Translation inhibition drives TXNIP expression
To determine whether TXNIP expression is generally correlated with protein synthesis, we investigated how the expression of several known translation regulators correlate with TXNIP expression. Using the Gene-tissue Expression Database (GTEx) and examining expression in blood, we identified a strong negative correlation between TXNIP and ribosomal protein L24 (RPL24), which has been shown to correlate well with global changes in protein synthesis in lymphocytes (28, 29), supporting the hypothesis that high translation rates suppress TXNIP expression (Figure 1A).
We next determined whether compounds that block translation at different steps regulate TXNIP expression. We found that treatment of Hela cells with three translation elongation inhibitors, cycloheximide (CHX), emetine, and puromycin, increased TXNIP expression dramatically (Figure 1B). Likewise, the translation initiation inhibitor RocA (30), induced TXNIP expression comparably to CHX (Figure 1C). As expected (6), TXNIP induction by RocA was accompanied by a decrease in glucose uptake (Figure 1D). siRNA-mediated knockdown of translation initiation factor EIF4E also increased TXNIP expression (Figure 1E-F), confirming our findings with the pharmacological inhibitors. It is counter-intuitive that TXNIP protein would accumulate following knockdown of EIF4E; however, TXNIP undergoes both cap-dependent and IRES-dependent translation (31). Therefore, we speculate that IRES-dependent translation accounts for the increase in TXNIP protein levels following EIF4E knockdown. Our previous studies demonstrated that mTORC1 suppresses MondoA transcriptional activity and TXNIP expression by competing for its obligate transcriptional partner Mlx (19). Consistent with our previous findings, the mTORC1 inhibitor Torin increased TXNIP expression, but this increase was much more modest than that observed with CHX (Figure 1G). This finding suggests that broad translation inhibitors like RocA and CHX increase TXNIP expression by a different mechanism than does Torin and their action is largely independent of mTORC1. Finally, CHX increased TXNIP expression in C2C12 and L6 myoblasts and HEK293T embryonic kidney cells (Figure 1H-J), suggesting that protein synthesis inhibitors generally increase TXNIP expression. Together these findings suggest that TXNIP expression, and consequently glucose uptake, is tightly linked to translation rate.
Protein synthesis inhibitors drive MondoA transcriptional activity
We next evaluated the involvement of MondoA in TXNIP induction in response to protein synthesis inhibition. CHX treatment increased TXNIP expression, in wildtype but not in MondoA-/- mouse embryonic fibroblasts (MEFs) (Figure 2A). Ectopic expression of MondoA in MondoA-/- MEFs rescued TXNIP induction (Figure 2B). We tested whether CHX increased MondoA transcriptional activity using several approaches. First, the nuclear localization of MondoA and the amount of MondoA on the TXNIP promoter increased following CHX treatment (Figure 2C-D). Second, CHX increased the expression from a TXNIP luciferase reporter construct in a manner that was strongly dependent on an intact CACGAG ChoRE about 80 bp upstream of the transcription start site (Figure 2E). Together these data demonstrate that CHX, and likely other protein synthesis inhibitors, drive MondoA nuclear accumulation, promoter binding and transcriptional activity.
Because MondoA transcriptional activity is strictly dependent on glucose (10, 11), we next determined the requirement for glucose in CHX-driven TXNIP expression. HeLa cells were treated with CHX in DMEM or in glucose-free DMEM. Surprisingly, TXNIP was induced in both media conditions (Figure 2F), suggesting that CHX might induce MondoA transcriptional activity independent of glucose. An alternate possibility is that Fetal Bovine Serum (FBS) contains sufficient glucose (~ 5 mM) such that when present in culture medium at 10% the resulting concentration of glucose (~0.5 mM) can support MondoA transcriptional activity. To test this hypothesis, we dialyzed FBS to remove small molecules including glucose and then treated cells with CHX in glucose-free DMEM + 10% dialyzed FBS. CHX did not increase TXNIP expression in medium containing dialyzed serum; however, adding glucose back to the medium that contained dialyzed serum rescued TXNIP induction (Figure 2G-H). CHX increased TXNIP expression at all glucose levels tested, and surprisingly decreased the threshold of glucose required for TXNIP induction ~ 5-fold (Figure 2H). Further, RocA, showed glucose-dependent changes in TXNIP expression (Supplemental Figure 1A). Thus, glucose is strictly required for CHX to increase MondoA transcriptional activity and also sensitizes MondoA transcriptional activity to lower glucose levels.
Protein synthesis inhibition drives G6P production
We next investigated how protein synthesis inhibitors increase MondoA transcriptional activity. We focused on a potential role for mitochondrial function and mtATP for three reasons: 1) protein translation is the most ATP-consuming biosynthetic reaction, 2) MondoA transcriptional activity depends on mtATP (13), 3) higher mtATP levels may sensitize MondoA transcriptional activity and TXNIP expression to lower levels of glucose by increasing levels of G6P (13). Consistent with a requirement for functional electron transport, inhibition of complex I with metformin completely abrogated TXNIP induction by CHX (Figure 3A). Likewise, and consistent with a requirement for mtATP, blocking the activity of ATP synthase (complex V) with oligomycin also robustly inhibited TXNIP expression (Figure 3A). To test the requirement of ATP synthesis further, we used siRNA to deplete ATP5I, which is an essential component of ATP synthase: our previous work established that ATP5I knockdown in HeLa cells blocks the production of mtATP (13). In this experimental context, ATP5I knockdown reduced background TXNIP expression and completely suppressed its induction by RocA (Figure 3B). We next determined how protein synthesis inhibition affects mtATP. We expressed a mitochondrial-targeted ATP FRET-biosensor (mitATEAM) in HeLa cells and used live cell imaging to quantify fluorescence (13, 17). Inhibiting protein synthesis by RocA lead to increased FRET signal indicating accumulation of ATP in the mitochondria (Figure 3C and Supplemental Figure 1B). These findings suggest a requirement for mtATP synthesis in driving TXNIP expression in response to protein synthesis inhibition.
Low pH medium increases MondoA transcriptional activity by increasing mtATP levels (13). In that work, we established that mtATP exits the mitochondrial matrix via a channel comprised of the adenine nucleotide transporter (ANT) and the voltage-dependent anion channel (VDAC), where it is used a substrate for VDAC-bound hexokinase II (HKII). Mitochondria-bound HKII then transfers a phosphate to cytoplasmic glucose to generate G6P resulting in a stimulation of MondoA transcriptional activity. We tested whether RocA induces TXNIP expression through a similar mechanism in three ways. First, expression of VDAC1(E72Q), which cannot interact with HKII and prevents HKII from interacting with mitochondria (13, 32), blocked the increase in TXNIP expression following RocA treatment (Figure 3D). By contrast, wildtype VDAC increased TXNIP expression in the presence of RocA. Second, methyl-jasmonate, which removes HKII from the outer membrane of mitochondria (33), blocked RocA induction of TXNIP (Figure 3E). Third, CHX lead to a dramatic reprogramming of metabolism, including significant changes in the levels of glycolytic and TCA cycle intermediates (Figures 3F-G and Supplemental Table 1). In particular, G6P levels increased more than 20-fold following CHX treatment (Figure 3G). Together these data are consistent with the model that protein synthesis inhibitors increase mtATP, which is subsequently exported from the mitochondrial matrix through the ANT/VDAC channel, ultimately increasing G6P levels to drive MondoA transcriptional activity.
MondoA and TXNIP are required for the cytotoxic effects of RocA
Because protein synthesis inhibitors are emerging as potential cancer therapeutics (4-7), we tested whether blocking protein synthesis induced TXNIP expression in cell lines with different oncogenic lesions. CHX induced TXNIP in MEFs and in MEFs that expressed an activated allele of HRAS (Figure 4A) (34). Further, TXNIP was induced by CHX-treatment in MEFs that lack the TSC2 tumor suppressor and in MDA-MDA-231 cells, which is a Triple Negative Breast Cancer (TNBC) cell line that harbors an inactivating mutation in TP53 and activating mutations in KRAS and BRAF (Figure 4B-C). Further, induction of c-Myc(T58A), which is a stabilized allele of c-Myc, did not block TXNIP induction in MDA-MB-231 cells (Figure 4C). RocA also increased TXNIP protein levels in HeLa cells, MDA-MB-157 cells, which is also a TNBC cell line, and in MBA-MB-231 cells (Figure 4D-F). Together these data demonstrate that RocA can induce TXNIP expression in a variety of cell lines and its action appears relatively independent of oncogenic burden.
The growth inhibitory effect of RocA has been tested primarily on multiple myeloma cell lines (6, 35). Consistent with a potential broad effect of RocA on cell growth, treatment of MDA-MB-157 and MDA-MB-231 breast cancer cells with 100 nM RocA resulted in a time-dependent reduction in cell viability such that virtually all the cells were dead after 4 days of treatment (Figure 5A). We expanded this analysis to 17 organoid cultures derived from breast cancer patients treated at Huntsman Cancer Institute. As with the cell lines, these Patient-Derived xenograft Organoids (PDxOs), showed sensitivity to RocA. 10 of 12 ER- models were sensitive to RocA, with consistently strong cytotoxicity around 50 nM (Figure 5B). Most of the Estrogen Receptor positive (ER+) models, with the exception of HCI-011, were also sensitive to RocA, but sensitivity was attenuated compared to the ER- models: HCI-003 was highly sensitive to RocA like the majority of the ER- models. Thus, RocA is broadly cytotoxic to breast cancer cells and appears to show preferential killing of cells from ER- breast cancers.
We next determined whether MondoA or TXNIP were required to mediate the cytotoxic effects of RocA. TXNIP-knockout MEFs were less susceptible to RocA than wild type MEFs (Figure 5C), consistent with the notion that TXNIP is a RocA effector. Likewise, TXNIP knockdown in MDA-MB-157 cells also partially blocked the cytotoxic effects of RocA (Figure 5D). Finally, we used CRISPR-Cas9 editing to generate HeLa cells that lack MondoA and conducted a RocA dose response experiment. While MondoA-knockout had no effect on cell proliferation in the absence of RocA (Supplemental Figure 1C), we observed that MondoA loss attenuated the cytotoxic effects of RocA and increased the IC50 of RocA from ~15 nM to ~25 nM (Figure 5E). Together these data suggest induction of MondoA transcriptional activity and the subsequent induction of TXNIP are required for the full cytotoxic effects of RocA. However, the effect of MondoA and TXNIP loss on RocA cytotoxicity, while significant, is subtle suggesting that other pathways must also contribute.
The role of MondoA in the transcriptional response to RocA
To understand the contribution of MondoA to the RocA-dependent transcriptional response, we conducted mRNA-sequencing on RNA prepared from wildtype or MondoA knockout Hela cells (HeLa:MKO) that had been treated for with 100nM RocA for 4 hrs. Using a 2-fold expression change cutoff and a p-value of ≤0.01, we identified 1,241 genes that were differentially regulated by RocA. Of these, 224 genes were not differentially regulated in the absence of MondoA. This finding suggests that approximately 20% (224/1241) of the RocA-driven transcriptome requires MondoA (Figure 6A): both up and downregulated genes were MondoA-dependent. We next used regression analysis to look for genes that were affected by RocA treatment and genotype. As expected, TXNIP was highly induced by RocA and its expression was highly dependent on MondoA (Figure 6A-B). Induction of the TXNIP paralog ARRDC4 by RocA was less robust but was also highly MondoA-dependent (Figure 6A-B). Pathways downregulated following RocA treatment of MondoA knockout cells included extracellular matrix organization and a number of signaling-related pathways (Figure 6C) (22). Pathways upregulated following RocA treatment of MondoA knockout cells also included extracellular matrix organization and several pathways involved in sterol biosynthesis. Finally, we conducted Gene Set Enrichment Analysis on the differentially regulated genes in HeLa and HeLa:MondoA-KO cells treated with RocA using 13445 pathways in the Molecular Signatures Database (23, 24). We identified 1033 gene sets that were enriched with a nominal p-value of ≤0.01. Leading edge analysis showed that pathways associated cell proliferation and cell movement were upregulated, and electron transport and ribosome-related pathways were downregulated in RocA-treated HeLa:MondoA-KO cells (Figure 6D). Together these data suggest that MondoA is required for the cellular transcriptional response to RocA treatment and may contribute to migratory and growth phenotypes driven by protein synthesis inhibitors.