CD16+) effector CD8 T cells with NK cell-like function in chronic hepatitis C virus infection. J. Immunol. 181 , 4219–4228 (2008). Freuchet, A., et al . Identification of human exTreg cells as CD16 + CD56 + cytotoxic CD4 + T cells. Nat. Immunol. 24 , 1748–1761 (2023). Bassi, C., et al . The 2016 update of the International Study Group (ISGPS) definition and grading of postoperative pancreatic fistula: 11 years After. Surgery 161 , 584–591 (2017). Hirono, S., et al . Risk factors for pancreatic fistula grade C after pancreatoduodenectomy: A large prospective, multicenter Japan-Taiwan collaboration study. J. Hepatobiliary Pancreat Sci. 27 , 622–631 (2020). McMillan, M. T., et al . The characterization and prediction of ISGPF Grade C fistulas following pancreatoduodenectomy. J. Gastrointest. Surg. 20 , 262–276 (2016). Hank, T., et al . Association between pancreatic fistula and long-term survival in the era of neoadjuvant chemotherapy. JAMA Surg. 154 , 943–951 (2019). Reinke, S., et al . Terminally differentiated CD8⁺ T cells negatively affect bone regeneration in humans. Sci. Transl. Med. 5 , 177ra36 (2013). Slade, M. S., Simmons, R. L., Yunis, E. & Greenberg, L. J. Immunodepression after major surgery in normal patients. Surgery 78 , 363–372 (1975). Cohen, J. T., Charpentier, K. P., Miner, T. J., Cioffi, W. G. & Beard, R. E. Lymphopenia following pancreaticoduodenectomy is associated with pancreatic fistula formation. Ann. Hepatobiliary Pancreat. Surg. 25 , 242–250 (2021). Chretien, A. S., et al . Natural killer defective maturation is associated with adverse clinical outcome in patients with acute myeloid leukemia. Front. Immunol. 8 , 573 (2017). Conway Morris, A., et al . Cell-surface signatures of immune dysfunction risk-stratify critically ill patients: INFECT study. Intensive Care Med. 44 , 627–635 (2018). Sharma, P. & Allison, J. P. Immune checkpoint targeting in cancer therapy: toward combination strategies with curative potential. Cell 161 , 205–214 (2015). Werba, G., et al. Single-cell RNA sequencing reveals the effects of chemotherapy on human pancreatic adenocarcinoma and its tumor microenvironment. Nat. Commun. 14 , 797 (2023). Middelburg, J., et al . The MHC-E peptide ligands for checkpoint CD94/NKG2A are governed by inflammatory signals, whereas LILRB1/2 receptors are peptide indifferent. Cell Rep. 42 , 113516 (2023). Hyung, J., et al . Tumor immune-gene expression profiles and peripheral immune phenotypes associated with clinical outcomes of locally advanced pancreatic cancer following folfirinox. ESMO Open 7 , 100484 (2022). Davern, M., et al . PD-1 blockade attenuates surgery-mediated immunosuppression and boosts Th1 immunity perioperatively in oesophagogastric junctional adenocarcinoma. Front. Immunol. 14 , 1150754 (2023). Hotchkiss, R. S., et al. Immune Checkpoint Inhibition in Sepsis: A Phase 1b Randomized, Placebo-Controlled, Single Ascending Dose Study of Antiprogrammed Cell Death-Ligand 1 antibody (BMS-936559). Crit. Care Med. 47 , 632–642 (2019). Balzano, G., et al. Total Pancreatectomy with islet Autotransplantation as an Alternative to high-risk pancreatojejunostomy after pancreaticoduodenectomy: A Prospective Randomized Trial. Ann. Surg. 277 , 894–903 (2023). Shinde, R. S., et al . External validation and comparison of the original, alternative and updated-alternative fistula risk scores for the prediction of postoperative pancreatic fistula after pancreatoduodenectomy. Pancreatology 20 , 751–756 (2020). Gu, Z., et al . Development and validation of a novel nomogram to predict postoperative pancreatic fistula after pancreatoduodenectomy using lasso-logistic regression: an international multi-institutional observational study. Int. J. Surg. 109 , 4027–4040 (2023). Ashraf Ganjouei, A., et al . A machine learning approach to predict postoperative pancreatic fistula after pancreaticoduodenectomy using only preoperatively known data. Ann. Surg. Oncol. 30 , 7738–7747 (2023). Verma, A., et al. Machine learning-based prediction of postoperative pancreatic fistula following pancreaticoduodenectomy. Ann. Surg. 280 , 325–331 (2024). Ingwersen, E. W., et al . Radiomics preoperative-Fistula risk score (RAD-FRS) for pancreatoduodenectomy: development and external validation. BJS Open 7 , zrad100 (2023). Becht, E., et al . Reverse-engineering flow-cytometry gating strategies for phenotypic labelling and high-performance cell sorting. Bioinformatics 35 , 301–308 (2019). Maslove, D. M., et al . Redefining critical illness. Nat. Med. 28 , 1141–1148 (2022). Collins, G. S., Reitsma, J. B., Altman, D. G. & Moons, K. G. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD Statement. BMC Med. 13 , 1 (2015). Van Unen, V., et al . Visual analysis of mass cytometry data by hierarchical stochastic neighbour embedding reveals rare cell types. Nat. Commun. 8 , 1740 (2017). Additional Declarations There is NO Competing Interest. Supplementary Files SupplementaryDigtalcontents.docx Cite Share Download PDF Status: Published Journal Publication published 11 Feb, 2026 Read the published version in Communications Medicine → Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5219663","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":365859403,"identity":"5024a01c-4eb2-4281-878b-79c95fe3a4d6","order_by":0,"name":"Jonathan Garnier","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBUlEQVRIiWNgGAWjYLCCBAYGZn5mCJsHJshPUItkM5oWyQZCNhkcQBPAqUXe/eyzBw8q7rEbH+c9+IGx7Y4Mf/vZgx8+7mGQMMehx/BMurlBwpliZrPDfMkSjG3PeCTO5CVLznjGICGDbi9MS0Mam0RiWwJQC48ZA2PbYR4DhhwzZp4DDHUSOBxm2P8MqOVfArNxM0wL/xsz5j8HGCRwaZGXANnSkMBswAzTIgG0hQGPFgMJoC0JxxKYJUB+STh3mEfixhtjyZ4DErht6U9jk/xRk5DM3w8MqA9lh+35+3MMP/w4YIPbFmiwJIMjMQEhgUsD0JYGCG2HFO+jYBSMglEwClABAEIjTWTVAICNAAAAAElFTkSuQmCC","orcid":"","institution":"Institut Paoli Calmettes","correspondingAuthor":true,"prefix":"","firstName":"Jonathan","middleName":"","lastName":"Garnier","suffix":""},{"id":365859404,"identity":"899815f9-e0c1-4382-9231-0cba44c8cf77","order_by":1,"name":"Gregoire Bellan","email":"","orcid":"","institution":"SurgeCare","correspondingAuthor":false,"prefix":"","firstName":"Gregoire","middleName":"","lastName":"Bellan","suffix":""},{"id":365859405,"identity":"d4beb88b-8dcc-4417-ad71-0d04ac1b8e96","order_by":2,"name":"Anais Palen","email":"","orcid":"","institution":"Institut Paoli-Calmettes","correspondingAuthor":false,"prefix":"","firstName":"Anais","middleName":"","lastName":"Palen","suffix":""},{"id":365859406,"identity":"007f9194-b63b-4dc8-986b-62459ccd968f","order_by":3,"name":"Xavier Durand","email":"","orcid":"https://orcid.org/0009-0005-2359-1987","institution":"SurgeCare","correspondingAuthor":false,"prefix":"","firstName":"Xavier","middleName":"","lastName":"Durand","suffix":""},{"id":365859407,"identity":"d48cfe23-a121-4caf-bc31-71cd34474fa4","order_by":4,"name":"Jacques Ewald","email":"","orcid":"","institution":"Institut Paoli-Calmettes","correspondingAuthor":false,"prefix":"","firstName":"Jacques","middleName":"","lastName":"Ewald","suffix":""},{"id":365859408,"identity":"ca157b70-171b-4cdf-a493-1977394ada99","order_by":5,"name":"Amira Ben Amara","email":"","orcid":"","institution":": Immunomonitoring Platform, Cancer Research Center of Marseille, CRCM, INSERM U1068, CNRS UMR7258, Aix-Marseille University U105, Marseille, France","correspondingAuthor":false,"prefix":"","firstName":"Amira","middleName":"Ben","lastName":"Amara","suffix":""},{"id":365859409,"identity":"55c08c08-32bf-4f58-a36f-a8c0a0a92888","order_by":6,"name":"Marie-Sarah Rouviere","email":"","orcid":"","institution":": Immunomonitoring Platform, Cancer Research Center of Marseille, CRCM, INSERM U1068, CNRS UMR7258, Aix-Marseille University U105, Marseille, France","correspondingAuthor":false,"prefix":"","firstName":"Marie-Sarah","middleName":"","lastName":"Rouviere","suffix":""},{"id":365859410,"identity":"a3901552-1022-4a28-b07a-86d916f7a126","order_by":7,"name":"Benjamin Choisy","email":"","orcid":"","institution":"SurgeCare","correspondingAuthor":false,"prefix":"","firstName":"Benjamin","middleName":"","lastName":"Choisy","suffix":""},{"id":365859411,"identity":"69519e67-45bf-42d5-90ec-e8480b8fe173","order_by":8,"name":"Franck Verdonk","email":"","orcid":"","institution":"SurgeCare","correspondingAuthor":false,"prefix":"","firstName":"Franck","middleName":"","lastName":"Verdonk","suffix":""},{"id":365859412,"identity":"fc512cdf-88ca-4ea0-95d8-7cc38e984458","order_by":9,"name":"Brice Gaudilliere","email":"","orcid":"https://orcid.org/0000-0002-3475-5706","institution":"Stanford University","correspondingAuthor":false,"prefix":"","firstName":"Brice","middleName":"","lastName":"Gaudilliere","suffix":""},{"id":365859413,"identity":"80735bed-8bb8-4dff-894e-f17d8d59a58a","order_by":10,"name":"Caroline Gouarné","email":"","orcid":"","institution":"Institut Paoli-Calmettes","correspondingAuthor":false,"prefix":"","firstName":"Caroline","middleName":"","lastName":"Gouarné","suffix":""},{"id":365859414,"identity":"a0280975-77ab-40cd-9eb1-85824eb5d3dc","order_by":11,"name":"Olivier Turrini","email":"","orcid":"","institution":"Institut Paoli-Calmettes","correspondingAuthor":false,"prefix":"","firstName":"Olivier","middleName":"","lastName":"Turrini","suffix":""},{"id":365859415,"identity":"ef15ed6c-e352-4d45-853a-e04203a9f2b3","order_by":12,"name":"Daniel Olive","email":"","orcid":"https://orcid.org/0000-0003-1299-4113","institution":"Team Immunity and Cancer, Centre de Recherche en Cancérologie de Marseille (CRCM), INSERM U1068, CNRS UMR 7258, Aix-Marseille University UM105 and Paoli-Calmettes Institute, Marseille, France","correspondingAuthor":false,"prefix":"","firstName":"Daniel","middleName":"","lastName":"Olive","suffix":""},{"id":365859416,"identity":"07dbbd3a-0a85-44bc-b1e6-f47b6725f658","order_by":13,"name":"Anne Sophie Chretien","email":"","orcid":"https://orcid.org/0000-0003-4685-5568","institution":"Paoli Calmettes Institute","correspondingAuthor":false,"prefix":"","firstName":"Anne","middleName":"Sophie","lastName":"Chretien","suffix":""}],"badges":[],"createdAt":"2024-10-07 16:35:29","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5219663/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5219663/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s43856-026-01422-y","type":"published","date":"2026-02-11T05:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":78881764,"identity":"bfe8d171-95cd-4d95-95ea-89f373a14dba","added_by":"auto","created_at":"2025-03-20 08:48:04","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":142685,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWorkflow from PBMCs to POPF prediction. \u003c/strong\u003eFor\u003cstrong\u003e \u003c/strong\u003e22 patients undergoing intent-to-treat pancreatoduodenectomy, blood samples and clinical outcomes data are collected preoperatively (baseline) and at the indicated time points postoperatively. After erythrocyte lysis, peripheral immune cells are cryopreserved and thawed at the end of the study. After thawing, peripheral immune cells are stained with cell-phenotyping and intracellular cell-signaling antibodies and analyzed using mass cytometry. Normalization and batch correction are performed to minimize the potential batch effects. Unsupervised bootstrapped clustering of immune-cell subsets is performed using hierarchical-stochastic neighbor embedding to identify differential immune-cell dynamics. Artificial intelligence algorithm pipelines are based on the previous clustering to compare, preoperatively, patients who will subsequently have POPF after PD to determine which clusters and clusters-interplay contained most of the variability between the two situations. Created with BioRender.com. Ab, antibody; AI, artificial intelligence; CyTOF, mass cytometry; h-SNE, hierarchical-stochastic neighbor embedding; PBMCs, peripheral blood mononuclear cells; PD, pancreatoduodenectomy; PreopD, preoperative day; POD, postoperative day; POPF, postoperative pancreatic fistula; ua-FRS, updated alternative fistula risk score.\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5219663/v1/5b4ae06b128a6e880e7c6f9c.jpg"},{"id":78881763,"identity":"ab187b00-64f7-4fbe-a8e9-86109df84a57","added_by":"auto","created_at":"2025-03-20 08:48:04","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":166181,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePreoperative immune cartography in the IMMUNOPANC trial (n=22). \u003c/strong\u003eData are included for 22 patients who underwent pancreatoduodenectomy. Left panel: Color h-SNE maps representing the different phenotypes of peripheral lymphocytes NK, CD8+, and CD4+ T-cell clustering following pancreatectomy (16, 18, and 21 clusters, respectively) using Gaussian mean shift clustering. Right panels: density maps displaying the local probability density of hSNE-embedded cells for identified clusters before pancreatoduodenectomy. Black dots represent the centroids of the clusters identified in the PreopD samples. Cell clusters are color-coded according to their directional differences. h-SNE, hierarchical-stochastic neighbor embedding; NK, natural killer; PreopD, preoperative day.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5219663/v1/8fb994e7af039c9f13de7d4a.jpg"},{"id":78881766,"identity":"5083bdd6-0d96-4f1d-bc79-40398b934f9e","added_by":"auto","created_at":"2025-03-20 08:48:04","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":174361,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIntegrated modeling of preoperative immune signature to separate patients with and without POPF after PD. (A)\u003c/strong\u003e The Stabl algorithm is used to select the most significant variables in a preoperative setting (PreopD) to separate patients with POPF (n=8) or without POPF (n=14). Boxplots depict the variable chosen by Stabl in subpopulations of natural killer cells and CD8+ and CD4+ T lymphocytes. Red dots represent patients with POPF; gray dots represent patients without POPF. The Y-axis indicates the absolute count of cells (10\u003csup\u003e6\u003c/sup\u003e/L). The chord diagram depicts the correlation between model features of different data layers of lymphocytes (natural killer cells, CD8+ T cells, and CD4+ T cells). Data layers are highlighted using the color scheme. \u003cstrong\u003e(B)\u003c/strong\u003e Selected variables with Stabl differentiating patients with and without a POPF are integrated using a multivariable logistic regression (LR) method. The LR model output is evaluated using a five-fold cross-validation procedure. Derived models separated patients with effective POPF with high performance (Stabl + logistic regression, AUROC=0.81 95% CI [0.59–0.98]). The reference line represents the performance of a random guess. \u003cstrong\u003e(C)\u003c/strong\u003e Two-sided Mann–Whitney rank-sum test of the model outputs,\u003cem\u003e P\u003c/em\u003e=2.04e-02. 95% CI, 95% confidence interval; AUC, Area Under the Curve; EM, effector memory; exT\u003csub\u003ereg\u003c/sub\u003e, exhausted regulatory T lymphocytes; h-SNE, hierarchical-stochastic neighbor embedding; NK, natural killer; PreopD, preoperative day; POPF, Postoperative Pancreatic Fistula; PRC, Precision-Recall; ROC, receiver operating characteristic; T\u003csub\u003eEMRA\u003c/sub\u003e, EM T lymphocytes re-expressing CD45RA; T\u003csub\u003ereg\u003c/sub\u003e, regulatory T lymphocytes. \u003csup\u003e#\u003c/sup\u003eCorresponding cluster in the h-SNE analysis.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5219663/v1/95280bdd63b58ffaa4a7961a.jpg"},{"id":78882562,"identity":"f251c8d6-f75c-423b-80ff-f078df68a861","added_by":"auto","created_at":"2025-03-20 08:56:04","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":136596,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparison of derived models for POPF prediction. \u003c/strong\u003eThe reference line represents the performance of a random guess.\u003cstrong\u003e (A) \u003c/strong\u003ePFLS: Stabl + logistic regression.\u003cstrong\u003e (B) \u003c/strong\u003eua-FRS for pancreatoduodenectomy. \u003cstrong\u003e(C)\u003c/strong\u003e Combining preoperative PFLS and intra-operative ua-FRS significantly improved the prediction. AUC, Area Under the Curve; FRS, Fistula Risk Score; POPF, Postoperative Pancreatic Fistula; PRC, Precision-Recall; ROC, receiver operating characteristic.\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5219663/v1/aff16c9785394f388fa0e4d7.jpg"},{"id":105101956,"identity":"2359f1ca-de16-4d92-ae91-ee15bd403839","added_by":"auto","created_at":"2026-03-21 07:10:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1667776,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5219663/v1/fd4f3804-694f-49c5-802e-f5ece963cdaa.pdf"},{"id":78881767,"identity":"d4c58f62-8698-4c0d-910d-cc966aa58617","added_by":"auto","created_at":"2025-03-20 08:48:04","extension":"docx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":1673320,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryDigtalcontents.docx","url":"https://assets-eu.researchsquare.com/files/rs-5219663/v1/a4e452e584fc7a599d405582.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Preoperative Lymphocyte Signature Predicts Pancreatic Fistula After Pancreatoduodenectomy","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003ePostoperative pancreatic fistula (POPF) following pancreatectomy is a major healthcare problem.\u003csup\u003e1,2\u003c/sup\u003e Multiple factors are associated with POPF, including clinical characteristics (age,\u003csup\u003e3\u003c/sup\u003e sex,\u003csup\u003e4\u003c/sup\u003e high body mass index (BMI),\u003csup\u003e5\u003c/sup\u003e and sarcopenia\u003csup\u003e6\u003c/sup\u003e), pancreas-related risk factors (pancreatic texture\u003csup\u003e3\u003c/sup\u003e and main duct size\u003csup\u003e7\u003c/sup\u003e), and intraoperative blood loss.\u003csup\u003e8\u003c/sup\u003e National studies\u003csup\u003e9,10\u003c/sup\u003e have emphasized the need for pancreatic surgery to be performed in an expert center to reduce postoperative mortality and failure-to-rescue\u003csup\u003e11\u003c/sup\u003e rates owing to POPF. However, despite being performed at high-volume expert centers, clinically relevant POPF continues to occur, resulting in a significant healthcare burden. Robust tools to preoperatively predict POPF are lacking. Several efforts have focused on the intraoperative prediction of POPF, including developing a Fistula Risk Score (FRS),\u003csup\u003e8\u003c/sup\u003e and an updated alternative FRS (ua-FRS)\u003csup\u003e12\u003c/sup\u003e for pancreatoduodenectomy (PD). Conversely, some factors are influenced by subjectivity (especially parenchyma texture), and mitigation strategies could modify the estimated risk.\u003csup\u003e13\u003c/sup\u003e Objective postoperative biochemical markers, such as drain fluid amylase (DFA),\u003csup\u003e14\u003c/sup\u003e immune factors including C-reactive protein,\u003csup\u003e15\u003c/sup\u003e or immune status as the neutrophil-to-lymphocyte ratio (NLR),\u003csup\u003e16\u003c/sup\u003e are associated with POPF and the patient clinical course. The investigation of these intrinsic pathophysiological effects is limited,\u003csup\u003e17\u003c/sup\u003e and given the interactive and interconnected nature of the surgical immune response,\u003csup\u003e18\u003c/sup\u003e it may not be captured by a single straightforward risk score. Thus, a comprehensive understanding of contributing mechanisms, more accurate predictive tools, novel risk stratification, preventive strategies, therapeutic targets, and disruptive technology are needed.\u003c/p\u003e \u003cp\u003eApplication of high-dimensional omics technologies, such as mass cytometry by time-of-flight (CyTOF), has enabled the identification of patient-specific immune signatures that predict various surgical outcomes with high accuracy, including surgical-site infections after bowel resection,\u003csup\u003e19\u003c/sup\u003e protracted functional recovery,\u003csup\u003e20\u003c/sup\u003e and postoperative cognitive decline after primary hip arthroplasty.\u003csup\u003e21\u003c/sup\u003e This ability to evaluate a patient\u0026rsquo;s specific immune states before surgery has tremendous translational potential in assessing preoperative interventions to alter the patient\u0026rsquo;s immune state and surgical recovery. Therefore, better characterization of the immune system in patients with and without POPF is crucial for identifying predictive biomarkers and new therapeutic targets.\u003c/p\u003e \u003cp\u003eThis study aimed to determine the efficacy of utilizing high-dimensional analysis of the patient\u0026rsquo;s preoperative peripheral immune system, combining mass cytometry and sparse machine learning (ML), in predicting POPF after PD.\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy participant characteristics\u003c/h2\u003e \u003cp\u003eTwenty-two patients underwent PD; among them, 4 (18%) had pancreatic ductal adenocarcinoma (PDAC). Eight (36%) patients had POPF (grade B, n\u0026thinsp;=\u0026thinsp;7 and grade C, n\u0026thinsp;=\u0026thinsp;1), with a median delay of 3 days (range, 3\u0026ndash;9 days).\u003c/p\u003e \u003cp\u003ePatient characteristics are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Patients with POPF had significantly higher ua-FRS (33.5% vs. 17%, respectively; \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.014) and softer pancreatic parenchyma (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.018). Comparative pre-, intra-, and post-operative characteristics of patients with and without POPF are detailed in Supplementary Table\u0026nbsp;2. The overall 90-day mortality rate was 4.5%.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eStudy participant characteristics and postoperative outcomes of pancreatectomy\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eParticipants, No (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eSex\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eMale\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11 (40)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eFemale\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11 (50)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eAge, mean (SD), years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e67.7\u0026thinsp;\u0026plusmn;\u0026thinsp;12\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eReport of surgery owing to COVID-19\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003ePathology\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eNo tumor\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (4.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eNeoplasia\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21 (95.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eNeoplasia including PDAC\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (19)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eSoft pancreatic parenchyma texture\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14 (64)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eMain pancreatic duct diameter\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3.00\u0026nbsp;[2.00;\u0026nbsp;7.00]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eua-FRS score\u003csup\u003e*\u003c/sup\u003e, median (IQR 25\u0026ndash;75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26.5 [16;34]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eOperative duration, mean (SD), min\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e443\u0026thinsp;\u0026plusmn;\u0026thinsp;80.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eOutcomes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eClavien\u0026thinsp;\u0026ge;\u0026thinsp;3 (%)\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e4 (18)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eHemorrhage\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 (13)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003ePOPF\u003c/em\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e8 (36)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eTime to diagnosis of POPF\u003c/em\u003e\u003csup\u003e\u003cem\u003e\u0026pound;\u003c/em\u003e\u003c/sup\u003e, median (IQR 25\u0026ndash;75), days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e3 [3;4]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eGrade B\u003c/em\u003e\u003csup\u003e\u003cem\u003e\u0026pound;\u003c/em\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7 (0.875)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e\u003cem\u003eGrade C\u003c/em\u003e\u003csup\u003e\u003cem\u003e\u0026pound;\u003c/em\u003e\u003c/sup\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (0.125)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eLength of hospitalization, median (IQR 25\u0026ndash;75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e18.0 [16.0; 23.5]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003e90-day mortality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (4.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003eCOVID-19, coronavirus disease; FRS, Fistula Risk Score; IQR, interquartile range; PDAC, pancreatic ductal adenocarcinoma; POPF, postoperative pancreatic fistula; SD, standard deviation; \u0026pound;, calculated on the POPF population\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003ePredictive model accurately identifies patients at risk for POPF based on their preoperative immune state\u003c/h3\u003e\n\u003cp\u003eUsing the h-SNE approach in the whole IMMUNOPANC cohort (n\u0026thinsp;=\u0026thinsp;39), we previously identified three mass cytometry data layers, comprising cell frequencies and maturation states, of six subpopulations of NK lymphocytes (16 clusters; Supplementary Fig.\u0026nbsp;2) and four subpopulations of both CD8\u0026thinsp;+\u0026thinsp;T lymphocytes (18 clusters; Supplementary Fig.\u0026nbsp;3) and CD4\u0026thinsp;+\u0026thinsp;T lymphocytes (21 clusters; Supplementary Fig.\u0026nbsp;4). Overall, 1,965 immune features were quantified, corresponding to either the frequency or the signaling activity of 36 receptors within 55 adaptive immune-cell subsets. We investigated the immune profiling of patient samples collected preoperatively to predict the subsequent development of POPF. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e2\u003c/span\u003e describes the preoperative immune cartography in patients who underwent PD (n\u0026thinsp;=\u0026thinsp;22). We employed Stabl,\u003csup\u003e22\u003c/sup\u003e a sparse ML method that combines multivariable predictive modeling with a data-driven feature selection process, using LR, RR, or RFC. The multivariable predictive model comprised eleven specific lymphocyte populations (clusters), as follows: six, three, and two in each lymphocyte family (NK, CD8\u0026thinsp;+\u0026thinsp;T cells, and CD4\u0026thinsp;+\u0026thinsp;T cells, respectively, Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eA), highlighting preoperative features predisposing patients to POPF after PD. The analysis yielded a model accurately classifying patients with and without subsequent POPF (Stabl\u0026thinsp;+\u0026thinsp;LR: area under the receiver operating characteristic curve [AUROC]\u0026thinsp;=\u0026thinsp;0.81 [0.59, 0.98], \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.04e-02, Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eB, C). Figure S5 represents the comparison of the different prediction models on PreopD, with no improvement with more complex models such as RR (Stabl\u0026thinsp;+\u0026thinsp;RR: AUROC\u0026thinsp;=\u0026thinsp;0.77 [0.51, 0.95], \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;4.5e-02) or RFC (Stabl\u0026thinsp;+\u0026thinsp;RF: AUROC\u0026thinsp;=\u0026thinsp;0.73 [0.47, 0.93], \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;8.9e-02). Given our objective of achieving the highest possible accuracy in predicting the occurrence of POPF, the identification of true positive cases was of utmost importance. For example, when the recall was set at 90%, the specificity, PPV, and NPV of the PFLS were 70%, 64%, and 90%, respectively.\u003c/p\u003e\n\u003ch3\u003ePreoperative NK, CD4+, and CD8 + T cell attributes associated with POPF\u003c/h3\u003e\n\u003cp\u003eWe identified the POPF immune fingerprint (pancreatic fistula lymphocyte signature [PFLS]) by comparing patients preoperatively with and without subsequent POPF in a very timely and concise period and in an expert center to reduce every other bias.\u003c/p\u003e \u003cp\u003eKey differences in cell counts between patients with and without subsequent POPF were observed as an increase in early and late-stage NK cells, phenotypically defined as NKG2A\u0026thinsp;+\u0026thinsp;CD158ah- (clusters #6 and #13) and CD158ah+ (cluster #3), respectively, as well as NK CD56\u003csup\u003ebright\u003c/sup\u003e NKG2A\u0026thinsp;+\u0026thinsp;cells (cluster #7). Furthermore, it is associated with a decrease in memory-like NKG2C\u0026thinsp;+\u0026thinsp;CD57\u0026thinsp;+\u0026thinsp;NK cells (clusters #2 and 14).\u003c/p\u003e \u003cp\u003eRegarding CD8\u0026thinsp;+\u0026thinsp;T cells, effector memory (T\u003csub\u003eEM\u003c/sub\u003e) CD8\u0026thinsp;+\u0026thinsp;T cells expressing butyrophilin 3 and CD57, a marker defining a population of T cells that contributes to long-term immunological memory,\u003csup\u003e23\u003c/sup\u003e were enriched in patients with POPF (cluster #5). We also observed enrichment in a cluster of T\u003csub\u003eEMRA\u003c/sub\u003e CD8\u0026thinsp;+\u0026thinsp;CD56\u0026thinsp;+\u0026thinsp;CD16\u0026thinsp;+\u0026thinsp;T cells (cluster #11), a phenotypic hallmark of T cells with NK cell-like functional properties.\u003csup\u003e24\u003c/sup\u003e These cells showed enhanced TCR-independent/CD16-dependent degranulation potential and expressed both CD57 and the exhaustion marker TIGIT. Notably, another cluster of T\u003csub\u003eEMRA\u003c/sub\u003e TIGIT\u0026thinsp;+\u0026thinsp;CD56- was decreased in patients with POPF (cluster #7).\u003c/p\u003e \u003cp\u003eRegarding CD4\u0026thinsp;+\u0026thinsp;T cells, exhausted regulatory T cells (T\u003csub\u003ereg\u003c/sub\u003e) expressing CD16, CD56, CD127, CCR7, and TIGIT (cluster #18) were higher in patients with POPF, a population with reduced immunosuppressive potential recently identified in humans by Freuchet et al.\u003csup\u003e25\u003c/sup\u003e Stabl also identified a cluster of T\u003csub\u003ereg\u003c/sub\u003e with phenotypic characteristics of highly immunosuppressive potential, expressing the marker of residency CD103, the immune checkpoints ICOS and cytotoxic T-lymphocyte-associated protein-4 (CTLA-4), and the ectonucleotidase CD39 (cluster #17).\u003c/p\u003e \u003cp\u003eTo better understand the biological implications of the model, Stabl-selected features were organized on a chord diagram (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). The most interconnected T lymphocyte features were CD56\u0026thinsp;+\u0026thinsp;CD16\u0026thinsp;+\u0026thinsp;TIGIT\u0026thinsp;+\u0026thinsp;T lymphocytes (clusters #11 for CD8\u0026thinsp;+\u0026thinsp;T\u003csub\u003eEMRA\u003c/sub\u003e and #18 for CD4\u0026thinsp;+\u0026thinsp;exT\u003csub\u003ereg\u003c/sub\u003e, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\rho\\:\\)\u003c/span\u003e\u003c/span\u003e=0.72) and memory-like NK NKG2C+ (clusters #2 and #14, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\rho\\:\\)\u003c/span\u003e\u003c/span\u003e=0.61). Another correlation was found between CLTA4\u0026thinsp;+\u0026thinsp;CD4\u0026thinsp;+\u0026thinsp;T\u003csub\u003ereg\u003c/sub\u003e (cluster #17) and both memory-like NK (clusters \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\#2,\\:\\rho\\:\\)\u003c/span\u003e\u003c/span\u003e=0.57 and #14, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\rho\\:\\)\u003c/span\u003e\u003c/span\u003e=0.51).\u003c/p\u003e\n\u003ch3\u003ePredictive performance of the FRS in combination with mass cytometry lymphocyte signature\u003c/h3\u003e\n\u003cp\u003eA multivariable confounder analysis, including demographic and clinical factors known to influence POPF development, was performed but was underpowered (Supplementary Table\u0026nbsp;3). Comparing different pre-, intra-, and postoperative scores (Supplementary Table\u0026nbsp;4), the PFLS AUROC was 0.81, comparable to that of the ua-FRS. DFA and NLR on POD1 and POD3, respectively, showed similar or inferior AUROC values (0.86 and 0.72, respectively). Combining preoperative PFLS and intraoperative ua-FRS improved the prediction significantly (AUROC\u0026thinsp;=\u0026thinsp;0.86, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;7.35e-03). Figure\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows the variation in AUROC and prediction from PFLS and ua-FRS to the FRS-PFLS combination.\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003ePOPF is the most prevalent, as well as life-threatening, complication, affecting 10\u0026ndash;30% of patients undergoing PD.\u003csup\u003e8,26,27\u003c/sup\u003e Furthermore, in patients with pancreatic tumors, the occurrence of POPF could lead to a complete failure of the oncologic strategy by delaying or annihilating the delivery of the indicated adjuvant chemotherapy.\u003csup\u003e28\u003c/sup\u003e Once POPF occurs, it is associated with a significant reduction in long-term survival.\u003csup\u003e29\u003c/sup\u003e However, the effective treatment of POPF remains a significant challenge in pancreatic surgery. POPF results in the loss of mechanical anastomotic integrity; therefore, screening the immune parameters involved in tissue repair\u003csup\u003e18,30\u003c/sup\u003e may improve the current prediction system based on clinical and subjective parameters. Precise prediction models for POPF development can provide proactive strategies for management and early intervention, as well as facilitate transparency in patient information.\u003c/p\u003e \u003cp\u003eThis study is an analysis of the circulating lymphocyte populations in patients before major pancreatic surgery (PD), aiming to identify immunological factors associated with the occurrence of subsequent POPF. The PFLS is the first sparse multivariable immune signature integrating high-dimensional mass cytometry data preoperatively in a novel ML pipeline to accurately identify patients at risk for developing subsequent POPF after PD. The predictive model comprised 11 single-cell features, was internally validated, and showed the benefit of combining with the currently available ua-FRS model (AUROC of 0.86).\u003c/p\u003e \u003cp\u003ePostoperative immune suppression has been known for some time\u003csup\u003e31\u003c/sup\u003e; however, studies showing the correlation between lymphocytes and complications after pancreatic surgery are scarce,\u003csup\u003e32\u003c/sup\u003e with small sample sizes from single institutions and without external validation. To advance this field, we propose an assessment of a patient\u0026rsquo;s immune profile in predicting POPF, aiming to improve the FRS prediction. This pilot study was motivated by a unique situation combining the following: 1) on-site mass cytometry expertise, enabling high-dimensional analysis of the phenotypic characteristics of immune cells; 2) pancreatic surgery expert centers rendering the best postoperative care possible; and 3) a new powerful ML pipeline that overcomes traditional statistical limitations by combining advanced multivariable analysis with Stabl and cross-validation assessment. To identify pertinent components of the intrinsic biology that may differentiate patients with POPF from those with complications, a variety of dimensions were evaluated, including clinical factors, immune-cell phenotype, and maturation profiles.\u003c/p\u003e \u003cp\u003eThe single-cell resolution afforded by mass cytometry provided new insights into cell-type-specific subtypes, which may contribute to the pathogenesis of POPF. The following three major immune dysfunctions characterize patients at risk for POPF: 1) a major increase in the NKG2A\u0026thinsp;+\u0026thinsp;NK population (four on six selected clusters); 2) a decrease in memory-like (NKG2C\u0026thinsp;+\u0026thinsp;CD57+) NK populations, which could be associated with NK hypomaturation\u003csup\u003e33\u003c/sup\u003e and a low response to infection; 3) an immune-depressive state with an increase in highly intercorrelated TIGIT\u0026thinsp;+\u0026thinsp;populations, CD8\u0026thinsp;+\u0026thinsp;CD56\u0026thinsp;+\u0026thinsp;CD16\u0026thinsp;+\u0026thinsp;T\u003csub\u003eEMRA\u003c/sub\u003e cells, and CD4\u0026thinsp;+\u0026thinsp;exT\u003csub\u003ereg\u003c/sub\u003e, which may be immunosuppressive and associated with an increased risk of subsequent nosocomial infection\u003csup\u003e34\u003c/sup\u003e and improper healing. Furthermore, the increase in the T\u003csub\u003ereg\u003c/sub\u003e CTLA4\u0026thinsp;+\u0026thinsp;cluster in the non-POPF group could be owing to the increased proportion of PDAC in this population as a surrogate marker indicative of a more assertive underlying disease (less prevalent POPF in this subgroup because of the nature of the pancreatic parenchyma).\u003c/p\u003e \u003cp\u003eA major mechanism for promoting the immunosuppressive state and progression in cancer involves the upregulation of pivotal immune inhibitory checkpoint pathways in T lymphocytes, such as CTLA-4/B7 or PD-1/PD-L1 in many neoplasias,\u003csup\u003e35\u003c/sup\u003e or more recently, TIGIT/PVR\u003csup\u003e36\u003c/sup\u003e in the PDAC tumoral microenvironment. Recently, the NKG2A axis has emerged as a novel immune checkpoint suppressing the cytolytic function of NK lymphocytes,\u003csup\u003e37\u003c/sup\u003e and is currently being investigated in clinical trials, including therapy-resistant non-small cell lung cancer, among others. Here, we revealed, for the first time, that similar circulating peripheral lymphocyte populations preoperatively are associated with severe postoperative complications. This could partly explain the dreadful situation of early recurrence after a surgical complication and the role of host immunity in such situations.\u003csup\u003e38\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eThere are several implications of a better POPF stratification of high-risk patients on modifications of the clinical course before or during the surgery. Coupling POPF prediction with a structured prehabilitation program, which focuses on enhancing patients\u0026rsquo; physical and immune resilience before surgery, may further reduce complications. These findings could also pave the way for immunotherapy, alone or combined with anti-CTLA4, anti-TIGIT, or anti-NKG2A antibodies in the perioperative context, to restore antitumor immunity\u003csup\u003e39\u003c/sup\u003e and prevent postoperative complications, similar to PD-1/PD-L1 pathway inhibition in sepsis.\u003csup\u003e40\u003c/sup\u003e Intraoperatively, there are fewer options but an extensive mitigation strategy could be adopted based on the patient\u0026rsquo;s risk profile. In very high-risk cases, ultimately, total pancreatectomy combined with intra-portal islet transplantation for cancer\u003csup\u003e41\u003c/sup\u003e (ongoing French study, TPIAT-01 NCT05116072) might be considered as a more definitive approach.\u003c/p\u003e \u003cp\u003eA large validation cohort from India\u003csup\u003e42\u003c/sup\u003e revealed the ua-FRS AUROC to be 0.70, highlighting the need to optimize the POPF prediction model. On the one hand, to mitigate the issue of multicollinearity and prevent overfitting, the least absolute shrinkage and selection operator regression\u003csup\u003e43\u003c/sup\u003e (AUROC 0.85) or nonparametric models\u003csup\u003e44,45\u003c/sup\u003e, including random forest, neural networks, and extreme gradient boosting (AUROC 0.69\u0026ndash;0.79), have been introduced. However, although the discriminatory power of the model may improve with the inclusion of postoperative indicators, this is detrimental to its clinical applicability. First, the excessive number of variables considered diminishes practicality; second, using postoperative indicators linked to POPF diagnosis (such as DFA) or occurring later in the postoperative pathway (such as delayed gastric emptying, site infection, or pathology) weakens its usability. On the other hand, predicting pancreatic characteristics preoperatively using a radiomics-based FRS is a recent promising challenge benefiting from ML integration.\u003csup\u003e46\u003c/sup\u003e\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eStrengths and limitations\u003c/h2\u003e \u003cp\u003eTo our knowledge, this is the first study combining high-dimensional mass cytometry analysis and a new sparse ML algorithm in pancreatic surgery. However, there are some limitations. First, one could argue that the POPF rate is relatively high (36%) for an expert center. It should be highlighted that this study focuses on a high-risk cohort without neoadjuvant treatment, characterized by a low PDAC rate (four patients, 19%), with mainly soft pancreatic parenchyma (64%) and high ua-FRS (median 26.5%). Second, although mass cytometry enables simultaneous detection of up to 50 parameters at the single-cell level, our study is hypothesis-driven; using fully unsupervised methods, such as single-cell RNA sequencing, to explore the whole immunome may enable the identification of additional critical parameters. Third, future experiments should integrate functional studies to validate the biological relevance of changes in phenotypes and their relationship to clinical outcomes. Fourth, the results should be converted to more widespread and faster technologies, such as flow cytometry, for bedside use. Recently, a backgating algorithm, Hypergate, was developed to minimize the number of markers for the detection of a population of interest.\u003csup\u003e47\u003c/sup\u003e A combination of both Hypergate and Stabl will enable the restriction of these investigations to a minimal panel of parameters likely to predict POPF, thereby facilitating the routine transfer of our prediction system. Finally, although mitigated by cross-validation, statistical-power assessment, and significant results, a small sample size can affect the statistical reproducibility and representativeness of the cohort; thereby, the findings require external validation. A prospective multicenter validation trial with the start-up SurgeCare is ongoing.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eAnalyzing the patient\u0026rsquo;s immunome before PD was found to have robust predictive potential to identify those at risk of developing POPF. Similarly, to reframe critical illness,\u003csup\u003e48\u003c/sup\u003e de-emphasizing POPF syndromes and focusing on the underlying biological changes will lead to better understanding, treatment, and prevention for patients and cost reduction for society.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eEthical statements, study design, and participants\u003c/h2\u003e \u003cp\u003eThis was a post-hoc analysis of the prospective IMMUNOPANC trial (NCT03978702). Briefly, patients undergoing pancreatectomy at the Paoli-Calmettes Institute were enrolled between February 23, 2021, and July 27, 2021, after approval by the Institutional Review Board (IMMUNOPANC IPC 2018-051) and the French ethics committee (CPP SUD-EST 6). Patients provided written informed consent before study participation. This study adhered to tenets of the Declaration of Helsinki, complied with the 2015 Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis Or Diagnosis statement,\u003csup\u003e49\u003c/sup\u003e and followed the European Directive 2001/20/CE and the General Data Protection Regulation 2016/679. Inclusion and exclusion criteria are reported in the Supplementary Methods. Figure\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows the study flowchart from patient blood sampling to computer analysis after batch normalization.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eOutcomes\u003c/h2\u003e \u003cp\u003eThe primary clinical endpoint was the occurrence of POPF after PD, defined as a clinically relevant pancreatic fistula according to the International Study Group in Pancreatic Surgery (ISGPS) 2016 classification,\u003csup\u003e26\u003c/sup\u003e and graded as B or C.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eData collection\u003c/h2\u003e \u003cp\u003eVariables collected are detailed in the Supplementary Methods. Patients were categorized into the following two groups: a control group (non-POPF) and a POPF group. All patients with PD had an estimated POPF risk based on the ua-FRS,\u003csup\u003e12\u003c/sup\u003e calculated prospectively and intraoperatively after pancreatic transection. Laboratory assessments used for POPF prediction were DFA and NLR on postoperative day 1 (POD1) and POD3, respectively, and were chosen based on externally validated scores.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eStudy workflow, batch effect correction, and hierarchical-stochastic neighbor embedding (h-SNE)\u003c/h2\u003e \u003cp\u003ePeripheral blood mononuclear cells (PBMCs) were thawed and processed as reported in the Supplementary Table\u0026nbsp;1 (mass cytometry panel) and Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e1\u003c/span\u003e (gating strategy to identify immune cells of interest in mass cytometry). Data acquisition was performed with Helios\u0026trade; (Standard Biotools) at a rate of 300\u0026ndash;400 events per second, and a total of 0.8\u0026ndash;1.3 x 10\u003csup\u003e6\u003c/sup\u003e events were collected from each patient time point. Raw .fcs files were manually pretreated using FlowJo v10.8.1. Live cells were manually gated using the gating strategy (see Supplementary Fig.\u0026nbsp;1). The generated raw FCS files were preprocessed before batch correction. Calibration beads were removed, and DNA-positive cells were identified (\u003csup\u003e191\u003c/sup\u003eIr and \u003csup\u003e193\u003c/sup\u003eIr). Dead cells were excluded based on cisplatin positivity, and the gating strategy was applied to manually gate the live cells. Doublets were excluded using Gaussian parameters\u0026mdash;event length and residual. Batch normalization was performed using R-based CytoBatchNorm (R script CyTOF Batch Adjust) because the samples were acquired in in 18 batches over 1 year. Channel-specific batch-to-batch variation was evaluated using anchor samples, and adjustment factors were transferred to patient samples. For h-SNE analyses, consensus files were generated for each group of patients with a fixed number of cells to obtain a representative and balanced view of all patient groups. Data were arcsinh-transformed with a cofactor of 5. Patient clusters were defined using h-SNE analysis with default settings (30 perplexity and 1,000 iterations). The non-supervised dimensionality reduction algorithm of \u003cem\u003eh-SNE\u003c/em\u003e implemented in Cytosplore (V2.2.1),\u003csup\u003e50\u003c/sup\u003e yielding three data layers comprising cell frequencies and maturation states of natural killer (NK) lymphocytes (16 clusters), CD8\u0026thinsp;+\u0026thinsp;T lymphocytes (18 clusters), and CD4\u0026thinsp;+\u0026thinsp;T lymphocytes (21 clusters) before and after pancreatectomy (n\u0026thinsp;=\u0026thinsp;39; Supplementary Figs.\u0026nbsp;2, 3, and 4 present the annotation of NK cell clusters, CD8\u0026thinsp;+\u0026thinsp;T cell clusters, and CD4\u0026thinsp;+\u0026thinsp;T cell clusters, respectively). Clusters defined by h-SNE were used for further supervised analysis in the specific PD population (n\u0026thinsp;=\u0026thinsp;22). Blood samples collected on the preoperative day were used to calculate the absolute cell count for each immune cluster. Samples from each patient were stained and processed simultaneously on the CyTOF to minimize variations between measurements.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003ePower analysis\u003c/h2\u003e \u003cp\u003eA power analysis was performed to determine the required sample size based on the following parameters: an estimated AUROC of more than 0.90, indicating a high level of predictivity; a proportion of 0.40, representing the expected prevalence of POPF within our cohort; a confidence interval width of 0.25; and a confidence level of 90%. The resultant sample size required for our study was determined to be 21 patients.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eUnivariable analysis methods are described in the Supplementary Methods. A multivariable analysis was performed using the Stabl algorithm (Biomics, version 2, Paris, France),\u003csup\u003e22\u003c/sup\u003e a supervised ML framework that enables sparse, reliable, and predictive feature selection in high-dimensionality datasets. Preprocessing is described in the Supplementary Methods. To establish a predictive model forecasting the occurrence of postoperative POPF, the following three ML predictive classifiers were fitted: LR, ridge regression (RR), and random forest classifier (RFC). To assess our model\u0026rsquo;s predictive performance, we used a stratified 5-fold cross-validation approach (20% data are tested at each fold) to ensure that repartition of the observed outcome was preserved in each fold (implemented with the Python package scikit-learn v.1.2). The model performance was evaluated using precision, specificity, sensitivity/recall, positive predictive value (PPV), negative predictive value (NPV), AUROC, area under the precision-recall curve, and significance tested using an unpaired Mann\u0026ndash;Whitney test. Metrics were performed on median predictions of the model over cross-validation. A post-hoc confounder analysis was performed using sex, BMI, and pancreatic texture with LR.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eData visualization\u003c/h2\u003e \u003cp\u003eA chord diagram representation was used to visualize inter-omic correlations based on the 22 PreopD samples using the Python Holoviews v1.15 library. We used the Spearman correlation coefficient (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\rho\\:\\)\u003c/span\u003e\u003c/span\u003e) to characterize the strength of the correlation.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was supported by the Fondation ARC (grant ARC#2022-00154 for ASC) and the Groupement d’intérêt scientifique -Infrastructures pour la Biologie, la Santé et l’Agronomie (GIS IBiSA). The team “Immunity and Cancer” was labeled “Equipe Fondation pour la Recherche Médicale (FRM) #2018-00198” (for D.O.).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization: Jonathan Garnier, Olivier Turrini, Anne-Sophie Chrétien, and Daniel Olive;\u003c/p\u003e\n\u003cp\u003eData curation: Jonathan Garnier, Olivier Turrini, Marie Sarah Rouvière, and Amira Ben-Amara;\u003c/p\u003e\n\u003cp\u003eFormal analysis: Jonathan Garnier, Anne-Sophie Chrétien, Marie Sarah Rouvière, Amira Ben-Amara, Gregoire Bellan, and Xavier Durand;\u003c/p\u003e\n\u003cp\u003eInvestigation: Jonathan Garnier, Olivier Turrini, and Anne-Sophie Chrétien;\u003c/p\u003e\n\u003cp\u003eMethodology: Jonathan Garnier, Olivier Turrini, Anne-Sophie Chrétien, Daniel Olive, Gregoire Bellan, and Xavier Durand;\u003c/p\u003e\n\u003cp\u003eProject administration: Jonathan Garnier, Olivier Turrini, and Caroline Gouarné;\u003c/p\u003e\n\u003cp\u003eResources: Jonathan Garnier, Olivier Turrini, Anaïs Palen, Jacques Ewald, Caroline Gouarné, Anne-Sophie Chrétien, Daniel Olive, Frank Verdonk, and\u0026nbsp;Brice Gaudilliere;\u003c/p\u003e\n\u003cp\u003eSoftware: Jonathan Garnier, Anne-Sophie Chrétien, Benjamin Choisy, Gregoire Bellan, and Xavier Durand;\u003c/p\u003e\n\u003cp\u003eWriting – original draft: Jonathan Garnier, Anne-Sophie Chrétien, Gregoire Bellan, and Xavier Durand;\u003c/p\u003e\n\u003cp\u003eWriting – review \u0026amp; editing: all authors.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests Statement:\u0026nbsp;\u003c/strong\u003eThe authors declare no conflicts of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u003c/strong\u003e: The datasets generated and/or analyzed during the current study are not publicly available because of patient privacy concerns but are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eMa, L. W., \u003cem\u003eet al\u003c/em\u003e. The cost of postoperative pancreatic fistula versus the cost of pasireotide: results from a prospective randomized trial. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e265\u003c/strong\u003e, 11\u0026ndash;16 (2017).\u003c/li\u003e\n\u003cli\u003eWilliamsson, C., Ansari, D., Andersson, R. \u0026amp; Tingstedt, B. Postoperative pancreatic fistula-impact on outcome, hospital cost and effects of centralization. \u003cem\u003eHPB (Oxford)\u003c/em\u003e\u003cstrong\u003e19\u003c/strong\u003e, 436\u0026ndash;442 (2017).\u003c/li\u003e\n\u003cli\u003eWellner, U. F., \u003cem\u003eet al\u003c/em\u003e. A simple scoring system based on clinical factors related to pancreatic texture predicts postoperative pancreatic fistula preoperatively. \u003cem\u003eHPB (Oxford)\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 696\u0026ndash;702 (2010).\u003c/li\u003e\n\u003cli\u003eYamamoto, Y. \u003cem\u003eet al.\u003c/em\u003e A preoperative predictive scoring system for postoperative pancreatic fistula after pancreaticoduodenectomy. \u003cem\u003eWorld J. Surg.\u003c/em\u003e\u003cstrong\u003e35\u003c/strong\u003e, 2747\u0026ndash;2755 (2011).\u003c/li\u003e\n\u003cli\u003eRoberts, K. J., \u003cem\u003eet al\u003c/em\u003e. Scoring system to predict pancreatic fistula after pancreaticoduodenectomy: a UK multicenter study. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e261\u003c/strong\u003e, 1191\u0026ndash;1197 (2015).\u003c/li\u003e\n\u003cli\u003eNishida, Y., \u003cem\u003eet al\u003c/em\u003e. Preoperative sarcopenia strongly influences the risk of postoperative pancreatic fistula formation after pancreaticoduodenectomy. \u003cem\u003eJ. Gastrointest. Surg.\u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 1586\u0026ndash;1594 (2016).\u003c/li\u003e\n\u003cli\u003eSchuh, F., \u003cem\u003eet al.\u003c/em\u003e A Simple Classification of Pancreatic Duct Size and Texture Predicts postoperative Pancreatic Fistula: A classification of the International Study Group of Pancreatic Surgery (ISGPS). \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e277\u003c/strong\u003e, e597\u0026ndash;e608 (2023).\u003c/li\u003e\n\u003cli\u003eCallery, M. P., Pratt, W. B., Kent, T. S., Chaikof, E. L. \u0026amp; Vollmer, C. M. A prospectively validated clinical risk score accurately predicts pancreatic fistula after pancreatoduodenectomy. \u003cem\u003eJ. Am. Coll. Surg.\u003c/em\u003e\u003cstrong\u003e216\u003c/strong\u003e, 1\u0026ndash;14 (2013).\u003c/li\u003e\n\u003cli\u003eNimptsch, U., Krautz, C., Weber, G. F., Mansky, T. \u0026amp; Gr\u0026uuml;tzmann, R. Nationwide in-hospital mortality following pancreatic surgery in Germany is higher than anticipated. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e264\u003c/strong\u003e, 1082\u0026ndash;1090 (2016).\u003c/li\u003e\n\u003cli\u003eFarges, O. \u003cem\u003eet al.\u003c/em\u003e The theory and practice of pancreatic surgery in France. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e266\u003c/strong\u003e, 797\u0026ndash;804 (2017).\u003c/li\u003e\n\u003cli\u003eEl Amrani, M., \u003cem\u003eet al\u003c/em\u003e. Referring patients to expert centers after pancreatectomy is too late to improve outcome. Inter-hospital transfer analysis in nationwide study of 19,938 patients. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e272\u003c/strong\u003e, 723\u0026ndash;730 (2020).\u003c/li\u003e\n\u003cli\u003eMungroop, T. H., \u003cem\u003eet al\u003c/em\u003e. Updated alternative fistula risk score (ua-FRS) to include minimally invasive pancreatoduodenectomy: pan-European validation. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e273\u003c/strong\u003e, 334\u0026ndash;340 (2021).\u003c/li\u003e\n\u003cli\u003eCasciani, F., Bassi, C. \u0026amp; Vollmer, C. M. Decision points in pancreatoduodenectomy: insights from the contemporary experts on prevention, mitigation, and management of postoperative pancreatic fistula. \u003cem\u003eSurgery\u003c/em\u003e\u003cstrong\u003e170\u003c/strong\u003e, 889\u0026ndash;909 (2021).\u003c/li\u003e\n\u003cli\u003eVen Fong, Z. V., \u003cem\u003eet al\u003c/em\u003e. Early drain removal\u0026mdash;the middle ground between the Drain versus No Drain debate in patients undergoing pancreaticoduodenectomy: A prospective validation study. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e262\u003c/strong\u003e, 378\u0026ndash;383 (2015).\u003c/li\u003e\n\u003cli\u003eVan Hilst, J. \u003cem\u003eet al\u003c/em\u003e. The inflammatory response after laparoscopic and open pancreatoduodenectomy and the association with complications in a multicenter randomized controlled trial. \u003cem\u003eHPB (Oxford)\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 1453\u0026ndash;1461 (2019).\u003c/li\u003e\n\u003cli\u003eGarnier, J., \u003cem\u003eet al\u003c/em\u003e. Establishment and external validation of neutrophil-to-lymphocyte ratio in excluding postoperative pancreatic fistula after pancreatoduodenectomy. \u003cem\u003eBJS Open\u003c/em\u003e\u003cstrong\u003e7\u003c/strong\u003e, zrac124 (2023).\u003c/li\u003e\n\u003cli\u003eS\u0026oslash;reide, K., Healey, A. J., Mole, D. J. \u0026amp; Parks, R. W. Pre-, peri- and post-operative factors for the development of pancreatic fistula after pancreatic surgery. \u003cem\u003eHPB (Oxford)\u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 1621\u0026ndash;1631 (2019).\u003c/li\u003e\n\u003cli\u003eVerdonk, F., \u003cem\u003eet al\u003c/em\u003e. Measuring the human immune response to surgery: multiomics for the prediction of postoperative outcomes. \u003cem\u003eCurr. Opin. Crit. Care\u003c/em\u003e\u003cstrong\u003e27\u003c/strong\u003e, 717\u0026ndash;725 (2021).\u003c/li\u003e\n\u003cli\u003eRumer, K. K., \u003cem\u003eet al\u003c/em\u003e. Integrated single-cell and plasma proteomic modeling to predict surgical site complications: A prospective cohort study. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e275\u003c/strong\u003e, 582\u0026ndash;590 (2022).\u003c/li\u003e\n\u003cli\u003eFragiadakis, G. K., \u003cem\u003eet al\u003c/em\u003e. Patient-specific immune states before surgery are strong correlates of surgical recovery. \u003cem\u003eAnesthesiology\u003c/em\u003e\u003cstrong\u003e123\u003c/strong\u003e, 1241\u0026ndash;1255 (2015).\u003c/li\u003e\n\u003cli\u003eVerdonk, F., \u003cem\u003eet al\u003c/em\u003e. An immune signature of postoperative cognitive decline in elderly patients. \u003cem\u003ebioRxiv\u003c/em\u003e (2024).\u003c/li\u003e\n\u003cli\u003eH\u0026eacute;dou, J., \u003cem\u003eet al\u003c/em\u003e. Discovery of sparse, reliable omic biomarkers with Stabl. \u003cem\u003eNat. Biotechnol.\u003c/em\u003e (2024).\u003c/li\u003e\n\u003cli\u003eAhmed, R., \u003cem\u003eet al\u003c/em\u003e. CD57\u0026thinsp;+\u0026thinsp;memory T cells proliferate in vivo. \u003cem\u003eCell Rep.\u003c/em\u003e\u003cstrong\u003e33\u003c/strong\u003e, 108501 (2020).\u003c/li\u003e\n\u003cli\u003eBj\u0026ouml;rkstr\u0026ouml;m, N. K., \u003cem\u003eet al\u003c/em\u003e. Elevated numbers of Fc\u0026gamma;RIIIA+ (CD16+) effector CD8 T cells with NK cell-like function in chronic hepatitis C virus infection. \u003cem\u003eJ. Immunol.\u003c/em\u003e\u003cstrong\u003e181\u003c/strong\u003e, 4219\u0026ndash;4228 (2008).\u003c/li\u003e\n\u003cli\u003eFreuchet, A., \u003cem\u003eet al\u003c/em\u003e. Identification of human exTreg cells as CD16\u0026thinsp;+\u0026thinsp;CD56\u0026thinsp;+\u0026thinsp;cytotoxic CD4\u0026thinsp;+\u0026thinsp;T cells. \u003cem\u003eNat. Immunol.\u003c/em\u003e\u003cstrong\u003e24\u003c/strong\u003e, 1748\u0026ndash;1761 (2023).\u003c/li\u003e\n\u003cli\u003eBassi, C., \u003cem\u003eet al\u003c/em\u003e. The 2016 update of the International Study Group (ISGPS) definition and grading of postoperative pancreatic fistula: 11 years After. \u003cem\u003eSurgery\u003c/em\u003e\u003cstrong\u003e161\u003c/strong\u003e, 584\u0026ndash;591 (2017).\u003c/li\u003e\n\u003cli\u003eHirono, S., \u003cem\u003eet al\u003c/em\u003e. Risk factors for pancreatic fistula grade C after pancreatoduodenectomy: A large prospective, multicenter Japan-Taiwan collaboration study. \u003cem\u003eJ. Hepatobiliary Pancreat Sci.\u003c/em\u003e\u003cstrong\u003e27\u003c/strong\u003e, 622\u0026ndash;631 (2020).\u003c/li\u003e\n\u003cli\u003eMcMillan, M. T., \u003cem\u003eet al\u003c/em\u003e. The characterization and prediction of ISGPF Grade C fistulas following pancreatoduodenectomy. \u003cem\u003eJ. Gastrointest. Surg.\u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 262\u0026ndash;276 (2016).\u003c/li\u003e\n\u003cli\u003eHank, T., \u003cem\u003eet al\u003c/em\u003e. Association between pancreatic fistula and long-term survival in the era of neoadjuvant chemotherapy. \u003cem\u003eJAMA Surg.\u003c/em\u003e\u003cstrong\u003e154\u003c/strong\u003e, 943\u0026ndash;951 (2019).\u003c/li\u003e\n\u003cli\u003eReinke, S., \u003cem\u003eet al\u003c/em\u003e. Terminally differentiated CD8⁺ T cells negatively affect bone regeneration in humans. \u003cem\u003eSci. Transl. Med.\u003c/em\u003e\u003cstrong\u003e5\u003c/strong\u003e, 177ra36 (2013).\u003c/li\u003e\n\u003cli\u003eSlade, M. S., Simmons, R. L., Yunis, E. \u0026amp; Greenberg, L. J. Immunodepression after major surgery in normal patients. \u003cem\u003eSurgery\u003c/em\u003e\u003cstrong\u003e78\u003c/strong\u003e, 363\u0026ndash;372 (1975).\u003c/li\u003e\n\u003cli\u003eCohen, J. T., Charpentier, K. P., Miner, T. J., Cioffi, W. G. \u0026amp; Beard, R. E. Lymphopenia following pancreaticoduodenectomy is associated with pancreatic fistula formation. \u003cem\u003eAnn. Hepatobiliary Pancreat. Surg.\u003c/em\u003e\u003cstrong\u003e25\u003c/strong\u003e, 242\u0026ndash;250 (2021).\u003c/li\u003e\n\u003cli\u003eChretien, A. S., \u003cem\u003eet al\u003c/em\u003e. Natural killer defective maturation is associated with adverse clinical outcome in patients with acute myeloid leukemia. \u003cem\u003eFront. Immunol.\u003c/em\u003e\u003cstrong\u003e8\u003c/strong\u003e, 573 (2017).\u003c/li\u003e\n\u003cli\u003eConway Morris, A., \u003cem\u003eet al\u003c/em\u003e. Cell-surface signatures of immune dysfunction risk-stratify critically ill patients: INFECT study. \u003cem\u003eIntensive Care Med.\u003c/em\u003e\u003cstrong\u003e44\u003c/strong\u003e, 627\u0026ndash;635 (2018).\u003c/li\u003e\n\u003cli\u003eSharma, P. \u0026amp; Allison, J. P. Immune checkpoint targeting in cancer therapy: toward combination strategies with curative potential. \u003cem\u003eCell\u003c/em\u003e\u003cstrong\u003e161\u003c/strong\u003e, 205\u0026ndash;214 (2015).\u003c/li\u003e\n\u003cli\u003eWerba, G., \u003cem\u003eet al.\u003c/em\u003e Single-cell RNA sequencing reveals the effects of chemotherapy on human pancreatic adenocarcinoma and its tumor microenvironment. \u003cem\u003eNat. Commun.\u003c/em\u003e\u003cstrong\u003e14\u003c/strong\u003e, 797 (2023).\u003c/li\u003e\n\u003cli\u003eMiddelburg, J., \u003cem\u003eet al\u003c/em\u003e. The MHC-E peptide ligands for checkpoint CD94/NKG2A are governed by inflammatory signals, whereas LILRB1/2 receptors are peptide indifferent. \u003cem\u003eCell Rep.\u003c/em\u003e\u003cstrong\u003e42\u003c/strong\u003e, 113516 (2023).\u003c/li\u003e\n\u003cli\u003eHyung, J., \u003cem\u003eet al\u003c/em\u003e. Tumor immune-gene expression profiles and peripheral immune phenotypes associated with clinical outcomes of locally advanced pancreatic cancer following folfirinox. \u003cem\u003eESMO Open\u003c/em\u003e\u003cstrong\u003e7\u003c/strong\u003e, 100484 (2022).\u003c/li\u003e\n\u003cli\u003eDavern, M., \u003cem\u003eet al\u003c/em\u003e. PD-1 blockade attenuates surgery-mediated immunosuppression and boosts Th1 immunity perioperatively in oesophagogastric junctional adenocarcinoma. \u003cem\u003eFront. Immunol.\u003c/em\u003e\u003cstrong\u003e14\u003c/strong\u003e, 1150754 (2023).\u003c/li\u003e\n\u003cli\u003eHotchkiss, R. S., \u003cem\u003eet al.\u003c/em\u003e Immune Checkpoint Inhibition in Sepsis: A Phase 1b Randomized, Placebo-Controlled, Single Ascending Dose Study of Antiprogrammed Cell Death-Ligand 1 antibody (BMS-936559). \u003cem\u003eCrit. Care Med.\u003c/em\u003e\u003cstrong\u003e47\u003c/strong\u003e, 632\u0026ndash;642 (2019).\u003c/li\u003e\n\u003cli\u003eBalzano, G., \u003cem\u003eet al.\u003c/em\u003e Total Pancreatectomy with islet Autotransplantation as an Alternative to high-risk pancreatojejunostomy after pancreaticoduodenectomy: A Prospective Randomized Trial. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e277\u003c/strong\u003e, 894\u0026ndash;903 (2023).\u003c/li\u003e\n\u003cli\u003eShinde, R. S., \u003cem\u003eet al\u003c/em\u003e. External validation and comparison of the original, alternative and updated-alternative fistula risk scores for the prediction of postoperative pancreatic fistula after pancreatoduodenectomy. \u003cem\u003ePancreatology\u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 751\u0026ndash;756 (2020).\u003c/li\u003e\n\u003cli\u003eGu, Z., \u003cem\u003eet al\u003c/em\u003e. Development and validation of a novel nomogram to predict postoperative pancreatic fistula after pancreatoduodenectomy using lasso-logistic regression: an international multi-institutional observational study. \u003cem\u003eInt. J. Surg.\u003c/em\u003e\u003cstrong\u003e109\u003c/strong\u003e, 4027\u0026ndash;4040 (2023).\u003c/li\u003e\n\u003cli\u003eAshraf Ganjouei, A., \u003cem\u003eet al\u003c/em\u003e. A machine learning approach to predict postoperative pancreatic fistula after pancreaticoduodenectomy using only preoperatively known data. \u003cem\u003eAnn. Surg. Oncol.\u003c/em\u003e\u003cstrong\u003e30\u003c/strong\u003e, 7738\u0026ndash;7747 (2023).\u003c/li\u003e\n\u003cli\u003eVerma, A., \u003cem\u003eet al.\u003c/em\u003e Machine learning-based prediction of postoperative pancreatic fistula following pancreaticoduodenectomy. \u003cem\u003eAnn. Surg.\u003c/em\u003e\u003cstrong\u003e280\u003c/strong\u003e, 325\u0026ndash;331 (2024).\u003c/li\u003e\n\u003cli\u003eIngwersen, E. W., \u003cem\u003eet al\u003c/em\u003e. Radiomics preoperative-Fistula risk score (RAD-FRS) for pancreatoduodenectomy: development and external validation. \u003cem\u003eBJS Open\u003c/em\u003e\u003cstrong\u003e7\u003c/strong\u003e, zrad100 (2023).\u003c/li\u003e\n\u003cli\u003eBecht, E., \u003cem\u003eet al\u003c/em\u003e. Reverse-engineering flow-cytometry gating strategies for phenotypic labelling and high-performance cell sorting. \u003cem\u003eBioinformatics\u003c/em\u003e\u003cstrong\u003e35\u003c/strong\u003e, 301\u0026ndash;308 (2019).\u003c/li\u003e\n\u003cli\u003eMaslove, D. M., \u003cem\u003eet al\u003c/em\u003e. Redefining critical illness. \u003cem\u003eNat. Med.\u003c/em\u003e\u003cstrong\u003e28\u003c/strong\u003e, 1141\u0026ndash;1148 (2022).\u003c/li\u003e\n\u003cli\u003eCollins, G. S., Reitsma, J. B., Altman, D. G. \u0026amp; Moons, K. G. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD Statement. \u003cem\u003eBMC Med.\u003c/em\u003e\u003cstrong\u003e13\u003c/strong\u003e, 1 (2015).\u003c/li\u003e\n\u003cli\u003eVan Unen, V., \u003cem\u003eet al\u003c/em\u003e. Visual analysis of mass cytometry data by hierarchical stochastic neighbour embedding reveals rare cell types. \u003cem\u003eNat. Commun.\u003c/em\u003e\u003cstrong\u003e8\u003c/strong\u003e, 1740 (2017).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-5219663/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5219663/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePostoperative pancreatic fistula (POPF) is the major driver of postoperative morbidity after pancreatoduodenectomy (PD). However, current preoperative prediction models lack precision. This study aimed to determine the ability of a high dimensional analysis from the patient\u0026rsquo;s peripheral immune system before PD using mass cytometry and sparse machine learning (ML), to predict POPF. Twenty-two patients in the prospective IMMUNOPANC trial (NCT03978702) underwent PD. Blood samples collected preoperatively were analyzed by combining single-cell mass cytometry and a new sparse ML pipeline, Stabl, to identify the most relevant POPF-predictive features. The logistic regression model output was evaluated using a five-fold cross-validation procedure. Eight (36%) patients experienced POPF (grade B, n\u0026thinsp;=\u0026thinsp;7; grade C, n\u0026thinsp;=\u0026thinsp;1). The multivariable predictive model comprised 11 features\u0026mdash;six natural killer, three CD8\u0026thinsp;+\u0026thinsp;T, and two CD4\u0026thinsp;+\u0026thinsp;T lymphocyte cell clusters\u0026mdash;revealing a preoperative POPF lymphocyte signature (Pancreatic Fistula Lymphocyte Signature, PFLS). The Stabl algorithm identified a predictive model classifying POPF patients with high performance (area under the receiver operating characteristic curve\u0026thinsp;=\u0026thinsp;0.81, \u003cem\u003eP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.04e-02). In summary, preoperative circulating immune-cell composition can predict POPF in patients undergoing pancreatoduodenectomy. Clinical application of the PFLS could potentially help identify high-risk populations and mitigate POPF risk.\u003c/p\u003e","manuscriptTitle":"Preoperative Lymphocyte Signature Predicts Pancreatic Fistula After Pancreatoduodenectomy","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-20 08:47:59","doi":"10.21203/rs.3.rs-5219663/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"communications-medicine","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"commsmed","sideBox":"Learn more about [Communications Medicine](http://www.nature.com/commsmed)","snPcode":"43856","submissionUrl":"https://mts-commsmed.nature.com/cgi-bin/main.plex","title":"Communications Medicine","twitterHandle":"@commsmedicine","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Communications Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"210c4a16-2e6e-4b3b-832c-fb67dfea2ded","owner":[],"postedDate":"March 20th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":38920052,"name":"Biological sciences/Immunology/Translational immunology"},{"id":38920053,"name":"Health sciences/Medical research/Outcomes research"}],"tags":[],"updatedAt":"2026-03-21T07:10:43+00:00","versionOfRecord":{"articleIdentity":"rs-5219663","link":"https://doi.org/10.1038/s43856-026-01422-y","journal":{"identity":"communications-medicine","isVorOnly":false,"title":"Communications Medicine"},"publishedOn":"2026-02-11 05:00:00","publishedOnDateReadable":"February 11th, 2026"},"versionCreatedAt":"2025-03-20 08:47:59","video":"","vorDoi":"10.1038/s43856-026-01422-y","vorDoiUrl":"https://doi.org/10.1038/s43856-026-01422-y","workflowStages":[]},"version":"v1","identity":"rs-5219663","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5219663","identity":"rs-5219663","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}