Culturing and maintenance of stem cells

hPSCs were cultured in E8 media (Thermo Fisher Scientific) on plates coated with 1% (v/v) Geltrex (Thermo Fisher Scientific). The hPSCs were incubated in a humidified 37 °C, 5% CO2 incubator, and the media was replenished every 24 h. Passaging was performed with 5 µM EDTA (Thermo Fisher Scientific) in DPBS (Thermo Fisher Scientific) and the hPSCs were replated as small colonies at dilutions from 1:3 − 1:5. Quality control of the hPSCs was regularly performed by IF imaging, RT-qPCR and flow cytometry for selected pluripotency markers. All cultured cells were checked for mycoplasma (BioNordica).

scHSC differentiation

The scHSCs were generated as previously described [37] based on published protocols for scHSC 2D differentiation [32, 33]. The differentiation was performed on four hPSC lines: WAe001-A (H1) (scHSC_1, WiCell Research Institute), UCSFi001-A (WTC11) (scHSC_2, Corelli Institute for Medical Research), WTSIi046-A (HPSI0214i-wibj_2) (scHSC_3, Wellcome Trust Sanger Institute), and HMGUi001-A (XM001) (scHSC_4, Helmholtz Zentrum München). As we have previously shown that different hPSC lines lead to variations in hPSC-derived HSC models [37], four hPSC lines were used to account for hPSC line-dependent variability. A schematic of the 12-day differentiation protocol is presented in Fig. 1A. See also Additional File 2 for media compositions. The differentiation media was changed every 48 h or every 24 h if a new media composition was required.

hPSCs were seeded as single cells and cultured on plates coated with 1% (v/v) Geltrex (Thermo Fisher Scientific) in E8 media (Thermo Fisher Scientific) supplemented with 10 µM Rock inhibitor (STEMCELL technologies). The differentiation protocol was initiated at approximately 25% confluency after 24–48 h of incubation. On day 0, the cells were incubated in DMEM/ F-12 medium with Glutamax™ supplement (Thermo Fisher Scientific) with 1% (v/v) MEM Non-Essential Amino Acids Solution (Thermo Fisher Scientific) and 1% (v/v) B-27™ Supplement (Thermo Fisher Scientific) and growth factors Activin A at 100 ng/mL (Peprotech), CHIR at 3 µM (Tocris), and BMP-4 at 20 ng/mL (Peprotech). From day 1 to day 5, the cells were grown in a basal mesoderm medium consisting of DMEM/ F-12 medium with Glutamax™ supplement containing 1% (v/v) MEM Non-Essential Amino Acids Solution, 1% (v/v) B-27™ Supplement (Thermo Fisher Scientific), 0.025% Insulin-Transferrin-Selenium (ITS-G) (Thermo Fisher Scientific), 2.5 µM Dexamethasone (Merck Sigma-Aldrich) and 100 µM 2-Phospho-L-ascorbic acid trisodium salt (Merck Sigma-Aldrich). This basal mesoderm media was supplemented with 20 ng/mL BMP-4 on day 1, and 20 ng/mL BMP-4, 20 ng/mL FGF-1 (Peprotech), and 20 ng/mL FGF-3 (R&D Systems) on days 2–5. The differentiating cells were passaged on day 5 and 10 µM Rock inhibitor was added to the differentiation medium for 24 h. Accutase (Thermo Fisher Scientific) was used for detachment and the cells were plated as single cells in a ratio of 1:3 on plates coated with 1% (v/v) Geltrex. From day 6 to day 12, the cells were grown in a basal HSC medium consisting of DMEM/ F-12 medium with Glutamax™ supplement containing 1% (v/v) MEM Non-Essential Amino Acids Solution, 1% (v/v) Fetal Bovine Serum (FBS) (Thermo Fisher Scientific), 0.025% ITS-G, 2.5 µM Dexamethasone, and 100 µM 2-Phospho-L-ascorbic acid trisodium salt. This basal HSC media was supplemented with 20 ng/mL FGF-1, 20 ng/mL FGF-3, 100 µM Palmitic acid (PA) (Merck Sigma-Aldrich), and 5 µM Retinol (ROL) (Merck Sigma-Aldrich) on days 6–12.

Human pHSC culture

Commercially available pHSCs were purchased from iXCells Biotechnologies (two donors) and BeCytes (one donor) and cultured according to the manufacturer’s instructions on plates coated with 1% (v/v) Geltrex. The pHSCs from iXCells Biotechnologies were used for IF imaging and the radioactive substrate oxidation assays. The pHSCs from BeCytes were used for cytokine multiplex analysis, VA ELISA, and lactate secretion. pHSCs from all donors were used in qPCR analysis and HT LD image analysis. A 24-hour treatment was used on pHSCs from BeCytes since this donor material was sensitive to the starvation and starvation markedly reduced growth.

The pHSCs from iXCells Biotechnologies were initially plated in Stellate Cell Growth Medium (iXCells Biotechnologies) and pHSCs from BeCytes were thawed in NPCs Thawing Medium (All Types) (BeCytes) before culture in Stellate Cell Growth Medium (BeCytes) after 24 h. Treatments were initiated 24–48 h after thawing, see “scHSC and pHSC culture, treatments, and activation”. An aliquot of the pHSCs was collected before plating for downstream transcriptomic analysis.

scHSC and pHSC culture, treatments, and activation

Differentiated scHSCs and human pHSCs were grown in four conditions: “Control”, “Starvation”, “TGF-β”, and “Starvation + TGF-β”. The “Control” medium consisted of DMEM/ F-12 medium with Glutamax™ supplement containing 1% (v/v) MEM Non-Essential Amino Acids Solution, 1% (v/v) FBS (Thermo Fisher Scientific), 0.025% ITS-G, 2.5 µM Dexamethasone, 100 µM 2-Phospho-L-ascorbic acid trisodium salt, 100 µM PA, and 5 µM ROL. In the “Starvation” and “Starvation + TGF-β” media, PA and ROL supplementation was removed and FBS was substituted for 1% (v/v) Bovine Serum Albumin (BSA) (VWR, SEQENS). In the “TGF-β” and “Starvation + TGF-β” media, TGF-β (Peprotech) was added at a concentration of 25 ng/mL, a concentration used in several scHSC activation models [32, 35, 37]. The treatments were chosen to investigate the individual and combinational effects of short-term ROL and PA depletion and TGF-β-induced activation. See also Additional File 2 for media compositions. The length of the treatments was 48 h if not otherwise stated and media was replenished every 48 h. The 48-hour TGF-β-induced activation model was chosen to avoid adverse effects of long-term cultivation.

mRNA sequencing by Novogene

Cultured cells were detached with Trypsin-EDTA (Merck Sigma-Aldrich) before total RNA isolation by RNeasy Mini kit (Qiagen). The proper concentration and volume of samples were sent to Novogene UK Cambridge Sequencing Center for mRNA sequencing with WBI-Quantification. Novogene performed RNA sample quality control, mRNA library preparation (poly A enrichment), and Illumina sequencing PE150 (6G raw data per sample). Novogene also performed standard bioinformatics analysis including data quality control and data filtering, mapping to reference genome, gene expression quantification and correlation analysis, differential expression analysis, and enrichment analysis including GSEA enrichment analysis.

RT-qPCR

RNA from scHSCs was isolated by Trizol™ (ThermoFisher Scientific) and RNA from pHSCs was isolated using RNeasy Micro kit (Qiagen). cDNA was generated using a High-Capacity cDNA Reverse Transcription Kit (ThermoFisher Scientific). Gene expression was subsequently determined by RT-qPCR analysis on a ViiA 7 (ThermoFisher Scientific) thermocycler using TaqMan probes and TaqMan Gene Expression Master Mix (ThermoFisher Scientific). GAPDH was used as the housekeeping gene.

Immunofluorescence confocal microscopy

Cells were fixed on glass slides with 4% paraformaldehyde (PFA) (Merck Sigma-Aldrich) for 10 min. The samples were then permeabilized and blocked in a blocking solution made of 0.1% (v/v) Triton-X (Merck Sigma-Aldrich) and 10% (v/v) Fetal Bovine Serum (Thermo Fisher Scientific) diluted in DPBS (Thermo Fisher Scientific). Next, the cells were incubated overnight at 4 oC with primary antibodies properly diluted (see Materials section) in blocking solution before staining with secondary antibodies, and nuclear staining with DAPI (Thermo Fisher Scientific) was performed in the dark for 1 h at room temperature. The stained cells were then mounted on glass slides with mounting media (5% gelatin in 1:1 distilled water and glycerol). Confocal imaging was performed with an LSM700 (Zeiss, Germany) confocal microscope using standard filter sets and laser lines with a 40x oil immersion objective. Stained samples were stored at -20 oC.

α-SMA signal intensity analysis (Python)

α-SMA intensity of confocal images was quantified by an automated Python3 (03.09.2012) script. The script is on GitHub: https://github.com/ingridwilhelmsen/a-SMA_analysis. Antibody dilutions and confocal microscope settings were equal for all samples to ensure that the input material was quantifiable.

Flow cytometry

Flow cytometry was performed at the Flow Cytometry Core Facility at Oslo University Hospital, Gaustad, on a BD LSRFortessa™ (BD Biosciences) Cell Sorter and analysis was performed using FlowJo (10.9.0).

The cells were fixed as single cells by incubation in 4% paraformaldehyde (PFA) (Merck Sigma-Aldrich) for 15 min before blocking in a blocking solution made of 10% (v/v) Fetal Bovine Serum (Thermo Fisher Scientific) diluted in DPBS (Thermo Fisher Scientific) for 30 min. For subsequent steps, cells were washed and stained in a FACS buffer containing DPBS and 0.1% BSA. The cells were stained overnight with 1:50 diluted antibody against PDGFR-β (R&D Systems) before secondary antibody staining with 1:250 diluted Alexa Fluor® 488 AffiniPure Donkey Anti-Goat IgG (H + L) (Jackson ImmunoResearch) for 60 min. The cells were washed two times after each staining. PDGFR-β was detected in the flow cytometer by an Alexa Fluor 488-A laser and autofluorescence was assessed using a Pacific Blue 405 laser with a 450/50 band-pass filter, which is adequate but suboptimal for detection of VA which is autofluorescent at 330 nm [41]. A total of 10,000 cells were counted per sample. hPSC samples stained with only secondary antibody were used as negative controls.

Cytokine multiplex discovery

Conditioned media from scHSCs was collected after 48 h of treatment if not otherwise stated and stored at -80 oC until the assay was performed. Separate wells in 96-well plates were regarded as technical replicates.

The Human Luminex® Discovery Assay (R&D Systems/Bio-Techne) was used for the detection of 12 selected cytokines (CCL2, CCL3, CCL4, CCL5, CCL21, CXCL1, HGF, IL-1 beta, IL-6, IL-8, IL-10, TNF-alpha) according to the manufacturer’s instructions and the readout was performed with a Bio-Plex 200 (Bio-Rad) reader. Values that were extrapolated outside the range of the assay were excluded from downstream analysis.

UV autofluorescence image analysis

VA emits autofluorescence at 330 nm and is thus detectable under UV light [41]. Images were acquired using an AxioVert.A1 microscope equipped with a Colibri7 LED light source using the UV line (Excitation: 385/30, Emission: QBP 425/30) of Filter Set 90 HE LED, a 10x A-Plan 0.25 NA objective, and a CMOS digital camera (all from Carl Zeiss). The exposure time was set to 50 ms. To minimize bleaching of the UV signal [41], all images were acquired within 1 s of UV light exposure. Confluent areas were independently chosen under bright field before UV light acquisition. Separate wells in 96-well plates were regarded as technical replicates. The images were analyzed in ImageJ (Java 8, 32-bit) by an automated macro implementing thresholding by using the Triangle method [42] as previously described [37].

ELISA for Procollagen C-peptide 1

Conditioned media from scHSCs was collected after 48 h of treatment if not otherwise stated and stored at -80 oC until the assay was performed. Separate wells in 96-well plates were regarded as technical replicates.

The Procollagen I C-Peptide ELISA kit (Takara Bio Inc) was used according to the manufacturer’s instructions and the readout was performed by measuring absorbance at 450 nm with a Wallac Victor2 1420 multilabel counter (Perkin Elmer/Wallac) reader.

ELISA for human vitamin A

Cultured media was removed from the cells after 48 h if not otherwise stated. The cells were then washed with DBPS and subsequently snap-frozen at -80 oC until the assay was performed. Separate wells in 96-well plates were regarded as technical replicates.

The cells were lysed by RIPA lysis and extraction buffer (ThermoFisher Scientific) before intracellular Vitamin A levels were measured by a Human Vitamin A ELISA Kit (Colorimetric) (Novus Biologicals//Bio-Techne) according to the manufacturer’s instructions. The readout was performed by measuring absorbance at 450 nm with a Wallac Victor2 1420 multilabel counter (Perkin Elmer/Wallac) reader.

Lactate detection

Lactate was measured both on conditioned media and intracellularly on cell lysates from RIPA lysis and extraction buffer lysis. Analysis was performed after 48 h of treatment if not otherwise stated.

A Lactate Glo-Assay (Promega) was used for the detection of lactate according to the manufacturer’s instructions. The readout was performed with a Glomax®-Multi + Detection system (Promega).

Protein quantification

Protein levels were measured to normalize the results obtained from assays measuring cytokine release, Procollagen C-Peptide I release, intracellular Vitamin A, radioactive substrate oxidation, oxygen consumption rate (OCR) and extracellular acidification rate (ECAR), and lactate levels. The results were normalized to 100 mg/mL protein as measured by Pierce™ BCA Protein Assay Kit (Thermo Fisher Scientific) on cell lysates lysed by RIPA lysis and extraction buffer (Thermo Fisher Scientific). For radioactive substrate oxidation, all results were adjusted for protein content, measured by the Bio-Rad protein assay using a VICTOR™ Nivo Multilabel Plate Reader (PerkinElmer).

Radioactive substrate oxidation assay

The evaluation of glucose and oleic acid metabolism was conducted by radioactive substrate oxidation assay as previously described [43]. Following cell activation, media was replaced by D-[14C(U)]glucose (0.5 µCi/ml, 200 µM) or [1-14 C]oleic acid (OA) (0.5 µCi/ml, 100 µM) substrates during 4 h and CO2 trapping was performed. In brief, a 96-well UniFilter® microplate, activated for the capture of CO2 by the addition of 1 M NaOH, was mounted on top of the 96-well cell-cultured plate. After 4 h of trapping, the cells were washed with PBS and harvested in 0.1 M NaOH. The 14CO2 trapped in the filter and 14C remaining in cells (cell-associated (CA)) was measured by the addition of scintillation fluid (Ultima Gold XR) and counted on a 2450 MicroBeta2 scintillation counter. Cellular uptake was counted as the sum of 14CO2 and CA radioactivity measurements: CO2 + CA. Fractional oxidation was calculated as CO2/uptake.

Oxygen consumption rate (OCR), extracellular acidification rate (ECAR), and glycolysis inhibition

The OCR and ECAR were analyzed with a Seahorse XFe24 Analyzer (Agilent). Differentiating scHSCs were plated in XFe 24-well plates (Agilent) on day 5 of differentiation during the mid-differentiation passage.

The cells were washed and incubated at 37 oC in an incubator without CO2 with Seahorse XF DMEM assay medium (Matriks AS) supplemented with glucose (10 mM), glutamine (2 mM), and pyruvate (1 mM) for 1 h before the assay. The assay was performed with serial injections of oligomycin (1.5 µM, Cell Signaling Technology), FCCP (1 µM, Merck Sigma-Aldrich), and a mixture of rotenone (0.5 µM, Merck Sigma-Aldrich) and antimycin A (0.5 µM, Merck Sigma-Aldrich). The baseline was determined in the Seahorse assay media during 5 cycles of measurement with 8-minute intervals (3 min mix, 2 min wait and 3 min measurement). The OCR and ECAR values were normalized to BCA values. Unless otherwise stated, the OCR and ECAR values of each scHSC lineage were normalized to the baseline of the “Control” condition of each respective cell line – resulting in the unit “% of baseline” – to allow for comparison of the results.

2-Deoxy-D-glucose (2-DG, Merck Sigma-Aldrich) was used at a final concentration of 50 mM to inhibit glycolysis. 2-DG was injected using the Seahorse XFe24 Analyzer during 5 cycles of measurement with 8-minute intervals (3 min mix, 2 min wait and 3 min measurement) before serial injection of oligomycin, FCCP, and Rotenon/ Antimycin A to assess the effect of 2-DG on the basal mitochondrial respiration.

Holotomographic lipid analysis

Holotomographic (HT), live-cell 3D imaging for the detection of lipids was performed using the Tomocube HT-X1 system that measures refractive index as imaging contrast. HT imaging allows live identification of subcellular 3D structures based on their refractive index with unprecedented resolution [44]. Lipids have a high refractive index and are therefore easily identified and quantified in HT imaging (representative images in Fig. 2E). The live cells were imaged in Cellvis 6 or 12 Well Glass Bottom Plates (Cellvis) and images were segmented using a Machine Learning (ML) model (XGBoost). The model was trained on a sub-set of the images where cells were segmented from the background using the local sharpness of the image as computed by a local 2D laplacian operator. The ML model was trained and validated on 60% and 20% of the already segmented data respectively and the threshold for positive detection was set as per maximizing the F-score. Testing was performed on the remaining 20% of the images and led to an area under the receiver operating characteristic (AUROC) of 0.997, ensuring good detection.

Lipid droplets, which present a relatively higher refractive index compared to the rest of the image, were detected using a 3D Laplacian of Gaussian on the segmented images to detect the spherical volumes of higher refractive index which are typical of lipid droplets.

Mitochondria analysis

The mitochondria were stained for 30 min using MitoTracker™ Deep Red FM - Special Packaging (ThermoFisher Scientific) at a dilution of 1:5000 (200 nM) according to the manufacturer’s instructions. The fluorescent signal was then imaged with the Tomocube HT-X1 system.

Mitochondria were first segmented using a standard threshold whose value was computed by Otsu’s algorithm [45] on each stack of images and then skeletonized using Zhang’s algorithm [46]. Skeletonized stacks were labeled (connectivity of three) and the mean size and number of resulting clusters were quantified.

Confocal Raman spectroscopy

Cells grown on glass coverslips were treated as indicated for 48 h, fixed (4% PFA, 10 min), and kept in PBS. Raman spectra were recorded using a confocal Raman microscope (alpha300R, WITec) equipped with a 75 mW 532 nm laser (approximately 58.4–59.2 mW at the sample), a spectrometer (UHTS300S, WITec) with a 600 groove/mm grating and a thermoelectrically cooled back-illuminated charge-coupled device camera (Oxford Instruments - Andor) using a ×63/1.0 NA water-immersion objective lens (W Plan-Apochromat, Zeiss). Cells were scanned using a Large Area Scan of 200 × 200 µM with 200 points and 200 lines using an integration time of 0.5 s and Topography correction.

Pre-processing of the scans was performed in Python3 by first removing the cosmic rays using a local median filter and the baseline using an asymmetrically reweighted penalized least squares smoothing (arPLS) [47] as implemented by pybaselines [48] which was computed on the mean spectrum of each scan. Every spectrum was renormalized so that the integral of the signal over the full wavenumber range was equal to unity.

Amounts of VA and lipids were estimated by the integrals of the Raman signal in the spectral windows of [1554, 1615] cm− 1 [49, 50] and [2800, 3000] cm− 1 [51, 52] respectively (see Fig. 2D, left). The correlation between these two compounds was estimated using Pearson’s correlation coefficient across the whole scans.

Statistical analysis

Independent cell lines are treated as biological replicates, denoted “n”, and separate wells within the same differentiation and cell line are treated as technical replicates, denoted “N”.

Statistical significance was determined using the statistical analysis unpaired t-test with Welch’s correction in the GraphPad Prism (10.2.0) software. When stated, outliers were identified and removed from analysis using ROUT, Q = 1%.

Results presented as bar graphs are expressed as mean ± standard deviation (SD). When values are mentioned in-text, they are stated as mean ± SD. Boxplots visually represent the five-number summary of the data, encompassing the minimum and maximum values with the whiskers, the first and third quartiles (Q1 and Q3) of the data within the box, and the median (Q2) at the line in the box. The error bars on line graphs represent the SD unless otherwise stated. All data points used to make the bar graphs and boxplots are visualized in the figures. Data from the different cell lines are colored as follows throughout the figure: Yellow for scHSC_1, green for scHSC_2, magenta for scHSC_3, and blue for scHSC_4.

All p-values are written on the graphs. In general, p-values ≤ 0.05 were considered of interest. In enrichment analysis of mRNA seq data, FDR q-values ≤ 0.25 were considered of interest.

Figure creation

The schematic figures were created with BioRender (biorender.com). Graphs were created with GraphPad Prism (10.2.0) or using Python3 (03.09.2012).