A broad kinase inhibitor library live cell imaging screen using liver, kidney and cardiac hiPSC-ICAM1-eGFP reporter lineages identifies liabilities for TNFα pathway modulation

Abstract

Kinase inhibitors (KIs) are an increasing class of drugs targeting protein kinase activity in various pathologies, but predominantly in cancer and autoimmune diseases. Despite successful clinical applications, one of the challenges that remains is their safety, since many KIs lack selectivity and inhibit multi kinases or block pathways that are also important for homeostasis of healthy cells. This may lead to adverse reactions in several critical target organs of toxicity, including heart, liver and kidney. Inflammatory cytokine-mediated intracellular signaling is critical for the control of cell survival and adverse outcome and are modulated by diverse kinases. Likewise, inhibition of kinase activities by on- or off-target effects of KIs may contribute to adverse effects. In this study we systematically screened 760 different KIs targeting the majority of the kinome for their ability to affect TNFα-mediated expression of ICAM1 in hiPSC-derived renal proximal tubule-like cells (PTLCs), cardiomyocytes-like cells (CMLC) and hepatocyte-like cells (HLC). Palbociclib (CDK4/6), miltefosine (PI3K/AKT/mTOR), gilteritinib (FLT3/AXL), and erdafitinib (FGFR) led to increased expression of the ICAM1-eGFP reporter and the release of chemokines. Kinase inhibitor activity data from the ChEMBL database indicated off-target kinase inhibitor activity associated with enhanced ICAM1-eGFP expression. Finally, clinically relevant KIs that enhanced ICAM1-eGFP expression in PTLCs were found to be associated with renal adverse events in patients. Our overall findings support the application of imaging-based hiPSC-derived ICAM1-eGFP reporter covering different differentiated cell lineages to identify liabilities of novel kinase inhibitors to modulate TNFα pathways.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00204-025-04226-6.

Introduction

Dysregulation of protein kinase activity is associated with many diseases and contributes to the development of cancer and neurological, cardiovascular and autoimmune disorders. Therefore, targeting the kinome via kinase inhibitors (KIs) has become a critical therapeutic strategy (Cohen et al. 2021). About 30% of the drug development efforts are focusing on kinase inhibitors and a total of 72 KIs were approved in 2022 (Roskoski 2023). Despite the great potential of these small molecule KIs, it remains challenging to develop selective candidates that are effective and safe (P. Y. Lee et al. 2023). A significant number of KIs exhibit multi kinase inhibition, leading to increased liability to adverse reactions that can be of mild to life threatening nature and can affect numerous target organs (Shyam Sunder et al. 2023). The heart, liver and kidney are most frequently affected by on- or off-target adverse effects (Xiong et al. 2022; Viganò et al. 2023; Grela-Wojewoda et al. 2022). Immune responses that are controlled by inflammatory cytokine signaling are critical in various target organ toxicities (Hutchins et al. 2024; Woolbright and Jaeschke 2018; Volarevic et al. 2019). Given the critical importance of various kinases in controlling cytokine signaling, inflammatory signaling is at risk for modulation by KIs, in particular for those KIs that are less specific (Shyam Sunder et al. 2023). So far, a systematic evaluation of the effect of KIs on cytokine signaling in different target organs is missing.

TNFα signaling regulates a broad range of biological activities, including cell proliferation, inflammation and cell death. TNFα interacts with the cell surface receptors TNFR1 and TNFR2 and triggers receptor trimerization leading to an associated signaling complex consisting of TRADD, TRAF2, c-IAPs and RIPK1 (Webster and Vucic 2020b). Subsequent engagement of the LUBAC complex, consisting of SHARPIN, HOIL-1L and HOIP, leads to linear ubiquitination of LUBAC, RIPK1, TNFR1 and NF-κB essential modulator (NEMO). Thereafter, a signaling kinase complex is formed consisting of the kinases IKKα, IKKβ and adaptor protein NEMO, and the ubiquitin binding proteins TAK1 and TAB2/3, leading to phosphorylation and degradation of the inhibitor of the κB (IκB) kinase (IKK) complex (Webster and Vucic 2020b).This enables translocation of the transcription factor Nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) to the nucleus. In addition, TAK1 phosphorylates mitogen-activated protein kinase kinases (MAPKKs) leading to activation of the JUN N-terminal kinase (JNK) and the p38 pathway (Brenner et al. 2015). All these pathways contribute to the induction of downstream targets involved in inflammation, and pro-survival signaling (Webster and Vucic 2020a). Expression of cytokines, chemokines and cell adhesion molecules, including intracellular adhesion molecule 1 (ICAM1), enhances the recruitment of pro-inflammatory immune cells (Dustin et al. 1986; Singh et al. 2023). Sustained inflammatory signaling can also lead to the induction of cell death via apoptosis, necrosis and necroptosis and thus contribute to tissue injury (van Loo and Bertrand 2023; Yang et al. 2015). TNFα signaling, and in a broader sense inflammation, plays a pivotal role in drug-induced organ toxicity (Campana et al. 2021; Miller et al. 2010; Fredriksson et al. 2011; Giustarini et al. 2020) and liver regeneration upon injury (Böhm et al. 2010). In addition, inflammation has been shown to play a role in doxorubicin-induced cardiotoxicity and cisplatin-induced renal proximal-tubule toxicity (L. Wang et al. 2016; Benedetti et al. 2013a). In the context of KI toxicity, involvement of inflammatory signaling in the mechanism of toxicity of the tyrosine KIs afatinib, ponatinib and sorafenib have been described (Huan Wang et al. 2023). Furthermore, VEGF inhibitors and other KIs (e.g. ibrutinib and belvarafenib) have been linked to acute interstitial nephritis (Estrada et al. 2019; Markóth et al. 2021; Abu Amer, Avila-Casado, and Kitchlu 2023). Lastly, the TNFα signaling pathway is dependent on the activity of various kinases, including RIPKs (Cuny and Degterev 2021), IKKs (Antonia, Hagan, and Baldwin 2021), MAPKs (Canovas and Nebreda 2021) and could therefore be impacted by KIs via on- or off-target pharmacology. Therefore, KIs may have a liability for modulating the TNFα signaling pathways in normal cells representing different critical organs.

Here we performed a systematic screening of 760 pre-clinical and clinical relevant KIs that are covering a majority of the kinome on the modulation of the TNFα signaling pathway. We used human-induced pluripotent stem cells (hiPSC) that expressed an endogenously tagged ICAM1-eGFP reporter and were differentiated in hepatocyte-like cells (HLC) (Boon et al. 2020), cardiomyocyte-like cells (CMLC) (Van Den Berg et al. 2014) and proximal tubule-like cells (PTLC) (Chandrasekaran et al. 2021). Sixteen KIs were identified as enhancers of ICAM1 expression particular in PTLCs, but not HLC and CMLC. Importantly, we were able to associate twelve of these KIs with observed (inflammatory) renal adverse events in clinical trials. Furthermore, we found that eight of these KIs also enhanced pro-inflammatory CXCL8 and CCL20 expression in PTLC. We therefore suggest that our hiPSC screening model is a relevant pre-clinical tool to identify TNFα pathway modulating agents that might indicate potential adverse events.

Materials and methods

Reagents

All growth factors for hiPSC differentiations were obtained from Peprotech. CHIR99021 (S1263) and and TTNPB (S4627) were purchased at SelleckChem. XAV-939 (3748) was obtained from Tocris.

Cell culture and differentiations

The hiPSC Sigma HC3x wildtype (https://hpscreg.eu/cell-line/SIGi001-A-23) and ICAM1-eGFP (https://hpscreg.eu/cell-line/SIGi001-A-24) lines were cultured on Matrigel (Corning) coated plates in mTesR1 medium (Stem Cell Technologies, Vancouver, Canada) and medium was refreshed daily. Cells were split twice a week using ReleSR (Stem Cell Technologies). The hiPSC line with doxycycline-inducible hepatic transcription factors HNF1A, PROX1 and FOXA3 was a kind gift from the Verfaillie laboratory, Leuven, Belgium (Boon et al. 2020). hiPSC differentiation towards hepatocyte-like cells (HLCs), cardiomyocyte-like cells (CMLCs) and proximal tubule-like cells (PTLCs) was performed as previously described (Van Den Berg et al. 2014; Chandrasekaran et al. 2021) (Fig. 1A). For HLC differentiation, 9000 cells were directly plated on SCREENSTAR 96 well black imaging plates (655866, Greiner Bio one). To generate CMLCs, hiPSC were initially plated at a density of 200,000 cells/12 well and subsequently, 150.000 cells/96well were reseeded on day 14. 3.0·10⁶ cells per T75 were seeded for PTLC differentiation and were replated on day 16 at a density of 52,000 cells per 96-well of a SCREENSTAR plate to enable confocal imaging.

Fig. 1.

Characterization of differentiated iPSCs towards hepatocyte, cardiomyocyte and proximal tubular like cells. A Overview of differentiation schemes for hepatocyte-like cells (HLCs), cardiomyocyte-like cells (CMLCs) and proximal tubular-like cells (PTLCs) B mRNA expression levels of cell type specific markers. Data is log2 counts per million normalized. Asterisks indicate significant (fdr p-adjusted ≤ 0.01) expression compared to towards induced pluripotent stem cells (hiPSC) control. C mRNA expression levels of TNF signaling pathway related (target) genes. Data is log2 counts per million normalized. D Kinome tree showing the expression of various kinases per cell type. Blue indicates undifferentiated iPSCs, orange indicate HLCs, red indicates CMLCs and green indicates PTLCs. Size of the dots indicate the log2 normalized counts of each kinase gene. See supplementary Table 1 for the normalized count data

TempO-Seq analysis

After complete differentiation, cells were treated with 0.1% PBS or indicated TNFα concentrations for a period of 24 h. Thereafter, cells were lysed using 1× TempO-Seq buffer for 15 min at room temperature, followed by storage at -80°C. Whole transcriptome gene expression data was acquired using targeted RNA sequencing technology (BioSpyder Technologies, Inc., Carlsbad, CA, USA). Raw counts were normalized in R using the DEeSeq2 package, followed by log2 fold change normalization (Love et al. 2014). Genes were considered differentially expressed if the false discovery adjusted p value ≤ 0.05.

High-throughput screening of kinase inhibitor library

CMLCs and PTLCs were pre-stained with 100 ng/mL Hoechst₃₃₃₄₂ for a period of 1 h. Next, medium was removed and samples were exposed up to 10 μM of the kinase inhibitor library (L1200, SelleckChem, version June 2019). 100 nM propidium iodide (P1304MP, ThermoFisher) was added to the exposure medium, to visualize cell death induction. After 8 h of incubation, cells were challenged with TNFα at the indicated concentrations for a period of 16 h. For the last two hours of the TNFα challenge, nuclei of HLCs were stained with 100 ng/mL Hoechst₃₃₃₄₂. Then, cells were imaged on a Nikon Eclipse Ti2 LIPSI microscope, equipped with a 408, 488, 561 and 633 nm laser, using four images per well with an average of 1034 ± 278, 1014 ± 198, 765 ± 149 cells/image for PTLCs, CMLCs and HLCs, respectively. After 48 h, were imaged again to follow-up on further increase of the ICAM1-eGFP signal and cell death induction. The primary screen consisted of one replicate per hiPSC lineage. The secondary screen in PTLCs was performed using three independent biological replicates.

Image analysis and hit selection

Nuclei were segmented using watershed-masked clustering, as described previously (Yan and Verbeek 2012). Segmented nuclei and the other images were then quantified using CellProfiler 4.2.1. The integrated intensity eGFP and area per object of the nuclei and PI positive staining were calculated. To calculate the fraction PI positive cells, the PI positive area was divided by the nuclear area on a cell-by-cell basis. A cell was considered PI positive if 10% of the nuclear area was covered by PI staining.

For the eGFP intensity hit selection, first the samples where excessive (> 35%) cell death was observed were removed from the primary analysis. Compounds exceeding this threshold were included in the secondary screen. Then, a LOESS regression was applied on the TNFα dose–response curve to determine the threshold for ICAM1 upregulation, compared to standard TNF conditions. The threshold was calculated per cell type and set at the level where TNF_standard + 95% CI ≤ TNF_dose – 95% CI, where standard indicates the used concentration for the TNFα challenge in all samples. All compounds exceeding the threshold were considered an enhanced hit. For downregulating compounds, a similar threshold was set at TNF_dose – 95% CI ≤ TNF_standard + 95% CI. For the validation screen, a similar approach was chosen. Here, all hit compounds were screened in a dose–response between 0.1 nM and 10 μM. Hit candidates were considered validated if the threshold was exceeded at any concentration, independent of their potency.

LegendPlex assay

Upon hiPSC differentiation, cells were exposed to 10 µM of the selected kinase inhibitors. Eight hours post treatment, cells were challenged with TNFα at indicated concentrations. After sixteen hours, supernatant was collected and spun down to remove cell debris and stored at -80 ℃ until further usage. Chemokine secretion was determined using the LEGENDplex™ HU Proinflam. Chemokine Panel 1 (740985, Biolegend) according to manufacturer’s instructions. Samples were diluted twice and chemokine concentrations were measured on an Aurora spectral flow cytometer equipped with three lasers and a plate loader (Cytek). Statistics were performed using a student’s t-test. P-values were adjusted (p-adj) using the false discovery rate. Chemokine secretion was considered significantly changed if p-adj ≤ 0.05.

Adverse event data

We used ChEMBL data, version 33, to normalize the kinase inhibitors and to generate a compound dictionary. We downloaded the data from Clinical Trials from the Clinical Trials Transformation Initiative (CTTI, https://aact.ctti-clinicaltrials.org/download) on August 2, 2023. We parsed the fields “name” and “description” from the “interventions.txt” file, and the brief_title, and official_title from the file studies.txt to keep clinical trials in which there were mentions of the drugs or their synonyms. Reported events (reported_events.txt) were extracted from clinical trials mentioning the kinase inhibitors and kept the clinical trials in which at least one subject was affected. We used the category organ_system to filter ADRs belonging to “Renal and urinary disorders”. The data was downloaded from the FDA Adverse Event Reporting System (FAERS, https://open.fda.gov/data/downloads/) on August 2, 2023. We kept reports where the drug was reported as Primary Suspect Drug and matched the names of the drugs to the dictionary generated using ChEMBL. We normalized the names of the adverse events to MedDRA codes using the Unified Medical Language System Metathesaurus. We used the semantic relationships to extract also the codes corresponding to the category “Renal”.

Results

Human iPSC-derived PTLCs, CMLCs and HLCs show differences in the expression levels of protein kinases

To gain insight in the applicability of an hiPSC-based screening system for kinase inhibitor safety assessment, we first characterized the different hiPSC-derived cell lineages. Induced pluripotent stem cells were differentiated to PTLCs, CMLCs and HLCs following previously established (Boon et al. 2020; Chandrasekaran et al. 2021; Van Den Berg et al. 2014) protocols (Fig. 1A) and RNA expression was analyzed by targeted whole transcriptome sequencing. First, we analyzed cell type specific markers compared to undifferentiated hiPSCs and determined that most markers were significantly (fdr p-adj ≤ 0.01) expressed in their specific cell lineage, except for AQP1 in PTLCs (fdr p-adj = 0.096; Fig. 1B). Next, we compared the gene expression of signaling molecules that are downstream of TNFR1, with a focus on genes enriched for the canonical NF-κB pathway according to the DoRoThea (Garcia-Alonso et al. 2019) database (Fig. 1C). The relative expression of genes involved in the TNFα signaling pathway (Fig. 1C, upper panel) was similar between all cell types, independent of TNFα stimulation. The expression of TNFα pathway NF-κB target genes was generally higher in HLCs compared to CMLCs and PTLCs (Fig. 1C, lower panel). The latter two lineages showed similar expression levels for most of the target genes. Given our focus on kinase inhibitors, we also examined the expression of the kinome in the three different lineages. We extracted the expression data of 536 human kinases (Supp. table 1) and projected the expression of these kinases on a kinome map (Manning et al. 2002; Eid et al. 2017). Although the expression levels of various kinases are different between cell types, in general, similar kinases were expressed (Fig. 1C).

ICAM1 reporter can be used as read-out for TNF target gene expression in all cell types

To monitor any enhancement of TNFα target gene expression that can be linked to adversity upon KI exposure and TNFα challenge, we used fluorescent reporters for TNFα-induced activation of the NF-κB pathway. The NF-κB target genes ICAM1 and VCAM1 encode surface glycoproteins expressed under inflammatory conditions and act as a major regulators of leukocyte attractions (Singh et al. 2023). For both targets, an eGFP reporter was generated (Supp. Figure 1 and Supp. Table 2) in hiPSCs using CRISPR-Cas9 assisted tagging of the endogenous proteins and both reporters were differentiated in PTLCs, CMLCs and HLCs, followed by 24 h exposure to TNFα and subsequent confocal imaging. In all three cell types, the ICAM1-eGFP reporter was expressed at higher levels than the VCAM1-eGFP reporter (Fig. 2A–C) and the VCAM1-eGFP reporter did not show a dose–response in CMLCs and HLCs (Fig. 2B, C). Importantly, most inhibitors that target the IKK complex upstream of NF-κB activation were able to block ICAM1-eGFP induction by TNFα in all three cell lineages (Supp. Fig. S2). For further studies we used the ICAM1-eGFP reporter to screen kinase inhibitors for their effects on TNFα signaling.

Fig. 2.

TNFα-induced response in iPSC ICAM1-eGFP and VCAM1-eGFP reporters. Normalized ICAM1-eGFP and VCAM1-eGFP expression of PTLCs (A), CMLCs (B) and HLCs (C) after TNFα stimulation for 24 h with indicated concentrations. The average value ± standard deviation (grey area) is shown for 3 independent biological replicates. Scale bar: 50 µm

Proximal tubular cells are most susceptible towards KI-induced changes of ICAM1-eGFP expression

760 kinase inhibitors were screened for their potential to interfere with TNFα-induced ICAM1 expression in the three different lineages (Fig. 3A). Upon differentiation, cells were treated for eight hours with each individual KI, followed by a TNFα challenge. Forty-eight hours after KI exposure, eGFP levels were measured. Thresholds for hit identification were then calculated based on point-of-departure approach as used previously (Vlasveld et al. 2024). Here, we have used the dose–response curve of TNFα treated cells and calculated the PoD of the ICAM1-eGFP expression in the various lineages to identify compounds that reduce TNF-target gene levels (Fig. 3B, dashed red line). Most of the KIs that blocked ICAM1-eGFP expression overlapped between the three lineages (Fig. 3C, Supplemental Table S3). Next, we checked the intended targets of the various inhibitors and their stage in clinical development based on the manufacturer’s compound annotations and ChemBL data (Fig. 3D, E). The top three most frequently identified intended targets for these compounds were related to protein tyrosine kinase signaling, PI3K/AKT/mTOR and angiogenesis pathway signaling, among other pathways (Fig. 3D). For these three respective pathways, ten, six and seven inhibitors already entered clinical trials (Fig. 3E).

Fig. 3.

High-throughput screening of kinase inhibitors that repress ICAM1-eGFP expression in HLC, CMLC and PTLC lineages. A Flowchart of the screen. B Point-of-departure (PoD) calculations per reporter. Cells were treated with indicated concentrations of TNFα. Thereafter, a loess curve-fit was performed with estimated standard errors between the lowest and highest TNFα concentration. The black dashed vertical line indicates the used TNFα concentration per cell type. The PoD (i.e. threshold for reduced ICAM1-eGFP expression) was set at the point where the predicted ICAM1-eGFP expression – standard error was larger than the ICAM1-eGFP + standard error at the used TNFα concentration. The PoD is indicated by the horizontal red dashed line. C Venn diagram showing the identified hits in the primary screen for each iPSC lineage. D Pathways that are targeted by compounds identified in the primary screen. E Clinical phase of the identified kinase inhibitors in the primary screen

Next, we also assessed enhancers of ICAM1-eGFP expression using the same approach. Again, the threshold for TNF target gene upregulation was calculated based on the point-of-departure approach, now for enhancement compared to control conditions (Fig. 4A, dashed red line). Interestingly, here little overlap of KIs that significantly enhanced ICAM1-eGFP expression was found between different cell types (Fig. 4B, Supplemental Table S3). Targets that enhanced ICAM1-eGFP induction included PI3K/mTOR AKT signaling (e.g. bimiralisib, GDC-0084, pictisilib) and cell cycle inhibitors (e.g. palbociclib isethionate, SU9516) (Fig. 4C). Remarkably, half of the identified PI3K/AKT/mTOR (12/24) and cell cycle inhibitors (6/12) has already entered clinical trials (Fig. 4D).

Fig. 4.

High-throughput screening of kinase inhibitors that enhance ICAM1-eGFP expression in HLC, CMLC and PTLC lineages. A Point-of-departure (PoD) calculations per lineage. Cells were treated with indicated concentrations of TNFα. Thereafter, a loess curve-fit was performed with estimated standard errors between the lowest and highest TNFα concentration. The black dashed vertical line indicates the used TNFα concentration per cell type. The PoD (i.e. threshold for enhanced ICAM1-eGFP expression) was set at the point where the predicted ICAM1-eGFP expression – standard error was larger than the ICAM1-eGFP + standard error at the used TNFα concentration. The PoD is indicated by the horizontal red dashed line. B Venn diagram showing the identified hits in the primary screen for each hiPSC lineage. C Pathways that are targeted by compounds identified in the primary screen. D Clinical phase of the identified kinase inhibitors in the primary screen. E Concentration–response curves of validated hits in PTLC-ICAM1-eGFP reporter cells. Data shown indicates fold-change normalized to 0.1% DMSO + 2.5 ng/mL TNFα, indicated by the horizontal black line. F Example images of PTLC samples of control conditions and after KI + TNF exposure (data shown in E). For the compounds, example images are shown for samples treated using 10 μM of the indicated compound and 2.5 ng/mL TNF. Scale bar: 50 µm. Data is shown as mean ± standard deviation (grey area) for three independent biological replicates

Since PTLCs were most sensitive for identification of the effects of KIs, possibly due to relatively high sensitivity towards TNFα, we further focused on this cell type and performed a secondary screen using a concentration–response analysis to address the potency of the different KIs. We particular focused on the enhancers of the ICAM1-eGFP induction that have entered clinical trials, since this could impact on stimulation of inflammatory responses (Kellum et al. 2021). In the secondary screen, a significant enhancement of ICAM1-eGFP expression for 16/68 of the candidate hits was found and these compounds were considered validated hits (Fig. 4E, F); for most of these KIs enhancement of ICAM1 expression was observed between 1 and 10 μM. For comparison of our screening concentrations to observed human maximum blood plasma concentrations (Cmax), we searched in literature for clinical trial reports. For ten of the validated hits, a Cmax ranging between 0.26 and 78.3 μM was found (Table 1), suggesting that the concentrations used in the secondary screen fall within a physiologically relevant range (Janku et al. 2024; Markham 2019; Wen et al. 2020; James et al. 2020; Flaherty et al. 2012; Sarker et al. 2015; Seo et al. 2017; Wheler et al. 2017; Mar Castro et al. 2017). For the validated compounds, different targets were impacted, including PI3K/mTOR related targets (bimiralisib, GDC-0084, GSK1059615), CDK related targets (palbociclib isethionate, SU9516, XL413) and cMET (BMS-777607).

Table 1.

Overview of maximum blood plasma (Cmax) concentrations of validated hits

Name	Target	Molecular weight (g/mol)	Clinical phase	Maximum dose	Cmax (ng/mL)	Cmax (μM)	References
Bimiralisib	PI3K	411.4	2	200 mg	1,200	2.92	Janku et al. (2024)
BMS-777607	cMET	512.9	Pre-clinical	NA	NA	NA	NA
Edafitinib	FGFR	446.6	4	12 mg	2,690	6.02	Markham (2019)
GDC-0084	PI3K	382.4	2	65 mg	99.4	0.26	Wen et al. (2020)
Gilteritinib	FLT3	552.7	4	120 mg	282	0.51	James et al. (2020)
GSK1059615	PI3K	333.4	Pre-clinical	NA	NA	NA	NA
Miltefosine	AKT	407.6	4	150 mg	31,900	78.3	Mar Castro et al. (2017)
Palbociblib	CDK4/6	447.5	4	150 mg	91	0.20	Flaherty et al. (2012)
Pictisilib	PI3K	513.6	2	450 mg	1,001	1.95	Sarker et al. (2015)
Pilaralisib	PI3K	541.0	2	600 mg	21,600	39.9	Wheler et al. (2017)
PS-1145	IKB/IKK	322.8	Pre-clinical	NA	NA	NA	NA
Radotinib	BCR-ABL	530.5	3	400 mg	250	0.47	Seo et al. (2017)
SNS-314	AURKA	430.9	1	NA	NA	NA	NA
SU9516	CDK1	241.3	1	NA	NA	NA	NA
XL413	CDC7	326.2	2	100 mg	5,993.5	18.4	Clinical trials database, NCT00886782

Overview of all validated KI that enhance ICAM1-eGFP expression. NA indicates no pharmacokinetic data could be found. For each compound, the maximum tested dose and related Cmax in the referred study was considered

Validated hits are associated with renal adverse events in the clinic

Increased inflammation can lead to an increase in leukocyte recruitment to the site of injury and thereby contribute to adverse effects in acute kidney injury (Kellum et al. 2021). Therefore, we questioned whether the identified KIs that enhance ICAM1 signaling are associated with renal adverse events in the clinic or clinical trials. For this, the FDA adverse events reporting system (faers, https://open.fda.gov/data/downloads/) and clinical trial transformation initiative database (www.clinicaltrials.gov) were mined for any renal adverse event. Four of the identified hits (palbociclib, miltefosine, gilteritinib and erdafitinib) have FDA approval and are currently on the market (Fig. 5A). Both miltefosine and gilteritinib have black box warnings for teratogenicity and carcinogenicity, respectively, but not for renal toxicity. As expected, for these marketed drugs we could find more reports of renal adverse events than for the drugs that are at earlier stages of clinical development (Fig. 5B). Although data is limited for the remaining KIs, most of those drugs already showed some (inflammatory-related) renal adverse events during phase 1/2 clinical trials. Of the marketed KIs, both palbociclib and gilteritinib are frequently associated with acute kidney injury and impairment of renal function in the clinic (Fig. 5C). Miltefosine is associated with impairment of renal function and tubulointerstitial nephritis. Of the drugs that are currently in early stages of clinical trials, bimiralisib and pictilisib are associated with renal injury and renal failure, respectively. For other KIs, mostly less severe kidney related events were identified (e.g. proteinuria, dysuria, glucose in urine), although the limited sample size for these compounds does not allow strong conclusions. Moreover, more severe kidney disease in future clinical use cannot be ruled out upon observation of these biomarkers. For example, proteinuria can be an early marker for more severe (kidney) pathologies (Hemmelgarn et al. 2010; Kramer et al. 2009).

Fig. 5.

Clinical renal injury associations of ICAM1-eGFP enhancing kinase inhibitors. A Clinical stage of identified kinase inhibitors. Color indicates whether marketed (phase 4) compounds have a black box warning. B Total number of renal injury events identified in faers or from clinical trials (www.clinicaltrials.gov) per kinase inhibitor. C Renal injury events identified for each compound in faers or within the clinical trial database. Red marked adverse events are associated with inflammation. A darker color indicates a higher frequency of the mentioned pathology by the indicated compound

Enhanced ICAM1 expression is indicative for increased chemokine secretion

As upregulation of ICAM1 expression is indicative for increased leukocyte attraction (Dustin et al. 1986), we tested whether TNFα-stimulated PTLCs induce the transcription and increased protein secretion of chemokines and cytokines, with a focus on those that are available within the BioLegend LegendPlex pro-inflammatory panels (Fig. 6A). Despite high IL23A mRNA levels upon treatment with TNFα, transcription of the selected cytokine genes was not significantly induced and IL-18 expression showed even significant downregulation (fdr p-adj = 0.04). Expression of the chemokine genes CXCL5, CXCL10, CXCL1, CCL2 and CCL11 was significantly (fdr p-adj ≤ 0.05) upregulated by TNFα, whereas for most other selected chemokine genes mRNA levels showed an upward but non-significant trend (Fig. 6A). In line with these findings, TNFα exposed PTLCs showed increased chemokine secretion in a concentration-dependent manner (Fig. 6B). We then determined whether the KIs that enhanced ICAM1-eGFP induction also increased chemokine secretion. Prior treatment of PTCLs with several KIs, including miltefosine, bimirasilib, and pictisilib, enhanced the TNFα-induced secretion of CCL20 and CXCL8 (Fig. 6C). Interestingly, an increase in CCL20 and CXCL8 secretion could also be observed for compounds that did not strongly affect ICAM1 expression in PTLCs, including AMG319 and CMLC hit KX2-391 (Kawai et al. 2009).

Fig. 6.

Effect of ICAM1-eGFP enhancing kinase inhibitors on chemokine and cytokine expression in PTLCs. A mRNA expression of chemokines and cytokines of the Biolegend Legendplex panels in proximal tubular cells in unstimulated and TNFα-stimulated conditions. Error bar indicates standard deviation based on three independent biological replicates. Asterisks indicate significant (fdr p-adjusted ≤ 0.05) differential expression upon TNFα treatment compared to unstimulated conditions. B Concentration of various chemokines in PTLC medium upon stimulation with different concentrations of TNFα. PTLCs were stimulated with indicated concentrations TNFα for 16 h. Thereafter, medium was harvested and chemokine concentration was measured using the LegendPlex assay. Vertical line indicates a concentration of 2.5 ng/mL TNFα, which is the concentration that was used for the high-throughput screen in Fig. 3. Data is shown for a single biological replicate. C Concentration in PTLC medium of selected chemokines after treatment of PTLCs with mentioned compounds and TNFα. PTLCs were treated with compounds for 8 h, followed by stimulation with 2.5 ng/mL TNFα for 16 h. Then, medium was harvested and chemokine concentration was measured using the LegendPlex assay. Data was fold change normalized towards the 0.1% DMSO + 2.5 ng/mL TNFα control. Red indicates increased chemokine expression, blue indicates decreased chemokine expression. NaN (grey) indicates that no significant secretion of chemokines was measured. The average value of two independent biological replicates is shown

Identified kinase inhibitors enhance TNF target gene expression via off-target effects

We wondered about the selectivity of the candidate enhancer hits and characterized them by mining the ChEMBL database (https://www.ebi.ac.uk/chembl/) for possible on- and off-targets. pChEMBL values were used to identify candidate KI targets by comparing inhibitory or dissociation constants (Fig. 7A) or IC50 values (Fig. 7B) for these targets. Out of the 16 KIs that increased TNFα signaling in PTLCs, three compounds were found in the database: pictilisib, BMS-777607 and GSK1059615. For pictisilib and BMS-777607 it was validated that these compounds had the highest affinity for their intended target (PI3K/AKT and cMET, respectively). Furthermore, many other kinases were also identified as targets of pictisilib and BMS-777607 within the range of our screening concentrations (0.1–10 μM; Fig. 4F). For GSK1059615 no data was available on the intended target (PI3K/AKT/mTOR), but many off-targets (e.g. CLK1, CLK2, TAOK1) were reported within screening concentration range. BMS-777607 (AXL, MET, MST1R, TYRO3 inhibitor) had many off-targets that were closely related to the intended target in the tyrosine kinase domain, shown by the close proximity of the off-targets in the kinome tree (Fig. 7C). On the contrary, pictilisib (PI3K/mTOR inhibitor) had many structurally different targets throughout the kinome, suggesting that this compound might be non-specific. Similarly, GSK1059615 (AKT1/mTOR inhibitor) also has quite diverse off-targets with high affinity, although the number of potential off-targets was less compared to pictilisib. We verified whether these potential intended and off-targets are also present in PTLCs, CMLCs and HLC. mRNA expression of these KI targets was highly similar and the majority of these genes showed moderate to high expression values (Suppl. Fig. S3). The various targets for these three inhibitors clustered together in the kinome map: pictilisib (CMGC kinases), BMS-777607 (tyrosine kinases, TK) and GSK1059615 (AGC kinases; Fig. 7C), all identified off-targets were mapped on the kinome map (Eid et al. 2017). Next, we used StringDB to evaluate whether the off-targets were associated with the IKK-complex (CHUK, IKBKB, IKBKG), IκBα (NFKBIA) and the genes for canonical NF-κB (RELA, P65) signaling (Szklarczyk et al. 2023). To minimize the possibility of finding false positive relationships, nodes were only included if (i) there was a high confidence (≥ 0.7) for a relationship between genes and (ii) the relationship has been described in various literature, shown in experiments or in databases. Of relevance, 29/75 candidate off-target kinases showed an (in)direct relationship with the NF-κB signaling pathway components (Fig. 7D). The expression of the different kinases were not affected by TNFα treatment (Supplemental Fig. S4).

Fig. 7.

Kinome target interaction of ICAM1-eGFP enhancing kinase inhibitors. A and B Ki/Kd (A) or IC50 (B) values of identified targets for selected compounds picitisilib, GSK1059615 and BMS-77607. Arrows indicate the intentend targets of the corresponding compound (indicated by the color). pChEMBL value is indicated by the color, with increasing intensity being a higher value. Grey color indicates that the compound does not target the gene. In the case multiple pChEMBL values were found, the highest value is shown in the heatmap. C Kinome tree representing the kinases targeted by the compounds. Color of the dots is similar to the presented colors for the kinase inhibitors shown at A, with orange being pictisilib, red being BMS-77607 and green being GSK1059615. Size of the dots is for aesthetic purposes. D StringDB plot shows the connection between potential targets of the compounds and key drives of the TNF signaling pathway. Colors correspond to the same compounds as shown in A–C

Discussion

In this study we assessed the vulnerability of kinase inhibitors to modulate the TNFα signaling pathway and assessed the effect of 760 KIs that cover the majority of the kinome on the expression of ICAM1-eGFP in hiPSC-derived PTLC, CMLC and HLC lineages. We uncovered different KIs that inhibit ICAM1 expression and are common between the three lineages. Most candidates enhanced ICAM1-eGFP expression with little overlap between different cell types and strongest effects were found in PTLCs. Three KIs, including the clinical applied miltefosine, also enhanced TNFα-induced chemokine CCL20 and CXCL8 expression. Three clinical marketed KIs that enhanced ICAM1-eGFP expression in PTLCs have demonstrated increased reporting of adverse renal effects. Additionally, unintended kinase inhibitor activity was associated with three inhibitors that enhanced ICAM1-eGFP expression. Our results indicate the liability of kinase inhibitors to modulate TNFα-mediated signaling and promote inflammatory responses. We propose a pre-clinical testing strategy to select KIs that demonstrate an unintended modulation of TNFα signaling for further evaluation. This could involve different cell lineages derived from hiPSC-ICAM1-eGFP reporters in combination with high throughput imaging and complementary orthogonal assays to measure chemokine secretion.

In our studies we used hiPSC-derived PTLCs, CMLCs and HLCs where ICAM1 was endogenously tagged with eGFP with an isogenic background. Yet, we identified different KIs that affected TNFα-induced ICAM1 expression across the three different lineages. This could relate to slight differences in mRNA expression of the various intended kinase targets between the different hiPSC-derived cell types. Additionally, differential expression of various transporter proteins in the differentiated cell lines could lead to differences in intracellular compound concentrations (Erdei et al. 2014). Although the induction of TNFα pathway downstream targets was largely similar between the three lineages, we cannot exclude that the regulation of the TNFα-signaling pathway is different between these cell types. Given these differences, this would advocate to assess liability of candidate KIs for TNFα-signaling modulation in different lineages susceptible for KI-related adverse drug reactions, which is well possible using our iPSC-based approach. However, some frequently observed KI toxicities like intestinal toxicity are likely non-specific consequences of their general proliferation-suppressive effects, which should be taken into account when setting up these assays (Shyam Sunder et al. 2023).

In the present study we did not identify a specific class of KIs that affect TNFα-induced ICAM1-eGFP expression in PTLCs. This would suggest that unintended kinase inhibition is a risk factor that contributes to the modulation of TNFα-signaling. For three KIs that enhanced ICAM1-eGFP expression we observed diverse kinase targets in ChEMBL (BMS-777607, GSK1059615 and pictisilib; see Fig. 7). Moreover, for most KIs that we identified there is no qualitative and quantitative kinase inhibitor information available, making a generalization difficult. Typically, highly selective kinase inhibitors are preferred for clinical application, although also broad-spectrum kinase inhibition are successfully used in a clinical setting (Giordano and Petrelli 2008). Further detailed datasets on kinase inhibitor spectrum and potency will be required to better understand the relationship between selectivity and specificity of KIs and adversity liabilities related to modulation of inflammatory signaling.

The presented secondary screen was performed using a physiological relevant concentration range for the identified kinase inhibitors. Among the validated hits, increased ICAM1-eGFP expression was observed at concentrations near the Cmax and up to 50-fold higher, as exemplified by palbociclib. Therefore, we conclude that the KI concentrations that induce TNFα-mediated ICAM1-eGFP expression may have clinical relevance. This may in particular be the case in the context of renal toxicity, since KIs are also often substrates for various drug transporters involved in renal clearance by the proximal tubular cells (Huang et al. 2020; Hulin et al. 2024). Additionally, KIs may be used in combination with classical anticancer drugs that cause renal toxicity involving inflammatory signaling, such as observed for cisplatin (Benedetti et al. 2013; Benedetti et al. 2013b). Additionally, for the liver the first pass effect after oral administration typically results in higher concentration in the liver compared to general plasma Cmax levels (Wright et al. 2015).

Treatment with several KIs (e.g. PS-1145, miltefosine, GSK1059615) resulted in enhanced ICAM1 expression, as well as increased CXCL8 expression levels (Fig. 6D). Enhanced levels of CXCL8 in vivo are associated with neutrophil recruitment via the CXCR8/ICAM1 axis (Chung and Lan 2011). During this innate immune response, IL-17 is produced by PTLCs which leads to synergistic production of other chemokines, including CCL20 (J. W. Lee et al. 2008). In the current study CCL20 is reported to be enhanced upon treatment with various KIs, including kidney injury-associated drugs pictilisib and bimiralisib (Fig. 5C). In mouse models, upregulation of this chemokine is associated with T cell recruitment, renal tissue injury and loss of kidney function (Turner et al. 2010).

Using the ChEMBL biological activity database, we identified targets for three KIs (pictisilib, GSK1059615, BMS-77607) with an affinity within therapeutic ranges (Fig. 7A, B). Moreover, for most of these targets we identified a link with key components of the NF-κB signaling pathway (Fig. 7D). Although this data provides information about possible mechanisms in general, we did not establish whether the inhibition of these kinases is directly driving the enhanced ICAM1-eGFP expression in PTLCs. Given that the (off-)targets of the three selected KIs are dispersed throughout the kinome and show little overlap with TNFα-related genes, it seems unlikely that these compounds intrinsically promote TNFα signaling. Since the PI3K/AKT/mTOR pathway (target of pictilisib and GSK1059615; Fig. 7A, B) and TYRO3, AXL and MER (TAM) receptor kinases (target of BMS-777607; Fig. 7A, B) are known to be able to modulate TNFα signaling outcome (Post et al. 2021; Hongshuang Wang et al. 2024), we anticipate that the differential target gene expression is mediated via these pathways.

The broad-spectrum effects of these inhibitors demonstrate the complexity in uncovering the exact mechanisms underlying the enhanced TNFα-mediated ICAM1-eGFP expression. Here we used a specific ICAM1-eGFP reporter that allows specific assessment of TNFα-signaling modulation in a high throughput live-cell imaging-based strategy. Novel cost-effective high throughput transcriptomics approaches including TempO-Seq or DRUG-seq (Ye et al. 2018) could facilitate additional orthogonal mechanisms underpinning the effects of the KIs on TNFα signaling, as well as other pathways that we did not cover in the current study.

In conclusion, we demonstrate the feasibility and applicability of hiPSC-ICAM1-eGFP differentiated in different cell lineages to assess the effect of KIs on TNFα-signaling. We anticipate that this reporter can be applied to assess the effect of other small molecules or therapeutic modalities on inflammatory signaling. Future perspectives may involve the implementation of these reporters in 3D hiPSC-derived organoids models of kidney, liver and heart models in combination with isogenic hiPSC-derived immune cells to investigate the effect of kinase inhibitors on inflammatory response in human physiological relevant models.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

hiPSC

Human induced pluripotent stem cell

HLC

Hepatocyte-like cell

PTLC

Proximal tubular-like cell

CMLC

Cardiomyocyte-like cells

TNFα

Tumor necrosis factor alpha

KIs

Kinase inhibitors

NF-κB

Nuclear factor kappa B

NEMO

NF-κB essential modulator

IκB

Inhibitor of the κB

IKK

Inhibitor of the κB kinase

MAPKK

Mitogen-activated protein kinase kinases

JNK

JUN N-terminal kinase

ICAM1

Intracellular adhesion molecule 1

eGFP

Enhanced green fluorescent protein

Cmax

Maximum blood plasma concentration

Funding

The work received funding from the Innovative Medicines Initiative 2 (IMI2) Joint Undertaking for the eTRANSAFE (agreement grant 777365) project, the innovation program and EFPIA; the EC Horizon2020 EU-ToxRisk project (grant number 681002), the EC Horizon2020 RISK-HUNT3R project (grant number 964537) and from the KU Leuven (grant number FWO-SBO-QPG-359638-iPSC-LIMIC).

Declarations

Conflict of interest

The authors declare no conflict of interest.