Modeling hepatocellular carcinoma and its microenvironment on a chip

TL;DR

Researchers developed a liver cancer model on a chip to study drug effects on tumors and their environment.

Contribution

The study introduces a patient-derived chip model of hepatocellular carcinoma to evaluate drug responses in tumor and microenvironment cells.

Findings

Sorafenib and lenvatinib reduced culture viability and altered vascular beds but did not affect tumor cells.

Atorvastatin reduced viability but did not impact vascular bed organization.

Tocilizumab, galunisertib, and vactosertib decreased IL6 levels, a key HCC prognostic marker.

Abstract

Hepatocellular carcinoma (HCC) is the most common type of liver cancer. Its incidence is increasing and is closely related to advanced liver disease. Interactions in the HCC microenvironment between tumor cells and the associated stroma actively regulate tumor initiation, progression, metastasis, and therapy response. Effective drug development increasingly requires advanced models that can be utilized in the earliest stages of compound and target discovery. Here we report a phenotypic screen on an advanced HCC patient-derived chip (PDChip) model. The vascularized HCC PDChip models include relevant cellular players of the HCC microenvironment. We assessed the effect of 28 treatment conditions on a panel of 8 primary HCC tumors and 2 cell lines. Approximately 1200 HCC PDchips were grown under perfusion flow, exposed to treatments, and subsequently assessed for viability, tumor-associated vasculature responses and chemokine and cytokine changes. Although the SoC therapeutics sorafenib and lenvatinib reduced culture viability and produced profound changes in the organization of the vascular beds, they did not affect the tumor cell population in these cultures. Atorvastatin, a 3-hydroxy-3-methylglutaryl-coenzyme A (HMG-CoA) reductase inhibitor, reduced PDChips viability but did not affect vascular bed organization. Sorafenib, lenvatinib and atorvastatin also affected chemokine and cytokine release. Tocilizumab, galunisertib, and vactosertib decreased the level of IL6, a relevant prognostic marker for HCC, while IL6 was increased by halofuginone. In conclusion, HCC PDChip models enabled a detailed evaluation of drug-induced responses in the tumor and associated microenvironment, highlighting their importance in preclinical research for understanding diseases and developing new drugs.

Figures

Generation and characterization of HCC PDChips.A Schematic representation of HCC tissue processing and PDChips generation using the OrganoPlate Graft. DTCs, CAFs, and HUVEC were mixed in fibrin and dispensed into the chamber of the OrganoPlate Graft microfluidic platform where they were cultured for 6 days, followed by exposure to a selection of compounds for 3 additional days. To assess compound responses, we evaluated cell viability, vascular network organization and chemokine and cytokine levels. B Phase contrast images of tumor compartment and lining vasculature on day 1 and 6 of HCC PDChi

HCC PDChip model viability in response to CAF targeting and SoC.A Scatterplot shows replicates reproducibility (*r* = 0.8236) of PDChips viability results. DMSO control viability distribution across plates is shown in (B). B Shows DMSO control viability percentage (values normalized to average of DMSO control viability/plate, *n* = 6). C Heatmap showing viability after treatment of the tumor compartment of HCC models. D Shows viability after treatment of the vascular compartment of HCC cultures. Heatmaps represent viability percentages (normalized to average of DMSO controls). Heatmaps include

Treatment-induced vascular response in HCC PDChips.A HCC PDChip models associated-vasculature organization in response to sorafenib and lenvatinib. HCC cultures were immunostained for CD31 (red), and confocal imaged. B Vasculature of HCC cultures was immunostained for CD31 and imaged using confocal microscopy (CD31 immunostaining for all cultures and treatment conditions can be found in Supplementary Fig. 2) (*n* = 30, for DMSO control; *n* = 10, for IgG1 control; *n* = 10, for each treatment condition). Abbreviation of drug names: Gal 0.1 µM and 1 µM (Galunisertib), Vac 0.1 µM and 1 µM (Vacto

Immunomodulatory response in HCC PDChip models.Supernatant of HCC cultures was collected and analyzed for expression of a panel of 13 cytokines and chemokines and IL6 by Luminex. A Significant changes in cytokine and chemokine production in HCC PDChip models in response to treatments. B Heatmap shows an overview of cytokine and chemokine responses of HCC cultures in response to treatments. (*n* = 48, for DMSO control; *n* = 16, for IgG1 control; *n* = 8–24, for treatment conditions). Statistical significance was determined using Wilcoxon tests, with *p*-values adjusted using the Benjamini–Hoch

Full text

Introduction

Hepatocellular carcinoma (HCC) is the third most common cause of cancer-related deaths worldwide and is a result of the malignant transformation of hepatocytes. HCC development is associated with viral disease, obesity, diabetes, and metabolic dysfunction-associated steatohepatitis (MASH), with ~90% of HCC developing on a background of cirrhosis [1, 2]. Until recently, sorafenib was the only first-line therapy approved by the Food and Drug Administration (FDA) for the treatment of advanced HCC [3]. Currently, several multikinase inhibitors such as first-line treatment lenvatinib, and second-line treatments regorafenib, cabozantinib, and an anti-VEGFR2 antibody, ramucirumab, have also received FDA approval for advanced HCC [4]. However, the median overall survival for patients undergoing treatment remains under 15 months [4, 5]. More recently, the FDA has approved nivolumab and pembrolizumab, and the combinations of nivolumab/ipilimumab, atezolizumab/bevacizumab, and tremelimumab/durvalumab, as second-line for sorafenib-pretreated HCC [6]. Nonetheless, tumor response rates for immune checkpoint inhibitors (ICIs) and combination therapies are reported to be only in the 8–20% range [7].Thus, a large unmet medical need remains for novel and effective (immuno)therapeutic strategies [8]. Overall, HCC remains a highly lethal malignancy for the 40% of affected patients diagnosed with advanced-stage disease.

The tumor microenvironment (TME) of HCC is complex and has been shown to play a crucial role in HCC disease progression and response to treatment. Among relevant elements of the HCC TME are endothelial cells, stromal cells including liver stellate cells and cancer-associated fibroblasts (CAFs), and the tumor immune infiltrate. Advances in the understanding of the HCC TME indicate that these components potentially regulate each other and together influence extracellular matrix (ECM) remodeling, metastasis, cancer stemness, and therapeutic resistance [9]. Commonly used HCC in vitro models such as cell lines, HCC-derived organoids, spheroids, and other 3D models, currently lack immune and supporting cells, only partially recapitulating disease complexity. Another important aspect, often overlooked by many in vitro platforms, is the ability to cultivate samples from different HCC patients thus assessing patient diversity and patient-specific responses [10].

Given the limitations of HCC models, Organ-on-a-Chip (OoC) technology will potentially contribute to narrowing the gap between the relevance of HCC in vitro models employed in preclinical studies and in vivo biology as it allows for the inclusion of several key TME parameters such as the presence of vasculature, perfusion, supporting cell types, and the resulting cellular interactions and tissue organization.

On-chip systems have been used to model various cancers, including breast [11], pancreatic [12–14], colorectal, and lung [15], providing advanced platforms for studying tumor cell dynamics and drug responses. Liver-focused on-chip systems have successfully simulated healthy and diseased states, replicating key features of liver tissue organization and metabolism. HCC-specific culture systems incorporating the HepG2 cell line have been used to model drug delivery [16] and bone metastasis [17]. More recently, patient-derived HCC on-chip has been employed to study the impact of oxygen concentration on tumor heterogeneity and drug responsiveness. This system has also been engrafted in mice, facilitating the growth of engrafted HCC tissue and generating multi-spot chip PDX models [18, 19].

Here, we advance the field of tumor modeling by combining a scalable, automation-compatible, microfluidic platform, the OrganoPlate, with HCC patient-derived tissue to simulate the cellular complexity of HCC and HCC patient heterogeneity in vitro. This comprehensive model setup, consisting of dissociated tumor tissue from HCC patients and HCC cell lines, combined with CAFs and vasculature, resulted in the generation of vascularized HCC on chip cultures able to produce the biomarker alpha-fetoprotein (AFP), and several chemokines and cytokines including IL6 and CCL2. HCC patient-derived chips (PDChips) were generated with samples from 8 different donors and 2 cell lines, and used to screen 28 different treatment conditions. The drug panel was composed of HCC SoC drugs, and a selection of drugs targeting several pathways in tumor cells as well as in the supporting cells, including endothelial cells and CAFs. Subsequently, all treatment responses were phenotypically characterized by assessing a combination of cell viability, vascular morphology and chemokine and cytokine production.

With this study, we demonstrate that the TME can be effectively modeled in a chip format, encompassing a range of cell types such as tumor and supporting cells, as well as important ECM components. Yielding robust cultures that can be screened across multiple donors and therapies. PDChip culture conditions facilitate relevant cellular interactions and organization, recapitulating disease complexity and patient diversity, allowing further insight into the targeting of HCC and its microenvironment. We believe that a broader implementation of tumor PDChip models will advance the development of novel and effective therapeutic alternatives.

Results

HCC PDChip models

To improve selection and evaluation of treatments targeting HCC, we developed an HCC PDChip model incorporating patient-derived tumor cells and components of the tumor microenvironment (TME) (Fig. 1A). This comprehensive model setup includes primary dissociated HCC cells or HCC cell lines, HCC-derived CAFs, and endothelial cells. HCC tumors (stages I and II), Huh7 and HLE cell lines used to generate the HCC PDChip models are described in Table 1. Cultures were generated in the OrganoPlate Graft platform, containing 64 chips in each Automation-compatible plate. Figure 1A shows the composition of the HCC PDChip model, the culture setup, and the readouts used in this study. HCC models were generated manually, whereas drug exposure, supernatant collection, addition of reagents, and culture fixation were automated and performed using a Biomek i5 pipetting robot (Beckman Coulter) and a MultiFlo FX non-contact dispenser (BioTek).Fig. 1. Generation and characterization of HCC PDChips.A Schematic representation of HCC tissue processing and PDChips generation using the OrganoPlate Graft. DTCs, CAFs, and HUVEC were mixed in fibrin and dispensed into the chamber of the OrganoPlate Graft microfluidic platform where they were cultured for 6 days, followed by exposure to a selection of compounds for 3 additional days. To assess compound responses, we evaluated cell viability, vascular network organization and chemokine and cytokine levels. B Phase contrast images of tumor compartment and lining vasculature on day 1 and 6 of HCC PDChips, these cultures were generated using samples from 8 primary HCC tumors and 2 HCC cell lines. B Also shows the bottom right quadrant of the tumor compartment of a HCC3 PDChip, images show that cells are poorly organized on day 1, while on day 6 culture shows tumor aggregates and an organized vascular network. C Confocal microscopy of HCC5 PDChip shows the presence of tumor cells (albumin, in green) and vasculature (VE-Cadherin, in red). D Confocal microscopy of a HCC6 PDChip shows the presence of CAFs and vascular network, vasculature is immunostained for CD31 (red), CAFs are immunostained for aSMA (green) and nuclei are labeled with Hoechst (blue). Overlay image shows that CAFs seem to line vascular structures. E Shows the presence of AFP, IL6 and CCL2 (ng/mL) in HCC patient-derived and HCC cell lines models, HCC patients (N = 8) and HCC cell lines (N = 2; Huh7 and HLE). AFP, IL6 and CCL2 were measured in the supernatant collected from DMSO controls for all donors and cell lines on day 9. Statistical significance was determined using unpaired Welch-corrected t-test to compare AFP, IL6, and CCL2 levels between HCC patient cultures and HCC cell lines. Adjusted p-values are expressed as follows: ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05.Table 1HCC patients information.PatientType of tumor and stageGenderRaceEthnicityAgeHCC1HCC, IMaleWhiteCaucasian67HCC2HCC, IMaleWhiteCaucasian73HCC3HCC,I/IIFemaleWhiteCaucasian67HCC4HCC, IIMaleWhiteNon-Hispanic/Latino70HCC5HCC, IIMaleWhiteNon-Hispanic/Latino69HCC6HCC, I-BMaleWhiteNon-Hispanic/Latino82HCC7HCC, IMaleWhiteNon-Hispanic/Latino63HCC8HCC, IIMaleWhiteNon-Hispanic/Latino69Cell linesHuh7MaleAsian/Japanese57HLEMaleAsian/Japanese68

Dissociated primary HCC cells or cell lines, CAFs and HUVECs were resuspended in fibrin, loaded into a chip, and allowed to self-organize generating the tumor compartment (Fig. 1A) (Bonanini et al., 2025). Side channels were loaded with endothelial cells, forming the vascular compartment (Fig. 1A). HCC cultures developed well-organized vascular structures over time (Fig. 1B, C, Supplementary Fig. 1). HCC tumor aggregates were visualized on phase contrast images on day 6 (Fig. 1B). In addition, different cellular compartments were identified through immunostaining (Fig. 1C). Interaction between tumor cell aggregates (albumin^+^) and vasculature (VE-Cadherin^+^) could be observed (Fig. 1C). CAFs (α-SMA^+^) were observed lining the tumor vasculature (CD31^+^) as well as side endothelial tubules (perfusion compartment) (Fig. 1D). Most HCC patient-derived cultures (6/8) presented levels of the HCC biomarker AFP > 200 ng/ml, and relevant secreted mediators such as IL6 and CCL2 were also present in these cultures. Very low levels of AFP were detected in the Huh7 and HLE cell line cultures.

HCC PDChips showed organized vasculature, and the presence of CAFs and tumor cells. AFP, IL6 and CCL2 levels in the supernatant differ across HCC patients. Next, cultures were exposed to several treatment conditions from day 6 to day 9.

HCC PDChips viability in response to a drug panel

HCC PDChips were exposed to a panel of 28 treatment conditions (Table 2), including the HCC SoC treatments sorafenib and lenvatinib, and a selection of single and combination treatments (Table 2). Cultures were exposed on day 6 for 72 h, after which supernatant was collected (Luminex analyses), cell viability evaluated (Alamar blue assay), and cultures were fixated for immunostaining.Table 2. Compounds selection.CompoundConcentrationTargetManufacturerReferencesControlsDMSO (Vehicle control)Sigma-Aldrich, D8418IgG1 10 μg/mL (mAb control)Selleck Chemicals, A2051Small molecule compoundsSorafenib1 and 10 μMMulti-TKI/SOCSelleck Chemicals, S7397[41]WZ8111 and 10 μMCXCR4 inhibitorSelleck ChemicalsS2912[42]UNBS516210 μMpan-antagonist of CXCL chemokine expressionSelleck Chemicals, S8869[43]Atorvastatin5 μMHMG-CoA reductase inhibitorSelleck Chemicals, S5715[44]Lenvatinib0.1 and1 μMPDGFR, VEGFR, FGFR inhibitorTargetMol, E7080[45]SH-4-540.1 and 1 μMSTAT3 blockerSelleck Chemicals, S7337[46]Crizotinib0.1 and 1 μMC-MET inhibitorSelleck Chemicals,S1068[47]Galunisertib0.1 and 1 μMTGFRBISelleck Chemicals, S2230[48]Vactosertib0.1 and 1 μMTGF-β receptor ALK4/ALK5Selleck Chemicals, S7530[49]LY20903140.1 and 1 μMGSK3 inhibitorSelleck Chemicals, S7063[50]Halofuginone0.1 μMInhibitor of collagen a1(I) and MMP-2 gene expressionMedChem Express, HY-N1584[34]Monoclonal antibodiesAtezolizumab5 μg/mL anti-PDL1Selleck Chemicals,A2004[51]Bevacizumab5 μg/mL anti-VEGF-ASelleck Chemicals,A2006[51]Tocilizumab5 μg/mLanti-IL-6RSelleck Chemicals,A2012[52]CombinationsAtezolizumabBevacizumab5 μg/mL5 μg/mLSelleck Chemicals,A2004Selleck Chemicals,A2006WZ811SH-4-541 μM0.1 μMSelleck ChemicalsS2912Selleck Chemicals, S7337UNBS5162 Halofuginone10 μM0.1 μMSelleck Chemicals, S8869MedChem Express, HY-N1584VactosertibSH-4-540.1 μM0.1 μMSelleck Chemicals, S7530Selleck Chemicals, S7337UNS5162SH-4-5410 μM0.1 μMSelleck Chemicals, S8869Selleck Chemicals, S7337WZ811LY20903141 μM0.1 μMSelleck ChemicalsS2912Selleck Chemicals, S7063

Viability assessment in the tumor and vascular compartments showed high reproducibility, indicating a comparable cell number and response between replicates (r = 0.8236) (Fig. 2A). Viability of DMSO controls was assessed to benchmark the distribution of the data per plate. To allow data analyses across different batches/plates, viability data was normalized to the average of the DMSO control condition. In Fig. 2B, DMSO control normalized data (normalized to average of DMSO control/plate) is shown.Fig. 2HCC PDChip model viability in response to CAF targeting and SoC.A Scatterplot shows replicates reproducibility (r = 0.8236) of PDChips viability results. DMSO control viability distribution across plates is shown in (B). B Shows DMSO control viability percentage (values normalized to average of DMSO control viability/plate, n = 6). C Heatmap showing viability after treatment of the tumor compartment of HCC models. D Shows viability after treatment of the vascular compartment of HCC cultures. Heatmaps represent viability percentages (normalized to average of DMSO controls). Heatmaps include n = 12 for DMSO control; n = 4 for IgG1 control; n = 4 for each treatment condition. For HCC4 n = 6 for DMSO control; n = 2 for IgG1 control; n = 2 for each treatment condition were included. E Representative confocal image of Albumin and Hoechst staining and respective segmentation are shown. Albumin and Hoechst stained cultures were imaged on an ImageXPress Micro Confocal XLS (Molecular Devices) and images were analyzed using INCarta image analysis software (Molecular Devices, version 2.1). F, G Show the quantification of albumin positive cells (tumor cells) in HCC PDChips in response to drug panel, (n = 24, for DMSO control; n = 8, for IgG1 control; n = 8, for each treatment condition).

Viability of the tumor compartment decreased in response to sorafenib 10 µM (71%), lenvatinib 1 µM (80%), atorvastatin 5 µM (47%), halofuginone 0.1 µM (72%) and UNBS5162 10 µM (72%) (Fig. 2C). These drugs also decreased the viability of the vascular compartment (Fig. 2D). After the Alamar blue viability assay, HCC PDChips were fixated and immunostaining for albumin, VE-Cadherin and CD31 was performed. Cultures were imaged on an ImageXPress Micro Confocal XLS (Molecular Devices) and images were analyzed using INCarta software (Molecular Devices). Tumor cell population in primary HCC-derived cultures showed a consistent distribution of albumin+ cells (Fig. 2E), compared to nearly absent levels in Huh7 cultures and a very inconsistent distribution in HLE cultures. Albumin+ cell population (cancer cells) was not significantly decreased in HCC patient-derived cultures in response to any of the treatment conditions (Fig. 2E, F), suggesting that targeting of the TME or cancer cells by our drug panel did not result in HCC cell death.

Combined, these data provide further insight into how the whole culture, the vasculature, and the tumor cell population respond to compounds and compound combinations, including SoC treatments. The effect of these compounds on the organization of the tumor-associated vasculature was further investigated.

Response of tumor-associated vasculature to treatments

Angiogenesis plays a significant role in HCC carcinogenesis and progression [20]. Therefore, it was crucial to assess the vascular response of HCC PDChips to treatments. To study the vascular response to the drug panel, CD31 immunostaining was performed, and cultures were imaged by confocal microscopy (Supplementary Fig. 2). Images of sorafenib and lenvatinib (HCC SoC) exposed cultures showed a clear effect of these drugs on the organization of the tumor-associated vasculature (Fig. 3). Morphological changes promoted by these drugs are in line with their anti-angiogenic effects. Next, images from CD31 stained cultures (Supplementary Fig. 2) were analyzed to assess parameters related to vasculature organization. First, a segmentation pipeline was applied to extract relevant parameters and evaluate compound-induced responses (Fig. 3A). Several descriptors associated with vasculature organization were assessed (Table 3). tSNE and PCA plots including data extracted from analyses of vasculature images showed that sorafenib and lenvatinib induced similar changes on evaluated descriptors (Fig. 3B, C). Assessment of tumor-associated vasculature confirmed morphological changes in line with sorafenib and lenvatinib anti-angiogenic effects. In Fig. 3E–H selected parameters influenced by these drugs are shown (total vessel density, total vessel area, branching index and total vessel length). Additionally, LY2090314 0.1 μM decreased vascular branching (Fig. 3F). Despite the clear effect of Atorvastatin on cell viability of tumor and vascular compartments, this drug did not affect tumor-associated vasculature organization (Fig. 3G).Fig. 3. Treatment-induced vascular response in HCC PDChips.A HCC PDChip models associated-vasculature organization in response to sorafenib and lenvatinib. HCC cultures were immunostained for CD31 (red), and confocal imaged. B Vasculature of HCC cultures was immunostained for CD31 and imaged using confocal microscopy (CD31 immunostaining for all cultures and treatment conditions can be found in Supplementary Fig. 2) (n = 30, for DMSO control; n = 10, for IgG1 control; n = 10, for each treatment condition). Abbreviation of drug names: Gal 0.1 µM and 1 µM (Galunisertib), Vac 0.1 µM and 1 µM (Vactosertib), LY 0.1 µM and 1 µM (LY2090314), WZ 1 µM and 10 µM (WZ811), Ator 5 µM (Atorvastatin), SH 0.1 µM and 1 µM (SH-4-54), Sora 0.1 µM and 1 µM (Sorafenib), Atez 5 µg/mL (Atezolizumab), Beva 5 µg/mL (Bevacizumab), Halo 0.1 µM (Halofuginone), UNB 10 µM (UNBS5162), Lenv 0.1 µM and 1 µM (Lenvatinib), Toci 5 µg/mL (Tocilizumab), Cri 0.1 µM and 1 µM (Crizotinib), Atez-Beva (Atezolizumab 5 µg/mL-Bevacizumab 5 µg/mL), Vac-SH (Vactosertib 0.1 µM -SH-4-54 0.1 µM), UNB-SH (UNBS5162 10 µM -SH-4-54 0.1 µM), WZ-SH (WZ811 1 µM - SH-4-54 0.1 µM), UNB-Halo (UNBS5162 10 µM -Halofuginone 0.1 µM), WZ-LY (WZ811 1 µM - LY2090314 0.1 µM). DMSO (Dimethyl sulfoxide 0.1%), and IgG1 (Immunoglobulin G1 10 µg/mL) were used as vehicle controls. Atezolizumab 5 µg/mL, Bevacizumab 5 µg/mL and Tocilizumab 5 µg/mL, also contains IgG1 5 µg/mL. Images were processed and 15 vascular organization-related descriptors were extracted. C Unsupervised linear (PCA), and D non-linear (t-SNE) reduction of the data show a clustering of sorafenib and lenvatinib. E–H Shows sorafenib, lenvatinib, LY2090314 and atorvastatin-induced changes in total vessel density, total vessel area, total vessel length and branching index. E–H include HCC patient-derived and HCC cell line culture results. Statistical significance was determined using either the Kruskal–Wallis test followed by Dunn’s post hoc test or ANOVA followed by Tukey’s post hoc test, depending on whether assumptions for parametric tests were met, with p-values adjusted using the Bonferroni correction. Adjusted p-values are expressed as follows: ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05.Table 3. Parameters and descriptors used for phenotypical assessment of tumor-associated vascular network.ParameterDescriptorExplant areaThe area of the convex hull of the vesselsThick/Thin/Total Vessel areaThe area of the vesselsThick/Thin/Total Vessel densityThe vessel area divided by the area of the center (area within the glass whole) of the tumor compartmentTotal number of junctionsNumber of branching points of the vessel skeletonBranching indexThe number of vessel junctions normalized by the vessel areaTotal vessel lengthThe sum of all vessel skeleton branches’ lengthAverage vessel lengthThe mean of all vessel skeleton branches’ lengthAverage vessel widthThe vessel area divided by the total vessel lengthThick – Total vessel area ratioRatio of thick and total vessel areaInside – Outside the center of the tumor compartment total vessel area ratioRatio of vessels located inside and outside the center of the tumor compartmentInside – Outside the center of the tumor compartment thick vessel area ratioRatio of thick vessels located inside and outside the center of the tumor compartment

HCC PDChips exhibit a very organized vascular network, which is only globally affected by SoC, sorafenib and lenvatinib. Next, cytokines and chemokines levels in HCC PDChips in response to the drug panel were analyzed.

Drug-induced immunomodulatory response in HCC PDChips

The HCC TME is highly complex, and the combinations of different cell types and resulting interactions have a significant influence on tumor progression, drug sensitivity, and the tumor immune landscape. Considering that supporting cells such as CAFs and endothelial cells are known to influence immune cell recruitment and function, we measured a panel of immunosuppressive (CXCL1, CXCL8, CXCL12, CCL2, CCL20, IL4, IL6) and immunostimulatory (IL21, TNF, CXCL10, CXCL11, CCL3 and CCL4) chemokines and cytokines, and the HCC biomarker AFP. Culture supernatant levels of these analytes were measured in response to the different treatments.

Supernatant analysis results show significant changes in multiple chemokines and cytokines (Figs. 2 and 4) induced by treatments in HCC patient cultures. A heatmap visualization (Fig. 4B) provides an overview of the effect of different treatments on chemokine and cytokine levels in HCC patient cultures. HCC biomarker AFP remained unaffected by treatments. Sorafenib exposure resulted in reduced levels of several chemokines and cytokines (CCL4, CXCL1, CXCL10, IL21, IL4 and TNFα) potentially promoting an immunosuppressive effect. A similar response was observed in the presence of lenvatinib, although CXCL12 was increased in response to lenvatinib 0.1uM. Atorvastatin induced a decrease in most tested analytes, this was likely attributable to its effect on culture viability.Fig. 4. Immunomodulatory response in HCC PDChip models.Supernatant of HCC cultures was collected and analyzed for expression of a panel of 13 cytokines and chemokines and IL6 by Luminex. A Significant changes in cytokine and chemokine production in HCC PDChip models in response to treatments. B Heatmap shows an overview of cytokine and chemokine responses of HCC cultures in response to treatments. (n = 48, for DMSO control; n = 16, for IgG1 control; n = 8–24, for treatment conditions). Statistical significance was determined using Wilcoxon tests, with p-values adjusted using the Benjamini–Hochberg correction. Adjusted p-values are expressed as follows: ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05.

IL6, which is a prognostic biomarker [21] in HCC, was significantly affected by several treatments, showing reduced levels in response to tocilizumab, galunisertib and vactosertib, whereas halofuginone alone or in combination with UNBS5162 led to increased IL6 levels. Halofuginone also promoted a decrease in IL4, IL21 and TNFα.

These data suggest that targeting of the supporting cell compartment influences relevant immune mediators, and that the use of such compounds could potentially impact patient anti-tumor immune responses.

Discussion

HCC remains challenging to treat, with a 5-year overall survival after diagnosis observed in only 18% of patients. This low survival rate is usually associated with late diagnosis, and limited treatment options [22]. Approved treatment options target different HCC tumor components, including the tumor, vascular, and immune compartments, indicating that the modulation of the TME is essential for effectively treating this tumor type [23]. Equally relevant is the variability across patients, which underscores the need for patient-derived models and for tailored therapeutic management of specific patient subsets [10, 24].

Several in vivo and in vitro models for HCC have been developed to understand HCC biology and drug sensitivities. While these models contribute in a relevant manner to the understanding of HCC biology, recapitulating HCC complexity and HCC patient diversity is still challenging and requires further efforts.

Here we report on an HCC PDChip setup that brings us a step closer to creating more comprehensive HCC models that are able to capture relevant HCC cellular self-organization and interactions. HCC patient-derived dissociated tumor tissue were cultured in presence of endothelial cells and CAFs. These cultures were generated on a scalable OoC platform and maintained for 9 days. Cellular interactions within the microfluidic platform led to vascularized tumor constructs, with HCC aggregates being enveloped and traversed by a lumenized and interconnected vascular plexus, in association with CAFs. Biomarkers AFP and IL6, as well as CCL2 were present in the supernatant of these tumor constructs (Fig. 1).

Vascularized HCC PDChips were exposed to a panel of treatment conditions that included SoC, single and combination treatments (Table 2). Drugs were selected for their ability to target tumor cells, endothelial cells and/or CAFs, and potentially modify cultures viability, organization as well as chemokine and cytokine landscape.

PDChips showed decreased viability in response to sorafenib 10 µM, lenvatinib 1 µM, halofuginone 0.1 µM, UNBS5162 10 µM and atorvastatin 5 µM; while WZ811 1 µM in combination with LY2090314 0.1 µM increased viability of HCC cultures (Fig. 2D). This effect remained significant when the whole group of HCC patient response was compared to controls (Fig. 4A, B). Quantification of tumor cells (albumin+ cell population) in these cultures suggested that, on average, this population remained unaffected by tested treatments as well as SoC (sorafenib and lenvatinib). Response to SoC seems to align with clinical data, sorafenib and lenvatinib have been shown in multiple trials to prolong median survival and the time to progression. However, most responders exhibit stable disease rather than significant tumor regression in response to these drugs [3, 25]. Atezolizumab plus bevacizumab did not affect HCC PDChip cultures viability, which could be a result of underrepresentation of immune cells in the DTC population or poor maintenance of this cell population under the culture conditions used. Maintenance of immune cells after tissue processing is challenging. Some immune cells may persist and could potentially be cultivated by adapting culture conditions. An additional strategy involves the inclusion of matched immune cells, which would enable evaluation of immunotherapeutics. This could be achieved through the expansion of tumor-infiltrating lymphocytes (TILs) or the inclusion of matched peripheral blood mononuclear cells (PBMCs). However, both approaches are limited by the availability of patient material and the success of TIL expansion, which is often variable and time consuming.

Next, we phenotypically characterized the associated vasculature of treated HCC tumor constructs. Cultures exposed to SoC, i.e. sorafenib and lenvatinib, were characterized by disorganization of the vascular beds compared to the control condition (Fig. 3). This response could potentially be associated with the decrease of tumor vascularity normally observed in response to these compounds in patients [23].LY2090314, a GSK3a/b inhibitor, specifically affected tumor vasculature branching.

As discussed previously, atorvastatin decreased HCC culture viability and did not result in abnormal vasculature. However, this drug did promote an increase in total vessel density, total vessel area, and total vessel length, suggesting a modulatory effect on vascular architecture. These findings align with the complex, dose- and context-dependent responses of endothelial cells to statins, which include anti-angiogenic effects, cytoprotection, and promotion of vessel maturation [26–29].

Culture supernatant was profiled using a Luminex panel. Treatments did not influence AFP levels (Fig. 4A). Lenvatinib and sorafenib in general negatively influenced chemokine and cytokine production (Fig. 4A, B). A similar response to sorafenib has also been observed previously in explant cultures [30]. Interestingly, IL6, which is an important mediator in HCC associated with a poorer response to sorafenib, regorafenib as well as atezolizumab plus bevacizumab [31–33], was decreased in HCC cultures supernatant in response to atorvastatin, galunisertib, vactosertib and tocilizumab. In contrast, halofuginone increased IL6 levels. In addition, this compound, known for its anti-fibrotic effects, decreased the levels of TNFα, IL4 and IL21. This was also observed when halofuginone was used in combination with UNBS5162. Halofuginone has demonstrated anti-fibrotic activity by reducing collagen I gene expression, regulating MMPs activity, increasing TIMP synthesis, and inhibiting TGFβ-driven signaling [34, 35]. Additionally, halofuginone inhibits prolyl-tRNA synthetase, activating the amino acid starvation response (AAR) pathway, leading to increased ATF4 expression [36]. ATF4 has been shown to link metabolic stress to increased IL6 expression [37], which may explain the observed rise in IL6 levels in HCC cultures treated with halofuginone.

In addition, atorvastatin treatment led to a decrease in several cytokines and chemokines, including IL6 (as discussed above) and CCL2 levels, which may indicate an effect on the cancer-associated fibroblast (CAF) population. However, the lack of reduction in AFP levels suggests a limited impact on the tumor cell population itself. This differential response highlights the importance of evaluating multiple components of the tumor microenvironment when assessing drug effects.

While patient-derived models are invaluable for cancer research due to their clinical relevance, they come with several limitations. Tumor heterogeneity poses a significant challenge, as variations between patients can limit standardization. Additionally, the complexity and cost of developing and maintaining these models are substantial. Scalability is another challenge, making it difficult to use these models for high-throughput screening or large-scale studies. Despite these challenges, patient-derived models remain a crucial tool, offering insights that are often more clinically relevant than traditional cell lines or animal models. However, detailed analyses of these tumor models enabled by the combination of several techniques seem essential to understanding model responses to different molecules as well as leveraging on the unmatched capabilities of on chip systems to recreate complex cellular interactions in vitro. In addition, the protocols presented in this study may support future clinical development and implementation of this state-of-the-art platform to create patient avatars and enable individualized therapy selection.

Altogether, our model setup enables testing of tumor, CAF and vascular targeting molecules, potentially also allows for the understanding of the role of tumor-vasculature or tumor-CAF interactions in tumor cell behavior and drug responses. HCC PDChip reported here has the potential to provide a platform that includes intra- and interpatient variability with sufficient scalability and ease of use for industrial and clinical implementation.

Materials and methods

Cells

Human umbilical vein endothelial cells (HUVEC, Lonza, C2519AS) were cultured in complete MV2, consisting of Endothelial Cell Growth Medium MV2 (PromoCell, C-22022) supplemented with 1% penicillin/streptomycin (Sigma Aldrich, P4333).

Cancer-associated fibroblasts (CAFs) from donor HCC3 were isolated in-house and cultured in a combination medium composed of 67% complete MV2 and 33% complete EMEM, consisting of EMEM (ATCC, 30-2003) supplemented with 10% fetal bovine serum (FBS, Gibco, 16140-071) and 1% penicillin and streptomycin solution (Sigma Aldrich, P4333).

Dissociated tissue cells (DTCs) from HCC tumor tissues HCC1 and HCC3 were obtained in collaboration with the Department of Surgery, Erasmus MC-University Medical Center, The Netherlands. Tumor biopsies collected during surgical removal of the tumor for curative intent, were kept on ice until use. METC approval (MEC2013-143) and written informed consent to use the biopsies for research purposes was provided by the patients. Samples were confirmed to be of tumor-origin with histopathological assessment. These were dissociated in-house following the protocol described below (Tissue dissociation in Supplemental information) and cryopreserved in freezing medium (90% FBS, 10% DMSO) until seeding. DTCs from donors HCC2, HCC4, HCC5, HCC6, HCC7, and HCC8 were purchased from Discovery Life Sciences and cryopreserved until seeding (Table 1). Informed consent was obtained from all donors by the tissue provider, and ethical oversight was managed by the tissue provider and MIMETAS.

HCC cells lines Huh7 (JCRB0403) and HLE (JCRB0404) were obtained from the Japanese Collection of Research Bioresources Cell Bank. These were cultured in DMEM supplemented with 10% fetal bovine serum (FBS,Gibco 16140-071) and 1% of penicillin and streptomycin solution (Sigma Aldrich, P4333).

Tissue dissociation

Fresh HCC tissue was dissociated as follows: the tissue was washed with Ad+++ (advanced DMEM/F12, Thermo Fisher Scientific,12634010) supplemented with 1% penicillin and streptomycin and cut into 3–5 mm pieces. The pieces were then collected in DBSA, DMEM supplemented with 1% bovine serum albumin (BSA, Sigma Aldrich, A2153) and centrifuged for 10 min at 300 × g. The supernatant was discarded, and the pellet was resuspended in digestion medium (Ad+++ supplemented with 2.5 mg/mL collagenase IV (STEMCELL Technologies, #07427) and incubated at 37 °C with vigorous shaking. After 15 min, DNase (Sigma, D5025) was added to the digestion medium and incubated for 30 min at 37 °C with vigorous shaking. The digestion was stopped by addition of complete EMEM and the cell suspension was passed through a 70 µm cell strainer. The resulting single-cell suspension was centrifuged for 10 min at 400 × g, the supernatant was discarded, and the cell pellet was resuspended in ammonium chloride solution (STEMCELL Technologies,07800) and incubated for 2 min to lyse any remaining red blood cells. The lysis was stopped by addition of DBSA, then cells were counted and either cryopreserved in freezing medium or cultured for fibroblast isolation.

Fibroblast isolation

Cancer-associated fibroblasts (CAF) were isolated from donor HCC3 dissociated tissue cells (DTC). Following HCC tissue dissociation, cells were cultured in a T25 flask in 67% complete MV2 and 33% complete EMEM, consisting of EMEM (ATCC, 30-2003) supplemented with 10% fetal bovine serum (FBS, Gibco, 16140-071) and 1% penicillin/streptomycin solution (Sigma Aldrich, P4333) until formation of fibroblast colonies, as established by observation under light microscopy based on morphology. Cells were then passaged and cultured until passage 4, when cell population consisted mostly of fibroblasts as determined by morphology and α-SMA expression (detected by immunofluorescence).

Generation of HCC PDChips

Cells were seeded into an OrganoPlate^®^ Graft (MIMETAS B.V., 6401-400-B). Seeding was performed as follows; HUVEC and CAF were cultured for one passage prior to seeding (seeded at passage 6 and 5, respectively). On the day of seeding, cells were detached using Trypsin/EDTA (Lonza, CC-5012), neutralized using complete MV2, centrifuged for 5 min at 300 × g, and resuspended in complete MV2 to the desired concentration. DTC were thawed in complete MV2, centrifuged for 5 min at 300 × g and resuspended in complete MV2 to the desired concentration. Thrombin (Enzyme Research Laboratories, HT 1002A) was diluted in complete MV2 to a concentration of 1 U/mL. DTC, HUVEC, CAF, thrombin, and fibrinogen (Enzyme Research Laboratories, FIB1) were kept on ice through the seeding process. The seeding mixture was prepared using the following ratios: DTC (3250 cells/μL), HUVEC (6500 cells/μL), and CAF (3250 cells/μL), thrombin (0.1 U/mL), and fibrinogen (5 mg/mL). Different cells types were combined first, followed by addition of fibrinogen and thrombin immediately prior to seeding. Seeding mixture (1.35 μL) was dispensed in the graft chamber and allowed to polymerize at room temperature for 10 min. Following polymerization, 1 μL HUVEC suspension (10,000 cells/μL in complete MV2) was added to each perfusion channel inlet and allowed to fill the perfusion channels before dispensing 50 μL complete MV2 to the perfusion channel inlets and to the graft chamber. The plate was incubated static at 37 °C for ~2 h to allow for HUVEC attachment in the perfusion channels. 50 μL complete MV2 was then dispensed in the perfusion outlets and the plate was incubated at 37 °C on the OrganoFlow at 14° at an 8-min interval. On the next day, all medium was removed from the plate and replaced with 50 μL/well complete ECGM2 (Endothelial Cell Growth Medium 2, PromoCell C-22011, supplemented with 1% p/s). Complete ECGM2 was refreshed in all wells every 48 h until compound exposure.

Compound exposure

The drug panel (Table 2) was composed of HCC SoC as well as a selection of compounds able to target several signaling pathways in tumor cells as well as the supporting endothelial cells and CAFs. Compound exposure was performed on day 6 of culture, when culture medium was removed and replaced with complete ECGM2 containing treatment compounds or controls according to the table below (Table 2). Each treatment condition and monoclonal antibody control was assigned to two chips in each plate (four chips per donor), while our vehicle control was assigned to six chips in each plate (twelve chips per donor). Compound addition to the plate was performed semi-automatically using a pipetting robot (Biomek i5; Beckman Coulter) to mix and dispense the reagents. Compound exposure was maintained for 72 h, until the end of the culture (day 9).

Alamar blue viability assay

On culture day 9, cell supernatant was collected from each well using a pipetting robot (Biomek i5; Beckman Coulter) and stored at −80 °C until cytokine analysis. Following supernatant collection, AlamarBlue cell viability reagent (Thermo Fisher Scientific, DAL1100) diluted 1:10 in culture medium was added to each well using a non-contact dispenser (MultiFlo FX; BioTek) and incubated at 37 °C for 2 h on the OrganoFlow^®^ rocker (MIMETAS B.V., MI-OFPR-L) at 14° at an 8 min interval. Following incubation, fluorescence was quantified at 544/590 nm using the Spark Cyto plate reader (Tecan Life Sciences).

Immunohistochemistry

After Alamar blue assay cultures were fixed as follows; all medium was removed from each well, replaced with fixation solution 3.7% formaldehyde (Thermo Fisher, 033314.K2) in HBSS with calcium and magnesium (Thermo Scientific, 14025092), and incubated at room temperature for 15 min. Fixation solution was removed, and each well was washed three times with phosphate buffered saline (PBS). Following the third wash, 50 μL PBS were added to each well and the plate was sealed using Parafilm and stored at 4 °C until immunofluorescent staining. For the immunofluorescent staining of endothelial, tumor, and fibroblast markers, plates were permeabilized and blocked for 2 h with blocking buffer (1% Triton X-100 + 3% BSA in PBS), followed by primary antibody incubation for Albumin-FITC (A80-229F, Bethyl), VE-Cadherin (Ab33168, Abcam), CD31 (M0823, Dako) and α-SMA (A2547, Sigma). Nuclei were stained with Hoechst (H3570, Thermo Fisher). Primary antibodies were diluted in antibody buffer (0.3% Triton X-100 3% BSA in PBS) overnight. Primary antibodies were washed three times with washing buffer (0.3% tritonX-100 in PBS) and incubated with Donkey anti-rabbit 647 (A31573, Thermo Fisher) or Goat anti-mouse 647 (A21236, Thermo Fisher) secondary antibodies in antibody buffer overnight. Secondary antibodies were washed three times with washing buffer and once with PBS. Following the washing steps, PBS was added to all wells and the plate was imaged using the ImageXPress XLS Micro Confocal (Molecular Devices).

Luminex analysis

Medium from culture day 9 was collected for cytokine and chemokine analysis. The samples were analyzed for quantification of AFP, CCL2, CCL3, CCL4, CCL20, CXCL1, CXCL10, CXCL11, IL-4, IL-8, IL-21, CXCL12, TNFα with the Human ProcartaPlex Mix&Match 13-plex (ThermoFisher Scientific) and for quantification of IL-6 with ProcartaPlex Human IL-6 (ThermoFisher Scientific). The assay was carried out following the protocol provided by the manufacturer. Raw data was processed using the ProcartaPlex Analysis App (ThermoFisher Scientific).

Vascular bed analysis

To assess phenotypical changes in the vascular network of HCC cultures, images were processed using FIJI, and segmented using a trainable image processing plugin Labkit [38]. Vascular network images (visualized through CD31 immunostaining) were preprocessed, and background signal was removed using the rolling ball background subtraction algorithm [39]. This step removed unwanted artifacts from the vessel images. Next, images were fed to a classifier trained with the Labkit plugin, resulting in three segmentation classes: background, thin vessels and thick vessels (Fig. 1B). After being segmented, the images were postprocessed, where signal outside of the area of interest (tumor compartment) and artifacts smaller than a certain threshold were removed, this improved the quality of the segmentation. The cleaned segmented network was then skeletonized to extract vessel descriptors further described in Table 3. Further data analysis was performed using Python The Python packages Pandas (Mckinney, 2010), NumPy [40] and Scikit-learn (Pedregosa et al., 2011) were used for data processing and Matplotlib was used for visualizations.

Data analysis

The number of replicates and selected readouts were determined based on a pilot experiment, this preliminary study was used to assess the reproducibility of on-chip culture generation and readouts. Data were excluded only in cases of clear outliers or technical failures that compromised assay performance.

Data was processed in Excel and graphs were generated in GraphPad, R studio and Python. Raw Luminex data was processed using the ProcartaPlex Analysis App (ThermoFisher Scientific). Data analysis and visualization were performed using R version 4.4.2 and RStudio version 2024.12.0 using the tidyverse package set. Normality was assessed using Shapiro–Wilk test. Statistical analyses were performed using pairwise Wilcoxon rank-sum test with Benjamini–Hochberg correction for multiple testing. Comparison of AFP, IL6, and CCL2 levels between control conditions of HCC patient and HCC cell line cultures was performed using unpaired Welch-corrected t-test. Vascular beds analyses were performed using CD31 images, these were segmented using a Labkit based segmentation model in FIJI. Albumin expression was quantified using INCarta image analysis software (Molecular Devices, version 2.1). A U-Net CNN model based on ResNet was trained using the SINAP module. The model accurately segmented the Albumin staining under a range of different expression levels. The total area expressed per chip was extracted, linked with experimental treatment metadata for comparison through bar charts. For the comparison of viability, tumor cell population, vasculature and morphology independent analyses were conducted. One-Way ANOVA followed by Tukey’s HSD post-hoc test was used for gaussian data, or Kruskal–Wallis followed by Dunn’s post-hoc test for non-gaussian data was used. All p-values were adjusted for multiple comparisons using Bonferroni correction, with adjusted values (q-values) ≤0.05 considered statistically significant.

Supplementary information

Supplemmental Information