Tumor Heterogeneity in VHL Drives Metastasis in Clear Cell Renal Cell Carcinoma

To study the impact of intratumoral VHL heterogeneity observed in patient ccRCC primary tumors, we engineered VHL gene deletion in three RCC models, including a new primary tumor cell line derived from an aggressive metastatic ccRCC. The VHL gene-deleted (VHL-KO) cells underwent epithelial-to-mesenchymal transition (EMT) and showed diminished proliferation and tumorigenicity compared to the parental, VHL-expressing (VHL+) cells. Renal tumors with either VHL+ or VHL-KO cells alone exhibit minimal metastatic potential. Interestingly, tumors with both cells displayed rampant lung metastasis, highlighting a novel cooperative metastatic mechanism. The poorly proliferative VHL-KO cells stimulated the proliferation, EMT and motility of neighboring VHL+ cells. We found that periostin (POSTN), a protein product overexpressed and secreted by VHL- cells, promoted metastasis by enhancing the motility of VHL-WT cells and facilitating vascular escape of tumor cells. Genetic deletion or antibody blockade of POSTN dramatically suppressed lung metastases in our preclinical models. Our work suggests a new strategy to halt progression in ccRCC by disrupting the critical metastatic crosstalk between heterogeneous cell populations within a tumor.


Introduction
Renal cell carcinoma (RCC) is the most common cancer of the kidney, consistently ranking amongst the top ten most prevalent malignancies in the world. Annually diagnosed cases exceed 70,000 in the US and 350,000 worldwide 1  Detailed studies of von Hippel Lindau (VHL) disease (a rare hereditary cancer syndrome manifesting as renal, CNS, adrenal, and pancreatic tumors) led to the identi cation and cloning of the VHL tumorsuppressor gene 4,5 . VHL also plays an integral role in the pathogenesis of sporadic, non-familial ccRCC, as somatic mutations of VHL have been reported in as many as 90% of cases [6][7][8][9][10] . Seminal research over the past two decades has unraveled the VHL protein's intricate and important function as an E3 ubiquitin ligase that targets the alpha subunit of hypoxia-inducible transcription factors (HIF-αs) for oxygendependent degradation 11,12 . Although constitutive activation of the HIF pathway due to VHL loss has been implicated as an oncogenic driver 13,14 , the precise oncogenic mechanism of VHL loss remains elusive. Numerous mouse models of renal-tubule-targeted VHL deletion have failed to generate renal lesions beyond preneoplastic cysts, even when combined with deletion of other tumor suppressor genes such as PTEN or p53 [15][16][17][18][19] . It is clear that the loss of VHL function upregulates both HIF1a and HIF2a, however these two paralogs appear to have distinct, and often contrary, roles in their gene regulatory activities 20,21 . Recent research suggests that HIF2α plays an oncogenic role, whereas HIF1α suppresses ccRCC tumors 22 . These opposing oncogenic roles remain an area of active debate [23][24][25] .
The potential contribution of VHL loss and its downstream effectors to metastatic progression is also poorly de ned, as there is no signi cant correlation between VHL mutational status and clinical outcome 26 . We and others have reported that silencing or deleting the VHL gene consistently results in epithelial-to-mesenchymal transition (EMT) [27][28][29] . EMT is an embryonic program used by polarized epithelial cells to break away from cell-cell contacts and basement membrane attachment, enabling cellular migration to distant sites 30 . It is highly reminiscent of the process that carcinomas adopt during metastatic spread 31 , although the direct role of EMT in cancer metastasis is actively debated 32,33 .
Our VHL-deleted ccRCC models revealed a new concept: EMT contributes to metastatic dissemination, but indirectly. Speci cally, inactivation of VHL renders VHLtumor cells highly motile but non- proliferative. Yet, these VHLcells drive metastasis by producing key soluble factors, including periostin (POSTN), which disrupts the vasculature and induces dissemination and growth of VHL-expressing (VHL + ) cells. This cooperative mechanism of metastasis, which occurs as a result of interactions between two distinct cell populations, highlights the active role of intratumoral heterogeneity in cancer aggression. A better understanding of the signals that mediate these interactions could identify novel therapeutic interventions to halt the deadly metastatic spread of ccRCC.

Intratumoral heterogeneity in VHL expression is prevalent in ccRCC
The mutation or inactivation of the VHL gene is recognized as a trunk lesion in the majority of ccRCC cases 9 . However, it is unclear whether VHL loss or inactivation is uniform throughout the entire tumor. VHL protein expression was analyzed by immunohistochemical analyses (IHC) of large tissue sections of a total of 26 ccRCC cases, collected by surgical resection at our institution over the past ve years (Table 1, Supplemental Figure 1A). The most recent 10 cases (#17 to 26) were freshly harvested tumor specimens collected from consecutive surgeries performed by a single surgeon (Table 1 and Supplemental Figure 1B). Intratumoral heterogeneity in VHL protein expression was commonly observed as none of the ccRCC cases displayed a uniform loss of VHL throughout the entire tumor (Table 1 and Supplemental Figure 1A). Representative images of H&E and VHL stains of consecutive cases, #20, 21, 22, and 23, are shown in Figure 1A, 1B, 1C and 1D. Case #20 contained the least amount of VHL positivity of 10%, while case #23 is a benign oncocytoma that is uniformly VHL positive.
To investigate VHL and other oncogenic mutations of ccRCC in detail, we established primary tumor cell lines and patient-derived xenografts (PDXs) in chorioallantoic membrane (CAM) model from freshly harvested surgical specimens (cases #18-26). The rate of PDX engraftment in CAM is e cient at an 80% success rate 34 . Here, we focused on the most aggressive case of metastatic ccRCC, case #22 (Table  1). This patient presented with a 10-cm, Fuhrman grade 4, primary tumor and bilateral lung metastases and succumbed to the disease within one year of nephrectomy despite multiple surgical and pharmacological interventions. As shown in Figures 1C, VHL expression in the primary tumor of #22 was highly heterogeneous, with VHL + cells juxtaposed to VHLcells. The CAM PDXs established from small tumor pieces of the primary tumor from case #22 revealed the presence of intermixed VHL + and VHLtumor cells (Supplemental Figure 1C). We successfully generated a primary tumor cell line from case #22, which consisted an equal mixture of VHL + and VHLtumor cells upon initial tumor dissociation (P0).
However, only the VHL + tumor cells were able to propagate continually, up to passage 20 currently (data not shown). Whole exome sequencing (WES) showed that this #22 cell line shared more than 80% of COSMIC annotated mutations found in the parental tumor (Supplemental Figure 1D). The VHL gene sequence of the #22 cell line possessed a homozygous in-frame T506C transition, leading to a L169P amino acid substitution (Supplemental Figure 1E). L169P is a common variant, with 14 occurrences in 446 patients analyzed 35 . However, it has not been biochemically characterized and its effect on VHL function is unknown. Our preliminary studies showed this L169P variant is comparable to wildtype VHL in its protein stability and in its ability to degrade HIF1a protein in an oxygen dependent manner (Supplemental Figure 1F, 1G, 1H, 1I). Common ccRCC driver mutations (such as VHL, TP53, BAP1, PBRM1, and SETD2) were analyzed by WES of four different tumor areas of the primary #22 tumor and compared to its derivative cell line as shown in Figure 1E. Variant allele frequencies (VAFs) and copy number ratios (CNR) showed clonal missense mutations in the VHL gene and frameshift mutations in the BAP1 gene in the #22 cell line ( Figure 1E). However, these two mutations were subclonal with varying contributions in the four tumor areas, suggesting cellular heterogeneity within the primary tumor.
Next, we analyzed single cell sequencing of primary tumor tissue and dissociated primary tumor cells from case #22. Heterogeneous VHL positivity was con rmed and is shown in Figure 1F and 1G, which cross-validates the VHL staining in Figure 1C demonstrating both VHL positive and VHL negative populations in the primary tumor. Next, we utilized the TCGA database to further investigate VHL heterogeneity. The CNR at the VHL locus was analyzed in the TCGA KIRC (Kidney Renal Clear Cell Carcinoma) cohort (n = 459) after adjusting for both tumor purity and ploidy. The peak CNR value (average for the mixture of cells analyzed) was between -1.1 and -0.4, suggesting that many purity and ploidy corrected TCGA samples had subclonal, single-copy loss of VHL, as a value of -1.1 represents a two-copy loss and -0.4 represents one-copy loss (shown as the dotted lines, Figure 1H). The VAFs of somatic VHL mutations in the TCGA KIRC cohort (n = 148) after adjusting for both tumor purity and ploidy also displayed a wide spectrum spanning between 0.5 and 1, indicating subclonal mutations ( Figure 1I). These results use consensus purity and ploidy estimates from three computational algorithms and IHC analysis (see Methods). We further validated that consistent evidence of subclonal VHL copy number loss and mutation can be found after adjusting with only one of the computational measures of tumor purity (ABSOLUTE) (Supplemental Figure 1J and 1K).
Collectively, IHC and genomic pro ling indicates that VHL protein expression and gene mutation is heterogenous within individual human ccRCC tumors.

Metastasis requires cooperation between VHL-KO and VHL-WT RCC cells
To study the functional interaction between VHL + and VHL -RCC cells, we used the CRISPR/Cas9 system to delete the VHL gene in the murine RCC RENCA (RC) and human ACHN cell line as previously reported 36 and in the primary cell line #22 here. The rst VHL-deleted RENCA line (denoted as RVN) was created by transducing RC cells with VHL-targeted lentiCRISPR (29). RVN cells underwent EMT and developed rampant lung metastases upon intrarenal implantation and were much more aggressive than the parental RENCA cells 36 . Unexpectedly, lung metastases from RVN tumors consisted largely of VHL-expressing cells with minor pockets of VHL-deleted, MMP-9 + cells (Supplemental Figure 2A). This result alerted us to the non-clonal nature of RVN line and that both VHL positive and negative cells are required to produce metastasis. We therefore selected several clones with bi-allelic VHL gene deletion, generated through transient expression of CRISPR/Cas9 as described in Hu et al. 36 . The clonal VHL-knockout line is denoted as RC-VHL-KO, and the parental, VHL + control treated line is denoted as RC-VHL-WT.
Next, we established renal tumors with either RC-VHL-WT cells, RC-VHL-KO cells, or a 1:1 mixture of the two cell lines. The growth and dissemination of these tumors in mice were monitored by bioluminescence imaging (BLI) of the re y luciferase marker gene. RC-VHL-WT and mixed primary tumors grew well, but VHL-KO tumors barely grew (Figure 2A, B). Interestingly, the mixed tumor bearing mice exhibited prominent thoracic metastatic BLI signals (Figure 2A, C). These mice also suffered tumor cachexia with signi cant weight loss (Supplemental Figure 2B). Lung metastasis was not observed in the VHL-KO tumor group (Supplemental Figures 2C). Detailed histological analyses revealed the mixedtumor group exhibited greatly increased numbers and sizes of lung metastases compared to the VHL-WT tumor group, which could be easily appreciated even at low magni cation ( Figure 2D) 36 . The in vivo growth and metastatic behavior of the RC tumors were further veri ed in the avian chorioallantoic membrane (CAM) tumor system 37,38 . The CAM model substantiated the poor growth of VHL-KO tumors in comparison to VHL-WT and 1:1 mixed tumors ( Figure 2E). Although the mixed tumors exhibited an insigni cant reduction in tumor size compared to VHL-WT tumors, they demonstrated a signi cant increase in circulating tumor cells, re ecting the heightened metastatic potential of the mixed tumors ( Figure 2F).
Next, we ascertained whether this novel cooperative mechanism of metastasis could be operating in human ccRCC models. The same CRISPR/Cas9 lentiviral system was employed to knock out the VHL gene in the human ACHN (AC) cell line (29,34), a widely used human RCC cell line known to express wildtype VHL protein 39 . Consistent with the ndings in the RC model 29 , the clonal AC-VHL-KO line exhibited an EMT cellular morphology (Supplemental Figure 2D), elevated expression of EMT markers (Supplemental Figure 2E). Tumors derived from both AC-VHL-WT and 1:1 VHL-WT:VHL-KO cells grew well after intrarenal implantation ( Figure 2G). However, when assessed by gross tissue inspection and detailed histology, lung metastases were observed only in the mixed-tumor group ( Figure 2G). Of note, the AC cells grew more slowly than RC cells in mice, resulting in smaller primary tumors and lung metastases.
The same CRISPR/Cas9 approach was applied to the VHL + primary tumor cells of case #22 to generate a VHL-deleted derivative line, #22 VHL-KO, denoting the engineered knockout of VHL gene. The gene edited biallelic frameshift mutation in #22 VHL-KO cells were con rmed by DNA sequencing (Supplemental Figure 1E). CAM tumors were established for #22 VHL + primary cells, #22 VHL-KO or 1:1 mixture of these 2 cell types (Supplemental Figure 2F). Importantly, the metastatic potential of the #22 CAM tumors, assessed by presence of tumor cells in the chick embryo, was only signi cantly increased in the mixed tumor group ( Figure 2H), paralleling the ndings of RC and AC model. Further analyses of differential single cell gene expression in VHL-KO/VHL + cells for these three models, #22, ACHN and RENCA, showed a congruent 200-geneset pattern of up-and down-regulation for a wide spectrum of genes ( Figure 2I). Similar congruent patterns in functional enrichments, such as sumoylation of transcription factor, and depletions, such as wnt signaling pathway and apoptotic cleavage of cellular proteins, were also observed across ccRCC cells from the different sources ( Figure 2J).
Collectively, the data from the VHL-deleted ccRCC models, #22, ACHN and RENCA, revealed that the cooperative interactions between two distinct populations of tumor cells (VHLand VHL + cells) are required to produce distant metastases.

VHLor VHL-KO cells induce the proliferation of VHL + tumor cells
An immediate question raised by the cooperative metastatic models (Figure 2) is the nature of the crosstalk between the two cell populations leading to induction of metastasis. We and others have reported that VHL gene deletion causes RCC tumor cells to undergo EMT 29 and slow proliferation, as VHL-KO cells derived from the murine RC 36 or the human AC (34) models grew slower than their parental VHL-WT cells in cell culture ( Figure 3A, B) and in vivo (Figure 2A, B, E). The primary cells of #22 also displayed the same growth pattern with the #22 VHL-KO grew signi cantly slower than the parental VHL + cells ( Figure 3C). In transwell co-cultures, the presence of VHL-KO cells induced the proliferation of VHL+ or VHL-WT cells in all 3 models, suggesting the in uence of soluble or paracrine factors ( Figure 3A, D and E). Cellular proliferation assessed by the rate of Ki67 staining was greater than 3 times higher in the VHLpositive (31.4%) than the VHL-negative areas (7.8%) of the 1:1 mixed RC renal tumor ( Figure 3F) and lung metastases (Supplemental Figure 3A). Analysis of human ccRCC tumor #22 revealed the same pattern, with Ki67 positivity rate more than 7 times higher in the VHL-positive (26.6%) than the VHL-negative tumor areas (1.4%) ( Figure 3G). The HALO in ltration analysis module was applied to decipher the spatial relationship between the different cell populations. The highest concentration of Ki-67 + cells resided at the edges of VHL + areas that were in closest proximity to and under the greatest potential paracrine in uence of the VHLcells ( Figure 3H). This is consistent with the in vitro ndings that VHLcells are promoting the proliferation of VHL + cells.
If inducing cellular proliferation of VHL + tumor cells is a critical metastasis-promoting function of the VHLcells, then metastatic lesions would be expected to be dominated by VHL + cells. In fact, this is the precise nding in lung metastases of RC model and case #22. The high prevalence of VHL + cells was observed in large lung metastases ( Figure 3I) and very small metastatic lesions ( Figure 3J) of 1:1 mixed RC tumors, and was assessed to exceed 95% ( Figure 3K), respectively by IHC, immuno uorescent staining (IF) and ow cytometry. The dominant presence of VHL-WT cells in the very small metastatic lesion suggested that the expansion and intravasation of VHL-WT cells occurred early in the metastatic cascade (Supplemental Figure 3B). Paralleling the preclinical scenario, histological analyses of the lung metastases of case #22 showed a high prevalence of VHL + cells, far exceeding the ratio of VHL+/VHLcells in the primary tumor ( Figure 3L, 3M and Table 1).
Collectively, our preclinical and clinical ndings suggest that VHL-KO or VHL-tumor cells drive the cooperative metastasis by inducing proliferation of VHL-WT or VHL+ cells.

VHL-KO cells induce the EMT and the motility of VHL-WT tumor cells
The deletion of VHL gene consistently induced the EMT program in VHL-KO RCC models as previously reported 29 , manifested in mesenchymal cell morphology, elevated EMT gene expression, and a prominent increase in cell motility (29,34). We reasoned that a pro-metastatic paracrine in uence of VHL-KO cells could also induce EMT and motility of the VHL-WT cells. Coculture of RC-VHL-KO cells and RC-VHL-WT cells led to upregulation of EMT markers such as N-Cad, MMP-9, and SMA and downregulation of the epithelial marker E-Cad in the RC-VHL-WT cells ( Figure 4A). We measured the motility of mStrawberry-labeled RC-VHL-WT or EGFP-labeled RC-VHL-KO cells alone ( Figure 4B) or in cocultures ( Figure 4C) using time-lapse live-cell microscopy. VHL-KO cells migrated much faster than VHL-WT cells ( Figure 4D and Supplementary Videos 1A, 1B, 1C, and 1D), and the migration of VHL-WT cells was greatly enhanced in coculture with VHL-KO cells ( Figure 4D). We further performed the migration assay in a 3D system allowing cancer cells to migrate through an extracellular matrix. Paralleling the results of the 2D system, VHL-KO cells also migrated faster than VHL-WT cells in 3D (Supplemental Figure 4A and Supplementary Videos 2A, 2B, 2C, and 2D). Conditioned medium from VHL-KO cells was also able to enhance the motility of VHL-WT cells, albeit to a lesser extent than coculture ( Figure 4E and Supplementary Video 3). Together, these results show that EMT + VHL -KO cells are able to increase the proliferation and the motility of neighboring VHL-WT cells via soluble factors and cell-cell contact, promoting the aggressive, metastatic behavior of the tumor.
Periostin is a soluble factor secreted by VHL-KO cells that promotes metastasis Next, we sought to clarify the signals downstream of VHL loss that could be governing metastasis. RNA sequencing identi ed four HIF1a-regulated genes that are coordinately upregulated upon VHL gene deletion and that predicted a very poor patient survival in the TCGA RCC (KIRC) database 29 . Amongst the four identi ed HIF1a-regulated genes, we focused on POSTN because it encodes a secreted cell-adhesion protein upregulated in EMT with known paracrine activity that confers aggressive and metastatic behavior 40,41 . POSTN is upregulated in kidney cancer 42 and is a poor prognostic indicator for RCC (Supplemental Figure 5A). However, the functional role of POSTN in RCC tumorigenesis has not been de ned. POSTN was upregulated in RC-VHL-KO relative to RC-VHL-WT cells at the RNA (Supplemental Figures  The mechanism of action of POSTN was further investigated with add-back experiments. The addition of recombinant POSTN to VHL-WT cells signi cantly promoted their motility, which was blocked by the cyclic-peptide integrin inhibitor, cilengitide ( Figure 4I and Supplemental Figure 5D). The addition of recombinant POSTN to VHL-WT cells activated focal adhesion kinase (FAK) via phosphorylation at Tyr 397, which was blocked by cilengitide ( Figure 4J). In RC-VHL-WT and RC-VHL-KO coculture, cilengitide disrupted the enhanced motility of VHL-WT cells in a dose dependent manner ( Figure 4K and Supplementary Videos 7A, 7B, 7C and 7D). These ndings from our preclinical model suggested that POSTN secreted by the VHL-KO cells could be an important mediator of the cooperative metastatic mechanism. Hence, it is critical to verify the correlation of VHL loss with the upregulation of POSTN expression in ccRCC human tumors.
As shown in Figure 5A, POSTN colocalized with RC-VHL-KO cells but not with RC-VHL-WT cells in serial sections of a large metastatic lesion from a mixed RENCA tumor. IHC of VHL and POSTN in serial sections of primary tumor of case #22 ( Figure 5B) showed that VHL + areas were POSTNwhereas VHLareas stained positive for POSTN ( Figure 5C). Furthermore, quantitative analysis of VHL + POSTNand VHL -POSTN + cells in this case #22 using the HALO image analysis software con rmed the distinct spatial distribution of these two populations (Figures 5D). IF staining of the lung metastatic lesions of case #22 clearly showed the predominance of VHL + cells (red uorescence), which were excluded from the POSTN + cells (green uorescence, Figure 5E). This reciprocal relationship was further con rmed in the retroperitoneal lymph node metastasis of case #17 ( Figure 5F, Table I). The increased RNA expression of POSTN upon VHL knockout in #22 primary cells was further validated by qRT-PCR ( Figure 5G). Analysis of the VHL and POSTN expression pattern in a tissue microarray (TMA) constructed from over 300 ccRCC patients who underwent nephrectomy at our institution 44 , con rmed the reciprocal relationship in approximately 30% of cases (Supplemental Figure 5E). The small sampling areas of TMA rendered the majority of the samples not informative.
The spatial arrangement of VHL + and VHLpopulations and their relationship to POSTN expression in clinical specimens recapitulated the ndings in our preclinical metastatic ccRCC model ( Figures 1C and   5B and Supplemental Figures 2A, 5F). These ndings suggest that POSTN secreted by VHLcells could be a key paracrine metastatic mediator in clinical disease.

VHL-KO cells and POSTN cause vascular destruction enhancing intravasation
Detailed analysis of tumor-cell invasion into the circulation (i.e., circulatory tumor cells [CTCs]) in our metastatic preclinical model revealed that the presence of VHL-KO cells in mixed primary renal tumors enhanced the number of VHL-WT cells escaping into the circulation (Supplemental Figure 6A, 3C). These results suggested that VHL-KO cells could also exert an in uence on vascular intravasation step of metastasis by disrupting the endothelial cell barrier as described in a recent study by Strilic and colleagues 45 . To examine the possible role of this mechanism in our tumor model, we established a 3D intravasation assay with a layer of either VHL-WT, VHL-KO, or a 1:1 mixture of the two placed above a layer of human umbilical vein endothelial cells (HUVEC), separated by a thin layer of Matrigel. After 48 hours, the integrity of the endothelial cell layer was tabulated, which showed signi cant endothelial cell destruction when cocultured with either VHL-KO or the 1:1 mixture but not with VHL-WT cells alone ( Figure 6A). The molecular signals involved in the destruction of endothelial cells by the VHL-KO cells was further investigated. A wide range of tumor cell models are reported to induce necroptosis in endothelial cells, but RCC models have not been investigated 45 . Thus, we examined if our RC cells were able to induce either necroptosis or apoptosis in HUVEC endothelial cells. Coculturing with RC-VHL-WT or RC-VHL-KO cells did not activate the necroptosis markers MLKL or RIP in HUVECs ( Figure 6B, C). However, RC-VHL-KO cells induced a robust apoptotic response in endothelial cells, as indicated by cleaved caspase 3 ( Figure 6B) and apoptosis reporter assays ( Figure 6D). Importantly, the anti-POSTN blocking antibody blunted the HUVEC cell apoptosis induced by VHL-KO cells (Figures 6B and 6D). We further assessed endothelial destruction and vascular leakage in vivo in the CAM tumor system with the Miles assay. As shown in Figure 6E, the vasculature of the mixed tumors was leakier than that of the VHL-WT tumors.
In total, these results highlight that the POSTN protein secreted by VHL-KO cells could augment the escape of tumor cells into the blood circulation by destroying the endothelial cell barrier.

Inhibition of POSTN blocks metastasis in ccRCC models
Given the multiple paracrine in uences exerted by POSTN on the metastatic cascade, we assessed whether blocking POSTN with genetic and pharmacological approaches could suppress metastasis in vivo. The RC-VHL/POSTN-KO double-knockout cell line exhibited reduced promotion of VHL-WT cell motility in coculture ( Figure 4G). Renal tumors containing 1:1 mixtures of VHL-WT cells and VHL/POSTN-KO cells showed reduced lung metastasis comparing mixed tumors with VHL-WT and VHL-KO cells, as measured by BLI ( Figure 7A). We further determined if the anti-POSTN-blocking antibody could therapeutically inhibit the development of lung metastasis. As shown in Figures 7B-E, treatment with anti-POSTN antibody MPC5B4 greatly suppressed lung metastasis as assessed by gross lung morphology ( Figure 7B), lung weight ( Figure 7C), histological assessment by H&E ( Figure 7D) and IF staining ( Figure 7E). The POSTN blocking treatment did not impact primary tumor growth signi cantly ( Figure 7F). We further assessed the anti-POSTN antibody treatment on the CAM PDX mixed tumor with 1:1 VHL+ and VHL-KO cells of case #22. Administration of MPC5B4 to the CAM PDX mixed tumors signi cantly reduced tumor cell dissemination into the chick embryo liver ( Figure 7G and 7H) without signi cant reduction in primary tumor growth ( Figure 7I). This nding fully recapitulated the therapeutic impact of MPC5B4 in the RC preclinical model. This showed that POSTN is likely a critical paracrine factor secreted by VHL-KO cells that exerts multiple prometastatic in uences, such as the EMT and motility in VHL-WT cells and the destruction of adjacent blood vessels ( Figure 8A). Thus, targeted blockade of POSTN appears to be a promising approach to inhibit the deadly metastatic process in ccRCC.
Taken together, our VHL-deleted ccRCC models reveal a novel metastatic mechanism that relies on cooperative interactions between two distinct populations of tumor cells: VHL-KO (VHL -) and VHL-WT (VHL + ) cells ( Figure 8B). VHL-KO cells displayed an EMT and highly motile phenotype, but grow poorly in vivo. However, VHL-KO cells induced an aggressive behavior in VHL-WT cells that promoted rampant lung metastases composed predominantly of VHL-WT cells. Many of the phenotypes observed in our preclinical model were consistent in clinical tumor specimens, in particular, the loss of VHL resulted in the upregulation of POSTN. These results lend credibility to a cooperative mechanism of metastasis operating in human ccRCC. The discovery of the novel cooperative metastatic mechanism and the demonstration of the critical metastatic cross talk open up novel therapeutic avenues to control metastases in ccRCC.

Discussion
In RCC and other epithelial cancers, metastasis is the major cause of mortality. The complex nature of metastasis coupled with an incomplete understanding of its mechanisms pose signi cant challenges to devising effective treatments. For the last three decades, the progression model has been the most common, prevailing concept for how metastasis occurs 46 . This model postulates that multiple progressive mutational events occur to enable a small fraction of cells to acquire full metastatic potential. Subsequent studies showed that clonal evolution and selection can enhance not only metastatic potential but also achieve metastatic site speci city 47,48 . The cooperative model of metastasis uncovered here proposes a distinctly different mechanism in which signals between two populations of tumor cells, rather than clonal progression of a single population, is needed to achieve metastasis.
In the three metastatic ccRCC models reported here, the prerequisite prometastatic interactions occur between VHL-KO (VHL -) and VHL-WT (VHL + ) cells. Intriguingly, EMT + VHL-KO cells are themselves poorly proliferative, but serve as the metastatic driver to induce aggressive behavior in the normally nonmetastatic VHL-WT cells. We further found that the loss of VHL function upregulates HIF1a, which in turn stimulates the production of POSTN. POSTN serves as the critical, paracrine, metastasis-promoting factor by not only inducing EMT and motility in neighboring VHL-WT cells, but also causing vascular destruction to facilitate the escape of tumor cells into the circulation. Given that POSTN can impact the metastatic process in multiple ways, we showed that blocking POSTN's function can, indeed, halt lung metastasis in our models. The clinical relevance of the cooperative mechanism of metastasis is supported by the fact that human tumor samples consistently showed intratumoral heterogeneity with VHLand VHL + tumor cell clusters in individual cases of ccRCC. We further demonstrated that POSTN overexpression is observed in VHL-nonexpressing areas of ccRCC tumors similar to our preclinical model.
The identi cation of the direct and instrumental role of POSTN in metastasis is a novel and signi cant nding. POSTN, also known as osteoblast-speci c factor 2, is a ubiquitous, secreted, stromal protein that promotes integrin-dependent cell adhesion and motility during bone and cardiac development 49,50 .
POSTN overexpression is observed under EMT and hypoxic conditions 50,51 , the conditions of our VHL-KO cells. POSTN was reported to bridge the colonization of breast cancer cells to their terminal lungmetastatic site 41 , placing POSTN's involvement at the distal end of the metastatic cascade. This result differs from ours in that POSTN acts at the tumor-proximal intravasation step. We have not fully investigated POSTN's role in the latter steps of the metastatic cascade in our ccRCC models. An anti-POSTN neutralizing antibody has also been shown to be effective in blocking metastatic progression in an ovarian cancer model 52 . As such, the clinical applicability of a POSTN-targeted therapeutic approach to block metastasis clearly warrants further investigation. The cooperative "team work" concept between distinct populations of tumor cells to advance the disease has also been reported in a recent study of breast cancer 53 . In this study, overexpression of EMT transcription factors that activate Hedgehog/GLI signaling promoted aggressive behavior in non-EMT cells in a paracrine manner. The cooperation between EMT and non-EMT cells reported is highly reminiscent of the crosstalk in our VHL-KO and VHL-WT model, but with POSTN as the functional mediator in our model. The cooperative, metastatic model proposed here and by others 53 emphasizes the need for different strategies to search for novel therapeutic targets.
Intratumoral heterogeneity in gene expression is widely recognized in ccRCC, and high-power gene sequencing technologies and bioinformatics have been applied to the study of this disease 54,55 . Despite these advances in the study of ccRCC, untangling the signaling pathways to distill the key cross talk signals remains very challenging. A potentially fruitful approach could be to rst separate the distinct populations based on cellular morphology or protein biomarker(s) and then interrogate the expression in the distinct populations. Furthermore, in pathways such as VHL and HIFa that involve extensive posttranscriptional regulation of protein stability, it would be prudent to integrate protein expression with gene-expression analyses to gain a comprehensive view of the tumor biology. In sum, the study reported here provides an alternate idea of how the complex task of metastasis can be achieved by a heterogeneous tumor such as ccRCC. This cooperative model can guide the search for better and more effective treatments to block metastases and address a clearly unmet need in the eld of cancer research.

Cells, plasmids, and reagents
The RENCA (RC) cell line was purchased from ATCC and was maintained in RPMI-1640 supplemented with 10% fetal bovine serum and 1X penicillin/streptomycin (Thermo Fisher, CA, USA, catalog number: 15140122). All CRISPR/Cas9-mediated knockout RC cell lines were selected with puromycin and clonally puri ed via single-cell cloning in a 96-well plate. A lentiviral vector encoding HA-tagged mStrawberry (modi ed from pSicoR, Addgene, MA, USA, catalog number: 11579) was used to label RC-VHL-WT cells, while a vector with the same backbone encoding ag-tagged EGFP was used to label RC-VHL-KO, RC-VHL/HIF1A-KO, and RC-VHL/POSTN-KO cells. In addition, for in vivo studies, all cell lines were also marked with lentivirus expressing re y luciferase to permit BLI. pGL3-basic was from Promega (CA, USA, catalog number: E1751) and was enzymatically digested with MluI and XhoI. The periostin promoter was cloned from the genomic DNA of RC cells with the following primers: forward -CGACGCGTTAAGGTGGACAGTGAGGAAGACACA, reverse -CCGCTCGAGTTGAGAAGAACGAGAGTAGAGATTTTAGG. The control renilla luciferase vector was pRL-TK from Promega (CA, USA, catalog number: E2231). The plasmid for overexpressing constitutively-active HIF1A was from Addgene (MA, USA, catalog number: 44028).
Time-lapse microscopy for 2D scratch assay and 3D migration assay A total of 1x10^5 tumor cells (e.g., 5x10^4 cells each of VHL-WT and VHL-KO cells) were grown on a 24well plate until reaching 90% con uence. The bottom of each well was scratched with the end of a 200 µl tip to form a gap. The cell migration was monitored continuously with a Nikon Eclipse Ti-E time-lapse microscope using a 10X objective, and a humidi ed, 37 °C environment containing 5% CO 2 . Speci c elds of interest were set and recorded at 15 minutes intervals for 20 hours using the FITC and TRITC channels.
Nikon elements software was used to measure the migration speed of cells in each group.
Transwell chambers (0.4 μm pore size, Thermo Fisher, catalog number: CLS3470-48EA) were assembled in a 24-well plate. One milliliter of RPMI-1640 medium supplemented with 10% fetal bovine serum and 50 Microscope images were taken of ve random elds of each well with a 10X objective in DAPI and TRITC channels and quanti ed with ImageJ.
For apoptosis evaluation, HUVECs were cultured in Transwell chambers as described above. After 48 hours, the plates were equilibrated at room temperature for 10 minutes, and 200 μL of Caspase-Glo 3/7 reagent (Promega, catalog number: G8090) was added to each well. After being placed on a shaker at 300-500 RPM for 30 seconds, the reaction was incubated at room temperature for 1 hour and then analyzed for luminescence with a Synergy HT microplate reader (BioTek).

Cell proliferation assay
Cell proliferation was measured using the MTS assay and direct cell counting. For both assays, cells in log phase were counted and seeded on day 0 at a density of 1000 cells per well onto a 96-well plate, or 500 cells per well onto a 384-well plate. For the MTS assay, cell numbers were evaluated every 24 hours on days 1, 2, 3, 4, 5, and 6 using the MTS kit (Promega, CA, USA, catalog number: G3582) and measured with a Multiskan MK3 microplate reader (Thermo, USA). For direct cell counting, an ImageXpress workstation was used to photograph each well of a 384-well plate and count the DAPI-stained cells. CAM tumor xenograft model, renal tumor implantation, and anti-periostin treatment studies in mice.
Establishment of CAM tumor xenografts and their analyses were performed as previously described 34,56,57 . Intrarenal implantation of 1x10^6 total RC or AC tumor cells was performed as previously described 29,36 . One week after implanting a 1:1 mixture of VHL-WT to VHL-KO cells, 10 mg/kg of MPC5B4 mAb was injected via tail vein three times per week for 4 weeks. The animals were imaged and sacri ced. Tissues were harvested, xed, para n embedded, and cut for histological analyses.

IHC and IF staining
Slides were baked at 65 °C for 20 minutes and depara nized through three 10min incubations in xylene then rehydrated in stepwise dilutions of ethanol from 100% to 50% followed by water. Citrate buffer was used for antigen retrieval in a vegetable steamer for 25 minutes. Blocking used 1% BSA, and the In addition, spatial quanti cation analyses for slide case #22 were performed using HALO TM Image Analysis program by Indica Labs (USA). An initial positive stain tissue marker analysis was conducted with proper nuclear segmentation and dye threshold intensities. Under HALO TM 3.0 Spatial Analysis Module, in ltration and density heat map algorithms were used to establish spatial relationships important for VHL-POSTN paracrine crosstalk.

Flow cytometry
Primary tumors and lungs of mice were dissected, minced into small pieces, and digested with 0.2% collagenous II at 37 °C on a 100 RPM shaker. The cell suspensions were passed through 70-µm cell strainers. The digested cells were stained with Hoechst 33342 for 15 minutes and analyzed by ow cytometry. Similarly, chicken and mouse blood were collected and lysed with red blood cell lysis buffer (BD Bioscience, USA, catalog number: 555899). Cells were then analyzed by ow cytometry for mStrawberry and EGFP expression.
Isolation and cultivation of primary ccRCC tumor cells.
With the consent of patients, primary ccRCC tumor samples were collected and chopped into pieces with sterile scissors and scalpels into RPMI-1640. Tissue chunks were transferred to a 15-mL conical tube and centrifuged at 300 x g at room temperature for 5 minutes. The supernatants were carefully discarded and the tissue pellet was resuspended in 2.6-mL prediluted 3 U/L Liberase TM (Sigma-Aldrich, catalog number: 5401119001) in RPMI-1640 media. The tube was placed on a100 RPM rotator at 37 °C for 1 hour. When the tissue was fully digested and no chunks were visible, cells were centrifuged at 300 x g at room temperature for 5 minutes. The pellet was further treated with prediluted 1X red blood cell lysis buffer (BD Biosciences, catalog number: 555899) in sterile water for 15 minutes and washed once with PBS. Cells were resuspended in RPMI-1640 supplemented with 10% fetal bovine serum and 1X penicillin/streptomycin and cultured in a humidi ed, 5% CO 2 incubator at 37 °C.

Human ccRCC patient specimens
The tissue microarray was constructed from a cohort of 357 patients who underwent nephrectomy for sporadic RCC at UCLA between 1989 and 2000, as previously described 44 . Clinical data, including age, gender, Eastern Cooperative Oncology Group performance status, and pathologic data (including tumornode-metastasis stage, histologic subtype, and Fuhrman grade) were collected for each case. This study was approved by the UCLA Institutional Review Board. include cells with total counts greater than 2,000, total number of expressed genes greater than 1,000, and percentage of mitochondrial reads less than twenty percent. Unnormalized raw count data was input to scImpute v0.0.9 for identi cation of read dropout events and imputation 67 . Count matrices were then processed using the Seurat v4 work ow for data transformation and scaling, dimensionality reduction through uniform manifold approximation and projection (UMAP), and differential expression through the MAST package [68][69][70] . Geneset enrichment analysis (GSEA) was then carried out using FGSEA v1.14.0 on a gene list pre-ranked with signed log10 p-values from the differential expression results 71 73 . Reads were selectively aligned to the GENCODE vM25 mouse reference transcriptome with corrections for sequence-speci c and GC content biases. Gene count data were then processed using the DESeq2 v1.22.2 package for normalization and differential expression analysis 74 . As with single cell data, GSEA was carried out using FGSEA v1.14.0 on pre-ranked gene lists. For RENCA models the signed

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.