Immediate variations in high‐dimensional circulating immune cells following pancreatectomy

Abstract

Objectives

Despite advancements in surgical techniques and perioperative care, pancreatic surgery remains a high‐risk procedure with significant morbidity and mortality. Furthermore, understanding its impact on the immune system is essential for designing strategies that interact with it. The aim of this study was to elucidate the immunomodulation that occurs following pancreatectomy.

Methods

Patients were recruited for the IMMUNOPANC trial (NCT03978702). We performed mass cytometry to analyse the circulating immune subpopulations and integrated the data using the hierarchical‐stochastic neighbour embedding clustering analysis and Stabl algorithm.

Results

Among the enrolled 39 patients, 33 (84.5%) had undergone pancreatectomy for neoplasia including 13 (39%) with pancreatic ductal adenocarcinoma. Twelve (32%) patients developed pancreatic fistula with a 90‐day mortality rate of 2.5%. We phenotyped 156 samples and observed a significant increase in myeloid cells (26% vs. 34%, P = 0.018) after pancreatectomy. Natural killer cell proportions decreased on postoperative day one (POD1) compared with the preoperative levels (7% vs. 12%, P < 0.001). Similarly, both CD8⁺ and CD4⁺ T‐cell proportions decreased significantly post‐surgery (25% and 40%, respectively, P < 0.001). During a narrow window, we observed the alteration of NK cells, contraction of CD8⁺ T cells, and increase in the proportion of naive CD4⁺ T cells. These changes may be a result of the immune response to surgery but could also reflect lymphoid organ demargination.

Conclusion

This study presents the first analysis of peripheral immune system trajectories following pancreatic surgery. Understanding these dynamics would facilitate the restoration of postoperative immunity, which is potentially crucial for recovery.

Patients were recruited in the frame of the IMMUNOPANC trial (NCT03978702). We used mass cytometry to analyse circulating immune subpopulations, and integrated the data using hierarchical‐stochastic neighbour embedding clustering analysis. This study presents the first analysis of peripheral immune system trajectories following pancreatic surgery.

Introduction

Pancreatic surgical intervention for cancer poses a double risk. On the one hand, despite advancements in surgical techniques and improvements in perioperative care, it remains a high‐risk procedure with substantial morbidity (> 80% ¹ ) and mortality (8% ² ) when not performed in specialised centres. ³ On the other hand, pancreatic ductal adenocarcinoma (PDAC) continues to be a formidable clinical challenge with increasing incidence ⁴ and a low 13% 5‐year survival rate reported in 2021. ⁵ Although immunotherapy has yielded notable results in other cancer types such as melanoma ⁶ and non‐small‐cell lung cancer, ⁷ its effectiveness is limited in PDAC. This may be primarily attributed to three major obstacles: (1) dual deficiency with relatively low tumor mutational burden, leading to neoantigen scarcity and PDAC cell‐induced downregulation of the class I major histocompatibility complex (MHC‐I), ⁸ which are essential for antigen presentation to cytotoxic CD8⁺ T cells; (2) immunosuppressive tumor microenvironment with dense stromal components comprising cancer‐associated fibroblasts, extracellular matrix proteins and immunosuppressive myeloid cells (e.g. tumor‐associated macrophages and myeloid‐derived suppressor cells), which physically and functionally exclude effector T cells ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ ; (3) dysfunctional infiltrating T cells such as exhaustion phenotypes ¹⁵ characterised by upregulation of inhibitory receptors, such as programmed cell death protein 1 (PD‐1), lymphocyte activation gene 3 (LAG‐3), T‐cell immunoglobulin and mucin domain‐containing protein (TIM‐3), loss of effector functions (e.g. cytokine production and cytotoxicity) and metabolic dysregulation. Therefore, a catalyst is needed to induce a response. A key approach to improving the clinical efficacy of immunotherapy in PDAC is the incorporation of additional therapies that can transform the ‘cold’ tumor microenvironment to a ‘hot’ one with T‐cell infiltration and proinflammatory cytokine production. ¹⁶ , ¹⁷ To assess the immune efficiency of this strategy, the impact of pancreatic surgery on immune populations needs to be understood and a baseline needs to be established for comparison. However, limited data are available on this subject as few studies have investigated the influence of pancreatic surgery on the immune system. Moreover, these studies often adopt a reductionist approach that focuses on cell counts without a detailed analysis of immune cell phenotype ¹⁸ or interleukin secretion. ¹⁹ Therefore, the immune response to tissue damage induced by surgical procedures needs to be systematically characterised. Mass cytometry (CyTOF) enables single‐cell phenotyping similar to flow cytometry but without the limitations caused by spectral overlap from fluorochromes. ²⁰ , ²¹ CyTOF has been used to investigate the peripheral immune system in patients undergoing elective hip arthroplasty ²² or non‐cancer bowel resection. ²³

In this study, we used high‐dimensional mass cytometry to characterise the immunomodulatory effects of pancreatectomy. Our primary objective was to provide the first comprehensive profile of peripheral immune cell distribution and elucidate the trajectories of the immune system following a pancreatectomy.

Results

Participant characteristics and ‘shock wave’ from surgery

A total of 39 patients who underwent pancreatic surgery were prospectively enrolled in the study. Among them, 33 (84%) had undergone pancreatectomy for neoplasia including 13 (39%) with PDAC. Two patients underwent explorative laparotomy, and 37 patients underwent pancreatectomy. Patient characteristics are presented in Table 1, and the study workflow is outlined in Figure 1 starting from patient blood sampling up to the final computer‐based analysis. During the postoperative week, neutrophil and lymphocyte counts varied significantly compared with the preoperative values (Supplementary table 2 and Figure 2). A detailed analysis of lymphocyte subpopulations in the first five patients and an in‐depth examination of activation, inhibitory or maturation markers for natural killer (NK) and T cells indicated the same conclusion: considerable changes occurred following pancreatectomy with an increase or decrease observed depending on the specific marker. However, these values returned to baseline levels after POD1 with similar frequencies noted between POD3 and POD7 (Supplementary figure 3). This suggests a narrow window of perceptible immune changes.

Table 1.

Study participant characteristics and postoperative pancreatectomy outcomes

	Participants, no (%)
Sex
Male	18
Female	21
Age [mean (SD), years]	65 (± 13)
Report of surgery because of COVID 19	0
Surgery
Explorative laparotomy	2 (5)
Pancreatoduodenectomy	22 (56)
Distal pancreatectomy	14 (36)
With splenectomy	7 (18)
Total pancreatectomy	1 (2.5)
Pathology
No tumor	1 (2.5)
Benign tumor	5 (13)
Neoplasia	33 (84.5)
Including PDAC	13 (33)
Operative duration [mean (SD), h]	6 (± 2)
Outcomes
Clavien ≥3 (%)	8 (20)
Haemorrhage	6 (15)
POPF	12 (30)
Grade B	9 (23)
Grade C	3 (7)
Length of hospitalisation [mean (SD), days]	17 (± 8)
90‐day mortality	1 (2.5)

PDAC, pancreatic ductal adenocarcinoma; POPF, postoperative pancreatic fistula.

Figure 1.

Study workflow. Thirty‐nine patients underwent intent‐to‐treat surgery (pancreatectomy n = 36 with one total pancreatectomy; explorative laparotomy n = 2). Peripheral blood and clinical outcome data were collected prior to surgery (baseline) and at the indicated time points after surgery. The peripheral immune cells were cryopreserved after erythrocyte lysis. For further analysis, the peripheral immune cells were thawed, stained with cell‐phenotyping and intracellular cell‐signalling antibodies, and analysed using mass cytometry. Normalisation and batch‐correction were performed to minimise potential batch effects. Unsupervised bootstrapped clustering of immune cell subsets was performed using hierarchical‐stochastic neighbour embedding to identify differential immune cell dynamics. Further exploratory analyses were performed using an artificial intelligence algorithm based on previous clustering to compare exploratory laparotomy and pancreatectomy and determine which clusters and cluster‐interplay showed most variability between the two scenarios. Visual created using BioRender.com. AB, antibody; AI, artificial intelligence; CyTOF, mass cytometry; h‐SNE, hierarchical‐stochastic neighbour embedding; PBMCs, peripheral blood mononuclear cells; POD, postoperative day; PreopD, preoperative day.

Figure 2.

Variations induced by surgery in frequency of major immune populations. (a) Immune cells were clustered based on the expression of lineage markers using unsupervised h‐SNE analysis. The clusters were projected into two dimensions, and major immune cell compartments were identified based on phenotypic marker expression. (b) h‐SNE enables identification of myeloid cells (CD33⁺), B cells (CD33⁻ CD19⁺), NK cells (CD33⁻ CD³⁻ CD56^+/−), CD8⁺ T cells (CD33⁻ CD3⁺ CD8⁺), CD4⁺ T cells (CD33⁻ CD3⁺ CD4⁺), and Vd2⁺ gd T cells (CD33⁻ CD3⁺ Vd2⁺). (c) Boxplots depicting significant differences in cell frequency of each immune cell lineage at different time points. Non‐represented populations showed no significant variation. Results for all 39 surgical patients are presented as interquartile ranges, median, and whiskers from minimum to maximum. Samples are paired. The Y‐axis indicates the % of CD45⁺ cells; the X‐axis indicates blood samples for the patient or healthy individual (HV) at specific perioperative time points. Purple: HV, blue: preoperative day (PreopD) sample, green: per‐operative sample before the pancreatectomy (POD 0‐B), orange: per‐operative sample after the pancreatectomy (POD 0‐A), red: postoperative day 1 sample (POD1). P‐values indicate the significance between different groups. Statistical analysis was performed using analysis of variance (ANOVA) with Tukey or Games–Howell's multiple comparisons test. gd, gamma delta; h‐SNE, hierarchical‐stochastic neighbour embedding; HV, healthy volunteers; NK, natural killer; uc, unconventional. *P < 0.05, **P < 0.01, ***P < 0.001.

Based on these observations, we selected four time points for further analysis: before surgery to establish the patient baseline (PreopD), immediately after anaesthesia and skin incision to minimise confounding factors (POD 0‐B), immediately after pancreatectomy (POD 0‐A) and POD1. We focussed on the immediate effects of pancreatectomy before POD2. We performed mass cytometry to analyse 156 prospectively collected peripheral blood mononuclear cell (PBMC) samples from 39 patients. The hierarchical stochastic neighbour embedding (hSNE) algorithm ²⁴ was used to create two‐dimensional maps from high‐dimensional data to provide an unsupervised view of perioperative immunity. This analysis identified nine distinct meta clusters and helped visualise major immune cell types within the CD45⁺ compartment (Figure 2). The proportion of myeloid cells significantly increased on POD1 following pancreatectomy compared with that on PreopD (34% vs. 26%, P = 0.02) and was higher than that of the healthy volunteers (HVs) (34% vs. 19%, P < 0.001). Simultaneously, the frequency of NK cells increased from 12% to 20% (P = 0.005) immediately after pancreatectomy, but on POD1, it decreased below the preoperative level (7% vs. 12%, respectively; P < 0.001) or that in HVs (7% vs. 9%; ns). Both CD8⁺ and CD4⁺ T‐cell counts decreased significantly after pancreatectomy, as evidenced by a 25% decrease in CD8⁺ (POD 0‐A vs. POD 0‐B, 12% vs. 19%, P < 0.001) and a 40% decrease in CD4⁺ T‐cell counts (POD 0‐A vs. POD 0‐B, 15% vs. 28%, P < 0.001); however, these levels returned to baseline levels on POD1 (no significant difference compared with that on PreopD). Notably, the preoperative CD4⁺ T‐cell level in this cohort was significantly lower than that in HVs (25% vs. 36%, respectively; P < 0.01). All significant perioperative variations are presented in Figure 2.

In‐depth characterisation of immediate variations among lymphocyte subpopulations

Using the h‐SNE dimensionality reduction algorithm, ²⁴ we identified 16 clusters of NK cells that were classified into six subpopulations (Figure 3), 18 clusters of CD8⁺ T cells (Figure 4) and 21 clusters of CD4⁺ T cells (Figure 5) that were classified into four major subpopulations depending on their differentiation profiles. We observed significant changes in the frequencies of several cell subsets during the immediate perioperative period, which were visualised using the density map (Figure 6a). Compared with the baseline levels, the frequencies of the CD56⁻ CD16⁺ unconventional NK cells (Figure 6b) showed a gradual and significant increase after pancreatectomy in clusters #8 (P = 0.03) and #15 BTN2⁺ (P < 0.03). DNAM‐1⁺ frequency increased transiently in cluster #9 (P < 0.01) but returned to the baseline at POD1. The frequency of CD56^bright NK cells (cluster #7) significantly decreased after pancreatectomy (P < 0.01). Similarly, among the CD8⁺ T cells (Figure 6c), the frequencies of T effector memory T_EM (cluster #2, P = 0.02) and T central memory T_CM (clusters #6, P = 0.03 and #13 T_CM CD28+, P = 0.04) cells significantly decreased immediately after pancreatectomy but were promptly reestablished on POD1. The frequencies in cluster #11 (NK‐like T effector memory re‐expressing CD45RA T_EMRA CD16⁺ CD56⁺, P = 0.01) increased significantly immediately after pancreatectomy and returned to baseline at POD1. The only significant change in CD4⁺ T‐cell subset frequencies during the perioperative period occurred in the naive population (Figure 6d) in clusters #18 (P = 0.01), #20 (P = 0.02), and #21 (P = 0.03). Taken together, in‐depth characterisation of the immediate variations among lymphocyte subpopulations showed an increase in unconventional NK cells, NK‐like T_EMRA, and naïve CD4^v T cells and a decrease in immature NK cells, T_EM and T_CM CD8⁺ T cells.

Figure 3.

Annotation of NK cell clusters. (a) NK cells were clustered based on the expression of phenotypic markers using unsupervised h‐SNE analysis. The clusters were projected into two dimensions, and 16 NK cell clusters were identified based on phenotypic marker expression. (b) Identification of six NK cell subpopulations. (c) Heatmap showing normalised and batch‐corrected expression of phenotypic markers for the 16 NK cell h‐SNE clusters. (d) Summary of the major phenotypic characteristics used to define NK cell subsets. NK, natural killer; Uc, unconventional.

Figure 4.

Annotation of CD8⁺ T cell clusters. (a) CD8⁺ T cells were clustered based on the expression of phenotypic markers using unsupervised h‐SNE analysis. The clusters were projected into two dimensions, and 18 CD8⁺ T cell compartments were identified based on phenotypic marker expression. (b) Identification of four major subpopulations of CD8⁺ T cells. (c) Heatmap showing the normalised and batch‐corrected marker expression for 18 CD8⁺ T cell Cytosplore clusters. Clusters were formed based on surface marker expression profiles. (d) Surface marker combinations used to identify CD8⁺ T cell subsets. CM, central memory; EM, effector memory; EMRA, effector memory CD45RA⁺; NK, natural killer.

Figure 5.

Annotation of CD4⁺ T cell clusters. (a) CD4⁺ T cells were clustered using Cytosplore based on the expression of various phenotypic markers obtained using unsupervised h‐SNE analysis. The clusters were projected into two dimensions, and 21 CD4⁺ T cell compartments were identified based on phenotypic marker expression. (b) Identification of four major subpopulations of CD4⁺ T cells. (c) Heatmap showing the normalised and batch‐corrected marker expression for the 21 CD4⁺ T cell Cytosplore clusters. Clusters are formed based on surface marker expression profiles. (d) Surface marker combinations used to identify CD4⁺ T cell subsets. CM, central memory; EM, effector memory; Tregs, regulatory T cells.

Figure 6.

Major postoperative variations following pancreatectomy. (a) Variations in the density map pertaining to lymphocytes NK, CD8⁺, and CD4⁺ clustering after pancreatectomy. A density map indicating the local probability density of h‐SNE‐embedded cells; black dots represent centroids of clusters identified using Gaussian mean shift clustering. Cell clusters are colour‐coded according to the directional differences. (b) Major variations in NK clustering. (c) Major variation in CD8⁺ T cells. (d) Major variation in CD4⁺ T cells. Samples are paired. The Y‐axis indicates % of NK, CD8⁺, or CD4⁺ T‐cell populations as described in the gating strategy; the X‐axis indicates the blood sample from the patient or HV at specific perioperative time points. Purple: HV, blue: preoperative day (PreopD) sample, green: per‐operative sample before the pancreatectomy (POD 0‐B), orange: per‐operative sample after pancreatectomy (POD 0‐A), red: postoperative day 1 sample (POD1). P‐values indicate the significance between different groups. Statistical analysis was performed using analysis of variance (ANOVA) with Tukey/Games–Howell's multiple comparisons test or Kruskal–Wallis test; *P < 0.05, **P < 0.01, ***P < 0.001. h‐SNE, hierarchical‐stochastic neighbour embedding; HV, healthy volunteers; NK, natural killer; T_CM, central memory T cells; T_EM, effector memory T cells; T_EMRA, T_EM re‐expressing CD45RA; (#) indicate the corresponding cluster in the population HSNE analysis (Figures 3, 4, 5).

Integrated modelling of POD1 immune phenotypes differentiates patients based on pancreatectomy status

Finally, the extremely rare scenario of including explorative laparotomy patients warranted a discovery study. We used a proprietary artificial intelligence‐powered data integration platform to identify the pancreatectomy fingerprint by comparing patients who underwent explorative laparotomy with those who underwent effective pancreatectomy on POD1 (Supplementary results 1 and Supplementary figure 5). Samples from POD1 after pancreatectomy, compared to those from explorative laparotomy (Supplementary figure 4 and figure 5a), showed a distinct immune profile. This included lower frequencies of immature and memory‐like NK cells (#13: NKG2A⁺ CD57⁺ CD158ah⁻; #14: NKG2C⁺ CD57⁺), low frequencies of naïve T cells (CD45RA⁺ CD27⁻ CCR7⁺) with an immunosuppressive phenotype (CD39⁺ CD73⁺ and high BTLA expression), CD8⁺ and CD4⁺ T lymphocytes (#15 and #21), and CD8⁺ TEM cells expressing butyrophilin 3 (BTN3⁺) (#2: CD56⁻; #17: CD56⁺). In contrast, pancreatectomy samples had increased frequencies of tissue‐resident TEM cells (CD45RA⁻ CD27⁻ CCR7⁻ CD103⁺) in both CD8⁺ and CD4⁺ subsets (clusters #16 and #5, respectively).

Discussion

The single‐cell analysis of immune events before and shortly after pancreatectomy provided a system‐level view of immune mechanisms associated with the performance of pancreatic surgery in an expert centre. Multiple biological dimensions (including immune cell composition, activation, inhibition and maturation states) were assessed to capture the complex biology of patients undergoing similar major surgeries under general anaesthesia but with different underlying pathologies or postoperative outcomes.

Here, we explored two complementary approaches to assess the impact of pancreatectomy on circulating immune cells with a particular focus on lymphocyte populations. First, we analysed the longitudinal changes in circulating immune cells over time. Second, we performed an explorative study to compare lymphocyte subpopulations on POD1 between patients who had undergone pancreatectomy and those in the control group who had undergone an explorative laparotomy without resection.

Two major outcomes were observed while analysing the characteristics of the early immune response: (1) exacerbation of the response of unconventional NK cells and T_EM CD103⁺ T cells and (2) immunosuppressive response primarily involving immature and memory‐like NK cells, memory CD8⁺ T cell compartment and naïve T cells. However, these values returned to baseline levels after POD1 with similar frequencies noted from POD3 to POD7 (Supplementary figure 3). This suggests a narrow observational window.

Mass cytometric analysis showed a distinct phenotypic shift in NK cells following surgery such as a significant decrease in immature and memory‐like NK subpopulations and global increase in unconventional CD56⁻ CD16⁺ NK cells. This population has been described in acute myeloid leukaemia and is associated with adverse clinical outcomes and lowered leukaemia‐free survival. ²⁵ NK cells are a population of interest in solid tumors, and NK cell dysfunction is involved in postoperative metastases. ²⁶ , ²⁷ Moreover, NK cell count depletion is an independent negative prognostic factor in patients with colon cancer. ²⁸ This shift of balance from CD56^bright to unconventional CD56⁻ CD16⁺ NK potentially favors cytotoxic function at the expense of regulatory cytokine production. Long‐term follow‐up of patient survival is required to correlate the immediate shift in NK cells with prognosis in specific pancreatic pathology.

Integrin CD103 is a marker of tissue‐resident memory T cells ²⁹ and tumor‐reactive CD8⁺ tumor‐infiltrating lymphocytes. ³⁰ In the present study, the CD103⁺ T‐cell phenotype was associated with the increase in specific subpopulations of T_EM lymphocytes, CD8⁺ and CD4⁺ T cells, after pancreatectomy compared with that after explorative laparotomy. We hypothesised that these populations could originate from tumor cells because no significant associated variations were observed in central memory T cells, which would suggest that these cells gave rise to the incriminated T_EM CD103⁺ lymphocytes. ³¹

Recent investigation of laparoscopic surgical resections of colorectal cancer using CyTOF technology ³² or RNA sequencing for predicting postoperative pneumonia acquisition following major abdominal surgery ³³ showed that the postoperative immune response was predominantly suppressive. Specifically, a decrease of circulating NK cells and T cells with an intermediate increase in CD56^bright CD16⁺ NK cell subset was observed after colorectal surgery. Moreover, Donlon et al. highlighted the fact that in oesophageal surgery for cancer, ³⁴ the frequency of naïve T cells significantly increased in circulation post‐esophagectomy from POD0 to POD7 (P < 0.01) with a significant decrease in effector memory T cells. Overall, these findings concur with ours that surgery induces a ‘shock’ for the memory T‐cell compartment, and the naive CD4⁺ compartment is particularly sensitive to surgical stress with lymphoid organ demargination. This emphasises the need for subpopulation clustering to gain a better understanding.

In this study, the decrease in the different CD8⁺ T_EM subpopulations theoretically explains the increased susceptibility to infections after pancreatic surgery. Notably, this subpopulation showed increased expression of immune checkpoint proteins such as BTN3, which is a therapeutic target in cancer and a prognostic factor in PDAC. ³⁵ Similarly, the C39⁺ CD73⁺ BTLA⁺ phenotype of naïve T cells was primarily immunosuppressive because BTLA activation inhibits the function of CD8⁺ cancer‐specific T cells. ³⁶

The perioperative period is marked by a storm of (potential) tumor cell circulation and immunosuppression, and identifying the window of pro‐inflammatory and immunosuppressive shift would afford an opportunity to minimise the impact of the surgery. The alteration of NK cells, contraction of CD8⁺ T cells, and increase in naive CD4⁺ T cells observed in this study may be indicative of the immune response to surgery but also of lymphoid organ demargination. Functionally, these changes may be associated with higher IL‐6, IL‐7, IL‐10 levels and shift towards a Th2 > Th1 response. ³⁷ Furthermore, the postoperative increase in unconventional CD56⁻CD16⁺ NK cells may be a consequence of surgical stress and systemic inflammation rather than antitumor recruitment alone and could transiently dampen pro‐inflammatory functions. However, it could be a potential target for therapeutic antibodies.

Overall, the findings of the conventional analysis with univariate comparison between different time points and those of AI‐powered data integration are complementary and show similarities, reinforcing their clinical relevance. Notably, the small sample size, particularly in the explorative laparotomy group (n = 2), is a significant limitation of this study, which is quite exploratory by nature. This could have introduced bias and may limit the generalisability of the findings. Hence, future studies with larger cohorts are needed to confirm these results. However, the unusually rare cohort of explorative laparotomy included in this study could be viewed as a case control of pancreatic surgery, which has not been reported in the literature and presents valuable information. The derived model differentiated patients who underwent pancreatectomy from those who underwent explorative laparotomy with high performance, as indicated by an AUC of 0.94 on POD1. The non‐significance of the other models at different time points acted as a surrogate marker, which indicated that the observed variability on POD1 was effectively because of pancreatectomy. Even though model performance could not be directly compared because this is the first study to evaluate such a scenario, the single‐cell resolution from mass cytometry has provided new insights regarding cell‐type‐specific responses.

To the best of our knowledge, this is the first study to elucidate the immune impact at the cellular level following pancreatic surgery at an expert centre with multiple longitudinal time‐point monitoring and a large prospective cohort. We have highlighted the transient margination of immune populations and recirculation of resident populations within an extremely short timeframe following the surgery. NK cells and T cells play crucial roles in controlling tumor growth and immune response to pancreatic cancer. ³⁸ Therefore, the significant decrease following pancreatectomy could potentially create a window of opportunity for residual circulating cancer cells to proliferate and metastasize. To counteract this, clinicians could potentially administer immunotherapies that boost the function of NK and T cells post‐surgery to attenuate immunuosuppression. ³⁹ These approaches could include the use of immune checkpoint inhibitors for PD‐1/PD‐L1 pathway inhibition in sepsis, ⁴⁰ or adoptive cell transfer. ⁴¹ Another approach is to view surgery as a ‘local ablation treatment’ that releases tumoral antigens and immune cells, thereby providing a unique opportunity to gain temporary access to tissue populations that are usually difficult to reach to potentially modify antitumor immunity through combined immunotherapy ⁴² or cell infusion (NK therapy ⁴³ or $\mathit{\gamma \delta }$ T cells ⁴⁴ ). However, these strategies need to be carefully timed to coincide with the immune suppression window following surgery and tailored to the individual patient's immune response and tumor characteristics. Further studies are needed to determine the feasibility and effectiveness of these approaches in the context of pancreatic surgery and cancer.

This study has certain limitations. First, reproducibility and availability are major challenges as the premise includes an expert centre for both pancreatic surgery and immunomonitoring. Second, although mass cytometry enables simultaneous detection of up to 50 parameters at the single‐cell level, our study is hypothesis‐driven. Therefore, the use of fully unsupervised methods, such as single‐cell RNA sequencing, to analyse the whole immunome may help identify additional critical parameters. Third, the study is limited to a single analytical platform and lacks functional analysis and tissue samples, which could validate the correlations. Fourth, potential confounding factors that could have influenced our results, such as underlying patient health conditions, age or pathology, were not discussed in this study. Hence, future studies should consider these factors to provide a more comprehensive understanding of the immune response following pancreatectomy. Finally, the migration of the studied populations from the circulating compartment (nodes, spleen, and thymus) to other sites could explain the observed decrease in frequency of these subpopulations.

In conclusion, this study provides initial insights into the immune response following pancreatectomy. Overall, these results act as a baseline that could help facilitate further evaluation of the immunomodulatory effects of neoadjuvant treatment. Further studies are warranted to determine the impacts in clinically relevant scenarios such as post‐operative pancreatic fistula prediction, immune effect of splenectomy, or impact of surgery on prognosis. Future studies must be conducted in multi‐centre settings using larger sample sizes with consideration of confounding factors and a more detailed exploration of the clinical implications of our findings to validate these results.

Methods

Ethical statements, study design and participants

In this prospective trial, patients who underwent pancreatectomy at the Paoli–Calmettes Institute (Marseille, France) were enrolled following approval from the Institutional Review Board (IMMUNOPANC IPC 2018‐051) and the French ethics committee (CPP SUD‐EST 6). Written informed consent was obtained from each participant prior to inclusion. The study is registered at clinicaltrial.gov (NCT03978702). Our study protocol strictly adhered to the principles outlined in the Declaration of Helsinki and followed the European Directive 2001/20/CE, the General Data Protection Regulation 2016/679, and the reporting guidelines specified in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE ⁴⁵ ) framework. We included patients aged ≥ 18 years who were both willing and capable of signing a written consent form and who underwent pancreatectomy for either benign or malignant conditions. Patients who had undergone neoadjuvant chemotherapy, were pregnant or breastfeeding, or had a history of autoimmune diseases were excluded from the study. In total, 40 patients were enrolled in the study between February 23, 2021 and July 27, 2021. None of these patients were lost to follow‐up over the 1‐year study period. One patient who was initially included withdrew consent before the surgical procedure.

Data collection, blood sample collection and study workflow

Surgical procedures in this study were either performed laparoscopically (left pancreatectomy) or through laparotomy (pancreatoduodenectomy, PD). The abdominal cavity was thoroughly examined to confirm the absence of carcinomatosis, liver metastasis and para‐aortic lymph node metastasis. ⁴⁶ Consequently, the study exclusively included non‐metastatic patients, and all pancreatectomies were performed with curative intent. In cases where surgery was not recommended owing to perioperative contraindications, the patients were assigned to the exploratory laparotomy control group (n = 2). The variables evaluated in all patients were age, sex, coronavirus disease status (the trial occurred in 2021; so, preoperative coronavirus was frequent and could change the immune status), operative duration, blood loss, final pathology, pre‐ and postoperative blood samples with neutrophil and lymphocyte counts, C‐reactive protein, procalcitonin and postoperative drain fluid amylase. Postoperative complications were graded using the Clavien–Dindo classification ⁴⁷ and International Study Group of Pancreatic Surgery criteria for postoperative pancreatic fistula ⁴⁸ and haemorrhage ⁴⁹ ; additionally, 90‐day postoperative mortality rates and length of hospital stay were recorded. An expert centre was defined as an institution performing more than 50 PD procedures/year. ⁵⁰

Peripheral blood mononuclear cells (PBMCs) were collected at various time points and cryopreserved in a solution of 90% albumin and 10% dimethyl sulfoxide. These time points were as follows: 1 day before surgery (preoperative day [PreopD]; n = 39), immediately after abdominal incision (postoperative day 0, just before pancreatectomy [POD 0‐B]; n = 39), after pancreatectomy (POD 0‐A; n = 39), and on the first (POD1; n = 39), third (POD3; n = 39), and seventh postoperative day (POD7; n = 39). Additionally, PBMC samples were obtained from age‐matched HVs (n = 17). Handling, conditioning, and storage of these samples were overseen by the Paoli–Calmettes Tumor Bank, authorised by the French Ministry of Research under #AC‐2007‐33. Thawing and processing of PBMC samples were done by following established procedures. ²⁵ , ⁵¹ Briefly, PBMCs were washed with Roswell Park Memorial Institute (RPMI) medium 1640 supplemented with 10% fetal calf serum (FCS) and incubated at 37°C under 5% CO₂ for 30 min in RPMI 1640 supplemented with 2% FCS and Pierce Universal Nuclease 25 kU (Thermo Fisher Scientific) diluted in a 1:10000 ratio. Next, the cells were incubated with 1 μM cisplatin to stain the dead cells (Standard Biotools, San Francisco, California, USA). The phenotype markers were assessed concurrently to characterise major innate and adaptative immune cell subsets using two panels comprising 32 and 37 antibodies developed for deep phenotyping of NK and CD4⁺, CD8⁺, and Vδ2⁺ γδ T cells (Supplementary table 1). Antibodies for mass cytometry were either purchased from Standard Biotools or were conjugated from commercially available purified antibodies to the appropriate metal isotope using the MaxPar X8 Polymer or MCP9 Polymer kits (Standard Biotools).

Staining was performed as previously described by Ben Amara et al. ⁵¹ Overall, 0.5–3 million PBMCs were incubated for 1 h at 4 °C with the extracellular antibodies. Next, the cells were fixed using 2% paraformaldehyde (PFA), washed, and permeabilised with the Foxp3 Staining Buffer Set (eBioscience) for 10 min at 4°C. Intracellular aspecific epitopes were blocked with 0.5 mg/mL Human Fc Block for 10 min at 4°C before incubation with the mixture of intracellular antibodies for 30 min at 4°C in Foxp3 Staining Buffer. The cells were washed, and the samples were incubated with intracellular antibodies, followed by another wash and overnight incubation with 125 nM iridium intercalator (Standard Biotools) in 2% PFA. Finally, the cells were diluted in EQ™ Four Element Calibration Beads (Standard Biotools), which were used for normalisation over time.

The reference samples comprising buffy‐coat‐derived PBMCs (purchased from Etablissement Français du Sang [EFS], Marseille, France) were used at a concentration of 1.0 × 10⁶ cells/mL. The reference samples were treated in the same manner as those of the patient samples, and one reference sample was included in each batch of the quantified patient samples.

Data acquisition was performed at a rate of 300–400 events per second using Helios™ (Fluidigm, Standard Biotools), with 0.8–1.3 × 10⁶ events collected at each time point. Raw .fcs files underwent manual pretreatment using FlowJo v10.8.1. The generated raw FCS files were preprocessed before batch correction. Calibration beads were removed, and DNA‐positive cells were identified (¹⁹¹Ir and ¹⁹³Ir). Dead cells were excluded based on cisplatin positivity, and live cells were manually gated according to a predetermined gating strategy (Supplementary figure 1). Doublets were excluded using Gaussian parameters, event length and residual. All patient samples were stained and processed simultaneously on the CyTOF to minimise variations between measurements.

Batch effect correction and hierarchical‐stochastic neighbour embedding (h‐SNE)

A preprocessing step was applied using CytoBatchNorm (R script CyTOF Batch Adjust) ⁵² to correct the batch effects resulting from the 1‐year sample acquisition period, which involved 32 different batches. Channel‐specific batch‐to‐batch variation was evaluated using anchor samples, and adjustment factors were transferred to patient samples. Hierarchical‐stochastic neighbour embedding (h‐SNE) non‐supervised analysis was performed using Cytosplore V2.2.1, ²⁴ which enabled relevant clustering for use in further supervised analysis. For h‐SNE analyses, consensus files were generated for each group of patients using a fixed number of cells to obtain a representative and balanced view of all patient groups. Data were arcsinh‐transformed using a cofactor of 5. The patient clusters were defined using h‐SNE analysis with default settings (30 perplexity and 1000 iterations).

Study endpoint and two‐step methodology

The primary objective of this study was to elucidate the immunological alterations that occur after pancreatic resection with the goal of creating an ‘immune cartography of pancreatectomy’. Our secondary focus was on generating a comprehensive profile of NK cells and T lymphocytes owing to their significance in pancreatic cancer. We used a two‐step methodology to accomplish these objectives. The first step involved identifying the most suitable time points for analysis based on data from the first five patients. This approach aimed to maximise the observed variability while minimising the need for further analyses. In the second step, we performed a detailed analysis of the data obtained at the selected time points for the remaining 34 patients.

Statistical analysis

Data were analysed using IBM SPSS Statistics for Windows version 29 (IBM Corp., Armonk, NY, USA). The data are summarised as counts (frequencies) for categorical variables and as median [range] or mean (standard deviation) for quantitative variables. The characteristics of the populations were compared using Fisher's exact test or the chi‐squared test for qualitative variables and the Mann–Whitney or t‐test for quantitative variables. The Benjamini–Hochberg correction was used for multiple hypothesis testing. We used the analysis of variance with subsequent pairwise group comparisons with either the post hoc Tukey or Games–Howell tests for immune cluster comparison. All tests were two‐sided. The alpha for all tests was set at 0.05, and P < 0.05 was required for significance.

Statistical analysis for discovery study

Owing to the high dimensionality of the dataset, the differences in single‐cell features between patients who experienced effective pancreatectomy on POD1 and those who did not (explorative laparotomy group) were identified using a multivariable analysis performed using the Stabl algorithm (Biomics, version 2, Paris, France), ⁵³ which is a supervised machine‐learning framework that enables sparse, reliable and predictive feature selection. A specific preprocessing strategy was applied to each omic. No values were missing. Each preprocessing pipeline item was implemented using the Python package scikit‐learn v.1.2. Monte Carlo cross‐validation (with two repetitions of a 2‐fold cross‐validation scheme) was implemented using the Python package scikit‐learn v.1.2. Model performance was assessed by calculating the AUC. Significance was tested using an unpaired Mann–Whitney test on the model‐derived cross‐validated values.

Data visualisation

The trajectories of the immune circulating cells were derived from the arcsinh transformation of cell surface receptor expression and were represented as median (splines) with interquartile range (shaded area) using the R library splines v3.3.3 and ggplot2 v3.5.0.

We used a chord diagram representation to visualise inter‐omic correlations using the Python Holoviews v1.15 library. The correlation matrix of the features was computed based on the 39 POD1 samples. We used the Spearman correlation coefficient ($\rho$) to characterise the strength of the correlation.

Author contributions

Jonathan Garnier: Conceptualization; data curation; formal analysis; methodology; project administration; resources; software; writing – original draft. Anaïs Palen: Resources; writing – review and editing. Xavier Durand: Formal analysis; methodology; software; writing – review and editing. Jacques Ewald: Resources; writing – review and editing. Amira Ben Amara: Data curation; formal analysis; writing – review and editing. Marie Sarah Rouvière: Data curation; formal analysis; writing – review and editing. Samuel Granjeaud: Formal analysis; writing – review and editing. Gregoire Bellan: Formal analysis; methodology; software; writing – review and editing. Benjamin Choisy: Software; writing – review and editing. Franck Verdonk: Resources; writing – review and editing. Brice Gaudilliere: Resources; writing – review and editing. Caroline Gouarne: Project administration; resources; writing – review and editing. Olivier Turrini: Conceptualization; data curation; methodology; project administration; resources; writing – review and editing. Daniel Olive: Conceptualization; methodology; resources; writing – review and editing. Anne Sophie Chretien: Conceptualization; formal analysis; investigation; methodology; resources; software; writing – original draft.

Conflict of interest

DO is a cofounder and shareholder of ImCheck Therapeutics, Alderaan Biotechnology, and Emergence Therapeutics and has received research funds from ImCheck Therapeutics, Alderaan Biotechnology, Cellectis, and Emergence Therapeutics. The other authors declare no conflicts of interest.

Supporting information

Supplementary figure 1

Supplementary figure 2

Supplementary figure 3

Supplementary figure 4

Supplementary figure 5

Supplementary table 1

Supplementary table 2

Supplementary results 1

Data availability statement

The datasets generated and/or analysed during this study are not publicly available because of patient privacy concerns but are available from the corresponding author upon reasonable request.