Blood and lymphatic systems are segregated by the FLCN tumor suppressor

Blood and lymphatic vessels structurally bear a strong resemblance but never share a lumen, thus maintaining their distinct functions. Although lymphatic vessels initially arise from embryonic veins, the molecular mechanism that maintains separation of these two systems has not been elucidated. Here, we show that genetic deficiency of Folliculin, a tumor suppressor, leads to misconnection of blood and lymphatic vessels in mice and humans. Absence of Folliculin results in the appearance of lymphatic-biased venous endothelial cells caused by ectopic expression of Prox1, a master transcription factor for lymphatic specification. Mechanistically, this phenotype is ascribed to nuclear translocation of the basic helix-loop-helix transcription factor Transcription Factor E3 (TFE3), binding to a regulatory element of Prox1, thereby enhancing its venous expression. Overall, these data demonstrate that Folliculin acts as a gatekeeper that maintains separation of blood and lymphatic vessels by limiting the plasticity of committed endothelial cells.

T he two major circulatory systems of blood and lymphatic vasculature are discretely distributed throughout the body. While the former provides tissues with oxygen and nutrients, the latter drains the interstitial fluid from the tissue spaces to return it to the bloodstream 1 , transports immune cells, absorbs dietary lipids 2 , and clears waste from the brain 3 . The structures of the two systems, the blood vessel (BV) and the lymphatic vessel (LV), are histologically similar but anatomically do not share the same lumen, except at the lymphovenous valve located near the venous angle, the final junction of collecting lymph ducts, and subclavian veins 4,5 . LVs arise mostly from preexisting veins 6,7 , although populations of lymphatic progenitors of non − venous origin have been identified 8,9 . The expression of Prox1, a master transcription factor controlling lymphatic specification, determines both the initiation and maintenance of the identity as a lymphatic endothelial cell (LEC), contributing to the separation of the blood and lymphatic systems 10,11 . Platelets activated by committed LECs also contribute to this separation 5,12,13 . However, the molecular mechanisms that ultimately maintain separation of these two circulatory systems have not been fully identified.
Folliculin (FLCN) has long been studied as a tumor suppressor gene responsible for the human autosomal dominant genetic disorder Birt-Hogg-Dubé (BHD) syndrome 14 . Affected individuals are at risk of developing of kidney cancers, lung cysts, and cutaneous fibrofolliculomas 15,16 . FLCN encoded by the FLCN gene, as a complex with its binding proteins FNIP1/2, acts as a GTPase activating protein (GAP) for Rag GTPases, thereby transducing amino acid levels to mTOR complex 1 (mTORC1) [17][18][19] and secluding the basic helix-loop-helix (bHLH) transcription factor and member of the MiTF family, Transcription Factor E3 (TFE3), in the cytoplasm 20,21 . Recently, FLCN has been revealed as the central player in the control of pluripotency of embryonic stem (ES) cells 21,22 . In this study, based on a unique phenotype of Flcn − deficient mice, we have identified the fundamental mechanism maintaining the separation of the blood and lymphatic vascular systems through limiting the plasticity of committed endothelial cells.
Flcn deficiency leads to ectopic Prox1 expression in veins. Next, we established a technique to visualize whole-mount tail skins ( Fig. 2a-e; Supplementary Fig. 2a-k). This technique allowed us to easily visualize two streams of arterial-venous-lymphatic bundles on both sides of the tail, and Cdh5-BAC-Cre ERT2 excised Flcnflox alleles in both blood and lymphatic endothelial cells ( Supplementary Fig. 2l). With this technique, the lymphatic dilation in Flcn iΔEC mice was apparent (Fig. 2b, c). Notably, Prox1 was ectopically expressed in VECs of Flcn iΔEC mice, some even with equal intensity to that seen in their LECs (Fig. 2d-j). Prox1 expression was also observed in retinal veins and the vena cava in Flcn iΔEC pups and cervical veins of Flcn ΔEHC embryos (Fig. 2k-p), suggesting that this phenomenon is widespread. When we deleted Flcn in adult endothelial cells, we observed similar ectopic expression of Prox1 in VECs ( Supplementary Fig. 1m, n). Considering that we did not detect lymphatic hyperplasia and blood filling in adult Flcn iΔEC mice, some compensation mechanism might have occurred after weaning and masked the phenotypes seen in their neonatal period. Consistent with the knowledge that transgenic Prox1 overexpression induces an LV phenotype in VECs 25 , mature LEC markers like Pdpn and Vegfr3 26 were slightly but significantly upregulated in Prox1 + VECs of Flcn iΔEC mice ( Supplementary Fig. 2m-o). In high magnification views, we observed abnormal sprouting from LVs toward veins as well as from veins toward LVs in Flcn iΔEC mice (Supplementary Fig. 2p-u). We also detected occasional bridging of endothelial filopodia between LVs and veins in Flcn iΔEC mice ( Supplementary  Fig. 2u), suggesting they somehow attracted each other. The EdU incorporation assay showed increased LEC but not VEC proliferation in Flcn iΔEC mice ( Fig. 2q-v). This hyper proliferation of LECs is likely responsible for lymphatic enlargement, and is caused by increased Vegfr3 expression (Supplementary Fig. 2m-o) enhancing ligand responsiveness.
Flcn deficiency gives rise to lymphatic-biased venous endothelial cells. To elucidate more precisely the cellular status, we employed single-cell RNA sequencing (scRNA-seq) on CD31 + cells isolated from the mesentery ( Supplementary Fig. 3a, b). The Uniform Manifold Approximation and Projection (UMAP) of cells revealed interconnected distribution of gene expression profiles suggesting continuities and discontinuities between each cell type (Fig. 3a). Subsequent clustering analysis identified non-endothelial and six endothelial clusters (Fig. 3b, c) that could be assigned as capillary, venous, or arterial endothelial cells (one cluster each) or LECs (three adjacent clusters) based on the expression of known endothelial subtype markers (Fig. 3c, d; Supplementary Fig. 3c). Most of the CD31 + non-endothelial cells were hematopoietic cells as they expressed CD45 ( Supplementary Fig. 3c-e). Notably, cells from Flcn iΔEC but not control mice showed a unique cell population within VECs, which expressed Prox1 (Fig. 3e-i). Comparison of transcriptional profiles revealed that this Prox1 + VECs most resembled VECs and showed a transitional state between a venous and lymphatic endothelial fate in terms of those specific markers (Fig. 3j, k; Supplementary Fig. 4a). Considering the elevated expression of mature lymphatic markers in veins of Flcn iΔEC mice ( Supplementary Fig. 2m-o) we named this population as LEC-biased VECs. Among those listed genes, we confirmed the upregulation of Cldn11 and downregulation of vWF in LEC-biased VECs, namely, VECs found in Flcn iΔEC mice, by immunohistochemistry ( Supplementary Fig. 4b-h). Extraction of genes characterizing the cluster of Prox1 + VECs did not highlight any gene specifically expressed in this cell population; rather, most were strongly expressed in VECs and LECs ( Supplementary Fig. 4i, j; Supplementary Data 1 and 2).
Ectopic Prox1 in veins accounts for the BV-LV misconnection in Flcn iΔEC mice. In agreement with the lymphatic blood filling in Flcn iΔEC mice, intracardiac injection of lectin revealed its influx into LVs, suggesting that somewhere BVs and LVs were mixed (Fig. 4a). In section specimens, LVs and veins shared a lumen perfused with lectin in Flcn iΔEC mice ( Fig. 4b-h). Transgenic overexpression of Prox1 in embryonic endothelial cells also resulted in similar blood-filled lymphatics ( Fig. 4i- Fig. 6a-h). At P6, venous Prox1 expression and blood-filled lymphatics were already evident in Flcn iΔEC mice without subcutaneous hemorrhage ( Supplementary  Fig. 6i-l). These data suggest that erythrocytes, which entered LVs, eventually extravasated from the inter-cellular gap between LECs. Next, we examined the lymphovenous valve (LVV) located at the venous angle as the cause of lymphatic blood filling. In immunohistochemistry of LVVs, we could not detect apparent abnormalities of the LVV structure including the two-layer structure of VECs and LECs in Flcn iΔEC mice at P8 ( Fig. 4p; Supplementary Fig. 6m-r). Under a stereo microscope, no apparent blood filling was observed in the thoracic ducts (TDs) of either control or Flcn iΔEC mice ( Supplementary Fig. 6s-v), suggesting impaired function of the LVV is less likely to be the cause of lymphatic blood filling in Flcn iΔEC pups. Importantly, Prox1 deletion in endothelial cells normalized the phenotypes of Flcn iΔEC mice ( Fig. 4q-t), suggesting their aberrant Prox1 expression was causative. Interestingly, mice lacking Prox1 in endothelial cells showed absence of lymphatic valves, although they did not have apparent impairment in overall structures of lymphatic vascular networks in accordance with the strong expression of Prox1 in lymphatic valves at this stage. The LECspecific Flcn deletion (Flcn iΔLEC ) led only to lymphatic hyperplasia but not blood-filled LVs and lethality ( Fig. 4u-z),  Flcn iΔEC mice show reddish colored skin (arrowheads in n) as well as dilated and blood-filled LVs (arrowheads in o). q-x Bright field views and whole-mount or section samples of mesentery at P8. Erythrocytes (closed arrowheads) are detected not only in arteries (A) and veins (V) but also in LVs of Flcn iΔEC mice. Scale bars: 5 mm (o); 1 mm (a-c, g, h, q, r); 200 µm (i-l, o, s, t); 50 µm (d, e, u-x). **P < 0.01; *P < 0.05; ND not detected. Data presented are the mean ± SD. The comparisons between the averages of the two groups were evaluated using the two-sided Student's t-test. Source data are provided as a Source Data file.
TFE3 binds the regulatory element of PROX1 directly. To determine the mechanism for Prox1 upregulation in Flcn iΔEC mice we focused on Tfe3, which intracellular localization and degradation is well known to be regulated by Flcn 21,22 . In the wild-type tail skin, Tfe3 immunoreactivity was abundant in mesenchymal cells, but barely visible in BVs or LVs (Fig. 5a, b). However, the VECs and LECs in Flcn iΔEC mice showed strong expression of Tfe3, which was localized in endothelial nuclei ( Fig. 5c-l). The specificity of this staining was confirmed using Tfe3 deficient mice (Supplementary Fig. 7a-d). Quantitative PCR analysis indicated that the mRNA expression of Tfe3 in isolated BECs was not different between control and Flcn iΔEC mice (Fig. 5m), suggesting nontranscriptional regulation like nuclear translocation and lysosomal degradation 21,22 is responsible for increased Tfe3 immunoreactivity. In control veins, Tfe3 proteins were colocalized with RagC, suggesting that Flcn secludes Tfe3 in the cytoplasm by binding the GDP form of RagC/D ( Fig. 5n-u; Supplementary Fig. 7e-n), as has previously been found in embryonic stem cells 17,21 . Analysis of scRNA-seq data showed expressions of Flcn and Tfe3 were significantly higher in VECs than arterial endothelial cells ( Supplementary Fig. 3f). This expression tendency might be, at least in part, the determinants for the VEC-inclined phenotype in Flcn iΔEC mice. Similarly, LECs had more abundant expression of Flcn than arterial endothelial cells, but the expression of Tfe3 was comparable between them ( Supplementary Fig. 3f), suggesting nontranscriptional regulation of Tfe3 like nuclear translocation and lysosomal degradation. We performed chromatin immunoprecipitation sequencing (ChIPseq) analysis using HK-2 cells. We found that TFE3 bound to the 150 bp genomic sequence just upstream of the transcription start site of PROX1 (hereafter referred to as PROX1 − 0.15 kb; Fig. 5v), which is distinct from the previously identified binding regions for Sox18, Gata2, FoxC2, and Nfatc1 27,28 . PROX1-0.15 kb has two E-box (enhancer box) sequences, suggesting a function as a regulatory element driven by a bHLH transcription factor (Fig. 5w). Indeed, doxycycline-induced TFE3 overexpression in HEK293 cells increased the expression of PROX1 (Fig. 5x), although the late induction of PROX1 may suggest the contribution of some indirect effect. The luciferase reporter assay confirmed the significance of E-box sequences located in PROX1-0.15 kb during TFE3-driven PROX1 transcription (Fig. 5y). endothelial cells. ChIP analysis of HUVECs confirmed that endogenous TFE3 binds PROX1-0.15 kb (Fig. 6a, b). Next, we performed the same ChIP assay on HUVECs treated with small interfering (si)RNAs targeting TFE3, confirming that this binding occurred specifically in endogenous TFE3 (Fig. 6c, d). To evaluate the impact of the FLCN-TFE3 axis on the expression of PROX1, we knocked down FLCN with or without TFE3 in HUVECs. We found si-FLCN significantly increased PROX1 transcription, and its effect was normalized by si-TFE3 (Fig. 6e, f, g). Supporting this observation, with si-FLCN treatment, TFE3 + cells largely correspond to PROX1 + cells ( Fig. 6h-m). Tfe3 deletion in vivo completely normalized blood-filled LVs, and partially normalized the lymphatic hyperplasia of Flcn iΔEC mice ( Fig. 6n-w). Flcn is reported to negatively regulate PGC1α and mitochondrial biogenesis in several organs, and biochemically functions in association with mTORC1 14,17,21,29 . Moreover, Prox1 expression is suppressed by rapamycin in a lymphangiectasia model 30 , suggesting some interaction among Prox1, mTor, and Flcn. We, therefore, created mice with deleted Flcn and deleted Pgc1α or Mtor in endothelial cells (Flcn;Pgc1α iΔEC and Flcn;Mtor iΔEC ), but the phenotype of these deficient mice was indistinguishable from that of Flcn iΔEC mice ( Supplementary Fig. 8).

Discussion
In this study, we found that strict limitation of plasticity between LECs and VECs governed by the FLCN-TFE3 pathway ensures the segregation of the lymphatic and blood circulatory systems. FLCN, as a complex with FNIP1/2, acts as a GAP for RagC/D and secludes TFE3 in the cytoplasm 21 . In our study of deficient mice, without Flcn, Tfe3 was translocated into the nucleus and ectopically upregulated Prox1 by binding to its promoter through the E-box in VECs (Fig. 7o). Genetic regulators of Prox1 have been intensively studied. Sox18 directly activates Prox1 transcription by binding to its proximal promoter 4.5 kb downstream from the transcription start site (Prox1 + 4.5 kb) 27 . In addition, a putative enhancer element of Prox1 (-11 kb upstream enhancer element) is bound by Gata2, Foxc2, and Nfatc1 28 , genes identified to be important for lymphatic development 31,32 . Considering that Tfe3-knockout mice do not show lymphatic defects, Tfe3 binding to the Prox1 − 0.15 kb might be dispensable for normal development, but may have an important role in pathological settings, in particular when Tfe3 is abnormally activated.
Mature LEC markers like Pdpn and Vegfr3 were slightly upregulated in LEC-biased VECs in Flcn iΔEC mice, but were far weaker than in their LECs, suggesting Prox1 expression is not sufficient to fully convert VECs to LECs. Such an ambiguous endothelial population exists in the interface of BV-LV junctions. In the budding process of initial lymphatics, bipotential Prox1 + precursor cells in veins asymmetrically divide into two daughters; one becomes Prox1 high LECs, which start to express mature LEC The comparisons between the averages of the two groups were evaluated using the two-sided Student's t-test. **P < 0.01; *P < 0.05; ND, not detected. Data are presented as the mean ± SD. Source data are provided as a Source Data file.
markers, and the other downregulates Prox1 and remains in the vein 33 . In the lymphovenous valve, Prox1 + endothelial cells line on both sides of the valve, but the mature LEC markers were only expressed on the lymphatic side 4 , suggesting LEC-baised VECs have an affinity to bona fide LECs. We suppose this Flcn function is not related to tumor suppressor activity, but is related to the regulatory function of the stem cell pluripotency and quiescence 21,22,34 . Our data indicate Flcn acts as a gatekeeper limiting the plasticity in committed endothelial cells between venous and lymphatic ones. As lymphatic endothelial cells differentiate from venous ones 6,7 , our data fit with the concept that Flcn orchestrates the differentiation status of various cells through nuclear translocation of Tfe3.
In the future, it is quite interesting to analyze the relevance of our data in mice to human BHD pathology, in particular the possibility that bialleic FLCN inactivation occurs in VECs of BHD lungs. Otherwise, monoalleic FLCN inactivation might be sufficient to render VECs vulnerable to environmental alterations such as bullae formation caused by epithelial damage. Affected VECs might be readily biased to LECs.
Overall, the data presented herein identify the fundamental molecular mechanism that maintains separation of the blood and lymphatic vascular systems. These findings may provide the foundation for the development of novel therapeutic approaches for human lymphatic disorders as well as to prevent cancer metastasis.

Methods
Mouse and matings. Animal experiments were approved by the regional animal study committees (Keio University) and were performed in accordance with the Guidelines of Keio University for Animal and Recombinant DNA experiments. The Tie2-Cre 24 , Prox1-Cre ERT27 , Prox1-flox 35 , Pgc1α-flox 36 , Tfe3 +/-37 , CAG-LSL-EGFP 38 , Cdh5-BAC-Cre ERT239 , and Flcn − flox 40 mice have been described previously. mTor-flox mice were obtained from the Jackson Laboratory (Stock No: 011009). All mice were crossed with C57BL/6 J mice more than eight times and     h-m Immunocytochemistry of HUVECs. PROX1 + cells largely correspond to TFE3 + cells (arrowheads). n-w Immunohistochemical analysis of tail whole mounts (n-q) or sections (r-u) at P8 and quantification (n = 3). Venous Prox1 expression (arrows) and blood-filled lymphatics (closed arrowheads) are detected in Flcn iΔEC mice. Hyperplastic LVs in Flcn iΔEC Tfe3 -/y are not blood-filled (open arrowhead). Scale bars: 200 µm (n-q); 50 µm (h-m, r-u). The comparisons between the averages of the two groups were evaluated using the two-sided Student's t-test. **P < 0.01; *P < 0.05; ND not detected. Data are presented as the mean ± SD. Source data are provided as a Source Data file. Unprocessed original scans of blots are shown in Source Data file.
Preparation of whole-mount and section tissues. For preparation of wholemount tail samples, the distal half of the tail was cutoff and an incision made in the ventral side with a scalpel. The coccyx and muscles were surgically removed and then the samples were flattened for 10 min in 4% paraformaldehyde (PFA) in PBS.
To prepare whole-mount skin samples from embryos, cutaneous tissues were removed from underlying muscles and flattened for 10 min in 4% PFA/PBS. For preparation of whole-mount mesenteric samples, intestines were unfolded 'to complete the wheel', according to a previous paper 41 . Isolated tails, skins, and mesenteries were re-fixed overnight before staining with antibodies. For section specimens, tissues were dissected out from mice and fixed overnight in 4% PFA in PBS. All whole-mount or section tissues were stained as described below.
Immunostaining. Immunohistochemical staining of whole-mount samples and tissue sections was performed as described previously 42  For double labeling with primary rabbit antibodies, the conjugated Fab fragments method was used (https://www.jacksonimmuno.com/ technical/products/protocols/double-labeling-same-species-primary/example-a).
Briefly, an excess amount of Alexa Fluor 488-conjugated AffiniPure Fab fragment donkey anti-rabbit IgG (Jackson ImmunoResearch, 711-547-003; 1:100) was applied to achieve effective blocking of the first primary antibody. To analyze cell proliferation in vivo, an EdU incorporation assay using Click-iT EdU Imaging Kits (Invitrogen) was performed according to the manufacturer's instructions. Briefly, 50 μl of EdU dissolved in DMSO/PBS (final concentration, 0.5 mg/ml) was injected intraperitoneally into P8 mice 6 h before sacrifice. Cell culture. HUVECs were cultured in EGM-2 medium (Cambrex, East Rutherford, NJ). HEK293 cells were cultured in DMEM supplemented with 10% fetal bovine serum (FBS). HK-2, a human kidney proximal tubule cell line, which expresses hemagglutinin (HA)-TFE3 in a doxycycline-dependent manner, was generated by two sequential lentiviral transductions using a Tet-On 3 G System (Clontech) and cultured in Advanced DMEM/F12 medium with 1.5% Tetracycline-Free FBS (Clontech), penicillin − streptomycin (100 U/ml), and selection antibiotics, 2 μg/ml blasticidin S and 0.8 μg/ml puromycin. For RNA interference, HUVECs were washed once with OptiMEM (Life Technologies) and transfected with 40 nM siRNA Duplex (Sigma-Aldrich) using 6 μl/ml Lipofectamine RNAiMAX (Invitrogen) in OptiMEM according to the manufacturer's instructions. After 24 h, the transfection medium was removed, and complete culture medium was added. Cells were cultured for a further 96 h and subsequently a second transfection was performed before each experiment. The FlexiTube GeneSolution was used for FLCN (SI03048402, QIAGEN, Hilden, Germany) and ON-TARGETplus siRNA for TFE3 (L-009363-00-0005, Dharmacon, Cambridge, UK). As a control, a siRNA-negative control duplex oligonucleotide (ThermoFisher Waltham, MA) was used.
Single-cell RNA sequencing. At P8, CD31 + cells were isolated from the mesentery of three mice per genotype using Dynabeads (Veritas). Libraries were constructed using a commercial microdroplet-based platform, Chromium Single Cell 3' v3 (10×Genomics, Pleasanton, CA) according to the manufacturer's instructions. A HiSeq3000 platform (Illumina, San Diego, CA) was used to generate 28 and 91 base paired-end reads. RNA-seq tags from the Chromium experiments were aligned using Cell Ranger 3.0 software (10×Genomics). Using Cell Ranger, sequences with low quality and PCR duplicates were removed. The scRNA-seq data were processed and analyzed by Seurat package v3.5.2 43 . UMAP was used for the dimensionality reduction that captures both local and global structure in scRNAseq data. Clusters were visualized in two dimensions as feature or violin plots, and annotated based on the expression of canonical endothelial subtype markers. Clustering itself does not include any supervison such as excluding specific genes or cell populations, but the determination of cell types such as LECs, VECs, AECs was arbitrarily conducted based on the known expression profiles of marker genes. Hierarchical clustering was performed by the Euclidean distance and group average method. Gene expression was averaged in each cluster and the correlation matrix calculated by Pearson correlations. The heatmap was created using the pandas v0.22.0, seaborn v0.8 package. To construct the lists of genes up-or downregulated in LEC-biased VECs, differentially expressed genes were extracted using the Wilcoxon Rank Sum test in the Seurat package. To select genes characterizing the cluster of LEC-biased VECs, genes differentially expressed when compared with all other clusters were extracted using the Seurat package.
Quantitative real-time RT-PCR analysis. Total RNA was prepared from cultured cells or mouse blood endothelial cells (Pdpn -LYVE1 -CD31 + ) isolated from the mesentery using Dynabeads as described above. Reverse transcription was performed using Superscript II (Invitrogen). Quantitative PCR assays were performed on an ABI 7500 Fast Real-Time PCR System using TaqMan Fast Universal PCR master mix (Applied Biosystems) and TaqMan Gene Expression Assay Mix with mouse Tfe3 (Mm00552681_m1), human PROX1 (Hs00160463_m1), or TFE3 (Hs00232406_m1). A mouse β-Actin (Mm00607939_s1) or a human β-ACTIN (Hs00357333_gl) assay mix served as an endogenous control. Data were analyzed using 7500 Fast System SDS Software 1.3.1 (Applied Biosystems). Each experiment was performed with four replicates from each sample and the results were averaged.
Confocal microscopy. Fluorescent images were obtained using a confocal laser scanning microscope (FV1000; Olympus, Tokyo, Japan). Quantification of cells or parameters of interest was conducted in a 1250 × 1250 µm field of view for vessel density and %LV area, and a 200 × 200 µm field of view for counting EdU + cells using FV10-ASW 3.0 Viewer (Olympus). To quantify the fluorescent intensity of indicated antibodies (Tfe3, CD31, Pdpn, Vegfr3, and Lyve1) using the Image J software (NIH, Bethesda, MD, USA), we measured the "Mean" values in randomly chosen four 50 × 50 µm fields of view in each image and were averaged. Blood-filled lymphatics were counted as the number of cross sections of LV tubes containing one or more erythrocyte in the entire embryo or tail section in scanned images. To get the serial images for z-stack slices, multiple slices horizontally imaged from the same field of view at 0.3-μm intervals were captured.
ChIP sequencing analysis. HA-TFE3 inducible HK-2 cell lines were cultured with 200 ng/ml doxycycline for 24 h, and cross-linked with 1% formaldehyde at room temperature for 5 min followed by incubation with 125 mM glycine. Nuclei preparation, chromatin digestion by micrococcal nuclease, chromatin immunoprecipitation with anti-HA antibody (3F10, Roche), and DNA purification were performed with a SimpleChIP Plus Enzymatic Chromatin IP Kit (Cell Signaling) following the manufacturer's protocol. DNA libraries required for high-throughput sequencing were prepared from DNA obtained after ChIP using the NEBNext Ultra DNA Library Prep Kit and NEBNext Multiplex Oligos for Illumina (New England BioLabs). Multiplexed libraries were subjected to cluster generation and sequencing using a NextSeq 500 Kit (75 cycles) with the NextSeq desktop sequencing system (Illumina). Peak detection was performed using the MACS algorithm (PMID:18798982) in Strand NGS software (Strand Life Sciences). HA-TFE3 binding was identified by the significant enrichment of each signal over input-DNA peaks at a P cutoff value of 10 −5 . Visualization of ChIP-seq data was performed using Strand NGS software.
ChIP assay. Approximately 1 × 10 7 HUVECs were fixed in 1% formalin for use in ChIP assays. The fixed cells were sonicated to obtain fragmented chromatin between 200 and 1000 bp in length. An aliquot of sonicated chromatin was put aside to be the input control. ChIP assays were performed using ChIP-IT Express (Active Motif, Tokyo, Japan) with 0.3 mg/ml of anti-TFE3 antibody (1:100, Sigma-Aldrich, HPA023881) or Rabbit IgG as a control. Primers used to amplify a 110 bp product from the human PROX1-0.15 kb region were 5'-TGTGACGTG CAGTCTTCCTG-3' and 3'-CGGCTGCAATGGTGTATTA-5'. A primer set for detecting the human GAPDH promoter region (5'-CCACATCGCTCAGACAC CAT-3' and 3'-CCCGCAAGGCTCGTAGAC -5') was used as a negative control. To quantify the relative abundance of ChIP signals, the intensity of the human PROX1-0.15 kb band was measured using Image J Software (NIH, Bethesda, MD, USA). All primer sequences are listed in Supplementary Data 3.
Luciferase reporter gene assay. The 293GPG packaging cells were transfected with pMY-TFE3GR-ires-GFP plasmid vector. HEK293A cells stably expressing TFE3GR fusion proteins were established by infecting the cells with the viral supernatant from 293GPG packaging cells after doxycycline removal. Transduction of >95% GFP-positive cells was confirmed by fluorescent microscopy. A 2000 bp fragment from the 5' region of the human PROX1 gene was cloned into the pGL4.23 luciferase vector. Two E-box sequences in the PROX1 promoter construct were mutated using a standard mutagenesis protocol and designed PCR primer sets. Reporter plasmids were cotransfected into HEK293A cells along with the phRL-TK vector as an internal control, using PEI-Max (Polysciences, Warrington, PA). The transfection medium was changed to fresh medium either with or without 0.1 μg/ml dexamethasone (DEX) after 24 h, and cultured for 12 h. Luciferase activity was measured and analyzed using the Dual-Luciferase Reporter Assay System, according to the manufacturer's protocols (Promega, Madison, WI).
Human samples. Patients who underwent surgical removal of pulmonary cysts were enrolled in this study based on written informed consents for histological and biochemical analyses. All patients received genetic counseling and were diagnosed as BHD. The study was approved by the Institutional Review Board of Yokohama City University (approval number, A110929001) and Keio University School of Medicine Ethics Committee (approval number, 20170369). The human study was conducted in compliance with the Japanese regulation, Ethical Guidelines for Medical and Health Research Involving Human Subjects, in addition to internationally established principles of the Declaration of Helsinki. Resected lung tissues (n = 3) were fixed with 10% buffered formalin and embedded in paraffin. Cystic lesions and surrounding noncystic pulmonary tissues from patients with BHD (n = 3) and spontaneous bullae (n = 3) were snap-frozen and stored in liquid nitrogen prior to western blotting. Four-micrometer thick paraffin sections were subjected to immunohistochemistry. After deparaffinization and rehydration, sections were autoclaved at 121°C for 15 min.
Statistics and reproducibility. Results are expressed as the mean ± standard deviation (SD). The comparisons between the averages of the two groups were evaluated using the two-tailed Student's t-test. P values of <0.05 were considered statistically significant. For histological analyses, at least three but typically more independent samples were quantified or qualitatively analyzed with each experimental repeat yielding highly similar results.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
For scRNA-seq data, raw data will be available in GEO (GSE133512). For ChIP-seq (Fig. 4p), raw data are available in GEO (GSE135490). There is no restriction on data availability. Source data are provided with this paper.