Thymidylate synthase maintains the de-differentiated state of aggressive breast cancers

Cancer cells frequently boost nucleotide metabolism (NM) to support their increased proliferation, but the consequences of elevated NM on tumor de-differentiation are mostly unexplored. Here, we identified a role for thymidylate synthase (TS), a NM enzyme and established drug target, in cancer cell de-differentiation and investigated its clinical significance in breast cancer (BC). In vitro, TS knockdown increased the population of CD24+ differentiated cells, and attenuated migration and sphere-formation. RNA-seq profiling indicated a repression of epithelial-to-mesenchymal transition (EMT) signature genes upon TS knockdown, and TS-deficient cells showed an increased ability to invade and metastasize in vivo, consistent with the occurrence of a partial EMT phenotype. Mechanistically, TS enzymatic activity was found essential for the maintenance of the EMT/stem-like state by fueling a DPYD-dependent pyrimidine catabolism. In patient tissues, TS levels were found significantly higher in poorly differentiated and in triple negative BC (TNBC), and strongly correlated with worse prognosis. The present study provides the rationale to study in-depth the role of NM at the crossroads of proliferation and differentiation, and depicts new avenues for the design of novel drug combinations for the treatment of BC.

Nucleotide metabolism (NM) is classically viewed as a motor of cellular proliferation (Lane and Fan, 2015). Cancer cells are, in fact, highly dependent on the de novo synthesis of nucleotides to produce sufficient DNA and RNA precursors to support their growth, and some cancer-promoting signaling pathways have been shown to regulate NM (Tong et al., 2009). However, a few recent studies have suggested that nucleotides-generating metabolic pathways may also serve as regulators of cancer stemness (Bageritz et al., 2014;Morgenroth et al., 2014), opening the possibility that some NM enzymes are implicated in cancer cell de-differentiation.
Thymidylate Synthase (TS) is the enzyme that catalyzes the conversion of deoxyuridine monophosphate (dUMP) to thymidine monophosphate (dTMP or thymidylate). Since this reaction provides the sole de novo pathway for thymidylate production, TS is essential for DNA synthesis and repair, and its absence blocks proliferation and causes cell death (Costi et al., 2005;Wilson et al., 2008). We previously discovered that TS expression is correlated with the EMT phenotype in the NCI-60 transcriptomic database by using a pan-cancer EMT gene ratio (Vimentin/E-Cadherin, VIM/CDH1) (Siddiqui et al., 2017), but the mechanistic involvement of the TS enzymatic activity on EMT/CSCs has never been shown.
Here, we report a novel fundamental role of TS in maintaining the de-differentiated phenotype of BC cells and its differential expression in the BC subtypes, with several potential therapeutic implications.

TS is a marker of more aggressive and EMT-driven breast cancer.
In order to study the association of TS expression with EMT markers in BC, we employed a VIM/CDH1 ratio to classify the BC cell lines belonging to the CCLE dataset (n=52) into epithelial, mesenchymal or intermediate phenotypes ( Fig. 1A and Supplementary Table 1. Comparing epithelial (VIM/CDH1<2) with mesenchymal cells (>2), a significantly higher expression of TYMS mRNA was found in the latter (p<0.005, Fig. 1B). Then, in order to test TS gene expression in the spectrum of BC patients, we analyzed 3 independent GEO datasets, and found that TS expression was significantly different among the BC subtypes. Normal-like samples or the well-differentiated tumors (like luminal A) exhibited low TS expression, whereas high TS levels were found in basal-like BC (Fig. 1C). BC with a basal-like gene signature are primarily triple-negative (TN), and are frequently enriched for markers of CSCs and EMT markers (Taube et al., 2010).
In order to test the clinical significance of the TS protein and its association with the aggressive phenotype in BC, immunohistochemistry (IHC) was performed on formalin-fixed paraffin-embedded samples from 120 BC patients. Patients' characteristics are shown in the Supplementary Table 1. TS staining was quantified by an IHC score and a significantly higher expression was found in the TNBC, as compared to Luminal-A (Fig. 1D, ANOVA p<0.01) and in high grade (G3) compared to low grade tumors ( Fig. 1E-F, ANOVA, p<0.01). A moderate level of correlation was found between TS and Ki67 proliferation marker (Rs=0.48, Supplementary Fig 1A). Analysis of the prognostic values using large patient datasets indicated TS expression as a marker for poor overall survival in BC (all subtypes, Fig. 1G). Of note, TS prognostically stratified luminal A and luminal B patients (Fig. 1H) as well as lower grade patients (Fig. 1I), while no association was not found in more aggressive BC (Supplementary Fig 1B).

BC.
The patient data prompted us to test whether TS plays a role in maintaining the dedifferentiated state of TNBC. Hence, we transduced the TNBC cell line MDA-MB-231 with lentiviruses containing non-overlapping shRNA sequences to knockdown TS ( Fig. 2A). CD44/CD24 surface staining followed by FACS quantification indicated an increase in the population of differentiated CD24 + cells, with a concomitant decrease in the CD44 + CD24 -BC stem cell population (Fig. 2B). In order to carefully monitor the effects on cell growth, the confluency of infected cells was examined with real-time proliferation assays upon TS depletion. The results showed a significant suppression of proliferation in cells infected with the shRNA that delivered a strong TS repression (shTS#2), while the sequence with a mild knockdown (shTS#1) did not alter the cells' growth ( Fig. 2C). By contrast, a shRNA (shTS#3) sequence that induced a complete TS elimination in this cell line caused a massive growth arrest and death ( Supplementary Fig 1C-D), in line with the lifeessential role of TS. To determine if the partially TS-deficient cells had reduced ability to migrate, we performed a real-time wound-healing assay where cell migration is corrected for proliferation outside the wound to account for changes in cell growth rate upon TS repression. The results showed a significant loss of migratory ability upon TS knockdown ( Fig. 2D-E) (Fig. 2M). In addition, to monitor the intravasation ability, a CAM assay on chicken embryos was performed (Fig. 2N), which confirmed an increased ability of TS-deficient cells seeded on the upper CAM to intravasate into the lower CAM (Fig. 2O). These observations clearly pointed at a strong impact of TS on BC differentiation and alteration of cells' behavior, which we aimed to further molecularly characterize.

TS knockdown induces loss of EMT and correlates with less aggressive BC.
To further delineate the molecular pathways regulated via TS and involved in mediating the de-differentiation program in BC, we subjected shTS#1 MDA-MB-231 cells to RNA-sequencing. By setting a stringent cut-off value of 2-fold to identify differentially-expressed genes (DEGs) compared to non-targeting infected cells (pLKO), we found 73 and 84 genes down and up-regulated, respectively (Supplementary Table 1). Pathway analysis (GSEA) revealed that EMT was the most significantly deregulated pathway, followed by TNF-α/NFκ-β signaling ( Fig.   3A), known to be functionally connected with EMT in BC (Li et al., 2012). RNA-seq data were validated by qPCR and (Fig. 3B, Supplementary Fig. 2A) were used to explore the prognostic significance of the gene signature associated with TSknockdown. A system biology approach was adopted to investigate different BC gene expression datasets, dividing the patients by low or high KD score (corresponding to high or low TS, respectively). As a further confirmation of a strong association with EMT, a previously established signature of genes upregulated during EMT (Sarrio et al., 2008) was found significantly enriched in patients with a low TS KD score (Fig. 3C). Moreover, a gene set predicting poor prognosis was found enriched in patients with low KD score (Fig. 3D), and similar associations with the grade of differentiation were found using both the TS KD score and TS expression to segregate the patients (Supplementary Fig. 2B-C). Survival analysis confirmed the strong prognostic impact of the TS signature, indicating a poorer prognosis for the patients with a lower score or higher TS in all subtypes (Fig 3E), and a trend for a significant impact in TNBC (Fig 3F). The KD score was also significantly correlated with all the major histopathological variables in BC, including histology, grade of differentiation and size of the tumors ( Fig. 3G-J). All these data together indicate that TS promotes EMT-driven aggressive BCs.
TS enzymatic activity and thymidine catabolism are essential for maintaining BC de-differentiation.
We then aimed at determining the impact of TS enzymatic activity on dedifferentiation. TS is the only de novo source of thymidylate (dTMP), which is either further phosphorylated to maintain the dNTP pools for DNA synthesis, or directed to degradation via sequential phosphorolytic cleavages. MDA-MB-231 shTS cells with the lowest level of TS knockdown and no major proliferation defect, but with a discernable change in the CD24 + population (shTS#1), where subjected to dNTP quantification. Consistent with the previously observed normal growth, no significant change in dTTP or other dNTPs was detected (Fig. 4A). This suggested that the differentiation observed in shTS cells was not caused by dNTP imbalance, as shown in other CSC models upon knockdown of specific NM-related enzymes (Bageritz et al., 2014), and that while a baseline TS level is required for the cells to grow, increased TS activity could be partly involved in other cellular functions. To functionally investigate the role of TS enzymatic activity, we reconstituted shTS#1 cells with either a wild type (TS wt ) expressing silent mutation in shRNA binding region or an enzymatically inactive (TS R50C ) form of TS (Rahman et al., 2004).
Western blotting confirmed the successful transduction ( Fig. 4B) and TS enzymatic activity assay confirmed the loss of catalytic activity in the TS R50C overexpressing cells (Fig. 4C). Strikingly, MDA-MB-231 cells overexpressing the TS wt enzyme increased proliferation (Fig. 4D), reduced the population of differentiated CD24 + cells (Fig. 4E), formed more mammospheres (Fig. 4F) and were significantly more migratory (Fig. 4G) compared to TS R50C cells overexpressing the catalytically inactive TS. These results clearly indicated that the TS enzymatic activity is essential for the maintenance of the EMT/CSCs phenotype. We therefore hypothesized that TS mediates de-differentiation via the nucleotide catabolism ( Fig.   4H), rather than through the pathway leading to DNA synthesis. This was supported by the results of a recent shRNA screen on human immortalized breast cells, which revealed that pyrimidine catabolism mediated by dihydropyrimidine dehydrogenase (DPYD) was responsible for the EMT phenotype (Shaul et al., 2014).
We therefore speculated the existence of a TS-DPYD axis controlling EMT/CSCs in BC cells. Analysis of expression data from the CCLE dataset reinforced this hypothesis, by showing a marked increase in mesenchymal-like BC cells not only for DPYD, but also for the NT5E (CD73), which is the upstream 5'-nucleotidase responsible for catalyzing the first step of dTMP degradation (Fig. 4I). In order to functionally prove our hypothesis, we knocked down DPYD (Fig. 4J) and observed a significant increase in the population of CD24 + cells (Fig. 4K) and a loss of migratory ability (Fig. 4L), in line with what we observed in TS-deficient cells.
However, the CD24 + enriching effect of DPYD knockdown could not be reverted by overexpressing TS (Fig. 4M), indicating that a functional DPYD is required for TS to promote de-differentiation. In summary, TS enzymatic activity and the rate of pyrimidine catabolism are essential for the maintenance of the BCSC phenotype.
We therefore propose a model in which dTMP produced by TS-overexpressing cancer cells is not only metabolized to support the uncontrolled proliferation, but can also partially sustain de-differentiation and EMT via DPYD-based pyrimidines degradation ( Fig. 4N).

DISCUSSION
The aggressiveness and the de-differentiated phenotype of neoplastic cells have been strongly connected with alterations in specific metabolic pathways, especially those involved in the transformation of glucose (Colvin et al., 2016;Morita et al., 2018;Schwab et al., 2018;Sciacovelli et al., 2016), while the contribution of other pathways is still majorly unexplored. Elevation in NM is typically associated with the tumor cells' increased demand for DNA precursors to sustain the uncontrolled proliferation (Lane and Fan, 2015). However, a few studies have shown that some NM enzymes are functionally involved in de-differentiation processes, like the ectonucleotidase ENPP1 in the maintenance of the CSCs-like state in glioblastoma (Bageritz et al., 2014). Similarly, treatment with non-toxic doses of TS-inhibiting drugs was found sufficient to induce a differentiation in multiple myeloma CSCs and sensitize the cells for radiotherapy (Morgenroth et al., 2014). These observations, together with the first demonstration of a connection between TS and EMT (Siddiqui et al., 2017), prompted us to investigate the EMT-driven TNBC model (Mani et al., 2008;Taube et al., 2010). Functional experiments clearly indicates that loss of TS altered the de-differentiated phenotype of TNBCs, reducing CD44 + CD24cells, and suppressing migratory and sphere-forming ability. Consistently, this was accompanied by a robust suppression of EMT-associated genes, as evaluated by RNA-seq analysis. In vivo, two independent animal models, confirmed by an in vitro  (Kang et al., 2018). Nevertheless, both the immunohistochemical quantification of TS protein levels and the TS knockdown scores derived from the RNA-seq data were strongly associated with BC dedifferentiation and prognosis, indicating a pivotal role of TS in the malignancy of BC. Interestingly, only a moderate co-expression between TS and the proliferation marker Ki67 was found, in line with previous data from other cancers (Ceppi et al., 2008;Monica et al., 2009), supporting our model that TS activity could be implicated in other cellular functions independent of proliferation. We found, in fact, that the EMT/CSC suppression imposed by a level of TS knockdown which did not perturb cells' proliferation and dNTP balance was rescued by overexpressing wildtype TS, but not by a catalytically inactive mutant. This was particularly important because it ruled out the possible contribution of non-enzymatic activities of TS (Liu et al., 2002), directly pointing at the enzymatic activity as the EMT/CSC driving force. Since dihydropyrimidine accumulation was previously shown to control EMT in BC (Shaul et al., 2014), we tested and demonstrated the hypothesis that high TS enzymatic activity in cancer cells sustains de-differentiation and EMT via a DPYD-dependent pyrimidine catabolism. Several follow-up studies are needed, for instance, to 1) address the impact of additional dTMP-transforming enzymes (like NT5E/CD73) and of salvage pathways; 2) identify the regulatory pathways upstream of the TS-DPYD axis, and the involvement of EMT transcription factors; 3) test a similar function for TS in other malignancies; 4) decipher the contribution of TS dysregulation to cancer-relevant pathways other than EMT (like inflammation).
In any case, our data assert that the classical role of TS as a mere proliferation marker needs to be revisited. For instance the previously identified oncogenic and tumor-initiating role of TS (Bertino and Banerjee, 2004;Rahman et al., 2004) could be explained with the direct/indirect control of de-differentiation and stemness.
More efforts will need to be dedicated to clarify how the cancer cells balance proliferation and differentiation, and to what exact extent NM genes are implicated in these processes.
TS is a well-established target of chemotherapy, being inhibited by drugs like 5fluorouracil (5-FU) or by folate analogues (Wilson et al., 2014), and its overexpression in tumors represents a major mechanism of chemo-resistance.
From the translational point of view, our finding that TS levels can be significantly different among BC subtypes contradicts earlier works (Pestalozzi et al., 1997) and may be useful to improve the treatment strategies. In our study, TS was found higher in the aggressive BC and in high-grade tumors, in line with previous observations in non-small cell lung (NSCLC) and in gastro-entero-pancreatic cancers (Ceppi et al., 2008;Ceppi et al., 2006;Monica et al., 2009), all representing tumors in which this information was found clinically very important for predicting the efficacy of anti-TS drugs (Scagliotti et al., 2008;Selvaggi and Scagliotti, 2009      Western blot analysis. Cells were lysed in RIPA lysis buffer and quantified using the Pierce BCA protein assay kit (both from Thermo-Fisher). Proteins lysates (10-35μg) were resolved on 10%SDS-PAGE gels and transferred to PVDF membrane (Thermo-Fisher). Membranes were blocked in 5%Milk (BioRad) in 1XTBS-T.
Membranes were then incubated in primary antibodies diluted in blocking solution at 4°C overnight. Western blot antibody for TS (clone: EPR4545) and DPYD (clone: EPR8811) are from Abcam and β-Actin (clone: 8H10D10) is from Cell Signaling.
After incubation with secondary antibodies (Southern Biotech), detection was performed using the ECL method (Pierce ECL western blotting substrate, Thermo-Fisher) and developed on X-Ray film (Thermo-Fisher) using a chemiluminescence imager, AGFA CP100.

Cell Surface Staining and FACS. Anti-CD44-FITC and anti-CD24-PE antibodies
were purchased from Biolegend. For CD44/CD24 staining, 400,000 cells were plated in 6-wells and allowed to grow overnight. Cells were trypsinized and collected in Imaging System (Perkin Elmer). Analysis was performed with live imaging software by measuring photon flux. Lungs were collected and fixed in 10% buffered formalin and processed to obtain paraffin blocks. Three-micron thick sections of formalin fixed paraffin embedded samples were stained using Hematoxylin and eosin (H&E).
The mice experiments were performed at the Bilket University Ankara and were approved by Animal Ethics Committee of the university. Deoxynucleotide Triphosphate Quantification. The cellular dNTP levels were determined by the RT-based dNTP assay (Diamond et al., 2004). Briefly, the cellular dNTPs in experimental triplicates were extracted by methanol, and the determined dNTP amounts were normalized for an equal cell number (1x10 6 ).

Real Time HUVEC Invasion Assay (in vitro
TS Enzyme Activity Quantification. TS was quantified in MDA-MB-231 cells as previously described (Pluim et al., 2013Pluim et al., 2013