Exosomal circZNF451 restrains anti-PD1 treatment in lung adenocarcinoma via polarizing macrophages by complexing with TRIM56 and FXR1

Six LUAD cell lines A549, H1299, Calu-3, H1975, H1395, and LLC, HEK-293 T cells, and THP-1 cells were provided by the cell bank of the Chinese Academy of Sciences. Cells were cultured with DMEM or RPMI-1640 (Gibco, USA) with 10% fetal bovine serum (Gibco, USA) at 37℃ and 5% CO₂.

All plasmids used in this study were constructed by Genomeditech (Shanghai, China). Lentivirus production and infection were performed using a lentiviral packaging kit and liposomal transfection reagent (Yeasen, Shanghai). Sequences of shcircZNF451, siFXR1, siTRIM56, siELF4, and siIRF4 are listed in Supplementary Table 3.

ALL tumor tissues and peripheral blood were obtained from patients at Fudan University Zhongshan Hospital who underwent surgery between 2008 and 2010. The tumor tissues were frozen by liquid nitrogen and stored under -80 ℃. The sera of the peripheral blood were stored under -20 ℃. The tissue microarray (TMA) consisted of 113 LUAD tissues and peripheral tumor tissues. Histopathology was evaluated by two experienced pathologists. Patient clinical characteristics are described in Supplementary Table 2. All patients signed the consent form, and all human and animal work was approved by the Ethics Committee of Fudan University Zhongshan Hospital.

Exosomes from the sera of LUAD patients and culture medium of LUAD cell lines were extracted with ExoQuick exosome precipitation solution (SBI, USA) according to the manufacturer’s protocol. Isolated exosomes were further used for quantitative real-time PCR (qRT-PCR) and circRNA sequencing.

Western blotting, qRT-PCR, and IHC were performed according to our previous study [13], and details are provided in the Supplementary Methods. The antibodies used in this study were presented in supplementary Table 4. The primers for qRT-PCR were presented in supplementary Table 5. The original images of western blotting were presented in additional file 4.

The RNase R was used to digest the linear mRNA, while the circRNAs could endure its digestion. The RNase A could be used to digest the circRNAs. A total of 2 μg of RNA was incubated with 10 U of RNase R or RNase A (Thermo Fisher, USA) for approximately 30 min. Digested RNA was further purified with a TRIeasy Plus total RNA kit (Yeasen, Shanghai) and analyzed by qRT-PCR.

To determine the localization of circZNF451, FXR1, and TRIM56 in THP-1 cells, cells were fixed with 4% formaldehyde, prehybridized, and further hybridized with fluorescein-labeled circZNF451 probe for 12 h at 37℃. Cells were blocked with goat serum for 30 min and incubated with primary antibody for 1 h. After washing with phosphate-buffered saline (PBS), cells were incubated with Alexa Fluor 488-conjugated secondary antibody (Yeasen, Shanghai) for 1 h. Cells were sealed with mounting medium plus DAPI, and images were captured with a fluorescence microscope. The probe of circZNF451 was listed in supplementary Table 5. To evaluate the expression of circZNF451 in the LUAD and peri-tumor tissues of TMA stained with FISH, we selected four represented areas in each dot of the TMA. The mean optical intensity of each area was calculated by the Image-Pro Plus software (Media Cybernetics; Silver Spring, MD, USA) and the average of the mean optical intensity in the four areas was regarded as the score for the expression of circZNF451.

Cytokine production in THP-1 cell supernatants was measured using the human cytokine/chemokine microarray (Luminex, USA). After the stimulation by LPS for 24 h, concentrations of interleukin-1β (IL-1β), IL-1 receptor antagonist (IL-1Ra), granulocyte–macrophage colony-stimulating factor (GM-CSF), CXCL-1, and IL-10 in the supernatant were measured by enzyme-linked immunosorbent assay (ELISA). ELISA kits used are listed in Supplementary Table 4.

To analyze immune cell infiltration in subcutaneous tumors, tumor tissues were ground into single cells, filtered with a 70-μm cell strainer, and purified by Percoll gradient (Yeasen, Shanghai). Then, ~ 2 × 10⁶ cells were washed with PBS and stained with antibodies and Fc blocker (BioLegend) for 30 min on ice. For Foxp3 staining in Treg cells, cells were permeabilized with 0.1% Triton X-100 and stained with antibody for 90 min. To stain cytokines in the cytoplasm, cells were stimulated with phorbol myristate acetate (PMA; 10 ng/mL) and ionomycin (1 μg/mL) and blocked by brefeldin A (1 μg/mL) for 4 h and stained with surface marker antibodies for 30 min. Cells were fixed, incubated with 0.2% saponin buffer for 20 min, and stained with cytokine antibodies for 30 min. Samples were analyzed with a Fortessa flow cytometer (BD), and data were analyzed with FlowJo V10.6. Antibodies used for flow cytometry are listed in Supplementary Table 4.

A total of 1 × 10⁶ macrophages cells (stimulated by 100 μg/mL PMA for 24 h in THP-1 cells) were seeded in the lower chamber of six-well Transwell chambers (Corning, USA), and 5 × 10⁵ LUAD cells were seeded in the upper chamber. After 48 h, the supernatant and macrophages were collected for further analysis. After isolation of CD8⁺T cells from healthy donors by using anti-CD8 immunomagnetic beads, cells were stimulated with anti-CD3 (10 μg/mL), anti-CD28 (1.5 μg/mL), and IL-2 (200 U/mL) and cultured with supernatant from the coculture system for 48 h. The proliferation of CD8⁺ T cells was detected by CFSE according to the manufacturer’s protocol (biolegend, USA) and analyzed by the flow cytometry.

To evaluate interactions between FXR1 and TRIM56, THP-1 cells were lysed with co-IP buffer supplemented with proteinase, RNase, and phosphatase inhibitors and centrifuged at 12,000 rpm for 15 min. The supernatant was incubated with the beads, which were incubated with FXR1 antibody. After 12 h, beads were washed with lysis buffer, mixed with sodium dodecyl sulfate (SDS) loading buffer, and analyzed by Western blotting.

The RNA pulldown assay was performed with an RNA antisense purification kit from BersinBio (Guangzhou, China). Briefly, a total of 2 × 10⁷ cells were lysed with lysis buffer supplemented with proteinase and RNase inhibitor and centrifuged at 16,000 × g for 10 min. The supernatant was incubated with 40 pmol of biotinylated circZNF451 probe for 3 h at 37℃. Twenty microliters of streptavidin beads were washed and hybridized with the supernatant for 30 min at room temperature. Enriched proteins bound to beads were eluted and used for mass spectrometry and western blotting.

RIP was performed using a Magna RIP kit (Millipore, USA) according to the manufacturer’s protocol. A total of 10 [7] cells were fixed, lysed, and centrifuged, and the supernatant was hybridized with the streptavidin beads for 12 h. RNA was eluted from the beads, and qRT-PCR was performed.

Cells (6 × 10⁶) were cross-linked with 1% formaldehyde for 10 min and quenched with 0.2 g of glycine. ChIP was performed with a SimpleChIP Plus sonication chromatin IP kit (Cell Signaling Technology, USA) according to the manufacturer’s protocol. The mixture was digested, and DNA was sonicated into 200- to 800-base pair (bp) fragments. After incubation with ELF4 antibody (Santa Cruz, USA) for 12 h, the DNA–protein complexes were purified with magnetic beads, and eluted DNA was analyzed by qRT-PCR.

Wild-type (WT) and mutant IRF4 promoters were copied into pEZX-FR01 and cotransfected into HEK293T cells with the ELF4 overexpression plasmid. After 48 h, cells were lysed, and a luciferase reporter assay kit (Promega, USA) was used to measure Renilla and firefly luciferase activity.

C57BL/6 J mice were obtained from the Vital River (Beijing, China). C57BL/6-Lyz2^CreERT2 and C57BL/6 J-ELF4^{em1(flox)Smoc} mice were constructed by the Shanghai Model Organisms Center (Shanghai, China). A total of forty-five mice were used in this study and each group contained five ones. The mice were fed in a pathogen-free environment. Before the subcutaneous implantation of LLC cells or sacrifice, the mice were anesthetized by 1% Pentobarbital sodium (80 mg/kg). Mice with conditional knockout of ELF4 in macrophages were constructed by C57BL/6-Lyz2^CreERT2 × C57BL/6 J-ELF4^{em1(flox)Smoc} cross. ELF4 knockout in macrophages was performed by tamoxifen administration (2 mg/mouse) for 5 days. The subcutaneous xenograft model is detailed in the Supplementary Methods.

SPSS 23.0 (IBM, USA) and GraphPad Prism 8.0 were used to analyze statistical differences. The clinical characteristics were analyzed by chi-squared tests. Differences between two groups were analyzed by Student’s t test, and Spearman correlation analysis was performed to determine correlation coefficients. For three groups or more, the one-way ANOVA test was used to compare the difference between groups. Univariate and multivariate Cox regression analysis was applied to find the independent risk factors for LUAD prognosis. Kaplan–Meier and log-rank tests were used to compare survival in different groups. A P value of < 0.05 was regarded as statistically significant; *P < 0.05; **P < 0.01; ***P < 0.001.

To illustrate the potential role of exosomal circRNAs in immunotherapy resistance, the circRNA sequencing was performed to determine the level of circRNAs in peripheral blood from 6 LUAD patients treated with anti-PD1; of these, three patients were classified as progression disease (PD), and three patients were classified as partial response (PR) according to the RECIST1.1 evaluation (Supplementary Table 1) [14]. Fifty-five circRNAs exhibited differential abundance between PD and PR patients in the peripheral blood. Of these, 23 circRNAs were significantly upregulated in patients with PD, and 32 circRNAs were downregulated (FC_(PD/PR) > 2, P < 0.05). Figure 1A). Then, the exosomes in the sera of the six patients were extracted (Fig. 1B). The exosomal markers TSG101 and CD63 were verified by western blot (Fig. 1C) and the level of 4 most significantly changed circRNAs (FC > 4, P < 0.05) were detected via qRT-PCR. The results suggested that the level of has-circ-0002638 (circZNF451) was increased in the exosomes of PD patients compared to that in PR patients (Fig. S1A). Detection of circZNF451 in biopsied tumor tissues via qRT-PCR (n = 6) also showed that the expression of circZNF451 in exosomes and tumor tissues is highly consistent (Fig. S1B). Sanger sequencing identified the closed-loop structure of circZNF451 (Fig. 1D).

We further detected the expression of circZNF451 in LUAD cell lines and exosomes in the supernatants from corresponding cells via qRT-PCR. High level of circZNF451 was observed in A459 and H1299 cells, and low level was observed in H1395 and H1975 cells (Fig. 1E and S1C). Moreover, circZNF451 remained stable after incubation with RNase R (Fig. 1F). FISH staining of circZNF451 confirmed localization to the cytoplasm of different LUAD cell lines. We also separated the RNA from the nuclear and cytoplasm of H1395-circZNF451 and H1975-circZNF451 and found that the overexpression of circZNF451 was mainly detected in the cytoplasm (Fig. 1H and S1D). Importantly, we extracted the exosomes from the sera of 4 LUAD patients before and after surgery and identified the dramatically declined level of exosomal circZNF451 after the resection of malignant lesions (Fig. 1G). The situations of the four patients were presented in additional file 3.

We further examined the role of circZNF451 in the prognosis of LUAD patients. FISH staining of circZNF451 in a TMA containing 113 LUAD and peritumor tissues revealed increased circZNF451 levels in tumor tissues (Fig. 1I). Cox regression and Kaplan–Meier analyses established circZNF451 as an independent factor contributing to poor LUAD prognosis (Fig. 1J, K). Thus, our results suggested that the level of circZNF451 was elevated in the exosomes of advanced LUAD patients resistant to PD1 blockade and highlighted the potential influence of circZNF451 on the progression of LUAD.

To further explore the potential impacts of circZNF451 on the TME, the IHC staining of CD4⁺T, CD8⁺T, CD56⁺NK, Foxp3⁺Treg, CD11C⁺dendritic cell (DC), CD163⁺macrophage and CD86⁺macrophage was performed in the 113 LUAD tissues. We found that the infiltration of CD8⁺T and CD86⁺macrophage was exhausted while the infiltration of Foxp3⁺Treg and CD163⁺macrophage was increased in circZNF451^high group (Fig. S2A). The expression of circZNF451 was positively correlated with the infiltration of CD163⁺ macrophage but negatively correlated with that of CD8⁺ T and CD86⁺ macrophage (Fig. 2A). To further verify these results, we constructed the LLC-circZNF451 cell lines (Fig. S2B) and an in vivo model in immunocompetent mice (C57BL/6 J) was developed with subcutaneous implantation of LLC-circZNF451 cells. CircZNF451 accelerated growth of subcutaneous tumors and lead to shorter survival in mice (Fig. 2B, C). Flow cytometry analysis confirmed the exhaustion of CD8⁺ T cells and increased infiltration of TAMs in tumors with circZNF451 overexpression (Fig. 2D). CircZNF451 also induced conversion of cytotoxic lymphocytes to exhausted CD8⁺ T cells, with evidence of increased expression of inhibitory immune checkpoint PD1 and TIM3 and decreased secretion of cytotoxic cytokines interferon-γ (IFN-γ) and granzyme B (GZMB; Fig. 2E, F). Additionally, we discovered increased infiltration of CD163⁺ macrophages and exhaustion of CD86⁺ macrophages after circZNF451 overexpression (Fig. 2G), suggesting that exosomal circZNF451 might elicit M2 polarization of macrophages.

Furthermore, we investigated the impacts of exosomal circZNF451 enrichment on macrophages and CD8⁺ T cells and constructed a coculture model by culturing LUAD cell lines with CD8⁺ T cells or macrophages (PMA-stimulated THP-1 cells; Fig. 2H). The circZNF451 was silenced in A549 and H1299 (high level of endogenous circZNF451) and overexpressed in H1395 and H1975 (low level of endogenous circZNF451, Fig. S2C). Coculture of H1395-circZNF451 cells and CD8⁺ T cells from healthy donors did not impair the secretion of IFN-γ and GZMB in CD8⁺ T cells (Fig. S2D), but coculture of H1395-circZNF451 cells and LPS-stimulated macrophages induced the enhanced expression of M2 polarization markers CD163, Arg1 and IL-10 (Fig. 2I and S2E). The qRT-PCR showed enhanced level of circZNF451 in macrophages after co-culturing with H1395-circZNF451 cells while the administration of exosome inhibitor GW4869 (20 μM) abolished the enrichment (Fig. S2F). The administration of the GW4869 in the coculture model also abolished M2 polarization of macrophages (Fig. S2G, H). Moreover, the supernatant from the macrophage/H1395-circZNF451 cells induced the dysfunction of cytotoxic CD8⁺T cells (Fig. S2I). The CFSE assay also confirmed the impaired proliferation of CD8⁺ T cells cultured by the supernatant from the macrophage/H1395-circZNF451 cells (Fig. S2J). We further identified the above results by coculturing macrophages with A549. Similarly, the knockdown of circZNF451 in A549 diminished the M2 polarization of macrophages, which also rescued the dysfunction and proliferation of CD8⁺ T cells (Fig. S2K-N). In addition, cytokines and chemokines in the supernatant of the coculture model were measured with a Luminex human cytokine/chemokine microarray, and upregulation of anti-inflammatory cytokines, including IL-10 and IL-1Ra, and downregulation of proinflammatory cytokines, including IL-1β, GM-CSF (CSF2), and CXCL1, were identified after coculturing with H1395-circZNF451 cells (Fig. 2J). These results demonstrate that circZNF451 can induce M2 polarization of macrophages, which further elicits the dysfunction of CD8⁺ T cells and resets an immune-suppressed TME.

CircRNAs can act as microRNA (miRNA) sponges. However, a RIP assay showed that circZNF451 could not be enriched by AGO2, which suggests that circZNF451 may not function as a miRNA sponge (Fig. 3A). Thus, we further tested whether circZNF451 function by interacting with proteins. An RNA pulldown assay was performed with a biotinylated circZNF451 probe in macrophages, and the silver staining showed a specific band at around 70 kDa (Fig. S3A). Mass spectrometry analysis of the band showed that the top three enriched proteins were HNRNPM, FXR1, and EIF3L, and AGO2 was not included (Additional File 1). The mass spectrometry data was further overlapped with the predicted candidates in circInteractome and the most detected RNA binding motifs in catRAPID database, which strongly suggested FXR1 as the downstream of circZNF451 (Fig. S3B). RNA pulldown and RIP assays further confirmed the specific interaction between circZNF451 and FXR1 (Fig. 3B, C). After coculturing with A549 cells for 48 h, RNA FISH and immunofluorescence staining of circZNF451 and FXR1 confirmed colocalization in the cytoplasm of macrophages (Fig. 3D). Furthermore, coculture with A549-shcircZNF451 and H1299-shcircZNF451 cells enhanced FXR1 protein levels in macrophages (PMA-stimulated THP-1 cells), while coculture with H1395-circZNF451 and H1975-circZNF451 cells showed the opposite trend, which was reversed after GW4869 administration (20 μM; Fig. 3E and S3C). Nonetheless, FXR1 mRNA expression did not change (Fig. 3F), suggesting that circZNF451 regulates expression of FXR1 protein.

FXR1 contains two KH domain (KH1 and KH2) and one RGG box, which are responsible for the interaction with RNAs [15]. To identify the exact RNA-binding domain for circZNF451, we constructed three Flag-tagged vectors encoding the fragments of FXR1. After co-transfection with circZNF451 into HEK293T cells, RIP assay showed that KH2 was mainly responsible for binding to circZNF451 (Fig. 3G). Here, we also explored the binding motif of FXR1 in the RBPsuite database and identified one potential binding region for FXR1 in the circZNF451 sequence (Fig.... S3D). We therefore mutated the FXR1-binding site in circZNF451 (Fig. S3E), which dramatically impaired the interaction between circZNF451 and FXR1 in macrophage (Fig. 3H).

Regulation of FXR1 protein rather than mRNA expression by circZNF451 indicates circZNF451 may destabilize FXR1 protein in macrophages. Indeed, circZNF451 overexpression downregulated FXR1 protein levels, administration of the protease inhibitor MG132 partially restored it (Fig. 3I). Overexpression of circZNF451 significantly enhanced FXR1 ubiquitination, which was reversed by circZNF451 mutation or knockdown. Administration of GW4869 (20 μM) also abolished the enhanced ubiquitination of FXR1 induced by circZNF451 (Fig. 3J). Together, these results suggest that LUAD-derived exosomal circZNF451 may drive FXR1 ubiquitination in macrophages.

To identify the ubiquitinating enzyme for FXR1, we reanalyzed the mass spectrometry data and identified two E3 ligases, TRIM56 and TRIM25. RNA pulldown and RIP confirmed that circZNF451 could specifically interacted with TRIM56 (Fig. 4A, B). We further performed a Co-IP assay in macrophages (PMA-stimulated THP-1 cells) and found that FXR1 could bind TRIM56 after coculturing with A549 for 48 h (Fig. 4C). Knockdown of TRIM56 in macrophages abolished the reduction of FXR1 mediated by H1395/H1975-circZNF451(Fig. 4D). Consistently, ubiquitination of FXR1 was weakened following TRIM56 knockdown (Fig. 4E).

The C terminus (amino acids 693–750) of TRIM56 is a region critical for RNA binding [16]. Thus, we constructed a Flag-tagged mutant TRIM56 with the deletion of C-terminal and transfected it into macrophages. After coculture with A549 cells, Western blotting showed that the transfection of flag-TRIM56 enhanced the downregulation of FXR1 while the mutation of TRIM56 rescued it (Fig. 4F). RNA pulldown and RIP assays showed that mutation of TRIM56 abolished the interaction between circZNF451 and TRIM56 in macrophage (Fig. 4G and S4A). We also observed that mutation in circZNF451 and digestion with RNase A impeded the interaction between TRIM56 and FXR1 in macrophages, which were consistent with the results confirmed by the knockdown of circZNF451 in A549 and H1299 (Fig. 4H and S4B). RNA FISH and immunofluorescence showed that circZNF451, FXR1, and TRIM56 were mainly presented in the cytoplasm of macrophages (Fig. 4I). These data suggest that circZNF451 functions as a scaffold to enhance the interaction between TRIM56 and FXR1 proteins in macrophages.

Interactions between circZNF451, FXR1, and TRIM56 indicate the impacts of the FXR1–circZNF451–TRIM56 complex on the macrophage immune phenotype. We first knocked down FXR1 in macrophage (PMA-stimulated THP-1 cells) with short interfering RNA (Fig. 5A). After LPS stimulation, flow cytometry analysis showed that reduction of FXR1 leads to M2 polarization of macrophages (Fig. 5B and S4C). Supernatant from macrophage-siFXR1 cells further induced cytotoxic dysfunction of CD8⁺ T cells, with decreased secretion of IFN-γ and GZMB (Fig. 5C). We also detected cytokines in the supernatant via ELISA and observed a significant upregulation of anti-inflammatory cytokines (e.g., IL-1Ra and IL-10) and a reduction in pro-inflammatory cytokines (e.g., GM-CSF, CXCL1, and IL-1β) after silencing FXR1 in macrophages (Fig. 5D).

By contrast, silencing TRIM56 in macrophage inhibited M2 polarization after coculturing with H1395-circZNF451 cells (Fig. 5E, F), and supernatant from the coculture system rescued the cytotoxic function of CD8⁺ T cells inhibited by circZNF451 overexpression (Fig. 5G). Additionally, GM-CSF, CXCL1, and IL-1β levels were elevated in the supernatant, while IL-1Ra and IL-10 levels were dramatically reduced after silencing TRIM56 in macrophages (Fig. 5H). However, overexpression of mutant TRIM56 had no impact on M2 polarization and production of pro- or anti-inflammatory cytokines in macrophages after coculture with A549 cells (Fig. 5I and S4D, E).

The RBP FXR1 regulates the stability of downstream mRNAs [17, 18]. Thus, RNA sequencing was performed to detect the profiles of mRNAs in macrophages (PMA-stimulated THP-1 cells) after coculture with H1395-circZNF451 and H1395-control cells. We found that the macrophages co-cultured with H1395-circZNF451 and stimulated by LPS upregulated ARG1, MRC1 (CD206), CD163, CD274 (PD-L1), IL-Ra (IL-1RN) and IL-10, while the expression of GM-CSF (CSF2), CXCL1, and IL-β was reduced (Fig. 6A). KEGG and GO analysis revealed a significant change in immune-related responses (Fig. 6B). Further analysis revealed that 65 immune-related genes among the identified differently expressed mRNAs overlapped with the predicted candidates of FXR1 in the ENCORI database, and 10 potential candidates were identified to be the downstream of FXR1 in macrophages. Because ELF4 was demonstrated the most significant shift in expression among the 10 candidates, and previous study identified a critical role for ELF4 in regulating inflammatory responses [19], we speculated that FXR1 degradation might enhance the anti-inflammatory phenotype of macrophages via ELF4 (Fig. 6C, Additional File 2). Previous studies reported the binding of 3’UTR was important for FXR1 to promote the degradation of the target genes. Thus, we constructed the vectors containing the 3’UTR of ELF4 and transferred it into the macrophages. The dual-luciferase reporter assay identified that the relative luciferase activity was obviously weakened while the co-transfection with the siFXR1#3 partially restored it, which confirmed that FXR1 could bind the 3’UTR of ELF4 to regulate its stability (Figure S4F). Interestingly, we found that both the overexpression of circZNF451 in LUAD and silencing of FXR1 in macrophages could contribute to enhanced expression of ELF4 (Fig. S5A, B). Also, immunofluorescence staining of ELF4 and CD163 in the TMA suggested a positive correlation between the expression of circZNF451 and infiltration of ELF4⁺CD163⁺ macrophages (Fig. 6D).

Here, we further identify the impact of ELF4 on M2 polarization of macrophages. By interfering the ELF4 in macrophages, we found that the reduction of ELF4 dramatically abrogated M2 polarization endowed by H1395- circZNF451 (Fig. 6E–G). Silencing of ELF4 also reduced the production of IL-1Ra and IL-10, but increased the secretion of GM-CSF, CXCL1, and IL-1β in the macrophage supernatant (Fig. S5C).

Analysis of ChIP-seq data from a published study [20] revealed that recruitment of ELF4 led to a significant induction of interferon regulatory factors (IRFs), including IRF4, which was significantly increased in our RNA-seq data (Fig. 6A). Previous studies had identified IRF4 as an important mediator of M2 polarization of macrophages [21], suggesting that ELF4 could elicit M2 polarization of macrophages via activation of IRF4. We identified the ELF4 binding motif in JASPAR and found three putative binding sites, B1 (− 351 to − 346), B2 (− 460 to − 455), and B3 (− 701 to − 696), with the core sequence of GGAA within a 2,000-bp distance from the IRF4 transcription start site (Fig. 6H). The ChIP assay showed that B1 had the highest binding affinity with ELF4 (Fig. 6I). We next constructed vectors of the promoter containing the mutation of the three binding sites (GGAA to AAAA), and luciferase activities in HEK293T cells were most significantly inhibited in the vectors containing the mutant sequence in B1 (Fig. 6J). Knockdown of ELF4 also decreased IRF4 expression in macrophages at both the mRNA and protein levels after coculture with H1395-circZNF451 cells (Fig. 6K and S5D). Moreover, knockdown of circZNF451 in LUAD cell lines decreased IRF4 expression in macrophage, while overexpression of circZNF451 showed the opposite trend, which was reversed by administration of GW4869 (Fig. 6L and S5E). Silencing of IRF4 also inhibited M2 polarization in macrophages cocultured with H1395-circZNF451 (Fig. 6M). Consistently, macrophage cytokine levels followed similar patterns as the knockdown of ELF4 (Fig. S5F). Collectively, these results suggest that degradation of FXR1 by circZNF451 activates the ELF4–IRF4 pathway in macrophages.

Considering the potential regulation of ELF4 on the phenotype of macrophages, we speculated that ELF4 knockout might rescue the immune-suppressed TME induced by exosomal circZNF451 and enhance sensitivity to PD1 blockade. Thus, we constructed a mouse model with conditional knockout of ELF4 in macrophages via C57BL/6 J-Lyz2^CreERT2 and C57BL/6 J-ELF4^{em1(flox)Smoc} mice, which were subcutaneously implanted with LLC-circZNF451 cells. PD1 blockade was administered 7 days later at an interval of 3 days. Knockout of ELF4 in macrophages dramatically enhanced the antitumor effect of PD1 blockade and inhibited tumor growth (Fig. S6A). The IHC staining showed that PD1 blockade enhanced CD8⁺ T cell infiltration and led to the exhaustion of CD163⁺ macrophages in ELF4-knockout mice with overexpression of circZNF451 (Fig. 7A). Consistently, the flow cytometry analysis also confirmed the similar results and identified the recovery of cytotoxic CD8⁺ T cell after the administration of PD1 blockade in ELF4-knockout mice with overexpression of circZNF451 (Fig. 7B, C and S6B).

To further verify the impacts of exosomal circZNF451 on the TME and immunotherapy, we selected ten LUAD patients receiving PD1 blockade (Opdivo) treatment, in whom six were evaluated as PD and four were evaluated as PR (Fig. 7D and Supplementary Table 6). Sera and biopsied tumor tissues of the ten patients were collected for analysis. Enrichment of circZNF451 in exosomes was significantly higher in PD patients and was consistent with the expression of circZNF451 in paired tumor tissue (Fig. 7E, F), and enrichment of circZNF451 in exosomes was negatively associated with tumor shrinkage (Fig. 7G). The mIF staining of CD8⁺ T cells and ELF4⁺CD163⁺ macrophages showed high infiltration of ELF4⁺CD163⁺ macrophages and low infiltration of CD8⁺ T cells in the tumor tissues of PD patients (Fig. 7H).

Here, we show that in LUAD, tumor-derived exosomal circZNF451 may elicit ubiquitination of FXR1 via the E3 ligase TRIM56 in macrophages. Degradation of FXR1 further reshapes the immune-suppressed TME via activation on the downstream ELF4–IRF4 pathway, which enhances the anti-inflammatory phenotype of macrophages and dysfunction of cytotoxic CD8⁺ T cells (Fig. 8) [22].