Akt regulation of glycolysis mediates bioenergetic stability in epithelial cells

Cells use multiple feedback controls to regulate metabolism in response to nutrient and signaling inputs. However, feedback creates the potential for unstable network responses. We examined how concentrations of key metabolites and signaling pathways interact to maintain homeostasis in proliferating human cells, using fluorescent reporters for AMPK activity, Akt activity, and cytosolic NADH/NAD+ redox. Across various conditions, including glycolytic or mitochondrial inhibition or cell proliferation, we observed distinct patterns of AMPK activity, including both stable adaptation and highly dynamic behaviors such as periodic oscillations and irregular fluctuations that indicate a failure to reach a steady state. Fluctuations in AMPK activity, Akt activity, and cytosolic NADH/NAD+ redox state were temporally linked in individual cells adapting to metabolic perturbations. By monitoring single-cell dynamics in each of these contexts, we identified PI3K/Akt regulation of glycolysis as a multifaceted modulator of single-cell metabolic dynamics that is required to maintain metabolic stability in proliferating cells.


Introduction 31
A central function of cellular metabolic regulation is to ensure an adequate supply of metabolites 32 for bioenergetics and biosynthetic processes. To maintain metabolic homeostasis, cells utilize feedback 33 loops at multiple levels in an integrated metabolic-signaling network. For instance, glycolysis is 34 predominantly regulated by feedback control at the level of phosphofructokinase, which senses the 35 availability of ATP and the respiratory intermediate citrate. Additionally, in response to ATP depletion, 36 the energy-sensing kinase AMPK stimulates glucose uptake and suppresses energy-consuming 37 processes (Hardie, 2008). The goal of these homeostatic pathways is to respond to bioenergetic stress 38 by increasing or decreasing the appropriate metabolic fluxes to return the cell to a state with stable and 39 sufficient levels of key metabolites. While bioenergetic stress can occur when any of a number of 40 metabolites becomes critically limited, we focus in this study on the key metabolite ATP because of its 41 broad importance as an energy source for cellular processes, and because AMPK activity can be used 42 as a reliable indicator of low ATP:AMP ratios within the cell. We therefore use the term bioenergetic 43 stress here to indicate a situation in which the concentration of available ATP is reduced, as indicated 44 by AMPK activation. 45 Bioenergetic stress can result from a loss of ATP production, such as when nutrients become 46 limited or metabolic pathways are inhibited by a pharmacological agent. Alternatively, ATP depletion 47 can also result from an increase in ATP usage, such as when anabolic processes are engaged during 48 cell growth. Because anabolic processes such as protein translation can use a large fraction (20-30%) 49 of cellular ATP (Buttgereit and Brand, 1995; Rolfe and Brown, 1997), it is unsurprising that cellular 50 proliferation and metabolic regulation are tightly linked (Gatenby and Gillies, 2004; Wang et al., 1976). 51 Growth factor (GF) stimulation activates the PI3K/Akt pathway, which plays a key role in proliferation by 52 stimulating both cell cycle progression and mTOR activity, leading to increased protein translation. 53 Simultaneously, Akt activity promotes glucose metabolism by stimulating the activity of hexokinase 54 (Roberts et al., 2013) and phosphofructokinase (Novellasdemunt et al., 2013)  1994; Tornheim and Lowenstein, 1973;Yang et al., 2008), confirming the potential for instability during 71 metabolic adaptation. However, in epithelial cells, little is known about the kinetic relationships between 72 signaling and metabolic activity that allow proliferation and other anabolic processes to proceed in step 73 with energy production. 74 To understand the kinetics of homeostasis, single-cell data are needed because of the potential 75 for metabolic state to vary even among genetically identical cells. Events that are asynchronous among 76 cells, and subpopulations with differential behaviors, are not apparent in the population mean due to 77 their tendency to "average out" (Purvis and Lahav, 2013). Until recently, dynamics in metabolism could To assess the dynamics of the cytosolic NADH-NAD + redox state, we utilized the fluorescent 118 biosensor Peredox, which is based on a circularly-permuted green fluorescent protein T-Sapphire 119 conjugated to the bacterial NADH-binding protein Rex (Hung et al., 2011). To maintain compatibility 120 with red-wavelength reporters for dual imaging and to simplify cell tracking, we generated a nuclear-121 targeted Peredox fused to the YFP mCitrine (Fig. 1E). NADH is a major redox cofactor in glycolysis, 122 which generates NADH from NAD + via the glyceraldehyde-3-phosphate dehydrogenase (GAPDH) 123 reaction in the cytosol. As NADH and NAD + exchange freely between nuclear and cytosolic 124 compartments, Peredox nuclear signal reports the cytosolic NADH-NAD + redox state and serves as an 125 indicator of glycolytic activity (Hung et al., 2011). Once normalized by the fused mCitrine signal to 126 correct for variations in biosensor expression, Peredox nuclear signal is thus defined as the "NADH 127 index." To verify cytosolic NADH-NAD + redox sensing, we exploited the lactate dehydrogenase reaction 128 to interconvert between pyruvate and lactate with concomitant exchange between NADH and NAD + . 1H). A control reporter with a mutation in the NADH binding site (Y98D) predicted to abrogate NADH 135 binding failed to respond to the same conditions (Fig. S1E). 136 To track PI3K/Akt pathway activity, we constructed a reporter based on the Forkhead 137 transcription factor FOXO3a. Akt phosphorylation of FOXO3a promotes its cytoplasmic retention; with low Akt activity, dephosphorylated FOXO3a translocates to the nucleus (Brunet et al., 1999;Tran et al., 139 2002). To monitor Akt activity, we thus fused a red fluorescent protein mCherry to a truncated FOXO3a 140 gene in which transcriptional activity was abrogated to minimize any interference on endogenous gene 141 transcription, a strategy previously shown to specifically report Akt activity (Gross and Rotwein, 2016; 142 Maryu et al., 2016). We refer to this construct as AKT-KTR; the Akt activity indicated by its cytosolic-to-143 nuclear fluorescence ratio is referred to as the 'Akt index' (Fig. 1I). Upon insulin application following 144 GF deprivation, MCF10A-AKT-KTR cells showed an abrupt increase in Akt index (Fig. 1K insulin treatment induced the strongest activation of Akt index, while EGF produced a more moderate 155 activation ( Fig. 2A). Glucose uptake and NADH index were also highest in insulin-treated cells, 156 intermediate in EGF-treated cells, and lowest in the absence of GFs (Figs. 2B, 2C), while the average 157 AMPK index correlated inversely with Akt index (Fig. 2D). Because EGF stimulated proliferation more 158 strongly than insulin (Fig. S2A), these metabolic parameters correlated poorly with proliferative rate. 159 Together, these results suggest that the increased rates of glucose uptake and metabolism stimulated 160 by Akt activity are the primary determinants of metabolic status under each GF. Accordingly, treatment 161 of insulin-cultured cells with an Akt or PI3K inhibitor decreased glucose uptake ( GF conditions, we lacked a framework to quantify and interpret these dynamics; we therefore turned to 166 defined metabolic perturbations as a tool to first establish basic homeostatic responses for single cells. To assess the range of individual cellular responses to specific bioenergetic challenges, we 170 exposed MCF10A-AMPKAR2 cells to a panel of metabolic inhibitors, including oligomycin (an inhibitor 171 of the mitochondrial F0/F1 ATPase), carbonyl cyanide m-chlorophenyl hydrazonesodium (CCCP, a 172 mitochondrial proton gradient uncoupler), and iodoacetate (IA; an alkylating agent that inhibits the 173 glycolytic enzyme GAPDH with minimal effects on other cellular thiols at <100 µM (Schmidt and Dringen, 2009)). As expected, each of these compounds rapidly raised the mean AMPK index in a 175 dose-dependent manner in cells cultured in growth medium, confirming that both glycolysis and 176 oxidative phosphorylation contribute to ATP production in proliferating MCF10A cells (Fig. 3, 177 Supplemental Movies 1-3). However, each inhibitor induced strikingly different kinetics at the single-cell 178 level. Notably, IA induced periodic oscillations of AMPK index, most evident at intermediate (5-10 µM) 179 IA concentrations in which oscillations were sustained for as many as 50 cycles over 20 hours (Fig.  180 3A). These fluctuations of AMPK activity were not synchronized among individual cells and thus not 181 apparent in the population average measurements. The asynchronous nature of these fluctuations 182 argued against imaging artifacts or environmental fluctuations, which would affect all cells 183 simultaneously. In contrast, oligomycin induced an immediate increase in AMPK index that peaked at 184 ~40 minutes but then fell, followed by a series of irregular pulses of AMPK activity ranging in duration 185 from 1-3 hours (Fig. 3B). 186 Because pulsatile AMPK activities were a common feature of the single-cell response to 187 multiple perturbations, we developed a "fluctuation score" to quantify the cumulative intensity of 188 fluctuations for each cell over time ( Fig. S3A and Methods section). Oligomycin and IA-treated cells 189 showed significantly increased fluctuation scores relative to untreated cells (Fig. 3D)

Temporally coordinated oscillatory dynamics in bioenergetics and signaling upon inhibition of glycolysis 211
To understand why cells fail to reach stable adaptation under some conditions, we focused first 212 on the rapid oscillations triggered by IA treatment (Fig. 4A). The average period of these oscillations 213 ranged from 18 minutes at 20-40 µM to 30 minutes at 5 µM (Fig. 4B). For IA at 10 µM or greater, the 214 percentage of cells displaying oscillations (defined as 5 or more successive pulses with a spacing of 1 215 hr or less) was >95%; this percentage fell to <60% at 5 µM IA, and no oscillation was detected at an increase in AMPK index, and peaks of Akt and AMPK index remained shifted by 0.5 cycles 230 thereafter (Fig. 4G). Based on these relative phase shifts, we constructed a composite diagram of the 231 relationship between the three parameters (Fig. 4H). Thus, single-cell oscillations in PI3K/Akt activity, 232 AMPK activity, and glycolytic NADH production were temporally coordinated, suggesting that feedback 233 regulation tightly links these processes on the scale of minutes and leads to a persistent cycling of each 234

pathway. 235
We hypothesized that IA-induced oscillations in AMPK activity and NADH/NAD + ratio resulted 236 from oscillations in glycolytic flux, triggered by feedback-driven increases in the entry of glucose into 237 glycolysis upon GAPDH inhibition and flux reduction by IA. We therefore compared the fluctuation 238 scores for IA-treated cells in the presence of varying extracellular concentrations of glucose and 239 pyruvate. In the absence of glucose, IA treatment failed to induce oscillations in AMPK or NADH index 240 (Figs. S4B, S4C). The incidence of oscillations, and the corresponding fluctuation score, increased with 241 the extracellular glucose concentration in a dose-dependent manner, reaching a maximum at 4-5 mM, 242 while the average period and amplitude remained essentially constant. In contrast, pyruvate alone, 243 although capable of serving as an ATP source for MCF10A cells (Fig. S1B) was unable to sustain IA-244 induced oscillations (Fig. S4D). Pyruvate also had no effect on IA-induced AMPK index oscillations in 245 the presence of glucose (Fig. S4D), although it rendered NADH index oscillations undetectable by lowering the resting NADH/NAD + ratio (Fig. S4E). Together, these observations indicate that pyruvate 247 does not fuel ATP production at a high enough rate to impact the rapid oscillatory changes during IA 248 treatment. The data support the conclusion that these rapid oscillatory dynamics originate from 249 changes in flux in glycolysis, with downstream metabolic processes playing little role. 250 Oscillations often arise in feedback systems in which there is a delay between induction of 251 feedback and recovery of the feedback-controlled variables (Glass et al., 1988). The co-oscillation of 252 Akt activity along with AMPK and NADH index suggests the involvement of a complex feedback 253 structure involving PI3K/Akt, AMPK, and also mTOR ( Consistent with this idea, multiple inhibitors of mTOR and PI3K activity suppressed IA-induced NADH 255 oscillations (Fig. 4I). Suppression was most potent with BEZ-235, which inhibits PI3K, mTORC1, and 256 mTORC2 activity, and strong but somewhat less potent with Torin1, which inhibits both mTORC1 and 257 mTORC2. Both rapamycin, which inhibits mTORC1 alone, and BKM-120, which inhibits only PI3K, had with no subsequent adaptation, followed by cell death in 100% of cells within 12 hours. In 17.5 mM 276 glucose (the baseline concentration for MCF10A media), oligomycin induced a rapid initial pulse of 277 AMPK activity and subsequent adaptation, with >75% of cells returning to baseline AMPK index within 278 2 hours. Following this adaptation, cells displayed regular pulsatile dynamics in AMPK index, with an 279 average period of ~2.5 hours; the first two pulses of AMPK index were highly synchronous among cells, 280 followed by gradual de-synchronization. As with the initial pulse, each burst of AMPK activity lasted 2-4 281 hours, suggesting that continuing oligomycin treatment induced ongoing bioenergetic challenges, which were nevertheless overcome by cells maintained at 17.5 mM glucose. At glucose concentrations of 3.4 283 mM and 1.7 mM, cells were unable to achieve full adaptation, with <25% and <10% of cells returning to 284 baseline within 2 hours, and subsequent pulses in AMPK index were relatively dampened and 285 prolonged. Thus, under high glucose levels, recovery of ATP levels occurs quickly and completely, but 286 gives rise to recurring pulses of AMPK activity, suggesting that changes in the rate of ATP production 287 by glucose metabolism via glycolysis generate these pulses. 288 We next examined how GF regulation of glucose metabolism impacts these dynamics (Fig. 5B). 289 In the presence of glutamine, treatment with insulin strongly suppressed oligomycin-induced pulses in 290 AMPKAR index, relative to non-GF treated cells. While EGF moderately increased the strength of 291 pulses, co-treatment with both EGF and insulin led to suppression of pulses (Figs. 5B, 5C). This 292 suppression was negated by co-treatment with Akt inhibitor, which strongly enhanced oligomycin-293 induced AMPKAR pulses (Fig. S5C). Similarly, in the presence of insulin, moderate inhibition of 294 glycolysis with a dose of IA too low to independently stimulate oscillations resulted in amplification of 295 oligomycin-induced AMPK pulses (Fig. S5D). Thus, the enhancement of glucose metabolism by insulin-296 mediated signaling is capable of attenuating recurrent ATP shortages following adaptation to 297

oligomycin. 298
The pronounced and relatively regular nature of oligomycin-induced AMPK index oscillations in 299 the presence of glucose alone (Fig. 5A), as compared to the more irregular pulses seen in complete 300 growth medium (Fig. 3A), suggested that glutamine or pyruvate, which are present in complete 301 medium, may also influence oligomycin-stimulated oscillations. While pyruvate had no effect on the 302 kinetics of oligomycin response, the amplitude and regularity of pulsing were strongly attenuated when 303 glutamine was present (Figs. S5E, S5F). When we performed the same experiment in the presence of 304 different GF stimuli, we found that glutamine was required for the suppression of oscillations by insulin 305 (Figs. 5D, 5E). However, regardless of the presence of insulin, cells cultured in the presence of 306 glutamine without glucose failed to recover their AMPK index and died within 12 hours (Fig. S5F), 307 suggesting that glutamine may play a role in supporting glucose metabolism, but cannot alone provide 308 sufficient ATP in the absence of oxidative phosphorylation. 309 Altogether, these data suggest a model whereby, in the presence of high glucose, feedback 310 regulation of glucose metabolism via glycolysis upon oligomycin treatment leads to recovery of ATP 311 levels ( Fig. S5G). However, this sharp increase in rate triggers negative feedback regulation of 312 glycolysis, causing ATP levels to fall again (Fig. S5G, top). Insulin and glutamine counter these 313 negative feedbacks, allowing glycolysis to continue at a high rate and thereby maintain high ATP levels 314 (Fig. S5G, bottom). In contrast, in the presence of low glucose, oligomycin cannot stimulate a large 315 enough increase in glycolysis to rapidly restore ATP levels, and negative feedback is not triggered, 316 resulting in ATP levels that remain at lower, but stable, levels ( Fig. S5G, middle). 317

Akt-stimulated glucose uptake is required for bioenergetic stability in proliferation 319
Finally, we returned to the question of how GF stimulation and proliferation impact bioenergetic 320 stability even in the absence of overt metabolic perturbations. We quantified the AMPK index fluctuation 321 scores for GF-stimulated MCF-10A cells; we found that while AMPK index fluctuations under these 322 conditions were less pronounced than in IA-or oligomycin-stressed cells, significantly more fluctuations 323 occurred in non-GF-or EGF-treated cells relative to cells treated with insulin or a combination of insulin 324 and EGF (Figs. 6A, 6B). To understand the basis for these fluctuations, we first used the Geminin cell 325 cycle reporter to examine cell cycle-dependent differences in AMPK index. In EGF-stimulated cells in 326 which AMPK fluctuations were most prominent, we compared the AMPK fluctuation scores between 327 G0/G1 and S/G2/M phases of the cell cycle ( Fig. 6C) and found that AMPK activity was significantly 328 more pulsatile in G0/G1 cells. However, this moderate difference between cell cycle phases does not 329 explain the overall effect of GFs on AMPK kinetics: compared to EGF-treated cells, insulin-treated cells 330 are more likely to be in G0/G1 but have a lower probability of AMPK index fluctuations. We therefore 331 next investigated the involvement of Akt by simultaneously monitoring both Akt and AMPK in dual 332 reporter MCF10A-AMPKAR2/AKT-KTR cells. Analysis of fluctuations in both reporters on a cell-by-cell 333 basis revealed a high frequency of inverse events, with a pulse in AMPK index mirrored by a decrease 334 in Akt index (Fig. 6D). Cross-correlation analysis indicated that such inverse events were highly 335 overrepresented in the population relative to their expected occurrences at random (Fig. 6E), 336 suggesting that AMPK fluctuations may result at least in part from rises and falls in Akt activity and the 337 associated rate of glucose uptake. and NADH availability can influence functions such as DNA synthesis and gene expression, 357 understanding these metabolic dynamics, rather than simply average or baseline concentrations, will be 358 crucial in developing an integrated model for the control of cellular metabolism and growth. 359 Feedbacks in metabolic control enable the cell to maintain adequate levels of key metabolites 360 under non-ideal circumstances. Because ATP plays a central role in providing energy for many 361 essential cellular processes, even short lapses in availability can potentially compromise cellular 362 function and viability; it is likely that evolution has selected for feedback kinetics that rapidly reverse any 363 decrease in ATP to prevent levels from falling dangerously low. Consistent with this idea, we find that 364 cells provided with different fuel sources (glucose, glutamine, and pyruvate) are able to adapt and 365 maintain steady levels of AMPK activity with few fluctuations, albeit at different set points that depend 366 on the fuel source (Figs. S1B, S1C). However, optimization for such rapid and efficient adaptation 367 comes with the potential that for certain conditions stable adaptation cannot be achieved, and unstable 368 Our results point to a central role for glycolysis in mediating metabolic stability. Despite its 387 relative inefficiency in ATP yield per molecule of glucose, glycolysis can, at least under certain 388 conditions, produce ATP at a faster rate than oxidative phosphorylation if sufficient glucose is available; 389 this ability is best documented in muscle cells during anaerobic activity but could conceivably extend to other situations such as hypoxic cells within a tumor (Liberti and Locasale, 2016). Our results also 391 implicate insulin and the PI3K/Akt pathways as controlling factors in bioenergetic stability, consistent 392 with their stimulatory effect on glucose uptake and glycolytic flux. The data presented here suggest that 393 the capacity for rapid ATP production by glycolysis can play both positive and negative roles in 394 bioenergetic stability. For example, insulin enhances the occurrence of IA-induced oscillations (Fig. 4I), 395 but has a suppressive role for oligomycin-induced oscillations (Fig. 5C). In our oscillation models (Figs. 396 S4G, S4H, and S5G), this difference is consistent with the configuration of the network in each case. In 397 the case of IA, where a bottleneck is imposed between upper and lower glycolysis, higher glucose input 398 stimulated by insulin would increase both the maximum rate for ATP production and the strength of We speculate that in the epithelial cells examined here, a core glycolytic oscillator becomes entrained 414 through feedback connections to these additional regulatory pathways that are central to growth and 415 homeostasis in this cell type. Our data also suggest that metabolic oscillations may be a wider 416 phenomenon than previously thought, as we demonstrate their occurrence in cells not typically 417 considered highly metabolically active, and also find that they may occur with heterogeneous phasing 418 that makes oscillations impossible to detect without single cell methods. The tools developed here will 419 be of use in detecting and analyzing similar oscillations in other cell types and conditions. 420 421 Implications of energetic stability in GF signaling, carcinogenesis, and pharmacotherapy 422 Given that over 90% of human tumors arise in epithelial tissue and that abnormal cell 423 proliferation underlies carcinogenesis, understanding metabolic requirements for proliferating epithelial 424 cells can have profound implications in oncology research. Our findings offer a potential explanation for 425 the metabolic advantage conferred by aerobic glycolysis in tumors and proliferating cells. Existing hypotheses for why aerobic glycolysis is common in proliferating cells include rapid ATP generation by 427 glycolysis (though less efficient than oxidative phosphorylation), as well as increased fluxes to glycolytic 428 intermediates for biosynthesis (Sullivan et al., 2015). We find that in the presence of EGF, where ATP 429 and NADH are low but proliferative rate is high, cells display increased bioenergetic instability and 430 sensitivity to inhibition of oxidative respiration, which can be reversed by insulin-mediated activation of For a subset of the data, we additionally verified the automated tracking results manually. After cell 512 tracking with the YFP images, the coordinates were applied to the other fluorescent channels. The 513 nuclear masks were eroded by 1 μm to ensure the exclusion of cytoplasmic pixels; the nuclear T-514 Sapphire, CFP, YFP, and RFP signals were calculated as the mean pixel values within the nuclear 515 masks in the respective images. The cytoplasmic CFP, YFP, and RFP signals were calculated as the 516 mean pixel value within a cytoplasmic "donut" mask, which consisted of an outer rim 3-4 μm from the 517 nuclear mask and the inner rim as the perimeter of the eroded nuclear mask or 2-3 μm from the original 518 nuclear mask. NADH index was calculated as a ratio of the background-subtracted nuclear T-Sapphire 519 to YFP signal. Akt index was calculated as a ratio of the background-subtracted nuclear RFP to 520 cytoplasmic RFP signal. AMPK index was calculated as the ratio of the background subtracted 521 cytoplasmic CFP to YFP ratio; because this ratio is linearly related to the fraction of unphosphorylated            B. Decrease in glucose uptake upon Pi3K/Akt inhibition. Glucose depletion from the medium was 899 assayed immediately following a 2 hour period during which the cells were exposed to the indicated 900 conditions. All measurements were made in the presence of EGF and Insulin, and measurements are 901 normalized to the DMSO condition.