Glutamine deficiency in solid tumor cells confers resistance to ribosomal RNA synthesis inhibitors

Ribosome biogenesis is an energetically expensive program that is dictated by nutrient availability. Here we report that nutrient deprivation severely impairs precursor ribosomal RNA (pre-rRNA) processing and leads to the accumulation of unprocessed rRNAs. Upon nutrient restoration, pre-rRNAs stored under starvation are processed into mature rRNAs that are utilized for ribosome biogenesis. Failure to accumulate pre-rRNAs under nutrient stress leads to perturbed ribosome assembly upon nutrient restoration and subsequent apoptosis via uL5/uL18-mediated activation of p53. Restoration of glutamine alone activates p53 by triggering uL5/uL18 translation. Induction of uL5/uL18 protein synthesis by glutamine is dependent on the translation factor eukaryotic elongation factor 2 (eEF2), which is in turn dependent on Raf/MEK/ERK signaling. Depriving cells of glutamine prevents the activation of p53 by rRNA synthesis inhibitors. Our data reveals a mechanism that tumor cells can exploit to suppress p53-mediated apoptosis during fluctuations in environmental nutrient availability.

Solid tumors outstrip their blood supply and develop tissue microenvironments that are depleted of oxygen and nutrients1,2. Such microenvironments are characterized by chronic hypoxia and are typically located more than 180 µm away from blood vessels3,4. Another type of hypoxia—known as ‘cycling hypoxia’—arises from transient shutdown of immature vasculature and can lead to ‘reoxygenation injury’, where increased free radical synthesis causes oxidative stress and tissue damage5,6. Along with hypoxia, an unstable blood supply also causes substantial fluctuations in nutrient availability in the tumor microenvironment7. Nutrient deprivation severely inhibits tumor cell proliferation but selects for aggressive cells that display increased angiogenic and metastatic ability8,9.

In order to preserve energy balance during metabolic stress, tumor cells evolve adaptive mechanisms to attenuate ATP-costly processes under nutrient deprivation10,11,12. These adaptations involve the inhibition of protein synthesis, which is the most energy-consuming process in the cell12. For instance, the eukaryotic elongation factor 2 kinase (eEF2K) is activated under nutrient restriction and inhibits the translation elongation factor eEF2, thereby suppressing overall protein synthesis and promoting cell survival under nutrient depletion2. Metabolic stress inhibits mammalian target of rapamycin complex 1 (mTORC1)13. This impairs mRNA translation initiation by activation of the eIF4E-binding protein 1 (4E-BP1), which inhibits the cap-binding translation initiation factor eIF4E13. Nutrient stress also triggers the integrated stress response (ISR), which leads to the inactivation of the translation factor eIF2α, thereby inhibiting translation initiation and promoting the synthesis of stress-adaptive proteins including ATF414. Another critical adaptation is the suppression of rRNA synthesis by RNA Polymerase I (Pol I), which comprises up to 60% of total cellular transcription15,16. mTORC1 and AMP-activated protein kinase (AMPK) reciprocally regulate pre-rRNA biosynthesis by phosphorylating TIF-IA, the protein complex that anchors Pol I to the rDNA promoter15,16. While oncogenic adaptations to nutrient deprivation have been well characterized, it is unknown how starving tumor cells resume proliferative capacities upon nutrient restoration.

Ribosome biogenesis is a highly sophisticated pathway that requires the production of four rRNAs and eighty ribosomal proteins17,18. Hundreds of protein trans-acting factors and small nucleolar RNAs aid the assembly process17,18. The process starts in the nucleolus where Pol I synthesizes a long polycistronic pre-rRNA that undergoes rapid folding, modification, processing, and associations with ribosomal proteins19,20. In unstressed cells, the tumor suppressor p53 is maintained at low levels by constitutive proteasomal degradation promoted by MDM2 (HDM2 in humans, henceforth denoted MDM2)21. When ribosome biogenesis is compromised, ribosomal components freely accumulate, including the ribosomal proteins uL5 (formerly known as RPL11) and uL18 (RPL5), which, together with the 5S rRNA, form a stable trimeric complex that binds and inhibits MDM2, leading to p53 stabilization and subsequent p53-mediated cell cycle arrest or apoptosis22,23,24.

Given the importance of ribosome biogenesis in cancer cell growth, highly potent first-in-class Pol I inhibitors—such as CX-5461 and BMH-21—have been developed to block rRNA synthesis and stabilize p5325,26. These inhibitors have the advantage of stabilizing p53 without inducing severe DNA damage, which is beneficial over mainstay genotoxic chemotherapies. Surprisingly, despite the importance of ribosome biogenesis to all cell types, single-agent therapy with CX-5461 has proven more effective against p53 wild-type hematological malignancies than against p53 wild-type solid tumors25,27,28,29. Notably, CX-5461 has been shown to possess antitumor activity against select solid tumor cell lines30. However, given the lack of data demonstrating in vivo on-target drug activity (pre-rRNA depletion, p53 stabilization), it is not entirely clear whether the antitumor activity is due to on-target or secondary effects.

Here, we report that nutrient deprivation severely impairs pre-rRNA processing and causes the accumulation of unprocessed rRNAs. Upon nutrient refeeding, protein synthesis resumes and newly synthesized ribosomal proteins (including uL5/uL18) associate with stored pre-rRNAs for ribosome assembly. Failure to accumulate pre-rRNAs under nutrient starvation leads to cell-lethal p53 stabilization, which is caused by uL5/uL18-mediated inhibition of MDM2. We also show that the activation of p53 by uL5/uL18 is dependent on glutamine availability. Glutamine mediates the production of unassembled uL5/uL18 by activating Ras/Raf/MEK/ERK signaling, which suppresses eEF2K to allow global eEF2-mediated translation. Our study identifies pre-rRNA accumulation as a mechanism by which cells use to stay alive during fluctuations in environmental nutrient availability. Finally, we show that glutamine deprivation prevents the activation of p53 by rRNA synthesis inhibitors, strongly suggesting that glutamine-deficient solid tumors display inherent resistance to small molecules that target Pol I.

Nutrient deprivation upregulates pre-rRNA expression

We first sought to understand how nutrient stress affects the rate of pre-rRNA synthesis by Pol I. We used the 5-fluorouridine (5FU) labeling assay to measure nascent pre-rRNA synthesis (see “Methods”). 5FU is a nucleoside analog that is incorporated into nascent RNAs, however, as the majority of newly synthesized RNA is pre-rRNA, 5FU labeling mainly occurs in the nucleolus (Supplementary Fig. 1a). Of note, this assay measures the rate of pre-rRNA synthesis (i.e. Pol I activity) rather than total pre-rRNA expression. To study nutrient stress, a panel of cell lines (colorectal HCT116, melanoma A375, lung adenocarcinoma A549, osteosarcoma U2OS, gastric MKN45) were cultured in medium lacking glucose, amino acids, and serum. After 24 h of starvation, the cells were pulsed with 5FU and processed for immunofluorescence. 5FU incorporation was significantly suppressed by nutrient deprivation (ND) in all cell lines tested (Fig. 1a). Thus, Pol I-mediated pre-rRNA synthesis is abrogated by metabolic stress.

a 5FU labeling under nutrient deprivation (24 h). Cancer cell lines were cultured under ND (-Glucose–Amino acids–FBS) for 24 h, and pre-rRNA synthesis was measured by 5FU metabolic labeling. Data represent the mean ± SD of three images. P values (left to right): <0.0001, 2 × 10−4, <0.0001, <0.0001, 0.0001 (∗∗∗p < 0.001; ∗∗∗∗p < 0.0001, statistical analysis by two-way ANOVA). Scale bar, 10 μm. b Schematic of pre-rRNA processing. 47S processing produces intermediate precursors which mature to 18S, 28S, and 5.8S rRNAs. The primary 47S pre-rRNA contains two external transcribed spacers (5′ETS and 3′ETS) and two internal transcriber spacers (ITS1 and ITS2). c Left: Pre-rRNA expression in HCT116 treated to CX-5461 or ND using northern blotting (ITS1 Probe). Right: Quantification of 47S/45S and 30S intermediates in response to CX-5461(10 µM) /ND. The full northern blot is shown in Supplementary Fig. 1b. d Left: Pre-rRNA expression in 3T3-L1 cells treated to CX-5461 or ND using northern blotting (ITS1 Probe). Right: Quantification of 47S/45S and 34S intermediates in response to CX-5461/ND. The full northern blot is shown in Supplementary Fig. 1c. e Expression of pre-rRNAs (47S/45S/30 S species) after CX-5461 (10 µM, 8 h) exposure. Data represent mean ± SD of n = 5 technical replicates. P values (left to right): <0.0001, <0.0001, <0.0001, <0.0001 (∗∗∗p < 0.0001, statistical analysis by one-way ANOVA). f Expression of pre-rRNAs (47S/45S/30S species) after 24 h ND in HCT116, A549, U2OS, and MKN45 cells, as assessed using qRT-PCR. Data represent mean ± SD of n = 3 technical replicates. P values (left to right): 0.0132, 0.0213, 0.0115, 0.0040 (∗p < 0.05; ∗∗p < 0.01, statistical analysis by two-way ANOVA).

Full size image

We next tested the effect of ND on the expression of each pre-rRNA species (Fig. 1b). HCT116 cells were subjected to ND for 24 h and pre-rRNA expression was analyzed by northern blotting. The Pol I inhibitor CX-5461 was used as a positive control for the inhibition of pre-rRNA synthesis. As expected, CX-5461 swiftly depleted the expression of all pre-rRNA intermediates within 2 h (Fig. 1c). This acute effect is due to the short half-lives of the pre-rRNAs (approximately 30 min), which is in turn caused by rapid processing30. Surprisingly, despite the inhibitory effect of ND on pre-rRNA synthesis, the expression of large precursors (47S, 45S) did not decrease after 24 h ND (Fig. 1c). ND also induced the robust upregulation of the downstream 30S species by approximately two-fold (Fig. 1c). Other precursors (21S, 18S-E) were slightly downregulated after 24 h of ND but were not decreased to the extent observed in CX-5461 treated cells (Fig. 1c, Supplementary Fig. 1b). Similarly, in mouse 3T3-L1 fibroblasts, ND did not affect 47 S/45S levels and upregulated the 34S intermediate (equivalent to human 30S) (Fig. 1d). Taken together, these results show that suppressed pre-rRNA synthesis under ND does not lead to decreased pre-rRNA expression.

We next used qRT-PCR to quantify the effect of ND on pre-rRNA levels. Given that the 30 S intermediate is upregulated by ND (Fig. 1c), we selected primers that measure the total abundance of the 47S, 45S, and 30S precursors. These primers bind across a cleavage site of the 5′-external transcribed sequence (5′ETS) of the pre-rRNA (Supplementary Fig. 1d). CX-5461 significantly decreased pre-rRNA expression (Fig. 1e). In contrast, 24 h ND significantly upregulated pre-rRNA levels in HCT116, A549, and MKN45 cells (Fig. 1f).

Pre-rRNA processing is impaired under ND

We hypothesized that upregulated pre-rRNA levels under ND is a direct consequence of impaired pre-rRNA processing. We used the L-(methyl-3H)-methionine metabolic labeling assay to assess pre-rRNA processing31. Control and ND HCT116 cells were pulsed with L-(methyl-3H)-methionine for 30 min and then chased with excess cold methionine (Fig. 2a). The maturation of L-(methyl-3H)-methionine labeled pre-rRNAs was tracked over 4 h. In control cells, the expression of large precursors (47S, 45S) and downstream intermediates (32S, 30S) quickly decreased over time, indicating that these precursors were metabolized into mature 18S/28S rRNAs (Fig. 2a). In contrast, ND caused the robust accumulation of the 47S, 45S, 32S, and 30S precursors (Fig. 2a), demonstrating that these intermediates did not mature to 18S/28S rRNAs. Accordingly, the production of 18S and 28S rRNAs was strongly impaired under ND (Fig. 2b). Remarkably, adding back the non-essential amino acid glutamine alone to starved cells fully rescued 18S and 28S rRNA production (Fig. 2b). This indicates that unprocessed rRNAs that accumulate under ND are utilized for ribosome biogenesis during nutrient restoration.

a Analysis of pre-rRNA processing kinetics in control or ND HCT116 cells. HCT116 cells that were pretreated to 3.5 h ND were subsequently pulse labeled with L-(methyl-3H)-methionine for 30 min and chased with excess cold methionine. To assess the effect of nutrient restoration on pre-rRNAs accumulated under ND, Gln was added back to the ND medium 2 h after the cold methionine chase (“ND + Gln”). b Quantification of the 28S/18S bands from Fig. 1f (as indicated by the red triangles). The black arrow at the 2 h time point indicates the time at which glutamine was added to starved (ND) cells. c ND increases the pre-rRNA half-life (λ) as assessed using qRT-PCR. HCT116 cells were pretreated to (24 h) ND, transcription was stopped by (20 µM) actinomycin D, and total RNA at the indicated time points for qRT-PCR analysis. Data represent mean ± SD of n = 3 technical replicates. P values (left to right): 0.006, 0.0007, 3.2 × 10−6 (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). d ND slows pre-rRNA processing in A375, A549, U2OS, MKN45 cells. Data represent mean ± SD of n = 3 technical replicates. P values (left to right): A375, 0.00081, 7.2 × 10−6, A549, 0.0035, 9.4 × 10−5, U2OS, 0.0015, 0.0031, MKN45, 0.00025, 0.00031 (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). e HCT116 tumor pre-rRNA half-life as determined using qRT-PCR. Data represent mean ± SD of n = 3 technical replicates. f Expression of pre-rRNAs, 18S, and 28S in HCT116 tumors after actinomycin D (1.33 mM) intratumoral injection. Data represent mean ± SD of n = 4 technical replicates. P values (left to right): 6.5 × 10−5, 0.20, 0.12 (ns, not significant; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). g Schematic of core and periphery regions of established HCT116 tumors (approximate size 1000 mm3). h Expression of HIF-1α in HCT116 periphery and core tissues as assessed using western blotting. i Pre-rRNA processing is faster in peripheral tissues compared to core tissues, as determined using qRT-PCR. Data represent mean ± SD of n = 3 technical replicates. P values (left to right): 0.046, 0.00064, 0.00016 (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001).

Full size image

To rigorously confirm that ND slows pre-rRNA processing, we measured the pre-rRNA half-life using qRT-PCR and the 5′ETS primers (Supplementary Fig. 1d). Transcription was blocked in control and ND HCT116 cells using actinomycin D (20 μM), and pre-rRNA expression was monitored over 1 h (see “Methods”). In control HCT116 cells, the pre-rRNA half-life was approximately 28 min, while cells subjected to ND displayed a significantly longer half-life of 78 min (Fig. 2c). Similarly, ND significantly increased the pre-rRNA half-life by at least two-fold in A375, A549, U2OS, and MKN45 cells (Fig. 2d).

Solid tumors are insufficiently vascularized with microenvironments that are strongly deprived of nutrients7. We therefore sought to estimate the half-life of pre-rRNAs in vivo. To investigate this, we injected a highly concentrated solution of actinomycin D (1.33 mM) directly into HCT116 tumors to acutely halt transcription. After injection, tumor tissues were found to contain >100 μM Actinomycin D, indicating sufficient drug diffusion (Supplementary Fig. 1e). Further, since the inhibition of Pol I transcription causes nucleolar proteins to translocate to the nucleoplasm, a process known as nucleolar disruption32,33, we assessed the localization of nucleolin in tumor cryosections. Indeed, control tumors exhibited many intact nucleolin-positive nucleoli, but actinomycin D-treated tumors displayed nucleolin staining in the cytoplasm (Supplementary Fig. 2a). Next, to assess the stability of pre-rRNAs in vivo, HCT116 tumors were intratumorally injected with actinomycin D and tissues were collected over 8 h. RNA was extracted from the tissues and pre-rRNA expression was analyzed by qRT-PCR. Tumoral pre-rRNAs displayed a half-life of approximately 3.8 h (Fig. 2e). Further, actinomycin D had no significant effect on 18S and 28S expression (Fig. 2f). Next, since it has been shown that the cores of solid tumors are more severely depleted of nutrients compared to the periphery34, we reasoned that core tumor cells would display slower pre-rRNA processing kinetics than cells in the periphery. To test this, large HCT116 tumors (>1000 mm3) were injected with actinomycin D and peripheral and core tissues were dissected to differentiate tumor cells under high or low nutrient conditions, respectively (Fig. 2g)34. Core tissues expressed higher levels of HIF-1α, consistent with the notion that core regions of solid tumors are hypoxic (Fig. 2h). Furthermore, metabolomic analysis showed that core tissues had significantly lower levels of several metabolites including glutamine, arginine, serine, and histidine (Supplementary Fig. 2e). Finally, in line with our hypothesis, peripheral pre-rRNAs exhibited a half-life of 3.5 h while core pre-rRNAs displayed a significantly longer half-life of 8.3 h (Fig. 2i).

Pre-rRNA accumulation is an adaptive response to nutrient restoration

Since pre-rRNAs that accumulate under ND are utilized for ribosome biogenesis during nutrient (glutamine) feedback (Fig. 2b), we hypothesized that pre-rRNA accumulation may be important to the resumption of ribosome assembly during nutrient recovery. To address this, we tested whether inhibiting pre-rRNA expression under ND leads to perturbed ribosome assembly during metabolic recovery. First, to prevent pre-rRNA accumulation, HCT116 or A375 cells were placed under ND with co-treatment of vehicle, CX-5461, or BMH-21 (a second Pol I inhibitor). Inhibition of pre-rRNA synthesis with CX-5461 or BMH-21 effectively blocked pre-rRNA accumulation under ND (Fig. 3a). Notably, ND alone caused high cell death, but this was not exacerbated by CX-5461 or BMH-21 co-treatment (Fig. 3b). Next, to induce metabolic recovery, the ‘pre-rRNA-depleted’ cells (ND + CX-5461 or ND + BMH-21) were transferred to drug-free nutrient-rich media for 8 h (Fig. 3c). The fidelity of ribosome assembly was evaluated by assessing the expression of unassembled ribosomal proteins, which accumulate upon defective ribosome biogenesis35. To evaluate ribosome-free proteins, protein lysates were depleted of large ribosomes using sucrose cushion ultracentrifugation, and the resulting ribosome-free lysates were analyzed by western blotting (Fig. 3d). We focused on the expression of the key ribosomal proteins uL5 (formerly RPL11) and uL18 (RPL5) because these two proteins have been shown to specifically accumulate in the ribosome-free fraction upon impaired ribosome biogenesis35. Ribosome-free uL5/uL18 was undetectable in ND vehicle cells subjected to nutrient restoration (NR), indicating that newly translated uL5/uL18 were efficiently incorporated into ribosomes during metabolic recovery (Fig. 3e). In contrast, unassembled uL5/uL18 was expressed in pre-rRNA-depleted cells (ND + CX-5461) subjected to NR (Fig. 3e). This showed that these ribosomal proteins were not incorporated into ribosomes. Moreover, halting translation with cycloheximide (CHX) prevented the accumulation of unassembled uL5/uL18 (Fig. 3e). This is consistent with the proposed model that ongoing protein synthesis is required for the accumulation of unassembled uL5/uL1835. We further assessed whether perturbed ribosome assembly was associated with nucleolar disruption. Indeed, nucleolin was found in the nucleoplasm in pre-rRNA-depleted cells exposed to NR, whereas the staining was confined to the nucleolus in control cells (Supplementary Fig. 3a).

a Pre-rRNA expression after treatment with vehicle (NaH2PO4), CX-5461 (10 µM), or BMH-21 (1 µM) under 24 h ND. Data represent mean ± SD of n = 3 technical replicates. P values (left to right): HCT116, <0.0001, <0.0001, A375, <0.0001, <0.0001, statistical analysis by two-way ANOVA. b Percentage of cell death in HCT116 cells cultured in control or ND medium with vehicle (NaH2PO4), CX-5461 (10 µM), or BMH-21 (1 µM) as assessed using trypan blue exclusion assay. Data represent mean ± SD of n = 3 biological replicates. (ns: not significant). P values (left to right): ND, <0.0001, <0.0001, statistical analysis by two-way ANOVA. c Schematic showing ND and nutrient restoration (NR) treatments. To inhibit pre-rRNA expression, HCT116 or A375 cells were first placed under ND in the presence of vehicle, CX-5461, or BMH-21. Cells pretreated to ND + CX-5461 (10 µM)/ND + BMH-21 (1 µM) represent “pre-rRNA-depleted” cells. To induce metabolic recovery, pre-starved cells were washed three times with PBS and were allowed to rest in drug-free DMEM (10% FBS). d In order to assess the expression of ribosome-free proteins, input lysates were clarified of large ribosomes using sucrose cushion ultracentrifugation. The resulting ribosome-free lysates were analyzed using western blotting. e Expression of ribosome-free r-proteins in metabolically recovering HCT116 cells. f NR induces cell death in pre-rRNA-depleted HCT116 and A375 cells. Representative microscopy images were taken after 24 h of NR. Scale bar, 100 μm. g Percentage of cell death of metabolically recovering cells as determined using trypan blue exclusion assays. Data represent mean ± SD of n = 3 biological replicates. P-values (left to right): HCT116, <0.0001, <0.0001, A375, <0.0001, <0.0001, statistical analysis by two-way ANOVA. h NR induces apoptosis in pre-rRNA-depleted cells, which is inhibited by the pan-caspase inhibitor zVAD-FMK (1 µM). i Pre-rRNA synthesis resumes after CX-5461 (10 µM)/BMH-21 (1 µM) drug washout and NR. Data represent mean ± SD of n = 3 independent images. P values (left to right): 0.077, 0.0422, statistical analysis by two-way ANOVA. Scale bar, 10 μm. For a, b, g, i, ns, not significant; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

Full size image

To assess the consequence of impaired ribosome assembly, cells depleted of pre-rRNAs were exposed to prolonged NR (24 h) and cell recovery was initially assessed using light microscopy. Starved control (ND + Vehicle) HCT116 and A375 cells resumed proliferation after NR, as shown by increased confluence. In striking contrast, starved pre-rRNA-depleted cells exposed to NR did not resume proliferation and further exhibited features of apoptosis such as reduced cell size and rounded morphology (Fig. 3f). Trypan blue exclusion assays revealed profound levels of cell death in pre-rRNA-depleted cells undergoing NR but not in control cells (Fig. 3g). The presence of cleaved caspase-3 and PARP indicated that NR-mediated cell death was apoptosis, detection of which was abrogated by the addition of the pan-caspase inhibitor zVAD-FMK (Fig. 3h). Furthermore, we observed that pre-rRNA-depleted cells resumed pre-rRNA synthesis upon NR (Fig. 3i), indicating that the effects of CX-5461 and BMH-21 are reversible. Taken together, these results show that failure to accumulate pre-rRNAs under ND leads to perturbed ribosomal subunit assembly and apoptosis during NR.

Nucleolar surveillance drives NR-mediated apoptosis

Disruptions in ribosome assembly cause unassembled uL5/uL18 to accumulate and bind to MDM2 in an inhibitory manner, leading to the stabilization of p53 and activation of p53-dependent apoptosis22,23,24,25,35,36,37. Known as the nucleolar surveillance pathway, we hypothesized that this mode cell death was the primary mechanism by which NR induces apoptosis in pre-rRNA-depleted cells. In line with this, pre-rRNA-depleted (ND + CX-5461/ND + BMH-21) cells subjected to NR accumulated high levels of p53 whereas this was not observed in starved vehicle (ND + Vehicle) cells (Fig. 4a, Supplementary Fig. 3b). Notably, p53 was not stabilized under ND + CX-5461 or ND + BMH-21, likely due to reduced uL5/uL18 translation (Supplementary Fig. 3c). We next immunoprecipitated MDM2 to assess the expression of associated uL5/uL18. MDM2 was undetectable in control ND cells, but was upregulated in pre-rRNA-depleted cells exposed to NR (Fig. 4b). This upregulation is due to MDM2 being a downstream transcriptional target of p53. Importantly, in pre-rRNA-depleted cells, precipitation of MDM2 showed high levels of co-immunoprecipitated uL5/uL18 (Fig. 4b).

a NR induces p53 in pre-rRNA depleted HCT116 cells as assessed using western blotting. b Co-immunoprecipitation of p53, uL5, and uL18 with MDM2. c Dual knockdown of uL5 and uL18 by siRNA inhibits NR-mediated activation of p53 and apoptotic markers. d NR induces apoptosis in HCT116 p53+/+ cells but not p53−/− cells. e NR induces apoptosis in pre-rRNA depleted HCT116 p53+/+ cells but not HCT116 p53−/− cells, as determined using Annexin V-PI flow cytometry. Data represent mean ± SD of n = 3 technical replicates. P values (left to right): p53+/+, 7.6 × 10−7, 7.9 × 10−5, p53−/−, <0.0001, <0.0001 (ns, not significant; ∗∗∗∗p < 0.0001, statistical analysis by two-way ANOVA). f (I) pre-rRNA intermediates accumulate in nutrient deprived cancer cells. In parallel, ND suppresses global mRNA translation. Following nutrient addition (NR), accumulated pre-rRNAs are bound by newly translated uL5 and uL18 for ribosome assembly. (II) Treatment with CX-5461 (1 µM) prevents cancer cells from accumulating pre-rRNAs under ND. Following NR, upon the absence of endogenous pre-rRNAs, uL5 and uL18 binds to MDM2, leading to p53 stabilization and nutrient-induced apoptosis.

Full size image

We next tested whether compromising the uL5/uL18-dependent nucleolar surveillance pathway could alleviate NR-mediated apoptosis. We therefore performed a double knockdown of uL5 and uL18 using two independent sets of siRNAs and assessed the activation of p53 and apoptotic markers. Dual depletion of uL5 and uL18 completely blocked the stabilization of p53 and greatly reduced PARP and Caspase 3 cleavage (Fig. 4c). Likewise, dual knockdown of uL5 and uL18 also decreased NR-mediated activation of p53 in A375 and A549 cells (Supplementary Fig. 3d). Independently silencing uL5 or uL18 alone also strongly inhibited p53 (Supplementary Fig. 3e). Further support for the role of p53 in NR-mediated apoptosis was obtained using p53 wild-type or knockout HCT116 cells. Pre-rRNA-depleted HCT116 p53 −/− cells exposed to NR exhibited lower levels of apoptosis compared to wild-type counterparts as shown by decreased activation of apoptotic markers and Annexin V staining (Fig. 4d, e). Finally, siRNAs targeting p53 or PUMA (the downstream effector of p53) also strongly alleviated NR-induced apoptosis in HCT116 and A375 cells (Supplementary Figs. 4a–c, 5a).

Taken together, these data show that cancer cells accumulate unprocessed rRNAs under starvation in order to escape p53-mediated apoptosis during nutrient refeeding (Fig. 4f). Pre-rRNAs that accumulate during starvation are utilized for ribosome biogenesis upon metabolic recovery (see also pulse-chase labeling Fig. 2a, b). Failure to accumulate pre-rRNAs under starvation results in unassembled uL5/uL18 binding to MDM2, leading to nutrient-induced p53-mediated apoptosis (Fig. 4f).

Glutamine metabolically activates p53 by inducing uL5/uL18 translation

The activation of p53 through the nucleolar surveillance pathway requires ongoing de novo protein synthesis of uL5 and uL1835,38. Thus, the accumulation of p53 serves as a proxy marker for active uL5/uL18 mRNA translation. Given this, we found that restoring glutamine alone was sufficient to rescue p53 in pre-rRNA-depleted HCT116 and A375 cells (Fig. 5a, Supplementary Fig. 5b). In contrast, restoration of leucine, tyrosine, and methionine had no effect on p53 (Supplementary Fig. 5c). In addition, dual knockdown of uL5/uL18 inhibited the activation of p53 by glutamine (Supplementary Fig. 5d). These results indicate that glutamine metabolically activates p53 in pre-rRNA-depleted cells by inducing uL5/uL18 protein synthesis. Finally, glutamine restoration also reactivated mTORC1, as shown by the increase of S6K (Thr 389) phosphorylation, a known mTORC1 substrate (Fig. 5a). This finding is consistent with the role of glutamine in activating mTORC139,40. This prompted us to examine whether mTORC1 is required for the metabolic stabilization of p53. To test this, pre-rRNA-depleted cells were re-fed glutamine in the presence of rapamycin or Torin1, two highly potent and specific mTORC1 inhibitors. However, rapamycin or Torin1 had no inhibitory effect on p53 (Fig. 5b). We therefore wondered whether other pathways that regulate protein synthesis, such as the Ras/Raf/MEK/ERK signaling cascade, might regulate glutamine-mediated synthesis of uL5/uL18. In support of this, glutamine restoration rescued MEK and ERK phosphorylation in pre-rRNA-depleted A375 cells (Fig. 5c). To test whether the metabolic stabilization of p53 is dependent on Raf/MEK/ERK signaling, pre-rRNA-depleted cells were re-fed glutamine in combination with a pan-Raf inhibitor, LY3009210 (400 nM) or AZ628 (500 nM). Strikingly, both pan-Raf inhibitors blocked the activation of p53 by glutamine (Fig. 5d). As expected, LY3009210 or AZ628 treatment strongly decreased ERK and MEK phosphorylation (Fig. 5d).

a Glutamine (gln) activates p53 in pre-rRNA depleted HCT116 cells as assessed by western blotting. HCT116 were cultured in ND medium with vehicle (NaH2PO4), CX-5461 (10 µM), or BMH-21 (1 µM) for 24 h. After 24 h the old treatment media was replaced with ND media containing PBS or Gln (4 mM) for 8 h. b Inhibiting mTORC1 with Rapamycin (1 µM) or Torin1 (1 µM) does not decrease the stabilization of p53 by glutamine. c Glutamine restoration rescues MEK and ERK phosphorylation in pre-rRNA-depleted A375 cells. d Inhibiting Raf with LY3009210 (400 nM) or AZ628 (500 nM) inhibits the metabolic activation of p53 by glutamine (gln). e Schematic: Glutamine restoration activates Ras/Raf/MEK/ERK signaling, which inhibits eEF2K. Upon eEF2K inhibition, eEF2 remains in its active form to facilitate global protein synthesis (including uL5/uL18 translation). Pharmacological inhibition of Raf/MEK/ERK leads to eEF2k activation, which in turn phosphorylates and inactivates eEF2. f Inhibiting MEK (10 µM U0126, 50 µM PD98059) or ERK (1 µM LY3214996, 0.1 µM GDC-0994) inhibits the metabolic activation of p53 by glutamine (gln). g Inhibiting Raf (400 nM LY3009210), MEK (10 µM U0126) and ERK (1 µM LY3214996) inhibits the metabolic activation of p53 by glutamine (gln) in A375 cells. h Expression of polysome-bound r-protein mRNAs. HCT116 cells were placed under ND for 24 h. After 24 h the old treatment media was replaced with ND media containing 4 mM Gln, LY3009210 (400 nM), or Gln + LY3009210. Polysomes were collected via ultracentrifugation on a 40% sucrose cushion and total RNA was extracted and assayed using qRT-PCR for uL5, uL18, uL23, and S3. Data represent mean ± SD of n = 4 technical replicates. P-values (left to right): uL5, <0.0001, 0.0003, <0.0001, uL18, <0.0001, 0.0052, uL23, <0.0001, 0.0003, <0.0001, S3, <0.0001, <0.0001, <0.0001 (∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001, statistical analysis by two-way ANOVA).

Full size image

We next sought to understand the potential regulatory effects of ERK signaling on uL5/uL18 translation. Given that mTORC1 primarily regulates translation initiation41, and that mTORC1 suppression did not inhibit p53 (Fig. 5b), we reasoned that ERK might regulate uL5/uL18 via the elongation step of translation. Indeed, in unstressed cells, ERK facilitates mRNA translation by inhibiting eukaryotic elongation factor-2 kinase (eEF2K), a key negative regulator of protein synthesis (Fig. 5e)42,43. Upon eEF2K inhibition, its downstream target eukaryotic elongation factor-2 (eEF2) remains in its active unphosphorylated form to mediate global translation elongation (Fig. 5e)42,43. Thus, we hypothesized that uL5/uL18 translation is dependent on ERK-mediated activation of eEF2 (Fig. 5e). We therefore assayed levels of eEF2 phosphorylation; phosphorylated eEF2 (Thr56) being the inactivated form that leads to arrest of general protein synthesis42,43. Inhibiting Raf with LY3009210 or AZ628 increased eEF2 phosphorylation in both starved control and pre-rRNA-depleted cells (Fig. 5d). We also tested if inhibiting the downstream kinases MEK and ERK would produce the same effect on eEF2 and p53. Pharmacological inhibition of MEK (U0126, PD98059) or ERK (LY3214996, GDC-0994) similarly induced eEF2 phosphorylation and p53 inhibition (Fig. 5f). Likewise, in A375 cells, inhibition of Raf (LY3009210), MEK (U0126), and ERK (LY3214996) completely blocked p53 expression (Fig. 5g). Next, to assess the effect of Raf inhibition on uL5/uL18 mRNA translation, we measured the expression of translationally active polysome-bound mRNAs (see “Methods”). ND alone decreased the proportion of polysome-bound uL5/uL18 mRNAs (Fig. 5h). Importantly, glutamine restoration rescued uL5/uL18 mRNA translation, but was inhibited by co-incubation with LY3009210 (Fig. 5h). Consistent with the role of eEF2 in general mRNA translation42,43, a similar effect was observed with uL23 and S3 polysome-bound mRNAs (Fig. 5h).

Glutamine deprivation confers cancer cells resistance to RNA Pol I inhibitors by blocking the nucleolar surveillance pathway

Since glutamine restoration stabilized p53 in pre-rRNA-depleted cells (Fig. 5a), we next tested the effect of glutamine deprivation on the activation of p53 by CX-5461. Strikingly, glutamine deprivation blocked the activation of p53 by 1 μM CX-5461, while the withdrawal of glucose or other amino acids had no effect (Fig. 6a, Supplementary Fig. 6a). Glutamine deprivation did not inhibit p53 stabilization by the MDM2 inhibitor AMG 232 (Supplementary Fig. 6b). Suppression of p53 transcriptional activity under glutamine deprivation was confirmed by measuring p21 mRNA (Supplementary Fig. 6c). The inhibition of p53 was not due to reduced uL5/uL18 translation, as glutamine deprivation did not noticeably inhibit uL5/uL18 nascent protein synthesis (Supplementary Fig. 6d). Interestingly, 1 μM CX-5461 failed to decrease pre-rRNA expression under glutamine deprivation (Supplementary Fig. 6e), however, increasing the dose to 10 μM rescued pre-rRNA inhibition and p53 activation (Supplementary Fig. 6f, g). CX-5461 caused uL5/uL18 to co-immunoprecipitate with MDM2 under control conditions but not under glutamine deprivation (Supplementary Fig. 6h). Glutamine deprivation inhibited the activation of p53 by BMH-21 and two other Pol I inhibitors, CX-3543 and actinomycin D (Fig. 6b). Glutamine deprivation also inhibited p53 in A375, A549, U2OS, LNCaP, and MKN45 cells (Fig. 6c), and conferred resistance to CX-5461 in HCT116 cells (Fig. 6d) as well as LNCaP and MKN45 cells (Supplementary Fig. 7a). Glucose deprivation, which does not inhibit p53, did not confer resistance to CX-5461 (Fig. 6d). Overexpression of p53 under glutamine deprivation rescued the growth suppressive effects of CX-5461 (Supplementary Fig. 7b). In addition, rescuing p53 with the non-genotoxic MDM2 inhibitor AMG 232 also inhibited the viability of glutamine-deprived cells (Supplementary Fig. 7c).

a Glutamine deprivation inhibits the activation of p53 by CX-5461 treatment. HCT116 cells were cultured in the respective starvation medium for 16 h followed by CX-5461 (1 µM) for 8 h. b Glutamine deprivation inhibits p53 activation by BMH-21 (1 µM), CX-3543 (1 µM), and ActD as assessed by western blotting. c Glutamine deprivation inhibits p53 activation by CX-5461 (1 µM) in A375, A549, U2OS, LNCaP, and MKN45 cells. d Deprivation of glutamine but not glucose conferred resistance to CX-5461 (1 µM) as assessed using MTT assay. Data points of each treatment were normalized to the respective vehicle-treated metabolic control. Data represent mean ± SD of n = 6 biological replicates. e p53 activation by CX-5461 was suppressed under 0.4 mM glutamine as assessed by western blotting. f CX-5461 failed to inhibit HCT116 tumor growth. Mice were orally dosed with vehicle (NaH2PO4), CX-5461 (50 mg/kg), or AMG 232 (50 mg/kg) once daily three times per week. Drug treatment started 5 days post implantation on tumors of approximately 75 mm3. Data represent mean ± SEM of n = 8 mice/group. P values: Vehicle/CX-5461, 0.078, Vehicle/AMG 232, 2.4 × 10−4 (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). g CX-5461 does not acutely stabilize p53 in HCT116 tumors. Mice were IP injected with vehicle (NaH2PO4), CX-5461 (50, 100, 200 mg/kg), or AMG 232 (50 mg/kg) and tumors were harvested after 8 h for western blot analysis. (h) p53 IHC staining in A375 tumors following vehicle or drug IP administration. Scale bar represents 50 µm.

Full size image

Solid tumors contain low levels of glutamine due to rapid consumption by tumor cells34. Thus, we predicted that CX-5461 would fail to stabilize p53 in vivo. Indeed, p53 was suppressed under 0.4 mM glutamine (Fig. 6e), which is approximately the concentration of glutamine found in tumor xenografts34. We selected HCT116 as our model cell line, since CX-5461 has been shown to potently inhibit HCT116 in vitro cell proliferation at an IC50 167 nM30. Oral gavage of CX-5461 at the highest long-term tolerable dose (50 mg/kg) failed to significantly inhibit tumor growth (Fig. 6f). We used AMG 232 as a positive control for p53-mediated tumor suppression. Oral gavage of AMG 232 (50 mg/kg) at the same dose and schedule as CX-5461 significantly suppressed HCT116 tumor growth (Fig. 6f). Furthermore, IP injection of CX-5461 up to 200 mg/kg to HCT116 tumors failed to activate p53 while AMG 232 (50 mg/kg) robustly induced p53 (Fig. 6g). Similarly, AMG 232 but not CX-5461 stabilized p53 in A375 tumors (Fig. 6h). These results indicate that tumoral glutamine deficiency hampers the antitumor activity of Pol I inhibitors by blocking the p53-dependent nucleolar surveillance pathway.

Unstable vasculature in solid tumors leads to the development of tissue microenvironments that fluctuate in nutrient availability7. Tumor cells residing in such environments adapt to nutrient depletion by suppressing processes that heavily utilize ATP such as ribosome biogenesis10,11,12, but how cells recover during nutrient restoration remains to be elucidated. Here, we report that metabolically stressed mammalian cells accumulate unprocessed rRNAs in order to escape nucleolar surveillance upon nutrient stress termination. We first observed that ND inhibits pre-rRNA processing, leading to the accumulation of pre-rRNA intermediates in nutrient starved cells. Upon nutrient restoration, recovering cells resume ribosome biogenesis and pre-rRNAs accumulated during stress are converted into mature rRNAs. Cells that fail to accumulate pre-rRNAs under ND undergo cell-lethal p53 activation during nutrient refeeding. Mechanistically, p53 is activated by unassembled uL5/uL18 binding and inhibiting MDM2. Our data shows that pre-rRNA accumulation during nutrient deprivation is a stress adaptation that restrains the MDM2-binding activity of uL5 and uL18 during fluctuations in nutrient availability.

During the preparation of this manuscript, CX-5461 was reported to possess additional inhibitory activities against Topoisomerase II44,45. This presents a limitation to our manuscript since we employed CX-5461 to inhibit pre-rRNA accumulation under ND. We addressed this limitation by employing BMH-21 alongside CX-5461 in our experiments.

We further identified that the activation of p53 through the uL5/uL18-dependent nucleolar surveillance pathway is dependent on exogenous glutamine. In pre-starved cells depleted of pre-rRNAs, refeeding glutamine was sufficient to stabilize p53, compatible with rescue of uL5/uL18 translation. Surprisingly, despite the known importance of mTORC1 in the translational control of 5′TOP mRNAs (that encode ribosomal proteins including uL5 and uL18)46,47, rapamycin or Torin1 treatment did not affect the stabilization of p53 by glutamine. Since the accumulation of p53 protein is a marker of uL5 and uL18 synthesis35,38, this observation suggested that uL5/uL18 translation was not solely dependent on mTORC1. Rather, we found that pharmacologically inhibiting Raf/MEK/ERK completely blocked the accumulation of p53. Raf/MEK/ERK inhibition appeared to arrest the synthesis of uL5 and uL18 by inactivating eEF2. Altogether, these results suggest that glutamine-induced synthesis of uL5 and uL18 is facilitated by the Ras/Raf/MEK/ERK-eEF2K-eEF2 axis.

Finally, we found that glutamine deprivation confers resistance to CX-5461 by blocking the stabilization of p53. This finding was supported in vivo where CX-5461 failed to stabilize p53, which could explain the little inhibitor effect on HCT116 tumor growth, particularly given the reported low glutamine tumor microenvironment34. In line with our findings, another study observed that p53 wild-type HCT116 tumors are highly resistant to CX-546148. This is despite another study demonstrating that HCT116 cells are remarkably sensitive to CX-5461 in vitro30. Thus, these results indicate that CX-5461 may display limited therapeutic benefit to solid tumors due to tumoral glutamine deficiency. Notably, glutamine deprivation did not decrease uL5/uL18 protein synthesis, indicating that the mechanism of p53 inhibition is independent of uL5/uL18 mRNA translation. Altogether, our data shows that tumoral glutamine deficiency hampers the effectiveness of Pol I inhibitors by blocking the nucleolar surveillance pathway. These findings offer an explanation as to why the efficacy of CX-5461 is not associated with p53 wild-type status in solid tumor types. Further, these results may explain why CX-5461 is highly effective against blood cancers25,27,28,29, in which severe pathological nutrient deprivation does not occur. Indeed, plasma glutamine concentrations are approximately 0.8 mM49, which is twice higher than the concentration at which p53 was found to be inhibited in this study (Fig. 4e). Further, since the rate of ribosome biogenesis is dependent on nutrient availability, this implies that cells of hematological malignancies possess higher rates of ribosome biogenesis than cells of solid tumors and are therefore more vulnerable to Pol I inhibitors.