p53 Gain-of-Function Mutation Induces Metastasis via BRD4-Dependent CSF-1 Expression

Tumor metastasis is a major barrier to effective cancer therapy, and metastasis is associated with more than 90% of cancer-related mortality (1, 2). Both intrinsic factors, including genomic instability and epigenetic alterations, and extrinsic factors, such as microenvironmental cues, have been implicated in the metastatic proclivity of cancer cells (3). The precise biological mechanisms underlying metastasis, however, require more extensive elucidation in order to improve the potential for novel therapeutics.

TP53 (Trp53 in mice) that encodes the p53 protein is mutated frequently in human cancers, and TP53 mutations correlate with higher metastatic rates and poor patient survival (4). Approximately 70% of TP53 mutations are missense, resulting in loss of wild-type p53 tumor suppressor activity and/or acquisition of dominant-negative (DN) functions (5, 6). Additionally, some missense mutations may result in gain-of-function properties that are protumorigenic as a result of DNA binding, interactions with other transcription factors and chromatin modifiers or modulation of p63, an ortholog of p53 (7–11). The most frequent “hotspot” TP53 missense mutations cluster within the p53 DNA-binding domain and include codons 175 (codon 172 in mice, which is the focus of this study), 245, 248, 249, 273, and 282 (12). These hotspot TP53 mutations have been reported to have different roles in promoting cancer cell proliferation, survival, migration, invasion, epithelial-to-mesenchymal transition (EMT), angiogenesis, and metastasis (4, 13–15). However, it is not clear how these mutations foster metastasis, catalyzing our desire to elucidate the underlying mechanisms.

Esophageal cancer is highly aggressive and is the sixth leading cause of cancer-related mortality worldwide. It is projected that there will be 957,000 new esophageal cancer cases worldwide by 2040 (16). Esophageal cancer comprises two predominant subtypes, esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC), with ESCC being more prevalent globally (17, 18). Most patients with ESCC or EAC present with metastases, primarily to the lungs (18). TP53 mutations are found in approximately 80% of ESCC cases, with the majority of mutations located in the DNA-binding domain (19, 20).

How TP53 mutations can modulate growth factors, cytokines, and chemokines to foster ESCC metastasis requires elucidation. Our novel study establishes colony-stimulating factor-1 (CSF-1) as one of the secreted cytokines that potentiates ESCC tumor cell invasion and lung metastasis in a Trp53^R172H-dependent manner. Canonically, CSF-1 recruits M2-like tumor-associated macrophages (TAM) and CD11b⁺Gr-1^LowLy6C^High myeloid-derived suppressor cells (MDSC) by binding to its tyrosine kinase receptor colony-stimulating factor-1 receptor (CSF-1R; refs. 21, 22). Although it has not been studied extensively, CSF-1R can be expressed on epithelial cells, regulating downstream pathways involved in proliferation, differentiation, and metastasis (23, 24).

We have demonstrated that p53-R172H occupies the Csf-1 promoter and regulates its transcription. The Csf-1 gene locus contains high levels of histone 3 lysine 27 acetylation (H3K27ac), a promoter and enhancer marker that is indicative of active transcription (25, 26). The epigenetic reader bromodomain-containing domain 4 (BRD4), a member of the bromodomain and extraterminal (BET) protein family, binds to acetylated lysine residues at the promoter and super-enhancers, leading to the assembly of the transcription complex and the transcription initiation of oncogenes such as c-Myc and Fosl (27). Therefore, we hypothesized that BRD4 might interact with p53-R172H to induce Csf-1 expression. We have used genetic deletion and complementary pharmacologic inhibition studies to demonstrate that loss of CSF-1 or BRD4 significantly reduces tumor cell invasion and lung metastasis through regulation of Stat3 activation and EMT in lung metastasis in a p53-R172H–dependent manner. Therefore, we advocate targeting the CSF-1/CSF-1R axis in patients with metastatic ESCC, an area bereft of impactful therapeutics. Given the common genomic properties of ESCC, head/neck SCC, and lung SCC (28–31), our nomination of the CSF-1/CSF-1R axis may be applicable to therapeutic approaches for other SCCs.

We generated a 4-nitroquinoline N-oxide (4-NQO) and Trp53^R172H-driven ESCC mouse model and derived cell lines from esophageal tumors of Trp53^+/+, Trp53^−/−, and Trp53^R172H/− mice after 4-NQO administration in drinking water (32). 4-NQO is a water-soluble quinoline derivative that causes DNA adduct formation and production of reactive oxygen species, resulting in oral-esophageal squamous cell carcinoma in mice (33). To better understand potential mediators of p53-R172H–driven metastasis, we performed bulk RNA-seq of metastatic ESCC cells to the lungs with p53^R172H/− (control) or Trp53^R172H depletion (shCtrl-M and shTrp53-M, respectively). We verified p53 protein levels in the different experimental conditions (Supplementary Fig. S1A; ref. 34).

From the RNA sequencing (RNA-seq) data, we found 260 upregulated genes and 291 downregulated genes in shCtrl-M compared with shTrp53-M (P_adj < 0.05 and |log₂ fold change|>2; Fig. 1A). As expected, Trp53 was among the most differentially expressed genes, which served as an internal control (Fig. 1A and B). Csf-1 was upregulated (∼2.8-fold) in shCtrl-M cells compared with shTrp53-M cells and was among the most upregulated genes, including the metallopeptidase inhibitor Timp1 that has been demonstrated previously to contribute to metastasis through senescence reprogramming (ref. 35; Fig. 1A and B; Supplementary Fig. S1B). Furthermore, we have identified genes that are downregulated in the primary and metastatic tumors (shCtrl and shTrp53), such as tight junction protein MAL and relevant proteins for vesicle trafficking and membrane link domain 3 (Marveld3). This gene has been demonstrated to reduce cell migration, EMT, and metastasis in non–small cell lung cancer (NSCLC) through the suppression of the Wnt/β-catenin signaling pathway (ref. 36; Supplementary Fig. S1C).

In a complementary approach, we conducted a cytokine array analysis on conditioned media from shCtrl-M (n = 5 cell lines) and shTrp53-M cells (n = 3 cell lines). Among the 62 cytokines examined, we identified FAS ligand, Fractalkine (CX3CL1), interleukin 6 (IL6), IL9, IL10, and CSF-1 with higher secretion from control cells, suggesting their potential involvement in Trp53^R172H-dependent metastasis (Fig. 1C). Of these six cytokines, CSF-1 and Fractalkine were secreted at the highest concentrations, but as mentioned, only Csf-1 was among the top upregulated genes. We assessed CSF-1 secretion by enzyme-linked immunosorbent assay (ELISA) and found that metastatic tumor cells with Trp53^R172H/− secreted more CSF-1 than those with Trp53^R172H/− depletion (Fig. 1D). Additionally, primary ESCC cells with Trp53^R172H/− secreted higher CSF-1 levels compared with either Trp53^−/− or Trp53^+/+ cells (Fig. 1E).

To elucidate the relationship between p53-R172H and Csf-1, we conducted ChIP-qPCR for p53 in five Trp53^R172H/− metastatic ESCC cell lines and found that the Csf-1 promoter region was highly enriched with p53-R172H occupancy, suggesting that Csf-1 is a gene target of p53-R172H (Fig. 1F). These data indicate that p53-R172H regulates Csf-1 transcription and suggest that CSF-1 may be a potential mediator of Trp53^R172H/−-dependent ESCC metastasis.

To study the functional role of the CSF-1/CSF-1R signaling in ESCC metastasis, we deleted Csf-1 in mouse Trp53^R172H/− tumor cells using a ribonucleoprotein (RNP)-based CRISPR/Cas9 approach (Supplementary Fig. S1D and S1E; Fig. 2A). We detected no changes in the expression of IL34, which encodes the other ligand that binds CSF-1R, upon the deletion of Csf-1 (Supplementary Fig. S1F; refs. 37, 38). Deletion of Csf-1 in Trp53^R172H/− tumor cells significantly reduced the number of invading tumor cells (Fig. 2B).

Next, we assessed the effects of CSF-1 on tumor progression in athymic nude mice (CD1^nu/nu) using the same cell lines described above. We first examined primary tumor growth following subcutaneous implantation and found that Csf-1 deletion reduced tumor growth starting at day 12 after injection (Fig. 2C), resulting in a 2- to 3-fold decrease in tumor weight at the endpoint (day 28; Fig. 2D). We then stained the formalin-fixed, paraffin-embedded (FFPE) tumor sections for Ki-67 and found that the nontarget (NT) controls had a higher percentage of Ki-67⁺ tumor cells than both Csf-1 knockout (KO) clones (Fig. 2E). Although these results are encouraging, cross-talk might occur between flank tumors that can influence tumor progression (39).

To investigate the contribution of Trp53^R172H/−-dependent CSF-1 to ESCC metastasis, we conducted lateral tail-vein injections with 1.0 × 10⁶Csf-1 KO or NT control Trp53^R172H/− tumor cells and assessed the tumor burden of lung metastases at the end of Week 8 (Fig. 2F). Csf-1 deletion significantly reduced the metastatic tumor burden in the lungs (Fig. 2G). Provocatively, when CSF-1 was overexpressed in Trp53^−/− and Trp53^+/+ cells (Supplementary Fig. S2A and S2B), there were no significant changes in the lung metastases as quantified by the number of YFP⁺ metastatic foci and the percentage of mice with metastatic tumors after injection with these cells (Supplementary Fig. S2C and S2D). Upon repeating the lateral tail-vein injections with shTrp53/Csf-1 KO cells and shTrp53/NT controls (Supplementary Fig. S2E), no difference was observed in the number of metastatic foci in the lungs (Fig. 2H). Overall, our data demonstrate that Trp53^R172H-dependent Csf-1 upregulation has a functional role in tumor cell invasion and metastasis.

Although it is known that CSF-1 mediates the cross-talk between tumor cells and macrophages (21, 22), the functional consequences of CSF-1/CSF-1R autocrine signaling have not been explored. We found that Csf-1r was expressed in metastatic and primary ESCC cells (Fig. 3A; Supplementary Fig. S2F), and to decipher its functional role, we treated Trp53^R172H/−, Trp53^−/- or Trp53^+/+ primary tumor cells with a mouse Csf-1r (AFS98) monoclonal neutralizing antibody. This treatment reduced the number of invading Trp53^R172H/− tumor cells, but it did not affect the invasive properties of Trp53^−/- or Trp53^+/+ cells (Fig. 3B).

To delineate effectors of the CSF-1/CSF-1R signaling, we identified gene-expression patterns that correlated with CSF-1 expression in The Cancer Genome Atlas (TCGA) SCC cases (including lung SCC, head and neck SCC, esophageal SCC, cervical SCC, and a subset of bladder cancers classified as basal squamous). Unbiased gene set enrichment analysis (GSEA) revealed that IL6/JAK/STAT3, EMT, and angiogenesis correlate significantly with high CSF-1 expression, and have been implicated in the mediation of tumor metastasis (Fig. 3C and D; Supplementary Fig. S3A and S3B). The EMT pathway was also enriched in shCtrl-M compared with shTrp53-M as revealed in our RNA-seq data with the ESCC tumor cells harboring Trp53^R172H/- (Supplementary Fig. S3C).

To pursue the connection between CSF-1 signaling and Stat3, we examined the levels of total Stat3 and phosphorylated Stat3 (Tyr705), which are associated with the activation of STAT3 signaling and EMT processes (40). Phosphorylated Stat3 levels were higher in cells expressing p53-R172H than in those expressing wild-type p53 or lacking p53 (Fig. 3E and F). Upon CSF-1R inhibition, both total and phosphorylated STAT3 decreased only in p53-R172H cells, and phosphorylated STAT3 levels were reduced to levels similar to those in p53-null or wild-type p53 cells (Fig. 3E–G). In addition, we evaluated the EMT pathway and discovered that treatment with CSF-1R neutralizing antibody downregulated Twist1 and Zeb1 in Trp53^R172H/− cells, but not p53-null or wild-type p53 cells (Fig. 3H). Metastatic lung lesions with Trp53^R172H Csf-1 KO #1 and KO #2 had a lower nuclear STAT3 H-score than lesions with intact CSF-1 (Fig. 3I). Upon CSF-1 neutralizing antibody treatment, the level of E-Cadherin (E-Cad) was elevated only in metastatic shCtrl-M cells with p53-R172H (Fig. 3J). Furthermore, the percentage of E-Cad^High cells was increased in the metastatic lung lesions with Trp53^R172H/− upon Csf-1 deletion (Fig. 3K). Similarly, metastatic tumors with Trp53^R172H had a higher H-score for another epithelial marker pan-cytokeratin (pan-CK) when Csf-1 was deleted (Supplementary Fig. S3D). Interestingly, in addition to E-Cad and pan-CK expression, Trp53^R172H/- metastatic lung lesions with Csf-1 KO had lower elongated cell scores than the control lesions, indicating a more epithelial morphology upon deletion of Csf-1 (Supplementary Fig. S3E). Overall, our results suggest that blocking the p53–R172H-dependent CSF-1/CSF-1R signaling axis maintains metastatic tumor cells in the epithelial state.

To elucidate the mechanism by which Trp53^R172H regulates Csf-1 transcription, we conducted Cleavage Under Targets and Release Using Nuclease sequencing (CUT&RUN-seq) for H3K27ac, an active enhancer and promoter marker (41), in Trp53^R172H/- and Trp53^+/+ ESCC cells. Principal component analysis (PCA) based upon H3K27ac profiles demonstrated that the ESCC cells with Trp53^R172H/− separated distinctively from the Trp53^+/+ cells (Supplementary Fig. S3F). Genome-wide analysis revealed a total of 904 regions with enriched H3K27ac and 1,255 regions with depleted H3K27ac in Trp53^R172H/− tumor cells compared with Trp53^+/+ cells (P_adj < 0.05; Supplementary Fig. S3G and S3H). Notably, H3K27ac was reduced across the Cdh1 gene body encoding the E-Cad and was enriched across the Vim gene body encoding the Vimentin in Trp53^R172H/− cells compared with Trp53^+/+ cells (Supplementary Fig. S3I). H3K27ac was enriched across the Csf-1 locus including the promoter site in Trp53^R172H/- ESCC cells compared with Trp53^+/+ cells (Fig. 4A), suggesting that the Csf-1 locus is active in the pre­sence of p53-R172H.

The epigenetic reader Brd4 binds to acetylated histone lysine residues at promoters and super-enhancers to regulate transcription initiation (27). We, therefore, investigated whether BRD4 was recruited to enriched H3K27ac sites in ESCC cells to induce Csf-1 transcription. We found that global BRD4 levels in Trp53^R172H/− tumor cells were ∼2-fold higher than those in Trp53^−/− cells and ∼3-fold higher than in Trp53^+/+ cells (Fig. 4B). To investigate the relationship between BRD4 and p53, we performed a proximity ligation assay (PLA) in mouse Trp53^R172H/−, Trp53^+/+, and Trp53^−/− ESCC cells and found a close proximity between BRD4 and p53-R172H, but not with wild-type p53 (Fig. 4C). To test whether these findings could be extended to human cells, we deleted endogenous TP53^R110L in TE11 esophageal cancer cells (42), and then ectopically expressed p53-R175H (Fig. 4D). We also deleted endogenous TP53^R175H in LS123 colorectal cancer cells (43) (Fig. 4E). Using these human cancer cell lines, we observed higher PLA signals in cells expressing BRD4 and p53-R175H than in cells lacking p53 or expressing p53-R110 L (Fig. 4F).

Next, cells were treated with a Brd4 small molecular inhibitor, JQ1 (44), or a PROTAC, ZXH-3-26, which selectively binds BRD4 and Cereblon, the substrate adaptor of the CUL4A E3 ligase complex that results in BRD4 ubiquitination and subsequent proteasomal degradation (45). In two separate Trp53^R172H/− ESCC cell lines, the PLA signal virtually disappeared upon BRD4 degradation, indicating that BRD4 was required for the PLA signal (Fig. 4G and H). As a control, C-MYC, which is known to be regulated by BRD4 (46), was downregulated (Fig. 4G). No difference was detected with another BET protein, namely, BRD2 (47, 48), suggesting the specificity of ZXH-3-26 (Supplementary Fig. S4A). JQ1 or ZXH-3-26 treatment significantly reduced Csf-1 expression (Fig. 4I) and secretion (Fig. 4J). These results indicated that BRD4 is a critical coregulator of p53-R172H–dependent Csf-1 expression.

As a complementary approach, we utilized two shRNAs to deplete BRD4 and verified that they effectively knocked down Brd4 in two mouse Trp53^R172H/− ESCC cell lines at both mRNA (Supplementary Fig. S4B) and protein (Fig. 5A) levels, although not affecting Brd2 (Supplementary Fig. S4C). Similar to ZXH-3-26 treatment of these cells, C-MYC was downregulated upon BRD4 depletion (Fig. 5A). Trp53^R172H/− ESCC cells with Brd4 knockdown secreted less CSF-1 (Fig. 5B) and had decreased invasion capability (Fig. 5C), which were recapitulated upon treating parental cells with pooled siRNAs targeting Brd4 (Supplementary Fig. S4D–S4G).

To investigate the role of BRD4 in lung metastasis, we treated mice with JQ1 or the vehicle control every other day between days 7 and 53 after tail-vein injection of Trp53^R172H/− ESCC cells (Fig. 5D). We observed a dramatic decrease in lung metastasis upon JQ1 treatment (Fig. 5E). Mechanistically, JQ1 treatment decreased nuclear Stat3 in metastatic tumors, indicating reduced activation of the protein and supporting the notion that BRD4 and CSF-1 are in the same axis in regulating STAT3 (Fig. 5F), compared with the control group. It is noteworthy that E-Cad expression increased upon depletion of Brd4 in Trp53^R172/- tumor cells (Fig. 5G), and the percentage of E-Cad^High cells increased in lung metastatic lesions upon treatment with JQ1 (Fig. 5H).

Furthermore, we collected blood sera from these experimental mice with lung metastasis at day 56 to assess circulating CSF-1 levels. Interestingly, although control mice with metastatic tumors had elevated circulating CSF-1 levels (2,800 pg/mL on average) compared with mice without any tumors, JQ1 treatment reduced the mean CSF-1 level to ∼1,800 pg/mL, which is approximately the level in mice without any tumors (Fig. 5I). These results provided an in vivo support for our hypothesis that BRD4 is an upstream regulator of CSF-1 expression in promoting Trp53^R172H-driven lung metastasis.

We next conducted an in-depth analysis of CSF-1, CSF-1R, and p53 expression in human ESCC samples. Initially, we stained tissue microarrays (TMA) containing 148 ESCC and 47 normal tissues for CSF-1, CSF-1R, and p53. Both CSF-1 (Fig. 6A) and CSF-1R (Supplementary Fig. S5A) levels were significantly higher in tumors than in adjacent normal tissues. Moreover, ESCCs with higher p53 expression, in most cases indicating mutant p53 leading to more stable p53, whose accumulation is regarded as a hallmark of cancer cells (49), had significantly elevated CSF-1 expression (Fig. 6B).

We then interrogated the TCGA for SCCs as noted previously (30) as TP53 mutations are the most frequently identified gene mutations in SCCs (50). We found that cases with “hotspot” missense p53 mutations had higher CSF-1 expression than those with non-hotspot or frameshift p53 mutations (Fig. 6C). Interestingly, when we focused on cases with p53-R175H, there was a significant increase in CSF-1 compared with cases with wild-type p53 or cases with non-hotspot missense and frameshift p53 mutations, suggesting a robust link between the p53-R175H mutation and CSF-1 expression (Fig. 6D).

We further studied the clinical outcomes of ESCC cases in the context of tumor CSF-1 expression. In our ESCC TMA cohort, poorly differentiated ESCC had significantly higher CSF-1 expression than well-differentiated ESCC (Fig. 6E). Remarkably, stage IV tumors had higher CSF-1 expression than stage I tumors (Fig. 6F). Analysis of the TCGA-ESCC dataset reinforced that the tumors in later stages (IIIa/b/c) had elevated CSF-1 mRNA expression compared with stage Ia/b tumors (Supplementary Fig. S5B). In addition, Kaplan–Meier survival analysis of the TCGA-ESCC cohort showed that the overall survival of CSF-1-high ESCC cases was significantly lower (Fig. 6G), which was supported by the analyses of the correlation between CSF-1 score and probability of overall survival in the ESCC TMAs (Supplementary Fig. S5C).

We next sought to test the therapeutic efficacy of inhibiting CSF-1R in vivo with a neutralizing antibody (clone AFS98) for its potential to reduce lung metastasis. We injected Trp53^R172H/− cells into athymic nude mice via tail-vein, and on day 7 after injection, we began treatment with either the CSF-1R neutralizing antibody or an equivalent concentration of IgG controls every other day for 8 weeks (Fig. 6H). At the predetermined final time point (day 56), we found that the metastatic burden in the lungs was significantly decreased in animals treated with CSF-1R neutralizing antibody (Fig. 6I).

Notably, analysis of TCGA data demonstrated a positive correlation between CSF-1 and CSF-1R expression and leukocyte infiltration, with the major component being macrophages (51, 52) in ESCC cases (Supplementary Fig. S5D and S5E). Therefore, we investigated how CSF-1 might modulate the tumor microenvironment. We observed a marked increase in the migration of Raw264.7 macrophages when they were cultured with Trp53^R172H/− tumor cells compared with Trp53^−/- or Trp53^+/+ cells (Fig. 6J). The number of migrated macro­phages correlated with CSF-1 levels in conditioned media from these cocultures: The CSF-1 concentration was approximately 350 pg/mL upon coculturing macrophages with Trp53^R172H/- tumor cells, which was ∼2- and ∼20-fold higher than macro­phage cocultures with Trp53^−/- and Trp53^+/+ cells, respectively (Fig. 6K).

To elucidate the changes in tumor-infiltrating macro­phages at the metastatic site, we conducted Vectra Multiplex Immunofluorescence (IF) in the lung tissues from Fig. 6L for total macrophages (F4/80), M2 macrophages (Cd163 and Cd206), and tumors (YFP). We observed reduced infiltration of Cd163⁺Cd206⁺ M2-polarized F4/80⁺ macrophages in the YFP⁺ metastatic tumors from animals treated with the CSF-1R neutralizing antibody (Fig. 6L). These findings support the role of CSF-1/CSF-1R signaling in the cross-talk between tumor cells and macrophages, in addition to the tumor cell-intrinsic properties. Overall, our findings underscore the therapeutic potential of targeting CSF-1R in metastatic ESCC in future clinical trials.

We have demonstrated that Trp53^R172H-dependent Csf-1 expression as an effector in mediating ESCC cell invasion and lung metastasis. Increased CSF-1 secretion in primary and metastatic tumor cells depends upon Trp53^R172H/−, but not on wild-type p53 or the absence of p53 (p53-null). We propose that p53-R172H regulates the transcription of Csf-1 as it interacts with the Csf-1 promoter.

Growing evidence has demonstrated the protumorigenic impact of missense p53 mutations in various cancer types (9). In many cases, such oncogenic activities are obtained in partnership with other transcription factors and chromatin-modifying proteins (11). In support of this notion, we discovered that Brd4, which binds to acetylated histone lysine residues (27), interacts with p53-R172H at the Csf-1 locus and is a coregulator in mediating the Csf-1 transcription and promoting metastasis.

Although our study demonstrates that p53-R172H binds to the promoter of Csf-1 to induce its transcription, it is likely that other proteins might be involved in this mechanism. We identified previously that YAP/TAZ signaling is a regulator of p53-R172H–mediated lung metastasis of ESCC (34). Interestingly, YAP1 has been reported to physically interact with BRD4 to mediate transcriptional activation, in which tumor cells depend upon transcriptional regulators to support their prooncogenic needs (53). Additionally, it has been suggested that nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) can recruit mutant p53, which in turn recruits BRD4 to active enhancers and trigger transcription of proinflammatory genes (54). Thus, NF-κB might play a role in activating p53-R172H-dependent CSF-1/CSF-1R signaling and promote lung metastasis in ESCC.

CSF-1R is expressed in epithelial cells, activating downstream pathways such as phosphatidylinositol 3-kinase (PI3K) and RAS/RAF/MAP and SRC family kinases (38, 55–57). The cell-autonomous mechanism we propose here suggests that Trp53^R172H/−-dependent CSF-1/CSF-1R signaling promotes the activation of STAT3 and EMT in metastatic lung lesions and is reinforced by TCGA analysis (Fig. 3C and D; Supplementary Fig. S3A and S3B). Our studies suggest that loss of Csf-1 maintains cells primarily in an epithelial state. This is consistent with our previous findings on metastatic organotropism, demonstrating a selective pressure for tumor cells with low E-Cad levels to metastasize to the lungs (58).

In a recent study, Yellapu and colleagues (59) analyzed the RNA-seq data of triple-negative breast cancer (TNBC) cell lines, including MDA-MB-231, HCC-1806, and SUM-159, which were cotreated with JQ1 and the bromodomain BAZ2A/B-specific inhibitor GSK2801 to evaluate the alterations in gene expression. They identified TNF, JAK–STAT, IL17, and NF-κB pathways as being downregulated upon cotreatment with JQ1 and GSK2801 (59). One of the genes that was enriched in these pathways was vascular endothelial growth factor A (VEGFA; ref. 59), which has been shown to be increased by p53-R175H and p53-R273H upon recruitment of long noncoding RNA (lncRNA) metastasis-associated lung adenocarcinoma transcript 1 (MALAT1) in breast cancers (60). Furthermore, platelet-derived growth factor receptor β (PDGFRβ) was enriched in these downregulated pathways (59), which has been demonstrated to be modulated by p53-R175H, and enhance invasion and lung metastasis in pancreatic ductal adenocarcinoma (PDAC) (61). These pathways point to putative gene targets of mutant p53-BRD4 coregulation that might contribute to ESCC-mediated lung metastasis.

It is important to note that we do not dismiss the contribution of CSF-1/CSF-1R–mediated paracrine signaling during lung metastasis. Both in silico TCGA analysis and our in vivo studies with the characterization of macrophage subpopulations upon blocking CSF-1R support the non–cell-autonomous functions of CSF-1/CSF-1R signaling. This suggests a choreographed interplay between cell-autonomous and non–cell-autonomous effects of CSF-1.

Various strategies to target p53 have been proposed, such as reestablishing wild-type p53 functions, destabilizing mutant versions of p53, inhibiting negative regulators of the wild-type p53, MDM2, and MDMX to activate the tumor-suppressive function of wild-type p53, and disrupting its interaction with other proteins (9, 62). However, most strategies to date have been ineffective. Therefore, focusing on the downstream effectors of mutant p53 has merit. We demonstrated a marked reduction in lung metastasis of Trp53^R172H/− ESCC cells upon treatment with a neutralizing antibody targeting CSF-1R. To that end, inhibiting the CSF-1/CSF-1R signaling pathway may open up new therapeutic approaches for patients with metastatic ESCC and even other SCCs, given the high frequency of p53-R175H mutations in these SCCs (30, 50). Treating tumor cells with a PROTAC that degrades BRD4 reduces invasion and CSF-1 secretion and treating mice with lung metastases with JQ1 decreases not only lung metastases but also CSF-1 levels in circulation. Therefore, CSF-1 may represent a possible biomarker for metastatic ESCC and help to stratify patients for CSF-1R therapy.

In summary, we propose p53-R172H/BRD4 coregulation of Csf-1 transcriptional expression and signaling through CSF-1R as a mechanism contributing to ESCC invasion and lung metastasis (Supplementary Fig. S6). Based on our results that are substantiated through multiple complementary genetic, pharmacologic, and in vivo approaches, we nominate this pathway as a potential therapeutic avenue for metastatic ESCC patients who intrinsically harbor a dismal prognosis.

Primary and metastatic ESCC cells isolated from the mouse models were cultured in keratinocyte serum-free media supplemented with 50 μg/mL bovine pituitary extract, 5 ng/mL human recombinant EGF, 0.018 mmol/L CaCl₂, and 1% penicillin/streptomycin (KSFM; Gibco, 37010022). For drug treatments, cells were treated with 10 μg/mL CSF-1R neutralizing antibody (InVivo­MAb anti-mouse CSF-1R; clone AFS98; Bio X Cell, BE0213; RRID: AB_2687699) or IgG control (IgG2a isotype control; Bio X Cell, BE0089; RRID: AB_1107769) for 24 hours. To target BRD4, the cells were treated with 1 μmol/L ZXH-3-26 (MedChemExpress, HY-122826), 250 nmol/L JQ1 (Selleckchem, S7110) or an equal volume of DMSO for 24 hours.

LS123 human colon cancer cells were cultured in Eagle's minimum essential medium (EMEM; ATCC, 30-2003; RRID: CVCL_1383) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. TE11 human ESCC cells (RRID: CVCL_TE11) were cultured in RPMI-1640 (Gibco, 11875093) supplemented with 10% FBS and 1% penicillin/streptomycin. All mouse and human cell lines were tested for Mycoplasma frequently with MycoAlert Mycoplasma Detection Assays (Lonza, LT07-318) and the short tandem repeat profiling was conducted on human cell lines through ATCC (ATCC-135-XV).

For the deletion of Csf-1, the ESCC cell lines were nucleofected with 2.5 μg of Spyfi Cas9 (Aldevron, 9214) and two pairs of guide RNAs (Synthego; Supplementary Table S1) using a RNP approach. The Amaxa Lonza 4D Nucleofector X-Unit (Lonza, AAF-1003X) and P3 Primary Cell kit (V4XP-3024) protocol CB-150 were used for the nucleofections. Forty-eight hours after nucleofection, the mixed population was sorted into 96-well plates as single cells and incubated for 3 weeks before the colonies were expanded and genotyped for the assessment of indels introduced. CSF-1 expression and secretion in the colonies with homozygous indels were verified by qRT-PCR and ELISA.

Deletion of TP53 in LS123 or TE11 cells was achieved by infecting pLentiCRISPR v2 (GenScript) with human TP53 single-guide RNA (Supplementary Table S1) or an empty control vector. The cells were sorted 48 hours after the infection into single-cell colonies in 96-well plates and treated with 1.5 μg/mL or 2.0 μg/mL Puromycin for 2 weeks, respectively. Viable colonies were expanded and the p53 levels were verified by immunoblotting.

The Csf-1 (Myc-DDK-tagged) cDNA (Origene, MR226827) was cloned into the pLenti-C-mRFP-P2A-Puro-Lenti vector (Origene, PS100094) at the SgfI-MluI restriction enzyme sites. The plasmid was amplified and verified using the forward primer 5″-GGACTTTCCAAAATGTCG-3″. The cells were infected with either 1 μg of the Csf-1 overexpression plasmid or an empty control vector. After infection, the cells were treated with 2 μg/mL Puromycin for 2 weeks. The Csf-1 expression was verified by qRT-PCR and ELISA.

To achieve Brd4 knockdown with shRNA, cells were transduced with pGFP-C-shLenti lentiviral particles with NT, shBrd4 A, and shBrd4 B (Origene, TR30023; Supplementary Table S1). After 48 hours, stable colonies were selected using 2 μg/mL Puromycin (Sigma-Aldrich). qRT-PCR and Western blotting were performed to confirm knockdown efficiency.

For Brd4 knockdown with siRNA, Accell SMARTPool targeting mouse Brd4 (Horizon Discovery; Supplementary Table S1) was used according to the manufacturer's instructions. Briefly, the cells were treated with 1 μmol/L siBrd4 or NT control (Horizon Discovery; Supplementary Table S1) in siRNA Delivery Media (Horizon Discovery, B-005000) and incubated for 72 hours at 37°C with 5% CO₂. qRT-PCR and Western blotting were performed to confirm knockdown efficiency after 96 hours.

L2-Cre; LSL-Trp53^+/+, Trp53^−/−, or Trp53^R172H/−; Rosa26^{lox-Stop-lox-YFP} mice were generated and treated with 4-NQO as described previously (32). All animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) at Columbia University, and all experiments were conducted in compliance with the NIH guidelines for animal research.

We used lateral tail-vein injections to model lung metastasis as described previously (34). Briefly, 1.0 × 10⁶ dissociated single cells were resuspended in 100 μL of cold phosphate-buffered saline (PBS) and injected into the distal end of the lateral tail vein of 7-week-old athymic nude mice (CrTac:NCr-Foxn1^nu, Taconic Bioscience; RRID: IMSR_TAC:NCRNU). Throughout the experiments, mice were monitored daily for signs of pain, discomfort, and/or weight loss. To assess metastatic burden, mice were euthanized at 8 weeks by CO₂ and cervical dislocation. The lungs were harvested and flushed with 20 mL of Heparin-PBS to reduce red blood cells, imaged with Keyence BZ-X800 (Keyence) microscope, and the metastatic burden was quantified as the number of YFP⁺ metastatic foci. CSF-1R neutralizing antibody (150 μg; InVivoMAb anti-mouse CSF-1R; clone AFS98; Bio X Cell, BE0213; RRID: AB_2687699) or equivalent IgG (IgG2a isotype control; Bio X Cell, BE0089; RRID: AB_1107769) was injected intraperitoneally between days 7 and 53 after the initial tail-vein injections. 100 mg JQ1 (Selleckchem, S7110) was dissolved in 1 mL of DMSO and then added to 19 mL of 20% captisol (20% SBE-β-CD in saline) to yield 5 mg/mL. 50 mg/kg of JQ1 or an equivalent volume of vehicle control was injected intraperitoneally every other day between days 7 and 53 after the initial tail-vein injections.

For subcutaneous injections, after anesthesia, 3.0 × 10⁶ tumor cells resuspended in 100 μL of 1:1 cold Matrigel:PBS were injected into the right and left dorsal flanks of the athymic nude mice (CrTac:NCr-Foxn1^nu, Taconic Bioscience; RRID: IMSR_TAC:NCRNU). One week after the subcutaneous transplantation, tumor growth was measured every other day. On day 28, mice were sacrificed, tumors were harvested, and their weights were measured, and harvested tumors were analyzed by histology and IHC.

The Transwell invasion assay was conducted using the Corning BioCoat Matrigel invasion chamber (Corning, 354481). The 1.0 × 10⁵ cells resuspended in 500 μL of serum-free KSFM were seeded in the upper chamber, and 500 μL of standard KSFM was added under the chamber. After 24-hour incubation at 37°C, the invading cells were fixed with 70% ethanol for 15 minutes and stained with 2% Crystal Violet for 10 minutes at room temperature. The number of invading cells was counted using the Keyence BZ-X800 (Keyence) microscope.

RNA samples were extracted using the QIAGEN Mini RNeasy kit (#74104) according to the manufacturer's instructions. cDNA from 1 μg of RNA was synthesized with iScript Reverse Transcription Supermix (Bio-Rad, 1708841). qRT-PCR was performed using the StepOnePlus real-time PCR system (Applied Biosystems) and TaqMan Universal PCR Master Mix (Applied Biosystems, 4304437) and TaqMan Gene-Expression Assays with the best coverage as listed in Supplementary Table S1. Relative expression was analyzed using the comparative Ct method (ΔΔCt) and compared with PPIA, a control endogenous gene.

The immunoblotting was performed as described previously (63). Briefly, cell pellets were resuspended in a 50–150 μL of cell lysis buffer [20 mmol/L Tris–HCl (pH 7.5), 150 mmol/L NaCl, 1 mmol/L Na₂EDTA, 1 mmol/L EGTA, 1% (vol/vol) Triton X-100, 2.5 mmol/L sodium pyrophosphate, 1 mmol/L β-glycerophosphate, 1 mmol/L Na₃VO₄, and 1× complete EDTA-free protease inhibitor cocktail (Roche)]. Protein lysates were quantified with Bio-Rad protein assay, and 0.5–2 μg/μL solutions were prepared using 4× Laemmli sample buffer (Bio-Rad). 10–30 μg of samples were loaded into 4% to 15% mini-protean TGX precast gels (Bio-Rad) and transferred to 0.2-μm nitrocellulose membrane with Trans-blot turbo mini transfer packs (Bio-Rad). After blocking in 5% milk in PBS for 60 minutes, membranes were incubated with the primary antibody for 16 hours at 4°C, followed by incubation with a secondary antibody for 60 minutes at room temperature (Supplementary Table S2). Antibodies were diluted in 5% bovine serum albumin (BSA) in PBS + 0.1% tween20. The membranes were imaged using the Chemidoc imaging system (Bio-Rad).

All the tissues were fixed in buffered zinc formalin (Thermo Fisher, 22-050-259) for 18 hours at 4°C, washed in PBS for 30 minutes, and placed in 70% ethanol. Antigen retrieval was performed on dewaxed and rehydrated FFPE sections using 10 mmol/L boiling citric acid buffer (pH = 6; IHC) or R-Buffer A (Fisher Scientific, 50311711; immunofluorescence) in a pressure cooker. For IHC staining, endogenous peroxidases were quenched with 3% H₂O₂ (Sigma-Aldrich, H1009), followed by blocking with Avidin (Sigma-Aldrich, A9275), Biotin (Sigma-Aldrich, B4501), and Starting Block T20 buffer (Thermo Fisher, 37543) for 15 each at room temperature. The tissue sections were incubated with primary antibodies (Supplementary Table S2) for approximately 16 hours at 4°C, and followed by incubation with biotinylated secondary antibodies (Vector Laboratories; Supplementary Table S2) for 30 minutes at 37°C. All antibodies were diluted in PBS + 1% BSA + 0.3% TritonX buffer. Slides were then incubated with ABC reagent (Vector Laboratories, PK-6100) for 30 minutes at 37°C, treated with a DAB substrate kit (Vector Laboratories, SK4100), counterstained with hematoxylin, dehydrated, and mounted for imaging.

For immunofluorescence staining, FFPE tissue sections or cells that were fixed with 4% PFA and permeabilized with 0.1% TritonX were blocked with PBS + 5% donkey serum + 0.3% TritonX blocking buffer for 1 hour at room temperature. All the primary antibodies were diluted in the same blocking buffer and incubated for approximately 16 hours at 4°C, followed by incubation with Alexa Fluor antibodies for 1 hour at room temperature (Supplementary Table S2). Brightfield and fluorescent images were taken with either Keyence BZ-X800 (Keyence) or Leica AT2 whole-slide digital imaging (Leica).

For CSF-1R flow cytometry, the mouse ESCC and Raw264.7 cells were dissociated into single cells with Hank's Enzyme Free Cell Dissociation Solution (EMD Millipore, S-004-C) and filtered through a 100-μm strainer. Cells were stained with anti–CSF-1R or isotype control (Supplementary Table S2) in a staining solution at 4°C for 30 minutes. Cells were washed three times in staining solution and then stained with SYTOX Red (Thermo Fisher, S34859) for flow-cytometric analysis.

On FFPE lung tissue sections, Vectra-6plex multiplex immunofluorescence was conducted with antibodies with specific Opal fluorophores directed against YFP, F4/80, Cd163, and Cd206 for metastatic tumor cells, total and M2-polarized macrophages (antibodies listed in Supplementary Table S2). The stained sections were imaged using a multispectral camera (VECTRA Quantitative Pathology Workstation), followed by signal unmixing, segmentation, and single-cell quantification using QuPath analysis software (RRID: SCR_018257).

PLA assays were conducted using the Duolink In Situ Red Starter kit (Sigma-Aldrich, DUO92101), according to the manufacturer's instructions. The primary antibodies and dilutions used are listed in Supplementary Table S2. ImageJ software (RRID:SCR_003070) was used to quantify PLA signal per nuclei.

The metastatic ESCC cells were grown in monoculture for 48 hours and their conditioned media were collected after centrifuging for 10 minutes at 4°C and 2,000 rpm. The relative concentrations of 62 cytokines were measured using the mouse cytokine array C3 (Raybiotech, AAM-CYT-3-2).

The standard culture medium was replaced by a serum-free medium 24 hours after the initial seeding of the cells. The mono- and cocultures were incubated at 37°C for 48 hours, and the conditioned medium was collected by centrifugation for 10 minutes at 2,000 rpm at 4°C. Secreted CSF-1 levels were measured using the Mouse M-CSF Quantikine ELISA Kit (R&D Systems, MMC00B), according to the manufacturer's instructions. The absorbance was read at 450 nm with a correction wavelength of 540 nm, and the final absorbance was calculated by subtracting the 540 nm correction from the 450 nm readout.

3.0 × 10⁷ primary and metastatic ESCC cells were collected and washed once in PBS and then crosslinked in 10 mL of 1% formaldehyde in PBS solution (Thermo Fisher, 28908) for 10 minutes at room temperature. The reaction was stopped by adding 0.5 mL glycine at a final concentration of 125 mmol/L for 5 minutes at room temperature. Chromatin immunoprecipitation, including cell lysis, sonication, immunoprecipitation, and purification, was conducted using an anti-p53 antibody (CM5, Supplementary Table S2) as previously described (64). The ChIP products were quantified using the StepOnePlus real-time PCR system (Applied Biosystems) using the reaction mix with the Power SYBR Green PCR Master Mix (Applied Biosystems, 4367659) and primers as listed in Supplementary Table S1. The values were normalized to the amount of DNA detected in the input samples by qRT-PCR.

Well-annotated TMAs representing surgically removed paired primary ESCC tumors and adjacent normal mucosa were obtained from institutional review board (IRB)–approved and deidentified therapy-naive patients (n = 148), as described previously (65, 66).

TCGA was accessed for the analysis of survival, tumor staging, gene expression, and mutation data in ESCC and SCC cases at https://gdc.cancer.gov/about-data/publications/pancanatlas. SCC tumors are listed in (30) and separate esophageal adenocarcinoma (EAC) from ESCC cases. The analysis of leukocyte infiltration is described previously (51).

CUT&RUN-seq sample preparation was performed following a published protocol (67). Briefly, 3.0 × 10⁵ ESCC tumor cells per condition were harvested and lightly crosslinked with 0.1% formaldehyde in PBS solution (Thermo Fisher, 28908) for 1 minute at room temperature. Cells bound to activated Biomag Plus Concavalin A beads (Bangs Laboratories, BP531) were incubated with primary antibodies for IgG or H3K27ac on a nutator overnight at 4°C (Supplementary Table S2). After binding pAG-MNase (EpiCypher, 15-1016), the targeted chromatin was digested with 100 mmol/L CaCl₂ for 30 minutes at 0°C, and the DNA was isolated using phenol-chloroform extraction. E. coli spike-in DNA (EpiCypher, 18-1401) was used for NGS normalization. The library preparation on the purified DNA was conducted with the NEBNext Ultra II DNA Library Prep Kit for Illumina (BioLabs, E7103) following the manufacturer's instructions.

Reads were aligned to the mouse reference genome mm10 using Bowtie v2.2.8 (RRID: SCR_005476) with parameters -q -I 50 -X 700 –very-sensitive-local –local –no-mixed –no-unal –no-discordant, and reads quality was assessed using fastQC. Statistics of aligned reads for each sample are provided in Supplementary Table S3. Binary alignment map (BAM) files were generated with samtools v1.9 (RRID: SCR_002105) and were used in downstream analysis. MACS2 v2.1.0 was used to call significant peaks. Peaks within ENCODE blacklisted regions were removed. Coverage tracks were generated from BAM files using deepTools (68), bamCoverage with parameters –binsize 1 and scaled with DEseq2 scaling factors. The BAM files were merged into a single BAM and significant peaks were called using MACS2 narrowPeak 2.1.0 to generate a bed file with the universe of regions. featureCounts from subread v2.0.1 (RRID: SCR_009803) was used to report the count of alignments from each sample's BAMs at the universe of regions and generate a counts matrix. Next, the R package DEseq2 v1.34 (RRID: SCR_015687) was used to identify differentially enriched peaks and generate PCA and MA plots. Differential regions were called if the adjusted P value was less than 0.05. Heat maps of genomic regions were generated with deepTools 3.2.1 (RRID: SCR_016366). The command computeMatrix was used to calculate scores at genomic regions and to generate a matrix file for use with plotHeatmap to generate plots.

All pathologic analyses were performed by Dr. Andres J. Klein-Szanto (Histopathology Facility, Fox Chase Cancer Center) in accordance with the consensus report and recommendations for pathologic analysis. The quantitative evaluation of positive cells (Ki-67, p53, CSF-1, CSF-1R) or nuclei (STAT3), as well as the score of elongated tumor cells, was performed by counting manually 300 to 500 cells per case on every slide in a blinded fashion. For IHC analysis, 0 indicated negative staining; 1, weak staining; 2, moderate staining; and 3, strong staining. The H-score was calculated as [3 × (% strong staining)] + [2 × (% moderate staining)] + [1 × (% weak staining)], yielding a range of 0–300.

STAR 2.7.10 was used to align trimmed fastq files to the reference genome (mm10), and Picard CollectRnaSeqMetrics module (Picard) was used to evaluate alignments and perform initial QC. Gene expression was quantified with HTSeq, and these data were imported into Rstudio (R 3.5) and used as an input file for DeSeq2 (RRID: SCR_015687) analysis to determine differentially expressed genes across different conditions. Differentially expressed genes were used as input for the Volcano plot and GSEA.

All data are presented as the mean ± standard error mean (SEM), and sample size and replicates are indicated in the text and figure legends. Statistical significance was set at P < 0.05. The statistical analyses, including Student unpaired t test, one-way ANOVA, and Mann–Whitney U test were performed using GraphPad Prism 9.0 (GraphPad Software; RRID: SCR_002798), and the false discovery rate method was used for the adjustment of P values (P_adj). For the correlation analyses, Spearman correlation coefficients were used. Power analysis was conducted with our Biostatistics Shared Resources.

CUT&RUN-seq raw and processed data reported in this article have been deposited at the NCBI Gene-Expression Omnibus (GEO; RRID:S CR_005012) with accession number GSE232757 and are publicly available as the date of publication. This paper also analyzes existing, publicly available data from TCGA and can be accessed from The Genomic Data Common, https://gdc.cancer.gov/about-data/publications/pancanatlas. This paper does not report original code. The scripts and parameters of each step can be provided upon request to the authors.

A.M. Taylor reports grants from Ono Pharmaceutical outside the submitted work. J.J. Manfredi reports grants from the NCI during the conduct of the study. No disclosures were reported by the other authors.

G. Efe: Conceptualization, resources, data curation, software, formal analysis, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. K.J. Dunbar: Data curation, formal analysis, validation, investigation, writing–review and editing. K. Sugiura: Data curation, validation, methodology. K. Cunningham: Data curation, formal analysis, validation, investigation. S. Carcamo: Data curation, software, formal analysis, writing–review and editing. S. Karaiskos: Data curation, software, formal analysis, writing–review and editing. Q. Tang: Resources, data curation. R. Cruz-Acuña: Data curation, visualization. L. Resnick-Silverman: Resources. J. Peura: Data curation, writing–review and editing. C. Lu: Conceptualization, writing–review and editing. D. Hasson: Conceptualization, software, writing–review and editing. A.J. Klein-Szanto: Data curation, formal analysis, investigation, writing–review and editing. A.M. Taylor: Resources, data curation, software, formal analysis, writing–review and editing. J.J. Manfredi: Conceptualization, resources, writing–review and editing. C. Prives: Conceptualization, resources, writing–review and editing. A.K. Rustgi: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This work was funded by NIH/NCI grants P01-CA098101 (A.K. Rustgi, G. Efe, K.J. Dunbar, K.M. Cunningham, and R. Cruz-Acuña), P30CA013696 (A.K. Rustgi), and R01CA272903-01 (A.K. Rustgi), the American Cancer Society Research Professorship (A.K. Rustgi), the NIH/NCI fellowship F31CA275369-02 (G. Efe), the NIH/NCI grant R35CA220526 (C. Prives), and Office of Research Infrastructure of the NIH under award number S10OD026880 (D. Hasson and S. Carcamo). We thank the Confocal and Special Microscopy, Molecular Pathology, Human Immune Monitoring and Biostatistics Shared Resources at the Herbert Irving Comprehensive Cancer Center (HICCC), and Flow Cytometry Core at the Columbia Stem Cell Initiative (CSCI) of Columbia University Irving Medical Center. We also thank Theresa Swayne (Columbia University) for technical assistance and input on confocal imaging and quantification of multiplex immunofluorescence. This work was supported in part by the Bioinformatics for Next-Generation Sequencing (BiNGS) shared resource facility within the Tisch Cancer Institute at the Icahn School of Medicine at Mount Sinai, which is partially funded by the NIH/NCI grant P30CA196521. This work was also supported in part through the computational resources and staff expertise provided by Scientific Computing at the Icahn School of Medicine at Mount Sinai and supported by the Clinical and Translational Science Awards (CTSA) grant UL1TR004419 from the National Center for Advancing Translational Sciences (NCATS).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.