Glucose binds and activates NSUN2 to promote translation and epidermal differentiation

Abstract

Elevations in intracellular glucose concentrations are essential for epithelial cell differentiation by mechanisms that are not fully understood. Glucose has recently been found to directly bind several proteins to alter their functions to enhance differentiation. Among the newly identified glucose-binding proteins is NSUN2, an RNA-binding protein that we identified as indispensable for epidermal differentiation. Glucose was found to bind conserved sequences within NSUN2, enhancing its binding to S-adenosyl-L-methionine and boosting its enzymatic activity. Additionally, glucose enhanced NSUN2’s proximity to proteins involved in mRNA translation, with NSUN2 modulating global messenger RNA (mRNA) translation, particularly that of key pro-differentiation mRNAs containing m⁵C modifications, such as GRHL3. Glucose thus engages diverse molecular mechanisms beyond its energetic roles to facilitate cellular differentiation processes.

Graphical Abstract

Graphical Abstract.

Introduction

Multi-layered stratified epithelia line various organs in cutaneous, digestive, respiratory, ocular and genitourinary tissues (1). In these tissues, undifferentiated cells in the basal epithelial layer bound to an underlying basement membrane cease dividing and migrate outward to undergo terminal differentiation at the epithelial surface (2). The molecular mechanisms governing this process, which is disrupted in numerous human neoplastic and inflammatory disorders, are still being unraveled. The epidermis is the stratified epithelium of the skin that serves as a barrier against desiccation and other environmental factors (3–5). Post-mitotic suprabasal keratinocytes undergo differentiation to produce specialized proteins and lipids, contributing to the formation of the stratum corneum barrier (6). Disruption in epidermal homeostasis and barrier function characterizes prevalent polygenic skin disorders, as well as monogenic diseases resulting from inherited mutations (7–9). Numerous regulators play essential roles in this process, including specific transcription factors (TFs) (10–12) and RNA-binding proteins (RBPs) (13,14), which modulate the expression of ∼3500 genes undergoing dynamic changes during epidermal differentiation (15). However, the full extent of essential regulators involved in the differentiation of the epidermis and other stratified epithelia remains unknown.

The abundant monosaccharide glucose, traditionally recognized as a universal currency for energy storage, transport and utilization, has recently been implicated in essential non-energetic roles, particularly in differentiation processes (16). Recent studies have shed light on these new roles for glucose, sparking interest in unraveling the molecular mechanisms underlying its actions. Intriguingly, intracellular glucose elevation has been observed in various stratified epithelia (16). In skin, glucose elevation plays a crucial role in differentiation processes by modulating the oligomerization and functions of specific proteins, such as the IRF6 TF and the DDX21 DEAD-box RNA helicase (13,16). Glucose binding facilitates IRF6 oligomerization and disrupts DDX21 dimers, influencing their interactions with small molecules and nucleotides. For instance, glucose abolishes the ATP binding of DDX21 and promotes the DNA binding of IRF6. Despite its small size, glucose interacts with specific amino acids within these proteins, exerting control over their affinities and oligomerization. However, it remains unclear whether the actions of glucose in this context are a unique phenomenon or part of a broader paradigm whereby glucose impacts multiple related proteins involved in cellular differentiation. Further investigation is needed to elucidate the full extent of glucose’s regulatory role in modulating protein functions during differentiation.

NSUN2, an RNA cytosine C5-methyltransferase, plays a critical role in various cellular processes by methylating cytosine to 5-methylcytosine (m⁵C) in different RNA species, including transferRNA (tRNA) (17) and messenger RNA (mRNA) (18). Its functions span diverse areas such as epidermal stem cell differentiation (19), testis differentiation (20) and cancer metastasis (21). Specifically, in epidermal stem cell differentiation, NSUN2 is involved in maintaining the balance between stem cell self-renewal and differentiation, particularly in mouse hair follicle stem cells (19). Interestingly, glucose has been identified as a direct binder of NSUN2, driving its oligomerization and promoting its enzymatic activity in cancer cells (22). The revelation of NSUN2 as a glucose sensor, combined with both glucose and NSUN2 functioning in epidermal stem cell differentiation, suggests that the glucose-NSUN2 interaction could impact human skin differentiation. Investigating this could provide valuable insights into the molecular mechanisms underlying epidermal differentiation and shed light on the broader regulatory roles of glucose in cellular processes.

In this study, we elucidated the vital role of glucose binding to NSUN2 in human epidermal differentiation and identified mechanistic features of how NSUN2 regulates this process. Our findings revealed that glucose directly binds to NSUN2 at N-terminal sequences, specifically at NSUN2 residues K28/R29. This glucose binding induces conformational changes in NSUN2, particularly in its S-adenosyl-L-methionine (SAM) binding domain, thereby enhancing the protein’s affinity for SAM and activating its enzymatic activity. Moreover, glucose binding also promotes the oligomerization of NSUN2, leading to its activation (22). With increased activity, NSUN2 augments global m⁵C RNA methylation levels (22) and exhibits enhanced binding to pro-differentiation mRNAs, such as GRHL3 mRNA. This enhanced binding facilitates the m⁵C modification of these target RNAs. Additionally, glucose facilitates NSUN2 binding to protein translation complexes, thereby promoting the translation of pro-differentiation RNAs, including GRHL3. Overall, our study highlights how glucose-mediated modulation of protein function engages diverse molecular mechanisms to enable cellular differentiation, with a specific focus on the role of NSUN2 in this process. These findings expand our understanding of the regulatory roles of glucose in cellular differentiation.

Materials and methods

Cells and organotypic culture

HEK293T (Takara Bio, #632180) cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM) media (Gibco, #11995–065) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin at 37°C with 5% CO₂. Primary human keratinocytes were isolated from fresh surgically discarded newborn foreskin, and keratinocyte progenitor cells were cultured in 50% (v:v) complete Keratinocyte-Serum-Free Media (SFM) (Gibco, #17005–042) and 50% (v:v) Medium 154 (Gibco, #M-154–500). Human skin keratinocyte cells were induced to differentiate by the addition of 1.2 mM calcium for 3 days at full confluence in the 50/50 media if not notified specifically. Cells with low glucose and the corresponding normal glucose condition were maintained in glucose-free DMEM (Gibco, #11966025) supplied with SFM supplement (rEGF: Gibco, #10450–013; BPE: 13028–014), 1X sodium pyruvate (Gibco, 11360070) and a final concentration of 0.5 g/L (‘low’ glucose, Gibco, A2494001) or 4.5 g/L glucose (‘normal’ glucose) with adding 1.2 mM calcium during differentiation for 3 days. All the ‘low’ and ‘normal’ conditions indicated referred to day 3 differentiated cells if not specified. All the keratinocyte progenitor cells were maintained in 50/50 media if not specified. Organotypic regeneration of human epidermal tissues was performed as previously described (23) using Keratinocyte Growth Medium (KGM). Day 7 epidermal tissues were collected. Human keratinocytes isolated from newborn foreskin were used if not notified specifically. For skin primary cells, biological replicates were performed in all cases from at least two independent, unrelated donors. All cells were tested regularly for Mycoplasma contamination; none tested positive throughout the studies.

Gene transfer and knockdown

To produce viruses, HEK293T cells in 10 cm plates were transfected with 7 μg of each retroviral expression construct, using Lipofectamine 3000 (Invitrogen, L3000015). Viral supernatants were collected 48 h after transfection and concentrated using Lenti-X concentrator (TaKaRa, 631231). For infection, the optimized viral titer was added to primary human keratinocytes and HEK293T cells together with polybrene (5 μg/ml). The next day, fresh media were added, and cells were selected using puromycin (1 μg/ml). Small hairpin RNAs (shRNAs) targeting NSUN2 were designed based on Sigma sequences (https://www.sigmaaldrich.com/US/en/semi-configurators/shrna?activeLink=productSearch) and were cloned into pLKO.1-puro plasmid (Addgene, #10878). The targeting sequences are listed below.

shScramble, 5′-TCCTAAGGTTAAGTCGCCCTCG-3′;

shNSUN2-1, 5′-GAGCGATGCCTTAGGATATTA-3′;

shNSUN2-2, 5′-TGCAGTGTCCCATCGTCTTAT-3′;

shNSUN2-3, 5′-TGAGAAGATGAAGGTTATTAA-3′.

Recombinant protein purification

All recombinant target proteins used in this study were produced in HEK293T cells. Briefly, 20 μg of pLEX plasmids with tagged protein constructs was transfected per 15 cm plate of ∼80% confluent HEK293T cells, using Lipofectamine 3000 (Invitrogen). Forty-eight hours after transfection, cells were harvested, lysed on ice for 30 min in lysis buffer (50 mM Tris–HCl, pH 7.5, 300 mM NaCl, 1 mM EDTA, 1% Triton X-100, 1X Protease Inhibitor Cocktail [Sigma, P8340]), then sonicated for 3 cycles of 10 s at an amplitude of 10% followed by a 10 s pause. The lysate was centrifuged at 16 000 g for 10 min and quantified. The lysate was added to a saturating volume of anti-FLAG M2 affinity gel (Millipore Sigma, A2220)—a volume determined by a small scale, pilot purification for each protein. The IP was performed overnight at 4°C and the next morning was washed three times with wash buffer (50 mM Tris–HCl, pH 7.5, 3 mM EDTA, 0.5% NP-40, 500 mM NaCl, 10% glycerol, 0.1 mM Dithiothreitol (DTT)). Before elution, M2 beads were primed with two washes of Elution Buffer (1X PBS), then purified protein was eluted using 0.5 mg/mL 3X FLAG peptide in Elution Buffer. Elution was performed once at 1.5X bead volume for 1 hour. Eluates were concentrated using 3k molecular weight cutoff Amicon columns (Millipore, UFC500396). Finally, the target protein concentration was determined by running purified protein and a BSA standard curve on a Bis-Tris gel and staining using InstantBlue Coomassie Protein Stain (Abcam, ab119211).

Microscale thermophoresis (MST)

Recombinant protein was labeled using the Monolith His-Tag Labeling Kit RED-tris-NTA (Nanotemper Technologies, MO-L018) following the manufacturer’s protocol at a 1:2 dye to protein ratio, in a reaction with a final concentration of 100 nM labeled protein. Protein concentrations were determined by BSA standard curve quantification. Labeling was assayed via a capillary scan at 60% LED power, and protein was diluted to 600 units using the Nanotemper-provided 1x Phosphate Buffered Saline with Tween 20 (PBST); this resulted in a target protein concentration of ∼50 nM. Labeled protein was mixed with ligand and incubated for 5 min at room temperature before being loaded into Monolith NT.115 Capillaries or NT.115 Premium Capillaries (NanoTemper Technologies, MO-K022 and MO-K025). Microscale thermophoresis (MST) was measured using a Monolith NT.115 instrument (NanoTemper Technologies) at an ambient temperature of 25°C. Instrument parameters were 60% excitation power and Medium MST power. Data from at least three independently pipetted measurements were analyzed (MO.Affinity Analysis software, NanoTemper Technologies) using the fraction bound as indicated in the respective figures. When indicated, labeled protein was pre-incubated with freshly prepared galactose (Sigma, G6404), glucose (Sigma, 49 139), 3-O-methyl-D-glucopyranose (3OMG, Sigma, M4879) or S-(5′-adenosyl)-L-methionine chloride dihydrochloride (Sigma, A7007) at room temperature for 5 min. The sequences for RNA are as follows: tRNA: 5′-GUA GCU CAG UGG UAG AGC GCG UGC UUA GCA UGU ACG AGG UCC CGG GUU CAA UCC CCG GC-3′. GRHL1 and GRHL3 RNAs are produced from in vitro transcription.

Cellular thermal shift assay (CETSA)

Cellular thermal shift assay (CETSA) was performed following published research with minor modifications (24,25). HEK293T cells were transfected with 20 μg FHH-NSUN2 plasmid in 15 cm plates for 8 h and then media was changed to glucose-free DMEM (Gibco, #11966025) supplied with 10% dialyzed FBS and 1X sodium pyruvate (Gibco, 11360070) for overnight incubation. Cells were then harvested and lysed with lysis buffer (1× TBS, 1.5 mM MgCl₂, 0.5% NP-40, 1× protease inhibitor [Roche, 11836170001]) for 15 min at 4°C and cleared by centrifugation at 20 000 × g for 15 min (4°C). The soluble fractions were diluted to 4–5 mg/ml using lysis buffer and allowed to warm to room temperature (RT). Glucose dissolved in PBS was added to 1 mM final concentration, or the same amount of PBS was added as a control, and the samples were allowed to incubate at RT for 20 min. Each sample was divided into aliquots and heated to eight temperatures (4, 40, 44, 48, 52, 56, 60 and 64°C) for 3 min. All samples were centrifuged at 20 000 g for 1 h 30 min (4°C). The soluble fractions were collected for western blot analysis.

Biotin-glucose probe synthesis

Glucose (360 mg), 4-dimethylaminopyridine (11.0 mg) and N-hydroxysuccinimidylbiotin (341 mg) were dissolved in dry dimethylformamide (10 ml) and heated in an oil bath at 45°C while stirring. Additional aliquots of glucose (180 mg) were added every 4–6 h until a NHS-biotin peak was no longer observed by mass spectrometry. The solvent was removed, and the resulting compound was purified by column chromatography on silica using dichloromethane/methanol/water (65:25:4) as eluent. Calculated mass [M + H]: 407.1483; found 407.1478. Proton and carbon NMR spectra were complex due to the equilibrium among the α- and β-anomers of the pyranose and furanose forms of the molecule.

Biotin-glucose pull down

For cell lysate experiments, a 15 cm plate of HEK293T cells were transfected with 30 μg pLEX FHH-NSUN2 plasmid and 90 μg polyethylenimine (PEI). Media were changed after 16 h and cells were collected after 36 h in E1A buffer (50 mM HEPES, pH 7.6, 250 mM NaCl, 0.1% Nonidet P-40, 5 mM EDTA, 1X Protease Inhibitor Cocktail). Lysate was cleared by centrifugation at 16 000 g for 10 min. Protein concentration was measured by Bradford Assay and 600 μg of lysate was used for each experiment (total 300 μl).

Dynabeads™ MyOne™ Streptavidin C1 (Thermo Fisher) were washed 3X with 10X PBS, then incubated with 500 pmoles of biotin (Sigma) or biotin-glucose per 100 μg beads for 30 min at 25°C. Beads were washed 3X with 10X PBS, then incubated with 1 μg of recombinant FHH-NSUN2 protein in either PBS or PBS + 10 mM glucose for 30 min at 25°C rotating. For cell lysate experiments, the washed glucose-conjugated beads were resuspended in E1A buffer, then incubated with 600 μg lysate per 200 μg beads in either PBS or PBS + 10 mM glucose for 30 min at 25°C, rotating. Beads were washed with 10X PBS four times, then eluted with 2X NuPAGE™ LDS Sample Buffer + 4 mM Biotin at 70°C with 1000 rpm shaking for 15 min. Equal volumes were evaluated by western blot staining for HA.

Fluorescence-based binding assay

Dye-conjugated glucose was prepared by reacting 200 μM AZDye 488 Alkyne (Click ChemTools, 1277–1) and 400 μM 2-azido-2-deoxy-D-glucose (Santa Cruz Biotechnology, sc-256068) in 1X PBS for 1 h with 0.1 mM CuSO₄, 0.5 mM tris-hydroxypropyltriazolylmethylamine (THPTA) and 15 mM sodium ascorbate at room temperature. The completeness of the reaction was checked by thin layer chromatography (TLC) using acetonitrile (ACN) buffer. Fluorescence was measured with 100 nM glucose conjugated AZDye 488 and a concentration gradient of FHH-NSUN2 recombinant protein. Fluorescence intensity was measured on a Tecan Infinite M1000 with excitation at 470 nm and emission at 525 nm with 5 nm bandwidth, gain at 50, z-position at 26 000 μm using 384-well plates. Fluorescence intensity was decreased with addition of NSUN2 protein. Total intensities were used to calculate K_d by (i) dividing the fluorescence value in each well by the 0 nM protein well of the same replicate, (ii) multiplying intensity by -1, then (iii) adding the absolute value of the lowest intensity to make the lowest value zero and (iv) fitting a 1:1 binding curve to the data using the python package pybindingcurve (26). A negative control was run against 100 nM unconjugated AZDye 488 Alkyne (Click ChemTools, 1277–1) by adding 5 μM FHH-NSUN2 recombinant protein. No decrease in total intensity was observed for the negative control.

Quantitative RT-PCR expression analysis

For quantitative reverse transcriptase-polymerase chain reaction (RT-PCR) (qRT-PCR), total RNA was extracted using RNeasy Plus (QIAGEN, 74136) and subsequently subjected to reverse transcription using the iScript cDNA Synthesis Kit (Bio-Rad, 1708891). qRT-PCR analysis was performed using the LightCycler 480 II System (Roche) with the SYBR Green Master Mix (Thermo Fishier, K0223). Samples were run in duplicate and normalized to levels of 60S ribosomal protein L32 for each reaction. The following primers were used for qPCR analysis:

Gene	Forward primer	Reverse primer
CDK1	TGCTTATGCAGGATTCCAGGTT	CATGTACTGACCAGGAGGGA
CDKN1A	GGGTCGAAAACGGCGGCAGA	CCTCGCGCTTCCAGGACTGC
DNMT1	AAGCCCGTAGAGTGGGAATG	GCTAGGTGAAGGTTCAGGCTT
FLG	AAAGAGCTGAAGGAACTTCTGG	AACCATATCTGGGTCATCTGG
GRHL1	GCCTACCCACTCCATCAAGA	GAGTCTGGAGTTCGCCTTTG
GRHL3	GGTGTTCATCGGCGTAAACT	CCCAAGCCACAGTCATAGGT
IVL	TGCCTGAGCAAGAATGTGAG	TGCTCTGGGTTTTCTGCTTT
KRT1	GAAGTCTCGAGAAAGGGAGCA	ATGGGTTCTAGTGGAGGTATCTA
KRT10	GCAAATTGAGAGCCTGACTG	CAGTGGACACATTTCGAAGG
KLF4	TCCACAACTTCCAGTCACCC	AGAACAGATGGGGTCTGTGAC
LCE3D	GCTGCTTCCTGAACCAC	GGGAACTCATGCATCAAG
L32	AGGCATTGACAACAGGGTTC	GTTGCACATCAGCAGCACTT
MAF	TATGCCCAGTCCTGCCGCTT	CGCTGCTCGAGCCGTTTTCT
MAFB	GACGCAGCTCATTCAGCAG	CCGGAGTTGGCGAGTTTCT
NSUN2	CCTTTCCAGAGGGATTTGT	TCTCCACTGCAAGGGACAT
ZNF750	AGCTCGCCTGAGTGTGAC	TGCAGACTCTGGCCTGTA

Western blot and immunofluorescence

For immunoblot analysis, proteins were quantified by Bradford assay (Bio-Rad, 5000201). Approximately 10–20 μg of cell lysates were loaded per lane on a Bis-Tris gel and transferred to a nitrocellulose membrane at 4°C. The resulting membrane was blocked with LI-COR blocking buffer (PBS or TBS) at room temperature for 1 h. The membrane was then incubated with primary antibody at 4°C overnight. Membranes were washed with TBS-T and incubated with secondary goat anti-mouse and goat anti-rabbit antibodies (LI-COR Biosciences) at a dilution of 1:4000 for 1 h at room temperature. After incubation with secondary antibodies, membranes were washed with TBS-T and visualized using the Odyssey CLx Infrared Imaging System (LI-COR Biosciences). Quantification was performed using Licor ImageStudioLite software (LI-COR Biosciences).

For immunofluorescence staining, tissue sections (7 μm thick) were fixed using either acetone or methanol. Primary antibodies were incubated at 4°C overnight, and secondary antibodies were incubated at room temperature for 1 h. Nuclear was stained by Duolink® In Situ Mounting Medium with 4′,6-diamidino-2-phenylindole (DAPI) (Sigma, DUO82040-5ML). All images were taken and processed using a Zeiss microscope with or without ApoTome.2.

The antibodies used in this study include anti-NSUN2 (ProteinTech, 20854–1-AP), Rb-anti-HA (CST, #3724), anti-Actin (Sigma, A2228), anti-KRT1 (Covance, PRB-149P), anti-KRT10 (Neomarkers, MS611P), anti-Filaggrin (Abcam, ab17808), anti-RPLP0 (Santa Cruz Biotechnology, sc-293260) and anti-Strept (LI-COR, 926-32230).

RNA-seq

For RNA-seq of NSUN2 knockdown samples, total RNA was extracted using the RNeasy Plus kit (QIAGEN, 74136), and libraries were prepared with QuantSeq 3′ mRNA-Seq V2 Library Prep Kit FWD with UDI (Lexogen, 191.96), following the manufacturer’s protocol. RNA-seq was performed on a NovaSeq 6000, paired-end, 150 bp length, with ∼30 M reads for each sample.

Quantification of 5-methylcytidine (m⁵C) and N⁶-methyladenosine (m⁶A) in RNA

mRNA was purified by using Dynabeads™ mRNA DIRECT™ Purification Kit (Invitrogen, 61012) following the manufacturer’s protocol. RNA samples (100 ng each) were digested with 1 unit of nuclease P1 (Sigma-Aldrich, N8630) in a 25 μl buffer containing 25 mM NaCl and 2.5 mM ZnCl₂ at 37°C for 2 h. To the resulting mixture were then added 1 unit Quick calf intestinal alkaline phosphatase (New England Biolabs, M0525) and 1 M ammonium bicarbonate (NH₄HCO₃). The mixture was incubated at 37°C for another 2 h and dried using a Speed-vac.

A stable isotope-dilution method was used for the accurate quantification of global levels of m⁵C and m⁶A in RNA. To 10 ng of RNA digestion mixture were added 3.77 pmol [¹³C₅]-labeled cytidine ([¹³C₅]-rC) (Cambridge Isotope Laboratories, CLM-3679–0.05), 10.3 fmol [¹³C₅]-m⁵C (synthesized), 2.825 pmol [¹³C₅]-labeled adenosine ([¹³C₅]-rA) (Cambridge Isotope Laboratories, CLM-3678–0.05) and 32.51 fmol [D₃]-m⁶A (LGC, TRC-M275897). Water was added to the mixture to make the total volume 100 μl. Chloroform extraction was subsequently used to remove enzymes from the digestion mixture, and the aqueous layer was dried using a Speed-Vac, and the dried residues were reconstituted in 90% ACN for desalting. After drying, about one-fourth of the nucleoside mixture was injected for Liquid Chromatography Tandem Mass Spectrometry (LC-MS/MS) analysis.

The nLC-MS/MS measurements were conducted on a TSQ-Altis triple-quadrupole mass spectrometer operated in the multiple-reaction monitoring (MRM) mode and coupled with a Dionex Ultimate 3000 HPLC module (Thermo Fisher Scientific, Inc.) with a PepMap Neo Trap-Cartridge (100 Å, 5 μm, 0.3 × 5 mm, Thermo Fisher Scientific) and an in-house prepared analytical column (75 μm × 200 mm) packed with Zorbax SB-C18 (5 μm beads, 100 Å in pore size, Agilent Technologies). The LC separation was conducted by using 0.1% formic acid in H₂O (mobile phase A) and 0.1% formic acid in ACN (mobile phase B). The electrospray voltage was configured to 2.0 kV, while the ion transport tube temperature was regulated at 275°C. Precursor and fragment ion selection widths were established at 0.7 and 0.4 m/z units, respectively, with a collision energy being set to 20 V. For the measurements of m⁵C and m⁶A, the samples were loaded onto the trapping column with mobile phase A at a flow rate of 2.5 μl/min in 8.5 min, and the nucleoside mixture was separated on the analytical column by using a 36.5-min gradient of 1–20% B in 9.2 min, 20–60% B in 9.2 min, 60–95% B in 0.6 min, then, at 95% B for 9 min, 95–1% B in 0.5 min, and finally, at 1% B for 8 min. The flow rate was 300 nl/min.

The calibration curves for the quantifications of modified ribonucleosides (m⁵C and m⁶A) were constructed by spiking different amounts of unlabeled standard nucleosides (rC and m⁵C, rA and m⁶A) and fixed amounts labeled standards, i.e. 3.77 pmol of [¹³C₅]-rC, 10.3 fmol of [¹³C₅]-m⁵C, 2.825 pmol of [¹³C₅]-rA and 32.51 fmol of [D₃]-m⁶A. The mixtures were then subjected to LC-MS/MS analyses under the same conditions as described above for the samples. The levels of m⁵C and m⁶A are represented as percentage values relative to rC and rA, respectively.

UVC crosslinking

For UVC crosslinking samples, NSUN2 recombinant protein (30 ng/μl) was incubated with or without 1 mM glucose at room temperature for 15 min and was crosslinked by UVC (254 nm, UV Stratalinker 2400) at 0.3 J/cm² in 2.5 μl droplets on a parafilm, crosslinking a total of 20 droplets per sample. All 20 droplets for each sample were combined for in-gel digestion.

BioID

In keratinocytes, bioID was performed with cells containing stable virus integrations of wild-type NSUN2 at endogenous levels or, as the control, uORF-BASU-NLS-GFP. Cells were labeled with 100 μM biotin for 5–6 h and harvested by washing twice with 1X ice cold PBS and lysed on ice for 30 min (50 mM Tris–HCl, pH 7.5, 300 mM NaCl, 1 mM EDTA, 1% Triton X-100, 1X Protease Inhibitor Cocktail), and sonicated for 3 cycles of 10 s at 10% amplitude, followed by 10 s pauses. After sonication, cell lysates were centrifuged through a 3 kDa filter (Millipore, UFC900324) to remove excess biotin and to exchange the buffer to 50 mM Tris–HCl, pH 7.5, 500 mM NaCl, 0.2% sodiumdodecyl sulfate (SDS), containing 1X Protease Inhibitor Cocktail. Cell lysates were incubated with MagReSyn Streptavidin beads (Resyn Biosciences, MR-STV010) with 25 μl of beads per mg of lysate for 2 h at room temperature or overnight at 4°C. Beads were then washed with buffer 1 (3% Lithiumdodecyl sulphate (LDS)) twice, buffer 2 (50 mM HEPES, pH 7.5, 500 mM NaCl, 1 mM EDTA, 0.1% Na-DOC, 1% TritonX-100) once and buffer 3 (10 mM Tris–HCl, pH 7.5, 250 mM LiCl, 1 mM EDTA, 0.5% Na-DOC, 0.5% NP-40) once, with a final wash of 50 mM Tris–HCl, pH 7.5. After washing, beads were then subjected to on-bead digestion.

Sample preparation and LC-MS/MS analysis

For on-bead digestion, bead-captured protein was reduced with dithiothreitol and alkylated with iodoacetamide. Processed proteins were subsequently digested overnight with Trypsin/Lys-C (Promega, V5071) at an enzyme/substrate ratio of 1:100 in 50 mM NH₄HCO₃ (pH 8.5) at 37°C.

For in-gel digestion, samples were loaded onto a 10% Bis-Tris gel. After electrophoresis, the desired gels were cut into slices, reduced in-gel with dithiothreitol and alkylated with iodoacetamide. Processed proteins were subsequently digested in-gel, overnight, with Trypsin/Lys-C (Promega, V5071) at an enzyme/substrate ratio of 1:100 in 50 mM NH₄HCO₃ (pH 8.5) at 37°C. The next day, peptides were recovered from gels first using a solution containing 5% acetic acid in H₂O and second with a solution containing 2.5% acetic acid in an equal volume mixture of CH₃CN and H₂O.

All the resulting peptide mixture was subsequently dried in a Speed-vac and desalted using OMIX C18 pipet tips (Agilent Technologies, A57003100). LC-MS/MS experiments were conducted on a Q Exactive Plus mass spectrometer equipped with an UltiMate 3000 UPLC system (Thermo Fisher Scientific).

Samples were automatically loaded at 3 μl/min onto a precolumn (150 μm i.d. and 3.5 cm in length) packed with ReproSil-Pur 120 C18-AQ stationary-phase material (5 μm in particle size, 120 Å in pore size, Dr Maisch). The precolumn was connected to a 20-cm fused-silica analytical column (PicoTip Emitter, New Objective, 75 μm i.d.) packed with 3 μm C18 beads (ReproSil-Pur 120 C18-AQ, Dr. Maisch). Peptides were then resolved using a 180-min gradient of 2–45% ACN in 0.1% formic acid. The flow rate was maintained at 300 nl/min.

The mass spectrometer was operated in a data-dependent acquisition mode. Full-scan mass spectra were acquired in the range of m/z 350–1500 using the Orbitrap analyzer at a resolution of 70 000 at m/z 200. Up to the 25 most abundant ions found in MS with a charge state of 2 or above were sequentially isolated and collisionally activated in the HCD cell with a normalized collision energy of 28 to yield MS/MS.

Mass spectrometry data analysis

For bioID, Maxquant (27) (v. 2.0.1.0) was used to analyze the LC-MS and MS/MS data for protein identification and quantification in LFQ mode, searching against the human Uniprot database, Proteome ID UP000005640 (reviewed and with isoforms). The maximum number of miss-cleavages for trypsin was two per peptide. Cysteine carbamidomethylation and methionine oxidation were set as fixed and variable modifications, respectively. The tolerances in mass accuracy were 20 ppm for both MS and MS/MS. The maximum false discovery rates (FDRs) were set at 0.01 at both peptide and protein levels, and the minimum required peptide length was six amino acids. Ratios were calculated from normalizing the intensity to the BASU-NLS-GFP control for each condition in each biological replicate. Proteins with an average ratio >2, and enriched in all the three individual biological replicates, were considered as bioID enriched targets.

For UVC crosslinking-MS, raw files were first searched by Maxquant (27) (v. 2.0.2.0) using NSUN2 protein sequence FASTA. Label-free quantification was applied. The maximum number of miss-cleavages for trypsin was two per peptide. Cysteine carbamidomethylation and methionine oxidation were set as fixed and variable modifications, respectively. The tolerances in mass accuracy were 20 ppm for both MS and MS/MS. The maximum FDRs were set at 0.01 at both peptide and protein levels, and the minimum required peptide length was six amino acids. Unique peptides were quantified and examined manually with plotting the chromatography at 10 ppm m/z window to identify if any had changed abundance.

Fast protein liquid chromatography (FPLC)

HEK293T cells from two 10 cm plates (70% confluency) were harvested, washed with ice cold PBS and lysed 5–10 volumes of ice-cold lysis buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 1.5 mM MgCl₂, 0.5 mM DTT, 1.25X protease inhibitors and 0.6% Triton X-100). Lysates were rotated at 25°C on with 5 U/ml of benzonase (Sigma, E1014-25KU) with or without 1 mM glucose for ∼30 min and then clarified by centrifugation for 10 min (maximum speed) at 4°C. Protein complexes were separated with a Superose 6 Increase 10/300 column (Cytiva 29–0915-96, 10/300 GL) on a AKTA Pure protein purification system attached to the fraction collector F9-C. The column was equilibrated with two column volumes of elution buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 1.5 mM MgCl₂, 0.5 mM DTT, 1.25X protease inhibitors and 0.6% Triton X-100) and then run using one column volume of elution buffer after sample loading (200 μl loop) to elute protein complexes. Automatic sample collection was set to 250 ml/fraction. A mixture of thyroglobulin, g-globulin, ovalbumin, myoglobin and vitamin B12 was used as gel filtration standards (BioRad, #1 511 901).

Immunoprecipitation

Approximately 3 million cells were harvested and washed with ice-cold PBS. Cells were lysed with lysis buffer (50 mM Tris–HCl, pH 7.5, 250 mM NaCl, 10% glycerol, 0.1% NP-40, 1X protease inhibitor) on ice for 20–30 minutes with occasional vortexing. Lysates were passed through 27.5 g needles and centrifuged for 15 min at 13 200 rpm, at 4°C. Approximately 300 μg protein from this resulting nuclear lysate was combined with an equal volume of Tris buffer (50 mM Tris–HCl, pH 7.5, 1X protease inhibitor) for a final concentration of ∼1 mg/ml protein. Normalized nuclear lysates were combined with 2 μg of either anti-NSUN2 (ProteinTech, 20854–1-AP), IgG (CST, #3900) or anti-RPLP0 (SCBT, sc-293260) and rotated at 4°C overnight. Lysates were further incubated with 20 μl protein G dynabeads (Invitrogen, 10004D) for 2 h at 4°C and beads washed three times with wash buffer (50 mM Tris–HCl, pH 7.5, 150 mM NaCl, 0.02% NP-40). After washing, beads were boiled with 1x LDS sample buffer and analyzed by western blot.

Proximity ligation assay (PLA)

Protein protein interaction in living cells was measured by proximity ligation assay (PLA) with Duolink In Situ Orange Starter Kit Mouse/Rabbit (DUO92102-1KT) according to the manufacturer’s instructions. Briefly, anti-NSUN2 (ProteinTech, 20854–1-AP) and anti-RPLP0 (SCBT, sc-293260) antibodies were applied for PLA primary antibody incubation. After Duolink PLA probe incubation, ligation and amplification, PLA samples were imaged by a Zeiss LSM880 confocal microscope.

In vitro transcription

Full-length transcripts for GRHL1 and GRHL3 were cloned into linearized pGeneArt1 (pGA.1, Synthesized from Life Technologies, P01505112) vector. Approximately 1 μg linearized transcript-containing pGA.1 vectors were used as a template in an in vitro transcription reaction for 2 h at 37°C, using the MEGAscript T7 transcription kit (ThermoFisher, AM1334) following the manufacturer’s instruction. The resulting mixture was diluted with 50 μl RNase-free water, and the in vitro transcribed RNA was then purified using RNeasy Plus kit (QIAGEN, 74 136). RNA integrity was tested by agarose-formaldehyde gel electrophoresis.

In vitro translation assay

In vitro translation assay was performed using Transcend™ Non-Radioactive Translation Detection System following the manufacturer’s protocol. Approximately 1.5 μg full-length RNAs were used as a template in an in vitro translation assay for 1 h at 30°C, using the Rabbit Reticulocyte Lysate (Promega, L4960), RNasin® Ribonuclease Inhibitor (Promega, N2511), Amino Acid Mixtures (Promega, L4461) and Transcend™ tRNA (Promega, L506A), with or without 500 ng NSUN2 or NSUN2^K28G/R29G. The resulting mixture (1 μl) was diluted with 9 μl 1XLDS sample buffer and heated to 90°C for 2 min. Samples proceeded to western blot analysis.

Cross-linking immunoprecipitation (CLIP-seq)

CLIP-seq was performed on HA-tagged NSUN2 in keratinocytes according to the published easyCLIP protocol (28,29), with 0.05% NP-40 in NT2 buffer, and the PCR was performed in two stages (described below). Briefly, L3 adapters include a phosphorylated 5′ end with a sequence of random UMI RNA nucleotides (rN), a barcode (B) and an azide group at the 3′ end (/3AzideN/) for dye conjugation and blocking 3′ ligation, in the sequence /5Phos/rNrNrN rNrNB BBB BAG ATC GGA AGA GCA CAC GTC AAA AAA AAA AAA AAA AAA AAA AAA /3AzideN/. L3 adapters were column purified and pre-conjugated to DBCO-dyes before use. L5 adapters had the sequence /5AzideN/ TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG GTA TAG rNrNrN rNrNrN rN, where /5AzideN/ denotes a 5′ azide group for dye conjugation and ligation blocking.

Keratinocytes were washed once with ice-cold PBS before being UVC crosslinked at 254 nm in 10 cm dishes or 6-well plates in a Stratalinker at 0.3 J/cm². Cells were then collected in ∼1 ml of 4°C CLIP lysis buffer and ∼2 mg clarified lysate per replicate was mixed with 22 μl anti-HA magnetic beads (ThermoFisher #88 837) and incubated for an hour. After washing, beads were treated with 0.02 U/μl Ambion RNAse I (ThermoFisher, #AM2294, (30° for ∼3 min), dephosphorylated at the 3′ end by PNK, and L3 adapters were ligated overnight at 16°. Following L3 adapter ligation, replicates were combined, 5′ phosphorylated and ligated to L5 oligos. RNA–protein complexes were separated by SDS-PAGE, transferred to nitrocellulose, and RNA was extracted from the membranes using proteinase K. RNA was then purified by capturing the poly(A) tail of the L3 adapters, reverse transcribed and PCR amplified.

Two rounds of PCR were performed. The initial round used the primers TCG TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG and 0GTT CAG ACG TGT GCT CTT CCG ATC T, and the PCR program was 5 s at 98°C, followed by 5 cycles of: 5 s at 98°C, 5 s at 58°C, 45 s at 72°C, then 6–10 cycles of: 5 s at 98°C, 5 s at 60°C, 45 s at 72°C. PCR products were then separated on a 4% E-gel (ThermoFisher, #G401004) to isolate products with >15 nt inserts. A second round of PCR was then performed as 5 cycles of PCR with full-length Illumina indexing primers (29) for 5 s at 98°C, 1 min at 68°C. The resulting PCR product was purified using a column. Sequencing was on a NovaSeq X plus, with paired-end 150 bp reads, at ∼30 million reads per sample.

Ribo-seq

Cells were harvested, lysed on ice for 30 min with lysis buffer (20 mM Tris–HCl, pH 7.4, 150 mM NaCl, 5 mM MgCl₂, 1 mM DTT, 100 μg/ml cycloheximide, 1% Triton X-100 and Turbo DNase I 25 U/ml) and centrifuged for 10 min at 20 000 g, 4°C. Apprximately 0.25 μl RNase I (Invitrogen, AM2294) was added to 100 μl cell lysate. Lysate was incubated for 45 min at room temperature with gentle mixing, before 3.3 μl SUPERase*In RNase inhibitor (ThermoFishier, AM2696) was added to stop nuclease digestion. RNA was purified with a ZR small-RNA PAGE Recovery Kit (Zymo) following the manufacturer’s protocol. The resulting RNA was incubated in 1X PNK buffer, 3 mM ATP and 1 U/μl (1 μl) PNK (NEB) at 37°C for 30 min, before purification with the same small-RNA recovery kit. Purified RNA was then loaded onto a 15% TBE-urea gel (EC6885BOX) with the running buffer (1XTBE) kept at ∼55°C. The ∼25–45 nt band was then cut out for ZR small-RNA PAGE Recovery Kit purification. The purified RNA was then subjected to library preparation using NEBNext® Small RNA Library Prep kit (E7330S), following the manufacturer’s protocol. Sequencing was performed on a NovaSeq X plus, with paired-end 150 bp reads, at ∼50 million reads per sample.

Sequencing data analysis

RNA-seq analysis. For all RNA-seq analyses, Gencode v39 annotations and the GRCh38 genome were utilized. Following the manufacturer’s guidelines, Lexogen reads were processed, aligned to the the Gencode v39 genome (‘basic’ annotations) with STAR (30) (v. 2.7), and gene quantification was performed using FeatureCounts (31) (part of Rsubread v.2.10.5). The resulting counts were analyzed using DESeq2 (32) (v.1.38.3).

CLIP-seq analysis. Inline barcodes and unique molecular identifiers (UMIs) were transferred to the read names using Cutadapt (33) (v. 1.8.1), which also clipped adapter sequences. All mapping was performed using STAR (v. 2.7). Initially, reads were mapped to a genome constructed from the rDNA locus sequence (Genbank accession U13369.1) using EndToEnd alignment. Reads that did not map were subsequently aligned to a genome of curated repetitive elements from Dfam release 3.3 (https://www.dfam.org/releases/Dfam_3.3/families/) (34) with the EndToEnd parameter and excluding splicing. Remaining unmapped reads were then mapped to the Gencode v39 genome with default settings and the parameter –limitOutSJcollapsed 2000000. UMI_tools was used to collapse read duplicates based on UMIs in the read names, filtering reads with a MAPQ score over 10 and discarding the second read in each pair. Conversion of formats to bedgraph, wig and bigwig was achieved using bedtools (version 2.30.0). Reads were assigned to genes using htseq-count on deduplicated bam files and only considering exons, after which reads not mapping to an exon were assigned to overlapping introns.

HOMER motif discovery. The MACS2 algorithm was run to identify peaks in genomic and transcriptomic bam files. The parameters set were –nomodel -q 0.01 –bw 100. Subsequently, peaks intersecting with non-coding RNAs or lacking a distinct strand orientation (a minimum of 5-fold enrichment in one strand) were discarded. Peaks on chromosomes other than the autosomes, X, Y and mitochondrial genome were also excluded. Peaks exhibiting a FDR <0.01 were retained. Exonic peaks were defined as those overlapping an exon by ≥40% their length. Control sequence files, comprising a minimum of 20 000 sequences, were generated with random bases that matched the lengths of the peaks. The findMotifs perl script from the HOMER suite was used to search for motifs ranging from 6 to 8 nucleotides in length, with the parameters -rna and -homer1.

Ribo-seq QC. For ribosome profiling QC, NEB small adapters were removed with k = 13 ktrim = r useshortkmers = t mink = 5 qtrim = t trimq = 10 minlength = 20, using the adapters file:

>read1adapter

AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC

>read2adapter

GATCGTCGGACTGTAGAACTCTGAACGTGTAGATCTCGGTGGTCGCCGTATCATT

Reads were then mapped to the hg38 genome with bowtie2. Reads were filtered for MAPQ, primary alignments and canonical chromosomes with samtools. Reads were subsequently filtered to remove reads overlapping ncRNA. The R package ribosomeProfilingQC (35) was used for analysis, following the package vignette. Reads with mapped inserts of length 30–36 had the strongest enrichment of the CDS and so were used for all analysis.

Ribo-seq translational efficiency. For translational efficiency analysis, NEB small adapters were removed in fastq files as for Ribosome profiling QC. The general analysis pipeline was a snakemake that performed steps of a published analysis guide (36), both for Ribo-seq and paired RNA-seq data. Reads were filtered for MAPQ, primary alignments and canonical chromosomes were selected for, and reads overlapping ncRNA removed, using the same methods as for ribosome profiling QC. Only Ribo-seq reads of length 30–36 were used. Reads with a P site mapping to a CDS were quantified according to the published method (36) using the ENSEMBL database v.92. The reads-per-gene tables for RNA-seq and Ribo-seq were analyzed for differential translation efficiency (TE) by xtail (37) using the docker container provided by the authors (R v.3.6.1).

Statistical analysis of biological data

All results are presented as the mean with SD unless notified specifically. Statistics were assessed using a Mann–Whitney U-test [motif enrichment]. P-values were calculated from two-tailed Student’s t-test if not notified specifically. All the GO analysis were performed with Enrichr and all the adjusted P-values for GO analysis were exported from Enrichr (38–40) if not stated specifically. Markers of significance are as follows: #, P> 0.05; *, 0.01 < P≤ 0.05; **, 0.001 < P≤ 0.01; ***, P≤ 0.001.

Results

Glucose binds NSUN2

Glucose polymer affinity chromatography and azido-glucose pull-down recently identified NSUN2 as a novel glucose-binding protein (13,16). To assess if this is a direct interaction, MST was performed with purified recombinant Flag-HA-His (FHH) tagged NSUN2 protein and glucose (Figure 1A and Supplementary Figure S1A). Glucose exhibited significant binding affinity to NSUN2, whereas galactose, a glucose diastereomer, did not show comparable binding (Figure 1A), indicating that the shape of the hexose molecule is crucial for NSUN2 binding. The non-metabolizable glucose analog, 3-O-methyl-D-glucose (3OMG), which can rescue impaired epidermal differentiation due to glucose restriction (16), displayed a similar binding affinity to NSUN2 as glucose (Figure 1A). A fluorescence-based assay also suggested a dissociation constant below the glucose concentration in differentiated keratinocytes (Supplementary Figure S1B). To perform additional evaluation of glucose binding, CETSAs (24,41) were conducted. CETSA demonstrated that glucose increased the thermal stability and solubility of NSUN2 (Supplementary Figure S1C and D). To further test the glucose-NSUN2 interaction, a biotin glucose probe was applied (22). Recombinant FHH-NSUN2 protein was enriched in the biotin-glucose pulldown compared to the biotin-only pulldown, and this binding was abolished in the presence of glucose (Supplementary Figure S1E–G), suggesting the pulldown of biotin glucose is glucose-specific. This specificity was further validated in cell lysate pulldown assays, where glucose disrupted the interaction between the probe and NSUN2 (Supplementary Figure S1H). These findings provide strong evidence that glucose directly binds to the NSUN2 protein.

Figure 1.

Glucose binds to NSUN2 and NSUN2 is essential for differentiation. (A) MST quantifies NSUN2’s binding affinity for glucose, 3OMG and galactose. (B) Overlap of downregulated genes (FDR < 0.05) between the two shRNAs for NSUN2 versus shScramble in RNA-seq of human epidermal organoid tissues. (C) Biological process GO terms for RNAs downregulated upon NSUN2 depletion (the 897 gene intersection in Figure 1B). (D) Changes in progenitor and differentiation markers upon NSUN2 loss in 2D day 3 differentiated keratinocytes and in skin organoid tissues (RNA-seq); differentiation markers shown include all genes with a GO term of keratinocyte differentiation, regulation of keratinocyte differentiation, or keratinization, with a fold-change >4 in any biological replicate. (E) Keratin 1 (K1), Keratin 10 (K10), Filaggrin (FLG), NSUN2 and Col7 immunostaining in regenerated human skin organoid tissues, treated with NSUN2 shRNAs or control shRNA; dotted line, basement membrane. Error bars = S.D. from ≥3 biological replicates.

NSUN2 is essential for epidermal differentiation

Given the requirement for glucose elevation in epidermal differentiation, a role for NSUN2 in this process was next explored. Depletion of NSUN2 in human epidermal keratinocytes cultured in differentiating conditions of confluence and elevated calcium (42) resulted in a failure to induce the full expression of differentiation genes, as evidenced by RNA-sequencing (Supplementary Figure S1I and J). Similarly, RNA-seq and qPCR in regenerated human skin organoid tissues (14,15,43–45) revealed that NSUN2 loss led to impaired expression of canonical mRNAs typically induced during epidermal differentiation (Figure 1B–D and Supplementary Figure S1K). RNA-seq experiments were performed with the two most efficient shRNA knockdowns (Supplementary Figure S1I and K). Moreover, early differentiation markers such as keratin 1 and keratin 10 proteins showed reduced protein expression in suprabasal layers, while the late differentiation marker filaggrin was altogether absent, indicating a marked deficiency in terminal differentiation (Figure 1E). These observations point to an essential role for NSUN2 in epidermal differentiation.

Glucose binding promotes NSUN2 binding to SAM

NSUN2, a methyltransferase, utilizes S-adenosyl-L-methionine (SAM) as a donor for methyl group transfer to RNA. It was recently shown that glucose promotes NSUN2 oligomerization and activates its methyltransferase activity in tumorigenesis and immunotherapy resistance (22). The same work identified the N-terminal amino acids 1–28 of NSUN2 as required for glucose binding (22). To identify the NSUN2 amino acids required for glucose binding, UVC crosslinking mass spectrometry (MS) was performed. This identified a peptide (GEAGWEGGYPEIVKENK) from amino acids 30–46 with decreased abundance after UVC crosslinking with added glucose (Figure 2A and Supplementary Figure S2A). Combined with the known glucose binding site at amino acids 1–28 (22), these data implicated K28/R29 as potentially important for NSUN2 binding to glucose. This was confirmed by substituting both amino acids with glycine (termed NSUN2^K28G/R29G), which abolished glucose binding (Figure 2B). Intactness of NSUN2^K28/R29 is thus indispensable for glucose binding.

Figure 2.

Glucose binding to NSUN2 requires amino acids K28/R29. (A) Relative abundance of the peptides with no glucose modification, calculated from the intensity obtained from 1 mM glucose/no-glucose, both with UVC crosslinking. The peptides with increased, no changes and decreased abundance after glucose treatment has been labeled with different colors. (B) MST of NSUN2 WT and NSUN2^K28G/R29G mutant binding affinity for glucose. (C) Molecular docking by UCSF Chimera shows the predicted binding site of SAM to NSUN2 (structure: AF-Q08J23-F1 from AlphaFold). Peptide sequences with increased accessibility were highlighted in the structure. (D) MST of NSUN2 binding affinity for SAM ± 350 μM glucose or galactose. (E) Percentage of m⁵C/rC with NSUN2 shRNAs and control shRNA treatment from two biological replicates and two technical replicates. (F) Percentage of m⁵C/rC in keratinocyte progenitor cells (Prog.) and day 3 differentiated cells (Diff.) from four biological replicates and two technical replicates. Error bars = S.D. from ≥3 biological replicates if not notified specifically. *, 0.01 < P≤ 0.05; **, 0.001 < P≤ 0.01; ***, P≤ 0.001.

UVC crosslinking MS revealed peptides with increased abundance after adding glucose, concentrated around a specific binding pocket (Figure 2A). Autodocking using UCSF Chimera identified these peptides as surrounding the SAM binding site (Figure 2C), close to the reported beta-sheet structure (46), suggesting that glucose might be altering the SAM binding pocket. SAM binding was therefore explored, revealing that wild-type NSUN2 exhibited low SAM-binding affinity (Figure 2D). Physiological concentrations of glucose (350 μM) (16) increased NSUN2 binding to SAM by 12.8-fold, whereas the galactose hexose control had minimal effects (Figure 2D). The NSUN2^K28G/R29G glucose-binding mutant showed minimal SAM binding and, in contrast to wild-type NSUN2, glucose did not increase its binding affinity for SAM (Supplementary Figure S2B). These data suggest that glucose binding enhances NSUN2 binding to SAM through a conformation shift.

NSUN2 is a SAM-binding methyltransferase that utilizes the methyl donor SAM as a cofactor to methylate proteins, small molecules, lipids and nucleic acids (47). Thus, SAM binding is directly related to the methyltransferase activity of the NSUN2 protein. Therefore, the activity of NSUN2 was further assessed by measuring global m⁵C levels. Depletion of NSUN2 resulted in a decrease in m⁵C in both total RNA and mRNA, without affecting the alternative RNA modification m⁶A (Figure 2E and Supplementary Figure S2C–E). Conversely, increasing intracellular glucose concentrations during differentiation was correlated with a 1.3-fold increase in m⁵C in total RNA, with a 2-fold increase in m⁵C seen in mRNA (Figure 2F). In contrast, no increase in m⁶A level was observed during differentiation; in fact, there was a slight decrease in m⁶A levels of mRNA (Supplementary Figure S2F). These results are consistent with NSUN2 activation during differentiation, leading to increased m⁵C modification in RNA, particularly in mRNA.

Glucose promotes NSUN2 binding to specific mRNAs

In light of NSUN2’s RNA-binding properties, its RNA-binding specificity was next explored using crosslinking and immunoprecipitation followed by sequencing (CLIP-seq, specifically easyCLIP-seq) (28). This identified enrichment of a GGUUCRAD RNA motif (Figure 3A) (where R denotes G or A, and D denotes A, G or U), with the published CNGGG NSUN2-dependent m⁵C site motif (48) also highly enriched (Supplementary Figure S3A). Most reads were mapped to tRNA or rRNA, with tRNA being the most frequent interaction (Supplementary Figure S3B and Supplementary Table S1), consistent with NSUN2’s known role as a tRNA and rRNA modification factor (49). tRNA crosslinking in easyCLIP-seq correlated well (Spearman 0.77) with published miCLIP-seq data, despite miCLIP differing in representing an enzymatic capture (Supplementary Figure S3C) (50). Glucose did not significantly alter the binding of NSUN2 to tRNA, rRNA or mRNA in CLIP-seq (Supplementary Figure S3B). Moreover, MST showed that the binding affinity of tRNA towards NSUN2 remained unchanged with or without glucose (Supplementary Figure S3D). However, within mRNA, glucose moved NSUN2 contacts from introns to exons (Figure 3B), indicating that NSUN2 shifted from pre-mRNA to mature mRNA upon glucose binding.

Figure 3.

Glucose regulates NSUN2 binding to specific pro-differentiation mRNA. (A) Motif identified by HOMER de novo searching of NSUN2 CLIP-seq peaks. (B) Binding of NSUN2 to mRNA exons or introns in undifferentiated (Prog.) and differentiated human keratinocytes grown in low (Low Glc.) or normal glucose (Normal Glc.). (C) Heatmap of the binding of NSUN2 to the NSUN2-mediated m⁵C levels of known RNAs in differentiated keratinocytes grown in normal versus low glucose (N/L). (D) Binding traces of NSUN2 to RNY1 RNA in differentiated keratinocytes grown in low or normal glucose from two independent donors. (E) Heatmap of the binding of NSUN2 to pro-differentiation mRNAs in the exons or introns in differentiated keratinocytes grown in normal versus low glucose (N/L). (F) Binding traces of NSUN2 to JUP, SPRR1B and S100A9 mRNAs in differentiated keratinocytes grown in low or normal glucose. Error bars, standard deviations from ≥3 biological replicates. **, 0.001 < P≤ 0.01; ***, P≤ 0.001.

Next, the binding of NSUN2 to known RNA targets with NSUN2-mediated m⁵C modifications was examined (50). Known NSUN2 target non-coding RNAs exhibited modestly higher binding in cells maintained under normal glucose conditions compared to low glucose conditions (Figure 3C), supporting the role of glucose in facilitating NSUN2 modification of known m⁵C RNA targets. Among these targets, RNY1, a small non-coding Y RNA transcribed by the same Pol III polymerase as tRNA and that enables DNA replication (51), showed the most enrichment in normal glucose conditions compared to low glucose conditions (Figure 3C), especially in regions with m⁵C modification on the RNA (52) (Figure 3D). NSUN2 also exhibited increased binding to specific pro-differentiation mRNAs, such as in the JUP, SPRR1B and S100A9 mRNAs, where cytosine methylation is also high (Figure 3E and F). Additionally, the majority of m⁵C modifications are located in 3′UTRs (Supplementary Figure S3E) (52). During differentiation, increased intracellular glucose concentrations led to increased NSUN2 engagement of the 3′UTR of mRNA, accompanied by decreased binding to the coding sequence (CDS) (Supplementary Figure S3F). These findings suggest a connection between NSUN2’s interaction with m⁵C sites on mRNA and epidermal differentiation.

Glucose promotes NSUN2 binding to translation-related proteins

To further elucidate the role of glucose regulation in NSUN2 association with pro-differentiation RNAs, a BioID experiment (in which a mutated biotin ligase is fused to the protein of interest and biotinylates the proximal proteome (53)) was conducted in differentiated keratinocytes cultured in normal or low glucose media (Supplementary Figure S4A). BioID analysis nominated 254 NSUN2 proximal proteins, with more proximal proteins detected in normal glucose conditions than with glucose restriction (Supplementary Figure S4B and Supplementary Table S2). Gene ontology (GO) analysis indicated that glucose promotes the association of NSUN2 with translation-related proteins, whereas glucose restriction led to an enrichment of NSUN2 associations with proteins involved in the negative regulation of translation (Figure 4A). Comparing the proteins associated with NSUN2 under normal glucose to the glucose restricted conditions that still support full cell viability (16) revealed enrichment of GO terms related to protein translation (Figure 4B and Supplementary Figure S4C), with many translation-related proteins significantly enriched in normal glucose versus low glucose conditions (Figure 4C and D). The interaction between NSUN2 and the translational protein RPLP0 was validated using co-immunoprecipitation (co-IP) and PLAs. Co-IP revealed a strong interaction between the two proteins (Figure 4E and F), and PLA further confirmed the NSUN2-RPLP0 association in differentiated keratinocytes (Figure 4G and H). The NSUN2 complex was further investigated using fast protein liquid chromatography (FPLC), which showed that with glucose binding, NSUN2 assembled into a larger molecular weight complex (Figure 4I and J; Supplementary Figure S4D). The known m⁵C reader YTHDF proteins (54) were also enriched in the NSUN2 proximal proteome under normal glucose conditions (Figure 4C), raising the possibility that glucose may modulate m⁵C reading activity. These findings suggest that glucose modulates the proximity of NSUN2 to translation-related proteins and implicates NSUN2 in translational regulation.

Figure 4.

Glucose promotes NSUN2 binding to translation-related complex. (A) Adjusted P-values of top biological process GO terms for the genes enriched in NSUN2 proximal proteomes in low or normal glucose conditions compared to corresponding GFP. (B) Adjusted P-values of top biological process GO terms for the genes enriched in NSUN2 proximal proteomes in normal glucose conditions compared to low glucose using the ratios of NSUN2 bioID normalized to the corresponding GFP bioID. (C) Volcano plot of the protein enrichment in normal glucose versus low glucose of NSUN2 bioID (normalized to corresponding GFP bioID) of the 254 enriched proteins. Some of the translation-related proteins and some other noteworthy proteins were highlighted. (D) Heatmap of the translation-related proteins enriched in NSUN2 proximal proteomes in low glucose or normal glucose conditions. (E) Western blot of NSUN2 and RPLP0 co-IP in day 3 differentiated cells. (F) Relative co-purification enrichment in NSUN2 and RPLP0 co-IP normalized to IgG control. (G) PLA signal for the interaction between NSUN2 and RPLP0 in differentiated cells maintained in normal glucose. (H) Quantitative analysis of PLA signal. (I) Western blot of NSUN2 and β-actin in FPLC from 14.25 to 19.25 ml, ±1 mM glucose incubation in the cell lysates. (J) Distribution of NSUN2 protein in FPLC, ±1 mM glucose in the cell lysates, from western blot. Error bars, standard deviations from ≥3 biological replicates. #, P> 0.05; *, 0.01 < P≤ 0.05; **, 0.001 < P≤ 0.01; ***, P≤ 0.001.

NSUN2 promotes the translation of pro-differentiation mRNAs

To further investigate how NSUN2 regulates protein translation, Ribo-seq experiments (55,56) were performed in differentiated cells with and without NSUN2 depletion (Supplementary Table S3). The resulting reads exhibited the three-nucleotide periodicity and CDS enrichment of good-quality Ribo-seq data (Supplementary Figure S5A–C). TE was calculated by comparing Ribo-seq data to mRNA levels obtained from mRNA-seq (37). Following NSUN2 knockdown, the most significant changes were decreases in TE (Figure 5A), suggesting that NSUN2 promotes translation. To account for potential biases from large mRNA level changes, we focused on mRNAs where TE was significantly altered while mRNA levels remained stable to within log2 fold changes of 0.5 (41%), identifying these as mRNAs with unbiased TE. TE was significantly decreased for these unbiased mRNAs (Mann–Whitney P= 10⁻⁴ versus all RNAs, Figure 5B). GO analysis further revealed an enrichment of terms related to skin and epidermal development among the mRNAs with reduced, unbiased TE upon NSUN2 depletion (Figure 5C), supporting the role of NSUN2 in promoting the translation of pro-differentiation mRNAs. NSUN2 exhibited stronger binding to mRNAs with decreased TE upon NSUN2 knockdown (Figure 5D), suggesting that NSUN2 directly binds to these mRNAs and promotes their translation.

Figure 5.

Glucose regulates NSUN2 and drives translation of pro-differentiation mRNAs. (A) Volcano plot of TE differences between shNSUN2 and shScramble. (B) Kernel density smoothed histogram of the change in translational efficiency upon NSUN2 knock-down for mRNA with significant efficiency changes (FDR < 0.1), but limited change in mRNA levels (|log2|<0.5). mRNAs with large changes in abundance upon NSUN2 loss were removed to avoid possible technical biases. P-value represents the change in TE in significant genes being different from all genes. (C) Adjusted P-values of top biological process GO terms for the genes with significantly decreased TE (FDR < 0.1) in shNSUN2 versus shScramble and limited (|log2|<0.5) mRNA level changes. (D) Box and scatter plots of the mean reads from easyCLIP-seq in day 3 differentiated cells maintained in normal glucose for RNAs with have significantly decreased TE (FDR < 0.1) in shNSUN2 versus shScramble and limited (|log2|<0.5) mRNA level changes (P-value, two-sided Mann–Whitney). (E) Binding traces of NSUN2 across GRHL3 mRNA in differentiated keratinocytes grown in low (black) or normal (red) glucose. (F) Western blot analysis of anti-Streptavidin representing GRHL1 or GRHL3 proteins, with or without adding FHH-NSUN2, from in vitro translation. (G) Quantitative analysis of the western blot result of GRHL1 (n = 3) and GRHL3 (n = 8) intensities from in vitro translation, normalizing to the condition without FHH-NSUN2 protein. (H) MST of NSUN2 binding to GRHL1 or GRHL3 full-length mRNAs. (I) Western blot analysis of anti-Streptavidin, representing GRHL3 protein with or without adding FHH-NSUN2^K28G/R29G. (J) Quantitative analysis of the western blot of GRHL3 intensity, normalizing to the condition without FHH-NSUN2^K28G/R29G protein (n = 5). Error bars, standard deviations from ≥3 biological replicates. #, P> 0.05; **, 0.001 < P≤ 0.01.

Given that the TE of many pro-differentiation mRNAs could be influenced by large changes in mRNA levels, an in vitro translation assay was performed to further investigate the translation of selected pro-differentiation mRNAs. Because NSUN2 exhibited glucose-dependent binding to GRHL3 mRNA at m⁵C modification sites, especially on the 3′UTR (Figure 5E), and given the central importance of GRHL3 for keratinocyte differentiation (57–59), the translation of GRHL3 protein was examined. The presence of NSUN2 modestly enhanced the TE of GRHL3 mRNA, while the levels of the control GRHL1 remained unchanged (Figure 5F and G). The observation that full-length GRHL3 mRNA binds 12-fold more tightly to NSUN2 than GRHL1 (Figure 5H) suggests that the differential effect on translation could be attributed to differences in binding affinity toward NSUN2. When the glucose-binding mutant NSUN2^K28G/R29G was introduced, however, the translation of GRHL3 mRNA was unaltered (Figure 5I and J), implying that glucose binding may be important for NSUN2-mediated promotion of protein translation.

Discussion

Here, we identify an essential role for the NSUN2 RNA-binding protein in epidermal differentiation facilitated by glucose. Glucose binding to the N-terminal region of NSUN2 enhances its binding to SAM and stimulates its enzymatic activity. Additionally, glucose promotes NSUN2’s association with translation complexes and increased translation of pro-differentiation mRNAs, like GRHL3. This multifaceted engagement of glucose with NSUN2 underscores the diverse molecular mechanisms through which RNA-binding proteins facilitate cellular differentiation.

The discovery of non-metabolic glucose actions has uncovered new roles for glucose in cellular processes, including epidermal differentiation (16) and tumorigenesis (22), distinct from its energy-related functions. Glucose binding has been shown to exert significant effects by modulating protein-small molecule, protein–nucleic acid or protein–protein interactions, namely glucose’s competitive binding to the ATP-binding domain of DDX21 (13), its facilitation of DNA-binding by IRF6 (16), and its promotion of NSUN2 multimerization and enzymatic activation (22). These findings suggest the existence of additional glucose-binding proteins and potential glucose-binding domains involved in various cellular processes, pointing to a broader role for glucose in cellular regulation beyond its traditional metabolic functions.

The NSUN2 binding motif found in this study perfectly matches the RNA motif of HRSP12, a protein that binds to YTHDF2 to orchestrate m⁶A-dependent endoribonucleolytic RNA decay (60). Specifically, BioID also showed that NSUN2 binds YTHDF proteins strongly in the elevated intracellular glucose concentrations seen in differentiated epithelial cells. YTHDF2 was also identified as an m⁵C reader protein (54), and YTHDF proteins have been known to facilitate translation of mRNA (61–63). Since m⁶A in mRNA would promote translation (64,65), the decrease in m⁶A levels in mRNA during differentiation, as indicated by our data (Supplementary Figure S2F), suggests that the facilitation of translation in differentiated keratinocytes is independent of m⁶A modification. These findings suggest that NSUN2 might cooperate with YTHDF proteins to traffic m⁵C-modified RNA to the translational complex for protein translation.

NSUN2 is predominantly found in the nucleolus but also appears in other cellular locations, such as the mitotic spindle during mitosis (66). Consistent with its nucleolar location, NSUN2 is involved in rRNA production and modifies tRNA (49), making it a part of the protein synthesis process. During epidermal differentiation, this process is altered, as analyses of polysome profiles revealed that key differentiation genes, including involucrin, are bound to heavy polysomes during differentiation, despite decreased general protein synthesis (67). The increased NSUN2 association with pro-epidermal differentiation RNAs such as vault RNAs (68) (Figure 3C), and a notable increase in NSUN2 proximity to translation factors, indicate a potential mechanism whereby NSUN2 contributes to vault RNA-regulated protein translation (69) during differentiation. Moreover, the observed alterations in total mRNA modification levels during epithelial differentiation imply a global reprogramming of the epitranscriptome, which is concurrent with changes in polysome profiles. Further investigation is warranted to fully understand the functional significance of these modifications and their impact on cellular transitions.

Data availability

All the mass spectrometry raw files have been deposited to the ProteomeXchange Consortium via the PeptideAtlas (70) partner repository with the dataset identifier PASS05886. All the sequencing data presented in the current publication have been deposited in and are available from the dbGaP database under dbGaP accession phs003767.v1.p1. Any additional information required to reanalyze the data reported in this paper is available from the Lead Contact upon request.

Supplementary data

Supplementary Data are available at NAR Online.

Funding

NIAMS/NIH [AR043799, AR045192 to P.A.K., R35 ES031707 to Y.W., K01AR084077 to W.M., K01AR082351 to D.F.P., K01AR070895 to V.L.-P., T32GM007276 to L.M.M.]; U.S. Department of Veterans Affairs [BX001409 to P.A.K.]. Funding for open access charge: NIAMS [AR043799].

Conflict of interest statement. None declared.