Abstract

   The small G protein Arf1 has been identified as playing a selective
   role in supporting cancer stem cells (CSCs), making it an attractive
   target for cancer therapy. However, the current Arf1 inhibitors have
   limited translational potential due to their high toxicity and low
   specificity. In this study, two new potent small‐molecule inhibitors of
   Arf1, identified as DU101 and DU102, for cancer therapy are introduced.
   Preclinical tumor models demonstrate that these inhibitors triggered a
   cascade of aging in CSCs and enhance anti‐tumor immunity in mouse
   cancer and PDX models. Through single‐cell sequencing, the remodeling
   of the tumor immune microenvironment induced by these new Arf1
   inhibitors is analyzed and an increase in tumor‐associated CD8+ CD4+
   double‐positive T (DPT) cells is identified. These DPT cells exhibit
   superior features of active CD8 single‐positive T cells and a higher
   percentage of TCF1+PD‐1+, characteristic of stem‐like T cells. The
   frequency of tumor‐infiltrating stem‐like DPT cells correlates with
   better disease‐free survival (DFS) in cancer patients, indicating that
   these inhibitors may offer a novel cancer immunotherapy strategy by
   converting the cold tumor immune microenvironment into a hot one, thus
   expanding the potential for immunotherapy in cancer patients.

   Keywords: anti‐tumor immunity, Arf1 inhibitor, cancer stem cell,
   superior T cells, trans‐cellular signaling
     __________________________________________________________________

   Potent small‐molecule inhibitors of Arf1, designated as DU101 and
   DU102, have been discovered to initiate a series of aging processes in
   CSCs and systemically remodeled immune microenvironment including
   increasing the population of tumor‐associated CD8+ CD4+ double‐positive
   T (DPT) cells in preclinical mouse cancer and PDX models.

   graphic file with name ADVS-11-2404442-g006.jpg

1. Introduction

   The small G protein Arf1, a member of the ADP‐ribosylation factor (Arf)
   small G protein family, is localized at the interface of multiple
   organelles. Its role involves coordinating organelle communication,
   fatty acid metabolism, and maintaining mitochondrial homeostasis.^[
   [44]^1 ^] Research has revealed high expression of Arf1 in various
   types of human cancers, including but not limited to breast,
   hepatocellular, colon, and prostate cancer.^[ [45]^2 , [46]^3 ^] And,
   elevated Arf1 expression is inversely linked to the prognosis of
   individuals diagnosed with cancer.^[ [47]^2 , [48]^4 , [49]^5 ^]

   Our recent study has revealed that the Arf1‐regulated lipid metabolism
   plays a crucial role in sustaining the homeostasis of cancer stem cells
   and neurons selectively.^[ [50]^2 , [51]^6 , [52]^7 , [53]^8 , [54]^9 ,
   [55]^10 ^] In Drosophila, the knockout of Arf1 in differentiated cells
   of the intestine and Malpighian Tubules (kidney) yields no discernible
   phenotype. However, the knockout of Arf1 in stem cells triggers a
   cascade of cellular organelle aging (exhibiting hallmarks of aging),^[
   [56]^11 , [57]^12 , [58]^13 ^] including lipid droplet (LD)
   accumulation, mitochondrial damage, reactive oxygen species (ROS)
   production, endoplasmic reticulum (ER) stress, and lysosomal protein
   aggregation. Abnormal stem cells in Drosophila melanogaster activate
   neighboring enterocyte (EC) cells by transmitting danger signals,
   resulting in the eradication of damaged stem cells through a process of
   killing, engulfing, and removal.^[ [59]^6 , [60]^7 ^] In mice, knockout
   of Arf1 in hepatocytes with Albumin‐Cre (Alb‐Cre) or normal adult
   intestinal stem cells with Lgr5‐CreER does not have any phenotype,
   while knocking down Arf1 in cancer stem cells (CSCs) of the intestine
   with Lgr5‐CreER or of the liver with Axin2‐CreER elicits a similar
   organelle aging cascade.^[ [61]^2 ^] The abnormal CSCs then send danger
   signals to activate antigen‐presenting dendritic cells (DCs) in the
   tumor immune microenvironment.^[ [62]^2 , [63]^14 ^] In contrast to
   current therapies, Arf1‐targeted therapy offers a dual anticancer
   mechanism by promoting CSC aging and enhancing the anti‐tumor activity
   of immune cells. However, further investigation is required to
   understand the anti‐tumor immunity response regulated by Arf1‐targeted
   therapy.

   The high toxicity and low specificity of the existing Arf1 inhibitors,
   such as Golgicide A (GCA), Brefeldin A (BFA), and Exo2 have greatly
   limited their translational potential and related research.^[ [64]^15 ,
   [65]^16 , [66]^17 , [67]^18 ^] Genetic screens in model organisms have
   been extensively used to study gene's biological functions.^[ [68]^19
   ^] To date, the utilization of such screens for the identification of
   target genes associated with tumor immunity has been limited. Our prior
   investigations demonstrated the selective eradication of
   RasV12‐transformed stem cell tumors in Drosophila Malpighian tubules
   (MT, fly kidney) through necrosis upon the knockdown of the
   COPI/Arf1‐lipolysis pathway. This process preserved differentiated
   cells. Notably, administering existing Arf1 inhibitors to flies
   resulted in the targeted elimination of tumorigenic stem cells while
   leaving normal stem cells unharmed.^[ [69]^6 , [70]^7 ^]

   Here, in the Drosophila stem cell tumor system, a screening of over 100
   chemicals, modified based on Exo2, led to the discovery of two new
   potent small‐molecule inhibitors of Arf1 for cancer therapy, denoted as
   DU101 and DU102. Demonstrations indicated that these newly designed
   Arf1 inhibitors exhibit low toxicity and robust anti‐tumor activities
   in various mouse tumor models, including liver, colon, breast, and
   melanoma, as well as PDX tumor models. Treatment with the inhibitors
   notably augmented T cell tumor infiltration, augmented the activation
   of CD8+ T cells via ATP, and induced the formation of tumor‐associated
   PD‐1+ TCF1+ CD8+ CD4+ stem‐like DPT cells. Consequently, the new Arf1
   inhibitors present an auspicious strategy for cancer immunotherapy and
   are presently under evaluation for clinical trials in patients with
   advanced solid tumors.

2. Results

2.1. Discovery of the Novel Arf1 Inhibitors

   Using a Drosophila stem cell tumor system, we screened and discovered
   two new potent small‐molecule inhibitors of Arf1, DU101, and DU102
   (Figure [71] 1a–d; Figure [72]S1a–c, Supporting Information details are
   described in STAR methods), which specifically and effectively killed
   the cancer stem cells. We first validated the biochemical and
   biophysical effects of DU101 and DU102 on Arf1. It was reported that
   the Sec7 domain of Golgi‐associated Arf‐GEF (GBF1) interacts with
   GDP‐bound Arf1 to regulate GDP‐GTP exchange for Arf1 activation.^[
   [73]^20 ^] We performed high‐precision molecular docking analysis and
   found that both DU101 and DU102 bound to the catalytic pocket of the
   Arf1‐GDP‐Sec7 (PDB code: 1RE0) complex at 2.40 Å (Figure [74]1b,d). We
   then directly examined the binding efficiency of DU101 or DU102 to the
   Sec7 domain of GBF1 by isothermal titration calorimetry and found that
   both DU101 and DU102 effectively bound the purified Arf1‐GDP‐Sec7
   domain (Figure [75]S2a,b, Supporting Information). Drug affinity
   responsive target stability (DARTS) is applied to track and identify
   target proteins based on the stability changes when the target protein
   binds to small molecules.^[ [76]^21 ^] By DARTS, we showed that DU101
   and DU102 could significantly enhance the stability of Arf1 in a
   dosage‐dependent way. As shown in Figure [77]S2c,d (Supporting
   Information), only the protein level of Arf1 increased in cells treated
   with proteinase combined with DU101 or DU102 indicating a specific
   protecting effect of Arf1 stability of these inhibitors. While the
   other ARF isoforms we tested, including Arf3, Arf5, Arf6, and pan‐Ras,
   the protein levels showed no obvious difference compared control group
   with DU101 or DU102 treatment group in cells with proteinase treatment
   (Figure [78]S2c,d, Supporting Information). For the Arf1 activity
   assay, the GST‐GGA3‐PBD fusion protein selectively binds the activated
   form of GTP‐bound Arf1 and Arf1 activity in cells can be examined by
   pulling down GST‐hGGA3‐Arf1.^[ [79]^22 ^] We found that DU101 or DU102
   treatment significantly inhibited Arf1 activity (Figure [80]1e,f).
   These results implied that our novel Arf1 inhibitor could inhibit Arf1
   activity by binding to Arf1 and Arf‐GEF.

Figure 1.

   Figure 1
   [81]Open in a new tab

   DU101 and DU102 are new potent inhibitors of Arf1. Chemical structure
   of the compound DU101. Representation of DU101 docking onto a published
   Sec7 domain of the Arf1‐GEF cocrystal structure (PDB ID 1RE0, 2.40 Å).
   (c) Chemical structure of the compound DU102. (d) Representation of
   DU102 docking onto a published Sec7 domain of the Arf1‐GEF cocrystal
   structure. (e) DU101 and DU102 caused a decrease in GBF1‐dependent Arf1
   activation. Huh7 cells were treated with DMSO, BFA, DU101 or DU102.
   Arf1 activities were assessed by GGA3 pull‐down and western blotting
   assays. (f) Quantification of the data in (e). Data are shown as the
   mean ± SEM. Student's t‐test. ***p < 0.001; n.s., not significant. BFA,
   brefeldin A.

   To evaluate the cell toxicity of DU101 and DU102, we examined the cell
   viability of the normal stem cells in Drosophila, as well as CT26, 4T1,
   B16‐F10, and LLC cells after treatments with DMSO and DU101 or DU102,
   and found that DU101 and DU102 exhibited low cell toxicity
   (Figures [82]S1b and [83]S2e–h, Supporting Information). To evaluate
   the toxicity in vivo, we compared different organs from mice with or
   without DU102 treatment. By hematoxylin‐eosin (HE) staining, no
   significant histomorphology alterations were discovered
   (Figure [84]S2i, Supporting Information). Together, these results
   demonstrated that the two new Arf1 inhibitors, DU101 and DU102, have
   low toxicity, and are potentially effective Arf1 inhibitors.

2.2. The New Arf1 Inhibitors Induced Tumor Regression and Prolonged Lifespan
in Multiple Mouse and PDX Tumor Models

   To investigate the therapeutic effect of DU101 and DU102 on tumors, we
   generated allografts of mouse liver cancer Hepa1‐6 cells in C57 mice
   and then treated the mice with DMSO, DU101 or DU102, respectively. We
   observed dramatically reduced tumor size and volume (Figure [85] 2a–c)
   but no decrease in body weight (Figure [86]2d) in mice treated with
   DU101 and DU102 at a dose of 5 mg kg^−1 for 2 weeks by gavage in
   comparison with mice treated with DMSO.

Figure 2.

   Figure 2
   [87]Open in a new tab

   DU101 and DU102 induced tumor regression and prolonged lifespan in
   mouse and PDX liver tumor models. a) Images of tumor sizes from Hepa1‐6
   cells transplanted into C57BL/6J mice and treated with the indicated
   reagents. b–d) Bar graphs of tumor weights (b), curves of tumor volumes
   (c), and bar graphs of the body weight of mice described in (a). (n = 6
   mice in each group). e) Experimental setup for the MYC‐ON mice. f–i)
   Representative liver images of MYC‐OFF or MYC‐ON mice treated with
   DMSO, DU101, or DU102 (f), liver surface tumor counts (g), liver
   weights (h), and survival curves (i) of MYC‐ON mice with the indicated
   treatments. (n = 10 mice in each group). j) Experimental setup for
   generating humanized immune and liver tumor PDX mice. k–l)
   Representative tumor images (k) and tumor weights (l) of humanized
   immune and liver tumor PDX mice treated with the indicated reagents. (n
   = 5 mice in each group). m) Experimental setup for the second tumor
   challenge experiment. n‐o) Representative tumor growth curve (n) and
   tumor weights (o) of mice with the first CT26 tumor cell challenge
   treated with the indicated reagents. (n = 5 mice in each group). p)
   Tumor growth curve for each mouse with second CT26 tumor cell
   challenge. (n = 5 mice in each group). Data are shown as the mean ±
   SEM. Student's t‐test, D‐Kaplan‒Meier, and log‐rank test (i). *p <
   0.05, **p < 0.01, ****p < 0.0001; n.s., not significant.

   We further examined the anti‐tumor effect of DU101 and DU102 in our
   previously used liver CSC transgenic mouse model,^[ [88]^2 ^]
   Axin2‐CreER driven MYC‐ON liver tumor mouse model. MYC‐ON mice were
   switched to a normal diet to induce tumor formation at the age of 6
   weeks and then treated with DMSO, DU101, or DU102 at the dose of 5 mg
   kg^−1 by gavage for 6 weeks (Figure [89]2e). We found that DU101 and
   DU102 treatment significantly reduced liver tumors (Figure [90]2f–h)
   and extended the lifespan of the MYC‐ON mice (Figure [91]2i) in
   comparison with DMSO treatment.

   To exam the general application of these inhibitors, we also treated
   allografts of mouse colon cancer CT‐26 cells, mouse breast cancer 4T1
   cells, and mouse melanoma B16‐F10 cells with DMSO, DU101 or DU102,
   which contain 98.3%, 40.1% and 76.6% of the CD133^+CD44^hi cells
   (potential CSC markers) respectively.^[ [92]^2 ^] We found that DU101
   or DU102 treatment also dramatically regressed tumors but had no
   effects on mouse body weights in comparison with mice treated with DMSO
   in the three mouse allograft tumor models (Figure [93]S3a–i, Supporting
   Information).

   To investigate the anti‐tumor effect of the new Arf1 inhibitors on
   human tumors, we seeded mouse livers with fresh human liver tumor
   samples and generated liver cancer patient‐derived xenografts (PDXs) in
   humanized HSC‐NCG mice (Figure [94]2j), which retained the major
   characteristics of the original human tumors and tumor immune
   microenvironment. We then treated humanized mice with DMSO or DU102 at
   5 mg kg^−1 by oral gavage for one week. In comparison with DMSO
   treatment, we found that DU102 treatment significantly reduced the
   tumor sizes of liver PDX (Figure [95]2k–l; Figure [96]S4a, Supporting
   Information) but had no impact on the mouse body weights
   (Figure [97]S4b, Supporting Information) in the humanized HSC‐NCG mice.

   To test whether DU102 treatment generated immune memory, we performed a
   second tumor challenge in DU102‐treated mice as the scheme showed
   (Figure [98]2m). Our results showed that DU102 treatment could
   effectively inhibit tumor growth after the first tumor challenge
   (Figure [99]2n–o). After the excision of the tumor, a second tumor
   challenge was performed. Mice treated with DU102 have a low
   tumorigenesis rate (1 in 5 compared with 4 in 5 Figure [100]2p). Thus,
   the DU102 treatment supports immune memory.

   To demonstrate the specificity of DU101 and DU102, we treated BALB/c
   mice with allograft tumors from the Arf1‐deficient CT26 cells
   (Figure [101]S4c–e, Supporting Information), control CT26 cells
   (Figure [102]S4f,j, Supporting Information), Arf4‐deficient CT26 cells
   (Figure [103]S4g,h,l, Supporting Information), Arf5‐deficient CT26
   cells (Figure [104]S4g,i,m, Supporting Information) and Arf6‐deficient
   CT26 cells (Figure [105]S4g,j,n, Supporting Information). DU101 or
   DU102 treatment did not further affect tumor and body weights in the
   Arf1‐deficient CT26 cell tumor‐bearing mice, indicating that DU101 and
   DU102 were specific Arf1 inhibitors. While, in other groups, DU102
   treatment could still effectively inhibit tumor growth. Collectively,
   the above results demonstrated that the new Arf1‐specific inhibitors
   DU101 and DU102 could significantly regress tumors with minimal toxic
   effects in multiple mouse and liver PDX models. Thus, the inhibitors
   might have great translational potential.

2.3. The New Arf1 Inhibitors Induced Tumor Cell Aging Cascade

   We previously reported that genetic ablation of Arf1 in Drosophila stem
   cells and mouse CSCs resulted in cell aging cascade,^[ [106]^2 ,
   [107]^6 , [108]^7 , [109]^14 ^] including LD accumulation, ROS
   production, mitochondrial damage, ER stress, and lysosomal protein
   aggregation. We then compared the effect of Arf1 inhibition to genetic
   deletion and found that treatment of CT26 cells with DU101 and DU102
   dramatically reduced the mitochondrial membrane potential
   (Figure [110]S5a,b, Supporting Information), increased ROS production
   as assayed by elevated Mitosox level (Figure [111]S5c,d, Supporting
   Information), induced expression of the lysosome protein LAMP1
   (Figure [112]S5c,d, Supporting Information), and increased protein
   aggregation as illustrated by increased staining of Proteostat dye
   (Figure [113]S5e,f, Supporting Information), in comparison with those
   in CT26 cells treated with DMSO control. The cellular organelle aging
   cascade triggered by Arf1 knockout has some similarities with commonly
   known cellular senescence.^[ [114]^23 ^] Senescence‐associated
   β‐galactosidase (SA‐β‐gal) is a molecular marker of senescent cells.^[
   [115]^24 ^] We stained SA‐β‐gal and did not find a change in cells
   treated with the Arf1 inhibitors in comparison with cells treated with
   DMSO (Figure [116]S5g,h, Supporting Information). These data suggest
   that, consistent with previously reported effects of genetic ablation
   of Arf1, the new Arf1 inhibitors DU101 and DU102 also promote cell
   aging cascade, which might be different from the commonly known
   cellular senescence.

2.4. The New Arf1 Inhibitors Promoted T Cell Infiltration, Activation, and
Memory T Cell Population

   In the previous study, we found that knocking down the Arf1 pathway in
   mouse CSCs with Lgr5‐CreER and Axin2‐CreER activated a systemic
   anti‐tumor immune response to destroy tumors.^[ [117]^2 ^] To
   demonstrate that the antitumor activity of DU101 and DU102 specifically
   relies on immune response, we first treated nude mice with allograft
   tumors from CT26 cells and found that DU101 or DU102 treatment did not
   affect tumor weights (Figure [118]S6a–d, Supporting Information).
   Further, flow cytometric analysis revealed that CD3^+ T cells were
   significantly increased in liver tumors of MYC‐ON mice after DU101 or
   DU102 treatment (Figure [119]S6e,f, Supporting Information) as well as
   after Arf1 knockout (Figure [120]S6g, Supporting Information). CD3^+ T
   cells were also significantly elevated in Hepa1‐6 allografts with DU101
   or DU102 treatment (Figure [121]S6h, Supporting Information). These
   data demonstrated that DU101 and DU102 treatment promoted T‐cell
   infiltration into tumors.

   We also reported that Arf1 knockdown in tumor cells resulted in the
   release of DAMPs, including ATP^2 (Figure [122]S7a,b, Supporting
   Information). We found that the released ATP level was also
   significantly increased in tumor cells with DU101 or DU102 treatment or
   Arf1 knockdown, which further stimulated T cell activation through
   P2RX7 (Figure [123]S7a,b, Supporting Information). By detecting the
   phosphorylation level of well‐established TCR downstream kinases
   including ZAP70, LAT, and ERK, we found that T cells were activated
   after culturing with medium from tumor cells treated with either DU101
   or DU102 or with Arf1 knockdown (Figure [124]S7c,d, Supporting
   Information). Simply adding ATP into the culture medium could
   reappearance the phenotype of elevated phosphorylation level
   (Figure [125]S7e, Supporting Information). Furthermore, T cell
   activation was blocked after adding apyrase to the medium, indicating
   that the released ATP from the Arf1‐ablated tumor cells mediated T cell
   activation (Figure [126]S7f–j, Supporting Information). Treated T cells
   with P2RX7 inhibitor, OxATP, also reversed T cell activation phenotype
   (Figure [127]S7k, Supporting Information). The ratio of CD44^+
   CD62L^+ central memory (T[CM]) cells was higher in hepatocellular
   carcinoma (HCC) tumors from MYC‐ON mice as well as in colon tumors from
   CT‐26 allograft mice after DU101 or DU102 treatment in comparison with
   DMSO treatment (Figure [128]S7l–o, Supporting Information). Together,
   our data suggested that the treatment with the new Arf1 inhibitors did
   not directly kill tumor cells rather than enhanced anti‐tumor immunity
   through promoting T cell tumor infiltration, T cell activation through
   releasing ATP and T cell differentiation toward a durable memory T cell
   population, consistent with our previously reported effects of genetic
   ablation of Arf1.

2.5. The New Arf1 Inhibitors Inflame the Tumor Microenvironment

   We next profiled a systematic analysis of the effects of DU101 and
   DU102 treatment on tumor‐infiltrating immune cells as well as cell‐cell
   interaction using single‐cell transcriptional profiling. To this, we
   performed single‐cell RNA sequencing (scRNA‐seq) with sorted CD45^+
   immune cells and CD45^− non‐immune cells from MYC‐ON liver tumors from
   DU101‐treated, DU102‐treated, and DMSO‐treated mice. We mixed sorted
   CD45^+ immune cells and CD45^− non‐immune cells at the ratio of 10:1
   for scRNA‐seq in each group, and obtained 35,468 cells in total after
   quality control and removing doublets. To find out the main components
   of TME, we integrated samples from different treatment groups,
   performed unsupervised clustering, and obtained 24 clusters of cells
   (Figure [129] 3a). We then annotated these cells using classical and
   well‐known markers and defined them as 13 cell types (Figure [130]3b;
   Figure [131]S8a, Supporting Information). Comparing the cell proportion
   between different treatment groups, we found that the proportion of T
   cells significantly increased after DU101 or DU102 treatment in
   comparison with that of DMSO treatment, while the proportion of myeloid
   cells decreased (Figure [132]3c,d; Figure [133]S8b, Supporting
   Information). These results suggested that the Arf1 inhibitors boosted
   tumor‐killing T cells.

Figure 3.

   Figure 3
   [134]Open in a new tab

   The new Arf1 inhibitors inflame the tumor microenvironment. a) Uniform
   manifold approximation and projection (UMAP) of immune and non‐immune
   cells in TME. b) Annotation of immune and non‐immune cells in TME. c)
   Cell density projection of immune and non‐immune cells in each
   treatment group. d) Bar plots showing the proportion of different types
   of immune cells in each treatment group. e) Annotation of Myeloid cells
   in TME based on selected cell markers. f) Bar plots showing the
   proportion of different subtypes of Myeloid cells in each treatment
   group. g) Annotation of DCs in TME based on selected cell markers. h)
   GO analysis of the up‐regulated genes of cDC1s in DU101 and DU102
   treatment groups.

   We further examined myeloid cells, which accounted for the largest
   proportion of immune cells other than T cells. We obtained myeloid
   cells and performed unsupervised clustering and got 23 cell clusters
   (Figure [135]S8c, Supporting Information). We then annotated myeloid
   cells based on their top feature genes and got 10 subtypes
   (Figure [136]3e; Figure [137]S8d, Supporting Information). We observed
   an increase in the Pglyrp1 expressing TAMs and a decrease in the Spp1
   and Igf1 expressing TAMs after DU101 and DU102 treatment
   (Figure [138]3f). The increased Pglyrp1 expressing TAMs exhibited
   pro‐inflammatory features and showed great ability in T cell
   recruitment and activation, as they highly express T cell chemokines
   Cxcl9 and Cxcl10, and type I IFN pathway genes Isg15 and Ifit2. In
   contrast, the decreased population of TAM_Spp1 and TAM_Igf1 showed
   immunosuppressive features, as they highly expressed genes associated
   with angiogenesis and immune regulatory factors, such as Arg1 and Spp1.
   To further evaluate the functions of myeloid cells, we collected
   previously reported macrophage M1 and M2 polarization gene lists^[
   [139]^25 ^] and downloaded IFN production and response gene lists from
   the GSEA database. We found that the expression of M1 polarization
   genes, type‐I, and type‐II IFN production and response pathways were
   up‐regulated after DU101 and DU102 treatment, while M2 polarization
   genes were down‐regulated after DU101 and DU102 treatment
   (Figure [140]S8e,f, Supporting Information). At last, we performed GO
   analysis on DEGs of myeloid cells after DU101 and DU102 treatment and
   found that pathways related to defense response, cytokine signaling
   pathway, and IFNβ pathway were enriched (Figure [141]S8g, Supporting
   Information).

   DCs play a crucial part in antigen‐presenting and innate immune
   response. We performed unsupervised clustering and annotated the 19
   clusters of cells as cDC1, cDC2, cDC3, and pDC (Figure [142]3g;
   Figure [143]S8h, Supporting Information). Because cDC1s are the major
   type of cells that mainly present antigen to CD8 T cells, we performed
   GO analysis on the up‐regulated DEGs of cDC1s after DU101 and DU102
   treatment, and observed enrichment in antigen‐presenting pathway,
   suggesting an enhanced antigen presentation from cDC1s to CD8 T cells
   (Figure [144]3h). Furthermore, we selected several genes related to DC
   activation, antigen presentation, and IFN signaling pathway and
   observed their up‐regulation in different subtypes of DCs
   (Figure [145]S8i, Supporting Information). These results demonstrated
   that myeloid cells were polarized into M1 pro‐inflammatory types and
   DCs were activated with enhanced antigen‐presenting ability after DU101
   and DU102, the Arf1 inhibitors programmed myeloid subsets into a
   proinflammatory phenotype that is characteristic of an inflamed tumor
   microenvironment (TME).

   The above data together demonstrated that the Arf1 inhibitors induced
   infiltration of anti‐tumor M1 macrophages, activated dendritic cells
   (DC), and reduced infiltration of tumor‐promoting M2 macrophages. Thus,
   the Arf1 inhibitors promoted proinflammatory remodeling of the TME to
   support anti‐tumor immunity.

2.6. The New Arf1 Inhibitors Reprogrammed T Cells with Superior Anti‐Tumor
Activity

   To further investigate the changes in T cells after treatment of the
   new Arf1 inhibitors, we obtained the annotated T cells and projected
   them to a reference mouse T cell atlas using ProjecTILs (Figure [146]
   4a; Figure [147]S9a, Supporting Information). We found that the
   proportion of CD8 T cells with effector and memory‐like features
   (CD8_EffectorMemory) increased, while the CD8 T cells with
   exhaustion‐associated features (CD8_Tex) and Tregs decreased after
   DU101 or DU102 treatment (Figure [148]4b). There was also a distinct
   increase in the ratio of CD8_EffectorMemory T cells to CD8_Tex cells,
   the ratio of CD8_EffectorMemory T cells to CD8_NaiveLike cells, and the
   ratio of the sum of Th1 cells and Tfh cells to Tregs (Figure [149]4c),
   suggesting that the T cells were more activated and had better tumor
   killing ability in addition to increased overall infiltration. Next, we
   analyzed the differentially expressed genes in CD8 T cells and found
   that there was a significant increase in the cytotoxicity‐related genes
   such as Gzma, Gzmb, Gzmk, and Nkg7, the T cell activation‐related genes
   Stat1 and Ifng, and the T cell chemotaxis related gene Cxcr6
   (Figure [150]4d). We found that the cytotoxicity‐related genes and T
   cell activation‐related genes were up‐regulated most in CD8_EarlyActiv
   and CD8_EffectorMemory T cells (Figure [151]4e). After performing GSEA
   pathway enrichment analysis on the up‐regulated DEGs in DU101 and DU102
   treatment groups, we found a significant enrichment in T cell
   proliferation, leukocyte cytotoxicity, lymphocyte migration, response
   to IFNγ, JAK‐STAT pathway activation, and so on (Figure [152]4f,g).
   Thus, DU101 and DU102 treatments reprogrammed CD8^+ T cells into more
   activated T cells with elevated effector function, reduced exhaustion,
   increased proliferation, and stronger cytotoxic activity.

Figure 4.

   Figure 4
   [153]Open in a new tab

   Treatment with the new Arf1 inhibitors induced T cell activation and
   blocked T cell exhaustion. a) Projection of T cells into reference T
   cell atlas by ProjecTILs. From left to right: Standard annotation of T
   cell subtypes in reference T cell atlas, projection of T cells in DMSO
   treatment group, DU101 treatment group, and DU102 treatment group. The
   contour lines show the cell density of projected T cells. b) Bar plots
   showing the proportion of different subtypes of T cells in each
   treatment group. c) The ratio of the proportion between different
   subtypes of T cells in each treatment group. Left: ratio of
   CD8_T_EffectorMemory cells versus CD8_Tex cells. Middle: ratio of
   CD8_T_EffectorMemory cells versus CD8_NaiveLike cells. Right: ratio of
   Th1 and Tfh cells versus Treg cells. d) Volcano plot showing
   up‐regulated and down‐regulated genes of CD8 T cells in DU101 and DU102
   treatment groups. Threshold for Super Anno: |avg_log2FC| > 0.3 and
   ‐Log10. p value ≥ 5. Threshold for Normal DEGs: 0.2 ≤ |avg_log2FC| ≤0.3
   and p value < 0.05. e) Heatmap showing expression of selected genes in
   different T cell subtypes in each treatment group. f,g) GSEA analysis
   of up‐regulated genes of CD8 T cells in DU101 and DU102 treatment
   groups. f) Three most representative up‐regulated pathways of CD8 T
   cells in DU101 and DU102 treatment groups. g) Ridge plots showing other
   13 up‐regulated pathways of CD8 T cells in DU101 and DU102 treatment
   groups.

2.7. The New Arf1 Inhibitors Promote a Population of PD‐1^+ TCF1^+ CD8^+
CD4^+ Stem‐Like T Cells in Liver Tumors

   Interestingly, we noticed a subpopulation of T cells with special CD4
   and CD8 double positive also significantly increased in the DU101 or
   DU102 treated group in our scRNA‐seq analysis. First, we confirmed the
   significant increase in the special CD4 and CD8 double‐positive cells
   (DPT cells) in MYC‐ON liver tumors after DU101 and DU102 treatment or
   after Arf1 gene deletion but not in spleens by FACS analysis
   (Figure [154]S9b–e, Supporting Information). Immunofluorescence tissue
   staining of tumors from MYC‐ON mice also verified that the DPT cells
   were significantly increased in tumors from DU101 or DU102‐treated mice
   in comparison with those from vehicle‐treated mice (Figure [155] 5a).
   We then perform a systematic analysis of the CD8^+ CD4^+ T cell
   subpopulation. To get a pure cluster of CD8^+ CD4^+ T cells, we
   re‐clustered all T cells using a supervised pseudotime‐ordered analysis
   only using Cd4, Cd8a, and Cd8b1. T cells were clustered into 2 states
   of CD8 single positive T (SPT) cells, 1 state of CD4 SPT cells, 1 state
   of CD8^+ CD4^+ T cells, and several states of CD8 CD4 double negative T
   (DNT) cells (Figure [156]5b–c; Figure [157]S9f, Supporting
   Information). According to feature gene comparison, the DPT cells were
   more similar to CD8 SPT cells (Figure [158]5d; Figure [159]S9g,
   Supporting Information). Specifically, DPT cells highly expressed cell
   cytotoxic genes (Nkg7, Gzmk, Prf1, Eomes), T cell chemokine receptor
   (Cxcr6), and stem‐like T cell marker (Tcf7) (Figure [160]5d;
   Figure [161]S9h, Supporting Information). Worth noting, we found that
   the percentage of stem‐like T cells (PD‐1^+ TCF1^+) was robustly
   increased in DPT cells after DU101 or DU102 treatment
   (Figure [162]5e–g). Consistently, the percentage of PD‐1^+ TCF1^+
   stem‐like T cells was higher in DPT cells than in CD8 SPT cells in CT26
   allografts (Figure [163]5h–i). Additionally, we found that DPT cells
   exhibited a greater potential for central memory T cell differentiation
   in vivo than CD8 SPT cells (Figure [164]5j).

Figure 5.

   Figure 5
   [165]Open in a new tab

   The new Arf1 inhibitors promote a population of PD‐1+TCF1+CD8+ CD4+
   stem‐like T cells in liver tumors. Immunofluorescence staining (IF) of
   liver tumors from MYC‐ON mice after treatment with DMSO, DU101 or
   DU102. Red, CD4; Green, CD8; DAPI, Blue. White dotted circles outlined
   the CD8^+ CD4^+ T cells. Scale bar, 50 µm. (n = 3 mice). Data are shown
   as mean ± SEM. Student's t‐test. *p < 0.05, **p < 0.01, ***p < 0.001;
   n.s., not significant. a) Supervised clustering of T cells according to
   the expression of Cd4, Cd8a, and Cd8b1. b) Annotation of T cell states
   based on their expression of Cd4, Cd8a, and Cd8b1. c) Heatmap showing
   the top feature genes of CD8 SPT, CD4 SPT, DPT, and DNT cells. d)
   Supervised clustering of DPT cells according to the expression of Tcf7
   and Pdcd1. e)Annotation of DPT cell states based on their expression of
   Tcf7 and Pdcd1. f) The ratio of stem‐like DPT cells versus other DPT
   cells in each treatment group. g) FACS Analysis of PD‐1^+ TCF1^+
   stem‐like T cells in tumor infiltrated CD8^+ CD4^− T and CD8^+ CD4^+ T
   cells. h) Analysis of PD‐1^+ TCF1^+ stem‐like T cells in
   tumor‐infiltrated CD8^+ CD4^− T and CD8^+ CD4^+ T cells. i) Analysis of
   CD62L^+ CD44^+ central memory T cells in tumor‐infiltrated CD8^+ CD4^−
   T and CD8^+ CD4^+ T cells.

   In summary, we found that treatment of the new Arf1 inhibitors DU101
   and DU102 enriched a CD8CD4 DPT cell population. The DPT cells have
   higher positive percentage of the stem‐like T cell marker TCF1 and T
   cell exhaustion marker PD‐1 (PD‐1^+ TCF1^+ CD8^+ CD4^+) and also had
   synergetic expression of memory T cell markers (including Sell/CD62L,
   CCR7, CD27, CD28, and IL7R), as well as activated T cell markers
   (including CD44, CD69, EOMES, and GZMB). The PD‐1^+TCF1^+CD8^+ CD4^+ T
   cells, first reported here according to our knowledge, might comprise a
   novel class of super anti‐tumor T cells.

2.8. CD8^+CD4^+ T Cells Exhibited Higher Activity and Stronger Cytotoxic
Function Than CD8^+ CD4^− T Cells

   We further performed GSEA analysis on the feature genes of these four
   (CD8 SPT, CD4 SPT, DPT, and DNT) T cell types, and found that DPT cells
   had superior characteristics in response to tumor cells, MHC protein
   binding, αβ T cell activation, and T cell selection, suggesting that
   DPT cells were more activated and probably had stronger anti‐tumor
   ability (Figure [166] 6a; Figure [167]S10a, Supporting Information). To
   verify the activity of the DPT cells, we compared the activation marker
   expression level and cytokine expression levels of DPT cells with the
   CD8 SPT cells. The DPT cells displayed higher expression of CD69, GZMB,
   and IL‐2 than the SPT cells with or without anti‐TCR stimulation both
   in vitro (Figure [168]6b; Figure [169]S10b, Supporting Information) and
   in CT26 allografts (Figure [170]6c). Though the percentage of CD69
   showed no difference between DPT cells and CD8 SPT cells with anti‐TCR
   stimulation in vitro, the mean fluorescence is higher in DPT cells upon
   anti‐TCR activation (Figure [171]S10b, Supporting Information). These
   data together suggested that DPT cells are more active potent cytotoxic
   T cells than CD8 SPT cells.

Figure 6.

   Figure 6
   [172]Open in a new tab

   CD8^+CD4^+ T cells exhibited higher activity and stronger cytotoxic
   function than CD8^+ CD4^− T cells. a) GSEA analysis of the top feature
   genes of CD8 SPT, CD4 SPT, DPT, and DNT cells. b) Quantification of
   CD69, GZMB, IFNγ and IL‐2 in CD8^+ CD4^− or CD8^+ CD4^+ T cells after
   stimulation with anti‐CD3 and anti‐CD28 antibodies for 24 h. (n = 3).
   c) Quantification of the percentage of CD69^+ T cells, GZMB^+ T cells,
   and IL‐2^+ T cells in CD8^+ CD4^−, or CD8^+ CD4^+ T cells from CT26
   allograft tumors. (n = 10). d) The setup of adoptive T cell assay and
   anti‐CD4 treatment. e) Representative images of luciferase‐labeled
   tumors fluorescence in the two treatment groups of B16‐F10 allografts.
   (n = 5). f) Quantification of the total tumor fluorescence at Day 10.
   g) Representative tumor images of B16‐F10‐OVA‐Luciferase allografts
   after treatment with IgG or anti‐CD4. (n = 5). h) Tumor weights of
   B16‐F10‐OVA‐Luciferase allografts with IgG or anti‐CD4 treatment. (n =
   5). i) Quantification of the percentage of CD8^+ CD4^+ T cells in
   tumor. (j) Quantification of the percentage of CD8^+ CD4^+ T cells in
   the spleen. k) Quantification of the mean fluorescence of CD69^+ T
   cells in the tumor. Data are shown as mean ± SEM. Student's t‐test. *p
   < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001; n.s., not
   significant.

   To characterize the role of DPT cells in T cell‐mediated anti‐tumor
   immunity in vivo, we transplanted CD8^+ T cells purified from OT1 mice
   into immunodeficient mice with B16F10‐OVA tumors and then eliminated
   DPT cells by anti‐CD4 treatment (Figure [173]6d). We found that
   anti‐CD4 treatment significantly reduced the tumor‐killing function of
   the T cells, and tumor regression was slower in mice treated with
   anti‐CD4 than in mice treated with IgG (Figure [174]6e,f). The tumor
   size and weight were also larger in mice with anti‐CD4 treatment
   (Figure [175]6g–h). Accordingly, the DPT percentage was decreased both
   in the spleen and tumor in mice with anti‐CD4 treatment in comparison
   with IgG treatment (Figure [176]6i,j; Figure [177]S10c–d, Supporting
   Information). Furthermore, the activation level of tumor‐infiltrated T
   cells, evaluated by the CD69 expression level, was decreased in mice
   with anti‐CD4 treatment in comparison with IgG treatment
   (Figure [178]6k; Figure [179]S10e, Supporting Information).

2.9. The Stem‐Like DPT Cells Exist in Human Liver Tumors and Correlate with
Better Disease‐Free Survival (DFS)

   We next tried to reveal whether this kind of DTP cells also existed in
   liver tumor patient samples. We utilized the deeply integrated human
   single‐cell omics data (DISCO), which was a public dataset of human
   scRNA‐seq data. We downloaded the liver tissue scRNA‐seq data from
   DISCO, filtered low‐quality cells, removed doublets, and isolated the
   potential DPT cells based on the positive expression of CD3, CD4, and
   CD8. We then constructed the first human liver tissue DPT cell atlas
   with 4062 high‐quality DPT cells from 28 projects and 139 samples. The
   supervised clustering was performed based on the expression of CD4,
   CD8A, CD8B, TCF7, PDCD1, and divided total DPT cells into 32 clusters
   (Figure [180] 7a,b). The DPT cells in cluster 17 expressed both TCF7
   and PDCD1, which were identified as stem‐like DPT cells
   (Figure [181]7c,d). We clustered the samples and divided them into
   eight groups, and found that the stem‐like DPT cells had higher
   infiltration in tumor samples than normal tissues, which might be a
   result of tumor antigen stimulation (Figure [182]7e). We used the
   expression top ten feature genes of the stem‐like DPT cells together
   with CD3D, CD3E, CD3G, CD4, CD8A, and CD8B as gene signatures to
   analyze its association in human liver tumors with LIHC patients'
   prognosis. We find patients with higher expression of the stem‐like DPT
   gene signature tended to have better disease‐free survival (DFS)
   (Figure [183]7f), suggesting that the stem‐like DPT cells in human
   liver tumors could also mediate anti‐tumor immune response.

Figure 7.

   Figure 7
   [184]Open in a new tab

   Construction of human liver tissue DPT cell atlas. a) Supervised
   clustering of human liver DPT atlas based on the expression level of
   CD4, CD8A, CD8B, TCF7, and PDCD1. b) Projects involved in the human
   liver DPT atlas. c) Annotation of DPT subtypes based on the expression
   of TCF7 and PDCD1 showed in (d). d) Feature plots showing the gene
   expression of TCF7 and PDCD1 in human liver DPT atlas. e) Bar plots
   showing the proportion of stem‐like DPTs in each type of patient
   sample. f) Kaplan–Meier plots showing the association of the expression
   of feature genes of stem‐like DPT cells in LIHC patients with a
   prognosis of disease‐free survival (DFS). g) Model of targeting Arf1 in
   tumor cells to induce systemic antitumor immunity.

3. Discussion

3.1. A Potential Novel Dual Anticancer Therapy

   In this study, we conducted a robust genetic screen in a Drosophila
   stem cell tumor system, testing over 100 modified chemicals from Exo2.
   This effort led to the identification of DU101 and DU102 as two potent
   small‐molecule inhibitors of Arf1 for cancer therapy. In both mouse
   models of cancer and pre‐clinical PDX tumor models, treatment with
   DU102 resulted in significant anti‐tumor immunity. We demonstrated that
   DU102 induces aging in cancer stem cells and triggers the release of
   various factors that remodel the tumor immune microenvironment.
   Consequently, it leads to the infiltration of anti‐tumor M1
   macrophages, activated dendritic cells (DC) with enhanced
   antigen‐presenting ability, and CD8 effector T cells. Moreover, DU102
   reduces the infiltration of tumor‐promoting M2 macrophages and
   exhausted CD8 T cells. Additionally, DU102 enhances T cell cytotoxicity
   and promotes the transformation of T cells to CD8CD4 double‐positive T
   (DPT) cells (Figure [185]7g).

3.2. The Effect of DU101 and DU102 May Target Tumor Cells with Stem Cell
Features

   The incidence of CSCs is generally less than 1% in most tumors. The
   question at hand is how the new Arf1 inhibitors in a small number of
   CSC cells could potentially trigger a robust systemic anti‐tumor
   response. In a recent study, it was discovered that pyroptosis of fewer
   than 15% of tumor cells was adequate to eradicate the entire grafted
   4T1 mammary tumor.^[ [186]^26 ^] However, it's uncertain whether the
   new Arf1 inhibitors specifically target cancer stem cells (CSCs) to
   induce systemic anti‐tumor immunity due to the ongoing debate about CSC
   markers and their identity. The observed effects of DU101 and DU102 may
   have originated from CSC stress or cellular organelle aging crisis,
   which could have led to the release of danger signals or factors. These
   could have triggered the activation of cytotoxic CD8+ T cells,
   resulting in the elimination of CSCs and a shift from an
   immunosuppressive tumor microenvironment (TME) to a favorable
   immunogenic state. Additionally, the release of cancer‐associated
   antigens (CAAs) by the initial dying cells could have reactivated
   immune cells in the TME, initiating a new wave of immune responses and
   enhancing their magnitude with each subsequent round.^[ [187]^26 ^]

3.3. The Novel Stem‐Like Double‐Positive T Cells Displayed Robust Anti‐Tumor
Functionality

   Previously, the focus has shifted towards stem cell‐like T cells as a
   result of their notable self‐renewal capacity, the ability for
   multi‐differentiation, and enduring effector function in the realm of
   adoptive therapy to combat tumors.^[ [188]^27 ^] TCF1+ PD‐1+ CD8+
   stem‐like T cells have been identified as maintaining enhanced
   responses to immune checkpoint inhibitors (ICBs) and adoptive cell
   transfer (ACT) in therapeutic animal models and human subjects with
   tumors.^[ [189]^28 , [190]^29 , [191]^30 , [192]^31 , [193]^32 ,
   [194]^33 , [195]^34 ^] Based on compelling evidence, it is clear that
   both CD8+ and CD4+ T cells are essential components in the body's
   anti‐tumor immune response.^[ [196]^35 ^] The DPT cells are known to
   potentially have additive effects. Nevertheless, recent research has
   indicated the presence of these cells in various human tumors,
   particularly in the leading‐edge area, which is situated adjacent to
   the malignant‐benign boundary of human hepatocellular carcinoma
   (HCC).^[ [197]^36 , [198]^37 ^] Human hepatocellular carcinoma (HCC)
   DPT cells demonstrate a high expression of PD‐1 and exhibit an
   activated phenotype along with robust functionality. Survival analysis
   indicates that HCC patients with a higher number of DPT cells and PD‐1+
   DPT cells experience significantly improved overall survival and
   recurrence‐free survival rates,^[ [199]^37 ^] suggesting that the DPT
   cells are superior anti‐tumor T cells.

   Studies pertaining to the lineage of DPT cells have yielded
   inconclusive findings. In the context of tumor development, the
   introduction of CD8+ T cells into B16 mice revealed distinct
   differentiation and functionalities of the DPT cells, emphasizing the
   adaptive nature of lineage commitment mechanisms within the peripheral
   environment.^[ [200]^36 ^] The TCR analysis of tumor‐infiltrated T
   cells in HCC patient samples indicated that the DPT cells likely
   originated from the same lineage as CD8 SPT cells, suggesting a
   probable transformation of DPT cells from intra‐tumoral SPT cells.^[
   [201]^37 ^] In our study, we conducted a comparison of cell markers and
   differentially expressed genes. Our findings indicate that DPT cells
   demonstrate a gene expression profile more closely related to cells
   from CD8 SPT than CD4 SPT. Furthermore, our supervised pseudo‐time and
   cell trajectory analysis revealed a stronger resemblance in expression
   patterns between DPT cells and CCD8 SPT than with CCD4 SPT cells. This
   suggests a flexible differentiation between DPT cells and CD8 SPT and
   offers a potential strategy for enhancing DPT cells using small
   molecular chemicals. Given the low percentage of DPT cells, it is
   imperative to explore strategies for promoting DPT differentiation for
   potential clinical translation. Further analysis into the regulation of
   differentiation between DPT cells and CD8 SPT cells, along with careful
   consideration of their inclusion during the sequencing and analysis of
   TILs from patient samples, is warranted.

   In conclusion, the newly developed Arf1 inhibitors in this study show
   promise as a novel strategy for cancer immunotherapy. With their strong
   pharmacology and positive safety profile in preclinical toxicology
   studies, a new and improved derivative (AOB01) of DU101 and DU102 is
   currently undergoing evaluation in Investigational New Drug (IND)
   studies, both as a monotherapy and in combination with anti‐PD‐1 for
   solid tumors. As far as we are aware, this new Arf1 inhibitor is the
   pioneering drug that activates a dual anticancer therapy by promoting
   cancer stem cell aging and anti‐tumor immunity and may lay the
   groundwork for additional therapeutics targeting this dual process.

4. Experimental Section

Experimental Model and Subject Details—Mice

   For the study, MYC‐overexpressing liver cancer mice were generated by
   breeding B6.Cg‐Tg(Cebpb‐Tta)5Bjd/J (JAX:003563) and
   FVB/N‐Tg(tetO‐MYC)36aBop/J (JAX:019376) mice, both obtained from The
   Jackson Laboratory.
   NOD/ShiLtJGpt‐Prkdc^em26Cd52Il2rg^em26Cd22Hr^em1Cin8936/Gpt mice (Cat#
   T003257) were also purchased and CD34^+
   HSC‐NOD/ShiLtJGpt‐Prkdc^em26Cd52Il2rg^em26Cd22/Gpt mice (Cat# T037620)
   were humanized from GemPharmatech in Nanjing, China. Additionally, OT1
   mice were kindly provided by Prof. Yu Xiaofei of Fudan University.
   Age‐matched male and female mice were used for all experiments, with
   6‐week‐old mice being the starting point for all experiments.
   Furthermore, PDX transplantation was performed using 12‐week‐old
   humanized mice. All animals were housed in specific pathogen‐free
   facilities, and all procedures were conducted in compliance with the
   Animal Care and Use Committee (ACUC) at Fudan University.

Experimental Model and Subject Details—Cells

   Mouse colorectal carcinoma CT26 cells (Cat# CRL‐2638) derived from
   BALB/c mice, mouse melanoma B16‐F10 cells (Cat# CRL‐6475) derived from
   C57BL/6J mice, mouse stage IV breast cancer 4T1 cells (Cat# CRL‐2539)
   derived from BALB/c mice, human embryonic kidney HEK 293T/17 cells
   (Cat# CRL‐11268), mouse hepatoma Hepa1‐6 cells (Cat# CRL‐1830) derived
   from C57BL/6J mice, and mouse Lewis lung carcinoma LLC1 cells (Cat#
   CRL‐1642) derived from C57BL/6J mice were all obtained from the
   American Tissue Culture Collection (ATCC). B16‐F10 cells expressing OVA
   antigen were kindly provided by Dr. Gaofeng Fan of Shanghaitech
   University. CT26 cells and 4T1 cells were cultured in RPMI 1640 (SIGMA,
   Lot# RNBK0103) containing 10% fetal bovine serum (FBS) (HyClone, Cat#
   SH30396.03) and 100 units mL^−1 penicillin/streptomycin (Gibco, Cat#
   15140122). B16‐F10 cells, HEK 293T/17 cells, Hepa1‐6 cells, and LLC1
   cells were grown in DMEM (Sigma, Lot# RNBK3466) supplemented with 10%
   FBS and 100 units mL^−1 penicillin/streptomycin. All cells were
   cultivated in a humidified atmosphere containing 5% CO[2] at 37 °C.

   Primary cultures of mouse T cells from the spleen were performed as
   previously described.^[ [202]^27 ^] Male and female mice at 8–12 weeks
   old were euthanized by CO[2], and the spleen was collected and
   dissected by grinding. CD8^+ T cells were separated using the EasySepTM
   Mouse CD8 T Cell isolation Kit (Stemcell, cat. No. 19853). T cells were
   re‐suspended and activated in anti‐CD3 and anti‐CD28 mAbs‐coated plates
   (invivomab) for 48 h. Activated T cells were maintained with IL‐2
   (10 ng mL^−1) and 2‐mercaptoethanol and cultured with a fresh complete
   medium.

Genomic Mouse Models

   In the context of the liver tumor model in mice, as previously
   documented in the scientific literature under PubMed ID: 31924786,
   Cebpb‐tTA/tet‐o‐MYC/Axin2‐CreER/Arf1f/f mice was employed to
   specifically ablate the Arf1 gene in Axin2+ liver cancer stem cells
   using Axin2‐CreER in the MYC‐ON mice. The alimentation provided in both
   the mating cage and the wean cages consisted of a Grain‐Based Rodent
   Diet containing 200 mg kg^−1 doxycycline (Bio‐Serv, Cat# 14‐727‐450).
   Tumorigenesis was initiated in 6‐week‐old mice through a transition to
   a normal diet. In Cebpb‐Tta/tet‐o‐MYC/Axin2‐CreER/Arf1f/f mice, the
   excision of the Arf1 gene from the stem cells was achieved via
   intraperitoneal injection of 10 mg kg^−1 tamoxifen (Sigma–Aldrich, Cat#
   T5648) daily for a period of five consecutive days. The euthanasia of
   the mice was performed subsequent to the manifestation of tumors, at
   10–12 weeks.

Engrafted Tumor Models

   For grafts of mouse tumor cells in BALB/c or C57BL/6J background mice,
   littermate mice were separated into groups, and 5 × 10^5 CT26 cells, 5
   × 10^5 4T1 cells, 5 × 10^5 B16‐F10 cells or 5 × 10^5 Hepa1‐6 cells were
   injected subcutaneously (s.c.) into the mouse's left or right side.
   DU101 or DU102 (5 mg k^−1g or 1% DMSO) was given to mice by gavage
   daily. Tumor volume was measured every 2–3 days by a digital caliper
   (ULINE, Cat# H‐7352), and the tumor volume was calculated using the
   formula ½ × longitudinal diameter (length) × the greatest transverse
   diameter (width).^[ [203]^2 ^] Mice were euthanized when the tumor
   volume reached 2000 mm^3.

Patient‐Derived Xenograft (PDX) Implantation

   The liver tumor patient‐derived xenograft (PDX) experiments were
   conducted in strict accordance with ethical guidelines, adhering to the
   principles outlined in the Declaration of Helsinki. Informed consent
   was obtained from all patients or their legal guardians. The primary
   liver tumor specimens originated from patients with liver cancer who
   had undergone clinical surgery at Zhong Shan Hospital, Fudan University
   in Shanghai, China. Following surgical resection, the liver tumors were
   sectioned into 1 mm^3 fragments and immersed in sterile 1xPBS. These
   minute tumor fragments were subsequently subcutaneously transplanted
   into anesthetized 11‐week‐old humanized mice. The mice were housed in
   controlled, pathogen‐free environments with a 12‐h dark/light cycle,
   and daily monitoring of tumor growth was performed.

Phenotypic Screen for Small Molecules That Kill Stem Cell Tumors in
Drosophila

   As previously demonstrated that knockdowns of the COPI/Arf1‐lipolysis
   pathway selectively killed Ras^V12‐transformed stem cell tumors in
   Drosophila Malpighian tubules (MT, fly kidney) through necrosis but
   spared differentiated cells. Feeding flies with the existing Arf1
   inhibitors selectively killed tumorigenic stem cells but not normal
   stem cells.^[ [204]^3 ^] Using the Drosophila stem cell tumor system, a
   phenotypic assay was developed using fluorescent confocal microscopy
   and screened new compounds that killed Ras^V12‐induced stem cell
   tumors.

   The screening was conducted in 96‐well plates, modified from a
   published screening procedure successfully conducted in adult
   Drosophila.^[ [205]^28 ^] In brief, i) after clonal induction (ACI)
   using the positively marked mosaic lineage (PMML) labeling technique^[
   [206]^29 ^] in adult Drosophila, flies with Ras^V12‐transformed renal
   and nephric stem cells (RNSCs) were cultured at 29 °C for 4 days to
   allow for tumor growth; ii) fly food was boiled, cooled to 37 °C,
   aliquoted into plates at 100 µL well^−1; (iii) drugs were added at
   0.5 µL well^−1 and mixed by pipetting up and down five times, resulting
   in a final compound concentration of ≈50 µM; and iv) after food
   solidification, flies with Ras^V12‐transformed RNSCs were added to the
   wells (three females per well) and cultured for 3 days at 29 °C. The
   flies were dissected after incubation and viewed under a dissecting
   microscope, and a digital camera was used to capture images of tumors
   using white light and GFP fluorescence. The data were analyzed using
   MetaMorph (Molecular Devices) to manually set a pixel intensity
   threshold that included the GFP‐labeled tumors but not background
   fluorescence. The data for each compound were stored in a Microsoft
   Access database. Drugs that reduced GFP levels by 50% or greater were
   retested in three wells with nine flies. Drugs that reduced GFP levels
   by 50% or more in all three replicates relative to DMSO were scored as
   hits.

   The classic Arf1 inhibitor brefeldin A (BFA) and the small molecule
   4‐hydroxy‐3‐methoxybenzaldehyde
   (5,6,7,8‐tetrahydro[1]benzothieno‐[2,3‐d]pyrimidin‐4‐yl) hydrazone
   (Exo2) stimulate biological changes that are virtually
   indistinguishable. Exo2 is much more amenable to chemical
   modification.^[ [207]^30 ^] A range of existing and newly synthesized
   Exo2 analogs were screened and it was found that DU101 and DU102 were
   effective and had relatively low toxicity (Patent No.
   PCT/CN2021/110 373).

Generation of Cell Lines

   The detailed sequences of the shRNA oligonucleotides are listed in
   Table [208]S1. The double oligonucleotides were annealed at 95 °C for
   20 min and cloned into the pLKO.1‐Puro lentivirus vector. After
   sequencing, the pLKO.1 plasmids containing different shRNAs were
   obtained. Lentiviruses carrying the pLKO.1 plasmids were then produced
   by co‐transfecting HEK293T cells with two helper plasmids: pVSVG and
   pMD2.G. The lentiviral particles were harvested from the medium after
   48 h of co‐transfection. The cell medium was passed through a 0.22‐µm
   filter, and the lentivirus was collected into Eppendorf tubes and kept
   at −20 °C for short‐term storage or −80 °C for long‐term storage. The
   cells were infected and selected with 5 µg mL^−1 puromycin treatment
   (ThermoFisher, Cat# A1113803) for 2 days. Stable cell lines were
   cultured in 1.0 µg mL^−1 puromycin.

Immunohistochemistry (IHC) and Immunofluorescence (IF) Analyses

   Mouse tissue and tumors were dissected and fixed in 4% paraformaldehyde
   at 4 °C overnight after mice were euthanized by CO[2] for 20 min.
   Tumors were dehydrated by gradient alcohols or 30% sucrose and then
   embedded in paraffin wax or OCT (Tissue‐Tek; Sakura Finetek USA, Cat#
   4583). 5um paraffin sections were generated by a Leica RM2125
   Microtome. 10um frozen sections were generated by a Leica CM1950.
   Paraffin sections were dewaxed with xylene and hydrated with gradient
   alcohol. Antigens were retrieved by sodium citrate solution (Solarbio,
   Cat# C1032) in a 100 °C cooker for 25 min. The sections were washed in
   1xPBS three times for 5 min each and blocked in 3% FBS for 1 h at room
   temperature. Then, the sections were incubated with primary antibodies
   at 4 °C overnight. The following day, sections were washed in 1xPBS
   three times for 5 min each. For IHC, the paraffin sections were
   incubated with secondary antibodies for 1 h at room temperature and
   then stained with a DAB horseradish peroxidase color development kit
   (Beyotime, Cat# P0202). For IF staining, the frozen sections that had
   been incubated with primary antibody were then stained with fluorescent
   secondary antibody at room temperature and protected from light for 1
   h. The sections were examined with an Olympus inverted fluorescence
   microscope (NEW IX73) or a Zeiss (LSM710) confocal microscope.

T Cell Depletion

   Mice were treated with 200 µg per mouse of rat anti‐CD4 (clone GK1.5,
   BioXcell, Cat#BP0003‐1) or 200 µg per mouse of rat IgG (BioXcell,
   Cat#BP0090) isotype control antibodies. All antibodies were diluted in
   PBS and i.p. injected into mice for 3 consecutive days per week, for
   two consecutive weeks.

T Cell Isolation and Culture

   To isolate spleen cells, a mouse was euthanized by CO[2] for 20 min,
   and the spleen was harvested, minced into small pieces, and placed in
   RPMI 1640 medium. The minced spleen was dissociated by pipetting up and
   down with a 5‐mL syringe plunger. The cells were then passed through a
   40‐µm strainer (JET BIOFIL, Cat# CSS013040), followed by the lysis of
   red blood cells. The isolated cells were washed twice in cell staining
   buffer (0.5% BSA in 1 × PBS). CD8‐positive T cells were purified with
   the EasySep Mouse CD8+ T Cell Isolation Kit (Stemcell, Cat# 19853).
   After centrifugation, T cells were resuspended in RPMI 1640 medium, 10%
   FBS, 100 units mL^−1 penicillin/streptomycin, 4 mM L‐glutamine (Gibco,
   Cat#25030081), 1 mM sodium pyruvate (Gibco, Cat# 11360070), 50 µM
   2‐mercaptoethanol (Sigma–Aldrich, Cat#M3180), 25 mM HEPES (Gibco,
   Cat#15630130), and 1 × MEM nonessential amino acid solution (Gibco,
   Cat#11140050). T cell suspensions were placed in plates that had been
   incubated one day earlier with anti‐CD3 and anti‐CD28 antibodies at a
   final concentration of 3 µg mL^−1. Two days later, recombinant mouse
   IL‐2 (Genescript, Cat# Z02764) at a final concentration of 10 ng mL^−1
   was added to the medium. Mouse T cells were cultured in a humidified
   atmosphere containing 5% CO[2] at 37 °C.

T Cell Activation Assay

   Tumor cells were cultured with 10 µM DU101 or DU102 for 24 h and
   changed to normal medium for another 24 h. Then, the cell medium was
   collected and centrifuged for 3 min at 1,000 rpm and stored at 4 °C. To
   digest ATP, the cell medium was treated with 10 U mL^−1 Apyrase (NEB,
   Cat# M0398) at 30 °C for 30 min. T cells (6 × 10^6) were cultured with
   cell conditioned medium for 24 h. The T cells were then collected by
   centrifugation at 500 × g for 3 min, resuspended in 600 µL RPMI 1640
   medium without FBS, and cultured at 37 °C for 60 min. The T cells were
   divided into 3 portions and stimulated with anti‐CD3 antibody for 30 s
   at 37 °C. Then, the T cells were treated with a secondary antibody
   (Thermo Fisher, Cat# [209]A18891) containing sodium orthovanadate for
   the indicated time. The T cells were lysed with RIPA buffer and boiled
   at 95 °C for 20 min. The protein samples were centrifuged at 12 000 rpm
   for 10 min at 4 °C and subjected to western blotting.

T Cell Adoptive Experiment

   CD8^+ T cells isolated from OT1 mice were treated with anti‐CD3 and
   anti‐CD28 for 2 days. 1 × 10^6 B16‐F10‐OVA‐Luciferase cells were
   subcutaneously injected into NCG mice. After 4 days, 1 × 10^6 CD8^+T
   cells were adoptively transferred into tumor‐bearing mice via tail vein
   injection.

Tumor‐Infiltrated Immune Cell Isolation

   Mouse liver tumor‐infiltrated immune cells were isolated by mincing
   them into small pieces and dissociating them with a tumor dissociation
   kit (Miltenyi Biotec, Cat# 130‐096‐730). The cells were collected and
   passed through a 40‐µm strainer into a 50‐mL tube. The single‐cell
   suspension was centrifuged for 5 min at 2000 rpm at room temperature.
   The pellet was washed with RBC lysis buffer (Beyotime, Cat# C3702) for
   3 min, and the cells were spun for 5 min at 2000 rpm at 4 °C and
   resuspended in RPMI medium to prepare the cells for staining for flow
   cytometry.

Fluorescence‐Activated Cell Sorting (FACS)

   The cells were suspended in a cell staining buffer (4abio, Cat#
   FXP005). DAPI (Sigma‒Aldrich, Cat# D9542) or a Zombie NIR Fixable kit
   (Biolegend, Cat# 423105) was used to identify live or dead cells. Cell
   surface staining was performed by adding appropriately conjugated
   fluorescent primary antibodies to the cells and incubating them at room
   temperature for 1 h in the dark. For intracellular staining, the cells
   were first fixed in 4% paraformaldehyde and permeabilized with 0.1%
   Triton X‐100 (Beyotime, Cat# ST795), and then intracellular staining
   was performed according to the corresponding protocols. After staining,
   suspended cells were analyzed by flow cytometry. Flow cytometry was
   performed using an LSR Fortessa (BD). Data were analyzed and presented
   with FlowJo_V10 software. The antibodies used are listed in Table
   [210]S2 (Supporting Information).

Plasmids

   The pET21b pVSVG, pMD2.G, and pLKO.1‐puro were kindly provided by Dr.
   Gaofeng Fan of Shanghaitech University. The other plasmids were
   generated in‐house; the detailed sequences are shown in Table [211]S1
   (Supporting Information). Briefly, all of the shRNAs were synthesized
   by Sangon, and the pLKO.1‐Puro plasmids were double digested by the
   AgeI‐HF (New England Biolabs, Cat# R3552S) and EcoRI‐HF (New England
   Biolabs, Cat#R3101S) enzymes. The shRNAs and vector were ligated by T4
   DNA ligase (New England Biolabs, Cat# M0202S). Positive clones were
   sequenced by Sangon to verify that the ligation was correct.

Western Blotting

   Cell proteins were released by RIPA lysis buffer (0.1% SDS, 150 mM
   NaCl, 0.5% sodium deoxycholate, 1 mM EDTA, 1 mM EGTA, 1.0% Triton
   X‐100, and 50 mM Tris pH 8.0) containing 1 mM PMSF (Beyotime, Cat#
   ST505). The proteins were then centrifuged at 12 000 × g for 3 min at 4
   °C. The supernatant protein concentration in the lysates was determined
   by a Protein Assay kit (Bio‐Rad, Cat# 5000006). All samples were
   adjusted to equal concentrations, 5x SDS loading buffer was added, and
   the sample was boiled at 100 °C for 15 min. Equal volumes of protein
   samples were loaded onto protein gels and separated by electrophoresis
   with an 80 V PowerPac Basic Power Supply (Bio‐Rad) for 100 min, then
   transferred to a nitrocellulose membrane (GE Healthcare, Cat#
   10600014). The membrane was blocked with 5.0% skimmed milk powder for 1
   h and then incubated with the appropriate primary antibody at 4 °C
   overnight with shaking. The next day, the membrane was washed with 1 x
   PBST three times for 5 min and incubated with the appropriate secondary
   HRP antibody in 5% milk for 1 h. The membrane was then developed with
   an Omni‐ECLFemto Light Chemiluminescence Kit (Epizyme, Cat# SQ201 L)
   and imaged with an imaging system (Tanon 5200).

RNA Extraction and qRT‐PCR

   For quantitative reverse transcription polymerase chain reaction, total
   RNA was extracted from the cells with a kit (Beyotime, Cat# R0011), and
   cDNA was synthesized by the Hifair V One‐Step RT‐gDNA Digestion
   SuperMix for qPCR Kit (Yeason, Cat# 11142ES60). The primers used for
   qRT‐PCR are listed in Table [212]S1 (Supporting Information). The total
   cDNA was amplified by Hieff qPCR SYBR Green Master Mix (No Rox)
   (Yeason, Cat# 11201ES08) with a CFX96 Touch Real‐Time PCR Detection
   System (Bio‐Rad). The 18S gene was used as an internal control. The
   results were calculated as 2^−ΔCT.

Imaging of Grafted Tumors

   Mouse melanoma B16F10‐OVA cells were transfected with pmeLUC plasmids
   and selected with 200 µg mL^−1 G418 (Beyotime, Cat# ST082) for 20 days
   to generate a stable cell line. NCG mice were s.c. grafted with 1 ×
   10^6 B16F10‐OVA‐LUC cells. D‐luciferin (100 mg kg^−1) was injected into
   the mice i.p. 10 min before imaging. The mouse tumor sites were
   photographed with a Xenogen IVIS Spectrum at the animal facility of the
   Institute of Developmental Biology and Molecular Medicine of Fudan
   University (Shanghai, China). The in vivo bioluminescence of the tumor
   region was quantified using Living ImageH software.

Extracellular ATP Test

   Cells were cultured in RPMI164 medium overnight, and the cells were
   treated with 10 µM DU102 or DU101 the next day. After 48 h of
   treatment, the cell medium was collected and centrifuged at 4 °C at
   1000 rpm for 10 min to clear dead cells. The supernatant was collected
   for the ATP test according to the manufacturer's instructions. Briefly,
   10 µL of CCK‐8 solution was added to 100 µL of the sample. Then, the
   samples were incubated for 1 h at 37 °C. The absorbance was measured at
   450 nm with a microplate reader.

Protein Purification

   To prepare Sec7 and Arf1 proteins, the gene fragment of Sec7 or Arf1
   was ligated into vector pET‐21b (+). After validating the correct
   sequence, the recombinant plasmids were transformed into BL21(DE3)
   competent cells (Transgen, Cat# CD601). Isopropyl‐β‐D‐thiogalactoside
   was added to the bacteria at a final concentration of 1 mM when the
   OD600 reached 0.6 to induce protein expression. After incubating at
   16 °C for 16 h, the bacteria were collected, and the His‐tagged
   proteins were extracted using a protein purification kit (Beyotime,
   Cat# P2229S) according to the manufacturer's instructions.

Arf1 Activity Assay

   Cells were cultured in RPMI 1640 overnight and transfected with the
   p23‐GGA3‐FLAG plasmid the next day. After 24 h, the cells were treated
   with GCA, BFA, DU101 or DU102. After 24 h of treatment, the cells were
   washed with 1X PBS and lysed with IP buffer (25 mM Tris‐HCl, pH 7.4),
   150 mM NaCl, 1% NP‐40, 1 mM EDTA, 5% glycerol, and protease inhibitor).
   The cell lysates were centrifuged at 13 000 × g for 10 min, and the
   supernatant was transferred to a new 1.5 mL centrifuge tube. The
   protein concentration was detected by a BCA Protein Assay Kit
   (Beyotime, Cat# P0012). An equal amount of protein was used in the
   GGA3‐IP experiment. 500 µL supernatant was mixed with 10 µL of 0.5 M
   EDTA pH 8.0, 5 µL of 10 mM GTPγS, and 5 µL of 100 mM GDP and vortexed.
   The mixture was incubated at 30 °C for 15 min with constant agitation.
   To terminate the reaction, 32 µL of 1 M MgCl2 was added to the mixture
   and vortexed. Then, Flag‐tagged magnetic beads (Beyotime, Cat# P2181M)
   were added to the mixture and incubated at 4 °C overnight. After
   incubation, the magnetic beads were washed 3 times, resuspended in
   protein loading buffer, and examined by immunoblotting with anti‐Arf1
   antibody (Proteintech, Cat# 10790‐1‐AP). Protein bands were visualized
   with an Omni‐ECL Femto Light Chemiluminescence Kit (Epizyme, Cat# SQ201
   L) on a Bio‐Rad ChemiDoc Touch (America).

Single‐Cell Sequencing

   MYC‐ON mice treated with DU101, DU102, or DMSO for 6 weeks were
   euthanized. Liver tumors were minced into small pieces and dissociated
   with a mouse Tumor Dissociation Kit (Miltenyi Biotec, Cat#
   130‐096‐730). The single‐cell suspension was centrifuged for 5 min at
   2000 rpm at room temperature. This pellet was washed with RBC lysis
   buffer (Beyotime, Cat# C3702) for 3 min, and the cells were spun for 5
   min at 2000 rpm at 4 °C. Then, the cells were stained with the fixable
   viability dye ZomBie NIR and anti‐mouse CD45‐PE antibody as well as
   anti‐mouse CD326 (Ep‐CAM)‐FITC, and sorted by BECKMAN COULTER MoFlo
   XDP. FACS sorted CD45+ cells and CD45‐ cells at the ratio of 10:1, and
   cells were sequenced by the 10x Genomics Chromium Single Cell platform.
   The barcoded libraries were prepared following the manufacturer's
   guidelines. Subsequently, 3′ mRNA libraries were created and subjected
   to sequencing on one NextSeq run and one NovaSeq SP run. Each sample
   yielded a count of over 315 million reads during sequencing. The
   sequencing was set up as a 28 cycles + 75 cycles non‐symmetric run on
   the NovaSeq. Demultiplexing was performed, allowing for one mismatch in
   the barcodes. More than 97.2% of bases in the unique molecular
   identifier (UMI) have Q30 or above.

Quality Control of scRNA‐seq Data

   Computational analyses were performed in R (v4.2.1) and Python (v3.10).
   For quality control, first SoupX^[ [213]^38 ^] was used to quantify and
   remove contamination caused by ambient mRNAs after Cellranger.^[
   [214]^39 ^] Then Scrublet^[ [215]^40 ^] was used to detect doublets in
   single‐cell RNA‐seq data. Then a Seurat object was created using Seurat
   (v4.2.0)^[ [216]^41 ^] and only the cells predicted to be “FALSE” by
   Scrublet were retained. To exclude low‐quality cells and erythroid
   cells before hierarchical clustering, only the cells that met the
   following requirements were retained: i) 250 < nFeature_RNA < 6000, ii)
   percent.mt < 15, iii) percent.hb < 0.5. To exclude genes with extremely
   low expression, only the genes with more than 10 TRUE values were
   retained.

Multiple Sample Integration and Unsupervised Clustering

   For unsupervised clustering of all tumor‐resident cells, Seurat object
   was normalized using NormalizeData. 2000 highly variable genes were
   selected using FindVariableFeatures. Then the data were scaled in all
   genes using Scaledata and performed Principal Components Analysis (PCA)
   with RunPCA. Harmony^[ [217]^42 ^] was used to integrate three samples
   (DU101, DU102 and DMSO). FindNeighbors were performed on the top 40
   principal components and RunTSNE or RunUMAP were used for data
   visualization.

Supervised Clustering Analysis

   Monocle 2^[ [218]^43 ^] was used to perform supervised clustering on T
   cells based on the expression of Cd4, Cd8a, and Cd8b1 to isolate DPT
   cells and supervised clustering on DPT cells was performed based on
   expression of Tcf7 and Pdcd1 to isolate stem‐like DPT cells. In detail,
   ordering_genes were set as Cd4, Cd8a, and Cd8b1, or Tcf7 and Pdcd1, and
   run setOrderingFilter with parameters as ordering_genes.

Public scRNA‐seq Dataset Analysis and DPT Atlas Construction

   DISCOtoolkit ([219]https://github.com/JinmiaoChenLab/DISCOtoolkit) was
   used to filter and download data from DISCO dataset.^[ [220]^44 ^] The
   FilterDiscoMetadata function was used with the parameter tissue = liver
   and DownloadDiscoData was used to download the data. First low‐quality
   cells were excluded and only kept cells with “percent.mt” less than 15.
   Then T cells were isolated based on the non‐zero expression of CD3D
   CD3E or CD3G. Next, the doublets were removed using DoubletFinder.^[
   [221]^45 ^] In the end, DPT cells were isolated from total T cells
   based on the non‐zero expression of CD4 and CD8 genes (CD8A or CD8B).
   The same procedures were followed as the mouse data analysis was done
   and unsupervised and supervised clustering was performed to construct
   the human liver tissue DPT cell atlas.

Drug Affinity Responsive Target Stability (DARTS)

   The drug affinity responsive target stability experiment was performed
   as previously reported^[ [222]^31 , [223]^32 ^] Briefly, Cell lysates
   from independent biological replicates were aliquoted in equivalent
   volumes containing 100 µg of protein and incubated for 30 min at 25 °C
   with or without DU101 or DU102. Proteinase K from Tritirachium
   album (Sigma–Aldrich, St. Louis, MO, USA) was added simultaneously to
   all samples at a proteinase K: substrate mass ratio of 1:100 and
   incubated at 25 °C for 30 min.

Molecular Docking

   Relevant targets were entered on the UniProt website
   ([224]https://www.uniprot.org/) to search and download the human
   protein structure with a eutectic molecule (PDB ID: 1RE0) as the
   docking receptor, and then AutoDock Tools 1.5.6 software was used to
   dehydrate the protein, hydrogenate it, calculate the Gasteiger charge
   and save it as a PDBQT file. Small molecule compounds were downloaded
   from PubChem for the 2D structure
   ([225]https://pubchem.ncbi.nlm.nih.gov/) with ChemBio3D energy
   minimization and AutoDock Tools 1.5.6 software to add atomic charge and
   distribution of atom types for the compound. The rotation center was
   selected, and all flexible keys were rotated by default and saved in
   PDBQT format as the docking ligand. With proteins as receptors and
   small molecules as ligands, the active sites of molecular docking were
   determined according to the coordinates of ligands in the target
   protein complex. Center_x = 42.01, center_y = 7.38, center_z = 38.49,
   size_x = 43, size_y = 52, size_z = 40, and AutoDock Vina was used for
   semiflexible docking. The default values were used for the other
   parameters unless otherwise specified. Finally, Discovery Studio
   Visualizer 2019 was used to analyze the results of the conformation
   with the highest score, and PyMOL was used to make correlation
   diagrams.

Quantification and Statistical Analysis

   Please refer to the figure captions for comprehensive experimental
   details, encompassing descriptions of the samples (cells, mice), and
   statistical information. The figure annotations specify the number of
   mice used, while the cell experiments were conducted with triplicate
   samples in two independent trials. Data analysis was carried out using
   GraphPad Prism 8.0 software (GraphPad Software). The data were
   presented as the mean ± SEM (standard error of the mean). Statistical
   distinctions were examined using two‐tailed Student's t‐tests and
   one‐way ANOVA (Bonferroni posttest), with significance established at a
   p‐value of ≤ 0.05.

Conflict of Interest

   The authors declare no conflict of interest.

Author Contributions

   Y.W., Q.L., Y.D., C.L., and J.Y. contributed equally to this work. Y.
   W., G.S., and S.X.H. conceived and designed the experiments. G. W., T.
   Y., C. L., and Y. W. performed the experiments. T. Y., C. L., Q. L., Y.
   W., and S.X.H. analyzed the data. T. Y., C. L., Q. L., Y. W., and
   S.X.H. wrote the manuscript.

Supporting information

   Supporting Information
   [226]ADVS-11-2404442-s001.pdf^ (3MB, pdf)

Acknowledgements