Targeting ALDH1A1 to enhance the efficacy of KRAS-targeted therapy through ferroptosis

Abstract

KRAS is among the most commonly mutated oncogenes in human malignancies. Although the advent of sotorasib and adagrasib, has lifted the “undruggable” stigma of KRAS, the resistance to KRAS inhibitors quickly becomes a major issue. Here, we reported that aldehyde dehydrogenase 1 family member A1 (ALDH1A1), an enzyme in retinoic acid biosynthesis and redox balance, increases in response to KRAS inhibitors and confers resistance in a range of cancer types. KRAS inhibitors' efficacy is significantly improved in sensitive or drug-resistant cells, patient-derived organoids (PDO), and xenograft models by ALDH1A1 knockout, loss of enzyme function, or inhibitor. Furthermore, we discovered that ALDH1A1 suppresses the efficacy of KRAS inhibitors by counteracting ferroptosis. ALDH1A1 detoxicates deleterious aldehydes, boosts the synthesis of NADH and retinoic acid (RA), and improves RARA function. ALDH1A1 also activates the CREB1/GPX4 pathway, stimulates the production of lipid droplets in a pH-dependent manner, and subsequently prevents ferroptosis induced by KRAS inhibitors. Meanwhile, we established that GTF2I is dephosphorylated at S784 via ERK by KRAS inhibitors, which hinders its nuclear translocation and mediates ALDH1A1's upregulation in response to KRAS inhibitors. In summary, the results offer valuable insights into targeting ALDH1A1 to enhance the effectiveness of KRAS-targeted therapy through ferroptosis in cancer treatment.

1. Introduction

Kirsten rats arcomaviral oncogene homolog (KRAS) is one of the most frequently mutated oncogenes in various cancers, especially in pancreatic adenocarcinoma (95 %), colorectal cancer (35 %), and lung cancer (25 %) [1,2]. In recent years, the development of small molecule inhibitors, sotorasib [3] and adagrasib [4], has overcome the “undruggable” challenge and brought hope to patients with KRAS-mutant tumors [[4], [5], [6], [7]]. However, the development of drug resistance will limit the efficacy of targeted therapies [8].

Ferroptosis is a unique form of regulated cell death characterized by an imbalance of reactive oxygen species (ROS) production and the peroxidation of polyunsaturated fatty acids (PUFA) [[9], [10], [11]]. In recent years, there has been a growing focus on studying the role of ferroptosis in tumor treatment resistance [[12], [13], [14]], and it can enhance the efficacy of radiotherapy [15], chemotherapy [16] and immunotherapy [17]. Previous studies mentioned that KRAS mutations alter cells' lipid and cysteine metabolism, conferring resistance to ferroptosis [18,19]. However, the relationship between KRAS-targeted therapy and ferroptosis, as well as the role of ferroptosis in KRAS inhibitors' resistance, remains unclear.

The aldehyde dehydrogenase (ALDH) superfamily comprises 19 isozymes with critical physiological functions belonging to the nicotinamide adenine dinucleotide phosphate (NADP)-dependent enzyme family. Their roles in redox homeostasis have been widely reported [20]. ALDH1A1 is the most prominent isozyme of the ALDH family, which regulates the metabolism of retinal to retinoic acid (RA), maintains low ROS levels, and enhances the detoxification of toxic aldehydes [21,22].

Herein, we found that ALDH1A1, by inhibiting ferroptosis, mediates resistance to KRAS-targeted therapies. We demonstrated that blocking ALDH1A1 activity enhances the effectiveness of KRAS-targeted drugs in various KRAS mutant cell lines, drug-resistant cells, patient-derived organoids (PDO), and animal models. Elucidation of the underlying mechanisms reveals that ALDH1A1 promotes the production of aldehyde metabolites and elevates the levels of NADH and RA, ultimately activating receptor α (RARA), which functions as a suppressor of ferroptosis [23]. Additionally, ALDH1A1 suppresses KRAS inhibitors-induced ferroptosis by activating the cAMP responsive element binding protein 1(CREB1)/glutathione peroxidase 4 (GPX4) pathway and promoting lipid droplet (LD) formation via reducing intracellular pH (pHi). Our findings also indicated that ALDH1A1 is upregulated through cytoplasmic retention of the transcriptional repressor GTF2I, because of dephosphorylation of general transcription factor IIi (GTF2I) at S784 via ERK/MSK by KRAS inhibitors, and finally promotes drug resistance. The comprehensive study identified a new mechanism for KRAS-targeted drug resistance and suggested ALDH1A1 as a potential co-target to improve the clinical outcomes of KRAS-targeted therapies.

2. Results

2.1. ALDH1A1 promotes resistance to KRAS-targeted inhibitors relying on enzyme activity

ALDH1A1 was upregulated in MIAPACA2 and H2122 cell lines after treatment with KRAS G12C inhibitor sotorasib or adagrasib for 72 h, as shown by RNA-seq analysis (Fig. 1A and B). The upregulation observed was verified in previous results [24] (Figs. S1A–B) and was consistent across various concentrations of KRAS inhibitors (Fig. S1C). We validated the upregulation in 10 other KRAS G12C-mutant cell lines from various cancers (Fig. 1C, Fig. S1D). Enzyme activity of ALDH1A1 also increased after the treatment (Fig. 1D and E).

Fig. 1.

ALDH1A1 promotes resistance to KRAS G12C inhibitors relying on enzyme activity. (A) Venn plot displayed 36 candidates from the intersection of differentially expressed genes detected by RNA sequencing between the DMSO and the KRAS G12C inhibitors (sotorasib or adagraisb) treatment in MIAPACA2 and H2122 cells. The cells were treated for 72 h with IC50 (sotorasib: MIAPACA2, IC50 = 40 nM; H2122, IC50 = 100 nM; adagrasib: MIAPACA2, IC50 = 60 nM; H2122, IC50 = 180 nM). (B)The volcano plots showed that ALDH1A1 was upregulated in both cells treated with KRAS G12C inhibitors (|log2FC| > 0.4 and p < 0.05). (C) qRT-PCR and WB verified the ALDH1A1 upregulation in twelve KRAS G12C mutant cell lines treated with sotorasib or adagraisb for 72 h (pancreatic adenocarcinoma cell line MIAPACA2; bladder urothelial carcinoma cell line UMUC3; esophageal squamous cell carcinoma cell line KYSE410; lung adenocarcinoma cell lines H2122, LU65, LU99, SW1573, and H1792; colorectal adenocarcinoma cell lines JVE-015, TT1TKB, SW837, and SW1463) (n = 3). (D-E) Relative ALDH1A1 activity enhanced in MIAPACA2 and H2122 cells in response to sotorasib or adagraisb (sotorasib: MIAPACA2, IC50 = 40 nM, IC90 = 1000 nM, Peak serum concentration = 10000 nM; H2122, IC50 = 100 nM, IC90 = 5000 nM, Peak serum concentration = 10000 nM; adagrasib: MIAPACA2, IC50 = 60 nM, IC90 = 1500 nM, Peak serum concentration = 6000 nM; H2122, IC50 = 180 nM, IC90 = 5000 nM, Peak serum concentration = 6000 nM, 72 h), determined via ALDH Activity Assay Kit (D) and ALDEFLUOR assay (E) (representative flow cytometry dot plots are shown) (n = 3). (F) Crucial active site 193K for ALDH1A1 was mutated from K to Q/R, blocking the ALDH1A1 activity but not the protein expression. (G-H) MIAPACA2 and H2122 stable cell lines were established with the ALDH1A1 knockout (KO) and overexpression (OE) with or without mutants (K193Q/R). The ALDH1A1 enzyme activity was decreased extremely in the ALDH1A1-KO, K193Q/R group and increased in ALDH1A1-OE (representative flow cytometry dot plots are shown) (n = 3). (I)Relative viability of MIAPACA2 and H2122 cells showed that ALDH1A1 KO and loss of enzyme function (K193Q/R) increased the effect of gradient concentrations of KRAS G12C inhibitors for 72 h, while ALDH1A1 OE did the opposite (n = 3). (J) Relative viability of MIAPACA2 and H2122 cells showed that ALDH1A1 inhibitor CM10 (500 nM) increased the effect of gradient concentrations of KRAS G12C inhibitors for 72 h (n = 3). Data were analyzed by two-way ANOVA or Student’s t-test and were presented by mean ± SD.

To further investigate the role of ALDH1A1 in modulating the sensitivity of KRAS G12C inhibitors, we knocked out ALDH1A1 in MIAPACA2 and H2122 cells and overexpressed it with or without a key catalytic site mutation (K193Q/R) (Fig. 1F). The K193 site mutations abolished enzyme activity but not protein level (Fig. 1G and H). ALDH1A1 knockout (KO) enhanced the effect of sotorasib and adagrasib, while ALDH1A1 overexpression (OE) restored it, only in the wild-type form, indicating that ALDH1A1 enzymatic activity mediates resistance to KRAS G12C inhibitors (Fig. 1I). The ALDH1A1 inhibitor CM10 [25] also increased the sensitivity in enzymatic dependent manner (Fig. 1J, Figs. S1E–F).

We verified the results in sotorasib or adagrasib-resistant cells (IC50 is more than ten times higher than that of parental cells), generated by 9-months exposure to the inhibitors (Figs. S1G–H). Resistant cells with ALDH1A1-KO and OE were constructed (Fig. S1I). Both ALDH1A1-KO and CM10 increased the resistant cells’ sensitivity to KRAS G12C inhibitors, and OE reversed it (Figs. S1J–K). To test the generality of inhibiting ALDH1A1, we assessed the sensitivity of parental MIAPACA2 and H2122 cells to pan-KRAS inhibitors BI-2493 [26] and that of KRAS G12D mutant cells (HPAC and GP2D) to KRAS G12D inhibitor MRTX1133 [27]. The combination of ALDH1A1 inhibitor CM10 improved the sensitization effect (Figs. S1L–M), suggesting a potential combination therapy in different KRAS inhibitors.

2.2. Knockout or pharmacological inhibition of ALDH1A1 enhances the effectiveness of KRAS-targeted inhibitors in PDOs and in vivo

To recapitulate the histological and molecular characteristics of primary PAAD and LUAD tissues, we cultured PAAD and LUAD PDOs (Fig. 2A and B, Fig. S1N) [28]. ALDH1A1 KO and OE-PDOs were constructed (Fig. 2C). Morphology (Fig. 2D–H) and luminescence measurement (Fig. 2E–I) showed that ALDH1A1 KO resulted in more pronounced cell death (5-fold change) than the NC group when treated with sotorasib or adagrasib, while ALDH1A1 OE conferred resistance. CM10 also increased PDOs sensitivity to KRAS G12C inhibitors (Fig. 2F–G, J-K).

Fig. 2.

Inhibition of ALDH1A1 confers sensitivity to KRAS G12C inhibitors in PDOs and in vivo. (A) Flow chart of KRAS G12C mutant PAAD and LUAD patient-derived organoids (PDOs) Establishment. (B) HE staining images of PAAD and LUAD patient tissues and PDOs (scale bars, 50 μm). (C) Top: WB showed the successful construction of ALDH1A1 KO and OE of PAAD and LUAD PDOs. Below: representative brightfield images of PAAD and LUAD PDOs (scale bars, 100 μm). (D) Representative brightfield images of PAAD PDOs with ALDH1A1 KO or OE, treated with KRAS G12C inhibitors for 72 h (scale bars, 100 μm). (E) Luminescence measurement showed the relative viability of PAAD with different concentrations of sotorasib and adagrasib for 72 h. ALDH1A1 KO caused more severe death in PDOs, which could be abolished by overexpression. (F–G) Representative brightfield images (F) and luminescence measurement (G) of PAAD PDOs incubated with or without CM10 (1000 nM), and treated with sotorasib (300 nM) or adagrasib (300 nM) for 72 h (scale bars, 100 μm). (H) Representative brightfield images of LUAD PDOs with ALDH1A1 KO or OE, treated with KRAS G12C inhibitors for 72 h (scale bars, 100 μm). (I) Luminescence measurement showed the relative viability of PAAD with different concentrations of sotorasib and adagrasib for 72 h. (J–K) Representative brightfield images (J) and luminescence measurement (K) of LUAD PDOs incubated with or without CM10 (1000 nM), and treated with sotorasib (300 nM) or adagrasib (300 nM) for 72 h (scale bars, 100 μm). (L) Schematic design of xenografted tumor assay. (M–O) Tumor images (M), growth curves (N), and tumor weight at the end of the treatment (O) showed that ALDH1A1 OE rescued the ALDH1A1 KO-induced sensitivity to KRAS G12C inhibitors in vivo. (P) Top: Administration of sotorasib, adagrasib, and CM10 for nude mice. Below: Tumor images of each group, growth curves (Q), and Tumor weight (R) showed that CM10 combined with KRAS G12C inhibitors repressed the tumor growth derived from MIAPACA2 cells more than KRAS inhibitors treatment alone. Data were analyzed by two-way ANOVA or Student's t-test and were presented by mean ± SD.

The benefit of knockout or pharmacological inhibition of ALDH1A1 was evaluated in vivo using xenograft models. Sotorasib and adagrasib reduced tumor growth compared to the vehicle group. ALDH1A1 KO further reduced tumor volume and weight in the KRAS inhibitors treatment groups (almost ten-fold change), while the tumors with ALDH1A1-OE cells were bigger and faster-growing (Fig. 2L–O). The combined administration of CM10 and KRAS-targeted inhibitors resulted in a remarkable reduction in tumor size, achieving a tumor burden that was one-eighth of that observed in the monotherapy groups (Fig. 2P–R). These results suggested that combined therapy with ALDH1A1 inhibition improves the effect of KRAS-targeted drugs in vivo.

2.3. ALDH1A1 confers resistance to KRAS-targeted inhibitors by inhibiting ferroptosis

MIAPACA2 and H2122 cells were treated with KRAS inhibitors in combination with ferroptosis inhibitors, ferrostatin-1 (fer-1) and deferoxamine (DFO), apoptosis inhibitor Z-VAD-FMK (Z-VAD), and necrosis inhibitor necrosulfonamide (necro) for 72 h. Ferroptosis inhibitors reversed KRAS inhibitors-induced cell death significantly, resulting in a cell survival rate of 80 % (Fig. 3A, Fig. S1O). The combination of KRAS inhibitors and ferroptosis inducers (FIN) (RSL3 and IKE) did not have a lethal synergistic effect (Fig. S2A). KRAS-targeted inhibitors induce ferroptosis could be one of the reasons. We confirmed the finding with elevated intracellular lipid peroxidation and malondialdehyde (MDA) levels after sotorasib or adagrasib treatment, and reversal by ferroptosis inhibitors (Fig. 3B and C). Ferroptosis inhibitors’ reversal extent on cell death caused by KRAS inhibitors showed no difference when ALDH1A1 was knockout, unlike apoptosis and necrosis inhibitors, which suggested that ALDH1A1 caused resistance to KRAS inhibitors through ferroptosis (Fig. 3A).

Fig. 3.

ALDH1A1 confers resistance to KRAS-targeted inhibitors by inhibiting ferroptosis. (A) Cell viability of MIAPACA2 and H2122 cells (ALDH1A1-NC, KO) treated with sotorasib or adagrasib alone, or combined with fer-1 (500 nM) or DFO (5000 nM) or Z-VAD (500 nM) or necrosulfonamide (250 nM) for 72 h (n = 3). Only fer-1 and DFO reversed the KRAS inhibitors-induced cell death and showed no difference in ALDH1A1 KO groups. (B–C) Co-treatment with fer-1 (500 nM) or DFO (5000 nM) abolished the KRAS G12C inhibitors-enhanced lipid peroxidation (B) and MDA (C) levels of MIAPACA2 and H2122 cells (IC50, 72 h) (n = 3). (D–G) Lipid peroxidation (D) and MDA (E) levels in MIAPACA2 and H2122 cells increased in ALDH1A1-KO, K193Q/R groups, and CM10-treated group (500 nM) (F–G), under the treatment of sotorasib or adagrasib (IC50, 72 h).ALDH1A1-OE reverses the increase. (n = 3) (H) CM10 and KRAS inhibitors combination increased the intracellular ferrous iron level. (n = 3) (I) DFO rescued mitochondrial shrinkage. CM10 exacerbated and mitochondrial shrinkage (MIAPACA2 cell) caused by sotorasib or adagrasib (IC50, 72 h), displayed by transmission electron microscopy images (scale bars, 10 μm). Data were analyzed by Student's t-test and were presented by mean ± SD.

We then validated that ALDH1A1 KO, abolish enzyme activity (K193Q/R), and CM10 treatment sensitized cells to FINs, and increased lipid peroxidation and MDA levels. ALDH1A1 OE did the opposite (Figure S2B-C, Figure S2D-G). In addition, no ALDH1A1 protein interaction with key ferroptosis factors was detected by Co-Immunoprecipitation Mass Spectrum (CoIP-MS) (Table S1). These results indicated that ALDH1A1 inhibits ferroptosis through its enzyme activity. Similarly, when treated with KRAS inhibitors, cells with ALDH1A1-KO, K193Q/R, and CM10-treated showed higher lipid peroxidation and MDA levels, while ALDH1A1-OE cells displayed approximately half the levels of these oxidative stress markers (Fig. 3D–G). Regardless of the concentration of KRAS-targeting drugs employed, the elevation of lipid peroxidation and MDA levels observed in ALDH1A1-KO cells and the corresponding reduction in ALDH1A1-OE cells remained consistent (Figs. S2H–I). Furthermore, the co-treatment of CM10 and KRAS inhibitors resulted in a significant evaluation of intracellular ferrous iron level by approximately 30 % (Fig. 3H) and severe mitochondrial shrinkage with denser membranes (Fig. 3I), which are reliable indicators of ferroptosis severity. The results demonstrated that ALDH1A1 attenuates KRAS inhibitors-induced ferroptosis and confers drug resistance.

2.4. ALDH1A1 inhibits ferroptosis by degradation of toxic aldehydes, production of NADH, and activation of RARA

We have proved that KRAS inhibitors induce ferroptosis with increased MDA level. Here, we further investigated the level of other aldehydes, including retinal and 4-hydroxy-2-nonena (4-HNE). ALDH1A1 KO or CM10 treatment significantly enhanced the toxic aldehydes produced with KRAS inhibitors, while ALDH1A1-OE decreased the level (Fig. 4A and B), confirming the role of ALDH1A1 in detoxification. ALDH1A1 was the main enzyme converting retinal to RA, with aldehydes donating and NAD accepting electrons, producing carboxylic acids and NADH. We measured the RA content and NAD+/NADH ratio. As expected, ALDH1A1 KO or inhibition kept the cells with low RA level, while ALDH1A1 overexpression increased RA production (Fig. 4C). The NAD+/NADH ratio change followed the aldehydes level (Fig. 4D).

Fig. 4.

ALDH1A1 inhibits ferroptosis through detoxifying aldehydes, promoting production of NADH and improving RARA function. (A–D) Measurement of retinal (A), 4-HNE (B), retinoic acid (C), and NADH (D) production in MIAPACA3 and H2122 cells with ALDH1A1 NC, inhibitor (CM10:500 nM), KO and OE under sotorasib or adagrasib treatment (IC50,IC90, Peak serum concentration, 72 h) (n = 3). (E) Relative viability showed retinoic acid receptor α (RARA) resisted KRAS G12C mutant cells to ferroptosis caused by RSL3, IKE, sotorasib and adagtasib (IC50), demonstrated by medications with RARA inhibitors AGN193109 (1000 nM) or activator ATRA (1000 nM) for 72 h. (n = 3) (F) Lipid peroxidation of KRAS G12C mutant cells increased when added AGN193109 (1000 nM) to sotorasib or adagrasib (IC50) for 72 h. (n = 3) (G) Dual-luciferase assays reflected significant RARA activity depression in ALDH1A1-KO and CM10 (500 nM) treated groups (MIAPACA2 and H2122 cells) under KRAS G12C inhibitors medications (IC50). (n = 3) Data were analyzed by Student's t-test and were presented by mean ± SD.

The inhibition of RARA with AGN193109 sensitized cells to KRAS inhibitors and FINs, while RARA activator all-trans-retinoic acid (ATRA) caused resistance (Fig. 4E). Also, lipid peroxidation induced by KRAS inhibitors was markedly potentiated in the presence of RARA inhibitors, exhibiting a twofold increase (Fig. 4F), indicating that RARA was the important suppressor of ferroptosis in KRAS inhibitors. Furthermore, we verified the RARA activity by dual-luciferase assay. As the RA level, RARA was less activated in the ALDH1A1-KO or CM10 group under the treatment of KRAS inhibitors, and ALDH1A1-OE increased RARA activity (Fig. 4G). We demonstrated that ALDH1A1 inhibits ferroptosis by metabolizing retinal to RA, promoting NADH production and activating RARA.

2.5. ALDH1A inhibits ferroptosis via pH-dependent activation of the CREB1/GPX4 axis and promotion of lipid droplet formation

ALDH family oxidizes aldehydes to carboxylic acids, which dissociate into carboxylate and hydrogen ions [21] (Fig. S3A). We discovered that KRAS inhibitors elevated pHi, ALDH1A1 KO or CM10 treatment upregulated it further, and ALDH1A1 OE made the cells acidic (Fig. 5A), indicating that ALDH1A1 decreased the pHi upregulated by KRAS inhibitors. We adjusted the extracellular pH (pHe) by adding HCl or NaOH. Lower pHe conferred resistance to FINs and KRAS inhibitors (Fig. 5B and C), with lower lipid peroxidation and MDA levels, which suggested that lower pH makes cells more resistant to ferroptosis (Fig. 5D and E).

Fig. 5.

ALDH1A1 inhibits ferroptosis through activation of the CREB1/GPX4 axis and promotion of lipid droplet formation depending on lower pH. (A)ALDH1A1-KO or pharmacological inhibition (CM10:500 nM) increased intracellular pH (pHi) detected by BCECF-AM probe in MIAPACA2 and H2122 cells treated with KRAS G12C inhibitors, while ALDH1A1 OE did the opposite (n = 3). (B–C) Relative viability revealed the lower pHe rendered KRAS G12C mutant cells resistant to RSL3, IKE (B), sotorasib and adagrasib (C) (pHe = 7.5, 7.0, 6.5, 72 h) (n = 3). (D–E) Lipid peroxidation (D) and MDA (E) levels in MIAPACA2 and H2122 cells at indicating pH (pHe = 7.5, 7.0, 6.5) verified lower pHe inhibits ferroptosis (n = 3). (F) qRT-PCR and WB screened out that GPX4 upregulated by acidic conditions, not FSP1 and SLC7A11 (n = 3). (G) MIAPACA2 and H2122 cells were maintained at indicating pH (pHe = 7.5, 7.0, 6.5) for 72 h, and the WB results showed that lower pHe upregulated GPX4 expression. (H)ALDH1A1 KO reduced CREB1 phosphorylation and GPX4 expression, and ALDH1A1 OE activated CREB1/GPX4 axis, while HCL supplement abolished the reduction. (I–J) Representative pictures (MIAPACA2 and H2122 cells) (I) and quantification (J) of Lipi-Green-stained lipid droplet showed lower pHe promoted LD formation upon maintained at indicating pH for 24 h (pHe = 7.5, 7.0, 6.5) (scale bars, 10 μm) (n = 3). (K–L)ALDH1A1 KO and inhibition (CM10:500 nM, 72 h) repressed LD formation, and ALDH1A1 OE promoted it, while HCL supplement reversed the suppression. Representative pictures (MIAPACA2 and H2122 cells) (K) and quantification (L) of Lipi-Green-stained lipid droplet were shown (scale bars, 10 μm) (n = 3). Data were analyzed by Student's t-test and were presented by mean ± SD. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

To further elucidate the role of pH, we investigated the expression of essential ferroptosis factors: GPX4, ferroptosis suppressor protein 1 (FSP1), and solute carrier family 7 member 11(SLC7A11) under different pHe. Only GPX4 increased significantly at lower pHe (Fig. 5F). Since GPX4 was upregulated in both mRNA and protein levels (Fig. 5G), and the cAMP pathway was enriched among differentially expressed genes (DEGs) based on ALDH1A1 expression (Fig. S3B), we considered transcription factor CREB1 regulation as previous study [29]. GPX4 was downregulated in the CREB1 knockdown group and severely decreased with KRAS inhibitors (Figs. S3C–D). JASPAR network [30] prediction (Fig. S3F), ChIP-Seq data in the Encode database [31] (Fig. S3E), our ChIP-qPCR assays (Figs. S3G–H), and luciferase experiments (Fig. S3I) all confirmed that CREB1 transcriptionally regulates GPX4. ALDH1A1 KO reduced CREB1 phosphorylation and GPX4 expression, but OE and HCL supplementation reversed this effect, implying that ALDH1A1 modulates the CREB1-GPX4 axis in a pH-dependent manner (Fig. 5H).

Previous studies have shown that lipid droplet (LD), active organelles regulating lipid metabolism, were modulated by pHi and inhibited ferroptosis [[32], [33], [34], [35]]. We validated that under acidic conditions, there was an increase in the capacity of MIAPACA2 and H2122 to capture polyunsaturated fatty acids (PUFAs) into LDs (Figs. S3J–K), and LDs formation was increased (Fig. 5I and J). Notably, the key gene Diacylglycerol-O-Acyltransferase Homolog 1 (DGAT1) [36,37] involved in LD formation was upregulated under acidic conditions, suggesting a mechanism in which LDs are modulated by pH and regulate ferroptosis (Fig. S3L). ALDH1A1 KO decreased LDs formation, while OE and HCL addition rescued this decrease (Fig. 5K and L). Thus, ALDH1A1 can lower pHi, upregulate the CREB1-GPX4 axis and LD biogenesis, and inhibit ferroptosis to resist KRAS inhibitors.

2.6. KRAS-targeted inhibitors upregulate ALDH1A1 expression by inhibiting GTF2I phosphorylation at S784 via ERK/MSK

We first treated the cells with sotorasib and adagrasib for 72 h, then withdrew the inhibitors for 72 h, and redosed them for 72 h. ALDH1A1 expression increased after the first and third treatments but decreased after the withdrawal (Fig. 6A), indicating that KRAS inhibitors induced ALDH1A1 upregulation. We treated the cells with ERK inhibitor SCH7 72984 and AKT inhibitor MK-2206 for 72 h, and only ERK inhibition upregulated ALDH1A1 expression, suggesting that KRAS inhibitors upregulated ALDH1A1 expression via ERK pathway (Fig. 6B). As ALDH1A1 was upregulated both in mRNA and protein levels, we considered transcription factors (TF) involved in response to KRAS inhibitors. Using DNA pull-down with high-throughput mass spectrometry and the ChIP-Seq data in Encode, we identified two TFs, GTF2I and zinc finger protein 384 (ZNF384), binding to ALDH1A1 promoter (Fig. 6C–Table S2). GTF2I knockdown increased ALDH1A1 expression, while ZNF384 had no effect (Figure S3M − N, Fig. 6D and E), which showed GTF2I as a transcriptional repressor to ALDH1A1 (Fig. 6F).

Fig. 6.

KRAS G12C inhibitors upregulate ALDH1A1 expression by inhibiting GTF2I phosphorylation at S784 via ERK and impair nuclear translocation. (A) qRT-PCR and WB showed the ALDH1A1 expression alteration following the treatment of sotorasib or adagrasib (IC50, 72 h), withdrawal of the inhibitors (72 h), and retreated with them (IC50, 72 h). (n = 3) (B) qRT-PCR and WB showed the ALDH1A1 expression decreased when treated with KRAS G12C inhibitors (IC50) or ERK inhibitor SCH7 72984 (IC50 = 50 nM), not AKT inhibitor MK-2206 (IC50 = 100 nM) for 72 h. (n = 3) (C) Genome-wide data of ALDH1A1 from the ENCODE database showed the ZNF784 and GTF2I binding peak in the promotor region close to the transcription start site (TSS) of ALDH1A1 after screening via DNA pull down with MS. (D–E) qRT-PCR (D) and WB (E) showed ZNF384, GTF2I, ALDH1A1 expression altered upon knockdown (KD) by three siRNAs together in MIAPACA2 and H2122 cells (n = 3). ALDH1A1 expression increased only in GTF2I-KD cells. (F) Dual-luciferase assays showed the fluorescence intensity of GTF2I-NC and KD in ALDH1A1-WT or mutated type of the promoter region in the two cell lines (n = 3). (G) qRT-PCR and WB showed no alteration of ALDH1A1 expression level when treated with sotorasib, adagrasib or SCH7 72984 (IC50) for 72 h in two cell lines (n = 3). (H) qRT-PCR showed ALDH1A1 expression altered upon treatment with inhibitors sotorasib (IC50), adagrasib (IC50), BRD7389 (for RSK, IC50 = 1.5 μM), RMM-46 (for MSK, IC50 = 200 nM), and ECT-206 (for MNK, IC50 = 70 nM) for 72 h in MIAPACA2 and H2122 cells (n = 3). ALDH1A1 expression increased when treated with KRAS inhibitors and RMM-46. (I-J) In vitro phosphorylation assay validated decrease of p-ERK, p-GTF2I at S784 and upregulated expression of ALDH1A1 in two cell lines treated with sotorasib, adagrasib or SCH7 72984 (IC50) for 72 h, and p-MSK, p-GTF2I at S784 and upregulated expression of ALDH1A1 with RMM-46, while ERK, MSK and GTF2I showed no difference (I) The alteration were abolished in GTF2I–S784A (a phospho-deficient mutant) and GTF2I–S784D (a phospho-mimetic mutant) group (J). (K) The immunofluorescence images and fluorescence intensity of each diagonal line in the selected area showed the GTFI location in MIAPACA2 and H2122 cells treated with sotorasib, adagrasib or SCH7 72984 (IC50) for 72 h (scale bars, 20 μm). Data were analyzed by Student's t-test and were presented by mean ± SD.

The expression of GTF2I did not change upon treatment with KRAS inhibitors (Fig. 6G), which inspired us to investigate post-translational modifications. We scanned PhosphoSite and found that S784 was the most frequently reported phosphorylation site of GTF2I. After enriching GTF2I with Co-IP, we discovered a significant decrease in GTF2I phosphorylation with a S784 phospho-specific antibody upon treatment with KRAS and ERK inhibitors (Fig. S3O). Validation showed that KRAS inhibitors decreased the phosphorylation of ERK and GTF2I–S784 (Fig. 6I). Based on GTF2I S784's regulation by basophilic kinases, we screened RSK, MSK, and MNK from ERK-regulated basophilic kinases [[38], [39], [40]]. Using inhibitors BRD7389 (for RSK), RMM-46 (for MSK), and ECT-206 (for MNK), we discovered that only MSK inhibitor upregulate ALDH1A1 (Fig. 6H). Validation showed RMM-46 reduced GTF2I phosphorylation without affecting ERK, while ERK inhibitor decreased MSK phosphorylation. Combination of RMM-46 with KRAS-targeted drugs or ERK inhibitor did not enhance effects (Fig. 6I), suggesting MSK operates within the ERK-GTF2I pathway. KRAS-targeted drugs block the ERK/MSK/GTF2I–S784 phosphorylation cascade, enabling ALDH1A1 upregulation. This effect was abolished in GTF2I–S784A (a phospho-deficient mutant) and -S784D (a phospho-mimetic mutant) mutants (Fig. 6J), confirming S784 dephosphorylation's role (Fig. 6J). GTF2I was reported to be a transcriptional repressor when phosphorylated [[41], [42], [43]]. As the immunofluorescence results exhibited, GTF2I S784-WT showed decreased nuclear translocation with the inhibitors. GTF2I was higher in the S784A cytoplasm and lower in the S784D cytoplasm, with no difference when KRAS and ERK were inhibited (Fig. 6K). In conclusion, we demonstrated that KRAS-targeted inhibitors suppress ERK/MSK and GTF2I–S784 phosphorylation, which block GTF2I nuclear translocation and upregulate ALDH1A1.

Taken together, our investigation substantiated that KRAS inhibitors lead to the dephosphorylation and the nuclear translocation of GTF2I via ERK/MSK, and the subsequent upregulation of ALDH1A1. The heightened enzymatic activity of ALDH1A1 inhibits ferroptosis and confers resistance to KRAS-targeted therapies both in vitro and in vivo, by the detoxification of aldehydes and activating RARA, inducing pHi reduction and activating CREB1-GPX4, as well as the formation of lipid droplet, which suggest a potential therapeutic strategy to improve the outcomes of patients with KRAS-mutant cancers (Fig. 7A).

Fig. 7.

Targeting ALDH1A1 to enhance the efficacy of KRAS-targeted therapy through ferroptosis (A) Mechanism diagram for ALDH1A1 promoting resistance to KRAS inhibitors through ferroptosis.

3. Discussion

KRAS-targeted drugs, such as sotorasib and adagrasib, are promising therapies for cancers with KRAS mutations, which are common in lung, pancreatic, and colorectal cancers. However, these inhibitors also face significant resistance problems due to KRAS secondary mutations [44,45], or off-target resistance (activation of upstream, downstream, or parallel pathways), histological transformation (adenocarcinoma to squamous-cell carcinoma), and tumor microenvironment changes [46]. Current studies are focusing on various strategies to overcome resistance, including combination blockade of upstream and downstream genes, like epidermal growth factor receptor (EGFR), Src homology 2 domain-containing protein tyrosine phosphatase (SHP2) [46], combination of other therapy such as carboplatin and pemetrexed [47], pembrolizumab or atezolizumab [48], and new pan-KRAS inhibitors and KRAS G12D inhibitors [26,27].

Previous studies have reported the KRAS mutant cells showed resistance to ferroptosis by reprogramming glutamine metabolism. Oncogenic KRAS were reported to upregulate glutamate oxaloacetate transaminases 1 and 2 (GOT1 and GOT2) and promote NADPH production via malic enzyme 1 (ME1) [49]. Also, KRAS mutant cancers have Nuclear Factor erythroid 2-Related Factor 2 (NRF2) antioxidant system activation [50] and thus upregulation of SLC7A11 [[51], [52], [53]], which protects cells from oxidative stress by increasing intracellular cysteine uptake and glutathione (GSH) synthesis and promoting ROS detoxification [19]. The GPX4 system, which is also upregulated in KRAS mutant tumors, is the most crucial intracellular mechanism against ferroptosis [54,55]. GPX4 uses GSH to reduce lipid hydroperoxides (L-OOH) to lipid alcohols (L-OH), thereby halting lipid peroxidation [56]. Additionally, KRAS mutation in lung cancer enhances the expression of acyl-coenzyme A synthetase long-chain family member 3 (ACSL3) to alter lipid metabolism, leading to monounsaturated fatty acid-containing phospholipid (MUFA-PL) production and ferroptosis resistance [57]. In this research, we demonstrated KRAS-targeted inhibitors inducing ferroptosis. We also found that the combination of KRAS inhibitors and FINs increased the cytotoxicity. Further investigation of the role of ALDH1A1 in ferroptosis and the enhancement of ferroptosis by inhibiting ALDH1A1 are significant for improving the efficacy of KRAS inhibitors.

Increasing evidence indicates that transcriptional activation of ALDH1A1 is associated with acquired resistance of tumor cells [58,59], such as in lung adenocarcinoma (LUAD), where ALDH1A1 mediates the upregulation of GPX4, modulating ROS metabolism and leading to resistance to tyrosine kinase inhibitors (TKIs) of EGFR [60]. Among the inducers of ferroptosis, ROS can initiate the generation of toxic aldehydes such as 4-HNE or MDA derived from lipid peroxidation, which are metabolized and regulated by ALDH. Previous reports have shown low ROS levels confer resistance to cancer cells [61,62]. Moreover, targeting the S100A9-ALDH1A1-RA signaling pathway can inhibit brain metastasis recurrence in lung cancer with EGFR mutations [63].

In our research, we demonstrated that ALDH1A1 inhibits KRAS inhibitors-induced ferroptosis by modulating multiple aspects of cellular metabolism, including detoxicating aldehydes, enhancing retinoic acid and NADH biosynthesis to maintain low ROS level, and activating RARA. We also demonstrated that ALDH1A1 decreased the pHi, which was upregulated by KRAS inhibitors, and acidity confers ferroptosis resistance. Chronic acidosis induces lipid droplet (LD) formation, while LDs affect ferroptosis sensitivity, such as LD-derived lipids fuel phospholipid synthesis [64], LD biogenesis prevents PUFA toxicity in breast cancer [65] and lipophagy-mediated LD degradation enhances GPX4 inhibition-induced ferroptosis in hepatocytes [66]. We demonstrated that GPX4 expression and lipid droplet formation were suppressed with higher pH and ALDH1A1 inhibition. At the same time, the HCL added to modulated the pHe rescued the suppression, indicating that ALDH1A1 modulates ferroptosis via pH.

Several issues warrant further discussion and investigation. One of the major limitations of this study is the scarcity of clinical samples from patients who underwent KRAS-targeted therapies, because they are still not approved in China. Another limitation is the incomplete understanding of the molecular mechanism by which KRAS induces ferroptosis. Further studies are needed to elucidate the role of ferroptosis in the response and resistance to KRAS-targeted therapies. In conclusion, our study provides important insights into the possibility of targeting ALDH1A1 to enhance the effectiveness of KRAS-targeted therapy through ferroptosis in various cancer treatment.

4. Methods

4.1. Cell culture and reagents

Lung adenocarcinoma (LUAD) cell line H2122 was obtained from Beyotime (Zhejiang, China); LU99 and LU65 were from ChuanQiu Biotechnology (Shanghai, China), and SW1573 were from Meisen Cell (Zhejiang, China). Colorectal adenocarcinoma (COADREAD) cell line SW837 was obtained from Beyotime; JVE-015 was from DSMZ (Braunschweig, Germany), and TT1TKB was from RIKEN BioResource Center (Ibaraki, Japan). Esophageal squamous cell carcinoma (ESCC) cell line KYSE410 was obtained from Meisen Cell. Other cell lines, including pancreatic adenocarcinoma (PAAD) cell lines MIAPACA2 and HPAC, LUAD cell line H1792, COADREAD cell line SW1463 and GP2D, bladder urothelial carcinoma (BLCA) cell line UMUC3 and human embryonic kidney 293 (HEK293T) cells were obtained from the cell bank of the Chinese Academy of Sciences. The cells were cultured in high-glucose DMEM (Hyclone, UT, USA) supplemented with 10 % fetal bovine serum (Every Green, Hangzhou, Zhejiang, China), 0.1 mg/ml streptomycin and 100 U/ml penicillin (Sangon Biotech, Shanghai, China) and were maintained in a humidified atmosphere with 95 % air and 5 % CO₂.

RSL3, Imidazole ketone erastin (IKE), KRAS G12C inhibitors sotorasib and adagrasib, KRAS G12D inhibitor MRTX1133, other reagents, including ferrostatin-1 (fer-1), Z-VAD-FMK (Z-VAD), deferoxamine (DFO), necrosulfonamide (necro), ERK inhibitor SCH7 72984, AKT inhibitor MK-2206, RSK inhibitor BRD7389 and MSK inhibitor RMM-46 were purchased from TargetMol (Shanghai, China). Pan-KRAS inhibitor BI-2493, RARA inhibitor AGN193109, and MNK inhibitor ECT-206 were purchased from MedChemExpress (Shanghai, China). RARA activator all-trans-retinoic acid (ATRA) was obtained from Beyotime (Zhejiang, China).

The sotorasib and adagrasib-resistant cells were established by exposing the cells to the inhibitors for 9 months and measuring the IC50 values. We obtained four resistant cell lines with more than a 10-fold increase in IC50 compared to the parental cells: MIAPACA2-Sotorasib-R, MIAPACA2-Adagrasib-R, H2122-Sotorasib-R, and H2122-Adagrasib-R.

4.2. RNA sequencing (RNA-seq) and bioinformatic analysis

RNA-seq was performed by oebiotech (Shanghai, China) as described before [23,67,68]. In brief, after MIAPACA2 and H2122 cells were treated with DMSO, sotorasib (sotorasib: MIAPACA2, IC50 = 40 nM; H2122, IC50 = 100 nM; adagrasib: MIAPACA2, IC50 = 60 nM; H2122, IC50 = 180 nM) for 72 h, the total RNA was extracted using TRIzol reagent (Tiangen, Beijing, China). RNA-seq and the following bioinformatic analyses were performed using the Illumina HiSeq platform (Illumina, USA) as previously described [69]. Analysis and visualization of the differentially expressed genes (DEG) were conducted using the limma and ggplot 2 packages in R software. We established the criteria as the p-value <0.05 and |log 2 fold change| > 0.5.

4.3. RNA extraction and quantitative real-time PCR (qRT-PCR)

RNA was extracted using TRIzol reagent (Tiangen) following the manufacturer's protocol. Complementary DNA was synthesized from RNA using the Hifair® II First-strand cDNA Synthesis Kit (YEASEN, Shanghai, China). The expression of mRNA was quantified by qRT-PCR using the Hieff® qPCR SYBR Green Master Mix (YEASEN, Shanghai, China) on the ABI QuantStudio 5 real-time PCR system (Thermo Fisher, Waltham, MA, USA). Each sample was replicated three times, and the mRNA expression was normalized to β-actin using the 2-ΔΔCT method. The primers were obtained from Sangon Biotech (Shanghai, China) and are shown in Table S3.

4.4. Western blot (WB) experiments

RIPA buffer (Beyotime) with a protease and two phosphatase inhibitor cocktails (TargetMol) were used to lyse cells on ice for 10 min and extract protein. BCA assay (YEASEN) was used to measure protein quantity, and SDS-PAGE (YAMAY Biotech, Shanghai, China) was used to separate protein before transferring it to PVDF membranes (Millipore, Billerica, MA, USA). The membranes were incubated overnight with a specific primary antibody at 4 °C after blocking with 5 % non-fat milk for 1 h. The membranes were washed 3 times with TBST solution and then incubated with HRP-labeled secondary antibody (1:2500, Beyotime) for 1 h. Moon Chemiluminescence Kit (Beyotime) was used to visualize protein bands. A polyclonal antibody against p-GTF2I S784 was obtained from ABclonal Technology. Briefly, rabbits were immunized five times with the phosphorylated peptide FRRP-(p-S)-TFG-C. Following the immunizations, the serum was collected and subjected to both positive and negative purification using the phosphorylated peptide and a control peptide (PFRRPSTFG-C), respectively. This process allowed us to obtain a polyclonal antibody specifically targeting the phosphorylation of GTF2I at S784. The effectiveness of this antibody was tested through dot blot experiments. The primary antibodies used in this study are shown in Table S4.

4.5. ALDH1A1 activity measurement

The ALDH Activity Assay Kit (Boxbio Science & Technology Co., Beijing, China) was used. The substrate AH was oxidized by ALDH to AcOH, while NAD⁺ was reduced to NADH + H⁺ in the reaction system. The NADH content was detected by its absorbance at 340 nm. The ALDH activity was expressed as nmol NADH/min per mg protein. Furthermore, ALDH activity in single tumor cells was measured using the ALDEFLUOR Kit (Stemcell Technologies, Canada) according to the manufacturer's instructions. BODIPY-aminoacetaldehyde, a non-toxic fluorescent substrate that can penetrate intact and viable cells, was metabolized by ALDH into BODIPY-aminoacetate, a fluorescent product that accumulated intracellularly. The fluorescence intensity was measured using an Accuri 6 cytometer (BD Biosciences, San Diego, USA).

4.6. Transfection of lentiviruses, siRNAs, and plasmids

Plasmids and lentivirus vectors of ALDH1A1-sgRNA (ALDH1A1-KO) were designed and constructed by Genechem Technology (Shanghai, China). Lentivirus vectors mediate the overexpression of ALDH1A1 with or without mutated K to Q/R at 193 residue of ALDH1A1 were constructed (ALDH1A1-OE, ALDH1A1-K193Q, ALDH1A1-K193R) based on the KO cell lines (MIAPACA2 and H2122) and were used to infect the cell lines. The 2.5 μM puromycin (Beyotime) was used to select stable cell lines for 48 h. Three siRNAs respectively targeting GTF2I (si-GTF2I-1, si-GTF2I-2, si-GTF2I-3), ZNF384 (si-ZNF384-1, si-ZNF384-2, si-ZNF384-3), CREB1 (si-CREB1-1, si-CREB1-2, si-CREB1-3), and two negative control siRNAs (siNC-1 and siNC-2) (Ribobio, Guangzhou, China) were designed and transfected into cells using lipo 8000 (Beyotime) and Opti-MEM (Thermo Fisher Scientific). GTF2I-WT, GTF2I–S784A and GTF2I–S784D vectors were transdected into HEK293T cells. All the information on sgRNA, siRNA, and primers is listed in Table S3.

4.7. Cytotoxicity assays

Cells were seeded in 96-well plates at 3000 cells per well and incubated for 24 h for adhesion and proliferation. Various doses of KRAS-targeted inhibitors, RSL3, IKE, fer-1, Z-VAD, DFO, necro, CM10, ATRA, AGN193109, were used to treat the cells for 72 h. The Cell Counting Kit-8 (TargetMol) was used to gauge the activity of cellular dehydrogenases as an indicator of cell viability. The optical density (OD) was read by a microplate reader, and the number of viable cells in each well was calculated.

4.8. Establishment of patient-derived organoids (PDO) and drug sensitivity tests

Fresh tumor tissues from PAAD and LUAD patients with pathological diagnosis who underwent surgery at Zhongshan Hospital, Fudan University, were obtained with approval from the hospital's ethics committee (B2022-180 R). Tissues were minced, washed with cold buffer, and digested enzymatically using PDO culture media kit (Absin, Shanghai, China) for human pancreatic cancer and lung cancer. The digest was centrifuged, and the pellet was mixed with matrigel (YEASEN) at a 25:1 ratio. The mixture was seeded in 24-well plates and solid field at 37 °C for 10 min. PDO culture media (500 μL) was added to each well to generate PDOs, which were fixed in 4 % paraformaldehyde for HE stain. The knockout of ALDH1A1 in PDOs was conducted by CRISPER/Cas-9 and the overexpression using lentivirus. Drug sensitivity was tested by incubating PDOs in 96-well plates with the drugs for 72 h. Cell viability was measured using the CellTiter-Lumi Luminescent 3D Assay Kit (Beyotime) (Fig. 2A).

4.9. Animal experiments

Animal experiments followed the Animal Ethics Committee guidelines of Zhongshan Hospital, Fudan University (B2023-181 R). Nude mice were kept in a pathogen-free laminar flow cabinet. MIAPACA2 cells (2 × 10⁶) expressing NC, ALDH1A1-KO, and ALDH1A1-OE were suspended in 100 μL phosphate-buffered saline (PBS) and injected subcutaneously into the right flank of 6-week male mice. Once tumors reached approximately 150 mm³ in size, mice were randomized into three treatment groups: vehicle, sotorasib (30 mg/kg), or adagrasib (30 mg/kg), and received oral gavage daily for 2 weeks (Fig. 2H). Mice were injected subcutaneously with MIAPACA2 cells (2 × 10⁶) and were treated with vehicle, sotorasib (30 mg/kg), or adagrasib (30 mg/kg) with or without CM10 (10 mg/kg), which was administered intraperitoneally every other day for 7 times (Fig. 2L). Tumor size was measured biweekly, and tumor volume was calculated as (length × width²)/2. Mice were euthanized after 2 weeks of drug administration.

4.10. Detection of lipid peroxidation and cell iron

A variety of agents were used to pretreat cells cultured in 12-well plates. The cells were then rinsed with PBS (Beyotime) and suspended in DMEM medium with 200 μM BODIPY 581/591C¹¹ dye (Thermo Fisher) for lipid peroxidation analysis. After incubating at 37 °C for 30 min, the cells were washed 3 times with PBS. The FITC channel of the Accuri 6 cytometer (BD, USA) was used to measure lipid peroxidation levels. FerroOrange dye (1 μM, Dojindo Molecular Technologies, Kumamoto, Japan) was used to incubate the cells for 30 min at 37 °C to detect Fe²⁺. Fluorescence was measured using the PE channel of the same cytometer. FlowJo software (TreeStar Inc, OR, USA) was used to process and visualize data.

4.11. Retinol, RA, malondialdehyde (MDA), and 4-hydroxy-2-nonena (4-HNE) measurement

The extent of ferroptosis was evaluated by quantifying the concentrations of retinal, RA, MDA, and 4-HNE in each experimental group. The Retinaldehyde ELISA Assay Kit (JIwei Biological Technology, Shanghai, China), RA ELISA Assay Kit (JIwei Biological Technology, Shanghai, China), MDA Assay Kit (Beyotime), and Lipid Peroxidation (4-HNE) Assay Kit (Abcam) were used to measure the levels of AH, MDA, and 4-HNE in cell lysates, following the manufacturer's instructions.

4.12. Transmission electron microscopy

Cell structure was preserved by crosslinking proteins in cells that were cultured in dishes of 6-cm diameter using a solution of glutaraldehyde. The cells were rinsed 3 times with 0.1 M PBS (pH 7.4) to eliminate any remaining glutaraldehyde and then fixed further with PBS containing osmium tetroxide. The samples were dehydrated in ethanol solutions of increasing concentration, embedded in resin, and cured for 48 h at 60 °C in an oven. Ultrathin sections were sliced with an ultramicrotome and stained with uranyl acetate and lead citrate to enhance visibility. The sections were dried for a night and then examined under a transmission electron microscope (Hitachi, Japan).

4.13. Co-Immunoprecipitation Mass Spectrum (Co-IP-MS)

The cells were cultured in 9-cm dishes and were lysed by ice-cold NP-40 buffer (Beyotime) containing protease and phosphatase inhibitors (TargetMol). The lysates were centrifuged at 13,000×g for 15 min at 4 °C. Pierce Crosslink Magnetic IP/Co-IP Kit (Thermo Fisher) was used to conduct the Co-IP according to the manufacturer's instructions. Briefly, the protein A/G magnetic beads cross-linked with ALDH1A1 (1:50 dilution, Abways) or IgG antibodies were added to the lysates and incubated overnight at 4 °C with gentle rotation. The beads were washed 3 times with NP-40 buffer and collected with a magnetic stand. The eluted proteins were analyzed by MS (oebiotech).

4.14. Detection of NAD⁺/NADH

The NAD⁺/NADH Assay Kit (Beyotime) was used to measure the NAD⁺/NADH ratio following the manufacturer's instructions. MIAPACA2 and H2122 cells were collected, rinsed with PBS, and lysed with NAD⁺/NADH extraction buffer. The lysates were centrifuged at 12,000×g for 5 min at 4 °C and split into two parts to measure NADH alone and total NADH (comprising NAD⁺ and NADH). An ALDH working solution was mixed with the liquid part of each sample and incubated at 37 °C for 10 min. The chromogen solution was added and incubated in the dark at 37 °C for 30 min. The OD was measured at 450 nm, and the NAD⁺/NADH ratio was calculated y as (Total NADH - NADH)/NADH. The experiment was performed 3 times.

4.15. Dual-luciferase reporter analysis

Luciferase reporter gene assays were performed using the promotor regions of ALDH1A1 and GPX4. The promoter region of ALDH1A1 spanning from −1200 to +300 base pairs and GPX4, spanning from −900 to +300 base pairs relative to the transcription start site (TSS), as well as the corresponding mutant sequence, was cloned into the phy-811 luciferase reporter vector (Hanyin Technology, Shanghai, China). HEK-293T cells with either NC and GTF2I knockdown (KD) were co-transfected with ALDH1A1-WT, or mutant plasmids, while the CREB1-KD was transfected with GPX4-WT or mutant plasmids. A renilla luciferase reporter plasmid was used as an internal control. Lipo8000 (Beyotime) was used as the transfection reagent, and the cell confluency was kept at 60–80 %. After 48 h post-transfection, dual-luciferase reporter assays were carried out with a Luciferase Reporter Gene Assay kit (Beyotime) following the manufacturer's instructions. The measurement of luciferase activity was performed using a FlexStation 3 Microplate Reader (Molecular Devices, San Jose, California, USA). The activity of RARA was tested using the pRAR-TA-Luc (Beyotime).

4.16. Live cell staining of LDs

To measure LD formation, cells were plated in 12-well plates and exposed to 0.1 mmol Lipi-Green (Dojindo Molecular Technologies), a fluorescent dye that stains neutral lipids, for 1 h at 37 °C and 5 % CO2. Fluorescence images of the stained cells were acquired using an Olympus IX71 microscope (Olympus, Japan). The LD area per cell was calculated by normalizing the total LD area by the cell number with Image J software.

4.17. Chromatin immunoprecipitation (ChIP) analysis

The SimpleChIP® Plus Enzymatic Chromatin IP Kit (Cell Signaling Technology, USA) was used to perform the ChIP according to the manufacturer's protocol. Briefly, cells were crosslinked with formaldehyde to preserve DNA-protein interactions. Chromatin was digested with micrococcal nuclease to generate 150–900 bp DNA/protein fragments. Immunoprecipitation was performed using either control IgG or anti-CREB1 antibody (1:50 dilution, Abways). Protein-DNA complexes were collected by protein G magnetic beads, and eluted chromatin was used to crosslink reversal. DNA fragments were isolated with spin columns and quantified by qRT-PCR with the following primer pairs: BS1-Primer: F: CGCAGTCGCCAACAACAAGTC; R: TTTCCGCGCCTCCTTTCCCA; BS2-Primer: F: GCGAGTTGGAGAAACCAAACCC, R: ATGCTCGCTTGTGTCTAGGAGG.

4.18. DNA pull-down and mass spectrum

The DNA probes of ALDH1A1 promotor region from Genechem Technology was amplified with TaKaRa LA Taq® with GC Buffer (Takara Bio, Kyoto, Japan). Biotin or non-biotin groups were used to label the probes as described before [70]. The amplified probes were purified by gel extraction after DNA electrophoresis to increase their purity with the SanPrep Column DNA Gel Extraction Kit (Sangon Biotech). Southern blot was conducted to determine the concentration of probes. The labeled DNA probes were incubated with magnetic beads for 30min at room temperature with gentle rotation. We washed the beads three times with BS/THES buffer (THES buffer:50 mM Tris-HCl, pH 7.5, 10 mM EDTA, 140 mM NaCl, 20 % sucrose, 5 × BS buffer: 50 mM HEPES pH 7.5, 25mMCaCl2, 250 mM KCl, 60 % glycerol) to remove unbound DNA. We added the cell lysates to the beads and incubated them overnight at 4 °C with gentle rotation. The beads were washed 3 times with washing buffer to remove non-specific proteins. The eluted proteins were analyzed by mass spectrum (oebiotech).

4.19. Immunofluorescence

The cells were fixed with 4 % paraformaldehyde for 15 min at room temperature and permeabilized with 0.5 % Triton X-100 for 20 min. The non-specific binding sites were blocked with 1 % bovine serum albumin (BSA) for 1 h and the cells were incubated with primary antibodies of GTF2I (1:100, Abways) overnight at 4 °C. The cells were washed 3 times with PBST and incubated with secondary antibodies conjugated with fluorescent dyes for 1 h at room temperature. The cells were washed 3 times with PBST and were mounted with a DAPI-containing mounting medium (Beyotime). We acquired fluorescence images using an Olympus IX71 microscope (Olympus).

4.20. Data analysis

All data analysis was performed using GraphPad Prism 9 (GraphPad Software, La Jolla, CA) and R (version 4.1.2). Reads were mapped to genes using TopHat (v.2.0.13) and HISAT2 (v.2.1.0), and raw data were normalized to fragments per kilobase of exon per million mapped reads (FPKM) for downstream analysis. Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis on DEGs between ALDH1A1 high and low groups downloaded from TCGA were performed using the ‘cluster profile’ R package. Adjusted p-value<0.05 and false discovery rate (FDR) < 0.05 were set as the cutoff. Continuous variables were compared using Student's t-test or two-way ANOVA, as appropriate. A p-value of less than 0.05 was considered statistically significant.

Ethics approval and consent to participate

All patients included in the study have signed informed consent in compliance with the Declaration of Helsinki. This study was approved by the Research Ethics Committee of Zhongshan Hospital, Fudan University (approval number: B2022-180 R).

Funding

This research was funded by The Biomedical Technology Supporting Foundation of Shanghai (No. 22S11900300), the Natural Science Foundation of Shanghai (No. 22ZR1411900) and the Science and Technology Commission of Fujian Province of China (No. 2021D014).

Data availability

TCGA database (https://gdc.xenahubs.net), ASPAR network resource (https://jaspar.genereg.net/), ENCODE database (https://www.encodeproject.org/), PhosphoSitePlus (https://www.phosphosite.org/homeAction.action). The inquiry of original data can be directed to the corresponding authors for rational reasons.

CRediT authorship contribution statement

Yunyi Bian: Conceptualization, Investigation, Visualization, Writing – original draft, Writing – review & editing, Formal analysis. Guangyao Shan: Methodology, Software, Validation. Guoshu Bi: Data curation, Formal analysis, Visualization. Jiaqi Liang: Methodology, Resources, Software. Zhengyang Hu: Validation, Visualization. Qihai Sui: Data curation, Investigation, Methodology, Resources. Haochun Shi: Formal analysis, Validation, Visualization. Zhaolin Zheng: Formal analysis, Software. Guangyu Yao: Data curation, Funding acquisition. Qun Wang: Funding acquisition, Project administration, Supervision. Hong Fan: Conceptualization, Funding acquisition, Project administration, Supervision. Cheng Zhan: Conceptualization, Project administration.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: