Close this search box.

Chip collection of hepatocellular carcinoma based on O2 heterogeneity from patient tissue – Nature Communications

For the clinical experiments, following approved protocols from the Institutional Review Board (IRB) of Yonsei University College of Medicine (IRB: 4-2016-0728). Animal experiments were conducted in accordance with the guidelines set by the Institutional Animal Care and Use Committee (IACUC) at Yonsei University College of Medicine (#2021-0176).

Tissue and cell samples from hepatocellular carcinoma (HCC) patients

Twelve patients with HCC consented to donate tissue during surgical tumor removal. Diagnosed in the early stages according to the Barcelona Clinic Liver Cancer system43, patients were classified into typical and irregular rim-like enhancement (IRE) HCC groups based on medical records and MRI findings. Typical HCC exhibited uniform arterial phase enhancement above the liver’s overall level, while IRE HCC displayed irregular enhancement with a bright peripheral border and dim center20. The quantitative analysis of MRI contrast intensity was conducted using the segmentation plot plugin from ImageJ (Fiji version 1. 52i, National Institute of Health, MD, USA) along each midsection. Tumors were resected within two months of diagnosis, followed by histological analysis to evaluate tumor characteristics, such as level of differentiation, microvascular invasion, and portal vein invasion. Diagnostic patient information included age, tumor mass size, viral infection status, and tumor marker levels (AFP and PIVKA-II). Regular post-operative monitoring involved CT, MRI, and tumor marker tests to detect recurrence. HCC tissues obtained during surgical resection were immediately transported to the laboratory in cold culture media. Upon arrival, the tissues were minced using grinder tips (G50; Coyote Bioscience, BJ, China), centrifuged at 1330 ×g for 3 min, and weighed. Subsequently, cells were extracted by incubating minced tissue with collagenase type I (210 U mL−1, 17100-017; Gibco, Carlsbad, CA, USA) and DNase I (10 μg mL−1, 11284932001; Sigma-Aldrich, St. Louis, MO, USA) in Dulbecco’s modified Eagle’s medium-high glucose with L-glutamine and without sodium pyruvate (DMEM, 11965-092; Gibco), supplemented with 1% (v v−1) penicillin-streptomycin (PS, 15140-122; Gibco) for 1 h at 37 °C under continuous shaking. After incubation, the mixture was filtered using a cell strainer with a 70 μm pore diameter (352350; Corning, NY, USA), and collagenase was deactivated by washing the cells three times in DMEM medium with 10% heat inactivated fetal bovine serum (FBS, 16000-044; Gibco). Cells were then cultured in DMEM supplemented with epithelial growth factor (EGF, 20 ng mL−1, PHG0311; Thermo Fisher Scientific, Waltham, MA, USA) and basic fibroblast growth factor (b-FGF, 10 ng mL−1,100-18C; PeproTech, Rocky Hill, NJ, USA) for 5–10 passages. Additionally, the HCC cell line Hep3B, obtained from the Korean Cell Line Research Foundation (88064; Seoul, Republic of Korea), was cultured in DMEM at 37 °C with 20% O2 and 5% CO2.

Production of the oxygen gradient chip

For the microchannel networks, microfibers were produced by spinning a PCL-PVAc-PEG copolymer (Soluplus®, 30446233; BASF, Ludwigshafen, Germany) dissolved in a 66% w v−1 MeOH solution using a custom device. The diameter of the fibers was adjusted by varying the rotational speed (140–176 g) of the spinning device42,44,45,46. Subsequently, the fibers were placed in a polydimethylsiloxane (PDMS, 31-000810-02; Dow corning, Midland, MI, USA) mold measuring 2 × 2 × 0.5 cm at a density ranging from 0 to 3 mg mL−1. The bottom plate of the mold was sealed onto a glass slide (1000412; Marienfeld, Lauda-Königshofen, Germany). Inlet and outlet holes for media perfusion were created by piercing the left and right ends of the PDMS mold with an 18G blunt needle (0721092; Korea Vaccine, Seoul, Republic of Korea). Two pairs of inlets and outlets were arranged in parallel on the upper (normoxia) and lower (hypoxia) sides of the mold to allow simultaneous perfusion of normoxic and hypoxic media through the corresponding inlets. Next, a solution containing gelatin (5.5% w v−1, G1890; Sigma-Aldrich) and microbial transglutaminase (1% w v1, mTG, 1201-50; Modernist Pantry LLC, Eliot, ME, USA) in a volume ratio of 9:1 was poured onto the Soluplus® microfibers within the PDMS mold. The mixture was then cross-linked for 30 min at 37 °C. The fibers were dissolved through a gel-to-solution transition at room temperature (RT) below the LCST of 38 °C and thoroughly perfused with PBS wash. Each inlet and outlet was created by inserting luer tubes into the hydrogel. This setup facilitated media perfusion into the inlet, circulation through the microchannel network, and subsequent outflow through the outlet, creating a closed circulation system. The chip was cultured by perfusing each type of media at a flow rate of 4.71 mL min−1 using a peristaltic pump (BT100-1L; Longer Precision Pump Co., Ltd, Amersham, UK) with daily media change. Hypoxic media were generated by deoxygenating normoxic media through incubation in a Hypoxystation (Whitley H35; West Yorkshire, UK) with 0.1% O2 and 5% CO2 for 4 h. Microchannel visualization was achieved by perfusing green fluorescence microspheres (0.2 μm, TetraSpeck, T7280; Thermo Fisher Scientific) with PBS (1:250) under confocal imaging (LSM 980; Zeiss, Oberkochen, Germany) with Zen 3.3 blue edition. Hydrogel porosity was assessed using scanning electron microscopy (Merlin, Zeiss), followed by quantitative image analyses using ImageJ (Fiji).

Computational fluidic dynamics (CFD)

CFD modeling was employed to determine the diffusion coefficient (D) of the hydrogel as a function of fiber density. It was also used to calculate the oxygen gradient within the chip based on cellular oxygen consumption. Initially, a single-inlet chip was used to experimentally derive the D in relation to the hydrogel porosity by varying the fiber density. This approach helped in isolating the effects of microchannel porosity from other parameters. To create an oxygen-free chip, the hydrogel was incubated in a Hypoxystation with 0.1% O2 and 5% CO2 for 4 h to remove oxygen. An oxygen dissolve meter (DM-1; CAS, Gyeonggi-do, Republic of Korea) was used to confirm the resulting anoxic conditions. The single-inlet chip was then perfused with normoxia media for 1 min while varying the fiber density (0, 1, 2, and 3 mg mL−1). Oxygen profiling was performed at incremental distances from the inlet using a Microx 4 system (PM-PSt7; PreSens Precision Sensing GmbH, Regensburg, Germany). The 3D chip model was created using computer-aided design (CAD) software (SpaceClaim; ANSYS, Canonsburg, PA, USA). Two pairs (two-inlet chips) of cylindrical inlets and outlets, each 1 mm in diameter and 20 mm in length, were connected to a cuboid hydrogel with dimensions of 20 mm (width) × 20 mm (length) × 5 mm (thickness), consistent with the actual chip size. Finite element analysis via Fluent Meshing (2020R, ANSYS, USA) generated a total of 143,710 polyhedral mesh elements. The diffusion of oxygen gas within the chip hydrogel was calculated under continuous media (water) flow conditions using the media density (1007 kg m−3). viscosity (0.958 × 10−3 kg m−1 s−1), and the diffusion coefficient of oxygen-media (DOM: 2.88 × 10−9 m2 s−1 at 37 °C)47. This calculation assumed an equal diffusion coefficient for oxygen and the media. Darcy’s law was used as the governing equation for hydrogel porosity, which is represented as:

$${{{{{bf{Q}}}}}}={{-}}{{{{{bf{kappa }}}}}}cdot {{{{{rm{A}}}}}}cdot Delta {{{{{rm{P}}}}}}/{{{{{rm{mu }}}}}}cdot {{{{{rm{L}}}}}}$$


where Q denotes the flow rate, A signifies the cross-sectional area, L represents the length of the system, ∆P stands for the pressure drop, μ is the media viscosity, and κ represents the viscous resistance. According to Darcy’s law, the pressure drop is proportional to the flow rate but inversely proportional to the viscous resistance of non-Newtonian flow, especially at low Reynolds numbers. The hydrogel’s viscous resistance was set to 7.5 × 1011 m−248, and porosity values of 38.8%, 44.4%, 54.6%, and 73% based on experimental data were used to reflect variations in fiber density (0, 1, 2, and 3 mg mL−1), respectively. The inlet boundary conditions for CFD analysis included the inlet flow velocity (100 mm s−1) and the oxygen concentration of normoxic media (5.45 mg mL−1). Transient simulations of the single-inlet chip system were conducted over 3000 iterations with 0.1-s intervals. The resulting color-coded volume was rendered using chart visualization to illustrate the oxygen concentration gradient within the chip. Moreover, a regression curve was plotted to establish a correlation between oxygen levels (y-axis) and diffusion coefficients (x-axis). This facilitated the calculation of hydrogel diffusion coefficients as a function of the fiber density by inputting probe-recorded oxygen concentration values on the y-axis of the regression curve. The two-inlet chip with embedded HCC tissue was simulated by incorporating tissue oxygen consumption into the CFD model. Tissue oxygen consumption was represented by a hypoxic oxygen influx modeled from the upper wall of the chip, allowing for the determination of the influx volume (Supplementary Fig. 3). The modeling was carried out using boundary conditions at the normoxia inlet (O2 concentration = 5.45 mg mL−1, inlet velocity = 100 mm s−1), the hypoxia inlet (O2 concentration = 0 mg mL−1, inlet velocity = 0 mm s−1), and the upper wall influx (0.007 mL s−1). The calculation process went through 1000 iterations via steady-state simulation, with the results presented using color-coded contour diagrams and oxygen concentration plots. Additionally, the effects of oxygen diffusion on the viability of Hep3B cells were investigated by culturing 5 × 106 cells in the chip for 5 days. A CCK-8 assay (1:10 ratio, Dojindo Molecular Technologies, Inc., Rockville, MD, USA) was performed, and the absorbance readings were taken at 450 nm using a plate reader (SpectraMax GeminiTM XPS/EM; Molecular Devices LLC, San Jose, CA, USA).

Tissue culture in the chip and immunohistochemistry

Tissue pieces (50 mg mL−1) were obtained from HCC patient tissues using a biopsy punch (1 mm in diameter, BP-10F; Kai Medical, TX, USA). The pieces were mixed with gelatin hydrogel and embedded in the chip at 37 °C, followed by media perfusion for 4 days. Tissues were then fixed with 4% paraformaldehyde (PFA, CNP015-0550; CellNest, Hanam-si, Gyeonggi-do, Republic of Korea) in PBS for 2 h at RT, washed three times with PBS, and then embedded in paraffin before being sectioned into 4 μm thick blocks. Hematoxylin and eosin (H&E) staining was conducted according to standard protocols, and optical imaging was performed using an inverted microscope (DMi8 M; Leica, Wetzlar, Germany).

Deparaffinization and rehydration of the tissue sections were accomplished using a series of dilutions with xylene and ethanol (100, 95, 80, and 70% v v−1 in deionized water). Antigen retrieval for CD31 was carried out using a low-pH buffer (k8005; Agilent Dako, Santa Clara, CA, USA), while a high-pH buffer (k800421-2; Agilent Dako) was used for K19, CD34, and AFP. CAIX did not require antigen retrieval. Endogenous peroxidases were deactivated by incubating the samples with a 3% H2O2 solution (H1009; Sigma-Aldrich) for 10 min, followed by washing with tris-buffered saline (TBS, ML023-03; Welgene, Gyeongsan-si, Gyeongsangbuk-do, Republic of Korea) and blocking with 5% bovine serum albumin (BSA, A0100-005; GenDEPOT, Altair, TX, USA) in PBS. The samples were then incubated with primary antibodies for CAIX (1:1000, NB100-417, Novus Biological LLC), CD31 (1:100, mouse monoclonal, NB600-562, Novus Biological LLC), K19 (1:1000, mouse monoclonal, ab9221, Abcam), CD34 (1:50, GA63261-2, Agilent Dako), and AFP (1:100, rabbit monoclonal, ab169552, Abcam) at RT for 1 h. Subsequently, the samples were incubated with HRP-labeled secondary antibodies (1:5000, anti-rabbit polymer, k4003; Agilent Dako) at RT for 20 min. Afterward, the samples were treated with DAB development solution (k3468; Agilent Dako) for 5 min, washed with deionized water, and counterstained with hematoxylin (k8008; Agilent Dako) before optical imaging. Cell apoptosis was assessed using the Click-iT Plus TUNEL assay (C10617; Thermo Fisher Scientific) following the manufacturer’s instructions. Confocal imaging (LSM 980; Zeiss) was performed with Zen 3.3 blue edition, and quantitative image analysis was conducted using ImageJ (Fiji).

Determination of cell preference to oxygen levels

HCC cells from typical (#6) and IRE (#9) HCC tissue were labeled with DiI (green) and DiO (red), respectively, using the Vybrant Multicolor cell-labeling kit (V22889; Thermo Fisher Scientific) at a ratio of 1:500 for 30 min. Subsequently, the cells were embedded and cultured in the chip at a concentration of 5 × 106 cells mL−1 for 7 days. After harvesting the hydrogels, they were washed with PBS three times and subjected to confocal imaging using tile and z-stack scanning (LSM 980; Zeiss) with Zen 3.3 blue edition. Quantitative image analysis was conducted using ImageJ (Fiji). Furthermore, HCC cells (typical #6 and IRE #9) were cultured in 24-well plates at a density of 5 × 104 cells well−1 for 3 days under either normoxic (20% O2) or hypoxic (0.1% O2) conditions. Following culture, the cells were fixed with 4% PFA in PBS at RT for 30 min and washed three times with PBS. Then, the cells were permeabilized using 0.1% Triton X-100 (T8787; Sigma-Aldrich) in PBS at RT for 1 h, followed by another three washes with PBS. F-actin staining was achieved by treating the cells with Alexa Fluor 488 phalloidin (1:500, A12379; Thermo Fisher Scientific) at RT for 1 h, followed by three washes with PBS. Confocal imaging with Zen 3.3 blue edition and quantitative image analysis by ImageJ (Fiji) were performed after 4’,6-diamidino-2-phenylindole, dihydrochloride (DAPI, R37606; Thermo Fisher Scientific) counterstaining. HCC cells (IRE #8) were cultured in the chip for 7 days and fixed overnight at 4 °C with 4% PFA in PBS. After three washes with PBS, the cells were permeabilized for 10 min using 0.5% Triton X-100 under perfusion with a 26 G syringe, followed by incubation at RT for 1 h. Subsequently, the samples were blocked for 1 h at RT using 1% BSA and 0.2% Triton X-100 in PBS. Primary antibodies for CD34 (1:200, mouse monoclonal, sc-7324; Santa Cruz Biotechnology, Dallas, TX, USA) and CAIX (1: 500, rabbit monoclonal, NB100-417; Novus Biological LLC), serving as markers of microvascular invasion and stemness, respectively, were applied to the samples and incubated overnight at 4 °C. After three washes with PBS, the samples were treated with secondary antibodies: fluorescein (FITC)-conjugated AffiniPure (1:250, goat anti-mouse IgG, 115-095-003; Jackson Immuno Research, West Grove, PA, USA) for CD34 and Alexa Fluor 594-conjugated AffiniPure (1:250, goat anti-rabbit IgG, 111-585-003, Jackson Immuno Research) for CAIX for 2 h at RT. Following another three washes with PBS, the samples were stained with DAPI for nucleus visualization and subjected to confocal imaging (LSM 980; Zeiss) in z-stack and tile scanning with Zen 3.3 blue edition. Quantitative image analysis was performed using ImageJ (Fiji).


Total RNA was extracted from cells using TRIzol reagent (15596018; Thermo Fisher Scientific), followed by reverse transcription of 1 μg of RNA into cDNA using AccuPower Cycle Script RT Pre-mix (Bioneer, Daejeon, Republic of Korea). The entire primer sequences were designed using the National Center for Biotechnology Information via Primer-BLAST. Oligonucleotides were purchased from Cosmogenetech (Seongdong-gu, Seoul, Republic of Korea). The StepOne Real-Time PCR system (Applied Biosystems, Foster City, CA, USA) was then employed for 40 cycles of target gene amplification with SYBR Green, cDNA, and primers (Table S2, Supporting Information). Gene expression levels were determined by comparing them with the Ct value of the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH) using StepOnePlus version 2.3. The resulting data were presented as a heatmap after converting the Ct values into Log2Fold values, and visualization was carried out using the RStudio (Version: 2023.06.1+524; Boston, MA, USA) heatmap.2 function.

Western blot

Proteins were extracted through incubation of samples in RIPA lysis buffer (89900; Thermo Fisher Scientific) supplemented with 1X Protease and Phosphatase Inhibitor Cocktail (78440; Thermo Fisher Scientific) on ice for 2 h. This was followed by syringe shattering with needles of varying diameters (18–26 G). Protein concentrations were then determined using a BCA protein assay kit (23227; Thermo Fisher Scientific). The protein samples were loaded onto SDS-PAGE (4–15% w v−1) polyacrylamide gel electrophoresis (4561094; Bio-Rad, Hercules, CA, USA) and then electro-transferred onto a nitrocellulose membrane using the iBlot 2 Gel Transfer Device (Invitrogen, Carlsbad, CA, USA). After blocking with a solution of 5% nonfat dry milk in TBS containing 0.1% Tween-20 (P9416; Sigma-Aldrich) for 1 h at room temperature, the membrane was incubated overnight at 4 °C with primary antibodies for CAIX (1:500, rabbit polyclonal, NB100-417; Novus Biologicals LLC, Centennial, CO, USA), ABCB1 (1:250, mouse monoclonal, sc-55510; Santa Cruz Biotechnology), and β-actin (1:500, mouse monoclonal, sc-47778; Santa Cruz Biotechnology). After three washes with TBS-T, the samples were exposed to secondary antibodies at a dilution of 1:5000. This included an anti-rabbit IgG antibody (ab6721; Abcam) for CAIX and an anti-mouse IgG antibody (ab6708; Abcam) for ABCB1 and β-actin. Protein bands on the membranes were observed using a chemiluminescence imaging system (ImageQuant LAS 4000; GE Healthcare Life Sciences, Chicago, IL, USA), and subjected to quantitative image analysis using ImageJ (Fiji).

Proteinase (MMP1) activity

Proteins were initially extracted and stored at −80 °C, then homogenized using 100 μL of cell lysis butter. After a 5-min incubation on ice, the samples were subjected to centrifugation at 16,000 × g at 4 °C for 10 min. The resulting supernatant was then transferred to a pre-chilled tube, and its concentration was determined using a BCA protein assay kit. To assess MMP1 activity, a fluorometric-based collagenase (collagen degradation/zymography) assay kit (ab234624, Abcam) was employed according to the manufacturer’s instructions. Fluorescence intensity was measured using a fluorometer (Varioskan Flash 3001, Thermo Fisher Scientific) at an excitation/emission wavelength of 490/520 nm in the kinetic mode at 37 °C for 90 min with 4-min intervals.

IC50 of anti-cancer drugs

HCC cells were cultured in 96-well plates at a density of 1 × 104 cells per well and allowed to reach 90% confluency over three days. Afterward, the cells were treated with Doxorubicin (D1515, Sigma-Aldrich) dissolved in distilled water at concentrations ranging from 0 to 1 µM or Sorafenib (SML2653, Sigma-Aldrich) dissolved in DMSO at concentrations ranging from 0 to 40 µM for 24 h at 37 °C with 5% CO2. Cell viability was assessed using a CCK-8 assay, performed at a 1:10 dilution, and the IC50 value for each drug was determined using the trendline.

Dual gradient chip

The dual gradient chip was developed by establishing a drug gradient perpendicular to the oxygen gradient. Initially, HCC tissues from IRE #8 and #9, at a concentration of 50 mg mL−1, were cultured on the chip for 7 days under the oxygen gradient. Subsequently, the cells were exposed to drug perfusion through one inlet of the chip for 3 days, with the concentration set 3 times higher than their respective IC50 value. At the same time, drug-free media was perfused through the other inlet to establish the drug gradient. Cell viability was assessed using a live/dead assay (L3224; Thermo Fisher Scientific) followed by confocal imaging (LSM 980; Zeiss) with Zen 3.3 blue edition. The regions of interest (ROI) were analyzed using the ImageJ plugin, and visualization was performed using the RStudio heatmap.2 functions excluding the clustering step.

Oxygen profiling in vivo

Male BALB/c nude mice aged six weeks, were obtained from Orient Bio (Seoul, Republic of Korea) and acclimated to a pathogen-free environment with a 12-h light/dark cycle, ambient temperature and humidity. Organ oxygen levels were assessed using the Microx 4 oxygen profiling system in three distinct groups: normal limb (NL), subcutaneous (SC), and ischemic limb (IL). Anesthesia was administered through intraperitoneal injection of zoletil (50 mg kg−1) and xylazine (10 mg kg−1). In the NL and SC groups, a small incision (<1 mm) was made in the right hindlimb and right subcutaneous region, respectively. In the IL group, ischemia was induced by ligating the femoral artery of the right hindlimb at the proximal and distal points using 4-0 silk, followed by resection of the artery segment between the ligation points. Subsequently, oxygen levels were measured using the system probe, and the mice were euthanized in a CO2 chamber.

Chip PDX model with systemic drug response

A microchannel hydrogel (2 × 2 × 2 mm) containing IRE #8 tissues (50 mg mL−1), epithelial growth factor (EGF, 20 ng mL−1), and basic fibroblast growth factor (b-FGF, 10 ng mL−1), was surgically implanted into the NL, SC, and IL regions of mice for 6 weeks. Two weeks after the chip implantation, the mice were injected with Doxorubicin in distilled water (100 μL, 3 mg kg−1) through the tail vein. This injection was repeated after 2 weeks, resulting in a total of two doses. Additionally, Sorafenib in DMSO (100 μL, 30 mg kg−1) was orally administered to each mouse. Drug responsiveness was evaluated by comparing the effects of the vehicle control (consisting of no drugs in DMSO and saline) with each specific drug (Doxorubicin or Sorafenib). After euthanasia using CO2 gas, tumor tissues were collected, and the tumor volume was measured using digital calipers. The x, y, and z-axis values were calculated and normalized to the volume of the tumor tissue at day 0, which served as the baseline (i.e., the volume of the mother tissue).

Statistical analysis

Data analysis was conducted using Excel (version 16.0.17531.20004), SigmaPlot 12.0 (Systat Software Inc., San Jose, CA, USA) and Prism 8.0.1 (GraphPad Software, Boston, MA, USA). The data are presented as means ± standard error of the mean. Paired comparisons were assessed using a two-tailed Student’s t-test, while multiple comparisons were carried out using one-way ANOVA with Tukey’s significant difference post-hoc test. Statistical significance was indicated as exact P-value. Clinical data were evaluated using the t-test of independent samples, with equal variances pooled for typical and IRE patients. Events, such as viral infection and recurrence ratio, were analyzed using the chi-square test. Histological differences were examined using Fisher’s exact test. Data were normalized and transformed with corresponding n numbers as described in each figure legend.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.