CoChat

High-throughput human microfluidic organoid-on-a-chip platform for modeling liver diseases and screening nanotherapeutics.

Open access

The DOI matches its Crossref record; 44 of the 44 references it cites were found in a registry or an index; 12 of the 17 sources its claims cite were checked for support.

Abstract

We combined healthy human liver tissue-derived organoids with a high-throughput three-lane OrganoPlate microfluidic format to establish a perfused organoid Liver-on-Chip (HepLoC) featuring 3D luminal tubules under continuous flow... As a proof-of-concept for drug-induced liver injury, troglitazone induced dose-dependent hepatocyte injury... To model metabolic liver disease, free fatty acids triggered lipid droplet accumulation...

Overview

CoChat Summary

TL;DR

The paper builds a high-throughput human liver organoid-on-a-chip (HepLoC) by integrating healthy human liver tissue–derived organoids into a perfused OrganoPlate microfluidic system. HepLoC forms mature hepatocyte-like tubules with functional bile canaliculi and is used to model troglitazone-induced cholestasis, free-fatty-acid–induced early MAFLD features, and to screen nanotherapeutics.

Key findings

  • Organoid-derived cells self-organized into 3D luminal tubules in chip after about 1 week of dynamic culture, and FITC-dextran leakage was delayed in fully confluent tubules cultured for 6 days, indicating a functional epithelial barrier.
  • HepLoC showed stronger hepatocyte maturation than 3D organoids, with increased mature hepatocyte markers and functions: higher ALB and HNF4A expression, detectable albumin and AAT secretion, and enhanced urea synthesis and ammonia elimination, approaching primary hepatocytes.
  • HepLoC formed functional bile canaliculi: CDF accumulated between cell junctions in HepLoC but largely stayed inside LoC, and TEM showed larger tight junctions and bile canaliculi in HepLoC.
  • Troglitazone induced dose-dependent hepatocyte injury with decreased viability, increased LDH, MRP2 and ZO-1 mislocalization, impaired bile acid transport, and tight-junction disruption; free fatty acids caused lipid droplet accumulation, increased triglycerides and ROS, and upregulated TNFA and IL6 while largely preserving viability.
  • In the screening application, reference hepatotoxins and LNP components were discriminated by viability assays; DSPE-PEG 2000 was biocompatible, DOTAP chloride was hepatotoxic, and D-Lin-MC3-DMA was safe at therapeutically relevant concentrations but toxic at higher doses. Curcumin liposomes reduced lipid accumulation more effectively than free curcumin with minimal hepatotoxicity.

Methods

  • Human liver organoids were established from healthy human liver tissue, expanded in Matrigel/Ultimatrix, then seeded as single cells into the upper perfusion channels of a 3-lane OrganoPlate with collagen I in the central channel; chips were cultured under dynamic flow on an OrganoFlow system.
  • Hepatocyte differentiation used a 2-step protocol: 1 week in BMP7-supplemented expansion medium followed by 1 week in hepatocyte differentiation medium containing factors including EGF, HGF, A83-01, DAPT, dexamethasone, BMP7, and FGF19.
  • Main analyses included immunofluorescence, ELISA, urea and ammonia assays, FITC-dextran barrier testing, CDFDA-based bile canaliculi imaging, TEM, qRT-PCR, RNA-seq/GSEA, ROS staining, and PrestoBlue viability assays for drug and nanoparticle screening.

Limitations

  • The paper states that HepLoC is “primarily” parenchymal-cell based and therefore a simplified version of the complex liver microenvironment; it says future versions should add non-parenchymal cells such as hepatic stellate cells and Kupffer cells.
  • The paper states that HepLoC still does not reach primary hepatocyte levels for some functions, and that further differentiation protocols need to be developed.
  • Evident from the text: the paper uses healthy donor liver tissue and models early-stage MAFLD and cholestasis in vitro, so it does not directly test advanced disease, patient-specific disease variation, or in vivo outcomes.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1186/s12951-026-04477-0

Cites

44 of 44 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Image 1
CC BY-NC-ND · From the PDF
Fig. 1 Schematic representation of LPC organoid tubule culture against ECM gel in the chip plate. ( A ), The process of establishment of human liver organoids from healthy liver tissues, and liver organoids were identified as LGR5 positive. Bottom panel shows the growth view of organoids from single cells. ( B ), The bright view of liver organoids of Passage 3 and 15. Bar = 1000 μm. ( C ), Representative microscopic image of co-immunofluorescence staining of KI67, SOX9 and HNF4A, EPCAM and KRT7 respectively in liver organoids. Bar = 50 μm. ( D ), The view of the OrganoPlate composed of 40 microfluidic chip units positioned underneath a 360-microtiter plate. Each unit has a 3 × 3 array of wells in one microfluidic. There are three channels in each unit (top and bottom microfluidic channels and middle gel channel), ECM was loaded in the middle channel( E -S1). ( E -S2), Liver organoids were cultured for 7 days and were digested into single cells and seeded in the top channel. ( E -S3). OrganoPlate placed the side stand in the incubator to allow cell attachments, ( E -S4), OrganoPlate was cultured on the rocker with an inclination of 7° and an interval of 8 min for 7 days. T-D view: Top to down view, F-B view: Front to back view
Fig. 1 Schematic representation of LPC organoid tubule culture against ECM gel in the chip plate. ( A ), The process of establishment of human liver organoids from healthy liver tissues, and liver organoids were identified as LGR5 positive. Bottom panel shows the growth view of organoids from single cells. ( B ), The bright view of liver organoids of Passage 3 and 15. Bar = 1000 μm. ( C ), Representative microscopic image of co-immunofluorescence staining of KI67, SOX9 and HNF4A, EPCAM and KRT7 respectively in liver organoids. Bar = 50 μm. ( D ), The view of the OrganoPlate composed of 40 microfluidic chip units positioned underneath a 360-microtiter plate. Each unit has a 3 × 3 array of wells in one microfluidic. There are three channels in each unit (top and bottom microfluidic channels and middle gel channel), ECM was loaded in the middle channel( E -S1). ( E -S2), Liver organoids were cultured for 7 days and were digested into single cells and seeded in the top channel. ( E -S3). OrganoPlate placed the side stand in the incubator to allow cell attachments, ( E -S4), OrganoPlate was cultured on the rocker with an inclination of 7° and an interval of 8 min for 7 days. T-D view: Top to down view, F-B view: Front to back viewCC BY-NC-ND · From the PDF
Fig. 2 Establishing 3D lumen structure with human liver organoids-derived cells in the chips. ( A ), Representative images of the lumen formed by liver organoid-derived cells in the top channel, including upper layer and bottom layer. Scale bar = 50 μm. ( B ), Maximum intensity projection and cross sections (left panel and bottom panel) of a cellular tube, nuclei are visualized with DAPI (blue). ( C ), The 3D reconstruction of nuclei of organoid-derived cells in the chips, including a reconstruction of a cross section orthogonal (y-z) and parallel (x-z) to the sprouting direction, in which the lumen was apparent. Nuclei were visualized with 4',6-diamidino-2-phenylindole (DAPI). ( D ), Representative images of the organoids-derived cells in the chips stained against KI67 (green), nuclei counterstained with DAPI (blue). Scale bars, 50 μm. ( E ), Fluorescence microscopy images of cellular tubes that cultured for 3 days and 6 days after adding 150-kD FITC-labeled dextran. Images were taken every 5 min from 0 to 20 min, Data represent mean ± SD, N = 4
Fig. 2 Establishing 3D lumen structure with human liver organoids-derived cells in the chips. ( A ), Representative images of the lumen formed by liver organoid-derived cells in the top channel, including upper layer and bottom layer. Scale bar = 50 μm. ( B ), Maximum intensity projection and cross sections (left panel and bottom panel) of a cellular tube, nuclei are visualized with DAPI (blue). ( C ), The 3D reconstruction of nuclei of organoid-derived cells in the chips, including a reconstruction of a cross section orthogonal (y-z) and parallel (x-z) to the sprouting direction, in which the lumen was apparent. Nuclei were visualized with 4',6-diamidino-2-phenylindole (DAPI). ( D ), Representative images of the organoids-derived cells in the chips stained against KI67 (green), nuclei counterstained with DAPI (blue). Scale bars, 50 μm. ( E ), Fluorescence microscopy images of cellular tubes that cultured for 3 days and 6 days after adding 150-kD FITC-labeled dextran. Images were taken every 5 min from 0 to 20 min, Data represent mean ± SD, N = 4CC BY-NC-ND · From the PDF
Fig. 3 Differentiating organoid-derived cells to mature hepatocytes in the chips. ( A ), qPCR analyses for the expression of LGR5 , KRT19 , HNF4A , and ALB in before and after differentiation of organoids and LoC. Primary hepatocytes (PH) were used as compared. Data represent mean, N = 5 (using organoids from three different donors with five technical replicates), * p < 0.05. ( B ) Albumin secretion, AAT secretion, urea production and ammonia elimination after 24 h of culture, measured from supernatants. Data represent mean ± SD, N = 5 as above. ( C, D and E ). Representative images of LoC and HepLoC stained against LGR5 (green) and MRP2 (red), KRT19 (green) and BSEP (red), ALB (green) and HNF4A (red), respectively. Nuclei were counterstained with DAPI (blue). Scale bars, 50 μm
Fig. 3 Differentiating organoid-derived cells to mature hepatocytes in the chips. ( A ), qPCR analyses for the expression of LGR5 , KRT19 , HNF4A , and ALB in before and after differentiation of organoids and LoC. Primary hepatocytes (PH) were used as compared. Data represent mean, N = 5 (using organoids from three different donors with five technical replicates), * p < 0.05. ( B ) Albumin secretion, AAT secretion, urea production and ammonia elimination after 24 h of culture, measured from supernatants. Data represent mean ± SD, N = 5 as above. ( C, D and E ). Representative images of LoC and HepLoC stained against LGR5 (green) and MRP2 (red), KRT19 (green) and BSEP (red), ALB (green) and HNF4A (red), respectively. Nuclei were counterstained with DAPI (blue). Scale bars, 50 μmCC BY-NC-ND · From the PDF
Fig. 4 HepLoC promote hepatocyte maturing than 3D liver organoids. ( A ) Principal component analysis (PCA) of bulk RNA-seq profiles from HepLoC cultures and 3D liver organoids (Hep-Org). ( B ) Volcano plot of differentially expressed genes between HepLoC and HepOrg. Red and blue dots show genes up-regulated and down-regulated in HepLoC, respectively; representative liver-enriched genes are labelled, N = 3 using organoids from three different donors. ( C ), Tissue/cell-type-specific gene-set enrichment analysis of HepLoC and Hep-Org. ( D ), Gene Ontology (GO) enrichment analysis of differentially expressed genes in HepLoC and Hep-Org. ( E ), KEGG pathway enrichment analysis of differentially expressed genes between HepLoC and Hep-Org, colours denote functional categories. ( F ), Heatmaps showing normalized expression of representative genes from hallmark and pathway gene sets related to bile acid metabolism, fatty acid metabolism, xenobiotic and drug metabolism (cytochrome P450), biological oxidations, alcohol metabolic processes and lipid transport in HepLoC and LoC samples. Colour scale indicates relative expression level (blue, low; red, high)
Fig. 4 HepLoC promote hepatocyte maturing than 3D liver organoids. ( A ) Principal component analysis (PCA) of bulk RNA-seq profiles from HepLoC cultures and 3D liver organoids (Hep-Org). ( B ) Volcano plot of differentially expressed genes between HepLoC and HepOrg. Red and blue dots show genes up-regulated and down-regulated in HepLoC, respectively; representative liver-enriched genes are labelled, N = 3 using organoids from three different donors. ( C ), Tissue/cell-type-specific gene-set enrichment analysis of HepLoC and Hep-Org. ( D ), Gene Ontology (GO) enrichment analysis of differentially expressed genes in HepLoC and Hep-Org. ( E ), KEGG pathway enrichment analysis of differentially expressed genes between HepLoC and Hep-Org, colours denote functional categories. ( F ), Heatmaps showing normalized expression of representative genes from hallmark and pathway gene sets related to bile acid metabolism, fatty acid metabolism, xenobiotic and drug metabolism (cytochrome P450), biological oxidations, alcohol metabolic processes and lipid transport in HepLoC and LoC samples. Colour scale indicates relative expression level (blue, low; red, high)CC BY-NC-ND · From the PDF
Fig. 5 Visualization of the bile canaliculi network through CDFDA in the HepLoC. ( A ), Schematic for visualizing the bile canaliculi network by CDFDA. CDFDA enters hepatocytes via passive diffusion and is then hydrolyzed by intracellular esterases, yielding fluorescent CDF that contains a fluorophore; this product is subsequently effluxed into the bile canaliculi via the MRP2 transporter. ( B ), Time-lapse imaging of bile canaliculi network in LoC and HepLoC at 5-minute intervals for up to 30 min of CDFDA (1 µmol/L) incubation. ( C ), Representative transmission electron microscopy (TEM) images of LoC and HepLoC, Scale bars, 500 nm
Fig. 5 Visualization of the bile canaliculi network through CDFDA in the HepLoC. ( A ), Schematic for visualizing the bile canaliculi network by CDFDA. CDFDA enters hepatocytes via passive diffusion and is then hydrolyzed by intracellular esterases, yielding fluorescent CDF that contains a fluorophore; this product is subsequently effluxed into the bile canaliculi via the MRP2 transporter. ( B ), Time-lapse imaging of bile canaliculi network in LoC and HepLoC at 5-minute intervals for up to 30 min of CDFDA (1 µmol/L) incubation. ( C ), Representative transmission electron microscopy (TEM) images of LoC and HepLoC, Scale bars, 500 nmCC BY-NC-ND · From the PDF
Fig. 6 Modelling drug-induced cholestasis in LoC. (A), Schematic of drug-induced cholestasis visualised using Troglitazone. Troglitazone (TGZ) incubation results in BSEP inhibition of cells in HepLoC, which hinders the transporting of CDF into canaliculi. In addition, the derivative of TGZ metabolism, Troglitazone sulfate (TS), inhibits MRP2 transporters and disrupts CDF efflux, resulting in increased intracellular accumulation of CDF after Troglitazone exposure. (B), The cytotoxic response measured by the viability of cells and LDH was evaluated before and after the exposure of Troglitazone (TGZ-HepLoC) for 6 h by presto blue. Data represent mean ± SD, N = 5 as above. Cell viability of HepLoC was used as control. (C), Representative images of HepLoC and TGZ-HepLoC stained against MRP2 (green) and ZO-1 (red), nuclei were counterstained with DAPI (blue). Magnified panels (right) correspond to specific outlined regions (left). Scale bars, 50 μm. (D) Time-lapse imaging of bile canaliculi network in HepLoC and TGZ-HepLoC at 30 min of CDFDA (1 µmol/L) incubation. (E), Representative transmission electron microscopy images of HepLoC and TGZ-HepLoC, Scale bars, 200 nm
Fig. 6 Modelling drug-induced cholestasis in LoC. (A), Schematic of drug-induced cholestasis visualised using Troglitazone. Troglitazone (TGZ) incubation results in BSEP inhibition of cells in HepLoC, which hinders the transporting of CDF into canaliculi. In addition, the derivative of TGZ metabolism, Troglitazone sulfate (TS), inhibits MRP2 transporters and disrupts CDF efflux, resulting in increased intracellular accumulation of CDF after Troglitazone exposure. (B), The cytotoxic response measured by the viability of cells and LDH was evaluated before and after the exposure of Troglitazone (TGZ-HepLoC) for 6 h by presto blue. Data represent mean ± SD, N = 5 as above. Cell viability of HepLoC was used as control. (C), Representative images of HepLoC and TGZ-HepLoC stained against MRP2 (green) and ZO-1 (red), nuclei were counterstained with DAPI (blue). Magnified panels (right) correspond to specific outlined regions (left). Scale bars, 50 μm. (D) Time-lapse imaging of bile canaliculi network in HepLoC and TGZ-HepLoC at 30 min of CDFDA (1 µmol/L) incubation. (E), Representative transmission electron microscopy images of HepLoC and TGZ-HepLoC, Scale bars, 200 nmCC BY-NC-ND · From the PDF
Fig. 7 Modeling of early-stage MAFLD in HepLoC. (A) , Representative images of HepLoC that treated with or without a cocktail of FFA for 3 days (FFAHepLoC). Scale bar, 1000 μm. ( B ), The cytotoxic response measured by the viability of cells was evaluated in HepLoC and FFA-HepLoC by Presto blue. Cell viability of HepLoC was used as control. Data represent mean ±SD, N = 5. ( C ), Intracellular triglyceride concentration of HepLoC and FFA-HepLoC, Data represent mean ± SD, N = 5. ( D ) qPCR-based analysis of human PPARα, ACACA, FASN, TNFA and IL6 gene expression. 18 S was used as the housekeeping gene. Data represent mean ± SD. * p < 0.05; ** p < 0.01 as compared to HepLoC. ( E ), GSEA analysis of hallmark fatty acid metabolism, oxidative phosphorylation and TNFA signaling via NFκB between HepLoC and FFA-HepLoC. ( F ), Representative images of HepLoC and FFA-HepLoC stained with BODIPY 493/503 to detect lipid droplets (green); nuclei were visualized with DAPI (blue). Scale bar: 50 μm. Magnified panels (right) correspond to specific outlined regions (left). ( G ), Representative images of HepLoC and FFA-HepLoC stained to detect ROS (green); nuclei were visualized with Hoechst (blue). Scale bar: 50 μm. Magnified panels (right) correspond to specific outlined regions (left)
Fig. 7 Modeling of early-stage MAFLD in HepLoC. (A) , Representative images of HepLoC that treated with or without a cocktail of FFA for 3 days (FFAHepLoC). Scale bar, 1000 μm. ( B ), The cytotoxic response measured by the viability of cells was evaluated in HepLoC and FFA-HepLoC by Presto blue. Cell viability of HepLoC was used as control. Data represent mean ±SD, N = 5. ( C ), Intracellular triglyceride concentration of HepLoC and FFA-HepLoC, Data represent mean ± SD, N = 5. ( D ) qPCR-based analysis of human PPARα, ACACA, FASN, TNFA and IL6 gene expression. 18 S was used as the housekeeping gene. Data represent mean ± SD. * p < 0.05; ** p < 0.01 as compared to HepLoC. ( E ), GSEA analysis of hallmark fatty acid metabolism, oxidative phosphorylation and TNFA signaling via NFκB between HepLoC and FFA-HepLoC. ( F ), Representative images of HepLoC and FFA-HepLoC stained with BODIPY 493/503 to detect lipid droplets (green); nuclei were visualized with DAPI (blue). Scale bar: 50 μm. Magnified panels (right) correspond to specific outlined regions (left). ( G ), Representative images of HepLoC and FFA-HepLoC stained to detect ROS (green); nuclei were visualized with Hoechst (blue). Scale bar: 50 μm. Magnified panels (right) correspond to specific outlined regions (left)CC BY-NC-ND · From the PDF
Fig. 8 Screening putative MAFLD drugs via FFA-HepLoC. (A) Dose-response curves of HepLoC exposed for 2 days to reference hepatotoxic compounds (acetaminophen, tetracycline), candidate MAFLD therapeutics (Fenofibrate, Obeticholic acid, recombinant hFGF19) and Curcumin and Curculosomes. Cell viability was quantified by metabolic viability assay, Cell viability of 0 was used as control. Data represent mean ± SD, N = 3. (B), Representative images of HepLoC and FFA-HepLoC stained with BODIPY 493/503 to detect lipid droplets (green); nuclei were visualized with DAPI (blue). Scale bar: 50 μm. (C), Intracellular triglyceride levels in FFA-HepLoC treated with vehicle or candidate MAFLD drugs, Data represent mean ± SD, N = 3. *p < 0.05
Fig. 8 Screening putative MAFLD drugs via FFA-HepLoC. (A) Dose-response curves of HepLoC exposed for 2 days to reference hepatotoxic compounds (acetaminophen, tetracycline), candidate MAFLD therapeutics (Fenofibrate, Obeticholic acid, recombinant hFGF19) and Curcumin and Curculosomes. Cell viability was quantified by metabolic viability assay, Cell viability of 0 was used as control. Data represent mean ± SD, N = 3. (B), Representative images of HepLoC and FFA-HepLoC stained with BODIPY 493/503 to detect lipid droplets (green); nuclei were visualized with DAPI (blue). Scale bar: 50 μm. (C), Intracellular triglyceride levels in FFA-HepLoC treated with vehicle or candidate MAFLD drugs, Data represent mean ± SD, N = 3. *p < 0.05CC BY-NC-ND · From the PDF

Verification

Claims the paper makes

7 Valid · 2 Not valid · 2 Not checked · 127 Own results and interpretations

78 more claims

Create a free account to see all 138 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 8
  • It may state two separate claims as one. 49

Create a free account to see these claims.

Citations

Cites

Cited by

See all 4

Metadata

Authors
Kong, D., Yu, S., Heegsma, J., Blokzijl, T., Qin, T., Pan, Y., Blokzijl, H., de Meijer, V. E., Duan, Y., Faber, K. N., Xia, Q., & He, K.
Affiliations
  • University Medical Center Groningen (NL)
  • Shanghai Jiao Tong University (CN)
Corresponding author
  • Kong, Defu
  • Faber, Klaas Nico
  • Xia, Qiang
  • He, Kang
Venue
Journal of nanobiotechnology
Published
2026-05-02
Volume
24
Issue
1
ISSN
1477-3155 (ISSN-L)
OpenAlex
W7160001015
Cited by
4
References
44
Citations by year
2026: 4
FWCI
5.98
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Nanobiotechnology (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
  • University of Groningen research database (University of Groningen / Centre for Information Technology) (submittedVersion, cc-by-nc-nd)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Funders
  • Shanghai Natural Science Foundation (23ZR1438600)
  • Shanghai Research Center for Organ Transplantation, Category A Key Research Center of Shanghai Municipality (2022ZZ01016)
  • Dynamic Evolution and Regulatory Mechanisms of Host Immune Response Before and After Liver Transplantation
  • Special Program of the National Natural Science Foundation of China (82241221)
Topics
  • Physical Sciences › Engineering › Biomedical Engineering › 3D Printing in Biomedical Research
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Pharmacology › Drug-Induced Hepatotoxicity and Protection
  • Health Sciences › Medicine › Surgery › Tissue Engineering and Regenerative Medicine
Keywords
liver-on-a-chip, organoids, hepatocyte differentiation, drug-induced liver injury, cholestatic liver injury, non-alcoholic fatty liver disease, free fatty acids, lipid droplets, hepatotoxicity screening, nanotherapeutics, MRP2, tight junction disruption, bile canaliculi, high-throughput screening
MeSH terms
Troglitazone, Multidrug Resistance-Associated Protein 2, Microphysiological Systems, Cell Differentiation, Drug Evaluation, Preclinical, Humans, Liver, Liver Diseases, Organoids, Hepatocytes, Microfluidics, Nanomedicine, Chemical and Drug Induced Liver Injury, High-Throughput Screening Assays

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.