The study protocol was approved by the ethics committee at Ulm University (ethical grant no. 30-20). The protocol was conducted following the Declaration of Helsinki. All study participants were extensively informed by an attending physician and gave their written, explicit consent. Data of study participants was pseudonymized.

Female, adult (≥ 18 years) patients that underwent bariatric surgery at Ulm University Hospital between 2020 and 2022 and met the criteria according to the “German S3-guideline: Surgery for obesity and metabolic diseases” (BMI ≥ 40 kg/m² or BMI ≥ 35 kg/m² with one or more obesity-associated comorbidities) were included in the study. Patients suffering from inflammatory bowel disease, systemic inflammatory disease, acute infections, cancer, autoimmune disease or receiving immunosuppressive therapy were excluded from the study. Study participants included in the patient group were of non-Hispanic White ethnicity.

Initial baseline comparison (prior to surgery) using conventional flow cytometry included 38 female patients and 10 healthy, female age-matched controls (CTRL). The CTRL group consisted of non-Hispanic White normal weight (BMI ≥ 18.5 and BMI ≤ 25.0 kg/m²), female employees engaged at Ulm University Hospital. According to a self-declaration of health, study participants of the CTRL group did not suffer from any acute and chronic diseases or physical restrictions. Furthermore, study participants of the CTRL group did not receive medications including antihypertensives, antidepressants, antidiabetics, statins, vitamin supplements, thyroid hormones, proton-pump inhibitors, or anti-inflammatory drugs. Additionally, clinical blood parameters were determined for the CTRL cohort to exclude inflammatory and metabolic disorders ( Supplementary Table 1 ).

To assess the effect of bariatric surgery on the peripheral blood immune system, a subcohort of 12 female patients was randomly selected and samples were acquired at baseline, 1 – 2 months post-surgery (p.s.), 3 - 5 months p.s., 6 - 8 months p.s. and 9 - 11 months p.s. using high-dimensional mass cytometry by time of flight (CyTOF), whole-blood-based quantitative reverse transcription polymerase chain reaction (qRT-PCR) and bead-based immunoassays. The previously described control cohort was used as reference. Four weeks prior to surgery, patients were on a strict diet consisting of 40% carbohydrates, 25% protein and 35% fat as well as a daily minimum of 30 g dietary fibers. The aim of the diet was to reduce liver size making surgery more feasible and consequently minimizing complications.

EDTA-blood samples were collected one day prior to surgery and at the respective time-points within follow-up examinations at Ulm University Hospital and subsequently processed in the laboratory. Patients and controls were not in a controlled fasted state at blood draw.

The BMI in kg/m² describes the ratio of the person’s weight in kilograms to the squared height in meters. Body weight and body height were determined prior to surgery and at the respective time-points during follow-up examinations at Ulm University Hospital. Blood samples were taken routinely and analyzed in the Department of Clinical Chemistry at Ulm University Hospital (accredited according to DIN EN ISO 15189). Reference values were also obtained from the Department of Clinical Chemistry at Ulm University Hospital.

Peripheral blood mononuclear cells (PBMCs) were isolated from 3 mL EDTA-blood using Ficoll-Paque™ PLUS (Cytiva). Blood was mixed with equal volume of PBS (Gibco), layered over Ficoll and centrifuged for 20 min and 300 x g at room temperature (RT) with breaks off. PBMCs were aspirated and washed twice with 10 mL PBS and once with 1 mL staining buffer (1 x PBS, 1% BSA, 2 mM EDTA, 0.05% NaN₃) each time followed by a centrifugation step (300 x g, 8 min, RT). Staining of 1.0 x 10⁶ cells was performed in the dark for 30 min at 4°C using 100 µL staining buffer containing fluorescent labelled antibodies (dilution 1:100) specific to CD3 (VioBlue, Miltenyi Biotec, 130-114-519), CD4 (PE, Miltenyi Biotec, 130-113-225), CD8 (APC, Miltenyi Biotec, 130-110-679), CD19 (FITC, Miltenyi Biotec, 130-113-645), CD56 (PE-Vio 770, Miltenyi Biotec, 130-113-313), CD14 (APC-Vio 770, Miltenyi Biotec, 130-110-522), CD16 (VioGreen, Miltenyi Biotec, 130-113-397) and HLA-DR (PerCP-Vio 700, Miltenyi Biotec, 130-111-793). Cells were washed with 2 mL staining buffer, centrifuged (300 x g, 8 min, RT), resuspended in 500 µL staining buffer and subsequently acquired on a MACSQuant^® Analyzer 10 Flow Cytometer (Miltenyi Biotech). Compensation was performed using instrument-specific automated compensation with single-stained compensation beads. To exclude background staining and unspecific binding of antibodies, isotype controls have been measured. All samples were acquired with the same voltage settings. Analysis of data was performed using Flowlogic™ 8.4 software (Inivai Technologies) according to the provided gating scheme ( Supplementary Figure 1 ). To analyze and exclude age-related effects on major immune cell subsets, patient cohort was subdivided into two age groups (20 – 45 years and 45 – 61 years) ( Supplementary Figure 2 ).

One mL of EDTA-blood was mixed with 1 mL SmartTube Proteomic Stabilizer PROT1 (SMART TUBE Inc.), incubated for 10 min at RT and stored at -80°C. Samples were thawed and erythrocyte lysis was performed using 1x Thaw-Lysis buffer (SMART TUBE Inc.) according to manufacturer’s specifications. After lysis, cells were resuspended in 2 mL CyFACS buffer (1 x PBS (Rockland, MB-008), 1% BSA, 2 mM EDTA, 0.05% NaN₃) and cell concentration was determined. A total number of 1.5 x 10⁶ cells per sample were used for the staining procedure. One mL CyFACS buffer was added, cells were centrifuged (600 x g, 8 min, RT) and supernatant was discarded. To prevent unspecific antibody staining, Fc receptor (FcR) block was performed by adding 3 µL of human FcR Blocking Reagent (Miltenyi) to the cells followed by an incubation for 20 min at RT. Master mix containing the antibodies listed in Supplementary Table 2 was prepared shortly prior to staining and filtered with 0.1 µM spin filter units (Merck).

One hundred µL of master mix was added to each sample and incubated for 30 min at 4°C. Titration of antibody panel was performed prior to the first experiments to determine the antibody concentration. Antibodies that were not purchased from Fluidigm were conjugated in-house using the respective Maxpar X8 Antibody Labeling Kit (Fluidigm) according to manufacturer’s specifications.

Cells were washed two times with 1 mL CyFACS buffer (600 x g, 8 min, RT). Fixation was performed by adding 1 mL of 3% methanol-free paraformaldehyde (PFA) (Thermo Scientific Pierce) in CyPBS followed by an incubation for 2 h at 4°C. Samples were washed twice with 1 mL CyPBS (600 x g, 8 min, RT). Cells were incubated with 0.5 mL of 0.2% Cell-ID Intercalator-Ir (Fluidigm) in Maxpar Fix and Perm Buffer (Fluidigm) for 20 min at RT followed by two washes with 2 mL CyPBS (600 x g, 8 min, RT). Cells were frozen in 1 mL cold freezing medium (10% DMSO in fetal calf serum) and stored at -80°C until day of acquisition.

On the day of acquisition, samples were thawed in a cold-water bath and washed once with 1 mL CyFACS containing Benzonase/Nuclease (Sigma Aldrich, 1:10,000), once with 1 mL CyFACS without Benzonase and three times with Maxpar Water (Fluidigm) each time centrifuged at 600 x g for 8 min at RT. Samples were adapted to a cell concentration of 10⁶ cells/mL and 300,000 cells were measured at 300 events/s on a Helios system (Fluidigm). EQ Four Element Calibration Beads (Fluidigm) were used for normalization over time.

Anchor samples consisting of buffy-coat derived PBMCs (purchased from Institute for Clinical Transfusion Medicine and Immunogenetics (IKT), Ulm, Germany) were adapted to a concentration of 1.0 x 10⁶ cells/mL. One mL aliquots were mixed with 1 mL SmartTube Proteomic Stabilizer PROT1 (SMART TUBE Inc.) and frozen at -80°C. Anchor samples were treated the same way as patient samples and one anchor was included in each batch of measured patient samples.

Generated raw FCS files were preprocessed prior to normalization ( Supplementary Figure 2A ). Calibration beads were removed, and DNA positive cells were identified (¹⁹¹Ir and ¹⁹³Ir). Doublets were excluded using gaussian parameters event length and residual (24). Batch normalization was performed using R-based CyTOF Batch Adjust workflow (25) adapted for Windows system. Channel specific batch-to-batch variation was evaluated using anchor samples and adjustment factors were transferred to patient samples ( Supplementary Figure 2B ).

After batch normalization, FCS files were uploaded to Cytobank (Beckman Coulter), transformed (arcsinh, co-factor = 5) and analyzed using manual gating as well as machine learning algorithms including dimensionality reduction via Uniform Manifold Approximation and Projection (UMAP) (26) and clustering via Self-Organizing Map (SOM) algorithm FlowSOM (27). Twenty thousand cells per sample were included in the automated analysis. For the illustration of the UMAP density-plots, FCS files were exported and concatenated according to their corresponding group using the CATALYSTLite online application (28).

In order to analyze populations of interest at a higher resolution, manually gated populations were exported from Cytobank as raw value FCS files. Validation of the gating was performed by comparing percentages of the manually gated populations with the percentages of populations identified with FlowSOM using linear regression analysis ( Supplementary Figure 3 ). For monocytes, manual gating seemed to be prone to slight underestimation of the population. This might be due to high heterogeneity within the monocyte population that could be better depicted with the algorithm (29). Mean frequencies of manually gated CD45⁺ cells can be found in the supplement ( Supplementary Table 3 ).

For the detailed analysis of specific cell populations the Spectre R package was used (30) with instructions and source code provided at https://github.com/ImmuneDynamics/spectre. The validated population of interest was exported from Cytobank as raw value FCS files. The arcsinh transformed (co-factor 5) dataset was merged into a single data.table, with keywords denoting the sample, group, and other factors added to each row (cell). The FlowSOM algorithm (27) was then run on the merged dataset to cluster the data, where every cell is assigned to a specific cluster and metacluster. Subsequently, the data was randomly downsampled to 10,000 cells per group and analyzed by the dimensionality reduction algorithm UMAP (26) for cellular visualization. Annotated clusters were depicted in a heatmap as normalized abundances with respect to the mean of the CTRL population. Statistical analysis was performed as described in the corresponding chapter “statistical analysis”.

Marker expression of clusters was shown as volcano plots generated with an adapted version of the EnhancedVolcano R script (31). Marker expression of each analyzed surface marker is shown as log₂ fold change (FC) calculated based on the cluster-specific mean fluorescence intensities (MFIs) of the respective groups. log₂ FC (x-axis) is plotted against the statistical significance (y-axis) shown as -log₁₀ p. Statistical significance was calculated based on cluster-specific MFIs using an unpaired two-sided Wilcoxon test (α = 0.05).

A list of the immune cell populations that have been identified via mass cytometry as well as the corresponding surface markers that have been used for identification is shown in the supplement ( Supplementary Table 4 ).

The R scripts used for the analysis of the data are available as PDF documents in the supplement.

Two hundred µL of EDTA-blood were mixed with equal volumes of Monarch^® DNA/RNA Protection Reagent (New England BioLabs), vortexed and stored at -80°C. RNA isolation was performed according to manufacturer’s specifications. Samples were thawed at RT and 10 µL proteinase K were added. Samples were vortexed and incubated for 30 min at RT. Four hundred µL isopropanol were added, vortexed, transferred to an RNA purification column and centrifuged (16,000 x g, 30 sec). For removal of residual genomic DNA on-column DNAse I treatment was performed. Priming buffer (0.5 mL) was added followed by two wash steps with RNA Wash Buffer. RNA was eluted with 50 µL nuclease-free water. RNA concentration was determined using QIAxpert System.

One hundred and fifty ng total RNA were transcribed into complementary DNA (cDNA) using the AffinityScript Multiple Temperature cDNA Synthesis Kit (Agilent Technologies) with oligo(dT) primers according to manufacturer’s specifications. Samples were adjusted to a final concentration of 1 ng/µL with nuclease-free water. Prior to qRT-PCR, samples were tested for genomic DNA (gDNA) contamination using exon-specific primers (Forward: 5’-TCT GCC GTT TTC CGT AGG ACT CTC-3’, Reverse: 5’-CCC TGG ATG TGA CAG CTC CCC-3’) and cDNA functionality using intron-specific primers (Forward: 5’-GGC ATC CTC ACC CTG AAG TA-3’, Reverse: 5’- GTC AGG CAG CTC GTA GCT CT-3’) for ACTB in conventional PCR. After exclusion of gDNA contamination and confirmation of cDNA functionality, qRT-PCR was performed using QuantiTect SYBR Green PCR kit (Qiagen) and QuantiTect Primer assays specific for GAPDH (QT00273322), RPLP0 (QT00075012), LEPR (QT00006524), ADIPOR1 (QT00002352), GIPR (QT00033138) and GHRL (QT00041377). GAPDH and RPLP0 were validated and used as housekeeping genes ( Supplementary Figure 4 ). Melting curves of PCR products were evaluated and should have been between 72°C to 86°C according to manufacturer’s specifications. Calculated ΔCT values of patients were normalized to the mean ΔCT value of control population resulting in depicted x-fold values.

EDTA-blood was centrifuged at 2,600 x g for 15 min and plasma was stored at -80°C. Samples were thawed and analyzed in duplicates using the LEGENDplex Human Metabolic Panel with a V-bottom plate (BioLegend). The measured analytes included adiponectin (beads A4), adipsin (beads A5), leptin (beads B4) and resistin (beads B7). Staining and set-up were performed according to the manufacturer’s specifications. Data was acquired on MACSQuant^® Analyzer 10 Flow Cytometer (Miltenyi Biotech) and analysis was performed with the Legendplex Cloud-based software (BioLegend, Version 2022-07-15).

Enzyme-linked Immunosorbent Assays (ELISAs) were performed with plasma samples using commercially available ELISA kits for Ghrelin (ThermoFisher, # BMS2192) and GIP (Merck, EZHGIP-54K) according to the manufacturer’s specifications. Absorption was determined at 450 nm in Tecan Spark^® Multimode Reader.

GraphPad Prism 7.04 (GraphPad Software, Inc.) was used for statistical analysis. The specific tests performed are described below each graph. When comparing two groups, a two-tailed Student’s t test (α = 0.05) was performed if requirements of normal distribution (Shapiro-Wilk test) and homogeneity of variances (F-test) were given. Otherwise, a Welch’s test or a Mann-Whitney test were performed. When comparing more than two groups, a one-way ANOVA with Geisser-Greenhouse correction followed by an uncorrected Fisher’s LSD test (α = 0.05) was performed when normal distribution (Shapiro-Wilk test) was given. Otherwise, a Kruskal-Wallis test with Geisser-Greenhouse correction followed by an uncorrected Dunn’s test was performed. The specific method is stated in the figure legend. For each experiment, the sample sizes are depicted in the figure or in the figure legend. The following indicators were used for all statistical tests: * indicates p < 0.05, ** indicates p < 0.01, *** indicates p < 0.001, and **** indicates p < 0.0001. Data is depicted either as boxplots showing full range or as heatmaps showing either row z-score normalized data or min-max scaled data (surface marker expression), unless otherwise stated.

The health status of 38 female patients with morbid obesity who underwent bariatric surgery was assessed by determining pre-existing comorbidities, blood pressure, medication, as well as several blood parameters ( Table 1 ). The median BMI of the cohort was 48.9 kg/m². The predominant procedure performed was sleeve gastrectomy (79%) followed by gastric bypass (21%). Patients with morbid obesity frequently suffered from diabetes and its preliminary stages (66%) and hyperlipoproteinemia (39%), which is also reflected in the blood values. Considering the measured blood parameters, patients showed typical obesity-associated signs of a low-grade chronic inflammation characterized by significantly elevated levels of both circulating C-reactive protein (CRP) (median= 7.7 mg/L) and circulating fibrinogen (median = 3.7 g/L) (32). The blood fat values for cholesterol (11 patients ≥ 5 mmol/L), triglycerides (13 patients ≥ 1.7), and low-density lipoprotein (LDL) (12 patients ≥ 3 mmol/L) were shifted towards the upper limit of normal levels, whereas the value of high-density lipoprotein (HDL) (11 patients ≤ 1.2 mmol/L) was shifted towards the lower limit of normal levels.

Apart from clinical blood parameters, PBMCs were characterized using flow cytometry. A control cohort (CTRL) consisting of 10 healthy, lean (BMI ≤ 25 kg/m²), age- and gender-matched volunteers ( Figures 1A, B ) was established serving as a reference for the patient group with morbid obesity (Baseline). T cells (CD3⁺) ( Figure 1C ) including CD4+ T helper cells ( Figure 1D ) and CD8+ cytotoxic T cells ( Figure 1E ) as well as NK cells (Lin^-, CD56⁺) ( Figure 1G ) tend to be slightly decreased in patients with morbid obesity. However, statistically these shifts were not significant due to high variation within the cohorts. B cells (CD19⁺) ( Figure 1F ) and monocytes (HLA-DR⁺, CD14⁺, CD16^var) ( Figure 1I ) tend to be increased in patients with morbid obesity. Within the monocyte compartment the HLA-DR^-, CD14⁺ cells ( Figure 1H ), referred to as immunosuppressive myeloid-derived suppressor cells (MDSCs) (15, 33), and the classical monocytes (CM) ( Figure 1L ) were significantly increased in patients with morbid obesity, whereas no difference was observed for intermediate monocytes (IM) ( Figure 1K ) and non-classical monocytes (NCM) ( Figure 1J ).

To investigate the influence of bariatric surgery on the health status of patients with morbid obesity a subcohort consisting of 12 female patients with morbid obesity undergoing gastric sleeve or gastric bypass surgery was established ( Figure 2A ). Blood samples were taken prior to surgery (Baseline) and within aftercare checkups 1 - 2 months post-surgery (p.s.), 3 - 5 months p.s., 6 - 8 months p.s., and 9 – 11 months p.s. as illustrated in Figure 2A . The previously described gender- and age-matched control cohort (CTRL) was used as a reference for normal BMI ( Figure 2B ), adipocytokine levels, and immune cell levels determined via mass cytometry. For the time progression of weight loss ( Figure 2B ) and clinical blood parameters ( Figure 2C ) values were compared to the baseline.

Compared to baseline, bariatric surgery resulted in significant weight loss ( Figure 2B ) after only 1 - 2 months p.s.. Additionally, compared to baseline, inflammatory blood parameters like CRP significantly decreased within the first 3 – 5 months p.s. and the concentration of circulating leukocytes significantly decreased as soon as 1 – 2 months p.s. ( Figure 2C ), whereas fibrinogen tends to be non-significantly decline 9 – 11 months p.s. Furthermore, metabolic parameters including triglyceride levels as well as the insulin levels significantly declined after surgery, reaching statistical significance 3 – 5 months p.s. (triglycerides) or rather 6 – 8 months p.s. (insulin) compared to baseline. Whereas LDL levels seemed to be less affected by bariatric surgery, HDL concentration slightly elevated 9 – 11 months p.s., without reaching statistical significance compared to baseline ( Figure 2C ).

In order to assess the impact of the bariatric surgery on plasma levels of the adipocytokines and gastrointestinal hormones, bead-based immunoassays, and ELISAs were used to investigate adiponectin, adipsin, leptin and resistin, as well as ghrelin and GIP ( Figure 3A ). Moreover, the gene expression levels of the respective adipocytokine receptors ADIPOR1 and LEPR were analyzed in whole blood samples ( Figure 3B ). A healthy, lean (BMI ≤ 25 kg/m²), gender- and age-matched control cohort (CTRL, n = 10) ( Figure 2B ) was used to determine reference levels. Regarding the adipocytokine levels, adiponectin tend to be non-significantly lower at baseline compared to the CTRL but normalized after 1 – 2 months p.s.. However, adipsin did not show any alteration at baseline compared to the CTRL. Moreover, leptin was significantly elevated at baseline compared to the CTRL and continuously declined to reach CTRL levels at 2 – 5 months p.s.. Apart from that, resistin was slightly increased at baseline compared to the CTRL but did normalize within 9 – 11 months p.s. Interestingly, gene expression of the receptors ADIPOR1 and LEPR ( Figure 3B ) seemed to be regulated in an opposite manner regarding the respective protein levels. The gene expression of ADIPOR1 seemed to be slightly, but non-significantly, upregulated in the peripheral blood immune cells of patients with morbid obesity compared to the CTRL but reached CTRL levels at 3 – 5 months p.s.. Similarly, LEPR expression level tended to be slightly, but non-significantly, decreased in the peripheral blood immune cells of patients with morbid obesity compared the CTRL but showed significantly increased levels 1 – 2 months p.s., reaching CTRL levels 9 – 11 months p.s. Ghrelin levels were already slightly, non-significantly decreased at baseline compared to the CTRL and further decreased after surgery not showing any sign of rising to CTRL levels within the observed period. Apart from that, GIP levels seemed to be slightly, but non-significantly, decreased in patients with morbid obesity compared to the CTRL and were not affected by bariatric surgery.

High-dimensional mass cytometry was used to analyze peripheral blood immune cell populations in 12 female patients with morbid obesity undergoing bariatric surgery ( Figure 2A ). A healthy, lean (BMI ≤ 25 kg/m²), gender- and age-matched control cohort (CTRL, n = 10) ( Figure 2B ) was used to determine reference levels. Major immune cell populations were detected using dimensionality reduction via UMAP and clustering with FlowSOM algorithm ( Figure 4A ). Clusters were manually annotated according to the expression patterns of surface markers CD45, CD66b, CD3, CD4, CD8a, NKG2D, CD19, CD33, CD56, HLA-DR, CD16, CD14, CD64, CD11b, CD11c, CD123 ( Figure 4B ). Illustration of UMAP density plots revealed shifts in the major immune cell populations of patients with morbid obesity compared to the CTRL cohort conspicuously visible in CD4+ T cells, granulocytes and monocytes ( Figure 4C ). Additionally, abundances of the detected immune cell populations normalized to the respective mean of the CTRL population were determined ( Figure 4D ). Using mass cytometry, NK cells (Lin-, CD16⁺, CD56⁺, NKG2D⁺) were shown to be significantly decreased in the condition of morbid obesity compared to the CTRL population. Moreover, bariatric surgery led to a further decline of the NK cell population which did not recover within 9 – 11 months p.s. ( Figure 4E ). Cross verification was performed using conventional flow cytometry to confirm the results obtained from mass cytometry ( Figure 4E ).

Due to the constant decrease of NK cell levels after surgery and the visible shifts in CD4+ T cells and monocytes, that were also detectable by flow cytometry, these immune cell populations were further investigated using the Spectre R package.

The shift observed in the CD4+ T cells was analyzed at a higher resolution using the Spectre R package (30). Prior to the analysis, manually gated CD4+ T cells were compared to the CD4+ T cell population identified with unsupervised clustering using FlowSOM ( Figure 4A ) to verify the defined populations ( Supplementary Figure 3 ). Linear regression analysis (Slope = 0.9584, R² = 0.9915) showed a high conformity of manual and automated gating. Accordingly, manually gated CD4+ T cells were selected for further analysis and clustered using FlowSOM followed by visualization with UMAP ( Figure 5A ) using the markers CD45, CD3, CD4, CD45RA, CD197, CD27, CD127, CD95, CD62L and CD28. Seven clusters were defined according to different surface marker expression ( Figures 5B–D ). Cluster 1 identified CD45RA⁺, CD62L⁺, CD27⁺ and CD197⁺ naïve T cells (34). Furthermore, cluster 2, 3, 4 and 5 identified CD45RA^-, CD62L⁺, CD27⁺ central memory T cells, whereas cluster 6 and 7 identified CD45RA^-, CD62L^- and CD27^- effector memory T cells (35, 36). The time-progression of the density plots ( Figure 5E ) revealed shifts in the mentioned CD4+ T cell subpopulations when comparing CTRL and baseline. Naïve T cell levels (Cluster 1) tend to be slightly, but non-significantly, decreased in patients with morbid obesity compared to the CTRL population and further declined after surgery reaching significantly decreased levels 1 – 2 months p.s. ( Figure 5C ). In contrast, CD127^high central memory (Cluster 2) and CD127^high effector memory (Cluster 7) T cell subsets were significantly expanded in patients with morbid obesity. Interestingly, 1 – 2 months p.s. both subsets already decreased, reaching a minimum at 6 – 8 months p.s., also detectable in the corresponding density plots. Nevertheless, 9 – 11 months p.s., a disturbance of the CD4+ T cell compartment was still detectable characterized by decreased naïve T cells and increased effector memory subsets (Cluster 6 and 7).

According to the resulting volcano plots ( Figure 5F ), an elevated expression of CD127 could be observed in several distinct clusters in patients with morbid obesity, even 9 – 11 months p.s., hinting towards a possible obesity-associated dysfunction within the regulation of this receptor.

As the ratio of NK cells significantly decreased after bariatric surgery, the NK cell population was further analyzed. Comparable to the workflow described for CD4+ T cell, manually gated NK cells were compared to the NK cells identified via unsupervised clustering using FlowSOM ( Supplementary Figure 3 ). Manually gated NK cells were clustered using FlowSOM followed by visualization with UMAP ( Figure 6A ) using the markers CD45, CD56, CD62L, CD183, CD27, NKG2D, and CD16. Three distinct clusters were defined according to their surface marker expression ( Figures 6B–D ). Cluster 1 defined the CD56^high, CD16^low cytokine-producing NK cell subset, whereas cluster 2 and 3 defined the CD56^dim, CD16⁺ cytolytic NK cell subset (37, 38). Apart from that, the CD56^high NK cell subset was described to express CD62L and CD27 (39). The cytotoxic NK cell subset could be further distinguished by the expression of NKG2D (cluster 3) (40), which is known to be an activating cell surface receptor (41). Regarding the density plot ( Figure 6E ), a considerable disturbance could not be observed in the NK cells department. The normalized abundances of the mentioned clusters ( Figure 6C ) revealed a non-significant shift towards CD56^dim, CD16⁺, NKG2D⁺ in patients with morbid obesity, which normalized rapidly after bariatric surgery. Additionally, the expression levels of NKG2D, CD11c, and CD223, also known as lymphocyte activation gene-3 (LAG-3), were significantly increased in the CD56^dim, CD16⁺, NKG2D⁺ subset ( Figure 6F ) from patients with morbid obesity compared to the CTRL population. Within 9 – 11 months p.s., the expression of CD223 normalized in the CD56^dim, CD16⁺ NKG2D⁺, whereas the expression of NKG2D was still slightly elevated.

The last subset that was investigated more specifically was the monocyte compartment. As previously described, manually gated monocytes were compared to the monocyte population determined via unsupervised clustering ( Supplementary Figure 3 ). The manually gated monocytes were clustered into five distinct clusters based on surface marker expression of CD33, CD64, CD62L, CD11b, CD14, CD16, HLA-DR, CD45, and CD183 ( Figures 7A–D ). Cluster 1 and 2 defined monocytes with lower HLA-DR expression but higher expression of CD62L. CD62L is a recruitment marker highly expressed on the classical monocyte subsets (42). Compared to cluster 2, cluster 1 showed higher CD183, CD11b, and CD33 expression and was significantly increased in patients with morbid obesity compared to the CTRL. Moreover, CD183, also known as CXCR3, is a chemokine receptor involved in the migration of monocytes into inflamed tissue (42, 43). Cluster 1, representing a CD183⁺ CD62L⁺ monocyte subset, was significantly enlarged in patients with morbid obesity but slowly diminished after bariatric surgery ( Figure 7C ). Besides that, cluster 3 (CD16⁺) and cluster 5 representing low expression of CD14 and CD62L, as well as high expression of HLA-DR were significantly decreased in patients with morbid obesity most likely due to the observed shift towards cluster 1 ( Figure 7E ). However, these cells did not return to CTRL levels after surgery. Moreover, a significantly decreased expression of the surface marker HLA-DR was observed on cluster 1 and 2 ( Figure 7F ). Importantly, the significant decrease in HLA-DR expression on the monocytes of patients with morbid obesity was also detectable in the flow cytometry data (data not shown). After 9 – 11 months p.s., cluster 1 kept the low HLA-DR expression level. Nevertheless, a decreased expression of TREM1, an activating receptor of monocytes that is involved in mediating inflammation (44), was observed in cluster 1 and 5 at 9 – 11 months p.s.