PCK1 and DHODH drive colorectal cancer liver metastatic colonization and nucleotide biosynthesis under hypoxia

Colorectal cancer (CRC) is a major cause of human death. Mortality is primarily due to metastatic organ colonization, with liver being the primary organ affected. We modeled metastatic CRC (mCRC) liver colonization using patient-derived primary and metastatic tumor xenografts (PDX). Such PDX modeling predicted patient survival outcomes. In vivo selection of multiple PDXs for enhanced metastatic capacity upregulated the gluconeogenic enzyme PCK1, which enhanced metastatic hypoxic survival by driving anabolic pyrimidine nucleotide biosynthesis. Consistently, highly metastatic tumors upregulated multiple pyrimidine biosynthesis intermediary metabolites. Therapeutic inhibition of the pyrimidine biosynthetic enzyme DHODH with oral leflunomide substantially impaired CRC liver metastatic colonization and hypoxic survival. Our findings provide a potential mechanistic basis for the epidemiologic association of anti-gluconeogenic drugs with improved CRC metastasis outcomes, reveal the exploitation of a gluconeogenesis enzyme for pyrimidine biosynthesis during hypoxia, and implicate DHODH and PCK1 as metabolic therapeutic targets in colorectal cancer metastasis.

gluconeogenic drugs with improved CRC metastasis outcomes, reveal the exploitation of a gluconeogenesis enzyme for pyrimidine biosynthesis during hypoxia, and implicate DHODH and PCK1 as metabolic therapeutic targets in colorectal cancer metastasis.

Introduction
Colorectal cancer (CRC) is a leading global cause of cancer-related death. An estimated 145,000 Americans will be diagnosed with CRC and roughly 51,000 will die from CRC in 2019 (1). The majority of these deaths are due to distant metastatic disease with the liver being the most common distant organ colonized (1). While the prognosis of patients diagnosed with non-metastatic or locoregional disease is relatively better, a significant fraction of such patients will nonetheless experience subsequent metastatic progression. A key clinical need is identifying which patients will develop metastatic disease since few prognostic indicators of disease progression exist.
Additionally, novel targeted therapies for the prevention and treatment of metastatic CRC are a major need.
While cell lines established from human CRC have provided important insights into the biology of CRC, tissue culture drift, adaptation, and evolution could restrict their relationship to the pathophysiology of patient tumors (2). Patient-derived xenograft (PDX) models, which allow for the growth of patient tumor samples in immunodeficient mice, can capture the endogenous diversity within a tumor as well as the patient-topatient variability of metastatic cancer. Prior work has revealed that breast cancer, melanoma and non-small cell lung cancer PDX modeling can predict human clinical outcomes (3,4). These models have revealed the potential of individualized mice models for prognostication and tailoring of therapies. Past colorectal cancer PDX models were subcutaneous implantations (5)(6)(7). Such subcutaneous PDX tumors are useful for tumor growth studies, but do not metastasize and are not exposed to the pathophysiologically relevant and restrictive conditions of the hepatic microenvironment.
While orthotopic PDX implantation serves as a good model for studying metastasis in melanoma and breast cancer, this approach has limited feasibility in colorectal cancer, as the orthotopic tumor kills the host due to the obstructive size that implanted tumors reach prior to liver metastatic colonization (3). Thus, a clinically relevant PDX model of CRC that models the critical process of liver metastatic colonization is an important need.
Beyond predicting clinical outcomes and therapeutic responses, PDX models of mCRC could allow for the identification of molecules that contribute to metastatic colonization through the use of in vivo selection. This process subjects a parental population (e.g. cancer cells, bacteria, yeast) to a severe physiologic bottleneck such that only those cells with the requisite gene expression states survive (8). Such selection is repeated iteratively to enrich for cells that are best adapted to the new microenvironment.
By engrafting patient-derived CRC primary and metastatic tumors of diverse mutational backgrounds, we observed that subcutaneous tumor engraftment efficiency and liver colonization capacity, but not subcutaneous tumor growth rate, was associated with patient overall survival. By performing liver-specific in vivo selection with CRC PDXs, we enriched for cells optimized for growth in the liver microenvironment. Gene expression analysis revealed the gluconeogenesis enzyme PCK1 to be a robust driver of liver metastatic colonization that is over-expressed in metastatic colorectal cancer.
Mechanistic studies revealed that PCK1 enhances anabolic pyrimidine nucleotide biosynthesis, which enables cancer cell growth in the context of hypoxia-a key feature of the liver microenvironment. Consistent with these observations, molecular and pharmacologic inhibition of PCK1 or the pyrimidine biosynthetic gene DHODH inhibited colorectal cancer liver metastatic colonization.

Liver growth and engraftment rates of CRC PDXs predict patient outcomes
In order to establish a PDX model of CRC liver metastatic colonization, a small sample of colorectal cancer tissue, taken either from a primary or metastatic site, was dissociated and injected subcutaneously into the flanks of NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ (Nod-Scid-Gamma; NSG) mice within two hours of surgical resection at MSKCC. Thirty-one subjects provided forty tumor samples; 48.3% of the subjects' samples engrafted (Suppl. Table 1). The majority of subjects in this study were classified as Stage IV colorectal cancer according to the American Joint Committee on Cancer (AJCC). However, AJCC stage was not associated with increased xenograft engraftment (p = 0.35; χ 2 test). The engraftment rates for tumor tissues that originated from the colon and the liver were similar (40% vs 37.5%; Suppl. Table 1 We found that subcutaneous tumor engraftment was associated with worse patient survival (p=0.045; Fig.1A). The time from subcutaneous tumor implantation to tumor harvest ranged from 35 to 88 days (Fig.1B). Among those CRC tumors that did grow subcutaneously, the time to reach the pre-determined tumor size (1,000 mm 3 ) was not significantly associated with patient survival (p=0.27; Fig.1C). When the estimated subcutaneous tumor volume reached 1,000 mm 3 , the mice were euthanized, and the tumors were removed. For each sample, a portion of the xenografted tumor was set aside for cryopreservation, and the rest of the tumor was dissociated into a single cell suspension for portal circulation injection via the spleen. Portal circulation injection has been demonstrated to be a reliable means of establishing liver growth via hematogenous spread of CRC cells, simulating the entry of cells into the portal circulation which is typical of clinical CRC progression. After injection of cells, we observed the mice until they were deemed ill by increased abdominal girth, slow movement, and pale footpads, at which point we proceeded to euthanization and tumor extractions. Successful liver metastatic colonization was achieved upon injection of 15/17 patient samples. The time to mouse sacrifice for the CRC patient-derived liver xenografts ranged from 51 to 407 days (Fig.1D) and did not correlate with subcutaneous tumor growth rates (R 2 =0.046, p=0.50; Fig.1E). The mCRC liver PDXs fell into two biologically distinct groups based on their growth rates: one set grew quickly, requiring mouse euthanasia within three months of implantation; the other set grew more slowly, requiring animal sacrifice after six months, or even one-year, postengraftment (Fig.S1A). Importantly, these two groups of PDXs exhibited similar growth rates when implanted subcutaneously (p=0.09; Fig.S1B). This suggests distinct selective pressures for PDX growth existing in the liver relative to the subcutaneous microenvironment. We found that the liver colonization model mimicked clinical outcomes, as patients whose xenografts rapidly colonized mouse livers fared poorly relative to patients whose xenografts colonized the liver slowly or not at all (p=0.031; Fig.1F). Taken together, these results establish that the CRC liver metastasis PDX modeling described above is prognostic of clinical outcomes.
A key objection to cell line xenografts is that the histology of animal tumors is often not representative of clinical sample histology. Contrary to this, we observed that both subcutaneous and liver engrafted tumors re-capitulated the architecture of the primary tumor from which they were derived (Fig.S2). CLR4 was established from a poorly differentiated liver metastatic colon adenocarcinoma; it remained poorly differentiated in both the subcutaneous and liver xenografts (Fig.S2). Similarly, CLR32 and CLR28 were derived from moderate-to-well differentiated primary colon and peritoneal metastatic adenocarcinomas and retained their moderate-to-well differentiated histology when passaged subcutaneously and hepatically (Fig.S2).

Generation of in-vivo selected highly liver metastatic PDXs
We next performed liver-directed in vivo selection through iterative splenic injections of four distinct CRC PDXs with varying mutational and metastatic backgrounds to obtain derivatives with increased capacity for liver colonization and growth (Fig.2). Tumors were only passaged in vivo without the use of in vitro culture. When a mouse bearing a liver colonization graft had met its pre-determined endpoint, it was euthanized and the liver tumor was removed and dissociated into a single cell suspension in a similar manner to that of the subcutaneous tumors described above. Dissociated cells were subsequently injected into the spleen of another mouse to generate a secondgeneration liver metastatic derivative. This process was repeated multiple times to create a highly metastatic derivative for each of the four distinct CRC PDXs. The number of rounds of in vivo selection varied between tumor samples (range: [5][6][7][8][9][10][11][12][13] and in general tended to represent the number of rounds required to plateau enhanced metastatic colonization capacity. In the last round of in vivo selection, a cohort of mice was subjected to portal circulation injection with either the parental CRC PDX cells or the liver-metastatic derivative CRC PDX cells in order to assess the relative liver colonization capacities among the liver-metastatic derivatives. In each of the four CRC PDX comparisons, the in vivo-selected CRC PDX liver metastatic derivatives colonized the mouse liver more efficiently than their parental counterparts (Fig.2). The two extreme isogenic populations of each patient, the parental CRC PDX and its livermetastatic derivative, were then subjected to transcriptomic and metabolite profiling as described below to identify candidate regulators of metastatic colonization.

Candidate metastasis promoting genes identified through transcriptomic profiling of metastatic CRC PDXs
We first sought to identify candidate mCRC liver colonization promoters through mRNA sequencing and differential gene expression analyses from parental CRC PDXs (anatomical locations included subcutaneous graft, cecal graft, or first-generation liver graft) and last generation liver-metastatic CRC PDXs. Comparisons between liver metastatic derivatives and their parental counterparts allowed for isogenic comparisons.
A phylogenetic tree using complete clustering and Euclidian distance function based upon the gene expression profiles demonstrated that isogenic pairs mostly clustered together with one exception (Fig. S3A).
Using gene set enrichment analysis (GSEA) (14), we interrogated each of the four pairs of tumors individually and as a composite to identify cancer-related pathways and signatures that were significantly altered in liver metastatic derivatives compared to their isogenic parental xenografts. The hypoxia signature was found to be upregulated in all of the liver metastatic derivatives individually and in the composite, where it was the most significantly enriched gene signature (normalized enrichment score (NES)=2.12, q-value<0.001; Fig.S3B). Upregulation of hypoxia genes in the liver metastatic derivatives is consistent with previous reports demonstrating that hypoxia exerts selective pressure in the liver metastatic microenvironment (13,15).
With each CRC PDX pair, we identified upregulated genes in each liver-metastatic derivative compared to its parental counterpart through a generalized linear model. The number of upregulated genes (p<0.05) in the liver-metastatic derivatives ranged from 200 (CLR28) to 345 (CLR27) out of a possible list of more than 12,000 genes. Fisher's combined probability test was used to construct a list of candidate liver colonization promoting genes that were statistically significantly upregulated across the four pairs of CRC PDXs with an effect size of greater than 1.5 log2 fold change (logFC). Using this approach, we identified 24 highly up-regulated genes in the liver metastatic derivatives ( Fig.S3C), with the ten most highly up-regulated genes annotated on the volcano plot ( Fig.3A). Interestingly, two of the top ten up-regulated genes (IFITM1, and CKB) have been previously implicated as promoters of colorectal cancer metastasis (13,16). The most common 'druggable' targets for cancer therapeutics are enzymes and cell-surface receptors. In the list of candidate genes, three were enzymes (ACSL6, CKB and PCK1) and one was a cell-surface receptor (CDHR1).
One of the genes on this list, creatine kinase-brain (CKB), was identified by us in a prior study using established colorectal cancer cell lines and shown to regulate tumoral phosphocreatine and ATP levels in the hypoxic microenvironment of the liver (13). Of the remaining three enzymes on our list, we chose to focus below on evaluating the role of PCK1 (phosphoenolpyruvate carboxykinase 1) given the availability of a pharmacological inhibitor and its heightened expression in normal liver (17), suggesting potential mimicry of hepatocytes by colorectal cancer cells during adaptation to the liver microenvironment.
We next investigated whether our 24-gene candidate CRC liver colonization signature was enriched in liver metastases from patients with colorectal cancer by querying a publicly available dataset in which transcriptomes of primary CRC tumors and liver metastases were profiled. Of the twenty-four genes, twenty-two were represented in this previously published dataset (18). We binned the patient data based upon differential gene expression in the primary CRC tumors versus the CRC liver metastatic tumors.
The up-regulated genes were significantly enriched (p=0.007) in the bin with the most upregulated genes in CRC liver metastases (Fig.3B) (19), supporting the clinical relevance of our in vivo-selected CRC PDX liver colonization mouse model. In further support of the clinical relevance of our findings, we found that the gene expression upregulation in our metastatic CRC system significantly correlated (rho=0.39, p=0.047) with the gene expression up-regulation in human liver CRC metastases relative to CRC primary tumors (Fig.3C). Interestingly, PCK1 was highly up-regulated in human CRC liver metastases relative to primary tumors. QPCR quantification confirmed PCK1 gene expression up-regulation in liver metastatic derivatives relative to isogenic parental counterparts (Fig.3D). We analyzed other publicly available colorectal cancer gene expression datasets and consistently observed PCK1 to be significantly upregulated

PCK1 promotes colorectal cancer liver metastatic colonization
We next performed functional in vivo studies using human colorectal cancer cell lines in which PCK1 expression was modulated through stable gene knockdown or overexpression. Depletion of PCK1 in SW480 cells by two independent shRNAs significantly impaired (p<0.0001 in both comparison) colorectal cancer liver metastatic colonization of cells introduced into the portal circulation of NSG mice (Fig.4A). PCK1 depletion in another colorectal cell line (LS174T) also significantly decreased (p<0.0001) liver metastatic colonization (Fig.4B). Conversely, PCK1 over-expression in SW480 cells significantly increased (p=0.003) liver metastatic colonization (Fig.4C). In contrast, PCK1 depletion did not impact subcutaneous tumor growth in the SW480 or LS174T cell lines (Fig.4D). To assess whether PCK1 modulation regulated cancer progression in a fully immunocompetent model as well, we depleted PCK1 in the murine colorectal cancer cell line CT26. Consistent with our observations in human cancer lines, PCK1 depletion decreased murine colorectal cancer cell liver colonization in an immune competent model (p=0.039 and p=0.005 for shCTRL vs shPCK1064, shCTRL vs shPCK1-66, respectively) and did not impair in vitro proliferation under basal cell culture conditions (Fig.4E).
We next sought to determine the cellular mechanism by which PCK1 impacts metastatic colonization; that is, whether PCK1 influences initial colorectal cancer cell liver colonization, apoptosis, or population growth. To identify whether initial liver colonization was the sole step in the metastatic cascade influenced by PCK1 or whether it could provide continued impact on colorectal cancer liver growth, we generated SW480 cells expressing an inducible PCK1 shRNA (Fig.S4A). Four days after portal systemic injection of cancer cells, at which time CRC cells have extravasated into the liver and begun initial outgrowth, we began administering a doxycycline or a control diet. We found that even after the initial liver colonization phase (days 0-4), PCK1 depletion continued to impair (p=0.004) colorectal cancer metastatic liver growth ( Fig.S4B-C). We did not observe increased apoptosis using the caspase 3/7 reporter in both PCK1 depleted cell populations in vivo (Fig.S4D). Evaluation of the in vivo growth rate through natural log slope calculations demonstrated that in each PCK1 modulation experiment, either knockdown or overexpression, in which a luciferase reporter was used, the rate of growth after the first measured time point (day 4-7) did not equal the rate of growth of the controls (Fig.S4E). These results reveal that PCK1 promotes the rate of metastatic growth in vivo.

Metabolic profiling reveals PCK1-dependent pyrimidine nucleotide biosynthesis in CRC under hypoxia
Given our identification of PCK1 as a metabolic regulator of CRC liver metastatic colonization as well as the enrichment of a hypoxic signature by GSEA in highly metastatic PDXs, we speculated that perhaps PCK1 promotes metabolic adaptation  (23,24). In vivo selected cancer cells can alter cellular metabolism in order to better respond to the metastatic microenvironment (13,15). To search for such adaptive metastatic metabolic alterations that associate with enhanced PCK1 expression, we performed metabolomic profiling of the four highly/poorly metastatic CRC PDX pairs. Unsupervised hierarchical clustering analysis was then performed on the differentially expressed metabolite profiles for each pair. Interestingly, the most salient observation was increased abundance in three out of four PDX pairs of multiple nucleoside base precursors and specifically metabolites in the pyrimidine biosynthetic pathway (Fig.5C). These metabolites comprised orotate, dihydroorotate, and ureidopropionate. These findings reveal that metastatic colonization by human CRC cells selects for induction of multiple metabolites in the pyrimidine biosynthetic pathway.
We hypothesized that perhaps enhanced levels of pyrimidine precursors were selected for in metastatic CRC cells to enable adaptation to hypoxia where precursors for pyrimidine biosynthesis such as aspartate are known to become depleted (23,25).
Without such an adaptation, cells would experience deficits in pyrimidine bases and consequently nucleotide pools, which would curb growth. How might PCK1 upregulation contribute to maintenance of nucleotide pools? Nucleotides contain nitrogenous bases covalently coupled to ribose and phosphate. PCK1 was previously shown to promote ribose generation by CRC cells under pathophysiological levels of glucose via the pentose phosphate pathway (26). We thus hypothesized that PCK1 The above findings reveal that liver metastatic CRC cells enhance pyrimidine levels and that PCK1 drives pyrimidine nucleotide levels under hypoxia. These findings suggest that hypoxia acts as a barrier to growth for metastatic CRC by limiting pyrimidine nucleoside levels. To directly test this, we determined if the growth defect of PCK1 depletion upon hypoxia could be rescued by the pyrimidine nucleoside uridine. Indeed, supplementation of CRC cells with uridine rescued the hypoxic growth defect observed upon PCK1 depletion (Fig.5G). These findings reveal PCK1 induction to be a mechanism employed by CRC cells to enhance pyrimidine nucleotide levels under hypoxia to promote growth.

Inhibition of PCK1 or DHODH suppresses CRC liver metastatic colonization
Due to the strong reduction in mCRC liver colonization observed upon PCK1 depletion, we hypothesized that PCK1 inhibition may represent a potential therapeutic strategy for impairing CRC metastatic progression. We performed in vivo proof-of-principle experiments in two independent CRC cell lines with a PCK1-inhibitor, 3mercaptopicolinic acid (3-MPA) (27). We treated CRC cells in vitro for 24 hours at a dose that did not alter cell proliferation in vitro (Fig.S6A). The following day, mice were subjected to portal circulation injections with either control or 3-MPA-treated cells. On day one, we repeated the 3-MPA or control gavage. We found that even such short-term treatment of 3-MPA decreased colorectal cancer liver colonization in this model (Fig.6B, Fig.S6C) (p=0.008; p=0.005 for Fig.6B and Fig.S6C respectively). Taken together, these results indicate that PCK1 promotes colorectal cancer liver colonization and represents a potential target for which therapeutics could be developed as a means of reducing CRC for metastatic relapse.
Dihydroorotate Dehydrogenase is a key enzyme in the metabolic pathway that reduces dihydroorotate to orotate, which is ultimately converted to the pyrimidine nucleotides UTP and CTP (Fig.6C). To further confirm that pyrimidine biosynthesis promotes CRC hypoxic growth, we sought to assess CRC growth upon DHODH inhibition. Leflunomide is an approved, well-tolerated, and high-affinity (Kd=12nM) small-molecule inhibitor of DHODH used in the treatment of rheumatoid arthritis. Leflunomide treatment significantly impaired CRC growth in the context of hypoxia-an effect that was more significant under hypoxia than normoxia (Fig.S6D-E). These findings confirm that metastatic CRC cell growth under hypoxia is sensitive to pyrimidine biosynthesis inhibition.
Our findings as a whole suggest that metastatic CRC liver metastatic colonization may be sensitive to inhibition of the pyrimidine biosynthetic pathway. To directly test this, we first depleted highly metastatic LVM3b CRC cells of DHODH (Fig.S6F). DHODH depletion substantially reduced CRC liver metastatic colonization (Fig.6D), revealing a critical role for DHODH activity and pyrimidine biosynthesis in CRC liver metastatic colonization. To determine if leflunomide can therapeutically inhibit CRC liver metastasis, we treated animals with a dose of this drug similar to that used for rheumatoid arthritis (7.5mg/kg body weight). Treatment of animals injected with highly metastatic Lvm3b cells with leflunomide caused a ~90-fold reduction in CRC liver metastatic colonization (Fig.6E). The leflunomide treated mice experienced significantly longer survival (p=0.006) than the control mice (Fig.6F). Importantly, these cells are known to be highly resistant to 5-FU (28), the backbone chemotherapeutic used in CRC, revealing that inhibition of DHODH can exert therapeutic benefit despite cellular resistance to an anti-metabolite that targets the pyrimidine pathway. Leflunomide treatment only modestly impacted primary tumor growth by two distinct CRC populations ( Fig.S6G-H), suggesting a preferential sensitivity of CRC cells to leflunomide-mediated DHODH inhibition during liver metastatic colonization. To determine if the metastatic colonization defect caused by leflunomide treatment is caused by pyrimidine depletion, we tested cell growth suppression in the presence or absence of uridine-the downstream metabolic product of the pyrimidine biosynthetic pathway. This revealed that the impaired growth upon hypoxia was rescued upon uridine supplementation (Fig.6G). Importantly, leflunomide treatment impaired proliferation significantly more in the context of hypoxia than under normoxia ( Fig.S6D-E). These observations reveal enhanced dependence of highly metastatic cells on pyrimidine biosynthesis and reveal upregulation of metabolites in this pathway as a selective adaptive trait of highly metastatic CRC cells. Overall, these findings identify DHODH as a therapeutic target in CRC progression and provide proof-of-concept for use of leflunomide in therapeutic inhibition of CRC metastatic progression.

Discussion
Colorectal cancer remains a challenging disease despite multiple advances over the last six decades. Some patients with metastatic CRC can experience regression responses to current therapies, though most succumb to their disease within three years. Given that most colorectal cancer deaths occur as a result of complications of metastatic disease, a model that can predict which patients with advanced CRC harbor more aggressive disease could aid in appropriately positioning patients for experimental clinical trials. The objectives of our study were two-fold: to develop a colorectal cancer liver metastasis patient-derived xenograft model, and to employ this model to identify candidate genes that may drive colorectal cancer liver colonization.
Most patient-derived xenograft models consist of subcutaneous tumor tissue implantation. Similar to others, we found that successful subcutaneous tumor engraftment associated with worse patient survival in those with colorectal cancer (7).
However, among those tumors in our study that did engraft subcutaneously, the subcutaneous tumor growth rate did not significantly correlate with patient survival. In contrast, we found that liver metastasis growth rate was significantly correlated with patient survival. The reason for this discrepancy in the prognostic power of subcutaneous tumor growth versus liver metastatic growth is likely the greater selective pressure inherent to the liver microenvironment. This collection of clinically predictive colorectal cancer liver metastatic PDX models represents a valuable resource for the cancer community.
PCK1 is the rate-limiting enzyme in gluconeogenesis and is often upregulated in patients with metabolic syndrome and diabetes mellitus. Epidemiologic data suggests that those patients with diabetes that are on metformin, a gluconeogenic-antagonist, exhibit improved colorectal cancer clinical outcomes relative to their metformin-free counterparts (29)(30)(31). Our observations suggest one potential mechanistic basis for the sensitivity of CRC metastatic progression to inhibition of this pathway.
Metabolic rewiring in cancer has been well-established to provide tumor cells with the necessary nutrients and anabolic components to sustain proliferative and energetic demands (32)(33)(34). While numerous pathways are involved in metabolic reprogramming, metabolic shunting into pathways including glucose metabolism, the citric acid (TCA) cycle, and lipogenesis largely support macromolecule synthesis for cancer cells (35)(36)(37)(38).
In line with these notions, there have been two reports on PCK1 and its role in cancer. silencing elicited the opposite phenotype in culture (39). Using cell culture metabolomics, Montal et al. recently described a mechanism by which PCK1 promotes colorectal cancer growth through its increased ability to metabolize glutamine into lipids and ribose (26). PCK1 silencing in a colorectal cancer cell line in vitro was shown to decrease glutamine utilization and TCA cycle flux (26). They further found that cells with increased expression of PCK1 consumed more glucose and produced more lactate.
The authors performed PCK1 staining on a primary colorectal cancer tissue microarray, finding that PCK1 was overexpressed in many primary CRC biopsies, but PCK1 expression was not associated with tumor grade. Our findings demonstrate a major role for PCK1 in liver metastatic colonization by CRC. While we did not find evidence of Second, we provide the first reported evidence that PCK1 can promote hypoxic survival.
Third, we uncover a key role for PCK1 and gluconeogenesis in pyrimidine biosynthesis under hypoxia. Our findings as well as the previously reported findings on PCK1 and PCK2 support important roles for these gluconeogenesis enzymes in cancer initiation, progression, and potential novel therapies.
Because metabolic programs are altered within tumor cells in the tumor microenvironment, metabolic liabilities emerge that provide therapeutic opportunities (39,43,44). Past elegant work by White et al. implicated DHODH as a regulator of melanoma formation via its effects on transcriptional elongation (45). More recent work has implicated DHODH as a regulator of differentiation in certain myeloid leukemias and pancreatic adenocarcinoma (46,47). Furthermore, Bajzikova et al. found that de-novo pyrimidine biosynthesis is essential for mouse breast cancer tumorigenesis in a DHODH dependent manner (48). Our work reveals that beyond effects on cell growth in vitro and primary tumor growth, CRC metastatic progression selects for upregulation of pyrimidine biosynthesis. Moreover, the use of leflunomide to therapeutically target DHODH has been implicated under various cancer contexts as a metabolic inhibitor (49,50). Here, we observe that molecular or pharmacological inhibition with leflunomide of this pathway strongly impairs CRC metastatic colonization relative to primary tumor growth. Our work reveals that hypoxia enhances the sensitivity of cells to DHODH inhibition, suggesting that enhanced pyrimidine biosynthesis enables enhanced growth upon hypoxia-a key feature of the hepatic tumor microenvironment. 5-Fluorouracil (5-FU) was the first chemotherapeutic to demonstrate efficacy in reducing the risk of CRC recurrence (51). This agent remains the backbone of the current FOLFOX regimen, which is administered to patients after surgical resection to reduce the risk of metastatic relapse. Interestingly, 5-FU targets thymidylate synthase, an enzyme downstream of DHODH in the pyrimidine biosynthetic pathway-supporting our premise of dependence of and susceptibility to inhibition of this pathway in CRC metastasis. Despite its activity, a large fraction of patients treated with 5-FU nonetheless relapse. Multiple mechanisms of resistance to 5-FU have been described (52). Our findings demonstrate that inhibition of DHODH can suppress metastatic progression of a CRC cell line that is resistant to 5-FU-revealing promise for clinical testing of this agent in patients at high risk for relapse and whose tumors may exhibit resistance to 5-FU. Overall, our work reveals that PDX modeling of CRC can be predictive of clinical survival outcomes; that integration of PDX modeling with in vivo selection can give rise to highly metastatic PDX derivatives which can be profiled transcriptomically and metabolically to identify key drivers of metastatic progression; and that PCK1 and DHODH represent key metabolic drivers of CRC metastasis and therapeutic targets in CRC. In vitro cell growth assays CT26 cells that had been stably transduced with PCK1-targetting shRNA hairpins or control hairpins were grown in vitro for 3 days and counted on day 3 using the Sceptor 2.0 automated Cell counter (Millipore).

In vitro hypoxia cell growth assays
Lvm3b cells or LS174T cells were grown under normoxia for 24 hours followed by incubation for 5 days under 0.5% oxygen and then counted using the Sceptor 2.0 automated Cell Counter (Millipore).

3-Mercaptopicolinic Acid in vitro growth assay
LS174T cells were seeded in 6-well plates. On day 1, the media was replaced with either control media or media supplemented with 1mM 3MPA. On day 2, all the media was replaced with control media. The experiment was terminated on day 5.
Twenty-four hour exposure to 1mM 3MPA in media does not alter LS174T cell growth in vitro. 2 x 10 4 LS174T cells were seeded in triplicate. On day 1, the media was replaced with either control media or media supplemented with 1mM 3MPA. On day 2, all the media was replaced with control media. The experiment was terminated on day 5. aligned to the reference transcriptome (Hg19) using TopHat (v2). Cufflinks (v2) was used to estimate transcript abundances. Upon merging assemblies (Cuffmerge), comparison of samples was made using Cuffdiff (v2) to determine genes that were differentially expressed between parental and liver-metastatic derivative xenografts.
Fisher's method was used to determine genes that were differentially expressed across all analyzed gene sets.

Gene expression profile clustering
Correlation matrix of gene expression profiles from RNA sequencing were generated using Spearman's correlation coefficient. Clustering was performed in R using Euclidean distance and complete agglomeration method.

Gene Set Enrichment Analysis (GSEA)
Each isogenic tumor pair (parental and liver-metastatic derivative) was evaluated for changes in the Hallmark gene sets using GSEA (v2.2.1, Broad Institute). Additionally, a composite gene set using Fisher's method as described in the section above was analyzed using GSEA.

Statistics
Kaplan-Meier analysis was used to evaluate patient survival based upon PDX parameters. Sample size in mouse experiments was chosen based on the biological variability observed with a given genotype. Non-parametric tests were used when normality could not be assumed. Mann Whitney test and t test were used when comparing independent shRNAs to shControl. One-tailed tests were used when a difference was predicted to be in one direction; otherwise, a two-tailed test was used. A P value less than or equal to 0.05 was considered significant. (*: P<0.05, **:p<0.01, ***:p<0.001, and ****: p<0.0001)Error bars represent SEM unless otherwise indicated.

Study Approval
Approval for the study was obtained through the MSKCC Institutional Review    tumor reached the threshold size, it was removed from the mouse, dissociated into a single-cell suspension, and injected into the spleens of another set of mice as a means of introducing the colorectal cancer cells into the portal circulation. When the mice were deemed ill, the liver tumors were removed, dissociated, and re-injected to establish a next-generation liver derivative. This was repeated numerous times (Range: 5-13) with each PDX sample to obtain a distant liver metastatic CRC PDX derivative requiring euthanasia of mice in 3 weeks after cancer cells injection. Each of the CRC PDX liver derivatives grew significantly faster in the livers compared to their parent samples.        these genes were significantly upregulated across the liver-metastatic derivatives compared to their parental counterparts. Mann-Whitney test, Bonferroni adjusted) (E) The in vivo growth rate of colorectal cancer cells in the liver, as calculated through the natural log of the liver photon flux ratio, was altered upon PCK1 modulation (p<0.0001, p<0.0001, p<0.0001, p=0.03, p<0.0001, and p<0.0001 for SW480 shCTRL vs shPCK1-3, vs shPCK1-4, LS174T shCTRL vs shPCK1-5, SW480 pbabe-empty vs PCK1 OE, SW480 shCTRL vs shPCK1-4 Dox+, vs shPCK1-4 Dox-, respectively, Analysis of covariance(ANCOVA).