ATF4 renders human T-cell acute lymphoblastic leukemia cell resistance to FGFR1 inhibitors through amino acid metabolic reprogramming

Abnormalities of FGFR1 have been reported in multiple malignancies, suggesting FGFR1 as a potential target for precision treatment, but drug resistance remains a formidable obstacle. In this study, we explored whether FGFR1 acted a therapeutic target in human T-cell acute lymphoblastic leukemia (T-ALL) and the molecular mechanisms underlying T-ALL cell resistance to FGFR1 inhibitors. We showed that FGFR1 was significantly upregulated in human T-ALL and inversely correlated with the prognosis of patients. Knockdown of FGFR1 suppressed T-ALL growth and progression both in vitro and in vivo. However, the T-ALL cells were resistant to FGFR1 inhibitors AZD4547 and PD-166866 even though FGFR1 signaling was specifically inhibited in the early stage. Mechanistically, we found that FGFR1 inhibitors markedly increased the expression of ATF4, which was a major initiator for T-ALL resistance to FGFR1 inhibitors. We further revealed that FGFR1 inhibitors induced expression of ATF4 through enhancing chromatin accessibility combined with translational activation via the GCN2-eIF2α pathway. Subsequently, ATF4 remodeled the amino acid metabolism by stimulating the expression of multiple metabolic genes ASNS, ASS1, PHGDH and SLC1A5, maintaining the activation of mTORC1, which contributed to the drug resistance in T-ALL cells. Targeting FGFR1 and mTOR exhibited synergistically anti-leukemic efficacy. These results reveal that FGFR1 is a potential therapeutic target in human T-ALL, and ATF4-mediated amino acid metabolic reprogramming contributes to the FGFR1 inhibitor resistance. Synergistically inhibiting FGFR1 and mTOR can overcome this obstacle in T-ALL therapy.


INTRODUCTION
T-cell acute lymphoblastic leukemia (T-ALL), a hematological malignancy derived from T-cell progenitors, accounting for approximately 10%-15% of pediatric and 20%-25% of adult ALL cases [1,2].Although the overall survival rate with current therapeutic progress has reached 80% in pediatrics but is <50% in adults, the relapsed or refractory symptoms are always tough challenges for T-ALL treatment [3,4].Therefore, a better understanding of abnormalities and drug resistance is essential for T-ALL precision therapy.
The fibroblast growth factor receptor 1 (FGFR1), which belongs to the receptor tyrosine kinase (RTK) superfamily [5,6], can regulate the proliferation, differentiation, and survival of cells after binding to its ligands (FGFs) [7,8].Aberrant FGFR1 has been reported in many cancers, including urothelial cancer [9], breast cancer [10], lung cancer [11], ovarian cancer [12], and Ewing sarcoma [13].In hematological malignancies, abnormalities of FGFR1 were described in stem cell leukemia-lymphoma syndrome (SCLL), a typical myeloproliferative disorder that has been called myeloid/lymphoid neoplasms with eosinophilia and rearrangement of FGFR1 in the recent WHO classification [14].These patients possibly deteriorated to B-or T-lymphoma/leukemia or AML and often had a poor prognosis [14,15].As described, ZNF198-FGFR1 and FGFR1OP2-FGFR1 could induce myeloproliferative disorder and T-lymphoblastic leukemia in mice [16,17].CUX1-FGFR1 fusion was detected in T-lymphoblastic leukemia/ lymphoma, which could be specifically repressed by the TKI-258, an inhibitor of receptor tyrosine kinase (RTKi) [18].These results suggest that the continuous activation of FGFR1 contributes to the formation of T-cell leukemia/lymphoma and that FGFR1-targeted therapy is a promising strategy for the treatment of T-cell malignancies with aberrant FGFR1 expression.
FGFR1 inhibitors exhibit potential tumoricidal activity in multiple tumors, but the intrinsic and acquired drug resistance is also serious challenge for FGFR1-directed targeted therapy [19].As reported, the V561M mutation, a mutation in the ATP binding site of FGFR1, could reduce the binding affinity of FGFR1 inhibitors with FGFR1 and contribute to the drug resistance [20].The deletion of exon 6 in the PTEN gene was found in resistant FGFR1-driven leukemia cells, leading to the premature termination of PTEN translation and the upregulation of the PI3K-Akt signaling pathway [21].Additionally, the activation of PUMA protein was compromised after leukemia cells acquired resistance against FGFR1 inhibitors in SCLL, which led to the incapacity of TKI-induced apoptosis [22].However, these mechanisms cannot fully encapsulate the mechanism of drug resistance, and a deeper understanding is still needed.
In our study, we found that the FGFR1 was upregulated and was essential for the survival and proliferation of T-ALL cells, whereas the leukemia cells had intrinsic resistance to FGFR1 inhibitors.Mechanistically, we identified that ATF4 initiated the survival-promoting amino acid metabolism reprogramming and boosted the activation of mTORC1, which contributed to drug resistance and survival.We also revealed the major causes of ATF4 upregulation at both the translation and transcription levels, and further found that targeting mTORC1 could synergistically overcome the drug resistance.Briefly, we addressed the phenotypic, molecular, and metabolic responses of human T-ALL cells to FGFR1 targeting and further proposed an optimal combined strategy for FGFR1-directed therapy.

Gene knockdown
The gene knockdown was performed through short hairpin RNA (shRNA) based on lentiviral vectors as described previously [23].The sequences of shRNAs were shown in Supplementary Table 3.

Western blot analysis
To examine the expression and phosphorylation of specific proteins, the total proteins from different samples were used for Western blot analysis as described previously [24].Antibodies used in this study were listed in Supplementary Table 4.

Flow cytometry
The flow cytometry protocol has been described previously [23,25].The cells of bone marrow were collected and washed for three times, then 1 million cells were incubated with antihuman CD7 antibody (BD Biosciences, 562635) for 30 min at 4 °C.After washing three times, each sample was resuspended with PBS containing 7-AAD for dead cell exclusion.Then, the samples were detected in Beckman CytoFLEX system.

Surface sensing of translation (SUnSET) assay
To examine the efficiency of protein synthesis we performed the SUnSET [26].In brief, the samples were incubated with puromycin (10 µg/ml) for 30 min in a cell culture incubator, then were washed with PBS three times and harvested for total protein extraction.The translational efficiency was detected using an anti-puromycin antibody (1:10000), and other steps were consistent with the Western blot protocol.
ATAC-seq analysis ATAC-seq analysis has been described previously [27].Briefly, Jurkat cells were treated with AZD4547 (2 μM) for 96 h, and ATACseq libraries were generated with Hyperactive In-Situ ChIP Library Prep Kit for Illumina (Vazyme, TD 901) according to its protocol, then the ATAC-seq libraries were sequenced on PE150 dataset.
The clean data were generated by Trim Galore.ATAC-seq reads were aligned to the hg38 genome with Bowtie2 [32].Then, BAM files were generated using Samtools [33], PCR duplicates were removed by Sambamba [34] and mitochondrial genomes were filtered out by Samtools [35].Signal peaks were generated by MACS2 [36].The motifs analysis was performed by using findMotifs.pl of Homor [37].

Targeted metabolomics analysis
We analyzed amino acids and their metabolites using the targeted metabolomics method.Briefly, 3 million cells were harvested and resuspended in 500 μL precooled 80% methanol/water, further suffered repeated freezing and thawing three times, the vortex was needed after thawing in each freeze-thaw cycle.The samples were centrifuged at 12,000 g/min for 10 min at 4 °C, then the supernatant was collected and centrifuged again.After centrifugation, the supernatant was transferred to Protein Precipitation Plate (Thermo Scientific, 90036) preparing for further LC-MS analysis by an LC-ESI-MS/MS system (UPLC, ExionLC AD, MS, QTRAP® 6500 + LC-MS/MS) with ACQUITY BEH Amide (2.1 × 100 mm, 1.7 μm).
Mass spectrometric data were analyzed using Analyst and MultiQuant.The species and quantity (ng/mL) of metabolites were calculated through each standard curve, and were normalized according to their protein concentration.Differential metabolites were selected with P < 0.05 and |log2 FoldChange | > 0.3.Metabolites were annotated by the KEGG compound database and mapped to the KEGG pathway database.Then the related significant pathways were identified by metabolite sets enrichment analysis (MSEA), and P values were calculated by hypergeometric test.

Drug screening
To screen the synergistic drugs with AZD4547, we screened the Approved Drug Library (TargetMol, L1000) using Jurkat cells.In brief, the 2 × 10 4 Jurkat cells were seeded in flat tissue culture plates (96 wells) and were treated with AZD4547 (2 μM), each drug of the library (3 μM) or combination of these two drugs for 72 h.For the second round of screening, the combination of each drug was annotated specifically (Supplementary Table 2).The DMSO was added synchronously as the negative control.To examine the cell viability, the cells were incubated with Cell Counting Kit-8 (Yeasen, 40203ES92) following its protocol, and then detected at 450 nm (OD).The synergy of AZD4547 combined with each drug was measured by the CDI (coefficient of drug interaction), and the CDI < 1 indicated a synergistic effect [38].Targeted pathways of each drug were annotated according to the product manual, and the first pathway was selected when the drug targeted multiple pathways.The P-value of pathway enrichment was calculated by the hypergeometric test.

Mouse models
The 6 to 8-week-old NCG mice (Strain NO.T001475, NOD/ ShiLtJGpt-Prkdc em26Cd52 Il2rg em26Cd22 /Gpt) were purchased from GemPharmatech (Nanjing, China).This study in vivo was approved by the Animal Ethics Committee of Wuhan University of Science and Technology.The protocol for Jurkat-luciferase cell-derived xenograft CDX models has been described previously [23].CDX mice were treated with AZD4547 30 mg/kg every 2 days (Aladdin, 1035270-39-3), PD-166866 30 mg/kg every 2 days (MCE, HY-101296), rapamycin 3 mg/kg every 2 days (MCE, HY-10219) or combination of AZD4547 and rapamycin intraperitoneally, PBS buffer was used as control.The progression of T-ALL was followed through imaging weekly using Imaging Systems (Spectral Instrument System, Ami HTX).The endpoint of the experiments was that weight loss exceeded 20% of the body weight of a similar normal animal.Then these mice were euthanized and the cells of bone marrow were collected and the invasion of T-ALL malignancies was detected with anti-human CD7 antibody (BD Biosciences, 562635).Statistical analysis Specific statistical analyses have been described in each section.For cell experiments, statistical significance between two groups was performed by the unpaired Student's t-test.one-way ANOVA was used to examine more than two groups.Three independent replicates have been performed unless otherwise specified, and data are represented as means ± SD.Statistical analyses were calculated using GraphPad Prism.For clinical survival data analysis, the optimal cutoff values were determined using the maximally selected rank statistics, and the survival curves were tested with the log-rank test.

FGFR1 is upregulated in T-ALL and associated with dismal prognosis in patients
To identify potential targets for the treatment of T-ALL, we analyzed gene expression profiles of our previous T-ALL cohort [39] and another T-ALL RNA-seq dataset [40], and identified 226 genes and 264 genes that were upregulated in T-ALL, respectively (Fig. 1a and Supplementary Fig. 1a).Thirty-five common genes were shared by both upregulated groups, several of which were involved in the progression or precision therapy of T-ALL as previously reported, such as STMN1, CD99, SOX4, MYB, CDK6 and TOX (Supplementary Fig. 1b).However, the functional roles of FGFR1, one of the significantly upregulated 35 genes, remain largely unclear in T-ALL.(Fig. 1b, c).We also detected the expression of the FGFR family (FGFR1-4) in T-ALL, and found that only FGFR1 was expressed in T-ALL (Supplementary Fig. 1c).
Next, we confirmed that FGFR1 was upregulated in T-ALL cell lines and primary T-ALL blasts compared to normal T cells at both transcriptional and translational levels (Fig. 1d, e).Since the upregulation of FGFR1 has been reported in numerous malignancies [41], we summarized the transcription level of FGFR1 in various types of leukemia or lymphoma through the Cancer Cell Line Encyclopedia database (CCLE), and revealed that the FGFR1 was most expressed in T-ALL compared with other hematopoietic malignancies (Fig. 1f).Importantly, T-ALL patients with higher FGFR1 expression were more prone to relapse or other adverse events and had shorter survival times than patients with lower FGFR1 expression (TARGET, phs000464) (Fig. 1g, h).The comparable results were observed in a B-ALL cohort (TARGET, phs000463) (Supplementary Fig. 1d, e).Taken together, FGFR1 was substantially elevated in T-ALL and negatively related to the prognosis of the T-ALL patients.
FGFR1 is a potential therapeutic target in T-ALL, while leukemia cells are resistant to FGFR1 inhibitors Subsequently, we examined the functional roles of FGFR1 in T-ALL.After silencing the FGFR1 in human T-ALL cell lines through RNA interference, the proliferation and survival of T-ALL cells were observably impaired (Fig. 2a-d).Then, we inhibited the function of FGFR1 through its inhibitors and found that all T-ALL cell lines were insensitive to the FGFR1 inhibitor AZD4547 (IC 50 > 3 µM) and was unresponsive to another FGFR1 inhibitor PD-166866 compared to normal T cells (low FGFR1 expression) (Fig. 2e, f).T-ALL cells still maintained survival and proliferation under long-term observation compared to those that were completely inhibited in the FGFR1 knockdown group (Fig. 2g).However, the phosphorylation of Akt and S6 (classical downstream kinases of FGFR1 [8]) was reduced after AZD4547 treatment, which was consistent with the FGFR1 knockdown group (Fig. 2h, i).Additionally, we examined the IC 50 of FGFR1 inhibitors (AZD4547 and PD-166866) and protein levels of FGFR1 in different T-ALL cell lines, and found that there were significant associations between the efficacy of FGFR1 inhibitors and protein levels of FGFR1 (Supplementary Fig. 2a, b).The inhibition by AZD4547 was not obvious after FGFR1 knockdown (Supplementary Fig. 2c).Together, these results suggested that although the FGFR1 signaling was specifically inhibited by its inhibitor, T-ALL cells still exhibited pro-survival ability.
To further evaluate the antileukemic efficiency of blocking FGFR1 in vivo, we established cell-derived xenografts based on fluorescence-labeled Jurkat or FGFR1-knockdown Jurkat cells.As the schematic diagram shown, AZD4547 (30 mg/kg every 2 days) or PD-166866 (30 mg/kg every 2 days) was administered from Day 14 to Day 28 (Fig. 2j).Neither AZD4547 nor PD-166866 suppressed the growth of human T-ALL cells, while the FGFR1 knockdown significantly inhibited the progression of T-ALL cells in vivo (Fig. 2k and Supplementary Fig. 2d).Similarly, the survival time of xenografts was extended in the FGFR1 knockdown group, but not in the AZD4547 or PD-166866 treatment groups (Fig. 2l).We also analyzed the bone marrow invasion through flow cytometry analysis, and found that neither of the FGFR1 inhibitors significantly prevented the leukemia cell invasion into the bone marrow compared to the FGFR1 knockdown group (Fig. 2m, n).Together, FGFR1 was essential for T-ALL progression but the leukemia cells were resistant to FGFR1 inhibitors.

ATF4 is essential for the resistance against FGFR1 inhibitors
The vastly different outcomes of the FGFR1 inhibitor groups and FGFR1 knockdown group both in vitro and in vivo prompted us to explore the pro-survival causes of the inhibitor groups (Fig. 2g, k).We performed RNA-seq in Jurkat cells after FGFR1 knockdown or treatment with FGFR1 inhibitors.The differentially expressed genes (DEGs) compared with the control group were summarized in the heatmap, and the DEGs were similar between the AZD4547 and PD-166866 treatment groups (Fig. 3a).A Venn diagram depicted that 598 DEGs were specifically enriched in the two FGFR1 inhibitor groups, but not in the FGFR1 knockdown group (Fig. 3b).Then, we focused on the 570 differentially expressed genes (DEGs) and found that these were significantly enriched in several pathways about survival-promoting metabolism, including amino acid metabolism, carbon metabolism and glucose, and fructose metabolism (Fig. 3c).However, the enriched pathways of the DEGs in the FGFR1 knockdown group were the transcriptional misregulation in cancer and the FoxO signaling pathway, which are related to gene expression and cell fate decisions [42,43], and were dramatically different from those pathways in the groups treated with FGFR1 inhibitors (Supplementary Fig. 3a).
To determine whether there was a sponsor gene to drive these transformations of 570 DEGs, we analyzed the upstream transcription factors (TFs) through ChEA3 [30].Among the predicted TFs, the ATF4 scored the highest with the minimum P-value (1.82 × 10 −16 ) and maximum gene coverage (Fig. 3d).However, the TFs enriched by DEGs in the FGFR1 knockdown group were obviously distinct (Supplementary Fig. 3b).The transcription and protein levels of ATF4 was gradually increased over time after AZD4547 and PD-166866 treatment in human T-ALL cell lines (Fig. 3e-g).Additionally, we found that FGFR1 inhibitor could not induce ATF4 upregulation after FGFR1   knockdown (Supplementary Fig. 3c), suggesting that the upregulation of ATF4 was not independent of FGFR1.The Jurkat cells were more sensitive to AZD4547 after ATF4 knockdown than in the control group (Fig. 3h).We also found that the antileukemic efficiency of AZD4547 was increased with ATF4 knockdown in vivo (Fig. 3i, j and Supplementary Fig. 3d).To amplify the feature and make it feasible to identify critical factors of drug resistance, we further developed a more resistant Jurkat cell line (Jurkat-AZD) through continuous exposure to AZD4547 for a long time (Supplementary Fig. 3e).Consistently, the Jurkat-AZD cells became less resistant to AZD4547 after ATF4 knockdown (Fig. 3k).These results indicated that ATF4 was important for the resistance of T-ALL cells to FGFR1 inhibitors.

ATF4 is induced by enhanced chromatin accessibility combined with GCN2-mediated translational activation
To investigate the molecular mechanism underlying ATF4 upregulation after FGFR1 inhibitor treatment, we performed ATAC-seq with high spatial resolution.Overall, the average ATAC-seq signal around the transcription start sites (TSSs) was increased after AZD4547 treatment (Fig. 4a, b).The distribution of the average ATAC-seq signal in different regions also changed, especially increased in the promotor regions (Fig. 4c).We identified the significant active TFs (top 20) based on RNA-seq data (Fig. 4d), and screened the motifs of transcription factors that were significantly enriched in the promotor regions using ATACseq data synchronously (Fig. 4e).Together with 570 DEGs in both FGFR1 inhibitors groups, we found that only two common TFs were shared in all three groups, and one was ATF4 (Fig. 4f).These results were consistent with our above data that ATF4 was at the vital place for initiating these transcriptional changes.Subsequently, we examined the change in the ATF4 promotor region and found that the chromatin accessibility was enhanced after AZD4547 treatment (Fig. 4g).Additionally, the chromatin accessibility around some of typical genes (ASS1, ASNS, SLC1A5, and GARS1) was enhanced after AZD4547 treatment correspondingly (Supplementary Fig. 4a-d).Together with the upregulation of ATF4 mRNA (Fig. 3e), these data indicated that increased chromatin accessibility in its promotor region was a potential cause of the transcriptional upregulation of ATF4.
The translation of ATF4 can be induced by eIF2α through the integrated stress response (ISR) when cells suffer multiple survival pressures [44].Amino acid deprivation and endoplasmic reticulum (ER) stress can activate eIF2α through GCN2 and PERK, respectively, which are closely related to amino acid metabolism and proteostasis [45].Here, we examined the possible involvement of these kinases in the initiation of ATF4 translation.Based on 570 DEGs from RNA-seq, we observed that the amino acid deprivation pathway was highly enriched, which is closely related to the GCN2 kinase (Fig. 4h).The increase of ATF4 was ceased after GCN2 knockdown (Fig. 4i), but was not entirely suppressed after PERK knockdown (Fig. 4k).We further functionally determined the role of GCN2 in drug resistance and found that Jurkat cells were more sensitive to AZD4547 after GCN2 knockdown (Fig. 4j), but not after PERK knockdown (Fig. 4l).Similar results were obtained when we combined AZD4547 with the GCN2 inhibitor SP600125 [46,47] (Fig. 4m).These results indicated that the translational activation of ATF4 mediated by FGFR1 inhibitors was mainly due to the GCN2-eIF2α pathway.

ATF4 is a crucial initiator that drives the reprogramming of amino acid metabolism
We have indicated that the DEGs in the FGFR1 inhibitor groups were enriched in metabolic pathways (Fig. 3c) and identified that the ATF4 was the initiator of these transformations (Fig. 3d and 4f).Here, we further identified that these typical DEGs were mainly divided into three categories, including kinases related to metabolism, transporters, and aminoacyl-tRNA biosynthesis, and all of these DEGs were significantly elevated in both groups treated with FGFR1 inhibitors (Fig. 5a).The mRNA levels of these typical genes were dramatically increased and blocked after knockdown of ATF4 (Fig. 5b-d).Next, we selected several typical genes to confirm the tendencies in the protein levels, and found that the protein levels of ASNS, ASS1, PHGDH, and SLC1A5 were significantly increased, with the protein of ATF4 increasing ahead of these proteins (Fig. 5e), In addition, these upregulations of proteins were similar in the more resistant Jurkat cells (Jurkat-AZD) (Fig. 5f).These AZD4547 induced upregulations were also interdicted after ATF4 knockdown (Fig. 5g).Together, these results suggested that the upregulation of these metabolic genes was induced by ATF4 when cells were exposed to FGFR1 inhibitors.
Since these typical DEGs had relevance to amino acid biosynthesis and uptake, we next quantified the intracellular amino acids and their metabolites through the targeted metabolomics analysis (Supplementary Table 1).Plentiful amino acids and metabolites significantly increased in the more resistant Jurkat cells (Jurkat-AZD) and came down after ATF4 knockdown.In particular, Asn, Arg and other essential amino acids (EAAs) that were related to ASNS, ASS1, and SLC1A5 (Fig. 5h and Supplementary Fig. 5a).Subsequently, we systematically identified that these differential amino acids, and found that these differential metabolites were mainly enriched in the pathways related to central carbon metabolism (related to PHGDH [48]), alanine/aspartate/glutamate/glycine/serine/threonine metabolism, mineral absorption and the TCA cycle, all of which are essential for cell survival and proliferation [49].Importantly, these metabolic pathways were enhanced in the more resistant Jurkat cells (Jurkat-AZD) and fell back after ATF4 knockdown (Fig. 5i), suggesting that ATF4 upregulated the intracellular quantities of amino acids under FGFR1 inhibitor treatment.

Targeting mTOR can overcome the resistance against FGFR1 inhibitors
To investigate the strategy of concomitant medications and overcome this resistance therapeutically, we performed drug screening (2059 approved drugs).After two rounds of screening, 30 drugs that had synergistic effects with AZD4547 were identified with a coefficient of drug interaction (CDI) < 1 [38] (Fig. 6a, Supplementary Table 2).The targets of these drugs were significantly enriched in angiogenesis and PI3K/Akt/mTOR signaling (Fig. 6b).Since the PI3K/ Akt/mTOR pathway is relevant to amino acid metabolism [50], we focused on this pathway and found that all of these three drugs (temsirolimus, rapamycin, and zotarolimus) were the inhibitors of mTOR (Fig. 6c).
Rapamycin is a typical mTOR inhibitor and anticancer drug.Here, we selected rapamycin for further verification.The combination of rapamycin and AZD4547 could significantly inhibited the viability of T-ALL cells, including Jurkat (CDI minimum = 0.12), MOLT-4 (CDI minimum = 0.35) and MOLT-16 (CDI minimum = 0.45) cells (Fig. 6d and Supplementary Fig. 6a, b).To further confirm the combination effect of targeting FGFR1 and mTOR, we used FGFR1 knockdown to replace its inhibitor, and found that synergy with the mTOR inhibitor also existed (Fig. S6c), which complemented their synergistic effect.However, PKI-587, a PI3K inhibitor, did not exhibit obvious synergy with AZD4547 (Supplementary Fig. 6d), indicating that the mTOR pathway, not the PI3K/AKT pathway, contributed to the FGFR1 drug resistance.We also found that the Jurkat-AZD cells were more sensitive to rapamycin (Fig. 6e).These results indicated that mTOR was an indispensable effector for drug resistance, and the resistance to FGFR1 inhibitors could enhance the sensitivity of cells to rapamycin.
We further found that ATF4 was induced by FGFR1 inhibitors in other FGFR1-upregulated malignancies, such as NCI-H1299 (NSCLC), OVCAR-8 (ovarian cancer), and SW620 (colorectal cancer) cells (Supplementary Fig. 6e, f), and the strategy of combining of these inhibitors was also applicable to these cell lines (Supplementary Fig. 6g-i).These results suggested that the resistance to f Venn diagram depicting the number of common genes in different groups.the "Motifs" was the group of motifs enriched in promotor, the "RNA-seq" was the group of 570 common DEGs shared in both AZD4547 and PD-166866 groups based on RNA-seq data, the "TFs activity (Top 20)" was the group of top 20 differentially activated TFs.g Signal of Chip-seq and ATAC-seq around the chromosomal region of ATF4.h Gene Set Enrichment Analysis (GSEA) of 570 common DEGs shared in both groups of AZD4547 and PD-166866.i, k The protein levels of GCN2, p-eIF2α, eIF2α, ATF4 in Jurkat cells with (i) GCN2 knockdown and (k) PERK knockdown.Cells were harvested after 72 h with GCN2-sh1 or PERK-sh1 or non-target lentivirus vector transfection (shC) and 6 h treatment with AZD4547 (2 μM) or DMSO.HSP90 was used as the internal control.Relative cell growth of GCN2 knockdown (j), or PERK knockdown (l) Jurkat cells compared with control Jurkat cells (shC), cell counting was performed at 72 h after AZD4547 treatment.m Relative cell growth of Jurkat cells treated with AZD4547 combined SP600125.Cell counting was performed at 72 h after treatment.Data are mean ± SD (Two-tailed unpaired Student's t test, * P < 0.05, ** P < 0.01, ns = no significance).Fig. 5 ATF4 is a crucial initiator to upregulate the metabolic genes and remodel the amino acid metabolism.a Heatmap depicting the differentially expressed genes (DEGs) about metabolism based on RNA-seq data.The color indicates the Z-score of different genes expression.Relative mRNA levels of DEGs in the group of (b) typical kinases, (c) transporters, and (d) aminoacyl-tRNA biosynthesis after ATF4 knockdown and AZD4547 treatment.Jurkat cells were harvested for qPCR analysis at 48 h after AZD4547 (2 μM) treatment.e Time series analysis of protein levels of ATF4, ASNS, ASS1, PHGDH and SLC1A5 in Jurkat cells exposed to AZD4547 (2 μM).The β-Tubulin (β-Tub) was used as the internal control.f Protein levels of ATF4, ASNS, ASS1, PHGDH, and SLC1A5 in Jurkat and the more resistant Jurkat cells (Jurkat-AZD) with or without AZD4547 (2 μM) treatment for 48 h.g The protein levels of ATF4, ASNS, ASS1, PHGDH and SLC1A5 in ATF4 knockdown Jurkat cells (ATF4-sh1, ATF4-sh2), non-target lentivirus vector (shC) was used as the control, the cells were harvested at 48 h after AZD4547 (2 μM) treatment.h Differential amino acid metabolites of different groups through targeted metabolomics analysis.The Jurkat cells were harvested after 48 h of transfection.The color indicates the Z-score of the quantity of metabolites.i Pathway analysis of the differential amino acid metabolites.Differential abundance score depicting the average, gross transformations of all metabolites in each pathway.the score of 1 indicates all measured metabolites increase in the pathway, and -1 indicates all measured metabolites decrease in the pathway.Data are mean ± SD.FGFR1 inhibitors mediated by ATF4 was universal in many malignancies, and that targeting mTOR could synergistically overcome this resistance.
To further confirm the synergistic effect in vivo, we established the Jurkat-luci cell-derived xenograft mouse models (CDX).AZD4547 (30 mg/kg every 2 days), rapamycin (3 mg/kg every 2 days) or the combination of these two drugs was administered intraperitoneally from Day 17 to Day 29 (Fig. 6f).The T-ALL progression in the combination group was significantly inhibited compared with that in the other groups (Fig. 6g, h).The combination of AZD4547 and rapamycin prolonged the survival time of the CDX (Fig. 6i), and significantly reduced the bone marrow invasion of T-ALL cells (Fig. 6j, k).Together, our data indicated that synergistically targeting FGFR1 and mTOR could inhibit the progression of T-ALL cells in vitro and in vivo.

Reprogramming of amino acid metabolism induces the activation of mTORC1
The synergistic effect of AZD4547 and rapamycin suggested that compensatory activation of mTOR might exist.We further monitored the phosphorylation ribosome protein S6 (p-S6), a reliable marker of mTORC1 activation, for a long time with AZD4547 treatment.The phosphorylation of S6 was decreased in the first few days, but restored after a longer period of AZD4547 treatment (Fig. 7a).We also found that the phosphorylation of Akt was rescued (Supplementary Fig. 7a), but jointly inhibiting PI3K (a directed upstream of Akt) and FGFR1 did not exhibit synergistically antileukemic efficiency (Supplementary Fig. 6d), which suggested that the rescue in upstream of the PI3K/Akt/mTOR pathway seems not indispensable in the FGFR1 inhibitor resistance.We further used the more resistant Jurkat cells (Jurkat-AZD) to amplify the feature of drug resistance and identified the crucial factors and found that the phosphorylation of S6 of the Jurkat-AZD cells was resistant to the inhibitory effect of AZD4547 (Fig. 7b).Since the ribosome protein S6 is involved in translation, we also found that the translational efficiency of total proteins was impaired in Jurkat cells, but did not decrease in the more resistant Jurkat cells (Jurkat-AZD) (Supplementary Fig. 7b, c).These results highlighted that the compensatory activation of mTORC1 was essential for the resistance to FGFR1 inhibitors.Fig. 7 The enhanced amino acid metabolism induced the activation of mTORC1.a Time series analysis of protein levels of ATF4, the phosphorylation levels of S6 and ribosomal protein S6 (S6) in Jurkat cells after AZD4547 treatment (right).DMSO was used as control (left).b Protein levels of ATF4, the phosphorylation levels of S6 and S6 in Jurkat or the more resistant Jurkat (Jurkat-AZD) cells with or without AZD4547 (2 μM), the β-Tubulin (β-Tub) was used as the internal control.c Protein levels of ATF4, ASS1, ASNS, PHGDH, SLC1A5 and phosphorylation level of S6 in ATF4 knockdown Jurkat cells.The cells were harvested at 72 h after ATF4 knockdown lentivirus vectors transfection, the β-Tubulin (β-Tub) was used as the internal control.Protein levels of the S6 and phosphorylation level of S6 (left) and relative cell growth (right) in the more resistant Jurkat (Jurkat-AZD) cells after (d) ASNS knockdown, (e) ASS1 knockdown, (f) SLC1A5 knockdown, or (g) PHGDH knockdown.Total proteins were harvested at 72 h after lentivirus vectors transfection, and the cell counting was performed at 72 h after different concentrations of AZD4547.Data are mean ± SD (Two-tailed unpaired Student's t test, *P < 0.05, **P < 0.01, ***P < 0.001).
To explore the connection between amino acid metabolism and the activation of mTORC1, we knocked down the ATF4 in the more resistant Jurkat cells (Jurkat-AZD), which had high ATF4 expression, and found that the phosphorylation of S6 and typical proteins (ASNS, ASS1, SLC1A5, and PHGDH) was significantly reduced (Fig. 7c).The translational efficiency was dramatically reduced after ATF4 knockdown (Supplementary Fig. 7d).Importantly, we found that the phosphorylation of S6 was significantly decreased after knockdown of ASNS, ASS1, SLC1A5, and PHGDH respectively and the Jurkat-AZD cells became more sensitive to AZD4547 and PD-166866 (Fig. 7d-g and Supplementary Fig. 7e-h).In summary, the reprogramming of amino acid metabolism induced the activation of mTORC1 and further contributed to resistance against FGFR1 inhibitors.

DISCUSSION
Targeting FGFR signaling by tyrosine kinase inhibitors (TKIs) is a successful therapeutic strategy in numerous tumors with FGFR abnormalities [51].Three FGFR inhibitors that are applied for mutations or translocations of FGFR2 or FGFR3 have been approved for urothelial cancer or cholangiocarcinoma [52][53][54].However, no inhibitor applied to any type of aberrations of FGFR1 has been approved, although the FGFR1 aberrations make up 49% of all abnormalities of the FGFR family [41].Due to drug resistance, only approximately 11% of FGFR1-amplified tumors respond to FGFR1 inhibitors, and several malignancies exhibit insufficient tumor shrinkage [55].In our manuscript, we identified that the FGFR1 was significantly upregulated in T-ALL and was negatively related to the prognosis of the patients.We further confirmed that FGFR1 was essential for T-ALL progression, and was a promising target for T-ALL therapy.However, neither the FGFR1 inhibitors AZD4547 nor PD-166866 exhibited antileukemic efficacy in T-ALL treatment, although the FGFR1 signaling was specifically inhibited by its inhibitors, and the AZD4547 had reached phase II clinical investigations, suggesting that T-ALL malignancies carrying FGFR1 overexpression have intrinsic resistance to FGFR1 inhibitors.Furthermore, our results addressed the detailed mechanism of the drug resistance to FGFR1 inhibitors and provide a potential avenue to overcome the obstacle in malignancies with FGFR1 overexpression.
ATF4 is a core stress-induced transcription factor of the integrated stress response (ISR), which is important for metabolic, and redox processes and can adapt cells to adverse stresses, including proteostasis defects, nutrient deprivation, viral infection and redox imbalances [44,56].Previous studies showed that ATF4 could drive resistance to sorafenib by preventing ferroptosis and resistance to gemcitabine via the TGF-beta1/SMAD2/3 pathway [57,58].Here, we identified that ATF4 was significantly increased during FGFR1 inhibitors treatment and was essential for resistance to FGFR1 inhibitors.Under a variety of stress conditions, the translation of ATF4 is activated by p-eIF2α mainly through four kinases, including GCN2, PERK, PKR, and HRI.Especially, GCN2 and PERK are closely related to amino acid or protein metabolism [59,60].Here, we found that the translational upregulation of ATF4 was mainly dependent on the GCN2-eIF2α pathway under FGFR1 inhibitors treatment.In addition, ATF4 mRNA was dramatically increased after treatment with FGFR1 inhibitors.Since the mRNA transcription is closely related to transposase accessibility [61], we performed ATAC-seq to describe the transposase accessibility in whole chromatin.The average transposase accessibility was increased in the regions of total promotors especially in the promotor of ATF4 after exposure to FGFR1 inhibitors, which appropriately contributed to the transcriptional upregulation of ATF4.These results elaborated the mechanism underlying the upregulation of ATF4 at both the transcriptional and translational levels.
Amino acid metabolism is closely related to the survival and progression of malignancies [62,63].ASNS mediated the resistance of leukemia to L-asparaginase by regulating asparagine synthesis [64].SLC1A5 contributed to the resistance of apatinib in non-small cell lung cancer through the reprogramming of glutamine metabolism [65].Here, we found that many survivalpromoting metabolisms, especially the amino acid metabolism and its metabolites, were upregulated after FGFR1 inhibitors treatment.More importantly, we identified that ATF4 was the sponsor of these transformations.These upregulations of survivalpromoting metabolism were interrupted after ATF4 knockdown, which revealed the initiator role of ATF4 in these metabolic processes and also provided a promising strategy for targeted therapy.Although blocking TFs remains a challenge in terms of treatment, the success of DB1976, a PU.1 inhibitor that could competitively blocks the motif of PU.1, suggests that a fresh idea for ATF4-targeted therapy to overcome the FGFR1 inhibitor resistance [66].
Drug screening is the process of identifying and optimizing prospective medications in high-throughput assays, which are also used to screen a specific biological functions [67].To identify the main effectors that mediate T-ALL cells resistance to FGFR1 inhibitors and determine a synergistic strategy on the treatment side, we screened the approved drug library, which comprises 2059 approved drugs for clinical treatment.We further identified that combined targeting mTORC1 and FGFR1 significantly inhibited the survival and proliferation of T-ALL cells, and the compensatory activation of mTORC1 contributed to the drug resistance.Interestingly, our previous study revealed that the T-ALL cells were resistant to the single rapamycin treatment, especially in the T-ALL leukemia stem cells [25], and the essential role of FGFR1 in stem cell leukemia [68].It is worth trying FGFR1 inhibitors combined with rapamycin for leukemia stem cell debulking.
mTORC1 is a master regulator that couples amino acid availability to cell survival, growth and drug resistance.Multiple factors modulate mTORC1 activity, such as growth factors, stress, energy status, and amino acids [50,69].Here, we revealed that inhibiting pivotal metabolic processes through knockdown of the key kinases (ASNS, ASS1, SLC1A5, and PHGDH) could directly inhibit mTORC1 activity and enhance susceptibility to FGFR1 Fig. 8 Schematic diagram of the resistance against FGFR1 inhibitors.The FGFR1 inhibitors specifically block FGFR1 signaling, but also induced the expression of ATF4 through enhancing chromatin accessibility transcriptionally combined with activating translation via the GCN2-eIF2α pathway.Then, ATF4 remodeled the amino acid metabolism by stimulating the expression of multiple metabolic genes, further maintained the activation of mTORC1, which contributed to the resistance to FGFR1 inhibitor in T-ALL.
inhibitors.As previously stated, the increase in Arg, Asn, and other EAAs could boost the mTORC1 activation [70][71][72][73].These findings implied that the upregulated amino acid biosynthesis and EAA uptake mediated by ATF4 were essential for mTORC1 activation and resistance to FGFR1 inhibitors.Moreover, these results also suggested that blocking ASNS, ASS1, SLC1A5 or PHGDH was a promising solution for drug resistance, which needs further confirmation.
Besides, we had successfully cases that targeted CD99 or CD30 to eliminate T-ALL or Hodgkin lymphoma respectively [23,74].Since FGFR1 is a membrane protein that is upregulated in T-ALL compared with its very low expression levels in normal T cells, it is possible to try chimeric antigen receptor (CAR) based immunotherapy.

CONCLUSION
In summary, our study identified that FGFR1 was a potential therapeutic target in T-ALL and elucidated that ATF4-induced amino acid metabolic reprogramming mediated the FGFR1targeted resistance (Fig. 8).Synergistically inhibition of FGFR1 and mTOR could overcome this resistance in T-ALL therapy.

Fig. 1
Fig. 1 FGFR1 is upregulated in T-ALL and negatively correlated with the prognosis of patients.a Heatmap depicting 226 upregulated expressed genes in a T-ALL cohort (CNCB, HRA000122).Upregulated genes were selected with P < 0.05, log2 FoldChange > 50.b Fragments Per Kilobase Million (FPKM) of FGFR1 in a T-ALL cohort with 12 normal T cell samples (CNCB, HRA000122).c mRNA expression levels of FGFR1 from an RNA-seq data about T-ALL cohort with 7 normal bone marrow samples (GEO, GSE26713).d Relative mRNA expression of FGFR1 in T-ALL cell lines and normal T cell samples through qPCR.e Protein levels of FGFR1 in T-ALL cell lines (left) and primary T-ALL blasts (right) using Western blot, the β-Tubulin (β-Tub) was used as an internal control.f mRNA expression levels of FGFR1 in different hematopoietic malignancies, data from the Cancer Cell Line Encyclopedia (CCLE).g Event-free survival of T-ALL patients with higher or lower expression of FGFR1, data from TARGET, phs000464.h Overall survival of T-ALL patients with higher or lower expression of FGFR1, data from TARGET, phs000464.Data are mean ± SD (Two-tailed unpaired Student's t test, *** P < 0.001).

Fig. 2
Fig. 2 FGFR1 is a potential therapeutic target in T-ALL but leukemia cells are resistant to FGFR1 inhibitors.a, c Knockdown efficiency of FGFR1 in (a) Jurkat or (c) MOLT-4 cells.Cells were harvested at 72 h after transfection using lentivirus vectors, non-target lentivirus vector (shC) was used as the negative control.b, d Relative cell growth of (b) Jurkat or (d) MOLT-4 cells after FGFR1 knockdown, cell counting was performed on day 1, 3, and 5 respectively.e, f The relative cell growth of T-ALL cell lines and normal T cell, cell counting was performed at 72 h after treatment with different concentration gradients of FGFR1 inhibitors (e) AZD4547, and (f) PD-166866.g Cell growth of Jurkat cells after FGFR1 knockdown or AZD4547 (3 μM) treatment.h Phosphorylation of Akt and S6 in Jurkat cells after FGFR1 knockdown.Cells were harvested at 72 h after transfection using lentivirus vectors.i Phosphorylation of Akt and S6 in Jurkat cells.Cells were harvested at 72 h after AZD4547 (3 μM) treatment.j-n Jurkat cell-derived xenograft (CDX) experiment in NCG mice, 5 × 10 6 luciferase labeled FGFR1 knockdown Jurkat cells (Jurkat-shFGFR1) or luciferase labeled control Jurkat cells (Jurkat-shC) that generated with non-specific shRNA were injected through the tail vein.PBS, AZD4547 (30 mg/kg every 2 days), or PD-166866 (30 mg/kg every 2 days) was intraperitoneally administered from day 14 to day 28 in Jurkat-shC mice models.Mice in group of FGFR1-sh1 were generated with Jurkat-FGFR1-sh1 and treated with PBS accordingly.j Schematic outline of cell-derived xenograft (CDX).k Bioluminescent imaging of cell-derived xenograft (CDX).l Kaplan-Meier survival curves of Jurkat cellderived xenograft mice (CDX).The endpoint was that weight loss exceeded 20% of the body weight of a similar normal animal according to guidelines of the Canadian Council on Animal Care (CCAC).m Analysis of bone marrow invasion by using anti-human CD7 antibody through flow cytometry.n Quantification of bone marrow invasion.Data are mean ± SD (Two-tailed unpaired Student's t test, *** P < 0.001, ns = no significance).

Fig. 3
Fig. 3 ATF4 is essential for the resistance of T-ALL against FGFR1 inhibitors.a Heatmap of the differentially expressed genes (DEGs).Jurkat-shC or Jurkat-shFGFR1 cells were treated with AZD4547 (2 μM), PD-166866 (4 μM), or DMSO for 36 h.DEGs were filtrated with P < 0.05 and |log 2 FoldChange | > 0.5.The color indicates the Z-score of different genes expression.b Venn diagram depicting the number of DEGs shared in different groups.c KEGG pathway analysis of 570 common DEGs that were shared in both AZD4547 and PD-166866 treatment groups.d Transcription factors enrichment analysis of 570 common DEGs that shared in both AZD4547 and PD-166866 treatment groups.e Time series analysis of the mRNA levels of ATF4 in Jurkat exposed to AZD4547 (2 μM) through qPCR.f, g Time series analysis of the protein levels of ATF4 in Jurkat (up) or MOLT-4 (down) exposed to (f) AZD4547 (2 μM) and (g) PD-166866 (4 μM).The β-Tubulin (β-Tub) was used as the internal control.h The protein levels of ATF4 (left) and relative cell growth (right) of Jurkat after ATF4 knockdown.Cells were treated with different concentrations of AZD4547 and cell counting was performed at 72 h after treatment.i, j Jurkat cell-derived xenograft experiment in NCG mice, 5 × 10 6 luciferase labeled ATF4 knockdown Jurkat cells (Jurkat-shATF4-Luci) or luciferase labeled control Jurkat cells (Jurkat-shC-Luci) were injected through the tail vein.PBS or AZD4547 (30 mg/kg every 2 days) was intraperitoneally administered from day 16 to day 26.i Schematic outline of cell-derived xenograft.j Bioluminescent imaging of Jurkat-derived xenograft.k The protein levels of ATF4 (left) and relative cell growth (right) of more resistant Jurkat cells (Jurkat-AZD) after ATF4 knockdown.Cells were treated with AZD4547 and cell counting was performed at 72 h after treatment.Data are mean ± SD (Two-tailed unpaired Student's t test, * P < 0.05, ** P < 0.01, *** P < 0.001).

Fig. 4
Fig. 4 Expression of ATF4 is stimulated by increased chromatin accessibility combined with GCN2-mediated translational activation.a Average ATAC-seq signal and b quantification of average ATAC-seq signal around transcription start site (TSS) in Jurkat cells after 96 h with AZD4547 (2 μM) treatment.DMSO was used as the control.n = 2. c Distribution of ATAC-seq signal in promotor, exon, intron, 3'UTR, and distal intergenic.d Top 20 significantly activated transcription factors (TFs) in each group based on RNA-seq data, the color indicates Z-scaled average TFs activity score.e Rank of motifs enriched in the region of promotor based on ATAC-seq.fVenn diagram depicting the number of common genes in different groups.the "Motifs" was the group of motifs enriched in promotor, the "RNA-seq" was the group of 570 common DEGs shared in both AZD4547 and PD-166866 groups based on RNA-seq data, the "TFs activity (Top 20)" was the group of top 20 differentially activated TFs.g Signal of Chip-seq and ATAC-seq around the chromosomal region of ATF4.h Gene Set Enrichment Analysis (GSEA) of 570 common DEGs shared in both groups of AZD4547 and PD-166866.i, k The protein levels of GCN2, p-eIF2α, eIF2α, ATF4 in Jurkat cells with (i) GCN2 knockdown and (k) PERK knockdown.Cells were harvested after 72 h with GCN2-sh1 or PERK-sh1 or non-target lentivirus vector transfection (shC) and 6 h treatment with AZD4547 (2 μM) or DMSO.HSP90 was used as the internal control.Relative cell growth of GCN2 knockdown (j), or PERK knockdown (l) Jurkat cells compared with control Jurkat cells (shC), cell counting was performed at 72 h after AZD4547 treatment.m Relative cell growth of Jurkat cells treated with AZD4547 combined SP600125.Cell counting was performed at 72 h after treatment.Data are mean ± SD (Two-tailed unpaired Student's t test, * P < 0.05, ** P < 0.01, ns = no significance).

Fig. 6
Fig. 6 Targeting mTOR could overcome the resistance against FGFR1 inhibitors.a Schematic of drug screening that AZD4547 combined with 2059 approved drugs.The Jurkat cells were treated with AZD4547 (2 μM) or each drug or a combination of these two drugs for 72 h.Cell viability was detected using CCK-8 kit.The synergistic efficiencies were calculated using the coefficient of drug interaction (CDI), CDI < 1 indicated a synergistic effect.b Target pathways enrichment analysis of 30 synergistic drugs.c List of drugs enriched in PI3K/Akt/mTOR signaling.d Synergistic inhibition effect of AZD4547 and rapamycin in Jurkat cells.The cell counting was performed at 72 h after AZD4547 and Rapamycin treatment.e Relative cell growth of Jurkat and the more resistant Jurkat cells (Jurkat-AZD) exposed to Rapamycin for 72 h.f-k Jurkat cell-derived xenograft (CDX) experiment in NCG mice, 5 × 10 6 luciferase labeled Jurkat cells were injected through the tail vein, AZD4547 (30 mg/kg every 2 days) or PD-166866 (30 mg/kg every 2 days) or combination of these two inhibitors was intraperitoneally administered from day 17 to day 29, the control group was treated with PBS.f Schematic of cell-derived xenograft (CDX).g Bioluminescent imaging of Jurkat-derived xenograft mice.h Total photon flux of bioluminescent of Jurkat-derived xenograft mice at day 29.i Kaplan-Meier survival curves of Jurkat-derived xenograft mice in different groups.The endpoint was that weight loss exceeded 20% of the body weight of a similar normal animal.j Analysis of bone marrow invasion using anti-human CD7 antibody through flow cytometry.k Quantification of bone marrow invasion in different groups.Data are mean ± SD (Two-tailed unpaired Student's t test, *P < 0.05, **P < 0.01, *** P < 0.001).