KLF5 regulates actin remodeling to enhance the metastasis of nasopharyngeal carcinoma

Transcription factors (TFs) engage in various cellular essential processes including differentiation, growth and migration. However, the master TF involved in distant metastasis of nasopharyngeal carcinoma (NPC) remains largely unclear. Here we show that KLF5 regulates actin remodeling to enhance NPC metastasis. We analyzed the msVIPER algorithm-generated transcriptional regulatory networks and identi�ed KLF5 as a master TF of metastatic NPC linked to poor clinical outcomes. KLF5 regulates actin remodeling and lamellipodia formation to promote the metastasis of NPC cells in vitro and in vivo. Mechanistically, KLF5 preferentially occupies distal enhancer regions of ACTN4 to activate its transcription, whereby decoding the informative DNA sequences. ACTN4, extensively localized within actin cytoskeleton, facilitates dense and branched actin networks and lamellipodia formation at the cell leading edge, empowering cells to migrate faster. Collectively, our �ndings reveal that KLF5 controls robust transcription program of ACTN4 to modulate actin remodeling and augment cell motility which enhances NPC metastasis, and provide new potential biomarkers and therapeutic interventions for NPC.


Introduction
Metastasis, the heterogeneous and systemic disease, compromises the function of distant organs colonized by aggressive cancer cells, causing disruption of homeostasis and eventually death [1].Cancer metastasis is a cell motility-based multistep pathological cascade during which cancer cells disseminate from the primary tumor, hijack tumor microenvironment, intravasate and extravasate from vessels, enter and exit dormancy, accommodate to various tissue conditions and colonize distant organs [1,2].During metastasis, orchestration of actin cytoskeleton dynamics renders cancer cells sensing and responding to physical stimuli to augment cell motility and dissemination [3][4][5][6].Cells precisely reshape by actin remodeling to resist uid shear stress in bodily circulatory systems, survive and seed metastatic lesions [3,5].
As the basis of metastasis, cell migration encompasses a series of cellular events.After integrating and transmitting cell-intrinsic stimuli and environment-generated cues, metastatic cells initiate the polarization of actin cytoskeletal machinery at the cell front and subsequently form protrusions [6][7][8][9].
Protrusions at the leading edge driven by rearranged dense actin networks generate asymmetric or polarized membrane tension to direct cell migration [7,10,11].Intriguingly, actin-rich nascent membrane protrusions extend from areas with low density of membrane-proximal F-actin [12].Lamellipodia, a atter type of protrusion with fan-like architecture, is dominated by Arp2/3-branched actin, attributed to activation of nucleation-promoting factors by Rho-family GTPases [13,14].Another mechanism of actindriven migration shows that, the orchestrated retrograde actin ow creates retrograde shear forces, propelling cells to migrate following the environmental topography [9].
In response to promigratory signals, rapid transcription modulation by transcription factors (TFs) during cell migration facilitates adaption of cells facing changing extracellular milieu [15].Human Krüppel-like factors (KLFs) are a family of 17 transcription factors interacting with DNA through conserved triple C2H2 zinc ngers (ZnFs) domain [16].KLFs control various key cellular processes including differentiation, in ammation and migration, with exclusive and redundant functions of each KLF [16][17][18][19].In cancers, derailed expression and function of KLFs under divergent contexts implicate cancer pathogenesis, heterogeneity, therapeutic resistance, recurrence and metastasis [16,20].A recent study revealed the KLF4-mediated exquisite transcription modulation model involving enhancers [21].In the reprogramming of mouse embryonic broblasts, KLF4 reorganizes the chromatin structure and rewires three-dimensional enhancer loops, leading to adaptive transcriptional changes [21].KLFs cross-regulate and co-opt in cancer development and malignant progression, albeit their intricate regulatory networks and effects remain to be fully determined.
Nasopharyngeal carcinoma (NPC) is a nasopharynx epithelium originated carcinoma with a high metastatic proclivity and special geographic distribution that in 2020, > 75% new cases occurred in Eastern Asia and South-Eastern Asia, especially in China [22,23].After rst-line induction chemotherapy plus concurrent chemoradiotherapy treatment, approximately 25% patients still developed recurrence or distant metastasis [24].For recurrent or metastatic NPC patients receiving rst-line treatment, the median progression-free survival of 7.6 months was suboptimal [25].However, the master TF involved in distant metastasis of NPC remains elusive.
In this study, we discover that KLF5 operates as a master TF and preferentially occupies distal enhancer regions to regulate transcription output in metastatic NPC.We further show that KLF5 transcriptionally activates ACTN4 by binding to its conserved enhancer to facilitate actin remodeling, extending of actinrich lamellipodia and distant metastasis of NPC.

KLF5 is a master TF of metastatic NPC
To investigate the master TF engaged in distant metastasis of NPC, we rst leveraged the Algorithm for the Reconstruction of Accurate Cellular Networks through Adaptive Partitioning (ARACNe-AP) to build transcriptional interactomes [26] with a multitude of TF-target interplay derived from gene expression pro le of curated datasets (Fig. 1A).Based on the interactomes across 1,639 known or likely TFs [27], we generated the intricate TFs regulatory networks by Virtual Inference of Protein-activity by Enriched Regulon analysis based on multiple samples (msVIPER) algorithm to assess transcriptional activity of master TFs [28] and identi ed 192 candidates for distant metastasis of NPC (Fig. 1A and Supplementary Table S1, 2).We delightedly found that three KLFs (KLF5, KLF7 and KLF3) in 192 active TFs presumably promotes NPC metastasis (Supplementary Fig. S1A).Compared to other KLFs, the expression of KLF5 ranked top in NPC samples and those with distant metastasis (Fig. 1B, C).Given its highest activity and expression among three active KLFs in metastatic samples of NPC (Supplementary Fig. S1A-E), we speculated that KLF5 operates as a master TF and promotes distant metastasis of NPC.
We next analyzed two public datasets (GSE103611 and GSE13597) and identi ed KLF5 high and KLF5 low human NPC groups, and KLF5 high group showed a larger proportion of patients with distant metastasis and stage III-IV disease (Fig. 1D-G).Gene set enrichment analysis (GSEA) of the transcriptome of these two groups revealed a pronounced enrichment of the metastasis relevant gene set in KLF5 high NPC group (Fig. 1H).These results indicate that NPC patients with higher expression and transcriptional activity of KLF5 are prone to develop distant metastasis and progress to an advanced stage.

KLF5 enhances metastatic potential of NPC cells in vitro and in vivo
We compared the expression level of KLF5 in immortalized N2-Tert nasopharyngeal mucosa cells and 6 NPC cell lines.The results showed that KLF5 expressed higher in NPC cells and samples than normal control (Fig. 1I, J and Supplementary Fig. S2A).Since KLF5 accelerated cell proliferation and malignant progression of pancreatic ductal adenocarcinoma [29], short hairpin RNA (shRNA)-mediated knockdown (KD) of KLF5 was used to investigate the role of KLF5 in the carcinogenesis of hyperproliferative NPC.Cell viability and colony formation results showed that, knockdown of KLF5 did not alter the proliferation kinetics of NPC cells (Fig. 2A and Supplementary Fig. S2B-E).Therefore, we focused on the role of KLF5 in distant metastasis of NPC and the underlying mechanism.Compared with control cells, knockdown of KLF5 triggered an indolent metastatic phenotype that suppressing cell migration and invasion (Fig. 2B and Supplementary Fig. S2F).Moreover, depletion of KLF5 signi cantly increased the time required for wound closure on 2D surfaces (Fig. 2C and Supplementary Fig. S2G, H).Accordingly, overexpression of KLF5 potentiated the motility of NPC cells in vitro (Fig. 2D, E and Supplementary Fig. S2I, J).We further constructed the lung metastatic model and inguinal lymph node metastatic model wherein KLF5-KD or control NPC cells were injected into the tail vein or footpad of immunocompromised mice, respectively.
Compared to the control, mice intravenously injected with KLF5-KD NPC cells shouldered less lung metastatic burden showing reduction of overt metastatic lesions (Fig. 2F).Consistently, knockdown of KLF5 precluded metastatic dissemination of NPC cells from primary tumor in footpad to groin, leading to smaller tumor-draining lymph nodes (Fig. 2H).Knockdown of KLF5 constrained NPC cell in ltration of normal tissues including skin, muscle and lung in metastatic models (Fig. 2G, I).Collectively, our data suggests that KLF5 enhances migratory propensity of cells in vitro and metastatic outbreak in vivo.

KLF5 regulates actin remodeling and lamellipodia formation
Next, we sought to identify the molecular features and pathways determining KLF5-enhanced metastatic potential of NPC.Kyoto Encyclopedia of Genes and Genomes (KEGG), Gene Ontology (GO) and Reactome pathway enrichment analysis of KLF5 high and KLF5 low human NPCs in different datasets revealed that the actin cytoskeleton reorganization, Rho-family GTPases and cell motility associated pathways are enriched in KLF5 high group (Fig. 3A, B).We applied GESA analysis to compare the transcriptome of these two groups, and observed a pronounced enrichment of gene sets germane to actin cytoskeleton, wound healing, Rho-family GTPases and EMT in KLF5 high group (Fig. 3C, D and Supplementary Fig. S3A-C).
Diverse guidance cues converge into asymmetric or polarized traction force generation to direct cell migration, with leading-edge protrusions acting as one pivotal executor [4].Most protrusions are driven by the continuous polymerization and depolymerization of branched and linear arrays of actin, of which lamellipodia appears atter and is dominated by Arp2/3-branched actin [4,14,30].To evaluate whether KLF5 augments cell motility by tuning actin remodeling and lamellipodia formation, we rst implemented RNA-sequencing (RNA-seq) analysis of KLF5-KD and control NPC cells.By KEGG, GO and GSEA analysis on the transcriptome of KLF5-KD and control cells, we identi ed enriched functional categories of actin cytoskeleton, focal adhesion, wound healing and cell migration regulation (Fig. 3E, F and Supplementary Fig. S3D-F).Second, we conducted knockdown or overexpression of KLF5 in another individual assay to verify the transcriptomic changes occurring in KLF5-depleted metastatic cells.Knockdown of KLF5 inhibited expression of many key genes in the pathway of actin cytoskeleton regulation, whereas overexpression of KLF5 elicited contrast effects (Fig. 3G, H and Supplementary Fig. S3G, H).Aligned with in silico analysis, depletion of KLF5 interrupted the dynamics of actin networks and induced unstructured actin polymerization, attenuated lamellipodia growth and unpolarized cell shapes (Fig. 3I-K).Accordingly, control cells plated on 2D surfaces generated denser and more branched cortical actin lament arrays in cytoplasm and lamellipodia at the leading edge, forming larger F-actin-rich lamellipodia and increasing cell spreading area (Fig. 3I-K).Taken together, these results demonstrate that KLF5-dependent actin branching drives lamellipodia growth and cell spreading, enhancing the cell motility in vitro.

KLF5 preferentially occupies distal enhancer regions
Given the canonical role of KLF5 in interpreting the genome, we employed calibrated chromatin immunoprecipitation sequencing (ChIP-seq) analysis to estimate the sequence preference and genomic binding loci of KLF5.To depict KLF5-related epigenetic landscape of NPC cells, we mined genome-wide pro les of NPC cell assessed by ChIP-seq (GSE95749) and analyzed distinct signal distribution pattern of epigenetic marks in the vicinity of KLF5 peaks.Based on signal of chromatin marks in C666-1 cells, the regions of KLF5 peaks were divided into three clusters: promoter (10.7%), enhancer (74.6%) and other regions (without marks, 14.7%) (Fig. 4A).Clusters of KLF5 peaks displayed similar distribution in HK1 cells (Supplementary Fig. S4A).Most KLF5-occupied promoter and enhancer regions extensively overlapped with accessible regions inferred by assay for transposase-accessible chromatin (ATAC), suggestive of their accessibility to transcription factors and related collaborators (Fig. 4A, B and Supplementary Fig. S4B).
Among grouped KLF5 peaks, H3K4me1 indicated enhancer, H3K4me3 indicated promoter and H3K27ac indicated enhancer and promoter in all regions, respectively (Supplementary Fig. S4C-E).In H3K27ac + regions, the H3K4me3:H3K27ac ratio was higher than H3K4me1:H3K27ac in promoter cluster, whereas enhancer cluster exhibited contrast manifestation and other unde ned regions exhibited no signi cant difference (Fig. 4C).While location of KLF5 peaks annotated to speci ed genomic regions, we observed similar effect that H3K4me1:H3K27ac was higher in intergenic regions and gene body, and H3K4me3:H3K27ac was higher in promoters (Fig. 4C).Approximately 15% of KLF5 peaks were located within compacted chromatin lacking ATAC signals (Fig. 4A, B).The metagene analysis showed that H3K27ac, H3K4me1 and H3K4me3 modi cations exhibited bimodal distribution pattern around KLF5 peak summit in enhancer or promoter cluster, while ATAC signal exhibited unimodal distribution (Fig. 4B and Supplementary Fig. S4A).These results suggested that KLF5 occupancy may propel nucleosome removal or further DNA unwrapping, rendering relevant DNA surfaces accessible to KLF5 and other DNAbinding proteins.

KLF5 transcriptionally activates ACTN4 by occupying its enhancer
To identify the prominent target transcriptionally activated by KLF5 during cell migration, we analyzed the biological function of KLF5 cis-acting targets inferred by ChIP-seq through GO and KEGG pathway enrichment analysis, and found these genes implicated actin cytoskeleton regulation, lamellipodia formation and cell migration, which was consistent with analysis on KLF5-grouped transcriptome of NPC cells and samples (Fig. 5A and Fig. 3A-F).Among candidates involved in actin cytoskeleton regulation (ACTN4, ITGB5, VCL, VAV2, BAIAP2), alpha-actinin 4 (ACTN4), a protein engaging with dynamics of actin laments [31][32][33][34], pro ciently recruited KLF5 to its gene body (Fig. 5B).With visualization of ChIP-seq data of KLF5 and histone modi cations and ATAC-seq data in NPC cells on UCSC genome browser, we observed that, in ACTN4 gene body, enhancer marks (H3K4me1 and H3K27ac) and ATAC signal was enriched in the KLF5-binding region with relatively low H3K4me3 signal (Fig. 5C and Supplementary Fig. S5A).Notably, KLF5 ChIP-seq for cells with various inherent characteristics uncovered similar binding peaks in ACTN4 gene body (Supplementary Fig. S5B), ranging from squamous cell carcinoma (SCC), gastric cancer (GC), colorectal cancer (CRC) to pancreatic ductal adenocarcinoma (PDAC).Simultaneously, the KLF5-binding region showed similar chromatin marks distribution and DNase I hypersensitivity in diverse cell types (Supplementary Fig. S5C), indicating this region as a conserved ACTN4 enhancer captured by KLF5 and recruitment of KLF5 to ACTN4 enhancer is a pervasive and heretofore unde ned phenomenon.
For orthogonal enhancer activity assays, we generated a luciferase reporter incorporating ACTN4 promoter and reporters incorporating two tandem gene cassettes: ACTN4 promoter followed by fulllength or truncated ACTN4 enhancer with one motif.Compared to pGL3-Basic reporter, introduction of ACTN4 regulatory sequences substantially activated luciferase transcription, with strongest activation by full-length ACTN4 enhancer (Fig. 5G, H and Supplementary Fig. S6H, I).Exogenous KLF5 further enhanced the luciferase activity (Fig. 5G, H and Supplementary Fig. S6H, I).We next mutated both two motifs and found diminished luciferase activity upon the mutant (Fig. 5I and Supplementary Fig. S6H, I), indicating that KLF5-bound loci dominate transcriptional activation ability of the ACTN4 enhancer.
Analysis of Hi-C data of human umbilical vein endothelial cells (HUVEC) in a 3D-genome Interaction Viewer and database [38] showed the direct interplay between ACTN4 enhancer and ACTN4 promoter (Supplementary Fig. S7).These results together demonstrate that KLF5 occupies ACTN4 enhancer and facilitates ACTN4 enhancer-promoter interplay to activate ACTN4 transcription, potentially through chromatin looping or compartmentalization.

ACTN4-actin conjunction enhances lamellipodia formation and cell motility
Since KLF5 transcriptionally activated ACTN4 in NPC cells, the positive correlation between KLF5 and ACTN4 expression was con rmed in NPC samples (Fig. 6A, B and Supplementary Fig. S8A, B).ACTN4 was upregulated in NPC samples, analogous to KLF5 expression pattern (Supplementary Fig. S8C).As opposed to ACTN4 low group, ACTN4 high group matched higher potential for metastatic progression (Fig. 6C and Supplementary Fig. S8D).GSEA analysis revealed the metastasis relevant gene set enrichment in ACTN4 high group (Fig. 6D), suggesting that ACTN4 engages metastatic pathways in NPC patients.We next knocked down ACTN4 in NPC cells, which mimicked KLF5-silenced metastatic phenotypes of inferior capacity to migrate and invade in short-term migration and wound healing assays in vitro (Fig. 6E-H and Supplementary Fig. S8E-G), indicating that ACTN4 imparts enhanced metastatic potential to NPC cells.Accordingly, overexpression of ACTN4 conveyed augmented metastatic activity (Fig. 6I, J and Supplementary Fig. S8H-J).Introduction of ACTN4 partially rescued the impaired metastatic activity of KLF5-KD NPC cells (Fig. 6K, L and Supplementary Fig. S8K, L).
In view of our data showing that KLF5 regulated actin remodeling and lamellipodia formation, as well as the well-established role of ACTN4 in actin dynamics and cell motility [31][32][33][34], we speculated that ACTN4, transcriptionally activated by KLF5, crosslinks with actin laments, triggering denser lamellipodial actin networks and lamellipodia growth.We next applied GSEA analysis of the expression pro les from ACTN4 high and ACTN4 low NPC groups, which revealed an active state of actin networks and lamellipodia organization at the cell leading edge underlying faster cell migration in ACTN4 high NPC group (Supplementary Fig. S8M).We further performed immuno uorescence analysis to detect the distribution of ACTN4 and actin laments in ACTN4-overexpressed NPC cells.ACTN4 colocalized with Factin but not with dissociative actin in the whole cell (Fig. 7A).Intriguingly, exogenous ACTN4 induced accumulation of actin laments in lamellipodia rather than in the whole cell (Fig. 7A, B).Moreover, larger area of lamellipodia were detected in cells with exogenous ACTN4, enhancing cell spread on 2D surface (Fig. 7A, B).Radial line pro le analysis showed that, F-actin intensively localized throughout the extending lamellipodia at the cell edge, and ACTN4 displayed similar distribution pattern that the uorescence intensity increased with distance from nucleus (Fig. 7C).In line with previous study [31], exogenous ACTN4 substantially colocalized with highly branched and dense actin lament networks, especially at the cell leading edge and membrane lamellipodia (Fig. 7C).Colocalization analysis corroborated the extensive colocalization between ACTN4 and F-actin in HONE-1 (Pearson's correlation value: 0.8; Manders' colocalization coe cients: tM1 = 0.908 and tM2 = 0.901) and SUNE-1 cells (Pearson's correlation value: 0.7; Manders' colocalization coe cients: tM1 = 0.836 and tM2 = 0.819).
Knockdown of ACTN4 impaired the ACTN4-actin conjunction and formation of branched actin networks, causing defective cell morphology and lamellipodia extending (Fig. 7D).Collectively, these results demonstrate the critical role of ACTN4 in remodeling actin networks at the cell edge to facilitate lamellipodia formation.

Discussion
In this study, we elucidate that KLF5 is a master TF that preferentially occupies the ACTN4 enhancer to activate its transcription, which regulates the formation of actin-driven lamellipodia and enhances the intrinsic metastatic potential of NPC cells in vitro and in vivo (Fig. 8).Through analyzing transcriptional interactome and regulatory network, we identi ed KLF5 as a master TF governing transcription program in NPC metastasis and exacerbation.Emerging evidence reveals that KLF5 promotes cancer metastasis under diverse contexts through divergent pathways [39][40][41].Our analysis suggests that NPC patients bearing high KLF5 burden are prone to develop distant metastasis and progress to an advanced stage.
By systematically analyzing the KLF5-related epigenetic landscape assessed by ChIP-seq and ATAC-seq, we discover the previously undetermined phenomenon that a large group of KLF5 peaks (74.6%) occur at distal regions coinciding with enhancer cis-regulatory elements in NPC, indicating its occupancy preference for distal enhancer regions.Enrichment of H3K27ac and ATAC signal in KLF5-captured promoters and enhancer regions reveals the chromatin accessibility and transcription active state of these regions associated genes.Accumulating evidence has linked KLF4 with dynamic enhancer organization which directs spatiotemporal gene expression programs in pluripotent stem cells [21,42,43].Previous chromatin pro ling studies showed that KLF4 cooperating with reprogramming factors mainly occupies enhancers in reprogramming and established pluripotent stem cells [43].KLF4 orchestrates the long-range chromatin loops rendering physical contacts between enhancers and promoters through interaction with architectural proteins such as cohesion [21,42].Recruitment of pioneer factors to chromatinized binding sites opens and remodels the chromatin and facilitates subsequent binding of other TFs and cofactors [44,45].KLF4 is a pioneer factor promoting reprogramming and the impaired pluripotency due to loss of KLF4 can be compensated by other KLFs (e.g., KLF5), indicating potential pioneer factor identity of them [46,47].Given the bimodal distribution of histone marks and unimodal distribution of ATAC signal around KLF5 peak summit, whether KLF5 operates as a pioneer factor and reads out nucleosomal DNA in compacted chromatin to facilitate appropriate nucleosome occupancy and position needs further study.
Orchestrated by transcription modulation, cell migration predominantly depends on precise tuning of global actin ow at the cell front and rear [6, 48].Our analysis shows that KLF5 interprets the genome to respond to promigratory cues, triggering rapid modulation of genes involved in actin cytoskeleton transcriptionally.Among these candidates, ACTN4 pro ciently recruits KLF5 to its gene body.Integrating ChIP-seq data of chromatin marks and KLF5 under various contexts, we discovered a heretofore unrecognized and conserved enhancer element in ACTN4 gene body occupied by KLF5.Intriguingly, introducing truncated ACTN4 enhancer with one motif slightly impaired the transcriptional activation strength of ACTN4 promoter with or without ectopic KLF5, presumably attributed to sequence integrity requirement for stabilizing the enhancer-promoter interplay.Alternatively, the truncated ACTN4 enhancer lacks potential to recruit su cient KLF5 and co-binding transcription activators capturing anking sequences, hence some repressive TFs and corepressors partially substitute for KLF5.Given canonical regulatory modes of long-range enhancer-promoter crosstalk [49][50][51], KLF5 on ACTN4 enhancer may recruit coactivators and promote the chromatin loop extrusion or genome compartmentalization to synergistically drive transcription.While the structure of KLF5-DNA complex remains undetermined, we simulated the docking of putative KLF5 structure and its binding motif in ACTN4 enhancer inferred by ChIP-seq and found a high con dence of the KLF5-motif docking model.Additional essential biochemical information will be obtained by high-resolution structure research of puri ed KLF5 and its partners.
Recent comprehensive measurements of human transcriptional effector domains revealed that KLF5 boasts activation and repression domains, rendering the bifunctional potential of KLF5 in speci c contexts [37].Thus, in response to distinct cues, KLF5 potentially oscillates between activating and repressing activity at various genomic loci and the regulatory output can be delineated by stoichiometries.
The transcription activation of ACTN4 by KLF5 explains the impaired actin network structure, actin-rich lamellipodia, cell spreading and directed migration caused by depletion of KLF5 in NPC cells.ACTN4actin conjunction empowers metastatic NPC cells to generate actin-driven lamellipodia and e ciently migrate faster to seek congenial milieu to survive and grow.Though studying KLF5-ACTN4 facilitated cell migration using 2D systems reveals the mechanism of how KLF5 enhances the metastatic potential of cells in vitro and in vivo, utilization of tools and more complex systems in more physiological contexts can deepen our understanding of migration principles underlying metastasis, through which providing promising therapeutic avenues to control metastasis -the overwhelming cause of cancer-associated death.
To prevent metastasis and eliminate established metastatic lesions, numerous agents have been developed to identify and drug targets in metastatic cascade, of which TFs are switching from 'undruggable' to 'druggable' [52][53][54].However, clinical validation of TF-targeting agents and incorporating them into the standard of care remains problematic, owing to their cytostatic manifestation in preclinical models and compensatory pathways involving requisite steps in metastasis [52].Therefore, rational combination therapies are e cacious in targeting metastasis.KLF5, the dysregulated master TF in metastatic NPC, represents a unique potential therapeutic target controlling metastasis, with challenges of developing inhibitors targeting its protein-DNA and protein-protein interplay [54].Though inhibitors targeting KLF5-DNA binding remain to be developed, a recent effort overcame the di culty of inhibitor recognizing convex and highly positively charged DNA binding interfaces [54,55].An agent targeting RUNX-DNA binding increased survival of mouse xenograft models [55].Inhibitors interrupting binding of KLF5 on ACTN4 enhancer and other loci may serve as means to mitigate metastases of NPC.Moreover, utilizing the induced proximity principle and ubiquitin-proteasome system, the proteolysis-targeting chimaera molecules potentially degrade KLF5 through a ligand binding KLF5 and another ligand recruiting ubiquitin ligases [56].
In summary, our discoveries underscore the prometastatic role of KLF5, a master TF in NPC, and provide mechanistic insights to identify potential biomarkers and therapeutic interventions to shrink metastases.Novel agents targeting mutated or dysregulated KLF5 and other master TFs in combination with rst-line therapies will improve outcome of patients with metastatic disease.

Transcriptional interactome and regulatory network inference
The transcriptional interactomes were generated by ARACNe-AP algorithm [26] from 4 GEO datasets pro led by RNA-seq or microarray: GSE118719, GSE68799, GSE13597 and GSE103611.ARACNe-AP was run with 100 reproducible bootstrap iterations using retrieved gene expression pro les and 1,639 TFs as prede ned input, with parameters setting to no DPI (Data Processing Inequality) tolerance and MI (Mutual Information) p-value threshold of 10 − 8 .Inferred signi cant KLF-target interactions were ltered (P < 0.05) for subsequent analysis.With input of transcriptional interactomes and gene expression pro les, regulatory networks were reverse engineered to output regulons.

TFs activity inference
TFs activity pro les were estimated by msVIPER algorithm [28], which tests regulon enrichment on gene expression signatures.By comparing NPC samples under speci c contexts (e.g., metastatic versus nonmetastatic), gene expression signatures were generated to identify potential role of TFs in NPC progression.After randomly permuted samples of gene expression pro les 1,000 times, null models were produced by using signatures generated with permutation iterations.Comparing each regulon enrichment score to a null model, TFs activity was inferred as normalized enrichment score, facilitating the identi cation of master TFs in distant metastasis of NPC.

RNA extraction, RT-PCR and qPCR
Total RNA extracted from cells using either TRIzol reagent (Invitrogen) or RNA Quick Puri cation kit (ESscience) according to the manufacturer's instructions, following RNA quality and quantity measurement by NanoDrop 2000 (Thermo Fisher Scienti c).cDNA was generated using HiScript III RT SuperMix for qPCR (Vazyme) according to the manufacturer's instructions.cDNA was diluted 1:10 in distilled water and 2 µl was used per qPCR reaction.
qRT-PCR was performed using SYBR qPCR Master Mix (Vazyme) on CFX96 Touch sequence detection system (Bio-Rad).Primers were designed using the Primer-BLAST tool at NCBI.Housekeeping gene mRNA level (GAPDH) was used for normalization.The mRNA levels of all genes were quanti ed using the 2 −ΔΔCt method to infer the difference.qPCR primer sets are listed in Supplementary Table S3.
Western blotting microscope (NIKON Eclipse Ti2-U).Images were analysed using Fiji/ImageJ by drawing a line indicating the migrating fronts on two sides of the scratch wound and comparing healing rate between groups.

Transwell migration and invasion assays
For Transwell migration and invasion assays, 8 µm pore size transwell chambers (corning) were coated with or without Matrigel (BD Biosciences) according to the manufacturer's instructions.Cells resuspended in 200 µl serum-free medium were seeded into the upper chambers with 3.5 × 10 4 HONE-1 cells or 4.5 × 10 4 SUNE-1 cells for migration assay and 7 × 10 4 HONE-1 cells or 9 × 10 4 SUNE-1 cells for invasion assay.Lower chambers were lled with 500 µl 10% FBS-added medium.After incubation for 12-14 hours (migration assay) or 20-22 hours (invasion assay), the migrated or invaded cells were xed with methanol for 10 min, stained with hematoxylin for 2 h and imaged by inverted microscope (NIKON Eclipse Ti2-U).Cell migration and invasion capability were de ned as relative number of migrated or invaded cells manually counted.

In vivo metastasis models
In metastasis experiments were performed in accordance with the guidelines of the Institutional Animal Care and Use Ethics Committee of SYSUCC (L025504202207003).BALB/c nude mice (female, 4-6 weeks old, 15g) were obtained from Charles River Laboratories (Beijing, China) and maintained at the Animal Experiment Center of the Sun Yat-sen University.
For the lung metastasis model, KLF5-KD or control HONE-1 cells (1 × 10 6 cells) in PBS were injected into the tail vein of mice (n = 6 per group).One month after injection, mice were euthanized and their lungs were dissected and analyzed for metastasis.
For the inguinal lymph node metastasis model, KLF5-KD or control HONE-1 and SUNE-1 cells (2 × 10 5 cells) in PBS were inoculated into the footpads of mice (n = 6 per group and n = 10 per group).One month after inoculation, mice were euthanized and their footpad primary tumors and inguinal tumor-draining lymph nodes were dissected and analyzed for metastasis.
The primary tumors and lungs were xed in 4% formaldehyde for 48 h.Tissues were embedded in para n, sectioned into 5 µm pieces, and mounted on slides for hematoxylin and eosin staining.Slides were imaged using an Olympus scanning system (VS200).
Mean intensity of F-actin and spreading area in the whole cell and lamellipodia were recorded through manually drawing the free-form regions of interest (ROI) using Fiji/ImageJ.For radial line pro le analysis, line scans of captured images were generated through lamellipodial outlines and the cell edge was applied as a reference to evaluate the spatial distribution of ACTN4 and F-actin.The Pearson correlation coe cient (r) was calculated between the mean intensity of F-actin and ACTN4 at different distance.For colocalization analysis, images of cells expressing ectopic ACTN4 were exported to Fiji and colocalization e ciency and speci city of ACTN4 and F-actin was analyzed using the Fiji Coloc2 plugin.

ChIP
ChIP assay was performed using Pierce Magnetic ChIP Kit (26157, Thermo Fisher Scienti c) following the manufacturer's instructions.Brie y, 4 × 10 6 cultured HONE-1 or SUNE-1 cells were harvested and washed with ice-cold PBS twice.Cells were xed in 2ml of 1% formaldehyde for 10 min at room temperature, and Glycine Solution (10×) was added to a nal concentration of 1× to quench crosslink for 5 min at room temperature, following two washes with ice-cold PBS.Cells were collected by mechanically scrapping in 1mL of ice-cold PBS with 10 µL of the Halt Cocktail and centrifuged at 3,000 × g for 5 min at 4°C.After removing PBS, crosslinked cells were lysed using Membrane Extraction Buffer containing protease/phosphatase inhibitors and digested using MNase.After sonication on ice to break nuclear membrane, solutions were centrifuged at 9,000 × g for 5 min at 4°C. 10% of the supernatants containing the digested chromatin was stored as input sample, and the remaining supernatants were incubated with primary antibodies: 1 µL IgG and 3-5 µg anti-KLF5 (rabbit; Sigma-Aldrich, 09822), anti-HA (rabbit; Abcam, ab9110) or anti-H3K27ac (rabbit; ab177178) at 4°C overnight with mixing.Protein A/G Magnetic Beads were added and incubated for 2 hours at 4°C with mixing.Beads were collected with a magnetic stand and were washed three times with IP Wash Buffer 1 and once with IP Wash Buffer 2. The protein-DNA complexes were eluted from beads and de-crosslinked using IP Elution Buffer containing NaCl and Proteinase K for 3 hours at 65°C.DNA samples were then puri ed and subjected to qPCR or ChIP-seq library generation.Primers were provided in Supplementary Table S6.

Dual-luciferase reporter assay
Promoters of ACTN4 (1,000 bp upstream of the transcription start site) and truncated ACTN4 enhancer containing motif1/2 or full-length ACTN4 enhancer or mutant ACTN4 enhancer (all bases of motifs replaced with adenosines) were cloned into pGL3-basic luciferase reporter plasmid (Promega).Inserted sequences were provided in Supplementary Table S7.Cells were seeded into 24-well plates and cotransfected with plasmids encoding an empty vector or HA-tagged KLF5 (400 ng) and luciferase reporter plasmids (100 ng) and a Renilla luciferase reporter (Addgene, 10 ng) using Lipofectamine 3000 (Invitrogen).

Motif enrichment analysis
For motif discovery, the motifs in KLF5 peaks of ChIP-seq were discovered using MEME (v.4.10.1) with the signi cance cutoff E value < 10 − 8 .For motif comparison, the discovered motifs in KLF5 peaks of ChIP-seq were compared to known motifs in HOCOMOCOv11 database using Tomtom (v.5.5.3).

RNA-seq analysis
For RNA-seq analysis of HONE-1 cells with or without knockdown of KLF5, total RNA was extracted from cells using TRIzol reagent (Invitrogen).Puri ed mRNA using Dynabeads Oligo (dT) (Thermo Fisher) was fragmented by NEBNext Magnesium RNA fragmentation module (NEB) and reverse-transcribed to create the cDNA by SuperScript™ II Reverse Transcriptase (Invitrogen), following U-labeled second-stranded DNAs synthesis with E. coli DNA polymerase I (NEB), RNase H (NEB) and dUTP Solution (Thermo Fisher).The nal cDNA libraries were constructed by random hexamer primed cDNA synthesis, PCR ampli cation and size selection as 300 ± 50 bp.The cDNA libraries were sequenced on the Illumina Novaseq™ 6000 platform according to the manufacturer's instructions.Raw reads were ltered and quality-veri ed using cutadapt (v.1.9.1) and FastQC (v.0.11.9) before alignment.Reads were aligned and assembled using HISAT2 (v.2.2.1) and StringTie (v.2.1.6)against the human genome GRCh38.Transcriptomes from all samples were merged to reconstruct a comprehensive transcriptome using gffcompare software (v.0.9.8).
Transcript-level and gene-level were quanti ed using StringTie and ballgown.Normalization of raw count data and genes differential expression analysis were performed using the DESeq2 software.The genes with the parameter of false discovery rate (FDR) below 0.05 and absolute fold change ≥ 2 were considered differentially expressed genes (DEGs), subjected to following GO, KEGG and GSEA analysis.The RNA-seq data for this study are available for download from the Gene Expression Omnibus (GEO) repository (GSE243952).

GO and KEGG analysis
GO and KEGG enrichment analysis were performed against Human MSigDB (v2023.1.Hs) collections c2 (KEGG subset of Canonical pathways) and c5 (Gene Ontology gene sets) for genes ranging from DEGs from RNA-seq of HONE-1 cells with or without knockdown of KLF5, DEGs identi ed from KLF5 hi and Promega), and re y luciferase activity was normalized to Renilla luciferase activity.Luciferase mRNA level in different groups were evaluated through RNA extraction, RT-PCR and qPCR.
Cells were lysed after 24-48h of transfection by Dual-Luciferase Reporter Assay System (E1910, Promega) following the manufacturer's instructions.Luciferase activity was measured by GloMax Navigator (