A comprehensive list of the chemicals, reagents, and antibodies used, along with the companies these were acquired from, the catalog numbers, and the concentrations/dilutions used are provided in Supplementary Table S1.

Telomerase-immortalized, human retinal pigment epithelium cells (RPE-1, ATCC CRL-4000) were cultured as described (20). Briefly, cells (passage ~8–24) were plated to ~70% confluency in complete DMEM (1 g/L glucose supplemented with 1X non-essential amino acids and 10% (v/v) heat-inactivated fetal calf serum (FCS)) 1 day before experimental treatments. RPE-1 cells stably expressing GFP-LC3, mtKeima and mito-roGFP have been described previously (25). Starvation media was as follows: no glucose (Corning cat#17-207-CV no glucose DMEM supplemented with pyruvate, glutamine, non-essential amino acids, and 10% (v/v) dialyzed-heat-inactivated fetal calf serum (dFCS)), no serum (Corning, 17-207-CV supplemented with 1 g/L glucose, pyruvate, glutamine and non-essential amino acids), no amino acids or Ringer’s (Alfa Aesar Kreb’s-Ringer bicarbonate-buffered solution cat# J67591 with or without 10% (v/v) dFCS). siRNA transfection was performed as previously described (24). 25 μM SFN treatment was used exclusively throughout this manuscript for the indicated incubation times.

GFP-LC3, mito-RoGFP, mtKeima, LTR, and MTG/TMRE flow cytometry were done as described (25), and FlowJo 10.3 software was used for gating analysis and quantification for all flow cytometry studies.

Mitochondrial membrane potential was determined in RPE-1 cells treated with vehicle (DMSO), SFN, or different nutrient deprivations for 4 or 24 h and co-labeled during the final 30 min of treatment with 5 nM tetramethylrhodamine ethyl ester (TMRE) and 200 nM MitoTracker Green™ (MTG). Cells were trypsinized, resuspended in complete media lacking Phenol Red, filtered through 100 μm mesh and subjected to flow cytometry. MTG was visualized with a 488 nm laser and a 510/21 nm filter. TMRE was visualized with the 561 nm laser and a 582/15 nm filter. Membrane potential changes were calculated using the ratio between the TMRE (red) and MTG (green) signals. Mitochondrial mass of RPE-1 cells treated for 24 h with either vehicle (DMSO), SFN, or different nutrient deprivations was measured by incubating cells in 200 nM MTG for the final 30 min of the 24 h treatments. Cells were processed for flow cytometry and mitochondrial mass was calculated by median fluorescence intensity (MFI) and graphed as fraction of control.

To assess mitochondrial redox stress, RPE-1 cells stably expressing mito-roGFP were treated for 4 h with SFN, FCCP, or different nutrient deprivations and then processed for flow cytometry using the 405 nm laser with 525/50 nm filter to visualize oxidized mito-GFP and the 488 nm laser with 530/30 nm filter to visualize reduced mito-GFP.

To quantify mitophagy, RPE-1 cells stably expressing mitochondria-targeted (mt)-mKeima and YFP-parkin were treated for 4 h at 37°C with vehicle, SFN, different nutrient deprivations or FCCP. Cells were then trypsinized, resuspended in 10% fetal calf serum (FCS) and 0.5 mM EDTA in PBS, passed through 100 μm mesh, and subjected to flow cytometry. YFP-Parkin positive cells were detected with 488 nm excitation and a 530/30 nm emission filter. Mt-mKeima was detected with a 488 nm excitation laser and a 695/40 nm emission filter for mitochondrial localization, and a 561 nm laser with 670/30 nm filter for lysosomal localization. YFP-Parkin-positive cells were plotted in FlowJo as acidic against neutral and two gates defined the cells exhibiting predominately lysosomal or mitochondrial mt-mKeima localization.

For autophagy assays, RPE-1 cells stably expressing GFP-LC3 were treated with SFN, the indicated pharmacological treatments, or starved of various nutrients for 24 h. Cells were then trypsinized, resuspended in 0.05% (v/v) Saponin/PBS to leak out soluble GFP-LCS not associated with autophagolysosomes, pelleted at 500 × g for 5 min, resuspended in PBS, filtered and subjected to flow cytometry.

Lysosomal mass was quantified by LysoTracker Red (LTR) uptake in RPE-1 cells treated with vehicle, SFN, Rapa, or different nutrient deprivations for 16 h. Cells were incubated with 50 nM LTR for the final 30 min of the 16 h. Cells were resuspended in PBS + 2% FCS and subjected to flow cytometry using the 561 nm laser and 582/15 nm filter.

Western blotting was performed as previously described (22, 23, 25) with the antibodies listed in Supplementary Table S1. Samples were solubilized in 2-times concentrated Laemmli buffer (100 mM Tris,pH 6.7; 2% (w/v) SDS; 25% (v/v) glycerol; 0.001% (w/v) Pyrinin Y; 0.008% (w/v) Bromophenol Blue; 285 mM b-mercaptoethanol) and heated in a boiling water bath for 5 min prior to being loaded onto gels for reducing, denaturing SDS-PAGE. All depicted Western blots are representative of at least 3 independent experiments and the migration of molecular weight markers (kDa) are indicated to left side of blots.

RNA was isolated using the Zymo Direct-zol microRNA Isolation Kit and cDNA was produced using the Quanta qScript cDNA Supermix, both per manufacturers’ instructions. qPCR was performed with PowerUp SYBR Green (Applied Biosystems, Inc.) using the primers listed in Supplementary Table S2.

Lysosomes were visualized in live cells by incubating in 50 nM LysoTracker Red for 30 min.

Unincorporated dye was removed by replacing the media and images were captured on a Nikon TE2000 inverted microscope as described (25).

RPE-1 cells were plated in 12-well dishes containing sterile glass coverslips. After the indicated treatments, cells were fixed in 4% (w/v) paraformaldehyde/PBS for 20 min, permeabilized in 0.2% (v/v) Triton X-100/PBS for 5 min, blocked in 3% (w/v) bovine serum albumin/PBS and probed overnight with anti-pPDH1 diluted 1:250. The following day, cells were washed in PBS and incubated in Alexa₄₈₈-conjugated donkey anti-rabbit IgG and Hoechst 33342, mounted and sealed on microscope slides, and imaged as described (22, 23, 25).

Following treatment with DMSO, 25 μM SFN or 5 mM 2-deoxy-d-glucose (2-DG) for the indicated times, RPE-1 cells were washed with warm media lacking both serum and glucose, incubated with Glucose Uptake Probe-Green (Dojindo,Inc. catalog# UP02) diluted 1:500 in glucose-and serum-free media for 15 min, washed with ice cold 1X WI buffer (Dojindo, Inc.), scraped into 1X WI buffer, filtered through 70 μm mesh, and subjected to flow cytometry on a BD FacsCelesta cell sorter equipped with a 488 nm laser and processed with Flowjo software.

For assessing lactate secretion per 10,000 cells, RPE-1 cells were plated in 24 well plates and incubated with 25 μM SFN or DMSO vehicle for 8 h after which media was removed for analysis (initial 8 h measurement) and replaced with fresh media containing 25 μM SFN or DMSO vehicle for an additional 16 h (for a total of 24 h treatment). Media was again removed for analyses (final 16 h measurement). Both the initial 8 h and final 16 h media collections were subjected to centrifugation at 4°C (500xg for 5 min) to remove cellular debris. Clarified supernatants were then snap frozen in liquid nitrogen. To determine cell number, remaining adherent cells were fixed and stained with 0.5% (w/v) crystal violet in 25% (v/v) methanol for 20 min, and then washed, air-dried, dissolved in methanol and the OD₅₇₀ was read and compared against a standard curve made using known numbers of cells. The lactate concentration in the previously snap frozen media was assayed using the L-Lactate Assay Kit I (Eton Biosciences, Inc.) and normalized to cell number.

RPE-1 cells (1.5×10⁶ for LC/MS and 0.5×10⁶ for GC/MS, 5 replicates/treatment/analysis) in complete DMEM were treated with vehicle or 25 μM SFN for 4 h. Cells were then washed with ice cold PBS before being snap frozen in liquid nitrogen. LC–MS analysis was performed as previously published (27) on an Agilent 6,546 LC/Q-TOF coupled to an Agilent 1,290 Infinity II LC. Chromatographic separation was performed on an Agilent InfinityLab Poroshell 120 HILIC-Z, 2.1 × 150 mm, 2.7 μm column, coupled with a UHPLC Guard, HILIC-Z, 2.1 mm × 5 mm, 2.7 μm, at 15°C, with a total run time of 29 min. GC/MS was performed as previously described (28).

RPE-1 cells were treated with vehicle or 25 μM SFN for 4 h before being lysed on ice in TPER (Thermo Scientific, Inc. catalog# 78510) supplemented with a proteinase inhibitor cocktail containing AEBSF hydrochloride, aprotinin, bestatin, E-64, leupeptin hemisulfate, and pepstatin A (EZBlock™, Biovision, Inc). Insoluble cellular debris was removed by centrifugation at 4°C (16,000 x g for 15 min) and soluble, extracted MondoA/MLX was captured by overnight incubation at 4°C with an anti-MLX antibody. Immunoprecipitates were washed, solubilized, and resolved by denaturing, reducing SDS-PAGE. MondoA was visualized with anti-MondoA antibody.

CHIP was performed as described in (29). 1×10⁷ RPE-1 cells were treated with vehicle or 25 μM SFN for 4 h, trypsinized, washed and fixed in 1% (w/v) formaldehyde in PBS at room temp for 5 min before the reaction was stopped by the addition of 125 mM glycine. Cells were washed and DNA was sheared using a Covaris E220. Chromatin DNA (chDNA) was mixed with anti-MondoA (MLXIP) antibody and Protein G magnetic beads overnight at 4°C. Bound chDNA was eluted, cleaned up with Zymo CHIP DNA Clean and Concentrator and used for CHIP qPCR with the following primers: TXNIP-CHIP7: 5’-CTC GCG TGG CTC TTC TG-3′ and TXNIP-CHIP8: 5’-GCA GGA GGC GGA AAC GT-3′. ΔCt was calculated by normalizing to input chDNA.

RPE-1 cells (50,000 cells/replicate, 5 replicates/treatment) in complete DMEM were treated with vehicle or 25 μM SFN for 4 h. RNA was isolated as described above and sequenced in the OMRF Clinical Genomics Center. The Lexogen QuantSeq sequencing data was aligned using a pipeline established by Bluebee, the Lexogen associated data analysis tool. Ensemble gene IDs were annotated using the biomart library in R for the GRCh38 (human) build. Read-count normalization and differentially expressed analyses were performed using the edgeR package from Bioconductor. Expression values quantile normalized with the voomWithDreamWeights function were analyzed for differential expression, accounting for repeated measure dependencies, using the standard functions of the limma package. Moderate t-test p-values were adjusted for multiple testing using the false discovery rate (FDR) method and FDR (q.value) <0.05 was used to filter significant differences. For a comparative look at fasting effect, raw fastq files and count data on HeLa cells under nutrient starvation protocol were downloaded from GEO repository, (accession code GSE211066), processed and analyzed alongside our sequencing data. Differentially expressed genes (DEGs) were defined by having an adjusted p-value (q- value) < 0.05 and a log2FC < −1 or log2FC > 1. Using the ggVennDiagram” package in R, DEGs from both analyses were compared via Venn diagram and the percentage of Starvation DEGs found to be in common with SFN DEGs was calculated.

Statistical differences were measured with unpaired Student’s t-test or Analysis of Variance (ANOVA) that was appropriate for each data set and analysis. Normal distribution was verified by Shapiro–Wilk test and a Mann–Whitney nonparametric test was used when normal distribution was not present. Error bars represent standard deviations, and all experimental data derived from at least 3 independent replicates.

We first compared the impacts of SFN versus different nutrient deprivations on mitochondrial mass and membrane potential. RPE-1 cells were labeled with MitoTracker Green (MTG) to mark total mitochondria and with tetramethylrhodamine ethyl ester (TMRE) to label mitochondria with intact membrane potential. Flow cytometry was used to quantify the uptake of both dyes, and FCCP treatment served as a positive control to show that the loss of mitochondrial membrane potential was readily detectable in this assay (Figure 1A, left graphs top and bottom). Membrane potential was measured at 4 h and 24 h post-treatment but mitochondrial mass was measured only at 24 h to provide sufficient time for mitochondrial biogenesis. SFN treatment caused a very modest but statistically significant loss of membrane potential at the 4 h time point but preserved mitochondrial membrane potential and increased mitochondrial mass after 24 h, most closely reflecting the mitochondrial response to AA deprivation (Figure 1A). In contrast, cells incubated in media lacking glucose showed reduced mitochondrial membrane potential, and serum deprivation increased mitochondrial mass but reduced mitochondrial membrane potential (Figure 1A).

In complementary experiments, we compared SFN treatment to nutrient deprivations to assess the extent of mitochondrial matrix oxidation, a proxy of mitochondrial redox status. These assays used RPE-1 cells stably expressing mito-roGFP, a mitochondrially-targeted GFP variant that harbors two redox-sensing cysteines (30). These cysteines modulate the excitation profile of the GFP, thereby allowing the redox state of the mitochondrial matrix to be quantified by measuring fluorescence intensity at the oxidizing and reducing excitation wavelengths (20). Four hour of SFN suppressed the fraction of mitochondria with an oxidized matrix (Figure 1B). Compared to DMSO-treated cells without starvation, both AA and glucose starvation also significantly decreased the percent of mitochondria with oxidized matrices but to a lesser extent than SFN. AA starvation looked most similar to SFN in this assay although all starvation conditions tested had statistically significant increases in oxidized mitochondrial matrices compared to SFN (Figure 1B).

This resistance to an oxidized mitochondrial matrix by SFN led us to test the hypothesis that the isothiocyanate could protect against mitochondrial oxidation induced by an exogenous oxidative challenge. Mito-roGFP RPE-1 cells were pre-treated for 4 h with SFN or DMSO followed by a 1 h exposure to 5 mM H₂O₂ to induce mitochondrial oxidation. Cells subjected to various nutrient starvations were likewise challenged with H₂O₂ (Figure 1C). In contrast to cells deprived of nutrients, versus complete media (CM), SFN prevented mitochondrial oxidation induced by H₂O₂ (Figure 1C). We next examined if SFN impacted homeostatic levels of mitochondria being targeted to lysosomes for mitophagy using RPE-1 cells stably expressing mitochondria-targeted Keima (mtKeima) and the parkin E3 ligase. mtKeima is a pH-dependent fluorescent protein containing the mitochondrial targeting sequence from COX VIII. mtKeima excites at 440 nm when mitochondria are in the cytoplasm and at 586 nm when mitochondria have been engulfed by lysosomes. The emission wavelength for mtKeima is ~620 nm and is independent of pH (31). As homeostatic mitophagy is minimal in unchallenged RPE-1 cells, stable over-expression of parkin accelerates mitophagy (31) to enable the evaluation and comparison of acute treatments (i.e., 4 h). FCCP treatment validated the sensitivity of the assay for tracking mitochondrial turnover by mitophagy in the time frame tested (Figure 1D). SFN, AA starvation, and serum deprivation all individually increased the percent of cells undergoing mitophagy whereas glucose deprivation was comparable to unstarved, DMSO-treated cells (Figure 1D). Together, these data show that SFN preserves mitochondrial membrane potential, limits mitochondrial matrix oxidation, and promotes homeostatic mitophagic flux (i.e., mitophagy and biogenesis), thereby mirroring aspects of nutrient deprivation, with the most similarities to AA starvation.

Nutrient signaling through the mechanistic target of rapamycin (mTOR) involves two integrated kinase hubs, mTORC1 and mTORC2. These complexes function as primary sensors of amino acids (mTORC1) and growth factors (mTORC2) (reviewed in (32)). In time-course experiments analyzing the acute effects of SFN on nutrient sensing, we tracked the phosphorylation of multiple proteins in the insulin signaling cascade with a focus on phosphatidylinositol 3-kinase (PI-3 K), Akt, and downstream substrates and effectors. These experiments compared SFN to the pharmacological inhibitors of insulin signaling GSK2334470 (GSK233), AZD8055 (AZD), and copanlisib (Cop). Targets of each inhibitor are diagrammed in Figure 2A. GSK233 inhibits phosphoinositide-dependent kinase-1 (PDK1), the kinase that activates Akt via phosphorylation of threonine 308 (T308), whereas AZD inhibits both mTORC1 and mTORC2, and Cop inhibits PI-3 K.

Compared to DMSO, phosphorylation of T308 on Akt was decreased by SFN, GSK233, and AZD (Figure 2B, lanes 1–4). To facilitate visualizing T308 phosphorylation, the signal was amplified by co-treating cells for the last 60 min of the 90 min incubation with GDC-0068 (GDC), an agent that blocks phosphatase access to T308 (33). This signal enhancement confirmed that SFN reduced T308 phosphorylation (Figure 2B, lane 5 versus 6) albeit to a lesser extent than the PDK1 inhibitor, GSK233 (Figure 2B, lane 6 versus 7), and that AZD altered the pattern of Akt phosphorylation (Figure 2B, lane 5 versus 8). T308 phosphorylation primes Akt to phosphorylate threonine 86 (T86) of Sin, an activating subunit of mTORC2. In turn, activated mTORC2 feeds back to phosphorylate Akt on serine 473 (S473) to generate fully activated Akt, which then phosphorylates and inhibits the downstream effectors PRAS and TSC2 (34, 35) (Figure 2A). In kinetic studies with homeostatic phosphorylation levels represented by time point ‘0’, the reduction by SFN of the priming T308 phosphorylation on Akt (Figure 2B) led to reduced phosphorylation of T86 on Sin and S473 on Akt by mTORC2 (Figure 2C, lanes 1–5). Consistent with dampening Akt activation, SFN reduced phosphorylation of the downstream substrates PRAS and TSC2 over the course of 60 min (Figure 2C, lanes 1–5). Notably, TSC2 is part of a complex that inhibits mTORC1, and is a phosphorylation substrate of AMPK (36). When compared to pharmacological inhibitors, the kinetics and specificity of phosphorylation loss mediated by SFN for each protein analyzed most closely approximated those observed when PDK1 was pharmacologically inhibited by GSK233 (Figure 2C, lanes 1–5 compared to 6–10). PI-3 K is the kinase directly upstream of PDK1 and Cop inhibits PI-3 K as well as the generation of PIP₃, which activates mTORC2 through direct binding to the PH-domain of Sin (37). Therefore, in cells treated with Cop, the loss of pSin T86 and pAKT S473 occurred within 5 min (Figure 2C, lanes a-e). Likewise, because AZD inhibits mTORC2, pSin T86 is largely preserved while loss of pAKT S473 also occurs within the first 5 min and pS6 decreased by 1 h (Figure 2C, lanes f-j). Together, these data show that SFN elicits a temporal phosphorylation pattern resembling the PDK1 inhibitor, GSK233. In contrast to the relatively rapid inhibition of mTORC2 signaling by SFN, phosphorylation of S6 (pS6) downstream of mTORC1 was unchanged during the first 60 min of treatment (Figure 2C, lanes 1 and 5). However, 8 h of SFN markedly reduced pS6 levels with a concomitant increase in the mTORC1 inhibitor, Sestrin 2. (Figure 2D, lane 4). The elevation of Sestrin 2 by SFN is mediated in part by stability/turnover. Sestrin 2 is stabilized by SFN (Figure 2E, lane 1 versus 3) and by MG132 in the absence of SFN (Figure 2E, lane 1 versus 2) but inhibition of the proteasome in the presence of SFN did not further increase Sestrin 2 levels (Figure 2E, lane 3 versus 4), indicating that the phytochemical limits the proteasomal degradation of Sestrin 2. Elevated Sestrin 2 is also mediated at the transcriptional level as SFN increased Sestrin 2 mRNA by ~10-fold (Figure 2F).

We next tested the hypothesis that SFN would activate autophagy since suppression of mTORC1/2 relieves autophagy inhibition to promote the salvaging of cellular bioenergetic substrates and building blocks (38). In addition, SFN has been reported to both activate [e.g., (39, 40)] and inhibit [e.g., (41)] autophagy based on Western blotting for microtubule-associated protein 1A/1B-light chain 3 (LC3) to measure the lipidated and unlipidated forms (LC3-I and LC3-II, respectively). Increased LC3-II levels are typically interpreted as a proxy of elevated autophagic flux. However, because LC3-II accumulation can occur through either an increase in autophagic flux or conversely, a block in autophagy progression downstream of autophagosome formation, we used RPE-1 cells stably expressing GFP-tagged LC3 (GFP-LC3) (25) in combination with bafilomycin A1 (Baf) and flow cytometry to determine the consequences of SFN on autophagic flux. Baf is a macrolide antibiotic that blocks the autophagy pathway by preventing fusion of autophagosomes to lysosomes and by inhibiting the vacuolar H⁺ ATPase to prevent the acidification of lysosomes and protein degradation (42). Eight hours of SFN treatment led to increased GFP-LC3 puncta as well as increased LysoTracker Red™ (LTR)-positive puncta, a marker of lysosomes (Figure 3A). To further analyze autophagic flux, RPE-1 cells stably expressing GFP-LC3 were treated with SFN, the indicated pharmacological treatments, or deprived of different nutrients for 24 h. Parallel wells of cells were co-treated with Baf during the final 6 h of the 24 h incubation. Prior to quantifying GFP-LC3 levels by flow cytometry, cells were permeabilized with saponin to leak out soluble GFP-LC3 not incorporated into autophagic vesicles. As Baf inhibits the degradation of autophagic cargoes, the amount of GFP-LC3 in cells incubated with Baf represents the “total” amount of autophagic traffic that entered the system during the final 6 h of the experiment. GFP-LC3 levels in cells not treated with Baf represent “residual” traffic actively transiting through the autophagy system, and the difference (i.e., total – residual) represents the amount of “cleared” autophagic traffic. Compared to established inducers of autophagy including rapamycin (Rapa), inhibitors of insulin signaling (GSK233, AZD, Cop), and nutrient deprivations, SFN treatment yielded greater amounts of total and residual GFP-LC3 autophagic vesicles (Figure 3B). With the exception of serum deprivation (− serum), SFN induced greater clearance of autophagic vesicles compared to the other treatments (Figure 3B).

The amplification of autophagic flux by SFN was further corroborated by testing whether combining the phytochemical with pharmacological inducers of autophagy (Figure 3C) or with nutrient deprivations that induce autophagy (Figure 3D) would further augment vesicular GFP-LC3 beyond the levels reached by each individual treatment. RPE-1 cells stably expressing GFP-LC3 were incubated for 24 h with inducers of autophagy and compared to the same treatments combined with SFN. Remarkably, co-treatment with SFN and pharmacological inducers of autophagy elevated GFP-LC3 levels with the exception of Cop, which approached significance (p = 0.092; Figure 3C). Moreover, depriving cells of serum and/or AA induced autophagy and SFN also amplified these inductions (Figure 3D).

We next tested if the increased autophagic flux induced by SFN was accompanied by increased cellular lysosomal content, as indicated by fluorescence microscopy of LTR-labeled cells (Figure 3A). We quantified lysosomal content by LTR uptake using flow cytometry, comparing SFN to nutrient deprivations and to Rapa (Figure 3E). Cells were treated with SFN, Rapa, or starved of specific nutrients overnight and then stained with LTR. Notably, neither Rapa nor AA starvation increased lysosomal content (Figure 3E gray and green bars, respectively) indicating that lysosomal biogenesis was responsive to the inhibition of insulin signaling but not inhibition of mTORC1 by Rapa or AA deprivation in this assay. SFN robustly increased lysosomal content, consistent with the phytochemical inducing the transcription factor EB (TFEB)-directed CLEAR transcriptional program downstream of mTOR, as recently reported (43). Using qPCR, we confirmed that 24 h of treatment with SFN induced the expression of canonical CLEAR target genes (ATP6V, LAMP1, TFEB, CTSA, and MCOLN1) by 1.5-to 4.5-fold over vehicle treatment (Figure 3F).

We next interrogated the time-dependent impacts of SFN on glucose uptake and metabolism based on work showing SFN, through its induction of Nrf2 transcriptional activity, induces genes encoding the rate limiting enzymes of pathways that shunt glucose out of glycolysis (44). We quantified glucose uptake using flow cytometry to measure the intracellular accumulation of a fluorescent glucose analog added to the media. Glucose uptake was reduced by SFN at 1 h, 4 h, and 8 h post-treatment similarly to 2-deoxy-glucose (2-DG), an established inhibitor of glucose uptake (Figure 4A). This reduced glucose uptake correlated with decreased extracellular lactate levels during the initial 8 h of SFN exposure (Figure 4B). However, glucose uptake recovered within 24 h of SFN treatment, with levels comparable to DMSO-treated cells (Figure 4A), and lactate secretion rebounded accordingly (Figure 4B, final 16 h). Complementary metabolomic analysis of glycolytic and tricarboxylic acid (TCA) cycle intermediates showed that 4 h of SFN led to the accumulation of intracellular glucose and the two downstream glycolytic intermediates, glucose-6-phosphate (G6P) and fructose-6-phosphate (F6P) (Figure 4C). Other assayed glycolytic intermediates were unchanged. TCA metabolite analysis revealed increased citrate and aconitate whereas a-ketoglutarate, succinate, fumarate and malate were all decreased (Figure 4D).

Concomitant with these effects on glycolysis and the TCA cycle, SFN reduced phosphorylation of PDH (Figure 4E), the mitochondrial enzyme that converts pyruvate to acetyl-CoA for the TCA cycle. Phosphorylation of PDH (pPDH) inactivates the enzyme and dephosphorylation converts PDH to an active form (45, 46). Immunofluorescence assays with an antibody recognizing Ser293 phosphorylation of PDH (pPDH), a proxy of PDH activity (47, 48), revealed that 4 h of SFN treatment robustly reduced the percentage of cells with mitochondria-associated pPDH (Figure 4E). AA deprivation or serum starvation both increased pPDH whereas glucose starvation reduced pPDH (Figure 4E). Thus, SFN most closely mimicked glucose starvation in this assay.

The accumulation of G6P, the first oxidation product of glycolysis, can elevate the expression of thioredoxin (TXN)-interacting protein (TXNIP), a central regulator of glucose entry into cells (49–53). By modulating glucose transporter (GLUT) endocytosis, increased levels of TXNIP reduce cellular glucose uptake whereas decreased TXNIP abundance promotes glucose uptake into cells [e.g., (49, 52)]. Based on SFN treatment acutely reducing glucose uptake but then restoring it by 24 h (Figure 4A), we tested the hypothesis that this restoration was mediated by reducing TXNIP despite increased G6P levels. After confirming that TXNIP mRNA and protein expression are glucose-responsive and dependent in RPE-1 cells (Supplementary Figure S1A), we examined TXNIP levels by Western blotting lysates from RPE-1 cells treated with the phytochemical for different lengths of time. Compared to vehicle-treated cells, 8 h and 24 h of SFN treatment dramatically reduced TXNIP protein expression (Figure 5A, lanes 3 versus 4 and 5 versus 6). Notably, the elevation of TXNIP at 8 and 24 h in the DMSO samples compared to the 1 h time point is consistent with glucose being taken up by the RPE-1 cells initially after replating into fresh media followed by an increase in TXNIP to restrict further uptake at the later time points (Figure 5A, lanes 3 and 5). Co-treatment with the proteasome inhibitor, MG132, stabilized TXNIP indicating that proteasome-mediated degradation of this protein was not adversely impacted by SFN (Figure 5B, lanes 3 versus 4) (49, 51). In contrast, co-administration of SFN with the global transcription inhibitor, actinomycin D (Act D), normalized TXNIP expression between DMSO and SFN in the presence and absence of MG132 (Figure 5B, lanes 5 versus 7 and lanes 6 versus 8), indicating that the SFN-mediated decrease in TXNIP occurs at the transcriptional level. qPCR analysis confirmed that 4 h of SFN robustly decreased TXNIP mRNA expression as well as the mRNA levels of the TXNIP paralogue, ARRDC4 (Figure 5C).

After ruling out that SFN was not accelerating the degradation of TXNIP mRNA (Supplementary Figure S1B), we examined the glucose-dependent transcription complex MondoA/Mlx. This heterodimeric transcription factor mediates TXNIP and ARRDC4 mRNA expression by binding a carbohydrate response element within their respective promoters (52). 1, 8, or 24 h of SFN treatment did not statistically significantly decrease MondoA or Mlx protein levels (Figure 5A blots and graph). Endogenous protein co-immunoprecipitations showed that SFN also does not disrupt complex formation between MondoA and Mlx (Supplementary Figure S1C). Chromatin immunoprecipitation (CHIP) assays however revealed that SFN reduced MondoA residency on the TXNIP promoter by ~30% (Figure 5D).

As Nrf2 can bind to the TXNIP promoter (54), we next tested the hypothesis that SFN-mediated suppression of TXNIP mRNA and protein levels is Nrf2 dependent. After confirming the efficiency of Nrf2 mRNA knockdown by siRNA using qPCR (Supplementary Figure S1D), we found that TXNIP mRNA and protein suppression by SFN were comparable between control (siCON) and Nrf2 (siNrf2) knockdown cells (Figures 5E, F, respectively). Additionally, stabilization of Nrf2 by other activators had only a very modest or no effect on TXNIP mRNA levels despite robustly inducing the Nrf2 target gene, heme-oxygenase 1 (HO-1) (Figure 5G). These data show that the suppression of TXNIP mRNA and protein expression by SFN is independent of Nrf2 induction.

Ayer and colleagues demonstrated that 2-DG induces the Mondo/Mlx transcriptional activation of TXNIP (52). This glucose derivative is phosphorylated by hexokinase to produce 2-DG-6-phosphate, a product that cannot be further processed for glucose oxidation and is a relatively poor substrate for glucose-6-phosphate dehydrogenase (G6PDH) (55). We therefore tested the hypothesis that SFN could suppress 2-DG-mediated TXNIP induction. Four hours of 2-DG followed by 4 h of DMSO elevated TXNIP mRNA expression ~8-fold but 4 h of SFN treatment suppressed this TXNIP induction (Figure 5H). Likewise, TXNIP transcription is also increased by treatment with Trichostatin A (TSA) (Figure 5I), a Class I/II histone deacetylase inhibitor (56). Pretreatment for 4 h with SFN prevented the TSA-induced transcriptional increase of TXNIP (Figure 5I). These data demonstrate that SFN suppresses TXNIP mRNA induction and does so independently of Nrf2 activation.

The hexosamine biosynthetic pathway (HBP) diverts F6P from glycolysis into an anabolic pathway for de novo generation of uridine diphosphate-N-acetyl glucosamine (UDP-GlcNAc), the substrate transferred to recipient proteins for N-and O-linked glycosylation reactions (reviewed in (57)). Notably, HBP activation can increase TXNIP mRNA expression (58). Because SFN treatment increased F6P levels (Figure 4C), we tested the hypothesis that the phytochemical inhibits glutamine F6P amidotransferase (GFAT), the first and rate-limiting enzyme of the HBP that converts F6P to glucosamine-6-phosphate. Consistent with this hypothesis, the suppression of TXNIP mRNA expression by SFN was reversed by supplying cells with exogenous glucosamine and this rescue was dose-dependent (Figure 6A). Glucosamine is phosphorylated by hexokinase to produce glucosamine-6-phosphate, the product produced by GFAT from F6P and glutamine. Thus, providing glucosamine to cells bypasses GFAT inhibition and increases TXNIP mRNA expression. This interpretation was independently confirmed using two GFAT pharmacological inhibitors, 6-diazo-5-oxo-L-norleucine (Don) and azaserine (Aza). Inhibiting GFAT with Don or Aza suppressed TXNIP mRNA expression ~ 80% (Figure 6B), and exogenous glucosamine likewise overcame this GFAT block leading to elevated TXNIP (Figure 6C). Consistent with the HBP being integrated with the unfolded protein response (UPR) and ER stress (59), 4 h of SFN upregulated the mRNA expression of the UPR-and ER stress responsive transcription factors XBP1 and ATF4 (Figure 6D). In contrast, inhibition of GFAT with either Don or Aza failed to increase XBP1 mRNA expression and only modestly increased ATF4 mRNA expression. These data demonstrate that the suppression of TXNIP mRNA expression by SFN likely results from inhibiting GFAT.

We further interrogated the transcriptional starvation responses elicited using RNA-seq analysis of RPE-1 cells treated with SFN for 4 h (raw data file accessible via GEO number GSE275316). Ingenuity Pathway Analysis (IPA) showed that the top canonical pathways impacted were general transcription machinery, iron uptake and transport, signaling by insulin receptor, unfolded protein response, and chaperone-mediated autophagy signaling (Figure 7A). Gene ontology (GO) enrichment analysis of canonical starvation pathways showed that SFN reduced mRNA expression of many components of general transcription, but selectively activated a number of transcriptionally-modulated stress and starvation-related response pathways including the UPR, ER stress, autophagy, and amino acid regulation of mTORC1 (Figure 7B). Highly induced genes included chaperones (i.e., heat shock proteins A1A, A1B, and A6), numerous ATG genes of the autophagy pathway as well as XBP1 (see Supplementary Table S3 – list of differentially expressed genes (DEGs) ranked by p-value). In addition to playing a central role for the UPR, XBP1 is a nexus for directing responses to starvation (60).

Four hours of SFN treatment decreased mitochondrial biogenesis pathway as identified by IPA (Figure 7B), although there was robust induction of GA binding protein transcription factor (GABPB1) (Figure 7C), an inducer of gene targets that promote mitochondrial biogenesis (61). The CLEAR signaling pathway was also suppressed at 4 h, based on decreased mRNA expression of the CLEAR inducing transcription factors, TFEB and MITF. Yet, 24 h of SFN increased TFEB mRNA expression along with a set of CLEAR genes (Figure 3E). Notably, Cesana et al. observed that it took 6 h for TFEB to be transcriptionally upregulated in starved HeLa cells and that the early growth response 1 (EGR1) protein promotes TFEB transcriptional activity (62). EGR1 mRNA was robustly increased (> 16-fold) by 4 h of SFN (Figure 7C). Additional changes consistent with a nutrient deprivation response were upregulation of several inhibitors of mTOR including DNA damage-inducible transcript 4 protein (DDIT4 (aka REDD1); >16-fold increase) and sestrin 2 (> 8-fold increase) (Figure 7C), confirming the reduced mTOR signaling and elevated sestrin 2 detected by Western blotting and qPCR (Figure 2). Both of these mTORC1 inhibitors are downstream targets of the ATF4 transcription factor (63), which was likewise elevated (Figures 6D, 7C). These RNA-seq data show that acute SFN treatment induces numerous transcriptional responses elicited by fasting.