Creation of a 5-HT-producing cell line
B16F0 mouse melanoma cells from American Type Culture Collection (ATCC, USA) were genetically modified to produce 5-HT. The B16F0 line was selected for three key characteristics: (1) its syngeneic nature with C57BL/6 mice, enabling studies in immunocompetent animals, (2) its production of melanin pigments, allowing macroscopic tracking of potential tumor dissemination to the lungs (which could indicate extension to left heart valves), and (3) its established use in tumor growth studies30,31,32. The design, synthesis of the transcript sequence, and generation of the knock-in cell lines were performed by Thermo Fisher Scientific (Carlsbad, USA) using a lentiviral polycistronic vector. The resulting genetically modified murine cell line is referred to as B16F0-Tph1.
The genetic construct (Fig. 1a) was engineered for optimal expression and monitoring capabilities. A CMV promoter was chosen to drive strong, constitutive expression in mammalian cells. The construct contained three key elements in a single reading frame: the TPH1 gene for 5-HT synthesis, a GFP reporter gene for real-time expression monitoring, and a blasticidin resistance gene for stable selection. The expression vector was designed with a P2A self-cleaving peptide sequence between GFP and TPH1 to ensure stoichiometric production of independent proteins from a single transcript, while maintaining stable selection through an independently-promoted blasticidin resistance gene33.
Validation of successful genetic modification was performed through multiple complementary approaches. GFP expression was quantified by flow cytometry and confirmed through fluorescence microscopy (Fig. 1b). Functional TPH1 expression was verified by measuring 5-HT production in culture supernatants using HPLC. Comparison between B16F0-Tph1 cells and non-transfected controls revealed significant differences in 5-HT production levels. Based on observed decreases in 5-HT production in later passages, we used cells only between passages 1–2 for all experiments.
Model development and optimization
Two systematic pilot experiments were conducted to optimize the protocol. The first pilot evaluated tumor burden and survival with varying cell numbers (500–1000–2500–10,000 cells, n = 2 per group). Mice receiving ≥ 2500 cells (n = 4) showed extensive peritoneal carcinomatosis and were euthanized at 4 weeks with intraperitoneal injection of Euthasol 140 mg/kg and death was confirmed by cervical dislocation.
The second pilot determined optimal exposure duration using 1000 cells (n = 3 per group). The melanin pigmentation of B16F0 cells enabled clear macroscopic identification of tumor spread. In this experiment, tumor progression was analyzed as a function of postoperative time: mice operated with the same number of cells were sacrificed sequentially weekly at 5, 6, and 7 weeks postoperatively to establish the kinetics of dissemination of B16F0-Tph1 cells in vivo. In the same experiment, 5-HT was assessed in blood taken from the right ventricle or retro-orbitally (systemic blood) in mice injected with B16F0-Tph1 cells (serotonin-producing melanoma) compared to mice injected with non-modified B16F0 cells (standard melanoma) to verify that the right heart is exposed exclusively to elevated 5-HT compared to systemic circulation in the treated group.
For the main experiment, we selected 1000 cells with 5-week exposure as optimal parameters. Three experimental groups were established: mice injected with B16F0-Tph1 cells (serotonin-producing melanoma), mice injected with non-modified B16F0 cells (standard melanoma), and sham-operated control mice. For the main comparative analyses, the study included 7 B16F0-Tph1 mice and 7 sham-operated controls, with mice randomized across housing units to prevent cage-specific effects. Individual identification was maintained through uniquely numbered ear tags.
The male mice, aged 9 weeks, were sourced from JANVIER LABS and were of the same genetic background as the B16F0 melanocytic cell line. They were housed for 1 week in the animal facility of INSERM, LVTS1148, Bichat and had the same body weight at the start of the experiment (21.1 ± 3.07 g in B16F0-Tph1 vs 22.6 ± 3.3 g in sham mice). Following anesthesia (ketamine 100 mg/kg, xylazine 20 mg/kg), a right subcostal incision exposed the liver. Cell suspension (2 µL PBS) was precisely delivered using a Hamilton® fixed-needle syringe (10 µL, 51 mm, 23 s gauge) at 45° angle to 2 mm depth. To prevent immediate tumor dissemination and ensure hemostasis, Surgicel Fibrillar hemostat (3 × 3 mm) was applied at the injection site. Post-operative care included buprenorphine analgesia (0.1 mg/kg, subcutaneous) and daily monitoring for signs of distress. Blinding was used for each step of the experimental process, and all samples were coded prior to analysis so that the treatment group could not be identified before analysis was completed.
Ethics
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and in accordance with ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines. The protocol was approved by the French Ministry of Education and Research (APAFIS no. 35025) and complies with European Union regulations. Sample size was determined by power calculation (power = 0.8, alpha = 0.05) which indicated a minimum requirement of 7 animals per group. Based on a 40% attrition rate observed in pilot studies, we initially included 10 animals per group to ensure adequate statistical power for the final analysis. All surgery was performed by the same operator, under anesthesia, and all efforts were made to minimize suffering. The humane endpoint criteria established for our experiments were as follows: inactivity, social isolation, arched posture, piloerection, aggressiveness, hind-limb paralysis, weight loss > 20%, skin lesions covering more than 10% of the body, or ascites detected on palpation. Animals exhibiting four or more simultaneous humane endpoint criteria, or any humane endpoint criterion for three consecutive days, were euthanized. Animals presenting fewer than four criteria received immediate analgesic treatment with buprenorphine (0.1 mg/kg, subcutaneously). If humane endpoints were reached, euthanasia was performed the same day. At the conclusion of the experiment, mice were euthanized under deep anesthesia using ketamine (150 mg/kg) and xylazine (30 mg/kg) administered intraperitoneally, followed by cervical dislocation to ensure humane endpoints. All procedures were approved by the institutional animal care committee of INSERM, LVTS1148, Bichat. GS has completed regulatory training in animal experimentation as a project designer “Concepteurs de projets expérimentaux utilisant des animaux vivants” at the Paris-Sorbonne University.
5-HT Measurement
In our murine model, we measured both urinary 5-HT and 5-HIAA to comprehensively characterize serotonin metabolism, as the mouse model may have different metabolic patterns compared to human patients. Twenty-four hours before planned termination, mice were placed in individual metabolic cages for urine collection. Urine samples were immediately stored at − 20 °C. The analysis was performed at the Biochemistry and Molecular Biology Laboratory of Lariboisière Hospital (Paris, France). Prior to analysis, samples were thawed, centrifuged (2000 g, 10 min, 4 °C), and acidified to pH 2–3. 5-HT and its metabolite 5-HIAA were quantified using High-Performance Liquid Chromatography (HPLC) with electrochemical detection. Chromatographic separation was performed on an Ultimate 3000 HPLC system using a reversed-phase C18 column (150 × 4.6 mm, 5 μm particle size) maintained at 30 °C. The mobile phase consisted of phosphate buffer and methanol (85:15 v/v) at a flow rate of 1.0 mL/min. External standards (50–5000 nmol/L) were used for calibration, and urinary results were normalized to creatinine concentration. All measurements were performed in triplicate to ensure analytical reliability.
The same analytical approach was used to measure 5-HT in cell culture supernatants during the in vitro validation phase as well as in the blood samples collected and processed using identical protocols, except for the creatinine normalization step.
High-resolution episcopic microscopy
Hearts were arrested in diastole by one-minute immersion in saturated potassium chloride solution and fixed in 4% paraformaldehyde for 24 h. Due to heart length exceeding the standard 5 mm depth resin mold, the apex was removed to focus analysis on the mid-ventricle region and the base containing all four valves. Following PBS washing, samples underwent dehydration through graded ethanol series (30%, 50%, 70%, 90%, 100%, 1 h each) and were embedded in JB-4 resin containing eosin fluorescent dye.
HREM imaging was performed in the Small Animal Histology and Morphology Platform (Necker). Sequential sectioning (1.75 µm) and imaging utilized systematic block-face photography under brightfield. The comprehensive image processing workflow began with initial acquisition of brightfield images at 2048 × 2048 pixels averaging 3 captures per slice for approximately 2000 sections per heart. Image processing continued with registration using ImageJ’s StackReg plugin, followed by background subtraction with a 50-pixel rolling ball algorithm, contrast enhancement at 0.3% saturation, and 3D median filtering with a 2 × 2 × 2 kernel. Final deconvolution in Imaris employed an iterative constrained Richardson-Lucy algorithm with 10 iterations, a signal-to-noise ratio of 20, automatic background correction, and a quality threshold of 0.05.
Morphometric analysis
Standardized valve measurements were performed using 3D Slicer software (version 4.11) on HREM reconstructions. Three-dimensional reconstruction followed a rigorous protocol beginning with data import and pre-processing of HREM image stacks (2000 sections/heart) into 3D Slicer. Semi-automatic segmentation using threshold-based algorithms was performed, with manual refinement at valve boundaries at a resolution of 10.4 µm (X,Y) × 1.75 µm (Z).
Multi-plane visualization in the Three-dimensional reconstruction of each cardiac valve was illustrated in (Supplementary Figures 2–6) enabled measurement of valve volumes as well as thickness measurements across each cusp/leaflet, from base to free border. Cardiac valve volumes were measured using images taken at 10.4 µm intervals in the X–Y plane and 1.75 µm along the Z axis. Using 3D Slicer 5.8.1 software, we defined regions of interest around each valve by examining the entire deconvoluted heart dataset. We ensured comprehensive inclusion of the target valve through visual inspection and manual addition/exclusion of relevant valve structures. Before quantification, we normalized image intensities across all specimens through histogram matching. The volume measurement involved applying optimized color thresholds to the light channel, followed by voxel counting within each valve region slice-by-slice. The Fill Between Slices tool provided interpolation between manually segmented areas, resulting in volumetric data. Data are expressed as mm3. An example of the output is shown in (Supplementary Figure 7).
The thickness measurements included 18 measurements (6 equidistant points from free edge to base for each leaflet/cusp in tricuspid and semilunar valves), and 12 measurements (6 equidistant points per leaflet) in the mitral valve. For measurement validation, inter-observer reliability was established between two independent observers with an intraclass correlation coefficient exceeding 0.90, supported by Bland–Altman analysis. Intra-observer reproducibility was confirmed through repeated measurements at a 1-week interval, yielding a coefficient of variation below 5%. All measurements were taken perpendicular to each leaflet’s long axis (Supplementary Figure 7, an example of measurement points on tricuspid valve), with the first point positioned at valve free edge, the last point at leaflet insertion in the valvular annulus (base), intermediate points equally spaced, and all measurements taken at a 90° angle to leaflet surface. Additional qualitative assessment documented characteristic features of carcinoid heart disease.
For grayscale intensity profiles, valve tissue was assessed using Fiji/ImageJ software (NIH, version 2.9.0). Regions of interest (ROIs) were manually drawn around valve structures in HREM images with consistent selection parameters across all specimens. Intensity profiles were generated using the histogram function, with standardized brightness and contrast settings (window level: 127, window width: 255) applied uniformly across all specimens. Comparative histograms were created to directly visualize the distribution of grayscale intensity values between B16F0-Tph1 and sham control valves. We note that these intensity values reflect optical properties of the tissue and are not direct measures of physical density or ECM composition.
Valve leaflet length and Chordae tendineae dimensions (length and thickness) in the tricuspid valve measurements were performed using 3D Slicer software. The complete contour of each valve leaflet was manually traced from base to free edge, following the natural curvature of the structure. Scale calibration was applied based on the HREM system setting. Total excursion length was measured along the midline of each leaflet/cusp from the annular attachment point to the distal free edge, using consistent anatomical landmarks (ventricular insertion point to distal coaptation margin) across all specimens. Each valve was measured three times independently by two investigators blinded to the experimental groups, and the average measurements were used for analysis.
Statistical analysis
All statistical analyses and data visualizations were performed using GraphPad Prism (version 9.0, GraphPad Software, San Diego, CA) and JMP (version 15.0, SAS Institute, Cary, NC). After confirming non-normal distribution of valve thickness measurements (D’Agostino-Pearson test), we employed non-parametric methods for group comparisons.
Urinary concentrations of 5-HT and 5-HIAA were normalized to creatinine to account for variations in urine concentration, and these ratios were subsequently log-transformed to reduce data variability and approximate normal distribution. This transformation was particularly effective for the 5-HT/creatinine ratio, reducing the coefficient of variation from over 50% in raw values to approximately 10% in log-transformed data.
Between-group comparisons (Sham vs B16F0-Tph1, n = 7 per group) were conducted using the Kolmogorov–Smirnov test due to its sensitivity to both distribution shape and location differences, which was essential for capturing the heterogeneous pattern of valve remodelling in our model. Kruskal–Wallis tests were used to assess potential asymmetry in valve thickness distribution from free edge to base.
Linear regression analysis was performed to assess the relationship between log-transformed urinary 5-HT/creatinine ratios and right heart valve thickness. The strength of association was determined by the coefficient of determination (R2), and significance was assessed at p < 0.05. Threshold values for both log(5-HT/creatinine) and valve thickness were calculated using stepwise regression and profiler analysis to identify optimal cutpoints that discriminated between normal and pathological valve states. Diagnostic performance of these thresholds was evaluated by calculating sensitivity and specificity for detecting pathological valve thickening in both pulmonary and tricuspid valves.
For broader correlation analyses, Spearman’s rank correlation was employed to accommodate non-linear associations and potential outliers in the biomarker data. For the threshold effect analysis in the discussion, we employed segmental regression with a data-driven cutpoint at a log(5-HT/creatinine) value of approximately 6.3, based on scatterplot visualization revealing distinct relationship patterns above and below this value.
Data are presented as individual data points with median lines ± IQR to indicate central tendency. Statistical significance was set at p < 0.05, with p-values adjusted for multiple comparisons using the false discovery rate method to balance Type I and Type II error risks in this exploratory model characterization.