Mitochondrial One-Carbon Flux has a Growth-Independent Role in Promoting Breast Cancer Metastasis

28 Abstract 29 Progression of primary cancer to metastatic disease is the most common cause of death in cancer patients 30 with minimal treatment options available. Canonical drugs mainly target the proliferative capacity of cancer 31 cells, which often leaves slow-proliferating, persistent cancer cells unaffected. Thus, we aimed to identify 32 metabolic determinants that enable cell plasticity and foster treatment resistance and tumor escape. 33 Using a panel of anti-cancer drugs, we uncovered that antifolates, despite inducing strong growth arrest, did 34 not impact the cancer cell’s motility potential, indicating that nucleotide synthesis is dispensable for cell motility. 35 Prolonged treatment even selected for more motile cancer subpopulations. We found that cytosolic inhibition 36 of DHFR by MTX only abrogates cytosolic folate cycle, while mitochondrial one-carbon cycle remains highly 37 active. Despite a decreased cellular demand for biomass production, de novo serine synthesis and formate 38 overflow are increased, suggesting that mitochondria provide a protective environment that allows serine 39 catabolism to support cellular motility during nucleotide synthesis inhibition. 40 Enhanced motility of growth-arrested cells was reduced by inhibition of PHGDH-dependent de novo serine 41 synthesis and genetic silencing of mitochondrial one-carbon cycle. In vivo targeting of mitochondrial one- 42 carbon cycle and formate overflow strongly and significantly reduced lung metastasis formation in an orthotopic 43 breast cancer model. In summary, we identified mitochondrial serine catabolism as a targetable, growth- 44 independent metabolic vulnerability to limit metastatic progression. ImageStudioLite Software Vers.5.2 (LI-COR) was used for 548 image analysis. Antibodies used for Western blot analysis in this study: MTHFD1L (16113-1-AP) from 549 Proteintech; vimentin (3390), β -actin (3700), and MTHFD2 (41377) from Cell Signaling Technology; PHGDH 550 (HPA021241), PSAT1 (HPA042924), PSPH (HPA020376), SHMT1 (HPA023314), and SHMT2 (HPA020549) 551 from Sigma Aldrich; IRDye 680RD Goat Anti-Mouse IgG (H+L) and IRDye 800CW Donkey Anti-Rabbit IgG (H+L) from LI-COR.


51
Many primary tumors can be targeted by specific treatment schemes. Thus survival rates of patients harboring 52 primary tumors steadily increases. However, in case of tumor relapse and metastatic disease, effective 53 therapies are mostly lacking with the consequence that secondary tumors account for the majority of cancer 54 deaths [1]. 55 Classical chemotherapeutic approaches to counteract cancer growth aim to target biomass production by 56 interfering with the synthesis of proteins, lipids, and nucleotides and thereby directly interfere with central 57 pathways of cancer cell metabolism. Consequently, it is well studied that beneficial metabolic reprogramming 58 is fundamental for cancer cells to maximize the synthesis of essential lipids, proteins and nucleotides in order 59 to drive the excessive growth rates observed within the primary tumor and secondary metastasis [2,3]. 60 However, changes in cancer cell metabolism during stages of reduced biomass production, for example upon 61 chemotherapy or when cancer cells transition along the metastatic cascade, are far less described [3]. 62 Targeting metabolic cues to not only reduce cancer growth but also to metabolically interfere with invasive and 63 migratory properties of cancer cells represents a desirable therapeutic approach to ultimately prevent 64 metastatic progression of the initial disease. We have previously shown that a reduction of biomass demand 65 can result in increased formate release from cancer cells to their surrounding microenvironment [4]. The 66 mechanistic basis for this formate overflow phenomenon particularly in growth inhibited cancer cells, is the 67 rate of serine catabolism through serine one-carbon cycle (SOC). 68 Within this metabolic pathway, which spans between cytosol and mitochondrion [5], extracellular and de novo 69 synthesized serine is catabolized to support all three essential pillars of metabolism: biomass production, redox 70 equivalent and bioenergy generation [6]. Especially metastasizing cancer cells need to selectively and 71 dynamically adopt their cellular metabolism to fulfil these three major metabolic constraints along every step 72 of the metastatic cascade [3,7]. Previous studies have shown that metabolic rewiring during the growth-73 independent process of tumor cell escape and dissemination is dominated by bioenergetic and redox demands 74 rather than biosynthetic needs [3,7,8]. Importantly, serine plasticity allows to readjust the metabolic outputs 75 of serine catabolism through the various routes of SOC towards either of the three metabolic hallmarks [6]. 76 For example, upon high growth rates, serine, glycine and formate production from SOC is required to support 77 the synthesis of anabolic building blocks such as proteins, lipids, heme, glutathione, and nucleotides. In 78 contrast, upon energy limitation, most of serine derived carbon is excreted in the form of formate or CO2 and 79 SOC is maximized towards energy production in form of NADH and ATP [4]. Alternatively, upon oxidative 80 stress, the demand for glycine to support GSH synthesis is prioritized [9]. We have previously shown that the 81 rate of serine catabolism to formate is significantly increased in adenomas of the small intestine and breast 82 tumors compared to normal tissue and, that increased formate levels can increase the invasiveness of cancer 83 cells [10]. In addition, other labs also reported a role of serine metabolism in the context of metastasis [11][12][13]. 84 However, these recent works focused mainly on the relevance of serine de novo synthesis via 85 phosphoglycerate dehydrogenase (PHGDH) itself instead of addressing the subsequent catabolism of serine 86 via SOC for metastatic progression. Mechanistic studies on the role of SOC in context of metastasis were 87 mainly focused on the proliferative capacities essential for metastatic outgrowth [11,12,14] rather than 88 analyzing metabolic requirements that support the growth-independent metastatic processes of dissemination. 89 In the absence of such growth demands and under certain metabolic stress conditions, we have previously demonstrated that cells increase their rate of formate overflow via SOC [4,10], which further manifests that 91 regulation of SOC is a key metabolic switch especially in, but not limited to, the growth-independent stages of 92 metastatic outgrowth. 93 In the present study, our goal was to identify metabolic processes that are specifically important to support 94 cancer cell motility in the growth independent process of metastatic dissemination. We have discovered, that 95 antifolates, while having strong growth arresting properties do not decrease the cellular motility potential. Using 96 MTX as a tool compound to separate metabolic processes of growth from motility processes, we have identified 97 SOC as a central metabolic pathway that supports cellular motility potential. Specifically, the mitochondrial 98 SOC is required to support the observed motility, while the cytosolic part, which is essential for proliferation, is 99 dispensable for the cellular motility potential. Finally, we provide evidence that formate overflow via 100 mitochondrial SOC is a targetable growth-independent vulnerability to limit breast cancer metastasis. This 101 identified metabolic vulnerability might have clinical implications for metastatic progression in context of 102 growth-inhibiting chemotherapies. 103 104

105
Inhibition of major anabolic synthesis routes differentially impacts the motility potential of cancer cells 106 To assess the relative importance of different metabolic pathways for cancer cell migration, we employed a 107 panel of metabolic perturbations targeting glycolysis, mitochondrial oxidative phosphorylation (OxPhos) as well 108 as lipid, protein, and nucleotide synthesis ( Figure 1A). All metabolic interventions effectively inhibited the 109 growth and cell cycle progression of MDA-MB468 breast cancer cells and the chosen drug concentrations had 110 none to mild toxic effects after 48 h treatment ( Figure 1B-C, S1A). Inhibition of glycolysis, OxPhos, lipid, and 111 protein synthesis significantly reduced migration of MDA-MB468 cells ( Figure 1D). Surprisingly and in contrast 112 to all other growth-inhibiting conditions, inhibition of nucleotide synthesis with various drugs did not diminish 113 migration of MDA-MB468 cells ( Figure 1E). Calculation of the area under curve (AUC) allowed us to 114 quantitatively compare wound closure upon the diverse metabolic perturbations over time (Figure 1D, E) and 115 confirmed that wound closure was significantly reduced upon rotenone (Rot), galactose (Gal), simvastatin 116 (SIM) and sirolimus (rapamycin, SIR) treatment. However, treatment with the antifolates methotrexate (MTX) 117 and pemetrexed (PEM), as well as with hydroxyurea (HU) and clofarabine (CLO) did not affect or did even 118 significantly increase wound closure in scratch assays in comparison to fully proliferative, untreated control 119 cells ( Figure 1D, E). This divergent effect of nucleotide synthesis inhibition from all other metabolic 120 perturbations was further realized, when directly correlating our data on MDA-MB468 cell migration with cell 121 growth in response to treatment (Figure 1F, G). While cell migration correlates with reduced cell growth for all 122 other metabolic perturbations, nucleotide synthesis inhibition does not reduce cell migration in correlation to 123 growth inhibition (Figure 1F, G). This finding was verified in LN229 glioblastoma and 4T1 breast cancer cells 124 ( Figure S1B-D). To further confirm this finding, we assessed migration of MDA-MB468 cells in a trans-well 125 assay. Similar to our previous findings, there was no observable reduction of cell migration in response to MTX 126 or PEM ( Figure 1H). Furthermore, using ECM-collagen coating in trans-well assay, we found that nucleotide 127 synthesis inhibition had no impact on cancer cell invasion ( Figure 1H). These findings were also confirmed 128 using LN229 and 4T1 cells (Figure S1E, F). Given this pro-migratory effect of nucleotide synthesis inhibition 129 in growth-arrested cancer cells, we generated MTX-resistant MDA-MB468 cells by long-term cultivation in 130 50 nM MTX for 2 months. The resulting MTX-resistant MDA-MB468 cells proliferate in the presence of 131 increasing concentrations of MTX and show enhanced migratory capacity when compared to the parental 132 MDA-MB468 cell line (Figure 1I, J). 133 In conclusion, the perturbation of any major anabolic route results in growth repression, its impact on cell 134 migration however highly depends on the targeted metabolic pathway, with nucleotide synthesis being an 135 ineffective target to abrogate cell motility. Considering that cell proliferation contributes to the experimental 136 outcome in scratch and trans-well assays, cell migration must in fact be increased upon nucleotide synthesis 137 inhibition to compensate for the reduced growth rate. Different means of metabolic growth repression thus 138 differentially impact cell motility and do not necessarily result in decreased cell migration. Furthermore, 139 prolonged nucleotide synthesis inhibition in drug-resistant cells acts as a selection pressure for more motile 140 cancer subpopulations. 141

MTX treatment triggers a ROS-driven EMT phenotype 142
One of the most established and extensively described nucleotide synthesis-inhibiting clinical drugs is the 143 dihydrofolate reductase inhibitor MTX. MTX impedes the regeneration of tetrahydrofolate (THF), which is a 144 central enzymatic cofactor during purine and pyrimidine synthesis [15]. As MTX induced the strongest growth 145 repression without impacting cell death (Figure 1B, C) and had no effect on cell migration ( Figure 1E), we 146 selected it as a tool compound to identify metabolic processes that are essential to support cell migration and 147 metastasis independent of cell proliferation. We tested multiple MTX concentrations and chose a concentration 148 of 50 nM, which is not only significantly growth-and cell cycle-arresting in MDA-MB468 cells ( Figure 1B, S1A, 149 S2A), but has also been shown to be within the therapeutic window [16][17][18][19]. We observed that 48 h MTX  we also found superoxide dismutase (SOD) to be one of the most significantly upregulated enzymes in 164 response to treatment ( Figure 2C, S2F), which corresponds to prior reports that depict MTX as a potent 165 inducer of oxidative stress [20][21][22]. We validated these prior findings by measuring a moderate but significant 166 increase of mitochondrial and cytosolic reactive oxidant species (ROS) in MDA-MB468 cells in response to 167 MTX ( Figure 2D, E). Moreover, the mesenchymal marker protein vimentin (VIM) and transforming growth 168 factor-beta2 (TGFB2) were found to be highly upregulated in the global proteomics approach following MTX 169 treatment ( Figure 2C, S2F). This finding corresponds to earlier reports, which highlight the potential of 170 oxidative stress as a driver of cancer cell migration and the transformation of cancer cells via epithelial-171 mesenchymal transition (EMT) [6,[23][24][25][26][27][28]. We thus analyzed expression levels of the major regulators of EMT 172 -the transcription factors ZEB1, ZEB2, SNAIL, and SLUG -in MDA-MB468 cells and found that ZEB1 and 173 ZEB2 expression as well as VIM expression was upregulated in a time-dependent manner in response to MTX 174 treatment ( Figure 2F). Similar upregulation of ZEB1, ZEB2, and VIM expression was observed in the MTX- ROS levels after MTX acts as a pro-migratory stimulus that subsequently results in the establishment and 187 long-term selection of phenotypically altered cancer cell subpopulations. However, research in the past has 188 raised awareness, that antioxidant treatment is not a promising cancer treatment strategy with mixed, and not 189 always beneficial, outcomes in clinical trials [30,31]. Therefore, our aim was to further investigate the metabolic 190 mechanisms that support enhanced cell migration in the presence of the nucleotide synthesis and growth-191 inhibiting tool compound MTX. 192

MTX-treated cells sustain high metabolic rates and enhance de novo serine synthesis 193
Having established that growth arrested MDA-MB468 cells maintain full migratory and invasive properties 194 under MTX treatment, we wanted to define metabolic mechanisms that maintain this pro-invasive EMT 195 phenotype. To this end, we focused on central carbon metabolism using [U-13 C]glutamine and [U-13 C]glucose 196 to monitor glycolytic activity and glucose-or glutamine-derived carbon oxidation through the TCA cycle ( Figure  197 3A, D). Absolute uptake rates of glutamine and glutamate were sustained in response to MTX treatment 198 ( Figure 3B), while the relative flux of [U-13 C]glutamine throughout the TCA-cycle was significantly increased 199 in response to MTX ( Figure 3C). Absolute quantification of uptake and release rates of glucose and lactate 200 also revealed sustained high glycolytic rates upon MTX treatment ( Figure 3E). This came as a surprise, as 201 lactate release rates were previously shown to generally correlate with cell growth rates [32]. Here however, 202 MTX-treated and growth-arrested cells maintained their glycolytic rate and associated lactate release rates at 203 constant high levels. This indicates a constant generation of glycolysis-derived ATP even in the presence of 204 reduced energetic demand for anabolic reactions under growth-arresting conditions. Consequently, [U-205 13 C]glucose distribution within the TCA-cycle was comparable in untreated and MTX-treated MDA-MB468 cells 206 ( Figure 3F) as well as in 4T1 and LN229 cells (Figure S3A, B). It has previously been shown that MTX inhibits 207 OCR in HCT116 cells [33]. We could replicate this finding and observed a ~ 30 % reduction in OCR in HCT116 208 cells upon MTX ( Figure S3C). However, OCR in MDA-MB468, LN229, and 4T1 cells was sustained at control 209 level in response to MTX (Figure 3G), which indicates cell line specific effects of MTX on OCR. Overall, these 210 findings indicate that, despite decreased growth rates, central carbon metabolism and mitochondrial 211 respiration are sustained in response to MTX treatment, with an additional increase in oxidative glutamine 212 metabolism. Furthermore, we observed that MTX treatment significantly increased the relative flux of glucose 213 to serine via the serine de novo synthesis pathway despite decreased anabolic demands for nucleotide 214 synthesis ( Figure 3H, S3D). The serine de novo synthesis pathway is catalyzed by three enzymes -PHGDH, 215 PSPH, and PSAT1, whose protein levels remained unchanged in response to MTX treatment in MDA-MB468, 216 MDA-MB231, 4T1 and LN229 cells (Figure S3E-H). In agreement with Diehl et al. [34] who showed that 217 changes in NAD + /NADH ratio can increase PHGDH activity and in consequence serine de novo synthesis 218 rates, we observed an increased NAD + /NADH ratio upon MTX ( Figure 3I). Congruently, yet somewhat 219 surprisingly, MTX treatment did also increase the proportion of labeled, extracellular formate derived from [U-220 13 C]glucose through serine and the mitochondrial SOC ( Figure 3J). This may indicate that the mitochondrial 221 part of the SOC is not as inhibited by MTX treatment as its cytosolic part. In summary, we demonstrate that, 222 despite decreased metabolic demand for biomass production, MTX-treated cells sustain high metabolic rates 223 comparable to fully proliferating cells. Additionally, treated cells exhibit increased rates of serine synthesis and 224 formate excretion. 225

Mitochondria protect SOC-dependent serine catabolism in the presence of MTX 226
Given our finding of increased [U-13 C]glucose-derived formate ( Figure 3J), we hypothesized that the influence 227 of MTX-mediated DHFR inhibition on mitochondrial SOC is, in contrast to cytosolic SOC, limited ( Figure 4A). 228 To test this hypothesis, we employed [U-13 C]serine tracing to further characterize serine flux through SOC. We 229 observed that despite inducing growth-arrest, MTX did not alter the rates of serine consumption and glycine 230 release in MDA-MB468 or LN229 cells (Figure 4B, S4A). This is a remarkable observation for two reasons: 231  To comparatively demonstrate how sole or dual inhibition of cytosolic and mitochondrial SOC affects serine 251 MID, we analyzed a panel of CRISPR generated knockouts in HAP1 cells in which either the cytosolic 252 (SHMT1), the mitochondrial (MFT, SHMT2), or both compartments (FPGS, MFT+SHMT1) of SOC were 253 abrogated [38]. In concordance with our previous findings and resulting conclusion that MTX solely abolishes 254 cytosolic SOC, cytosolic SHMT1 KO did not eliminate intermediary M+1 and M+2 serine isotoplogues from [U-255 13 C]serine, whereas mitochondrial or combined cytosolic and mitochondrial inhibition of SOC upon MFT KO, 256 SHMT2 KO, FPGS KO and MFT+SHMT1 KO resulted in a complete loss of M+1 and M+2 serine isotopologues 257 ( Figure 4F). SHMT1 KO reduced serine M+1 and M+2 isotopologue abundance comparable to MTX treatment 258 ( Figure 4F). In mutants harboring a full inhibition of SOC or mitochondrial SOC, we additionally observed an 259 increase of the serine to glycine ratio, whereas MTX treatment and SHMT1 KO slightly reduced serine to 260 glycine ratio ( Figure 4G). Overall, these results indicate that mitochondria provide a protected cellular 261 environment that permits serine catabolism with formate overflow via SOC even in the presence of the 262 antifolate MTX. 263

Targeting of serine de novo synthesis decreases MTX-dependent cancer cell migration 264
Having confirmed increased serine synthesis from glucose and subsequent formate overflow through 265 mitochondrial SOC in response to MTX treatment, we sought to determine whether interfering with this 266 metabolic pathway is a means to counteract enhanced cell migration upon MTX. To identify a suitable inhibitor 267 of serine de novo synthesis, we tested multiple available allosteric and one competitive inhibitor of PHGDH. 268 We found that the competitive inhibitor BI-4916 (BI) had superior efficiency when compared to the allosteric 269  indicating that serine is not only essential to support biomass production but also to support the metabolic 285 program required for migration ( Figure 5F). Furthermore, while neither BI treatment alone nor serine and 286 glycine starvation alone were sufficient to prevent the induction of the previously observed EMT-phenotype in 287 response to MTX (Figure 2G, 5G), combined inhibition and starvation did minimize ZEB1, ZEB2, and VIM 288 upregulation ( Figure 5G). This indicated that the efficient restriction of both, extracellular and de novo-289 synthesized serine to inhibit mitochondrial SOC, is effective to prevent the pro-migratory, phenotypic change 290 upon MTX. Notably, serine and glycine starvation did not further increase cell death after combined MTX and 291 BI treatment ( Figure 5H). Since MTX-treated cells were already in full growth arrest, the additional perturbation 292 with S/G starvation and/or PHGDH inhibition reveals that serine catabolism fulfills an important function to 293 support cell motility. This function is independent of serine's anabolic function to support biosynthesis 294 processes. (Figure S5E). Using the MTX-resistant MDA-MB468 cell line, we found that BI treatment was also 295 sufficient to reduce the migratory capacity of such selected, pro-migratory, and potentially metastatic cancer 296 subpopulations ( Figure 5I). In summary, depletion of de novo-synthesized serine using a specific PHGDH 297 inhibitor uncovered that the mitochondrial SOC is essential to sustain maximal cancer cell migration under 298 growth-arresting stimuli such as nucleotide synthesis inhibition by MTX. This implies that the mitochondrial 299 SOC represents a metabolic vulnerability that can be targeted to limit cell migration during certain 300 chemotherapeutic interventions ( Figure 5J).  were not affected by MTHFD2 KD, which further confirmed that the observed reduction upon MTX and 313 MTHFD2 KD was not related to enhanced cell death or further reduced growth rates, but solely based on an 314 inherent reduction of migratory capacity upon loss of mitochondrial SOC activity (Figure 6B, 6D). In summary, 315 we uncovered that mitochondrial SOC acts as a metabolic facilitator of cell migration under stress conditions 316 such as chemotherapeutic nucleotide-synthesis inhibition. Based on these observations, we speculate that 317 different growth inhibiting and stress inducing conditions, such as antifolate therapy or energy limitation [4], 318 can trigger increased rates of formate overflow and concomitant accumulation of formate within the tumor 319 microenvironment. Such increased concentrations of formate might act in an autocrine and/or paracrine 320 manner to trigger a pro-invasive phenotype, which represents the first stage to enter the metastasis process. 321 Conclusively, targeting formate overflow emerges as a conceivable intervention to limit metastasis formation. 322

Genetic targeting of mitochondrial SOC reduces metastasis formation in vivo 323
We sought to confirm our hypothesis on the central role of mitochondrial SOC as a facilitator of cancer cell 324 migration and ultimately metastasis formation in vivo. To specifically inhibit formate overflow, we generated a 325 stable knockdown of Mthfd1l in 4T1 breast cancer cells. The engineered cells exhibited a functional knockdown 326 of the enzyme as seen from the prominent reduction in formate release rates when compared to non-targeting 327 shRNA transfected 4T1 cells (Figure S7A). In line with our previous reports [4, 10], knockdown of Mthfd1l had 328 no effect on cell proliferation in vitro ( Figure S7B) and reduced invasion ( Figure S7C). To test if formate 329 overflow suppression can reduce metastasis formation in vivo, we injected both murine cell lines into the 330 mammary fat pads of immunocompetent BALB/c mice and monitored primary tumor growth over 6 weeks 331 ( Figure 7A). In concordance with in vitro data, primary tumor growth in vivo and tumor weight at the endpoint 332 were not affected by Mthfd1l knockdown (Figure 7B, C). After 6 weeks, mice were sacrificed and lung tissue 333 was stained with H&E staining. Staining shows that micrometastases formation was visibly reduced in 334 response to MTHFD1L KD ( Figure 7D). In addition, the number of macroscopic lung metastases were counted 335 and we found primary tumors derived from Mthfd1l-depleted 4T1 cells had significantly less pulmonary 336 metastases compared to 4T1 cells expressing MTHFD1L (Figure 7E). While all mice that were injected with 337 MTHFD1L-expressing 4T1 cells, exhibited lung metastases over 50% of mice injected with Mthfd1l knockdown 338 cells did not show metastatic lesions (Figure 7E). These in vivo findings further strengthen the key relevance 339 of mitochondrial SOC and specifically mitochondrial formate generation as a cornerstone that supports cancer 340 cell migration independent of primary tumor growth and cancer cell proliferation rates. 341 342

343
In this study, we uncovered a growth-independent function for serine and its mitochondrial catabolism that 344 drives cancer cell dissemination. Importantly, by using MTX as a tool compound to study migration 345 independent of proliferation, we could show that mitochondria sustain an autarkic SOC that is sufficient to 346 support the cells migratory capacity. This finding implies the dismal possibility that chemotherapeutic 347 approaches with nucleotide synthesis inhibiting drugs, which are undertaken to reduce primary cancer growth, 348 might not sufficiently suppress the motility capacities of remaining cancer cells. The fact, that in our hands, 349 breast cancer cells even showed a transitioning phenotype and were selected over time to be drug resistant, 350 while at the same time showing enhanced migratory potential, could warrant further work to comparatively 351 investigate metastatic disease progression after such therapies. Of note, aside from its use in cancer therapy, 352 the antifolate MTX represents the "anchor drug" for chronic treatment of autoimmune diseases such as 353 rheumatoid arthritis. Here, intracellular erythrocyte concentration of MTX and mean MTX plasma concentration 354 after chronic, low-dose MTX were previously reported to be in a comparable, even slightly higher, nM range 355 to our chosen drug concentration of 50 nM [40,41]. This implies an additional, potential significance of our 356 findings in cancer patients that undergo chronic MTX therapy for an arthritic comorbidity and might warrant 357 further clinical retrospective and prospective investigations beyond the scope of this study. Our aim here was 358 however to comprehensively analyze the metabolic pathways that foster such sustained cell migration under 359 growth-arresting conditions. Having previously reported that formate overflow via SOC is a characteristic of 360 invasive cancer cells [10], and knowing that MTX has a direct inhibiting effect on SOC and proliferation, 361 studying this metabolic pathway emerged as a conceivable starting point for such further mechanistic 362 investigations. Within SOC, the non-essential amino acid serine is catabolized in the mitochondrion to glycine 363 and formyl-tetrahydrofolate (formyl-THF), which are required building blocks for cytosolic purine and thymidine 364 (a pyrimidine) synthesis. During thymidine synthesis, the co-factor THF is oxidized to dihydrofolate (DHF), 365 which subsequently requires regeneration to keep the folate in its bioactive form. This reduction of DHF to THF 366 is dependent on the enzyme DHF reductase (DHFR), which is a target of MTX. Hence, upon MTX treatment 367 the folate co-factor accumulates in its inactive form DHF, which prevents nucleotide synthesis and cell growth 368 through obstruction of SOC activity [5] [42]. In agreement, our tool compound MTX effectively abolishes SOC-369 dependent nucleotide synthesis in the cytosol. However, using [U 13 C]serine-and [U 13 C]glucose-assisted metabolic flux analyses, we found that mitochondrial one-carbon cycle was preserved under antifolate-induced 371 nucleotide synthesis inhibition. Such compartmentalization of SOC is on the one hand enabled by the lack of 372 oxidation of THF to DHF in the mitochondrial compartment, but is also based on the chemical modification of 373 folate species and their transport activities across the mitochondrial membrane. For maximal biologic activity, 374 folate species need to be polyglutamated by folylpolyglutamate synthase (FPGS) and the resulting 375 polyglutamate species were shown to be poorly transported across the mitochondrial membrane [43,44]. 376 Consequently, polyglutamated folate species can be chemically trapped within the mitochondria to sustain an 377 autarkic mitochondrial SOC especially upon growth arrest and subsequent unaltered mitochondrial content. 378 Furthermore, there have been implications that the mitochondrial folate transporter is specific for reduced 379 folates with the possibility of limiting the transport of MTX and folic acid itself across the mitochondrial 380 membrane [45].Taken together, such compartmentalization of SOC and autarkic function of mitochondrial 381 SOC emerges as a selective advantage upon perturbation of cytosolic SOC and indicates an additional 382 metabolic functionality of the mitochondrial SOC pathway during growth-arresting conditions. Specifically, 383 plasticity of serine catabolism through this sustained mitochondrial pathway could offer a metabolic reserve in 384 the form of ATP and redox equivalent generation that might allow growth-arrested cancer cells to undergo 385 epithelial-mesenchymal transition and to enhance their migratory capacity in order to escape from sites of 386 nutrient deprivation or chemotherapeutic targeting. In concert with this notion, we identified mitochondrial SOC 387 as a targetable metabolic vulnerability by using genetic and pharmacologic inhibition of mitochondrial one-388 carbon cycle activity to reduce migration in vitro and metastasis formation in vivo. These findings are in line 389 with previous reports indicating that serine de novo synthesis rates and serine uptake rates are associated 390 with disease progression and even metastatic dissemination [11][12][13]46]. However to our knowledge, this is 391 the first report to show that it is indeed the enhanced activity of mitochondrial serine catabolism to formate via 392 SOC which translates these enhanced serine synthesis rates or serine uptake rates to a pro-migratory signal 393 downstream. Based on these findings, we hypothesize that serine plasticity and the alternative usage of serine 394 catabolism in the mitochondria can selectively promote formate overflow in response to extrinsic or intrinsic 395 stress stimuli such as growth inhibition or nutrient deprivation. Consequently, the resulting increase in local 396 formate concentrations in the TME could subsequently serve as a local stress signal in an auto-or paracrine 397 manner to promote the formation of invasive cancer cells. These cells are, as a result, more prone and capable 398 in escaping a non-beneficial primary TME, which eventually advances dissemination and metastasis formation. 399 Thus, restricting mitochondrial SOC, which is at the root of this stress signaling cascade, emerges as a 400 promising, preventive strategy to hinder metastatic tumor progression early on.  Table S2. GC-MS 485 chromatograms were processed using Agilent MassHunter Quantitative Analysis for GC-MS, Version B.08.00. 486 Final determination of release rates was performed as described in [4]. Target ions (m/z) and dwell times are shown in Table S3. 499 The MetaboliteDetector software package (Version 3.220180913) was used for mass spectrometric data post 500 processing, quantification, MID calculations, correction of natural isotope abundance, and determinations of 501 fractional carbon contributions [47]. 502

Analysis of Medium Exchange Rates: 503
Polar metabolites of the culture medium were derivatized for 90 min at 45 °C with 20 μl of methoxyamine (c = 504 20 mg/ml) in pyridine under continuous shaking and subsequently for 90 min at 45 °C with 20 μl of MSTFA. 505 GC-MS analysis was performed using an Agilent 7890B GC coupled to an Agilent 5977A Mass Selective 506 Detector (Agilent Technologies). A sample volume of 1 μl was injected into a Split/Splitless inlet, operating in 507 splitless mode at 270 °C. Gas chromatograph was equipped with a 30 m (I.D. 250 μm, film 0.25 μm) ZB-35MS 508 capillary column with 5 m guard column (Phenomenex). Helium was used as carrier gas with a constant flow 509 rate of 1.2 ml/min. GC oven temperature program: 90° C for 1 min, 9° C/min to 270° C, 25° C/min to 320° C 510 and held for 7 min. Total run time was 30 min. Transfer line temperature was set to 280° C. MSD was operated 511 under electron ionization at 70 eV. MS source was held at 230° C and the quadrupole at 150° C. Full scan 512 mass spectra were acquired between m/z 70 and 700. 513 The MetaboliteDetector software package (Version 3.220180913) was used for quantification. Briefly, peak 514 areas of all isotopologues of defined quantification ions were summed up and divided by the peak area of the 515 internal standard for normalization. In addition, a calibration curve was prepared to calculate absolute 516 concentrations. Absolute uptake and release rates were calculated as described in [4] 517

Western Blot 537
Total cell lysates were prepared by 30 min incubation of cell pellets on ice in cell lysis buffer (150 mM NaCl, 538 1 mM EDTA, 50 mM Tris-HCl, 1% NP-40). Lysis efficiency was maximized by sonification. Lysis solution was 539 centrifuged at 13,000 g for 15 min at 4°C and supernatant was collected and stored at -80°C. Protein

Poly-L-lysine Coating 554
Poly-L-lysine (P1274) was purchased from Sigma-Aldrich and reconstituted at 500 μg/ml in H2OMQ. Wells were 555 coated with PLL prior to Seahorse measurement and migration analysis by Scratch assay. To this end, PLL 556 solution was diluted 1:20 in H2OMQ and added to the plates at least 1 h prior to seeding. Following incubation 557 with PLL and before cells were seeded, plates were washed twice with H2OMQ and allowed to dry under the 558 hood. 559

Seahorse Measurements 560
The day prior to measuring OCR, 40,000 cells were seeded on poly-L-lysine coated plates. Cells were treated 561 as indicated. XF96 Extracellular Flux Analyzer (Seahorse Bioscience) was used to measure basal OCR 562 following manufacturer's instructions. Basal OCR was normalized to the protein concentration in the wells 563 following the protocol described in [49] using Bradford assay. 564

Flow Cytometric Analysis of Cell Cycle Distribution and Cell Death 565
200,000 cells were seeded in 2 ml DMEM and treated the subsequent day as indicated. After incubation, 566 medium was collected and cells were washed with PBS. PBS fraction was collected and cells were detached 567 with trypsin. Trypsination was stopped by DMEM addition and this fraction as well as a subsequent PBS 568 washing fraction were collected. The combined solutions were centrifuged and washed with PBS. 569

Cell Cycle Distribution
Subsequent centrifugation yielded a pellet that was resuspended in 100 μl PBS and fixed with ice-cold 80% 571 EtOH. Fixed cells were stored at -20°C for at least 1 h and maximum 5 days prior to measurement. Cells were 572 centrifuged, pellet was incubated for 1 h in 200 μl RNAse A in PBS (30 μg/ml) at RT. Immediately prior to 573 measurement, 98 μl propidium iodide (PI) in PBS (50 μg/ml) was added. Flow cytometric analysis was 574 performed using BD FACSCanto and BD FACSDiva software. Analysis was performed in FlowJo. 575 Cell Death Analysis 576 A fresh pellet following centrifugation was resuspended in 50 μl AnnexinV-FITC staining solution (5% 577 AnnexinV-FITC in AnnexinV binding buffer (10 mM HEPES pH 7.4, 140 mM NaCl, 2.5 mM CaCl2, 0.1% BSA 578 in ddH2O)) and incubated for 15 min on ice in the dark. 450 μl PI-staining solution (1.1 μg/ml PI in AnnexinV 579 binding buffer) was added immediately prior to measurement using BD FACSCanto and BD FACSDiva 580 software (for wild-type cells) or NovoCyte Quanteon (for GFP-positive transfected cells). Analysis was 581 performed in FlowJo. 582

Flow Cytometric Analysis of ROS Levels 583
250,000 MDA-MB468 cells were seeded in 2 ml medium in 6 well plates and treated as indicated. Following 584 incubation, adherent cells were detached with trypsin, centrifuged at 350 g for 5 min and washed with warm 585 DMEM. Live cells were stained in 100 μl DMEM supplemented with 1:2,000 DAPI and 1:500 DCFDA for 30 min 586 at 37°C. Following incubation, samples were centrifuged at 350 g for 5 min and washed with PBS. Following 587 centrifugation at 350 g for 5 min, cells were resuspended in 100 μl PBS and measured using the BD 588 LSRFortessa TM system and BD FACSDiva software. Data analysis was performed using FlowJo software. 589

Cell Proliferation 590
Cell proliferation was determined using the IncuCyte® Live-Cell Analysis system (Essen Bioscience). Cell 591 proliferation was either determined as measurement of cell density (confluence) or and through quantification 592 of red objects by using the IncuCyte ® NucLight Rapid Red Reagent (Essen Bioscience). Staining was 593 performed according to manufacturer's instructions. Viable cell number was determined by trypan blue staining 594 and automatic counting using a Countess™ Cell Counting Chamber Slide (Thermo Fisher). 595

Migration Assay (Wound Closure) 596
Cell migration was determined using the IncuCyte® Scratch Wound Assay system for 96 well plates (Essen 597 Bioscience). 96 well image lock plates (Essen Bioscience) were PLL coated prior to seeding of 40,000 cells 598 per well. Wound application was performed 24 h post seeding using the 96-pin IncuCyte WoundMaker Tool 599 (Essen Bioscience) and indicated individual treatment conditions were applied simultaneously. Measurement 600 for individual treatment conditions was performed in biological replicates (n = 4 -8) and repeated as 601 independent experiments as stated for the respective experiment (see figure legends). 602

Invasion and Migration Assay (Transwell Assay) 603
Cellular invasion and migration was analyzed by Boyden chamber assay as described in Meiser et al., 2018 604 [10]. Transwells were coated with ECM-Collagen for invasion assays and un-coated for migration assay.   injected with either 4T1 SCR or 4T1 MTHFD1L KD cells. Primary tumor growth was monitored between day 7 648 and day 35 of the experiment. Weight was monitored over time and no weight loss was observed. Experiment 649 was terminated after 6 weeks and lung and liver were prepared for examination of metastatic outgrowth. No 650 metastases were found in the liver. Macroscopic lung metastases were blindly counted under a microscope 651 and microscopic lung metastases were visualized by H&E staining. 652

H&E staining 653
For H&E staining, snap-frozen lungs were cut in 15 μm sections using Leica Cryostat (Leica Biosystems) and 654 embedded in Tissue-Tek ® O.C.T (Sakura). 5 sections in a distance of 150 μm to each other were stained per 655 lung. Staining was performed according to established protocols. Briefly, selected sections were dehydrated 656 with MeOH and stained with Gill 2 hematoxylin. Sections were neutralized with successive washes of tap 657 water, hard water (10g MgSO4 and 0.7g NaHCO3 per L), and distilled water. Subsequently, sections were 658 stained with Eosin-solution and dehydrated through successive washes with 80%, 95%, 100% EtOH, and Xylol 659 prior to mounting. 660 Proteomics 661 SILAC (Stable Isotope Labeling with Amino acids in Cell culture) strategy was used for the proteomic analysis. 662 In general, MDA-MB468 cells were cultivated in DMEM-F12 SILAC medium supplemented either with Lys 0 663 and Arg 0 (light channel) or Lys 8 and Arg 10 (heavy channel). After 6 passages, labeling efficiency of heavy 664 channel was checked using LC-MS/MS. Cells in light channel were treated with 50 nM MTX for 24h and 72h, 665 while heavy channel was used as control. Cell pellets of biological duplicates were collected after the 666 treatment. Proteins were extracted in lysis buffer (50mM ammonium bicarbonate, 6M Urea, 2M Thio-urea, pH 667 8) following a 30 min incubation at 4 °C in the presence of protease inhibitors (cOmplete™ EDTA-free Protease 668 Inhibitor Cocktail, Roche). Following centrifugation at 16,000 g for 10 min, supernatants were taken for protein 669 quantification. Samples from light channel (50 μg protein) were mixed with control heavy channel (50 μg 670 protein) for protein reduction (5 mM DTT, 1 h incubation at 37°C) and alkylation (10 mM IAA, 45min in dark at 671 room temperature). Protein digestion was performed with Lys-C (FUJIFILM Wako, 125-05061) at 1:30 ratio 672 (enzyme/protein substances) for 4 h at 37 °C, then samples were diluted 4 times with 50 mM ammonium 673 bicarbonate and digested overnight with 1 μg of trypsin at 37 °C. Digestion was terminated through addition of 674 formic acid (1% final concentration). Digested peptides were cleaned up with reverse phase Sep-Pak C18 1 675 cc Vac Cartridge (Waters, WAT054955) and eluted with 1 mL 50% ACN. Eluted peptides were dried by 676 Speedvac (Thermo Fisher Scientific) and re-suspended in 0.1% formic acid. Peptide concentration was 677 measured with Nanodrop. Peptides were measured by LC-MS/MS on Q-Exactive HF mass spectrometer 678 (Thermo Fisher) connected to a Dionex Ultimate 3000 (Thermo Fisher). 500 ng of peptides were loaded onto 679 a trap column (Acclaim PepMap 75 μm x 2 cm, C18, 3 μm) and separated on a 25 cm Acclaim pepmap RSLC 680 column (75 μm x 25 cm, C18, 2 μm) using a 150 min gradient (2% to 90% acetonitrile) with a flow rate of 0.3 681 μL/min. MS data were acquired in data dependent mode (DDA). Survey scans of peptide precursors from 375 682 to 1500 m/z were performed at 70,000 resolution with a 3×10 6 ion count target and the top 12 abundant peaks 683 from survey scan were selected for fragmentation. Tandem MS was performed by isolation at 1.4 m/z with the 684 quadrupole, HCD fragmentation with a normalized collision energy of 28. The MS2 ion count target was set to 685 1×10 5 and the max injection time was 45 ms. Only precursors with a charge state of 2-7 were sampled for 686 MS2. The dynamic exclusion duration was set to 20 s with a 10 ppm mass tolerance around the selected 687 precursor and its isotopes. Each sample was analyzed twice as technical replicates. All raw data was analyzed 688 with MaxQuant (version 1.6.7.0) and searched with Andromeda against the Homo sapiens database from 689 Uniprot. The minimal peptide length was set to 7 amino acids and the maximum of 3 missed cleavages were 690 allowed. The search included variable modifications of methionine oxidation and N-terminal acetylation, 691 deamidation (N and Q) and fixed modification of carbamidomethyl cysteine. The "Match between run" was 692 checked within 1 min retention time window. Mass tolerance for peptide precursor and fragments were set as 693 10 ppm and 20 ppm, respectively. The FDR was set to 0.01 for peptide and protein identifications. SILAC 694 based protein quantification (MaxQuant built-in) was used for quantitative evaluation of identified protein. 695 ProTIGY (https://github.com/broadinstitute/protigy), an R based tool, was used for differential analysis of 696 MaxQuant output. 697

Statistics 698
Unpaired t-test with Welch's correction was applied for pairwise comparison (two-sided) using GraphPad 699 Software Vers.8. For normalization, data points of one experiment were either normalized to the untreated 700 control or divided by the global mean of this individual experiment. We define one n as one independent 701 biological experiment (consisting of several wells, e.g. triplicate wells for all stable isotope tracing experiments). 702 The mean of one experiment was considered as one n.