Repurposing the Antidepressant Sertraline as SHMT Inhibitor to Suppress Serine/Glycine Synthesis–Addicted Breast Tumor Growth

This article is featured in Highlights of This Issue, p. 1

Rewiring of energy metabolism, exemplified by the Warburg effect, is one of the hallmarks of cancer (1). Although normal cells often rely on serine/glycine uptake from their environment, several cancer subtypes produce their own serine/glycine via intracellular serine/glycine synthesis and become addicted to this own production (2, 3). In general, serine/glycine synthesis (Fig. 1) consists of two processes: de novo serine synthesis from glucose and interconversion of serine into glycine. De novo serine synthesis branches from glycolysis, with the glycolytic intermediate 3-phosphoglycerate being converted into serine via three consecutive enzymatic reactions catalyzed by phosphoglycerate dehydrogenase (PHGDH), phosphoserine aminotransferase 1 (PSAT1), and phosphoserine phosphatase (PSPH). Thereafter, serine is catabolized into glycine, facilitated by serine hydroxymethyltransferase 1 or 2 (SHMT1/2), with 1 and 2 referring to the cell compartment in which the reaction takes place, i.e., the cytosol or the mitochondria (2, 3). By relying on intracellular serine/glycine synthesis, cancer cells feed their high requirements to generate formate for purine synthesis, produce reductive equivalents to control redox homeostasis, regulate DNA demethylation, and support lipid metabolism (3).

A cancer type well known for its dependency on serine/glycine synthesis is breast cancer, where 6% of the patient samples display copy-number gains of the PHGDH gene. Furthermore, 70% of estrogen receptor–negative breast tumors have increased PHGDH protein levels, and inhibition of PHGDH via RNA interference or PHGDH inhibitors impairs cell proliferation and survival (4–8). Besides PHGDH, SHMT2 is also known to be exploited by breast cancer cells as SHMT2 expression levels are positively correlated with breast cancer grade (9). Additionally, SHMT2 is identified as a direct target gene of the MYC oncogene (10). MYC-driven stimulation of de novo serine synthesis by transcriptional upregulation of PHGDH, PSAT1, and PSPH is also critical for sustaining survival and rapid proliferation of cancer cells under nutrient deprived conditions (11). Apart from MYC, the oncogene KRAS, the tumor suppressor p53, the mTOR–ATF4 axis, and the T-cell leukemia–associated R98S mutation in ribosomal protein L10 (RPL10 R98S) have also been shown to enhance serine/glycine synthesis in cancer cells (12–15).

Evidence for addiction to serine/glycine synthesis in cancer subsets, including triple-negative breast cancer and T-cell leukemia that are both currently treated with toxic intensive chemotherapy regimens, is growing. This highlights the necessity to develop novel therapeutic intervention strategies for these cancers, specifically focusing on targeting serine/glycine synthesis. PHGDH and SHMT inhibitors have been identified but did not enter clinical trials due to unfavorable pharmacokinetic profiles or because they have only recently been developed (7, 8, 16–18). Further efforts to identify clinically applicable drugs targeting this pathway are therefore required.

In this study, we made use of a lower eukaryotic yeast model system that specifically upregulates serine/glycine synthesis in response to sublethal stress. Using this platform, we discovered two repurposed compounds, sertraline and thimerosal (chemical structures in Fig. 5E; ref. 19), that show selective toxicity to serine/glycine synthesis-addicted cancer cell lines, while they had no effect on cancer and normal lymphoid cell lines that take up serine and glycine from their environment. Moreover, this work reports the potential application of the clinically used antidepressant sertraline as adjuvant therapeutic agent to treat serine/glycine synthesis–addicted cancers, especially when combined with drugs causing mitochondrial dysfunction.

Breast cancer cell lines MDA-MB-231 (ATCC #HTB-26, RRID:CVCL_0062), MDA-MB-468 (ATCC #HTB-130, RRID:CVCL_0419), MCF7 (ATCC #HTB-22, RRID:CVCL_0031), and HCC70 (ATCC #CRL-2315, RRID:CVCL_1270) were cultured at 37°C and 5% CO₂ in DMEM (Gibco #41965039) supplemented with 10% fetal bovine serum (FBS; Gibco #10270-106). Because of many serial passages, MCF7 cell authenticity was confirmed by Microsynth AG. Other cell lines were freshly obtained from ATCC. The Ba/F3 pro-B cell line (DSMZ #ACC-300, RRID:CVCL_0161) was obtained from Leibniz-Institute DSMZ and grown in RPMI-1640 (Gibco #21875091) supplemented with 10% FBS and 10 ng/mL interleukin 3 (IL3; Miltenyi). CRISPR-Cas9 engineering of Ba/F3 cells is described in the supplementary information. All cell lines were tested for Mycoplasma with the MycoAlert Mycoplasma Detection Kit (Lonza #LT07–418) before use and were cultured for maximum 49 days after thawing. Serine-depleted media correspond to DMEM without serine and glycine (US Biological Life Sciences #D9802-01), supplemented with 4.5 g/L glucose, 3.7 g/L sodium bicarbonate, 1:100 glutamax and 10% dialyzed serum (Gibco #A3382001).

Stock solutions (1,000×) of each compound were made and stored at −20°C. Sertraline (Sigma-Aldrich #S6319), NCT-503 (Sigma-Aldrich #SML1659), rotenone (Sanbio #13995-100), antimycin A (Sigma-Aldrich #A8674), and artemether (TCI Europe #A2190) were dissolved in DMSO, whereas thimerosal (Sigma-Aldrich #T8784), benzalkonium chloride (Sigma-Aldrich #12060), and bupropion (TCI Europe #B3649) were dissolved in water. Sertraline's activity can differ between batches, causing variability in the exact dose that is required to obtain the observed effects.

MDA-MB-231 (7,500 cells/well), MDA-MB-468 (7,500 cells/well), MCF7 (4,000 cells/well), and HCC70 (4,500 cells/well) cells (100 μL) were seeded in 96-well plates (Sigma-Aldrich #Z707902-108EA) and incubated at 37°C for 24 hours to obtain optimal adherence to the surface. Next, 100 μL of 2× compound solutions, diluted in DMEM with 10% FBS, were added. During the following 5 days, cell proliferation was assessed by real-time imaging of confluency on an IncuCyte Zoom system (Essen BioScience). Using the exponential portions of the resulting growth curves, we calculated the growth rate (GR) under various drug conditions. The formula below was used, in which t1 and t2 are two time points within the exponential growth phase: GR (1/h) = [ln(confluency t2) – ln(confluency t1)]/[time (t2) – time (t1)].

Three RPL10 WT versus three RPL10 R98S clones were plated and cultured for 72 hours to exhaust the medium. Afterward, the clones were incubated with sertraline (7.3 μmol/L) or thimerosal (1 μmol/L). Relative cell viability was measured, using flow cytometry, after 48 hours of compound treatment.

MDA-MB-468 cells (150,000) were plated in 3 mL of DMEM in 6-well plates (VWR #734-0948). After 24 hours of incubation at 37°C, cells were washed with PBS, and 3 mL of tracing medium (glucose-free DMEM with 10% dialyzed serum and 4.5 g/l ¹³C₆-glucose) was added. Subsequently, cells were incubated at 37°C during 24 hours (thimerosal and NCT-503) or 72 hours (sertraline and/or artemether). Metabolites for the subsequent mass spectrometry analysis were prepared by quenching the cells in liquid nitrogen followed by a cold two-phase methanol–water–chloroform extraction (20, 21). Phase separation was achieved by centrifugation at 4°C (24 × 3.75 × g, 10 minutes). The methanol–water phase containing polar metabolites was separated and dried using a vacuum concentrator at 4°C overnight. Dried metabolite samples were stored at −80°C. Polar metabolites were analyzed by GC-MS and LC-MS, as described in the supplementary information.

PHGDH activity upon drug treatment was tested using human PHGDH (BPS Bioscience #71079) and a colorimetric PHGDH activity kit (BioVision #K569). PHGDH inhibitor NCT-503 served as a positive control (7). Human PHGDH was diluted in water to a concentration of 0.15 mg/mL. Next, 5 μL of this PHGDH enzyme solution and 10 μL sertraline, thimerosal, or NCT-503 was added in a 96-well plate (flat bottom). Subsequently, 35 μL PHGDH assay buffer (BioVision #K569) and 50 μL PHGDH reaction mix (prepared as described in #K569, BioVision) was added. Afterward, absorbance at 450 nm was measured over time. In between measurements, the plate was incubated at 37°C, protected from light. Finally, PHGDH activity was calculated between two time points within the linear range of the internal standard (as described in #K569, BioVision).

Sertraline was modeled using MOE (Chemical Computing Group; MOE, RRID:SCR_014882) with the MMFF94x force field. The structures of the putative enzymes present in serine/glycine synthesis were obtained from the RCSB database (PDB ID SHMT1: 1BJ4, SHMT2: 5V7I). The bioactive conformations were chosen for each enzyme (SHMT1, SHMT2 as dimers) and optimized in MOE using protonate_3D. Docking was performed using GOLD8 software (ref. 22; GOLD, RRID:SCR_000188). Specifically, sertraline was docked in the presence of the pyridoxal 5′-phosphate (PLP) cofactor. Furthermore, the known SHMT inhibitor (pyrazolopyran scaffold, SHIN1) and a library of FDA-approved compounds (17, 23) were docked in the presence of the PLP ligand. Conformational restraints were applied to the ligand by disallowing the flipping of ring conformations and planar R-NR1R2 groups to ensure the rigidity of sertraline. Each ligand was docked 10 times into each enzyme, and the score was calculated using CHEMPLP scoring function. The interactions of sertraline with SHMT1/2 were analyzed and visualized using PyMOL version 1.8 (Schrödinger, 2015; PyMOL, RRID:SCR_000305). The docked conformations of the ligands in SHMT1/2 were superimposed using PyMOL to compare the interactions. Ligand efficiency scores were calculated to normalize each docking score by the number of heavy atoms in each ligand.

MDA-MB-468 cells (150,000) were plated in 2 mL of DMEM in 6-well plates (VWR #734-0948). After 24 hours of incubation at 37°C, cells were washed with PBS, and 2 mL of serine-free DMEM (US Biological Life Sciences #D9802-01), supplemented with 4.5 g/L glucose, 3.7 g/L sodium bicarbonate, 400 μmol/L [2,3,3-²H]-serine, 400 μmol/L glycine, 1:100 glutamax, and 10% dialyzed serum, was added for 48 hours. Metabolites for the subsequent mass spectrometry analysis were prepared by quenching the cells in liquid nitrogen followed by a cold two-phase methanol–water–chloroform extraction (20, 21). Phase separation was achieved by centrifugation at 4°C (24 × 3.75 × g, 10 minutes). The methanol–water phase containing polar metabolites was separated and dried using a vacuum concentrator at 4°C overnight. Dried metabolite samples were stored at −80°C. Polar metabolites were analyzed by GC-MS and LC-MS, as described in the supplementary information.

Recombinant human SHMT1 (Novusbio #NBP1-72548) was diluted in buffer (20 mmol/L Tris-HCl pH 8.0 and 100 mmol/L NaCl) to a concentration of 5 μmol/L. Thermal scanning was performed in clear 96-well PCR plates within each well 4 μmol/L SHMT1, 5× SYPRO Orange (Invitrogen #S6650; 5,000×) and 1.5 μL sertraline (7 mmol/L stock in 50% DMSO) in a total volume of 15 μL. In each well, the DMSO background was normalized to 5%. Dual SHMT1/2 inhibitor SHIN1, at a concentration of 250 μmol/L, served as a positive control (17). Unfolding was monitored by real-time PCR (CFX Connect, Bio-Rad Laboratories) and fluorescence intensity was measured every 20 seconds in the FRET channel over a linear 20°C to 95°C gradient of 1.5°C/min. Plots of the first derivative of fluorescence versus temperature were generated in CFX software.

Recombinant human SHMT2 (OriGene #TP760127) was fluorescently labeled using the Monolith NT protein labeling kit Red-NHS (NanoTemper Technologies). For microscale thermophoresis (MST), the SHMT2 enzyme concentration was kept constant at 50 nmol/L, whereas sertraline was titrated down from 730 μmol/L to 0.36 μmol/L. The DMSO background was kept constant at 5%, and dual SHMT1/2 inhibitor SHIN1 (500 μmol/L) served as a positive control (17). The MST buffer contained 25 mmol/L Tris-HCl pH 8.0, 150 mmol/L NaCl, 0.5% glycerol, and 0.05% sarkosyl for determining sertraline's binding affinity to SHMT2. MST with SHIN1 was performed in 25 mmol/L Tris-HCl pH 8.0, 150 mmol/L NaCl, 0.1% glycerol, and 0.01% sarkosyl. Measurements were performed using Monolith NT automated premium capillaries and recorded on a Monolith NT automated instrument (NanoTemper Technologies) with a pico-red laser channel at 7% excitation power and “medium” MST power. The affinity constant (K_d) and experimental data fitting was performed using MO.Affinity analysis software. The thermophoretic movement of bound and unbound state superposed linearly. Therefore, the fraction bound (f) is described as F_norm = (1 − f) F_norm,unbound + (f) F_norm,bound, where F_norm is the normalized fluorescence, F_norm,unbound corresponds to the normalized fluorescence of the unbound state, and F_{norm, bound} is the normalized fluorescence of the bound state.

MDA-MB-468 cells (150,000) were plated in 2 mL of DMEM in 6-well plates (VWR #734-0948). The day after (day 0), cells were washed with PBS, and 2 mL of fresh DMEM was added. After 72 hours (day 3), medium samples were taken (0.5–1 mL). The cells were counted on days 0 and 3, using an automated cell counter. Medium samples were analyzed by GC-MS, as described in the supplementary information. Specifically, uptake and secretion rates were calculated by subtracting metabolite concentrations of treated samples from those of culture media incubated for the same amount of time but without cells. Finally, results were normalized for both cell number and growth rate.

Cell death was quantified using a combined Zombie Aqua–eFluor506 and Annexin V–PE staining. MDA-MB-468 cells (150,000) were plated in 2 mL of DMEM in 6-well plates (VWR #734-0948) and incubated at 37°C for 24 hours to obtain optimal adherence to the surface. The day after, cells were treated with 2 mL of fresh DMEM containing sertraline (5, 7.5, or 10 μmol/L) and/or artemether (80 μmol/L). Upon 24 hours of treatment, cells were collected in a 96-well U-bottom plate and washed with PBS. Afterward, cell viability was analyzed following incubation with Zombie Aqua (BioLegend #423102, 1:1,000 in PBS) for 20 minutes at RT in the dark, an Annexin V binding buffer (Invitrogen #V13246) washing step and Annexin V (IQ Products #IQP-120R, 1:100 in binding buffer) staining at RT in the dark for 15 minutes. All samples were acquired using a FACSCanto II flow cytometer and analyzed with FlowJo V10 (FlowJo, RRID:SCR_008520).

Cell cycle was analyzed by bromodeoxyuridine (BrdUrd) incorporation and propidium iodide (PI) staining. MDA-MB-468 cells (150,000) were plated in 2 mL of DMEM in 6-well plates (VWR #734-0948) and incubated at 37°C for 24 hours to obtain optimal adherence to the surface. The day after, cells were treated with 2 mL of fresh DMEM containing sertraline (5, 7.5, or 10 μmol/L) and/or artemether (80 μmol/L). Upon 24 hours of treatment, cells were incubated with 10 μmol/L BrdUrd (eBioscience #00-4440-51A) at 37°C for 1 hour. Next, cells were collected using trypsin and washed with PBS. Afterward, BrdUrd incorporation and PI staining were analyzed following 70% ice-cold ethanol fixation, 2 mol/L HCl denaturation, 0.5 mol/L EDTA (pH 8.0) neutralization, BrdUrd-FITC antibody (BioLegend #364104, RRID:AB_2564481) staining for 20 minutes at RT and 20 minutes incubation with PI solution (100 μg/mL) at RT. All samples were acquired using a FACSCanto II flow cytometer and analyzed with FlowJo V10 (FlowJo, RRID:SCR_008520).

Animal experiments were approved by the local ethics committee (P262-2015). Breast cancer cells (3 × 10⁶) were injected subcutaneously in the left (MDA-MB-231) and right (MDA-MB-468) flanks in a 1:1 mixture with matrigel (Corning). The mice (IMSR #JAX:005557, RRID:IMSR_JAX:005557) were monitored on a daily basis and sacrificed after 28 days. Mice received treatments on days 7, 9, 11, 13, 15, 20, and 24. Therapy was administered via intraperitoneal injections at dosages of 2.5 mg/kg sertraline and/or 40 mg/kg artemether. Control mice were treated with DMSO.

All statistical analyses were performed using GraphPad Prism 8 software (GraphPad Prism, RRID:SCR_002798), and data are presented as mean ± standard deviation (SD). Specific statistical tests used for each experiment are mentioned in the figure legends. Results were considered to be statistically significant if the adjusted P value was < 0.05 (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001).

To address the need for clinically applicable drugs targeting serine/glycine synthesis, we focused on drug repurposing with the aim to develop novel therapies that can rapidly enter the clinical practice as their safety and pharmacokinetics have been validated in patients before. To do this, we made use of the lower eukaryotic yeast Candida albicans, which specifically upregulates serine/glycine synthesis as a tolerance mechanism against sublethal stress (Supplementary Fig. S1A and S1B; refs. 24, 25), and hypothesized that compounds resensitizing this yeast to the antifungal stress can be potent inhibitors of serine/glycine synthesis. Subsequently, “resensitizing hits” were further validated in a cancer context, both in vitro and in vivo, by investigating their selective antiproliferative effects on serine/glycine synthesis–addicted cancer cell lines and delineating their exact mode-of-action (Fig. 1). Interestingly, the serotonin reuptake inhibitor sertraline was identified as a resensitizing agent, implying that it might stem from inhibition of serine/glycine synthesis (Supplementary Fig. S1C). This drug caught our attention, as sertraline is widely used in the clinic as antidepressant and its safety for chronic use is well documented.

To investigate whether sertraline is a specific inhibitor of serine/glycine synthesis, we tested its efficacy on cancer cell lines representing two subtypes of breast cancer, being those that rely on serine/glycine synthesis (MDA-MB-468) and those that depend on serine/glycine uptake (MDA-MB-231; refs. 6, 26). MDA-MB-468 cells typically express higher transcript and protein levels of serine/glycine synthesis enzymes (6, 26), and flux through this metabolic pathway is associated with proliferation and sustained survival (6). We performed ¹³C₆-glucose tracing to monitor the activity of serine/glycine synthesis in these cell lines. In such experiments, synthesis of serine and glycine can be monitored by following carbon-13 stable isotope incorporation from glucose, with glucose-derived serine and glycine showing mass shifts of 3 and 2 units (M + 3 serine and M + 2 glycine), respectively. Cellular uptake of unlabeled serine and glycine from the cell culture medium will, on the other hand, result in unlabeled (M + 0) serine and glycine, whereas interconversion between serine and glycine, catalyzed by SHMT1/2, will result in partially labeled serine (M + 1 and M + 2) and glycine (M + 1; Fig. 3D). Our data confirmed that MDA-MB-468 cells showed around 10% serine M + 3 and glycine M + 2 contribution from labeled glucose, whereas serine and glycine were unlabeled in MDA-MB-231 cells (Supplementary Fig. S2A). MDA-MB-468 cells are thus able to endogenously produce serine and glycine, whereas MDA-MB-231 cells are not. In agreement with this, real-time monitoring of cell confluency of both cell lines treated with NCT-503, an established PHDGH inhibitor (7), showed dose-dependent inhibition of MDA-MB-468 proliferation, whereas minimal effects on MDA-MB-231 were observed (Supplementary Fig. S2B and S2C). Accordingly, treatment of MDA-MB-468 cells with sertraline induced a dose-dependent impairment of cell proliferation, whereas MDA-MB-231 cells were hardly affected (Fig. 2; Supplementary Fig. S3). These results indicate that compounds resensitizing the yeast model system can specifically inhibit the proliferation of serine/glycine synthesis–addicted breast cancer cells.

Besides sertraline, a previously performed screening of 1,600 off-patent drugs and other bioactive compounds (Pharmakon 1,600 repurposing library) using the same yeast model system as described above resulted in the identification of 49 other “resensitizing hits” (Supplementary Table S1; refs. 24, 27). We hypothesized that this list might include other potent serine/glycine synthesis inhibitors (25). To select additional candidates for validation in serine/glycine synthesis–addicted breast cancer cell lines, all 49 agents were ranked into three classes based on their yeast resensitizing capacity (high, intermediate, and low). One representative agent of each class was retained for further validation, taking into account their clinical usage, as well as the presence of sulfhydryl containing groups that are proven to be effective for inhibition of PHGDH enzyme activity (8). Quaternary ammonium compounds came out as the top “hits” in the highest ranked class, of which we selected benzalkonium chloride. Out of the intermediate class of resensitizing agents, we selected thimerosal due to the presence of a reactive sulfhydryl group in its chemical structure. Bupropion, representing the lowest ranked class, was also retained for testing because it is clinically used as antidepressant, similar to sertraline. Thimerosal and benzalkonium chloride showed selective inhibitory activity against serine/glycine synthesis–addicted MDA-MB-468 cells (Fig. 2; Supplementary Fig. S3). Conversely, bupropion did not show any anticancer activity, not even at high concentrations (Fig. 2; Supplementary Fig. S3).

To validate these results, we selected a second pair of breast cancer cell lines, of which one is dependent on serine/glycine synthesis (HCC70) and one favors serine/glycine uptake from the microenvironment (MCF7; refs. 6, 26). As expected, PHGDH inhibitor NCT-503 induced a stronger inhibition of proliferation in HCC70 than in MCF7, especially at lower drug doses (Supplementary Fig. S4A and S4B). In agreement with previous results, sertraline, thimerosal, and benzalkonium chloride impaired HCC70 cell proliferation in a dose-dependent manner, whereas there was no or minor toxicity on MCF7 cells (Supplementary Fig. S5A and S5B). Overall, thimerosal showed the highest selectivity in targeting serine/glycine synthesis–addicted breast cancer cells as compared with sertraline and benzalkonium chloride, whereas sertraline, as an antidepressant, has highest potential for clinical use in cancer patients. We therefore selected both compounds for further characterization of their mode-of-action.

Next, we verified whether sertraline and thimerosal would be applicable as agents that target serine/glycine synthesis–addicted cancers other than breast cancer. The ribosomal RPL10 R98S mutation is detected in 8% of pediatric T-cell acute leukemia patients (28). In leukemic cells carrying this mutation, the mutant ribosomes accumulate on PSPH mRNAs, leading to elevated PSPH protein translation associated with enhanced serine/glycine synthesis. We showed that RPL10 R98S leukemic cancer cells depend on this elevated PSPH expression to support their proliferation and in vivo expansion (14). As such, RPL10 R98S leukemias represent a second cancer subgroup that is addicted to intracellular serine/glycine synthesis and we hypothesized that RPL10 R98S-mutant cells might be more sensitive to sertraline and thimerosal. We used mouse pro-B Ba/F3 cells in which the endogenous Rpl10 locus was engineered with CRISPR/Cas9 technology to express the WT or R98S-mutant form of RPL10, comparing three independent single-cell clones of each genotype. RPL10 R98S clones showed selective inhibition of cell survival upon sertraline or thimerosal treatment as compared with the RPL10 WT clones (Fig. 3A). These results confirm applicability of sertraline and thimerosal in serine/glycine synthesis–addicted cancer types other than breast cancer and suggest that these agents may target different serine/glycine synthesis–addicted cancer subtypes.

To define the mode-of-action of both compounds, we first evaluated whether serine depletion from the culture medium increases the sensitivity of MDA-MB-468 breast cancer cells to sertraline and thimerosal by enforced dependency on intracellular serine/glycine synthesis. Only sertraline induced a significantly stronger impairment of MDA-MB-468 cell proliferation under serine-depleted conditions (Fig. 3B; Supplementary Fig. S6A and S6B). A similar trend was observed with NCT-503 (Supplementary Fig. S6C and S6D).

Given that MDA-MB-468 cells rely more on synthesis than uptake, inhibition of serine/glycine synthesis could cause changes in intracellular serine/glycine abundance, depending on the specific synthesis enzyme that is targeted. Although reductions in enzymatic activity of PHGDH, PSAT1, or PSPH are expected to result in reduced intracellular concentrations of both serine and glycine, targeting SHMT1/2 is rather expected to affect levels of glycine only. Mass spectrometry analysis revealed that sertraline did not affect the intracellular levels of serine in MDA-MB-468 cells, and that it solely reduced the abundance of downstream glycine (Fig. 3C). In contrast, thimerosal decreased serine abundance to the same extent as glycine, as seen for NCT-503 (Fig. 3C; Supplementary Fig. S6E).

We subsequently performed ¹³C₆-glucose tracing to monitor the effects of sertraline and thimerosal on overall serine/glycine synthesis in more detail (Fig. 3D). Our data showed that sertraline reduced de novo synthesized M + 3 serine, and that it targets serine to glycine interconversion in MDA-MB-468 cells, characterized by a reduced fractional labeling of M + 2 glycine and M + 1 serine (Fig. 3E). In contrast to sertraline, thimerosal did not significantly reduce the fraction of glucose-derived M + 3 serine and M + 2 glycine (Fig. 3E). In parallel, NCT-503 only caused minor changes in fractional glucose contribution of M + 3 serine and M + 2 glycine that could not solely account for the lower intracellular serine and glycine abundances (Supplementary Fig. S6E and S6F). Because we assess with this tracing the ratio between synthesis and uptake, our data may indicate a simultaneous downregulation of serine/glycine synthesis and uptake upon thimerosal or NCT-503 treatment.

Collectively, we reasoned that thimerosal, similar to NCT-503, is more likely to affect one of the three first biosynthetic enzymes. For sertraline, our data point toward inhibition of serine to glycine interconversion, implying that SHMT1/2 can be its potential target. In line with this hypothesis, steady-state levels of serine (Fig. 3C) may originate from a block in conversion to glycine and subsequent downregulation of de novo serine synthesis and uptake.

To gain additional insights in the exact serine/glycine synthesis enzyme targeted by sertraline and thimerosal, we first performed an in vitro enzymatic PHGDH assay. Sertraline did not show any effect on PHGDH activity, further excluding PHGDH as sertraline's potential target (Fig. 4A). Similar to NCT-503, thimerosal was able to reduce PHGDH activity (Fig. 4A). Notably, the thimerosal concentration inducing 50% inhibition (IC₅₀ = 0.1517 μmol/L) was even 10-fold lower than this of NCT-503 (IC₅₀ = 1.747 μmol/L; Fig. 4A; Supplementary Fig. S7A).

Because our in vitro enzymatic PHGDH assay had excluded PHGDH as sertraline's target and previous data already pointed toward more downstream inhibition of serine/glycine synthesis (Fig. 3), we focused further on SHMT1/2. First, we computationally docked sertraline to SHMT1 and SHMT2. It has been shown that SHMT is an ubiquitous pyridoxal 5′-phosphate (PLP)-dependent enzyme (17, 29). Therefore, docking scores were determined with the PLP cofactor inside the binding pocket (Table 1). As a reference, the already reported plant-derived and nonclinically used dual SHMT1/2 inhibitor with a pyrazolopyran scaffold, SHIN1, was used (17, 30). High docking scores were obtained for sertraline, in the range of the compounds characterized by a pyrazolopyran scaffold. Moreover, aligning both dockings showed that sertraline potentially binds in the exact same pocket as SHIN1 and has similar interactions with SHMT1/2, namely, a hydrogen bond with its −NH group (Fig. 4B).

Next, a virtual library screen was performed by docking a library of 6,392 FDA-approved ligands to SHMT1/2 (23). We observed that sertraline scores are average for both SHMT1 and SHMT2 (Supplementary Fig. S7B). Because sertraline is a very small molecule with only one option to create a high scoring hydrogen bond, the scoring function will assign a lower docking score to sertraline as compared with larger molecules with more hydrophilic groups. We therefore calculated the ligand efficiency score, which adds a normalization for the amount of heavy atoms in each ligand to the docking score. This significantly improved the prediction for sertraline binding to SHMT1/2 and even assigned a higher score to sertraline as compared with the clinically used SHMT inhibiting antifolates pemetrexed and lometrexol (Supplementary Fig. S7B; Supplementary Tables S2 and S3).

To experimentally measure cellular SHMT activity upon sertraline treatment, we performed isotopic hydrogen tracing (Fig. 4C) in which MDA-MB-468 cells were incubated with [2,3,3-²H]-serine, followed by analysis of ²H incorporation in glycine and thymidine triphosphate (dTTP; ref. 31). These tracings revealed that sertraline increased the intracellular fraction of M + 3 serine and decreased the fraction of M + 0, M + 1, and M + 2 serine (Fig. 4D). Although M + 3 serine is used by SHMT1/2 to convert into glycine, partially labeled serine (M + 1 and M + 2) only arises from converting glycine back into serine. Furthermore, only M + 1 dTTP was detected in MDA-MB-468 (Supplementary Fig. S7C), supporting that these cells process their 1C units only in the mitochondria via SHMT2 (31, 32). Interestingly, the M + 1 dTTP fraction was decreased upon sertraline treatment (Fig. 4E). Accordingly, binding of sertraline to SHMT2 was confirmed by MST (K_d = 13.1 μmol/L; Fig. 4F), using SHIN1 as a positive control (Supplementary Fig. S7D; ref. 17). In addition, sertraline decreased the melting temperature of SHMT1 as measured by a thermal shift assay (Fig. 4G), consistent with decreased stability of SHMT1 upon binding of SHIN1 (Supplementary Fig. S7E). These findings identify sertraline as dual SHMT1/2 inhibitor.

In contrast to what may be expected from an SHMT inhibitor, sertraline did not increase the intracellular serine levels (Fig. 3C). To look closer into this, we determined the impact of sertraline on serine/glycine uptake and secretion rates in MDA-MB-468 cells (Fig. 4H). Our data showed a dose-dependent reduction of net serine uptake. Upon the highest sertraline dose, we even observed net serine secretion. These observations can explain the lack of serine accumulation in these breast cancer cells. Whether these effects are due to direct serine secretion and/or a sensory feedback mechanism of pyruvate kinase (PKM2), an important regulator of the glycolytic flux, remains to be determined. Furthermore, measuring glycine uptake and secretion rates revealed that sertraline-treated MDA-MB-468 cells have the tendency to take up less glycine (Fig. 4H). The strength of blocking both the serine into glycine interconversion and glycine uptake is what makes SHMT inhibitors more cytotoxic, explaining sertraline's potent activity on MDA-MB-468 cells, as was previously also observed using SHIN1 (17). Together, these results support that sertraline inhibits SHMT1/2 activity and glycine uptake, leading to reduced net serine uptake and de novo serine synthesis, to prevent serine accumulation.

In contrast to thimerosal, for which clinical applications are limited to topical use because of a toxic mercury group in its structure, sertraline is a small-molecule antidepressant. Our findings support that sertraline can be an attractive adjuvant therapeutic agent to specifically treat serine/glycine synthesis–addicted cancers.

It has been established that mitochondrial dysfunction causes changes in one-carbon metabolism and that cancer cells will depend even more on serine/glycine synthesis upon mitochondrial inhibition (33, 34). Furthermore, serine/glycine synthesis and the tricarboxylic acid (TCA) cycle are strongly interconnected, implying that targeting both serine/glycine synthesis and mitochondrial metabolism might further metabolically disrupt cancer cells. As a proof-of-concept, we investigated the effect of sertraline on MDA-MB-468 proliferation when combined with suboptimal dosages of mitochondrial inhibitors rotenone and antimycin A, targeting respiratory chain complexes I and III, respectively (35). Although sertraline monotherapy caused around 20% reduction in cell proliferation, combining sertraline with rotenone or antimycin A further decreased proliferation of MDA-MB-468. No synergistic effect of both combinations on MDA-MB-231 proliferation was observed (Fig. 5A; Supplementary Fig. S8A–S8C).

To further analyze sertraline's clinical potential in combination therapy, we tested sertraline in combination with a clinically used and less generally toxic drug than rotenone and antimycin A. Artemether, an antimalarial agent, has already been shown to have potent anticancer activity (36, 37). Interestingly, artemisinin, which belongs to the same structural class as artemether, was identified as activator of AMPK, suggesting a possible inhibition of mitochondrial function (38). By performing ¹³C₆-glucose tracing of artemether-treated MDA-MB-468 cells (Supplementary Fig. S9), we confirmed inhibitory action of artemether on mitochondrial TCA cycle activity, as evidenced by decreased contribution of labeled glucose into TCA cycle metabolites, especially the M + 4 variants. Besides, a slight reduction of the M + 3 serine fraction was detected, but this was lower than the observed inhibitory effect on TCA cycle intermediates. Additionally, glucose contribution into M + 3 lactate and M + 3 alanine was elevated upon artemether treatment, which was not observed with sertraline. The latter suggests that overall TCA cycle activity is reduced, and pyruvate will use alternative routes such as its reduction to lactate, supporting that artemether acts as mild mitochondrial inhibitor. Corresponding to what was observed using the well-established mitochondrial inhibitors rotenone and antimycin A, the artemether–sertraline combination indeed further decreased MDA-MB-468 cell proliferation as compared with single drug treatment, whereas minimal effects were observed on MDA-MB-231 cells (Fig. 5A; Supplementary Fig. S8B and S8C). ¹³C₆-glucose tracing confirmed the superior effect of this novel combination therapy as the contribution of labeled glucose into TCA cycle metabolites was strongly reduced (Supplementary Fig. S9). In contrast, the reduction in M + 3 serine and M + 2 glycine was identical to that of sertraline monotherapy, confirming that the inhibitory effect on serine/glycine synthesis is rather due to sertraline than to artemether.

Because the artemether–sertraline combination mainly reduced the MDA-MB-468 proliferation rate, we hypothesized that it interferes with cell-cycle progression rather than inducing cell death. Indeed, BrdUrd staining showed a significant decrease in BrdUrd incorporation in MDA-MB-468 cells upon combination treatment, compared with monotherapy (Fig. 5B and C; Supplementary Fig. S10B). Moreover, PI cell-cycle analysis confirmed a G₁–S cell-cycle arrest, with 20% more MDA-MB-468 cells in G₁-phase upon combination treatment (Fig. 5B and C; Supplementary Fig. S10B). Only 5% of MDA-MB-468 cells were double positive for the Annexin V and Zombie Aqua cell death markers after combination treatment (Supplementary Fig. S10C), supporting limited effects on cell apoptosis. Apart from the artemether–sertraline combination, minor effects on cell-cycle were observed with a suboptimal dose of sertraline that increases with increased doses of sertraline (Supplementary Fig. S10A and S10B). As such, artemether strengthens sertraline's inhibitory effect on G₁–S cell-cycle progression, highlighting the potential of a mitochondrial inhibitor in combination with sertraline.

To evaluate the therapeutic potential of these findings more profoundly, we tested the efficacy of the artemether–sertraline combination in an in vivo mouse model. To this end, we implanted MDA-MB-231 and MDA-MB-468 cells in the opposite flanks of immunodeficient NOD-SCID/IL2γ^−/− (NSG) mice and treated the animals with sertraline (2.5 mg/kg), artemether (40 mg/kg), or a combination of both. After 4 weeks, only the MDA-MB-468 mouse xenografted tumors showed significant differences between treatment groups, and artemether significantly reduced MDA-MB-468 tumor growth (Fig. 5D). Most notably, the artemether–sertraline combination caused an even stronger inhibition of MDA-MB-468 tumor formation as compared with monotherapy (Fig. 5D). No changes in MDA-MB-231 tumor growth were observed between treatment groups within the same mice (Fig. 5D). Collectively, this highlights the potency of a mitochondrial inhibitor, such as artemether, in combination with sertraline to treat serine/glycine synthesis-addicted cancers.

Our data support upregulation of serine/glycine synthesis as a general mechanism to deal with cellular stress that has been evolutionarily conserved from the lower eukaryote Candida albicans up to human cancer cells. Hence, this yeast model system can be used for rapid and low-cost screening for compounds targeting serine/glycine synthesis. Further characterization of the mode-of-action of “resensitizing hits” revealed that sertraline and thimerosal target intracellular serine/glycine synthesis enzymes SHMT1/2 and PHGDH, respectively (Fig. 5E).

Although thimerosal showed potent PHGDH inhibition with 10-fold lower IC₅₀ values as compared with NCT-503, its clinical applications are limited to topical use due to the presence of a toxic mercury group in its chemical structure. Therefore, we tested thimerosal derivatives that do not contain a mercury group (methyl thiosalicylate, thiosalicylic acid, and sodium thiosalicylate) for their potential to inhibit MDA-MB-468 proliferation. However, these compounds did not affect MDA-MB-468 proliferation, nor could they inhibit in vitro PHGDH enzymatic activity. In agreement with these observations, it was previously described that thimerosal's sulfhydryl reactive properties, which are expected to be important for inhibiting PHGDH activity, are attributable to its mercury-containing chemical structure (8, 39). In line with an important role for sulfhydryl reactive properties in inhibiting PHGDH, compound CBR-9480, being very similar in chemical structure with thimerosal, was also picked up as a potent PHGDH inhibitor (8). As opposed to thimerosal, the antidepressant sertraline has high clinical potential as targeting agent for serine/glycine synthesis–addicted cancers.

Besides T-cell leukemia (14, 16) and breast cancer, upregulation of serine/glycine synthesis is also known in other cancer types such as melanoma, glioblastoma, prostate, testis, ovary, liver, kidney, lung, and pancreas cancers (40, 41). Expression levels of serine/glycine synthesis enzymes reaching above 4-fold increased expression as compared with normal tissue controls (based on our Ba/F3 RPL10 WT and R98S model) may serve as a biomarker to identify serine/glycine synthesis–addicted cancers that may benefit from sertraline treatment.

Whereas antitumor activity of sertraline has previously been described (42–44), our findings pinpoint the exact mode-of-action of its anticancer activity, and support that this agent could also be an attractive adjuvant therapeutic agent to specifically treat the subset of serine/glycine synthesis–addicted cancers. Moreover, suppression of serine/glycine synthesis has been shown to resensitize therapy-resistant tumors to anticancer treatment. In particular, PHGDH targeting resulted in resensitization to doxorubicin, vemurafenib, and HIF2α-antagonist in therapy-resistant triple-negative breast cancer cells, melanoma with oncogenic BRAF V600E mutations, and advanced renal cell carcinoma, respectively (45–47). Additionally, sertraline passes the blood–brain barrier, which may even open opportunities to target serine/glycine synthesis–addicted brain tumors.

The doses at which we observed activity against serine/glycine synthesis–addicted cancer cell lines are in the range of concentrations at which sertraline's antitumor activity was described by others (42–44). Clinical use of sertraline as adjuvant anticancer therapy will only be possible if the dosages required for anticancer efficacy are not reaching toxicity in humans. As an antidepressant, sertraline is administered in a dosage ranging from 50 to 200 mg/day, resulting in serum concentrations between 0.065 and 0.54 μmol/L (48). Considering that the average body weight of an adult in Europe is 70.8 kg (49), sertraline doses vary between 0.7 and 2.8 mg/kg. Furthermore, sertraline has a linear pharmacokinetic profile and daily oral intake of 400 mg (5.6 mg/kg) is well tolerated in patients (43, 48). In our in vivo experiments, we treated the mice with 2.5 mg/kg sertraline, which thus is in the range of the administered doses in humans. Hassell and colleagues even used 60 mg/kg sertraline in mice, and no adverse effects were observed (43). This suggests that therapeutics doses of sertraline can be achieved in cancer patients.

Serine/glycine uptake-dependent cancers can become addicted to serine/glycine synthesis upon serine starvation (50). However, metabolic adaptation of synthesis-addicted cancer cells by reuptake of serine/glycine from their environment under long-term SHMT or PHGDH inhibition has not been reported yet. Such a potential resistance mechanism again underscores the importance of using a combination therapy as we propose.

In conclusion, we identified the widely used antidepressant sertraline as a novel inhibitor of serine/glycine synthesis enzyme SHMT that demonstrated high efficacy against metabolically addicted cancers. Whereas previously identified inhibitors of serine/glycine synthesis enzymes did not reach clinical trials, sertraline is a clinically used drug that can safely be used in humans. Our results suggest that sertraline may have applications as adjuvant therapeutic intervention strategy for serine/glycine synthesis–addicted cancers, especially when combined with drugs attacking other central nodes of cancer cell metabolism such as mitochondrial inhibitors. Collectively, this work provides a novel and cost-efficient treatment option for the rapidly growing list of serine/glycine synthesis–addicted cancers.

S.L. Geeraerts reports grants from Fonds Wetenschappelijk Onderzoek (FWO; 1S14517N) during the conduct of the study and has a patent for pharmacological targeting of serine/glycine synthesis (PCT/EP2019/072826 27.08.2019) pending. K.R. Kampen reports a patent for pharmacological targeting of serine/glycine synthesis (PCT/EP2019/072826 27.08.2019) pending. G. Rinaldi reports grants from Kom op tegen Kanker and FWO during the conduct of the study. N. Louros reports grants from Fund for Scientific Research Flanders post-doctoral fellowship (FWO 12P0919N) during the conduct of the study. E. Heylen reports grants from Fonds voor Wetenschappelijk Onderzoek (FWO) during the conduct of the study. S. Vereecke reports grants from FWO—Vlaanderen during the conduct of the study. B. Verbelen reports grants from FWO during the conduct of the study. S.-M. Fendt reports grants from KU Leuven Methusalem, co-funding and grants from Fonds Baillet Latour during the conduct of the study, as well as other from Merck, Black Belt Tx, and Fund+ outside the submitted work. B.P.A. Cammue reports a patent for pharmacological targeting of serine/glycine synthesis (PCT/EP2019/072826 27.08.2019) pending. K. Thevissen reports a patent for pharmacological targeting of serine/glycine synthesis (PCT/EP2019/072826 27.08.2019) pending. K. De Keersmaecker reports grants from Stichting Tegen Kanker and KU Leuven Research Council during the conduct of the study, and has a patent for pharmacological targeting of serine/glycine synthesis (PCT/EP2019/072826 27.08.2019) pending. No disclosures were reported by the other authors.

S.L. Geeraerts: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, and writing–review and editing. K.R. Kampen: Conceptualization, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, and writing–review and editing. G. Rinaldi: Formal analysis and investigation. P. Gupta: Formal analysis, investigation, and visualization. M. Planque: Formal analysis and investigation. N. Louros: Formal analysis and investigation. E. Heylen: Investigation. K. De Cremer: Conceptualization. K. De Brucker: Conceptualization. S. Vereecke: Investigation. B. Verbelen: Investigation. P. Vermeersch: Investigation. J. Schymkowitz: Supervision. F. Rousseau: Supervision. D. Cassiman: Conceptualization. S.-M. Fendt: Supervision, methodology, and writing–original draft. A. Voet: Supervision, investigation, methodology, and writing–original draft. B.P.A. Cammue: Conceptualization, resources, supervision, funding acquisition, and writing–original draft. K. Thevissen: Conceptualization, resources, supervision, investigation, methodology, writing–original draft, and writing–review and editing. K. De Keersmaecker: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, and writing–review and editing.

We thank Annelies Peeters and Tamara Davenne for technical support. S.L. Geeraerts received a SB PhD fellowship from “Fonds Wetenschappelijk Onderzoek (FWO)” (1S14517N). K.R. Kampen was supported by a PDM postdoctoral mandate fellowship obtained from the KU Leuven, the “Emmanuel van der Schueren” postdoctoral fellowship from “Kom op tegen Kanker,” and a research grant from FWO (FWO KAN2018 1501419N). G. Rinaldi is supported by consecutive PhD fellowships from the “Emmanuel van der Schueren – Kom op tegen Kanker” foundation and FWO (1137117N and 1137119N). P. Gupta and A. Voet acknowledge a research grant from FWO (G0F9316N Odysseus). The Switch Laboratory was supported by grants from the European Research Council under the European Union's Horizon 2020 Framework Programme ERC Grant Agreement [647458 (MANGO) to J. Schymkowitz]; the Flanders Institute for Biotechnology (VIB; grant no. C0401); the Industrial Research Fund of KU Leuven [“Industrieel Onderzoeksfonds (IOF)”]; the Funds for Scientific Research Flanders (FWO; Hercules Foundation grant AKUL/15/34–G0H1716N); and the Flanders Agency for Innovation by Science and Technology (IWT; SBO grant 60839). N. Louros was funded by Fund for Scientific Research Flanders Post-doctoral fellowship (FWO 12P0919N). E. Heylen received an FWO PhD fellowship fundamental research (1106121N). S. Vereecke and B. Verbelen are SB PhD fellows at FWO (1S07118N and 1S49817N). P. Vermeersch and D. Cassiman are senior clinical investigators of the Research Foundation–Flanders. S.-M. Fendt acknowledges FWO funding and KU Leuven Methusalem co-funding. K. Thevissen received a mandate as Innovation Manager from IOF Fund of KU Leuven. This research was funded by a grant from “Stichting Tegen Kanker” (2016-112) and by funding from the KU Leuven Research Council (C1 grant C14/18/104) to K. De Keersmaecker.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.