Fragile X messenger ribonucleoprotein 1 (FMRP) regulates glycolytic gene expression under chronic hypoxia in HCT116 cells

Abstract

Oxygen shortage, known as hypoxia, occurs commonly in both physiological and pathological conditions. Transcriptional regulation by hypoxia-inducible factors is a dominant regulatory mechanism controlling hypoxia-responsive genes during acute hypoxia; however, recent studies suggest that post-transcriptional regulation, including RNA degradation, also involves hypoxia-induced gene expression during the chronic hypoxia. In this study, we developed a method to quantify the contributions of RNA synthesis and degradation to differential gene expression, and identified 102 genes mainly regulated via RNA degradation under chronic hypoxia in HCT116 cells. Bioinformatics analysis showed that the genes mainly regulated by RNA degradation were involved in glycolysis. We examined changes in the RNA-binding ability of RNA-binding proteins by RNA interactome capture and statistical analysis using public databases. We identified fragile X messenger ribonucleoprotein 1 (FMRP) as an RNA-binding protein involved in the chronic hypoxia-induced increase in mRNAs encoding rate-limiting enzymes. This study emphasizes the importance of post-transcriptional gene regulation under chronic hypoxia in HCT116 cells.

Introduction

Aerobic organisms utilize oxygen to produce chemical energy for the tricarboxylic acid cycle (TCA cycle) and electron transport chain^1,2,3,4,5. Adenosine triphosphate (ATP) supplies the chemical energy for diverse biochemical reactions^6,7. Low oxygen conditions, referred to as hypoxia, thus causes a crisis for aerobic organisms. Hypoxia occurs under both physiological and pathological situations^8,9; for instance, hypoxic conditions are observed in embryonic development^10,11. Normal mammalian development occurs in a hypoxic environment, and the hypoxic environment is absolutely required in aspects of developmental morphogenesis for placental and fetal heart¹¹. In addition, an insufficient vascular supply caused by rapid growth of tumor tissues results in hypoxic regions within tumor tissues, thus enhancing the epithelial-to-mesenchymal transition of cells, which in turn increases cell motility and metastasis^12,13. Pathological hypoxic conditions are also observed in chronic heart and kidney diseases⁹ and in cardiovascular diseases¹⁴.

To avoid the crisis caused by oxygen deficiency, cells have developed various mechanisms to respond to hypoxia. Regulation of transcription modules by hypoxia-inducible factors (HIFs) is a widely studied hypoxia-response mechanism. Among these, HIF1, which plays a crucial role in the hypoxia response, was originally identified as a protein binding to the hypoxia response element (HRE) of the erythropoietin gene (EPO)^15,16,17. During hypoxia, the regulatory subunit of HIF1, HIF1α, is stabilized and transported into the nucleus to regulate the transcription of target genes such as VEGF and GLUT1^18,19. The regulatory subunit HIF2α is highly homologous to HIF1α, and is specifically expressed in certain tissues such as the endothelium, lung, and brain. EPO was reported to be regulated by HIF2α under stress conditions in vivo²⁰. HIF1 regulates cell cycle arrest and apoptosis upon hypoxia^21,22, and inflammation during liver injury²³. Although HIF1 is known to be a dominant regulatory factor in acute hypoxia (several hours), recent studies have suggested that the chronic hypoxic response (several days) is regulated by other mechanisms. HIF1α protein levels increase rapidly at the onset of hypoxia and return to basal levels under prolonged (chronic) hypoxia²⁴. We previously reported that chronic hypoxia activates transcription factors other than HIF1, including cAMP-response element binding protein (CREB) and nuclear factor-kB (NF-kB), contributing to tumor malignancy²⁴.

Despite limited reports, post-transcriptional mechanisms, such as RNA stability, have also been shown to be involved in acute hypoxic responses^25,26,27. Some hypoxia-responsive RNAs, including VEGF, EPO, and TH, are stabilized under acute hypoxic conditions, and some RNA-binding proteins (RBPs), such as HuR (also known as ELAVL1), are known to be involved in the stability of these RNAs^{27,28,29,30,31}. These results suggest that gene expression profiles under chronic hypoxia are regulated by distinct mechanisms, including the regulation of RNA stability, in contrast to the HIF1-dominant acute phase. Nevertheless, the landscape of gene expression regulation via post-transcriptional mechanisms such as RNA degradation, in response to chronic hypoxia, remains unclear.

Recent developments in techniques, including next-generation sequencing (NGS), have enabled us to measure RNA kinetics, comprising RNA synthesis and degradation, as well as gene expression levels^{32,33,34,35,36,37,38,39,40,41}. For instance, SLAM-seq, TimeLapse-seq, and TUC-seq enable in situ labeling of intracellular RNAs with 4-thiouridine (4sU) followed by base conversion to distinguish newly synthesized RNAs and measuring RNA synthesis rate^42,43,44. BRIC-seq enables pulsed labeling of intracellular RNAs to quantify RNA degradation rates^45,46. We recently developed Dyrec-seq to quantify RNA synthesis and degradation rates simultaneously and comprehensively by multi-labeling of endogenous RNAs with both 4sU and 5′-bromouridine (BrU)⁴⁷, and revealed that RNA degradation affected chronological gene expression patterns. The combined measurement of these kinetic parameters enables to reveal the contributions of transcriptional and post-transcriptional regulation to differential gene expression. Moreover, recent approaches to the proteome-wide identification of RBPs based on mass spectrometry (MS) enable the quantification of the RNA-binding properties of RBPs, as a major factor regulating RNA kinetics^48,49. These approaches have led to the development of enhanced RNA interactome capture (eRIC)-MS for accurate quantification of the RNA-binding properties of RBPs⁵⁰, enabling us to identify RBPs involved in post-transcriptional regulation under specific conditions.

In this study, we hypothesized that post-transcriptional regulation, especially RNA degradation via RBPs, controls the regulated expression of hypoxia-responsive genes in chronic hypoxia. We initially calculated RNA synthesis and degradation simultaneously in HCT116 human colorectal carcinoma-derived cells under chronic hypoxic conditions, and quantified the respective contributions of RNA synthesis and degradation to the differential gene expression. Our study led to a novel insight that mRNAs encoding enzymes involved in glycolysis (glycolytic mRNAs) is regulated via RNA degradation through an RBP, named fragile X messenger ribonucleoprotein 1 (FMRP) which is coded by FMR1 gene. This study highlights the involvement of post-transcriptional regulation via the specific RBP in the enhancement of glycolysis in chronic hypoxia.

Results

Chronic hypoxia in HCT116 cells

To investigate the chronological response of HCT116 cells in response to hypoxic conditions, we cultured the cells under hypoxic condition (1% O₂), and examined the temporal changes in RNA and protein expression. HIF1α protein levels increased up to 4 h after hypoxia and then diminished within 48 h (Fig S1A). RNA levels of the early hypoxia-response factors GLUT1 and CA9 increased continuously from 0 to 48 h (Fig S1B and S1C), while RNA levels of the chronic hypoxia-response factor MMP1²⁴ only increased at 48 h (Fig S1D). We also examined the gene expression profile of HCT116 cells under chronic hypoxia by RNA-seq analysis. To examine differences in gene expression profiles, we analyzed the RNA expression profiles using principal component analysis (PCA) (Fig S1E and S1F). The first principal component was largely altered at 24 and 48 h after hypoxic treatment, indicating that the hypoxia-induced expression profile changed within 24 h and remained changed until at least 48 h (Fig S1F). The first principal component was also largely changed at 72 h, possibly reflecting the effects of cell death. On the basis of this PCA, we considered that the cellular state in response to hypoxia reached a steady state between 24 and 48 h. We therefore defined chronic hypoxia as 36 h after applying hypoxic conditions in this study.

Simultaneous calculation of RNA synthesis and degradation rates in HCT116 cells under normoxia and chronic hypoxia states

We quantified RNA synthesis and degradation rates in HCT116 cells under chronic hypoxia by SLAM-seq, which allowed the identification of newly synthesized transcripts with 4sU labeling⁴². The labelled RNAs were reacted with iodoacetamide (IAA), a thiol-reactive compound, to attach a carboxyamidomethyl group to the thiol group in the incorporated 4sU by nucleophilic substitution (S_N2) reaction (alkylation). Because the alkylated 4sU is paired with guanine instead of adenine during reverse transcription, a 4sU incorporated into the RNA is detected as a mutation (T-to-C conversion). Determination of the 3′ untranslated regions (UTRs) of the alkylated RNA using QuantSeq, a poly(A)-tail dependent massive sequencing technique⁵¹, enabled us to quantify the expression levels of RNAs and those labeled with 4sU simultaneously (Fig. 1A, B). We determined the appropriate concentration of 4sU for labeling HCT116 cells by measuring the 4sU incorporation ratio and cell viability under the indicated 4sU concentrations (Fig S1G and S1H). The incorporation ratio of 4sU reached a plateau at approximately 2.5% with 50 µM 4sU in 12 h (Fig S1G). Treatment with up to 100 μM 4sU resulted in > 80% of proliferation (Fig S1H). On the basis of these results, we used 100 µM 4sU for RNA labeling in this study. Moreover, we confirmed efficient alkylation of 4-thiouracil, a nucleobase that compose 4sU, using IAA, by the shift in absorption spectrum (Fig S1I). High-performance liquid chromatography (HPLC) measurement of nucleoside samples derived from RNA collected from cells labeled with 100 µM 4sU and those alkylated with IAA treatment indicated a clear shift in the maximum absorption wavelength (Fig S1J), indicating efficient alkylation of 4sU incorporated into the RNAs. T-to-C conversions obtained from the alkylated 4sU-labeled RNAs increased approximately four-fold compared with those obtained from unalkylated RNAs, indicating the identifiability of 4sU-labeled RNAs in NGS (Fig S1K). Note that the “T-to-C conversion” in RNA-seq reads derived from genes located on the reverse strand for the reference genome is identified as “A-to-G conversion” during bioinformatics analysis.

Fig. 1

RNA synthesis and degradation rates. (A) Schematic diagram of 4sU labeling. HCT116 cells were cultured under hypoxic (1% O₂) conditions for 36 h, followed by incubation for a further 12 h after adding 4sU at a final concentration of 100 µM. The abundance of 4sU-labeled RNAs increased in a time-dependent manner (red line), according to the RNA synthesis and degradation rates. The expression level ($X$) of each RNA reached a plateau upon adaptation to hypoxia (gray line) (B) Base conversions in 4sU-labeled RNAs. The 4sU incorporated in intracellular RNAs during transcription is alkylated by IAA and pairs to guanosine (G) instead of adenine (A) on reverse transcription. Because G pairs with cytosine (C) during PCR as preparation for NGS analysis, alkylated 4sUs are read as C (T-to-C conversion), enabling us to distinguish 4sU-labeled RNAs in NGS analysis. (C-D) Joint distribution of RNA synthesis and degradation rates in normoxic (C) and hypoxic (D) samples. Histograms of individual rates are shown at top and right of scatter plots. The x- and y-axes indicate RNA synthesis rate (${k}_{s}$) and degradation rate (${k}_{d}$), respectively.

We collected RNAs for SLAM-seq according to the above conditions (Fig. 1A, B). Briefly, HCT116 cells were cultured under hypoxic (1% O₂) and normoxic conditions (21% O₂) for 36 h, followed by the addition of 4sU to the medium to a final concentration of 100 µM (Fig. 1A). RNAs were collected at 0, 1, 2, 4, 8, and 12 h after addition of 4sU, and the purified RNAs were subjected to alkylation of 4sUs with IAA. The collected RNA samples were provided to QuantSeq to determine the sequences of the 3′UTRs. We detected more than 30 × 10⁶ reads in all samples (Fig S2A). For alignment of the reads using the SLUMDUNK tool⁴² on the reference genome (hg38), we found that > 50% of the reads were aligned singly (Fig S2B). Finally, we counted T-to-C conversions (and A-to-G conversions) detected in the QuantSeq data for the alkylated 4sU-labeled RNA. Detection of T-to-C conversions by the SLAMDUNK tool indicated a labeling-time-dependent increase in T-to-C conversions up to approximately 8.0% (Fig S2C), resulting in approximately 40% of reads including T-to-C conversions (Fig S2D).

Based on the labeling-time-dependent increase in T-to-C conversions (and A-to-G conversions), we calculated the RNA synthesis and degradation rates in HCT116 cells simultaneously. First, we identified expressed genes based on the QuantSeq results, with gene expression levels quantified as counts per million (CPM). The CPM values were distributed unimodally in a range > 0 in individual samples (Fig S2E). Here, we considered genes with a CPM > 0 as expressed genes, and identified 11,969 RNAs derived from expressed genes in all samples (Table S1). We then simultaneously calculated the synthesis and degradation rates for RNAs derived from individual genes using the SLAM-seq data. The RNA synthesis rate, ${k}_{s}$, and degradation rate, ${k}_{d}$, were defined as the abundance of RNA molecules synthesized per min and the ratio of RNA molecules degraded per min, respectively. The expression level of an RNA is determined as the ratio of the RNA synthesis and degradation rates, and the shape of the curve is determined by the degradation rate. Namely, we can calculate the ${k}_{s}$ and ${k}_{d}$ values based on the fitting curve on the temporal increase of newly synthesized RNAs identified based on 4sU incorporation, combined with RNA expression level (see below). When the RNA expression level is at a steady state, the amount of newly synthesized RNA (${x}_{t}$) at each time point ($t$) is as follows:

$$\begin{array}{c}{x}_{t}=\frac{{k}_{s}}{{k}_{d}}\left(1-{e}^{-{k}_{d}t}\right)\end{array}$$

(1)

We can therefore calculate ${k}_{s}$ and ${k}_{d}$ by fitting the time series of 4sU-labeled RNA and expression levels estimated from SLAM-seq to Eq. 1. Prior to calculating ${k}_{s}$ and ${k}_{d}$ values, we extracted the genes in steady state after 4sU treatment (see “Materials and methods”), because a steady state of RNA expression level is required to estimate the ${k}_{s}$ and ${k}_{d}$ values. A total of 9541 genes were in steady state in both normoxic and hypoxic samples, among the 11,969 genes expressed in all samples (Table S1). For the RNAs derived from these genes, we calculated the ${k}_{s}$ and ${k}_{d}$ values at the genome-wide level by fitting the time series of 4sU-labeled RNA to Eq. 1. We also adopted RNAs with a good fit between their actual expression levels and expression levels predicted from the estimated ${k}_{s}$ and ${k}_{d}$ values (see “Materials and methods”). We were therefore able to calculate the RNA synthesis and RNA degradation rates of 8,961 and 8,479 RNAs for normoxic and hypoxic samples, respectively (Table S2). The estimated ${k}_{s}$ and ${k}_{d}$ values both obeyed a log Gaussian distribution, and no mutual correlation of ${k}_{s}$ and ${k}_{d}$ was observed in either sample (Fig. 1C, D).

Some differentially expressed genes are regulated via RNA degradation under chronic hypoxia

We aimed to estimate the relative contributions of RNA synthesis and degradation to the differential expression of RNAs, based on the calculated RNA synthesis and degradation rates (Fig. 2A). We identified RNAs that were differentially expressed in chronic hypoxia by setting a threshold change in CPM of > 1.5-fold or < 2/threefold between normoxic and hypoxic samples and a false discovery rate (FDR) of < 0.05 (paired t test). A total of 1,330 RNAs among the 11,969 genes expressed in all samples were identified as differentially expressed in hypoxia (Fig. 2B and Table S3), and the RNA synthesis and degradation rates for 1160 of these RNAs were estimated in both normoxic and hypoxic samples.

Fig. 2

Contributions of RNA synthesis and degradation rates to differential gene expression. (A) Calculation of contributions of RNA kinetics (RNA synthesis and degradation rates) to differential gene expression. RNA expression level was determined as the ratio of ${k}_{s}$ and ${k}_{d}$ values, and the difference in logarithmic expression level can thus be decomposed into ${k}_{s}$ and ${k}_{d}$. (B) Identification of DEGs in cells under chronic hypoxia. Genes with a > 1.5-fold or < 2/threefold change in expression level under hypoxia compared with that under normoxia and FDR (paired t-test) < 0.05 were extracted as DEGs. The x- and y-axes indicate log-fold change and log-FDR, respectively. Red and blue dots indicate increased (193) and decreased (1137) genes, respectively. (C) Histogram of contribution of RNA degradation to differential expression (${\rho }_{d}$). The x- and y-axes indicate ${\rho }_{d}$ value and number of RNAs, respectively. Red and blue represent up-regulated and down-regulated RNAs, respectively. The 102 RNAs whose expression levels were regulated mainly via RNA degradation (${\rho }_{d}$ > 60%) are identified. (D) Functional enrichment analysis of mRNAs whose abundances were regulated via RNA degradation under chronic hypoxia. Functions with an FDR < 0.05 are indicated. (E) GSEA of glycolysis involved in distribution of ${\rho }_{d}$ values.

We then quantified the contributions of RNA synthesis and degradation to the differential expression of RNAs. When $t$ in Eq. 1 approaches infinity, the abundance of newly synthesized RNAs, ${x}_{t}$, approaches asymptotically to expression level ($X$), thus:

$$\begin{aligned} X & = \mathop {\lim }\limits_{t \to \infty } \frac{{k_{s} }}{{k_{d} }}\left( {1 - e^{{ - k_{d} t}} } \right) \\ & \begin{array}{*{20}c} { = \frac{{k_{s} }}{{k_{d} }}} \\ \end{array} \\ \end{aligned}$$

(2)

To quantify the relationships of differential RNA expression between two conditions (e.g., normoxia vs. hypoxia) and synthesis or degradation rates, we moved Eq. 2 to log space:

$$\begin{array}{c}\Delta \text{log}\left(X\right)=\Delta \text{log}\left({k}_{s}\right)-\Delta \text{log}\left({k}_{d}\right)\end{array}$$

(3)

The sum of the contributions of RNA synthesis and degradation to the differential expression of RNAs should be one, and we therefore defined the contributions of RNA synthesis, ${\rho }_{s}$, and degradation, ${\rho }_{d}$, as relative values of $\Delta \text{log}\left({k}_{s}\right)$ or $\Delta \text{log}\left({k}_{d}\right)$ over $\Delta \text{log}\left(X\right)$:

$${\rho }_{s}=\frac{\Delta \text{log}\left({k}_{s}\right)}{\Delta \text{log}\left(X\right)}$$

(4)

and

$${\rho }_{d}=\frac{\Delta \text{log}\left({k}_{d}\right)}{\Delta \text{log}\left(X\right)}.$$

(5)

These definitions satisfy the premise that the sum of the contributions is equal to one. Estimation of the RNA synthesis and degradation rates thus enabled us to quantify their relative contributions. The estimated ${\rho }_{d}$ values obeyed an exponential unimodal distribution with 0% of mode (Fig. 2C), indicating that most gene expression levels were regulated via transcriptional regulation. However, 102 RNAs had ${\rho }_{d}$ values > 60%, indicating that their differential expression was mainly regulated via RNA degradation. Moreover, increased RNAs tended to have larger ${\rho }_{d}$ values (Fig. 2C), indicating that their increased expression in chronic hypoxia tended to be caused by RNA stabilization.

Regulation of RNA degradation involves glycolytic enhancement under chronic hypoxia

To determine the cellular functions regulated via either RNA synthesis or degradation under chronic hypoxia, we performed functional enrichment analysis for RNAs mainly regulated via either mechanism (Fig. 2D and S3, and Table S4). Differentially expressed RNAs mainly regulated via RNA synthesis were significantly enriched (FDR < 0.05) in RNA processing, lipid metabolism, and alternative splicing (Fig S3), while differentially expressed RNAs mainly regulated via RNA degradation were significantly enriched in glycolysis (Fig. 2D and S3). We also carried out gene set enrichment analysis (GSEA)⁵² based on the ${\rho }_{d}$ values to examine cooperativeness of RNA degradation on cellular functions. GSEA statistically tests the homogeneities of differential expression of RNAs involved in specific biological functions. A uniform distribution in fold change of RNAs involved in a specific term makes the p-value of the term larger, while an uneven distribution makes the p-value smaller. We used the ${\rho }_{d}$ value from change in RNA degradation rates and expression level instead of the fold change, to approach cooperative regulation of RNA degradation (see “Materials and methods”). GSEA suggested that RNAs with high ${\rho }_{d}$ values (differential expressed genes mainly regulated by RNA degradation) were significantly related to glycolysis (Fig. 2E and Table S5), as with the results of the functional enrichment analysis.

Changes in glycolysis in response to hypoxic conditions are important for cellular adaptation to hypoxia^53,54. Although the enhancement of glycolysis by acute hypoxia for up to a few hours is considered to be predominantly regulated transcriptionally via HIF⁵³, the mechanisms regulating glycolysis in chronic hypoxia remain unclear. The current findings suggest that regulation of RNA degradation, rather than transcriptional regulation, is the main mechanism responsible for the enhancement of glycolysis under chronic hypoxia. In support of this finding, metabolic enzymes encoded by mRNAs with large ${\rho }_{d}$ values, i.e., mainly regulated via RNA degradation, were involved in glycolysis in Kyoto Encyclopedia of Genes and Genomes (KEGG)^55,56,57 Metabolic pathways (hsa01100) (Fig. 3A, B), consistent with the results of the functional enrichment analysis. Moreover, mRNAs encoding rate-limiting enzymes for glycolysis, such as hexokinase-1 (HK1), phosphofructokinase, liver type (PFKL), and pyruvate kinase M1/2 (PKM), were among those with large ${\rho }_{d}$ values. In contrast, enzymes encoded by RNAs with small ${\rho }_{d}$ values, i.e., mainly regulated by transcriptional regulation, were distributed on lipid and amino acid metabolism. We confirmed that up-regulated mRNAs with high ${\rho }_{d}$ values (red in Fig. 3A) were mapped mainly to carbon metabolism, thus supporting the idea that mRNAs involved in glycolysis were stabilized during chronic hypoxia. Measurement of individual RNA levels also indicated significant increases in mRNAs encoding rate-limiting enzymes (HK1, PFKL, and PKM mRNAs) (Fig. 3C), and calculation of the half-lives of these mRNAs indicated that they were stabilized in response to chronic hypoxia (Fig. 3D). Protein levels of HK1, PFKL, and PKM were also increased under the same conditions (Fig. 3E), accompanied by the accumulation of intracellular lactate (Fig. 3F). These results suggested that expression levels of mRNAs encoding rate-limiting enzymes for glycolysis were increased by RNA stabilization, resulting in increased abundances of these enzymes to enhance glycolysis.

Fig. 3

Contribution of RNA degradation-mediated gene regulation to differential expression of RNAs in glycolysis. (A and B) Contribution of RNA degradation to differential gene expression (${\rho }_{d}$) for the up-regulated (A) and down-regulated RNAs (B) projected onto the KEGG metabolic pathways (hsa01100). Nodes and edges indicate metabolites and enzymes/reactions, respectively. Red and blue edges indicate enzymes whose RNA levels were regulated via RNA degradation (${\rho }_{d}$ > 60%) and transcriptional regulation (${\rho }_{d}$ < 40%), respectively. (C) Relative expression of mRNAs encoding rate-limiting enzymes for glycolysis (n = 3). RNA expression levels were normalized to ACTB in the same sample. Error bars indicate standard deviations. (D) Changes in half-lives under chronic hypoxic conditions. Light and dark gray bars indicate calculated half-lives under normoxic and hypoxic conditions, respectively. (E) Western blotting of indicated proteins in HCT116 cells under normoxic or hypoxic conditions (left). For serial dilutions, #1 of normoxia sample with indicated dilutions were used for quantification of each protein. Band intensities were quantified (right) (n = 3). The y-axis indicates relative protein abundance normalized to intensity of β-actin in the same sample. Light and dark gray bars indicate relative protein abundances in normoxic and hypoxic conditions, respectively. (F) Relative intracellular lactate abundance in normoxic and hypoxic samples (n = 3). Error bars indicate standard deviation. *p < 0.05, **p < 0.01.

Identification of candidate RBPs involved in chronic hypoxia by eRIC-MS

We investigated how the degradation of glycolytic mRNAs was regulated under chronic hypoxia by focusing on RBPs, as proteins involved in the regulation of RNA fate, including degradation^58,59. We examined changes in mRNA binding to individual RBPs under chronic hypoxia using eRIC-MS, in which poly(A)-tailed mRNAs were collected and the proteins bound to these mRNAs were purified and identified by MS (Fig. 4A, see “Materials and methods”). Briefly, we cultured HCT116 cell for 36 h in normoxic or hypoxic conditions, followed by treatment with ultraviolet rays (UV) to crosslink RNA-RBP complexes. We then isolated the poly(A)-tailed RNAs using oligo(dT) and analyzed them using BioAnalyzer. Ribosomal RNAs were removed and poly(A)-tailed RNAs were highly enriched (Fig S4A). RNase treatment eluted RBPs binding to the poly(A)-tailed RNAs from oligo(dT). We also obtained RBPs from cells without UV crosslinking as negative controls in normoxic or hypoxic conditions. We performed proteomic analysis using MS with technical duplications for all samples under four treatments: normoxia with UV, normoxia without UV, hypoxia with UV, and hypoxia without UV, with biological triplicates, to obtain profiles of RBPs bound to poly(A)-tailed RNAs (Table S6).

Fig. 4

Landscape of RBP–mRNA binding under normoxic and hypoxic conditions. (A) Scheme of eRIC-MS. HCT cells cultured under normoxic or hypoxic conditions are exposed to UV light to crosslink RNAs and RBPs. Cells are lysed under denaturing conditions, and poly(A)-tailed RNAs crosslinked with their binding proteins are captured using oligo(dT) probes. Proteins bound to poly(A)-tailed RNAs are eluted by RNase, followed by MS analysis. (B) Identification of proteins reliably binding to poly(A)-tailed RNAs. The x- and y-axes in each panel indicate log-fold change and log-FDR, respectively. Blue and orange dots indicate proteins significantly enriched in samples with UV crosslinking compared with proteins in samples without UV (FDR < 0.05) in normoxic and chronic hypoxic conditions, respectively. (C) Representative molecular functions involved in RNA binding that were significantly enriched for proteins binding to poly(A)-tailed RNAs. (D) Distribution of fold change in yield of proteins bound to poly(A)-tailed RNAs from cells under hypoxic conditions compared with yield of those from cells under normoxic conditions. Proteins sorted according to fold changes. Gray and red dots indicate proteins with < 1.5-fold change (no change) and > 1.5-fold change (activated), respectively.

Following the acquisition of the samples in biological triplicates, we rigorously examined the reproducibility of the protein profiles by comparing technical duplicates of the MS data obtained from two independent analyses of the same samples. We found that the protein profiles for two samples (the second sample of normoxia without UV and the first sample of hypoxia with UV) were less reproducible (Fig. S4B), and therefore omitted these samples from subsequent analyses. MS data frequently contain several proteins as noise due to non-specific binding during precipitation and purification. To identify RBPs specifically binding to poly(A)-tailed RNAs, we performed statistical test for the abundance of proteins compared with that in the corresponding negative controls (without UV). Among the obtained RBPs with two or more unique peptides, 349 and 258 were significantly enriched in the normoxic and hypoxic samples with UV crosslinking, respectively (Fig. 4B, see “Materials and methods”). Gene ontology (GO) analysis of the significantly enriched RBPs in the obtained protein samples confirmed that we mainly obtained RBPs from eRIC-MS approach (Fig. 4C).

Based on the eRIC-MS data, we identified RBPs whose binding to poly(A)-tailed RNAs was changed by chronic hypoxia. Comparing the RBP abundances between hypoxic and normoxic conditions indicated that 93 and two RBPs, among the total 351 identified RBPs, were normoxia- and hypoxia-specific, respectively, and 257 RBPs were common to both conditions (Table S7). Among the 257 RBPs identified in normoxic and hypoxic conditions commonly, 43 were activated (increased > 1.5-fold) under chronic hypoxic conditions (Fig. 4D and Table S7).

Identification of FMRP as stability regulators for glycolytic mRNAs

To identify the RBPs responsible for stabilizing glycolytic mRNAs in response to chronic hypoxia, we performed a hypergeometric test integrating databases with the SLAM-seq data (Fig. 5A). Using the Reactome database^60,61,62, which collects information on biological pathways, to identify mRNAs encoding glycolytic enzymes, we identified 72 glycolytic mRNAs (Reactome mRNAs). By surveying the Reactome mRNAs among the 6,140 mRNAs stabilized during chronic hypoxia obtained from SLAM-seq analysis, we identified 40 Reactome mRNAs that were stabilized in hypoxia (stabilized Reactome mRNAs). We then aimed to identify candidate RBPs that potentially stabilize these 40 mRNAs. We initially listed the RBP-target mRNAs using 150 RBPs analyzed in the ENCODE eCLIP-seq study⁶³, which identified candidate target mRNAs regulated by the 150 ENCODE RBPs. The hypergeometric test showed that 75 of these 150 ENCODE RBPs potentially regulate the 40 stabilized Reactome mRNAs (FDR < 0.05) (Fig. 5B).

Fig. 5

Identification of RBPs regulating RNA abundance of rate-limiting enzymes. (A) Scheme of hypergeometric test to identify candidate RBPs stabilizing glycolytic mRNAs under chronic hypoxia. Among 72 glycolytic mRNAs identified from the Reactome database (Reactome mRNAs), 40 were extracted as stabilized under chronic hypoxia based on SLAM-seq data (stabilized Reactome mRNAs). A total of 150 RBPs studied by the ENCODE project were analyzed to identify their target RNAs. Hypergeometric test of 40 stabilized Reactome mRNAs against target mRNAs of 150 ENCODE RBPs was performed to identify candidate RBPs stabilizing glycolytic mRNAs under hypoxia. (B) RBPs for which the 40 stabilized Reactome mRNAs were significantly enriched against the target RNAs compared with whole transcriptome. RBPs with FDR < 0.05 are indicated. Red bars indicated the RBPs identified as activated RBPs in eRIC-MC. (C) Venn diagram of 43 activated RBPs identified based on eRIC-MS (Fig. 4D), and 75 ENCODE RBPs for which stabilized Reactome mRNAs were significantly enriched against the target RNAs identified based on the hypergeometric test (Fig. 5B). (D) Differential expression of mRNAs encoding rate-limiting enzymes for glycolysis with depletion of FMRP (siFMRP), in normoxia and hypoxia conditions (n = 3). The y-axis indicates fold-change of RNA abundance normalized to average RNA abundance with siControl (siCont) in normoxia condition. RNA expression levels were normalized to ACTB in same sample. Error bars indicate standard deviations.

Cross-referencing the 75 ENCODE RBPs extracted from the hypergeometric test (Fig. 5B) and 43 activated RBPs identified by eRIC-MS (Fig. 4D) implicated two RBPs, IGF2BP2 and FMRP, potentially regulating the stability of glycolytic mRNAs such as HK1, PFKL, and PKM (Fig. 5C, and Table S8). We examined the effects of the two activated RBPs on expression of the glycolytic mRNAs by depletion assay. As described above, most of the glycolytic mRNAs were increased under chronic hypoxia (Table S2). We therefore hypothesized that the activated RBPs IGF2BP2 and FMRP act as stabilizing factors in chronic hypoxia condition for the glycolytic mRNAs, and examined the effects of depletion of the RBPs on the RNAs encoding the rate-limiting enzymes, as representatives of the glycolytic mRNAs. Expression levels of all the rate-limiting enzymes were significantly increased under hypoxic conditions compared with normoxia conditions; however, the expression levels were not changed under hypoxia after knocking-down of FMRP, indicating that knock-down of FMRP canceled the increase in rate-limiting enzymes under chronic hypoxia conditions (Fig. 5D and S5A). In contrast, knocking-down of IGF2BP2 did not cancel the increase in rate-limiting enzymes under chronic hypoxia (Fig S5B and S5C). These results indicated that FMRP, but not IGF2BP2, was involved in the stabilization of mRNAs encoding rate-limiting enzymes for glycolysis under chronic hypoxia conditions.

Finally, we examined the mechanism by which FMRP stabilizes mRNAs encoding rate-limiting enzymes in chronic hypoxia. The eRIC-MS results suggested that FMRP behaved as a stabilizing factor, and FMRP binding to mRNAs was increased under chronic hypoxia. The increased abundance of RBPs binding to target RNAs may be caused by upregulating protein levels of the RBPs and/or by enhancing the binding activity of the RBPs. We found no change in the abundance of FMRP protein under chronic hypoxia (Fig S5D), suggesting that FMRP binding to mRNAs encoding rate-limiting enzymes for glycolysis was regulated via a post-translational mechanism, such as phosphorylation, localization, or other mechanisms. These results provide novel insights into the mechanism through which FMRP upregulates mRNAs encoding rate-limiting enzymes for glycolysis in HCT116 cells adapted to chronic hypoxia; i.e., the RNA-binding activity of the RNA-stabilizing factor, FMRP, is enhanced in response to chronic hypoxia post-translationally, and binding FMRP stabilizes the mRNAs to upregulate the expression levels of HK1, PFKL, and PKM (Fig. 6).

Fig. 6

Schematic diagram of enhancement of glycolysis under chronic hypoxia. Schematic diagram indicating enhancement of glycolysis in chronic hypoxia. Upon chronic hypoxia, FMRP binds to HK1, PFKL and PKM mRNAs to stabilize them, resulting in enhanced glycolysis to adapt to hypoxia.

Discussion

Differential gene expression, accompanied by responses to extracellular stimulation and physiological processes such as development, is generally considered to be regulated via alterations in gene transcription; however, post-transcriptional regulation, including RNA degradation, has recently been shown to play important roles in the regulation of gene expression^64,65,66. In this study, we developed a procedure to quantify the respective contributions of RNA synthesis (${\rho }_{s}$) and RNA degradation (${\rho }_{d}$) to differential gene expression simultaneously. We identified 1003 and 102 genes in HCT116 cells whose differential expression levels were mainly regulated via RNA synthesis and degradation, respectively, under chronic hypoxia. Regarding the genes mainly regulated via RNA degradation, functional enrichment analysis revealed that these genes were significantly enriched in glycolysis, and most of these mRNAs were stabilized; i.e., expression levels of glycolytic RNAs were regulated by mRNA stabilization during adaptation to chronic hypoxia. This represents a reasonable strategy for enhancing glycolysis while avoiding further consumption of ATP. Furthermore, integration of eRIC-MS to identify RBPs binding mRNAs under specific conditions, with statistical analysis using public databases followed by depletion assays, indicated that FMRP is an important factor controlling the increase in rate-limiting enzymes for glycolysis under chronic hypoxia via regulation of RNA stability.

The current results suggest that glycolysis is enhanced during chronic hypoxia via regulation of mRNA stability by RBP: the RNA stabilizing factor FMRP bind to the mRNAs encoding rate-limiting enzymes for glycolysis in response to chronic hypoxia, to protect the RNA from degradation (Fig. 6). FMRP is an RBP related to the incurable neurological disease, fragile X syndrome, and is considered to be responsible for neuronal development and synaptic plasticity via the regulation of mRNA degradation and alternative splicing^67,68,69. We recently reported that FMRP regulates neuronal differentiation by repressing the nonsense mediated mRNA decay (NMD) pathway⁷⁰ and sequestering mRNA from translation and deadenylation⁷¹. FMRP shapes the proteome and transcriptome under several pathological hypoxic conditions, such as resistance of leukemia cells in a hypoxic microenvironment⁷², acute kidney injury caused by hypoxia-reoxygenation⁷³, and encephalopathy of prematurity following perinatal hypoxia–ischemia⁷⁴. Various mechanisms, including remodeling of the translational process, have been proposed to modulate the proteome and transcriptome, but the details remain unclear. The current study showed that FMRP regulated RNA stability in chronic hypoxia, suggesting that FMRP may act as a factor coupling RNA stability and translation in chronic hypoxic responses. Note that our proposal is limited by the fact that the current experiments has not provided direct evidence as to whether FMRP directly binds to stabilize the mRNAs encoding the rate-limiting enzymes of glycolysis under chronic hypoxia.

Recent developments in measurement techniques have enabled us to acquire information on RNA kinetics, including RNA synthesis and degradation^{42,43,44,45,46}. Nevertheless, it is independent matters that the RNA kinetics are altered, and that changes in RNA kinetics affect gene expression. For instance, mutation-dependent rapid RNA degradation leads to up-regulated transcription of related genes, including the mutated gene itself. This compensation mechanism allows destabilization of RNA to increase transcription to maintain robust gene expression levels⁷⁵. This report demonstrates that transcription and RNA degradation are highly linked. The current approach for determining the quantity of respective alterations in RNA synthesis and degradation on differential gene expression (${\rho }_{s}$ and ${\rho }_{d}$) enabled us to clarify the effects of RNA kinetics on gene expression quantitatively and comprehensively. Although this “regulatory analysis approach” has been introduced in metabolic studies⁷⁶, to the best of our knowledge, this is the first time it has been applied to analyze gene expression. Differential gene expression in response to hypoxia is generally considered to be controlled by transcriptional regulation by HIFs; however, the current approach revealed that the expression of glycolytic mRNAs during chronic hypoxia is regulated via changes in RNA stability by RBPs, such as FMRP. RNA kinetics regulates various pathological and physiological cell functions, including responses to extracellular stress or signals and cell differentiation or developmental processes. The current approach will provide substantial insights into how the regulation of RNA kinetics contributes to cellular behavior in relation to pathological and physiological phenomena.

In conclusion, we developed a procedure to quantify the respective contributions of RNA synthesis and degradation to differential gene expression, and showed that enhancement of glycolysis in response to chronic hypoxia is regulated via alterations in RNA stability. Moreover, we found that FMRP is involved in increases of mRNAs encoding the rate-limiting enzymes for glycolysis. Regulation of glycolysis plays an important role in the adaptation to hypoxia^77,78,79. These findings provide substantial insights into cellular regulatory mechanisms under chronic hypoxia, and may thus contribute to the development of cancer treatment strategies and medicines.

Materials and methods

Human colorectal carcinoma derived HCT116 cell line

The human colorectal carcinoma derived HCT116 cell line used in this study were purchased from ATCC, and were passaged and stored in the laboratory. In this study, we used only cells with less than 20 passages for the experiments.

Cell culture and hypoxia treatment

HCT116 cells were cultured with Dulbecco’s modified Eagle’s medium (DMEM) (Wako, Cat #044-29765), supplying with 10% fetal bovine serum (FBS) (Life Technologies, Cat #F7524). FBS was heat-inactivated at 56 °C for 30 min. HCT116 cells were cultured in a humidified incubator (Thermo Fisher Scientific, MODEL #370, REL #1, S/N #310370-4133) with 5% CO₂ at 37 °C. For hypoxia treatment, the cells were precultured for 24 h, followed by culturing in hypoxia chamber (RUSKINN, UM-025, Version #2.0-_CSC2.01) with 1% O₂, 5% CO₂, and 94% N₂ at 37 °C for specified time.

Collection of RNA and protein samples

For collection of RNAs samples, the cells were washed twice with phosphate-buffered saline (PBS) and then total RNA was isolated using RNAiso Plus (Takara, #9109) according to the manufacturer’s protocol. Where appropriate, genomic DNA included in RNA samples are removed by Recombinant DNase I treatment (Recombinant DNase I (Takara, #2270), DNase I Buffer (Takara, #2270), Recombinant RNase Inhibitor (Takara, #2313), 37 °C, 1 h), and RNAs were purified using RNAiso Plus again. For collection of protein samples, intracellular proteins were collected using 2 × SDS sample buffer (20% glycerol, 100 mM Tris–HCl (pH 6.8), 40 mg/mL SDS, 0.01% Bromophenol Blue, 12% 2-mercaptoethanol), followed by fragmentation of genomic DNAs using sonication (15 s, three times) and heat denaturation (95 °C, 5 min).

Reverse transcription-quantitative polymerase chain reaction (RT-qPCR)

We reverse-transcribed appropriate amount of total RNA purified from cell lysate using 5 × PrimeScript RT Master Mix (Takara, #9109) according to the manufacturer’s protocol, and then quantified expression level of individual genes by PCR using SYBR Premix Ex TaqII (Perfect Real Time) (Takara, RR041) with the cDNA as templates. For the PCR, we used Thermal Cycler Dice Real Time system II (Takara, TP900) with 2 step PCR mode (95 °C for 5 s, and 60 °C for 30 s, 40 cycles). The Expression levels were quantified with ΔΔCt methods against corresponding control condition and internal control. Primer sequences used for the PCR are indicated in Table 1.

Table 1 Primer sequences for quantitative polymerase chain reaction.

SDS polyacrylamide gel electrophoresis (PAGE), SYPRO Ruby staining, and western plotting (WB)

Proteins samples collected from the cells were separated by SDS-PAGE with Mini-protean TGX Gels (BIO-RAD, Cat#4561086) or hand-made gels consisted of 3% stacking gel (for one large gel: ultrapure water 3.0 mL, 40% acrylamide bis mixed 0.34 mL, 4 × wide range buffer 1.13 mL, 5% APS 90 μL, TEMED 9.0 μL) and 8% of separating gel (for one large gel: ultrapure water 5.4 mL, 40% acrylamide bis mixed 2.0 mL, 4 × wide range buffer 2.5 mL, 5% APS 100 μL, TEMED 6.0 μL), subjected to SYPRO Ruby (Invitrogen, S12000) staining or WB. For SYPRO Ruby staining, the gel was incubated twice in 25 mL of fix solution (12.5 mL 100% methanol, 1.75 mL acetic acid in nuclease-free water) on shaker at RT for 15 min. Then we stained the gel with 25 mL of SYPRO Ruby in microwave for 15 s, and incubated on shaker at RT for 3 min (avoid light); continued to heat the gel in microwave for 10 s and incubated on shaker at RT for 3 min (avoid light); after that, heat the gel in microwave for another 10 s and incubate on shake at RT for 30 min (avoid light). The SYPRO Ruby stained gel was washed with 25 mL wash solution (2.5 mL 100% methanol, 1.75 mL acetic acid in nuclease-free water) on shaker at RT for 30 min, and then stained proteins were detected with LAS 4000 mini (Fujifilm). For WB, the separated proteins are transferred on Immobilon-P PVDF membrane (Millipore, MA) with Trans-Blot SD Semi-Dry Transfer CeLL (Bio-Rad, #1703940) (0.05A, 45 min or 120 min). The membranes were incubated with antibodies diluted to corresponding dilutions (Table 2) at room temperature. Corresponding secondary antibodies are applied and then intensities of bands were detected with LAS 4000 mini.

Table 2 Properties and dilution of antibodies.

RNA sequencing (RNA-seq) analysis

We sequenced 100 ng of total RNA samples purified from cell lysate using Nova-seq 6000 (Illumine). The sequencings were requested to Macrogen Japan (https://www.macrogen-japan.co.jp/). Briefly, poly(A)-tailed RNAs were enriched using oligo(dT)-conjugate beads, and qualities of the enriched poly(A)-tailed RNAs were assessed using Bioanalyzer (Agilent). The sequence libraries were prepared with TruSeq stranded mRNA Sample Prep Kit (Illumine, RS-122-2101), and then standard Illumina protocols were used to generate 150-bp paired end read libraries that were sequenced on the Nova-seq 6000 platform. Based on the RNA-seq data, we estimated expression level of individual genes. We used HISAT2 (ver. 2.1.0)⁸⁰ with hg38 genomic sequences as references to align the sequenced fragments. The gene expression profiles were quantified using StringTie (ver. 1.3.4d)^81,82. Both alignment and quantification of gene expression profiles were performed with default parameters of HISAT2 and StringTie tools.

Principal component analysis (PCA)

We performed PCA on the gene expression profiles estimated from the RNA-seq data and eRIC-MS data to identify latent variables in the data sets. PCA was performed using prcomp function in R with default parameters. Each dataset was standardized to make the mean and variance constant.

RNA digestion and HPLC measurement

We performed HPLC measurement according to a previous report⁸³. Total RNAs purified from cell lysate were treated with DNase, followed by digestion to nucleotides (0.1 mM DL-dithiothreitol solution (DTT) (Sigma-Aldrich, #3483-12-3), 13.8 mM MgCl2 (Wako, #136-03995), 34.6 mM Tris–HCl (pH 7.5) (Invitrogen, #15567027), 1.6U Alkaline Phosphatase (E. coli C75) (BAP) (Takara, #2120), 0.2U Phosphodiesterase I (Worthington, LS003926), 37 °C for 16 h). After purification, peaks of the nucleotides were determined using Prominence HPLC system (Shimadzu) with HPLC buffer A (3% acetonitrile (Wako, #015-08633), 0.1 M TEAA (Wako, #202-02646)) and HPLC buffer B (900 mL acetonitrile (Wako, #015-08633), and 100 mL ultrapure water). Concentration of the nucleotides and nucleotide analogs were quantified based on area of the peaks.

Alkylation of 4sUracil

We alkylated the 20 μg of 4sUracil by reacting with 10 mM of IAA under optimal conditions (50% DMSO (SIGMA #D-8418), 10 mM iodoacetamide (Wako #095-02151), 50 mM sodium phosphate buffer pH8.0 (1 M NaH₂PO₄ (Nacalai Tesque #317-18) 4.66 mL, 1 M Na₂HPO₄ (Wako #196-02835) 340 μL, up to 10 mL with UltraPure Distilled Water (Invitrogen #10977-015)), for 15 min at 50 °C). The reaction was halted by adding 100 mM of DTT, and the spectrum of absorbance was measured using e-Spect (Malcom). The absorbance at 400 nm of wavelength was measured as the reference wavelength.

SLAM-seq

RNA labeling and IAA treatment

We seeded HCT116 cells with 7.5 × 10⁴ cells/mL of concentration on 12 well plate (Thermo Scientific #150628). Following 24 h of preculture, we cultured the cells under normoxic and hypoxic condition for 36 h. We added final concentration of 100 μM 4sU into the culture medium to label newly synthesized RNAs. The cells were collected after 0, 1, 2, 4, 8, 12 h after the 4sU addition, and the total RNAs were purified. We alkylated the 20 μg of total RNA by reacting with 10 mM of IAA under optimal conditions (50% DMSO (50% DMSO (SIGMA #D-8418), 10 mM iodoacetamide (Wako #095-02151), 50 mM sodium phosphate buffer pH8.0 (1 M NaH₂PO₄ (Nacalai Tesque #317-18) 4.66 mL, 1 M Na₂HPO₄ (Wako #196-02835) 340 μL, up to 10 mL with UltraPure Distilled Water (Invitrogen #10977-015)), for 15 min at 50 °C). Quality of the total RNAs were assessed after ethanol precipitation. We requested DNA Tech (https://dnatech.genomecenter.ucdavis.edu/) to provide the total RNA to QuantSeq analysis. The QuantSeq analysis were performed twice for each sample.

Detection of T-to-C conversion by SLAMDUNK tool

We quantified the expression level and newly synthesized RNA level based on the QuantSeq data of each time point using SLAMDUNK tool (ver. 0.3.3)⁴², a pipeline for analysis of SLAM-seq data. Since QuantSeq sequences 3’ end of RNA in poly(A)-dependent manner, we aligned the QuantSeq data to comprehensive 3’ untranslated regions (3’UTRs) sequences generated based on human genome sequences and annotation (hg38) obtained from Ensemble database (release 92)⁸⁴. We performed the alignment with default parameters of the SLAMDUNK tool. Briefly, we trimmed twelve bases from the 5′ end as adaptor-clipped reads, and then removed four and more subsequent adenines from the 3′ end as remaining poly(A)-tail. VarScan (ver. 2.4.1)⁸⁵ included in the SLAMDUNK tool regards a mismatch as a SNP when it has 0.8 and more of variant fraction and tenfold and more coverage cutoff. Through these filters, we counted total number of reads and the number of those including T-to-C conversions aligned on the 3′ UTRs of individual genes. Since 4sU pairs with guanine (G) during reverse transcription instead of adenine (A), the 4sU-labeled reads were identified as those including T-to-C conversions. Since QuantSeq generates one read from one RNA, the number of reads including and not including T-to-C mutations correspond to the numbers of 4sU-labeled and unlabeled RNAs, respectively. Therefore, we counted the numbers of reads and those including T-to-C conversion corresponded to each gene, and normalized as count per million (CPM).

Identification of RNAs in a steady state

To remove the cells whose expression is fluctuated by biological or mechanical effect during 4sU labeling, we extracted the genes in a steady state with the procedure we developed previously⁴⁷. Briefly, since the expression of a gene in a steady state changes dependent on only white noise, the sum of angles (SoA) formed by the lines connecting each time point is relatively small, whereas the SoA values in a differentially expressed gene (DEG) with a constant trend is relatively large. We calculated the SoA values from the angles formed by the lines connecting certain time points and neighboring ones, within a time series of expression level for each gene. Then, we calculated empirical p-values of the SoA by comparing those when all time points are randomly rearranged. We calculated FDR from the empirical p-values using Storey’s procedure⁸⁶. The genes whose FDR values are less than 0.001 were identified as genes in a steady state.

Identification of DEGs using paired t-test

We identified DEGs based on CPM of individual RNAs inferred from the SLAM-seq data. To avoid effect of 4sU labeling of gene expression, we tested differences in gene expression levels in cells in normoxia and hypoxia with paired t-test for each time point. Based on the p-values, we calculated FDR using Storey procedure⁸⁶. Among the genes expressed in all time points, we identified the genes whose FDR are less than 0.01 as DEGs.

Inference of RNA synthesis and degradation rates

According to Eqs. 1 and 2, expression level of RNA is determined as ratio of RNA transcription rate, ${k}_{s}$, and degradation rate, ${k}_{d}$, and increase curves of 4sU-labeled RNAs depend on ${k}_{d}$. Therefore, fitting of the time series of 4sU-labeled RNAs on Eq. 1 enables us to infer ${k}_{s}$ and ${k}_{d}$ values. For RNAs derived from genes in steady states, we obtained the ${k}_{s}$ and ${k}_{d}$ values by fitting of the time series of 4sU-labeled RNAs on Eq. 1 according to previous report⁴⁷, in genome-wide manner. Briefly, we performed fitting by combining an evolutionary algorithm (genetic algorithm) and hill climbing (L-BFGS-B algorithm), and evaluating with the least squares method in Python 2.7. The genetic algorithm was implemented using the DEAP library⁸⁷ with a generation number of 200, population number of 50, crossover probability of 0.5, and mutation probability of 0.2. The L-BFGS-B algorithm was implemented using the minimize module in the SciPy package, in which the parameters estimated by the genetic algorithm are given as initial parameters. The fitness in each gene was evaluated as the correlation of actual newly synthesized RNA levels with estimated values. The probability of the null hypothesis that a population correlation coefficient is equivalent to zero was calculated for each gene using the OLS module in the StatsModels package⁸⁸, and the ${k}_{s}$ and ${k}_{d}$ values of the genes whose FDR as determined by Storey’s procedure⁸⁶ was less than 10⁻⁵ were extracted.

Functional enrichment analysis

For functional enrichment analysis of focused genes and protein, we performed functional enrichment analysis using DAVID tool (ver. 6.8) (https://david.ncifcrf.gov/). As functional term, we utilized Biological Process (GOTERM_BP_DIRECT), Cellular Component (GOTERM_CC_DIRECT), Molecular Function (GOTERM_MF_DIRECT), UniProt keywords (UP_KEYWORDS), and KEGG pathway (KEGG_PATHWAY). The p-values indicating enrichment were calculated based on modified hypergeometric test^89,90. We selected optimal gene list as the background. The functional terms whose FDR values are less than thresholds were identified as those significantly enriched.

GSEA

The relationship between the biological function of genes and ${\rho }_{d}$ values, contribution of RNA degradation rate on differential expression, were examined using GSEA (ver. 4.0.3)⁵². For each individual gene, 1 was used as the control value and genes were ranked based on the ratio of the ${\rho }_{d}$ value to the control (i.e., the original ${\rho }_{d}$ value). Enrichment score (ES) for each term in the GSEA hallmark was calculated using default parameters, and compared with the distribution of ES values for random set of 10,000 genes to calculate empirical p-values. The empirical p-values were corrected as FDR. The terms whose FDR values are less than 0.05 were identified as those significantly enriched.

Lactate assay

Intracellular lactate was quantified using Lactate Assay Kit-WST (Doujin, #343-09281). The HCT cells were cultured in corresponding conditions with six of replicates and the medium were removed, and cell lysates were prepared with 0.1% Triton solution. We then added 20 μL of lactate standard solution and 80 μL of working solution to 20 μL of the cell lysate, followed by incubation on 37 °C for 30 min. The absorbance at 450 nm of wavelength was measured using an absorbance microplate reader (Tecan).

eRIC-MS

Coupling of the capture probe to Dynabeads

Dynabeads™ MyOne™ Carboxylic Acid (Invitrogen, #65012) was washed with double volumes of 100 mM MES (pH 4.8) and vortexed for 5 to 10 s. We removed the supernatant with a magnet stand and resuspended the Dynabeads in 30 μL of 100 mM MES (pH 4.8). Based on previous reports⁹¹, we utilized HPLC purified capture probe consisted of primary amine and C6 linker, followed by 20 thymidine nucleotides, in which every other nucleotide is a locked nucleic acid (LNA) (Exiqon); /ssH5AmC6-LNA10T10/T(L)TT(L)TT(L)TT(L)TT(L)TT(L)TT(L)TT(L)TT(L)TT(L)T (T(L): LNA thymidine, T: DNA thymidine)⁹². We prepared the LNA oligo(dT) to 97.2 μL/sample with 10.8 μL of 1M MES (pH 4.8) and 27 μL of 500mg/mL N-(3-Dimethylaminopropyl)-N’-ethylcarbodiimide hydrochloride (EDC) (Sigma-Aldrich, #E7750). The LNA oligo(dT) was coupled on Dynabeads by gently rotating mixture of 600 μL of Dynabeads slurry (10 mg/mL) and 97.2 μL of LNA oligo(dT) (200 μM) at room temperature for 3 h. The Dynabeads were then washed with two volumes of 250 mM Tris buffer (pH 8.0) and 0.01% Tween 20 for more the 30 min twice. The LNA oligo(dT) coupled Dynabeads were stored in 0.1% PBS-Tween at 4 °C.

UV crosslink

The medium culturing HCT116 cells on the 15 cm dish was removed and cells were washed with 10 mL cold PBS for twice. To avoid effect of exposure on normoxic environment, the hypoxia treated cells were sealed with hybrid-bag after removing the PBS completely in the hypoxia chamber, and transferred from the chamber. For normoxia treated cells, hybrid-bag was also used in UV crosslink. 2000 mJ/cm² of UV was used in crosslink and irradiation was omitted in controls without UV crosslink (without UV). Immediately before sample collection, we added 1.0 mM DTT (Sigma-Aldrich, #43816) and 0.5 U/μL of recombinant RNase inhibitor (RRI) (Takara, #2313B) into prepared lysis buffer (10 mM Tris–HCl (pH 7.5) (Invitrogen, #15567027), 10 mM NaCl, 0.02% (w/v) Digitonin (Fujifilm, 043-21371), 1 mM EDTA pH 8.0 (Invitrogen, MA)). After irradiation, we opened the sealed hybrid-bag were, and kept the irradiation cell on ice. We then added 2.0 mL of cold hypo lysis buffer and collected cells with scraper (Corning). The cell lysates were mixed for 10 times and transferred into 5 mL centrifuge tubes. Cell lysates from two of 15 cm dishes were collected for one eRIC-MS sample. Above operations were done on ice quickly.

Capture of RNA-RBP conjugations

Cell lysates were mixed by inversion at 4 °C for 10 min. Cell lysates were then centrifuged at 1000g for 5 min at 4 °C to remove the nucleolus, the supernatants were transferred to new tubes, and continued to centrifuge at 15,000g for 5 min at 4 °C to remove the organelles like mitochondria. We transferred the supernatants to new tubes and added 500 mM of lithium chloride (LiCl) (Sigma-Aldrich, #L9650), and inverted tubes completely before adding 0.5% lithium dodecyl sulfate (LiDS) (Sigma-Aldrich, #L9781). The mixture was Incubated at 60 °C for 15 min, and quickly cooled down on ice for 5 min. The mixtures were clarified with centrifuge at 15,000g for 5 min at 4 °C and the supernatants were transferred into new tubes. We stored 40 μL and 50 μL supernatants as input for later RNA and protein analysis, respectively, and DTT was added to the remaining supernatants for a final concentration to be 5 mM. The remaining supernatants were mixed with the LNA oligo(dT) coupled Dynabeads washed with five volumes of the hypo lysis buffer for 3 times before usage, and gently rotated at 40 °C for 1 h to capture RNA–protein complexes.

Elution of RNA-RBP complex

We collected the Dynabeads capturing the RNA–protein complexes with a magnetic stand, and transferred supernatants to new tubes for later analysis. Beads were subjected to successive rounds of washes with wash buffer 1 (20 mM Tris–HCl (pH 7.5), 500 mM LiCl, 1 mM EDTA, 5 mM DTT, and 0.1% (w/v) LiDS), wash buffer 2 (20 mM Tris–HCl (pH 7.5), 500 mM LiCl, 1 mM EDTA, 5 mM DTT, and 0.02% (v/v) NP40), and wash buffer 3 (20 mM Tris–HCl (pH 7.5), 200 mM LiCl, 1 mM EDTA, 5 mM DTT, and 0.02% (v/v) NP40) for twice with gentle rotation at 40 °C for each 5 min. Pre-elution was performed in 440 μL nuclease-free water at 40 °C for 5 min. Afterwards, the beads suspension was divided into two groups: 400 μL of RNase-mediated elution for protein analysis; and 40 μL of heat-mediated elution for RNA/DNA analyses. For RNase-mediated elution, beads were resuspended in 400 μL of RNase buffer (0.25 μL of RNase mixture in 400 μL nuclease-free water; RNase mixture: 1 μg/μL RNase A, 40 U/μL RNase T1, 50 mm Tris (pH 7.0), 50 mm NaCl (Invitrogen, #AM9759), and 50% glycerol), and incubated at 37 °C for 30 min. For heat-mediated elution, beads were resuspended in 40 μL nuclease-free water, and incubated at 95 °C for 5 min. We took the supernatants immediately after beads were collected with a magnetic stand. To confirm the effect of above elution, heat-mediated second elution was conducted with above two groups. After that, stored all samples at -80 °C.

MS measurement

Following a confirmation that ribosomal RNAs were removed from the heat-mediate elution using a BioAnalyzer, and that no abnormalities such as contaminants occurs from a part of the RNase-mediate elution using SDS-page and SYPRO Ruby staining, the SDS in the RNase-mediate elution samples was removed using the methanol–chloroform protein precipitation method. Briefly, four volumes of methanol, one volume of chloroform, and three volumes of water were added to the eluted sample and mixed thoroughly. The samples were centrifuged at 15,000 rpm for 10 min, and the water phase was removed carefully, and then four volumes of methanol was added to the samples, and the samples were centrifuged at 15,000 rpm for 10 min. After that, the supernatant was removed, and the pellet was washed with 100% ice-cold acetone once. The precipitated protein was re-dissolved in guanidine hydrochloride and reduced with Tris (2-carboxyethyl) phosphine hydrochloride, alkylated with iodoacetamide, followed by digestion with lysyl endopeptidase and trypsin. The digested peptide mixture was applied to a Mightysil-PR-18 (Kanto Chemical) frit-less column (45 × 0.150 mm ID), and separated using a 0–40% gradient of acetonitrile containing 0.1% formic acid for 80 min at a flow rate of 100 nL/min, and the eluted peptides were sprayed into a mass spectrometer (Triple TOF 5600 + ; AB Sciex) directly. MS and MS/MS spectra were obtained using the information-dependent mode. Up to 25 precursor ions above an intensity threshold of 50 counts/sec were selected for MS/MS analyses from each survey scan. All MS/MS spectra were searched against protein sequences of the RefSeq (NCBI) human protein database (RDB) using Proteome discoverer 2.2, and decoy sequences were then selected with FDR < 1%.

Identification of RBPs

RBPs reliably binding to poly(A)-tailed RNAs were identified by comparing eRIC-MS data from same condition with irradiation (with UV) and without irradiation (without UV). The protein sample is acquired as biological triplicates in each condition, and then technical duplicates of the MS data analyzed from the same samples are compared to discard improperly measured outlying samples. Among the peptides detected in both of with and without UV in appropriately measured samples, signal intensities of those with two and more unique peptides number, using one-side Mann–Whitney U test. The p-values are corrected as FDR with Storey’s procedure⁸⁶. To compensate lower detection power of Mann–Whitney U test, a non-parametric test, we adapted 5% as threshold of the FDR. RBPs including one and more peptides with FDRs less than the threshold were considered as those reliably binding to poly(A)-tailed RNAs. Note that, among the twelve samples (triplication for four conditions), one sample from group of “hypoxia with UV” and the other from group of “normoxia without UV” were excluded from later analysis because of lower reproducibility. Abundance of the RBPs were estimated as the value of signal intensity with the total normalized to 1,000,000.

Hypergeometric test to extract RBPs targeting stabilized glycolytic mRNAs

The glycolytic mRNAs were identified based on Reactome^60,61,62, a database for biological pathways (https://reactome.org/download/current/Ensembl2Reactome_All_Levels.txt). The data obtained from Reactome database contains Ensembl gene IDs and relating biological pathways. We extracted entries corresponded to Ensembl gene IDs as whole transcriptome (background) of the hypergeometric test. Among these entries, we identified the entries corresponded to “Glycolysis” as glycolytic mRNAs (named as “Reactome mRNAs”), and extracted those whose ${k}_{d}$ values were decreased in chronic hypoxia condition compared with normoxia as stabilized Reactome mRNAs. For each RBP included in ENCODE, a database of functional elements of human genome⁶³, we corresponded target RNAs and calculated p-value for enrichment of the stabilized Reactome mRNAs included in the targets against the background with hypergeometric test. Based on the p-values, we calculated FDR using Storey procedure⁸⁶.

Depletion assay

HCT116 cells were seeded in 24-well plates at concentration of 2.5 × 10⁴ cells/well with DMEM (Wako, Cat# 044–29,765) supplying with 10% FBS ((Life Technologies, Cat# F7524), and incubated appropriate time in the humidified incubator (Normoxic condition) or in the hypoxia chamber (hypoxic condition). 1 μM of siRNAs (Table 3) were transfected into the HCT116 cells by using Opti-MEM (Gibco, Cat# 31,985–070) and Lipofectamine RNAimax Reagent (Invitrogen, REF# 13,778–500) according to manufacturer’s protocol, followed by 48 h of incubation. The cells were washed twice with PBS and then total RNA was isolated using RNAiso Plus (Takara, #9109) according to the manufacturer’s protocol. Expression level of individual genes were quantified as described above, following the reverse-transcription.

Table 3 siRNAs used in this study.