Genome editing of TXNIP in human pluripotent stem cells for the generation of hepatocyte-like cells and insulin-producing islet-like aggregates

Abstract

Background

Thioredoxin-interacting protein (TXNIP) plays a role in regulating endoplasmic reticulum (ER) and oxidative stress, which disrupt glucose homeostasis in diabetes. However, the impact of TXNIP deficiency on the differentiation and functionality of human stem cell-derived somatic metabolic cells remains unclear.

Methods

We used CRISPR-Cas12a genome editing to generate TXNIP-deficient (TXNIP^−/−) H1 human embryonic stem cells (H1-hESCs). These cells were differentiated into hepatocyte-like cells (HLCs) and stem-cell-derived insulin-producing islets (SC-islets). The maturation and functionality TXNIP^−/− and TXNIP^+/+ SC-islets were assessed by implantation under the kidney capsule of male or female NOD-SCID mice.

Results

TXNIP deficiency significantly increased H1-hESC proliferation without affecting pluripotency, viability, or differentiation potential into HLCs and SC-islets. Bulk RNA-sequencing of thapsigargin-treated TXNIP^−/− and TXNIP^+/+ hESCs revealed differential expression of stress-responsive genes, with enriched apoptosis-related pathways in TXNIP^+/+ cells, but minimal transcriptional changes specific to TXNIP deficiency. In HLCs, TXNIP deletion reduced albumin secretion and insulin signalling, as indicated by decreased AKT phosphorylation, while showing no differences in glycolytic activity or lipid metabolism markers. Under thapsigargin-induced ER stress, TXNIP^−/− HLCs exhibited transiently reduced eIF2α phosphorylation and lower BiP expression, suggesting compromised adaptive responses to prolonged stress. SC-islets derived from TXNIP^−/− hESCs showed comparable viability, endocrine cell composition, and cytokine responses to TXNIP^+/+ islets. Following IFNα or IFNγ treatment, STAT1 phosphorylation was increased in TXNIP^−/− SC-islets, indicating that IFN signalling remained intact despite TXNIP deficiency. Upon implantation into NOD-SCID mice, both TXNIP^−/− and TXNIP^+/+ SC-islets produced human C-peptide and responded to glucose stimulation. However, TXNIP^−/− SC-islets did not demonstrate enhanced glycaemic control or glucose-stimulated insulin secretion compared to controls.

Conclusions

Our study demonstrates that TXNIP deficiency does not improve the differentiation or functionality of HLCs and SC-islets. We present the generation and characterisation of TXNIP^−/− and TXNIP^+/+ H1-hESCs, HLCs, and SC-islets as valuable models for future studies on the role of TXNIP in metabolic cell biology.

Background

Globally, 537 million adults are living with diabetes, and the incidence is expected to increase by approximately 46% by 2045 (International Diabetes Federation, www.idf.org). Diabetes is characterised by either insulin resistance with β-cell deficiency (type 2 diabetes) or autoimmune destruction of the β cells (type 1 diabetes). Both endoplasmic reticulum (ER) and oxidative stress are key mechanisms underlying hepatocyte and β-cell dysfunction in type 2 diabetes [1, 2]. Thus, strategies aimed at enhancing β-cell function and alleviating insulin resistance hold great potential for advancing diabetes therapies.

Thioredoxin-interacting protein (TXNIP) is a modulator of oxidative signalling within the cell. TXNIP inhibits both cytosolic and mitochondrial thioredoxin (TRX), thereby promoting accumulation of reactive oxygen species (ROS), leading to oxidative stress [3]. In addition, TXNIP is an important mediator of ER stress, being upregulated downstream in response to all three branches of the unfolded protein response (UPR) [4]. TXNIP also plays a central role in regulating cytoplasmic glucose homeostasis by controlling GLUT1 glucose transporter [5]. In pancreatic β-cells, TXNIP expression increases under hyperglycaemic conditions, promoting apoptosis and contributing to diabetes progression; conversely, TXNIP deficiency protects against diabetes in animal models [6, 7]. By contrast, the role of TXNIP in hepatocyte dysfunction remains less well understood, although it is recognized as a critical regulator of glucose homeostasis and lipid metabolism in the liver [8, 9].

Recent work demonstrated that hyperglycaemia-induced persistent hypomethylation of the TXNIP promoter sustains oxidative stress and inflammation, reinforcing TXNIP’s significance as a biomarker and therapeutic target [10,11,12]. Verapamil, an antihypertensive drug, reduces TXNIP expression, rescuing β-cell apoptosis and preventing diabetes onset in obese and streptozotocin (STZ) mouse models [13]. Clinical trials suggest verapamil improves β-cell function in type 1 diabetes patients [14, 15]. However, verapamil is not a specific TXNIP inhibitor and cannot be used in patients with hypotension and ventricular dysfunction. Thus, a small molecule (SRI-37330) has recently been developed as a targeted TXNIP inhibitor, showing promising results in reversing diabetes and hepatic steatosis in mice [16].

Human pancreatic islet transplantation, a promising therapeutic strategy for diabetes, faces challenges including immunosuppressive treatment, donor islet shortage, and graft loss [17]. Human stem cell-derived insulin-producing islets (SC-islets) represent promising alternatives to overcome these limitations. Genome editing approaches, including CRISPR-Cas, may further enhance SC-islets graft survival and resilience post-transplantation [18].

Txnip deletion promotes pluripotency of mouse stem cells during fibroblast reprogramming and reduces spontaneous differentiation [19]. In contrast, TXNIP is constitutively active in human induced pluripotent stem cells (hiPSCs) and is upregulated during differentiation in a p53-independent manner [20]. This suggests a context-dependent role of TXNIP in stem cell biology. Given that diabetes involves dysfunction in both hepatic and pancreatic tissues affected by ER and oxidative stress, we hypothesized that TXNIP deficiency could enhance the differentiation and functionality of human stem cell-derived metabolic cells. In this study, we generated and characterized TXNIP^−/− and TXNIP^+/+ human embryonic stem cells (hESCs), differentiating them into hepatocyte-like cells (HLCs) and SC-islets. Investigating these two distinct metabolic lineages, enabled us to determine whether TXNIP regulates metabolic cell differentiation and function, or instead exerts cell-specific effects.

Methods

The present work has been reported in line with the ARRIVE guidelines 2.0.

Ethics statement

The experimental procedures were approved by Belgian Regulations for Animal Care, and the animal protocols underwent approval from the Commision d’Etique du Bien-Être Animal (CEBEA), Faculté de Médecine, Université Libre de Bruxelles (dossiers No. 732N and 802N). All animal experiments were performed in strict compliance with the Guide for the Care and Use of Laboratory Animals. H1-hESCs (WiCell, Madison, WI) are registered with hPSCreg for use in the European Union.

CRISPR-Cas12a-mediated TXNIP knockout

gRNAs flanking exon 3 of TXNIP gene were designed using Benchling, CRISPOR, CHOPCHOP, and Cas-Designer software. H1-hESCs were electroporated using the Neon transfection system as previously described [21]. Briefly, the Alt-R Cas12a (Cpf1) Ultra nuclease (Integrated DNA Technologies, Cat#10001272) was combined with Alt-R CRISPR-Cas12a (Cpf1) crRNA (Supplementary Table S1) targeting TXNIP to generate ribonucleoprotein editing complexes before electroporation. TXNIP gene modification was confirmed by PCR with internal and external primers (Supplementary Table S2). Single-cell clones were selected, expanded, and screened by Sanger sequencing (Eurofins Genomics) to confirm the editing site and the integrity of the non-edited wild-type alleles. Off-target effects were assessed using CRISPOR, followed by PCR amplification and Sanger sequencing (Eurofins Genomics) of top predicted sites to confirm the absence of unintended mutations. Sequencing primers are listed in Supplementary Table S2.

Stem cell culturing

Selected H1-hESC TXNIP^−/− and TXNIP^+/+ clones were thawed in mTeSR Plus medium (STEMCELL Technologies Cat#100–0276) supplemented with Y-27632 ROCK pathway inhibitor (STEMCELL Technologies, Cat#72304) and seeded into 3.5-cm dishes. Once cells reached 70–80% confluence, they were detached using 0.5 mM EDTA (Invitrogen, Cat#15575020) and passaged at a 1:6–1:10 ratio into new Matrigel-coated vessels (Corning, Cat#356231).

mRNA extraction and qPCR

Poly(A) + mRNA was extracted from H1-hESCs and cells from different stages during HLCs and SC-islets differentiation using Dynabeads mRNA DIRECT kit (Invitrogen, Cat#61012). The differentiation timeline and gene expression analysis were conducted based on an established protocol [21, 22]. The extracted mRNA was reverse transcribed with a reverse transcriptase kit (Eurogentec, Cat#RT-RTCK-03). qPCR was performed with SYBR Green reagent (Bio-Rad Laboratories, Cat#1725274) using a Bio-Rad CFX machine (Bio-Rad Laboratories). β-actin (for H1-hESCs), Cyclophilin G (for SC-islets), and TATA-Box Binding Protein (for HLCs) were used as the endogenous housekeeping gene to calculate ΔCt values, which were normalized to undifferentiated H1-hESCs to derive ΔΔCt values and relative gene expression. Oligonucleotide sequences for qPCR primers are listed in Supplementary Table S2.

Protein extraction and western blotting

Total protein lysates were prepared using Cell Lysis Buffer (Cell Signaling Technology, Cat#9803S), supplemented with Halt Protease Inhibitor Cocktail (Thermo Fisher Scientific, Cat#78442). Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, Cat#PI23227) was used to perform protein quantification. Protein lysates (10–40 μg) were separated by 8% and 12% polyacrylamide gels and were successively transferred to 0.22 μm nitrocellulose membranes (Bio-Rad Laboratories, Cat#1620112). Membranes were blocked in 5% milk buffer and incubated in primary antibodies (Supplementary Table S3) diluted in BSA-containing buffer (1X TBS, 0.3% Tween 20, 5% BSA, 0.2% NaN₃). HRP-conjugated goat-anti rabbit and goat-anti-mouse IgG secondary antibodies (Supplementary Table S4) were used for the detection of protein bands. Bands were visualized using the Amersham ImageQuant 800 Western blot imaging system (Cytiva Life Science).

Immunofluorescence

H1-hESCs, HLCs and SC-islets were seeded or differentiated in 8-well culture slides (Corning) and fixed with 4% PFA. Cells were permeabilized with 0.5% Triton X-100 in DPBS, followed by incubation with UltraVision Protein Block (Fisher Scientific Cat#15169666) to reduce nonspecific background staining. Primary antibodies (Supplementary Table S3) were diluted in 0.1% Tween in DPBS and incubated overnight at 4 °C. After washing with DPBS, secondary antibodies (Supplementary Table S4) were diluted in 0.1% Tween in DPBS and incubated for 1 h at room temperature in the dark, followed by washing with DPBS. Slides were mounted with VECTASHIELD Antifade Mounting Medium with DAPI (Vector Laboratories, Cat#VEC.H-1200) and imaged with Zeiss Observer coupled with Colibri 5 Multicolor LED Light Source (Carl Zeiss Microscopy, Gmbh). For image acquisition, ZEN 3.2 software (Carl Zeiss Microscopy, Gmbh) was used.

Karyotyping

Cells were incubated in mTeSR Plus medium supplemented with 0.1 μg/mL Colcemid (Thermo Fisher Scientific, Cat#15210040) for 3 h at 37 °C. Then cells were washed and incubated with Accutase (Capricorn Scientific, Cat#ACC-1B) to obtain single-cell suspension. Following Accutase neutralization with DMEM/F-12 (Thermo Fisher Scientific, Cat#31331093), cells were centrifuged, and the supernatant was discarded. The pellet was then treated with pre-warmed 75 mM KCl (Sigma-Aldrich, Cat#P9541) and incubated for 10 min at 37°C. After centrifugation, the pellet was fixed for 10 min with a 3:1 methanol/glacial acetic acid solution at room temperature. Karyotype analysis was performed at the Center of Human Genetics at the Université libre de Bruxelles (ULB).

Colony-forming assay

TXNIP^−/− and TXNIP^+/+ isogenic H1-hESCs were dissociated into single-cell suspension with Gentle Cell Dissociation Reagent (STEMCELL Technologies, Cat#100–0485). Cells were seeded in Matrigel-coated 6-well plates in technical triplicates in the presence of Y-27632 ROCK pathway inhibitor. Medium was changed every two days. After two weeks, the resultant colonies were fixed with 4% PFA solution containing 0.5% crystal violet. Colony number and stained area were quantified using CellProfiler [23].

Viability and SYTOX Green cell death assays

H1-hESC viability was assessed by suspending the stem cells in single cells with Gentle Cell Dissociation Reagent. Cells were stained with Trypan Blue 0.4% (Thermo Fisher Scientific, Cat# 15–250-061), and viable cells were counted using the LUNA-FL™ Dual Fluorescence Cell Counter (Logos Biosystems).

H1-hESCs cell death was assessed by staining dead cells with SYTOX Green staining. Briefly, on day 1, 10,000 H1-hESCs were seeded in Matrigel-coated 96-well plates in triplicates and in the presence of ROCK inhibitor. On day 2, the medium was refreshed, and on day 3, dead cells were stained with SYTOX Green Nucleic Acid Stain (Thermo Fisher Scientific, Cat#S7020). After measurement of induced fluorescence, cells were permeabilized with 0.2% Triton X-100 for 1 h to induce total cell death, and fluorescence was measured again to determine the maximum signal. The percentage of cell death was assessed via the following formula:

$$Cell\;death \% = \frac{{\left( {induced\;fluorescence - background\;fluorescence} \right)}}{{\left( {\max imal\;fluorescence - background\;fluorescence} \right)}} \times 100\%$$

Bulk RNA-sequencing (RNA-seq) and bioinformatic analysis

Total RNA extraction was performed with RNeasy Mini Kit (Qiagen, Cat#74104) following manufacturer’s instructions. RNA quality was assessed with 4150 TapeStation System (Agilent). cDNA synthesis, library preparation, sequencing, and alignment to the human genome were performed by the BRIGHTcore facility (Brussels, Belgium). Illumina NovaSeq 600 was used for sequencing. Data preprocessing addressed missing values, duplicate IDs, low counts (< 3 in any sample), and low variability (coefficient of variation < 0.5) to ensure high RNA-seq data quality. Normalization was performed using log₂ counts per million (CPM) with edgeR (v3.36.0). Principal Component Analysis (PCA) was used to assess sample clustering and identify batch effects. Then, batch correction was integrated into the differential expression analysis model using LIMMA (v3.50.1), incorporating "Biological Replicates" as a covariate to account for batch effects. Differential expression analysis identified genes with an adjusted p -value < 0.05 and specific log₂ fold change thresholds: |log₂ fold change|> 1.0 for downregulated (< -1.0) and upregulated (> 1.0) genes. Volcano plots were generated using GraphPad Prism (v10). Functional insights were derived through Gene Set Enrichment Analysis (GSEA) using ClusterProfiler (v4.2.0) with the GO and KEGG databases (2023 versions). The detailed results are provided as Supplementary Data sets S1 and S2. Overrepresentation analysis was performed using Enrichr with the MSigDB Hallmark database. P-values were adjusted using the Benjamini–Hochberg method to control the false discovery rate. All analyses were conducted in R (v4.1.2; R Core Team, 2021).

H1-hESCs differentiation into HLCs, albumin concentration measurement, HLC viability and treatments

CRISPR-Cas12a-edited H1-hESCs were differentiated into HLCs following a previously described protocol [24]. H1-hESCs were detached using Gentle Cell Dissociation Reagent, seeded in laminin 521-coated plates (BioLamina, Cat#LN521-05), and allowed to reach optimal confluency before the start of the differentiation protocol. Albumin concentration in the cell culture medium over differentiation was measured with the Human ALB / Serum albumin ELISA Kit (Sigma-Aldrich, Cat# RAB0603). HLCs viability was assessed by suspending HLCs in single cells with Accumax Cell Dissociation Solution (Capricorn Scientific, Cat#A7089-100ML), staining with Acridine Orange/Propidium Iodide Stain (Logos Biosystems, Cat#F23001) and counting with LUNA-FL™ Dual Fluorescence Cell Counter (Logos Biosystems).

For insulin treatment, HLCs were serum-starved for 4 h, followed by 100 nM human insulin (Sigma-Aldrich, Cat#I9278-5ML) incubation for 15 min and 1 h. For thapsigargin (Sigma-Aldrich, Cat#T9033-0.5MG) treatment, HLCs were incubated with 1 μM thapsigargin for 8 and 24 h.

Assessment of HLCs glucose metabolism dynamics by HYlight

HYlight [25, 26], a fluorescent biosensor that assesses glycolysis in real time by responding to changes in fructose 1,6-bisphosphate, was employed in HLCs. Cells were transfected with the pCS2 + _HYlight plasmid using Lipofectamine 3000 (Thermo Fisher). 1 h before imaging, cells were deprived of glucose using Seahorse XF DMEM medium (Agilent) with a pH of 7.4, 2 mM glutamine, no phenol red, and no glucose. Media was changed once again prior to the assay. Cell imaging was conducted using a Nikon Eclipse Ti2 inverted microscope, which was outfitted with a Nikon AX confocal system and a 20 × objective (NA 0.8, Plan Apo λD 20 × OFN 25 DIC N2). Cells expressing HYlight were imaged using 488 nm and 405 nm wavelengths, emission was captured from 510–530 nm and cells were maintained at 37 °C without CO₂. Images were acquired every 2 min. Image processing was performed by using NIS-Elements (Nikon) and single-cell regions of interest (ROIs) were manually selected. The fluorescence intensity at 488 and 405 channels was measured for each ROI over time, extracted, and used to calculate the excitation ratio (R = F488/F405). The fluorescence ratio was normalized using the lowest value of each data set as 0 and the maximum value as 1. In the glycolytic stress test, cells underwent imaging for 5 min, followed by sequential treatments: 10 mM Glucose (Sigma-Aldrich) for 20 min, 5 µM oligomycin (MedChem Express) for the next 15 min, and 50 mM 2-deoxy-D-glucose (2-DG) (Sigma-Aldrich) for the final 10 min. A minimum of n = 3 with 50 cells per replicate were analysed.

H1-hESCs differentiation into SC-islets and cytokine treatments

CRISPR-Cas12a-edited H1-hESCs were differentiated into SC-islets as previously described [21, 27,28,29]. Briefly, TXNIP^−/− and TXNIP^+/+ clones were dissociated into a single-cell suspension using Accutase. 2 million cells were seeded in 6-well plates and 250,000 cells in 8-wells culture slides. Cells were differentiated using a 7-stage protocol. The first four stages of the differentiation, until pancreatic progenitor stage, were conducted in monolayer. Subsequently, cells were dispersed into a single-cell suspension and 1 million cells were seeded into wells of 24-Aggrewells plates for the final stages of differentiation.

For cytokine treatments, 650 spheroids per condition were resuspended in 1 mL of Ham’s F-10, GlutaMax medium (Gibco, Cat#41550) supplemented with 0.75% bovine serum albumin fraction V fatty acid-free (Roche, Cat#1077583500), 5 μL GlutaMAX (Gibco, Cat#35050–38), and 100 U/mL penicillin–streptomycin (Gibco). Spheroids were transferred to non-adherent 24-well plates and placed on a rocker at 220 rpm inside a humidified incubator. For the pulse and chase treatments, 2000 U/mL IFNα (Peprotech) or 1000 U/mL IFNγ (Peprotech) were added to the medium for 1 h. After the pulse, the medium was replaced, and the spheroid pellets were collected at 0, 2, 5, 8, and 24 h post-treatment for protein extraction.

Mice

Mice were housed and managed in compliance with the Belgian Regulations for Animal Care, and the animal protocols underwent approval from the Commision d’Etique du Bien-Être Animal (CEBEA), Faculté de Médecine, Université Libre de Bruxelles (dossiers No. 732N and 802N). The number of mice used was decided based on own previous experiments. Mice were housed at 22 °C with ad libitum access to food and water. For anaesthesia, we used 4–5% isoflurane with a gas flow rate of 1.0–1.5 LPM and maintained using 1–3% isoflurane. Reflexes were checked before starting and during the procedure. After the surgery, a buprenorphine injection (0.05 mg/kg in PBS) was administered intramuscularly. Explanted kidneys were collected at the time of nephrectomy. Mice were euthanized using CO₂ inhalation following standard ethical guidelines.

SC-islet implantation in NOD-SCID mice

3,000 SC-islets (~ 2.25 million cells) TXNIP^−/− or TXNIP^+/+ (control) were implanted under the kidney capsule of 8–12 weeks male or female mice and allowed to mature for 16 weeks. Once SC-islets reached maturity, 2 doses of 150 µg/g of STZ were intraperitoneally injected to destroy the murine β cells. One week after the second STZ injection, the engrafted kidney was removed to induce diabetes. Human C-peptide levels were measured at week 8. Random glycaemia was measured once at weeks 8, 12, and 16, and then daily after STZ treatment until three days after diabetes was detected. All random glycaemia measurements were taken at a standardized time each day (9:00–10:00 AM) to account for diurnal variations. Glucose tolerance test (GTT) and glucose-stimulated insulin secretion (GSIS) assays were performed at weeks 12 and 16 post-implantation. For these assays, mice were fasted for 6 h and subsequently administered 2 µg/g of D(+)-Glucose (Millipore, Cat#108342) dissolved in 200 µl of DPBS via intraperitoneal injection.

Blood samples were collected from the tail tip and glycaemia was measured using a glucometer (Accu-Check Performa, Roche, Basel, Switzerland). Additionally, 10–15 drops of blood were collected in capillary EDTA tubes (Sarstedt) for the measurement of human C-peptide using the Ultrasensitive human C-peptide ELISA (Mercodia, Cat#10–1141-01).

Statistical analysis

The results are presented as mean ± standard error of the mean (SEM) unless specified otherwise. For comparison between the two groups, a Student’s t-test was used. For comparisons involving more than two groups on one independent variable, one-way ANOVA was used. Two-way ANOVA was used for comparisons between groups split by two independent variables. Multiple comparisons were corrected with Tukey’s test. Statistical analyses were performed using Prism software (GraphPad Software, Inc, La Jolla, CA, USA). Differences were regarded as statistically significant if *p < 0.05; **p < 0.01; ***p < 0.001.

Results

Generation and validation of pluripotent TXNIP ^−/− and TXNIP ^+/+ H1-hESC lines

The male human embryonic stem cell line H1 (H1-hESCs) was used to delete exon 3 of the TXNIP gene using CRISPR-Cas12a, as described previously [21] (Fig. 1A). Clone purity and genotype were validated by PCR analysis with internal (P1-P2) and external (P3-P4) primers (Fig. 1B). Sequencing of the genomic DNA from the selected TXNIP^−/− single-cell clone revealed a 241 bp deletion and a 3 bp insertion within the targeted region (Fig. 1C). The complete absence of TXNIP transcript and protein expression was confirmed by qPCR and western blot, respectively (Fig. 1D-E).

Fig. 1

Generation of TXNIP^−/− and TXNIP^+/+ H1-hESCs. A Schematic representation of the CRISPR-Cas12a-mediated gene editing strategy targeting exon 3 of the TXNIP gene in H1-hESCs along with the PCR primer binding regions. B PCR analysis using internal and external primers showing the successful deletion of exon 3 in the TXNIP^+/−, and TXNIP^−/− clones, with TXNIP^+/+ as the wild-type control. C Confirmation of the deleted exon 3 region by Sanger sequencing in selected TXNIP^−/− and TXNIP^+/+ clones from the PCR analysis. D qPCR showing absence of TXNIP mRNA expression in TXNIP^−/− H1-hESCs (n = 4, mean ± SEM, **p  < 0.01, unpaired student’s t-test). E Representative western blot and quantification confirming the absence of TXNIP protein expression in TXNIP^−/− H1-hESCs (n = 3, mean ± SEM, *< 0.05, unpaired student’s t-test). Full-length blots are presented in Supplementary Figure S5A. F Representative immunofluorescence images showing comparable expression of pluripotency markers (OCT4, NANOG, TRA-1–60, and SSEA-4) in TXNIP^−/− and TXNIP^+/+ H1-hESCs. Scale bar, 50 μm. G Karyotyping analysis indicates chromosomal integrity is maintained in TXNIP^−/− and TXNIP^+/+ H1-hESCs. H Representative immunofluorescence images showing successful differentiation of TXNIP^−/− and TXNIP^+/+ hESCs into endoderm (SOX17), mesoderm (VIM), and ectoderm (TUBB3). Scale bar, 50 μm

To ensure the pluripotency of the generated clones, immunofluorescence staining was performed, demonstrating consistent expression of key markers, including Octamer-binding transcription factor 4 (OCT4), homeobox protein nanog (NANOG), podocalyxin (TRA1-60) and stage-specific embryonic antigen-4 (SSEA4), in both TXNIP^−/− and TXNIP^+/+ H1-hESCs (Fig. 1F). Karyotype analysis further confirmed chromosomal integrity of the single-cell clones following CRISPR editing (Fig. 1G).

The differentiation potential of the generated TXNIP^−/− and TXNIP^+/+ clones was assessed by inducing differentiation into the three primary germ layers. Successful differentiation was assessed by immunofluorescence staining of endoderm marker transcription factor SRY-box 17 (SOX17), mesoderm marker vimentin (VIM), and ectoderm marker tubulin beta-3 chain (TUBB3) (Fig. 1H). To rule out off-target effects, we performed CRISPOR-based off-target analysis followed by Sanger sequencing, confirming no unintended mutations in both the TXNIP clones used in this study and an independently generated second TXNIP^−/− clone (Supplementary Figure S1A). The second clone was also validated for pluripotency and differentiation potential (Supplementary Figure S1B-F). Overall, these results show the successful generation of functionally pluripotent TXNIP^−/− and TXNIP^+/+ H1-hESCs, establishing a reliable platform for differentiation studies.

TXNIP deficiency increases H1-hESCs proliferation without major differences in thapsigargin-regulated genes

To evaluate cellular function of H1-hESCs following TXNIP loss, the colony growth was assessed. Quantification of the colony-forming assay showed increased proliferation in TXNIP-deficient cells (Fig. 2A). Furthermore, TXNIP knockout did not impact H1-hESC viability (Fig. 2B) or cell death (Fig. 2C), indicating that the increased proliferation observed in TXNIP-deficient cells was linked to the acquisition of higher proliferative potential rather than an increase in cell viability or induction of cell death. Our result is in line with the role of TXNIP observed in colony formation of reprogrammed mouse embryonic fibroblasts [19].

Fig. 2

TXNIP deficiency increases H1-hESCs proliferation but does not improve cellular function upon ER stress. A Colony-forming assay shows increased proliferation in TXNIP^−/− H1-hESCs compared to TXNIP^+/+ cells (n = 3, mean ± SEM, *p  < 0.05, unpaired student’s t-test). B Cell viability assessment reveals no significant differences between TXNIP^−/− and TXNIP^+/+ H1-hESCs (n = 6, mean ± SEM, ns > 0.05, unpaired student’s t-test). C SYTOX Green assay indicates comparable levels of cell death between TXNIP^−/− and TXNIP^+/+ H1-hESCs (n = 3, mean ± SEM, ns > 0.05, unpaired student’s t-test). D Representative western blot and quantification showing thapsigargin-induced TXNIP expression in TXNIP^+/+ cells but not in TXNIP^−/− cells (n = 3, mean ± SEM, ****p  < 0.0001, unpaired student’s t-test). Full-length blots are presented in Supplementary Figure S5B. E Venn diagrams showing the number of significantly upregulated and downregulated differentially expressed genes (DEGs, defined as log₂FC > 1 and p_adj < 0.05) under different conditions. The left diagram compares significantly downregulated genes between TXNIP^−/− and TXNIP^+/+ H1-hESCs treated with thapsigargin. The right diagram compares significantly upregulated genes between TXNIP^−/− and TXNIP^+/+ H1-hESCs treated with thapsigargin. F Volcano plots showing significantly dysregulated genes in thapsigargin-treated versus DMSO-treated cells, analysed separately for TXNIP^−/− and TXNIP^+/+ cells. DEGs are defined as log₂FC > 1 and –log₁₀(p_adj) > 1.3. G GSEA performed on thapsigargin-treated TXNIP^+/+ H1-hESCs, highlighting significant pathway enrichment using KEGG and Reactome databases (p_adj < 0.05). H GSEA performed on thapsigargin-treated TXNIP^−/− H1-hESCs, highlighting pathway enrichment using KEGG and Reactome databases (p_adj < 0.05). I Overrepresentation analysis using Enrichr of differentially upregulated genes (log₂FC > 1 and p_adj < 0.05) on thapsigargin-treated TXNIP^−/− and TXNIP^+/+ H1-hESCs, highlighting significant pathway enrichment based on MSigDB—Hallmark gene sets (The dotted vertical line represents p_adj < 0.05 to indicate statistical significance). J Volcano plots showing significantly dysregulated genes in TXNIP^−/− versus TXNIP^+/+, analysed separately for DMSO-treated and thapsigargin-treated cells. DEGs are defined as log₂FC > 1 and –log₁₀(p_adj) > 1.3

TXNIP expression was significantly upregulated in TXNIP^+/+ cells after an 8 h thapsigargin treatment (Fig. 2D), consistent with prior observations in pancreatic β-cells [30]. RNA-seq analysis confirmed TXNIP knockout and its reduced transcript levels in TXNIP^−/− cells under both DMSO and thapsigargin treatments (Supplementary Figure S2A), where incomplete degradation of nonfunctional transcripts was observed.

Both genotypes exhibited similar transcriptional responses to thapsigargin, with 1,543 and 1,682 genes upregulated in TXNIP^+/+ and TXNIP^−/− cells, respectively, and 1,303 shared differentially expressed genes (DEGs) (Fig. 2E). Key ER stress-related genes, such as MANF, EIF2AK3, and DDIT3, were induced in both genotypes (Fig. 2F), indicating an intact ER stress response. Gene set enrichment analysis (GSEA) revealed similar enrichment of ER stress pathways in both TXNIP^−/− and TXNIP^+/+ cells (Fig. 2G-H). However, apoptosis-related pathways reached significance (adjusted p  = 0.043) exclusively in TXNIP^+/+ cells and not in TXNIP^−/− cells ((adjusted p  = 0.094) (Fig. 2I). Among the genes driving this apoptotic signature, IL-6, a cytokine linked to apoptotic signalling, was significantly upregulated in TXNIP^+/+ cells compared to TXNIP^−/− cells. This is consistent with previous studies linking TXNIP to ER stress-mediated inflammasome activation and IL-1β and its downstream effector IL-6 production [31]. Details of enriched gene sets and pathways (adjusted p  < 0.05) are provided in Supplementary Dataset S1.

Minimal gene dysregulation was observed between TXNIP^−/− and TXNIP^+/+ cells under DMSO and thapsigargin conditions, with only 16 and 13 significant DEGs, respectively (Fig. 2J). Among the uniquely upregulated genes in TXNIP^−/− cells were NLRP2 and NLRP7, both linked to inflammation via IL-1β release [32, 33], and several long non-coding RNAs. Interestingly, TXNIP^−/− cells upregulated ERAS, an ESC-specific RAS isoform associated with tumour-like proliferation [34], in response to thapsigargin treatment (Fig. 2J). This may partly explain the enhanced proliferation observed in TXNIP-deficient cells. However, immunoblotting for ERK phosphorylation revealed no evidenced pathway activation, suggesting that additional post-transcriptional mechanisms may drive this phenotype (Supplementary Figure S2B). DEGs significantly upregulated or downregulated in TXNIP^−/− and TXNIP^+/+ cells following thapsigargin treatment (log₂FC ≤ 1, adjusted p  < 0.05) are listed in Supplementary Dataset S2.

These findings demonstrate that TXNIP deficiency enhances proliferation without significantly altering the transcriptional response to thapsigargin-induced ER stress. The increased ERAS expression possibly contributed to the proliferative phenotype. Importantly, TXNIP loss does not improve the ER stress response in stem cells.

TXNIP deficiency does not enhance insulin signalling or ER stress response in HLCs

To investigate the role of TXNIP in human hepatocyte differentiation and functionality, TXNIP^−/− and TXNIP^+/+ H1-hESCs were differentiated into HLCs following a multi-stage differentiation protocol (Fig. 3A, Supplementary Fig. 3). Immunofluorescence confirmed the robust expression of SOX17 at day 4 (definitive endoderm), alpha-fetoprotein (AFP) and HNF4A at day 9 (hepatoblast), and hepatocyte markers albumin (ALB) and HNF4A at day 25 (mature HLCs) (Fig. 3B). These transitions were further supported by qPCR analysis of NANOG, HNF4A, and ALB expression (Fig. 3E). Albumin release into the culture began at day 19 onwards (Fig. 3C), with TXNIP^−/− HLCs showing reduced albumin secretion compared to TXNIP^+/+ HLCs at days 23 and 25 (Fig. 3C). This reduction in albumin release suggests a potential role for TXNIP in regulating liver-specific secretory pathways in hepatocytes, without affecting cell viability (Fig. 3D). Stage-wise characterisation showed that TXNIP expression, low in undifferentiated H1-hESCs, progressively increased during differentiation, peaking in mature HLCs at days 17 and 25 (Fig. 3F). Despite this, cytoplasmic ALB levels in TXNIP^−/− and TXNIP^+/+ HLCs were comparable (Fig. 3F), indicating that reduced albumin release in TXNIP^−/− HLCs was not due to impaired synthesis but potentially linked to disruptions in secretion mechanisms.

Fig. 3

Differentiation of TXNIP^−/− and TXNIP^+/+ H1-hESCs into hepatocyte-like cells. A Timeline and schematic representation of multi-stage differentiation protocol for generating TXNIP^−/− and TXNIP^+/+ HLCs from H1-hESCs. B Representative immunofluorescence images showing the expression of stage-specific differentiation markers for definitive endoderm (SOX17), hepatoblast (HNF4A and AFP), and mature hepatocytes (HNF4A and ALB) confirming successful differentiation in TXNIP^−/− and TXNIP^+/+ HLCs. Scale bar, 50 μm. C ELISA quantification showing albumin secretion over time shows reduced secretion TXNIP^−/− HLCs at later stages in the supernatant over HLC differentiation starting at day 13. (n = 4, mean ± SEM, *p  < 0.05, ordinary two-way ANOVA, Tukey’s multiple comparison test). D Cell viability assessment shows no significant differences between TXNIP^−/− and TXNIP^+/+ HLCs. (n = 3, mean ± SEM, unpaired student’s t-test). E Relative gene expression levels of NANOG, HNF4A, and ALB at days 0, 4, 9, 17 and 25 show consistent differentiation dynamics across genotypes. TATA-box-binding protein (TBP) was used as the endogenous housekeeping gene for ΔCt calculation, and ΔΔCt values were normalized to undifferentiated TXNIP^+/+ H1-hESCs to determine relative gene expression (mean ± SEM, unpaired student’s t-test). F Representative immunoblots and quantifications showing stage-wise expression of TXNIP, ALB, BiP, and lipid metabolism markers (ACC1, FASN, CD36, and PPARγ) in TXNIP^−/− and TXNIP^+/+ H1-hESCs differentiation (n = 3–5, mean ± SEM, *p  < 0.05, **p  < 0.01, ***p  < 0.001, ****p  < 0.0001, ordinary two-way ANOVA, Tukey’s multiple comparison test). Full-length blots are presented in Supplementary Figure S5C

To explore whether TXNIP influences lipid metabolism and ER stress during HLC differentiation, we assessed the expression of key lipid metabolism-related markers, including acetyl-CoA carboxylase 1 (ACC1), fatty acid synthase (FASN), and platelet glycoprotein 4 (CD36), as well as peroxisome proliferator-activated receptor gamma (PPARγ), a regulator of lipid storage and adipogenesis. In addition, we measured the expression of binding immunoglobulin protein (BiP), an ER chaperone that serves as an indicator of ER stress. PPARγ expression was transiently elevated at the hepatoblast stage (day 9) in TXNIP deficiency but evened out as hepatoblasts matured into HLCs. For other lipid metabolism markers (ACC1, FASN, CD36) and the ER stress marker BiP, no significant difference between TXNIP^−/− and TXNIP^+/+ genotypes was observed throughout the HLC differentiation (Fig. 3F). These findings suggest that while TXNIP deficiency may transiently alter lipid regulatory signalling during early differentiation stages, it does not appear to have a lasting impact on lipid metabolism in mature HLCs.

To evaluate insulin signalling, mature TXNIP^−/− and TXNIP^+/+ HLCs were starved and stimulated with insulin. While phosphorylation of the insulin receptor (p-IR) remained unchanged, a significant decrease in protein kinase B (AKT) phosphorylation was observed at 60 min post-insulin stimulation in TXNIP^−/− HLCs (Fig. 4A). This suggests that TXNIP deficiency may affect insulin sensitivity downstream of the insulin receptor at the level of AKT activation.

Fig. 4

TXNIP deficiency in HLCs is not linked to an improved insulin and ER stress response. A Representative western blots and quantifications of insulin signalling in TXNIP^−/− and TXNIP^+/+ HLCs following insulin (100 nM) stimulation. (n = 4, mean ± SEM, *p  < 0.05, ordinary two-way ANOVA, Tukey’s multiple comparison test). Full-length blots are presented in Supplementary Figure S5D. B Representative western blots and quantifications of ER stress markers in TXNIP^−/− and TXNIP^+/+ HLCs after treatment with thapsigargin (TG) for 8 and 24 h. (n = 3–4, mean ± SEM, *p  < 0.05, **p  < 0.01, ***p  < 0.001, ****p  < 0.0001, ordinary two-way ANOVA, Tukey’s multiple comparison test). Full-length blots are presented in Supplementary Figure S6A-C. C Assessment of TXNIP^−/− and TXNIP^+/+ HLCs glycolytic dynamics. HLCs were transfected with HYlight biosensor and starved (S) before glucose stimulation (G). Oligomycin was employed to reach maximum glycolytic capacity (M). Higher fluorescent intensity correlated with increased glycolysis. Solid lines represent the mean across cells, while dots represent the mean ± SD (n = 3)

To assess ER stress, HLCs were treated with thapsigargin. TXNIP^+/+ cells showed the expected upregulation of TXNIP (Fig. 4B). Both genotypes upregulated BiP, but TXNIP^−/− HLCs displayed reduced BiP levels at 24 h, indicating the role of TXNIP in sustaining BiP expression via the ATF6 pathway. Early PERK activation (p-eIF2α) was reduced in TXNIP^−/− cells but normalized by 24 h, while elevated p-JNK levels in TXNIP^−/− cells suggested compensatory IRE1α activation. p38 phosphorylation remained unchanged between genotypes (Fig. 4B).

TXNIP deficiency has been reported to induce glycolytic energy production [19]. Given the established role of TXNIP as a critical mediator linking oxidative stress and glucose homeostasis, we also sought to explore its impact on glycolytic activity in HLCs using the HYlight biosensor. This method enables real-time tracking of glucose metabolism by monitoring fructose 1,6-bisphosphate, a key metabolite in the glycolytic pathway [26]. However, no significant differences were observed between TXNIP^−/− and TXNIP^+/+ HLCs (Fig. 4C).

In summary, TXNIP deficiency selectively impacts stress-response pathways, including BiP expression and PERK activation, without disrupting key processes in hepatocyte development or basal function. Additionally, TXNIP^−/− HLCs do not exhibit enhanced performance in insulin signalling, glucose metabolism, or stress resilience, underscoring its limited role in conferring functional advantages under these conditions.

TXNIP deficiency does not impair SC-islet differentiation

To investigate the role of TXNIP in human β-cell development and functionality, TXNIP^−/− and TXNIP^+/+ H1-hESCs were differentiated into SC-islets using a multi-stage protocol (Fig. 5A). Successful differentiation was assessed by immunofluorescence staining. Stage 1 marks the transition from pluripotency to the endodermal lineage; treated H1-hESCs were stained SOX17⁺/OCT4⁻, confirming their successful differentiation into definitive endoderm. By stage 4, cells expressed both PDX1 and NKX6.1 markers, indicating successful progression to pancreatic progenitor differentiation. From stage 4 onwards, cells were cultured in static microwells as aggregates until stage 7 to form three-dimensional β-like cell spheroids (SC-islets). SC-islets contained 25–65% insulin-positive cells, along with a small percentage of cells positive for glucagon, somatostatin and polyhormonal, as determined by immunocytochemistry (Fig. 5B). Quantitative analysis showed no significant differences in the proportions of marker-positive cell populations between TXNIP^−/− and TXNIP^+/+ SC-islets (Fig. 5C). Additionally, TXNIP deficiency did not impact cell viability at either stage 4 or stage 7 of differentiation (Fig. 5D). Cells progressed along a normal developmental pathway during differentiation, transiently expressing endocrine progenitor marker neurogenin 3 (NGN3) at Stage 5. mRNA transcript levels of NGN3 and INS were similar between the two genotypes, indicating consistent expression of endocrine markers (Fig. 5E). TXNIP expression gradually increased during β-cell differentiation peaking at stages 6 and 7 (Fig. 5F). Taken together, these data indicate that TXNIP deficiency does not alter the differentiation of H1-hESCs into SC-islets in vitro.

Fig. 5

Differentiation of TXNIP^−/− and TXNIP^+/+ H1-hESCs into SC-islets and their interferon response. A Schematic representation of the multi-stage differentiation protocol for generating TXNIP^−/− and TXNIP^+/+ SC-islets from H1-hESCs. Cells are differentiated as a 2D planar culture until stage 4 (pancreatic progenitors) and are then differentiated as aggregates in static microwells (3D). B Representative immunofluorescence images showing the expression of stage-specific differentiation markers for definitive endoderm (SOX17), pancreatic progenitors (PDX1/NKX6.1), and SC-islets (INS/GCG/SST), confirming successful stage-specific differentiation. Scale bar, 50 μm. C Quantification of the percentage of SC-islet marker-positive cell populations of insulin (INS +), glucagon (GCG +), somatostatin (SST +), and polyhormonal (PH) show comparable endocrine lineage specification in TXNIP^−/− and TXNIP^+/+ SC-islets (n = 6, mean ± SEM, unpaired student’s t-test). D Cell viability assessment at stage 4 pancreatic progenitor cells and stage 7 SC-islets shows no significant differences between TXNIP^−/− and TXNIP^+/+ (n = 6, mean ± SEM, unpaired student’s t-test). E Relative gene expression levels of NGN3 and INS at stages S4, S5, S6, and S7 show consistent differentiation dynamics across genotypes. Cyclophilin was used as the endogenous housekeeping gene for ΔCt calculation, and ΔΔCt values were normalized to undifferentiated TXNIP^+/+ H1-hESCs to determine relative gene expression (mean ± SEM, unpaired student’s t-test). F Representative western blot and quantification showing stage-wise TXNIP expression over β-like cell differentiation. (n = 3, mean ± SEM, ****p  < 0.0001, ordinary two-way ANOVA, Tukey’s multiple comparison test). Full-length blots are presented in Supplementary Figure S7A. G Representative western blot showing STAT1 signalling response in TXNIP^−/− and TXNIP^+/+ SC-islets following pulse and chase treatment with IFNα and IFNγ (NP = no pulse; n = 2). Full-length blots are presented in Supplementary Figure S7B

Given the established sensitivity of β cells to inflammatory cytokines in the diabetic context, we next sought to investigate whether TXNIP deficiency modulates the inflammatory response in SC-islets. TXNIP^−/− and TXNIP^+/+ SC-islets were treated with a 1 h pulse of cytokines IFNα or IFNγ, and the signalling response was assessed by quantifying the levels of p-STAT1 by immunoblotting [28] (Fig. 5G). TXNIP expression in control cells showed dynamic regulation, with increased levels observed at early time points (0–5 h) post-cytokine treatment, followed by a gradual decline during the intermediate and late phases (8–24 h) [35]. To investigate the temporal dynamics of STAT1 activation, p-STAT1 levels were assessed in TXNIP^−/− and TXNIP^+/+ cells at multiple time points (0 h, 2 h, 5 h, 8 h, and 24 h) following IFNα or IFNγ stimulation. During the early response phase (0–5 h), STAT1 phosphorylation was rapidly induced in TXNIP^+/+ cells, with maximal activation observed at these time points. In the sustained response phase (5–8 h), p-STAT1 levels began to decline, consistent with normal signal attenuation, and returned to near-basal levels by the late response phase (24 h) for both IFNα or IFNγ stimulation. Interestingly, in TXNIP^−/− cells, the induction of STAT1 phosphorylation by both cytokines was attenuated, with lower p-STAT1 levels observed during the early response phase (0–5 h) and intermediate response phase (8 h) compared to TXNIP^+/+ cells. However, by the late response phase (24 h), p-STAT1 levels in TXNIP^−/− cells became comparable to basal levels, similar to TXNIP^+/+ cells. In addition to p-STAT1, total STAT1 levels increased in response to IFNα or IFNγ, reaching maximal levels at 24 h, regardless of TXNIP status (Fig. 5G). These findings suggest that TXNIP is induced by interferon signalling and may contribute to the rapid and sustained phosphorylation of STAT1.

In vivo implantation of TXNIP-deficient SC-islets in immunodeficient mice does not show enhanced functional maturation

To investigate the in vivo maturation and functionality of TXNIP-deficient SC-islets, we implanted TXNIP^−/− and TXNIP^+/+ aggregates under the kidney capsule of immunodeficient NOD-SCID mice. Over 16 weeks, we tracked implanted cell performance by measuring glucose and human C-peptide levels (Fig. 6A). Glycaemic measurements taken at weeks 8, 12, and 16 post-implantations showed a progressive lowering of blood glucose from mouse (~ 6–8 mM) to human physiological levels (~ 4 mM) starting at week 12 in both female and male mice. This suggested the successful humanization of glycaemic control in the mice (Fig. 6B). We did not observe any differences in glycaemia, human C-peptide levels, or their dynamics between male and female mice implanted with TXNIP^−/− and TXNIP^+/+ SC-islets at 12- and 16-weeks post-implantation (Fig. 6C). This indicates that TXNIP deficiency does not affect the in vivo maturation of SC-islets (Fig. 6B–C). To assess SC-islet in vivo functionality, GTT and GSIS assays were performed. GTT assays showed effective glucose clearance in both male and female mice, with no significant differences between TXNIP^−/− and TXNIP^+/+ SC-islets (Fig. 6D). Similarly, GSIS assays showed an insulin response to glucose challenge, with human C-peptide peaking at 30 min, post-glucose injection in both genotypes and sexes (Fig. 6E). These results suggest that TXNIP deficiency does not influence SC-islet functionality in vivo. In addition, GTT and GSIS assays performed at week 12 post-implantation showed comparable results to those at week 16 in both male and female mice (Supplementary Figure S4A-B). After 18 weeks, murine β cells were ablated in both male and female mice using two high doses of STZ, followed by daily body weight and glycaemic measurements to monitor blood glucose levels (Fig. 6F, Supplementary Figure S4C). By week 20, the functionality of the implanted grafts was assessed through unilateral nephrectomy, which resulted in the development of hyperglycaemia in both TXNIP^−/− and TXNIP^+/+ grafted SC-islets in mice, confirming the dependence on the implanted graft for glycaemic regulation (Fig. 6F). Notably, prior to nephrectomy, blood glucose levels in male mice remained at human-equivalent levels, irrespective of TXNIP expression status (Fig. 6F). Immunohistochemical analysis of TXNIP^−/− or TXNIP^+/+ grafts in mice showed SC-islets containing cells expressing insulin, glucagon, and somatostatin, indicating successful engraftment and functionality of the implanted SC-islets (Fig. 6G). Together, these results showed that the generated TXNIP^−/− and TXNIP^+/+ SC-islets are functionally mature in vivo in both male and female mice starting from week 12 post-implantation, with both maturation and functionality being independent of TXNIP expression status.

Fig. 6

TXNIP deficiency does not enhance the in vivo functionality or maturation of SC-islets in NOD-SCID mice post-implantation. A Schematic showing TXNIP^−/− and TXNIP^+/+ SC-islet implantation and timeline of functional assessments. Streptozotocin (STZ) injection leads to murine pancreas ablation. Implanted SC-islets take over glycaemic control. B Random glycaemia measurements at weeks 8, 12, and 16 after TXNIP^−/− and TXNIP^+/+ SC-islet implantation in male and female NOD-SCID mice (♂ n = 4–7, ♀ n = 5–7, mean ± SEM). C Fasting human C-peptide measurements at weeks 12 and 16 after TXNIP^−/− and TXNIP^+/+ SC-islet implantation in male and female NOD-SCID mice (♂ n = 6–8, ♀ n = 7–10, mean ± SEM, ordinary two-way ANOVA, Tukey’s multiple comparison test). D GTT with the corresponding area under the curve (AUC) at week 16 post-implantation of TXNIP^−/− and TXNIP^+/+ SC-islet in male and female NOD-SCID mice (♂ n = 6–8, ♀ n = 6–10, mean ± SEM, unpaired student’s t-test). E GSIS with the corresponding AUC at week 16 post-implantation of TXNIP^−/− and TXNIP^+/+ SC-islet in male and female NOD-SCID mice (♂ n = 5–8, ♀ n = 4–16, mean ± SEM, unpaired student’s t-test). F Random glycaemia of implanted TXNIP^−/− and TXNIP^+/+ SC-islet male NOD-SCID mice. On the left of the graph break, measurements at weeks 8, 12 and 16 show lowering of blood glucose concentration to human levels. On the right, glycaemia is followed daily after STZ injection. After survival nephrectomy, glycaemia was measured daily until euthanasia to confirm diabetes development (n = 3–4, mean ± SEM, unpaired student’s t-test). G Immunohistochemical staining of explanted TXNIP^−/− and TXNIP^+/+ SC-islet confirms expression of insulin, glucagon, and somatostatin, indicating successful engraftment and functionality with no differences between the genotypes. Scale bar, 50 μm

Discussion

The objective of this study was to generate human TXNIP knockout stem cells for the first time and investigate their role in the differentiation and functionality of metabolic cells, exploring TXNIP as a potential therapeutic target. We developed and characterised TXNIP^−/− and TXNIP^+/+ H1-hESCs, and differentiated them into two human metabolic cell types: HLCs and SC-islets. Our findings indicate that while TXNIP deficiency influences cellular proliferation in H1-hESCs, it does not affect the differentiation of HLCs or SC-islets.

Stem cells share several features with cancer cells, such as tumorigenicity, self-renewal, pluripotency, and genetic instability. The increased colony-forming capacity observed in TXNIP-deficient stem cells underscores TXNIP established role as a tumour suppressor in cancer development, despite evidence suggesting its context-dependent role in tumorigenicity [36]. TXNIP deficiency induced pluripotency and differentiation in mouse stem cells through glucose-mediated histone acetylation, increased glycolysis, and elevated OCT4 activity [19]. However, murine Txnip knockout stem cells produced smaller teratomas than controls when injected in immunodeficient mice [19]. Further studies are warranted to explore the effects of TXNIP on human stem cell pluripotency, teratoma formation, and characterisation. Interestingly, RNA-seq analysis revealed that TXNIP deficiency caused minimal transcriptional changes under both basal and thapsigargin-induced ER stress conditions, suggesting that its role in regulating stem cell proliferation may involve post-transcriptional mechanisms. Notably, the upregulation of ERAS, an ESC-specific RAS isoform linked to tumour-like proliferation, in TXNIP^−/− cells provides a potential explanation for their increased proliferation. However, the lack of MAPK activation, as indicated by unchanged ERK phosphorylation, implies that alternative signalling pathways may compensate for TXNIP loss.

Despite comparable transcriptional responses to ER stress across genotypes, pathway-specific differences emerged. Thapsigargin-treated TXNIP^+/+ cells showed enrichment of apoptosis-related pathways, including upregulation of IL-6, a cytokine linked to inflammasome activation and cell death [4]. In contrast, TXNIP^−/− cells displayed reduced apoptotic signalling, potentially contributing to their enhanced proliferative capacity. These results highlight the complexity of TXNIP’s regulatory role in stem cells and underscore the need for further investigation into how it modulates growth and stress responses.

In the context of hepatocyte differentiation, TXNIP deficiency did not impair the development of mature HLCs. However, reduced albumin secretion in TXNIP^−/− HLCs during late differentiation stages, despite comparable intracellular synthesis, suggests a role for TXNIP in hepatocyte-specific secretion pathways. TXNIP expression progressively increased during differentiation, consistent with its potential role in maintaining hepatic functionality. Interestingly, TXNIP-deficient cells showed transient upregulation of PPARγ at the hepatoblast stage (day 9), a master regulator of lipid metabolism often linked to fat accumulation in obesity and diabetes [37]. TXNIP has been reported as a target of PPARγ in melanoma cells [38], suggesting a potential regulatory interplay. However, our results showed that TXNIP-dependent PPARγ upregulation was limited to the hepatoblast stage, with levels normalizing in mature hepatocyte-like cells. While PPARγ’s role in hepatoblasts remains understudied, investigating its regulatory interactions during hepatocyte differentiation could reveal new insights into lipid metabolism and potential long-term effects in mature hepatocytes. Functional assessments of insulin signalling revealed reduced AKT phosphorylation in TXNIP^−/− HLCs following insulin stimulation. Additionally, lower BiP expression and delayed eIF2α phosphorylation during thapsigargin-induced stress suggest a role TXNIP role in maintaining ER stress homeostasis. The compensatory increase in p-JNK in TXNIP^−/− HLCs suggests activation of alternative IRE1α-mediated stress pathways, but without functional advantages. These findings emphasize that TXNIP deficiency does not hinder hepatocyte differentiation but can affect secretory, insulin signalling, and stress-response mechanisms.

During the differentiation of H1-hESCs into SC-islets, the absence of TXNIP did not impact the cell commitment to insulin-, glucagon-, and somatostatin-positive cells, nor did it affect cell viability, consistent with observations in HLCs differentiation. In β-cell biology, TXNIP serves as a key signalling hub between ER stress and inflammation. TXNIP was found to be induced by ER stress [30]. Moreover, TXNIP expression is upregulated by IFNγ in rodent INS-1 cells and human pancreatic islets [35]. In our study, cytokine-mediated inflammatory stimulation by IFNα triggered robust and sustained STAT1 activation in TXNIP^+/+ SC-islets, consistent with its role in β-cell inflammatory signalling [39]. Although, STAT1 phosphorylation occurred during early and intermediate response phases in TXNIP^-/- SC-islets, the attenuated phosphorylation levels suggest potential modulation of inflammatory pathways. Future research is necessary to clarify the functional implications of these observations, particularly in the context of diabetes-related inflammation.

Previous studies in rodent β-cell lines and mouse models have demonstrated that reducing or deleting TXNIP enhances GSIS and improves glucose homeostasis [40,41,42]. However, in vitro stem cell-derived β-like cells may not fully recapitulate the TXNIP expression patterns or glucose responsiveness of mature islets. Recent work indicates that high-glucose-induced TXNIP expression in stem cell-derived β-like cells is only about 1.5–2-fold [43], compared with ~10-fold in mature primary human islets [42, 44, 45]. Thus, TXNIP^−/− and TXNIP^+/+ SC-islets were implanted in vivo to allow full maturation [46]. We hypothesized that TXNIP deficiency could alleviate implantation-related stress and enhance SC-islet maturation in vivo. However, our implantation studies revealed that both TXNIP^−/− and TXNIP^+/+ SC-islets successfully engrafted in NOD-SCID mice, with comparable glycaemic control, human C-peptide levels, and GSIS. These findings indicate that TXNIP is not essential for SC-islet functionality or adaptation post-implantation.

Overall, while TXNIP influences inflammatory signalling in vitro, its deficiency does not enhance SC-islet maturation or glucose homeostasis in vivo. These results highlight the context-dependent roles of TXNIP in β-cell biology and suggest its functional redundancy in SC-islet differentiation and implantation under non-diabetic conditions.

One important limitation of our study is the lack of evaluation of TXNIP deficiency in embryonic stem cells with different genetic backgrounds or in patient-derived human-induced pluripotent stem cells. In the present study, we used immunodeficient NOD-SCID mice to study the intrinsic effects of TXNIP deficiency on the function and maturation of SC-islets, removing confounding factors from immune-mediated graft rejection. These mice do not recapitulate human immune responses, future work should incorporate clinically relevant models such as immunosuppressed immunocompetent NOD mice or humanized mouse models (e.g., Hu-PBL-NSG-MHC^null) [47]. These investigations would require comprehensive screening and analysis, which extend beyond the scope of this work.

Conclusions

The present study demonstrated that TXNIP deficiency enhances stem cell proliferation but does not significantly improve the differentiation or functional maturation of HLCs or SC-islets under the conditions tested. We have successfully established the first TXNIP-deficient H1-hESC lines for generating these two metabolic cell types and characterizing their stress responses. Our lines offer a platform for future in-depth mechanistic investigations of TXNIP in oxidative stress, ER stress, inflammasome activation, and glucose and lipid homeostasis. They can also serve as a robust system to explore gene–gene interactions, where TXNIP deficiency could be combined with genetic modifications targeting other metabolic regulators (e.g., the rescue of β-cell dysfunction in Wfs1-deficient mice upon TXNIP deletion [48]. Additionally, these lines can facilitate drug-screening efforts, offering valuable negative controls to test the specificity and efficacy of novel TXNIP-targeting compounds (e.g., SRI-37330 [16]). Moreover, our lines can also allow the study of epigenetic mechanisms involved in metabolic memory, particularly regarding TXNIP promoter methylation patterns and their long-term effects on oxidative stress, inflammation, and metabolic function (11).